net/core/filter.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Linux Socket Filter - Kernel level socket filtering
   4  *
   5  * Based on the design of the Berkeley Packet Filter. The new
   6  * internal format has been designed by PLUMgrid:
   7  *
   8  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   9  *
  10  * Authors:
  11  *
  12  *      Jay Schulist <jschlst@samba.org>
  13  *      Alexei Starovoitov <ast@plumgrid.com>
  14  *      Daniel Borkmann <dborkman@redhat.com>
  15  *
  16  * Andi Kleen - Fix a few bad bugs and races.
  17  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/types.h>
  22 #include <linux/mm.h>
  23 #include <linux/fcntl.h>
  24 #include <linux/socket.h>
  25 #include <linux/sock_diag.h>
  26 #include <linux/in.h>
  27 #include <linux/inet.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/if_packet.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/gfp.h>
  32 #include <net/inet_common.h>
  33 #include <net/ip.h>
  34 #include <net/protocol.h>
  35 #include <net/netlink.h>
  36 #include <linux/skbuff.h>
  37 #include <linux/skmsg.h>
  38 #include <net/sock.h>
  39 #include <net/flow_dissector.h>
  40 #include <linux/errno.h>
  41 #include <linux/timer.h>
  42 #include <linux/uaccess.h>
  43 #include <asm/unaligned.h>
  44 #include <asm/cmpxchg.h>
  45 #include <linux/filter.h>
  46 #include <linux/ratelimit.h>
  47 #include <linux/seccomp.h>
  48 #include <linux/if_vlan.h>
  49 #include <linux/bpf.h>
  50 #include <net/sch_generic.h>
  51 #include <net/cls_cgroup.h>
  52 #include <net/dst_metadata.h>
  53 #include <net/dst.h>
  54 #include <net/sock_reuseport.h>
  55 #include <net/busy_poll.h>
  56 #include <net/tcp.h>
  57 #include <net/xfrm.h>
  58 #include <net/udp.h>
  59 #include <linux/bpf_trace.h>
  60 #include <net/xdp_sock.h>
  61 #include <linux/inetdevice.h>
  62 #include <net/inet_hashtables.h>
  63 #include <net/inet6_hashtables.h>
  64 #include <net/ip_fib.h>
  65 #include <net/nexthop.h>
  66 #include <net/flow.h>
  67 #include <net/arp.h>
  68 #include <net/ipv6.h>
  69 #include <net/net_namespace.h>
  70 #include <linux/seg6_local.h>
  71 #include <net/seg6.h>
  72 #include <net/seg6_local.h>
  73 #include <net/lwtunnel.h>
  74 #include <net/ipv6_stubs.h>
  75 #include <net/bpf_sk_storage.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(sizeof_field(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(sizeof_field(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(sizeof_field(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(sizeof_field(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(sizeof_field(struct net_device, ifindex) != 4);
 342                 BUILD_BUG_ON(sizeof_field(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(sizeof_field(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(sizeof_field(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 (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_4(bpf_flow_dissector_load_bytes,
1732            const struct bpf_flow_dissector *, ctx, u32, offset,
1733            void *, to, u32, len)
1734 {
1735         void *ptr;
1736
1737         if (unlikely(offset > 0xffff))
1738                 goto err_clear;
1739
1740         if (unlikely(!ctx->skb))
1741                 goto err_clear;
1742
1743         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1744         if (unlikely(!ptr))
1745                 goto err_clear;
1746         if (ptr != to)
1747                 memcpy(to, ptr, len);
1748
1749         return 0;
1750 err_clear:
1751         memset(to, 0, len);
1752         return -EFAULT;
1753 }
1754
1755 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1756         .func           = bpf_flow_dissector_load_bytes,
1757         .gpl_only       = false,
1758         .ret_type       = RET_INTEGER,
1759         .arg1_type      = ARG_PTR_TO_CTX,
1760         .arg2_type      = ARG_ANYTHING,
1761         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1762         .arg4_type      = ARG_CONST_SIZE,
1763 };
1764
1765 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1766            u32, offset, void *, to, u32, len, u32, start_header)
1767 {
1768         u8 *end = skb_tail_pointer(skb);
1769         u8 *net = skb_network_header(skb);
1770         u8 *mac = skb_mac_header(skb);
1771         u8 *ptr;
1772
1773         if (unlikely(offset > 0xffff || len > (end - mac)))
1774                 goto err_clear;
1775
1776         switch (start_header) {
1777         case BPF_HDR_START_MAC:
1778                 ptr = mac + offset;
1779                 break;
1780         case BPF_HDR_START_NET:
1781                 ptr = net + offset;
1782                 break;
1783         default:
1784                 goto err_clear;
1785         }
1786
1787         if (likely(ptr >= mac && ptr + len <= end)) {
1788                 memcpy(to, ptr, len);
1789                 return 0;
1790         }
1791
1792 err_clear:
1793         memset(to, 0, len);
1794         return -EFAULT;
1795 }
1796
1797 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1798         .func           = bpf_skb_load_bytes_relative,
1799         .gpl_only       = false,
1800         .ret_type       = RET_INTEGER,
1801         .arg1_type      = ARG_PTR_TO_CTX,
1802         .arg2_type      = ARG_ANYTHING,
1803         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1804         .arg4_type      = ARG_CONST_SIZE,
1805         .arg5_type      = ARG_ANYTHING,
1806 };
1807
1808 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1809 {
1810         /* Idea is the following: should the needed direct read/write
1811          * test fail during runtime, we can pull in more data and redo
1812          * again, since implicitly, we invalidate previous checks here.
1813          *
1814          * Or, since we know how much we need to make read/writeable,
1815          * this can be done once at the program beginning for direct
1816          * access case. By this we overcome limitations of only current
1817          * headroom being accessible.
1818          */
1819         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1820 }
1821
1822 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1823         .func           = bpf_skb_pull_data,
1824         .gpl_only       = false,
1825         .ret_type       = RET_INTEGER,
1826         .arg1_type      = ARG_PTR_TO_CTX,
1827         .arg2_type      = ARG_ANYTHING,
1828 };
1829
1830 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1831 {
1832         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1833 }
1834
1835 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1836         .func           = bpf_sk_fullsock,
1837         .gpl_only       = false,
1838         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1839         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1840 };
1841
1842 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1843                                            unsigned int write_len)
1844 {
1845         int err = __bpf_try_make_writable(skb, write_len);
1846
1847         bpf_compute_data_end_sk_skb(skb);
1848         return err;
1849 }
1850
1851 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1852 {
1853         /* Idea is the following: should the needed direct read/write
1854          * test fail during runtime, we can pull in more data and redo
1855          * again, since implicitly, we invalidate previous checks here.
1856          *
1857          * Or, since we know how much we need to make read/writeable,
1858          * this can be done once at the program beginning for direct
1859          * access case. By this we overcome limitations of only current
1860          * headroom being accessible.
1861          */
1862         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1863 }
1864
1865 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1866         .func           = sk_skb_pull_data,
1867         .gpl_only       = false,
1868         .ret_type       = RET_INTEGER,
1869         .arg1_type      = ARG_PTR_TO_CTX,
1870         .arg2_type      = ARG_ANYTHING,
1871 };
1872
1873 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1874            u64, from, u64, to, u64, flags)
1875 {
1876         __sum16 *ptr;
1877
1878         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1879                 return -EINVAL;
1880         if (unlikely(offset > 0xffff || offset & 1))
1881                 return -EFAULT;
1882         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1883                 return -EFAULT;
1884
1885         ptr = (__sum16 *)(skb->data + offset);
1886         switch (flags & BPF_F_HDR_FIELD_MASK) {
1887         case 0:
1888                 if (unlikely(from != 0))
1889                         return -EINVAL;
1890
1891                 csum_replace_by_diff(ptr, to);
1892                 break;
1893         case 2:
1894                 csum_replace2(ptr, from, to);
1895                 break;
1896         case 4:
1897                 csum_replace4(ptr, from, to);
1898                 break;
1899         default:
1900                 return -EINVAL;
1901         }
1902
1903         return 0;
1904 }
1905
1906 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1907         .func           = bpf_l3_csum_replace,
1908         .gpl_only       = false,
1909         .ret_type       = RET_INTEGER,
1910         .arg1_type      = ARG_PTR_TO_CTX,
1911         .arg2_type      = ARG_ANYTHING,
1912         .arg3_type      = ARG_ANYTHING,
1913         .arg4_type      = ARG_ANYTHING,
1914         .arg5_type      = ARG_ANYTHING,
1915 };
1916
1917 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1918            u64, from, u64, to, u64, flags)
1919 {
1920         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1921         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1922         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1923         __sum16 *ptr;
1924
1925         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1926                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1927                 return -EINVAL;
1928         if (unlikely(offset > 0xffff || offset & 1))
1929                 return -EFAULT;
1930         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1931                 return -EFAULT;
1932
1933         ptr = (__sum16 *)(skb->data + offset);
1934         if (is_mmzero && !do_mforce && !*ptr)
1935                 return 0;
1936
1937         switch (flags & BPF_F_HDR_FIELD_MASK) {
1938         case 0:
1939                 if (unlikely(from != 0))
1940                         return -EINVAL;
1941
1942                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1943                 break;
1944         case 2:
1945                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1946                 break;
1947         case 4:
1948                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1949                 break;
1950         default:
1951                 return -EINVAL;
1952         }
1953
1954         if (is_mmzero && !*ptr)
1955                 *ptr = CSUM_MANGLED_0;
1956         return 0;
1957 }
1958
1959 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1960         .func           = bpf_l4_csum_replace,
1961         .gpl_only       = false,
1962         .ret_type       = RET_INTEGER,
1963         .arg1_type      = ARG_PTR_TO_CTX,
1964         .arg2_type      = ARG_ANYTHING,
1965         .arg3_type      = ARG_ANYTHING,
1966         .arg4_type      = ARG_ANYTHING,
1967         .arg5_type      = ARG_ANYTHING,
1968 };
1969
1970 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1971            __be32 *, to, u32, to_size, __wsum, seed)
1972 {
1973         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1974         u32 diff_size = from_size + to_size;
1975         int i, j = 0;
1976
1977         /* This is quite flexible, some examples:
1978          *
1979          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1980          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1981          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
1982          *
1983          * Even for diffing, from_size and to_size don't need to be equal.
1984          */
1985         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1986                      diff_size > sizeof(sp->diff)))
1987                 return -EINVAL;
1988
1989         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1990                 sp->diff[j] = ~from[i];
1991         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
1992                 sp->diff[j] = to[i];
1993
1994         return csum_partial(sp->diff, diff_size, seed);
1995 }
1996
1997 static const struct bpf_func_proto bpf_csum_diff_proto = {
1998         .func           = bpf_csum_diff,
1999         .gpl_only       = false,
2000         .pkt_access     = true,
2001         .ret_type       = RET_INTEGER,
2002         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
2003         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2004         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
2005         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2006         .arg5_type      = ARG_ANYTHING,
2007 };
2008
2009 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2010 {
2011         /* The interface is to be used in combination with bpf_csum_diff()
2012          * for direct packet writes. csum rotation for alignment as well
2013          * as emulating csum_sub() can be done from the eBPF program.
2014          */
2015         if (skb->ip_summed == CHECKSUM_COMPLETE)
2016                 return (skb->csum = csum_add(skb->csum, csum));
2017
2018         return -ENOTSUPP;
2019 }
2020
2021 static const struct bpf_func_proto bpf_csum_update_proto = {
2022         .func           = bpf_csum_update,
2023         .gpl_only       = false,
2024         .ret_type       = RET_INTEGER,
2025         .arg1_type      = ARG_PTR_TO_CTX,
2026         .arg2_type      = ARG_ANYTHING,
2027 };
2028
2029 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2030 {
2031         return dev_forward_skb(dev, skb);
2032 }
2033
2034 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2035                                       struct sk_buff *skb)
2036 {
2037         int ret = ____dev_forward_skb(dev, skb);
2038
2039         if (likely(!ret)) {
2040                 skb->dev = dev;
2041                 ret = netif_rx(skb);
2042         }
2043
2044         return ret;
2045 }
2046
2047 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2048 {
2049         int ret;
2050
2051         if (dev_xmit_recursion()) {
2052                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2053                 kfree_skb(skb);
2054                 return -ENETDOWN;
2055         }
2056
2057         skb->dev = dev;
2058         skb->tstamp = 0;
2059
2060         dev_xmit_recursion_inc();
2061         ret = dev_queue_xmit(skb);
2062         dev_xmit_recursion_dec();
2063
2064         return ret;
2065 }
2066
2067 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2068                                  u32 flags)
2069 {
2070         unsigned int mlen = skb_network_offset(skb);
2071
2072         if (mlen) {
2073                 __skb_pull(skb, mlen);
2074
2075                 /* At ingress, the mac header has already been pulled once.
2076                  * At egress, skb_pospull_rcsum has to be done in case that
2077                  * the skb is originated from ingress (i.e. a forwarded skb)
2078                  * to ensure that rcsum starts at net header.
2079                  */
2080                 if (!skb_at_tc_ingress(skb))
2081                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2082         }
2083         skb_pop_mac_header(skb);
2084         skb_reset_mac_len(skb);
2085         return flags & BPF_F_INGRESS ?
2086                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2087 }
2088
2089 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2090                                  u32 flags)
2091 {
2092         /* Verify that a link layer header is carried */
2093         if (unlikely(skb->mac_header >= skb->network_header)) {
2094                 kfree_skb(skb);
2095                 return -ERANGE;
2096         }
2097
2098         bpf_push_mac_rcsum(skb);
2099         return flags & BPF_F_INGRESS ?
2100                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2101 }
2102
2103 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2104                           u32 flags)
2105 {
2106         if (dev_is_mac_header_xmit(dev))
2107                 return __bpf_redirect_common(skb, dev, flags);
2108         else
2109                 return __bpf_redirect_no_mac(skb, dev, flags);
2110 }
2111
2112 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2113 {
2114         struct net_device *dev;
2115         struct sk_buff *clone;
2116         int ret;
2117
2118         if (unlikely(flags & ~(BPF_F_INGRESS)))
2119                 return -EINVAL;
2120
2121         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2122         if (unlikely(!dev))
2123                 return -EINVAL;
2124
2125         clone = skb_clone(skb, GFP_ATOMIC);
2126         if (unlikely(!clone))
2127                 return -ENOMEM;
2128
2129         /* For direct write, we need to keep the invariant that the skbs
2130          * we're dealing with need to be uncloned. Should uncloning fail
2131          * here, we need to free the just generated clone to unclone once
2132          * again.
2133          */
2134         ret = bpf_try_make_head_writable(skb);
2135         if (unlikely(ret)) {
2136                 kfree_skb(clone);
2137                 return -ENOMEM;
2138         }
2139
2140         return __bpf_redirect(clone, dev, flags);
2141 }
2142
2143 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2144         .func           = bpf_clone_redirect,
2145         .gpl_only       = false,
2146         .ret_type       = RET_INTEGER,
2147         .arg1_type      = ARG_PTR_TO_CTX,
2148         .arg2_type      = ARG_ANYTHING,
2149         .arg3_type      = ARG_ANYTHING,
2150 };
2151
2152 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2153 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2154
2155 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2156 {
2157         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2158
2159         if (unlikely(flags & ~(BPF_F_INGRESS)))
2160                 return TC_ACT_SHOT;
2161
2162         ri->flags = flags;
2163         ri->tgt_index = ifindex;
2164
2165         return TC_ACT_REDIRECT;
2166 }
2167
2168 int skb_do_redirect(struct sk_buff *skb)
2169 {
2170         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2171         struct net_device *dev;
2172
2173         dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
2174         ri->tgt_index = 0;
2175         if (unlikely(!dev)) {
2176                 kfree_skb(skb);
2177                 return -EINVAL;
2178         }
2179
2180         return __bpf_redirect(skb, dev, ri->flags);
2181 }
2182
2183 static const struct bpf_func_proto bpf_redirect_proto = {
2184         .func           = bpf_redirect,
2185         .gpl_only       = false,
2186         .ret_type       = RET_INTEGER,
2187         .arg1_type      = ARG_ANYTHING,
2188         .arg2_type      = ARG_ANYTHING,
2189 };
2190
2191 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2192 {
2193         msg->apply_bytes = bytes;
2194         return 0;
2195 }
2196
2197 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2198         .func           = bpf_msg_apply_bytes,
2199         .gpl_only       = false,
2200         .ret_type       = RET_INTEGER,
2201         .arg1_type      = ARG_PTR_TO_CTX,
2202         .arg2_type      = ARG_ANYTHING,
2203 };
2204
2205 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2206 {
2207         msg->cork_bytes = bytes;
2208         return 0;
2209 }
2210
2211 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2212         .func           = bpf_msg_cork_bytes,
2213         .gpl_only       = false,
2214         .ret_type       = RET_INTEGER,
2215         .arg1_type      = ARG_PTR_TO_CTX,
2216         .arg2_type      = ARG_ANYTHING,
2217 };
2218
2219 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2220            u32, end, u64, flags)
2221 {
2222         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2223         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2224         struct scatterlist *sge;
2225         u8 *raw, *to, *from;
2226         struct page *page;
2227
2228         if (unlikely(flags || end <= start))
2229                 return -EINVAL;
2230
2231         /* First find the starting scatterlist element */
2232         i = msg->sg.start;
2233         do {
2234                 offset += len;
2235                 len = sk_msg_elem(msg, i)->length;
2236                 if (start < offset + len)
2237                         break;
2238                 sk_msg_iter_var_next(i);
2239         } while (i != msg->sg.end);
2240
2241         if (unlikely(start >= offset + len))
2242                 return -EINVAL;
2243
2244         first_sge = i;
2245         /* The start may point into the sg element so we need to also
2246          * account for the headroom.
2247          */
2248         bytes_sg_total = start - offset + bytes;
2249         if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2250                 goto out;
2251
2252         /* At this point we need to linearize multiple scatterlist
2253          * elements or a single shared page. Either way we need to
2254          * copy into a linear buffer exclusively owned by BPF. Then
2255          * place the buffer in the scatterlist and fixup the original
2256          * entries by removing the entries now in the linear buffer
2257          * and shifting the remaining entries. For now we do not try
2258          * to copy partial entries to avoid complexity of running out
2259          * of sg_entry slots. The downside is reading a single byte
2260          * will copy the entire sg entry.
2261          */
2262         do {
2263                 copy += sk_msg_elem(msg, i)->length;
2264                 sk_msg_iter_var_next(i);
2265                 if (bytes_sg_total <= copy)
2266                         break;
2267         } while (i != msg->sg.end);
2268         last_sge = i;
2269
2270         if (unlikely(bytes_sg_total > copy))
2271                 return -EINVAL;
2272
2273         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2274                            get_order(copy));
2275         if (unlikely(!page))
2276                 return -ENOMEM;
2277
2278         raw = page_address(page);
2279         i = first_sge;
2280         do {
2281                 sge = sk_msg_elem(msg, i);
2282                 from = sg_virt(sge);
2283                 len = sge->length;
2284                 to = raw + poffset;
2285
2286                 memcpy(to, from, len);
2287                 poffset += len;
2288                 sge->length = 0;
2289                 put_page(sg_page(sge));
2290
2291                 sk_msg_iter_var_next(i);
2292         } while (i != last_sge);
2293
2294         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2295
2296         /* To repair sg ring we need to shift entries. If we only
2297          * had a single entry though we can just replace it and
2298          * be done. Otherwise walk the ring and shift the entries.
2299          */
2300         WARN_ON_ONCE(last_sge == first_sge);
2301         shift = last_sge > first_sge ?
2302                 last_sge - first_sge - 1 :
2303                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2304         if (!shift)
2305                 goto out;
2306
2307         i = first_sge;
2308         sk_msg_iter_var_next(i);
2309         do {
2310                 u32 move_from;
2311
2312                 if (i + shift >= NR_MSG_FRAG_IDS)
2313                         move_from = i + shift - NR_MSG_FRAG_IDS;
2314                 else
2315                         move_from = i + shift;
2316                 if (move_from == msg->sg.end)
2317                         break;
2318
2319                 msg->sg.data[i] = msg->sg.data[move_from];
2320                 msg->sg.data[move_from].length = 0;
2321                 msg->sg.data[move_from].page_link = 0;
2322                 msg->sg.data[move_from].offset = 0;
2323                 sk_msg_iter_var_next(i);
2324         } while (1);
2325
2326         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2327                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2328                       msg->sg.end - shift;
2329 out:
2330         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2331         msg->data_end = msg->data + bytes;
2332         return 0;
2333 }
2334
2335 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2336         .func           = bpf_msg_pull_data,
2337         .gpl_only       = false,
2338         .ret_type       = RET_INTEGER,
2339         .arg1_type      = ARG_PTR_TO_CTX,
2340         .arg2_type      = ARG_ANYTHING,
2341         .arg3_type      = ARG_ANYTHING,
2342         .arg4_type      = ARG_ANYTHING,
2343 };
2344
2345 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2346            u32, len, u64, flags)
2347 {
2348         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2349         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2350         u8 *raw, *to, *from;
2351         struct page *page;
2352
2353         if (unlikely(flags))
2354                 return -EINVAL;
2355
2356         /* First find the starting scatterlist element */
2357         i = msg->sg.start;
2358         do {
2359                 offset += l;
2360                 l = sk_msg_elem(msg, i)->length;
2361
2362                 if (start < offset + l)
2363                         break;
2364                 sk_msg_iter_var_next(i);
2365         } while (i != msg->sg.end);
2366
2367         if (start >= offset + l)
2368                 return -EINVAL;
2369
2370         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2371
2372         /* If no space available will fallback to copy, we need at
2373          * least one scatterlist elem available to push data into
2374          * when start aligns to the beginning of an element or two
2375          * when it falls inside an element. We handle the start equals
2376          * offset case because its the common case for inserting a
2377          * header.
2378          */
2379         if (!space || (space == 1 && start != offset))
2380                 copy = msg->sg.data[i].length;
2381
2382         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2383                            get_order(copy + len));
2384         if (unlikely(!page))
2385                 return -ENOMEM;
2386
2387         if (copy) {
2388                 int front, back;
2389
2390                 raw = page_address(page);
2391
2392                 psge = sk_msg_elem(msg, i);
2393                 front = start - offset;
2394                 back = psge->length - front;
2395                 from = sg_virt(psge);
2396
2397                 if (front)
2398                         memcpy(raw, from, front);
2399
2400                 if (back) {
2401                         from += front;
2402                         to = raw + front + len;
2403
2404                         memcpy(to, from, back);
2405                 }
2406
2407                 put_page(sg_page(psge));
2408         } else if (start - offset) {
2409                 psge = sk_msg_elem(msg, i);
2410                 rsge = sk_msg_elem_cpy(msg, i);
2411
2412                 psge->length = start - offset;
2413                 rsge.length -= psge->length;
2414                 rsge.offset += start;
2415
2416                 sk_msg_iter_var_next(i);
2417                 sg_unmark_end(psge);
2418                 sg_unmark_end(&rsge);
2419                 sk_msg_iter_next(msg, end);
2420         }
2421
2422         /* Slot(s) to place newly allocated data */
2423         new = i;
2424
2425         /* Shift one or two slots as needed */
2426         if (!copy) {
2427                 sge = sk_msg_elem_cpy(msg, i);
2428
2429                 sk_msg_iter_var_next(i);
2430                 sg_unmark_end(&sge);
2431                 sk_msg_iter_next(msg, end);
2432
2433                 nsge = sk_msg_elem_cpy(msg, i);
2434                 if (rsge.length) {
2435                         sk_msg_iter_var_next(i);
2436                         nnsge = sk_msg_elem_cpy(msg, i);
2437                 }
2438
2439                 while (i != msg->sg.end) {
2440                         msg->sg.data[i] = sge;
2441                         sge = nsge;
2442                         sk_msg_iter_var_next(i);
2443                         if (rsge.length) {
2444                                 nsge = nnsge;
2445                                 nnsge = sk_msg_elem_cpy(msg, i);
2446                         } else {
2447                                 nsge = sk_msg_elem_cpy(msg, i);
2448                         }
2449                 }
2450         }
2451
2452         /* Place newly allocated data buffer */
2453         sk_mem_charge(msg->sk, len);
2454         msg->sg.size += len;
2455         __clear_bit(new, &msg->sg.copy);
2456         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2457         if (rsge.length) {
2458                 get_page(sg_page(&rsge));
2459                 sk_msg_iter_var_next(new);
2460                 msg->sg.data[new] = rsge;
2461         }
2462
2463         sk_msg_compute_data_pointers(msg);
2464         return 0;
2465 }
2466
2467 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2468         .func           = bpf_msg_push_data,
2469         .gpl_only       = false,
2470         .ret_type       = RET_INTEGER,
2471         .arg1_type      = ARG_PTR_TO_CTX,
2472         .arg2_type      = ARG_ANYTHING,
2473         .arg3_type      = ARG_ANYTHING,
2474         .arg4_type      = ARG_ANYTHING,
2475 };
2476
2477 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2478 {
2479         int prev;
2480
2481         do {
2482                 prev = i;
2483                 sk_msg_iter_var_next(i);
2484                 msg->sg.data[prev] = msg->sg.data[i];
2485         } while (i != msg->sg.end);
2486
2487         sk_msg_iter_prev(msg, end);
2488 }
2489
2490 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2491 {
2492         struct scatterlist tmp, sge;
2493
2494         sk_msg_iter_next(msg, end);
2495         sge = sk_msg_elem_cpy(msg, i);
2496         sk_msg_iter_var_next(i);
2497         tmp = sk_msg_elem_cpy(msg, i);
2498
2499         while (i != msg->sg.end) {
2500                 msg->sg.data[i] = sge;
2501                 sk_msg_iter_var_next(i);
2502                 sge = tmp;
2503                 tmp = sk_msg_elem_cpy(msg, i);
2504         }
2505 }
2506
2507 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2508            u32, len, u64, flags)
2509 {
2510         u32 i = 0, l = 0, space, offset = 0;
2511         u64 last = start + len;
2512         int pop;
2513
2514         if (unlikely(flags))
2515                 return -EINVAL;
2516
2517         /* First find the starting scatterlist element */
2518         i = msg->sg.start;
2519         do {
2520                 offset += l;
2521                 l = sk_msg_elem(msg, i)->length;
2522
2523                 if (start < offset + l)
2524                         break;
2525                 sk_msg_iter_var_next(i);
2526         } while (i != msg->sg.end);
2527
2528         /* Bounds checks: start and pop must be inside message */
2529         if (start >= offset + l || last >= msg->sg.size)
2530                 return -EINVAL;
2531
2532         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2533
2534         pop = len;
2535         /* --------------| offset
2536          * -| start      |-------- len -------|
2537          *
2538          *  |----- a ----|-------- pop -------|----- b ----|
2539          *  |______________________________________________| length
2540          *
2541          *
2542          * a:   region at front of scatter element to save
2543          * b:   region at back of scatter element to save when length > A + pop
2544          * pop: region to pop from element, same as input 'pop' here will be
2545          *      decremented below per iteration.
2546          *
2547          * Two top-level cases to handle when start != offset, first B is non
2548          * zero and second B is zero corresponding to when a pop includes more
2549          * than one element.
2550          *
2551          * Then if B is non-zero AND there is no space allocate space and
2552          * compact A, B regions into page. If there is space shift ring to
2553          * the rigth free'ing the next element in ring to place B, leaving
2554          * A untouched except to reduce length.
2555          */
2556         if (start != offset) {
2557                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2558                 int a = start;
2559                 int b = sge->length - pop - a;
2560
2561                 sk_msg_iter_var_next(i);
2562
2563                 if (pop < sge->length - a) {
2564                         if (space) {
2565                                 sge->length = a;
2566                                 sk_msg_shift_right(msg, i);
2567                                 nsge = sk_msg_elem(msg, i);
2568                                 get_page(sg_page(sge));
2569                                 sg_set_page(nsge,
2570                                             sg_page(sge),
2571                                             b, sge->offset + pop + a);
2572                         } else {
2573                                 struct page *page, *orig;
2574                                 u8 *to, *from;
2575
2576                                 page = alloc_pages(__GFP_NOWARN |
2577                                                    __GFP_COMP   | GFP_ATOMIC,
2578                                                    get_order(a + b));
2579                                 if (unlikely(!page))
2580                                         return -ENOMEM;
2581
2582                                 sge->length = a;
2583                                 orig = sg_page(sge);
2584                                 from = sg_virt(sge);
2585                                 to = page_address(page);
2586                                 memcpy(to, from, a);
2587                                 memcpy(to + a, from + a + pop, b);
2588                                 sg_set_page(sge, page, a + b, 0);
2589                                 put_page(orig);
2590                         }
2591                         pop = 0;
2592                 } else if (pop >= sge->length - a) {
2593                         sge->length = a;
2594                         pop -= (sge->length - a);
2595                 }
2596         }
2597
2598         /* From above the current layout _must_ be as follows,
2599          *
2600          * -| offset
2601          * -| start
2602          *
2603          *  |---- pop ---|---------------- b ------------|
2604          *  |____________________________________________| length
2605          *
2606          * Offset and start of the current msg elem are equal because in the
2607          * previous case we handled offset != start and either consumed the
2608          * entire element and advanced to the next element OR pop == 0.
2609          *
2610          * Two cases to handle here are first pop is less than the length
2611          * leaving some remainder b above. Simply adjust the element's layout
2612          * in this case. Or pop >= length of the element so that b = 0. In this
2613          * case advance to next element decrementing pop.
2614          */
2615         while (pop) {
2616                 struct scatterlist *sge = sk_msg_elem(msg, i);
2617
2618                 if (pop < sge->length) {
2619                         sge->length -= pop;
2620                         sge->offset += pop;
2621                         pop = 0;
2622                 } else {
2623                         pop -= sge->length;
2624                         sk_msg_shift_left(msg, i);
2625                 }
2626                 sk_msg_iter_var_next(i);
2627         }
2628
2629         sk_mem_uncharge(msg->sk, len - pop);
2630         msg->sg.size -= (len - pop);
2631         sk_msg_compute_data_pointers(msg);
2632         return 0;
2633 }
2634
2635 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2636         .func           = bpf_msg_pop_data,
2637         .gpl_only       = false,
2638         .ret_type       = RET_INTEGER,
2639         .arg1_type      = ARG_PTR_TO_CTX,
2640         .arg2_type      = ARG_ANYTHING,
2641         .arg3_type      = ARG_ANYTHING,
2642         .arg4_type      = ARG_ANYTHING,
2643 };
2644
2645 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2646 {
2647         return task_get_classid(skb);
2648 }
2649
2650 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2651         .func           = bpf_get_cgroup_classid,
2652         .gpl_only       = false,
2653         .ret_type       = RET_INTEGER,
2654         .arg1_type      = ARG_PTR_TO_CTX,
2655 };
2656
2657 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2658 {
2659         return dst_tclassid(skb);
2660 }
2661
2662 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2663         .func           = bpf_get_route_realm,
2664         .gpl_only       = false,
2665         .ret_type       = RET_INTEGER,
2666         .arg1_type      = ARG_PTR_TO_CTX,
2667 };
2668
2669 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2670 {
2671         /* If skb_clear_hash() was called due to mangling, we can
2672          * trigger SW recalculation here. Later access to hash
2673          * can then use the inline skb->hash via context directly
2674          * instead of calling this helper again.
2675          */
2676         return skb_get_hash(skb);
2677 }
2678
2679 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2680         .func           = bpf_get_hash_recalc,
2681         .gpl_only       = false,
2682         .ret_type       = RET_INTEGER,
2683         .arg1_type      = ARG_PTR_TO_CTX,
2684 };
2685
2686 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2687 {
2688         /* After all direct packet write, this can be used once for
2689          * triggering a lazy recalc on next skb_get_hash() invocation.
2690          */
2691         skb_clear_hash(skb);
2692         return 0;
2693 }
2694
2695 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2696         .func           = bpf_set_hash_invalid,
2697         .gpl_only       = false,
2698         .ret_type       = RET_INTEGER,
2699         .arg1_type      = ARG_PTR_TO_CTX,
2700 };
2701
2702 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2703 {
2704         /* Set user specified hash as L4(+), so that it gets returned
2705          * on skb_get_hash() call unless BPF prog later on triggers a
2706          * skb_clear_hash().
2707          */
2708         __skb_set_sw_hash(skb, hash, true);
2709         return 0;
2710 }
2711
2712 static const struct bpf_func_proto bpf_set_hash_proto = {
2713         .func           = bpf_set_hash,
2714         .gpl_only       = false,
2715         .ret_type       = RET_INTEGER,
2716         .arg1_type      = ARG_PTR_TO_CTX,
2717         .arg2_type      = ARG_ANYTHING,
2718 };
2719
2720 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2721            u16, vlan_tci)
2722 {
2723         int ret;
2724
2725         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2726                      vlan_proto != htons(ETH_P_8021AD)))
2727                 vlan_proto = htons(ETH_P_8021Q);
2728
2729         bpf_push_mac_rcsum(skb);
2730         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2731         bpf_pull_mac_rcsum(skb);
2732
2733         bpf_compute_data_pointers(skb);
2734         return ret;
2735 }
2736
2737 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2738         .func           = bpf_skb_vlan_push,
2739         .gpl_only       = false,
2740         .ret_type       = RET_INTEGER,
2741         .arg1_type      = ARG_PTR_TO_CTX,
2742         .arg2_type      = ARG_ANYTHING,
2743         .arg3_type      = ARG_ANYTHING,
2744 };
2745
2746 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2747 {
2748         int ret;
2749
2750         bpf_push_mac_rcsum(skb);
2751         ret = skb_vlan_pop(skb);
2752         bpf_pull_mac_rcsum(skb);
2753
2754         bpf_compute_data_pointers(skb);
2755         return ret;
2756 }
2757
2758 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2759         .func           = bpf_skb_vlan_pop,
2760         .gpl_only       = false,
2761         .ret_type       = RET_INTEGER,
2762         .arg1_type      = ARG_PTR_TO_CTX,
2763 };
2764
2765 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2766 {
2767         /* Caller already did skb_cow() with len as headroom,
2768          * so no need to do it here.
2769          */
2770         skb_push(skb, len);
2771         memmove(skb->data, skb->data + len, off);
2772         memset(skb->data + off, 0, len);
2773
2774         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2775          * needed here as it does not change the skb->csum
2776          * result for checksum complete when summing over
2777          * zeroed blocks.
2778          */
2779         return 0;
2780 }
2781
2782 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2783 {
2784         /* skb_ensure_writable() is not needed here, as we're
2785          * already working on an uncloned skb.
2786          */
2787         if (unlikely(!pskb_may_pull(skb, off + len)))
2788                 return -ENOMEM;
2789
2790         skb_postpull_rcsum(skb, skb->data + off, len);
2791         memmove(skb->data + len, skb->data, off);
2792         __skb_pull(skb, len);
2793
2794         return 0;
2795 }
2796
2797 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2798 {
2799         bool trans_same = skb->transport_header == skb->network_header;
2800         int ret;
2801
2802         /* There's no need for __skb_push()/__skb_pull() pair to
2803          * get to the start of the mac header as we're guaranteed
2804          * to always start from here under eBPF.
2805          */
2806         ret = bpf_skb_generic_push(skb, off, len);
2807         if (likely(!ret)) {
2808                 skb->mac_header -= len;
2809                 skb->network_header -= len;
2810                 if (trans_same)
2811                         skb->transport_header = skb->network_header;
2812         }
2813
2814         return ret;
2815 }
2816
2817 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2818 {
2819         bool trans_same = skb->transport_header == skb->network_header;
2820         int ret;
2821
2822         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2823         ret = bpf_skb_generic_pop(skb, off, len);
2824         if (likely(!ret)) {
2825                 skb->mac_header += len;
2826                 skb->network_header += len;
2827                 if (trans_same)
2828                         skb->transport_header = skb->network_header;
2829         }
2830
2831         return ret;
2832 }
2833
2834 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2835 {
2836         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2837         u32 off = skb_mac_header_len(skb);
2838         int ret;
2839
2840         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2841                 return -ENOTSUPP;
2842
2843         ret = skb_cow(skb, len_diff);
2844         if (unlikely(ret < 0))
2845                 return ret;
2846
2847         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2848         if (unlikely(ret < 0))
2849                 return ret;
2850
2851         if (skb_is_gso(skb)) {
2852                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2853
2854                 /* SKB_GSO_TCPV4 needs to be changed into
2855                  * SKB_GSO_TCPV6.
2856                  */
2857                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2858                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2859                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2860                 }
2861
2862                 /* Due to IPv6 header, MSS needs to be downgraded. */
2863                 skb_decrease_gso_size(shinfo, len_diff);
2864                 /* Header must be checked, and gso_segs recomputed. */
2865                 shinfo->gso_type |= SKB_GSO_DODGY;
2866                 shinfo->gso_segs = 0;
2867         }
2868
2869         skb->protocol = htons(ETH_P_IPV6);
2870         skb_clear_hash(skb);
2871
2872         return 0;
2873 }
2874
2875 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2876 {
2877         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2878         u32 off = skb_mac_header_len(skb);
2879         int ret;
2880
2881         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2882                 return -ENOTSUPP;
2883
2884         ret = skb_unclone(skb, GFP_ATOMIC);
2885         if (unlikely(ret < 0))
2886                 return ret;
2887
2888         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2889         if (unlikely(ret < 0))
2890                 return ret;
2891
2892         if (skb_is_gso(skb)) {
2893                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2894
2895                 /* SKB_GSO_TCPV6 needs to be changed into
2896                  * SKB_GSO_TCPV4.
2897                  */
2898                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2899                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2900                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2901                 }
2902
2903                 /* Due to IPv4 header, MSS can be upgraded. */
2904                 skb_increase_gso_size(shinfo, len_diff);
2905                 /* Header must be checked, and gso_segs recomputed. */
2906                 shinfo->gso_type |= SKB_GSO_DODGY;
2907                 shinfo->gso_segs = 0;
2908         }
2909
2910         skb->protocol = htons(ETH_P_IP);
2911         skb_clear_hash(skb);
2912
2913         return 0;
2914 }
2915
2916 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2917 {
2918         __be16 from_proto = skb->protocol;
2919
2920         if (from_proto == htons(ETH_P_IP) &&
2921               to_proto == htons(ETH_P_IPV6))
2922                 return bpf_skb_proto_4_to_6(skb);
2923
2924         if (from_proto == htons(ETH_P_IPV6) &&
2925               to_proto == htons(ETH_P_IP))
2926                 return bpf_skb_proto_6_to_4(skb);
2927
2928         return -ENOTSUPP;
2929 }
2930
2931 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2932            u64, flags)
2933 {
2934         int ret;
2935
2936         if (unlikely(flags))
2937                 return -EINVAL;
2938
2939         /* General idea is that this helper does the basic groundwork
2940          * needed for changing the protocol, and eBPF program fills the
2941          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2942          * and other helpers, rather than passing a raw buffer here.
2943          *
2944          * The rationale is to keep this minimal and without a need to
2945          * deal with raw packet data. F.e. even if we would pass buffers
2946          * here, the program still needs to call the bpf_lX_csum_replace()
2947          * helpers anyway. Plus, this way we keep also separation of
2948          * concerns, since f.e. bpf_skb_store_bytes() should only take
2949          * care of stores.
2950          *
2951          * Currently, additional options and extension header space are
2952          * not supported, but flags register is reserved so we can adapt
2953          * that. For offloads, we mark packet as dodgy, so that headers
2954          * need to be verified first.
2955          */
2956         ret = bpf_skb_proto_xlat(skb, proto);
2957         bpf_compute_data_pointers(skb);
2958         return ret;
2959 }
2960
2961 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2962         .func           = bpf_skb_change_proto,
2963         .gpl_only       = false,
2964         .ret_type       = RET_INTEGER,
2965         .arg1_type      = ARG_PTR_TO_CTX,
2966         .arg2_type      = ARG_ANYTHING,
2967         .arg3_type      = ARG_ANYTHING,
2968 };
2969
2970 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2971 {
2972         /* We only allow a restricted subset to be changed for now. */
2973         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2974                      !skb_pkt_type_ok(pkt_type)))
2975                 return -EINVAL;
2976
2977         skb->pkt_type = pkt_type;
2978         return 0;
2979 }
2980
2981 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2982         .func           = bpf_skb_change_type,
2983         .gpl_only       = false,
2984         .ret_type       = RET_INTEGER,
2985         .arg1_type      = ARG_PTR_TO_CTX,
2986         .arg2_type      = ARG_ANYTHING,
2987 };
2988
2989 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2990 {
2991         switch (skb->protocol) {
2992         case htons(ETH_P_IP):
2993                 return sizeof(struct iphdr);
2994         case htons(ETH_P_IPV6):
2995                 return sizeof(struct ipv6hdr);
2996         default:
2997                 return ~0U;
2998         }
2999 }
3000
3001 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3002                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3003
3004 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3005                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3006                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3007                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3008                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3009                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3010
3011 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3012                             u64 flags)
3013 {
3014         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3015         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3016         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3017         unsigned int gso_type = SKB_GSO_DODGY;
3018         int ret;
3019
3020         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3021                 /* udp gso_size delineates datagrams, only allow if fixed */
3022                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3023                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3024                         return -ENOTSUPP;
3025         }
3026
3027         ret = skb_cow_head(skb, len_diff);
3028         if (unlikely(ret < 0))
3029                 return ret;
3030
3031         if (encap) {
3032                 if (skb->protocol != htons(ETH_P_IP) &&
3033                     skb->protocol != htons(ETH_P_IPV6))
3034                         return -ENOTSUPP;
3035
3036                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3037                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3038                         return -EINVAL;
3039
3040                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3041                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3042                         return -EINVAL;
3043
3044                 if (skb->encapsulation)
3045                         return -EALREADY;
3046
3047                 mac_len = skb->network_header - skb->mac_header;
3048                 inner_net = skb->network_header;
3049                 if (inner_mac_len > len_diff)
3050                         return -EINVAL;
3051                 inner_trans = skb->transport_header;
3052         }
3053
3054         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3055         if (unlikely(ret < 0))
3056                 return ret;
3057
3058         if (encap) {
3059                 skb->inner_mac_header = inner_net - inner_mac_len;
3060                 skb->inner_network_header = inner_net;
3061                 skb->inner_transport_header = inner_trans;
3062                 skb_set_inner_protocol(skb, skb->protocol);
3063
3064                 skb->encapsulation = 1;
3065                 skb_set_network_header(skb, mac_len);
3066
3067                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3068                         gso_type |= SKB_GSO_UDP_TUNNEL;
3069                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3070                         gso_type |= SKB_GSO_GRE;
3071                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3072                         gso_type |= SKB_GSO_IPXIP6;
3073                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3074                         gso_type |= SKB_GSO_IPXIP4;
3075
3076                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3077                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3078                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3079                                         sizeof(struct ipv6hdr) :
3080                                         sizeof(struct iphdr);
3081
3082                         skb_set_transport_header(skb, mac_len + nh_len);
3083                 }
3084
3085                 /* Match skb->protocol to new outer l3 protocol */
3086                 if (skb->protocol == htons(ETH_P_IP) &&
3087                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3088                         skb->protocol = htons(ETH_P_IPV6);
3089                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3090                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3091                         skb->protocol = htons(ETH_P_IP);
3092         }
3093
3094         if (skb_is_gso(skb)) {
3095                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3096
3097                 /* Due to header grow, MSS needs to be downgraded. */
3098                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3099                         skb_decrease_gso_size(shinfo, len_diff);
3100
3101                 /* Header must be checked, and gso_segs recomputed. */
3102                 shinfo->gso_type |= gso_type;
3103                 shinfo->gso_segs = 0;
3104         }
3105
3106         return 0;
3107 }
3108
3109 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3110                               u64 flags)
3111 {
3112         int ret;
3113
3114         if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO)
3115                 return -EINVAL;
3116
3117         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3118                 /* udp gso_size delineates datagrams, only allow if fixed */
3119                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3120                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3121                         return -ENOTSUPP;
3122         }
3123
3124         ret = skb_unclone(skb, GFP_ATOMIC);
3125         if (unlikely(ret < 0))
3126                 return ret;
3127
3128         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3129         if (unlikely(ret < 0))
3130                 return ret;
3131
3132         if (skb_is_gso(skb)) {
3133                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3134
3135                 /* Due to header shrink, MSS can be upgraded. */
3136                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3137                         skb_increase_gso_size(shinfo, len_diff);
3138
3139                 /* Header must be checked, and gso_segs recomputed. */
3140                 shinfo->gso_type |= SKB_GSO_DODGY;
3141                 shinfo->gso_segs = 0;
3142         }
3143
3144         return 0;
3145 }
3146
3147 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3148 {
3149         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3150                           SKB_MAX_ALLOC;
3151 }
3152
3153 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3154            u32, mode, u64, flags)
3155 {
3156         u32 len_cur, len_diff_abs = abs(len_diff);
3157         u32 len_min = bpf_skb_net_base_len(skb);
3158         u32 len_max = __bpf_skb_max_len(skb);
3159         __be16 proto = skb->protocol;
3160         bool shrink = len_diff < 0;
3161         u32 off;
3162         int ret;
3163
3164         if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK))
3165                 return -EINVAL;
3166         if (unlikely(len_diff_abs > 0xfffU))
3167                 return -EFAULT;
3168         if (unlikely(proto != htons(ETH_P_IP) &&
3169                      proto != htons(ETH_P_IPV6)))
3170                 return -ENOTSUPP;
3171
3172         off = skb_mac_header_len(skb);
3173         switch (mode) {
3174         case BPF_ADJ_ROOM_NET:
3175                 off += bpf_skb_net_base_len(skb);
3176                 break;
3177         case BPF_ADJ_ROOM_MAC:
3178                 break;
3179         default:
3180                 return -ENOTSUPP;
3181         }
3182
3183         len_cur = skb->len - skb_network_offset(skb);
3184         if ((shrink && (len_diff_abs >= len_cur ||
3185                         len_cur - len_diff_abs < len_min)) ||
3186             (!shrink && (skb->len + len_diff_abs > len_max &&
3187                          !skb_is_gso(skb))))
3188                 return -ENOTSUPP;
3189
3190         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3191                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3192
3193         bpf_compute_data_pointers(skb);
3194         return ret;
3195 }
3196
3197 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3198         .func           = bpf_skb_adjust_room,
3199         .gpl_only       = false,
3200         .ret_type       = RET_INTEGER,
3201         .arg1_type      = ARG_PTR_TO_CTX,
3202         .arg2_type      = ARG_ANYTHING,
3203         .arg3_type      = ARG_ANYTHING,
3204         .arg4_type      = ARG_ANYTHING,
3205 };
3206
3207 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3208 {
3209         u32 min_len = skb_network_offset(skb);
3210
3211         if (skb_transport_header_was_set(skb))
3212                 min_len = skb_transport_offset(skb);
3213         if (skb->ip_summed == CHECKSUM_PARTIAL)
3214                 min_len = skb_checksum_start_offset(skb) +
3215                           skb->csum_offset + sizeof(__sum16);
3216         return min_len;
3217 }
3218
3219 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3220 {
3221         unsigned int old_len = skb->len;
3222         int ret;
3223
3224         ret = __skb_grow_rcsum(skb, new_len);
3225         if (!ret)
3226                 memset(skb->data + old_len, 0, new_len - old_len);
3227         return ret;
3228 }
3229
3230 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3231 {
3232         return __skb_trim_rcsum(skb, new_len);
3233 }
3234
3235 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3236                                         u64 flags)
3237 {
3238         u32 max_len = __bpf_skb_max_len(skb);
3239         u32 min_len = __bpf_skb_min_len(skb);
3240         int ret;
3241
3242         if (unlikely(flags || new_len > max_len || new_len < min_len))
3243                 return -EINVAL;
3244         if (skb->encapsulation)
3245                 return -ENOTSUPP;
3246
3247         /* The basic idea of this helper is that it's performing the
3248          * needed work to either grow or trim an skb, and eBPF program
3249          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3250          * bpf_lX_csum_replace() and others rather than passing a raw
3251          * buffer here. This one is a slow path helper and intended
3252          * for replies with control messages.
3253          *
3254          * Like in bpf_skb_change_proto(), we want to keep this rather
3255          * minimal and without protocol specifics so that we are able
3256          * to separate concerns as in bpf_skb_store_bytes() should only
3257          * be the one responsible for writing buffers.
3258          *
3259          * It's really expected to be a slow path operation here for
3260          * control message replies, so we're implicitly linearizing,
3261          * uncloning and drop offloads from the skb by this.
3262          */
3263         ret = __bpf_try_make_writable(skb, skb->len);
3264         if (!ret) {
3265                 if (new_len > skb->len)
3266                         ret = bpf_skb_grow_rcsum(skb, new_len);
3267                 else if (new_len < skb->len)
3268                         ret = bpf_skb_trim_rcsum(skb, new_len);
3269                 if (!ret && skb_is_gso(skb))
3270                         skb_gso_reset(skb);
3271         }
3272         return ret;
3273 }
3274
3275 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3276            u64, flags)
3277 {
3278         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3279
3280         bpf_compute_data_pointers(skb);
3281         return ret;
3282 }
3283
3284 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3285         .func           = bpf_skb_change_tail,
3286         .gpl_only       = false,
3287         .ret_type       = RET_INTEGER,
3288         .arg1_type      = ARG_PTR_TO_CTX,
3289         .arg2_type      = ARG_ANYTHING,
3290         .arg3_type      = ARG_ANYTHING,
3291 };
3292
3293 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3294            u64, flags)
3295 {
3296         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3297
3298         bpf_compute_data_end_sk_skb(skb);
3299         return ret;
3300 }
3301
3302 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3303         .func           = sk_skb_change_tail,
3304         .gpl_only       = false,
3305         .ret_type       = RET_INTEGER,
3306         .arg1_type      = ARG_PTR_TO_CTX,
3307         .arg2_type      = ARG_ANYTHING,
3308         .arg3_type      = ARG_ANYTHING,
3309 };
3310
3311 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3312                                         u64 flags)
3313 {
3314         u32 max_len = __bpf_skb_max_len(skb);
3315         u32 new_len = skb->len + head_room;
3316         int ret;
3317
3318         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3319                      new_len < skb->len))
3320                 return -EINVAL;
3321
3322         ret = skb_cow(skb, head_room);
3323         if (likely(!ret)) {
3324                 /* Idea for this helper is that we currently only
3325                  * allow to expand on mac header. This means that
3326                  * skb->protocol network header, etc, stay as is.
3327                  * Compared to bpf_skb_change_tail(), we're more
3328                  * flexible due to not needing to linearize or
3329                  * reset GSO. Intention for this helper is to be
3330                  * used by an L3 skb that needs to push mac header
3331                  * for redirection into L2 device.
3332                  */
3333                 __skb_push(skb, head_room);
3334                 memset(skb->data, 0, head_room);
3335                 skb_reset_mac_header(skb);
3336         }
3337
3338         return ret;
3339 }
3340
3341 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3342            u64, flags)
3343 {
3344         int ret = __bpf_skb_change_head(skb, head_room, flags);
3345
3346         bpf_compute_data_pointers(skb);
3347         return ret;
3348 }
3349
3350 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3351         .func           = bpf_skb_change_head,
3352         .gpl_only       = false,
3353         .ret_type       = RET_INTEGER,
3354         .arg1_type      = ARG_PTR_TO_CTX,
3355         .arg2_type      = ARG_ANYTHING,
3356         .arg3_type      = ARG_ANYTHING,
3357 };
3358
3359 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3360            u64, flags)
3361 {
3362         int ret = __bpf_skb_change_head(skb, head_room, flags);
3363
3364         bpf_compute_data_end_sk_skb(skb);
3365         return ret;
3366 }
3367
3368 static const struct bpf_func_proto sk_skb_change_head_proto = {
3369         .func           = sk_skb_change_head,
3370         .gpl_only       = false,
3371         .ret_type       = RET_INTEGER,
3372         .arg1_type      = ARG_PTR_TO_CTX,
3373         .arg2_type      = ARG_ANYTHING,
3374         .arg3_type      = ARG_ANYTHING,
3375 };
3376 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3377 {
3378         return xdp_data_meta_unsupported(xdp) ? 0 :
3379                xdp->data - xdp->data_meta;
3380 }
3381
3382 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3383 {
3384         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3385         unsigned long metalen = xdp_get_metalen(xdp);
3386         void *data_start = xdp_frame_end + metalen;
3387         void *data = xdp->data + offset;
3388
3389         if (unlikely(data < data_start ||
3390                      data > xdp->data_end - ETH_HLEN))
3391                 return -EINVAL;
3392
3393         if (metalen)
3394                 memmove(xdp->data_meta + offset,
3395                         xdp->data_meta, metalen);
3396         xdp->data_meta += offset;
3397         xdp->data = data;
3398
3399         return 0;
3400 }
3401
3402 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3403         .func           = bpf_xdp_adjust_head,
3404         .gpl_only       = false,
3405         .ret_type       = RET_INTEGER,
3406         .arg1_type      = ARG_PTR_TO_CTX,
3407         .arg2_type      = ARG_ANYTHING,
3408 };
3409
3410 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3411 {
3412         void *data_end = xdp->data_end + offset;
3413
3414         /* only shrinking is allowed for now. */
3415         if (unlikely(offset >= 0))
3416                 return -EINVAL;
3417
3418         if (unlikely(data_end < xdp->data + ETH_HLEN))
3419                 return -EINVAL;
3420
3421         xdp->data_end = data_end;
3422
3423         return 0;
3424 }
3425
3426 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3427         .func           = bpf_xdp_adjust_tail,
3428         .gpl_only       = false,
3429         .ret_type       = RET_INTEGER,
3430         .arg1_type      = ARG_PTR_TO_CTX,
3431         .arg2_type      = ARG_ANYTHING,
3432 };
3433
3434 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3435 {
3436         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3437         void *meta = xdp->data_meta + offset;
3438         unsigned long metalen = xdp->data - meta;
3439
3440         if (xdp_data_meta_unsupported(xdp))
3441                 return -ENOTSUPP;
3442         if (unlikely(meta < xdp_frame_end ||
3443                      meta > xdp->data))
3444                 return -EINVAL;
3445         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3446                      (metalen > 32)))
3447                 return -EACCES;
3448
3449         xdp->data_meta = meta;
3450
3451         return 0;
3452 }
3453
3454 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3455         .func           = bpf_xdp_adjust_meta,
3456         .gpl_only       = false,
3457         .ret_type       = RET_INTEGER,
3458         .arg1_type      = ARG_PTR_TO_CTX,
3459         .arg2_type      = ARG_ANYTHING,
3460 };
3461
3462 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3463                             struct bpf_map *map, struct xdp_buff *xdp)
3464 {
3465         switch (map->map_type) {
3466         case BPF_MAP_TYPE_DEVMAP:
3467         case BPF_MAP_TYPE_DEVMAP_HASH:
3468                 return dev_map_enqueue(fwd, xdp, dev_rx);
3469         case BPF_MAP_TYPE_CPUMAP:
3470                 return cpu_map_enqueue(fwd, xdp, dev_rx);
3471         case BPF_MAP_TYPE_XSKMAP:
3472                 return __xsk_map_redirect(fwd, xdp);
3473         default:
3474                 return -EBADRQC;
3475         }
3476         return 0;
3477 }
3478
3479 void xdp_do_flush(void)
3480 {
3481         __dev_flush();
3482         __cpu_map_flush();
3483         __xsk_map_flush();
3484 }
3485 EXPORT_SYMBOL_GPL(xdp_do_flush);
3486
3487 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3488 {
3489         switch (map->map_type) {
3490         case BPF_MAP_TYPE_DEVMAP:
3491                 return __dev_map_lookup_elem(map, index);
3492         case BPF_MAP_TYPE_DEVMAP_HASH:
3493                 return __dev_map_hash_lookup_elem(map, index);
3494         case BPF_MAP_TYPE_CPUMAP:
3495                 return __cpu_map_lookup_elem(map, index);
3496         case BPF_MAP_TYPE_XSKMAP:
3497                 return __xsk_map_lookup_elem(map, index);
3498         default:
3499                 return NULL;
3500         }
3501 }
3502
3503 void bpf_clear_redirect_map(struct bpf_map *map)
3504 {
3505         struct bpf_redirect_info *ri;
3506         int cpu;
3507
3508         for_each_possible_cpu(cpu) {
3509                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3510                 /* Avoid polluting remote cacheline due to writes if
3511                  * not needed. Once we pass this test, we need the
3512                  * cmpxchg() to make sure it hasn't been changed in
3513                  * the meantime by remote CPU.
3514                  */
3515                 if (unlikely(READ_ONCE(ri->map) == map))
3516                         cmpxchg(&ri->map, map, NULL);
3517         }
3518 }
3519
3520 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3521                     struct bpf_prog *xdp_prog)
3522 {
3523         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3524         struct bpf_map *map = READ_ONCE(ri->map);
3525         u32 index = ri->tgt_index;
3526         void *fwd = ri->tgt_value;
3527         int err;
3528
3529         ri->tgt_index = 0;
3530         ri->tgt_value = NULL;
3531         WRITE_ONCE(ri->map, NULL);
3532
3533         if (unlikely(!map)) {
3534                 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3535                 if (unlikely(!fwd)) {
3536                         err = -EINVAL;
3537                         goto err;
3538                 }
3539
3540                 err = dev_xdp_enqueue(fwd, xdp, dev);
3541         } else {
3542                 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
3543         }
3544
3545         if (unlikely(err))
3546                 goto err;
3547
3548         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3549         return 0;
3550 err:
3551         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3552         return err;
3553 }
3554 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3555
3556 static int xdp_do_generic_redirect_map(struct net_device *dev,
3557                                        struct sk_buff *skb,
3558                                        struct xdp_buff *xdp,
3559                                        struct bpf_prog *xdp_prog,
3560                                        struct bpf_map *map)
3561 {
3562         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3563         u32 index = ri->tgt_index;
3564         void *fwd = ri->tgt_value;
3565         int err = 0;
3566
3567         ri->tgt_index = 0;
3568         ri->tgt_value = NULL;
3569         WRITE_ONCE(ri->map, NULL);
3570
3571         if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
3572             map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
3573                 struct bpf_dtab_netdev *dst = fwd;
3574
3575                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3576                 if (unlikely(err))
3577                         goto err;
3578         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3579                 struct xdp_sock *xs = fwd;
3580
3581                 err = xsk_generic_rcv(xs, xdp);
3582                 if (err)
3583                         goto err;
3584                 consume_skb(skb);
3585         } else {
3586                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3587                 err = -EBADRQC;
3588                 goto err;
3589         }
3590
3591         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3592         return 0;
3593 err:
3594         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3595         return err;
3596 }
3597
3598 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3599                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3600 {
3601         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3602         struct bpf_map *map = READ_ONCE(ri->map);
3603         u32 index = ri->tgt_index;
3604         struct net_device *fwd;
3605         int err = 0;
3606
3607         if (map)
3608                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3609                                                    map);
3610         ri->tgt_index = 0;
3611         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3612         if (unlikely(!fwd)) {
3613                 err = -EINVAL;
3614                 goto err;
3615         }
3616
3617         err = xdp_ok_fwd_dev(fwd, skb->len);
3618         if (unlikely(err))
3619                 goto err;
3620
3621         skb->dev = fwd;
3622         _trace_xdp_redirect(dev, xdp_prog, index);
3623         generic_xdp_tx(skb, xdp_prog);
3624         return 0;
3625 err:
3626         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3627         return err;
3628 }
3629
3630 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3631 {
3632         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3633
3634         if (unlikely(flags))
3635                 return XDP_ABORTED;
3636
3637         ri->flags = flags;
3638         ri->tgt_index = ifindex;
3639         ri->tgt_value = NULL;
3640         WRITE_ONCE(ri->map, NULL);
3641
3642         return XDP_REDIRECT;
3643 }
3644
3645 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3646         .func           = bpf_xdp_redirect,
3647         .gpl_only       = false,
3648         .ret_type       = RET_INTEGER,
3649         .arg1_type      = ARG_ANYTHING,
3650         .arg2_type      = ARG_ANYTHING,
3651 };
3652
3653 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3654            u64, flags)
3655 {
3656         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3657
3658         /* Lower bits of the flags are used as return code on lookup failure */
3659         if (unlikely(flags > XDP_TX))
3660                 return XDP_ABORTED;
3661
3662         ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
3663         if (unlikely(!ri->tgt_value)) {
3664                 /* If the lookup fails we want to clear out the state in the
3665                  * redirect_info struct completely, so that if an eBPF program
3666                  * performs multiple lookups, the last one always takes
3667                  * precedence.
3668                  */
3669                 WRITE_ONCE(ri->map, NULL);
3670                 return flags;
3671         }
3672
3673         ri->flags = flags;
3674         ri->tgt_index = ifindex;
3675         WRITE_ONCE(ri->map, map);
3676
3677         return XDP_REDIRECT;
3678 }
3679
3680 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3681         .func           = bpf_xdp_redirect_map,
3682         .gpl_only       = false,
3683         .ret_type       = RET_INTEGER,
3684         .arg1_type      = ARG_CONST_MAP_PTR,
3685         .arg2_type      = ARG_ANYTHING,
3686         .arg3_type      = ARG_ANYTHING,
3687 };
3688
3689 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3690                                   unsigned long off, unsigned long len)
3691 {
3692         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3693
3694         if (unlikely(!ptr))
3695                 return len;
3696         if (ptr != dst_buff)
3697                 memcpy(dst_buff, ptr, len);
3698
3699         return 0;
3700 }
3701
3702 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3703            u64, flags, void *, meta, u64, meta_size)
3704 {
3705         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3706
3707         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3708                 return -EINVAL;
3709         if (unlikely(!skb || skb_size > skb->len))
3710                 return -EFAULT;
3711
3712         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3713                                 bpf_skb_copy);
3714 }
3715
3716 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3717         .func           = bpf_skb_event_output,
3718         .gpl_only       = true,
3719         .ret_type       = RET_INTEGER,
3720         .arg1_type      = ARG_PTR_TO_CTX,
3721         .arg2_type      = ARG_CONST_MAP_PTR,
3722         .arg3_type      = ARG_ANYTHING,
3723         .arg4_type      = ARG_PTR_TO_MEM,
3724         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3725 };
3726
3727 static int bpf_skb_output_btf_ids[5];
3728 const struct bpf_func_proto bpf_skb_output_proto = {
3729         .func           = bpf_skb_event_output,
3730         .gpl_only       = true,
3731         .ret_type       = RET_INTEGER,
3732         .arg1_type      = ARG_PTR_TO_BTF_ID,
3733         .arg2_type      = ARG_CONST_MAP_PTR,
3734         .arg3_type      = ARG_ANYTHING,
3735         .arg4_type      = ARG_PTR_TO_MEM,
3736         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3737         .btf_id         = bpf_skb_output_btf_ids,
3738 };
3739
3740 static unsigned short bpf_tunnel_key_af(u64 flags)
3741 {
3742         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3743 }
3744
3745 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3746            u32, size, u64, flags)
3747 {
3748         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3749         u8 compat[sizeof(struct bpf_tunnel_key)];
3750         void *to_orig = to;
3751         int err;
3752
3753         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3754                 err = -EINVAL;
3755                 goto err_clear;
3756         }
3757         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3758                 err = -EPROTO;
3759                 goto err_clear;
3760         }
3761         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3762                 err = -EINVAL;
3763                 switch (size) {
3764                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3765                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3766                         goto set_compat;
3767                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3768                         /* Fixup deprecated structure layouts here, so we have
3769                          * a common path later on.
3770                          */
3771                         if (ip_tunnel_info_af(info) != AF_INET)
3772                                 goto err_clear;
3773 set_compat:
3774                         to = (struct bpf_tunnel_key *)compat;
3775                         break;
3776                 default:
3777                         goto err_clear;
3778                 }
3779         }
3780
3781         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3782         to->tunnel_tos = info->key.tos;
3783         to->tunnel_ttl = info->key.ttl;
3784         to->tunnel_ext = 0;
3785
3786         if (flags & BPF_F_TUNINFO_IPV6) {
3787                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3788                        sizeof(to->remote_ipv6));
3789                 to->tunnel_label = be32_to_cpu(info->key.label);
3790         } else {
3791                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3792                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3793                 to->tunnel_label = 0;
3794         }
3795
3796         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3797                 memcpy(to_orig, to, size);
3798
3799         return 0;
3800 err_clear:
3801         memset(to_orig, 0, size);
3802         return err;
3803 }
3804
3805 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3806         .func           = bpf_skb_get_tunnel_key,
3807         .gpl_only       = false,
3808         .ret_type       = RET_INTEGER,
3809         .arg1_type      = ARG_PTR_TO_CTX,
3810         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3811         .arg3_type      = ARG_CONST_SIZE,
3812         .arg4_type      = ARG_ANYTHING,
3813 };
3814
3815 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3816 {
3817         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3818         int err;
3819
3820         if (unlikely(!info ||
3821                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3822                 err = -ENOENT;
3823                 goto err_clear;
3824         }
3825         if (unlikely(size < info->options_len)) {
3826                 err = -ENOMEM;
3827                 goto err_clear;
3828         }
3829
3830         ip_tunnel_info_opts_get(to, info);
3831         if (size > info->options_len)
3832                 memset(to + info->options_len, 0, size - info->options_len);
3833
3834         return info->options_len;
3835 err_clear:
3836         memset(to, 0, size);
3837         return err;
3838 }
3839
3840 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3841         .func           = bpf_skb_get_tunnel_opt,
3842         .gpl_only       = false,
3843         .ret_type       = RET_INTEGER,
3844         .arg1_type      = ARG_PTR_TO_CTX,
3845         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3846         .arg3_type      = ARG_CONST_SIZE,
3847 };
3848
3849 static struct metadata_dst __percpu *md_dst;
3850
3851 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3852            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3853 {
3854         struct metadata_dst *md = this_cpu_ptr(md_dst);
3855         u8 compat[sizeof(struct bpf_tunnel_key)];
3856         struct ip_tunnel_info *info;
3857
3858         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3859                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3860                 return -EINVAL;
3861         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3862                 switch (size) {
3863                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3864                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3865                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3866                         /* Fixup deprecated structure layouts here, so we have
3867                          * a common path later on.
3868                          */
3869                         memcpy(compat, from, size);
3870                         memset(compat + size, 0, sizeof(compat) - size);
3871                         from = (const struct bpf_tunnel_key *) compat;
3872                         break;
3873                 default:
3874                         return -EINVAL;
3875                 }
3876         }
3877         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3878                      from->tunnel_ext))
3879                 return -EINVAL;
3880
3881         skb_dst_drop(skb);
3882         dst_hold((struct dst_entry *) md);
3883         skb_dst_set(skb, (struct dst_entry *) md);
3884
3885         info = &md->u.tun_info;
3886         memset(info, 0, sizeof(*info));
3887         info->mode = IP_TUNNEL_INFO_TX;
3888
3889         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3890         if (flags & BPF_F_DONT_FRAGMENT)
3891                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3892         if (flags & BPF_F_ZERO_CSUM_TX)
3893                 info->key.tun_flags &= ~TUNNEL_CSUM;
3894         if (flags & BPF_F_SEQ_NUMBER)
3895                 info->key.tun_flags |= TUNNEL_SEQ;
3896
3897         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3898         info->key.tos = from->tunnel_tos;
3899         info->key.ttl = from->tunnel_ttl;
3900
3901         if (flags & BPF_F_TUNINFO_IPV6) {
3902                 info->mode |= IP_TUNNEL_INFO_IPV6;
3903                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3904                        sizeof(from->remote_ipv6));
3905                 info->key.label = cpu_to_be32(from->tunnel_label) &
3906                                   IPV6_FLOWLABEL_MASK;
3907         } else {
3908                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3909         }
3910
3911         return 0;
3912 }
3913
3914 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3915         .func           = bpf_skb_set_tunnel_key,
3916         .gpl_only       = false,
3917         .ret_type       = RET_INTEGER,
3918         .arg1_type      = ARG_PTR_TO_CTX,
3919         .arg2_type      = ARG_PTR_TO_MEM,
3920         .arg3_type      = ARG_CONST_SIZE,
3921         .arg4_type      = ARG_ANYTHING,
3922 };
3923
3924 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3925            const u8 *, from, u32, size)
3926 {
3927         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3928         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3929
3930         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3931                 return -EINVAL;
3932         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3933                 return -ENOMEM;
3934
3935         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3936
3937         return 0;
3938 }
3939
3940 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3941         .func           = bpf_skb_set_tunnel_opt,
3942         .gpl_only       = false,
3943         .ret_type       = RET_INTEGER,
3944         .arg1_type      = ARG_PTR_TO_CTX,
3945         .arg2_type      = ARG_PTR_TO_MEM,
3946         .arg3_type      = ARG_CONST_SIZE,
3947 };
3948
3949 static const struct bpf_func_proto *
3950 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3951 {
3952         if (!md_dst) {
3953                 struct metadata_dst __percpu *tmp;
3954
3955                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3956                                                 METADATA_IP_TUNNEL,
3957                                                 GFP_KERNEL);
3958                 if (!tmp)
3959                         return NULL;
3960                 if (cmpxchg(&md_dst, NULL, tmp))
3961                         metadata_dst_free_percpu(tmp);
3962         }
3963
3964         switch (which) {
3965         case BPF_FUNC_skb_set_tunnel_key:
3966                 return &bpf_skb_set_tunnel_key_proto;
3967         case BPF_FUNC_skb_set_tunnel_opt:
3968                 return &bpf_skb_set_tunnel_opt_proto;
3969         default:
3970                 return NULL;
3971         }
3972 }
3973
3974 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3975            u32, idx)
3976 {
3977         struct bpf_array *array = container_of(map, struct bpf_array, map);
3978         struct cgroup *cgrp;
3979         struct sock *sk;
3980
3981         sk = skb_to_full_sk(skb);
3982         if (!sk || !sk_fullsock(sk))
3983                 return -ENOENT;
3984         if (unlikely(idx >= array->map.max_entries))
3985                 return -E2BIG;
3986
3987         cgrp = READ_ONCE(array->ptrs[idx]);
3988         if (unlikely(!cgrp))
3989                 return -EAGAIN;
3990
3991         return sk_under_cgroup_hierarchy(sk, cgrp);
3992 }
3993
3994 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3995         .func           = bpf_skb_under_cgroup,
3996         .gpl_only       = false,
3997         .ret_type       = RET_INTEGER,
3998         .arg1_type      = ARG_PTR_TO_CTX,
3999         .arg2_type      = ARG_CONST_MAP_PTR,
4000         .arg3_type      = ARG_ANYTHING,
4001 };
4002
4003 #ifdef CONFIG_SOCK_CGROUP_DATA
4004 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4005 {
4006         struct sock *sk = skb_to_full_sk(skb);
4007         struct cgroup *cgrp;
4008
4009         if (!sk || !sk_fullsock(sk))
4010                 return 0;
4011
4012         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4013         return cgroup_id(cgrp);
4014 }
4015
4016 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4017         .func           = bpf_skb_cgroup_id,
4018         .gpl_only       = false,
4019         .ret_type       = RET_INTEGER,
4020         .arg1_type      = ARG_PTR_TO_CTX,
4021 };
4022
4023 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4024            ancestor_level)
4025 {
4026         struct sock *sk = skb_to_full_sk(skb);
4027         struct cgroup *ancestor;
4028         struct cgroup *cgrp;
4029
4030         if (!sk || !sk_fullsock(sk))
4031                 return 0;
4032
4033         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4034         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4035         if (!ancestor)
4036                 return 0;
4037
4038         return cgroup_id(ancestor);
4039 }
4040
4041 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4042         .func           = bpf_skb_ancestor_cgroup_id,
4043         .gpl_only       = false,
4044         .ret_type       = RET_INTEGER,
4045         .arg1_type      = ARG_PTR_TO_CTX,
4046         .arg2_type      = ARG_ANYTHING,
4047 };
4048 #endif
4049
4050 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4051                                   unsigned long off, unsigned long len)
4052 {
4053         memcpy(dst_buff, src_buff + off, len);
4054         return 0;
4055 }
4056
4057 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4058            u64, flags, void *, meta, u64, meta_size)
4059 {
4060         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4061
4062         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4063                 return -EINVAL;
4064         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4065                 return -EFAULT;
4066
4067         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4068                                 xdp_size, bpf_xdp_copy);
4069 }
4070
4071 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4072         .func           = bpf_xdp_event_output,
4073         .gpl_only       = true,
4074         .ret_type       = RET_INTEGER,
4075         .arg1_type      = ARG_PTR_TO_CTX,
4076         .arg2_type      = ARG_CONST_MAP_PTR,
4077         .arg3_type      = ARG_ANYTHING,
4078         .arg4_type      = ARG_PTR_TO_MEM,
4079         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4080 };
4081
4082 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4083 {
4084         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4085 }
4086
4087 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4088         .func           = bpf_get_socket_cookie,
4089         .gpl_only       = false,
4090         .ret_type       = RET_INTEGER,
4091         .arg1_type      = ARG_PTR_TO_CTX,
4092 };
4093
4094 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4095 {
4096         return sock_gen_cookie(ctx->sk);
4097 }
4098
4099 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4100         .func           = bpf_get_socket_cookie_sock_addr,
4101         .gpl_only       = false,
4102         .ret_type       = RET_INTEGER,
4103         .arg1_type      = ARG_PTR_TO_CTX,
4104 };
4105
4106 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4107 {
4108         return sock_gen_cookie(ctx->sk);
4109 }
4110
4111 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4112         .func           = bpf_get_socket_cookie_sock_ops,
4113         .gpl_only       = false,
4114         .ret_type       = RET_INTEGER,
4115         .arg1_type      = ARG_PTR_TO_CTX,
4116 };
4117
4118 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4119 {
4120         struct sock *sk = sk_to_full_sk(skb->sk);
4121         kuid_t kuid;
4122
4123         if (!sk || !sk_fullsock(sk))
4124                 return overflowuid;
4125         kuid = sock_net_uid(sock_net(sk), sk);
4126         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4127 }
4128
4129 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4130         .func           = bpf_get_socket_uid,
4131         .gpl_only       = false,
4132         .ret_type       = RET_INTEGER,
4133         .arg1_type      = ARG_PTR_TO_CTX,
4134 };
4135
4136 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4137            struct bpf_map *, map, u64, flags, void *, data, u64, size)
4138 {
4139         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4140                 return -EINVAL;
4141
4142         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4143 }
4144
4145 static const struct bpf_func_proto bpf_sockopt_event_output_proto =  {
4146         .func           = bpf_sockopt_event_output,
4147         .gpl_only       = true,
4148         .ret_type       = RET_INTEGER,
4149         .arg1_type      = ARG_PTR_TO_CTX,
4150         .arg2_type      = ARG_CONST_MAP_PTR,
4151         .arg3_type      = ARG_ANYTHING,
4152         .arg4_type      = ARG_PTR_TO_MEM,
4153         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4154 };
4155
4156 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4157            int, level, int, optname, char *, optval, int, optlen)
4158 {
4159         struct sock *sk = bpf_sock->sk;
4160         int ret = 0;
4161         int val;
4162
4163         if (!sk_fullsock(sk))
4164                 return -EINVAL;
4165
4166         if (level == SOL_SOCKET) {
4167                 if (optlen != sizeof(int))
4168                         return -EINVAL;
4169                 val = *((int *)optval);
4170
4171                 /* Only some socketops are supported */
4172                 switch (optname) {
4173                 case SO_RCVBUF:
4174                         val = min_t(u32, val, sysctl_rmem_max);
4175                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4176                         WRITE_ONCE(sk->sk_rcvbuf,
4177                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4178                         break;
4179                 case SO_SNDBUF:
4180                         val = min_t(u32, val, sysctl_wmem_max);
4181                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4182                         WRITE_ONCE(sk->sk_sndbuf,
4183                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4184                         break;
4185                 case SO_MAX_PACING_RATE: /* 32bit version */
4186                         if (val != ~0U)
4187                                 cmpxchg(&sk->sk_pacing_status,
4188                                         SK_PACING_NONE,
4189                                         SK_PACING_NEEDED);
4190                         sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4191                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4192                                                  sk->sk_max_pacing_rate);
4193                         break;
4194                 case SO_PRIORITY:
4195                         sk->sk_priority = val;
4196                         break;
4197                 case SO_RCVLOWAT:
4198                         if (val < 0)
4199                                 val = INT_MAX;
4200                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4201                         break;
4202                 case SO_MARK:
4203                         if (sk->sk_mark != val) {
4204                                 sk->sk_mark = val;
4205                                 sk_dst_reset(sk);
4206                         }
4207                         break;
4208                 default:
4209                         ret = -EINVAL;
4210                 }
4211 #ifdef CONFIG_INET
4212         } else if (level == SOL_IP) {
4213                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4214                         return -EINVAL;
4215
4216                 val = *((int *)optval);
4217                 /* Only some options are supported */
4218                 switch (optname) {
4219                 case IP_TOS:
4220                         if (val < -1 || val > 0xff) {
4221                                 ret = -EINVAL;
4222                         } else {
4223                                 struct inet_sock *inet = inet_sk(sk);
4224
4225                                 if (val == -1)
4226                                         val = 0;
4227                                 inet->tos = val;
4228                         }
4229                         break;
4230                 default:
4231                         ret = -EINVAL;
4232                 }
4233 #if IS_ENABLED(CONFIG_IPV6)
4234         } else if (level == SOL_IPV6) {
4235                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4236                         return -EINVAL;
4237
4238                 val = *((int *)optval);
4239                 /* Only some options are supported */
4240                 switch (optname) {
4241                 case IPV6_TCLASS:
4242                         if (val < -1 || val > 0xff) {
4243                                 ret = -EINVAL;
4244                         } else {
4245                                 struct ipv6_pinfo *np = inet6_sk(sk);
4246
4247                                 if (val == -1)
4248                                         val = 0;
4249                                 np->tclass = val;
4250                         }
4251                         break;
4252                 default:
4253                         ret = -EINVAL;
4254                 }
4255 #endif
4256         } else if (level == SOL_TCP &&
4257                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4258                 if (optname == TCP_CONGESTION) {
4259                         char name[TCP_CA_NAME_MAX];
4260                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4261
4262                         strncpy(name, optval, min_t(long, optlen,
4263                                                     TCP_CA_NAME_MAX-1));
4264                         name[TCP_CA_NAME_MAX-1] = 0;
4265                         ret = tcp_set_congestion_control(sk, name, false,
4266                                                          reinit, true);
4267                 } else {
4268                         struct tcp_sock *tp = tcp_sk(sk);
4269
4270                         if (optlen != sizeof(int))
4271                                 return -EINVAL;
4272
4273                         val = *((int *)optval);
4274                         /* Only some options are supported */
4275                         switch (optname) {
4276                         case TCP_BPF_IW:
4277                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4278                                         ret = -EINVAL;
4279                                 else
4280                                         tp->snd_cwnd = val;
4281                                 break;
4282                         case TCP_BPF_SNDCWND_CLAMP:
4283                                 if (val <= 0) {
4284                                         ret = -EINVAL;
4285                                 } else {
4286                                         tp->snd_cwnd_clamp = val;
4287                                         tp->snd_ssthresh = val;
4288                                 }
4289                                 break;
4290                         case TCP_SAVE_SYN:
4291                                 if (val < 0 || val > 1)
4292                                         ret = -EINVAL;
4293                                 else
4294                                         tp->save_syn = val;
4295                                 break;
4296                         default:
4297                                 ret = -EINVAL;
4298                         }
4299                 }
4300 #endif
4301         } else {
4302                 ret = -EINVAL;
4303         }
4304         return ret;
4305 }
4306
4307 static const struct bpf_func_proto bpf_setsockopt_proto = {
4308         .func           = bpf_setsockopt,
4309         .gpl_only       = false,
4310         .ret_type       = RET_INTEGER,
4311         .arg1_type      = ARG_PTR_TO_CTX,
4312         .arg2_type      = ARG_ANYTHING,
4313         .arg3_type      = ARG_ANYTHING,
4314         .arg4_type      = ARG_PTR_TO_MEM,
4315         .arg5_type      = ARG_CONST_SIZE,
4316 };
4317
4318 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4319            int, level, int, optname, char *, optval, int, optlen)
4320 {
4321         struct sock *sk = bpf_sock->sk;
4322
4323         if (!sk_fullsock(sk))
4324                 goto err_clear;
4325 #ifdef CONFIG_INET
4326         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4327                 struct inet_connection_sock *icsk;
4328                 struct tcp_sock *tp;
4329
4330                 switch (optname) {
4331                 case TCP_CONGESTION:
4332                         icsk = inet_csk(sk);
4333
4334                         if (!icsk->icsk_ca_ops || optlen <= 1)
4335                                 goto err_clear;
4336                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4337                         optval[optlen - 1] = 0;
4338                         break;
4339                 case TCP_SAVED_SYN:
4340                         tp = tcp_sk(sk);
4341
4342                         if (optlen <= 0 || !tp->saved_syn ||
4343                             optlen > tp->saved_syn[0])
4344                                 goto err_clear;
4345                         memcpy(optval, tp->saved_syn + 1, optlen);
4346                         break;
4347                 default:
4348                         goto err_clear;
4349                 }
4350         } else if (level == SOL_IP) {
4351                 struct inet_sock *inet = inet_sk(sk);
4352
4353                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4354                         goto err_clear;
4355
4356                 /* Only some options are supported */
4357                 switch (optname) {
4358                 case IP_TOS:
4359                         *((int *)optval) = (int)inet->tos;
4360                         break;
4361                 default:
4362                         goto err_clear;
4363                 }
4364 #if IS_ENABLED(CONFIG_IPV6)
4365         } else if (level == SOL_IPV6) {
4366                 struct ipv6_pinfo *np = inet6_sk(sk);
4367
4368                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4369                         goto err_clear;
4370
4371                 /* Only some options are supported */
4372                 switch (optname) {
4373                 case IPV6_TCLASS:
4374                         *((int *)optval) = (int)np->tclass;
4375                         break;
4376                 default:
4377                         goto err_clear;
4378                 }
4379 #endif
4380         } else {
4381                 goto err_clear;
4382         }
4383         return 0;
4384 #endif
4385 err_clear:
4386         memset(optval, 0, optlen);
4387         return -EINVAL;
4388 }
4389
4390 static const struct bpf_func_proto bpf_getsockopt_proto = {
4391         .func           = bpf_getsockopt,
4392         .gpl_only       = false,
4393         .ret_type       = RET_INTEGER,
4394         .arg1_type      = ARG_PTR_TO_CTX,
4395         .arg2_type      = ARG_ANYTHING,
4396         .arg3_type      = ARG_ANYTHING,
4397         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4398         .arg5_type      = ARG_CONST_SIZE,
4399 };
4400
4401 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4402            int, argval)
4403 {
4404         struct sock *sk = bpf_sock->sk;
4405         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4406
4407         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4408                 return -EINVAL;
4409
4410         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4411
4412         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4413 }
4414
4415 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4416         .func           = bpf_sock_ops_cb_flags_set,
4417         .gpl_only       = false,
4418         .ret_type       = RET_INTEGER,
4419         .arg1_type      = ARG_PTR_TO_CTX,
4420         .arg2_type      = ARG_ANYTHING,
4421 };
4422
4423 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4424 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4425
4426 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4427            int, addr_len)
4428 {
4429 #ifdef CONFIG_INET
4430         struct sock *sk = ctx->sk;
4431         int err;
4432
4433         /* Binding to port can be expensive so it's prohibited in the helper.
4434          * Only binding to IP is supported.
4435          */
4436         err = -EINVAL;
4437         if (addr_len < offsetofend(struct sockaddr, sa_family))
4438                 return err;
4439         if (addr->sa_family == AF_INET) {
4440                 if (addr_len < sizeof(struct sockaddr_in))
4441                         return err;
4442                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4443                         return err;
4444                 return __inet_bind(sk, addr, addr_len, true, false);
4445 #if IS_ENABLED(CONFIG_IPV6)
4446         } else if (addr->sa_family == AF_INET6) {
4447                 if (addr_len < SIN6_LEN_RFC2133)
4448                         return err;
4449                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4450                         return err;
4451                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4452                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4453                  */
4454                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4455 #endif /* CONFIG_IPV6 */
4456         }
4457 #endif /* CONFIG_INET */
4458
4459         return -EAFNOSUPPORT;
4460 }
4461
4462 static const struct bpf_func_proto bpf_bind_proto = {
4463         .func           = bpf_bind,
4464         .gpl_only       = false,
4465         .ret_type       = RET_INTEGER,
4466         .arg1_type      = ARG_PTR_TO_CTX,
4467         .arg2_type      = ARG_PTR_TO_MEM,
4468         .arg3_type      = ARG_CONST_SIZE,
4469 };
4470
4471 #ifdef CONFIG_XFRM
4472 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4473            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4474 {
4475         const struct sec_path *sp = skb_sec_path(skb);
4476         const struct xfrm_state *x;
4477
4478         if (!sp || unlikely(index >= sp->len || flags))
4479                 goto err_clear;
4480
4481         x = sp->xvec[index];
4482
4483         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4484                 goto err_clear;
4485
4486         to->reqid = x->props.reqid;
4487         to->spi = x->id.spi;
4488         to->family = x->props.family;
4489         to->ext = 0;
4490
4491         if (to->family == AF_INET6) {
4492                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4493                        sizeof(to->remote_ipv6));
4494         } else {
4495                 to->remote_ipv4 = x->props.saddr.a4;
4496                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4497         }
4498
4499         return 0;
4500 err_clear:
4501         memset(to, 0, size);
4502         return -EINVAL;
4503 }
4504
4505 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4506         .func           = bpf_skb_get_xfrm_state,
4507         .gpl_only       = false,
4508         .ret_type       = RET_INTEGER,
4509         .arg1_type      = ARG_PTR_TO_CTX,
4510         .arg2_type      = ARG_ANYTHING,
4511         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4512         .arg4_type      = ARG_CONST_SIZE,
4513         .arg5_type      = ARG_ANYTHING,
4514 };
4515 #endif
4516
4517 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4518 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4519                                   const struct neighbour *neigh,
4520                                   const struct net_device *dev)
4521 {
4522         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4523         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4524         params->h_vlan_TCI = 0;
4525         params->h_vlan_proto = 0;
4526         params->ifindex = dev->ifindex;
4527
4528         return 0;
4529 }
4530 #endif
4531
4532 #if IS_ENABLED(CONFIG_INET)
4533 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4534                                u32 flags, bool check_mtu)
4535 {
4536         struct fib_nh_common *nhc;
4537         struct in_device *in_dev;
4538         struct neighbour *neigh;
4539         struct net_device *dev;
4540         struct fib_result res;
4541         struct flowi4 fl4;
4542         int err;
4543         u32 mtu;
4544
4545         dev = dev_get_by_index_rcu(net, params->ifindex);
4546         if (unlikely(!dev))
4547                 return -ENODEV;
4548
4549         /* verify forwarding is enabled on this interface */
4550         in_dev = __in_dev_get_rcu(dev);
4551         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4552                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4553
4554         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4555                 fl4.flowi4_iif = 1;
4556                 fl4.flowi4_oif = params->ifindex;
4557         } else {
4558                 fl4.flowi4_iif = params->ifindex;
4559                 fl4.flowi4_oif = 0;
4560         }
4561         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4562         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4563         fl4.flowi4_flags = 0;
4564
4565         fl4.flowi4_proto = params->l4_protocol;
4566         fl4.daddr = params->ipv4_dst;
4567         fl4.saddr = params->ipv4_src;
4568         fl4.fl4_sport = params->sport;
4569         fl4.fl4_dport = params->dport;
4570
4571         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4572                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4573                 struct fib_table *tb;
4574
4575                 tb = fib_get_table(net, tbid);
4576                 if (unlikely(!tb))
4577                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4578
4579                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4580         } else {
4581                 fl4.flowi4_mark = 0;
4582                 fl4.flowi4_secid = 0;
4583                 fl4.flowi4_tun_key.tun_id = 0;
4584                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4585
4586                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4587         }
4588
4589         if (err) {
4590                 /* map fib lookup errors to RTN_ type */
4591                 if (err == -EINVAL)
4592                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4593                 if (err == -EHOSTUNREACH)
4594                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4595                 if (err == -EACCES)
4596                         return BPF_FIB_LKUP_RET_PROHIBIT;
4597
4598                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4599         }
4600
4601         if (res.type != RTN_UNICAST)
4602                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4603
4604         if (fib_info_num_path(res.fi) > 1)
4605                 fib_select_path(net, &res, &fl4, NULL);
4606
4607         if (check_mtu) {
4608                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4609                 if (params->tot_len > mtu)
4610                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4611         }
4612
4613         nhc = res.nhc;
4614
4615         /* do not handle lwt encaps right now */
4616         if (nhc->nhc_lwtstate)
4617                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4618
4619         dev = nhc->nhc_dev;
4620
4621         params->rt_metric = res.fi->fib_priority;
4622
4623         /* xdp and cls_bpf programs are run in RCU-bh so
4624          * rcu_read_lock_bh is not needed here
4625          */
4626         if (likely(nhc->nhc_gw_family != AF_INET6)) {
4627                 if (nhc->nhc_gw_family)
4628                         params->ipv4_dst = nhc->nhc_gw.ipv4;
4629
4630                 neigh = __ipv4_neigh_lookup_noref(dev,
4631                                                  (__force u32)params->ipv4_dst);
4632         } else {
4633                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
4634
4635                 params->family = AF_INET6;
4636                 *dst = nhc->nhc_gw.ipv6;
4637                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4638         }
4639
4640         if (!neigh)
4641                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4642
4643         return bpf_fib_set_fwd_params(params, neigh, dev);
4644 }
4645 #endif
4646
4647 #if IS_ENABLED(CONFIG_IPV6)
4648 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4649                                u32 flags, bool check_mtu)
4650 {
4651         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4652         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4653         struct fib6_result res = {};
4654         struct neighbour *neigh;
4655         struct net_device *dev;
4656         struct inet6_dev *idev;
4657         struct flowi6 fl6;
4658         int strict = 0;
4659         int oif, err;
4660         u32 mtu;
4661
4662         /* link local addresses are never forwarded */
4663         if (rt6_need_strict(dst) || rt6_need_strict(src))
4664                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4665
4666         dev = dev_get_by_index_rcu(net, params->ifindex);
4667         if (unlikely(!dev))
4668                 return -ENODEV;
4669
4670         idev = __in6_dev_get_safely(dev);
4671         if (unlikely(!idev || !idev->cnf.forwarding))
4672                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4673
4674         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4675                 fl6.flowi6_iif = 1;
4676                 oif = fl6.flowi6_oif = params->ifindex;
4677         } else {
4678                 oif = fl6.flowi6_iif = params->ifindex;
4679                 fl6.flowi6_oif = 0;
4680                 strict = RT6_LOOKUP_F_HAS_SADDR;
4681         }
4682         fl6.flowlabel = params->flowinfo;
4683         fl6.flowi6_scope = 0;
4684         fl6.flowi6_flags = 0;
4685         fl6.mp_hash = 0;
4686
4687         fl6.flowi6_proto = params->l4_protocol;
4688         fl6.daddr = *dst;
4689         fl6.saddr = *src;
4690         fl6.fl6_sport = params->sport;
4691         fl6.fl6_dport = params->dport;
4692
4693         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4694                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4695                 struct fib6_table *tb;
4696
4697                 tb = ipv6_stub->fib6_get_table(net, tbid);
4698                 if (unlikely(!tb))
4699                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4700
4701                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
4702                                                    strict);
4703         } else {
4704                 fl6.flowi6_mark = 0;
4705                 fl6.flowi6_secid = 0;
4706                 fl6.flowi6_tun_key.tun_id = 0;
4707                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4708
4709                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
4710         }
4711
4712         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
4713                      res.f6i == net->ipv6.fib6_null_entry))
4714                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4715
4716         switch (res.fib6_type) {
4717         /* only unicast is forwarded */
4718         case RTN_UNICAST:
4719                 break;
4720         case RTN_BLACKHOLE:
4721                 return BPF_FIB_LKUP_RET_BLACKHOLE;
4722         case RTN_UNREACHABLE:
4723                 return BPF_FIB_LKUP_RET_UNREACHABLE;
4724         case RTN_PROHIBIT:
4725                 return BPF_FIB_LKUP_RET_PROHIBIT;
4726         default:
4727                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4728         }
4729
4730         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
4731                                     fl6.flowi6_oif != 0, NULL, strict);
4732
4733         if (check_mtu) {
4734                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
4735                 if (params->tot_len > mtu)
4736                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4737         }
4738
4739         if (res.nh->fib_nh_lws)
4740                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4741
4742         if (res.nh->fib_nh_gw_family)
4743                 *dst = res.nh->fib_nh_gw6;
4744
4745         dev = res.nh->fib_nh_dev;
4746         params->rt_metric = res.f6i->fib6_metric;
4747
4748         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4749          * not needed here.
4750          */
4751         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4752         if (!neigh)
4753                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4754
4755         return bpf_fib_set_fwd_params(params, neigh, dev);
4756 }
4757 #endif
4758
4759 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4760            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4761 {
4762         if (plen < sizeof(*params))
4763                 return -EINVAL;
4764
4765         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4766                 return -EINVAL;
4767
4768         switch (params->family) {
4769 #if IS_ENABLED(CONFIG_INET)
4770         case AF_INET:
4771                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4772                                            flags, true);
4773 #endif
4774 #if IS_ENABLED(CONFIG_IPV6)
4775         case AF_INET6:
4776                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4777                                            flags, true);
4778 #endif
4779         }
4780         return -EAFNOSUPPORT;
4781 }
4782
4783 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4784         .func           = bpf_xdp_fib_lookup,
4785         .gpl_only       = true,
4786         .ret_type       = RET_INTEGER,
4787         .arg1_type      = ARG_PTR_TO_CTX,
4788         .arg2_type      = ARG_PTR_TO_MEM,
4789         .arg3_type      = ARG_CONST_SIZE,
4790         .arg4_type      = ARG_ANYTHING,
4791 };
4792
4793 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4794            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4795 {
4796         struct net *net = dev_net(skb->dev);
4797         int rc = -EAFNOSUPPORT;
4798
4799         if (plen < sizeof(*params))
4800                 return -EINVAL;
4801
4802         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4803                 return -EINVAL;
4804
4805         switch (params->family) {
4806 #if IS_ENABLED(CONFIG_INET)
4807         case AF_INET:
4808                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4809                 break;
4810 #endif
4811 #if IS_ENABLED(CONFIG_IPV6)
4812         case AF_INET6:
4813                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4814                 break;
4815 #endif
4816         }
4817
4818         if (!rc) {
4819                 struct net_device *dev;
4820
4821                 dev = dev_get_by_index_rcu(net, params->ifindex);
4822                 if (!is_skb_forwardable(dev, skb))
4823                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4824         }
4825
4826         return rc;
4827 }
4828
4829 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4830         .func           = bpf_skb_fib_lookup,
4831         .gpl_only       = true,
4832         .ret_type       = RET_INTEGER,
4833         .arg1_type      = ARG_PTR_TO_CTX,
4834         .arg2_type      = ARG_PTR_TO_MEM,
4835         .arg3_type      = ARG_CONST_SIZE,
4836         .arg4_type      = ARG_ANYTHING,
4837 };
4838
4839 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4840 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4841 {
4842         int err;
4843         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4844
4845         if (!seg6_validate_srh(srh, len))
4846                 return -EINVAL;
4847
4848         switch (type) {
4849         case BPF_LWT_ENCAP_SEG6_INLINE:
4850                 if (skb->protocol != htons(ETH_P_IPV6))
4851                         return -EBADMSG;
4852
4853                 err = seg6_do_srh_inline(skb, srh);
4854                 break;
4855         case BPF_LWT_ENCAP_SEG6:
4856                 skb_reset_inner_headers(skb);
4857                 skb->encapsulation = 1;
4858                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4859                 break;
4860         default:
4861                 return -EINVAL;
4862         }
4863
4864         bpf_compute_data_pointers(skb);
4865         if (err)
4866                 return err;
4867
4868         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4869         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4870
4871         return seg6_lookup_nexthop(skb, NULL, 0);
4872 }
4873 #endif /* CONFIG_IPV6_SEG6_BPF */
4874
4875 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4876 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4877                              bool ingress)
4878 {
4879         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4880 }
4881 #endif
4882
4883 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4884            u32, len)
4885 {
4886         switch (type) {
4887 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4888         case BPF_LWT_ENCAP_SEG6:
4889         case BPF_LWT_ENCAP_SEG6_INLINE:
4890                 return bpf_push_seg6_encap(skb, type, hdr, len);
4891 #endif
4892 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4893         case BPF_LWT_ENCAP_IP:
4894                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
4895 #endif
4896         default:
4897                 return -EINVAL;
4898         }
4899 }
4900
4901 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4902            void *, hdr, u32, len)
4903 {
4904         switch (type) {
4905 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4906         case BPF_LWT_ENCAP_IP:
4907                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
4908 #endif
4909         default:
4910                 return -EINVAL;
4911         }
4912 }
4913
4914 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4915         .func           = bpf_lwt_in_push_encap,
4916         .gpl_only       = false,
4917         .ret_type       = RET_INTEGER,
4918         .arg1_type      = ARG_PTR_TO_CTX,
4919         .arg2_type      = ARG_ANYTHING,
4920         .arg3_type      = ARG_PTR_TO_MEM,
4921         .arg4_type      = ARG_CONST_SIZE
4922 };
4923
4924 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
4925         .func           = bpf_lwt_xmit_push_encap,
4926         .gpl_only       = false,
4927         .ret_type       = RET_INTEGER,
4928         .arg1_type      = ARG_PTR_TO_CTX,
4929         .arg2_type      = ARG_ANYTHING,
4930         .arg3_type      = ARG_PTR_TO_MEM,
4931         .arg4_type      = ARG_CONST_SIZE
4932 };
4933
4934 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4935 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4936            const void *, from, u32, len)
4937 {
4938         struct seg6_bpf_srh_state *srh_state =
4939                 this_cpu_ptr(&seg6_bpf_srh_states);
4940         struct ipv6_sr_hdr *srh = srh_state->srh;
4941         void *srh_tlvs, *srh_end, *ptr;
4942         int srhoff = 0;
4943
4944         if (srh == NULL)
4945                 return -EINVAL;
4946
4947         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4948         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4949
4950         ptr = skb->data + offset;
4951         if (ptr >= srh_tlvs && ptr + len <= srh_end)
4952                 srh_state->valid = false;
4953         else if (ptr < (void *)&srh->flags ||
4954                  ptr + len > (void *)&srh->segments)
4955                 return -EFAULT;
4956
4957         if (unlikely(bpf_try_make_writable(skb, offset + len)))
4958                 return -EFAULT;
4959         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4960                 return -EINVAL;
4961         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4962
4963         memcpy(skb->data + offset, from, len);
4964         return 0;
4965 }
4966
4967 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4968         .func           = bpf_lwt_seg6_store_bytes,
4969         .gpl_only       = false,
4970         .ret_type       = RET_INTEGER,
4971         .arg1_type      = ARG_PTR_TO_CTX,
4972         .arg2_type      = ARG_ANYTHING,
4973         .arg3_type      = ARG_PTR_TO_MEM,
4974         .arg4_type      = ARG_CONST_SIZE
4975 };
4976
4977 static void bpf_update_srh_state(struct sk_buff *skb)
4978 {
4979         struct seg6_bpf_srh_state *srh_state =
4980                 this_cpu_ptr(&seg6_bpf_srh_states);
4981         int srhoff = 0;
4982
4983         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4984                 srh_state->srh = NULL;
4985         } else {
4986                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4987                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4988                 srh_state->valid = true;
4989         }
4990 }
4991
4992 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4993            u32, action, void *, param, u32, param_len)
4994 {
4995         struct seg6_bpf_srh_state *srh_state =
4996                 this_cpu_ptr(&seg6_bpf_srh_states);
4997         int hdroff = 0;
4998         int err;
4999
5000         switch (action) {
5001         case SEG6_LOCAL_ACTION_END_X:
5002                 if (!seg6_bpf_has_valid_srh(skb))
5003                         return -EBADMSG;
5004                 if (param_len != sizeof(struct in6_addr))
5005                         return -EINVAL;
5006                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5007         case SEG6_LOCAL_ACTION_END_T:
5008                 if (!seg6_bpf_has_valid_srh(skb))
5009                         return -EBADMSG;
5010                 if (param_len != sizeof(int))
5011                         return -EINVAL;
5012                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5013         case SEG6_LOCAL_ACTION_END_DT6:
5014                 if (!seg6_bpf_has_valid_srh(skb))
5015                         return -EBADMSG;
5016                 if (param_len != sizeof(int))
5017                         return -EINVAL;
5018
5019                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5020                         return -EBADMSG;
5021                 if (!pskb_pull(skb, hdroff))
5022                         return -EBADMSG;
5023
5024                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5025                 skb_reset_network_header(skb);
5026                 skb_reset_transport_header(skb);
5027                 skb->encapsulation = 0;
5028
5029                 bpf_compute_data_pointers(skb);
5030                 bpf_update_srh_state(skb);
5031                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5032         case SEG6_LOCAL_ACTION_END_B6:
5033                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5034                         return -EBADMSG;
5035                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5036                                           param, param_len);
5037                 if (!err)
5038                         bpf_update_srh_state(skb);
5039
5040                 return err;
5041         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5042                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5043                         return -EBADMSG;
5044                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5045                                           param, param_len);
5046                 if (!err)
5047                         bpf_update_srh_state(skb);
5048
5049                 return err;
5050         default:
5051                 return -EINVAL;
5052         }
5053 }
5054
5055 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5056         .func           = bpf_lwt_seg6_action,
5057         .gpl_only       = false,
5058         .ret_type       = RET_INTEGER,
5059         .arg1_type      = ARG_PTR_TO_CTX,
5060         .arg2_type      = ARG_ANYTHING,
5061         .arg3_type      = ARG_PTR_TO_MEM,
5062         .arg4_type      = ARG_CONST_SIZE
5063 };
5064
5065 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5066            s32, len)
5067 {
5068         struct seg6_bpf_srh_state *srh_state =
5069                 this_cpu_ptr(&seg6_bpf_srh_states);
5070         struct ipv6_sr_hdr *srh = srh_state->srh;
5071         void *srh_end, *srh_tlvs, *ptr;
5072         struct ipv6hdr *hdr;
5073         int srhoff = 0;
5074         int ret;
5075
5076         if (unlikely(srh == NULL))
5077                 return -EINVAL;
5078
5079         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5080                         ((srh->first_segment + 1) << 4));
5081         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5082                         srh_state->hdrlen);
5083         ptr = skb->data + offset;
5084
5085         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5086                 return -EFAULT;
5087         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5088                 return -EFAULT;
5089
5090         if (len > 0) {
5091                 ret = skb_cow_head(skb, len);
5092                 if (unlikely(ret < 0))
5093                         return ret;
5094
5095                 ret = bpf_skb_net_hdr_push(skb, offset, len);
5096         } else {
5097                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5098         }
5099
5100         bpf_compute_data_pointers(skb);
5101         if (unlikely(ret < 0))
5102                 return ret;
5103
5104         hdr = (struct ipv6hdr *)skb->data;
5105         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5106
5107         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5108                 return -EINVAL;
5109         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5110         srh_state->hdrlen += len;
5111         srh_state->valid = false;
5112         return 0;
5113 }
5114
5115 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5116         .func           = bpf_lwt_seg6_adjust_srh,
5117         .gpl_only       = false,
5118         .ret_type       = RET_INTEGER,
5119         .arg1_type      = ARG_PTR_TO_CTX,
5120         .arg2_type      = ARG_ANYTHING,
5121         .arg3_type      = ARG_ANYTHING,
5122 };
5123 #endif /* CONFIG_IPV6_SEG6_BPF */
5124
5125 #ifdef CONFIG_INET
5126 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5127                               int dif, int sdif, u8 family, u8 proto)
5128 {
5129         bool refcounted = false;
5130         struct sock *sk = NULL;
5131
5132         if (family == AF_INET) {
5133                 __be32 src4 = tuple->ipv4.saddr;
5134                 __be32 dst4 = tuple->ipv4.daddr;
5135
5136                 if (proto == IPPROTO_TCP)
5137                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5138                                            src4, tuple->ipv4.sport,
5139                                            dst4, tuple->ipv4.dport,
5140                                            dif, sdif, &refcounted);
5141                 else
5142                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5143                                                dst4, tuple->ipv4.dport,
5144                                                dif, sdif, &udp_table, NULL);
5145 #if IS_ENABLED(CONFIG_IPV6)
5146         } else {
5147                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5148                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5149
5150                 if (proto == IPPROTO_TCP)
5151                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5152                                             src6, tuple->ipv6.sport,
5153                                             dst6, ntohs(tuple->ipv6.dport),
5154                                             dif, sdif, &refcounted);
5155                 else if (likely(ipv6_bpf_stub))
5156                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5157                                                             src6, tuple->ipv6.sport,
5158                                                             dst6, tuple->ipv6.dport,
5159                                                             dif, sdif,
5160                                                             &udp_table, NULL);
5161 #endif
5162         }
5163
5164         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5165                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5166                 sk = NULL;
5167         }
5168         return sk;
5169 }
5170
5171 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5172  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5173  * Returns the socket as an 'unsigned long' to simplify the casting in the
5174  * callers to satisfy BPF_CALL declarations.
5175  */
5176 static struct sock *
5177 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5178                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5179                  u64 flags)
5180 {
5181         struct sock *sk = NULL;
5182         u8 family = AF_UNSPEC;
5183         struct net *net;
5184         int sdif;
5185
5186         if (len == sizeof(tuple->ipv4))
5187                 family = AF_INET;
5188         else if (len == sizeof(tuple->ipv6))
5189                 family = AF_INET6;
5190         else
5191                 return NULL;
5192
5193         if (unlikely(family == AF_UNSPEC || flags ||
5194                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5195                 goto out;
5196
5197         if (family == AF_INET)
5198                 sdif = inet_sdif(skb);
5199         else
5200                 sdif = inet6_sdif(skb);
5201
5202         if ((s32)netns_id < 0) {
5203                 net = caller_net;
5204                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5205         } else {
5206                 net = get_net_ns_by_id(caller_net, netns_id);
5207                 if (unlikely(!net))
5208                         goto out;
5209                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5210                 put_net(net);
5211         }
5212
5213 out:
5214         return sk;
5215 }
5216
5217 static struct sock *
5218 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5219                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5220                 u64 flags)
5221 {
5222         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5223                                            ifindex, proto, netns_id, flags);
5224
5225         if (sk) {
5226                 sk = sk_to_full_sk(sk);
5227                 if (!sk_fullsock(sk)) {
5228                         sock_gen_put(sk);
5229                         return NULL;
5230                 }
5231         }
5232
5233         return sk;
5234 }
5235
5236 static struct sock *
5237 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5238                u8 proto, u64 netns_id, u64 flags)
5239 {
5240         struct net *caller_net;
5241         int ifindex;
5242
5243         if (skb->dev) {
5244                 caller_net = dev_net(skb->dev);
5245                 ifindex = skb->dev->ifindex;
5246         } else {
5247                 caller_net = sock_net(skb->sk);
5248                 ifindex = 0;
5249         }
5250
5251         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5252                                 netns_id, flags);
5253 }
5254
5255 static struct sock *
5256 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5257               u8 proto, u64 netns_id, u64 flags)
5258 {
5259         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5260                                          flags);
5261
5262         if (sk) {
5263                 sk = sk_to_full_sk(sk);
5264                 if (!sk_fullsock(sk)) {
5265                         sock_gen_put(sk);
5266                         return NULL;
5267                 }
5268         }
5269
5270         return sk;
5271 }
5272
5273 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5274            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5275 {
5276         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5277                                              netns_id, flags);
5278 }
5279
5280 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5281         .func           = bpf_skc_lookup_tcp,
5282         .gpl_only       = false,
5283         .pkt_access     = true,
5284         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5285         .arg1_type      = ARG_PTR_TO_CTX,
5286         .arg2_type      = ARG_PTR_TO_MEM,
5287         .arg3_type      = ARG_CONST_SIZE,
5288         .arg4_type      = ARG_ANYTHING,
5289         .arg5_type      = ARG_ANYTHING,
5290 };
5291
5292 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5293            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5294 {
5295         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5296                                             netns_id, flags);
5297 }
5298
5299 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5300         .func           = bpf_sk_lookup_tcp,
5301         .gpl_only       = false,
5302         .pkt_access     = true,
5303         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5304         .arg1_type      = ARG_PTR_TO_CTX,
5305         .arg2_type      = ARG_PTR_TO_MEM,
5306         .arg3_type      = ARG_CONST_SIZE,
5307         .arg4_type      = ARG_ANYTHING,
5308         .arg5_type      = ARG_ANYTHING,
5309 };
5310
5311 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5312            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5313 {
5314         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5315                                             netns_id, flags);
5316 }
5317
5318 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5319         .func           = bpf_sk_lookup_udp,
5320         .gpl_only       = false,
5321         .pkt_access     = true,
5322         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5323         .arg1_type      = ARG_PTR_TO_CTX,
5324         .arg2_type      = ARG_PTR_TO_MEM,
5325         .arg3_type      = ARG_CONST_SIZE,
5326         .arg4_type      = ARG_ANYTHING,
5327         .arg5_type      = ARG_ANYTHING,
5328 };
5329
5330 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5331 {
5332         /* Only full sockets have sk->sk_flags. */
5333         if (!sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE))
5334                 sock_gen_put(sk);
5335         return 0;
5336 }
5337
5338 static const struct bpf_func_proto bpf_sk_release_proto = {
5339         .func           = bpf_sk_release,
5340         .gpl_only       = false,
5341         .ret_type       = RET_INTEGER,
5342         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5343 };
5344
5345 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5346            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5347 {
5348         struct net *caller_net = dev_net(ctx->rxq->dev);
5349         int ifindex = ctx->rxq->dev->ifindex;
5350
5351         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5352                                               ifindex, IPPROTO_UDP, netns_id,
5353                                               flags);
5354 }
5355
5356 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5357         .func           = bpf_xdp_sk_lookup_udp,
5358         .gpl_only       = false,
5359         .pkt_access     = true,
5360         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5361         .arg1_type      = ARG_PTR_TO_CTX,
5362         .arg2_type      = ARG_PTR_TO_MEM,
5363         .arg3_type      = ARG_CONST_SIZE,
5364         .arg4_type      = ARG_ANYTHING,
5365         .arg5_type      = ARG_ANYTHING,
5366 };
5367
5368 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
5369            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5370 {
5371         struct net *caller_net = dev_net(ctx->rxq->dev);
5372         int ifindex = ctx->rxq->dev->ifindex;
5373
5374         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
5375                                                ifindex, IPPROTO_TCP, netns_id,
5376                                                flags);
5377 }
5378
5379 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
5380         .func           = bpf_xdp_skc_lookup_tcp,
5381         .gpl_only       = false,
5382         .pkt_access     = true,
5383         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5384         .arg1_type      = ARG_PTR_TO_CTX,
5385         .arg2_type      = ARG_PTR_TO_MEM,
5386         .arg3_type      = ARG_CONST_SIZE,
5387         .arg4_type      = ARG_ANYTHING,
5388         .arg5_type      = ARG_ANYTHING,
5389 };
5390
5391 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5392            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5393 {
5394         struct net *caller_net = dev_net(ctx->rxq->dev);
5395         int ifindex = ctx->rxq->dev->ifindex;
5396
5397         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5398                                               ifindex, IPPROTO_TCP, netns_id,
5399                                               flags);
5400 }
5401
5402 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5403         .func           = bpf_xdp_sk_lookup_tcp,
5404         .gpl_only       = false,
5405         .pkt_access     = true,
5406         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5407         .arg1_type      = ARG_PTR_TO_CTX,
5408         .arg2_type      = ARG_PTR_TO_MEM,
5409         .arg3_type      = ARG_CONST_SIZE,
5410         .arg4_type      = ARG_ANYTHING,
5411         .arg5_type      = ARG_ANYTHING,
5412 };
5413
5414 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5415            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5416 {
5417         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
5418                                                sock_net(ctx->sk), 0,
5419                                                IPPROTO_TCP, netns_id, flags);
5420 }
5421
5422 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
5423         .func           = bpf_sock_addr_skc_lookup_tcp,
5424         .gpl_only       = false,
5425         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5426         .arg1_type      = ARG_PTR_TO_CTX,
5427         .arg2_type      = ARG_PTR_TO_MEM,
5428         .arg3_type      = ARG_CONST_SIZE,
5429         .arg4_type      = ARG_ANYTHING,
5430         .arg5_type      = ARG_ANYTHING,
5431 };
5432
5433 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5434            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5435 {
5436         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5437                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
5438                                               netns_id, flags);
5439 }
5440
5441 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5442         .func           = bpf_sock_addr_sk_lookup_tcp,
5443         .gpl_only       = false,
5444         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5445         .arg1_type      = ARG_PTR_TO_CTX,
5446         .arg2_type      = ARG_PTR_TO_MEM,
5447         .arg3_type      = ARG_CONST_SIZE,
5448         .arg4_type      = ARG_ANYTHING,
5449         .arg5_type      = ARG_ANYTHING,
5450 };
5451
5452 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5453            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5454 {
5455         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5456                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
5457                                               netns_id, flags);
5458 }
5459
5460 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5461         .func           = bpf_sock_addr_sk_lookup_udp,
5462         .gpl_only       = false,
5463         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5464         .arg1_type      = ARG_PTR_TO_CTX,
5465         .arg2_type      = ARG_PTR_TO_MEM,
5466         .arg3_type      = ARG_CONST_SIZE,
5467         .arg4_type      = ARG_ANYTHING,
5468         .arg5_type      = ARG_ANYTHING,
5469 };
5470
5471 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5472                                   struct bpf_insn_access_aux *info)
5473 {
5474         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
5475                                           icsk_retransmits))
5476                 return false;
5477
5478         if (off % size != 0)
5479                 return false;
5480
5481         switch (off) {
5482         case offsetof(struct bpf_tcp_sock, bytes_received):
5483         case offsetof(struct bpf_tcp_sock, bytes_acked):
5484                 return size == sizeof(__u64);
5485         default:
5486                 return size == sizeof(__u32);
5487         }
5488 }
5489
5490 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5491                                     const struct bpf_insn *si,
5492                                     struct bpf_insn *insn_buf,
5493                                     struct bpf_prog *prog, u32 *target_size)
5494 {
5495         struct bpf_insn *insn = insn_buf;
5496
5497 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
5498         do {                                                            \
5499                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
5500                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
5501                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5502                                       si->dst_reg, si->src_reg,         \
5503                                       offsetof(struct tcp_sock, FIELD)); \
5504         } while (0)
5505
5506 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
5507         do {                                                            \
5508                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
5509                                           FIELD) >                      \
5510                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
5511                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
5512                                         struct inet_connection_sock,    \
5513                                         FIELD),                         \
5514                                       si->dst_reg, si->src_reg,         \
5515                                       offsetof(                         \
5516                                         struct inet_connection_sock,    \
5517                                         FIELD));                        \
5518         } while (0)
5519
5520         if (insn > insn_buf)
5521                 return insn - insn_buf;
5522
5523         switch (si->off) {
5524         case offsetof(struct bpf_tcp_sock, rtt_min):
5525                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
5526                              sizeof(struct minmax));
5527                 BUILD_BUG_ON(sizeof(struct minmax) <
5528                              sizeof(struct minmax_sample));
5529
5530                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5531                                       offsetof(struct tcp_sock, rtt_min) +
5532                                       offsetof(struct minmax_sample, v));
5533                 break;
5534         case offsetof(struct bpf_tcp_sock, snd_cwnd):
5535                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
5536                 break;
5537         case offsetof(struct bpf_tcp_sock, srtt_us):
5538                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
5539                 break;
5540         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
5541                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
5542                 break;
5543         case offsetof(struct bpf_tcp_sock, rcv_nxt):
5544                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
5545                 break;
5546         case offsetof(struct bpf_tcp_sock, snd_nxt):
5547                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
5548                 break;
5549         case offsetof(struct bpf_tcp_sock, snd_una):
5550                 BPF_TCP_SOCK_GET_COMMON(snd_una);
5551                 break;
5552         case offsetof(struct bpf_tcp_sock, mss_cache):
5553                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
5554                 break;
5555         case offsetof(struct bpf_tcp_sock, ecn_flags):
5556                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
5557                 break;
5558         case offsetof(struct bpf_tcp_sock, rate_delivered):
5559                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
5560                 break;
5561         case offsetof(struct bpf_tcp_sock, rate_interval_us):
5562                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
5563                 break;
5564         case offsetof(struct bpf_tcp_sock, packets_out):
5565                 BPF_TCP_SOCK_GET_COMMON(packets_out);
5566                 break;
5567         case offsetof(struct bpf_tcp_sock, retrans_out):
5568                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
5569                 break;
5570         case offsetof(struct bpf_tcp_sock, total_retrans):
5571                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
5572                 break;
5573         case offsetof(struct bpf_tcp_sock, segs_in):
5574                 BPF_TCP_SOCK_GET_COMMON(segs_in);
5575                 break;
5576         case offsetof(struct bpf_tcp_sock, data_segs_in):
5577                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
5578                 break;
5579         case offsetof(struct bpf_tcp_sock, segs_out):
5580                 BPF_TCP_SOCK_GET_COMMON(segs_out);
5581                 break;
5582         case offsetof(struct bpf_tcp_sock, data_segs_out):
5583                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
5584                 break;
5585         case offsetof(struct bpf_tcp_sock, lost_out):
5586                 BPF_TCP_SOCK_GET_COMMON(lost_out);
5587                 break;
5588         case offsetof(struct bpf_tcp_sock, sacked_out):
5589                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
5590                 break;
5591         case offsetof(struct bpf_tcp_sock, bytes_received):
5592                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
5593                 break;
5594         case offsetof(struct bpf_tcp_sock, bytes_acked):
5595                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
5596                 break;
5597         case offsetof(struct bpf_tcp_sock, dsack_dups):
5598                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
5599                 break;
5600         case offsetof(struct bpf_tcp_sock, delivered):
5601                 BPF_TCP_SOCK_GET_COMMON(delivered);
5602                 break;
5603         case offsetof(struct bpf_tcp_sock, delivered_ce):
5604                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
5605                 break;
5606         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
5607                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
5608                 break;
5609         }
5610
5611         return insn - insn_buf;
5612 }
5613
5614 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5615 {
5616         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5617                 return (unsigned long)sk;
5618
5619         return (unsigned long)NULL;
5620 }
5621
5622 const struct bpf_func_proto bpf_tcp_sock_proto = {
5623         .func           = bpf_tcp_sock,
5624         .gpl_only       = false,
5625         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
5626         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5627 };
5628
5629 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
5630 {
5631         sk = sk_to_full_sk(sk);
5632
5633         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
5634                 return (unsigned long)sk;
5635
5636         return (unsigned long)NULL;
5637 }
5638
5639 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
5640         .func           = bpf_get_listener_sock,
5641         .gpl_only       = false,
5642         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5643         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5644 };
5645
5646 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5647 {
5648         unsigned int iphdr_len;
5649
5650         if (skb->protocol == cpu_to_be16(ETH_P_IP))
5651                 iphdr_len = sizeof(struct iphdr);
5652         else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5653                 iphdr_len = sizeof(struct ipv6hdr);
5654         else
5655                 return 0;
5656
5657         if (skb_headlen(skb) < iphdr_len)
5658                 return 0;
5659
5660         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5661                 return 0;
5662
5663         return INET_ECN_set_ce(skb);
5664 }
5665
5666 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5667                                   struct bpf_insn_access_aux *info)
5668 {
5669         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
5670                 return false;
5671
5672         if (off % size != 0)
5673                 return false;
5674
5675         switch (off) {
5676         default:
5677                 return size == sizeof(__u32);
5678         }
5679 }
5680
5681 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
5682                                     const struct bpf_insn *si,
5683                                     struct bpf_insn *insn_buf,
5684                                     struct bpf_prog *prog, u32 *target_size)
5685 {
5686         struct bpf_insn *insn = insn_buf;
5687
5688 #define BPF_XDP_SOCK_GET(FIELD)                                         \
5689         do {                                                            \
5690                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
5691                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
5692                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
5693                                       si->dst_reg, si->src_reg,         \
5694                                       offsetof(struct xdp_sock, FIELD)); \
5695         } while (0)
5696
5697         switch (si->off) {
5698         case offsetof(struct bpf_xdp_sock, queue_id):
5699                 BPF_XDP_SOCK_GET(queue_id);
5700                 break;
5701         }
5702
5703         return insn - insn_buf;
5704 }
5705
5706 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5707         .func           = bpf_skb_ecn_set_ce,
5708         .gpl_only       = false,
5709         .ret_type       = RET_INTEGER,
5710         .arg1_type      = ARG_PTR_TO_CTX,
5711 };
5712
5713 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5714            struct tcphdr *, th, u32, th_len)
5715 {
5716 #ifdef CONFIG_SYN_COOKIES
5717         u32 cookie;
5718         int ret;
5719
5720         if (unlikely(th_len < sizeof(*th)))
5721                 return -EINVAL;
5722
5723         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
5724         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5725                 return -EINVAL;
5726
5727         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5728                 return -EINVAL;
5729
5730         if (!th->ack || th->rst || th->syn)
5731                 return -ENOENT;
5732
5733         if (tcp_synq_no_recent_overflow(sk))
5734                 return -ENOENT;
5735
5736         cookie = ntohl(th->ack_seq) - 1;
5737
5738         switch (sk->sk_family) {
5739         case AF_INET:
5740                 if (unlikely(iph_len < sizeof(struct iphdr)))
5741                         return -EINVAL;
5742
5743                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
5744                 break;
5745
5746 #if IS_BUILTIN(CONFIG_IPV6)
5747         case AF_INET6:
5748                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5749                         return -EINVAL;
5750
5751                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
5752                 break;
5753 #endif /* CONFIG_IPV6 */
5754
5755         default:
5756                 return -EPROTONOSUPPORT;
5757         }
5758
5759         if (ret > 0)
5760                 return 0;
5761
5762         return -ENOENT;
5763 #else
5764         return -ENOTSUPP;
5765 #endif
5766 }
5767
5768 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
5769         .func           = bpf_tcp_check_syncookie,
5770         .gpl_only       = true,
5771         .pkt_access     = true,
5772         .ret_type       = RET_INTEGER,
5773         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5774         .arg2_type      = ARG_PTR_TO_MEM,
5775         .arg3_type      = ARG_CONST_SIZE,
5776         .arg4_type      = ARG_PTR_TO_MEM,
5777         .arg5_type      = ARG_CONST_SIZE,
5778 };
5779
5780 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5781            struct tcphdr *, th, u32, th_len)
5782 {
5783 #ifdef CONFIG_SYN_COOKIES
5784         u32 cookie;
5785         u16 mss;
5786
5787         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
5788                 return -EINVAL;
5789
5790         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5791                 return -EINVAL;
5792
5793         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5794                 return -ENOENT;
5795
5796         if (!th->syn || th->ack || th->fin || th->rst)
5797                 return -EINVAL;
5798
5799         if (unlikely(iph_len < sizeof(struct iphdr)))
5800                 return -EINVAL;
5801
5802         /* Both struct iphdr and struct ipv6hdr have the version field at the
5803          * same offset so we can cast to the shorter header (struct iphdr).
5804          */
5805         switch (((struct iphdr *)iph)->version) {
5806         case 4:
5807                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
5808                         return -EINVAL;
5809
5810                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
5811                 break;
5812
5813 #if IS_BUILTIN(CONFIG_IPV6)
5814         case 6:
5815                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5816                         return -EINVAL;
5817
5818                 if (sk->sk_family != AF_INET6)
5819                         return -EINVAL;
5820
5821                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
5822                 break;
5823 #endif /* CONFIG_IPV6 */
5824
5825         default:
5826                 return -EPROTONOSUPPORT;
5827         }
5828         if (mss == 0)
5829                 return -ENOENT;
5830
5831         return cookie | ((u64)mss << 32);
5832 #else
5833         return -EOPNOTSUPP;
5834 #endif /* CONFIG_SYN_COOKIES */
5835 }
5836
5837 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
5838         .func           = bpf_tcp_gen_syncookie,
5839         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
5840         .pkt_access     = true,
5841         .ret_type       = RET_INTEGER,
5842         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5843         .arg2_type      = ARG_PTR_TO_MEM,
5844         .arg3_type      = ARG_CONST_SIZE,
5845         .arg4_type      = ARG_PTR_TO_MEM,
5846         .arg5_type      = ARG_CONST_SIZE,
5847 };
5848
5849 #endif /* CONFIG_INET */
5850
5851 bool bpf_helper_changes_pkt_data(void *func)
5852 {
5853         if (func == bpf_skb_vlan_push ||
5854             func == bpf_skb_vlan_pop ||
5855             func == bpf_skb_store_bytes ||
5856             func == bpf_skb_change_proto ||
5857             func == bpf_skb_change_head ||
5858             func == sk_skb_change_head ||
5859             func == bpf_skb_change_tail ||
5860             func == sk_skb_change_tail ||
5861             func == bpf_skb_adjust_room ||
5862             func == bpf_skb_pull_data ||
5863             func == sk_skb_pull_data ||
5864             func == bpf_clone_redirect ||
5865             func == bpf_l3_csum_replace ||
5866             func == bpf_l4_csum_replace ||
5867             func == bpf_xdp_adjust_head ||
5868             func == bpf_xdp_adjust_meta ||
5869             func == bpf_msg_pull_data ||
5870             func == bpf_msg_push_data ||
5871             func == bpf_msg_pop_data ||
5872             func == bpf_xdp_adjust_tail ||
5873 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5874             func == bpf_lwt_seg6_store_bytes ||
5875             func == bpf_lwt_seg6_adjust_srh ||
5876             func == bpf_lwt_seg6_action ||
5877 #endif
5878             func == bpf_lwt_in_push_encap ||
5879             func == bpf_lwt_xmit_push_encap)
5880                 return true;
5881
5882         return false;
5883 }
5884
5885 const struct bpf_func_proto *
5886 bpf_base_func_proto(enum bpf_func_id func_id)
5887 {
5888         switch (func_id) {
5889         case BPF_FUNC_map_lookup_elem:
5890                 return &bpf_map_lookup_elem_proto;
5891         case BPF_FUNC_map_update_elem:
5892                 return &bpf_map_update_elem_proto;
5893         case BPF_FUNC_map_delete_elem:
5894                 return &bpf_map_delete_elem_proto;
5895         case BPF_FUNC_map_push_elem:
5896                 return &bpf_map_push_elem_proto;
5897         case BPF_FUNC_map_pop_elem:
5898                 return &bpf_map_pop_elem_proto;
5899         case BPF_FUNC_map_peek_elem:
5900                 return &bpf_map_peek_elem_proto;
5901         case BPF_FUNC_get_prandom_u32:
5902                 return &bpf_get_prandom_u32_proto;
5903         case BPF_FUNC_get_smp_processor_id:
5904                 return &bpf_get_raw_smp_processor_id_proto;
5905         case BPF_FUNC_get_numa_node_id:
5906                 return &bpf_get_numa_node_id_proto;
5907         case BPF_FUNC_tail_call:
5908                 return &bpf_tail_call_proto;
5909         case BPF_FUNC_ktime_get_ns:
5910                 return &bpf_ktime_get_ns_proto;
5911         default:
5912                 break;
5913         }
5914
5915         if (!capable(CAP_SYS_ADMIN))
5916                 return NULL;
5917
5918         switch (func_id) {
5919         case BPF_FUNC_spin_lock:
5920                 return &bpf_spin_lock_proto;
5921         case BPF_FUNC_spin_unlock:
5922                 return &bpf_spin_unlock_proto;
5923         case BPF_FUNC_trace_printk:
5924                 return bpf_get_trace_printk_proto();
5925         case BPF_FUNC_jiffies64:
5926                 return &bpf_jiffies64_proto;
5927         default:
5928                 return NULL;
5929         }
5930 }
5931
5932 static const struct bpf_func_proto *
5933 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5934 {
5935         switch (func_id) {
5936         /* inet and inet6 sockets are created in a process
5937          * context so there is always a valid uid/gid
5938          */
5939         case BPF_FUNC_get_current_uid_gid:
5940                 return &bpf_get_current_uid_gid_proto;
5941         case BPF_FUNC_get_local_storage:
5942                 return &bpf_get_local_storage_proto;
5943         default:
5944                 return bpf_base_func_proto(func_id);
5945         }
5946 }
5947
5948 static const struct bpf_func_proto *
5949 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5950 {
5951         switch (func_id) {
5952         /* inet and inet6 sockets are created in a process
5953          * context so there is always a valid uid/gid
5954          */
5955         case BPF_FUNC_get_current_uid_gid:
5956                 return &bpf_get_current_uid_gid_proto;
5957         case BPF_FUNC_bind:
5958                 switch (prog->expected_attach_type) {
5959                 case BPF_CGROUP_INET4_CONNECT:
5960                 case BPF_CGROUP_INET6_CONNECT:
5961                         return &bpf_bind_proto;
5962                 default:
5963                         return NULL;
5964                 }
5965         case BPF_FUNC_get_socket_cookie:
5966                 return &bpf_get_socket_cookie_sock_addr_proto;
5967         case BPF_FUNC_get_local_storage:
5968                 return &bpf_get_local_storage_proto;
5969 #ifdef CONFIG_INET
5970         case BPF_FUNC_sk_lookup_tcp:
5971                 return &bpf_sock_addr_sk_lookup_tcp_proto;
5972         case BPF_FUNC_sk_lookup_udp:
5973                 return &bpf_sock_addr_sk_lookup_udp_proto;
5974         case BPF_FUNC_sk_release:
5975                 return &bpf_sk_release_proto;
5976         case BPF_FUNC_skc_lookup_tcp:
5977                 return &bpf_sock_addr_skc_lookup_tcp_proto;
5978 #endif /* CONFIG_INET */
5979         case BPF_FUNC_sk_storage_get:
5980                 return &bpf_sk_storage_get_proto;
5981         case BPF_FUNC_sk_storage_delete:
5982                 return &bpf_sk_storage_delete_proto;
5983         default:
5984                 return bpf_base_func_proto(func_id);
5985         }
5986 }
5987
5988 static const struct bpf_func_proto *
5989 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5990 {
5991         switch (func_id) {
5992         case BPF_FUNC_skb_load_bytes:
5993                 return &bpf_skb_load_bytes_proto;
5994         case BPF_FUNC_skb_load_bytes_relative:
5995                 return &bpf_skb_load_bytes_relative_proto;
5996         case BPF_FUNC_get_socket_cookie:
5997                 return &bpf_get_socket_cookie_proto;
5998         case BPF_FUNC_get_socket_uid:
5999                 return &bpf_get_socket_uid_proto;
6000         case BPF_FUNC_perf_event_output:
6001                 return &bpf_skb_event_output_proto;
6002         default:
6003                 return bpf_base_func_proto(func_id);
6004         }
6005 }
6006
6007 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6008 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6009
6010 static const struct bpf_func_proto *
6011 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6012 {
6013         switch (func_id) {
6014         case BPF_FUNC_get_local_storage:
6015                 return &bpf_get_local_storage_proto;
6016         case BPF_FUNC_sk_fullsock:
6017                 return &bpf_sk_fullsock_proto;
6018         case BPF_FUNC_sk_storage_get:
6019                 return &bpf_sk_storage_get_proto;
6020         case BPF_FUNC_sk_storage_delete:
6021                 return &bpf_sk_storage_delete_proto;
6022         case BPF_FUNC_perf_event_output:
6023                 return &bpf_skb_event_output_proto;
6024 #ifdef CONFIG_SOCK_CGROUP_DATA
6025         case BPF_FUNC_skb_cgroup_id:
6026                 return &bpf_skb_cgroup_id_proto;
6027 #endif
6028 #ifdef CONFIG_INET
6029         case BPF_FUNC_tcp_sock:
6030                 return &bpf_tcp_sock_proto;
6031         case BPF_FUNC_get_listener_sock:
6032                 return &bpf_get_listener_sock_proto;
6033         case BPF_FUNC_skb_ecn_set_ce:
6034                 return &bpf_skb_ecn_set_ce_proto;
6035 #endif
6036         default:
6037                 return sk_filter_func_proto(func_id, prog);
6038         }
6039 }
6040
6041 static const struct bpf_func_proto *
6042 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6043 {
6044         switch (func_id) {
6045         case BPF_FUNC_skb_store_bytes:
6046                 return &bpf_skb_store_bytes_proto;
6047         case BPF_FUNC_skb_load_bytes:
6048                 return &bpf_skb_load_bytes_proto;
6049         case BPF_FUNC_skb_load_bytes_relative:
6050                 return &bpf_skb_load_bytes_relative_proto;
6051         case BPF_FUNC_skb_pull_data:
6052                 return &bpf_skb_pull_data_proto;
6053         case BPF_FUNC_csum_diff:
6054                 return &bpf_csum_diff_proto;
6055         case BPF_FUNC_csum_update:
6056                 return &bpf_csum_update_proto;
6057         case BPF_FUNC_l3_csum_replace:
6058                 return &bpf_l3_csum_replace_proto;
6059         case BPF_FUNC_l4_csum_replace:
6060                 return &bpf_l4_csum_replace_proto;
6061         case BPF_FUNC_clone_redirect:
6062                 return &bpf_clone_redirect_proto;
6063         case BPF_FUNC_get_cgroup_classid:
6064                 return &bpf_get_cgroup_classid_proto;
6065         case BPF_FUNC_skb_vlan_push:
6066                 return &bpf_skb_vlan_push_proto;
6067         case BPF_FUNC_skb_vlan_pop:
6068                 return &bpf_skb_vlan_pop_proto;
6069         case BPF_FUNC_skb_change_proto:
6070                 return &bpf_skb_change_proto_proto;
6071         case BPF_FUNC_skb_change_type:
6072                 return &bpf_skb_change_type_proto;
6073         case BPF_FUNC_skb_adjust_room:
6074                 return &bpf_skb_adjust_room_proto;
6075         case BPF_FUNC_skb_change_tail:
6076                 return &bpf_skb_change_tail_proto;
6077         case BPF_FUNC_skb_get_tunnel_key:
6078                 return &bpf_skb_get_tunnel_key_proto;
6079         case BPF_FUNC_skb_set_tunnel_key:
6080                 return bpf_get_skb_set_tunnel_proto(func_id);
6081         case BPF_FUNC_skb_get_tunnel_opt:
6082                 return &bpf_skb_get_tunnel_opt_proto;
6083         case BPF_FUNC_skb_set_tunnel_opt:
6084                 return bpf_get_skb_set_tunnel_proto(func_id);
6085         case BPF_FUNC_redirect:
6086                 return &bpf_redirect_proto;
6087         case BPF_FUNC_get_route_realm:
6088                 return &bpf_get_route_realm_proto;
6089         case BPF_FUNC_get_hash_recalc:
6090                 return &bpf_get_hash_recalc_proto;
6091         case BPF_FUNC_set_hash_invalid:
6092                 return &bpf_set_hash_invalid_proto;
6093         case BPF_FUNC_set_hash:
6094                 return &bpf_set_hash_proto;
6095         case BPF_FUNC_perf_event_output:
6096                 return &bpf_skb_event_output_proto;
6097         case BPF_FUNC_get_smp_processor_id:
6098                 return &bpf_get_smp_processor_id_proto;
6099         case BPF_FUNC_skb_under_cgroup:
6100                 return &bpf_skb_under_cgroup_proto;
6101         case BPF_FUNC_get_socket_cookie:
6102                 return &bpf_get_socket_cookie_proto;
6103         case BPF_FUNC_get_socket_uid:
6104                 return &bpf_get_socket_uid_proto;
6105         case BPF_FUNC_fib_lookup:
6106                 return &bpf_skb_fib_lookup_proto;
6107         case BPF_FUNC_sk_fullsock:
6108                 return &bpf_sk_fullsock_proto;
6109         case BPF_FUNC_sk_storage_get:
6110                 return &bpf_sk_storage_get_proto;
6111         case BPF_FUNC_sk_storage_delete:
6112                 return &bpf_sk_storage_delete_proto;
6113 #ifdef CONFIG_XFRM
6114         case BPF_FUNC_skb_get_xfrm_state:
6115                 return &bpf_skb_get_xfrm_state_proto;
6116 #endif
6117 #ifdef CONFIG_SOCK_CGROUP_DATA
6118         case BPF_FUNC_skb_cgroup_id:
6119                 return &bpf_skb_cgroup_id_proto;
6120         case BPF_FUNC_skb_ancestor_cgroup_id:
6121                 return &bpf_skb_ancestor_cgroup_id_proto;
6122 #endif
6123 #ifdef CONFIG_INET
6124         case BPF_FUNC_sk_lookup_tcp:
6125                 return &bpf_sk_lookup_tcp_proto;
6126         case BPF_FUNC_sk_lookup_udp:
6127                 return &bpf_sk_lookup_udp_proto;
6128         case BPF_FUNC_sk_release:
6129                 return &bpf_sk_release_proto;
6130         case BPF_FUNC_tcp_sock:
6131                 return &bpf_tcp_sock_proto;
6132         case BPF_FUNC_get_listener_sock:
6133                 return &bpf_get_listener_sock_proto;
6134         case BPF_FUNC_skc_lookup_tcp:
6135                 return &bpf_skc_lookup_tcp_proto;
6136         case BPF_FUNC_tcp_check_syncookie:
6137                 return &bpf_tcp_check_syncookie_proto;
6138         case BPF_FUNC_skb_ecn_set_ce:
6139                 return &bpf_skb_ecn_set_ce_proto;
6140         case BPF_FUNC_tcp_gen_syncookie:
6141                 return &bpf_tcp_gen_syncookie_proto;
6142 #endif
6143         default:
6144                 return bpf_base_func_proto(func_id);
6145         }
6146 }
6147
6148 static const struct bpf_func_proto *
6149 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6150 {
6151         switch (func_id) {
6152         case BPF_FUNC_perf_event_output:
6153                 return &bpf_xdp_event_output_proto;
6154         case BPF_FUNC_get_smp_processor_id:
6155                 return &bpf_get_smp_processor_id_proto;
6156         case BPF_FUNC_csum_diff:
6157                 return &bpf_csum_diff_proto;
6158         case BPF_FUNC_xdp_adjust_head:
6159                 return &bpf_xdp_adjust_head_proto;
6160         case BPF_FUNC_xdp_adjust_meta:
6161                 return &bpf_xdp_adjust_meta_proto;
6162         case BPF_FUNC_redirect:
6163                 return &bpf_xdp_redirect_proto;
6164         case BPF_FUNC_redirect_map:
6165                 return &bpf_xdp_redirect_map_proto;
6166         case BPF_FUNC_xdp_adjust_tail:
6167                 return &bpf_xdp_adjust_tail_proto;
6168         case BPF_FUNC_fib_lookup:
6169                 return &bpf_xdp_fib_lookup_proto;
6170 #ifdef CONFIG_INET
6171         case BPF_FUNC_sk_lookup_udp:
6172                 return &bpf_xdp_sk_lookup_udp_proto;
6173         case BPF_FUNC_sk_lookup_tcp:
6174                 return &bpf_xdp_sk_lookup_tcp_proto;
6175         case BPF_FUNC_sk_release:
6176                 return &bpf_sk_release_proto;
6177         case BPF_FUNC_skc_lookup_tcp:
6178                 return &bpf_xdp_skc_lookup_tcp_proto;
6179         case BPF_FUNC_tcp_check_syncookie:
6180                 return &bpf_tcp_check_syncookie_proto;
6181         case BPF_FUNC_tcp_gen_syncookie:
6182                 return &bpf_tcp_gen_syncookie_proto;
6183 #endif
6184         default:
6185                 return bpf_base_func_proto(func_id);
6186         }
6187 }
6188
6189 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
6190 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
6191
6192 static const struct bpf_func_proto *
6193 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6194 {
6195         switch (func_id) {
6196         case BPF_FUNC_setsockopt:
6197                 return &bpf_setsockopt_proto;
6198         case BPF_FUNC_getsockopt:
6199                 return &bpf_getsockopt_proto;
6200         case BPF_FUNC_sock_ops_cb_flags_set:
6201                 return &bpf_sock_ops_cb_flags_set_proto;
6202         case BPF_FUNC_sock_map_update:
6203                 return &bpf_sock_map_update_proto;
6204         case BPF_FUNC_sock_hash_update:
6205                 return &bpf_sock_hash_update_proto;
6206         case BPF_FUNC_get_socket_cookie:
6207                 return &bpf_get_socket_cookie_sock_ops_proto;
6208         case BPF_FUNC_get_local_storage:
6209                 return &bpf_get_local_storage_proto;
6210         case BPF_FUNC_perf_event_output:
6211                 return &bpf_sockopt_event_output_proto;
6212         case BPF_FUNC_sk_storage_get:
6213                 return &bpf_sk_storage_get_proto;
6214         case BPF_FUNC_sk_storage_delete:
6215                 return &bpf_sk_storage_delete_proto;
6216 #ifdef CONFIG_INET
6217         case BPF_FUNC_tcp_sock:
6218                 return &bpf_tcp_sock_proto;
6219 #endif /* CONFIG_INET */
6220         default:
6221                 return bpf_base_func_proto(func_id);
6222         }
6223 }
6224
6225 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
6226 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
6227
6228 static const struct bpf_func_proto *
6229 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6230 {
6231         switch (func_id) {
6232         case BPF_FUNC_msg_redirect_map:
6233                 return &bpf_msg_redirect_map_proto;
6234         case BPF_FUNC_msg_redirect_hash:
6235                 return &bpf_msg_redirect_hash_proto;
6236         case BPF_FUNC_msg_apply_bytes:
6237                 return &bpf_msg_apply_bytes_proto;
6238         case BPF_FUNC_msg_cork_bytes:
6239                 return &bpf_msg_cork_bytes_proto;
6240         case BPF_FUNC_msg_pull_data:
6241                 return &bpf_msg_pull_data_proto;
6242         case BPF_FUNC_msg_push_data:
6243                 return &bpf_msg_push_data_proto;
6244         case BPF_FUNC_msg_pop_data:
6245                 return &bpf_msg_pop_data_proto;
6246         default:
6247                 return bpf_base_func_proto(func_id);
6248         }
6249 }
6250
6251 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
6252 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
6253
6254 static const struct bpf_func_proto *
6255 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6256 {
6257         switch (func_id) {
6258         case BPF_FUNC_skb_store_bytes:
6259                 return &bpf_skb_store_bytes_proto;
6260         case BPF_FUNC_skb_load_bytes:
6261                 return &bpf_skb_load_bytes_proto;
6262         case BPF_FUNC_skb_pull_data:
6263                 return &sk_skb_pull_data_proto;
6264         case BPF_FUNC_skb_change_tail:
6265                 return &sk_skb_change_tail_proto;
6266         case BPF_FUNC_skb_change_head:
6267                 return &sk_skb_change_head_proto;
6268         case BPF_FUNC_get_socket_cookie:
6269                 return &bpf_get_socket_cookie_proto;
6270         case BPF_FUNC_get_socket_uid:
6271                 return &bpf_get_socket_uid_proto;
6272         case BPF_FUNC_sk_redirect_map:
6273                 return &bpf_sk_redirect_map_proto;
6274         case BPF_FUNC_sk_redirect_hash:
6275                 return &bpf_sk_redirect_hash_proto;
6276         case BPF_FUNC_perf_event_output:
6277                 return &bpf_skb_event_output_proto;
6278 #ifdef CONFIG_INET
6279         case BPF_FUNC_sk_lookup_tcp:
6280                 return &bpf_sk_lookup_tcp_proto;
6281         case BPF_FUNC_sk_lookup_udp:
6282                 return &bpf_sk_lookup_udp_proto;
6283         case BPF_FUNC_sk_release:
6284                 return &bpf_sk_release_proto;
6285         case BPF_FUNC_skc_lookup_tcp:
6286                 return &bpf_skc_lookup_tcp_proto;
6287 #endif
6288         default:
6289                 return bpf_base_func_proto(func_id);
6290         }
6291 }
6292
6293 static const struct bpf_func_proto *
6294 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6295 {
6296         switch (func_id) {
6297         case BPF_FUNC_skb_load_bytes:
6298                 return &bpf_flow_dissector_load_bytes_proto;
6299         default:
6300                 return bpf_base_func_proto(func_id);
6301         }
6302 }
6303
6304 static const struct bpf_func_proto *
6305 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6306 {
6307         switch (func_id) {
6308         case BPF_FUNC_skb_load_bytes:
6309                 return &bpf_skb_load_bytes_proto;
6310         case BPF_FUNC_skb_pull_data:
6311                 return &bpf_skb_pull_data_proto;
6312         case BPF_FUNC_csum_diff:
6313                 return &bpf_csum_diff_proto;
6314         case BPF_FUNC_get_cgroup_classid:
6315                 return &bpf_get_cgroup_classid_proto;
6316         case BPF_FUNC_get_route_realm:
6317                 return &bpf_get_route_realm_proto;
6318         case BPF_FUNC_get_hash_recalc:
6319                 return &bpf_get_hash_recalc_proto;
6320         case BPF_FUNC_perf_event_output:
6321                 return &bpf_skb_event_output_proto;
6322         case BPF_FUNC_get_smp_processor_id:
6323                 return &bpf_get_smp_processor_id_proto;
6324         case BPF_FUNC_skb_under_cgroup:
6325                 return &bpf_skb_under_cgroup_proto;
6326         default:
6327                 return bpf_base_func_proto(func_id);
6328         }
6329 }
6330
6331 static const struct bpf_func_proto *
6332 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6333 {
6334         switch (func_id) {
6335         case BPF_FUNC_lwt_push_encap:
6336                 return &bpf_lwt_in_push_encap_proto;
6337         default:
6338                 return lwt_out_func_proto(func_id, prog);
6339         }
6340 }
6341
6342 static const struct bpf_func_proto *
6343 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6344 {
6345         switch (func_id) {
6346         case BPF_FUNC_skb_get_tunnel_key:
6347                 return &bpf_skb_get_tunnel_key_proto;
6348         case BPF_FUNC_skb_set_tunnel_key:
6349                 return bpf_get_skb_set_tunnel_proto(func_id);
6350         case BPF_FUNC_skb_get_tunnel_opt:
6351                 return &bpf_skb_get_tunnel_opt_proto;
6352         case BPF_FUNC_skb_set_tunnel_opt:
6353                 return bpf_get_skb_set_tunnel_proto(func_id);
6354         case BPF_FUNC_redirect:
6355                 return &bpf_redirect_proto;
6356         case BPF_FUNC_clone_redirect:
6357                 return &bpf_clone_redirect_proto;
6358         case BPF_FUNC_skb_change_tail:
6359                 return &bpf_skb_change_tail_proto;
6360         case BPF_FUNC_skb_change_head:
6361                 return &bpf_skb_change_head_proto;
6362         case BPF_FUNC_skb_store_bytes:
6363                 return &bpf_skb_store_bytes_proto;
6364         case BPF_FUNC_csum_update:
6365                 return &bpf_csum_update_proto;
6366         case BPF_FUNC_l3_csum_replace:
6367                 return &bpf_l3_csum_replace_proto;
6368         case BPF_FUNC_l4_csum_replace:
6369                 return &bpf_l4_csum_replace_proto;
6370         case BPF_FUNC_set_hash_invalid:
6371                 return &bpf_set_hash_invalid_proto;
6372         case BPF_FUNC_lwt_push_encap:
6373                 return &bpf_lwt_xmit_push_encap_proto;
6374         default:
6375                 return lwt_out_func_proto(func_id, prog);
6376         }
6377 }
6378
6379 static const struct bpf_func_proto *
6380 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6381 {
6382         switch (func_id) {
6383 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6384         case BPF_FUNC_lwt_seg6_store_bytes:
6385                 return &bpf_lwt_seg6_store_bytes_proto;
6386         case BPF_FUNC_lwt_seg6_action:
6387                 return &bpf_lwt_seg6_action_proto;
6388         case BPF_FUNC_lwt_seg6_adjust_srh:
6389                 return &bpf_lwt_seg6_adjust_srh_proto;
6390 #endif
6391         default:
6392                 return lwt_out_func_proto(func_id, prog);
6393         }
6394 }
6395
6396 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
6397                                     const struct bpf_prog *prog,
6398                                     struct bpf_insn_access_aux *info)
6399 {
6400         const int size_default = sizeof(__u32);
6401
6402         if (off < 0 || off >= sizeof(struct __sk_buff))
6403                 return false;
6404
6405         /* The verifier guarantees that size > 0. */
6406         if (off % size != 0)
6407                 return false;
6408
6409         switch (off) {
6410         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6411                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
6412                         return false;
6413                 break;
6414         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
6415         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
6416         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
6417         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
6418         case bpf_ctx_range(struct __sk_buff, data):
6419         case bpf_ctx_range(struct __sk_buff, data_meta):
6420         case bpf_ctx_range(struct __sk_buff, data_end):
6421                 if (size != size_default)
6422                         return false;
6423                 break;
6424         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6425                 return false;
6426         case bpf_ctx_range(struct __sk_buff, tstamp):
6427                 if (size != sizeof(__u64))
6428                         return false;
6429                 break;
6430         case offsetof(struct __sk_buff, sk):
6431                 if (type == BPF_WRITE || size != sizeof(__u64))
6432                         return false;
6433                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
6434                 break;
6435         default:
6436                 /* Only narrow read access allowed for now. */
6437                 if (type == BPF_WRITE) {
6438                         if (size != size_default)
6439                                 return false;
6440                 } else {
6441                         bpf_ctx_record_field_size(info, size_default);
6442                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6443                                 return false;
6444                 }
6445         }
6446
6447         return true;
6448 }
6449
6450 static bool sk_filter_is_valid_access(int off, int size,
6451                                       enum bpf_access_type type,
6452                                       const struct bpf_prog *prog,
6453                                       struct bpf_insn_access_aux *info)
6454 {
6455         switch (off) {
6456         case bpf_ctx_range(struct __sk_buff, tc_classid):
6457         case bpf_ctx_range(struct __sk_buff, data):
6458         case bpf_ctx_range(struct __sk_buff, data_meta):
6459         case bpf_ctx_range(struct __sk_buff, data_end):
6460         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6461         case bpf_ctx_range(struct __sk_buff, tstamp):
6462         case bpf_ctx_range(struct __sk_buff, wire_len):
6463                 return false;
6464         }
6465
6466         if (type == BPF_WRITE) {
6467                 switch (off) {
6468                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6469                         break;
6470                 default:
6471                         return false;
6472                 }
6473         }
6474
6475         return bpf_skb_is_valid_access(off, size, type, prog, info);
6476 }
6477
6478 static bool cg_skb_is_valid_access(int off, int size,
6479                                    enum bpf_access_type type,
6480                                    const struct bpf_prog *prog,
6481                                    struct bpf_insn_access_aux *info)
6482 {
6483         switch (off) {
6484         case bpf_ctx_range(struct __sk_buff, tc_classid):
6485         case bpf_ctx_range(struct __sk_buff, data_meta):
6486         case bpf_ctx_range(struct __sk_buff, wire_len):
6487                 return false;
6488         case bpf_ctx_range(struct __sk_buff, data):
6489         case bpf_ctx_range(struct __sk_buff, data_end):
6490                 if (!capable(CAP_SYS_ADMIN))
6491                         return false;
6492                 break;
6493         }
6494
6495         if (type == BPF_WRITE) {
6496                 switch (off) {
6497                 case bpf_ctx_range(struct __sk_buff, mark):
6498                 case bpf_ctx_range(struct __sk_buff, priority):
6499                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6500                         break;
6501                 case bpf_ctx_range(struct __sk_buff, tstamp):
6502                         if (!capable(CAP_SYS_ADMIN))
6503                                 return false;
6504                         break;
6505                 default:
6506                         return false;
6507                 }
6508         }
6509
6510         switch (off) {
6511         case bpf_ctx_range(struct __sk_buff, data):
6512                 info->reg_type = PTR_TO_PACKET;
6513                 break;
6514         case bpf_ctx_range(struct __sk_buff, data_end):
6515                 info->reg_type = PTR_TO_PACKET_END;
6516                 break;
6517         }
6518
6519         return bpf_skb_is_valid_access(off, size, type, prog, info);
6520 }
6521
6522 static bool lwt_is_valid_access(int off, int size,
6523                                 enum bpf_access_type type,
6524                                 const struct bpf_prog *prog,
6525                                 struct bpf_insn_access_aux *info)
6526 {
6527         switch (off) {
6528         case bpf_ctx_range(struct __sk_buff, tc_classid):
6529         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6530         case bpf_ctx_range(struct __sk_buff, data_meta):
6531         case bpf_ctx_range(struct __sk_buff, tstamp):
6532         case bpf_ctx_range(struct __sk_buff, wire_len):
6533                 return false;
6534         }
6535
6536         if (type == BPF_WRITE) {
6537                 switch (off) {
6538                 case bpf_ctx_range(struct __sk_buff, mark):
6539                 case bpf_ctx_range(struct __sk_buff, priority):
6540                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6541                         break;
6542                 default:
6543                         return false;
6544                 }
6545         }
6546
6547         switch (off) {
6548         case bpf_ctx_range(struct __sk_buff, data):
6549                 info->reg_type = PTR_TO_PACKET;
6550                 break;
6551         case bpf_ctx_range(struct __sk_buff, data_end):
6552                 info->reg_type = PTR_TO_PACKET_END;
6553                 break;
6554         }
6555
6556         return bpf_skb_is_valid_access(off, size, type, prog, info);
6557 }
6558
6559 /* Attach type specific accesses */
6560 static bool __sock_filter_check_attach_type(int off,
6561                                             enum bpf_access_type access_type,
6562                                             enum bpf_attach_type attach_type)
6563 {
6564         switch (off) {
6565         case offsetof(struct bpf_sock, bound_dev_if):
6566         case offsetof(struct bpf_sock, mark):
6567         case offsetof(struct bpf_sock, priority):
6568                 switch (attach_type) {
6569                 case BPF_CGROUP_INET_SOCK_CREATE:
6570                         goto full_access;
6571                 default:
6572                         return false;
6573                 }
6574         case bpf_ctx_range(struct bpf_sock, src_ip4):
6575                 switch (attach_type) {
6576                 case BPF_CGROUP_INET4_POST_BIND:
6577                         goto read_only;
6578                 default:
6579                         return false;
6580                 }
6581         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6582                 switch (attach_type) {
6583                 case BPF_CGROUP_INET6_POST_BIND:
6584                         goto read_only;
6585                 default:
6586                         return false;
6587                 }
6588         case bpf_ctx_range(struct bpf_sock, src_port):
6589                 switch (attach_type) {
6590                 case BPF_CGROUP_INET4_POST_BIND:
6591                 case BPF_CGROUP_INET6_POST_BIND:
6592                         goto read_only;
6593                 default:
6594                         return false;
6595                 }
6596         }
6597 read_only:
6598         return access_type == BPF_READ;
6599 full_access:
6600         return true;
6601 }
6602
6603 bool bpf_sock_common_is_valid_access(int off, int size,
6604                                      enum bpf_access_type type,
6605                                      struct bpf_insn_access_aux *info)
6606 {
6607         switch (off) {
6608         case bpf_ctx_range_till(struct bpf_sock, type, priority):
6609                 return false;
6610         default:
6611                 return bpf_sock_is_valid_access(off, size, type, info);
6612         }
6613 }
6614
6615 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6616                               struct bpf_insn_access_aux *info)
6617 {
6618         const int size_default = sizeof(__u32);
6619
6620         if (off < 0 || off >= sizeof(struct bpf_sock))
6621                 return false;
6622         if (off % size != 0)
6623                 return false;
6624
6625         switch (off) {
6626         case offsetof(struct bpf_sock, state):
6627         case offsetof(struct bpf_sock, family):
6628         case offsetof(struct bpf_sock, type):
6629         case offsetof(struct bpf_sock, protocol):
6630         case offsetof(struct bpf_sock, dst_port):
6631         case offsetof(struct bpf_sock, src_port):
6632         case bpf_ctx_range(struct bpf_sock, src_ip4):
6633         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6634         case bpf_ctx_range(struct bpf_sock, dst_ip4):
6635         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6636                 bpf_ctx_record_field_size(info, size_default);
6637                 return bpf_ctx_narrow_access_ok(off, size, size_default);
6638         }
6639
6640         return size == size_default;
6641 }
6642
6643 static bool sock_filter_is_valid_access(int off, int size,
6644                                         enum bpf_access_type type,
6645                                         const struct bpf_prog *prog,
6646                                         struct bpf_insn_access_aux *info)
6647 {
6648         if (!bpf_sock_is_valid_access(off, size, type, info))
6649                 return false;
6650         return __sock_filter_check_attach_type(off, type,
6651                                                prog->expected_attach_type);
6652 }
6653
6654 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6655                              const struct bpf_prog *prog)
6656 {
6657         /* Neither direct read nor direct write requires any preliminary
6658          * action.
6659          */
6660         return 0;
6661 }
6662
6663 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6664                                 const struct bpf_prog *prog, int drop_verdict)
6665 {
6666         struct bpf_insn *insn = insn_buf;
6667
6668         if (!direct_write)
6669                 return 0;
6670
6671         /* if (!skb->cloned)
6672          *       goto start;
6673          *
6674          * (Fast-path, otherwise approximation that we might be
6675          *  a clone, do the rest in helper.)
6676          */
6677         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6678         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6679         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6680
6681         /* ret = bpf_skb_pull_data(skb, 0); */
6682         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6683         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6684         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6685                                BPF_FUNC_skb_pull_data);
6686         /* if (!ret)
6687          *      goto restore;
6688          * return TC_ACT_SHOT;
6689          */
6690         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6691         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6692         *insn++ = BPF_EXIT_INSN();
6693
6694         /* restore: */
6695         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6696         /* start: */
6697         *insn++ = prog->insnsi[0];
6698
6699         return insn - insn_buf;
6700 }
6701
6702 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6703                           struct bpf_insn *insn_buf)
6704 {
6705         bool indirect = BPF_MODE(orig->code) == BPF_IND;
6706         struct bpf_insn *insn = insn_buf;
6707
6708         /* We're guaranteed here that CTX is in R6. */
6709         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6710         if (!indirect) {
6711                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6712         } else {
6713                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6714                 if (orig->imm)
6715                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6716         }
6717
6718         switch (BPF_SIZE(orig->code)) {
6719         case BPF_B:
6720                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6721                 break;
6722         case BPF_H:
6723                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6724                 break;
6725         case BPF_W:
6726                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6727                 break;
6728         }
6729
6730         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6731         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6732         *insn++ = BPF_EXIT_INSN();
6733
6734         return insn - insn_buf;
6735 }
6736
6737 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6738                                const struct bpf_prog *prog)
6739 {
6740         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6741 }
6742
6743 static bool tc_cls_act_is_valid_access(int off, int size,
6744                                        enum bpf_access_type type,
6745                                        const struct bpf_prog *prog,
6746                                        struct bpf_insn_access_aux *info)
6747 {
6748         if (type == BPF_WRITE) {
6749                 switch (off) {
6750                 case bpf_ctx_range(struct __sk_buff, mark):
6751                 case bpf_ctx_range(struct __sk_buff, tc_index):
6752                 case bpf_ctx_range(struct __sk_buff, priority):
6753                 case bpf_ctx_range(struct __sk_buff, tc_classid):
6754                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6755                 case bpf_ctx_range(struct __sk_buff, tstamp):
6756                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6757                         break;
6758                 default:
6759                         return false;
6760                 }
6761         }
6762
6763         switch (off) {
6764         case bpf_ctx_range(struct __sk_buff, data):
6765                 info->reg_type = PTR_TO_PACKET;
6766                 break;
6767         case bpf_ctx_range(struct __sk_buff, data_meta):
6768                 info->reg_type = PTR_TO_PACKET_META;
6769                 break;
6770         case bpf_ctx_range(struct __sk_buff, data_end):
6771                 info->reg_type = PTR_TO_PACKET_END;
6772                 break;
6773         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6774                 return false;
6775         }
6776
6777         return bpf_skb_is_valid_access(off, size, type, prog, info);
6778 }
6779
6780 static bool __is_valid_xdp_access(int off, int size)
6781 {
6782         if (off < 0 || off >= sizeof(struct xdp_md))
6783                 return false;
6784         if (off % size != 0)
6785                 return false;
6786         if (size != sizeof(__u32))
6787                 return false;
6788
6789         return true;
6790 }
6791
6792 static bool xdp_is_valid_access(int off, int size,
6793                                 enum bpf_access_type type,
6794                                 const struct bpf_prog *prog,
6795                                 struct bpf_insn_access_aux *info)
6796 {
6797         if (type == BPF_WRITE) {
6798                 if (bpf_prog_is_dev_bound(prog->aux)) {
6799                         switch (off) {
6800                         case offsetof(struct xdp_md, rx_queue_index):
6801                                 return __is_valid_xdp_access(off, size);
6802                         }
6803                 }
6804                 return false;
6805         }
6806
6807         switch (off) {
6808         case offsetof(struct xdp_md, data):
6809                 info->reg_type = PTR_TO_PACKET;
6810                 break;
6811         case offsetof(struct xdp_md, data_meta):
6812                 info->reg_type = PTR_TO_PACKET_META;
6813                 break;
6814         case offsetof(struct xdp_md, data_end):
6815                 info->reg_type = PTR_TO_PACKET_END;
6816                 break;
6817         }
6818
6819         return __is_valid_xdp_access(off, size);
6820 }
6821
6822 void bpf_warn_invalid_xdp_action(u32 act)
6823 {
6824         const u32 act_max = XDP_REDIRECT;
6825
6826         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6827                   act > act_max ? "Illegal" : "Driver unsupported",
6828                   act);
6829 }
6830 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6831
6832 static bool sock_addr_is_valid_access(int off, int size,
6833                                       enum bpf_access_type type,
6834                                       const struct bpf_prog *prog,
6835                                       struct bpf_insn_access_aux *info)
6836 {
6837         const int size_default = sizeof(__u32);
6838
6839         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6840                 return false;
6841         if (off % size != 0)
6842                 return false;
6843
6844         /* Disallow access to IPv6 fields from IPv4 contex and vise
6845          * versa.
6846          */
6847         switch (off) {
6848         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6849                 switch (prog->expected_attach_type) {
6850                 case BPF_CGROUP_INET4_BIND:
6851                 case BPF_CGROUP_INET4_CONNECT:
6852                 case BPF_CGROUP_UDP4_SENDMSG:
6853                 case BPF_CGROUP_UDP4_RECVMSG:
6854                         break;
6855                 default:
6856                         return false;
6857                 }
6858                 break;
6859         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6860                 switch (prog->expected_attach_type) {
6861                 case BPF_CGROUP_INET6_BIND:
6862                 case BPF_CGROUP_INET6_CONNECT:
6863                 case BPF_CGROUP_UDP6_SENDMSG:
6864                 case BPF_CGROUP_UDP6_RECVMSG:
6865                         break;
6866                 default:
6867                         return false;
6868                 }
6869                 break;
6870         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6871                 switch (prog->expected_attach_type) {
6872                 case BPF_CGROUP_UDP4_SENDMSG:
6873                         break;
6874                 default:
6875                         return false;
6876                 }
6877                 break;
6878         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6879                                 msg_src_ip6[3]):
6880                 switch (prog->expected_attach_type) {
6881                 case BPF_CGROUP_UDP6_SENDMSG:
6882                         break;
6883                 default:
6884                         return false;
6885                 }
6886                 break;
6887         }
6888
6889         switch (off) {
6890         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6891         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6892         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6893         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6894                                 msg_src_ip6[3]):
6895                 if (type == BPF_READ) {
6896                         bpf_ctx_record_field_size(info, size_default);
6897
6898                         if (bpf_ctx_wide_access_ok(off, size,
6899                                                    struct bpf_sock_addr,
6900                                                    user_ip6))
6901                                 return true;
6902
6903                         if (bpf_ctx_wide_access_ok(off, size,
6904                                                    struct bpf_sock_addr,
6905                                                    msg_src_ip6))
6906                                 return true;
6907
6908                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6909                                 return false;
6910                 } else {
6911                         if (bpf_ctx_wide_access_ok(off, size,
6912                                                    struct bpf_sock_addr,
6913                                                    user_ip6))
6914                                 return true;
6915
6916                         if (bpf_ctx_wide_access_ok(off, size,
6917                                                    struct bpf_sock_addr,
6918                                                    msg_src_ip6))
6919                                 return true;
6920
6921                         if (size != size_default)
6922                                 return false;
6923                 }
6924                 break;
6925         case bpf_ctx_range(struct bpf_sock_addr, user_port):
6926                 if (size != size_default)
6927                         return false;
6928                 break;
6929         case offsetof(struct bpf_sock_addr, sk):
6930                 if (type != BPF_READ)
6931                         return false;
6932                 if (size != sizeof(__u64))
6933                         return false;
6934                 info->reg_type = PTR_TO_SOCKET;
6935                 break;
6936         default:
6937                 if (type == BPF_READ) {
6938                         if (size != size_default)
6939                                 return false;
6940                 } else {
6941                         return false;
6942                 }
6943         }
6944
6945         return true;
6946 }
6947
6948 static bool sock_ops_is_valid_access(int off, int size,
6949                                      enum bpf_access_type type,
6950                                      const struct bpf_prog *prog,
6951                                      struct bpf_insn_access_aux *info)
6952 {
6953         const int size_default = sizeof(__u32);
6954
6955         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
6956                 return false;
6957
6958         /* The verifier guarantees that size > 0. */
6959         if (off % size != 0)
6960                 return false;
6961
6962         if (type == BPF_WRITE) {
6963                 switch (off) {
6964                 case offsetof(struct bpf_sock_ops, reply):
6965                 case offsetof(struct bpf_sock_ops, sk_txhash):
6966                         if (size != size_default)
6967                                 return false;
6968                         break;
6969                 default:
6970                         return false;
6971                 }
6972         } else {
6973                 switch (off) {
6974                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
6975                                         bytes_acked):
6976                         if (size != sizeof(__u64))
6977                                 return false;
6978                         break;
6979                 case offsetof(struct bpf_sock_ops, sk):
6980                         if (size != sizeof(__u64))
6981                                 return false;
6982                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
6983                         break;
6984                 default:
6985                         if (size != size_default)
6986                                 return false;
6987                         break;
6988                 }
6989         }
6990
6991         return true;
6992 }
6993
6994 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
6995                            const struct bpf_prog *prog)
6996 {
6997         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
6998 }
6999
7000 static bool sk_skb_is_valid_access(int off, int size,
7001                                    enum bpf_access_type type,
7002                                    const struct bpf_prog *prog,
7003                                    struct bpf_insn_access_aux *info)
7004 {
7005         switch (off) {
7006         case bpf_ctx_range(struct __sk_buff, tc_classid):
7007         case bpf_ctx_range(struct __sk_buff, data_meta):
7008         case bpf_ctx_range(struct __sk_buff, tstamp):
7009         case bpf_ctx_range(struct __sk_buff, wire_len):
7010                 return false;
7011         }
7012
7013         if (type == BPF_WRITE) {
7014                 switch (off) {
7015                 case bpf_ctx_range(struct __sk_buff, tc_index):
7016                 case bpf_ctx_range(struct __sk_buff, priority):
7017                         break;
7018                 default:
7019                         return false;
7020                 }
7021         }
7022
7023         switch (off) {
7024         case bpf_ctx_range(struct __sk_buff, mark):
7025                 return false;
7026         case bpf_ctx_range(struct __sk_buff, data):
7027                 info->reg_type = PTR_TO_PACKET;
7028                 break;
7029         case bpf_ctx_range(struct __sk_buff, data_end):
7030                 info->reg_type = PTR_TO_PACKET_END;
7031                 break;
7032         }
7033
7034         return bpf_skb_is_valid_access(off, size, type, prog, info);
7035 }
7036
7037 static bool sk_msg_is_valid_access(int off, int size,
7038                                    enum bpf_access_type type,
7039                                    const struct bpf_prog *prog,
7040                                    struct bpf_insn_access_aux *info)
7041 {
7042         if (type == BPF_WRITE)
7043                 return false;
7044
7045         if (off % size != 0)
7046                 return false;
7047
7048         switch (off) {
7049         case offsetof(struct sk_msg_md, data):
7050                 info->reg_type = PTR_TO_PACKET;
7051                 if (size != sizeof(__u64))
7052                         return false;
7053                 break;
7054         case offsetof(struct sk_msg_md, data_end):
7055                 info->reg_type = PTR_TO_PACKET_END;
7056                 if (size != sizeof(__u64))
7057                         return false;
7058                 break;
7059         case bpf_ctx_range(struct sk_msg_md, family):
7060         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
7061         case bpf_ctx_range(struct sk_msg_md, local_ip4):
7062         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
7063         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
7064         case bpf_ctx_range(struct sk_msg_md, remote_port):
7065         case bpf_ctx_range(struct sk_msg_md, local_port):
7066         case bpf_ctx_range(struct sk_msg_md, size):
7067                 if (size != sizeof(__u32))
7068                         return false;
7069                 break;
7070         default:
7071                 return false;
7072         }
7073         return true;
7074 }
7075
7076 static bool flow_dissector_is_valid_access(int off, int size,
7077                                            enum bpf_access_type type,
7078                                            const struct bpf_prog *prog,
7079                                            struct bpf_insn_access_aux *info)
7080 {
7081         const int size_default = sizeof(__u32);
7082
7083         if (off < 0 || off >= sizeof(struct __sk_buff))
7084                 return false;
7085
7086         if (type == BPF_WRITE)
7087                 return false;
7088
7089         switch (off) {
7090         case bpf_ctx_range(struct __sk_buff, data):
7091                 if (size != size_default)
7092                         return false;
7093                 info->reg_type = PTR_TO_PACKET;
7094                 return true;
7095         case bpf_ctx_range(struct __sk_buff, data_end):
7096                 if (size != size_default)
7097                         return false;
7098                 info->reg_type = PTR_TO_PACKET_END;
7099                 return true;
7100         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7101                 if (size != sizeof(__u64))
7102                         return false;
7103                 info->reg_type = PTR_TO_FLOW_KEYS;
7104                 return true;
7105         default:
7106                 return false;
7107         }
7108 }
7109
7110 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
7111                                              const struct bpf_insn *si,
7112                                              struct bpf_insn *insn_buf,
7113                                              struct bpf_prog *prog,
7114                                              u32 *target_size)
7115
7116 {
7117         struct bpf_insn *insn = insn_buf;
7118
7119         switch (si->off) {
7120         case offsetof(struct __sk_buff, data):
7121                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
7122                                       si->dst_reg, si->src_reg,
7123                                       offsetof(struct bpf_flow_dissector, data));
7124                 break;
7125
7126         case offsetof(struct __sk_buff, data_end):
7127                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
7128                                       si->dst_reg, si->src_reg,
7129                                       offsetof(struct bpf_flow_dissector, data_end));
7130                 break;
7131
7132         case offsetof(struct __sk_buff, flow_keys):
7133                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
7134                                       si->dst_reg, si->src_reg,
7135                                       offsetof(struct bpf_flow_dissector, flow_keys));
7136                 break;
7137         }
7138
7139         return insn - insn_buf;
7140 }
7141
7142 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
7143                                   const struct bpf_insn *si,
7144                                   struct bpf_insn *insn_buf,
7145                                   struct bpf_prog *prog, u32 *target_size)
7146 {
7147         struct bpf_insn *insn = insn_buf;
7148         int off;
7149
7150         switch (si->off) {
7151         case offsetof(struct __sk_buff, len):
7152                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7153                                       bpf_target_off(struct sk_buff, len, 4,
7154                                                      target_size));
7155                 break;
7156
7157         case offsetof(struct __sk_buff, protocol):
7158                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7159                                       bpf_target_off(struct sk_buff, protocol, 2,
7160                                                      target_size));
7161                 break;
7162
7163         case offsetof(struct __sk_buff, vlan_proto):
7164                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7165                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
7166                                                      target_size));
7167                 break;
7168
7169         case offsetof(struct __sk_buff, priority):
7170                 if (type == BPF_WRITE)
7171                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7172                                               bpf_target_off(struct sk_buff, priority, 4,
7173                                                              target_size));
7174                 else
7175                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7176                                               bpf_target_off(struct sk_buff, priority, 4,
7177                                                              target_size));
7178                 break;
7179
7180         case offsetof(struct __sk_buff, ingress_ifindex):
7181                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7182                                       bpf_target_off(struct sk_buff, skb_iif, 4,
7183                                                      target_size));
7184                 break;
7185
7186         case offsetof(struct __sk_buff, ifindex):
7187                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7188                                       si->dst_reg, si->src_reg,
7189                                       offsetof(struct sk_buff, dev));
7190                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
7191                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7192                                       bpf_target_off(struct net_device, ifindex, 4,
7193                                                      target_size));
7194                 break;
7195
7196         case offsetof(struct __sk_buff, hash):
7197                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7198                                       bpf_target_off(struct sk_buff, hash, 4,
7199                                                      target_size));
7200                 break;
7201
7202         case offsetof(struct __sk_buff, mark):
7203                 if (type == BPF_WRITE)
7204                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7205                                               bpf_target_off(struct sk_buff, mark, 4,
7206                                                              target_size));
7207                 else
7208                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7209                                               bpf_target_off(struct sk_buff, mark, 4,
7210                                                              target_size));
7211                 break;
7212
7213         case offsetof(struct __sk_buff, pkt_type):
7214                 *target_size = 1;
7215                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7216                                       PKT_TYPE_OFFSET());
7217                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
7218 #ifdef __BIG_ENDIAN_BITFIELD
7219                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
7220 #endif
7221                 break;
7222
7223         case offsetof(struct __sk_buff, queue_mapping):
7224                 if (type == BPF_WRITE) {
7225                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
7226                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7227                                               bpf_target_off(struct sk_buff,
7228                                                              queue_mapping,
7229                                                              2, target_size));
7230                 } else {
7231                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7232                                               bpf_target_off(struct sk_buff,
7233                                                              queue_mapping,
7234                                                              2, target_size));
7235                 }
7236                 break;
7237
7238         case offsetof(struct __sk_buff, vlan_present):
7239                 *target_size = 1;
7240                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7241                                       PKT_VLAN_PRESENT_OFFSET());
7242                 if (PKT_VLAN_PRESENT_BIT)
7243                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
7244                 if (PKT_VLAN_PRESENT_BIT < 7)
7245                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
7246                 break;
7247
7248         case offsetof(struct __sk_buff, vlan_tci):
7249                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7250                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
7251                                                      target_size));
7252                 break;
7253
7254         case offsetof(struct __sk_buff, cb[0]) ...
7255              offsetofend(struct __sk_buff, cb[4]) - 1:
7256                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
7257                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
7258                               offsetof(struct qdisc_skb_cb, data)) %
7259                              sizeof(__u64));
7260
7261                 prog->cb_access = 1;
7262                 off  = si->off;
7263                 off -= offsetof(struct __sk_buff, cb[0]);
7264                 off += offsetof(struct sk_buff, cb);
7265                 off += offsetof(struct qdisc_skb_cb, data);
7266                 if (type == BPF_WRITE)
7267                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
7268                                               si->src_reg, off);
7269                 else
7270                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
7271                                               si->src_reg, off);
7272                 break;
7273
7274         case offsetof(struct __sk_buff, tc_classid):
7275                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
7276
7277                 off  = si->off;
7278                 off -= offsetof(struct __sk_buff, tc_classid);
7279                 off += offsetof(struct sk_buff, cb);
7280                 off += offsetof(struct qdisc_skb_cb, tc_classid);
7281                 *target_size = 2;
7282                 if (type == BPF_WRITE)
7283                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
7284                                               si->src_reg, off);
7285                 else
7286                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
7287                                               si->src_reg, off);
7288                 break;
7289
7290         case offsetof(struct __sk_buff, data):
7291                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
7292                                       si->dst_reg, si->src_reg,
7293                                       offsetof(struct sk_buff, data));
7294                 break;
7295
7296         case offsetof(struct __sk_buff, data_meta):
7297                 off  = si->off;
7298                 off -= offsetof(struct __sk_buff, data_meta);
7299                 off += offsetof(struct sk_buff, cb);
7300                 off += offsetof(struct bpf_skb_data_end, data_meta);
7301                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7302                                       si->src_reg, off);
7303                 break;
7304
7305         case offsetof(struct __sk_buff, data_end):
7306                 off  = si->off;
7307                 off -= offsetof(struct __sk_buff, data_end);
7308                 off += offsetof(struct sk_buff, cb);
7309                 off += offsetof(struct bpf_skb_data_end, data_end);
7310                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7311                                       si->src_reg, off);
7312                 break;
7313
7314         case offsetof(struct __sk_buff, tc_index):
7315 #ifdef CONFIG_NET_SCHED
7316                 if (type == BPF_WRITE)
7317                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7318                                               bpf_target_off(struct sk_buff, tc_index, 2,
7319                                                              target_size));
7320                 else
7321                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7322                                               bpf_target_off(struct sk_buff, tc_index, 2,
7323                                                              target_size));
7324 #else
7325                 *target_size = 2;
7326                 if (type == BPF_WRITE)
7327                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
7328                 else
7329                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7330 #endif
7331                 break;
7332
7333         case offsetof(struct __sk_buff, napi_id):
7334 #if defined(CONFIG_NET_RX_BUSY_POLL)
7335                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7336                                       bpf_target_off(struct sk_buff, napi_id, 4,
7337                                                      target_size));
7338                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
7339                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7340 #else
7341                 *target_size = 4;
7342                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7343 #endif
7344                 break;
7345         case offsetof(struct __sk_buff, family):
7346                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
7347
7348                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7349                                       si->dst_reg, si->src_reg,
7350                                       offsetof(struct sk_buff, sk));
7351                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7352                                       bpf_target_off(struct sock_common,
7353                                                      skc_family,
7354                                                      2, target_size));
7355                 break;
7356         case offsetof(struct __sk_buff, remote_ip4):
7357                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
7358
7359                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7360                                       si->dst_reg, si->src_reg,
7361                                       offsetof(struct sk_buff, sk));
7362                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7363                                       bpf_target_off(struct sock_common,
7364                                                      skc_daddr,
7365                                                      4, target_size));
7366                 break;
7367         case offsetof(struct __sk_buff, local_ip4):
7368                 BUILD_BUG_ON(sizeof_field(struct sock_common,
7369                                           skc_rcv_saddr) != 4);
7370
7371                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7372                                       si->dst_reg, si->src_reg,
7373                                       offsetof(struct sk_buff, sk));
7374                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7375                                       bpf_target_off(struct sock_common,
7376                                                      skc_rcv_saddr,
7377                                                      4, target_size));
7378                 break;
7379         case offsetof(struct __sk_buff, remote_ip6[0]) ...
7380              offsetof(struct __sk_buff, remote_ip6[3]):
7381 #if IS_ENABLED(CONFIG_IPV6)
7382                 BUILD_BUG_ON(sizeof_field(struct sock_common,
7383                                           skc_v6_daddr.s6_addr32[0]) != 4);
7384
7385                 off = si->off;
7386                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
7387
7388                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7389                                       si->dst_reg, si->src_reg,
7390                                       offsetof(struct sk_buff, sk));
7391                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7392                                       offsetof(struct sock_common,
7393                                                skc_v6_daddr.s6_addr32[0]) +
7394                                       off);
7395 #else
7396                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7397 #endif
7398                 break;
7399         case offsetof(struct __sk_buff, local_ip6[0]) ...
7400              offsetof(struct __sk_buff, local_ip6[3]):
7401 #if IS_ENABLED(CONFIG_IPV6)
7402                 BUILD_BUG_ON(sizeof_field(struct sock_common,
7403                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7404
7405                 off = si->off;
7406                 off -= offsetof(struct __sk_buff, local_ip6[0]);
7407
7408                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7409                                       si->dst_reg, si->src_reg,
7410                                       offsetof(struct sk_buff, sk));
7411                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7412                                       offsetof(struct sock_common,
7413                                                skc_v6_rcv_saddr.s6_addr32[0]) +
7414                                       off);
7415 #else
7416                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7417 #endif
7418                 break;
7419
7420         case offsetof(struct __sk_buff, remote_port):
7421                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
7422
7423                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7424                                       si->dst_reg, si->src_reg,
7425                                       offsetof(struct sk_buff, sk));
7426                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7427                                       bpf_target_off(struct sock_common,
7428                                                      skc_dport,
7429                                                      2, target_size));
7430 #ifndef __BIG_ENDIAN_BITFIELD
7431                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7432 #endif
7433                 break;
7434
7435         case offsetof(struct __sk_buff, local_port):
7436                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
7437
7438                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7439                                       si->dst_reg, si->src_reg,
7440                                       offsetof(struct sk_buff, sk));
7441                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7442                                       bpf_target_off(struct sock_common,
7443                                                      skc_num, 2, target_size));
7444                 break;
7445
7446         case offsetof(struct __sk_buff, tstamp):
7447                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
7448
7449                 if (type == BPF_WRITE)
7450                         *insn++ = BPF_STX_MEM(BPF_DW,
7451                                               si->dst_reg, si->src_reg,
7452                                               bpf_target_off(struct sk_buff,
7453                                                              tstamp, 8,
7454                                                              target_size));
7455                 else
7456                         *insn++ = BPF_LDX_MEM(BPF_DW,
7457                                               si->dst_reg, si->src_reg,
7458                                               bpf_target_off(struct sk_buff,
7459                                                              tstamp, 8,
7460                                                              target_size));
7461                 break;
7462
7463         case offsetof(struct __sk_buff, gso_segs):
7464                 /* si->dst_reg = skb_shinfo(SKB); */
7465 #ifdef NET_SKBUFF_DATA_USES_OFFSET
7466                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7467                                       BPF_REG_AX, si->src_reg,
7468                                       offsetof(struct sk_buff, end));
7469                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
7470                                       si->dst_reg, si->src_reg,
7471                                       offsetof(struct sk_buff, head));
7472                 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
7473 #else
7474                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7475                                       si->dst_reg, si->src_reg,
7476                                       offsetof(struct sk_buff, end));
7477 #endif
7478                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
7479                                       si->dst_reg, si->dst_reg,
7480                                       bpf_target_off(struct skb_shared_info,
7481                                                      gso_segs, 2,
7482                                                      target_size));
7483                 break;
7484         case offsetof(struct __sk_buff, wire_len):
7485                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
7486
7487                 off = si->off;
7488                 off -= offsetof(struct __sk_buff, wire_len);
7489                 off += offsetof(struct sk_buff, cb);
7490                 off += offsetof(struct qdisc_skb_cb, pkt_len);
7491                 *target_size = 4;
7492                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
7493                 break;
7494
7495         case offsetof(struct __sk_buff, sk):
7496                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7497                                       si->dst_reg, si->src_reg,
7498                                       offsetof(struct sk_buff, sk));
7499                 break;
7500         }
7501
7502         return insn - insn_buf;
7503 }
7504
7505 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7506                                 const struct bpf_insn *si,
7507                                 struct bpf_insn *insn_buf,
7508                                 struct bpf_prog *prog, u32 *target_size)
7509 {
7510         struct bpf_insn *insn = insn_buf;
7511         int off;
7512
7513         switch (si->off) {
7514         case offsetof(struct bpf_sock, bound_dev_if):
7515                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
7516
7517                 if (type == BPF_WRITE)
7518                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7519                                         offsetof(struct sock, sk_bound_dev_if));
7520                 else
7521                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7522                                       offsetof(struct sock, sk_bound_dev_if));
7523                 break;
7524
7525         case offsetof(struct bpf_sock, mark):
7526                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
7527
7528                 if (type == BPF_WRITE)
7529                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7530                                         offsetof(struct sock, sk_mark));
7531                 else
7532                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7533                                       offsetof(struct sock, sk_mark));
7534                 break;
7535
7536         case offsetof(struct bpf_sock, priority):
7537                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
7538
7539                 if (type == BPF_WRITE)
7540                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7541                                         offsetof(struct sock, sk_priority));
7542                 else
7543                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7544                                       offsetof(struct sock, sk_priority));
7545                 break;
7546
7547         case offsetof(struct bpf_sock, family):
7548                 *insn++ = BPF_LDX_MEM(
7549                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7550                         si->dst_reg, si->src_reg,
7551                         bpf_target_off(struct sock_common,
7552                                        skc_family,
7553                                        sizeof_field(struct sock_common,
7554                                                     skc_family),
7555                                        target_size));
7556                 break;
7557
7558         case offsetof(struct bpf_sock, type):
7559                 *insn++ = BPF_LDX_MEM(
7560                         BPF_FIELD_SIZEOF(struct sock, sk_type),
7561                         si->dst_reg, si->src_reg,
7562                         bpf_target_off(struct sock, sk_type,
7563                                        sizeof_field(struct sock, sk_type),
7564                                        target_size));
7565                 break;
7566
7567         case offsetof(struct bpf_sock, protocol):
7568                 *insn++ = BPF_LDX_MEM(
7569                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
7570                         si->dst_reg, si->src_reg,
7571                         bpf_target_off(struct sock, sk_protocol,
7572                                        sizeof_field(struct sock, sk_protocol),
7573                                        target_size));
7574                 break;
7575
7576         case offsetof(struct bpf_sock, src_ip4):
7577                 *insn++ = BPF_LDX_MEM(
7578                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7579                         bpf_target_off(struct sock_common, skc_rcv_saddr,
7580                                        sizeof_field(struct sock_common,
7581                                                     skc_rcv_saddr),
7582                                        target_size));
7583                 break;
7584
7585         case offsetof(struct bpf_sock, dst_ip4):
7586                 *insn++ = BPF_LDX_MEM(
7587                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7588                         bpf_target_off(struct sock_common, skc_daddr,
7589                                        sizeof_field(struct sock_common,
7590                                                     skc_daddr),
7591                                        target_size));
7592                 break;
7593
7594         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7595 #if IS_ENABLED(CONFIG_IPV6)
7596                 off = si->off;
7597                 off -= offsetof(struct bpf_sock, src_ip6[0]);
7598                 *insn++ = BPF_LDX_MEM(
7599                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7600                         bpf_target_off(
7601                                 struct sock_common,
7602                                 skc_v6_rcv_saddr.s6_addr32[0],
7603                                 sizeof_field(struct sock_common,
7604                                              skc_v6_rcv_saddr.s6_addr32[0]),
7605                                 target_size) + off);
7606 #else
7607                 (void)off;
7608                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7609 #endif
7610                 break;
7611
7612         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7613 #if IS_ENABLED(CONFIG_IPV6)
7614                 off = si->off;
7615                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7616                 *insn++ = BPF_LDX_MEM(
7617                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7618                         bpf_target_off(struct sock_common,
7619                                        skc_v6_daddr.s6_addr32[0],
7620                                        sizeof_field(struct sock_common,
7621                                                     skc_v6_daddr.s6_addr32[0]),
7622                                        target_size) + off);
7623 #else
7624                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7625                 *target_size = 4;
7626 #endif
7627                 break;
7628
7629         case offsetof(struct bpf_sock, src_port):
7630                 *insn++ = BPF_LDX_MEM(
7631                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7632                         si->dst_reg, si->src_reg,
7633                         bpf_target_off(struct sock_common, skc_num,
7634                                        sizeof_field(struct sock_common,
7635                                                     skc_num),
7636                                        target_size));
7637                 break;
7638
7639         case offsetof(struct bpf_sock, dst_port):
7640                 *insn++ = BPF_LDX_MEM(
7641                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7642                         si->dst_reg, si->src_reg,
7643                         bpf_target_off(struct sock_common, skc_dport,
7644                                        sizeof_field(struct sock_common,
7645                                                     skc_dport),
7646                                        target_size));
7647                 break;
7648
7649         case offsetof(struct bpf_sock, state):
7650                 *insn++ = BPF_LDX_MEM(
7651                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7652                         si->dst_reg, si->src_reg,
7653                         bpf_target_off(struct sock_common, skc_state,
7654                                        sizeof_field(struct sock_common,
7655                                                     skc_state),
7656                                        target_size));
7657                 break;
7658         }
7659
7660         return insn - insn_buf;
7661 }
7662
7663 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7664                                          const struct bpf_insn *si,
7665                                          struct bpf_insn *insn_buf,
7666                                          struct bpf_prog *prog, u32 *target_size)
7667 {
7668         struct bpf_insn *insn = insn_buf;
7669
7670         switch (si->off) {
7671         case offsetof(struct __sk_buff, ifindex):
7672                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7673                                       si->dst_reg, si->src_reg,
7674                                       offsetof(struct sk_buff, dev));
7675                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7676                                       bpf_target_off(struct net_device, ifindex, 4,
7677                                                      target_size));
7678                 break;
7679         default:
7680                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7681                                               target_size);
7682         }
7683
7684         return insn - insn_buf;
7685 }
7686
7687 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7688                                   const struct bpf_insn *si,
7689                                   struct bpf_insn *insn_buf,
7690                                   struct bpf_prog *prog, u32 *target_size)
7691 {
7692         struct bpf_insn *insn = insn_buf;
7693
7694         switch (si->off) {
7695         case offsetof(struct xdp_md, data):
7696                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7697                                       si->dst_reg, si->src_reg,
7698                                       offsetof(struct xdp_buff, data));
7699                 break;
7700         case offsetof(struct xdp_md, data_meta):
7701                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7702                                       si->dst_reg, si->src_reg,
7703                                       offsetof(struct xdp_buff, data_meta));
7704                 break;
7705         case offsetof(struct xdp_md, data_end):
7706                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7707                                       si->dst_reg, si->src_reg,
7708                                       offsetof(struct xdp_buff, data_end));
7709                 break;
7710         case offsetof(struct xdp_md, ingress_ifindex):
7711                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7712                                       si->dst_reg, si->src_reg,
7713                                       offsetof(struct xdp_buff, rxq));
7714                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7715                                       si->dst_reg, si->dst_reg,
7716                                       offsetof(struct xdp_rxq_info, dev));
7717                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7718                                       offsetof(struct net_device, ifindex));
7719                 break;
7720         case offsetof(struct xdp_md, rx_queue_index):
7721                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7722                                       si->dst_reg, si->src_reg,
7723                                       offsetof(struct xdp_buff, rxq));
7724                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7725                                       offsetof(struct xdp_rxq_info,
7726                                                queue_index));
7727                 break;
7728         }
7729
7730         return insn - insn_buf;
7731 }
7732
7733 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7734  * context Structure, F is Field in context structure that contains a pointer
7735  * to Nested Structure of type NS that has the field NF.
7736  *
7737  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7738  * sure that SIZE is not greater than actual size of S.F.NF.
7739  *
7740  * If offset OFF is provided, the load happens from that offset relative to
7741  * offset of NF.
7742  */
7743 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
7744         do {                                                                   \
7745                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
7746                                       si->src_reg, offsetof(S, F));            \
7747                 *insn++ = BPF_LDX_MEM(                                         \
7748                         SIZE, si->dst_reg, si->dst_reg,                        \
7749                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
7750                                        target_size)                            \
7751                                 + OFF);                                        \
7752         } while (0)
7753
7754 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
7755         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
7756                                              BPF_FIELD_SIZEOF(NS, NF), 0)
7757
7758 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7759  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7760  *
7761  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7762  * "register" since two registers available in convert_ctx_access are not
7763  * enough: we can't override neither SRC, since it contains value to store, nor
7764  * DST since it contains pointer to context that may be used by later
7765  * instructions. But we need a temporary place to save pointer to nested
7766  * structure whose field we want to store to.
7767  */
7768 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
7769         do {                                                                   \
7770                 int tmp_reg = BPF_REG_9;                                       \
7771                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
7772                         --tmp_reg;                                             \
7773                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
7774                         --tmp_reg;                                             \
7775                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
7776                                       offsetof(S, TF));                        \
7777                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
7778                                       si->dst_reg, offsetof(S, F));            \
7779                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
7780                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
7781                                        target_size)                            \
7782                                 + OFF);                                        \
7783                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
7784                                       offsetof(S, TF));                        \
7785         } while (0)
7786
7787 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7788                                                       TF)                      \
7789         do {                                                                   \
7790                 if (type == BPF_WRITE) {                                       \
7791                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
7792                                                          OFF, TF);             \
7793                 } else {                                                       \
7794                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
7795                                 S, NS, F, NF, SIZE, OFF);  \
7796                 }                                                              \
7797         } while (0)
7798
7799 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
7800         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
7801                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7802
7803 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7804                                         const struct bpf_insn *si,
7805                                         struct bpf_insn *insn_buf,
7806                                         struct bpf_prog *prog, u32 *target_size)
7807 {
7808         struct bpf_insn *insn = insn_buf;
7809         int off;
7810
7811         switch (si->off) {
7812         case offsetof(struct bpf_sock_addr, user_family):
7813                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7814                                             struct sockaddr, uaddr, sa_family);
7815                 break;
7816
7817         case offsetof(struct bpf_sock_addr, user_ip4):
7818                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7819                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7820                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7821                 break;
7822
7823         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7824                 off = si->off;
7825                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7826                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7827                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7828                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7829                         tmp_reg);
7830                 break;
7831
7832         case offsetof(struct bpf_sock_addr, user_port):
7833                 /* To get port we need to know sa_family first and then treat
7834                  * sockaddr as either sockaddr_in or sockaddr_in6.
7835                  * Though we can simplify since port field has same offset and
7836                  * size in both structures.
7837                  * Here we check this invariant and use just one of the
7838                  * structures if it's true.
7839                  */
7840                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7841                              offsetof(struct sockaddr_in6, sin6_port));
7842                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
7843                              sizeof_field(struct sockaddr_in6, sin6_port));
7844                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7845                                                      struct sockaddr_in6, uaddr,
7846                                                      sin6_port, tmp_reg);
7847                 break;
7848
7849         case offsetof(struct bpf_sock_addr, family):
7850                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7851                                             struct sock, sk, sk_family);
7852                 break;
7853
7854         case offsetof(struct bpf_sock_addr, type):
7855                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7856                                             struct sock, sk, sk_type);
7857                 break;
7858
7859         case offsetof(struct bpf_sock_addr, protocol):
7860                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7861                                             struct sock, sk, sk_protocol);
7862                 break;
7863
7864         case offsetof(struct bpf_sock_addr, msg_src_ip4):
7865                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7866                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7867                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7868                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7869                 break;
7870
7871         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7872                                 msg_src_ip6[3]):
7873                 off = si->off;
7874                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7875                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7876                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7877                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7878                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7879                 break;
7880         case offsetof(struct bpf_sock_addr, sk):
7881                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
7882                                       si->dst_reg, si->src_reg,
7883                                       offsetof(struct bpf_sock_addr_kern, sk));
7884                 break;
7885         }
7886
7887         return insn - insn_buf;
7888 }
7889
7890 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7891                                        const struct bpf_insn *si,
7892                                        struct bpf_insn *insn_buf,
7893                                        struct bpf_prog *prog,
7894                                        u32 *target_size)
7895 {
7896         struct bpf_insn *insn = insn_buf;
7897         int off;
7898
7899 /* Helper macro for adding read access to tcp_sock or sock fields. */
7900 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
7901         do {                                                                  \
7902                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
7903                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
7904                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7905                                                 struct bpf_sock_ops_kern,     \
7906                                                 is_fullsock),                 \
7907                                       si->dst_reg, si->src_reg,               \
7908                                       offsetof(struct bpf_sock_ops_kern,      \
7909                                                is_fullsock));                 \
7910                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
7911                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7912                                                 struct bpf_sock_ops_kern, sk),\
7913                                       si->dst_reg, si->src_reg,               \
7914                                       offsetof(struct bpf_sock_ops_kern, sk));\
7915                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
7916                                                        OBJ_FIELD),            \
7917                                       si->dst_reg, si->dst_reg,               \
7918                                       offsetof(OBJ, OBJ_FIELD));              \
7919         } while (0)
7920
7921 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
7922                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
7923
7924 /* Helper macro for adding write access to tcp_sock or sock fields.
7925  * The macro is called with two registers, dst_reg which contains a pointer
7926  * to ctx (context) and src_reg which contains the value that should be
7927  * stored. However, we need an additional register since we cannot overwrite
7928  * dst_reg because it may be used later in the program.
7929  * Instead we "borrow" one of the other register. We first save its value
7930  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
7931  * it at the end of the macro.
7932  */
7933 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
7934         do {                                                                  \
7935                 int reg = BPF_REG_9;                                          \
7936                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
7937                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
7938                 if (si->dst_reg == reg || si->src_reg == reg)                 \
7939                         reg--;                                                \
7940                 if (si->dst_reg == reg || si->src_reg == reg)                 \
7941                         reg--;                                                \
7942                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
7943                                       offsetof(struct bpf_sock_ops_kern,      \
7944                                                temp));                        \
7945                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7946                                                 struct bpf_sock_ops_kern,     \
7947                                                 is_fullsock),                 \
7948                                       reg, si->dst_reg,                       \
7949                                       offsetof(struct bpf_sock_ops_kern,      \
7950                                                is_fullsock));                 \
7951                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
7952                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7953                                                 struct bpf_sock_ops_kern, sk),\
7954                                       reg, si->dst_reg,                       \
7955                                       offsetof(struct bpf_sock_ops_kern, sk));\
7956                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
7957                                       reg, si->src_reg,                       \
7958                                       offsetof(OBJ, OBJ_FIELD));              \
7959                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
7960                                       offsetof(struct bpf_sock_ops_kern,      \
7961                                                temp));                        \
7962         } while (0)
7963
7964 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
7965         do {                                                                  \
7966                 if (TYPE == BPF_WRITE)                                        \
7967                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
7968                 else                                                          \
7969                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
7970         } while (0)
7971
7972         if (insn > insn_buf)
7973                 return insn - insn_buf;
7974
7975         switch (si->off) {
7976         case offsetof(struct bpf_sock_ops, op) ...
7977              offsetof(struct bpf_sock_ops, replylong[3]):
7978                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, op) !=
7979                              sizeof_field(struct bpf_sock_ops_kern, op));
7980                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
7981                              sizeof_field(struct bpf_sock_ops_kern, reply));
7982                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
7983                              sizeof_field(struct bpf_sock_ops_kern, replylong));
7984                 off = si->off;
7985                 off -= offsetof(struct bpf_sock_ops, op);
7986                 off += offsetof(struct bpf_sock_ops_kern, op);
7987                 if (type == BPF_WRITE)
7988                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7989                                               off);
7990                 else
7991                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7992                                               off);
7993                 break;
7994
7995         case offsetof(struct bpf_sock_ops, family):
7996                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
7997
7998                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7999                                               struct bpf_sock_ops_kern, sk),
8000                                       si->dst_reg, si->src_reg,
8001                                       offsetof(struct bpf_sock_ops_kern, sk));
8002                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8003                                       offsetof(struct sock_common, skc_family));
8004                 break;
8005
8006         case offsetof(struct bpf_sock_ops, remote_ip4):
8007                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8008
8009                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8010                                                 struct bpf_sock_ops_kern, sk),
8011                                       si->dst_reg, si->src_reg,
8012                                       offsetof(struct bpf_sock_ops_kern, sk));
8013                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8014                                       offsetof(struct sock_common, skc_daddr));
8015                 break;
8016
8017         case offsetof(struct bpf_sock_ops, local_ip4):
8018                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8019                                           skc_rcv_saddr) != 4);
8020
8021                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8022                                               struct bpf_sock_ops_kern, sk),
8023                                       si->dst_reg, si->src_reg,
8024                                       offsetof(struct bpf_sock_ops_kern, sk));
8025                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8026                                       offsetof(struct sock_common,
8027                                                skc_rcv_saddr));
8028                 break;
8029
8030         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
8031              offsetof(struct bpf_sock_ops, remote_ip6[3]):
8032 #if IS_ENABLED(CONFIG_IPV6)
8033                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8034                                           skc_v6_daddr.s6_addr32[0]) != 4);
8035
8036                 off = si->off;
8037                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
8038                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8039                                                 struct bpf_sock_ops_kern, sk),
8040                                       si->dst_reg, si->src_reg,
8041                                       offsetof(struct bpf_sock_ops_kern, sk));
8042                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8043                                       offsetof(struct sock_common,
8044                                                skc_v6_daddr.s6_addr32[0]) +
8045                                       off);
8046 #else
8047                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8048 #endif
8049                 break;
8050
8051         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
8052              offsetof(struct bpf_sock_ops, local_ip6[3]):
8053 #if IS_ENABLED(CONFIG_IPV6)
8054                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8055                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8056
8057                 off = si->off;
8058                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
8059                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8060                                                 struct bpf_sock_ops_kern, sk),
8061                                       si->dst_reg, si->src_reg,
8062                                       offsetof(struct bpf_sock_ops_kern, sk));
8063                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8064                                       offsetof(struct sock_common,
8065                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8066                                       off);
8067 #else
8068                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8069 #endif
8070                 break;
8071
8072         case offsetof(struct bpf_sock_ops, remote_port):
8073                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8074
8075                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8076                                                 struct bpf_sock_ops_kern, sk),
8077                                       si->dst_reg, si->src_reg,
8078                                       offsetof(struct bpf_sock_ops_kern, sk));
8079                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8080                                       offsetof(struct sock_common, skc_dport));
8081 #ifndef __BIG_ENDIAN_BITFIELD
8082                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8083 #endif
8084                 break;
8085
8086         case offsetof(struct bpf_sock_ops, local_port):
8087                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8088
8089                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8090                                                 struct bpf_sock_ops_kern, sk),
8091                                       si->dst_reg, si->src_reg,
8092                                       offsetof(struct bpf_sock_ops_kern, sk));
8093                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8094                                       offsetof(struct sock_common, skc_num));
8095                 break;
8096
8097         case offsetof(struct bpf_sock_ops, is_fullsock):
8098                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8099                                                 struct bpf_sock_ops_kern,
8100                                                 is_fullsock),
8101                                       si->dst_reg, si->src_reg,
8102                                       offsetof(struct bpf_sock_ops_kern,
8103                                                is_fullsock));
8104                 break;
8105
8106         case offsetof(struct bpf_sock_ops, state):
8107                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
8108
8109                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8110                                                 struct bpf_sock_ops_kern, sk),
8111                                       si->dst_reg, si->src_reg,
8112                                       offsetof(struct bpf_sock_ops_kern, sk));
8113                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
8114                                       offsetof(struct sock_common, skc_state));
8115                 break;
8116
8117         case offsetof(struct bpf_sock_ops, rtt_min):
8118                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
8119                              sizeof(struct minmax));
8120                 BUILD_BUG_ON(sizeof(struct minmax) <
8121                              sizeof(struct minmax_sample));
8122
8123                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8124                                                 struct bpf_sock_ops_kern, sk),
8125                                       si->dst_reg, si->src_reg,
8126                                       offsetof(struct bpf_sock_ops_kern, sk));
8127                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8128                                       offsetof(struct tcp_sock, rtt_min) +
8129                                       sizeof_field(struct minmax_sample, t));
8130                 break;
8131
8132         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
8133                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
8134                                    struct tcp_sock);
8135                 break;
8136
8137         case offsetof(struct bpf_sock_ops, sk_txhash):
8138                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
8139                                           struct sock, type);
8140                 break;
8141         case offsetof(struct bpf_sock_ops, snd_cwnd):
8142                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
8143                 break;
8144         case offsetof(struct bpf_sock_ops, srtt_us):
8145                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
8146                 break;
8147         case offsetof(struct bpf_sock_ops, snd_ssthresh):
8148                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
8149                 break;
8150         case offsetof(struct bpf_sock_ops, rcv_nxt):
8151                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
8152                 break;
8153         case offsetof(struct bpf_sock_ops, snd_nxt):
8154                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
8155                 break;
8156         case offsetof(struct bpf_sock_ops, snd_una):
8157                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
8158                 break;
8159         case offsetof(struct bpf_sock_ops, mss_cache):
8160                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
8161                 break;
8162         case offsetof(struct bpf_sock_ops, ecn_flags):
8163                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
8164                 break;
8165         case offsetof(struct bpf_sock_ops, rate_delivered):
8166                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
8167                 break;
8168         case offsetof(struct bpf_sock_ops, rate_interval_us):
8169                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
8170                 break;
8171         case offsetof(struct bpf_sock_ops, packets_out):
8172                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
8173                 break;
8174         case offsetof(struct bpf_sock_ops, retrans_out):
8175                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
8176                 break;
8177         case offsetof(struct bpf_sock_ops, total_retrans):
8178                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
8179                 break;
8180         case offsetof(struct bpf_sock_ops, segs_in):
8181                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
8182                 break;
8183         case offsetof(struct bpf_sock_ops, data_segs_in):
8184                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
8185                 break;
8186         case offsetof(struct bpf_sock_ops, segs_out):
8187                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
8188                 break;
8189         case offsetof(struct bpf_sock_ops, data_segs_out):
8190                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
8191                 break;
8192         case offsetof(struct bpf_sock_ops, lost_out):
8193                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
8194                 break;
8195         case offsetof(struct bpf_sock_ops, sacked_out):
8196                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
8197                 break;
8198         case offsetof(struct bpf_sock_ops, bytes_received):
8199                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
8200                 break;
8201         case offsetof(struct bpf_sock_ops, bytes_acked):
8202                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
8203                 break;
8204         case offsetof(struct bpf_sock_ops, sk):
8205                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8206                                                 struct bpf_sock_ops_kern,
8207                                                 is_fullsock),
8208                                       si->dst_reg, si->src_reg,
8209                                       offsetof(struct bpf_sock_ops_kern,
8210                                                is_fullsock));
8211                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8212                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8213                                                 struct bpf_sock_ops_kern, sk),
8214                                       si->dst_reg, si->src_reg,
8215                                       offsetof(struct bpf_sock_ops_kern, sk));
8216                 break;
8217         }
8218         return insn - insn_buf;
8219 }
8220
8221 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
8222                                      const struct bpf_insn *si,
8223                                      struct bpf_insn *insn_buf,
8224                                      struct bpf_prog *prog, u32 *target_size)
8225 {
8226         struct bpf_insn *insn = insn_buf;
8227         int off;
8228
8229         switch (si->off) {
8230         case offsetof(struct __sk_buff, data_end):
8231                 off  = si->off;
8232                 off -= offsetof(struct __sk_buff, data_end);
8233                 off += offsetof(struct sk_buff, cb);
8234                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
8235                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8236                                       si->src_reg, off);
8237                 break;
8238         default:
8239                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8240                                               target_size);
8241         }
8242
8243         return insn - insn_buf;
8244 }
8245
8246 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
8247                                      const struct bpf_insn *si,
8248                                      struct bpf_insn *insn_buf,
8249                                      struct bpf_prog *prog, u32 *target_size)
8250 {
8251         struct bpf_insn *insn = insn_buf;
8252 #if IS_ENABLED(CONFIG_IPV6)
8253         int off;
8254 #endif
8255
8256         /* convert ctx uses the fact sg element is first in struct */
8257         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
8258
8259         switch (si->off) {
8260         case offsetof(struct sk_msg_md, data):
8261                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
8262                                       si->dst_reg, si->src_reg,
8263                                       offsetof(struct sk_msg, data));
8264                 break;
8265         case offsetof(struct sk_msg_md, data_end):
8266                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
8267                                       si->dst_reg, si->src_reg,
8268                                       offsetof(struct sk_msg, data_end));
8269                 break;
8270         case offsetof(struct sk_msg_md, family):
8271                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8272
8273                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8274                                               struct sk_msg, sk),
8275                                       si->dst_reg, si->src_reg,
8276                                       offsetof(struct sk_msg, sk));
8277                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8278                                       offsetof(struct sock_common, skc_family));
8279                 break;
8280
8281         case offsetof(struct sk_msg_md, remote_ip4):
8282                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8283
8284                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8285                                                 struct sk_msg, sk),
8286                                       si->dst_reg, si->src_reg,
8287                                       offsetof(struct sk_msg, sk));
8288                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8289                                       offsetof(struct sock_common, skc_daddr));
8290                 break;
8291
8292         case offsetof(struct sk_msg_md, local_ip4):
8293                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8294                                           skc_rcv_saddr) != 4);
8295
8296                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8297                                               struct sk_msg, sk),
8298                                       si->dst_reg, si->src_reg,
8299                                       offsetof(struct sk_msg, sk));
8300                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8301                                       offsetof(struct sock_common,
8302                                                skc_rcv_saddr));
8303                 break;
8304
8305         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
8306              offsetof(struct sk_msg_md, remote_ip6[3]):
8307 #if IS_ENABLED(CONFIG_IPV6)
8308                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8309                                           skc_v6_daddr.s6_addr32[0]) != 4);
8310
8311                 off = si->off;
8312                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
8313                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8314                                                 struct sk_msg, sk),
8315                                       si->dst_reg, si->src_reg,
8316                                       offsetof(struct sk_msg, sk));
8317                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8318                                       offsetof(struct sock_common,
8319                                                skc_v6_daddr.s6_addr32[0]) +
8320                                       off);
8321 #else
8322                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8323 #endif
8324                 break;
8325
8326         case offsetof(struct sk_msg_md, local_ip6[0]) ...
8327              offsetof(struct sk_msg_md, local_ip6[3]):
8328 #if IS_ENABLED(CONFIG_IPV6)
8329                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8330                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8331
8332                 off = si->off;
8333                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
8334                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8335                                                 struct sk_msg, sk),
8336                                       si->dst_reg, si->src_reg,
8337                                       offsetof(struct sk_msg, sk));
8338                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8339                                       offsetof(struct sock_common,
8340                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8341                                       off);
8342 #else
8343                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8344 #endif
8345                 break;
8346
8347         case offsetof(struct sk_msg_md, remote_port):
8348                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8349
8350                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8351                                                 struct sk_msg, sk),
8352                                       si->dst_reg, si->src_reg,
8353                                       offsetof(struct sk_msg, sk));
8354                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8355                                       offsetof(struct sock_common, skc_dport));
8356 #ifndef __BIG_ENDIAN_BITFIELD
8357                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8358 #endif
8359                 break;
8360
8361         case offsetof(struct sk_msg_md, local_port):
8362                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8363
8364                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8365                                                 struct sk_msg, sk),
8366                                       si->dst_reg, si->src_reg,
8367                                       offsetof(struct sk_msg, sk));
8368                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8369                                       offsetof(struct sock_common, skc_num));
8370                 break;
8371
8372         case offsetof(struct sk_msg_md, size):
8373                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
8374                                       si->dst_reg, si->src_reg,
8375                                       offsetof(struct sk_msg_sg, size));
8376                 break;
8377         }
8378
8379         return insn - insn_buf;
8380 }
8381
8382 const struct bpf_verifier_ops sk_filter_verifier_ops = {
8383         .get_func_proto         = sk_filter_func_proto,
8384         .is_valid_access        = sk_filter_is_valid_access,
8385         .convert_ctx_access     = bpf_convert_ctx_access,
8386         .gen_ld_abs             = bpf_gen_ld_abs,
8387 };
8388
8389 const struct bpf_prog_ops sk_filter_prog_ops = {
8390         .test_run               = bpf_prog_test_run_skb,
8391 };
8392
8393 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
8394         .get_func_proto         = tc_cls_act_func_proto,
8395         .is_valid_access        = tc_cls_act_is_valid_access,
8396         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
8397         .gen_prologue           = tc_cls_act_prologue,
8398         .gen_ld_abs             = bpf_gen_ld_abs,
8399 };
8400
8401 const struct bpf_prog_ops tc_cls_act_prog_ops = {
8402         .test_run               = bpf_prog_test_run_skb,
8403 };
8404
8405 const struct bpf_verifier_ops xdp_verifier_ops = {
8406         .get_func_proto         = xdp_func_proto,
8407         .is_valid_access        = xdp_is_valid_access,
8408         .convert_ctx_access     = xdp_convert_ctx_access,
8409         .gen_prologue           = bpf_noop_prologue,
8410 };
8411
8412 const struct bpf_prog_ops xdp_prog_ops = {
8413         .test_run               = bpf_prog_test_run_xdp,
8414 };
8415
8416 const struct bpf_verifier_ops cg_skb_verifier_ops = {
8417         .get_func_proto         = cg_skb_func_proto,
8418         .is_valid_access        = cg_skb_is_valid_access,
8419         .convert_ctx_access     = bpf_convert_ctx_access,
8420 };
8421
8422 const struct bpf_prog_ops cg_skb_prog_ops = {
8423         .test_run               = bpf_prog_test_run_skb,
8424 };
8425
8426 const struct bpf_verifier_ops lwt_in_verifier_ops = {
8427         .get_func_proto         = lwt_in_func_proto,
8428         .is_valid_access        = lwt_is_valid_access,
8429         .convert_ctx_access     = bpf_convert_ctx_access,
8430 };
8431
8432 const struct bpf_prog_ops lwt_in_prog_ops = {
8433         .test_run               = bpf_prog_test_run_skb,
8434 };
8435
8436 const struct bpf_verifier_ops lwt_out_verifier_ops = {
8437         .get_func_proto         = lwt_out_func_proto,
8438         .is_valid_access        = lwt_is_valid_access,
8439         .convert_ctx_access     = bpf_convert_ctx_access,
8440 };
8441
8442 const struct bpf_prog_ops lwt_out_prog_ops = {
8443         .test_run               = bpf_prog_test_run_skb,
8444 };
8445
8446 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
8447         .get_func_proto         = lwt_xmit_func_proto,
8448         .is_valid_access        = lwt_is_valid_access,
8449         .convert_ctx_access     = bpf_convert_ctx_access,
8450         .gen_prologue           = tc_cls_act_prologue,
8451 };
8452
8453 const struct bpf_prog_ops lwt_xmit_prog_ops = {
8454         .test_run               = bpf_prog_test_run_skb,
8455 };
8456
8457 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
8458         .get_func_proto         = lwt_seg6local_func_proto,
8459         .is_valid_access        = lwt_is_valid_access,
8460         .convert_ctx_access     = bpf_convert_ctx_access,
8461 };
8462
8463 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
8464         .test_run               = bpf_prog_test_run_skb,
8465 };
8466
8467 const struct bpf_verifier_ops cg_sock_verifier_ops = {
8468         .get_func_proto         = sock_filter_func_proto,
8469         .is_valid_access        = sock_filter_is_valid_access,
8470         .convert_ctx_access     = bpf_sock_convert_ctx_access,
8471 };
8472
8473 const struct bpf_prog_ops cg_sock_prog_ops = {
8474 };
8475
8476 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
8477         .get_func_proto         = sock_addr_func_proto,
8478         .is_valid_access        = sock_addr_is_valid_access,
8479         .convert_ctx_access     = sock_addr_convert_ctx_access,
8480 };
8481
8482 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
8483 };
8484
8485 const struct bpf_verifier_ops sock_ops_verifier_ops = {
8486         .get_func_proto         = sock_ops_func_proto,
8487         .is_valid_access        = sock_ops_is_valid_access,
8488         .convert_ctx_access     = sock_ops_convert_ctx_access,
8489 };
8490
8491 const struct bpf_prog_ops sock_ops_prog_ops = {
8492 };
8493
8494 const struct bpf_verifier_ops sk_skb_verifier_ops = {
8495         .get_func_proto         = sk_skb_func_proto,
8496         .is_valid_access        = sk_skb_is_valid_access,
8497         .convert_ctx_access     = sk_skb_convert_ctx_access,
8498         .gen_prologue           = sk_skb_prologue,
8499 };
8500
8501 const struct bpf_prog_ops sk_skb_prog_ops = {
8502 };
8503
8504 const struct bpf_verifier_ops sk_msg_verifier_ops = {
8505         .get_func_proto         = sk_msg_func_proto,
8506         .is_valid_access        = sk_msg_is_valid_access,
8507         .convert_ctx_access     = sk_msg_convert_ctx_access,
8508         .gen_prologue           = bpf_noop_prologue,
8509 };
8510
8511 const struct bpf_prog_ops sk_msg_prog_ops = {
8512 };
8513
8514 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
8515         .get_func_proto         = flow_dissector_func_proto,
8516         .is_valid_access        = flow_dissector_is_valid_access,
8517         .convert_ctx_access     = flow_dissector_convert_ctx_access,
8518 };
8519
8520 const struct bpf_prog_ops flow_dissector_prog_ops = {
8521         .test_run               = bpf_prog_test_run_flow_dissector,
8522 };
8523
8524 int sk_detach_filter(struct sock *sk)
8525 {
8526         int ret = -ENOENT;
8527         struct sk_filter *filter;
8528
8529         if (sock_flag(sk, SOCK_FILTER_LOCKED))
8530                 return -EPERM;
8531
8532         filter = rcu_dereference_protected(sk->sk_filter,
8533                                            lockdep_sock_is_held(sk));
8534         if (filter) {
8535                 RCU_INIT_POINTER(sk->sk_filter, NULL);
8536                 sk_filter_uncharge(sk, filter);
8537                 ret = 0;
8538         }
8539
8540         return ret;
8541 }
8542 EXPORT_SYMBOL_GPL(sk_detach_filter);
8543
8544 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
8545                   unsigned int len)
8546 {
8547         struct sock_fprog_kern *fprog;
8548         struct sk_filter *filter;
8549         int ret = 0;
8550
8551         lock_sock(sk);
8552         filter = rcu_dereference_protected(sk->sk_filter,
8553                                            lockdep_sock_is_held(sk));
8554         if (!filter)
8555                 goto out;
8556
8557         /* We're copying the filter that has been originally attached,
8558          * so no conversion/decode needed anymore. eBPF programs that
8559          * have no original program cannot be dumped through this.
8560          */
8561         ret = -EACCES;
8562         fprog = filter->prog->orig_prog;
8563         if (!fprog)
8564                 goto out;
8565
8566         ret = fprog->len;
8567         if (!len)
8568                 /* User space only enquires number of filter blocks. */
8569                 goto out;
8570
8571         ret = -EINVAL;
8572         if (len < fprog->len)
8573                 goto out;
8574
8575         ret = -EFAULT;
8576         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8577                 goto out;
8578
8579         /* Instead of bytes, the API requests to return the number
8580          * of filter blocks.
8581          */
8582         ret = fprog->len;
8583 out:
8584         release_sock(sk);
8585         return ret;
8586 }
8587
8588 #ifdef CONFIG_INET
8589 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8590                                     struct sock_reuseport *reuse,
8591                                     struct sock *sk, struct sk_buff *skb,
8592                                     u32 hash)
8593 {
8594         reuse_kern->skb = skb;
8595         reuse_kern->sk = sk;
8596         reuse_kern->selected_sk = NULL;
8597         reuse_kern->data_end = skb->data + skb_headlen(skb);
8598         reuse_kern->hash = hash;
8599         reuse_kern->reuseport_id = reuse->reuseport_id;
8600         reuse_kern->bind_inany = reuse->bind_inany;
8601 }
8602
8603 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8604                                   struct bpf_prog *prog, struct sk_buff *skb,
8605                                   u32 hash)
8606 {
8607         struct sk_reuseport_kern reuse_kern;
8608         enum sk_action action;
8609
8610         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8611         action = BPF_PROG_RUN(prog, &reuse_kern);
8612
8613         if (action == SK_PASS)
8614                 return reuse_kern.selected_sk;
8615         else
8616                 return ERR_PTR(-ECONNREFUSED);
8617 }
8618
8619 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8620            struct bpf_map *, map, void *, key, u32, flags)
8621 {
8622         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
8623         struct sock_reuseport *reuse;
8624         struct sock *selected_sk;
8625
8626         selected_sk = map->ops->map_lookup_elem(map, key);
8627         if (!selected_sk)
8628                 return -ENOENT;
8629
8630         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8631         if (!reuse) {
8632                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
8633                  * The only (!reuse) case here is - the sk has already been
8634                  * unhashed (e.g. by close()), so treat it as -ENOENT.
8635                  *
8636                  * Other maps (e.g. sock_map) do not provide this guarantee and
8637                  * the sk may never be in the reuseport group to begin with.
8638                  */
8639                 return is_sockarray ? -ENOENT : -EINVAL;
8640         }
8641
8642         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8643                 struct sock *sk = reuse_kern->sk;
8644
8645                 if (sk->sk_protocol != selected_sk->sk_protocol)
8646                         return -EPROTOTYPE;
8647                 else if (sk->sk_family != selected_sk->sk_family)
8648                         return -EAFNOSUPPORT;
8649
8650                 /* Catch all. Likely bound to a different sockaddr. */
8651                 return -EBADFD;
8652         }
8653
8654         reuse_kern->selected_sk = selected_sk;
8655
8656         return 0;
8657 }
8658
8659 static const struct bpf_func_proto sk_select_reuseport_proto = {
8660         .func           = sk_select_reuseport,
8661         .gpl_only       = false,
8662         .ret_type       = RET_INTEGER,
8663         .arg1_type      = ARG_PTR_TO_CTX,
8664         .arg2_type      = ARG_CONST_MAP_PTR,
8665         .arg3_type      = ARG_PTR_TO_MAP_KEY,
8666         .arg4_type      = ARG_ANYTHING,
8667 };
8668
8669 BPF_CALL_4(sk_reuseport_load_bytes,
8670            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8671            void *, to, u32, len)
8672 {
8673         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8674 }
8675
8676 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8677         .func           = sk_reuseport_load_bytes,
8678         .gpl_only       = false,
8679         .ret_type       = RET_INTEGER,
8680         .arg1_type      = ARG_PTR_TO_CTX,
8681         .arg2_type      = ARG_ANYTHING,
8682         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
8683         .arg4_type      = ARG_CONST_SIZE,
8684 };
8685
8686 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8687            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8688            void *, to, u32, len, u32, start_header)
8689 {
8690         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8691                                                len, start_header);
8692 }
8693
8694 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8695         .func           = sk_reuseport_load_bytes_relative,
8696         .gpl_only       = false,
8697         .ret_type       = RET_INTEGER,
8698         .arg1_type      = ARG_PTR_TO_CTX,
8699         .arg2_type      = ARG_ANYTHING,
8700         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
8701         .arg4_type      = ARG_CONST_SIZE,
8702         .arg5_type      = ARG_ANYTHING,
8703 };
8704
8705 static const struct bpf_func_proto *
8706 sk_reuseport_func_proto(enum bpf_func_id func_id,
8707                         const struct bpf_prog *prog)
8708 {
8709         switch (func_id) {
8710         case BPF_FUNC_sk_select_reuseport:
8711                 return &sk_select_reuseport_proto;
8712         case BPF_FUNC_skb_load_bytes:
8713                 return &sk_reuseport_load_bytes_proto;
8714         case BPF_FUNC_skb_load_bytes_relative:
8715                 return &sk_reuseport_load_bytes_relative_proto;
8716         default:
8717                 return bpf_base_func_proto(func_id);
8718         }
8719 }
8720
8721 static bool
8722 sk_reuseport_is_valid_access(int off, int size,
8723                              enum bpf_access_type type,
8724                              const struct bpf_prog *prog,
8725                              struct bpf_insn_access_aux *info)
8726 {
8727         const u32 size_default = sizeof(__u32);
8728
8729         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8730             off % size || type != BPF_READ)
8731                 return false;
8732
8733         switch (off) {
8734         case offsetof(struct sk_reuseport_md, data):
8735                 info->reg_type = PTR_TO_PACKET;
8736                 return size == sizeof(__u64);
8737
8738         case offsetof(struct sk_reuseport_md, data_end):
8739                 info->reg_type = PTR_TO_PACKET_END;
8740                 return size == sizeof(__u64);
8741
8742         case offsetof(struct sk_reuseport_md, hash):
8743                 return size == size_default;
8744
8745         /* Fields that allow narrowing */
8746         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
8747                 if (size < sizeof_field(struct sk_buff, protocol))
8748                         return false;
8749                 /* fall through */
8750         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
8751         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
8752         case bpf_ctx_range(struct sk_reuseport_md, len):
8753                 bpf_ctx_record_field_size(info, size_default);
8754                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8755
8756         default:
8757                 return false;
8758         }
8759 }
8760
8761 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
8762         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8763                               si->dst_reg, si->src_reg,                 \
8764                               bpf_target_off(struct sk_reuseport_kern, F, \
8765                                              sizeof_field(struct sk_reuseport_kern, F), \
8766                                              target_size));             \
8767         })
8768
8769 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
8770         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
8771                                     struct sk_buff,                     \
8772                                     skb,                                \
8773                                     SKB_FIELD)
8774
8775 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
8776         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
8777                                     struct sock,                        \
8778                                     sk,                                 \
8779                                     SK_FIELD)
8780
8781 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8782                                            const struct bpf_insn *si,
8783                                            struct bpf_insn *insn_buf,
8784                                            struct bpf_prog *prog,
8785                                            u32 *target_size)
8786 {
8787         struct bpf_insn *insn = insn_buf;
8788
8789         switch (si->off) {
8790         case offsetof(struct sk_reuseport_md, data):
8791                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8792                 break;
8793
8794         case offsetof(struct sk_reuseport_md, len):
8795                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8796                 break;
8797
8798         case offsetof(struct sk_reuseport_md, eth_protocol):
8799                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8800                 break;
8801
8802         case offsetof(struct sk_reuseport_md, ip_protocol):
8803                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
8804                 break;
8805
8806         case offsetof(struct sk_reuseport_md, data_end):
8807                 SK_REUSEPORT_LOAD_FIELD(data_end);
8808                 break;
8809
8810         case offsetof(struct sk_reuseport_md, hash):
8811                 SK_REUSEPORT_LOAD_FIELD(hash);
8812                 break;
8813
8814         case offsetof(struct sk_reuseport_md, bind_inany):
8815                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8816                 break;
8817         }
8818
8819         return insn - insn_buf;
8820 }
8821
8822 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8823         .get_func_proto         = sk_reuseport_func_proto,
8824         .is_valid_access        = sk_reuseport_is_valid_access,
8825         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
8826 };
8827
8828 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8829 };
8830 #endif /* CONFIG_INET */
8831
8832 DEFINE_BPF_DISPATCHER(bpf_dispatcher_xdp)
8833
8834 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
8835 {
8836         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(bpf_dispatcher_xdp),
8837                                    prev_prog, prog);
8838 }