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         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3847         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3848                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3849                 return -EINVAL;
3850         }
3851
3852         if (unlikely(data_end < xdp->data + ETH_HLEN))
3853                 return -EINVAL;
3854
3855         /* Clear memory area on grow, can contain uninit kernel memory */
3856         if (offset > 0)
3857                 memset(xdp->data_end, 0, offset);
3858
3859         xdp->data_end = data_end;
3860
3861         return 0;
3862 }
3863
3864 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3865         .func           = bpf_xdp_adjust_tail,
3866         .gpl_only       = false,
3867         .ret_type       = RET_INTEGER,
3868         .arg1_type      = ARG_PTR_TO_CTX,
3869         .arg2_type      = ARG_ANYTHING,
3870 };
3871
3872 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3873 {
3874         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3875         void *meta = xdp->data_meta + offset;
3876         unsigned long metalen = xdp->data - meta;
3877
3878         if (xdp_data_meta_unsupported(xdp))
3879                 return -ENOTSUPP;
3880         if (unlikely(meta < xdp_frame_end ||
3881                      meta > xdp->data))
3882                 return -EINVAL;
3883         if (unlikely(xdp_metalen_invalid(metalen)))
3884                 return -EACCES;
3885
3886         xdp->data_meta = meta;
3887
3888         return 0;
3889 }
3890
3891 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3892         .func           = bpf_xdp_adjust_meta,
3893         .gpl_only       = false,
3894         .ret_type       = RET_INTEGER,
3895         .arg1_type      = ARG_PTR_TO_CTX,
3896         .arg2_type      = ARG_ANYTHING,
3897 };
3898
3899 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3900  * below:
3901  *
3902  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3903  *    of the redirect and store it (along with some other metadata) in a per-CPU
3904  *    struct bpf_redirect_info.
3905  *
3906  * 2. When the program returns the XDP_REDIRECT return code, the driver will
3907  *    call xdp_do_redirect() which will use the information in struct
3908  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
3909  *    bulk queue structure.
3910  *
3911  * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3912  *    which will flush all the different bulk queues, thus completing the
3913  *    redirect.
3914  *
3915  * Pointers to the map entries will be kept around for this whole sequence of
3916  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3917  * the core code; instead, the RCU protection relies on everything happening
3918  * inside a single NAPI poll sequence, which means it's between a pair of calls
3919  * to local_bh_disable()/local_bh_enable().
3920  *
3921  * The map entries are marked as __rcu and the map code makes sure to
3922  * dereference those pointers with rcu_dereference_check() in a way that works
3923  * for both sections that to hold an rcu_read_lock() and sections that are
3924  * called from NAPI without a separate rcu_read_lock(). The code below does not
3925  * use RCU annotations, but relies on those in the map code.
3926  */
3927 void xdp_do_flush(void)
3928 {
3929         __dev_flush();
3930         __cpu_map_flush();
3931         __xsk_map_flush();
3932 }
3933 EXPORT_SYMBOL_GPL(xdp_do_flush);
3934
3935 void bpf_clear_redirect_map(struct bpf_map *map)
3936 {
3937         struct bpf_redirect_info *ri;
3938         int cpu;
3939
3940         for_each_possible_cpu(cpu) {
3941                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3942                 /* Avoid polluting remote cacheline due to writes if
3943                  * not needed. Once we pass this test, we need the
3944                  * cmpxchg() to make sure it hasn't been changed in
3945                  * the meantime by remote CPU.
3946                  */
3947                 if (unlikely(READ_ONCE(ri->map) == map))
3948                         cmpxchg(&ri->map, map, NULL);
3949         }
3950 }
3951
3952 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
3953 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
3954
3955 u32 xdp_master_redirect(struct xdp_buff *xdp)
3956 {
3957         struct net_device *master, *slave;
3958         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3959
3960         master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
3961         slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
3962         if (slave && slave != xdp->rxq->dev) {
3963                 /* The target device is different from the receiving device, so
3964                  * redirect it to the new device.
3965                  * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
3966                  * drivers to unmap the packet from their rx ring.
3967                  */
3968                 ri->tgt_index = slave->ifindex;
3969                 ri->map_id = INT_MAX;
3970                 ri->map_type = BPF_MAP_TYPE_UNSPEC;
3971                 return XDP_REDIRECT;
3972         }
3973         return XDP_TX;
3974 }
3975 EXPORT_SYMBOL_GPL(xdp_master_redirect);
3976
3977 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3978                     struct bpf_prog *xdp_prog)
3979 {
3980         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3981         enum bpf_map_type map_type = ri->map_type;
3982         void *fwd = ri->tgt_value;
3983         u32 map_id = ri->map_id;
3984         struct bpf_map *map;
3985         int err;
3986
3987         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3988         ri->map_type = BPF_MAP_TYPE_UNSPEC;
3989
3990         switch (map_type) {
3991         case BPF_MAP_TYPE_DEVMAP:
3992                 fallthrough;
3993         case BPF_MAP_TYPE_DEVMAP_HASH:
3994                 map = READ_ONCE(ri->map);
3995                 if (unlikely(map)) {
3996                         WRITE_ONCE(ri->map, NULL);
3997                         err = dev_map_enqueue_multi(xdp, dev, map,
3998                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
3999                 } else {
4000                         err = dev_map_enqueue(fwd, xdp, dev);
4001                 }
4002                 break;
4003         case BPF_MAP_TYPE_CPUMAP:
4004                 err = cpu_map_enqueue(fwd, xdp, dev);
4005                 break;
4006         case BPF_MAP_TYPE_XSKMAP:
4007                 err = __xsk_map_redirect(fwd, xdp);
4008                 break;
4009         case BPF_MAP_TYPE_UNSPEC:
4010                 if (map_id == INT_MAX) {
4011                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4012                         if (unlikely(!fwd)) {
4013                                 err = -EINVAL;
4014                                 break;
4015                         }
4016                         err = dev_xdp_enqueue(fwd, xdp, dev);
4017                         break;
4018                 }
4019                 fallthrough;
4020         default:
4021                 err = -EBADRQC;
4022         }
4023
4024         if (unlikely(err))
4025                 goto err;
4026
4027         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4028         return 0;
4029 err:
4030         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4031         return err;
4032 }
4033 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4034
4035 static int xdp_do_generic_redirect_map(struct net_device *dev,
4036                                        struct sk_buff *skb,
4037                                        struct xdp_buff *xdp,
4038                                        struct bpf_prog *xdp_prog,
4039                                        void *fwd,
4040                                        enum bpf_map_type map_type, u32 map_id)
4041 {
4042         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4043         struct bpf_map *map;
4044         int err;
4045
4046         switch (map_type) {
4047         case BPF_MAP_TYPE_DEVMAP:
4048                 fallthrough;
4049         case BPF_MAP_TYPE_DEVMAP_HASH:
4050                 map = READ_ONCE(ri->map);
4051                 if (unlikely(map)) {
4052                         WRITE_ONCE(ri->map, NULL);
4053                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4054                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4055                 } else {
4056                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4057                 }
4058                 if (unlikely(err))
4059                         goto err;
4060                 break;
4061         case BPF_MAP_TYPE_XSKMAP:
4062                 err = xsk_generic_rcv(fwd, xdp);
4063                 if (err)
4064                         goto err;
4065                 consume_skb(skb);
4066                 break;
4067         case BPF_MAP_TYPE_CPUMAP:
4068                 err = cpu_map_generic_redirect(fwd, skb);
4069                 if (unlikely(err))
4070                         goto err;
4071                 break;
4072         default:
4073                 err = -EBADRQC;
4074                 goto err;
4075         }
4076
4077         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4078         return 0;
4079 err:
4080         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4081         return err;
4082 }
4083
4084 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4085                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4086 {
4087         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4088         enum bpf_map_type map_type = ri->map_type;
4089         void *fwd = ri->tgt_value;
4090         u32 map_id = ri->map_id;
4091         int err;
4092
4093         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4094         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4095
4096         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4097                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4098                 if (unlikely(!fwd)) {
4099                         err = -EINVAL;
4100                         goto err;
4101                 }
4102
4103                 err = xdp_ok_fwd_dev(fwd, skb->len);
4104                 if (unlikely(err))
4105                         goto err;
4106
4107                 skb->dev = fwd;
4108                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4109                 generic_xdp_tx(skb, xdp_prog);
4110                 return 0;
4111         }
4112
4113         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4114 err:
4115         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4116         return err;
4117 }
4118
4119 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4120 {
4121         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4122
4123         if (unlikely(flags))
4124                 return XDP_ABORTED;
4125
4126         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4127          * by map_idr) is used for ifindex based XDP redirect.
4128          */
4129         ri->tgt_index = ifindex;
4130         ri->map_id = INT_MAX;
4131         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4132
4133         return XDP_REDIRECT;
4134 }
4135
4136 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4137         .func           = bpf_xdp_redirect,
4138         .gpl_only       = false,
4139         .ret_type       = RET_INTEGER,
4140         .arg1_type      = ARG_ANYTHING,
4141         .arg2_type      = ARG_ANYTHING,
4142 };
4143
4144 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4145            u64, flags)
4146 {
4147         return map->ops->map_redirect(map, ifindex, flags);
4148 }
4149
4150 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4151         .func           = bpf_xdp_redirect_map,
4152         .gpl_only       = false,
4153         .ret_type       = RET_INTEGER,
4154         .arg1_type      = ARG_CONST_MAP_PTR,
4155         .arg2_type      = ARG_ANYTHING,
4156         .arg3_type      = ARG_ANYTHING,
4157 };
4158
4159 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4160                                   unsigned long off, unsigned long len)
4161 {
4162         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4163
4164         if (unlikely(!ptr))
4165                 return len;
4166         if (ptr != dst_buff)
4167                 memcpy(dst_buff, ptr, len);
4168
4169         return 0;
4170 }
4171
4172 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4173            u64, flags, void *, meta, u64, meta_size)
4174 {
4175         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4176
4177         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4178                 return -EINVAL;
4179         if (unlikely(!skb || skb_size > skb->len))
4180                 return -EFAULT;
4181
4182         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4183                                 bpf_skb_copy);
4184 }
4185
4186 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4187         .func           = bpf_skb_event_output,
4188         .gpl_only       = true,
4189         .ret_type       = RET_INTEGER,
4190         .arg1_type      = ARG_PTR_TO_CTX,
4191         .arg2_type      = ARG_CONST_MAP_PTR,
4192         .arg3_type      = ARG_ANYTHING,
4193         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4194         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4195 };
4196
4197 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4198
4199 const struct bpf_func_proto bpf_skb_output_proto = {
4200         .func           = bpf_skb_event_output,
4201         .gpl_only       = true,
4202         .ret_type       = RET_INTEGER,
4203         .arg1_type      = ARG_PTR_TO_BTF_ID,
4204         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4205         .arg2_type      = ARG_CONST_MAP_PTR,
4206         .arg3_type      = ARG_ANYTHING,
4207         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4208         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4209 };
4210
4211 static unsigned short bpf_tunnel_key_af(u64 flags)
4212 {
4213         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4214 }
4215
4216 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4217            u32, size, u64, flags)
4218 {
4219         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4220         u8 compat[sizeof(struct bpf_tunnel_key)];
4221         void *to_orig = to;
4222         int err;
4223
4224         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4225                 err = -EINVAL;
4226                 goto err_clear;
4227         }
4228         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4229                 err = -EPROTO;
4230                 goto err_clear;
4231         }
4232         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4233                 err = -EINVAL;
4234                 switch (size) {
4235                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4236                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4237                         goto set_compat;
4238                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4239                         /* Fixup deprecated structure layouts here, so we have
4240                          * a common path later on.
4241                          */
4242                         if (ip_tunnel_info_af(info) != AF_INET)
4243                                 goto err_clear;
4244 set_compat:
4245                         to = (struct bpf_tunnel_key *)compat;
4246                         break;
4247                 default:
4248                         goto err_clear;
4249                 }
4250         }
4251
4252         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4253         to->tunnel_tos = info->key.tos;
4254         to->tunnel_ttl = info->key.ttl;
4255         to->tunnel_ext = 0;
4256
4257         if (flags & BPF_F_TUNINFO_IPV6) {
4258                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4259                        sizeof(to->remote_ipv6));
4260                 to->tunnel_label = be32_to_cpu(info->key.label);
4261         } else {
4262                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4263                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4264                 to->tunnel_label = 0;
4265         }
4266
4267         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4268                 memcpy(to_orig, to, size);
4269
4270         return 0;
4271 err_clear:
4272         memset(to_orig, 0, size);
4273         return err;
4274 }
4275
4276 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4277         .func           = bpf_skb_get_tunnel_key,
4278         .gpl_only       = false,
4279         .ret_type       = RET_INTEGER,
4280         .arg1_type      = ARG_PTR_TO_CTX,
4281         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4282         .arg3_type      = ARG_CONST_SIZE,
4283         .arg4_type      = ARG_ANYTHING,
4284 };
4285
4286 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4287 {
4288         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4289         int err;
4290
4291         if (unlikely(!info ||
4292                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4293                 err = -ENOENT;
4294                 goto err_clear;
4295         }
4296         if (unlikely(size < info->options_len)) {
4297                 err = -ENOMEM;
4298                 goto err_clear;
4299         }
4300
4301         ip_tunnel_info_opts_get(to, info);
4302         if (size > info->options_len)
4303                 memset(to + info->options_len, 0, size - info->options_len);
4304
4305         return info->options_len;
4306 err_clear:
4307         memset(to, 0, size);
4308         return err;
4309 }
4310
4311 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4312         .func           = bpf_skb_get_tunnel_opt,
4313         .gpl_only       = false,
4314         .ret_type       = RET_INTEGER,
4315         .arg1_type      = ARG_PTR_TO_CTX,
4316         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4317         .arg3_type      = ARG_CONST_SIZE,
4318 };
4319
4320 static struct metadata_dst __percpu *md_dst;
4321
4322 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4323            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4324 {
4325         struct metadata_dst *md = this_cpu_ptr(md_dst);
4326         u8 compat[sizeof(struct bpf_tunnel_key)];
4327         struct ip_tunnel_info *info;
4328
4329         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4330                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4331                 return -EINVAL;
4332         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4333                 switch (size) {
4334                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4335                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4336                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4337                         /* Fixup deprecated structure layouts here, so we have
4338                          * a common path later on.
4339                          */
4340                         memcpy(compat, from, size);
4341                         memset(compat + size, 0, sizeof(compat) - size);
4342                         from = (const struct bpf_tunnel_key *) compat;
4343                         break;
4344                 default:
4345                         return -EINVAL;
4346                 }
4347         }
4348         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4349                      from->tunnel_ext))
4350                 return -EINVAL;
4351
4352         skb_dst_drop(skb);
4353         dst_hold((struct dst_entry *) md);
4354         skb_dst_set(skb, (struct dst_entry *) md);
4355
4356         info = &md->u.tun_info;
4357         memset(info, 0, sizeof(*info));
4358         info->mode = IP_TUNNEL_INFO_TX;
4359
4360         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4361         if (flags & BPF_F_DONT_FRAGMENT)
4362                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4363         if (flags & BPF_F_ZERO_CSUM_TX)
4364                 info->key.tun_flags &= ~TUNNEL_CSUM;
4365         if (flags & BPF_F_SEQ_NUMBER)
4366                 info->key.tun_flags |= TUNNEL_SEQ;
4367
4368         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4369         info->key.tos = from->tunnel_tos;
4370         info->key.ttl = from->tunnel_ttl;
4371
4372         if (flags & BPF_F_TUNINFO_IPV6) {
4373                 info->mode |= IP_TUNNEL_INFO_IPV6;
4374                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4375                        sizeof(from->remote_ipv6));
4376                 info->key.label = cpu_to_be32(from->tunnel_label) &
4377                                   IPV6_FLOWLABEL_MASK;
4378         } else {
4379                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4380         }
4381
4382         return 0;
4383 }
4384
4385 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4386         .func           = bpf_skb_set_tunnel_key,
4387         .gpl_only       = false,
4388         .ret_type       = RET_INTEGER,
4389         .arg1_type      = ARG_PTR_TO_CTX,
4390         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4391         .arg3_type      = ARG_CONST_SIZE,
4392         .arg4_type      = ARG_ANYTHING,
4393 };
4394
4395 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4396            const u8 *, from, u32, size)
4397 {
4398         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4399         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4400
4401         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4402                 return -EINVAL;
4403         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4404                 return -ENOMEM;
4405
4406         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4407
4408         return 0;
4409 }
4410
4411 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4412         .func           = bpf_skb_set_tunnel_opt,
4413         .gpl_only       = false,
4414         .ret_type       = RET_INTEGER,
4415         .arg1_type      = ARG_PTR_TO_CTX,
4416         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4417         .arg3_type      = ARG_CONST_SIZE,
4418 };
4419
4420 static const struct bpf_func_proto *
4421 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4422 {
4423         if (!md_dst) {
4424                 struct metadata_dst __percpu *tmp;
4425
4426                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4427                                                 METADATA_IP_TUNNEL,
4428                                                 GFP_KERNEL);
4429                 if (!tmp)
4430                         return NULL;
4431                 if (cmpxchg(&md_dst, NULL, tmp))
4432                         metadata_dst_free_percpu(tmp);
4433         }
4434
4435         switch (which) {
4436         case BPF_FUNC_skb_set_tunnel_key:
4437                 return &bpf_skb_set_tunnel_key_proto;
4438         case BPF_FUNC_skb_set_tunnel_opt:
4439                 return &bpf_skb_set_tunnel_opt_proto;
4440         default:
4441                 return NULL;
4442         }
4443 }
4444
4445 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4446            u32, idx)
4447 {
4448         struct bpf_array *array = container_of(map, struct bpf_array, map);
4449         struct cgroup *cgrp;
4450         struct sock *sk;
4451
4452         sk = skb_to_full_sk(skb);
4453         if (!sk || !sk_fullsock(sk))
4454                 return -ENOENT;
4455         if (unlikely(idx >= array->map.max_entries))
4456                 return -E2BIG;
4457
4458         cgrp = READ_ONCE(array->ptrs[idx]);
4459         if (unlikely(!cgrp))
4460                 return -EAGAIN;
4461
4462         return sk_under_cgroup_hierarchy(sk, cgrp);
4463 }
4464
4465 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4466         .func           = bpf_skb_under_cgroup,
4467         .gpl_only       = false,
4468         .ret_type       = RET_INTEGER,
4469         .arg1_type      = ARG_PTR_TO_CTX,
4470         .arg2_type      = ARG_CONST_MAP_PTR,
4471         .arg3_type      = ARG_ANYTHING,
4472 };
4473
4474 #ifdef CONFIG_SOCK_CGROUP_DATA
4475 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4476 {
4477         struct cgroup *cgrp;
4478
4479         sk = sk_to_full_sk(sk);
4480         if (!sk || !sk_fullsock(sk))
4481                 return 0;
4482
4483         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4484         return cgroup_id(cgrp);
4485 }
4486
4487 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4488 {
4489         return __bpf_sk_cgroup_id(skb->sk);
4490 }
4491
4492 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4493         .func           = bpf_skb_cgroup_id,
4494         .gpl_only       = false,
4495         .ret_type       = RET_INTEGER,
4496         .arg1_type      = ARG_PTR_TO_CTX,
4497 };
4498
4499 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4500                                               int ancestor_level)
4501 {
4502         struct cgroup *ancestor;
4503         struct cgroup *cgrp;
4504
4505         sk = sk_to_full_sk(sk);
4506         if (!sk || !sk_fullsock(sk))
4507                 return 0;
4508
4509         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4510         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4511         if (!ancestor)
4512                 return 0;
4513
4514         return cgroup_id(ancestor);
4515 }
4516
4517 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4518            ancestor_level)
4519 {
4520         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4521 }
4522
4523 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4524         .func           = bpf_skb_ancestor_cgroup_id,
4525         .gpl_only       = false,
4526         .ret_type       = RET_INTEGER,
4527         .arg1_type      = ARG_PTR_TO_CTX,
4528         .arg2_type      = ARG_ANYTHING,
4529 };
4530
4531 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4532 {
4533         return __bpf_sk_cgroup_id(sk);
4534 }
4535
4536 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4537         .func           = bpf_sk_cgroup_id,
4538         .gpl_only       = false,
4539         .ret_type       = RET_INTEGER,
4540         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4541 };
4542
4543 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4544 {
4545         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4546 }
4547
4548 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4549         .func           = bpf_sk_ancestor_cgroup_id,
4550         .gpl_only       = false,
4551         .ret_type       = RET_INTEGER,
4552         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4553         .arg2_type      = ARG_ANYTHING,
4554 };
4555 #endif
4556
4557 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4558                                   unsigned long off, unsigned long len)
4559 {
4560         memcpy(dst_buff, src_buff + off, len);
4561         return 0;
4562 }
4563
4564 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4565            u64, flags, void *, meta, u64, meta_size)
4566 {
4567         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4568
4569         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4570                 return -EINVAL;
4571         if (unlikely(!xdp ||
4572                      xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4573                 return -EFAULT;
4574
4575         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4576                                 xdp_size, bpf_xdp_copy);
4577 }
4578
4579 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4580         .func           = bpf_xdp_event_output,
4581         .gpl_only       = true,
4582         .ret_type       = RET_INTEGER,
4583         .arg1_type      = ARG_PTR_TO_CTX,
4584         .arg2_type      = ARG_CONST_MAP_PTR,
4585         .arg3_type      = ARG_ANYTHING,
4586         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4587         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4588 };
4589
4590 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4591
4592 const struct bpf_func_proto bpf_xdp_output_proto = {
4593         .func           = bpf_xdp_event_output,
4594         .gpl_only       = true,
4595         .ret_type       = RET_INTEGER,
4596         .arg1_type      = ARG_PTR_TO_BTF_ID,
4597         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4598         .arg2_type      = ARG_CONST_MAP_PTR,
4599         .arg3_type      = ARG_ANYTHING,
4600         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4601         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4602 };
4603
4604 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4605 {
4606         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4607 }
4608
4609 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4610         .func           = bpf_get_socket_cookie,
4611         .gpl_only       = false,
4612         .ret_type       = RET_INTEGER,
4613         .arg1_type      = ARG_PTR_TO_CTX,
4614 };
4615
4616 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4617 {
4618         return __sock_gen_cookie(ctx->sk);
4619 }
4620
4621 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4622         .func           = bpf_get_socket_cookie_sock_addr,
4623         .gpl_only       = false,
4624         .ret_type       = RET_INTEGER,
4625         .arg1_type      = ARG_PTR_TO_CTX,
4626 };
4627
4628 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4629 {
4630         return __sock_gen_cookie(ctx);
4631 }
4632
4633 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4634         .func           = bpf_get_socket_cookie_sock,
4635         .gpl_only       = false,
4636         .ret_type       = RET_INTEGER,
4637         .arg1_type      = ARG_PTR_TO_CTX,
4638 };
4639
4640 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4641 {
4642         return sk ? sock_gen_cookie(sk) : 0;
4643 }
4644
4645 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4646         .func           = bpf_get_socket_ptr_cookie,
4647         .gpl_only       = false,
4648         .ret_type       = RET_INTEGER,
4649         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4650 };
4651
4652 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4653 {
4654         return __sock_gen_cookie(ctx->sk);
4655 }
4656
4657 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4658         .func           = bpf_get_socket_cookie_sock_ops,
4659         .gpl_only       = false,
4660         .ret_type       = RET_INTEGER,
4661         .arg1_type      = ARG_PTR_TO_CTX,
4662 };
4663
4664 static u64 __bpf_get_netns_cookie(struct sock *sk)
4665 {
4666         const struct net *net = sk ? sock_net(sk) : &init_net;
4667
4668         return net->net_cookie;
4669 }
4670
4671 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4672 {
4673         return __bpf_get_netns_cookie(ctx);
4674 }
4675
4676 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4677         .func           = bpf_get_netns_cookie_sock,
4678         .gpl_only       = false,
4679         .ret_type       = RET_INTEGER,
4680         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4681 };
4682
4683 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4684 {
4685         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4686 }
4687
4688 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4689         .func           = bpf_get_netns_cookie_sock_addr,
4690         .gpl_only       = false,
4691         .ret_type       = RET_INTEGER,
4692         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4693 };
4694
4695 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4696 {
4697         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4698 }
4699
4700 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4701         .func           = bpf_get_netns_cookie_sock_ops,
4702         .gpl_only       = false,
4703         .ret_type       = RET_INTEGER,
4704         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4705 };
4706
4707 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4708 {
4709         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4710 }
4711
4712 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4713         .func           = bpf_get_netns_cookie_sk_msg,
4714         .gpl_only       = false,
4715         .ret_type       = RET_INTEGER,
4716         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4717 };
4718
4719 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4720 {
4721         struct sock *sk = sk_to_full_sk(skb->sk);
4722         kuid_t kuid;
4723
4724         if (!sk || !sk_fullsock(sk))
4725                 return overflowuid;
4726         kuid = sock_net_uid(sock_net(sk), sk);
4727         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4728 }
4729
4730 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4731         .func           = bpf_get_socket_uid,
4732         .gpl_only       = false,
4733         .ret_type       = RET_INTEGER,
4734         .arg1_type      = ARG_PTR_TO_CTX,
4735 };
4736
4737 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4738                            char *optval, int optlen)
4739 {
4740         char devname[IFNAMSIZ];
4741         int val, valbool;
4742         struct net *net;
4743         int ifindex;
4744         int ret = 0;
4745
4746         if (!sk_fullsock(sk))
4747                 return -EINVAL;
4748
4749         sock_owned_by_me(sk);
4750
4751         if (level == SOL_SOCKET) {
4752                 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4753                         return -EINVAL;
4754                 val = *((int *)optval);
4755                 valbool = val ? 1 : 0;
4756
4757                 /* Only some socketops are supported */
4758                 switch (optname) {
4759                 case SO_RCVBUF:
4760                         val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
4761                         val = min_t(int, val, INT_MAX / 2);
4762                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4763                         WRITE_ONCE(sk->sk_rcvbuf,
4764                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4765                         break;
4766                 case SO_SNDBUF:
4767                         val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
4768                         val = min_t(int, val, INT_MAX / 2);
4769                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4770                         WRITE_ONCE(sk->sk_sndbuf,
4771                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4772                         break;
4773                 case SO_MAX_PACING_RATE: /* 32bit version */
4774                         if (val != ~0U)
4775                                 cmpxchg(&sk->sk_pacing_status,
4776                                         SK_PACING_NONE,
4777                                         SK_PACING_NEEDED);
4778                         sk->sk_max_pacing_rate = (val == ~0U) ?
4779                                                  ~0UL : (unsigned int)val;
4780                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4781                                                  sk->sk_max_pacing_rate);
4782                         break;
4783                 case SO_PRIORITY:
4784                         sk->sk_priority = val;
4785                         break;
4786                 case SO_RCVLOWAT:
4787                         if (val < 0)
4788                                 val = INT_MAX;
4789                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4790                         break;
4791                 case SO_MARK:
4792                         if (sk->sk_mark != val) {
4793                                 sk->sk_mark = val;
4794                                 sk_dst_reset(sk);
4795                         }
4796                         break;
4797                 case SO_BINDTODEVICE:
4798                         optlen = min_t(long, optlen, IFNAMSIZ - 1);
4799                         strncpy(devname, optval, optlen);
4800                         devname[optlen] = 0;
4801
4802                         ifindex = 0;
4803                         if (devname[0] != '\0') {
4804                                 struct net_device *dev;
4805
4806                                 ret = -ENODEV;
4807
4808                                 net = sock_net(sk);
4809                                 dev = dev_get_by_name(net, devname);
4810                                 if (!dev)
4811                                         break;
4812                                 ifindex = dev->ifindex;
4813                                 dev_put(dev);
4814                         }
4815                         fallthrough;
4816                 case SO_BINDTOIFINDEX:
4817                         if (optname == SO_BINDTOIFINDEX)
4818                                 ifindex = val;
4819                         ret = sock_bindtoindex(sk, ifindex, false);
4820                         break;
4821                 case SO_KEEPALIVE:
4822                         if (sk->sk_prot->keepalive)
4823                                 sk->sk_prot->keepalive(sk, valbool);
4824                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4825                         break;
4826                 case SO_REUSEPORT:
4827                         sk->sk_reuseport = valbool;
4828                         break;
4829                 default:
4830                         ret = -EINVAL;
4831                 }
4832 #ifdef CONFIG_INET
4833         } else if (level == SOL_IP) {
4834                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4835                         return -EINVAL;
4836
4837                 val = *((int *)optval);
4838                 /* Only some options are supported */
4839                 switch (optname) {
4840                 case IP_TOS:
4841                         if (val < -1 || val > 0xff) {
4842                                 ret = -EINVAL;
4843                         } else {
4844                                 struct inet_sock *inet = inet_sk(sk);
4845
4846                                 if (val == -1)
4847                                         val = 0;
4848                                 inet->tos = val;
4849                         }
4850                         break;
4851                 default:
4852                         ret = -EINVAL;
4853                 }
4854 #if IS_ENABLED(CONFIG_IPV6)
4855         } else if (level == SOL_IPV6) {
4856                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4857                         return -EINVAL;
4858
4859                 val = *((int *)optval);
4860                 /* Only some options are supported */
4861                 switch (optname) {
4862                 case IPV6_TCLASS:
4863                         if (val < -1 || val > 0xff) {
4864                                 ret = -EINVAL;
4865                         } else {
4866                                 struct ipv6_pinfo *np = inet6_sk(sk);
4867
4868                                 if (val == -1)
4869                                         val = 0;
4870                                 np->tclass = val;
4871                         }
4872                         break;
4873                 default:
4874                         ret = -EINVAL;
4875                 }
4876 #endif
4877         } else if (level == SOL_TCP &&
4878                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4879                 if (optname == TCP_CONGESTION) {
4880                         char name[TCP_CA_NAME_MAX];
4881
4882                         strncpy(name, optval, min_t(long, optlen,
4883                                                     TCP_CA_NAME_MAX-1));
4884                         name[TCP_CA_NAME_MAX-1] = 0;
4885                         ret = tcp_set_congestion_control(sk, name, false, true);
4886                 } else {
4887                         struct inet_connection_sock *icsk = inet_csk(sk);
4888                         struct tcp_sock *tp = tcp_sk(sk);
4889                         unsigned long timeout;
4890
4891                         if (optlen != sizeof(int))
4892                                 return -EINVAL;
4893
4894                         val = *((int *)optval);
4895                         /* Only some options are supported */
4896                         switch (optname) {
4897                         case TCP_BPF_IW:
4898                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4899                                         ret = -EINVAL;
4900                                 else
4901                                         tcp_snd_cwnd_set(tp, val);
4902                                 break;
4903                         case TCP_BPF_SNDCWND_CLAMP:
4904                                 if (val <= 0) {
4905                                         ret = -EINVAL;
4906                                 } else {
4907                                         tp->snd_cwnd_clamp = val;
4908                                         tp->snd_ssthresh = val;
4909                                 }
4910                                 break;
4911                         case TCP_BPF_DELACK_MAX:
4912                                 timeout = usecs_to_jiffies(val);
4913                                 if (timeout > TCP_DELACK_MAX ||
4914                                     timeout < TCP_TIMEOUT_MIN)
4915                                         return -EINVAL;
4916                                 inet_csk(sk)->icsk_delack_max = timeout;
4917                                 break;
4918                         case TCP_BPF_RTO_MIN:
4919                                 timeout = usecs_to_jiffies(val);
4920                                 if (timeout > TCP_RTO_MIN ||
4921                                     timeout < TCP_TIMEOUT_MIN)
4922                                         return -EINVAL;
4923                                 inet_csk(sk)->icsk_rto_min = timeout;
4924                                 break;
4925                         case TCP_SAVE_SYN:
4926                                 if (val < 0 || val > 1)
4927                                         ret = -EINVAL;
4928                                 else
4929                                         tp->save_syn = val;
4930                                 break;
4931                         case TCP_KEEPIDLE:
4932                                 ret = tcp_sock_set_keepidle_locked(sk, val);
4933                                 break;
4934                         case TCP_KEEPINTVL:
4935                                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4936                                         ret = -EINVAL;
4937                                 else
4938                                         tp->keepalive_intvl = val * HZ;
4939                                 break;
4940                         case TCP_KEEPCNT:
4941                                 if (val < 1 || val > MAX_TCP_KEEPCNT)
4942                                         ret = -EINVAL;
4943                                 else
4944                                         tp->keepalive_probes = val;
4945                                 break;
4946                         case TCP_SYNCNT:
4947                                 if (val < 1 || val > MAX_TCP_SYNCNT)
4948                                         ret = -EINVAL;
4949                                 else
4950                                         icsk->icsk_syn_retries = val;
4951                                 break;
4952                         case TCP_USER_TIMEOUT:
4953                                 if (val < 0)
4954                                         ret = -EINVAL;
4955                                 else
4956                                         icsk->icsk_user_timeout = val;
4957                                 break;
4958                         case TCP_NOTSENT_LOWAT:
4959                                 tp->notsent_lowat = val;
4960                                 sk->sk_write_space(sk);
4961                                 break;
4962                         case TCP_WINDOW_CLAMP:
4963                                 ret = tcp_set_window_clamp(sk, val);
4964                                 break;
4965                         default:
4966                                 ret = -EINVAL;
4967                         }
4968                 }
4969 #endif
4970         } else {
4971                 ret = -EINVAL;
4972         }
4973         return ret;
4974 }
4975
4976 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4977                            char *optval, int optlen)
4978 {
4979         if (!sk_fullsock(sk))
4980                 goto err_clear;
4981
4982         sock_owned_by_me(sk);
4983
4984         if (level == SOL_SOCKET) {
4985                 if (optlen != sizeof(int))
4986                         goto err_clear;
4987
4988                 switch (optname) {
4989                 case SO_MARK:
4990                         *((int *)optval) = sk->sk_mark;
4991                         break;
4992                 case SO_PRIORITY:
4993                         *((int *)optval) = sk->sk_priority;
4994                         break;
4995                 case SO_BINDTOIFINDEX:
4996                         *((int *)optval) = sk->sk_bound_dev_if;
4997                         break;
4998                 case SO_REUSEPORT:
4999                         *((int *)optval) = sk->sk_reuseport;
5000                         break;
5001                 default:
5002                         goto err_clear;
5003                 }
5004 #ifdef CONFIG_INET
5005         } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
5006                 struct inet_connection_sock *icsk;
5007                 struct tcp_sock *tp;
5008
5009                 switch (optname) {
5010                 case TCP_CONGESTION:
5011                         icsk = inet_csk(sk);
5012
5013                         if (!icsk->icsk_ca_ops || optlen <= 1)
5014                                 goto err_clear;
5015                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
5016                         optval[optlen - 1] = 0;
5017                         break;
5018                 case TCP_SAVED_SYN:
5019                         tp = tcp_sk(sk);
5020
5021                         if (optlen <= 0 || !tp->saved_syn ||
5022                             optlen > tcp_saved_syn_len(tp->saved_syn))
5023                                 goto err_clear;
5024                         memcpy(optval, tp->saved_syn->data, optlen);
5025                         break;
5026                 default:
5027                         goto err_clear;
5028                 }
5029         } else if (level == SOL_IP) {
5030                 struct inet_sock *inet = inet_sk(sk);
5031
5032                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
5033                         goto err_clear;
5034
5035                 /* Only some options are supported */
5036                 switch (optname) {
5037                 case IP_TOS:
5038                         *((int *)optval) = (int)inet->tos;
5039                         break;
5040                 default:
5041                         goto err_clear;
5042                 }
5043 #if IS_ENABLED(CONFIG_IPV6)
5044         } else if (level == SOL_IPV6) {
5045                 struct ipv6_pinfo *np = inet6_sk(sk);
5046
5047                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5048                         goto err_clear;
5049
5050                 /* Only some options are supported */
5051                 switch (optname) {
5052                 case IPV6_TCLASS:
5053                         *((int *)optval) = (int)np->tclass;
5054                         break;
5055                 default:
5056                         goto err_clear;
5057                 }
5058 #endif
5059 #endif
5060         } else {
5061                 goto err_clear;
5062         }
5063         return 0;
5064 err_clear:
5065         memset(optval, 0, optlen);
5066         return -EINVAL;
5067 }
5068
5069 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5070            int, optname, char *, optval, int, optlen)
5071 {
5072         if (level == SOL_TCP && optname == TCP_CONGESTION) {
5073                 if (optlen >= sizeof("cdg") - 1 &&
5074                     !strncmp("cdg", optval, optlen))
5075                         return -ENOTSUPP;
5076         }
5077
5078         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5079 }
5080
5081 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5082         .func           = bpf_sk_setsockopt,
5083         .gpl_only       = false,
5084         .ret_type       = RET_INTEGER,
5085         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5086         .arg2_type      = ARG_ANYTHING,
5087         .arg3_type      = ARG_ANYTHING,
5088         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5089         .arg5_type      = ARG_CONST_SIZE,
5090 };
5091
5092 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5093            int, optname, char *, optval, int, optlen)
5094 {
5095         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5096 }
5097
5098 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5099         .func           = bpf_sk_getsockopt,
5100         .gpl_only       = false,
5101         .ret_type       = RET_INTEGER,
5102         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5103         .arg2_type      = ARG_ANYTHING,
5104         .arg3_type      = ARG_ANYTHING,
5105         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5106         .arg5_type      = ARG_CONST_SIZE,
5107 };
5108
5109 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5110            int, level, int, optname, char *, optval, int, optlen)
5111 {
5112         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5113 }
5114
5115 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5116         .func           = bpf_sock_addr_setsockopt,
5117         .gpl_only       = false,
5118         .ret_type       = RET_INTEGER,
5119         .arg1_type      = ARG_PTR_TO_CTX,
5120         .arg2_type      = ARG_ANYTHING,
5121         .arg3_type      = ARG_ANYTHING,
5122         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5123         .arg5_type      = ARG_CONST_SIZE,
5124 };
5125
5126 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5127            int, level, int, optname, char *, optval, int, optlen)
5128 {
5129         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5130 }
5131
5132 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5133         .func           = bpf_sock_addr_getsockopt,
5134         .gpl_only       = false,
5135         .ret_type       = RET_INTEGER,
5136         .arg1_type      = ARG_PTR_TO_CTX,
5137         .arg2_type      = ARG_ANYTHING,
5138         .arg3_type      = ARG_ANYTHING,
5139         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5140         .arg5_type      = ARG_CONST_SIZE,
5141 };
5142
5143 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5144            int, level, int, optname, char *, optval, int, optlen)
5145 {
5146         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5147 }
5148
5149 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5150         .func           = bpf_sock_ops_setsockopt,
5151         .gpl_only       = false,
5152         .ret_type       = RET_INTEGER,
5153         .arg1_type      = ARG_PTR_TO_CTX,
5154         .arg2_type      = ARG_ANYTHING,
5155         .arg3_type      = ARG_ANYTHING,
5156         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5157         .arg5_type      = ARG_CONST_SIZE,
5158 };
5159
5160 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5161                                 int optname, const u8 **start)
5162 {
5163         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5164         const u8 *hdr_start;
5165         int ret;
5166
5167         if (syn_skb) {
5168                 /* sk is a request_sock here */
5169
5170                 if (optname == TCP_BPF_SYN) {
5171                         hdr_start = syn_skb->data;
5172                         ret = tcp_hdrlen(syn_skb);
5173                 } else if (optname == TCP_BPF_SYN_IP) {
5174                         hdr_start = skb_network_header(syn_skb);
5175                         ret = skb_network_header_len(syn_skb) +
5176                                 tcp_hdrlen(syn_skb);
5177                 } else {
5178                         /* optname == TCP_BPF_SYN_MAC */
5179                         hdr_start = skb_mac_header(syn_skb);
5180                         ret = skb_mac_header_len(syn_skb) +
5181                                 skb_network_header_len(syn_skb) +
5182                                 tcp_hdrlen(syn_skb);
5183                 }
5184         } else {
5185                 struct sock *sk = bpf_sock->sk;
5186                 struct saved_syn *saved_syn;
5187
5188                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5189                         /* synack retransmit. bpf_sock->syn_skb will
5190                          * not be available.  It has to resort to
5191                          * saved_syn (if it is saved).
5192                          */
5193                         saved_syn = inet_reqsk(sk)->saved_syn;
5194                 else
5195                         saved_syn = tcp_sk(sk)->saved_syn;
5196
5197                 if (!saved_syn)
5198                         return -ENOENT;
5199
5200                 if (optname == TCP_BPF_SYN) {
5201                         hdr_start = saved_syn->data +
5202                                 saved_syn->mac_hdrlen +
5203                                 saved_syn->network_hdrlen;
5204                         ret = saved_syn->tcp_hdrlen;
5205                 } else if (optname == TCP_BPF_SYN_IP) {
5206                         hdr_start = saved_syn->data +
5207                                 saved_syn->mac_hdrlen;
5208                         ret = saved_syn->network_hdrlen +
5209                                 saved_syn->tcp_hdrlen;
5210                 } else {
5211                         /* optname == TCP_BPF_SYN_MAC */
5212
5213                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5214                         if (!saved_syn->mac_hdrlen)
5215                                 return -ENOENT;
5216
5217                         hdr_start = saved_syn->data;
5218                         ret = saved_syn->mac_hdrlen +
5219                                 saved_syn->network_hdrlen +
5220                                 saved_syn->tcp_hdrlen;
5221                 }
5222         }
5223
5224         *start = hdr_start;
5225         return ret;
5226 }
5227
5228 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5229            int, level, int, optname, char *, optval, int, optlen)
5230 {
5231         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5232             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5233                 int ret, copy_len = 0;
5234                 const u8 *start;
5235
5236                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5237                 if (ret > 0) {
5238                         copy_len = ret;
5239                         if (optlen < copy_len) {
5240                                 copy_len = optlen;
5241                                 ret = -ENOSPC;
5242                         }
5243
5244                         memcpy(optval, start, copy_len);
5245                 }
5246
5247                 /* Zero out unused buffer at the end */
5248                 memset(optval + copy_len, 0, optlen - copy_len);
5249
5250                 return ret;
5251         }
5252
5253         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5254 }
5255
5256 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5257         .func           = bpf_sock_ops_getsockopt,
5258         .gpl_only       = false,
5259         .ret_type       = RET_INTEGER,
5260         .arg1_type      = ARG_PTR_TO_CTX,
5261         .arg2_type      = ARG_ANYTHING,
5262         .arg3_type      = ARG_ANYTHING,
5263         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5264         .arg5_type      = ARG_CONST_SIZE,
5265 };
5266
5267 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5268            int, argval)
5269 {
5270         struct sock *sk = bpf_sock->sk;
5271         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5272
5273         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5274                 return -EINVAL;
5275
5276         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5277
5278         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5279 }
5280
5281 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5282         .func           = bpf_sock_ops_cb_flags_set,
5283         .gpl_only       = false,
5284         .ret_type       = RET_INTEGER,
5285         .arg1_type      = ARG_PTR_TO_CTX,
5286         .arg2_type      = ARG_ANYTHING,
5287 };
5288
5289 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5290 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5291
5292 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5293            int, addr_len)
5294 {
5295 #ifdef CONFIG_INET
5296         struct sock *sk = ctx->sk;
5297         u32 flags = BIND_FROM_BPF;
5298         int err;
5299
5300         err = -EINVAL;
5301         if (addr_len < offsetofend(struct sockaddr, sa_family))
5302                 return err;
5303         if (addr->sa_family == AF_INET) {
5304                 if (addr_len < sizeof(struct sockaddr_in))
5305                         return err;
5306                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5307                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5308                 return __inet_bind(sk, addr, addr_len, flags);
5309 #if IS_ENABLED(CONFIG_IPV6)
5310         } else if (addr->sa_family == AF_INET6) {
5311                 if (addr_len < SIN6_LEN_RFC2133)
5312                         return err;
5313                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5314                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5315                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5316                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5317                  */
5318                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5319 #endif /* CONFIG_IPV6 */
5320         }
5321 #endif /* CONFIG_INET */
5322
5323         return -EAFNOSUPPORT;
5324 }
5325
5326 static const struct bpf_func_proto bpf_bind_proto = {
5327         .func           = bpf_bind,
5328         .gpl_only       = false,
5329         .ret_type       = RET_INTEGER,
5330         .arg1_type      = ARG_PTR_TO_CTX,
5331         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5332         .arg3_type      = ARG_CONST_SIZE,
5333 };
5334
5335 #ifdef CONFIG_XFRM
5336 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5337            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5338 {
5339         const struct sec_path *sp = skb_sec_path(skb);
5340         const struct xfrm_state *x;
5341
5342         if (!sp || unlikely(index >= sp->len || flags))
5343                 goto err_clear;
5344
5345         x = sp->xvec[index];
5346
5347         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5348                 goto err_clear;
5349
5350         to->reqid = x->props.reqid;
5351         to->spi = x->id.spi;
5352         to->family = x->props.family;
5353         to->ext = 0;
5354
5355         if (to->family == AF_INET6) {
5356                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5357                        sizeof(to->remote_ipv6));
5358         } else {
5359                 to->remote_ipv4 = x->props.saddr.a4;
5360                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5361         }
5362
5363         return 0;
5364 err_clear:
5365         memset(to, 0, size);
5366         return -EINVAL;
5367 }
5368
5369 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5370         .func           = bpf_skb_get_xfrm_state,
5371         .gpl_only       = false,
5372         .ret_type       = RET_INTEGER,
5373         .arg1_type      = ARG_PTR_TO_CTX,
5374         .arg2_type      = ARG_ANYTHING,
5375         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5376         .arg4_type      = ARG_CONST_SIZE,
5377         .arg5_type      = ARG_ANYTHING,
5378 };
5379 #endif
5380
5381 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5382 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5383                                   const struct neighbour *neigh,
5384                                   const struct net_device *dev, u32 mtu)
5385 {
5386         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5387         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5388         params->h_vlan_TCI = 0;
5389         params->h_vlan_proto = 0;
5390         if (mtu)
5391                 params->mtu_result = mtu; /* union with tot_len */
5392
5393         return 0;
5394 }
5395 #endif
5396
5397 #if IS_ENABLED(CONFIG_INET)
5398 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5399                                u32 flags, bool check_mtu)
5400 {
5401         struct fib_nh_common *nhc;
5402         struct in_device *in_dev;
5403         struct neighbour *neigh;
5404         struct net_device *dev;
5405         struct fib_result res;
5406         struct flowi4 fl4;
5407         u32 mtu = 0;
5408         int err;
5409
5410         dev = dev_get_by_index_rcu(net, params->ifindex);
5411         if (unlikely(!dev))
5412                 return -ENODEV;
5413
5414         /* verify forwarding is enabled on this interface */
5415         in_dev = __in_dev_get_rcu(dev);
5416         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5417                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5418
5419         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5420                 fl4.flowi4_iif = 1;
5421                 fl4.flowi4_oif = params->ifindex;
5422         } else {
5423                 fl4.flowi4_iif = params->ifindex;
5424                 fl4.flowi4_oif = 0;
5425         }
5426         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5427         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5428         fl4.flowi4_flags = 0;
5429
5430         fl4.flowi4_proto = params->l4_protocol;
5431         fl4.daddr = params->ipv4_dst;
5432         fl4.saddr = params->ipv4_src;
5433         fl4.fl4_sport = params->sport;
5434         fl4.fl4_dport = params->dport;
5435         fl4.flowi4_multipath_hash = 0;
5436
5437         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5438                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5439                 struct fib_table *tb;
5440
5441                 tb = fib_get_table(net, tbid);
5442                 if (unlikely(!tb))
5443                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5444
5445                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5446         } else {
5447                 fl4.flowi4_mark = 0;
5448                 fl4.flowi4_secid = 0;
5449                 fl4.flowi4_tun_key.tun_id = 0;
5450                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5451
5452                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5453         }
5454
5455         if (err) {
5456                 /* map fib lookup errors to RTN_ type */
5457                 if (err == -EINVAL)
5458                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5459                 if (err == -EHOSTUNREACH)
5460                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5461                 if (err == -EACCES)
5462                         return BPF_FIB_LKUP_RET_PROHIBIT;
5463
5464                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5465         }
5466
5467         if (res.type != RTN_UNICAST)
5468                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5469
5470         if (fib_info_num_path(res.fi) > 1)
5471                 fib_select_path(net, &res, &fl4, NULL);
5472
5473         if (check_mtu) {
5474                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5475                 if (params->tot_len > mtu) {
5476                         params->mtu_result = mtu; /* union with tot_len */
5477                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5478                 }
5479         }
5480
5481         nhc = res.nhc;
5482
5483         /* do not handle lwt encaps right now */
5484         if (nhc->nhc_lwtstate)
5485                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5486
5487         dev = nhc->nhc_dev;
5488
5489         params->rt_metric = res.fi->fib_priority;
5490         params->ifindex = dev->ifindex;
5491
5492         /* xdp and cls_bpf programs are run in RCU-bh so
5493          * rcu_read_lock_bh is not needed here
5494          */
5495         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5496                 if (nhc->nhc_gw_family)
5497                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5498
5499                 neigh = __ipv4_neigh_lookup_noref(dev,
5500                                                  (__force u32)params->ipv4_dst);
5501         } else {
5502                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5503
5504                 params->family = AF_INET6;
5505                 *dst = nhc->nhc_gw.ipv6;
5506                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5507         }
5508
5509         if (!neigh)
5510                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5511
5512         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5513 }
5514 #endif
5515
5516 #if IS_ENABLED(CONFIG_IPV6)
5517 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5518                                u32 flags, bool check_mtu)
5519 {
5520         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5521         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5522         struct fib6_result res = {};
5523         struct neighbour *neigh;
5524         struct net_device *dev;
5525         struct inet6_dev *idev;
5526         struct flowi6 fl6;
5527         int strict = 0;
5528         int oif, err;
5529         u32 mtu = 0;
5530
5531         /* link local addresses are never forwarded */
5532         if (rt6_need_strict(dst) || rt6_need_strict(src))
5533                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5534
5535         dev = dev_get_by_index_rcu(net, params->ifindex);
5536         if (unlikely(!dev))
5537                 return -ENODEV;
5538
5539         idev = __in6_dev_get_safely(dev);
5540         if (unlikely(!idev || !idev->cnf.forwarding))
5541                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5542
5543         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5544                 fl6.flowi6_iif = 1;
5545                 oif = fl6.flowi6_oif = params->ifindex;
5546         } else {
5547                 oif = fl6.flowi6_iif = params->ifindex;
5548                 fl6.flowi6_oif = 0;
5549                 strict = RT6_LOOKUP_F_HAS_SADDR;
5550         }
5551         fl6.flowlabel = params->flowinfo;
5552         fl6.flowi6_scope = 0;
5553         fl6.flowi6_flags = 0;
5554         fl6.mp_hash = 0;
5555
5556         fl6.flowi6_proto = params->l4_protocol;
5557         fl6.daddr = *dst;
5558         fl6.saddr = *src;
5559         fl6.fl6_sport = params->sport;
5560         fl6.fl6_dport = params->dport;
5561
5562         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5563                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5564                 struct fib6_table *tb;
5565
5566                 tb = ipv6_stub->fib6_get_table(net, tbid);
5567                 if (unlikely(!tb))
5568                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5569
5570                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5571                                                    strict);
5572         } else {
5573                 fl6.flowi6_mark = 0;
5574                 fl6.flowi6_secid = 0;
5575                 fl6.flowi6_tun_key.tun_id = 0;
5576                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5577
5578                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5579         }
5580
5581         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5582                      res.f6i == net->ipv6.fib6_null_entry))
5583                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5584
5585         switch (res.fib6_type) {
5586         /* only unicast is forwarded */
5587         case RTN_UNICAST:
5588                 break;
5589         case RTN_BLACKHOLE:
5590                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5591         case RTN_UNREACHABLE:
5592                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5593         case RTN_PROHIBIT:
5594                 return BPF_FIB_LKUP_RET_PROHIBIT;
5595         default:
5596                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5597         }
5598
5599         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5600                                     fl6.flowi6_oif != 0, NULL, strict);
5601
5602         if (check_mtu) {
5603                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5604                 if (params->tot_len > mtu) {
5605                         params->mtu_result = mtu; /* union with tot_len */
5606                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5607                 }
5608         }
5609
5610         if (res.nh->fib_nh_lws)
5611                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5612
5613         if (res.nh->fib_nh_gw_family)
5614                 *dst = res.nh->fib_nh_gw6;
5615
5616         dev = res.nh->fib_nh_dev;
5617         params->rt_metric = res.f6i->fib6_metric;
5618         params->ifindex = dev->ifindex;
5619
5620         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5621          * not needed here.
5622          */
5623         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5624         if (!neigh)
5625                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5626
5627         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5628 }
5629 #endif
5630
5631 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5632            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5633 {
5634         if (plen < sizeof(*params))
5635                 return -EINVAL;
5636
5637         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5638                 return -EINVAL;
5639
5640         switch (params->family) {
5641 #if IS_ENABLED(CONFIG_INET)
5642         case AF_INET:
5643                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5644                                            flags, true);
5645 #endif
5646 #if IS_ENABLED(CONFIG_IPV6)
5647         case AF_INET6:
5648                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5649                                            flags, true);
5650 #endif
5651         }
5652         return -EAFNOSUPPORT;
5653 }
5654
5655 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5656         .func           = bpf_xdp_fib_lookup,
5657         .gpl_only       = true,
5658         .ret_type       = RET_INTEGER,
5659         .arg1_type      = ARG_PTR_TO_CTX,
5660         .arg2_type      = ARG_PTR_TO_MEM,
5661         .arg3_type      = ARG_CONST_SIZE,
5662         .arg4_type      = ARG_ANYTHING,
5663 };
5664
5665 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5666            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5667 {
5668         struct net *net = dev_net(skb->dev);
5669         int rc = -EAFNOSUPPORT;
5670         bool check_mtu = false;
5671
5672         if (plen < sizeof(*params))
5673                 return -EINVAL;
5674
5675         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5676                 return -EINVAL;
5677
5678         if (params->tot_len)
5679                 check_mtu = true;
5680
5681         switch (params->family) {
5682 #if IS_ENABLED(CONFIG_INET)
5683         case AF_INET:
5684                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5685                 break;
5686 #endif
5687 #if IS_ENABLED(CONFIG_IPV6)
5688         case AF_INET6:
5689                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5690                 break;
5691 #endif
5692         }
5693
5694         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5695                 struct net_device *dev;
5696
5697                 /* When tot_len isn't provided by user, check skb
5698                  * against MTU of FIB lookup resulting net_device
5699                  */
5700                 dev = dev_get_by_index_rcu(net, params->ifindex);
5701                 if (!is_skb_forwardable(dev, skb))
5702                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5703
5704                 params->mtu_result = dev->mtu; /* union with tot_len */
5705         }
5706
5707         return rc;
5708 }
5709
5710 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5711         .func           = bpf_skb_fib_lookup,
5712         .gpl_only       = true,
5713         .ret_type       = RET_INTEGER,
5714         .arg1_type      = ARG_PTR_TO_CTX,
5715         .arg2_type      = ARG_PTR_TO_MEM,
5716         .arg3_type      = ARG_CONST_SIZE,
5717         .arg4_type      = ARG_ANYTHING,
5718 };
5719
5720 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5721                                             u32 ifindex)
5722 {
5723         struct net *netns = dev_net(dev_curr);
5724
5725         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5726         if (ifindex == 0)
5727                 return dev_curr;
5728
5729         return dev_get_by_index_rcu(netns, ifindex);
5730 }
5731
5732 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5733            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5734 {
5735         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5736         struct net_device *dev = skb->dev;
5737         int skb_len, dev_len;
5738         int mtu;
5739
5740         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5741                 return -EINVAL;
5742
5743         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5744                 return -EINVAL;
5745
5746         dev = __dev_via_ifindex(dev, ifindex);
5747         if (unlikely(!dev))
5748                 return -ENODEV;
5749
5750         mtu = READ_ONCE(dev->mtu);
5751
5752         dev_len = mtu + dev->hard_header_len;
5753
5754         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5755         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5756
5757         skb_len += len_diff; /* minus result pass check */
5758         if (skb_len <= dev_len) {
5759                 ret = BPF_MTU_CHK_RET_SUCCESS;
5760                 goto out;
5761         }
5762         /* At this point, skb->len exceed MTU, but as it include length of all
5763          * segments, it can still be below MTU.  The SKB can possibly get
5764          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
5765          * must choose if segs are to be MTU checked.
5766          */
5767         if (skb_is_gso(skb)) {
5768                 ret = BPF_MTU_CHK_RET_SUCCESS;
5769
5770                 if (flags & BPF_MTU_CHK_SEGS &&
5771                     !skb_gso_validate_network_len(skb, mtu))
5772                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5773         }
5774 out:
5775         /* BPF verifier guarantees valid pointer */
5776         *mtu_len = mtu;
5777
5778         return ret;
5779 }
5780
5781 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5782            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5783 {
5784         struct net_device *dev = xdp->rxq->dev;
5785         int xdp_len = xdp->data_end - xdp->data;
5786         int ret = BPF_MTU_CHK_RET_SUCCESS;
5787         int mtu, dev_len;
5788
5789         /* XDP variant doesn't support multi-buffer segment check (yet) */
5790         if (unlikely(flags))
5791                 return -EINVAL;
5792
5793         dev = __dev_via_ifindex(dev, ifindex);
5794         if (unlikely(!dev))
5795                 return -ENODEV;
5796
5797         mtu = READ_ONCE(dev->mtu);
5798
5799         /* Add L2-header as dev MTU is L3 size */
5800         dev_len = mtu + dev->hard_header_len;
5801
5802         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5803         if (*mtu_len)
5804                 xdp_len = *mtu_len + dev->hard_header_len;
5805
5806         xdp_len += len_diff; /* minus result pass check */
5807         if (xdp_len > dev_len)
5808                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5809
5810         /* BPF verifier guarantees valid pointer */
5811         *mtu_len = mtu;
5812
5813         return ret;
5814 }
5815
5816 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5817         .func           = bpf_skb_check_mtu,
5818         .gpl_only       = true,
5819         .ret_type       = RET_INTEGER,
5820         .arg1_type      = ARG_PTR_TO_CTX,
5821         .arg2_type      = ARG_ANYTHING,
5822         .arg3_type      = ARG_PTR_TO_INT,
5823         .arg4_type      = ARG_ANYTHING,
5824         .arg5_type      = ARG_ANYTHING,
5825 };
5826
5827 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5828         .func           = bpf_xdp_check_mtu,
5829         .gpl_only       = true,
5830         .ret_type       = RET_INTEGER,
5831         .arg1_type      = ARG_PTR_TO_CTX,
5832         .arg2_type      = ARG_ANYTHING,
5833         .arg3_type      = ARG_PTR_TO_INT,
5834         .arg4_type      = ARG_ANYTHING,
5835         .arg5_type      = ARG_ANYTHING,
5836 };
5837
5838 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5839 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5840 {
5841         int err;
5842         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5843
5844         if (!seg6_validate_srh(srh, len, false))
5845                 return -EINVAL;
5846
5847         switch (type) {
5848         case BPF_LWT_ENCAP_SEG6_INLINE:
5849                 if (skb->protocol != htons(ETH_P_IPV6))
5850                         return -EBADMSG;
5851
5852                 err = seg6_do_srh_inline(skb, srh);
5853                 break;
5854         case BPF_LWT_ENCAP_SEG6:
5855                 skb_reset_inner_headers(skb);
5856                 skb->encapsulation = 1;
5857                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5858                 break;
5859         default:
5860                 return -EINVAL;
5861         }
5862
5863         bpf_compute_data_pointers(skb);
5864         if (err)
5865                 return err;
5866
5867         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5868
5869         return seg6_lookup_nexthop(skb, NULL, 0);
5870 }
5871 #endif /* CONFIG_IPV6_SEG6_BPF */
5872
5873 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5874 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5875                              bool ingress)
5876 {
5877         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5878 }
5879 #endif
5880
5881 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5882            u32, len)
5883 {
5884         switch (type) {
5885 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5886         case BPF_LWT_ENCAP_SEG6:
5887         case BPF_LWT_ENCAP_SEG6_INLINE:
5888                 return bpf_push_seg6_encap(skb, type, hdr, len);
5889 #endif
5890 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5891         case BPF_LWT_ENCAP_IP:
5892                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5893 #endif
5894         default:
5895                 return -EINVAL;
5896         }
5897 }
5898
5899 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5900            void *, hdr, u32, len)
5901 {
5902         switch (type) {
5903 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5904         case BPF_LWT_ENCAP_IP:
5905                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5906 #endif
5907         default:
5908                 return -EINVAL;
5909         }
5910 }
5911
5912 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5913         .func           = bpf_lwt_in_push_encap,
5914         .gpl_only       = false,
5915         .ret_type       = RET_INTEGER,
5916         .arg1_type      = ARG_PTR_TO_CTX,
5917         .arg2_type      = ARG_ANYTHING,
5918         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5919         .arg4_type      = ARG_CONST_SIZE
5920 };
5921
5922 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5923         .func           = bpf_lwt_xmit_push_encap,
5924         .gpl_only       = false,
5925         .ret_type       = RET_INTEGER,
5926         .arg1_type      = ARG_PTR_TO_CTX,
5927         .arg2_type      = ARG_ANYTHING,
5928         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5929         .arg4_type      = ARG_CONST_SIZE
5930 };
5931
5932 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5933 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5934            const void *, from, u32, len)
5935 {
5936         struct seg6_bpf_srh_state *srh_state =
5937                 this_cpu_ptr(&seg6_bpf_srh_states);
5938         struct ipv6_sr_hdr *srh = srh_state->srh;
5939         void *srh_tlvs, *srh_end, *ptr;
5940         int srhoff = 0;
5941
5942         if (srh == NULL)
5943                 return -EINVAL;
5944
5945         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5946         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5947
5948         ptr = skb->data + offset;
5949         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5950                 srh_state->valid = false;
5951         else if (ptr < (void *)&srh->flags ||
5952                  ptr + len > (void *)&srh->segments)
5953                 return -EFAULT;
5954
5955         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5956                 return -EFAULT;
5957         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5958                 return -EINVAL;
5959         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5960
5961         memcpy(skb->data + offset, from, len);
5962         return 0;
5963 }
5964
5965 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5966         .func           = bpf_lwt_seg6_store_bytes,
5967         .gpl_only       = false,
5968         .ret_type       = RET_INTEGER,
5969         .arg1_type      = ARG_PTR_TO_CTX,
5970         .arg2_type      = ARG_ANYTHING,
5971         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5972         .arg4_type      = ARG_CONST_SIZE
5973 };
5974
5975 static void bpf_update_srh_state(struct sk_buff *skb)
5976 {
5977         struct seg6_bpf_srh_state *srh_state =
5978                 this_cpu_ptr(&seg6_bpf_srh_states);
5979         int srhoff = 0;
5980
5981         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5982                 srh_state->srh = NULL;
5983         } else {
5984                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5985                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5986                 srh_state->valid = true;
5987         }
5988 }
5989
5990 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5991            u32, action, void *, param, u32, param_len)
5992 {
5993         struct seg6_bpf_srh_state *srh_state =
5994                 this_cpu_ptr(&seg6_bpf_srh_states);
5995         int hdroff = 0;
5996         int err;
5997
5998         switch (action) {
5999         case SEG6_LOCAL_ACTION_END_X:
6000                 if (!seg6_bpf_has_valid_srh(skb))
6001                         return -EBADMSG;
6002                 if (param_len != sizeof(struct in6_addr))
6003                         return -EINVAL;
6004                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6005         case SEG6_LOCAL_ACTION_END_T:
6006                 if (!seg6_bpf_has_valid_srh(skb))
6007                         return -EBADMSG;
6008                 if (param_len != sizeof(int))
6009                         return -EINVAL;
6010                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6011         case SEG6_LOCAL_ACTION_END_DT6:
6012                 if (!seg6_bpf_has_valid_srh(skb))
6013                         return -EBADMSG;
6014                 if (param_len != sizeof(int))
6015                         return -EINVAL;
6016
6017                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6018                         return -EBADMSG;
6019                 if (!pskb_pull(skb, hdroff))
6020                         return -EBADMSG;
6021
6022                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6023                 skb_reset_network_header(skb);
6024                 skb_reset_transport_header(skb);
6025                 skb->encapsulation = 0;
6026
6027                 bpf_compute_data_pointers(skb);
6028                 bpf_update_srh_state(skb);
6029                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6030         case SEG6_LOCAL_ACTION_END_B6:
6031                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6032                         return -EBADMSG;
6033                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6034                                           param, param_len);
6035                 if (!err)
6036                         bpf_update_srh_state(skb);
6037
6038                 return err;
6039         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6040                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6041                         return -EBADMSG;
6042                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6043                                           param, param_len);
6044                 if (!err)
6045                         bpf_update_srh_state(skb);
6046
6047                 return err;
6048         default:
6049                 return -EINVAL;
6050         }
6051 }
6052
6053 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6054         .func           = bpf_lwt_seg6_action,
6055         .gpl_only       = false,
6056         .ret_type       = RET_INTEGER,
6057         .arg1_type      = ARG_PTR_TO_CTX,
6058         .arg2_type      = ARG_ANYTHING,
6059         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6060         .arg4_type      = ARG_CONST_SIZE
6061 };
6062
6063 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6064            s32, len)
6065 {
6066         struct seg6_bpf_srh_state *srh_state =
6067                 this_cpu_ptr(&seg6_bpf_srh_states);
6068         struct ipv6_sr_hdr *srh = srh_state->srh;
6069         void *srh_end, *srh_tlvs, *ptr;
6070         struct ipv6hdr *hdr;
6071         int srhoff = 0;
6072         int ret;
6073
6074         if (unlikely(srh == NULL))
6075                 return -EINVAL;
6076
6077         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6078                         ((srh->first_segment + 1) << 4));
6079         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6080                         srh_state->hdrlen);
6081         ptr = skb->data + offset;
6082
6083         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6084                 return -EFAULT;
6085         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6086                 return -EFAULT;
6087
6088         if (len > 0) {
6089                 ret = skb_cow_head(skb, len);
6090                 if (unlikely(ret < 0))
6091                         return ret;
6092
6093                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6094         } else {
6095                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6096         }
6097
6098         bpf_compute_data_pointers(skb);
6099         if (unlikely(ret < 0))
6100                 return ret;
6101
6102         hdr = (struct ipv6hdr *)skb->data;
6103         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6104
6105         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6106                 return -EINVAL;
6107         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6108         srh_state->hdrlen += len;
6109         srh_state->valid = false;
6110         return 0;
6111 }
6112
6113 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6114         .func           = bpf_lwt_seg6_adjust_srh,
6115         .gpl_only       = false,
6116         .ret_type       = RET_INTEGER,
6117         .arg1_type      = ARG_PTR_TO_CTX,
6118         .arg2_type      = ARG_ANYTHING,
6119         .arg3_type      = ARG_ANYTHING,
6120 };
6121 #endif /* CONFIG_IPV6_SEG6_BPF */
6122
6123 #ifdef CONFIG_INET
6124 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6125                               int dif, int sdif, u8 family, u8 proto)
6126 {
6127         bool refcounted = false;
6128         struct sock *sk = NULL;
6129
6130         if (family == AF_INET) {
6131                 __be32 src4 = tuple->ipv4.saddr;
6132                 __be32 dst4 = tuple->ipv4.daddr;
6133
6134                 if (proto == IPPROTO_TCP)
6135                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6136                                            src4, tuple->ipv4.sport,
6137                                            dst4, tuple->ipv4.dport,
6138                                            dif, sdif, &refcounted);
6139                 else
6140                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6141                                                dst4, tuple->ipv4.dport,
6142                                                dif, sdif, &udp_table, NULL);
6143 #if IS_ENABLED(CONFIG_IPV6)
6144         } else {
6145                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6146                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6147
6148                 if (proto == IPPROTO_TCP)
6149                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6150                                             src6, tuple->ipv6.sport,
6151                                             dst6, ntohs(tuple->ipv6.dport),
6152                                             dif, sdif, &refcounted);
6153                 else if (likely(ipv6_bpf_stub))
6154                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6155                                                             src6, tuple->ipv6.sport,
6156                                                             dst6, tuple->ipv6.dport,
6157                                                             dif, sdif,
6158                                                             &udp_table, NULL);
6159 #endif
6160         }
6161
6162         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6163                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6164                 sk = NULL;
6165         }
6166         return sk;
6167 }
6168
6169 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6170  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6171  * Returns the socket as an 'unsigned long' to simplify the casting in the
6172  * callers to satisfy BPF_CALL declarations.
6173  */
6174 static struct sock *
6175 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6176                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6177                  u64 flags)
6178 {
6179         struct sock *sk = NULL;
6180         u8 family = AF_UNSPEC;
6181         struct net *net;
6182         int sdif;
6183
6184         if (len == sizeof(tuple->ipv4))
6185                 family = AF_INET;
6186         else if (len == sizeof(tuple->ipv6))
6187                 family = AF_INET6;
6188         else
6189                 return NULL;
6190
6191         if (unlikely(family == AF_UNSPEC || flags ||
6192                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6193                 goto out;
6194
6195         if (family == AF_INET)
6196                 sdif = inet_sdif(skb);
6197         else
6198                 sdif = inet6_sdif(skb);
6199
6200         if ((s32)netns_id < 0) {
6201                 net = caller_net;
6202                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6203         } else {
6204                 net = get_net_ns_by_id(caller_net, netns_id);
6205                 if (unlikely(!net))
6206                         goto out;
6207                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6208                 put_net(net);
6209         }
6210
6211 out:
6212         return sk;
6213 }
6214
6215 static struct sock *
6216 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6217                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6218                 u64 flags)
6219 {
6220         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6221                                            ifindex, proto, netns_id, flags);
6222
6223         if (sk) {
6224                 struct sock *sk2 = sk_to_full_sk(sk);
6225
6226                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6227                  * sock refcnt is decremented to prevent a request_sock leak.
6228                  */
6229                 if (!sk_fullsock(sk2))
6230                         sk2 = NULL;
6231                 if (sk2 != sk) {
6232                         sock_gen_put(sk);
6233                         /* Ensure there is no need to bump sk2 refcnt */
6234                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6235                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6236                                 return NULL;
6237                         }
6238                         sk = sk2;
6239                 }
6240         }
6241
6242         return sk;
6243 }
6244
6245 static struct sock *
6246 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6247                u8 proto, u64 netns_id, u64 flags)
6248 {
6249         struct net *caller_net;
6250         int ifindex;
6251
6252         if (skb->dev) {
6253                 caller_net = dev_net(skb->dev);
6254                 ifindex = skb->dev->ifindex;
6255         } else {
6256                 caller_net = sock_net(skb->sk);
6257                 ifindex = 0;
6258         }
6259
6260         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6261                                 netns_id, flags);
6262 }
6263
6264 static struct sock *
6265 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6266               u8 proto, u64 netns_id, u64 flags)
6267 {
6268         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6269                                          flags);
6270
6271         if (sk) {
6272                 struct sock *sk2 = sk_to_full_sk(sk);
6273
6274                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6275                  * sock refcnt is decremented to prevent a request_sock leak.
6276                  */
6277                 if (!sk_fullsock(sk2))
6278                         sk2 = NULL;
6279                 if (sk2 != sk) {
6280                         sock_gen_put(sk);
6281                         /* Ensure there is no need to bump sk2 refcnt */
6282                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6283                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6284                                 return NULL;
6285                         }
6286                         sk = sk2;
6287                 }
6288         }
6289
6290         return sk;
6291 }
6292
6293 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6294            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6295 {
6296         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6297                                              netns_id, flags);
6298 }
6299
6300 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6301         .func           = bpf_skc_lookup_tcp,
6302         .gpl_only       = false,
6303         .pkt_access     = true,
6304         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6305         .arg1_type      = ARG_PTR_TO_CTX,
6306         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6307         .arg3_type      = ARG_CONST_SIZE,
6308         .arg4_type      = ARG_ANYTHING,
6309         .arg5_type      = ARG_ANYTHING,
6310 };
6311
6312 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6313            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6314 {
6315         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6316                                             netns_id, flags);
6317 }
6318
6319 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6320         .func           = bpf_sk_lookup_tcp,
6321         .gpl_only       = false,
6322         .pkt_access     = true,
6323         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6324         .arg1_type      = ARG_PTR_TO_CTX,
6325         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6326         .arg3_type      = ARG_CONST_SIZE,
6327         .arg4_type      = ARG_ANYTHING,
6328         .arg5_type      = ARG_ANYTHING,
6329 };
6330
6331 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6332            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6333 {
6334         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6335                                             netns_id, flags);
6336 }
6337
6338 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6339         .func           = bpf_sk_lookup_udp,
6340         .gpl_only       = false,
6341         .pkt_access     = true,
6342         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6343         .arg1_type      = ARG_PTR_TO_CTX,
6344         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6345         .arg3_type      = ARG_CONST_SIZE,
6346         .arg4_type      = ARG_ANYTHING,
6347         .arg5_type      = ARG_ANYTHING,
6348 };
6349
6350 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6351 {
6352         if (sk && sk_is_refcounted(sk))
6353                 sock_gen_put(sk);
6354         return 0;
6355 }
6356
6357 static const struct bpf_func_proto bpf_sk_release_proto = {
6358         .func           = bpf_sk_release,
6359         .gpl_only       = false,
6360         .ret_type       = RET_INTEGER,
6361         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6362 };
6363
6364 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6365            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6366 {
6367         struct net *caller_net = dev_net(ctx->rxq->dev);
6368         int ifindex = ctx->rxq->dev->ifindex;
6369
6370         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6371                                               ifindex, IPPROTO_UDP, netns_id,
6372                                               flags);
6373 }
6374
6375 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6376         .func           = bpf_xdp_sk_lookup_udp,
6377         .gpl_only       = false,
6378         .pkt_access     = true,
6379         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6380         .arg1_type      = ARG_PTR_TO_CTX,
6381         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6382         .arg3_type      = ARG_CONST_SIZE,
6383         .arg4_type      = ARG_ANYTHING,
6384         .arg5_type      = ARG_ANYTHING,
6385 };
6386
6387 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6388            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6389 {
6390         struct net *caller_net = dev_net(ctx->rxq->dev);
6391         int ifindex = ctx->rxq->dev->ifindex;
6392
6393         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6394                                                ifindex, IPPROTO_TCP, netns_id,
6395                                                flags);
6396 }
6397
6398 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6399         .func           = bpf_xdp_skc_lookup_tcp,
6400         .gpl_only       = false,
6401         .pkt_access     = true,
6402         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6403         .arg1_type      = ARG_PTR_TO_CTX,
6404         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6405         .arg3_type      = ARG_CONST_SIZE,
6406         .arg4_type      = ARG_ANYTHING,
6407         .arg5_type      = ARG_ANYTHING,
6408 };
6409
6410 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6411            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6412 {
6413         struct net *caller_net = dev_net(ctx->rxq->dev);
6414         int ifindex = ctx->rxq->dev->ifindex;
6415
6416         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6417                                               ifindex, IPPROTO_TCP, netns_id,
6418                                               flags);
6419 }
6420
6421 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6422         .func           = bpf_xdp_sk_lookup_tcp,
6423         .gpl_only       = false,
6424         .pkt_access     = true,
6425         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6426         .arg1_type      = ARG_PTR_TO_CTX,
6427         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6428         .arg3_type      = ARG_CONST_SIZE,
6429         .arg4_type      = ARG_ANYTHING,
6430         .arg5_type      = ARG_ANYTHING,
6431 };
6432
6433 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6434            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6435 {
6436         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6437                                                sock_net(ctx->sk), 0,
6438                                                IPPROTO_TCP, netns_id, flags);
6439 }
6440
6441 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6442         .func           = bpf_sock_addr_skc_lookup_tcp,
6443         .gpl_only       = false,
6444         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6445         .arg1_type      = ARG_PTR_TO_CTX,
6446         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6447         .arg3_type      = ARG_CONST_SIZE,
6448         .arg4_type      = ARG_ANYTHING,
6449         .arg5_type      = ARG_ANYTHING,
6450 };
6451
6452 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6453            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6454 {
6455         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6456                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6457                                               netns_id, flags);
6458 }
6459
6460 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6461         .func           = bpf_sock_addr_sk_lookup_tcp,
6462         .gpl_only       = false,
6463         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6464         .arg1_type      = ARG_PTR_TO_CTX,
6465         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6466         .arg3_type      = ARG_CONST_SIZE,
6467         .arg4_type      = ARG_ANYTHING,
6468         .arg5_type      = ARG_ANYTHING,
6469 };
6470
6471 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6472            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6473 {
6474         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6475                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6476                                               netns_id, flags);
6477 }
6478
6479 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6480         .func           = bpf_sock_addr_sk_lookup_udp,
6481         .gpl_only       = false,
6482         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6483         .arg1_type      = ARG_PTR_TO_CTX,
6484         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6485         .arg3_type      = ARG_CONST_SIZE,
6486         .arg4_type      = ARG_ANYTHING,
6487         .arg5_type      = ARG_ANYTHING,
6488 };
6489
6490 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6491                                   struct bpf_insn_access_aux *info)
6492 {
6493         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6494                                           icsk_retransmits))
6495                 return false;
6496
6497         if (off % size != 0)
6498                 return false;
6499
6500         switch (off) {
6501         case offsetof(struct bpf_tcp_sock, bytes_received):
6502         case offsetof(struct bpf_tcp_sock, bytes_acked):
6503                 return size == sizeof(__u64);
6504         default:
6505                 return size == sizeof(__u32);
6506         }
6507 }
6508
6509 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6510                                     const struct bpf_insn *si,
6511                                     struct bpf_insn *insn_buf,
6512                                     struct bpf_prog *prog, u32 *target_size)
6513 {
6514         struct bpf_insn *insn = insn_buf;
6515
6516 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6517         do {                                                            \
6518                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6519                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6520                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6521                                       si->dst_reg, si->src_reg,         \
6522                                       offsetof(struct tcp_sock, FIELD)); \
6523         } while (0)
6524
6525 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6526         do {                                                            \
6527                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6528                                           FIELD) >                      \
6529                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6530                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6531                                         struct inet_connection_sock,    \
6532                                         FIELD),                         \
6533                                       si->dst_reg, si->src_reg,         \
6534                                       offsetof(                         \
6535                                         struct inet_connection_sock,    \
6536                                         FIELD));                        \
6537         } while (0)
6538
6539         if (insn > insn_buf)
6540                 return insn - insn_buf;
6541
6542         switch (si->off) {
6543         case offsetof(struct bpf_tcp_sock, rtt_min):
6544                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6545                              sizeof(struct minmax));
6546                 BUILD_BUG_ON(sizeof(struct minmax) <
6547                              sizeof(struct minmax_sample));
6548
6549                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6550                                       offsetof(struct tcp_sock, rtt_min) +
6551                                       offsetof(struct minmax_sample, v));
6552                 break;
6553         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6554                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6555                 break;
6556         case offsetof(struct bpf_tcp_sock, srtt_us):
6557                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6558                 break;
6559         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6560                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6561                 break;
6562         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6563                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6564                 break;
6565         case offsetof(struct bpf_tcp_sock, snd_nxt):
6566                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6567                 break;
6568         case offsetof(struct bpf_tcp_sock, snd_una):
6569                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6570                 break;
6571         case offsetof(struct bpf_tcp_sock, mss_cache):
6572                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6573                 break;
6574         case offsetof(struct bpf_tcp_sock, ecn_flags):
6575                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6576                 break;
6577         case offsetof(struct bpf_tcp_sock, rate_delivered):
6578                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6579                 break;
6580         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6581                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6582                 break;
6583         case offsetof(struct bpf_tcp_sock, packets_out):
6584                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6585                 break;
6586         case offsetof(struct bpf_tcp_sock, retrans_out):
6587                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6588                 break;
6589         case offsetof(struct bpf_tcp_sock, total_retrans):
6590                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6591                 break;
6592         case offsetof(struct bpf_tcp_sock, segs_in):
6593                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6594                 break;
6595         case offsetof(struct bpf_tcp_sock, data_segs_in):
6596                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6597                 break;
6598         case offsetof(struct bpf_tcp_sock, segs_out):
6599                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6600                 break;
6601         case offsetof(struct bpf_tcp_sock, data_segs_out):
6602                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6603                 break;
6604         case offsetof(struct bpf_tcp_sock, lost_out):
6605                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6606                 break;
6607         case offsetof(struct bpf_tcp_sock, sacked_out):
6608                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6609                 break;
6610         case offsetof(struct bpf_tcp_sock, bytes_received):
6611                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6612                 break;
6613         case offsetof(struct bpf_tcp_sock, bytes_acked):
6614                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6615                 break;
6616         case offsetof(struct bpf_tcp_sock, dsack_dups):
6617                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6618                 break;
6619         case offsetof(struct bpf_tcp_sock, delivered):
6620                 BPF_TCP_SOCK_GET_COMMON(delivered);
6621                 break;
6622         case offsetof(struct bpf_tcp_sock, delivered_ce):
6623                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6624                 break;
6625         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6626                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6627                 break;
6628         }
6629
6630         return insn - insn_buf;
6631 }
6632
6633 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6634 {
6635         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6636                 return (unsigned long)sk;
6637
6638         return (unsigned long)NULL;
6639 }
6640
6641 const struct bpf_func_proto bpf_tcp_sock_proto = {
6642         .func           = bpf_tcp_sock,
6643         .gpl_only       = false,
6644         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6645         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6646 };
6647
6648 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6649 {
6650         sk = sk_to_full_sk(sk);
6651
6652         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6653                 return (unsigned long)sk;
6654
6655         return (unsigned long)NULL;
6656 }
6657
6658 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6659         .func           = bpf_get_listener_sock,
6660         .gpl_only       = false,
6661         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6662         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6663 };
6664
6665 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6666 {
6667         unsigned int iphdr_len;
6668
6669         switch (skb_protocol(skb, true)) {
6670         case cpu_to_be16(ETH_P_IP):
6671                 iphdr_len = sizeof(struct iphdr);
6672                 break;
6673         case cpu_to_be16(ETH_P_IPV6):
6674                 iphdr_len = sizeof(struct ipv6hdr);
6675                 break;
6676         default:
6677                 return 0;
6678         }
6679
6680         if (skb_headlen(skb) < iphdr_len)
6681                 return 0;
6682
6683         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6684                 return 0;
6685
6686         return INET_ECN_set_ce(skb);
6687 }
6688
6689 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6690                                   struct bpf_insn_access_aux *info)
6691 {
6692         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6693                 return false;
6694
6695         if (off % size != 0)
6696                 return false;
6697
6698         switch (off) {
6699         default:
6700                 return size == sizeof(__u32);
6701         }
6702 }
6703
6704 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6705                                     const struct bpf_insn *si,
6706                                     struct bpf_insn *insn_buf,
6707                                     struct bpf_prog *prog, u32 *target_size)
6708 {
6709         struct bpf_insn *insn = insn_buf;
6710
6711 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6712         do {                                                            \
6713                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6714                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6715                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6716                                       si->dst_reg, si->src_reg,         \
6717                                       offsetof(struct xdp_sock, FIELD)); \
6718         } while (0)
6719
6720         switch (si->off) {
6721         case offsetof(struct bpf_xdp_sock, queue_id):
6722                 BPF_XDP_SOCK_GET(queue_id);
6723                 break;
6724         }
6725
6726         return insn - insn_buf;
6727 }
6728
6729 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6730         .func           = bpf_skb_ecn_set_ce,
6731         .gpl_only       = false,
6732         .ret_type       = RET_INTEGER,
6733         .arg1_type      = ARG_PTR_TO_CTX,
6734 };
6735
6736 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6737            struct tcphdr *, th, u32, th_len)
6738 {
6739 #ifdef CONFIG_SYN_COOKIES
6740         u32 cookie;
6741         int ret;
6742
6743         if (unlikely(!sk || th_len < sizeof(*th)))
6744                 return -EINVAL;
6745
6746         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6747         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6748                 return -EINVAL;
6749
6750         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6751                 return -EINVAL;
6752
6753         if (!th->ack || th->rst || th->syn)
6754                 return -ENOENT;
6755
6756         if (unlikely(iph_len < sizeof(struct iphdr)))
6757                 return -EINVAL;
6758
6759         if (tcp_synq_no_recent_overflow(sk))
6760                 return -ENOENT;
6761
6762         cookie = ntohl(th->ack_seq) - 1;
6763
6764         /* Both struct iphdr and struct ipv6hdr have the version field at the
6765          * same offset so we can cast to the shorter header (struct iphdr).
6766          */
6767         switch (((struct iphdr *)iph)->version) {
6768         case 4:
6769                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6770                         return -EINVAL;
6771
6772                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6773                 break;
6774
6775 #if IS_BUILTIN(CONFIG_IPV6)
6776         case 6:
6777                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6778                         return -EINVAL;
6779
6780                 if (sk->sk_family != AF_INET6)
6781                         return -EINVAL;
6782
6783                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6784                 break;
6785 #endif /* CONFIG_IPV6 */
6786
6787         default:
6788                 return -EPROTONOSUPPORT;
6789         }
6790
6791         if (ret > 0)
6792                 return 0;
6793
6794         return -ENOENT;
6795 #else
6796         return -ENOTSUPP;
6797 #endif
6798 }
6799
6800 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6801         .func           = bpf_tcp_check_syncookie,
6802         .gpl_only       = true,
6803         .pkt_access     = true,
6804         .ret_type       = RET_INTEGER,
6805         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6806         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6807         .arg3_type      = ARG_CONST_SIZE,
6808         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6809         .arg5_type      = ARG_CONST_SIZE,
6810 };
6811
6812 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6813            struct tcphdr *, th, u32, th_len)
6814 {
6815 #ifdef CONFIG_SYN_COOKIES
6816         u32 cookie;
6817         u16 mss;
6818
6819         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6820                 return -EINVAL;
6821
6822         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6823                 return -EINVAL;
6824
6825         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6826                 return -ENOENT;
6827
6828         if (!th->syn || th->ack || th->fin || th->rst)
6829                 return -EINVAL;
6830
6831         if (unlikely(iph_len < sizeof(struct iphdr)))
6832                 return -EINVAL;
6833
6834         /* Both struct iphdr and struct ipv6hdr have the version field at the
6835          * same offset so we can cast to the shorter header (struct iphdr).
6836          */
6837         switch (((struct iphdr *)iph)->version) {
6838         case 4:
6839                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6840                         return -EINVAL;
6841
6842                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6843                 break;
6844
6845 #if IS_BUILTIN(CONFIG_IPV6)
6846         case 6:
6847                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6848                         return -EINVAL;
6849
6850                 if (sk->sk_family != AF_INET6)
6851                         return -EINVAL;
6852
6853                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6854                 break;
6855 #endif /* CONFIG_IPV6 */
6856
6857         default:
6858                 return -EPROTONOSUPPORT;
6859         }
6860         if (mss == 0)
6861                 return -ENOENT;
6862
6863         return cookie | ((u64)mss << 32);
6864 #else
6865         return -EOPNOTSUPP;
6866 #endif /* CONFIG_SYN_COOKIES */
6867 }
6868
6869 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6870         .func           = bpf_tcp_gen_syncookie,
6871         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
6872         .pkt_access     = true,
6873         .ret_type       = RET_INTEGER,
6874         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6875         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6876         .arg3_type      = ARG_CONST_SIZE,
6877         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6878         .arg5_type      = ARG_CONST_SIZE,
6879 };
6880
6881 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6882 {
6883         if (!sk || flags != 0)
6884                 return -EINVAL;
6885         if (!skb_at_tc_ingress(skb))
6886                 return -EOPNOTSUPP;
6887         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6888                 return -ENETUNREACH;
6889         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6890                 return -ESOCKTNOSUPPORT;
6891         if (sk_is_refcounted(sk) &&
6892             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6893                 return -ENOENT;
6894
6895         skb_orphan(skb);
6896         skb->sk = sk;
6897         skb->destructor = sock_pfree;
6898
6899         return 0;
6900 }
6901
6902 static const struct bpf_func_proto bpf_sk_assign_proto = {
6903         .func           = bpf_sk_assign,
6904         .gpl_only       = false,
6905         .ret_type       = RET_INTEGER,
6906         .arg1_type      = ARG_PTR_TO_CTX,
6907         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6908         .arg3_type      = ARG_ANYTHING,
6909 };
6910
6911 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6912                                     u8 search_kind, const u8 *magic,
6913                                     u8 magic_len, bool *eol)
6914 {
6915         u8 kind, kind_len;
6916
6917         *eol = false;
6918
6919         while (op < opend) {
6920                 kind = op[0];
6921
6922                 if (kind == TCPOPT_EOL) {
6923                         *eol = true;
6924                         return ERR_PTR(-ENOMSG);
6925                 } else if (kind == TCPOPT_NOP) {
6926                         op++;
6927                         continue;
6928                 }
6929
6930                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6931                         /* Something is wrong in the received header.
6932                          * Follow the TCP stack's tcp_parse_options()
6933                          * and just bail here.
6934                          */
6935                         return ERR_PTR(-EFAULT);
6936
6937                 kind_len = op[1];
6938                 if (search_kind == kind) {
6939                         if (!magic_len)
6940                                 return op;
6941
6942                         if (magic_len > kind_len - 2)
6943                                 return ERR_PTR(-ENOMSG);
6944
6945                         if (!memcmp(&op[2], magic, magic_len))
6946                                 return op;
6947                 }
6948
6949                 op += kind_len;
6950         }
6951
6952         return ERR_PTR(-ENOMSG);
6953 }
6954
6955 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6956            void *, search_res, u32, len, u64, flags)
6957 {
6958         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6959         const u8 *op, *opend, *magic, *search = search_res;
6960         u8 search_kind, search_len, copy_len, magic_len;
6961         int ret;
6962
6963         /* 2 byte is the minimal option len except TCPOPT_NOP and
6964          * TCPOPT_EOL which are useless for the bpf prog to learn
6965          * and this helper disallow loading them also.
6966          */
6967         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6968                 return -EINVAL;
6969
6970         search_kind = search[0];
6971         search_len = search[1];
6972
6973         if (search_len > len || search_kind == TCPOPT_NOP ||
6974             search_kind == TCPOPT_EOL)
6975                 return -EINVAL;
6976
6977         if (search_kind == TCPOPT_EXP || search_kind == 253) {
6978                 /* 16 or 32 bit magic.  +2 for kind and kind length */
6979                 if (search_len != 4 && search_len != 6)
6980                         return -EINVAL;
6981                 magic = &search[2];
6982                 magic_len = search_len - 2;
6983         } else {
6984                 if (search_len)
6985                         return -EINVAL;
6986                 magic = NULL;
6987                 magic_len = 0;
6988         }
6989
6990         if (load_syn) {
6991                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6992                 if (ret < 0)
6993                         return ret;
6994
6995                 opend = op + ret;
6996                 op += sizeof(struct tcphdr);
6997         } else {
6998                 if (!bpf_sock->skb ||
6999                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7000                         /* This bpf_sock->op cannot call this helper */
7001                         return -EPERM;
7002
7003                 opend = bpf_sock->skb_data_end;
7004                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7005         }
7006
7007         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7008                                 &eol);
7009         if (IS_ERR(op))
7010                 return PTR_ERR(op);
7011
7012         copy_len = op[1];
7013         ret = copy_len;
7014         if (copy_len > len) {
7015                 ret = -ENOSPC;
7016                 copy_len = len;
7017         }
7018
7019         memcpy(search_res, op, copy_len);
7020         return ret;
7021 }
7022
7023 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7024         .func           = bpf_sock_ops_load_hdr_opt,
7025         .gpl_only       = false,
7026         .ret_type       = RET_INTEGER,
7027         .arg1_type      = ARG_PTR_TO_CTX,
7028         .arg2_type      = ARG_PTR_TO_MEM,
7029         .arg3_type      = ARG_CONST_SIZE,
7030         .arg4_type      = ARG_ANYTHING,
7031 };
7032
7033 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7034            const void *, from, u32, len, u64, flags)
7035 {
7036         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7037         const u8 *op, *new_op, *magic = NULL;
7038         struct sk_buff *skb;
7039         bool eol;
7040
7041         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7042                 return -EPERM;
7043
7044         if (len < 2 || flags)
7045                 return -EINVAL;
7046
7047         new_op = from;
7048         new_kind = new_op[0];
7049         new_kind_len = new_op[1];
7050
7051         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7052             new_kind == TCPOPT_EOL)
7053                 return -EINVAL;
7054
7055         if (new_kind_len > bpf_sock->remaining_opt_len)
7056                 return -ENOSPC;
7057
7058         /* 253 is another experimental kind */
7059         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7060                 if (new_kind_len < 4)
7061                         return -EINVAL;
7062                 /* Match for the 2 byte magic also.
7063                  * RFC 6994: the magic could be 2 or 4 bytes.
7064                  * Hence, matching by 2 byte only is on the
7065                  * conservative side but it is the right
7066                  * thing to do for the 'search-for-duplication'
7067                  * purpose.
7068                  */
7069                 magic = &new_op[2];
7070                 magic_len = 2;
7071         }
7072
7073         /* Check for duplication */
7074         skb = bpf_sock->skb;
7075         op = skb->data + sizeof(struct tcphdr);
7076         opend = bpf_sock->skb_data_end;
7077
7078         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7079                                 &eol);
7080         if (!IS_ERR(op))
7081                 return -EEXIST;
7082
7083         if (PTR_ERR(op) != -ENOMSG)
7084                 return PTR_ERR(op);
7085
7086         if (eol)
7087                 /* The option has been ended.  Treat it as no more
7088                  * header option can be written.
7089                  */
7090                 return -ENOSPC;
7091
7092         /* No duplication found.  Store the header option. */
7093         memcpy(opend, from, new_kind_len);
7094
7095         bpf_sock->remaining_opt_len -= new_kind_len;
7096         bpf_sock->skb_data_end += new_kind_len;
7097
7098         return 0;
7099 }
7100
7101 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7102         .func           = bpf_sock_ops_store_hdr_opt,
7103         .gpl_only       = false,
7104         .ret_type       = RET_INTEGER,
7105         .arg1_type      = ARG_PTR_TO_CTX,
7106         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7107         .arg3_type      = ARG_CONST_SIZE,
7108         .arg4_type      = ARG_ANYTHING,
7109 };
7110
7111 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7112            u32, len, u64, flags)
7113 {
7114         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7115                 return -EPERM;
7116
7117         if (flags || len < 2)
7118                 return -EINVAL;
7119
7120         if (len > bpf_sock->remaining_opt_len)
7121                 return -ENOSPC;
7122
7123         bpf_sock->remaining_opt_len -= len;
7124
7125         return 0;
7126 }
7127
7128 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7129         .func           = bpf_sock_ops_reserve_hdr_opt,
7130         .gpl_only       = false,
7131         .ret_type       = RET_INTEGER,
7132         .arg1_type      = ARG_PTR_TO_CTX,
7133         .arg2_type      = ARG_ANYTHING,
7134         .arg3_type      = ARG_ANYTHING,
7135 };
7136
7137 #endif /* CONFIG_INET */
7138
7139 bool bpf_helper_changes_pkt_data(void *func)
7140 {
7141         if (func == bpf_skb_vlan_push ||
7142             func == bpf_skb_vlan_pop ||
7143             func == bpf_skb_store_bytes ||
7144             func == bpf_skb_change_proto ||
7145             func == bpf_skb_change_head ||
7146             func == sk_skb_change_head ||
7147             func == bpf_skb_change_tail ||
7148             func == sk_skb_change_tail ||
7149             func == bpf_skb_adjust_room ||
7150             func == sk_skb_adjust_room ||
7151             func == bpf_skb_pull_data ||
7152             func == sk_skb_pull_data ||
7153             func == bpf_clone_redirect ||
7154             func == bpf_l3_csum_replace ||
7155             func == bpf_l4_csum_replace ||
7156             func == bpf_xdp_adjust_head ||
7157             func == bpf_xdp_adjust_meta ||
7158             func == bpf_msg_pull_data ||
7159             func == bpf_msg_push_data ||
7160             func == bpf_msg_pop_data ||
7161             func == bpf_xdp_adjust_tail ||
7162 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7163             func == bpf_lwt_seg6_store_bytes ||
7164             func == bpf_lwt_seg6_adjust_srh ||
7165             func == bpf_lwt_seg6_action ||
7166 #endif
7167 #ifdef CONFIG_INET
7168             func == bpf_sock_ops_store_hdr_opt ||
7169 #endif
7170             func == bpf_lwt_in_push_encap ||
7171             func == bpf_lwt_xmit_push_encap)
7172                 return true;
7173
7174         return false;
7175 }
7176
7177 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7178 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7179
7180 static const struct bpf_func_proto *
7181 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7182 {
7183         switch (func_id) {
7184         /* inet and inet6 sockets are created in a process
7185          * context so there is always a valid uid/gid
7186          */
7187         case BPF_FUNC_get_current_uid_gid:
7188                 return &bpf_get_current_uid_gid_proto;
7189         case BPF_FUNC_get_local_storage:
7190                 return &bpf_get_local_storage_proto;
7191         case BPF_FUNC_get_socket_cookie:
7192                 return &bpf_get_socket_cookie_sock_proto;
7193         case BPF_FUNC_get_netns_cookie:
7194                 return &bpf_get_netns_cookie_sock_proto;
7195         case BPF_FUNC_perf_event_output:
7196                 return &bpf_event_output_data_proto;
7197         case BPF_FUNC_get_current_pid_tgid:
7198                 return &bpf_get_current_pid_tgid_proto;
7199         case BPF_FUNC_get_current_comm:
7200                 return &bpf_get_current_comm_proto;
7201 #ifdef CONFIG_CGROUPS
7202         case BPF_FUNC_get_current_cgroup_id:
7203                 return &bpf_get_current_cgroup_id_proto;
7204         case BPF_FUNC_get_current_ancestor_cgroup_id:
7205                 return &bpf_get_current_ancestor_cgroup_id_proto;
7206 #endif
7207 #ifdef CONFIG_CGROUP_NET_CLASSID
7208         case BPF_FUNC_get_cgroup_classid:
7209                 return &bpf_get_cgroup_classid_curr_proto;
7210 #endif
7211         case BPF_FUNC_sk_storage_get:
7212                 return &bpf_sk_storage_get_cg_sock_proto;
7213         case BPF_FUNC_ktime_get_coarse_ns:
7214                 return &bpf_ktime_get_coarse_ns_proto;
7215         default:
7216                 return bpf_base_func_proto(func_id);
7217         }
7218 }
7219
7220 static const struct bpf_func_proto *
7221 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7222 {
7223         switch (func_id) {
7224         /* inet and inet6 sockets are created in a process
7225          * context so there is always a valid uid/gid
7226          */
7227         case BPF_FUNC_get_current_uid_gid:
7228                 return &bpf_get_current_uid_gid_proto;
7229         case BPF_FUNC_bind:
7230                 switch (prog->expected_attach_type) {
7231                 case BPF_CGROUP_INET4_CONNECT:
7232                 case BPF_CGROUP_INET6_CONNECT:
7233                         return &bpf_bind_proto;
7234                 default:
7235                         return NULL;
7236                 }
7237         case BPF_FUNC_get_socket_cookie:
7238                 return &bpf_get_socket_cookie_sock_addr_proto;
7239         case BPF_FUNC_get_netns_cookie:
7240                 return &bpf_get_netns_cookie_sock_addr_proto;
7241         case BPF_FUNC_get_local_storage:
7242                 return &bpf_get_local_storage_proto;
7243         case BPF_FUNC_perf_event_output:
7244                 return &bpf_event_output_data_proto;
7245         case BPF_FUNC_get_current_pid_tgid:
7246                 return &bpf_get_current_pid_tgid_proto;
7247         case BPF_FUNC_get_current_comm:
7248                 return &bpf_get_current_comm_proto;
7249 #ifdef CONFIG_CGROUPS
7250         case BPF_FUNC_get_current_cgroup_id:
7251                 return &bpf_get_current_cgroup_id_proto;
7252         case BPF_FUNC_get_current_ancestor_cgroup_id:
7253                 return &bpf_get_current_ancestor_cgroup_id_proto;
7254 #endif
7255 #ifdef CONFIG_CGROUP_NET_CLASSID
7256         case BPF_FUNC_get_cgroup_classid:
7257                 return &bpf_get_cgroup_classid_curr_proto;
7258 #endif
7259 #ifdef CONFIG_INET
7260         case BPF_FUNC_sk_lookup_tcp:
7261                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7262         case BPF_FUNC_sk_lookup_udp:
7263                 return &bpf_sock_addr_sk_lookup_udp_proto;
7264         case BPF_FUNC_sk_release:
7265                 return &bpf_sk_release_proto;
7266         case BPF_FUNC_skc_lookup_tcp:
7267                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7268 #endif /* CONFIG_INET */
7269         case BPF_FUNC_sk_storage_get:
7270                 return &bpf_sk_storage_get_proto;
7271         case BPF_FUNC_sk_storage_delete:
7272                 return &bpf_sk_storage_delete_proto;
7273         case BPF_FUNC_setsockopt:
7274                 switch (prog->expected_attach_type) {
7275                 case BPF_CGROUP_INET4_BIND:
7276                 case BPF_CGROUP_INET6_BIND:
7277                 case BPF_CGROUP_INET4_CONNECT:
7278                 case BPF_CGROUP_INET6_CONNECT:
7279                 case BPF_CGROUP_UDP4_RECVMSG:
7280                 case BPF_CGROUP_UDP6_RECVMSG:
7281                 case BPF_CGROUP_UDP4_SENDMSG:
7282                 case BPF_CGROUP_UDP6_SENDMSG:
7283                 case BPF_CGROUP_INET4_GETPEERNAME:
7284                 case BPF_CGROUP_INET6_GETPEERNAME:
7285                 case BPF_CGROUP_INET4_GETSOCKNAME:
7286                 case BPF_CGROUP_INET6_GETSOCKNAME:
7287                         return &bpf_sock_addr_setsockopt_proto;
7288                 default:
7289                         return NULL;
7290                 }
7291         case BPF_FUNC_getsockopt:
7292                 switch (prog->expected_attach_type) {
7293                 case BPF_CGROUP_INET4_BIND:
7294                 case BPF_CGROUP_INET6_BIND:
7295                 case BPF_CGROUP_INET4_CONNECT:
7296                 case BPF_CGROUP_INET6_CONNECT:
7297                 case BPF_CGROUP_UDP4_RECVMSG:
7298                 case BPF_CGROUP_UDP6_RECVMSG:
7299                 case BPF_CGROUP_UDP4_SENDMSG:
7300                 case BPF_CGROUP_UDP6_SENDMSG:
7301                 case BPF_CGROUP_INET4_GETPEERNAME:
7302                 case BPF_CGROUP_INET6_GETPEERNAME:
7303                 case BPF_CGROUP_INET4_GETSOCKNAME:
7304                 case BPF_CGROUP_INET6_GETSOCKNAME:
7305                         return &bpf_sock_addr_getsockopt_proto;
7306                 default:
7307                         return NULL;
7308                 }
7309         default:
7310                 return bpf_sk_base_func_proto(func_id);
7311         }
7312 }
7313
7314 static const struct bpf_func_proto *
7315 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7316 {
7317         switch (func_id) {
7318         case BPF_FUNC_skb_load_bytes:
7319                 return &bpf_skb_load_bytes_proto;
7320         case BPF_FUNC_skb_load_bytes_relative:
7321                 return &bpf_skb_load_bytes_relative_proto;
7322         case BPF_FUNC_get_socket_cookie:
7323                 return &bpf_get_socket_cookie_proto;
7324         case BPF_FUNC_get_socket_uid:
7325                 return &bpf_get_socket_uid_proto;
7326         case BPF_FUNC_perf_event_output:
7327                 return &bpf_skb_event_output_proto;
7328         default:
7329                 return bpf_sk_base_func_proto(func_id);
7330         }
7331 }
7332
7333 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7334 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7335
7336 static const struct bpf_func_proto *
7337 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7338 {
7339         switch (func_id) {
7340         case BPF_FUNC_get_local_storage:
7341                 return &bpf_get_local_storage_proto;
7342         case BPF_FUNC_sk_fullsock:
7343                 return &bpf_sk_fullsock_proto;
7344         case BPF_FUNC_sk_storage_get:
7345                 return &bpf_sk_storage_get_proto;
7346         case BPF_FUNC_sk_storage_delete:
7347                 return &bpf_sk_storage_delete_proto;
7348         case BPF_FUNC_perf_event_output:
7349                 return &bpf_skb_event_output_proto;
7350 #ifdef CONFIG_SOCK_CGROUP_DATA
7351         case BPF_FUNC_skb_cgroup_id:
7352                 return &bpf_skb_cgroup_id_proto;
7353         case BPF_FUNC_skb_ancestor_cgroup_id:
7354                 return &bpf_skb_ancestor_cgroup_id_proto;
7355         case BPF_FUNC_sk_cgroup_id:
7356                 return &bpf_sk_cgroup_id_proto;
7357         case BPF_FUNC_sk_ancestor_cgroup_id:
7358                 return &bpf_sk_ancestor_cgroup_id_proto;
7359 #endif
7360 #ifdef CONFIG_INET
7361         case BPF_FUNC_sk_lookup_tcp:
7362                 return &bpf_sk_lookup_tcp_proto;
7363         case BPF_FUNC_sk_lookup_udp:
7364                 return &bpf_sk_lookup_udp_proto;
7365         case BPF_FUNC_sk_release:
7366                 return &bpf_sk_release_proto;
7367         case BPF_FUNC_skc_lookup_tcp:
7368                 return &bpf_skc_lookup_tcp_proto;
7369         case BPF_FUNC_tcp_sock:
7370                 return &bpf_tcp_sock_proto;
7371         case BPF_FUNC_get_listener_sock:
7372                 return &bpf_get_listener_sock_proto;
7373         case BPF_FUNC_skb_ecn_set_ce:
7374                 return &bpf_skb_ecn_set_ce_proto;
7375 #endif
7376         default:
7377                 return sk_filter_func_proto(func_id, prog);
7378         }
7379 }
7380
7381 static const struct bpf_func_proto *
7382 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7383 {
7384         switch (func_id) {
7385         case BPF_FUNC_skb_store_bytes:
7386                 return &bpf_skb_store_bytes_proto;
7387         case BPF_FUNC_skb_load_bytes:
7388                 return &bpf_skb_load_bytes_proto;
7389         case BPF_FUNC_skb_load_bytes_relative:
7390                 return &bpf_skb_load_bytes_relative_proto;
7391         case BPF_FUNC_skb_pull_data:
7392                 return &bpf_skb_pull_data_proto;
7393         case BPF_FUNC_csum_diff:
7394                 return &bpf_csum_diff_proto;
7395         case BPF_FUNC_csum_update:
7396                 return &bpf_csum_update_proto;
7397         case BPF_FUNC_csum_level:
7398                 return &bpf_csum_level_proto;
7399         case BPF_FUNC_l3_csum_replace:
7400                 return &bpf_l3_csum_replace_proto;
7401         case BPF_FUNC_l4_csum_replace:
7402                 return &bpf_l4_csum_replace_proto;
7403         case BPF_FUNC_clone_redirect:
7404                 return &bpf_clone_redirect_proto;
7405         case BPF_FUNC_get_cgroup_classid:
7406                 return &bpf_get_cgroup_classid_proto;
7407         case BPF_FUNC_skb_vlan_push:
7408                 return &bpf_skb_vlan_push_proto;
7409         case BPF_FUNC_skb_vlan_pop:
7410                 return &bpf_skb_vlan_pop_proto;
7411         case BPF_FUNC_skb_change_proto:
7412                 return &bpf_skb_change_proto_proto;
7413         case BPF_FUNC_skb_change_type:
7414                 return &bpf_skb_change_type_proto;
7415         case BPF_FUNC_skb_adjust_room:
7416                 return &bpf_skb_adjust_room_proto;
7417         case BPF_FUNC_skb_change_tail:
7418                 return &bpf_skb_change_tail_proto;
7419         case BPF_FUNC_skb_change_head:
7420                 return &bpf_skb_change_head_proto;
7421         case BPF_FUNC_skb_get_tunnel_key:
7422                 return &bpf_skb_get_tunnel_key_proto;
7423         case BPF_FUNC_skb_set_tunnel_key:
7424                 return bpf_get_skb_set_tunnel_proto(func_id);
7425         case BPF_FUNC_skb_get_tunnel_opt:
7426                 return &bpf_skb_get_tunnel_opt_proto;
7427         case BPF_FUNC_skb_set_tunnel_opt:
7428                 return bpf_get_skb_set_tunnel_proto(func_id);
7429         case BPF_FUNC_redirect:
7430                 return &bpf_redirect_proto;
7431         case BPF_FUNC_redirect_neigh:
7432                 return &bpf_redirect_neigh_proto;
7433         case BPF_FUNC_redirect_peer:
7434                 return &bpf_redirect_peer_proto;
7435         case BPF_FUNC_get_route_realm:
7436                 return &bpf_get_route_realm_proto;
7437         case BPF_FUNC_get_hash_recalc:
7438                 return &bpf_get_hash_recalc_proto;
7439         case BPF_FUNC_set_hash_invalid:
7440                 return &bpf_set_hash_invalid_proto;
7441         case BPF_FUNC_set_hash:
7442                 return &bpf_set_hash_proto;
7443         case BPF_FUNC_perf_event_output:
7444                 return &bpf_skb_event_output_proto;
7445         case BPF_FUNC_get_smp_processor_id:
7446                 return &bpf_get_smp_processor_id_proto;
7447         case BPF_FUNC_skb_under_cgroup:
7448                 return &bpf_skb_under_cgroup_proto;
7449         case BPF_FUNC_get_socket_cookie:
7450                 return &bpf_get_socket_cookie_proto;
7451         case BPF_FUNC_get_socket_uid:
7452                 return &bpf_get_socket_uid_proto;
7453         case BPF_FUNC_fib_lookup:
7454                 return &bpf_skb_fib_lookup_proto;
7455         case BPF_FUNC_check_mtu:
7456                 return &bpf_skb_check_mtu_proto;
7457         case BPF_FUNC_sk_fullsock:
7458                 return &bpf_sk_fullsock_proto;
7459         case BPF_FUNC_sk_storage_get:
7460                 return &bpf_sk_storage_get_proto;
7461         case BPF_FUNC_sk_storage_delete:
7462                 return &bpf_sk_storage_delete_proto;
7463 #ifdef CONFIG_XFRM
7464         case BPF_FUNC_skb_get_xfrm_state:
7465                 return &bpf_skb_get_xfrm_state_proto;
7466 #endif
7467 #ifdef CONFIG_CGROUP_NET_CLASSID
7468         case BPF_FUNC_skb_cgroup_classid:
7469                 return &bpf_skb_cgroup_classid_proto;
7470 #endif
7471 #ifdef CONFIG_SOCK_CGROUP_DATA
7472         case BPF_FUNC_skb_cgroup_id:
7473                 return &bpf_skb_cgroup_id_proto;
7474         case BPF_FUNC_skb_ancestor_cgroup_id:
7475                 return &bpf_skb_ancestor_cgroup_id_proto;
7476 #endif
7477 #ifdef CONFIG_INET
7478         case BPF_FUNC_sk_lookup_tcp:
7479                 return &bpf_sk_lookup_tcp_proto;
7480         case BPF_FUNC_sk_lookup_udp:
7481                 return &bpf_sk_lookup_udp_proto;
7482         case BPF_FUNC_sk_release:
7483                 return &bpf_sk_release_proto;
7484         case BPF_FUNC_tcp_sock:
7485                 return &bpf_tcp_sock_proto;
7486         case BPF_FUNC_get_listener_sock:
7487                 return &bpf_get_listener_sock_proto;
7488         case BPF_FUNC_skc_lookup_tcp:
7489                 return &bpf_skc_lookup_tcp_proto;
7490         case BPF_FUNC_tcp_check_syncookie:
7491                 return &bpf_tcp_check_syncookie_proto;
7492         case BPF_FUNC_skb_ecn_set_ce:
7493                 return &bpf_skb_ecn_set_ce_proto;
7494         case BPF_FUNC_tcp_gen_syncookie:
7495                 return &bpf_tcp_gen_syncookie_proto;
7496         case BPF_FUNC_sk_assign:
7497                 return &bpf_sk_assign_proto;
7498 #endif
7499         default:
7500                 return bpf_sk_base_func_proto(func_id);
7501         }
7502 }
7503
7504 static const struct bpf_func_proto *
7505 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7506 {
7507         switch (func_id) {
7508         case BPF_FUNC_perf_event_output:
7509                 return &bpf_xdp_event_output_proto;
7510         case BPF_FUNC_get_smp_processor_id:
7511                 return &bpf_get_smp_processor_id_proto;
7512         case BPF_FUNC_csum_diff:
7513                 return &bpf_csum_diff_proto;
7514         case BPF_FUNC_xdp_adjust_head:
7515                 return &bpf_xdp_adjust_head_proto;
7516         case BPF_FUNC_xdp_adjust_meta:
7517                 return &bpf_xdp_adjust_meta_proto;
7518         case BPF_FUNC_redirect:
7519                 return &bpf_xdp_redirect_proto;
7520         case BPF_FUNC_redirect_map:
7521                 return &bpf_xdp_redirect_map_proto;
7522         case BPF_FUNC_xdp_adjust_tail:
7523                 return &bpf_xdp_adjust_tail_proto;
7524         case BPF_FUNC_fib_lookup:
7525                 return &bpf_xdp_fib_lookup_proto;
7526         case BPF_FUNC_check_mtu:
7527                 return &bpf_xdp_check_mtu_proto;
7528 #ifdef CONFIG_INET
7529         case BPF_FUNC_sk_lookup_udp:
7530                 return &bpf_xdp_sk_lookup_udp_proto;
7531         case BPF_FUNC_sk_lookup_tcp:
7532                 return &bpf_xdp_sk_lookup_tcp_proto;
7533         case BPF_FUNC_sk_release:
7534                 return &bpf_sk_release_proto;
7535         case BPF_FUNC_skc_lookup_tcp:
7536                 return &bpf_xdp_skc_lookup_tcp_proto;
7537         case BPF_FUNC_tcp_check_syncookie:
7538                 return &bpf_tcp_check_syncookie_proto;
7539         case BPF_FUNC_tcp_gen_syncookie:
7540                 return &bpf_tcp_gen_syncookie_proto;
7541 #endif
7542         default:
7543                 return bpf_sk_base_func_proto(func_id);
7544         }
7545 }
7546
7547 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7548 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7549
7550 static const struct bpf_func_proto *
7551 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7552 {
7553         switch (func_id) {
7554         case BPF_FUNC_setsockopt:
7555                 return &bpf_sock_ops_setsockopt_proto;
7556         case BPF_FUNC_getsockopt:
7557                 return &bpf_sock_ops_getsockopt_proto;
7558         case BPF_FUNC_sock_ops_cb_flags_set:
7559                 return &bpf_sock_ops_cb_flags_set_proto;
7560         case BPF_FUNC_sock_map_update:
7561                 return &bpf_sock_map_update_proto;
7562         case BPF_FUNC_sock_hash_update:
7563                 return &bpf_sock_hash_update_proto;
7564         case BPF_FUNC_get_socket_cookie:
7565                 return &bpf_get_socket_cookie_sock_ops_proto;
7566         case BPF_FUNC_get_local_storage:
7567                 return &bpf_get_local_storage_proto;
7568         case BPF_FUNC_perf_event_output:
7569                 return &bpf_event_output_data_proto;
7570         case BPF_FUNC_sk_storage_get:
7571                 return &bpf_sk_storage_get_proto;
7572         case BPF_FUNC_sk_storage_delete:
7573                 return &bpf_sk_storage_delete_proto;
7574         case BPF_FUNC_get_netns_cookie:
7575                 return &bpf_get_netns_cookie_sock_ops_proto;
7576 #ifdef CONFIG_INET
7577         case BPF_FUNC_load_hdr_opt:
7578                 return &bpf_sock_ops_load_hdr_opt_proto;
7579         case BPF_FUNC_store_hdr_opt:
7580                 return &bpf_sock_ops_store_hdr_opt_proto;
7581         case BPF_FUNC_reserve_hdr_opt:
7582                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7583         case BPF_FUNC_tcp_sock:
7584                 return &bpf_tcp_sock_proto;
7585 #endif /* CONFIG_INET */
7586         default:
7587                 return bpf_sk_base_func_proto(func_id);
7588         }
7589 }
7590
7591 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7592 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7593
7594 static const struct bpf_func_proto *
7595 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7596 {
7597         switch (func_id) {
7598         case BPF_FUNC_msg_redirect_map:
7599                 return &bpf_msg_redirect_map_proto;
7600         case BPF_FUNC_msg_redirect_hash:
7601                 return &bpf_msg_redirect_hash_proto;
7602         case BPF_FUNC_msg_apply_bytes:
7603                 return &bpf_msg_apply_bytes_proto;
7604         case BPF_FUNC_msg_cork_bytes:
7605                 return &bpf_msg_cork_bytes_proto;
7606         case BPF_FUNC_msg_pull_data:
7607                 return &bpf_msg_pull_data_proto;
7608         case BPF_FUNC_msg_push_data:
7609                 return &bpf_msg_push_data_proto;
7610         case BPF_FUNC_msg_pop_data:
7611                 return &bpf_msg_pop_data_proto;
7612         case BPF_FUNC_perf_event_output:
7613                 return &bpf_event_output_data_proto;
7614         case BPF_FUNC_get_current_uid_gid:
7615                 return &bpf_get_current_uid_gid_proto;
7616         case BPF_FUNC_get_current_pid_tgid:
7617                 return &bpf_get_current_pid_tgid_proto;
7618         case BPF_FUNC_sk_storage_get:
7619                 return &bpf_sk_storage_get_proto;
7620         case BPF_FUNC_sk_storage_delete:
7621                 return &bpf_sk_storage_delete_proto;
7622         case BPF_FUNC_get_netns_cookie:
7623                 return &bpf_get_netns_cookie_sk_msg_proto;
7624 #ifdef CONFIG_CGROUPS
7625         case BPF_FUNC_get_current_cgroup_id:
7626                 return &bpf_get_current_cgroup_id_proto;
7627         case BPF_FUNC_get_current_ancestor_cgroup_id:
7628                 return &bpf_get_current_ancestor_cgroup_id_proto;
7629 #endif
7630 #ifdef CONFIG_CGROUP_NET_CLASSID
7631         case BPF_FUNC_get_cgroup_classid:
7632                 return &bpf_get_cgroup_classid_curr_proto;
7633 #endif
7634         default:
7635                 return bpf_sk_base_func_proto(func_id);
7636         }
7637 }
7638
7639 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7640 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7641
7642 static const struct bpf_func_proto *
7643 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7644 {
7645         switch (func_id) {
7646         case BPF_FUNC_skb_store_bytes:
7647                 return &bpf_skb_store_bytes_proto;
7648         case BPF_FUNC_skb_load_bytes:
7649                 return &bpf_skb_load_bytes_proto;
7650         case BPF_FUNC_skb_pull_data:
7651                 return &sk_skb_pull_data_proto;
7652         case BPF_FUNC_skb_change_tail:
7653                 return &sk_skb_change_tail_proto;
7654         case BPF_FUNC_skb_change_head:
7655                 return &sk_skb_change_head_proto;
7656         case BPF_FUNC_skb_adjust_room:
7657                 return &sk_skb_adjust_room_proto;
7658         case BPF_FUNC_get_socket_cookie:
7659                 return &bpf_get_socket_cookie_proto;
7660         case BPF_FUNC_get_socket_uid:
7661                 return &bpf_get_socket_uid_proto;
7662         case BPF_FUNC_sk_redirect_map:
7663                 return &bpf_sk_redirect_map_proto;
7664         case BPF_FUNC_sk_redirect_hash:
7665                 return &bpf_sk_redirect_hash_proto;
7666         case BPF_FUNC_perf_event_output:
7667                 return &bpf_skb_event_output_proto;
7668 #ifdef CONFIG_INET
7669         case BPF_FUNC_sk_lookup_tcp:
7670                 return &bpf_sk_lookup_tcp_proto;
7671         case BPF_FUNC_sk_lookup_udp:
7672                 return &bpf_sk_lookup_udp_proto;
7673         case BPF_FUNC_sk_release:
7674                 return &bpf_sk_release_proto;
7675         case BPF_FUNC_skc_lookup_tcp:
7676                 return &bpf_skc_lookup_tcp_proto;
7677 #endif
7678         default:
7679                 return bpf_sk_base_func_proto(func_id);
7680         }
7681 }
7682
7683 static const struct bpf_func_proto *
7684 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7685 {
7686         switch (func_id) {
7687         case BPF_FUNC_skb_load_bytes:
7688                 return &bpf_flow_dissector_load_bytes_proto;
7689         default:
7690                 return bpf_sk_base_func_proto(func_id);
7691         }
7692 }
7693
7694 static const struct bpf_func_proto *
7695 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7696 {
7697         switch (func_id) {
7698         case BPF_FUNC_skb_load_bytes:
7699                 return &bpf_skb_load_bytes_proto;
7700         case BPF_FUNC_skb_pull_data:
7701                 return &bpf_skb_pull_data_proto;
7702         case BPF_FUNC_csum_diff:
7703                 return &bpf_csum_diff_proto;
7704         case BPF_FUNC_get_cgroup_classid:
7705                 return &bpf_get_cgroup_classid_proto;
7706         case BPF_FUNC_get_route_realm:
7707                 return &bpf_get_route_realm_proto;
7708         case BPF_FUNC_get_hash_recalc:
7709                 return &bpf_get_hash_recalc_proto;
7710         case BPF_FUNC_perf_event_output:
7711                 return &bpf_skb_event_output_proto;
7712         case BPF_FUNC_get_smp_processor_id:
7713                 return &bpf_get_smp_processor_id_proto;
7714         case BPF_FUNC_skb_under_cgroup:
7715                 return &bpf_skb_under_cgroup_proto;
7716         default:
7717                 return bpf_sk_base_func_proto(func_id);
7718         }
7719 }
7720
7721 static const struct bpf_func_proto *
7722 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7723 {
7724         switch (func_id) {
7725         case BPF_FUNC_lwt_push_encap:
7726                 return &bpf_lwt_in_push_encap_proto;
7727         default:
7728                 return lwt_out_func_proto(func_id, prog);
7729         }
7730 }
7731
7732 static const struct bpf_func_proto *
7733 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7734 {
7735         switch (func_id) {
7736         case BPF_FUNC_skb_get_tunnel_key:
7737                 return &bpf_skb_get_tunnel_key_proto;
7738         case BPF_FUNC_skb_set_tunnel_key:
7739                 return bpf_get_skb_set_tunnel_proto(func_id);
7740         case BPF_FUNC_skb_get_tunnel_opt:
7741                 return &bpf_skb_get_tunnel_opt_proto;
7742         case BPF_FUNC_skb_set_tunnel_opt:
7743                 return bpf_get_skb_set_tunnel_proto(func_id);
7744         case BPF_FUNC_redirect:
7745                 return &bpf_redirect_proto;
7746         case BPF_FUNC_clone_redirect:
7747                 return &bpf_clone_redirect_proto;
7748         case BPF_FUNC_skb_change_tail:
7749                 return &bpf_skb_change_tail_proto;
7750         case BPF_FUNC_skb_change_head:
7751                 return &bpf_skb_change_head_proto;
7752         case BPF_FUNC_skb_store_bytes:
7753                 return &bpf_skb_store_bytes_proto;
7754         case BPF_FUNC_csum_update:
7755                 return &bpf_csum_update_proto;
7756         case BPF_FUNC_csum_level:
7757                 return &bpf_csum_level_proto;
7758         case BPF_FUNC_l3_csum_replace:
7759                 return &bpf_l3_csum_replace_proto;
7760         case BPF_FUNC_l4_csum_replace:
7761                 return &bpf_l4_csum_replace_proto;
7762         case BPF_FUNC_set_hash_invalid:
7763                 return &bpf_set_hash_invalid_proto;
7764         case BPF_FUNC_lwt_push_encap:
7765                 return &bpf_lwt_xmit_push_encap_proto;
7766         default:
7767                 return lwt_out_func_proto(func_id, prog);
7768         }
7769 }
7770
7771 static const struct bpf_func_proto *
7772 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7773 {
7774         switch (func_id) {
7775 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7776         case BPF_FUNC_lwt_seg6_store_bytes:
7777                 return &bpf_lwt_seg6_store_bytes_proto;
7778         case BPF_FUNC_lwt_seg6_action:
7779                 return &bpf_lwt_seg6_action_proto;
7780         case BPF_FUNC_lwt_seg6_adjust_srh:
7781                 return &bpf_lwt_seg6_adjust_srh_proto;
7782 #endif
7783         default:
7784                 return lwt_out_func_proto(func_id, prog);
7785         }
7786 }
7787
7788 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7789                                     const struct bpf_prog *prog,
7790                                     struct bpf_insn_access_aux *info)
7791 {
7792         const int size_default = sizeof(__u32);
7793
7794         if (off < 0 || off >= sizeof(struct __sk_buff))
7795                 return false;
7796
7797         /* The verifier guarantees that size > 0. */
7798         if (off % size != 0)
7799                 return false;
7800
7801         switch (off) {
7802         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7803                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7804                         return false;
7805                 break;
7806         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7807         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7808         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7809         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7810         case bpf_ctx_range(struct __sk_buff, data):
7811         case bpf_ctx_range(struct __sk_buff, data_meta):
7812         case bpf_ctx_range(struct __sk_buff, data_end):
7813                 if (size != size_default)
7814                         return false;
7815                 break;
7816         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7817                 return false;
7818         case bpf_ctx_range(struct __sk_buff, tstamp):
7819                 if (size != sizeof(__u64))
7820                         return false;
7821                 break;
7822         case offsetof(struct __sk_buff, sk):
7823                 if (type == BPF_WRITE || size != sizeof(__u64))
7824                         return false;
7825                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7826                 break;
7827         default:
7828                 /* Only narrow read access allowed for now. */
7829                 if (type == BPF_WRITE) {
7830                         if (size != size_default)
7831                                 return false;
7832                 } else {
7833                         bpf_ctx_record_field_size(info, size_default);
7834                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7835                                 return false;
7836                 }
7837         }
7838
7839         return true;
7840 }
7841
7842 static bool sk_filter_is_valid_access(int off, int size,
7843                                       enum bpf_access_type type,
7844                                       const struct bpf_prog *prog,
7845                                       struct bpf_insn_access_aux *info)
7846 {
7847         switch (off) {
7848         case bpf_ctx_range(struct __sk_buff, tc_classid):
7849         case bpf_ctx_range(struct __sk_buff, data):
7850         case bpf_ctx_range(struct __sk_buff, data_meta):
7851         case bpf_ctx_range(struct __sk_buff, data_end):
7852         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7853         case bpf_ctx_range(struct __sk_buff, tstamp):
7854         case bpf_ctx_range(struct __sk_buff, wire_len):
7855                 return false;
7856         }
7857
7858         if (type == BPF_WRITE) {
7859                 switch (off) {
7860                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7861                         break;
7862                 default:
7863                         return false;
7864                 }
7865         }
7866
7867         return bpf_skb_is_valid_access(off, size, type, prog, info);
7868 }
7869
7870 static bool cg_skb_is_valid_access(int off, int size,
7871                                    enum bpf_access_type type,
7872                                    const struct bpf_prog *prog,
7873                                    struct bpf_insn_access_aux *info)
7874 {
7875         switch (off) {
7876         case bpf_ctx_range(struct __sk_buff, tc_classid):
7877         case bpf_ctx_range(struct __sk_buff, data_meta):
7878         case bpf_ctx_range(struct __sk_buff, wire_len):
7879                 return false;
7880         case bpf_ctx_range(struct __sk_buff, data):
7881         case bpf_ctx_range(struct __sk_buff, data_end):
7882                 if (!bpf_capable())
7883                         return false;
7884                 break;
7885         }
7886
7887         if (type == BPF_WRITE) {
7888                 switch (off) {
7889                 case bpf_ctx_range(struct __sk_buff, mark):
7890                 case bpf_ctx_range(struct __sk_buff, priority):
7891                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7892                         break;
7893                 case bpf_ctx_range(struct __sk_buff, tstamp):
7894                         if (!bpf_capable())
7895                                 return false;
7896                         break;
7897                 default:
7898                         return false;
7899                 }
7900         }
7901
7902         switch (off) {
7903         case bpf_ctx_range(struct __sk_buff, data):
7904                 info->reg_type = PTR_TO_PACKET;
7905                 break;
7906         case bpf_ctx_range(struct __sk_buff, data_end):
7907                 info->reg_type = PTR_TO_PACKET_END;
7908                 break;
7909         }
7910
7911         return bpf_skb_is_valid_access(off, size, type, prog, info);
7912 }
7913
7914 static bool lwt_is_valid_access(int off, int size,
7915                                 enum bpf_access_type type,
7916                                 const struct bpf_prog *prog,
7917                                 struct bpf_insn_access_aux *info)
7918 {
7919         switch (off) {
7920         case bpf_ctx_range(struct __sk_buff, tc_classid):
7921         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7922         case bpf_ctx_range(struct __sk_buff, data_meta):
7923         case bpf_ctx_range(struct __sk_buff, tstamp):
7924         case bpf_ctx_range(struct __sk_buff, wire_len):
7925                 return false;
7926         }
7927
7928         if (type == BPF_WRITE) {
7929                 switch (off) {
7930                 case bpf_ctx_range(struct __sk_buff, mark):
7931                 case bpf_ctx_range(struct __sk_buff, priority):
7932                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7933                         break;
7934                 default:
7935                         return false;
7936                 }
7937         }
7938
7939         switch (off) {
7940         case bpf_ctx_range(struct __sk_buff, data):
7941                 info->reg_type = PTR_TO_PACKET;
7942                 break;
7943         case bpf_ctx_range(struct __sk_buff, data_end):
7944                 info->reg_type = PTR_TO_PACKET_END;
7945                 break;
7946         }
7947
7948         return bpf_skb_is_valid_access(off, size, type, prog, info);
7949 }
7950
7951 /* Attach type specific accesses */
7952 static bool __sock_filter_check_attach_type(int off,
7953                                             enum bpf_access_type access_type,
7954                                             enum bpf_attach_type attach_type)
7955 {
7956         switch (off) {
7957         case offsetof(struct bpf_sock, bound_dev_if):
7958         case offsetof(struct bpf_sock, mark):
7959         case offsetof(struct bpf_sock, priority):
7960                 switch (attach_type) {
7961                 case BPF_CGROUP_INET_SOCK_CREATE:
7962                 case BPF_CGROUP_INET_SOCK_RELEASE:
7963                         goto full_access;
7964                 default:
7965                         return false;
7966                 }
7967         case bpf_ctx_range(struct bpf_sock, src_ip4):
7968                 switch (attach_type) {
7969                 case BPF_CGROUP_INET4_POST_BIND:
7970                         goto read_only;
7971                 default:
7972                         return false;
7973                 }
7974         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7975                 switch (attach_type) {
7976                 case BPF_CGROUP_INET6_POST_BIND:
7977                         goto read_only;
7978                 default:
7979                         return false;
7980                 }
7981         case bpf_ctx_range(struct bpf_sock, src_port):
7982                 switch (attach_type) {
7983                 case BPF_CGROUP_INET4_POST_BIND:
7984                 case BPF_CGROUP_INET6_POST_BIND:
7985                         goto read_only;
7986                 default:
7987                         return false;
7988                 }
7989         }
7990 read_only:
7991         return access_type == BPF_READ;
7992 full_access:
7993         return true;
7994 }
7995
7996 bool bpf_sock_common_is_valid_access(int off, int size,
7997                                      enum bpf_access_type type,
7998                                      struct bpf_insn_access_aux *info)
7999 {
8000         switch (off) {
8001         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8002                 return false;
8003         default:
8004                 return bpf_sock_is_valid_access(off, size, type, info);
8005         }
8006 }
8007
8008 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8009                               struct bpf_insn_access_aux *info)
8010 {
8011         const int size_default = sizeof(__u32);
8012         int field_size;
8013
8014         if (off < 0 || off >= sizeof(struct bpf_sock))
8015                 return false;
8016         if (off % size != 0)
8017                 return false;
8018
8019         switch (off) {
8020         case offsetof(struct bpf_sock, state):
8021         case offsetof(struct bpf_sock, family):
8022         case offsetof(struct bpf_sock, type):
8023         case offsetof(struct bpf_sock, protocol):
8024         case offsetof(struct bpf_sock, src_port):
8025         case offsetof(struct bpf_sock, rx_queue_mapping):
8026         case bpf_ctx_range(struct bpf_sock, src_ip4):
8027         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8028         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8029         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8030                 bpf_ctx_record_field_size(info, size_default);
8031                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8032         case bpf_ctx_range(struct bpf_sock, dst_port):
8033                 field_size = size == size_default ?
8034                         size_default : sizeof_field(struct bpf_sock, dst_port);
8035                 bpf_ctx_record_field_size(info, field_size);
8036                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8037         case offsetofend(struct bpf_sock, dst_port) ...
8038              offsetof(struct bpf_sock, dst_ip4) - 1:
8039                 return false;
8040         }
8041
8042         return size == size_default;
8043 }
8044
8045 static bool sock_filter_is_valid_access(int off, int size,
8046                                         enum bpf_access_type type,
8047                                         const struct bpf_prog *prog,
8048                                         struct bpf_insn_access_aux *info)
8049 {
8050         if (!bpf_sock_is_valid_access(off, size, type, info))
8051                 return false;
8052         return __sock_filter_check_attach_type(off, type,
8053                                                prog->expected_attach_type);
8054 }
8055
8056 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8057                              const struct bpf_prog *prog)
8058 {
8059         /* Neither direct read nor direct write requires any preliminary
8060          * action.
8061          */
8062         return 0;
8063 }
8064
8065 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8066                                 const struct bpf_prog *prog, int drop_verdict)
8067 {
8068         struct bpf_insn *insn = insn_buf;
8069
8070         if (!direct_write)
8071                 return 0;
8072
8073         /* if (!skb->cloned)
8074          *       goto start;
8075          *
8076          * (Fast-path, otherwise approximation that we might be
8077          *  a clone, do the rest in helper.)
8078          */
8079         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
8080         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8081         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8082
8083         /* ret = bpf_skb_pull_data(skb, 0); */
8084         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8085         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8086         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8087                                BPF_FUNC_skb_pull_data);
8088         /* if (!ret)
8089          *      goto restore;
8090          * return TC_ACT_SHOT;
8091          */
8092         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8093         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8094         *insn++ = BPF_EXIT_INSN();
8095
8096         /* restore: */
8097         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8098         /* start: */
8099         *insn++ = prog->insnsi[0];
8100
8101         return insn - insn_buf;
8102 }
8103
8104 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8105                           struct bpf_insn *insn_buf)
8106 {
8107         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8108         struct bpf_insn *insn = insn_buf;
8109
8110         if (!indirect) {
8111                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8112         } else {
8113                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8114                 if (orig->imm)
8115                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8116         }
8117         /* We're guaranteed here that CTX is in R6. */
8118         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8119
8120         switch (BPF_SIZE(orig->code)) {
8121         case BPF_B:
8122                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8123                 break;
8124         case BPF_H:
8125                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8126                 break;
8127         case BPF_W:
8128                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8129                 break;
8130         }
8131
8132         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8133         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8134         *insn++ = BPF_EXIT_INSN();
8135
8136         return insn - insn_buf;
8137 }
8138
8139 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8140                                const struct bpf_prog *prog)
8141 {
8142         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8143 }
8144
8145 static bool tc_cls_act_is_valid_access(int off, int size,
8146                                        enum bpf_access_type type,
8147                                        const struct bpf_prog *prog,
8148                                        struct bpf_insn_access_aux *info)
8149 {
8150         if (type == BPF_WRITE) {
8151                 switch (off) {
8152                 case bpf_ctx_range(struct __sk_buff, mark):
8153                 case bpf_ctx_range(struct __sk_buff, tc_index):
8154                 case bpf_ctx_range(struct __sk_buff, priority):
8155                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8156                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8157                 case bpf_ctx_range(struct __sk_buff, tstamp):
8158                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8159                         break;
8160                 default:
8161                         return false;
8162                 }
8163         }
8164
8165         switch (off) {
8166         case bpf_ctx_range(struct __sk_buff, data):
8167                 info->reg_type = PTR_TO_PACKET;
8168                 break;
8169         case bpf_ctx_range(struct __sk_buff, data_meta):
8170                 info->reg_type = PTR_TO_PACKET_META;
8171                 break;
8172         case bpf_ctx_range(struct __sk_buff, data_end):
8173                 info->reg_type = PTR_TO_PACKET_END;
8174                 break;
8175         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8176                 return false;
8177         }
8178
8179         return bpf_skb_is_valid_access(off, size, type, prog, info);
8180 }
8181
8182 static bool __is_valid_xdp_access(int off, int size)
8183 {
8184         if (off < 0 || off >= sizeof(struct xdp_md))
8185                 return false;
8186         if (off % size != 0)
8187                 return false;
8188         if (size != sizeof(__u32))
8189                 return false;
8190
8191         return true;
8192 }
8193
8194 static bool xdp_is_valid_access(int off, int size,
8195                                 enum bpf_access_type type,
8196                                 const struct bpf_prog *prog,
8197                                 struct bpf_insn_access_aux *info)
8198 {
8199         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8200                 switch (off) {
8201                 case offsetof(struct xdp_md, egress_ifindex):
8202                         return false;
8203                 }
8204         }
8205
8206         if (type == BPF_WRITE) {
8207                 if (bpf_prog_is_dev_bound(prog->aux)) {
8208                         switch (off) {
8209                         case offsetof(struct xdp_md, rx_queue_index):
8210                                 return __is_valid_xdp_access(off, size);
8211                         }
8212                 }
8213                 return false;
8214         }
8215
8216         switch (off) {
8217         case offsetof(struct xdp_md, data):
8218                 info->reg_type = PTR_TO_PACKET;
8219                 break;
8220         case offsetof(struct xdp_md, data_meta):
8221                 info->reg_type = PTR_TO_PACKET_META;
8222                 break;
8223         case offsetof(struct xdp_md, data_end):
8224                 info->reg_type = PTR_TO_PACKET_END;
8225                 break;
8226         }
8227
8228         return __is_valid_xdp_access(off, size);
8229 }
8230
8231 void bpf_warn_invalid_xdp_action(u32 act)
8232 {
8233         const u32 act_max = XDP_REDIRECT;
8234
8235         pr_warn_once("%s XDP return value %u, expect packet loss!\n",
8236                      act > act_max ? "Illegal" : "Driver unsupported",
8237                      act);
8238 }
8239 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8240
8241 static bool sock_addr_is_valid_access(int off, int size,
8242                                       enum bpf_access_type type,
8243                                       const struct bpf_prog *prog,
8244                                       struct bpf_insn_access_aux *info)
8245 {
8246         const int size_default = sizeof(__u32);
8247
8248         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8249                 return false;
8250         if (off % size != 0)
8251                 return false;
8252
8253         /* Disallow access to IPv6 fields from IPv4 contex and vise
8254          * versa.
8255          */
8256         switch (off) {
8257         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8258                 switch (prog->expected_attach_type) {
8259                 case BPF_CGROUP_INET4_BIND:
8260                 case BPF_CGROUP_INET4_CONNECT:
8261                 case BPF_CGROUP_INET4_GETPEERNAME:
8262                 case BPF_CGROUP_INET4_GETSOCKNAME:
8263                 case BPF_CGROUP_UDP4_SENDMSG:
8264                 case BPF_CGROUP_UDP4_RECVMSG:
8265                         break;
8266                 default:
8267                         return false;
8268                 }
8269                 break;
8270         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8271                 switch (prog->expected_attach_type) {
8272                 case BPF_CGROUP_INET6_BIND:
8273                 case BPF_CGROUP_INET6_CONNECT:
8274                 case BPF_CGROUP_INET6_GETPEERNAME:
8275                 case BPF_CGROUP_INET6_GETSOCKNAME:
8276                 case BPF_CGROUP_UDP6_SENDMSG:
8277                 case BPF_CGROUP_UDP6_RECVMSG:
8278                         break;
8279                 default:
8280                         return false;
8281                 }
8282                 break;
8283         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8284                 switch (prog->expected_attach_type) {
8285                 case BPF_CGROUP_UDP4_SENDMSG:
8286                         break;
8287                 default:
8288                         return false;
8289                 }
8290                 break;
8291         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8292                                 msg_src_ip6[3]):
8293                 switch (prog->expected_attach_type) {
8294                 case BPF_CGROUP_UDP6_SENDMSG:
8295                         break;
8296                 default:
8297                         return false;
8298                 }
8299                 break;
8300         }
8301
8302         switch (off) {
8303         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8304         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8305         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8306         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8307                                 msg_src_ip6[3]):
8308         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8309                 if (type == BPF_READ) {
8310                         bpf_ctx_record_field_size(info, size_default);
8311
8312                         if (bpf_ctx_wide_access_ok(off, size,
8313                                                    struct bpf_sock_addr,
8314                                                    user_ip6))
8315                                 return true;
8316
8317                         if (bpf_ctx_wide_access_ok(off, size,
8318                                                    struct bpf_sock_addr,
8319                                                    msg_src_ip6))
8320                                 return true;
8321
8322                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8323                                 return false;
8324                 } else {
8325                         if (bpf_ctx_wide_access_ok(off, size,
8326                                                    struct bpf_sock_addr,
8327                                                    user_ip6))
8328                                 return true;
8329
8330                         if (bpf_ctx_wide_access_ok(off, size,
8331                                                    struct bpf_sock_addr,
8332                                                    msg_src_ip6))
8333                                 return true;
8334
8335                         if (size != size_default)
8336                                 return false;
8337                 }
8338                 break;
8339         case offsetof(struct bpf_sock_addr, sk):
8340                 if (type != BPF_READ)
8341                         return false;
8342                 if (size != sizeof(__u64))
8343                         return false;
8344                 info->reg_type = PTR_TO_SOCKET;
8345                 break;
8346         default:
8347                 if (type == BPF_READ) {
8348                         if (size != size_default)
8349                                 return false;
8350                 } else {
8351                         return false;
8352                 }
8353         }
8354
8355         return true;
8356 }
8357
8358 static bool sock_ops_is_valid_access(int off, int size,
8359                                      enum bpf_access_type type,
8360                                      const struct bpf_prog *prog,
8361                                      struct bpf_insn_access_aux *info)
8362 {
8363         const int size_default = sizeof(__u32);
8364
8365         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8366                 return false;
8367
8368         /* The verifier guarantees that size > 0. */
8369         if (off % size != 0)
8370                 return false;
8371
8372         if (type == BPF_WRITE) {
8373                 switch (off) {
8374                 case offsetof(struct bpf_sock_ops, reply):
8375                 case offsetof(struct bpf_sock_ops, sk_txhash):
8376                         if (size != size_default)
8377                                 return false;
8378                         break;
8379                 default:
8380                         return false;
8381                 }
8382         } else {
8383                 switch (off) {
8384                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8385                                         bytes_acked):
8386                         if (size != sizeof(__u64))
8387                                 return false;
8388                         break;
8389                 case offsetof(struct bpf_sock_ops, sk):
8390                         if (size != sizeof(__u64))
8391                                 return false;
8392                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8393                         break;
8394                 case offsetof(struct bpf_sock_ops, skb_data):
8395                         if (size != sizeof(__u64))
8396                                 return false;
8397                         info->reg_type = PTR_TO_PACKET;
8398                         break;
8399                 case offsetof(struct bpf_sock_ops, skb_data_end):
8400                         if (size != sizeof(__u64))
8401                                 return false;
8402                         info->reg_type = PTR_TO_PACKET_END;
8403                         break;
8404                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8405                         bpf_ctx_record_field_size(info, size_default);
8406                         return bpf_ctx_narrow_access_ok(off, size,
8407                                                         size_default);
8408                 default:
8409                         if (size != size_default)
8410                                 return false;
8411                         break;
8412                 }
8413         }
8414
8415         return true;
8416 }
8417
8418 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8419                            const struct bpf_prog *prog)
8420 {
8421         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8422 }
8423
8424 static bool sk_skb_is_valid_access(int off, int size,
8425                                    enum bpf_access_type type,
8426                                    const struct bpf_prog *prog,
8427                                    struct bpf_insn_access_aux *info)
8428 {
8429         switch (off) {
8430         case bpf_ctx_range(struct __sk_buff, tc_classid):
8431         case bpf_ctx_range(struct __sk_buff, data_meta):
8432         case bpf_ctx_range(struct __sk_buff, tstamp):
8433         case bpf_ctx_range(struct __sk_buff, wire_len):
8434                 return false;
8435         }
8436
8437         if (type == BPF_WRITE) {
8438                 switch (off) {
8439                 case bpf_ctx_range(struct __sk_buff, tc_index):
8440                 case bpf_ctx_range(struct __sk_buff, priority):
8441                         break;
8442                 default:
8443                         return false;
8444                 }
8445         }
8446
8447         switch (off) {
8448         case bpf_ctx_range(struct __sk_buff, mark):
8449                 return false;
8450         case bpf_ctx_range(struct __sk_buff, data):
8451                 info->reg_type = PTR_TO_PACKET;
8452                 break;
8453         case bpf_ctx_range(struct __sk_buff, data_end):
8454                 info->reg_type = PTR_TO_PACKET_END;
8455                 break;
8456         }
8457
8458         return bpf_skb_is_valid_access(off, size, type, prog, info);
8459 }
8460
8461 static bool sk_msg_is_valid_access(int off, int size,
8462                                    enum bpf_access_type type,
8463                                    const struct bpf_prog *prog,
8464                                    struct bpf_insn_access_aux *info)
8465 {
8466         if (type == BPF_WRITE)
8467                 return false;
8468
8469         if (off % size != 0)
8470                 return false;
8471
8472         switch (off) {
8473         case offsetof(struct sk_msg_md, data):
8474                 info->reg_type = PTR_TO_PACKET;
8475                 if (size != sizeof(__u64))
8476                         return false;
8477                 break;
8478         case offsetof(struct sk_msg_md, data_end):
8479                 info->reg_type = PTR_TO_PACKET_END;
8480                 if (size != sizeof(__u64))
8481                         return false;
8482                 break;
8483         case offsetof(struct sk_msg_md, sk):
8484                 if (size != sizeof(__u64))
8485                         return false;
8486                 info->reg_type = PTR_TO_SOCKET;
8487                 break;
8488         case bpf_ctx_range(struct sk_msg_md, family):
8489         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8490         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8491         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8492         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8493         case bpf_ctx_range(struct sk_msg_md, remote_port):
8494         case bpf_ctx_range(struct sk_msg_md, local_port):
8495         case bpf_ctx_range(struct sk_msg_md, size):
8496                 if (size != sizeof(__u32))
8497                         return false;
8498                 break;
8499         default:
8500                 return false;
8501         }
8502         return true;
8503 }
8504
8505 static bool flow_dissector_is_valid_access(int off, int size,
8506                                            enum bpf_access_type type,
8507                                            const struct bpf_prog *prog,
8508                                            struct bpf_insn_access_aux *info)
8509 {
8510         const int size_default = sizeof(__u32);
8511
8512         if (off < 0 || off >= sizeof(struct __sk_buff))
8513                 return false;
8514
8515         if (type == BPF_WRITE)
8516                 return false;
8517
8518         switch (off) {
8519         case bpf_ctx_range(struct __sk_buff, data):
8520                 if (size != size_default)
8521                         return false;
8522                 info->reg_type = PTR_TO_PACKET;
8523                 return true;
8524         case bpf_ctx_range(struct __sk_buff, data_end):
8525                 if (size != size_default)
8526                         return false;
8527                 info->reg_type = PTR_TO_PACKET_END;
8528                 return true;
8529         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8530                 if (size != sizeof(__u64))
8531                         return false;
8532                 info->reg_type = PTR_TO_FLOW_KEYS;
8533                 return true;
8534         default:
8535                 return false;
8536         }
8537 }
8538
8539 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8540                                              const struct bpf_insn *si,
8541                                              struct bpf_insn *insn_buf,
8542                                              struct bpf_prog *prog,
8543                                              u32 *target_size)
8544
8545 {
8546         struct bpf_insn *insn = insn_buf;
8547
8548         switch (si->off) {
8549         case offsetof(struct __sk_buff, data):
8550                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8551                                       si->dst_reg, si->src_reg,
8552                                       offsetof(struct bpf_flow_dissector, data));
8553                 break;
8554
8555         case offsetof(struct __sk_buff, data_end):
8556                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8557                                       si->dst_reg, si->src_reg,
8558                                       offsetof(struct bpf_flow_dissector, data_end));
8559                 break;
8560
8561         case offsetof(struct __sk_buff, flow_keys):
8562                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8563                                       si->dst_reg, si->src_reg,
8564                                       offsetof(struct bpf_flow_dissector, flow_keys));
8565                 break;
8566         }
8567
8568         return insn - insn_buf;
8569 }
8570
8571 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8572                                                   struct bpf_insn *insn)
8573 {
8574         /* si->dst_reg = skb_shinfo(SKB); */
8575 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8576         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8577                               BPF_REG_AX, si->src_reg,
8578                               offsetof(struct sk_buff, end));
8579         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8580                               si->dst_reg, si->src_reg,
8581                               offsetof(struct sk_buff, head));
8582         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8583 #else
8584         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8585                               si->dst_reg, si->src_reg,
8586                               offsetof(struct sk_buff, end));
8587 #endif
8588
8589         return insn;
8590 }
8591
8592 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8593                                   const struct bpf_insn *si,
8594                                   struct bpf_insn *insn_buf,
8595                                   struct bpf_prog *prog, u32 *target_size)
8596 {
8597         struct bpf_insn *insn = insn_buf;
8598         int off;
8599
8600         switch (si->off) {
8601         case offsetof(struct __sk_buff, len):
8602                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8603                                       bpf_target_off(struct sk_buff, len, 4,
8604                                                      target_size));
8605                 break;
8606
8607         case offsetof(struct __sk_buff, protocol):
8608                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8609                                       bpf_target_off(struct sk_buff, protocol, 2,
8610                                                      target_size));
8611                 break;
8612
8613         case offsetof(struct __sk_buff, vlan_proto):
8614                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8615                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
8616                                                      target_size));
8617                 break;
8618
8619         case offsetof(struct __sk_buff, priority):
8620                 if (type == BPF_WRITE)
8621                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8622                                               bpf_target_off(struct sk_buff, priority, 4,
8623                                                              target_size));
8624                 else
8625                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8626                                               bpf_target_off(struct sk_buff, priority, 4,
8627                                                              target_size));
8628                 break;
8629
8630         case offsetof(struct __sk_buff, ingress_ifindex):
8631                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8632                                       bpf_target_off(struct sk_buff, skb_iif, 4,
8633                                                      target_size));
8634                 break;
8635
8636         case offsetof(struct __sk_buff, ifindex):
8637                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8638                                       si->dst_reg, si->src_reg,
8639                                       offsetof(struct sk_buff, dev));
8640                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8641                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8642                                       bpf_target_off(struct net_device, ifindex, 4,
8643                                                      target_size));
8644                 break;
8645
8646         case offsetof(struct __sk_buff, hash):
8647                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8648                                       bpf_target_off(struct sk_buff, hash, 4,
8649                                                      target_size));
8650                 break;
8651
8652         case offsetof(struct __sk_buff, mark):
8653                 if (type == BPF_WRITE)
8654                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8655                                               bpf_target_off(struct sk_buff, mark, 4,
8656                                                              target_size));
8657                 else
8658                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8659                                               bpf_target_off(struct sk_buff, mark, 4,
8660                                                              target_size));
8661                 break;
8662
8663         case offsetof(struct __sk_buff, pkt_type):
8664                 *target_size = 1;
8665                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8666                                       PKT_TYPE_OFFSET());
8667                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8668 #ifdef __BIG_ENDIAN_BITFIELD
8669                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8670 #endif
8671                 break;
8672
8673         case offsetof(struct __sk_buff, queue_mapping):
8674                 if (type == BPF_WRITE) {
8675                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8676                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8677                                               bpf_target_off(struct sk_buff,
8678                                                              queue_mapping,
8679                                                              2, target_size));
8680                 } else {
8681                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8682                                               bpf_target_off(struct sk_buff,
8683                                                              queue_mapping,
8684                                                              2, target_size));
8685                 }
8686                 break;
8687
8688         case offsetof(struct __sk_buff, vlan_present):
8689                 *target_size = 1;
8690                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8691                                       PKT_VLAN_PRESENT_OFFSET());
8692                 if (PKT_VLAN_PRESENT_BIT)
8693                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8694                 if (PKT_VLAN_PRESENT_BIT < 7)
8695                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8696                 break;
8697
8698         case offsetof(struct __sk_buff, vlan_tci):
8699                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8700                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
8701                                                      target_size));
8702                 break;
8703
8704         case offsetof(struct __sk_buff, cb[0]) ...
8705              offsetofend(struct __sk_buff, cb[4]) - 1:
8706                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8707                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8708                               offsetof(struct qdisc_skb_cb, data)) %
8709                              sizeof(__u64));
8710
8711                 prog->cb_access = 1;
8712                 off  = si->off;
8713                 off -= offsetof(struct __sk_buff, cb[0]);
8714                 off += offsetof(struct sk_buff, cb);
8715                 off += offsetof(struct qdisc_skb_cb, data);
8716                 if (type == BPF_WRITE)
8717                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8718                                               si->src_reg, off);
8719                 else
8720                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8721                                               si->src_reg, off);
8722                 break;
8723
8724         case offsetof(struct __sk_buff, tc_classid):
8725                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8726
8727                 off  = si->off;
8728                 off -= offsetof(struct __sk_buff, tc_classid);
8729                 off += offsetof(struct sk_buff, cb);
8730                 off += offsetof(struct qdisc_skb_cb, tc_classid);
8731                 *target_size = 2;
8732                 if (type == BPF_WRITE)
8733                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8734                                               si->src_reg, off);
8735                 else
8736                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8737                                               si->src_reg, off);
8738                 break;
8739
8740         case offsetof(struct __sk_buff, data):
8741                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8742                                       si->dst_reg, si->src_reg,
8743                                       offsetof(struct sk_buff, data));
8744                 break;
8745
8746         case offsetof(struct __sk_buff, data_meta):
8747                 off  = si->off;
8748                 off -= offsetof(struct __sk_buff, data_meta);
8749                 off += offsetof(struct sk_buff, cb);
8750                 off += offsetof(struct bpf_skb_data_end, data_meta);
8751                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8752                                       si->src_reg, off);
8753                 break;
8754
8755         case offsetof(struct __sk_buff, data_end):
8756                 off  = si->off;
8757                 off -= offsetof(struct __sk_buff, data_end);
8758                 off += offsetof(struct sk_buff, cb);
8759                 off += offsetof(struct bpf_skb_data_end, data_end);
8760                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8761                                       si->src_reg, off);
8762                 break;
8763
8764         case offsetof(struct __sk_buff, tc_index):
8765 #ifdef CONFIG_NET_SCHED
8766                 if (type == BPF_WRITE)
8767                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8768                                               bpf_target_off(struct sk_buff, tc_index, 2,
8769                                                              target_size));
8770                 else
8771                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8772                                               bpf_target_off(struct sk_buff, tc_index, 2,
8773                                                              target_size));
8774 #else
8775                 *target_size = 2;
8776                 if (type == BPF_WRITE)
8777                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8778                 else
8779                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8780 #endif
8781                 break;
8782
8783         case offsetof(struct __sk_buff, napi_id):
8784 #if defined(CONFIG_NET_RX_BUSY_POLL)
8785                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8786                                       bpf_target_off(struct sk_buff, napi_id, 4,
8787                                                      target_size));
8788                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8789                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8790 #else
8791                 *target_size = 4;
8792                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8793 #endif
8794                 break;
8795         case offsetof(struct __sk_buff, family):
8796                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8797
8798                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8799                                       si->dst_reg, si->src_reg,
8800                                       offsetof(struct sk_buff, sk));
8801                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8802                                       bpf_target_off(struct sock_common,
8803                                                      skc_family,
8804                                                      2, target_size));
8805                 break;
8806         case offsetof(struct __sk_buff, remote_ip4):
8807                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8808
8809                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8810                                       si->dst_reg, si->src_reg,
8811                                       offsetof(struct sk_buff, sk));
8812                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8813                                       bpf_target_off(struct sock_common,
8814                                                      skc_daddr,
8815                                                      4, target_size));
8816                 break;
8817         case offsetof(struct __sk_buff, local_ip4):
8818                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8819                                           skc_rcv_saddr) != 4);
8820
8821                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8822                                       si->dst_reg, si->src_reg,
8823                                       offsetof(struct sk_buff, sk));
8824                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8825                                       bpf_target_off(struct sock_common,
8826                                                      skc_rcv_saddr,
8827                                                      4, target_size));
8828                 break;
8829         case offsetof(struct __sk_buff, remote_ip6[0]) ...
8830              offsetof(struct __sk_buff, remote_ip6[3]):
8831 #if IS_ENABLED(CONFIG_IPV6)
8832                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8833                                           skc_v6_daddr.s6_addr32[0]) != 4);
8834
8835                 off = si->off;
8836                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8837
8838                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8839                                       si->dst_reg, si->src_reg,
8840                                       offsetof(struct sk_buff, sk));
8841                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8842                                       offsetof(struct sock_common,
8843                                                skc_v6_daddr.s6_addr32[0]) +
8844                                       off);
8845 #else
8846                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8847 #endif
8848                 break;
8849         case offsetof(struct __sk_buff, local_ip6[0]) ...
8850              offsetof(struct __sk_buff, local_ip6[3]):
8851 #if IS_ENABLED(CONFIG_IPV6)
8852                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8853                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8854
8855                 off = si->off;
8856                 off -= offsetof(struct __sk_buff, local_ip6[0]);
8857
8858                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8859                                       si->dst_reg, si->src_reg,
8860                                       offsetof(struct sk_buff, sk));
8861                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8862                                       offsetof(struct sock_common,
8863                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8864                                       off);
8865 #else
8866                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8867 #endif
8868                 break;
8869
8870         case offsetof(struct __sk_buff, remote_port):
8871                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8872
8873                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8874                                       si->dst_reg, si->src_reg,
8875                                       offsetof(struct sk_buff, sk));
8876                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8877                                       bpf_target_off(struct sock_common,
8878                                                      skc_dport,
8879                                                      2, target_size));
8880 #ifndef __BIG_ENDIAN_BITFIELD
8881                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8882 #endif
8883                 break;
8884
8885         case offsetof(struct __sk_buff, local_port):
8886                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8887
8888                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8889                                       si->dst_reg, si->src_reg,
8890                                       offsetof(struct sk_buff, sk));
8891                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8892                                       bpf_target_off(struct sock_common,
8893                                                      skc_num, 2, target_size));
8894                 break;
8895
8896         case offsetof(struct __sk_buff, tstamp):
8897                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8898
8899                 if (type == BPF_WRITE)
8900                         *insn++ = BPF_STX_MEM(BPF_DW,
8901                                               si->dst_reg, si->src_reg,
8902                                               bpf_target_off(struct sk_buff,
8903                                                              tstamp, 8,
8904                                                              target_size));
8905                 else
8906                         *insn++ = BPF_LDX_MEM(BPF_DW,
8907                                               si->dst_reg, si->src_reg,
8908                                               bpf_target_off(struct sk_buff,
8909                                                              tstamp, 8,
8910                                                              target_size));
8911                 break;
8912
8913         case offsetof(struct __sk_buff, gso_segs):
8914                 insn = bpf_convert_shinfo_access(si, insn);
8915                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8916                                       si->dst_reg, si->dst_reg,
8917                                       bpf_target_off(struct skb_shared_info,
8918                                                      gso_segs, 2,
8919                                                      target_size));
8920                 break;
8921         case offsetof(struct __sk_buff, gso_size):
8922                 insn = bpf_convert_shinfo_access(si, insn);
8923                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8924                                       si->dst_reg, si->dst_reg,
8925                                       bpf_target_off(struct skb_shared_info,
8926                                                      gso_size, 2,
8927                                                      target_size));
8928                 break;
8929         case offsetof(struct __sk_buff, wire_len):
8930                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8931
8932                 off = si->off;
8933                 off -= offsetof(struct __sk_buff, wire_len);
8934                 off += offsetof(struct sk_buff, cb);
8935                 off += offsetof(struct qdisc_skb_cb, pkt_len);
8936                 *target_size = 4;
8937                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8938                 break;
8939
8940         case offsetof(struct __sk_buff, sk):
8941                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8942                                       si->dst_reg, si->src_reg,
8943                                       offsetof(struct sk_buff, sk));
8944                 break;
8945         }
8946
8947         return insn - insn_buf;
8948 }
8949
8950 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8951                                 const struct bpf_insn *si,
8952                                 struct bpf_insn *insn_buf,
8953                                 struct bpf_prog *prog, u32 *target_size)
8954 {
8955         struct bpf_insn *insn = insn_buf;
8956         int off;
8957
8958         switch (si->off) {
8959         case offsetof(struct bpf_sock, bound_dev_if):
8960                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8961
8962                 if (type == BPF_WRITE)
8963                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8964                                         offsetof(struct sock, sk_bound_dev_if));
8965                 else
8966                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8967                                       offsetof(struct sock, sk_bound_dev_if));
8968                 break;
8969
8970         case offsetof(struct bpf_sock, mark):
8971                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8972
8973                 if (type == BPF_WRITE)
8974                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8975                                         offsetof(struct sock, sk_mark));
8976                 else
8977                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8978                                       offsetof(struct sock, sk_mark));
8979                 break;
8980
8981         case offsetof(struct bpf_sock, priority):
8982                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8983
8984                 if (type == BPF_WRITE)
8985                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8986                                         offsetof(struct sock, sk_priority));
8987                 else
8988                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8989                                       offsetof(struct sock, sk_priority));
8990                 break;
8991
8992         case offsetof(struct bpf_sock, family):
8993                 *insn++ = BPF_LDX_MEM(
8994                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8995                         si->dst_reg, si->src_reg,
8996                         bpf_target_off(struct sock_common,
8997                                        skc_family,
8998                                        sizeof_field(struct sock_common,
8999                                                     skc_family),
9000                                        target_size));
9001                 break;
9002
9003         case offsetof(struct bpf_sock, type):
9004                 *insn++ = BPF_LDX_MEM(
9005                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9006                         si->dst_reg, si->src_reg,
9007                         bpf_target_off(struct sock, sk_type,
9008                                        sizeof_field(struct sock, sk_type),
9009                                        target_size));
9010                 break;
9011
9012         case offsetof(struct bpf_sock, protocol):
9013                 *insn++ = BPF_LDX_MEM(
9014                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9015                         si->dst_reg, si->src_reg,
9016                         bpf_target_off(struct sock, sk_protocol,
9017                                        sizeof_field(struct sock, sk_protocol),
9018                                        target_size));
9019                 break;
9020
9021         case offsetof(struct bpf_sock, src_ip4):
9022                 *insn++ = BPF_LDX_MEM(
9023                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9024                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9025                                        sizeof_field(struct sock_common,
9026                                                     skc_rcv_saddr),
9027                                        target_size));
9028                 break;
9029
9030         case offsetof(struct bpf_sock, dst_ip4):
9031                 *insn++ = BPF_LDX_MEM(
9032                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9033                         bpf_target_off(struct sock_common, skc_daddr,
9034                                        sizeof_field(struct sock_common,
9035                                                     skc_daddr),
9036                                        target_size));
9037                 break;
9038
9039         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9040 #if IS_ENABLED(CONFIG_IPV6)
9041                 off = si->off;
9042                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9043                 *insn++ = BPF_LDX_MEM(
9044                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9045                         bpf_target_off(
9046                                 struct sock_common,
9047                                 skc_v6_rcv_saddr.s6_addr32[0],
9048                                 sizeof_field(struct sock_common,
9049                                              skc_v6_rcv_saddr.s6_addr32[0]),
9050                                 target_size) + off);
9051 #else
9052                 (void)off;
9053                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9054 #endif
9055                 break;
9056
9057         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9058 #if IS_ENABLED(CONFIG_IPV6)
9059                 off = si->off;
9060                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9061                 *insn++ = BPF_LDX_MEM(
9062                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9063                         bpf_target_off(struct sock_common,
9064                                        skc_v6_daddr.s6_addr32[0],
9065                                        sizeof_field(struct sock_common,
9066                                                     skc_v6_daddr.s6_addr32[0]),
9067                                        target_size) + off);
9068 #else
9069                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9070                 *target_size = 4;
9071 #endif
9072                 break;
9073
9074         case offsetof(struct bpf_sock, src_port):
9075                 *insn++ = BPF_LDX_MEM(
9076                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9077                         si->dst_reg, si->src_reg,
9078                         bpf_target_off(struct sock_common, skc_num,
9079                                        sizeof_field(struct sock_common,
9080                                                     skc_num),
9081                                        target_size));
9082                 break;
9083
9084         case offsetof(struct bpf_sock, dst_port):
9085                 *insn++ = BPF_LDX_MEM(
9086                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9087                         si->dst_reg, si->src_reg,
9088                         bpf_target_off(struct sock_common, skc_dport,
9089                                        sizeof_field(struct sock_common,
9090                                                     skc_dport),
9091                                        target_size));
9092                 break;
9093
9094         case offsetof(struct bpf_sock, state):
9095                 *insn++ = BPF_LDX_MEM(
9096                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9097                         si->dst_reg, si->src_reg,
9098                         bpf_target_off(struct sock_common, skc_state,
9099                                        sizeof_field(struct sock_common,
9100                                                     skc_state),
9101                                        target_size));
9102                 break;
9103         case offsetof(struct bpf_sock, rx_queue_mapping):
9104 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9105                 *insn++ = BPF_LDX_MEM(
9106                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9107                         si->dst_reg, si->src_reg,
9108                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9109                                        sizeof_field(struct sock,
9110                                                     sk_rx_queue_mapping),
9111                                        target_size));
9112                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9113                                       1);
9114                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9115 #else
9116                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9117                 *target_size = 2;
9118 #endif
9119                 break;
9120         }
9121
9122         return insn - insn_buf;
9123 }
9124
9125 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9126                                          const struct bpf_insn *si,
9127                                          struct bpf_insn *insn_buf,
9128                                          struct bpf_prog *prog, u32 *target_size)
9129 {
9130         struct bpf_insn *insn = insn_buf;
9131
9132         switch (si->off) {
9133         case offsetof(struct __sk_buff, ifindex):
9134                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9135                                       si->dst_reg, si->src_reg,
9136                                       offsetof(struct sk_buff, dev));
9137                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9138                                       bpf_target_off(struct net_device, ifindex, 4,
9139                                                      target_size));
9140                 break;
9141         default:
9142                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9143                                               target_size);
9144         }
9145
9146         return insn - insn_buf;
9147 }
9148
9149 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9150                                   const struct bpf_insn *si,
9151                                   struct bpf_insn *insn_buf,
9152                                   struct bpf_prog *prog, u32 *target_size)
9153 {
9154         struct bpf_insn *insn = insn_buf;
9155
9156         switch (si->off) {
9157         case offsetof(struct xdp_md, data):
9158                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9159                                       si->dst_reg, si->src_reg,
9160                                       offsetof(struct xdp_buff, data));
9161                 break;
9162         case offsetof(struct xdp_md, data_meta):
9163                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9164                                       si->dst_reg, si->src_reg,
9165                                       offsetof(struct xdp_buff, data_meta));
9166                 break;
9167         case offsetof(struct xdp_md, data_end):
9168                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9169                                       si->dst_reg, si->src_reg,
9170                                       offsetof(struct xdp_buff, data_end));
9171                 break;
9172         case offsetof(struct xdp_md, ingress_ifindex):
9173                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9174                                       si->dst_reg, si->src_reg,
9175                                       offsetof(struct xdp_buff, rxq));
9176                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9177                                       si->dst_reg, si->dst_reg,
9178                                       offsetof(struct xdp_rxq_info, dev));
9179                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9180                                       offsetof(struct net_device, ifindex));
9181                 break;
9182         case offsetof(struct xdp_md, rx_queue_index):
9183                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9184                                       si->dst_reg, si->src_reg,
9185                                       offsetof(struct xdp_buff, rxq));
9186                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9187                                       offsetof(struct xdp_rxq_info,
9188                                                queue_index));
9189                 break;
9190         case offsetof(struct xdp_md, egress_ifindex):
9191                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9192                                       si->dst_reg, si->src_reg,
9193                                       offsetof(struct xdp_buff, txq));
9194                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9195                                       si->dst_reg, si->dst_reg,
9196                                       offsetof(struct xdp_txq_info, dev));
9197                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9198                                       offsetof(struct net_device, ifindex));
9199                 break;
9200         }
9201
9202         return insn - insn_buf;
9203 }
9204
9205 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9206  * context Structure, F is Field in context structure that contains a pointer
9207  * to Nested Structure of type NS that has the field NF.
9208  *
9209  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9210  * sure that SIZE is not greater than actual size of S.F.NF.
9211  *
9212  * If offset OFF is provided, the load happens from that offset relative to
9213  * offset of NF.
9214  */
9215 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9216         do {                                                                   \
9217                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9218                                       si->src_reg, offsetof(S, F));            \
9219                 *insn++ = BPF_LDX_MEM(                                         \
9220                         SIZE, si->dst_reg, si->dst_reg,                        \
9221                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9222                                        target_size)                            \
9223                                 + OFF);                                        \
9224         } while (0)
9225
9226 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9227         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9228                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9229
9230 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9231  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9232  *
9233  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9234  * "register" since two registers available in convert_ctx_access are not
9235  * enough: we can't override neither SRC, since it contains value to store, nor
9236  * DST since it contains pointer to context that may be used by later
9237  * instructions. But we need a temporary place to save pointer to nested
9238  * structure whose field we want to store to.
9239  */
9240 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9241         do {                                                                   \
9242                 int tmp_reg = BPF_REG_9;                                       \
9243                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9244                         --tmp_reg;                                             \
9245                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9246                         --tmp_reg;                                             \
9247                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9248                                       offsetof(S, TF));                        \
9249                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9250                                       si->dst_reg, offsetof(S, F));            \
9251                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9252                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9253                                        target_size)                            \
9254                                 + OFF);                                        \
9255                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9256                                       offsetof(S, TF));                        \
9257         } while (0)
9258
9259 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9260                                                       TF)                      \
9261         do {                                                                   \
9262                 if (type == BPF_WRITE) {                                       \
9263                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9264                                                          OFF, TF);             \
9265                 } else {                                                       \
9266                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9267                                 S, NS, F, NF, SIZE, OFF);  \
9268                 }                                                              \
9269         } while (0)
9270
9271 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9272         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9273                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9274
9275 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9276                                         const struct bpf_insn *si,
9277                                         struct bpf_insn *insn_buf,
9278                                         struct bpf_prog *prog, u32 *target_size)
9279 {
9280         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9281         struct bpf_insn *insn = insn_buf;
9282
9283         switch (si->off) {
9284         case offsetof(struct bpf_sock_addr, user_family):
9285                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9286                                             struct sockaddr, uaddr, sa_family);
9287                 break;
9288
9289         case offsetof(struct bpf_sock_addr, user_ip4):
9290                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9291                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9292                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9293                 break;
9294
9295         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9296                 off = si->off;
9297                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9298                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9299                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9300                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9301                         tmp_reg);
9302                 break;
9303
9304         case offsetof(struct bpf_sock_addr, user_port):
9305                 /* To get port we need to know sa_family first and then treat
9306                  * sockaddr as either sockaddr_in or sockaddr_in6.
9307                  * Though we can simplify since port field has same offset and
9308                  * size in both structures.
9309                  * Here we check this invariant and use just one of the
9310                  * structures if it's true.
9311                  */
9312                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9313                              offsetof(struct sockaddr_in6, sin6_port));
9314                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9315                              sizeof_field(struct sockaddr_in6, sin6_port));
9316                 /* Account for sin6_port being smaller than user_port. */
9317                 port_size = min(port_size, BPF_LDST_BYTES(si));
9318                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9319                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9320                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9321                 break;
9322
9323         case offsetof(struct bpf_sock_addr, family):
9324                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9325                                             struct sock, sk, sk_family);
9326                 break;
9327
9328         case offsetof(struct bpf_sock_addr, type):
9329                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9330                                             struct sock, sk, sk_type);
9331                 break;
9332
9333         case offsetof(struct bpf_sock_addr, protocol):
9334                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9335                                             struct sock, sk, sk_protocol);
9336                 break;
9337
9338         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9339                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9340                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9341                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9342                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9343                 break;
9344
9345         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9346                                 msg_src_ip6[3]):
9347                 off = si->off;
9348                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9349                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9350                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9351                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9352                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9353                 break;
9354         case offsetof(struct bpf_sock_addr, sk):
9355                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9356                                       si->dst_reg, si->src_reg,
9357                                       offsetof(struct bpf_sock_addr_kern, sk));
9358                 break;
9359         }
9360
9361         return insn - insn_buf;
9362 }
9363
9364 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9365                                        const struct bpf_insn *si,
9366                                        struct bpf_insn *insn_buf,
9367                                        struct bpf_prog *prog,
9368                                        u32 *target_size)
9369 {
9370         struct bpf_insn *insn = insn_buf;
9371         int off;
9372
9373 /* Helper macro for adding read access to tcp_sock or sock fields. */
9374 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9375         do {                                                                  \
9376                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
9377                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9378                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9379                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9380                         reg--;                                                \
9381                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9382                         reg--;                                                \
9383                 if (si->dst_reg == si->src_reg) {                             \
9384                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9385                                           offsetof(struct bpf_sock_ops_kern,  \
9386                                           temp));                             \
9387                         fullsock_reg = reg;                                   \
9388                         jmp += 2;                                             \
9389                 }                                                             \
9390                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9391                                                 struct bpf_sock_ops_kern,     \
9392                                                 is_fullsock),                 \
9393                                       fullsock_reg, si->src_reg,              \
9394                                       offsetof(struct bpf_sock_ops_kern,      \
9395                                                is_fullsock));                 \
9396                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9397                 if (si->dst_reg == si->src_reg)                               \
9398                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9399                                       offsetof(struct bpf_sock_ops_kern,      \
9400                                       temp));                                 \
9401                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9402                                                 struct bpf_sock_ops_kern, sk),\
9403                                       si->dst_reg, si->src_reg,               \
9404                                       offsetof(struct bpf_sock_ops_kern, sk));\
9405                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9406                                                        OBJ_FIELD),            \
9407                                       si->dst_reg, si->dst_reg,               \
9408                                       offsetof(OBJ, OBJ_FIELD));              \
9409                 if (si->dst_reg == si->src_reg) {                             \
9410                         *insn++ = BPF_JMP_A(1);                               \
9411                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9412                                       offsetof(struct bpf_sock_ops_kern,      \
9413                                       temp));                                 \
9414                 }                                                             \
9415         } while (0)
9416
9417 #define SOCK_OPS_GET_SK()                                                             \
9418         do {                                                                  \
9419                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9420                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9421                         reg--;                                                \
9422                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9423                         reg--;                                                \
9424                 if (si->dst_reg == si->src_reg) {                             \
9425                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9426                                           offsetof(struct bpf_sock_ops_kern,  \
9427                                           temp));                             \
9428                         fullsock_reg = reg;                                   \
9429                         jmp += 2;                                             \
9430                 }                                                             \
9431                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9432                                                 struct bpf_sock_ops_kern,     \
9433                                                 is_fullsock),                 \
9434                                       fullsock_reg, si->src_reg,              \
9435                                       offsetof(struct bpf_sock_ops_kern,      \
9436                                                is_fullsock));                 \
9437                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9438                 if (si->dst_reg == si->src_reg)                               \
9439                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9440                                       offsetof(struct bpf_sock_ops_kern,      \
9441                                       temp));                                 \
9442                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9443                                                 struct bpf_sock_ops_kern, sk),\
9444                                       si->dst_reg, si->src_reg,               \
9445                                       offsetof(struct bpf_sock_ops_kern, sk));\
9446                 if (si->dst_reg == si->src_reg) {                             \
9447                         *insn++ = BPF_JMP_A(1);                               \
9448                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9449                                       offsetof(struct bpf_sock_ops_kern,      \
9450                                       temp));                                 \
9451                 }                                                             \
9452         } while (0)
9453
9454 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9455                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9456
9457 /* Helper macro for adding write access to tcp_sock or sock fields.
9458  * The macro is called with two registers, dst_reg which contains a pointer
9459  * to ctx (context) and src_reg which contains the value that should be
9460  * stored. However, we need an additional register since we cannot overwrite
9461  * dst_reg because it may be used later in the program.
9462  * Instead we "borrow" one of the other register. We first save its value
9463  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9464  * it at the end of the macro.
9465  */
9466 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9467         do {                                                                  \
9468                 int reg = BPF_REG_9;                                          \
9469                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9470                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9471                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9472                         reg--;                                                \
9473                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9474                         reg--;                                                \
9475                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9476                                       offsetof(struct bpf_sock_ops_kern,      \
9477                                                temp));                        \
9478                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9479                                                 struct bpf_sock_ops_kern,     \
9480                                                 is_fullsock),                 \
9481                                       reg, si->dst_reg,                       \
9482                                       offsetof(struct bpf_sock_ops_kern,      \
9483                                                is_fullsock));                 \
9484                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
9485                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9486                                                 struct bpf_sock_ops_kern, sk),\
9487                                       reg, si->dst_reg,                       \
9488                                       offsetof(struct bpf_sock_ops_kern, sk));\
9489                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
9490                                       reg, si->src_reg,                       \
9491                                       offsetof(OBJ, OBJ_FIELD));              \
9492                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
9493                                       offsetof(struct bpf_sock_ops_kern,      \
9494                                                temp));                        \
9495         } while (0)
9496
9497 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
9498         do {                                                                  \
9499                 if (TYPE == BPF_WRITE)                                        \
9500                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9501                 else                                                          \
9502                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9503         } while (0)
9504
9505         if (insn > insn_buf)
9506                 return insn - insn_buf;
9507
9508         switch (si->off) {
9509         case offsetof(struct bpf_sock_ops, op):
9510                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9511                                                        op),
9512                                       si->dst_reg, si->src_reg,
9513                                       offsetof(struct bpf_sock_ops_kern, op));
9514                 break;
9515
9516         case offsetof(struct bpf_sock_ops, replylong[0]) ...
9517              offsetof(struct bpf_sock_ops, replylong[3]):
9518                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9519                              sizeof_field(struct bpf_sock_ops_kern, reply));
9520                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9521                              sizeof_field(struct bpf_sock_ops_kern, replylong));
9522                 off = si->off;
9523                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9524                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9525                 if (type == BPF_WRITE)
9526                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9527                                               off);
9528                 else
9529                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9530                                               off);
9531                 break;
9532
9533         case offsetof(struct bpf_sock_ops, family):
9534                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9535
9536                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9537                                               struct bpf_sock_ops_kern, sk),
9538                                       si->dst_reg, si->src_reg,
9539                                       offsetof(struct bpf_sock_ops_kern, sk));
9540                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9541                                       offsetof(struct sock_common, skc_family));
9542                 break;
9543
9544         case offsetof(struct bpf_sock_ops, remote_ip4):
9545                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9546
9547                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9548                                                 struct bpf_sock_ops_kern, sk),
9549                                       si->dst_reg, si->src_reg,
9550                                       offsetof(struct bpf_sock_ops_kern, sk));
9551                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9552                                       offsetof(struct sock_common, skc_daddr));
9553                 break;
9554
9555         case offsetof(struct bpf_sock_ops, local_ip4):
9556                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9557                                           skc_rcv_saddr) != 4);
9558
9559                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9560                                               struct bpf_sock_ops_kern, sk),
9561                                       si->dst_reg, si->src_reg,
9562                                       offsetof(struct bpf_sock_ops_kern, sk));
9563                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9564                                       offsetof(struct sock_common,
9565                                                skc_rcv_saddr));
9566                 break;
9567
9568         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9569              offsetof(struct bpf_sock_ops, remote_ip6[3]):
9570 #if IS_ENABLED(CONFIG_IPV6)
9571                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9572                                           skc_v6_daddr.s6_addr32[0]) != 4);
9573
9574                 off = si->off;
9575                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9576                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9577                                                 struct bpf_sock_ops_kern, sk),
9578                                       si->dst_reg, si->src_reg,
9579                                       offsetof(struct bpf_sock_ops_kern, sk));
9580                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9581                                       offsetof(struct sock_common,
9582                                                skc_v6_daddr.s6_addr32[0]) +
9583                                       off);
9584 #else
9585                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9586 #endif
9587                 break;
9588
9589         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9590              offsetof(struct bpf_sock_ops, local_ip6[3]):
9591 #if IS_ENABLED(CONFIG_IPV6)
9592                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9593                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9594
9595                 off = si->off;
9596                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9597                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9598                                                 struct bpf_sock_ops_kern, sk),
9599                                       si->dst_reg, si->src_reg,
9600                                       offsetof(struct bpf_sock_ops_kern, sk));
9601                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9602                                       offsetof(struct sock_common,
9603                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9604                                       off);
9605 #else
9606                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9607 #endif
9608                 break;
9609
9610         case offsetof(struct bpf_sock_ops, remote_port):
9611                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9612
9613                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9614                                                 struct bpf_sock_ops_kern, sk),
9615                                       si->dst_reg, si->src_reg,
9616                                       offsetof(struct bpf_sock_ops_kern, sk));
9617                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9618                                       offsetof(struct sock_common, skc_dport));
9619 #ifndef __BIG_ENDIAN_BITFIELD
9620                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9621 #endif
9622                 break;
9623
9624         case offsetof(struct bpf_sock_ops, local_port):
9625                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9626
9627                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9628                                                 struct bpf_sock_ops_kern, sk),
9629                                       si->dst_reg, si->src_reg,
9630                                       offsetof(struct bpf_sock_ops_kern, sk));
9631                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9632                                       offsetof(struct sock_common, skc_num));
9633                 break;
9634
9635         case offsetof(struct bpf_sock_ops, is_fullsock):
9636                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9637                                                 struct bpf_sock_ops_kern,
9638                                                 is_fullsock),
9639                                       si->dst_reg, si->src_reg,
9640                                       offsetof(struct bpf_sock_ops_kern,
9641                                                is_fullsock));
9642                 break;
9643
9644         case offsetof(struct bpf_sock_ops, state):
9645                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9646
9647                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9648                                                 struct bpf_sock_ops_kern, sk),
9649                                       si->dst_reg, si->src_reg,
9650                                       offsetof(struct bpf_sock_ops_kern, sk));
9651                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9652                                       offsetof(struct sock_common, skc_state));
9653                 break;
9654
9655         case offsetof(struct bpf_sock_ops, rtt_min):
9656                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9657                              sizeof(struct minmax));
9658                 BUILD_BUG_ON(sizeof(struct minmax) <
9659                              sizeof(struct minmax_sample));
9660
9661                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9662                                                 struct bpf_sock_ops_kern, sk),
9663                                       si->dst_reg, si->src_reg,
9664                                       offsetof(struct bpf_sock_ops_kern, sk));
9665                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9666                                       offsetof(struct tcp_sock, rtt_min) +
9667                                       sizeof_field(struct minmax_sample, t));
9668                 break;
9669
9670         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9671                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9672                                    struct tcp_sock);
9673                 break;
9674
9675         case offsetof(struct bpf_sock_ops, sk_txhash):
9676                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9677                                           struct sock, type);
9678                 break;
9679         case offsetof(struct bpf_sock_ops, snd_cwnd):
9680                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9681                 break;
9682         case offsetof(struct bpf_sock_ops, srtt_us):
9683                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9684                 break;
9685         case offsetof(struct bpf_sock_ops, snd_ssthresh):
9686                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9687                 break;
9688         case offsetof(struct bpf_sock_ops, rcv_nxt):
9689                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9690                 break;
9691         case offsetof(struct bpf_sock_ops, snd_nxt):
9692                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9693                 break;
9694         case offsetof(struct bpf_sock_ops, snd_una):
9695                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9696                 break;
9697         case offsetof(struct bpf_sock_ops, mss_cache):
9698                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9699                 break;
9700         case offsetof(struct bpf_sock_ops, ecn_flags):
9701                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9702                 break;
9703         case offsetof(struct bpf_sock_ops, rate_delivered):
9704                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9705                 break;
9706         case offsetof(struct bpf_sock_ops, rate_interval_us):
9707                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9708                 break;
9709         case offsetof(struct bpf_sock_ops, packets_out):
9710                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9711                 break;
9712         case offsetof(struct bpf_sock_ops, retrans_out):
9713                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9714                 break;
9715         case offsetof(struct bpf_sock_ops, total_retrans):
9716                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9717                 break;
9718         case offsetof(struct bpf_sock_ops, segs_in):
9719                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9720                 break;
9721         case offsetof(struct bpf_sock_ops, data_segs_in):
9722                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9723                 break;
9724         case offsetof(struct bpf_sock_ops, segs_out):
9725                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9726                 break;
9727         case offsetof(struct bpf_sock_ops, data_segs_out):
9728                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9729                 break;
9730         case offsetof(struct bpf_sock_ops, lost_out):
9731                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9732                 break;
9733         case offsetof(struct bpf_sock_ops, sacked_out):
9734                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9735                 break;
9736         case offsetof(struct bpf_sock_ops, bytes_received):
9737                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9738                 break;
9739         case offsetof(struct bpf_sock_ops, bytes_acked):
9740                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9741                 break;
9742         case offsetof(struct bpf_sock_ops, sk):
9743                 SOCK_OPS_GET_SK();
9744                 break;
9745         case offsetof(struct bpf_sock_ops, skb_data_end):
9746                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9747                                                        skb_data_end),
9748                                       si->dst_reg, si->src_reg,
9749                                       offsetof(struct bpf_sock_ops_kern,
9750                                                skb_data_end));
9751                 break;
9752         case offsetof(struct bpf_sock_ops, skb_data):
9753                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9754                                                        skb),
9755                                       si->dst_reg, si->src_reg,
9756                                       offsetof(struct bpf_sock_ops_kern,
9757                                                skb));
9758                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9759                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9760                                       si->dst_reg, si->dst_reg,
9761                                       offsetof(struct sk_buff, data));
9762                 break;
9763         case offsetof(struct bpf_sock_ops, skb_len):
9764                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9765                                                        skb),
9766                                       si->dst_reg, si->src_reg,
9767                                       offsetof(struct bpf_sock_ops_kern,
9768                                                skb));
9769                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9770                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9771                                       si->dst_reg, si->dst_reg,
9772                                       offsetof(struct sk_buff, len));
9773                 break;
9774         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9775                 off = offsetof(struct sk_buff, cb);
9776                 off += offsetof(struct tcp_skb_cb, tcp_flags);
9777                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9778                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9779                                                        skb),
9780                                       si->dst_reg, si->src_reg,
9781                                       offsetof(struct bpf_sock_ops_kern,
9782                                                skb));
9783                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9784                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9785                                                        tcp_flags),
9786                                       si->dst_reg, si->dst_reg, off);
9787                 break;
9788         }
9789         return insn - insn_buf;
9790 }
9791
9792 /* data_end = skb->data + skb_headlen() */
9793 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9794                                                     struct bpf_insn *insn)
9795 {
9796         int reg;
9797         int temp_reg_off = offsetof(struct sk_buff, cb) +
9798                            offsetof(struct sk_skb_cb, temp_reg);
9799
9800         if (si->src_reg == si->dst_reg) {
9801                 /* We need an extra register, choose and save a register. */
9802                 reg = BPF_REG_9;
9803                 if (si->src_reg == reg || si->dst_reg == reg)
9804                         reg--;
9805                 if (si->src_reg == reg || si->dst_reg == reg)
9806                         reg--;
9807                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
9808         } else {
9809                 reg = si->dst_reg;
9810         }
9811
9812         /* reg = skb->data */
9813         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9814                               reg, si->src_reg,
9815                               offsetof(struct sk_buff, data));
9816         /* AX = skb->len */
9817         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9818                               BPF_REG_AX, si->src_reg,
9819                               offsetof(struct sk_buff, len));
9820         /* reg = skb->data + skb->len */
9821         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
9822         /* AX = skb->data_len */
9823         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9824                               BPF_REG_AX, si->src_reg,
9825                               offsetof(struct sk_buff, data_len));
9826
9827         /* reg = skb->data + skb->len - skb->data_len */
9828         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
9829
9830         if (si->src_reg == si->dst_reg) {
9831                 /* Restore the saved register */
9832                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
9833                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
9834                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
9835         }
9836
9837         return insn;
9838 }
9839
9840 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9841                                      const struct bpf_insn *si,
9842                                      struct bpf_insn *insn_buf,
9843                                      struct bpf_prog *prog, u32 *target_size)
9844 {
9845         struct bpf_insn *insn = insn_buf;
9846         int off;
9847
9848         switch (si->off) {
9849         case offsetof(struct __sk_buff, data_end):
9850                 insn = bpf_convert_data_end_access(si, insn);
9851                 break;
9852         case offsetof(struct __sk_buff, cb[0]) ...
9853              offsetofend(struct __sk_buff, cb[4]) - 1:
9854                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9855                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9856                               offsetof(struct sk_skb_cb, data)) %
9857                              sizeof(__u64));
9858
9859                 prog->cb_access = 1;
9860                 off  = si->off;
9861                 off -= offsetof(struct __sk_buff, cb[0]);
9862                 off += offsetof(struct sk_buff, cb);
9863                 off += offsetof(struct sk_skb_cb, data);
9864                 if (type == BPF_WRITE)
9865                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9866                                               si->src_reg, off);
9867                 else
9868                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9869                                               si->src_reg, off);
9870                 break;
9871
9872
9873         default:
9874                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9875                                               target_size);
9876         }
9877
9878         return insn - insn_buf;
9879 }
9880
9881 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9882                                      const struct bpf_insn *si,
9883                                      struct bpf_insn *insn_buf,
9884                                      struct bpf_prog *prog, u32 *target_size)
9885 {
9886         struct bpf_insn *insn = insn_buf;
9887 #if IS_ENABLED(CONFIG_IPV6)
9888         int off;
9889 #endif
9890
9891         /* convert ctx uses the fact sg element is first in struct */
9892         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9893
9894         switch (si->off) {
9895         case offsetof(struct sk_msg_md, data):
9896                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9897                                       si->dst_reg, si->src_reg,
9898                                       offsetof(struct sk_msg, data));
9899                 break;
9900         case offsetof(struct sk_msg_md, data_end):
9901                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9902                                       si->dst_reg, si->src_reg,
9903                                       offsetof(struct sk_msg, data_end));
9904                 break;
9905         case offsetof(struct sk_msg_md, family):
9906                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9907
9908                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9909                                               struct sk_msg, sk),
9910                                       si->dst_reg, si->src_reg,
9911                                       offsetof(struct sk_msg, sk));
9912                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9913                                       offsetof(struct sock_common, skc_family));
9914                 break;
9915
9916         case offsetof(struct sk_msg_md, remote_ip4):
9917                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9918
9919                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9920                                                 struct sk_msg, sk),
9921                                       si->dst_reg, si->src_reg,
9922                                       offsetof(struct sk_msg, sk));
9923                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9924                                       offsetof(struct sock_common, skc_daddr));
9925                 break;
9926
9927         case offsetof(struct sk_msg_md, local_ip4):
9928                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9929                                           skc_rcv_saddr) != 4);
9930
9931                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9932                                               struct sk_msg, sk),
9933                                       si->dst_reg, si->src_reg,
9934                                       offsetof(struct sk_msg, sk));
9935                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9936                                       offsetof(struct sock_common,
9937                                                skc_rcv_saddr));
9938                 break;
9939
9940         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9941              offsetof(struct sk_msg_md, remote_ip6[3]):
9942 #if IS_ENABLED(CONFIG_IPV6)
9943                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9944                                           skc_v6_daddr.s6_addr32[0]) != 4);
9945
9946                 off = si->off;
9947                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9948                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9949                                                 struct sk_msg, sk),
9950                                       si->dst_reg, si->src_reg,
9951                                       offsetof(struct sk_msg, sk));
9952                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9953                                       offsetof(struct sock_common,
9954                                                skc_v6_daddr.s6_addr32[0]) +
9955                                       off);
9956 #else
9957                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9958 #endif
9959                 break;
9960
9961         case offsetof(struct sk_msg_md, local_ip6[0]) ...
9962              offsetof(struct sk_msg_md, local_ip6[3]):
9963 #if IS_ENABLED(CONFIG_IPV6)
9964                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9965                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9966
9967                 off = si->off;
9968                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9969                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9970                                                 struct sk_msg, sk),
9971                                       si->dst_reg, si->src_reg,
9972                                       offsetof(struct sk_msg, sk));
9973                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9974                                       offsetof(struct sock_common,
9975                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9976                                       off);
9977 #else
9978                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9979 #endif
9980                 break;
9981
9982         case offsetof(struct sk_msg_md, remote_port):
9983                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9984
9985                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9986                                                 struct sk_msg, sk),
9987                                       si->dst_reg, si->src_reg,
9988                                       offsetof(struct sk_msg, sk));
9989                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9990                                       offsetof(struct sock_common, skc_dport));
9991 #ifndef __BIG_ENDIAN_BITFIELD
9992                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9993 #endif
9994                 break;
9995
9996         case offsetof(struct sk_msg_md, local_port):
9997                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9998
9999                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10000                                                 struct sk_msg, sk),
10001                                       si->dst_reg, si->src_reg,
10002                                       offsetof(struct sk_msg, sk));
10003                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10004                                       offsetof(struct sock_common, skc_num));
10005                 break;
10006
10007         case offsetof(struct sk_msg_md, size):
10008                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10009                                       si->dst_reg, si->src_reg,
10010                                       offsetof(struct sk_msg_sg, size));
10011                 break;
10012
10013         case offsetof(struct sk_msg_md, sk):
10014                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10015                                       si->dst_reg, si->src_reg,
10016                                       offsetof(struct sk_msg, sk));
10017                 break;
10018         }
10019
10020         return insn - insn_buf;
10021 }
10022
10023 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10024         .get_func_proto         = sk_filter_func_proto,
10025         .is_valid_access        = sk_filter_is_valid_access,
10026         .convert_ctx_access     = bpf_convert_ctx_access,
10027         .gen_ld_abs             = bpf_gen_ld_abs,
10028 };
10029
10030 const struct bpf_prog_ops sk_filter_prog_ops = {
10031         .test_run               = bpf_prog_test_run_skb,
10032 };
10033
10034 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10035         .get_func_proto         = tc_cls_act_func_proto,
10036         .is_valid_access        = tc_cls_act_is_valid_access,
10037         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10038         .gen_prologue           = tc_cls_act_prologue,
10039         .gen_ld_abs             = bpf_gen_ld_abs,
10040         .check_kfunc_call       = bpf_prog_test_check_kfunc_call,
10041 };
10042
10043 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10044         .test_run               = bpf_prog_test_run_skb,
10045 };
10046
10047 const struct bpf_verifier_ops xdp_verifier_ops = {
10048         .get_func_proto         = xdp_func_proto,
10049         .is_valid_access        = xdp_is_valid_access,
10050         .convert_ctx_access     = xdp_convert_ctx_access,
10051         .gen_prologue           = bpf_noop_prologue,
10052 };
10053
10054 const struct bpf_prog_ops xdp_prog_ops = {
10055         .test_run               = bpf_prog_test_run_xdp,
10056 };
10057
10058 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10059         .get_func_proto         = cg_skb_func_proto,
10060         .is_valid_access        = cg_skb_is_valid_access,
10061         .convert_ctx_access     = bpf_convert_ctx_access,
10062 };
10063
10064 const struct bpf_prog_ops cg_skb_prog_ops = {
10065         .test_run               = bpf_prog_test_run_skb,
10066 };
10067
10068 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10069         .get_func_proto         = lwt_in_func_proto,
10070         .is_valid_access        = lwt_is_valid_access,
10071         .convert_ctx_access     = bpf_convert_ctx_access,
10072 };
10073
10074 const struct bpf_prog_ops lwt_in_prog_ops = {
10075         .test_run               = bpf_prog_test_run_skb,
10076 };
10077
10078 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10079         .get_func_proto         = lwt_out_func_proto,
10080         .is_valid_access        = lwt_is_valid_access,
10081         .convert_ctx_access     = bpf_convert_ctx_access,
10082 };
10083
10084 const struct bpf_prog_ops lwt_out_prog_ops = {
10085         .test_run               = bpf_prog_test_run_skb,
10086 };
10087
10088 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10089         .get_func_proto         = lwt_xmit_func_proto,
10090         .is_valid_access        = lwt_is_valid_access,
10091         .convert_ctx_access     = bpf_convert_ctx_access,
10092         .gen_prologue           = tc_cls_act_prologue,
10093 };
10094
10095 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10096         .test_run               = bpf_prog_test_run_skb,
10097 };
10098
10099 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10100         .get_func_proto         = lwt_seg6local_func_proto,
10101         .is_valid_access        = lwt_is_valid_access,
10102         .convert_ctx_access     = bpf_convert_ctx_access,
10103 };
10104
10105 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10106         .test_run               = bpf_prog_test_run_skb,
10107 };
10108
10109 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10110         .get_func_proto         = sock_filter_func_proto,
10111         .is_valid_access        = sock_filter_is_valid_access,
10112         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10113 };
10114
10115 const struct bpf_prog_ops cg_sock_prog_ops = {
10116 };
10117
10118 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10119         .get_func_proto         = sock_addr_func_proto,
10120         .is_valid_access        = sock_addr_is_valid_access,
10121         .convert_ctx_access     = sock_addr_convert_ctx_access,
10122 };
10123
10124 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10125 };
10126
10127 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10128         .get_func_proto         = sock_ops_func_proto,
10129         .is_valid_access        = sock_ops_is_valid_access,
10130         .convert_ctx_access     = sock_ops_convert_ctx_access,
10131 };
10132
10133 const struct bpf_prog_ops sock_ops_prog_ops = {
10134 };
10135
10136 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10137         .get_func_proto         = sk_skb_func_proto,
10138         .is_valid_access        = sk_skb_is_valid_access,
10139         .convert_ctx_access     = sk_skb_convert_ctx_access,
10140         .gen_prologue           = sk_skb_prologue,
10141 };
10142
10143 const struct bpf_prog_ops sk_skb_prog_ops = {
10144 };
10145
10146 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10147         .get_func_proto         = sk_msg_func_proto,
10148         .is_valid_access        = sk_msg_is_valid_access,
10149         .convert_ctx_access     = sk_msg_convert_ctx_access,
10150         .gen_prologue           = bpf_noop_prologue,
10151 };
10152
10153 const struct bpf_prog_ops sk_msg_prog_ops = {
10154 };
10155
10156 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10157         .get_func_proto         = flow_dissector_func_proto,
10158         .is_valid_access        = flow_dissector_is_valid_access,
10159         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10160 };
10161
10162 const struct bpf_prog_ops flow_dissector_prog_ops = {
10163         .test_run               = bpf_prog_test_run_flow_dissector,
10164 };
10165
10166 int sk_detach_filter(struct sock *sk)
10167 {
10168         int ret = -ENOENT;
10169         struct sk_filter *filter;
10170
10171         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10172                 return -EPERM;
10173
10174         filter = rcu_dereference_protected(sk->sk_filter,
10175                                            lockdep_sock_is_held(sk));
10176         if (filter) {
10177                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10178                 sk_filter_uncharge(sk, filter);
10179                 ret = 0;
10180         }
10181
10182         return ret;
10183 }
10184 EXPORT_SYMBOL_GPL(sk_detach_filter);
10185
10186 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10187                   unsigned int len)
10188 {
10189         struct sock_fprog_kern *fprog;
10190         struct sk_filter *filter;
10191         int ret = 0;
10192
10193         lock_sock(sk);
10194         filter = rcu_dereference_protected(sk->sk_filter,
10195                                            lockdep_sock_is_held(sk));
10196         if (!filter)
10197                 goto out;
10198
10199         /* We're copying the filter that has been originally attached,
10200          * so no conversion/decode needed anymore. eBPF programs that
10201          * have no original program cannot be dumped through this.
10202          */
10203         ret = -EACCES;
10204         fprog = filter->prog->orig_prog;
10205         if (!fprog)
10206                 goto out;
10207
10208         ret = fprog->len;
10209         if (!len)
10210                 /* User space only enquires number of filter blocks. */
10211                 goto out;
10212
10213         ret = -EINVAL;
10214         if (len < fprog->len)
10215                 goto out;
10216
10217         ret = -EFAULT;
10218         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10219                 goto out;
10220
10221         /* Instead of bytes, the API requests to return the number
10222          * of filter blocks.
10223          */
10224         ret = fprog->len;
10225 out:
10226         release_sock(sk);
10227         return ret;
10228 }
10229
10230 #ifdef CONFIG_INET
10231 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10232                                     struct sock_reuseport *reuse,
10233                                     struct sock *sk, struct sk_buff *skb,
10234                                     struct sock *migrating_sk,
10235                                     u32 hash)
10236 {
10237         reuse_kern->skb = skb;
10238         reuse_kern->sk = sk;
10239         reuse_kern->selected_sk = NULL;
10240         reuse_kern->migrating_sk = migrating_sk;
10241         reuse_kern->data_end = skb->data + skb_headlen(skb);
10242         reuse_kern->hash = hash;
10243         reuse_kern->reuseport_id = reuse->reuseport_id;
10244         reuse_kern->bind_inany = reuse->bind_inany;
10245 }
10246
10247 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10248                                   struct bpf_prog *prog, struct sk_buff *skb,
10249                                   struct sock *migrating_sk,
10250                                   u32 hash)
10251 {
10252         struct sk_reuseport_kern reuse_kern;
10253         enum sk_action action;
10254
10255         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10256         action = bpf_prog_run(prog, &reuse_kern);
10257
10258         if (action == SK_PASS)
10259                 return reuse_kern.selected_sk;
10260         else
10261                 return ERR_PTR(-ECONNREFUSED);
10262 }
10263
10264 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10265            struct bpf_map *, map, void *, key, u32, flags)
10266 {
10267         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10268         struct sock_reuseport *reuse;
10269         struct sock *selected_sk;
10270
10271         selected_sk = map->ops->map_lookup_elem(map, key);
10272         if (!selected_sk)
10273                 return -ENOENT;
10274
10275         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10276         if (!reuse) {
10277                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10278                 if (sk_is_refcounted(selected_sk))
10279                         sock_put(selected_sk);
10280
10281                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10282                  * The only (!reuse) case here is - the sk has already been
10283                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10284                  *
10285                  * Other maps (e.g. sock_map) do not provide this guarantee and
10286                  * the sk may never be in the reuseport group to begin with.
10287                  */
10288                 return is_sockarray ? -ENOENT : -EINVAL;
10289         }
10290
10291         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10292                 struct sock *sk = reuse_kern->sk;
10293
10294                 if (sk->sk_protocol != selected_sk->sk_protocol)
10295                         return -EPROTOTYPE;
10296                 else if (sk->sk_family != selected_sk->sk_family)
10297                         return -EAFNOSUPPORT;
10298
10299                 /* Catch all. Likely bound to a different sockaddr. */
10300                 return -EBADFD;
10301         }
10302
10303         reuse_kern->selected_sk = selected_sk;
10304
10305         return 0;
10306 }
10307
10308 static const struct bpf_func_proto sk_select_reuseport_proto = {
10309         .func           = sk_select_reuseport,
10310         .gpl_only       = false,
10311         .ret_type       = RET_INTEGER,
10312         .arg1_type      = ARG_PTR_TO_CTX,
10313         .arg2_type      = ARG_CONST_MAP_PTR,
10314         .arg3_type      = ARG_PTR_TO_MAP_KEY,
10315         .arg4_type      = ARG_ANYTHING,
10316 };
10317
10318 BPF_CALL_4(sk_reuseport_load_bytes,
10319            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10320            void *, to, u32, len)
10321 {
10322         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10323 }
10324
10325 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10326         .func           = sk_reuseport_load_bytes,
10327         .gpl_only       = false,
10328         .ret_type       = RET_INTEGER,
10329         .arg1_type      = ARG_PTR_TO_CTX,
10330         .arg2_type      = ARG_ANYTHING,
10331         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10332         .arg4_type      = ARG_CONST_SIZE,
10333 };
10334
10335 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10336            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10337            void *, to, u32, len, u32, start_header)
10338 {
10339         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10340                                                len, start_header);
10341 }
10342
10343 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10344         .func           = sk_reuseport_load_bytes_relative,
10345         .gpl_only       = false,
10346         .ret_type       = RET_INTEGER,
10347         .arg1_type      = ARG_PTR_TO_CTX,
10348         .arg2_type      = ARG_ANYTHING,
10349         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10350         .arg4_type      = ARG_CONST_SIZE,
10351         .arg5_type      = ARG_ANYTHING,
10352 };
10353
10354 static const struct bpf_func_proto *
10355 sk_reuseport_func_proto(enum bpf_func_id func_id,
10356                         const struct bpf_prog *prog)
10357 {
10358         switch (func_id) {
10359         case BPF_FUNC_sk_select_reuseport:
10360                 return &sk_select_reuseport_proto;
10361         case BPF_FUNC_skb_load_bytes:
10362                 return &sk_reuseport_load_bytes_proto;
10363         case BPF_FUNC_skb_load_bytes_relative:
10364                 return &sk_reuseport_load_bytes_relative_proto;
10365         case BPF_FUNC_get_socket_cookie:
10366                 return &bpf_get_socket_ptr_cookie_proto;
10367         case BPF_FUNC_ktime_get_coarse_ns:
10368                 return &bpf_ktime_get_coarse_ns_proto;
10369         default:
10370                 return bpf_base_func_proto(func_id);
10371         }
10372 }
10373
10374 static bool
10375 sk_reuseport_is_valid_access(int off, int size,
10376                              enum bpf_access_type type,
10377                              const struct bpf_prog *prog,
10378                              struct bpf_insn_access_aux *info)
10379 {
10380         const u32 size_default = sizeof(__u32);
10381
10382         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10383             off % size || type != BPF_READ)
10384                 return false;
10385
10386         switch (off) {
10387         case offsetof(struct sk_reuseport_md, data):
10388                 info->reg_type = PTR_TO_PACKET;
10389                 return size == sizeof(__u64);
10390
10391         case offsetof(struct sk_reuseport_md, data_end):
10392                 info->reg_type = PTR_TO_PACKET_END;
10393                 return size == sizeof(__u64);
10394
10395         case offsetof(struct sk_reuseport_md, hash):
10396                 return size == size_default;
10397
10398         case offsetof(struct sk_reuseport_md, sk):
10399                 info->reg_type = PTR_TO_SOCKET;
10400                 return size == sizeof(__u64);
10401
10402         case offsetof(struct sk_reuseport_md, migrating_sk):
10403                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10404                 return size == sizeof(__u64);
10405
10406         /* Fields that allow narrowing */
10407         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10408                 if (size < sizeof_field(struct sk_buff, protocol))
10409                         return false;
10410                 fallthrough;
10411         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10412         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10413         case bpf_ctx_range(struct sk_reuseport_md, len):
10414                 bpf_ctx_record_field_size(info, size_default);
10415                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10416
10417         default:
10418                 return false;
10419         }
10420 }
10421
10422 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10423         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10424                               si->dst_reg, si->src_reg,                 \
10425                               bpf_target_off(struct sk_reuseport_kern, F, \
10426                                              sizeof_field(struct sk_reuseport_kern, F), \
10427                                              target_size));             \
10428         })
10429
10430 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10431         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10432                                     struct sk_buff,                     \
10433                                     skb,                                \
10434                                     SKB_FIELD)
10435
10436 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10437         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10438                                     struct sock,                        \
10439                                     sk,                                 \
10440                                     SK_FIELD)
10441
10442 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10443                                            const struct bpf_insn *si,
10444                                            struct bpf_insn *insn_buf,
10445                                            struct bpf_prog *prog,
10446                                            u32 *target_size)
10447 {
10448         struct bpf_insn *insn = insn_buf;
10449
10450         switch (si->off) {
10451         case offsetof(struct sk_reuseport_md, data):
10452                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10453                 break;
10454
10455         case offsetof(struct sk_reuseport_md, len):
10456                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10457                 break;
10458
10459         case offsetof(struct sk_reuseport_md, eth_protocol):
10460                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10461                 break;
10462
10463         case offsetof(struct sk_reuseport_md, ip_protocol):
10464                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10465                 break;
10466
10467         case offsetof(struct sk_reuseport_md, data_end):
10468                 SK_REUSEPORT_LOAD_FIELD(data_end);
10469                 break;
10470
10471         case offsetof(struct sk_reuseport_md, hash):
10472                 SK_REUSEPORT_LOAD_FIELD(hash);
10473                 break;
10474
10475         case offsetof(struct sk_reuseport_md, bind_inany):
10476                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10477                 break;
10478
10479         case offsetof(struct sk_reuseport_md, sk):
10480                 SK_REUSEPORT_LOAD_FIELD(sk);
10481                 break;
10482
10483         case offsetof(struct sk_reuseport_md, migrating_sk):
10484                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10485                 break;
10486         }
10487
10488         return insn - insn_buf;
10489 }
10490
10491 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10492         .get_func_proto         = sk_reuseport_func_proto,
10493         .is_valid_access        = sk_reuseport_is_valid_access,
10494         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
10495 };
10496
10497 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10498 };
10499
10500 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10501 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10502
10503 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10504            struct sock *, sk, u64, flags)
10505 {
10506         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10507                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10508                 return -EINVAL;
10509         if (unlikely(sk && sk_is_refcounted(sk)))
10510                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10511         if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10512                 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10513
10514         /* Check if socket is suitable for packet L3/L4 protocol */
10515         if (sk && sk->sk_protocol != ctx->protocol)
10516                 return -EPROTOTYPE;
10517         if (sk && sk->sk_family != ctx->family &&
10518             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10519                 return -EAFNOSUPPORT;
10520
10521         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10522                 return -EEXIST;
10523
10524         /* Select socket as lookup result */
10525         ctx->selected_sk = sk;
10526         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10527         return 0;
10528 }
10529
10530 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10531         .func           = bpf_sk_lookup_assign,
10532         .gpl_only       = false,
10533         .ret_type       = RET_INTEGER,
10534         .arg1_type      = ARG_PTR_TO_CTX,
10535         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
10536         .arg3_type      = ARG_ANYTHING,
10537 };
10538
10539 static const struct bpf_func_proto *
10540 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10541 {
10542         switch (func_id) {
10543         case BPF_FUNC_perf_event_output:
10544                 return &bpf_event_output_data_proto;
10545         case BPF_FUNC_sk_assign:
10546                 return &bpf_sk_lookup_assign_proto;
10547         case BPF_FUNC_sk_release:
10548                 return &bpf_sk_release_proto;
10549         default:
10550                 return bpf_sk_base_func_proto(func_id);
10551         }
10552 }
10553
10554 static bool sk_lookup_is_valid_access(int off, int size,
10555                                       enum bpf_access_type type,
10556                                       const struct bpf_prog *prog,
10557                                       struct bpf_insn_access_aux *info)
10558 {
10559         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10560                 return false;
10561         if (off % size != 0)
10562                 return false;
10563         if (type != BPF_READ)
10564                 return false;
10565
10566         switch (off) {
10567         case offsetof(struct bpf_sk_lookup, sk):
10568                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10569                 return size == sizeof(__u64);
10570
10571         case bpf_ctx_range(struct bpf_sk_lookup, family):
10572         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10573         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10574         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10575         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10576         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10577         case offsetof(struct bpf_sk_lookup, remote_port) ...
10578              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
10579         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10580                 bpf_ctx_record_field_size(info, sizeof(__u32));
10581                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10582
10583         default:
10584                 return false;
10585         }
10586 }
10587
10588 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10589                                         const struct bpf_insn *si,
10590                                         struct bpf_insn *insn_buf,
10591                                         struct bpf_prog *prog,
10592                                         u32 *target_size)
10593 {
10594         struct bpf_insn *insn = insn_buf;
10595
10596         switch (si->off) {
10597         case offsetof(struct bpf_sk_lookup, sk):
10598                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10599                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
10600                 break;
10601
10602         case offsetof(struct bpf_sk_lookup, family):
10603                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10604                                       bpf_target_off(struct bpf_sk_lookup_kern,
10605                                                      family, 2, target_size));
10606                 break;
10607
10608         case offsetof(struct bpf_sk_lookup, protocol):
10609                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10610                                       bpf_target_off(struct bpf_sk_lookup_kern,
10611                                                      protocol, 2, target_size));
10612                 break;
10613
10614         case offsetof(struct bpf_sk_lookup, remote_ip4):
10615                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10616                                       bpf_target_off(struct bpf_sk_lookup_kern,
10617                                                      v4.saddr, 4, target_size));
10618                 break;
10619
10620         case offsetof(struct bpf_sk_lookup, local_ip4):
10621                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10622                                       bpf_target_off(struct bpf_sk_lookup_kern,
10623                                                      v4.daddr, 4, target_size));
10624                 break;
10625
10626         case bpf_ctx_range_till(struct bpf_sk_lookup,
10627                                 remote_ip6[0], remote_ip6[3]): {
10628 #if IS_ENABLED(CONFIG_IPV6)
10629                 int off = si->off;
10630
10631                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10632                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10633                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10634                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10635                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10636                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10637 #else
10638                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10639 #endif
10640                 break;
10641         }
10642         case bpf_ctx_range_till(struct bpf_sk_lookup,
10643                                 local_ip6[0], local_ip6[3]): {
10644 #if IS_ENABLED(CONFIG_IPV6)
10645                 int off = si->off;
10646
10647                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10648                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10649                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10650                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10651                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10652                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10653 #else
10654                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10655 #endif
10656                 break;
10657         }
10658         case offsetof(struct bpf_sk_lookup, remote_port):
10659                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10660                                       bpf_target_off(struct bpf_sk_lookup_kern,
10661                                                      sport, 2, target_size));
10662                 break;
10663
10664         case offsetof(struct bpf_sk_lookup, local_port):
10665                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10666                                       bpf_target_off(struct bpf_sk_lookup_kern,
10667                                                      dport, 2, target_size));
10668                 break;
10669         }
10670
10671         return insn - insn_buf;
10672 }
10673
10674 const struct bpf_prog_ops sk_lookup_prog_ops = {
10675         .test_run = bpf_prog_test_run_sk_lookup,
10676 };
10677
10678 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10679         .get_func_proto         = sk_lookup_func_proto,
10680         .is_valid_access        = sk_lookup_is_valid_access,
10681         .convert_ctx_access     = sk_lookup_convert_ctx_access,
10682 };
10683
10684 #endif /* CONFIG_INET */
10685
10686 DEFINE_BPF_DISPATCHER(xdp)
10687
10688 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10689 {
10690         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10691 }
10692
10693 #ifdef CONFIG_DEBUG_INFO_BTF
10694 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10695 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10696 BTF_SOCK_TYPE_xxx
10697 #undef BTF_SOCK_TYPE
10698 #else
10699 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10700 #endif
10701
10702 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10703 {
10704         /* tcp6_sock type is not generated in dwarf and hence btf,
10705          * trigger an explicit type generation here.
10706          */
10707         BTF_TYPE_EMIT(struct tcp6_sock);
10708         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10709             sk->sk_family == AF_INET6)
10710                 return (unsigned long)sk;
10711
10712         return (unsigned long)NULL;
10713 }
10714
10715 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10716         .func                   = bpf_skc_to_tcp6_sock,
10717         .gpl_only               = false,
10718         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10719         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10720         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10721 };
10722
10723 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10724 {
10725         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10726                 return (unsigned long)sk;
10727
10728         return (unsigned long)NULL;
10729 }
10730
10731 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10732         .func                   = bpf_skc_to_tcp_sock,
10733         .gpl_only               = false,
10734         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10735         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10736         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10737 };
10738
10739 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10740 {
10741         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10742          * generated if CONFIG_INET=n. Trigger an explicit generation here.
10743          */
10744         BTF_TYPE_EMIT(struct inet_timewait_sock);
10745         BTF_TYPE_EMIT(struct tcp_timewait_sock);
10746
10747 #ifdef CONFIG_INET
10748         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10749                 return (unsigned long)sk;
10750 #endif
10751
10752 #if IS_BUILTIN(CONFIG_IPV6)
10753         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10754                 return (unsigned long)sk;
10755 #endif
10756
10757         return (unsigned long)NULL;
10758 }
10759
10760 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10761         .func                   = bpf_skc_to_tcp_timewait_sock,
10762         .gpl_only               = false,
10763         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10764         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10765         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10766 };
10767
10768 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10769 {
10770 #ifdef CONFIG_INET
10771         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10772                 return (unsigned long)sk;
10773 #endif
10774
10775 #if IS_BUILTIN(CONFIG_IPV6)
10776         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10777                 return (unsigned long)sk;
10778 #endif
10779
10780         return (unsigned long)NULL;
10781 }
10782
10783 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10784         .func                   = bpf_skc_to_tcp_request_sock,
10785         .gpl_only               = false,
10786         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10787         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10788         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10789 };
10790
10791 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10792 {
10793         /* udp6_sock type is not generated in dwarf and hence btf,
10794          * trigger an explicit type generation here.
10795          */
10796         BTF_TYPE_EMIT(struct udp6_sock);
10797         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10798             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10799                 return (unsigned long)sk;
10800
10801         return (unsigned long)NULL;
10802 }
10803
10804 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10805         .func                   = bpf_skc_to_udp6_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_UDP6],
10810 };
10811
10812 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10813 {
10814         return (unsigned long)sock_from_file(file);
10815 }
10816
10817 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10818 BTF_ID(struct, socket)
10819 BTF_ID(struct, file)
10820
10821 const struct bpf_func_proto bpf_sock_from_file_proto = {
10822         .func           = bpf_sock_from_file,
10823         .gpl_only       = false,
10824         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
10825         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
10826         .arg1_type      = ARG_PTR_TO_BTF_ID,
10827         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
10828 };
10829
10830 static const struct bpf_func_proto *
10831 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10832 {
10833         const struct bpf_func_proto *func;
10834
10835         switch (func_id) {
10836         case BPF_FUNC_skc_to_tcp6_sock:
10837                 func = &bpf_skc_to_tcp6_sock_proto;
10838                 break;
10839         case BPF_FUNC_skc_to_tcp_sock:
10840                 func = &bpf_skc_to_tcp_sock_proto;
10841                 break;
10842         case BPF_FUNC_skc_to_tcp_timewait_sock:
10843                 func = &bpf_skc_to_tcp_timewait_sock_proto;
10844                 break;
10845         case BPF_FUNC_skc_to_tcp_request_sock:
10846                 func = &bpf_skc_to_tcp_request_sock_proto;
10847                 break;
10848         case BPF_FUNC_skc_to_udp6_sock:
10849                 func = &bpf_skc_to_udp6_sock_proto;
10850                 break;
10851         case BPF_FUNC_ktime_get_coarse_ns:
10852                 return &bpf_ktime_get_coarse_ns_proto;
10853         default:
10854                 return bpf_base_func_proto(func_id);
10855         }
10856
10857         if (!perfmon_capable())
10858                 return NULL;
10859
10860         return func;
10861 }