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/atomic.h>
  21 #include <linux/module.h>
  22 #include <linux/types.h>
  23 #include <linux/mm.h>
  24 #include <linux/fcntl.h>
  25 #include <linux/socket.h>
  26 #include <linux/sock_diag.h>
  27 #include <linux/in.h>
  28 #include <linux/inet.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_packet.h>
  31 #include <linux/if_arp.h>
  32 #include <linux/gfp.h>
  33 #include <net/inet_common.h>
  34 #include <net/ip.h>
  35 #include <net/protocol.h>
  36 #include <net/netlink.h>
  37 #include <linux/skbuff.h>
  38 #include <linux/skmsg.h>
  39 #include <net/sock.h>
  40 #include <net/flow_dissector.h>
  41 #include <linux/errno.h>
  42 #include <linux/timer.h>
  43 #include <linux/uaccess.h>
  44 #include <asm/unaligned.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 <linux/btf.h>
  51 #include <net/sch_generic.h>
  52 #include <net/cls_cgroup.h>
  53 #include <net/dst_metadata.h>
  54 #include <net/dst.h>
  55 #include <net/sock_reuseport.h>
  56 #include <net/busy_poll.h>
  57 #include <net/tcp.h>
  58 #include <net/xfrm.h>
  59 #include <net/udp.h>
  60 #include <linux/bpf_trace.h>
  61 #include <net/xdp_sock.h>
  62 #include <linux/inetdevice.h>
  63 #include <net/inet_hashtables.h>
  64 #include <net/inet6_hashtables.h>
  65 #include <net/ip_fib.h>
  66 #include <net/nexthop.h>
  67 #include <net/flow.h>
  68 #include <net/arp.h>
  69 #include <net/ipv6.h>
  70 #include <net/net_namespace.h>
  71 #include <linux/seg6_local.h>
  72 #include <net/seg6.h>
  73 #include <net/seg6_local.h>
  74 #include <net/lwtunnel.h>
  75 #include <net/ipv6_stubs.h>
  76 #include <net/bpf_sk_storage.h>
  77 #include <net/transp_v6.h>
  78 #include <linux/btf_ids.h>
  79 #include <net/tls.h>
  80 #include <net/xdp.h>
  81
  82 static const struct bpf_func_proto *
  83 bpf_sk_base_func_proto(enum bpf_func_id func_id);
  84
  85 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
  86 {
  87         if (in_compat_syscall()) {
  88                 struct compat_sock_fprog f32;
  89
  90                 if (len != sizeof(f32))
  91                         return -EINVAL;
  92                 if (copy_from_sockptr(&f32, src, sizeof(f32)))
  93                         return -EFAULT;
  94                 memset(dst, 0, sizeof(*dst));
  95                 dst->len = f32.len;
  96                 dst->filter = compat_ptr(f32.filter);
  97         } else {
  98                 if (len != sizeof(*dst))
  99                         return -EINVAL;
 100                 if (copy_from_sockptr(dst, src, sizeof(*dst)))
 101                         return -EFAULT;
 102         }
 103
 104         return 0;
 105 }
 106 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
 107
 108 /**
 109  *      sk_filter_trim_cap - run a packet through a socket filter
 110  *      @sk: sock associated with &sk_buff
 111  *      @skb: buffer to filter
 112  *      @cap: limit on how short the eBPF program may trim the packet
 113  *
 114  * Run the eBPF program and then cut skb->data to correct size returned by
 115  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
 116  * than pkt_len we keep whole skb->data. This is the socket level
 117  * wrapper to bpf_prog_run. It returns 0 if the packet should
 118  * be accepted or -EPERM if the packet should be tossed.
 119  *
 120  */
 121 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
 122 {
 123         int err;
 124         struct sk_filter *filter;
 125
 126         /*
 127          * If the skb was allocated from pfmemalloc reserves, only
 128          * allow SOCK_MEMALLOC sockets to use it as this socket is
 129          * helping free memory
 130          */
 131         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 132                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 133                 return -ENOMEM;
 134         }
 135         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 136         if (err)
 137                 return err;
 138
 139         err = security_sock_rcv_skb(sk, skb);
 140         if (err)
 141                 return err;
 142
 143         rcu_read_lock();
 144         filter = rcu_dereference(sk->sk_filter);
 145         if (filter) {
 146                 struct sock *save_sk = skb->sk;
 147                 unsigned int pkt_len;
 148
 149                 skb->sk = sk;
 150                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 151                 skb->sk = save_sk;
 152                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 153         }
 154         rcu_read_unlock();
 155
 156         return err;
 157 }
 158 EXPORT_SYMBOL(sk_filter_trim_cap);
 159
 160 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 161 {
 162         return skb_get_poff(skb);
 163 }
 164
 165 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 166 {
 167         struct nlattr *nla;
 168
 169         if (skb_is_nonlinear(skb))
 170                 return 0;
 171
 172         if (skb->len < sizeof(struct nlattr))
 173                 return 0;
 174
 175         if (a > skb->len - sizeof(struct nlattr))
 176                 return 0;
 177
 178         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 179         if (nla)
 180                 return (void *) nla - (void *) skb->data;
 181
 182         return 0;
 183 }
 184
 185 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 186 {
 187         struct nlattr *nla;
 188
 189         if (skb_is_nonlinear(skb))
 190                 return 0;
 191
 192         if (skb->len < sizeof(struct nlattr))
 193                 return 0;
 194
 195         if (a > skb->len - sizeof(struct nlattr))
 196                 return 0;
 197
 198         nla = (struct nlattr *) &skb->data[a];
 199         if (nla->nla_len > skb->len - a)
 200                 return 0;
 201
 202         nla = nla_find_nested(nla, x);
 203         if (nla)
 204                 return (void *) nla - (void *) skb->data;
 205
 206         return 0;
 207 }
 208
 209 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 210            data, int, headlen, int, offset)
 211 {
 212         u8 tmp, *ptr;
 213         const int len = sizeof(tmp);
 214
 215         if (offset >= 0) {
 216                 if (headlen - offset >= len)
 217                         return *(u8 *)(data + offset);
 218                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 219                         return tmp;
 220         } else {
 221                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 222                 if (likely(ptr))
 223                         return *(u8 *)ptr;
 224         }
 225
 226         return -EFAULT;
 227 }
 228
 229 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 230            int, offset)
 231 {
 232         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 233                                          offset);
 234 }
 235
 236 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 237            data, int, headlen, int, offset)
 238 {
 239         u16 tmp, *ptr;
 240         const int len = sizeof(tmp);
 241
 242         if (offset >= 0) {
 243                 if (headlen - offset >= len)
 244                         return get_unaligned_be16(data + offset);
 245                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 246                         return be16_to_cpu(tmp);
 247         } else {
 248                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 249                 if (likely(ptr))
 250                         return get_unaligned_be16(ptr);
 251         }
 252
 253         return -EFAULT;
 254 }
 255
 256 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 257            int, offset)
 258 {
 259         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 260                                           offset);
 261 }
 262
 263 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 264            data, int, headlen, int, offset)
 265 {
 266         u32 tmp, *ptr;
 267         const int len = sizeof(tmp);
 268
 269         if (likely(offset >= 0)) {
 270                 if (headlen - offset >= len)
 271                         return get_unaligned_be32(data + offset);
 272                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 273                         return be32_to_cpu(tmp);
 274         } else {
 275                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 276                 if (likely(ptr))
 277                         return get_unaligned_be32(ptr);
 278         }
 279
 280         return -EFAULT;
 281 }
 282
 283 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 284            int, offset)
 285 {
 286         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 287                                           offset);
 288 }
 289
 290 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 291                               struct bpf_insn *insn_buf)
 292 {
 293         struct bpf_insn *insn = insn_buf;
 294
 295         switch (skb_field) {
 296         case SKF_AD_MARK:
 297                 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
 298
 299                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 300                                       offsetof(struct sk_buff, mark));
 301                 break;
 302
 303         case SKF_AD_PKTTYPE:
 304                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 305                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 306 #ifdef __BIG_ENDIAN_BITFIELD
 307                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 308 #endif
 309                 break;
 310
 311         case SKF_AD_QUEUE:
 312                 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
 313
 314                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 315                                       offsetof(struct sk_buff, queue_mapping));
 316                 break;
 317
 318         case SKF_AD_VLAN_TAG:
 319                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
 320
 321                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 322                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 323                                       offsetof(struct sk_buff, vlan_tci));
 324                 break;
 325         case SKF_AD_VLAN_TAG_PRESENT:
 326                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
 327                 if (PKT_VLAN_PRESENT_BIT)
 328                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
 329                 if (PKT_VLAN_PRESENT_BIT < 7)
 330                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 331                 break;
 332         }
 333
 334         return insn - insn_buf;
 335 }
 336
 337 static bool convert_bpf_extensions(struct sock_filter *fp,
 338                                    struct bpf_insn **insnp)
 339 {
 340         struct bpf_insn *insn = *insnp;
 341         u32 cnt;
 342
 343         switch (fp->k) {
 344         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 345                 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
 346
 347                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 348                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 349                                       offsetof(struct sk_buff, protocol));
 350                 /* A = ntohs(A) [emitting a nop or swap16] */
 351                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 352                 break;
 353
 354         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 355                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 356                 insn += cnt - 1;
 357                 break;
 358
 359         case SKF_AD_OFF + SKF_AD_IFINDEX:
 360         case SKF_AD_OFF + SKF_AD_HATYPE:
 361                 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
 362                 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
 363
 364                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 365                                       BPF_REG_TMP, BPF_REG_CTX,
 366                                       offsetof(struct sk_buff, dev));
 367                 /* if (tmp != 0) goto pc + 1 */
 368                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 369                 *insn++ = BPF_EXIT_INSN();
 370                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 371                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 372                                             offsetof(struct net_device, ifindex));
 373                 else
 374                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 375                                             offsetof(struct net_device, type));
 376                 break;
 377
 378         case SKF_AD_OFF + SKF_AD_MARK:
 379                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 380                 insn += cnt - 1;
 381                 break;
 382
 383         case SKF_AD_OFF + SKF_AD_RXHASH:
 384                 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
 385
 386                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 387                                     offsetof(struct sk_buff, hash));
 388                 break;
 389
 390         case SKF_AD_OFF + SKF_AD_QUEUE:
 391                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 392                 insn += cnt - 1;
 393                 break;
 394
 395         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 396                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 397                                          BPF_REG_A, BPF_REG_CTX, insn);
 398                 insn += cnt - 1;
 399                 break;
 400
 401         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 402                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 403                                          BPF_REG_A, BPF_REG_CTX, insn);
 404                 insn += cnt - 1;
 405                 break;
 406
 407         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 408                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
 409
 410                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 411                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 412                                       offsetof(struct sk_buff, vlan_proto));
 413                 /* A = ntohs(A) [emitting a nop or swap16] */
 414                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 415                 break;
 416
 417         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 418         case SKF_AD_OFF + SKF_AD_NLATTR:
 419         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 420         case SKF_AD_OFF + SKF_AD_CPU:
 421         case SKF_AD_OFF + SKF_AD_RANDOM:
 422                 /* arg1 = CTX */
 423                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 424                 /* arg2 = A */
 425                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 426                 /* arg3 = X */
 427                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 428                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 429                 switch (fp->k) {
 430                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 431                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 432                         break;
 433                 case SKF_AD_OFF + SKF_AD_NLATTR:
 434                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 435                         break;
 436                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 437                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 438                         break;
 439                 case SKF_AD_OFF + SKF_AD_CPU:
 440                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 441                         break;
 442                 case SKF_AD_OFF + SKF_AD_RANDOM:
 443                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 444                         bpf_user_rnd_init_once();
 445                         break;
 446                 }
 447                 break;
 448
 449         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 450                 /* A ^= X */
 451                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 452                 break;
 453
 454         default:
 455                 /* This is just a dummy call to avoid letting the compiler
 456                  * evict __bpf_call_base() as an optimization. Placed here
 457                  * where no-one bothers.
 458                  */
 459                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 460                 return false;
 461         }
 462
 463         *insnp = insn;
 464         return true;
 465 }
 466
 467 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 468 {
 469         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 470         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 471         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 472                       BPF_SIZE(fp->code) == BPF_W;
 473         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 474         const int ip_align = NET_IP_ALIGN;
 475         struct bpf_insn *insn = *insnp;
 476         int offset = fp->k;
 477
 478         if (!indirect &&
 479             ((unaligned_ok && offset >= 0) ||
 480              (!unaligned_ok && offset >= 0 &&
 481               offset + ip_align >= 0 &&
 482               offset + ip_align % size == 0))) {
 483                 bool ldx_off_ok = offset <= S16_MAX;
 484
 485                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 486                 if (offset)
 487                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 488                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 489                                       size, 2 + endian + (!ldx_off_ok * 2));
 490                 if (ldx_off_ok) {
 491                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 492                                               BPF_REG_D, offset);
 493                 } else {
 494                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 495                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 496                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 497                                               BPF_REG_TMP, 0);
 498                 }
 499                 if (endian)
 500                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 501                 *insn++ = BPF_JMP_A(8);
 502         }
 503
 504         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 505         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 506         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 507         if (!indirect) {
 508                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 509         } else {
 510                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 511                 if (fp->k)
 512                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 513         }
 514
 515         switch (BPF_SIZE(fp->code)) {
 516         case BPF_B:
 517                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 518                 break;
 519         case BPF_H:
 520                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 521                 break;
 522         case BPF_W:
 523                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 524                 break;
 525         default:
 526                 return false;
 527         }
 528
 529         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 530         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 531         *insn   = BPF_EXIT_INSN();
 532
 533         *insnp = insn;
 534         return true;
 535 }
 536
 537 /**
 538  *      bpf_convert_filter - convert filter program
 539  *      @prog: the user passed filter program
 540  *      @len: the length of the user passed filter program
 541  *      @new_prog: allocated 'struct bpf_prog' or NULL
 542  *      @new_len: pointer to store length of converted program
 543  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 544  *
 545  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 546  * style extended BPF (eBPF).
 547  * Conversion workflow:
 548  *
 549  * 1) First pass for calculating the new program length:
 550  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 551  *
 552  * 2) 2nd pass to remap in two passes: 1st pass finds new
 553  *    jump offsets, 2nd pass remapping:
 554  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 555  */
 556 static int bpf_convert_filter(struct sock_filter *prog, int len,
 557                               struct bpf_prog *new_prog, int *new_len,
 558                               bool *seen_ld_abs)
 559 {
 560         int new_flen = 0, pass = 0, target, i, stack_off;
 561         struct bpf_insn *new_insn, *first_insn = NULL;
 562         struct sock_filter *fp;
 563         int *addrs = NULL;
 564         u8 bpf_src;
 565
 566         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 567         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 568
 569         if (len <= 0 || len > BPF_MAXINSNS)
 570                 return -EINVAL;
 571
 572         if (new_prog) {
 573                 first_insn = new_prog->insnsi;
 574                 addrs = kcalloc(len, sizeof(*addrs),
 575                                 GFP_KERNEL | __GFP_NOWARN);
 576                 if (!addrs)
 577                         return -ENOMEM;
 578         }
 579
 580 do_pass:
 581         new_insn = first_insn;
 582         fp = prog;
 583
 584         /* Classic BPF related prologue emission. */
 585         if (new_prog) {
 586                 /* Classic BPF expects A and X to be reset first. These need
 587                  * to be guaranteed to be the first two instructions.
 588                  */
 589                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 590                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 591
 592                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 593                  * In eBPF case it's done by the compiler, here we need to
 594                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 595                  */
 596                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 597                 if (*seen_ld_abs) {
 598                         /* For packet access in classic BPF, cache skb->data
 599                          * in callee-saved BPF R8 and skb->len - skb->data_len
 600                          * (headlen) in BPF R9. Since classic BPF is read-only
 601                          * on CTX, we only need to cache it once.
 602                          */
 603                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 604                                                   BPF_REG_D, BPF_REG_CTX,
 605                                                   offsetof(struct sk_buff, data));
 606                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 607                                                   offsetof(struct sk_buff, len));
 608                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 609                                                   offsetof(struct sk_buff, data_len));
 610                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 611                 }
 612         } else {
 613                 new_insn += 3;
 614         }
 615
 616         for (i = 0; i < len; fp++, i++) {
 617                 struct bpf_insn tmp_insns[32] = { };
 618                 struct bpf_insn *insn = tmp_insns;
 619
 620                 if (addrs)
 621                         addrs[i] = new_insn - first_insn;
 622
 623                 switch (fp->code) {
 624                 /* All arithmetic insns and skb loads map as-is. */
 625                 case BPF_ALU | BPF_ADD | BPF_X:
 626                 case BPF_ALU | BPF_ADD | BPF_K:
 627                 case BPF_ALU | BPF_SUB | BPF_X:
 628                 case BPF_ALU | BPF_SUB | BPF_K:
 629                 case BPF_ALU | BPF_AND | BPF_X:
 630                 case BPF_ALU | BPF_AND | BPF_K:
 631                 case BPF_ALU | BPF_OR | BPF_X:
 632                 case BPF_ALU | BPF_OR | BPF_K:
 633                 case BPF_ALU | BPF_LSH | BPF_X:
 634                 case BPF_ALU | BPF_LSH | BPF_K:
 635                 case BPF_ALU | BPF_RSH | BPF_X:
 636                 case BPF_ALU | BPF_RSH | BPF_K:
 637                 case BPF_ALU | BPF_XOR | BPF_X:
 638                 case BPF_ALU | BPF_XOR | BPF_K:
 639                 case BPF_ALU | BPF_MUL | BPF_X:
 640                 case BPF_ALU | BPF_MUL | BPF_K:
 641                 case BPF_ALU | BPF_DIV | BPF_X:
 642                 case BPF_ALU | BPF_DIV | BPF_K:
 643                 case BPF_ALU | BPF_MOD | BPF_X:
 644                 case BPF_ALU | BPF_MOD | BPF_K:
 645                 case BPF_ALU | BPF_NEG:
 646                 case BPF_LD | BPF_ABS | BPF_W:
 647                 case BPF_LD | BPF_ABS | BPF_H:
 648                 case BPF_LD | BPF_ABS | BPF_B:
 649                 case BPF_LD | BPF_IND | BPF_W:
 650                 case BPF_LD | BPF_IND | BPF_H:
 651                 case BPF_LD | BPF_IND | BPF_B:
 652                         /* Check for overloaded BPF extension and
 653                          * directly convert it if found, otherwise
 654                          * just move on with mapping.
 655                          */
 656                         if (BPF_CLASS(fp->code) == BPF_LD &&
 657                             BPF_MODE(fp->code) == BPF_ABS &&
 658                             convert_bpf_extensions(fp, &insn))
 659                                 break;
 660                         if (BPF_CLASS(fp->code) == BPF_LD &&
 661                             convert_bpf_ld_abs(fp, &insn)) {
 662                                 *seen_ld_abs = true;
 663                                 break;
 664                         }
 665
 666                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 667                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 668                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 669                                 /* Error with exception code on div/mod by 0.
 670                                  * For cBPF programs, this was always return 0.
 671                                  */
 672                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 673                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 674                                 *insn++ = BPF_EXIT_INSN();
 675                         }
 676
 677                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 678                         break;
 679
 680                 /* Jump transformation cannot use BPF block macros
 681                  * everywhere as offset calculation and target updates
 682                  * require a bit more work than the rest, i.e. jump
 683                  * opcodes map as-is, but offsets need adjustment.
 684                  */
 685
 686 #define BPF_EMIT_JMP                                                    \
 687         do {                                                            \
 688                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 689                 s32 off;                                                \
 690                                                                         \
 691                 if (target >= len || target < 0)                        \
 692                         goto err;                                       \
 693                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 694                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 695                 off -= insn - tmp_insns;                                \
 696                 /* Reject anything not fitting into insn->off. */       \
 697                 if (off < off_min || off > off_max)                     \
 698                         goto err;                                       \
 699                 insn->off = off;                                        \
 700         } while (0)
 701
 702                 case BPF_JMP | BPF_JA:
 703                         target = i + fp->k + 1;
 704                         insn->code = fp->code;
 705                         BPF_EMIT_JMP;
 706                         break;
 707
 708                 case BPF_JMP | BPF_JEQ | BPF_K:
 709                 case BPF_JMP | BPF_JEQ | BPF_X:
 710                 case BPF_JMP | BPF_JSET | BPF_K:
 711                 case BPF_JMP | BPF_JSET | BPF_X:
 712                 case BPF_JMP | BPF_JGT | BPF_K:
 713                 case BPF_JMP | BPF_JGT | BPF_X:
 714                 case BPF_JMP | BPF_JGE | BPF_K:
 715                 case BPF_JMP | BPF_JGE | BPF_X:
 716                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 717                                 /* BPF immediates are signed, zero extend
 718                                  * immediate into tmp register and use it
 719                                  * in compare insn.
 720                                  */
 721                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 722
 723                                 insn->dst_reg = BPF_REG_A;
 724                                 insn->src_reg = BPF_REG_TMP;
 725                                 bpf_src = BPF_X;
 726                         } else {
 727                                 insn->dst_reg = BPF_REG_A;
 728                                 insn->imm = fp->k;
 729                                 bpf_src = BPF_SRC(fp->code);
 730                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 731                         }
 732
 733                         /* Common case where 'jump_false' is next insn. */
 734                         if (fp->jf == 0) {
 735                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 736                                 target = i + fp->jt + 1;
 737                                 BPF_EMIT_JMP;
 738                                 break;
 739                         }
 740
 741                         /* Convert some jumps when 'jump_true' is next insn. */
 742                         if (fp->jt == 0) {
 743                                 switch (BPF_OP(fp->code)) {
 744                                 case BPF_JEQ:
 745                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 746                                         break;
 747                                 case BPF_JGT:
 748                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 749                                         break;
 750                                 case BPF_JGE:
 751                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 752                                         break;
 753                                 default:
 754                                         goto jmp_rest;
 755                                 }
 756
 757                                 target = i + fp->jf + 1;
 758                                 BPF_EMIT_JMP;
 759                                 break;
 760                         }
 761 jmp_rest:
 762                         /* Other jumps are mapped into two insns: Jxx and JA. */
 763                         target = i + fp->jt + 1;
 764                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 765                         BPF_EMIT_JMP;
 766                         insn++;
 767
 768                         insn->code = BPF_JMP | BPF_JA;
 769                         target = i + fp->jf + 1;
 770                         BPF_EMIT_JMP;
 771                         break;
 772
 773                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 774                 case BPF_LDX | BPF_MSH | BPF_B: {
 775                         struct sock_filter tmp = {
 776                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 777                                 .k      = fp->k,
 778                         };
 779
 780                         *seen_ld_abs = true;
 781
 782                         /* X = A */
 783                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 784                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 785                         convert_bpf_ld_abs(&tmp, &insn);
 786                         insn++;
 787                         /* A &= 0xf */
 788                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 789                         /* A <<= 2 */
 790                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 791                         /* tmp = X */
 792                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 793                         /* X = A */
 794                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 795                         /* A = tmp */
 796                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 797                         break;
 798                 }
 799                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 800                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 801                  */
 802                 case BPF_RET | BPF_A:
 803                 case BPF_RET | BPF_K:
 804                         if (BPF_RVAL(fp->code) == BPF_K)
 805                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 806                                                         0, fp->k);
 807                         *insn = BPF_EXIT_INSN();
 808                         break;
 809
 810                 /* Store to stack. */
 811                 case BPF_ST:
 812                 case BPF_STX:
 813                         stack_off = fp->k * 4  + 4;
 814                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 815                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 816                                             -stack_off);
 817                         /* check_load_and_stores() verifies that classic BPF can
 818                          * load from stack only after write, so tracking
 819                          * stack_depth for ST|STX insns is enough
 820                          */
 821                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 822                                 new_prog->aux->stack_depth = stack_off;
 823                         break;
 824
 825                 /* Load from stack. */
 826                 case BPF_LD | BPF_MEM:
 827                 case BPF_LDX | BPF_MEM:
 828                         stack_off = fp->k * 4  + 4;
 829                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 830                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 831                                             -stack_off);
 832                         break;
 833
 834                 /* A = K or X = K */
 835                 case BPF_LD | BPF_IMM:
 836                 case BPF_LDX | BPF_IMM:
 837                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 838                                               BPF_REG_A : BPF_REG_X, fp->k);
 839                         break;
 840
 841                 /* X = A */
 842                 case BPF_MISC | BPF_TAX:
 843                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 844                         break;
 845
 846                 /* A = X */
 847                 case BPF_MISC | BPF_TXA:
 848                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 849                         break;
 850
 851                 /* A = skb->len or X = skb->len */
 852                 case BPF_LD | BPF_W | BPF_LEN:
 853                 case BPF_LDX | BPF_W | BPF_LEN:
 854                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 855                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 856                                             offsetof(struct sk_buff, len));
 857                         break;
 858
 859                 /* Access seccomp_data fields. */
 860                 case BPF_LDX | BPF_ABS | BPF_W:
 861                         /* A = *(u32 *) (ctx + K) */
 862                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 863                         break;
 864
 865                 /* Unknown instruction. */
 866                 default:
 867                         goto err;
 868                 }
 869
 870                 insn++;
 871                 if (new_prog)
 872                         memcpy(new_insn, tmp_insns,
 873                                sizeof(*insn) * (insn - tmp_insns));
 874                 new_insn += insn - tmp_insns;
 875         }
 876
 877         if (!new_prog) {
 878                 /* Only calculating new length. */
 879                 *new_len = new_insn - first_insn;
 880                 if (*seen_ld_abs)
 881                         *new_len += 4; /* Prologue bits. */
 882                 return 0;
 883         }
 884
 885         pass++;
 886         if (new_flen != new_insn - first_insn) {
 887                 new_flen = new_insn - first_insn;
 888                 if (pass > 2)
 889                         goto err;
 890                 goto do_pass;
 891         }
 892
 893         kfree(addrs);
 894         BUG_ON(*new_len != new_flen);
 895         return 0;
 896 err:
 897         kfree(addrs);
 898         return -EINVAL;
 899 }
 900
 901 /* Security:
 902  *
 903  * As we dont want to clear mem[] array for each packet going through
 904  * __bpf_prog_run(), we check that filter loaded by user never try to read
 905  * a cell if not previously written, and we check all branches to be sure
 906  * a malicious user doesn't try to abuse us.
 907  */
 908 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 909 {
 910         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 911         int pc, ret = 0;
 912
 913         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 914
 915         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 916         if (!masks)
 917                 return -ENOMEM;
 918
 919         memset(masks, 0xff, flen * sizeof(*masks));
 920
 921         for (pc = 0; pc < flen; pc++) {
 922                 memvalid &= masks[pc];
 923
 924                 switch (filter[pc].code) {
 925                 case BPF_ST:
 926                 case BPF_STX:
 927                         memvalid |= (1 << filter[pc].k);
 928                         break;
 929                 case BPF_LD | BPF_MEM:
 930                 case BPF_LDX | BPF_MEM:
 931                         if (!(memvalid & (1 << filter[pc].k))) {
 932                                 ret = -EINVAL;
 933                                 goto error;
 934                         }
 935                         break;
 936                 case BPF_JMP | BPF_JA:
 937                         /* A jump must set masks on target */
 938                         masks[pc + 1 + filter[pc].k] &= memvalid;
 939                         memvalid = ~0;
 940                         break;
 941                 case BPF_JMP | BPF_JEQ | BPF_K:
 942                 case BPF_JMP | BPF_JEQ | BPF_X:
 943                 case BPF_JMP | BPF_JGE | BPF_K:
 944                 case BPF_JMP | BPF_JGE | BPF_X:
 945                 case BPF_JMP | BPF_JGT | BPF_K:
 946                 case BPF_JMP | BPF_JGT | BPF_X:
 947                 case BPF_JMP | BPF_JSET | BPF_K:
 948                 case BPF_JMP | BPF_JSET | BPF_X:
 949                         /* A jump must set masks on targets */
 950                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 951                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 952                         memvalid = ~0;
 953                         break;
 954                 }
 955         }
 956 error:
 957         kfree(masks);
 958         return ret;
 959 }
 960
 961 static bool chk_code_allowed(u16 code_to_probe)
 962 {
 963         static const bool codes[] = {
 964                 /* 32 bit ALU operations */
 965                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 966                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 967                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 968                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 969                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 970                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 971                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 972                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 973                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 974                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 975                 [BPF_ALU | BPF_AND | BPF_K] = true,
 976                 [BPF_ALU | BPF_AND | BPF_X] = true,
 977                 [BPF_ALU | BPF_OR | BPF_K] = true,
 978                 [BPF_ALU | BPF_OR | BPF_X] = true,
 979                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 980                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 981                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 982                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 983                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 984                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 985                 [BPF_ALU | BPF_NEG] = true,
 986                 /* Load instructions */
 987                 [BPF_LD | BPF_W | BPF_ABS] = true,
 988                 [BPF_LD | BPF_H | BPF_ABS] = true,
 989                 [BPF_LD | BPF_B | BPF_ABS] = true,
 990                 [BPF_LD | BPF_W | BPF_LEN] = true,
 991                 [BPF_LD | BPF_W | BPF_IND] = true,
 992                 [BPF_LD | BPF_H | BPF_IND] = true,
 993                 [BPF_LD | BPF_B | BPF_IND] = true,
 994                 [BPF_LD | BPF_IMM] = true,
 995                 [BPF_LD | BPF_MEM] = true,
 996                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 997                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 998                 [BPF_LDX | BPF_IMM] = true,
 999                 [BPF_LDX | BPF_MEM] = true,
1000                 /* Store instructions */
1001                 [BPF_ST] = true,
1002                 [BPF_STX] = true,
1003                 /* Misc instructions */
1004                 [BPF_MISC | BPF_TAX] = true,
1005                 [BPF_MISC | BPF_TXA] = true,
1006                 /* Return instructions */
1007                 [BPF_RET | BPF_K] = true,
1008                 [BPF_RET | BPF_A] = true,
1009                 /* Jump instructions */
1010                 [BPF_JMP | BPF_JA] = true,
1011                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1012                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1013                 [BPF_JMP | BPF_JGE | BPF_K] = true,
1014                 [BPF_JMP | BPF_JGE | BPF_X] = true,
1015                 [BPF_JMP | BPF_JGT | BPF_K] = true,
1016                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1017                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1018                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1019         };
1020
1021         if (code_to_probe >= ARRAY_SIZE(codes))
1022                 return false;
1023
1024         return codes[code_to_probe];
1025 }
1026
1027 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1028                                 unsigned int flen)
1029 {
1030         if (filter == NULL)
1031                 return false;
1032         if (flen == 0 || flen > BPF_MAXINSNS)
1033                 return false;
1034
1035         return true;
1036 }
1037
1038 /**
1039  *      bpf_check_classic - verify socket filter code
1040  *      @filter: filter to verify
1041  *      @flen: length of filter
1042  *
1043  * Check the user's filter code. If we let some ugly
1044  * filter code slip through kaboom! The filter must contain
1045  * no references or jumps that are out of range, no illegal
1046  * instructions, and must end with a RET instruction.
1047  *
1048  * All jumps are forward as they are not signed.
1049  *
1050  * Returns 0 if the rule set is legal or -EINVAL if not.
1051  */
1052 static int bpf_check_classic(const struct sock_filter *filter,
1053                              unsigned int flen)
1054 {
1055         bool anc_found;
1056         int pc;
1057
1058         /* Check the filter code now */
1059         for (pc = 0; pc < flen; pc++) {
1060                 const struct sock_filter *ftest = &filter[pc];
1061
1062                 /* May we actually operate on this code? */
1063                 if (!chk_code_allowed(ftest->code))
1064                         return -EINVAL;
1065
1066                 /* Some instructions need special checks */
1067                 switch (ftest->code) {
1068                 case BPF_ALU | BPF_DIV | BPF_K:
1069                 case BPF_ALU | BPF_MOD | BPF_K:
1070                         /* Check for division by zero */
1071                         if (ftest->k == 0)
1072                                 return -EINVAL;
1073                         break;
1074                 case BPF_ALU | BPF_LSH | BPF_K:
1075                 case BPF_ALU | BPF_RSH | BPF_K:
1076                         if (ftest->k >= 32)
1077                                 return -EINVAL;
1078                         break;
1079                 case BPF_LD | BPF_MEM:
1080                 case BPF_LDX | BPF_MEM:
1081                 case BPF_ST:
1082                 case BPF_STX:
1083                         /* Check for invalid memory addresses */
1084                         if (ftest->k >= BPF_MEMWORDS)
1085                                 return -EINVAL;
1086                         break;
1087                 case BPF_JMP | BPF_JA:
1088                         /* Note, the large ftest->k might cause loops.
1089                          * Compare this with conditional jumps below,
1090                          * where offsets are limited. --ANK (981016)
1091                          */
1092                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1093                                 return -EINVAL;
1094                         break;
1095                 case BPF_JMP | BPF_JEQ | BPF_K:
1096                 case BPF_JMP | BPF_JEQ | BPF_X:
1097                 case BPF_JMP | BPF_JGE | BPF_K:
1098                 case BPF_JMP | BPF_JGE | BPF_X:
1099                 case BPF_JMP | BPF_JGT | BPF_K:
1100                 case BPF_JMP | BPF_JGT | BPF_X:
1101                 case BPF_JMP | BPF_JSET | BPF_K:
1102                 case BPF_JMP | BPF_JSET | BPF_X:
1103                         /* Both conditionals must be safe */
1104                         if (pc + ftest->jt + 1 >= flen ||
1105                             pc + ftest->jf + 1 >= flen)
1106                                 return -EINVAL;
1107                         break;
1108                 case BPF_LD | BPF_W | BPF_ABS:
1109                 case BPF_LD | BPF_H | BPF_ABS:
1110                 case BPF_LD | BPF_B | BPF_ABS:
1111                         anc_found = false;
1112                         if (bpf_anc_helper(ftest) & BPF_ANC)
1113                                 anc_found = true;
1114                         /* Ancillary operation unknown or unsupported */
1115                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1116                                 return -EINVAL;
1117                 }
1118         }
1119
1120         /* Last instruction must be a RET code */
1121         switch (filter[flen - 1].code) {
1122         case BPF_RET | BPF_K:
1123         case BPF_RET | BPF_A:
1124                 return check_load_and_stores(filter, flen);
1125         }
1126
1127         return -EINVAL;
1128 }
1129
1130 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1131                                       const struct sock_fprog *fprog)
1132 {
1133         unsigned int fsize = bpf_classic_proglen(fprog);
1134         struct sock_fprog_kern *fkprog;
1135
1136         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1137         if (!fp->orig_prog)
1138                 return -ENOMEM;
1139
1140         fkprog = fp->orig_prog;
1141         fkprog->len = fprog->len;
1142
1143         fkprog->filter = kmemdup(fp->insns, fsize,
1144                                  GFP_KERNEL | __GFP_NOWARN);
1145         if (!fkprog->filter) {
1146                 kfree(fp->orig_prog);
1147                 return -ENOMEM;
1148         }
1149
1150         return 0;
1151 }
1152
1153 static void bpf_release_orig_filter(struct bpf_prog *fp)
1154 {
1155         struct sock_fprog_kern *fprog = fp->orig_prog;
1156
1157         if (fprog) {
1158                 kfree(fprog->filter);
1159                 kfree(fprog);
1160         }
1161 }
1162
1163 static void __bpf_prog_release(struct bpf_prog *prog)
1164 {
1165         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1166                 bpf_prog_put(prog);
1167         } else {
1168                 bpf_release_orig_filter(prog);
1169                 bpf_prog_free(prog);
1170         }
1171 }
1172
1173 static void __sk_filter_release(struct sk_filter *fp)
1174 {
1175         __bpf_prog_release(fp->prog);
1176         kfree(fp);
1177 }
1178
1179 /**
1180  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1181  *      @rcu: rcu_head that contains the sk_filter to free
1182  */
1183 static void sk_filter_release_rcu(struct rcu_head *rcu)
1184 {
1185         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1186
1187         __sk_filter_release(fp);
1188 }
1189
1190 /**
1191  *      sk_filter_release - release a socket filter
1192  *      @fp: filter to remove
1193  *
1194  *      Remove a filter from a socket and release its resources.
1195  */
1196 static void sk_filter_release(struct sk_filter *fp)
1197 {
1198         if (refcount_dec_and_test(&fp->refcnt))
1199                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1200 }
1201
1202 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1203 {
1204         u32 filter_size = bpf_prog_size(fp->prog->len);
1205
1206         atomic_sub(filter_size, &sk->sk_omem_alloc);
1207         sk_filter_release(fp);
1208 }
1209
1210 /* try to charge the socket memory if there is space available
1211  * return true on success
1212  */
1213 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1214 {
1215         u32 filter_size = bpf_prog_size(fp->prog->len);
1216         int optmem_max = READ_ONCE(sysctl_optmem_max);
1217
1218         /* same check as in sock_kmalloc() */
1219         if (filter_size <= optmem_max &&
1220             atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1221                 atomic_add(filter_size, &sk->sk_omem_alloc);
1222                 return true;
1223         }
1224         return false;
1225 }
1226
1227 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1228 {
1229         if (!refcount_inc_not_zero(&fp->refcnt))
1230                 return false;
1231
1232         if (!__sk_filter_charge(sk, fp)) {
1233                 sk_filter_release(fp);
1234                 return false;
1235         }
1236         return true;
1237 }
1238
1239 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1240 {
1241         struct sock_filter *old_prog;
1242         struct bpf_prog *old_fp;
1243         int err, new_len, old_len = fp->len;
1244         bool seen_ld_abs = false;
1245
1246         /* We are free to overwrite insns et al right here as it
1247          * won't be used at this point in time anymore internally
1248          * after the migration to the internal BPF instruction
1249          * representation.
1250          */
1251         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1252                      sizeof(struct bpf_insn));
1253
1254         /* Conversion cannot happen on overlapping memory areas,
1255          * so we need to keep the user BPF around until the 2nd
1256          * pass. At this time, the user BPF is stored in fp->insns.
1257          */
1258         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1259                            GFP_KERNEL | __GFP_NOWARN);
1260         if (!old_prog) {
1261                 err = -ENOMEM;
1262                 goto out_err;
1263         }
1264
1265         /* 1st pass: calculate the new program length. */
1266         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1267                                  &seen_ld_abs);
1268         if (err)
1269                 goto out_err_free;
1270
1271         /* Expand fp for appending the new filter representation. */
1272         old_fp = fp;
1273         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1274         if (!fp) {
1275                 /* The old_fp is still around in case we couldn't
1276                  * allocate new memory, so uncharge on that one.
1277                  */
1278                 fp = old_fp;
1279                 err = -ENOMEM;
1280                 goto out_err_free;
1281         }
1282
1283         fp->len = new_len;
1284
1285         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1286         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1287                                  &seen_ld_abs);
1288         if (err)
1289                 /* 2nd bpf_convert_filter() can fail only if it fails
1290                  * to allocate memory, remapping must succeed. Note,
1291                  * that at this time old_fp has already been released
1292                  * by krealloc().
1293                  */
1294                 goto out_err_free;
1295
1296         fp = bpf_prog_select_runtime(fp, &err);
1297         if (err)
1298                 goto out_err_free;
1299
1300         kfree(old_prog);
1301         return fp;
1302
1303 out_err_free:
1304         kfree(old_prog);
1305 out_err:
1306         __bpf_prog_release(fp);
1307         return ERR_PTR(err);
1308 }
1309
1310 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1311                                            bpf_aux_classic_check_t trans)
1312 {
1313         int err;
1314
1315         fp->bpf_func = NULL;
1316         fp->jited = 0;
1317
1318         err = bpf_check_classic(fp->insns, fp->len);
1319         if (err) {
1320                 __bpf_prog_release(fp);
1321                 return ERR_PTR(err);
1322         }
1323
1324         /* There might be additional checks and transformations
1325          * needed on classic filters, f.e. in case of seccomp.
1326          */
1327         if (trans) {
1328                 err = trans(fp->insns, fp->len);
1329                 if (err) {
1330                         __bpf_prog_release(fp);
1331                         return ERR_PTR(err);
1332                 }
1333         }
1334
1335         /* Probe if we can JIT compile the filter and if so, do
1336          * the compilation of the filter.
1337          */
1338         bpf_jit_compile(fp);
1339
1340         /* JIT compiler couldn't process this filter, so do the
1341          * internal BPF translation for the optimized interpreter.
1342          */
1343         if (!fp->jited)
1344                 fp = bpf_migrate_filter(fp);
1345
1346         return fp;
1347 }
1348
1349 /**
1350  *      bpf_prog_create - create an unattached filter
1351  *      @pfp: the unattached filter that is created
1352  *      @fprog: the filter program
1353  *
1354  * Create a filter independent of any socket. We first run some
1355  * sanity checks on it to make sure it does not explode on us later.
1356  * If an error occurs or there is insufficient memory for the filter
1357  * a negative errno code is returned. On success the return is zero.
1358  */
1359 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1360 {
1361         unsigned int fsize = bpf_classic_proglen(fprog);
1362         struct bpf_prog *fp;
1363
1364         /* Make sure new filter is there and in the right amounts. */
1365         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1366                 return -EINVAL;
1367
1368         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1369         if (!fp)
1370                 return -ENOMEM;
1371
1372         memcpy(fp->insns, fprog->filter, fsize);
1373
1374         fp->len = fprog->len;
1375         /* Since unattached filters are not copied back to user
1376          * space through sk_get_filter(), we do not need to hold
1377          * a copy here, and can spare us the work.
1378          */
1379         fp->orig_prog = NULL;
1380
1381         /* bpf_prepare_filter() already takes care of freeing
1382          * memory in case something goes wrong.
1383          */
1384         fp = bpf_prepare_filter(fp, NULL);
1385         if (IS_ERR(fp))
1386                 return PTR_ERR(fp);
1387
1388         *pfp = fp;
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(bpf_prog_create);
1392
1393 /**
1394  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1395  *      @pfp: the unattached filter that is created
1396  *      @fprog: the filter program
1397  *      @trans: post-classic verifier transformation handler
1398  *      @save_orig: save classic BPF program
1399  *
1400  * This function effectively does the same as bpf_prog_create(), only
1401  * that it builds up its insns buffer from user space provided buffer.
1402  * It also allows for passing a bpf_aux_classic_check_t handler.
1403  */
1404 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1405                               bpf_aux_classic_check_t trans, bool save_orig)
1406 {
1407         unsigned int fsize = bpf_classic_proglen(fprog);
1408         struct bpf_prog *fp;
1409         int err;
1410
1411         /* Make sure new filter is there and in the right amounts. */
1412         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1413                 return -EINVAL;
1414
1415         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1416         if (!fp)
1417                 return -ENOMEM;
1418
1419         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1420                 __bpf_prog_free(fp);
1421                 return -EFAULT;
1422         }
1423
1424         fp->len = fprog->len;
1425         fp->orig_prog = NULL;
1426
1427         if (save_orig) {
1428                 err = bpf_prog_store_orig_filter(fp, fprog);
1429                 if (err) {
1430                         __bpf_prog_free(fp);
1431                         return -ENOMEM;
1432                 }
1433         }
1434
1435         /* bpf_prepare_filter() already takes care of freeing
1436          * memory in case something goes wrong.
1437          */
1438         fp = bpf_prepare_filter(fp, trans);
1439         if (IS_ERR(fp))
1440                 return PTR_ERR(fp);
1441
1442         *pfp = fp;
1443         return 0;
1444 }
1445 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1446
1447 void bpf_prog_destroy(struct bpf_prog *fp)
1448 {
1449         __bpf_prog_release(fp);
1450 }
1451 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1452
1453 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1454 {
1455         struct sk_filter *fp, *old_fp;
1456
1457         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1458         if (!fp)
1459                 return -ENOMEM;
1460
1461         fp->prog = prog;
1462
1463         if (!__sk_filter_charge(sk, fp)) {
1464                 kfree(fp);
1465                 return -ENOMEM;
1466         }
1467         refcount_set(&fp->refcnt, 1);
1468
1469         old_fp = rcu_dereference_protected(sk->sk_filter,
1470                                            lockdep_sock_is_held(sk));
1471         rcu_assign_pointer(sk->sk_filter, fp);
1472
1473         if (old_fp)
1474                 sk_filter_uncharge(sk, old_fp);
1475
1476         return 0;
1477 }
1478
1479 static
1480 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1481 {
1482         unsigned int fsize = bpf_classic_proglen(fprog);
1483         struct bpf_prog *prog;
1484         int err;
1485
1486         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1487                 return ERR_PTR(-EPERM);
1488
1489         /* Make sure new filter is there and in the right amounts. */
1490         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1491                 return ERR_PTR(-EINVAL);
1492
1493         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1494         if (!prog)
1495                 return ERR_PTR(-ENOMEM);
1496
1497         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1498                 __bpf_prog_free(prog);
1499                 return ERR_PTR(-EFAULT);
1500         }
1501
1502         prog->len = fprog->len;
1503
1504         err = bpf_prog_store_orig_filter(prog, fprog);
1505         if (err) {
1506                 __bpf_prog_free(prog);
1507                 return ERR_PTR(-ENOMEM);
1508         }
1509
1510         /* bpf_prepare_filter() already takes care of freeing
1511          * memory in case something goes wrong.
1512          */
1513         return bpf_prepare_filter(prog, NULL);
1514 }
1515
1516 /**
1517  *      sk_attach_filter - attach a socket filter
1518  *      @fprog: the filter program
1519  *      @sk: the socket to use
1520  *
1521  * Attach the user's filter code. We first run some sanity checks on
1522  * it to make sure it does not explode on us later. If an error
1523  * occurs or there is insufficient memory for the filter a negative
1524  * errno code is returned. On success the return is zero.
1525  */
1526 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1527 {
1528         struct bpf_prog *prog = __get_filter(fprog, sk);
1529         int err;
1530
1531         if (IS_ERR(prog))
1532                 return PTR_ERR(prog);
1533
1534         err = __sk_attach_prog(prog, sk);
1535         if (err < 0) {
1536                 __bpf_prog_release(prog);
1537                 return err;
1538         }
1539
1540         return 0;
1541 }
1542 EXPORT_SYMBOL_GPL(sk_attach_filter);
1543
1544 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1545 {
1546         struct bpf_prog *prog = __get_filter(fprog, sk);
1547         int err;
1548
1549         if (IS_ERR(prog))
1550                 return PTR_ERR(prog);
1551
1552         if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1553                 err = -ENOMEM;
1554         else
1555                 err = reuseport_attach_prog(sk, prog);
1556
1557         if (err)
1558                 __bpf_prog_release(prog);
1559
1560         return err;
1561 }
1562
1563 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 {
1565         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1566                 return ERR_PTR(-EPERM);
1567
1568         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 }
1570
1571 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 {
1573         struct bpf_prog *prog = __get_bpf(ufd, sk);
1574         int err;
1575
1576         if (IS_ERR(prog))
1577                 return PTR_ERR(prog);
1578
1579         err = __sk_attach_prog(prog, sk);
1580         if (err < 0) {
1581                 bpf_prog_put(prog);
1582                 return err;
1583         }
1584
1585         return 0;
1586 }
1587
1588 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 {
1590         struct bpf_prog *prog;
1591         int err;
1592
1593         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1594                 return -EPERM;
1595
1596         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1597         if (PTR_ERR(prog) == -EINVAL)
1598                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1599         if (IS_ERR(prog))
1600                 return PTR_ERR(prog);
1601
1602         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1603                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1604                  * bpf prog (e.g. sockmap).  It depends on the
1605                  * limitation imposed by bpf_prog_load().
1606                  * Hence, sysctl_optmem_max is not checked.
1607                  */
1608                 if ((sk->sk_type != SOCK_STREAM &&
1609                      sk->sk_type != SOCK_DGRAM) ||
1610                     (sk->sk_protocol != IPPROTO_UDP &&
1611                      sk->sk_protocol != IPPROTO_TCP) ||
1612                     (sk->sk_family != AF_INET &&
1613                      sk->sk_family != AF_INET6)) {
1614                         err = -ENOTSUPP;
1615                         goto err_prog_put;
1616                 }
1617         } else {
1618                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1619                 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1620                         err = -ENOMEM;
1621                         goto err_prog_put;
1622                 }
1623         }
1624
1625         err = reuseport_attach_prog(sk, prog);
1626 err_prog_put:
1627         if (err)
1628                 bpf_prog_put(prog);
1629
1630         return err;
1631 }
1632
1633 void sk_reuseport_prog_free(struct bpf_prog *prog)
1634 {
1635         if (!prog)
1636                 return;
1637
1638         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1639                 bpf_prog_put(prog);
1640         else
1641                 bpf_prog_destroy(prog);
1642 }
1643
1644 struct bpf_scratchpad {
1645         union {
1646                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1647                 u8     buff[MAX_BPF_STACK];
1648         };
1649 };
1650
1651 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1652
1653 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1654                                           unsigned int write_len)
1655 {
1656         return skb_ensure_writable(skb, write_len);
1657 }
1658
1659 static inline int bpf_try_make_writable(struct sk_buff *skb,
1660                                         unsigned int write_len)
1661 {
1662         int err = __bpf_try_make_writable(skb, write_len);
1663
1664         bpf_compute_data_pointers(skb);
1665         return err;
1666 }
1667
1668 static int bpf_try_make_head_writable(struct sk_buff *skb)
1669 {
1670         return bpf_try_make_writable(skb, skb_headlen(skb));
1671 }
1672
1673 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1674 {
1675         if (skb_at_tc_ingress(skb))
1676                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1677 }
1678
1679 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1680 {
1681         if (skb_at_tc_ingress(skb))
1682                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1683 }
1684
1685 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1686            const void *, from, u32, len, u64, flags)
1687 {
1688         void *ptr;
1689
1690         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1691                 return -EINVAL;
1692         if (unlikely(offset > INT_MAX))
1693                 return -EFAULT;
1694         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1695                 return -EFAULT;
1696
1697         ptr = skb->data + offset;
1698         if (flags & BPF_F_RECOMPUTE_CSUM)
1699                 __skb_postpull_rcsum(skb, ptr, len, offset);
1700
1701         memcpy(ptr, from, len);
1702
1703         if (flags & BPF_F_RECOMPUTE_CSUM)
1704                 __skb_postpush_rcsum(skb, ptr, len, offset);
1705         if (flags & BPF_F_INVALIDATE_HASH)
1706                 skb_clear_hash(skb);
1707
1708         return 0;
1709 }
1710
1711 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1712         .func           = bpf_skb_store_bytes,
1713         .gpl_only       = false,
1714         .ret_type       = RET_INTEGER,
1715         .arg1_type      = ARG_PTR_TO_CTX,
1716         .arg2_type      = ARG_ANYTHING,
1717         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1718         .arg4_type      = ARG_CONST_SIZE,
1719         .arg5_type      = ARG_ANYTHING,
1720 };
1721
1722 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1723            void *, to, u32, len)
1724 {
1725         void *ptr;
1726
1727         if (unlikely(offset > INT_MAX))
1728                 goto err_clear;
1729
1730         ptr = skb_header_pointer(skb, offset, len, to);
1731         if (unlikely(!ptr))
1732                 goto err_clear;
1733         if (ptr != to)
1734                 memcpy(to, ptr, len);
1735
1736         return 0;
1737 err_clear:
1738         memset(to, 0, len);
1739         return -EFAULT;
1740 }
1741
1742 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1743         .func           = bpf_skb_load_bytes,
1744         .gpl_only       = false,
1745         .ret_type       = RET_INTEGER,
1746         .arg1_type      = ARG_PTR_TO_CTX,
1747         .arg2_type      = ARG_ANYTHING,
1748         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1749         .arg4_type      = ARG_CONST_SIZE,
1750 };
1751
1752 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1753            const struct bpf_flow_dissector *, ctx, u32, offset,
1754            void *, to, u32, len)
1755 {
1756         void *ptr;
1757
1758         if (unlikely(offset > 0xffff))
1759                 goto err_clear;
1760
1761         if (unlikely(!ctx->skb))
1762                 goto err_clear;
1763
1764         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1765         if (unlikely(!ptr))
1766                 goto err_clear;
1767         if (ptr != to)
1768                 memcpy(to, ptr, len);
1769
1770         return 0;
1771 err_clear:
1772         memset(to, 0, len);
1773         return -EFAULT;
1774 }
1775
1776 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1777         .func           = bpf_flow_dissector_load_bytes,
1778         .gpl_only       = false,
1779         .ret_type       = RET_INTEGER,
1780         .arg1_type      = ARG_PTR_TO_CTX,
1781         .arg2_type      = ARG_ANYTHING,
1782         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1783         .arg4_type      = ARG_CONST_SIZE,
1784 };
1785
1786 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1787            u32, offset, void *, to, u32, len, u32, start_header)
1788 {
1789         u8 *end = skb_tail_pointer(skb);
1790         u8 *start, *ptr;
1791
1792         if (unlikely(offset > 0xffff))
1793                 goto err_clear;
1794
1795         switch (start_header) {
1796         case BPF_HDR_START_MAC:
1797                 if (unlikely(!skb_mac_header_was_set(skb)))
1798                         goto err_clear;
1799                 start = skb_mac_header(skb);
1800                 break;
1801         case BPF_HDR_START_NET:
1802                 start = skb_network_header(skb);
1803                 break;
1804         default:
1805                 goto err_clear;
1806         }
1807
1808         ptr = start + offset;
1809
1810         if (likely(ptr + len <= end)) {
1811                 memcpy(to, ptr, len);
1812                 return 0;
1813         }
1814
1815 err_clear:
1816         memset(to, 0, len);
1817         return -EFAULT;
1818 }
1819
1820 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1821         .func           = bpf_skb_load_bytes_relative,
1822         .gpl_only       = false,
1823         .ret_type       = RET_INTEGER,
1824         .arg1_type      = ARG_PTR_TO_CTX,
1825         .arg2_type      = ARG_ANYTHING,
1826         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1827         .arg4_type      = ARG_CONST_SIZE,
1828         .arg5_type      = ARG_ANYTHING,
1829 };
1830
1831 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1832 {
1833         /* Idea is the following: should the needed direct read/write
1834          * test fail during runtime, we can pull in more data and redo
1835          * again, since implicitly, we invalidate previous checks here.
1836          *
1837          * Or, since we know how much we need to make read/writeable,
1838          * this can be done once at the program beginning for direct
1839          * access case. By this we overcome limitations of only current
1840          * headroom being accessible.
1841          */
1842         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1843 }
1844
1845 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1846         .func           = bpf_skb_pull_data,
1847         .gpl_only       = false,
1848         .ret_type       = RET_INTEGER,
1849         .arg1_type      = ARG_PTR_TO_CTX,
1850         .arg2_type      = ARG_ANYTHING,
1851 };
1852
1853 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1854 {
1855         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1856 }
1857
1858 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1859         .func           = bpf_sk_fullsock,
1860         .gpl_only       = false,
1861         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1862         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1863 };
1864
1865 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1866                                            unsigned int write_len)
1867 {
1868         return __bpf_try_make_writable(skb, write_len);
1869 }
1870
1871 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1872 {
1873         /* Idea is the following: should the needed direct read/write
1874          * test fail during runtime, we can pull in more data and redo
1875          * again, since implicitly, we invalidate previous checks here.
1876          *
1877          * Or, since we know how much we need to make read/writeable,
1878          * this can be done once at the program beginning for direct
1879          * access case. By this we overcome limitations of only current
1880          * headroom being accessible.
1881          */
1882         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1883 }
1884
1885 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1886         .func           = sk_skb_pull_data,
1887         .gpl_only       = false,
1888         .ret_type       = RET_INTEGER,
1889         .arg1_type      = ARG_PTR_TO_CTX,
1890         .arg2_type      = ARG_ANYTHING,
1891 };
1892
1893 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1894            u64, from, u64, to, u64, flags)
1895 {
1896         __sum16 *ptr;
1897
1898         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1899                 return -EINVAL;
1900         if (unlikely(offset > 0xffff || offset & 1))
1901                 return -EFAULT;
1902         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1903                 return -EFAULT;
1904
1905         ptr = (__sum16 *)(skb->data + offset);
1906         switch (flags & BPF_F_HDR_FIELD_MASK) {
1907         case 0:
1908                 if (unlikely(from != 0))
1909                         return -EINVAL;
1910
1911                 csum_replace_by_diff(ptr, to);
1912                 break;
1913         case 2:
1914                 csum_replace2(ptr, from, to);
1915                 break;
1916         case 4:
1917                 csum_replace4(ptr, from, to);
1918                 break;
1919         default:
1920                 return -EINVAL;
1921         }
1922
1923         return 0;
1924 }
1925
1926 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1927         .func           = bpf_l3_csum_replace,
1928         .gpl_only       = false,
1929         .ret_type       = RET_INTEGER,
1930         .arg1_type      = ARG_PTR_TO_CTX,
1931         .arg2_type      = ARG_ANYTHING,
1932         .arg3_type      = ARG_ANYTHING,
1933         .arg4_type      = ARG_ANYTHING,
1934         .arg5_type      = ARG_ANYTHING,
1935 };
1936
1937 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1938            u64, from, u64, to, u64, flags)
1939 {
1940         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1941         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1942         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1943         __sum16 *ptr;
1944
1945         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1946                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1947                 return -EINVAL;
1948         if (unlikely(offset > 0xffff || offset & 1))
1949                 return -EFAULT;
1950         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1951                 return -EFAULT;
1952
1953         ptr = (__sum16 *)(skb->data + offset);
1954         if (is_mmzero && !do_mforce && !*ptr)
1955                 return 0;
1956
1957         switch (flags & BPF_F_HDR_FIELD_MASK) {
1958         case 0:
1959                 if (unlikely(from != 0))
1960                         return -EINVAL;
1961
1962                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1963                 break;
1964         case 2:
1965                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1966                 break;
1967         case 4:
1968                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1969                 break;
1970         default:
1971                 return -EINVAL;
1972         }
1973
1974         if (is_mmzero && !*ptr)
1975                 *ptr = CSUM_MANGLED_0;
1976         return 0;
1977 }
1978
1979 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1980         .func           = bpf_l4_csum_replace,
1981         .gpl_only       = false,
1982         .ret_type       = RET_INTEGER,
1983         .arg1_type      = ARG_PTR_TO_CTX,
1984         .arg2_type      = ARG_ANYTHING,
1985         .arg3_type      = ARG_ANYTHING,
1986         .arg4_type      = ARG_ANYTHING,
1987         .arg5_type      = ARG_ANYTHING,
1988 };
1989
1990 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1991            __be32 *, to, u32, to_size, __wsum, seed)
1992 {
1993         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1994         u32 diff_size = from_size + to_size;
1995         int i, j = 0;
1996
1997         /* This is quite flexible, some examples:
1998          *
1999          * from_size == 0, to_size > 0,  seed := csum --> pushing data
2000          * from_size > 0,  to_size == 0, seed := csum --> pulling data
2001          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2002          *
2003          * Even for diffing, from_size and to_size don't need to be equal.
2004          */
2005         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2006                      diff_size > sizeof(sp->diff)))
2007                 return -EINVAL;
2008
2009         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2010                 sp->diff[j] = ~from[i];
2011         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2012                 sp->diff[j] = to[i];
2013
2014         return csum_partial(sp->diff, diff_size, seed);
2015 }
2016
2017 static const struct bpf_func_proto bpf_csum_diff_proto = {
2018         .func           = bpf_csum_diff,
2019         .gpl_only       = false,
2020         .pkt_access     = true,
2021         .ret_type       = RET_INTEGER,
2022         .arg1_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2023         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2024         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2026         .arg5_type      = ARG_ANYTHING,
2027 };
2028
2029 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2030 {
2031         /* The interface is to be used in combination with bpf_csum_diff()
2032          * for direct packet writes. csum rotation for alignment as well
2033          * as emulating csum_sub() can be done from the eBPF program.
2034          */
2035         if (skb->ip_summed == CHECKSUM_COMPLETE)
2036                 return (skb->csum = csum_add(skb->csum, csum));
2037
2038         return -ENOTSUPP;
2039 }
2040
2041 static const struct bpf_func_proto bpf_csum_update_proto = {
2042         .func           = bpf_csum_update,
2043         .gpl_only       = false,
2044         .ret_type       = RET_INTEGER,
2045         .arg1_type      = ARG_PTR_TO_CTX,
2046         .arg2_type      = ARG_ANYTHING,
2047 };
2048
2049 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2050 {
2051         /* The interface is to be used in combination with bpf_skb_adjust_room()
2052          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2053          * is passed as flags, for example.
2054          */
2055         switch (level) {
2056         case BPF_CSUM_LEVEL_INC:
2057                 __skb_incr_checksum_unnecessary(skb);
2058                 break;
2059         case BPF_CSUM_LEVEL_DEC:
2060                 __skb_decr_checksum_unnecessary(skb);
2061                 break;
2062         case BPF_CSUM_LEVEL_RESET:
2063                 __skb_reset_checksum_unnecessary(skb);
2064                 break;
2065         case BPF_CSUM_LEVEL_QUERY:
2066                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2067                        skb->csum_level : -EACCES;
2068         default:
2069                 return -EINVAL;
2070         }
2071
2072         return 0;
2073 }
2074
2075 static const struct bpf_func_proto bpf_csum_level_proto = {
2076         .func           = bpf_csum_level,
2077         .gpl_only       = false,
2078         .ret_type       = RET_INTEGER,
2079         .arg1_type      = ARG_PTR_TO_CTX,
2080         .arg2_type      = ARG_ANYTHING,
2081 };
2082
2083 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2084 {
2085         return dev_forward_skb_nomtu(dev, skb);
2086 }
2087
2088 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2089                                       struct sk_buff *skb)
2090 {
2091         int ret = ____dev_forward_skb(dev, skb, false);
2092
2093         if (likely(!ret)) {
2094                 skb->dev = dev;
2095                 ret = netif_rx(skb);
2096         }
2097
2098         return ret;
2099 }
2100
2101 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2102 {
2103         int ret;
2104
2105         if (dev_xmit_recursion()) {
2106                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2107                 kfree_skb(skb);
2108                 return -ENETDOWN;
2109         }
2110
2111         skb->dev = dev;
2112         skb->tstamp = 0;
2113
2114         dev_xmit_recursion_inc();
2115         ret = dev_queue_xmit(skb);
2116         dev_xmit_recursion_dec();
2117
2118         return ret;
2119 }
2120
2121 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2122                                  u32 flags)
2123 {
2124         unsigned int mlen = skb_network_offset(skb);
2125
2126         if (unlikely(skb->len <= mlen)) {
2127                 kfree_skb(skb);
2128                 return -ERANGE;
2129         }
2130
2131         if (mlen) {
2132                 __skb_pull(skb, mlen);
2133                 if (unlikely(!skb->len)) {
2134                         kfree_skb(skb);
2135                         return -ERANGE;
2136                 }
2137
2138                 /* At ingress, the mac header has already been pulled once.
2139                  * At egress, skb_pospull_rcsum has to be done in case that
2140                  * the skb is originated from ingress (i.e. a forwarded skb)
2141                  * to ensure that rcsum starts at net header.
2142                  */
2143                 if (!skb_at_tc_ingress(skb))
2144                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2145         }
2146         skb_pop_mac_header(skb);
2147         skb_reset_mac_len(skb);
2148         return flags & BPF_F_INGRESS ?
2149                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2150 }
2151
2152 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2153                                  u32 flags)
2154 {
2155         /* Verify that a link layer header is carried */
2156         if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2157                 kfree_skb(skb);
2158                 return -ERANGE;
2159         }
2160
2161         bpf_push_mac_rcsum(skb);
2162         return flags & BPF_F_INGRESS ?
2163                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2164 }
2165
2166 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2167                           u32 flags)
2168 {
2169         if (dev_is_mac_header_xmit(dev))
2170                 return __bpf_redirect_common(skb, dev, flags);
2171         else
2172                 return __bpf_redirect_no_mac(skb, dev, flags);
2173 }
2174
2175 #if IS_ENABLED(CONFIG_IPV6)
2176 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2177                             struct net_device *dev, struct bpf_nh_params *nh)
2178 {
2179         u32 hh_len = LL_RESERVED_SPACE(dev);
2180         const struct in6_addr *nexthop;
2181         struct dst_entry *dst = NULL;
2182         struct neighbour *neigh;
2183
2184         if (dev_xmit_recursion()) {
2185                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2186                 goto out_drop;
2187         }
2188
2189         skb->dev = dev;
2190         skb->tstamp = 0;
2191
2192         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2193                 skb = skb_expand_head(skb, hh_len);
2194                 if (!skb)
2195                         return -ENOMEM;
2196         }
2197
2198         rcu_read_lock_bh();
2199         if (!nh) {
2200                 dst = skb_dst(skb);
2201                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2202                                       &ipv6_hdr(skb)->daddr);
2203         } else {
2204                 nexthop = &nh->ipv6_nh;
2205         }
2206         neigh = ip_neigh_gw6(dev, nexthop);
2207         if (likely(!IS_ERR(neigh))) {
2208                 int ret;
2209
2210                 sock_confirm_neigh(skb, neigh);
2211                 dev_xmit_recursion_inc();
2212                 ret = neigh_output(neigh, skb, false);
2213                 dev_xmit_recursion_dec();
2214                 rcu_read_unlock_bh();
2215                 return ret;
2216         }
2217         rcu_read_unlock_bh();
2218         if (dst)
2219                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2220 out_drop:
2221         kfree_skb(skb);
2222         return -ENETDOWN;
2223 }
2224
2225 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2226                                    struct bpf_nh_params *nh)
2227 {
2228         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2229         struct net *net = dev_net(dev);
2230         int err, ret = NET_XMIT_DROP;
2231
2232         if (!nh) {
2233                 struct dst_entry *dst;
2234                 struct flowi6 fl6 = {
2235                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2236                         .flowi6_mark  = skb->mark,
2237                         .flowlabel    = ip6_flowinfo(ip6h),
2238                         .flowi6_oif   = dev->ifindex,
2239                         .flowi6_proto = ip6h->nexthdr,
2240                         .daddr        = ip6h->daddr,
2241                         .saddr        = ip6h->saddr,
2242                 };
2243
2244                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2245                 if (IS_ERR(dst))
2246                         goto out_drop;
2247
2248                 skb_dst_set(skb, dst);
2249         } else if (nh->nh_family != AF_INET6) {
2250                 goto out_drop;
2251         }
2252
2253         err = bpf_out_neigh_v6(net, skb, dev, nh);
2254         if (unlikely(net_xmit_eval(err)))
2255                 dev->stats.tx_errors++;
2256         else
2257                 ret = NET_XMIT_SUCCESS;
2258         goto out_xmit;
2259 out_drop:
2260         dev->stats.tx_errors++;
2261         kfree_skb(skb);
2262 out_xmit:
2263         return ret;
2264 }
2265 #else
2266 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2267                                    struct bpf_nh_params *nh)
2268 {
2269         kfree_skb(skb);
2270         return NET_XMIT_DROP;
2271 }
2272 #endif /* CONFIG_IPV6 */
2273
2274 #if IS_ENABLED(CONFIG_INET)
2275 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2276                             struct net_device *dev, struct bpf_nh_params *nh)
2277 {
2278         u32 hh_len = LL_RESERVED_SPACE(dev);
2279         struct neighbour *neigh;
2280         bool is_v6gw = false;
2281
2282         if (dev_xmit_recursion()) {
2283                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2284                 goto out_drop;
2285         }
2286
2287         skb->dev = dev;
2288         skb->tstamp = 0;
2289
2290         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2291                 skb = skb_expand_head(skb, hh_len);
2292                 if (!skb)
2293                         return -ENOMEM;
2294         }
2295
2296         rcu_read_lock_bh();
2297         if (!nh) {
2298                 struct dst_entry *dst = skb_dst(skb);
2299                 struct rtable *rt = container_of(dst, struct rtable, dst);
2300
2301                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2302         } else if (nh->nh_family == AF_INET6) {
2303                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2304                 is_v6gw = true;
2305         } else if (nh->nh_family == AF_INET) {
2306                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2307         } else {
2308                 rcu_read_unlock_bh();
2309                 goto out_drop;
2310         }
2311
2312         if (likely(!IS_ERR(neigh))) {
2313                 int ret;
2314
2315                 sock_confirm_neigh(skb, neigh);
2316                 dev_xmit_recursion_inc();
2317                 ret = neigh_output(neigh, skb, is_v6gw);
2318                 dev_xmit_recursion_dec();
2319                 rcu_read_unlock_bh();
2320                 return ret;
2321         }
2322         rcu_read_unlock_bh();
2323 out_drop:
2324         kfree_skb(skb);
2325         return -ENETDOWN;
2326 }
2327
2328 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2329                                    struct bpf_nh_params *nh)
2330 {
2331         const struct iphdr *ip4h = ip_hdr(skb);
2332         struct net *net = dev_net(dev);
2333         int err, ret = NET_XMIT_DROP;
2334
2335         if (!nh) {
2336                 struct flowi4 fl4 = {
2337                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2338                         .flowi4_mark  = skb->mark,
2339                         .flowi4_tos   = RT_TOS(ip4h->tos),
2340                         .flowi4_oif   = dev->ifindex,
2341                         .flowi4_proto = ip4h->protocol,
2342                         .daddr        = ip4h->daddr,
2343                         .saddr        = ip4h->saddr,
2344                 };
2345                 struct rtable *rt;
2346
2347                 rt = ip_route_output_flow(net, &fl4, NULL);
2348                 if (IS_ERR(rt))
2349                         goto out_drop;
2350                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2351                         ip_rt_put(rt);
2352                         goto out_drop;
2353                 }
2354
2355                 skb_dst_set(skb, &rt->dst);
2356         }
2357
2358         err = bpf_out_neigh_v4(net, skb, dev, nh);
2359         if (unlikely(net_xmit_eval(err)))
2360                 dev->stats.tx_errors++;
2361         else
2362                 ret = NET_XMIT_SUCCESS;
2363         goto out_xmit;
2364 out_drop:
2365         dev->stats.tx_errors++;
2366         kfree_skb(skb);
2367 out_xmit:
2368         return ret;
2369 }
2370 #else
2371 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2372                                    struct bpf_nh_params *nh)
2373 {
2374         kfree_skb(skb);
2375         return NET_XMIT_DROP;
2376 }
2377 #endif /* CONFIG_INET */
2378
2379 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2380                                 struct bpf_nh_params *nh)
2381 {
2382         struct ethhdr *ethh = eth_hdr(skb);
2383
2384         if (unlikely(skb->mac_header >= skb->network_header))
2385                 goto out;
2386         bpf_push_mac_rcsum(skb);
2387         if (is_multicast_ether_addr(ethh->h_dest))
2388                 goto out;
2389
2390         skb_pull(skb, sizeof(*ethh));
2391         skb_unset_mac_header(skb);
2392         skb_reset_network_header(skb);
2393
2394         if (skb->protocol == htons(ETH_P_IP))
2395                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2396         else if (skb->protocol == htons(ETH_P_IPV6))
2397                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2398 out:
2399         kfree_skb(skb);
2400         return -ENOTSUPP;
2401 }
2402
2403 /* Internal, non-exposed redirect flags. */
2404 enum {
2405         BPF_F_NEIGH     = (1ULL << 1),
2406         BPF_F_PEER      = (1ULL << 2),
2407         BPF_F_NEXTHOP   = (1ULL << 3),
2408 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2409 };
2410
2411 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2412 {
2413         struct net_device *dev;
2414         struct sk_buff *clone;
2415         int ret;
2416
2417         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2418                 return -EINVAL;
2419
2420         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2421         if (unlikely(!dev))
2422                 return -EINVAL;
2423
2424         clone = skb_clone(skb, GFP_ATOMIC);
2425         if (unlikely(!clone))
2426                 return -ENOMEM;
2427
2428         /* For direct write, we need to keep the invariant that the skbs
2429          * we're dealing with need to be uncloned. Should uncloning fail
2430          * here, we need to free the just generated clone to unclone once
2431          * again.
2432          */
2433         ret = bpf_try_make_head_writable(skb);
2434         if (unlikely(ret)) {
2435                 kfree_skb(clone);
2436                 return -ENOMEM;
2437         }
2438
2439         return __bpf_redirect(clone, dev, flags);
2440 }
2441
2442 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2443         .func           = bpf_clone_redirect,
2444         .gpl_only       = false,
2445         .ret_type       = RET_INTEGER,
2446         .arg1_type      = ARG_PTR_TO_CTX,
2447         .arg2_type      = ARG_ANYTHING,
2448         .arg3_type      = ARG_ANYTHING,
2449 };
2450
2451 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2452 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2453
2454 int skb_do_redirect(struct sk_buff *skb)
2455 {
2456         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2457         struct net *net = dev_net(skb->dev);
2458         struct net_device *dev;
2459         u32 flags = ri->flags;
2460
2461         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2462         ri->tgt_index = 0;
2463         ri->flags = 0;
2464         if (unlikely(!dev))
2465                 goto out_drop;
2466         if (flags & BPF_F_PEER) {
2467                 const struct net_device_ops *ops = dev->netdev_ops;
2468
2469                 if (unlikely(!ops->ndo_get_peer_dev ||
2470                              !skb_at_tc_ingress(skb)))
2471                         goto out_drop;
2472                 dev = ops->ndo_get_peer_dev(dev);
2473                 if (unlikely(!dev ||
2474                              !(dev->flags & IFF_UP) ||
2475                              net_eq(net, dev_net(dev))))
2476                         goto out_drop;
2477                 skb->dev = dev;
2478                 return -EAGAIN;
2479         }
2480         return flags & BPF_F_NEIGH ?
2481                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2482                                     &ri->nh : NULL) :
2483                __bpf_redirect(skb, dev, flags);
2484 out_drop:
2485         kfree_skb(skb);
2486         return -EINVAL;
2487 }
2488
2489 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2490 {
2491         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2492
2493         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2494                 return TC_ACT_SHOT;
2495
2496         ri->flags = flags;
2497         ri->tgt_index = ifindex;
2498
2499         return TC_ACT_REDIRECT;
2500 }
2501
2502 static const struct bpf_func_proto bpf_redirect_proto = {
2503         .func           = bpf_redirect,
2504         .gpl_only       = false,
2505         .ret_type       = RET_INTEGER,
2506         .arg1_type      = ARG_ANYTHING,
2507         .arg2_type      = ARG_ANYTHING,
2508 };
2509
2510 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2511 {
2512         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2513
2514         if (unlikely(flags))
2515                 return TC_ACT_SHOT;
2516
2517         ri->flags = BPF_F_PEER;
2518         ri->tgt_index = ifindex;
2519
2520         return TC_ACT_REDIRECT;
2521 }
2522
2523 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2524         .func           = bpf_redirect_peer,
2525         .gpl_only       = false,
2526         .ret_type       = RET_INTEGER,
2527         .arg1_type      = ARG_ANYTHING,
2528         .arg2_type      = ARG_ANYTHING,
2529 };
2530
2531 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2532            int, plen, u64, flags)
2533 {
2534         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2535
2536         if (unlikely((plen && plen < sizeof(*params)) || flags))
2537                 return TC_ACT_SHOT;
2538
2539         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2540         ri->tgt_index = ifindex;
2541
2542         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2543         if (plen)
2544                 memcpy(&ri->nh, params, sizeof(ri->nh));
2545
2546         return TC_ACT_REDIRECT;
2547 }
2548
2549 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2550         .func           = bpf_redirect_neigh,
2551         .gpl_only       = false,
2552         .ret_type       = RET_INTEGER,
2553         .arg1_type      = ARG_ANYTHING,
2554         .arg2_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2555         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2556         .arg4_type      = ARG_ANYTHING,
2557 };
2558
2559 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2560 {
2561         msg->apply_bytes = bytes;
2562         return 0;
2563 }
2564
2565 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2566         .func           = bpf_msg_apply_bytes,
2567         .gpl_only       = false,
2568         .ret_type       = RET_INTEGER,
2569         .arg1_type      = ARG_PTR_TO_CTX,
2570         .arg2_type      = ARG_ANYTHING,
2571 };
2572
2573 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2574 {
2575         msg->cork_bytes = bytes;
2576         return 0;
2577 }
2578
2579 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2580         .func           = bpf_msg_cork_bytes,
2581         .gpl_only       = false,
2582         .ret_type       = RET_INTEGER,
2583         .arg1_type      = ARG_PTR_TO_CTX,
2584         .arg2_type      = ARG_ANYTHING,
2585 };
2586
2587 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2588            u32, end, u64, flags)
2589 {
2590         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2591         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2592         struct scatterlist *sge;
2593         u8 *raw, *to, *from;
2594         struct page *page;
2595
2596         if (unlikely(flags || end <= start))
2597                 return -EINVAL;
2598
2599         /* First find the starting scatterlist element */
2600         i = msg->sg.start;
2601         do {
2602                 offset += len;
2603                 len = sk_msg_elem(msg, i)->length;
2604                 if (start < offset + len)
2605                         break;
2606                 sk_msg_iter_var_next(i);
2607         } while (i != msg->sg.end);
2608
2609         if (unlikely(start >= offset + len))
2610                 return -EINVAL;
2611
2612         first_sge = i;
2613         /* The start may point into the sg element so we need to also
2614          * account for the headroom.
2615          */
2616         bytes_sg_total = start - offset + bytes;
2617         if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2618                 goto out;
2619
2620         /* At this point we need to linearize multiple scatterlist
2621          * elements or a single shared page. Either way we need to
2622          * copy into a linear buffer exclusively owned by BPF. Then
2623          * place the buffer in the scatterlist and fixup the original
2624          * entries by removing the entries now in the linear buffer
2625          * and shifting the remaining entries. For now we do not try
2626          * to copy partial entries to avoid complexity of running out
2627          * of sg_entry slots. The downside is reading a single byte
2628          * will copy the entire sg entry.
2629          */
2630         do {
2631                 copy += sk_msg_elem(msg, i)->length;
2632                 sk_msg_iter_var_next(i);
2633                 if (bytes_sg_total <= copy)
2634                         break;
2635         } while (i != msg->sg.end);
2636         last_sge = i;
2637
2638         if (unlikely(bytes_sg_total > copy))
2639                 return -EINVAL;
2640
2641         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2642                            get_order(copy));
2643         if (unlikely(!page))
2644                 return -ENOMEM;
2645
2646         raw = page_address(page);
2647         i = first_sge;
2648         do {
2649                 sge = sk_msg_elem(msg, i);
2650                 from = sg_virt(sge);
2651                 len = sge->length;
2652                 to = raw + poffset;
2653
2654                 memcpy(to, from, len);
2655                 poffset += len;
2656                 sge->length = 0;
2657                 put_page(sg_page(sge));
2658
2659                 sk_msg_iter_var_next(i);
2660         } while (i != last_sge);
2661
2662         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2663
2664         /* To repair sg ring we need to shift entries. If we only
2665          * had a single entry though we can just replace it and
2666          * be done. Otherwise walk the ring and shift the entries.
2667          */
2668         WARN_ON_ONCE(last_sge == first_sge);
2669         shift = last_sge > first_sge ?
2670                 last_sge - first_sge - 1 :
2671                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2672         if (!shift)
2673                 goto out;
2674
2675         i = first_sge;
2676         sk_msg_iter_var_next(i);
2677         do {
2678                 u32 move_from;
2679
2680                 if (i + shift >= NR_MSG_FRAG_IDS)
2681                         move_from = i + shift - NR_MSG_FRAG_IDS;
2682                 else
2683                         move_from = i + shift;
2684                 if (move_from == msg->sg.end)
2685                         break;
2686
2687                 msg->sg.data[i] = msg->sg.data[move_from];
2688                 msg->sg.data[move_from].length = 0;
2689                 msg->sg.data[move_from].page_link = 0;
2690                 msg->sg.data[move_from].offset = 0;
2691                 sk_msg_iter_var_next(i);
2692         } while (1);
2693
2694         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2695                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2696                       msg->sg.end - shift;
2697 out:
2698         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2699         msg->data_end = msg->data + bytes;
2700         return 0;
2701 }
2702
2703 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2704         .func           = bpf_msg_pull_data,
2705         .gpl_only       = false,
2706         .ret_type       = RET_INTEGER,
2707         .arg1_type      = ARG_PTR_TO_CTX,
2708         .arg2_type      = ARG_ANYTHING,
2709         .arg3_type      = ARG_ANYTHING,
2710         .arg4_type      = ARG_ANYTHING,
2711 };
2712
2713 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2714            u32, len, u64, flags)
2715 {
2716         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2717         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2718         u8 *raw, *to, *from;
2719         struct page *page;
2720
2721         if (unlikely(flags))
2722                 return -EINVAL;
2723
2724         if (unlikely(len == 0))
2725                 return 0;
2726
2727         /* First find the starting scatterlist element */
2728         i = msg->sg.start;
2729         do {
2730                 offset += l;
2731                 l = sk_msg_elem(msg, i)->length;
2732
2733                 if (start < offset + l)
2734                         break;
2735                 sk_msg_iter_var_next(i);
2736         } while (i != msg->sg.end);
2737
2738         if (start >= offset + l)
2739                 return -EINVAL;
2740
2741         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2742
2743         /* If no space available will fallback to copy, we need at
2744          * least one scatterlist elem available to push data into
2745          * when start aligns to the beginning of an element or two
2746          * when it falls inside an element. We handle the start equals
2747          * offset case because its the common case for inserting a
2748          * header.
2749          */
2750         if (!space || (space == 1 && start != offset))
2751                 copy = msg->sg.data[i].length;
2752
2753         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2754                            get_order(copy + len));
2755         if (unlikely(!page))
2756                 return -ENOMEM;
2757
2758         if (copy) {
2759                 int front, back;
2760
2761                 raw = page_address(page);
2762
2763                 psge = sk_msg_elem(msg, i);
2764                 front = start - offset;
2765                 back = psge->length - front;
2766                 from = sg_virt(psge);
2767
2768                 if (front)
2769                         memcpy(raw, from, front);
2770
2771                 if (back) {
2772                         from += front;
2773                         to = raw + front + len;
2774
2775                         memcpy(to, from, back);
2776                 }
2777
2778                 put_page(sg_page(psge));
2779         } else if (start - offset) {
2780                 psge = sk_msg_elem(msg, i);
2781                 rsge = sk_msg_elem_cpy(msg, i);
2782
2783                 psge->length = start - offset;
2784                 rsge.length -= psge->length;
2785                 rsge.offset += start;
2786
2787                 sk_msg_iter_var_next(i);
2788                 sg_unmark_end(psge);
2789                 sg_unmark_end(&rsge);
2790                 sk_msg_iter_next(msg, end);
2791         }
2792
2793         /* Slot(s) to place newly allocated data */
2794         new = i;
2795
2796         /* Shift one or two slots as needed */
2797         if (!copy) {
2798                 sge = sk_msg_elem_cpy(msg, i);
2799
2800                 sk_msg_iter_var_next(i);
2801                 sg_unmark_end(&sge);
2802                 sk_msg_iter_next(msg, end);
2803
2804                 nsge = sk_msg_elem_cpy(msg, i);
2805                 if (rsge.length) {
2806                         sk_msg_iter_var_next(i);
2807                         nnsge = sk_msg_elem_cpy(msg, i);
2808                 }
2809
2810                 while (i != msg->sg.end) {
2811                         msg->sg.data[i] = sge;
2812                         sge = nsge;
2813                         sk_msg_iter_var_next(i);
2814                         if (rsge.length) {
2815                                 nsge = nnsge;
2816                                 nnsge = sk_msg_elem_cpy(msg, i);
2817                         } else {
2818                                 nsge = sk_msg_elem_cpy(msg, i);
2819                         }
2820                 }
2821         }
2822
2823         /* Place newly allocated data buffer */
2824         sk_mem_charge(msg->sk, len);
2825         msg->sg.size += len;
2826         __clear_bit(new, &msg->sg.copy);
2827         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2828         if (rsge.length) {
2829                 get_page(sg_page(&rsge));
2830                 sk_msg_iter_var_next(new);
2831                 msg->sg.data[new] = rsge;
2832         }
2833
2834         sk_msg_compute_data_pointers(msg);
2835         return 0;
2836 }
2837
2838 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2839         .func           = bpf_msg_push_data,
2840         .gpl_only       = false,
2841         .ret_type       = RET_INTEGER,
2842         .arg1_type      = ARG_PTR_TO_CTX,
2843         .arg2_type      = ARG_ANYTHING,
2844         .arg3_type      = ARG_ANYTHING,
2845         .arg4_type      = ARG_ANYTHING,
2846 };
2847
2848 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2849 {
2850         int prev;
2851
2852         do {
2853                 prev = i;
2854                 sk_msg_iter_var_next(i);
2855                 msg->sg.data[prev] = msg->sg.data[i];
2856         } while (i != msg->sg.end);
2857
2858         sk_msg_iter_prev(msg, end);
2859 }
2860
2861 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2862 {
2863         struct scatterlist tmp, sge;
2864
2865         sk_msg_iter_next(msg, end);
2866         sge = sk_msg_elem_cpy(msg, i);
2867         sk_msg_iter_var_next(i);
2868         tmp = sk_msg_elem_cpy(msg, i);
2869
2870         while (i != msg->sg.end) {
2871                 msg->sg.data[i] = sge;
2872                 sk_msg_iter_var_next(i);
2873                 sge = tmp;
2874                 tmp = sk_msg_elem_cpy(msg, i);
2875         }
2876 }
2877
2878 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2879            u32, len, u64, flags)
2880 {
2881         u32 i = 0, l = 0, space, offset = 0;
2882         u64 last = start + len;
2883         int pop;
2884
2885         if (unlikely(flags))
2886                 return -EINVAL;
2887
2888         /* First find the starting scatterlist element */
2889         i = msg->sg.start;
2890         do {
2891                 offset += l;
2892                 l = sk_msg_elem(msg, i)->length;
2893
2894                 if (start < offset + l)
2895                         break;
2896                 sk_msg_iter_var_next(i);
2897         } while (i != msg->sg.end);
2898
2899         /* Bounds checks: start and pop must be inside message */
2900         if (start >= offset + l || last >= msg->sg.size)
2901                 return -EINVAL;
2902
2903         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2904
2905         pop = len;
2906         /* --------------| offset
2907          * -| start      |-------- len -------|
2908          *
2909          *  |----- a ----|-------- pop -------|----- b ----|
2910          *  |______________________________________________| length
2911          *
2912          *
2913          * a:   region at front of scatter element to save
2914          * b:   region at back of scatter element to save when length > A + pop
2915          * pop: region to pop from element, same as input 'pop' here will be
2916          *      decremented below per iteration.
2917          *
2918          * Two top-level cases to handle when start != offset, first B is non
2919          * zero and second B is zero corresponding to when a pop includes more
2920          * than one element.
2921          *
2922          * Then if B is non-zero AND there is no space allocate space and
2923          * compact A, B regions into page. If there is space shift ring to
2924          * the rigth free'ing the next element in ring to place B, leaving
2925          * A untouched except to reduce length.
2926          */
2927         if (start != offset) {
2928                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2929                 int a = start;
2930                 int b = sge->length - pop - a;
2931
2932                 sk_msg_iter_var_next(i);
2933
2934                 if (pop < sge->length - a) {
2935                         if (space) {
2936                                 sge->length = a;
2937                                 sk_msg_shift_right(msg, i);
2938                                 nsge = sk_msg_elem(msg, i);
2939                                 get_page(sg_page(sge));
2940                                 sg_set_page(nsge,
2941                                             sg_page(sge),
2942                                             b, sge->offset + pop + a);
2943                         } else {
2944                                 struct page *page, *orig;
2945                                 u8 *to, *from;
2946
2947                                 page = alloc_pages(__GFP_NOWARN |
2948                                                    __GFP_COMP   | GFP_ATOMIC,
2949                                                    get_order(a + b));
2950                                 if (unlikely(!page))
2951                                         return -ENOMEM;
2952
2953                                 sge->length = a;
2954                                 orig = sg_page(sge);
2955                                 from = sg_virt(sge);
2956                                 to = page_address(page);
2957                                 memcpy(to, from, a);
2958                                 memcpy(to + a, from + a + pop, b);
2959                                 sg_set_page(sge, page, a + b, 0);
2960                                 put_page(orig);
2961                         }
2962                         pop = 0;
2963                 } else if (pop >= sge->length - a) {
2964                         pop -= (sge->length - a);
2965                         sge->length = a;
2966                 }
2967         }
2968
2969         /* From above the current layout _must_ be as follows,
2970          *
2971          * -| offset
2972          * -| start
2973          *
2974          *  |---- pop ---|---------------- b ------------|
2975          *  |____________________________________________| length
2976          *
2977          * Offset and start of the current msg elem are equal because in the
2978          * previous case we handled offset != start and either consumed the
2979          * entire element and advanced to the next element OR pop == 0.
2980          *
2981          * Two cases to handle here are first pop is less than the length
2982          * leaving some remainder b above. Simply adjust the element's layout
2983          * in this case. Or pop >= length of the element so that b = 0. In this
2984          * case advance to next element decrementing pop.
2985          */
2986         while (pop) {
2987                 struct scatterlist *sge = sk_msg_elem(msg, i);
2988
2989                 if (pop < sge->length) {
2990                         sge->length -= pop;
2991                         sge->offset += pop;
2992                         pop = 0;
2993                 } else {
2994                         pop -= sge->length;
2995                         sk_msg_shift_left(msg, i);
2996                 }
2997                 sk_msg_iter_var_next(i);
2998         }
2999
3000         sk_mem_uncharge(msg->sk, len - pop);
3001         msg->sg.size -= (len - pop);
3002         sk_msg_compute_data_pointers(msg);
3003         return 0;
3004 }
3005
3006 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3007         .func           = bpf_msg_pop_data,
3008         .gpl_only       = false,
3009         .ret_type       = RET_INTEGER,
3010         .arg1_type      = ARG_PTR_TO_CTX,
3011         .arg2_type      = ARG_ANYTHING,
3012         .arg3_type      = ARG_ANYTHING,
3013         .arg4_type      = ARG_ANYTHING,
3014 };
3015
3016 #ifdef CONFIG_CGROUP_NET_CLASSID
3017 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3018 {
3019         return __task_get_classid(current);
3020 }
3021
3022 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3023         .func           = bpf_get_cgroup_classid_curr,
3024         .gpl_only       = false,
3025         .ret_type       = RET_INTEGER,
3026 };
3027
3028 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3029 {
3030         struct sock *sk = skb_to_full_sk(skb);
3031
3032         if (!sk || !sk_fullsock(sk))
3033                 return 0;
3034
3035         return sock_cgroup_classid(&sk->sk_cgrp_data);
3036 }
3037
3038 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3039         .func           = bpf_skb_cgroup_classid,
3040         .gpl_only       = false,
3041         .ret_type       = RET_INTEGER,
3042         .arg1_type      = ARG_PTR_TO_CTX,
3043 };
3044 #endif
3045
3046 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3047 {
3048         return task_get_classid(skb);
3049 }
3050
3051 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3052         .func           = bpf_get_cgroup_classid,
3053         .gpl_only       = false,
3054         .ret_type       = RET_INTEGER,
3055         .arg1_type      = ARG_PTR_TO_CTX,
3056 };
3057
3058 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3059 {
3060         return dst_tclassid(skb);
3061 }
3062
3063 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3064         .func           = bpf_get_route_realm,
3065         .gpl_only       = false,
3066         .ret_type       = RET_INTEGER,
3067         .arg1_type      = ARG_PTR_TO_CTX,
3068 };
3069
3070 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3071 {
3072         /* If skb_clear_hash() was called due to mangling, we can
3073          * trigger SW recalculation here. Later access to hash
3074          * can then use the inline skb->hash via context directly
3075          * instead of calling this helper again.
3076          */
3077         return skb_get_hash(skb);
3078 }
3079
3080 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3081         .func           = bpf_get_hash_recalc,
3082         .gpl_only       = false,
3083         .ret_type       = RET_INTEGER,
3084         .arg1_type      = ARG_PTR_TO_CTX,
3085 };
3086
3087 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3088 {
3089         /* After all direct packet write, this can be used once for
3090          * triggering a lazy recalc on next skb_get_hash() invocation.
3091          */
3092         skb_clear_hash(skb);
3093         return 0;
3094 }
3095
3096 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3097         .func           = bpf_set_hash_invalid,
3098         .gpl_only       = false,
3099         .ret_type       = RET_INTEGER,
3100         .arg1_type      = ARG_PTR_TO_CTX,
3101 };
3102
3103 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3104 {
3105         /* Set user specified hash as L4(+), so that it gets returned
3106          * on skb_get_hash() call unless BPF prog later on triggers a
3107          * skb_clear_hash().
3108          */
3109         __skb_set_sw_hash(skb, hash, true);
3110         return 0;
3111 }
3112
3113 static const struct bpf_func_proto bpf_set_hash_proto = {
3114         .func           = bpf_set_hash,
3115         .gpl_only       = false,
3116         .ret_type       = RET_INTEGER,
3117         .arg1_type      = ARG_PTR_TO_CTX,
3118         .arg2_type      = ARG_ANYTHING,
3119 };
3120
3121 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3122            u16, vlan_tci)
3123 {
3124         int ret;
3125
3126         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3127                      vlan_proto != htons(ETH_P_8021AD)))
3128                 vlan_proto = htons(ETH_P_8021Q);
3129
3130         bpf_push_mac_rcsum(skb);
3131         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3132         bpf_pull_mac_rcsum(skb);
3133
3134         bpf_compute_data_pointers(skb);
3135         return ret;
3136 }
3137
3138 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3139         .func           = bpf_skb_vlan_push,
3140         .gpl_only       = false,
3141         .ret_type       = RET_INTEGER,
3142         .arg1_type      = ARG_PTR_TO_CTX,
3143         .arg2_type      = ARG_ANYTHING,
3144         .arg3_type      = ARG_ANYTHING,
3145 };
3146
3147 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3148 {
3149         int ret;
3150
3151         bpf_push_mac_rcsum(skb);
3152         ret = skb_vlan_pop(skb);
3153         bpf_pull_mac_rcsum(skb);
3154
3155         bpf_compute_data_pointers(skb);
3156         return ret;
3157 }
3158
3159 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3160         .func           = bpf_skb_vlan_pop,
3161         .gpl_only       = false,
3162         .ret_type       = RET_INTEGER,
3163         .arg1_type      = ARG_PTR_TO_CTX,
3164 };
3165
3166 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3167 {
3168         /* Caller already did skb_cow() with len as headroom,
3169          * so no need to do it here.
3170          */
3171         skb_push(skb, len);
3172         memmove(skb->data, skb->data + len, off);
3173         memset(skb->data + off, 0, len);
3174
3175         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3176          * needed here as it does not change the skb->csum
3177          * result for checksum complete when summing over
3178          * zeroed blocks.
3179          */
3180         return 0;
3181 }
3182
3183 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3184 {
3185         void *old_data;
3186
3187         /* skb_ensure_writable() is not needed here, as we're
3188          * already working on an uncloned skb.
3189          */
3190         if (unlikely(!pskb_may_pull(skb, off + len)))
3191                 return -ENOMEM;
3192
3193         old_data = skb->data;
3194         __skb_pull(skb, len);
3195         skb_postpull_rcsum(skb, old_data + off, len);
3196         memmove(skb->data, old_data, off);
3197
3198         return 0;
3199 }
3200
3201 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3202 {
3203         bool trans_same = skb->transport_header == skb->network_header;
3204         int ret;
3205
3206         /* There's no need for __skb_push()/__skb_pull() pair to
3207          * get to the start of the mac header as we're guaranteed
3208          * to always start from here under eBPF.
3209          */
3210         ret = bpf_skb_generic_push(skb, off, len);
3211         if (likely(!ret)) {
3212                 skb->mac_header -= len;
3213                 skb->network_header -= len;
3214                 if (trans_same)
3215                         skb->transport_header = skb->network_header;
3216         }
3217
3218         return ret;
3219 }
3220
3221 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3222 {
3223         bool trans_same = skb->transport_header == skb->network_header;
3224         int ret;
3225
3226         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3227         ret = bpf_skb_generic_pop(skb, off, len);
3228         if (likely(!ret)) {
3229                 skb->mac_header += len;
3230                 skb->network_header += len;
3231                 if (trans_same)
3232                         skb->transport_header = skb->network_header;
3233         }
3234
3235         return ret;
3236 }
3237
3238 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3239 {
3240         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3241         u32 off = skb_mac_header_len(skb);
3242         int ret;
3243
3244         ret = skb_cow(skb, len_diff);
3245         if (unlikely(ret < 0))
3246                 return ret;
3247
3248         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3249         if (unlikely(ret < 0))
3250                 return ret;
3251
3252         if (skb_is_gso(skb)) {
3253                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3254
3255                 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3256                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3257                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3258                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3259                 }
3260         }
3261
3262         skb->protocol = htons(ETH_P_IPV6);
3263         skb_clear_hash(skb);
3264
3265         return 0;
3266 }
3267
3268 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3269 {
3270         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3271         u32 off = skb_mac_header_len(skb);
3272         int ret;
3273
3274         ret = skb_unclone(skb, GFP_ATOMIC);
3275         if (unlikely(ret < 0))
3276                 return ret;
3277
3278         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3279         if (unlikely(ret < 0))
3280                 return ret;
3281
3282         if (skb_is_gso(skb)) {
3283                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3284
3285                 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3286                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3287                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3288                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3289                 }
3290         }
3291
3292         skb->protocol = htons(ETH_P_IP);
3293         skb_clear_hash(skb);
3294
3295         return 0;
3296 }
3297
3298 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3299 {
3300         __be16 from_proto = skb->protocol;
3301
3302         if (from_proto == htons(ETH_P_IP) &&
3303               to_proto == htons(ETH_P_IPV6))
3304                 return bpf_skb_proto_4_to_6(skb);
3305
3306         if (from_proto == htons(ETH_P_IPV6) &&
3307               to_proto == htons(ETH_P_IP))
3308                 return bpf_skb_proto_6_to_4(skb);
3309
3310         return -ENOTSUPP;
3311 }
3312
3313 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3314            u64, flags)
3315 {
3316         int ret;
3317
3318         if (unlikely(flags))
3319                 return -EINVAL;
3320
3321         /* General idea is that this helper does the basic groundwork
3322          * needed for changing the protocol, and eBPF program fills the
3323          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3324          * and other helpers, rather than passing a raw buffer here.
3325          *
3326          * The rationale is to keep this minimal and without a need to
3327          * deal with raw packet data. F.e. even if we would pass buffers
3328          * here, the program still needs to call the bpf_lX_csum_replace()
3329          * helpers anyway. Plus, this way we keep also separation of
3330          * concerns, since f.e. bpf_skb_store_bytes() should only take
3331          * care of stores.
3332          *
3333          * Currently, additional options and extension header space are
3334          * not supported, but flags register is reserved so we can adapt
3335          * that. For offloads, we mark packet as dodgy, so that headers
3336          * need to be verified first.
3337          */
3338         ret = bpf_skb_proto_xlat(skb, proto);
3339         bpf_compute_data_pointers(skb);
3340         return ret;
3341 }
3342
3343 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3344         .func           = bpf_skb_change_proto,
3345         .gpl_only       = false,
3346         .ret_type       = RET_INTEGER,
3347         .arg1_type      = ARG_PTR_TO_CTX,
3348         .arg2_type      = ARG_ANYTHING,
3349         .arg3_type      = ARG_ANYTHING,
3350 };
3351
3352 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3353 {
3354         /* We only allow a restricted subset to be changed for now. */
3355         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3356                      !skb_pkt_type_ok(pkt_type)))
3357                 return -EINVAL;
3358
3359         skb->pkt_type = pkt_type;
3360         return 0;
3361 }
3362
3363 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3364         .func           = bpf_skb_change_type,
3365         .gpl_only       = false,
3366         .ret_type       = RET_INTEGER,
3367         .arg1_type      = ARG_PTR_TO_CTX,
3368         .arg2_type      = ARG_ANYTHING,
3369 };
3370
3371 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3372 {
3373         switch (skb->protocol) {
3374         case htons(ETH_P_IP):
3375                 return sizeof(struct iphdr);
3376         case htons(ETH_P_IPV6):
3377                 return sizeof(struct ipv6hdr);
3378         default:
3379                 return ~0U;
3380         }
3381 }
3382
3383 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3384                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3385
3386 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3387                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3388                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3389                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3390                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3391                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3392                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3393
3394 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3395                             u64 flags)
3396 {
3397         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3398         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3399         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3400         unsigned int gso_type = SKB_GSO_DODGY;
3401         int ret;
3402
3403         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3404                 /* udp gso_size delineates datagrams, only allow if fixed */
3405                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3406                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3407                         return -ENOTSUPP;
3408         }
3409
3410         ret = skb_cow_head(skb, len_diff);
3411         if (unlikely(ret < 0))
3412                 return ret;
3413
3414         if (encap) {
3415                 if (skb->protocol != htons(ETH_P_IP) &&
3416                     skb->protocol != htons(ETH_P_IPV6))
3417                         return -ENOTSUPP;
3418
3419                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3420                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3421                         return -EINVAL;
3422
3423                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3424                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3425                         return -EINVAL;
3426
3427                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3428                     inner_mac_len < ETH_HLEN)
3429                         return -EINVAL;
3430
3431                 if (skb->encapsulation)
3432                         return -EALREADY;
3433
3434                 mac_len = skb->network_header - skb->mac_header;
3435                 inner_net = skb->network_header;
3436                 if (inner_mac_len > len_diff)
3437                         return -EINVAL;
3438                 inner_trans = skb->transport_header;
3439         }
3440
3441         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3442         if (unlikely(ret < 0))
3443                 return ret;
3444
3445         if (encap) {
3446                 skb->inner_mac_header = inner_net - inner_mac_len;
3447                 skb->inner_network_header = inner_net;
3448                 skb->inner_transport_header = inner_trans;
3449
3450                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3451                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3452                 else
3453                         skb_set_inner_protocol(skb, skb->protocol);
3454
3455                 skb->encapsulation = 1;
3456                 skb_set_network_header(skb, mac_len);
3457
3458                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3459                         gso_type |= SKB_GSO_UDP_TUNNEL;
3460                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3461                         gso_type |= SKB_GSO_GRE;
3462                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3463                         gso_type |= SKB_GSO_IPXIP6;
3464                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3465                         gso_type |= SKB_GSO_IPXIP4;
3466
3467                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3468                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3469                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3470                                         sizeof(struct ipv6hdr) :
3471                                         sizeof(struct iphdr);
3472
3473                         skb_set_transport_header(skb, mac_len + nh_len);
3474                 }
3475
3476                 /* Match skb->protocol to new outer l3 protocol */
3477                 if (skb->protocol == htons(ETH_P_IP) &&
3478                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479                         skb->protocol = htons(ETH_P_IPV6);
3480                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3481                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3482                         skb->protocol = htons(ETH_P_IP);
3483         }
3484
3485         if (skb_is_gso(skb)) {
3486                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3487
3488                 /* Due to header grow, MSS needs to be downgraded. */
3489                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3490                         skb_decrease_gso_size(shinfo, len_diff);
3491
3492                 /* Header must be checked, and gso_segs recomputed. */
3493                 shinfo->gso_type |= gso_type;
3494                 shinfo->gso_segs = 0;
3495         }
3496
3497         return 0;
3498 }
3499
3500 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3501                               u64 flags)
3502 {
3503         int ret;
3504
3505         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3506                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3507                 return -EINVAL;
3508
3509         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3510                 /* udp gso_size delineates datagrams, only allow if fixed */
3511                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3512                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3513                         return -ENOTSUPP;
3514         }
3515
3516         ret = skb_unclone(skb, GFP_ATOMIC);
3517         if (unlikely(ret < 0))
3518                 return ret;
3519
3520         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3521         if (unlikely(ret < 0))
3522                 return ret;
3523
3524         if (skb_is_gso(skb)) {
3525                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3526
3527                 /* Due to header shrink, MSS can be upgraded. */
3528                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3529                         skb_increase_gso_size(shinfo, len_diff);
3530
3531                 /* Header must be checked, and gso_segs recomputed. */
3532                 shinfo->gso_type |= SKB_GSO_DODGY;
3533                 shinfo->gso_segs = 0;
3534         }
3535
3536         return 0;
3537 }
3538
3539 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3540
3541 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3542            u32, mode, u64, flags)
3543 {
3544         u32 len_diff_abs = abs(len_diff);
3545         bool shrink = len_diff < 0;
3546         int ret = 0;
3547
3548         if (unlikely(flags || mode))
3549                 return -EINVAL;
3550         if (unlikely(len_diff_abs > 0xfffU))
3551                 return -EFAULT;
3552
3553         if (!shrink) {
3554                 ret = skb_cow(skb, len_diff);
3555                 if (unlikely(ret < 0))
3556                         return ret;
3557                 __skb_push(skb, len_diff_abs);
3558                 memset(skb->data, 0, len_diff_abs);
3559         } else {
3560                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3561                         return -ENOMEM;
3562                 __skb_pull(skb, len_diff_abs);
3563         }
3564         if (tls_sw_has_ctx_rx(skb->sk)) {
3565                 struct strp_msg *rxm = strp_msg(skb);
3566
3567                 rxm->full_len += len_diff;
3568         }
3569         return ret;
3570 }
3571
3572 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3573         .func           = sk_skb_adjust_room,
3574         .gpl_only       = false,
3575         .ret_type       = RET_INTEGER,
3576         .arg1_type      = ARG_PTR_TO_CTX,
3577         .arg2_type      = ARG_ANYTHING,
3578         .arg3_type      = ARG_ANYTHING,
3579         .arg4_type      = ARG_ANYTHING,
3580 };
3581
3582 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3583            u32, mode, u64, flags)
3584 {
3585         u32 len_cur, len_diff_abs = abs(len_diff);
3586         u32 len_min = bpf_skb_net_base_len(skb);
3587         u32 len_max = BPF_SKB_MAX_LEN;
3588         __be16 proto = skb->protocol;
3589         bool shrink = len_diff < 0;
3590         u32 off;
3591         int ret;
3592
3593         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3594                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3595                 return -EINVAL;
3596         if (unlikely(len_diff_abs > 0xfffU))
3597                 return -EFAULT;
3598         if (unlikely(proto != htons(ETH_P_IP) &&
3599                      proto != htons(ETH_P_IPV6)))
3600                 return -ENOTSUPP;
3601
3602         off = skb_mac_header_len(skb);
3603         switch (mode) {
3604         case BPF_ADJ_ROOM_NET:
3605                 off += bpf_skb_net_base_len(skb);
3606                 break;
3607         case BPF_ADJ_ROOM_MAC:
3608                 break;
3609         default:
3610                 return -ENOTSUPP;
3611         }
3612
3613         len_cur = skb->len - skb_network_offset(skb);
3614         if ((shrink && (len_diff_abs >= len_cur ||
3615                         len_cur - len_diff_abs < len_min)) ||
3616             (!shrink && (skb->len + len_diff_abs > len_max &&
3617                          !skb_is_gso(skb))))
3618                 return -ENOTSUPP;
3619
3620         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3621                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3622         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3623                 __skb_reset_checksum_unnecessary(skb);
3624
3625         bpf_compute_data_pointers(skb);
3626         return ret;
3627 }
3628
3629 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3630         .func           = bpf_skb_adjust_room,
3631         .gpl_only       = false,
3632         .ret_type       = RET_INTEGER,
3633         .arg1_type      = ARG_PTR_TO_CTX,
3634         .arg2_type      = ARG_ANYTHING,
3635         .arg3_type      = ARG_ANYTHING,
3636         .arg4_type      = ARG_ANYTHING,
3637 };
3638
3639 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3640 {
3641         u32 min_len = skb_network_offset(skb);
3642
3643         if (skb_transport_header_was_set(skb))
3644                 min_len = skb_transport_offset(skb);
3645         if (skb->ip_summed == CHECKSUM_PARTIAL)
3646                 min_len = skb_checksum_start_offset(skb) +
3647                           skb->csum_offset + sizeof(__sum16);
3648         return min_len;
3649 }
3650
3651 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3652 {
3653         unsigned int old_len = skb->len;
3654         int ret;
3655
3656         ret = __skb_grow_rcsum(skb, new_len);
3657         if (!ret)
3658                 memset(skb->data + old_len, 0, new_len - old_len);
3659         return ret;
3660 }
3661
3662 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3663 {
3664         return __skb_trim_rcsum(skb, new_len);
3665 }
3666
3667 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3668                                         u64 flags)
3669 {
3670         u32 max_len = BPF_SKB_MAX_LEN;
3671         u32 min_len = __bpf_skb_min_len(skb);
3672         int ret;
3673
3674         if (unlikely(flags || new_len > max_len || new_len < min_len))
3675                 return -EINVAL;
3676         if (skb->encapsulation)
3677                 return -ENOTSUPP;
3678
3679         /* The basic idea of this helper is that it's performing the
3680          * needed work to either grow or trim an skb, and eBPF program
3681          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3682          * bpf_lX_csum_replace() and others rather than passing a raw
3683          * buffer here. This one is a slow path helper and intended
3684          * for replies with control messages.
3685          *
3686          * Like in bpf_skb_change_proto(), we want to keep this rather
3687          * minimal and without protocol specifics so that we are able
3688          * to separate concerns as in bpf_skb_store_bytes() should only
3689          * be the one responsible for writing buffers.
3690          *
3691          * It's really expected to be a slow path operation here for
3692          * control message replies, so we're implicitly linearizing,
3693          * uncloning and drop offloads from the skb by this.
3694          */
3695         ret = __bpf_try_make_writable(skb, skb->len);
3696         if (!ret) {
3697                 if (new_len > skb->len)
3698                         ret = bpf_skb_grow_rcsum(skb, new_len);
3699                 else if (new_len < skb->len)
3700                         ret = bpf_skb_trim_rcsum(skb, new_len);
3701                 if (!ret && skb_is_gso(skb))
3702                         skb_gso_reset(skb);
3703         }
3704         return ret;
3705 }
3706
3707 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3708            u64, flags)
3709 {
3710         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3711
3712         bpf_compute_data_pointers(skb);
3713         return ret;
3714 }
3715
3716 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3717         .func           = bpf_skb_change_tail,
3718         .gpl_only       = false,
3719         .ret_type       = RET_INTEGER,
3720         .arg1_type      = ARG_PTR_TO_CTX,
3721         .arg2_type      = ARG_ANYTHING,
3722         .arg3_type      = ARG_ANYTHING,
3723 };
3724
3725 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3726            u64, flags)
3727 {
3728         return __bpf_skb_change_tail(skb, new_len, flags);
3729 }
3730
3731 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3732         .func           = sk_skb_change_tail,
3733         .gpl_only       = false,
3734         .ret_type       = RET_INTEGER,
3735         .arg1_type      = ARG_PTR_TO_CTX,
3736         .arg2_type      = ARG_ANYTHING,
3737         .arg3_type      = ARG_ANYTHING,
3738 };
3739
3740 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3741                                         u64 flags)
3742 {
3743         u32 max_len = BPF_SKB_MAX_LEN;
3744         u32 new_len = skb->len + head_room;
3745         int ret;
3746
3747         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3748                      new_len < skb->len))
3749                 return -EINVAL;
3750
3751         ret = skb_cow(skb, head_room);
3752         if (likely(!ret)) {
3753                 /* Idea for this helper is that we currently only
3754                  * allow to expand on mac header. This means that
3755                  * skb->protocol network header, etc, stay as is.
3756                  * Compared to bpf_skb_change_tail(), we're more
3757                  * flexible due to not needing to linearize or
3758                  * reset GSO. Intention for this helper is to be
3759                  * used by an L3 skb that needs to push mac header
3760                  * for redirection into L2 device.
3761                  */
3762                 __skb_push(skb, head_room);
3763                 memset(skb->data, 0, head_room);
3764                 skb_reset_mac_header(skb);
3765                 skb_reset_mac_len(skb);
3766         }
3767
3768         return ret;
3769 }
3770
3771 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3772            u64, flags)
3773 {
3774         int ret = __bpf_skb_change_head(skb, head_room, flags);
3775
3776         bpf_compute_data_pointers(skb);
3777         return ret;
3778 }
3779
3780 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3781         .func           = bpf_skb_change_head,
3782         .gpl_only       = false,
3783         .ret_type       = RET_INTEGER,
3784         .arg1_type      = ARG_PTR_TO_CTX,
3785         .arg2_type      = ARG_ANYTHING,
3786         .arg3_type      = ARG_ANYTHING,
3787 };
3788
3789 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3790            u64, flags)
3791 {
3792         return __bpf_skb_change_head(skb, head_room, flags);
3793 }
3794
3795 static const struct bpf_func_proto sk_skb_change_head_proto = {
3796         .func           = sk_skb_change_head,
3797         .gpl_only       = false,
3798         .ret_type       = RET_INTEGER,
3799         .arg1_type      = ARG_PTR_TO_CTX,
3800         .arg2_type      = ARG_ANYTHING,
3801         .arg3_type      = ARG_ANYTHING,
3802 };
3803 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3804 {
3805         return xdp_data_meta_unsupported(xdp) ? 0 :
3806                xdp->data - xdp->data_meta;
3807 }
3808
3809 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3810 {
3811         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3812         unsigned long metalen = xdp_get_metalen(xdp);
3813         void *data_start = xdp_frame_end + metalen;
3814         void *data = xdp->data + offset;
3815
3816         if (unlikely(data < data_start ||
3817                      data > xdp->data_end - ETH_HLEN))
3818                 return -EINVAL;
3819
3820         if (metalen)
3821                 memmove(xdp->data_meta + offset,
3822                         xdp->data_meta, metalen);
3823         xdp->data_meta += offset;
3824         xdp->data = data;
3825
3826         return 0;
3827 }
3828
3829 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3830         .func           = bpf_xdp_adjust_head,
3831         .gpl_only       = false,
3832         .ret_type       = RET_INTEGER,
3833         .arg1_type      = ARG_PTR_TO_CTX,
3834         .arg2_type      = ARG_ANYTHING,
3835 };
3836
3837 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3838 {
3839         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3840         void *data_end = xdp->data_end + offset;
3841
3842         /* Notice that xdp_data_hard_end have reserved some tailroom */
3843         if (unlikely(data_end > data_hard_end))
3844                 return -EINVAL;
3845
3846         if (unlikely(data_end < xdp->data + ETH_HLEN))
3847                 return -EINVAL;
3848
3849         /* Clear memory area on grow, can contain uninit kernel memory */
3850         if (offset > 0)
3851                 memset(xdp->data_end, 0, offset);
3852
3853         xdp->data_end = data_end;
3854
3855         return 0;
3856 }
3857
3858 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3859         .func           = bpf_xdp_adjust_tail,
3860         .gpl_only       = false,
3861         .ret_type       = RET_INTEGER,
3862         .arg1_type      = ARG_PTR_TO_CTX,
3863         .arg2_type      = ARG_ANYTHING,
3864 };
3865
3866 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3867 {
3868         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3869         void *meta = xdp->data_meta + offset;
3870         unsigned long metalen = xdp->data - meta;
3871
3872         if (xdp_data_meta_unsupported(xdp))
3873                 return -ENOTSUPP;
3874         if (unlikely(meta < xdp_frame_end ||
3875                      meta > xdp->data))
3876                 return -EINVAL;
3877         if (unlikely(xdp_metalen_invalid(metalen)))
3878                 return -EACCES;
3879
3880         xdp->data_meta = meta;
3881
3882         return 0;
3883 }
3884
3885 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3886         .func           = bpf_xdp_adjust_meta,
3887         .gpl_only       = false,
3888         .ret_type       = RET_INTEGER,
3889         .arg1_type      = ARG_PTR_TO_CTX,
3890         .arg2_type      = ARG_ANYTHING,
3891 };
3892
3893 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3894  * below:
3895  *
3896  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3897  *    of the redirect and store it (along with some other metadata) in a per-CPU
3898  *    struct bpf_redirect_info.
3899  *
3900  * 2. When the program returns the XDP_REDIRECT return code, the driver will
3901  *    call xdp_do_redirect() which will use the information in struct
3902  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
3903  *    bulk queue structure.
3904  *
3905  * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3906  *    which will flush all the different bulk queues, thus completing the
3907  *    redirect.
3908  *
3909  * Pointers to the map entries will be kept around for this whole sequence of
3910  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3911  * the core code; instead, the RCU protection relies on everything happening
3912  * inside a single NAPI poll sequence, which means it's between a pair of calls
3913  * to local_bh_disable()/local_bh_enable().
3914  *
3915  * The map entries are marked as __rcu and the map code makes sure to
3916  * dereference those pointers with rcu_dereference_check() in a way that works
3917  * for both sections that to hold an rcu_read_lock() and sections that are
3918  * called from NAPI without a separate rcu_read_lock(). The code below does not
3919  * use RCU annotations, but relies on those in the map code.
3920  */
3921 void xdp_do_flush(void)
3922 {
3923         __dev_flush();
3924         __cpu_map_flush();
3925         __xsk_map_flush();
3926 }
3927 EXPORT_SYMBOL_GPL(xdp_do_flush);
3928
3929 void bpf_clear_redirect_map(struct bpf_map *map)
3930 {
3931         struct bpf_redirect_info *ri;
3932         int cpu;
3933
3934         for_each_possible_cpu(cpu) {
3935                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3936                 /* Avoid polluting remote cacheline due to writes if
3937                  * not needed. Once we pass this test, we need the
3938                  * cmpxchg() to make sure it hasn't been changed in
3939                  * the meantime by remote CPU.
3940                  */
3941                 if (unlikely(READ_ONCE(ri->map) == map))
3942                         cmpxchg(&ri->map, map, NULL);
3943         }
3944 }
3945
3946 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3947 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3948
3949 u32 xdp_master_redirect(struct xdp_buff *xdp)
3950 {
3951         struct net_device *master, *slave;
3952         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3953
3954         master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3955         slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3956         if (slave && slave != xdp->rxq->dev) {
3957                 /* The target device is different from the receiving device, so
3958                  * redirect it to the new device.
3959                  * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3960                  * drivers to unmap the packet from their rx ring.
3961                  */
3962                 ri->tgt_index = slave->ifindex;
3963                 ri->map_id = INT_MAX;
3964                 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3965                 return XDP_REDIRECT;
3966         }
3967         return XDP_TX;
3968 }
3969 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3970
3971 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3972                     struct bpf_prog *xdp_prog)
3973 {
3974         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3975         enum bpf_map_type map_type = ri->map_type;
3976         void *fwd = ri->tgt_value;
3977         u32 map_id = ri->map_id;
3978         struct bpf_map *map;
3979         int err;
3980
3981         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3982         ri->map_type = BPF_MAP_TYPE_UNSPEC;
3983
3984         switch (map_type) {
3985         case BPF_MAP_TYPE_DEVMAP:
3986                 fallthrough;
3987         case BPF_MAP_TYPE_DEVMAP_HASH:
3988                 map = READ_ONCE(ri->map);
3989                 if (unlikely(map)) {
3990                         WRITE_ONCE(ri->map, NULL);
3991                         err = dev_map_enqueue_multi(xdp, dev, map,
3992                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
3993                 } else {
3994                         err = dev_map_enqueue(fwd, xdp, dev);
3995                 }
3996                 break;
3997         case BPF_MAP_TYPE_CPUMAP:
3998                 err = cpu_map_enqueue(fwd, xdp, dev);
3999                 break;
4000         case BPF_MAP_TYPE_XSKMAP:
4001                 err = __xsk_map_redirect(fwd, xdp);
4002                 break;
4003         case BPF_MAP_TYPE_UNSPEC:
4004                 if (map_id == INT_MAX) {
4005                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4006                         if (unlikely(!fwd)) {
4007                                 err = -EINVAL;
4008                                 break;
4009                         }
4010                         err = dev_xdp_enqueue(fwd, xdp, dev);
4011                         break;
4012                 }
4013                 fallthrough;
4014         default:
4015                 err = -EBADRQC;
4016         }
4017
4018         if (unlikely(err))
4019                 goto err;
4020
4021         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4022         return 0;
4023 err:
4024         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4025         return err;
4026 }
4027 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4028
4029 static int xdp_do_generic_redirect_map(struct net_device *dev,
4030                                        struct sk_buff *skb,
4031                                        struct xdp_buff *xdp,
4032                                        struct bpf_prog *xdp_prog,
4033                                        void *fwd,
4034                                        enum bpf_map_type map_type, u32 map_id)
4035 {
4036         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4037         struct bpf_map *map;
4038         int err;
4039
4040         switch (map_type) {
4041         case BPF_MAP_TYPE_DEVMAP:
4042                 fallthrough;
4043         case BPF_MAP_TYPE_DEVMAP_HASH:
4044                 map = READ_ONCE(ri->map);
4045                 if (unlikely(map)) {
4046                         WRITE_ONCE(ri->map, NULL);
4047                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4048                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4049                 } else {
4050                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4051                 }
4052                 if (unlikely(err))
4053                         goto err;
4054                 break;
4055         case BPF_MAP_TYPE_XSKMAP:
4056                 err = xsk_generic_rcv(fwd, xdp);
4057                 if (err)
4058                         goto err;
4059                 consume_skb(skb);
4060                 break;
4061         case BPF_MAP_TYPE_CPUMAP:
4062                 err = cpu_map_generic_redirect(fwd, skb);
4063                 if (unlikely(err))
4064                         goto err;
4065                 break;
4066         default:
4067                 err = -EBADRQC;
4068                 goto err;
4069         }
4070
4071         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4072         return 0;
4073 err:
4074         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4075         return err;
4076 }
4077
4078 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4079                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4080 {
4081         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4082         enum bpf_map_type map_type = ri->map_type;
4083         void *fwd = ri->tgt_value;
4084         u32 map_id = ri->map_id;
4085         int err;
4086
4087         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4088         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4089
4090         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4091                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4092                 if (unlikely(!fwd)) {
4093                         err = -EINVAL;
4094                         goto err;
4095                 }
4096
4097                 err = xdp_ok_fwd_dev(fwd, skb->len);
4098                 if (unlikely(err))
4099                         goto err;
4100
4101                 skb->dev = fwd;
4102                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4103                 generic_xdp_tx(skb, xdp_prog);
4104                 return 0;
4105         }
4106
4107         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4108 err:
4109         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4110         return err;
4111 }
4112
4113 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4114 {
4115         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4116
4117         if (unlikely(flags))
4118                 return XDP_ABORTED;
4119
4120         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4121          * by map_idr) is used for ifindex based XDP redirect.
4122          */
4123         ri->tgt_index = ifindex;
4124         ri->map_id = INT_MAX;
4125         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4126
4127         return XDP_REDIRECT;
4128 }
4129
4130 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4131         .func           = bpf_xdp_redirect,
4132         .gpl_only       = false,
4133         .ret_type       = RET_INTEGER,
4134         .arg1_type      = ARG_ANYTHING,
4135         .arg2_type      = ARG_ANYTHING,
4136 };
4137
4138 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4139            u64, flags)
4140 {
4141         return map->ops->map_redirect(map, ifindex, flags);
4142 }
4143
4144 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4145         .func           = bpf_xdp_redirect_map,
4146         .gpl_only       = false,
4147         .ret_type       = RET_INTEGER,
4148         .arg1_type      = ARG_CONST_MAP_PTR,
4149         .arg2_type      = ARG_ANYTHING,
4150         .arg3_type      = ARG_ANYTHING,
4151 };
4152
4153 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4154                                   unsigned long off, unsigned long len)
4155 {
4156         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4157
4158         if (unlikely(!ptr))
4159                 return len;
4160         if (ptr != dst_buff)
4161                 memcpy(dst_buff, ptr, len);
4162
4163         return 0;
4164 }
4165
4166 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4167            u64, flags, void *, meta, u64, meta_size)
4168 {
4169         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4170
4171         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4172                 return -EINVAL;
4173         if (unlikely(!skb || skb_size > skb->len))
4174                 return -EFAULT;
4175
4176         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4177                                 bpf_skb_copy);
4178 }
4179
4180 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4181         .func           = bpf_skb_event_output,
4182         .gpl_only       = true,
4183         .ret_type       = RET_INTEGER,
4184         .arg1_type      = ARG_PTR_TO_CTX,
4185         .arg2_type      = ARG_CONST_MAP_PTR,
4186         .arg3_type      = ARG_ANYTHING,
4187         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4188         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4189 };
4190
4191 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4192
4193 const struct bpf_func_proto bpf_skb_output_proto = {
4194         .func           = bpf_skb_event_output,
4195         .gpl_only       = true,
4196         .ret_type       = RET_INTEGER,
4197         .arg1_type      = ARG_PTR_TO_BTF_ID,
4198         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4199         .arg2_type      = ARG_CONST_MAP_PTR,
4200         .arg3_type      = ARG_ANYTHING,
4201         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4202         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4203 };
4204
4205 static unsigned short bpf_tunnel_key_af(u64 flags)
4206 {
4207         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4208 }
4209
4210 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4211            u32, size, u64, flags)
4212 {
4213         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4214         u8 compat[sizeof(struct bpf_tunnel_key)];
4215         void *to_orig = to;
4216         int err;
4217
4218         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4219                 err = -EINVAL;
4220                 goto err_clear;
4221         }
4222         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4223                 err = -EPROTO;
4224                 goto err_clear;
4225         }
4226         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4227                 err = -EINVAL;
4228                 switch (size) {
4229                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4230                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4231                         goto set_compat;
4232                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4233                         /* Fixup deprecated structure layouts here, so we have
4234                          * a common path later on.
4235                          */
4236                         if (ip_tunnel_info_af(info) != AF_INET)
4237                                 goto err_clear;
4238 set_compat:
4239                         to = (struct bpf_tunnel_key *)compat;
4240                         break;
4241                 default:
4242                         goto err_clear;
4243                 }
4244         }
4245
4246         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4247         to->tunnel_tos = info->key.tos;
4248         to->tunnel_ttl = info->key.ttl;
4249         to->tunnel_ext = 0;
4250
4251         if (flags & BPF_F_TUNINFO_IPV6) {
4252                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4253                        sizeof(to->remote_ipv6));
4254                 to->tunnel_label = be32_to_cpu(info->key.label);
4255         } else {
4256                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4257                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4258                 to->tunnel_label = 0;
4259         }
4260
4261         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4262                 memcpy(to_orig, to, size);
4263
4264         return 0;
4265 err_clear:
4266         memset(to_orig, 0, size);
4267         return err;
4268 }
4269
4270 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4271         .func           = bpf_skb_get_tunnel_key,
4272         .gpl_only       = false,
4273         .ret_type       = RET_INTEGER,
4274         .arg1_type      = ARG_PTR_TO_CTX,
4275         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4276         .arg3_type      = ARG_CONST_SIZE,
4277         .arg4_type      = ARG_ANYTHING,
4278 };
4279
4280 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4281 {
4282         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4283         int err;
4284
4285         if (unlikely(!info ||
4286                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4287                 err = -ENOENT;
4288                 goto err_clear;
4289         }
4290         if (unlikely(size < info->options_len)) {
4291                 err = -ENOMEM;
4292                 goto err_clear;
4293         }
4294
4295         ip_tunnel_info_opts_get(to, info);
4296         if (size > info->options_len)
4297                 memset(to + info->options_len, 0, size - info->options_len);
4298
4299         return info->options_len;
4300 err_clear:
4301         memset(to, 0, size);
4302         return err;
4303 }
4304
4305 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4306         .func           = bpf_skb_get_tunnel_opt,
4307         .gpl_only       = false,
4308         .ret_type       = RET_INTEGER,
4309         .arg1_type      = ARG_PTR_TO_CTX,
4310         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4311         .arg3_type      = ARG_CONST_SIZE,
4312 };
4313
4314 static struct metadata_dst __percpu *md_dst;
4315
4316 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4317            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4318 {
4319         struct metadata_dst *md = this_cpu_ptr(md_dst);
4320         u8 compat[sizeof(struct bpf_tunnel_key)];
4321         struct ip_tunnel_info *info;
4322
4323         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4324                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4325                 return -EINVAL;
4326         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4327                 switch (size) {
4328                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4329                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4330                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4331                         /* Fixup deprecated structure layouts here, so we have
4332                          * a common path later on.
4333                          */
4334                         memcpy(compat, from, size);
4335                         memset(compat + size, 0, sizeof(compat) - size);
4336                         from = (const struct bpf_tunnel_key *) compat;
4337                         break;
4338                 default:
4339                         return -EINVAL;
4340                 }
4341         }
4342         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4343                      from->tunnel_ext))
4344                 return -EINVAL;
4345
4346         skb_dst_drop(skb);
4347         dst_hold((struct dst_entry *) md);
4348         skb_dst_set(skb, (struct dst_entry *) md);
4349
4350         info = &md->u.tun_info;
4351         memset(info, 0, sizeof(*info));
4352         info->mode = IP_TUNNEL_INFO_TX;
4353
4354         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4355         if (flags & BPF_F_DONT_FRAGMENT)
4356                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4357         if (flags & BPF_F_ZERO_CSUM_TX)
4358                 info->key.tun_flags &= ~TUNNEL_CSUM;
4359         if (flags & BPF_F_SEQ_NUMBER)
4360                 info->key.tun_flags |= TUNNEL_SEQ;
4361
4362         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4363         info->key.tos = from->tunnel_tos;
4364         info->key.ttl = from->tunnel_ttl;
4365
4366         if (flags & BPF_F_TUNINFO_IPV6) {
4367                 info->mode |= IP_TUNNEL_INFO_IPV6;
4368                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4369                        sizeof(from->remote_ipv6));
4370                 info->key.label = cpu_to_be32(from->tunnel_label) &
4371                                   IPV6_FLOWLABEL_MASK;
4372         } else {
4373                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4374         }
4375
4376         return 0;
4377 }
4378
4379 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4380         .func           = bpf_skb_set_tunnel_key,
4381         .gpl_only       = false,
4382         .ret_type       = RET_INTEGER,
4383         .arg1_type      = ARG_PTR_TO_CTX,
4384         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4385         .arg3_type      = ARG_CONST_SIZE,
4386         .arg4_type      = ARG_ANYTHING,
4387 };
4388
4389 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4390            const u8 *, from, u32, size)
4391 {
4392         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4393         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4394
4395         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4396                 return -EINVAL;
4397         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4398                 return -ENOMEM;
4399
4400         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4401
4402         return 0;
4403 }
4404
4405 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4406         .func           = bpf_skb_set_tunnel_opt,
4407         .gpl_only       = false,
4408         .ret_type       = RET_INTEGER,
4409         .arg1_type      = ARG_PTR_TO_CTX,
4410         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4411         .arg3_type      = ARG_CONST_SIZE,
4412 };
4413
4414 static const struct bpf_func_proto *
4415 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4416 {
4417         if (!md_dst) {
4418                 struct metadata_dst __percpu *tmp;
4419
4420                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4421                                                 METADATA_IP_TUNNEL,
4422                                                 GFP_KERNEL);
4423                 if (!tmp)
4424                         return NULL;
4425                 if (cmpxchg(&md_dst, NULL, tmp))
4426                         metadata_dst_free_percpu(tmp);
4427         }
4428
4429         switch (which) {
4430         case BPF_FUNC_skb_set_tunnel_key:
4431                 return &bpf_skb_set_tunnel_key_proto;
4432         case BPF_FUNC_skb_set_tunnel_opt:
4433                 return &bpf_skb_set_tunnel_opt_proto;
4434         default:
4435                 return NULL;
4436         }
4437 }
4438
4439 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4440            u32, idx)
4441 {
4442         struct bpf_array *array = container_of(map, struct bpf_array, map);
4443         struct cgroup *cgrp;
4444         struct sock *sk;
4445
4446         sk = skb_to_full_sk(skb);
4447         if (!sk || !sk_fullsock(sk))
4448                 return -ENOENT;
4449         if (unlikely(idx >= array->map.max_entries))
4450                 return -E2BIG;
4451
4452         cgrp = READ_ONCE(array->ptrs[idx]);
4453         if (unlikely(!cgrp))
4454                 return -EAGAIN;
4455
4456         return sk_under_cgroup_hierarchy(sk, cgrp);
4457 }
4458
4459 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4460         .func           = bpf_skb_under_cgroup,
4461         .gpl_only       = false,
4462         .ret_type       = RET_INTEGER,
4463         .arg1_type      = ARG_PTR_TO_CTX,
4464         .arg2_type      = ARG_CONST_MAP_PTR,
4465         .arg3_type      = ARG_ANYTHING,
4466 };
4467
4468 #ifdef CONFIG_SOCK_CGROUP_DATA
4469 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4470 {
4471         struct cgroup *cgrp;
4472
4473         sk = sk_to_full_sk(sk);
4474         if (!sk || !sk_fullsock(sk))
4475                 return 0;
4476
4477         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4478         return cgroup_id(cgrp);
4479 }
4480
4481 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4482 {
4483         return __bpf_sk_cgroup_id(skb->sk);
4484 }
4485
4486 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4487         .func           = bpf_skb_cgroup_id,
4488         .gpl_only       = false,
4489         .ret_type       = RET_INTEGER,
4490         .arg1_type      = ARG_PTR_TO_CTX,
4491 };
4492
4493 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4494                                               int ancestor_level)
4495 {
4496         struct cgroup *ancestor;
4497         struct cgroup *cgrp;
4498
4499         sk = sk_to_full_sk(sk);
4500         if (!sk || !sk_fullsock(sk))
4501                 return 0;
4502
4503         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4504         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4505         if (!ancestor)
4506                 return 0;
4507
4508         return cgroup_id(ancestor);
4509 }
4510
4511 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4512            ancestor_level)
4513 {
4514         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4515 }
4516
4517 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4518         .func           = bpf_skb_ancestor_cgroup_id,
4519         .gpl_only       = false,
4520         .ret_type       = RET_INTEGER,
4521         .arg1_type      = ARG_PTR_TO_CTX,
4522         .arg2_type      = ARG_ANYTHING,
4523 };
4524
4525 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4526 {
4527         return __bpf_sk_cgroup_id(sk);
4528 }
4529
4530 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4531         .func           = bpf_sk_cgroup_id,
4532         .gpl_only       = false,
4533         .ret_type       = RET_INTEGER,
4534         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4535 };
4536
4537 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4538 {
4539         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4540 }
4541
4542 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4543         .func           = bpf_sk_ancestor_cgroup_id,
4544         .gpl_only       = false,
4545         .ret_type       = RET_INTEGER,
4546         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4547         .arg2_type      = ARG_ANYTHING,
4548 };
4549 #endif
4550
4551 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4552                                   unsigned long off, unsigned long len)
4553 {
4554         memcpy(dst_buff, src_buff + off, len);
4555         return 0;
4556 }
4557
4558 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4559            u64, flags, void *, meta, u64, meta_size)
4560 {
4561         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4562
4563         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4564                 return -EINVAL;
4565         if (unlikely(!xdp ||
4566                      xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4567                 return -EFAULT;
4568
4569         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4570                                 xdp_size, bpf_xdp_copy);
4571 }
4572
4573 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4574         .func           = bpf_xdp_event_output,
4575         .gpl_only       = true,
4576         .ret_type       = RET_INTEGER,
4577         .arg1_type      = ARG_PTR_TO_CTX,
4578         .arg2_type      = ARG_CONST_MAP_PTR,
4579         .arg3_type      = ARG_ANYTHING,
4580         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4581         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4582 };
4583
4584 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4585
4586 const struct bpf_func_proto bpf_xdp_output_proto = {
4587         .func           = bpf_xdp_event_output,
4588         .gpl_only       = true,
4589         .ret_type       = RET_INTEGER,
4590         .arg1_type      = ARG_PTR_TO_BTF_ID,
4591         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4592         .arg2_type      = ARG_CONST_MAP_PTR,
4593         .arg3_type      = ARG_ANYTHING,
4594         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4595         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4596 };
4597
4598 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4599 {
4600         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4601 }
4602
4603 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4604         .func           = bpf_get_socket_cookie,
4605         .gpl_only       = false,
4606         .ret_type       = RET_INTEGER,
4607         .arg1_type      = ARG_PTR_TO_CTX,
4608 };
4609
4610 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4611 {
4612         return __sock_gen_cookie(ctx->sk);
4613 }
4614
4615 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4616         .func           = bpf_get_socket_cookie_sock_addr,
4617         .gpl_only       = false,
4618         .ret_type       = RET_INTEGER,
4619         .arg1_type      = ARG_PTR_TO_CTX,
4620 };
4621
4622 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4623 {
4624         return __sock_gen_cookie(ctx);
4625 }
4626
4627 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4628         .func           = bpf_get_socket_cookie_sock,
4629         .gpl_only       = false,
4630         .ret_type       = RET_INTEGER,
4631         .arg1_type      = ARG_PTR_TO_CTX,
4632 };
4633
4634 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4635 {
4636         return sk ? sock_gen_cookie(sk) : 0;
4637 }
4638
4639 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4640         .func           = bpf_get_socket_ptr_cookie,
4641         .gpl_only       = false,
4642         .ret_type       = RET_INTEGER,
4643         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4644 };
4645
4646 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4647 {
4648         return __sock_gen_cookie(ctx->sk);
4649 }
4650
4651 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4652         .func           = bpf_get_socket_cookie_sock_ops,
4653         .gpl_only       = false,
4654         .ret_type       = RET_INTEGER,
4655         .arg1_type      = ARG_PTR_TO_CTX,
4656 };
4657
4658 static u64 __bpf_get_netns_cookie(struct sock *sk)
4659 {
4660         const struct net *net = sk ? sock_net(sk) : &init_net;
4661
4662         return net->net_cookie;
4663 }
4664
4665 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4666 {
4667         return __bpf_get_netns_cookie(ctx);
4668 }
4669
4670 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4671         .func           = bpf_get_netns_cookie_sock,
4672         .gpl_only       = false,
4673         .ret_type       = RET_INTEGER,
4674         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4675 };
4676
4677 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4678 {
4679         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4680 }
4681
4682 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4683         .func           = bpf_get_netns_cookie_sock_addr,
4684         .gpl_only       = false,
4685         .ret_type       = RET_INTEGER,
4686         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4687 };
4688
4689 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4690 {
4691         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4692 }
4693
4694 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4695         .func           = bpf_get_netns_cookie_sock_ops,
4696         .gpl_only       = false,
4697         .ret_type       = RET_INTEGER,
4698         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4699 };
4700
4701 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4702 {
4703         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4704 }
4705
4706 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4707         .func           = bpf_get_netns_cookie_sk_msg,
4708         .gpl_only       = false,
4709         .ret_type       = RET_INTEGER,
4710         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4711 };
4712
4713 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4714 {
4715         struct sock *sk = sk_to_full_sk(skb->sk);
4716         kuid_t kuid;
4717
4718         if (!sk || !sk_fullsock(sk))
4719                 return overflowuid;
4720         kuid = sock_net_uid(sock_net(sk), sk);
4721         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4722 }
4723
4724 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4725         .func           = bpf_get_socket_uid,
4726         .gpl_only       = false,
4727         .ret_type       = RET_INTEGER,
4728         .arg1_type      = ARG_PTR_TO_CTX,
4729 };
4730
4731 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4732                            char *optval, int optlen)
4733 {
4734         char devname[IFNAMSIZ];
4735         int val, valbool;
4736         struct net *net;
4737         int ifindex;
4738         int ret = 0;
4739
4740         if (!sk_fullsock(sk))
4741                 return -EINVAL;
4742
4743         sock_owned_by_me(sk);
4744
4745         if (level == SOL_SOCKET) {
4746                 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4747                         return -EINVAL;
4748                 val = *((int *)optval);
4749                 valbool = val ? 1 : 0;
4750
4751                 /* Only some socketops are supported */
4752                 switch (optname) {
4753                 case SO_RCVBUF:
4754                         val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
4755                         val = min_t(int, val, INT_MAX / 2);
4756                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4757                         WRITE_ONCE(sk->sk_rcvbuf,
4758                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4759                         break;
4760                 case SO_SNDBUF:
4761                         val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
4762                         val = min_t(int, val, INT_MAX / 2);
4763                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4764                         WRITE_ONCE(sk->sk_sndbuf,
4765                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4766                         break;
4767                 case SO_MAX_PACING_RATE: /* 32bit version */
4768                         if (val != ~0U)
4769                                 cmpxchg(&sk->sk_pacing_status,
4770                                         SK_PACING_NONE,
4771                                         SK_PACING_NEEDED);
4772                         sk->sk_max_pacing_rate = (val == ~0U) ?
4773                                                  ~0UL : (unsigned int)val;
4774                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4775                                                  sk->sk_max_pacing_rate);
4776                         break;
4777                 case SO_PRIORITY:
4778                         sk->sk_priority = val;
4779                         break;
4780                 case SO_RCVLOWAT:
4781                         if (val < 0)
4782                                 val = INT_MAX;
4783                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4784                         break;
4785                 case SO_MARK:
4786                         if (sk->sk_mark != val) {
4787                                 sk->sk_mark = val;
4788                                 sk_dst_reset(sk);
4789                         }
4790                         break;
4791                 case SO_BINDTODEVICE:
4792                         optlen = min_t(long, optlen, IFNAMSIZ - 1);
4793                         strncpy(devname, optval, optlen);
4794                         devname[optlen] = 0;
4795
4796                         ifindex = 0;
4797                         if (devname[0] != '\0') {
4798                                 struct net_device *dev;
4799
4800                                 ret = -ENODEV;
4801
4802                                 net = sock_net(sk);
4803                                 dev = dev_get_by_name(net, devname);
4804                                 if (!dev)
4805                                         break;
4806                                 ifindex = dev->ifindex;
4807                                 dev_put(dev);
4808                         }
4809                         fallthrough;
4810                 case SO_BINDTOIFINDEX:
4811                         if (optname == SO_BINDTOIFINDEX)
4812                                 ifindex = val;
4813                         ret = sock_bindtoindex(sk, ifindex, false);
4814                         break;
4815                 case SO_KEEPALIVE:
4816                         if (sk->sk_prot->keepalive)
4817                                 sk->sk_prot->keepalive(sk, valbool);
4818                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4819                         break;
4820                 case SO_REUSEPORT:
4821                         sk->sk_reuseport = valbool;
4822                         break;
4823                 default:
4824                         ret = -EINVAL;
4825                 }
4826 #ifdef CONFIG_INET
4827         } else if (level == SOL_IP) {
4828                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4829                         return -EINVAL;
4830
4831                 val = *((int *)optval);
4832                 /* Only some options are supported */
4833                 switch (optname) {
4834                 case IP_TOS:
4835                         if (val < -1 || val > 0xff) {
4836                                 ret = -EINVAL;
4837                         } else {
4838                                 struct inet_sock *inet = inet_sk(sk);
4839
4840                                 if (val == -1)
4841                                         val = 0;
4842                                 inet->tos = val;
4843                         }
4844                         break;
4845                 default:
4846                         ret = -EINVAL;
4847                 }
4848 #if IS_ENABLED(CONFIG_IPV6)
4849         } else if (level == SOL_IPV6) {
4850                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4851                         return -EINVAL;
4852
4853                 val = *((int *)optval);
4854                 /* Only some options are supported */
4855                 switch (optname) {
4856                 case IPV6_TCLASS:
4857                         if (val < -1 || val > 0xff) {
4858                                 ret = -EINVAL;
4859                         } else {
4860                                 struct ipv6_pinfo *np = inet6_sk(sk);
4861
4862                                 if (val == -1)
4863                                         val = 0;
4864                                 np->tclass = val;
4865                         }
4866                         break;
4867                 default:
4868                         ret = -EINVAL;
4869                 }
4870 #endif
4871         } else if (level == SOL_TCP &&
4872                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4873                 if (optname == TCP_CONGESTION) {
4874                         char name[TCP_CA_NAME_MAX];
4875
4876                         strncpy(name, optval, min_t(long, optlen,
4877                                                     TCP_CA_NAME_MAX-1));
4878                         name[TCP_CA_NAME_MAX-1] = 0;
4879                         ret = tcp_set_congestion_control(sk, name, false, true);
4880                 } else {
4881                         struct inet_connection_sock *icsk = inet_csk(sk);
4882                         struct tcp_sock *tp = tcp_sk(sk);
4883                         unsigned long timeout;
4884
4885                         if (optlen != sizeof(int))
4886                                 return -EINVAL;
4887
4888                         val = *((int *)optval);
4889                         /* Only some options are supported */
4890                         switch (optname) {
4891                         case TCP_BPF_IW:
4892                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4893                                         ret = -EINVAL;
4894                                 else
4895                                         tcp_snd_cwnd_set(tp, val);
4896                                 break;
4897                         case TCP_BPF_SNDCWND_CLAMP:
4898                                 if (val <= 0) {
4899                                         ret = -EINVAL;
4900                                 } else {
4901                                         tp->snd_cwnd_clamp = val;
4902                                         tp->snd_ssthresh = val;
4903                                 }
4904                                 break;
4905                         case TCP_BPF_DELACK_MAX:
4906                                 timeout = usecs_to_jiffies(val);
4907                                 if (timeout > TCP_DELACK_MAX ||
4908                                     timeout < TCP_TIMEOUT_MIN)
4909                                         return -EINVAL;
4910                                 inet_csk(sk)->icsk_delack_max = timeout;
4911                                 break;
4912                         case TCP_BPF_RTO_MIN:
4913                                 timeout = usecs_to_jiffies(val);
4914                                 if (timeout > TCP_RTO_MIN ||
4915                                     timeout < TCP_TIMEOUT_MIN)
4916                                         return -EINVAL;
4917                                 inet_csk(sk)->icsk_rto_min = timeout;
4918                                 break;
4919                         case TCP_SAVE_SYN:
4920                                 if (val < 0 || val > 1)
4921                                         ret = -EINVAL;
4922                                 else
4923                                         tp->save_syn = val;
4924                                 break;
4925                         case TCP_KEEPIDLE:
4926                                 ret = tcp_sock_set_keepidle_locked(sk, val);
4927                                 break;
4928                         case TCP_KEEPINTVL:
4929                                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4930                                         ret = -EINVAL;
4931                                 else
4932                                         tp->keepalive_intvl = val * HZ;
4933                                 break;
4934                         case TCP_KEEPCNT:
4935                                 if (val < 1 || val > MAX_TCP_KEEPCNT)
4936                                         ret = -EINVAL;
4937                                 else
4938                                         tp->keepalive_probes = val;
4939                                 break;
4940                         case TCP_SYNCNT:
4941                                 if (val < 1 || val > MAX_TCP_SYNCNT)
4942                                         ret = -EINVAL;
4943                                 else
4944                                         icsk->icsk_syn_retries = val;
4945                                 break;
4946                         case TCP_USER_TIMEOUT:
4947                                 if (val < 0)
4948                                         ret = -EINVAL;
4949                                 else
4950                                         icsk->icsk_user_timeout = val;
4951                                 break;
4952                         case TCP_NOTSENT_LOWAT:
4953                                 tp->notsent_lowat = val;
4954                                 sk->sk_write_space(sk);
4955                                 break;
4956                         case TCP_WINDOW_CLAMP:
4957                                 ret = tcp_set_window_clamp(sk, val);
4958                                 break;
4959                         default:
4960                                 ret = -EINVAL;
4961                         }
4962                 }
4963 #endif
4964         } else {
4965                 ret = -EINVAL;
4966         }
4967         return ret;
4968 }
4969
4970 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4971                            char *optval, int optlen)
4972 {
4973         if (!sk_fullsock(sk))
4974                 goto err_clear;
4975
4976         sock_owned_by_me(sk);
4977
4978         if (level == SOL_SOCKET) {
4979                 if (optlen != sizeof(int))
4980                         goto err_clear;
4981
4982                 switch (optname) {
4983                 case SO_MARK:
4984                         *((int *)optval) = sk->sk_mark;
4985                         break;
4986                 case SO_PRIORITY:
4987                         *((int *)optval) = sk->sk_priority;
4988                         break;
4989                 case SO_BINDTOIFINDEX:
4990                         *((int *)optval) = sk->sk_bound_dev_if;
4991                         break;
4992                 case SO_REUSEPORT:
4993                         *((int *)optval) = sk->sk_reuseport;
4994                         break;
4995                 default:
4996                         goto err_clear;
4997                 }
4998 #ifdef CONFIG_INET
4999         } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
5000                 struct inet_connection_sock *icsk;
5001                 struct tcp_sock *tp;
5002
5003                 switch (optname) {
5004                 case TCP_CONGESTION:
5005                         icsk = inet_csk(sk);
5006
5007                         if (!icsk->icsk_ca_ops || optlen <= 1)
5008                                 goto err_clear;
5009                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5010                         optval[optlen - 1] = 0;
5011                         break;
5012                 case TCP_SAVED_SYN:
5013                         tp = tcp_sk(sk);
5014
5015                         if (optlen <= 0 || !tp->saved_syn ||
5016                             optlen > tcp_saved_syn_len(tp->saved_syn))
5017                                 goto err_clear;
5018                         memcpy(optval, tp->saved_syn->data, optlen);
5019                         break;
5020                 default:
5021                         goto err_clear;
5022                 }
5023         } else if (level == SOL_IP) {
5024                 struct inet_sock *inet = inet_sk(sk);
5025
5026                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5027                         goto err_clear;
5028
5029                 /* Only some options are supported */
5030                 switch (optname) {
5031                 case IP_TOS:
5032                         *((int *)optval) = (int)inet->tos;
5033                         break;
5034                 default:
5035                         goto err_clear;
5036                 }
5037 #if IS_ENABLED(CONFIG_IPV6)
5038         } else if (level == SOL_IPV6) {
5039                 struct ipv6_pinfo *np = inet6_sk(sk);
5040
5041                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5042                         goto err_clear;
5043
5044                 /* Only some options are supported */
5045                 switch (optname) {
5046                 case IPV6_TCLASS:
5047                         *((int *)optval) = (int)np->tclass;
5048                         break;
5049                 default:
5050                         goto err_clear;
5051                 }
5052 #endif
5053 #endif
5054         } else {
5055                 goto err_clear;
5056         }
5057         return 0;
5058 err_clear:
5059         memset(optval, 0, optlen);
5060         return -EINVAL;
5061 }
5062
5063 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5064            int, optname, char *, optval, int, optlen)
5065 {
5066         if (level == SOL_TCP && optname == TCP_CONGESTION) {
5067                 if (optlen >= sizeof("cdg") - 1 &&
5068                     !strncmp("cdg", optval, optlen))
5069                         return -ENOTSUPP;
5070         }
5071
5072         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5073 }
5074
5075 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5076         .func           = bpf_sk_setsockopt,
5077         .gpl_only       = false,
5078         .ret_type       = RET_INTEGER,
5079         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5080         .arg2_type      = ARG_ANYTHING,
5081         .arg3_type      = ARG_ANYTHING,
5082         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5083         .arg5_type      = ARG_CONST_SIZE,
5084 };
5085
5086 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5087            int, optname, char *, optval, int, optlen)
5088 {
5089         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5090 }
5091
5092 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5093         .func           = bpf_sk_getsockopt,
5094         .gpl_only       = false,
5095         .ret_type       = RET_INTEGER,
5096         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5097         .arg2_type      = ARG_ANYTHING,
5098         .arg3_type      = ARG_ANYTHING,
5099         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5100         .arg5_type      = ARG_CONST_SIZE,
5101 };
5102
5103 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5104            int, level, int, optname, char *, optval, int, optlen)
5105 {
5106         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5107 }
5108
5109 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5110         .func           = bpf_sock_addr_setsockopt,
5111         .gpl_only       = false,
5112         .ret_type       = RET_INTEGER,
5113         .arg1_type      = ARG_PTR_TO_CTX,
5114         .arg2_type      = ARG_ANYTHING,
5115         .arg3_type      = ARG_ANYTHING,
5116         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5117         .arg5_type      = ARG_CONST_SIZE,
5118 };
5119
5120 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5121            int, level, int, optname, char *, optval, int, optlen)
5122 {
5123         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5124 }
5125
5126 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5127         .func           = bpf_sock_addr_getsockopt,
5128         .gpl_only       = false,
5129         .ret_type       = RET_INTEGER,
5130         .arg1_type      = ARG_PTR_TO_CTX,
5131         .arg2_type      = ARG_ANYTHING,
5132         .arg3_type      = ARG_ANYTHING,
5133         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5134         .arg5_type      = ARG_CONST_SIZE,
5135 };
5136
5137 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5138            int, level, int, optname, char *, optval, int, optlen)
5139 {
5140         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5141 }
5142
5143 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5144         .func           = bpf_sock_ops_setsockopt,
5145         .gpl_only       = false,
5146         .ret_type       = RET_INTEGER,
5147         .arg1_type      = ARG_PTR_TO_CTX,
5148         .arg2_type      = ARG_ANYTHING,
5149         .arg3_type      = ARG_ANYTHING,
5150         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5151         .arg5_type      = ARG_CONST_SIZE,
5152 };
5153
5154 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5155                                 int optname, const u8 **start)
5156 {
5157         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5158         const u8 *hdr_start;
5159         int ret;
5160
5161         if (syn_skb) {
5162                 /* sk is a request_sock here */
5163
5164                 if (optname == TCP_BPF_SYN) {
5165                         hdr_start = syn_skb->data;
5166                         ret = tcp_hdrlen(syn_skb);
5167                 } else if (optname == TCP_BPF_SYN_IP) {
5168                         hdr_start = skb_network_header(syn_skb);
5169                         ret = skb_network_header_len(syn_skb) +
5170                                 tcp_hdrlen(syn_skb);
5171                 } else {
5172                         /* optname == TCP_BPF_SYN_MAC */
5173                         hdr_start = skb_mac_header(syn_skb);
5174                         ret = skb_mac_header_len(syn_skb) +
5175                                 skb_network_header_len(syn_skb) +
5176                                 tcp_hdrlen(syn_skb);
5177                 }
5178         } else {
5179                 struct sock *sk = bpf_sock->sk;
5180                 struct saved_syn *saved_syn;
5181
5182                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5183                         /* synack retransmit. bpf_sock->syn_skb will
5184                          * not be available.  It has to resort to
5185                          * saved_syn (if it is saved).
5186                          */
5187                         saved_syn = inet_reqsk(sk)->saved_syn;
5188                 else
5189                         saved_syn = tcp_sk(sk)->saved_syn;
5190
5191                 if (!saved_syn)
5192                         return -ENOENT;
5193
5194                 if (optname == TCP_BPF_SYN) {
5195                         hdr_start = saved_syn->data +
5196                                 saved_syn->mac_hdrlen +
5197                                 saved_syn->network_hdrlen;
5198                         ret = saved_syn->tcp_hdrlen;
5199                 } else if (optname == TCP_BPF_SYN_IP) {
5200                         hdr_start = saved_syn->data +
5201                                 saved_syn->mac_hdrlen;
5202                         ret = saved_syn->network_hdrlen +
5203                                 saved_syn->tcp_hdrlen;
5204                 } else {
5205                         /* optname == TCP_BPF_SYN_MAC */
5206
5207                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5208                         if (!saved_syn->mac_hdrlen)
5209                                 return -ENOENT;
5210
5211                         hdr_start = saved_syn->data;
5212                         ret = saved_syn->mac_hdrlen +
5213                                 saved_syn->network_hdrlen +
5214                                 saved_syn->tcp_hdrlen;
5215                 }
5216         }
5217
5218         *start = hdr_start;
5219         return ret;
5220 }
5221
5222 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5223            int, level, int, optname, char *, optval, int, optlen)
5224 {
5225         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5226             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5227                 int ret, copy_len = 0;
5228                 const u8 *start;
5229
5230                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5231                 if (ret > 0) {
5232                         copy_len = ret;
5233                         if (optlen < copy_len) {
5234                                 copy_len = optlen;
5235                                 ret = -ENOSPC;
5236                         }
5237
5238                         memcpy(optval, start, copy_len);
5239                 }
5240
5241                 /* Zero out unused buffer at the end */
5242                 memset(optval + copy_len, 0, optlen - copy_len);
5243
5244                 return ret;
5245         }
5246
5247         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5248 }
5249
5250 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5251         .func           = bpf_sock_ops_getsockopt,
5252         .gpl_only       = false,
5253         .ret_type       = RET_INTEGER,
5254         .arg1_type      = ARG_PTR_TO_CTX,
5255         .arg2_type      = ARG_ANYTHING,
5256         .arg3_type      = ARG_ANYTHING,
5257         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5258         .arg5_type      = ARG_CONST_SIZE,
5259 };
5260
5261 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5262            int, argval)
5263 {
5264         struct sock *sk = bpf_sock->sk;
5265         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5266
5267         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5268                 return -EINVAL;
5269
5270         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5271
5272         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5273 }
5274
5275 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5276         .func           = bpf_sock_ops_cb_flags_set,
5277         .gpl_only       = false,
5278         .ret_type       = RET_INTEGER,
5279         .arg1_type      = ARG_PTR_TO_CTX,
5280         .arg2_type      = ARG_ANYTHING,
5281 };
5282
5283 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5284 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5285
5286 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5287            int, addr_len)
5288 {
5289 #ifdef CONFIG_INET
5290         struct sock *sk = ctx->sk;
5291         u32 flags = BIND_FROM_BPF;
5292         int err;
5293
5294         err = -EINVAL;
5295         if (addr_len < offsetofend(struct sockaddr, sa_family))
5296                 return err;
5297         if (addr->sa_family == AF_INET) {
5298                 if (addr_len < sizeof(struct sockaddr_in))
5299                         return err;
5300                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5301                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5302                 return __inet_bind(sk, addr, addr_len, flags);
5303 #if IS_ENABLED(CONFIG_IPV6)
5304         } else if (addr->sa_family == AF_INET6) {
5305                 if (addr_len < SIN6_LEN_RFC2133)
5306                         return err;
5307                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5308                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5309                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5310                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5311                  */
5312                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5313 #endif /* CONFIG_IPV6 */
5314         }
5315 #endif /* CONFIG_INET */
5316
5317         return -EAFNOSUPPORT;
5318 }
5319
5320 static const struct bpf_func_proto bpf_bind_proto = {
5321         .func           = bpf_bind,
5322         .gpl_only       = false,
5323         .ret_type       = RET_INTEGER,
5324         .arg1_type      = ARG_PTR_TO_CTX,
5325         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5326         .arg3_type      = ARG_CONST_SIZE,
5327 };
5328
5329 #ifdef CONFIG_XFRM
5330 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5331            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5332 {
5333         const struct sec_path *sp = skb_sec_path(skb);
5334         const struct xfrm_state *x;
5335
5336         if (!sp || unlikely(index >= sp->len || flags))
5337                 goto err_clear;
5338
5339         x = sp->xvec[index];
5340
5341         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5342                 goto err_clear;
5343
5344         to->reqid = x->props.reqid;
5345         to->spi = x->id.spi;
5346         to->family = x->props.family;
5347         to->ext = 0;
5348
5349         if (to->family == AF_INET6) {
5350                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5351                        sizeof(to->remote_ipv6));
5352         } else {
5353                 to->remote_ipv4 = x->props.saddr.a4;
5354                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5355         }
5356
5357         return 0;
5358 err_clear:
5359         memset(to, 0, size);
5360         return -EINVAL;
5361 }
5362
5363 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5364         .func           = bpf_skb_get_xfrm_state,
5365         .gpl_only       = false,
5366         .ret_type       = RET_INTEGER,
5367         .arg1_type      = ARG_PTR_TO_CTX,
5368         .arg2_type      = ARG_ANYTHING,
5369         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5370         .arg4_type      = ARG_CONST_SIZE,
5371         .arg5_type      = ARG_ANYTHING,
5372 };
5373 #endif
5374
5375 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5376 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5377                                   const struct neighbour *neigh,
5378                                   const struct net_device *dev, u32 mtu)
5379 {
5380         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5381         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5382         params->h_vlan_TCI = 0;
5383         params->h_vlan_proto = 0;
5384         if (mtu)
5385                 params->mtu_result = mtu; /* union with tot_len */
5386
5387         return 0;
5388 }
5389 #endif
5390
5391 #if IS_ENABLED(CONFIG_INET)
5392 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5393                                u32 flags, bool check_mtu)
5394 {
5395         struct fib_nh_common *nhc;
5396         struct in_device *in_dev;
5397         struct neighbour *neigh;
5398         struct net_device *dev;
5399         struct fib_result res;
5400         struct flowi4 fl4;
5401         u32 mtu = 0;
5402         int err;
5403
5404         dev = dev_get_by_index_rcu(net, params->ifindex);
5405         if (unlikely(!dev))
5406                 return -ENODEV;
5407
5408         /* verify forwarding is enabled on this interface */
5409         in_dev = __in_dev_get_rcu(dev);
5410         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5411                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5412
5413         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5414                 fl4.flowi4_iif = 1;
5415                 fl4.flowi4_oif = params->ifindex;
5416         } else {
5417                 fl4.flowi4_iif = params->ifindex;
5418                 fl4.flowi4_oif = 0;
5419         }
5420         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5421         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5422         fl4.flowi4_flags = 0;
5423
5424         fl4.flowi4_proto = params->l4_protocol;
5425         fl4.daddr = params->ipv4_dst;
5426         fl4.saddr = params->ipv4_src;
5427         fl4.fl4_sport = params->sport;
5428         fl4.fl4_dport = params->dport;
5429         fl4.flowi4_multipath_hash = 0;
5430
5431         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5432                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5433                 struct fib_table *tb;
5434
5435                 tb = fib_get_table(net, tbid);
5436                 if (unlikely(!tb))
5437                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5438
5439                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5440         } else {
5441                 fl4.flowi4_mark = 0;
5442                 fl4.flowi4_secid = 0;
5443                 fl4.flowi4_tun_key.tun_id = 0;
5444                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5445
5446                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5447         }
5448
5449         if (err) {
5450                 /* map fib lookup errors to RTN_ type */
5451                 if (err == -EINVAL)
5452                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5453                 if (err == -EHOSTUNREACH)
5454                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5455                 if (err == -EACCES)
5456                         return BPF_FIB_LKUP_RET_PROHIBIT;
5457
5458                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5459         }
5460
5461         if (res.type != RTN_UNICAST)
5462                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5463
5464         if (fib_info_num_path(res.fi) > 1)
5465                 fib_select_path(net, &res, &fl4, NULL);
5466
5467         if (check_mtu) {
5468                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5469                 if (params->tot_len > mtu) {
5470                         params->mtu_result = mtu; /* union with tot_len */
5471                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5472                 }
5473         }
5474
5475         nhc = res.nhc;
5476
5477         /* do not handle lwt encaps right now */
5478         if (nhc->nhc_lwtstate)
5479                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5480
5481         dev = nhc->nhc_dev;
5482
5483         params->rt_metric = res.fi->fib_priority;
5484         params->ifindex = dev->ifindex;
5485
5486         /* xdp and cls_bpf programs are run in RCU-bh so
5487          * rcu_read_lock_bh is not needed here
5488          */
5489         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5490                 if (nhc->nhc_gw_family)
5491                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5492
5493                 neigh = __ipv4_neigh_lookup_noref(dev,
5494                                                  (__force u32)params->ipv4_dst);
5495         } else {
5496                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5497
5498                 params->family = AF_INET6;
5499                 *dst = nhc->nhc_gw.ipv6;
5500                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5501         }
5502
5503         if (!neigh || !(neigh->nud_state & NUD_VALID))
5504                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5505
5506         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5507 }
5508 #endif
5509
5510 #if IS_ENABLED(CONFIG_IPV6)
5511 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5512                                u32 flags, bool check_mtu)
5513 {
5514         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5515         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5516         struct fib6_result res = {};
5517         struct neighbour *neigh;
5518         struct net_device *dev;
5519         struct inet6_dev *idev;
5520         struct flowi6 fl6;
5521         int strict = 0;
5522         int oif, err;
5523         u32 mtu = 0;
5524
5525         /* link local addresses are never forwarded */
5526         if (rt6_need_strict(dst) || rt6_need_strict(src))
5527                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5528
5529         dev = dev_get_by_index_rcu(net, params->ifindex);
5530         if (unlikely(!dev))
5531                 return -ENODEV;
5532
5533         idev = __in6_dev_get_safely(dev);
5534         if (unlikely(!idev || !idev->cnf.forwarding))
5535                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5536
5537         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5538                 fl6.flowi6_iif = 1;
5539                 oif = fl6.flowi6_oif = params->ifindex;
5540         } else {
5541                 oif = fl6.flowi6_iif = params->ifindex;
5542                 fl6.flowi6_oif = 0;
5543                 strict = RT6_LOOKUP_F_HAS_SADDR;
5544         }
5545         fl6.flowlabel = params->flowinfo;
5546         fl6.flowi6_scope = 0;
5547         fl6.flowi6_flags = 0;
5548         fl6.mp_hash = 0;
5549
5550         fl6.flowi6_proto = params->l4_protocol;
5551         fl6.daddr = *dst;
5552         fl6.saddr = *src;
5553         fl6.fl6_sport = params->sport;
5554         fl6.fl6_dport = params->dport;
5555
5556         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5557                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5558                 struct fib6_table *tb;
5559
5560                 tb = ipv6_stub->fib6_get_table(net, tbid);
5561                 if (unlikely(!tb))
5562                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5563
5564                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5565                                                    strict);
5566         } else {
5567                 fl6.flowi6_mark = 0;
5568                 fl6.flowi6_secid = 0;
5569                 fl6.flowi6_tun_key.tun_id = 0;
5570                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5571
5572                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5573         }
5574
5575         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5576                      res.f6i == net->ipv6.fib6_null_entry))
5577                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5578
5579         switch (res.fib6_type) {
5580         /* only unicast is forwarded */
5581         case RTN_UNICAST:
5582                 break;
5583         case RTN_BLACKHOLE:
5584                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5585         case RTN_UNREACHABLE:
5586                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5587         case RTN_PROHIBIT:
5588                 return BPF_FIB_LKUP_RET_PROHIBIT;
5589         default:
5590                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5591         }
5592
5593         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5594                                     fl6.flowi6_oif != 0, NULL, strict);
5595
5596         if (check_mtu) {
5597                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5598                 if (params->tot_len > mtu) {
5599                         params->mtu_result = mtu; /* union with tot_len */
5600                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5601                 }
5602         }
5603
5604         if (res.nh->fib_nh_lws)
5605                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5606
5607         if (res.nh->fib_nh_gw_family)
5608                 *dst = res.nh->fib_nh_gw6;
5609
5610         dev = res.nh->fib_nh_dev;
5611         params->rt_metric = res.f6i->fib6_metric;
5612         params->ifindex = dev->ifindex;
5613
5614         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5615          * not needed here.
5616          */
5617         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5618         if (!neigh || !(neigh->nud_state & NUD_VALID))
5619                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5620
5621         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5622 }
5623 #endif
5624
5625 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5626            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5627 {
5628         if (plen < sizeof(*params))
5629                 return -EINVAL;
5630
5631         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5632                 return -EINVAL;
5633
5634         switch (params->family) {
5635 #if IS_ENABLED(CONFIG_INET)
5636         case AF_INET:
5637                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5638                                            flags, true);
5639 #endif
5640 #if IS_ENABLED(CONFIG_IPV6)
5641         case AF_INET6:
5642                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5643                                            flags, true);
5644 #endif
5645         }
5646         return -EAFNOSUPPORT;
5647 }
5648
5649 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5650         .func           = bpf_xdp_fib_lookup,
5651         .gpl_only       = true,
5652         .ret_type       = RET_INTEGER,
5653         .arg1_type      = ARG_PTR_TO_CTX,
5654         .arg2_type      = ARG_PTR_TO_MEM,
5655         .arg3_type      = ARG_CONST_SIZE,
5656         .arg4_type      = ARG_ANYTHING,
5657 };
5658
5659 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5660            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5661 {
5662         struct net *net = dev_net(skb->dev);
5663         int rc = -EAFNOSUPPORT;
5664         bool check_mtu = false;
5665
5666         if (plen < sizeof(*params))
5667                 return -EINVAL;
5668
5669         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5670                 return -EINVAL;
5671
5672         if (params->tot_len)
5673                 check_mtu = true;
5674
5675         switch (params->family) {
5676 #if IS_ENABLED(CONFIG_INET)
5677         case AF_INET:
5678                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5679                 break;
5680 #endif
5681 #if IS_ENABLED(CONFIG_IPV6)
5682         case AF_INET6:
5683                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5684                 break;
5685 #endif
5686         }
5687
5688         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5689                 struct net_device *dev;
5690
5691                 /* When tot_len isn't provided by user, check skb
5692                  * against MTU of FIB lookup resulting net_device
5693                  */
5694                 dev = dev_get_by_index_rcu(net, params->ifindex);
5695                 if (!is_skb_forwardable(dev, skb))
5696                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5697
5698                 params->mtu_result = dev->mtu; /* union with tot_len */
5699         }
5700
5701         return rc;
5702 }
5703
5704 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5705         .func           = bpf_skb_fib_lookup,
5706         .gpl_only       = true,
5707         .ret_type       = RET_INTEGER,
5708         .arg1_type      = ARG_PTR_TO_CTX,
5709         .arg2_type      = ARG_PTR_TO_MEM,
5710         .arg3_type      = ARG_CONST_SIZE,
5711         .arg4_type      = ARG_ANYTHING,
5712 };
5713
5714 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5715                                             u32 ifindex)
5716 {
5717         struct net *netns = dev_net(dev_curr);
5718
5719         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5720         if (ifindex == 0)
5721                 return dev_curr;
5722
5723         return dev_get_by_index_rcu(netns, ifindex);
5724 }
5725
5726 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5727            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5728 {
5729         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5730         struct net_device *dev = skb->dev;
5731         int skb_len, dev_len;
5732         int mtu;
5733
5734         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5735                 return -EINVAL;
5736
5737         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5738                 return -EINVAL;
5739
5740         dev = __dev_via_ifindex(dev, ifindex);
5741         if (unlikely(!dev))
5742                 return -ENODEV;
5743
5744         mtu = READ_ONCE(dev->mtu);
5745
5746         dev_len = mtu + dev->hard_header_len;
5747
5748         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5749         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5750
5751         skb_len += len_diff; /* minus result pass check */
5752         if (skb_len <= dev_len) {
5753                 ret = BPF_MTU_CHK_RET_SUCCESS;
5754                 goto out;
5755         }
5756         /* At this point, skb->len exceed MTU, but as it include length of all
5757          * segments, it can still be below MTU.  The SKB can possibly get
5758          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
5759          * must choose if segs are to be MTU checked.
5760          */
5761         if (skb_is_gso(skb)) {
5762                 ret = BPF_MTU_CHK_RET_SUCCESS;
5763
5764                 if (flags & BPF_MTU_CHK_SEGS &&
5765                     !skb_gso_validate_network_len(skb, mtu))
5766                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5767         }
5768 out:
5769         /* BPF verifier guarantees valid pointer */
5770         *mtu_len = mtu;
5771
5772         return ret;
5773 }
5774
5775 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5776            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5777 {
5778         struct net_device *dev = xdp->rxq->dev;
5779         int xdp_len = xdp->data_end - xdp->data;
5780         int ret = BPF_MTU_CHK_RET_SUCCESS;
5781         int mtu, dev_len;
5782
5783         /* XDP variant doesn't support multi-buffer segment check (yet) */
5784         if (unlikely(flags))
5785                 return -EINVAL;
5786
5787         dev = __dev_via_ifindex(dev, ifindex);
5788         if (unlikely(!dev))
5789                 return -ENODEV;
5790
5791         mtu = READ_ONCE(dev->mtu);
5792
5793         /* Add L2-header as dev MTU is L3 size */
5794         dev_len = mtu + dev->hard_header_len;
5795
5796         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5797         if (*mtu_len)
5798                 xdp_len = *mtu_len + dev->hard_header_len;
5799
5800         xdp_len += len_diff; /* minus result pass check */
5801         if (xdp_len > dev_len)
5802                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5803
5804         /* BPF verifier guarantees valid pointer */
5805         *mtu_len = mtu;
5806
5807         return ret;
5808 }
5809
5810 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5811         .func           = bpf_skb_check_mtu,
5812         .gpl_only       = true,
5813         .ret_type       = RET_INTEGER,
5814         .arg1_type      = ARG_PTR_TO_CTX,
5815         .arg2_type      = ARG_ANYTHING,
5816         .arg3_type      = ARG_PTR_TO_INT,
5817         .arg4_type      = ARG_ANYTHING,
5818         .arg5_type      = ARG_ANYTHING,
5819 };
5820
5821 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5822         .func           = bpf_xdp_check_mtu,
5823         .gpl_only       = true,
5824         .ret_type       = RET_INTEGER,
5825         .arg1_type      = ARG_PTR_TO_CTX,
5826         .arg2_type      = ARG_ANYTHING,
5827         .arg3_type      = ARG_PTR_TO_INT,
5828         .arg4_type      = ARG_ANYTHING,
5829         .arg5_type      = ARG_ANYTHING,
5830 };
5831
5832 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5833 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5834 {
5835         int err;
5836         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5837
5838         if (!seg6_validate_srh(srh, len, false))
5839                 return -EINVAL;
5840
5841         switch (type) {
5842         case BPF_LWT_ENCAP_SEG6_INLINE:
5843                 if (skb->protocol != htons(ETH_P_IPV6))
5844                         return -EBADMSG;
5845
5846                 err = seg6_do_srh_inline(skb, srh);
5847                 break;
5848         case BPF_LWT_ENCAP_SEG6:
5849                 skb_reset_inner_headers(skb);
5850                 skb->encapsulation = 1;
5851                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5852                 break;
5853         default:
5854                 return -EINVAL;
5855         }
5856
5857         bpf_compute_data_pointers(skb);
5858         if (err)
5859                 return err;
5860
5861         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5862
5863         return seg6_lookup_nexthop(skb, NULL, 0);
5864 }
5865 #endif /* CONFIG_IPV6_SEG6_BPF */
5866
5867 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5868 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5869                              bool ingress)
5870 {
5871         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5872 }
5873 #endif
5874
5875 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5876            u32, len)
5877 {
5878         switch (type) {
5879 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5880         case BPF_LWT_ENCAP_SEG6:
5881         case BPF_LWT_ENCAP_SEG6_INLINE:
5882                 return bpf_push_seg6_encap(skb, type, hdr, len);
5883 #endif
5884 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5885         case BPF_LWT_ENCAP_IP:
5886                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5887 #endif
5888         default:
5889                 return -EINVAL;
5890         }
5891 }
5892
5893 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5894            void *, hdr, u32, len)
5895 {
5896         switch (type) {
5897 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5898         case BPF_LWT_ENCAP_IP:
5899                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5900 #endif
5901         default:
5902                 return -EINVAL;
5903         }
5904 }
5905
5906 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5907         .func           = bpf_lwt_in_push_encap,
5908         .gpl_only       = false,
5909         .ret_type       = RET_INTEGER,
5910         .arg1_type      = ARG_PTR_TO_CTX,
5911         .arg2_type      = ARG_ANYTHING,
5912         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5913         .arg4_type      = ARG_CONST_SIZE
5914 };
5915
5916 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5917         .func           = bpf_lwt_xmit_push_encap,
5918         .gpl_only       = false,
5919         .ret_type       = RET_INTEGER,
5920         .arg1_type      = ARG_PTR_TO_CTX,
5921         .arg2_type      = ARG_ANYTHING,
5922         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5923         .arg4_type      = ARG_CONST_SIZE
5924 };
5925
5926 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5927 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5928            const void *, from, u32, len)
5929 {
5930         struct seg6_bpf_srh_state *srh_state =
5931                 this_cpu_ptr(&seg6_bpf_srh_states);
5932         struct ipv6_sr_hdr *srh = srh_state->srh;
5933         void *srh_tlvs, *srh_end, *ptr;
5934         int srhoff = 0;
5935
5936         if (srh == NULL)
5937                 return -EINVAL;
5938
5939         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5940         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5941
5942         ptr = skb->data + offset;
5943         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5944                 srh_state->valid = false;
5945         else if (ptr < (void *)&srh->flags ||
5946                  ptr + len > (void *)&srh->segments)
5947                 return -EFAULT;
5948
5949         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5950                 return -EFAULT;
5951         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5952                 return -EINVAL;
5953         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5954
5955         memcpy(skb->data + offset, from, len);
5956         return 0;
5957 }
5958
5959 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5960         .func           = bpf_lwt_seg6_store_bytes,
5961         .gpl_only       = false,
5962         .ret_type       = RET_INTEGER,
5963         .arg1_type      = ARG_PTR_TO_CTX,
5964         .arg2_type      = ARG_ANYTHING,
5965         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5966         .arg4_type      = ARG_CONST_SIZE
5967 };
5968
5969 static void bpf_update_srh_state(struct sk_buff *skb)
5970 {
5971         struct seg6_bpf_srh_state *srh_state =
5972                 this_cpu_ptr(&seg6_bpf_srh_states);
5973         int srhoff = 0;
5974
5975         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5976                 srh_state->srh = NULL;
5977         } else {
5978                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5979                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5980                 srh_state->valid = true;
5981         }
5982 }
5983
5984 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5985            u32, action, void *, param, u32, param_len)
5986 {
5987         struct seg6_bpf_srh_state *srh_state =
5988                 this_cpu_ptr(&seg6_bpf_srh_states);
5989         int hdroff = 0;
5990         int err;
5991
5992         switch (action) {
5993         case SEG6_LOCAL_ACTION_END_X:
5994                 if (!seg6_bpf_has_valid_srh(skb))
5995                         return -EBADMSG;
5996                 if (param_len != sizeof(struct in6_addr))
5997                         return -EINVAL;
5998                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5999         case SEG6_LOCAL_ACTION_END_T:
6000                 if (!seg6_bpf_has_valid_srh(skb))
6001                         return -EBADMSG;
6002                 if (param_len != sizeof(int))
6003                         return -EINVAL;
6004                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6005         case SEG6_LOCAL_ACTION_END_DT6:
6006                 if (!seg6_bpf_has_valid_srh(skb))
6007                         return -EBADMSG;
6008                 if (param_len != sizeof(int))
6009                         return -EINVAL;
6010
6011                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6012                         return -EBADMSG;
6013                 if (!pskb_pull(skb, hdroff))
6014                         return -EBADMSG;
6015
6016                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6017                 skb_reset_network_header(skb);
6018                 skb_reset_transport_header(skb);
6019                 skb->encapsulation = 0;
6020
6021                 bpf_compute_data_pointers(skb);
6022                 bpf_update_srh_state(skb);
6023                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6024         case SEG6_LOCAL_ACTION_END_B6:
6025                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6026                         return -EBADMSG;
6027                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6028                                           param, param_len);
6029                 if (!err)
6030                         bpf_update_srh_state(skb);
6031
6032                 return err;
6033         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6034                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6035                         return -EBADMSG;
6036                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6037                                           param, param_len);
6038                 if (!err)
6039                         bpf_update_srh_state(skb);
6040
6041                 return err;
6042         default:
6043                 return -EINVAL;
6044         }
6045 }
6046
6047 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6048         .func           = bpf_lwt_seg6_action,
6049         .gpl_only       = false,
6050         .ret_type       = RET_INTEGER,
6051         .arg1_type      = ARG_PTR_TO_CTX,
6052         .arg2_type      = ARG_ANYTHING,
6053         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6054         .arg4_type      = ARG_CONST_SIZE
6055 };
6056
6057 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6058            s32, len)
6059 {
6060         struct seg6_bpf_srh_state *srh_state =
6061                 this_cpu_ptr(&seg6_bpf_srh_states);
6062         struct ipv6_sr_hdr *srh = srh_state->srh;
6063         void *srh_end, *srh_tlvs, *ptr;
6064         struct ipv6hdr *hdr;
6065         int srhoff = 0;
6066         int ret;
6067
6068         if (unlikely(srh == NULL))
6069                 return -EINVAL;
6070
6071         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6072                         ((srh->first_segment + 1) << 4));
6073         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6074                         srh_state->hdrlen);
6075         ptr = skb->data + offset;
6076
6077         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6078                 return -EFAULT;
6079         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6080                 return -EFAULT;
6081
6082         if (len > 0) {
6083                 ret = skb_cow_head(skb, len);
6084                 if (unlikely(ret < 0))
6085                         return ret;
6086
6087                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6088         } else {
6089                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6090         }
6091
6092         bpf_compute_data_pointers(skb);
6093         if (unlikely(ret < 0))
6094                 return ret;
6095
6096         hdr = (struct ipv6hdr *)skb->data;
6097         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6098
6099         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6100                 return -EINVAL;
6101         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6102         srh_state->hdrlen += len;
6103         srh_state->valid = false;
6104         return 0;
6105 }
6106
6107 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6108         .func           = bpf_lwt_seg6_adjust_srh,
6109         .gpl_only       = false,
6110         .ret_type       = RET_INTEGER,
6111         .arg1_type      = ARG_PTR_TO_CTX,
6112         .arg2_type      = ARG_ANYTHING,
6113         .arg3_type      = ARG_ANYTHING,
6114 };
6115 #endif /* CONFIG_IPV6_SEG6_BPF */
6116
6117 #ifdef CONFIG_INET
6118 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6119                               int dif, int sdif, u8 family, u8 proto)
6120 {
6121         bool refcounted = false;
6122         struct sock *sk = NULL;
6123
6124         if (family == AF_INET) {
6125                 __be32 src4 = tuple->ipv4.saddr;
6126                 __be32 dst4 = tuple->ipv4.daddr;
6127
6128                 if (proto == IPPROTO_TCP)
6129                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6130                                            src4, tuple->ipv4.sport,
6131                                            dst4, tuple->ipv4.dport,
6132                                            dif, sdif, &refcounted);
6133                 else
6134                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6135                                                dst4, tuple->ipv4.dport,
6136                                                dif, sdif, &udp_table, NULL);
6137 #if IS_ENABLED(CONFIG_IPV6)
6138         } else {
6139                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6140                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6141
6142                 if (proto == IPPROTO_TCP)
6143                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6144                                             src6, tuple->ipv6.sport,
6145                                             dst6, ntohs(tuple->ipv6.dport),
6146                                             dif, sdif, &refcounted);
6147                 else if (likely(ipv6_bpf_stub))
6148                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6149                                                             src6, tuple->ipv6.sport,
6150                                                             dst6, tuple->ipv6.dport,
6151                                                             dif, sdif,
6152                                                             &udp_table, NULL);
6153 #endif
6154         }
6155
6156         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6157                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6158                 sk = NULL;
6159         }
6160         return sk;
6161 }
6162
6163 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6164  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6165  * Returns the socket as an 'unsigned long' to simplify the casting in the
6166  * callers to satisfy BPF_CALL declarations.
6167  */
6168 static struct sock *
6169 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6170                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6171                  u64 flags, int sdif)
6172 {
6173         struct sock *sk = NULL;
6174         struct net *net;
6175         u8 family;
6176
6177         if (len == sizeof(tuple->ipv4))
6178                 family = AF_INET;
6179         else if (len == sizeof(tuple->ipv6))
6180                 family = AF_INET6;
6181         else
6182                 return NULL;
6183
6184         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6185                 goto out;
6186
6187         if (sdif < 0) {
6188                 if (family == AF_INET)
6189                         sdif = inet_sdif(skb);
6190                 else
6191                         sdif = inet6_sdif(skb);
6192         }
6193
6194         if ((s32)netns_id < 0) {
6195                 net = caller_net;
6196                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6197         } else {
6198                 net = get_net_ns_by_id(caller_net, netns_id);
6199                 if (unlikely(!net))
6200                         goto out;
6201                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6202                 put_net(net);
6203         }
6204
6205 out:
6206         return sk;
6207 }
6208
6209 static struct sock *
6210 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6211                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6212                 u64 flags, int sdif)
6213 {
6214         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6215                                            ifindex, proto, netns_id, flags,
6216                                            sdif);
6217
6218         if (sk) {
6219                 struct sock *sk2 = sk_to_full_sk(sk);
6220
6221                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6222                  * sock refcnt is decremented to prevent a request_sock leak.
6223                  */
6224                 if (!sk_fullsock(sk2))
6225                         sk2 = NULL;
6226                 if (sk2 != sk) {
6227                         sock_gen_put(sk);
6228                         /* Ensure there is no need to bump sk2 refcnt */
6229                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6230                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6231                                 return NULL;
6232                         }
6233                         sk = sk2;
6234                 }
6235         }
6236
6237         return sk;
6238 }
6239
6240 static struct sock *
6241 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6242                u8 proto, u64 netns_id, u64 flags)
6243 {
6244         struct net *caller_net;
6245         int ifindex;
6246
6247         if (skb->dev) {
6248                 caller_net = dev_net(skb->dev);
6249                 ifindex = skb->dev->ifindex;
6250         } else {
6251                 caller_net = sock_net(skb->sk);
6252                 ifindex = 0;
6253         }
6254
6255         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6256                                 netns_id, flags, -1);
6257 }
6258
6259 static struct sock *
6260 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6261               u8 proto, u64 netns_id, u64 flags)
6262 {
6263         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6264                                          flags);
6265
6266         if (sk) {
6267                 struct sock *sk2 = sk_to_full_sk(sk);
6268
6269                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6270                  * sock refcnt is decremented to prevent a request_sock leak.
6271                  */
6272                 if (!sk_fullsock(sk2))
6273                         sk2 = NULL;
6274                 if (sk2 != sk) {
6275                         sock_gen_put(sk);
6276                         /* Ensure there is no need to bump sk2 refcnt */
6277                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6278                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6279                                 return NULL;
6280                         }
6281                         sk = sk2;
6282                 }
6283         }
6284
6285         return sk;
6286 }
6287
6288 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6289            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6290 {
6291         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6292                                              netns_id, flags);
6293 }
6294
6295 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6296         .func           = bpf_skc_lookup_tcp,
6297         .gpl_only       = false,
6298         .pkt_access     = true,
6299         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6300         .arg1_type      = ARG_PTR_TO_CTX,
6301         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6302         .arg3_type      = ARG_CONST_SIZE,
6303         .arg4_type      = ARG_ANYTHING,
6304         .arg5_type      = ARG_ANYTHING,
6305 };
6306
6307 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6308            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6309 {
6310         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6311                                             netns_id, flags);
6312 }
6313
6314 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6315         .func           = bpf_sk_lookup_tcp,
6316         .gpl_only       = false,
6317         .pkt_access     = true,
6318         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6319         .arg1_type      = ARG_PTR_TO_CTX,
6320         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6321         .arg3_type      = ARG_CONST_SIZE,
6322         .arg4_type      = ARG_ANYTHING,
6323         .arg5_type      = ARG_ANYTHING,
6324 };
6325
6326 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6327            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6328 {
6329         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6330                                             netns_id, flags);
6331 }
6332
6333 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6334         .func           = bpf_sk_lookup_udp,
6335         .gpl_only       = false,
6336         .pkt_access     = true,
6337         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6338         .arg1_type      = ARG_PTR_TO_CTX,
6339         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6340         .arg3_type      = ARG_CONST_SIZE,
6341         .arg4_type      = ARG_ANYTHING,
6342         .arg5_type      = ARG_ANYTHING,
6343 };
6344
6345 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6346            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6347 {
6348         struct net_device *dev = skb->dev;
6349         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6350         struct net *caller_net = dev_net(dev);
6351
6352         return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6353                                                ifindex, IPPROTO_TCP, netns_id,
6354                                                flags, sdif);
6355 }
6356
6357 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6358         .func           = bpf_tc_skc_lookup_tcp,
6359         .gpl_only       = false,
6360         .pkt_access     = true,
6361         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6362         .arg1_type      = ARG_PTR_TO_CTX,
6363         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6364         .arg3_type      = ARG_CONST_SIZE,
6365         .arg4_type      = ARG_ANYTHING,
6366         .arg5_type      = ARG_ANYTHING,
6367 };
6368
6369 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6370            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6371 {
6372         struct net_device *dev = skb->dev;
6373         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6374         struct net *caller_net = dev_net(dev);
6375
6376         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6377                                               ifindex, IPPROTO_TCP, netns_id,
6378                                               flags, sdif);
6379 }
6380
6381 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6382         .func           = bpf_tc_sk_lookup_tcp,
6383         .gpl_only       = false,
6384         .pkt_access     = true,
6385         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6386         .arg1_type      = ARG_PTR_TO_CTX,
6387         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6388         .arg3_type      = ARG_CONST_SIZE,
6389         .arg4_type      = ARG_ANYTHING,
6390         .arg5_type      = ARG_ANYTHING,
6391 };
6392
6393 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6394            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6395 {
6396         struct net_device *dev = skb->dev;
6397         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6398         struct net *caller_net = dev_net(dev);
6399
6400         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6401                                               ifindex, IPPROTO_UDP, netns_id,
6402                                               flags, sdif);
6403 }
6404
6405 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6406         .func           = bpf_tc_sk_lookup_udp,
6407         .gpl_only       = false,
6408         .pkt_access     = true,
6409         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6410         .arg1_type      = ARG_PTR_TO_CTX,
6411         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6412         .arg3_type      = ARG_CONST_SIZE,
6413         .arg4_type      = ARG_ANYTHING,
6414         .arg5_type      = ARG_ANYTHING,
6415 };
6416
6417 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6418 {
6419         if (sk && sk_is_refcounted(sk))
6420                 sock_gen_put(sk);
6421         return 0;
6422 }
6423
6424 static const struct bpf_func_proto bpf_sk_release_proto = {
6425         .func           = bpf_sk_release,
6426         .gpl_only       = false,
6427         .ret_type       = RET_INTEGER,
6428         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6429 };
6430
6431 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6432            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6433 {
6434         struct net_device *dev = ctx->rxq->dev;
6435         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6436         struct net *caller_net = dev_net(dev);
6437
6438         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6439                                               ifindex, IPPROTO_UDP, netns_id,
6440                                               flags, sdif);
6441 }
6442
6443 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6444         .func           = bpf_xdp_sk_lookup_udp,
6445         .gpl_only       = false,
6446         .pkt_access     = true,
6447         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6448         .arg1_type      = ARG_PTR_TO_CTX,
6449         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6450         .arg3_type      = ARG_CONST_SIZE,
6451         .arg4_type      = ARG_ANYTHING,
6452         .arg5_type      = ARG_ANYTHING,
6453 };
6454
6455 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6456            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6457 {
6458         struct net_device *dev = ctx->rxq->dev;
6459         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6460         struct net *caller_net = dev_net(dev);
6461
6462         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6463                                                ifindex, IPPROTO_TCP, netns_id,
6464                                                flags, sdif);
6465 }
6466
6467 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6468         .func           = bpf_xdp_skc_lookup_tcp,
6469         .gpl_only       = false,
6470         .pkt_access     = true,
6471         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6472         .arg1_type      = ARG_PTR_TO_CTX,
6473         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6474         .arg3_type      = ARG_CONST_SIZE,
6475         .arg4_type      = ARG_ANYTHING,
6476         .arg5_type      = ARG_ANYTHING,
6477 };
6478
6479 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6480            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6481 {
6482         struct net_device *dev = ctx->rxq->dev;
6483         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6484         struct net *caller_net = dev_net(dev);
6485
6486         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6487                                               ifindex, IPPROTO_TCP, netns_id,
6488                                               flags, sdif);
6489 }
6490
6491 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6492         .func           = bpf_xdp_sk_lookup_tcp,
6493         .gpl_only       = false,
6494         .pkt_access     = true,
6495         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6496         .arg1_type      = ARG_PTR_TO_CTX,
6497         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6498         .arg3_type      = ARG_CONST_SIZE,
6499         .arg4_type      = ARG_ANYTHING,
6500         .arg5_type      = ARG_ANYTHING,
6501 };
6502
6503 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6504            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6505 {
6506         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6507                                                sock_net(ctx->sk), 0,
6508                                                IPPROTO_TCP, netns_id, flags,
6509                                                -1);
6510 }
6511
6512 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6513         .func           = bpf_sock_addr_skc_lookup_tcp,
6514         .gpl_only       = false,
6515         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6516         .arg1_type      = ARG_PTR_TO_CTX,
6517         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6518         .arg3_type      = ARG_CONST_SIZE,
6519         .arg4_type      = ARG_ANYTHING,
6520         .arg5_type      = ARG_ANYTHING,
6521 };
6522
6523 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6524            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6525 {
6526         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6527                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6528                                               netns_id, flags, -1);
6529 }
6530
6531 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6532         .func           = bpf_sock_addr_sk_lookup_tcp,
6533         .gpl_only       = false,
6534         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6535         .arg1_type      = ARG_PTR_TO_CTX,
6536         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6537         .arg3_type      = ARG_CONST_SIZE,
6538         .arg4_type      = ARG_ANYTHING,
6539         .arg5_type      = ARG_ANYTHING,
6540 };
6541
6542 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6543            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6544 {
6545         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6546                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6547                                               netns_id, flags, -1);
6548 }
6549
6550 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6551         .func           = bpf_sock_addr_sk_lookup_udp,
6552         .gpl_only       = false,
6553         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6554         .arg1_type      = ARG_PTR_TO_CTX,
6555         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6556         .arg3_type      = ARG_CONST_SIZE,
6557         .arg4_type      = ARG_ANYTHING,
6558         .arg5_type      = ARG_ANYTHING,
6559 };
6560
6561 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6562                                   struct bpf_insn_access_aux *info)
6563 {
6564         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6565                                           icsk_retransmits))
6566                 return false;
6567
6568         if (off % size != 0)
6569                 return false;
6570
6571         switch (off) {
6572         case offsetof(struct bpf_tcp_sock, bytes_received):
6573         case offsetof(struct bpf_tcp_sock, bytes_acked):
6574                 return size == sizeof(__u64);
6575         default:
6576                 return size == sizeof(__u32);
6577         }
6578 }
6579
6580 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6581                                     const struct bpf_insn *si,
6582                                     struct bpf_insn *insn_buf,
6583                                     struct bpf_prog *prog, u32 *target_size)
6584 {
6585         struct bpf_insn *insn = insn_buf;
6586
6587 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6588         do {                                                            \
6589                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6590                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6591                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6592                                       si->dst_reg, si->src_reg,         \
6593                                       offsetof(struct tcp_sock, FIELD)); \
6594         } while (0)
6595
6596 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6597         do {                                                            \
6598                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6599                                           FIELD) >                      \
6600                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6601                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6602                                         struct inet_connection_sock,    \
6603                                         FIELD),                         \
6604                                       si->dst_reg, si->src_reg,         \
6605                                       offsetof(                         \
6606                                         struct inet_connection_sock,    \
6607                                         FIELD));                        \
6608         } while (0)
6609
6610         if (insn > insn_buf)
6611                 return insn - insn_buf;
6612
6613         switch (si->off) {
6614         case offsetof(struct bpf_tcp_sock, rtt_min):
6615                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6616                              sizeof(struct minmax));
6617                 BUILD_BUG_ON(sizeof(struct minmax) <
6618                              sizeof(struct minmax_sample));
6619
6620                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6621                                       offsetof(struct tcp_sock, rtt_min) +
6622                                       offsetof(struct minmax_sample, v));
6623                 break;
6624         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6625                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6626                 break;
6627         case offsetof(struct bpf_tcp_sock, srtt_us):
6628                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6629                 break;
6630         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6631                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6632                 break;
6633         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6634                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6635                 break;
6636         case offsetof(struct bpf_tcp_sock, snd_nxt):
6637                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6638                 break;
6639         case offsetof(struct bpf_tcp_sock, snd_una):
6640                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6641                 break;
6642         case offsetof(struct bpf_tcp_sock, mss_cache):
6643                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6644                 break;
6645         case offsetof(struct bpf_tcp_sock, ecn_flags):
6646                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6647                 break;
6648         case offsetof(struct bpf_tcp_sock, rate_delivered):
6649                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6650                 break;
6651         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6652                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6653                 break;
6654         case offsetof(struct bpf_tcp_sock, packets_out):
6655                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6656                 break;
6657         case offsetof(struct bpf_tcp_sock, retrans_out):
6658                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6659                 break;
6660         case offsetof(struct bpf_tcp_sock, total_retrans):
6661                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6662                 break;
6663         case offsetof(struct bpf_tcp_sock, segs_in):
6664                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6665                 break;
6666         case offsetof(struct bpf_tcp_sock, data_segs_in):
6667                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6668                 break;
6669         case offsetof(struct bpf_tcp_sock, segs_out):
6670                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6671                 break;
6672         case offsetof(struct bpf_tcp_sock, data_segs_out):
6673                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6674                 break;
6675         case offsetof(struct bpf_tcp_sock, lost_out):
6676                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6677                 break;
6678         case offsetof(struct bpf_tcp_sock, sacked_out):
6679                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6680                 break;
6681         case offsetof(struct bpf_tcp_sock, bytes_received):
6682                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6683                 break;
6684         case offsetof(struct bpf_tcp_sock, bytes_acked):
6685                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6686                 break;
6687         case offsetof(struct bpf_tcp_sock, dsack_dups):
6688                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6689                 break;
6690         case offsetof(struct bpf_tcp_sock, delivered):
6691                 BPF_TCP_SOCK_GET_COMMON(delivered);
6692                 break;
6693         case offsetof(struct bpf_tcp_sock, delivered_ce):
6694                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6695                 break;
6696         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6697                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6698                 break;
6699         }
6700
6701         return insn - insn_buf;
6702 }
6703
6704 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6705 {
6706         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6707                 return (unsigned long)sk;
6708
6709         return (unsigned long)NULL;
6710 }
6711
6712 const struct bpf_func_proto bpf_tcp_sock_proto = {
6713         .func           = bpf_tcp_sock,
6714         .gpl_only       = false,
6715         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6716         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6717 };
6718
6719 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6720 {
6721         sk = sk_to_full_sk(sk);
6722
6723         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6724                 return (unsigned long)sk;
6725
6726         return (unsigned long)NULL;
6727 }
6728
6729 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6730         .func           = bpf_get_listener_sock,
6731         .gpl_only       = false,
6732         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6733         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6734 };
6735
6736 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6737 {
6738         unsigned int iphdr_len;
6739
6740         switch (skb_protocol(skb, true)) {
6741         case cpu_to_be16(ETH_P_IP):
6742                 iphdr_len = sizeof(struct iphdr);
6743                 break;
6744         case cpu_to_be16(ETH_P_IPV6):
6745                 iphdr_len = sizeof(struct ipv6hdr);
6746                 break;
6747         default:
6748                 return 0;
6749         }
6750
6751         if (skb_headlen(skb) < iphdr_len)
6752                 return 0;
6753
6754         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6755                 return 0;
6756
6757         return INET_ECN_set_ce(skb);
6758 }
6759
6760 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6761                                   struct bpf_insn_access_aux *info)
6762 {
6763         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6764                 return false;
6765
6766         if (off % size != 0)
6767                 return false;
6768
6769         switch (off) {
6770         default:
6771                 return size == sizeof(__u32);
6772         }
6773 }
6774
6775 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6776                                     const struct bpf_insn *si,
6777                                     struct bpf_insn *insn_buf,
6778                                     struct bpf_prog *prog, u32 *target_size)
6779 {
6780         struct bpf_insn *insn = insn_buf;
6781
6782 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6783         do {                                                            \
6784                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6785                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6786                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6787                                       si->dst_reg, si->src_reg,         \
6788                                       offsetof(struct xdp_sock, FIELD)); \
6789         } while (0)
6790
6791         switch (si->off) {
6792         case offsetof(struct bpf_xdp_sock, queue_id):
6793                 BPF_XDP_SOCK_GET(queue_id);
6794                 break;
6795         }
6796
6797         return insn - insn_buf;
6798 }
6799
6800 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6801         .func           = bpf_skb_ecn_set_ce,
6802         .gpl_only       = false,
6803         .ret_type       = RET_INTEGER,
6804         .arg1_type      = ARG_PTR_TO_CTX,
6805 };
6806
6807 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6808            struct tcphdr *, th, u32, th_len)
6809 {
6810 #ifdef CONFIG_SYN_COOKIES
6811         u32 cookie;
6812         int ret;
6813
6814         if (unlikely(!sk || th_len < sizeof(*th)))
6815                 return -EINVAL;
6816
6817         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6818         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6819                 return -EINVAL;
6820
6821         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6822                 return -EINVAL;
6823
6824         if (!th->ack || th->rst || th->syn)
6825                 return -ENOENT;
6826
6827         if (unlikely(iph_len < sizeof(struct iphdr)))
6828                 return -EINVAL;
6829
6830         if (tcp_synq_no_recent_overflow(sk))
6831                 return -ENOENT;
6832
6833         cookie = ntohl(th->ack_seq) - 1;
6834
6835         /* Both struct iphdr and struct ipv6hdr have the version field at the
6836          * same offset so we can cast to the shorter header (struct iphdr).
6837          */
6838         switch (((struct iphdr *)iph)->version) {
6839         case 4:
6840                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6841                         return -EINVAL;
6842
6843                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6844                 break;
6845
6846 #if IS_BUILTIN(CONFIG_IPV6)
6847         case 6:
6848                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6849                         return -EINVAL;
6850
6851                 if (sk->sk_family != AF_INET6)
6852                         return -EINVAL;
6853
6854                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6855                 break;
6856 #endif /* CONFIG_IPV6 */
6857
6858         default:
6859                 return -EPROTONOSUPPORT;
6860         }
6861
6862         if (ret > 0)
6863                 return 0;
6864
6865         return -ENOENT;
6866 #else
6867         return -ENOTSUPP;
6868 #endif
6869 }
6870
6871 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6872         .func           = bpf_tcp_check_syncookie,
6873         .gpl_only       = true,
6874         .pkt_access     = true,
6875         .ret_type       = RET_INTEGER,
6876         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6877         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6878         .arg3_type      = ARG_CONST_SIZE,
6879         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6880         .arg5_type      = ARG_CONST_SIZE,
6881 };
6882
6883 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6884            struct tcphdr *, th, u32, th_len)
6885 {
6886 #ifdef CONFIG_SYN_COOKIES
6887         u32 cookie;
6888         u16 mss;
6889
6890         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6891                 return -EINVAL;
6892
6893         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6894                 return -EINVAL;
6895
6896         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6897                 return -ENOENT;
6898
6899         if (!th->syn || th->ack || th->fin || th->rst)
6900                 return -EINVAL;
6901
6902         if (unlikely(iph_len < sizeof(struct iphdr)))
6903                 return -EINVAL;
6904
6905         /* Both struct iphdr and struct ipv6hdr have the version field at the
6906          * same offset so we can cast to the shorter header (struct iphdr).
6907          */
6908         switch (((struct iphdr *)iph)->version) {
6909         case 4:
6910                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6911                         return -EINVAL;
6912
6913                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6914                 break;
6915
6916 #if IS_BUILTIN(CONFIG_IPV6)
6917         case 6:
6918                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6919                         return -EINVAL;
6920
6921                 if (sk->sk_family != AF_INET6)
6922                         return -EINVAL;
6923
6924                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6925                 break;
6926 #endif /* CONFIG_IPV6 */
6927
6928         default:
6929                 return -EPROTONOSUPPORT;
6930         }
6931         if (mss == 0)
6932                 return -ENOENT;
6933
6934         return cookie | ((u64)mss << 32);
6935 #else
6936         return -EOPNOTSUPP;
6937 #endif /* CONFIG_SYN_COOKIES */
6938 }
6939
6940 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6941         .func           = bpf_tcp_gen_syncookie,
6942         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
6943         .pkt_access     = true,
6944         .ret_type       = RET_INTEGER,
6945         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6946         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6947         .arg3_type      = ARG_CONST_SIZE,
6948         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6949         .arg5_type      = ARG_CONST_SIZE,
6950 };
6951
6952 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6953 {
6954         if (!sk || flags != 0)
6955                 return -EINVAL;
6956         if (!skb_at_tc_ingress(skb))
6957                 return -EOPNOTSUPP;
6958         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6959                 return -ENETUNREACH;
6960         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6961                 return -ESOCKTNOSUPPORT;
6962         if (sk_unhashed(sk))
6963                 return -EOPNOTSUPP;
6964         if (sk_is_refcounted(sk) &&
6965             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6966                 return -ENOENT;
6967
6968         skb_orphan(skb);
6969         skb->sk = sk;
6970         skb->destructor = sock_pfree;
6971
6972         return 0;
6973 }
6974
6975 static const struct bpf_func_proto bpf_sk_assign_proto = {
6976         .func           = bpf_sk_assign,
6977         .gpl_only       = false,
6978         .ret_type       = RET_INTEGER,
6979         .arg1_type      = ARG_PTR_TO_CTX,
6980         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6981         .arg3_type      = ARG_ANYTHING,
6982 };
6983
6984 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6985                                     u8 search_kind, const u8 *magic,
6986                                     u8 magic_len, bool *eol)
6987 {
6988         u8 kind, kind_len;
6989
6990         *eol = false;
6991
6992         while (op < opend) {
6993                 kind = op[0];
6994
6995                 if (kind == TCPOPT_EOL) {
6996                         *eol = true;
6997                         return ERR_PTR(-ENOMSG);
6998                 } else if (kind == TCPOPT_NOP) {
6999                         op++;
7000                         continue;
7001                 }
7002
7003                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7004                         /* Something is wrong in the received header.
7005                          * Follow the TCP stack's tcp_parse_options()
7006                          * and just bail here.
7007                          */
7008                         return ERR_PTR(-EFAULT);
7009
7010                 kind_len = op[1];
7011                 if (search_kind == kind) {
7012                         if (!magic_len)
7013                                 return op;
7014
7015                         if (magic_len > kind_len - 2)
7016                                 return ERR_PTR(-ENOMSG);
7017
7018                         if (!memcmp(&op[2], magic, magic_len))
7019                                 return op;
7020                 }
7021
7022                 op += kind_len;
7023         }
7024
7025         return ERR_PTR(-ENOMSG);
7026 }
7027
7028 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7029            void *, search_res, u32, len, u64, flags)
7030 {
7031         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7032         const u8 *op, *opend, *magic, *search = search_res;
7033         u8 search_kind, search_len, copy_len, magic_len;
7034         int ret;
7035
7036         /* 2 byte is the minimal option len except TCPOPT_NOP and
7037          * TCPOPT_EOL which are useless for the bpf prog to learn
7038          * and this helper disallow loading them also.
7039          */
7040         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7041                 return -EINVAL;
7042
7043         search_kind = search[0];
7044         search_len = search[1];
7045
7046         if (search_len > len || search_kind == TCPOPT_NOP ||
7047             search_kind == TCPOPT_EOL)
7048                 return -EINVAL;
7049
7050         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7051                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7052                 if (search_len != 4 && search_len != 6)
7053                         return -EINVAL;
7054                 magic = &search[2];
7055                 magic_len = search_len - 2;
7056         } else {
7057                 if (search_len)
7058                         return -EINVAL;
7059                 magic = NULL;
7060                 magic_len = 0;
7061         }
7062
7063         if (load_syn) {
7064                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7065                 if (ret < 0)
7066                         return ret;
7067
7068                 opend = op + ret;
7069                 op += sizeof(struct tcphdr);
7070         } else {
7071                 if (!bpf_sock->skb ||
7072                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7073                         /* This bpf_sock->op cannot call this helper */
7074                         return -EPERM;
7075
7076                 opend = bpf_sock->skb_data_end;
7077                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7078         }
7079
7080         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7081                                 &eol);
7082         if (IS_ERR(op))
7083                 return PTR_ERR(op);
7084
7085         copy_len = op[1];
7086         ret = copy_len;
7087         if (copy_len > len) {
7088                 ret = -ENOSPC;
7089                 copy_len = len;
7090         }
7091
7092         memcpy(search_res, op, copy_len);
7093         return ret;
7094 }
7095
7096 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7097         .func           = bpf_sock_ops_load_hdr_opt,
7098         .gpl_only       = false,
7099         .ret_type       = RET_INTEGER,
7100         .arg1_type      = ARG_PTR_TO_CTX,
7101         .arg2_type      = ARG_PTR_TO_MEM,
7102         .arg3_type      = ARG_CONST_SIZE,
7103         .arg4_type      = ARG_ANYTHING,
7104 };
7105
7106 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7107            const void *, from, u32, len, u64, flags)
7108 {
7109         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7110         const u8 *op, *new_op, *magic = NULL;
7111         struct sk_buff *skb;
7112         bool eol;
7113
7114         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7115                 return -EPERM;
7116
7117         if (len < 2 || flags)
7118                 return -EINVAL;
7119
7120         new_op = from;
7121         new_kind = new_op[0];
7122         new_kind_len = new_op[1];
7123
7124         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7125             new_kind == TCPOPT_EOL)
7126                 return -EINVAL;
7127
7128         if (new_kind_len > bpf_sock->remaining_opt_len)
7129                 return -ENOSPC;
7130
7131         /* 253 is another experimental kind */
7132         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7133                 if (new_kind_len < 4)
7134                         return -EINVAL;
7135                 /* Match for the 2 byte magic also.
7136                  * RFC 6994: the magic could be 2 or 4 bytes.
7137                  * Hence, matching by 2 byte only is on the
7138                  * conservative side but it is the right
7139                  * thing to do for the 'search-for-duplication'
7140                  * purpose.
7141                  */
7142                 magic = &new_op[2];
7143                 magic_len = 2;
7144         }
7145
7146         /* Check for duplication */
7147         skb = bpf_sock->skb;
7148         op = skb->data + sizeof(struct tcphdr);
7149         opend = bpf_sock->skb_data_end;
7150
7151         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7152                                 &eol);
7153         if (!IS_ERR(op))
7154                 return -EEXIST;
7155
7156         if (PTR_ERR(op) != -ENOMSG)
7157                 return PTR_ERR(op);
7158
7159         if (eol)
7160                 /* The option has been ended.  Treat it as no more
7161                  * header option can be written.
7162                  */
7163                 return -ENOSPC;
7164
7165         /* No duplication found.  Store the header option. */
7166         memcpy(opend, from, new_kind_len);
7167
7168         bpf_sock->remaining_opt_len -= new_kind_len;
7169         bpf_sock->skb_data_end += new_kind_len;
7170
7171         return 0;
7172 }
7173
7174 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7175         .func           = bpf_sock_ops_store_hdr_opt,
7176         .gpl_only       = false,
7177         .ret_type       = RET_INTEGER,
7178         .arg1_type      = ARG_PTR_TO_CTX,
7179         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7180         .arg3_type      = ARG_CONST_SIZE,
7181         .arg4_type      = ARG_ANYTHING,
7182 };
7183
7184 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7185            u32, len, u64, flags)
7186 {
7187         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7188                 return -EPERM;
7189
7190         if (flags || len < 2)
7191                 return -EINVAL;
7192
7193         if (len > bpf_sock->remaining_opt_len)
7194                 return -ENOSPC;
7195
7196         bpf_sock->remaining_opt_len -= len;
7197
7198         return 0;
7199 }
7200
7201 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7202         .func           = bpf_sock_ops_reserve_hdr_opt,
7203         .gpl_only       = false,
7204         .ret_type       = RET_INTEGER,
7205         .arg1_type      = ARG_PTR_TO_CTX,
7206         .arg2_type      = ARG_ANYTHING,
7207         .arg3_type      = ARG_ANYTHING,
7208 };
7209
7210 #endif /* CONFIG_INET */
7211
7212 bool bpf_helper_changes_pkt_data(void *func)
7213 {
7214         if (func == bpf_skb_vlan_push ||
7215             func == bpf_skb_vlan_pop ||
7216             func == bpf_skb_store_bytes ||
7217             func == bpf_skb_change_proto ||
7218             func == bpf_skb_change_head ||
7219             func == sk_skb_change_head ||
7220             func == bpf_skb_change_tail ||
7221             func == sk_skb_change_tail ||
7222             func == bpf_skb_adjust_room ||
7223             func == sk_skb_adjust_room ||
7224             func == bpf_skb_pull_data ||
7225             func == sk_skb_pull_data ||
7226             func == bpf_clone_redirect ||
7227             func == bpf_l3_csum_replace ||
7228             func == bpf_l4_csum_replace ||
7229             func == bpf_xdp_adjust_head ||
7230             func == bpf_xdp_adjust_meta ||
7231             func == bpf_msg_pull_data ||
7232             func == bpf_msg_push_data ||
7233             func == bpf_msg_pop_data ||
7234             func == bpf_xdp_adjust_tail ||
7235 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7236             func == bpf_lwt_seg6_store_bytes ||
7237             func == bpf_lwt_seg6_adjust_srh ||
7238             func == bpf_lwt_seg6_action ||
7239 #endif
7240 #ifdef CONFIG_INET
7241             func == bpf_sock_ops_store_hdr_opt ||
7242 #endif
7243             func == bpf_lwt_in_push_encap ||
7244             func == bpf_lwt_xmit_push_encap)
7245                 return true;
7246
7247         return false;
7248 }
7249
7250 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7251 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7252
7253 static const struct bpf_func_proto *
7254 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7255 {
7256         switch (func_id) {
7257         /* inet and inet6 sockets are created in a process
7258          * context so there is always a valid uid/gid
7259          */
7260         case BPF_FUNC_get_current_uid_gid:
7261                 return &bpf_get_current_uid_gid_proto;
7262         case BPF_FUNC_get_local_storage:
7263                 return &bpf_get_local_storage_proto;
7264         case BPF_FUNC_get_socket_cookie:
7265                 return &bpf_get_socket_cookie_sock_proto;
7266         case BPF_FUNC_get_netns_cookie:
7267                 return &bpf_get_netns_cookie_sock_proto;
7268         case BPF_FUNC_perf_event_output:
7269                 return &bpf_event_output_data_proto;
7270         case BPF_FUNC_get_current_pid_tgid:
7271                 return &bpf_get_current_pid_tgid_proto;
7272         case BPF_FUNC_get_current_comm:
7273                 return &bpf_get_current_comm_proto;
7274 #ifdef CONFIG_CGROUPS
7275         case BPF_FUNC_get_current_cgroup_id:
7276                 return &bpf_get_current_cgroup_id_proto;
7277         case BPF_FUNC_get_current_ancestor_cgroup_id:
7278                 return &bpf_get_current_ancestor_cgroup_id_proto;
7279 #endif
7280 #ifdef CONFIG_CGROUP_NET_CLASSID
7281         case BPF_FUNC_get_cgroup_classid:
7282                 return &bpf_get_cgroup_classid_curr_proto;
7283 #endif
7284         case BPF_FUNC_sk_storage_get:
7285                 return &bpf_sk_storage_get_cg_sock_proto;
7286         case BPF_FUNC_ktime_get_coarse_ns:
7287                 return &bpf_ktime_get_coarse_ns_proto;
7288         default:
7289                 return bpf_base_func_proto(func_id);
7290         }
7291 }
7292
7293 static const struct bpf_func_proto *
7294 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7295 {
7296         switch (func_id) {
7297         /* inet and inet6 sockets are created in a process
7298          * context so there is always a valid uid/gid
7299          */
7300         case BPF_FUNC_get_current_uid_gid:
7301                 return &bpf_get_current_uid_gid_proto;
7302         case BPF_FUNC_bind:
7303                 switch (prog->expected_attach_type) {
7304                 case BPF_CGROUP_INET4_CONNECT:
7305                 case BPF_CGROUP_INET6_CONNECT:
7306                         return &bpf_bind_proto;
7307                 default:
7308                         return NULL;
7309                 }
7310         case BPF_FUNC_get_socket_cookie:
7311                 return &bpf_get_socket_cookie_sock_addr_proto;
7312         case BPF_FUNC_get_netns_cookie:
7313                 return &bpf_get_netns_cookie_sock_addr_proto;
7314         case BPF_FUNC_get_local_storage:
7315                 return &bpf_get_local_storage_proto;
7316         case BPF_FUNC_perf_event_output:
7317                 return &bpf_event_output_data_proto;
7318         case BPF_FUNC_get_current_pid_tgid:
7319                 return &bpf_get_current_pid_tgid_proto;
7320         case BPF_FUNC_get_current_comm:
7321                 return &bpf_get_current_comm_proto;
7322 #ifdef CONFIG_CGROUPS
7323         case BPF_FUNC_get_current_cgroup_id:
7324                 return &bpf_get_current_cgroup_id_proto;
7325         case BPF_FUNC_get_current_ancestor_cgroup_id:
7326                 return &bpf_get_current_ancestor_cgroup_id_proto;
7327 #endif
7328 #ifdef CONFIG_CGROUP_NET_CLASSID
7329         case BPF_FUNC_get_cgroup_classid:
7330                 return &bpf_get_cgroup_classid_curr_proto;
7331 #endif
7332 #ifdef CONFIG_INET
7333         case BPF_FUNC_sk_lookup_tcp:
7334                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7335         case BPF_FUNC_sk_lookup_udp:
7336                 return &bpf_sock_addr_sk_lookup_udp_proto;
7337         case BPF_FUNC_sk_release:
7338                 return &bpf_sk_release_proto;
7339         case BPF_FUNC_skc_lookup_tcp:
7340                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7341 #endif /* CONFIG_INET */
7342         case BPF_FUNC_sk_storage_get:
7343                 return &bpf_sk_storage_get_proto;
7344         case BPF_FUNC_sk_storage_delete:
7345                 return &bpf_sk_storage_delete_proto;
7346         case BPF_FUNC_setsockopt:
7347                 switch (prog->expected_attach_type) {
7348                 case BPF_CGROUP_INET4_BIND:
7349                 case BPF_CGROUP_INET6_BIND:
7350                 case BPF_CGROUP_INET4_CONNECT:
7351                 case BPF_CGROUP_INET6_CONNECT:
7352                 case BPF_CGROUP_UDP4_RECVMSG:
7353                 case BPF_CGROUP_UDP6_RECVMSG:
7354                 case BPF_CGROUP_UDP4_SENDMSG:
7355                 case BPF_CGROUP_UDP6_SENDMSG:
7356                 case BPF_CGROUP_INET4_GETPEERNAME:
7357                 case BPF_CGROUP_INET6_GETPEERNAME:
7358                 case BPF_CGROUP_INET4_GETSOCKNAME:
7359                 case BPF_CGROUP_INET6_GETSOCKNAME:
7360                         return &bpf_sock_addr_setsockopt_proto;
7361                 default:
7362                         return NULL;
7363                 }
7364         case BPF_FUNC_getsockopt:
7365                 switch (prog->expected_attach_type) {
7366                 case BPF_CGROUP_INET4_BIND:
7367                 case BPF_CGROUP_INET6_BIND:
7368                 case BPF_CGROUP_INET4_CONNECT:
7369                 case BPF_CGROUP_INET6_CONNECT:
7370                 case BPF_CGROUP_UDP4_RECVMSG:
7371                 case BPF_CGROUP_UDP6_RECVMSG:
7372                 case BPF_CGROUP_UDP4_SENDMSG:
7373                 case BPF_CGROUP_UDP6_SENDMSG:
7374                 case BPF_CGROUP_INET4_GETPEERNAME:
7375                 case BPF_CGROUP_INET6_GETPEERNAME:
7376                 case BPF_CGROUP_INET4_GETSOCKNAME:
7377                 case BPF_CGROUP_INET6_GETSOCKNAME:
7378                         return &bpf_sock_addr_getsockopt_proto;
7379                 default:
7380                         return NULL;
7381                 }
7382         default:
7383                 return bpf_sk_base_func_proto(func_id);
7384         }
7385 }
7386
7387 static const struct bpf_func_proto *
7388 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7389 {
7390         switch (func_id) {
7391         case BPF_FUNC_skb_load_bytes:
7392                 return &bpf_skb_load_bytes_proto;
7393         case BPF_FUNC_skb_load_bytes_relative:
7394                 return &bpf_skb_load_bytes_relative_proto;
7395         case BPF_FUNC_get_socket_cookie:
7396                 return &bpf_get_socket_cookie_proto;
7397         case BPF_FUNC_get_socket_uid:
7398                 return &bpf_get_socket_uid_proto;
7399         case BPF_FUNC_perf_event_output:
7400                 return &bpf_skb_event_output_proto;
7401         default:
7402                 return bpf_sk_base_func_proto(func_id);
7403         }
7404 }
7405
7406 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7407 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7408
7409 static const struct bpf_func_proto *
7410 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7411 {
7412         switch (func_id) {
7413         case BPF_FUNC_get_local_storage:
7414                 return &bpf_get_local_storage_proto;
7415         case BPF_FUNC_sk_fullsock:
7416                 return &bpf_sk_fullsock_proto;
7417         case BPF_FUNC_sk_storage_get:
7418                 return &bpf_sk_storage_get_proto;
7419         case BPF_FUNC_sk_storage_delete:
7420                 return &bpf_sk_storage_delete_proto;
7421         case BPF_FUNC_perf_event_output:
7422                 return &bpf_skb_event_output_proto;
7423 #ifdef CONFIG_SOCK_CGROUP_DATA
7424         case BPF_FUNC_skb_cgroup_id:
7425                 return &bpf_skb_cgroup_id_proto;
7426         case BPF_FUNC_skb_ancestor_cgroup_id:
7427                 return &bpf_skb_ancestor_cgroup_id_proto;
7428         case BPF_FUNC_sk_cgroup_id:
7429                 return &bpf_sk_cgroup_id_proto;
7430         case BPF_FUNC_sk_ancestor_cgroup_id:
7431                 return &bpf_sk_ancestor_cgroup_id_proto;
7432 #endif
7433 #ifdef CONFIG_INET
7434         case BPF_FUNC_sk_lookup_tcp:
7435                 return &bpf_sk_lookup_tcp_proto;
7436         case BPF_FUNC_sk_lookup_udp:
7437                 return &bpf_sk_lookup_udp_proto;
7438         case BPF_FUNC_sk_release:
7439                 return &bpf_sk_release_proto;
7440         case BPF_FUNC_skc_lookup_tcp:
7441                 return &bpf_skc_lookup_tcp_proto;
7442         case BPF_FUNC_tcp_sock:
7443                 return &bpf_tcp_sock_proto;
7444         case BPF_FUNC_get_listener_sock:
7445                 return &bpf_get_listener_sock_proto;
7446         case BPF_FUNC_skb_ecn_set_ce:
7447                 return &bpf_skb_ecn_set_ce_proto;
7448 #endif
7449         default:
7450                 return sk_filter_func_proto(func_id, prog);
7451         }
7452 }
7453
7454 static const struct bpf_func_proto *
7455 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7456 {
7457         switch (func_id) {
7458         case BPF_FUNC_skb_store_bytes:
7459                 return &bpf_skb_store_bytes_proto;
7460         case BPF_FUNC_skb_load_bytes:
7461                 return &bpf_skb_load_bytes_proto;
7462         case BPF_FUNC_skb_load_bytes_relative:
7463                 return &bpf_skb_load_bytes_relative_proto;
7464         case BPF_FUNC_skb_pull_data:
7465                 return &bpf_skb_pull_data_proto;
7466         case BPF_FUNC_csum_diff:
7467                 return &bpf_csum_diff_proto;
7468         case BPF_FUNC_csum_update:
7469                 return &bpf_csum_update_proto;
7470         case BPF_FUNC_csum_level:
7471                 return &bpf_csum_level_proto;
7472         case BPF_FUNC_l3_csum_replace:
7473                 return &bpf_l3_csum_replace_proto;
7474         case BPF_FUNC_l4_csum_replace:
7475                 return &bpf_l4_csum_replace_proto;
7476         case BPF_FUNC_clone_redirect:
7477                 return &bpf_clone_redirect_proto;
7478         case BPF_FUNC_get_cgroup_classid:
7479                 return &bpf_get_cgroup_classid_proto;
7480         case BPF_FUNC_skb_vlan_push:
7481                 return &bpf_skb_vlan_push_proto;
7482         case BPF_FUNC_skb_vlan_pop:
7483                 return &bpf_skb_vlan_pop_proto;
7484         case BPF_FUNC_skb_change_proto:
7485                 return &bpf_skb_change_proto_proto;
7486         case BPF_FUNC_skb_change_type:
7487                 return &bpf_skb_change_type_proto;
7488         case BPF_FUNC_skb_adjust_room:
7489                 return &bpf_skb_adjust_room_proto;
7490         case BPF_FUNC_skb_change_tail:
7491                 return &bpf_skb_change_tail_proto;
7492         case BPF_FUNC_skb_change_head:
7493                 return &bpf_skb_change_head_proto;
7494         case BPF_FUNC_skb_get_tunnel_key:
7495                 return &bpf_skb_get_tunnel_key_proto;
7496         case BPF_FUNC_skb_set_tunnel_key:
7497                 return bpf_get_skb_set_tunnel_proto(func_id);
7498         case BPF_FUNC_skb_get_tunnel_opt:
7499                 return &bpf_skb_get_tunnel_opt_proto;
7500         case BPF_FUNC_skb_set_tunnel_opt:
7501                 return bpf_get_skb_set_tunnel_proto(func_id);
7502         case BPF_FUNC_redirect:
7503                 return &bpf_redirect_proto;
7504         case BPF_FUNC_redirect_neigh:
7505                 return &bpf_redirect_neigh_proto;
7506         case BPF_FUNC_redirect_peer:
7507                 return &bpf_redirect_peer_proto;
7508         case BPF_FUNC_get_route_realm:
7509                 return &bpf_get_route_realm_proto;
7510         case BPF_FUNC_get_hash_recalc:
7511                 return &bpf_get_hash_recalc_proto;
7512         case BPF_FUNC_set_hash_invalid:
7513                 return &bpf_set_hash_invalid_proto;
7514         case BPF_FUNC_set_hash:
7515                 return &bpf_set_hash_proto;
7516         case BPF_FUNC_perf_event_output:
7517                 return &bpf_skb_event_output_proto;
7518         case BPF_FUNC_get_smp_processor_id:
7519                 return &bpf_get_smp_processor_id_proto;
7520         case BPF_FUNC_skb_under_cgroup:
7521                 return &bpf_skb_under_cgroup_proto;
7522         case BPF_FUNC_get_socket_cookie:
7523                 return &bpf_get_socket_cookie_proto;
7524         case BPF_FUNC_get_socket_uid:
7525                 return &bpf_get_socket_uid_proto;
7526         case BPF_FUNC_fib_lookup:
7527                 return &bpf_skb_fib_lookup_proto;
7528         case BPF_FUNC_check_mtu:
7529                 return &bpf_skb_check_mtu_proto;
7530         case BPF_FUNC_sk_fullsock:
7531                 return &bpf_sk_fullsock_proto;
7532         case BPF_FUNC_sk_storage_get:
7533                 return &bpf_sk_storage_get_proto;
7534         case BPF_FUNC_sk_storage_delete:
7535                 return &bpf_sk_storage_delete_proto;
7536 #ifdef CONFIG_XFRM
7537         case BPF_FUNC_skb_get_xfrm_state:
7538                 return &bpf_skb_get_xfrm_state_proto;
7539 #endif
7540 #ifdef CONFIG_CGROUP_NET_CLASSID
7541         case BPF_FUNC_skb_cgroup_classid:
7542                 return &bpf_skb_cgroup_classid_proto;
7543 #endif
7544 #ifdef CONFIG_SOCK_CGROUP_DATA
7545         case BPF_FUNC_skb_cgroup_id:
7546                 return &bpf_skb_cgroup_id_proto;
7547         case BPF_FUNC_skb_ancestor_cgroup_id:
7548                 return &bpf_skb_ancestor_cgroup_id_proto;
7549 #endif
7550 #ifdef CONFIG_INET
7551         case BPF_FUNC_sk_lookup_tcp:
7552                 return &bpf_tc_sk_lookup_tcp_proto;
7553         case BPF_FUNC_sk_lookup_udp:
7554                 return &bpf_tc_sk_lookup_udp_proto;
7555         case BPF_FUNC_sk_release:
7556                 return &bpf_sk_release_proto;
7557         case BPF_FUNC_tcp_sock:
7558                 return &bpf_tcp_sock_proto;
7559         case BPF_FUNC_get_listener_sock:
7560                 return &bpf_get_listener_sock_proto;
7561         case BPF_FUNC_skc_lookup_tcp:
7562                 return &bpf_tc_skc_lookup_tcp_proto;
7563         case BPF_FUNC_tcp_check_syncookie:
7564                 return &bpf_tcp_check_syncookie_proto;
7565         case BPF_FUNC_skb_ecn_set_ce:
7566                 return &bpf_skb_ecn_set_ce_proto;
7567         case BPF_FUNC_tcp_gen_syncookie:
7568                 return &bpf_tcp_gen_syncookie_proto;
7569         case BPF_FUNC_sk_assign:
7570                 return &bpf_sk_assign_proto;
7571 #endif
7572         default:
7573                 return bpf_sk_base_func_proto(func_id);
7574         }
7575 }
7576
7577 static const struct bpf_func_proto *
7578 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7579 {
7580         switch (func_id) {
7581         case BPF_FUNC_perf_event_output:
7582                 return &bpf_xdp_event_output_proto;
7583         case BPF_FUNC_get_smp_processor_id:
7584                 return &bpf_get_smp_processor_id_proto;
7585         case BPF_FUNC_csum_diff:
7586                 return &bpf_csum_diff_proto;
7587         case BPF_FUNC_xdp_adjust_head:
7588                 return &bpf_xdp_adjust_head_proto;
7589         case BPF_FUNC_xdp_adjust_meta:
7590                 return &bpf_xdp_adjust_meta_proto;
7591         case BPF_FUNC_redirect:
7592                 return &bpf_xdp_redirect_proto;
7593         case BPF_FUNC_redirect_map:
7594                 return &bpf_xdp_redirect_map_proto;
7595         case BPF_FUNC_xdp_adjust_tail:
7596                 return &bpf_xdp_adjust_tail_proto;
7597         case BPF_FUNC_fib_lookup:
7598                 return &bpf_xdp_fib_lookup_proto;
7599         case BPF_FUNC_check_mtu:
7600                 return &bpf_xdp_check_mtu_proto;
7601 #ifdef CONFIG_INET
7602         case BPF_FUNC_sk_lookup_udp:
7603                 return &bpf_xdp_sk_lookup_udp_proto;
7604         case BPF_FUNC_sk_lookup_tcp:
7605                 return &bpf_xdp_sk_lookup_tcp_proto;
7606         case BPF_FUNC_sk_release:
7607                 return &bpf_sk_release_proto;
7608         case BPF_FUNC_skc_lookup_tcp:
7609                 return &bpf_xdp_skc_lookup_tcp_proto;
7610         case BPF_FUNC_tcp_check_syncookie:
7611                 return &bpf_tcp_check_syncookie_proto;
7612         case BPF_FUNC_tcp_gen_syncookie:
7613                 return &bpf_tcp_gen_syncookie_proto;
7614 #endif
7615         default:
7616                 return bpf_sk_base_func_proto(func_id);
7617         }
7618 }
7619
7620 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7621 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7622
7623 static const struct bpf_func_proto *
7624 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7625 {
7626         switch (func_id) {
7627         case BPF_FUNC_setsockopt:
7628                 return &bpf_sock_ops_setsockopt_proto;
7629         case BPF_FUNC_getsockopt:
7630                 return &bpf_sock_ops_getsockopt_proto;
7631         case BPF_FUNC_sock_ops_cb_flags_set:
7632                 return &bpf_sock_ops_cb_flags_set_proto;
7633         case BPF_FUNC_sock_map_update:
7634                 return &bpf_sock_map_update_proto;
7635         case BPF_FUNC_sock_hash_update:
7636                 return &bpf_sock_hash_update_proto;
7637         case BPF_FUNC_get_socket_cookie:
7638                 return &bpf_get_socket_cookie_sock_ops_proto;
7639         case BPF_FUNC_get_local_storage:
7640                 return &bpf_get_local_storage_proto;
7641         case BPF_FUNC_perf_event_output:
7642                 return &bpf_event_output_data_proto;
7643         case BPF_FUNC_sk_storage_get:
7644                 return &bpf_sk_storage_get_proto;
7645         case BPF_FUNC_sk_storage_delete:
7646                 return &bpf_sk_storage_delete_proto;
7647         case BPF_FUNC_get_netns_cookie:
7648                 return &bpf_get_netns_cookie_sock_ops_proto;
7649 #ifdef CONFIG_INET
7650         case BPF_FUNC_load_hdr_opt:
7651                 return &bpf_sock_ops_load_hdr_opt_proto;
7652         case BPF_FUNC_store_hdr_opt:
7653                 return &bpf_sock_ops_store_hdr_opt_proto;
7654         case BPF_FUNC_reserve_hdr_opt:
7655                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7656         case BPF_FUNC_tcp_sock:
7657                 return &bpf_tcp_sock_proto;
7658 #endif /* CONFIG_INET */
7659         default:
7660                 return bpf_sk_base_func_proto(func_id);
7661         }
7662 }
7663
7664 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7665 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7666
7667 static const struct bpf_func_proto *
7668 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7669 {
7670         switch (func_id) {
7671         case BPF_FUNC_msg_redirect_map:
7672                 return &bpf_msg_redirect_map_proto;
7673         case BPF_FUNC_msg_redirect_hash:
7674                 return &bpf_msg_redirect_hash_proto;
7675         case BPF_FUNC_msg_apply_bytes:
7676                 return &bpf_msg_apply_bytes_proto;
7677         case BPF_FUNC_msg_cork_bytes:
7678                 return &bpf_msg_cork_bytes_proto;
7679         case BPF_FUNC_msg_pull_data:
7680                 return &bpf_msg_pull_data_proto;
7681         case BPF_FUNC_msg_push_data:
7682                 return &bpf_msg_push_data_proto;
7683         case BPF_FUNC_msg_pop_data:
7684                 return &bpf_msg_pop_data_proto;
7685         case BPF_FUNC_perf_event_output:
7686                 return &bpf_event_output_data_proto;
7687         case BPF_FUNC_get_current_uid_gid:
7688                 return &bpf_get_current_uid_gid_proto;
7689         case BPF_FUNC_get_current_pid_tgid:
7690                 return &bpf_get_current_pid_tgid_proto;
7691         case BPF_FUNC_sk_storage_get:
7692                 return &bpf_sk_storage_get_proto;
7693         case BPF_FUNC_sk_storage_delete:
7694                 return &bpf_sk_storage_delete_proto;
7695         case BPF_FUNC_get_netns_cookie:
7696                 return &bpf_get_netns_cookie_sk_msg_proto;
7697 #ifdef CONFIG_CGROUPS
7698         case BPF_FUNC_get_current_cgroup_id:
7699                 return &bpf_get_current_cgroup_id_proto;
7700         case BPF_FUNC_get_current_ancestor_cgroup_id:
7701                 return &bpf_get_current_ancestor_cgroup_id_proto;
7702 #endif
7703 #ifdef CONFIG_CGROUP_NET_CLASSID
7704         case BPF_FUNC_get_cgroup_classid:
7705                 return &bpf_get_cgroup_classid_curr_proto;
7706 #endif
7707         default:
7708                 return bpf_sk_base_func_proto(func_id);
7709         }
7710 }
7711
7712 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7713 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7714
7715 static const struct bpf_func_proto *
7716 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7717 {
7718         switch (func_id) {
7719         case BPF_FUNC_skb_store_bytes:
7720                 return &bpf_skb_store_bytes_proto;
7721         case BPF_FUNC_skb_load_bytes:
7722                 return &bpf_skb_load_bytes_proto;
7723         case BPF_FUNC_skb_pull_data:
7724                 return &sk_skb_pull_data_proto;
7725         case BPF_FUNC_skb_change_tail:
7726                 return &sk_skb_change_tail_proto;
7727         case BPF_FUNC_skb_change_head:
7728                 return &sk_skb_change_head_proto;
7729         case BPF_FUNC_skb_adjust_room:
7730                 return &sk_skb_adjust_room_proto;
7731         case BPF_FUNC_get_socket_cookie:
7732                 return &bpf_get_socket_cookie_proto;
7733         case BPF_FUNC_get_socket_uid:
7734                 return &bpf_get_socket_uid_proto;
7735         case BPF_FUNC_sk_redirect_map:
7736                 return &bpf_sk_redirect_map_proto;
7737         case BPF_FUNC_sk_redirect_hash:
7738                 return &bpf_sk_redirect_hash_proto;
7739         case BPF_FUNC_perf_event_output:
7740                 return &bpf_skb_event_output_proto;
7741 #ifdef CONFIG_INET
7742         case BPF_FUNC_sk_lookup_tcp:
7743                 return &bpf_sk_lookup_tcp_proto;
7744         case BPF_FUNC_sk_lookup_udp:
7745                 return &bpf_sk_lookup_udp_proto;
7746         case BPF_FUNC_sk_release:
7747                 return &bpf_sk_release_proto;
7748         case BPF_FUNC_skc_lookup_tcp:
7749                 return &bpf_skc_lookup_tcp_proto;
7750 #endif
7751         default:
7752                 return bpf_sk_base_func_proto(func_id);
7753         }
7754 }
7755
7756 static const struct bpf_func_proto *
7757 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7758 {
7759         switch (func_id) {
7760         case BPF_FUNC_skb_load_bytes:
7761                 return &bpf_flow_dissector_load_bytes_proto;
7762         default:
7763                 return bpf_sk_base_func_proto(func_id);
7764         }
7765 }
7766
7767 static const struct bpf_func_proto *
7768 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7769 {
7770         switch (func_id) {
7771         case BPF_FUNC_skb_load_bytes:
7772                 return &bpf_skb_load_bytes_proto;
7773         case BPF_FUNC_skb_pull_data:
7774                 return &bpf_skb_pull_data_proto;
7775         case BPF_FUNC_csum_diff:
7776                 return &bpf_csum_diff_proto;
7777         case BPF_FUNC_get_cgroup_classid:
7778                 return &bpf_get_cgroup_classid_proto;
7779         case BPF_FUNC_get_route_realm:
7780                 return &bpf_get_route_realm_proto;
7781         case BPF_FUNC_get_hash_recalc:
7782                 return &bpf_get_hash_recalc_proto;
7783         case BPF_FUNC_perf_event_output:
7784                 return &bpf_skb_event_output_proto;
7785         case BPF_FUNC_get_smp_processor_id:
7786                 return &bpf_get_smp_processor_id_proto;
7787         case BPF_FUNC_skb_under_cgroup:
7788                 return &bpf_skb_under_cgroup_proto;
7789         default:
7790                 return bpf_sk_base_func_proto(func_id);
7791         }
7792 }
7793
7794 static const struct bpf_func_proto *
7795 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7796 {
7797         switch (func_id) {
7798         case BPF_FUNC_lwt_push_encap:
7799                 return &bpf_lwt_in_push_encap_proto;
7800         default:
7801                 return lwt_out_func_proto(func_id, prog);
7802         }
7803 }
7804
7805 static const struct bpf_func_proto *
7806 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7807 {
7808         switch (func_id) {
7809         case BPF_FUNC_skb_get_tunnel_key:
7810                 return &bpf_skb_get_tunnel_key_proto;
7811         case BPF_FUNC_skb_set_tunnel_key:
7812                 return bpf_get_skb_set_tunnel_proto(func_id);
7813         case BPF_FUNC_skb_get_tunnel_opt:
7814                 return &bpf_skb_get_tunnel_opt_proto;
7815         case BPF_FUNC_skb_set_tunnel_opt:
7816                 return bpf_get_skb_set_tunnel_proto(func_id);
7817         case BPF_FUNC_redirect:
7818                 return &bpf_redirect_proto;
7819         case BPF_FUNC_clone_redirect:
7820                 return &bpf_clone_redirect_proto;
7821         case BPF_FUNC_skb_change_tail:
7822                 return &bpf_skb_change_tail_proto;
7823         case BPF_FUNC_skb_change_head:
7824                 return &bpf_skb_change_head_proto;
7825         case BPF_FUNC_skb_store_bytes:
7826                 return &bpf_skb_store_bytes_proto;
7827         case BPF_FUNC_csum_update:
7828                 return &bpf_csum_update_proto;
7829         case BPF_FUNC_csum_level:
7830                 return &bpf_csum_level_proto;
7831         case BPF_FUNC_l3_csum_replace:
7832                 return &bpf_l3_csum_replace_proto;
7833         case BPF_FUNC_l4_csum_replace:
7834                 return &bpf_l4_csum_replace_proto;
7835         case BPF_FUNC_set_hash_invalid:
7836                 return &bpf_set_hash_invalid_proto;
7837         case BPF_FUNC_lwt_push_encap:
7838                 return &bpf_lwt_xmit_push_encap_proto;
7839         default:
7840                 return lwt_out_func_proto(func_id, prog);
7841         }
7842 }
7843
7844 static const struct bpf_func_proto *
7845 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7846 {
7847         switch (func_id) {
7848 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7849         case BPF_FUNC_lwt_seg6_store_bytes:
7850                 return &bpf_lwt_seg6_store_bytes_proto;
7851         case BPF_FUNC_lwt_seg6_action:
7852                 return &bpf_lwt_seg6_action_proto;
7853         case BPF_FUNC_lwt_seg6_adjust_srh:
7854                 return &bpf_lwt_seg6_adjust_srh_proto;
7855 #endif
7856         default:
7857                 return lwt_out_func_proto(func_id, prog);
7858         }
7859 }
7860
7861 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7862                                     const struct bpf_prog *prog,
7863                                     struct bpf_insn_access_aux *info)
7864 {
7865         const int size_default = sizeof(__u32);
7866
7867         if (off < 0 || off >= sizeof(struct __sk_buff))
7868                 return false;
7869
7870         /* The verifier guarantees that size > 0. */
7871         if (off % size != 0)
7872                 return false;
7873
7874         switch (off) {
7875         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7876                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7877                         return false;
7878                 break;
7879         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7880         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7881         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7882         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7883         case bpf_ctx_range(struct __sk_buff, data):
7884         case bpf_ctx_range(struct __sk_buff, data_meta):
7885         case bpf_ctx_range(struct __sk_buff, data_end):
7886                 if (size != size_default)
7887                         return false;
7888                 break;
7889         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7890                 return false;
7891         case bpf_ctx_range(struct __sk_buff, tstamp):
7892                 if (size != sizeof(__u64))
7893                         return false;
7894                 break;
7895         case offsetof(struct __sk_buff, sk):
7896                 if (type == BPF_WRITE || size != sizeof(__u64))
7897                         return false;
7898                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7899                 break;
7900         default:
7901                 /* Only narrow read access allowed for now. */
7902                 if (type == BPF_WRITE) {
7903                         if (size != size_default)
7904                                 return false;
7905                 } else {
7906                         bpf_ctx_record_field_size(info, size_default);
7907                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7908                                 return false;
7909                 }
7910         }
7911
7912         return true;
7913 }
7914
7915 static bool sk_filter_is_valid_access(int off, int size,
7916                                       enum bpf_access_type type,
7917                                       const struct bpf_prog *prog,
7918                                       struct bpf_insn_access_aux *info)
7919 {
7920         switch (off) {
7921         case bpf_ctx_range(struct __sk_buff, tc_classid):
7922         case bpf_ctx_range(struct __sk_buff, data):
7923         case bpf_ctx_range(struct __sk_buff, data_meta):
7924         case bpf_ctx_range(struct __sk_buff, data_end):
7925         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7926         case bpf_ctx_range(struct __sk_buff, tstamp):
7927         case bpf_ctx_range(struct __sk_buff, wire_len):
7928                 return false;
7929         }
7930
7931         if (type == BPF_WRITE) {
7932                 switch (off) {
7933                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7934                         break;
7935                 default:
7936                         return false;
7937                 }
7938         }
7939
7940         return bpf_skb_is_valid_access(off, size, type, prog, info);
7941 }
7942
7943 static bool cg_skb_is_valid_access(int off, int size,
7944                                    enum bpf_access_type type,
7945                                    const struct bpf_prog *prog,
7946                                    struct bpf_insn_access_aux *info)
7947 {
7948         switch (off) {
7949         case bpf_ctx_range(struct __sk_buff, tc_classid):
7950         case bpf_ctx_range(struct __sk_buff, data_meta):
7951         case bpf_ctx_range(struct __sk_buff, wire_len):
7952                 return false;
7953         case bpf_ctx_range(struct __sk_buff, data):
7954         case bpf_ctx_range(struct __sk_buff, data_end):
7955                 if (!bpf_capable())
7956                         return false;
7957                 break;
7958         }
7959
7960         if (type == BPF_WRITE) {
7961                 switch (off) {
7962                 case bpf_ctx_range(struct __sk_buff, mark):
7963                 case bpf_ctx_range(struct __sk_buff, priority):
7964                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7965                         break;
7966                 case bpf_ctx_range(struct __sk_buff, tstamp):
7967                         if (!bpf_capable())
7968                                 return false;
7969                         break;
7970                 default:
7971                         return false;
7972                 }
7973         }
7974
7975         switch (off) {
7976         case bpf_ctx_range(struct __sk_buff, data):
7977                 info->reg_type = PTR_TO_PACKET;
7978                 break;
7979         case bpf_ctx_range(struct __sk_buff, data_end):
7980                 info->reg_type = PTR_TO_PACKET_END;
7981                 break;
7982         }
7983
7984         return bpf_skb_is_valid_access(off, size, type, prog, info);
7985 }
7986
7987 static bool lwt_is_valid_access(int off, int size,
7988                                 enum bpf_access_type type,
7989                                 const struct bpf_prog *prog,
7990                                 struct bpf_insn_access_aux *info)
7991 {
7992         switch (off) {
7993         case bpf_ctx_range(struct __sk_buff, tc_classid):
7994         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7995         case bpf_ctx_range(struct __sk_buff, data_meta):
7996         case bpf_ctx_range(struct __sk_buff, tstamp):
7997         case bpf_ctx_range(struct __sk_buff, wire_len):
7998                 return false;
7999         }
8000
8001         if (type == BPF_WRITE) {
8002                 switch (off) {
8003                 case bpf_ctx_range(struct __sk_buff, mark):
8004                 case bpf_ctx_range(struct __sk_buff, priority):
8005                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8006                         break;
8007                 default:
8008                         return false;
8009                 }
8010         }
8011
8012         switch (off) {
8013         case bpf_ctx_range(struct __sk_buff, data):
8014                 info->reg_type = PTR_TO_PACKET;
8015                 break;
8016         case bpf_ctx_range(struct __sk_buff, data_end):
8017                 info->reg_type = PTR_TO_PACKET_END;
8018                 break;
8019         }
8020
8021         return bpf_skb_is_valid_access(off, size, type, prog, info);
8022 }
8023
8024 /* Attach type specific accesses */
8025 static bool __sock_filter_check_attach_type(int off,
8026                                             enum bpf_access_type access_type,
8027                                             enum bpf_attach_type attach_type)
8028 {
8029         switch (off) {
8030         case offsetof(struct bpf_sock, bound_dev_if):
8031         case offsetof(struct bpf_sock, mark):
8032         case offsetof(struct bpf_sock, priority):
8033                 switch (attach_type) {
8034                 case BPF_CGROUP_INET_SOCK_CREATE:
8035                 case BPF_CGROUP_INET_SOCK_RELEASE:
8036                         goto full_access;
8037                 default:
8038                         return false;
8039                 }
8040         case bpf_ctx_range(struct bpf_sock, src_ip4):
8041                 switch (attach_type) {
8042                 case BPF_CGROUP_INET4_POST_BIND:
8043                         goto read_only;
8044                 default:
8045                         return false;
8046                 }
8047         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8048                 switch (attach_type) {
8049                 case BPF_CGROUP_INET6_POST_BIND:
8050                         goto read_only;
8051                 default:
8052                         return false;
8053                 }
8054         case bpf_ctx_range(struct bpf_sock, src_port):
8055                 switch (attach_type) {
8056                 case BPF_CGROUP_INET4_POST_BIND:
8057                 case BPF_CGROUP_INET6_POST_BIND:
8058                         goto read_only;
8059                 default:
8060                         return false;
8061                 }
8062         }
8063 read_only:
8064         return access_type == BPF_READ;
8065 full_access:
8066         return true;
8067 }
8068
8069 bool bpf_sock_common_is_valid_access(int off, int size,
8070                                      enum bpf_access_type type,
8071                                      struct bpf_insn_access_aux *info)
8072 {
8073         switch (off) {
8074         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8075                 return false;
8076         default:
8077                 return bpf_sock_is_valid_access(off, size, type, info);
8078         }
8079 }
8080
8081 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8082                               struct bpf_insn_access_aux *info)
8083 {
8084         const int size_default = sizeof(__u32);
8085         int field_size;
8086
8087         if (off < 0 || off >= sizeof(struct bpf_sock))
8088                 return false;
8089         if (off % size != 0)
8090                 return false;
8091
8092         switch (off) {
8093         case offsetof(struct bpf_sock, state):
8094         case offsetof(struct bpf_sock, family):
8095         case offsetof(struct bpf_sock, type):
8096         case offsetof(struct bpf_sock, protocol):
8097         case offsetof(struct bpf_sock, src_port):
8098         case offsetof(struct bpf_sock, rx_queue_mapping):
8099         case bpf_ctx_range(struct bpf_sock, src_ip4):
8100         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8101         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8102         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8103                 bpf_ctx_record_field_size(info, size_default);
8104                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8105         case bpf_ctx_range(struct bpf_sock, dst_port):
8106                 field_size = size == size_default ?
8107                         size_default : sizeof_field(struct bpf_sock, dst_port);
8108                 bpf_ctx_record_field_size(info, field_size);
8109                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8110         case offsetofend(struct bpf_sock, dst_port) ...
8111              offsetof(struct bpf_sock, dst_ip4) - 1:
8112                 return false;
8113         }
8114
8115         return size == size_default;
8116 }
8117
8118 static bool sock_filter_is_valid_access(int off, int size,
8119                                         enum bpf_access_type type,
8120                                         const struct bpf_prog *prog,
8121                                         struct bpf_insn_access_aux *info)
8122 {
8123         if (!bpf_sock_is_valid_access(off, size, type, info))
8124                 return false;
8125         return __sock_filter_check_attach_type(off, type,
8126                                                prog->expected_attach_type);
8127 }
8128
8129 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8130                              const struct bpf_prog *prog)
8131 {
8132         /* Neither direct read nor direct write requires any preliminary
8133          * action.
8134          */
8135         return 0;
8136 }
8137
8138 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8139                                 const struct bpf_prog *prog, int drop_verdict)
8140 {
8141         struct bpf_insn *insn = insn_buf;
8142
8143         if (!direct_write)
8144                 return 0;
8145
8146         /* if (!skb->cloned)
8147          *       goto start;
8148          *
8149          * (Fast-path, otherwise approximation that we might be
8150          *  a clone, do the rest in helper.)
8151          */
8152         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
8153         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8154         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8155
8156         /* ret = bpf_skb_pull_data(skb, 0); */
8157         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8158         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8159         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8160                                BPF_FUNC_skb_pull_data);
8161         /* if (!ret)
8162          *      goto restore;
8163          * return TC_ACT_SHOT;
8164          */
8165         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8166         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8167         *insn++ = BPF_EXIT_INSN();
8168
8169         /* restore: */
8170         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8171         /* start: */
8172         *insn++ = prog->insnsi[0];
8173
8174         return insn - insn_buf;
8175 }
8176
8177 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8178                           struct bpf_insn *insn_buf)
8179 {
8180         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8181         struct bpf_insn *insn = insn_buf;
8182
8183         if (!indirect) {
8184                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8185         } else {
8186                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8187                 if (orig->imm)
8188                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8189         }
8190         /* We're guaranteed here that CTX is in R6. */
8191         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8192
8193         switch (BPF_SIZE(orig->code)) {
8194         case BPF_B:
8195                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8196                 break;
8197         case BPF_H:
8198                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8199                 break;
8200         case BPF_W:
8201                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8202                 break;
8203         }
8204
8205         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8206         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8207         *insn++ = BPF_EXIT_INSN();
8208
8209         return insn - insn_buf;
8210 }
8211
8212 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8213                                const struct bpf_prog *prog)
8214 {
8215         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8216 }
8217
8218 static bool tc_cls_act_is_valid_access(int off, int size,
8219                                        enum bpf_access_type type,
8220                                        const struct bpf_prog *prog,
8221                                        struct bpf_insn_access_aux *info)
8222 {
8223         if (type == BPF_WRITE) {
8224                 switch (off) {
8225                 case bpf_ctx_range(struct __sk_buff, mark):
8226                 case bpf_ctx_range(struct __sk_buff, tc_index):
8227                 case bpf_ctx_range(struct __sk_buff, priority):
8228                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8229                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8230                 case bpf_ctx_range(struct __sk_buff, tstamp):
8231                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8232                         break;
8233                 default:
8234                         return false;
8235                 }
8236         }
8237
8238         switch (off) {
8239         case bpf_ctx_range(struct __sk_buff, data):
8240                 info->reg_type = PTR_TO_PACKET;
8241                 break;
8242         case bpf_ctx_range(struct __sk_buff, data_meta):
8243                 info->reg_type = PTR_TO_PACKET_META;
8244                 break;
8245         case bpf_ctx_range(struct __sk_buff, data_end):
8246                 info->reg_type = PTR_TO_PACKET_END;
8247                 break;
8248         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8249                 return false;
8250         }
8251
8252         return bpf_skb_is_valid_access(off, size, type, prog, info);
8253 }
8254
8255 static bool __is_valid_xdp_access(int off, int size)
8256 {
8257         if (off < 0 || off >= sizeof(struct xdp_md))
8258                 return false;
8259         if (off % size != 0)
8260                 return false;
8261         if (size != sizeof(__u32))
8262                 return false;
8263
8264         return true;
8265 }
8266
8267 static bool xdp_is_valid_access(int off, int size,
8268                                 enum bpf_access_type type,
8269                                 const struct bpf_prog *prog,
8270                                 struct bpf_insn_access_aux *info)
8271 {
8272         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8273                 switch (off) {
8274                 case offsetof(struct xdp_md, egress_ifindex):
8275                         return false;
8276                 }
8277         }
8278
8279         if (type == BPF_WRITE) {
8280                 if (bpf_prog_is_dev_bound(prog->aux)) {
8281                         switch (off) {
8282                         case offsetof(struct xdp_md, rx_queue_index):
8283                                 return __is_valid_xdp_access(off, size);
8284                         }
8285                 }
8286                 return false;
8287         }
8288
8289         switch (off) {
8290         case offsetof(struct xdp_md, data):
8291                 info->reg_type = PTR_TO_PACKET;
8292                 break;
8293         case offsetof(struct xdp_md, data_meta):
8294                 info->reg_type = PTR_TO_PACKET_META;
8295                 break;
8296         case offsetof(struct xdp_md, data_end):
8297                 info->reg_type = PTR_TO_PACKET_END;
8298                 break;
8299         }
8300
8301         return __is_valid_xdp_access(off, size);
8302 }
8303
8304 void bpf_warn_invalid_xdp_action(u32 act)
8305 {
8306         const u32 act_max = XDP_REDIRECT;
8307
8308         pr_warn_once("%s XDP return value %u, expect packet loss!\n",
8309                      act > act_max ? "Illegal" : "Driver unsupported",
8310                      act);
8311 }
8312 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8313
8314 static bool sock_addr_is_valid_access(int off, int size,
8315                                       enum bpf_access_type type,
8316                                       const struct bpf_prog *prog,
8317                                       struct bpf_insn_access_aux *info)
8318 {
8319         const int size_default = sizeof(__u32);
8320
8321         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8322                 return false;
8323         if (off % size != 0)
8324                 return false;
8325
8326         /* Disallow access to IPv6 fields from IPv4 contex and vise
8327          * versa.
8328          */
8329         switch (off) {
8330         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8331                 switch (prog->expected_attach_type) {
8332                 case BPF_CGROUP_INET4_BIND:
8333                 case BPF_CGROUP_INET4_CONNECT:
8334                 case BPF_CGROUP_INET4_GETPEERNAME:
8335                 case BPF_CGROUP_INET4_GETSOCKNAME:
8336                 case BPF_CGROUP_UDP4_SENDMSG:
8337                 case BPF_CGROUP_UDP4_RECVMSG:
8338                         break;
8339                 default:
8340                         return false;
8341                 }
8342                 break;
8343         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8344                 switch (prog->expected_attach_type) {
8345                 case BPF_CGROUP_INET6_BIND:
8346                 case BPF_CGROUP_INET6_CONNECT:
8347                 case BPF_CGROUP_INET6_GETPEERNAME:
8348                 case BPF_CGROUP_INET6_GETSOCKNAME:
8349                 case BPF_CGROUP_UDP6_SENDMSG:
8350                 case BPF_CGROUP_UDP6_RECVMSG:
8351                         break;
8352                 default:
8353                         return false;
8354                 }
8355                 break;
8356         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8357                 switch (prog->expected_attach_type) {
8358                 case BPF_CGROUP_UDP4_SENDMSG:
8359                         break;
8360                 default:
8361                         return false;
8362                 }
8363                 break;
8364         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8365                                 msg_src_ip6[3]):
8366                 switch (prog->expected_attach_type) {
8367                 case BPF_CGROUP_UDP6_SENDMSG:
8368                         break;
8369                 default:
8370                         return false;
8371                 }
8372                 break;
8373         }
8374
8375         switch (off) {
8376         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8377         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8378         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8379         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8380                                 msg_src_ip6[3]):
8381         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8382                 if (type == BPF_READ) {
8383                         bpf_ctx_record_field_size(info, size_default);
8384
8385                         if (bpf_ctx_wide_access_ok(off, size,
8386                                                    struct bpf_sock_addr,
8387                                                    user_ip6))
8388                                 return true;
8389
8390                         if (bpf_ctx_wide_access_ok(off, size,
8391                                                    struct bpf_sock_addr,
8392                                                    msg_src_ip6))
8393                                 return true;
8394
8395                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8396                                 return false;
8397                 } else {
8398                         if (bpf_ctx_wide_access_ok(off, size,
8399                                                    struct bpf_sock_addr,
8400                                                    user_ip6))
8401                                 return true;
8402
8403                         if (bpf_ctx_wide_access_ok(off, size,
8404                                                    struct bpf_sock_addr,
8405                                                    msg_src_ip6))
8406                                 return true;
8407
8408                         if (size != size_default)
8409                                 return false;
8410                 }
8411                 break;
8412         case offsetof(struct bpf_sock_addr, sk):
8413                 if (type != BPF_READ)
8414                         return false;
8415                 if (size != sizeof(__u64))
8416                         return false;
8417                 info->reg_type = PTR_TO_SOCKET;
8418                 break;
8419         default:
8420                 if (type == BPF_READ) {
8421                         if (size != size_default)
8422                                 return false;
8423                 } else {
8424                         return false;
8425                 }
8426         }
8427
8428         return true;
8429 }
8430
8431 static bool sock_ops_is_valid_access(int off, int size,
8432                                      enum bpf_access_type type,
8433                                      const struct bpf_prog *prog,
8434                                      struct bpf_insn_access_aux *info)
8435 {
8436         const int size_default = sizeof(__u32);
8437
8438         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8439                 return false;
8440
8441         /* The verifier guarantees that size > 0. */
8442         if (off % size != 0)
8443                 return false;
8444
8445         if (type == BPF_WRITE) {
8446                 switch (off) {
8447                 case offsetof(struct bpf_sock_ops, reply):
8448                 case offsetof(struct bpf_sock_ops, sk_txhash):
8449                         if (size != size_default)
8450                                 return false;
8451                         break;
8452                 default:
8453                         return false;
8454                 }
8455         } else {
8456                 switch (off) {
8457                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8458                                         bytes_acked):
8459                         if (size != sizeof(__u64))
8460                                 return false;
8461                         break;
8462                 case offsetof(struct bpf_sock_ops, sk):
8463                         if (size != sizeof(__u64))
8464                                 return false;
8465                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8466                         break;
8467                 case offsetof(struct bpf_sock_ops, skb_data):
8468                         if (size != sizeof(__u64))
8469                                 return false;
8470                         info->reg_type = PTR_TO_PACKET;
8471                         break;
8472                 case offsetof(struct bpf_sock_ops, skb_data_end):
8473                         if (size != sizeof(__u64))
8474                                 return false;
8475                         info->reg_type = PTR_TO_PACKET_END;
8476                         break;
8477                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8478                         bpf_ctx_record_field_size(info, size_default);
8479                         return bpf_ctx_narrow_access_ok(off, size,
8480                                                         size_default);
8481                 default:
8482                         if (size != size_default)
8483                                 return false;
8484                         break;
8485                 }
8486         }
8487
8488         return true;
8489 }
8490
8491 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8492                            const struct bpf_prog *prog)
8493 {
8494         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8495 }
8496
8497 static bool sk_skb_is_valid_access(int off, int size,
8498                                    enum bpf_access_type type,
8499                                    const struct bpf_prog *prog,
8500                                    struct bpf_insn_access_aux *info)
8501 {
8502         switch (off) {
8503         case bpf_ctx_range(struct __sk_buff, tc_classid):
8504         case bpf_ctx_range(struct __sk_buff, data_meta):
8505         case bpf_ctx_range(struct __sk_buff, tstamp):
8506         case bpf_ctx_range(struct __sk_buff, wire_len):
8507                 return false;
8508         }
8509
8510         if (type == BPF_WRITE) {
8511                 switch (off) {
8512                 case bpf_ctx_range(struct __sk_buff, tc_index):
8513                 case bpf_ctx_range(struct __sk_buff, priority):
8514                         break;
8515                 default:
8516                         return false;
8517                 }
8518         }
8519
8520         switch (off) {
8521         case bpf_ctx_range(struct __sk_buff, mark):
8522                 return false;
8523         case bpf_ctx_range(struct __sk_buff, data):
8524                 info->reg_type = PTR_TO_PACKET;
8525                 break;
8526         case bpf_ctx_range(struct __sk_buff, data_end):
8527                 info->reg_type = PTR_TO_PACKET_END;
8528                 break;
8529         }
8530
8531         return bpf_skb_is_valid_access(off, size, type, prog, info);
8532 }
8533
8534 static bool sk_msg_is_valid_access(int off, int size,
8535                                    enum bpf_access_type type,
8536                                    const struct bpf_prog *prog,
8537                                    struct bpf_insn_access_aux *info)
8538 {
8539         if (type == BPF_WRITE)
8540                 return false;
8541
8542         if (off % size != 0)
8543                 return false;
8544
8545         switch (off) {
8546         case offsetof(struct sk_msg_md, data):
8547                 info->reg_type = PTR_TO_PACKET;
8548                 if (size != sizeof(__u64))
8549                         return false;
8550                 break;
8551         case offsetof(struct sk_msg_md, data_end):
8552                 info->reg_type = PTR_TO_PACKET_END;
8553                 if (size != sizeof(__u64))
8554                         return false;
8555                 break;
8556         case offsetof(struct sk_msg_md, sk):
8557                 if (size != sizeof(__u64))
8558                         return false;
8559                 info->reg_type = PTR_TO_SOCKET;
8560                 break;
8561         case bpf_ctx_range(struct sk_msg_md, family):
8562         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8563         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8564         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8565         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8566         case bpf_ctx_range(struct sk_msg_md, remote_port):
8567         case bpf_ctx_range(struct sk_msg_md, local_port):
8568         case bpf_ctx_range(struct sk_msg_md, size):
8569                 if (size != sizeof(__u32))
8570                         return false;
8571                 break;
8572         default:
8573                 return false;
8574         }
8575         return true;
8576 }
8577
8578 static bool flow_dissector_is_valid_access(int off, int size,
8579                                            enum bpf_access_type type,
8580                                            const struct bpf_prog *prog,
8581                                            struct bpf_insn_access_aux *info)
8582 {
8583         const int size_default = sizeof(__u32);
8584
8585         if (off < 0 || off >= sizeof(struct __sk_buff))
8586                 return false;
8587
8588         if (type == BPF_WRITE)
8589                 return false;
8590
8591         switch (off) {
8592         case bpf_ctx_range(struct __sk_buff, data):
8593                 if (size != size_default)
8594                         return false;
8595                 info->reg_type = PTR_TO_PACKET;
8596                 return true;
8597         case bpf_ctx_range(struct __sk_buff, data_end):
8598                 if (size != size_default)
8599                         return false;
8600                 info->reg_type = PTR_TO_PACKET_END;
8601                 return true;
8602         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8603                 if (size != sizeof(__u64))
8604                         return false;
8605                 info->reg_type = PTR_TO_FLOW_KEYS;
8606                 return true;
8607         default:
8608                 return false;
8609         }
8610 }
8611
8612 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8613                                              const struct bpf_insn *si,
8614                                              struct bpf_insn *insn_buf,
8615                                              struct bpf_prog *prog,
8616                                              u32 *target_size)
8617
8618 {
8619         struct bpf_insn *insn = insn_buf;
8620
8621         switch (si->off) {
8622         case offsetof(struct __sk_buff, data):
8623                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8624                                       si->dst_reg, si->src_reg,
8625                                       offsetof(struct bpf_flow_dissector, data));
8626                 break;
8627
8628         case offsetof(struct __sk_buff, data_end):
8629                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8630                                       si->dst_reg, si->src_reg,
8631                                       offsetof(struct bpf_flow_dissector, data_end));
8632                 break;
8633
8634         case offsetof(struct __sk_buff, flow_keys):
8635                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8636                                       si->dst_reg, si->src_reg,
8637                                       offsetof(struct bpf_flow_dissector, flow_keys));
8638                 break;
8639         }
8640
8641         return insn - insn_buf;
8642 }
8643
8644 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8645                                                   struct bpf_insn *insn)
8646 {
8647         /* si->dst_reg = skb_shinfo(SKB); */
8648 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8649         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8650                               BPF_REG_AX, si->src_reg,
8651                               offsetof(struct sk_buff, end));
8652         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8653                               si->dst_reg, si->src_reg,
8654                               offsetof(struct sk_buff, head));
8655         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8656 #else
8657         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8658                               si->dst_reg, si->src_reg,
8659                               offsetof(struct sk_buff, end));
8660 #endif
8661
8662         return insn;
8663 }
8664
8665 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8666                                   const struct bpf_insn *si,
8667                                   struct bpf_insn *insn_buf,
8668                                   struct bpf_prog *prog, u32 *target_size)
8669 {
8670         struct bpf_insn *insn = insn_buf;
8671         int off;
8672
8673         switch (si->off) {
8674         case offsetof(struct __sk_buff, len):
8675                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8676                                       bpf_target_off(struct sk_buff, len, 4,
8677                                                      target_size));
8678                 break;
8679
8680         case offsetof(struct __sk_buff, protocol):
8681                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8682                                       bpf_target_off(struct sk_buff, protocol, 2,
8683                                                      target_size));
8684                 break;
8685
8686         case offsetof(struct __sk_buff, vlan_proto):
8687                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8688                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
8689                                                      target_size));
8690                 break;
8691
8692         case offsetof(struct __sk_buff, priority):
8693                 if (type == BPF_WRITE)
8694                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8695                                               bpf_target_off(struct sk_buff, priority, 4,
8696                                                              target_size));
8697                 else
8698                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8699                                               bpf_target_off(struct sk_buff, priority, 4,
8700                                                              target_size));
8701                 break;
8702
8703         case offsetof(struct __sk_buff, ingress_ifindex):
8704                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8705                                       bpf_target_off(struct sk_buff, skb_iif, 4,
8706                                                      target_size));
8707                 break;
8708
8709         case offsetof(struct __sk_buff, ifindex):
8710                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8711                                       si->dst_reg, si->src_reg,
8712                                       offsetof(struct sk_buff, dev));
8713                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8714                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8715                                       bpf_target_off(struct net_device, ifindex, 4,
8716                                                      target_size));
8717                 break;
8718
8719         case offsetof(struct __sk_buff, hash):
8720                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8721                                       bpf_target_off(struct sk_buff, hash, 4,
8722                                                      target_size));
8723                 break;
8724
8725         case offsetof(struct __sk_buff, mark):
8726                 if (type == BPF_WRITE)
8727                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8728                                               bpf_target_off(struct sk_buff, mark, 4,
8729                                                              target_size));
8730                 else
8731                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8732                                               bpf_target_off(struct sk_buff, mark, 4,
8733                                                              target_size));
8734                 break;
8735
8736         case offsetof(struct __sk_buff, pkt_type):
8737                 *target_size = 1;
8738                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8739                                       PKT_TYPE_OFFSET());
8740                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8741 #ifdef __BIG_ENDIAN_BITFIELD
8742                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8743 #endif
8744                 break;
8745
8746         case offsetof(struct __sk_buff, queue_mapping):
8747                 if (type == BPF_WRITE) {
8748                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8749                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8750                                               bpf_target_off(struct sk_buff,
8751                                                              queue_mapping,
8752                                                              2, target_size));
8753                 } else {
8754                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8755                                               bpf_target_off(struct sk_buff,
8756                                                              queue_mapping,
8757                                                              2, target_size));
8758                 }
8759                 break;
8760
8761         case offsetof(struct __sk_buff, vlan_present):
8762                 *target_size = 1;
8763                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8764                                       PKT_VLAN_PRESENT_OFFSET());
8765                 if (PKT_VLAN_PRESENT_BIT)
8766                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8767                 if (PKT_VLAN_PRESENT_BIT < 7)
8768                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8769                 break;
8770
8771         case offsetof(struct __sk_buff, vlan_tci):
8772                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8773                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
8774                                                      target_size));
8775                 break;
8776
8777         case offsetof(struct __sk_buff, cb[0]) ...
8778              offsetofend(struct __sk_buff, cb[4]) - 1:
8779                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8780                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8781                               offsetof(struct qdisc_skb_cb, data)) %
8782                              sizeof(__u64));
8783
8784                 prog->cb_access = 1;
8785                 off  = si->off;
8786                 off -= offsetof(struct __sk_buff, cb[0]);
8787                 off += offsetof(struct sk_buff, cb);
8788                 off += offsetof(struct qdisc_skb_cb, data);
8789                 if (type == BPF_WRITE)
8790                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8791                                               si->src_reg, off);
8792                 else
8793                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8794                                               si->src_reg, off);
8795                 break;
8796
8797         case offsetof(struct __sk_buff, tc_classid):
8798                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8799
8800                 off  = si->off;
8801                 off -= offsetof(struct __sk_buff, tc_classid);
8802                 off += offsetof(struct sk_buff, cb);
8803                 off += offsetof(struct qdisc_skb_cb, tc_classid);
8804                 *target_size = 2;
8805                 if (type == BPF_WRITE)
8806                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8807                                               si->src_reg, off);
8808                 else
8809                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8810                                               si->src_reg, off);
8811                 break;
8812
8813         case offsetof(struct __sk_buff, data):
8814                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8815                                       si->dst_reg, si->src_reg,
8816                                       offsetof(struct sk_buff, data));
8817                 break;
8818
8819         case offsetof(struct __sk_buff, data_meta):
8820                 off  = si->off;
8821                 off -= offsetof(struct __sk_buff, data_meta);
8822                 off += offsetof(struct sk_buff, cb);
8823                 off += offsetof(struct bpf_skb_data_end, data_meta);
8824                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8825                                       si->src_reg, off);
8826                 break;
8827
8828         case offsetof(struct __sk_buff, data_end):
8829                 off  = si->off;
8830                 off -= offsetof(struct __sk_buff, data_end);
8831                 off += offsetof(struct sk_buff, cb);
8832                 off += offsetof(struct bpf_skb_data_end, data_end);
8833                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8834                                       si->src_reg, off);
8835                 break;
8836
8837         case offsetof(struct __sk_buff, tc_index):
8838 #ifdef CONFIG_NET_SCHED
8839                 if (type == BPF_WRITE)
8840                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8841                                               bpf_target_off(struct sk_buff, tc_index, 2,
8842                                                              target_size));
8843                 else
8844                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8845                                               bpf_target_off(struct sk_buff, tc_index, 2,
8846                                                              target_size));
8847 #else
8848                 *target_size = 2;
8849                 if (type == BPF_WRITE)
8850                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8851                 else
8852                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8853 #endif
8854                 break;
8855
8856         case offsetof(struct __sk_buff, napi_id):
8857 #if defined(CONFIG_NET_RX_BUSY_POLL)
8858                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8859                                       bpf_target_off(struct sk_buff, napi_id, 4,
8860                                                      target_size));
8861                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8862                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8863 #else
8864                 *target_size = 4;
8865                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8866 #endif
8867                 break;
8868         case offsetof(struct __sk_buff, family):
8869                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8870
8871                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8872                                       si->dst_reg, si->src_reg,
8873                                       offsetof(struct sk_buff, sk));
8874                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8875                                       bpf_target_off(struct sock_common,
8876                                                      skc_family,
8877                                                      2, target_size));
8878                 break;
8879         case offsetof(struct __sk_buff, remote_ip4):
8880                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8881
8882                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8883                                       si->dst_reg, si->src_reg,
8884                                       offsetof(struct sk_buff, sk));
8885                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8886                                       bpf_target_off(struct sock_common,
8887                                                      skc_daddr,
8888                                                      4, target_size));
8889                 break;
8890         case offsetof(struct __sk_buff, local_ip4):
8891                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8892                                           skc_rcv_saddr) != 4);
8893
8894                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8895                                       si->dst_reg, si->src_reg,
8896                                       offsetof(struct sk_buff, sk));
8897                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8898                                       bpf_target_off(struct sock_common,
8899                                                      skc_rcv_saddr,
8900                                                      4, target_size));
8901                 break;
8902         case offsetof(struct __sk_buff, remote_ip6[0]) ...
8903              offsetof(struct __sk_buff, remote_ip6[3]):
8904 #if IS_ENABLED(CONFIG_IPV6)
8905                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8906                                           skc_v6_daddr.s6_addr32[0]) != 4);
8907
8908                 off = si->off;
8909                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8910
8911                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8912                                       si->dst_reg, si->src_reg,
8913                                       offsetof(struct sk_buff, sk));
8914                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8915                                       offsetof(struct sock_common,
8916                                                skc_v6_daddr.s6_addr32[0]) +
8917                                       off);
8918 #else
8919                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8920 #endif
8921                 break;
8922         case offsetof(struct __sk_buff, local_ip6[0]) ...
8923              offsetof(struct __sk_buff, local_ip6[3]):
8924 #if IS_ENABLED(CONFIG_IPV6)
8925                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8926                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8927
8928                 off = si->off;
8929                 off -= offsetof(struct __sk_buff, local_ip6[0]);
8930
8931                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8932                                       si->dst_reg, si->src_reg,
8933                                       offsetof(struct sk_buff, sk));
8934                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8935                                       offsetof(struct sock_common,
8936                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8937                                       off);
8938 #else
8939                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8940 #endif
8941                 break;
8942
8943         case offsetof(struct __sk_buff, remote_port):
8944                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8945
8946                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8947                                       si->dst_reg, si->src_reg,
8948                                       offsetof(struct sk_buff, sk));
8949                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8950                                       bpf_target_off(struct sock_common,
8951                                                      skc_dport,
8952                                                      2, target_size));
8953 #ifndef __BIG_ENDIAN_BITFIELD
8954                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8955 #endif
8956                 break;
8957
8958         case offsetof(struct __sk_buff, local_port):
8959                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8960
8961                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8962                                       si->dst_reg, si->src_reg,
8963                                       offsetof(struct sk_buff, sk));
8964                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8965                                       bpf_target_off(struct sock_common,
8966                                                      skc_num, 2, target_size));
8967                 break;
8968
8969         case offsetof(struct __sk_buff, tstamp):
8970                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8971
8972                 if (type == BPF_WRITE)
8973                         *insn++ = BPF_STX_MEM(BPF_DW,
8974                                               si->dst_reg, si->src_reg,
8975                                               bpf_target_off(struct sk_buff,
8976                                                              tstamp, 8,
8977                                                              target_size));
8978                 else
8979                         *insn++ = BPF_LDX_MEM(BPF_DW,
8980                                               si->dst_reg, si->src_reg,
8981                                               bpf_target_off(struct sk_buff,
8982                                                              tstamp, 8,
8983                                                              target_size));
8984                 break;
8985
8986         case offsetof(struct __sk_buff, gso_segs):
8987                 insn = bpf_convert_shinfo_access(si, insn);
8988                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8989                                       si->dst_reg, si->dst_reg,
8990                                       bpf_target_off(struct skb_shared_info,
8991                                                      gso_segs, 2,
8992                                                      target_size));
8993                 break;
8994         case offsetof(struct __sk_buff, gso_size):
8995                 insn = bpf_convert_shinfo_access(si, insn);
8996                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8997                                       si->dst_reg, si->dst_reg,
8998                                       bpf_target_off(struct skb_shared_info,
8999                                                      gso_size, 2,
9000                                                      target_size));
9001                 break;
9002         case offsetof(struct __sk_buff, wire_len):
9003                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9004
9005                 off = si->off;
9006                 off -= offsetof(struct __sk_buff, wire_len);
9007                 off += offsetof(struct sk_buff, cb);
9008                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9009                 *target_size = 4;
9010                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9011                 break;
9012
9013         case offsetof(struct __sk_buff, sk):
9014                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9015                                       si->dst_reg, si->src_reg,
9016                                       offsetof(struct sk_buff, sk));
9017                 break;
9018         }
9019
9020         return insn - insn_buf;
9021 }
9022
9023 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9024                                 const struct bpf_insn *si,
9025                                 struct bpf_insn *insn_buf,
9026                                 struct bpf_prog *prog, u32 *target_size)
9027 {
9028         struct bpf_insn *insn = insn_buf;
9029         int off;
9030
9031         switch (si->off) {
9032         case offsetof(struct bpf_sock, bound_dev_if):
9033                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9034
9035                 if (type == BPF_WRITE)
9036                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9037                                         offsetof(struct sock, sk_bound_dev_if));
9038                 else
9039                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9040                                       offsetof(struct sock, sk_bound_dev_if));
9041                 break;
9042
9043         case offsetof(struct bpf_sock, mark):
9044                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9045
9046                 if (type == BPF_WRITE)
9047                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9048                                         offsetof(struct sock, sk_mark));
9049                 else
9050                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9051                                       offsetof(struct sock, sk_mark));
9052                 break;
9053
9054         case offsetof(struct bpf_sock, priority):
9055                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9056
9057                 if (type == BPF_WRITE)
9058                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9059                                         offsetof(struct sock, sk_priority));
9060                 else
9061                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9062                                       offsetof(struct sock, sk_priority));
9063                 break;
9064
9065         case offsetof(struct bpf_sock, family):
9066                 *insn++ = BPF_LDX_MEM(
9067                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9068                         si->dst_reg, si->src_reg,
9069                         bpf_target_off(struct sock_common,
9070                                        skc_family,
9071                                        sizeof_field(struct sock_common,
9072                                                     skc_family),
9073                                        target_size));
9074                 break;
9075
9076         case offsetof(struct bpf_sock, type):
9077                 *insn++ = BPF_LDX_MEM(
9078                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9079                         si->dst_reg, si->src_reg,
9080                         bpf_target_off(struct sock, sk_type,
9081                                        sizeof_field(struct sock, sk_type),
9082                                        target_size));
9083                 break;
9084
9085         case offsetof(struct bpf_sock, protocol):
9086                 *insn++ = BPF_LDX_MEM(
9087                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9088                         si->dst_reg, si->src_reg,
9089                         bpf_target_off(struct sock, sk_protocol,
9090                                        sizeof_field(struct sock, sk_protocol),
9091                                        target_size));
9092                 break;
9093
9094         case offsetof(struct bpf_sock, src_ip4):
9095                 *insn++ = BPF_LDX_MEM(
9096                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9097                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9098                                        sizeof_field(struct sock_common,
9099                                                     skc_rcv_saddr),
9100                                        target_size));
9101                 break;
9102
9103         case offsetof(struct bpf_sock, dst_ip4):
9104                 *insn++ = BPF_LDX_MEM(
9105                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9106                         bpf_target_off(struct sock_common, skc_daddr,
9107                                        sizeof_field(struct sock_common,
9108                                                     skc_daddr),
9109                                        target_size));
9110                 break;
9111
9112         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9113 #if IS_ENABLED(CONFIG_IPV6)
9114                 off = si->off;
9115                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9116                 *insn++ = BPF_LDX_MEM(
9117                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9118                         bpf_target_off(
9119                                 struct sock_common,
9120                                 skc_v6_rcv_saddr.s6_addr32[0],
9121                                 sizeof_field(struct sock_common,
9122                                              skc_v6_rcv_saddr.s6_addr32[0]),
9123                                 target_size) + off);
9124 #else
9125                 (void)off;
9126                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9127 #endif
9128                 break;
9129
9130         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9131 #if IS_ENABLED(CONFIG_IPV6)
9132                 off = si->off;
9133                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9134                 *insn++ = BPF_LDX_MEM(
9135                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9136                         bpf_target_off(struct sock_common,
9137                                        skc_v6_daddr.s6_addr32[0],
9138                                        sizeof_field(struct sock_common,
9139                                                     skc_v6_daddr.s6_addr32[0]),
9140                                        target_size) + off);
9141 #else
9142                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9143                 *target_size = 4;
9144 #endif
9145                 break;
9146
9147         case offsetof(struct bpf_sock, src_port):
9148                 *insn++ = BPF_LDX_MEM(
9149                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9150                         si->dst_reg, si->src_reg,
9151                         bpf_target_off(struct sock_common, skc_num,
9152                                        sizeof_field(struct sock_common,
9153                                                     skc_num),
9154                                        target_size));
9155                 break;
9156
9157         case offsetof(struct bpf_sock, dst_port):
9158                 *insn++ = BPF_LDX_MEM(
9159                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9160                         si->dst_reg, si->src_reg,
9161                         bpf_target_off(struct sock_common, skc_dport,
9162                                        sizeof_field(struct sock_common,
9163                                                     skc_dport),
9164                                        target_size));
9165                 break;
9166
9167         case offsetof(struct bpf_sock, state):
9168                 *insn++ = BPF_LDX_MEM(
9169                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9170                         si->dst_reg, si->src_reg,
9171                         bpf_target_off(struct sock_common, skc_state,
9172                                        sizeof_field(struct sock_common,
9173                                                     skc_state),
9174                                        target_size));
9175                 break;
9176         case offsetof(struct bpf_sock, rx_queue_mapping):
9177 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9178                 *insn++ = BPF_LDX_MEM(
9179                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9180                         si->dst_reg, si->src_reg,
9181                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9182                                        sizeof_field(struct sock,
9183                                                     sk_rx_queue_mapping),
9184                                        target_size));
9185                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9186                                       1);
9187                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9188 #else
9189                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9190                 *target_size = 2;
9191 #endif
9192                 break;
9193         }
9194
9195         return insn - insn_buf;
9196 }
9197
9198 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9199                                          const struct bpf_insn *si,
9200                                          struct bpf_insn *insn_buf,
9201                                          struct bpf_prog *prog, u32 *target_size)
9202 {
9203         struct bpf_insn *insn = insn_buf;
9204
9205         switch (si->off) {
9206         case offsetof(struct __sk_buff, ifindex):
9207                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9208                                       si->dst_reg, si->src_reg,
9209                                       offsetof(struct sk_buff, dev));
9210                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9211                                       bpf_target_off(struct net_device, ifindex, 4,
9212                                                      target_size));
9213                 break;
9214         default:
9215                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9216                                               target_size);
9217         }
9218
9219         return insn - insn_buf;
9220 }
9221
9222 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9223                                   const struct bpf_insn *si,
9224                                   struct bpf_insn *insn_buf,
9225                                   struct bpf_prog *prog, u32 *target_size)
9226 {
9227         struct bpf_insn *insn = insn_buf;
9228
9229         switch (si->off) {
9230         case offsetof(struct xdp_md, data):
9231                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9232                                       si->dst_reg, si->src_reg,
9233                                       offsetof(struct xdp_buff, data));
9234                 break;
9235         case offsetof(struct xdp_md, data_meta):
9236                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9237                                       si->dst_reg, si->src_reg,
9238                                       offsetof(struct xdp_buff, data_meta));
9239                 break;
9240         case offsetof(struct xdp_md, data_end):
9241                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9242                                       si->dst_reg, si->src_reg,
9243                                       offsetof(struct xdp_buff, data_end));
9244                 break;
9245         case offsetof(struct xdp_md, ingress_ifindex):
9246                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9247                                       si->dst_reg, si->src_reg,
9248                                       offsetof(struct xdp_buff, rxq));
9249                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9250                                       si->dst_reg, si->dst_reg,
9251                                       offsetof(struct xdp_rxq_info, dev));
9252                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9253                                       offsetof(struct net_device, ifindex));
9254                 break;
9255         case offsetof(struct xdp_md, rx_queue_index):
9256                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9257                                       si->dst_reg, si->src_reg,
9258                                       offsetof(struct xdp_buff, rxq));
9259                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9260                                       offsetof(struct xdp_rxq_info,
9261                                                queue_index));
9262                 break;
9263         case offsetof(struct xdp_md, egress_ifindex):
9264                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9265                                       si->dst_reg, si->src_reg,
9266                                       offsetof(struct xdp_buff, txq));
9267                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9268                                       si->dst_reg, si->dst_reg,
9269                                       offsetof(struct xdp_txq_info, dev));
9270                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9271                                       offsetof(struct net_device, ifindex));
9272                 break;
9273         }
9274
9275         return insn - insn_buf;
9276 }
9277
9278 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9279  * context Structure, F is Field in context structure that contains a pointer
9280  * to Nested Structure of type NS that has the field NF.
9281  *
9282  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9283  * sure that SIZE is not greater than actual size of S.F.NF.
9284  *
9285  * If offset OFF is provided, the load happens from that offset relative to
9286  * offset of NF.
9287  */
9288 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9289         do {                                                                   \
9290                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9291                                       si->src_reg, offsetof(S, F));            \
9292                 *insn++ = BPF_LDX_MEM(                                         \
9293                         SIZE, si->dst_reg, si->dst_reg,                        \
9294                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9295                                        target_size)                            \
9296                                 + OFF);                                        \
9297         } while (0)
9298
9299 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9300         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9301                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9302
9303 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9304  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9305  *
9306  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9307  * "register" since two registers available in convert_ctx_access are not
9308  * enough: we can't override neither SRC, since it contains value to store, nor
9309  * DST since it contains pointer to context that may be used by later
9310  * instructions. But we need a temporary place to save pointer to nested
9311  * structure whose field we want to store to.
9312  */
9313 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9314         do {                                                                   \
9315                 int tmp_reg = BPF_REG_9;                                       \
9316                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9317                         --tmp_reg;                                             \
9318                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9319                         --tmp_reg;                                             \
9320                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9321                                       offsetof(S, TF));                        \
9322                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9323                                       si->dst_reg, offsetof(S, F));            \
9324                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9325                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9326                                        target_size)                            \
9327                                 + OFF);                                        \
9328                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9329                                       offsetof(S, TF));                        \
9330         } while (0)
9331
9332 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9333                                                       TF)                      \
9334         do {                                                                   \
9335                 if (type == BPF_WRITE) {                                       \
9336                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9337                                                          OFF, TF);             \
9338                 } else {                                                       \
9339                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9340                                 S, NS, F, NF, SIZE, OFF);  \
9341                 }                                                              \
9342         } while (0)
9343
9344 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9345         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9346                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9347
9348 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9349                                         const struct bpf_insn *si,
9350                                         struct bpf_insn *insn_buf,
9351                                         struct bpf_prog *prog, u32 *target_size)
9352 {
9353         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9354         struct bpf_insn *insn = insn_buf;
9355
9356         switch (si->off) {
9357         case offsetof(struct bpf_sock_addr, user_family):
9358                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9359                                             struct sockaddr, uaddr, sa_family);
9360                 break;
9361
9362         case offsetof(struct bpf_sock_addr, user_ip4):
9363                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9364                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9365                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9366                 break;
9367
9368         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9369                 off = si->off;
9370                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9371                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9372                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9373                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9374                         tmp_reg);
9375                 break;
9376
9377         case offsetof(struct bpf_sock_addr, user_port):
9378                 /* To get port we need to know sa_family first and then treat
9379                  * sockaddr as either sockaddr_in or sockaddr_in6.
9380                  * Though we can simplify since port field has same offset and
9381                  * size in both structures.
9382                  * Here we check this invariant and use just one of the
9383                  * structures if it's true.
9384                  */
9385                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9386                              offsetof(struct sockaddr_in6, sin6_port));
9387                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9388                              sizeof_field(struct sockaddr_in6, sin6_port));
9389                 /* Account for sin6_port being smaller than user_port. */
9390                 port_size = min(port_size, BPF_LDST_BYTES(si));
9391                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9392                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9393                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9394                 break;
9395
9396         case offsetof(struct bpf_sock_addr, family):
9397                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9398                                             struct sock, sk, sk_family);
9399                 break;
9400
9401         case offsetof(struct bpf_sock_addr, type):
9402                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9403                                             struct sock, sk, sk_type);
9404                 break;
9405
9406         case offsetof(struct bpf_sock_addr, protocol):
9407                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9408                                             struct sock, sk, sk_protocol);
9409                 break;
9410
9411         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9412                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9413                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9414                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9415                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9416                 break;
9417
9418         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9419                                 msg_src_ip6[3]):
9420                 off = si->off;
9421                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9422                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9423                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9424                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9425                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9426                 break;
9427         case offsetof(struct bpf_sock_addr, sk):
9428                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9429                                       si->dst_reg, si->src_reg,
9430                                       offsetof(struct bpf_sock_addr_kern, sk));
9431                 break;
9432         }
9433
9434         return insn - insn_buf;
9435 }
9436
9437 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9438                                        const struct bpf_insn *si,
9439                                        struct bpf_insn *insn_buf,
9440                                        struct bpf_prog *prog,
9441                                        u32 *target_size)
9442 {
9443         struct bpf_insn *insn = insn_buf;
9444         int off;
9445
9446 /* Helper macro for adding read access to tcp_sock or sock fields. */
9447 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9448         do {                                                                  \
9449                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
9450                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9451                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9452                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9453                         reg--;                                                \
9454                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9455                         reg--;                                                \
9456                 if (si->dst_reg == si->src_reg) {                             \
9457                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9458                                           offsetof(struct bpf_sock_ops_kern,  \
9459                                           temp));                             \
9460                         fullsock_reg = reg;                                   \
9461                         jmp += 2;                                             \
9462                 }                                                             \
9463                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9464                                                 struct bpf_sock_ops_kern,     \
9465                                                 is_fullsock),                 \
9466                                       fullsock_reg, si->src_reg,              \
9467                                       offsetof(struct bpf_sock_ops_kern,      \
9468                                                is_fullsock));                 \
9469                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9470                 if (si->dst_reg == si->src_reg)                               \
9471                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9472                                       offsetof(struct bpf_sock_ops_kern,      \
9473                                       temp));                                 \
9474                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9475                                                 struct bpf_sock_ops_kern, sk),\
9476                                       si->dst_reg, si->src_reg,               \
9477                                       offsetof(struct bpf_sock_ops_kern, sk));\
9478                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9479                                                        OBJ_FIELD),            \
9480                                       si->dst_reg, si->dst_reg,               \
9481                                       offsetof(OBJ, OBJ_FIELD));              \
9482                 if (si->dst_reg == si->src_reg) {                             \
9483                         *insn++ = BPF_JMP_A(1);                               \
9484                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9485                                       offsetof(struct bpf_sock_ops_kern,      \
9486                                       temp));                                 \
9487                 }                                                             \
9488         } while (0)
9489
9490 #define SOCK_OPS_GET_SK()                                                             \
9491         do {                                                                  \
9492                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9493                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9494                         reg--;                                                \
9495                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9496                         reg--;                                                \
9497                 if (si->dst_reg == si->src_reg) {                             \
9498                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9499                                           offsetof(struct bpf_sock_ops_kern,  \
9500                                           temp));                             \
9501                         fullsock_reg = reg;                                   \
9502                         jmp += 2;                                             \
9503                 }                                                             \
9504                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9505                                                 struct bpf_sock_ops_kern,     \
9506                                                 is_fullsock),                 \
9507                                       fullsock_reg, si->src_reg,              \
9508                                       offsetof(struct bpf_sock_ops_kern,      \
9509                                                is_fullsock));                 \
9510                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9511                 if (si->dst_reg == si->src_reg)                               \
9512                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9513                                       offsetof(struct bpf_sock_ops_kern,      \
9514                                       temp));                                 \
9515                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9516                                                 struct bpf_sock_ops_kern, sk),\
9517                                       si->dst_reg, si->src_reg,               \
9518                                       offsetof(struct bpf_sock_ops_kern, sk));\
9519                 if (si->dst_reg == si->src_reg) {                             \
9520                         *insn++ = BPF_JMP_A(1);                               \
9521                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9522                                       offsetof(struct bpf_sock_ops_kern,      \
9523                                       temp));                                 \
9524                 }                                                             \
9525         } while (0)
9526
9527 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9528                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9529
9530 /* Helper macro for adding write access to tcp_sock or sock fields.
9531  * The macro is called with two registers, dst_reg which contains a pointer
9532  * to ctx (context) and src_reg which contains the value that should be
9533  * stored. However, we need an additional register since we cannot overwrite
9534  * dst_reg because it may be used later in the program.
9535  * Instead we "borrow" one of the other register. We first save its value
9536  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9537  * it at the end of the macro.
9538  */
9539 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9540         do {                                                                  \
9541                 int reg = BPF_REG_9;                                          \
9542                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9543                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9544                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9545                         reg--;                                                \
9546                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9547                         reg--;                                                \
9548                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9549                                       offsetof(struct bpf_sock_ops_kern,      \
9550                                                temp));                        \
9551                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9552                                                 struct bpf_sock_ops_kern,     \
9553                                                 is_fullsock),                 \
9554                                       reg, si->dst_reg,                       \
9555                                       offsetof(struct bpf_sock_ops_kern,      \
9556                                                is_fullsock));                 \
9557                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
9558                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9559                                                 struct bpf_sock_ops_kern, sk),\
9560                                       reg, si->dst_reg,                       \
9561                                       offsetof(struct bpf_sock_ops_kern, sk));\
9562                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
9563                                       reg, si->src_reg,                       \
9564                                       offsetof(OBJ, OBJ_FIELD));              \
9565                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
9566                                       offsetof(struct bpf_sock_ops_kern,      \
9567                                                temp));                        \
9568         } while (0)
9569
9570 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
9571         do {                                                                  \
9572                 if (TYPE == BPF_WRITE)                                        \
9573                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9574                 else                                                          \
9575                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9576         } while (0)
9577
9578         if (insn > insn_buf)
9579                 return insn - insn_buf;
9580
9581         switch (si->off) {
9582         case offsetof(struct bpf_sock_ops, op):
9583                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9584                                                        op),
9585                                       si->dst_reg, si->src_reg,
9586                                       offsetof(struct bpf_sock_ops_kern, op));
9587                 break;
9588
9589         case offsetof(struct bpf_sock_ops, replylong[0]) ...
9590              offsetof(struct bpf_sock_ops, replylong[3]):
9591                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9592                              sizeof_field(struct bpf_sock_ops_kern, reply));
9593                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9594                              sizeof_field(struct bpf_sock_ops_kern, replylong));
9595                 off = si->off;
9596                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9597                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9598                 if (type == BPF_WRITE)
9599                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9600                                               off);
9601                 else
9602                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9603                                               off);
9604                 break;
9605
9606         case offsetof(struct bpf_sock_ops, family):
9607                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9608
9609                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9610                                               struct bpf_sock_ops_kern, sk),
9611                                       si->dst_reg, si->src_reg,
9612                                       offsetof(struct bpf_sock_ops_kern, sk));
9613                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9614                                       offsetof(struct sock_common, skc_family));
9615                 break;
9616
9617         case offsetof(struct bpf_sock_ops, remote_ip4):
9618                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9619
9620                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9621                                                 struct bpf_sock_ops_kern, sk),
9622                                       si->dst_reg, si->src_reg,
9623                                       offsetof(struct bpf_sock_ops_kern, sk));
9624                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9625                                       offsetof(struct sock_common, skc_daddr));
9626                 break;
9627
9628         case offsetof(struct bpf_sock_ops, local_ip4):
9629                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9630                                           skc_rcv_saddr) != 4);
9631
9632                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9633                                               struct bpf_sock_ops_kern, sk),
9634                                       si->dst_reg, si->src_reg,
9635                                       offsetof(struct bpf_sock_ops_kern, sk));
9636                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9637                                       offsetof(struct sock_common,
9638                                                skc_rcv_saddr));
9639                 break;
9640
9641         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9642              offsetof(struct bpf_sock_ops, remote_ip6[3]):
9643 #if IS_ENABLED(CONFIG_IPV6)
9644                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9645                                           skc_v6_daddr.s6_addr32[0]) != 4);
9646
9647                 off = si->off;
9648                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9649                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9650                                                 struct bpf_sock_ops_kern, sk),
9651                                       si->dst_reg, si->src_reg,
9652                                       offsetof(struct bpf_sock_ops_kern, sk));
9653                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9654                                       offsetof(struct sock_common,
9655                                                skc_v6_daddr.s6_addr32[0]) +
9656                                       off);
9657 #else
9658                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9659 #endif
9660                 break;
9661
9662         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9663              offsetof(struct bpf_sock_ops, local_ip6[3]):
9664 #if IS_ENABLED(CONFIG_IPV6)
9665                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9666                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9667
9668                 off = si->off;
9669                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9670                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9671                                                 struct bpf_sock_ops_kern, sk),
9672                                       si->dst_reg, si->src_reg,
9673                                       offsetof(struct bpf_sock_ops_kern, sk));
9674                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9675                                       offsetof(struct sock_common,
9676                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9677                                       off);
9678 #else
9679                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9680 #endif
9681                 break;
9682
9683         case offsetof(struct bpf_sock_ops, remote_port):
9684                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9685
9686                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9687                                                 struct bpf_sock_ops_kern, sk),
9688                                       si->dst_reg, si->src_reg,
9689                                       offsetof(struct bpf_sock_ops_kern, sk));
9690                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9691                                       offsetof(struct sock_common, skc_dport));
9692 #ifndef __BIG_ENDIAN_BITFIELD
9693                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9694 #endif
9695                 break;
9696
9697         case offsetof(struct bpf_sock_ops, local_port):
9698                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9699
9700                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9701                                                 struct bpf_sock_ops_kern, sk),
9702                                       si->dst_reg, si->src_reg,
9703                                       offsetof(struct bpf_sock_ops_kern, sk));
9704                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9705                                       offsetof(struct sock_common, skc_num));
9706                 break;
9707
9708         case offsetof(struct bpf_sock_ops, is_fullsock):
9709                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9710                                                 struct bpf_sock_ops_kern,
9711                                                 is_fullsock),
9712                                       si->dst_reg, si->src_reg,
9713                                       offsetof(struct bpf_sock_ops_kern,
9714                                                is_fullsock));
9715                 break;
9716
9717         case offsetof(struct bpf_sock_ops, state):
9718                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9719
9720                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9721                                                 struct bpf_sock_ops_kern, sk),
9722                                       si->dst_reg, si->src_reg,
9723                                       offsetof(struct bpf_sock_ops_kern, sk));
9724                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9725                                       offsetof(struct sock_common, skc_state));
9726                 break;
9727
9728         case offsetof(struct bpf_sock_ops, rtt_min):
9729                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9730                              sizeof(struct minmax));
9731                 BUILD_BUG_ON(sizeof(struct minmax) <
9732                              sizeof(struct minmax_sample));
9733
9734                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9735                                                 struct bpf_sock_ops_kern, sk),
9736                                       si->dst_reg, si->src_reg,
9737                                       offsetof(struct bpf_sock_ops_kern, sk));
9738                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9739                                       offsetof(struct tcp_sock, rtt_min) +
9740                                       sizeof_field(struct minmax_sample, t));
9741                 break;
9742
9743         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9744                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9745                                    struct tcp_sock);
9746                 break;
9747
9748         case offsetof(struct bpf_sock_ops, sk_txhash):
9749                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9750                                           struct sock, type);
9751                 break;
9752         case offsetof(struct bpf_sock_ops, snd_cwnd):
9753                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9754                 break;
9755         case offsetof(struct bpf_sock_ops, srtt_us):
9756                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9757                 break;
9758         case offsetof(struct bpf_sock_ops, snd_ssthresh):
9759                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9760                 break;
9761         case offsetof(struct bpf_sock_ops, rcv_nxt):
9762                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9763                 break;
9764         case offsetof(struct bpf_sock_ops, snd_nxt):
9765                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9766                 break;
9767         case offsetof(struct bpf_sock_ops, snd_una):
9768                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9769                 break;
9770         case offsetof(struct bpf_sock_ops, mss_cache):
9771                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9772                 break;
9773         case offsetof(struct bpf_sock_ops, ecn_flags):
9774                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9775                 break;
9776         case offsetof(struct bpf_sock_ops, rate_delivered):
9777                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9778                 break;
9779         case offsetof(struct bpf_sock_ops, rate_interval_us):
9780                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9781                 break;
9782         case offsetof(struct bpf_sock_ops, packets_out):
9783                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9784                 break;
9785         case offsetof(struct bpf_sock_ops, retrans_out):
9786                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9787                 break;
9788         case offsetof(struct bpf_sock_ops, total_retrans):
9789                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9790                 break;
9791         case offsetof(struct bpf_sock_ops, segs_in):
9792                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9793                 break;
9794         case offsetof(struct bpf_sock_ops, data_segs_in):
9795                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9796                 break;
9797         case offsetof(struct bpf_sock_ops, segs_out):
9798                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9799                 break;
9800         case offsetof(struct bpf_sock_ops, data_segs_out):
9801                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9802                 break;
9803         case offsetof(struct bpf_sock_ops, lost_out):
9804                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9805                 break;
9806         case offsetof(struct bpf_sock_ops, sacked_out):
9807                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9808                 break;
9809         case offsetof(struct bpf_sock_ops, bytes_received):
9810                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9811                 break;
9812         case offsetof(struct bpf_sock_ops, bytes_acked):
9813                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9814                 break;
9815         case offsetof(struct bpf_sock_ops, sk):
9816                 SOCK_OPS_GET_SK();
9817                 break;
9818         case offsetof(struct bpf_sock_ops, skb_data_end):
9819                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9820                                                        skb_data_end),
9821                                       si->dst_reg, si->src_reg,
9822                                       offsetof(struct bpf_sock_ops_kern,
9823                                                skb_data_end));
9824                 break;
9825         case offsetof(struct bpf_sock_ops, skb_data):
9826                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9827                                                        skb),
9828                                       si->dst_reg, si->src_reg,
9829                                       offsetof(struct bpf_sock_ops_kern,
9830                                                skb));
9831                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9832                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9833                                       si->dst_reg, si->dst_reg,
9834                                       offsetof(struct sk_buff, data));
9835                 break;
9836         case offsetof(struct bpf_sock_ops, skb_len):
9837                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9838                                                        skb),
9839                                       si->dst_reg, si->src_reg,
9840                                       offsetof(struct bpf_sock_ops_kern,
9841                                                skb));
9842                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9843                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9844                                       si->dst_reg, si->dst_reg,
9845                                       offsetof(struct sk_buff, len));
9846                 break;
9847         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9848                 off = offsetof(struct sk_buff, cb);
9849                 off += offsetof(struct tcp_skb_cb, tcp_flags);
9850                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9851                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9852                                                        skb),
9853                                       si->dst_reg, si->src_reg,
9854                                       offsetof(struct bpf_sock_ops_kern,
9855                                                skb));
9856                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9857                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9858                                                        tcp_flags),
9859                                       si->dst_reg, si->dst_reg, off);
9860                 break;
9861         }
9862         return insn - insn_buf;
9863 }
9864
9865 /* data_end = skb->data + skb_headlen() */
9866 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9867                                                     struct bpf_insn *insn)
9868 {
9869         int reg;
9870         int temp_reg_off = offsetof(struct sk_buff, cb) +
9871                            offsetof(struct sk_skb_cb, temp_reg);
9872
9873         if (si->src_reg == si->dst_reg) {
9874                 /* We need an extra register, choose and save a register. */
9875                 reg = BPF_REG_9;
9876                 if (si->src_reg == reg || si->dst_reg == reg)
9877                         reg--;
9878                 if (si->src_reg == reg || si->dst_reg == reg)
9879                         reg--;
9880                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
9881         } else {
9882                 reg = si->dst_reg;
9883         }
9884
9885         /* reg = skb->data */
9886         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9887                               reg, si->src_reg,
9888                               offsetof(struct sk_buff, data));
9889         /* AX = skb->len */
9890         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9891                               BPF_REG_AX, si->src_reg,
9892                               offsetof(struct sk_buff, len));
9893         /* reg = skb->data + skb->len */
9894         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
9895         /* AX = skb->data_len */
9896         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9897                               BPF_REG_AX, si->src_reg,
9898                               offsetof(struct sk_buff, data_len));
9899
9900         /* reg = skb->data + skb->len - skb->data_len */
9901         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
9902
9903         if (si->src_reg == si->dst_reg) {
9904                 /* Restore the saved register */
9905                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
9906                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
9907                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
9908         }
9909
9910         return insn;
9911 }
9912
9913 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9914                                      const struct bpf_insn *si,
9915                                      struct bpf_insn *insn_buf,
9916                                      struct bpf_prog *prog, u32 *target_size)
9917 {
9918         struct bpf_insn *insn = insn_buf;
9919         int off;
9920
9921         switch (si->off) {
9922         case offsetof(struct __sk_buff, data_end):
9923                 insn = bpf_convert_data_end_access(si, insn);
9924                 break;
9925         case offsetof(struct __sk_buff, cb[0]) ...
9926              offsetofend(struct __sk_buff, cb[4]) - 1:
9927                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9928                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9929                               offsetof(struct sk_skb_cb, data)) %
9930                              sizeof(__u64));
9931
9932                 prog->cb_access = 1;
9933                 off  = si->off;
9934                 off -= offsetof(struct __sk_buff, cb[0]);
9935                 off += offsetof(struct sk_buff, cb);
9936                 off += offsetof(struct sk_skb_cb, data);
9937                 if (type == BPF_WRITE)
9938                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9939                                               si->src_reg, off);
9940                 else
9941                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9942                                               si->src_reg, off);
9943                 break;
9944
9945
9946         default:
9947                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9948                                               target_size);
9949         }
9950
9951         return insn - insn_buf;
9952 }
9953
9954 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9955                                      const struct bpf_insn *si,
9956                                      struct bpf_insn *insn_buf,
9957                                      struct bpf_prog *prog, u32 *target_size)
9958 {
9959         struct bpf_insn *insn = insn_buf;
9960 #if IS_ENABLED(CONFIG_IPV6)
9961         int off;
9962 #endif
9963
9964         /* convert ctx uses the fact sg element is first in struct */
9965         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9966
9967         switch (si->off) {
9968         case offsetof(struct sk_msg_md, data):
9969                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9970                                       si->dst_reg, si->src_reg,
9971                                       offsetof(struct sk_msg, data));
9972                 break;
9973         case offsetof(struct sk_msg_md, data_end):
9974                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9975                                       si->dst_reg, si->src_reg,
9976                                       offsetof(struct sk_msg, data_end));
9977                 break;
9978         case offsetof(struct sk_msg_md, family):
9979                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9980
9981                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9982                                               struct sk_msg, sk),
9983                                       si->dst_reg, si->src_reg,
9984                                       offsetof(struct sk_msg, sk));
9985                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9986                                       offsetof(struct sock_common, skc_family));
9987                 break;
9988
9989         case offsetof(struct sk_msg_md, remote_ip4):
9990                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9991
9992                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9993                                                 struct sk_msg, sk),
9994                                       si->dst_reg, si->src_reg,
9995                                       offsetof(struct sk_msg, sk));
9996                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9997                                       offsetof(struct sock_common, skc_daddr));
9998                 break;
9999
10000         case offsetof(struct sk_msg_md, local_ip4):
10001                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10002                                           skc_rcv_saddr) != 4);
10003
10004                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10005                                               struct sk_msg, sk),
10006                                       si->dst_reg, si->src_reg,
10007                                       offsetof(struct sk_msg, sk));
10008                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10009                                       offsetof(struct sock_common,
10010                                                skc_rcv_saddr));
10011                 break;
10012
10013         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10014              offsetof(struct sk_msg_md, remote_ip6[3]):
10015 #if IS_ENABLED(CONFIG_IPV6)
10016                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10017                                           skc_v6_daddr.s6_addr32[0]) != 4);
10018
10019                 off = si->off;
10020                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10021                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10022                                                 struct sk_msg, sk),
10023                                       si->dst_reg, si->src_reg,
10024                                       offsetof(struct sk_msg, sk));
10025                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10026                                       offsetof(struct sock_common,
10027                                                skc_v6_daddr.s6_addr32[0]) +
10028                                       off);
10029 #else
10030                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10031 #endif
10032                 break;
10033
10034         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10035              offsetof(struct sk_msg_md, local_ip6[3]):
10036 #if IS_ENABLED(CONFIG_IPV6)
10037                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10038                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10039
10040                 off = si->off;
10041                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10042                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10043                                                 struct sk_msg, sk),
10044                                       si->dst_reg, si->src_reg,
10045                                       offsetof(struct sk_msg, sk));
10046                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10047                                       offsetof(struct sock_common,
10048                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10049                                       off);
10050 #else
10051                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10052 #endif
10053                 break;
10054
10055         case offsetof(struct sk_msg_md, remote_port):
10056                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10057
10058                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10059                                                 struct sk_msg, sk),
10060                                       si->dst_reg, si->src_reg,
10061                                       offsetof(struct sk_msg, sk));
10062                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10063                                       offsetof(struct sock_common, skc_dport));
10064 #ifndef __BIG_ENDIAN_BITFIELD
10065                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10066 #endif
10067                 break;
10068
10069         case offsetof(struct sk_msg_md, local_port):
10070                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10071
10072                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10073                                                 struct sk_msg, sk),
10074                                       si->dst_reg, si->src_reg,
10075                                       offsetof(struct sk_msg, sk));
10076                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10077                                       offsetof(struct sock_common, skc_num));
10078                 break;
10079
10080         case offsetof(struct sk_msg_md, size):
10081                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10082                                       si->dst_reg, si->src_reg,
10083                                       offsetof(struct sk_msg_sg, size));
10084                 break;
10085
10086         case offsetof(struct sk_msg_md, sk):
10087                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10088                                       si->dst_reg, si->src_reg,
10089                                       offsetof(struct sk_msg, sk));
10090                 break;
10091         }
10092
10093         return insn - insn_buf;
10094 }
10095
10096 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10097         .get_func_proto         = sk_filter_func_proto,
10098         .is_valid_access        = sk_filter_is_valid_access,
10099         .convert_ctx_access     = bpf_convert_ctx_access,
10100         .gen_ld_abs             = bpf_gen_ld_abs,
10101 };
10102
10103 const struct bpf_prog_ops sk_filter_prog_ops = {
10104         .test_run               = bpf_prog_test_run_skb,
10105 };
10106
10107 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10108         .get_func_proto         = tc_cls_act_func_proto,
10109         .is_valid_access        = tc_cls_act_is_valid_access,
10110         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10111         .gen_prologue           = tc_cls_act_prologue,
10112         .gen_ld_abs             = bpf_gen_ld_abs,
10113         .check_kfunc_call       = bpf_prog_test_check_kfunc_call,
10114 };
10115
10116 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10117         .test_run               = bpf_prog_test_run_skb,
10118 };
10119
10120 const struct bpf_verifier_ops xdp_verifier_ops = {
10121         .get_func_proto         = xdp_func_proto,
10122         .is_valid_access        = xdp_is_valid_access,
10123         .convert_ctx_access     = xdp_convert_ctx_access,
10124         .gen_prologue           = bpf_noop_prologue,
10125 };
10126
10127 const struct bpf_prog_ops xdp_prog_ops = {
10128         .test_run               = bpf_prog_test_run_xdp,
10129 };
10130
10131 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10132         .get_func_proto         = cg_skb_func_proto,
10133         .is_valid_access        = cg_skb_is_valid_access,
10134         .convert_ctx_access     = bpf_convert_ctx_access,
10135 };
10136
10137 const struct bpf_prog_ops cg_skb_prog_ops = {
10138         .test_run               = bpf_prog_test_run_skb,
10139 };
10140
10141 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10142         .get_func_proto         = lwt_in_func_proto,
10143         .is_valid_access        = lwt_is_valid_access,
10144         .convert_ctx_access     = bpf_convert_ctx_access,
10145 };
10146
10147 const struct bpf_prog_ops lwt_in_prog_ops = {
10148         .test_run               = bpf_prog_test_run_skb,
10149 };
10150
10151 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10152         .get_func_proto         = lwt_out_func_proto,
10153         .is_valid_access        = lwt_is_valid_access,
10154         .convert_ctx_access     = bpf_convert_ctx_access,
10155 };
10156
10157 const struct bpf_prog_ops lwt_out_prog_ops = {
10158         .test_run               = bpf_prog_test_run_skb,
10159 };
10160
10161 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10162         .get_func_proto         = lwt_xmit_func_proto,
10163         .is_valid_access        = lwt_is_valid_access,
10164         .convert_ctx_access     = bpf_convert_ctx_access,
10165         .gen_prologue           = tc_cls_act_prologue,
10166 };
10167
10168 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10169         .test_run               = bpf_prog_test_run_skb,
10170 };
10171
10172 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10173         .get_func_proto         = lwt_seg6local_func_proto,
10174         .is_valid_access        = lwt_is_valid_access,
10175         .convert_ctx_access     = bpf_convert_ctx_access,
10176 };
10177
10178 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10179         .test_run               = bpf_prog_test_run_skb,
10180 };
10181
10182 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10183         .get_func_proto         = sock_filter_func_proto,
10184         .is_valid_access        = sock_filter_is_valid_access,
10185         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10186 };
10187
10188 const struct bpf_prog_ops cg_sock_prog_ops = {
10189 };
10190
10191 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10192         .get_func_proto         = sock_addr_func_proto,
10193         .is_valid_access        = sock_addr_is_valid_access,
10194         .convert_ctx_access     = sock_addr_convert_ctx_access,
10195 };
10196
10197 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10198 };
10199
10200 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10201         .get_func_proto         = sock_ops_func_proto,
10202         .is_valid_access        = sock_ops_is_valid_access,
10203         .convert_ctx_access     = sock_ops_convert_ctx_access,
10204 };
10205
10206 const struct bpf_prog_ops sock_ops_prog_ops = {
10207 };
10208
10209 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10210         .get_func_proto         = sk_skb_func_proto,
10211         .is_valid_access        = sk_skb_is_valid_access,
10212         .convert_ctx_access     = sk_skb_convert_ctx_access,
10213         .gen_prologue           = sk_skb_prologue,
10214 };
10215
10216 const struct bpf_prog_ops sk_skb_prog_ops = {
10217 };
10218
10219 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10220         .get_func_proto         = sk_msg_func_proto,
10221         .is_valid_access        = sk_msg_is_valid_access,
10222         .convert_ctx_access     = sk_msg_convert_ctx_access,
10223         .gen_prologue           = bpf_noop_prologue,
10224 };
10225
10226 const struct bpf_prog_ops sk_msg_prog_ops = {
10227 };
10228
10229 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10230         .get_func_proto         = flow_dissector_func_proto,
10231         .is_valid_access        = flow_dissector_is_valid_access,
10232         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10233 };
10234
10235 const struct bpf_prog_ops flow_dissector_prog_ops = {
10236         .test_run               = bpf_prog_test_run_flow_dissector,
10237 };
10238
10239 int sk_detach_filter(struct sock *sk)
10240 {
10241         int ret = -ENOENT;
10242         struct sk_filter *filter;
10243
10244         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10245                 return -EPERM;
10246
10247         filter = rcu_dereference_protected(sk->sk_filter,
10248                                            lockdep_sock_is_held(sk));
10249         if (filter) {
10250                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10251                 sk_filter_uncharge(sk, filter);
10252                 ret = 0;
10253         }
10254
10255         return ret;
10256 }
10257 EXPORT_SYMBOL_GPL(sk_detach_filter);
10258
10259 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10260                   unsigned int len)
10261 {
10262         struct sock_fprog_kern *fprog;
10263         struct sk_filter *filter;
10264         int ret = 0;
10265
10266         lock_sock(sk);
10267         filter = rcu_dereference_protected(sk->sk_filter,
10268                                            lockdep_sock_is_held(sk));
10269         if (!filter)
10270                 goto out;
10271
10272         /* We're copying the filter that has been originally attached,
10273          * so no conversion/decode needed anymore. eBPF programs that
10274          * have no original program cannot be dumped through this.
10275          */
10276         ret = -EACCES;
10277         fprog = filter->prog->orig_prog;
10278         if (!fprog)
10279                 goto out;
10280
10281         ret = fprog->len;
10282         if (!len)
10283                 /* User space only enquires number of filter blocks. */
10284                 goto out;
10285
10286         ret = -EINVAL;
10287         if (len < fprog->len)
10288                 goto out;
10289
10290         ret = -EFAULT;
10291         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10292                 goto out;
10293
10294         /* Instead of bytes, the API requests to return the number
10295          * of filter blocks.
10296          */
10297         ret = fprog->len;
10298 out:
10299         release_sock(sk);
10300         return ret;
10301 }
10302
10303 #ifdef CONFIG_INET
10304 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10305                                     struct sock_reuseport *reuse,
10306                                     struct sock *sk, struct sk_buff *skb,
10307                                     struct sock *migrating_sk,
10308                                     u32 hash)
10309 {
10310         reuse_kern->skb = skb;
10311         reuse_kern->sk = sk;
10312         reuse_kern->selected_sk = NULL;
10313         reuse_kern->migrating_sk = migrating_sk;
10314         reuse_kern->data_end = skb->data + skb_headlen(skb);
10315         reuse_kern->hash = hash;
10316         reuse_kern->reuseport_id = reuse->reuseport_id;
10317         reuse_kern->bind_inany = reuse->bind_inany;
10318 }
10319
10320 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10321                                   struct bpf_prog *prog, struct sk_buff *skb,
10322                                   struct sock *migrating_sk,
10323                                   u32 hash)
10324 {
10325         struct sk_reuseport_kern reuse_kern;
10326         enum sk_action action;
10327
10328         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10329         action = bpf_prog_run(prog, &reuse_kern);
10330
10331         if (action == SK_PASS)
10332                 return reuse_kern.selected_sk;
10333         else
10334                 return ERR_PTR(-ECONNREFUSED);
10335 }
10336
10337 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10338            struct bpf_map *, map, void *, key, u32, flags)
10339 {
10340         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10341         struct sock_reuseport *reuse;
10342         struct sock *selected_sk;
10343
10344         selected_sk = map->ops->map_lookup_elem(map, key);
10345         if (!selected_sk)
10346                 return -ENOENT;
10347
10348         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10349         if (!reuse) {
10350                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10351                 if (sk_is_refcounted(selected_sk))
10352                         sock_put(selected_sk);
10353
10354                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10355                  * The only (!reuse) case here is - the sk has already been
10356                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10357                  *
10358                  * Other maps (e.g. sock_map) do not provide this guarantee and
10359                  * the sk may never be in the reuseport group to begin with.
10360                  */
10361                 return is_sockarray ? -ENOENT : -EINVAL;
10362         }
10363
10364         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10365                 struct sock *sk = reuse_kern->sk;
10366
10367                 if (sk->sk_protocol != selected_sk->sk_protocol)
10368                         return -EPROTOTYPE;
10369                 else if (sk->sk_family != selected_sk->sk_family)
10370                         return -EAFNOSUPPORT;
10371
10372                 /* Catch all. Likely bound to a different sockaddr. */
10373                 return -EBADFD;
10374         }
10375
10376         reuse_kern->selected_sk = selected_sk;
10377
10378         return 0;
10379 }
10380
10381 static const struct bpf_func_proto sk_select_reuseport_proto = {
10382         .func           = sk_select_reuseport,
10383         .gpl_only       = false,
10384         .ret_type       = RET_INTEGER,
10385         .arg1_type      = ARG_PTR_TO_CTX,
10386         .arg2_type      = ARG_CONST_MAP_PTR,
10387         .arg3_type      = ARG_PTR_TO_MAP_KEY,
10388         .arg4_type      = ARG_ANYTHING,
10389 };
10390
10391 BPF_CALL_4(sk_reuseport_load_bytes,
10392            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10393            void *, to, u32, len)
10394 {
10395         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10396 }
10397
10398 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10399         .func           = sk_reuseport_load_bytes,
10400         .gpl_only       = false,
10401         .ret_type       = RET_INTEGER,
10402         .arg1_type      = ARG_PTR_TO_CTX,
10403         .arg2_type      = ARG_ANYTHING,
10404         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10405         .arg4_type      = ARG_CONST_SIZE,
10406 };
10407
10408 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10409            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10410            void *, to, u32, len, u32, start_header)
10411 {
10412         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10413                                                len, start_header);
10414 }
10415
10416 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10417         .func           = sk_reuseport_load_bytes_relative,
10418         .gpl_only       = false,
10419         .ret_type       = RET_INTEGER,
10420         .arg1_type      = ARG_PTR_TO_CTX,
10421         .arg2_type      = ARG_ANYTHING,
10422         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10423         .arg4_type      = ARG_CONST_SIZE,
10424         .arg5_type      = ARG_ANYTHING,
10425 };
10426
10427 static const struct bpf_func_proto *
10428 sk_reuseport_func_proto(enum bpf_func_id func_id,
10429                         const struct bpf_prog *prog)
10430 {
10431         switch (func_id) {
10432         case BPF_FUNC_sk_select_reuseport:
10433                 return &sk_select_reuseport_proto;
10434         case BPF_FUNC_skb_load_bytes:
10435                 return &sk_reuseport_load_bytes_proto;
10436         case BPF_FUNC_skb_load_bytes_relative:
10437                 return &sk_reuseport_load_bytes_relative_proto;
10438         case BPF_FUNC_get_socket_cookie:
10439                 return &bpf_get_socket_ptr_cookie_proto;
10440         case BPF_FUNC_ktime_get_coarse_ns:
10441                 return &bpf_ktime_get_coarse_ns_proto;
10442         default:
10443                 return bpf_base_func_proto(func_id);
10444         }
10445 }
10446
10447 static bool
10448 sk_reuseport_is_valid_access(int off, int size,
10449                              enum bpf_access_type type,
10450                              const struct bpf_prog *prog,
10451                              struct bpf_insn_access_aux *info)
10452 {
10453         const u32 size_default = sizeof(__u32);
10454
10455         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10456             off % size || type != BPF_READ)
10457                 return false;
10458
10459         switch (off) {
10460         case offsetof(struct sk_reuseport_md, data):
10461                 info->reg_type = PTR_TO_PACKET;
10462                 return size == sizeof(__u64);
10463
10464         case offsetof(struct sk_reuseport_md, data_end):
10465                 info->reg_type = PTR_TO_PACKET_END;
10466                 return size == sizeof(__u64);
10467
10468         case offsetof(struct sk_reuseport_md, hash):
10469                 return size == size_default;
10470
10471         case offsetof(struct sk_reuseport_md, sk):
10472                 info->reg_type = PTR_TO_SOCKET;
10473                 return size == sizeof(__u64);
10474
10475         case offsetof(struct sk_reuseport_md, migrating_sk):
10476                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10477                 return size == sizeof(__u64);
10478
10479         /* Fields that allow narrowing */
10480         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10481                 if (size < sizeof_field(struct sk_buff, protocol))
10482                         return false;
10483                 fallthrough;
10484         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10485         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10486         case bpf_ctx_range(struct sk_reuseport_md, len):
10487                 bpf_ctx_record_field_size(info, size_default);
10488                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10489
10490         default:
10491                 return false;
10492         }
10493 }
10494
10495 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10496         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10497                               si->dst_reg, si->src_reg,                 \
10498                               bpf_target_off(struct sk_reuseport_kern, F, \
10499                                              sizeof_field(struct sk_reuseport_kern, F), \
10500                                              target_size));             \
10501         })
10502
10503 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10504         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10505                                     struct sk_buff,                     \
10506                                     skb,                                \
10507                                     SKB_FIELD)
10508
10509 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10510         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10511                                     struct sock,                        \
10512                                     sk,                                 \
10513                                     SK_FIELD)
10514
10515 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10516                                            const struct bpf_insn *si,
10517                                            struct bpf_insn *insn_buf,
10518                                            struct bpf_prog *prog,
10519                                            u32 *target_size)
10520 {
10521         struct bpf_insn *insn = insn_buf;
10522
10523         switch (si->off) {
10524         case offsetof(struct sk_reuseport_md, data):
10525                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10526                 break;
10527
10528         case offsetof(struct sk_reuseport_md, len):
10529                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10530                 break;
10531
10532         case offsetof(struct sk_reuseport_md, eth_protocol):
10533                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10534                 break;
10535
10536         case offsetof(struct sk_reuseport_md, ip_protocol):
10537                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10538                 break;
10539
10540         case offsetof(struct sk_reuseport_md, data_end):
10541                 SK_REUSEPORT_LOAD_FIELD(data_end);
10542                 break;
10543
10544         case offsetof(struct sk_reuseport_md, hash):
10545                 SK_REUSEPORT_LOAD_FIELD(hash);
10546                 break;
10547
10548         case offsetof(struct sk_reuseport_md, bind_inany):
10549                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10550                 break;
10551
10552         case offsetof(struct sk_reuseport_md, sk):
10553                 SK_REUSEPORT_LOAD_FIELD(sk);
10554                 break;
10555
10556         case offsetof(struct sk_reuseport_md, migrating_sk):
10557                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10558                 break;
10559         }
10560
10561         return insn - insn_buf;
10562 }
10563
10564 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10565         .get_func_proto         = sk_reuseport_func_proto,
10566         .is_valid_access        = sk_reuseport_is_valid_access,
10567         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
10568 };
10569
10570 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10571 };
10572
10573 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10574 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10575
10576 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10577            struct sock *, sk, u64, flags)
10578 {
10579         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10580                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10581                 return -EINVAL;
10582         if (unlikely(sk && sk_is_refcounted(sk)))
10583                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10584         if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10585                 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10586
10587         /* Check if socket is suitable for packet L3/L4 protocol */
10588         if (sk && sk->sk_protocol != ctx->protocol)
10589                 return -EPROTOTYPE;
10590         if (sk && sk->sk_family != ctx->family &&
10591             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10592                 return -EAFNOSUPPORT;
10593
10594         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10595                 return -EEXIST;
10596
10597         /* Select socket as lookup result */
10598         ctx->selected_sk = sk;
10599         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10600         return 0;
10601 }
10602
10603 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10604         .func           = bpf_sk_lookup_assign,
10605         .gpl_only       = false,
10606         .ret_type       = RET_INTEGER,
10607         .arg1_type      = ARG_PTR_TO_CTX,
10608         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
10609         .arg3_type      = ARG_ANYTHING,
10610 };
10611
10612 static const struct bpf_func_proto *
10613 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10614 {
10615         switch (func_id) {
10616         case BPF_FUNC_perf_event_output:
10617                 return &bpf_event_output_data_proto;
10618         case BPF_FUNC_sk_assign:
10619                 return &bpf_sk_lookup_assign_proto;
10620         case BPF_FUNC_sk_release:
10621                 return &bpf_sk_release_proto;
10622         default:
10623                 return bpf_sk_base_func_proto(func_id);
10624         }
10625 }
10626
10627 static bool sk_lookup_is_valid_access(int off, int size,
10628                                       enum bpf_access_type type,
10629                                       const struct bpf_prog *prog,
10630                                       struct bpf_insn_access_aux *info)
10631 {
10632         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10633                 return false;
10634         if (off % size != 0)
10635                 return false;
10636         if (type != BPF_READ)
10637                 return false;
10638
10639         switch (off) {
10640         case offsetof(struct bpf_sk_lookup, sk):
10641                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10642                 return size == sizeof(__u64);
10643
10644         case bpf_ctx_range(struct bpf_sk_lookup, family):
10645         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10646         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10647         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10648         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10649         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10650         case offsetof(struct bpf_sk_lookup, remote_port) ...
10651              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
10652         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10653                 bpf_ctx_record_field_size(info, sizeof(__u32));
10654                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10655
10656         default:
10657                 return false;
10658         }
10659 }
10660
10661 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10662                                         const struct bpf_insn *si,
10663                                         struct bpf_insn *insn_buf,
10664                                         struct bpf_prog *prog,
10665                                         u32 *target_size)
10666 {
10667         struct bpf_insn *insn = insn_buf;
10668
10669         switch (si->off) {
10670         case offsetof(struct bpf_sk_lookup, sk):
10671                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10672                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
10673                 break;
10674
10675         case offsetof(struct bpf_sk_lookup, family):
10676                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10677                                       bpf_target_off(struct bpf_sk_lookup_kern,
10678                                                      family, 2, target_size));
10679                 break;
10680
10681         case offsetof(struct bpf_sk_lookup, protocol):
10682                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10683                                       bpf_target_off(struct bpf_sk_lookup_kern,
10684                                                      protocol, 2, target_size));
10685                 break;
10686
10687         case offsetof(struct bpf_sk_lookup, remote_ip4):
10688                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10689                                       bpf_target_off(struct bpf_sk_lookup_kern,
10690                                                      v4.saddr, 4, target_size));
10691                 break;
10692
10693         case offsetof(struct bpf_sk_lookup, local_ip4):
10694                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10695                                       bpf_target_off(struct bpf_sk_lookup_kern,
10696                                                      v4.daddr, 4, target_size));
10697                 break;
10698
10699         case bpf_ctx_range_till(struct bpf_sk_lookup,
10700                                 remote_ip6[0], remote_ip6[3]): {
10701 #if IS_ENABLED(CONFIG_IPV6)
10702                 int off = si->off;
10703
10704                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10705                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10706                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10707                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10708                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10709                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10710 #else
10711                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10712 #endif
10713                 break;
10714         }
10715         case bpf_ctx_range_till(struct bpf_sk_lookup,
10716                                 local_ip6[0], local_ip6[3]): {
10717 #if IS_ENABLED(CONFIG_IPV6)
10718                 int off = si->off;
10719
10720                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10721                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10722                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10723                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10724                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10725                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10726 #else
10727                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10728 #endif
10729                 break;
10730         }
10731         case offsetof(struct bpf_sk_lookup, remote_port):
10732                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10733                                       bpf_target_off(struct bpf_sk_lookup_kern,
10734                                                      sport, 2, target_size));
10735                 break;
10736
10737         case offsetof(struct bpf_sk_lookup, local_port):
10738                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10739                                       bpf_target_off(struct bpf_sk_lookup_kern,
10740                                                      dport, 2, target_size));
10741                 break;
10742         }
10743
10744         return insn - insn_buf;
10745 }
10746
10747 const struct bpf_prog_ops sk_lookup_prog_ops = {
10748         .test_run = bpf_prog_test_run_sk_lookup,
10749 };
10750
10751 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10752         .get_func_proto         = sk_lookup_func_proto,
10753         .is_valid_access        = sk_lookup_is_valid_access,
10754         .convert_ctx_access     = sk_lookup_convert_ctx_access,
10755 };
10756
10757 #endif /* CONFIG_INET */
10758
10759 DEFINE_BPF_DISPATCHER(xdp)
10760
10761 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10762 {
10763         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10764 }
10765
10766 #ifdef CONFIG_DEBUG_INFO_BTF
10767 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10768 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10769 BTF_SOCK_TYPE_xxx
10770 #undef BTF_SOCK_TYPE
10771 #else
10772 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10773 #endif
10774
10775 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10776 {
10777         /* tcp6_sock type is not generated in dwarf and hence btf,
10778          * trigger an explicit type generation here.
10779          */
10780         BTF_TYPE_EMIT(struct tcp6_sock);
10781         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10782             sk->sk_family == AF_INET6)
10783                 return (unsigned long)sk;
10784
10785         return (unsigned long)NULL;
10786 }
10787
10788 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10789         .func                   = bpf_skc_to_tcp6_sock,
10790         .gpl_only               = false,
10791         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10792         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10793         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10794 };
10795
10796 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10797 {
10798         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10799                 return (unsigned long)sk;
10800
10801         return (unsigned long)NULL;
10802 }
10803
10804 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10805         .func                   = bpf_skc_to_tcp_sock,
10806         .gpl_only               = false,
10807         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10808         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10809         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10810 };
10811
10812 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10813 {
10814         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10815          * generated if CONFIG_INET=n. Trigger an explicit generation here.
10816          */
10817         BTF_TYPE_EMIT(struct inet_timewait_sock);
10818         BTF_TYPE_EMIT(struct tcp_timewait_sock);
10819
10820 #ifdef CONFIG_INET
10821         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10822                 return (unsigned long)sk;
10823 #endif
10824
10825 #if IS_BUILTIN(CONFIG_IPV6)
10826         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10827                 return (unsigned long)sk;
10828 #endif
10829
10830         return (unsigned long)NULL;
10831 }
10832
10833 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10834         .func                   = bpf_skc_to_tcp_timewait_sock,
10835         .gpl_only               = false,
10836         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10837         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10838         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10839 };
10840
10841 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10842 {
10843 #ifdef CONFIG_INET
10844         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10845                 return (unsigned long)sk;
10846 #endif
10847
10848 #if IS_BUILTIN(CONFIG_IPV6)
10849         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10850                 return (unsigned long)sk;
10851 #endif
10852
10853         return (unsigned long)NULL;
10854 }
10855
10856 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10857         .func                   = bpf_skc_to_tcp_request_sock,
10858         .gpl_only               = false,
10859         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10860         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10861         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10862 };
10863
10864 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10865 {
10866         /* udp6_sock type is not generated in dwarf and hence btf,
10867          * trigger an explicit type generation here.
10868          */
10869         BTF_TYPE_EMIT(struct udp6_sock);
10870         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10871             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10872                 return (unsigned long)sk;
10873
10874         return (unsigned long)NULL;
10875 }
10876
10877 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10878         .func                   = bpf_skc_to_udp6_sock,
10879         .gpl_only               = false,
10880         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10881         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10882         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10883 };
10884
10885 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10886 {
10887         return (unsigned long)sock_from_file(file);
10888 }
10889
10890 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10891 BTF_ID(struct, socket)
10892 BTF_ID(struct, file)
10893
10894 const struct bpf_func_proto bpf_sock_from_file_proto = {
10895         .func           = bpf_sock_from_file,
10896         .gpl_only       = false,
10897         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
10898         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
10899         .arg1_type      = ARG_PTR_TO_BTF_ID,
10900         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
10901 };
10902
10903 static const struct bpf_func_proto *
10904 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10905 {
10906         const struct bpf_func_proto *func;
10907
10908         switch (func_id) {
10909         case BPF_FUNC_skc_to_tcp6_sock:
10910                 func = &bpf_skc_to_tcp6_sock_proto;
10911                 break;
10912         case BPF_FUNC_skc_to_tcp_sock:
10913                 func = &bpf_skc_to_tcp_sock_proto;
10914                 break;
10915         case BPF_FUNC_skc_to_tcp_timewait_sock:
10916                 func = &bpf_skc_to_tcp_timewait_sock_proto;
10917                 break;
10918         case BPF_FUNC_skc_to_tcp_request_sock:
10919                 func = &bpf_skc_to_tcp_request_sock_proto;
10920                 break;
10921         case BPF_FUNC_skc_to_udp6_sock:
10922                 func = &bpf_skc_to_udp6_sock_proto;
10923                 break;
10924         case BPF_FUNC_ktime_get_coarse_ns:
10925                 return &bpf_ktime_get_coarse_ns_proto;
10926         default:
10927                 return bpf_base_func_proto(func_id);
10928         }
10929
10930         if (!perfmon_capable())
10931                 return NULL;
10932
10933         return func;
10934 }