net/core/filter.c

   1 /*
   2  * Linux Socket Filter - Kernel level socket filtering
   3  *
   4  * Based on the design of the Berkeley Packet Filter. The new
   5  * internal format has been designed by PLUMgrid:
   6  *
   7  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   8  *
   9  * Authors:
  10  *
  11  *      Jay Schulist <jschlst@samba.org>
  12  *      Alexei Starovoitov <ast@plumgrid.com>
  13  *      Daniel Borkmann <dborkman@redhat.com>
  14  *
  15  * This program is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU General Public License
  17  * as published by the Free Software Foundation; either version
  18  * 2 of the License, or (at your option) any later version.
  19  *
  20  * Andi Kleen - Fix a few bad bugs and races.
  21  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/fcntl.h>
  28 #include <linux/socket.h>
  29 #include <linux/sock_diag.h>
  30 #include <linux/in.h>
  31 #include <linux/inet.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/if_packet.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/gfp.h>
  36 #include <net/inet_common.h>
  37 #include <net/ip.h>
  38 #include <net/protocol.h>
  39 #include <net/netlink.h>
  40 #include <linux/skbuff.h>
  41 #include <linux/skmsg.h>
  42 #include <net/sock.h>
  43 #include <net/flow_dissector.h>
  44 #include <linux/errno.h>
  45 #include <linux/timer.h>
  46 #include <linux/uaccess.h>
  47 #include <asm/unaligned.h>
  48 #include <asm/cmpxchg.h>
  49 #include <linux/filter.h>
  50 #include <linux/ratelimit.h>
  51 #include <linux/seccomp.h>
  52 #include <linux/if_vlan.h>
  53 #include <linux/bpf.h>
  54 #include <net/sch_generic.h>
  55 #include <net/cls_cgroup.h>
  56 #include <net/dst_metadata.h>
  57 #include <net/dst.h>
  58 #include <net/sock_reuseport.h>
  59 #include <net/busy_poll.h>
  60 #include <net/tcp.h>
  61 #include <net/xfrm.h>
  62 #include <net/udp.h>
  63 #include <linux/bpf_trace.h>
  64 #include <net/xdp_sock.h>
  65 #include <linux/inetdevice.h>
  66 #include <net/inet_hashtables.h>
  67 #include <net/inet6_hashtables.h>
  68 #include <net/ip_fib.h>
  69 #include <net/flow.h>
  70 #include <net/arp.h>
  71 #include <net/ipv6.h>
  72 #include <net/net_namespace.h>
  73 #include <linux/seg6_local.h>
  74 #include <net/seg6.h>
  75 #include <net/seg6_local.h>
  76
  77 /**
  78  *      sk_filter_trim_cap - run a packet through a socket filter
  79  *      @sk: sock associated with &sk_buff
  80  *      @skb: buffer to filter
  81  *      @cap: limit on how short the eBPF program may trim the packet
  82  *
  83  * Run the eBPF program and then cut skb->data to correct size returned by
  84  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
  85  * than pkt_len we keep whole skb->data. This is the socket level
  86  * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
  87  * be accepted or -EPERM if the packet should be tossed.
  88  *
  89  */
  90 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
  91 {
  92         int err;
  93         struct sk_filter *filter;
  94
  95         /*
  96          * If the skb was allocated from pfmemalloc reserves, only
  97          * allow SOCK_MEMALLOC sockets to use it as this socket is
  98          * helping free memory
  99          */
 100         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 101                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 102                 return -ENOMEM;
 103         }
 104         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 105         if (err)
 106                 return err;
 107
 108         err = security_sock_rcv_skb(sk, skb);
 109         if (err)
 110                 return err;
 111
 112         rcu_read_lock();
 113         filter = rcu_dereference(sk->sk_filter);
 114         if (filter) {
 115                 struct sock *save_sk = skb->sk;
 116                 unsigned int pkt_len;
 117
 118                 skb->sk = sk;
 119                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 120                 skb->sk = save_sk;
 121                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 122         }
 123         rcu_read_unlock();
 124
 125         return err;
 126 }
 127 EXPORT_SYMBOL(sk_filter_trim_cap);
 128
 129 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 130 {
 131         return skb_get_poff(skb);
 132 }
 133
 134 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 135 {
 136         struct nlattr *nla;
 137
 138         if (skb_is_nonlinear(skb))
 139                 return 0;
 140
 141         if (skb->len < sizeof(struct nlattr))
 142                 return 0;
 143
 144         if (a > skb->len - sizeof(struct nlattr))
 145                 return 0;
 146
 147         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 148         if (nla)
 149                 return (void *) nla - (void *) skb->data;
 150
 151         return 0;
 152 }
 153
 154 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 155 {
 156         struct nlattr *nla;
 157
 158         if (skb_is_nonlinear(skb))
 159                 return 0;
 160
 161         if (skb->len < sizeof(struct nlattr))
 162                 return 0;
 163
 164         if (a > skb->len - sizeof(struct nlattr))
 165                 return 0;
 166
 167         nla = (struct nlattr *) &skb->data[a];
 168         if (nla->nla_len > skb->len - a)
 169                 return 0;
 170
 171         nla = nla_find_nested(nla, x);
 172         if (nla)
 173                 return (void *) nla - (void *) skb->data;
 174
 175         return 0;
 176 }
 177
 178 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 179            data, int, headlen, int, offset)
 180 {
 181         u8 tmp, *ptr;
 182         const int len = sizeof(tmp);
 183
 184         if (offset >= 0) {
 185                 if (headlen - offset >= len)
 186                         return *(u8 *)(data + offset);
 187                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 188                         return tmp;
 189         } else {
 190                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 191                 if (likely(ptr))
 192                         return *(u8 *)ptr;
 193         }
 194
 195         return -EFAULT;
 196 }
 197
 198 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 199            int, offset)
 200 {
 201         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 202                                          offset);
 203 }
 204
 205 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 206            data, int, headlen, int, offset)
 207 {
 208         u16 tmp, *ptr;
 209         const int len = sizeof(tmp);
 210
 211         if (offset >= 0) {
 212                 if (headlen - offset >= len)
 213                         return get_unaligned_be16(data + offset);
 214                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 215                         return be16_to_cpu(tmp);
 216         } else {
 217                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 218                 if (likely(ptr))
 219                         return get_unaligned_be16(ptr);
 220         }
 221
 222         return -EFAULT;
 223 }
 224
 225 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 226            int, offset)
 227 {
 228         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 229                                           offset);
 230 }
 231
 232 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 233            data, int, headlen, int, offset)
 234 {
 235         u32 tmp, *ptr;
 236         const int len = sizeof(tmp);
 237
 238         if (likely(offset >= 0)) {
 239                 if (headlen - offset >= len)
 240                         return get_unaligned_be32(data + offset);
 241                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 242                         return be32_to_cpu(tmp);
 243         } else {
 244                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 245                 if (likely(ptr))
 246                         return get_unaligned_be32(ptr);
 247         }
 248
 249         return -EFAULT;
 250 }
 251
 252 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 253            int, offset)
 254 {
 255         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 256                                           offset);
 257 }
 258
 259 BPF_CALL_0(bpf_get_raw_cpu_id)
 260 {
 261         return raw_smp_processor_id();
 262 }
 263
 264 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 265         .func           = bpf_get_raw_cpu_id,
 266         .gpl_only       = false,
 267         .ret_type       = RET_INTEGER,
 268 };
 269
 270 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 271                               struct bpf_insn *insn_buf)
 272 {
 273         struct bpf_insn *insn = insn_buf;
 274
 275         switch (skb_field) {
 276         case SKF_AD_MARK:
 277                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
 278
 279                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 280                                       offsetof(struct sk_buff, mark));
 281                 break;
 282
 283         case SKF_AD_PKTTYPE:
 284                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 285                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 286 #ifdef __BIG_ENDIAN_BITFIELD
 287                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 288 #endif
 289                 break;
 290
 291         case SKF_AD_QUEUE:
 292                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
 293
 294                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 295                                       offsetof(struct sk_buff, queue_mapping));
 296                 break;
 297
 298         case SKF_AD_VLAN_TAG:
 299                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
 300
 301                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 302                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 303                                       offsetof(struct sk_buff, vlan_tci));
 304                 break;
 305         case SKF_AD_VLAN_TAG_PRESENT:
 306                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
 307                 if (PKT_VLAN_PRESENT_BIT)
 308                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
 309                 if (PKT_VLAN_PRESENT_BIT < 7)
 310                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 311                 break;
 312         }
 313
 314         return insn - insn_buf;
 315 }
 316
 317 static bool convert_bpf_extensions(struct sock_filter *fp,
 318                                    struct bpf_insn **insnp)
 319 {
 320         struct bpf_insn *insn = *insnp;
 321         u32 cnt;
 322
 323         switch (fp->k) {
 324         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 325                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
 326
 327                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 328                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 329                                       offsetof(struct sk_buff, protocol));
 330                 /* A = ntohs(A) [emitting a nop or swap16] */
 331                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 332                 break;
 333
 334         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 335                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 336                 insn += cnt - 1;
 337                 break;
 338
 339         case SKF_AD_OFF + SKF_AD_IFINDEX:
 340         case SKF_AD_OFF + SKF_AD_HATYPE:
 341                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
 342                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
 343
 344                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 345                                       BPF_REG_TMP, BPF_REG_CTX,
 346                                       offsetof(struct sk_buff, dev));
 347                 /* if (tmp != 0) goto pc + 1 */
 348                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 349                 *insn++ = BPF_EXIT_INSN();
 350                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 351                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 352                                             offsetof(struct net_device, ifindex));
 353                 else
 354                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 355                                             offsetof(struct net_device, type));
 356                 break;
 357
 358         case SKF_AD_OFF + SKF_AD_MARK:
 359                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 360                 insn += cnt - 1;
 361                 break;
 362
 363         case SKF_AD_OFF + SKF_AD_RXHASH:
 364                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
 365
 366                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 367                                     offsetof(struct sk_buff, hash));
 368                 break;
 369
 370         case SKF_AD_OFF + SKF_AD_QUEUE:
 371                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 372                 insn += cnt - 1;
 373                 break;
 374
 375         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 376                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 377                                          BPF_REG_A, BPF_REG_CTX, insn);
 378                 insn += cnt - 1;
 379                 break;
 380
 381         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 382                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 383                                          BPF_REG_A, BPF_REG_CTX, insn);
 384                 insn += cnt - 1;
 385                 break;
 386
 387         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 388                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
 389
 390                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 391                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 392                                       offsetof(struct sk_buff, vlan_proto));
 393                 /* A = ntohs(A) [emitting a nop or swap16] */
 394                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 395                 break;
 396
 397         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 398         case SKF_AD_OFF + SKF_AD_NLATTR:
 399         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 400         case SKF_AD_OFF + SKF_AD_CPU:
 401         case SKF_AD_OFF + SKF_AD_RANDOM:
 402                 /* arg1 = CTX */
 403                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 404                 /* arg2 = A */
 405                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 406                 /* arg3 = X */
 407                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 408                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 409                 switch (fp->k) {
 410                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 411                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 412                         break;
 413                 case SKF_AD_OFF + SKF_AD_NLATTR:
 414                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 415                         break;
 416                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 417                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 418                         break;
 419                 case SKF_AD_OFF + SKF_AD_CPU:
 420                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 421                         break;
 422                 case SKF_AD_OFF + SKF_AD_RANDOM:
 423                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 424                         bpf_user_rnd_init_once();
 425                         break;
 426                 }
 427                 break;
 428
 429         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 430                 /* A ^= X */
 431                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 432                 break;
 433
 434         default:
 435                 /* This is just a dummy call to avoid letting the compiler
 436                  * evict __bpf_call_base() as an optimization. Placed here
 437                  * where no-one bothers.
 438                  */
 439                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 440                 return false;
 441         }
 442
 443         *insnp = insn;
 444         return true;
 445 }
 446
 447 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 448 {
 449         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 450         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 451         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 452                       BPF_SIZE(fp->code) == BPF_W;
 453         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 454         const int ip_align = NET_IP_ALIGN;
 455         struct bpf_insn *insn = *insnp;
 456         int offset = fp->k;
 457
 458         if (!indirect &&
 459             ((unaligned_ok && offset >= 0) ||
 460              (!unaligned_ok && offset >= 0 &&
 461               offset + ip_align >= 0 &&
 462               offset + ip_align % size == 0))) {
 463                 bool ldx_off_ok = offset <= S16_MAX;
 464
 465                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 466                 if (offset)
 467                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 468                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 469                                       size, 2 + endian + (!ldx_off_ok * 2));
 470                 if (ldx_off_ok) {
 471                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 472                                               BPF_REG_D, offset);
 473                 } else {
 474                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 475                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 476                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 477                                               BPF_REG_TMP, 0);
 478                 }
 479                 if (endian)
 480                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 481                 *insn++ = BPF_JMP_A(8);
 482         }
 483
 484         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 485         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 486         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 487         if (!indirect) {
 488                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 489         } else {
 490                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 491                 if (fp->k)
 492                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 493         }
 494
 495         switch (BPF_SIZE(fp->code)) {
 496         case BPF_B:
 497                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 498                 break;
 499         case BPF_H:
 500                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 501                 break;
 502         case BPF_W:
 503                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 504                 break;
 505         default:
 506                 return false;
 507         }
 508
 509         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 510         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 511         *insn   = BPF_EXIT_INSN();
 512
 513         *insnp = insn;
 514         return true;
 515 }
 516
 517 /**
 518  *      bpf_convert_filter - convert filter program
 519  *      @prog: the user passed filter program
 520  *      @len: the length of the user passed filter program
 521  *      @new_prog: allocated 'struct bpf_prog' or NULL
 522  *      @new_len: pointer to store length of converted program
 523  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 524  *
 525  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 526  * style extended BPF (eBPF).
 527  * Conversion workflow:
 528  *
 529  * 1) First pass for calculating the new program length:
 530  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 531  *
 532  * 2) 2nd pass to remap in two passes: 1st pass finds new
 533  *    jump offsets, 2nd pass remapping:
 534  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 535  */
 536 static int bpf_convert_filter(struct sock_filter *prog, int len,
 537                               struct bpf_prog *new_prog, int *new_len,
 538                               bool *seen_ld_abs)
 539 {
 540         int new_flen = 0, pass = 0, target, i, stack_off;
 541         struct bpf_insn *new_insn, *first_insn = NULL;
 542         struct sock_filter *fp;
 543         int *addrs = NULL;
 544         u8 bpf_src;
 545
 546         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 547         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 548
 549         if (len <= 0 || len > BPF_MAXINSNS)
 550                 return -EINVAL;
 551
 552         if (new_prog) {
 553                 first_insn = new_prog->insnsi;
 554                 addrs = kcalloc(len, sizeof(*addrs),
 555                                 GFP_KERNEL | __GFP_NOWARN);
 556                 if (!addrs)
 557                         return -ENOMEM;
 558         }
 559
 560 do_pass:
 561         new_insn = first_insn;
 562         fp = prog;
 563
 564         /* Classic BPF related prologue emission. */
 565         if (new_prog) {
 566                 /* Classic BPF expects A and X to be reset first. These need
 567                  * to be guaranteed to be the first two instructions.
 568                  */
 569                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 570                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 571
 572                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 573                  * In eBPF case it's done by the compiler, here we need to
 574                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 575                  */
 576                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 577                 if (*seen_ld_abs) {
 578                         /* For packet access in classic BPF, cache skb->data
 579                          * in callee-saved BPF R8 and skb->len - skb->data_len
 580                          * (headlen) in BPF R9. Since classic BPF is read-only
 581                          * on CTX, we only need to cache it once.
 582                          */
 583                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 584                                                   BPF_REG_D, BPF_REG_CTX,
 585                                                   offsetof(struct sk_buff, data));
 586                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 587                                                   offsetof(struct sk_buff, len));
 588                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 589                                                   offsetof(struct sk_buff, data_len));
 590                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 591                 }
 592         } else {
 593                 new_insn += 3;
 594         }
 595
 596         for (i = 0; i < len; fp++, i++) {
 597                 struct bpf_insn tmp_insns[32] = { };
 598                 struct bpf_insn *insn = tmp_insns;
 599
 600                 if (addrs)
 601                         addrs[i] = new_insn - first_insn;
 602
 603                 switch (fp->code) {
 604                 /* All arithmetic insns and skb loads map as-is. */
 605                 case BPF_ALU | BPF_ADD | BPF_X:
 606                 case BPF_ALU | BPF_ADD | BPF_K:
 607                 case BPF_ALU | BPF_SUB | BPF_X:
 608                 case BPF_ALU | BPF_SUB | BPF_K:
 609                 case BPF_ALU | BPF_AND | BPF_X:
 610                 case BPF_ALU | BPF_AND | BPF_K:
 611                 case BPF_ALU | BPF_OR | BPF_X:
 612                 case BPF_ALU | BPF_OR | BPF_K:
 613                 case BPF_ALU | BPF_LSH | BPF_X:
 614                 case BPF_ALU | BPF_LSH | BPF_K:
 615                 case BPF_ALU | BPF_RSH | BPF_X:
 616                 case BPF_ALU | BPF_RSH | BPF_K:
 617                 case BPF_ALU | BPF_XOR | BPF_X:
 618                 case BPF_ALU | BPF_XOR | BPF_K:
 619                 case BPF_ALU | BPF_MUL | BPF_X:
 620                 case BPF_ALU | BPF_MUL | BPF_K:
 621                 case BPF_ALU | BPF_DIV | BPF_X:
 622                 case BPF_ALU | BPF_DIV | BPF_K:
 623                 case BPF_ALU | BPF_MOD | BPF_X:
 624                 case BPF_ALU | BPF_MOD | BPF_K:
 625                 case BPF_ALU | BPF_NEG:
 626                 case BPF_LD | BPF_ABS | BPF_W:
 627                 case BPF_LD | BPF_ABS | BPF_H:
 628                 case BPF_LD | BPF_ABS | BPF_B:
 629                 case BPF_LD | BPF_IND | BPF_W:
 630                 case BPF_LD | BPF_IND | BPF_H:
 631                 case BPF_LD | BPF_IND | BPF_B:
 632                         /* Check for overloaded BPF extension and
 633                          * directly convert it if found, otherwise
 634                          * just move on with mapping.
 635                          */
 636                         if (BPF_CLASS(fp->code) == BPF_LD &&
 637                             BPF_MODE(fp->code) == BPF_ABS &&
 638                             convert_bpf_extensions(fp, &insn))
 639                                 break;
 640                         if (BPF_CLASS(fp->code) == BPF_LD &&
 641                             convert_bpf_ld_abs(fp, &insn)) {
 642                                 *seen_ld_abs = true;
 643                                 break;
 644                         }
 645
 646                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 647                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 648                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 649                                 /* Error with exception code on div/mod by 0.
 650                                  * For cBPF programs, this was always return 0.
 651                                  */
 652                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 653                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 654                                 *insn++ = BPF_EXIT_INSN();
 655                         }
 656
 657                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 658                         break;
 659
 660                 /* Jump transformation cannot use BPF block macros
 661                  * everywhere as offset calculation and target updates
 662                  * require a bit more work than the rest, i.e. jump
 663                  * opcodes map as-is, but offsets need adjustment.
 664                  */
 665
 666 #define BPF_EMIT_JMP                                                    \
 667         do {                                                            \
 668                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 669                 s32 off;                                                \
 670                                                                         \
 671                 if (target >= len || target < 0)                        \
 672                         goto err;                                       \
 673                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 674                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 675                 off -= insn - tmp_insns;                                \
 676                 /* Reject anything not fitting into insn->off. */       \
 677                 if (off < off_min || off > off_max)                     \
 678                         goto err;                                       \
 679                 insn->off = off;                                        \
 680         } while (0)
 681
 682                 case BPF_JMP | BPF_JA:
 683                         target = i + fp->k + 1;
 684                         insn->code = fp->code;
 685                         BPF_EMIT_JMP;
 686                         break;
 687
 688                 case BPF_JMP | BPF_JEQ | BPF_K:
 689                 case BPF_JMP | BPF_JEQ | BPF_X:
 690                 case BPF_JMP | BPF_JSET | BPF_K:
 691                 case BPF_JMP | BPF_JSET | BPF_X:
 692                 case BPF_JMP | BPF_JGT | BPF_K:
 693                 case BPF_JMP | BPF_JGT | BPF_X:
 694                 case BPF_JMP | BPF_JGE | BPF_K:
 695                 case BPF_JMP | BPF_JGE | BPF_X:
 696                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 697                                 /* BPF immediates are signed, zero extend
 698                                  * immediate into tmp register and use it
 699                                  * in compare insn.
 700                                  */
 701                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 702
 703                                 insn->dst_reg = BPF_REG_A;
 704                                 insn->src_reg = BPF_REG_TMP;
 705                                 bpf_src = BPF_X;
 706                         } else {
 707                                 insn->dst_reg = BPF_REG_A;
 708                                 insn->imm = fp->k;
 709                                 bpf_src = BPF_SRC(fp->code);
 710                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 711                         }
 712
 713                         /* Common case where 'jump_false' is next insn. */
 714                         if (fp->jf == 0) {
 715                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 716                                 target = i + fp->jt + 1;
 717                                 BPF_EMIT_JMP;
 718                                 break;
 719                         }
 720
 721                         /* Convert some jumps when 'jump_true' is next insn. */
 722                         if (fp->jt == 0) {
 723                                 switch (BPF_OP(fp->code)) {
 724                                 case BPF_JEQ:
 725                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 726                                         break;
 727                                 case BPF_JGT:
 728                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 729                                         break;
 730                                 case BPF_JGE:
 731                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 732                                         break;
 733                                 default:
 734                                         goto jmp_rest;
 735                                 }
 736
 737                                 target = i + fp->jf + 1;
 738                                 BPF_EMIT_JMP;
 739                                 break;
 740                         }
 741 jmp_rest:
 742                         /* Other jumps are mapped into two insns: Jxx and JA. */
 743                         target = i + fp->jt + 1;
 744                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 745                         BPF_EMIT_JMP;
 746                         insn++;
 747
 748                         insn->code = BPF_JMP | BPF_JA;
 749                         target = i + fp->jf + 1;
 750                         BPF_EMIT_JMP;
 751                         break;
 752
 753                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 754                 case BPF_LDX | BPF_MSH | BPF_B: {
 755                         struct sock_filter tmp = {
 756                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 757                                 .k      = fp->k,
 758                         };
 759
 760                         *seen_ld_abs = true;
 761
 762                         /* X = A */
 763                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 764                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 765                         convert_bpf_ld_abs(&tmp, &insn);
 766                         insn++;
 767                         /* A &= 0xf */
 768                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 769                         /* A <<= 2 */
 770                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 771                         /* tmp = X */
 772                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 773                         /* X = A */
 774                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 775                         /* A = tmp */
 776                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 777                         break;
 778                 }
 779                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 780                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 781                  */
 782                 case BPF_RET | BPF_A:
 783                 case BPF_RET | BPF_K:
 784                         if (BPF_RVAL(fp->code) == BPF_K)
 785                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 786                                                         0, fp->k);
 787                         *insn = BPF_EXIT_INSN();
 788                         break;
 789
 790                 /* Store to stack. */
 791                 case BPF_ST:
 792                 case BPF_STX:
 793                         stack_off = fp->k * 4  + 4;
 794                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 795                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 796                                             -stack_off);
 797                         /* check_load_and_stores() verifies that classic BPF can
 798                          * load from stack only after write, so tracking
 799                          * stack_depth for ST|STX insns is enough
 800                          */
 801                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 802                                 new_prog->aux->stack_depth = stack_off;
 803                         break;
 804
 805                 /* Load from stack. */
 806                 case BPF_LD | BPF_MEM:
 807                 case BPF_LDX | BPF_MEM:
 808                         stack_off = fp->k * 4  + 4;
 809                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 810                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 811                                             -stack_off);
 812                         break;
 813
 814                 /* A = K or X = K */
 815                 case BPF_LD | BPF_IMM:
 816                 case BPF_LDX | BPF_IMM:
 817                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 818                                               BPF_REG_A : BPF_REG_X, fp->k);
 819                         break;
 820
 821                 /* X = A */
 822                 case BPF_MISC | BPF_TAX:
 823                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 824                         break;
 825
 826                 /* A = X */
 827                 case BPF_MISC | BPF_TXA:
 828                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 829                         break;
 830
 831                 /* A = skb->len or X = skb->len */
 832                 case BPF_LD | BPF_W | BPF_LEN:
 833                 case BPF_LDX | BPF_W | BPF_LEN:
 834                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 835                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 836                                             offsetof(struct sk_buff, len));
 837                         break;
 838
 839                 /* Access seccomp_data fields. */
 840                 case BPF_LDX | BPF_ABS | BPF_W:
 841                         /* A = *(u32 *) (ctx + K) */
 842                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 843                         break;
 844
 845                 /* Unknown instruction. */
 846                 default:
 847                         goto err;
 848                 }
 849
 850                 insn++;
 851                 if (new_prog)
 852                         memcpy(new_insn, tmp_insns,
 853                                sizeof(*insn) * (insn - tmp_insns));
 854                 new_insn += insn - tmp_insns;
 855         }
 856
 857         if (!new_prog) {
 858                 /* Only calculating new length. */
 859                 *new_len = new_insn - first_insn;
 860                 if (*seen_ld_abs)
 861                         *new_len += 4; /* Prologue bits. */
 862                 return 0;
 863         }
 864
 865         pass++;
 866         if (new_flen != new_insn - first_insn) {
 867                 new_flen = new_insn - first_insn;
 868                 if (pass > 2)
 869                         goto err;
 870                 goto do_pass;
 871         }
 872
 873         kfree(addrs);
 874         BUG_ON(*new_len != new_flen);
 875         return 0;
 876 err:
 877         kfree(addrs);
 878         return -EINVAL;
 879 }
 880
 881 /* Security:
 882  *
 883  * As we dont want to clear mem[] array for each packet going through
 884  * __bpf_prog_run(), we check that filter loaded by user never try to read
 885  * a cell if not previously written, and we check all branches to be sure
 886  * a malicious user doesn't try to abuse us.
 887  */
 888 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 889 {
 890         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 891         int pc, ret = 0;
 892
 893         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 894
 895         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 896         if (!masks)
 897                 return -ENOMEM;
 898
 899         memset(masks, 0xff, flen * sizeof(*masks));
 900
 901         for (pc = 0; pc < flen; pc++) {
 902                 memvalid &= masks[pc];
 903
 904                 switch (filter[pc].code) {
 905                 case BPF_ST:
 906                 case BPF_STX:
 907                         memvalid |= (1 << filter[pc].k);
 908                         break;
 909                 case BPF_LD | BPF_MEM:
 910                 case BPF_LDX | BPF_MEM:
 911                         if (!(memvalid & (1 << filter[pc].k))) {
 912                                 ret = -EINVAL;
 913                                 goto error;
 914                         }
 915                         break;
 916                 case BPF_JMP | BPF_JA:
 917                         /* A jump must set masks on target */
 918                         masks[pc + 1 + filter[pc].k] &= memvalid;
 919                         memvalid = ~0;
 920                         break;
 921                 case BPF_JMP | BPF_JEQ | BPF_K:
 922                 case BPF_JMP | BPF_JEQ | BPF_X:
 923                 case BPF_JMP | BPF_JGE | BPF_K:
 924                 case BPF_JMP | BPF_JGE | BPF_X:
 925                 case BPF_JMP | BPF_JGT | BPF_K:
 926                 case BPF_JMP | BPF_JGT | BPF_X:
 927                 case BPF_JMP | BPF_JSET | BPF_K:
 928                 case BPF_JMP | BPF_JSET | BPF_X:
 929                         /* A jump must set masks on targets */
 930                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 931                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 932                         memvalid = ~0;
 933                         break;
 934                 }
 935         }
 936 error:
 937         kfree(masks);
 938         return ret;
 939 }
 940
 941 static bool chk_code_allowed(u16 code_to_probe)
 942 {
 943         static const bool codes[] = {
 944                 /* 32 bit ALU operations */
 945                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 946                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 947                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 948                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 949                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 950                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 951                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 952                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 953                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 954                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 955                 [BPF_ALU | BPF_AND | BPF_K] = true,
 956                 [BPF_ALU | BPF_AND | BPF_X] = true,
 957                 [BPF_ALU | BPF_OR | BPF_K] = true,
 958                 [BPF_ALU | BPF_OR | BPF_X] = true,
 959                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 960                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 961                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 962                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 963                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 964                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 965                 [BPF_ALU | BPF_NEG] = true,
 966                 /* Load instructions */
 967                 [BPF_LD | BPF_W | BPF_ABS] = true,
 968                 [BPF_LD | BPF_H | BPF_ABS] = true,
 969                 [BPF_LD | BPF_B | BPF_ABS] = true,
 970                 [BPF_LD | BPF_W | BPF_LEN] = true,
 971                 [BPF_LD | BPF_W | BPF_IND] = true,
 972                 [BPF_LD | BPF_H | BPF_IND] = true,
 973                 [BPF_LD | BPF_B | BPF_IND] = true,
 974                 [BPF_LD | BPF_IMM] = true,
 975                 [BPF_LD | BPF_MEM] = true,
 976                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 977                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 978                 [BPF_LDX | BPF_IMM] = true,
 979                 [BPF_LDX | BPF_MEM] = true,
 980                 /* Store instructions */
 981                 [BPF_ST] = true,
 982                 [BPF_STX] = true,
 983                 /* Misc instructions */
 984                 [BPF_MISC | BPF_TAX] = true,
 985                 [BPF_MISC | BPF_TXA] = true,
 986                 /* Return instructions */
 987                 [BPF_RET | BPF_K] = true,
 988                 [BPF_RET | BPF_A] = true,
 989                 /* Jump instructions */
 990                 [BPF_JMP | BPF_JA] = true,
 991                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
 992                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
 993                 [BPF_JMP | BPF_JGE | BPF_K] = true,
 994                 [BPF_JMP | BPF_JGE | BPF_X] = true,
 995                 [BPF_JMP | BPF_JGT | BPF_K] = true,
 996                 [BPF_JMP | BPF_JGT | BPF_X] = true,
 997                 [BPF_JMP | BPF_JSET | BPF_K] = true,
 998                 [BPF_JMP | BPF_JSET | BPF_X] = true,
 999         };
1000
1001         if (code_to_probe >= ARRAY_SIZE(codes))
1002                 return false;
1003
1004         return codes[code_to_probe];
1005 }
1006
1007 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1008                                 unsigned int flen)
1009 {
1010         if (filter == NULL)
1011                 return false;
1012         if (flen == 0 || flen > BPF_MAXINSNS)
1013                 return false;
1014
1015         return true;
1016 }
1017
1018 /**
1019  *      bpf_check_classic - verify socket filter code
1020  *      @filter: filter to verify
1021  *      @flen: length of filter
1022  *
1023  * Check the user's filter code. If we let some ugly
1024  * filter code slip through kaboom! The filter must contain
1025  * no references or jumps that are out of range, no illegal
1026  * instructions, and must end with a RET instruction.
1027  *
1028  * All jumps are forward as they are not signed.
1029  *
1030  * Returns 0 if the rule set is legal or -EINVAL if not.
1031  */
1032 static int bpf_check_classic(const struct sock_filter *filter,
1033                              unsigned int flen)
1034 {
1035         bool anc_found;
1036         int pc;
1037
1038         /* Check the filter code now */
1039         for (pc = 0; pc < flen; pc++) {
1040                 const struct sock_filter *ftest = &filter[pc];
1041
1042                 /* May we actually operate on this code? */
1043                 if (!chk_code_allowed(ftest->code))
1044                         return -EINVAL;
1045
1046                 /* Some instructions need special checks */
1047                 switch (ftest->code) {
1048                 case BPF_ALU | BPF_DIV | BPF_K:
1049                 case BPF_ALU | BPF_MOD | BPF_K:
1050                         /* Check for division by zero */
1051                         if (ftest->k == 0)
1052                                 return -EINVAL;
1053                         break;
1054                 case BPF_ALU | BPF_LSH | BPF_K:
1055                 case BPF_ALU | BPF_RSH | BPF_K:
1056                         if (ftest->k >= 32)
1057                                 return -EINVAL;
1058                         break;
1059                 case BPF_LD | BPF_MEM:
1060                 case BPF_LDX | BPF_MEM:
1061                 case BPF_ST:
1062                 case BPF_STX:
1063                         /* Check for invalid memory addresses */
1064                         if (ftest->k >= BPF_MEMWORDS)
1065                                 return -EINVAL;
1066                         break;
1067                 case BPF_JMP | BPF_JA:
1068                         /* Note, the large ftest->k might cause loops.
1069                          * Compare this with conditional jumps below,
1070                          * where offsets are limited. --ANK (981016)
1071                          */
1072                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1073                                 return -EINVAL;
1074                         break;
1075                 case BPF_JMP | BPF_JEQ | BPF_K:
1076                 case BPF_JMP | BPF_JEQ | BPF_X:
1077                 case BPF_JMP | BPF_JGE | BPF_K:
1078                 case BPF_JMP | BPF_JGE | BPF_X:
1079                 case BPF_JMP | BPF_JGT | BPF_K:
1080                 case BPF_JMP | BPF_JGT | BPF_X:
1081                 case BPF_JMP | BPF_JSET | BPF_K:
1082                 case BPF_JMP | BPF_JSET | BPF_X:
1083                         /* Both conditionals must be safe */
1084                         if (pc + ftest->jt + 1 >= flen ||
1085                             pc + ftest->jf + 1 >= flen)
1086                                 return -EINVAL;
1087                         break;
1088                 case BPF_LD | BPF_W | BPF_ABS:
1089                 case BPF_LD | BPF_H | BPF_ABS:
1090                 case BPF_LD | BPF_B | BPF_ABS:
1091                         anc_found = false;
1092                         if (bpf_anc_helper(ftest) & BPF_ANC)
1093                                 anc_found = true;
1094                         /* Ancillary operation unknown or unsupported */
1095                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1096                                 return -EINVAL;
1097                 }
1098         }
1099
1100         /* Last instruction must be a RET code */
1101         switch (filter[flen - 1].code) {
1102         case BPF_RET | BPF_K:
1103         case BPF_RET | BPF_A:
1104                 return check_load_and_stores(filter, flen);
1105         }
1106
1107         return -EINVAL;
1108 }
1109
1110 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1111                                       const struct sock_fprog *fprog)
1112 {
1113         unsigned int fsize = bpf_classic_proglen(fprog);
1114         struct sock_fprog_kern *fkprog;
1115
1116         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1117         if (!fp->orig_prog)
1118                 return -ENOMEM;
1119
1120         fkprog = fp->orig_prog;
1121         fkprog->len = fprog->len;
1122
1123         fkprog->filter = kmemdup(fp->insns, fsize,
1124                                  GFP_KERNEL | __GFP_NOWARN);
1125         if (!fkprog->filter) {
1126                 kfree(fp->orig_prog);
1127                 return -ENOMEM;
1128         }
1129
1130         return 0;
1131 }
1132
1133 static void bpf_release_orig_filter(struct bpf_prog *fp)
1134 {
1135         struct sock_fprog_kern *fprog = fp->orig_prog;
1136
1137         if (fprog) {
1138                 kfree(fprog->filter);
1139                 kfree(fprog);
1140         }
1141 }
1142
1143 static void __bpf_prog_release(struct bpf_prog *prog)
1144 {
1145         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1146                 bpf_prog_put(prog);
1147         } else {
1148                 bpf_release_orig_filter(prog);
1149                 bpf_prog_free(prog);
1150         }
1151 }
1152
1153 static void __sk_filter_release(struct sk_filter *fp)
1154 {
1155         __bpf_prog_release(fp->prog);
1156         kfree(fp);
1157 }
1158
1159 /**
1160  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1161  *      @rcu: rcu_head that contains the sk_filter to free
1162  */
1163 static void sk_filter_release_rcu(struct rcu_head *rcu)
1164 {
1165         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1166
1167         __sk_filter_release(fp);
1168 }
1169
1170 /**
1171  *      sk_filter_release - release a socket filter
1172  *      @fp: filter to remove
1173  *
1174  *      Remove a filter from a socket and release its resources.
1175  */
1176 static void sk_filter_release(struct sk_filter *fp)
1177 {
1178         if (refcount_dec_and_test(&fp->refcnt))
1179                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1180 }
1181
1182 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1183 {
1184         u32 filter_size = bpf_prog_size(fp->prog->len);
1185
1186         atomic_sub(filter_size, &sk->sk_omem_alloc);
1187         sk_filter_release(fp);
1188 }
1189
1190 /* try to charge the socket memory if there is space available
1191  * return true on success
1192  */
1193 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1194 {
1195         u32 filter_size = bpf_prog_size(fp->prog->len);
1196
1197         /* same check as in sock_kmalloc() */
1198         if (filter_size <= sysctl_optmem_max &&
1199             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1200                 atomic_add(filter_size, &sk->sk_omem_alloc);
1201                 return true;
1202         }
1203         return false;
1204 }
1205
1206 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1207 {
1208         if (!refcount_inc_not_zero(&fp->refcnt))
1209                 return false;
1210
1211         if (!__sk_filter_charge(sk, fp)) {
1212                 sk_filter_release(fp);
1213                 return false;
1214         }
1215         return true;
1216 }
1217
1218 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1219 {
1220         struct sock_filter *old_prog;
1221         struct bpf_prog *old_fp;
1222         int err, new_len, old_len = fp->len;
1223         bool seen_ld_abs = false;
1224
1225         /* We are free to overwrite insns et al right here as it
1226          * won't be used at this point in time anymore internally
1227          * after the migration to the internal BPF instruction
1228          * representation.
1229          */
1230         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1231                      sizeof(struct bpf_insn));
1232
1233         /* Conversion cannot happen on overlapping memory areas,
1234          * so we need to keep the user BPF around until the 2nd
1235          * pass. At this time, the user BPF is stored in fp->insns.
1236          */
1237         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1238                            GFP_KERNEL | __GFP_NOWARN);
1239         if (!old_prog) {
1240                 err = -ENOMEM;
1241                 goto out_err;
1242         }
1243
1244         /* 1st pass: calculate the new program length. */
1245         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1246                                  &seen_ld_abs);
1247         if (err)
1248                 goto out_err_free;
1249
1250         /* Expand fp for appending the new filter representation. */
1251         old_fp = fp;
1252         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1253         if (!fp) {
1254                 /* The old_fp is still around in case we couldn't
1255                  * allocate new memory, so uncharge on that one.
1256                  */
1257                 fp = old_fp;
1258                 err = -ENOMEM;
1259                 goto out_err_free;
1260         }
1261
1262         fp->len = new_len;
1263
1264         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1265         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1266                                  &seen_ld_abs);
1267         if (err)
1268                 /* 2nd bpf_convert_filter() can fail only if it fails
1269                  * to allocate memory, remapping must succeed. Note,
1270                  * that at this time old_fp has already been released
1271                  * by krealloc().
1272                  */
1273                 goto out_err_free;
1274
1275         fp = bpf_prog_select_runtime(fp, &err);
1276         if (err)
1277                 goto out_err_free;
1278
1279         kfree(old_prog);
1280         return fp;
1281
1282 out_err_free:
1283         kfree(old_prog);
1284 out_err:
1285         __bpf_prog_release(fp);
1286         return ERR_PTR(err);
1287 }
1288
1289 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1290                                            bpf_aux_classic_check_t trans)
1291 {
1292         int err;
1293
1294         fp->bpf_func = NULL;
1295         fp->jited = 0;
1296
1297         err = bpf_check_classic(fp->insns, fp->len);
1298         if (err) {
1299                 __bpf_prog_release(fp);
1300                 return ERR_PTR(err);
1301         }
1302
1303         /* There might be additional checks and transformations
1304          * needed on classic filters, f.e. in case of seccomp.
1305          */
1306         if (trans) {
1307                 err = trans(fp->insns, fp->len);
1308                 if (err) {
1309                         __bpf_prog_release(fp);
1310                         return ERR_PTR(err);
1311                 }
1312         }
1313
1314         /* Probe if we can JIT compile the filter and if so, do
1315          * the compilation of the filter.
1316          */
1317         bpf_jit_compile(fp);
1318
1319         /* JIT compiler couldn't process this filter, so do the
1320          * internal BPF translation for the optimized interpreter.
1321          */
1322         if (!fp->jited)
1323                 fp = bpf_migrate_filter(fp);
1324
1325         return fp;
1326 }
1327
1328 /**
1329  *      bpf_prog_create - create an unattached filter
1330  *      @pfp: the unattached filter that is created
1331  *      @fprog: the filter program
1332  *
1333  * Create a filter independent of any socket. We first run some
1334  * sanity checks on it to make sure it does not explode on us later.
1335  * If an error occurs or there is insufficient memory for the filter
1336  * a negative errno code is returned. On success the return is zero.
1337  */
1338 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1339 {
1340         unsigned int fsize = bpf_classic_proglen(fprog);
1341         struct bpf_prog *fp;
1342
1343         /* Make sure new filter is there and in the right amounts. */
1344         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1345                 return -EINVAL;
1346
1347         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1348         if (!fp)
1349                 return -ENOMEM;
1350
1351         memcpy(fp->insns, fprog->filter, fsize);
1352
1353         fp->len = fprog->len;
1354         /* Since unattached filters are not copied back to user
1355          * space through sk_get_filter(), we do not need to hold
1356          * a copy here, and can spare us the work.
1357          */
1358         fp->orig_prog = NULL;
1359
1360         /* bpf_prepare_filter() already takes care of freeing
1361          * memory in case something goes wrong.
1362          */
1363         fp = bpf_prepare_filter(fp, NULL);
1364         if (IS_ERR(fp))
1365                 return PTR_ERR(fp);
1366
1367         *pfp = fp;
1368         return 0;
1369 }
1370 EXPORT_SYMBOL_GPL(bpf_prog_create);
1371
1372 /**
1373  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1374  *      @pfp: the unattached filter that is created
1375  *      @fprog: the filter program
1376  *      @trans: post-classic verifier transformation handler
1377  *      @save_orig: save classic BPF program
1378  *
1379  * This function effectively does the same as bpf_prog_create(), only
1380  * that it builds up its insns buffer from user space provided buffer.
1381  * It also allows for passing a bpf_aux_classic_check_t handler.
1382  */
1383 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1384                               bpf_aux_classic_check_t trans, bool save_orig)
1385 {
1386         unsigned int fsize = bpf_classic_proglen(fprog);
1387         struct bpf_prog *fp;
1388         int err;
1389
1390         /* Make sure new filter is there and in the right amounts. */
1391         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1392                 return -EINVAL;
1393
1394         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1395         if (!fp)
1396                 return -ENOMEM;
1397
1398         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1399                 __bpf_prog_free(fp);
1400                 return -EFAULT;
1401         }
1402
1403         fp->len = fprog->len;
1404         fp->orig_prog = NULL;
1405
1406         if (save_orig) {
1407                 err = bpf_prog_store_orig_filter(fp, fprog);
1408                 if (err) {
1409                         __bpf_prog_free(fp);
1410                         return -ENOMEM;
1411                 }
1412         }
1413
1414         /* bpf_prepare_filter() already takes care of freeing
1415          * memory in case something goes wrong.
1416          */
1417         fp = bpf_prepare_filter(fp, trans);
1418         if (IS_ERR(fp))
1419                 return PTR_ERR(fp);
1420
1421         *pfp = fp;
1422         return 0;
1423 }
1424 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1425
1426 void bpf_prog_destroy(struct bpf_prog *fp)
1427 {
1428         __bpf_prog_release(fp);
1429 }
1430 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1431
1432 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1433 {
1434         struct sk_filter *fp, *old_fp;
1435
1436         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1437         if (!fp)
1438                 return -ENOMEM;
1439
1440         fp->prog = prog;
1441
1442         if (!__sk_filter_charge(sk, fp)) {
1443                 kfree(fp);
1444                 return -ENOMEM;
1445         }
1446         refcount_set(&fp->refcnt, 1);
1447
1448         old_fp = rcu_dereference_protected(sk->sk_filter,
1449                                            lockdep_sock_is_held(sk));
1450         rcu_assign_pointer(sk->sk_filter, fp);
1451
1452         if (old_fp)
1453                 sk_filter_uncharge(sk, old_fp);
1454
1455         return 0;
1456 }
1457
1458 static
1459 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1460 {
1461         unsigned int fsize = bpf_classic_proglen(fprog);
1462         struct bpf_prog *prog;
1463         int err;
1464
1465         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1466                 return ERR_PTR(-EPERM);
1467
1468         /* Make sure new filter is there and in the right amounts. */
1469         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1470                 return ERR_PTR(-EINVAL);
1471
1472         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1473         if (!prog)
1474                 return ERR_PTR(-ENOMEM);
1475
1476         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1477                 __bpf_prog_free(prog);
1478                 return ERR_PTR(-EFAULT);
1479         }
1480
1481         prog->len = fprog->len;
1482
1483         err = bpf_prog_store_orig_filter(prog, fprog);
1484         if (err) {
1485                 __bpf_prog_free(prog);
1486                 return ERR_PTR(-ENOMEM);
1487         }
1488
1489         /* bpf_prepare_filter() already takes care of freeing
1490          * memory in case something goes wrong.
1491          */
1492         return bpf_prepare_filter(prog, NULL);
1493 }
1494
1495 /**
1496  *      sk_attach_filter - attach a socket filter
1497  *      @fprog: the filter program
1498  *      @sk: the socket to use
1499  *
1500  * Attach the user's filter code. We first run some sanity checks on
1501  * it to make sure it does not explode on us later. If an error
1502  * occurs or there is insufficient memory for the filter a negative
1503  * errno code is returned. On success the return is zero.
1504  */
1505 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1506 {
1507         struct bpf_prog *prog = __get_filter(fprog, sk);
1508         int err;
1509
1510         if (IS_ERR(prog))
1511                 return PTR_ERR(prog);
1512
1513         err = __sk_attach_prog(prog, sk);
1514         if (err < 0) {
1515                 __bpf_prog_release(prog);
1516                 return err;
1517         }
1518
1519         return 0;
1520 }
1521 EXPORT_SYMBOL_GPL(sk_attach_filter);
1522
1523 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1524 {
1525         struct bpf_prog *prog = __get_filter(fprog, sk);
1526         int err;
1527
1528         if (IS_ERR(prog))
1529                 return PTR_ERR(prog);
1530
1531         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1532                 err = -ENOMEM;
1533         else
1534                 err = reuseport_attach_prog(sk, prog);
1535
1536         if (err)
1537                 __bpf_prog_release(prog);
1538
1539         return err;
1540 }
1541
1542 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1543 {
1544         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1545                 return ERR_PTR(-EPERM);
1546
1547         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1548 }
1549
1550 int sk_attach_bpf(u32 ufd, struct sock *sk)
1551 {
1552         struct bpf_prog *prog = __get_bpf(ufd, sk);
1553         int err;
1554
1555         if (IS_ERR(prog))
1556                 return PTR_ERR(prog);
1557
1558         err = __sk_attach_prog(prog, sk);
1559         if (err < 0) {
1560                 bpf_prog_put(prog);
1561                 return err;
1562         }
1563
1564         return 0;
1565 }
1566
1567 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1568 {
1569         struct bpf_prog *prog;
1570         int err;
1571
1572         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1573                 return -EPERM;
1574
1575         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1576         if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1577                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1578         if (IS_ERR(prog))
1579                 return PTR_ERR(prog);
1580
1581         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1582                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1583                  * bpf prog (e.g. sockmap).  It depends on the
1584                  * limitation imposed by bpf_prog_load().
1585                  * Hence, sysctl_optmem_max is not checked.
1586                  */
1587                 if ((sk->sk_type != SOCK_STREAM &&
1588                      sk->sk_type != SOCK_DGRAM) ||
1589                     (sk->sk_protocol != IPPROTO_UDP &&
1590                      sk->sk_protocol != IPPROTO_TCP) ||
1591                     (sk->sk_family != AF_INET &&
1592                      sk->sk_family != AF_INET6)) {
1593                         err = -ENOTSUPP;
1594                         goto err_prog_put;
1595                 }
1596         } else {
1597                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1598                 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1599                         err = -ENOMEM;
1600                         goto err_prog_put;
1601                 }
1602         }
1603
1604         err = reuseport_attach_prog(sk, prog);
1605 err_prog_put:
1606         if (err)
1607                 bpf_prog_put(prog);
1608
1609         return err;
1610 }
1611
1612 void sk_reuseport_prog_free(struct bpf_prog *prog)
1613 {
1614         if (!prog)
1615                 return;
1616
1617         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1618                 bpf_prog_put(prog);
1619         else
1620                 bpf_prog_destroy(prog);
1621 }
1622
1623 struct bpf_scratchpad {
1624         union {
1625                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1626                 u8     buff[MAX_BPF_STACK];
1627         };
1628 };
1629
1630 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1631
1632 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1633                                           unsigned int write_len)
1634 {
1635         return skb_ensure_writable(skb, write_len);
1636 }
1637
1638 static inline int bpf_try_make_writable(struct sk_buff *skb,
1639                                         unsigned int write_len)
1640 {
1641         int err = __bpf_try_make_writable(skb, write_len);
1642
1643         bpf_compute_data_pointers(skb);
1644         return err;
1645 }
1646
1647 static int bpf_try_make_head_writable(struct sk_buff *skb)
1648 {
1649         return bpf_try_make_writable(skb, skb_headlen(skb));
1650 }
1651
1652 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1653 {
1654         if (skb_at_tc_ingress(skb))
1655                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1656 }
1657
1658 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1659 {
1660         if (skb_at_tc_ingress(skb))
1661                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1662 }
1663
1664 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1665            const void *, from, u32, len, u64, flags)
1666 {
1667         void *ptr;
1668
1669         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1670                 return -EINVAL;
1671         if (unlikely(offset > 0xffff))
1672                 return -EFAULT;
1673         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1674                 return -EFAULT;
1675
1676         ptr = skb->data + offset;
1677         if (flags & BPF_F_RECOMPUTE_CSUM)
1678                 __skb_postpull_rcsum(skb, ptr, len, offset);
1679
1680         memcpy(ptr, from, len);
1681
1682         if (flags & BPF_F_RECOMPUTE_CSUM)
1683                 __skb_postpush_rcsum(skb, ptr, len, offset);
1684         if (flags & BPF_F_INVALIDATE_HASH)
1685                 skb_clear_hash(skb);
1686
1687         return 0;
1688 }
1689
1690 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1691         .func           = bpf_skb_store_bytes,
1692         .gpl_only       = false,
1693         .ret_type       = RET_INTEGER,
1694         .arg1_type      = ARG_PTR_TO_CTX,
1695         .arg2_type      = ARG_ANYTHING,
1696         .arg3_type      = ARG_PTR_TO_MEM,
1697         .arg4_type      = ARG_CONST_SIZE,
1698         .arg5_type      = ARG_ANYTHING,
1699 };
1700
1701 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1702            void *, to, u32, len)
1703 {
1704         void *ptr;
1705
1706         if (unlikely(offset > 0xffff))
1707                 goto err_clear;
1708
1709         ptr = skb_header_pointer(skb, offset, len, to);
1710         if (unlikely(!ptr))
1711                 goto err_clear;
1712         if (ptr != to)
1713                 memcpy(to, ptr, len);
1714
1715         return 0;
1716 err_clear:
1717         memset(to, 0, len);
1718         return -EFAULT;
1719 }
1720
1721 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1722         .func           = bpf_skb_load_bytes,
1723         .gpl_only       = false,
1724         .ret_type       = RET_INTEGER,
1725         .arg1_type      = ARG_PTR_TO_CTX,
1726         .arg2_type      = ARG_ANYTHING,
1727         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1728         .arg4_type      = ARG_CONST_SIZE,
1729 };
1730
1731 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1732            u32, offset, void *, to, u32, len, u32, start_header)
1733 {
1734         u8 *end = skb_tail_pointer(skb);
1735         u8 *net = skb_network_header(skb);
1736         u8 *mac = skb_mac_header(skb);
1737         u8 *ptr;
1738
1739         if (unlikely(offset > 0xffff || len > (end - mac)))
1740                 goto err_clear;
1741
1742         switch (start_header) {
1743         case BPF_HDR_START_MAC:
1744                 ptr = mac + offset;
1745                 break;
1746         case BPF_HDR_START_NET:
1747                 ptr = net + offset;
1748                 break;
1749         default:
1750                 goto err_clear;
1751         }
1752
1753         if (likely(ptr >= mac && ptr + len <= end)) {
1754                 memcpy(to, ptr, len);
1755                 return 0;
1756         }
1757
1758 err_clear:
1759         memset(to, 0, len);
1760         return -EFAULT;
1761 }
1762
1763 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1764         .func           = bpf_skb_load_bytes_relative,
1765         .gpl_only       = false,
1766         .ret_type       = RET_INTEGER,
1767         .arg1_type      = ARG_PTR_TO_CTX,
1768         .arg2_type      = ARG_ANYTHING,
1769         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1770         .arg4_type      = ARG_CONST_SIZE,
1771         .arg5_type      = ARG_ANYTHING,
1772 };
1773
1774 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1775 {
1776         /* Idea is the following: should the needed direct read/write
1777          * test fail during runtime, we can pull in more data and redo
1778          * again, since implicitly, we invalidate previous checks here.
1779          *
1780          * Or, since we know how much we need to make read/writeable,
1781          * this can be done once at the program beginning for direct
1782          * access case. By this we overcome limitations of only current
1783          * headroom being accessible.
1784          */
1785         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1786 }
1787
1788 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1789         .func           = bpf_skb_pull_data,
1790         .gpl_only       = false,
1791         .ret_type       = RET_INTEGER,
1792         .arg1_type      = ARG_PTR_TO_CTX,
1793         .arg2_type      = ARG_ANYTHING,
1794 };
1795
1796 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1797                                            unsigned int write_len)
1798 {
1799         int err = __bpf_try_make_writable(skb, write_len);
1800
1801         bpf_compute_data_end_sk_skb(skb);
1802         return err;
1803 }
1804
1805 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1806 {
1807         /* Idea is the following: should the needed direct read/write
1808          * test fail during runtime, we can pull in more data and redo
1809          * again, since implicitly, we invalidate previous checks here.
1810          *
1811          * Or, since we know how much we need to make read/writeable,
1812          * this can be done once at the program beginning for direct
1813          * access case. By this we overcome limitations of only current
1814          * headroom being accessible.
1815          */
1816         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1817 }
1818
1819 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1820         .func           = sk_skb_pull_data,
1821         .gpl_only       = false,
1822         .ret_type       = RET_INTEGER,
1823         .arg1_type      = ARG_PTR_TO_CTX,
1824         .arg2_type      = ARG_ANYTHING,
1825 };
1826
1827 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1828            u64, from, u64, to, u64, flags)
1829 {
1830         __sum16 *ptr;
1831
1832         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1833                 return -EINVAL;
1834         if (unlikely(offset > 0xffff || offset & 1))
1835                 return -EFAULT;
1836         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1837                 return -EFAULT;
1838
1839         ptr = (__sum16 *)(skb->data + offset);
1840         switch (flags & BPF_F_HDR_FIELD_MASK) {
1841         case 0:
1842                 if (unlikely(from != 0))
1843                         return -EINVAL;
1844
1845                 csum_replace_by_diff(ptr, to);
1846                 break;
1847         case 2:
1848                 csum_replace2(ptr, from, to);
1849                 break;
1850         case 4:
1851                 csum_replace4(ptr, from, to);
1852                 break;
1853         default:
1854                 return -EINVAL;
1855         }
1856
1857         return 0;
1858 }
1859
1860 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1861         .func           = bpf_l3_csum_replace,
1862         .gpl_only       = false,
1863         .ret_type       = RET_INTEGER,
1864         .arg1_type      = ARG_PTR_TO_CTX,
1865         .arg2_type      = ARG_ANYTHING,
1866         .arg3_type      = ARG_ANYTHING,
1867         .arg4_type      = ARG_ANYTHING,
1868         .arg5_type      = ARG_ANYTHING,
1869 };
1870
1871 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1872            u64, from, u64, to, u64, flags)
1873 {
1874         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1875         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1876         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1877         __sum16 *ptr;
1878
1879         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1880                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1881                 return -EINVAL;
1882         if (unlikely(offset > 0xffff || offset & 1))
1883                 return -EFAULT;
1884         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1885                 return -EFAULT;
1886
1887         ptr = (__sum16 *)(skb->data + offset);
1888         if (is_mmzero && !do_mforce && !*ptr)
1889                 return 0;
1890
1891         switch (flags & BPF_F_HDR_FIELD_MASK) {
1892         case 0:
1893                 if (unlikely(from != 0))
1894                         return -EINVAL;
1895
1896                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1897                 break;
1898         case 2:
1899                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1900                 break;
1901         case 4:
1902                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1903                 break;
1904         default:
1905                 return -EINVAL;
1906         }
1907
1908         if (is_mmzero && !*ptr)
1909                 *ptr = CSUM_MANGLED_0;
1910         return 0;
1911 }
1912
1913 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1914         .func           = bpf_l4_csum_replace,
1915         .gpl_only       = false,
1916         .ret_type       = RET_INTEGER,
1917         .arg1_type      = ARG_PTR_TO_CTX,
1918         .arg2_type      = ARG_ANYTHING,
1919         .arg3_type      = ARG_ANYTHING,
1920         .arg4_type      = ARG_ANYTHING,
1921         .arg5_type      = ARG_ANYTHING,
1922 };
1923
1924 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1925            __be32 *, to, u32, to_size, __wsum, seed)
1926 {
1927         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1928         u32 diff_size = from_size + to_size;
1929         int i, j = 0;
1930
1931         /* This is quite flexible, some examples:
1932          *
1933          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1934          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1935          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
1936          *
1937          * Even for diffing, from_size and to_size don't need to be equal.
1938          */
1939         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1940                      diff_size > sizeof(sp->diff)))
1941                 return -EINVAL;
1942
1943         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1944                 sp->diff[j] = ~from[i];
1945         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
1946                 sp->diff[j] = to[i];
1947
1948         return csum_partial(sp->diff, diff_size, seed);
1949 }
1950
1951 static const struct bpf_func_proto bpf_csum_diff_proto = {
1952         .func           = bpf_csum_diff,
1953         .gpl_only       = false,
1954         .pkt_access     = true,
1955         .ret_type       = RET_INTEGER,
1956         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
1957         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1958         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
1959         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
1960         .arg5_type      = ARG_ANYTHING,
1961 };
1962
1963 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1964 {
1965         /* The interface is to be used in combination with bpf_csum_diff()
1966          * for direct packet writes. csum rotation for alignment as well
1967          * as emulating csum_sub() can be done from the eBPF program.
1968          */
1969         if (skb->ip_summed == CHECKSUM_COMPLETE)
1970                 return (skb->csum = csum_add(skb->csum, csum));
1971
1972         return -ENOTSUPP;
1973 }
1974
1975 static const struct bpf_func_proto bpf_csum_update_proto = {
1976         .func           = bpf_csum_update,
1977         .gpl_only       = false,
1978         .ret_type       = RET_INTEGER,
1979         .arg1_type      = ARG_PTR_TO_CTX,
1980         .arg2_type      = ARG_ANYTHING,
1981 };
1982
1983 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1984 {
1985         return dev_forward_skb(dev, skb);
1986 }
1987
1988 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1989                                       struct sk_buff *skb)
1990 {
1991         int ret = ____dev_forward_skb(dev, skb);
1992
1993         if (likely(!ret)) {
1994                 skb->dev = dev;
1995                 ret = netif_rx(skb);
1996         }
1997
1998         return ret;
1999 }
2000
2001 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2002 {
2003         int ret;
2004
2005         if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
2006                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2007                 kfree_skb(skb);
2008                 return -ENETDOWN;
2009         }
2010
2011         skb->dev = dev;
2012
2013         __this_cpu_inc(xmit_recursion);
2014         ret = dev_queue_xmit(skb);
2015         __this_cpu_dec(xmit_recursion);
2016
2017         return ret;
2018 }
2019
2020 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2021                                  u32 flags)
2022 {
2023         unsigned int mlen = skb_network_offset(skb);
2024
2025         if (mlen) {
2026                 __skb_pull(skb, mlen);
2027
2028                 /* At ingress, the mac header has already been pulled once.
2029                  * At egress, skb_pospull_rcsum has to be done in case that
2030                  * the skb is originated from ingress (i.e. a forwarded skb)
2031                  * to ensure that rcsum starts at net header.
2032                  */
2033                 if (!skb_at_tc_ingress(skb))
2034                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2035         }
2036         skb_pop_mac_header(skb);
2037         skb_reset_mac_len(skb);
2038         return flags & BPF_F_INGRESS ?
2039                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2040 }
2041
2042 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2043                                  u32 flags)
2044 {
2045         /* Verify that a link layer header is carried */
2046         if (unlikely(skb->mac_header >= skb->network_header)) {
2047                 kfree_skb(skb);
2048                 return -ERANGE;
2049         }
2050
2051         bpf_push_mac_rcsum(skb);
2052         return flags & BPF_F_INGRESS ?
2053                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2054 }
2055
2056 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2057                           u32 flags)
2058 {
2059         if (dev_is_mac_header_xmit(dev))
2060                 return __bpf_redirect_common(skb, dev, flags);
2061         else
2062                 return __bpf_redirect_no_mac(skb, dev, flags);
2063 }
2064
2065 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2066 {
2067         struct net_device *dev;
2068         struct sk_buff *clone;
2069         int ret;
2070
2071         if (unlikely(flags & ~(BPF_F_INGRESS)))
2072                 return -EINVAL;
2073
2074         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2075         if (unlikely(!dev))
2076                 return -EINVAL;
2077
2078         clone = skb_clone(skb, GFP_ATOMIC);
2079         if (unlikely(!clone))
2080                 return -ENOMEM;
2081
2082         /* For direct write, we need to keep the invariant that the skbs
2083          * we're dealing with need to be uncloned. Should uncloning fail
2084          * here, we need to free the just generated clone to unclone once
2085          * again.
2086          */
2087         ret = bpf_try_make_head_writable(skb);
2088         if (unlikely(ret)) {
2089                 kfree_skb(clone);
2090                 return -ENOMEM;
2091         }
2092
2093         return __bpf_redirect(clone, dev, flags);
2094 }
2095
2096 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2097         .func           = bpf_clone_redirect,
2098         .gpl_only       = false,
2099         .ret_type       = RET_INTEGER,
2100         .arg1_type      = ARG_PTR_TO_CTX,
2101         .arg2_type      = ARG_ANYTHING,
2102         .arg3_type      = ARG_ANYTHING,
2103 };
2104
2105 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2106 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2107
2108 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2109 {
2110         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2111
2112         if (unlikely(flags & ~(BPF_F_INGRESS)))
2113                 return TC_ACT_SHOT;
2114
2115         ri->ifindex = ifindex;
2116         ri->flags = flags;
2117
2118         return TC_ACT_REDIRECT;
2119 }
2120
2121 int skb_do_redirect(struct sk_buff *skb)
2122 {
2123         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2124         struct net_device *dev;
2125
2126         dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2127         ri->ifindex = 0;
2128         if (unlikely(!dev)) {
2129                 kfree_skb(skb);
2130                 return -EINVAL;
2131         }
2132
2133         return __bpf_redirect(skb, dev, ri->flags);
2134 }
2135
2136 static const struct bpf_func_proto bpf_redirect_proto = {
2137         .func           = bpf_redirect,
2138         .gpl_only       = false,
2139         .ret_type       = RET_INTEGER,
2140         .arg1_type      = ARG_ANYTHING,
2141         .arg2_type      = ARG_ANYTHING,
2142 };
2143
2144 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2145 {
2146         msg->apply_bytes = bytes;
2147         return 0;
2148 }
2149
2150 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2151         .func           = bpf_msg_apply_bytes,
2152         .gpl_only       = false,
2153         .ret_type       = RET_INTEGER,
2154         .arg1_type      = ARG_PTR_TO_CTX,
2155         .arg2_type      = ARG_ANYTHING,
2156 };
2157
2158 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2159 {
2160         msg->cork_bytes = bytes;
2161         return 0;
2162 }
2163
2164 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2165         .func           = bpf_msg_cork_bytes,
2166         .gpl_only       = false,
2167         .ret_type       = RET_INTEGER,
2168         .arg1_type      = ARG_PTR_TO_CTX,
2169         .arg2_type      = ARG_ANYTHING,
2170 };
2171
2172 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2173            u32, end, u64, flags)
2174 {
2175         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2176         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2177         struct scatterlist *sge;
2178         u8 *raw, *to, *from;
2179         struct page *page;
2180
2181         if (unlikely(flags || end <= start))
2182                 return -EINVAL;
2183
2184         /* First find the starting scatterlist element */
2185         i = msg->sg.start;
2186         do {
2187                 len = sk_msg_elem(msg, i)->length;
2188                 if (start < offset + len)
2189                         break;
2190                 offset += len;
2191                 sk_msg_iter_var_next(i);
2192         } while (i != msg->sg.end);
2193
2194         if (unlikely(start >= offset + len))
2195                 return -EINVAL;
2196
2197         first_sge = i;
2198         /* The start may point into the sg element so we need to also
2199          * account for the headroom.
2200          */
2201         bytes_sg_total = start - offset + bytes;
2202         if (!msg->sg.copy[i] && bytes_sg_total <= len)
2203                 goto out;
2204
2205         /* At this point we need to linearize multiple scatterlist
2206          * elements or a single shared page. Either way we need to
2207          * copy into a linear buffer exclusively owned by BPF. Then
2208          * place the buffer in the scatterlist and fixup the original
2209          * entries by removing the entries now in the linear buffer
2210          * and shifting the remaining entries. For now we do not try
2211          * to copy partial entries to avoid complexity of running out
2212          * of sg_entry slots. The downside is reading a single byte
2213          * will copy the entire sg entry.
2214          */
2215         do {
2216                 copy += sk_msg_elem(msg, i)->length;
2217                 sk_msg_iter_var_next(i);
2218                 if (bytes_sg_total <= copy)
2219                         break;
2220         } while (i != msg->sg.end);
2221         last_sge = i;
2222
2223         if (unlikely(bytes_sg_total > copy))
2224                 return -EINVAL;
2225
2226         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2227                            get_order(copy));
2228         if (unlikely(!page))
2229                 return -ENOMEM;
2230
2231         raw = page_address(page);
2232         i = first_sge;
2233         do {
2234                 sge = sk_msg_elem(msg, i);
2235                 from = sg_virt(sge);
2236                 len = sge->length;
2237                 to = raw + poffset;
2238
2239                 memcpy(to, from, len);
2240                 poffset += len;
2241                 sge->length = 0;
2242                 put_page(sg_page(sge));
2243
2244                 sk_msg_iter_var_next(i);
2245         } while (i != last_sge);
2246
2247         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2248
2249         /* To repair sg ring we need to shift entries. If we only
2250          * had a single entry though we can just replace it and
2251          * be done. Otherwise walk the ring and shift the entries.
2252          */
2253         WARN_ON_ONCE(last_sge == first_sge);
2254         shift = last_sge > first_sge ?
2255                 last_sge - first_sge - 1 :
2256                 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2257         if (!shift)
2258                 goto out;
2259
2260         i = first_sge;
2261         sk_msg_iter_var_next(i);
2262         do {
2263                 u32 move_from;
2264
2265                 if (i + shift >= MAX_MSG_FRAGS)
2266                         move_from = i + shift - MAX_MSG_FRAGS;
2267                 else
2268                         move_from = i + shift;
2269                 if (move_from == msg->sg.end)
2270                         break;
2271
2272                 msg->sg.data[i] = msg->sg.data[move_from];
2273                 msg->sg.data[move_from].length = 0;
2274                 msg->sg.data[move_from].page_link = 0;
2275                 msg->sg.data[move_from].offset = 0;
2276                 sk_msg_iter_var_next(i);
2277         } while (1);
2278
2279         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2280                       msg->sg.end - shift + MAX_MSG_FRAGS :
2281                       msg->sg.end - shift;
2282 out:
2283         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2284         msg->data_end = msg->data + bytes;
2285         return 0;
2286 }
2287
2288 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2289         .func           = bpf_msg_pull_data,
2290         .gpl_only       = false,
2291         .ret_type       = RET_INTEGER,
2292         .arg1_type      = ARG_PTR_TO_CTX,
2293         .arg2_type      = ARG_ANYTHING,
2294         .arg3_type      = ARG_ANYTHING,
2295         .arg4_type      = ARG_ANYTHING,
2296 };
2297
2298 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2299            u32, len, u64, flags)
2300 {
2301         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2302         u32 new, i = 0, l, space, copy = 0, offset = 0;
2303         u8 *raw, *to, *from;
2304         struct page *page;
2305
2306         if (unlikely(flags))
2307                 return -EINVAL;
2308
2309         /* First find the starting scatterlist element */
2310         i = msg->sg.start;
2311         do {
2312                 l = sk_msg_elem(msg, i)->length;
2313
2314                 if (start < offset + l)
2315                         break;
2316                 offset += l;
2317                 sk_msg_iter_var_next(i);
2318         } while (i != msg->sg.end);
2319
2320         if (start >= offset + l)
2321                 return -EINVAL;
2322
2323         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2324
2325         /* If no space available will fallback to copy, we need at
2326          * least one scatterlist elem available to push data into
2327          * when start aligns to the beginning of an element or two
2328          * when it falls inside an element. We handle the start equals
2329          * offset case because its the common case for inserting a
2330          * header.
2331          */
2332         if (!space || (space == 1 && start != offset))
2333                 copy = msg->sg.data[i].length;
2334
2335         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2336                            get_order(copy + len));
2337         if (unlikely(!page))
2338                 return -ENOMEM;
2339
2340         if (copy) {
2341                 int front, back;
2342
2343                 raw = page_address(page);
2344
2345                 psge = sk_msg_elem(msg, i);
2346                 front = start - offset;
2347                 back = psge->length - front;
2348                 from = sg_virt(psge);
2349
2350                 if (front)
2351                         memcpy(raw, from, front);
2352
2353                 if (back) {
2354                         from += front;
2355                         to = raw + front + len;
2356
2357                         memcpy(to, from, back);
2358                 }
2359
2360                 put_page(sg_page(psge));
2361         } else if (start - offset) {
2362                 psge = sk_msg_elem(msg, i);
2363                 rsge = sk_msg_elem_cpy(msg, i);
2364
2365                 psge->length = start - offset;
2366                 rsge.length -= psge->length;
2367                 rsge.offset += start;
2368
2369                 sk_msg_iter_var_next(i);
2370                 sg_unmark_end(psge);
2371                 sk_msg_iter_next(msg, end);
2372         }
2373
2374         /* Slot(s) to place newly allocated data */
2375         new = i;
2376
2377         /* Shift one or two slots as needed */
2378         if (!copy) {
2379                 sge = sk_msg_elem_cpy(msg, i);
2380
2381                 sk_msg_iter_var_next(i);
2382                 sg_unmark_end(&sge);
2383                 sk_msg_iter_next(msg, end);
2384
2385                 nsge = sk_msg_elem_cpy(msg, i);
2386                 if (rsge.length) {
2387                         sk_msg_iter_var_next(i);
2388                         nnsge = sk_msg_elem_cpy(msg, i);
2389                 }
2390
2391                 while (i != msg->sg.end) {
2392                         msg->sg.data[i] = sge;
2393                         sge = nsge;
2394                         sk_msg_iter_var_next(i);
2395                         if (rsge.length) {
2396                                 nsge = nnsge;
2397                                 nnsge = sk_msg_elem_cpy(msg, i);
2398                         } else {
2399                                 nsge = sk_msg_elem_cpy(msg, i);
2400                         }
2401                 }
2402         }
2403
2404         /* Place newly allocated data buffer */
2405         sk_mem_charge(msg->sk, len);
2406         msg->sg.size += len;
2407         msg->sg.copy[new] = false;
2408         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2409         if (rsge.length) {
2410                 get_page(sg_page(&rsge));
2411                 sk_msg_iter_var_next(new);
2412                 msg->sg.data[new] = rsge;
2413         }
2414
2415         sk_msg_compute_data_pointers(msg);
2416         return 0;
2417 }
2418
2419 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2420         .func           = bpf_msg_push_data,
2421         .gpl_only       = false,
2422         .ret_type       = RET_INTEGER,
2423         .arg1_type      = ARG_PTR_TO_CTX,
2424         .arg2_type      = ARG_ANYTHING,
2425         .arg3_type      = ARG_ANYTHING,
2426         .arg4_type      = ARG_ANYTHING,
2427 };
2428
2429 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2430 {
2431         int prev;
2432
2433         do {
2434                 prev = i;
2435                 sk_msg_iter_var_next(i);
2436                 msg->sg.data[prev] = msg->sg.data[i];
2437         } while (i != msg->sg.end);
2438
2439         sk_msg_iter_prev(msg, end);
2440 }
2441
2442 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2443 {
2444         struct scatterlist tmp, sge;
2445
2446         sk_msg_iter_next(msg, end);
2447         sge = sk_msg_elem_cpy(msg, i);
2448         sk_msg_iter_var_next(i);
2449         tmp = sk_msg_elem_cpy(msg, i);
2450
2451         while (i != msg->sg.end) {
2452                 msg->sg.data[i] = sge;
2453                 sk_msg_iter_var_next(i);
2454                 sge = tmp;
2455                 tmp = sk_msg_elem_cpy(msg, i);
2456         }
2457 }
2458
2459 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2460            u32, len, u64, flags)
2461 {
2462         u32 i = 0, l, space, offset = 0;
2463         u64 last = start + len;
2464         int pop;
2465
2466         if (unlikely(flags))
2467                 return -EINVAL;
2468
2469         /* First find the starting scatterlist element */
2470         i = msg->sg.start;
2471         do {
2472                 l = sk_msg_elem(msg, i)->length;
2473
2474                 if (start < offset + l)
2475                         break;
2476                 offset += l;
2477                 sk_msg_iter_var_next(i);
2478         } while (i != msg->sg.end);
2479
2480         /* Bounds checks: start and pop must be inside message */
2481         if (start >= offset + l || last >= msg->sg.size)
2482                 return -EINVAL;
2483
2484         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2485
2486         pop = len;
2487         /* --------------| offset
2488          * -| start      |-------- len -------|
2489          *
2490          *  |----- a ----|-------- pop -------|----- b ----|
2491          *  |______________________________________________| length
2492          *
2493          *
2494          * a:   region at front of scatter element to save
2495          * b:   region at back of scatter element to save when length > A + pop
2496          * pop: region to pop from element, same as input 'pop' here will be
2497          *      decremented below per iteration.
2498          *
2499          * Two top-level cases to handle when start != offset, first B is non
2500          * zero and second B is zero corresponding to when a pop includes more
2501          * than one element.
2502          *
2503          * Then if B is non-zero AND there is no space allocate space and
2504          * compact A, B regions into page. If there is space shift ring to
2505          * the rigth free'ing the next element in ring to place B, leaving
2506          * A untouched except to reduce length.
2507          */
2508         if (start != offset) {
2509                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2510                 int a = start;
2511                 int b = sge->length - pop - a;
2512
2513                 sk_msg_iter_var_next(i);
2514
2515                 if (pop < sge->length - a) {
2516                         if (space) {
2517                                 sge->length = a;
2518                                 sk_msg_shift_right(msg, i);
2519                                 nsge = sk_msg_elem(msg, i);
2520                                 get_page(sg_page(sge));
2521                                 sg_set_page(nsge,
2522                                             sg_page(sge),
2523                                             b, sge->offset + pop + a);
2524                         } else {
2525                                 struct page *page, *orig;
2526                                 u8 *to, *from;
2527
2528                                 page = alloc_pages(__GFP_NOWARN |
2529                                                    __GFP_COMP   | GFP_ATOMIC,
2530                                                    get_order(a + b));
2531                                 if (unlikely(!page))
2532                                         return -ENOMEM;
2533
2534                                 sge->length = a;
2535                                 orig = sg_page(sge);
2536                                 from = sg_virt(sge);
2537                                 to = page_address(page);
2538                                 memcpy(to, from, a);
2539                                 memcpy(to + a, from + a + pop, b);
2540                                 sg_set_page(sge, page, a + b, 0);
2541                                 put_page(orig);
2542                         }
2543                         pop = 0;
2544                 } else if (pop >= sge->length - a) {
2545                         sge->length = a;
2546                         pop -= (sge->length - a);
2547                 }
2548         }
2549
2550         /* From above the current layout _must_ be as follows,
2551          *
2552          * -| offset
2553          * -| start
2554          *
2555          *  |---- pop ---|---------------- b ------------|
2556          *  |____________________________________________| length
2557          *
2558          * Offset and start of the current msg elem are equal because in the
2559          * previous case we handled offset != start and either consumed the
2560          * entire element and advanced to the next element OR pop == 0.
2561          *
2562          * Two cases to handle here are first pop is less than the length
2563          * leaving some remainder b above. Simply adjust the element's layout
2564          * in this case. Or pop >= length of the element so that b = 0. In this
2565          * case advance to next element decrementing pop.
2566          */
2567         while (pop) {
2568                 struct scatterlist *sge = sk_msg_elem(msg, i);
2569
2570                 if (pop < sge->length) {
2571                         sge->length -= pop;
2572                         sge->offset += pop;
2573                         pop = 0;
2574                 } else {
2575                         pop -= sge->length;
2576                         sk_msg_shift_left(msg, i);
2577                 }
2578                 sk_msg_iter_var_next(i);
2579         }
2580
2581         sk_mem_uncharge(msg->sk, len - pop);
2582         msg->sg.size -= (len - pop);
2583         sk_msg_compute_data_pointers(msg);
2584         return 0;
2585 }
2586
2587 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2588         .func           = bpf_msg_pop_data,
2589         .gpl_only       = false,
2590         .ret_type       = RET_INTEGER,
2591         .arg1_type      = ARG_PTR_TO_CTX,
2592         .arg2_type      = ARG_ANYTHING,
2593         .arg3_type      = ARG_ANYTHING,
2594         .arg4_type      = ARG_ANYTHING,
2595 };
2596
2597 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2598 {
2599         return task_get_classid(skb);
2600 }
2601
2602 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2603         .func           = bpf_get_cgroup_classid,
2604         .gpl_only       = false,
2605         .ret_type       = RET_INTEGER,
2606         .arg1_type      = ARG_PTR_TO_CTX,
2607 };
2608
2609 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2610 {
2611         return dst_tclassid(skb);
2612 }
2613
2614 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2615         .func           = bpf_get_route_realm,
2616         .gpl_only       = false,
2617         .ret_type       = RET_INTEGER,
2618         .arg1_type      = ARG_PTR_TO_CTX,
2619 };
2620
2621 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2622 {
2623         /* If skb_clear_hash() was called due to mangling, we can
2624          * trigger SW recalculation here. Later access to hash
2625          * can then use the inline skb->hash via context directly
2626          * instead of calling this helper again.
2627          */
2628         return skb_get_hash(skb);
2629 }
2630
2631 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2632         .func           = bpf_get_hash_recalc,
2633         .gpl_only       = false,
2634         .ret_type       = RET_INTEGER,
2635         .arg1_type      = ARG_PTR_TO_CTX,
2636 };
2637
2638 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2639 {
2640         /* After all direct packet write, this can be used once for
2641          * triggering a lazy recalc on next skb_get_hash() invocation.
2642          */
2643         skb_clear_hash(skb);
2644         return 0;
2645 }
2646
2647 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2648         .func           = bpf_set_hash_invalid,
2649         .gpl_only       = false,
2650         .ret_type       = RET_INTEGER,
2651         .arg1_type      = ARG_PTR_TO_CTX,
2652 };
2653
2654 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2655 {
2656         /* Set user specified hash as L4(+), so that it gets returned
2657          * on skb_get_hash() call unless BPF prog later on triggers a
2658          * skb_clear_hash().
2659          */
2660         __skb_set_sw_hash(skb, hash, true);
2661         return 0;
2662 }
2663
2664 static const struct bpf_func_proto bpf_set_hash_proto = {
2665         .func           = bpf_set_hash,
2666         .gpl_only       = false,
2667         .ret_type       = RET_INTEGER,
2668         .arg1_type      = ARG_PTR_TO_CTX,
2669         .arg2_type      = ARG_ANYTHING,
2670 };
2671
2672 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2673            u16, vlan_tci)
2674 {
2675         int ret;
2676
2677         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2678                      vlan_proto != htons(ETH_P_8021AD)))
2679                 vlan_proto = htons(ETH_P_8021Q);
2680
2681         bpf_push_mac_rcsum(skb);
2682         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2683         bpf_pull_mac_rcsum(skb);
2684
2685         bpf_compute_data_pointers(skb);
2686         return ret;
2687 }
2688
2689 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2690         .func           = bpf_skb_vlan_push,
2691         .gpl_only       = false,
2692         .ret_type       = RET_INTEGER,
2693         .arg1_type      = ARG_PTR_TO_CTX,
2694         .arg2_type      = ARG_ANYTHING,
2695         .arg3_type      = ARG_ANYTHING,
2696 };
2697
2698 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2699 {
2700         int ret;
2701
2702         bpf_push_mac_rcsum(skb);
2703         ret = skb_vlan_pop(skb);
2704         bpf_pull_mac_rcsum(skb);
2705
2706         bpf_compute_data_pointers(skb);
2707         return ret;
2708 }
2709
2710 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2711         .func           = bpf_skb_vlan_pop,
2712         .gpl_only       = false,
2713         .ret_type       = RET_INTEGER,
2714         .arg1_type      = ARG_PTR_TO_CTX,
2715 };
2716
2717 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2718 {
2719         /* Caller already did skb_cow() with len as headroom,
2720          * so no need to do it here.
2721          */
2722         skb_push(skb, len);
2723         memmove(skb->data, skb->data + len, off);
2724         memset(skb->data + off, 0, len);
2725
2726         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2727          * needed here as it does not change the skb->csum
2728          * result for checksum complete when summing over
2729          * zeroed blocks.
2730          */
2731         return 0;
2732 }
2733
2734 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2735 {
2736         /* skb_ensure_writable() is not needed here, as we're
2737          * already working on an uncloned skb.
2738          */
2739         if (unlikely(!pskb_may_pull(skb, off + len)))
2740                 return -ENOMEM;
2741
2742         skb_postpull_rcsum(skb, skb->data + off, len);
2743         memmove(skb->data + len, skb->data, off);
2744         __skb_pull(skb, len);
2745
2746         return 0;
2747 }
2748
2749 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2750 {
2751         bool trans_same = skb->transport_header == skb->network_header;
2752         int ret;
2753
2754         /* There's no need for __skb_push()/__skb_pull() pair to
2755          * get to the start of the mac header as we're guaranteed
2756          * to always start from here under eBPF.
2757          */
2758         ret = bpf_skb_generic_push(skb, off, len);
2759         if (likely(!ret)) {
2760                 skb->mac_header -= len;
2761                 skb->network_header -= len;
2762                 if (trans_same)
2763                         skb->transport_header = skb->network_header;
2764         }
2765
2766         return ret;
2767 }
2768
2769 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2770 {
2771         bool trans_same = skb->transport_header == skb->network_header;
2772         int ret;
2773
2774         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2775         ret = bpf_skb_generic_pop(skb, off, len);
2776         if (likely(!ret)) {
2777                 skb->mac_header += len;
2778                 skb->network_header += len;
2779                 if (trans_same)
2780                         skb->transport_header = skb->network_header;
2781         }
2782
2783         return ret;
2784 }
2785
2786 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2787 {
2788         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2789         u32 off = skb_mac_header_len(skb);
2790         int ret;
2791
2792         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2793         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2794                 return -ENOTSUPP;
2795
2796         ret = skb_cow(skb, len_diff);
2797         if (unlikely(ret < 0))
2798                 return ret;
2799
2800         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2801         if (unlikely(ret < 0))
2802                 return ret;
2803
2804         if (skb_is_gso(skb)) {
2805                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2806
2807                 /* SKB_GSO_TCPV4 needs to be changed into
2808                  * SKB_GSO_TCPV6.
2809                  */
2810                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2811                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2812                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2813                 }
2814
2815                 /* Due to IPv6 header, MSS needs to be downgraded. */
2816                 skb_decrease_gso_size(shinfo, len_diff);
2817                 /* Header must be checked, and gso_segs recomputed. */
2818                 shinfo->gso_type |= SKB_GSO_DODGY;
2819                 shinfo->gso_segs = 0;
2820         }
2821
2822         skb->protocol = htons(ETH_P_IPV6);
2823         skb_clear_hash(skb);
2824
2825         return 0;
2826 }
2827
2828 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2829 {
2830         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2831         u32 off = skb_mac_header_len(skb);
2832         int ret;
2833
2834         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2835         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2836                 return -ENOTSUPP;
2837
2838         ret = skb_unclone(skb, GFP_ATOMIC);
2839         if (unlikely(ret < 0))
2840                 return ret;
2841
2842         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2843         if (unlikely(ret < 0))
2844                 return ret;
2845
2846         if (skb_is_gso(skb)) {
2847                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2848
2849                 /* SKB_GSO_TCPV6 needs to be changed into
2850                  * SKB_GSO_TCPV4.
2851                  */
2852                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2853                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2854                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2855                 }
2856
2857                 /* Due to IPv4 header, MSS can be upgraded. */
2858                 skb_increase_gso_size(shinfo, len_diff);
2859                 /* Header must be checked, and gso_segs recomputed. */
2860                 shinfo->gso_type |= SKB_GSO_DODGY;
2861                 shinfo->gso_segs = 0;
2862         }
2863
2864         skb->protocol = htons(ETH_P_IP);
2865         skb_clear_hash(skb);
2866
2867         return 0;
2868 }
2869
2870 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2871 {
2872         __be16 from_proto = skb->protocol;
2873
2874         if (from_proto == htons(ETH_P_IP) &&
2875               to_proto == htons(ETH_P_IPV6))
2876                 return bpf_skb_proto_4_to_6(skb);
2877
2878         if (from_proto == htons(ETH_P_IPV6) &&
2879               to_proto == htons(ETH_P_IP))
2880                 return bpf_skb_proto_6_to_4(skb);
2881
2882         return -ENOTSUPP;
2883 }
2884
2885 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2886            u64, flags)
2887 {
2888         int ret;
2889
2890         if (unlikely(flags))
2891                 return -EINVAL;
2892
2893         /* General idea is that this helper does the basic groundwork
2894          * needed for changing the protocol, and eBPF program fills the
2895          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2896          * and other helpers, rather than passing a raw buffer here.
2897          *
2898          * The rationale is to keep this minimal and without a need to
2899          * deal with raw packet data. F.e. even if we would pass buffers
2900          * here, the program still needs to call the bpf_lX_csum_replace()
2901          * helpers anyway. Plus, this way we keep also separation of
2902          * concerns, since f.e. bpf_skb_store_bytes() should only take
2903          * care of stores.
2904          *
2905          * Currently, additional options and extension header space are
2906          * not supported, but flags register is reserved so we can adapt
2907          * that. For offloads, we mark packet as dodgy, so that headers
2908          * need to be verified first.
2909          */
2910         ret = bpf_skb_proto_xlat(skb, proto);
2911         bpf_compute_data_pointers(skb);
2912         return ret;
2913 }
2914
2915 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2916         .func           = bpf_skb_change_proto,
2917         .gpl_only       = false,
2918         .ret_type       = RET_INTEGER,
2919         .arg1_type      = ARG_PTR_TO_CTX,
2920         .arg2_type      = ARG_ANYTHING,
2921         .arg3_type      = ARG_ANYTHING,
2922 };
2923
2924 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2925 {
2926         /* We only allow a restricted subset to be changed for now. */
2927         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2928                      !skb_pkt_type_ok(pkt_type)))
2929                 return -EINVAL;
2930
2931         skb->pkt_type = pkt_type;
2932         return 0;
2933 }
2934
2935 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2936         .func           = bpf_skb_change_type,
2937         .gpl_only       = false,
2938         .ret_type       = RET_INTEGER,
2939         .arg1_type      = ARG_PTR_TO_CTX,
2940         .arg2_type      = ARG_ANYTHING,
2941 };
2942
2943 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2944 {
2945         switch (skb->protocol) {
2946         case htons(ETH_P_IP):
2947                 return sizeof(struct iphdr);
2948         case htons(ETH_P_IPV6):
2949                 return sizeof(struct ipv6hdr);
2950         default:
2951                 return ~0U;
2952         }
2953 }
2954
2955 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2956 {
2957         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2958         int ret;
2959
2960         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2961         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2962                 return -ENOTSUPP;
2963
2964         ret = skb_cow(skb, len_diff);
2965         if (unlikely(ret < 0))
2966                 return ret;
2967
2968         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2969         if (unlikely(ret < 0))
2970                 return ret;
2971
2972         if (skb_is_gso(skb)) {
2973                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2974
2975                 /* Due to header grow, MSS needs to be downgraded. */
2976                 skb_decrease_gso_size(shinfo, len_diff);
2977                 /* Header must be checked, and gso_segs recomputed. */
2978                 shinfo->gso_type |= SKB_GSO_DODGY;
2979                 shinfo->gso_segs = 0;
2980         }
2981
2982         return 0;
2983 }
2984
2985 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2986 {
2987         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2988         int ret;
2989
2990         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2991         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2992                 return -ENOTSUPP;
2993
2994         ret = skb_unclone(skb, GFP_ATOMIC);
2995         if (unlikely(ret < 0))
2996                 return ret;
2997
2998         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2999         if (unlikely(ret < 0))
3000                 return ret;
3001
3002         if (skb_is_gso(skb)) {
3003                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3004
3005                 /* Due to header shrink, MSS can be upgraded. */
3006                 skb_increase_gso_size(shinfo, len_diff);
3007                 /* Header must be checked, and gso_segs recomputed. */
3008                 shinfo->gso_type |= SKB_GSO_DODGY;
3009                 shinfo->gso_segs = 0;
3010         }
3011
3012         return 0;
3013 }
3014
3015 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3016 {
3017         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3018                           SKB_MAX_ALLOC;
3019 }
3020
3021 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
3022 {
3023         bool trans_same = skb->transport_header == skb->network_header;
3024         u32 len_cur, len_diff_abs = abs(len_diff);
3025         u32 len_min = bpf_skb_net_base_len(skb);
3026         u32 len_max = __bpf_skb_max_len(skb);
3027         __be16 proto = skb->protocol;
3028         bool shrink = len_diff < 0;
3029         int ret;
3030
3031         if (unlikely(len_diff_abs > 0xfffU))
3032                 return -EFAULT;
3033         if (unlikely(proto != htons(ETH_P_IP) &&
3034                      proto != htons(ETH_P_IPV6)))
3035                 return -ENOTSUPP;
3036
3037         len_cur = skb->len - skb_network_offset(skb);
3038         if (skb_transport_header_was_set(skb) && !trans_same)
3039                 len_cur = skb_network_header_len(skb);
3040         if ((shrink && (len_diff_abs >= len_cur ||
3041                         len_cur - len_diff_abs < len_min)) ||
3042             (!shrink && (skb->len + len_diff_abs > len_max &&
3043                          !skb_is_gso(skb))))
3044                 return -ENOTSUPP;
3045
3046         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
3047                        bpf_skb_net_grow(skb, len_diff_abs);
3048
3049         bpf_compute_data_pointers(skb);
3050         return ret;
3051 }
3052
3053 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3054            u32, mode, u64, flags)
3055 {
3056         if (unlikely(flags))
3057                 return -EINVAL;
3058         if (likely(mode == BPF_ADJ_ROOM_NET))
3059                 return bpf_skb_adjust_net(skb, len_diff);
3060
3061         return -ENOTSUPP;
3062 }
3063
3064 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3065         .func           = bpf_skb_adjust_room,
3066         .gpl_only       = false,
3067         .ret_type       = RET_INTEGER,
3068         .arg1_type      = ARG_PTR_TO_CTX,
3069         .arg2_type      = ARG_ANYTHING,
3070         .arg3_type      = ARG_ANYTHING,
3071         .arg4_type      = ARG_ANYTHING,
3072 };
3073
3074 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3075 {
3076         u32 min_len = skb_network_offset(skb);
3077
3078         if (skb_transport_header_was_set(skb))
3079                 min_len = skb_transport_offset(skb);
3080         if (skb->ip_summed == CHECKSUM_PARTIAL)
3081                 min_len = skb_checksum_start_offset(skb) +
3082                           skb->csum_offset + sizeof(__sum16);
3083         return min_len;
3084 }
3085
3086 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3087 {
3088         unsigned int old_len = skb->len;
3089         int ret;
3090
3091         ret = __skb_grow_rcsum(skb, new_len);
3092         if (!ret)
3093                 memset(skb->data + old_len, 0, new_len - old_len);
3094         return ret;
3095 }
3096
3097 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3098 {
3099         return __skb_trim_rcsum(skb, new_len);
3100 }
3101
3102 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3103                                         u64 flags)
3104 {
3105         u32 max_len = __bpf_skb_max_len(skb);
3106         u32 min_len = __bpf_skb_min_len(skb);
3107         int ret;
3108
3109         if (unlikely(flags || new_len > max_len || new_len < min_len))
3110                 return -EINVAL;
3111         if (skb->encapsulation)
3112                 return -ENOTSUPP;
3113
3114         /* The basic idea of this helper is that it's performing the
3115          * needed work to either grow or trim an skb, and eBPF program
3116          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3117          * bpf_lX_csum_replace() and others rather than passing a raw
3118          * buffer here. This one is a slow path helper and intended
3119          * for replies with control messages.
3120          *
3121          * Like in bpf_skb_change_proto(), we want to keep this rather
3122          * minimal and without protocol specifics so that we are able
3123          * to separate concerns as in bpf_skb_store_bytes() should only
3124          * be the one responsible for writing buffers.
3125          *
3126          * It's really expected to be a slow path operation here for
3127          * control message replies, so we're implicitly linearizing,
3128          * uncloning and drop offloads from the skb by this.
3129          */
3130         ret = __bpf_try_make_writable(skb, skb->len);
3131         if (!ret) {
3132                 if (new_len > skb->len)
3133                         ret = bpf_skb_grow_rcsum(skb, new_len);
3134                 else if (new_len < skb->len)
3135                         ret = bpf_skb_trim_rcsum(skb, new_len);
3136                 if (!ret && skb_is_gso(skb))
3137                         skb_gso_reset(skb);
3138         }
3139         return ret;
3140 }
3141
3142 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3143            u64, flags)
3144 {
3145         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3146
3147         bpf_compute_data_pointers(skb);
3148         return ret;
3149 }
3150
3151 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3152         .func           = bpf_skb_change_tail,
3153         .gpl_only       = false,
3154         .ret_type       = RET_INTEGER,
3155         .arg1_type      = ARG_PTR_TO_CTX,
3156         .arg2_type      = ARG_ANYTHING,
3157         .arg3_type      = ARG_ANYTHING,
3158 };
3159
3160 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3161            u64, flags)
3162 {
3163         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3164
3165         bpf_compute_data_end_sk_skb(skb);
3166         return ret;
3167 }
3168
3169 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3170         .func           = sk_skb_change_tail,
3171         .gpl_only       = false,
3172         .ret_type       = RET_INTEGER,
3173         .arg1_type      = ARG_PTR_TO_CTX,
3174         .arg2_type      = ARG_ANYTHING,
3175         .arg3_type      = ARG_ANYTHING,
3176 };
3177
3178 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3179                                         u64 flags)
3180 {
3181         u32 max_len = __bpf_skb_max_len(skb);
3182         u32 new_len = skb->len + head_room;
3183         int ret;
3184
3185         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3186                      new_len < skb->len))
3187                 return -EINVAL;
3188
3189         ret = skb_cow(skb, head_room);
3190         if (likely(!ret)) {
3191                 /* Idea for this helper is that we currently only
3192                  * allow to expand on mac header. This means that
3193                  * skb->protocol network header, etc, stay as is.
3194                  * Compared to bpf_skb_change_tail(), we're more
3195                  * flexible due to not needing to linearize or
3196                  * reset GSO. Intention for this helper is to be
3197                  * used by an L3 skb that needs to push mac header
3198                  * for redirection into L2 device.
3199                  */
3200                 __skb_push(skb, head_room);
3201                 memset(skb->data, 0, head_room);
3202                 skb_reset_mac_header(skb);
3203         }
3204
3205         return ret;
3206 }
3207
3208 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3209            u64, flags)
3210 {
3211         int ret = __bpf_skb_change_head(skb, head_room, flags);
3212
3213         bpf_compute_data_pointers(skb);
3214         return ret;
3215 }
3216
3217 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3218         .func           = bpf_skb_change_head,
3219         .gpl_only       = false,
3220         .ret_type       = RET_INTEGER,
3221         .arg1_type      = ARG_PTR_TO_CTX,
3222         .arg2_type      = ARG_ANYTHING,
3223         .arg3_type      = ARG_ANYTHING,
3224 };
3225
3226 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3227            u64, flags)
3228 {
3229         int ret = __bpf_skb_change_head(skb, head_room, flags);
3230
3231         bpf_compute_data_end_sk_skb(skb);
3232         return ret;
3233 }
3234
3235 static const struct bpf_func_proto sk_skb_change_head_proto = {
3236         .func           = sk_skb_change_head,
3237         .gpl_only       = false,
3238         .ret_type       = RET_INTEGER,
3239         .arg1_type      = ARG_PTR_TO_CTX,
3240         .arg2_type      = ARG_ANYTHING,
3241         .arg3_type      = ARG_ANYTHING,
3242 };
3243 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3244 {
3245         return xdp_data_meta_unsupported(xdp) ? 0 :
3246                xdp->data - xdp->data_meta;
3247 }
3248
3249 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3250 {
3251         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3252         unsigned long metalen = xdp_get_metalen(xdp);
3253         void *data_start = xdp_frame_end + metalen;
3254         void *data = xdp->data + offset;
3255
3256         if (unlikely(data < data_start ||
3257                      data > xdp->data_end - ETH_HLEN))
3258                 return -EINVAL;
3259
3260         if (metalen)
3261                 memmove(xdp->data_meta + offset,
3262                         xdp->data_meta, metalen);
3263         xdp->data_meta += offset;
3264         xdp->data = data;
3265
3266         return 0;
3267 }
3268
3269 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3270         .func           = bpf_xdp_adjust_head,
3271         .gpl_only       = false,
3272         .ret_type       = RET_INTEGER,
3273         .arg1_type      = ARG_PTR_TO_CTX,
3274         .arg2_type      = ARG_ANYTHING,
3275 };
3276
3277 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3278 {
3279         void *data_end = xdp->data_end + offset;
3280
3281         /* only shrinking is allowed for now. */
3282         if (unlikely(offset >= 0))
3283                 return -EINVAL;
3284
3285         if (unlikely(data_end < xdp->data + ETH_HLEN))
3286                 return -EINVAL;
3287
3288         xdp->data_end = data_end;
3289
3290         return 0;
3291 }
3292
3293 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3294         .func           = bpf_xdp_adjust_tail,
3295         .gpl_only       = false,
3296         .ret_type       = RET_INTEGER,
3297         .arg1_type      = ARG_PTR_TO_CTX,
3298         .arg2_type      = ARG_ANYTHING,
3299 };
3300
3301 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3302 {
3303         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3304         void *meta = xdp->data_meta + offset;
3305         unsigned long metalen = xdp->data - meta;
3306
3307         if (xdp_data_meta_unsupported(xdp))
3308                 return -ENOTSUPP;
3309         if (unlikely(meta < xdp_frame_end ||
3310                      meta > xdp->data))
3311                 return -EINVAL;
3312         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3313                      (metalen > 32)))
3314                 return -EACCES;
3315
3316         xdp->data_meta = meta;
3317
3318         return 0;
3319 }
3320
3321 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3322         .func           = bpf_xdp_adjust_meta,
3323         .gpl_only       = false,
3324         .ret_type       = RET_INTEGER,
3325         .arg1_type      = ARG_PTR_TO_CTX,
3326         .arg2_type      = ARG_ANYTHING,
3327 };
3328
3329 static int __bpf_tx_xdp(struct net_device *dev,
3330                         struct bpf_map *map,
3331                         struct xdp_buff *xdp,
3332                         u32 index)
3333 {
3334         struct xdp_frame *xdpf;
3335         int err, sent;
3336
3337         if (!dev->netdev_ops->ndo_xdp_xmit) {
3338                 return -EOPNOTSUPP;
3339         }
3340
3341         err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3342         if (unlikely(err))
3343                 return err;
3344
3345         xdpf = convert_to_xdp_frame(xdp);
3346         if (unlikely(!xdpf))
3347                 return -EOVERFLOW;
3348
3349         sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3350         if (sent <= 0)
3351                 return sent;
3352         return 0;
3353 }
3354
3355 static noinline int
3356 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3357                      struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3358 {
3359         struct net_device *fwd;
3360         u32 index = ri->ifindex;
3361         int err;
3362
3363         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3364         ri->ifindex = 0;
3365         if (unlikely(!fwd)) {
3366                 err = -EINVAL;
3367                 goto err;
3368         }
3369
3370         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3371         if (unlikely(err))
3372                 goto err;
3373
3374         _trace_xdp_redirect(dev, xdp_prog, index);
3375         return 0;
3376 err:
3377         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3378         return err;
3379 }
3380
3381 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3382                             struct bpf_map *map,
3383                             struct xdp_buff *xdp,
3384                             u32 index)
3385 {
3386         int err;
3387
3388         switch (map->map_type) {
3389         case BPF_MAP_TYPE_DEVMAP: {
3390                 struct bpf_dtab_netdev *dst = fwd;
3391
3392                 err = dev_map_enqueue(dst, xdp, dev_rx);
3393                 if (unlikely(err))
3394                         return err;
3395                 __dev_map_insert_ctx(map, index);
3396                 break;
3397         }
3398         case BPF_MAP_TYPE_CPUMAP: {
3399                 struct bpf_cpu_map_entry *rcpu = fwd;
3400
3401                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3402                 if (unlikely(err))
3403                         return err;
3404                 __cpu_map_insert_ctx(map, index);
3405                 break;
3406         }
3407         case BPF_MAP_TYPE_XSKMAP: {
3408                 struct xdp_sock *xs = fwd;
3409
3410                 err = __xsk_map_redirect(map, xdp, xs);
3411                 return err;
3412         }
3413         default:
3414                 break;
3415         }
3416         return 0;
3417 }
3418
3419 void xdp_do_flush_map(void)
3420 {
3421         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3422         struct bpf_map *map = ri->map_to_flush;
3423
3424         ri->map_to_flush = NULL;
3425         if (map) {
3426                 switch (map->map_type) {
3427                 case BPF_MAP_TYPE_DEVMAP:
3428                         __dev_map_flush(map);
3429                         break;
3430                 case BPF_MAP_TYPE_CPUMAP:
3431                         __cpu_map_flush(map);
3432                         break;
3433                 case BPF_MAP_TYPE_XSKMAP:
3434                         __xsk_map_flush(map);
3435                         break;
3436                 default:
3437                         break;
3438                 }
3439         }
3440 }
3441 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3442
3443 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3444 {
3445         switch (map->map_type) {
3446         case BPF_MAP_TYPE_DEVMAP:
3447                 return __dev_map_lookup_elem(map, index);
3448         case BPF_MAP_TYPE_CPUMAP:
3449                 return __cpu_map_lookup_elem(map, index);
3450         case BPF_MAP_TYPE_XSKMAP:
3451                 return __xsk_map_lookup_elem(map, index);
3452         default:
3453                 return NULL;
3454         }
3455 }
3456
3457 void bpf_clear_redirect_map(struct bpf_map *map)
3458 {
3459         struct bpf_redirect_info *ri;
3460         int cpu;
3461
3462         for_each_possible_cpu(cpu) {
3463                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3464                 /* Avoid polluting remote cacheline due to writes if
3465                  * not needed. Once we pass this test, we need the
3466                  * cmpxchg() to make sure it hasn't been changed in
3467                  * the meantime by remote CPU.
3468                  */
3469                 if (unlikely(READ_ONCE(ri->map) == map))
3470                         cmpxchg(&ri->map, map, NULL);
3471         }
3472 }
3473
3474 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3475                                struct bpf_prog *xdp_prog, struct bpf_map *map,
3476                                struct bpf_redirect_info *ri)
3477 {
3478         u32 index = ri->ifindex;
3479         void *fwd = NULL;
3480         int err;
3481
3482         ri->ifindex = 0;
3483         WRITE_ONCE(ri->map, NULL);
3484
3485         fwd = __xdp_map_lookup_elem(map, index);
3486         if (unlikely(!fwd)) {
3487                 err = -EINVAL;
3488                 goto err;
3489         }
3490         if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3491                 xdp_do_flush_map();
3492
3493         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3494         if (unlikely(err))
3495                 goto err;
3496
3497         ri->map_to_flush = map;
3498         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3499         return 0;
3500 err:
3501         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3502         return err;
3503 }
3504
3505 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3506                     struct bpf_prog *xdp_prog)
3507 {
3508         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3509         struct bpf_map *map = READ_ONCE(ri->map);
3510
3511         if (likely(map))
3512                 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3513
3514         return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3515 }
3516 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3517
3518 static int xdp_do_generic_redirect_map(struct net_device *dev,
3519                                        struct sk_buff *skb,
3520                                        struct xdp_buff *xdp,
3521                                        struct bpf_prog *xdp_prog,
3522                                        struct bpf_map *map)
3523 {
3524         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3525         u32 index = ri->ifindex;
3526         void *fwd = NULL;
3527         int err = 0;
3528
3529         ri->ifindex = 0;
3530         WRITE_ONCE(ri->map, NULL);
3531
3532         fwd = __xdp_map_lookup_elem(map, index);
3533         if (unlikely(!fwd)) {
3534                 err = -EINVAL;
3535                 goto err;
3536         }
3537
3538         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3539                 struct bpf_dtab_netdev *dst = fwd;
3540
3541                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3542                 if (unlikely(err))
3543                         goto err;
3544         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3545                 struct xdp_sock *xs = fwd;
3546
3547                 err = xsk_generic_rcv(xs, xdp);
3548                 if (err)
3549                         goto err;
3550                 consume_skb(skb);
3551         } else {
3552                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3553                 err = -EBADRQC;
3554                 goto err;
3555         }
3556
3557         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3558         return 0;
3559 err:
3560         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3561         return err;
3562 }
3563
3564 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3565                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3566 {
3567         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3568         struct bpf_map *map = READ_ONCE(ri->map);
3569         u32 index = ri->ifindex;
3570         struct net_device *fwd;
3571         int err = 0;
3572
3573         if (map)
3574                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3575                                                    map);
3576         ri->ifindex = 0;
3577         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3578         if (unlikely(!fwd)) {
3579                 err = -EINVAL;
3580                 goto err;
3581         }
3582
3583         err = xdp_ok_fwd_dev(fwd, skb->len);
3584         if (unlikely(err))
3585                 goto err;
3586
3587         skb->dev = fwd;
3588         _trace_xdp_redirect(dev, xdp_prog, index);
3589         generic_xdp_tx(skb, xdp_prog);
3590         return 0;
3591 err:
3592         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3593         return err;
3594 }
3595 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3596
3597 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3598 {
3599         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3600
3601         if (unlikely(flags))
3602                 return XDP_ABORTED;
3603
3604         ri->ifindex = ifindex;
3605         ri->flags = flags;
3606         WRITE_ONCE(ri->map, NULL);
3607
3608         return XDP_REDIRECT;
3609 }
3610
3611 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3612         .func           = bpf_xdp_redirect,
3613         .gpl_only       = false,
3614         .ret_type       = RET_INTEGER,
3615         .arg1_type      = ARG_ANYTHING,
3616         .arg2_type      = ARG_ANYTHING,
3617 };
3618
3619 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3620            u64, flags)
3621 {
3622         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3623
3624         if (unlikely(flags))
3625                 return XDP_ABORTED;
3626
3627         ri->ifindex = ifindex;
3628         ri->flags = flags;
3629         WRITE_ONCE(ri->map, map);
3630
3631         return XDP_REDIRECT;
3632 }
3633
3634 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3635         .func           = bpf_xdp_redirect_map,
3636         .gpl_only       = false,
3637         .ret_type       = RET_INTEGER,
3638         .arg1_type      = ARG_CONST_MAP_PTR,
3639         .arg2_type      = ARG_ANYTHING,
3640         .arg3_type      = ARG_ANYTHING,
3641 };
3642
3643 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3644                                   unsigned long off, unsigned long len)
3645 {
3646         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3647
3648         if (unlikely(!ptr))
3649                 return len;
3650         if (ptr != dst_buff)
3651                 memcpy(dst_buff, ptr, len);
3652
3653         return 0;
3654 }
3655
3656 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3657            u64, flags, void *, meta, u64, meta_size)
3658 {
3659         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3660
3661         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3662                 return -EINVAL;
3663         if (unlikely(skb_size > skb->len))
3664                 return -EFAULT;
3665
3666         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3667                                 bpf_skb_copy);
3668 }
3669
3670 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3671         .func           = bpf_skb_event_output,
3672         .gpl_only       = true,
3673         .ret_type       = RET_INTEGER,
3674         .arg1_type      = ARG_PTR_TO_CTX,
3675         .arg2_type      = ARG_CONST_MAP_PTR,
3676         .arg3_type      = ARG_ANYTHING,
3677         .arg4_type      = ARG_PTR_TO_MEM,
3678         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3679 };
3680
3681 static unsigned short bpf_tunnel_key_af(u64 flags)
3682 {
3683         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3684 }
3685
3686 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3687            u32, size, u64, flags)
3688 {
3689         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3690         u8 compat[sizeof(struct bpf_tunnel_key)];
3691         void *to_orig = to;
3692         int err;
3693
3694         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3695                 err = -EINVAL;
3696                 goto err_clear;
3697         }
3698         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3699                 err = -EPROTO;
3700                 goto err_clear;
3701         }
3702         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3703                 err = -EINVAL;
3704                 switch (size) {
3705                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3706                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3707                         goto set_compat;
3708                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3709                         /* Fixup deprecated structure layouts here, so we have
3710                          * a common path later on.
3711                          */
3712                         if (ip_tunnel_info_af(info) != AF_INET)
3713                                 goto err_clear;
3714 set_compat:
3715                         to = (struct bpf_tunnel_key *)compat;
3716                         break;
3717                 default:
3718                         goto err_clear;
3719                 }
3720         }
3721
3722         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3723         to->tunnel_tos = info->key.tos;
3724         to->tunnel_ttl = info->key.ttl;
3725         to->tunnel_ext = 0;
3726
3727         if (flags & BPF_F_TUNINFO_IPV6) {
3728                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3729                        sizeof(to->remote_ipv6));
3730                 to->tunnel_label = be32_to_cpu(info->key.label);
3731         } else {
3732                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3733                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3734                 to->tunnel_label = 0;
3735         }
3736
3737         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3738                 memcpy(to_orig, to, size);
3739
3740         return 0;
3741 err_clear:
3742         memset(to_orig, 0, size);
3743         return err;
3744 }
3745
3746 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3747         .func           = bpf_skb_get_tunnel_key,
3748         .gpl_only       = false,
3749         .ret_type       = RET_INTEGER,
3750         .arg1_type      = ARG_PTR_TO_CTX,
3751         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3752         .arg3_type      = ARG_CONST_SIZE,
3753         .arg4_type      = ARG_ANYTHING,
3754 };
3755
3756 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3757 {
3758         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3759         int err;
3760
3761         if (unlikely(!info ||
3762                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3763                 err = -ENOENT;
3764                 goto err_clear;
3765         }
3766         if (unlikely(size < info->options_len)) {
3767                 err = -ENOMEM;
3768                 goto err_clear;
3769         }
3770
3771         ip_tunnel_info_opts_get(to, info);
3772         if (size > info->options_len)
3773                 memset(to + info->options_len, 0, size - info->options_len);
3774
3775         return info->options_len;
3776 err_clear:
3777         memset(to, 0, size);
3778         return err;
3779 }
3780
3781 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3782         .func           = bpf_skb_get_tunnel_opt,
3783         .gpl_only       = false,
3784         .ret_type       = RET_INTEGER,
3785         .arg1_type      = ARG_PTR_TO_CTX,
3786         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3787         .arg3_type      = ARG_CONST_SIZE,
3788 };
3789
3790 static struct metadata_dst __percpu *md_dst;
3791
3792 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3793            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3794 {
3795         struct metadata_dst *md = this_cpu_ptr(md_dst);
3796         u8 compat[sizeof(struct bpf_tunnel_key)];
3797         struct ip_tunnel_info *info;
3798
3799         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3800                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3801                 return -EINVAL;
3802         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3803                 switch (size) {
3804                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3805                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3806                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3807                         /* Fixup deprecated structure layouts here, so we have
3808                          * a common path later on.
3809                          */
3810                         memcpy(compat, from, size);
3811                         memset(compat + size, 0, sizeof(compat) - size);
3812                         from = (const struct bpf_tunnel_key *) compat;
3813                         break;
3814                 default:
3815                         return -EINVAL;
3816                 }
3817         }
3818         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3819                      from->tunnel_ext))
3820                 return -EINVAL;
3821
3822         skb_dst_drop(skb);
3823         dst_hold((struct dst_entry *) md);
3824         skb_dst_set(skb, (struct dst_entry *) md);
3825
3826         info = &md->u.tun_info;
3827         memset(info, 0, sizeof(*info));
3828         info->mode = IP_TUNNEL_INFO_TX;
3829
3830         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3831         if (flags & BPF_F_DONT_FRAGMENT)
3832                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3833         if (flags & BPF_F_ZERO_CSUM_TX)
3834                 info->key.tun_flags &= ~TUNNEL_CSUM;
3835         if (flags & BPF_F_SEQ_NUMBER)
3836                 info->key.tun_flags |= TUNNEL_SEQ;
3837
3838         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3839         info->key.tos = from->tunnel_tos;
3840         info->key.ttl = from->tunnel_ttl;
3841
3842         if (flags & BPF_F_TUNINFO_IPV6) {
3843                 info->mode |= IP_TUNNEL_INFO_IPV6;
3844                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3845                        sizeof(from->remote_ipv6));
3846                 info->key.label = cpu_to_be32(from->tunnel_label) &
3847                                   IPV6_FLOWLABEL_MASK;
3848         } else {
3849                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3850         }
3851
3852         return 0;
3853 }
3854
3855 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3856         .func           = bpf_skb_set_tunnel_key,
3857         .gpl_only       = false,
3858         .ret_type       = RET_INTEGER,
3859         .arg1_type      = ARG_PTR_TO_CTX,
3860         .arg2_type      = ARG_PTR_TO_MEM,
3861         .arg3_type      = ARG_CONST_SIZE,
3862         .arg4_type      = ARG_ANYTHING,
3863 };
3864
3865 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3866            const u8 *, from, u32, size)
3867 {
3868         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3869         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3870
3871         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3872                 return -EINVAL;
3873         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3874                 return -ENOMEM;
3875
3876         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3877
3878         return 0;
3879 }
3880
3881 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3882         .func           = bpf_skb_set_tunnel_opt,
3883         .gpl_only       = false,
3884         .ret_type       = RET_INTEGER,
3885         .arg1_type      = ARG_PTR_TO_CTX,
3886         .arg2_type      = ARG_PTR_TO_MEM,
3887         .arg3_type      = ARG_CONST_SIZE,
3888 };
3889
3890 static const struct bpf_func_proto *
3891 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3892 {
3893         if (!md_dst) {
3894                 struct metadata_dst __percpu *tmp;
3895
3896                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3897                                                 METADATA_IP_TUNNEL,
3898                                                 GFP_KERNEL);
3899                 if (!tmp)
3900                         return NULL;
3901                 if (cmpxchg(&md_dst, NULL, tmp))
3902                         metadata_dst_free_percpu(tmp);
3903         }
3904
3905         switch (which) {
3906         case BPF_FUNC_skb_set_tunnel_key:
3907                 return &bpf_skb_set_tunnel_key_proto;
3908         case BPF_FUNC_skb_set_tunnel_opt:
3909                 return &bpf_skb_set_tunnel_opt_proto;
3910         default:
3911                 return NULL;
3912         }
3913 }
3914
3915 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3916            u32, idx)
3917 {
3918         struct bpf_array *array = container_of(map, struct bpf_array, map);
3919         struct cgroup *cgrp;
3920         struct sock *sk;
3921
3922         sk = skb_to_full_sk(skb);
3923         if (!sk || !sk_fullsock(sk))
3924                 return -ENOENT;
3925         if (unlikely(idx >= array->map.max_entries))
3926                 return -E2BIG;
3927
3928         cgrp = READ_ONCE(array->ptrs[idx]);
3929         if (unlikely(!cgrp))
3930                 return -EAGAIN;
3931
3932         return sk_under_cgroup_hierarchy(sk, cgrp);
3933 }
3934
3935 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3936         .func           = bpf_skb_under_cgroup,
3937         .gpl_only       = false,
3938         .ret_type       = RET_INTEGER,
3939         .arg1_type      = ARG_PTR_TO_CTX,
3940         .arg2_type      = ARG_CONST_MAP_PTR,
3941         .arg3_type      = ARG_ANYTHING,
3942 };
3943
3944 #ifdef CONFIG_SOCK_CGROUP_DATA
3945 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3946 {
3947         struct sock *sk = skb_to_full_sk(skb);
3948         struct cgroup *cgrp;
3949
3950         if (!sk || !sk_fullsock(sk))
3951                 return 0;
3952
3953         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3954         return cgrp->kn->id.id;
3955 }
3956
3957 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3958         .func           = bpf_skb_cgroup_id,
3959         .gpl_only       = false,
3960         .ret_type       = RET_INTEGER,
3961         .arg1_type      = ARG_PTR_TO_CTX,
3962 };
3963
3964 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3965            ancestor_level)
3966 {
3967         struct sock *sk = skb_to_full_sk(skb);
3968         struct cgroup *ancestor;
3969         struct cgroup *cgrp;
3970
3971         if (!sk || !sk_fullsock(sk))
3972                 return 0;
3973
3974         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3975         ancestor = cgroup_ancestor(cgrp, ancestor_level);
3976         if (!ancestor)
3977                 return 0;
3978
3979         return ancestor->kn->id.id;
3980 }
3981
3982 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3983         .func           = bpf_skb_ancestor_cgroup_id,
3984         .gpl_only       = false,
3985         .ret_type       = RET_INTEGER,
3986         .arg1_type      = ARG_PTR_TO_CTX,
3987         .arg2_type      = ARG_ANYTHING,
3988 };
3989 #endif
3990
3991 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3992                                   unsigned long off, unsigned long len)
3993 {
3994         memcpy(dst_buff, src_buff + off, len);
3995         return 0;
3996 }
3997
3998 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3999            u64, flags, void *, meta, u64, meta_size)
4000 {
4001         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4002
4003         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4004                 return -EINVAL;
4005         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4006                 return -EFAULT;
4007
4008         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4009                                 xdp_size, bpf_xdp_copy);
4010 }
4011
4012 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4013         .func           = bpf_xdp_event_output,
4014         .gpl_only       = true,
4015         .ret_type       = RET_INTEGER,
4016         .arg1_type      = ARG_PTR_TO_CTX,
4017         .arg2_type      = ARG_CONST_MAP_PTR,
4018         .arg3_type      = ARG_ANYTHING,
4019         .arg4_type      = ARG_PTR_TO_MEM,
4020         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4021 };
4022
4023 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4024 {
4025         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4026 }
4027
4028 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4029         .func           = bpf_get_socket_cookie,
4030         .gpl_only       = false,
4031         .ret_type       = RET_INTEGER,
4032         .arg1_type      = ARG_PTR_TO_CTX,
4033 };
4034
4035 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4036 {
4037         return sock_gen_cookie(ctx->sk);
4038 }
4039
4040 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4041         .func           = bpf_get_socket_cookie_sock_addr,
4042         .gpl_only       = false,
4043         .ret_type       = RET_INTEGER,
4044         .arg1_type      = ARG_PTR_TO_CTX,
4045 };
4046
4047 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4048 {
4049         return sock_gen_cookie(ctx->sk);
4050 }
4051
4052 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4053         .func           = bpf_get_socket_cookie_sock_ops,
4054         .gpl_only       = false,
4055         .ret_type       = RET_INTEGER,
4056         .arg1_type      = ARG_PTR_TO_CTX,
4057 };
4058
4059 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4060 {
4061         struct sock *sk = sk_to_full_sk(skb->sk);
4062         kuid_t kuid;
4063
4064         if (!sk || !sk_fullsock(sk))
4065                 return overflowuid;
4066         kuid = sock_net_uid(sock_net(sk), sk);
4067         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4068 }
4069
4070 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4071         .func           = bpf_get_socket_uid,
4072         .gpl_only       = false,
4073         .ret_type       = RET_INTEGER,
4074         .arg1_type      = ARG_PTR_TO_CTX,
4075 };
4076
4077 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4078            struct bpf_map *, map, u64, flags, void *, data, u64, size)
4079 {
4080         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4081                 return -EINVAL;
4082
4083         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4084 }
4085
4086 static const struct bpf_func_proto bpf_sockopt_event_output_proto =  {
4087         .func           = bpf_sockopt_event_output,
4088         .gpl_only       = true,
4089         .ret_type       = RET_INTEGER,
4090         .arg1_type      = ARG_PTR_TO_CTX,
4091         .arg2_type      = ARG_CONST_MAP_PTR,
4092         .arg3_type      = ARG_ANYTHING,
4093         .arg4_type      = ARG_PTR_TO_MEM,
4094         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4095 };
4096
4097 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4098            int, level, int, optname, char *, optval, int, optlen)
4099 {
4100         struct sock *sk = bpf_sock->sk;
4101         int ret = 0;
4102         int val;
4103
4104         if (!sk_fullsock(sk))
4105                 return -EINVAL;
4106
4107         if (level == SOL_SOCKET) {
4108                 if (optlen != sizeof(int))
4109                         return -EINVAL;
4110                 val = *((int *)optval);
4111
4112                 /* Only some socketops are supported */
4113                 switch (optname) {
4114                 case SO_RCVBUF:
4115                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4116                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
4117                         break;
4118                 case SO_SNDBUF:
4119                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4120                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
4121                         break;
4122                 case SO_MAX_PACING_RATE: /* 32bit version */
4123                         if (val != ~0U)
4124                                 cmpxchg(&sk->sk_pacing_status,
4125                                         SK_PACING_NONE,
4126                                         SK_PACING_NEEDED);
4127                         sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4128                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4129                                                  sk->sk_max_pacing_rate);
4130                         break;
4131                 case SO_PRIORITY:
4132                         sk->sk_priority = val;
4133                         break;
4134                 case SO_RCVLOWAT:
4135                         if (val < 0)
4136                                 val = INT_MAX;
4137                         sk->sk_rcvlowat = val ? : 1;
4138                         break;
4139                 case SO_MARK:
4140                         if (sk->sk_mark != val) {
4141                                 sk->sk_mark = val;
4142                                 sk_dst_reset(sk);
4143                         }
4144                         break;
4145                 default:
4146                         ret = -EINVAL;
4147                 }
4148 #ifdef CONFIG_INET
4149         } else if (level == SOL_IP) {
4150                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4151                         return -EINVAL;
4152
4153                 val = *((int *)optval);
4154                 /* Only some options are supported */
4155                 switch (optname) {
4156                 case IP_TOS:
4157                         if (val < -1 || val > 0xff) {
4158                                 ret = -EINVAL;
4159                         } else {
4160                                 struct inet_sock *inet = inet_sk(sk);
4161
4162                                 if (val == -1)
4163                                         val = 0;
4164                                 inet->tos = val;
4165                         }
4166                         break;
4167                 default:
4168                         ret = -EINVAL;
4169                 }
4170 #if IS_ENABLED(CONFIG_IPV6)
4171         } else if (level == SOL_IPV6) {
4172                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4173                         return -EINVAL;
4174
4175                 val = *((int *)optval);
4176                 /* Only some options are supported */
4177                 switch (optname) {
4178                 case IPV6_TCLASS:
4179                         if (val < -1 || val > 0xff) {
4180                                 ret = -EINVAL;
4181                         } else {
4182                                 struct ipv6_pinfo *np = inet6_sk(sk);
4183
4184                                 if (val == -1)
4185                                         val = 0;
4186                                 np->tclass = val;
4187                         }
4188                         break;
4189                 default:
4190                         ret = -EINVAL;
4191                 }
4192 #endif
4193         } else if (level == SOL_TCP &&
4194                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4195                 if (optname == TCP_CONGESTION) {
4196                         char name[TCP_CA_NAME_MAX];
4197                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4198
4199                         strncpy(name, optval, min_t(long, optlen,
4200                                                     TCP_CA_NAME_MAX-1));
4201                         name[TCP_CA_NAME_MAX-1] = 0;
4202                         ret = tcp_set_congestion_control(sk, name, false,
4203                                                          reinit);
4204                 } else {
4205                         struct tcp_sock *tp = tcp_sk(sk);
4206
4207                         if (optlen != sizeof(int))
4208                                 return -EINVAL;
4209
4210                         val = *((int *)optval);
4211                         /* Only some options are supported */
4212                         switch (optname) {
4213                         case TCP_BPF_IW:
4214                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4215                                         ret = -EINVAL;
4216                                 else
4217                                         tp->snd_cwnd = val;
4218                                 break;
4219                         case TCP_BPF_SNDCWND_CLAMP:
4220                                 if (val <= 0) {
4221                                         ret = -EINVAL;
4222                                 } else {
4223                                         tp->snd_cwnd_clamp = val;
4224                                         tp->snd_ssthresh = val;
4225                                 }
4226                                 break;
4227                         case TCP_SAVE_SYN:
4228                                 if (val < 0 || val > 1)
4229                                         ret = -EINVAL;
4230                                 else
4231                                         tp->save_syn = val;
4232                                 break;
4233                         default:
4234                                 ret = -EINVAL;
4235                         }
4236                 }
4237 #endif
4238         } else {
4239                 ret = -EINVAL;
4240         }
4241         return ret;
4242 }
4243
4244 static const struct bpf_func_proto bpf_setsockopt_proto = {
4245         .func           = bpf_setsockopt,
4246         .gpl_only       = false,
4247         .ret_type       = RET_INTEGER,
4248         .arg1_type      = ARG_PTR_TO_CTX,
4249         .arg2_type      = ARG_ANYTHING,
4250         .arg3_type      = ARG_ANYTHING,
4251         .arg4_type      = ARG_PTR_TO_MEM,
4252         .arg5_type      = ARG_CONST_SIZE,
4253 };
4254
4255 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4256            int, level, int, optname, char *, optval, int, optlen)
4257 {
4258         struct sock *sk = bpf_sock->sk;
4259
4260         if (!sk_fullsock(sk))
4261                 goto err_clear;
4262 #ifdef CONFIG_INET
4263         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4264                 struct inet_connection_sock *icsk;
4265                 struct tcp_sock *tp;
4266
4267                 switch (optname) {
4268                 case TCP_CONGESTION:
4269                         icsk = inet_csk(sk);
4270
4271                         if (!icsk->icsk_ca_ops || optlen <= 1)
4272                                 goto err_clear;
4273                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4274                         optval[optlen - 1] = 0;
4275                         break;
4276                 case TCP_SAVED_SYN:
4277                         tp = tcp_sk(sk);
4278
4279                         if (optlen <= 0 || !tp->saved_syn ||
4280                             optlen > tp->saved_syn[0])
4281                                 goto err_clear;
4282                         memcpy(optval, tp->saved_syn + 1, optlen);
4283                         break;
4284                 default:
4285                         goto err_clear;
4286                 }
4287         } else if (level == SOL_IP) {
4288                 struct inet_sock *inet = inet_sk(sk);
4289
4290                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4291                         goto err_clear;
4292
4293                 /* Only some options are supported */
4294                 switch (optname) {
4295                 case IP_TOS:
4296                         *((int *)optval) = (int)inet->tos;
4297                         break;
4298                 default:
4299                         goto err_clear;
4300                 }
4301 #if IS_ENABLED(CONFIG_IPV6)
4302         } else if (level == SOL_IPV6) {
4303                 struct ipv6_pinfo *np = inet6_sk(sk);
4304
4305                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4306                         goto err_clear;
4307
4308                 /* Only some options are supported */
4309                 switch (optname) {
4310                 case IPV6_TCLASS:
4311                         *((int *)optval) = (int)np->tclass;
4312                         break;
4313                 default:
4314                         goto err_clear;
4315                 }
4316 #endif
4317         } else {
4318                 goto err_clear;
4319         }
4320         return 0;
4321 #endif
4322 err_clear:
4323         memset(optval, 0, optlen);
4324         return -EINVAL;
4325 }
4326
4327 static const struct bpf_func_proto bpf_getsockopt_proto = {
4328         .func           = bpf_getsockopt,
4329         .gpl_only       = false,
4330         .ret_type       = RET_INTEGER,
4331         .arg1_type      = ARG_PTR_TO_CTX,
4332         .arg2_type      = ARG_ANYTHING,
4333         .arg3_type      = ARG_ANYTHING,
4334         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4335         .arg5_type      = ARG_CONST_SIZE,
4336 };
4337
4338 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4339            int, argval)
4340 {
4341         struct sock *sk = bpf_sock->sk;
4342         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4343
4344         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4345                 return -EINVAL;
4346
4347         if (val)
4348                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4349
4350         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4351 }
4352
4353 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4354         .func           = bpf_sock_ops_cb_flags_set,
4355         .gpl_only       = false,
4356         .ret_type       = RET_INTEGER,
4357         .arg1_type      = ARG_PTR_TO_CTX,
4358         .arg2_type      = ARG_ANYTHING,
4359 };
4360
4361 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4362 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4363
4364 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4365            int, addr_len)
4366 {
4367 #ifdef CONFIG_INET
4368         struct sock *sk = ctx->sk;
4369         int err;
4370
4371         /* Binding to port can be expensive so it's prohibited in the helper.
4372          * Only binding to IP is supported.
4373          */
4374         err = -EINVAL;
4375         if (addr->sa_family == AF_INET) {
4376                 if (addr_len < sizeof(struct sockaddr_in))
4377                         return err;
4378                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4379                         return err;
4380                 return __inet_bind(sk, addr, addr_len, true, false);
4381 #if IS_ENABLED(CONFIG_IPV6)
4382         } else if (addr->sa_family == AF_INET6) {
4383                 if (addr_len < SIN6_LEN_RFC2133)
4384                         return err;
4385                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4386                         return err;
4387                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4388                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4389                  */
4390                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4391 #endif /* CONFIG_IPV6 */
4392         }
4393 #endif /* CONFIG_INET */
4394
4395         return -EAFNOSUPPORT;
4396 }
4397
4398 static const struct bpf_func_proto bpf_bind_proto = {
4399         .func           = bpf_bind,
4400         .gpl_only       = false,
4401         .ret_type       = RET_INTEGER,
4402         .arg1_type      = ARG_PTR_TO_CTX,
4403         .arg2_type      = ARG_PTR_TO_MEM,
4404         .arg3_type      = ARG_CONST_SIZE,
4405 };
4406
4407 #ifdef CONFIG_XFRM
4408 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4409            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4410 {
4411         const struct sec_path *sp = skb_sec_path(skb);
4412         const struct xfrm_state *x;
4413
4414         if (!sp || unlikely(index >= sp->len || flags))
4415                 goto err_clear;
4416
4417         x = sp->xvec[index];
4418
4419         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4420                 goto err_clear;
4421
4422         to->reqid = x->props.reqid;
4423         to->spi = x->id.spi;
4424         to->family = x->props.family;
4425         to->ext = 0;
4426
4427         if (to->family == AF_INET6) {
4428                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4429                        sizeof(to->remote_ipv6));
4430         } else {
4431                 to->remote_ipv4 = x->props.saddr.a4;
4432                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4433         }
4434
4435         return 0;
4436 err_clear:
4437         memset(to, 0, size);
4438         return -EINVAL;
4439 }
4440
4441 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4442         .func           = bpf_skb_get_xfrm_state,
4443         .gpl_only       = false,
4444         .ret_type       = RET_INTEGER,
4445         .arg1_type      = ARG_PTR_TO_CTX,
4446         .arg2_type      = ARG_ANYTHING,
4447         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4448         .arg4_type      = ARG_CONST_SIZE,
4449         .arg5_type      = ARG_ANYTHING,
4450 };
4451 #endif
4452
4453 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4454 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4455                                   const struct neighbour *neigh,
4456                                   const struct net_device *dev)
4457 {
4458         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4459         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4460         params->h_vlan_TCI = 0;
4461         params->h_vlan_proto = 0;
4462         params->ifindex = dev->ifindex;
4463
4464         return 0;
4465 }
4466 #endif
4467
4468 #if IS_ENABLED(CONFIG_INET)
4469 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4470                                u32 flags, bool check_mtu)
4471 {
4472         struct in_device *in_dev;
4473         struct neighbour *neigh;
4474         struct net_device *dev;
4475         struct fib_result res;
4476         struct fib_nh *nh;
4477         struct flowi4 fl4;
4478         int err;
4479         u32 mtu;
4480
4481         dev = dev_get_by_index_rcu(net, params->ifindex);
4482         if (unlikely(!dev))
4483                 return -ENODEV;
4484
4485         /* verify forwarding is enabled on this interface */
4486         in_dev = __in_dev_get_rcu(dev);
4487         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4488                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4489
4490         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4491                 fl4.flowi4_iif = 1;
4492                 fl4.flowi4_oif = params->ifindex;
4493         } else {
4494                 fl4.flowi4_iif = params->ifindex;
4495                 fl4.flowi4_oif = 0;
4496         }
4497         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4498         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4499         fl4.flowi4_flags = 0;
4500
4501         fl4.flowi4_proto = params->l4_protocol;
4502         fl4.daddr = params->ipv4_dst;
4503         fl4.saddr = params->ipv4_src;
4504         fl4.fl4_sport = params->sport;
4505         fl4.fl4_dport = params->dport;
4506
4507         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4508                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4509                 struct fib_table *tb;
4510
4511                 tb = fib_get_table(net, tbid);
4512                 if (unlikely(!tb))
4513                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4514
4515                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4516         } else {
4517                 fl4.flowi4_mark = 0;
4518                 fl4.flowi4_secid = 0;
4519                 fl4.flowi4_tun_key.tun_id = 0;
4520                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4521
4522                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4523         }
4524
4525         if (err) {
4526                 /* map fib lookup errors to RTN_ type */
4527                 if (err == -EINVAL)
4528                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4529                 if (err == -EHOSTUNREACH)
4530                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4531                 if (err == -EACCES)
4532                         return BPF_FIB_LKUP_RET_PROHIBIT;
4533
4534                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4535         }
4536
4537         if (res.type != RTN_UNICAST)
4538                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4539
4540         if (res.fi->fib_nhs > 1)
4541                 fib_select_path(net, &res, &fl4, NULL);
4542
4543         if (check_mtu) {
4544                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4545                 if (params->tot_len > mtu)
4546                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4547         }
4548
4549         nh = &res.fi->fib_nh[res.nh_sel];
4550
4551         /* do not handle lwt encaps right now */
4552         if (nh->nh_lwtstate)
4553                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4554
4555         dev = nh->nh_dev;
4556         if (nh->nh_gw)
4557                 params->ipv4_dst = nh->nh_gw;
4558
4559         params->rt_metric = res.fi->fib_priority;
4560
4561         /* xdp and cls_bpf programs are run in RCU-bh so
4562          * rcu_read_lock_bh is not needed here
4563          */
4564         neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4565         if (!neigh)
4566                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4567
4568         return bpf_fib_set_fwd_params(params, neigh, dev);
4569 }
4570 #endif
4571
4572 #if IS_ENABLED(CONFIG_IPV6)
4573 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4574                                u32 flags, bool check_mtu)
4575 {
4576         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4577         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4578         struct neighbour *neigh;
4579         struct net_device *dev;
4580         struct inet6_dev *idev;
4581         struct fib6_info *f6i;
4582         struct flowi6 fl6;
4583         int strict = 0;
4584         int oif;
4585         u32 mtu;
4586
4587         /* link local addresses are never forwarded */
4588         if (rt6_need_strict(dst) || rt6_need_strict(src))
4589                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4590
4591         dev = dev_get_by_index_rcu(net, params->ifindex);
4592         if (unlikely(!dev))
4593                 return -ENODEV;
4594
4595         idev = __in6_dev_get_safely(dev);
4596         if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4597                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4598
4599         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4600                 fl6.flowi6_iif = 1;
4601                 oif = fl6.flowi6_oif = params->ifindex;
4602         } else {
4603                 oif = fl6.flowi6_iif = params->ifindex;
4604                 fl6.flowi6_oif = 0;
4605                 strict = RT6_LOOKUP_F_HAS_SADDR;
4606         }
4607         fl6.flowlabel = params->flowinfo;
4608         fl6.flowi6_scope = 0;
4609         fl6.flowi6_flags = 0;
4610         fl6.mp_hash = 0;
4611
4612         fl6.flowi6_proto = params->l4_protocol;
4613         fl6.daddr = *dst;
4614         fl6.saddr = *src;
4615         fl6.fl6_sport = params->sport;
4616         fl6.fl6_dport = params->dport;
4617
4618         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4619                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4620                 struct fib6_table *tb;
4621
4622                 tb = ipv6_stub->fib6_get_table(net, tbid);
4623                 if (unlikely(!tb))
4624                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4625
4626                 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4627         } else {
4628                 fl6.flowi6_mark = 0;
4629                 fl6.flowi6_secid = 0;
4630                 fl6.flowi6_tun_key.tun_id = 0;
4631                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4632
4633                 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4634         }
4635
4636         if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4637                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4638
4639         if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4640                 switch (f6i->fib6_type) {
4641                 case RTN_BLACKHOLE:
4642                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4643                 case RTN_UNREACHABLE:
4644                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4645                 case RTN_PROHIBIT:
4646                         return BPF_FIB_LKUP_RET_PROHIBIT;
4647                 default:
4648                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4649                 }
4650         }
4651
4652         if (f6i->fib6_type != RTN_UNICAST)
4653                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4654
4655         if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4656                 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4657                                                        fl6.flowi6_oif, NULL,
4658                                                        strict);
4659
4660         if (check_mtu) {
4661                 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4662                 if (params->tot_len > mtu)
4663                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4664         }
4665
4666         if (f6i->fib6_nh.nh_lwtstate)
4667                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4668
4669         if (f6i->fib6_flags & RTF_GATEWAY)
4670                 *dst = f6i->fib6_nh.nh_gw;
4671
4672         dev = f6i->fib6_nh.nh_dev;
4673         params->rt_metric = f6i->fib6_metric;
4674
4675         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4676          * not needed here. Can not use __ipv6_neigh_lookup_noref here
4677          * because we need to get nd_tbl via the stub
4678          */
4679         neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4680                                       ndisc_hashfn, dst, dev);
4681         if (!neigh)
4682                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4683
4684         return bpf_fib_set_fwd_params(params, neigh, dev);
4685 }
4686 #endif
4687
4688 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4689            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4690 {
4691         if (plen < sizeof(*params))
4692                 return -EINVAL;
4693
4694         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4695                 return -EINVAL;
4696
4697         switch (params->family) {
4698 #if IS_ENABLED(CONFIG_INET)
4699         case AF_INET:
4700                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4701                                            flags, true);
4702 #endif
4703 #if IS_ENABLED(CONFIG_IPV6)
4704         case AF_INET6:
4705                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4706                                            flags, true);
4707 #endif
4708         }
4709         return -EAFNOSUPPORT;
4710 }
4711
4712 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4713         .func           = bpf_xdp_fib_lookup,
4714         .gpl_only       = true,
4715         .ret_type       = RET_INTEGER,
4716         .arg1_type      = ARG_PTR_TO_CTX,
4717         .arg2_type      = ARG_PTR_TO_MEM,
4718         .arg3_type      = ARG_CONST_SIZE,
4719         .arg4_type      = ARG_ANYTHING,
4720 };
4721
4722 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4723            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4724 {
4725         struct net *net = dev_net(skb->dev);
4726         int rc = -EAFNOSUPPORT;
4727
4728         if (plen < sizeof(*params))
4729                 return -EINVAL;
4730
4731         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4732                 return -EINVAL;
4733
4734         switch (params->family) {
4735 #if IS_ENABLED(CONFIG_INET)
4736         case AF_INET:
4737                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4738                 break;
4739 #endif
4740 #if IS_ENABLED(CONFIG_IPV6)
4741         case AF_INET6:
4742                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4743                 break;
4744 #endif
4745         }
4746
4747         if (!rc) {
4748                 struct net_device *dev;
4749
4750                 dev = dev_get_by_index_rcu(net, params->ifindex);
4751                 if (!is_skb_forwardable(dev, skb))
4752                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4753         }
4754
4755         return rc;
4756 }
4757
4758 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4759         .func           = bpf_skb_fib_lookup,
4760         .gpl_only       = true,
4761         .ret_type       = RET_INTEGER,
4762         .arg1_type      = ARG_PTR_TO_CTX,
4763         .arg2_type      = ARG_PTR_TO_MEM,
4764         .arg3_type      = ARG_CONST_SIZE,
4765         .arg4_type      = ARG_ANYTHING,
4766 };
4767
4768 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4769 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4770 {
4771         int err;
4772         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4773
4774         if (!seg6_validate_srh(srh, len))
4775                 return -EINVAL;
4776
4777         switch (type) {
4778         case BPF_LWT_ENCAP_SEG6_INLINE:
4779                 if (skb->protocol != htons(ETH_P_IPV6))
4780                         return -EBADMSG;
4781
4782                 err = seg6_do_srh_inline(skb, srh);
4783                 break;
4784         case BPF_LWT_ENCAP_SEG6:
4785                 skb_reset_inner_headers(skb);
4786                 skb->encapsulation = 1;
4787                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4788                 break;
4789         default:
4790                 return -EINVAL;
4791         }
4792
4793         bpf_compute_data_pointers(skb);
4794         if (err)
4795                 return err;
4796
4797         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4798         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4799
4800         return seg6_lookup_nexthop(skb, NULL, 0);
4801 }
4802 #endif /* CONFIG_IPV6_SEG6_BPF */
4803
4804 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4805            u32, len)
4806 {
4807         switch (type) {
4808 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4809         case BPF_LWT_ENCAP_SEG6:
4810         case BPF_LWT_ENCAP_SEG6_INLINE:
4811                 return bpf_push_seg6_encap(skb, type, hdr, len);
4812 #endif
4813         default:
4814                 return -EINVAL;
4815         }
4816 }
4817
4818 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4819         .func           = bpf_lwt_push_encap,
4820         .gpl_only       = false,
4821         .ret_type       = RET_INTEGER,
4822         .arg1_type      = ARG_PTR_TO_CTX,
4823         .arg2_type      = ARG_ANYTHING,
4824         .arg3_type      = ARG_PTR_TO_MEM,
4825         .arg4_type      = ARG_CONST_SIZE
4826 };
4827
4828 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4829 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4830            const void *, from, u32, len)
4831 {
4832         struct seg6_bpf_srh_state *srh_state =
4833                 this_cpu_ptr(&seg6_bpf_srh_states);
4834         struct ipv6_sr_hdr *srh = srh_state->srh;
4835         void *srh_tlvs, *srh_end, *ptr;
4836         int srhoff = 0;
4837
4838         if (srh == NULL)
4839                 return -EINVAL;
4840
4841         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4842         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4843
4844         ptr = skb->data + offset;
4845         if (ptr >= srh_tlvs && ptr + len <= srh_end)
4846                 srh_state->valid = false;
4847         else if (ptr < (void *)&srh->flags ||
4848                  ptr + len > (void *)&srh->segments)
4849                 return -EFAULT;
4850
4851         if (unlikely(bpf_try_make_writable(skb, offset + len)))
4852                 return -EFAULT;
4853         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4854                 return -EINVAL;
4855         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4856
4857         memcpy(skb->data + offset, from, len);
4858         return 0;
4859 }
4860
4861 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4862         .func           = bpf_lwt_seg6_store_bytes,
4863         .gpl_only       = false,
4864         .ret_type       = RET_INTEGER,
4865         .arg1_type      = ARG_PTR_TO_CTX,
4866         .arg2_type      = ARG_ANYTHING,
4867         .arg3_type      = ARG_PTR_TO_MEM,
4868         .arg4_type      = ARG_CONST_SIZE
4869 };
4870
4871 static void bpf_update_srh_state(struct sk_buff *skb)
4872 {
4873         struct seg6_bpf_srh_state *srh_state =
4874                 this_cpu_ptr(&seg6_bpf_srh_states);
4875         int srhoff = 0;
4876
4877         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4878                 srh_state->srh = NULL;
4879         } else {
4880                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4881                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4882                 srh_state->valid = true;
4883         }
4884 }
4885
4886 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4887            u32, action, void *, param, u32, param_len)
4888 {
4889         struct seg6_bpf_srh_state *srh_state =
4890                 this_cpu_ptr(&seg6_bpf_srh_states);
4891         int hdroff = 0;
4892         int err;
4893
4894         switch (action) {
4895         case SEG6_LOCAL_ACTION_END_X:
4896                 if (!seg6_bpf_has_valid_srh(skb))
4897                         return -EBADMSG;
4898                 if (param_len != sizeof(struct in6_addr))
4899                         return -EINVAL;
4900                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4901         case SEG6_LOCAL_ACTION_END_T:
4902                 if (!seg6_bpf_has_valid_srh(skb))
4903                         return -EBADMSG;
4904                 if (param_len != sizeof(int))
4905                         return -EINVAL;
4906                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4907         case SEG6_LOCAL_ACTION_END_DT6:
4908                 if (!seg6_bpf_has_valid_srh(skb))
4909                         return -EBADMSG;
4910                 if (param_len != sizeof(int))
4911                         return -EINVAL;
4912
4913                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4914                         return -EBADMSG;
4915                 if (!pskb_pull(skb, hdroff))
4916                         return -EBADMSG;
4917
4918                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4919                 skb_reset_network_header(skb);
4920                 skb_reset_transport_header(skb);
4921                 skb->encapsulation = 0;
4922
4923                 bpf_compute_data_pointers(skb);
4924                 bpf_update_srh_state(skb);
4925                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4926         case SEG6_LOCAL_ACTION_END_B6:
4927                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4928                         return -EBADMSG;
4929                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4930                                           param, param_len);
4931                 if (!err)
4932                         bpf_update_srh_state(skb);
4933
4934                 return err;
4935         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4936                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4937                         return -EBADMSG;
4938                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4939                                           param, param_len);
4940                 if (!err)
4941                         bpf_update_srh_state(skb);
4942
4943                 return err;
4944         default:
4945                 return -EINVAL;
4946         }
4947 }
4948
4949 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4950         .func           = bpf_lwt_seg6_action,
4951         .gpl_only       = false,
4952         .ret_type       = RET_INTEGER,
4953         .arg1_type      = ARG_PTR_TO_CTX,
4954         .arg2_type      = ARG_ANYTHING,
4955         .arg3_type      = ARG_PTR_TO_MEM,
4956         .arg4_type      = ARG_CONST_SIZE
4957 };
4958
4959 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4960            s32, len)
4961 {
4962         struct seg6_bpf_srh_state *srh_state =
4963                 this_cpu_ptr(&seg6_bpf_srh_states);
4964         struct ipv6_sr_hdr *srh = srh_state->srh;
4965         void *srh_end, *srh_tlvs, *ptr;
4966         struct ipv6hdr *hdr;
4967         int srhoff = 0;
4968         int ret;
4969
4970         if (unlikely(srh == NULL))
4971                 return -EINVAL;
4972
4973         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4974                         ((srh->first_segment + 1) << 4));
4975         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4976                         srh_state->hdrlen);
4977         ptr = skb->data + offset;
4978
4979         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4980                 return -EFAULT;
4981         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4982                 return -EFAULT;
4983
4984         if (len > 0) {
4985                 ret = skb_cow_head(skb, len);
4986                 if (unlikely(ret < 0))
4987                         return ret;
4988
4989                 ret = bpf_skb_net_hdr_push(skb, offset, len);
4990         } else {
4991                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4992         }
4993
4994         bpf_compute_data_pointers(skb);
4995         if (unlikely(ret < 0))
4996                 return ret;
4997
4998         hdr = (struct ipv6hdr *)skb->data;
4999         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5000
5001         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5002                 return -EINVAL;
5003         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5004         srh_state->hdrlen += len;
5005         srh_state->valid = false;
5006         return 0;
5007 }
5008
5009 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5010         .func           = bpf_lwt_seg6_adjust_srh,
5011         .gpl_only       = false,
5012         .ret_type       = RET_INTEGER,
5013         .arg1_type      = ARG_PTR_TO_CTX,
5014         .arg2_type      = ARG_ANYTHING,
5015         .arg3_type      = ARG_ANYTHING,
5016 };
5017 #endif /* CONFIG_IPV6_SEG6_BPF */
5018
5019 #ifdef CONFIG_INET
5020 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5021                               int dif, int sdif, u8 family, u8 proto)
5022 {
5023         bool refcounted = false;
5024         struct sock *sk = NULL;
5025
5026         if (family == AF_INET) {
5027                 __be32 src4 = tuple->ipv4.saddr;
5028                 __be32 dst4 = tuple->ipv4.daddr;
5029
5030                 if (proto == IPPROTO_TCP)
5031                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5032                                            src4, tuple->ipv4.sport,
5033                                            dst4, tuple->ipv4.dport,
5034                                            dif, sdif, &refcounted);
5035                 else
5036                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5037                                                dst4, tuple->ipv4.dport,
5038                                                dif, sdif, &udp_table, NULL);
5039 #if IS_ENABLED(CONFIG_IPV6)
5040         } else {
5041                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5042                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5043
5044                 if (proto == IPPROTO_TCP)
5045                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5046                                             src6, tuple->ipv6.sport,
5047                                             dst6, ntohs(tuple->ipv6.dport),
5048                                             dif, sdif, &refcounted);
5049                 else if (likely(ipv6_bpf_stub))
5050                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5051                                                             src6, tuple->ipv6.sport,
5052                                                             dst6, tuple->ipv6.dport,
5053                                                             dif, sdif,
5054                                                             &udp_table, NULL);
5055 #endif
5056         }
5057
5058         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5059                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5060                 sk = NULL;
5061         }
5062         return sk;
5063 }
5064
5065 /* bpf_sk_lookup performs the core lookup for different types of sockets,
5066  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5067  * Returns the socket as an 'unsigned long' to simplify the casting in the
5068  * callers to satisfy BPF_CALL declarations.
5069  */
5070 static unsigned long
5071 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5072                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5073                 u64 flags)
5074 {
5075         struct sock *sk = NULL;
5076         u8 family = AF_UNSPEC;
5077         struct net *net;
5078         int sdif;
5079
5080         family = len == sizeof(tuple->ipv4) ? AF_INET : AF_INET6;
5081         if (unlikely(family == AF_UNSPEC || flags ||
5082                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5083                 goto out;
5084
5085         if (family == AF_INET)
5086                 sdif = inet_sdif(skb);
5087         else
5088                 sdif = inet6_sdif(skb);
5089
5090         if ((s32)netns_id < 0) {
5091                 net = caller_net;
5092                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5093         } else {
5094                 net = get_net_ns_by_id(caller_net, netns_id);
5095                 if (unlikely(!net))
5096                         goto out;
5097                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5098                 put_net(net);
5099         }
5100
5101         if (sk)
5102                 sk = sk_to_full_sk(sk);
5103 out:
5104         return (unsigned long) sk;
5105 }
5106
5107 static unsigned long
5108 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5109               u8 proto, u64 netns_id, u64 flags)
5110 {
5111         struct net *caller_net;
5112         int ifindex;
5113
5114         if (skb->dev) {
5115                 caller_net = dev_net(skb->dev);
5116                 ifindex = skb->dev->ifindex;
5117         } else {
5118                 caller_net = sock_net(skb->sk);
5119                 ifindex = 0;
5120         }
5121
5122         return __bpf_sk_lookup(skb, tuple, len, caller_net, ifindex,
5123                               proto, netns_id, flags);
5124 }
5125
5126 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5127            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5128 {
5129         return bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, netns_id, flags);
5130 }
5131
5132 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5133         .func           = bpf_sk_lookup_tcp,
5134         .gpl_only       = false,
5135         .pkt_access     = true,
5136         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5137         .arg1_type      = ARG_PTR_TO_CTX,
5138         .arg2_type      = ARG_PTR_TO_MEM,
5139         .arg3_type      = ARG_CONST_SIZE,
5140         .arg4_type      = ARG_ANYTHING,
5141         .arg5_type      = ARG_ANYTHING,
5142 };
5143
5144 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5145            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5146 {
5147         return bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, netns_id, flags);
5148 }
5149
5150 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5151         .func           = bpf_sk_lookup_udp,
5152         .gpl_only       = false,
5153         .pkt_access     = true,
5154         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5155         .arg1_type      = ARG_PTR_TO_CTX,
5156         .arg2_type      = ARG_PTR_TO_MEM,
5157         .arg3_type      = ARG_CONST_SIZE,
5158         .arg4_type      = ARG_ANYTHING,
5159         .arg5_type      = ARG_ANYTHING,
5160 };
5161
5162 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5163 {
5164         if (!sock_flag(sk, SOCK_RCU_FREE))
5165                 sock_gen_put(sk);
5166         return 0;
5167 }
5168
5169 static const struct bpf_func_proto bpf_sk_release_proto = {
5170         .func           = bpf_sk_release,
5171         .gpl_only       = false,
5172         .ret_type       = RET_INTEGER,
5173         .arg1_type      = ARG_PTR_TO_SOCKET,
5174 };
5175
5176 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5177            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5178 {
5179         struct net *caller_net = dev_net(ctx->rxq->dev);
5180         int ifindex = ctx->rxq->dev->ifindex;
5181
5182         return __bpf_sk_lookup(NULL, tuple, len, caller_net, ifindex,
5183                               IPPROTO_UDP, netns_id, flags);
5184 }
5185
5186 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5187         .func           = bpf_xdp_sk_lookup_udp,
5188         .gpl_only       = false,
5189         .pkt_access     = true,
5190         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5191         .arg1_type      = ARG_PTR_TO_CTX,
5192         .arg2_type      = ARG_PTR_TO_MEM,
5193         .arg3_type      = ARG_CONST_SIZE,
5194         .arg4_type      = ARG_ANYTHING,
5195         .arg5_type      = ARG_ANYTHING,
5196 };
5197
5198 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5199            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5200 {
5201         struct net *caller_net = dev_net(ctx->rxq->dev);
5202         int ifindex = ctx->rxq->dev->ifindex;
5203
5204         return __bpf_sk_lookup(NULL, tuple, len, caller_net, ifindex,
5205                               IPPROTO_TCP, netns_id, flags);
5206 }
5207
5208 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5209         .func           = bpf_xdp_sk_lookup_tcp,
5210         .gpl_only       = false,
5211         .pkt_access     = true,
5212         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5213         .arg1_type      = ARG_PTR_TO_CTX,
5214         .arg2_type      = ARG_PTR_TO_MEM,
5215         .arg3_type      = ARG_CONST_SIZE,
5216         .arg4_type      = ARG_ANYTHING,
5217         .arg5_type      = ARG_ANYTHING,
5218 };
5219
5220 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5221            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5222 {
5223         return __bpf_sk_lookup(NULL, tuple, len, sock_net(ctx->sk), 0,
5224                                IPPROTO_TCP, netns_id, flags);
5225 }
5226
5227 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5228         .func           = bpf_sock_addr_sk_lookup_tcp,
5229         .gpl_only       = false,
5230         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5231         .arg1_type      = ARG_PTR_TO_CTX,
5232         .arg2_type      = ARG_PTR_TO_MEM,
5233         .arg3_type      = ARG_CONST_SIZE,
5234         .arg4_type      = ARG_ANYTHING,
5235         .arg5_type      = ARG_ANYTHING,
5236 };
5237
5238 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5239            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5240 {
5241         return __bpf_sk_lookup(NULL, tuple, len, sock_net(ctx->sk), 0,
5242                                IPPROTO_UDP, netns_id, flags);
5243 }
5244
5245 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5246         .func           = bpf_sock_addr_sk_lookup_udp,
5247         .gpl_only       = false,
5248         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5249         .arg1_type      = ARG_PTR_TO_CTX,
5250         .arg2_type      = ARG_PTR_TO_MEM,
5251         .arg3_type      = ARG_CONST_SIZE,
5252         .arg4_type      = ARG_ANYTHING,
5253         .arg5_type      = ARG_ANYTHING,
5254 };
5255
5256 #endif /* CONFIG_INET */
5257
5258 bool bpf_helper_changes_pkt_data(void *func)
5259 {
5260         if (func == bpf_skb_vlan_push ||
5261             func == bpf_skb_vlan_pop ||
5262             func == bpf_skb_store_bytes ||
5263             func == bpf_skb_change_proto ||
5264             func == bpf_skb_change_head ||
5265             func == sk_skb_change_head ||
5266             func == bpf_skb_change_tail ||
5267             func == sk_skb_change_tail ||
5268             func == bpf_skb_adjust_room ||
5269             func == bpf_skb_pull_data ||
5270             func == sk_skb_pull_data ||
5271             func == bpf_clone_redirect ||
5272             func == bpf_l3_csum_replace ||
5273             func == bpf_l4_csum_replace ||
5274             func == bpf_xdp_adjust_head ||
5275             func == bpf_xdp_adjust_meta ||
5276             func == bpf_msg_pull_data ||
5277             func == bpf_msg_push_data ||
5278             func == bpf_msg_pop_data ||
5279             func == bpf_xdp_adjust_tail ||
5280 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5281             func == bpf_lwt_seg6_store_bytes ||
5282             func == bpf_lwt_seg6_adjust_srh ||
5283             func == bpf_lwt_seg6_action ||
5284 #endif
5285             func == bpf_lwt_push_encap)
5286                 return true;
5287
5288         return false;
5289 }
5290
5291 static const struct bpf_func_proto *
5292 bpf_base_func_proto(enum bpf_func_id func_id)
5293 {
5294         switch (func_id) {
5295         case BPF_FUNC_map_lookup_elem:
5296                 return &bpf_map_lookup_elem_proto;
5297         case BPF_FUNC_map_update_elem:
5298                 return &bpf_map_update_elem_proto;
5299         case BPF_FUNC_map_delete_elem:
5300                 return &bpf_map_delete_elem_proto;
5301         case BPF_FUNC_map_push_elem:
5302                 return &bpf_map_push_elem_proto;
5303         case BPF_FUNC_map_pop_elem:
5304                 return &bpf_map_pop_elem_proto;
5305         case BPF_FUNC_map_peek_elem:
5306                 return &bpf_map_peek_elem_proto;
5307         case BPF_FUNC_get_prandom_u32:
5308                 return &bpf_get_prandom_u32_proto;
5309         case BPF_FUNC_get_smp_processor_id:
5310                 return &bpf_get_raw_smp_processor_id_proto;
5311         case BPF_FUNC_get_numa_node_id:
5312                 return &bpf_get_numa_node_id_proto;
5313         case BPF_FUNC_tail_call:
5314                 return &bpf_tail_call_proto;
5315         case BPF_FUNC_ktime_get_ns:
5316                 return &bpf_ktime_get_ns_proto;
5317         case BPF_FUNC_trace_printk:
5318                 if (capable(CAP_SYS_ADMIN))
5319                         return bpf_get_trace_printk_proto();
5320                 /* else, fall through */
5321         default:
5322                 return NULL;
5323         }
5324 }
5325
5326 static const struct bpf_func_proto *
5327 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5328 {
5329         switch (func_id) {
5330         /* inet and inet6 sockets are created in a process
5331          * context so there is always a valid uid/gid
5332          */
5333         case BPF_FUNC_get_current_uid_gid:
5334                 return &bpf_get_current_uid_gid_proto;
5335         case BPF_FUNC_get_local_storage:
5336                 return &bpf_get_local_storage_proto;
5337         default:
5338                 return bpf_base_func_proto(func_id);
5339         }
5340 }
5341
5342 static const struct bpf_func_proto *
5343 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5344 {
5345         switch (func_id) {
5346         /* inet and inet6 sockets are created in a process
5347          * context so there is always a valid uid/gid
5348          */
5349         case BPF_FUNC_get_current_uid_gid:
5350                 return &bpf_get_current_uid_gid_proto;
5351         case BPF_FUNC_bind:
5352                 switch (prog->expected_attach_type) {
5353                 case BPF_CGROUP_INET4_CONNECT:
5354                 case BPF_CGROUP_INET6_CONNECT:
5355                         return &bpf_bind_proto;
5356                 default:
5357                         return NULL;
5358                 }
5359         case BPF_FUNC_get_socket_cookie:
5360                 return &bpf_get_socket_cookie_sock_addr_proto;
5361         case BPF_FUNC_get_local_storage:
5362                 return &bpf_get_local_storage_proto;
5363 #ifdef CONFIG_INET
5364         case BPF_FUNC_sk_lookup_tcp:
5365                 return &bpf_sock_addr_sk_lookup_tcp_proto;
5366         case BPF_FUNC_sk_lookup_udp:
5367                 return &bpf_sock_addr_sk_lookup_udp_proto;
5368         case BPF_FUNC_sk_release:
5369                 return &bpf_sk_release_proto;
5370 #endif /* CONFIG_INET */
5371         default:
5372                 return bpf_base_func_proto(func_id);
5373         }
5374 }
5375
5376 static const struct bpf_func_proto *
5377 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5378 {
5379         switch (func_id) {
5380         case BPF_FUNC_skb_load_bytes:
5381                 return &bpf_skb_load_bytes_proto;
5382         case BPF_FUNC_skb_load_bytes_relative:
5383                 return &bpf_skb_load_bytes_relative_proto;
5384         case BPF_FUNC_get_socket_cookie:
5385                 return &bpf_get_socket_cookie_proto;
5386         case BPF_FUNC_get_socket_uid:
5387                 return &bpf_get_socket_uid_proto;
5388         default:
5389                 return bpf_base_func_proto(func_id);
5390         }
5391 }
5392
5393 static const struct bpf_func_proto *
5394 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5395 {
5396         switch (func_id) {
5397         case BPF_FUNC_get_local_storage:
5398                 return &bpf_get_local_storage_proto;
5399         default:
5400                 return sk_filter_func_proto(func_id, prog);
5401         }
5402 }
5403
5404 static const struct bpf_func_proto *
5405 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5406 {
5407         switch (func_id) {
5408         case BPF_FUNC_skb_store_bytes:
5409                 return &bpf_skb_store_bytes_proto;
5410         case BPF_FUNC_skb_load_bytes:
5411                 return &bpf_skb_load_bytes_proto;
5412         case BPF_FUNC_skb_load_bytes_relative:
5413                 return &bpf_skb_load_bytes_relative_proto;
5414         case BPF_FUNC_skb_pull_data:
5415                 return &bpf_skb_pull_data_proto;
5416         case BPF_FUNC_csum_diff:
5417                 return &bpf_csum_diff_proto;
5418         case BPF_FUNC_csum_update:
5419                 return &bpf_csum_update_proto;
5420         case BPF_FUNC_l3_csum_replace:
5421                 return &bpf_l3_csum_replace_proto;
5422         case BPF_FUNC_l4_csum_replace:
5423                 return &bpf_l4_csum_replace_proto;
5424         case BPF_FUNC_clone_redirect:
5425                 return &bpf_clone_redirect_proto;
5426         case BPF_FUNC_get_cgroup_classid:
5427                 return &bpf_get_cgroup_classid_proto;
5428         case BPF_FUNC_skb_vlan_push:
5429                 return &bpf_skb_vlan_push_proto;
5430         case BPF_FUNC_skb_vlan_pop:
5431                 return &bpf_skb_vlan_pop_proto;
5432         case BPF_FUNC_skb_change_proto:
5433                 return &bpf_skb_change_proto_proto;
5434         case BPF_FUNC_skb_change_type:
5435                 return &bpf_skb_change_type_proto;
5436         case BPF_FUNC_skb_adjust_room:
5437                 return &bpf_skb_adjust_room_proto;
5438         case BPF_FUNC_skb_change_tail:
5439                 return &bpf_skb_change_tail_proto;
5440         case BPF_FUNC_skb_get_tunnel_key:
5441                 return &bpf_skb_get_tunnel_key_proto;
5442         case BPF_FUNC_skb_set_tunnel_key:
5443                 return bpf_get_skb_set_tunnel_proto(func_id);
5444         case BPF_FUNC_skb_get_tunnel_opt:
5445                 return &bpf_skb_get_tunnel_opt_proto;
5446         case BPF_FUNC_skb_set_tunnel_opt:
5447                 return bpf_get_skb_set_tunnel_proto(func_id);
5448         case BPF_FUNC_redirect:
5449                 return &bpf_redirect_proto;
5450         case BPF_FUNC_get_route_realm:
5451                 return &bpf_get_route_realm_proto;
5452         case BPF_FUNC_get_hash_recalc:
5453                 return &bpf_get_hash_recalc_proto;
5454         case BPF_FUNC_set_hash_invalid:
5455                 return &bpf_set_hash_invalid_proto;
5456         case BPF_FUNC_set_hash:
5457                 return &bpf_set_hash_proto;
5458         case BPF_FUNC_perf_event_output:
5459                 return &bpf_skb_event_output_proto;
5460         case BPF_FUNC_get_smp_processor_id:
5461                 return &bpf_get_smp_processor_id_proto;
5462         case BPF_FUNC_skb_under_cgroup:
5463                 return &bpf_skb_under_cgroup_proto;
5464         case BPF_FUNC_get_socket_cookie:
5465                 return &bpf_get_socket_cookie_proto;
5466         case BPF_FUNC_get_socket_uid:
5467                 return &bpf_get_socket_uid_proto;
5468         case BPF_FUNC_fib_lookup:
5469                 return &bpf_skb_fib_lookup_proto;
5470 #ifdef CONFIG_XFRM
5471         case BPF_FUNC_skb_get_xfrm_state:
5472                 return &bpf_skb_get_xfrm_state_proto;
5473 #endif
5474 #ifdef CONFIG_SOCK_CGROUP_DATA
5475         case BPF_FUNC_skb_cgroup_id:
5476                 return &bpf_skb_cgroup_id_proto;
5477         case BPF_FUNC_skb_ancestor_cgroup_id:
5478                 return &bpf_skb_ancestor_cgroup_id_proto;
5479 #endif
5480 #ifdef CONFIG_INET
5481         case BPF_FUNC_sk_lookup_tcp:
5482                 return &bpf_sk_lookup_tcp_proto;
5483         case BPF_FUNC_sk_lookup_udp:
5484                 return &bpf_sk_lookup_udp_proto;
5485         case BPF_FUNC_sk_release:
5486                 return &bpf_sk_release_proto;
5487 #endif
5488         default:
5489                 return bpf_base_func_proto(func_id);
5490         }
5491 }
5492
5493 static const struct bpf_func_proto *
5494 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5495 {
5496         switch (func_id) {
5497         case BPF_FUNC_perf_event_output:
5498                 return &bpf_xdp_event_output_proto;
5499         case BPF_FUNC_get_smp_processor_id:
5500                 return &bpf_get_smp_processor_id_proto;
5501         case BPF_FUNC_csum_diff:
5502                 return &bpf_csum_diff_proto;
5503         case BPF_FUNC_xdp_adjust_head:
5504                 return &bpf_xdp_adjust_head_proto;
5505         case BPF_FUNC_xdp_adjust_meta:
5506                 return &bpf_xdp_adjust_meta_proto;
5507         case BPF_FUNC_redirect:
5508                 return &bpf_xdp_redirect_proto;
5509         case BPF_FUNC_redirect_map:
5510                 return &bpf_xdp_redirect_map_proto;
5511         case BPF_FUNC_xdp_adjust_tail:
5512                 return &bpf_xdp_adjust_tail_proto;
5513         case BPF_FUNC_fib_lookup:
5514                 return &bpf_xdp_fib_lookup_proto;
5515 #ifdef CONFIG_INET
5516         case BPF_FUNC_sk_lookup_udp:
5517                 return &bpf_xdp_sk_lookup_udp_proto;
5518         case BPF_FUNC_sk_lookup_tcp:
5519                 return &bpf_xdp_sk_lookup_tcp_proto;
5520         case BPF_FUNC_sk_release:
5521                 return &bpf_sk_release_proto;
5522 #endif
5523         default:
5524                 return bpf_base_func_proto(func_id);
5525         }
5526 }
5527
5528 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
5529 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
5530
5531 static const struct bpf_func_proto *
5532 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5533 {
5534         switch (func_id) {
5535         case BPF_FUNC_setsockopt:
5536                 return &bpf_setsockopt_proto;
5537         case BPF_FUNC_getsockopt:
5538                 return &bpf_getsockopt_proto;
5539         case BPF_FUNC_sock_ops_cb_flags_set:
5540                 return &bpf_sock_ops_cb_flags_set_proto;
5541         case BPF_FUNC_sock_map_update:
5542                 return &bpf_sock_map_update_proto;
5543         case BPF_FUNC_sock_hash_update:
5544                 return &bpf_sock_hash_update_proto;
5545         case BPF_FUNC_get_socket_cookie:
5546                 return &bpf_get_socket_cookie_sock_ops_proto;
5547         case BPF_FUNC_get_local_storage:
5548                 return &bpf_get_local_storage_proto;
5549         case BPF_FUNC_perf_event_output:
5550                 return &bpf_sockopt_event_output_proto;
5551         default:
5552                 return bpf_base_func_proto(func_id);
5553         }
5554 }
5555
5556 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
5557 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
5558
5559 static const struct bpf_func_proto *
5560 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5561 {
5562         switch (func_id) {
5563         case BPF_FUNC_msg_redirect_map:
5564                 return &bpf_msg_redirect_map_proto;
5565         case BPF_FUNC_msg_redirect_hash:
5566                 return &bpf_msg_redirect_hash_proto;
5567         case BPF_FUNC_msg_apply_bytes:
5568                 return &bpf_msg_apply_bytes_proto;
5569         case BPF_FUNC_msg_cork_bytes:
5570                 return &bpf_msg_cork_bytes_proto;
5571         case BPF_FUNC_msg_pull_data:
5572                 return &bpf_msg_pull_data_proto;
5573         case BPF_FUNC_msg_push_data:
5574                 return &bpf_msg_push_data_proto;
5575         case BPF_FUNC_msg_pop_data:
5576                 return &bpf_msg_pop_data_proto;
5577         default:
5578                 return bpf_base_func_proto(func_id);
5579         }
5580 }
5581
5582 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
5583 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
5584
5585 static const struct bpf_func_proto *
5586 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5587 {
5588         switch (func_id) {
5589         case BPF_FUNC_skb_store_bytes:
5590                 return &bpf_skb_store_bytes_proto;
5591         case BPF_FUNC_skb_load_bytes:
5592                 return &bpf_skb_load_bytes_proto;
5593         case BPF_FUNC_skb_pull_data:
5594                 return &sk_skb_pull_data_proto;
5595         case BPF_FUNC_skb_change_tail:
5596                 return &sk_skb_change_tail_proto;
5597         case BPF_FUNC_skb_change_head:
5598                 return &sk_skb_change_head_proto;
5599         case BPF_FUNC_get_socket_cookie:
5600                 return &bpf_get_socket_cookie_proto;
5601         case BPF_FUNC_get_socket_uid:
5602                 return &bpf_get_socket_uid_proto;
5603         case BPF_FUNC_sk_redirect_map:
5604                 return &bpf_sk_redirect_map_proto;
5605         case BPF_FUNC_sk_redirect_hash:
5606                 return &bpf_sk_redirect_hash_proto;
5607 #ifdef CONFIG_INET
5608         case BPF_FUNC_sk_lookup_tcp:
5609                 return &bpf_sk_lookup_tcp_proto;
5610         case BPF_FUNC_sk_lookup_udp:
5611                 return &bpf_sk_lookup_udp_proto;
5612         case BPF_FUNC_sk_release:
5613                 return &bpf_sk_release_proto;
5614 #endif
5615         default:
5616                 return bpf_base_func_proto(func_id);
5617         }
5618 }
5619
5620 static const struct bpf_func_proto *
5621 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5622 {
5623         switch (func_id) {
5624         case BPF_FUNC_skb_load_bytes:
5625                 return &bpf_skb_load_bytes_proto;
5626         default:
5627                 return bpf_base_func_proto(func_id);
5628         }
5629 }
5630
5631 static const struct bpf_func_proto *
5632 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5633 {
5634         switch (func_id) {
5635         case BPF_FUNC_skb_load_bytes:
5636                 return &bpf_skb_load_bytes_proto;
5637         case BPF_FUNC_skb_pull_data:
5638                 return &bpf_skb_pull_data_proto;
5639         case BPF_FUNC_csum_diff:
5640                 return &bpf_csum_diff_proto;
5641         case BPF_FUNC_get_cgroup_classid:
5642                 return &bpf_get_cgroup_classid_proto;
5643         case BPF_FUNC_get_route_realm:
5644                 return &bpf_get_route_realm_proto;
5645         case BPF_FUNC_get_hash_recalc:
5646                 return &bpf_get_hash_recalc_proto;
5647         case BPF_FUNC_perf_event_output:
5648                 return &bpf_skb_event_output_proto;
5649         case BPF_FUNC_get_smp_processor_id:
5650                 return &bpf_get_smp_processor_id_proto;
5651         case BPF_FUNC_skb_under_cgroup:
5652                 return &bpf_skb_under_cgroup_proto;
5653         default:
5654                 return bpf_base_func_proto(func_id);
5655         }
5656 }
5657
5658 static const struct bpf_func_proto *
5659 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5660 {
5661         switch (func_id) {
5662         case BPF_FUNC_lwt_push_encap:
5663                 return &bpf_lwt_push_encap_proto;
5664         default:
5665                 return lwt_out_func_proto(func_id, prog);
5666         }
5667 }
5668
5669 static const struct bpf_func_proto *
5670 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5671 {
5672         switch (func_id) {
5673         case BPF_FUNC_skb_get_tunnel_key:
5674                 return &bpf_skb_get_tunnel_key_proto;
5675         case BPF_FUNC_skb_set_tunnel_key:
5676                 return bpf_get_skb_set_tunnel_proto(func_id);
5677         case BPF_FUNC_skb_get_tunnel_opt:
5678                 return &bpf_skb_get_tunnel_opt_proto;
5679         case BPF_FUNC_skb_set_tunnel_opt:
5680                 return bpf_get_skb_set_tunnel_proto(func_id);
5681         case BPF_FUNC_redirect:
5682                 return &bpf_redirect_proto;
5683         case BPF_FUNC_clone_redirect:
5684                 return &bpf_clone_redirect_proto;
5685         case BPF_FUNC_skb_change_tail:
5686                 return &bpf_skb_change_tail_proto;
5687         case BPF_FUNC_skb_change_head:
5688                 return &bpf_skb_change_head_proto;
5689         case BPF_FUNC_skb_store_bytes:
5690                 return &bpf_skb_store_bytes_proto;
5691         case BPF_FUNC_csum_update:
5692                 return &bpf_csum_update_proto;
5693         case BPF_FUNC_l3_csum_replace:
5694                 return &bpf_l3_csum_replace_proto;
5695         case BPF_FUNC_l4_csum_replace:
5696                 return &bpf_l4_csum_replace_proto;
5697         case BPF_FUNC_set_hash_invalid:
5698                 return &bpf_set_hash_invalid_proto;
5699         default:
5700                 return lwt_out_func_proto(func_id, prog);
5701         }
5702 }
5703
5704 static const struct bpf_func_proto *
5705 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5706 {
5707         switch (func_id) {
5708 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5709         case BPF_FUNC_lwt_seg6_store_bytes:
5710                 return &bpf_lwt_seg6_store_bytes_proto;
5711         case BPF_FUNC_lwt_seg6_action:
5712                 return &bpf_lwt_seg6_action_proto;
5713         case BPF_FUNC_lwt_seg6_adjust_srh:
5714                 return &bpf_lwt_seg6_adjust_srh_proto;
5715 #endif
5716         default:
5717                 return lwt_out_func_proto(func_id, prog);
5718         }
5719 }
5720
5721 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5722                                     const struct bpf_prog *prog,
5723                                     struct bpf_insn_access_aux *info)
5724 {
5725         const int size_default = sizeof(__u32);
5726
5727         if (off < 0 || off >= sizeof(struct __sk_buff))
5728                 return false;
5729
5730         /* The verifier guarantees that size > 0. */
5731         if (off % size != 0)
5732                 return false;
5733
5734         switch (off) {
5735         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5736                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5737                         return false;
5738                 break;
5739         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5740         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5741         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5742         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5743         case bpf_ctx_range(struct __sk_buff, data):
5744         case bpf_ctx_range(struct __sk_buff, data_meta):
5745         case bpf_ctx_range(struct __sk_buff, data_end):
5746                 if (size != size_default)
5747                         return false;
5748                 break;
5749         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5750                 if (size != sizeof(__u64))
5751                         return false;
5752                 break;
5753         case bpf_ctx_range(struct __sk_buff, tstamp):
5754                 if (size != sizeof(__u64))
5755                         return false;
5756                 break;
5757         default:
5758                 /* Only narrow read access allowed for now. */
5759                 if (type == BPF_WRITE) {
5760                         if (size != size_default)
5761                                 return false;
5762                 } else {
5763                         bpf_ctx_record_field_size(info, size_default);
5764                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5765                                 return false;
5766                 }
5767         }
5768
5769         return true;
5770 }
5771
5772 static bool sk_filter_is_valid_access(int off, int size,
5773                                       enum bpf_access_type type,
5774                                       const struct bpf_prog *prog,
5775                                       struct bpf_insn_access_aux *info)
5776 {
5777         switch (off) {
5778         case bpf_ctx_range(struct __sk_buff, tc_classid):
5779         case bpf_ctx_range(struct __sk_buff, data):
5780         case bpf_ctx_range(struct __sk_buff, data_meta):
5781         case bpf_ctx_range(struct __sk_buff, data_end):
5782         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5783         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5784         case bpf_ctx_range(struct __sk_buff, tstamp):
5785         case bpf_ctx_range(struct __sk_buff, wire_len):
5786                 return false;
5787         }
5788
5789         if (type == BPF_WRITE) {
5790                 switch (off) {
5791                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5792                         break;
5793                 default:
5794                         return false;
5795                 }
5796         }
5797
5798         return bpf_skb_is_valid_access(off, size, type, prog, info);
5799 }
5800
5801 static bool cg_skb_is_valid_access(int off, int size,
5802                                    enum bpf_access_type type,
5803                                    const struct bpf_prog *prog,
5804                                    struct bpf_insn_access_aux *info)
5805 {
5806         switch (off) {
5807         case bpf_ctx_range(struct __sk_buff, tc_classid):
5808         case bpf_ctx_range(struct __sk_buff, data_meta):
5809         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5810         case bpf_ctx_range(struct __sk_buff, wire_len):
5811                 return false;
5812         case bpf_ctx_range(struct __sk_buff, data):
5813         case bpf_ctx_range(struct __sk_buff, data_end):
5814                 if (!capable(CAP_SYS_ADMIN))
5815                         return false;
5816                 break;
5817         }
5818
5819         if (type == BPF_WRITE) {
5820                 switch (off) {
5821                 case bpf_ctx_range(struct __sk_buff, mark):
5822                 case bpf_ctx_range(struct __sk_buff, priority):
5823                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5824                         break;
5825                 case bpf_ctx_range(struct __sk_buff, tstamp):
5826                         if (!capable(CAP_SYS_ADMIN))
5827                                 return false;
5828                         break;
5829                 default:
5830                         return false;
5831                 }
5832         }
5833
5834         switch (off) {
5835         case bpf_ctx_range(struct __sk_buff, data):
5836                 info->reg_type = PTR_TO_PACKET;
5837                 break;
5838         case bpf_ctx_range(struct __sk_buff, data_end):
5839                 info->reg_type = PTR_TO_PACKET_END;
5840                 break;
5841         }
5842
5843         return bpf_skb_is_valid_access(off, size, type, prog, info);
5844 }
5845
5846 static bool lwt_is_valid_access(int off, int size,
5847                                 enum bpf_access_type type,
5848                                 const struct bpf_prog *prog,
5849                                 struct bpf_insn_access_aux *info)
5850 {
5851         switch (off) {
5852         case bpf_ctx_range(struct __sk_buff, tc_classid):
5853         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5854         case bpf_ctx_range(struct __sk_buff, data_meta):
5855         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
5856         case bpf_ctx_range(struct __sk_buff, tstamp):
5857         case bpf_ctx_range(struct __sk_buff, wire_len):
5858                 return false;
5859         }
5860
5861         if (type == BPF_WRITE) {
5862                 switch (off) {
5863                 case bpf_ctx_range(struct __sk_buff, mark):
5864                 case bpf_ctx_range(struct __sk_buff, priority):
5865                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5866                         break;
5867                 default:
5868                         return false;
5869                 }
5870         }
5871
5872         switch (off) {
5873         case bpf_ctx_range(struct __sk_buff, data):
5874                 info->reg_type = PTR_TO_PACKET;
5875                 break;
5876         case bpf_ctx_range(struct __sk_buff, data_end):
5877                 info->reg_type = PTR_TO_PACKET_END;
5878                 break;
5879         }
5880
5881         return bpf_skb_is_valid_access(off, size, type, prog, info);
5882 }
5883
5884 /* Attach type specific accesses */
5885 static bool __sock_filter_check_attach_type(int off,
5886                                             enum bpf_access_type access_type,
5887                                             enum bpf_attach_type attach_type)
5888 {
5889         switch (off) {
5890         case offsetof(struct bpf_sock, bound_dev_if):
5891         case offsetof(struct bpf_sock, mark):
5892         case offsetof(struct bpf_sock, priority):
5893                 switch (attach_type) {
5894                 case BPF_CGROUP_INET_SOCK_CREATE:
5895                         goto full_access;
5896                 default:
5897                         return false;
5898                 }
5899         case bpf_ctx_range(struct bpf_sock, src_ip4):
5900                 switch (attach_type) {
5901                 case BPF_CGROUP_INET4_POST_BIND:
5902                         goto read_only;
5903                 default:
5904                         return false;
5905                 }
5906         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5907                 switch (attach_type) {
5908                 case BPF_CGROUP_INET6_POST_BIND:
5909                         goto read_only;
5910                 default:
5911                         return false;
5912                 }
5913         case bpf_ctx_range(struct bpf_sock, src_port):
5914                 switch (attach_type) {
5915                 case BPF_CGROUP_INET4_POST_BIND:
5916                 case BPF_CGROUP_INET6_POST_BIND:
5917                         goto read_only;
5918                 default:
5919                         return false;
5920                 }
5921         }
5922 read_only:
5923         return access_type == BPF_READ;
5924 full_access:
5925         return true;
5926 }
5927
5928 static bool __sock_filter_check_size(int off, int size,
5929                                      struct bpf_insn_access_aux *info)
5930 {
5931         const int size_default = sizeof(__u32);
5932
5933         switch (off) {
5934         case bpf_ctx_range(struct bpf_sock, src_ip4):
5935         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5936                 bpf_ctx_record_field_size(info, size_default);
5937                 return bpf_ctx_narrow_access_ok(off, size, size_default);
5938         }
5939
5940         return size == size_default;
5941 }
5942
5943 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5944                               struct bpf_insn_access_aux *info)
5945 {
5946         if (off < 0 || off >= sizeof(struct bpf_sock))
5947                 return false;
5948         if (off % size != 0)
5949                 return false;
5950         if (!__sock_filter_check_size(off, size, info))
5951                 return false;
5952         return true;
5953 }
5954
5955 static bool sock_filter_is_valid_access(int off, int size,
5956                                         enum bpf_access_type type,
5957                                         const struct bpf_prog *prog,
5958                                         struct bpf_insn_access_aux *info)
5959 {
5960         if (!bpf_sock_is_valid_access(off, size, type, info))
5961                 return false;
5962         return __sock_filter_check_attach_type(off, type,
5963                                                prog->expected_attach_type);
5964 }
5965
5966 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
5967                              const struct bpf_prog *prog)
5968 {
5969         /* Neither direct read nor direct write requires any preliminary
5970          * action.
5971          */
5972         return 0;
5973 }
5974
5975 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5976                                 const struct bpf_prog *prog, int drop_verdict)
5977 {
5978         struct bpf_insn *insn = insn_buf;
5979
5980         if (!direct_write)
5981                 return 0;
5982
5983         /* if (!skb->cloned)
5984          *       goto start;
5985          *
5986          * (Fast-path, otherwise approximation that we might be
5987          *  a clone, do the rest in helper.)
5988          */
5989         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5990         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5991         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5992
5993         /* ret = bpf_skb_pull_data(skb, 0); */
5994         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5995         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5996         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5997                                BPF_FUNC_skb_pull_data);
5998         /* if (!ret)
5999          *      goto restore;
6000          * return TC_ACT_SHOT;
6001          */
6002         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6003         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6004         *insn++ = BPF_EXIT_INSN();
6005
6006         /* restore: */
6007         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6008         /* start: */
6009         *insn++ = prog->insnsi[0];
6010
6011         return insn - insn_buf;
6012 }
6013
6014 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6015                           struct bpf_insn *insn_buf)
6016 {
6017         bool indirect = BPF_MODE(orig->code) == BPF_IND;
6018         struct bpf_insn *insn = insn_buf;
6019
6020         /* We're guaranteed here that CTX is in R6. */
6021         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6022         if (!indirect) {
6023                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6024         } else {
6025                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6026                 if (orig->imm)
6027                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6028         }
6029
6030         switch (BPF_SIZE(orig->code)) {
6031         case BPF_B:
6032                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6033                 break;
6034         case BPF_H:
6035                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6036                 break;
6037         case BPF_W:
6038                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6039                 break;
6040         }
6041
6042         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6043         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6044         *insn++ = BPF_EXIT_INSN();
6045
6046         return insn - insn_buf;
6047 }
6048
6049 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6050                                const struct bpf_prog *prog)
6051 {
6052         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6053 }
6054
6055 static bool tc_cls_act_is_valid_access(int off, int size,
6056                                        enum bpf_access_type type,
6057                                        const struct bpf_prog *prog,
6058                                        struct bpf_insn_access_aux *info)
6059 {
6060         if (type == BPF_WRITE) {
6061                 switch (off) {
6062                 case bpf_ctx_range(struct __sk_buff, mark):
6063                 case bpf_ctx_range(struct __sk_buff, tc_index):
6064                 case bpf_ctx_range(struct __sk_buff, priority):
6065                 case bpf_ctx_range(struct __sk_buff, tc_classid):
6066                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6067                 case bpf_ctx_range(struct __sk_buff, tstamp):
6068                         break;
6069                 default:
6070                         return false;
6071                 }
6072         }
6073
6074         switch (off) {
6075         case bpf_ctx_range(struct __sk_buff, data):
6076                 info->reg_type = PTR_TO_PACKET;
6077                 break;
6078         case bpf_ctx_range(struct __sk_buff, data_meta):
6079                 info->reg_type = PTR_TO_PACKET_META;
6080                 break;
6081         case bpf_ctx_range(struct __sk_buff, data_end):
6082                 info->reg_type = PTR_TO_PACKET_END;
6083                 break;
6084         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6085         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6086                 return false;
6087         }
6088
6089         return bpf_skb_is_valid_access(off, size, type, prog, info);
6090 }
6091
6092 static bool __is_valid_xdp_access(int off, int size)
6093 {
6094         if (off < 0 || off >= sizeof(struct xdp_md))
6095                 return false;
6096         if (off % size != 0)
6097                 return false;
6098         if (size != sizeof(__u32))
6099                 return false;
6100
6101         return true;
6102 }
6103
6104 static bool xdp_is_valid_access(int off, int size,
6105                                 enum bpf_access_type type,
6106                                 const struct bpf_prog *prog,
6107                                 struct bpf_insn_access_aux *info)
6108 {
6109         if (type == BPF_WRITE) {
6110                 if (bpf_prog_is_dev_bound(prog->aux)) {
6111                         switch (off) {
6112                         case offsetof(struct xdp_md, rx_queue_index):
6113                                 return __is_valid_xdp_access(off, size);
6114                         }
6115                 }
6116                 return false;
6117         }
6118
6119         switch (off) {
6120         case offsetof(struct xdp_md, data):
6121                 info->reg_type = PTR_TO_PACKET;
6122                 break;
6123         case offsetof(struct xdp_md, data_meta):
6124                 info->reg_type = PTR_TO_PACKET_META;
6125                 break;
6126         case offsetof(struct xdp_md, data_end):
6127                 info->reg_type = PTR_TO_PACKET_END;
6128                 break;
6129         }
6130
6131         return __is_valid_xdp_access(off, size);
6132 }
6133
6134 void bpf_warn_invalid_xdp_action(u32 act)
6135 {
6136         const u32 act_max = XDP_REDIRECT;
6137
6138         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6139                   act > act_max ? "Illegal" : "Driver unsupported",
6140                   act);
6141 }
6142 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6143
6144 static bool sock_addr_is_valid_access(int off, int size,
6145                                       enum bpf_access_type type,
6146                                       const struct bpf_prog *prog,
6147                                       struct bpf_insn_access_aux *info)
6148 {
6149         const int size_default = sizeof(__u32);
6150
6151         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6152                 return false;
6153         if (off % size != 0)
6154                 return false;
6155
6156         /* Disallow access to IPv6 fields from IPv4 contex and vise
6157          * versa.
6158          */
6159         switch (off) {
6160         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6161                 switch (prog->expected_attach_type) {
6162                 case BPF_CGROUP_INET4_BIND:
6163                 case BPF_CGROUP_INET4_CONNECT:
6164                 case BPF_CGROUP_UDP4_SENDMSG:
6165                         break;
6166                 default:
6167                         return false;
6168                 }
6169                 break;
6170         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6171                 switch (prog->expected_attach_type) {
6172                 case BPF_CGROUP_INET6_BIND:
6173                 case BPF_CGROUP_INET6_CONNECT:
6174                 case BPF_CGROUP_UDP6_SENDMSG:
6175                         break;
6176                 default:
6177                         return false;
6178                 }
6179                 break;
6180         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6181                 switch (prog->expected_attach_type) {
6182                 case BPF_CGROUP_UDP4_SENDMSG:
6183                         break;
6184                 default:
6185                         return false;
6186                 }
6187                 break;
6188         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6189                                 msg_src_ip6[3]):
6190                 switch (prog->expected_attach_type) {
6191                 case BPF_CGROUP_UDP6_SENDMSG:
6192                         break;
6193                 default:
6194                         return false;
6195                 }
6196                 break;
6197         }
6198
6199         switch (off) {
6200         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6201         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6202         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6203         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6204                                 msg_src_ip6[3]):
6205                 /* Only narrow read access allowed for now. */
6206                 if (type == BPF_READ) {
6207                         bpf_ctx_record_field_size(info, size_default);
6208                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6209                                 return false;
6210                 } else {
6211                         if (size != size_default)
6212                                 return false;
6213                 }
6214                 break;
6215         case bpf_ctx_range(struct bpf_sock_addr, user_port):
6216                 if (size != size_default)
6217                         return false;
6218                 break;
6219         default:
6220                 if (type == BPF_READ) {
6221                         if (size != size_default)
6222                                 return false;
6223                 } else {
6224                         return false;
6225                 }
6226         }
6227
6228         return true;
6229 }
6230
6231 static bool sock_ops_is_valid_access(int off, int size,
6232                                      enum bpf_access_type type,
6233                                      const struct bpf_prog *prog,
6234                                      struct bpf_insn_access_aux *info)
6235 {
6236         const int size_default = sizeof(__u32);
6237
6238         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6239                 return false;
6240
6241         /* The verifier guarantees that size > 0. */
6242         if (off % size != 0)
6243                 return false;
6244
6245         if (type == BPF_WRITE) {
6246                 switch (off) {
6247                 case offsetof(struct bpf_sock_ops, reply):
6248                 case offsetof(struct bpf_sock_ops, sk_txhash):
6249                         if (size != size_default)
6250                                 return false;
6251                         break;
6252                 default:
6253                         return false;
6254                 }
6255         } else {
6256                 switch (off) {
6257                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6258                                         bytes_acked):
6259                         if (size != sizeof(__u64))
6260                                 return false;
6261                         break;
6262                 default:
6263                         if (size != size_default)
6264                                 return false;
6265                         break;
6266                 }
6267         }
6268
6269         return true;
6270 }
6271
6272 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6273                            const struct bpf_prog *prog)
6274 {
6275         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6276 }
6277
6278 static bool sk_skb_is_valid_access(int off, int size,
6279                                    enum bpf_access_type type,
6280                                    const struct bpf_prog *prog,
6281                                    struct bpf_insn_access_aux *info)
6282 {
6283         switch (off) {
6284         case bpf_ctx_range(struct __sk_buff, tc_classid):
6285         case bpf_ctx_range(struct __sk_buff, data_meta):
6286         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6287         case bpf_ctx_range(struct __sk_buff, tstamp):
6288         case bpf_ctx_range(struct __sk_buff, wire_len):
6289                 return false;
6290         }
6291
6292         if (type == BPF_WRITE) {
6293                 switch (off) {
6294                 case bpf_ctx_range(struct __sk_buff, tc_index):
6295                 case bpf_ctx_range(struct __sk_buff, priority):
6296                         break;
6297                 default:
6298                         return false;
6299                 }
6300         }
6301
6302         switch (off) {
6303         case bpf_ctx_range(struct __sk_buff, mark):
6304                 return false;
6305         case bpf_ctx_range(struct __sk_buff, data):
6306                 info->reg_type = PTR_TO_PACKET;
6307                 break;
6308         case bpf_ctx_range(struct __sk_buff, data_end):
6309                 info->reg_type = PTR_TO_PACKET_END;
6310                 break;
6311         }
6312
6313         return bpf_skb_is_valid_access(off, size, type, prog, info);
6314 }
6315
6316 static bool sk_msg_is_valid_access(int off, int size,
6317                                    enum bpf_access_type type,
6318                                    const struct bpf_prog *prog,
6319                                    struct bpf_insn_access_aux *info)
6320 {
6321         if (type == BPF_WRITE)
6322                 return false;
6323
6324         if (off % size != 0)
6325                 return false;
6326
6327         switch (off) {
6328         case offsetof(struct sk_msg_md, data):
6329                 info->reg_type = PTR_TO_PACKET;
6330                 if (size != sizeof(__u64))
6331                         return false;
6332                 break;
6333         case offsetof(struct sk_msg_md, data_end):
6334                 info->reg_type = PTR_TO_PACKET_END;
6335                 if (size != sizeof(__u64))
6336                         return false;
6337                 break;
6338         case bpf_ctx_range(struct sk_msg_md, family):
6339         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
6340         case bpf_ctx_range(struct sk_msg_md, local_ip4):
6341         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
6342         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
6343         case bpf_ctx_range(struct sk_msg_md, remote_port):
6344         case bpf_ctx_range(struct sk_msg_md, local_port):
6345         case bpf_ctx_range(struct sk_msg_md, size):
6346                 if (size != sizeof(__u32))
6347                         return false;
6348                 break;
6349         default:
6350                 return false;
6351         }
6352         return true;
6353 }
6354
6355 static bool flow_dissector_is_valid_access(int off, int size,
6356                                            enum bpf_access_type type,
6357                                            const struct bpf_prog *prog,
6358                                            struct bpf_insn_access_aux *info)
6359 {
6360         if (type == BPF_WRITE) {
6361                 switch (off) {
6362                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6363                         break;
6364                 default:
6365                         return false;
6366                 }
6367         }
6368
6369         switch (off) {
6370         case bpf_ctx_range(struct __sk_buff, data):
6371                 info->reg_type = PTR_TO_PACKET;
6372                 break;
6373         case bpf_ctx_range(struct __sk_buff, data_end):
6374                 info->reg_type = PTR_TO_PACKET_END;
6375                 break;
6376         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6377                 info->reg_type = PTR_TO_FLOW_KEYS;
6378                 break;
6379         case bpf_ctx_range(struct __sk_buff, tc_classid):
6380         case bpf_ctx_range(struct __sk_buff, data_meta):
6381         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6382         case bpf_ctx_range(struct __sk_buff, tstamp):
6383         case bpf_ctx_range(struct __sk_buff, wire_len):
6384                 return false;
6385         }
6386
6387         return bpf_skb_is_valid_access(off, size, type, prog, info);
6388 }
6389
6390 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
6391                                   const struct bpf_insn *si,
6392                                   struct bpf_insn *insn_buf,
6393                                   struct bpf_prog *prog, u32 *target_size)
6394 {
6395         struct bpf_insn *insn = insn_buf;
6396         int off;
6397
6398         switch (si->off) {
6399         case offsetof(struct __sk_buff, len):
6400                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6401                                       bpf_target_off(struct sk_buff, len, 4,
6402                                                      target_size));
6403                 break;
6404
6405         case offsetof(struct __sk_buff, protocol):
6406                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6407                                       bpf_target_off(struct sk_buff, protocol, 2,
6408                                                      target_size));
6409                 break;
6410
6411         case offsetof(struct __sk_buff, vlan_proto):
6412                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6413                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
6414                                                      target_size));
6415                 break;
6416
6417         case offsetof(struct __sk_buff, priority):
6418                 if (type == BPF_WRITE)
6419                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6420                                               bpf_target_off(struct sk_buff, priority, 4,
6421                                                              target_size));
6422                 else
6423                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6424                                               bpf_target_off(struct sk_buff, priority, 4,
6425                                                              target_size));
6426                 break;
6427
6428         case offsetof(struct __sk_buff, ingress_ifindex):
6429                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6430                                       bpf_target_off(struct sk_buff, skb_iif, 4,
6431                                                      target_size));
6432                 break;
6433
6434         case offsetof(struct __sk_buff, ifindex):
6435                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6436                                       si->dst_reg, si->src_reg,
6437                                       offsetof(struct sk_buff, dev));
6438                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
6439                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6440                                       bpf_target_off(struct net_device, ifindex, 4,
6441                                                      target_size));
6442                 break;
6443
6444         case offsetof(struct __sk_buff, hash):
6445                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6446                                       bpf_target_off(struct sk_buff, hash, 4,
6447                                                      target_size));
6448                 break;
6449
6450         case offsetof(struct __sk_buff, mark):
6451                 if (type == BPF_WRITE)
6452                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6453                                               bpf_target_off(struct sk_buff, mark, 4,
6454                                                              target_size));
6455                 else
6456                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6457                                               bpf_target_off(struct sk_buff, mark, 4,
6458                                                              target_size));
6459                 break;
6460
6461         case offsetof(struct __sk_buff, pkt_type):
6462                 *target_size = 1;
6463                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6464                                       PKT_TYPE_OFFSET());
6465                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
6466 #ifdef __BIG_ENDIAN_BITFIELD
6467                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
6468 #endif
6469                 break;
6470
6471         case offsetof(struct __sk_buff, queue_mapping):
6472                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6473                                       bpf_target_off(struct sk_buff, queue_mapping, 2,
6474                                                      target_size));
6475                 break;
6476
6477         case offsetof(struct __sk_buff, vlan_present):
6478                 *target_size = 1;
6479                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6480                                       PKT_VLAN_PRESENT_OFFSET());
6481                 if (PKT_VLAN_PRESENT_BIT)
6482                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
6483                 if (PKT_VLAN_PRESENT_BIT < 7)
6484                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
6485                 break;
6486
6487         case offsetof(struct __sk_buff, vlan_tci):
6488                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6489                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
6490                                                      target_size));
6491                 break;
6492
6493         case offsetof(struct __sk_buff, cb[0]) ...
6494              offsetofend(struct __sk_buff, cb[4]) - 1:
6495                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
6496                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
6497                               offsetof(struct qdisc_skb_cb, data)) %
6498                              sizeof(__u64));
6499
6500                 prog->cb_access = 1;
6501                 off  = si->off;
6502                 off -= offsetof(struct __sk_buff, cb[0]);
6503                 off += offsetof(struct sk_buff, cb);
6504                 off += offsetof(struct qdisc_skb_cb, data);
6505                 if (type == BPF_WRITE)
6506                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
6507                                               si->src_reg, off);
6508                 else
6509                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
6510                                               si->src_reg, off);
6511                 break;
6512
6513         case offsetof(struct __sk_buff, tc_classid):
6514                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
6515
6516                 off  = si->off;
6517                 off -= offsetof(struct __sk_buff, tc_classid);
6518                 off += offsetof(struct sk_buff, cb);
6519                 off += offsetof(struct qdisc_skb_cb, tc_classid);
6520                 *target_size = 2;
6521                 if (type == BPF_WRITE)
6522                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
6523                                               si->src_reg, off);
6524                 else
6525                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
6526                                               si->src_reg, off);
6527                 break;
6528
6529         case offsetof(struct __sk_buff, data):
6530                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
6531                                       si->dst_reg, si->src_reg,
6532                                       offsetof(struct sk_buff, data));
6533                 break;
6534
6535         case offsetof(struct __sk_buff, data_meta):
6536                 off  = si->off;
6537                 off -= offsetof(struct __sk_buff, data_meta);
6538                 off += offsetof(struct sk_buff, cb);
6539                 off += offsetof(struct bpf_skb_data_end, data_meta);
6540                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6541                                       si->src_reg, off);
6542                 break;
6543
6544         case offsetof(struct __sk_buff, data_end):
6545                 off  = si->off;
6546                 off -= offsetof(struct __sk_buff, data_end);
6547                 off += offsetof(struct sk_buff, cb);
6548                 off += offsetof(struct bpf_skb_data_end, data_end);
6549                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6550                                       si->src_reg, off);
6551                 break;
6552
6553         case offsetof(struct __sk_buff, tc_index):
6554 #ifdef CONFIG_NET_SCHED
6555                 if (type == BPF_WRITE)
6556                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6557                                               bpf_target_off(struct sk_buff, tc_index, 2,
6558                                                              target_size));
6559                 else
6560                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6561                                               bpf_target_off(struct sk_buff, tc_index, 2,
6562                                                              target_size));
6563 #else
6564                 *target_size = 2;
6565                 if (type == BPF_WRITE)
6566                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
6567                 else
6568                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6569 #endif
6570                 break;
6571
6572         case offsetof(struct __sk_buff, napi_id):
6573 #if defined(CONFIG_NET_RX_BUSY_POLL)
6574                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6575                                       bpf_target_off(struct sk_buff, napi_id, 4,
6576                                                      target_size));
6577                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
6578                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6579 #else
6580                 *target_size = 4;
6581                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6582 #endif
6583                 break;
6584         case offsetof(struct __sk_buff, family):
6585                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6586
6587                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6588                                       si->dst_reg, si->src_reg,
6589                                       offsetof(struct sk_buff, sk));
6590                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6591                                       bpf_target_off(struct sock_common,
6592                                                      skc_family,
6593                                                      2, target_size));
6594                 break;
6595         case offsetof(struct __sk_buff, remote_ip4):
6596                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6597
6598                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6599                                       si->dst_reg, si->src_reg,
6600                                       offsetof(struct sk_buff, sk));
6601                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6602                                       bpf_target_off(struct sock_common,
6603                                                      skc_daddr,
6604                                                      4, target_size));
6605                 break;
6606         case offsetof(struct __sk_buff, local_ip4):
6607                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6608                                           skc_rcv_saddr) != 4);
6609
6610                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6611                                       si->dst_reg, si->src_reg,
6612                                       offsetof(struct sk_buff, sk));
6613                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6614                                       bpf_target_off(struct sock_common,
6615                                                      skc_rcv_saddr,
6616                                                      4, target_size));
6617                 break;
6618         case offsetof(struct __sk_buff, remote_ip6[0]) ...
6619              offsetof(struct __sk_buff, remote_ip6[3]):
6620 #if IS_ENABLED(CONFIG_IPV6)
6621                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6622                                           skc_v6_daddr.s6_addr32[0]) != 4);
6623
6624                 off = si->off;
6625                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
6626
6627                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6628                                       si->dst_reg, si->src_reg,
6629                                       offsetof(struct sk_buff, sk));
6630                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6631                                       offsetof(struct sock_common,
6632                                                skc_v6_daddr.s6_addr32[0]) +
6633                                       off);
6634 #else
6635                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6636 #endif
6637                 break;
6638         case offsetof(struct __sk_buff, local_ip6[0]) ...
6639              offsetof(struct __sk_buff, local_ip6[3]):
6640 #if IS_ENABLED(CONFIG_IPV6)
6641                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6642                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6643
6644                 off = si->off;
6645                 off -= offsetof(struct __sk_buff, local_ip6[0]);
6646
6647                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6648                                       si->dst_reg, si->src_reg,
6649                                       offsetof(struct sk_buff, sk));
6650                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6651                                       offsetof(struct sock_common,
6652                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6653                                       off);
6654 #else
6655                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6656 #endif
6657                 break;
6658
6659         case offsetof(struct __sk_buff, remote_port):
6660                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6661
6662                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6663                                       si->dst_reg, si->src_reg,
6664                                       offsetof(struct sk_buff, sk));
6665                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6666                                       bpf_target_off(struct sock_common,
6667                                                      skc_dport,
6668                                                      2, target_size));
6669 #ifndef __BIG_ENDIAN_BITFIELD
6670                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6671 #endif
6672                 break;
6673
6674         case offsetof(struct __sk_buff, local_port):
6675                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6676
6677                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6678                                       si->dst_reg, si->src_reg,
6679                                       offsetof(struct sk_buff, sk));
6680                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6681                                       bpf_target_off(struct sock_common,
6682                                                      skc_num, 2, target_size));
6683                 break;
6684
6685         case offsetof(struct __sk_buff, flow_keys):
6686                 off  = si->off;
6687                 off -= offsetof(struct __sk_buff, flow_keys);
6688                 off += offsetof(struct sk_buff, cb);
6689                 off += offsetof(struct qdisc_skb_cb, flow_keys);
6690                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6691                                       si->src_reg, off);
6692                 break;
6693
6694         case offsetof(struct __sk_buff, tstamp):
6695                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
6696
6697                 if (type == BPF_WRITE)
6698                         *insn++ = BPF_STX_MEM(BPF_DW,
6699                                               si->dst_reg, si->src_reg,
6700                                               bpf_target_off(struct sk_buff,
6701                                                              tstamp, 8,
6702                                                              target_size));
6703                 else
6704                         *insn++ = BPF_LDX_MEM(BPF_DW,
6705                                               si->dst_reg, si->src_reg,
6706                                               bpf_target_off(struct sk_buff,
6707                                                              tstamp, 8,
6708                                                              target_size));
6709                 break;
6710
6711         case offsetof(struct __sk_buff, wire_len):
6712                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
6713
6714                 off = si->off;
6715                 off -= offsetof(struct __sk_buff, wire_len);
6716                 off += offsetof(struct sk_buff, cb);
6717                 off += offsetof(struct qdisc_skb_cb, pkt_len);
6718                 *target_size = 4;
6719                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
6720         }
6721
6722         return insn - insn_buf;
6723 }
6724
6725 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
6726                                 const struct bpf_insn *si,
6727                                 struct bpf_insn *insn_buf,
6728                                 struct bpf_prog *prog, u32 *target_size)
6729 {
6730         struct bpf_insn *insn = insn_buf;
6731         int off;
6732
6733         switch (si->off) {
6734         case offsetof(struct bpf_sock, bound_dev_if):
6735                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
6736
6737                 if (type == BPF_WRITE)
6738                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6739                                         offsetof(struct sock, sk_bound_dev_if));
6740                 else
6741                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6742                                       offsetof(struct sock, sk_bound_dev_if));
6743                 break;
6744
6745         case offsetof(struct bpf_sock, mark):
6746                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
6747
6748                 if (type == BPF_WRITE)
6749                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6750                                         offsetof(struct sock, sk_mark));
6751                 else
6752                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6753                                       offsetof(struct sock, sk_mark));
6754                 break;
6755
6756         case offsetof(struct bpf_sock, priority):
6757                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
6758
6759                 if (type == BPF_WRITE)
6760                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6761                                         offsetof(struct sock, sk_priority));
6762                 else
6763                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6764                                       offsetof(struct sock, sk_priority));
6765                 break;
6766
6767         case offsetof(struct bpf_sock, family):
6768                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
6769
6770                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6771                                       offsetof(struct sock, sk_family));
6772                 break;
6773
6774         case offsetof(struct bpf_sock, type):
6775                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6776                                       offsetof(struct sock, __sk_flags_offset));
6777                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6778                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6779                 break;
6780
6781         case offsetof(struct bpf_sock, protocol):
6782                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6783                                       offsetof(struct sock, __sk_flags_offset));
6784                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6785                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
6786                 break;
6787
6788         case offsetof(struct bpf_sock, src_ip4):
6789                 *insn++ = BPF_LDX_MEM(
6790                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6791                         bpf_target_off(struct sock_common, skc_rcv_saddr,
6792                                        FIELD_SIZEOF(struct sock_common,
6793                                                     skc_rcv_saddr),
6794                                        target_size));
6795                 break;
6796
6797         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6798 #if IS_ENABLED(CONFIG_IPV6)
6799                 off = si->off;
6800                 off -= offsetof(struct bpf_sock, src_ip6[0]);
6801                 *insn++ = BPF_LDX_MEM(
6802                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6803                         bpf_target_off(
6804                                 struct sock_common,
6805                                 skc_v6_rcv_saddr.s6_addr32[0],
6806                                 FIELD_SIZEOF(struct sock_common,
6807                                              skc_v6_rcv_saddr.s6_addr32[0]),
6808                                 target_size) + off);
6809 #else
6810                 (void)off;
6811                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6812 #endif
6813                 break;
6814
6815         case offsetof(struct bpf_sock, src_port):
6816                 *insn++ = BPF_LDX_MEM(
6817                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6818                         si->dst_reg, si->src_reg,
6819                         bpf_target_off(struct sock_common, skc_num,
6820                                        FIELD_SIZEOF(struct sock_common,
6821                                                     skc_num),
6822                                        target_size));
6823                 break;
6824         }
6825
6826         return insn - insn_buf;
6827 }
6828
6829 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6830                                          const struct bpf_insn *si,
6831                                          struct bpf_insn *insn_buf,
6832                                          struct bpf_prog *prog, u32 *target_size)
6833 {
6834         struct bpf_insn *insn = insn_buf;
6835
6836         switch (si->off) {
6837         case offsetof(struct __sk_buff, ifindex):
6838                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6839                                       si->dst_reg, si->src_reg,
6840                                       offsetof(struct sk_buff, dev));
6841                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6842                                       bpf_target_off(struct net_device, ifindex, 4,
6843                                                      target_size));
6844                 break;
6845         default:
6846                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6847                                               target_size);
6848         }
6849
6850         return insn - insn_buf;
6851 }
6852
6853 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6854                                   const struct bpf_insn *si,
6855                                   struct bpf_insn *insn_buf,
6856                                   struct bpf_prog *prog, u32 *target_size)
6857 {
6858         struct bpf_insn *insn = insn_buf;
6859
6860         switch (si->off) {
6861         case offsetof(struct xdp_md, data):
6862                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6863                                       si->dst_reg, si->src_reg,
6864                                       offsetof(struct xdp_buff, data));
6865                 break;
6866         case offsetof(struct xdp_md, data_meta):
6867                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6868                                       si->dst_reg, si->src_reg,
6869                                       offsetof(struct xdp_buff, data_meta));
6870                 break;
6871         case offsetof(struct xdp_md, data_end):
6872                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6873                                       si->dst_reg, si->src_reg,
6874                                       offsetof(struct xdp_buff, data_end));
6875                 break;
6876         case offsetof(struct xdp_md, ingress_ifindex):
6877                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6878                                       si->dst_reg, si->src_reg,
6879                                       offsetof(struct xdp_buff, rxq));
6880                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6881                                       si->dst_reg, si->dst_reg,
6882                                       offsetof(struct xdp_rxq_info, dev));
6883                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6884                                       offsetof(struct net_device, ifindex));
6885                 break;
6886         case offsetof(struct xdp_md, rx_queue_index):
6887                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6888                                       si->dst_reg, si->src_reg,
6889                                       offsetof(struct xdp_buff, rxq));
6890                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6891                                       offsetof(struct xdp_rxq_info,
6892                                                queue_index));
6893                 break;
6894         }
6895
6896         return insn - insn_buf;
6897 }
6898
6899 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6900  * context Structure, F is Field in context structure that contains a pointer
6901  * to Nested Structure of type NS that has the field NF.
6902  *
6903  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6904  * sure that SIZE is not greater than actual size of S.F.NF.
6905  *
6906  * If offset OFF is provided, the load happens from that offset relative to
6907  * offset of NF.
6908  */
6909 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
6910         do {                                                                   \
6911                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
6912                                       si->src_reg, offsetof(S, F));            \
6913                 *insn++ = BPF_LDX_MEM(                                         \
6914                         SIZE, si->dst_reg, si->dst_reg,                        \
6915                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6916                                        target_size)                            \
6917                                 + OFF);                                        \
6918         } while (0)
6919
6920 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
6921         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
6922                                              BPF_FIELD_SIZEOF(NS, NF), 0)
6923
6924 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6925  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6926  *
6927  * It doesn't support SIZE argument though since narrow stores are not
6928  * supported for now.
6929  *
6930  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6931  * "register" since two registers available in convert_ctx_access are not
6932  * enough: we can't override neither SRC, since it contains value to store, nor
6933  * DST since it contains pointer to context that may be used by later
6934  * instructions. But we need a temporary place to save pointer to nested
6935  * structure whose field we want to store to.
6936  */
6937 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF)                \
6938         do {                                                                   \
6939                 int tmp_reg = BPF_REG_9;                                       \
6940                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6941                         --tmp_reg;                                             \
6942                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6943                         --tmp_reg;                                             \
6944                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
6945                                       offsetof(S, TF));                        \
6946                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
6947                                       si->dst_reg, offsetof(S, F));            \
6948                 *insn++ = BPF_STX_MEM(                                         \
6949                         BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg,        \
6950                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6951                                        target_size)                            \
6952                                 + OFF);                                        \
6953                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
6954                                       offsetof(S, TF));                        \
6955         } while (0)
6956
6957 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6958                                                       TF)                      \
6959         do {                                                                   \
6960                 if (type == BPF_WRITE) {                                       \
6961                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF,    \
6962                                                          TF);                  \
6963                 } else {                                                       \
6964                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
6965                                 S, NS, F, NF, SIZE, OFF);  \
6966                 }                                                              \
6967         } while (0)
6968
6969 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
6970         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
6971                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6972
6973 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6974                                         const struct bpf_insn *si,
6975                                         struct bpf_insn *insn_buf,
6976                                         struct bpf_prog *prog, u32 *target_size)
6977 {
6978         struct bpf_insn *insn = insn_buf;
6979         int off;
6980
6981         switch (si->off) {
6982         case offsetof(struct bpf_sock_addr, user_family):
6983                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6984                                             struct sockaddr, uaddr, sa_family);
6985                 break;
6986
6987         case offsetof(struct bpf_sock_addr, user_ip4):
6988                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6989                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6990                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6991                 break;
6992
6993         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6994                 off = si->off;
6995                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6996                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6997                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6998                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6999                         tmp_reg);
7000                 break;
7001
7002         case offsetof(struct bpf_sock_addr, user_port):
7003                 /* To get port we need to know sa_family first and then treat
7004                  * sockaddr as either sockaddr_in or sockaddr_in6.
7005                  * Though we can simplify since port field has same offset and
7006                  * size in both structures.
7007                  * Here we check this invariant and use just one of the
7008                  * structures if it's true.
7009                  */
7010                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7011                              offsetof(struct sockaddr_in6, sin6_port));
7012                 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7013                              FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7014                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7015                                                      struct sockaddr_in6, uaddr,
7016                                                      sin6_port, tmp_reg);
7017                 break;
7018
7019         case offsetof(struct bpf_sock_addr, family):
7020                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7021                                             struct sock, sk, sk_family);
7022                 break;
7023
7024         case offsetof(struct bpf_sock_addr, type):
7025                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7026                         struct bpf_sock_addr_kern, struct sock, sk,
7027                         __sk_flags_offset, BPF_W, 0);
7028                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7029                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7030                 break;
7031
7032         case offsetof(struct bpf_sock_addr, protocol):
7033                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7034                         struct bpf_sock_addr_kern, struct sock, sk,
7035                         __sk_flags_offset, BPF_W, 0);
7036                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7037                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7038                                         SK_FL_PROTO_SHIFT);
7039                 break;
7040
7041         case offsetof(struct bpf_sock_addr, msg_src_ip4):
7042                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7043                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7044                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7045                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7046                 break;
7047
7048         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7049                                 msg_src_ip6[3]):
7050                 off = si->off;
7051                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7052                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7053                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7054                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7055                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7056                 break;
7057         }
7058
7059         return insn - insn_buf;
7060 }
7061
7062 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7063                                        const struct bpf_insn *si,
7064                                        struct bpf_insn *insn_buf,
7065                                        struct bpf_prog *prog,
7066                                        u32 *target_size)
7067 {
7068         struct bpf_insn *insn = insn_buf;
7069         int off;
7070
7071         switch (si->off) {
7072         case offsetof(struct bpf_sock_ops, op) ...
7073              offsetof(struct bpf_sock_ops, replylong[3]):
7074                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
7075                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
7076                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
7077                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
7078                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
7079                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
7080                 off = si->off;
7081                 off -= offsetof(struct bpf_sock_ops, op);
7082                 off += offsetof(struct bpf_sock_ops_kern, op);
7083                 if (type == BPF_WRITE)
7084                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7085                                               off);
7086                 else
7087                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7088                                               off);
7089                 break;
7090
7091         case offsetof(struct bpf_sock_ops, family):
7092                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7093
7094                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7095                                               struct bpf_sock_ops_kern, sk),
7096                                       si->dst_reg, si->src_reg,
7097                                       offsetof(struct bpf_sock_ops_kern, sk));
7098                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7099                                       offsetof(struct sock_common, skc_family));
7100                 break;
7101
7102         case offsetof(struct bpf_sock_ops, remote_ip4):
7103                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7104
7105                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7106                                                 struct bpf_sock_ops_kern, sk),
7107                                       si->dst_reg, si->src_reg,
7108                                       offsetof(struct bpf_sock_ops_kern, sk));
7109                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7110                                       offsetof(struct sock_common, skc_daddr));
7111                 break;
7112
7113         case offsetof(struct bpf_sock_ops, local_ip4):
7114                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7115                                           skc_rcv_saddr) != 4);
7116
7117                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7118                                               struct bpf_sock_ops_kern, sk),
7119                                       si->dst_reg, si->src_reg,
7120                                       offsetof(struct bpf_sock_ops_kern, sk));
7121                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7122                                       offsetof(struct sock_common,
7123                                                skc_rcv_saddr));
7124                 break;
7125
7126         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
7127              offsetof(struct bpf_sock_ops, remote_ip6[3]):
7128 #if IS_ENABLED(CONFIG_IPV6)
7129                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7130                                           skc_v6_daddr.s6_addr32[0]) != 4);
7131
7132                 off = si->off;
7133                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
7134                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7135                                                 struct bpf_sock_ops_kern, sk),
7136                                       si->dst_reg, si->src_reg,
7137                                       offsetof(struct bpf_sock_ops_kern, sk));
7138                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7139                                       offsetof(struct sock_common,
7140                                                skc_v6_daddr.s6_addr32[0]) +
7141                                       off);
7142 #else
7143                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7144 #endif
7145                 break;
7146
7147         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
7148              offsetof(struct bpf_sock_ops, local_ip6[3]):
7149 #if IS_ENABLED(CONFIG_IPV6)
7150                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7151                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7152
7153                 off = si->off;
7154                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
7155                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7156                                                 struct bpf_sock_ops_kern, sk),
7157                                       si->dst_reg, si->src_reg,
7158                                       offsetof(struct bpf_sock_ops_kern, sk));
7159                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7160                                       offsetof(struct sock_common,
7161                                                skc_v6_rcv_saddr.s6_addr32[0]) +
7162                                       off);
7163 #else
7164                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7165 #endif
7166                 break;
7167
7168         case offsetof(struct bpf_sock_ops, remote_port):
7169                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7170
7171                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7172                                                 struct bpf_sock_ops_kern, sk),
7173                                       si->dst_reg, si->src_reg,
7174                                       offsetof(struct bpf_sock_ops_kern, sk));
7175                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7176                                       offsetof(struct sock_common, skc_dport));
7177 #ifndef __BIG_ENDIAN_BITFIELD
7178                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7179 #endif
7180                 break;
7181
7182         case offsetof(struct bpf_sock_ops, local_port):
7183                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7184
7185                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7186                                                 struct bpf_sock_ops_kern, sk),
7187                                       si->dst_reg, si->src_reg,
7188                                       offsetof(struct bpf_sock_ops_kern, sk));
7189                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7190                                       offsetof(struct sock_common, skc_num));
7191                 break;
7192
7193         case offsetof(struct bpf_sock_ops, is_fullsock):
7194                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7195                                                 struct bpf_sock_ops_kern,
7196                                                 is_fullsock),
7197                                       si->dst_reg, si->src_reg,
7198                                       offsetof(struct bpf_sock_ops_kern,
7199                                                is_fullsock));
7200                 break;
7201
7202         case offsetof(struct bpf_sock_ops, state):
7203                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
7204
7205                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7206                                                 struct bpf_sock_ops_kern, sk),
7207                                       si->dst_reg, si->src_reg,
7208                                       offsetof(struct bpf_sock_ops_kern, sk));
7209                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
7210                                       offsetof(struct sock_common, skc_state));
7211                 break;
7212
7213         case offsetof(struct bpf_sock_ops, rtt_min):
7214                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
7215                              sizeof(struct minmax));
7216                 BUILD_BUG_ON(sizeof(struct minmax) <
7217                              sizeof(struct minmax_sample));
7218
7219                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7220                                                 struct bpf_sock_ops_kern, sk),
7221                                       si->dst_reg, si->src_reg,
7222                                       offsetof(struct bpf_sock_ops_kern, sk));
7223                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7224                                       offsetof(struct tcp_sock, rtt_min) +
7225                                       FIELD_SIZEOF(struct minmax_sample, t));
7226                 break;
7227
7228 /* Helper macro for adding read access to tcp_sock or sock fields. */
7229 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
7230         do {                                                                  \
7231                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
7232                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
7233                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7234                                                 struct bpf_sock_ops_kern,     \
7235                                                 is_fullsock),                 \
7236                                       si->dst_reg, si->src_reg,               \
7237                                       offsetof(struct bpf_sock_ops_kern,      \
7238                                                is_fullsock));                 \
7239                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
7240                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7241                                                 struct bpf_sock_ops_kern, sk),\
7242                                       si->dst_reg, si->src_reg,               \
7243                                       offsetof(struct bpf_sock_ops_kern, sk));\
7244                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
7245                                                        OBJ_FIELD),            \
7246                                       si->dst_reg, si->dst_reg,               \
7247                                       offsetof(OBJ, OBJ_FIELD));              \
7248         } while (0)
7249
7250 /* Helper macro for adding write access to tcp_sock or sock fields.
7251  * The macro is called with two registers, dst_reg which contains a pointer
7252  * to ctx (context) and src_reg which contains the value that should be
7253  * stored. However, we need an additional register since we cannot overwrite
7254  * dst_reg because it may be used later in the program.
7255  * Instead we "borrow" one of the other register. We first save its value
7256  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7257  * it at the end of the macro.
7258  */
7259 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
7260         do {                                                                  \
7261                 int reg = BPF_REG_9;                                          \
7262                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
7263                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
7264                 if (si->dst_reg == reg || si->src_reg == reg)                 \
7265                         reg--;                                                \
7266                 if (si->dst_reg == reg || si->src_reg == reg)                 \
7267                         reg--;                                                \
7268                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
7269                                       offsetof(struct bpf_sock_ops_kern,      \
7270                                                temp));                        \
7271                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7272                                                 struct bpf_sock_ops_kern,     \
7273                                                 is_fullsock),                 \
7274                                       reg, si->dst_reg,                       \
7275                                       offsetof(struct bpf_sock_ops_kern,      \
7276                                                is_fullsock));                 \
7277                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
7278                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7279                                                 struct bpf_sock_ops_kern, sk),\
7280                                       reg, si->dst_reg,                       \
7281                                       offsetof(struct bpf_sock_ops_kern, sk));\
7282                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
7283                                       reg, si->src_reg,                       \
7284                                       offsetof(OBJ, OBJ_FIELD));              \
7285                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
7286                                       offsetof(struct bpf_sock_ops_kern,      \
7287                                                temp));                        \
7288         } while (0)
7289
7290 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
7291         do {                                                                  \
7292                 if (TYPE == BPF_WRITE)                                        \
7293                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
7294                 else                                                          \
7295                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
7296         } while (0)
7297
7298         case offsetof(struct bpf_sock_ops, snd_cwnd):
7299                 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
7300                 break;
7301
7302         case offsetof(struct bpf_sock_ops, srtt_us):
7303                 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
7304                 break;
7305
7306         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
7307                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
7308                                    struct tcp_sock);
7309                 break;
7310
7311         case offsetof(struct bpf_sock_ops, snd_ssthresh):
7312                 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
7313                 break;
7314
7315         case offsetof(struct bpf_sock_ops, rcv_nxt):
7316                 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
7317                 break;
7318
7319         case offsetof(struct bpf_sock_ops, snd_nxt):
7320                 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
7321                 break;
7322
7323         case offsetof(struct bpf_sock_ops, snd_una):
7324                 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
7325                 break;
7326
7327         case offsetof(struct bpf_sock_ops, mss_cache):
7328                 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
7329                 break;
7330
7331         case offsetof(struct bpf_sock_ops, ecn_flags):
7332                 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
7333                 break;
7334
7335         case offsetof(struct bpf_sock_ops, rate_delivered):
7336                 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
7337                                    struct tcp_sock);
7338                 break;
7339
7340         case offsetof(struct bpf_sock_ops, rate_interval_us):
7341                 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
7342                                    struct tcp_sock);
7343                 break;
7344
7345         case offsetof(struct bpf_sock_ops, packets_out):
7346                 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
7347                 break;
7348
7349         case offsetof(struct bpf_sock_ops, retrans_out):
7350                 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
7351                 break;
7352
7353         case offsetof(struct bpf_sock_ops, total_retrans):
7354                 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
7355                                    struct tcp_sock);
7356                 break;
7357
7358         case offsetof(struct bpf_sock_ops, segs_in):
7359                 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
7360                 break;
7361
7362         case offsetof(struct bpf_sock_ops, data_segs_in):
7363                 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
7364                 break;
7365
7366         case offsetof(struct bpf_sock_ops, segs_out):
7367                 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
7368                 break;
7369
7370         case offsetof(struct bpf_sock_ops, data_segs_out):
7371                 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
7372                                    struct tcp_sock);
7373                 break;
7374
7375         case offsetof(struct bpf_sock_ops, lost_out):
7376                 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
7377                 break;
7378
7379         case offsetof(struct bpf_sock_ops, sacked_out):
7380                 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
7381                 break;
7382
7383         case offsetof(struct bpf_sock_ops, sk_txhash):
7384                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
7385                                           struct sock, type);
7386                 break;
7387
7388         case offsetof(struct bpf_sock_ops, bytes_received):
7389                 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
7390                                    struct tcp_sock);
7391                 break;
7392
7393         case offsetof(struct bpf_sock_ops, bytes_acked):
7394                 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
7395                 break;
7396
7397         }
7398         return insn - insn_buf;
7399 }
7400
7401 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
7402                                      const struct bpf_insn *si,
7403                                      struct bpf_insn *insn_buf,
7404                                      struct bpf_prog *prog, u32 *target_size)
7405 {
7406         struct bpf_insn *insn = insn_buf;
7407         int off;
7408
7409         switch (si->off) {
7410         case offsetof(struct __sk_buff, data_end):
7411                 off  = si->off;
7412                 off -= offsetof(struct __sk_buff, data_end);
7413                 off += offsetof(struct sk_buff, cb);
7414                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
7415                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7416                                       si->src_reg, off);
7417                 break;
7418         default:
7419                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7420                                               target_size);
7421         }
7422
7423         return insn - insn_buf;
7424 }
7425
7426 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
7427                                      const struct bpf_insn *si,
7428                                      struct bpf_insn *insn_buf,
7429                                      struct bpf_prog *prog, u32 *target_size)
7430 {
7431         struct bpf_insn *insn = insn_buf;
7432 #if IS_ENABLED(CONFIG_IPV6)
7433         int off;
7434 #endif
7435
7436         /* convert ctx uses the fact sg element is first in struct */
7437         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
7438
7439         switch (si->off) {
7440         case offsetof(struct sk_msg_md, data):
7441                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
7442                                       si->dst_reg, si->src_reg,
7443                                       offsetof(struct sk_msg, data));
7444                 break;
7445         case offsetof(struct sk_msg_md, data_end):
7446                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
7447                                       si->dst_reg, si->src_reg,
7448                                       offsetof(struct sk_msg, data_end));
7449                 break;
7450         case offsetof(struct sk_msg_md, family):
7451                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7452
7453                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7454                                               struct sk_msg, sk),
7455                                       si->dst_reg, si->src_reg,
7456                                       offsetof(struct sk_msg, sk));
7457                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7458                                       offsetof(struct sock_common, skc_family));
7459                 break;
7460
7461         case offsetof(struct sk_msg_md, remote_ip4):
7462                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7463
7464                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7465                                                 struct sk_msg, sk),
7466                                       si->dst_reg, si->src_reg,
7467                                       offsetof(struct sk_msg, sk));
7468                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7469                                       offsetof(struct sock_common, skc_daddr));
7470                 break;
7471
7472         case offsetof(struct sk_msg_md, local_ip4):
7473                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7474                                           skc_rcv_saddr) != 4);
7475
7476                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7477                                               struct sk_msg, sk),
7478                                       si->dst_reg, si->src_reg,
7479                                       offsetof(struct sk_msg, sk));
7480                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7481                                       offsetof(struct sock_common,
7482                                                skc_rcv_saddr));
7483                 break;
7484
7485         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
7486              offsetof(struct sk_msg_md, remote_ip6[3]):
7487 #if IS_ENABLED(CONFIG_IPV6)
7488                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7489                                           skc_v6_daddr.s6_addr32[0]) != 4);
7490
7491                 off = si->off;
7492                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
7493                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7494                                                 struct sk_msg, sk),
7495                                       si->dst_reg, si->src_reg,
7496                                       offsetof(struct sk_msg, sk));
7497                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7498                                       offsetof(struct sock_common,
7499                                                skc_v6_daddr.s6_addr32[0]) +
7500                                       off);
7501 #else
7502                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7503 #endif
7504                 break;
7505
7506         case offsetof(struct sk_msg_md, local_ip6[0]) ...
7507              offsetof(struct sk_msg_md, local_ip6[3]):
7508 #if IS_ENABLED(CONFIG_IPV6)
7509                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7510                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7511
7512                 off = si->off;
7513                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
7514                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7515                                                 struct sk_msg, sk),
7516                                       si->dst_reg, si->src_reg,
7517                                       offsetof(struct sk_msg, sk));
7518                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7519                                       offsetof(struct sock_common,
7520                                                skc_v6_rcv_saddr.s6_addr32[0]) +
7521                                       off);
7522 #else
7523                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7524 #endif
7525                 break;
7526
7527         case offsetof(struct sk_msg_md, remote_port):
7528                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7529
7530                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7531                                                 struct sk_msg, sk),
7532                                       si->dst_reg, si->src_reg,
7533                                       offsetof(struct sk_msg, sk));
7534                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7535                                       offsetof(struct sock_common, skc_dport));
7536 #ifndef __BIG_ENDIAN_BITFIELD
7537                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7538 #endif
7539                 break;
7540
7541         case offsetof(struct sk_msg_md, local_port):
7542                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7543
7544                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7545                                                 struct sk_msg, sk),
7546                                       si->dst_reg, si->src_reg,
7547                                       offsetof(struct sk_msg, sk));
7548                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7549                                       offsetof(struct sock_common, skc_num));
7550                 break;
7551
7552         case offsetof(struct sk_msg_md, size):
7553                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
7554                                       si->dst_reg, si->src_reg,
7555                                       offsetof(struct sk_msg_sg, size));
7556                 break;
7557         }
7558
7559         return insn - insn_buf;
7560 }
7561
7562 const struct bpf_verifier_ops sk_filter_verifier_ops = {
7563         .get_func_proto         = sk_filter_func_proto,
7564         .is_valid_access        = sk_filter_is_valid_access,
7565         .convert_ctx_access     = bpf_convert_ctx_access,
7566         .gen_ld_abs             = bpf_gen_ld_abs,
7567 };
7568
7569 const struct bpf_prog_ops sk_filter_prog_ops = {
7570         .test_run               = bpf_prog_test_run_skb,
7571 };
7572
7573 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
7574         .get_func_proto         = tc_cls_act_func_proto,
7575         .is_valid_access        = tc_cls_act_is_valid_access,
7576         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
7577         .gen_prologue           = tc_cls_act_prologue,
7578         .gen_ld_abs             = bpf_gen_ld_abs,
7579 };
7580
7581 const struct bpf_prog_ops tc_cls_act_prog_ops = {
7582         .test_run               = bpf_prog_test_run_skb,
7583 };
7584
7585 const struct bpf_verifier_ops xdp_verifier_ops = {
7586         .get_func_proto         = xdp_func_proto,
7587         .is_valid_access        = xdp_is_valid_access,
7588         .convert_ctx_access     = xdp_convert_ctx_access,
7589         .gen_prologue           = bpf_noop_prologue,
7590 };
7591
7592 const struct bpf_prog_ops xdp_prog_ops = {
7593         .test_run               = bpf_prog_test_run_xdp,
7594 };
7595
7596 const struct bpf_verifier_ops cg_skb_verifier_ops = {
7597         .get_func_proto         = cg_skb_func_proto,
7598         .is_valid_access        = cg_skb_is_valid_access,
7599         .convert_ctx_access     = bpf_convert_ctx_access,
7600 };
7601
7602 const struct bpf_prog_ops cg_skb_prog_ops = {
7603         .test_run               = bpf_prog_test_run_skb,
7604 };
7605
7606 const struct bpf_verifier_ops lwt_in_verifier_ops = {
7607         .get_func_proto         = lwt_in_func_proto,
7608         .is_valid_access        = lwt_is_valid_access,
7609         .convert_ctx_access     = bpf_convert_ctx_access,
7610 };
7611
7612 const struct bpf_prog_ops lwt_in_prog_ops = {
7613         .test_run               = bpf_prog_test_run_skb,
7614 };
7615
7616 const struct bpf_verifier_ops lwt_out_verifier_ops = {
7617         .get_func_proto         = lwt_out_func_proto,
7618         .is_valid_access        = lwt_is_valid_access,
7619         .convert_ctx_access     = bpf_convert_ctx_access,
7620 };
7621
7622 const struct bpf_prog_ops lwt_out_prog_ops = {
7623         .test_run               = bpf_prog_test_run_skb,
7624 };
7625
7626 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
7627         .get_func_proto         = lwt_xmit_func_proto,
7628         .is_valid_access        = lwt_is_valid_access,
7629         .convert_ctx_access     = bpf_convert_ctx_access,
7630         .gen_prologue           = tc_cls_act_prologue,
7631 };
7632
7633 const struct bpf_prog_ops lwt_xmit_prog_ops = {
7634         .test_run               = bpf_prog_test_run_skb,
7635 };
7636
7637 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
7638         .get_func_proto         = lwt_seg6local_func_proto,
7639         .is_valid_access        = lwt_is_valid_access,
7640         .convert_ctx_access     = bpf_convert_ctx_access,
7641 };
7642
7643 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
7644         .test_run               = bpf_prog_test_run_skb,
7645 };
7646
7647 const struct bpf_verifier_ops cg_sock_verifier_ops = {
7648         .get_func_proto         = sock_filter_func_proto,
7649         .is_valid_access        = sock_filter_is_valid_access,
7650         .convert_ctx_access     = bpf_sock_convert_ctx_access,
7651 };
7652
7653 const struct bpf_prog_ops cg_sock_prog_ops = {
7654 };
7655
7656 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
7657         .get_func_proto         = sock_addr_func_proto,
7658         .is_valid_access        = sock_addr_is_valid_access,
7659         .convert_ctx_access     = sock_addr_convert_ctx_access,
7660 };
7661
7662 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
7663 };
7664
7665 const struct bpf_verifier_ops sock_ops_verifier_ops = {
7666         .get_func_proto         = sock_ops_func_proto,
7667         .is_valid_access        = sock_ops_is_valid_access,
7668         .convert_ctx_access     = sock_ops_convert_ctx_access,
7669 };
7670
7671 const struct bpf_prog_ops sock_ops_prog_ops = {
7672 };
7673
7674 const struct bpf_verifier_ops sk_skb_verifier_ops = {
7675         .get_func_proto         = sk_skb_func_proto,
7676         .is_valid_access        = sk_skb_is_valid_access,
7677         .convert_ctx_access     = sk_skb_convert_ctx_access,
7678         .gen_prologue           = sk_skb_prologue,
7679 };
7680
7681 const struct bpf_prog_ops sk_skb_prog_ops = {
7682 };
7683
7684 const struct bpf_verifier_ops sk_msg_verifier_ops = {
7685         .get_func_proto         = sk_msg_func_proto,
7686         .is_valid_access        = sk_msg_is_valid_access,
7687         .convert_ctx_access     = sk_msg_convert_ctx_access,
7688         .gen_prologue           = bpf_noop_prologue,
7689 };
7690
7691 const struct bpf_prog_ops sk_msg_prog_ops = {
7692 };
7693
7694 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
7695         .get_func_proto         = flow_dissector_func_proto,
7696         .is_valid_access        = flow_dissector_is_valid_access,
7697         .convert_ctx_access     = bpf_convert_ctx_access,
7698 };
7699
7700 const struct bpf_prog_ops flow_dissector_prog_ops = {
7701 };
7702
7703 int sk_detach_filter(struct sock *sk)
7704 {
7705         int ret = -ENOENT;
7706         struct sk_filter *filter;
7707
7708         if (sock_flag(sk, SOCK_FILTER_LOCKED))
7709                 return -EPERM;
7710
7711         filter = rcu_dereference_protected(sk->sk_filter,
7712                                            lockdep_sock_is_held(sk));
7713         if (filter) {
7714                 RCU_INIT_POINTER(sk->sk_filter, NULL);
7715                 sk_filter_uncharge(sk, filter);
7716                 ret = 0;
7717         }
7718
7719         return ret;
7720 }
7721 EXPORT_SYMBOL_GPL(sk_detach_filter);
7722
7723 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7724                   unsigned int len)
7725 {
7726         struct sock_fprog_kern *fprog;
7727         struct sk_filter *filter;
7728         int ret = 0;
7729
7730         lock_sock(sk);
7731         filter = rcu_dereference_protected(sk->sk_filter,
7732                                            lockdep_sock_is_held(sk));
7733         if (!filter)
7734                 goto out;
7735
7736         /* We're copying the filter that has been originally attached,
7737          * so no conversion/decode needed anymore. eBPF programs that
7738          * have no original program cannot be dumped through this.
7739          */
7740         ret = -EACCES;
7741         fprog = filter->prog->orig_prog;
7742         if (!fprog)
7743                 goto out;
7744
7745         ret = fprog->len;
7746         if (!len)
7747                 /* User space only enquires number of filter blocks. */
7748                 goto out;
7749
7750         ret = -EINVAL;
7751         if (len < fprog->len)
7752                 goto out;
7753
7754         ret = -EFAULT;
7755         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
7756                 goto out;
7757
7758         /* Instead of bytes, the API requests to return the number
7759          * of filter blocks.
7760          */
7761         ret = fprog->len;
7762 out:
7763         release_sock(sk);
7764         return ret;
7765 }
7766
7767 #ifdef CONFIG_INET
7768 struct sk_reuseport_kern {
7769         struct sk_buff *skb;
7770         struct sock *sk;
7771         struct sock *selected_sk;
7772         void *data_end;
7773         u32 hash;
7774         u32 reuseport_id;
7775         bool bind_inany;
7776 };
7777
7778 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
7779                                     struct sock_reuseport *reuse,
7780                                     struct sock *sk, struct sk_buff *skb,
7781                                     u32 hash)
7782 {
7783         reuse_kern->skb = skb;
7784         reuse_kern->sk = sk;
7785         reuse_kern->selected_sk = NULL;
7786         reuse_kern->data_end = skb->data + skb_headlen(skb);
7787         reuse_kern->hash = hash;
7788         reuse_kern->reuseport_id = reuse->reuseport_id;
7789         reuse_kern->bind_inany = reuse->bind_inany;
7790 }
7791
7792 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
7793                                   struct bpf_prog *prog, struct sk_buff *skb,
7794                                   u32 hash)
7795 {
7796         struct sk_reuseport_kern reuse_kern;
7797         enum sk_action action;
7798
7799         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
7800         action = BPF_PROG_RUN(prog, &reuse_kern);
7801
7802         if (action == SK_PASS)
7803                 return reuse_kern.selected_sk;
7804         else
7805                 return ERR_PTR(-ECONNREFUSED);
7806 }
7807
7808 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
7809            struct bpf_map *, map, void *, key, u32, flags)
7810 {
7811         struct sock_reuseport *reuse;
7812         struct sock *selected_sk;
7813
7814         selected_sk = map->ops->map_lookup_elem(map, key);
7815         if (!selected_sk)
7816                 return -ENOENT;
7817
7818         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
7819         if (!reuse)
7820                 /* selected_sk is unhashed (e.g. by close()) after the
7821                  * above map_lookup_elem().  Treat selected_sk has already
7822                  * been removed from the map.
7823                  */
7824                 return -ENOENT;
7825
7826         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
7827                 struct sock *sk;
7828
7829                 if (unlikely(!reuse_kern->reuseport_id))
7830                         /* There is a small race between adding the
7831                          * sk to the map and setting the
7832                          * reuse_kern->reuseport_id.
7833                          * Treat it as the sk has not been added to
7834                          * the bpf map yet.
7835                          */
7836                         return -ENOENT;
7837
7838                 sk = reuse_kern->sk;
7839                 if (sk->sk_protocol != selected_sk->sk_protocol)
7840                         return -EPROTOTYPE;
7841                 else if (sk->sk_family != selected_sk->sk_family)
7842                         return -EAFNOSUPPORT;
7843
7844                 /* Catch all. Likely bound to a different sockaddr. */
7845                 return -EBADFD;
7846         }
7847
7848         reuse_kern->selected_sk = selected_sk;
7849
7850         return 0;
7851 }
7852
7853 static const struct bpf_func_proto sk_select_reuseport_proto = {
7854         .func           = sk_select_reuseport,
7855         .gpl_only       = false,
7856         .ret_type       = RET_INTEGER,
7857         .arg1_type      = ARG_PTR_TO_CTX,
7858         .arg2_type      = ARG_CONST_MAP_PTR,
7859         .arg3_type      = ARG_PTR_TO_MAP_KEY,
7860         .arg4_type      = ARG_ANYTHING,
7861 };
7862
7863 BPF_CALL_4(sk_reuseport_load_bytes,
7864            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7865            void *, to, u32, len)
7866 {
7867         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
7868 }
7869
7870 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
7871         .func           = sk_reuseport_load_bytes,
7872         .gpl_only       = false,
7873         .ret_type       = RET_INTEGER,
7874         .arg1_type      = ARG_PTR_TO_CTX,
7875         .arg2_type      = ARG_ANYTHING,
7876         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7877         .arg4_type      = ARG_CONST_SIZE,
7878 };
7879
7880 BPF_CALL_5(sk_reuseport_load_bytes_relative,
7881            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7882            void *, to, u32, len, u32, start_header)
7883 {
7884         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
7885                                                len, start_header);
7886 }
7887
7888 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
7889         .func           = sk_reuseport_load_bytes_relative,
7890         .gpl_only       = false,
7891         .ret_type       = RET_INTEGER,
7892         .arg1_type      = ARG_PTR_TO_CTX,
7893         .arg2_type      = ARG_ANYTHING,
7894         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7895         .arg4_type      = ARG_CONST_SIZE,
7896         .arg5_type      = ARG_ANYTHING,
7897 };
7898
7899 static const struct bpf_func_proto *
7900 sk_reuseport_func_proto(enum bpf_func_id func_id,
7901                         const struct bpf_prog *prog)
7902 {
7903         switch (func_id) {
7904         case BPF_FUNC_sk_select_reuseport:
7905                 return &sk_select_reuseport_proto;
7906         case BPF_FUNC_skb_load_bytes:
7907                 return &sk_reuseport_load_bytes_proto;
7908         case BPF_FUNC_skb_load_bytes_relative:
7909                 return &sk_reuseport_load_bytes_relative_proto;
7910         default:
7911                 return bpf_base_func_proto(func_id);
7912         }
7913 }
7914
7915 static bool
7916 sk_reuseport_is_valid_access(int off, int size,
7917                              enum bpf_access_type type,
7918                              const struct bpf_prog *prog,
7919                              struct bpf_insn_access_aux *info)
7920 {
7921         const u32 size_default = sizeof(__u32);
7922
7923         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
7924             off % size || type != BPF_READ)
7925                 return false;
7926
7927         switch (off) {
7928         case offsetof(struct sk_reuseport_md, data):
7929                 info->reg_type = PTR_TO_PACKET;
7930                 return size == sizeof(__u64);
7931
7932         case offsetof(struct sk_reuseport_md, data_end):
7933                 info->reg_type = PTR_TO_PACKET_END;
7934                 return size == sizeof(__u64);
7935
7936         case offsetof(struct sk_reuseport_md, hash):
7937                 return size == size_default;
7938
7939         /* Fields that allow narrowing */
7940         case offsetof(struct sk_reuseport_md, eth_protocol):
7941                 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
7942                         return false;
7943                 /* fall through */
7944         case offsetof(struct sk_reuseport_md, ip_protocol):
7945         case offsetof(struct sk_reuseport_md, bind_inany):
7946         case offsetof(struct sk_reuseport_md, len):
7947                 bpf_ctx_record_field_size(info, size_default);
7948                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7949
7950         default:
7951                 return false;
7952         }
7953 }
7954
7955 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
7956         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7957                               si->dst_reg, si->src_reg,                 \
7958                               bpf_target_off(struct sk_reuseport_kern, F, \
7959                                              FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7960                                              target_size));             \
7961         })
7962
7963 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
7964         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
7965                                     struct sk_buff,                     \
7966                                     skb,                                \
7967                                     SKB_FIELD)
7968
7969 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
7970         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern,  \
7971                                              struct sock,               \
7972                                              sk,                        \
7973                                              SK_FIELD, BPF_SIZE, EXTRA_OFF)
7974
7975 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
7976                                            const struct bpf_insn *si,
7977                                            struct bpf_insn *insn_buf,
7978                                            struct bpf_prog *prog,
7979                                            u32 *target_size)
7980 {
7981         struct bpf_insn *insn = insn_buf;
7982
7983         switch (si->off) {
7984         case offsetof(struct sk_reuseport_md, data):
7985                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
7986                 break;
7987
7988         case offsetof(struct sk_reuseport_md, len):
7989                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
7990                 break;
7991
7992         case offsetof(struct sk_reuseport_md, eth_protocol):
7993                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
7994                 break;
7995
7996         case offsetof(struct sk_reuseport_md, ip_protocol):
7997                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7998                 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
7999                                                     BPF_W, 0);
8000                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8001                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8002                                         SK_FL_PROTO_SHIFT);
8003                 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
8004                  * aware.  No further narrowing or masking is needed.
8005                  */
8006                 *target_size = 1;
8007                 break;
8008
8009         case offsetof(struct sk_reuseport_md, data_end):
8010                 SK_REUSEPORT_LOAD_FIELD(data_end);
8011                 break;
8012
8013         case offsetof(struct sk_reuseport_md, hash):
8014                 SK_REUSEPORT_LOAD_FIELD(hash);
8015                 break;
8016
8017         case offsetof(struct sk_reuseport_md, bind_inany):
8018                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8019                 break;
8020         }
8021
8022         return insn - insn_buf;
8023 }
8024
8025 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8026         .get_func_proto         = sk_reuseport_func_proto,
8027         .is_valid_access        = sk_reuseport_is_valid_access,
8028         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
8029 };
8030
8031 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8032 };
8033 #endif /* CONFIG_INET */