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/bpf_verifier.h> 22 #include <linux/module.h> 23 #include <linux/types.h> 24 #include <linux/mm.h> 25 #include <linux/fcntl.h> 26 #include <linux/socket.h> 27 #include <linux/sock_diag.h> 28 #include <linux/in.h> 29 #include <linux/inet.h> 30 #include <linux/netdevice.h> 31 #include <linux/if_packet.h> 32 #include <linux/if_arp.h> 33 #include <linux/gfp.h> 34 #include <net/inet_common.h> 35 #include <net/ip.h> 36 #include <net/protocol.h> 37 #include <net/netlink.h> 38 #include <linux/skbuff.h> 39 #include <linux/skmsg.h> 40 #include <net/sock.h> 41 #include <net/flow_dissector.h> 42 #include <linux/errno.h> 43 #include <linux/timer.h> 44 #include <linux/uaccess.h> 45 #include <linux/unaligned.h> 46 #include <linux/filter.h> 47 #include <linux/ratelimit.h> 48 #include <linux/seccomp.h> 49 #include <linux/if_vlan.h> 50 #include <linux/bpf.h> 51 #include <linux/btf.h> 52 #include <net/sch_generic.h> 53 #include <net/cls_cgroup.h> 54 #include <net/dst_metadata.h> 55 #include <net/dst.h> 56 #include <net/sock_reuseport.h> 57 #include <net/busy_poll.h> 58 #include <net/tcp.h> 59 #include <net/gre.h> 60 #include <net/xfrm.h> 61 #include <net/udp.h> 62 #include <linux/bpf_trace.h> 63 #include <net/xdp_sock.h> 64 #include <linux/inetdevice.h> 65 #include <net/inet_hashtables.h> 66 #include <net/inet6_hashtables.h> 67 #include <net/ip_fib.h> 68 #include <net/nexthop.h> 69 #include <net/flow.h> 70 #include <net/arp.h> 71 #include <net/ipv6.h> 72 #include <net/net_namespace.h> 73 #include <linux/seg6_local.h> 74 #include <net/seg6.h> 75 #include <net/seg6_local.h> 76 #include <net/lwtunnel.h> 77 #include <net/bpf_sk_storage.h> 78 #include <net/transp_v6.h> 79 #include <linux/btf_ids.h> 80 #include <net/tls.h> 81 #include <net/xdp.h> 82 #include <net/mptcp.h> 83 #include <net/netfilter/nf_conntrack_bpf.h> 84 #include <net/netkit.h> 85 #include <linux/un.h> 86 #include <net/xdp_sock_drv.h> 87 #include <net/inet_dscp.h> 88 #include <linux/icmpv6.h> 89 #include <net/icmp.h> 90 #include <net/ip6_route.h> 91 92 #include "dev.h" 93 94 /* Keep the struct bpf_fib_lookup small so that it fits into a cacheline */ 95 static_assert(sizeof(struct bpf_fib_lookup) == 64, "struct bpf_fib_lookup size check"); 96 97 static const struct bpf_func_proto * 98 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog); 99 100 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len) 101 { 102 if (in_compat_syscall()) { 103 struct compat_sock_fprog f32; 104 105 if (len != sizeof(f32)) 106 return -EINVAL; 107 if (copy_from_sockptr(&f32, src, sizeof(f32))) 108 return -EFAULT; 109 memset(dst, 0, sizeof(*dst)); 110 dst->len = f32.len; 111 dst->filter = compat_ptr(f32.filter); 112 } else { 113 if (len != sizeof(*dst)) 114 return -EINVAL; 115 if (copy_from_sockptr(dst, src, sizeof(*dst))) 116 return -EFAULT; 117 } 118 119 return 0; 120 } 121 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user); 122 123 /** 124 * sk_filter_trim_cap - run a packet through a socket filter 125 * @sk: sock associated with &sk_buff 126 * @skb: buffer to filter 127 * @cap: limit on how short the eBPF program may trim the packet 128 * 129 * Run the eBPF program and then cut skb->data to correct size returned by 130 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 131 * than pkt_len we keep whole skb->data. This is the socket level 132 * wrapper to bpf_prog_run. It returns 0 if the packet should 133 * be accepted or a drop_reason if the packet should be tossed. 134 * 135 */ 136 enum skb_drop_reason 137 sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 138 { 139 enum skb_drop_reason drop_reason; 140 struct sk_filter *filter; 141 int err; 142 143 /* 144 * If the skb was allocated from pfmemalloc reserves, only 145 * allow SOCK_MEMALLOC sockets to use it as this socket is 146 * helping free memory 147 */ 148 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 149 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 150 return SKB_DROP_REASON_PFMEMALLOC; 151 } 152 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 153 if (err) 154 return SKB_DROP_REASON_SOCKET_FILTER; 155 156 err = security_sock_rcv_skb(sk, skb); 157 if (err) 158 return SKB_DROP_REASON_SECURITY_HOOK; 159 160 drop_reason = 0; 161 rcu_read_lock(); 162 filter = rcu_dereference(sk->sk_filter); 163 if (filter) { 164 struct sock *save_sk = skb->sk; 165 unsigned int pkt_len; 166 167 skb->sk = sk; 168 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 169 skb->sk = save_sk; 170 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 171 if (err) 172 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 173 } 174 rcu_read_unlock(); 175 176 return drop_reason; 177 } 178 EXPORT_SYMBOL(sk_filter_trim_cap); 179 180 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb) 181 { 182 return skb_get_poff(skb); 183 } 184 185 BPF_CALL_3(bpf_skb_get_nlattr, 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 = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 199 if (nla) 200 return (void *) nla - (void *) skb->data; 201 202 return 0; 203 } 204 205 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 206 { 207 struct nlattr *nla; 208 209 if (skb_is_nonlinear(skb)) 210 return 0; 211 212 if (skb->len < sizeof(struct nlattr)) 213 return 0; 214 215 if (a > skb->len - sizeof(struct nlattr)) 216 return 0; 217 218 nla = (struct nlattr *) &skb->data[a]; 219 if (!nla_ok(nla, skb->len - a)) 220 return 0; 221 222 nla = nla_find_nested(nla, x); 223 if (nla) 224 return (void *) nla - (void *) skb->data; 225 226 return 0; 227 } 228 229 static int bpf_skb_load_helper_convert_offset(const struct sk_buff *skb, int offset) 230 { 231 if (likely(offset >= 0)) 232 return offset; 233 234 if (offset >= SKF_NET_OFF) 235 return offset - SKF_NET_OFF + skb_network_offset(skb); 236 237 if (offset >= SKF_LL_OFF && skb_mac_header_was_set(skb)) 238 return offset - SKF_LL_OFF + skb_mac_offset(skb); 239 240 return INT_MIN; 241 } 242 243 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *, 244 data, int, headlen, int, offset) 245 { 246 u8 tmp; 247 const int len = sizeof(tmp); 248 249 offset = bpf_skb_load_helper_convert_offset(skb, offset); 250 if (offset == INT_MIN) 251 return -EFAULT; 252 253 if (headlen - offset >= len) 254 return *(u8 *)(data + offset); 255 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 256 return tmp; 257 else 258 return -EFAULT; 259 } 260 261 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb, 262 int, offset) 263 { 264 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len, 265 offset); 266 } 267 268 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *, 269 data, int, headlen, int, offset) 270 { 271 __be16 tmp; 272 const int len = sizeof(tmp); 273 274 offset = bpf_skb_load_helper_convert_offset(skb, offset); 275 if (offset == INT_MIN) 276 return -EFAULT; 277 278 if (headlen - offset >= len) 279 return get_unaligned_be16(data + offset); 280 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 281 return be16_to_cpu(tmp); 282 else 283 return -EFAULT; 284 } 285 286 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb, 287 int, offset) 288 { 289 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len, 290 offset); 291 } 292 293 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *, 294 data, int, headlen, int, offset) 295 { 296 __be32 tmp; 297 const int len = sizeof(tmp); 298 299 offset = bpf_skb_load_helper_convert_offset(skb, offset); 300 if (offset == INT_MIN) 301 return -EFAULT; 302 303 if (headlen - offset >= len) 304 return get_unaligned_be32(data + offset); 305 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 306 return be32_to_cpu(tmp); 307 else 308 return -EFAULT; 309 } 310 311 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb, 312 int, offset) 313 { 314 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len, 315 offset); 316 } 317 318 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 319 struct bpf_insn *insn_buf) 320 { 321 struct bpf_insn *insn = insn_buf; 322 323 switch (skb_field) { 324 case SKF_AD_MARK: 325 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4); 326 327 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 328 offsetof(struct sk_buff, mark)); 329 break; 330 331 case SKF_AD_PKTTYPE: 332 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET); 333 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 334 #ifdef __BIG_ENDIAN_BITFIELD 335 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 336 #endif 337 break; 338 339 case SKF_AD_QUEUE: 340 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2); 341 342 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 343 offsetof(struct sk_buff, queue_mapping)); 344 break; 345 346 case SKF_AD_VLAN_TAG: 347 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2); 348 349 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 350 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 351 offsetof(struct sk_buff, vlan_tci)); 352 break; 353 case SKF_AD_VLAN_TAG_PRESENT: 354 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4); 355 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 356 offsetof(struct sk_buff, vlan_all)); 357 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1); 358 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1); 359 break; 360 } 361 362 return insn - insn_buf; 363 } 364 365 static bool convert_bpf_extensions(struct sock_filter *fp, 366 struct bpf_insn **insnp) 367 { 368 struct bpf_insn *insn = *insnp; 369 u32 cnt; 370 371 switch (fp->k) { 372 case SKF_AD_OFF + SKF_AD_PROTOCOL: 373 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2); 374 375 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 376 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 377 offsetof(struct sk_buff, protocol)); 378 /* A = ntohs(A) [emitting a nop or swap16] */ 379 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 380 break; 381 382 case SKF_AD_OFF + SKF_AD_PKTTYPE: 383 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 384 insn += cnt - 1; 385 break; 386 387 case SKF_AD_OFF + SKF_AD_IFINDEX: 388 case SKF_AD_OFF + SKF_AD_HATYPE: 389 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4); 390 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2); 391 392 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 393 BPF_REG_TMP, BPF_REG_CTX, 394 offsetof(struct sk_buff, dev)); 395 /* if (tmp != 0) goto pc + 1 */ 396 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 397 *insn++ = BPF_EXIT_INSN(); 398 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 399 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 400 offsetof(struct net_device, ifindex)); 401 else 402 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 403 offsetof(struct net_device, type)); 404 break; 405 406 case SKF_AD_OFF + SKF_AD_MARK: 407 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 408 insn += cnt - 1; 409 break; 410 411 case SKF_AD_OFF + SKF_AD_RXHASH: 412 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4); 413 414 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 415 offsetof(struct sk_buff, hash)); 416 break; 417 418 case SKF_AD_OFF + SKF_AD_QUEUE: 419 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 420 insn += cnt - 1; 421 break; 422 423 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 424 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 425 BPF_REG_A, BPF_REG_CTX, insn); 426 insn += cnt - 1; 427 break; 428 429 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 430 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 431 BPF_REG_A, BPF_REG_CTX, insn); 432 insn += cnt - 1; 433 break; 434 435 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 436 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2); 437 438 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 439 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 440 offsetof(struct sk_buff, vlan_proto)); 441 /* A = ntohs(A) [emitting a nop or swap16] */ 442 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 443 break; 444 445 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 446 case SKF_AD_OFF + SKF_AD_NLATTR: 447 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 448 case SKF_AD_OFF + SKF_AD_CPU: 449 case SKF_AD_OFF + SKF_AD_RANDOM: 450 /* arg1 = CTX */ 451 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 452 /* arg2 = A */ 453 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 454 /* arg3 = X */ 455 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 456 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 457 switch (fp->k) { 458 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 459 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset); 460 break; 461 case SKF_AD_OFF + SKF_AD_NLATTR: 462 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr); 463 break; 464 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 465 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest); 466 break; 467 case SKF_AD_OFF + SKF_AD_CPU: 468 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id); 469 break; 470 case SKF_AD_OFF + SKF_AD_RANDOM: 471 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 472 bpf_user_rnd_init_once(); 473 break; 474 } 475 break; 476 477 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 478 /* A ^= X */ 479 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 480 break; 481 482 default: 483 /* This is just a dummy call to avoid letting the compiler 484 * evict __bpf_call_base() as an optimization. Placed here 485 * where no-one bothers. 486 */ 487 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 488 return false; 489 } 490 491 *insnp = insn; 492 return true; 493 } 494 495 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp) 496 { 497 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS); 498 int size = bpf_size_to_bytes(BPF_SIZE(fp->code)); 499 bool endian = BPF_SIZE(fp->code) == BPF_H || 500 BPF_SIZE(fp->code) == BPF_W; 501 bool indirect = BPF_MODE(fp->code) == BPF_IND; 502 const int ip_align = NET_IP_ALIGN; 503 struct bpf_insn *insn = *insnp; 504 int offset = fp->k; 505 506 if (!indirect && 507 ((unaligned_ok && offset >= 0) || 508 (!unaligned_ok && offset >= 0 && 509 offset + ip_align >= 0 && 510 (offset + ip_align) % size == 0))) { 511 bool ldx_off_ok = offset <= S16_MAX; 512 513 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H); 514 if (offset) 515 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset); 516 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, 517 size, 2 + endian + (!ldx_off_ok * 2)); 518 if (ldx_off_ok) { 519 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 520 BPF_REG_D, offset); 521 } else { 522 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D); 523 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset); 524 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 525 BPF_REG_TMP, 0); 526 } 527 if (endian) 528 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8); 529 *insn++ = BPF_JMP_A(8); 530 } 531 532 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 533 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D); 534 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H); 535 if (!indirect) { 536 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset); 537 } else { 538 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X); 539 if (fp->k) 540 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset); 541 } 542 543 switch (BPF_SIZE(fp->code)) { 544 case BPF_B: 545 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8); 546 break; 547 case BPF_H: 548 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16); 549 break; 550 case BPF_W: 551 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32); 552 break; 553 default: 554 return false; 555 } 556 557 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2); 558 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 559 *insn = BPF_EXIT_INSN(); 560 561 *insnp = insn; 562 return true; 563 } 564 565 /** 566 * bpf_convert_filter - convert filter program 567 * @prog: the user passed filter program 568 * @len: the length of the user passed filter program 569 * @new_prog: allocated 'struct bpf_prog' or NULL 570 * @new_len: pointer to store length of converted program 571 * @seen_ld_abs: bool whether we've seen ld_abs/ind 572 * 573 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 574 * style extended BPF (eBPF). 575 * Conversion workflow: 576 * 577 * 1) First pass for calculating the new program length: 578 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs) 579 * 580 * 2) 2nd pass to remap in two passes: 1st pass finds new 581 * jump offsets, 2nd pass remapping: 582 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs) 583 */ 584 static int bpf_convert_filter(struct sock_filter *prog, int len, 585 struct bpf_prog *new_prog, int *new_len, 586 bool *seen_ld_abs) 587 { 588 int new_flen = 0, pass = 0, target, i, stack_off; 589 struct bpf_insn *new_insn, *first_insn = NULL; 590 struct sock_filter *fp; 591 int *addrs = NULL; 592 u8 bpf_src; 593 594 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 595 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 596 597 if (len <= 0 || len > BPF_MAXINSNS) 598 return -EINVAL; 599 600 if (new_prog) { 601 first_insn = new_prog->insnsi; 602 addrs = kzalloc_objs(*addrs, len, GFP_KERNEL | __GFP_NOWARN); 603 if (!addrs) 604 return -ENOMEM; 605 } 606 607 do_pass: 608 new_insn = first_insn; 609 fp = prog; 610 611 /* Classic BPF related prologue emission. */ 612 if (new_prog) { 613 /* Classic BPF expects A and X to be reset first. These need 614 * to be guaranteed to be the first two instructions. 615 */ 616 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 617 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 618 619 /* All programs must keep CTX in callee saved BPF_REG_CTX. 620 * In eBPF case it's done by the compiler, here we need to 621 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 622 */ 623 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 624 if (*seen_ld_abs) { 625 /* For packet access in classic BPF, cache skb->data 626 * in callee-saved BPF R8 and skb->len - skb->data_len 627 * (headlen) in BPF R9. Since classic BPF is read-only 628 * on CTX, we only need to cache it once. 629 */ 630 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 631 BPF_REG_D, BPF_REG_CTX, 632 offsetof(struct sk_buff, data)); 633 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX, 634 offsetof(struct sk_buff, len)); 635 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX, 636 offsetof(struct sk_buff, data_len)); 637 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP); 638 } 639 } else { 640 new_insn += 3; 641 } 642 643 for (i = 0; i < len; fp++, i++) { 644 struct bpf_insn tmp_insns[32] = { }; 645 struct bpf_insn *insn = tmp_insns; 646 647 if (addrs) 648 addrs[i] = new_insn - first_insn; 649 650 switch (fp->code) { 651 /* All arithmetic insns and skb loads map as-is. */ 652 case BPF_ALU | BPF_ADD | BPF_X: 653 case BPF_ALU | BPF_ADD | BPF_K: 654 case BPF_ALU | BPF_SUB | BPF_X: 655 case BPF_ALU | BPF_SUB | BPF_K: 656 case BPF_ALU | BPF_AND | BPF_X: 657 case BPF_ALU | BPF_AND | BPF_K: 658 case BPF_ALU | BPF_OR | BPF_X: 659 case BPF_ALU | BPF_OR | BPF_K: 660 case BPF_ALU | BPF_LSH | BPF_X: 661 case BPF_ALU | BPF_LSH | BPF_K: 662 case BPF_ALU | BPF_RSH | BPF_X: 663 case BPF_ALU | BPF_RSH | BPF_K: 664 case BPF_ALU | BPF_XOR | BPF_X: 665 case BPF_ALU | BPF_XOR | BPF_K: 666 case BPF_ALU | BPF_MUL | BPF_X: 667 case BPF_ALU | BPF_MUL | BPF_K: 668 case BPF_ALU | BPF_DIV | BPF_X: 669 case BPF_ALU | BPF_DIV | BPF_K: 670 case BPF_ALU | BPF_MOD | BPF_X: 671 case BPF_ALU | BPF_MOD | BPF_K: 672 case BPF_ALU | BPF_NEG: 673 case BPF_LD | BPF_ABS | BPF_W: 674 case BPF_LD | BPF_ABS | BPF_H: 675 case BPF_LD | BPF_ABS | BPF_B: 676 case BPF_LD | BPF_IND | BPF_W: 677 case BPF_LD | BPF_IND | BPF_H: 678 case BPF_LD | BPF_IND | BPF_B: 679 /* Check for overloaded BPF extension and 680 * directly convert it if found, otherwise 681 * just move on with mapping. 682 */ 683 if (BPF_CLASS(fp->code) == BPF_LD && 684 BPF_MODE(fp->code) == BPF_ABS && 685 convert_bpf_extensions(fp, &insn)) 686 break; 687 if (BPF_CLASS(fp->code) == BPF_LD && 688 convert_bpf_ld_abs(fp, &insn)) { 689 *seen_ld_abs = true; 690 break; 691 } 692 693 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) || 694 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) { 695 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X); 696 /* Error with exception code on div/mod by 0. 697 * For cBPF programs, this was always return 0. 698 */ 699 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2); 700 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 701 *insn++ = BPF_EXIT_INSN(); 702 } 703 704 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 705 break; 706 707 /* Jump transformation cannot use BPF block macros 708 * everywhere as offset calculation and target updates 709 * require a bit more work than the rest, i.e. jump 710 * opcodes map as-is, but offsets need adjustment. 711 */ 712 713 #define BPF_EMIT_JMP \ 714 do { \ 715 const s32 off_min = S16_MIN, off_max = S16_MAX; \ 716 s32 off; \ 717 \ 718 if (target >= len || target < 0) \ 719 goto err; \ 720 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 721 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 722 off -= insn - tmp_insns; \ 723 /* Reject anything not fitting into insn->off. */ \ 724 if (off < off_min || off > off_max) \ 725 goto err; \ 726 insn->off = off; \ 727 } while (0) 728 729 case BPF_JMP | BPF_JA: 730 target = i + fp->k + 1; 731 insn->code = fp->code; 732 BPF_EMIT_JMP; 733 break; 734 735 case BPF_JMP | BPF_JEQ | BPF_K: 736 case BPF_JMP | BPF_JEQ | BPF_X: 737 case BPF_JMP | BPF_JSET | BPF_K: 738 case BPF_JMP | BPF_JSET | BPF_X: 739 case BPF_JMP | BPF_JGT | BPF_K: 740 case BPF_JMP | BPF_JGT | BPF_X: 741 case BPF_JMP | BPF_JGE | BPF_K: 742 case BPF_JMP | BPF_JGE | BPF_X: 743 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 744 /* BPF immediates are signed, zero extend 745 * immediate into tmp register and use it 746 * in compare insn. 747 */ 748 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 749 750 insn->dst_reg = BPF_REG_A; 751 insn->src_reg = BPF_REG_TMP; 752 bpf_src = BPF_X; 753 } else { 754 insn->dst_reg = BPF_REG_A; 755 insn->imm = fp->k; 756 bpf_src = BPF_SRC(fp->code); 757 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 758 } 759 760 /* Common case where 'jump_false' is next insn. */ 761 if (fp->jf == 0) { 762 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 763 target = i + fp->jt + 1; 764 BPF_EMIT_JMP; 765 break; 766 } 767 768 /* Convert some jumps when 'jump_true' is next insn. */ 769 if (fp->jt == 0) { 770 switch (BPF_OP(fp->code)) { 771 case BPF_JEQ: 772 insn->code = BPF_JMP | BPF_JNE | bpf_src; 773 break; 774 case BPF_JGT: 775 insn->code = BPF_JMP | BPF_JLE | bpf_src; 776 break; 777 case BPF_JGE: 778 insn->code = BPF_JMP | BPF_JLT | bpf_src; 779 break; 780 default: 781 goto jmp_rest; 782 } 783 784 target = i + fp->jf + 1; 785 BPF_EMIT_JMP; 786 break; 787 } 788 jmp_rest: 789 /* Other jumps are mapped into two insns: Jxx and JA. */ 790 target = i + fp->jt + 1; 791 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 792 BPF_EMIT_JMP; 793 insn++; 794 795 insn->code = BPF_JMP | BPF_JA; 796 target = i + fp->jf + 1; 797 BPF_EMIT_JMP; 798 break; 799 800 /* ldxb 4 * ([14] & 0xf) is remapped into 6 insns. */ 801 case BPF_LDX | BPF_MSH | BPF_B: { 802 struct sock_filter tmp = { 803 .code = BPF_LD | BPF_ABS | BPF_B, 804 .k = fp->k, 805 }; 806 807 *seen_ld_abs = true; 808 809 /* X = A */ 810 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 811 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 812 convert_bpf_ld_abs(&tmp, &insn); 813 insn++; 814 /* A &= 0xf */ 815 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 816 /* A <<= 2 */ 817 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 818 /* tmp = X */ 819 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X); 820 /* X = A */ 821 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 822 /* A = tmp */ 823 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 824 break; 825 } 826 /* RET_K is remapped into 2 insns. RET_A case doesn't need an 827 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 828 */ 829 case BPF_RET | BPF_A: 830 case BPF_RET | BPF_K: 831 if (BPF_RVAL(fp->code) == BPF_K) 832 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 833 0, fp->k); 834 *insn = BPF_EXIT_INSN(); 835 break; 836 837 /* Store to stack. */ 838 case BPF_ST: 839 case BPF_STX: 840 stack_off = fp->k * 4 + 4; 841 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 842 BPF_ST ? BPF_REG_A : BPF_REG_X, 843 -stack_off); 844 /* check_load_and_stores() verifies that classic BPF can 845 * load from stack only after write, so tracking 846 * stack_depth for ST|STX insns is enough 847 */ 848 if (new_prog && new_prog->aux->stack_depth < stack_off) 849 new_prog->aux->stack_depth = stack_off; 850 break; 851 852 /* Load from stack. */ 853 case BPF_LD | BPF_MEM: 854 case BPF_LDX | BPF_MEM: 855 stack_off = fp->k * 4 + 4; 856 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 857 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 858 -stack_off); 859 break; 860 861 /* A = K or X = K */ 862 case BPF_LD | BPF_IMM: 863 case BPF_LDX | BPF_IMM: 864 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 865 BPF_REG_A : BPF_REG_X, fp->k); 866 break; 867 868 /* X = A */ 869 case BPF_MISC | BPF_TAX: 870 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 871 break; 872 873 /* A = X */ 874 case BPF_MISC | BPF_TXA: 875 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 876 break; 877 878 /* A = skb->len or X = skb->len */ 879 case BPF_LD | BPF_W | BPF_LEN: 880 case BPF_LDX | BPF_W | BPF_LEN: 881 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 882 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 883 offsetof(struct sk_buff, len)); 884 break; 885 886 /* Access seccomp_data fields. */ 887 case BPF_LDX | BPF_ABS | BPF_W: 888 /* A = *(u32 *) (ctx + K) */ 889 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 890 break; 891 892 /* Unknown instruction. */ 893 default: 894 goto err; 895 } 896 897 insn++; 898 if (new_prog) 899 memcpy(new_insn, tmp_insns, 900 sizeof(*insn) * (insn - tmp_insns)); 901 new_insn += insn - tmp_insns; 902 } 903 904 if (!new_prog) { 905 /* Only calculating new length. */ 906 *new_len = new_insn - first_insn; 907 if (*seen_ld_abs) 908 *new_len += 4; /* Prologue bits. */ 909 return 0; 910 } 911 912 pass++; 913 if (new_flen != new_insn - first_insn) { 914 new_flen = new_insn - first_insn; 915 if (pass > 2) 916 goto err; 917 goto do_pass; 918 } 919 920 kfree(addrs); 921 BUG_ON(*new_len != new_flen); 922 return 0; 923 err: 924 kfree(addrs); 925 return -EINVAL; 926 } 927 928 /* Security: 929 * 930 * As we dont want to clear mem[] array for each packet going through 931 * __bpf_prog_run(), we check that filter loaded by user never try to read 932 * a cell if not previously written, and we check all branches to be sure 933 * a malicious user doesn't try to abuse us. 934 */ 935 static int check_load_and_stores(const struct sock_filter *filter, int flen) 936 { 937 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 938 int pc, ret = 0; 939 940 BUILD_BUG_ON(BPF_MEMWORDS > 16); 941 942 masks = kmalloc_objs(*masks, flen); 943 if (!masks) 944 return -ENOMEM; 945 946 memset(masks, 0xff, flen * sizeof(*masks)); 947 948 for (pc = 0; pc < flen; pc++) { 949 memvalid &= masks[pc]; 950 951 switch (filter[pc].code) { 952 case BPF_ST: 953 case BPF_STX: 954 memvalid |= (1 << filter[pc].k); 955 break; 956 case BPF_LD | BPF_MEM: 957 case BPF_LDX | BPF_MEM: 958 if (!(memvalid & (1 << filter[pc].k))) { 959 ret = -EINVAL; 960 goto error; 961 } 962 break; 963 case BPF_JMP | BPF_JA: 964 /* A jump must set masks on target */ 965 masks[pc + 1 + filter[pc].k] &= memvalid; 966 memvalid = ~0; 967 break; 968 case BPF_JMP | BPF_JEQ | BPF_K: 969 case BPF_JMP | BPF_JEQ | BPF_X: 970 case BPF_JMP | BPF_JGE | BPF_K: 971 case BPF_JMP | BPF_JGE | BPF_X: 972 case BPF_JMP | BPF_JGT | BPF_K: 973 case BPF_JMP | BPF_JGT | BPF_X: 974 case BPF_JMP | BPF_JSET | BPF_K: 975 case BPF_JMP | BPF_JSET | BPF_X: 976 /* A jump must set masks on targets */ 977 masks[pc + 1 + filter[pc].jt] &= memvalid; 978 masks[pc + 1 + filter[pc].jf] &= memvalid; 979 memvalid = ~0; 980 break; 981 } 982 } 983 error: 984 kfree(masks); 985 return ret; 986 } 987 988 static bool chk_code_allowed(u16 code_to_probe) 989 { 990 static const bool codes[] = { 991 /* 32 bit ALU operations */ 992 [BPF_ALU | BPF_ADD | BPF_K] = true, 993 [BPF_ALU | BPF_ADD | BPF_X] = true, 994 [BPF_ALU | BPF_SUB | BPF_K] = true, 995 [BPF_ALU | BPF_SUB | BPF_X] = true, 996 [BPF_ALU | BPF_MUL | BPF_K] = true, 997 [BPF_ALU | BPF_MUL | BPF_X] = true, 998 [BPF_ALU | BPF_DIV | BPF_K] = true, 999 [BPF_ALU | BPF_DIV | BPF_X] = true, 1000 [BPF_ALU | BPF_MOD | BPF_K] = true, 1001 [BPF_ALU | BPF_MOD | BPF_X] = true, 1002 [BPF_ALU | BPF_AND | BPF_K] = true, 1003 [BPF_ALU | BPF_AND | BPF_X] = true, 1004 [BPF_ALU | BPF_OR | BPF_K] = true, 1005 [BPF_ALU | BPF_OR | BPF_X] = true, 1006 [BPF_ALU | BPF_XOR | BPF_K] = true, 1007 [BPF_ALU | BPF_XOR | BPF_X] = true, 1008 [BPF_ALU | BPF_LSH | BPF_K] = true, 1009 [BPF_ALU | BPF_LSH | BPF_X] = true, 1010 [BPF_ALU | BPF_RSH | BPF_K] = true, 1011 [BPF_ALU | BPF_RSH | BPF_X] = true, 1012 [BPF_ALU | BPF_NEG] = true, 1013 /* Load instructions */ 1014 [BPF_LD | BPF_W | BPF_ABS] = true, 1015 [BPF_LD | BPF_H | BPF_ABS] = true, 1016 [BPF_LD | BPF_B | BPF_ABS] = true, 1017 [BPF_LD | BPF_W | BPF_LEN] = true, 1018 [BPF_LD | BPF_W | BPF_IND] = true, 1019 [BPF_LD | BPF_H | BPF_IND] = true, 1020 [BPF_LD | BPF_B | BPF_IND] = true, 1021 [BPF_LD | BPF_IMM] = true, 1022 [BPF_LD | BPF_MEM] = true, 1023 [BPF_LDX | BPF_W | BPF_LEN] = true, 1024 [BPF_LDX | BPF_B | BPF_MSH] = true, 1025 [BPF_LDX | BPF_IMM] = true, 1026 [BPF_LDX | BPF_MEM] = true, 1027 /* Store instructions */ 1028 [BPF_ST] = true, 1029 [BPF_STX] = true, 1030 /* Misc instructions */ 1031 [BPF_MISC | BPF_TAX] = true, 1032 [BPF_MISC | BPF_TXA] = true, 1033 /* Return instructions */ 1034 [BPF_RET | BPF_K] = true, 1035 [BPF_RET | BPF_A] = true, 1036 /* Jump instructions */ 1037 [BPF_JMP | BPF_JA] = true, 1038 [BPF_JMP | BPF_JEQ | BPF_K] = true, 1039 [BPF_JMP | BPF_JEQ | BPF_X] = true, 1040 [BPF_JMP | BPF_JGE | BPF_K] = true, 1041 [BPF_JMP | BPF_JGE | BPF_X] = true, 1042 [BPF_JMP | BPF_JGT | BPF_K] = true, 1043 [BPF_JMP | BPF_JGT | BPF_X] = true, 1044 [BPF_JMP | BPF_JSET | BPF_K] = true, 1045 [BPF_JMP | BPF_JSET | BPF_X] = true, 1046 }; 1047 1048 if (code_to_probe >= ARRAY_SIZE(codes)) 1049 return false; 1050 1051 return codes[code_to_probe]; 1052 } 1053 1054 static bool bpf_check_basics_ok(const struct sock_filter *filter, 1055 unsigned int flen) 1056 { 1057 if (filter == NULL) 1058 return false; 1059 if (flen == 0 || flen > BPF_MAXINSNS) 1060 return false; 1061 1062 return true; 1063 } 1064 1065 /** 1066 * bpf_check_classic - verify socket filter code 1067 * @filter: filter to verify 1068 * @flen: length of filter 1069 * 1070 * Check the user's filter code. If we let some ugly 1071 * filter code slip through kaboom! The filter must contain 1072 * no references or jumps that are out of range, no illegal 1073 * instructions, and must end with a RET instruction. 1074 * 1075 * All jumps are forward as they are not signed. 1076 * 1077 * Returns 0 if the rule set is legal or -EINVAL if not. 1078 */ 1079 static int bpf_check_classic(const struct sock_filter *filter, 1080 unsigned int flen) 1081 { 1082 bool anc_found; 1083 int pc; 1084 1085 /* Check the filter code now */ 1086 for (pc = 0; pc < flen; pc++) { 1087 const struct sock_filter *ftest = &filter[pc]; 1088 1089 /* May we actually operate on this code? */ 1090 if (!chk_code_allowed(ftest->code)) 1091 return -EINVAL; 1092 1093 /* Some instructions need special checks */ 1094 switch (ftest->code) { 1095 case BPF_ALU | BPF_DIV | BPF_K: 1096 case BPF_ALU | BPF_MOD | BPF_K: 1097 /* Check for division by zero */ 1098 if (ftest->k == 0) 1099 return -EINVAL; 1100 break; 1101 case BPF_ALU | BPF_LSH | BPF_K: 1102 case BPF_ALU | BPF_RSH | BPF_K: 1103 if (ftest->k >= 32) 1104 return -EINVAL; 1105 break; 1106 case BPF_LD | BPF_MEM: 1107 case BPF_LDX | BPF_MEM: 1108 case BPF_ST: 1109 case BPF_STX: 1110 /* Check for invalid memory addresses */ 1111 if (ftest->k >= BPF_MEMWORDS) 1112 return -EINVAL; 1113 break; 1114 case BPF_JMP | BPF_JA: 1115 /* Note, the large ftest->k might cause loops. 1116 * Compare this with conditional jumps below, 1117 * where offsets are limited. --ANK (981016) 1118 */ 1119 if (ftest->k >= (unsigned int)(flen - pc - 1)) 1120 return -EINVAL; 1121 break; 1122 case BPF_JMP | BPF_JEQ | BPF_K: 1123 case BPF_JMP | BPF_JEQ | BPF_X: 1124 case BPF_JMP | BPF_JGE | BPF_K: 1125 case BPF_JMP | BPF_JGE | BPF_X: 1126 case BPF_JMP | BPF_JGT | BPF_K: 1127 case BPF_JMP | BPF_JGT | BPF_X: 1128 case BPF_JMP | BPF_JSET | BPF_K: 1129 case BPF_JMP | BPF_JSET | BPF_X: 1130 /* Both conditionals must be safe */ 1131 if (pc + ftest->jt + 1 >= flen || 1132 pc + ftest->jf + 1 >= flen) 1133 return -EINVAL; 1134 break; 1135 case BPF_LD | BPF_W | BPF_ABS: 1136 case BPF_LD | BPF_H | BPF_ABS: 1137 case BPF_LD | BPF_B | BPF_ABS: 1138 anc_found = false; 1139 if (bpf_anc_helper(ftest) & BPF_ANC) 1140 anc_found = true; 1141 /* Ancillary operation unknown or unsupported */ 1142 if (anc_found == false && ftest->k >= SKF_AD_OFF) 1143 return -EINVAL; 1144 } 1145 } 1146 1147 /* Last instruction must be a RET code */ 1148 switch (filter[flen - 1].code) { 1149 case BPF_RET | BPF_K: 1150 case BPF_RET | BPF_A: 1151 return check_load_and_stores(filter, flen); 1152 } 1153 1154 return -EINVAL; 1155 } 1156 1157 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 1158 const struct sock_fprog *fprog) 1159 { 1160 unsigned int fsize = bpf_classic_proglen(fprog); 1161 struct sock_fprog_kern *fkprog; 1162 1163 fp->orig_prog = kmalloc_obj(*fkprog); 1164 if (!fp->orig_prog) 1165 return -ENOMEM; 1166 1167 fkprog = fp->orig_prog; 1168 fkprog->len = fprog->len; 1169 1170 fkprog->filter = kmemdup(fp->insns, fsize, 1171 GFP_KERNEL | __GFP_NOWARN); 1172 if (!fkprog->filter) { 1173 kfree(fp->orig_prog); 1174 return -ENOMEM; 1175 } 1176 1177 return 0; 1178 } 1179 1180 static void bpf_release_orig_filter(struct bpf_prog *fp) 1181 { 1182 struct sock_fprog_kern *fprog = fp->orig_prog; 1183 1184 if (fprog) { 1185 kfree(fprog->filter); 1186 kfree(fprog); 1187 } 1188 } 1189 1190 static void __bpf_prog_release(struct bpf_prog *prog) 1191 { 1192 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 1193 bpf_prog_put(prog); 1194 } else { 1195 bpf_release_orig_filter(prog); 1196 bpf_prog_free(prog); 1197 } 1198 } 1199 1200 static void __sk_filter_release(struct sk_filter *fp) 1201 { 1202 __bpf_prog_release(fp->prog); 1203 kfree(fp); 1204 } 1205 1206 /** 1207 * sk_filter_release_rcu - Release a socket filter by rcu_head 1208 * @rcu: rcu_head that contains the sk_filter to free 1209 */ 1210 static void sk_filter_release_rcu(struct rcu_head *rcu) 1211 { 1212 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 1213 1214 __sk_filter_release(fp); 1215 } 1216 1217 /** 1218 * sk_filter_release - release a socket filter 1219 * @fp: filter to remove 1220 * 1221 * Remove a filter from a socket and release its resources. 1222 */ 1223 static void sk_filter_release(struct sk_filter *fp) 1224 { 1225 if (refcount_dec_and_test(&fp->refcnt)) 1226 call_rcu(&fp->rcu, sk_filter_release_rcu); 1227 } 1228 1229 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1230 { 1231 u32 filter_size = bpf_prog_size(fp->prog->len); 1232 1233 atomic_sub(filter_size, &sk->sk_omem_alloc); 1234 sk_filter_release(fp); 1235 } 1236 1237 /* try to charge the socket memory if there is space available 1238 * return true on success 1239 */ 1240 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1241 { 1242 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 1243 u32 filter_size = bpf_prog_size(fp->prog->len); 1244 1245 /* same check as in sock_kmalloc() */ 1246 if (filter_size <= optmem_max && 1247 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) { 1248 atomic_add(filter_size, &sk->sk_omem_alloc); 1249 return true; 1250 } 1251 return false; 1252 } 1253 1254 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1255 { 1256 if (!refcount_inc_not_zero(&fp->refcnt)) 1257 return false; 1258 1259 if (!__sk_filter_charge(sk, fp)) { 1260 sk_filter_release(fp); 1261 return false; 1262 } 1263 return true; 1264 } 1265 1266 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 1267 { 1268 struct sock_filter *old_prog; 1269 struct bpf_prog *old_fp; 1270 int err, new_len, old_len = fp->len; 1271 bool seen_ld_abs = false; 1272 1273 /* We are free to overwrite insns et al right here as it won't be used at 1274 * this point in time anymore internally after the migration to the eBPF 1275 * instruction representation. 1276 */ 1277 BUILD_BUG_ON(sizeof(struct sock_filter) != 1278 sizeof(struct bpf_insn)); 1279 1280 /* Conversion cannot happen on overlapping memory areas, 1281 * so we need to keep the user BPF around until the 2nd 1282 * pass. At this time, the user BPF is stored in fp->insns. 1283 */ 1284 old_prog = kmemdup_array(fp->insns, old_len, sizeof(struct sock_filter), 1285 GFP_KERNEL | __GFP_NOWARN); 1286 if (!old_prog) { 1287 err = -ENOMEM; 1288 goto out_err; 1289 } 1290 1291 /* 1st pass: calculate the new program length. */ 1292 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len, 1293 &seen_ld_abs); 1294 if (err) 1295 goto out_err_free; 1296 1297 /* Expand fp for appending the new filter representation. */ 1298 old_fp = fp; 1299 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); 1300 if (!fp) { 1301 /* The old_fp is still around in case we couldn't 1302 * allocate new memory, so uncharge on that one. 1303 */ 1304 fp = old_fp; 1305 err = -ENOMEM; 1306 goto out_err_free; 1307 } 1308 1309 fp->len = new_len; 1310 1311 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ 1312 err = bpf_convert_filter(old_prog, old_len, fp, &new_len, 1313 &seen_ld_abs); 1314 if (err) 1315 /* 2nd bpf_convert_filter() can fail only if it fails 1316 * to allocate memory, remapping must succeed. Note, 1317 * that at this time old_fp has already been released 1318 * by krealloc(). 1319 */ 1320 goto out_err_free; 1321 1322 fp = bpf_prog_select_runtime(fp, &err); 1323 if (err) 1324 goto out_err_free; 1325 1326 kfree(old_prog); 1327 return fp; 1328 1329 out_err_free: 1330 kfree(old_prog); 1331 out_err: 1332 __bpf_prog_release(fp); 1333 return ERR_PTR(err); 1334 } 1335 1336 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, 1337 bpf_aux_classic_check_t trans) 1338 { 1339 int err; 1340 1341 fp->bpf_func = NULL; 1342 fp->jited = 0; 1343 1344 err = bpf_check_classic(fp->insns, fp->len); 1345 if (err) { 1346 __bpf_prog_release(fp); 1347 return ERR_PTR(err); 1348 } 1349 1350 /* There might be additional checks and transformations 1351 * needed on classic filters, f.e. in case of seccomp. 1352 */ 1353 if (trans) { 1354 err = trans(fp->insns, fp->len); 1355 if (err) { 1356 __bpf_prog_release(fp); 1357 return ERR_PTR(err); 1358 } 1359 } 1360 1361 /* Probe if we can JIT compile the filter and if so, do 1362 * the compilation of the filter. 1363 */ 1364 bpf_jit_compile(fp); 1365 1366 /* JIT compiler couldn't process this filter, so do the eBPF translation 1367 * for the optimized interpreter. 1368 */ 1369 if (!fp->jited) 1370 fp = bpf_migrate_filter(fp); 1371 1372 return fp; 1373 } 1374 1375 /** 1376 * bpf_prog_create - create an unattached filter 1377 * @pfp: the unattached filter that is created 1378 * @fprog: the filter program 1379 * 1380 * Create a filter independent of any socket. We first run some 1381 * sanity checks on it to make sure it does not explode on us later. 1382 * If an error occurs or there is insufficient memory for the filter 1383 * a negative errno code is returned. On success the return is zero. 1384 */ 1385 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) 1386 { 1387 unsigned int fsize = bpf_classic_proglen(fprog); 1388 struct bpf_prog *fp; 1389 1390 /* Make sure new filter is there and in the right amounts. */ 1391 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1392 return -EINVAL; 1393 1394 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1395 if (!fp) 1396 return -ENOMEM; 1397 1398 memcpy(fp->insns, fprog->filter, fsize); 1399 1400 fp->len = fprog->len; 1401 /* Since unattached filters are not copied back to user 1402 * space through sk_get_filter(), we do not need to hold 1403 * a copy here, and can spare us the work. 1404 */ 1405 fp->orig_prog = NULL; 1406 1407 /* bpf_prepare_filter() already takes care of freeing 1408 * memory in case something goes wrong. 1409 */ 1410 fp = bpf_prepare_filter(fp, NULL); 1411 if (IS_ERR(fp)) 1412 return PTR_ERR(fp); 1413 1414 *pfp = fp; 1415 return 0; 1416 } 1417 EXPORT_SYMBOL_GPL(bpf_prog_create); 1418 1419 /** 1420 * bpf_prog_create_from_user - create an unattached filter from user buffer 1421 * @pfp: the unattached filter that is created 1422 * @fprog: the filter program 1423 * @trans: post-classic verifier transformation handler 1424 * @save_orig: save classic BPF program 1425 * 1426 * This function effectively does the same as bpf_prog_create(), only 1427 * that it builds up its insns buffer from user space provided buffer. 1428 * It also allows for passing a bpf_aux_classic_check_t handler. 1429 */ 1430 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1431 bpf_aux_classic_check_t trans, bool save_orig) 1432 { 1433 unsigned int fsize = bpf_classic_proglen(fprog); 1434 struct bpf_prog *fp; 1435 int err; 1436 1437 /* Make sure new filter is there and in the right amounts. */ 1438 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1439 return -EINVAL; 1440 1441 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1442 if (!fp) 1443 return -ENOMEM; 1444 1445 if (copy_from_user(fp->insns, fprog->filter, fsize)) { 1446 __bpf_prog_free(fp); 1447 return -EFAULT; 1448 } 1449 1450 fp->len = fprog->len; 1451 fp->orig_prog = NULL; 1452 1453 if (save_orig) { 1454 err = bpf_prog_store_orig_filter(fp, fprog); 1455 if (err) { 1456 __bpf_prog_free(fp); 1457 return -ENOMEM; 1458 } 1459 } 1460 1461 /* bpf_prepare_filter() already takes care of freeing 1462 * memory in case something goes wrong. 1463 */ 1464 fp = bpf_prepare_filter(fp, trans); 1465 if (IS_ERR(fp)) 1466 return PTR_ERR(fp); 1467 1468 *pfp = fp; 1469 return 0; 1470 } 1471 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); 1472 1473 void bpf_prog_destroy(struct bpf_prog *fp) 1474 { 1475 __bpf_prog_release(fp); 1476 } 1477 EXPORT_SYMBOL_GPL(bpf_prog_destroy); 1478 1479 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) 1480 { 1481 struct sk_filter *fp, *old_fp; 1482 1483 fp = kmalloc_obj(*fp); 1484 if (!fp) 1485 return -ENOMEM; 1486 1487 fp->prog = prog; 1488 1489 if (!__sk_filter_charge(sk, fp)) { 1490 kfree(fp); 1491 return -ENOMEM; 1492 } 1493 refcount_set(&fp->refcnt, 1); 1494 1495 old_fp = rcu_dereference_protected(sk->sk_filter, 1496 lockdep_sock_is_held(sk)); 1497 rcu_assign_pointer(sk->sk_filter, fp); 1498 1499 if (old_fp) 1500 sk_filter_uncharge(sk, old_fp); 1501 1502 return 0; 1503 } 1504 1505 static 1506 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) 1507 { 1508 unsigned int fsize = bpf_classic_proglen(fprog); 1509 struct bpf_prog *prog; 1510 int err; 1511 1512 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1513 return ERR_PTR(-EPERM); 1514 1515 /* Make sure new filter is there and in the right amounts. */ 1516 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1517 return ERR_PTR(-EINVAL); 1518 1519 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1520 if (!prog) 1521 return ERR_PTR(-ENOMEM); 1522 1523 if (copy_from_user(prog->insns, fprog->filter, fsize)) { 1524 __bpf_prog_free(prog); 1525 return ERR_PTR(-EFAULT); 1526 } 1527 1528 prog->len = fprog->len; 1529 1530 err = bpf_prog_store_orig_filter(prog, fprog); 1531 if (err) { 1532 __bpf_prog_free(prog); 1533 return ERR_PTR(-ENOMEM); 1534 } 1535 1536 /* bpf_prepare_filter() already takes care of freeing 1537 * memory in case something goes wrong. 1538 */ 1539 return bpf_prepare_filter(prog, NULL); 1540 } 1541 1542 /** 1543 * sk_attach_filter - attach a socket filter 1544 * @fprog: the filter program 1545 * @sk: the socket to use 1546 * 1547 * Attach the user's filter code. We first run some sanity checks on 1548 * it to make sure it does not explode on us later. If an error 1549 * occurs or there is insufficient memory for the filter a negative 1550 * errno code is returned. On success the return is zero. 1551 */ 1552 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1553 { 1554 struct bpf_prog *prog = __get_filter(fprog, sk); 1555 int err; 1556 1557 if (IS_ERR(prog)) 1558 return PTR_ERR(prog); 1559 1560 err = __sk_attach_prog(prog, sk); 1561 if (err < 0) { 1562 __bpf_prog_release(prog); 1563 return err; 1564 } 1565 1566 return 0; 1567 } 1568 EXPORT_SYMBOL_GPL(sk_attach_filter); 1569 1570 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1571 { 1572 struct bpf_prog *prog = __get_filter(fprog, sk); 1573 int err, optmem_max; 1574 1575 if (IS_ERR(prog)) 1576 return PTR_ERR(prog); 1577 1578 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 1579 if (bpf_prog_size(prog->len) > optmem_max) 1580 err = -ENOMEM; 1581 else 1582 err = reuseport_attach_prog(sk, prog); 1583 1584 if (err) 1585 __bpf_prog_release(prog); 1586 1587 return err; 1588 } 1589 1590 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) 1591 { 1592 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1593 return ERR_PTR(-EPERM); 1594 1595 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1596 } 1597 1598 int sk_attach_bpf(u32 ufd, struct sock *sk) 1599 { 1600 struct bpf_prog *prog = __get_bpf(ufd, sk); 1601 int err; 1602 1603 if (IS_ERR(prog)) 1604 return PTR_ERR(prog); 1605 1606 err = __sk_attach_prog(prog, sk); 1607 if (err < 0) { 1608 bpf_prog_put(prog); 1609 return err; 1610 } 1611 1612 return 0; 1613 } 1614 1615 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) 1616 { 1617 struct bpf_prog *prog; 1618 int err, optmem_max; 1619 1620 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1621 return -EPERM; 1622 1623 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1624 if (PTR_ERR(prog) == -EINVAL) 1625 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT); 1626 if (IS_ERR(prog)) 1627 return PTR_ERR(prog); 1628 1629 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) { 1630 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER 1631 * bpf prog (e.g. sockmap). It depends on the 1632 * limitation imposed by bpf_prog_load(). 1633 * Hence, sysctl_optmem_max is not checked. 1634 */ 1635 if ((sk->sk_type != SOCK_STREAM && 1636 sk->sk_type != SOCK_DGRAM) || 1637 (sk->sk_protocol != IPPROTO_UDP && 1638 sk->sk_protocol != IPPROTO_TCP) || 1639 (sk->sk_family != AF_INET && 1640 sk->sk_family != AF_INET6)) { 1641 err = -ENOTSUPP; 1642 goto err_prog_put; 1643 } 1644 } else { 1645 /* BPF_PROG_TYPE_SOCKET_FILTER */ 1646 optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 1647 if (bpf_prog_size(prog->len) > optmem_max) { 1648 err = -ENOMEM; 1649 goto err_prog_put; 1650 } 1651 } 1652 1653 err = reuseport_attach_prog(sk, prog); 1654 err_prog_put: 1655 if (err) 1656 bpf_prog_put(prog); 1657 1658 return err; 1659 } 1660 1661 static void sk_reuseport_prog_free_rcu(struct rcu_head *rcu) 1662 { 1663 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); 1664 struct bpf_prog *prog = aux->prog; 1665 1666 bpf_release_orig_filter(prog); 1667 bpf_prog_free(prog); 1668 } 1669 1670 void sk_reuseport_prog_free(struct bpf_prog *prog) 1671 { 1672 if (!prog) 1673 return; 1674 1675 if (bpf_prog_was_classic(prog)) 1676 call_rcu(&prog->aux->rcu, sk_reuseport_prog_free_rcu); 1677 else 1678 bpf_prog_put(prog); 1679 } 1680 1681 static inline int __bpf_try_make_writable(struct sk_buff *skb, 1682 unsigned int write_len) 1683 { 1684 #ifdef CONFIG_DEBUG_NET 1685 /* Avoid a splat in pskb_may_pull_reason() */ 1686 if (write_len > INT_MAX) 1687 return -EINVAL; 1688 #endif 1689 return skb_ensure_writable(skb, write_len); 1690 } 1691 1692 static inline int bpf_try_make_writable(struct sk_buff *skb, 1693 unsigned int write_len) 1694 { 1695 int err = __bpf_try_make_writable(skb, write_len); 1696 1697 bpf_compute_data_pointers(skb); 1698 return err; 1699 } 1700 1701 static int bpf_try_make_head_writable(struct sk_buff *skb) 1702 { 1703 return bpf_try_make_writable(skb, skb_headlen(skb)); 1704 } 1705 1706 static inline void bpf_push_mac_rcsum(struct sk_buff *skb) 1707 { 1708 if (skb_at_tc_ingress(skb)) 1709 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1710 } 1711 1712 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) 1713 { 1714 if (skb_at_tc_ingress(skb)) 1715 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1716 } 1717 1718 /* 1719 * The source of a write through a dynptr may be a slice of the same skb. 1720 * Making the skb writable can move its head: follow the move. 1721 */ 1722 static int bpf_try_make_writable_from(struct sk_buff *skb, u32 write_len, 1723 const void **from) 1724 { 1725 const void *head = skb->head; 1726 bool in_head = *from >= head && 1727 *from < (const void *)skb_end_pointer(skb); 1728 int err; 1729 1730 err = bpf_try_make_writable(skb, write_len); 1731 if (!err && in_head) 1732 *from = skb->head + (*from - head); 1733 return err; 1734 } 1735 1736 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset, 1737 const void *, from, u32, len, u64, flags) 1738 { 1739 void *ptr; 1740 1741 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) 1742 return -EINVAL; 1743 if (unlikely(offset > INT_MAX)) 1744 return -EFAULT; 1745 if (unlikely(bpf_try_make_writable_from(skb, offset + len, &from))) 1746 return -EFAULT; 1747 1748 ptr = skb->data + offset; 1749 if (flags & BPF_F_RECOMPUTE_CSUM) 1750 __skb_postpull_rcsum(skb, ptr, len, offset); 1751 1752 /* through a dynptr the source may be an overlapping slice of the skb */ 1753 memmove(ptr, from, len); 1754 1755 if (flags & BPF_F_RECOMPUTE_CSUM) 1756 __skb_postpush_rcsum(skb, ptr, len, offset); 1757 if (flags & BPF_F_INVALIDATE_HASH) 1758 skb_clear_hash(skb); 1759 1760 return 0; 1761 } 1762 1763 static const struct bpf_func_proto bpf_skb_store_bytes_proto = { 1764 .func = bpf_skb_store_bytes, 1765 .gpl_only = false, 1766 .ret_type = RET_INTEGER, 1767 .arg1_type = ARG_PTR_TO_CTX, 1768 .arg2_type = ARG_ANYTHING, 1769 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 1770 .arg4_type = ARG_MEM_SIZE, 1771 .arg5_type = ARG_ANYTHING, 1772 }; 1773 1774 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1775 u32 len, u64 flags) 1776 { 1777 return ____bpf_skb_store_bytes(skb, offset, from, len, flags); 1778 } 1779 1780 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset, 1781 void *, to, u32, len) 1782 { 1783 void *ptr; 1784 1785 if (unlikely(offset > INT_MAX)) 1786 goto err_clear; 1787 1788 ptr = skb_header_pointer(skb, offset, len, to); 1789 if (unlikely(!ptr)) 1790 goto err_clear; 1791 if (ptr != to) 1792 memmove(to, ptr, len); 1793 1794 return 0; 1795 err_clear: 1796 memset(to, 0, len); 1797 return -EFAULT; 1798 } 1799 1800 static const struct bpf_func_proto bpf_skb_load_bytes_proto = { 1801 .func = bpf_skb_load_bytes, 1802 .gpl_only = false, 1803 .ret_type = RET_INTEGER, 1804 .arg1_type = ARG_PTR_TO_CTX, 1805 .arg2_type = ARG_ANYTHING, 1806 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1807 .arg4_type = ARG_MEM_SIZE, 1808 }; 1809 1810 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len) 1811 { 1812 return ____bpf_skb_load_bytes(skb, offset, to, len); 1813 } 1814 1815 BPF_CALL_4(bpf_flow_dissector_load_bytes, 1816 const struct bpf_flow_dissector *, ctx, u32, offset, 1817 void *, to, u32, len) 1818 { 1819 void *ptr; 1820 1821 if (unlikely(offset > 0xffff)) 1822 goto err_clear; 1823 1824 if (unlikely(!ctx->skb)) 1825 goto err_clear; 1826 1827 ptr = skb_header_pointer(ctx->skb, offset, len, to); 1828 if (unlikely(!ptr)) 1829 goto err_clear; 1830 if (ptr != to) 1831 memcpy(to, ptr, len); 1832 1833 return 0; 1834 err_clear: 1835 memset(to, 0, len); 1836 return -EFAULT; 1837 } 1838 1839 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = { 1840 .func = bpf_flow_dissector_load_bytes, 1841 .gpl_only = false, 1842 .ret_type = RET_INTEGER, 1843 .arg1_type = ARG_PTR_TO_CTX, 1844 .arg2_type = ARG_ANYTHING, 1845 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1846 .arg4_type = ARG_MEM_SIZE, 1847 }; 1848 1849 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb, 1850 u32, offset, void *, to, u32, len, u32, start_header) 1851 { 1852 u8 *end = skb_tail_pointer(skb); 1853 u8 *start, *ptr; 1854 1855 if (unlikely(offset > 0xffff)) 1856 goto err_clear; 1857 1858 switch (start_header) { 1859 case BPF_HDR_START_MAC: 1860 if (unlikely(!skb_mac_header_was_set(skb))) 1861 goto err_clear; 1862 start = skb_mac_header(skb); 1863 break; 1864 case BPF_HDR_START_NET: 1865 start = skb_network_header(skb); 1866 break; 1867 default: 1868 goto err_clear; 1869 } 1870 1871 ptr = start + offset; 1872 1873 if (likely(ptr + len <= end)) { 1874 memmove(to, ptr, len); 1875 return 0; 1876 } 1877 1878 err_clear: 1879 memset(to, 0, len); 1880 return -EFAULT; 1881 } 1882 1883 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = { 1884 .func = bpf_skb_load_bytes_relative, 1885 .gpl_only = false, 1886 .ret_type = RET_INTEGER, 1887 .arg1_type = ARG_PTR_TO_CTX, 1888 .arg2_type = ARG_ANYTHING, 1889 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1890 .arg4_type = ARG_MEM_SIZE, 1891 .arg5_type = ARG_ANYTHING, 1892 }; 1893 1894 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len) 1895 { 1896 /* Idea is the following: should the needed direct read/write 1897 * test fail during runtime, we can pull in more data and redo 1898 * again, since implicitly, we invalidate previous checks here. 1899 * 1900 * Or, since we know how much we need to make read/writeable, 1901 * this can be done once at the program beginning for direct 1902 * access case. By this we overcome limitations of only current 1903 * headroom being accessible. 1904 */ 1905 return bpf_try_make_writable(skb, len ? : skb_headlen(skb)); 1906 } 1907 1908 static const struct bpf_func_proto bpf_skb_pull_data_proto = { 1909 .func = bpf_skb_pull_data, 1910 .gpl_only = false, 1911 .ret_type = RET_INTEGER, 1912 .arg1_type = ARG_PTR_TO_CTX, 1913 .arg2_type = ARG_ANYTHING, 1914 }; 1915 1916 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk) 1917 { 1918 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL; 1919 } 1920 1921 static const struct bpf_func_proto bpf_sk_fullsock_proto = { 1922 .func = bpf_sk_fullsock, 1923 .gpl_only = false, 1924 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 1925 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 1926 }; 1927 1928 static inline int sk_skb_try_make_writable(struct sk_buff *skb, 1929 unsigned int write_len) 1930 { 1931 return __bpf_try_make_writable(skb, write_len); 1932 } 1933 1934 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len) 1935 { 1936 /* Idea is the following: should the needed direct read/write 1937 * test fail during runtime, we can pull in more data and redo 1938 * again, since implicitly, we invalidate previous checks here. 1939 * 1940 * Or, since we know how much we need to make read/writeable, 1941 * this can be done once at the program beginning for direct 1942 * access case. By this we overcome limitations of only current 1943 * headroom being accessible. 1944 */ 1945 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb)); 1946 } 1947 1948 static const struct bpf_func_proto sk_skb_pull_data_proto = { 1949 .func = sk_skb_pull_data, 1950 .gpl_only = false, 1951 .ret_type = RET_INTEGER, 1952 .arg1_type = ARG_PTR_TO_CTX, 1953 .arg2_type = ARG_ANYTHING, 1954 }; 1955 1956 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset, 1957 u64, from, u64, to, u64, flags) 1958 { 1959 __sum16 *ptr; 1960 1961 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) 1962 return -EINVAL; 1963 if (unlikely(offset > 0xffff || offset & 1)) 1964 return -EFAULT; 1965 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1966 return -EFAULT; 1967 1968 ptr = (__sum16 *)(skb->data + offset); 1969 switch (flags & BPF_F_HDR_FIELD_MASK) { 1970 case 0: 1971 if (unlikely(from != 0)) 1972 return -EINVAL; 1973 1974 csum_replace_by_diff(ptr, to); 1975 break; 1976 case 2: 1977 csum_replace2(ptr, from, to); 1978 break; 1979 case 4: 1980 csum_replace4(ptr, from, to); 1981 break; 1982 default: 1983 return -EINVAL; 1984 } 1985 1986 return 0; 1987 } 1988 1989 static const struct bpf_func_proto bpf_l3_csum_replace_proto = { 1990 .func = bpf_l3_csum_replace, 1991 .gpl_only = false, 1992 .ret_type = RET_INTEGER, 1993 .arg1_type = ARG_PTR_TO_CTX, 1994 .arg2_type = ARG_ANYTHING, 1995 .arg3_type = ARG_ANYTHING, 1996 .arg4_type = ARG_ANYTHING, 1997 .arg5_type = ARG_ANYTHING, 1998 }; 1999 2000 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset, 2001 u64, from, u64, to, u64, flags) 2002 { 2003 bool is_pseudo = flags & BPF_F_PSEUDO_HDR; 2004 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; 2005 bool do_mforce = flags & BPF_F_MARK_ENFORCE; 2006 bool is_ipv6 = flags & BPF_F_IPV6; 2007 __sum16 *ptr; 2008 2009 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE | 2010 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK | BPF_F_IPV6))) 2011 return -EINVAL; 2012 if (unlikely(offset > 0xffff || offset & 1)) 2013 return -EFAULT; 2014 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 2015 return -EFAULT; 2016 2017 ptr = (__sum16 *)(skb->data + offset); 2018 if (is_mmzero && !do_mforce && !*ptr) 2019 return 0; 2020 2021 switch (flags & BPF_F_HDR_FIELD_MASK) { 2022 case 0: 2023 if (unlikely(from != 0)) 2024 return -EINVAL; 2025 2026 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo, is_ipv6); 2027 break; 2028 case 2: 2029 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); 2030 break; 2031 case 4: 2032 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); 2033 break; 2034 default: 2035 return -EINVAL; 2036 } 2037 2038 if (is_mmzero && !*ptr) 2039 *ptr = CSUM_MANGLED_0; 2040 return 0; 2041 } 2042 2043 static const struct bpf_func_proto bpf_l4_csum_replace_proto = { 2044 .func = bpf_l4_csum_replace, 2045 .gpl_only = false, 2046 .ret_type = RET_INTEGER, 2047 .arg1_type = ARG_PTR_TO_CTX, 2048 .arg2_type = ARG_ANYTHING, 2049 .arg3_type = ARG_ANYTHING, 2050 .arg4_type = ARG_ANYTHING, 2051 .arg5_type = ARG_ANYTHING, 2052 }; 2053 2054 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size, 2055 __be32 *, to, u32, to_size, __wsum, seed) 2056 { 2057 /* This is quite flexible, some examples: 2058 * 2059 * from_size == 0, to_size > 0, seed := csum --> pushing data 2060 * from_size > 0, to_size == 0, seed := csum --> pulling data 2061 * from_size > 0, to_size > 0, seed := 0 --> diffing data 2062 * 2063 * Even for diffing, from_size and to_size don't need to be equal. 2064 */ 2065 2066 __wsum ret = seed; 2067 2068 if (from_size && to_size) 2069 ret = csum_sub(csum_partial(to, to_size, ret), 2070 csum_partial(from, from_size, 0)); 2071 else if (to_size) 2072 ret = csum_partial(to, to_size, ret); 2073 2074 else if (from_size) 2075 ret = ~csum_partial(from, from_size, ~ret); 2076 2077 return csum_from32to16((__force unsigned int)ret); 2078 } 2079 2080 static const struct bpf_func_proto bpf_csum_diff_proto = { 2081 .func = bpf_csum_diff, 2082 .gpl_only = false, 2083 .pkt_access = true, 2084 .ret_type = RET_INTEGER, 2085 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2086 .arg2_type = ARG_MEM_SIZE_OR_ZERO, 2087 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2088 .arg4_type = ARG_MEM_SIZE_OR_ZERO, 2089 .arg5_type = ARG_ANYTHING, 2090 }; 2091 2092 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum) 2093 { 2094 /* The interface is to be used in combination with bpf_csum_diff() 2095 * for direct packet writes. csum rotation for alignment as well 2096 * as emulating csum_sub() can be done from the eBPF program. 2097 */ 2098 if (skb->ip_summed == CHECKSUM_COMPLETE) 2099 return (skb->csum = csum_add(skb->csum, csum)); 2100 2101 return -ENOTSUPP; 2102 } 2103 2104 static const struct bpf_func_proto bpf_csum_update_proto = { 2105 .func = bpf_csum_update, 2106 .gpl_only = false, 2107 .ret_type = RET_INTEGER, 2108 .arg1_type = ARG_PTR_TO_CTX, 2109 .arg2_type = ARG_ANYTHING, 2110 }; 2111 2112 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level) 2113 { 2114 /* The interface is to be used in combination with bpf_skb_adjust_room() 2115 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET 2116 * is passed as flags, for example. 2117 */ 2118 switch (level) { 2119 case BPF_CSUM_LEVEL_INC: 2120 __skb_incr_checksum_unnecessary(skb); 2121 break; 2122 case BPF_CSUM_LEVEL_DEC: 2123 __skb_decr_checksum_unnecessary(skb); 2124 break; 2125 case BPF_CSUM_LEVEL_RESET: 2126 __skb_reset_checksum_unnecessary(skb); 2127 break; 2128 case BPF_CSUM_LEVEL_QUERY: 2129 return skb->ip_summed == CHECKSUM_UNNECESSARY ? 2130 skb->csum_level : -EACCES; 2131 default: 2132 return -EINVAL; 2133 } 2134 2135 return 0; 2136 } 2137 2138 static const struct bpf_func_proto bpf_csum_level_proto = { 2139 .func = bpf_csum_level, 2140 .gpl_only = false, 2141 .ret_type = RET_INTEGER, 2142 .arg1_type = ARG_PTR_TO_CTX, 2143 .arg2_type = ARG_ANYTHING, 2144 }; 2145 2146 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) 2147 { 2148 return dev_forward_skb_nomtu(dev, skb); 2149 } 2150 2151 static inline int __bpf_rx_skb_no_mac(struct net_device *dev, 2152 struct sk_buff *skb) 2153 { 2154 int ret = ____dev_forward_skb(dev, skb, false); 2155 2156 if (likely(!ret)) { 2157 skb->dev = dev; 2158 ret = netif_rx(skb); 2159 } 2160 2161 return ret; 2162 } 2163 2164 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) 2165 { 2166 int ret; 2167 2168 if (dev_xmit_recursion()) { 2169 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2170 kfree_skb(skb); 2171 return -ENETDOWN; 2172 } 2173 2174 skb->dev = dev; 2175 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb)); 2176 skb_clear_tstamp(skb); 2177 2178 dev_xmit_recursion_inc(); 2179 ret = dev_queue_xmit(skb); 2180 dev_xmit_recursion_dec(); 2181 2182 return ret; 2183 } 2184 2185 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 2186 u32 flags) 2187 { 2188 unsigned int mlen = skb_network_offset(skb); 2189 2190 if (unlikely(skb->len <= mlen)) { 2191 kfree_skb(skb); 2192 return -ERANGE; 2193 } 2194 2195 if (mlen) { 2196 __skb_pull(skb, mlen); 2197 2198 /* At ingress, the mac header has already been pulled once. 2199 * At egress, skb_pospull_rcsum has to be done in case that 2200 * the skb is originated from ingress (i.e. a forwarded skb) 2201 * to ensure that rcsum starts at net header. 2202 */ 2203 if (!skb_at_tc_ingress(skb)) 2204 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 2205 } 2206 skb_pop_mac_header(skb); 2207 skb_reset_mac_len(skb); 2208 return flags & BPF_F_INGRESS ? 2209 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 2210 } 2211 2212 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 2213 u32 flags) 2214 { 2215 /* Verify that a link layer header is carried */ 2216 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) { 2217 kfree_skb(skb); 2218 return -ERANGE; 2219 } 2220 2221 bpf_push_mac_rcsum(skb); 2222 return flags & BPF_F_INGRESS ? 2223 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 2224 } 2225 2226 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 2227 u32 flags) 2228 { 2229 if (dev_is_mac_header_xmit(dev)) 2230 return __bpf_redirect_common(skb, dev, flags); 2231 else 2232 return __bpf_redirect_no_mac(skb, dev, flags); 2233 } 2234 2235 #if IS_ENABLED(CONFIG_IPV6) 2236 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb, 2237 struct net_device *dev, struct bpf_nh_params *nh) 2238 { 2239 u32 hh_len = LL_RESERVED_SPACE(dev); 2240 const struct in6_addr *nexthop; 2241 struct dst_entry *dst = NULL; 2242 struct neighbour *neigh; 2243 2244 if (dev_xmit_recursion()) { 2245 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2246 goto out_drop; 2247 } 2248 2249 skb->dev = dev; 2250 skb_clear_tstamp(skb); 2251 2252 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2253 skb = skb_expand_head(skb, hh_len); 2254 if (!skb) 2255 return -ENOMEM; 2256 } 2257 2258 if (unlikely(!ipv6_mod_enabled())) 2259 goto out_drop; 2260 2261 rcu_read_lock(); 2262 if (!nh) { 2263 dst = skb_dst(skb); 2264 nexthop = rt6_nexthop(dst_rt6_info(dst), 2265 &ipv6_hdr(skb)->daddr); 2266 } else { 2267 nexthop = &nh->ipv6_nh; 2268 } 2269 neigh = ip_neigh_gw6(dev, nexthop); 2270 if (likely(!IS_ERR(neigh))) { 2271 int ret; 2272 2273 sock_confirm_neigh(skb, neigh); 2274 local_bh_disable(); 2275 dev_xmit_recursion_inc(); 2276 ret = neigh_output(neigh, skb, false); 2277 dev_xmit_recursion_dec(); 2278 local_bh_enable(); 2279 rcu_read_unlock(); 2280 return ret; 2281 } 2282 rcu_read_unlock(); 2283 if (dst) 2284 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 2285 out_drop: 2286 kfree_skb(skb); 2287 return -ENETDOWN; 2288 } 2289 2290 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2291 struct bpf_nh_params *nh) 2292 { 2293 struct net *net = dev_net(dev); 2294 int err, ret = NET_XMIT_DROP; 2295 const struct ipv6hdr *ip6h; 2296 2297 if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h)))) 2298 goto out_drop; 2299 ip6h = ipv6_hdr(skb); 2300 2301 if (!nh) { 2302 struct dst_entry *dst; 2303 struct flowi6 fl6 = { 2304 .flowi6_flags = FLOWI_FLAG_ANYSRC, 2305 .flowi6_mark = skb->mark, 2306 .flowlabel = ip6_flowinfo(ip6h), 2307 .flowi6_oif = dev->ifindex, 2308 .flowi6_proto = ip6h->nexthdr, 2309 .daddr = ip6h->daddr, 2310 .saddr = ip6h->saddr, 2311 }; 2312 2313 dst = ip6_dst_lookup_flow(net, NULL, &fl6, NULL); 2314 if (IS_ERR(dst)) 2315 goto out_drop; 2316 2317 skb_dst_drop(skb); 2318 skb_dst_set(skb, dst); 2319 } else if (nh->nh_family != AF_INET6) { 2320 goto out_drop; 2321 } 2322 2323 err = bpf_out_neigh_v6(net, skb, dev, nh); 2324 if (unlikely(net_xmit_eval(err))) 2325 dev_core_stats_tx_dropped_inc(dev); 2326 else 2327 ret = NET_XMIT_SUCCESS; 2328 goto out_xmit; 2329 out_drop: 2330 dev_core_stats_tx_dropped_inc(dev); 2331 kfree_skb(skb); 2332 out_xmit: 2333 return ret; 2334 } 2335 #else 2336 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2337 struct bpf_nh_params *nh) 2338 { 2339 kfree_skb(skb); 2340 return NET_XMIT_DROP; 2341 } 2342 #endif /* CONFIG_IPV6 */ 2343 2344 #if IS_ENABLED(CONFIG_INET) 2345 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb, 2346 struct net_device *dev, struct bpf_nh_params *nh) 2347 { 2348 u32 hh_len = LL_RESERVED_SPACE(dev); 2349 struct neighbour *neigh; 2350 bool is_v6gw = false; 2351 2352 if (dev_xmit_recursion()) { 2353 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2354 goto out_drop; 2355 } 2356 2357 skb->dev = dev; 2358 skb_clear_tstamp(skb); 2359 2360 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2361 skb = skb_expand_head(skb, hh_len); 2362 if (!skb) 2363 return -ENOMEM; 2364 } 2365 2366 rcu_read_lock(); 2367 if (!nh) { 2368 struct rtable *rt = skb_rtable(skb); 2369 2370 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 2371 } else if (nh->nh_family == AF_INET6) { 2372 if (unlikely(!ipv6_mod_enabled())) { 2373 rcu_read_unlock(); 2374 goto out_drop; 2375 } 2376 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh); 2377 is_v6gw = true; 2378 } else if (nh->nh_family == AF_INET) { 2379 neigh = ip_neigh_gw4(dev, nh->ipv4_nh); 2380 } else { 2381 rcu_read_unlock(); 2382 goto out_drop; 2383 } 2384 2385 if (likely(!IS_ERR(neigh))) { 2386 int ret; 2387 2388 sock_confirm_neigh(skb, neigh); 2389 local_bh_disable(); 2390 dev_xmit_recursion_inc(); 2391 ret = neigh_output(neigh, skb, is_v6gw); 2392 dev_xmit_recursion_dec(); 2393 local_bh_enable(); 2394 rcu_read_unlock(); 2395 return ret; 2396 } 2397 rcu_read_unlock(); 2398 out_drop: 2399 kfree_skb(skb); 2400 return -ENETDOWN; 2401 } 2402 2403 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2404 struct bpf_nh_params *nh) 2405 { 2406 struct net *net = dev_net(dev); 2407 int err, ret = NET_XMIT_DROP; 2408 const struct iphdr *ip4h; 2409 2410 if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h)))) 2411 goto out_drop; 2412 ip4h = ip_hdr(skb); 2413 2414 if (!nh) { 2415 struct flowi4 fl4 = { 2416 .flowi4_flags = FLOWI_FLAG_ANYSRC, 2417 .flowi4_mark = skb->mark, 2418 .flowi4_dscp = ip4h_dscp(ip4h), 2419 .flowi4_oif = dev->ifindex, 2420 .flowi4_proto = ip4h->protocol, 2421 .daddr = ip4h->daddr, 2422 .saddr = ip4h->saddr, 2423 }; 2424 struct rtable *rt; 2425 2426 rt = ip_route_output_flow(net, &fl4, NULL); 2427 if (IS_ERR(rt)) 2428 goto out_drop; 2429 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 2430 ip_rt_put(rt); 2431 goto out_drop; 2432 } 2433 2434 skb_dst_drop(skb); 2435 skb_dst_set(skb, &rt->dst); 2436 } 2437 2438 err = bpf_out_neigh_v4(net, skb, dev, nh); 2439 if (unlikely(net_xmit_eval(err))) 2440 dev_core_stats_tx_dropped_inc(dev); 2441 else 2442 ret = NET_XMIT_SUCCESS; 2443 goto out_xmit; 2444 out_drop: 2445 dev_core_stats_tx_dropped_inc(dev); 2446 kfree_skb(skb); 2447 out_xmit: 2448 return ret; 2449 } 2450 #else 2451 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2452 struct bpf_nh_params *nh) 2453 { 2454 kfree_skb(skb); 2455 return NET_XMIT_DROP; 2456 } 2457 #endif /* CONFIG_INET */ 2458 2459 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev, 2460 struct bpf_nh_params *nh) 2461 { 2462 if (unlikely(skb_network_offset(skb) < ETH_HLEN || 2463 !pskb_may_pull(skb, ETH_HLEN))) 2464 goto out; 2465 bpf_push_mac_rcsum(skb); 2466 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) 2467 goto out; 2468 2469 skb_pull(skb, ETH_HLEN); 2470 skb_unset_mac_header(skb); 2471 skb_reset_network_header(skb); 2472 2473 if (skb->protocol == htons(ETH_P_IP)) 2474 return __bpf_redirect_neigh_v4(skb, dev, nh); 2475 else if (skb->protocol == htons(ETH_P_IPV6)) 2476 return __bpf_redirect_neigh_v6(skb, dev, nh); 2477 out: 2478 kfree_skb(skb); 2479 return -ENOTSUPP; 2480 } 2481 2482 /* Internal, non-exposed redirect flags. */ 2483 enum { 2484 BPF_F_NEIGH = (1ULL << 16), 2485 BPF_F_PEER = (1ULL << 17), 2486 BPF_F_NEXTHOP = (1ULL << 18), 2487 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP) 2488 }; 2489 2490 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 2491 { 2492 struct net_device *dev; 2493 struct sk_buff *clone; 2494 int ret; 2495 2496 BUILD_BUG_ON(BPF_F_REDIRECT_INTERNAL & BPF_F_REDIRECT_FLAGS); 2497 2498 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2499 return -EINVAL; 2500 2501 /* BPF test infra's convert___skb_to_skb() can create type-less 2502 * GSO packets. gso_features_check() will detect this as a bad 2503 * offload. However, lets not leak them out in the first place. 2504 */ 2505 if (unlikely(skb_is_gso(skb) && !skb_shinfo(skb)->gso_type)) 2506 return -EBADMSG; 2507 2508 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 2509 if (unlikely(!dev)) 2510 return -EINVAL; 2511 2512 clone = skb_clone(skb, GFP_ATOMIC); 2513 if (unlikely(!clone)) 2514 return -ENOMEM; 2515 2516 /* For direct write, we need to keep the invariant that the skbs 2517 * we're dealing with need to be uncloned. Should uncloning fail 2518 * here, we need to free the just generated clone to unclone once 2519 * again. 2520 */ 2521 ret = bpf_try_make_head_writable(skb); 2522 if (unlikely(ret)) { 2523 kfree_skb(clone); 2524 return -ENOMEM; 2525 } 2526 2527 return __bpf_redirect(clone, dev, flags); 2528 } 2529 2530 static const struct bpf_func_proto bpf_clone_redirect_proto = { 2531 .func = bpf_clone_redirect, 2532 .gpl_only = false, 2533 .ret_type = RET_INTEGER, 2534 .arg1_type = ARG_PTR_TO_CTX, 2535 .arg2_type = ARG_ANYTHING, 2536 .arg3_type = ARG_ANYTHING, 2537 }; 2538 2539 static struct net_device *skb_get_peer_dev(struct net_device *dev) 2540 { 2541 const struct net_device_ops *ops = dev->netdev_ops; 2542 2543 if (likely(ops->ndo_get_peer_dev)) 2544 return INDIRECT_CALL_1(ops->ndo_get_peer_dev, 2545 netkit_peer_dev, dev); 2546 return NULL; 2547 } 2548 2549 int skb_do_redirect(struct sk_buff *skb) 2550 { 2551 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 2552 struct net *net = dev_net(skb->dev); 2553 struct net_device *dev; 2554 u32 flags = ri->flags; 2555 2556 dev = dev_get_by_index_rcu(net, ri->tgt_index); 2557 ri->tgt_index = 0; 2558 ri->flags = 0; 2559 if (unlikely(!dev)) 2560 goto out_drop; 2561 if (flags & BPF_F_PEER) { 2562 dev = skb_get_peer_dev(dev); 2563 if (unlikely(!dev || 2564 !(dev->flags & IFF_UP) || 2565 net_eq(net, dev_net(dev)))) 2566 goto out_drop; 2567 skb_scrub_packet(skb, false); 2568 if (flags & BPF_F_EGRESS) 2569 return __bpf_redirect(skb, dev, 0); 2570 if (unlikely(!skb_at_tc_ingress(skb))) 2571 goto out_drop; 2572 skb->dev = dev; 2573 dev_sw_netstats_rx_add(dev, skb->len); 2574 return -EAGAIN; 2575 } 2576 return flags & BPF_F_NEIGH ? 2577 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ? 2578 &ri->nh : NULL) : 2579 __bpf_redirect(skb, dev, flags); 2580 out_drop: 2581 kfree_skb(skb); 2582 return -EINVAL; 2583 } 2584 2585 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 2586 { 2587 struct bpf_redirect_info *ri; 2588 2589 if (unlikely(!bpf_net_ctx_get() || 2590 (flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))) 2591 return TC_ACT_SHOT; 2592 2593 ri = bpf_net_ctx_get_ri(); 2594 ri->flags = flags; 2595 ri->tgt_index = ifindex; 2596 2597 return TC_ACT_REDIRECT; 2598 } 2599 2600 static const struct bpf_func_proto bpf_redirect_proto = { 2601 .func = bpf_redirect, 2602 .gpl_only = false, 2603 .ret_type = RET_INTEGER, 2604 .arg1_type = ARG_ANYTHING, 2605 .arg2_type = ARG_ANYTHING, 2606 }; 2607 2608 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags) 2609 { 2610 struct bpf_redirect_info *ri; 2611 2612 if (unlikely(!bpf_net_ctx_get() || (flags & ~BPF_F_EGRESS))) 2613 return TC_ACT_SHOT; 2614 2615 ri = bpf_net_ctx_get_ri(); 2616 ri->flags = BPF_F_PEER | flags; 2617 ri->tgt_index = ifindex; 2618 2619 return TC_ACT_REDIRECT; 2620 } 2621 2622 static const struct bpf_func_proto bpf_redirect_peer_proto = { 2623 .func = bpf_redirect_peer, 2624 .gpl_only = false, 2625 .ret_type = RET_INTEGER, 2626 .arg1_type = ARG_ANYTHING, 2627 .arg2_type = ARG_ANYTHING, 2628 }; 2629 2630 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params, 2631 int, plen, u64, flags) 2632 { 2633 struct bpf_redirect_info *ri; 2634 2635 if (unlikely((plen && plen < sizeof(*params)) || 2636 !bpf_net_ctx_get() || flags)) 2637 return TC_ACT_SHOT; 2638 2639 ri = bpf_net_ctx_get_ri(); 2640 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0); 2641 ri->tgt_index = ifindex; 2642 2643 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params)); 2644 if (plen) 2645 memcpy(&ri->nh, params, sizeof(ri->nh)); 2646 2647 return TC_ACT_REDIRECT; 2648 } 2649 2650 static const struct bpf_func_proto bpf_redirect_neigh_proto = { 2651 .func = bpf_redirect_neigh, 2652 .gpl_only = false, 2653 .ret_type = RET_INTEGER, 2654 .arg1_type = ARG_ANYTHING, 2655 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2656 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 2657 .arg4_type = ARG_ANYTHING, 2658 }; 2659 2660 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes) 2661 { 2662 msg->apply_bytes = bytes; 2663 return 0; 2664 } 2665 2666 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = { 2667 .func = bpf_msg_apply_bytes, 2668 .gpl_only = false, 2669 .ret_type = RET_INTEGER, 2670 .arg1_type = ARG_PTR_TO_CTX, 2671 .arg2_type = ARG_ANYTHING, 2672 }; 2673 2674 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes) 2675 { 2676 msg->cork_bytes = bytes; 2677 return 0; 2678 } 2679 2680 static bool sk_msg_elem_is_copy(const struct sk_msg *msg, u32 i) 2681 { 2682 return test_bit(i, msg->sg.copy); 2683 } 2684 2685 static void sk_msg_clear_elem_copy(struct sk_msg *msg, u32 i) 2686 { 2687 __clear_bit(i, msg->sg.copy); 2688 } 2689 2690 static void sk_msg_set_elem_copy(struct sk_msg *msg, u32 i, bool sg_copy) 2691 { 2692 __assign_bit(i, msg->sg.copy, sg_copy); 2693 } 2694 2695 static void sk_msg_clear_copy_range(struct sk_msg *msg, u32 start, u32 end) 2696 { 2697 while (start != end) { 2698 sk_msg_clear_elem_copy(msg, start); 2699 sk_msg_iter_var_next(start); 2700 } 2701 } 2702 2703 static void sk_msg_sg_move(struct sk_msg *msg, u32 dst, u32 src) 2704 { 2705 msg->sg.data[dst] = msg->sg.data[src]; 2706 2707 sk_msg_set_elem_copy(msg, dst, 2708 sk_msg_elem_is_copy(msg, src)); 2709 } 2710 2711 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = { 2712 .func = bpf_msg_cork_bytes, 2713 .gpl_only = false, 2714 .ret_type = RET_INTEGER, 2715 .arg1_type = ARG_PTR_TO_CTX, 2716 .arg2_type = ARG_ANYTHING, 2717 }; 2718 2719 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start, 2720 u32, end, u64, flags) 2721 { 2722 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start; 2723 u32 first_sge, last_sge, i, shift, bytes_sg_total; 2724 struct scatterlist *sge; 2725 u8 *raw, *to, *from; 2726 struct page *page; 2727 2728 if (unlikely(flags || end <= start)) 2729 return -EINVAL; 2730 2731 /* First find the starting scatterlist element */ 2732 i = msg->sg.start; 2733 do { 2734 offset += len; 2735 len = sk_msg_elem(msg, i)->length; 2736 if (start < offset + len) 2737 break; 2738 sk_msg_iter_var_next(i); 2739 } while (i != msg->sg.end); 2740 2741 if (unlikely(start >= offset + len)) 2742 return -EINVAL; 2743 2744 first_sge = i; 2745 /* The start may point into the sg element so we need to also 2746 * account for the headroom. 2747 */ 2748 bytes_sg_total = start - offset + bytes; 2749 if (!sk_msg_elem_is_copy(msg, i) && bytes_sg_total <= len) 2750 goto out; 2751 2752 /* At this point we need to linearize multiple scatterlist 2753 * elements or a single shared page. Either way we need to 2754 * copy into a linear buffer exclusively owned by BPF. Then 2755 * place the buffer in the scatterlist and fixup the original 2756 * entries by removing the entries now in the linear buffer 2757 * and shifting the remaining entries. For now we do not try 2758 * to copy partial entries to avoid complexity of running out 2759 * of sg_entry slots. The downside is reading a single byte 2760 * will copy the entire sg entry. 2761 */ 2762 do { 2763 copy += sk_msg_elem(msg, i)->length; 2764 sk_msg_iter_var_next(i); 2765 if (bytes_sg_total <= copy) 2766 break; 2767 } while (i != msg->sg.end); 2768 last_sge = i; 2769 2770 if (unlikely(bytes_sg_total > copy)) 2771 return -EINVAL; 2772 2773 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2774 get_order(copy)); 2775 if (unlikely(!page)) 2776 return -ENOMEM; 2777 2778 raw = page_address(page); 2779 i = first_sge; 2780 do { 2781 sge = sk_msg_elem(msg, i); 2782 from = sg_virt(sge); 2783 len = sge->length; 2784 to = raw + poffset; 2785 2786 memcpy(to, from, len); 2787 poffset += len; 2788 sge->length = 0; 2789 put_page(sg_page(sge)); 2790 sk_msg_clear_elem_copy(msg, i); 2791 2792 sk_msg_iter_var_next(i); 2793 } while (i != last_sge); 2794 2795 sg_set_page(&msg->sg.data[first_sge], page, copy, 0); 2796 sk_msg_clear_elem_copy(msg, first_sge); 2797 2798 /* To repair sg ring we need to shift entries. If we only 2799 * had a single entry though we can just replace it and 2800 * be done. Otherwise walk the ring and shift the entries. 2801 */ 2802 WARN_ON_ONCE(last_sge == first_sge); 2803 shift = last_sge > first_sge ? 2804 last_sge - first_sge - 1 : 2805 NR_MSG_FRAG_IDS - first_sge + last_sge - 1; 2806 if (!shift) { 2807 sk_msg_clear_elem_copy(msg, msg->sg.end); 2808 goto out; 2809 } 2810 2811 i = first_sge; 2812 sk_msg_iter_var_next(i); 2813 sk_msg_clear_copy_range(msg, i, last_sge); 2814 2815 i = first_sge; 2816 sk_msg_iter_var_next(i); 2817 do { 2818 u32 move_from; 2819 2820 if (i + shift >= NR_MSG_FRAG_IDS) 2821 move_from = i + shift - NR_MSG_FRAG_IDS; 2822 else 2823 move_from = i + shift; 2824 if (move_from == msg->sg.end) 2825 break; 2826 2827 sk_msg_sg_move(msg, i, move_from); 2828 msg->sg.data[move_from].length = 0; 2829 msg->sg.data[move_from].page_link = 0; 2830 msg->sg.data[move_from].offset = 0; 2831 sk_msg_clear_elem_copy(msg, move_from); 2832 sk_msg_iter_var_next(i); 2833 } while (1); 2834 2835 msg->sg.end = msg->sg.end - shift > msg->sg.end ? 2836 msg->sg.end - shift + NR_MSG_FRAG_IDS : 2837 msg->sg.end - shift; 2838 sk_msg_clear_elem_copy(msg, msg->sg.end); 2839 out: 2840 sk_msg_reset_curr(msg); 2841 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset; 2842 msg->data_end = msg->data + bytes; 2843 return 0; 2844 } 2845 2846 static const struct bpf_func_proto bpf_msg_pull_data_proto = { 2847 .func = bpf_msg_pull_data, 2848 .gpl_only = false, 2849 .ret_type = RET_INTEGER, 2850 .arg1_type = ARG_PTR_TO_CTX, 2851 .arg2_type = ARG_ANYTHING, 2852 .arg3_type = ARG_ANYTHING, 2853 .arg4_type = ARG_ANYTHING, 2854 }; 2855 2856 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start, 2857 u32, len, u64, flags) 2858 { 2859 bool sge_copy = false, nsge_copy = false, nnsge_copy = false; 2860 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge; 2861 u32 new, i = 0, l = 0, space, copy = 0, offset = 0; 2862 bool rsge_copy = false; 2863 u8 *raw, *to, *from; 2864 struct page *page; 2865 2866 if (unlikely(flags)) 2867 return -EINVAL; 2868 2869 if (unlikely(len == 0)) 2870 return 0; 2871 2872 /* First find the starting scatterlist element */ 2873 i = msg->sg.start; 2874 do { 2875 offset += l; 2876 l = sk_msg_elem(msg, i)->length; 2877 2878 if (start < offset + l) 2879 break; 2880 sk_msg_iter_var_next(i); 2881 } while (i != msg->sg.end); 2882 2883 if (start > offset + l) 2884 return -EINVAL; 2885 2886 /* A push at the end of the msg works on the last element */ 2887 if (i == msg->sg.end && start != offset) 2888 sk_msg_iter_var_prev(i); 2889 2890 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2891 2892 /* If no space available will fallback to copy, we need at 2893 * least one scatterlist elem available to push data into 2894 * when start aligns to the beginning of an element or two 2895 * when it falls inside an element. We handle the start equals 2896 * offset case because its the common case for inserting a 2897 * header. 2898 */ 2899 if (!space || (space == 1 && start != offset && start < offset + l)) 2900 copy = msg->sg.data[i].length; 2901 2902 if (unlikely(copy + len < copy)) 2903 return -EINVAL; 2904 2905 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2906 get_order(copy + len)); 2907 if (unlikely(!page)) 2908 return -ENOMEM; 2909 2910 if (copy) { 2911 int front, back; 2912 2913 raw = page_address(page); 2914 2915 psge = sk_msg_elem(msg, i); 2916 front = start - offset; 2917 back = psge->length - front; 2918 from = sg_virt(psge); 2919 2920 if (front) 2921 memcpy(raw, from, front); 2922 2923 if (back) { 2924 from += front; 2925 to = raw + front + len; 2926 2927 memcpy(to, from, back); 2928 } 2929 2930 put_page(sg_page(psge)); 2931 new = i; 2932 goto place_new; 2933 } 2934 2935 if (start - offset) { 2936 psge = sk_msg_elem(msg, i); 2937 rsge = sk_msg_elem_cpy(msg, i); 2938 rsge_copy = sk_msg_elem_is_copy(msg, i); 2939 2940 psge->length = start - offset; 2941 rsge.length -= psge->length; 2942 rsge.offset += start - offset; 2943 2944 sk_msg_iter_var_next(i); 2945 sg_unmark_end(psge); 2946 sg_unmark_end(&rsge); 2947 } 2948 2949 /* Slot(s) to place newly allocated data */ 2950 sk_msg_iter_next(msg, end); 2951 new = i; 2952 sk_msg_iter_var_next(i); 2953 2954 if (i == msg->sg.end) { 2955 if (!rsge.length) 2956 goto place_new; 2957 sk_msg_iter_next(msg, end); 2958 goto place_new; 2959 } 2960 2961 /* Shift one or two slots as needed */ 2962 sge = sk_msg_elem_cpy(msg, new); 2963 sg_unmark_end(&sge); 2964 sge_copy = sk_msg_elem_is_copy(msg, new); 2965 2966 nsge = sk_msg_elem_cpy(msg, i); 2967 nsge_copy = sk_msg_elem_is_copy(msg, i); 2968 if (rsge.length) { 2969 sk_msg_iter_var_next(i); 2970 nnsge = sk_msg_elem_cpy(msg, i); 2971 nnsge_copy = sk_msg_elem_is_copy(msg, i); 2972 sk_msg_iter_next(msg, end); 2973 } 2974 2975 while (i != msg->sg.end) { 2976 msg->sg.data[i] = sge; 2977 sk_msg_set_elem_copy(msg, i, sge_copy); 2978 sge = nsge; 2979 sge_copy = nsge_copy; 2980 sk_msg_iter_var_next(i); 2981 if (rsge.length) { 2982 nsge = nnsge; 2983 nsge_copy = nnsge_copy; 2984 nnsge = sk_msg_elem_cpy(msg, i); 2985 nnsge_copy = sk_msg_elem_is_copy(msg, i); 2986 } else { 2987 nsge = sk_msg_elem_cpy(msg, i); 2988 nsge_copy = sk_msg_elem_is_copy(msg, i); 2989 } 2990 } 2991 2992 place_new: 2993 /* Place newly allocated data buffer */ 2994 sk_mem_charge(msg->sk, len); 2995 msg->sg.size += len; 2996 sk_msg_clear_elem_copy(msg, new); 2997 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2998 /* A free slot can have the end mark */ 2999 sg_unmark_end(&msg->sg.data[new]); 3000 if (rsge.length) { 3001 get_page(sg_page(&rsge)); 3002 sk_msg_iter_var_next(new); 3003 msg->sg.data[new] = rsge; 3004 sk_msg_set_elem_copy(msg, new, rsge_copy); 3005 } 3006 sk_msg_clear_elem_copy(msg, msg->sg.end); 3007 3008 sk_msg_reset_curr(msg); 3009 sk_msg_compute_data_pointers(msg); 3010 return 0; 3011 } 3012 3013 static const struct bpf_func_proto bpf_msg_push_data_proto = { 3014 .func = bpf_msg_push_data, 3015 .gpl_only = false, 3016 .ret_type = RET_INTEGER, 3017 .arg1_type = ARG_PTR_TO_CTX, 3018 .arg2_type = ARG_ANYTHING, 3019 .arg3_type = ARG_ANYTHING, 3020 .arg4_type = ARG_ANYTHING, 3021 }; 3022 3023 static void sk_msg_shift_left(struct sk_msg *msg, int i) 3024 { 3025 struct scatterlist *sge = sk_msg_elem(msg, i); 3026 int prev; 3027 3028 put_page(sg_page(sge)); 3029 do { 3030 prev = i; 3031 sk_msg_iter_var_next(i); 3032 sk_msg_sg_move(msg, prev, i); 3033 } while (i != msg->sg.end); 3034 3035 sk_msg_iter_prev(msg, end); 3036 sk_msg_clear_elem_copy(msg, msg->sg.end); 3037 } 3038 3039 static void sk_msg_shift_right(struct sk_msg *msg, int i) 3040 { 3041 struct scatterlist tmp, sge; 3042 bool tmp_copy, sge_copy; 3043 3044 sk_msg_iter_next(msg, end); 3045 sge = sk_msg_elem_cpy(msg, i); 3046 sge_copy = sk_msg_elem_is_copy(msg, i); 3047 sk_msg_iter_var_next(i); 3048 tmp = sk_msg_elem_cpy(msg, i); 3049 tmp_copy = sk_msg_elem_is_copy(msg, i); 3050 3051 while (i != msg->sg.end) { 3052 msg->sg.data[i] = sge; 3053 sk_msg_set_elem_copy(msg, i, sge_copy); 3054 sk_msg_iter_var_next(i); 3055 sge = tmp; 3056 sge_copy = tmp_copy; 3057 tmp = sk_msg_elem_cpy(msg, i); 3058 tmp_copy = sk_msg_elem_is_copy(msg, i); 3059 } 3060 sk_msg_clear_elem_copy(msg, msg->sg.end); 3061 } 3062 3063 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 3064 u32, len, u64, flags) 3065 { 3066 u32 i = 0, l = 0, space, offset = 0; 3067 u64 last = (u64)start + len; 3068 u32 pop; 3069 3070 if (unlikely(flags)) 3071 return -EINVAL; 3072 3073 if (unlikely(len == 0)) 3074 return 0; 3075 3076 /* First find the starting scatterlist element */ 3077 i = msg->sg.start; 3078 do { 3079 offset += l; 3080 l = sk_msg_elem(msg, i)->length; 3081 3082 if (start < offset + l) 3083 break; 3084 sk_msg_iter_var_next(i); 3085 } while (i != msg->sg.end); 3086 3087 /* Bounds checks: start and pop must be inside message */ 3088 if (start >= offset + l || last > msg->sg.size) 3089 return -EINVAL; 3090 3091 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 3092 3093 pop = len; 3094 /* --------------| offset 3095 * -| start |-------- len -------| 3096 * 3097 * |----- a ----|-------- pop -------|----- b ----| 3098 * |______________________________________________| length 3099 * 3100 * 3101 * a: region at front of scatter element to save 3102 * b: region at back of scatter element to save when length > A + pop 3103 * pop: region to pop from element, same as input 'pop' here will be 3104 * decremented below per iteration. 3105 * 3106 * Two top-level cases to handle when start != offset, first B is non 3107 * zero and second B is zero corresponding to when a pop includes more 3108 * than one element. 3109 * 3110 * Then if B is non-zero AND there is no space allocate space and 3111 * compact A, B regions into page. If there is space shift ring to 3112 * the right free'ing the next element in ring to place B, leaving 3113 * A untouched except to reduce length. 3114 */ 3115 if (start != offset) { 3116 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 3117 bool sge_copy = sk_msg_elem_is_copy(msg, i); 3118 int a = start - offset; 3119 int b = sge->length - pop - a; 3120 u32 sge_idx = i; 3121 3122 sk_msg_iter_var_next(i); 3123 3124 if (b > 0) { 3125 if (space) { 3126 sge->length = a; 3127 sk_msg_shift_right(msg, i); 3128 nsge = sk_msg_elem(msg, i); 3129 get_page(sg_page(sge)); 3130 sg_set_page(nsge, 3131 sg_page(sge), 3132 b, sge->offset + pop + a); 3133 sg_unmark_end(nsge); 3134 sk_msg_set_elem_copy(msg, i, sge_copy); 3135 } else { 3136 struct page *page, *orig; 3137 u8 *to, *from; 3138 3139 page = alloc_pages(__GFP_NOWARN | 3140 __GFP_COMP | GFP_ATOMIC, 3141 get_order(a + b)); 3142 if (unlikely(!page)) 3143 return -ENOMEM; 3144 3145 orig = sg_page(sge); 3146 from = sg_virt(sge); 3147 to = page_address(page); 3148 memcpy(to, from, a); 3149 memcpy(to + a, from + a + pop, b); 3150 sg_set_page(sge, page, a + b, 0); 3151 sk_msg_clear_elem_copy(msg, sge_idx); 3152 put_page(orig); 3153 } 3154 pop = 0; 3155 } else { 3156 pop -= (sge->length - a); 3157 sge->length = a; 3158 } 3159 } 3160 3161 /* From above the current layout _must_ be as follows, 3162 * 3163 * -| offset 3164 * -| start 3165 * 3166 * |---- pop ---|---------------- b ------------| 3167 * |____________________________________________| length 3168 * 3169 * Offset and start of the current msg elem are equal because in the 3170 * previous case we handled offset != start and either consumed the 3171 * entire element and advanced to the next element OR pop == 0. 3172 * 3173 * Two cases to handle here are first pop is less than the length 3174 * leaving some remainder b above. Simply adjust the element's layout 3175 * in this case. Or pop >= length of the element so that b = 0. In this 3176 * case advance to next element decrementing pop. 3177 */ 3178 while (pop) { 3179 struct scatterlist *sge = sk_msg_elem(msg, i); 3180 3181 if (pop < sge->length) { 3182 sge->length -= pop; 3183 sge->offset += pop; 3184 pop = 0; 3185 } else { 3186 pop -= sge->length; 3187 sk_msg_shift_left(msg, i); 3188 } 3189 } 3190 3191 sk_mem_uncharge(msg->sk, len - pop); 3192 msg->sg.size -= (len - pop); 3193 sk_msg_reset_curr(msg); 3194 sk_msg_compute_data_pointers(msg); 3195 return 0; 3196 } 3197 3198 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 3199 .func = bpf_msg_pop_data, 3200 .gpl_only = false, 3201 .ret_type = RET_INTEGER, 3202 .arg1_type = ARG_PTR_TO_CTX, 3203 .arg2_type = ARG_ANYTHING, 3204 .arg3_type = ARG_ANYTHING, 3205 .arg4_type = ARG_ANYTHING, 3206 }; 3207 3208 #ifdef CONFIG_CGROUP_NET_CLASSID 3209 BPF_CALL_0(bpf_get_cgroup_classid_curr) 3210 { 3211 return __task_get_classid(current); 3212 } 3213 3214 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { 3215 .func = bpf_get_cgroup_classid_curr, 3216 .gpl_only = false, 3217 .ret_type = RET_INTEGER, 3218 }; 3219 3220 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb) 3221 { 3222 struct sock *sk = skb_to_full_sk(skb); 3223 3224 if (!sk || !sk_fullsock(sk)) 3225 return 0; 3226 3227 return sock_cgroup_classid(&sk->sk_cgrp_data); 3228 } 3229 3230 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = { 3231 .func = bpf_skb_cgroup_classid, 3232 .gpl_only = false, 3233 .ret_type = RET_INTEGER, 3234 .arg1_type = ARG_PTR_TO_CTX, 3235 }; 3236 #endif 3237 3238 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 3239 { 3240 return task_get_classid(skb); 3241 } 3242 3243 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 3244 .func = bpf_get_cgroup_classid, 3245 .gpl_only = false, 3246 .ret_type = RET_INTEGER, 3247 .arg1_type = ARG_PTR_TO_CTX, 3248 }; 3249 3250 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 3251 { 3252 return dst_tclassid(skb); 3253 } 3254 3255 static const struct bpf_func_proto bpf_get_route_realm_proto = { 3256 .func = bpf_get_route_realm, 3257 .gpl_only = false, 3258 .ret_type = RET_INTEGER, 3259 .arg1_type = ARG_PTR_TO_CTX, 3260 }; 3261 3262 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 3263 { 3264 /* If skb_clear_hash() was called due to mangling, we can 3265 * trigger SW recalculation here. Later access to hash 3266 * can then use the inline skb->hash via context directly 3267 * instead of calling this helper again. 3268 */ 3269 return skb_get_hash(skb); 3270 } 3271 3272 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 3273 .func = bpf_get_hash_recalc, 3274 .gpl_only = false, 3275 .ret_type = RET_INTEGER, 3276 .arg1_type = ARG_PTR_TO_CTX, 3277 }; 3278 3279 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 3280 { 3281 /* After all direct packet write, this can be used once for 3282 * triggering a lazy recalc on next skb_get_hash() invocation. 3283 */ 3284 skb_clear_hash(skb); 3285 return 0; 3286 } 3287 3288 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 3289 .func = bpf_set_hash_invalid, 3290 .gpl_only = false, 3291 .ret_type = RET_INTEGER, 3292 .arg1_type = ARG_PTR_TO_CTX, 3293 }; 3294 3295 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 3296 { 3297 /* Set user specified hash as L4(+), so that it gets returned 3298 * on skb_get_hash() call unless BPF prog later on triggers a 3299 * skb_clear_hash(). 3300 */ 3301 __skb_set_sw_hash(skb, hash, true); 3302 return 0; 3303 } 3304 3305 static const struct bpf_func_proto bpf_set_hash_proto = { 3306 .func = bpf_set_hash, 3307 .gpl_only = false, 3308 .ret_type = RET_INTEGER, 3309 .arg1_type = ARG_PTR_TO_CTX, 3310 .arg2_type = ARG_ANYTHING, 3311 }; 3312 3313 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 3314 u16, vlan_tci) 3315 { 3316 int ret; 3317 3318 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 3319 vlan_proto != htons(ETH_P_8021AD))) 3320 vlan_proto = htons(ETH_P_8021Q); 3321 3322 bpf_push_mac_rcsum(skb); 3323 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 3324 bpf_pull_mac_rcsum(skb); 3325 skb_reset_mac_len(skb); 3326 3327 bpf_compute_data_pointers(skb); 3328 return ret; 3329 } 3330 3331 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 3332 .func = bpf_skb_vlan_push, 3333 .gpl_only = false, 3334 .ret_type = RET_INTEGER, 3335 .arg1_type = ARG_PTR_TO_CTX, 3336 .arg2_type = ARG_ANYTHING, 3337 .arg3_type = ARG_ANYTHING, 3338 }; 3339 3340 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 3341 { 3342 int ret; 3343 3344 bpf_push_mac_rcsum(skb); 3345 ret = skb_vlan_pop(skb); 3346 bpf_pull_mac_rcsum(skb); 3347 3348 bpf_compute_data_pointers(skb); 3349 return ret; 3350 } 3351 3352 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 3353 .func = bpf_skb_vlan_pop, 3354 .gpl_only = false, 3355 .ret_type = RET_INTEGER, 3356 .arg1_type = ARG_PTR_TO_CTX, 3357 }; 3358 3359 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 3360 { 3361 /* The bytes in front of @off are moved with skb->data. */ 3362 if (unlikely(off > skb->len || !pskb_may_pull(skb, off))) 3363 return -ENOMEM; 3364 3365 /* Caller already did skb_cow() with meta_len+len as headroom, 3366 * so no need to do it here. 3367 */ 3368 skb_push(skb, len); 3369 skb_postpush_data_move(skb, len, off); 3370 memset(skb->data + off, 0, len); 3371 3372 /* No skb_postpush_rcsum(skb, skb->data + off, len) 3373 * needed here as it does not change the skb->csum 3374 * result for checksum complete when summing over 3375 * zeroed blocks. 3376 */ 3377 return 0; 3378 } 3379 3380 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 3381 { 3382 void *old_data; 3383 3384 /* skb_ensure_writable() is not needed here, as we're 3385 * already working on an uncloned skb. The first test 3386 * keeps off + len from wrapping. 3387 */ 3388 if (unlikely(off > skb->len || !pskb_may_pull(skb, off + len))) 3389 return -ENOMEM; 3390 3391 old_data = skb->data; 3392 __skb_pull(skb, len); 3393 skb_postpull_rcsum(skb, old_data + off, len); 3394 skb_postpull_data_move(skb, len, off); 3395 3396 return 0; 3397 } 3398 3399 /* 3400 * bpf_skb_net_hdr_push() inserts bytes at @off, bpf_skb_net_hdr_pop() removes 3401 * bytes at @off. Both move the bytes in front of @off together with skb->data, 3402 * and leave the rest of the packet where it is. Return true if the transport 3403 * header is in the part that moves. A transport header that is equal to the 3404 * network header stays equal to it. 3405 */ 3406 static bool bpf_skb_trans_hdr_moves(const struct sk_buff *skb, u32 off) 3407 { 3408 if (skb->transport_header == skb->network_header) 3409 return true; 3410 3411 return skb_transport_header_was_set(skb) && 3412 (u32)skb_transport_offset(skb) < off; 3413 } 3414 3415 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 3416 { 3417 bool trans_moves = bpf_skb_trans_hdr_moves(skb, off); 3418 int ret; 3419 3420 /* There's no need for __skb_push()/__skb_pull() pair to 3421 * get to the start of the mac header as we're guaranteed 3422 * to always start from here under eBPF. 3423 */ 3424 ret = bpf_skb_generic_push(skb, off, len); 3425 if (likely(!ret)) { 3426 skb->mac_header -= len; 3427 skb->network_header -= len; 3428 if (trans_moves) 3429 skb->transport_header -= len; 3430 } 3431 3432 return ret; 3433 } 3434 3435 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 3436 { 3437 bool trans_moves = bpf_skb_trans_hdr_moves(skb, off); 3438 int ret; 3439 3440 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 3441 ret = bpf_skb_generic_pop(skb, off, len); 3442 if (likely(!ret)) { 3443 skb->mac_header += len; 3444 skb->network_header += len; 3445 if (trans_moves) { 3446 skb->transport_header += len; 3447 } else if (skb_transport_header_was_set(skb) && 3448 skb_transport_offset(skb) < (int)off) { 3449 /* It was in the removed bytes or before skb->data. */ 3450 skb_set_transport_header(skb, off); 3451 } 3452 /* Do not leave the inner transport header in front of @off. */ 3453 if (skb->encapsulation && 3454 skb_inner_transport_offset(skb) < (int)off) 3455 skb_set_inner_transport_header(skb, off); 3456 } 3457 3458 return ret; 3459 } 3460 3461 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 3462 { 3463 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3464 const u8 meta_len = skb_metadata_len(skb); 3465 u32 off = skb_mac_header_len(skb); 3466 int ret; 3467 3468 ret = skb_cow(skb, meta_len + len_diff); 3469 if (unlikely(ret < 0)) 3470 return ret; 3471 3472 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3473 if (unlikely(ret < 0)) 3474 return ret; 3475 3476 if (skb_is_gso(skb)) { 3477 struct skb_shared_info *shinfo = skb_shinfo(skb); 3478 3479 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */ 3480 if (shinfo->gso_type & SKB_GSO_TCPV4) { 3481 shinfo->gso_type &= ~SKB_GSO_TCPV4; 3482 shinfo->gso_type |= SKB_GSO_TCPV6; 3483 } 3484 shinfo->gso_type |= SKB_GSO_DODGY; 3485 } 3486 3487 skb->protocol = htons(ETH_P_IPV6); 3488 skb_clear_hash(skb); 3489 3490 return 0; 3491 } 3492 3493 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 3494 { 3495 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3496 u32 off = skb_mac_header_len(skb); 3497 int ret; 3498 3499 ret = skb_unclone(skb, GFP_ATOMIC); 3500 if (unlikely(ret < 0)) 3501 return ret; 3502 3503 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3504 if (unlikely(ret < 0)) 3505 return ret; 3506 3507 if (skb_is_gso(skb)) { 3508 struct skb_shared_info *shinfo = skb_shinfo(skb); 3509 3510 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */ 3511 if (shinfo->gso_type & SKB_GSO_TCPV6) { 3512 shinfo->gso_type &= ~SKB_GSO_TCPV6; 3513 shinfo->gso_type |= SKB_GSO_TCPV4; 3514 } 3515 shinfo->gso_type |= SKB_GSO_DODGY; 3516 } 3517 3518 skb->protocol = htons(ETH_P_IP); 3519 skb_clear_hash(skb); 3520 3521 return 0; 3522 } 3523 3524 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 3525 { 3526 __be16 from_proto = skb->protocol; 3527 3528 if (from_proto == htons(ETH_P_IP) && 3529 to_proto == htons(ETH_P_IPV6)) 3530 return bpf_skb_proto_4_to_6(skb); 3531 3532 if (from_proto == htons(ETH_P_IPV6) && 3533 to_proto == htons(ETH_P_IP)) 3534 return bpf_skb_proto_6_to_4(skb); 3535 3536 return -ENOTSUPP; 3537 } 3538 3539 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 3540 u64, flags) 3541 { 3542 int ret; 3543 3544 if (unlikely(flags)) 3545 return -EINVAL; 3546 3547 /* General idea is that this helper does the basic groundwork 3548 * needed for changing the protocol, and eBPF program fills the 3549 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 3550 * and other helpers, rather than passing a raw buffer here. 3551 * 3552 * The rationale is to keep this minimal and without a need to 3553 * deal with raw packet data. F.e. even if we would pass buffers 3554 * here, the program still needs to call the bpf_lX_csum_replace() 3555 * helpers anyway. Plus, this way we keep also separation of 3556 * concerns, since f.e. bpf_skb_store_bytes() should only take 3557 * care of stores. 3558 * 3559 * Currently, additional options and extension header space are 3560 * not supported, but flags register is reserved so we can adapt 3561 * that. For offloads, we mark packet as dodgy, so that headers 3562 * need to be verified first. 3563 */ 3564 ret = bpf_skb_proto_xlat(skb, proto); 3565 bpf_compute_data_pointers(skb); 3566 if (ret) 3567 return ret; 3568 3569 if (skb_valid_dst(skb)) 3570 skb_dst_drop(skb); 3571 3572 return 0; 3573 } 3574 3575 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 3576 .func = bpf_skb_change_proto, 3577 .gpl_only = false, 3578 .ret_type = RET_INTEGER, 3579 .arg1_type = ARG_PTR_TO_CTX, 3580 .arg2_type = ARG_ANYTHING, 3581 .arg3_type = ARG_ANYTHING, 3582 }; 3583 3584 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 3585 { 3586 /* We only allow a restricted subset to be changed for now. */ 3587 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 3588 !skb_pkt_type_ok(pkt_type))) 3589 return -EINVAL; 3590 3591 skb->pkt_type = pkt_type; 3592 return 0; 3593 } 3594 3595 static const struct bpf_func_proto bpf_skb_change_type_proto = { 3596 .func = bpf_skb_change_type, 3597 .gpl_only = false, 3598 .ret_type = RET_INTEGER, 3599 .arg1_type = ARG_PTR_TO_CTX, 3600 .arg2_type = ARG_ANYTHING, 3601 }; 3602 3603 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3604 { 3605 switch (skb->protocol) { 3606 case htons(ETH_P_IP): 3607 return sizeof(struct iphdr); 3608 case htons(ETH_P_IPV6): 3609 return sizeof(struct ipv6hdr); 3610 default: 3611 return ~0U; 3612 } 3613 } 3614 3615 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3616 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3617 3618 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \ 3619 BPF_F_ADJ_ROOM_DECAP_L3_IPV6) 3620 3621 #define BPF_F_ADJ_ROOM_DECAP_L4_MASK (BPF_F_ADJ_ROOM_DECAP_L4_UDP | \ 3622 BPF_F_ADJ_ROOM_DECAP_L4_GRE) 3623 3624 #define BPF_F_ADJ_ROOM_DECAP_IPXIP_MASK (BPF_F_ADJ_ROOM_DECAP_IPXIP4 | \ 3625 BPF_F_ADJ_ROOM_DECAP_IPXIP6) 3626 3627 #define BPF_F_ADJ_ROOM_ENCAP_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3628 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3629 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3630 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \ 3631 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3632 BPF_ADJ_ROOM_ENCAP_L2_MASK)) 3633 3634 #define BPF_F_ADJ_ROOM_DECAP_MASK (BPF_F_ADJ_ROOM_DECAP_L3_MASK | \ 3635 BPF_F_ADJ_ROOM_DECAP_L4_MASK | \ 3636 BPF_F_ADJ_ROOM_DECAP_IPXIP_MASK) 3637 3638 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3639 BPF_F_ADJ_ROOM_ENCAP_MASK | \ 3640 BPF_F_ADJ_ROOM_DECAP_MASK | \ 3641 BPF_F_ADJ_ROOM_NO_CSUM_RESET) 3642 3643 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3644 u64 flags) 3645 { 3646 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3647 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3648 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3649 const u8 meta_len = skb_metadata_len(skb); 3650 unsigned int gso_type = SKB_GSO_DODGY; 3651 int ret; 3652 3653 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3654 /* udp gso_size delineates datagrams, only allow if fixed */ 3655 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3656 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3657 return -ENOTSUPP; 3658 } 3659 3660 /* 3661 * The MSS is lowered by len_diff. Leave the least that the stack lets a 3662 * TCP GSO packet have, or gso_segs can overflow. 3663 */ 3664 if (skb_is_gso(skb) && !(flags & BPF_F_ADJ_ROOM_FIXED_GSO) && 3665 skb_shinfo(skb)->gso_size < len_diff + TCP_MIN_GSO_SIZE) 3666 return -ENOTSUPP; 3667 3668 ret = skb_cow_head(skb, meta_len + len_diff); 3669 if (unlikely(ret < 0)) 3670 return ret; 3671 3672 if (encap) { 3673 if (skb->protocol != htons(ETH_P_IP) && 3674 skb->protocol != htons(ETH_P_IPV6)) 3675 return -ENOTSUPP; 3676 3677 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3678 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3679 return -EINVAL; 3680 3681 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3682 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3683 return -EINVAL; 3684 3685 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH && 3686 inner_mac_len < ETH_HLEN) 3687 return -EINVAL; 3688 3689 if (skb->encapsulation) 3690 return -EALREADY; 3691 3692 mac_len = skb->network_header - skb->mac_header; 3693 inner_net = skb->network_header; 3694 if (inner_mac_len > len_diff) 3695 return -EINVAL; 3696 inner_trans = skb->transport_header; 3697 } 3698 3699 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3700 if (unlikely(ret < 0)) 3701 return ret; 3702 3703 if (encap) { 3704 skb->inner_mac_header = inner_net - inner_mac_len; 3705 skb->inner_network_header = inner_net; 3706 skb->inner_transport_header = inner_trans; 3707 3708 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH) 3709 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 3710 else 3711 skb_set_inner_protocol(skb, skb->protocol); 3712 3713 skb->encapsulation = 1; 3714 skb_set_network_header(skb, mac_len); 3715 3716 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3717 gso_type |= SKB_GSO_UDP_TUNNEL; 3718 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3719 gso_type |= SKB_GSO_GRE; 3720 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3721 gso_type |= SKB_GSO_IPXIP6; 3722 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3723 gso_type |= SKB_GSO_IPXIP4; 3724 3725 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3726 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3727 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3728 sizeof(struct ipv6hdr) : 3729 sizeof(struct iphdr); 3730 3731 skb_set_transport_header(skb, mac_len + nh_len); 3732 } 3733 3734 /* Match skb->protocol to new outer l3 protocol */ 3735 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3736 skb->protocol = htons(ETH_P_IPV6); 3737 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3738 skb->protocol = htons(ETH_P_IP); 3739 3740 if (skb_valid_dst(skb)) 3741 skb_dst_drop(skb); 3742 } 3743 3744 if (skb_is_gso(skb)) { 3745 struct skb_shared_info *shinfo = skb_shinfo(skb); 3746 3747 /* Header must be checked, and gso_segs recomputed. */ 3748 shinfo->gso_type |= gso_type; 3749 shinfo->gso_segs = 0; 3750 3751 /* Due to header growth, MSS needs to be downgraded. 3752 * There is a BUG_ON() when segmenting the frag_list with 3753 * head_frag true, so linearize the skb after downgrading 3754 * the MSS. 3755 */ 3756 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) { 3757 skb_decrease_gso_size(shinfo, len_diff); 3758 if (shinfo->frag_list) 3759 return skb_linearize(skb); 3760 } 3761 } 3762 3763 return 0; 3764 } 3765 3766 /* 3767 * Code that went by the inner transport header while skb->encapsulation was set 3768 * goes by the transport header from now on. So take the inner one over, unless 3769 * it is the smaller one, as it is if nobody has set it. Do not leave the result 3770 * in front of @off, where the bytes were removed. 3771 */ 3772 static void bpf_skb_clear_encap(struct sk_buff *skb, u32 off) 3773 { 3774 int thoff; 3775 3776 if (!skb->encapsulation) 3777 return; 3778 3779 thoff = max(skb->inner_transport_header, skb->transport_header) - 3780 skb_headroom(skb); 3781 skb_set_transport_header(skb, max_t(int, thoff, off)); 3782 skb->encapsulation = 0; 3783 } 3784 3785 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3786 u64 flags) 3787 { 3788 bool decap = flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK; 3789 int ret; 3790 3791 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_DECAP_MASK | 3792 BPF_F_ADJ_ROOM_FIXED_GSO | 3793 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3794 return -EINVAL; 3795 3796 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3797 /* udp gso_size delineates datagrams, only allow if fixed */ 3798 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3799 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3800 return -ENOTSUPP; 3801 } 3802 3803 /* 3804 * The MSS is raised by len_diff. It has 16 bits, and GSO_BY_FRAGS is 3805 * not a size. 3806 */ 3807 if (skb_is_gso(skb) && !(flags & BPF_F_ADJ_ROOM_FIXED_GSO) && 3808 skb_shinfo(skb)->gso_size + len_diff >= GSO_BY_FRAGS) 3809 return -ENOTSUPP; 3810 3811 ret = skb_unclone(skb, GFP_ATOMIC); 3812 if (unlikely(ret < 0)) 3813 return ret; 3814 3815 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3816 if (unlikely(ret < 0)) 3817 return ret; 3818 3819 if (decap) { 3820 /* Match skb->protocol to new outer l3 protocol */ 3821 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6) 3822 skb->protocol = htons(ETH_P_IPV6); 3823 else if (flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4) 3824 skb->protocol = htons(ETH_P_IP); 3825 3826 if (skb_valid_dst(skb)) 3827 skb_dst_drop(skb); 3828 } 3829 3830 if (skb_is_gso(skb)) { 3831 struct skb_shared_info *shinfo = skb_shinfo(skb); 3832 3833 /* Due to header shrink, MSS can be upgraded. */ 3834 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3835 skb_increase_gso_size(shinfo, len_diff); 3836 3837 /* Selective GSO flag clearing based on decap type. 3838 * Only clear the flags for the tunnel layer being removed. 3839 */ 3840 if ((flags & BPF_F_ADJ_ROOM_DECAP_L4_UDP) && 3841 (shinfo->gso_type & (SKB_GSO_UDP_TUNNEL | 3842 SKB_GSO_UDP_TUNNEL_CSUM))) 3843 shinfo->gso_type &= ~(SKB_GSO_UDP_TUNNEL | 3844 SKB_GSO_UDP_TUNNEL_CSUM); 3845 if ((flags & BPF_F_ADJ_ROOM_DECAP_L4_GRE) && 3846 (shinfo->gso_type & (SKB_GSO_GRE | SKB_GSO_GRE_CSUM))) 3847 shinfo->gso_type &= ~(SKB_GSO_GRE | 3848 SKB_GSO_GRE_CSUM); 3849 if ((flags & BPF_F_ADJ_ROOM_DECAP_IPXIP4) && 3850 (shinfo->gso_type & SKB_GSO_IPXIP4)) 3851 shinfo->gso_type &= ~SKB_GSO_IPXIP4; 3852 if ((flags & BPF_F_ADJ_ROOM_DECAP_IPXIP6) && 3853 (shinfo->gso_type & SKB_GSO_IPXIP6)) 3854 shinfo->gso_type &= ~SKB_GSO_IPXIP6; 3855 3856 /* Clear encapsulation flag only when no tunnel GSO flags remain */ 3857 if (flags & (BPF_F_ADJ_ROOM_DECAP_L4_MASK | 3858 BPF_F_ADJ_ROOM_DECAP_IPXIP_MASK)) { 3859 if (!(shinfo->gso_type & (SKB_GSO_UDP_TUNNEL | 3860 SKB_GSO_UDP_TUNNEL_CSUM | 3861 SKB_GSO_GRE | 3862 SKB_GSO_GRE_CSUM | 3863 SKB_GSO_IPXIP4 | 3864 SKB_GSO_IPXIP6 | 3865 SKB_GSO_ESP))) 3866 bpf_skb_clear_encap(skb, off); 3867 } 3868 3869 /* Header must be checked, and gso_segs recomputed. */ 3870 shinfo->gso_type |= SKB_GSO_DODGY; 3871 shinfo->gso_segs = 0; 3872 } else { 3873 /* For non-GSO packets, clear encapsulation if decap flags are set */ 3874 if (flags & (BPF_F_ADJ_ROOM_DECAP_L4_MASK | 3875 BPF_F_ADJ_ROOM_DECAP_IPXIP_MASK)) 3876 bpf_skb_clear_encap(skb, off); 3877 } 3878 3879 return 0; 3880 } 3881 3882 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC 3883 3884 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3885 u32, mode, u64, flags) 3886 { 3887 u32 len_diff_abs = abs(len_diff); 3888 bool shrink = len_diff < 0; 3889 int ret = 0; 3890 3891 if (unlikely(flags || mode)) 3892 return -EINVAL; 3893 if (unlikely(len_diff_abs > 0xfffU)) 3894 return -EFAULT; 3895 3896 if (!shrink) { 3897 ret = skb_cow(skb, len_diff); 3898 if (unlikely(ret < 0)) 3899 return ret; 3900 __skb_push(skb, len_diff_abs); 3901 memset(skb->data, 0, len_diff_abs); 3902 } else { 3903 if (unlikely(!pskb_may_pull(skb, len_diff_abs))) 3904 return -ENOMEM; 3905 __skb_pull(skb, len_diff_abs); 3906 } 3907 if (tls_sw_has_ctx_rx(skb->sk)) { 3908 struct strp_msg *rxm = strp_msg(skb); 3909 3910 rxm->full_len += len_diff; 3911 } 3912 return ret; 3913 } 3914 3915 static const struct bpf_func_proto sk_skb_adjust_room_proto = { 3916 .func = sk_skb_adjust_room, 3917 .gpl_only = false, 3918 .ret_type = RET_INTEGER, 3919 .arg1_type = ARG_PTR_TO_CTX, 3920 .arg2_type = ARG_ANYTHING, 3921 .arg3_type = ARG_ANYTHING, 3922 .arg4_type = ARG_ANYTHING, 3923 }; 3924 3925 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3926 u32, mode, u64, flags) 3927 { 3928 u32 len_cur, len_diff_abs = abs(len_diff); 3929 u32 len_min = bpf_skb_net_base_len(skb); 3930 u32 len_max = BPF_SKB_MAX_LEN; 3931 __be16 proto = skb->protocol; 3932 bool shrink = len_diff < 0; 3933 u32 off; 3934 int ret; 3935 3936 if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK)) 3937 return -EINVAL; 3938 if (unlikely(len_diff_abs > 0xfffU)) 3939 return -EFAULT; 3940 if (unlikely(proto != htons(ETH_P_IP) && 3941 proto != htons(ETH_P_IPV6))) 3942 return -ENOTSUPP; 3943 3944 off = skb_mac_header_len(skb); 3945 switch (mode) { 3946 case BPF_ADJ_ROOM_NET: 3947 off += bpf_skb_net_base_len(skb); 3948 break; 3949 case BPF_ADJ_ROOM_MAC: 3950 break; 3951 default: 3952 return -ENOTSUPP; 3953 } 3954 3955 if (flags & BPF_F_ADJ_ROOM_DECAP_MASK) { 3956 u32 len_decap_min = 0; 3957 3958 if (!shrink) 3959 return -EINVAL; 3960 3961 /* Reject mutually exclusive decap flag pairs. */ 3962 if ((flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) == 3963 BPF_F_ADJ_ROOM_DECAP_L3_MASK) 3964 return -EINVAL; 3965 3966 if ((flags & BPF_F_ADJ_ROOM_DECAP_L4_MASK) == 3967 BPF_F_ADJ_ROOM_DECAP_L4_MASK) 3968 return -EINVAL; 3969 3970 if ((flags & BPF_F_ADJ_ROOM_DECAP_IPXIP_MASK) == 3971 BPF_F_ADJ_ROOM_DECAP_IPXIP_MASK) 3972 return -EINVAL; 3973 3974 /* Reject mutually exclusive decap tunnel type flags. */ 3975 if ((flags & BPF_F_ADJ_ROOM_DECAP_L4_MASK) && 3976 (flags & BPF_F_ADJ_ROOM_DECAP_IPXIP_MASK)) 3977 return -EINVAL; 3978 3979 if (flags & BPF_F_ADJ_ROOM_DECAP_L4_MASK) 3980 len_decap_min += bpf_skb_net_base_len(skb); 3981 3982 if (flags & BPF_F_ADJ_ROOM_DECAP_L4_UDP) 3983 len_decap_min += sizeof(struct udphdr); 3984 3985 if (flags & BPF_F_ADJ_ROOM_DECAP_L4_GRE) 3986 len_decap_min += sizeof(struct gre_base_hdr); 3987 3988 if (flags & BPF_F_ADJ_ROOM_DECAP_IPXIP4) 3989 len_decap_min += sizeof(struct iphdr); 3990 3991 if (flags & BPF_F_ADJ_ROOM_DECAP_IPXIP6) 3992 len_decap_min += sizeof(struct ipv6hdr); 3993 3994 if (len_diff_abs < len_decap_min) 3995 return -EINVAL; 3996 3997 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4) 3998 len_min = sizeof(struct iphdr); 3999 4000 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6) 4001 len_min = sizeof(struct ipv6hdr); 4002 } 4003 4004 len_cur = skb->len - skb_network_offset(skb); 4005 if ((shrink && (len_diff_abs >= len_cur || 4006 len_cur - len_diff_abs < len_min)) || 4007 (!shrink && (skb->len + len_diff_abs > len_max && 4008 !skb_is_gso(skb)))) 4009 return -ENOTSUPP; 4010 4011 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 4012 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 4013 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET)) 4014 __skb_reset_checksum_unnecessary(skb); 4015 4016 bpf_compute_data_pointers(skb); 4017 return ret; 4018 } 4019 4020 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 4021 .func = bpf_skb_adjust_room, 4022 .gpl_only = false, 4023 .ret_type = RET_INTEGER, 4024 .arg1_type = ARG_PTR_TO_CTX, 4025 .arg2_type = ARG_ANYTHING, 4026 .arg3_type = ARG_ANYTHING, 4027 .arg4_type = ARG_ANYTHING, 4028 }; 4029 4030 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 4031 { 4032 int offset = skb_network_offset(skb); 4033 u32 min_len = 0; 4034 4035 if (offset > 0) 4036 min_len = offset; 4037 if (skb_transport_header_was_set(skb)) { 4038 offset = skb_transport_offset(skb); 4039 if (offset > 0) 4040 min_len = offset; 4041 } 4042 return min_len; 4043 } 4044 4045 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 4046 { 4047 unsigned int old_len = skb->len; 4048 int ret; 4049 4050 ret = __skb_grow_rcsum(skb, new_len); 4051 if (!ret) 4052 memset(skb->data + old_len, 0, new_len - old_len); 4053 return ret; 4054 } 4055 4056 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 4057 { 4058 if (skb->ip_summed == CHECKSUM_PARTIAL && 4059 new_len < skb_checksum_start_offset(skb) + skb->csum_offset + 4060 sizeof(__sum16)) 4061 skb->ip_summed = CHECKSUM_NONE; 4062 4063 return __skb_trim_rcsum(skb, new_len); 4064 } 4065 4066 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 4067 u64 flags) 4068 { 4069 u32 max_len = BPF_SKB_MAX_LEN; 4070 u32 min_len = __bpf_skb_min_len(skb); 4071 int ret; 4072 4073 if (unlikely(flags || new_len > max_len || new_len < min_len)) 4074 return -EINVAL; 4075 if (skb->encapsulation) 4076 return -ENOTSUPP; 4077 4078 /* The basic idea of this helper is that it's performing the 4079 * needed work to either grow or trim an skb, and eBPF program 4080 * rewrites the rest via helpers like bpf_skb_store_bytes(), 4081 * bpf_lX_csum_replace() and others rather than passing a raw 4082 * buffer here. This one is a slow path helper and intended 4083 * for replies with control messages. 4084 * 4085 * Like in bpf_skb_change_proto(), we want to keep this rather 4086 * minimal and without protocol specifics so that we are able 4087 * to separate concerns as in bpf_skb_store_bytes() should only 4088 * be the one responsible for writing buffers. 4089 * 4090 * It's really expected to be a slow path operation here for 4091 * control message replies, so we're implicitly linearizing, 4092 * uncloning and drop offloads from the skb by this. 4093 */ 4094 ret = __bpf_try_make_writable(skb, skb->len); 4095 if (!ret) { 4096 if (new_len > skb->len) 4097 ret = bpf_skb_grow_rcsum(skb, new_len); 4098 else if (new_len < skb->len) 4099 ret = bpf_skb_trim_rcsum(skb, new_len); 4100 if (!ret && skb_is_gso(skb)) 4101 skb_gso_reset(skb); 4102 } 4103 return ret; 4104 } 4105 4106 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 4107 u64, flags) 4108 { 4109 int ret = __bpf_skb_change_tail(skb, new_len, flags); 4110 4111 bpf_compute_data_pointers(skb); 4112 return ret; 4113 } 4114 4115 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 4116 .func = bpf_skb_change_tail, 4117 .gpl_only = false, 4118 .ret_type = RET_INTEGER, 4119 .arg1_type = ARG_PTR_TO_CTX, 4120 .arg2_type = ARG_ANYTHING, 4121 .arg3_type = ARG_ANYTHING, 4122 }; 4123 4124 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 4125 u64, flags) 4126 { 4127 return __bpf_skb_change_tail(skb, new_len, flags); 4128 } 4129 4130 static const struct bpf_func_proto sk_skb_change_tail_proto = { 4131 .func = sk_skb_change_tail, 4132 .gpl_only = false, 4133 .ret_type = RET_INTEGER, 4134 .arg1_type = ARG_PTR_TO_CTX, 4135 .arg2_type = ARG_ANYTHING, 4136 .arg3_type = ARG_ANYTHING, 4137 }; 4138 4139 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 4140 u64 flags) 4141 { 4142 const u8 meta_len = skb_metadata_len(skb); 4143 u32 max_len = BPF_SKB_MAX_LEN; 4144 u32 new_len = skb->len + head_room; 4145 int ret; 4146 4147 if (unlikely(flags || (int)head_room < 0 || 4148 (!skb_is_gso(skb) && new_len > max_len) || 4149 new_len < skb->len)) 4150 return -EINVAL; 4151 4152 ret = skb_cow(skb, meta_len + head_room); 4153 if (likely(!ret)) { 4154 /* Idea for this helper is that we currently only 4155 * allow to expand on mac header. This means that 4156 * skb->protocol network header, etc, stay as is. 4157 * Compared to bpf_skb_change_tail(), we're more 4158 * flexible due to not needing to linearize or 4159 * reset GSO. Intention for this helper is to be 4160 * used by an L3 skb that needs to push mac header 4161 * for redirection into L2 device. 4162 */ 4163 __skb_push(skb, head_room); 4164 skb_postpush_data_move(skb, head_room, 0); 4165 memset(skb->data, 0, head_room); 4166 skb_reset_mac_header(skb); 4167 skb_reset_mac_len(skb); 4168 } 4169 4170 return ret; 4171 } 4172 4173 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 4174 u64, flags) 4175 { 4176 int ret = __bpf_skb_change_head(skb, head_room, flags); 4177 4178 bpf_compute_data_pointers(skb); 4179 return ret; 4180 } 4181 4182 static const struct bpf_func_proto bpf_skb_change_head_proto = { 4183 .func = bpf_skb_change_head, 4184 .gpl_only = false, 4185 .ret_type = RET_INTEGER, 4186 .arg1_type = ARG_PTR_TO_CTX, 4187 .arg2_type = ARG_ANYTHING, 4188 .arg3_type = ARG_ANYTHING, 4189 }; 4190 4191 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 4192 u64, flags) 4193 { 4194 return __bpf_skb_change_head(skb, head_room, flags); 4195 } 4196 4197 static const struct bpf_func_proto sk_skb_change_head_proto = { 4198 .func = sk_skb_change_head, 4199 .gpl_only = false, 4200 .ret_type = RET_INTEGER, 4201 .arg1_type = ARG_PTR_TO_CTX, 4202 .arg2_type = ARG_ANYTHING, 4203 .arg3_type = ARG_ANYTHING, 4204 }; 4205 4206 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp) 4207 { 4208 return xdp_get_buff_len(xdp); 4209 } 4210 4211 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = { 4212 .func = bpf_xdp_get_buff_len, 4213 .gpl_only = false, 4214 .ret_type = RET_INTEGER, 4215 .arg1_type = ARG_PTR_TO_CTX, 4216 }; 4217 4218 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff) 4219 4220 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = { 4221 .func = bpf_xdp_get_buff_len, 4222 .gpl_only = false, 4223 .arg1_type = ARG_PTR_TO_BTF_ID, 4224 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0], 4225 }; 4226 4227 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 4228 { 4229 return xdp_data_meta_unsupported(xdp) ? 0 : 4230 xdp->data - xdp->data_meta; 4231 } 4232 4233 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 4234 { 4235 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 4236 unsigned long metalen = xdp_get_metalen(xdp); 4237 void *data_start = xdp_frame_end + metalen; 4238 void *data = xdp->data + offset; 4239 4240 if (unlikely(data < data_start || 4241 data > xdp->data_end - ETH_HLEN)) 4242 return -EINVAL; 4243 4244 if (metalen) 4245 memmove(xdp->data_meta + offset, 4246 xdp->data_meta, metalen); 4247 xdp->data_meta += offset; 4248 xdp->data = data; 4249 4250 return 0; 4251 } 4252 4253 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 4254 .func = bpf_xdp_adjust_head, 4255 .gpl_only = false, 4256 .ret_type = RET_INTEGER, 4257 .arg1_type = ARG_PTR_TO_CTX, 4258 .arg2_type = ARG_ANYTHING, 4259 }; 4260 4261 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 4262 void *buf, unsigned long len, bool flush) 4263 { 4264 unsigned long ptr_len, ptr_off = 0; 4265 skb_frag_t *next_frag, *end_frag; 4266 struct skb_shared_info *sinfo; 4267 void *src, *dst; 4268 u8 *ptr_buf; 4269 4270 if (likely(xdp->data_end - xdp->data >= off + len)) { 4271 src = flush ? buf : xdp->data + off; 4272 dst = flush ? xdp->data + off : buf; 4273 memcpy(dst, src, len); 4274 return; 4275 } 4276 4277 sinfo = xdp_get_shared_info_from_buff(xdp); 4278 end_frag = &sinfo->frags[sinfo->nr_frags]; 4279 next_frag = &sinfo->frags[0]; 4280 4281 ptr_len = xdp->data_end - xdp->data; 4282 ptr_buf = xdp->data; 4283 4284 while (true) { 4285 if (off < ptr_off + ptr_len) { 4286 unsigned long copy_off = off - ptr_off; 4287 unsigned long copy_len = min(len, ptr_len - copy_off); 4288 4289 src = flush ? buf : ptr_buf + copy_off; 4290 dst = flush ? ptr_buf + copy_off : buf; 4291 memcpy(dst, src, copy_len); 4292 4293 off += copy_len; 4294 len -= copy_len; 4295 buf += copy_len; 4296 } 4297 4298 if (!len || next_frag == end_frag) 4299 break; 4300 4301 ptr_off += ptr_len; 4302 ptr_buf = skb_frag_address(next_frag); 4303 ptr_len = skb_frag_size(next_frag); 4304 next_frag++; 4305 } 4306 } 4307 4308 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 4309 { 4310 u32 size = xdp->data_end - xdp->data; 4311 struct skb_shared_info *sinfo; 4312 void *addr = xdp->data; 4313 int i; 4314 4315 if (unlikely(offset > 0xffff || len > 0xffff)) 4316 return ERR_PTR(-EFAULT); 4317 4318 if (unlikely(offset + len > xdp_get_buff_len(xdp))) 4319 return ERR_PTR(-EINVAL); 4320 4321 if (likely(offset < size)) /* linear area */ 4322 goto out; 4323 4324 sinfo = xdp_get_shared_info_from_buff(xdp); 4325 offset -= size; 4326 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */ 4327 u32 frag_size = skb_frag_size(&sinfo->frags[i]); 4328 4329 if (offset < frag_size) { 4330 addr = skb_frag_address(&sinfo->frags[i]); 4331 size = frag_size; 4332 break; 4333 } 4334 offset -= frag_size; 4335 } 4336 out: 4337 return offset + len <= size ? addr + offset : NULL; 4338 } 4339 4340 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset, 4341 void *, buf, u32, len) 4342 { 4343 void *ptr; 4344 4345 ptr = bpf_xdp_pointer(xdp, offset, len); 4346 if (IS_ERR(ptr)) 4347 return PTR_ERR(ptr); 4348 4349 if (!ptr) 4350 bpf_xdp_copy_buf(xdp, offset, buf, len, false); 4351 else 4352 memmove(buf, ptr, len); 4353 4354 return 0; 4355 } 4356 4357 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = { 4358 .func = bpf_xdp_load_bytes, 4359 .gpl_only = false, 4360 .ret_type = RET_INTEGER, 4361 .arg1_type = ARG_PTR_TO_CTX, 4362 .arg2_type = ARG_ANYTHING, 4363 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 4364 .arg4_type = ARG_MEM_SIZE, 4365 }; 4366 4367 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len) 4368 { 4369 return ____bpf_xdp_load_bytes(xdp, offset, buf, len); 4370 } 4371 4372 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset, 4373 void *, buf, u32, len) 4374 { 4375 void *ptr; 4376 4377 ptr = bpf_xdp_pointer(xdp, offset, len); 4378 if (IS_ERR(ptr)) 4379 return PTR_ERR(ptr); 4380 4381 if (!ptr) 4382 bpf_xdp_copy_buf(xdp, offset, buf, len, true); 4383 else 4384 /* through a dynptr the source may be an overlapping slice */ 4385 memmove(ptr, buf, len); 4386 4387 return 0; 4388 } 4389 4390 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = { 4391 .func = bpf_xdp_store_bytes, 4392 .gpl_only = false, 4393 .ret_type = RET_INTEGER, 4394 .arg1_type = ARG_PTR_TO_CTX, 4395 .arg2_type = ARG_ANYTHING, 4396 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4397 .arg4_type = ARG_MEM_SIZE, 4398 }; 4399 4400 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len) 4401 { 4402 return ____bpf_xdp_store_bytes(xdp, offset, buf, len); 4403 } 4404 4405 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset) 4406 { 4407 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4408 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1]; 4409 struct xdp_rxq_info *rxq = xdp->rxq; 4410 int tailroom; 4411 4412 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz) 4413 return -EOPNOTSUPP; 4414 4415 tailroom = rxq->frag_size - skb_frag_size(frag) - 4416 skb_frag_off(frag) % rxq->frag_size; 4417 WARN_ON_ONCE(tailroom < 0); 4418 if (unlikely(offset > tailroom)) 4419 return -EINVAL; 4420 4421 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset); 4422 skb_frag_size_add(frag, offset); 4423 sinfo->xdp_frags_size += offset; 4424 if (rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL) 4425 xsk_buff_get_tail(xdp)->data_end += offset; 4426 4427 return 0; 4428 } 4429 4430 static struct xdp_buff *bpf_xdp_shrink_data_zc(struct xdp_buff *xdp, int shrink, 4431 bool tail, bool release) 4432 { 4433 struct xdp_buff *zc_frag = tail ? xsk_buff_get_tail(xdp) : 4434 xsk_buff_get_head(xdp); 4435 4436 if (release) { 4437 xsk_buff_del_frag(zc_frag); 4438 } else { 4439 if (tail) 4440 zc_frag->data_end -= shrink; 4441 else 4442 zc_frag->data += shrink; 4443 } 4444 4445 return zc_frag; 4446 } 4447 4448 static bool bpf_xdp_shrink_data(struct xdp_buff *xdp, skb_frag_t *frag, 4449 int shrink, bool tail) 4450 { 4451 enum xdp_mem_type mem_type = xdp->rxq->mem.type; 4452 bool release = skb_frag_size(frag) == shrink; 4453 netmem_ref netmem = skb_frag_netmem(frag); 4454 struct xdp_buff *zc_frag = NULL; 4455 4456 if (mem_type == MEM_TYPE_XSK_BUFF_POOL) { 4457 netmem = 0; 4458 zc_frag = bpf_xdp_shrink_data_zc(xdp, shrink, tail, release); 4459 } 4460 4461 if (release) { 4462 __xdp_return(netmem, mem_type, false, zc_frag); 4463 } else { 4464 if (!tail) 4465 skb_frag_off_add(frag, shrink); 4466 skb_frag_size_sub(frag, shrink); 4467 } 4468 4469 return release; 4470 } 4471 4472 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset) 4473 { 4474 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4475 int i, n_frags_free = 0, len_free = 0; 4476 4477 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN)) 4478 return -EINVAL; 4479 4480 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) { 4481 skb_frag_t *frag = &sinfo->frags[i]; 4482 int shrink = min_t(int, offset, skb_frag_size(frag)); 4483 4484 len_free += shrink; 4485 offset -= shrink; 4486 if (bpf_xdp_shrink_data(xdp, frag, shrink, true)) 4487 n_frags_free++; 4488 } 4489 sinfo->nr_frags -= n_frags_free; 4490 sinfo->xdp_frags_size -= len_free; 4491 4492 if (unlikely(!sinfo->nr_frags)) { 4493 xdp_buff_clear_frags_flag(xdp); 4494 xdp_buff_clear_frag_pfmemalloc(xdp); 4495 xdp->data_end -= offset; 4496 } 4497 4498 return 0; 4499 } 4500 4501 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 4502 { 4503 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */ 4504 void *data_end = xdp->data_end + offset; 4505 4506 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */ 4507 if (offset < 0) 4508 return bpf_xdp_frags_shrink_tail(xdp, -offset); 4509 4510 return bpf_xdp_frags_increase_tail(xdp, offset); 4511 } 4512 4513 /* Notice that xdp_data_hard_end have reserved some tailroom */ 4514 if (unlikely(data_end > data_hard_end)) 4515 return -EINVAL; 4516 4517 if (unlikely(data_end < xdp->data + ETH_HLEN)) 4518 return -EINVAL; 4519 4520 /* Clear memory area on grow, can contain uninit kernel memory */ 4521 if (offset > 0) 4522 memset(xdp->data_end, 0, offset); 4523 4524 xdp->data_end = data_end; 4525 4526 return 0; 4527 } 4528 4529 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 4530 .func = bpf_xdp_adjust_tail, 4531 .gpl_only = false, 4532 .ret_type = RET_INTEGER, 4533 .arg1_type = ARG_PTR_TO_CTX, 4534 .arg2_type = ARG_ANYTHING, 4535 }; 4536 4537 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 4538 { 4539 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 4540 void *meta = xdp->data_meta + offset; 4541 unsigned long metalen = xdp->data - meta; 4542 4543 if (xdp_data_meta_unsupported(xdp)) 4544 return -ENOTSUPP; 4545 if (unlikely(meta < xdp_frame_end || 4546 meta > xdp->data)) 4547 return -EINVAL; 4548 if (unlikely(xdp_metalen_invalid(metalen))) 4549 return -EACCES; 4550 4551 xdp->data_meta = meta; 4552 4553 return 0; 4554 } 4555 4556 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 4557 .func = bpf_xdp_adjust_meta, 4558 .gpl_only = false, 4559 .ret_type = RET_INTEGER, 4560 .arg1_type = ARG_PTR_TO_CTX, 4561 .arg2_type = ARG_ANYTHING, 4562 }; 4563 4564 /** 4565 * DOC: xdp redirect 4566 * 4567 * XDP_REDIRECT works by a three-step process, implemented in the functions 4568 * below: 4569 * 4570 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target 4571 * of the redirect and store it (along with some other metadata) in a per-CPU 4572 * struct bpf_redirect_info. 4573 * 4574 * 2. When the program returns the XDP_REDIRECT return code, the driver will 4575 * call xdp_do_redirect() which will use the information in struct 4576 * bpf_redirect_info to actually enqueue the frame into a map type-specific 4577 * bulk queue structure. 4578 * 4579 * 3. Before exiting its NAPI poll loop, the driver will call 4580 * xdp_do_flush(), which will flush all the different bulk queues, 4581 * thus completing the redirect. Note that xdp_do_flush() must be 4582 * called before napi_complete_done() in the driver, as the 4583 * XDP_REDIRECT logic relies on being inside a single NAPI instance 4584 * through to the xdp_do_flush() call for RCU protection of all 4585 * in-kernel data structures. 4586 */ 4587 /* 4588 * Pointers to the map entries will be kept around for this whole sequence of 4589 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in 4590 * the core code; instead, the RCU protection relies on everything happening 4591 * inside a single NAPI poll sequence, which means it's between a pair of calls 4592 * to local_bh_disable()/local_bh_enable(). 4593 * 4594 * The map entries are marked as __rcu and the map code makes sure to 4595 * dereference those pointers with rcu_dereference_check() in a way that works 4596 * for both sections that to hold an rcu_read_lock() and sections that are 4597 * called from NAPI without a separate rcu_read_lock(). The code below does not 4598 * use RCU annotations, but relies on those in the map code. 4599 */ 4600 void xdp_do_flush(void) 4601 { 4602 struct list_head *lh_map, *lh_dev, *lh_xsk; 4603 4604 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk); 4605 if (lh_dev) 4606 __dev_flush(lh_dev); 4607 if (lh_map) 4608 __cpu_map_flush(lh_map); 4609 if (lh_xsk) 4610 __xsk_map_flush(lh_xsk); 4611 } 4612 EXPORT_SYMBOL_GPL(xdp_do_flush); 4613 4614 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL) 4615 void xdp_do_check_flushed(struct napi_struct *napi) 4616 { 4617 struct list_head *lh_map, *lh_dev, *lh_xsk; 4618 bool missed = false; 4619 4620 bpf_net_ctx_get_all_used_flush_lists(&lh_map, &lh_dev, &lh_xsk); 4621 if (lh_dev) { 4622 __dev_flush(lh_dev); 4623 missed = true; 4624 } 4625 if (lh_map) { 4626 __cpu_map_flush(lh_map); 4627 missed = true; 4628 } 4629 if (lh_xsk) { 4630 __xsk_map_flush(lh_xsk); 4631 missed = true; 4632 } 4633 4634 WARN_ONCE(missed, "Missing xdp_do_flush() invocation after NAPI by %ps\n", 4635 napi->poll); 4636 } 4637 #endif 4638 4639 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 4640 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key); 4641 4642 u32 xdp_master_redirect(struct xdp_buff *xdp) 4643 { 4644 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 4645 struct net_device *master, *slave; 4646 4647 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev); 4648 if (unlikely(!master || !(master->flags & IFF_UP))) 4649 return XDP_ABORTED; 4650 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp); 4651 if (slave && slave != xdp->rxq->dev) { 4652 /* The target device is different from the receiving device, so 4653 * redirect it to the new device. 4654 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled 4655 * drivers to unmap the packet from their rx ring. 4656 */ 4657 ri->tgt_index = slave->ifindex; 4658 ri->map_id = INT_MAX; 4659 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4660 return XDP_REDIRECT; 4661 } 4662 return XDP_TX; 4663 } 4664 EXPORT_SYMBOL_GPL(xdp_master_redirect); 4665 4666 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri, 4667 const struct net_device *dev, 4668 struct xdp_buff *xdp, 4669 const struct bpf_prog *xdp_prog) 4670 { 4671 enum bpf_map_type map_type = ri->map_type; 4672 void *fwd = ri->tgt_value; 4673 u32 map_id = ri->map_id; 4674 int err; 4675 4676 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4677 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4678 4679 err = __xsk_map_redirect(fwd, xdp); 4680 if (unlikely(err)) 4681 goto err; 4682 4683 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4684 return 0; 4685 err: 4686 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4687 return err; 4688 } 4689 4690 static __always_inline int 4691 __xdp_do_redirect_frame(struct bpf_redirect_info *ri, struct net_device *dev, 4692 struct xdp_frame *xdpf, 4693 const struct bpf_prog *xdp_prog) 4694 { 4695 enum bpf_map_type map_type = ri->map_type; 4696 void *fwd = ri->tgt_value; 4697 u32 map_id = ri->map_id; 4698 u32 flags = ri->flags; 4699 struct bpf_map *map; 4700 int err; 4701 4702 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4703 ri->flags = 0; 4704 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4705 4706 if (unlikely(!xdpf)) { 4707 err = -EOVERFLOW; 4708 goto err; 4709 } 4710 4711 switch (map_type) { 4712 case BPF_MAP_TYPE_DEVMAP: 4713 fallthrough; 4714 case BPF_MAP_TYPE_DEVMAP_HASH: 4715 if (unlikely(flags & BPF_F_BROADCAST)) { 4716 map = READ_ONCE(ri->map); 4717 4718 /* The map pointer is cleared when the map is being torn 4719 * down by dev_map_free() 4720 */ 4721 if (unlikely(!map)) { 4722 err = -ENOENT; 4723 break; 4724 } 4725 4726 WRITE_ONCE(ri->map, NULL); 4727 err = dev_map_enqueue_multi(xdpf, dev, map, 4728 flags & BPF_F_EXCLUDE_INGRESS); 4729 } else { 4730 err = dev_map_enqueue(fwd, xdpf, dev); 4731 } 4732 break; 4733 case BPF_MAP_TYPE_CPUMAP: 4734 err = cpu_map_enqueue(fwd, xdpf, dev); 4735 break; 4736 case BPF_MAP_TYPE_UNSPEC: 4737 if (map_id == INT_MAX) { 4738 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4739 if (unlikely(!fwd)) { 4740 err = -EINVAL; 4741 break; 4742 } 4743 err = dev_xdp_enqueue(fwd, xdpf, dev); 4744 break; 4745 } 4746 fallthrough; 4747 default: 4748 err = -EBADRQC; 4749 } 4750 4751 if (unlikely(err)) 4752 goto err; 4753 4754 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4755 return 0; 4756 err: 4757 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4758 return err; 4759 } 4760 4761 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 4762 const struct bpf_prog *xdp_prog) 4763 { 4764 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 4765 enum bpf_map_type map_type = ri->map_type; 4766 4767 if (map_type == BPF_MAP_TYPE_XSKMAP) 4768 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4769 4770 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp), 4771 xdp_prog); 4772 } 4773 EXPORT_SYMBOL_GPL(xdp_do_redirect); 4774 4775 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp, 4776 struct xdp_frame *xdpf, 4777 const struct bpf_prog *xdp_prog) 4778 { 4779 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 4780 enum bpf_map_type map_type = ri->map_type; 4781 4782 if (map_type == BPF_MAP_TYPE_XSKMAP) 4783 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4784 4785 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog); 4786 } 4787 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame); 4788 4789 static int xdp_do_generic_redirect_map(struct net_device *dev, 4790 struct sk_buff *skb, 4791 struct xdp_buff *xdp, 4792 const struct bpf_prog *xdp_prog, 4793 void *fwd, enum bpf_map_type map_type, 4794 u32 map_id, u32 flags) 4795 { 4796 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 4797 struct bpf_map *map; 4798 int err; 4799 4800 switch (map_type) { 4801 case BPF_MAP_TYPE_DEVMAP: 4802 fallthrough; 4803 case BPF_MAP_TYPE_DEVMAP_HASH: 4804 if (unlikely(flags & BPF_F_BROADCAST)) { 4805 map = READ_ONCE(ri->map); 4806 4807 /* The map pointer is cleared when the map is being torn 4808 * down by dev_map_free() 4809 */ 4810 if (unlikely(!map)) { 4811 err = -ENOENT; 4812 break; 4813 } 4814 4815 WRITE_ONCE(ri->map, NULL); 4816 err = dev_map_redirect_multi(dev, skb, xdp_prog, map, 4817 flags & BPF_F_EXCLUDE_INGRESS); 4818 } else { 4819 err = dev_map_generic_redirect(fwd, skb, xdp_prog); 4820 } 4821 if (unlikely(err)) 4822 goto err; 4823 break; 4824 case BPF_MAP_TYPE_XSKMAP: 4825 err = xsk_generic_rcv(fwd, xdp); 4826 if (err) 4827 goto err; 4828 consume_skb(skb); 4829 break; 4830 case BPF_MAP_TYPE_CPUMAP: 4831 err = cpu_map_generic_redirect(fwd, skb); 4832 if (unlikely(err)) 4833 goto err; 4834 break; 4835 default: 4836 err = -EBADRQC; 4837 goto err; 4838 } 4839 4840 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4841 return 0; 4842 err: 4843 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4844 return err; 4845 } 4846 4847 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 4848 struct xdp_buff *xdp, 4849 const struct bpf_prog *xdp_prog) 4850 { 4851 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 4852 enum bpf_map_type map_type = ri->map_type; 4853 void *fwd = ri->tgt_value; 4854 u32 map_id = ri->map_id; 4855 u32 flags = ri->flags; 4856 int err; 4857 4858 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4859 ri->flags = 0; 4860 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4861 4862 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) { 4863 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4864 if (unlikely(!fwd)) { 4865 err = -EINVAL; 4866 goto err; 4867 } 4868 4869 err = xdp_ok_fwd_dev(fwd, skb->len); 4870 if (unlikely(err)) 4871 goto err; 4872 4873 skb->dev = fwd; 4874 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index); 4875 generic_xdp_tx(skb, xdp_prog); 4876 return 0; 4877 } 4878 4879 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id, flags); 4880 err: 4881 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err); 4882 return err; 4883 } 4884 4885 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 4886 { 4887 struct bpf_redirect_info *ri = bpf_net_ctx_get_ri(); 4888 4889 if (unlikely(flags)) 4890 return XDP_ABORTED; 4891 4892 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated 4893 * by map_idr) is used for ifindex based XDP redirect. 4894 */ 4895 ri->tgt_index = ifindex; 4896 ri->map_id = INT_MAX; 4897 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4898 4899 return XDP_REDIRECT; 4900 } 4901 4902 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 4903 .func = bpf_xdp_redirect, 4904 .gpl_only = false, 4905 .ret_type = RET_INTEGER, 4906 .arg1_type = ARG_ANYTHING, 4907 .arg2_type = ARG_ANYTHING, 4908 }; 4909 4910 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key, 4911 u64, flags) 4912 { 4913 return map->ops->map_redirect(map, key, flags); 4914 } 4915 4916 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 4917 .func = bpf_xdp_redirect_map, 4918 .gpl_only = false, 4919 .ret_type = RET_INTEGER, 4920 .arg1_type = ARG_CONST_MAP_PTR, 4921 .arg2_type = ARG_ANYTHING, 4922 .arg3_type = ARG_ANYTHING, 4923 }; 4924 4925 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 4926 unsigned long off, unsigned long len) 4927 { 4928 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 4929 4930 if (unlikely(!ptr)) 4931 return len; 4932 if (ptr != dst_buff) 4933 memcpy(dst_buff, ptr, len); 4934 4935 return 0; 4936 } 4937 4938 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 4939 u64, flags, void *, meta, u64, meta_size) 4940 { 4941 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4942 4943 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4944 return -EINVAL; 4945 if (unlikely(!skb || skb_size > skb->len)) 4946 return -EFAULT; 4947 4948 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 4949 bpf_skb_copy); 4950 } 4951 4952 static const struct bpf_func_proto bpf_skb_event_output_proto = { 4953 .func = bpf_skb_event_output, 4954 .gpl_only = true, 4955 .ret_type = RET_INTEGER, 4956 .arg1_type = ARG_PTR_TO_CTX, 4957 .arg2_type = ARG_CONST_MAP_PTR, 4958 .arg3_type = ARG_ANYTHING, 4959 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4960 .arg5_type = ARG_MEM_SIZE_OR_ZERO, 4961 }; 4962 4963 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff) 4964 4965 const struct bpf_func_proto bpf_skb_output_proto = { 4966 .func = bpf_skb_event_output, 4967 .gpl_only = true, 4968 .ret_type = RET_INTEGER, 4969 .arg1_type = ARG_PTR_TO_BTF_ID, 4970 .arg1_btf_id = &bpf_skb_output_btf_ids[0], 4971 .arg2_type = ARG_CONST_MAP_PTR, 4972 .arg3_type = ARG_ANYTHING, 4973 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4974 .arg5_type = ARG_MEM_SIZE_OR_ZERO, 4975 }; 4976 4977 static unsigned short bpf_tunnel_key_af(u64 flags) 4978 { 4979 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 4980 } 4981 4982 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 4983 u32, size, u64, flags) 4984 { 4985 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4986 u8 compat[sizeof(struct bpf_tunnel_key)]; 4987 void *to_orig = to; 4988 int err; 4989 4990 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 | 4991 BPF_F_TUNINFO_FLAGS)))) { 4992 err = -EINVAL; 4993 goto err_clear; 4994 } 4995 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4996 err = -EPROTO; 4997 goto err_clear; 4998 } 4999 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 5000 err = -EINVAL; 5001 switch (size) { 5002 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 5003 case offsetof(struct bpf_tunnel_key, tunnel_label): 5004 case offsetof(struct bpf_tunnel_key, tunnel_ext): 5005 goto set_compat; 5006 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 5007 /* Fixup deprecated structure layouts here, so we have 5008 * a common path later on. 5009 */ 5010 if (ip_tunnel_info_af(info) != AF_INET) 5011 goto err_clear; 5012 set_compat: 5013 to = (struct bpf_tunnel_key *)compat; 5014 break; 5015 default: 5016 goto err_clear; 5017 } 5018 } 5019 5020 to->tunnel_id = be64_to_cpu(info->key.tun_id); 5021 to->tunnel_tos = info->key.tos; 5022 to->tunnel_ttl = info->key.ttl; 5023 if (flags & BPF_F_TUNINFO_FLAGS) 5024 to->tunnel_flags = ip_tunnel_flags_to_be16(info->key.tun_flags); 5025 else 5026 to->tunnel_ext = 0; 5027 5028 if (flags & BPF_F_TUNINFO_IPV6) { 5029 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 5030 sizeof(to->remote_ipv6)); 5031 memcpy(to->local_ipv6, &info->key.u.ipv6.dst, 5032 sizeof(to->local_ipv6)); 5033 to->tunnel_label = be32_to_cpu(info->key.label); 5034 } else { 5035 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 5036 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5037 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst); 5038 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3); 5039 to->tunnel_label = 0; 5040 } 5041 5042 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 5043 memcpy(to_orig, to, size); 5044 5045 return 0; 5046 err_clear: 5047 memset(to_orig, 0, size); 5048 return err; 5049 } 5050 5051 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 5052 .func = bpf_skb_get_tunnel_key, 5053 .gpl_only = false, 5054 .ret_type = RET_INTEGER, 5055 .arg1_type = ARG_PTR_TO_CTX, 5056 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 5057 .arg3_type = ARG_MEM_SIZE, 5058 .arg4_type = ARG_ANYTHING, 5059 }; 5060 5061 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 5062 { 5063 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 5064 int err; 5065 5066 if (unlikely(!info || 5067 !ip_tunnel_is_options_present(info->key.tun_flags))) { 5068 err = -ENOENT; 5069 goto err_clear; 5070 } 5071 if (unlikely(size < info->options_len)) { 5072 err = -ENOMEM; 5073 goto err_clear; 5074 } 5075 5076 ip_tunnel_info_opts_get(to, info); 5077 if (size > info->options_len) 5078 memset(to + info->options_len, 0, size - info->options_len); 5079 5080 return info->options_len; 5081 err_clear: 5082 memset(to, 0, size); 5083 return err; 5084 } 5085 5086 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 5087 .func = bpf_skb_get_tunnel_opt, 5088 .gpl_only = false, 5089 .ret_type = RET_INTEGER, 5090 .arg1_type = ARG_PTR_TO_CTX, 5091 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 5092 .arg3_type = ARG_MEM_SIZE, 5093 }; 5094 5095 static struct metadata_dst __percpu *md_dst; 5096 5097 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 5098 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 5099 { 5100 struct metadata_dst *md = this_cpu_ptr(md_dst); 5101 u8 compat[sizeof(struct bpf_tunnel_key)]; 5102 struct ip_tunnel_info *info; 5103 5104 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 5105 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER | 5106 BPF_F_NO_TUNNEL_KEY))) 5107 return -EINVAL; 5108 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 5109 switch (size) { 5110 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 5111 case offsetof(struct bpf_tunnel_key, tunnel_label): 5112 case offsetof(struct bpf_tunnel_key, tunnel_ext): 5113 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 5114 /* Fixup deprecated structure layouts here, so we have 5115 * a common path later on. 5116 */ 5117 memcpy(compat, from, size); 5118 memset(compat + size, 0, sizeof(compat) - size); 5119 from = (const struct bpf_tunnel_key *) compat; 5120 break; 5121 default: 5122 return -EINVAL; 5123 } 5124 } 5125 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 5126 from->tunnel_ext)) 5127 return -EINVAL; 5128 5129 skb_dst_drop(skb); 5130 dst_hold((struct dst_entry *) md); 5131 skb_dst_set(skb, (struct dst_entry *) md); 5132 5133 info = &md->u.tun_info; 5134 memset(info, 0, sizeof(*info)); 5135 info->mode = IP_TUNNEL_INFO_TX; 5136 5137 __set_bit(IP_TUNNEL_NOCACHE_BIT, info->key.tun_flags); 5138 __assign_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, info->key.tun_flags, 5139 flags & BPF_F_DONT_FRAGMENT); 5140 __assign_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags, 5141 !(flags & BPF_F_ZERO_CSUM_TX)); 5142 __assign_bit(IP_TUNNEL_SEQ_BIT, info->key.tun_flags, 5143 flags & BPF_F_SEQ_NUMBER); 5144 __assign_bit(IP_TUNNEL_KEY_BIT, info->key.tun_flags, 5145 !(flags & BPF_F_NO_TUNNEL_KEY)); 5146 5147 info->key.tun_id = cpu_to_be64(from->tunnel_id); 5148 info->key.tos = from->tunnel_tos; 5149 info->key.ttl = from->tunnel_ttl; 5150 5151 if (flags & BPF_F_TUNINFO_IPV6) { 5152 info->mode |= IP_TUNNEL_INFO_IPV6; 5153 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 5154 sizeof(from->remote_ipv6)); 5155 memcpy(&info->key.u.ipv6.src, from->local_ipv6, 5156 sizeof(from->local_ipv6)); 5157 info->key.label = cpu_to_be32(from->tunnel_label) & 5158 IPV6_FLOWLABEL_MASK; 5159 } else { 5160 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 5161 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4); 5162 info->key.flow_flags = FLOWI_FLAG_ANYSRC; 5163 } 5164 5165 return 0; 5166 } 5167 5168 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 5169 .func = bpf_skb_set_tunnel_key, 5170 .gpl_only = false, 5171 .ret_type = RET_INTEGER, 5172 .arg1_type = ARG_PTR_TO_CTX, 5173 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5174 .arg3_type = ARG_MEM_SIZE, 5175 .arg4_type = ARG_ANYTHING, 5176 }; 5177 5178 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 5179 const u8 *, from, u32, size) 5180 { 5181 struct ip_tunnel_info *info = skb_tunnel_info(skb); 5182 const struct metadata_dst *md = this_cpu_ptr(md_dst); 5183 IP_TUNNEL_DECLARE_FLAGS(present) = { }; 5184 5185 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 5186 return -EINVAL; 5187 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 5188 return -ENOMEM; 5189 5190 ip_tunnel_set_options_present(present); 5191 ip_tunnel_info_opts_set(info, from, size, present); 5192 5193 return 0; 5194 } 5195 5196 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 5197 .func = bpf_skb_set_tunnel_opt, 5198 .gpl_only = false, 5199 .ret_type = RET_INTEGER, 5200 .arg1_type = ARG_PTR_TO_CTX, 5201 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5202 .arg3_type = ARG_MEM_SIZE, 5203 }; 5204 5205 static const struct bpf_func_proto * 5206 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 5207 { 5208 if (!md_dst) { 5209 struct metadata_dst __percpu *tmp; 5210 5211 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 5212 METADATA_IP_TUNNEL, 5213 GFP_KERNEL); 5214 if (!tmp) 5215 return NULL; 5216 if (cmpxchg(&md_dst, NULL, tmp)) 5217 metadata_dst_free_percpu(tmp); 5218 } 5219 5220 switch (which) { 5221 case BPF_FUNC_skb_set_tunnel_key: 5222 return &bpf_skb_set_tunnel_key_proto; 5223 case BPF_FUNC_skb_set_tunnel_opt: 5224 return &bpf_skb_set_tunnel_opt_proto; 5225 default: 5226 return NULL; 5227 } 5228 } 5229 5230 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 5231 u32, idx) 5232 { 5233 struct bpf_array *array = container_of(map, struct bpf_array, map); 5234 struct cgroup *cgrp; 5235 struct sock *sk; 5236 5237 sk = skb_to_full_sk(skb); 5238 if (!sk || !sk_fullsock(sk)) 5239 return -ENOENT; 5240 if (unlikely(idx >= array->map.max_entries)) 5241 return -E2BIG; 5242 5243 cgrp = READ_ONCE(array->ptrs[idx]); 5244 if (unlikely(!cgrp)) 5245 return -EAGAIN; 5246 5247 return sk_under_cgroup_hierarchy(sk, cgrp); 5248 } 5249 5250 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 5251 .func = bpf_skb_under_cgroup, 5252 .gpl_only = false, 5253 .ret_type = RET_INTEGER, 5254 .arg1_type = ARG_PTR_TO_CTX, 5255 .arg2_type = ARG_CONST_MAP_PTR, 5256 .arg3_type = ARG_ANYTHING, 5257 }; 5258 5259 #ifdef CONFIG_SOCK_CGROUP_DATA 5260 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 5261 { 5262 struct cgroup *cgrp; 5263 5264 sk = sk_to_full_sk(sk); 5265 if (!sk || !sk_fullsock(sk)) 5266 return 0; 5267 5268 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 5269 return cgroup_id(cgrp); 5270 } 5271 5272 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 5273 { 5274 return __bpf_sk_cgroup_id(skb->sk); 5275 } 5276 5277 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 5278 .func = bpf_skb_cgroup_id, 5279 .gpl_only = false, 5280 .ret_type = RET_INTEGER, 5281 .arg1_type = ARG_PTR_TO_CTX, 5282 }; 5283 5284 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 5285 int ancestor_level) 5286 { 5287 struct cgroup *ancestor; 5288 struct cgroup *cgrp; 5289 5290 sk = sk_to_full_sk(sk); 5291 if (!sk || !sk_fullsock(sk)) 5292 return 0; 5293 5294 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 5295 ancestor = cgroup_ancestor(cgrp, ancestor_level); 5296 if (!ancestor) 5297 return 0; 5298 5299 return cgroup_id(ancestor); 5300 } 5301 5302 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 5303 ancestor_level) 5304 { 5305 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 5306 } 5307 5308 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 5309 .func = bpf_skb_ancestor_cgroup_id, 5310 .gpl_only = false, 5311 .ret_type = RET_INTEGER, 5312 .arg1_type = ARG_PTR_TO_CTX, 5313 .arg2_type = ARG_ANYTHING, 5314 }; 5315 5316 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 5317 { 5318 return __bpf_sk_cgroup_id(sk); 5319 } 5320 5321 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 5322 .func = bpf_sk_cgroup_id, 5323 .gpl_only = false, 5324 .ret_type = RET_INTEGER, 5325 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5326 }; 5327 5328 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 5329 { 5330 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 5331 } 5332 5333 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 5334 .func = bpf_sk_ancestor_cgroup_id, 5335 .gpl_only = false, 5336 .ret_type = RET_INTEGER, 5337 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5338 .arg2_type = ARG_ANYTHING, 5339 }; 5340 #endif 5341 5342 static unsigned long bpf_xdp_copy(void *dst, const void *ctx, 5343 unsigned long off, unsigned long len) 5344 { 5345 struct xdp_buff *xdp = (struct xdp_buff *)ctx; 5346 5347 bpf_xdp_copy_buf(xdp, off, dst, len, false); 5348 return 0; 5349 } 5350 5351 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 5352 u64, flags, void *, meta, u64, meta_size) 5353 { 5354 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 5355 5356 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 5357 return -EINVAL; 5358 5359 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp))) 5360 return -EFAULT; 5361 5362 return bpf_event_output(map, flags, meta, meta_size, xdp, 5363 xdp_size, bpf_xdp_copy); 5364 } 5365 5366 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 5367 .func = bpf_xdp_event_output, 5368 .gpl_only = true, 5369 .ret_type = RET_INTEGER, 5370 .arg1_type = ARG_PTR_TO_CTX, 5371 .arg2_type = ARG_CONST_MAP_PTR, 5372 .arg3_type = ARG_ANYTHING, 5373 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5374 .arg5_type = ARG_MEM_SIZE_OR_ZERO, 5375 }; 5376 5377 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 5378 5379 const struct bpf_func_proto bpf_xdp_output_proto = { 5380 .func = bpf_xdp_event_output, 5381 .gpl_only = true, 5382 .ret_type = RET_INTEGER, 5383 .arg1_type = ARG_PTR_TO_BTF_ID, 5384 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 5385 .arg2_type = ARG_CONST_MAP_PTR, 5386 .arg3_type = ARG_ANYTHING, 5387 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5388 .arg5_type = ARG_MEM_SIZE_OR_ZERO, 5389 }; 5390 5391 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 5392 { 5393 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 5394 } 5395 5396 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 5397 .func = bpf_get_socket_cookie, 5398 .gpl_only = false, 5399 .ret_type = RET_INTEGER, 5400 .arg1_type = ARG_PTR_TO_CTX, 5401 }; 5402 5403 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 5404 { 5405 return __sock_gen_cookie(ctx->sk); 5406 } 5407 5408 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 5409 .func = bpf_get_socket_cookie_sock_addr, 5410 .gpl_only = false, 5411 .ret_type = RET_INTEGER, 5412 .arg1_type = ARG_PTR_TO_CTX, 5413 }; 5414 5415 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 5416 { 5417 return __sock_gen_cookie(ctx); 5418 } 5419 5420 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 5421 .func = bpf_get_socket_cookie_sock, 5422 .gpl_only = false, 5423 .ret_type = RET_INTEGER, 5424 .arg1_type = ARG_PTR_TO_CTX, 5425 }; 5426 5427 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk) 5428 { 5429 return sk ? sock_gen_cookie(sk) : 0; 5430 } 5431 5432 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = { 5433 .func = bpf_get_socket_ptr_cookie, 5434 .gpl_only = false, 5435 .ret_type = RET_INTEGER, 5436 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL, 5437 }; 5438 5439 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 5440 { 5441 return __sock_gen_cookie(ctx->sk); 5442 } 5443 5444 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 5445 .func = bpf_get_socket_cookie_sock_ops, 5446 .gpl_only = false, 5447 .ret_type = RET_INTEGER, 5448 .arg1_type = ARG_PTR_TO_CTX, 5449 }; 5450 5451 static u64 __bpf_get_netns_cookie(struct sock *sk) 5452 { 5453 const struct net *net = sk ? sock_net(sk) : &init_net; 5454 5455 return net->net_cookie; 5456 } 5457 5458 BPF_CALL_1(bpf_get_netns_cookie, struct sk_buff *, skb) 5459 { 5460 return __bpf_get_netns_cookie(skb && skb->sk ? skb->sk : NULL); 5461 } 5462 5463 static const struct bpf_func_proto bpf_get_netns_cookie_proto = { 5464 .func = bpf_get_netns_cookie, 5465 .ret_type = RET_INTEGER, 5466 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5467 }; 5468 5469 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 5470 { 5471 return __bpf_get_netns_cookie(ctx); 5472 } 5473 5474 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 5475 .func = bpf_get_netns_cookie_sock, 5476 .gpl_only = false, 5477 .ret_type = RET_INTEGER, 5478 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5479 }; 5480 5481 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 5482 { 5483 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5484 } 5485 5486 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 5487 .func = bpf_get_netns_cookie_sock_addr, 5488 .gpl_only = false, 5489 .ret_type = RET_INTEGER, 5490 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5491 }; 5492 5493 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 5494 { 5495 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5496 } 5497 5498 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = { 5499 .func = bpf_get_netns_cookie_sock_ops, 5500 .gpl_only = false, 5501 .ret_type = RET_INTEGER, 5502 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5503 }; 5504 5505 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx) 5506 { 5507 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5508 } 5509 5510 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = { 5511 .func = bpf_get_netns_cookie_sk_msg, 5512 .gpl_only = false, 5513 .ret_type = RET_INTEGER, 5514 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5515 }; 5516 5517 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 5518 { 5519 struct sock *sk = sk_to_full_sk(skb->sk); 5520 kuid_t kuid; 5521 5522 if (!sk || !sk_fullsock(sk)) 5523 return overflowuid; 5524 kuid = sock_net_uid(sock_net(sk), sk); 5525 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 5526 } 5527 5528 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 5529 .func = bpf_get_socket_uid, 5530 .gpl_only = false, 5531 .ret_type = RET_INTEGER, 5532 .arg1_type = ARG_PTR_TO_CTX, 5533 }; 5534 5535 static int sk_bpf_set_get_cb_flags(struct sock *sk, char *optval, bool getopt) 5536 { 5537 u32 sk_bpf_cb_flags; 5538 5539 if (getopt) { 5540 *(u32 *)optval = sk->sk_bpf_cb_flags; 5541 return 0; 5542 } 5543 5544 sk_bpf_cb_flags = *(u32 *)optval; 5545 5546 if (sk_bpf_cb_flags & ~SK_BPF_CB_MASK) 5547 return -EINVAL; 5548 5549 sk->sk_bpf_cb_flags = sk_bpf_cb_flags; 5550 5551 return 0; 5552 } 5553 5554 static int sol_socket_sockopt(struct sock *sk, int optname, 5555 char *optval, int *optlen, 5556 bool getopt) 5557 { 5558 switch (optname) { 5559 case SO_REUSEADDR: 5560 case SO_SNDBUF: 5561 case SO_RCVBUF: 5562 case SO_KEEPALIVE: 5563 case SO_PRIORITY: 5564 case SO_REUSEPORT: 5565 case SO_RCVLOWAT: 5566 case SO_MARK: 5567 case SO_MAX_PACING_RATE: 5568 case SO_BINDTOIFINDEX: 5569 case SO_TXREHASH: 5570 case SK_BPF_CB_FLAGS: 5571 if (*optlen != sizeof(int)) 5572 return -EINVAL; 5573 break; 5574 case SO_BINDTODEVICE: 5575 break; 5576 default: 5577 return -EINVAL; 5578 } 5579 5580 if (optname == SK_BPF_CB_FLAGS) 5581 return sk_bpf_set_get_cb_flags(sk, optval, getopt); 5582 5583 if (getopt) { 5584 if (optname == SO_BINDTODEVICE) 5585 return -EINVAL; 5586 return sk_getsockopt(sk, SOL_SOCKET, optname, 5587 KERNEL_SOCKPTR(optval), 5588 KERNEL_SOCKPTR(optlen)); 5589 } 5590 5591 return sk_setsockopt(sk, SOL_SOCKET, optname, 5592 KERNEL_SOCKPTR(optval), *optlen); 5593 } 5594 5595 static int bpf_sol_tcp_getsockopt(struct sock *sk, int optname, 5596 char *optval, int optlen) 5597 { 5598 if (optlen != sizeof(int)) 5599 return -EINVAL; 5600 5601 switch (optname) { 5602 case TCP_BPF_SOCK_OPS_CB_FLAGS: { 5603 int cb_flags = tcp_sk(sk)->bpf_sock_ops_cb_flags; 5604 5605 memcpy(optval, &cb_flags, optlen); 5606 break; 5607 } 5608 case TCP_BPF_RTO_MIN: { 5609 int rto_min_us = jiffies_to_usecs(inet_csk(sk)->icsk_rto_min); 5610 5611 memcpy(optval, &rto_min_us, optlen); 5612 break; 5613 } 5614 case TCP_BPF_DELACK_MAX: { 5615 int delack_max_us = jiffies_to_usecs(inet_csk(sk)->icsk_delack_max); 5616 5617 memcpy(optval, &delack_max_us, optlen); 5618 break; 5619 } 5620 default: 5621 return -EINVAL; 5622 } 5623 5624 return 0; 5625 } 5626 5627 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname, 5628 char *optval, int optlen) 5629 { 5630 struct tcp_sock *tp = tcp_sk(sk); 5631 unsigned long timeout; 5632 int val; 5633 5634 if (optlen != sizeof(int)) 5635 return -EINVAL; 5636 5637 val = *(int *)optval; 5638 5639 /* Only some options are supported */ 5640 switch (optname) { 5641 case TCP_BPF_IW: 5642 if (val <= 0 || tp->data_segs_out > tp->syn_data) 5643 return -EINVAL; 5644 tcp_snd_cwnd_set(tp, min_t(u32, val, tp->snd_cwnd_clamp)); 5645 break; 5646 case TCP_BPF_SNDCWND_CLAMP: 5647 if (val <= 0) 5648 return -EINVAL; 5649 val = min_t(u32, val, TCP_MAX_SND_CWND); 5650 tp->snd_cwnd_clamp = val; 5651 WRITE_ONCE(tp->snd_ssthresh, val); 5652 break; 5653 case TCP_BPF_DELACK_MAX: 5654 timeout = usecs_to_jiffies(val); 5655 if (timeout > TCP_DELACK_MAX || 5656 timeout < TCP_TIMEOUT_MIN) 5657 return -EINVAL; 5658 inet_csk(sk)->icsk_delack_max = timeout; 5659 break; 5660 case TCP_BPF_RTO_MIN: 5661 timeout = usecs_to_jiffies(val); 5662 if (timeout > TCP_RTO_MIN || 5663 timeout < TCP_TIMEOUT_MIN) 5664 return -EINVAL; 5665 inet_csk(sk)->icsk_rto_min = timeout; 5666 break; 5667 case TCP_BPF_SOCK_OPS_CB_FLAGS: 5668 if (val & ~(BPF_SOCK_OPS_ALL_CB_FLAGS)) 5669 return -EINVAL; 5670 tp->bpf_sock_ops_cb_flags = val; 5671 break; 5672 default: 5673 return -EINVAL; 5674 } 5675 5676 return 0; 5677 } 5678 5679 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval, 5680 int *optlen, bool getopt) 5681 { 5682 struct tcp_sock *tp; 5683 int ret; 5684 5685 if (*optlen < 2) 5686 return -EINVAL; 5687 5688 if (getopt) { 5689 if (!inet_csk(sk)->icsk_ca_ops) 5690 return -EINVAL; 5691 /* BPF expects NULL-terminated tcp-cc string */ 5692 optval[--(*optlen)] = '\0'; 5693 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION, 5694 KERNEL_SOCKPTR(optval), 5695 KERNEL_SOCKPTR(optlen)); 5696 } 5697 5698 /* "cdg" is the only cc that alloc a ptr 5699 * in inet_csk_ca area. The bpf-tcp-cc may 5700 * overwrite this ptr after switching to cdg. 5701 */ 5702 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen)) 5703 return -ENOTSUPP; 5704 5705 /* It stops this looping 5706 * 5707 * .init => bpf_setsockopt(tcp_cc) => .init => 5708 * bpf_setsockopt(tcp_cc)" => .init => .... 5709 * 5710 * The second bpf_setsockopt(tcp_cc) is not allowed 5711 * in order to break the loop when both .init 5712 * are the same bpf prog. 5713 * 5714 * This applies even the second bpf_setsockopt(tcp_cc) 5715 * does not cause a loop. This limits only the first 5716 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to 5717 * pick a fallback cc (eg. peer does not support ECN) 5718 * and the second '.init' cannot fallback to 5719 * another. 5720 */ 5721 tp = tcp_sk(sk); 5722 if (tp->bpf_chg_cc_inprogress) 5723 return -EBUSY; 5724 5725 tp->bpf_chg_cc_inprogress = 1; 5726 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION, 5727 KERNEL_SOCKPTR(optval), *optlen); 5728 tp->bpf_chg_cc_inprogress = 0; 5729 return ret; 5730 } 5731 5732 static int sol_tcp_sockopt(struct sock *sk, int optname, 5733 char *optval, int *optlen, 5734 bool getopt) 5735 { 5736 if (!sk_is_tcp(sk)) 5737 return -EINVAL; 5738 5739 switch (optname) { 5740 case TCP_NODELAY: 5741 case TCP_MAXSEG: 5742 case TCP_KEEPIDLE: 5743 case TCP_KEEPINTVL: 5744 case TCP_KEEPCNT: 5745 case TCP_SYNCNT: 5746 case TCP_WINDOW_CLAMP: 5747 case TCP_THIN_LINEAR_TIMEOUTS: 5748 case TCP_USER_TIMEOUT: 5749 case TCP_NOTSENT_LOWAT: 5750 case TCP_SAVE_SYN: 5751 case TCP_RTO_MAX_MS: 5752 if (*optlen != sizeof(int)) 5753 return -EINVAL; 5754 break; 5755 case TCP_CONGESTION: 5756 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt); 5757 case TCP_SAVED_SYN: 5758 if (*optlen < 1) 5759 return -EINVAL; 5760 break; 5761 default: 5762 if (getopt) 5763 return bpf_sol_tcp_getsockopt(sk, optname, optval, *optlen); 5764 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen); 5765 } 5766 5767 if (getopt) { 5768 if (optname == TCP_SAVED_SYN) { 5769 struct tcp_sock *tp = tcp_sk(sk); 5770 5771 if (!tp->saved_syn || 5772 *optlen > tcp_saved_syn_len(tp->saved_syn)) 5773 return -EINVAL; 5774 memcpy(optval, tp->saved_syn->data, *optlen); 5775 /* It cannot free tp->saved_syn here because it 5776 * does not know if the user space still needs it. 5777 */ 5778 return 0; 5779 } 5780 5781 return do_tcp_getsockopt(sk, SOL_TCP, optname, 5782 KERNEL_SOCKPTR(optval), 5783 KERNEL_SOCKPTR(optlen)); 5784 } 5785 5786 return do_tcp_setsockopt(sk, SOL_TCP, optname, 5787 KERNEL_SOCKPTR(optval), *optlen); 5788 } 5789 5790 static bool sk_allows_sol_ip_sockopt(struct sock *sk) 5791 { 5792 switch (sk->sk_family) { 5793 case AF_INET: 5794 return true; 5795 case AF_INET6: 5796 /* Allow getting/setting sockopt for possible ipv4-mapped ipv6 socket. */ 5797 return sk->sk_type != SOCK_RAW && !ipv6_only_sock(sk); 5798 default: 5799 return false; 5800 } 5801 } 5802 5803 static int sol_ip_sockopt(struct sock *sk, int optname, 5804 char *optval, int *optlen, 5805 bool getopt) 5806 { 5807 if (!sk_allows_sol_ip_sockopt(sk)) 5808 return -EINVAL; 5809 5810 switch (optname) { 5811 case IP_TOS: 5812 case IP_TRANSPARENT: 5813 if (*optlen != sizeof(int)) 5814 return -EINVAL; 5815 break; 5816 default: 5817 return -EINVAL; 5818 } 5819 5820 if (getopt) 5821 return do_ip_getsockopt(sk, SOL_IP, optname, 5822 KERNEL_SOCKPTR(optval), 5823 KERNEL_SOCKPTR(optlen)); 5824 5825 return do_ip_setsockopt(sk, SOL_IP, optname, 5826 KERNEL_SOCKPTR(optval), *optlen); 5827 } 5828 5829 static int sol_ipv6_sockopt(struct sock *sk, int optname, 5830 char *optval, int *optlen, 5831 bool getopt) 5832 { 5833 if (sk->sk_family != AF_INET6) 5834 return -EINVAL; 5835 5836 switch (optname) { 5837 case IPV6_TCLASS: 5838 case IPV6_AUTOFLOWLABEL: 5839 case IPV6_TRANSPARENT: 5840 if (*optlen != sizeof(int)) 5841 return -EINVAL; 5842 break; 5843 default: 5844 return -EINVAL; 5845 } 5846 5847 if (getopt) 5848 return do_ipv6_getsockopt(sk, SOL_IPV6, optname, 5849 KERNEL_SOCKPTR(optval), 5850 KERNEL_SOCKPTR(optlen)); 5851 5852 return do_ipv6_setsockopt(sk, SOL_IPV6, optname, 5853 KERNEL_SOCKPTR(optval), *optlen); 5854 } 5855 5856 static int __bpf_setsockopt(struct sock *sk, int level, int optname, 5857 char *optval, int optlen) 5858 { 5859 if (!sk_fullsock(sk)) 5860 return -EINVAL; 5861 5862 if (level == SOL_SOCKET) 5863 return sol_socket_sockopt(sk, optname, optval, &optlen, false); 5864 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5865 return sol_ip_sockopt(sk, optname, optval, &optlen, false); 5866 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5867 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false); 5868 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5869 return sol_tcp_sockopt(sk, optname, optval, &optlen, false); 5870 5871 return -EINVAL; 5872 } 5873 5874 static bool is_locked_tcp_sock_ops(struct bpf_sock_ops_kern *bpf_sock) 5875 { 5876 return bpf_sock->op <= BPF_SOCK_OPS_WRITE_HDR_OPT_CB; 5877 } 5878 5879 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 5880 char *optval, int optlen) 5881 { 5882 if (sk_fullsock(sk)) 5883 sock_owned_by_me(sk); 5884 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5885 } 5886 5887 static int __bpf_getsockopt(struct sock *sk, int level, int optname, 5888 char *optval, int optlen) 5889 { 5890 int err, saved_optlen = optlen; 5891 5892 if (!sk_fullsock(sk)) { 5893 err = -EINVAL; 5894 goto done; 5895 } 5896 5897 if (level == SOL_SOCKET) 5898 err = sol_socket_sockopt(sk, optname, optval, &optlen, true); 5899 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5900 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true); 5901 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5902 err = sol_ip_sockopt(sk, optname, optval, &optlen, true); 5903 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5904 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true); 5905 else 5906 err = -EINVAL; 5907 5908 done: 5909 if (err) 5910 optlen = 0; 5911 if (optlen < saved_optlen) 5912 memset(optval + optlen, 0, saved_optlen - optlen); 5913 return err; 5914 } 5915 5916 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 5917 char *optval, int optlen) 5918 { 5919 if (sk_fullsock(sk)) 5920 sock_owned_by_me(sk); 5921 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5922 } 5923 5924 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, 5925 int, optname, char *, optval, int, optlen) 5926 { 5927 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5928 } 5929 5930 const struct bpf_func_proto bpf_sk_setsockopt_proto = { 5931 .func = bpf_sk_setsockopt, 5932 .gpl_only = false, 5933 .ret_type = RET_INTEGER, 5934 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5935 .arg2_type = ARG_ANYTHING, 5936 .arg3_type = ARG_ANYTHING, 5937 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5938 .arg5_type = ARG_MEM_SIZE, 5939 }; 5940 5941 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level, 5942 int, optname, char *, optval, int, optlen) 5943 { 5944 return _bpf_getsockopt(sk, level, optname, optval, optlen); 5945 } 5946 5947 const struct bpf_func_proto bpf_sk_getsockopt_proto = { 5948 .func = bpf_sk_getsockopt, 5949 .gpl_only = false, 5950 .ret_type = RET_INTEGER, 5951 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5952 .arg2_type = ARG_ANYTHING, 5953 .arg3_type = ARG_ANYTHING, 5954 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5955 .arg5_type = ARG_MEM_SIZE, 5956 }; 5957 5958 BPF_CALL_5(bpf_sk_setsockopt_nodelay, struct sock *, sk, int, level, 5959 int, optname, char *, optval, int, optlen) 5960 { 5961 /* 5962 * TCP_NODELAY triggers tcp_push_pending_frames() and re-enters 5963 * CA_EVENT_TX_START in bpf_tcp_cc. 5964 */ 5965 if (level == SOL_TCP && optname == TCP_NODELAY) 5966 return -EOPNOTSUPP; 5967 5968 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5969 } 5970 5971 const struct bpf_func_proto bpf_sk_setsockopt_nodelay_proto = { 5972 .func = bpf_sk_setsockopt_nodelay, 5973 .gpl_only = false, 5974 .ret_type = RET_INTEGER, 5975 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5976 .arg2_type = ARG_ANYTHING, 5977 .arg3_type = ARG_ANYTHING, 5978 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5979 .arg5_type = ARG_MEM_SIZE, 5980 }; 5981 5982 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level, 5983 int, optname, char *, optval, int, optlen) 5984 { 5985 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5986 } 5987 5988 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = { 5989 .func = bpf_unlocked_sk_setsockopt, 5990 .gpl_only = false, 5991 .ret_type = RET_INTEGER, 5992 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5993 .arg2_type = ARG_ANYTHING, 5994 .arg3_type = ARG_ANYTHING, 5995 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5996 .arg5_type = ARG_MEM_SIZE, 5997 }; 5998 5999 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level, 6000 int, optname, char *, optval, int, optlen) 6001 { 6002 return __bpf_getsockopt(sk, level, optname, optval, optlen); 6003 } 6004 6005 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = { 6006 .func = bpf_unlocked_sk_getsockopt, 6007 .gpl_only = false, 6008 .ret_type = RET_INTEGER, 6009 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6010 .arg2_type = ARG_ANYTHING, 6011 .arg3_type = ARG_ANYTHING, 6012 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 6013 .arg5_type = ARG_MEM_SIZE, 6014 }; 6015 6016 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 6017 int, level, int, optname, char *, optval, int, optlen) 6018 { 6019 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 6020 } 6021 6022 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 6023 .func = bpf_sock_addr_setsockopt, 6024 .gpl_only = false, 6025 .ret_type = RET_INTEGER, 6026 .arg1_type = ARG_PTR_TO_CTX, 6027 .arg2_type = ARG_ANYTHING, 6028 .arg3_type = ARG_ANYTHING, 6029 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6030 .arg5_type = ARG_MEM_SIZE, 6031 }; 6032 6033 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 6034 int, level, int, optname, char *, optval, int, optlen) 6035 { 6036 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 6037 } 6038 6039 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 6040 .func = bpf_sock_addr_getsockopt, 6041 .gpl_only = false, 6042 .ret_type = RET_INTEGER, 6043 .arg1_type = ARG_PTR_TO_CTX, 6044 .arg2_type = ARG_ANYTHING, 6045 .arg3_type = ARG_ANYTHING, 6046 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 6047 .arg5_type = ARG_MEM_SIZE, 6048 }; 6049 6050 static int sk_bpf_set_get_bypass_prot_mem(struct sock *sk, 6051 char *optval, int optlen, 6052 bool getopt) 6053 { 6054 int val; 6055 6056 if (optlen != sizeof(int)) 6057 return -EINVAL; 6058 6059 if (!sk_has_account(sk)) 6060 return -EOPNOTSUPP; 6061 6062 if (getopt) { 6063 *(int *)optval = sk->sk_bypass_prot_mem; 6064 return 0; 6065 } 6066 6067 val = *(int *)optval; 6068 if (val < 0 || val > 1) 6069 return -EINVAL; 6070 6071 sk->sk_bypass_prot_mem = val; 6072 return 0; 6073 } 6074 6075 BPF_CALL_5(bpf_sock_create_setsockopt, struct sock *, sk, int, level, 6076 int, optname, char *, optval, int, optlen) 6077 { 6078 if (level == SOL_SOCKET && optname == SK_BPF_BYPASS_PROT_MEM) 6079 return sk_bpf_set_get_bypass_prot_mem(sk, optval, optlen, false); 6080 6081 return __bpf_setsockopt(sk, level, optname, optval, optlen); 6082 } 6083 6084 static const struct bpf_func_proto bpf_sock_create_setsockopt_proto = { 6085 .func = bpf_sock_create_setsockopt, 6086 .gpl_only = false, 6087 .ret_type = RET_INTEGER, 6088 .arg1_type = ARG_PTR_TO_CTX, 6089 .arg2_type = ARG_ANYTHING, 6090 .arg3_type = ARG_ANYTHING, 6091 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6092 .arg5_type = ARG_MEM_SIZE, 6093 }; 6094 6095 BPF_CALL_5(bpf_sock_create_getsockopt, struct sock *, sk, int, level, 6096 int, optname, char *, optval, int, optlen) 6097 { 6098 if (level == SOL_SOCKET && optname == SK_BPF_BYPASS_PROT_MEM) { 6099 int err = sk_bpf_set_get_bypass_prot_mem(sk, optval, optlen, true); 6100 6101 if (err) 6102 memset(optval, 0, optlen); 6103 6104 return err; 6105 } 6106 6107 return __bpf_getsockopt(sk, level, optname, optval, optlen); 6108 } 6109 6110 static const struct bpf_func_proto bpf_sock_create_getsockopt_proto = { 6111 .func = bpf_sock_create_getsockopt, 6112 .gpl_only = false, 6113 .ret_type = RET_INTEGER, 6114 .arg1_type = ARG_PTR_TO_CTX, 6115 .arg2_type = ARG_ANYTHING, 6116 .arg3_type = ARG_ANYTHING, 6117 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 6118 .arg5_type = ARG_MEM_SIZE, 6119 }; 6120 6121 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 6122 int, level, int, optname, char *, optval, int, optlen) 6123 { 6124 if (!is_locked_tcp_sock_ops(bpf_sock)) 6125 return -EOPNOTSUPP; 6126 6127 /* TCP_NODELAY triggers tcp_push_pending_frames() and re-enters these callbacks. */ 6128 if ((bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB || 6129 bpf_sock->op == BPF_SOCK_OPS_WRITE_HDR_OPT_CB) && 6130 level == SOL_TCP && optname == TCP_NODELAY) 6131 return -EOPNOTSUPP; 6132 6133 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 6134 } 6135 6136 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 6137 .func = bpf_sock_ops_setsockopt, 6138 .gpl_only = false, 6139 .ret_type = RET_INTEGER, 6140 .arg1_type = ARG_PTR_TO_CTX, 6141 .arg2_type = ARG_ANYTHING, 6142 .arg3_type = ARG_ANYTHING, 6143 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6144 .arg5_type = ARG_MEM_SIZE, 6145 }; 6146 6147 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 6148 int optname, const u8 **start) 6149 { 6150 struct sk_buff *syn_skb = bpf_sock->syn_skb; 6151 const u8 *hdr_start; 6152 int ret; 6153 6154 if (syn_skb) { 6155 /* sk is a request_sock here */ 6156 6157 if (optname == TCP_BPF_SYN) { 6158 hdr_start = syn_skb->data; 6159 ret = tcp_hdrlen(syn_skb); 6160 } else if (optname == TCP_BPF_SYN_IP) { 6161 hdr_start = skb_network_header(syn_skb); 6162 ret = skb_network_header_len(syn_skb) + 6163 tcp_hdrlen(syn_skb); 6164 } else { 6165 /* optname == TCP_BPF_SYN_MAC */ 6166 hdr_start = skb_mac_header(syn_skb); 6167 ret = skb_mac_header_len(syn_skb) + 6168 skb_network_header_len(syn_skb) + 6169 tcp_hdrlen(syn_skb); 6170 } 6171 } else { 6172 struct sock *sk = bpf_sock->sk; 6173 struct saved_syn *saved_syn; 6174 6175 if (sk->sk_state == TCP_NEW_SYN_RECV) 6176 /* synack retransmit. bpf_sock->syn_skb will 6177 * not be available. It has to resort to 6178 * saved_syn (if it is saved). 6179 */ 6180 saved_syn = rcu_dereference(inet_reqsk(sk)->saved_syn); 6181 else 6182 saved_syn = tcp_sk(sk)->saved_syn; 6183 6184 if (!saved_syn) 6185 return -ENOENT; 6186 6187 if (optname == TCP_BPF_SYN) { 6188 hdr_start = saved_syn->data + 6189 saved_syn->mac_hdrlen + 6190 saved_syn->network_hdrlen; 6191 ret = saved_syn->tcp_hdrlen; 6192 } else if (optname == TCP_BPF_SYN_IP) { 6193 hdr_start = saved_syn->data + 6194 saved_syn->mac_hdrlen; 6195 ret = saved_syn->network_hdrlen + 6196 saved_syn->tcp_hdrlen; 6197 } else { 6198 /* optname == TCP_BPF_SYN_MAC */ 6199 6200 /* TCP_SAVE_SYN may not have saved the mac hdr */ 6201 if (!saved_syn->mac_hdrlen) 6202 return -ENOENT; 6203 6204 hdr_start = saved_syn->data; 6205 ret = saved_syn->mac_hdrlen + 6206 saved_syn->network_hdrlen + 6207 saved_syn->tcp_hdrlen; 6208 } 6209 } 6210 6211 *start = hdr_start; 6212 return ret; 6213 } 6214 6215 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 6216 int, level, int, optname, char *, optval, int, optlen) 6217 { 6218 if (!is_locked_tcp_sock_ops(bpf_sock)) 6219 return -EOPNOTSUPP; 6220 6221 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 6222 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 6223 int ret, copy_len = 0; 6224 const u8 *start; 6225 6226 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 6227 if (ret > 0) { 6228 copy_len = ret; 6229 if (optlen < copy_len) { 6230 copy_len = optlen; 6231 ret = -ENOSPC; 6232 } 6233 6234 memcpy(optval, start, copy_len); 6235 } 6236 6237 /* Zero out unused buffer at the end */ 6238 memset(optval + copy_len, 0, optlen - copy_len); 6239 6240 return ret; 6241 } 6242 6243 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 6244 } 6245 6246 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 6247 .func = bpf_sock_ops_getsockopt, 6248 .gpl_only = false, 6249 .ret_type = RET_INTEGER, 6250 .arg1_type = ARG_PTR_TO_CTX, 6251 .arg2_type = ARG_ANYTHING, 6252 .arg3_type = ARG_ANYTHING, 6253 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 6254 .arg5_type = ARG_MEM_SIZE, 6255 }; 6256 6257 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 6258 int, argval) 6259 { 6260 struct sock *sk = bpf_sock->sk; 6261 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 6262 6263 if (!is_locked_tcp_sock_ops(bpf_sock)) 6264 return -EOPNOTSUPP; 6265 6266 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 6267 return -EINVAL; 6268 6269 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 6270 6271 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 6272 } 6273 6274 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 6275 .func = bpf_sock_ops_cb_flags_set, 6276 .gpl_only = false, 6277 .ret_type = RET_INTEGER, 6278 .arg1_type = ARG_PTR_TO_CTX, 6279 .arg2_type = ARG_ANYTHING, 6280 }; 6281 6282 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 6283 int, addr_len) 6284 { 6285 #ifdef CONFIG_INET 6286 struct sock *sk = ctx->sk; 6287 u32 flags = BIND_FROM_BPF; 6288 int err; 6289 6290 err = -EINVAL; 6291 if (addr_len < offsetofend(struct sockaddr, sa_family)) 6292 return err; 6293 if (addr->sa_family == AF_INET) { 6294 if (addr_len < sizeof(struct sockaddr_in)) 6295 return err; 6296 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 6297 flags |= BIND_FORCE_ADDRESS_NO_PORT; 6298 return __inet_bind(sk, (struct sockaddr_unsized *)addr, addr_len, flags); 6299 #if IS_ENABLED(CONFIG_IPV6) 6300 } else if (addr->sa_family == AF_INET6) { 6301 if (addr_len < SIN6_LEN_RFC2133) 6302 return err; 6303 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 6304 flags |= BIND_FORCE_ADDRESS_NO_PORT; 6305 6306 return __inet6_bind(sk, (struct sockaddr_unsized *)addr, 6307 addr_len, flags); 6308 #endif /* CONFIG_IPV6 */ 6309 } 6310 #endif /* CONFIG_INET */ 6311 6312 return -EAFNOSUPPORT; 6313 } 6314 6315 static const struct bpf_func_proto bpf_bind_proto = { 6316 .func = bpf_bind, 6317 .gpl_only = false, 6318 .ret_type = RET_INTEGER, 6319 .arg1_type = ARG_PTR_TO_CTX, 6320 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6321 .arg3_type = ARG_MEM_SIZE, 6322 }; 6323 6324 #ifdef CONFIG_XFRM 6325 6326 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \ 6327 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)) 6328 6329 struct metadata_dst __percpu *xfrm_bpf_md_dst; 6330 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst); 6331 6332 #endif 6333 6334 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 6335 struct bpf_xfrm_state *, to, u32, size, u64, flags) 6336 { 6337 const struct sec_path *sp = skb_sec_path(skb); 6338 const struct xfrm_state *x; 6339 6340 if (!sp || unlikely(index >= sp->len || flags)) 6341 goto err_clear; 6342 6343 x = sp->xvec[index]; 6344 6345 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 6346 goto err_clear; 6347 6348 to->reqid = x->props.reqid; 6349 to->spi = x->id.spi; 6350 to->family = x->props.family; 6351 to->ext = 0; 6352 6353 if (to->family == AF_INET6) { 6354 memcpy(to->remote_ipv6, x->props.saddr.a6, 6355 sizeof(to->remote_ipv6)); 6356 } else { 6357 to->remote_ipv4 = x->props.saddr.a4; 6358 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 6359 } 6360 6361 return 0; 6362 err_clear: 6363 memset(to, 0, size); 6364 return -EINVAL; 6365 } 6366 6367 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 6368 .func = bpf_skb_get_xfrm_state, 6369 .gpl_only = false, 6370 .ret_type = RET_INTEGER, 6371 .arg1_type = ARG_PTR_TO_CTX, 6372 .arg2_type = ARG_ANYTHING, 6373 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 6374 .arg4_type = ARG_MEM_SIZE, 6375 .arg5_type = ARG_ANYTHING, 6376 }; 6377 #endif 6378 6379 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 6380 static int bpf_fib_set_fwd_params(struct net_device *dev, 6381 struct bpf_fib_lookup *params, 6382 u32 flags, u32 mtu, u32 in_ifindex) 6383 { 6384 params->h_vlan_TCI = 0; 6385 params->h_vlan_proto = 0; 6386 6387 #if IS_ENABLED(CONFIG_VLAN_8021Q) 6388 if ((flags & BPF_FIB_LOOKUP_VLAN) && is_vlan_dev(dev)) { 6389 struct net_device *real_dev = vlan_dev_priv(dev)->real_dev; 6390 6391 if (!is_vlan_dev(real_dev) && 6392 net_eq(dev_net(real_dev), dev_net(dev))) { 6393 params->h_vlan_proto = vlan_dev_vlan_proto(dev); 6394 params->h_vlan_TCI = htons(vlan_dev_vlan_id(dev)); 6395 params->ifindex = real_dev->ifindex; 6396 } else { 6397 params->ifindex = in_ifindex; 6398 return BPF_FIB_LKUP_RET_VLAN_FAILURE; 6399 } 6400 } 6401 #endif 6402 6403 if (mtu) 6404 params->mtu_result = mtu; /* union with tot_len */ 6405 6406 return 0; 6407 } 6408 6409 static struct net_device *bpf_fib_vlan_input_dev(struct net_device *dev, 6410 const struct bpf_fib_lookup *params) 6411 { 6412 __be16 proto = params->h_vlan_proto; 6413 struct net_device *vlan_dev; 6414 u16 vid; 6415 6416 if (proto != htons(ETH_P_8021Q) && proto != htons(ETH_P_8021AD)) 6417 return ERR_PTR(-EINVAL); 6418 6419 vid = ntohs(params->h_vlan_TCI) & VLAN_VID_MASK; 6420 vlan_dev = __vlan_find_dev_deep_rcu(dev, proto, vid); 6421 if (!vlan_dev || !(vlan_dev->flags & IFF_UP) || 6422 !net_eq(dev_net(vlan_dev), dev_net(dev))) 6423 return NULL; 6424 6425 return vlan_dev; 6426 } 6427 #endif 6428 6429 #if IS_ENABLED(CONFIG_INET) 6430 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 6431 u32 flags, bool check_mtu) 6432 { 6433 u32 in_ifindex = params->ifindex; 6434 struct neighbour *neigh = NULL; 6435 struct fib_nh_common *nhc; 6436 struct in_device *in_dev; 6437 struct net_device *dev; 6438 struct fib_result res; 6439 struct flowi4 fl4 = {}; 6440 u32 mtu = 0; 6441 int err; 6442 6443 dev = dev_get_by_index_rcu(net, params->ifindex); 6444 if (unlikely(!dev)) 6445 return -ENODEV; 6446 6447 if (flags & BPF_FIB_LOOKUP_VLAN_INPUT) { 6448 dev = bpf_fib_vlan_input_dev(dev, params); 6449 if (IS_ERR(dev)) 6450 return PTR_ERR(dev); 6451 if (!dev) 6452 return BPF_FIB_LKUP_RET_NOT_FWDED; 6453 } 6454 6455 /* verify forwarding is enabled on this interface */ 6456 in_dev = __in_dev_get_rcu(dev); 6457 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 6458 return BPF_FIB_LKUP_RET_FWD_DISABLED; 6459 6460 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 6461 fl4.flowi4_iif = 1; 6462 fl4.flowi4_oif = params->ifindex; 6463 } else { 6464 /* 6465 * dev->ifindex, not params->ifindex: VLAN_INPUT may have 6466 * resolved dev to a subinterface above. 6467 */ 6468 fl4.flowi4_iif = dev->ifindex; 6469 fl4.flowi4_oif = 0; 6470 } 6471 fl4.flowi4_dscp = inet_dsfield_to_dscp(params->tos); 6472 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 6473 fl4.flowi4_flags = 0; 6474 6475 fl4.flowi4_proto = params->l4_protocol; 6476 fl4.daddr = params->ipv4_dst; 6477 fl4.saddr = params->ipv4_src; 6478 fl4.fl4_sport = params->sport; 6479 fl4.fl4_dport = params->dport; 6480 fl4.flowi4_multipath_hash = 0; 6481 6482 if (flags & BPF_FIB_LOOKUP_DIRECT) { 6483 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 6484 struct fib_table *tb; 6485 6486 if (flags & BPF_FIB_LOOKUP_TBID) { 6487 tbid = params->tbid; 6488 /* zero out for vlan output */ 6489 params->tbid = 0; 6490 } 6491 6492 tb = fib_get_table(net, tbid); 6493 if (unlikely(!tb)) 6494 return BPF_FIB_LKUP_RET_NOT_FWDED; 6495 6496 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 6497 } else { 6498 if (flags & BPF_FIB_LOOKUP_MARK) 6499 fl4.flowi4_mark = params->mark; 6500 else 6501 fl4.flowi4_mark = 0; 6502 fl4.flowi4_secid = 0; 6503 fl4.flowi4_tun_key.tun_id = 0; 6504 fl4.flowi4_uid = sock_net_uid(net, NULL); 6505 6506 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 6507 } 6508 6509 if (err) { 6510 /* map fib lookup errors to RTN_ type */ 6511 if (err == -EINVAL) 6512 return BPF_FIB_LKUP_RET_BLACKHOLE; 6513 if (err == -EHOSTUNREACH) 6514 return BPF_FIB_LKUP_RET_UNREACHABLE; 6515 if (err == -EACCES) 6516 return BPF_FIB_LKUP_RET_PROHIBIT; 6517 6518 return BPF_FIB_LKUP_RET_NOT_FWDED; 6519 } 6520 6521 if (res.type != RTN_UNICAST) 6522 return BPF_FIB_LKUP_RET_NOT_FWDED; 6523 6524 if (fib_info_num_path(res.fi) > 1) 6525 fib_select_path(net, &res, &fl4, NULL); 6526 6527 if (check_mtu) { 6528 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 6529 if (params->tot_len > mtu) { 6530 params->mtu_result = mtu; /* union with tot_len */ 6531 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 6532 } 6533 } 6534 6535 nhc = res.nhc; 6536 6537 /* do not handle lwt encaps right now */ 6538 if (nhc->nhc_lwtstate) 6539 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 6540 6541 dev = nhc->nhc_dev; 6542 6543 params->rt_metric = res.fi->fib_priority; 6544 params->ifindex = dev->ifindex; 6545 6546 if (flags & BPF_FIB_LOOKUP_SRC) 6547 params->ipv4_src = fib_result_prefsrc(net, &res); 6548 6549 /* xdp and cls_bpf programs are run in RCU-bh so 6550 * rcu_read_lock_bh is not needed here 6551 */ 6552 if (likely(nhc->nhc_gw_family != AF_INET6)) { 6553 if (nhc->nhc_gw_family) 6554 params->ipv4_dst = nhc->nhc_gw.ipv4; 6555 } else { 6556 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 6557 6558 params->family = AF_INET6; 6559 *dst = nhc->nhc_gw.ipv6; 6560 } 6561 6562 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH) 6563 goto set_fwd_params; 6564 6565 if (likely(nhc->nhc_gw_family != AF_INET6)) 6566 neigh = __ipv4_neigh_lookup_noref(dev, 6567 (__force u32)params->ipv4_dst); 6568 else if (IS_ENABLED(CONFIG_IPV6)) 6569 neigh = __ipv6_neigh_lookup_noref(dev, params->ipv6_dst); 6570 6571 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID)) 6572 return BPF_FIB_LKUP_RET_NO_NEIGH; 6573 memcpy(params->dmac, neigh->ha, ETH_ALEN); 6574 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 6575 6576 set_fwd_params: 6577 return bpf_fib_set_fwd_params(dev, params, flags, mtu, in_ifindex); 6578 } 6579 #endif 6580 6581 #if IS_ENABLED(CONFIG_IPV6) 6582 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 6583 u32 flags, bool check_mtu) 6584 { 6585 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 6586 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 6587 u32 in_ifindex = params->ifindex; 6588 struct fib6_result res = {}; 6589 struct neighbour *neigh; 6590 struct net_device *dev; 6591 struct inet6_dev *idev; 6592 struct flowi6 fl6 = {}; 6593 int strict = 0; 6594 int oif, err; 6595 u32 mtu = 0; 6596 6597 /* link local addresses are never forwarded */ 6598 if (rt6_need_strict(dst) || rt6_need_strict(src)) 6599 return BPF_FIB_LKUP_RET_NOT_FWDED; 6600 6601 dev = dev_get_by_index_rcu(net, params->ifindex); 6602 if (unlikely(!dev)) 6603 return -ENODEV; 6604 6605 if (flags & BPF_FIB_LOOKUP_VLAN_INPUT) { 6606 dev = bpf_fib_vlan_input_dev(dev, params); 6607 if (IS_ERR(dev)) 6608 return PTR_ERR(dev); 6609 if (!dev) 6610 return BPF_FIB_LKUP_RET_NOT_FWDED; 6611 } 6612 6613 idev = __in6_dev_get_safely(dev); 6614 if (unlikely(!idev || !READ_ONCE(idev->cnf.forwarding))) 6615 return BPF_FIB_LKUP_RET_FWD_DISABLED; 6616 6617 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 6618 fl6.flowi6_iif = 1; 6619 oif = fl6.flowi6_oif = params->ifindex; 6620 } else { 6621 /* 6622 * dev->ifindex, not params->ifindex: VLAN_INPUT may have 6623 * resolved dev to a subinterface above. 6624 */ 6625 oif = dev->ifindex; 6626 fl6.flowi6_iif = oif; 6627 fl6.flowi6_oif = 0; 6628 strict = RT6_LOOKUP_F_HAS_SADDR; 6629 } 6630 fl6.flowlabel = params->flowinfo; 6631 fl6.flowi6_scope = 0; 6632 fl6.flowi6_flags = 0; 6633 fl6.mp_hash = 0; 6634 6635 fl6.flowi6_proto = params->l4_protocol; 6636 fl6.daddr = *dst; 6637 fl6.saddr = *src; 6638 fl6.fl6_sport = params->sport; 6639 fl6.fl6_dport = params->dport; 6640 6641 if (flags & BPF_FIB_LOOKUP_DIRECT) { 6642 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 6643 struct fib6_table *tb; 6644 6645 if (flags & BPF_FIB_LOOKUP_TBID) { 6646 tbid = params->tbid; 6647 /* zero out for vlan output */ 6648 params->tbid = 0; 6649 } 6650 6651 tb = fib6_get_table(net, tbid); 6652 if (unlikely(!tb)) 6653 return BPF_FIB_LKUP_RET_NOT_FWDED; 6654 6655 err = fib6_table_lookup(net, tb, oif, &fl6, &res, strict); 6656 } else { 6657 if (flags & BPF_FIB_LOOKUP_MARK) 6658 fl6.flowi6_mark = params->mark; 6659 else 6660 fl6.flowi6_mark = 0; 6661 fl6.flowi6_secid = 0; 6662 fl6.flowi6_tun_key.tun_id = 0; 6663 fl6.flowi6_uid = sock_net_uid(net, NULL); 6664 6665 err = fib6_lookup(net, oif, &fl6, &res, strict); 6666 } 6667 6668 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 6669 res.f6i == net->ipv6.fib6_null_entry)) 6670 return BPF_FIB_LKUP_RET_NOT_FWDED; 6671 6672 switch (res.fib6_type) { 6673 /* only unicast is forwarded */ 6674 case RTN_UNICAST: 6675 break; 6676 case RTN_BLACKHOLE: 6677 return BPF_FIB_LKUP_RET_BLACKHOLE; 6678 case RTN_UNREACHABLE: 6679 return BPF_FIB_LKUP_RET_UNREACHABLE; 6680 case RTN_PROHIBIT: 6681 return BPF_FIB_LKUP_RET_PROHIBIT; 6682 default: 6683 return BPF_FIB_LKUP_RET_NOT_FWDED; 6684 } 6685 6686 fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 6687 fl6.flowi6_oif != 0, NULL, strict); 6688 6689 if (check_mtu) { 6690 mtu = ip6_mtu_from_fib6(&res, dst, src); 6691 if (params->tot_len > mtu) { 6692 params->mtu_result = mtu; /* union with tot_len */ 6693 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 6694 } 6695 } 6696 6697 if (res.nh->fib_nh_lws) 6698 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 6699 6700 if (res.nh->fib_nh_gw_family) 6701 *dst = res.nh->fib_nh_gw6; 6702 6703 dev = res.nh->fib_nh_dev; 6704 params->rt_metric = res.f6i->fib6_metric; 6705 params->ifindex = dev->ifindex; 6706 6707 if (flags & BPF_FIB_LOOKUP_SRC) { 6708 if (res.f6i->fib6_prefsrc.plen) { 6709 *src = res.f6i->fib6_prefsrc.addr; 6710 } else { 6711 err = ipv6_dev_get_saddr(net, dev, &fl6.daddr, 0, src); 6712 if (err) 6713 return BPF_FIB_LKUP_RET_NO_SRC_ADDR; 6714 } 6715 } 6716 6717 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH) 6718 goto set_fwd_params; 6719 6720 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 6721 * not needed here. 6722 */ 6723 neigh = __ipv6_neigh_lookup_noref(dev, dst); 6724 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID)) 6725 return BPF_FIB_LKUP_RET_NO_NEIGH; 6726 memcpy(params->dmac, neigh->ha, ETH_ALEN); 6727 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 6728 6729 set_fwd_params: 6730 return bpf_fib_set_fwd_params(dev, params, flags, mtu, in_ifindex); 6731 } 6732 #endif 6733 6734 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \ 6735 BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID | \ 6736 BPF_FIB_LOOKUP_SRC | BPF_FIB_LOOKUP_MARK | \ 6737 BPF_FIB_LOOKUP_VLAN | BPF_FIB_LOOKUP_VLAN_INPUT) 6738 6739 static bool bpf_fib_lookup_flags_ok(u32 flags) 6740 { 6741 if (flags & ~BPF_FIB_LOOKUP_MASK) 6742 return false; 6743 6744 if ((flags & BPF_FIB_LOOKUP_VLAN_INPUT) && 6745 (flags & (BPF_FIB_LOOKUP_TBID | BPF_FIB_LOOKUP_OUTPUT))) 6746 return false; 6747 6748 return true; 6749 } 6750 6751 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 6752 struct bpf_fib_lookup *, params, int, plen, u32, flags) 6753 { 6754 if (plen < sizeof(*params)) 6755 return -EINVAL; 6756 6757 if (!bpf_fib_lookup_flags_ok(flags)) 6758 return -EINVAL; 6759 6760 switch (params->family) { 6761 #if IS_ENABLED(CONFIG_INET) 6762 case AF_INET: 6763 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 6764 flags, true); 6765 #endif 6766 #if IS_ENABLED(CONFIG_IPV6) 6767 case AF_INET6: 6768 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 6769 flags, true); 6770 #endif 6771 } 6772 return -EAFNOSUPPORT; 6773 } 6774 6775 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 6776 .func = bpf_xdp_fib_lookup, 6777 .gpl_only = true, 6778 .ret_type = RET_INTEGER, 6779 .arg1_type = ARG_PTR_TO_CTX, 6780 .arg2_type = ARG_PTR_TO_MEM | MEM_WRITE, 6781 .arg3_type = ARG_MEM_SIZE, 6782 .arg4_type = ARG_ANYTHING, 6783 }; 6784 6785 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 6786 struct bpf_fib_lookup *, params, int, plen, u32, flags) 6787 { 6788 struct net *net = dev_net(skb->dev); 6789 int rc = -EAFNOSUPPORT; 6790 bool check_mtu = false; 6791 6792 if (plen < sizeof(*params)) 6793 return -EINVAL; 6794 6795 if (!bpf_fib_lookup_flags_ok(flags)) 6796 return -EINVAL; 6797 6798 if (flags & BPF_FIB_LOOKUP_VLAN) 6799 return -EINVAL; 6800 6801 if (params->tot_len) 6802 check_mtu = true; 6803 6804 switch (params->family) { 6805 #if IS_ENABLED(CONFIG_INET) 6806 case AF_INET: 6807 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu); 6808 break; 6809 #endif 6810 #if IS_ENABLED(CONFIG_IPV6) 6811 case AF_INET6: 6812 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu); 6813 break; 6814 #endif 6815 } 6816 6817 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) { 6818 struct net_device *dev; 6819 6820 /* When tot_len isn't provided by user, check skb 6821 * against MTU of FIB lookup resulting net_device 6822 */ 6823 dev = dev_get_by_index_rcu(net, params->ifindex); 6824 if (unlikely(!dev)) 6825 return -ENODEV; 6826 if (!is_skb_forwardable(dev, skb)) 6827 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 6828 6829 params->mtu_result = dev->mtu; /* union with tot_len */ 6830 } 6831 6832 return rc; 6833 } 6834 6835 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 6836 .func = bpf_skb_fib_lookup, 6837 .gpl_only = true, 6838 .ret_type = RET_INTEGER, 6839 .arg1_type = ARG_PTR_TO_CTX, 6840 .arg2_type = ARG_PTR_TO_MEM | MEM_WRITE, 6841 .arg3_type = ARG_MEM_SIZE, 6842 .arg4_type = ARG_ANYTHING, 6843 }; 6844 6845 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr, 6846 u32 ifindex) 6847 { 6848 struct net *netns = dev_net(dev_curr); 6849 6850 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */ 6851 if (ifindex == 0) 6852 return dev_curr; 6853 6854 return dev_get_by_index_rcu(netns, ifindex); 6855 } 6856 6857 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb, 6858 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6859 { 6860 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6861 struct net_device *dev = skb->dev; 6862 int mtu, dev_len, skb_len; 6863 6864 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) 6865 return -EINVAL; 6866 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) 6867 return -EINVAL; 6868 6869 dev = __dev_via_ifindex(dev, ifindex); 6870 if (unlikely(!dev)) 6871 return -ENODEV; 6872 6873 mtu = READ_ONCE(dev->mtu); 6874 dev_len = mtu + dev->hard_header_len; 6875 6876 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6877 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len; 6878 6879 skb_len += len_diff; /* minus result pass check */ 6880 if (skb_len <= dev_len) { 6881 ret = BPF_MTU_CHK_RET_SUCCESS; 6882 goto out; 6883 } 6884 /* At this point, skb->len exceed MTU, but as it include length of all 6885 * segments, it can still be below MTU. The SKB can possibly get 6886 * re-segmented in transmit path (see validate_xmit_skb). Thus, user 6887 * must choose if segs are to be MTU checked. 6888 */ 6889 if (skb_is_gso(skb)) { 6890 ret = BPF_MTU_CHK_RET_SUCCESS; 6891 if (flags & BPF_MTU_CHK_SEGS) { 6892 if (!skb_transport_header_was_set(skb)) 6893 return -EINVAL; 6894 if (!skb_gso_validate_network_len(skb, mtu)) 6895 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG; 6896 } 6897 } 6898 out: 6899 *mtu_len = mtu; 6900 return ret; 6901 } 6902 6903 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp, 6904 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6905 { 6906 struct net_device *dev = xdp->rxq->dev; 6907 int xdp_len = xdp->data_end - xdp->data; 6908 int ret = BPF_MTU_CHK_RET_SUCCESS; 6909 int mtu, dev_len; 6910 6911 /* XDP variant doesn't support multi-buffer segment check (yet) */ 6912 if (unlikely(flags)) 6913 return -EINVAL; 6914 6915 dev = __dev_via_ifindex(dev, ifindex); 6916 if (unlikely(!dev)) 6917 return -ENODEV; 6918 6919 mtu = READ_ONCE(dev->mtu); 6920 dev_len = mtu + dev->hard_header_len; 6921 6922 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6923 if (*mtu_len) 6924 xdp_len = *mtu_len + dev->hard_header_len; 6925 6926 xdp_len += len_diff; /* minus result pass check */ 6927 if (xdp_len > dev_len) 6928 ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6929 6930 *mtu_len = mtu; 6931 return ret; 6932 } 6933 6934 static const struct bpf_func_proto bpf_skb_check_mtu_proto = { 6935 .func = bpf_skb_check_mtu, 6936 .gpl_only = true, 6937 .ret_type = RET_INTEGER, 6938 .arg1_type = ARG_PTR_TO_CTX, 6939 .arg2_type = ARG_ANYTHING, 6940 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_WRITE | MEM_ALIGNED, 6941 .arg3_size = sizeof(u32), 6942 .arg4_type = ARG_ANYTHING, 6943 .arg5_type = ARG_ANYTHING, 6944 }; 6945 6946 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = { 6947 .func = bpf_xdp_check_mtu, 6948 .gpl_only = true, 6949 .ret_type = RET_INTEGER, 6950 .arg1_type = ARG_PTR_TO_CTX, 6951 .arg2_type = ARG_ANYTHING, 6952 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_WRITE | MEM_ALIGNED, 6953 .arg3_size = sizeof(u32), 6954 .arg4_type = ARG_ANYTHING, 6955 .arg5_type = ARG_ANYTHING, 6956 }; 6957 6958 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6959 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 6960 { 6961 int err; 6962 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 6963 6964 /* With ipv6.disable=1 the netns has no seg6 data and no IPv6 tables. */ 6965 if (!ipv6_mod_enabled()) 6966 return -EAFNOSUPPORT; 6967 6968 if (!seg6_validate_srh(srh, len, false)) 6969 return -EINVAL; 6970 6971 /* 6972 * The encap mode names IPv6 as the inner protocol, and neither mode 6973 * updates skb->protocol. 6974 */ 6975 if (skb->protocol != htons(ETH_P_IPV6)) 6976 return -EBADMSG; 6977 6978 switch (type) { 6979 case BPF_LWT_ENCAP_SEG6_INLINE: 6980 err = seg6_do_srh_inline(skb, srh); 6981 break; 6982 case BPF_LWT_ENCAP_SEG6: 6983 skb_reset_inner_headers(skb); 6984 skb->encapsulation = 1; 6985 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 6986 break; 6987 default: 6988 return -EINVAL; 6989 } 6990 6991 bpf_compute_data_pointers(skb); 6992 if (err) 6993 return err; 6994 6995 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 6996 6997 return seg6_lookup_nexthop(skb, NULL, 0); 6998 } 6999 #endif /* CONFIG_IPV6_SEG6_BPF */ 7000 7001 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 7002 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 7003 bool ingress) 7004 { 7005 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 7006 } 7007 #endif 7008 7009 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 7010 u32, len) 7011 { 7012 switch (type) { 7013 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7014 case BPF_LWT_ENCAP_SEG6: 7015 case BPF_LWT_ENCAP_SEG6_INLINE: 7016 return bpf_push_seg6_encap(skb, type, hdr, len); 7017 #endif 7018 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 7019 case BPF_LWT_ENCAP_IP: 7020 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 7021 #endif 7022 default: 7023 return -EINVAL; 7024 } 7025 } 7026 7027 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 7028 void *, hdr, u32, len) 7029 { 7030 switch (type) { 7031 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 7032 case BPF_LWT_ENCAP_IP: 7033 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 7034 #endif 7035 default: 7036 return -EINVAL; 7037 } 7038 } 7039 7040 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 7041 .func = bpf_lwt_in_push_encap, 7042 .gpl_only = false, 7043 .ret_type = RET_INTEGER, 7044 .arg1_type = ARG_PTR_TO_CTX, 7045 .arg2_type = ARG_ANYTHING, 7046 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7047 .arg4_type = ARG_MEM_SIZE 7048 }; 7049 7050 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 7051 .func = bpf_lwt_xmit_push_encap, 7052 .gpl_only = false, 7053 .ret_type = RET_INTEGER, 7054 .arg1_type = ARG_PTR_TO_CTX, 7055 .arg2_type = ARG_ANYTHING, 7056 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7057 .arg4_type = ARG_MEM_SIZE 7058 }; 7059 7060 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7061 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 7062 const void *, from, u32, len) 7063 { 7064 struct seg6_bpf_srh_state *srh_state = 7065 this_cpu_ptr(&seg6_bpf_srh_states); 7066 struct ipv6_sr_hdr *srh = srh_state->srh; 7067 void *srh_tlvs, *srh_end, *ptr; 7068 int srhoff; 7069 7070 lockdep_assert_held(&srh_state->bh_lock); 7071 if (srh == NULL) 7072 return -EINVAL; 7073 7074 srhoff = (unsigned char *)srh - skb->data; 7075 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 7076 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 7077 7078 ptr = skb->data + offset; 7079 if (ptr >= srh_tlvs && ptr + len <= srh_end) 7080 srh_state->valid = false; 7081 else if (ptr < (void *)&srh->flags || 7082 ptr + len > (void *)&srh->segments) 7083 return -EFAULT; 7084 7085 if (unlikely(bpf_try_make_writable(skb, offset + len))) 7086 return -EFAULT; 7087 /* The head may have moved. */ 7088 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 7089 7090 memcpy(skb->data + offset, from, len); 7091 return 0; 7092 } 7093 7094 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 7095 .func = bpf_lwt_seg6_store_bytes, 7096 .gpl_only = false, 7097 .ret_type = RET_INTEGER, 7098 .arg1_type = ARG_PTR_TO_CTX, 7099 .arg2_type = ARG_ANYTHING, 7100 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7101 .arg4_type = ARG_MEM_SIZE 7102 }; 7103 7104 static void bpf_update_srh_state(struct sk_buff *skb) 7105 { 7106 struct seg6_bpf_srh_state *srh_state = 7107 this_cpu_ptr(&seg6_bpf_srh_states); 7108 7109 /* Behind a decapsulation this is a header that nobody has looked at */ 7110 srh_state->srh = seg6_get_srh(skb, 0); 7111 if (srh_state->srh) { 7112 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 7113 srh_state->valid = true; 7114 } 7115 /* it may have been pulled into the head */ 7116 bpf_compute_data_pointers(skb); 7117 } 7118 7119 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 7120 u32, action, void *, param, u32, param_len) 7121 { 7122 struct seg6_bpf_srh_state *srh_state = 7123 this_cpu_ptr(&seg6_bpf_srh_states); 7124 int hdroff = 0; 7125 int err; 7126 7127 lockdep_assert_held(&srh_state->bh_lock); 7128 switch (action) { 7129 case SEG6_LOCAL_ACTION_END_X: 7130 if (!seg6_bpf_has_valid_srh(skb)) 7131 return -EBADMSG; 7132 if (param_len != sizeof(struct in6_addr)) 7133 return -EINVAL; 7134 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 7135 case SEG6_LOCAL_ACTION_END_T: 7136 if (!seg6_bpf_has_valid_srh(skb)) 7137 return -EBADMSG; 7138 if (param_len != sizeof(int)) 7139 return -EINVAL; 7140 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 7141 case SEG6_LOCAL_ACTION_END_DT6: 7142 if (!seg6_bpf_has_valid_srh(skb)) 7143 return -EBADMSG; 7144 if (param_len != sizeof(int)) 7145 return -EINVAL; 7146 7147 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 7148 return -EBADMSG; 7149 /* As long as nothing has been changed, the program goes on */ 7150 if (!pskb_may_pull(skb, hdroff + sizeof(struct ipv6hdr))) { 7151 /* but the head may have moved */ 7152 bpf_update_srh_state(skb); 7153 return -EBADMSG; 7154 } 7155 7156 __skb_pull(skb, hdroff); 7157 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 7158 skb_reset_network_header(skb); 7159 skb_reset_transport_header(skb); 7160 skb->encapsulation = 0; 7161 7162 bpf_compute_data_pointers(skb); 7163 bpf_update_srh_state(skb); 7164 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 7165 case SEG6_LOCAL_ACTION_END_B6: 7166 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 7167 return -EBADMSG; 7168 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 7169 param, param_len); 7170 /* The header can be in place although err says otherwise */ 7171 bpf_update_srh_state(skb); 7172 7173 return err; 7174 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 7175 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 7176 return -EBADMSG; 7177 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 7178 param, param_len); 7179 /* The header can be in place although err says otherwise */ 7180 bpf_update_srh_state(skb); 7181 7182 return err; 7183 default: 7184 return -EINVAL; 7185 } 7186 } 7187 7188 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 7189 .func = bpf_lwt_seg6_action, 7190 .gpl_only = false, 7191 .ret_type = RET_INTEGER, 7192 .arg1_type = ARG_PTR_TO_CTX, 7193 .arg2_type = ARG_ANYTHING, 7194 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7195 .arg4_type = ARG_MEM_SIZE 7196 }; 7197 7198 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 7199 s32, len) 7200 { 7201 struct seg6_bpf_srh_state *srh_state = 7202 this_cpu_ptr(&seg6_bpf_srh_states); 7203 bool mac_set = skb_mac_header_was_set(skb); 7204 struct ipv6_sr_hdr *srh = srh_state->srh; 7205 void *srh_end, *srh_tlvs, *ptr; 7206 struct ipv6hdr *hdr; 7207 int srhoff; 7208 int ret; 7209 7210 lockdep_assert_held(&srh_state->bh_lock); 7211 if (unlikely(srh == NULL)) 7212 return -EINVAL; 7213 7214 srhoff = (unsigned char *)srh - skb->data; 7215 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 7216 ((srh->first_segment + 1) << 4)); 7217 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 7218 srh_state->hdrlen); 7219 ptr = skb->data + offset; 7220 7221 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 7222 return -EFAULT; 7223 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 7224 return -EFAULT; 7225 /* The length field of the SRH has 8 bits, in units of 8 bytes. */ 7226 if (unlikely(len > (U8_MAX << 3) - srh_state->hdrlen)) 7227 return -EFAULT; 7228 7229 /* 7230 * skb->data is the network header here. The MAC header is in front of 7231 * it, so bpf_skb_net_hdr_push() and bpf_skb_net_hdr_pop() do not move 7232 * it with the other headers. 7233 */ 7234 if (len > 0) { 7235 ret = skb_cow_head(skb, len + skb->mac_len); 7236 if (unlikely(ret < 0)) 7237 return ret; 7238 7239 /* The push can write over the MAC header, so move it first. */ 7240 if (mac_set) 7241 memmove(skb->data - len - skb->mac_len, 7242 skb_mac_header(skb), skb->mac_len); 7243 ret = bpf_skb_net_hdr_push(skb, offset, len); 7244 } else { 7245 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 7246 /* The pop has added -len to skb->mac_header. */ 7247 if (likely(!ret) && mac_set) 7248 memmove(skb->data - skb->mac_len, 7249 skb_mac_header(skb) + len, skb->mac_len); 7250 } 7251 7252 bpf_compute_data_pointers(skb); 7253 if (unlikely(ret < 0)) 7254 return ret; 7255 7256 if (mac_set) 7257 skb_set_mac_header(skb, -skb->mac_len); 7258 else 7259 skb_unset_mac_header(skb); 7260 7261 hdr = (struct ipv6hdr *)skb->data; 7262 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 7263 7264 /* The bytes in front of offset have moved together with skb->data. */ 7265 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 7266 srh_state->hdrlen += len; 7267 srh_state->valid = false; 7268 return 0; 7269 } 7270 7271 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 7272 .func = bpf_lwt_seg6_adjust_srh, 7273 .gpl_only = false, 7274 .ret_type = RET_INTEGER, 7275 .arg1_type = ARG_PTR_TO_CTX, 7276 .arg2_type = ARG_ANYTHING, 7277 .arg3_type = ARG_ANYTHING, 7278 }; 7279 #endif /* CONFIG_IPV6_SEG6_BPF */ 7280 7281 #ifdef CONFIG_INET 7282 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 7283 int dif, int sdif, u8 family, u8 proto) 7284 { 7285 bool refcounted = false; 7286 struct sock *sk = NULL; 7287 7288 if (family == AF_INET) { 7289 __be32 src4 = tuple->ipv4.saddr; 7290 __be32 dst4 = tuple->ipv4.daddr; 7291 7292 if (proto == IPPROTO_TCP) 7293 sk = __inet_lookup(net, NULL, 0, 7294 src4, tuple->ipv4.sport, 7295 dst4, tuple->ipv4.dport, 7296 dif, sdif, &refcounted); 7297 else 7298 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 7299 dst4, tuple->ipv4.dport, 7300 dif, sdif, NULL); 7301 #if IS_ENABLED(CONFIG_IPV6) 7302 } else { 7303 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 7304 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 7305 7306 if (proto == IPPROTO_TCP) 7307 sk = __inet6_lookup(net, NULL, 0, 7308 src6, tuple->ipv6.sport, 7309 dst6, ntohs(tuple->ipv6.dport), 7310 dif, sdif, &refcounted); 7311 else if (likely(ipv6_mod_enabled())) 7312 sk = __udp6_lib_lookup(net, src6, tuple->ipv6.sport, 7313 dst6, tuple->ipv6.dport, 7314 dif, sdif, NULL); 7315 #endif 7316 } 7317 7318 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 7319 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 7320 sk = NULL; 7321 } 7322 return sk; 7323 } 7324 7325 /* bpf_skc_lookup performs the core lookup for different types of sockets, 7326 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 7327 */ 7328 static struct sock * 7329 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 7330 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 7331 u64 flags, int sdif) 7332 { 7333 struct sock *sk = NULL; 7334 struct net *net; 7335 u8 family; 7336 7337 if (len == sizeof(tuple->ipv4)) 7338 family = AF_INET; 7339 else if (len == sizeof(tuple->ipv6)) 7340 family = AF_INET6; 7341 else 7342 return NULL; 7343 7344 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX))) 7345 goto out; 7346 7347 if (sdif < 0) { 7348 if (family == AF_INET) 7349 sdif = inet_sdif(skb); 7350 else 7351 sdif = inet6_sdif(skb); 7352 } 7353 7354 if ((s32)netns_id < 0) { 7355 net = caller_net; 7356 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 7357 } else { 7358 net = get_net_ns_by_id(caller_net, netns_id); 7359 if (unlikely(!net)) 7360 goto out; 7361 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 7362 put_net(net); 7363 } 7364 7365 out: 7366 return sk; 7367 } 7368 7369 static struct sock * 7370 bpf_sk_lookup_full_sk(struct sock *sk) 7371 { 7372 struct sock *sk2 = sk_to_full_sk(sk); 7373 7374 /* 7375 * sk_to_full_sk() may return sk->rsk_listener, make sure the original 7376 * sk sock refcnt is decremented to prevent a request_sock leak. 7377 */ 7378 if (sk2 != sk) { 7379 sock_gen_put(sk); 7380 /* Ensure there is no need to bump sk2 refcnt. */ 7381 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 7382 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 7383 return NULL; 7384 } 7385 sk = sk2; 7386 } 7387 7388 return sk; 7389 } 7390 7391 static struct sock * 7392 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 7393 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 7394 u64 flags, int sdif) 7395 { 7396 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 7397 ifindex, proto, netns_id, flags, 7398 sdif); 7399 if (sk) 7400 sk = bpf_sk_lookup_full_sk(sk); 7401 return sk; 7402 } 7403 7404 static struct sock * 7405 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 7406 u8 proto, u64 netns_id, u64 flags) 7407 { 7408 struct net *caller_net; 7409 int ifindex; 7410 7411 if (skb->dev) { 7412 caller_net = dev_net(skb->dev); 7413 ifindex = skb->dev->ifindex; 7414 } else { 7415 caller_net = sock_net(skb->sk); 7416 ifindex = 0; 7417 } 7418 7419 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 7420 netns_id, flags, -1); 7421 } 7422 7423 static struct sock * 7424 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 7425 u8 proto, u64 netns_id, u64 flags) 7426 { 7427 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 7428 flags); 7429 if (sk) 7430 sk = bpf_sk_lookup_full_sk(sk); 7431 return sk; 7432 } 7433 7434 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 7435 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7436 { 7437 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 7438 netns_id, flags); 7439 } 7440 7441 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 7442 .func = bpf_skc_lookup_tcp, 7443 .gpl_only = false, 7444 .pkt_access = true, 7445 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 7446 .arg1_type = ARG_PTR_TO_CTX, 7447 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7448 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7449 .arg4_type = ARG_ANYTHING, 7450 .arg5_type = ARG_ANYTHING, 7451 }; 7452 7453 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 7454 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7455 { 7456 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 7457 netns_id, flags); 7458 } 7459 7460 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 7461 .func = bpf_sk_lookup_tcp, 7462 .gpl_only = false, 7463 .pkt_access = true, 7464 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7465 .arg1_type = ARG_PTR_TO_CTX, 7466 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7467 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7468 .arg4_type = ARG_ANYTHING, 7469 .arg5_type = ARG_ANYTHING, 7470 }; 7471 7472 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 7473 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7474 { 7475 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 7476 netns_id, flags); 7477 } 7478 7479 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 7480 .func = bpf_sk_lookup_udp, 7481 .gpl_only = false, 7482 .pkt_access = true, 7483 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7484 .arg1_type = ARG_PTR_TO_CTX, 7485 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7486 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7487 .arg4_type = ARG_ANYTHING, 7488 .arg5_type = ARG_ANYTHING, 7489 }; 7490 7491 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb, 7492 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7493 { 7494 struct net_device *dev = skb->dev; 7495 int ifindex = dev->ifindex, sdif = dev_sdif(dev); 7496 struct net *caller_net = dev_net(dev); 7497 7498 return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net, 7499 ifindex, IPPROTO_TCP, netns_id, 7500 flags, sdif); 7501 } 7502 7503 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = { 7504 .func = bpf_tc_skc_lookup_tcp, 7505 .gpl_only = false, 7506 .pkt_access = true, 7507 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 7508 .arg1_type = ARG_PTR_TO_CTX, 7509 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7510 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7511 .arg4_type = ARG_ANYTHING, 7512 .arg5_type = ARG_ANYTHING, 7513 }; 7514 7515 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb, 7516 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7517 { 7518 struct net_device *dev = skb->dev; 7519 int ifindex = dev->ifindex, sdif = dev_sdif(dev); 7520 struct net *caller_net = dev_net(dev); 7521 7522 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net, 7523 ifindex, IPPROTO_TCP, netns_id, 7524 flags, sdif); 7525 } 7526 7527 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = { 7528 .func = bpf_tc_sk_lookup_tcp, 7529 .gpl_only = false, 7530 .pkt_access = true, 7531 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7532 .arg1_type = ARG_PTR_TO_CTX, 7533 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7534 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7535 .arg4_type = ARG_ANYTHING, 7536 .arg5_type = ARG_ANYTHING, 7537 }; 7538 7539 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb, 7540 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7541 { 7542 struct net_device *dev = skb->dev; 7543 int ifindex = dev->ifindex, sdif = dev_sdif(dev); 7544 struct net *caller_net = dev_net(dev); 7545 7546 return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net, 7547 ifindex, IPPROTO_UDP, netns_id, 7548 flags, sdif); 7549 } 7550 7551 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = { 7552 .func = bpf_tc_sk_lookup_udp, 7553 .gpl_only = false, 7554 .pkt_access = true, 7555 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7556 .arg1_type = ARG_PTR_TO_CTX, 7557 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7558 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7559 .arg4_type = ARG_ANYTHING, 7560 .arg5_type = ARG_ANYTHING, 7561 }; 7562 7563 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 7564 { 7565 if (sk && sk_is_refcounted(sk)) 7566 sock_gen_put(sk); 7567 return 0; 7568 } 7569 7570 static const struct bpf_func_proto bpf_sk_release_proto = { 7571 .func = bpf_sk_release, 7572 .gpl_only = false, 7573 .ret_type = RET_INTEGER, 7574 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE, 7575 }; 7576 7577 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 7578 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 7579 { 7580 struct net_device *dev = ctx->rxq->dev; 7581 int ifindex = dev->ifindex, sdif = dev_sdif(dev); 7582 struct net *caller_net = dev_net(dev); 7583 7584 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 7585 ifindex, IPPROTO_UDP, netns_id, 7586 flags, sdif); 7587 } 7588 7589 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 7590 .func = bpf_xdp_sk_lookup_udp, 7591 .gpl_only = false, 7592 .pkt_access = true, 7593 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7594 .arg1_type = ARG_PTR_TO_CTX, 7595 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7596 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7597 .arg4_type = ARG_ANYTHING, 7598 .arg5_type = ARG_ANYTHING, 7599 }; 7600 7601 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 7602 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 7603 { 7604 struct net_device *dev = ctx->rxq->dev; 7605 int ifindex = dev->ifindex, sdif = dev_sdif(dev); 7606 struct net *caller_net = dev_net(dev); 7607 7608 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 7609 ifindex, IPPROTO_TCP, netns_id, 7610 flags, sdif); 7611 } 7612 7613 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 7614 .func = bpf_xdp_skc_lookup_tcp, 7615 .gpl_only = false, 7616 .pkt_access = true, 7617 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 7618 .arg1_type = ARG_PTR_TO_CTX, 7619 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7620 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7621 .arg4_type = ARG_ANYTHING, 7622 .arg5_type = ARG_ANYTHING, 7623 }; 7624 7625 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 7626 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 7627 { 7628 struct net_device *dev = ctx->rxq->dev; 7629 int ifindex = dev->ifindex, sdif = dev_sdif(dev); 7630 struct net *caller_net = dev_net(dev); 7631 7632 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 7633 ifindex, IPPROTO_TCP, netns_id, 7634 flags, sdif); 7635 } 7636 7637 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 7638 .func = bpf_xdp_sk_lookup_tcp, 7639 .gpl_only = false, 7640 .pkt_access = true, 7641 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7642 .arg1_type = ARG_PTR_TO_CTX, 7643 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7644 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7645 .arg4_type = ARG_ANYTHING, 7646 .arg5_type = ARG_ANYTHING, 7647 }; 7648 7649 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 7650 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7651 { 7652 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 7653 sock_net(ctx->sk), 0, 7654 IPPROTO_TCP, netns_id, flags, 7655 -1); 7656 } 7657 7658 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 7659 .func = bpf_sock_addr_skc_lookup_tcp, 7660 .gpl_only = false, 7661 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 7662 .arg1_type = ARG_PTR_TO_CTX, 7663 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7664 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7665 .arg4_type = ARG_ANYTHING, 7666 .arg5_type = ARG_ANYTHING, 7667 }; 7668 7669 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 7670 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7671 { 7672 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 7673 sock_net(ctx->sk), 0, IPPROTO_TCP, 7674 netns_id, flags, -1); 7675 } 7676 7677 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 7678 .func = bpf_sock_addr_sk_lookup_tcp, 7679 .gpl_only = false, 7680 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7681 .arg1_type = ARG_PTR_TO_CTX, 7682 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7683 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7684 .arg4_type = ARG_ANYTHING, 7685 .arg5_type = ARG_ANYTHING, 7686 }; 7687 7688 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 7689 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 7690 { 7691 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 7692 sock_net(ctx->sk), 0, IPPROTO_UDP, 7693 netns_id, flags, -1); 7694 } 7695 7696 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 7697 .func = bpf_sock_addr_sk_lookup_udp, 7698 .gpl_only = false, 7699 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7700 .arg1_type = ARG_PTR_TO_CTX, 7701 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7702 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 7703 .arg4_type = ARG_ANYTHING, 7704 .arg5_type = ARG_ANYTHING, 7705 }; 7706 7707 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7708 struct bpf_insn_access_aux *info) 7709 { 7710 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 7711 icsk_retransmits)) 7712 return false; 7713 7714 if (off % size != 0) 7715 return false; 7716 7717 switch (off) { 7718 case bpf_ctx_range(struct bpf_tcp_sock, bytes_received): 7719 case bpf_ctx_range(struct bpf_tcp_sock, bytes_acked): 7720 return size == sizeof(__u64); 7721 default: 7722 return size == sizeof(__u32); 7723 } 7724 } 7725 7726 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 7727 const struct bpf_insn *si, 7728 struct bpf_insn *insn_buf, 7729 struct bpf_prog *prog, u32 *target_size) 7730 { 7731 struct bpf_insn *insn = insn_buf; 7732 7733 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 7734 do { \ 7735 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 7736 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 7737 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 7738 si->dst_reg, si->src_reg, \ 7739 offsetof(struct tcp_sock, FIELD)); \ 7740 } while (0) 7741 7742 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 7743 do { \ 7744 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 7745 FIELD) > \ 7746 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 7747 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 7748 struct inet_connection_sock, \ 7749 FIELD), \ 7750 si->dst_reg, si->src_reg, \ 7751 offsetof( \ 7752 struct inet_connection_sock, \ 7753 FIELD)); \ 7754 } while (0) 7755 7756 BTF_TYPE_EMIT(struct bpf_tcp_sock); 7757 7758 switch (si->off) { 7759 case offsetof(struct bpf_tcp_sock, rtt_min): 7760 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 7761 sizeof(struct minmax)); 7762 BUILD_BUG_ON(sizeof(struct minmax) < 7763 sizeof(struct minmax_sample)); 7764 7765 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7766 offsetof(struct tcp_sock, rtt_min) + 7767 offsetof(struct minmax_sample, v)); 7768 break; 7769 case offsetof(struct bpf_tcp_sock, snd_cwnd): 7770 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 7771 break; 7772 case offsetof(struct bpf_tcp_sock, srtt_us): 7773 BPF_TCP_SOCK_GET_COMMON(srtt_us); 7774 break; 7775 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 7776 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 7777 break; 7778 case offsetof(struct bpf_tcp_sock, rcv_nxt): 7779 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 7780 break; 7781 case offsetof(struct bpf_tcp_sock, snd_nxt): 7782 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 7783 break; 7784 case offsetof(struct bpf_tcp_sock, snd_una): 7785 BPF_TCP_SOCK_GET_COMMON(snd_una); 7786 break; 7787 case offsetof(struct bpf_tcp_sock, mss_cache): 7788 BPF_TCP_SOCK_GET_COMMON(mss_cache); 7789 break; 7790 case offsetof(struct bpf_tcp_sock, ecn_flags): 7791 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 7792 break; 7793 case offsetof(struct bpf_tcp_sock, rate_delivered): 7794 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 7795 break; 7796 case offsetof(struct bpf_tcp_sock, rate_interval_us): 7797 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 7798 break; 7799 case offsetof(struct bpf_tcp_sock, packets_out): 7800 BPF_TCP_SOCK_GET_COMMON(packets_out); 7801 break; 7802 case offsetof(struct bpf_tcp_sock, retrans_out): 7803 BPF_TCP_SOCK_GET_COMMON(retrans_out); 7804 break; 7805 case offsetof(struct bpf_tcp_sock, total_retrans): 7806 BPF_TCP_SOCK_GET_COMMON(total_retrans); 7807 break; 7808 case offsetof(struct bpf_tcp_sock, segs_in): 7809 BPF_TCP_SOCK_GET_COMMON(segs_in); 7810 break; 7811 case offsetof(struct bpf_tcp_sock, data_segs_in): 7812 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 7813 break; 7814 case offsetof(struct bpf_tcp_sock, segs_out): 7815 BPF_TCP_SOCK_GET_COMMON(segs_out); 7816 break; 7817 case offsetof(struct bpf_tcp_sock, data_segs_out): 7818 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 7819 break; 7820 case offsetof(struct bpf_tcp_sock, lost_out): 7821 BPF_TCP_SOCK_GET_COMMON(lost_out); 7822 break; 7823 case offsetof(struct bpf_tcp_sock, sacked_out): 7824 BPF_TCP_SOCK_GET_COMMON(sacked_out); 7825 break; 7826 case offsetof(struct bpf_tcp_sock, bytes_received): 7827 BPF_TCP_SOCK_GET_COMMON(bytes_received); 7828 break; 7829 case offsetof(struct bpf_tcp_sock, bytes_acked): 7830 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 7831 break; 7832 case offsetof(struct bpf_tcp_sock, dsack_dups): 7833 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 7834 break; 7835 case offsetof(struct bpf_tcp_sock, delivered): 7836 BPF_TCP_SOCK_GET_COMMON(delivered); 7837 break; 7838 case offsetof(struct bpf_tcp_sock, delivered_ce): 7839 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 7840 break; 7841 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 7842 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 7843 break; 7844 } 7845 7846 return insn - insn_buf; 7847 } 7848 7849 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 7850 { 7851 if (sk_fullsock(sk) && sk_is_tcp(sk)) 7852 return (unsigned long)sk; 7853 7854 return (unsigned long)NULL; 7855 } 7856 7857 const struct bpf_func_proto bpf_tcp_sock_proto = { 7858 .func = bpf_tcp_sock, 7859 .gpl_only = false, 7860 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 7861 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 7862 }; 7863 7864 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 7865 { 7866 sk = sk_to_full_sk(sk); 7867 7868 if (sk && sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 7869 return (unsigned long)sk; 7870 7871 return (unsigned long)NULL; 7872 } 7873 7874 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 7875 .func = bpf_get_listener_sock, 7876 .gpl_only = false, 7877 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7878 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 7879 }; 7880 7881 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 7882 { 7883 unsigned int iphdr_len; 7884 7885 switch (skb_protocol(skb, true)) { 7886 case cpu_to_be16(ETH_P_IP): 7887 iphdr_len = sizeof(struct iphdr); 7888 break; 7889 case cpu_to_be16(ETH_P_IPV6): 7890 iphdr_len = sizeof(struct ipv6hdr); 7891 break; 7892 default: 7893 return 0; 7894 } 7895 7896 if (skb_headlen(skb) < iphdr_len) 7897 return 0; 7898 7899 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 7900 return 0; 7901 7902 return INET_ECN_set_ce(skb); 7903 } 7904 7905 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7906 struct bpf_insn_access_aux *info) 7907 { 7908 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 7909 return false; 7910 7911 if (off % size != 0) 7912 return false; 7913 7914 switch (off) { 7915 default: 7916 return size == sizeof(__u32); 7917 } 7918 } 7919 7920 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 7921 const struct bpf_insn *si, 7922 struct bpf_insn *insn_buf, 7923 struct bpf_prog *prog, u32 *target_size) 7924 { 7925 struct bpf_insn *insn = insn_buf; 7926 7927 #define BPF_XDP_SOCK_GET(FIELD) \ 7928 do { \ 7929 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 7930 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 7931 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 7932 si->dst_reg, si->src_reg, \ 7933 offsetof(struct xdp_sock, FIELD)); \ 7934 } while (0) 7935 7936 BTF_TYPE_EMIT(struct bpf_xdp_sock); 7937 7938 switch (si->off) { 7939 case offsetof(struct bpf_xdp_sock, queue_id): 7940 BPF_XDP_SOCK_GET(queue_id); 7941 break; 7942 } 7943 7944 return insn - insn_buf; 7945 } 7946 7947 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 7948 .func = bpf_skb_ecn_set_ce, 7949 .gpl_only = false, 7950 .ret_type = RET_INTEGER, 7951 .arg1_type = ARG_PTR_TO_CTX, 7952 }; 7953 7954 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7955 struct tcphdr *, th, u32, th_len) 7956 { 7957 #ifdef CONFIG_SYN_COOKIES 7958 int ret; 7959 7960 if (unlikely(!sk || th_len < sizeof(*th))) 7961 return -EINVAL; 7962 7963 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 7964 if (sk->sk_state != TCP_LISTEN || sk->sk_protocol != IPPROTO_TCP) 7965 return -EINVAL; 7966 7967 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7968 return -EINVAL; 7969 7970 if (!th->ack || th->rst || th->syn) 7971 return -ENOENT; 7972 7973 if (unlikely(iph_len < sizeof(struct iphdr))) 7974 return -EINVAL; 7975 7976 if (tcp_synq_no_recent_overflow(sk)) 7977 return -ENOENT; 7978 7979 /* Both struct iphdr and struct ipv6hdr have the version field at the 7980 * same offset so we can cast to the shorter header (struct iphdr). 7981 */ 7982 switch (((struct iphdr *)iph)->version) { 7983 case 4: 7984 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7985 return -EINVAL; 7986 7987 ret = __cookie_v4_check((struct iphdr *)iph, th); 7988 break; 7989 7990 #if IS_ENABLED(CONFIG_IPV6) 7991 case 6: 7992 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7993 return -EINVAL; 7994 7995 if (sk->sk_family != AF_INET6) 7996 return -EINVAL; 7997 7998 ret = __cookie_v6_check((struct ipv6hdr *)iph, th); 7999 break; 8000 #endif /* CONFIG_IPV6 */ 8001 8002 default: 8003 return -EPROTONOSUPPORT; 8004 } 8005 8006 if (ret > 0) 8007 return 0; 8008 8009 return -ENOENT; 8010 #else 8011 return -ENOTSUPP; 8012 #endif 8013 } 8014 8015 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 8016 .func = bpf_tcp_check_syncookie, 8017 .gpl_only = true, 8018 .pkt_access = true, 8019 .ret_type = RET_INTEGER, 8020 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 8021 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 8022 .arg3_type = ARG_MEM_SIZE, 8023 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 8024 .arg5_type = ARG_MEM_SIZE, 8025 }; 8026 8027 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 8028 struct tcphdr *, th, u32, th_len) 8029 { 8030 #ifdef CONFIG_SYN_COOKIES 8031 u32 cookie; 8032 u16 mss; 8033 8034 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 8035 return -EINVAL; 8036 8037 if (sk->sk_state != TCP_LISTEN || sk->sk_protocol != IPPROTO_TCP) 8038 return -EINVAL; 8039 8040 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 8041 return -ENOENT; 8042 8043 if (!th->syn || th->ack || th->fin || th->rst) 8044 return -EINVAL; 8045 8046 if (unlikely(iph_len < sizeof(struct iphdr))) 8047 return -EINVAL; 8048 8049 /* Both struct iphdr and struct ipv6hdr have the version field at the 8050 * same offset so we can cast to the shorter header (struct iphdr). 8051 */ 8052 switch (((struct iphdr *)iph)->version) { 8053 case 4: 8054 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 8055 return -EINVAL; 8056 8057 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 8058 break; 8059 8060 #if IS_ENABLED(CONFIG_IPV6) 8061 case 6: 8062 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 8063 return -EINVAL; 8064 8065 if (sk->sk_family != AF_INET6) 8066 return -EINVAL; 8067 8068 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 8069 break; 8070 #endif /* CONFIG_IPV6 */ 8071 8072 default: 8073 return -EPROTONOSUPPORT; 8074 } 8075 if (mss == 0) 8076 return -ENOENT; 8077 8078 return cookie | ((u64)mss << 32); 8079 #else 8080 return -EOPNOTSUPP; 8081 #endif /* CONFIG_SYN_COOKIES */ 8082 } 8083 8084 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 8085 .func = bpf_tcp_gen_syncookie, 8086 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 8087 .pkt_access = true, 8088 .ret_type = RET_INTEGER, 8089 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 8090 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 8091 .arg3_type = ARG_MEM_SIZE, 8092 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 8093 .arg5_type = ARG_MEM_SIZE, 8094 }; 8095 8096 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 8097 { 8098 if (!sk || flags != 0) 8099 return -EINVAL; 8100 if (!skb_at_tc_ingress(skb)) 8101 return -EOPNOTSUPP; 8102 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 8103 return -ENETUNREACH; 8104 if (sk_unhashed(sk)) 8105 return -EOPNOTSUPP; 8106 if (sk_is_refcounted(sk) && 8107 unlikely(!refcount_inc_not_zero_acquire(&sk->sk_refcnt))) 8108 return -ENOENT; 8109 8110 skb_orphan(skb); 8111 skb->sk = sk; 8112 skb->destructor = sock_pfree; 8113 8114 return 0; 8115 } 8116 8117 static const struct bpf_func_proto bpf_sk_assign_proto = { 8118 .func = bpf_sk_assign, 8119 .gpl_only = false, 8120 .ret_type = RET_INTEGER, 8121 .arg1_type = ARG_PTR_TO_CTX, 8122 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 8123 .arg3_type = ARG_ANYTHING, 8124 }; 8125 8126 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 8127 u8 search_kind, const u8 *magic, 8128 u8 magic_len, bool *eol) 8129 { 8130 u8 kind, kind_len; 8131 8132 *eol = false; 8133 8134 while (op < opend) { 8135 kind = op[0]; 8136 8137 if (kind == TCPOPT_EOL) { 8138 *eol = true; 8139 return ERR_PTR(-ENOMSG); 8140 } else if (kind == TCPOPT_NOP) { 8141 op++; 8142 continue; 8143 } 8144 8145 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 8146 /* Something is wrong in the received header. 8147 * Follow the TCP stack's tcp_parse_options() 8148 * and just bail here. 8149 */ 8150 return ERR_PTR(-EFAULT); 8151 8152 kind_len = op[1]; 8153 if (search_kind == kind) { 8154 if (!magic_len) 8155 return op; 8156 8157 if (magic_len > kind_len - 2) 8158 return ERR_PTR(-ENOMSG); 8159 8160 if (!memcmp(&op[2], magic, magic_len)) 8161 return op; 8162 } 8163 8164 op += kind_len; 8165 } 8166 8167 return ERR_PTR(-ENOMSG); 8168 } 8169 8170 int __bpf_sock_ops_load_hdr_opt(struct bpf_sock_ops_kern *bpf_sock, 8171 void *search_res, u32 len, u64 flags) 8172 { 8173 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 8174 const u8 *op, *opend, *magic, *search = search_res; 8175 u8 search_kind, search_len, copy_len, magic_len; 8176 int ret; 8177 8178 /* 2 byte is the minimal option len except TCPOPT_NOP and 8179 * TCPOPT_EOL which are useless for the bpf prog to learn 8180 * and this helper disallow loading them also. 8181 */ 8182 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 8183 return -EINVAL; 8184 8185 search_kind = search[0]; 8186 search_len = search[1]; 8187 8188 if (search_len > len || search_kind == TCPOPT_NOP || 8189 search_kind == TCPOPT_EOL) 8190 return -EINVAL; 8191 8192 if (search_kind == TCPOPT_EXP || search_kind == 253) { 8193 /* 16 or 32 bit magic. +2 for kind and kind length */ 8194 if (search_len != 4 && search_len != 6) 8195 return -EINVAL; 8196 magic = &search[2]; 8197 magic_len = search_len - 2; 8198 } else { 8199 if (search_len) 8200 return -EINVAL; 8201 magic = NULL; 8202 magic_len = 0; 8203 } 8204 8205 if (load_syn) { 8206 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 8207 if (ret < 0) 8208 return ret; 8209 8210 opend = op + ret; 8211 op += sizeof(struct tcphdr); 8212 } else { 8213 if (!bpf_sock->skb || 8214 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 8215 /* This bpf_sock->op cannot call this helper */ 8216 return -EPERM; 8217 8218 opend = bpf_sock->skb_data_end; 8219 op = bpf_sock->skb->data + sizeof(struct tcphdr); 8220 } 8221 8222 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 8223 &eol); 8224 if (IS_ERR(op)) 8225 return PTR_ERR(op); 8226 8227 copy_len = op[1]; 8228 ret = copy_len; 8229 if (copy_len > len) { 8230 ret = -ENOSPC; 8231 copy_len = len; 8232 } 8233 8234 memcpy(search_res, op, copy_len); 8235 return ret; 8236 } 8237 8238 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 8239 void *, search_res, u32, len, u64, flags) 8240 { 8241 if (!is_locked_tcp_sock_ops(bpf_sock)) 8242 return -EOPNOTSUPP; 8243 8244 return __bpf_sock_ops_load_hdr_opt(bpf_sock, search_res, len, flags); 8245 } 8246 8247 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 8248 .func = bpf_sock_ops_load_hdr_opt, 8249 .gpl_only = false, 8250 .ret_type = RET_INTEGER, 8251 .arg1_type = ARG_PTR_TO_CTX, 8252 .arg2_type = ARG_PTR_TO_MEM | MEM_WRITE, 8253 .arg3_type = ARG_MEM_SIZE, 8254 .arg4_type = ARG_ANYTHING, 8255 }; 8256 8257 int __bpf_sock_ops_store_hdr_opt(struct bpf_sock_ops_kern *bpf_sock, 8258 const void *from, u32 len, u64 flags) 8259 { 8260 u8 new_kind, new_kind_len, magic_len = 0, *opend; 8261 const u8 *op, *new_op, *magic = NULL; 8262 struct sk_buff *skb; 8263 bool eol; 8264 8265 if (len < 2 || flags) 8266 return -EINVAL; 8267 8268 new_op = from; 8269 new_kind = new_op[0]; 8270 new_kind_len = new_op[1]; 8271 8272 if (new_kind_len > len || new_kind == TCPOPT_NOP || 8273 new_kind == TCPOPT_EOL) 8274 return -EINVAL; 8275 8276 if (new_kind_len > bpf_sock->remaining_opt_len) 8277 return -ENOSPC; 8278 8279 /* 253 is another experimental kind */ 8280 if (new_kind == TCPOPT_EXP || new_kind == 253) { 8281 if (new_kind_len < 4) 8282 return -EINVAL; 8283 /* Match for the 2 byte magic also. 8284 * RFC 6994: the magic could be 2 or 4 bytes. 8285 * Hence, matching by 2 byte only is on the 8286 * conservative side but it is the right 8287 * thing to do for the 'search-for-duplication' 8288 * purpose. 8289 */ 8290 magic = &new_op[2]; 8291 magic_len = 2; 8292 } 8293 8294 /* Check for duplication */ 8295 skb = bpf_sock->skb; 8296 op = skb->data + sizeof(struct tcphdr); 8297 opend = bpf_sock->skb_data_end; 8298 8299 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 8300 &eol); 8301 if (!IS_ERR(op)) 8302 return -EEXIST; 8303 8304 if (PTR_ERR(op) != -ENOMSG) 8305 return PTR_ERR(op); 8306 8307 if (eol) 8308 /* The option has been ended. Treat it as no more 8309 * header option can be written. 8310 */ 8311 return -ENOSPC; 8312 8313 /* No duplication found. Store the header option. */ 8314 memcpy(opend, from, new_kind_len); 8315 8316 bpf_sock->remaining_opt_len -= new_kind_len; 8317 bpf_sock->skb_data_end += new_kind_len; 8318 8319 return 0; 8320 } 8321 8322 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 8323 const void *, from, u32, len, u64, flags) 8324 { 8325 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 8326 return -EPERM; 8327 8328 return __bpf_sock_ops_store_hdr_opt(bpf_sock, from, len, flags); 8329 } 8330 8331 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 8332 .func = bpf_sock_ops_store_hdr_opt, 8333 .gpl_only = false, 8334 .ret_type = RET_INTEGER, 8335 .arg1_type = ARG_PTR_TO_CTX, 8336 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 8337 .arg3_type = ARG_MEM_SIZE, 8338 .arg4_type = ARG_ANYTHING, 8339 }; 8340 8341 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 8342 u32, len, u64, flags) 8343 { 8344 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 8345 return -EPERM; 8346 8347 if (flags || len < 2) 8348 return -EINVAL; 8349 8350 if (len > bpf_sock->remaining_opt_len) 8351 return -ENOSPC; 8352 8353 bpf_sock->remaining_opt_len -= len; 8354 8355 return 0; 8356 } 8357 8358 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 8359 .func = bpf_sock_ops_reserve_hdr_opt, 8360 .gpl_only = false, 8361 .ret_type = RET_INTEGER, 8362 .arg1_type = ARG_PTR_TO_CTX, 8363 .arg2_type = ARG_ANYTHING, 8364 .arg3_type = ARG_ANYTHING, 8365 }; 8366 8367 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, 8368 u64, tstamp, u32, tstamp_type) 8369 { 8370 /* skb_clear_delivery_time() is done for inet protocol */ 8371 if (skb->protocol != htons(ETH_P_IP) && 8372 skb->protocol != htons(ETH_P_IPV6)) 8373 return -EOPNOTSUPP; 8374 8375 switch (tstamp_type) { 8376 case BPF_SKB_CLOCK_REALTIME: 8377 skb->tstamp = tstamp; 8378 skb->tstamp_type = SKB_CLOCK_REALTIME; 8379 break; 8380 case BPF_SKB_CLOCK_MONOTONIC: 8381 if (!tstamp) 8382 return -EINVAL; 8383 skb->tstamp = tstamp; 8384 skb->tstamp_type = SKB_CLOCK_MONOTONIC; 8385 break; 8386 case BPF_SKB_CLOCK_TAI: 8387 if (!tstamp) 8388 return -EINVAL; 8389 skb->tstamp = tstamp; 8390 skb->tstamp_type = SKB_CLOCK_TAI; 8391 break; 8392 default: 8393 return -EINVAL; 8394 } 8395 8396 return 0; 8397 } 8398 8399 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = { 8400 .func = bpf_skb_set_tstamp, 8401 .gpl_only = false, 8402 .ret_type = RET_INTEGER, 8403 .arg1_type = ARG_PTR_TO_CTX, 8404 .arg2_type = ARG_ANYTHING, 8405 .arg3_type = ARG_ANYTHING, 8406 }; 8407 8408 #ifdef CONFIG_SYN_COOKIES 8409 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph, 8410 struct tcphdr *, th, u32, th_len) 8411 { 8412 u32 cookie; 8413 u16 mss; 8414 8415 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 8416 return -EINVAL; 8417 8418 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT; 8419 cookie = __cookie_v4_init_sequence(iph, th, &mss); 8420 8421 return cookie | ((u64)mss << 32); 8422 } 8423 8424 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = { 8425 .func = bpf_tcp_raw_gen_syncookie_ipv4, 8426 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */ 8427 .pkt_access = true, 8428 .ret_type = RET_INTEGER, 8429 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_RDONLY, 8430 .arg1_size = sizeof(struct iphdr), 8431 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 8432 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 8433 }; 8434 8435 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph, 8436 struct tcphdr *, th, u32, th_len) 8437 { 8438 #if IS_ENABLED(CONFIG_IPV6) 8439 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 8440 sizeof(struct ipv6hdr); 8441 u32 cookie; 8442 u16 mss; 8443 8444 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 8445 return -EINVAL; 8446 8447 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp; 8448 cookie = __cookie_v6_init_sequence(iph, th, &mss); 8449 8450 return cookie | ((u64)mss << 32); 8451 #else 8452 return -EPROTONOSUPPORT; 8453 #endif 8454 } 8455 8456 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = { 8457 .func = bpf_tcp_raw_gen_syncookie_ipv6, 8458 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */ 8459 .pkt_access = true, 8460 .ret_type = RET_INTEGER, 8461 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_RDONLY, 8462 .arg1_size = sizeof(struct ipv6hdr), 8463 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 8464 .arg3_type = ARG_MEM_SIZE_OR_ZERO, 8465 }; 8466 8467 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph, 8468 struct tcphdr *, th) 8469 { 8470 if (__cookie_v4_check(iph, th) > 0) 8471 return 0; 8472 8473 return -EACCES; 8474 } 8475 8476 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = { 8477 .func = bpf_tcp_raw_check_syncookie_ipv4, 8478 .gpl_only = true, /* __cookie_v4_check is GPL */ 8479 .pkt_access = true, 8480 .ret_type = RET_INTEGER, 8481 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_RDONLY, 8482 .arg1_size = sizeof(struct iphdr), 8483 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_RDONLY, 8484 .arg2_size = sizeof(struct tcphdr), 8485 }; 8486 8487 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph, 8488 struct tcphdr *, th) 8489 { 8490 #if IS_ENABLED(CONFIG_IPV6) 8491 if (__cookie_v6_check(iph, th) > 0) 8492 return 0; 8493 8494 return -EACCES; 8495 #else 8496 return -EPROTONOSUPPORT; 8497 #endif 8498 } 8499 8500 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = { 8501 .func = bpf_tcp_raw_check_syncookie_ipv6, 8502 .gpl_only = true, /* __cookie_v6_check is GPL */ 8503 .pkt_access = true, 8504 .ret_type = RET_INTEGER, 8505 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_RDONLY, 8506 .arg1_size = sizeof(struct ipv6hdr), 8507 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_RDONLY, 8508 .arg2_size = sizeof(struct tcphdr), 8509 }; 8510 #endif /* CONFIG_SYN_COOKIES */ 8511 8512 #endif /* CONFIG_INET */ 8513 8514 bool bpf_helper_changes_pkt_data(enum bpf_func_id func_id) 8515 { 8516 switch (func_id) { 8517 case BPF_FUNC_clone_redirect: 8518 case BPF_FUNC_l3_csum_replace: 8519 case BPF_FUNC_l4_csum_replace: 8520 case BPF_FUNC_lwt_push_encap: 8521 case BPF_FUNC_lwt_seg6_action: 8522 case BPF_FUNC_lwt_seg6_adjust_srh: 8523 case BPF_FUNC_lwt_seg6_store_bytes: 8524 case BPF_FUNC_msg_pop_data: 8525 case BPF_FUNC_msg_pull_data: 8526 case BPF_FUNC_msg_push_data: 8527 case BPF_FUNC_skb_adjust_room: 8528 case BPF_FUNC_skb_change_head: 8529 case BPF_FUNC_skb_change_proto: 8530 case BPF_FUNC_skb_change_tail: 8531 case BPF_FUNC_skb_pull_data: 8532 case BPF_FUNC_skb_store_bytes: 8533 case BPF_FUNC_skb_vlan_pop: 8534 case BPF_FUNC_skb_vlan_push: 8535 case BPF_FUNC_store_hdr_opt: 8536 case BPF_FUNC_xdp_adjust_head: 8537 case BPF_FUNC_xdp_adjust_meta: 8538 case BPF_FUNC_xdp_adjust_tail: 8539 /* tail-called program could call any of the above */ 8540 case BPF_FUNC_tail_call: 8541 return true; 8542 default: 8543 return false; 8544 } 8545 } 8546 8547 const struct bpf_func_proto bpf_event_output_data_proto __weak; 8548 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 8549 8550 static const struct bpf_func_proto * 8551 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8552 { 8553 const struct bpf_func_proto *func_proto; 8554 8555 func_proto = cgroup_common_func_proto(func_id, prog); 8556 if (func_proto) 8557 return func_proto; 8558 8559 switch (func_id) { 8560 case BPF_FUNC_get_socket_cookie: 8561 return &bpf_get_socket_cookie_sock_proto; 8562 case BPF_FUNC_get_netns_cookie: 8563 return &bpf_get_netns_cookie_sock_proto; 8564 case BPF_FUNC_perf_event_output: 8565 return &bpf_event_output_data_proto; 8566 case BPF_FUNC_sk_storage_get: 8567 return &bpf_sk_storage_get_cg_sock_proto; 8568 case BPF_FUNC_ktime_get_coarse_ns: 8569 return &bpf_ktime_get_coarse_ns_proto; 8570 case BPF_FUNC_setsockopt: 8571 switch (prog->expected_attach_type) { 8572 case BPF_CGROUP_INET_SOCK_CREATE: 8573 return &bpf_sock_create_setsockopt_proto; 8574 default: 8575 return NULL; 8576 } 8577 case BPF_FUNC_getsockopt: 8578 switch (prog->expected_attach_type) { 8579 case BPF_CGROUP_INET_SOCK_CREATE: 8580 return &bpf_sock_create_getsockopt_proto; 8581 default: 8582 return NULL; 8583 } 8584 default: 8585 return bpf_base_func_proto(func_id, prog); 8586 } 8587 } 8588 8589 static const struct bpf_func_proto * 8590 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8591 { 8592 const struct bpf_func_proto *func_proto; 8593 8594 func_proto = cgroup_common_func_proto(func_id, prog); 8595 if (func_proto) 8596 return func_proto; 8597 8598 switch (func_id) { 8599 case BPF_FUNC_bind: 8600 switch (prog->expected_attach_type) { 8601 case BPF_CGROUP_INET4_CONNECT: 8602 case BPF_CGROUP_INET6_CONNECT: 8603 return &bpf_bind_proto; 8604 default: 8605 return NULL; 8606 } 8607 case BPF_FUNC_get_socket_cookie: 8608 return &bpf_get_socket_cookie_sock_addr_proto; 8609 case BPF_FUNC_get_netns_cookie: 8610 return &bpf_get_netns_cookie_sock_addr_proto; 8611 case BPF_FUNC_perf_event_output: 8612 return &bpf_event_output_data_proto; 8613 #ifdef CONFIG_INET 8614 case BPF_FUNC_sk_lookup_tcp: 8615 return &bpf_sock_addr_sk_lookup_tcp_proto; 8616 case BPF_FUNC_sk_lookup_udp: 8617 return &bpf_sock_addr_sk_lookup_udp_proto; 8618 case BPF_FUNC_sk_release: 8619 return &bpf_sk_release_proto; 8620 case BPF_FUNC_skc_lookup_tcp: 8621 return &bpf_sock_addr_skc_lookup_tcp_proto; 8622 #endif /* CONFIG_INET */ 8623 case BPF_FUNC_sk_storage_get: 8624 return &bpf_sk_storage_get_proto; 8625 case BPF_FUNC_sk_storage_delete: 8626 return &bpf_sk_storage_delete_proto; 8627 case BPF_FUNC_setsockopt: 8628 switch (prog->expected_attach_type) { 8629 case BPF_CGROUP_INET4_BIND: 8630 case BPF_CGROUP_INET6_BIND: 8631 case BPF_CGROUP_INET4_CONNECT: 8632 case BPF_CGROUP_INET6_CONNECT: 8633 case BPF_CGROUP_UNIX_CONNECT: 8634 case BPF_CGROUP_UDP4_RECVMSG: 8635 case BPF_CGROUP_UDP6_RECVMSG: 8636 case BPF_CGROUP_UNIX_RECVMSG: 8637 case BPF_CGROUP_UDP4_SENDMSG: 8638 case BPF_CGROUP_UDP6_SENDMSG: 8639 case BPF_CGROUP_UNIX_SENDMSG: 8640 case BPF_CGROUP_INET4_GETPEERNAME: 8641 case BPF_CGROUP_INET6_GETPEERNAME: 8642 case BPF_CGROUP_INET4_GETSOCKNAME: 8643 case BPF_CGROUP_INET6_GETSOCKNAME: 8644 return &bpf_sock_addr_setsockopt_proto; 8645 default: 8646 return NULL; 8647 } 8648 case BPF_FUNC_getsockopt: 8649 switch (prog->expected_attach_type) { 8650 case BPF_CGROUP_INET4_BIND: 8651 case BPF_CGROUP_INET6_BIND: 8652 case BPF_CGROUP_INET4_CONNECT: 8653 case BPF_CGROUP_INET6_CONNECT: 8654 case BPF_CGROUP_UNIX_CONNECT: 8655 case BPF_CGROUP_UDP4_RECVMSG: 8656 case BPF_CGROUP_UDP6_RECVMSG: 8657 case BPF_CGROUP_UNIX_RECVMSG: 8658 case BPF_CGROUP_UDP4_SENDMSG: 8659 case BPF_CGROUP_UDP6_SENDMSG: 8660 case BPF_CGROUP_UNIX_SENDMSG: 8661 case BPF_CGROUP_INET4_GETPEERNAME: 8662 case BPF_CGROUP_INET6_GETPEERNAME: 8663 case BPF_CGROUP_INET4_GETSOCKNAME: 8664 case BPF_CGROUP_INET6_GETSOCKNAME: 8665 return &bpf_sock_addr_getsockopt_proto; 8666 default: 8667 return NULL; 8668 } 8669 default: 8670 return bpf_sk_base_func_proto(func_id, prog); 8671 } 8672 } 8673 8674 static const struct bpf_func_proto * 8675 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8676 { 8677 switch (func_id) { 8678 case BPF_FUNC_skb_load_bytes: 8679 return &bpf_skb_load_bytes_proto; 8680 case BPF_FUNC_skb_load_bytes_relative: 8681 return &bpf_skb_load_bytes_relative_proto; 8682 case BPF_FUNC_get_socket_cookie: 8683 return &bpf_get_socket_cookie_proto; 8684 case BPF_FUNC_get_netns_cookie: 8685 return &bpf_get_netns_cookie_proto; 8686 case BPF_FUNC_get_socket_uid: 8687 return &bpf_get_socket_uid_proto; 8688 case BPF_FUNC_perf_event_output: 8689 return &bpf_skb_event_output_proto; 8690 default: 8691 return bpf_sk_base_func_proto(func_id, prog); 8692 } 8693 } 8694 8695 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 8696 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 8697 8698 static const struct bpf_func_proto * 8699 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8700 { 8701 const struct bpf_func_proto *func_proto; 8702 8703 func_proto = cgroup_common_func_proto(func_id, prog); 8704 if (func_proto) 8705 return func_proto; 8706 8707 switch (func_id) { 8708 case BPF_FUNC_sk_fullsock: 8709 return &bpf_sk_fullsock_proto; 8710 case BPF_FUNC_sk_storage_get: 8711 return &bpf_sk_storage_get_proto; 8712 case BPF_FUNC_sk_storage_delete: 8713 return &bpf_sk_storage_delete_proto; 8714 case BPF_FUNC_perf_event_output: 8715 return &bpf_skb_event_output_proto; 8716 #ifdef CONFIG_SOCK_CGROUP_DATA 8717 case BPF_FUNC_skb_cgroup_id: 8718 return &bpf_skb_cgroup_id_proto; 8719 case BPF_FUNC_skb_ancestor_cgroup_id: 8720 return &bpf_skb_ancestor_cgroup_id_proto; 8721 case BPF_FUNC_sk_cgroup_id: 8722 return &bpf_sk_cgroup_id_proto; 8723 case BPF_FUNC_sk_ancestor_cgroup_id: 8724 return &bpf_sk_ancestor_cgroup_id_proto; 8725 #endif 8726 #ifdef CONFIG_INET 8727 case BPF_FUNC_sk_lookup_tcp: 8728 return &bpf_sk_lookup_tcp_proto; 8729 case BPF_FUNC_sk_lookup_udp: 8730 return &bpf_sk_lookup_udp_proto; 8731 case BPF_FUNC_sk_release: 8732 return &bpf_sk_release_proto; 8733 case BPF_FUNC_skc_lookup_tcp: 8734 return &bpf_skc_lookup_tcp_proto; 8735 case BPF_FUNC_tcp_sock: 8736 return &bpf_tcp_sock_proto; 8737 case BPF_FUNC_get_listener_sock: 8738 return &bpf_get_listener_sock_proto; 8739 case BPF_FUNC_skb_ecn_set_ce: 8740 return &bpf_skb_ecn_set_ce_proto; 8741 #endif 8742 default: 8743 return sk_filter_func_proto(func_id, prog); 8744 } 8745 } 8746 8747 static const struct bpf_func_proto * 8748 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8749 { 8750 switch (func_id) { 8751 case BPF_FUNC_skb_store_bytes: 8752 return &bpf_skb_store_bytes_proto; 8753 case BPF_FUNC_skb_load_bytes: 8754 return &bpf_skb_load_bytes_proto; 8755 case BPF_FUNC_skb_load_bytes_relative: 8756 return &bpf_skb_load_bytes_relative_proto; 8757 case BPF_FUNC_skb_pull_data: 8758 return &bpf_skb_pull_data_proto; 8759 case BPF_FUNC_csum_diff: 8760 return &bpf_csum_diff_proto; 8761 case BPF_FUNC_csum_update: 8762 return &bpf_csum_update_proto; 8763 case BPF_FUNC_csum_level: 8764 return &bpf_csum_level_proto; 8765 case BPF_FUNC_l3_csum_replace: 8766 return &bpf_l3_csum_replace_proto; 8767 case BPF_FUNC_l4_csum_replace: 8768 return &bpf_l4_csum_replace_proto; 8769 case BPF_FUNC_clone_redirect: 8770 return &bpf_clone_redirect_proto; 8771 case BPF_FUNC_get_cgroup_classid: 8772 return &bpf_get_cgroup_classid_proto; 8773 case BPF_FUNC_skb_vlan_push: 8774 return &bpf_skb_vlan_push_proto; 8775 case BPF_FUNC_skb_vlan_pop: 8776 return &bpf_skb_vlan_pop_proto; 8777 case BPF_FUNC_skb_change_proto: 8778 return &bpf_skb_change_proto_proto; 8779 case BPF_FUNC_skb_change_type: 8780 return &bpf_skb_change_type_proto; 8781 case BPF_FUNC_skb_adjust_room: 8782 return &bpf_skb_adjust_room_proto; 8783 case BPF_FUNC_skb_change_tail: 8784 return &bpf_skb_change_tail_proto; 8785 case BPF_FUNC_skb_change_head: 8786 return &bpf_skb_change_head_proto; 8787 case BPF_FUNC_skb_get_tunnel_key: 8788 return &bpf_skb_get_tunnel_key_proto; 8789 case BPF_FUNC_skb_set_tunnel_key: 8790 return bpf_get_skb_set_tunnel_proto(func_id); 8791 case BPF_FUNC_skb_get_tunnel_opt: 8792 return &bpf_skb_get_tunnel_opt_proto; 8793 case BPF_FUNC_skb_set_tunnel_opt: 8794 return bpf_get_skb_set_tunnel_proto(func_id); 8795 case BPF_FUNC_redirect: 8796 return &bpf_redirect_proto; 8797 case BPF_FUNC_redirect_neigh: 8798 return &bpf_redirect_neigh_proto; 8799 case BPF_FUNC_redirect_peer: 8800 return &bpf_redirect_peer_proto; 8801 case BPF_FUNC_get_route_realm: 8802 return &bpf_get_route_realm_proto; 8803 case BPF_FUNC_get_hash_recalc: 8804 return &bpf_get_hash_recalc_proto; 8805 case BPF_FUNC_set_hash_invalid: 8806 return &bpf_set_hash_invalid_proto; 8807 case BPF_FUNC_set_hash: 8808 return &bpf_set_hash_proto; 8809 case BPF_FUNC_perf_event_output: 8810 return &bpf_skb_event_output_proto; 8811 case BPF_FUNC_get_smp_processor_id: 8812 return &bpf_get_smp_processor_id_proto; 8813 case BPF_FUNC_skb_under_cgroup: 8814 return &bpf_skb_under_cgroup_proto; 8815 case BPF_FUNC_get_socket_cookie: 8816 return &bpf_get_socket_cookie_proto; 8817 case BPF_FUNC_get_netns_cookie: 8818 return &bpf_get_netns_cookie_proto; 8819 case BPF_FUNC_get_socket_uid: 8820 return &bpf_get_socket_uid_proto; 8821 case BPF_FUNC_fib_lookup: 8822 return &bpf_skb_fib_lookup_proto; 8823 case BPF_FUNC_check_mtu: 8824 return &bpf_skb_check_mtu_proto; 8825 case BPF_FUNC_sk_fullsock: 8826 return &bpf_sk_fullsock_proto; 8827 case BPF_FUNC_sk_storage_get: 8828 return &bpf_sk_storage_get_proto; 8829 case BPF_FUNC_sk_storage_delete: 8830 return &bpf_sk_storage_delete_proto; 8831 #ifdef CONFIG_XFRM 8832 case BPF_FUNC_skb_get_xfrm_state: 8833 return &bpf_skb_get_xfrm_state_proto; 8834 #endif 8835 #ifdef CONFIG_CGROUP_NET_CLASSID 8836 case BPF_FUNC_skb_cgroup_classid: 8837 return &bpf_skb_cgroup_classid_proto; 8838 #endif 8839 #ifdef CONFIG_SOCK_CGROUP_DATA 8840 case BPF_FUNC_skb_cgroup_id: 8841 return &bpf_skb_cgroup_id_proto; 8842 case BPF_FUNC_skb_ancestor_cgroup_id: 8843 return &bpf_skb_ancestor_cgroup_id_proto; 8844 #endif 8845 #ifdef CONFIG_INET 8846 case BPF_FUNC_sk_lookup_tcp: 8847 return &bpf_tc_sk_lookup_tcp_proto; 8848 case BPF_FUNC_sk_lookup_udp: 8849 return &bpf_tc_sk_lookup_udp_proto; 8850 case BPF_FUNC_sk_release: 8851 return &bpf_sk_release_proto; 8852 case BPF_FUNC_tcp_sock: 8853 return &bpf_tcp_sock_proto; 8854 case BPF_FUNC_get_listener_sock: 8855 return &bpf_get_listener_sock_proto; 8856 case BPF_FUNC_skc_lookup_tcp: 8857 return &bpf_tc_skc_lookup_tcp_proto; 8858 case BPF_FUNC_tcp_check_syncookie: 8859 return &bpf_tcp_check_syncookie_proto; 8860 case BPF_FUNC_skb_ecn_set_ce: 8861 return &bpf_skb_ecn_set_ce_proto; 8862 case BPF_FUNC_tcp_gen_syncookie: 8863 return &bpf_tcp_gen_syncookie_proto; 8864 case BPF_FUNC_sk_assign: 8865 return &bpf_sk_assign_proto; 8866 case BPF_FUNC_skb_set_tstamp: 8867 return &bpf_skb_set_tstamp_proto; 8868 #ifdef CONFIG_SYN_COOKIES 8869 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 8870 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 8871 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 8872 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 8873 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 8874 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 8875 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 8876 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 8877 #endif 8878 #endif 8879 default: 8880 return bpf_sk_base_func_proto(func_id, prog); 8881 } 8882 } 8883 8884 static const struct bpf_func_proto * 8885 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8886 { 8887 switch (func_id) { 8888 case BPF_FUNC_perf_event_output: 8889 return &bpf_xdp_event_output_proto; 8890 case BPF_FUNC_get_smp_processor_id: 8891 return &bpf_get_smp_processor_id_proto; 8892 case BPF_FUNC_csum_diff: 8893 return &bpf_csum_diff_proto; 8894 case BPF_FUNC_xdp_adjust_head: 8895 return &bpf_xdp_adjust_head_proto; 8896 case BPF_FUNC_xdp_adjust_meta: 8897 return &bpf_xdp_adjust_meta_proto; 8898 case BPF_FUNC_redirect: 8899 return &bpf_xdp_redirect_proto; 8900 case BPF_FUNC_redirect_map: 8901 return &bpf_xdp_redirect_map_proto; 8902 case BPF_FUNC_xdp_adjust_tail: 8903 return &bpf_xdp_adjust_tail_proto; 8904 case BPF_FUNC_xdp_get_buff_len: 8905 return &bpf_xdp_get_buff_len_proto; 8906 case BPF_FUNC_xdp_load_bytes: 8907 return &bpf_xdp_load_bytes_proto; 8908 case BPF_FUNC_xdp_store_bytes: 8909 return &bpf_xdp_store_bytes_proto; 8910 case BPF_FUNC_fib_lookup: 8911 return &bpf_xdp_fib_lookup_proto; 8912 case BPF_FUNC_check_mtu: 8913 return &bpf_xdp_check_mtu_proto; 8914 #ifdef CONFIG_INET 8915 case BPF_FUNC_sk_lookup_udp: 8916 return &bpf_xdp_sk_lookup_udp_proto; 8917 case BPF_FUNC_sk_lookup_tcp: 8918 return &bpf_xdp_sk_lookup_tcp_proto; 8919 case BPF_FUNC_sk_release: 8920 return &bpf_sk_release_proto; 8921 case BPF_FUNC_skc_lookup_tcp: 8922 return &bpf_xdp_skc_lookup_tcp_proto; 8923 case BPF_FUNC_tcp_check_syncookie: 8924 return &bpf_tcp_check_syncookie_proto; 8925 case BPF_FUNC_tcp_gen_syncookie: 8926 return &bpf_tcp_gen_syncookie_proto; 8927 #ifdef CONFIG_SYN_COOKIES 8928 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 8929 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 8930 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 8931 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 8932 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 8933 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 8934 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 8935 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 8936 #endif 8937 #endif 8938 default: 8939 return bpf_sk_base_func_proto(func_id, prog); 8940 } 8941 8942 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES) 8943 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The 8944 * kfuncs are defined in two different modules, and we want to be able 8945 * to use them interchangeably with the same BTF type ID. Because modules 8946 * can't de-duplicate BTF IDs between each other, we need the type to be 8947 * referenced in the vmlinux BTF or the verifier will get confused about 8948 * the different types. So we add this dummy type reference which will 8949 * be included in vmlinux BTF, allowing both modules to refer to the 8950 * same type ID. 8951 */ 8952 BTF_TYPE_EMIT(struct nf_conn___init); 8953 #endif 8954 } 8955 8956 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 8957 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 8958 8959 static const struct bpf_func_proto * 8960 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8961 { 8962 const struct bpf_func_proto *func_proto; 8963 8964 func_proto = cgroup_common_func_proto(func_id, prog); 8965 if (func_proto) 8966 return func_proto; 8967 8968 switch (func_id) { 8969 case BPF_FUNC_setsockopt: 8970 return &bpf_sock_ops_setsockopt_proto; 8971 case BPF_FUNC_getsockopt: 8972 return &bpf_sock_ops_getsockopt_proto; 8973 case BPF_FUNC_sock_ops_cb_flags_set: 8974 return &bpf_sock_ops_cb_flags_set_proto; 8975 case BPF_FUNC_sock_map_update: 8976 return &bpf_sock_map_update_proto; 8977 case BPF_FUNC_sock_hash_update: 8978 return &bpf_sock_hash_update_proto; 8979 case BPF_FUNC_get_socket_cookie: 8980 return &bpf_get_socket_cookie_sock_ops_proto; 8981 case BPF_FUNC_perf_event_output: 8982 return &bpf_event_output_data_proto; 8983 case BPF_FUNC_sk_storage_get: 8984 return &bpf_sk_storage_get_proto; 8985 case BPF_FUNC_sk_storage_delete: 8986 return &bpf_sk_storage_delete_proto; 8987 case BPF_FUNC_get_netns_cookie: 8988 return &bpf_get_netns_cookie_sock_ops_proto; 8989 #ifdef CONFIG_INET 8990 case BPF_FUNC_load_hdr_opt: 8991 return &bpf_sock_ops_load_hdr_opt_proto; 8992 case BPF_FUNC_store_hdr_opt: 8993 return &bpf_sock_ops_store_hdr_opt_proto; 8994 case BPF_FUNC_reserve_hdr_opt: 8995 return &bpf_sock_ops_reserve_hdr_opt_proto; 8996 case BPF_FUNC_tcp_sock: 8997 return &bpf_tcp_sock_proto; 8998 #endif /* CONFIG_INET */ 8999 default: 9000 return bpf_sk_base_func_proto(func_id, prog); 9001 } 9002 } 9003 9004 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 9005 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 9006 9007 static const struct bpf_func_proto * 9008 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 9009 { 9010 switch (func_id) { 9011 case BPF_FUNC_msg_redirect_map: 9012 return &bpf_msg_redirect_map_proto; 9013 case BPF_FUNC_msg_redirect_hash: 9014 return &bpf_msg_redirect_hash_proto; 9015 case BPF_FUNC_msg_apply_bytes: 9016 return &bpf_msg_apply_bytes_proto; 9017 case BPF_FUNC_msg_cork_bytes: 9018 return &bpf_msg_cork_bytes_proto; 9019 case BPF_FUNC_msg_pull_data: 9020 return &bpf_msg_pull_data_proto; 9021 case BPF_FUNC_msg_push_data: 9022 return &bpf_msg_push_data_proto; 9023 case BPF_FUNC_msg_pop_data: 9024 return &bpf_msg_pop_data_proto; 9025 case BPF_FUNC_perf_event_output: 9026 return &bpf_event_output_data_proto; 9027 case BPF_FUNC_sk_storage_get: 9028 return &bpf_sk_storage_get_proto; 9029 case BPF_FUNC_sk_storage_delete: 9030 return &bpf_sk_storage_delete_proto; 9031 case BPF_FUNC_get_netns_cookie: 9032 return &bpf_get_netns_cookie_sk_msg_proto; 9033 default: 9034 return bpf_sk_base_func_proto(func_id, prog); 9035 } 9036 } 9037 9038 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 9039 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 9040 9041 static const struct bpf_func_proto * 9042 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 9043 { 9044 switch (func_id) { 9045 case BPF_FUNC_skb_store_bytes: 9046 return &bpf_skb_store_bytes_proto; 9047 case BPF_FUNC_skb_load_bytes: 9048 return &bpf_skb_load_bytes_proto; 9049 case BPF_FUNC_skb_pull_data: 9050 return &sk_skb_pull_data_proto; 9051 case BPF_FUNC_skb_change_tail: 9052 return &sk_skb_change_tail_proto; 9053 case BPF_FUNC_skb_change_head: 9054 return &sk_skb_change_head_proto; 9055 case BPF_FUNC_skb_adjust_room: 9056 return &sk_skb_adjust_room_proto; 9057 case BPF_FUNC_get_socket_cookie: 9058 return &bpf_get_socket_cookie_proto; 9059 case BPF_FUNC_get_socket_uid: 9060 return &bpf_get_socket_uid_proto; 9061 case BPF_FUNC_sk_redirect_map: 9062 return &bpf_sk_redirect_map_proto; 9063 case BPF_FUNC_sk_redirect_hash: 9064 return &bpf_sk_redirect_hash_proto; 9065 case BPF_FUNC_perf_event_output: 9066 return &bpf_skb_event_output_proto; 9067 #ifdef CONFIG_INET 9068 case BPF_FUNC_sk_lookup_tcp: 9069 return &bpf_sk_lookup_tcp_proto; 9070 case BPF_FUNC_sk_lookup_udp: 9071 return &bpf_sk_lookup_udp_proto; 9072 case BPF_FUNC_sk_release: 9073 return &bpf_sk_release_proto; 9074 case BPF_FUNC_skc_lookup_tcp: 9075 return &bpf_skc_lookup_tcp_proto; 9076 #endif 9077 default: 9078 return bpf_sk_base_func_proto(func_id, prog); 9079 } 9080 } 9081 9082 static const struct bpf_func_proto * 9083 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 9084 { 9085 switch (func_id) { 9086 case BPF_FUNC_skb_load_bytes: 9087 return &bpf_flow_dissector_load_bytes_proto; 9088 default: 9089 return bpf_sk_base_func_proto(func_id, prog); 9090 } 9091 } 9092 9093 static const struct bpf_func_proto * 9094 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 9095 { 9096 switch (func_id) { 9097 case BPF_FUNC_skb_load_bytes: 9098 return &bpf_skb_load_bytes_proto; 9099 case BPF_FUNC_skb_pull_data: 9100 return &bpf_skb_pull_data_proto; 9101 case BPF_FUNC_csum_diff: 9102 return &bpf_csum_diff_proto; 9103 case BPF_FUNC_get_cgroup_classid: 9104 return &bpf_get_cgroup_classid_proto; 9105 case BPF_FUNC_get_route_realm: 9106 return &bpf_get_route_realm_proto; 9107 case BPF_FUNC_get_hash_recalc: 9108 return &bpf_get_hash_recalc_proto; 9109 case BPF_FUNC_perf_event_output: 9110 return &bpf_skb_event_output_proto; 9111 case BPF_FUNC_get_smp_processor_id: 9112 return &bpf_get_smp_processor_id_proto; 9113 case BPF_FUNC_skb_under_cgroup: 9114 return &bpf_skb_under_cgroup_proto; 9115 default: 9116 return bpf_sk_base_func_proto(func_id, prog); 9117 } 9118 } 9119 9120 static const struct bpf_func_proto * 9121 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 9122 { 9123 switch (func_id) { 9124 case BPF_FUNC_lwt_push_encap: 9125 return &bpf_lwt_in_push_encap_proto; 9126 default: 9127 return lwt_out_func_proto(func_id, prog); 9128 } 9129 } 9130 9131 static const struct bpf_func_proto * 9132 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 9133 { 9134 switch (func_id) { 9135 case BPF_FUNC_skb_get_tunnel_key: 9136 return &bpf_skb_get_tunnel_key_proto; 9137 case BPF_FUNC_skb_set_tunnel_key: 9138 return bpf_get_skb_set_tunnel_proto(func_id); 9139 case BPF_FUNC_skb_get_tunnel_opt: 9140 return &bpf_skb_get_tunnel_opt_proto; 9141 case BPF_FUNC_skb_set_tunnel_opt: 9142 return bpf_get_skb_set_tunnel_proto(func_id); 9143 case BPF_FUNC_redirect: 9144 return &bpf_redirect_proto; 9145 case BPF_FUNC_clone_redirect: 9146 return &bpf_clone_redirect_proto; 9147 case BPF_FUNC_skb_change_tail: 9148 return &bpf_skb_change_tail_proto; 9149 case BPF_FUNC_skb_change_head: 9150 return &bpf_skb_change_head_proto; 9151 case BPF_FUNC_skb_store_bytes: 9152 return &bpf_skb_store_bytes_proto; 9153 case BPF_FUNC_csum_update: 9154 return &bpf_csum_update_proto; 9155 case BPF_FUNC_csum_level: 9156 return &bpf_csum_level_proto; 9157 case BPF_FUNC_l3_csum_replace: 9158 return &bpf_l3_csum_replace_proto; 9159 case BPF_FUNC_l4_csum_replace: 9160 return &bpf_l4_csum_replace_proto; 9161 case BPF_FUNC_set_hash_invalid: 9162 return &bpf_set_hash_invalid_proto; 9163 case BPF_FUNC_lwt_push_encap: 9164 return &bpf_lwt_xmit_push_encap_proto; 9165 default: 9166 return lwt_out_func_proto(func_id, prog); 9167 } 9168 } 9169 9170 static const struct bpf_func_proto * 9171 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 9172 { 9173 switch (func_id) { 9174 case BPF_FUNC_skb_pull_data: 9175 return NULL; 9176 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 9177 case BPF_FUNC_lwt_seg6_store_bytes: 9178 return &bpf_lwt_seg6_store_bytes_proto; 9179 case BPF_FUNC_lwt_seg6_action: 9180 return &bpf_lwt_seg6_action_proto; 9181 case BPF_FUNC_lwt_seg6_adjust_srh: 9182 return &bpf_lwt_seg6_adjust_srh_proto; 9183 #endif 9184 default: 9185 return lwt_out_func_proto(func_id, prog); 9186 } 9187 } 9188 9189 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 9190 const struct bpf_prog *prog, 9191 struct bpf_insn_access_aux *info) 9192 { 9193 const int size_default = sizeof(__u32); 9194 9195 if (off < 0 || off >= sizeof(struct __sk_buff)) 9196 return false; 9197 9198 /* The verifier guarantees that size > 0. */ 9199 if (off % size != 0) 9200 return false; 9201 9202 switch (off) { 9203 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 9204 if (off + size > offsetofend(struct __sk_buff, cb[4])) 9205 return false; 9206 break; 9207 case bpf_ctx_range(struct __sk_buff, data): 9208 case bpf_ctx_range(struct __sk_buff, data_meta): 9209 case bpf_ctx_range(struct __sk_buff, data_end): 9210 if (info->is_ldsx || size != size_default) 9211 return false; 9212 break; 9213 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 9214 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 9215 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 9216 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 9217 if (size != size_default) 9218 return false; 9219 break; 9220 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 9221 return false; 9222 case bpf_ctx_range(struct __sk_buff, hwtstamp): 9223 if (type == BPF_WRITE || size != sizeof(__u64)) 9224 return false; 9225 break; 9226 case bpf_ctx_range(struct __sk_buff, tstamp): 9227 if (size != sizeof(__u64)) 9228 return false; 9229 break; 9230 case bpf_ctx_range_ptr(struct __sk_buff, sk): 9231 if (type == BPF_WRITE || size != sizeof(__u64)) 9232 return false; 9233 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 9234 break; 9235 case offsetof(struct __sk_buff, tstamp_type): 9236 return false; 9237 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1: 9238 /* Explicitly prohibit access to padding in __sk_buff. */ 9239 return false; 9240 default: 9241 /* Only narrow read access allowed for now. */ 9242 if (type == BPF_WRITE) { 9243 if (size != size_default) 9244 return false; 9245 } else { 9246 bpf_ctx_record_field_size(info, size_default); 9247 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 9248 return false; 9249 } 9250 } 9251 9252 return true; 9253 } 9254 9255 static bool sk_filter_is_valid_access(int off, int size, 9256 enum bpf_access_type type, 9257 const struct bpf_prog *prog, 9258 struct bpf_insn_access_aux *info) 9259 { 9260 switch (off) { 9261 case bpf_ctx_range(struct __sk_buff, tc_classid): 9262 case bpf_ctx_range(struct __sk_buff, data): 9263 case bpf_ctx_range(struct __sk_buff, data_meta): 9264 case bpf_ctx_range(struct __sk_buff, data_end): 9265 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 9266 case bpf_ctx_range(struct __sk_buff, tstamp): 9267 case bpf_ctx_range(struct __sk_buff, wire_len): 9268 case bpf_ctx_range(struct __sk_buff, hwtstamp): 9269 return false; 9270 } 9271 9272 if (type == BPF_WRITE) { 9273 switch (off) { 9274 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 9275 break; 9276 default: 9277 return false; 9278 } 9279 } 9280 9281 return bpf_skb_is_valid_access(off, size, type, prog, info); 9282 } 9283 9284 static bool cg_skb_is_valid_access(int off, int size, 9285 enum bpf_access_type type, 9286 const struct bpf_prog *prog, 9287 struct bpf_insn_access_aux *info) 9288 { 9289 switch (off) { 9290 case bpf_ctx_range(struct __sk_buff, tc_classid): 9291 case bpf_ctx_range(struct __sk_buff, data_meta): 9292 case bpf_ctx_range(struct __sk_buff, wire_len): 9293 return false; 9294 case bpf_ctx_range(struct __sk_buff, data): 9295 case bpf_ctx_range(struct __sk_buff, data_end): 9296 if (!bpf_token_capable(prog->aux->token, CAP_BPF)) 9297 return false; 9298 break; 9299 } 9300 9301 if (type == BPF_WRITE) { 9302 switch (off) { 9303 case bpf_ctx_range(struct __sk_buff, mark): 9304 case bpf_ctx_range(struct __sk_buff, priority): 9305 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 9306 break; 9307 case bpf_ctx_range(struct __sk_buff, tstamp): 9308 if (!bpf_token_capable(prog->aux->token, CAP_BPF)) 9309 return false; 9310 break; 9311 default: 9312 return false; 9313 } 9314 } 9315 9316 switch (off) { 9317 case bpf_ctx_range(struct __sk_buff, data): 9318 info->reg_type = PTR_TO_PACKET; 9319 break; 9320 case bpf_ctx_range(struct __sk_buff, data_end): 9321 info->reg_type = PTR_TO_PACKET_END; 9322 break; 9323 } 9324 9325 return bpf_skb_is_valid_access(off, size, type, prog, info); 9326 } 9327 9328 static bool lwt_is_valid_access(int off, int size, 9329 enum bpf_access_type type, 9330 const struct bpf_prog *prog, 9331 struct bpf_insn_access_aux *info) 9332 { 9333 switch (off) { 9334 case bpf_ctx_range(struct __sk_buff, tc_classid): 9335 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 9336 case bpf_ctx_range(struct __sk_buff, data_meta): 9337 case bpf_ctx_range(struct __sk_buff, tstamp): 9338 case bpf_ctx_range(struct __sk_buff, wire_len): 9339 case bpf_ctx_range(struct __sk_buff, hwtstamp): 9340 return false; 9341 } 9342 9343 if (type == BPF_WRITE) { 9344 switch (off) { 9345 case bpf_ctx_range(struct __sk_buff, mark): 9346 case bpf_ctx_range(struct __sk_buff, priority): 9347 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 9348 break; 9349 default: 9350 return false; 9351 } 9352 } 9353 9354 switch (off) { 9355 case bpf_ctx_range(struct __sk_buff, data): 9356 info->reg_type = PTR_TO_PACKET; 9357 break; 9358 case bpf_ctx_range(struct __sk_buff, data_end): 9359 info->reg_type = PTR_TO_PACKET_END; 9360 break; 9361 } 9362 9363 return bpf_skb_is_valid_access(off, size, type, prog, info); 9364 } 9365 9366 /* Attach type specific accesses */ 9367 static bool __sock_filter_check_attach_type(int off, 9368 enum bpf_access_type access_type, 9369 enum bpf_attach_type attach_type) 9370 { 9371 switch (off) { 9372 case offsetof(struct bpf_sock, bound_dev_if): 9373 case offsetof(struct bpf_sock, mark): 9374 case offsetof(struct bpf_sock, priority): 9375 switch (attach_type) { 9376 case BPF_CGROUP_INET_SOCK_CREATE: 9377 case BPF_CGROUP_INET_SOCK_RELEASE: 9378 goto full_access; 9379 default: 9380 return false; 9381 } 9382 case bpf_ctx_range(struct bpf_sock, src_ip4): 9383 switch (attach_type) { 9384 case BPF_CGROUP_INET4_POST_BIND: 9385 goto read_only; 9386 default: 9387 return false; 9388 } 9389 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9390 switch (attach_type) { 9391 case BPF_CGROUP_INET6_POST_BIND: 9392 goto read_only; 9393 default: 9394 return false; 9395 } 9396 case bpf_ctx_range(struct bpf_sock, src_port): 9397 switch (attach_type) { 9398 case BPF_CGROUP_INET4_POST_BIND: 9399 case BPF_CGROUP_INET6_POST_BIND: 9400 goto read_only; 9401 default: 9402 return false; 9403 } 9404 } 9405 read_only: 9406 return access_type == BPF_READ; 9407 full_access: 9408 return true; 9409 } 9410 9411 bool bpf_sock_common_is_valid_access(int off, int size, 9412 enum bpf_access_type type, 9413 struct bpf_insn_access_aux *info) 9414 { 9415 switch (off) { 9416 case bpf_ctx_range_till(struct bpf_sock, type, priority): 9417 return false; 9418 default: 9419 return bpf_sock_is_valid_access(off, size, type, info); 9420 } 9421 } 9422 9423 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 9424 struct bpf_insn_access_aux *info) 9425 { 9426 const int size_default = sizeof(__u32); 9427 int field_size; 9428 9429 if (off < 0 || off >= sizeof(struct bpf_sock)) 9430 return false; 9431 if (off % size != 0) 9432 return false; 9433 9434 switch (off) { 9435 case offsetof(struct bpf_sock, state): 9436 case offsetof(struct bpf_sock, family): 9437 case offsetof(struct bpf_sock, type): 9438 case offsetof(struct bpf_sock, protocol): 9439 case offsetof(struct bpf_sock, src_port): 9440 case offsetof(struct bpf_sock, rx_queue_mapping): 9441 case bpf_ctx_range(struct bpf_sock, src_ip4): 9442 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9443 case bpf_ctx_range(struct bpf_sock, dst_ip4): 9444 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 9445 bpf_ctx_record_field_size(info, size_default); 9446 return bpf_ctx_narrow_access_ok(off, size, size_default); 9447 case bpf_ctx_range(struct bpf_sock, dst_port): 9448 field_size = size == size_default ? 9449 size_default : sizeof_field(struct bpf_sock, dst_port); 9450 bpf_ctx_record_field_size(info, field_size); 9451 return bpf_ctx_narrow_access_ok(off, size, field_size); 9452 case offsetofend(struct bpf_sock, dst_port) ... 9453 offsetof(struct bpf_sock, dst_ip4) - 1: 9454 return false; 9455 } 9456 9457 return size == size_default; 9458 } 9459 9460 static bool sock_filter_is_valid_access(int off, int size, 9461 enum bpf_access_type type, 9462 const struct bpf_prog *prog, 9463 struct bpf_insn_access_aux *info) 9464 { 9465 if (!bpf_sock_is_valid_access(off, size, type, info)) 9466 return false; 9467 return __sock_filter_check_attach_type(off, type, 9468 prog->expected_attach_type); 9469 } 9470 9471 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 9472 const struct bpf_prog *prog) 9473 { 9474 /* Neither direct read nor direct write requires any preliminary 9475 * action. 9476 */ 9477 return 0; 9478 } 9479 9480 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 9481 const struct bpf_prog *prog, int drop_verdict) 9482 { 9483 struct bpf_insn *insn = insn_buf; 9484 9485 if (!direct_write) 9486 return 0; 9487 9488 /* if (!skb->cloned) 9489 * goto start; 9490 * 9491 * (Fast-path, otherwise approximation that we might be 9492 * a clone, do the rest in helper.) 9493 */ 9494 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET); 9495 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 9496 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 9497 9498 /* ret = bpf_skb_pull_data(skb, 0); */ 9499 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 9500 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 9501 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 9502 BPF_FUNC_skb_pull_data); 9503 /* if (!ret) 9504 * goto restore; 9505 * return TC_ACT_SHOT; 9506 */ 9507 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 9508 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 9509 *insn++ = BPF_EXIT_INSN(); 9510 9511 /* restore: */ 9512 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 9513 /* start: */ 9514 *insn++ = prog->insnsi[0]; 9515 9516 return insn - insn_buf; 9517 } 9518 9519 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 9520 struct bpf_insn *insn_buf) 9521 { 9522 bool indirect = BPF_MODE(orig->code) == BPF_IND; 9523 struct bpf_insn *insn = insn_buf; 9524 9525 if (!indirect) { 9526 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 9527 } else { 9528 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 9529 if (orig->imm) 9530 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 9531 } 9532 /* We're guaranteed here that CTX is in R6. */ 9533 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 9534 9535 switch (BPF_SIZE(orig->code)) { 9536 case BPF_B: 9537 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 9538 break; 9539 case BPF_H: 9540 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 9541 break; 9542 case BPF_W: 9543 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 9544 break; 9545 } 9546 9547 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 9548 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 9549 *insn++ = BPF_EXIT_INSN(); 9550 9551 return insn - insn_buf; 9552 } 9553 9554 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 9555 const struct bpf_prog *prog) 9556 { 9557 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 9558 } 9559 9560 static bool tc_cls_act_is_valid_access(int off, int size, 9561 enum bpf_access_type type, 9562 const struct bpf_prog *prog, 9563 struct bpf_insn_access_aux *info) 9564 { 9565 if (type == BPF_WRITE) { 9566 switch (off) { 9567 case bpf_ctx_range(struct __sk_buff, mark): 9568 case bpf_ctx_range(struct __sk_buff, tc_index): 9569 case bpf_ctx_range(struct __sk_buff, priority): 9570 case bpf_ctx_range(struct __sk_buff, tc_classid): 9571 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 9572 case bpf_ctx_range(struct __sk_buff, tstamp): 9573 case bpf_ctx_range(struct __sk_buff, queue_mapping): 9574 break; 9575 default: 9576 return false; 9577 } 9578 } 9579 9580 switch (off) { 9581 case bpf_ctx_range(struct __sk_buff, data): 9582 info->reg_type = PTR_TO_PACKET; 9583 break; 9584 case bpf_ctx_range(struct __sk_buff, data_meta): 9585 info->reg_type = PTR_TO_PACKET_META; 9586 break; 9587 case bpf_ctx_range(struct __sk_buff, data_end): 9588 info->reg_type = PTR_TO_PACKET_END; 9589 break; 9590 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 9591 return false; 9592 case offsetof(struct __sk_buff, tstamp_type): 9593 /* The convert_ctx_access() on reading and writing 9594 * __sk_buff->tstamp depends on whether the bpf prog 9595 * has used __sk_buff->tstamp_type or not. 9596 * Thus, we need to set prog->tstamp_type_access 9597 * earlier during is_valid_access() here. 9598 */ 9599 ((struct bpf_prog *)prog)->tstamp_type_access = 1; 9600 return size == sizeof(__u8); 9601 } 9602 9603 return bpf_skb_is_valid_access(off, size, type, prog, info); 9604 } 9605 9606 DEFINE_MUTEX(nf_conn_btf_access_lock); 9607 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock); 9608 9609 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 9610 const struct bpf_reg_state *reg, 9611 int off, int size); 9612 EXPORT_SYMBOL_GPL(nfct_btf_struct_access); 9613 9614 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log, 9615 const struct bpf_reg_state *reg, 9616 int off, int size) 9617 { 9618 int ret = -EACCES; 9619 9620 mutex_lock(&nf_conn_btf_access_lock); 9621 if (nfct_btf_struct_access) 9622 ret = nfct_btf_struct_access(log, reg, off, size); 9623 mutex_unlock(&nf_conn_btf_access_lock); 9624 9625 return ret; 9626 } 9627 9628 static bool __is_valid_xdp_access(int off, int size) 9629 { 9630 if (off < 0 || off >= sizeof(struct xdp_md)) 9631 return false; 9632 if (off % size != 0) 9633 return false; 9634 if (size != sizeof(__u32)) 9635 return false; 9636 9637 return true; 9638 } 9639 9640 static bool xdp_is_valid_access(int off, int size, 9641 enum bpf_access_type type, 9642 const struct bpf_prog *prog, 9643 struct bpf_insn_access_aux *info) 9644 { 9645 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 9646 switch (off) { 9647 case offsetof(struct xdp_md, egress_ifindex): 9648 return false; 9649 } 9650 } 9651 9652 if (type == BPF_WRITE) { 9653 if (bpf_prog_is_offloaded(prog->aux)) { 9654 switch (off) { 9655 case offsetof(struct xdp_md, rx_queue_index): 9656 return __is_valid_xdp_access(off, size); 9657 } 9658 } 9659 return false; 9660 } else { 9661 switch (off) { 9662 case offsetof(struct xdp_md, data_meta): 9663 case offsetof(struct xdp_md, data): 9664 case offsetof(struct xdp_md, data_end): 9665 if (info->is_ldsx) 9666 return false; 9667 } 9668 } 9669 9670 switch (off) { 9671 case offsetof(struct xdp_md, data): 9672 info->reg_type = PTR_TO_PACKET; 9673 break; 9674 case offsetof(struct xdp_md, data_meta): 9675 info->reg_type = PTR_TO_PACKET_META; 9676 break; 9677 case offsetof(struct xdp_md, data_end): 9678 info->reg_type = PTR_TO_PACKET_END; 9679 break; 9680 } 9681 9682 return __is_valid_xdp_access(off, size); 9683 } 9684 9685 void bpf_warn_invalid_xdp_action(const struct net_device *dev, 9686 const struct bpf_prog *prog, u32 act) 9687 { 9688 const u32 act_max = XDP_REDIRECT; 9689 9690 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n", 9691 act > act_max ? "Illegal" : "Driver unsupported", 9692 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A"); 9693 } 9694 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 9695 9696 static int xdp_btf_struct_access(struct bpf_verifier_log *log, 9697 const struct bpf_reg_state *reg, 9698 int off, int size) 9699 { 9700 int ret = -EACCES; 9701 9702 mutex_lock(&nf_conn_btf_access_lock); 9703 if (nfct_btf_struct_access) 9704 ret = nfct_btf_struct_access(log, reg, off, size); 9705 mutex_unlock(&nf_conn_btf_access_lock); 9706 9707 return ret; 9708 } 9709 9710 static bool sock_addr_is_valid_access(int off, int size, 9711 enum bpf_access_type type, 9712 const struct bpf_prog *prog, 9713 struct bpf_insn_access_aux *info) 9714 { 9715 const int size_default = sizeof(__u32); 9716 9717 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 9718 return false; 9719 if (off % size != 0) 9720 return false; 9721 9722 /* Disallow access to fields not belonging to the attach type's address 9723 * family. 9724 */ 9725 switch (off) { 9726 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 9727 switch (prog->expected_attach_type) { 9728 case BPF_CGROUP_INET4_BIND: 9729 case BPF_CGROUP_INET4_CONNECT: 9730 case BPF_CGROUP_INET4_GETPEERNAME: 9731 case BPF_CGROUP_INET4_GETSOCKNAME: 9732 case BPF_CGROUP_UDP4_SENDMSG: 9733 case BPF_CGROUP_UDP4_RECVMSG: 9734 break; 9735 default: 9736 return false; 9737 } 9738 break; 9739 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 9740 switch (prog->expected_attach_type) { 9741 case BPF_CGROUP_INET6_BIND: 9742 case BPF_CGROUP_INET6_CONNECT: 9743 case BPF_CGROUP_INET6_GETPEERNAME: 9744 case BPF_CGROUP_INET6_GETSOCKNAME: 9745 case BPF_CGROUP_UDP6_SENDMSG: 9746 case BPF_CGROUP_UDP6_RECVMSG: 9747 break; 9748 default: 9749 return false; 9750 } 9751 break; 9752 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 9753 switch (prog->expected_attach_type) { 9754 case BPF_CGROUP_UDP4_SENDMSG: 9755 break; 9756 default: 9757 return false; 9758 } 9759 break; 9760 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9761 msg_src_ip6[3]): 9762 switch (prog->expected_attach_type) { 9763 case BPF_CGROUP_UDP6_SENDMSG: 9764 break; 9765 default: 9766 return false; 9767 } 9768 break; 9769 } 9770 9771 switch (off) { 9772 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 9773 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 9774 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 9775 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9776 msg_src_ip6[3]): 9777 case bpf_ctx_range(struct bpf_sock_addr, user_port): 9778 if (type == BPF_READ) { 9779 bpf_ctx_record_field_size(info, size_default); 9780 9781 if (bpf_ctx_wide_access_ok(off, size, 9782 struct bpf_sock_addr, 9783 user_ip6)) 9784 return true; 9785 9786 if (bpf_ctx_wide_access_ok(off, size, 9787 struct bpf_sock_addr, 9788 msg_src_ip6)) 9789 return true; 9790 9791 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 9792 return false; 9793 } else { 9794 if (bpf_ctx_wide_access_ok(off, size, 9795 struct bpf_sock_addr, 9796 user_ip6)) 9797 return true; 9798 9799 if (bpf_ctx_wide_access_ok(off, size, 9800 struct bpf_sock_addr, 9801 msg_src_ip6)) 9802 return true; 9803 9804 if (size != size_default) 9805 return false; 9806 } 9807 break; 9808 case bpf_ctx_range_ptr(struct bpf_sock_addr, sk): 9809 if (type != BPF_READ) 9810 return false; 9811 if (size != sizeof(__u64)) 9812 return false; 9813 info->reg_type = PTR_TO_SOCKET; 9814 break; 9815 case bpf_ctx_range(struct bpf_sock_addr, user_family): 9816 case bpf_ctx_range(struct bpf_sock_addr, family): 9817 case bpf_ctx_range(struct bpf_sock_addr, type): 9818 case bpf_ctx_range(struct bpf_sock_addr, protocol): 9819 if (type != BPF_READ) 9820 return false; 9821 if (size != size_default) 9822 return false; 9823 break; 9824 default: 9825 return false; 9826 } 9827 9828 return true; 9829 } 9830 9831 static bool sock_ops_is_valid_access(int off, int size, 9832 enum bpf_access_type type, 9833 const struct bpf_prog *prog, 9834 struct bpf_insn_access_aux *info) 9835 { 9836 const int size_default = sizeof(__u32); 9837 9838 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 9839 return false; 9840 9841 /* The verifier guarantees that size > 0. */ 9842 if (off % size != 0) 9843 return false; 9844 9845 if (type == BPF_WRITE) { 9846 switch (off) { 9847 case offsetof(struct bpf_sock_ops, reply): 9848 case offsetof(struct bpf_sock_ops, sk_txhash): 9849 if (size != size_default) 9850 return false; 9851 break; 9852 default: 9853 return false; 9854 } 9855 } else { 9856 switch (off) { 9857 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 9858 bytes_acked): 9859 if (size != sizeof(__u64)) 9860 return false; 9861 break; 9862 case bpf_ctx_range_ptr(struct bpf_sock_ops, sk): 9863 if (size != sizeof(__u64)) 9864 return false; 9865 info->reg_type = PTR_TO_SOCKET_OR_NULL; 9866 break; 9867 case bpf_ctx_range_ptr(struct bpf_sock_ops, skb_data): 9868 if (size != sizeof(__u64)) 9869 return false; 9870 info->reg_type = PTR_TO_PACKET; 9871 break; 9872 case bpf_ctx_range_ptr(struct bpf_sock_ops, skb_data_end): 9873 if (size != sizeof(__u64)) 9874 return false; 9875 info->reg_type = PTR_TO_PACKET_END; 9876 break; 9877 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 9878 bpf_ctx_record_field_size(info, size_default); 9879 return bpf_ctx_narrow_access_ok(off, size, 9880 size_default); 9881 case bpf_ctx_range(struct bpf_sock_ops, skb_hwtstamp): 9882 if (size != sizeof(__u64)) 9883 return false; 9884 break; 9885 default: 9886 if (size != size_default) 9887 return false; 9888 break; 9889 } 9890 } 9891 9892 return true; 9893 } 9894 9895 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 9896 const struct bpf_prog *prog) 9897 { 9898 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 9899 } 9900 9901 static bool sk_skb_is_valid_access(int off, int size, 9902 enum bpf_access_type type, 9903 const struct bpf_prog *prog, 9904 struct bpf_insn_access_aux *info) 9905 { 9906 switch (off) { 9907 case bpf_ctx_range(struct __sk_buff, tc_classid): 9908 case bpf_ctx_range(struct __sk_buff, data_meta): 9909 case bpf_ctx_range(struct __sk_buff, tstamp): 9910 case bpf_ctx_range(struct __sk_buff, wire_len): 9911 case bpf_ctx_range(struct __sk_buff, hwtstamp): 9912 return false; 9913 } 9914 9915 if (type == BPF_WRITE) { 9916 switch (off) { 9917 case bpf_ctx_range(struct __sk_buff, tc_index): 9918 case bpf_ctx_range(struct __sk_buff, priority): 9919 break; 9920 default: 9921 return false; 9922 } 9923 } 9924 9925 switch (off) { 9926 case bpf_ctx_range(struct __sk_buff, mark): 9927 return false; 9928 case bpf_ctx_range(struct __sk_buff, data): 9929 info->reg_type = PTR_TO_PACKET; 9930 break; 9931 case bpf_ctx_range(struct __sk_buff, data_end): 9932 info->reg_type = PTR_TO_PACKET_END; 9933 break; 9934 } 9935 9936 return bpf_skb_is_valid_access(off, size, type, prog, info); 9937 } 9938 9939 static bool sk_msg_is_valid_access(int off, int size, 9940 enum bpf_access_type type, 9941 const struct bpf_prog *prog, 9942 struct bpf_insn_access_aux *info) 9943 { 9944 if (type == BPF_WRITE) 9945 return false; 9946 9947 if (off % size != 0) 9948 return false; 9949 9950 switch (off) { 9951 case bpf_ctx_range_ptr(struct sk_msg_md, data): 9952 info->reg_type = PTR_TO_PACKET; 9953 if (size != sizeof(__u64)) 9954 return false; 9955 break; 9956 case bpf_ctx_range_ptr(struct sk_msg_md, data_end): 9957 info->reg_type = PTR_TO_PACKET_END; 9958 if (size != sizeof(__u64)) 9959 return false; 9960 break; 9961 case bpf_ctx_range_ptr(struct sk_msg_md, sk): 9962 if (size != sizeof(__u64)) 9963 return false; 9964 info->reg_type = PTR_TO_SOCKET; 9965 break; 9966 case bpf_ctx_range(struct sk_msg_md, family): 9967 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 9968 case bpf_ctx_range(struct sk_msg_md, local_ip4): 9969 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 9970 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 9971 case bpf_ctx_range(struct sk_msg_md, remote_port): 9972 case bpf_ctx_range(struct sk_msg_md, local_port): 9973 case bpf_ctx_range(struct sk_msg_md, size): 9974 if (size != sizeof(__u32)) 9975 return false; 9976 break; 9977 default: 9978 return false; 9979 } 9980 return true; 9981 } 9982 9983 static bool flow_dissector_is_valid_access(int off, int size, 9984 enum bpf_access_type type, 9985 const struct bpf_prog *prog, 9986 struct bpf_insn_access_aux *info) 9987 { 9988 const int size_default = sizeof(__u32); 9989 9990 if (off < 0 || off >= sizeof(struct __sk_buff)) 9991 return false; 9992 9993 if (off % size != 0) 9994 return false; 9995 9996 if (type == BPF_WRITE) 9997 return false; 9998 9999 switch (off) { 10000 case bpf_ctx_range(struct __sk_buff, data): 10001 if (info->is_ldsx || size != size_default) 10002 return false; 10003 info->reg_type = PTR_TO_PACKET; 10004 return true; 10005 case bpf_ctx_range(struct __sk_buff, data_end): 10006 if (info->is_ldsx || size != size_default) 10007 return false; 10008 info->reg_type = PTR_TO_PACKET_END; 10009 return true; 10010 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 10011 if (size != sizeof(__u64)) 10012 return false; 10013 info->reg_type = PTR_TO_FLOW_KEYS; 10014 return true; 10015 default: 10016 return false; 10017 } 10018 } 10019 10020 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 10021 const struct bpf_insn *si, 10022 struct bpf_insn *insn_buf, 10023 struct bpf_prog *prog, 10024 u32 *target_size) 10025 10026 { 10027 struct bpf_insn *insn = insn_buf; 10028 10029 switch (si->off) { 10030 case offsetof(struct __sk_buff, data): 10031 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 10032 si->dst_reg, si->src_reg, 10033 offsetof(struct bpf_flow_dissector, data)); 10034 break; 10035 10036 case offsetof(struct __sk_buff, data_end): 10037 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 10038 si->dst_reg, si->src_reg, 10039 offsetof(struct bpf_flow_dissector, data_end)); 10040 break; 10041 10042 case offsetof(struct __sk_buff, flow_keys): 10043 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 10044 si->dst_reg, si->src_reg, 10045 offsetof(struct bpf_flow_dissector, flow_keys)); 10046 break; 10047 } 10048 10049 return insn - insn_buf; 10050 } 10051 10052 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si, 10053 struct bpf_insn *insn) 10054 { 10055 __u8 value_reg = si->dst_reg; 10056 __u8 skb_reg = si->src_reg; 10057 BUILD_BUG_ON(__SKB_CLOCK_MAX != (int)BPF_SKB_CLOCK_TAI); 10058 BUILD_BUG_ON(SKB_CLOCK_REALTIME != (int)BPF_SKB_CLOCK_REALTIME); 10059 BUILD_BUG_ON(SKB_CLOCK_MONOTONIC != (int)BPF_SKB_CLOCK_MONOTONIC); 10060 BUILD_BUG_ON(SKB_CLOCK_TAI != (int)BPF_SKB_CLOCK_TAI); 10061 *insn++ = BPF_LDX_MEM(BPF_B, value_reg, skb_reg, SKB_BF_MONO_TC_OFFSET); 10062 *insn++ = BPF_ALU32_IMM(BPF_AND, value_reg, SKB_TSTAMP_TYPE_MASK); 10063 #ifdef __BIG_ENDIAN_BITFIELD 10064 *insn++ = BPF_ALU32_IMM(BPF_RSH, value_reg, SKB_TSTAMP_TYPE_RSHIFT); 10065 #else 10066 BUILD_BUG_ON(!(SKB_TSTAMP_TYPE_MASK & 0x1)); 10067 #endif 10068 10069 return insn; 10070 } 10071 10072 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg, 10073 struct bpf_insn *insn) 10074 { 10075 /* si->dst_reg = skb_shinfo(SKB); */ 10076 #ifdef NET_SKBUFF_DATA_USES_OFFSET 10077 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 10078 BPF_REG_AX, skb_reg, 10079 offsetof(struct sk_buff, end)); 10080 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 10081 dst_reg, skb_reg, 10082 offsetof(struct sk_buff, head)); 10083 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX); 10084 #else 10085 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 10086 dst_reg, skb_reg, 10087 offsetof(struct sk_buff, end)); 10088 #endif 10089 10090 return insn; 10091 } 10092 10093 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, 10094 const struct bpf_insn *si, 10095 struct bpf_insn *insn) 10096 { 10097 __u8 value_reg = si->dst_reg; 10098 __u8 skb_reg = si->src_reg; 10099 10100 #ifdef CONFIG_NET_XGRESS 10101 /* If the tstamp_type is read, 10102 * the bpf prog is aware the tstamp could have delivery time. 10103 * Thus, read skb->tstamp as is if tstamp_type_access is true. 10104 */ 10105 if (!prog->tstamp_type_access) { 10106 /* AX is needed because src_reg and dst_reg could be the same */ 10107 __u8 tmp_reg = BPF_REG_AX; 10108 10109 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET); 10110 /* check if ingress mask bits is set */ 10111 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); 10112 *insn++ = BPF_JMP_A(4); 10113 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, SKB_TSTAMP_TYPE_MASK, 1); 10114 *insn++ = BPF_JMP_A(2); 10115 /* skb->tc_at_ingress && skb->tstamp_type, 10116 * read 0 as the (rcv) timestamp. 10117 */ 10118 *insn++ = BPF_MOV64_IMM(value_reg, 0); 10119 *insn++ = BPF_JMP_A(1); 10120 } 10121 #endif 10122 10123 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg, 10124 offsetof(struct sk_buff, tstamp)); 10125 return insn; 10126 } 10127 10128 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, 10129 const struct bpf_insn *si, 10130 struct bpf_insn *insn) 10131 { 10132 __u8 value_reg = si->src_reg; 10133 __u8 skb_reg = si->dst_reg; 10134 10135 #ifdef CONFIG_NET_XGRESS 10136 /* If the tstamp_type is read, 10137 * the bpf prog is aware the tstamp could have delivery time. 10138 * Thus, write skb->tstamp as is if tstamp_type_access is true. 10139 * Otherwise, writing at ingress will have to clear the 10140 * skb->tstamp_type bit also. 10141 */ 10142 if (!prog->tstamp_type_access) { 10143 __u8 tmp_reg = BPF_REG_AX; 10144 10145 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET); 10146 /* Writing __sk_buff->tstamp as ingress, goto <clear> */ 10147 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); 10148 /* goto <store> */ 10149 *insn++ = BPF_JMP_A(2); 10150 /* <clear>: skb->tstamp_type */ 10151 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_TSTAMP_TYPE_MASK); 10152 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET); 10153 } 10154 #endif 10155 10156 /* <store>: skb->tstamp = tstamp */ 10157 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM, 10158 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm); 10159 return insn; 10160 } 10161 10162 #define BPF_EMIT_STORE(size, si, off) \ 10163 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \ 10164 (si)->dst_reg, (si)->src_reg, (off), (si)->imm) 10165 10166 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 10167 const struct bpf_insn *si, 10168 struct bpf_insn *insn_buf, 10169 struct bpf_prog *prog, u32 *target_size) 10170 { 10171 struct bpf_insn *insn = insn_buf; 10172 int off; 10173 10174 switch (si->off) { 10175 case offsetof(struct __sk_buff, len): 10176 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10177 bpf_target_off(struct sk_buff, len, 4, 10178 target_size)); 10179 break; 10180 10181 case offsetof(struct __sk_buff, protocol): 10182 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10183 bpf_target_off(struct sk_buff, protocol, 2, 10184 target_size)); 10185 break; 10186 10187 case offsetof(struct __sk_buff, vlan_proto): 10188 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10189 bpf_target_off(struct sk_buff, vlan_proto, 2, 10190 target_size)); 10191 break; 10192 10193 case offsetof(struct __sk_buff, priority): 10194 if (type == BPF_WRITE) 10195 *insn++ = BPF_EMIT_STORE(BPF_W, si, 10196 bpf_target_off(struct sk_buff, priority, 4, 10197 target_size)); 10198 else 10199 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10200 bpf_target_off(struct sk_buff, priority, 4, 10201 target_size)); 10202 break; 10203 10204 case offsetof(struct __sk_buff, ingress_ifindex): 10205 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10206 bpf_target_off(struct sk_buff, skb_iif, 4, 10207 target_size)); 10208 break; 10209 10210 case offsetof(struct __sk_buff, ifindex): 10211 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 10212 si->dst_reg, si->src_reg, 10213 offsetof(struct sk_buff, dev)); 10214 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10215 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10216 bpf_target_off(struct net_device, ifindex, 4, 10217 target_size)); 10218 break; 10219 10220 case offsetof(struct __sk_buff, hash): 10221 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10222 bpf_target_off(struct sk_buff, hash, 4, 10223 target_size)); 10224 break; 10225 10226 case offsetof(struct __sk_buff, mark): 10227 if (type == BPF_WRITE) 10228 *insn++ = BPF_EMIT_STORE(BPF_W, si, 10229 bpf_target_off(struct sk_buff, mark, 4, 10230 target_size)); 10231 else 10232 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10233 bpf_target_off(struct sk_buff, mark, 4, 10234 target_size)); 10235 break; 10236 10237 case offsetof(struct __sk_buff, pkt_type): 10238 *target_size = 1; 10239 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 10240 PKT_TYPE_OFFSET); 10241 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 10242 #ifdef __BIG_ENDIAN_BITFIELD 10243 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 10244 #endif 10245 break; 10246 10247 case offsetof(struct __sk_buff, queue_mapping): 10248 if (type == BPF_WRITE) { 10249 u32 offset = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size); 10250 10251 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) { 10252 *insn++ = BPF_JMP_A(0); /* noop */ 10253 break; 10254 } 10255 10256 if (BPF_CLASS(si->code) == BPF_STX) 10257 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 10258 *insn++ = BPF_EMIT_STORE(BPF_H, si, offset); 10259 } else { 10260 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10261 bpf_target_off(struct sk_buff, 10262 queue_mapping, 10263 2, target_size)); 10264 } 10265 break; 10266 10267 case offsetof(struct __sk_buff, vlan_present): 10268 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10269 bpf_target_off(struct sk_buff, 10270 vlan_all, 4, target_size)); 10271 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10272 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1); 10273 break; 10274 10275 case offsetof(struct __sk_buff, vlan_tci): 10276 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10277 bpf_target_off(struct sk_buff, vlan_tci, 2, 10278 target_size)); 10279 break; 10280 10281 case offsetof(struct __sk_buff, cb[0]) ... 10282 offsetofend(struct __sk_buff, cb[4]) - 1: 10283 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 10284 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 10285 offsetof(struct qdisc_skb_cb, data)) % 10286 sizeof(__u64)); 10287 10288 prog->cb_access = 1; 10289 off = si->off; 10290 off -= offsetof(struct __sk_buff, cb[0]); 10291 off += offsetof(struct sk_buff, cb); 10292 off += offsetof(struct qdisc_skb_cb, data); 10293 if (type == BPF_WRITE) 10294 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off); 10295 else 10296 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 10297 si->src_reg, off); 10298 break; 10299 10300 case offsetof(struct __sk_buff, tc_classid): 10301 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 10302 10303 off = si->off; 10304 off -= offsetof(struct __sk_buff, tc_classid); 10305 off += offsetof(struct sk_buff, cb); 10306 off += offsetof(struct qdisc_skb_cb, tc_classid); 10307 *target_size = 2; 10308 if (type == BPF_WRITE) 10309 *insn++ = BPF_EMIT_STORE(BPF_H, si, off); 10310 else 10311 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 10312 si->src_reg, off); 10313 break; 10314 10315 case offsetof(struct __sk_buff, data): 10316 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10317 si->dst_reg, si->src_reg, 10318 offsetof(struct sk_buff, data)); 10319 break; 10320 10321 case offsetof(struct __sk_buff, data_meta): 10322 off = si->off; 10323 off -= offsetof(struct __sk_buff, data_meta); 10324 off += offsetof(struct sk_buff, cb); 10325 off += offsetof(struct bpf_skb_data_end, data_meta); 10326 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 10327 si->src_reg, off); 10328 break; 10329 10330 case offsetof(struct __sk_buff, data_end): 10331 off = si->off; 10332 off -= offsetof(struct __sk_buff, data_end); 10333 off += offsetof(struct sk_buff, cb); 10334 off += offsetof(struct bpf_skb_data_end, data_end); 10335 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 10336 si->src_reg, off); 10337 break; 10338 10339 case offsetof(struct __sk_buff, tc_index): 10340 #ifdef CONFIG_NET_SCHED 10341 if (type == BPF_WRITE) 10342 *insn++ = BPF_EMIT_STORE(BPF_H, si, 10343 bpf_target_off(struct sk_buff, tc_index, 2, 10344 target_size)); 10345 else 10346 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10347 bpf_target_off(struct sk_buff, tc_index, 2, 10348 target_size)); 10349 #else 10350 *target_size = 2; 10351 if (type == BPF_WRITE) 10352 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 10353 else 10354 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 10355 #endif 10356 break; 10357 10358 case offsetof(struct __sk_buff, napi_id): 10359 #if defined(CONFIG_NET_RX_BUSY_POLL) 10360 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10361 bpf_target_off(struct sk_buff, napi_id, 4, 10362 target_size)); 10363 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 10364 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 10365 #else 10366 *target_size = 4; 10367 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 10368 #endif 10369 break; 10370 case offsetof(struct __sk_buff, family): 10371 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10372 10373 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10374 si->dst_reg, si->src_reg, 10375 offsetof(struct sk_buff, sk)); 10376 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10377 bpf_target_off(struct sock_common, 10378 skc_family, 10379 2, target_size)); 10380 break; 10381 case offsetof(struct __sk_buff, remote_ip4): 10382 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10383 10384 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10385 si->dst_reg, si->src_reg, 10386 offsetof(struct sk_buff, sk)); 10387 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10388 bpf_target_off(struct sock_common, 10389 skc_daddr, 10390 4, target_size)); 10391 break; 10392 case offsetof(struct __sk_buff, local_ip4): 10393 BUILD_BUG_ON(sizeof_field(struct sock_common, 10394 skc_rcv_saddr) != 4); 10395 10396 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10397 si->dst_reg, si->src_reg, 10398 offsetof(struct sk_buff, sk)); 10399 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10400 bpf_target_off(struct sock_common, 10401 skc_rcv_saddr, 10402 4, target_size)); 10403 break; 10404 case offsetof(struct __sk_buff, remote_ip6[0]) ... 10405 offsetof(struct __sk_buff, remote_ip6[3]): 10406 #if IS_ENABLED(CONFIG_IPV6) 10407 BUILD_BUG_ON(sizeof_field(struct sock_common, 10408 skc_v6_daddr.s6_addr32[0]) != 4); 10409 10410 off = si->off; 10411 off -= offsetof(struct __sk_buff, remote_ip6[0]); 10412 10413 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10414 si->dst_reg, si->src_reg, 10415 offsetof(struct sk_buff, sk)); 10416 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10417 offsetof(struct sock_common, 10418 skc_v6_daddr.s6_addr32[0]) + 10419 off); 10420 #else 10421 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10422 #endif 10423 break; 10424 case offsetof(struct __sk_buff, local_ip6[0]) ... 10425 offsetof(struct __sk_buff, local_ip6[3]): 10426 #if IS_ENABLED(CONFIG_IPV6) 10427 BUILD_BUG_ON(sizeof_field(struct sock_common, 10428 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10429 10430 off = si->off; 10431 off -= offsetof(struct __sk_buff, local_ip6[0]); 10432 10433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10434 si->dst_reg, si->src_reg, 10435 offsetof(struct sk_buff, sk)); 10436 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10437 offsetof(struct sock_common, 10438 skc_v6_rcv_saddr.s6_addr32[0]) + 10439 off); 10440 #else 10441 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10442 #endif 10443 break; 10444 10445 case offsetof(struct __sk_buff, remote_port): 10446 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10447 10448 /* 10449 * The converted value is 4 bytes wide, not 2: on little 10450 * endian the 16-bit port is shifted into the high half. 10451 * Don't use bpf_target_off() here, it would report the 10452 * member size and the narrow-load masking in the fixup 10453 * pass would be skipped. 10454 */ 10455 *target_size = 4; 10456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10457 si->dst_reg, si->src_reg, 10458 offsetof(struct sk_buff, sk)); 10459 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10460 offsetof(struct sock_common, skc_dport)); 10461 #ifndef __BIG_ENDIAN_BITFIELD 10462 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10463 #endif 10464 break; 10465 10466 case offsetof(struct __sk_buff, local_port): 10467 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10468 10469 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10470 si->dst_reg, si->src_reg, 10471 offsetof(struct sk_buff, sk)); 10472 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10473 bpf_target_off(struct sock_common, 10474 skc_num, 2, target_size)); 10475 break; 10476 10477 case offsetof(struct __sk_buff, tstamp): 10478 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 10479 10480 if (type == BPF_WRITE) 10481 insn = bpf_convert_tstamp_write(prog, si, insn); 10482 else 10483 insn = bpf_convert_tstamp_read(prog, si, insn); 10484 break; 10485 10486 case offsetof(struct __sk_buff, tstamp_type): 10487 insn = bpf_convert_tstamp_type_read(si, insn); 10488 break; 10489 10490 case offsetof(struct __sk_buff, gso_segs): 10491 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 10492 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 10493 si->dst_reg, si->dst_reg, 10494 bpf_target_off(struct skb_shared_info, 10495 gso_segs, 2, 10496 target_size)); 10497 break; 10498 case offsetof(struct __sk_buff, gso_size): 10499 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 10500 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 10501 si->dst_reg, si->dst_reg, 10502 bpf_target_off(struct skb_shared_info, 10503 gso_size, 2, 10504 target_size)); 10505 break; 10506 case offsetof(struct __sk_buff, wire_len): 10507 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 10508 10509 off = si->off; 10510 off -= offsetof(struct __sk_buff, wire_len); 10511 off += offsetof(struct sk_buff, cb); 10512 off += offsetof(struct qdisc_skb_cb, pkt_len); 10513 *target_size = 4; 10514 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 10515 break; 10516 10517 case offsetof(struct __sk_buff, sk): 10518 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 10519 si->dst_reg, si->src_reg, 10520 offsetof(struct sk_buff, sk)); 10521 break; 10522 case offsetof(struct __sk_buff, hwtstamp): 10523 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8); 10524 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0); 10525 10526 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 10527 *insn++ = BPF_LDX_MEM(BPF_DW, 10528 si->dst_reg, si->dst_reg, 10529 bpf_target_off(struct skb_shared_info, 10530 hwtstamps, 8, 10531 target_size)); 10532 break; 10533 } 10534 10535 return insn - insn_buf; 10536 } 10537 10538 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 10539 const struct bpf_insn *si, 10540 struct bpf_insn *insn_buf, 10541 struct bpf_prog *prog, u32 *target_size) 10542 { 10543 struct bpf_insn *insn = insn_buf; 10544 int off; 10545 10546 switch (si->off) { 10547 case offsetof(struct bpf_sock, bound_dev_if): 10548 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 10549 10550 if (type == BPF_WRITE) 10551 *insn++ = BPF_EMIT_STORE(BPF_W, si, 10552 offsetof(struct sock, sk_bound_dev_if)); 10553 else 10554 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10555 offsetof(struct sock, sk_bound_dev_if)); 10556 break; 10557 10558 case offsetof(struct bpf_sock, mark): 10559 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 10560 10561 if (type == BPF_WRITE) 10562 *insn++ = BPF_EMIT_STORE(BPF_W, si, 10563 offsetof(struct sock, sk_mark)); 10564 else 10565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10566 offsetof(struct sock, sk_mark)); 10567 break; 10568 10569 case offsetof(struct bpf_sock, priority): 10570 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 10571 10572 if (type == BPF_WRITE) 10573 *insn++ = BPF_EMIT_STORE(BPF_W, si, 10574 offsetof(struct sock, sk_priority)); 10575 else 10576 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10577 offsetof(struct sock, sk_priority)); 10578 break; 10579 10580 case offsetof(struct bpf_sock, family): 10581 *insn++ = BPF_LDX_MEM( 10582 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 10583 si->dst_reg, si->src_reg, 10584 bpf_target_off(struct sock_common, 10585 skc_family, 10586 sizeof_field(struct sock_common, 10587 skc_family), 10588 target_size)); 10589 break; 10590 10591 case offsetof(struct bpf_sock, type): 10592 *insn++ = BPF_LDX_MEM( 10593 BPF_FIELD_SIZEOF(struct sock, sk_type), 10594 si->dst_reg, si->src_reg, 10595 bpf_target_off(struct sock, sk_type, 10596 sizeof_field(struct sock, sk_type), 10597 target_size)); 10598 break; 10599 10600 case offsetof(struct bpf_sock, protocol): 10601 *insn++ = BPF_LDX_MEM( 10602 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 10603 si->dst_reg, si->src_reg, 10604 bpf_target_off(struct sock, sk_protocol, 10605 sizeof_field(struct sock, sk_protocol), 10606 target_size)); 10607 break; 10608 10609 case offsetof(struct bpf_sock, src_ip4): 10610 *insn++ = BPF_LDX_MEM( 10611 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 10612 bpf_target_off(struct sock_common, skc_rcv_saddr, 10613 sizeof_field(struct sock_common, 10614 skc_rcv_saddr), 10615 target_size)); 10616 break; 10617 10618 case offsetof(struct bpf_sock, dst_ip4): 10619 *insn++ = BPF_LDX_MEM( 10620 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 10621 bpf_target_off(struct sock_common, skc_daddr, 10622 sizeof_field(struct sock_common, 10623 skc_daddr), 10624 target_size)); 10625 break; 10626 10627 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 10628 #if IS_ENABLED(CONFIG_IPV6) 10629 off = si->off; 10630 off -= offsetof(struct bpf_sock, src_ip6[0]); 10631 *insn++ = BPF_LDX_MEM( 10632 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 10633 bpf_target_off( 10634 struct sock_common, 10635 skc_v6_rcv_saddr.s6_addr32[0], 10636 sizeof_field(struct sock_common, 10637 skc_v6_rcv_saddr.s6_addr32[0]), 10638 target_size) + off); 10639 #else 10640 (void)off; 10641 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10642 #endif 10643 break; 10644 10645 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 10646 #if IS_ENABLED(CONFIG_IPV6) 10647 off = si->off; 10648 off -= offsetof(struct bpf_sock, dst_ip6[0]); 10649 *insn++ = BPF_LDX_MEM( 10650 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 10651 bpf_target_off(struct sock_common, 10652 skc_v6_daddr.s6_addr32[0], 10653 sizeof_field(struct sock_common, 10654 skc_v6_daddr.s6_addr32[0]), 10655 target_size) + off); 10656 #else 10657 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10658 *target_size = 4; 10659 #endif 10660 break; 10661 10662 case offsetof(struct bpf_sock, src_port): 10663 *insn++ = BPF_LDX_MEM( 10664 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 10665 si->dst_reg, si->src_reg, 10666 bpf_target_off(struct sock_common, skc_num, 10667 sizeof_field(struct sock_common, 10668 skc_num), 10669 target_size)); 10670 break; 10671 10672 case offsetof(struct bpf_sock, dst_port): 10673 *insn++ = BPF_LDX_MEM( 10674 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 10675 si->dst_reg, si->src_reg, 10676 bpf_target_off(struct sock_common, skc_dport, 10677 sizeof_field(struct sock_common, 10678 skc_dport), 10679 target_size)); 10680 break; 10681 10682 case offsetof(struct bpf_sock, state): 10683 *insn++ = BPF_LDX_MEM( 10684 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 10685 si->dst_reg, si->src_reg, 10686 bpf_target_off(struct sock_common, skc_state, 10687 sizeof_field(struct sock_common, 10688 skc_state), 10689 target_size)); 10690 break; 10691 case offsetof(struct bpf_sock, rx_queue_mapping): 10692 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 10693 *insn++ = BPF_LDX_MEM( 10694 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 10695 si->dst_reg, si->src_reg, 10696 bpf_target_off(struct sock, sk_rx_queue_mapping, 10697 sizeof_field(struct sock, 10698 sk_rx_queue_mapping), 10699 target_size)); 10700 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 10701 1); 10702 *insn++ = BPF_MOV32_IMM(si->dst_reg, -1); 10703 #else 10704 *insn++ = BPF_MOV32_IMM(si->dst_reg, -1); 10705 #endif 10706 *target_size = sizeof_field(struct bpf_sock, rx_queue_mapping); 10707 10708 break; 10709 } 10710 10711 return insn - insn_buf; 10712 } 10713 10714 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 10715 const struct bpf_insn *si, 10716 struct bpf_insn *insn_buf, 10717 struct bpf_prog *prog, u32 *target_size) 10718 { 10719 struct bpf_insn *insn = insn_buf; 10720 10721 switch (si->off) { 10722 case offsetof(struct __sk_buff, ifindex): 10723 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 10724 si->dst_reg, si->src_reg, 10725 offsetof(struct sk_buff, dev)); 10726 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10727 bpf_target_off(struct net_device, ifindex, 4, 10728 target_size)); 10729 break; 10730 default: 10731 return bpf_convert_ctx_access(type, si, insn_buf, prog, 10732 target_size); 10733 } 10734 10735 return insn - insn_buf; 10736 } 10737 10738 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 10739 const struct bpf_insn *si, 10740 struct bpf_insn *insn_buf, 10741 struct bpf_prog *prog, u32 *target_size) 10742 { 10743 struct bpf_insn *insn = insn_buf; 10744 10745 switch (si->off) { 10746 case offsetof(struct xdp_md, data): 10747 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 10748 si->dst_reg, si->src_reg, 10749 offsetof(struct xdp_buff, data)); 10750 break; 10751 case offsetof(struct xdp_md, data_meta): 10752 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 10753 si->dst_reg, si->src_reg, 10754 offsetof(struct xdp_buff, data_meta)); 10755 break; 10756 case offsetof(struct xdp_md, data_end): 10757 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 10758 si->dst_reg, si->src_reg, 10759 offsetof(struct xdp_buff, data_end)); 10760 break; 10761 case offsetof(struct xdp_md, ingress_ifindex): 10762 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 10763 si->dst_reg, si->src_reg, 10764 offsetof(struct xdp_buff, rxq)); 10765 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 10766 si->dst_reg, si->dst_reg, 10767 offsetof(struct xdp_rxq_info, dev)); 10768 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10769 offsetof(struct net_device, ifindex)); 10770 break; 10771 case offsetof(struct xdp_md, rx_queue_index): 10772 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 10773 si->dst_reg, si->src_reg, 10774 offsetof(struct xdp_buff, rxq)); 10775 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10776 offsetof(struct xdp_rxq_info, 10777 queue_index)); 10778 break; 10779 case offsetof(struct xdp_md, egress_ifindex): 10780 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 10781 si->dst_reg, si->src_reg, 10782 offsetof(struct xdp_buff, txq)); 10783 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 10784 si->dst_reg, si->dst_reg, 10785 offsetof(struct xdp_txq_info, dev)); 10786 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10787 offsetof(struct net_device, ifindex)); 10788 break; 10789 } 10790 10791 return insn - insn_buf; 10792 } 10793 10794 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 10795 * context Structure, F is Field in context structure that contains a pointer 10796 * to Nested Structure of type NS that has the field NF. 10797 * 10798 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 10799 * sure that SIZE is not greater than actual size of S.F.NF. 10800 * 10801 * If offset OFF is provided, the load happens from that offset relative to 10802 * offset of NF. 10803 */ 10804 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 10805 do { \ 10806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 10807 si->src_reg, offsetof(S, F)); \ 10808 *insn++ = BPF_LDX_MEM( \ 10809 SIZE, si->dst_reg, si->dst_reg, \ 10810 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 10811 target_size) \ 10812 + OFF); \ 10813 } while (0) 10814 10815 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 10816 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 10817 BPF_FIELD_SIZEOF(NS, NF), 0) 10818 10819 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 10820 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 10821 * 10822 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 10823 * "register" since two registers available in convert_ctx_access are not 10824 * enough: we can't override neither SRC, since it contains value to store, nor 10825 * DST since it contains pointer to context that may be used by later 10826 * instructions. But we need a temporary place to save pointer to nested 10827 * structure whose field we want to store to. 10828 */ 10829 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 10830 do { \ 10831 int tmp_reg = BPF_REG_9; \ 10832 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 10833 --tmp_reg; \ 10834 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 10835 --tmp_reg; \ 10836 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 10837 offsetof(S, TF)); \ 10838 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 10839 si->dst_reg, offsetof(S, F)); \ 10840 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \ 10841 tmp_reg, si->src_reg, \ 10842 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 10843 target_size) \ 10844 + OFF, \ 10845 si->imm); \ 10846 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 10847 offsetof(S, TF)); \ 10848 } while (0) 10849 10850 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 10851 TF) \ 10852 do { \ 10853 if (type == BPF_WRITE) { \ 10854 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 10855 OFF, TF); \ 10856 } else { \ 10857 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 10858 S, NS, F, NF, SIZE, OFF); \ 10859 } \ 10860 } while (0) 10861 10862 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 10863 const struct bpf_insn *si, 10864 struct bpf_insn *insn_buf, 10865 struct bpf_prog *prog, u32 *target_size) 10866 { 10867 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 10868 struct bpf_insn *insn = insn_buf; 10869 10870 switch (si->off) { 10871 case offsetof(struct bpf_sock_addr, user_family): 10872 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10873 struct sockaddr, uaddr, sa_family); 10874 break; 10875 10876 case offsetof(struct bpf_sock_addr, user_ip4): 10877 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10878 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 10879 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 10880 break; 10881 10882 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 10883 off = si->off; 10884 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 10885 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10886 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 10887 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 10888 tmp_reg); 10889 break; 10890 10891 case offsetof(struct bpf_sock_addr, user_port): 10892 /* To get port we need to know sa_family first and then treat 10893 * sockaddr as either sockaddr_in or sockaddr_in6. 10894 * Though we can simplify since port field has same offset and 10895 * size in both structures. 10896 * Here we check this invariant and use just one of the 10897 * structures if it's true. 10898 */ 10899 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 10900 offsetof(struct sockaddr_in6, sin6_port)); 10901 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 10902 sizeof_field(struct sockaddr_in6, sin6_port)); 10903 /* Account for sin6_port being smaller than user_port. */ 10904 port_size = min(port_size, BPF_LDST_BYTES(si)); 10905 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10906 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 10907 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 10908 break; 10909 10910 case offsetof(struct bpf_sock_addr, family): 10911 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10912 struct sock, sk, sk_family); 10913 break; 10914 10915 case offsetof(struct bpf_sock_addr, type): 10916 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10917 struct sock, sk, sk_type); 10918 break; 10919 10920 case offsetof(struct bpf_sock_addr, protocol): 10921 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10922 struct sock, sk, sk_protocol); 10923 break; 10924 10925 case offsetof(struct bpf_sock_addr, msg_src_ip4): 10926 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 10927 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10928 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 10929 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 10930 break; 10931 10932 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 10933 msg_src_ip6[3]): 10934 off = si->off; 10935 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 10936 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 10937 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10938 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 10939 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 10940 break; 10941 case offsetof(struct bpf_sock_addr, sk): 10942 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 10943 si->dst_reg, si->src_reg, 10944 offsetof(struct bpf_sock_addr_kern, sk)); 10945 break; 10946 } 10947 10948 return insn - insn_buf; 10949 } 10950 10951 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 10952 const struct bpf_insn *si, 10953 struct bpf_insn *insn_buf, 10954 struct bpf_prog *prog, 10955 u32 *target_size) 10956 { 10957 struct bpf_insn *insn = insn_buf; 10958 int off; 10959 10960 /* Helper macro for adding read access to tcp_sock or sock fields. */ 10961 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10962 do { \ 10963 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 10964 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10965 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10966 if (si->dst_reg == reg || si->src_reg == reg) \ 10967 reg--; \ 10968 if (si->dst_reg == reg || si->src_reg == reg) \ 10969 reg--; \ 10970 if (si->dst_reg == si->src_reg) { \ 10971 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10972 offsetof(struct bpf_sock_ops_kern, \ 10973 temp)); \ 10974 fullsock_reg = reg; \ 10975 jmp += 2; \ 10976 } \ 10977 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10978 struct bpf_sock_ops_kern, \ 10979 is_locked_tcp_sock), \ 10980 fullsock_reg, si->src_reg, \ 10981 offsetof(struct bpf_sock_ops_kern, \ 10982 is_locked_tcp_sock)); \ 10983 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10984 if (si->dst_reg == si->src_reg) \ 10985 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10986 offsetof(struct bpf_sock_ops_kern, \ 10987 temp)); \ 10988 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10989 struct bpf_sock_ops_kern, sk),\ 10990 si->dst_reg, si->src_reg, \ 10991 offsetof(struct bpf_sock_ops_kern, sk));\ 10992 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 10993 OBJ_FIELD), \ 10994 si->dst_reg, si->dst_reg, \ 10995 offsetof(OBJ, OBJ_FIELD)); \ 10996 if (si->dst_reg == si->src_reg) { \ 10997 *insn++ = BPF_JMP_A(2); \ 10998 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10999 offsetof(struct bpf_sock_ops_kern, \ 11000 temp)); \ 11001 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); \ 11002 } \ 11003 } while (0) 11004 11005 #define SOCK_OPS_GET_SK() \ 11006 do { \ 11007 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 11008 if (si->dst_reg == reg || si->src_reg == reg) \ 11009 reg--; \ 11010 if (si->dst_reg == reg || si->src_reg == reg) \ 11011 reg--; \ 11012 if (si->dst_reg == si->src_reg) { \ 11013 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 11014 offsetof(struct bpf_sock_ops_kern, \ 11015 temp)); \ 11016 fullsock_reg = reg; \ 11017 jmp += 2; \ 11018 } \ 11019 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 11020 struct bpf_sock_ops_kern, \ 11021 is_fullsock), \ 11022 fullsock_reg, si->src_reg, \ 11023 offsetof(struct bpf_sock_ops_kern, \ 11024 is_fullsock)); \ 11025 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 11026 if (si->dst_reg == si->src_reg) \ 11027 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 11028 offsetof(struct bpf_sock_ops_kern, \ 11029 temp)); \ 11030 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 11031 struct bpf_sock_ops_kern, sk),\ 11032 si->dst_reg, si->src_reg, \ 11033 offsetof(struct bpf_sock_ops_kern, sk));\ 11034 if (si->dst_reg == si->src_reg) { \ 11035 *insn++ = BPF_JMP_A(2); \ 11036 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 11037 offsetof(struct bpf_sock_ops_kern, \ 11038 temp)); \ 11039 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); \ 11040 } \ 11041 } while (0) 11042 11043 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 11044 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 11045 11046 /* Helper macro for adding write access to tcp_sock or sock fields. 11047 * The macro is called with two registers, dst_reg which contains a pointer 11048 * to ctx (context) and src_reg which contains the value that should be 11049 * stored. However, we need an additional register since we cannot overwrite 11050 * dst_reg because it may be used later in the program. 11051 * Instead we "borrow" one of the other register. We first save its value 11052 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 11053 * it at the end of the macro. 11054 */ 11055 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 11056 do { \ 11057 int reg = BPF_REG_9; \ 11058 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 11059 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 11060 if (si->dst_reg == reg || si->src_reg == reg) \ 11061 reg--; \ 11062 if (si->dst_reg == reg || si->src_reg == reg) \ 11063 reg--; \ 11064 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 11065 offsetof(struct bpf_sock_ops_kern, \ 11066 temp)); \ 11067 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 11068 struct bpf_sock_ops_kern, \ 11069 is_locked_tcp_sock), \ 11070 reg, si->dst_reg, \ 11071 offsetof(struct bpf_sock_ops_kern, \ 11072 is_locked_tcp_sock)); \ 11073 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 11074 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 11075 struct bpf_sock_ops_kern, sk),\ 11076 reg, si->dst_reg, \ 11077 offsetof(struct bpf_sock_ops_kern, sk));\ 11078 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \ 11079 BPF_MEM | BPF_CLASS(si->code), \ 11080 reg, si->src_reg, \ 11081 offsetof(OBJ, OBJ_FIELD), \ 11082 si->imm); \ 11083 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 11084 offsetof(struct bpf_sock_ops_kern, \ 11085 temp)); \ 11086 } while (0) 11087 11088 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 11089 do { \ 11090 if (TYPE == BPF_WRITE) \ 11091 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 11092 else \ 11093 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 11094 } while (0) 11095 11096 switch (si->off) { 11097 case offsetof(struct bpf_sock_ops, op): 11098 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 11099 op), 11100 si->dst_reg, si->src_reg, 11101 offsetof(struct bpf_sock_ops_kern, op)); 11102 break; 11103 11104 case offsetof(struct bpf_sock_ops, replylong[0]) ... 11105 offsetof(struct bpf_sock_ops, replylong[3]): 11106 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 11107 sizeof_field(struct bpf_sock_ops_kern, reply)); 11108 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 11109 sizeof_field(struct bpf_sock_ops_kern, replylong)); 11110 off = si->off; 11111 off -= offsetof(struct bpf_sock_ops, replylong[0]); 11112 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 11113 if (type == BPF_WRITE) 11114 *insn++ = BPF_EMIT_STORE(BPF_W, si, off); 11115 else 11116 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11117 off); 11118 break; 11119 11120 case offsetof(struct bpf_sock_ops, family): 11121 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 11122 11123 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11124 struct bpf_sock_ops_kern, sk), 11125 si->dst_reg, si->src_reg, 11126 offsetof(struct bpf_sock_ops_kern, sk)); 11127 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 11128 offsetof(struct sock_common, skc_family)); 11129 break; 11130 11131 case offsetof(struct bpf_sock_ops, remote_ip4): 11132 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 11133 11134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11135 struct bpf_sock_ops_kern, sk), 11136 si->dst_reg, si->src_reg, 11137 offsetof(struct bpf_sock_ops_kern, sk)); 11138 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11139 offsetof(struct sock_common, skc_daddr)); 11140 break; 11141 11142 case offsetof(struct bpf_sock_ops, local_ip4): 11143 BUILD_BUG_ON(sizeof_field(struct sock_common, 11144 skc_rcv_saddr) != 4); 11145 11146 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11147 struct bpf_sock_ops_kern, sk), 11148 si->dst_reg, si->src_reg, 11149 offsetof(struct bpf_sock_ops_kern, sk)); 11150 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11151 offsetof(struct sock_common, 11152 skc_rcv_saddr)); 11153 break; 11154 11155 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 11156 offsetof(struct bpf_sock_ops, remote_ip6[3]): 11157 #if IS_ENABLED(CONFIG_IPV6) 11158 BUILD_BUG_ON(sizeof_field(struct sock_common, 11159 skc_v6_daddr.s6_addr32[0]) != 4); 11160 11161 off = si->off; 11162 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 11163 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11164 struct bpf_sock_ops_kern, sk), 11165 si->dst_reg, si->src_reg, 11166 offsetof(struct bpf_sock_ops_kern, sk)); 11167 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11168 offsetof(struct sock_common, 11169 skc_v6_daddr.s6_addr32[0]) + 11170 off); 11171 #else 11172 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11173 #endif 11174 break; 11175 11176 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 11177 offsetof(struct bpf_sock_ops, local_ip6[3]): 11178 #if IS_ENABLED(CONFIG_IPV6) 11179 BUILD_BUG_ON(sizeof_field(struct sock_common, 11180 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 11181 11182 off = si->off; 11183 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 11184 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11185 struct bpf_sock_ops_kern, sk), 11186 si->dst_reg, si->src_reg, 11187 offsetof(struct bpf_sock_ops_kern, sk)); 11188 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11189 offsetof(struct sock_common, 11190 skc_v6_rcv_saddr.s6_addr32[0]) + 11191 off); 11192 #else 11193 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11194 #endif 11195 break; 11196 11197 case offsetof(struct bpf_sock_ops, remote_port): 11198 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 11199 11200 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11201 struct bpf_sock_ops_kern, sk), 11202 si->dst_reg, si->src_reg, 11203 offsetof(struct bpf_sock_ops_kern, sk)); 11204 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 11205 offsetof(struct sock_common, skc_dport)); 11206 #ifndef __BIG_ENDIAN_BITFIELD 11207 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 11208 #endif 11209 break; 11210 11211 case offsetof(struct bpf_sock_ops, local_port): 11212 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 11213 11214 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11215 struct bpf_sock_ops_kern, sk), 11216 si->dst_reg, si->src_reg, 11217 offsetof(struct bpf_sock_ops_kern, sk)); 11218 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 11219 offsetof(struct sock_common, skc_num)); 11220 break; 11221 11222 case offsetof(struct bpf_sock_ops, is_fullsock): 11223 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11224 struct bpf_sock_ops_kern, 11225 is_fullsock), 11226 si->dst_reg, si->src_reg, 11227 offsetof(struct bpf_sock_ops_kern, 11228 is_fullsock)); 11229 break; 11230 11231 case offsetof(struct bpf_sock_ops, state): 11232 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 11233 11234 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11235 struct bpf_sock_ops_kern, sk), 11236 si->dst_reg, si->src_reg, 11237 offsetof(struct bpf_sock_ops_kern, sk)); 11238 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 11239 offsetof(struct sock_common, skc_state)); 11240 break; 11241 11242 case offsetof(struct bpf_sock_ops, rtt_min): 11243 SOCK_OPS_GET_FIELD(rtt_min, rtt_min.s[0].v, struct tcp_sock); 11244 break; 11245 11246 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 11247 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 11248 struct tcp_sock); 11249 break; 11250 11251 case offsetof(struct bpf_sock_ops, sk_txhash): 11252 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 11253 struct sock, type); 11254 break; 11255 case offsetof(struct bpf_sock_ops, snd_cwnd): 11256 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 11257 break; 11258 case offsetof(struct bpf_sock_ops, srtt_us): 11259 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 11260 break; 11261 case offsetof(struct bpf_sock_ops, snd_ssthresh): 11262 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 11263 break; 11264 case offsetof(struct bpf_sock_ops, rcv_nxt): 11265 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 11266 break; 11267 case offsetof(struct bpf_sock_ops, snd_nxt): 11268 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 11269 break; 11270 case offsetof(struct bpf_sock_ops, snd_una): 11271 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 11272 break; 11273 case offsetof(struct bpf_sock_ops, mss_cache): 11274 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 11275 break; 11276 case offsetof(struct bpf_sock_ops, ecn_flags): 11277 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 11278 break; 11279 case offsetof(struct bpf_sock_ops, rate_delivered): 11280 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 11281 break; 11282 case offsetof(struct bpf_sock_ops, rate_interval_us): 11283 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 11284 break; 11285 case offsetof(struct bpf_sock_ops, packets_out): 11286 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 11287 break; 11288 case offsetof(struct bpf_sock_ops, retrans_out): 11289 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 11290 break; 11291 case offsetof(struct bpf_sock_ops, total_retrans): 11292 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 11293 break; 11294 case offsetof(struct bpf_sock_ops, segs_in): 11295 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 11296 break; 11297 case offsetof(struct bpf_sock_ops, data_segs_in): 11298 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 11299 break; 11300 case offsetof(struct bpf_sock_ops, segs_out): 11301 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 11302 break; 11303 case offsetof(struct bpf_sock_ops, data_segs_out): 11304 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 11305 break; 11306 case offsetof(struct bpf_sock_ops, lost_out): 11307 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 11308 break; 11309 case offsetof(struct bpf_sock_ops, sacked_out): 11310 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 11311 break; 11312 case offsetof(struct bpf_sock_ops, bytes_received): 11313 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 11314 break; 11315 case offsetof(struct bpf_sock_ops, bytes_acked): 11316 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 11317 break; 11318 case offsetof(struct bpf_sock_ops, sk): 11319 SOCK_OPS_GET_SK(); 11320 break; 11321 case offsetof(struct bpf_sock_ops, skb_data_end): 11322 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 11323 skb_data_end), 11324 si->dst_reg, si->src_reg, 11325 offsetof(struct bpf_sock_ops_kern, 11326 skb_data_end)); 11327 break; 11328 case offsetof(struct bpf_sock_ops, skb_data): 11329 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 11330 skb), 11331 si->dst_reg, si->src_reg, 11332 offsetof(struct bpf_sock_ops_kern, 11333 skb)); 11334 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11335 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 11336 si->dst_reg, si->dst_reg, 11337 offsetof(struct sk_buff, data)); 11338 break; 11339 case offsetof(struct bpf_sock_ops, skb_len): 11340 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 11341 skb), 11342 si->dst_reg, si->src_reg, 11343 offsetof(struct bpf_sock_ops_kern, 11344 skb)); 11345 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11346 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 11347 si->dst_reg, si->dst_reg, 11348 offsetof(struct sk_buff, len)); 11349 break; 11350 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 11351 off = offsetof(struct sk_buff, cb); 11352 off += offsetof(struct tcp_skb_cb, tcp_flags); 11353 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 11354 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 11355 skb), 11356 si->dst_reg, si->src_reg, 11357 offsetof(struct bpf_sock_ops_kern, 11358 skb)); 11359 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11360 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 11361 tcp_flags), 11362 si->dst_reg, si->dst_reg, off); 11363 break; 11364 case offsetof(struct bpf_sock_ops, skb_hwtstamp): { 11365 struct bpf_insn *jmp_on_null_skb; 11366 11367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 11368 skb), 11369 si->dst_reg, si->src_reg, 11370 offsetof(struct bpf_sock_ops_kern, 11371 skb)); 11372 /* Reserve one insn to test skb == NULL */ 11373 jmp_on_null_skb = insn++; 11374 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn); 11375 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 11376 bpf_target_off(struct skb_shared_info, 11377 hwtstamps, 8, 11378 target_size)); 11379 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 11380 insn - jmp_on_null_skb - 1); 11381 break; 11382 } 11383 } 11384 return insn - insn_buf; 11385 } 11386 11387 /* data_end = skb->data + skb_headlen() */ 11388 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si, 11389 struct bpf_insn *insn) 11390 { 11391 int reg; 11392 int temp_reg_off = offsetof(struct sk_buff, cb) + 11393 offsetof(struct sk_skb_cb, temp_reg); 11394 11395 if (si->src_reg == si->dst_reg) { 11396 /* We need an extra register, choose and save a register. */ 11397 reg = BPF_REG_9; 11398 if (si->src_reg == reg || si->dst_reg == reg) 11399 reg--; 11400 if (si->src_reg == reg || si->dst_reg == reg) 11401 reg--; 11402 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off); 11403 } else { 11404 reg = si->dst_reg; 11405 } 11406 11407 /* reg = skb->data */ 11408 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 11409 reg, si->src_reg, 11410 offsetof(struct sk_buff, data)); 11411 /* AX = skb->len */ 11412 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 11413 BPF_REG_AX, si->src_reg, 11414 offsetof(struct sk_buff, len)); 11415 /* reg = skb->data + skb->len */ 11416 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX); 11417 /* AX = skb->data_len */ 11418 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len), 11419 BPF_REG_AX, si->src_reg, 11420 offsetof(struct sk_buff, data_len)); 11421 11422 /* reg = skb->data + skb->len - skb->data_len */ 11423 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX); 11424 11425 if (si->src_reg == si->dst_reg) { 11426 /* Restore the saved register */ 11427 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg); 11428 *insn++ = BPF_MOV64_REG(si->dst_reg, reg); 11429 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off); 11430 } 11431 11432 return insn; 11433 } 11434 11435 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 11436 const struct bpf_insn *si, 11437 struct bpf_insn *insn_buf, 11438 struct bpf_prog *prog, u32 *target_size) 11439 { 11440 struct bpf_insn *insn = insn_buf; 11441 int off; 11442 11443 switch (si->off) { 11444 case offsetof(struct __sk_buff, data_end): 11445 insn = bpf_convert_data_end_access(si, insn); 11446 break; 11447 case offsetof(struct __sk_buff, cb[0]) ... 11448 offsetofend(struct __sk_buff, cb[4]) - 1: 11449 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20); 11450 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 11451 offsetof(struct sk_skb_cb, data)) % 11452 sizeof(__u64)); 11453 11454 prog->cb_access = 1; 11455 off = si->off; 11456 off -= offsetof(struct __sk_buff, cb[0]); 11457 off += offsetof(struct sk_buff, cb); 11458 off += offsetof(struct sk_skb_cb, data); 11459 if (type == BPF_WRITE) 11460 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off); 11461 else 11462 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 11463 si->src_reg, off); 11464 break; 11465 11466 11467 default: 11468 return bpf_convert_ctx_access(type, si, insn_buf, prog, 11469 target_size); 11470 } 11471 11472 return insn - insn_buf; 11473 } 11474 11475 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 11476 const struct bpf_insn *si, 11477 struct bpf_insn *insn_buf, 11478 struct bpf_prog *prog, u32 *target_size) 11479 { 11480 struct bpf_insn *insn = insn_buf; 11481 #if IS_ENABLED(CONFIG_IPV6) 11482 int off; 11483 #endif 11484 11485 /* convert ctx uses the fact sg element is first in struct */ 11486 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 11487 11488 switch (si->off) { 11489 case offsetof(struct sk_msg_md, data): 11490 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 11491 si->dst_reg, si->src_reg, 11492 offsetof(struct sk_msg, data)); 11493 break; 11494 case offsetof(struct sk_msg_md, data_end): 11495 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 11496 si->dst_reg, si->src_reg, 11497 offsetof(struct sk_msg, data_end)); 11498 break; 11499 case offsetof(struct sk_msg_md, family): 11500 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 11501 11502 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11503 struct sk_msg, sk), 11504 si->dst_reg, si->src_reg, 11505 offsetof(struct sk_msg, sk)); 11506 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 11507 offsetof(struct sock_common, skc_family)); 11508 break; 11509 11510 case offsetof(struct sk_msg_md, remote_ip4): 11511 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 11512 11513 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11514 struct sk_msg, sk), 11515 si->dst_reg, si->src_reg, 11516 offsetof(struct sk_msg, sk)); 11517 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11518 offsetof(struct sock_common, skc_daddr)); 11519 break; 11520 11521 case offsetof(struct sk_msg_md, local_ip4): 11522 BUILD_BUG_ON(sizeof_field(struct sock_common, 11523 skc_rcv_saddr) != 4); 11524 11525 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11526 struct sk_msg, sk), 11527 si->dst_reg, si->src_reg, 11528 offsetof(struct sk_msg, sk)); 11529 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11530 offsetof(struct sock_common, 11531 skc_rcv_saddr)); 11532 break; 11533 11534 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 11535 offsetof(struct sk_msg_md, remote_ip6[3]): 11536 #if IS_ENABLED(CONFIG_IPV6) 11537 BUILD_BUG_ON(sizeof_field(struct sock_common, 11538 skc_v6_daddr.s6_addr32[0]) != 4); 11539 11540 off = si->off; 11541 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 11542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11543 struct sk_msg, sk), 11544 si->dst_reg, si->src_reg, 11545 offsetof(struct sk_msg, sk)); 11546 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11547 offsetof(struct sock_common, 11548 skc_v6_daddr.s6_addr32[0]) + 11549 off); 11550 #else 11551 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11552 #endif 11553 break; 11554 11555 case offsetof(struct sk_msg_md, local_ip6[0]) ... 11556 offsetof(struct sk_msg_md, local_ip6[3]): 11557 #if IS_ENABLED(CONFIG_IPV6) 11558 BUILD_BUG_ON(sizeof_field(struct sock_common, 11559 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 11560 11561 off = si->off; 11562 off -= offsetof(struct sk_msg_md, local_ip6[0]); 11563 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11564 struct sk_msg, sk), 11565 si->dst_reg, si->src_reg, 11566 offsetof(struct sk_msg, sk)); 11567 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 11568 offsetof(struct sock_common, 11569 skc_v6_rcv_saddr.s6_addr32[0]) + 11570 off); 11571 #else 11572 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11573 #endif 11574 break; 11575 11576 case offsetof(struct sk_msg_md, remote_port): 11577 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 11578 11579 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11580 struct sk_msg, sk), 11581 si->dst_reg, si->src_reg, 11582 offsetof(struct sk_msg, sk)); 11583 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 11584 offsetof(struct sock_common, skc_dport)); 11585 #ifndef __BIG_ENDIAN_BITFIELD 11586 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 11587 #endif 11588 break; 11589 11590 case offsetof(struct sk_msg_md, local_port): 11591 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 11592 11593 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 11594 struct sk_msg, sk), 11595 si->dst_reg, si->src_reg, 11596 offsetof(struct sk_msg, sk)); 11597 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 11598 offsetof(struct sock_common, skc_num)); 11599 break; 11600 11601 case offsetof(struct sk_msg_md, size): 11602 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 11603 si->dst_reg, si->src_reg, 11604 offsetof(struct sk_msg_sg, size)); 11605 break; 11606 11607 case offsetof(struct sk_msg_md, sk): 11608 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 11609 si->dst_reg, si->src_reg, 11610 offsetof(struct sk_msg, sk)); 11611 break; 11612 } 11613 11614 return insn - insn_buf; 11615 } 11616 11617 const struct bpf_verifier_ops sk_filter_verifier_ops = { 11618 .get_func_proto = sk_filter_func_proto, 11619 .is_valid_access = sk_filter_is_valid_access, 11620 .convert_ctx_access = bpf_convert_ctx_access, 11621 .gen_ld_abs = bpf_gen_ld_abs, 11622 }; 11623 11624 const struct bpf_prog_ops sk_filter_prog_ops = { 11625 .test_run = bpf_prog_test_run_skb, 11626 }; 11627 11628 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 11629 .get_func_proto = tc_cls_act_func_proto, 11630 .is_valid_access = tc_cls_act_is_valid_access, 11631 .convert_ctx_access = tc_cls_act_convert_ctx_access, 11632 .gen_prologue = tc_cls_act_prologue, 11633 .gen_ld_abs = bpf_gen_ld_abs, 11634 .btf_struct_access = tc_cls_act_btf_struct_access, 11635 }; 11636 11637 const struct bpf_prog_ops tc_cls_act_prog_ops = { 11638 .test_run = bpf_prog_test_run_skb, 11639 }; 11640 11641 const struct bpf_verifier_ops xdp_verifier_ops = { 11642 .get_func_proto = xdp_func_proto, 11643 .is_valid_access = xdp_is_valid_access, 11644 .convert_ctx_access = xdp_convert_ctx_access, 11645 .gen_prologue = bpf_noop_prologue, 11646 .btf_struct_access = xdp_btf_struct_access, 11647 }; 11648 11649 const struct bpf_prog_ops xdp_prog_ops = { 11650 .test_run = bpf_prog_test_run_xdp, 11651 }; 11652 11653 const struct bpf_verifier_ops cg_skb_verifier_ops = { 11654 .get_func_proto = cg_skb_func_proto, 11655 .is_valid_access = cg_skb_is_valid_access, 11656 .convert_ctx_access = bpf_convert_ctx_access, 11657 }; 11658 11659 const struct bpf_prog_ops cg_skb_prog_ops = { 11660 .test_run = bpf_prog_test_run_skb, 11661 }; 11662 11663 const struct bpf_verifier_ops lwt_in_verifier_ops = { 11664 .get_func_proto = lwt_in_func_proto, 11665 .is_valid_access = lwt_is_valid_access, 11666 .convert_ctx_access = bpf_convert_ctx_access, 11667 }; 11668 11669 const struct bpf_prog_ops lwt_in_prog_ops = { 11670 .test_run = bpf_prog_test_run_skb, 11671 }; 11672 11673 const struct bpf_verifier_ops lwt_out_verifier_ops = { 11674 .get_func_proto = lwt_out_func_proto, 11675 .is_valid_access = lwt_is_valid_access, 11676 .convert_ctx_access = bpf_convert_ctx_access, 11677 }; 11678 11679 const struct bpf_prog_ops lwt_out_prog_ops = { 11680 .test_run = bpf_prog_test_run_skb, 11681 }; 11682 11683 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 11684 .get_func_proto = lwt_xmit_func_proto, 11685 .is_valid_access = lwt_is_valid_access, 11686 .convert_ctx_access = bpf_convert_ctx_access, 11687 .gen_prologue = tc_cls_act_prologue, 11688 }; 11689 11690 const struct bpf_prog_ops lwt_xmit_prog_ops = { 11691 .test_run = bpf_prog_test_run_skb, 11692 }; 11693 11694 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 11695 .get_func_proto = lwt_seg6local_func_proto, 11696 .is_valid_access = lwt_is_valid_access, 11697 .convert_ctx_access = bpf_convert_ctx_access, 11698 }; 11699 11700 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 11701 }; 11702 11703 const struct bpf_verifier_ops cg_sock_verifier_ops = { 11704 .get_func_proto = sock_filter_func_proto, 11705 .is_valid_access = sock_filter_is_valid_access, 11706 .convert_ctx_access = bpf_sock_convert_ctx_access, 11707 }; 11708 11709 const struct bpf_prog_ops cg_sock_prog_ops = { 11710 }; 11711 11712 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 11713 .get_func_proto = sock_addr_func_proto, 11714 .is_valid_access = sock_addr_is_valid_access, 11715 .convert_ctx_access = sock_addr_convert_ctx_access, 11716 }; 11717 11718 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 11719 }; 11720 11721 const struct bpf_verifier_ops sock_ops_verifier_ops = { 11722 .get_func_proto = sock_ops_func_proto, 11723 .is_valid_access = sock_ops_is_valid_access, 11724 .convert_ctx_access = sock_ops_convert_ctx_access, 11725 }; 11726 11727 const struct bpf_prog_ops sock_ops_prog_ops = { 11728 }; 11729 11730 const struct bpf_verifier_ops sk_skb_verifier_ops = { 11731 .get_func_proto = sk_skb_func_proto, 11732 .is_valid_access = sk_skb_is_valid_access, 11733 .convert_ctx_access = sk_skb_convert_ctx_access, 11734 .gen_prologue = sk_skb_prologue, 11735 }; 11736 11737 const struct bpf_prog_ops sk_skb_prog_ops = { 11738 }; 11739 11740 const struct bpf_verifier_ops sk_msg_verifier_ops = { 11741 .get_func_proto = sk_msg_func_proto, 11742 .is_valid_access = sk_msg_is_valid_access, 11743 .convert_ctx_access = sk_msg_convert_ctx_access, 11744 .gen_prologue = bpf_noop_prologue, 11745 }; 11746 11747 const struct bpf_prog_ops sk_msg_prog_ops = { 11748 }; 11749 11750 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 11751 .get_func_proto = flow_dissector_func_proto, 11752 .is_valid_access = flow_dissector_is_valid_access, 11753 .convert_ctx_access = flow_dissector_convert_ctx_access, 11754 }; 11755 11756 const struct bpf_prog_ops flow_dissector_prog_ops = { 11757 .test_run = bpf_prog_test_run_flow_dissector, 11758 }; 11759 11760 int sk_detach_filter(struct sock *sk) 11761 { 11762 int ret = -ENOENT; 11763 struct sk_filter *filter; 11764 11765 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 11766 return -EPERM; 11767 11768 filter = rcu_dereference_protected(sk->sk_filter, 11769 lockdep_sock_is_held(sk)); 11770 if (filter) { 11771 RCU_INIT_POINTER(sk->sk_filter, NULL); 11772 sk_filter_uncharge(sk, filter); 11773 ret = 0; 11774 } 11775 11776 return ret; 11777 } 11778 EXPORT_SYMBOL_GPL(sk_detach_filter); 11779 11780 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len) 11781 { 11782 struct sock_fprog_kern *fprog; 11783 struct sk_filter *filter; 11784 int ret = 0; 11785 11786 sockopt_lock_sock(sk); 11787 filter = rcu_dereference_protected(sk->sk_filter, 11788 lockdep_sock_is_held(sk)); 11789 if (!filter) 11790 goto out; 11791 11792 /* We're copying the filter that has been originally attached, 11793 * so no conversion/decode needed anymore. eBPF programs that 11794 * have no original program cannot be dumped through this. 11795 */ 11796 ret = -EACCES; 11797 fprog = filter->prog->orig_prog; 11798 if (!fprog) 11799 goto out; 11800 11801 ret = fprog->len; 11802 if (!len) 11803 /* User space only enquires number of filter blocks. */ 11804 goto out; 11805 11806 ret = -EINVAL; 11807 if (len < fprog->len) 11808 goto out; 11809 11810 ret = -EFAULT; 11811 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog))) 11812 goto out; 11813 11814 /* Instead of bytes, the API requests to return the number 11815 * of filter blocks. 11816 */ 11817 ret = fprog->len; 11818 out: 11819 sockopt_release_sock(sk); 11820 return ret; 11821 } 11822 11823 #ifdef CONFIG_INET 11824 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 11825 struct sock_reuseport *reuse, 11826 struct sock *sk, struct sk_buff *skb, 11827 struct sock *migrating_sk, 11828 u32 hash) 11829 { 11830 reuse_kern->skb = skb; 11831 reuse_kern->sk = sk; 11832 reuse_kern->selected_sk = NULL; 11833 reuse_kern->migrating_sk = migrating_sk; 11834 reuse_kern->data_end = skb->data + skb_headlen(skb); 11835 reuse_kern->hash = hash; 11836 reuse_kern->reuseport_id = reuse->reuseport_id; 11837 reuse_kern->bind_inany = reuse->bind_inany; 11838 } 11839 11840 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 11841 struct bpf_prog *prog, struct sk_buff *skb, 11842 struct sock *migrating_sk, 11843 u32 hash) 11844 { 11845 struct sk_reuseport_kern reuse_kern; 11846 enum sk_action action; 11847 11848 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash); 11849 action = bpf_prog_run(prog, &reuse_kern); 11850 11851 if (action == SK_PASS) 11852 return reuse_kern.selected_sk; 11853 else 11854 return ERR_PTR(-ECONNREFUSED); 11855 } 11856 11857 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 11858 struct bpf_map *, map, void *, key, u32, flags) 11859 { 11860 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 11861 struct sock_reuseport *reuse; 11862 struct sock *selected_sk; 11863 int err; 11864 11865 selected_sk = map->ops->map_lookup_elem(map, key); 11866 if (!selected_sk) 11867 return -ENOENT; 11868 11869 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 11870 if (!reuse) { 11871 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 11872 * The only (!reuse) case here is - the sk has already been 11873 * unhashed (e.g. by close()), so treat it as -ENOENT. 11874 * 11875 * Other maps (e.g. sock_map) do not provide this guarantee and 11876 * the sk may never be in the reuseport group to begin with. 11877 */ 11878 err = is_sockarray ? -ENOENT : -EINVAL; 11879 goto error; 11880 } 11881 11882 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 11883 struct sock *sk = reuse_kern->sk; 11884 11885 if (sk->sk_protocol != selected_sk->sk_protocol) { 11886 err = -EPROTOTYPE; 11887 } else if (sk->sk_family != selected_sk->sk_family) { 11888 err = -EAFNOSUPPORT; 11889 } else { 11890 /* Catch all. Likely bound to a different sockaddr. */ 11891 err = -EBADFD; 11892 } 11893 goto error; 11894 } 11895 11896 reuse_kern->selected_sk = selected_sk; 11897 11898 return 0; 11899 error: 11900 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 11901 if (sk_is_refcounted(selected_sk)) 11902 sock_put(selected_sk); 11903 11904 return err; 11905 } 11906 11907 static const struct bpf_func_proto sk_select_reuseport_proto = { 11908 .func = sk_select_reuseport, 11909 .gpl_only = false, 11910 .ret_type = RET_INTEGER, 11911 .arg1_type = ARG_PTR_TO_CTX, 11912 .arg2_type = ARG_CONST_MAP_PTR, 11913 .arg3_type = ARG_PTR_TO_MAP_KEY, 11914 .arg4_type = ARG_ANYTHING, 11915 }; 11916 11917 BPF_CALL_4(sk_reuseport_load_bytes, 11918 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 11919 void *, to, u32, len) 11920 { 11921 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 11922 } 11923 11924 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 11925 .func = sk_reuseport_load_bytes, 11926 .gpl_only = false, 11927 .ret_type = RET_INTEGER, 11928 .arg1_type = ARG_PTR_TO_CTX, 11929 .arg2_type = ARG_ANYTHING, 11930 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 11931 .arg4_type = ARG_MEM_SIZE, 11932 }; 11933 11934 BPF_CALL_5(sk_reuseport_load_bytes_relative, 11935 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 11936 void *, to, u32, len, u32, start_header) 11937 { 11938 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 11939 len, start_header); 11940 } 11941 11942 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 11943 .func = sk_reuseport_load_bytes_relative, 11944 .gpl_only = false, 11945 .ret_type = RET_INTEGER, 11946 .arg1_type = ARG_PTR_TO_CTX, 11947 .arg2_type = ARG_ANYTHING, 11948 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 11949 .arg4_type = ARG_MEM_SIZE, 11950 .arg5_type = ARG_ANYTHING, 11951 }; 11952 11953 static const struct bpf_func_proto * 11954 sk_reuseport_func_proto(enum bpf_func_id func_id, 11955 const struct bpf_prog *prog) 11956 { 11957 switch (func_id) { 11958 case BPF_FUNC_sk_select_reuseport: 11959 return &sk_select_reuseport_proto; 11960 case BPF_FUNC_skb_load_bytes: 11961 return &sk_reuseport_load_bytes_proto; 11962 case BPF_FUNC_skb_load_bytes_relative: 11963 return &sk_reuseport_load_bytes_relative_proto; 11964 case BPF_FUNC_get_socket_cookie: 11965 return &bpf_get_socket_ptr_cookie_proto; 11966 case BPF_FUNC_ktime_get_coarse_ns: 11967 return &bpf_ktime_get_coarse_ns_proto; 11968 default: 11969 return bpf_base_func_proto(func_id, prog); 11970 } 11971 } 11972 11973 static bool 11974 sk_reuseport_is_valid_access(int off, int size, 11975 enum bpf_access_type type, 11976 const struct bpf_prog *prog, 11977 struct bpf_insn_access_aux *info) 11978 { 11979 const u32 size_default = sizeof(__u32); 11980 11981 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 11982 off % size || type != BPF_READ) 11983 return false; 11984 11985 switch (off) { 11986 case offsetof(struct sk_reuseport_md, data): 11987 info->reg_type = PTR_TO_PACKET; 11988 return size == sizeof(__u64); 11989 11990 case offsetof(struct sk_reuseport_md, data_end): 11991 info->reg_type = PTR_TO_PACKET_END; 11992 return size == sizeof(__u64); 11993 11994 case offsetof(struct sk_reuseport_md, hash): 11995 return size == size_default; 11996 11997 case offsetof(struct sk_reuseport_md, sk): 11998 info->reg_type = PTR_TO_SOCKET; 11999 return size == sizeof(__u64); 12000 12001 case offsetof(struct sk_reuseport_md, migrating_sk): 12002 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 12003 return size == sizeof(__u64); 12004 12005 /* Fields that allow narrowing */ 12006 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 12007 if (size < sizeof_field(struct sk_buff, protocol)) 12008 return false; 12009 fallthrough; 12010 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 12011 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 12012 case bpf_ctx_range(struct sk_reuseport_md, len): 12013 bpf_ctx_record_field_size(info, size_default); 12014 return bpf_ctx_narrow_access_ok(off, size, size_default); 12015 12016 default: 12017 return false; 12018 } 12019 } 12020 12021 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 12022 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 12023 si->dst_reg, si->src_reg, \ 12024 bpf_target_off(struct sk_reuseport_kern, F, \ 12025 sizeof_field(struct sk_reuseport_kern, F), \ 12026 target_size)); \ 12027 }) 12028 12029 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 12030 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 12031 struct sk_buff, \ 12032 skb, \ 12033 SKB_FIELD) 12034 12035 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 12036 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 12037 struct sock, \ 12038 sk, \ 12039 SK_FIELD) 12040 12041 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 12042 const struct bpf_insn *si, 12043 struct bpf_insn *insn_buf, 12044 struct bpf_prog *prog, 12045 u32 *target_size) 12046 { 12047 struct bpf_insn *insn = insn_buf; 12048 12049 switch (si->off) { 12050 case offsetof(struct sk_reuseport_md, data): 12051 SK_REUSEPORT_LOAD_SKB_FIELD(data); 12052 break; 12053 12054 case offsetof(struct sk_reuseport_md, len): 12055 SK_REUSEPORT_LOAD_SKB_FIELD(len); 12056 break; 12057 12058 case offsetof(struct sk_reuseport_md, eth_protocol): 12059 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 12060 break; 12061 12062 case offsetof(struct sk_reuseport_md, ip_protocol): 12063 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 12064 break; 12065 12066 case offsetof(struct sk_reuseport_md, data_end): 12067 SK_REUSEPORT_LOAD_FIELD(data_end); 12068 break; 12069 12070 case offsetof(struct sk_reuseport_md, hash): 12071 SK_REUSEPORT_LOAD_FIELD(hash); 12072 break; 12073 12074 case offsetof(struct sk_reuseport_md, bind_inany): 12075 SK_REUSEPORT_LOAD_FIELD(bind_inany); 12076 break; 12077 12078 case offsetof(struct sk_reuseport_md, sk): 12079 SK_REUSEPORT_LOAD_FIELD(sk); 12080 break; 12081 12082 case offsetof(struct sk_reuseport_md, migrating_sk): 12083 SK_REUSEPORT_LOAD_FIELD(migrating_sk); 12084 break; 12085 } 12086 12087 return insn - insn_buf; 12088 } 12089 12090 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 12091 .get_func_proto = sk_reuseport_func_proto, 12092 .is_valid_access = sk_reuseport_is_valid_access, 12093 .convert_ctx_access = sk_reuseport_convert_ctx_access, 12094 }; 12095 12096 const struct bpf_prog_ops sk_reuseport_prog_ops = { 12097 }; 12098 12099 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 12100 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 12101 12102 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 12103 struct sock *, sk, u64, flags) 12104 { 12105 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 12106 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 12107 return -EINVAL; 12108 if (unlikely(sk && sk_is_refcounted(sk))) 12109 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 12110 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN)) 12111 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */ 12112 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE)) 12113 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */ 12114 12115 /* Check if socket is suitable for packet L3/L4 protocol */ 12116 if (sk && sk->sk_protocol != ctx->protocol) 12117 return -EPROTOTYPE; 12118 if (sk && sk->sk_family != ctx->family && 12119 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 12120 return -EAFNOSUPPORT; 12121 12122 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 12123 return -EEXIST; 12124 12125 /* Select socket as lookup result */ 12126 ctx->selected_sk = sk; 12127 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 12128 return 0; 12129 } 12130 12131 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 12132 .func = bpf_sk_lookup_assign, 12133 .gpl_only = false, 12134 .ret_type = RET_INTEGER, 12135 .arg1_type = ARG_PTR_TO_CTX, 12136 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 12137 .arg3_type = ARG_ANYTHING, 12138 }; 12139 12140 static const struct bpf_func_proto * 12141 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 12142 { 12143 switch (func_id) { 12144 case BPF_FUNC_perf_event_output: 12145 return &bpf_event_output_data_proto; 12146 case BPF_FUNC_sk_assign: 12147 return &bpf_sk_lookup_assign_proto; 12148 case BPF_FUNC_sk_release: 12149 return &bpf_sk_release_proto; 12150 default: 12151 return bpf_sk_base_func_proto(func_id, prog); 12152 } 12153 } 12154 12155 static bool sk_lookup_is_valid_access(int off, int size, 12156 enum bpf_access_type type, 12157 const struct bpf_prog *prog, 12158 struct bpf_insn_access_aux *info) 12159 { 12160 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 12161 return false; 12162 if (off % size != 0) 12163 return false; 12164 if (type != BPF_READ) 12165 return false; 12166 12167 switch (off) { 12168 case bpf_ctx_range_ptr(struct bpf_sk_lookup, sk): 12169 info->reg_type = PTR_TO_SOCKET_OR_NULL; 12170 return size == sizeof(__u64); 12171 12172 case bpf_ctx_range(struct bpf_sk_lookup, family): 12173 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 12174 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 12175 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 12176 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 12177 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 12178 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 12179 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex): 12180 bpf_ctx_record_field_size(info, sizeof(__u32)); 12181 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 12182 12183 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 12184 /* Allow 4-byte access to 2-byte field for backward compatibility */ 12185 if (size == sizeof(__u32)) 12186 return true; 12187 bpf_ctx_record_field_size(info, sizeof(__be16)); 12188 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16)); 12189 12190 case offsetofend(struct bpf_sk_lookup, remote_port) ... 12191 offsetof(struct bpf_sk_lookup, local_ip4) - 1: 12192 /* Allow access to zero padding for backward compatibility */ 12193 bpf_ctx_record_field_size(info, sizeof(__u16)); 12194 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16)); 12195 12196 default: 12197 return false; 12198 } 12199 } 12200 12201 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 12202 const struct bpf_insn *si, 12203 struct bpf_insn *insn_buf, 12204 struct bpf_prog *prog, 12205 u32 *target_size) 12206 { 12207 struct bpf_insn *insn = insn_buf; 12208 12209 switch (si->off) { 12210 case offsetof(struct bpf_sk_lookup, sk): 12211 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 12212 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 12213 break; 12214 12215 case offsetof(struct bpf_sk_lookup, family): 12216 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 12217 bpf_target_off(struct bpf_sk_lookup_kern, 12218 family, 2, target_size)); 12219 break; 12220 12221 case offsetof(struct bpf_sk_lookup, protocol): 12222 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 12223 bpf_target_off(struct bpf_sk_lookup_kern, 12224 protocol, 2, target_size)); 12225 break; 12226 12227 case offsetof(struct bpf_sk_lookup, remote_ip4): 12228 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 12229 bpf_target_off(struct bpf_sk_lookup_kern, 12230 v4.saddr, 4, target_size)); 12231 break; 12232 12233 case offsetof(struct bpf_sk_lookup, local_ip4): 12234 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 12235 bpf_target_off(struct bpf_sk_lookup_kern, 12236 v4.daddr, 4, target_size)); 12237 break; 12238 12239 case bpf_ctx_range_till(struct bpf_sk_lookup, 12240 remote_ip6[0], remote_ip6[3]): { 12241 #if IS_ENABLED(CONFIG_IPV6) 12242 int off = si->off; 12243 12244 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 12245 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 12246 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 12247 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 12248 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 12249 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 12250 #else 12251 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 12252 #endif 12253 break; 12254 } 12255 case bpf_ctx_range_till(struct bpf_sk_lookup, 12256 local_ip6[0], local_ip6[3]): { 12257 #if IS_ENABLED(CONFIG_IPV6) 12258 int off = si->off; 12259 12260 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 12261 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 12262 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 12263 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 12264 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 12265 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 12266 #else 12267 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 12268 #endif 12269 break; 12270 } 12271 case offsetof(struct bpf_sk_lookup, remote_port): 12272 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 12273 bpf_target_off(struct bpf_sk_lookup_kern, 12274 sport, 2, target_size)); 12275 break; 12276 12277 case offsetofend(struct bpf_sk_lookup, remote_port): 12278 *target_size = 2; 12279 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 12280 break; 12281 12282 case offsetof(struct bpf_sk_lookup, local_port): 12283 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 12284 bpf_target_off(struct bpf_sk_lookup_kern, 12285 dport, 2, target_size)); 12286 break; 12287 12288 case offsetof(struct bpf_sk_lookup, ingress_ifindex): 12289 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 12290 bpf_target_off(struct bpf_sk_lookup_kern, 12291 ingress_ifindex, 4, target_size)); 12292 break; 12293 } 12294 12295 return insn - insn_buf; 12296 } 12297 12298 const struct bpf_prog_ops sk_lookup_prog_ops = { 12299 .test_run = bpf_prog_test_run_sk_lookup, 12300 }; 12301 12302 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 12303 .get_func_proto = sk_lookup_func_proto, 12304 .is_valid_access = sk_lookup_is_valid_access, 12305 .convert_ctx_access = sk_lookup_convert_ctx_access, 12306 }; 12307 12308 #endif /* CONFIG_INET */ 12309 12310 DEFINE_BPF_DISPATCHER(xdp) 12311 12312 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 12313 { 12314 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 12315 } 12316 12317 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE) 12318 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 12319 BTF_SOCK_TYPE_xxx 12320 #undef BTF_SOCK_TYPE 12321 12322 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 12323 { 12324 /* tcp6_sock type is not generated in dwarf and hence btf, 12325 * trigger an explicit type generation here. 12326 */ 12327 BTF_TYPE_EMIT(struct tcp6_sock); 12328 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 12329 sk->sk_type == SOCK_STREAM && sk->sk_family == AF_INET6) 12330 return (unsigned long)sk; 12331 12332 return (unsigned long)NULL; 12333 } 12334 12335 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 12336 .func = bpf_skc_to_tcp6_sock, 12337 .gpl_only = false, 12338 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12339 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 12340 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 12341 }; 12342 12343 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 12344 { 12345 if (sk && sk_fullsock(sk) && sk_is_tcp(sk)) 12346 return (unsigned long)sk; 12347 12348 return (unsigned long)NULL; 12349 } 12350 12351 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 12352 .func = bpf_skc_to_tcp_sock, 12353 .gpl_only = false, 12354 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12355 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 12356 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 12357 }; 12358 12359 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 12360 { 12361 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 12362 * generated if CONFIG_INET=n. Trigger an explicit generation here. 12363 */ 12364 BTF_TYPE_EMIT(struct inet_timewait_sock); 12365 BTF_TYPE_EMIT(struct tcp_timewait_sock); 12366 12367 #ifdef CONFIG_INET 12368 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 12369 return (unsigned long)sk; 12370 #endif 12371 12372 #if IS_ENABLED(CONFIG_IPV6) 12373 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 12374 return (unsigned long)sk; 12375 #endif 12376 12377 return (unsigned long)NULL; 12378 } 12379 12380 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 12381 .func = bpf_skc_to_tcp_timewait_sock, 12382 .gpl_only = false, 12383 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12384 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 12385 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 12386 }; 12387 12388 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 12389 { 12390 /* Ensure BTF for the structs exists even with CONFIG_INET=n. */ 12391 BTF_TYPE_EMIT(struct inet_request_sock); 12392 BTF_TYPE_EMIT(struct tcp_request_sock); 12393 12394 #ifdef CONFIG_INET 12395 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 12396 return (unsigned long)sk; 12397 #endif 12398 12399 #if IS_ENABLED(CONFIG_IPV6) 12400 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 12401 return (unsigned long)sk; 12402 #endif 12403 12404 return (unsigned long)NULL; 12405 } 12406 12407 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 12408 .func = bpf_skc_to_tcp_request_sock, 12409 .gpl_only = false, 12410 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12411 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 12412 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 12413 }; 12414 12415 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 12416 { 12417 /* udp6_sock type is not generated in dwarf and hence btf, 12418 * trigger an explicit type generation here. 12419 */ 12420 BTF_TYPE_EMIT(struct udp6_sock); 12421 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 12422 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 12423 return (unsigned long)sk; 12424 12425 return (unsigned long)NULL; 12426 } 12427 12428 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 12429 .func = bpf_skc_to_udp6_sock, 12430 .gpl_only = false, 12431 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12432 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 12433 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 12434 }; 12435 12436 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk) 12437 { 12438 /* unix_sock type is not generated in dwarf and hence btf, 12439 * trigger an explicit type generation here. 12440 */ 12441 BTF_TYPE_EMIT(struct unix_sock); 12442 if (sk && sk_is_unix(sk)) 12443 return (unsigned long)sk; 12444 12445 return (unsigned long)NULL; 12446 } 12447 12448 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { 12449 .func = bpf_skc_to_unix_sock, 12450 .gpl_only = false, 12451 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12452 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 12453 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX], 12454 }; 12455 12456 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) 12457 { 12458 BTF_TYPE_EMIT(struct mptcp_sock); 12459 return (unsigned long)bpf_mptcp_sock_from_subflow(sk); 12460 } 12461 12462 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { 12463 .func = bpf_skc_to_mptcp_sock, 12464 .gpl_only = false, 12465 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12466 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 12467 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP], 12468 }; 12469 12470 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 12471 { 12472 return (unsigned long)sock_from_file(file); 12473 } 12474 12475 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 12476 BTF_ID(struct, socket) 12477 BTF_ID(struct, file) 12478 12479 const struct bpf_func_proto bpf_sock_from_file_proto = { 12480 .func = bpf_sock_from_file, 12481 .gpl_only = false, 12482 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 12483 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 12484 .arg1_type = ARG_PTR_TO_BTF_ID, 12485 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 12486 }; 12487 12488 static const struct bpf_func_proto * 12489 bpf_sk_base_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 12490 { 12491 const struct bpf_func_proto *func; 12492 12493 switch (func_id) { 12494 case BPF_FUNC_skc_to_tcp6_sock: 12495 func = &bpf_skc_to_tcp6_sock_proto; 12496 break; 12497 case BPF_FUNC_skc_to_tcp_sock: 12498 func = &bpf_skc_to_tcp_sock_proto; 12499 break; 12500 case BPF_FUNC_skc_to_tcp_timewait_sock: 12501 func = &bpf_skc_to_tcp_timewait_sock_proto; 12502 break; 12503 case BPF_FUNC_skc_to_tcp_request_sock: 12504 func = &bpf_skc_to_tcp_request_sock_proto; 12505 break; 12506 case BPF_FUNC_skc_to_udp6_sock: 12507 func = &bpf_skc_to_udp6_sock_proto; 12508 break; 12509 case BPF_FUNC_skc_to_unix_sock: 12510 func = &bpf_skc_to_unix_sock_proto; 12511 break; 12512 case BPF_FUNC_skc_to_mptcp_sock: 12513 func = &bpf_skc_to_mptcp_sock_proto; 12514 break; 12515 case BPF_FUNC_ktime_get_coarse_ns: 12516 return &bpf_ktime_get_coarse_ns_proto; 12517 default: 12518 return bpf_base_func_proto(func_id, prog); 12519 } 12520 12521 if (!bpf_token_capable(prog->aux->token, CAP_PERFMON)) 12522 return NULL; 12523 12524 return func; 12525 } 12526 12527 /** 12528 * bpf_skb_meta_pointer() - Gets a mutable pointer within the skb metadata area. 12529 * @skb: socket buffer carrying the metadata 12530 * @offset: offset into the metadata area, must be <= skb_metadata_len() 12531 */ 12532 void *bpf_skb_meta_pointer(struct sk_buff *skb, u32 offset) 12533 { 12534 return skb_metadata_end(skb) - skb_metadata_len(skb) + offset; 12535 } 12536 12537 int __bpf_skb_meta_store_bytes(struct sk_buff *skb, u32 offset, 12538 const void *from, u32 len, u64 flags) 12539 { 12540 if (unlikely(flags)) 12541 return -EINVAL; 12542 if (unlikely(bpf_try_make_writable_from(skb, 0, &from))) 12543 return -EFAULT; 12544 12545 memmove(bpf_skb_meta_pointer(skb, offset), from, len); 12546 return 0; 12547 } 12548 12549 __bpf_kfunc_start_defs(); 12550 __bpf_kfunc int bpf_dynptr_from_skb(struct __sk_buff *s, u64 flags, 12551 struct bpf_dynptr *ptr__uninit) 12552 { 12553 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit; 12554 struct sk_buff *skb = (struct sk_buff *)s; 12555 12556 if (flags) { 12557 bpf_dynptr_set_null(ptr); 12558 return -EINVAL; 12559 } 12560 12561 bpf_dynptr_init(ptr, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len); 12562 12563 return 0; 12564 } 12565 12566 /** 12567 * bpf_dynptr_from_skb_meta() - Initialize a dynptr to the skb metadata area. 12568 * @skb_: socket buffer carrying the metadata 12569 * @flags: future use, must be zero 12570 * @ptr__uninit: dynptr to initialize 12571 * 12572 * Set up a dynptr for access to the metadata area earlier allocated from the 12573 * XDP context with bpf_xdp_adjust_meta(). Serves as an alternative to 12574 * &__sk_buff->data_meta. 12575 * 12576 * Return: 12577 * * %0 - dynptr ready to use 12578 * * %-EINVAL - invalid flags, dynptr set to null 12579 */ 12580 __bpf_kfunc int bpf_dynptr_from_skb_meta(struct __sk_buff *skb_, u64 flags, 12581 struct bpf_dynptr *ptr__uninit) 12582 { 12583 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit; 12584 struct sk_buff *skb = (struct sk_buff *)skb_; 12585 12586 if (flags) { 12587 bpf_dynptr_set_null(ptr); 12588 return -EINVAL; 12589 } 12590 12591 bpf_dynptr_init(ptr, skb, BPF_DYNPTR_TYPE_SKB_META, 0, skb_metadata_len(skb)); 12592 12593 return 0; 12594 } 12595 12596 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_md *x, u64 flags, 12597 struct bpf_dynptr *ptr__uninit) 12598 { 12599 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit; 12600 struct xdp_buff *xdp = (struct xdp_buff *)x; 12601 12602 if (flags) { 12603 bpf_dynptr_set_null(ptr); 12604 return -EINVAL; 12605 } 12606 12607 bpf_dynptr_init(ptr, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp)); 12608 12609 return 0; 12610 } 12611 12612 __bpf_kfunc int bpf_sock_addr_set_sun_path(struct bpf_sock_addr_kern *sa_kern, 12613 const u8 *sun_path, u32 sun_path__sz) 12614 { 12615 struct sockaddr_un *un; 12616 12617 if (sa_kern->sk->sk_family != AF_UNIX) 12618 return -EINVAL; 12619 12620 /* We do not allow changing the address to unnamed or larger than the 12621 * maximum allowed address size for a unix sockaddr. 12622 */ 12623 if (sun_path__sz == 0 || sun_path__sz > UNIX_PATH_MAX) 12624 return -EINVAL; 12625 12626 un = (struct sockaddr_un *)sa_kern->uaddr; 12627 memcpy(un->sun_path, sun_path, sun_path__sz); 12628 sa_kern->uaddrlen = offsetof(struct sockaddr_un, sun_path) + sun_path__sz; 12629 12630 return 0; 12631 } 12632 12633 __bpf_kfunc int bpf_sk_assign_tcp_reqsk(struct __sk_buff *s, struct sock *sk, 12634 struct bpf_tcp_req_attrs *attrs, int attrs__sz) 12635 { 12636 #if IS_ENABLED(CONFIG_SYN_COOKIES) 12637 struct sk_buff *skb = (struct sk_buff *)s; 12638 const struct request_sock_ops *ops; 12639 struct inet_request_sock *ireq; 12640 struct tcp_request_sock *treq; 12641 struct request_sock *req; 12642 struct net *net; 12643 __u16 min_mss; 12644 u32 tsoff = 0; 12645 12646 if (attrs__sz != sizeof(*attrs) || 12647 attrs->reserved[0] || attrs->reserved[1] || attrs->reserved[2]) 12648 return -EINVAL; 12649 12650 if (!skb_at_tc_ingress(skb)) 12651 return -EINVAL; 12652 12653 net = dev_net(skb->dev); 12654 if (net != sock_net(sk)) 12655 return -ENETUNREACH; 12656 12657 switch (skb->protocol) { 12658 case htons(ETH_P_IP): 12659 ops = &tcp_request_sock_ops; 12660 min_mss = 536; 12661 break; 12662 #if IS_ENABLED(CONFIG_IPV6) 12663 case htons(ETH_P_IPV6): 12664 ops = &tcp6_request_sock_ops; 12665 min_mss = IPV6_MIN_MTU - 60; 12666 break; 12667 #endif 12668 default: 12669 return -EINVAL; 12670 } 12671 12672 if (sk->sk_type != SOCK_STREAM || sk->sk_state != TCP_LISTEN || 12673 sk_is_mptcp(sk)) 12674 return -EINVAL; 12675 12676 if (attrs->mss < min_mss) 12677 return -EINVAL; 12678 12679 if (attrs->wscale_ok) { 12680 if (!READ_ONCE(net->ipv4.sysctl_tcp_window_scaling)) 12681 return -EINVAL; 12682 12683 if (attrs->snd_wscale > TCP_MAX_WSCALE || 12684 attrs->rcv_wscale > TCP_MAX_WSCALE) 12685 return -EINVAL; 12686 } 12687 12688 if (attrs->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack)) 12689 return -EINVAL; 12690 12691 if (attrs->tstamp_ok) { 12692 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps)) 12693 return -EINVAL; 12694 12695 tsoff = attrs->rcv_tsecr - tcp_ns_to_ts(attrs->usec_ts_ok, tcp_clock_ns()); 12696 } 12697 12698 req = inet_reqsk_alloc(ops, sk, false); 12699 if (!req) 12700 return -ENOMEM; 12701 12702 ireq = inet_rsk(req); 12703 treq = tcp_rsk(req); 12704 12705 req->rsk_listener = sk; 12706 req->syncookie = 1; 12707 req->mss = attrs->mss; 12708 req->ts_recent = attrs->rcv_tsval; 12709 12710 ireq->snd_wscale = attrs->snd_wscale; 12711 ireq->rcv_wscale = attrs->rcv_wscale; 12712 ireq->tstamp_ok = !!attrs->tstamp_ok; 12713 ireq->sack_ok = !!attrs->sack_ok; 12714 ireq->wscale_ok = !!attrs->wscale_ok; 12715 ireq->ecn_ok = !!attrs->ecn_ok; 12716 12717 treq->req_usec_ts = !!attrs->usec_ts_ok; 12718 treq->ts_off = tsoff; 12719 12720 skb_orphan(skb); 12721 skb->sk = req_to_sk(req); 12722 skb->destructor = sock_pfree; 12723 12724 return 0; 12725 #else 12726 return -EOPNOTSUPP; 12727 #endif 12728 } 12729 12730 __bpf_kfunc int bpf_sock_ops_enable_tx_tstamp(struct bpf_sock_ops_kern *skops, 12731 u64 flags) 12732 { 12733 struct sk_buff *skb; 12734 12735 if (skops->op != BPF_SOCK_OPS_TSTAMP_SENDMSG_CB) 12736 return -EOPNOTSUPP; 12737 12738 if (flags) 12739 return -EINVAL; 12740 12741 skb = skops->skb; 12742 skb_shinfo(skb)->tx_flags |= SKBTX_BPF; 12743 TCP_SKB_CB(skb)->txstamp_ack |= TSTAMP_ACK_BPF; 12744 skb_shinfo(skb)->tskey = TCP_SKB_CB(skb)->seq + skb->len - 1; 12745 12746 return 0; 12747 } 12748 12749 /** 12750 * bpf_xdp_pull_data() - Pull in non-linear xdp data. 12751 * @x: &xdp_md associated with the XDP buffer 12752 * @len: length of data to be made directly accessible in the linear part 12753 * 12754 * Pull in data in case the XDP buffer associated with @x is non-linear and 12755 * not all @len are in the linear data area. 12756 * 12757 * Direct packet access allows reading and writing linear XDP data through 12758 * packet pointers (i.e., &xdp_md->data + offsets). The amount of data which 12759 * ends up in the linear part of the xdp_buff depends on the NIC and its 12760 * configuration. When a frag-capable XDP program wants to directly access 12761 * headers that may be in the non-linear area, call this kfunc to make sure 12762 * the data is available in the linear area. Alternatively, use dynptr or 12763 * bpf_xdp_{load,store}_bytes() to access data without pulling. 12764 * 12765 * This kfunc can also be used with bpf_xdp_adjust_head() to decapsulate 12766 * headers in the non-linear data area. 12767 * 12768 * A call to this kfunc may reduce headroom. If there is not enough tailroom 12769 * in the linear data area, metadata and data will be shifted down. 12770 * 12771 * A call to this kfunc is susceptible to change the buffer geometry. 12772 * Therefore, at load time, all checks on pointers previously done by the 12773 * verifier are invalidated and must be performed again, if the kfunc is used 12774 * in combination with direct packet access. 12775 * 12776 * Return: 12777 * * %0 - success 12778 * * %-EINVAL - invalid len 12779 */ 12780 __bpf_kfunc int bpf_xdp_pull_data(struct xdp_md *x, u32 len) 12781 { 12782 struct xdp_buff *xdp = (struct xdp_buff *)x; 12783 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 12784 int i, delta, shift, headroom, tailroom, n_frags_free = 0; 12785 void *data_hard_end = xdp_data_hard_end(xdp); 12786 int data_len = xdp->data_end - xdp->data; 12787 void *start; 12788 12789 if (len <= data_len) 12790 return 0; 12791 12792 if (unlikely(len > xdp_get_buff_len(xdp))) 12793 return -EINVAL; 12794 12795 start = xdp_data_meta_unsupported(xdp) ? xdp->data : xdp->data_meta; 12796 12797 headroom = start - xdp->data_hard_start - sizeof(struct xdp_frame); 12798 tailroom = data_hard_end - xdp->data_end; 12799 12800 delta = len - data_len; 12801 if (unlikely(delta > tailroom + headroom)) 12802 return -EINVAL; 12803 12804 shift = delta - tailroom; 12805 if (shift > 0) { 12806 memmove(start - shift, start, xdp->data_end - start); 12807 12808 xdp->data_meta -= shift; 12809 xdp->data -= shift; 12810 xdp->data_end -= shift; 12811 } 12812 12813 for (i = 0; i < sinfo->nr_frags && delta; i++) { 12814 skb_frag_t *frag = &sinfo->frags[i]; 12815 u32 shrink = min_t(u32, delta, skb_frag_size(frag)); 12816 12817 memcpy(xdp->data_end, skb_frag_address(frag), shrink); 12818 12819 xdp->data_end += shrink; 12820 sinfo->xdp_frags_size -= shrink; 12821 delta -= shrink; 12822 if (bpf_xdp_shrink_data(xdp, frag, shrink, false)) 12823 n_frags_free++; 12824 } 12825 12826 if (unlikely(n_frags_free)) { 12827 memmove(sinfo->frags, sinfo->frags + n_frags_free, 12828 (sinfo->nr_frags - n_frags_free) * sizeof(skb_frag_t)); 12829 12830 sinfo->nr_frags -= n_frags_free; 12831 12832 if (!sinfo->nr_frags) { 12833 xdp_buff_clear_frags_flag(xdp); 12834 xdp_buff_clear_frag_pfmemalloc(xdp); 12835 } 12836 } 12837 12838 return 0; 12839 } 12840 12841 /** 12842 * bpf_icmp_send - Send an ICMP control message 12843 * @skb_ctx: Packet that triggered the control message 12844 * @type: ICMP type (only ICMP_DEST_UNREACH/ICMPV6_DEST_UNREACH supported) 12845 * @code: ICMP code (0-15 except ICMP_FRAG_NEEDED for IPv4, 0-6 for IPv6) 12846 * 12847 * Sends an ICMP control message in response to the packet. The original packet 12848 * is cloned before sending the ICMP message, so the BPF program can still let 12849 * the packet pass if desired. 12850 * 12851 * Currently only ICMP_DEST_UNREACH (IPv4) and ICMPV6_DEST_UNREACH (IPv6) are 12852 * supported. 12853 * 12854 * Return: 0 on success (send attempt), negative error code on failure: 12855 * -EBUSY: Recursion detected 12856 * -EPROTONOSUPPORT: Non-IP protocol 12857 * -EOPNOTSUPP: Unsupported ICMP type 12858 * -EINVAL: Invalid code parameter 12859 * -ENETUNREACH: No usable route/dst for the ICMP reply 12860 * -ENOMEM: Memory allocation failed 12861 */ 12862 __bpf_kfunc int bpf_icmp_send(struct __sk_buff *skb_ctx, int type, int code) 12863 { 12864 struct sk_buff *skb = (struct sk_buff *)skb_ctx; 12865 struct sock *sk; 12866 12867 sk = skb_to_full_sk(skb); 12868 if (sk && sk->sk_kern_sock && 12869 (sk->sk_protocol == IPPROTO_ICMP || sk->sk_protocol == IPPROTO_ICMPV6)) 12870 return -EBUSY; 12871 12872 if (!skb_valid_dst(skb)) 12873 return -ENETUNREACH; 12874 12875 switch (skb->protocol) { 12876 #if IS_ENABLED(CONFIG_INET) 12877 case htons(ETH_P_IP): { 12878 struct sk_buff *nskb; 12879 12880 if (type != ICMP_DEST_UNREACH) 12881 return -EOPNOTSUPP; 12882 if (code < 0 || code > NR_ICMP_UNREACH || 12883 code == ICMP_FRAG_NEEDED) /* needs a valid next-hop MTU */ 12884 return -EINVAL; 12885 12886 nskb = skb_clone(skb, GFP_ATOMIC); 12887 if (!nskb) 12888 return -ENOMEM; 12889 12890 memset(IPCB(nskb), 0, sizeof(*IPCB(nskb))); 12891 icmp_send(nskb, type, code, 0); 12892 consume_skb(nskb); 12893 break; 12894 } 12895 #endif 12896 #if IS_ENABLED(CONFIG_IPV6) 12897 case htons(ETH_P_IPV6): { 12898 struct sk_buff *nskb; 12899 12900 if (type != ICMPV6_DEST_UNREACH) 12901 return -EOPNOTSUPP; 12902 if (code < 0 || code > ICMPV6_REJECT_ROUTE) 12903 return -EINVAL; 12904 12905 nskb = skb_clone(skb, GFP_ATOMIC); 12906 if (!nskb) 12907 return -ENOMEM; 12908 12909 memset(IP6CB(nskb), 0, sizeof(*IP6CB(nskb))); 12910 icmpv6_send(nskb, type, code, 0); 12911 consume_skb(nskb); 12912 break; 12913 } 12914 #endif 12915 default: 12916 return -EPROTONOSUPPORT; 12917 } 12918 12919 return 0; 12920 } 12921 12922 __bpf_kfunc_end_defs(); 12923 12924 int bpf_dynptr_from_skb_rdonly(struct __sk_buff *skb, u64 flags, 12925 struct bpf_dynptr *ptr__uninit) 12926 { 12927 struct bpf_dynptr_kern *ptr = (struct bpf_dynptr_kern *)ptr__uninit; 12928 int err; 12929 12930 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit); 12931 if (err) 12932 return err; 12933 12934 bpf_dynptr_set_rdonly(ptr); 12935 12936 return 0; 12937 } 12938 12939 BTF_KFUNCS_START(bpf_kfunc_check_set_skb) 12940 BTF_ID_FLAGS(func, bpf_dynptr_from_skb) 12941 BTF_KFUNCS_END(bpf_kfunc_check_set_skb) 12942 12943 BTF_KFUNCS_START(bpf_kfunc_check_set_skb_meta) 12944 BTF_ID_FLAGS(func, bpf_dynptr_from_skb_meta) 12945 BTF_KFUNCS_END(bpf_kfunc_check_set_skb_meta) 12946 12947 BTF_KFUNCS_START(bpf_kfunc_check_set_xdp) 12948 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp) 12949 BTF_ID_FLAGS(func, bpf_xdp_pull_data) 12950 BTF_KFUNCS_END(bpf_kfunc_check_set_xdp) 12951 12952 BTF_KFUNCS_START(bpf_kfunc_check_set_sock_addr) 12953 BTF_ID_FLAGS(func, bpf_sock_addr_set_sun_path) 12954 BTF_KFUNCS_END(bpf_kfunc_check_set_sock_addr) 12955 12956 BTF_KFUNCS_START(bpf_kfunc_check_set_tcp_reqsk) 12957 BTF_ID_FLAGS(func, bpf_sk_assign_tcp_reqsk) 12958 BTF_KFUNCS_END(bpf_kfunc_check_set_tcp_reqsk) 12959 12960 BTF_KFUNCS_START(bpf_kfunc_check_set_sock_ops) 12961 BTF_ID_FLAGS(func, bpf_sock_ops_enable_tx_tstamp) 12962 BTF_KFUNCS_END(bpf_kfunc_check_set_sock_ops) 12963 12964 BTF_KFUNCS_START(bpf_kfunc_check_set_icmp_send) 12965 BTF_ID_FLAGS(func, bpf_icmp_send) 12966 BTF_KFUNCS_END(bpf_kfunc_check_set_icmp_send) 12967 12968 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = { 12969 .owner = THIS_MODULE, 12970 .set = &bpf_kfunc_check_set_skb, 12971 }; 12972 12973 static const struct btf_kfunc_id_set bpf_kfunc_set_skb_meta = { 12974 .owner = THIS_MODULE, 12975 .set = &bpf_kfunc_check_set_skb_meta, 12976 }; 12977 12978 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = { 12979 .owner = THIS_MODULE, 12980 .set = &bpf_kfunc_check_set_xdp, 12981 }; 12982 12983 static const struct btf_kfunc_id_set bpf_kfunc_set_sock_addr = { 12984 .owner = THIS_MODULE, 12985 .set = &bpf_kfunc_check_set_sock_addr, 12986 }; 12987 12988 static const struct btf_kfunc_id_set bpf_kfunc_set_tcp_reqsk = { 12989 .owner = THIS_MODULE, 12990 .set = &bpf_kfunc_check_set_tcp_reqsk, 12991 }; 12992 12993 static const struct btf_kfunc_id_set bpf_kfunc_set_sock_ops = { 12994 .owner = THIS_MODULE, 12995 .set = &bpf_kfunc_check_set_sock_ops, 12996 }; 12997 12998 static const struct btf_kfunc_id_set bpf_kfunc_set_icmp_send = { 12999 .owner = THIS_MODULE, 13000 .set = &bpf_kfunc_check_set_icmp_send, 13001 }; 13002 13003 static int __init bpf_kfunc_init(void) 13004 { 13005 int ret; 13006 13007 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb); 13008 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb); 13009 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb); 13010 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb); 13011 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb); 13012 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb); 13013 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb); 13014 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb); 13015 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb); 13016 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb); 13017 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_kfunc_set_skb); 13018 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS, &bpf_kfunc_set_skb); 13019 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb_meta); 13020 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb_meta); 13021 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp); 13022 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SOCK_ADDR, 13023 &bpf_kfunc_set_sock_addr); 13024 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_tcp_reqsk); 13025 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_icmp_send); 13026 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCK_OPS, &bpf_kfunc_set_sock_ops); 13027 } 13028 late_initcall(bpf_kfunc_init); 13029 13030 __bpf_kfunc_start_defs(); 13031 13032 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code. 13033 * 13034 * The function expects a non-NULL pointer to a socket, and invokes the 13035 * protocol specific socket destroy handlers. 13036 * 13037 * The helper can only be called from BPF contexts that have acquired the socket 13038 * locks. 13039 * 13040 * Parameters: 13041 * @sock: Pointer to socket to be destroyed 13042 * 13043 * Return: 13044 * On error, may return EOPNOTSUPP, or whatever the protocol specific 13045 * destroy handler returns. 13046 * EOPNOTSUPP if protocol specific destroy handler is not supported. 13047 * 0 otherwise 13048 */ 13049 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock) 13050 { 13051 struct sock *sk = (struct sock *)sock; 13052 13053 /* The locking semantics that allow for synchronous execution of the 13054 * destroy handlers are only supported for TCP and UDP. 13055 * Supporting protocols will need to acquire sock lock in the BPF context 13056 * prior to invoking this kfunc. 13057 */ 13058 if (!sk->sk_prot->diag_destroy) 13059 return -EOPNOTSUPP; 13060 13061 if (sk_fullsock(sk) && 13062 sk->sk_protocol != IPPROTO_TCP && 13063 sk->sk_protocol != IPPROTO_UDP) 13064 return -EOPNOTSUPP; 13065 13066 return sk->sk_prot->diag_destroy(sk, ECONNABORTED); 13067 } 13068 13069 __bpf_kfunc_end_defs(); 13070 13071 BTF_KFUNCS_START(bpf_sk_iter_kfunc_ids) 13072 BTF_ID_FLAGS(func, bpf_sock_destroy) 13073 BTF_KFUNCS_END(bpf_sk_iter_kfunc_ids) 13074 13075 BTF_SET_START(bpf_sk_iter_locked_ids) 13076 #if defined(CONFIG_INET) && defined(CONFIG_PROC_FS) 13077 BTF_ID(func, bpf_iter_tcp) 13078 BTF_ID(func, bpf_iter_udp) 13079 #endif 13080 BTF_SET_END(bpf_sk_iter_locked_ids) 13081 13082 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id) 13083 { 13084 if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) && 13085 (prog->expected_attach_type != BPF_TRACE_ITER || 13086 !btf_id_set_contains(&bpf_sk_iter_locked_ids, prog->aux->attach_btf_id))) 13087 return -EACCES; 13088 13089 return 0; 13090 } 13091 13092 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = { 13093 .owner = THIS_MODULE, 13094 .set = &bpf_sk_iter_kfunc_ids, 13095 .filter = tracing_iter_filter, 13096 }; 13097 13098 static int init_subsystem(void) 13099 { 13100 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set); 13101 } 13102 late_initcall(init_subsystem); 13103