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