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(FIELD_SIZEOF(struct sk_buff, mark) != 4);
 278
 279                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 280                                       offsetof(struct sk_buff, mark));
 281                 break;
 282
 283         case SKF_AD_PKTTYPE:
 284                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 285                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 286 #ifdef __BIG_ENDIAN_BITFIELD
 287                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 288 #endif
 289                 break;
 290
 291         case SKF_AD_QUEUE:
 292                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
 293
 294                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 295                                       offsetof(struct sk_buff, queue_mapping));
 296                 break;
 297
 298         case SKF_AD_VLAN_TAG:
 299                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
 300
 301                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 302                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 303                                       offsetof(struct sk_buff, vlan_tci));
 304                 break;
 305         case SKF_AD_VLAN_TAG_PRESENT:
 306                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
 307                 if (PKT_VLAN_PRESENT_BIT)
 308                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
 309                 if (PKT_VLAN_PRESENT_BIT < 7)
 310                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 311                 break;
 312         }
 313
 314         return insn - insn_buf;
 315 }
 316
 317 static bool convert_bpf_extensions(struct sock_filter *fp,
 318                                    struct bpf_insn **insnp)
 319 {
 320         struct bpf_insn *insn = *insnp;
 321         u32 cnt;
 322
 323         switch (fp->k) {
 324         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 325                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
 326
 327                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 328                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 329                                       offsetof(struct sk_buff, protocol));
 330                 /* A = ntohs(A) [emitting a nop or swap16] */
 331                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 332                 break;
 333
 334         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 335                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 336                 insn += cnt - 1;
 337                 break;
 338
 339         case SKF_AD_OFF + SKF_AD_IFINDEX:
 340         case SKF_AD_OFF + SKF_AD_HATYPE:
 341                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
 342                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
 343
 344                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 345                                       BPF_REG_TMP, BPF_REG_CTX,
 346                                       offsetof(struct sk_buff, dev));
 347                 /* if (tmp != 0) goto pc + 1 */
 348                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 349                 *insn++ = BPF_EXIT_INSN();
 350                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 351                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 352                                             offsetof(struct net_device, ifindex));
 353                 else
 354                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 355                                             offsetof(struct net_device, type));
 356                 break;
 357
 358         case SKF_AD_OFF + SKF_AD_MARK:
 359                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 360                 insn += cnt - 1;
 361                 break;
 362
 363         case SKF_AD_OFF + SKF_AD_RXHASH:
 364                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
 365
 366                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 367                                     offsetof(struct sk_buff, hash));
 368                 break;
 369
 370         case SKF_AD_OFF + SKF_AD_QUEUE:
 371                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 372                 insn += cnt - 1;
 373                 break;
 374
 375         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 376                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 377                                          BPF_REG_A, BPF_REG_CTX, insn);
 378                 insn += cnt - 1;
 379                 break;
 380
 381         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 382                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 383                                          BPF_REG_A, BPF_REG_CTX, insn);
 384                 insn += cnt - 1;
 385                 break;
 386
 387         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 388                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
 389
 390                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 391                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 392                                       offsetof(struct sk_buff, vlan_proto));
 393                 /* A = ntohs(A) [emitting a nop or swap16] */
 394                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 395                 break;
 396
 397         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 398         case SKF_AD_OFF + SKF_AD_NLATTR:
 399         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 400         case SKF_AD_OFF + SKF_AD_CPU:
 401         case SKF_AD_OFF + SKF_AD_RANDOM:
 402                 /* arg1 = CTX */
 403                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 404                 /* arg2 = A */
 405                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 406                 /* arg3 = X */
 407                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 408                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 409                 switch (fp->k) {
 410                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 411                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 412                         break;
 413                 case SKF_AD_OFF + SKF_AD_NLATTR:
 414                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 415                         break;
 416                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 417                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 418                         break;
 419                 case SKF_AD_OFF + SKF_AD_CPU:
 420                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 421                         break;
 422                 case SKF_AD_OFF + SKF_AD_RANDOM:
 423                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 424                         bpf_user_rnd_init_once();
 425                         break;
 426                 }
 427                 break;
 428
 429         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 430                 /* A ^= X */
 431                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 432                 break;
 433
 434         default:
 435                 /* This is just a dummy call to avoid letting the compiler
 436                  * evict __bpf_call_base() as an optimization. Placed here
 437                  * where no-one bothers.
 438                  */
 439                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 440                 return false;
 441         }
 442
 443         *insnp = insn;
 444         return true;
 445 }
 446
 447 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 448 {
 449         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 450         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 451         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 452                       BPF_SIZE(fp->code) == BPF_W;
 453         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 454         const int ip_align = NET_IP_ALIGN;
 455         struct bpf_insn *insn = *insnp;
 456         int offset = fp->k;
 457
 458         if (!indirect &&
 459             ((unaligned_ok && offset >= 0) ||
 460              (!unaligned_ok && offset >= 0 &&
 461               offset + ip_align >= 0 &&
 462               offset + ip_align % size == 0))) {
 463                 bool ldx_off_ok = offset <= S16_MAX;
 464
 465                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 466                 if (offset)
 467                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 468                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 469                                       size, 2 + endian + (!ldx_off_ok * 2));
 470                 if (ldx_off_ok) {
 471                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 472                                               BPF_REG_D, offset);
 473                 } else {
 474                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 475                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 476                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 477                                               BPF_REG_TMP, 0);
 478                 }
 479                 if (endian)
 480                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 481                 *insn++ = BPF_JMP_A(8);
 482         }
 483
 484         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 485         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 486         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 487         if (!indirect) {
 488                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 489         } else {
 490                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 491                 if (fp->k)
 492                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 493         }
 494
 495         switch (BPF_SIZE(fp->code)) {
 496         case BPF_B:
 497                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 498                 break;
 499         case BPF_H:
 500                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 501                 break;
 502         case BPF_W:
 503                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 504                 break;
 505         default:
 506                 return false;
 507         }
 508
 509         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 510         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 511         *insn   = BPF_EXIT_INSN();
 512
 513         *insnp = insn;
 514         return true;
 515 }
 516
 517 /**
 518  *      bpf_convert_filter - convert filter program
 519  *      @prog: the user passed filter program
 520  *      @len: the length of the user passed filter program
 521  *      @new_prog: allocated 'struct bpf_prog' or NULL
 522  *      @new_len: pointer to store length of converted program
 523  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 524  *
 525  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 526  * style extended BPF (eBPF).
 527  * Conversion workflow:
 528  *
 529  * 1) First pass for calculating the new program length:
 530  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 531  *
 532  * 2) 2nd pass to remap in two passes: 1st pass finds new
 533  *    jump offsets, 2nd pass remapping:
 534  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 535  */
 536 static int bpf_convert_filter(struct sock_filter *prog, int len,
 537                               struct bpf_prog *new_prog, int *new_len,
 538                               bool *seen_ld_abs)
 539 {
 540         int new_flen = 0, pass = 0, target, i, stack_off;
 541         struct bpf_insn *new_insn, *first_insn = NULL;
 542         struct sock_filter *fp;
 543         int *addrs = NULL;
 544         u8 bpf_src;
 545
 546         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 547         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 548
 549         if (len <= 0 || len > BPF_MAXINSNS)
 550                 return -EINVAL;
 551
 552         if (new_prog) {
 553                 first_insn = new_prog->insnsi;
 554                 addrs = kcalloc(len, sizeof(*addrs),
 555                                 GFP_KERNEL | __GFP_NOWARN);
 556                 if (!addrs)
 557                         return -ENOMEM;
 558         }
 559
 560 do_pass:
 561         new_insn = first_insn;
 562         fp = prog;
 563
 564         /* Classic BPF related prologue emission. */
 565         if (new_prog) {
 566                 /* Classic BPF expects A and X to be reset first. These need
 567                  * to be guaranteed to be the first two instructions.
 568                  */
 569                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 570                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 571
 572                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 573                  * In eBPF case it's done by the compiler, here we need to
 574                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 575                  */
 576                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 577                 if (*seen_ld_abs) {
 578                         /* For packet access in classic BPF, cache skb->data
 579                          * in callee-saved BPF R8 and skb->len - skb->data_len
 580                          * (headlen) in BPF R9. Since classic BPF is read-only
 581                          * on CTX, we only need to cache it once.
 582                          */
 583                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 584                                                   BPF_REG_D, BPF_REG_CTX,
 585                                                   offsetof(struct sk_buff, data));
 586                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 587                                                   offsetof(struct sk_buff, len));
 588                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 589                                                   offsetof(struct sk_buff, data_len));
 590                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 591                 }
 592         } else {
 593                 new_insn += 3;
 594         }
 595
 596         for (i = 0; i < len; fp++, i++) {
 597                 struct bpf_insn tmp_insns[32] = { };
 598                 struct bpf_insn *insn = tmp_insns;
 599
 600                 if (addrs)
 601                         addrs[i] = new_insn - first_insn;
 602
 603                 switch (fp->code) {
 604                 /* All arithmetic insns and skb loads map as-is. */
 605                 case BPF_ALU | BPF_ADD | BPF_X:
 606                 case BPF_ALU | BPF_ADD | BPF_K:
 607                 case BPF_ALU | BPF_SUB | BPF_X:
 608                 case BPF_ALU | BPF_SUB | BPF_K:
 609                 case BPF_ALU | BPF_AND | BPF_X:
 610                 case BPF_ALU | BPF_AND | BPF_K:
 611                 case BPF_ALU | BPF_OR | BPF_X:
 612                 case BPF_ALU | BPF_OR | BPF_K:
 613                 case BPF_ALU | BPF_LSH | BPF_X:
 614                 case BPF_ALU | BPF_LSH | BPF_K:
 615                 case BPF_ALU | BPF_RSH | BPF_X:
 616                 case BPF_ALU | BPF_RSH | BPF_K:
 617                 case BPF_ALU | BPF_XOR | BPF_X:
 618                 case BPF_ALU | BPF_XOR | BPF_K:
 619                 case BPF_ALU | BPF_MUL | BPF_X:
 620                 case BPF_ALU | BPF_MUL | BPF_K:
 621                 case BPF_ALU | BPF_DIV | BPF_X:
 622                 case BPF_ALU | BPF_DIV | BPF_K:
 623                 case BPF_ALU | BPF_MOD | BPF_X:
 624                 case BPF_ALU | BPF_MOD | BPF_K:
 625                 case BPF_ALU | BPF_NEG:
 626                 case BPF_LD | BPF_ABS | BPF_W:
 627                 case BPF_LD | BPF_ABS | BPF_H:
 628                 case BPF_LD | BPF_ABS | BPF_B:
 629                 case BPF_LD | BPF_IND | BPF_W:
 630                 case BPF_LD | BPF_IND | BPF_H:
 631                 case BPF_LD | BPF_IND | BPF_B:
 632                         /* Check for overloaded BPF extension and
 633                          * directly convert it if found, otherwise
 634                          * just move on with mapping.
 635                          */
 636                         if (BPF_CLASS(fp->code) == BPF_LD &&
 637                             BPF_MODE(fp->code) == BPF_ABS &&
 638                             convert_bpf_extensions(fp, &insn))
 639                                 break;
 640                         if (BPF_CLASS(fp->code) == BPF_LD &&
 641                             convert_bpf_ld_abs(fp, &insn)) {
 642                                 *seen_ld_abs = true;
 643                                 break;
 644                         }
 645
 646                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 647                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 648                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 649                                 /* Error with exception code on div/mod by 0.
 650                                  * For cBPF programs, this was always return 0.
 651                                  */
 652                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 653                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 654                                 *insn++ = BPF_EXIT_INSN();
 655                         }
 656
 657                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 658                         break;
 659
 660                 /* Jump transformation cannot use BPF block macros
 661                  * everywhere as offset calculation and target updates
 662                  * require a bit more work than the rest, i.e. jump
 663                  * opcodes map as-is, but offsets need adjustment.
 664                  */
 665
 666 #define BPF_EMIT_JMP                                                    \
 667         do {                                                            \
 668                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 669                 s32 off;                                                \
 670                                                                         \
 671                 if (target >= len || target < 0)                        \
 672                         goto err;                                       \
 673                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 674                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 675                 off -= insn - tmp_insns;                                \
 676                 /* Reject anything not fitting into insn->off. */       \
 677                 if (off < off_min || off > off_max)                     \
 678                         goto err;                                       \
 679                 insn->off = off;                                        \
 680         } while (0)
 681
 682                 case BPF_JMP | BPF_JA:
 683                         target = i + fp->k + 1;
 684                         insn->code = fp->code;
 685                         BPF_EMIT_JMP;
 686                         break;
 687
 688                 case BPF_JMP | BPF_JEQ | BPF_K:
 689                 case BPF_JMP | BPF_JEQ | BPF_X:
 690                 case BPF_JMP | BPF_JSET | BPF_K:
 691                 case BPF_JMP | BPF_JSET | BPF_X:
 692                 case BPF_JMP | BPF_JGT | BPF_K:
 693                 case BPF_JMP | BPF_JGT | BPF_X:
 694                 case BPF_JMP | BPF_JGE | BPF_K:
 695                 case BPF_JMP | BPF_JGE | BPF_X:
 696                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 697                                 /* BPF immediates are signed, zero extend
 698                                  * immediate into tmp register and use it
 699                                  * in compare insn.
 700                                  */
 701                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 702
 703                                 insn->dst_reg = BPF_REG_A;
 704                                 insn->src_reg = BPF_REG_TMP;
 705                                 bpf_src = BPF_X;
 706                         } else {
 707                                 insn->dst_reg = BPF_REG_A;
 708                                 insn->imm = fp->k;
 709                                 bpf_src = BPF_SRC(fp->code);
 710                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 711                         }
 712
 713                         /* Common case where 'jump_false' is next insn. */
 714                         if (fp->jf == 0) {
 715                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 716                                 target = i + fp->jt + 1;
 717                                 BPF_EMIT_JMP;
 718                                 break;
 719                         }
 720
 721                         /* Convert some jumps when 'jump_true' is next insn. */
 722                         if (fp->jt == 0) {
 723                                 switch (BPF_OP(fp->code)) {
 724                                 case BPF_JEQ:
 725                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 726                                         break;
 727                                 case BPF_JGT:
 728                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 729                                         break;
 730                                 case BPF_JGE:
 731                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 732                                         break;
 733                                 default:
 734                                         goto jmp_rest;
 735                                 }
 736
 737                                 target = i + fp->jf + 1;
 738                                 BPF_EMIT_JMP;
 739                                 break;
 740                         }
 741 jmp_rest:
 742                         /* Other jumps are mapped into two insns: Jxx and JA. */
 743                         target = i + fp->jt + 1;
 744                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 745                         BPF_EMIT_JMP;
 746                         insn++;
 747
 748                         insn->code = BPF_JMP | BPF_JA;
 749                         target = i + fp->jf + 1;
 750                         BPF_EMIT_JMP;
 751                         break;
 752
 753                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 754                 case BPF_LDX | BPF_MSH | BPF_B: {
 755                         struct sock_filter tmp = {
 756                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 757                                 .k      = fp->k,
 758                         };
 759
 760                         *seen_ld_abs = true;
 761
 762                         /* X = A */
 763                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 764                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 765                         convert_bpf_ld_abs(&tmp, &insn);
 766                         insn++;
 767                         /* A &= 0xf */
 768                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 769                         /* A <<= 2 */
 770                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 771                         /* tmp = X */
 772                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 773                         /* X = A */
 774                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 775                         /* A = tmp */
 776                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 777                         break;
 778                 }
 779                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 780                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 781                  */
 782                 case BPF_RET | BPF_A:
 783                 case BPF_RET | BPF_K:
 784                         if (BPF_RVAL(fp->code) == BPF_K)
 785                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 786                                                         0, fp->k);
 787                         *insn = BPF_EXIT_INSN();
 788                         break;
 789
 790                 /* Store to stack. */
 791                 case BPF_ST:
 792                 case BPF_STX:
 793                         stack_off = fp->k * 4  + 4;
 794                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 795                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 796                                             -stack_off);
 797                         /* check_load_and_stores() verifies that classic BPF can
 798                          * load from stack only after write, so tracking
 799                          * stack_depth for ST|STX insns is enough
 800                          */
 801                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 802                                 new_prog->aux->stack_depth = stack_off;
 803                         break;
 804
 805                 /* Load from stack. */
 806                 case BPF_LD | BPF_MEM:
 807                 case BPF_LDX | BPF_MEM:
 808                         stack_off = fp->k * 4  + 4;
 809                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 810                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 811                                             -stack_off);
 812                         break;
 813
 814                 /* A = K or X = K */
 815                 case BPF_LD | BPF_IMM:
 816                 case BPF_LDX | BPF_IMM:
 817                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 818                                               BPF_REG_A : BPF_REG_X, fp->k);
 819                         break;
 820
 821                 /* X = A */
 822                 case BPF_MISC | BPF_TAX:
 823                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 824                         break;
 825
 826                 /* A = X */
 827                 case BPF_MISC | BPF_TXA:
 828                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 829                         break;
 830
 831                 /* A = skb->len or X = skb->len */
 832                 case BPF_LD | BPF_W | BPF_LEN:
 833                 case BPF_LDX | BPF_W | BPF_LEN:
 834                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 835                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 836                                             offsetof(struct sk_buff, len));
 837                         break;
 838
 839                 /* Access seccomp_data fields. */
 840                 case BPF_LDX | BPF_ABS | BPF_W:
 841                         /* A = *(u32 *) (ctx + K) */
 842                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 843                         break;
 844
 845                 /* Unknown instruction. */
 846                 default:
 847                         goto err;
 848                 }
 849
 850                 insn++;
 851                 if (new_prog)
 852                         memcpy(new_insn, tmp_insns,
 853                                sizeof(*insn) * (insn - tmp_insns));
 854                 new_insn += insn - tmp_insns;
 855         }
 856
 857         if (!new_prog) {
 858                 /* Only calculating new length. */
 859                 *new_len = new_insn - first_insn;
 860                 if (*seen_ld_abs)
 861                         *new_len += 4; /* Prologue bits. */
 862                 return 0;
 863         }
 864
 865         pass++;
 866         if (new_flen != new_insn - first_insn) {
 867                 new_flen = new_insn - first_insn;
 868                 if (pass > 2)
 869                         goto err;
 870                 goto do_pass;
 871         }
 872
 873         kfree(addrs);
 874         BUG_ON(*new_len != new_flen);
 875         return 0;
 876 err:
 877         kfree(addrs);
 878         return -EINVAL;
 879 }
 880
 881 /* Security:
 882  *
 883  * As we dont want to clear mem[] array for each packet going through
 884  * __bpf_prog_run(), we check that filter loaded by user never try to read
 885  * a cell if not previously written, and we check all branches to be sure
 886  * a malicious user doesn't try to abuse us.
 887  */
 888 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 889 {
 890         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 891         int pc, ret = 0;
 892
 893         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 894
 895         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 896         if (!masks)
 897                 return -ENOMEM;
 898
 899         memset(masks, 0xff, flen * sizeof(*masks));
 900
 901         for (pc = 0; pc < flen; pc++) {
 902                 memvalid &= masks[pc];
 903
 904                 switch (filter[pc].code) {
 905                 case BPF_ST:
 906                 case BPF_STX:
 907                         memvalid |= (1 << filter[pc].k);
 908                         break;
 909                 case BPF_LD | BPF_MEM:
 910                 case BPF_LDX | BPF_MEM:
 911                         if (!(memvalid & (1 << filter[pc].k))) {
 912                                 ret = -EINVAL;
 913                                 goto error;
 914                         }
 915                         break;
 916                 case BPF_JMP | BPF_JA:
 917                         /* A jump must set masks on target */
 918                         masks[pc + 1 + filter[pc].k] &= memvalid;
 919                         memvalid = ~0;
 920                         break;
 921                 case BPF_JMP | BPF_JEQ | BPF_K:
 922                 case BPF_JMP | BPF_JEQ | BPF_X:
 923                 case BPF_JMP | BPF_JGE | BPF_K:
 924                 case BPF_JMP | BPF_JGE | BPF_X:
 925                 case BPF_JMP | BPF_JGT | BPF_K:
 926                 case BPF_JMP | BPF_JGT | BPF_X:
 927                 case BPF_JMP | BPF_JSET | BPF_K:
 928                 case BPF_JMP | BPF_JSET | BPF_X:
 929                         /* A jump must set masks on targets */
 930                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 931                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 932                         memvalid = ~0;
 933                         break;
 934                 }
 935         }
 936 error:
 937         kfree(masks);
 938         return ret;
 939 }
 940
 941 static bool chk_code_allowed(u16 code_to_probe)
 942 {
 943         static const bool codes[] = {
 944                 /* 32 bit ALU operations */
 945                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 946                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 947                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 948                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 949                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 950                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 951                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 952                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 953                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 954                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 955                 [BPF_ALU | BPF_AND | BPF_K] = true,
 956                 [BPF_ALU | BPF_AND | BPF_X] = true,
 957                 [BPF_ALU | BPF_OR | BPF_K] = true,
 958                 [BPF_ALU | BPF_OR | BPF_X] = true,
 959                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 960                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 961                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 962                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 963                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 964                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 965                 [BPF_ALU | BPF_NEG] = true,
 966                 /* Load instructions */
 967                 [BPF_LD | BPF_W | BPF_ABS] = true,
 968                 [BPF_LD | BPF_H | BPF_ABS] = true,
 969                 [BPF_LD | BPF_B | BPF_ABS] = true,
 970                 [BPF_LD | BPF_W | BPF_LEN] = true,
 971                 [BPF_LD | BPF_W | BPF_IND] = true,
 972                 [BPF_LD | BPF_H | BPF_IND] = true,
 973                 [BPF_LD | BPF_B | BPF_IND] = true,
 974                 [BPF_LD | BPF_IMM] = true,
 975                 [BPF_LD | BPF_MEM] = true,
 976                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 977                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 978                 [BPF_LDX | BPF_IMM] = true,
 979                 [BPF_LDX | BPF_MEM] = true,
 980                 /* Store instructions */
 981                 [BPF_ST] = true,
 982                 [BPF_STX] = true,
 983                 /* Misc instructions */
 984                 [BPF_MISC | BPF_TAX] = true,
 985                 [BPF_MISC | BPF_TXA] = true,
 986                 /* Return instructions */
 987                 [BPF_RET | BPF_K] = true,
 988                 [BPF_RET | BPF_A] = true,
 989                 /* Jump instructions */
 990                 [BPF_JMP | BPF_JA] = true,
 991                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
 992                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
 993                 [BPF_JMP | BPF_JGE | BPF_K] = true,
 994                 [BPF_JMP | BPF_JGE | BPF_X] = true,
 995                 [BPF_JMP | BPF_JGT | BPF_K] = true,
 996                 [BPF_JMP | BPF_JGT | BPF_X] = true,
 997                 [BPF_JMP | BPF_JSET | BPF_K] = true,
 998                 [BPF_JMP | BPF_JSET | BPF_X] = true,
 999         };
1000
1001         if (code_to_probe >= ARRAY_SIZE(codes))
1002                 return false;
1003
1004         return codes[code_to_probe];
1005 }
1006
1007 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1008                                 unsigned int flen)
1009 {
1010         if (filter == NULL)
1011                 return false;
1012         if (flen == 0 || flen > BPF_MAXINSNS)
1013                 return false;
1014
1015         return true;
1016 }
1017
1018 /**
1019  *      bpf_check_classic - verify socket filter code
1020  *      @filter: filter to verify
1021  *      @flen: length of filter
1022  *
1023  * Check the user's filter code. If we let some ugly
1024  * filter code slip through kaboom! The filter must contain
1025  * no references or jumps that are out of range, no illegal
1026  * instructions, and must end with a RET instruction.
1027  *
1028  * All jumps are forward as they are not signed.
1029  *
1030  * Returns 0 if the rule set is legal or -EINVAL if not.
1031  */
1032 static int bpf_check_classic(const struct sock_filter *filter,
1033                              unsigned int flen)
1034 {
1035         bool anc_found;
1036         int pc;
1037
1038         /* Check the filter code now */
1039         for (pc = 0; pc < flen; pc++) {
1040                 const struct sock_filter *ftest = &filter[pc];
1041
1042                 /* May we actually operate on this code? */
1043                 if (!chk_code_allowed(ftest->code))
1044                         return -EINVAL;
1045
1046                 /* Some instructions need special checks */
1047                 switch (ftest->code) {
1048                 case BPF_ALU | BPF_DIV | BPF_K:
1049                 case BPF_ALU | BPF_MOD | BPF_K:
1050                         /* Check for division by zero */
1051                         if (ftest->k == 0)
1052                                 return -EINVAL;
1053                         break;
1054                 case BPF_ALU | BPF_LSH | BPF_K:
1055                 case BPF_ALU | BPF_RSH | BPF_K:
1056                         if (ftest->k >= 32)
1057                                 return -EINVAL;
1058                         break;
1059                 case BPF_LD | BPF_MEM:
1060                 case BPF_LDX | BPF_MEM:
1061                 case BPF_ST:
1062                 case BPF_STX:
1063                         /* Check for invalid memory addresses */
1064                         if (ftest->k >= BPF_MEMWORDS)
1065                                 return -EINVAL;
1066                         break;
1067                 case BPF_JMP | BPF_JA:
1068                         /* Note, the large ftest->k might cause loops.
1069                          * Compare this with conditional jumps below,
1070                          * where offsets are limited. --ANK (981016)
1071                          */
1072                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1073                                 return -EINVAL;
1074                         break;
1075                 case BPF_JMP | BPF_JEQ | BPF_K:
1076                 case BPF_JMP | BPF_JEQ | BPF_X:
1077                 case BPF_JMP | BPF_JGE | BPF_K:
1078                 case BPF_JMP | BPF_JGE | BPF_X:
1079                 case BPF_JMP | BPF_JGT | BPF_K:
1080                 case BPF_JMP | BPF_JGT | BPF_X:
1081                 case BPF_JMP | BPF_JSET | BPF_K:
1082                 case BPF_JMP | BPF_JSET | BPF_X:
1083                         /* Both conditionals must be safe */
1084                         if (pc + ftest->jt + 1 >= flen ||
1085                             pc + ftest->jf + 1 >= flen)
1086                                 return -EINVAL;
1087                         break;
1088                 case BPF_LD | BPF_W | BPF_ABS:
1089                 case BPF_LD | BPF_H | BPF_ABS:
1090                 case BPF_LD | BPF_B | BPF_ABS:
1091                         anc_found = false;
1092                         if (bpf_anc_helper(ftest) & BPF_ANC)
1093                                 anc_found = true;
1094                         /* Ancillary operation unknown or unsupported */
1095                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1096                                 return -EINVAL;
1097                 }
1098         }
1099
1100         /* Last instruction must be a RET code */
1101         switch (filter[flen - 1].code) {
1102         case BPF_RET | BPF_K:
1103         case BPF_RET | BPF_A:
1104                 return check_load_and_stores(filter, flen);
1105         }
1106
1107         return -EINVAL;
1108 }
1109
1110 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1111                                       const struct sock_fprog *fprog)
1112 {
1113         unsigned int fsize = bpf_classic_proglen(fprog);
1114         struct sock_fprog_kern *fkprog;
1115
1116         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1117         if (!fp->orig_prog)
1118                 return -ENOMEM;
1119
1120         fkprog = fp->orig_prog;
1121         fkprog->len = fprog->len;
1122
1123         fkprog->filter = kmemdup(fp->insns, fsize,
1124                                  GFP_KERNEL | __GFP_NOWARN);
1125         if (!fkprog->filter) {
1126                 kfree(fp->orig_prog);
1127                 return -ENOMEM;
1128         }
1129
1130         return 0;
1131 }
1132
1133 static void bpf_release_orig_filter(struct bpf_prog *fp)
1134 {
1135         struct sock_fprog_kern *fprog = fp->orig_prog;
1136
1137         if (fprog) {
1138                 kfree(fprog->filter);
1139                 kfree(fprog);
1140         }
1141 }
1142
1143 static void __bpf_prog_release(struct bpf_prog *prog)
1144 {
1145         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1146                 bpf_prog_put(prog);
1147         } else {
1148                 bpf_release_orig_filter(prog);
1149                 bpf_prog_free(prog);
1150         }
1151 }
1152
1153 static void __sk_filter_release(struct sk_filter *fp)
1154 {
1155         __bpf_prog_release(fp->prog);
1156         kfree(fp);
1157 }
1158
1159 /**
1160  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1161  *      @rcu: rcu_head that contains the sk_filter to free
1162  */
1163 static void sk_filter_release_rcu(struct rcu_head *rcu)
1164 {
1165         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1166
1167         __sk_filter_release(fp);
1168 }
1169
1170 /**
1171  *      sk_filter_release - release a socket filter
1172  *      @fp: filter to remove
1173  *
1174  *      Remove a filter from a socket and release its resources.
1175  */
1176 static void sk_filter_release(struct sk_filter *fp)
1177 {
1178         if (refcount_dec_and_test(&fp->refcnt))
1179                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1180 }
1181
1182 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1183 {
1184         u32 filter_size = bpf_prog_size(fp->prog->len);
1185
1186         atomic_sub(filter_size, &sk->sk_omem_alloc);
1187         sk_filter_release(fp);
1188 }
1189
1190 /* try to charge the socket memory if there is space available
1191  * return true on success
1192  */
1193 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1194 {
1195         u32 filter_size = bpf_prog_size(fp->prog->len);
1196
1197         /* same check as in sock_kmalloc() */
1198         if (filter_size <= sysctl_optmem_max &&
1199             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1200                 atomic_add(filter_size, &sk->sk_omem_alloc);
1201                 return true;
1202         }
1203         return false;
1204 }
1205
1206 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1207 {
1208         if (!refcount_inc_not_zero(&fp->refcnt))
1209                 return false;
1210
1211         if (!__sk_filter_charge(sk, fp)) {
1212                 sk_filter_release(fp);
1213                 return false;
1214         }
1215         return true;
1216 }
1217
1218 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1219 {
1220         struct sock_filter *old_prog;
1221         struct bpf_prog *old_fp;
1222         int err, new_len, old_len = fp->len;
1223         bool seen_ld_abs = false;
1224
1225         /* We are free to overwrite insns et al right here as it
1226          * won't be used at this point in time anymore internally
1227          * after the migration to the internal BPF instruction
1228          * representation.
1229          */
1230         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1231                      sizeof(struct bpf_insn));
1232
1233         /* Conversion cannot happen on overlapping memory areas,
1234          * so we need to keep the user BPF around until the 2nd
1235          * pass. At this time, the user BPF is stored in fp->insns.
1236          */
1237         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1238                            GFP_KERNEL | __GFP_NOWARN);
1239         if (!old_prog) {
1240                 err = -ENOMEM;
1241                 goto out_err;
1242         }
1243
1244         /* 1st pass: calculate the new program length. */
1245         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1246                                  &seen_ld_abs);
1247         if (err)
1248                 goto out_err_free;
1249
1250         /* Expand fp for appending the new filter representation. */
1251         old_fp = fp;
1252         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1253         if (!fp) {
1254                 /* The old_fp is still around in case we couldn't
1255                  * allocate new memory, so uncharge on that one.
1256                  */
1257                 fp = old_fp;
1258                 err = -ENOMEM;
1259                 goto out_err_free;
1260         }
1261
1262         fp->len = new_len;
1263
1264         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1265         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1266                                  &seen_ld_abs);
1267         if (err)
1268                 /* 2nd bpf_convert_filter() can fail only if it fails
1269                  * to allocate memory, remapping must succeed. Note,
1270                  * that at this time old_fp has already been released
1271                  * by krealloc().
1272                  */
1273                 goto out_err_free;
1274
1275         fp = bpf_prog_select_runtime(fp, &err);
1276         if (err)
1277                 goto out_err_free;
1278
1279         kfree(old_prog);
1280         return fp;
1281
1282 out_err_free:
1283         kfree(old_prog);
1284 out_err:
1285         __bpf_prog_release(fp);
1286         return ERR_PTR(err);
1287 }
1288
1289 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1290                                            bpf_aux_classic_check_t trans)
1291 {
1292         int err;
1293
1294         fp->bpf_func = NULL;
1295         fp->jited = 0;
1296
1297         err = bpf_check_classic(fp->insns, fp->len);
1298         if (err) {
1299                 __bpf_prog_release(fp);
1300                 return ERR_PTR(err);
1301         }
1302
1303         /* There might be additional checks and transformations
1304          * needed on classic filters, f.e. in case of seccomp.
1305          */
1306         if (trans) {
1307                 err = trans(fp->insns, fp->len);
1308                 if (err) {
1309                         __bpf_prog_release(fp);
1310                         return ERR_PTR(err);
1311                 }
1312         }
1313
1314         /* Probe if we can JIT compile the filter and if so, do
1315          * the compilation of the filter.
1316          */
1317         bpf_jit_compile(fp);
1318
1319         /* JIT compiler couldn't process this filter, so do the
1320          * internal BPF translation for the optimized interpreter.
1321          */
1322         if (!fp->jited)
1323                 fp = bpf_migrate_filter(fp);
1324
1325         return fp;
1326 }
1327
1328 /**
1329  *      bpf_prog_create - create an unattached filter
1330  *      @pfp: the unattached filter that is created
1331  *      @fprog: the filter program
1332  *
1333  * Create a filter independent of any socket. We first run some
1334  * sanity checks on it to make sure it does not explode on us later.
1335  * If an error occurs or there is insufficient memory for the filter
1336  * a negative errno code is returned. On success the return is zero.
1337  */
1338 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1339 {
1340         unsigned int fsize = bpf_classic_proglen(fprog);
1341         struct bpf_prog *fp;
1342
1343         /* Make sure new filter is there and in the right amounts. */
1344         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1345                 return -EINVAL;
1346
1347         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1348         if (!fp)
1349                 return -ENOMEM;
1350
1351         memcpy(fp->insns, fprog->filter, fsize);
1352
1353         fp->len = fprog->len;
1354         /* Since unattached filters are not copied back to user
1355          * space through sk_get_filter(), we do not need to hold
1356          * a copy here, and can spare us the work.
1357          */
1358         fp->orig_prog = NULL;
1359
1360         /* bpf_prepare_filter() already takes care of freeing
1361          * memory in case something goes wrong.
1362          */
1363         fp = bpf_prepare_filter(fp, NULL);
1364         if (IS_ERR(fp))
1365                 return PTR_ERR(fp);
1366
1367         *pfp = fp;
1368         return 0;
1369 }
1370 EXPORT_SYMBOL_GPL(bpf_prog_create);
1371
1372 /**
1373  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1374  *      @pfp: the unattached filter that is created
1375  *      @fprog: the filter program
1376  *      @trans: post-classic verifier transformation handler
1377  *      @save_orig: save classic BPF program
1378  *
1379  * This function effectively does the same as bpf_prog_create(), only
1380  * that it builds up its insns buffer from user space provided buffer.
1381  * It also allows for passing a bpf_aux_classic_check_t handler.
1382  */
1383 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1384                               bpf_aux_classic_check_t trans, bool save_orig)
1385 {
1386         unsigned int fsize = bpf_classic_proglen(fprog);
1387         struct bpf_prog *fp;
1388         int err;
1389
1390         /* Make sure new filter is there and in the right amounts. */
1391         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1392                 return -EINVAL;
1393
1394         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1395         if (!fp)
1396                 return -ENOMEM;
1397
1398         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1399                 __bpf_prog_free(fp);
1400                 return -EFAULT;
1401         }
1402
1403         fp->len = fprog->len;
1404         fp->orig_prog = NULL;
1405
1406         if (save_orig) {
1407                 err = bpf_prog_store_orig_filter(fp, fprog);
1408                 if (err) {
1409                         __bpf_prog_free(fp);
1410                         return -ENOMEM;
1411                 }
1412         }
1413
1414         /* bpf_prepare_filter() already takes care of freeing
1415          * memory in case something goes wrong.
1416          */
1417         fp = bpf_prepare_filter(fp, trans);
1418         if (IS_ERR(fp))
1419                 return PTR_ERR(fp);
1420
1421         *pfp = fp;
1422         return 0;
1423 }
1424 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1425
1426 void bpf_prog_destroy(struct bpf_prog *fp)
1427 {
1428         __bpf_prog_release(fp);
1429 }
1430 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1431
1432 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1433 {
1434         struct sk_filter *fp, *old_fp;
1435
1436         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1437         if (!fp)
1438                 return -ENOMEM;
1439
1440         fp->prog = prog;
1441
1442         if (!__sk_filter_charge(sk, fp)) {
1443                 kfree(fp);
1444                 return -ENOMEM;
1445         }
1446         refcount_set(&fp->refcnt, 1);
1447
1448         old_fp = rcu_dereference_protected(sk->sk_filter,
1449                                            lockdep_sock_is_held(sk));
1450         rcu_assign_pointer(sk->sk_filter, fp);
1451
1452         if (old_fp)
1453                 sk_filter_uncharge(sk, old_fp);
1454
1455         return 0;
1456 }
1457
1458 static
1459 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1460 {
1461         unsigned int fsize = bpf_classic_proglen(fprog);
1462         struct bpf_prog *prog;
1463         int err;
1464
1465         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1466                 return ERR_PTR(-EPERM);
1467
1468         /* Make sure new filter is there and in the right amounts. */
1469         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1470                 return ERR_PTR(-EINVAL);
1471
1472         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1473         if (!prog)
1474                 return ERR_PTR(-ENOMEM);
1475
1476         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1477                 __bpf_prog_free(prog);
1478                 return ERR_PTR(-EFAULT);
1479         }
1480
1481         prog->len = fprog->len;
1482
1483         err = bpf_prog_store_orig_filter(prog, fprog);
1484         if (err) {
1485                 __bpf_prog_free(prog);
1486                 return ERR_PTR(-ENOMEM);
1487         }
1488
1489         /* bpf_prepare_filter() already takes care of freeing
1490          * memory in case something goes wrong.
1491          */
1492         return bpf_prepare_filter(prog, NULL);
1493 }
1494
1495 /**
1496  *      sk_attach_filter - attach a socket filter
1497  *      @fprog: the filter program
1498  *      @sk: the socket to use
1499  *
1500  * Attach the user's filter code. We first run some sanity checks on
1501  * it to make sure it does not explode on us later. If an error
1502  * occurs or there is insufficient memory for the filter a negative
1503  * errno code is returned. On success the return is zero.
1504  */
1505 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1506 {
1507         struct bpf_prog *prog = __get_filter(fprog, sk);
1508         int err;
1509
1510         if (IS_ERR(prog))
1511                 return PTR_ERR(prog);
1512
1513         err = __sk_attach_prog(prog, sk);
1514         if (err < 0) {
1515                 __bpf_prog_release(prog);
1516                 return err;
1517         }
1518
1519         return 0;
1520 }
1521 EXPORT_SYMBOL_GPL(sk_attach_filter);
1522
1523 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1524 {
1525         struct bpf_prog *prog = __get_filter(fprog, sk);
1526         int err;
1527
1528         if (IS_ERR(prog))
1529                 return PTR_ERR(prog);
1530
1531         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1532                 err = -ENOMEM;
1533         else
1534                 err = reuseport_attach_prog(sk, prog);
1535
1536         if (err)
1537                 __bpf_prog_release(prog);
1538
1539         return err;
1540 }
1541
1542 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1543 {
1544         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1545                 return ERR_PTR(-EPERM);
1546
1547         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1548 }
1549
1550 int sk_attach_bpf(u32 ufd, struct sock *sk)
1551 {
1552         struct bpf_prog *prog = __get_bpf(ufd, sk);
1553         int err;
1554
1555         if (IS_ERR(prog))
1556                 return PTR_ERR(prog);
1557
1558         err = __sk_attach_prog(prog, sk);
1559         if (err < 0) {
1560                 bpf_prog_put(prog);
1561                 return err;
1562         }
1563
1564         return 0;
1565 }
1566
1567 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1568 {
1569         struct bpf_prog *prog;
1570         int err;
1571
1572         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1573                 return -EPERM;
1574
1575         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1576         if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1577                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1578         if (IS_ERR(prog))
1579                 return PTR_ERR(prog);
1580
1581         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1582                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1583                  * bpf prog (e.g. sockmap).  It depends on the
1584                  * limitation imposed by bpf_prog_load().
1585                  * Hence, sysctl_optmem_max is not checked.
1586                  */
1587                 if ((sk->sk_type != SOCK_STREAM &&
1588                      sk->sk_type != SOCK_DGRAM) ||
1589                     (sk->sk_protocol != IPPROTO_UDP &&
1590                      sk->sk_protocol != IPPROTO_TCP) ||
1591                     (sk->sk_family != AF_INET &&
1592                      sk->sk_family != AF_INET6)) {
1593                         err = -ENOTSUPP;
1594                         goto err_prog_put;
1595                 }
1596         } else {
1597                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1598                 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1599                         err = -ENOMEM;
1600                         goto err_prog_put;
1601                 }
1602         }
1603
1604         err = reuseport_attach_prog(sk, prog);
1605 err_prog_put:
1606         if (err)
1607                 bpf_prog_put(prog);
1608
1609         return err;
1610 }
1611
1612 void sk_reuseport_prog_free(struct bpf_prog *prog)
1613 {
1614         if (!prog)
1615                 return;
1616
1617         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1618                 bpf_prog_put(prog);
1619         else
1620                 bpf_prog_destroy(prog);
1621 }
1622
1623 struct bpf_scratchpad {
1624         union {
1625                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1626                 u8     buff[MAX_BPF_STACK];
1627         };
1628 };
1629
1630 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1631
1632 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1633                                           unsigned int write_len)
1634 {
1635         return skb_ensure_writable(skb, write_len);
1636 }
1637
1638 static inline int bpf_try_make_writable(struct sk_buff *skb,
1639                                         unsigned int write_len)
1640 {
1641         int err = __bpf_try_make_writable(skb, write_len);
1642
1643         bpf_compute_data_pointers(skb);
1644         return err;
1645 }
1646
1647 static int bpf_try_make_head_writable(struct sk_buff *skb)
1648 {
1649         return bpf_try_make_writable(skb, skb_headlen(skb));
1650 }
1651
1652 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1653 {
1654         if (skb_at_tc_ingress(skb))
1655                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1656 }
1657
1658 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1659 {
1660         if (skb_at_tc_ingress(skb))
1661                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1662 }
1663
1664 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1665            const void *, from, u32, len, u64, flags)
1666 {
1667         void *ptr;
1668
1669         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1670                 return -EINVAL;
1671         if (unlikely(offset > 0xffff))
1672                 return -EFAULT;
1673         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1674                 return -EFAULT;
1675
1676         ptr = skb->data + offset;
1677         if (flags & BPF_F_RECOMPUTE_CSUM)
1678                 __skb_postpull_rcsum(skb, ptr, len, offset);
1679
1680         memcpy(ptr, from, len);
1681
1682         if (flags & BPF_F_RECOMPUTE_CSUM)
1683                 __skb_postpush_rcsum(skb, ptr, len, offset);
1684         if (flags & BPF_F_INVALIDATE_HASH)
1685                 skb_clear_hash(skb);
1686
1687         return 0;
1688 }
1689
1690 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1691         .func           = bpf_skb_store_bytes,
1692         .gpl_only       = false,
1693         .ret_type       = RET_INTEGER,
1694         .arg1_type      = ARG_PTR_TO_CTX,
1695         .arg2_type      = ARG_ANYTHING,
1696         .arg3_type      = ARG_PTR_TO_MEM,
1697         .arg4_type      = ARG_CONST_SIZE,
1698         .arg5_type      = ARG_ANYTHING,
1699 };
1700
1701 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1702            void *, to, u32, len)
1703 {
1704         void *ptr;
1705
1706         if (unlikely(offset > 0xffff))
1707                 goto err_clear;
1708
1709         ptr = skb_header_pointer(skb, offset, len, to);
1710         if (unlikely(!ptr))
1711                 goto err_clear;
1712         if (ptr != to)
1713                 memcpy(to, ptr, len);
1714
1715         return 0;
1716 err_clear:
1717         memset(to, 0, len);
1718         return -EFAULT;
1719 }
1720
1721 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1722         .func           = bpf_skb_load_bytes,
1723         .gpl_only       = false,
1724         .ret_type       = RET_INTEGER,
1725         .arg1_type      = ARG_PTR_TO_CTX,
1726         .arg2_type      = ARG_ANYTHING,
1727         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1728         .arg4_type      = ARG_CONST_SIZE,
1729 };
1730
1731 BPF_CALL_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
2059         dev_xmit_recursion_inc();
2060         ret = dev_queue_xmit(skb);
2061         dev_xmit_recursion_dec();
2062
2063         return ret;
2064 }
2065
2066 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2067                                  u32 flags)
2068 {
2069         unsigned int mlen = skb_network_offset(skb);
2070
2071         if (mlen) {
2072                 __skb_pull(skb, mlen);
2073
2074                 /* At ingress, the mac header has already been pulled once.
2075                  * At egress, skb_pospull_rcsum has to be done in case that
2076                  * the skb is originated from ingress (i.e. a forwarded skb)
2077                  * to ensure that rcsum starts at net header.
2078                  */
2079                 if (!skb_at_tc_ingress(skb))
2080                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2081         }
2082         skb_pop_mac_header(skb);
2083         skb_reset_mac_len(skb);
2084         return flags & BPF_F_INGRESS ?
2085                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2086 }
2087
2088 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2089                                  u32 flags)
2090 {
2091         /* Verify that a link layer header is carried */
2092         if (unlikely(skb->mac_header >= skb->network_header)) {
2093                 kfree_skb(skb);
2094                 return -ERANGE;
2095         }
2096
2097         bpf_push_mac_rcsum(skb);
2098         return flags & BPF_F_INGRESS ?
2099                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2100 }
2101
2102 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2103                           u32 flags)
2104 {
2105         if (dev_is_mac_header_xmit(dev))
2106                 return __bpf_redirect_common(skb, dev, flags);
2107         else
2108                 return __bpf_redirect_no_mac(skb, dev, flags);
2109 }
2110
2111 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2112 {
2113         struct net_device *dev;
2114         struct sk_buff *clone;
2115         int ret;
2116
2117         if (unlikely(flags & ~(BPF_F_INGRESS)))
2118                 return -EINVAL;
2119
2120         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2121         if (unlikely(!dev))
2122                 return -EINVAL;
2123
2124         clone = skb_clone(skb, GFP_ATOMIC);
2125         if (unlikely(!clone))
2126                 return -ENOMEM;
2127
2128         /* For direct write, we need to keep the invariant that the skbs
2129          * we're dealing with need to be uncloned. Should uncloning fail
2130          * here, we need to free the just generated clone to unclone once
2131          * again.
2132          */
2133         ret = bpf_try_make_head_writable(skb);
2134         if (unlikely(ret)) {
2135                 kfree_skb(clone);
2136                 return -ENOMEM;
2137         }
2138
2139         return __bpf_redirect(clone, dev, flags);
2140 }
2141
2142 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2143         .func           = bpf_clone_redirect,
2144         .gpl_only       = false,
2145         .ret_type       = RET_INTEGER,
2146         .arg1_type      = ARG_PTR_TO_CTX,
2147         .arg2_type      = ARG_ANYTHING,
2148         .arg3_type      = ARG_ANYTHING,
2149 };
2150
2151 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2152 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2153
2154 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2155 {
2156         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2157
2158         if (unlikely(flags & ~(BPF_F_INGRESS)))
2159                 return TC_ACT_SHOT;
2160
2161         ri->flags = flags;
2162         ri->tgt_index = ifindex;
2163
2164         return TC_ACT_REDIRECT;
2165 }
2166
2167 int skb_do_redirect(struct sk_buff *skb)
2168 {
2169         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2170         struct net_device *dev;
2171
2172         dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index);
2173         ri->tgt_index = 0;
2174         if (unlikely(!dev)) {
2175                 kfree_skb(skb);
2176                 return -EINVAL;
2177         }
2178
2179         return __bpf_redirect(skb, dev, ri->flags);
2180 }
2181
2182 static const struct bpf_func_proto bpf_redirect_proto = {
2183         .func           = bpf_redirect,
2184         .gpl_only       = false,
2185         .ret_type       = RET_INTEGER,
2186         .arg1_type      = ARG_ANYTHING,
2187         .arg2_type      = ARG_ANYTHING,
2188 };
2189
2190 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2191 {
2192         msg->apply_bytes = bytes;
2193         return 0;
2194 }
2195
2196 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2197         .func           = bpf_msg_apply_bytes,
2198         .gpl_only       = false,
2199         .ret_type       = RET_INTEGER,
2200         .arg1_type      = ARG_PTR_TO_CTX,
2201         .arg2_type      = ARG_ANYTHING,
2202 };
2203
2204 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2205 {
2206         msg->cork_bytes = bytes;
2207         return 0;
2208 }
2209
2210 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2211         .func           = bpf_msg_cork_bytes,
2212         .gpl_only       = false,
2213         .ret_type       = RET_INTEGER,
2214         .arg1_type      = ARG_PTR_TO_CTX,
2215         .arg2_type      = ARG_ANYTHING,
2216 };
2217
2218 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2219            u32, end, u64, flags)
2220 {
2221         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2222         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2223         struct scatterlist *sge;
2224         u8 *raw, *to, *from;
2225         struct page *page;
2226
2227         if (unlikely(flags || end <= start))
2228                 return -EINVAL;
2229
2230         /* First find the starting scatterlist element */
2231         i = msg->sg.start;
2232         do {
2233                 len = sk_msg_elem(msg, i)->length;
2234                 if (start < offset + len)
2235                         break;
2236                 offset += len;
2237                 sk_msg_iter_var_next(i);
2238         } while (i != msg->sg.end);
2239
2240         if (unlikely(start >= offset + len))
2241                 return -EINVAL;
2242
2243         first_sge = i;
2244         /* The start may point into the sg element so we need to also
2245          * account for the headroom.
2246          */
2247         bytes_sg_total = start - offset + bytes;
2248         if (!msg->sg.copy[i] && bytes_sg_total <= len)
2249                 goto out;
2250
2251         /* At this point we need to linearize multiple scatterlist
2252          * elements or a single shared page. Either way we need to
2253          * copy into a linear buffer exclusively owned by BPF. Then
2254          * place the buffer in the scatterlist and fixup the original
2255          * entries by removing the entries now in the linear buffer
2256          * and shifting the remaining entries. For now we do not try
2257          * to copy partial entries to avoid complexity of running out
2258          * of sg_entry slots. The downside is reading a single byte
2259          * will copy the entire sg entry.
2260          */
2261         do {
2262                 copy += sk_msg_elem(msg, i)->length;
2263                 sk_msg_iter_var_next(i);
2264                 if (bytes_sg_total <= copy)
2265                         break;
2266         } while (i != msg->sg.end);
2267         last_sge = i;
2268
2269         if (unlikely(bytes_sg_total > copy))
2270                 return -EINVAL;
2271
2272         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2273                            get_order(copy));
2274         if (unlikely(!page))
2275                 return -ENOMEM;
2276
2277         raw = page_address(page);
2278         i = first_sge;
2279         do {
2280                 sge = sk_msg_elem(msg, i);
2281                 from = sg_virt(sge);
2282                 len = sge->length;
2283                 to = raw + poffset;
2284
2285                 memcpy(to, from, len);
2286                 poffset += len;
2287                 sge->length = 0;
2288                 put_page(sg_page(sge));
2289
2290                 sk_msg_iter_var_next(i);
2291         } while (i != last_sge);
2292
2293         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2294
2295         /* To repair sg ring we need to shift entries. If we only
2296          * had a single entry though we can just replace it and
2297          * be done. Otherwise walk the ring and shift the entries.
2298          */
2299         WARN_ON_ONCE(last_sge == first_sge);
2300         shift = last_sge > first_sge ?
2301                 last_sge - first_sge - 1 :
2302                 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2303         if (!shift)
2304                 goto out;
2305
2306         i = first_sge;
2307         sk_msg_iter_var_next(i);
2308         do {
2309                 u32 move_from;
2310
2311                 if (i + shift >= MAX_MSG_FRAGS)
2312                         move_from = i + shift - MAX_MSG_FRAGS;
2313                 else
2314                         move_from = i + shift;
2315                 if (move_from == msg->sg.end)
2316                         break;
2317
2318                 msg->sg.data[i] = msg->sg.data[move_from];
2319                 msg->sg.data[move_from].length = 0;
2320                 msg->sg.data[move_from].page_link = 0;
2321                 msg->sg.data[move_from].offset = 0;
2322                 sk_msg_iter_var_next(i);
2323         } while (1);
2324
2325         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2326                       msg->sg.end - shift + MAX_MSG_FRAGS :
2327                       msg->sg.end - shift;
2328 out:
2329         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2330         msg->data_end = msg->data + bytes;
2331         return 0;
2332 }
2333
2334 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2335         .func           = bpf_msg_pull_data,
2336         .gpl_only       = false,
2337         .ret_type       = RET_INTEGER,
2338         .arg1_type      = ARG_PTR_TO_CTX,
2339         .arg2_type      = ARG_ANYTHING,
2340         .arg3_type      = ARG_ANYTHING,
2341         .arg4_type      = ARG_ANYTHING,
2342 };
2343
2344 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2345            u32, len, u64, flags)
2346 {
2347         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2348         u32 new, i = 0, l, space, copy = 0, offset = 0;
2349         u8 *raw, *to, *from;
2350         struct page *page;
2351
2352         if (unlikely(flags))
2353                 return -EINVAL;
2354
2355         /* First find the starting scatterlist element */
2356         i = msg->sg.start;
2357         do {
2358                 l = sk_msg_elem(msg, i)->length;
2359
2360                 if (start < offset + l)
2361                         break;
2362                 offset += l;
2363                 sk_msg_iter_var_next(i);
2364         } while (i != msg->sg.end);
2365
2366         if (start >= offset + l)
2367                 return -EINVAL;
2368
2369         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2370
2371         /* If no space available will fallback to copy, we need at
2372          * least one scatterlist elem available to push data into
2373          * when start aligns to the beginning of an element or two
2374          * when it falls inside an element. We handle the start equals
2375          * offset case because its the common case for inserting a
2376          * header.
2377          */
2378         if (!space || (space == 1 && start != offset))
2379                 copy = msg->sg.data[i].length;
2380
2381         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2382                            get_order(copy + len));
2383         if (unlikely(!page))
2384                 return -ENOMEM;
2385
2386         if (copy) {
2387                 int front, back;
2388
2389                 raw = page_address(page);
2390
2391                 psge = sk_msg_elem(msg, i);
2392                 front = start - offset;
2393                 back = psge->length - front;
2394                 from = sg_virt(psge);
2395
2396                 if (front)
2397                         memcpy(raw, from, front);
2398
2399                 if (back) {
2400                         from += front;
2401                         to = raw + front + len;
2402
2403                         memcpy(to, from, back);
2404                 }
2405
2406                 put_page(sg_page(psge));
2407         } else if (start - offset) {
2408                 psge = sk_msg_elem(msg, i);
2409                 rsge = sk_msg_elem_cpy(msg, i);
2410
2411                 psge->length = start - offset;
2412                 rsge.length -= psge->length;
2413                 rsge.offset += start;
2414
2415                 sk_msg_iter_var_next(i);
2416                 sg_unmark_end(psge);
2417                 sk_msg_iter_next(msg, end);
2418         }
2419
2420         /* Slot(s) to place newly allocated data */
2421         new = i;
2422
2423         /* Shift one or two slots as needed */
2424         if (!copy) {
2425                 sge = sk_msg_elem_cpy(msg, i);
2426
2427                 sk_msg_iter_var_next(i);
2428                 sg_unmark_end(&sge);
2429                 sk_msg_iter_next(msg, end);
2430
2431                 nsge = sk_msg_elem_cpy(msg, i);
2432                 if (rsge.length) {
2433                         sk_msg_iter_var_next(i);
2434                         nnsge = sk_msg_elem_cpy(msg, i);
2435                 }
2436
2437                 while (i != msg->sg.end) {
2438                         msg->sg.data[i] = sge;
2439                         sge = nsge;
2440                         sk_msg_iter_var_next(i);
2441                         if (rsge.length) {
2442                                 nsge = nnsge;
2443                                 nnsge = sk_msg_elem_cpy(msg, i);
2444                         } else {
2445                                 nsge = sk_msg_elem_cpy(msg, i);
2446                         }
2447                 }
2448         }
2449
2450         /* Place newly allocated data buffer */
2451         sk_mem_charge(msg->sk, len);
2452         msg->sg.size += len;
2453         msg->sg.copy[new] = false;
2454         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2455         if (rsge.length) {
2456                 get_page(sg_page(&rsge));
2457                 sk_msg_iter_var_next(new);
2458                 msg->sg.data[new] = rsge;
2459         }
2460
2461         sk_msg_compute_data_pointers(msg);
2462         return 0;
2463 }
2464
2465 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2466         .func           = bpf_msg_push_data,
2467         .gpl_only       = false,
2468         .ret_type       = RET_INTEGER,
2469         .arg1_type      = ARG_PTR_TO_CTX,
2470         .arg2_type      = ARG_ANYTHING,
2471         .arg3_type      = ARG_ANYTHING,
2472         .arg4_type      = ARG_ANYTHING,
2473 };
2474
2475 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2476 {
2477         int prev;
2478
2479         do {
2480                 prev = i;
2481                 sk_msg_iter_var_next(i);
2482                 msg->sg.data[prev] = msg->sg.data[i];
2483         } while (i != msg->sg.end);
2484
2485         sk_msg_iter_prev(msg, end);
2486 }
2487
2488 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2489 {
2490         struct scatterlist tmp, sge;
2491
2492         sk_msg_iter_next(msg, end);
2493         sge = sk_msg_elem_cpy(msg, i);
2494         sk_msg_iter_var_next(i);
2495         tmp = sk_msg_elem_cpy(msg, i);
2496
2497         while (i != msg->sg.end) {
2498                 msg->sg.data[i] = sge;
2499                 sk_msg_iter_var_next(i);
2500                 sge = tmp;
2501                 tmp = sk_msg_elem_cpy(msg, i);
2502         }
2503 }
2504
2505 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2506            u32, len, u64, flags)
2507 {
2508         u32 i = 0, l, space, offset = 0;
2509         u64 last = start + len;
2510         int pop;
2511
2512         if (unlikely(flags))
2513                 return -EINVAL;
2514
2515         /* First find the starting scatterlist element */
2516         i = msg->sg.start;
2517         do {
2518                 l = sk_msg_elem(msg, i)->length;
2519
2520                 if (start < offset + l)
2521                         break;
2522                 offset += l;
2523                 sk_msg_iter_var_next(i);
2524         } while (i != msg->sg.end);
2525
2526         /* Bounds checks: start and pop must be inside message */
2527         if (start >= offset + l || last >= msg->sg.size)
2528                 return -EINVAL;
2529
2530         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2531
2532         pop = len;
2533         /* --------------| offset
2534          * -| start      |-------- len -------|
2535          *
2536          *  |----- a ----|-------- pop -------|----- b ----|
2537          *  |______________________________________________| length
2538          *
2539          *
2540          * a:   region at front of scatter element to save
2541          * b:   region at back of scatter element to save when length > A + pop
2542          * pop: region to pop from element, same as input 'pop' here will be
2543          *      decremented below per iteration.
2544          *
2545          * Two top-level cases to handle when start != offset, first B is non
2546          * zero and second B is zero corresponding to when a pop includes more
2547          * than one element.
2548          *
2549          * Then if B is non-zero AND there is no space allocate space and
2550          * compact A, B regions into page. If there is space shift ring to
2551          * the rigth free'ing the next element in ring to place B, leaving
2552          * A untouched except to reduce length.
2553          */
2554         if (start != offset) {
2555                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2556                 int a = start;
2557                 int b = sge->length - pop - a;
2558
2559                 sk_msg_iter_var_next(i);
2560
2561                 if (pop < sge->length - a) {
2562                         if (space) {
2563                                 sge->length = a;
2564                                 sk_msg_shift_right(msg, i);
2565                                 nsge = sk_msg_elem(msg, i);
2566                                 get_page(sg_page(sge));
2567                                 sg_set_page(nsge,
2568                                             sg_page(sge),
2569                                             b, sge->offset + pop + a);
2570                         } else {
2571                                 struct page *page, *orig;
2572                                 u8 *to, *from;
2573
2574                                 page = alloc_pages(__GFP_NOWARN |
2575                                                    __GFP_COMP   | GFP_ATOMIC,
2576                                                    get_order(a + b));
2577                                 if (unlikely(!page))
2578                                         return -ENOMEM;
2579
2580                                 sge->length = a;
2581                                 orig = sg_page(sge);
2582                                 from = sg_virt(sge);
2583                                 to = page_address(page);
2584                                 memcpy(to, from, a);
2585                                 memcpy(to + a, from + a + pop, b);
2586                                 sg_set_page(sge, page, a + b, 0);
2587                                 put_page(orig);
2588                         }
2589                         pop = 0;
2590                 } else if (pop >= sge->length - a) {
2591                         sge->length = a;
2592                         pop -= (sge->length - a);
2593                 }
2594         }
2595
2596         /* From above the current layout _must_ be as follows,
2597          *
2598          * -| offset
2599          * -| start
2600          *
2601          *  |---- pop ---|---------------- b ------------|
2602          *  |____________________________________________| length
2603          *
2604          * Offset and start of the current msg elem are equal because in the
2605          * previous case we handled offset != start and either consumed the
2606          * entire element and advanced to the next element OR pop == 0.
2607          *
2608          * Two cases to handle here are first pop is less than the length
2609          * leaving some remainder b above. Simply adjust the element's layout
2610          * in this case. Or pop >= length of the element so that b = 0. In this
2611          * case advance to next element decrementing pop.
2612          */
2613         while (pop) {
2614                 struct scatterlist *sge = sk_msg_elem(msg, i);
2615
2616                 if (pop < sge->length) {
2617                         sge->length -= pop;
2618                         sge->offset += pop;
2619                         pop = 0;
2620                 } else {
2621                         pop -= sge->length;
2622                         sk_msg_shift_left(msg, i);
2623                 }
2624                 sk_msg_iter_var_next(i);
2625         }
2626
2627         sk_mem_uncharge(msg->sk, len - pop);
2628         msg->sg.size -= (len - pop);
2629         sk_msg_compute_data_pointers(msg);
2630         return 0;
2631 }
2632
2633 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2634         .func           = bpf_msg_pop_data,
2635         .gpl_only       = false,
2636         .ret_type       = RET_INTEGER,
2637         .arg1_type      = ARG_PTR_TO_CTX,
2638         .arg2_type      = ARG_ANYTHING,
2639         .arg3_type      = ARG_ANYTHING,
2640         .arg4_type      = ARG_ANYTHING,
2641 };
2642
2643 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2644 {
2645         return task_get_classid(skb);
2646 }
2647
2648 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2649         .func           = bpf_get_cgroup_classid,
2650         .gpl_only       = false,
2651         .ret_type       = RET_INTEGER,
2652         .arg1_type      = ARG_PTR_TO_CTX,
2653 };
2654
2655 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2656 {
2657         return dst_tclassid(skb);
2658 }
2659
2660 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2661         .func           = bpf_get_route_realm,
2662         .gpl_only       = false,
2663         .ret_type       = RET_INTEGER,
2664         .arg1_type      = ARG_PTR_TO_CTX,
2665 };
2666
2667 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2668 {
2669         /* If skb_clear_hash() was called due to mangling, we can
2670          * trigger SW recalculation here. Later access to hash
2671          * can then use the inline skb->hash via context directly
2672          * instead of calling this helper again.
2673          */
2674         return skb_get_hash(skb);
2675 }
2676
2677 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2678         .func           = bpf_get_hash_recalc,
2679         .gpl_only       = false,
2680         .ret_type       = RET_INTEGER,
2681         .arg1_type      = ARG_PTR_TO_CTX,
2682 };
2683
2684 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2685 {
2686         /* After all direct packet write, this can be used once for
2687          * triggering a lazy recalc on next skb_get_hash() invocation.
2688          */
2689         skb_clear_hash(skb);
2690         return 0;
2691 }
2692
2693 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2694         .func           = bpf_set_hash_invalid,
2695         .gpl_only       = false,
2696         .ret_type       = RET_INTEGER,
2697         .arg1_type      = ARG_PTR_TO_CTX,
2698 };
2699
2700 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2701 {
2702         /* Set user specified hash as L4(+), so that it gets returned
2703          * on skb_get_hash() call unless BPF prog later on triggers a
2704          * skb_clear_hash().
2705          */
2706         __skb_set_sw_hash(skb, hash, true);
2707         return 0;
2708 }
2709
2710 static const struct bpf_func_proto bpf_set_hash_proto = {
2711         .func           = bpf_set_hash,
2712         .gpl_only       = false,
2713         .ret_type       = RET_INTEGER,
2714         .arg1_type      = ARG_PTR_TO_CTX,
2715         .arg2_type      = ARG_ANYTHING,
2716 };
2717
2718 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2719            u16, vlan_tci)
2720 {
2721         int ret;
2722
2723         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2724                      vlan_proto != htons(ETH_P_8021AD)))
2725                 vlan_proto = htons(ETH_P_8021Q);
2726
2727         bpf_push_mac_rcsum(skb);
2728         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2729         bpf_pull_mac_rcsum(skb);
2730
2731         bpf_compute_data_pointers(skb);
2732         return ret;
2733 }
2734
2735 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2736         .func           = bpf_skb_vlan_push,
2737         .gpl_only       = false,
2738         .ret_type       = RET_INTEGER,
2739         .arg1_type      = ARG_PTR_TO_CTX,
2740         .arg2_type      = ARG_ANYTHING,
2741         .arg3_type      = ARG_ANYTHING,
2742 };
2743
2744 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2745 {
2746         int ret;
2747
2748         bpf_push_mac_rcsum(skb);
2749         ret = skb_vlan_pop(skb);
2750         bpf_pull_mac_rcsum(skb);
2751
2752         bpf_compute_data_pointers(skb);
2753         return ret;
2754 }
2755
2756 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2757         .func           = bpf_skb_vlan_pop,
2758         .gpl_only       = false,
2759         .ret_type       = RET_INTEGER,
2760         .arg1_type      = ARG_PTR_TO_CTX,
2761 };
2762
2763 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2764 {
2765         /* Caller already did skb_cow() with len as headroom,
2766          * so no need to do it here.
2767          */
2768         skb_push(skb, len);
2769         memmove(skb->data, skb->data + len, off);
2770         memset(skb->data + off, 0, len);
2771
2772         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2773          * needed here as it does not change the skb->csum
2774          * result for checksum complete when summing over
2775          * zeroed blocks.
2776          */
2777         return 0;
2778 }
2779
2780 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2781 {
2782         /* skb_ensure_writable() is not needed here, as we're
2783          * already working on an uncloned skb.
2784          */
2785         if (unlikely(!pskb_may_pull(skb, off + len)))
2786                 return -ENOMEM;
2787
2788         skb_postpull_rcsum(skb, skb->data + off, len);
2789         memmove(skb->data + len, skb->data, off);
2790         __skb_pull(skb, len);
2791
2792         return 0;
2793 }
2794
2795 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2796 {
2797         bool trans_same = skb->transport_header == skb->network_header;
2798         int ret;
2799
2800         /* There's no need for __skb_push()/__skb_pull() pair to
2801          * get to the start of the mac header as we're guaranteed
2802          * to always start from here under eBPF.
2803          */
2804         ret = bpf_skb_generic_push(skb, off, len);
2805         if (likely(!ret)) {
2806                 skb->mac_header -= len;
2807                 skb->network_header -= len;
2808                 if (trans_same)
2809                         skb->transport_header = skb->network_header;
2810         }
2811
2812         return ret;
2813 }
2814
2815 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2816 {
2817         bool trans_same = skb->transport_header == skb->network_header;
2818         int ret;
2819
2820         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2821         ret = bpf_skb_generic_pop(skb, off, len);
2822         if (likely(!ret)) {
2823                 skb->mac_header += len;
2824                 skb->network_header += len;
2825                 if (trans_same)
2826                         skb->transport_header = skb->network_header;
2827         }
2828
2829         return ret;
2830 }
2831
2832 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2833 {
2834         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2835         u32 off = skb_mac_header_len(skb);
2836         int ret;
2837
2838         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2839                 return -ENOTSUPP;
2840
2841         ret = skb_cow(skb, len_diff);
2842         if (unlikely(ret < 0))
2843                 return ret;
2844
2845         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2846         if (unlikely(ret < 0))
2847                 return ret;
2848
2849         if (skb_is_gso(skb)) {
2850                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2851
2852                 /* SKB_GSO_TCPV4 needs to be changed into
2853                  * SKB_GSO_TCPV6.
2854                  */
2855                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2856                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2857                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2858                 }
2859
2860                 /* Due to IPv6 header, MSS needs to be downgraded. */
2861                 skb_decrease_gso_size(shinfo, len_diff);
2862                 /* Header must be checked, and gso_segs recomputed. */
2863                 shinfo->gso_type |= SKB_GSO_DODGY;
2864                 shinfo->gso_segs = 0;
2865         }
2866
2867         skb->protocol = htons(ETH_P_IPV6);
2868         skb_clear_hash(skb);
2869
2870         return 0;
2871 }
2872
2873 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2874 {
2875         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2876         u32 off = skb_mac_header_len(skb);
2877         int ret;
2878
2879         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
2880                 return -ENOTSUPP;
2881
2882         ret = skb_unclone(skb, GFP_ATOMIC);
2883         if (unlikely(ret < 0))
2884                 return ret;
2885
2886         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2887         if (unlikely(ret < 0))
2888                 return ret;
2889
2890         if (skb_is_gso(skb)) {
2891                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2892
2893                 /* SKB_GSO_TCPV6 needs to be changed into
2894                  * SKB_GSO_TCPV4.
2895                  */
2896                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2897                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2898                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2899                 }
2900
2901                 /* Due to IPv4 header, MSS can be upgraded. */
2902                 skb_increase_gso_size(shinfo, len_diff);
2903                 /* Header must be checked, and gso_segs recomputed. */
2904                 shinfo->gso_type |= SKB_GSO_DODGY;
2905                 shinfo->gso_segs = 0;
2906         }
2907
2908         skb->protocol = htons(ETH_P_IP);
2909         skb_clear_hash(skb);
2910
2911         return 0;
2912 }
2913
2914 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2915 {
2916         __be16 from_proto = skb->protocol;
2917
2918         if (from_proto == htons(ETH_P_IP) &&
2919               to_proto == htons(ETH_P_IPV6))
2920                 return bpf_skb_proto_4_to_6(skb);
2921
2922         if (from_proto == htons(ETH_P_IPV6) &&
2923               to_proto == htons(ETH_P_IP))
2924                 return bpf_skb_proto_6_to_4(skb);
2925
2926         return -ENOTSUPP;
2927 }
2928
2929 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2930            u64, flags)
2931 {
2932         int ret;
2933
2934         if (unlikely(flags))
2935                 return -EINVAL;
2936
2937         /* General idea is that this helper does the basic groundwork
2938          * needed for changing the protocol, and eBPF program fills the
2939          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2940          * and other helpers, rather than passing a raw buffer here.
2941          *
2942          * The rationale is to keep this minimal and without a need to
2943          * deal with raw packet data. F.e. even if we would pass buffers
2944          * here, the program still needs to call the bpf_lX_csum_replace()
2945          * helpers anyway. Plus, this way we keep also separation of
2946          * concerns, since f.e. bpf_skb_store_bytes() should only take
2947          * care of stores.
2948          *
2949          * Currently, additional options and extension header space are
2950          * not supported, but flags register is reserved so we can adapt
2951          * that. For offloads, we mark packet as dodgy, so that headers
2952          * need to be verified first.
2953          */
2954         ret = bpf_skb_proto_xlat(skb, proto);
2955         bpf_compute_data_pointers(skb);
2956         return ret;
2957 }
2958
2959 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2960         .func           = bpf_skb_change_proto,
2961         .gpl_only       = false,
2962         .ret_type       = RET_INTEGER,
2963         .arg1_type      = ARG_PTR_TO_CTX,
2964         .arg2_type      = ARG_ANYTHING,
2965         .arg3_type      = ARG_ANYTHING,
2966 };
2967
2968 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2969 {
2970         /* We only allow a restricted subset to be changed for now. */
2971         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2972                      !skb_pkt_type_ok(pkt_type)))
2973                 return -EINVAL;
2974
2975         skb->pkt_type = pkt_type;
2976         return 0;
2977 }
2978
2979 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2980         .func           = bpf_skb_change_type,
2981         .gpl_only       = false,
2982         .ret_type       = RET_INTEGER,
2983         .arg1_type      = ARG_PTR_TO_CTX,
2984         .arg2_type      = ARG_ANYTHING,
2985 };
2986
2987 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2988 {
2989         switch (skb->protocol) {
2990         case htons(ETH_P_IP):
2991                 return sizeof(struct iphdr);
2992         case htons(ETH_P_IPV6):
2993                 return sizeof(struct ipv6hdr);
2994         default:
2995                 return ~0U;
2996         }
2997 }
2998
2999 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3000                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3001
3002 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3003                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3004                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3005                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3006                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3007                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3008
3009 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3010                             u64 flags)
3011 {
3012         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3013         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3014         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3015         unsigned int gso_type = SKB_GSO_DODGY;
3016         int ret;
3017
3018         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3019                 /* udp gso_size delineates datagrams, only allow if fixed */
3020                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3021                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3022                         return -ENOTSUPP;
3023         }
3024
3025         ret = skb_cow_head(skb, len_diff);
3026         if (unlikely(ret < 0))
3027                 return ret;
3028
3029         if (encap) {
3030                 if (skb->protocol != htons(ETH_P_IP) &&
3031                     skb->protocol != htons(ETH_P_IPV6))
3032                         return -ENOTSUPP;
3033
3034                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3035                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3036                         return -EINVAL;
3037
3038                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3039                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3040                         return -EINVAL;
3041
3042                 if (skb->encapsulation)
3043                         return -EALREADY;
3044
3045                 mac_len = skb->network_header - skb->mac_header;
3046                 inner_net = skb->network_header;
3047                 if (inner_mac_len > len_diff)
3048                         return -EINVAL;
3049                 inner_trans = skb->transport_header;
3050         }
3051
3052         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3053         if (unlikely(ret < 0))
3054                 return ret;
3055
3056         if (encap) {
3057                 skb->inner_mac_header = inner_net - inner_mac_len;
3058                 skb->inner_network_header = inner_net;
3059                 skb->inner_transport_header = inner_trans;
3060                 skb_set_inner_protocol(skb, skb->protocol);
3061
3062                 skb->encapsulation = 1;
3063                 skb_set_network_header(skb, mac_len);
3064
3065                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3066                         gso_type |= SKB_GSO_UDP_TUNNEL;
3067                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3068                         gso_type |= SKB_GSO_GRE;
3069                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3070                         gso_type |= SKB_GSO_IPXIP6;
3071                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3072                         gso_type |= SKB_GSO_IPXIP4;
3073
3074                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3075                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3076                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3077                                         sizeof(struct ipv6hdr) :
3078                                         sizeof(struct iphdr);
3079
3080                         skb_set_transport_header(skb, mac_len + nh_len);
3081                 }
3082
3083                 /* Match skb->protocol to new outer l3 protocol */
3084                 if (skb->protocol == htons(ETH_P_IP) &&
3085                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3086                         skb->protocol = htons(ETH_P_IPV6);
3087                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3088                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3089                         skb->protocol = htons(ETH_P_IP);
3090         }
3091
3092         if (skb_is_gso(skb)) {
3093                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3094
3095                 /* Due to header grow, MSS needs to be downgraded. */
3096                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3097                         skb_decrease_gso_size(shinfo, len_diff);
3098
3099                 /* Header must be checked, and gso_segs recomputed. */
3100                 shinfo->gso_type |= gso_type;
3101                 shinfo->gso_segs = 0;
3102         }
3103
3104         return 0;
3105 }
3106
3107 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3108                               u64 flags)
3109 {
3110         int ret;
3111
3112         if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO)
3113                 return -EINVAL;
3114
3115         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3116                 /* udp gso_size delineates datagrams, only allow if fixed */
3117                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3118                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3119                         return -ENOTSUPP;
3120         }
3121
3122         ret = skb_unclone(skb, GFP_ATOMIC);
3123         if (unlikely(ret < 0))
3124                 return ret;
3125
3126         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3127         if (unlikely(ret < 0))
3128                 return ret;
3129
3130         if (skb_is_gso(skb)) {
3131                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3132
3133                 /* Due to header shrink, MSS can be upgraded. */
3134                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3135                         skb_increase_gso_size(shinfo, len_diff);
3136
3137                 /* Header must be checked, and gso_segs recomputed. */
3138                 shinfo->gso_type |= SKB_GSO_DODGY;
3139                 shinfo->gso_segs = 0;
3140         }
3141
3142         return 0;
3143 }
3144
3145 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
3146 {
3147         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
3148                           SKB_MAX_ALLOC;
3149 }
3150
3151 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3152            u32, mode, u64, flags)
3153 {
3154         u32 len_cur, len_diff_abs = abs(len_diff);
3155         u32 len_min = bpf_skb_net_base_len(skb);
3156         u32 len_max = __bpf_skb_max_len(skb);
3157         __be16 proto = skb->protocol;
3158         bool shrink = len_diff < 0;
3159         u32 off;
3160         int ret;
3161
3162         if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK))
3163                 return -EINVAL;
3164         if (unlikely(len_diff_abs > 0xfffU))
3165                 return -EFAULT;
3166         if (unlikely(proto != htons(ETH_P_IP) &&
3167                      proto != htons(ETH_P_IPV6)))
3168                 return -ENOTSUPP;
3169
3170         off = skb_mac_header_len(skb);
3171         switch (mode) {
3172         case BPF_ADJ_ROOM_NET:
3173                 off += bpf_skb_net_base_len(skb);
3174                 break;
3175         case BPF_ADJ_ROOM_MAC:
3176                 break;
3177         default:
3178                 return -ENOTSUPP;
3179         }
3180
3181         len_cur = skb->len - skb_network_offset(skb);
3182         if ((shrink && (len_diff_abs >= len_cur ||
3183                         len_cur - len_diff_abs < len_min)) ||
3184             (!shrink && (skb->len + len_diff_abs > len_max &&
3185                          !skb_is_gso(skb))))
3186                 return -ENOTSUPP;
3187
3188         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3189                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3190
3191         bpf_compute_data_pointers(skb);
3192         return ret;
3193 }
3194
3195 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3196         .func           = bpf_skb_adjust_room,
3197         .gpl_only       = false,
3198         .ret_type       = RET_INTEGER,
3199         .arg1_type      = ARG_PTR_TO_CTX,
3200         .arg2_type      = ARG_ANYTHING,
3201         .arg3_type      = ARG_ANYTHING,
3202         .arg4_type      = ARG_ANYTHING,
3203 };
3204
3205 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3206 {
3207         u32 min_len = skb_network_offset(skb);
3208
3209         if (skb_transport_header_was_set(skb))
3210                 min_len = skb_transport_offset(skb);
3211         if (skb->ip_summed == CHECKSUM_PARTIAL)
3212                 min_len = skb_checksum_start_offset(skb) +
3213                           skb->csum_offset + sizeof(__sum16);
3214         return min_len;
3215 }
3216
3217 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3218 {
3219         unsigned int old_len = skb->len;
3220         int ret;
3221
3222         ret = __skb_grow_rcsum(skb, new_len);
3223         if (!ret)
3224                 memset(skb->data + old_len, 0, new_len - old_len);
3225         return ret;
3226 }
3227
3228 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3229 {
3230         return __skb_trim_rcsum(skb, new_len);
3231 }
3232
3233 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3234                                         u64 flags)
3235 {
3236         u32 max_len = __bpf_skb_max_len(skb);
3237         u32 min_len = __bpf_skb_min_len(skb);
3238         int ret;
3239
3240         if (unlikely(flags || new_len > max_len || new_len < min_len))
3241                 return -EINVAL;
3242         if (skb->encapsulation)
3243                 return -ENOTSUPP;
3244
3245         /* The basic idea of this helper is that it's performing the
3246          * needed work to either grow or trim an skb, and eBPF program
3247          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3248          * bpf_lX_csum_replace() and others rather than passing a raw
3249          * buffer here. This one is a slow path helper and intended
3250          * for replies with control messages.
3251          *
3252          * Like in bpf_skb_change_proto(), we want to keep this rather
3253          * minimal and without protocol specifics so that we are able
3254          * to separate concerns as in bpf_skb_store_bytes() should only
3255          * be the one responsible for writing buffers.
3256          *
3257          * It's really expected to be a slow path operation here for
3258          * control message replies, so we're implicitly linearizing,
3259          * uncloning and drop offloads from the skb by this.
3260          */
3261         ret = __bpf_try_make_writable(skb, skb->len);
3262         if (!ret) {
3263                 if (new_len > skb->len)
3264                         ret = bpf_skb_grow_rcsum(skb, new_len);
3265                 else if (new_len < skb->len)
3266                         ret = bpf_skb_trim_rcsum(skb, new_len);
3267                 if (!ret && skb_is_gso(skb))
3268                         skb_gso_reset(skb);
3269         }
3270         return ret;
3271 }
3272
3273 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3274            u64, flags)
3275 {
3276         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3277
3278         bpf_compute_data_pointers(skb);
3279         return ret;
3280 }
3281
3282 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3283         .func           = bpf_skb_change_tail,
3284         .gpl_only       = false,
3285         .ret_type       = RET_INTEGER,
3286         .arg1_type      = ARG_PTR_TO_CTX,
3287         .arg2_type      = ARG_ANYTHING,
3288         .arg3_type      = ARG_ANYTHING,
3289 };
3290
3291 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3292            u64, flags)
3293 {
3294         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3295
3296         bpf_compute_data_end_sk_skb(skb);
3297         return ret;
3298 }
3299
3300 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3301         .func           = sk_skb_change_tail,
3302         .gpl_only       = false,
3303         .ret_type       = RET_INTEGER,
3304         .arg1_type      = ARG_PTR_TO_CTX,
3305         .arg2_type      = ARG_ANYTHING,
3306         .arg3_type      = ARG_ANYTHING,
3307 };
3308
3309 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3310                                         u64 flags)
3311 {
3312         u32 max_len = __bpf_skb_max_len(skb);
3313         u32 new_len = skb->len + head_room;
3314         int ret;
3315
3316         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3317                      new_len < skb->len))
3318                 return -EINVAL;
3319
3320         ret = skb_cow(skb, head_room);
3321         if (likely(!ret)) {
3322                 /* Idea for this helper is that we currently only
3323                  * allow to expand on mac header. This means that
3324                  * skb->protocol network header, etc, stay as is.
3325                  * Compared to bpf_skb_change_tail(), we're more
3326                  * flexible due to not needing to linearize or
3327                  * reset GSO. Intention for this helper is to be
3328                  * used by an L3 skb that needs to push mac header
3329                  * for redirection into L2 device.
3330                  */
3331                 __skb_push(skb, head_room);
3332                 memset(skb->data, 0, head_room);
3333                 skb_reset_mac_header(skb);
3334         }
3335
3336         return ret;
3337 }
3338
3339 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3340            u64, flags)
3341 {
3342         int ret = __bpf_skb_change_head(skb, head_room, flags);
3343
3344         bpf_compute_data_pointers(skb);
3345         return ret;
3346 }
3347
3348 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3349         .func           = bpf_skb_change_head,
3350         .gpl_only       = false,
3351         .ret_type       = RET_INTEGER,
3352         .arg1_type      = ARG_PTR_TO_CTX,
3353         .arg2_type      = ARG_ANYTHING,
3354         .arg3_type      = ARG_ANYTHING,
3355 };
3356
3357 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3358            u64, flags)
3359 {
3360         int ret = __bpf_skb_change_head(skb, head_room, flags);
3361
3362         bpf_compute_data_end_sk_skb(skb);
3363         return ret;
3364 }
3365
3366 static const struct bpf_func_proto sk_skb_change_head_proto = {
3367         .func           = sk_skb_change_head,
3368         .gpl_only       = false,
3369         .ret_type       = RET_INTEGER,
3370         .arg1_type      = ARG_PTR_TO_CTX,
3371         .arg2_type      = ARG_ANYTHING,
3372         .arg3_type      = ARG_ANYTHING,
3373 };
3374 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3375 {
3376         return xdp_data_meta_unsupported(xdp) ? 0 :
3377                xdp->data - xdp->data_meta;
3378 }
3379
3380 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3381 {
3382         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3383         unsigned long metalen = xdp_get_metalen(xdp);
3384         void *data_start = xdp_frame_end + metalen;
3385         void *data = xdp->data + offset;
3386
3387         if (unlikely(data < data_start ||
3388                      data > xdp->data_end - ETH_HLEN))
3389                 return -EINVAL;
3390
3391         if (metalen)
3392                 memmove(xdp->data_meta + offset,
3393                         xdp->data_meta, metalen);
3394         xdp->data_meta += offset;
3395         xdp->data = data;
3396
3397         return 0;
3398 }
3399
3400 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3401         .func           = bpf_xdp_adjust_head,
3402         .gpl_only       = false,
3403         .ret_type       = RET_INTEGER,
3404         .arg1_type      = ARG_PTR_TO_CTX,
3405         .arg2_type      = ARG_ANYTHING,
3406 };
3407
3408 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3409 {
3410         void *data_end = xdp->data_end + offset;
3411
3412         /* only shrinking is allowed for now. */
3413         if (unlikely(offset >= 0))
3414                 return -EINVAL;
3415
3416         if (unlikely(data_end < xdp->data + ETH_HLEN))
3417                 return -EINVAL;
3418
3419         xdp->data_end = data_end;
3420
3421         return 0;
3422 }
3423
3424 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3425         .func           = bpf_xdp_adjust_tail,
3426         .gpl_only       = false,
3427         .ret_type       = RET_INTEGER,
3428         .arg1_type      = ARG_PTR_TO_CTX,
3429         .arg2_type      = ARG_ANYTHING,
3430 };
3431
3432 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3433 {
3434         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3435         void *meta = xdp->data_meta + offset;
3436         unsigned long metalen = xdp->data - meta;
3437
3438         if (xdp_data_meta_unsupported(xdp))
3439                 return -ENOTSUPP;
3440         if (unlikely(meta < xdp_frame_end ||
3441                      meta > xdp->data))
3442                 return -EINVAL;
3443         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3444                      (metalen > 32)))
3445                 return -EACCES;
3446
3447         xdp->data_meta = meta;
3448
3449         return 0;
3450 }
3451
3452 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3453         .func           = bpf_xdp_adjust_meta,
3454         .gpl_only       = false,
3455         .ret_type       = RET_INTEGER,
3456         .arg1_type      = ARG_PTR_TO_CTX,
3457         .arg2_type      = ARG_ANYTHING,
3458 };
3459
3460 static int __bpf_tx_xdp(struct net_device *dev,
3461                         struct bpf_map *map,
3462                         struct xdp_buff *xdp,
3463                         u32 index)
3464 {
3465         struct xdp_frame *xdpf;
3466         int err, sent;
3467
3468         if (!dev->netdev_ops->ndo_xdp_xmit) {
3469                 return -EOPNOTSUPP;
3470         }
3471
3472         err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3473         if (unlikely(err))
3474                 return err;
3475
3476         xdpf = convert_to_xdp_frame(xdp);
3477         if (unlikely(!xdpf))
3478                 return -EOVERFLOW;
3479
3480         sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3481         if (sent <= 0)
3482                 return sent;
3483         return 0;
3484 }
3485
3486 static noinline int
3487 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3488                      struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3489 {
3490         struct net_device *fwd;
3491         u32 index = ri->tgt_index;
3492         int err;
3493
3494         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3495         ri->tgt_index = 0;
3496         if (unlikely(!fwd)) {
3497                 err = -EINVAL;
3498                 goto err;
3499         }
3500
3501         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3502         if (unlikely(err))
3503                 goto err;
3504
3505         _trace_xdp_redirect(dev, xdp_prog, index);
3506         return 0;
3507 err:
3508         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3509         return err;
3510 }
3511
3512 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3513                             struct bpf_map *map,
3514                             struct xdp_buff *xdp,
3515                             u32 index)
3516 {
3517         int err;
3518
3519         switch (map->map_type) {
3520         case BPF_MAP_TYPE_DEVMAP:
3521         case BPF_MAP_TYPE_DEVMAP_HASH: {
3522                 struct bpf_dtab_netdev *dst = fwd;
3523
3524                 err = dev_map_enqueue(dst, xdp, dev_rx);
3525                 if (unlikely(err))
3526                         return err;
3527                 break;
3528         }
3529         case BPF_MAP_TYPE_CPUMAP: {
3530                 struct bpf_cpu_map_entry *rcpu = fwd;
3531
3532                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3533                 if (unlikely(err))
3534                         return err;
3535                 break;
3536         }
3537         case BPF_MAP_TYPE_XSKMAP: {
3538                 struct xdp_sock *xs = fwd;
3539
3540                 err = __xsk_map_redirect(map, xdp, xs);
3541                 return err;
3542         }
3543         default:
3544                 break;
3545         }
3546         return 0;
3547 }
3548
3549 void xdp_do_flush_map(void)
3550 {
3551         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3552         struct bpf_map *map = ri->map_to_flush;
3553
3554         ri->map_to_flush = NULL;
3555         if (map) {
3556                 switch (map->map_type) {
3557                 case BPF_MAP_TYPE_DEVMAP:
3558                 case BPF_MAP_TYPE_DEVMAP_HASH:
3559                         __dev_map_flush(map);
3560                         break;
3561                 case BPF_MAP_TYPE_CPUMAP:
3562                         __cpu_map_flush(map);
3563                         break;
3564                 case BPF_MAP_TYPE_XSKMAP:
3565                         __xsk_map_flush(map);
3566                         break;
3567                 default:
3568                         break;
3569                 }
3570         }
3571 }
3572 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3573
3574 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3575 {
3576         switch (map->map_type) {
3577         case BPF_MAP_TYPE_DEVMAP:
3578                 return __dev_map_lookup_elem(map, index);
3579         case BPF_MAP_TYPE_DEVMAP_HASH:
3580                 return __dev_map_hash_lookup_elem(map, index);
3581         case BPF_MAP_TYPE_CPUMAP:
3582                 return __cpu_map_lookup_elem(map, index);
3583         case BPF_MAP_TYPE_XSKMAP:
3584                 return __xsk_map_lookup_elem(map, index);
3585         default:
3586                 return NULL;
3587         }
3588 }
3589
3590 void bpf_clear_redirect_map(struct bpf_map *map)
3591 {
3592         struct bpf_redirect_info *ri;
3593         int cpu;
3594
3595         for_each_possible_cpu(cpu) {
3596                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3597                 /* Avoid polluting remote cacheline due to writes if
3598                  * not needed. Once we pass this test, we need the
3599                  * cmpxchg() to make sure it hasn't been changed in
3600                  * the meantime by remote CPU.
3601                  */
3602                 if (unlikely(READ_ONCE(ri->map) == map))
3603                         cmpxchg(&ri->map, map, NULL);
3604         }
3605 }
3606
3607 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3608                                struct bpf_prog *xdp_prog, struct bpf_map *map,
3609                                struct bpf_redirect_info *ri)
3610 {
3611         u32 index = ri->tgt_index;
3612         void *fwd = ri->tgt_value;
3613         int err;
3614
3615         ri->tgt_index = 0;
3616         ri->tgt_value = NULL;
3617         WRITE_ONCE(ri->map, NULL);
3618
3619         if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3620                 xdp_do_flush_map();
3621
3622         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3623         if (unlikely(err))
3624                 goto err;
3625
3626         ri->map_to_flush = map;
3627         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3628         return 0;
3629 err:
3630         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3631         return err;
3632 }
3633
3634 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3635                     struct bpf_prog *xdp_prog)
3636 {
3637         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3638         struct bpf_map *map = READ_ONCE(ri->map);
3639
3640         if (likely(map))
3641                 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3642
3643         return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3644 }
3645 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3646
3647 static int xdp_do_generic_redirect_map(struct net_device *dev,
3648                                        struct sk_buff *skb,
3649                                        struct xdp_buff *xdp,
3650                                        struct bpf_prog *xdp_prog,
3651                                        struct bpf_map *map)
3652 {
3653         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3654         u32 index = ri->tgt_index;
3655         void *fwd = ri->tgt_value;
3656         int err = 0;
3657
3658         ri->tgt_index = 0;
3659         ri->tgt_value = NULL;
3660         WRITE_ONCE(ri->map, NULL);
3661
3662         if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
3663             map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
3664                 struct bpf_dtab_netdev *dst = fwd;
3665
3666                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3667                 if (unlikely(err))
3668                         goto err;
3669         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3670                 struct xdp_sock *xs = fwd;
3671
3672                 err = xsk_generic_rcv(xs, xdp);
3673                 if (err)
3674                         goto err;
3675                 consume_skb(skb);
3676         } else {
3677                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3678                 err = -EBADRQC;
3679                 goto err;
3680         }
3681
3682         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3683         return 0;
3684 err:
3685         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3686         return err;
3687 }
3688
3689 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3690                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3691 {
3692         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3693         struct bpf_map *map = READ_ONCE(ri->map);
3694         u32 index = ri->tgt_index;
3695         struct net_device *fwd;
3696         int err = 0;
3697
3698         if (map)
3699                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3700                                                    map);
3701         ri->tgt_index = 0;
3702         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3703         if (unlikely(!fwd)) {
3704                 err = -EINVAL;
3705                 goto err;
3706         }
3707
3708         err = xdp_ok_fwd_dev(fwd, skb->len);
3709         if (unlikely(err))
3710                 goto err;
3711
3712         skb->dev = fwd;
3713         _trace_xdp_redirect(dev, xdp_prog, index);
3714         generic_xdp_tx(skb, xdp_prog);
3715         return 0;
3716 err:
3717         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3718         return err;
3719 }
3720 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3721
3722 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3723 {
3724         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3725
3726         if (unlikely(flags))
3727                 return XDP_ABORTED;
3728
3729         ri->flags = flags;
3730         ri->tgt_index = ifindex;
3731         ri->tgt_value = NULL;
3732         WRITE_ONCE(ri->map, NULL);
3733
3734         return XDP_REDIRECT;
3735 }
3736
3737 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3738         .func           = bpf_xdp_redirect,
3739         .gpl_only       = false,
3740         .ret_type       = RET_INTEGER,
3741         .arg1_type      = ARG_ANYTHING,
3742         .arg2_type      = ARG_ANYTHING,
3743 };
3744
3745 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3746            u64, flags)
3747 {
3748         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3749
3750         /* Lower bits of the flags are used as return code on lookup failure */
3751         if (unlikely(flags > XDP_TX))
3752                 return XDP_ABORTED;
3753
3754         ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
3755         if (unlikely(!ri->tgt_value)) {
3756                 /* If the lookup fails we want to clear out the state in the
3757                  * redirect_info struct completely, so that if an eBPF program
3758                  * performs multiple lookups, the last one always takes
3759                  * precedence.
3760                  */
3761                 WRITE_ONCE(ri->map, NULL);
3762                 return flags;
3763         }
3764
3765         ri->flags = flags;
3766         ri->tgt_index = ifindex;
3767         WRITE_ONCE(ri->map, map);
3768
3769         return XDP_REDIRECT;
3770 }
3771
3772 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3773         .func           = bpf_xdp_redirect_map,
3774         .gpl_only       = false,
3775         .ret_type       = RET_INTEGER,
3776         .arg1_type      = ARG_CONST_MAP_PTR,
3777         .arg2_type      = ARG_ANYTHING,
3778         .arg3_type      = ARG_ANYTHING,
3779 };
3780
3781 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3782                                   unsigned long off, unsigned long len)
3783 {
3784         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3785
3786         if (unlikely(!ptr))
3787                 return len;
3788         if (ptr != dst_buff)
3789                 memcpy(dst_buff, ptr, len);
3790
3791         return 0;
3792 }
3793
3794 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3795            u64, flags, void *, meta, u64, meta_size)
3796 {
3797         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3798
3799         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3800                 return -EINVAL;
3801         if (unlikely(skb_size > skb->len))
3802                 return -EFAULT;
3803
3804         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3805                                 bpf_skb_copy);
3806 }
3807
3808 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3809         .func           = bpf_skb_event_output,
3810         .gpl_only       = true,
3811         .ret_type       = RET_INTEGER,
3812         .arg1_type      = ARG_PTR_TO_CTX,
3813         .arg2_type      = ARG_CONST_MAP_PTR,
3814         .arg3_type      = ARG_ANYTHING,
3815         .arg4_type      = ARG_PTR_TO_MEM,
3816         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3817 };
3818
3819 static unsigned short bpf_tunnel_key_af(u64 flags)
3820 {
3821         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3822 }
3823
3824 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3825            u32, size, u64, flags)
3826 {
3827         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3828         u8 compat[sizeof(struct bpf_tunnel_key)];
3829         void *to_orig = to;
3830         int err;
3831
3832         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3833                 err = -EINVAL;
3834                 goto err_clear;
3835         }
3836         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3837                 err = -EPROTO;
3838                 goto err_clear;
3839         }
3840         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3841                 err = -EINVAL;
3842                 switch (size) {
3843                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3844                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3845                         goto set_compat;
3846                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3847                         /* Fixup deprecated structure layouts here, so we have
3848                          * a common path later on.
3849                          */
3850                         if (ip_tunnel_info_af(info) != AF_INET)
3851                                 goto err_clear;
3852 set_compat:
3853                         to = (struct bpf_tunnel_key *)compat;
3854                         break;
3855                 default:
3856                         goto err_clear;
3857                 }
3858         }
3859
3860         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3861         to->tunnel_tos = info->key.tos;
3862         to->tunnel_ttl = info->key.ttl;
3863         to->tunnel_ext = 0;
3864
3865         if (flags & BPF_F_TUNINFO_IPV6) {
3866                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3867                        sizeof(to->remote_ipv6));
3868                 to->tunnel_label = be32_to_cpu(info->key.label);
3869         } else {
3870                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3871                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3872                 to->tunnel_label = 0;
3873         }
3874
3875         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3876                 memcpy(to_orig, to, size);
3877
3878         return 0;
3879 err_clear:
3880         memset(to_orig, 0, size);
3881         return err;
3882 }
3883
3884 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3885         .func           = bpf_skb_get_tunnel_key,
3886         .gpl_only       = false,
3887         .ret_type       = RET_INTEGER,
3888         .arg1_type      = ARG_PTR_TO_CTX,
3889         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3890         .arg3_type      = ARG_CONST_SIZE,
3891         .arg4_type      = ARG_ANYTHING,
3892 };
3893
3894 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3895 {
3896         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3897         int err;
3898
3899         if (unlikely(!info ||
3900                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3901                 err = -ENOENT;
3902                 goto err_clear;
3903         }
3904         if (unlikely(size < info->options_len)) {
3905                 err = -ENOMEM;
3906                 goto err_clear;
3907         }
3908
3909         ip_tunnel_info_opts_get(to, info);
3910         if (size > info->options_len)
3911                 memset(to + info->options_len, 0, size - info->options_len);
3912
3913         return info->options_len;
3914 err_clear:
3915         memset(to, 0, size);
3916         return err;
3917 }
3918
3919 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3920         .func           = bpf_skb_get_tunnel_opt,
3921         .gpl_only       = false,
3922         .ret_type       = RET_INTEGER,
3923         .arg1_type      = ARG_PTR_TO_CTX,
3924         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3925         .arg3_type      = ARG_CONST_SIZE,
3926 };
3927
3928 static struct metadata_dst __percpu *md_dst;
3929
3930 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3931            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3932 {
3933         struct metadata_dst *md = this_cpu_ptr(md_dst);
3934         u8 compat[sizeof(struct bpf_tunnel_key)];
3935         struct ip_tunnel_info *info;
3936
3937         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3938                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3939                 return -EINVAL;
3940         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3941                 switch (size) {
3942                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3943                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3944                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3945                         /* Fixup deprecated structure layouts here, so we have
3946                          * a common path later on.
3947                          */
3948                         memcpy(compat, from, size);
3949                         memset(compat + size, 0, sizeof(compat) - size);
3950                         from = (const struct bpf_tunnel_key *) compat;
3951                         break;
3952                 default:
3953                         return -EINVAL;
3954                 }
3955         }
3956         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3957                      from->tunnel_ext))
3958                 return -EINVAL;
3959
3960         skb_dst_drop(skb);
3961         dst_hold((struct dst_entry *) md);
3962         skb_dst_set(skb, (struct dst_entry *) md);
3963
3964         info = &md->u.tun_info;
3965         memset(info, 0, sizeof(*info));
3966         info->mode = IP_TUNNEL_INFO_TX;
3967
3968         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3969         if (flags & BPF_F_DONT_FRAGMENT)
3970                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3971         if (flags & BPF_F_ZERO_CSUM_TX)
3972                 info->key.tun_flags &= ~TUNNEL_CSUM;
3973         if (flags & BPF_F_SEQ_NUMBER)
3974                 info->key.tun_flags |= TUNNEL_SEQ;
3975
3976         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3977         info->key.tos = from->tunnel_tos;
3978         info->key.ttl = from->tunnel_ttl;
3979
3980         if (flags & BPF_F_TUNINFO_IPV6) {
3981                 info->mode |= IP_TUNNEL_INFO_IPV6;
3982                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3983                        sizeof(from->remote_ipv6));
3984                 info->key.label = cpu_to_be32(from->tunnel_label) &
3985                                   IPV6_FLOWLABEL_MASK;
3986         } else {
3987                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3988         }
3989
3990         return 0;
3991 }
3992
3993 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3994         .func           = bpf_skb_set_tunnel_key,
3995         .gpl_only       = false,
3996         .ret_type       = RET_INTEGER,
3997         .arg1_type      = ARG_PTR_TO_CTX,
3998         .arg2_type      = ARG_PTR_TO_MEM,
3999         .arg3_type      = ARG_CONST_SIZE,
4000         .arg4_type      = ARG_ANYTHING,
4001 };
4002
4003 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4004            const u8 *, from, u32, size)
4005 {
4006         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4007         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4008
4009         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4010                 return -EINVAL;
4011         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4012                 return -ENOMEM;
4013
4014         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4015
4016         return 0;
4017 }
4018
4019 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4020         .func           = bpf_skb_set_tunnel_opt,
4021         .gpl_only       = false,
4022         .ret_type       = RET_INTEGER,
4023         .arg1_type      = ARG_PTR_TO_CTX,
4024         .arg2_type      = ARG_PTR_TO_MEM,
4025         .arg3_type      = ARG_CONST_SIZE,
4026 };
4027
4028 static const struct bpf_func_proto *
4029 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4030 {
4031         if (!md_dst) {
4032                 struct metadata_dst __percpu *tmp;
4033
4034                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4035                                                 METADATA_IP_TUNNEL,
4036                                                 GFP_KERNEL);
4037                 if (!tmp)
4038                         return NULL;
4039                 if (cmpxchg(&md_dst, NULL, tmp))
4040                         metadata_dst_free_percpu(tmp);
4041         }
4042
4043         switch (which) {
4044         case BPF_FUNC_skb_set_tunnel_key:
4045                 return &bpf_skb_set_tunnel_key_proto;
4046         case BPF_FUNC_skb_set_tunnel_opt:
4047                 return &bpf_skb_set_tunnel_opt_proto;
4048         default:
4049                 return NULL;
4050         }
4051 }
4052
4053 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4054            u32, idx)
4055 {
4056         struct bpf_array *array = container_of(map, struct bpf_array, map);
4057         struct cgroup *cgrp;
4058         struct sock *sk;
4059
4060         sk = skb_to_full_sk(skb);
4061         if (!sk || !sk_fullsock(sk))
4062                 return -ENOENT;
4063         if (unlikely(idx >= array->map.max_entries))
4064                 return -E2BIG;
4065
4066         cgrp = READ_ONCE(array->ptrs[idx]);
4067         if (unlikely(!cgrp))
4068                 return -EAGAIN;
4069
4070         return sk_under_cgroup_hierarchy(sk, cgrp);
4071 }
4072
4073 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4074         .func           = bpf_skb_under_cgroup,
4075         .gpl_only       = false,
4076         .ret_type       = RET_INTEGER,
4077         .arg1_type      = ARG_PTR_TO_CTX,
4078         .arg2_type      = ARG_CONST_MAP_PTR,
4079         .arg3_type      = ARG_ANYTHING,
4080 };
4081
4082 #ifdef CONFIG_SOCK_CGROUP_DATA
4083 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4084 {
4085         struct sock *sk = skb_to_full_sk(skb);
4086         struct cgroup *cgrp;
4087
4088         if (!sk || !sk_fullsock(sk))
4089                 return 0;
4090
4091         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4092         return cgrp->kn->id.id;
4093 }
4094
4095 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4096         .func           = bpf_skb_cgroup_id,
4097         .gpl_only       = false,
4098         .ret_type       = RET_INTEGER,
4099         .arg1_type      = ARG_PTR_TO_CTX,
4100 };
4101
4102 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4103            ancestor_level)
4104 {
4105         struct sock *sk = skb_to_full_sk(skb);
4106         struct cgroup *ancestor;
4107         struct cgroup *cgrp;
4108
4109         if (!sk || !sk_fullsock(sk))
4110                 return 0;
4111
4112         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4113         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4114         if (!ancestor)
4115                 return 0;
4116
4117         return ancestor->kn->id.id;
4118 }
4119
4120 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4121         .func           = bpf_skb_ancestor_cgroup_id,
4122         .gpl_only       = false,
4123         .ret_type       = RET_INTEGER,
4124         .arg1_type      = ARG_PTR_TO_CTX,
4125         .arg2_type      = ARG_ANYTHING,
4126 };
4127 #endif
4128
4129 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4130                                   unsigned long off, unsigned long len)
4131 {
4132         memcpy(dst_buff, src_buff + off, len);
4133         return 0;
4134 }
4135
4136 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4137            u64, flags, void *, meta, u64, meta_size)
4138 {
4139         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4140
4141         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4142                 return -EINVAL;
4143         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4144                 return -EFAULT;
4145
4146         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4147                                 xdp_size, bpf_xdp_copy);
4148 }
4149
4150 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4151         .func           = bpf_xdp_event_output,
4152         .gpl_only       = true,
4153         .ret_type       = RET_INTEGER,
4154         .arg1_type      = ARG_PTR_TO_CTX,
4155         .arg2_type      = ARG_CONST_MAP_PTR,
4156         .arg3_type      = ARG_ANYTHING,
4157         .arg4_type      = ARG_PTR_TO_MEM,
4158         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4159 };
4160
4161 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4162 {
4163         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
4164 }
4165
4166 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4167         .func           = bpf_get_socket_cookie,
4168         .gpl_only       = false,
4169         .ret_type       = RET_INTEGER,
4170         .arg1_type      = ARG_PTR_TO_CTX,
4171 };
4172
4173 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4174 {
4175         return sock_gen_cookie(ctx->sk);
4176 }
4177
4178 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4179         .func           = bpf_get_socket_cookie_sock_addr,
4180         .gpl_only       = false,
4181         .ret_type       = RET_INTEGER,
4182         .arg1_type      = ARG_PTR_TO_CTX,
4183 };
4184
4185 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4186 {
4187         return sock_gen_cookie(ctx->sk);
4188 }
4189
4190 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4191         .func           = bpf_get_socket_cookie_sock_ops,
4192         .gpl_only       = false,
4193         .ret_type       = RET_INTEGER,
4194         .arg1_type      = ARG_PTR_TO_CTX,
4195 };
4196
4197 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4198 {
4199         struct sock *sk = sk_to_full_sk(skb->sk);
4200         kuid_t kuid;
4201
4202         if (!sk || !sk_fullsock(sk))
4203                 return overflowuid;
4204         kuid = sock_net_uid(sock_net(sk), sk);
4205         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4206 }
4207
4208 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4209         .func           = bpf_get_socket_uid,
4210         .gpl_only       = false,
4211         .ret_type       = RET_INTEGER,
4212         .arg1_type      = ARG_PTR_TO_CTX,
4213 };
4214
4215 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock,
4216            struct bpf_map *, map, u64, flags, void *, data, u64, size)
4217 {
4218         if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
4219                 return -EINVAL;
4220
4221         return bpf_event_output(map, flags, data, size, NULL, 0, NULL);
4222 }
4223
4224 static const struct bpf_func_proto bpf_sockopt_event_output_proto =  {
4225         .func           = bpf_sockopt_event_output,
4226         .gpl_only       = true,
4227         .ret_type       = RET_INTEGER,
4228         .arg1_type      = ARG_PTR_TO_CTX,
4229         .arg2_type      = ARG_CONST_MAP_PTR,
4230         .arg3_type      = ARG_ANYTHING,
4231         .arg4_type      = ARG_PTR_TO_MEM,
4232         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4233 };
4234
4235 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4236            int, level, int, optname, char *, optval, int, optlen)
4237 {
4238         struct sock *sk = bpf_sock->sk;
4239         int ret = 0;
4240         int val;
4241
4242         if (!sk_fullsock(sk))
4243                 return -EINVAL;
4244
4245         if (level == SOL_SOCKET) {
4246                 if (optlen != sizeof(int))
4247                         return -EINVAL;
4248                 val = *((int *)optval);
4249
4250                 /* Only some socketops are supported */
4251                 switch (optname) {
4252                 case SO_RCVBUF:
4253                         val = min_t(u32, val, sysctl_rmem_max);
4254                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4255                         WRITE_ONCE(sk->sk_rcvbuf,
4256                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4257                         break;
4258                 case SO_SNDBUF:
4259                         val = min_t(u32, val, sysctl_wmem_max);
4260                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4261                         WRITE_ONCE(sk->sk_sndbuf,
4262                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4263                         break;
4264                 case SO_MAX_PACING_RATE: /* 32bit version */
4265                         if (val != ~0U)
4266                                 cmpxchg(&sk->sk_pacing_status,
4267                                         SK_PACING_NONE,
4268                                         SK_PACING_NEEDED);
4269                         sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
4270                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4271                                                  sk->sk_max_pacing_rate);
4272                         break;
4273                 case SO_PRIORITY:
4274                         sk->sk_priority = val;
4275                         break;
4276                 case SO_RCVLOWAT:
4277                         if (val < 0)
4278                                 val = INT_MAX;
4279                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4280                         break;
4281                 case SO_MARK:
4282                         if (sk->sk_mark != val) {
4283                                 sk->sk_mark = val;
4284                                 sk_dst_reset(sk);
4285                         }
4286                         break;
4287                 default:
4288                         ret = -EINVAL;
4289                 }
4290 #ifdef CONFIG_INET
4291         } else if (level == SOL_IP) {
4292                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4293                         return -EINVAL;
4294
4295                 val = *((int *)optval);
4296                 /* Only some options are supported */
4297                 switch (optname) {
4298                 case IP_TOS:
4299                         if (val < -1 || val > 0xff) {
4300                                 ret = -EINVAL;
4301                         } else {
4302                                 struct inet_sock *inet = inet_sk(sk);
4303
4304                                 if (val == -1)
4305                                         val = 0;
4306                                 inet->tos = val;
4307                         }
4308                         break;
4309                 default:
4310                         ret = -EINVAL;
4311                 }
4312 #if IS_ENABLED(CONFIG_IPV6)
4313         } else if (level == SOL_IPV6) {
4314                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4315                         return -EINVAL;
4316
4317                 val = *((int *)optval);
4318                 /* Only some options are supported */
4319                 switch (optname) {
4320                 case IPV6_TCLASS:
4321                         if (val < -1 || val > 0xff) {
4322                                 ret = -EINVAL;
4323                         } else {
4324                                 struct ipv6_pinfo *np = inet6_sk(sk);
4325
4326                                 if (val == -1)
4327                                         val = 0;
4328                                 np->tclass = val;
4329                         }
4330                         break;
4331                 default:
4332                         ret = -EINVAL;
4333                 }
4334 #endif
4335         } else if (level == SOL_TCP &&
4336                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4337                 if (optname == TCP_CONGESTION) {
4338                         char name[TCP_CA_NAME_MAX];
4339                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4340
4341                         strncpy(name, optval, min_t(long, optlen,
4342                                                     TCP_CA_NAME_MAX-1));
4343                         name[TCP_CA_NAME_MAX-1] = 0;
4344                         ret = tcp_set_congestion_control(sk, name, false,
4345                                                          reinit, true);
4346                 } else {
4347                         struct tcp_sock *tp = tcp_sk(sk);
4348
4349                         if (optlen != sizeof(int))
4350                                 return -EINVAL;
4351
4352                         val = *((int *)optval);
4353                         /* Only some options are supported */
4354                         switch (optname) {
4355                         case TCP_BPF_IW:
4356                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4357                                         ret = -EINVAL;
4358                                 else
4359                                         tp->snd_cwnd = val;
4360                                 break;
4361                         case TCP_BPF_SNDCWND_CLAMP:
4362                                 if (val <= 0) {
4363                                         ret = -EINVAL;
4364                                 } else {
4365                                         tp->snd_cwnd_clamp = val;
4366                                         tp->snd_ssthresh = val;
4367                                 }
4368                                 break;
4369                         case TCP_SAVE_SYN:
4370                                 if (val < 0 || val > 1)
4371                                         ret = -EINVAL;
4372                                 else
4373                                         tp->save_syn = val;
4374                                 break;
4375                         default:
4376                                 ret = -EINVAL;
4377                         }
4378                 }
4379 #endif
4380         } else {
4381                 ret = -EINVAL;
4382         }
4383         return ret;
4384 }
4385
4386 static const struct bpf_func_proto bpf_setsockopt_proto = {
4387         .func           = bpf_setsockopt,
4388         .gpl_only       = false,
4389         .ret_type       = RET_INTEGER,
4390         .arg1_type      = ARG_PTR_TO_CTX,
4391         .arg2_type      = ARG_ANYTHING,
4392         .arg3_type      = ARG_ANYTHING,
4393         .arg4_type      = ARG_PTR_TO_MEM,
4394         .arg5_type      = ARG_CONST_SIZE,
4395 };
4396
4397 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4398            int, level, int, optname, char *, optval, int, optlen)
4399 {
4400         struct sock *sk = bpf_sock->sk;
4401
4402         if (!sk_fullsock(sk))
4403                 goto err_clear;
4404 #ifdef CONFIG_INET
4405         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4406                 struct inet_connection_sock *icsk;
4407                 struct tcp_sock *tp;
4408
4409                 switch (optname) {
4410                 case TCP_CONGESTION:
4411                         icsk = inet_csk(sk);
4412
4413                         if (!icsk->icsk_ca_ops || optlen <= 1)
4414                                 goto err_clear;
4415                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4416                         optval[optlen - 1] = 0;
4417                         break;
4418                 case TCP_SAVED_SYN:
4419                         tp = tcp_sk(sk);
4420
4421                         if (optlen <= 0 || !tp->saved_syn ||
4422                             optlen > tp->saved_syn[0])
4423                                 goto err_clear;
4424                         memcpy(optval, tp->saved_syn + 1, optlen);
4425                         break;
4426                 default:
4427                         goto err_clear;
4428                 }
4429         } else if (level == SOL_IP) {
4430                 struct inet_sock *inet = inet_sk(sk);
4431
4432                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4433                         goto err_clear;
4434
4435                 /* Only some options are supported */
4436                 switch (optname) {
4437                 case IP_TOS:
4438                         *((int *)optval) = (int)inet->tos;
4439                         break;
4440                 default:
4441                         goto err_clear;
4442                 }
4443 #if IS_ENABLED(CONFIG_IPV6)
4444         } else if (level == SOL_IPV6) {
4445                 struct ipv6_pinfo *np = inet6_sk(sk);
4446
4447                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4448                         goto err_clear;
4449
4450                 /* Only some options are supported */
4451                 switch (optname) {
4452                 case IPV6_TCLASS:
4453                         *((int *)optval) = (int)np->tclass;
4454                         break;
4455                 default:
4456                         goto err_clear;
4457                 }
4458 #endif
4459         } else {
4460                 goto err_clear;
4461         }
4462         return 0;
4463 #endif
4464 err_clear:
4465         memset(optval, 0, optlen);
4466         return -EINVAL;
4467 }
4468
4469 static const struct bpf_func_proto bpf_getsockopt_proto = {
4470         .func           = bpf_getsockopt,
4471         .gpl_only       = false,
4472         .ret_type       = RET_INTEGER,
4473         .arg1_type      = ARG_PTR_TO_CTX,
4474         .arg2_type      = ARG_ANYTHING,
4475         .arg3_type      = ARG_ANYTHING,
4476         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4477         .arg5_type      = ARG_CONST_SIZE,
4478 };
4479
4480 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4481            int, argval)
4482 {
4483         struct sock *sk = bpf_sock->sk;
4484         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4485
4486         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4487                 return -EINVAL;
4488
4489         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4490
4491         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4492 }
4493
4494 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4495         .func           = bpf_sock_ops_cb_flags_set,
4496         .gpl_only       = false,
4497         .ret_type       = RET_INTEGER,
4498         .arg1_type      = ARG_PTR_TO_CTX,
4499         .arg2_type      = ARG_ANYTHING,
4500 };
4501
4502 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4503 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4504
4505 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4506            int, addr_len)
4507 {
4508 #ifdef CONFIG_INET
4509         struct sock *sk = ctx->sk;
4510         int err;
4511
4512         /* Binding to port can be expensive so it's prohibited in the helper.
4513          * Only binding to IP is supported.
4514          */
4515         err = -EINVAL;
4516         if (addr_len < offsetofend(struct sockaddr, sa_family))
4517                 return err;
4518         if (addr->sa_family == AF_INET) {
4519                 if (addr_len < sizeof(struct sockaddr_in))
4520                         return err;
4521                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4522                         return err;
4523                 return __inet_bind(sk, addr, addr_len, true, false);
4524 #if IS_ENABLED(CONFIG_IPV6)
4525         } else if (addr->sa_family == AF_INET6) {
4526                 if (addr_len < SIN6_LEN_RFC2133)
4527                         return err;
4528                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4529                         return err;
4530                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4531                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4532                  */
4533                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4534 #endif /* CONFIG_IPV6 */
4535         }
4536 #endif /* CONFIG_INET */
4537
4538         return -EAFNOSUPPORT;
4539 }
4540
4541 static const struct bpf_func_proto bpf_bind_proto = {
4542         .func           = bpf_bind,
4543         .gpl_only       = false,
4544         .ret_type       = RET_INTEGER,
4545         .arg1_type      = ARG_PTR_TO_CTX,
4546         .arg2_type      = ARG_PTR_TO_MEM,
4547         .arg3_type      = ARG_CONST_SIZE,
4548 };
4549
4550 #ifdef CONFIG_XFRM
4551 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4552            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4553 {
4554         const struct sec_path *sp = skb_sec_path(skb);
4555         const struct xfrm_state *x;
4556
4557         if (!sp || unlikely(index >= sp->len || flags))
4558                 goto err_clear;
4559
4560         x = sp->xvec[index];
4561
4562         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4563                 goto err_clear;
4564
4565         to->reqid = x->props.reqid;
4566         to->spi = x->id.spi;
4567         to->family = x->props.family;
4568         to->ext = 0;
4569
4570         if (to->family == AF_INET6) {
4571                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4572                        sizeof(to->remote_ipv6));
4573         } else {
4574                 to->remote_ipv4 = x->props.saddr.a4;
4575                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4576         }
4577
4578         return 0;
4579 err_clear:
4580         memset(to, 0, size);
4581         return -EINVAL;
4582 }
4583
4584 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4585         .func           = bpf_skb_get_xfrm_state,
4586         .gpl_only       = false,
4587         .ret_type       = RET_INTEGER,
4588         .arg1_type      = ARG_PTR_TO_CTX,
4589         .arg2_type      = ARG_ANYTHING,
4590         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4591         .arg4_type      = ARG_CONST_SIZE,
4592         .arg5_type      = ARG_ANYTHING,
4593 };
4594 #endif
4595
4596 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4597 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4598                                   const struct neighbour *neigh,
4599                                   const struct net_device *dev)
4600 {
4601         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4602         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4603         params->h_vlan_TCI = 0;
4604         params->h_vlan_proto = 0;
4605         params->ifindex = dev->ifindex;
4606
4607         return 0;
4608 }
4609 #endif
4610
4611 #if IS_ENABLED(CONFIG_INET)
4612 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4613                                u32 flags, bool check_mtu)
4614 {
4615         struct fib_nh_common *nhc;
4616         struct in_device *in_dev;
4617         struct neighbour *neigh;
4618         struct net_device *dev;
4619         struct fib_result res;
4620         struct flowi4 fl4;
4621         int err;
4622         u32 mtu;
4623
4624         dev = dev_get_by_index_rcu(net, params->ifindex);
4625         if (unlikely(!dev))
4626                 return -ENODEV;
4627
4628         /* verify forwarding is enabled on this interface */
4629         in_dev = __in_dev_get_rcu(dev);
4630         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4631                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4632
4633         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4634                 fl4.flowi4_iif = 1;
4635                 fl4.flowi4_oif = params->ifindex;
4636         } else {
4637                 fl4.flowi4_iif = params->ifindex;
4638                 fl4.flowi4_oif = 0;
4639         }
4640         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4641         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4642         fl4.flowi4_flags = 0;
4643
4644         fl4.flowi4_proto = params->l4_protocol;
4645         fl4.daddr = params->ipv4_dst;
4646         fl4.saddr = params->ipv4_src;
4647         fl4.fl4_sport = params->sport;
4648         fl4.fl4_dport = params->dport;
4649
4650         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4651                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4652                 struct fib_table *tb;
4653
4654                 tb = fib_get_table(net, tbid);
4655                 if (unlikely(!tb))
4656                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4657
4658                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4659         } else {
4660                 fl4.flowi4_mark = 0;
4661                 fl4.flowi4_secid = 0;
4662                 fl4.flowi4_tun_key.tun_id = 0;
4663                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4664
4665                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4666         }
4667
4668         if (err) {
4669                 /* map fib lookup errors to RTN_ type */
4670                 if (err == -EINVAL)
4671                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4672                 if (err == -EHOSTUNREACH)
4673                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4674                 if (err == -EACCES)
4675                         return BPF_FIB_LKUP_RET_PROHIBIT;
4676
4677                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4678         }
4679
4680         if (res.type != RTN_UNICAST)
4681                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4682
4683         if (fib_info_num_path(res.fi) > 1)
4684                 fib_select_path(net, &res, &fl4, NULL);
4685
4686         if (check_mtu) {
4687                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4688                 if (params->tot_len > mtu)
4689                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4690         }
4691
4692         nhc = res.nhc;
4693
4694         /* do not handle lwt encaps right now */
4695         if (nhc->nhc_lwtstate)
4696                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4697
4698         dev = nhc->nhc_dev;
4699
4700         params->rt_metric = res.fi->fib_priority;
4701
4702         /* xdp and cls_bpf programs are run in RCU-bh so
4703          * rcu_read_lock_bh is not needed here
4704          */
4705         if (likely(nhc->nhc_gw_family != AF_INET6)) {
4706                 if (nhc->nhc_gw_family)
4707                         params->ipv4_dst = nhc->nhc_gw.ipv4;
4708
4709                 neigh = __ipv4_neigh_lookup_noref(dev,
4710                                                  (__force u32)params->ipv4_dst);
4711         } else {
4712                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
4713
4714                 params->family = AF_INET6;
4715                 *dst = nhc->nhc_gw.ipv6;
4716                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4717         }
4718
4719         if (!neigh)
4720                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4721
4722         return bpf_fib_set_fwd_params(params, neigh, dev);
4723 }
4724 #endif
4725
4726 #if IS_ENABLED(CONFIG_IPV6)
4727 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4728                                u32 flags, bool check_mtu)
4729 {
4730         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4731         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4732         struct fib6_result res = {};
4733         struct neighbour *neigh;
4734         struct net_device *dev;
4735         struct inet6_dev *idev;
4736         struct flowi6 fl6;
4737         int strict = 0;
4738         int oif, err;
4739         u32 mtu;
4740
4741         /* link local addresses are never forwarded */
4742         if (rt6_need_strict(dst) || rt6_need_strict(src))
4743                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4744
4745         dev = dev_get_by_index_rcu(net, params->ifindex);
4746         if (unlikely(!dev))
4747                 return -ENODEV;
4748
4749         idev = __in6_dev_get_safely(dev);
4750         if (unlikely(!idev || !idev->cnf.forwarding))
4751                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4752
4753         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4754                 fl6.flowi6_iif = 1;
4755                 oif = fl6.flowi6_oif = params->ifindex;
4756         } else {
4757                 oif = fl6.flowi6_iif = params->ifindex;
4758                 fl6.flowi6_oif = 0;
4759                 strict = RT6_LOOKUP_F_HAS_SADDR;
4760         }
4761         fl6.flowlabel = params->flowinfo;
4762         fl6.flowi6_scope = 0;
4763         fl6.flowi6_flags = 0;
4764         fl6.mp_hash = 0;
4765
4766         fl6.flowi6_proto = params->l4_protocol;
4767         fl6.daddr = *dst;
4768         fl6.saddr = *src;
4769         fl6.fl6_sport = params->sport;
4770         fl6.fl6_dport = params->dport;
4771
4772         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4773                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4774                 struct fib6_table *tb;
4775
4776                 tb = ipv6_stub->fib6_get_table(net, tbid);
4777                 if (unlikely(!tb))
4778                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4779
4780                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
4781                                                    strict);
4782         } else {
4783                 fl6.flowi6_mark = 0;
4784                 fl6.flowi6_secid = 0;
4785                 fl6.flowi6_tun_key.tun_id = 0;
4786                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4787
4788                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
4789         }
4790
4791         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
4792                      res.f6i == net->ipv6.fib6_null_entry))
4793                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4794
4795         switch (res.fib6_type) {
4796         /* only unicast is forwarded */
4797         case RTN_UNICAST:
4798                 break;
4799         case RTN_BLACKHOLE:
4800                 return BPF_FIB_LKUP_RET_BLACKHOLE;
4801         case RTN_UNREACHABLE:
4802                 return BPF_FIB_LKUP_RET_UNREACHABLE;
4803         case RTN_PROHIBIT:
4804                 return BPF_FIB_LKUP_RET_PROHIBIT;
4805         default:
4806                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4807         }
4808
4809         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
4810                                     fl6.flowi6_oif != 0, NULL, strict);
4811
4812         if (check_mtu) {
4813                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
4814                 if (params->tot_len > mtu)
4815                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4816         }
4817
4818         if (res.nh->fib_nh_lws)
4819                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4820
4821         if (res.nh->fib_nh_gw_family)
4822                 *dst = res.nh->fib_nh_gw6;
4823
4824         dev = res.nh->fib_nh_dev;
4825         params->rt_metric = res.f6i->fib6_metric;
4826
4827         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4828          * not needed here.
4829          */
4830         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
4831         if (!neigh)
4832                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4833
4834         return bpf_fib_set_fwd_params(params, neigh, dev);
4835 }
4836 #endif
4837
4838 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4839            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4840 {
4841         if (plen < sizeof(*params))
4842                 return -EINVAL;
4843
4844         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4845                 return -EINVAL;
4846
4847         switch (params->family) {
4848 #if IS_ENABLED(CONFIG_INET)
4849         case AF_INET:
4850                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4851                                            flags, true);
4852 #endif
4853 #if IS_ENABLED(CONFIG_IPV6)
4854         case AF_INET6:
4855                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4856                                            flags, true);
4857 #endif
4858         }
4859         return -EAFNOSUPPORT;
4860 }
4861
4862 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4863         .func           = bpf_xdp_fib_lookup,
4864         .gpl_only       = true,
4865         .ret_type       = RET_INTEGER,
4866         .arg1_type      = ARG_PTR_TO_CTX,
4867         .arg2_type      = ARG_PTR_TO_MEM,
4868         .arg3_type      = ARG_CONST_SIZE,
4869         .arg4_type      = ARG_ANYTHING,
4870 };
4871
4872 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4873            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4874 {
4875         struct net *net = dev_net(skb->dev);
4876         int rc = -EAFNOSUPPORT;
4877
4878         if (plen < sizeof(*params))
4879                 return -EINVAL;
4880
4881         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4882                 return -EINVAL;
4883
4884         switch (params->family) {
4885 #if IS_ENABLED(CONFIG_INET)
4886         case AF_INET:
4887                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4888                 break;
4889 #endif
4890 #if IS_ENABLED(CONFIG_IPV6)
4891         case AF_INET6:
4892                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4893                 break;
4894 #endif
4895         }
4896
4897         if (!rc) {
4898                 struct net_device *dev;
4899
4900                 dev = dev_get_by_index_rcu(net, params->ifindex);
4901                 if (!is_skb_forwardable(dev, skb))
4902                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4903         }
4904
4905         return rc;
4906 }
4907
4908 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4909         .func           = bpf_skb_fib_lookup,
4910         .gpl_only       = true,
4911         .ret_type       = RET_INTEGER,
4912         .arg1_type      = ARG_PTR_TO_CTX,
4913         .arg2_type      = ARG_PTR_TO_MEM,
4914         .arg3_type      = ARG_CONST_SIZE,
4915         .arg4_type      = ARG_ANYTHING,
4916 };
4917
4918 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4919 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4920 {
4921         int err;
4922         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4923
4924         if (!seg6_validate_srh(srh, len))
4925                 return -EINVAL;
4926
4927         switch (type) {
4928         case BPF_LWT_ENCAP_SEG6_INLINE:
4929                 if (skb->protocol != htons(ETH_P_IPV6))
4930                         return -EBADMSG;
4931
4932                 err = seg6_do_srh_inline(skb, srh);
4933                 break;
4934         case BPF_LWT_ENCAP_SEG6:
4935                 skb_reset_inner_headers(skb);
4936                 skb->encapsulation = 1;
4937                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4938                 break;
4939         default:
4940                 return -EINVAL;
4941         }
4942
4943         bpf_compute_data_pointers(skb);
4944         if (err)
4945                 return err;
4946
4947         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4948         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4949
4950         return seg6_lookup_nexthop(skb, NULL, 0);
4951 }
4952 #endif /* CONFIG_IPV6_SEG6_BPF */
4953
4954 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4955 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
4956                              bool ingress)
4957 {
4958         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
4959 }
4960 #endif
4961
4962 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4963            u32, len)
4964 {
4965         switch (type) {
4966 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4967         case BPF_LWT_ENCAP_SEG6:
4968         case BPF_LWT_ENCAP_SEG6_INLINE:
4969                 return bpf_push_seg6_encap(skb, type, hdr, len);
4970 #endif
4971 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4972         case BPF_LWT_ENCAP_IP:
4973                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
4974 #endif
4975         default:
4976                 return -EINVAL;
4977         }
4978 }
4979
4980 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
4981            void *, hdr, u32, len)
4982 {
4983         switch (type) {
4984 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
4985         case BPF_LWT_ENCAP_IP:
4986                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
4987 #endif
4988         default:
4989                 return -EINVAL;
4990         }
4991 }
4992
4993 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
4994         .func           = bpf_lwt_in_push_encap,
4995         .gpl_only       = false,
4996         .ret_type       = RET_INTEGER,
4997         .arg1_type      = ARG_PTR_TO_CTX,
4998         .arg2_type      = ARG_ANYTHING,
4999         .arg3_type      = ARG_PTR_TO_MEM,
5000         .arg4_type      = ARG_CONST_SIZE
5001 };
5002
5003 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5004         .func           = bpf_lwt_xmit_push_encap,
5005         .gpl_only       = false,
5006         .ret_type       = RET_INTEGER,
5007         .arg1_type      = ARG_PTR_TO_CTX,
5008         .arg2_type      = ARG_ANYTHING,
5009         .arg3_type      = ARG_PTR_TO_MEM,
5010         .arg4_type      = ARG_CONST_SIZE
5011 };
5012
5013 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5014 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5015            const void *, from, u32, len)
5016 {
5017         struct seg6_bpf_srh_state *srh_state =
5018                 this_cpu_ptr(&seg6_bpf_srh_states);
5019         struct ipv6_sr_hdr *srh = srh_state->srh;
5020         void *srh_tlvs, *srh_end, *ptr;
5021         int srhoff = 0;
5022
5023         if (srh == NULL)
5024                 return -EINVAL;
5025
5026         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5027         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5028
5029         ptr = skb->data + offset;
5030         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5031                 srh_state->valid = false;
5032         else if (ptr < (void *)&srh->flags ||
5033                  ptr + len > (void *)&srh->segments)
5034                 return -EFAULT;
5035
5036         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5037                 return -EFAULT;
5038         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5039                 return -EINVAL;
5040         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5041
5042         memcpy(skb->data + offset, from, len);
5043         return 0;
5044 }
5045
5046 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5047         .func           = bpf_lwt_seg6_store_bytes,
5048         .gpl_only       = false,
5049         .ret_type       = RET_INTEGER,
5050         .arg1_type      = ARG_PTR_TO_CTX,
5051         .arg2_type      = ARG_ANYTHING,
5052         .arg3_type      = ARG_PTR_TO_MEM,
5053         .arg4_type      = ARG_CONST_SIZE
5054 };
5055
5056 static void bpf_update_srh_state(struct sk_buff *skb)
5057 {
5058         struct seg6_bpf_srh_state *srh_state =
5059                 this_cpu_ptr(&seg6_bpf_srh_states);
5060         int srhoff = 0;
5061
5062         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5063                 srh_state->srh = NULL;
5064         } else {
5065                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5066                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5067                 srh_state->valid = true;
5068         }
5069 }
5070
5071 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5072            u32, action, void *, param, u32, param_len)
5073 {
5074         struct seg6_bpf_srh_state *srh_state =
5075                 this_cpu_ptr(&seg6_bpf_srh_states);
5076         int hdroff = 0;
5077         int err;
5078
5079         switch (action) {
5080         case SEG6_LOCAL_ACTION_END_X:
5081                 if (!seg6_bpf_has_valid_srh(skb))
5082                         return -EBADMSG;
5083                 if (param_len != sizeof(struct in6_addr))
5084                         return -EINVAL;
5085                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5086         case SEG6_LOCAL_ACTION_END_T:
5087                 if (!seg6_bpf_has_valid_srh(skb))
5088                         return -EBADMSG;
5089                 if (param_len != sizeof(int))
5090                         return -EINVAL;
5091                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5092         case SEG6_LOCAL_ACTION_END_DT6:
5093                 if (!seg6_bpf_has_valid_srh(skb))
5094                         return -EBADMSG;
5095                 if (param_len != sizeof(int))
5096                         return -EINVAL;
5097
5098                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5099                         return -EBADMSG;
5100                 if (!pskb_pull(skb, hdroff))
5101                         return -EBADMSG;
5102
5103                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5104                 skb_reset_network_header(skb);
5105                 skb_reset_transport_header(skb);
5106                 skb->encapsulation = 0;
5107
5108                 bpf_compute_data_pointers(skb);
5109                 bpf_update_srh_state(skb);
5110                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5111         case SEG6_LOCAL_ACTION_END_B6:
5112                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5113                         return -EBADMSG;
5114                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5115                                           param, param_len);
5116                 if (!err)
5117                         bpf_update_srh_state(skb);
5118
5119                 return err;
5120         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5121                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5122                         return -EBADMSG;
5123                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5124                                           param, param_len);
5125                 if (!err)
5126                         bpf_update_srh_state(skb);
5127
5128                 return err;
5129         default:
5130                 return -EINVAL;
5131         }
5132 }
5133
5134 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5135         .func           = bpf_lwt_seg6_action,
5136         .gpl_only       = false,
5137         .ret_type       = RET_INTEGER,
5138         .arg1_type      = ARG_PTR_TO_CTX,
5139         .arg2_type      = ARG_ANYTHING,
5140         .arg3_type      = ARG_PTR_TO_MEM,
5141         .arg4_type      = ARG_CONST_SIZE
5142 };
5143
5144 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5145            s32, len)
5146 {
5147         struct seg6_bpf_srh_state *srh_state =
5148                 this_cpu_ptr(&seg6_bpf_srh_states);
5149         struct ipv6_sr_hdr *srh = srh_state->srh;
5150         void *srh_end, *srh_tlvs, *ptr;
5151         struct ipv6hdr *hdr;
5152         int srhoff = 0;
5153         int ret;
5154
5155         if (unlikely(srh == NULL))
5156                 return -EINVAL;
5157
5158         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5159                         ((srh->first_segment + 1) << 4));
5160         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5161                         srh_state->hdrlen);
5162         ptr = skb->data + offset;
5163
5164         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5165                 return -EFAULT;
5166         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5167                 return -EFAULT;
5168
5169         if (len > 0) {
5170                 ret = skb_cow_head(skb, len);
5171                 if (unlikely(ret < 0))
5172                         return ret;
5173
5174                 ret = bpf_skb_net_hdr_push(skb, offset, len);
5175         } else {
5176                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5177         }
5178
5179         bpf_compute_data_pointers(skb);
5180         if (unlikely(ret < 0))
5181                 return ret;
5182
5183         hdr = (struct ipv6hdr *)skb->data;
5184         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5185
5186         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5187                 return -EINVAL;
5188         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5189         srh_state->hdrlen += len;
5190         srh_state->valid = false;
5191         return 0;
5192 }
5193
5194 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5195         .func           = bpf_lwt_seg6_adjust_srh,
5196         .gpl_only       = false,
5197         .ret_type       = RET_INTEGER,
5198         .arg1_type      = ARG_PTR_TO_CTX,
5199         .arg2_type      = ARG_ANYTHING,
5200         .arg3_type      = ARG_ANYTHING,
5201 };
5202 #endif /* CONFIG_IPV6_SEG6_BPF */
5203
5204 #ifdef CONFIG_INET
5205 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5206                               int dif, int sdif, u8 family, u8 proto)
5207 {
5208         bool refcounted = false;
5209         struct sock *sk = NULL;
5210
5211         if (family == AF_INET) {
5212                 __be32 src4 = tuple->ipv4.saddr;
5213                 __be32 dst4 = tuple->ipv4.daddr;
5214
5215                 if (proto == IPPROTO_TCP)
5216                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5217                                            src4, tuple->ipv4.sport,
5218                                            dst4, tuple->ipv4.dport,
5219                                            dif, sdif, &refcounted);
5220                 else
5221                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5222                                                dst4, tuple->ipv4.dport,
5223                                                dif, sdif, &udp_table, NULL);
5224 #if IS_ENABLED(CONFIG_IPV6)
5225         } else {
5226                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5227                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5228
5229                 if (proto == IPPROTO_TCP)
5230                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5231                                             src6, tuple->ipv6.sport,
5232                                             dst6, ntohs(tuple->ipv6.dport),
5233                                             dif, sdif, &refcounted);
5234                 else if (likely(ipv6_bpf_stub))
5235                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5236                                                             src6, tuple->ipv6.sport,
5237                                                             dst6, tuple->ipv6.dport,
5238                                                             dif, sdif,
5239                                                             &udp_table, NULL);
5240 #endif
5241         }
5242
5243         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5244                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5245                 sk = NULL;
5246         }
5247         return sk;
5248 }
5249
5250 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5251  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5252  * Returns the socket as an 'unsigned long' to simplify the casting in the
5253  * callers to satisfy BPF_CALL declarations.
5254  */
5255 static struct sock *
5256 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5257                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5258                  u64 flags)
5259 {
5260         struct sock *sk = NULL;
5261         u8 family = AF_UNSPEC;
5262         struct net *net;
5263         int sdif;
5264
5265         if (len == sizeof(tuple->ipv4))
5266                 family = AF_INET;
5267         else if (len == sizeof(tuple->ipv6))
5268                 family = AF_INET6;
5269         else
5270                 return NULL;
5271
5272         if (unlikely(family == AF_UNSPEC || flags ||
5273                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5274                 goto out;
5275
5276         if (family == AF_INET)
5277                 sdif = inet_sdif(skb);
5278         else
5279                 sdif = inet6_sdif(skb);
5280
5281         if ((s32)netns_id < 0) {
5282                 net = caller_net;
5283                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5284         } else {
5285                 net = get_net_ns_by_id(caller_net, netns_id);
5286                 if (unlikely(!net))
5287                         goto out;
5288                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5289                 put_net(net);
5290         }
5291
5292 out:
5293         return sk;
5294 }
5295
5296 static struct sock *
5297 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5298                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5299                 u64 flags)
5300 {
5301         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5302                                            ifindex, proto, netns_id, flags);
5303
5304         if (sk) {
5305                 sk = sk_to_full_sk(sk);
5306                 if (!sk_fullsock(sk)) {
5307                         if (!sock_flag(sk, SOCK_RCU_FREE))
5308                                 sock_gen_put(sk);
5309                         return NULL;
5310                 }
5311         }
5312
5313         return sk;
5314 }
5315
5316 static struct sock *
5317 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5318                u8 proto, u64 netns_id, u64 flags)
5319 {
5320         struct net *caller_net;
5321         int ifindex;
5322
5323         if (skb->dev) {
5324                 caller_net = dev_net(skb->dev);
5325                 ifindex = skb->dev->ifindex;
5326         } else {
5327                 caller_net = sock_net(skb->sk);
5328                 ifindex = 0;
5329         }
5330
5331         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
5332                                 netns_id, flags);
5333 }
5334
5335 static struct sock *
5336 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5337               u8 proto, u64 netns_id, u64 flags)
5338 {
5339         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
5340                                          flags);
5341
5342         if (sk) {
5343                 sk = sk_to_full_sk(sk);
5344                 if (!sk_fullsock(sk)) {
5345                         if (!sock_flag(sk, SOCK_RCU_FREE))
5346                                 sock_gen_put(sk);
5347                         return NULL;
5348                 }
5349         }
5350
5351         return sk;
5352 }
5353
5354 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
5355            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5356 {
5357         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
5358                                              netns_id, flags);
5359 }
5360
5361 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
5362         .func           = bpf_skc_lookup_tcp,
5363         .gpl_only       = false,
5364         .pkt_access     = true,
5365         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5366         .arg1_type      = ARG_PTR_TO_CTX,
5367         .arg2_type      = ARG_PTR_TO_MEM,
5368         .arg3_type      = ARG_CONST_SIZE,
5369         .arg4_type      = ARG_ANYTHING,
5370         .arg5_type      = ARG_ANYTHING,
5371 };
5372
5373 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
5374            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5375 {
5376         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
5377                                             netns_id, flags);
5378 }
5379
5380 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
5381         .func           = bpf_sk_lookup_tcp,
5382         .gpl_only       = false,
5383         .pkt_access     = true,
5384         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5385         .arg1_type      = ARG_PTR_TO_CTX,
5386         .arg2_type      = ARG_PTR_TO_MEM,
5387         .arg3_type      = ARG_CONST_SIZE,
5388         .arg4_type      = ARG_ANYTHING,
5389         .arg5_type      = ARG_ANYTHING,
5390 };
5391
5392 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
5393            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5394 {
5395         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
5396                                             netns_id, flags);
5397 }
5398
5399 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
5400         .func           = bpf_sk_lookup_udp,
5401         .gpl_only       = false,
5402         .pkt_access     = true,
5403         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5404         .arg1_type      = ARG_PTR_TO_CTX,
5405         .arg2_type      = ARG_PTR_TO_MEM,
5406         .arg3_type      = ARG_CONST_SIZE,
5407         .arg4_type      = ARG_ANYTHING,
5408         .arg5_type      = ARG_ANYTHING,
5409 };
5410
5411 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
5412 {
5413         if (!sock_flag(sk, SOCK_RCU_FREE))
5414                 sock_gen_put(sk);
5415         return 0;
5416 }
5417
5418 static const struct bpf_func_proto bpf_sk_release_proto = {
5419         .func           = bpf_sk_release,
5420         .gpl_only       = false,
5421         .ret_type       = RET_INTEGER,
5422         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5423 };
5424
5425 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
5426            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5427 {
5428         struct net *caller_net = dev_net(ctx->rxq->dev);
5429         int ifindex = ctx->rxq->dev->ifindex;
5430
5431         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5432                                               ifindex, IPPROTO_UDP, netns_id,
5433                                               flags);
5434 }
5435
5436 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
5437         .func           = bpf_xdp_sk_lookup_udp,
5438         .gpl_only       = false,
5439         .pkt_access     = true,
5440         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5441         .arg1_type      = ARG_PTR_TO_CTX,
5442         .arg2_type      = ARG_PTR_TO_MEM,
5443         .arg3_type      = ARG_CONST_SIZE,
5444         .arg4_type      = ARG_ANYTHING,
5445         .arg5_type      = ARG_ANYTHING,
5446 };
5447
5448 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
5449            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5450 {
5451         struct net *caller_net = dev_net(ctx->rxq->dev);
5452         int ifindex = ctx->rxq->dev->ifindex;
5453
5454         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
5455                                                ifindex, IPPROTO_TCP, netns_id,
5456                                                flags);
5457 }
5458
5459 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
5460         .func           = bpf_xdp_skc_lookup_tcp,
5461         .gpl_only       = false,
5462         .pkt_access     = true,
5463         .ret_type       = RET_PTR_TO_SOCK_COMMON_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 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
5472            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
5473 {
5474         struct net *caller_net = dev_net(ctx->rxq->dev);
5475         int ifindex = ctx->rxq->dev->ifindex;
5476
5477         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
5478                                               ifindex, IPPROTO_TCP, netns_id,
5479                                               flags);
5480 }
5481
5482 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
5483         .func           = bpf_xdp_sk_lookup_tcp,
5484         .gpl_only       = false,
5485         .pkt_access     = true,
5486         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5487         .arg1_type      = ARG_PTR_TO_CTX,
5488         .arg2_type      = ARG_PTR_TO_MEM,
5489         .arg3_type      = ARG_CONST_SIZE,
5490         .arg4_type      = ARG_ANYTHING,
5491         .arg5_type      = ARG_ANYTHING,
5492 };
5493
5494 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5495            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5496 {
5497         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
5498                                                sock_net(ctx->sk), 0,
5499                                                IPPROTO_TCP, netns_id, flags);
5500 }
5501
5502 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
5503         .func           = bpf_sock_addr_skc_lookup_tcp,
5504         .gpl_only       = false,
5505         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
5506         .arg1_type      = ARG_PTR_TO_CTX,
5507         .arg2_type      = ARG_PTR_TO_MEM,
5508         .arg3_type      = ARG_CONST_SIZE,
5509         .arg4_type      = ARG_ANYTHING,
5510         .arg5_type      = ARG_ANYTHING,
5511 };
5512
5513 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
5514            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5515 {
5516         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5517                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
5518                                               netns_id, flags);
5519 }
5520
5521 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
5522         .func           = bpf_sock_addr_sk_lookup_tcp,
5523         .gpl_only       = false,
5524         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5525         .arg1_type      = ARG_PTR_TO_CTX,
5526         .arg2_type      = ARG_PTR_TO_MEM,
5527         .arg3_type      = ARG_CONST_SIZE,
5528         .arg4_type      = ARG_ANYTHING,
5529         .arg5_type      = ARG_ANYTHING,
5530 };
5531
5532 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
5533            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
5534 {
5535         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
5536                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
5537                                               netns_id, flags);
5538 }
5539
5540 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
5541         .func           = bpf_sock_addr_sk_lookup_udp,
5542         .gpl_only       = false,
5543         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5544         .arg1_type      = ARG_PTR_TO_CTX,
5545         .arg2_type      = ARG_PTR_TO_MEM,
5546         .arg3_type      = ARG_CONST_SIZE,
5547         .arg4_type      = ARG_ANYTHING,
5548         .arg5_type      = ARG_ANYTHING,
5549 };
5550
5551 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5552                                   struct bpf_insn_access_aux *info)
5553 {
5554         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
5555                                           icsk_retransmits))
5556                 return false;
5557
5558         if (off % size != 0)
5559                 return false;
5560
5561         switch (off) {
5562         case offsetof(struct bpf_tcp_sock, bytes_received):
5563         case offsetof(struct bpf_tcp_sock, bytes_acked):
5564                 return size == sizeof(__u64);
5565         default:
5566                 return size == sizeof(__u32);
5567         }
5568 }
5569
5570 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
5571                                     const struct bpf_insn *si,
5572                                     struct bpf_insn *insn_buf,
5573                                     struct bpf_prog *prog, u32 *target_size)
5574 {
5575         struct bpf_insn *insn = insn_buf;
5576
5577 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
5578         do {                                                            \
5579                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, FIELD) >     \
5580                              FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5581                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
5582                                       si->dst_reg, si->src_reg,         \
5583                                       offsetof(struct tcp_sock, FIELD)); \
5584         } while (0)
5585
5586 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
5587         do {                                                            \
5588                 BUILD_BUG_ON(FIELD_SIZEOF(struct inet_connection_sock,  \
5589                                           FIELD) >                      \
5590                              FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \
5591                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
5592                                         struct inet_connection_sock,    \
5593                                         FIELD),                         \
5594                                       si->dst_reg, si->src_reg,         \
5595                                       offsetof(                         \
5596                                         struct inet_connection_sock,    \
5597                                         FIELD));                        \
5598         } while (0)
5599
5600         if (insn > insn_buf)
5601                 return insn - insn_buf;
5602
5603         switch (si->off) {
5604         case offsetof(struct bpf_tcp_sock, rtt_min):
5605                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
5606                              sizeof(struct minmax));
5607                 BUILD_BUG_ON(sizeof(struct minmax) <
5608                              sizeof(struct minmax_sample));
5609
5610                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5611                                       offsetof(struct tcp_sock, rtt_min) +
5612                                       offsetof(struct minmax_sample, v));
5613                 break;
5614         case offsetof(struct bpf_tcp_sock, snd_cwnd):
5615                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
5616                 break;
5617         case offsetof(struct bpf_tcp_sock, srtt_us):
5618                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
5619                 break;
5620         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
5621                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
5622                 break;
5623         case offsetof(struct bpf_tcp_sock, rcv_nxt):
5624                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
5625                 break;
5626         case offsetof(struct bpf_tcp_sock, snd_nxt):
5627                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
5628                 break;
5629         case offsetof(struct bpf_tcp_sock, snd_una):
5630                 BPF_TCP_SOCK_GET_COMMON(snd_una);
5631                 break;
5632         case offsetof(struct bpf_tcp_sock, mss_cache):
5633                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
5634                 break;
5635         case offsetof(struct bpf_tcp_sock, ecn_flags):
5636                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
5637                 break;
5638         case offsetof(struct bpf_tcp_sock, rate_delivered):
5639                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
5640                 break;
5641         case offsetof(struct bpf_tcp_sock, rate_interval_us):
5642                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
5643                 break;
5644         case offsetof(struct bpf_tcp_sock, packets_out):
5645                 BPF_TCP_SOCK_GET_COMMON(packets_out);
5646                 break;
5647         case offsetof(struct bpf_tcp_sock, retrans_out):
5648                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
5649                 break;
5650         case offsetof(struct bpf_tcp_sock, total_retrans):
5651                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
5652                 break;
5653         case offsetof(struct bpf_tcp_sock, segs_in):
5654                 BPF_TCP_SOCK_GET_COMMON(segs_in);
5655                 break;
5656         case offsetof(struct bpf_tcp_sock, data_segs_in):
5657                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
5658                 break;
5659         case offsetof(struct bpf_tcp_sock, segs_out):
5660                 BPF_TCP_SOCK_GET_COMMON(segs_out);
5661                 break;
5662         case offsetof(struct bpf_tcp_sock, data_segs_out):
5663                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
5664                 break;
5665         case offsetof(struct bpf_tcp_sock, lost_out):
5666                 BPF_TCP_SOCK_GET_COMMON(lost_out);
5667                 break;
5668         case offsetof(struct bpf_tcp_sock, sacked_out):
5669                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
5670                 break;
5671         case offsetof(struct bpf_tcp_sock, bytes_received):
5672                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
5673                 break;
5674         case offsetof(struct bpf_tcp_sock, bytes_acked):
5675                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
5676                 break;
5677         case offsetof(struct bpf_tcp_sock, dsack_dups):
5678                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
5679                 break;
5680         case offsetof(struct bpf_tcp_sock, delivered):
5681                 BPF_TCP_SOCK_GET_COMMON(delivered);
5682                 break;
5683         case offsetof(struct bpf_tcp_sock, delivered_ce):
5684                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
5685                 break;
5686         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
5687                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
5688                 break;
5689         }
5690
5691         return insn - insn_buf;
5692 }
5693
5694 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
5695 {
5696         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
5697                 return (unsigned long)sk;
5698
5699         return (unsigned long)NULL;
5700 }
5701
5702 const struct bpf_func_proto bpf_tcp_sock_proto = {
5703         .func           = bpf_tcp_sock,
5704         .gpl_only       = false,
5705         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
5706         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5707 };
5708
5709 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
5710 {
5711         sk = sk_to_full_sk(sk);
5712
5713         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
5714                 return (unsigned long)sk;
5715
5716         return (unsigned long)NULL;
5717 }
5718
5719 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
5720         .func           = bpf_get_listener_sock,
5721         .gpl_only       = false,
5722         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
5723         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5724 };
5725
5726 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
5727 {
5728         unsigned int iphdr_len;
5729
5730         if (skb->protocol == cpu_to_be16(ETH_P_IP))
5731                 iphdr_len = sizeof(struct iphdr);
5732         else if (skb->protocol == cpu_to_be16(ETH_P_IPV6))
5733                 iphdr_len = sizeof(struct ipv6hdr);
5734         else
5735                 return 0;
5736
5737         if (skb_headlen(skb) < iphdr_len)
5738                 return 0;
5739
5740         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
5741                 return 0;
5742
5743         return INET_ECN_set_ce(skb);
5744 }
5745
5746 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5747                                   struct bpf_insn_access_aux *info)
5748 {
5749         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
5750                 return false;
5751
5752         if (off % size != 0)
5753                 return false;
5754
5755         switch (off) {
5756         default:
5757                 return size == sizeof(__u32);
5758         }
5759 }
5760
5761 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
5762                                     const struct bpf_insn *si,
5763                                     struct bpf_insn *insn_buf,
5764                                     struct bpf_prog *prog, u32 *target_size)
5765 {
5766         struct bpf_insn *insn = insn_buf;
5767
5768 #define BPF_XDP_SOCK_GET(FIELD)                                         \
5769         do {                                                            \
5770                 BUILD_BUG_ON(FIELD_SIZEOF(struct xdp_sock, FIELD) >     \
5771                              FIELD_SIZEOF(struct bpf_xdp_sock, FIELD)); \
5772                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
5773                                       si->dst_reg, si->src_reg,         \
5774                                       offsetof(struct xdp_sock, FIELD)); \
5775         } while (0)
5776
5777         switch (si->off) {
5778         case offsetof(struct bpf_xdp_sock, queue_id):
5779                 BPF_XDP_SOCK_GET(queue_id);
5780                 break;
5781         }
5782
5783         return insn - insn_buf;
5784 }
5785
5786 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
5787         .func           = bpf_skb_ecn_set_ce,
5788         .gpl_only       = false,
5789         .ret_type       = RET_INTEGER,
5790         .arg1_type      = ARG_PTR_TO_CTX,
5791 };
5792
5793 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5794            struct tcphdr *, th, u32, th_len)
5795 {
5796 #ifdef CONFIG_SYN_COOKIES
5797         u32 cookie;
5798         int ret;
5799
5800         if (unlikely(th_len < sizeof(*th)))
5801                 return -EINVAL;
5802
5803         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
5804         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5805                 return -EINVAL;
5806
5807         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5808                 return -EINVAL;
5809
5810         if (!th->ack || th->rst || th->syn)
5811                 return -ENOENT;
5812
5813         if (tcp_synq_no_recent_overflow(sk))
5814                 return -ENOENT;
5815
5816         cookie = ntohl(th->ack_seq) - 1;
5817
5818         switch (sk->sk_family) {
5819         case AF_INET:
5820                 if (unlikely(iph_len < sizeof(struct iphdr)))
5821                         return -EINVAL;
5822
5823                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
5824                 break;
5825
5826 #if IS_BUILTIN(CONFIG_IPV6)
5827         case AF_INET6:
5828                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5829                         return -EINVAL;
5830
5831                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
5832                 break;
5833 #endif /* CONFIG_IPV6 */
5834
5835         default:
5836                 return -EPROTONOSUPPORT;
5837         }
5838
5839         if (ret > 0)
5840                 return 0;
5841
5842         return -ENOENT;
5843 #else
5844         return -ENOTSUPP;
5845 #endif
5846 }
5847
5848 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
5849         .func           = bpf_tcp_check_syncookie,
5850         .gpl_only       = true,
5851         .pkt_access     = true,
5852         .ret_type       = RET_INTEGER,
5853         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5854         .arg2_type      = ARG_PTR_TO_MEM,
5855         .arg3_type      = ARG_CONST_SIZE,
5856         .arg4_type      = ARG_PTR_TO_MEM,
5857         .arg5_type      = ARG_CONST_SIZE,
5858 };
5859
5860 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
5861            struct tcphdr *, th, u32, th_len)
5862 {
5863 #ifdef CONFIG_SYN_COOKIES
5864         u32 cookie;
5865         u16 mss;
5866
5867         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
5868                 return -EINVAL;
5869
5870         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
5871                 return -EINVAL;
5872
5873         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
5874                 return -ENOENT;
5875
5876         if (!th->syn || th->ack || th->fin || th->rst)
5877                 return -EINVAL;
5878
5879         if (unlikely(iph_len < sizeof(struct iphdr)))
5880                 return -EINVAL;
5881
5882         /* Both struct iphdr and struct ipv6hdr have the version field at the
5883          * same offset so we can cast to the shorter header (struct iphdr).
5884          */
5885         switch (((struct iphdr *)iph)->version) {
5886         case 4:
5887                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
5888                         return -EINVAL;
5889
5890                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
5891                 break;
5892
5893 #if IS_BUILTIN(CONFIG_IPV6)
5894         case 6:
5895                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
5896                         return -EINVAL;
5897
5898                 if (sk->sk_family != AF_INET6)
5899                         return -EINVAL;
5900
5901                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
5902                 break;
5903 #endif /* CONFIG_IPV6 */
5904
5905         default:
5906                 return -EPROTONOSUPPORT;
5907         }
5908         if (mss == 0)
5909                 return -ENOENT;
5910
5911         return cookie | ((u64)mss << 32);
5912 #else
5913         return -EOPNOTSUPP;
5914 #endif /* CONFIG_SYN_COOKIES */
5915 }
5916
5917 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
5918         .func           = bpf_tcp_gen_syncookie,
5919         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
5920         .pkt_access     = true,
5921         .ret_type       = RET_INTEGER,
5922         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
5923         .arg2_type      = ARG_PTR_TO_MEM,
5924         .arg3_type      = ARG_CONST_SIZE,
5925         .arg4_type      = ARG_PTR_TO_MEM,
5926         .arg5_type      = ARG_CONST_SIZE,
5927 };
5928
5929 #endif /* CONFIG_INET */
5930
5931 bool bpf_helper_changes_pkt_data(void *func)
5932 {
5933         if (func == bpf_skb_vlan_push ||
5934             func == bpf_skb_vlan_pop ||
5935             func == bpf_skb_store_bytes ||
5936             func == bpf_skb_change_proto ||
5937             func == bpf_skb_change_head ||
5938             func == sk_skb_change_head ||
5939             func == bpf_skb_change_tail ||
5940             func == sk_skb_change_tail ||
5941             func == bpf_skb_adjust_room ||
5942             func == bpf_skb_pull_data ||
5943             func == sk_skb_pull_data ||
5944             func == bpf_clone_redirect ||
5945             func == bpf_l3_csum_replace ||
5946             func == bpf_l4_csum_replace ||
5947             func == bpf_xdp_adjust_head ||
5948             func == bpf_xdp_adjust_meta ||
5949             func == bpf_msg_pull_data ||
5950             func == bpf_msg_push_data ||
5951             func == bpf_msg_pop_data ||
5952             func == bpf_xdp_adjust_tail ||
5953 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5954             func == bpf_lwt_seg6_store_bytes ||
5955             func == bpf_lwt_seg6_adjust_srh ||
5956             func == bpf_lwt_seg6_action ||
5957 #endif
5958             func == bpf_lwt_in_push_encap ||
5959             func == bpf_lwt_xmit_push_encap)
5960                 return true;
5961
5962         return false;
5963 }
5964
5965 static const struct bpf_func_proto *
5966 bpf_base_func_proto(enum bpf_func_id func_id)
5967 {
5968         switch (func_id) {
5969         case BPF_FUNC_map_lookup_elem:
5970                 return &bpf_map_lookup_elem_proto;
5971         case BPF_FUNC_map_update_elem:
5972                 return &bpf_map_update_elem_proto;
5973         case BPF_FUNC_map_delete_elem:
5974                 return &bpf_map_delete_elem_proto;
5975         case BPF_FUNC_map_push_elem:
5976                 return &bpf_map_push_elem_proto;
5977         case BPF_FUNC_map_pop_elem:
5978                 return &bpf_map_pop_elem_proto;
5979         case BPF_FUNC_map_peek_elem:
5980                 return &bpf_map_peek_elem_proto;
5981         case BPF_FUNC_get_prandom_u32:
5982                 return &bpf_get_prandom_u32_proto;
5983         case BPF_FUNC_get_smp_processor_id:
5984                 return &bpf_get_raw_smp_processor_id_proto;
5985         case BPF_FUNC_get_numa_node_id:
5986                 return &bpf_get_numa_node_id_proto;
5987         case BPF_FUNC_tail_call:
5988                 return &bpf_tail_call_proto;
5989         case BPF_FUNC_ktime_get_ns:
5990                 return &bpf_ktime_get_ns_proto;
5991         default:
5992                 break;
5993         }
5994
5995         if (!capable(CAP_SYS_ADMIN))
5996                 return NULL;
5997
5998         switch (func_id) {
5999         case BPF_FUNC_spin_lock:
6000                 return &bpf_spin_lock_proto;
6001         case BPF_FUNC_spin_unlock:
6002                 return &bpf_spin_unlock_proto;
6003         case BPF_FUNC_trace_printk:
6004                 return bpf_get_trace_printk_proto();
6005         default:
6006                 return NULL;
6007         }
6008 }
6009
6010 static const struct bpf_func_proto *
6011 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6012 {
6013         switch (func_id) {
6014         /* inet and inet6 sockets are created in a process
6015          * context so there is always a valid uid/gid
6016          */
6017         case BPF_FUNC_get_current_uid_gid:
6018                 return &bpf_get_current_uid_gid_proto;
6019         case BPF_FUNC_get_local_storage:
6020                 return &bpf_get_local_storage_proto;
6021         default:
6022                 return bpf_base_func_proto(func_id);
6023         }
6024 }
6025
6026 static const struct bpf_func_proto *
6027 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6028 {
6029         switch (func_id) {
6030         /* inet and inet6 sockets are created in a process
6031          * context so there is always a valid uid/gid
6032          */
6033         case BPF_FUNC_get_current_uid_gid:
6034                 return &bpf_get_current_uid_gid_proto;
6035         case BPF_FUNC_bind:
6036                 switch (prog->expected_attach_type) {
6037                 case BPF_CGROUP_INET4_CONNECT:
6038                 case BPF_CGROUP_INET6_CONNECT:
6039                         return &bpf_bind_proto;
6040                 default:
6041                         return NULL;
6042                 }
6043         case BPF_FUNC_get_socket_cookie:
6044                 return &bpf_get_socket_cookie_sock_addr_proto;
6045         case BPF_FUNC_get_local_storage:
6046                 return &bpf_get_local_storage_proto;
6047 #ifdef CONFIG_INET
6048         case BPF_FUNC_sk_lookup_tcp:
6049                 return &bpf_sock_addr_sk_lookup_tcp_proto;
6050         case BPF_FUNC_sk_lookup_udp:
6051                 return &bpf_sock_addr_sk_lookup_udp_proto;
6052         case BPF_FUNC_sk_release:
6053                 return &bpf_sk_release_proto;
6054         case BPF_FUNC_skc_lookup_tcp:
6055                 return &bpf_sock_addr_skc_lookup_tcp_proto;
6056 #endif /* CONFIG_INET */
6057         case BPF_FUNC_sk_storage_get:
6058                 return &bpf_sk_storage_get_proto;
6059         case BPF_FUNC_sk_storage_delete:
6060                 return &bpf_sk_storage_delete_proto;
6061         default:
6062                 return bpf_base_func_proto(func_id);
6063         }
6064 }
6065
6066 static const struct bpf_func_proto *
6067 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6068 {
6069         switch (func_id) {
6070         case BPF_FUNC_skb_load_bytes:
6071                 return &bpf_skb_load_bytes_proto;
6072         case BPF_FUNC_skb_load_bytes_relative:
6073                 return &bpf_skb_load_bytes_relative_proto;
6074         case BPF_FUNC_get_socket_cookie:
6075                 return &bpf_get_socket_cookie_proto;
6076         case BPF_FUNC_get_socket_uid:
6077                 return &bpf_get_socket_uid_proto;
6078         case BPF_FUNC_perf_event_output:
6079                 return &bpf_skb_event_output_proto;
6080         default:
6081                 return bpf_base_func_proto(func_id);
6082         }
6083 }
6084
6085 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
6086 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
6087
6088 static const struct bpf_func_proto *
6089 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6090 {
6091         switch (func_id) {
6092         case BPF_FUNC_get_local_storage:
6093                 return &bpf_get_local_storage_proto;
6094         case BPF_FUNC_sk_fullsock:
6095                 return &bpf_sk_fullsock_proto;
6096         case BPF_FUNC_sk_storage_get:
6097                 return &bpf_sk_storage_get_proto;
6098         case BPF_FUNC_sk_storage_delete:
6099                 return &bpf_sk_storage_delete_proto;
6100         case BPF_FUNC_perf_event_output:
6101                 return &bpf_skb_event_output_proto;
6102 #ifdef CONFIG_SOCK_CGROUP_DATA
6103         case BPF_FUNC_skb_cgroup_id:
6104                 return &bpf_skb_cgroup_id_proto;
6105 #endif
6106 #ifdef CONFIG_INET
6107         case BPF_FUNC_tcp_sock:
6108                 return &bpf_tcp_sock_proto;
6109         case BPF_FUNC_get_listener_sock:
6110                 return &bpf_get_listener_sock_proto;
6111         case BPF_FUNC_skb_ecn_set_ce:
6112                 return &bpf_skb_ecn_set_ce_proto;
6113 #endif
6114         default:
6115                 return sk_filter_func_proto(func_id, prog);
6116         }
6117 }
6118
6119 static const struct bpf_func_proto *
6120 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6121 {
6122         switch (func_id) {
6123         case BPF_FUNC_skb_store_bytes:
6124                 return &bpf_skb_store_bytes_proto;
6125         case BPF_FUNC_skb_load_bytes:
6126                 return &bpf_skb_load_bytes_proto;
6127         case BPF_FUNC_skb_load_bytes_relative:
6128                 return &bpf_skb_load_bytes_relative_proto;
6129         case BPF_FUNC_skb_pull_data:
6130                 return &bpf_skb_pull_data_proto;
6131         case BPF_FUNC_csum_diff:
6132                 return &bpf_csum_diff_proto;
6133         case BPF_FUNC_csum_update:
6134                 return &bpf_csum_update_proto;
6135         case BPF_FUNC_l3_csum_replace:
6136                 return &bpf_l3_csum_replace_proto;
6137         case BPF_FUNC_l4_csum_replace:
6138                 return &bpf_l4_csum_replace_proto;
6139         case BPF_FUNC_clone_redirect:
6140                 return &bpf_clone_redirect_proto;
6141         case BPF_FUNC_get_cgroup_classid:
6142                 return &bpf_get_cgroup_classid_proto;
6143         case BPF_FUNC_skb_vlan_push:
6144                 return &bpf_skb_vlan_push_proto;
6145         case BPF_FUNC_skb_vlan_pop:
6146                 return &bpf_skb_vlan_pop_proto;
6147         case BPF_FUNC_skb_change_proto:
6148                 return &bpf_skb_change_proto_proto;
6149         case BPF_FUNC_skb_change_type:
6150                 return &bpf_skb_change_type_proto;
6151         case BPF_FUNC_skb_adjust_room:
6152                 return &bpf_skb_adjust_room_proto;
6153         case BPF_FUNC_skb_change_tail:
6154                 return &bpf_skb_change_tail_proto;
6155         case BPF_FUNC_skb_get_tunnel_key:
6156                 return &bpf_skb_get_tunnel_key_proto;
6157         case BPF_FUNC_skb_set_tunnel_key:
6158                 return bpf_get_skb_set_tunnel_proto(func_id);
6159         case BPF_FUNC_skb_get_tunnel_opt:
6160                 return &bpf_skb_get_tunnel_opt_proto;
6161         case BPF_FUNC_skb_set_tunnel_opt:
6162                 return bpf_get_skb_set_tunnel_proto(func_id);
6163         case BPF_FUNC_redirect:
6164                 return &bpf_redirect_proto;
6165         case BPF_FUNC_get_route_realm:
6166                 return &bpf_get_route_realm_proto;
6167         case BPF_FUNC_get_hash_recalc:
6168                 return &bpf_get_hash_recalc_proto;
6169         case BPF_FUNC_set_hash_invalid:
6170                 return &bpf_set_hash_invalid_proto;
6171         case BPF_FUNC_set_hash:
6172                 return &bpf_set_hash_proto;
6173         case BPF_FUNC_perf_event_output:
6174                 return &bpf_skb_event_output_proto;
6175         case BPF_FUNC_get_smp_processor_id:
6176                 return &bpf_get_smp_processor_id_proto;
6177         case BPF_FUNC_skb_under_cgroup:
6178                 return &bpf_skb_under_cgroup_proto;
6179         case BPF_FUNC_get_socket_cookie:
6180                 return &bpf_get_socket_cookie_proto;
6181         case BPF_FUNC_get_socket_uid:
6182                 return &bpf_get_socket_uid_proto;
6183         case BPF_FUNC_fib_lookup:
6184                 return &bpf_skb_fib_lookup_proto;
6185         case BPF_FUNC_sk_fullsock:
6186                 return &bpf_sk_fullsock_proto;
6187         case BPF_FUNC_sk_storage_get:
6188                 return &bpf_sk_storage_get_proto;
6189         case BPF_FUNC_sk_storage_delete:
6190                 return &bpf_sk_storage_delete_proto;
6191 #ifdef CONFIG_XFRM
6192         case BPF_FUNC_skb_get_xfrm_state:
6193                 return &bpf_skb_get_xfrm_state_proto;
6194 #endif
6195 #ifdef CONFIG_SOCK_CGROUP_DATA
6196         case BPF_FUNC_skb_cgroup_id:
6197                 return &bpf_skb_cgroup_id_proto;
6198         case BPF_FUNC_skb_ancestor_cgroup_id:
6199                 return &bpf_skb_ancestor_cgroup_id_proto;
6200 #endif
6201 #ifdef CONFIG_INET
6202         case BPF_FUNC_sk_lookup_tcp:
6203                 return &bpf_sk_lookup_tcp_proto;
6204         case BPF_FUNC_sk_lookup_udp:
6205                 return &bpf_sk_lookup_udp_proto;
6206         case BPF_FUNC_sk_release:
6207                 return &bpf_sk_release_proto;
6208         case BPF_FUNC_tcp_sock:
6209                 return &bpf_tcp_sock_proto;
6210         case BPF_FUNC_get_listener_sock:
6211                 return &bpf_get_listener_sock_proto;
6212         case BPF_FUNC_skc_lookup_tcp:
6213                 return &bpf_skc_lookup_tcp_proto;
6214         case BPF_FUNC_tcp_check_syncookie:
6215                 return &bpf_tcp_check_syncookie_proto;
6216         case BPF_FUNC_skb_ecn_set_ce:
6217                 return &bpf_skb_ecn_set_ce_proto;
6218         case BPF_FUNC_tcp_gen_syncookie:
6219                 return &bpf_tcp_gen_syncookie_proto;
6220 #endif
6221         default:
6222                 return bpf_base_func_proto(func_id);
6223         }
6224 }
6225
6226 static const struct bpf_func_proto *
6227 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6228 {
6229         switch (func_id) {
6230         case BPF_FUNC_perf_event_output:
6231                 return &bpf_xdp_event_output_proto;
6232         case BPF_FUNC_get_smp_processor_id:
6233                 return &bpf_get_smp_processor_id_proto;
6234         case BPF_FUNC_csum_diff:
6235                 return &bpf_csum_diff_proto;
6236         case BPF_FUNC_xdp_adjust_head:
6237                 return &bpf_xdp_adjust_head_proto;
6238         case BPF_FUNC_xdp_adjust_meta:
6239                 return &bpf_xdp_adjust_meta_proto;
6240         case BPF_FUNC_redirect:
6241                 return &bpf_xdp_redirect_proto;
6242         case BPF_FUNC_redirect_map:
6243                 return &bpf_xdp_redirect_map_proto;
6244         case BPF_FUNC_xdp_adjust_tail:
6245                 return &bpf_xdp_adjust_tail_proto;
6246         case BPF_FUNC_fib_lookup:
6247                 return &bpf_xdp_fib_lookup_proto;
6248 #ifdef CONFIG_INET
6249         case BPF_FUNC_sk_lookup_udp:
6250                 return &bpf_xdp_sk_lookup_udp_proto;
6251         case BPF_FUNC_sk_lookup_tcp:
6252                 return &bpf_xdp_sk_lookup_tcp_proto;
6253         case BPF_FUNC_sk_release:
6254                 return &bpf_sk_release_proto;
6255         case BPF_FUNC_skc_lookup_tcp:
6256                 return &bpf_xdp_skc_lookup_tcp_proto;
6257         case BPF_FUNC_tcp_check_syncookie:
6258                 return &bpf_tcp_check_syncookie_proto;
6259         case BPF_FUNC_tcp_gen_syncookie:
6260                 return &bpf_tcp_gen_syncookie_proto;
6261 #endif
6262         default:
6263                 return bpf_base_func_proto(func_id);
6264         }
6265 }
6266
6267 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
6268 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
6269
6270 static const struct bpf_func_proto *
6271 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6272 {
6273         switch (func_id) {
6274         case BPF_FUNC_setsockopt:
6275                 return &bpf_setsockopt_proto;
6276         case BPF_FUNC_getsockopt:
6277                 return &bpf_getsockopt_proto;
6278         case BPF_FUNC_sock_ops_cb_flags_set:
6279                 return &bpf_sock_ops_cb_flags_set_proto;
6280         case BPF_FUNC_sock_map_update:
6281                 return &bpf_sock_map_update_proto;
6282         case BPF_FUNC_sock_hash_update:
6283                 return &bpf_sock_hash_update_proto;
6284         case BPF_FUNC_get_socket_cookie:
6285                 return &bpf_get_socket_cookie_sock_ops_proto;
6286         case BPF_FUNC_get_local_storage:
6287                 return &bpf_get_local_storage_proto;
6288         case BPF_FUNC_perf_event_output:
6289                 return &bpf_sockopt_event_output_proto;
6290         case BPF_FUNC_sk_storage_get:
6291                 return &bpf_sk_storage_get_proto;
6292         case BPF_FUNC_sk_storage_delete:
6293                 return &bpf_sk_storage_delete_proto;
6294 #ifdef CONFIG_INET
6295         case BPF_FUNC_tcp_sock:
6296                 return &bpf_tcp_sock_proto;
6297 #endif /* CONFIG_INET */
6298         default:
6299                 return bpf_base_func_proto(func_id);
6300         }
6301 }
6302
6303 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
6304 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
6305
6306 static const struct bpf_func_proto *
6307 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6308 {
6309         switch (func_id) {
6310         case BPF_FUNC_msg_redirect_map:
6311                 return &bpf_msg_redirect_map_proto;
6312         case BPF_FUNC_msg_redirect_hash:
6313                 return &bpf_msg_redirect_hash_proto;
6314         case BPF_FUNC_msg_apply_bytes:
6315                 return &bpf_msg_apply_bytes_proto;
6316         case BPF_FUNC_msg_cork_bytes:
6317                 return &bpf_msg_cork_bytes_proto;
6318         case BPF_FUNC_msg_pull_data:
6319                 return &bpf_msg_pull_data_proto;
6320         case BPF_FUNC_msg_push_data:
6321                 return &bpf_msg_push_data_proto;
6322         case BPF_FUNC_msg_pop_data:
6323                 return &bpf_msg_pop_data_proto;
6324         default:
6325                 return bpf_base_func_proto(func_id);
6326         }
6327 }
6328
6329 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
6330 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
6331
6332 static const struct bpf_func_proto *
6333 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6334 {
6335         switch (func_id) {
6336         case BPF_FUNC_skb_store_bytes:
6337                 return &bpf_skb_store_bytes_proto;
6338         case BPF_FUNC_skb_load_bytes:
6339                 return &bpf_skb_load_bytes_proto;
6340         case BPF_FUNC_skb_pull_data:
6341                 return &sk_skb_pull_data_proto;
6342         case BPF_FUNC_skb_change_tail:
6343                 return &sk_skb_change_tail_proto;
6344         case BPF_FUNC_skb_change_head:
6345                 return &sk_skb_change_head_proto;
6346         case BPF_FUNC_get_socket_cookie:
6347                 return &bpf_get_socket_cookie_proto;
6348         case BPF_FUNC_get_socket_uid:
6349                 return &bpf_get_socket_uid_proto;
6350         case BPF_FUNC_sk_redirect_map:
6351                 return &bpf_sk_redirect_map_proto;
6352         case BPF_FUNC_sk_redirect_hash:
6353                 return &bpf_sk_redirect_hash_proto;
6354         case BPF_FUNC_perf_event_output:
6355                 return &bpf_skb_event_output_proto;
6356 #ifdef CONFIG_INET
6357         case BPF_FUNC_sk_lookup_tcp:
6358                 return &bpf_sk_lookup_tcp_proto;
6359         case BPF_FUNC_sk_lookup_udp:
6360                 return &bpf_sk_lookup_udp_proto;
6361         case BPF_FUNC_sk_release:
6362                 return &bpf_sk_release_proto;
6363         case BPF_FUNC_skc_lookup_tcp:
6364                 return &bpf_skc_lookup_tcp_proto;
6365 #endif
6366         default:
6367                 return bpf_base_func_proto(func_id);
6368         }
6369 }
6370
6371 static const struct bpf_func_proto *
6372 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6373 {
6374         switch (func_id) {
6375         case BPF_FUNC_skb_load_bytes:
6376                 return &bpf_flow_dissector_load_bytes_proto;
6377         default:
6378                 return bpf_base_func_proto(func_id);
6379         }
6380 }
6381
6382 static const struct bpf_func_proto *
6383 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6384 {
6385         switch (func_id) {
6386         case BPF_FUNC_skb_load_bytes:
6387                 return &bpf_skb_load_bytes_proto;
6388         case BPF_FUNC_skb_pull_data:
6389                 return &bpf_skb_pull_data_proto;
6390         case BPF_FUNC_csum_diff:
6391                 return &bpf_csum_diff_proto;
6392         case BPF_FUNC_get_cgroup_classid:
6393                 return &bpf_get_cgroup_classid_proto;
6394         case BPF_FUNC_get_route_realm:
6395                 return &bpf_get_route_realm_proto;
6396         case BPF_FUNC_get_hash_recalc:
6397                 return &bpf_get_hash_recalc_proto;
6398         case BPF_FUNC_perf_event_output:
6399                 return &bpf_skb_event_output_proto;
6400         case BPF_FUNC_get_smp_processor_id:
6401                 return &bpf_get_smp_processor_id_proto;
6402         case BPF_FUNC_skb_under_cgroup:
6403                 return &bpf_skb_under_cgroup_proto;
6404         default:
6405                 return bpf_base_func_proto(func_id);
6406         }
6407 }
6408
6409 static const struct bpf_func_proto *
6410 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6411 {
6412         switch (func_id) {
6413         case BPF_FUNC_lwt_push_encap:
6414                 return &bpf_lwt_in_push_encap_proto;
6415         default:
6416                 return lwt_out_func_proto(func_id, prog);
6417         }
6418 }
6419
6420 static const struct bpf_func_proto *
6421 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6422 {
6423         switch (func_id) {
6424         case BPF_FUNC_skb_get_tunnel_key:
6425                 return &bpf_skb_get_tunnel_key_proto;
6426         case BPF_FUNC_skb_set_tunnel_key:
6427                 return bpf_get_skb_set_tunnel_proto(func_id);
6428         case BPF_FUNC_skb_get_tunnel_opt:
6429                 return &bpf_skb_get_tunnel_opt_proto;
6430         case BPF_FUNC_skb_set_tunnel_opt:
6431                 return bpf_get_skb_set_tunnel_proto(func_id);
6432         case BPF_FUNC_redirect:
6433                 return &bpf_redirect_proto;
6434         case BPF_FUNC_clone_redirect:
6435                 return &bpf_clone_redirect_proto;
6436         case BPF_FUNC_skb_change_tail:
6437                 return &bpf_skb_change_tail_proto;
6438         case BPF_FUNC_skb_change_head:
6439                 return &bpf_skb_change_head_proto;
6440         case BPF_FUNC_skb_store_bytes:
6441                 return &bpf_skb_store_bytes_proto;
6442         case BPF_FUNC_csum_update:
6443                 return &bpf_csum_update_proto;
6444         case BPF_FUNC_l3_csum_replace:
6445                 return &bpf_l3_csum_replace_proto;
6446         case BPF_FUNC_l4_csum_replace:
6447                 return &bpf_l4_csum_replace_proto;
6448         case BPF_FUNC_set_hash_invalid:
6449                 return &bpf_set_hash_invalid_proto;
6450         case BPF_FUNC_lwt_push_encap:
6451                 return &bpf_lwt_xmit_push_encap_proto;
6452         default:
6453                 return lwt_out_func_proto(func_id, prog);
6454         }
6455 }
6456
6457 static const struct bpf_func_proto *
6458 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6459 {
6460         switch (func_id) {
6461 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6462         case BPF_FUNC_lwt_seg6_store_bytes:
6463                 return &bpf_lwt_seg6_store_bytes_proto;
6464         case BPF_FUNC_lwt_seg6_action:
6465                 return &bpf_lwt_seg6_action_proto;
6466         case BPF_FUNC_lwt_seg6_adjust_srh:
6467                 return &bpf_lwt_seg6_adjust_srh_proto;
6468 #endif
6469         default:
6470                 return lwt_out_func_proto(func_id, prog);
6471         }
6472 }
6473
6474 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
6475                                     const struct bpf_prog *prog,
6476                                     struct bpf_insn_access_aux *info)
6477 {
6478         const int size_default = sizeof(__u32);
6479
6480         if (off < 0 || off >= sizeof(struct __sk_buff))
6481                 return false;
6482
6483         /* The verifier guarantees that size > 0. */
6484         if (off % size != 0)
6485                 return false;
6486
6487         switch (off) {
6488         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6489                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
6490                         return false;
6491                 break;
6492         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
6493         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
6494         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
6495         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
6496         case bpf_ctx_range(struct __sk_buff, data):
6497         case bpf_ctx_range(struct __sk_buff, data_meta):
6498         case bpf_ctx_range(struct __sk_buff, data_end):
6499                 if (size != size_default)
6500                         return false;
6501                 break;
6502         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
6503                 return false;
6504         case bpf_ctx_range(struct __sk_buff, tstamp):
6505                 if (size != sizeof(__u64))
6506                         return false;
6507                 break;
6508         case offsetof(struct __sk_buff, sk):
6509                 if (type == BPF_WRITE || size != sizeof(__u64))
6510                         return false;
6511                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
6512                 break;
6513         default:
6514                 /* Only narrow read access allowed for now. */
6515                 if (type == BPF_WRITE) {
6516                         if (size != size_default)
6517                                 return false;
6518                 } else {
6519                         bpf_ctx_record_field_size(info, size_default);
6520                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6521                                 return false;
6522                 }
6523         }
6524
6525         return true;
6526 }
6527
6528 static bool sk_filter_is_valid_access(int off, int size,
6529                                       enum bpf_access_type type,
6530                                       const struct bpf_prog *prog,
6531                                       struct bpf_insn_access_aux *info)
6532 {
6533         switch (off) {
6534         case bpf_ctx_range(struct __sk_buff, tc_classid):
6535         case bpf_ctx_range(struct __sk_buff, data):
6536         case bpf_ctx_range(struct __sk_buff, data_meta):
6537         case bpf_ctx_range(struct __sk_buff, data_end):
6538         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6539         case bpf_ctx_range(struct __sk_buff, tstamp):
6540         case bpf_ctx_range(struct __sk_buff, wire_len):
6541                 return false;
6542         }
6543
6544         if (type == BPF_WRITE) {
6545                 switch (off) {
6546                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6547                         break;
6548                 default:
6549                         return false;
6550                 }
6551         }
6552
6553         return bpf_skb_is_valid_access(off, size, type, prog, info);
6554 }
6555
6556 static bool cg_skb_is_valid_access(int off, int size,
6557                                    enum bpf_access_type type,
6558                                    const struct bpf_prog *prog,
6559                                    struct bpf_insn_access_aux *info)
6560 {
6561         switch (off) {
6562         case bpf_ctx_range(struct __sk_buff, tc_classid):
6563         case bpf_ctx_range(struct __sk_buff, data_meta):
6564         case bpf_ctx_range(struct __sk_buff, wire_len):
6565                 return false;
6566         case bpf_ctx_range(struct __sk_buff, data):
6567         case bpf_ctx_range(struct __sk_buff, data_end):
6568                 if (!capable(CAP_SYS_ADMIN))
6569                         return false;
6570                 break;
6571         }
6572
6573         if (type == BPF_WRITE) {
6574                 switch (off) {
6575                 case bpf_ctx_range(struct __sk_buff, mark):
6576                 case bpf_ctx_range(struct __sk_buff, priority):
6577                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6578                         break;
6579                 case bpf_ctx_range(struct __sk_buff, tstamp):
6580                         if (!capable(CAP_SYS_ADMIN))
6581                                 return false;
6582                         break;
6583                 default:
6584                         return false;
6585                 }
6586         }
6587
6588         switch (off) {
6589         case bpf_ctx_range(struct __sk_buff, data):
6590                 info->reg_type = PTR_TO_PACKET;
6591                 break;
6592         case bpf_ctx_range(struct __sk_buff, data_end):
6593                 info->reg_type = PTR_TO_PACKET_END;
6594                 break;
6595         }
6596
6597         return bpf_skb_is_valid_access(off, size, type, prog, info);
6598 }
6599
6600 static bool lwt_is_valid_access(int off, int size,
6601                                 enum bpf_access_type type,
6602                                 const struct bpf_prog *prog,
6603                                 struct bpf_insn_access_aux *info)
6604 {
6605         switch (off) {
6606         case bpf_ctx_range(struct __sk_buff, tc_classid):
6607         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6608         case bpf_ctx_range(struct __sk_buff, data_meta):
6609         case bpf_ctx_range(struct __sk_buff, tstamp):
6610         case bpf_ctx_range(struct __sk_buff, wire_len):
6611                 return false;
6612         }
6613
6614         if (type == BPF_WRITE) {
6615                 switch (off) {
6616                 case bpf_ctx_range(struct __sk_buff, mark):
6617                 case bpf_ctx_range(struct __sk_buff, priority):
6618                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6619                         break;
6620                 default:
6621                         return false;
6622                 }
6623         }
6624
6625         switch (off) {
6626         case bpf_ctx_range(struct __sk_buff, data):
6627                 info->reg_type = PTR_TO_PACKET;
6628                 break;
6629         case bpf_ctx_range(struct __sk_buff, data_end):
6630                 info->reg_type = PTR_TO_PACKET_END;
6631                 break;
6632         }
6633
6634         return bpf_skb_is_valid_access(off, size, type, prog, info);
6635 }
6636
6637 /* Attach type specific accesses */
6638 static bool __sock_filter_check_attach_type(int off,
6639                                             enum bpf_access_type access_type,
6640                                             enum bpf_attach_type attach_type)
6641 {
6642         switch (off) {
6643         case offsetof(struct bpf_sock, bound_dev_if):
6644         case offsetof(struct bpf_sock, mark):
6645         case offsetof(struct bpf_sock, priority):
6646                 switch (attach_type) {
6647                 case BPF_CGROUP_INET_SOCK_CREATE:
6648                         goto full_access;
6649                 default:
6650                         return false;
6651                 }
6652         case bpf_ctx_range(struct bpf_sock, src_ip4):
6653                 switch (attach_type) {
6654                 case BPF_CGROUP_INET4_POST_BIND:
6655                         goto read_only;
6656                 default:
6657                         return false;
6658                 }
6659         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6660                 switch (attach_type) {
6661                 case BPF_CGROUP_INET6_POST_BIND:
6662                         goto read_only;
6663                 default:
6664                         return false;
6665                 }
6666         case bpf_ctx_range(struct bpf_sock, src_port):
6667                 switch (attach_type) {
6668                 case BPF_CGROUP_INET4_POST_BIND:
6669                 case BPF_CGROUP_INET6_POST_BIND:
6670                         goto read_only;
6671                 default:
6672                         return false;
6673                 }
6674         }
6675 read_only:
6676         return access_type == BPF_READ;
6677 full_access:
6678         return true;
6679 }
6680
6681 bool bpf_sock_common_is_valid_access(int off, int size,
6682                                      enum bpf_access_type type,
6683                                      struct bpf_insn_access_aux *info)
6684 {
6685         switch (off) {
6686         case bpf_ctx_range_till(struct bpf_sock, type, priority):
6687                 return false;
6688         default:
6689                 return bpf_sock_is_valid_access(off, size, type, info);
6690         }
6691 }
6692
6693 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6694                               struct bpf_insn_access_aux *info)
6695 {
6696         const int size_default = sizeof(__u32);
6697
6698         if (off < 0 || off >= sizeof(struct bpf_sock))
6699                 return false;
6700         if (off % size != 0)
6701                 return false;
6702
6703         switch (off) {
6704         case offsetof(struct bpf_sock, state):
6705         case offsetof(struct bpf_sock, family):
6706         case offsetof(struct bpf_sock, type):
6707         case offsetof(struct bpf_sock, protocol):
6708         case offsetof(struct bpf_sock, dst_port):
6709         case offsetof(struct bpf_sock, src_port):
6710         case bpf_ctx_range(struct bpf_sock, src_ip4):
6711         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6712         case bpf_ctx_range(struct bpf_sock, dst_ip4):
6713         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
6714                 bpf_ctx_record_field_size(info, size_default);
6715                 return bpf_ctx_narrow_access_ok(off, size, size_default);
6716         }
6717
6718         return size == size_default;
6719 }
6720
6721 static bool sock_filter_is_valid_access(int off, int size,
6722                                         enum bpf_access_type type,
6723                                         const struct bpf_prog *prog,
6724                                         struct bpf_insn_access_aux *info)
6725 {
6726         if (!bpf_sock_is_valid_access(off, size, type, info))
6727                 return false;
6728         return __sock_filter_check_attach_type(off, type,
6729                                                prog->expected_attach_type);
6730 }
6731
6732 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
6733                              const struct bpf_prog *prog)
6734 {
6735         /* Neither direct read nor direct write requires any preliminary
6736          * action.
6737          */
6738         return 0;
6739 }
6740
6741 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
6742                                 const struct bpf_prog *prog, int drop_verdict)
6743 {
6744         struct bpf_insn *insn = insn_buf;
6745
6746         if (!direct_write)
6747                 return 0;
6748
6749         /* if (!skb->cloned)
6750          *       goto start;
6751          *
6752          * (Fast-path, otherwise approximation that we might be
6753          *  a clone, do the rest in helper.)
6754          */
6755         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
6756         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
6757         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
6758
6759         /* ret = bpf_skb_pull_data(skb, 0); */
6760         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
6761         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
6762         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
6763                                BPF_FUNC_skb_pull_data);
6764         /* if (!ret)
6765          *      goto restore;
6766          * return TC_ACT_SHOT;
6767          */
6768         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
6769         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
6770         *insn++ = BPF_EXIT_INSN();
6771
6772         /* restore: */
6773         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
6774         /* start: */
6775         *insn++ = prog->insnsi[0];
6776
6777         return insn - insn_buf;
6778 }
6779
6780 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
6781                           struct bpf_insn *insn_buf)
6782 {
6783         bool indirect = BPF_MODE(orig->code) == BPF_IND;
6784         struct bpf_insn *insn = insn_buf;
6785
6786         /* We're guaranteed here that CTX is in R6. */
6787         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
6788         if (!indirect) {
6789                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
6790         } else {
6791                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
6792                 if (orig->imm)
6793                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
6794         }
6795
6796         switch (BPF_SIZE(orig->code)) {
6797         case BPF_B:
6798                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
6799                 break;
6800         case BPF_H:
6801                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
6802                 break;
6803         case BPF_W:
6804                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
6805                 break;
6806         }
6807
6808         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
6809         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
6810         *insn++ = BPF_EXIT_INSN();
6811
6812         return insn - insn_buf;
6813 }
6814
6815 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
6816                                const struct bpf_prog *prog)
6817 {
6818         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
6819 }
6820
6821 static bool tc_cls_act_is_valid_access(int off, int size,
6822                                        enum bpf_access_type type,
6823                                        const struct bpf_prog *prog,
6824                                        struct bpf_insn_access_aux *info)
6825 {
6826         if (type == BPF_WRITE) {
6827                 switch (off) {
6828                 case bpf_ctx_range(struct __sk_buff, mark):
6829                 case bpf_ctx_range(struct __sk_buff, tc_index):
6830                 case bpf_ctx_range(struct __sk_buff, priority):
6831                 case bpf_ctx_range(struct __sk_buff, tc_classid):
6832                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6833                 case bpf_ctx_range(struct __sk_buff, tstamp):
6834                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
6835                         break;
6836                 default:
6837                         return false;
6838                 }
6839         }
6840
6841         switch (off) {
6842         case bpf_ctx_range(struct __sk_buff, data):
6843                 info->reg_type = PTR_TO_PACKET;
6844                 break;
6845         case bpf_ctx_range(struct __sk_buff, data_meta):
6846                 info->reg_type = PTR_TO_PACKET_META;
6847                 break;
6848         case bpf_ctx_range(struct __sk_buff, data_end):
6849                 info->reg_type = PTR_TO_PACKET_END;
6850                 break;
6851         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6852                 return false;
6853         }
6854
6855         return bpf_skb_is_valid_access(off, size, type, prog, info);
6856 }
6857
6858 static bool __is_valid_xdp_access(int off, int size)
6859 {
6860         if (off < 0 || off >= sizeof(struct xdp_md))
6861                 return false;
6862         if (off % size != 0)
6863                 return false;
6864         if (size != sizeof(__u32))
6865                 return false;
6866
6867         return true;
6868 }
6869
6870 static bool xdp_is_valid_access(int off, int size,
6871                                 enum bpf_access_type type,
6872                                 const struct bpf_prog *prog,
6873                                 struct bpf_insn_access_aux *info)
6874 {
6875         if (type == BPF_WRITE) {
6876                 if (bpf_prog_is_dev_bound(prog->aux)) {
6877                         switch (off) {
6878                         case offsetof(struct xdp_md, rx_queue_index):
6879                                 return __is_valid_xdp_access(off, size);
6880                         }
6881                 }
6882                 return false;
6883         }
6884
6885         switch (off) {
6886         case offsetof(struct xdp_md, data):
6887                 info->reg_type = PTR_TO_PACKET;
6888                 break;
6889         case offsetof(struct xdp_md, data_meta):
6890                 info->reg_type = PTR_TO_PACKET_META;
6891                 break;
6892         case offsetof(struct xdp_md, data_end):
6893                 info->reg_type = PTR_TO_PACKET_END;
6894                 break;
6895         }
6896
6897         return __is_valid_xdp_access(off, size);
6898 }
6899
6900 void bpf_warn_invalid_xdp_action(u32 act)
6901 {
6902         const u32 act_max = XDP_REDIRECT;
6903
6904         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
6905                   act > act_max ? "Illegal" : "Driver unsupported",
6906                   act);
6907 }
6908 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
6909
6910 static bool sock_addr_is_valid_access(int off, int size,
6911                                       enum bpf_access_type type,
6912                                       const struct bpf_prog *prog,
6913                                       struct bpf_insn_access_aux *info)
6914 {
6915         const int size_default = sizeof(__u32);
6916
6917         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
6918                 return false;
6919         if (off % size != 0)
6920                 return false;
6921
6922         /* Disallow access to IPv6 fields from IPv4 contex and vise
6923          * versa.
6924          */
6925         switch (off) {
6926         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6927                 switch (prog->expected_attach_type) {
6928                 case BPF_CGROUP_INET4_BIND:
6929                 case BPF_CGROUP_INET4_CONNECT:
6930                 case BPF_CGROUP_UDP4_SENDMSG:
6931                 case BPF_CGROUP_UDP4_RECVMSG:
6932                         break;
6933                 default:
6934                         return false;
6935                 }
6936                 break;
6937         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6938                 switch (prog->expected_attach_type) {
6939                 case BPF_CGROUP_INET6_BIND:
6940                 case BPF_CGROUP_INET6_CONNECT:
6941                 case BPF_CGROUP_UDP6_SENDMSG:
6942                 case BPF_CGROUP_UDP6_RECVMSG:
6943                         break;
6944                 default:
6945                         return false;
6946                 }
6947                 break;
6948         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6949                 switch (prog->expected_attach_type) {
6950                 case BPF_CGROUP_UDP4_SENDMSG:
6951                         break;
6952                 default:
6953                         return false;
6954                 }
6955                 break;
6956         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6957                                 msg_src_ip6[3]):
6958                 switch (prog->expected_attach_type) {
6959                 case BPF_CGROUP_UDP6_SENDMSG:
6960                         break;
6961                 default:
6962                         return false;
6963                 }
6964                 break;
6965         }
6966
6967         switch (off) {
6968         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
6969         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6970         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
6971         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6972                                 msg_src_ip6[3]):
6973                 if (type == BPF_READ) {
6974                         bpf_ctx_record_field_size(info, size_default);
6975
6976                         if (bpf_ctx_wide_access_ok(off, size,
6977                                                    struct bpf_sock_addr,
6978                                                    user_ip6))
6979                                 return true;
6980
6981                         if (bpf_ctx_wide_access_ok(off, size,
6982                                                    struct bpf_sock_addr,
6983                                                    msg_src_ip6))
6984                                 return true;
6985
6986                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
6987                                 return false;
6988                 } else {
6989                         if (bpf_ctx_wide_access_ok(off, size,
6990                                                    struct bpf_sock_addr,
6991                                                    user_ip6))
6992                                 return true;
6993
6994                         if (bpf_ctx_wide_access_ok(off, size,
6995                                                    struct bpf_sock_addr,
6996                                                    msg_src_ip6))
6997                                 return true;
6998
6999                         if (size != size_default)
7000                                 return false;
7001                 }
7002                 break;
7003         case bpf_ctx_range(struct bpf_sock_addr, user_port):
7004                 if (size != size_default)
7005                         return false;
7006                 break;
7007         case offsetof(struct bpf_sock_addr, sk):
7008                 if (type != BPF_READ)
7009                         return false;
7010                 if (size != sizeof(__u64))
7011                         return false;
7012                 info->reg_type = PTR_TO_SOCKET;
7013                 break;
7014         default:
7015                 if (type == BPF_READ) {
7016                         if (size != size_default)
7017                                 return false;
7018                 } else {
7019                         return false;
7020                 }
7021         }
7022
7023         return true;
7024 }
7025
7026 static bool sock_ops_is_valid_access(int off, int size,
7027                                      enum bpf_access_type type,
7028                                      const struct bpf_prog *prog,
7029                                      struct bpf_insn_access_aux *info)
7030 {
7031         const int size_default = sizeof(__u32);
7032
7033         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
7034                 return false;
7035
7036         /* The verifier guarantees that size > 0. */
7037         if (off % size != 0)
7038                 return false;
7039
7040         if (type == BPF_WRITE) {
7041                 switch (off) {
7042                 case offsetof(struct bpf_sock_ops, reply):
7043                 case offsetof(struct bpf_sock_ops, sk_txhash):
7044                         if (size != size_default)
7045                                 return false;
7046                         break;
7047                 default:
7048                         return false;
7049                 }
7050         } else {
7051                 switch (off) {
7052                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
7053                                         bytes_acked):
7054                         if (size != sizeof(__u64))
7055                                 return false;
7056                         break;
7057                 case offsetof(struct bpf_sock_ops, sk):
7058                         if (size != sizeof(__u64))
7059                                 return false;
7060                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
7061                         break;
7062                 default:
7063                         if (size != size_default)
7064                                 return false;
7065                         break;
7066                 }
7067         }
7068
7069         return true;
7070 }
7071
7072 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
7073                            const struct bpf_prog *prog)
7074 {
7075         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
7076 }
7077
7078 static bool sk_skb_is_valid_access(int off, int size,
7079                                    enum bpf_access_type type,
7080                                    const struct bpf_prog *prog,
7081                                    struct bpf_insn_access_aux *info)
7082 {
7083         switch (off) {
7084         case bpf_ctx_range(struct __sk_buff, tc_classid):
7085         case bpf_ctx_range(struct __sk_buff, data_meta):
7086         case bpf_ctx_range(struct __sk_buff, tstamp):
7087         case bpf_ctx_range(struct __sk_buff, wire_len):
7088                 return false;
7089         }
7090
7091         if (type == BPF_WRITE) {
7092                 switch (off) {
7093                 case bpf_ctx_range(struct __sk_buff, tc_index):
7094                 case bpf_ctx_range(struct __sk_buff, priority):
7095                         break;
7096                 default:
7097                         return false;
7098                 }
7099         }
7100
7101         switch (off) {
7102         case bpf_ctx_range(struct __sk_buff, mark):
7103                 return false;
7104         case bpf_ctx_range(struct __sk_buff, data):
7105                 info->reg_type = PTR_TO_PACKET;
7106                 break;
7107         case bpf_ctx_range(struct __sk_buff, data_end):
7108                 info->reg_type = PTR_TO_PACKET_END;
7109                 break;
7110         }
7111
7112         return bpf_skb_is_valid_access(off, size, type, prog, info);
7113 }
7114
7115 static bool sk_msg_is_valid_access(int off, int size,
7116                                    enum bpf_access_type type,
7117                                    const struct bpf_prog *prog,
7118                                    struct bpf_insn_access_aux *info)
7119 {
7120         if (type == BPF_WRITE)
7121                 return false;
7122
7123         if (off % size != 0)
7124                 return false;
7125
7126         switch (off) {
7127         case offsetof(struct sk_msg_md, data):
7128                 info->reg_type = PTR_TO_PACKET;
7129                 if (size != sizeof(__u64))
7130                         return false;
7131                 break;
7132         case offsetof(struct sk_msg_md, data_end):
7133                 info->reg_type = PTR_TO_PACKET_END;
7134                 if (size != sizeof(__u64))
7135                         return false;
7136                 break;
7137         case bpf_ctx_range(struct sk_msg_md, family):
7138         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
7139         case bpf_ctx_range(struct sk_msg_md, local_ip4):
7140         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
7141         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
7142         case bpf_ctx_range(struct sk_msg_md, remote_port):
7143         case bpf_ctx_range(struct sk_msg_md, local_port):
7144         case bpf_ctx_range(struct sk_msg_md, size):
7145                 if (size != sizeof(__u32))
7146                         return false;
7147                 break;
7148         default:
7149                 return false;
7150         }
7151         return true;
7152 }
7153
7154 static bool flow_dissector_is_valid_access(int off, int size,
7155                                            enum bpf_access_type type,
7156                                            const struct bpf_prog *prog,
7157                                            struct bpf_insn_access_aux *info)
7158 {
7159         const int size_default = sizeof(__u32);
7160
7161         if (off < 0 || off >= sizeof(struct __sk_buff))
7162                 return false;
7163
7164         if (type == BPF_WRITE)
7165                 return false;
7166
7167         switch (off) {
7168         case bpf_ctx_range(struct __sk_buff, data):
7169                 if (size != size_default)
7170                         return false;
7171                 info->reg_type = PTR_TO_PACKET;
7172                 return true;
7173         case bpf_ctx_range(struct __sk_buff, data_end):
7174                 if (size != size_default)
7175                         return false;
7176                 info->reg_type = PTR_TO_PACKET_END;
7177                 return true;
7178         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7179                 if (size != sizeof(__u64))
7180                         return false;
7181                 info->reg_type = PTR_TO_FLOW_KEYS;
7182                 return true;
7183         default:
7184                 return false;
7185         }
7186 }
7187
7188 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
7189                                              const struct bpf_insn *si,
7190                                              struct bpf_insn *insn_buf,
7191                                              struct bpf_prog *prog,
7192                                              u32 *target_size)
7193
7194 {
7195         struct bpf_insn *insn = insn_buf;
7196
7197         switch (si->off) {
7198         case offsetof(struct __sk_buff, data):
7199                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
7200                                       si->dst_reg, si->src_reg,
7201                                       offsetof(struct bpf_flow_dissector, data));
7202                 break;
7203
7204         case offsetof(struct __sk_buff, data_end):
7205                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
7206                                       si->dst_reg, si->src_reg,
7207                                       offsetof(struct bpf_flow_dissector, data_end));
7208                 break;
7209
7210         case offsetof(struct __sk_buff, flow_keys):
7211                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
7212                                       si->dst_reg, si->src_reg,
7213                                       offsetof(struct bpf_flow_dissector, flow_keys));
7214                 break;
7215         }
7216
7217         return insn - insn_buf;
7218 }
7219
7220 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
7221                                   const struct bpf_insn *si,
7222                                   struct bpf_insn *insn_buf,
7223                                   struct bpf_prog *prog, u32 *target_size)
7224 {
7225         struct bpf_insn *insn = insn_buf;
7226         int off;
7227
7228         switch (si->off) {
7229         case offsetof(struct __sk_buff, len):
7230                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7231                                       bpf_target_off(struct sk_buff, len, 4,
7232                                                      target_size));
7233                 break;
7234
7235         case offsetof(struct __sk_buff, protocol):
7236                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7237                                       bpf_target_off(struct sk_buff, protocol, 2,
7238                                                      target_size));
7239                 break;
7240
7241         case offsetof(struct __sk_buff, vlan_proto):
7242                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7243                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
7244                                                      target_size));
7245                 break;
7246
7247         case offsetof(struct __sk_buff, priority):
7248                 if (type == BPF_WRITE)
7249                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7250                                               bpf_target_off(struct sk_buff, priority, 4,
7251                                                              target_size));
7252                 else
7253                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7254                                               bpf_target_off(struct sk_buff, priority, 4,
7255                                                              target_size));
7256                 break;
7257
7258         case offsetof(struct __sk_buff, ingress_ifindex):
7259                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7260                                       bpf_target_off(struct sk_buff, skb_iif, 4,
7261                                                      target_size));
7262                 break;
7263
7264         case offsetof(struct __sk_buff, ifindex):
7265                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7266                                       si->dst_reg, si->src_reg,
7267                                       offsetof(struct sk_buff, dev));
7268                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
7269                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7270                                       bpf_target_off(struct net_device, ifindex, 4,
7271                                                      target_size));
7272                 break;
7273
7274         case offsetof(struct __sk_buff, hash):
7275                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7276                                       bpf_target_off(struct sk_buff, hash, 4,
7277                                                      target_size));
7278                 break;
7279
7280         case offsetof(struct __sk_buff, mark):
7281                 if (type == BPF_WRITE)
7282                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7283                                               bpf_target_off(struct sk_buff, mark, 4,
7284                                                              target_size));
7285                 else
7286                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7287                                               bpf_target_off(struct sk_buff, mark, 4,
7288                                                              target_size));
7289                 break;
7290
7291         case offsetof(struct __sk_buff, pkt_type):
7292                 *target_size = 1;
7293                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7294                                       PKT_TYPE_OFFSET());
7295                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
7296 #ifdef __BIG_ENDIAN_BITFIELD
7297                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
7298 #endif
7299                 break;
7300
7301         case offsetof(struct __sk_buff, queue_mapping):
7302                 if (type == BPF_WRITE) {
7303                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
7304                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7305                                               bpf_target_off(struct sk_buff,
7306                                                              queue_mapping,
7307                                                              2, target_size));
7308                 } else {
7309                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7310                                               bpf_target_off(struct sk_buff,
7311                                                              queue_mapping,
7312                                                              2, target_size));
7313                 }
7314                 break;
7315
7316         case offsetof(struct __sk_buff, vlan_present):
7317                 *target_size = 1;
7318                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
7319                                       PKT_VLAN_PRESENT_OFFSET());
7320                 if (PKT_VLAN_PRESENT_BIT)
7321                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
7322                 if (PKT_VLAN_PRESENT_BIT < 7)
7323                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
7324                 break;
7325
7326         case offsetof(struct __sk_buff, vlan_tci):
7327                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7328                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
7329                                                      target_size));
7330                 break;
7331
7332         case offsetof(struct __sk_buff, cb[0]) ...
7333              offsetofend(struct __sk_buff, cb[4]) - 1:
7334                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
7335                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
7336                               offsetof(struct qdisc_skb_cb, data)) %
7337                              sizeof(__u64));
7338
7339                 prog->cb_access = 1;
7340                 off  = si->off;
7341                 off -= offsetof(struct __sk_buff, cb[0]);
7342                 off += offsetof(struct sk_buff, cb);
7343                 off += offsetof(struct qdisc_skb_cb, data);
7344                 if (type == BPF_WRITE)
7345                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
7346                                               si->src_reg, off);
7347                 else
7348                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
7349                                               si->src_reg, off);
7350                 break;
7351
7352         case offsetof(struct __sk_buff, tc_classid):
7353                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
7354
7355                 off  = si->off;
7356                 off -= offsetof(struct __sk_buff, tc_classid);
7357                 off += offsetof(struct sk_buff, cb);
7358                 off += offsetof(struct qdisc_skb_cb, tc_classid);
7359                 *target_size = 2;
7360                 if (type == BPF_WRITE)
7361                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
7362                                               si->src_reg, off);
7363                 else
7364                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
7365                                               si->src_reg, off);
7366                 break;
7367
7368         case offsetof(struct __sk_buff, data):
7369                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
7370                                       si->dst_reg, si->src_reg,
7371                                       offsetof(struct sk_buff, data));
7372                 break;
7373
7374         case offsetof(struct __sk_buff, data_meta):
7375                 off  = si->off;
7376                 off -= offsetof(struct __sk_buff, data_meta);
7377                 off += offsetof(struct sk_buff, cb);
7378                 off += offsetof(struct bpf_skb_data_end, data_meta);
7379                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7380                                       si->src_reg, off);
7381                 break;
7382
7383         case offsetof(struct __sk_buff, data_end):
7384                 off  = si->off;
7385                 off -= offsetof(struct __sk_buff, data_end);
7386                 off += offsetof(struct sk_buff, cb);
7387                 off += offsetof(struct bpf_skb_data_end, data_end);
7388                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7389                                       si->src_reg, off);
7390                 break;
7391
7392         case offsetof(struct __sk_buff, tc_index):
7393 #ifdef CONFIG_NET_SCHED
7394                 if (type == BPF_WRITE)
7395                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
7396                                               bpf_target_off(struct sk_buff, tc_index, 2,
7397                                                              target_size));
7398                 else
7399                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
7400                                               bpf_target_off(struct sk_buff, tc_index, 2,
7401                                                              target_size));
7402 #else
7403                 *target_size = 2;
7404                 if (type == BPF_WRITE)
7405                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
7406                 else
7407                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7408 #endif
7409                 break;
7410
7411         case offsetof(struct __sk_buff, napi_id):
7412 #if defined(CONFIG_NET_RX_BUSY_POLL)
7413                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7414                                       bpf_target_off(struct sk_buff, napi_id, 4,
7415                                                      target_size));
7416                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
7417                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7418 #else
7419                 *target_size = 4;
7420                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
7421 #endif
7422                 break;
7423         case offsetof(struct __sk_buff, family):
7424                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7425
7426                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7427                                       si->dst_reg, si->src_reg,
7428                                       offsetof(struct sk_buff, sk));
7429                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7430                                       bpf_target_off(struct sock_common,
7431                                                      skc_family,
7432                                                      2, target_size));
7433                 break;
7434         case offsetof(struct __sk_buff, remote_ip4):
7435                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7436
7437                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7438                                       si->dst_reg, si->src_reg,
7439                                       offsetof(struct sk_buff, sk));
7440                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7441                                       bpf_target_off(struct sock_common,
7442                                                      skc_daddr,
7443                                                      4, target_size));
7444                 break;
7445         case offsetof(struct __sk_buff, local_ip4):
7446                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7447                                           skc_rcv_saddr) != 4);
7448
7449                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7450                                       si->dst_reg, si->src_reg,
7451                                       offsetof(struct sk_buff, sk));
7452                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7453                                       bpf_target_off(struct sock_common,
7454                                                      skc_rcv_saddr,
7455                                                      4, target_size));
7456                 break;
7457         case offsetof(struct __sk_buff, remote_ip6[0]) ...
7458              offsetof(struct __sk_buff, remote_ip6[3]):
7459 #if IS_ENABLED(CONFIG_IPV6)
7460                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7461                                           skc_v6_daddr.s6_addr32[0]) != 4);
7462
7463                 off = si->off;
7464                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
7465
7466                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7467                                       si->dst_reg, si->src_reg,
7468                                       offsetof(struct sk_buff, sk));
7469                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7470                                       offsetof(struct sock_common,
7471                                                skc_v6_daddr.s6_addr32[0]) +
7472                                       off);
7473 #else
7474                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7475 #endif
7476                 break;
7477         case offsetof(struct __sk_buff, local_ip6[0]) ...
7478              offsetof(struct __sk_buff, local_ip6[3]):
7479 #if IS_ENABLED(CONFIG_IPV6)
7480                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7481                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7482
7483                 off = si->off;
7484                 off -= offsetof(struct __sk_buff, local_ip6[0]);
7485
7486                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7487                                       si->dst_reg, si->src_reg,
7488                                       offsetof(struct sk_buff, sk));
7489                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7490                                       offsetof(struct sock_common,
7491                                                skc_v6_rcv_saddr.s6_addr32[0]) +
7492                                       off);
7493 #else
7494                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7495 #endif
7496                 break;
7497
7498         case offsetof(struct __sk_buff, remote_port):
7499                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7500
7501                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7502                                       si->dst_reg, si->src_reg,
7503                                       offsetof(struct sk_buff, sk));
7504                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7505                                       bpf_target_off(struct sock_common,
7506                                                      skc_dport,
7507                                                      2, target_size));
7508 #ifndef __BIG_ENDIAN_BITFIELD
7509                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7510 #endif
7511                 break;
7512
7513         case offsetof(struct __sk_buff, local_port):
7514                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7515
7516                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7517                                       si->dst_reg, si->src_reg,
7518                                       offsetof(struct sk_buff, sk));
7519                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7520                                       bpf_target_off(struct sock_common,
7521                                                      skc_num, 2, target_size));
7522                 break;
7523
7524         case offsetof(struct __sk_buff, tstamp):
7525                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8);
7526
7527                 if (type == BPF_WRITE)
7528                         *insn++ = BPF_STX_MEM(BPF_DW,
7529                                               si->dst_reg, si->src_reg,
7530                                               bpf_target_off(struct sk_buff,
7531                                                              tstamp, 8,
7532                                                              target_size));
7533                 else
7534                         *insn++ = BPF_LDX_MEM(BPF_DW,
7535                                               si->dst_reg, si->src_reg,
7536                                               bpf_target_off(struct sk_buff,
7537                                                              tstamp, 8,
7538                                                              target_size));
7539                 break;
7540
7541         case offsetof(struct __sk_buff, gso_segs):
7542                 /* si->dst_reg = skb_shinfo(SKB); */
7543 #ifdef NET_SKBUFF_DATA_USES_OFFSET
7544                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7545                                       BPF_REG_AX, si->src_reg,
7546                                       offsetof(struct sk_buff, end));
7547                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
7548                                       si->dst_reg, si->src_reg,
7549                                       offsetof(struct sk_buff, head));
7550                 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
7551 #else
7552                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
7553                                       si->dst_reg, si->src_reg,
7554                                       offsetof(struct sk_buff, end));
7555 #endif
7556                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
7557                                       si->dst_reg, si->dst_reg,
7558                                       bpf_target_off(struct skb_shared_info,
7559                                                      gso_segs, 2,
7560                                                      target_size));
7561                 break;
7562         case offsetof(struct __sk_buff, wire_len):
7563                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4);
7564
7565                 off = si->off;
7566                 off -= offsetof(struct __sk_buff, wire_len);
7567                 off += offsetof(struct sk_buff, cb);
7568                 off += offsetof(struct qdisc_skb_cb, pkt_len);
7569                 *target_size = 4;
7570                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
7571                 break;
7572
7573         case offsetof(struct __sk_buff, sk):
7574                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
7575                                       si->dst_reg, si->src_reg,
7576                                       offsetof(struct sk_buff, sk));
7577                 break;
7578         }
7579
7580         return insn - insn_buf;
7581 }
7582
7583 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
7584                                 const struct bpf_insn *si,
7585                                 struct bpf_insn *insn_buf,
7586                                 struct bpf_prog *prog, u32 *target_size)
7587 {
7588         struct bpf_insn *insn = insn_buf;
7589         int off;
7590
7591         switch (si->off) {
7592         case offsetof(struct bpf_sock, bound_dev_if):
7593                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
7594
7595                 if (type == BPF_WRITE)
7596                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7597                                         offsetof(struct sock, sk_bound_dev_if));
7598                 else
7599                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7600                                       offsetof(struct sock, sk_bound_dev_if));
7601                 break;
7602
7603         case offsetof(struct bpf_sock, mark):
7604                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
7605
7606                 if (type == BPF_WRITE)
7607                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7608                                         offsetof(struct sock, sk_mark));
7609                 else
7610                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7611                                       offsetof(struct sock, sk_mark));
7612                 break;
7613
7614         case offsetof(struct bpf_sock, priority):
7615                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
7616
7617                 if (type == BPF_WRITE)
7618                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
7619                                         offsetof(struct sock, sk_priority));
7620                 else
7621                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7622                                       offsetof(struct sock, sk_priority));
7623                 break;
7624
7625         case offsetof(struct bpf_sock, family):
7626                 *insn++ = BPF_LDX_MEM(
7627                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
7628                         si->dst_reg, si->src_reg,
7629                         bpf_target_off(struct sock_common,
7630                                        skc_family,
7631                                        FIELD_SIZEOF(struct sock_common,
7632                                                     skc_family),
7633                                        target_size));
7634                 break;
7635
7636         case offsetof(struct bpf_sock, type):
7637                 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2);
7638                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7639                                       offsetof(struct sock, __sk_flags_offset));
7640                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7641                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7642                 *target_size = 2;
7643                 break;
7644
7645         case offsetof(struct bpf_sock, protocol):
7646                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7647                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7648                                       offsetof(struct sock, __sk_flags_offset));
7649                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7650                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
7651                 *target_size = 1;
7652                 break;
7653
7654         case offsetof(struct bpf_sock, src_ip4):
7655                 *insn++ = BPF_LDX_MEM(
7656                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7657                         bpf_target_off(struct sock_common, skc_rcv_saddr,
7658                                        FIELD_SIZEOF(struct sock_common,
7659                                                     skc_rcv_saddr),
7660                                        target_size));
7661                 break;
7662
7663         case offsetof(struct bpf_sock, dst_ip4):
7664                 *insn++ = BPF_LDX_MEM(
7665                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7666                         bpf_target_off(struct sock_common, skc_daddr,
7667                                        FIELD_SIZEOF(struct sock_common,
7668                                                     skc_daddr),
7669                                        target_size));
7670                 break;
7671
7672         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7673 #if IS_ENABLED(CONFIG_IPV6)
7674                 off = si->off;
7675                 off -= offsetof(struct bpf_sock, src_ip6[0]);
7676                 *insn++ = BPF_LDX_MEM(
7677                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7678                         bpf_target_off(
7679                                 struct sock_common,
7680                                 skc_v6_rcv_saddr.s6_addr32[0],
7681                                 FIELD_SIZEOF(struct sock_common,
7682                                              skc_v6_rcv_saddr.s6_addr32[0]),
7683                                 target_size) + off);
7684 #else
7685                 (void)off;
7686                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7687 #endif
7688                 break;
7689
7690         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7691 #if IS_ENABLED(CONFIG_IPV6)
7692                 off = si->off;
7693                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
7694                 *insn++ = BPF_LDX_MEM(
7695                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
7696                         bpf_target_off(struct sock_common,
7697                                        skc_v6_daddr.s6_addr32[0],
7698                                        FIELD_SIZEOF(struct sock_common,
7699                                                     skc_v6_daddr.s6_addr32[0]),
7700                                        target_size) + off);
7701 #else
7702                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7703                 *target_size = 4;
7704 #endif
7705                 break;
7706
7707         case offsetof(struct bpf_sock, src_port):
7708                 *insn++ = BPF_LDX_MEM(
7709                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
7710                         si->dst_reg, si->src_reg,
7711                         bpf_target_off(struct sock_common, skc_num,
7712                                        FIELD_SIZEOF(struct sock_common,
7713                                                     skc_num),
7714                                        target_size));
7715                 break;
7716
7717         case offsetof(struct bpf_sock, dst_port):
7718                 *insn++ = BPF_LDX_MEM(
7719                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
7720                         si->dst_reg, si->src_reg,
7721                         bpf_target_off(struct sock_common, skc_dport,
7722                                        FIELD_SIZEOF(struct sock_common,
7723                                                     skc_dport),
7724                                        target_size));
7725                 break;
7726
7727         case offsetof(struct bpf_sock, state):
7728                 *insn++ = BPF_LDX_MEM(
7729                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
7730                         si->dst_reg, si->src_reg,
7731                         bpf_target_off(struct sock_common, skc_state,
7732                                        FIELD_SIZEOF(struct sock_common,
7733                                                     skc_state),
7734                                        target_size));
7735                 break;
7736         }
7737
7738         return insn - insn_buf;
7739 }
7740
7741 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
7742                                          const struct bpf_insn *si,
7743                                          struct bpf_insn *insn_buf,
7744                                          struct bpf_prog *prog, u32 *target_size)
7745 {
7746         struct bpf_insn *insn = insn_buf;
7747
7748         switch (si->off) {
7749         case offsetof(struct __sk_buff, ifindex):
7750                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
7751                                       si->dst_reg, si->src_reg,
7752                                       offsetof(struct sk_buff, dev));
7753                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7754                                       bpf_target_off(struct net_device, ifindex, 4,
7755                                                      target_size));
7756                 break;
7757         default:
7758                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7759                                               target_size);
7760         }
7761
7762         return insn - insn_buf;
7763 }
7764
7765 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
7766                                   const struct bpf_insn *si,
7767                                   struct bpf_insn *insn_buf,
7768                                   struct bpf_prog *prog, u32 *target_size)
7769 {
7770         struct bpf_insn *insn = insn_buf;
7771
7772         switch (si->off) {
7773         case offsetof(struct xdp_md, data):
7774                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
7775                                       si->dst_reg, si->src_reg,
7776                                       offsetof(struct xdp_buff, data));
7777                 break;
7778         case offsetof(struct xdp_md, data_meta):
7779                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
7780                                       si->dst_reg, si->src_reg,
7781                                       offsetof(struct xdp_buff, data_meta));
7782                 break;
7783         case offsetof(struct xdp_md, data_end):
7784                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
7785                                       si->dst_reg, si->src_reg,
7786                                       offsetof(struct xdp_buff, data_end));
7787                 break;
7788         case offsetof(struct xdp_md, ingress_ifindex):
7789                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7790                                       si->dst_reg, si->src_reg,
7791                                       offsetof(struct xdp_buff, rxq));
7792                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
7793                                       si->dst_reg, si->dst_reg,
7794                                       offsetof(struct xdp_rxq_info, dev));
7795                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7796                                       offsetof(struct net_device, ifindex));
7797                 break;
7798         case offsetof(struct xdp_md, rx_queue_index):
7799                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
7800                                       si->dst_reg, si->src_reg,
7801                                       offsetof(struct xdp_buff, rxq));
7802                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7803                                       offsetof(struct xdp_rxq_info,
7804                                                queue_index));
7805                 break;
7806         }
7807
7808         return insn - insn_buf;
7809 }
7810
7811 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
7812  * context Structure, F is Field in context structure that contains a pointer
7813  * to Nested Structure of type NS that has the field NF.
7814  *
7815  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
7816  * sure that SIZE is not greater than actual size of S.F.NF.
7817  *
7818  * If offset OFF is provided, the load happens from that offset relative to
7819  * offset of NF.
7820  */
7821 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
7822         do {                                                                   \
7823                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
7824                                       si->src_reg, offsetof(S, F));            \
7825                 *insn++ = BPF_LDX_MEM(                                         \
7826                         SIZE, si->dst_reg, si->dst_reg,                        \
7827                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
7828                                        target_size)                            \
7829                                 + OFF);                                        \
7830         } while (0)
7831
7832 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
7833         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
7834                                              BPF_FIELD_SIZEOF(NS, NF), 0)
7835
7836 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
7837  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
7838  *
7839  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
7840  * "register" since two registers available in convert_ctx_access are not
7841  * enough: we can't override neither SRC, since it contains value to store, nor
7842  * DST since it contains pointer to context that may be used by later
7843  * instructions. But we need a temporary place to save pointer to nested
7844  * structure whose field we want to store to.
7845  */
7846 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
7847         do {                                                                   \
7848                 int tmp_reg = BPF_REG_9;                                       \
7849                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
7850                         --tmp_reg;                                             \
7851                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
7852                         --tmp_reg;                                             \
7853                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
7854                                       offsetof(S, TF));                        \
7855                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
7856                                       si->dst_reg, offsetof(S, F));            \
7857                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
7858                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
7859                                        target_size)                            \
7860                                 + OFF);                                        \
7861                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
7862                                       offsetof(S, TF));                        \
7863         } while (0)
7864
7865 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
7866                                                       TF)                      \
7867         do {                                                                   \
7868                 if (type == BPF_WRITE) {                                       \
7869                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
7870                                                          OFF, TF);             \
7871                 } else {                                                       \
7872                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
7873                                 S, NS, F, NF, SIZE, OFF);  \
7874                 }                                                              \
7875         } while (0)
7876
7877 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
7878         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
7879                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
7880
7881 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
7882                                         const struct bpf_insn *si,
7883                                         struct bpf_insn *insn_buf,
7884                                         struct bpf_prog *prog, u32 *target_size)
7885 {
7886         struct bpf_insn *insn = insn_buf;
7887         int off;
7888
7889         switch (si->off) {
7890         case offsetof(struct bpf_sock_addr, user_family):
7891                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7892                                             struct sockaddr, uaddr, sa_family);
7893                 break;
7894
7895         case offsetof(struct bpf_sock_addr, user_ip4):
7896                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7897                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
7898                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
7899                 break;
7900
7901         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
7902                 off = si->off;
7903                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
7904                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7905                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
7906                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
7907                         tmp_reg);
7908                 break;
7909
7910         case offsetof(struct bpf_sock_addr, user_port):
7911                 /* To get port we need to know sa_family first and then treat
7912                  * sockaddr as either sockaddr_in or sockaddr_in6.
7913                  * Though we can simplify since port field has same offset and
7914                  * size in both structures.
7915                  * Here we check this invariant and use just one of the
7916                  * structures if it's true.
7917                  */
7918                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
7919                              offsetof(struct sockaddr_in6, sin6_port));
7920                 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
7921                              FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
7922                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
7923                                                      struct sockaddr_in6, uaddr,
7924                                                      sin6_port, tmp_reg);
7925                 break;
7926
7927         case offsetof(struct bpf_sock_addr, family):
7928                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
7929                                             struct sock, sk, sk_family);
7930                 break;
7931
7932         case offsetof(struct bpf_sock_addr, type):
7933                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7934                         struct bpf_sock_addr_kern, struct sock, sk,
7935                         __sk_flags_offset, BPF_W, 0);
7936                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
7937                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
7938                 break;
7939
7940         case offsetof(struct bpf_sock_addr, protocol):
7941                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
7942                         struct bpf_sock_addr_kern, struct sock, sk,
7943                         __sk_flags_offset, BPF_W, 0);
7944                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7945                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7946                                         SK_FL_PROTO_SHIFT);
7947                 break;
7948
7949         case offsetof(struct bpf_sock_addr, msg_src_ip4):
7950                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
7951                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7952                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
7953                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
7954                 break;
7955
7956         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
7957                                 msg_src_ip6[3]):
7958                 off = si->off;
7959                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
7960                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
7961                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
7962                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
7963                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
7964                 break;
7965         case offsetof(struct bpf_sock_addr, sk):
7966                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
7967                                       si->dst_reg, si->src_reg,
7968                                       offsetof(struct bpf_sock_addr_kern, sk));
7969                 break;
7970         }
7971
7972         return insn - insn_buf;
7973 }
7974
7975 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
7976                                        const struct bpf_insn *si,
7977                                        struct bpf_insn *insn_buf,
7978                                        struct bpf_prog *prog,
7979                                        u32 *target_size)
7980 {
7981         struct bpf_insn *insn = insn_buf;
7982         int off;
7983
7984 /* Helper macro for adding read access to tcp_sock or sock fields. */
7985 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
7986         do {                                                                  \
7987                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
7988                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
7989                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7990                                                 struct bpf_sock_ops_kern,     \
7991                                                 is_fullsock),                 \
7992                                       si->dst_reg, si->src_reg,               \
7993                                       offsetof(struct bpf_sock_ops_kern,      \
7994                                                is_fullsock));                 \
7995                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
7996                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
7997                                                 struct bpf_sock_ops_kern, sk),\
7998                                       si->dst_reg, si->src_reg,               \
7999                                       offsetof(struct bpf_sock_ops_kern, sk));\
8000                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
8001                                                        OBJ_FIELD),            \
8002                                       si->dst_reg, si->dst_reg,               \
8003                                       offsetof(OBJ, OBJ_FIELD));              \
8004         } while (0)
8005
8006 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
8007                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
8008
8009 /* Helper macro for adding write access to tcp_sock or sock fields.
8010  * The macro is called with two registers, dst_reg which contains a pointer
8011  * to ctx (context) and src_reg which contains the value that should be
8012  * stored. However, we need an additional register since we cannot overwrite
8013  * dst_reg because it may be used later in the program.
8014  * Instead we "borrow" one of the other register. We first save its value
8015  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
8016  * it at the end of the macro.
8017  */
8018 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
8019         do {                                                                  \
8020                 int reg = BPF_REG_9;                                          \
8021                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
8022                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
8023                 if (si->dst_reg == reg || si->src_reg == reg)                 \
8024                         reg--;                                                \
8025                 if (si->dst_reg == reg || si->src_reg == reg)                 \
8026                         reg--;                                                \
8027                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
8028                                       offsetof(struct bpf_sock_ops_kern,      \
8029                                                temp));                        \
8030                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
8031                                                 struct bpf_sock_ops_kern,     \
8032                                                 is_fullsock),                 \
8033                                       reg, si->dst_reg,                       \
8034                                       offsetof(struct bpf_sock_ops_kern,      \
8035                                                is_fullsock));                 \
8036                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
8037                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
8038                                                 struct bpf_sock_ops_kern, sk),\
8039                                       reg, si->dst_reg,                       \
8040                                       offsetof(struct bpf_sock_ops_kern, sk));\
8041                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
8042                                       reg, si->src_reg,                       \
8043                                       offsetof(OBJ, OBJ_FIELD));              \
8044                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
8045                                       offsetof(struct bpf_sock_ops_kern,      \
8046                                                temp));                        \
8047         } while (0)
8048
8049 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
8050         do {                                                                  \
8051                 if (TYPE == BPF_WRITE)                                        \
8052                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
8053                 else                                                          \
8054                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
8055         } while (0)
8056
8057         if (insn > insn_buf)
8058                 return insn - insn_buf;
8059
8060         switch (si->off) {
8061         case offsetof(struct bpf_sock_ops, op) ...
8062              offsetof(struct bpf_sock_ops, replylong[3]):
8063                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
8064                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
8065                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
8066                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
8067                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
8068                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
8069                 off = si->off;
8070                 off -= offsetof(struct bpf_sock_ops, op);
8071                 off += offsetof(struct bpf_sock_ops_kern, op);
8072                 if (type == BPF_WRITE)
8073                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8074                                               off);
8075                 else
8076                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8077                                               off);
8078                 break;
8079
8080         case offsetof(struct bpf_sock_ops, family):
8081                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
8082
8083                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8084                                               struct bpf_sock_ops_kern, sk),
8085                                       si->dst_reg, si->src_reg,
8086                                       offsetof(struct bpf_sock_ops_kern, sk));
8087                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8088                                       offsetof(struct sock_common, skc_family));
8089                 break;
8090
8091         case offsetof(struct bpf_sock_ops, remote_ip4):
8092                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
8093
8094                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8095                                                 struct bpf_sock_ops_kern, sk),
8096                                       si->dst_reg, si->src_reg,
8097                                       offsetof(struct bpf_sock_ops_kern, sk));
8098                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8099                                       offsetof(struct sock_common, skc_daddr));
8100                 break;
8101
8102         case offsetof(struct bpf_sock_ops, local_ip4):
8103                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8104                                           skc_rcv_saddr) != 4);
8105
8106                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8107                                               struct bpf_sock_ops_kern, sk),
8108                                       si->dst_reg, si->src_reg,
8109                                       offsetof(struct bpf_sock_ops_kern, sk));
8110                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8111                                       offsetof(struct sock_common,
8112                                                skc_rcv_saddr));
8113                 break;
8114
8115         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
8116              offsetof(struct bpf_sock_ops, remote_ip6[3]):
8117 #if IS_ENABLED(CONFIG_IPV6)
8118                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8119                                           skc_v6_daddr.s6_addr32[0]) != 4);
8120
8121                 off = si->off;
8122                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
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 sock_common,
8129                                                skc_v6_daddr.s6_addr32[0]) +
8130                                       off);
8131 #else
8132                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8133 #endif
8134                 break;
8135
8136         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
8137              offsetof(struct bpf_sock_ops, local_ip6[3]):
8138 #if IS_ENABLED(CONFIG_IPV6)
8139                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8140                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8141
8142                 off = si->off;
8143                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
8144                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8145                                                 struct bpf_sock_ops_kern, sk),
8146                                       si->dst_reg, si->src_reg,
8147                                       offsetof(struct bpf_sock_ops_kern, sk));
8148                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8149                                       offsetof(struct sock_common,
8150                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8151                                       off);
8152 #else
8153                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8154 #endif
8155                 break;
8156
8157         case offsetof(struct bpf_sock_ops, remote_port):
8158                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
8159
8160                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8161                                                 struct bpf_sock_ops_kern, sk),
8162                                       si->dst_reg, si->src_reg,
8163                                       offsetof(struct bpf_sock_ops_kern, sk));
8164                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8165                                       offsetof(struct sock_common, skc_dport));
8166 #ifndef __BIG_ENDIAN_BITFIELD
8167                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8168 #endif
8169                 break;
8170
8171         case offsetof(struct bpf_sock_ops, local_port):
8172                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
8173
8174                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8175                                                 struct bpf_sock_ops_kern, sk),
8176                                       si->dst_reg, si->src_reg,
8177                                       offsetof(struct bpf_sock_ops_kern, sk));
8178                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8179                                       offsetof(struct sock_common, skc_num));
8180                 break;
8181
8182         case offsetof(struct bpf_sock_ops, is_fullsock):
8183                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8184                                                 struct bpf_sock_ops_kern,
8185                                                 is_fullsock),
8186                                       si->dst_reg, si->src_reg,
8187                                       offsetof(struct bpf_sock_ops_kern,
8188                                                is_fullsock));
8189                 break;
8190
8191         case offsetof(struct bpf_sock_ops, state):
8192                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
8193
8194                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8195                                                 struct bpf_sock_ops_kern, sk),
8196                                       si->dst_reg, si->src_reg,
8197                                       offsetof(struct bpf_sock_ops_kern, sk));
8198                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
8199                                       offsetof(struct sock_common, skc_state));
8200                 break;
8201
8202         case offsetof(struct bpf_sock_ops, rtt_min):
8203                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
8204                              sizeof(struct minmax));
8205                 BUILD_BUG_ON(sizeof(struct minmax) <
8206                              sizeof(struct minmax_sample));
8207
8208                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8209                                                 struct bpf_sock_ops_kern, sk),
8210                                       si->dst_reg, si->src_reg,
8211                                       offsetof(struct bpf_sock_ops_kern, sk));
8212                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8213                                       offsetof(struct tcp_sock, rtt_min) +
8214                                       FIELD_SIZEOF(struct minmax_sample, t));
8215                 break;
8216
8217         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
8218                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
8219                                    struct tcp_sock);
8220                 break;
8221
8222         case offsetof(struct bpf_sock_ops, sk_txhash):
8223                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
8224                                           struct sock, type);
8225                 break;
8226         case offsetof(struct bpf_sock_ops, snd_cwnd):
8227                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
8228                 break;
8229         case offsetof(struct bpf_sock_ops, srtt_us):
8230                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
8231                 break;
8232         case offsetof(struct bpf_sock_ops, snd_ssthresh):
8233                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
8234                 break;
8235         case offsetof(struct bpf_sock_ops, rcv_nxt):
8236                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
8237                 break;
8238         case offsetof(struct bpf_sock_ops, snd_nxt):
8239                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
8240                 break;
8241         case offsetof(struct bpf_sock_ops, snd_una):
8242                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
8243                 break;
8244         case offsetof(struct bpf_sock_ops, mss_cache):
8245                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
8246                 break;
8247         case offsetof(struct bpf_sock_ops, ecn_flags):
8248                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
8249                 break;
8250         case offsetof(struct bpf_sock_ops, rate_delivered):
8251                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
8252                 break;
8253         case offsetof(struct bpf_sock_ops, rate_interval_us):
8254                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
8255                 break;
8256         case offsetof(struct bpf_sock_ops, packets_out):
8257                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
8258                 break;
8259         case offsetof(struct bpf_sock_ops, retrans_out):
8260                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
8261                 break;
8262         case offsetof(struct bpf_sock_ops, total_retrans):
8263                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
8264                 break;
8265         case offsetof(struct bpf_sock_ops, segs_in):
8266                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
8267                 break;
8268         case offsetof(struct bpf_sock_ops, data_segs_in):
8269                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
8270                 break;
8271         case offsetof(struct bpf_sock_ops, segs_out):
8272                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
8273                 break;
8274         case offsetof(struct bpf_sock_ops, data_segs_out):
8275                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
8276                 break;
8277         case offsetof(struct bpf_sock_ops, lost_out):
8278                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
8279                 break;
8280         case offsetof(struct bpf_sock_ops, sacked_out):
8281                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
8282                 break;
8283         case offsetof(struct bpf_sock_ops, bytes_received):
8284                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
8285                 break;
8286         case offsetof(struct bpf_sock_ops, bytes_acked):
8287                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
8288                 break;
8289         case offsetof(struct bpf_sock_ops, sk):
8290                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8291                                                 struct bpf_sock_ops_kern,
8292                                                 is_fullsock),
8293                                       si->dst_reg, si->src_reg,
8294                                       offsetof(struct bpf_sock_ops_kern,
8295                                                is_fullsock));
8296                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8297                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8298                                                 struct bpf_sock_ops_kern, sk),
8299                                       si->dst_reg, si->src_reg,
8300                                       offsetof(struct bpf_sock_ops_kern, sk));
8301                 break;
8302         }
8303         return insn - insn_buf;
8304 }
8305
8306 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
8307                                      const struct bpf_insn *si,
8308                                      struct bpf_insn *insn_buf,
8309                                      struct bpf_prog *prog, u32 *target_size)
8310 {
8311         struct bpf_insn *insn = insn_buf;
8312         int off;
8313
8314         switch (si->off) {
8315         case offsetof(struct __sk_buff, data_end):
8316                 off  = si->off;
8317                 off -= offsetof(struct __sk_buff, data_end);
8318                 off += offsetof(struct sk_buff, cb);
8319                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
8320                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8321                                       si->src_reg, off);
8322                 break;
8323         default:
8324                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8325                                               target_size);
8326         }
8327
8328         return insn - insn_buf;
8329 }
8330
8331 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
8332                                      const struct bpf_insn *si,
8333                                      struct bpf_insn *insn_buf,
8334                                      struct bpf_prog *prog, u32 *target_size)
8335 {
8336         struct bpf_insn *insn = insn_buf;
8337 #if IS_ENABLED(CONFIG_IPV6)
8338         int off;
8339 #endif
8340
8341         /* convert ctx uses the fact sg element is first in struct */
8342         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
8343
8344         switch (si->off) {
8345         case offsetof(struct sk_msg_md, data):
8346                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
8347                                       si->dst_reg, si->src_reg,
8348                                       offsetof(struct sk_msg, data));
8349                 break;
8350         case offsetof(struct sk_msg_md, data_end):
8351                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
8352                                       si->dst_reg, si->src_reg,
8353                                       offsetof(struct sk_msg, data_end));
8354                 break;
8355         case offsetof(struct sk_msg_md, family):
8356                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
8357
8358                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8359                                               struct sk_msg, sk),
8360                                       si->dst_reg, si->src_reg,
8361                                       offsetof(struct sk_msg, sk));
8362                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8363                                       offsetof(struct sock_common, skc_family));
8364                 break;
8365
8366         case offsetof(struct sk_msg_md, remote_ip4):
8367                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
8368
8369                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8370                                                 struct sk_msg, sk),
8371                                       si->dst_reg, si->src_reg,
8372                                       offsetof(struct sk_msg, sk));
8373                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8374                                       offsetof(struct sock_common, skc_daddr));
8375                 break;
8376
8377         case offsetof(struct sk_msg_md, local_ip4):
8378                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8379                                           skc_rcv_saddr) != 4);
8380
8381                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8382                                               struct sk_msg, sk),
8383                                       si->dst_reg, si->src_reg,
8384                                       offsetof(struct sk_msg, sk));
8385                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8386                                       offsetof(struct sock_common,
8387                                                skc_rcv_saddr));
8388                 break;
8389
8390         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
8391              offsetof(struct sk_msg_md, remote_ip6[3]):
8392 #if IS_ENABLED(CONFIG_IPV6)
8393                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8394                                           skc_v6_daddr.s6_addr32[0]) != 4);
8395
8396                 off = si->off;
8397                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
8398                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8399                                                 struct sk_msg, sk),
8400                                       si->dst_reg, si->src_reg,
8401                                       offsetof(struct sk_msg, sk));
8402                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8403                                       offsetof(struct sock_common,
8404                                                skc_v6_daddr.s6_addr32[0]) +
8405                                       off);
8406 #else
8407                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8408 #endif
8409                 break;
8410
8411         case offsetof(struct sk_msg_md, local_ip6[0]) ...
8412              offsetof(struct sk_msg_md, local_ip6[3]):
8413 #if IS_ENABLED(CONFIG_IPV6)
8414                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
8415                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8416
8417                 off = si->off;
8418                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
8419                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8420                                                 struct sk_msg, sk),
8421                                       si->dst_reg, si->src_reg,
8422                                       offsetof(struct sk_msg, sk));
8423                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8424                                       offsetof(struct sock_common,
8425                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8426                                       off);
8427 #else
8428                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8429 #endif
8430                 break;
8431
8432         case offsetof(struct sk_msg_md, remote_port):
8433                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
8434
8435                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8436                                                 struct sk_msg, sk),
8437                                       si->dst_reg, si->src_reg,
8438                                       offsetof(struct sk_msg, sk));
8439                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8440                                       offsetof(struct sock_common, skc_dport));
8441 #ifndef __BIG_ENDIAN_BITFIELD
8442                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8443 #endif
8444                 break;
8445
8446         case offsetof(struct sk_msg_md, local_port):
8447                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
8448
8449                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
8450                                                 struct sk_msg, sk),
8451                                       si->dst_reg, si->src_reg,
8452                                       offsetof(struct sk_msg, sk));
8453                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8454                                       offsetof(struct sock_common, skc_num));
8455                 break;
8456
8457         case offsetof(struct sk_msg_md, size):
8458                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
8459                                       si->dst_reg, si->src_reg,
8460                                       offsetof(struct sk_msg_sg, size));
8461                 break;
8462         }
8463
8464         return insn - insn_buf;
8465 }
8466
8467 const struct bpf_verifier_ops sk_filter_verifier_ops = {
8468         .get_func_proto         = sk_filter_func_proto,
8469         .is_valid_access        = sk_filter_is_valid_access,
8470         .convert_ctx_access     = bpf_convert_ctx_access,
8471         .gen_ld_abs             = bpf_gen_ld_abs,
8472 };
8473
8474 const struct bpf_prog_ops sk_filter_prog_ops = {
8475         .test_run               = bpf_prog_test_run_skb,
8476 };
8477
8478 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
8479         .get_func_proto         = tc_cls_act_func_proto,
8480         .is_valid_access        = tc_cls_act_is_valid_access,
8481         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
8482         .gen_prologue           = tc_cls_act_prologue,
8483         .gen_ld_abs             = bpf_gen_ld_abs,
8484 };
8485
8486 const struct bpf_prog_ops tc_cls_act_prog_ops = {
8487         .test_run               = bpf_prog_test_run_skb,
8488 };
8489
8490 const struct bpf_verifier_ops xdp_verifier_ops = {
8491         .get_func_proto         = xdp_func_proto,
8492         .is_valid_access        = xdp_is_valid_access,
8493         .convert_ctx_access     = xdp_convert_ctx_access,
8494         .gen_prologue           = bpf_noop_prologue,
8495 };
8496
8497 const struct bpf_prog_ops xdp_prog_ops = {
8498         .test_run               = bpf_prog_test_run_xdp,
8499 };
8500
8501 const struct bpf_verifier_ops cg_skb_verifier_ops = {
8502         .get_func_proto         = cg_skb_func_proto,
8503         .is_valid_access        = cg_skb_is_valid_access,
8504         .convert_ctx_access     = bpf_convert_ctx_access,
8505 };
8506
8507 const struct bpf_prog_ops cg_skb_prog_ops = {
8508         .test_run               = bpf_prog_test_run_skb,
8509 };
8510
8511 const struct bpf_verifier_ops lwt_in_verifier_ops = {
8512         .get_func_proto         = lwt_in_func_proto,
8513         .is_valid_access        = lwt_is_valid_access,
8514         .convert_ctx_access     = bpf_convert_ctx_access,
8515 };
8516
8517 const struct bpf_prog_ops lwt_in_prog_ops = {
8518         .test_run               = bpf_prog_test_run_skb,
8519 };
8520
8521 const struct bpf_verifier_ops lwt_out_verifier_ops = {
8522         .get_func_proto         = lwt_out_func_proto,
8523         .is_valid_access        = lwt_is_valid_access,
8524         .convert_ctx_access     = bpf_convert_ctx_access,
8525 };
8526
8527 const struct bpf_prog_ops lwt_out_prog_ops = {
8528         .test_run               = bpf_prog_test_run_skb,
8529 };
8530
8531 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
8532         .get_func_proto         = lwt_xmit_func_proto,
8533         .is_valid_access        = lwt_is_valid_access,
8534         .convert_ctx_access     = bpf_convert_ctx_access,
8535         .gen_prologue           = tc_cls_act_prologue,
8536 };
8537
8538 const struct bpf_prog_ops lwt_xmit_prog_ops = {
8539         .test_run               = bpf_prog_test_run_skb,
8540 };
8541
8542 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
8543         .get_func_proto         = lwt_seg6local_func_proto,
8544         .is_valid_access        = lwt_is_valid_access,
8545         .convert_ctx_access     = bpf_convert_ctx_access,
8546 };
8547
8548 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
8549         .test_run               = bpf_prog_test_run_skb,
8550 };
8551
8552 const struct bpf_verifier_ops cg_sock_verifier_ops = {
8553         .get_func_proto         = sock_filter_func_proto,
8554         .is_valid_access        = sock_filter_is_valid_access,
8555         .convert_ctx_access     = bpf_sock_convert_ctx_access,
8556 };
8557
8558 const struct bpf_prog_ops cg_sock_prog_ops = {
8559 };
8560
8561 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
8562         .get_func_proto         = sock_addr_func_proto,
8563         .is_valid_access        = sock_addr_is_valid_access,
8564         .convert_ctx_access     = sock_addr_convert_ctx_access,
8565 };
8566
8567 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
8568 };
8569
8570 const struct bpf_verifier_ops sock_ops_verifier_ops = {
8571         .get_func_proto         = sock_ops_func_proto,
8572         .is_valid_access        = sock_ops_is_valid_access,
8573         .convert_ctx_access     = sock_ops_convert_ctx_access,
8574 };
8575
8576 const struct bpf_prog_ops sock_ops_prog_ops = {
8577 };
8578
8579 const struct bpf_verifier_ops sk_skb_verifier_ops = {
8580         .get_func_proto         = sk_skb_func_proto,
8581         .is_valid_access        = sk_skb_is_valid_access,
8582         .convert_ctx_access     = sk_skb_convert_ctx_access,
8583         .gen_prologue           = sk_skb_prologue,
8584 };
8585
8586 const struct bpf_prog_ops sk_skb_prog_ops = {
8587 };
8588
8589 const struct bpf_verifier_ops sk_msg_verifier_ops = {
8590         .get_func_proto         = sk_msg_func_proto,
8591         .is_valid_access        = sk_msg_is_valid_access,
8592         .convert_ctx_access     = sk_msg_convert_ctx_access,
8593         .gen_prologue           = bpf_noop_prologue,
8594 };
8595
8596 const struct bpf_prog_ops sk_msg_prog_ops = {
8597 };
8598
8599 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
8600         .get_func_proto         = flow_dissector_func_proto,
8601         .is_valid_access        = flow_dissector_is_valid_access,
8602         .convert_ctx_access     = flow_dissector_convert_ctx_access,
8603 };
8604
8605 const struct bpf_prog_ops flow_dissector_prog_ops = {
8606         .test_run               = bpf_prog_test_run_flow_dissector,
8607 };
8608
8609 int sk_detach_filter(struct sock *sk)
8610 {
8611         int ret = -ENOENT;
8612         struct sk_filter *filter;
8613
8614         if (sock_flag(sk, SOCK_FILTER_LOCKED))
8615                 return -EPERM;
8616
8617         filter = rcu_dereference_protected(sk->sk_filter,
8618                                            lockdep_sock_is_held(sk));
8619         if (filter) {
8620                 RCU_INIT_POINTER(sk->sk_filter, NULL);
8621                 sk_filter_uncharge(sk, filter);
8622                 ret = 0;
8623         }
8624
8625         return ret;
8626 }
8627 EXPORT_SYMBOL_GPL(sk_detach_filter);
8628
8629 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
8630                   unsigned int len)
8631 {
8632         struct sock_fprog_kern *fprog;
8633         struct sk_filter *filter;
8634         int ret = 0;
8635
8636         lock_sock(sk);
8637         filter = rcu_dereference_protected(sk->sk_filter,
8638                                            lockdep_sock_is_held(sk));
8639         if (!filter)
8640                 goto out;
8641
8642         /* We're copying the filter that has been originally attached,
8643          * so no conversion/decode needed anymore. eBPF programs that
8644          * have no original program cannot be dumped through this.
8645          */
8646         ret = -EACCES;
8647         fprog = filter->prog->orig_prog;
8648         if (!fprog)
8649                 goto out;
8650
8651         ret = fprog->len;
8652         if (!len)
8653                 /* User space only enquires number of filter blocks. */
8654                 goto out;
8655
8656         ret = -EINVAL;
8657         if (len < fprog->len)
8658                 goto out;
8659
8660         ret = -EFAULT;
8661         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
8662                 goto out;
8663
8664         /* Instead of bytes, the API requests to return the number
8665          * of filter blocks.
8666          */
8667         ret = fprog->len;
8668 out:
8669         release_sock(sk);
8670         return ret;
8671 }
8672
8673 #ifdef CONFIG_INET
8674 struct sk_reuseport_kern {
8675         struct sk_buff *skb;
8676         struct sock *sk;
8677         struct sock *selected_sk;
8678         void *data_end;
8679         u32 hash;
8680         u32 reuseport_id;
8681         bool bind_inany;
8682 };
8683
8684 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
8685                                     struct sock_reuseport *reuse,
8686                                     struct sock *sk, struct sk_buff *skb,
8687                                     u32 hash)
8688 {
8689         reuse_kern->skb = skb;
8690         reuse_kern->sk = sk;
8691         reuse_kern->selected_sk = NULL;
8692         reuse_kern->data_end = skb->data + skb_headlen(skb);
8693         reuse_kern->hash = hash;
8694         reuse_kern->reuseport_id = reuse->reuseport_id;
8695         reuse_kern->bind_inany = reuse->bind_inany;
8696 }
8697
8698 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
8699                                   struct bpf_prog *prog, struct sk_buff *skb,
8700                                   u32 hash)
8701 {
8702         struct sk_reuseport_kern reuse_kern;
8703         enum sk_action action;
8704
8705         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
8706         action = BPF_PROG_RUN(prog, &reuse_kern);
8707
8708         if (action == SK_PASS)
8709                 return reuse_kern.selected_sk;
8710         else
8711                 return ERR_PTR(-ECONNREFUSED);
8712 }
8713
8714 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
8715            struct bpf_map *, map, void *, key, u32, flags)
8716 {
8717         struct sock_reuseport *reuse;
8718         struct sock *selected_sk;
8719
8720         selected_sk = map->ops->map_lookup_elem(map, key);
8721         if (!selected_sk)
8722                 return -ENOENT;
8723
8724         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
8725         if (!reuse)
8726                 /* selected_sk is unhashed (e.g. by close()) after the
8727                  * above map_lookup_elem().  Treat selected_sk has already
8728                  * been removed from the map.
8729                  */
8730                 return -ENOENT;
8731
8732         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
8733                 struct sock *sk;
8734
8735                 if (unlikely(!reuse_kern->reuseport_id))
8736                         /* There is a small race between adding the
8737                          * sk to the map and setting the
8738                          * reuse_kern->reuseport_id.
8739                          * Treat it as the sk has not been added to
8740                          * the bpf map yet.
8741                          */
8742                         return -ENOENT;
8743
8744                 sk = reuse_kern->sk;
8745                 if (sk->sk_protocol != selected_sk->sk_protocol)
8746                         return -EPROTOTYPE;
8747                 else if (sk->sk_family != selected_sk->sk_family)
8748                         return -EAFNOSUPPORT;
8749
8750                 /* Catch all. Likely bound to a different sockaddr. */
8751                 return -EBADFD;
8752         }
8753
8754         reuse_kern->selected_sk = selected_sk;
8755
8756         return 0;
8757 }
8758
8759 static const struct bpf_func_proto sk_select_reuseport_proto = {
8760         .func           = sk_select_reuseport,
8761         .gpl_only       = false,
8762         .ret_type       = RET_INTEGER,
8763         .arg1_type      = ARG_PTR_TO_CTX,
8764         .arg2_type      = ARG_CONST_MAP_PTR,
8765         .arg3_type      = ARG_PTR_TO_MAP_KEY,
8766         .arg4_type      = ARG_ANYTHING,
8767 };
8768
8769 BPF_CALL_4(sk_reuseport_load_bytes,
8770            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8771            void *, to, u32, len)
8772 {
8773         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
8774 }
8775
8776 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
8777         .func           = sk_reuseport_load_bytes,
8778         .gpl_only       = false,
8779         .ret_type       = RET_INTEGER,
8780         .arg1_type      = ARG_PTR_TO_CTX,
8781         .arg2_type      = ARG_ANYTHING,
8782         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
8783         .arg4_type      = ARG_CONST_SIZE,
8784 };
8785
8786 BPF_CALL_5(sk_reuseport_load_bytes_relative,
8787            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
8788            void *, to, u32, len, u32, start_header)
8789 {
8790         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
8791                                                len, start_header);
8792 }
8793
8794 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
8795         .func           = sk_reuseport_load_bytes_relative,
8796         .gpl_only       = false,
8797         .ret_type       = RET_INTEGER,
8798         .arg1_type      = ARG_PTR_TO_CTX,
8799         .arg2_type      = ARG_ANYTHING,
8800         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
8801         .arg4_type      = ARG_CONST_SIZE,
8802         .arg5_type      = ARG_ANYTHING,
8803 };
8804
8805 static const struct bpf_func_proto *
8806 sk_reuseport_func_proto(enum bpf_func_id func_id,
8807                         const struct bpf_prog *prog)
8808 {
8809         switch (func_id) {
8810         case BPF_FUNC_sk_select_reuseport:
8811                 return &sk_select_reuseport_proto;
8812         case BPF_FUNC_skb_load_bytes:
8813                 return &sk_reuseport_load_bytes_proto;
8814         case BPF_FUNC_skb_load_bytes_relative:
8815                 return &sk_reuseport_load_bytes_relative_proto;
8816         default:
8817                 return bpf_base_func_proto(func_id);
8818         }
8819 }
8820
8821 static bool
8822 sk_reuseport_is_valid_access(int off, int size,
8823                              enum bpf_access_type type,
8824                              const struct bpf_prog *prog,
8825                              struct bpf_insn_access_aux *info)
8826 {
8827         const u32 size_default = sizeof(__u32);
8828
8829         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
8830             off % size || type != BPF_READ)
8831                 return false;
8832
8833         switch (off) {
8834         case offsetof(struct sk_reuseport_md, data):
8835                 info->reg_type = PTR_TO_PACKET;
8836                 return size == sizeof(__u64);
8837
8838         case offsetof(struct sk_reuseport_md, data_end):
8839                 info->reg_type = PTR_TO_PACKET_END;
8840                 return size == sizeof(__u64);
8841
8842         case offsetof(struct sk_reuseport_md, hash):
8843                 return size == size_default;
8844
8845         /* Fields that allow narrowing */
8846         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
8847                 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
8848                         return false;
8849                 /* fall through */
8850         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
8851         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
8852         case bpf_ctx_range(struct sk_reuseport_md, len):
8853                 bpf_ctx_record_field_size(info, size_default);
8854                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8855
8856         default:
8857                 return false;
8858         }
8859 }
8860
8861 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
8862         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8863                               si->dst_reg, si->src_reg,                 \
8864                               bpf_target_off(struct sk_reuseport_kern, F, \
8865                                              FIELD_SIZEOF(struct sk_reuseport_kern, F), \
8866                                              target_size));             \
8867         })
8868
8869 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
8870         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
8871                                     struct sk_buff,                     \
8872                                     skb,                                \
8873                                     SKB_FIELD)
8874
8875 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
8876         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern,  \
8877                                              struct sock,               \
8878                                              sk,                        \
8879                                              SK_FIELD, BPF_SIZE, EXTRA_OFF)
8880
8881 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
8882                                            const struct bpf_insn *si,
8883                                            struct bpf_insn *insn_buf,
8884                                            struct bpf_prog *prog,
8885                                            u32 *target_size)
8886 {
8887         struct bpf_insn *insn = insn_buf;
8888
8889         switch (si->off) {
8890         case offsetof(struct sk_reuseport_md, data):
8891                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
8892                 break;
8893
8894         case offsetof(struct sk_reuseport_md, len):
8895                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
8896                 break;
8897
8898         case offsetof(struct sk_reuseport_md, eth_protocol):
8899                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
8900                 break;
8901
8902         case offsetof(struct sk_reuseport_md, ip_protocol):
8903                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
8904                 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
8905                                                     BPF_W, 0);
8906                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
8907                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
8908                                         SK_FL_PROTO_SHIFT);
8909                 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
8910                  * aware.  No further narrowing or masking is needed.
8911                  */
8912                 *target_size = 1;
8913                 break;
8914
8915         case offsetof(struct sk_reuseport_md, data_end):
8916                 SK_REUSEPORT_LOAD_FIELD(data_end);
8917                 break;
8918
8919         case offsetof(struct sk_reuseport_md, hash):
8920                 SK_REUSEPORT_LOAD_FIELD(hash);
8921                 break;
8922
8923         case offsetof(struct sk_reuseport_md, bind_inany):
8924                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
8925                 break;
8926         }
8927
8928         return insn - insn_buf;
8929 }
8930
8931 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
8932         .get_func_proto         = sk_reuseport_func_proto,
8933         .is_valid_access        = sk_reuseport_is_valid_access,
8934         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
8935 };
8936
8937 const struct bpf_prog_ops sk_reuseport_prog_ops = {
8938 };
8939 #endif /* CONFIG_INET */