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 <asm/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/xfrm.h> 60 #include <net/udp.h> 61 #include <linux/bpf_trace.h> 62 #include <net/xdp_sock.h> 63 #include <linux/inetdevice.h> 64 #include <net/inet_hashtables.h> 65 #include <net/inet6_hashtables.h> 66 #include <net/ip_fib.h> 67 #include <net/nexthop.h> 68 #include <net/flow.h> 69 #include <net/arp.h> 70 #include <net/ipv6.h> 71 #include <net/net_namespace.h> 72 #include <linux/seg6_local.h> 73 #include <net/seg6.h> 74 #include <net/seg6_local.h> 75 #include <net/lwtunnel.h> 76 #include <net/ipv6_stubs.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 85 static const struct bpf_func_proto * 86 bpf_sk_base_func_proto(enum bpf_func_id func_id); 87 88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len) 89 { 90 if (in_compat_syscall()) { 91 struct compat_sock_fprog f32; 92 93 if (len != sizeof(f32)) 94 return -EINVAL; 95 if (copy_from_sockptr(&f32, src, sizeof(f32))) 96 return -EFAULT; 97 memset(dst, 0, sizeof(*dst)); 98 dst->len = f32.len; 99 dst->filter = compat_ptr(f32.filter); 100 } else { 101 if (len != sizeof(*dst)) 102 return -EINVAL; 103 if (copy_from_sockptr(dst, src, sizeof(*dst))) 104 return -EFAULT; 105 } 106 107 return 0; 108 } 109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user); 110 111 /** 112 * sk_filter_trim_cap - run a packet through a socket filter 113 * @sk: sock associated with &sk_buff 114 * @skb: buffer to filter 115 * @cap: limit on how short the eBPF program may trim the packet 116 * 117 * Run the eBPF program and then cut skb->data to correct size returned by 118 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 119 * than pkt_len we keep whole skb->data. This is the socket level 120 * wrapper to bpf_prog_run. It returns 0 if the packet should 121 * be accepted or -EPERM if the packet should be tossed. 122 * 123 */ 124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 125 { 126 int err; 127 struct sk_filter *filter; 128 129 /* 130 * If the skb was allocated from pfmemalloc reserves, only 131 * allow SOCK_MEMALLOC sockets to use it as this socket is 132 * helping free memory 133 */ 134 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 135 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 136 return -ENOMEM; 137 } 138 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 139 if (err) 140 return err; 141 142 err = security_sock_rcv_skb(sk, skb); 143 if (err) 144 return err; 145 146 rcu_read_lock(); 147 filter = rcu_dereference(sk->sk_filter); 148 if (filter) { 149 struct sock *save_sk = skb->sk; 150 unsigned int pkt_len; 151 152 skb->sk = sk; 153 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 154 skb->sk = save_sk; 155 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 156 } 157 rcu_read_unlock(); 158 159 return err; 160 } 161 EXPORT_SYMBOL(sk_filter_trim_cap); 162 163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb) 164 { 165 return skb_get_poff(skb); 166 } 167 168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x) 169 { 170 struct nlattr *nla; 171 172 if (skb_is_nonlinear(skb)) 173 return 0; 174 175 if (skb->len < sizeof(struct nlattr)) 176 return 0; 177 178 if (a > skb->len - sizeof(struct nlattr)) 179 return 0; 180 181 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 182 if (nla) 183 return (void *) nla - (void *) skb->data; 184 185 return 0; 186 } 187 188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 189 { 190 struct nlattr *nla; 191 192 if (skb_is_nonlinear(skb)) 193 return 0; 194 195 if (skb->len < sizeof(struct nlattr)) 196 return 0; 197 198 if (a > skb->len - sizeof(struct nlattr)) 199 return 0; 200 201 nla = (struct nlattr *) &skb->data[a]; 202 if (nla->nla_len > skb->len - a) 203 return 0; 204 205 nla = nla_find_nested(nla, x); 206 if (nla) 207 return (void *) nla - (void *) skb->data; 208 209 return 0; 210 } 211 212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *, 213 data, int, headlen, int, offset) 214 { 215 u8 tmp, *ptr; 216 const int len = sizeof(tmp); 217 218 if (offset >= 0) { 219 if (headlen - offset >= len) 220 return *(u8 *)(data + offset); 221 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 222 return tmp; 223 } else { 224 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 225 if (likely(ptr)) 226 return *(u8 *)ptr; 227 } 228 229 return -EFAULT; 230 } 231 232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb, 233 int, offset) 234 { 235 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len, 236 offset); 237 } 238 239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *, 240 data, int, headlen, int, offset) 241 { 242 __be16 tmp, *ptr; 243 const int len = sizeof(tmp); 244 245 if (offset >= 0) { 246 if (headlen - offset >= len) 247 return get_unaligned_be16(data + offset); 248 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 249 return be16_to_cpu(tmp); 250 } else { 251 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 252 if (likely(ptr)) 253 return get_unaligned_be16(ptr); 254 } 255 256 return -EFAULT; 257 } 258 259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb, 260 int, offset) 261 { 262 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len, 263 offset); 264 } 265 266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *, 267 data, int, headlen, int, offset) 268 { 269 __be32 tmp, *ptr; 270 const int len = sizeof(tmp); 271 272 if (likely(offset >= 0)) { 273 if (headlen - offset >= len) 274 return get_unaligned_be32(data + offset); 275 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 276 return be32_to_cpu(tmp); 277 } else { 278 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 279 if (likely(ptr)) 280 return get_unaligned_be32(ptr); 281 } 282 283 return -EFAULT; 284 } 285 286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb, 287 int, offset) 288 { 289 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len, 290 offset); 291 } 292 293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 294 struct bpf_insn *insn_buf) 295 { 296 struct bpf_insn *insn = insn_buf; 297 298 switch (skb_field) { 299 case SKF_AD_MARK: 300 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4); 301 302 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 303 offsetof(struct sk_buff, mark)); 304 break; 305 306 case SKF_AD_PKTTYPE: 307 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET); 308 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 309 #ifdef __BIG_ENDIAN_BITFIELD 310 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 311 #endif 312 break; 313 314 case SKF_AD_QUEUE: 315 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2); 316 317 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 318 offsetof(struct sk_buff, queue_mapping)); 319 break; 320 321 case SKF_AD_VLAN_TAG: 322 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2); 323 324 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 325 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 326 offsetof(struct sk_buff, vlan_tci)); 327 break; 328 case SKF_AD_VLAN_TAG_PRESENT: 329 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4); 330 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 331 offsetof(struct sk_buff, vlan_all)); 332 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1); 333 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1); 334 break; 335 } 336 337 return insn - insn_buf; 338 } 339 340 static bool convert_bpf_extensions(struct sock_filter *fp, 341 struct bpf_insn **insnp) 342 { 343 struct bpf_insn *insn = *insnp; 344 u32 cnt; 345 346 switch (fp->k) { 347 case SKF_AD_OFF + SKF_AD_PROTOCOL: 348 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2); 349 350 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 351 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 352 offsetof(struct sk_buff, protocol)); 353 /* A = ntohs(A) [emitting a nop or swap16] */ 354 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 355 break; 356 357 case SKF_AD_OFF + SKF_AD_PKTTYPE: 358 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 359 insn += cnt - 1; 360 break; 361 362 case SKF_AD_OFF + SKF_AD_IFINDEX: 363 case SKF_AD_OFF + SKF_AD_HATYPE: 364 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4); 365 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2); 366 367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 368 BPF_REG_TMP, BPF_REG_CTX, 369 offsetof(struct sk_buff, dev)); 370 /* if (tmp != 0) goto pc + 1 */ 371 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 372 *insn++ = BPF_EXIT_INSN(); 373 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 374 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 375 offsetof(struct net_device, ifindex)); 376 else 377 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 378 offsetof(struct net_device, type)); 379 break; 380 381 case SKF_AD_OFF + SKF_AD_MARK: 382 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 383 insn += cnt - 1; 384 break; 385 386 case SKF_AD_OFF + SKF_AD_RXHASH: 387 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4); 388 389 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 390 offsetof(struct sk_buff, hash)); 391 break; 392 393 case SKF_AD_OFF + SKF_AD_QUEUE: 394 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 395 insn += cnt - 1; 396 break; 397 398 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 399 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 400 BPF_REG_A, BPF_REG_CTX, insn); 401 insn += cnt - 1; 402 break; 403 404 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 405 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 406 BPF_REG_A, BPF_REG_CTX, insn); 407 insn += cnt - 1; 408 break; 409 410 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 411 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2); 412 413 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 414 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 415 offsetof(struct sk_buff, vlan_proto)); 416 /* A = ntohs(A) [emitting a nop or swap16] */ 417 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 418 break; 419 420 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 421 case SKF_AD_OFF + SKF_AD_NLATTR: 422 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 423 case SKF_AD_OFF + SKF_AD_CPU: 424 case SKF_AD_OFF + SKF_AD_RANDOM: 425 /* arg1 = CTX */ 426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 427 /* arg2 = A */ 428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 429 /* arg3 = X */ 430 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 431 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 432 switch (fp->k) { 433 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 434 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset); 435 break; 436 case SKF_AD_OFF + SKF_AD_NLATTR: 437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr); 438 break; 439 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 440 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest); 441 break; 442 case SKF_AD_OFF + SKF_AD_CPU: 443 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id); 444 break; 445 case SKF_AD_OFF + SKF_AD_RANDOM: 446 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 447 bpf_user_rnd_init_once(); 448 break; 449 } 450 break; 451 452 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 453 /* A ^= X */ 454 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 455 break; 456 457 default: 458 /* This is just a dummy call to avoid letting the compiler 459 * evict __bpf_call_base() as an optimization. Placed here 460 * where no-one bothers. 461 */ 462 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 463 return false; 464 } 465 466 *insnp = insn; 467 return true; 468 } 469 470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp) 471 { 472 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS); 473 int size = bpf_size_to_bytes(BPF_SIZE(fp->code)); 474 bool endian = BPF_SIZE(fp->code) == BPF_H || 475 BPF_SIZE(fp->code) == BPF_W; 476 bool indirect = BPF_MODE(fp->code) == BPF_IND; 477 const int ip_align = NET_IP_ALIGN; 478 struct bpf_insn *insn = *insnp; 479 int offset = fp->k; 480 481 if (!indirect && 482 ((unaligned_ok && offset >= 0) || 483 (!unaligned_ok && offset >= 0 && 484 offset + ip_align >= 0 && 485 offset + ip_align % size == 0))) { 486 bool ldx_off_ok = offset <= S16_MAX; 487 488 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H); 489 if (offset) 490 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset); 491 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, 492 size, 2 + endian + (!ldx_off_ok * 2)); 493 if (ldx_off_ok) { 494 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 495 BPF_REG_D, offset); 496 } else { 497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D); 498 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset); 499 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 500 BPF_REG_TMP, 0); 501 } 502 if (endian) 503 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8); 504 *insn++ = BPF_JMP_A(8); 505 } 506 507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 508 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D); 509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H); 510 if (!indirect) { 511 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset); 512 } else { 513 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X); 514 if (fp->k) 515 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset); 516 } 517 518 switch (BPF_SIZE(fp->code)) { 519 case BPF_B: 520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8); 521 break; 522 case BPF_H: 523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16); 524 break; 525 case BPF_W: 526 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32); 527 break; 528 default: 529 return false; 530 } 531 532 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2); 533 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 534 *insn = BPF_EXIT_INSN(); 535 536 *insnp = insn; 537 return true; 538 } 539 540 /** 541 * bpf_convert_filter - convert filter program 542 * @prog: the user passed filter program 543 * @len: the length of the user passed filter program 544 * @new_prog: allocated 'struct bpf_prog' or NULL 545 * @new_len: pointer to store length of converted program 546 * @seen_ld_abs: bool whether we've seen ld_abs/ind 547 * 548 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 549 * style extended BPF (eBPF). 550 * Conversion workflow: 551 * 552 * 1) First pass for calculating the new program length: 553 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs) 554 * 555 * 2) 2nd pass to remap in two passes: 1st pass finds new 556 * jump offsets, 2nd pass remapping: 557 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs) 558 */ 559 static int bpf_convert_filter(struct sock_filter *prog, int len, 560 struct bpf_prog *new_prog, int *new_len, 561 bool *seen_ld_abs) 562 { 563 int new_flen = 0, pass = 0, target, i, stack_off; 564 struct bpf_insn *new_insn, *first_insn = NULL; 565 struct sock_filter *fp; 566 int *addrs = NULL; 567 u8 bpf_src; 568 569 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 570 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 571 572 if (len <= 0 || len > BPF_MAXINSNS) 573 return -EINVAL; 574 575 if (new_prog) { 576 first_insn = new_prog->insnsi; 577 addrs = kcalloc(len, sizeof(*addrs), 578 GFP_KERNEL | __GFP_NOWARN); 579 if (!addrs) 580 return -ENOMEM; 581 } 582 583 do_pass: 584 new_insn = first_insn; 585 fp = prog; 586 587 /* Classic BPF related prologue emission. */ 588 if (new_prog) { 589 /* Classic BPF expects A and X to be reset first. These need 590 * to be guaranteed to be the first two instructions. 591 */ 592 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 593 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 594 595 /* All programs must keep CTX in callee saved BPF_REG_CTX. 596 * In eBPF case it's done by the compiler, here we need to 597 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 598 */ 599 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 600 if (*seen_ld_abs) { 601 /* For packet access in classic BPF, cache skb->data 602 * in callee-saved BPF R8 and skb->len - skb->data_len 603 * (headlen) in BPF R9. Since classic BPF is read-only 604 * on CTX, we only need to cache it once. 605 */ 606 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 607 BPF_REG_D, BPF_REG_CTX, 608 offsetof(struct sk_buff, data)); 609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX, 610 offsetof(struct sk_buff, len)); 611 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX, 612 offsetof(struct sk_buff, data_len)); 613 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP); 614 } 615 } else { 616 new_insn += 3; 617 } 618 619 for (i = 0; i < len; fp++, i++) { 620 struct bpf_insn tmp_insns[32] = { }; 621 struct bpf_insn *insn = tmp_insns; 622 623 if (addrs) 624 addrs[i] = new_insn - first_insn; 625 626 switch (fp->code) { 627 /* All arithmetic insns and skb loads map as-is. */ 628 case BPF_ALU | BPF_ADD | BPF_X: 629 case BPF_ALU | BPF_ADD | BPF_K: 630 case BPF_ALU | BPF_SUB | BPF_X: 631 case BPF_ALU | BPF_SUB | BPF_K: 632 case BPF_ALU | BPF_AND | BPF_X: 633 case BPF_ALU | BPF_AND | BPF_K: 634 case BPF_ALU | BPF_OR | BPF_X: 635 case BPF_ALU | BPF_OR | BPF_K: 636 case BPF_ALU | BPF_LSH | BPF_X: 637 case BPF_ALU | BPF_LSH | BPF_K: 638 case BPF_ALU | BPF_RSH | BPF_X: 639 case BPF_ALU | BPF_RSH | BPF_K: 640 case BPF_ALU | BPF_XOR | BPF_X: 641 case BPF_ALU | BPF_XOR | BPF_K: 642 case BPF_ALU | BPF_MUL | BPF_X: 643 case BPF_ALU | BPF_MUL | BPF_K: 644 case BPF_ALU | BPF_DIV | BPF_X: 645 case BPF_ALU | BPF_DIV | BPF_K: 646 case BPF_ALU | BPF_MOD | BPF_X: 647 case BPF_ALU | BPF_MOD | BPF_K: 648 case BPF_ALU | BPF_NEG: 649 case BPF_LD | BPF_ABS | BPF_W: 650 case BPF_LD | BPF_ABS | BPF_H: 651 case BPF_LD | BPF_ABS | BPF_B: 652 case BPF_LD | BPF_IND | BPF_W: 653 case BPF_LD | BPF_IND | BPF_H: 654 case BPF_LD | BPF_IND | BPF_B: 655 /* Check for overloaded BPF extension and 656 * directly convert it if found, otherwise 657 * just move on with mapping. 658 */ 659 if (BPF_CLASS(fp->code) == BPF_LD && 660 BPF_MODE(fp->code) == BPF_ABS && 661 convert_bpf_extensions(fp, &insn)) 662 break; 663 if (BPF_CLASS(fp->code) == BPF_LD && 664 convert_bpf_ld_abs(fp, &insn)) { 665 *seen_ld_abs = true; 666 break; 667 } 668 669 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) || 670 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) { 671 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X); 672 /* Error with exception code on div/mod by 0. 673 * For cBPF programs, this was always return 0. 674 */ 675 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2); 676 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 677 *insn++ = BPF_EXIT_INSN(); 678 } 679 680 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 681 break; 682 683 /* Jump transformation cannot use BPF block macros 684 * everywhere as offset calculation and target updates 685 * require a bit more work than the rest, i.e. jump 686 * opcodes map as-is, but offsets need adjustment. 687 */ 688 689 #define BPF_EMIT_JMP \ 690 do { \ 691 const s32 off_min = S16_MIN, off_max = S16_MAX; \ 692 s32 off; \ 693 \ 694 if (target >= len || target < 0) \ 695 goto err; \ 696 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 697 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 698 off -= insn - tmp_insns; \ 699 /* Reject anything not fitting into insn->off. */ \ 700 if (off < off_min || off > off_max) \ 701 goto err; \ 702 insn->off = off; \ 703 } while (0) 704 705 case BPF_JMP | BPF_JA: 706 target = i + fp->k + 1; 707 insn->code = fp->code; 708 BPF_EMIT_JMP; 709 break; 710 711 case BPF_JMP | BPF_JEQ | BPF_K: 712 case BPF_JMP | BPF_JEQ | BPF_X: 713 case BPF_JMP | BPF_JSET | BPF_K: 714 case BPF_JMP | BPF_JSET | BPF_X: 715 case BPF_JMP | BPF_JGT | BPF_K: 716 case BPF_JMP | BPF_JGT | BPF_X: 717 case BPF_JMP | BPF_JGE | BPF_K: 718 case BPF_JMP | BPF_JGE | BPF_X: 719 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 720 /* BPF immediates are signed, zero extend 721 * immediate into tmp register and use it 722 * in compare insn. 723 */ 724 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 725 726 insn->dst_reg = BPF_REG_A; 727 insn->src_reg = BPF_REG_TMP; 728 bpf_src = BPF_X; 729 } else { 730 insn->dst_reg = BPF_REG_A; 731 insn->imm = fp->k; 732 bpf_src = BPF_SRC(fp->code); 733 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 734 } 735 736 /* Common case where 'jump_false' is next insn. */ 737 if (fp->jf == 0) { 738 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 739 target = i + fp->jt + 1; 740 BPF_EMIT_JMP; 741 break; 742 } 743 744 /* Convert some jumps when 'jump_true' is next insn. */ 745 if (fp->jt == 0) { 746 switch (BPF_OP(fp->code)) { 747 case BPF_JEQ: 748 insn->code = BPF_JMP | BPF_JNE | bpf_src; 749 break; 750 case BPF_JGT: 751 insn->code = BPF_JMP | BPF_JLE | bpf_src; 752 break; 753 case BPF_JGE: 754 insn->code = BPF_JMP | BPF_JLT | bpf_src; 755 break; 756 default: 757 goto jmp_rest; 758 } 759 760 target = i + fp->jf + 1; 761 BPF_EMIT_JMP; 762 break; 763 } 764 jmp_rest: 765 /* Other jumps are mapped into two insns: Jxx and JA. */ 766 target = i + fp->jt + 1; 767 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 768 BPF_EMIT_JMP; 769 insn++; 770 771 insn->code = BPF_JMP | BPF_JA; 772 target = i + fp->jf + 1; 773 BPF_EMIT_JMP; 774 break; 775 776 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ 777 case BPF_LDX | BPF_MSH | BPF_B: { 778 struct sock_filter tmp = { 779 .code = BPF_LD | BPF_ABS | BPF_B, 780 .k = fp->k, 781 }; 782 783 *seen_ld_abs = true; 784 785 /* X = A */ 786 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 787 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 788 convert_bpf_ld_abs(&tmp, &insn); 789 insn++; 790 /* A &= 0xf */ 791 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 792 /* A <<= 2 */ 793 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 794 /* tmp = X */ 795 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X); 796 /* X = A */ 797 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 798 /* A = tmp */ 799 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 800 break; 801 } 802 /* RET_K is remaped into 2 insns. RET_A case doesn't need an 803 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 804 */ 805 case BPF_RET | BPF_A: 806 case BPF_RET | BPF_K: 807 if (BPF_RVAL(fp->code) == BPF_K) 808 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 809 0, fp->k); 810 *insn = BPF_EXIT_INSN(); 811 break; 812 813 /* Store to stack. */ 814 case BPF_ST: 815 case BPF_STX: 816 stack_off = fp->k * 4 + 4; 817 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 818 BPF_ST ? BPF_REG_A : BPF_REG_X, 819 -stack_off); 820 /* check_load_and_stores() verifies that classic BPF can 821 * load from stack only after write, so tracking 822 * stack_depth for ST|STX insns is enough 823 */ 824 if (new_prog && new_prog->aux->stack_depth < stack_off) 825 new_prog->aux->stack_depth = stack_off; 826 break; 827 828 /* Load from stack. */ 829 case BPF_LD | BPF_MEM: 830 case BPF_LDX | BPF_MEM: 831 stack_off = fp->k * 4 + 4; 832 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 833 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 834 -stack_off); 835 break; 836 837 /* A = K or X = K */ 838 case BPF_LD | BPF_IMM: 839 case BPF_LDX | BPF_IMM: 840 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 841 BPF_REG_A : BPF_REG_X, fp->k); 842 break; 843 844 /* X = A */ 845 case BPF_MISC | BPF_TAX: 846 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 847 break; 848 849 /* A = X */ 850 case BPF_MISC | BPF_TXA: 851 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 852 break; 853 854 /* A = skb->len or X = skb->len */ 855 case BPF_LD | BPF_W | BPF_LEN: 856 case BPF_LDX | BPF_W | BPF_LEN: 857 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 858 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 859 offsetof(struct sk_buff, len)); 860 break; 861 862 /* Access seccomp_data fields. */ 863 case BPF_LDX | BPF_ABS | BPF_W: 864 /* A = *(u32 *) (ctx + K) */ 865 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 866 break; 867 868 /* Unknown instruction. */ 869 default: 870 goto err; 871 } 872 873 insn++; 874 if (new_prog) 875 memcpy(new_insn, tmp_insns, 876 sizeof(*insn) * (insn - tmp_insns)); 877 new_insn += insn - tmp_insns; 878 } 879 880 if (!new_prog) { 881 /* Only calculating new length. */ 882 *new_len = new_insn - first_insn; 883 if (*seen_ld_abs) 884 *new_len += 4; /* Prologue bits. */ 885 return 0; 886 } 887 888 pass++; 889 if (new_flen != new_insn - first_insn) { 890 new_flen = new_insn - first_insn; 891 if (pass > 2) 892 goto err; 893 goto do_pass; 894 } 895 896 kfree(addrs); 897 BUG_ON(*new_len != new_flen); 898 return 0; 899 err: 900 kfree(addrs); 901 return -EINVAL; 902 } 903 904 /* Security: 905 * 906 * As we dont want to clear mem[] array for each packet going through 907 * __bpf_prog_run(), we check that filter loaded by user never try to read 908 * a cell if not previously written, and we check all branches to be sure 909 * a malicious user doesn't try to abuse us. 910 */ 911 static int check_load_and_stores(const struct sock_filter *filter, int flen) 912 { 913 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 914 int pc, ret = 0; 915 916 BUILD_BUG_ON(BPF_MEMWORDS > 16); 917 918 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); 919 if (!masks) 920 return -ENOMEM; 921 922 memset(masks, 0xff, flen * sizeof(*masks)); 923 924 for (pc = 0; pc < flen; pc++) { 925 memvalid &= masks[pc]; 926 927 switch (filter[pc].code) { 928 case BPF_ST: 929 case BPF_STX: 930 memvalid |= (1 << filter[pc].k); 931 break; 932 case BPF_LD | BPF_MEM: 933 case BPF_LDX | BPF_MEM: 934 if (!(memvalid & (1 << filter[pc].k))) { 935 ret = -EINVAL; 936 goto error; 937 } 938 break; 939 case BPF_JMP | BPF_JA: 940 /* A jump must set masks on target */ 941 masks[pc + 1 + filter[pc].k] &= memvalid; 942 memvalid = ~0; 943 break; 944 case BPF_JMP | BPF_JEQ | BPF_K: 945 case BPF_JMP | BPF_JEQ | BPF_X: 946 case BPF_JMP | BPF_JGE | BPF_K: 947 case BPF_JMP | BPF_JGE | BPF_X: 948 case BPF_JMP | BPF_JGT | BPF_K: 949 case BPF_JMP | BPF_JGT | BPF_X: 950 case BPF_JMP | BPF_JSET | BPF_K: 951 case BPF_JMP | BPF_JSET | BPF_X: 952 /* A jump must set masks on targets */ 953 masks[pc + 1 + filter[pc].jt] &= memvalid; 954 masks[pc + 1 + filter[pc].jf] &= memvalid; 955 memvalid = ~0; 956 break; 957 } 958 } 959 error: 960 kfree(masks); 961 return ret; 962 } 963 964 static bool chk_code_allowed(u16 code_to_probe) 965 { 966 static const bool codes[] = { 967 /* 32 bit ALU operations */ 968 [BPF_ALU | BPF_ADD | BPF_K] = true, 969 [BPF_ALU | BPF_ADD | BPF_X] = true, 970 [BPF_ALU | BPF_SUB | BPF_K] = true, 971 [BPF_ALU | BPF_SUB | BPF_X] = true, 972 [BPF_ALU | BPF_MUL | BPF_K] = true, 973 [BPF_ALU | BPF_MUL | BPF_X] = true, 974 [BPF_ALU | BPF_DIV | BPF_K] = true, 975 [BPF_ALU | BPF_DIV | BPF_X] = true, 976 [BPF_ALU | BPF_MOD | BPF_K] = true, 977 [BPF_ALU | BPF_MOD | BPF_X] = true, 978 [BPF_ALU | BPF_AND | BPF_K] = true, 979 [BPF_ALU | BPF_AND | BPF_X] = true, 980 [BPF_ALU | BPF_OR | BPF_K] = true, 981 [BPF_ALU | BPF_OR | BPF_X] = true, 982 [BPF_ALU | BPF_XOR | BPF_K] = true, 983 [BPF_ALU | BPF_XOR | BPF_X] = true, 984 [BPF_ALU | BPF_LSH | BPF_K] = true, 985 [BPF_ALU | BPF_LSH | BPF_X] = true, 986 [BPF_ALU | BPF_RSH | BPF_K] = true, 987 [BPF_ALU | BPF_RSH | BPF_X] = true, 988 [BPF_ALU | BPF_NEG] = true, 989 /* Load instructions */ 990 [BPF_LD | BPF_W | BPF_ABS] = true, 991 [BPF_LD | BPF_H | BPF_ABS] = true, 992 [BPF_LD | BPF_B | BPF_ABS] = true, 993 [BPF_LD | BPF_W | BPF_LEN] = true, 994 [BPF_LD | BPF_W | BPF_IND] = true, 995 [BPF_LD | BPF_H | BPF_IND] = true, 996 [BPF_LD | BPF_B | BPF_IND] = true, 997 [BPF_LD | BPF_IMM] = true, 998 [BPF_LD | BPF_MEM] = true, 999 [BPF_LDX | BPF_W | BPF_LEN] = true, 1000 [BPF_LDX | BPF_B | BPF_MSH] = true, 1001 [BPF_LDX | BPF_IMM] = true, 1002 [BPF_LDX | BPF_MEM] = true, 1003 /* Store instructions */ 1004 [BPF_ST] = true, 1005 [BPF_STX] = true, 1006 /* Misc instructions */ 1007 [BPF_MISC | BPF_TAX] = true, 1008 [BPF_MISC | BPF_TXA] = true, 1009 /* Return instructions */ 1010 [BPF_RET | BPF_K] = true, 1011 [BPF_RET | BPF_A] = true, 1012 /* Jump instructions */ 1013 [BPF_JMP | BPF_JA] = true, 1014 [BPF_JMP | BPF_JEQ | BPF_K] = true, 1015 [BPF_JMP | BPF_JEQ | BPF_X] = true, 1016 [BPF_JMP | BPF_JGE | BPF_K] = true, 1017 [BPF_JMP | BPF_JGE | BPF_X] = true, 1018 [BPF_JMP | BPF_JGT | BPF_K] = true, 1019 [BPF_JMP | BPF_JGT | BPF_X] = true, 1020 [BPF_JMP | BPF_JSET | BPF_K] = true, 1021 [BPF_JMP | BPF_JSET | BPF_X] = true, 1022 }; 1023 1024 if (code_to_probe >= ARRAY_SIZE(codes)) 1025 return false; 1026 1027 return codes[code_to_probe]; 1028 } 1029 1030 static bool bpf_check_basics_ok(const struct sock_filter *filter, 1031 unsigned int flen) 1032 { 1033 if (filter == NULL) 1034 return false; 1035 if (flen == 0 || flen > BPF_MAXINSNS) 1036 return false; 1037 1038 return true; 1039 } 1040 1041 /** 1042 * bpf_check_classic - verify socket filter code 1043 * @filter: filter to verify 1044 * @flen: length of filter 1045 * 1046 * Check the user's filter code. If we let some ugly 1047 * filter code slip through kaboom! The filter must contain 1048 * no references or jumps that are out of range, no illegal 1049 * instructions, and must end with a RET instruction. 1050 * 1051 * All jumps are forward as they are not signed. 1052 * 1053 * Returns 0 if the rule set is legal or -EINVAL if not. 1054 */ 1055 static int bpf_check_classic(const struct sock_filter *filter, 1056 unsigned int flen) 1057 { 1058 bool anc_found; 1059 int pc; 1060 1061 /* Check the filter code now */ 1062 for (pc = 0; pc < flen; pc++) { 1063 const struct sock_filter *ftest = &filter[pc]; 1064 1065 /* May we actually operate on this code? */ 1066 if (!chk_code_allowed(ftest->code)) 1067 return -EINVAL; 1068 1069 /* Some instructions need special checks */ 1070 switch (ftest->code) { 1071 case BPF_ALU | BPF_DIV | BPF_K: 1072 case BPF_ALU | BPF_MOD | BPF_K: 1073 /* Check for division by zero */ 1074 if (ftest->k == 0) 1075 return -EINVAL; 1076 break; 1077 case BPF_ALU | BPF_LSH | BPF_K: 1078 case BPF_ALU | BPF_RSH | BPF_K: 1079 if (ftest->k >= 32) 1080 return -EINVAL; 1081 break; 1082 case BPF_LD | BPF_MEM: 1083 case BPF_LDX | BPF_MEM: 1084 case BPF_ST: 1085 case BPF_STX: 1086 /* Check for invalid memory addresses */ 1087 if (ftest->k >= BPF_MEMWORDS) 1088 return -EINVAL; 1089 break; 1090 case BPF_JMP | BPF_JA: 1091 /* Note, the large ftest->k might cause loops. 1092 * Compare this with conditional jumps below, 1093 * where offsets are limited. --ANK (981016) 1094 */ 1095 if (ftest->k >= (unsigned int)(flen - pc - 1)) 1096 return -EINVAL; 1097 break; 1098 case BPF_JMP | BPF_JEQ | BPF_K: 1099 case BPF_JMP | BPF_JEQ | BPF_X: 1100 case BPF_JMP | BPF_JGE | BPF_K: 1101 case BPF_JMP | BPF_JGE | BPF_X: 1102 case BPF_JMP | BPF_JGT | BPF_K: 1103 case BPF_JMP | BPF_JGT | BPF_X: 1104 case BPF_JMP | BPF_JSET | BPF_K: 1105 case BPF_JMP | BPF_JSET | BPF_X: 1106 /* Both conditionals must be safe */ 1107 if (pc + ftest->jt + 1 >= flen || 1108 pc + ftest->jf + 1 >= flen) 1109 return -EINVAL; 1110 break; 1111 case BPF_LD | BPF_W | BPF_ABS: 1112 case BPF_LD | BPF_H | BPF_ABS: 1113 case BPF_LD | BPF_B | BPF_ABS: 1114 anc_found = false; 1115 if (bpf_anc_helper(ftest) & BPF_ANC) 1116 anc_found = true; 1117 /* Ancillary operation unknown or unsupported */ 1118 if (anc_found == false && ftest->k >= SKF_AD_OFF) 1119 return -EINVAL; 1120 } 1121 } 1122 1123 /* Last instruction must be a RET code */ 1124 switch (filter[flen - 1].code) { 1125 case BPF_RET | BPF_K: 1126 case BPF_RET | BPF_A: 1127 return check_load_and_stores(filter, flen); 1128 } 1129 1130 return -EINVAL; 1131 } 1132 1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 1134 const struct sock_fprog *fprog) 1135 { 1136 unsigned int fsize = bpf_classic_proglen(fprog); 1137 struct sock_fprog_kern *fkprog; 1138 1139 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); 1140 if (!fp->orig_prog) 1141 return -ENOMEM; 1142 1143 fkprog = fp->orig_prog; 1144 fkprog->len = fprog->len; 1145 1146 fkprog->filter = kmemdup(fp->insns, fsize, 1147 GFP_KERNEL | __GFP_NOWARN); 1148 if (!fkprog->filter) { 1149 kfree(fp->orig_prog); 1150 return -ENOMEM; 1151 } 1152 1153 return 0; 1154 } 1155 1156 static void bpf_release_orig_filter(struct bpf_prog *fp) 1157 { 1158 struct sock_fprog_kern *fprog = fp->orig_prog; 1159 1160 if (fprog) { 1161 kfree(fprog->filter); 1162 kfree(fprog); 1163 } 1164 } 1165 1166 static void __bpf_prog_release(struct bpf_prog *prog) 1167 { 1168 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 1169 bpf_prog_put(prog); 1170 } else { 1171 bpf_release_orig_filter(prog); 1172 bpf_prog_free(prog); 1173 } 1174 } 1175 1176 static void __sk_filter_release(struct sk_filter *fp) 1177 { 1178 __bpf_prog_release(fp->prog); 1179 kfree(fp); 1180 } 1181 1182 /** 1183 * sk_filter_release_rcu - Release a socket filter by rcu_head 1184 * @rcu: rcu_head that contains the sk_filter to free 1185 */ 1186 static void sk_filter_release_rcu(struct rcu_head *rcu) 1187 { 1188 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 1189 1190 __sk_filter_release(fp); 1191 } 1192 1193 /** 1194 * sk_filter_release - release a socket filter 1195 * @fp: filter to remove 1196 * 1197 * Remove a filter from a socket and release its resources. 1198 */ 1199 static void sk_filter_release(struct sk_filter *fp) 1200 { 1201 if (refcount_dec_and_test(&fp->refcnt)) 1202 call_rcu(&fp->rcu, sk_filter_release_rcu); 1203 } 1204 1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1206 { 1207 u32 filter_size = bpf_prog_size(fp->prog->len); 1208 1209 atomic_sub(filter_size, &sk->sk_omem_alloc); 1210 sk_filter_release(fp); 1211 } 1212 1213 /* try to charge the socket memory if there is space available 1214 * return true on success 1215 */ 1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1217 { 1218 u32 filter_size = bpf_prog_size(fp->prog->len); 1219 int optmem_max = READ_ONCE(sysctl_optmem_max); 1220 1221 /* same check as in sock_kmalloc() */ 1222 if (filter_size <= optmem_max && 1223 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) { 1224 atomic_add(filter_size, &sk->sk_omem_alloc); 1225 return true; 1226 } 1227 return false; 1228 } 1229 1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1231 { 1232 if (!refcount_inc_not_zero(&fp->refcnt)) 1233 return false; 1234 1235 if (!__sk_filter_charge(sk, fp)) { 1236 sk_filter_release(fp); 1237 return false; 1238 } 1239 return true; 1240 } 1241 1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 1243 { 1244 struct sock_filter *old_prog; 1245 struct bpf_prog *old_fp; 1246 int err, new_len, old_len = fp->len; 1247 bool seen_ld_abs = false; 1248 1249 /* We are free to overwrite insns et al right here as it won't be used at 1250 * this point in time anymore internally after the migration to the eBPF 1251 * instruction representation. 1252 */ 1253 BUILD_BUG_ON(sizeof(struct sock_filter) != 1254 sizeof(struct bpf_insn)); 1255 1256 /* Conversion cannot happen on overlapping memory areas, 1257 * so we need to keep the user BPF around until the 2nd 1258 * pass. At this time, the user BPF is stored in fp->insns. 1259 */ 1260 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), 1261 GFP_KERNEL | __GFP_NOWARN); 1262 if (!old_prog) { 1263 err = -ENOMEM; 1264 goto out_err; 1265 } 1266 1267 /* 1st pass: calculate the new program length. */ 1268 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len, 1269 &seen_ld_abs); 1270 if (err) 1271 goto out_err_free; 1272 1273 /* Expand fp for appending the new filter representation. */ 1274 old_fp = fp; 1275 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); 1276 if (!fp) { 1277 /* The old_fp is still around in case we couldn't 1278 * allocate new memory, so uncharge on that one. 1279 */ 1280 fp = old_fp; 1281 err = -ENOMEM; 1282 goto out_err_free; 1283 } 1284 1285 fp->len = new_len; 1286 1287 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ 1288 err = bpf_convert_filter(old_prog, old_len, fp, &new_len, 1289 &seen_ld_abs); 1290 if (err) 1291 /* 2nd bpf_convert_filter() can fail only if it fails 1292 * to allocate memory, remapping must succeed. Note, 1293 * that at this time old_fp has already been released 1294 * by krealloc(). 1295 */ 1296 goto out_err_free; 1297 1298 fp = bpf_prog_select_runtime(fp, &err); 1299 if (err) 1300 goto out_err_free; 1301 1302 kfree(old_prog); 1303 return fp; 1304 1305 out_err_free: 1306 kfree(old_prog); 1307 out_err: 1308 __bpf_prog_release(fp); 1309 return ERR_PTR(err); 1310 } 1311 1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, 1313 bpf_aux_classic_check_t trans) 1314 { 1315 int err; 1316 1317 fp->bpf_func = NULL; 1318 fp->jited = 0; 1319 1320 err = bpf_check_classic(fp->insns, fp->len); 1321 if (err) { 1322 __bpf_prog_release(fp); 1323 return ERR_PTR(err); 1324 } 1325 1326 /* There might be additional checks and transformations 1327 * needed on classic filters, f.e. in case of seccomp. 1328 */ 1329 if (trans) { 1330 err = trans(fp->insns, fp->len); 1331 if (err) { 1332 __bpf_prog_release(fp); 1333 return ERR_PTR(err); 1334 } 1335 } 1336 1337 /* Probe if we can JIT compile the filter and if so, do 1338 * the compilation of the filter. 1339 */ 1340 bpf_jit_compile(fp); 1341 1342 /* JIT compiler couldn't process this filter, so do the eBPF translation 1343 * for the optimized interpreter. 1344 */ 1345 if (!fp->jited) 1346 fp = bpf_migrate_filter(fp); 1347 1348 return fp; 1349 } 1350 1351 /** 1352 * bpf_prog_create - create an unattached filter 1353 * @pfp: the unattached filter that is created 1354 * @fprog: the filter program 1355 * 1356 * Create a filter independent of any socket. We first run some 1357 * sanity checks on it to make sure it does not explode on us later. 1358 * If an error occurs or there is insufficient memory for the filter 1359 * a negative errno code is returned. On success the return is zero. 1360 */ 1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) 1362 { 1363 unsigned int fsize = bpf_classic_proglen(fprog); 1364 struct bpf_prog *fp; 1365 1366 /* Make sure new filter is there and in the right amounts. */ 1367 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1368 return -EINVAL; 1369 1370 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1371 if (!fp) 1372 return -ENOMEM; 1373 1374 memcpy(fp->insns, fprog->filter, fsize); 1375 1376 fp->len = fprog->len; 1377 /* Since unattached filters are not copied back to user 1378 * space through sk_get_filter(), we do not need to hold 1379 * a copy here, and can spare us the work. 1380 */ 1381 fp->orig_prog = NULL; 1382 1383 /* bpf_prepare_filter() already takes care of freeing 1384 * memory in case something goes wrong. 1385 */ 1386 fp = bpf_prepare_filter(fp, NULL); 1387 if (IS_ERR(fp)) 1388 return PTR_ERR(fp); 1389 1390 *pfp = fp; 1391 return 0; 1392 } 1393 EXPORT_SYMBOL_GPL(bpf_prog_create); 1394 1395 /** 1396 * bpf_prog_create_from_user - create an unattached filter from user buffer 1397 * @pfp: the unattached filter that is created 1398 * @fprog: the filter program 1399 * @trans: post-classic verifier transformation handler 1400 * @save_orig: save classic BPF program 1401 * 1402 * This function effectively does the same as bpf_prog_create(), only 1403 * that it builds up its insns buffer from user space provided buffer. 1404 * It also allows for passing a bpf_aux_classic_check_t handler. 1405 */ 1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1407 bpf_aux_classic_check_t trans, bool save_orig) 1408 { 1409 unsigned int fsize = bpf_classic_proglen(fprog); 1410 struct bpf_prog *fp; 1411 int err; 1412 1413 /* Make sure new filter is there and in the right amounts. */ 1414 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1415 return -EINVAL; 1416 1417 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1418 if (!fp) 1419 return -ENOMEM; 1420 1421 if (copy_from_user(fp->insns, fprog->filter, fsize)) { 1422 __bpf_prog_free(fp); 1423 return -EFAULT; 1424 } 1425 1426 fp->len = fprog->len; 1427 fp->orig_prog = NULL; 1428 1429 if (save_orig) { 1430 err = bpf_prog_store_orig_filter(fp, fprog); 1431 if (err) { 1432 __bpf_prog_free(fp); 1433 return -ENOMEM; 1434 } 1435 } 1436 1437 /* bpf_prepare_filter() already takes care of freeing 1438 * memory in case something goes wrong. 1439 */ 1440 fp = bpf_prepare_filter(fp, trans); 1441 if (IS_ERR(fp)) 1442 return PTR_ERR(fp); 1443 1444 *pfp = fp; 1445 return 0; 1446 } 1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); 1448 1449 void bpf_prog_destroy(struct bpf_prog *fp) 1450 { 1451 __bpf_prog_release(fp); 1452 } 1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy); 1454 1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) 1456 { 1457 struct sk_filter *fp, *old_fp; 1458 1459 fp = kmalloc(sizeof(*fp), GFP_KERNEL); 1460 if (!fp) 1461 return -ENOMEM; 1462 1463 fp->prog = prog; 1464 1465 if (!__sk_filter_charge(sk, fp)) { 1466 kfree(fp); 1467 return -ENOMEM; 1468 } 1469 refcount_set(&fp->refcnt, 1); 1470 1471 old_fp = rcu_dereference_protected(sk->sk_filter, 1472 lockdep_sock_is_held(sk)); 1473 rcu_assign_pointer(sk->sk_filter, fp); 1474 1475 if (old_fp) 1476 sk_filter_uncharge(sk, old_fp); 1477 1478 return 0; 1479 } 1480 1481 static 1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) 1483 { 1484 unsigned int fsize = bpf_classic_proglen(fprog); 1485 struct bpf_prog *prog; 1486 int err; 1487 1488 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1489 return ERR_PTR(-EPERM); 1490 1491 /* Make sure new filter is there and in the right amounts. */ 1492 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1493 return ERR_PTR(-EINVAL); 1494 1495 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1496 if (!prog) 1497 return ERR_PTR(-ENOMEM); 1498 1499 if (copy_from_user(prog->insns, fprog->filter, fsize)) { 1500 __bpf_prog_free(prog); 1501 return ERR_PTR(-EFAULT); 1502 } 1503 1504 prog->len = fprog->len; 1505 1506 err = bpf_prog_store_orig_filter(prog, fprog); 1507 if (err) { 1508 __bpf_prog_free(prog); 1509 return ERR_PTR(-ENOMEM); 1510 } 1511 1512 /* bpf_prepare_filter() already takes care of freeing 1513 * memory in case something goes wrong. 1514 */ 1515 return bpf_prepare_filter(prog, NULL); 1516 } 1517 1518 /** 1519 * sk_attach_filter - attach a socket filter 1520 * @fprog: the filter program 1521 * @sk: the socket to use 1522 * 1523 * Attach the user's filter code. We first run some sanity checks on 1524 * it to make sure it does not explode on us later. If an error 1525 * occurs or there is insufficient memory for the filter a negative 1526 * errno code is returned. On success the return is zero. 1527 */ 1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1529 { 1530 struct bpf_prog *prog = __get_filter(fprog, sk); 1531 int err; 1532 1533 if (IS_ERR(prog)) 1534 return PTR_ERR(prog); 1535 1536 err = __sk_attach_prog(prog, sk); 1537 if (err < 0) { 1538 __bpf_prog_release(prog); 1539 return err; 1540 } 1541 1542 return 0; 1543 } 1544 EXPORT_SYMBOL_GPL(sk_attach_filter); 1545 1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1547 { 1548 struct bpf_prog *prog = __get_filter(fprog, sk); 1549 int err; 1550 1551 if (IS_ERR(prog)) 1552 return PTR_ERR(prog); 1553 1554 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) 1555 err = -ENOMEM; 1556 else 1557 err = reuseport_attach_prog(sk, prog); 1558 1559 if (err) 1560 __bpf_prog_release(prog); 1561 1562 return err; 1563 } 1564 1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) 1566 { 1567 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1568 return ERR_PTR(-EPERM); 1569 1570 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1571 } 1572 1573 int sk_attach_bpf(u32 ufd, struct sock *sk) 1574 { 1575 struct bpf_prog *prog = __get_bpf(ufd, sk); 1576 int err; 1577 1578 if (IS_ERR(prog)) 1579 return PTR_ERR(prog); 1580 1581 err = __sk_attach_prog(prog, sk); 1582 if (err < 0) { 1583 bpf_prog_put(prog); 1584 return err; 1585 } 1586 1587 return 0; 1588 } 1589 1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) 1591 { 1592 struct bpf_prog *prog; 1593 int err; 1594 1595 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1596 return -EPERM; 1597 1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1599 if (PTR_ERR(prog) == -EINVAL) 1600 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT); 1601 if (IS_ERR(prog)) 1602 return PTR_ERR(prog); 1603 1604 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) { 1605 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER 1606 * bpf prog (e.g. sockmap). It depends on the 1607 * limitation imposed by bpf_prog_load(). 1608 * Hence, sysctl_optmem_max is not checked. 1609 */ 1610 if ((sk->sk_type != SOCK_STREAM && 1611 sk->sk_type != SOCK_DGRAM) || 1612 (sk->sk_protocol != IPPROTO_UDP && 1613 sk->sk_protocol != IPPROTO_TCP) || 1614 (sk->sk_family != AF_INET && 1615 sk->sk_family != AF_INET6)) { 1616 err = -ENOTSUPP; 1617 goto err_prog_put; 1618 } 1619 } else { 1620 /* BPF_PROG_TYPE_SOCKET_FILTER */ 1621 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) { 1622 err = -ENOMEM; 1623 goto err_prog_put; 1624 } 1625 } 1626 1627 err = reuseport_attach_prog(sk, prog); 1628 err_prog_put: 1629 if (err) 1630 bpf_prog_put(prog); 1631 1632 return err; 1633 } 1634 1635 void sk_reuseport_prog_free(struct bpf_prog *prog) 1636 { 1637 if (!prog) 1638 return; 1639 1640 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) 1641 bpf_prog_put(prog); 1642 else 1643 bpf_prog_destroy(prog); 1644 } 1645 1646 struct bpf_scratchpad { 1647 union { 1648 __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; 1649 u8 buff[MAX_BPF_STACK]; 1650 }; 1651 }; 1652 1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); 1654 1655 static inline int __bpf_try_make_writable(struct sk_buff *skb, 1656 unsigned int write_len) 1657 { 1658 return skb_ensure_writable(skb, write_len); 1659 } 1660 1661 static inline int bpf_try_make_writable(struct sk_buff *skb, 1662 unsigned int write_len) 1663 { 1664 int err = __bpf_try_make_writable(skb, write_len); 1665 1666 bpf_compute_data_pointers(skb); 1667 return err; 1668 } 1669 1670 static int bpf_try_make_head_writable(struct sk_buff *skb) 1671 { 1672 return bpf_try_make_writable(skb, skb_headlen(skb)); 1673 } 1674 1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb) 1676 { 1677 if (skb_at_tc_ingress(skb)) 1678 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1679 } 1680 1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) 1682 { 1683 if (skb_at_tc_ingress(skb)) 1684 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1685 } 1686 1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset, 1688 const void *, from, u32, len, u64, flags) 1689 { 1690 void *ptr; 1691 1692 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) 1693 return -EINVAL; 1694 if (unlikely(offset > INT_MAX)) 1695 return -EFAULT; 1696 if (unlikely(bpf_try_make_writable(skb, offset + len))) 1697 return -EFAULT; 1698 1699 ptr = skb->data + offset; 1700 if (flags & BPF_F_RECOMPUTE_CSUM) 1701 __skb_postpull_rcsum(skb, ptr, len, offset); 1702 1703 memcpy(ptr, from, len); 1704 1705 if (flags & BPF_F_RECOMPUTE_CSUM) 1706 __skb_postpush_rcsum(skb, ptr, len, offset); 1707 if (flags & BPF_F_INVALIDATE_HASH) 1708 skb_clear_hash(skb); 1709 1710 return 0; 1711 } 1712 1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = { 1714 .func = bpf_skb_store_bytes, 1715 .gpl_only = false, 1716 .ret_type = RET_INTEGER, 1717 .arg1_type = ARG_PTR_TO_CTX, 1718 .arg2_type = ARG_ANYTHING, 1719 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 1720 .arg4_type = ARG_CONST_SIZE, 1721 .arg5_type = ARG_ANYTHING, 1722 }; 1723 1724 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset, 1725 void *, to, u32, len) 1726 { 1727 void *ptr; 1728 1729 if (unlikely(offset > INT_MAX)) 1730 goto err_clear; 1731 1732 ptr = skb_header_pointer(skb, offset, len, to); 1733 if (unlikely(!ptr)) 1734 goto err_clear; 1735 if (ptr != to) 1736 memcpy(to, ptr, len); 1737 1738 return 0; 1739 err_clear: 1740 memset(to, 0, len); 1741 return -EFAULT; 1742 } 1743 1744 static const struct bpf_func_proto bpf_skb_load_bytes_proto = { 1745 .func = bpf_skb_load_bytes, 1746 .gpl_only = false, 1747 .ret_type = RET_INTEGER, 1748 .arg1_type = ARG_PTR_TO_CTX, 1749 .arg2_type = ARG_ANYTHING, 1750 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1751 .arg4_type = ARG_CONST_SIZE, 1752 }; 1753 1754 BPF_CALL_4(bpf_flow_dissector_load_bytes, 1755 const struct bpf_flow_dissector *, ctx, u32, offset, 1756 void *, to, u32, len) 1757 { 1758 void *ptr; 1759 1760 if (unlikely(offset > 0xffff)) 1761 goto err_clear; 1762 1763 if (unlikely(!ctx->skb)) 1764 goto err_clear; 1765 1766 ptr = skb_header_pointer(ctx->skb, offset, len, to); 1767 if (unlikely(!ptr)) 1768 goto err_clear; 1769 if (ptr != to) 1770 memcpy(to, ptr, len); 1771 1772 return 0; 1773 err_clear: 1774 memset(to, 0, len); 1775 return -EFAULT; 1776 } 1777 1778 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = { 1779 .func = bpf_flow_dissector_load_bytes, 1780 .gpl_only = false, 1781 .ret_type = RET_INTEGER, 1782 .arg1_type = ARG_PTR_TO_CTX, 1783 .arg2_type = ARG_ANYTHING, 1784 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1785 .arg4_type = ARG_CONST_SIZE, 1786 }; 1787 1788 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb, 1789 u32, offset, void *, to, u32, len, u32, start_header) 1790 { 1791 u8 *end = skb_tail_pointer(skb); 1792 u8 *start, *ptr; 1793 1794 if (unlikely(offset > 0xffff)) 1795 goto err_clear; 1796 1797 switch (start_header) { 1798 case BPF_HDR_START_MAC: 1799 if (unlikely(!skb_mac_header_was_set(skb))) 1800 goto err_clear; 1801 start = skb_mac_header(skb); 1802 break; 1803 case BPF_HDR_START_NET: 1804 start = skb_network_header(skb); 1805 break; 1806 default: 1807 goto err_clear; 1808 } 1809 1810 ptr = start + offset; 1811 1812 if (likely(ptr + len <= end)) { 1813 memcpy(to, ptr, len); 1814 return 0; 1815 } 1816 1817 err_clear: 1818 memset(to, 0, len); 1819 return -EFAULT; 1820 } 1821 1822 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = { 1823 .func = bpf_skb_load_bytes_relative, 1824 .gpl_only = false, 1825 .ret_type = RET_INTEGER, 1826 .arg1_type = ARG_PTR_TO_CTX, 1827 .arg2_type = ARG_ANYTHING, 1828 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1829 .arg4_type = ARG_CONST_SIZE, 1830 .arg5_type = ARG_ANYTHING, 1831 }; 1832 1833 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len) 1834 { 1835 /* Idea is the following: should the needed direct read/write 1836 * test fail during runtime, we can pull in more data and redo 1837 * again, since implicitly, we invalidate previous checks here. 1838 * 1839 * Or, since we know how much we need to make read/writeable, 1840 * this can be done once at the program beginning for direct 1841 * access case. By this we overcome limitations of only current 1842 * headroom being accessible. 1843 */ 1844 return bpf_try_make_writable(skb, len ? : skb_headlen(skb)); 1845 } 1846 1847 static const struct bpf_func_proto bpf_skb_pull_data_proto = { 1848 .func = bpf_skb_pull_data, 1849 .gpl_only = false, 1850 .ret_type = RET_INTEGER, 1851 .arg1_type = ARG_PTR_TO_CTX, 1852 .arg2_type = ARG_ANYTHING, 1853 }; 1854 1855 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk) 1856 { 1857 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL; 1858 } 1859 1860 static const struct bpf_func_proto bpf_sk_fullsock_proto = { 1861 .func = bpf_sk_fullsock, 1862 .gpl_only = false, 1863 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 1864 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 1865 }; 1866 1867 static inline int sk_skb_try_make_writable(struct sk_buff *skb, 1868 unsigned int write_len) 1869 { 1870 return __bpf_try_make_writable(skb, write_len); 1871 } 1872 1873 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len) 1874 { 1875 /* Idea is the following: should the needed direct read/write 1876 * test fail during runtime, we can pull in more data and redo 1877 * again, since implicitly, we invalidate previous checks here. 1878 * 1879 * Or, since we know how much we need to make read/writeable, 1880 * this can be done once at the program beginning for direct 1881 * access case. By this we overcome limitations of only current 1882 * headroom being accessible. 1883 */ 1884 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb)); 1885 } 1886 1887 static const struct bpf_func_proto sk_skb_pull_data_proto = { 1888 .func = sk_skb_pull_data, 1889 .gpl_only = false, 1890 .ret_type = RET_INTEGER, 1891 .arg1_type = ARG_PTR_TO_CTX, 1892 .arg2_type = ARG_ANYTHING, 1893 }; 1894 1895 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset, 1896 u64, from, u64, to, u64, flags) 1897 { 1898 __sum16 *ptr; 1899 1900 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) 1901 return -EINVAL; 1902 if (unlikely(offset > 0xffff || offset & 1)) 1903 return -EFAULT; 1904 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1905 return -EFAULT; 1906 1907 ptr = (__sum16 *)(skb->data + offset); 1908 switch (flags & BPF_F_HDR_FIELD_MASK) { 1909 case 0: 1910 if (unlikely(from != 0)) 1911 return -EINVAL; 1912 1913 csum_replace_by_diff(ptr, to); 1914 break; 1915 case 2: 1916 csum_replace2(ptr, from, to); 1917 break; 1918 case 4: 1919 csum_replace4(ptr, from, to); 1920 break; 1921 default: 1922 return -EINVAL; 1923 } 1924 1925 return 0; 1926 } 1927 1928 static const struct bpf_func_proto bpf_l3_csum_replace_proto = { 1929 .func = bpf_l3_csum_replace, 1930 .gpl_only = false, 1931 .ret_type = RET_INTEGER, 1932 .arg1_type = ARG_PTR_TO_CTX, 1933 .arg2_type = ARG_ANYTHING, 1934 .arg3_type = ARG_ANYTHING, 1935 .arg4_type = ARG_ANYTHING, 1936 .arg5_type = ARG_ANYTHING, 1937 }; 1938 1939 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset, 1940 u64, from, u64, to, u64, flags) 1941 { 1942 bool is_pseudo = flags & BPF_F_PSEUDO_HDR; 1943 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; 1944 bool do_mforce = flags & BPF_F_MARK_ENFORCE; 1945 __sum16 *ptr; 1946 1947 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE | 1948 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK))) 1949 return -EINVAL; 1950 if (unlikely(offset > 0xffff || offset & 1)) 1951 return -EFAULT; 1952 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1953 return -EFAULT; 1954 1955 ptr = (__sum16 *)(skb->data + offset); 1956 if (is_mmzero && !do_mforce && !*ptr) 1957 return 0; 1958 1959 switch (flags & BPF_F_HDR_FIELD_MASK) { 1960 case 0: 1961 if (unlikely(from != 0)) 1962 return -EINVAL; 1963 1964 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); 1965 break; 1966 case 2: 1967 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); 1968 break; 1969 case 4: 1970 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); 1971 break; 1972 default: 1973 return -EINVAL; 1974 } 1975 1976 if (is_mmzero && !*ptr) 1977 *ptr = CSUM_MANGLED_0; 1978 return 0; 1979 } 1980 1981 static const struct bpf_func_proto bpf_l4_csum_replace_proto = { 1982 .func = bpf_l4_csum_replace, 1983 .gpl_only = false, 1984 .ret_type = RET_INTEGER, 1985 .arg1_type = ARG_PTR_TO_CTX, 1986 .arg2_type = ARG_ANYTHING, 1987 .arg3_type = ARG_ANYTHING, 1988 .arg4_type = ARG_ANYTHING, 1989 .arg5_type = ARG_ANYTHING, 1990 }; 1991 1992 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size, 1993 __be32 *, to, u32, to_size, __wsum, seed) 1994 { 1995 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); 1996 u32 diff_size = from_size + to_size; 1997 int i, j = 0; 1998 1999 /* This is quite flexible, some examples: 2000 * 2001 * from_size == 0, to_size > 0, seed := csum --> pushing data 2002 * from_size > 0, to_size == 0, seed := csum --> pulling data 2003 * from_size > 0, to_size > 0, seed := 0 --> diffing data 2004 * 2005 * Even for diffing, from_size and to_size don't need to be equal. 2006 */ 2007 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || 2008 diff_size > sizeof(sp->diff))) 2009 return -EINVAL; 2010 2011 for (i = 0; i < from_size / sizeof(__be32); i++, j++) 2012 sp->diff[j] = ~from[i]; 2013 for (i = 0; i < to_size / sizeof(__be32); i++, j++) 2014 sp->diff[j] = to[i]; 2015 2016 return csum_partial(sp->diff, diff_size, seed); 2017 } 2018 2019 static const struct bpf_func_proto bpf_csum_diff_proto = { 2020 .func = bpf_csum_diff, 2021 .gpl_only = false, 2022 .pkt_access = true, 2023 .ret_type = RET_INTEGER, 2024 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2025 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 2026 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2027 .arg4_type = ARG_CONST_SIZE_OR_ZERO, 2028 .arg5_type = ARG_ANYTHING, 2029 }; 2030 2031 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum) 2032 { 2033 /* The interface is to be used in combination with bpf_csum_diff() 2034 * for direct packet writes. csum rotation for alignment as well 2035 * as emulating csum_sub() can be done from the eBPF program. 2036 */ 2037 if (skb->ip_summed == CHECKSUM_COMPLETE) 2038 return (skb->csum = csum_add(skb->csum, csum)); 2039 2040 return -ENOTSUPP; 2041 } 2042 2043 static const struct bpf_func_proto bpf_csum_update_proto = { 2044 .func = bpf_csum_update, 2045 .gpl_only = false, 2046 .ret_type = RET_INTEGER, 2047 .arg1_type = ARG_PTR_TO_CTX, 2048 .arg2_type = ARG_ANYTHING, 2049 }; 2050 2051 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level) 2052 { 2053 /* The interface is to be used in combination with bpf_skb_adjust_room() 2054 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET 2055 * is passed as flags, for example. 2056 */ 2057 switch (level) { 2058 case BPF_CSUM_LEVEL_INC: 2059 __skb_incr_checksum_unnecessary(skb); 2060 break; 2061 case BPF_CSUM_LEVEL_DEC: 2062 __skb_decr_checksum_unnecessary(skb); 2063 break; 2064 case BPF_CSUM_LEVEL_RESET: 2065 __skb_reset_checksum_unnecessary(skb); 2066 break; 2067 case BPF_CSUM_LEVEL_QUERY: 2068 return skb->ip_summed == CHECKSUM_UNNECESSARY ? 2069 skb->csum_level : -EACCES; 2070 default: 2071 return -EINVAL; 2072 } 2073 2074 return 0; 2075 } 2076 2077 static const struct bpf_func_proto bpf_csum_level_proto = { 2078 .func = bpf_csum_level, 2079 .gpl_only = false, 2080 .ret_type = RET_INTEGER, 2081 .arg1_type = ARG_PTR_TO_CTX, 2082 .arg2_type = ARG_ANYTHING, 2083 }; 2084 2085 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) 2086 { 2087 return dev_forward_skb_nomtu(dev, skb); 2088 } 2089 2090 static inline int __bpf_rx_skb_no_mac(struct net_device *dev, 2091 struct sk_buff *skb) 2092 { 2093 int ret = ____dev_forward_skb(dev, skb, false); 2094 2095 if (likely(!ret)) { 2096 skb->dev = dev; 2097 ret = netif_rx(skb); 2098 } 2099 2100 return ret; 2101 } 2102 2103 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) 2104 { 2105 int ret; 2106 2107 if (dev_xmit_recursion()) { 2108 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2109 kfree_skb(skb); 2110 return -ENETDOWN; 2111 } 2112 2113 skb->dev = dev; 2114 skb_clear_tstamp(skb); 2115 2116 dev_xmit_recursion_inc(); 2117 ret = dev_queue_xmit(skb); 2118 dev_xmit_recursion_dec(); 2119 2120 return ret; 2121 } 2122 2123 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 2124 u32 flags) 2125 { 2126 unsigned int mlen = skb_network_offset(skb); 2127 2128 if (unlikely(skb->len <= mlen)) { 2129 kfree_skb(skb); 2130 return -ERANGE; 2131 } 2132 2133 if (mlen) { 2134 __skb_pull(skb, mlen); 2135 2136 /* At ingress, the mac header has already been pulled once. 2137 * At egress, skb_pospull_rcsum has to be done in case that 2138 * the skb is originated from ingress (i.e. a forwarded skb) 2139 * to ensure that rcsum starts at net header. 2140 */ 2141 if (!skb_at_tc_ingress(skb)) 2142 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 2143 } 2144 skb_pop_mac_header(skb); 2145 skb_reset_mac_len(skb); 2146 return flags & BPF_F_INGRESS ? 2147 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 2148 } 2149 2150 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 2151 u32 flags) 2152 { 2153 /* Verify that a link layer header is carried */ 2154 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) { 2155 kfree_skb(skb); 2156 return -ERANGE; 2157 } 2158 2159 bpf_push_mac_rcsum(skb); 2160 return flags & BPF_F_INGRESS ? 2161 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 2162 } 2163 2164 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 2165 u32 flags) 2166 { 2167 if (dev_is_mac_header_xmit(dev)) 2168 return __bpf_redirect_common(skb, dev, flags); 2169 else 2170 return __bpf_redirect_no_mac(skb, dev, flags); 2171 } 2172 2173 #if IS_ENABLED(CONFIG_IPV6) 2174 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb, 2175 struct net_device *dev, struct bpf_nh_params *nh) 2176 { 2177 u32 hh_len = LL_RESERVED_SPACE(dev); 2178 const struct in6_addr *nexthop; 2179 struct dst_entry *dst = NULL; 2180 struct neighbour *neigh; 2181 2182 if (dev_xmit_recursion()) { 2183 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2184 goto out_drop; 2185 } 2186 2187 skb->dev = dev; 2188 skb_clear_tstamp(skb); 2189 2190 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2191 skb = skb_expand_head(skb, hh_len); 2192 if (!skb) 2193 return -ENOMEM; 2194 } 2195 2196 rcu_read_lock_bh(); 2197 if (!nh) { 2198 dst = skb_dst(skb); 2199 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst), 2200 &ipv6_hdr(skb)->daddr); 2201 } else { 2202 nexthop = &nh->ipv6_nh; 2203 } 2204 neigh = ip_neigh_gw6(dev, nexthop); 2205 if (likely(!IS_ERR(neigh))) { 2206 int ret; 2207 2208 sock_confirm_neigh(skb, neigh); 2209 dev_xmit_recursion_inc(); 2210 ret = neigh_output(neigh, skb, false); 2211 dev_xmit_recursion_dec(); 2212 rcu_read_unlock_bh(); 2213 return ret; 2214 } 2215 rcu_read_unlock_bh(); 2216 if (dst) 2217 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 2218 out_drop: 2219 kfree_skb(skb); 2220 return -ENETDOWN; 2221 } 2222 2223 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2224 struct bpf_nh_params *nh) 2225 { 2226 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 2227 struct net *net = dev_net(dev); 2228 int err, ret = NET_XMIT_DROP; 2229 2230 if (!nh) { 2231 struct dst_entry *dst; 2232 struct flowi6 fl6 = { 2233 .flowi6_flags = FLOWI_FLAG_ANYSRC, 2234 .flowi6_mark = skb->mark, 2235 .flowlabel = ip6_flowinfo(ip6h), 2236 .flowi6_oif = dev->ifindex, 2237 .flowi6_proto = ip6h->nexthdr, 2238 .daddr = ip6h->daddr, 2239 .saddr = ip6h->saddr, 2240 }; 2241 2242 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL); 2243 if (IS_ERR(dst)) 2244 goto out_drop; 2245 2246 skb_dst_set(skb, dst); 2247 } else if (nh->nh_family != AF_INET6) { 2248 goto out_drop; 2249 } 2250 2251 err = bpf_out_neigh_v6(net, skb, dev, nh); 2252 if (unlikely(net_xmit_eval(err))) 2253 dev->stats.tx_errors++; 2254 else 2255 ret = NET_XMIT_SUCCESS; 2256 goto out_xmit; 2257 out_drop: 2258 dev->stats.tx_errors++; 2259 kfree_skb(skb); 2260 out_xmit: 2261 return ret; 2262 } 2263 #else 2264 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2265 struct bpf_nh_params *nh) 2266 { 2267 kfree_skb(skb); 2268 return NET_XMIT_DROP; 2269 } 2270 #endif /* CONFIG_IPV6 */ 2271 2272 #if IS_ENABLED(CONFIG_INET) 2273 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb, 2274 struct net_device *dev, struct bpf_nh_params *nh) 2275 { 2276 u32 hh_len = LL_RESERVED_SPACE(dev); 2277 struct neighbour *neigh; 2278 bool is_v6gw = false; 2279 2280 if (dev_xmit_recursion()) { 2281 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2282 goto out_drop; 2283 } 2284 2285 skb->dev = dev; 2286 skb_clear_tstamp(skb); 2287 2288 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2289 skb = skb_expand_head(skb, hh_len); 2290 if (!skb) 2291 return -ENOMEM; 2292 } 2293 2294 rcu_read_lock_bh(); 2295 if (!nh) { 2296 struct dst_entry *dst = skb_dst(skb); 2297 struct rtable *rt = container_of(dst, struct rtable, dst); 2298 2299 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 2300 } else if (nh->nh_family == AF_INET6) { 2301 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh); 2302 is_v6gw = true; 2303 } else if (nh->nh_family == AF_INET) { 2304 neigh = ip_neigh_gw4(dev, nh->ipv4_nh); 2305 } else { 2306 rcu_read_unlock_bh(); 2307 goto out_drop; 2308 } 2309 2310 if (likely(!IS_ERR(neigh))) { 2311 int ret; 2312 2313 sock_confirm_neigh(skb, neigh); 2314 dev_xmit_recursion_inc(); 2315 ret = neigh_output(neigh, skb, is_v6gw); 2316 dev_xmit_recursion_dec(); 2317 rcu_read_unlock_bh(); 2318 return ret; 2319 } 2320 rcu_read_unlock_bh(); 2321 out_drop: 2322 kfree_skb(skb); 2323 return -ENETDOWN; 2324 } 2325 2326 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2327 struct bpf_nh_params *nh) 2328 { 2329 const struct iphdr *ip4h = ip_hdr(skb); 2330 struct net *net = dev_net(dev); 2331 int err, ret = NET_XMIT_DROP; 2332 2333 if (!nh) { 2334 struct flowi4 fl4 = { 2335 .flowi4_flags = FLOWI_FLAG_ANYSRC, 2336 .flowi4_mark = skb->mark, 2337 .flowi4_tos = RT_TOS(ip4h->tos), 2338 .flowi4_oif = dev->ifindex, 2339 .flowi4_proto = ip4h->protocol, 2340 .daddr = ip4h->daddr, 2341 .saddr = ip4h->saddr, 2342 }; 2343 struct rtable *rt; 2344 2345 rt = ip_route_output_flow(net, &fl4, NULL); 2346 if (IS_ERR(rt)) 2347 goto out_drop; 2348 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 2349 ip_rt_put(rt); 2350 goto out_drop; 2351 } 2352 2353 skb_dst_set(skb, &rt->dst); 2354 } 2355 2356 err = bpf_out_neigh_v4(net, skb, dev, nh); 2357 if (unlikely(net_xmit_eval(err))) 2358 dev->stats.tx_errors++; 2359 else 2360 ret = NET_XMIT_SUCCESS; 2361 goto out_xmit; 2362 out_drop: 2363 dev->stats.tx_errors++; 2364 kfree_skb(skb); 2365 out_xmit: 2366 return ret; 2367 } 2368 #else 2369 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2370 struct bpf_nh_params *nh) 2371 { 2372 kfree_skb(skb); 2373 return NET_XMIT_DROP; 2374 } 2375 #endif /* CONFIG_INET */ 2376 2377 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev, 2378 struct bpf_nh_params *nh) 2379 { 2380 struct ethhdr *ethh = eth_hdr(skb); 2381 2382 if (unlikely(skb->mac_header >= skb->network_header)) 2383 goto out; 2384 bpf_push_mac_rcsum(skb); 2385 if (is_multicast_ether_addr(ethh->h_dest)) 2386 goto out; 2387 2388 skb_pull(skb, sizeof(*ethh)); 2389 skb_unset_mac_header(skb); 2390 skb_reset_network_header(skb); 2391 2392 if (skb->protocol == htons(ETH_P_IP)) 2393 return __bpf_redirect_neigh_v4(skb, dev, nh); 2394 else if (skb->protocol == htons(ETH_P_IPV6)) 2395 return __bpf_redirect_neigh_v6(skb, dev, nh); 2396 out: 2397 kfree_skb(skb); 2398 return -ENOTSUPP; 2399 } 2400 2401 /* Internal, non-exposed redirect flags. */ 2402 enum { 2403 BPF_F_NEIGH = (1ULL << 1), 2404 BPF_F_PEER = (1ULL << 2), 2405 BPF_F_NEXTHOP = (1ULL << 3), 2406 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP) 2407 }; 2408 2409 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 2410 { 2411 struct net_device *dev; 2412 struct sk_buff *clone; 2413 int ret; 2414 2415 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2416 return -EINVAL; 2417 2418 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 2419 if (unlikely(!dev)) 2420 return -EINVAL; 2421 2422 clone = skb_clone(skb, GFP_ATOMIC); 2423 if (unlikely(!clone)) 2424 return -ENOMEM; 2425 2426 /* For direct write, we need to keep the invariant that the skbs 2427 * we're dealing with need to be uncloned. Should uncloning fail 2428 * here, we need to free the just generated clone to unclone once 2429 * again. 2430 */ 2431 ret = bpf_try_make_head_writable(skb); 2432 if (unlikely(ret)) { 2433 kfree_skb(clone); 2434 return -ENOMEM; 2435 } 2436 2437 return __bpf_redirect(clone, dev, flags); 2438 } 2439 2440 static const struct bpf_func_proto bpf_clone_redirect_proto = { 2441 .func = bpf_clone_redirect, 2442 .gpl_only = false, 2443 .ret_type = RET_INTEGER, 2444 .arg1_type = ARG_PTR_TO_CTX, 2445 .arg2_type = ARG_ANYTHING, 2446 .arg3_type = ARG_ANYTHING, 2447 }; 2448 2449 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 2450 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); 2451 2452 int skb_do_redirect(struct sk_buff *skb) 2453 { 2454 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2455 struct net *net = dev_net(skb->dev); 2456 struct net_device *dev; 2457 u32 flags = ri->flags; 2458 2459 dev = dev_get_by_index_rcu(net, ri->tgt_index); 2460 ri->tgt_index = 0; 2461 ri->flags = 0; 2462 if (unlikely(!dev)) 2463 goto out_drop; 2464 if (flags & BPF_F_PEER) { 2465 const struct net_device_ops *ops = dev->netdev_ops; 2466 2467 if (unlikely(!ops->ndo_get_peer_dev || 2468 !skb_at_tc_ingress(skb))) 2469 goto out_drop; 2470 dev = ops->ndo_get_peer_dev(dev); 2471 if (unlikely(!dev || 2472 !(dev->flags & IFF_UP) || 2473 net_eq(net, dev_net(dev)))) 2474 goto out_drop; 2475 skb->dev = dev; 2476 return -EAGAIN; 2477 } 2478 return flags & BPF_F_NEIGH ? 2479 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ? 2480 &ri->nh : NULL) : 2481 __bpf_redirect(skb, dev, flags); 2482 out_drop: 2483 kfree_skb(skb); 2484 return -EINVAL; 2485 } 2486 2487 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 2488 { 2489 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2490 2491 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2492 return TC_ACT_SHOT; 2493 2494 ri->flags = flags; 2495 ri->tgt_index = ifindex; 2496 2497 return TC_ACT_REDIRECT; 2498 } 2499 2500 static const struct bpf_func_proto bpf_redirect_proto = { 2501 .func = bpf_redirect, 2502 .gpl_only = false, 2503 .ret_type = RET_INTEGER, 2504 .arg1_type = ARG_ANYTHING, 2505 .arg2_type = ARG_ANYTHING, 2506 }; 2507 2508 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags) 2509 { 2510 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2511 2512 if (unlikely(flags)) 2513 return TC_ACT_SHOT; 2514 2515 ri->flags = BPF_F_PEER; 2516 ri->tgt_index = ifindex; 2517 2518 return TC_ACT_REDIRECT; 2519 } 2520 2521 static const struct bpf_func_proto bpf_redirect_peer_proto = { 2522 .func = bpf_redirect_peer, 2523 .gpl_only = false, 2524 .ret_type = RET_INTEGER, 2525 .arg1_type = ARG_ANYTHING, 2526 .arg2_type = ARG_ANYTHING, 2527 }; 2528 2529 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params, 2530 int, plen, u64, flags) 2531 { 2532 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2533 2534 if (unlikely((plen && plen < sizeof(*params)) || flags)) 2535 return TC_ACT_SHOT; 2536 2537 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0); 2538 ri->tgt_index = ifindex; 2539 2540 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params)); 2541 if (plen) 2542 memcpy(&ri->nh, params, sizeof(ri->nh)); 2543 2544 return TC_ACT_REDIRECT; 2545 } 2546 2547 static const struct bpf_func_proto bpf_redirect_neigh_proto = { 2548 .func = bpf_redirect_neigh, 2549 .gpl_only = false, 2550 .ret_type = RET_INTEGER, 2551 .arg1_type = ARG_ANYTHING, 2552 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2553 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 2554 .arg4_type = ARG_ANYTHING, 2555 }; 2556 2557 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes) 2558 { 2559 msg->apply_bytes = bytes; 2560 return 0; 2561 } 2562 2563 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = { 2564 .func = bpf_msg_apply_bytes, 2565 .gpl_only = false, 2566 .ret_type = RET_INTEGER, 2567 .arg1_type = ARG_PTR_TO_CTX, 2568 .arg2_type = ARG_ANYTHING, 2569 }; 2570 2571 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes) 2572 { 2573 msg->cork_bytes = bytes; 2574 return 0; 2575 } 2576 2577 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = { 2578 .func = bpf_msg_cork_bytes, 2579 .gpl_only = false, 2580 .ret_type = RET_INTEGER, 2581 .arg1_type = ARG_PTR_TO_CTX, 2582 .arg2_type = ARG_ANYTHING, 2583 }; 2584 2585 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start, 2586 u32, end, u64, flags) 2587 { 2588 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start; 2589 u32 first_sge, last_sge, i, shift, bytes_sg_total; 2590 struct scatterlist *sge; 2591 u8 *raw, *to, *from; 2592 struct page *page; 2593 2594 if (unlikely(flags || end <= start)) 2595 return -EINVAL; 2596 2597 /* First find the starting scatterlist element */ 2598 i = msg->sg.start; 2599 do { 2600 offset += len; 2601 len = sk_msg_elem(msg, i)->length; 2602 if (start < offset + len) 2603 break; 2604 sk_msg_iter_var_next(i); 2605 } while (i != msg->sg.end); 2606 2607 if (unlikely(start >= offset + len)) 2608 return -EINVAL; 2609 2610 first_sge = i; 2611 /* The start may point into the sg element so we need to also 2612 * account for the headroom. 2613 */ 2614 bytes_sg_total = start - offset + bytes; 2615 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len) 2616 goto out; 2617 2618 /* At this point we need to linearize multiple scatterlist 2619 * elements or a single shared page. Either way we need to 2620 * copy into a linear buffer exclusively owned by BPF. Then 2621 * place the buffer in the scatterlist and fixup the original 2622 * entries by removing the entries now in the linear buffer 2623 * and shifting the remaining entries. For now we do not try 2624 * to copy partial entries to avoid complexity of running out 2625 * of sg_entry slots. The downside is reading a single byte 2626 * will copy the entire sg entry. 2627 */ 2628 do { 2629 copy += sk_msg_elem(msg, i)->length; 2630 sk_msg_iter_var_next(i); 2631 if (bytes_sg_total <= copy) 2632 break; 2633 } while (i != msg->sg.end); 2634 last_sge = i; 2635 2636 if (unlikely(bytes_sg_total > copy)) 2637 return -EINVAL; 2638 2639 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2640 get_order(copy)); 2641 if (unlikely(!page)) 2642 return -ENOMEM; 2643 2644 raw = page_address(page); 2645 i = first_sge; 2646 do { 2647 sge = sk_msg_elem(msg, i); 2648 from = sg_virt(sge); 2649 len = sge->length; 2650 to = raw + poffset; 2651 2652 memcpy(to, from, len); 2653 poffset += len; 2654 sge->length = 0; 2655 put_page(sg_page(sge)); 2656 2657 sk_msg_iter_var_next(i); 2658 } while (i != last_sge); 2659 2660 sg_set_page(&msg->sg.data[first_sge], page, copy, 0); 2661 2662 /* To repair sg ring we need to shift entries. If we only 2663 * had a single entry though we can just replace it and 2664 * be done. Otherwise walk the ring and shift the entries. 2665 */ 2666 WARN_ON_ONCE(last_sge == first_sge); 2667 shift = last_sge > first_sge ? 2668 last_sge - first_sge - 1 : 2669 NR_MSG_FRAG_IDS - first_sge + last_sge - 1; 2670 if (!shift) 2671 goto out; 2672 2673 i = first_sge; 2674 sk_msg_iter_var_next(i); 2675 do { 2676 u32 move_from; 2677 2678 if (i + shift >= NR_MSG_FRAG_IDS) 2679 move_from = i + shift - NR_MSG_FRAG_IDS; 2680 else 2681 move_from = i + shift; 2682 if (move_from == msg->sg.end) 2683 break; 2684 2685 msg->sg.data[i] = msg->sg.data[move_from]; 2686 msg->sg.data[move_from].length = 0; 2687 msg->sg.data[move_from].page_link = 0; 2688 msg->sg.data[move_from].offset = 0; 2689 sk_msg_iter_var_next(i); 2690 } while (1); 2691 2692 msg->sg.end = msg->sg.end - shift > msg->sg.end ? 2693 msg->sg.end - shift + NR_MSG_FRAG_IDS : 2694 msg->sg.end - shift; 2695 out: 2696 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset; 2697 msg->data_end = msg->data + bytes; 2698 return 0; 2699 } 2700 2701 static const struct bpf_func_proto bpf_msg_pull_data_proto = { 2702 .func = bpf_msg_pull_data, 2703 .gpl_only = false, 2704 .ret_type = RET_INTEGER, 2705 .arg1_type = ARG_PTR_TO_CTX, 2706 .arg2_type = ARG_ANYTHING, 2707 .arg3_type = ARG_ANYTHING, 2708 .arg4_type = ARG_ANYTHING, 2709 }; 2710 2711 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start, 2712 u32, len, u64, flags) 2713 { 2714 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge; 2715 u32 new, i = 0, l = 0, space, copy = 0, offset = 0; 2716 u8 *raw, *to, *from; 2717 struct page *page; 2718 2719 if (unlikely(flags)) 2720 return -EINVAL; 2721 2722 if (unlikely(len == 0)) 2723 return 0; 2724 2725 /* First find the starting scatterlist element */ 2726 i = msg->sg.start; 2727 do { 2728 offset += l; 2729 l = sk_msg_elem(msg, i)->length; 2730 2731 if (start < offset + l) 2732 break; 2733 sk_msg_iter_var_next(i); 2734 } while (i != msg->sg.end); 2735 2736 if (start >= offset + l) 2737 return -EINVAL; 2738 2739 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2740 2741 /* If no space available will fallback to copy, we need at 2742 * least one scatterlist elem available to push data into 2743 * when start aligns to the beginning of an element or two 2744 * when it falls inside an element. We handle the start equals 2745 * offset case because its the common case for inserting a 2746 * header. 2747 */ 2748 if (!space || (space == 1 && start != offset)) 2749 copy = msg->sg.data[i].length; 2750 2751 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2752 get_order(copy + len)); 2753 if (unlikely(!page)) 2754 return -ENOMEM; 2755 2756 if (copy) { 2757 int front, back; 2758 2759 raw = page_address(page); 2760 2761 psge = sk_msg_elem(msg, i); 2762 front = start - offset; 2763 back = psge->length - front; 2764 from = sg_virt(psge); 2765 2766 if (front) 2767 memcpy(raw, from, front); 2768 2769 if (back) { 2770 from += front; 2771 to = raw + front + len; 2772 2773 memcpy(to, from, back); 2774 } 2775 2776 put_page(sg_page(psge)); 2777 } else if (start - offset) { 2778 psge = sk_msg_elem(msg, i); 2779 rsge = sk_msg_elem_cpy(msg, i); 2780 2781 psge->length = start - offset; 2782 rsge.length -= psge->length; 2783 rsge.offset += start; 2784 2785 sk_msg_iter_var_next(i); 2786 sg_unmark_end(psge); 2787 sg_unmark_end(&rsge); 2788 sk_msg_iter_next(msg, end); 2789 } 2790 2791 /* Slot(s) to place newly allocated data */ 2792 new = i; 2793 2794 /* Shift one or two slots as needed */ 2795 if (!copy) { 2796 sge = sk_msg_elem_cpy(msg, i); 2797 2798 sk_msg_iter_var_next(i); 2799 sg_unmark_end(&sge); 2800 sk_msg_iter_next(msg, end); 2801 2802 nsge = sk_msg_elem_cpy(msg, i); 2803 if (rsge.length) { 2804 sk_msg_iter_var_next(i); 2805 nnsge = sk_msg_elem_cpy(msg, i); 2806 } 2807 2808 while (i != msg->sg.end) { 2809 msg->sg.data[i] = sge; 2810 sge = nsge; 2811 sk_msg_iter_var_next(i); 2812 if (rsge.length) { 2813 nsge = nnsge; 2814 nnsge = sk_msg_elem_cpy(msg, i); 2815 } else { 2816 nsge = sk_msg_elem_cpy(msg, i); 2817 } 2818 } 2819 } 2820 2821 /* Place newly allocated data buffer */ 2822 sk_mem_charge(msg->sk, len); 2823 msg->sg.size += len; 2824 __clear_bit(new, msg->sg.copy); 2825 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2826 if (rsge.length) { 2827 get_page(sg_page(&rsge)); 2828 sk_msg_iter_var_next(new); 2829 msg->sg.data[new] = rsge; 2830 } 2831 2832 sk_msg_compute_data_pointers(msg); 2833 return 0; 2834 } 2835 2836 static const struct bpf_func_proto bpf_msg_push_data_proto = { 2837 .func = bpf_msg_push_data, 2838 .gpl_only = false, 2839 .ret_type = RET_INTEGER, 2840 .arg1_type = ARG_PTR_TO_CTX, 2841 .arg2_type = ARG_ANYTHING, 2842 .arg3_type = ARG_ANYTHING, 2843 .arg4_type = ARG_ANYTHING, 2844 }; 2845 2846 static void sk_msg_shift_left(struct sk_msg *msg, int i) 2847 { 2848 int prev; 2849 2850 do { 2851 prev = i; 2852 sk_msg_iter_var_next(i); 2853 msg->sg.data[prev] = msg->sg.data[i]; 2854 } while (i != msg->sg.end); 2855 2856 sk_msg_iter_prev(msg, end); 2857 } 2858 2859 static void sk_msg_shift_right(struct sk_msg *msg, int i) 2860 { 2861 struct scatterlist tmp, sge; 2862 2863 sk_msg_iter_next(msg, end); 2864 sge = sk_msg_elem_cpy(msg, i); 2865 sk_msg_iter_var_next(i); 2866 tmp = sk_msg_elem_cpy(msg, i); 2867 2868 while (i != msg->sg.end) { 2869 msg->sg.data[i] = sge; 2870 sk_msg_iter_var_next(i); 2871 sge = tmp; 2872 tmp = sk_msg_elem_cpy(msg, i); 2873 } 2874 } 2875 2876 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 2877 u32, len, u64, flags) 2878 { 2879 u32 i = 0, l = 0, space, offset = 0; 2880 u64 last = start + len; 2881 int pop; 2882 2883 if (unlikely(flags)) 2884 return -EINVAL; 2885 2886 /* First find the starting scatterlist element */ 2887 i = msg->sg.start; 2888 do { 2889 offset += l; 2890 l = sk_msg_elem(msg, i)->length; 2891 2892 if (start < offset + l) 2893 break; 2894 sk_msg_iter_var_next(i); 2895 } while (i != msg->sg.end); 2896 2897 /* Bounds checks: start and pop must be inside message */ 2898 if (start >= offset + l || last >= msg->sg.size) 2899 return -EINVAL; 2900 2901 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2902 2903 pop = len; 2904 /* --------------| offset 2905 * -| start |-------- len -------| 2906 * 2907 * |----- a ----|-------- pop -------|----- b ----| 2908 * |______________________________________________| length 2909 * 2910 * 2911 * a: region at front of scatter element to save 2912 * b: region at back of scatter element to save when length > A + pop 2913 * pop: region to pop from element, same as input 'pop' here will be 2914 * decremented below per iteration. 2915 * 2916 * Two top-level cases to handle when start != offset, first B is non 2917 * zero and second B is zero corresponding to when a pop includes more 2918 * than one element. 2919 * 2920 * Then if B is non-zero AND there is no space allocate space and 2921 * compact A, B regions into page. If there is space shift ring to 2922 * the rigth free'ing the next element in ring to place B, leaving 2923 * A untouched except to reduce length. 2924 */ 2925 if (start != offset) { 2926 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 2927 int a = start; 2928 int b = sge->length - pop - a; 2929 2930 sk_msg_iter_var_next(i); 2931 2932 if (pop < sge->length - a) { 2933 if (space) { 2934 sge->length = a; 2935 sk_msg_shift_right(msg, i); 2936 nsge = sk_msg_elem(msg, i); 2937 get_page(sg_page(sge)); 2938 sg_set_page(nsge, 2939 sg_page(sge), 2940 b, sge->offset + pop + a); 2941 } else { 2942 struct page *page, *orig; 2943 u8 *to, *from; 2944 2945 page = alloc_pages(__GFP_NOWARN | 2946 __GFP_COMP | GFP_ATOMIC, 2947 get_order(a + b)); 2948 if (unlikely(!page)) 2949 return -ENOMEM; 2950 2951 sge->length = a; 2952 orig = sg_page(sge); 2953 from = sg_virt(sge); 2954 to = page_address(page); 2955 memcpy(to, from, a); 2956 memcpy(to + a, from + a + pop, b); 2957 sg_set_page(sge, page, a + b, 0); 2958 put_page(orig); 2959 } 2960 pop = 0; 2961 } else if (pop >= sge->length - a) { 2962 pop -= (sge->length - a); 2963 sge->length = a; 2964 } 2965 } 2966 2967 /* From above the current layout _must_ be as follows, 2968 * 2969 * -| offset 2970 * -| start 2971 * 2972 * |---- pop ---|---------------- b ------------| 2973 * |____________________________________________| length 2974 * 2975 * Offset and start of the current msg elem are equal because in the 2976 * previous case we handled offset != start and either consumed the 2977 * entire element and advanced to the next element OR pop == 0. 2978 * 2979 * Two cases to handle here are first pop is less than the length 2980 * leaving some remainder b above. Simply adjust the element's layout 2981 * in this case. Or pop >= length of the element so that b = 0. In this 2982 * case advance to next element decrementing pop. 2983 */ 2984 while (pop) { 2985 struct scatterlist *sge = sk_msg_elem(msg, i); 2986 2987 if (pop < sge->length) { 2988 sge->length -= pop; 2989 sge->offset += pop; 2990 pop = 0; 2991 } else { 2992 pop -= sge->length; 2993 sk_msg_shift_left(msg, i); 2994 } 2995 sk_msg_iter_var_next(i); 2996 } 2997 2998 sk_mem_uncharge(msg->sk, len - pop); 2999 msg->sg.size -= (len - pop); 3000 sk_msg_compute_data_pointers(msg); 3001 return 0; 3002 } 3003 3004 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 3005 .func = bpf_msg_pop_data, 3006 .gpl_only = false, 3007 .ret_type = RET_INTEGER, 3008 .arg1_type = ARG_PTR_TO_CTX, 3009 .arg2_type = ARG_ANYTHING, 3010 .arg3_type = ARG_ANYTHING, 3011 .arg4_type = ARG_ANYTHING, 3012 }; 3013 3014 #ifdef CONFIG_CGROUP_NET_CLASSID 3015 BPF_CALL_0(bpf_get_cgroup_classid_curr) 3016 { 3017 return __task_get_classid(current); 3018 } 3019 3020 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { 3021 .func = bpf_get_cgroup_classid_curr, 3022 .gpl_only = false, 3023 .ret_type = RET_INTEGER, 3024 }; 3025 3026 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb) 3027 { 3028 struct sock *sk = skb_to_full_sk(skb); 3029 3030 if (!sk || !sk_fullsock(sk)) 3031 return 0; 3032 3033 return sock_cgroup_classid(&sk->sk_cgrp_data); 3034 } 3035 3036 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = { 3037 .func = bpf_skb_cgroup_classid, 3038 .gpl_only = false, 3039 .ret_type = RET_INTEGER, 3040 .arg1_type = ARG_PTR_TO_CTX, 3041 }; 3042 #endif 3043 3044 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 3045 { 3046 return task_get_classid(skb); 3047 } 3048 3049 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 3050 .func = bpf_get_cgroup_classid, 3051 .gpl_only = false, 3052 .ret_type = RET_INTEGER, 3053 .arg1_type = ARG_PTR_TO_CTX, 3054 }; 3055 3056 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 3057 { 3058 return dst_tclassid(skb); 3059 } 3060 3061 static const struct bpf_func_proto bpf_get_route_realm_proto = { 3062 .func = bpf_get_route_realm, 3063 .gpl_only = false, 3064 .ret_type = RET_INTEGER, 3065 .arg1_type = ARG_PTR_TO_CTX, 3066 }; 3067 3068 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 3069 { 3070 /* If skb_clear_hash() was called due to mangling, we can 3071 * trigger SW recalculation here. Later access to hash 3072 * can then use the inline skb->hash via context directly 3073 * instead of calling this helper again. 3074 */ 3075 return skb_get_hash(skb); 3076 } 3077 3078 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 3079 .func = bpf_get_hash_recalc, 3080 .gpl_only = false, 3081 .ret_type = RET_INTEGER, 3082 .arg1_type = ARG_PTR_TO_CTX, 3083 }; 3084 3085 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 3086 { 3087 /* After all direct packet write, this can be used once for 3088 * triggering a lazy recalc on next skb_get_hash() invocation. 3089 */ 3090 skb_clear_hash(skb); 3091 return 0; 3092 } 3093 3094 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 3095 .func = bpf_set_hash_invalid, 3096 .gpl_only = false, 3097 .ret_type = RET_INTEGER, 3098 .arg1_type = ARG_PTR_TO_CTX, 3099 }; 3100 3101 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 3102 { 3103 /* Set user specified hash as L4(+), so that it gets returned 3104 * on skb_get_hash() call unless BPF prog later on triggers a 3105 * skb_clear_hash(). 3106 */ 3107 __skb_set_sw_hash(skb, hash, true); 3108 return 0; 3109 } 3110 3111 static const struct bpf_func_proto bpf_set_hash_proto = { 3112 .func = bpf_set_hash, 3113 .gpl_only = false, 3114 .ret_type = RET_INTEGER, 3115 .arg1_type = ARG_PTR_TO_CTX, 3116 .arg2_type = ARG_ANYTHING, 3117 }; 3118 3119 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 3120 u16, vlan_tci) 3121 { 3122 int ret; 3123 3124 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 3125 vlan_proto != htons(ETH_P_8021AD))) 3126 vlan_proto = htons(ETH_P_8021Q); 3127 3128 bpf_push_mac_rcsum(skb); 3129 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 3130 bpf_pull_mac_rcsum(skb); 3131 3132 bpf_compute_data_pointers(skb); 3133 return ret; 3134 } 3135 3136 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 3137 .func = bpf_skb_vlan_push, 3138 .gpl_only = false, 3139 .ret_type = RET_INTEGER, 3140 .arg1_type = ARG_PTR_TO_CTX, 3141 .arg2_type = ARG_ANYTHING, 3142 .arg3_type = ARG_ANYTHING, 3143 }; 3144 3145 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 3146 { 3147 int ret; 3148 3149 bpf_push_mac_rcsum(skb); 3150 ret = skb_vlan_pop(skb); 3151 bpf_pull_mac_rcsum(skb); 3152 3153 bpf_compute_data_pointers(skb); 3154 return ret; 3155 } 3156 3157 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 3158 .func = bpf_skb_vlan_pop, 3159 .gpl_only = false, 3160 .ret_type = RET_INTEGER, 3161 .arg1_type = ARG_PTR_TO_CTX, 3162 }; 3163 3164 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 3165 { 3166 /* Caller already did skb_cow() with len as headroom, 3167 * so no need to do it here. 3168 */ 3169 skb_push(skb, len); 3170 memmove(skb->data, skb->data + len, off); 3171 memset(skb->data + off, 0, len); 3172 3173 /* No skb_postpush_rcsum(skb, skb->data + off, len) 3174 * needed here as it does not change the skb->csum 3175 * result for checksum complete when summing over 3176 * zeroed blocks. 3177 */ 3178 return 0; 3179 } 3180 3181 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 3182 { 3183 /* skb_ensure_writable() is not needed here, as we're 3184 * already working on an uncloned skb. 3185 */ 3186 if (unlikely(!pskb_may_pull(skb, off + len))) 3187 return -ENOMEM; 3188 3189 skb_postpull_rcsum(skb, skb->data + off, len); 3190 memmove(skb->data + len, skb->data, off); 3191 __skb_pull(skb, len); 3192 3193 return 0; 3194 } 3195 3196 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 3197 { 3198 bool trans_same = skb->transport_header == skb->network_header; 3199 int ret; 3200 3201 /* There's no need for __skb_push()/__skb_pull() pair to 3202 * get to the start of the mac header as we're guaranteed 3203 * to always start from here under eBPF. 3204 */ 3205 ret = bpf_skb_generic_push(skb, off, len); 3206 if (likely(!ret)) { 3207 skb->mac_header -= len; 3208 skb->network_header -= len; 3209 if (trans_same) 3210 skb->transport_header = skb->network_header; 3211 } 3212 3213 return ret; 3214 } 3215 3216 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 3217 { 3218 bool trans_same = skb->transport_header == skb->network_header; 3219 int ret; 3220 3221 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 3222 ret = bpf_skb_generic_pop(skb, off, len); 3223 if (likely(!ret)) { 3224 skb->mac_header += len; 3225 skb->network_header += len; 3226 if (trans_same) 3227 skb->transport_header = skb->network_header; 3228 } 3229 3230 return ret; 3231 } 3232 3233 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 3234 { 3235 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3236 u32 off = skb_mac_header_len(skb); 3237 int ret; 3238 3239 ret = skb_cow(skb, len_diff); 3240 if (unlikely(ret < 0)) 3241 return ret; 3242 3243 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3244 if (unlikely(ret < 0)) 3245 return ret; 3246 3247 if (skb_is_gso(skb)) { 3248 struct skb_shared_info *shinfo = skb_shinfo(skb); 3249 3250 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */ 3251 if (shinfo->gso_type & SKB_GSO_TCPV4) { 3252 shinfo->gso_type &= ~SKB_GSO_TCPV4; 3253 shinfo->gso_type |= SKB_GSO_TCPV6; 3254 } 3255 } 3256 3257 skb->protocol = htons(ETH_P_IPV6); 3258 skb_clear_hash(skb); 3259 3260 return 0; 3261 } 3262 3263 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 3264 { 3265 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3266 u32 off = skb_mac_header_len(skb); 3267 int ret; 3268 3269 ret = skb_unclone(skb, GFP_ATOMIC); 3270 if (unlikely(ret < 0)) 3271 return ret; 3272 3273 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3274 if (unlikely(ret < 0)) 3275 return ret; 3276 3277 if (skb_is_gso(skb)) { 3278 struct skb_shared_info *shinfo = skb_shinfo(skb); 3279 3280 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */ 3281 if (shinfo->gso_type & SKB_GSO_TCPV6) { 3282 shinfo->gso_type &= ~SKB_GSO_TCPV6; 3283 shinfo->gso_type |= SKB_GSO_TCPV4; 3284 } 3285 } 3286 3287 skb->protocol = htons(ETH_P_IP); 3288 skb_clear_hash(skb); 3289 3290 return 0; 3291 } 3292 3293 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 3294 { 3295 __be16 from_proto = skb->protocol; 3296 3297 if (from_proto == htons(ETH_P_IP) && 3298 to_proto == htons(ETH_P_IPV6)) 3299 return bpf_skb_proto_4_to_6(skb); 3300 3301 if (from_proto == htons(ETH_P_IPV6) && 3302 to_proto == htons(ETH_P_IP)) 3303 return bpf_skb_proto_6_to_4(skb); 3304 3305 return -ENOTSUPP; 3306 } 3307 3308 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 3309 u64, flags) 3310 { 3311 int ret; 3312 3313 if (unlikely(flags)) 3314 return -EINVAL; 3315 3316 /* General idea is that this helper does the basic groundwork 3317 * needed for changing the protocol, and eBPF program fills the 3318 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 3319 * and other helpers, rather than passing a raw buffer here. 3320 * 3321 * The rationale is to keep this minimal and without a need to 3322 * deal with raw packet data. F.e. even if we would pass buffers 3323 * here, the program still needs to call the bpf_lX_csum_replace() 3324 * helpers anyway. Plus, this way we keep also separation of 3325 * concerns, since f.e. bpf_skb_store_bytes() should only take 3326 * care of stores. 3327 * 3328 * Currently, additional options and extension header space are 3329 * not supported, but flags register is reserved so we can adapt 3330 * that. For offloads, we mark packet as dodgy, so that headers 3331 * need to be verified first. 3332 */ 3333 ret = bpf_skb_proto_xlat(skb, proto); 3334 bpf_compute_data_pointers(skb); 3335 return ret; 3336 } 3337 3338 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 3339 .func = bpf_skb_change_proto, 3340 .gpl_only = false, 3341 .ret_type = RET_INTEGER, 3342 .arg1_type = ARG_PTR_TO_CTX, 3343 .arg2_type = ARG_ANYTHING, 3344 .arg3_type = ARG_ANYTHING, 3345 }; 3346 3347 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 3348 { 3349 /* We only allow a restricted subset to be changed for now. */ 3350 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 3351 !skb_pkt_type_ok(pkt_type))) 3352 return -EINVAL; 3353 3354 skb->pkt_type = pkt_type; 3355 return 0; 3356 } 3357 3358 static const struct bpf_func_proto bpf_skb_change_type_proto = { 3359 .func = bpf_skb_change_type, 3360 .gpl_only = false, 3361 .ret_type = RET_INTEGER, 3362 .arg1_type = ARG_PTR_TO_CTX, 3363 .arg2_type = ARG_ANYTHING, 3364 }; 3365 3366 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3367 { 3368 switch (skb->protocol) { 3369 case htons(ETH_P_IP): 3370 return sizeof(struct iphdr); 3371 case htons(ETH_P_IPV6): 3372 return sizeof(struct ipv6hdr); 3373 default: 3374 return ~0U; 3375 } 3376 } 3377 3378 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3379 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3380 3381 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3382 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3383 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3384 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3385 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \ 3386 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3387 BPF_ADJ_ROOM_ENCAP_L2_MASK)) 3388 3389 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3390 u64 flags) 3391 { 3392 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3393 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3394 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3395 unsigned int gso_type = SKB_GSO_DODGY; 3396 int ret; 3397 3398 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3399 /* udp gso_size delineates datagrams, only allow if fixed */ 3400 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3401 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3402 return -ENOTSUPP; 3403 } 3404 3405 ret = skb_cow_head(skb, len_diff); 3406 if (unlikely(ret < 0)) 3407 return ret; 3408 3409 if (encap) { 3410 if (skb->protocol != htons(ETH_P_IP) && 3411 skb->protocol != htons(ETH_P_IPV6)) 3412 return -ENOTSUPP; 3413 3414 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3415 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3416 return -EINVAL; 3417 3418 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3419 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3420 return -EINVAL; 3421 3422 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH && 3423 inner_mac_len < ETH_HLEN) 3424 return -EINVAL; 3425 3426 if (skb->encapsulation) 3427 return -EALREADY; 3428 3429 mac_len = skb->network_header - skb->mac_header; 3430 inner_net = skb->network_header; 3431 if (inner_mac_len > len_diff) 3432 return -EINVAL; 3433 inner_trans = skb->transport_header; 3434 } 3435 3436 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3437 if (unlikely(ret < 0)) 3438 return ret; 3439 3440 if (encap) { 3441 skb->inner_mac_header = inner_net - inner_mac_len; 3442 skb->inner_network_header = inner_net; 3443 skb->inner_transport_header = inner_trans; 3444 3445 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH) 3446 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 3447 else 3448 skb_set_inner_protocol(skb, skb->protocol); 3449 3450 skb->encapsulation = 1; 3451 skb_set_network_header(skb, mac_len); 3452 3453 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3454 gso_type |= SKB_GSO_UDP_TUNNEL; 3455 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3456 gso_type |= SKB_GSO_GRE; 3457 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3458 gso_type |= SKB_GSO_IPXIP6; 3459 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3460 gso_type |= SKB_GSO_IPXIP4; 3461 3462 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3463 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3464 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3465 sizeof(struct ipv6hdr) : 3466 sizeof(struct iphdr); 3467 3468 skb_set_transport_header(skb, mac_len + nh_len); 3469 } 3470 3471 /* Match skb->protocol to new outer l3 protocol */ 3472 if (skb->protocol == htons(ETH_P_IP) && 3473 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3474 skb->protocol = htons(ETH_P_IPV6); 3475 else if (skb->protocol == htons(ETH_P_IPV6) && 3476 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3477 skb->protocol = htons(ETH_P_IP); 3478 } 3479 3480 if (skb_is_gso(skb)) { 3481 struct skb_shared_info *shinfo = skb_shinfo(skb); 3482 3483 /* Due to header grow, MSS needs to be downgraded. */ 3484 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3485 skb_decrease_gso_size(shinfo, len_diff); 3486 3487 /* Header must be checked, and gso_segs recomputed. */ 3488 shinfo->gso_type |= gso_type; 3489 shinfo->gso_segs = 0; 3490 } 3491 3492 return 0; 3493 } 3494 3495 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3496 u64 flags) 3497 { 3498 int ret; 3499 3500 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO | 3501 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3502 return -EINVAL; 3503 3504 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3505 /* udp gso_size delineates datagrams, only allow if fixed */ 3506 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3507 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3508 return -ENOTSUPP; 3509 } 3510 3511 ret = skb_unclone(skb, GFP_ATOMIC); 3512 if (unlikely(ret < 0)) 3513 return ret; 3514 3515 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3516 if (unlikely(ret < 0)) 3517 return ret; 3518 3519 if (skb_is_gso(skb)) { 3520 struct skb_shared_info *shinfo = skb_shinfo(skb); 3521 3522 /* Due to header shrink, MSS can be upgraded. */ 3523 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3524 skb_increase_gso_size(shinfo, len_diff); 3525 3526 /* Header must be checked, and gso_segs recomputed. */ 3527 shinfo->gso_type |= SKB_GSO_DODGY; 3528 shinfo->gso_segs = 0; 3529 } 3530 3531 return 0; 3532 } 3533 3534 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC 3535 3536 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3537 u32, mode, u64, flags) 3538 { 3539 u32 len_diff_abs = abs(len_diff); 3540 bool shrink = len_diff < 0; 3541 int ret = 0; 3542 3543 if (unlikely(flags || mode)) 3544 return -EINVAL; 3545 if (unlikely(len_diff_abs > 0xfffU)) 3546 return -EFAULT; 3547 3548 if (!shrink) { 3549 ret = skb_cow(skb, len_diff); 3550 if (unlikely(ret < 0)) 3551 return ret; 3552 __skb_push(skb, len_diff_abs); 3553 memset(skb->data, 0, len_diff_abs); 3554 } else { 3555 if (unlikely(!pskb_may_pull(skb, len_diff_abs))) 3556 return -ENOMEM; 3557 __skb_pull(skb, len_diff_abs); 3558 } 3559 if (tls_sw_has_ctx_rx(skb->sk)) { 3560 struct strp_msg *rxm = strp_msg(skb); 3561 3562 rxm->full_len += len_diff; 3563 } 3564 return ret; 3565 } 3566 3567 static const struct bpf_func_proto sk_skb_adjust_room_proto = { 3568 .func = sk_skb_adjust_room, 3569 .gpl_only = false, 3570 .ret_type = RET_INTEGER, 3571 .arg1_type = ARG_PTR_TO_CTX, 3572 .arg2_type = ARG_ANYTHING, 3573 .arg3_type = ARG_ANYTHING, 3574 .arg4_type = ARG_ANYTHING, 3575 }; 3576 3577 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3578 u32, mode, u64, flags) 3579 { 3580 u32 len_cur, len_diff_abs = abs(len_diff); 3581 u32 len_min = bpf_skb_net_base_len(skb); 3582 u32 len_max = BPF_SKB_MAX_LEN; 3583 __be16 proto = skb->protocol; 3584 bool shrink = len_diff < 0; 3585 u32 off; 3586 int ret; 3587 3588 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK | 3589 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3590 return -EINVAL; 3591 if (unlikely(len_diff_abs > 0xfffU)) 3592 return -EFAULT; 3593 if (unlikely(proto != htons(ETH_P_IP) && 3594 proto != htons(ETH_P_IPV6))) 3595 return -ENOTSUPP; 3596 3597 off = skb_mac_header_len(skb); 3598 switch (mode) { 3599 case BPF_ADJ_ROOM_NET: 3600 off += bpf_skb_net_base_len(skb); 3601 break; 3602 case BPF_ADJ_ROOM_MAC: 3603 break; 3604 default: 3605 return -ENOTSUPP; 3606 } 3607 3608 len_cur = skb->len - skb_network_offset(skb); 3609 if ((shrink && (len_diff_abs >= len_cur || 3610 len_cur - len_diff_abs < len_min)) || 3611 (!shrink && (skb->len + len_diff_abs > len_max && 3612 !skb_is_gso(skb)))) 3613 return -ENOTSUPP; 3614 3615 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3616 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3617 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET)) 3618 __skb_reset_checksum_unnecessary(skb); 3619 3620 bpf_compute_data_pointers(skb); 3621 return ret; 3622 } 3623 3624 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3625 .func = bpf_skb_adjust_room, 3626 .gpl_only = false, 3627 .ret_type = RET_INTEGER, 3628 .arg1_type = ARG_PTR_TO_CTX, 3629 .arg2_type = ARG_ANYTHING, 3630 .arg3_type = ARG_ANYTHING, 3631 .arg4_type = ARG_ANYTHING, 3632 }; 3633 3634 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3635 { 3636 u32 min_len = skb_network_offset(skb); 3637 3638 if (skb_transport_header_was_set(skb)) 3639 min_len = skb_transport_offset(skb); 3640 if (skb->ip_summed == CHECKSUM_PARTIAL) 3641 min_len = skb_checksum_start_offset(skb) + 3642 skb->csum_offset + sizeof(__sum16); 3643 return min_len; 3644 } 3645 3646 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3647 { 3648 unsigned int old_len = skb->len; 3649 int ret; 3650 3651 ret = __skb_grow_rcsum(skb, new_len); 3652 if (!ret) 3653 memset(skb->data + old_len, 0, new_len - old_len); 3654 return ret; 3655 } 3656 3657 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3658 { 3659 return __skb_trim_rcsum(skb, new_len); 3660 } 3661 3662 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3663 u64 flags) 3664 { 3665 u32 max_len = BPF_SKB_MAX_LEN; 3666 u32 min_len = __bpf_skb_min_len(skb); 3667 int ret; 3668 3669 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3670 return -EINVAL; 3671 if (skb->encapsulation) 3672 return -ENOTSUPP; 3673 3674 /* The basic idea of this helper is that it's performing the 3675 * needed work to either grow or trim an skb, and eBPF program 3676 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3677 * bpf_lX_csum_replace() and others rather than passing a raw 3678 * buffer here. This one is a slow path helper and intended 3679 * for replies with control messages. 3680 * 3681 * Like in bpf_skb_change_proto(), we want to keep this rather 3682 * minimal and without protocol specifics so that we are able 3683 * to separate concerns as in bpf_skb_store_bytes() should only 3684 * be the one responsible for writing buffers. 3685 * 3686 * It's really expected to be a slow path operation here for 3687 * control message replies, so we're implicitly linearizing, 3688 * uncloning and drop offloads from the skb by this. 3689 */ 3690 ret = __bpf_try_make_writable(skb, skb->len); 3691 if (!ret) { 3692 if (new_len > skb->len) 3693 ret = bpf_skb_grow_rcsum(skb, new_len); 3694 else if (new_len < skb->len) 3695 ret = bpf_skb_trim_rcsum(skb, new_len); 3696 if (!ret && skb_is_gso(skb)) 3697 skb_gso_reset(skb); 3698 } 3699 return ret; 3700 } 3701 3702 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3703 u64, flags) 3704 { 3705 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3706 3707 bpf_compute_data_pointers(skb); 3708 return ret; 3709 } 3710 3711 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3712 .func = bpf_skb_change_tail, 3713 .gpl_only = false, 3714 .ret_type = RET_INTEGER, 3715 .arg1_type = ARG_PTR_TO_CTX, 3716 .arg2_type = ARG_ANYTHING, 3717 .arg3_type = ARG_ANYTHING, 3718 }; 3719 3720 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3721 u64, flags) 3722 { 3723 return __bpf_skb_change_tail(skb, new_len, flags); 3724 } 3725 3726 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3727 .func = sk_skb_change_tail, 3728 .gpl_only = false, 3729 .ret_type = RET_INTEGER, 3730 .arg1_type = ARG_PTR_TO_CTX, 3731 .arg2_type = ARG_ANYTHING, 3732 .arg3_type = ARG_ANYTHING, 3733 }; 3734 3735 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3736 u64 flags) 3737 { 3738 u32 max_len = BPF_SKB_MAX_LEN; 3739 u32 new_len = skb->len + head_room; 3740 int ret; 3741 3742 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3743 new_len < skb->len)) 3744 return -EINVAL; 3745 3746 ret = skb_cow(skb, head_room); 3747 if (likely(!ret)) { 3748 /* Idea for this helper is that we currently only 3749 * allow to expand on mac header. This means that 3750 * skb->protocol network header, etc, stay as is. 3751 * Compared to bpf_skb_change_tail(), we're more 3752 * flexible due to not needing to linearize or 3753 * reset GSO. Intention for this helper is to be 3754 * used by an L3 skb that needs to push mac header 3755 * for redirection into L2 device. 3756 */ 3757 __skb_push(skb, head_room); 3758 memset(skb->data, 0, head_room); 3759 skb_reset_mac_header(skb); 3760 skb_reset_mac_len(skb); 3761 } 3762 3763 return ret; 3764 } 3765 3766 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3767 u64, flags) 3768 { 3769 int ret = __bpf_skb_change_head(skb, head_room, flags); 3770 3771 bpf_compute_data_pointers(skb); 3772 return ret; 3773 } 3774 3775 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3776 .func = bpf_skb_change_head, 3777 .gpl_only = false, 3778 .ret_type = RET_INTEGER, 3779 .arg1_type = ARG_PTR_TO_CTX, 3780 .arg2_type = ARG_ANYTHING, 3781 .arg3_type = ARG_ANYTHING, 3782 }; 3783 3784 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3785 u64, flags) 3786 { 3787 return __bpf_skb_change_head(skb, head_room, flags); 3788 } 3789 3790 static const struct bpf_func_proto sk_skb_change_head_proto = { 3791 .func = sk_skb_change_head, 3792 .gpl_only = false, 3793 .ret_type = RET_INTEGER, 3794 .arg1_type = ARG_PTR_TO_CTX, 3795 .arg2_type = ARG_ANYTHING, 3796 .arg3_type = ARG_ANYTHING, 3797 }; 3798 3799 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp) 3800 { 3801 return xdp_get_buff_len(xdp); 3802 } 3803 3804 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = { 3805 .func = bpf_xdp_get_buff_len, 3806 .gpl_only = false, 3807 .ret_type = RET_INTEGER, 3808 .arg1_type = ARG_PTR_TO_CTX, 3809 }; 3810 3811 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff) 3812 3813 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = { 3814 .func = bpf_xdp_get_buff_len, 3815 .gpl_only = false, 3816 .arg1_type = ARG_PTR_TO_BTF_ID, 3817 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0], 3818 }; 3819 3820 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3821 { 3822 return xdp_data_meta_unsupported(xdp) ? 0 : 3823 xdp->data - xdp->data_meta; 3824 } 3825 3826 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3827 { 3828 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3829 unsigned long metalen = xdp_get_metalen(xdp); 3830 void *data_start = xdp_frame_end + metalen; 3831 void *data = xdp->data + offset; 3832 3833 if (unlikely(data < data_start || 3834 data > xdp->data_end - ETH_HLEN)) 3835 return -EINVAL; 3836 3837 if (metalen) 3838 memmove(xdp->data_meta + offset, 3839 xdp->data_meta, metalen); 3840 xdp->data_meta += offset; 3841 xdp->data = data; 3842 3843 return 0; 3844 } 3845 3846 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3847 .func = bpf_xdp_adjust_head, 3848 .gpl_only = false, 3849 .ret_type = RET_INTEGER, 3850 .arg1_type = ARG_PTR_TO_CTX, 3851 .arg2_type = ARG_ANYTHING, 3852 }; 3853 3854 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 3855 void *buf, unsigned long len, bool flush) 3856 { 3857 unsigned long ptr_len, ptr_off = 0; 3858 skb_frag_t *next_frag, *end_frag; 3859 struct skb_shared_info *sinfo; 3860 void *src, *dst; 3861 u8 *ptr_buf; 3862 3863 if (likely(xdp->data_end - xdp->data >= off + len)) { 3864 src = flush ? buf : xdp->data + off; 3865 dst = flush ? xdp->data + off : buf; 3866 memcpy(dst, src, len); 3867 return; 3868 } 3869 3870 sinfo = xdp_get_shared_info_from_buff(xdp); 3871 end_frag = &sinfo->frags[sinfo->nr_frags]; 3872 next_frag = &sinfo->frags[0]; 3873 3874 ptr_len = xdp->data_end - xdp->data; 3875 ptr_buf = xdp->data; 3876 3877 while (true) { 3878 if (off < ptr_off + ptr_len) { 3879 unsigned long copy_off = off - ptr_off; 3880 unsigned long copy_len = min(len, ptr_len - copy_off); 3881 3882 src = flush ? buf : ptr_buf + copy_off; 3883 dst = flush ? ptr_buf + copy_off : buf; 3884 memcpy(dst, src, copy_len); 3885 3886 off += copy_len; 3887 len -= copy_len; 3888 buf += copy_len; 3889 } 3890 3891 if (!len || next_frag == end_frag) 3892 break; 3893 3894 ptr_off += ptr_len; 3895 ptr_buf = skb_frag_address(next_frag); 3896 ptr_len = skb_frag_size(next_frag); 3897 next_frag++; 3898 } 3899 } 3900 3901 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 3902 { 3903 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 3904 u32 size = xdp->data_end - xdp->data; 3905 void *addr = xdp->data; 3906 int i; 3907 3908 if (unlikely(offset > 0xffff || len > 0xffff)) 3909 return ERR_PTR(-EFAULT); 3910 3911 if (offset + len > xdp_get_buff_len(xdp)) 3912 return ERR_PTR(-EINVAL); 3913 3914 if (offset < size) /* linear area */ 3915 goto out; 3916 3917 offset -= size; 3918 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */ 3919 u32 frag_size = skb_frag_size(&sinfo->frags[i]); 3920 3921 if (offset < frag_size) { 3922 addr = skb_frag_address(&sinfo->frags[i]); 3923 size = frag_size; 3924 break; 3925 } 3926 offset -= frag_size; 3927 } 3928 out: 3929 return offset + len <= size ? addr + offset : NULL; 3930 } 3931 3932 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset, 3933 void *, buf, u32, len) 3934 { 3935 void *ptr; 3936 3937 ptr = bpf_xdp_pointer(xdp, offset, len); 3938 if (IS_ERR(ptr)) 3939 return PTR_ERR(ptr); 3940 3941 if (!ptr) 3942 bpf_xdp_copy_buf(xdp, offset, buf, len, false); 3943 else 3944 memcpy(buf, ptr, len); 3945 3946 return 0; 3947 } 3948 3949 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = { 3950 .func = bpf_xdp_load_bytes, 3951 .gpl_only = false, 3952 .ret_type = RET_INTEGER, 3953 .arg1_type = ARG_PTR_TO_CTX, 3954 .arg2_type = ARG_ANYTHING, 3955 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 3956 .arg4_type = ARG_CONST_SIZE, 3957 }; 3958 3959 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset, 3960 void *, buf, u32, len) 3961 { 3962 void *ptr; 3963 3964 ptr = bpf_xdp_pointer(xdp, offset, len); 3965 if (IS_ERR(ptr)) 3966 return PTR_ERR(ptr); 3967 3968 if (!ptr) 3969 bpf_xdp_copy_buf(xdp, offset, buf, len, true); 3970 else 3971 memcpy(ptr, buf, len); 3972 3973 return 0; 3974 } 3975 3976 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = { 3977 .func = bpf_xdp_store_bytes, 3978 .gpl_only = false, 3979 .ret_type = RET_INTEGER, 3980 .arg1_type = ARG_PTR_TO_CTX, 3981 .arg2_type = ARG_ANYTHING, 3982 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 3983 .arg4_type = ARG_CONST_SIZE, 3984 }; 3985 3986 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset) 3987 { 3988 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 3989 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1]; 3990 struct xdp_rxq_info *rxq = xdp->rxq; 3991 unsigned int tailroom; 3992 3993 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz) 3994 return -EOPNOTSUPP; 3995 3996 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag); 3997 if (unlikely(offset > tailroom)) 3998 return -EINVAL; 3999 4000 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset); 4001 skb_frag_size_add(frag, offset); 4002 sinfo->xdp_frags_size += offset; 4003 4004 return 0; 4005 } 4006 4007 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset) 4008 { 4009 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4010 int i, n_frags_free = 0, len_free = 0; 4011 4012 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN)) 4013 return -EINVAL; 4014 4015 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) { 4016 skb_frag_t *frag = &sinfo->frags[i]; 4017 int shrink = min_t(int, offset, skb_frag_size(frag)); 4018 4019 len_free += shrink; 4020 offset -= shrink; 4021 4022 if (skb_frag_size(frag) == shrink) { 4023 struct page *page = skb_frag_page(frag); 4024 4025 __xdp_return(page_address(page), &xdp->rxq->mem, 4026 false, NULL); 4027 n_frags_free++; 4028 } else { 4029 skb_frag_size_sub(frag, shrink); 4030 break; 4031 } 4032 } 4033 sinfo->nr_frags -= n_frags_free; 4034 sinfo->xdp_frags_size -= len_free; 4035 4036 if (unlikely(!sinfo->nr_frags)) { 4037 xdp_buff_clear_frags_flag(xdp); 4038 xdp->data_end -= offset; 4039 } 4040 4041 return 0; 4042 } 4043 4044 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 4045 { 4046 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */ 4047 void *data_end = xdp->data_end + offset; 4048 4049 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */ 4050 if (offset < 0) 4051 return bpf_xdp_frags_shrink_tail(xdp, -offset); 4052 4053 return bpf_xdp_frags_increase_tail(xdp, offset); 4054 } 4055 4056 /* Notice that xdp_data_hard_end have reserved some tailroom */ 4057 if (unlikely(data_end > data_hard_end)) 4058 return -EINVAL; 4059 4060 /* ALL drivers MUST init xdp->frame_sz, chicken check below */ 4061 if (unlikely(xdp->frame_sz > PAGE_SIZE)) { 4062 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz); 4063 return -EINVAL; 4064 } 4065 4066 if (unlikely(data_end < xdp->data + ETH_HLEN)) 4067 return -EINVAL; 4068 4069 /* Clear memory area on grow, can contain uninit kernel memory */ 4070 if (offset > 0) 4071 memset(xdp->data_end, 0, offset); 4072 4073 xdp->data_end = data_end; 4074 4075 return 0; 4076 } 4077 4078 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 4079 .func = bpf_xdp_adjust_tail, 4080 .gpl_only = false, 4081 .ret_type = RET_INTEGER, 4082 .arg1_type = ARG_PTR_TO_CTX, 4083 .arg2_type = ARG_ANYTHING, 4084 }; 4085 4086 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 4087 { 4088 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 4089 void *meta = xdp->data_meta + offset; 4090 unsigned long metalen = xdp->data - meta; 4091 4092 if (xdp_data_meta_unsupported(xdp)) 4093 return -ENOTSUPP; 4094 if (unlikely(meta < xdp_frame_end || 4095 meta > xdp->data)) 4096 return -EINVAL; 4097 if (unlikely(xdp_metalen_invalid(metalen))) 4098 return -EACCES; 4099 4100 xdp->data_meta = meta; 4101 4102 return 0; 4103 } 4104 4105 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 4106 .func = bpf_xdp_adjust_meta, 4107 .gpl_only = false, 4108 .ret_type = RET_INTEGER, 4109 .arg1_type = ARG_PTR_TO_CTX, 4110 .arg2_type = ARG_ANYTHING, 4111 }; 4112 4113 /** 4114 * DOC: xdp redirect 4115 * 4116 * XDP_REDIRECT works by a three-step process, implemented in the functions 4117 * below: 4118 * 4119 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target 4120 * of the redirect and store it (along with some other metadata) in a per-CPU 4121 * struct bpf_redirect_info. 4122 * 4123 * 2. When the program returns the XDP_REDIRECT return code, the driver will 4124 * call xdp_do_redirect() which will use the information in struct 4125 * bpf_redirect_info to actually enqueue the frame into a map type-specific 4126 * bulk queue structure. 4127 * 4128 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(), 4129 * which will flush all the different bulk queues, thus completing the 4130 * redirect. 4131 */ 4132 /* 4133 * Pointers to the map entries will be kept around for this whole sequence of 4134 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in 4135 * the core code; instead, the RCU protection relies on everything happening 4136 * inside a single NAPI poll sequence, which means it's between a pair of calls 4137 * to local_bh_disable()/local_bh_enable(). 4138 * 4139 * The map entries are marked as __rcu and the map code makes sure to 4140 * dereference those pointers with rcu_dereference_check() in a way that works 4141 * for both sections that to hold an rcu_read_lock() and sections that are 4142 * called from NAPI without a separate rcu_read_lock(). The code below does not 4143 * use RCU annotations, but relies on those in the map code. 4144 */ 4145 void xdp_do_flush(void) 4146 { 4147 __dev_flush(); 4148 __cpu_map_flush(); 4149 __xsk_map_flush(); 4150 } 4151 EXPORT_SYMBOL_GPL(xdp_do_flush); 4152 4153 void bpf_clear_redirect_map(struct bpf_map *map) 4154 { 4155 struct bpf_redirect_info *ri; 4156 int cpu; 4157 4158 for_each_possible_cpu(cpu) { 4159 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 4160 /* Avoid polluting remote cacheline due to writes if 4161 * not needed. Once we pass this test, we need the 4162 * cmpxchg() to make sure it hasn't been changed in 4163 * the meantime by remote CPU. 4164 */ 4165 if (unlikely(READ_ONCE(ri->map) == map)) 4166 cmpxchg(&ri->map, map, NULL); 4167 } 4168 } 4169 4170 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 4171 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key); 4172 4173 u32 xdp_master_redirect(struct xdp_buff *xdp) 4174 { 4175 struct net_device *master, *slave; 4176 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4177 4178 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev); 4179 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp); 4180 if (slave && slave != xdp->rxq->dev) { 4181 /* The target device is different from the receiving device, so 4182 * redirect it to the new device. 4183 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled 4184 * drivers to unmap the packet from their rx ring. 4185 */ 4186 ri->tgt_index = slave->ifindex; 4187 ri->map_id = INT_MAX; 4188 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4189 return XDP_REDIRECT; 4190 } 4191 return XDP_TX; 4192 } 4193 EXPORT_SYMBOL_GPL(xdp_master_redirect); 4194 4195 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri, 4196 struct net_device *dev, 4197 struct xdp_buff *xdp, 4198 struct bpf_prog *xdp_prog) 4199 { 4200 enum bpf_map_type map_type = ri->map_type; 4201 void *fwd = ri->tgt_value; 4202 u32 map_id = ri->map_id; 4203 int err; 4204 4205 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4206 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4207 4208 err = __xsk_map_redirect(fwd, xdp); 4209 if (unlikely(err)) 4210 goto err; 4211 4212 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4213 return 0; 4214 err: 4215 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4216 return err; 4217 } 4218 4219 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri, 4220 struct net_device *dev, 4221 struct xdp_frame *xdpf, 4222 struct bpf_prog *xdp_prog) 4223 { 4224 enum bpf_map_type map_type = ri->map_type; 4225 void *fwd = ri->tgt_value; 4226 u32 map_id = ri->map_id; 4227 struct bpf_map *map; 4228 int err; 4229 4230 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4231 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4232 4233 if (unlikely(!xdpf)) { 4234 err = -EOVERFLOW; 4235 goto err; 4236 } 4237 4238 switch (map_type) { 4239 case BPF_MAP_TYPE_DEVMAP: 4240 fallthrough; 4241 case BPF_MAP_TYPE_DEVMAP_HASH: 4242 map = READ_ONCE(ri->map); 4243 if (unlikely(map)) { 4244 WRITE_ONCE(ri->map, NULL); 4245 err = dev_map_enqueue_multi(xdpf, dev, map, 4246 ri->flags & BPF_F_EXCLUDE_INGRESS); 4247 } else { 4248 err = dev_map_enqueue(fwd, xdpf, dev); 4249 } 4250 break; 4251 case BPF_MAP_TYPE_CPUMAP: 4252 err = cpu_map_enqueue(fwd, xdpf, dev); 4253 break; 4254 case BPF_MAP_TYPE_UNSPEC: 4255 if (map_id == INT_MAX) { 4256 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4257 if (unlikely(!fwd)) { 4258 err = -EINVAL; 4259 break; 4260 } 4261 err = dev_xdp_enqueue(fwd, xdpf, dev); 4262 break; 4263 } 4264 fallthrough; 4265 default: 4266 err = -EBADRQC; 4267 } 4268 4269 if (unlikely(err)) 4270 goto err; 4271 4272 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4273 return 0; 4274 err: 4275 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4276 return err; 4277 } 4278 4279 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 4280 struct bpf_prog *xdp_prog) 4281 { 4282 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4283 enum bpf_map_type map_type = ri->map_type; 4284 4285 /* XDP_REDIRECT is not fully supported yet for xdp frags since 4286 * not all XDP capable drivers can map non-linear xdp_frame in 4287 * ndo_xdp_xmit. 4288 */ 4289 if (unlikely(xdp_buff_has_frags(xdp) && 4290 map_type != BPF_MAP_TYPE_CPUMAP)) 4291 return -EOPNOTSUPP; 4292 4293 if (map_type == BPF_MAP_TYPE_XSKMAP) 4294 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4295 4296 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp), 4297 xdp_prog); 4298 } 4299 EXPORT_SYMBOL_GPL(xdp_do_redirect); 4300 4301 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp, 4302 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog) 4303 { 4304 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4305 enum bpf_map_type map_type = ri->map_type; 4306 4307 if (map_type == BPF_MAP_TYPE_XSKMAP) 4308 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4309 4310 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog); 4311 } 4312 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame); 4313 4314 static int xdp_do_generic_redirect_map(struct net_device *dev, 4315 struct sk_buff *skb, 4316 struct xdp_buff *xdp, 4317 struct bpf_prog *xdp_prog, 4318 void *fwd, 4319 enum bpf_map_type map_type, u32 map_id) 4320 { 4321 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4322 struct bpf_map *map; 4323 int err; 4324 4325 switch (map_type) { 4326 case BPF_MAP_TYPE_DEVMAP: 4327 fallthrough; 4328 case BPF_MAP_TYPE_DEVMAP_HASH: 4329 map = READ_ONCE(ri->map); 4330 if (unlikely(map)) { 4331 WRITE_ONCE(ri->map, NULL); 4332 err = dev_map_redirect_multi(dev, skb, xdp_prog, map, 4333 ri->flags & BPF_F_EXCLUDE_INGRESS); 4334 } else { 4335 err = dev_map_generic_redirect(fwd, skb, xdp_prog); 4336 } 4337 if (unlikely(err)) 4338 goto err; 4339 break; 4340 case BPF_MAP_TYPE_XSKMAP: 4341 err = xsk_generic_rcv(fwd, xdp); 4342 if (err) 4343 goto err; 4344 consume_skb(skb); 4345 break; 4346 case BPF_MAP_TYPE_CPUMAP: 4347 err = cpu_map_generic_redirect(fwd, skb); 4348 if (unlikely(err)) 4349 goto err; 4350 break; 4351 default: 4352 err = -EBADRQC; 4353 goto err; 4354 } 4355 4356 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4357 return 0; 4358 err: 4359 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4360 return err; 4361 } 4362 4363 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 4364 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 4365 { 4366 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4367 enum bpf_map_type map_type = ri->map_type; 4368 void *fwd = ri->tgt_value; 4369 u32 map_id = ri->map_id; 4370 int err; 4371 4372 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4373 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4374 4375 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) { 4376 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4377 if (unlikely(!fwd)) { 4378 err = -EINVAL; 4379 goto err; 4380 } 4381 4382 err = xdp_ok_fwd_dev(fwd, skb->len); 4383 if (unlikely(err)) 4384 goto err; 4385 4386 skb->dev = fwd; 4387 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index); 4388 generic_xdp_tx(skb, xdp_prog); 4389 return 0; 4390 } 4391 4392 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id); 4393 err: 4394 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err); 4395 return err; 4396 } 4397 4398 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 4399 { 4400 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4401 4402 if (unlikely(flags)) 4403 return XDP_ABORTED; 4404 4405 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated 4406 * by map_idr) is used for ifindex based XDP redirect. 4407 */ 4408 ri->tgt_index = ifindex; 4409 ri->map_id = INT_MAX; 4410 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4411 4412 return XDP_REDIRECT; 4413 } 4414 4415 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 4416 .func = bpf_xdp_redirect, 4417 .gpl_only = false, 4418 .ret_type = RET_INTEGER, 4419 .arg1_type = ARG_ANYTHING, 4420 .arg2_type = ARG_ANYTHING, 4421 }; 4422 4423 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key, 4424 u64, flags) 4425 { 4426 return map->ops->map_redirect(map, key, flags); 4427 } 4428 4429 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 4430 .func = bpf_xdp_redirect_map, 4431 .gpl_only = false, 4432 .ret_type = RET_INTEGER, 4433 .arg1_type = ARG_CONST_MAP_PTR, 4434 .arg2_type = ARG_ANYTHING, 4435 .arg3_type = ARG_ANYTHING, 4436 }; 4437 4438 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 4439 unsigned long off, unsigned long len) 4440 { 4441 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 4442 4443 if (unlikely(!ptr)) 4444 return len; 4445 if (ptr != dst_buff) 4446 memcpy(dst_buff, ptr, len); 4447 4448 return 0; 4449 } 4450 4451 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 4452 u64, flags, void *, meta, u64, meta_size) 4453 { 4454 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4455 4456 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4457 return -EINVAL; 4458 if (unlikely(!skb || skb_size > skb->len)) 4459 return -EFAULT; 4460 4461 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 4462 bpf_skb_copy); 4463 } 4464 4465 static const struct bpf_func_proto bpf_skb_event_output_proto = { 4466 .func = bpf_skb_event_output, 4467 .gpl_only = true, 4468 .ret_type = RET_INTEGER, 4469 .arg1_type = ARG_PTR_TO_CTX, 4470 .arg2_type = ARG_CONST_MAP_PTR, 4471 .arg3_type = ARG_ANYTHING, 4472 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4473 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4474 }; 4475 4476 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff) 4477 4478 const struct bpf_func_proto bpf_skb_output_proto = { 4479 .func = bpf_skb_event_output, 4480 .gpl_only = true, 4481 .ret_type = RET_INTEGER, 4482 .arg1_type = ARG_PTR_TO_BTF_ID, 4483 .arg1_btf_id = &bpf_skb_output_btf_ids[0], 4484 .arg2_type = ARG_CONST_MAP_PTR, 4485 .arg3_type = ARG_ANYTHING, 4486 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4487 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4488 }; 4489 4490 static unsigned short bpf_tunnel_key_af(u64 flags) 4491 { 4492 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 4493 } 4494 4495 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 4496 u32, size, u64, flags) 4497 { 4498 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4499 u8 compat[sizeof(struct bpf_tunnel_key)]; 4500 void *to_orig = to; 4501 int err; 4502 4503 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 | 4504 BPF_F_TUNINFO_FLAGS)))) { 4505 err = -EINVAL; 4506 goto err_clear; 4507 } 4508 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4509 err = -EPROTO; 4510 goto err_clear; 4511 } 4512 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4513 err = -EINVAL; 4514 switch (size) { 4515 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4516 case offsetof(struct bpf_tunnel_key, tunnel_label): 4517 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4518 goto set_compat; 4519 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4520 /* Fixup deprecated structure layouts here, so we have 4521 * a common path later on. 4522 */ 4523 if (ip_tunnel_info_af(info) != AF_INET) 4524 goto err_clear; 4525 set_compat: 4526 to = (struct bpf_tunnel_key *)compat; 4527 break; 4528 default: 4529 goto err_clear; 4530 } 4531 } 4532 4533 to->tunnel_id = be64_to_cpu(info->key.tun_id); 4534 to->tunnel_tos = info->key.tos; 4535 to->tunnel_ttl = info->key.ttl; 4536 if (flags & BPF_F_TUNINFO_FLAGS) 4537 to->tunnel_flags = info->key.tun_flags; 4538 else 4539 to->tunnel_ext = 0; 4540 4541 if (flags & BPF_F_TUNINFO_IPV6) { 4542 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 4543 sizeof(to->remote_ipv6)); 4544 memcpy(to->local_ipv6, &info->key.u.ipv6.dst, 4545 sizeof(to->local_ipv6)); 4546 to->tunnel_label = be32_to_cpu(info->key.label); 4547 } else { 4548 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 4549 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4550 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst); 4551 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3); 4552 to->tunnel_label = 0; 4553 } 4554 4555 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 4556 memcpy(to_orig, to, size); 4557 4558 return 0; 4559 err_clear: 4560 memset(to_orig, 0, size); 4561 return err; 4562 } 4563 4564 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 4565 .func = bpf_skb_get_tunnel_key, 4566 .gpl_only = false, 4567 .ret_type = RET_INTEGER, 4568 .arg1_type = ARG_PTR_TO_CTX, 4569 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4570 .arg3_type = ARG_CONST_SIZE, 4571 .arg4_type = ARG_ANYTHING, 4572 }; 4573 4574 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 4575 { 4576 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4577 int err; 4578 4579 if (unlikely(!info || 4580 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 4581 err = -ENOENT; 4582 goto err_clear; 4583 } 4584 if (unlikely(size < info->options_len)) { 4585 err = -ENOMEM; 4586 goto err_clear; 4587 } 4588 4589 ip_tunnel_info_opts_get(to, info); 4590 if (size > info->options_len) 4591 memset(to + info->options_len, 0, size - info->options_len); 4592 4593 return info->options_len; 4594 err_clear: 4595 memset(to, 0, size); 4596 return err; 4597 } 4598 4599 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 4600 .func = bpf_skb_get_tunnel_opt, 4601 .gpl_only = false, 4602 .ret_type = RET_INTEGER, 4603 .arg1_type = ARG_PTR_TO_CTX, 4604 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4605 .arg3_type = ARG_CONST_SIZE, 4606 }; 4607 4608 static struct metadata_dst __percpu *md_dst; 4609 4610 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 4611 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 4612 { 4613 struct metadata_dst *md = this_cpu_ptr(md_dst); 4614 u8 compat[sizeof(struct bpf_tunnel_key)]; 4615 struct ip_tunnel_info *info; 4616 4617 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 4618 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER))) 4619 return -EINVAL; 4620 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4621 switch (size) { 4622 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4623 case offsetof(struct bpf_tunnel_key, tunnel_label): 4624 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4625 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4626 /* Fixup deprecated structure layouts here, so we have 4627 * a common path later on. 4628 */ 4629 memcpy(compat, from, size); 4630 memset(compat + size, 0, sizeof(compat) - size); 4631 from = (const struct bpf_tunnel_key *) compat; 4632 break; 4633 default: 4634 return -EINVAL; 4635 } 4636 } 4637 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 4638 from->tunnel_ext)) 4639 return -EINVAL; 4640 4641 skb_dst_drop(skb); 4642 dst_hold((struct dst_entry *) md); 4643 skb_dst_set(skb, (struct dst_entry *) md); 4644 4645 info = &md->u.tun_info; 4646 memset(info, 0, sizeof(*info)); 4647 info->mode = IP_TUNNEL_INFO_TX; 4648 4649 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 4650 if (flags & BPF_F_DONT_FRAGMENT) 4651 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 4652 if (flags & BPF_F_ZERO_CSUM_TX) 4653 info->key.tun_flags &= ~TUNNEL_CSUM; 4654 if (flags & BPF_F_SEQ_NUMBER) 4655 info->key.tun_flags |= TUNNEL_SEQ; 4656 4657 info->key.tun_id = cpu_to_be64(from->tunnel_id); 4658 info->key.tos = from->tunnel_tos; 4659 info->key.ttl = from->tunnel_ttl; 4660 4661 if (flags & BPF_F_TUNINFO_IPV6) { 4662 info->mode |= IP_TUNNEL_INFO_IPV6; 4663 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 4664 sizeof(from->remote_ipv6)); 4665 memcpy(&info->key.u.ipv6.src, from->local_ipv6, 4666 sizeof(from->local_ipv6)); 4667 info->key.label = cpu_to_be32(from->tunnel_label) & 4668 IPV6_FLOWLABEL_MASK; 4669 } else { 4670 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 4671 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4); 4672 info->key.flow_flags = FLOWI_FLAG_ANYSRC; 4673 } 4674 4675 return 0; 4676 } 4677 4678 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 4679 .func = bpf_skb_set_tunnel_key, 4680 .gpl_only = false, 4681 .ret_type = RET_INTEGER, 4682 .arg1_type = ARG_PTR_TO_CTX, 4683 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4684 .arg3_type = ARG_CONST_SIZE, 4685 .arg4_type = ARG_ANYTHING, 4686 }; 4687 4688 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4689 const u8 *, from, u32, size) 4690 { 4691 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4692 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4693 4694 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4695 return -EINVAL; 4696 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4697 return -ENOMEM; 4698 4699 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4700 4701 return 0; 4702 } 4703 4704 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4705 .func = bpf_skb_set_tunnel_opt, 4706 .gpl_only = false, 4707 .ret_type = RET_INTEGER, 4708 .arg1_type = ARG_PTR_TO_CTX, 4709 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4710 .arg3_type = ARG_CONST_SIZE, 4711 }; 4712 4713 static const struct bpf_func_proto * 4714 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4715 { 4716 if (!md_dst) { 4717 struct metadata_dst __percpu *tmp; 4718 4719 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4720 METADATA_IP_TUNNEL, 4721 GFP_KERNEL); 4722 if (!tmp) 4723 return NULL; 4724 if (cmpxchg(&md_dst, NULL, tmp)) 4725 metadata_dst_free_percpu(tmp); 4726 } 4727 4728 switch (which) { 4729 case BPF_FUNC_skb_set_tunnel_key: 4730 return &bpf_skb_set_tunnel_key_proto; 4731 case BPF_FUNC_skb_set_tunnel_opt: 4732 return &bpf_skb_set_tunnel_opt_proto; 4733 default: 4734 return NULL; 4735 } 4736 } 4737 4738 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4739 u32, idx) 4740 { 4741 struct bpf_array *array = container_of(map, struct bpf_array, map); 4742 struct cgroup *cgrp; 4743 struct sock *sk; 4744 4745 sk = skb_to_full_sk(skb); 4746 if (!sk || !sk_fullsock(sk)) 4747 return -ENOENT; 4748 if (unlikely(idx >= array->map.max_entries)) 4749 return -E2BIG; 4750 4751 cgrp = READ_ONCE(array->ptrs[idx]); 4752 if (unlikely(!cgrp)) 4753 return -EAGAIN; 4754 4755 return sk_under_cgroup_hierarchy(sk, cgrp); 4756 } 4757 4758 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4759 .func = bpf_skb_under_cgroup, 4760 .gpl_only = false, 4761 .ret_type = RET_INTEGER, 4762 .arg1_type = ARG_PTR_TO_CTX, 4763 .arg2_type = ARG_CONST_MAP_PTR, 4764 .arg3_type = ARG_ANYTHING, 4765 }; 4766 4767 #ifdef CONFIG_SOCK_CGROUP_DATA 4768 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 4769 { 4770 struct cgroup *cgrp; 4771 4772 sk = sk_to_full_sk(sk); 4773 if (!sk || !sk_fullsock(sk)) 4774 return 0; 4775 4776 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4777 return cgroup_id(cgrp); 4778 } 4779 4780 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4781 { 4782 return __bpf_sk_cgroup_id(skb->sk); 4783 } 4784 4785 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4786 .func = bpf_skb_cgroup_id, 4787 .gpl_only = false, 4788 .ret_type = RET_INTEGER, 4789 .arg1_type = ARG_PTR_TO_CTX, 4790 }; 4791 4792 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 4793 int ancestor_level) 4794 { 4795 struct cgroup *ancestor; 4796 struct cgroup *cgrp; 4797 4798 sk = sk_to_full_sk(sk); 4799 if (!sk || !sk_fullsock(sk)) 4800 return 0; 4801 4802 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4803 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4804 if (!ancestor) 4805 return 0; 4806 4807 return cgroup_id(ancestor); 4808 } 4809 4810 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4811 ancestor_level) 4812 { 4813 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 4814 } 4815 4816 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4817 .func = bpf_skb_ancestor_cgroup_id, 4818 .gpl_only = false, 4819 .ret_type = RET_INTEGER, 4820 .arg1_type = ARG_PTR_TO_CTX, 4821 .arg2_type = ARG_ANYTHING, 4822 }; 4823 4824 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 4825 { 4826 return __bpf_sk_cgroup_id(sk); 4827 } 4828 4829 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 4830 .func = bpf_sk_cgroup_id, 4831 .gpl_only = false, 4832 .ret_type = RET_INTEGER, 4833 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4834 }; 4835 4836 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 4837 { 4838 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 4839 } 4840 4841 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 4842 .func = bpf_sk_ancestor_cgroup_id, 4843 .gpl_only = false, 4844 .ret_type = RET_INTEGER, 4845 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4846 .arg2_type = ARG_ANYTHING, 4847 }; 4848 #endif 4849 4850 static unsigned long bpf_xdp_copy(void *dst, const void *ctx, 4851 unsigned long off, unsigned long len) 4852 { 4853 struct xdp_buff *xdp = (struct xdp_buff *)ctx; 4854 4855 bpf_xdp_copy_buf(xdp, off, dst, len, false); 4856 return 0; 4857 } 4858 4859 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4860 u64, flags, void *, meta, u64, meta_size) 4861 { 4862 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4863 4864 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4865 return -EINVAL; 4866 4867 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp))) 4868 return -EFAULT; 4869 4870 return bpf_event_output(map, flags, meta, meta_size, xdp, 4871 xdp_size, bpf_xdp_copy); 4872 } 4873 4874 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4875 .func = bpf_xdp_event_output, 4876 .gpl_only = true, 4877 .ret_type = RET_INTEGER, 4878 .arg1_type = ARG_PTR_TO_CTX, 4879 .arg2_type = ARG_CONST_MAP_PTR, 4880 .arg3_type = ARG_ANYTHING, 4881 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4882 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4883 }; 4884 4885 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 4886 4887 const struct bpf_func_proto bpf_xdp_output_proto = { 4888 .func = bpf_xdp_event_output, 4889 .gpl_only = true, 4890 .ret_type = RET_INTEGER, 4891 .arg1_type = ARG_PTR_TO_BTF_ID, 4892 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 4893 .arg2_type = ARG_CONST_MAP_PTR, 4894 .arg3_type = ARG_ANYTHING, 4895 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4896 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4897 }; 4898 4899 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4900 { 4901 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 4902 } 4903 4904 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4905 .func = bpf_get_socket_cookie, 4906 .gpl_only = false, 4907 .ret_type = RET_INTEGER, 4908 .arg1_type = ARG_PTR_TO_CTX, 4909 }; 4910 4911 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4912 { 4913 return __sock_gen_cookie(ctx->sk); 4914 } 4915 4916 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4917 .func = bpf_get_socket_cookie_sock_addr, 4918 .gpl_only = false, 4919 .ret_type = RET_INTEGER, 4920 .arg1_type = ARG_PTR_TO_CTX, 4921 }; 4922 4923 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4924 { 4925 return __sock_gen_cookie(ctx); 4926 } 4927 4928 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4929 .func = bpf_get_socket_cookie_sock, 4930 .gpl_only = false, 4931 .ret_type = RET_INTEGER, 4932 .arg1_type = ARG_PTR_TO_CTX, 4933 }; 4934 4935 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk) 4936 { 4937 return sk ? sock_gen_cookie(sk) : 0; 4938 } 4939 4940 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = { 4941 .func = bpf_get_socket_ptr_cookie, 4942 .gpl_only = false, 4943 .ret_type = RET_INTEGER, 4944 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4945 }; 4946 4947 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4948 { 4949 return __sock_gen_cookie(ctx->sk); 4950 } 4951 4952 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4953 .func = bpf_get_socket_cookie_sock_ops, 4954 .gpl_only = false, 4955 .ret_type = RET_INTEGER, 4956 .arg1_type = ARG_PTR_TO_CTX, 4957 }; 4958 4959 static u64 __bpf_get_netns_cookie(struct sock *sk) 4960 { 4961 const struct net *net = sk ? sock_net(sk) : &init_net; 4962 4963 return net->net_cookie; 4964 } 4965 4966 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 4967 { 4968 return __bpf_get_netns_cookie(ctx); 4969 } 4970 4971 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 4972 .func = bpf_get_netns_cookie_sock, 4973 .gpl_only = false, 4974 .ret_type = RET_INTEGER, 4975 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4976 }; 4977 4978 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4979 { 4980 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4981 } 4982 4983 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 4984 .func = bpf_get_netns_cookie_sock_addr, 4985 .gpl_only = false, 4986 .ret_type = RET_INTEGER, 4987 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4988 }; 4989 4990 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4991 { 4992 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4993 } 4994 4995 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = { 4996 .func = bpf_get_netns_cookie_sock_ops, 4997 .gpl_only = false, 4998 .ret_type = RET_INTEGER, 4999 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5000 }; 5001 5002 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx) 5003 { 5004 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5005 } 5006 5007 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = { 5008 .func = bpf_get_netns_cookie_sk_msg, 5009 .gpl_only = false, 5010 .ret_type = RET_INTEGER, 5011 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5012 }; 5013 5014 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 5015 { 5016 struct sock *sk = sk_to_full_sk(skb->sk); 5017 kuid_t kuid; 5018 5019 if (!sk || !sk_fullsock(sk)) 5020 return overflowuid; 5021 kuid = sock_net_uid(sock_net(sk), sk); 5022 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 5023 } 5024 5025 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 5026 .func = bpf_get_socket_uid, 5027 .gpl_only = false, 5028 .ret_type = RET_INTEGER, 5029 .arg1_type = ARG_PTR_TO_CTX, 5030 }; 5031 5032 static int sol_socket_sockopt(struct sock *sk, int optname, 5033 char *optval, int *optlen, 5034 bool getopt) 5035 { 5036 switch (optname) { 5037 case SO_REUSEADDR: 5038 case SO_SNDBUF: 5039 case SO_RCVBUF: 5040 case SO_KEEPALIVE: 5041 case SO_PRIORITY: 5042 case SO_REUSEPORT: 5043 case SO_RCVLOWAT: 5044 case SO_MARK: 5045 case SO_MAX_PACING_RATE: 5046 case SO_BINDTOIFINDEX: 5047 case SO_TXREHASH: 5048 if (*optlen != sizeof(int)) 5049 return -EINVAL; 5050 break; 5051 case SO_BINDTODEVICE: 5052 break; 5053 default: 5054 return -EINVAL; 5055 } 5056 5057 if (getopt) { 5058 if (optname == SO_BINDTODEVICE) 5059 return -EINVAL; 5060 return sk_getsockopt(sk, SOL_SOCKET, optname, 5061 KERNEL_SOCKPTR(optval), 5062 KERNEL_SOCKPTR(optlen)); 5063 } 5064 5065 return sk_setsockopt(sk, SOL_SOCKET, optname, 5066 KERNEL_SOCKPTR(optval), *optlen); 5067 } 5068 5069 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname, 5070 char *optval, int optlen) 5071 { 5072 struct tcp_sock *tp = tcp_sk(sk); 5073 unsigned long timeout; 5074 int val; 5075 5076 if (optlen != sizeof(int)) 5077 return -EINVAL; 5078 5079 val = *(int *)optval; 5080 5081 /* Only some options are supported */ 5082 switch (optname) { 5083 case TCP_BPF_IW: 5084 if (val <= 0 || tp->data_segs_out > tp->syn_data) 5085 return -EINVAL; 5086 tcp_snd_cwnd_set(tp, val); 5087 break; 5088 case TCP_BPF_SNDCWND_CLAMP: 5089 if (val <= 0) 5090 return -EINVAL; 5091 tp->snd_cwnd_clamp = val; 5092 tp->snd_ssthresh = val; 5093 break; 5094 case TCP_BPF_DELACK_MAX: 5095 timeout = usecs_to_jiffies(val); 5096 if (timeout > TCP_DELACK_MAX || 5097 timeout < TCP_TIMEOUT_MIN) 5098 return -EINVAL; 5099 inet_csk(sk)->icsk_delack_max = timeout; 5100 break; 5101 case TCP_BPF_RTO_MIN: 5102 timeout = usecs_to_jiffies(val); 5103 if (timeout > TCP_RTO_MIN || 5104 timeout < TCP_TIMEOUT_MIN) 5105 return -EINVAL; 5106 inet_csk(sk)->icsk_rto_min = timeout; 5107 break; 5108 default: 5109 return -EINVAL; 5110 } 5111 5112 return 0; 5113 } 5114 5115 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval, 5116 int *optlen, bool getopt) 5117 { 5118 struct tcp_sock *tp; 5119 int ret; 5120 5121 if (*optlen < 2) 5122 return -EINVAL; 5123 5124 if (getopt) { 5125 if (!inet_csk(sk)->icsk_ca_ops) 5126 return -EINVAL; 5127 /* BPF expects NULL-terminated tcp-cc string */ 5128 optval[--(*optlen)] = '\0'; 5129 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION, 5130 KERNEL_SOCKPTR(optval), 5131 KERNEL_SOCKPTR(optlen)); 5132 } 5133 5134 /* "cdg" is the only cc that alloc a ptr 5135 * in inet_csk_ca area. The bpf-tcp-cc may 5136 * overwrite this ptr after switching to cdg. 5137 */ 5138 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen)) 5139 return -ENOTSUPP; 5140 5141 /* It stops this looping 5142 * 5143 * .init => bpf_setsockopt(tcp_cc) => .init => 5144 * bpf_setsockopt(tcp_cc)" => .init => .... 5145 * 5146 * The second bpf_setsockopt(tcp_cc) is not allowed 5147 * in order to break the loop when both .init 5148 * are the same bpf prog. 5149 * 5150 * This applies even the second bpf_setsockopt(tcp_cc) 5151 * does not cause a loop. This limits only the first 5152 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to 5153 * pick a fallback cc (eg. peer does not support ECN) 5154 * and the second '.init' cannot fallback to 5155 * another. 5156 */ 5157 tp = tcp_sk(sk); 5158 if (tp->bpf_chg_cc_inprogress) 5159 return -EBUSY; 5160 5161 tp->bpf_chg_cc_inprogress = 1; 5162 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION, 5163 KERNEL_SOCKPTR(optval), *optlen); 5164 tp->bpf_chg_cc_inprogress = 0; 5165 return ret; 5166 } 5167 5168 static int sol_tcp_sockopt(struct sock *sk, int optname, 5169 char *optval, int *optlen, 5170 bool getopt) 5171 { 5172 if (sk->sk_prot->setsockopt != tcp_setsockopt) 5173 return -EINVAL; 5174 5175 switch (optname) { 5176 case TCP_NODELAY: 5177 case TCP_MAXSEG: 5178 case TCP_KEEPIDLE: 5179 case TCP_KEEPINTVL: 5180 case TCP_KEEPCNT: 5181 case TCP_SYNCNT: 5182 case TCP_WINDOW_CLAMP: 5183 case TCP_THIN_LINEAR_TIMEOUTS: 5184 case TCP_USER_TIMEOUT: 5185 case TCP_NOTSENT_LOWAT: 5186 case TCP_SAVE_SYN: 5187 if (*optlen != sizeof(int)) 5188 return -EINVAL; 5189 break; 5190 case TCP_CONGESTION: 5191 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt); 5192 case TCP_SAVED_SYN: 5193 if (*optlen < 1) 5194 return -EINVAL; 5195 break; 5196 default: 5197 if (getopt) 5198 return -EINVAL; 5199 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen); 5200 } 5201 5202 if (getopt) { 5203 if (optname == TCP_SAVED_SYN) { 5204 struct tcp_sock *tp = tcp_sk(sk); 5205 5206 if (!tp->saved_syn || 5207 *optlen > tcp_saved_syn_len(tp->saved_syn)) 5208 return -EINVAL; 5209 memcpy(optval, tp->saved_syn->data, *optlen); 5210 /* It cannot free tp->saved_syn here because it 5211 * does not know if the user space still needs it. 5212 */ 5213 return 0; 5214 } 5215 5216 return do_tcp_getsockopt(sk, SOL_TCP, optname, 5217 KERNEL_SOCKPTR(optval), 5218 KERNEL_SOCKPTR(optlen)); 5219 } 5220 5221 return do_tcp_setsockopt(sk, SOL_TCP, optname, 5222 KERNEL_SOCKPTR(optval), *optlen); 5223 } 5224 5225 static int sol_ip_sockopt(struct sock *sk, int optname, 5226 char *optval, int *optlen, 5227 bool getopt) 5228 { 5229 if (sk->sk_family != AF_INET) 5230 return -EINVAL; 5231 5232 switch (optname) { 5233 case IP_TOS: 5234 if (*optlen != sizeof(int)) 5235 return -EINVAL; 5236 break; 5237 default: 5238 return -EINVAL; 5239 } 5240 5241 if (getopt) 5242 return do_ip_getsockopt(sk, SOL_IP, optname, 5243 KERNEL_SOCKPTR(optval), 5244 KERNEL_SOCKPTR(optlen)); 5245 5246 return do_ip_setsockopt(sk, SOL_IP, optname, 5247 KERNEL_SOCKPTR(optval), *optlen); 5248 } 5249 5250 static int sol_ipv6_sockopt(struct sock *sk, int optname, 5251 char *optval, int *optlen, 5252 bool getopt) 5253 { 5254 if (sk->sk_family != AF_INET6) 5255 return -EINVAL; 5256 5257 switch (optname) { 5258 case IPV6_TCLASS: 5259 case IPV6_AUTOFLOWLABEL: 5260 if (*optlen != sizeof(int)) 5261 return -EINVAL; 5262 break; 5263 default: 5264 return -EINVAL; 5265 } 5266 5267 if (getopt) 5268 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname, 5269 KERNEL_SOCKPTR(optval), 5270 KERNEL_SOCKPTR(optlen)); 5271 5272 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname, 5273 KERNEL_SOCKPTR(optval), *optlen); 5274 } 5275 5276 static int __bpf_setsockopt(struct sock *sk, int level, int optname, 5277 char *optval, int optlen) 5278 { 5279 if (!sk_fullsock(sk)) 5280 return -EINVAL; 5281 5282 if (level == SOL_SOCKET) 5283 return sol_socket_sockopt(sk, optname, optval, &optlen, false); 5284 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5285 return sol_ip_sockopt(sk, optname, optval, &optlen, false); 5286 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5287 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false); 5288 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5289 return sol_tcp_sockopt(sk, optname, optval, &optlen, false); 5290 5291 return -EINVAL; 5292 } 5293 5294 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 5295 char *optval, int optlen) 5296 { 5297 if (sk_fullsock(sk)) 5298 sock_owned_by_me(sk); 5299 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5300 } 5301 5302 static int __bpf_getsockopt(struct sock *sk, int level, int optname, 5303 char *optval, int optlen) 5304 { 5305 int err, saved_optlen = optlen; 5306 5307 if (!sk_fullsock(sk)) { 5308 err = -EINVAL; 5309 goto done; 5310 } 5311 5312 if (level == SOL_SOCKET) 5313 err = sol_socket_sockopt(sk, optname, optval, &optlen, true); 5314 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5315 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true); 5316 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5317 err = sol_ip_sockopt(sk, optname, optval, &optlen, true); 5318 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5319 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true); 5320 else 5321 err = -EINVAL; 5322 5323 done: 5324 if (err) 5325 optlen = 0; 5326 if (optlen < saved_optlen) 5327 memset(optval + optlen, 0, saved_optlen - optlen); 5328 return err; 5329 } 5330 5331 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 5332 char *optval, int optlen) 5333 { 5334 if (sk_fullsock(sk)) 5335 sock_owned_by_me(sk); 5336 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5337 } 5338 5339 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, 5340 int, optname, char *, optval, int, optlen) 5341 { 5342 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5343 } 5344 5345 const struct bpf_func_proto bpf_sk_setsockopt_proto = { 5346 .func = bpf_sk_setsockopt, 5347 .gpl_only = false, 5348 .ret_type = RET_INTEGER, 5349 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5350 .arg2_type = ARG_ANYTHING, 5351 .arg3_type = ARG_ANYTHING, 5352 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5353 .arg5_type = ARG_CONST_SIZE, 5354 }; 5355 5356 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level, 5357 int, optname, char *, optval, int, optlen) 5358 { 5359 return _bpf_getsockopt(sk, level, optname, optval, optlen); 5360 } 5361 5362 const struct bpf_func_proto bpf_sk_getsockopt_proto = { 5363 .func = bpf_sk_getsockopt, 5364 .gpl_only = false, 5365 .ret_type = RET_INTEGER, 5366 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5367 .arg2_type = ARG_ANYTHING, 5368 .arg3_type = ARG_ANYTHING, 5369 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5370 .arg5_type = ARG_CONST_SIZE, 5371 }; 5372 5373 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level, 5374 int, optname, char *, optval, int, optlen) 5375 { 5376 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5377 } 5378 5379 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = { 5380 .func = bpf_unlocked_sk_setsockopt, 5381 .gpl_only = false, 5382 .ret_type = RET_INTEGER, 5383 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5384 .arg2_type = ARG_ANYTHING, 5385 .arg3_type = ARG_ANYTHING, 5386 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5387 .arg5_type = ARG_CONST_SIZE, 5388 }; 5389 5390 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level, 5391 int, optname, char *, optval, int, optlen) 5392 { 5393 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5394 } 5395 5396 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = { 5397 .func = bpf_unlocked_sk_getsockopt, 5398 .gpl_only = false, 5399 .ret_type = RET_INTEGER, 5400 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5401 .arg2_type = ARG_ANYTHING, 5402 .arg3_type = ARG_ANYTHING, 5403 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5404 .arg5_type = ARG_CONST_SIZE, 5405 }; 5406 5407 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 5408 int, level, int, optname, char *, optval, int, optlen) 5409 { 5410 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5411 } 5412 5413 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5414 .func = bpf_sock_addr_setsockopt, 5415 .gpl_only = false, 5416 .ret_type = RET_INTEGER, 5417 .arg1_type = ARG_PTR_TO_CTX, 5418 .arg2_type = ARG_ANYTHING, 5419 .arg3_type = ARG_ANYTHING, 5420 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5421 .arg5_type = ARG_CONST_SIZE, 5422 }; 5423 5424 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5425 int, level, int, optname, char *, optval, int, optlen) 5426 { 5427 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5428 } 5429 5430 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5431 .func = bpf_sock_addr_getsockopt, 5432 .gpl_only = false, 5433 .ret_type = RET_INTEGER, 5434 .arg1_type = ARG_PTR_TO_CTX, 5435 .arg2_type = ARG_ANYTHING, 5436 .arg3_type = ARG_ANYTHING, 5437 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5438 .arg5_type = ARG_CONST_SIZE, 5439 }; 5440 5441 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5442 int, level, int, optname, char *, optval, int, optlen) 5443 { 5444 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5445 } 5446 5447 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5448 .func = bpf_sock_ops_setsockopt, 5449 .gpl_only = false, 5450 .ret_type = RET_INTEGER, 5451 .arg1_type = ARG_PTR_TO_CTX, 5452 .arg2_type = ARG_ANYTHING, 5453 .arg3_type = ARG_ANYTHING, 5454 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5455 .arg5_type = ARG_CONST_SIZE, 5456 }; 5457 5458 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5459 int optname, const u8 **start) 5460 { 5461 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5462 const u8 *hdr_start; 5463 int ret; 5464 5465 if (syn_skb) { 5466 /* sk is a request_sock here */ 5467 5468 if (optname == TCP_BPF_SYN) { 5469 hdr_start = syn_skb->data; 5470 ret = tcp_hdrlen(syn_skb); 5471 } else if (optname == TCP_BPF_SYN_IP) { 5472 hdr_start = skb_network_header(syn_skb); 5473 ret = skb_network_header_len(syn_skb) + 5474 tcp_hdrlen(syn_skb); 5475 } else { 5476 /* optname == TCP_BPF_SYN_MAC */ 5477 hdr_start = skb_mac_header(syn_skb); 5478 ret = skb_mac_header_len(syn_skb) + 5479 skb_network_header_len(syn_skb) + 5480 tcp_hdrlen(syn_skb); 5481 } 5482 } else { 5483 struct sock *sk = bpf_sock->sk; 5484 struct saved_syn *saved_syn; 5485 5486 if (sk->sk_state == TCP_NEW_SYN_RECV) 5487 /* synack retransmit. bpf_sock->syn_skb will 5488 * not be available. It has to resort to 5489 * saved_syn (if it is saved). 5490 */ 5491 saved_syn = inet_reqsk(sk)->saved_syn; 5492 else 5493 saved_syn = tcp_sk(sk)->saved_syn; 5494 5495 if (!saved_syn) 5496 return -ENOENT; 5497 5498 if (optname == TCP_BPF_SYN) { 5499 hdr_start = saved_syn->data + 5500 saved_syn->mac_hdrlen + 5501 saved_syn->network_hdrlen; 5502 ret = saved_syn->tcp_hdrlen; 5503 } else if (optname == TCP_BPF_SYN_IP) { 5504 hdr_start = saved_syn->data + 5505 saved_syn->mac_hdrlen; 5506 ret = saved_syn->network_hdrlen + 5507 saved_syn->tcp_hdrlen; 5508 } else { 5509 /* optname == TCP_BPF_SYN_MAC */ 5510 5511 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5512 if (!saved_syn->mac_hdrlen) 5513 return -ENOENT; 5514 5515 hdr_start = saved_syn->data; 5516 ret = saved_syn->mac_hdrlen + 5517 saved_syn->network_hdrlen + 5518 saved_syn->tcp_hdrlen; 5519 } 5520 } 5521 5522 *start = hdr_start; 5523 return ret; 5524 } 5525 5526 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5527 int, level, int, optname, char *, optval, int, optlen) 5528 { 5529 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5530 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5531 int ret, copy_len = 0; 5532 const u8 *start; 5533 5534 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5535 if (ret > 0) { 5536 copy_len = ret; 5537 if (optlen < copy_len) { 5538 copy_len = optlen; 5539 ret = -ENOSPC; 5540 } 5541 5542 memcpy(optval, start, copy_len); 5543 } 5544 5545 /* Zero out unused buffer at the end */ 5546 memset(optval + copy_len, 0, optlen - copy_len); 5547 5548 return ret; 5549 } 5550 5551 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5552 } 5553 5554 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5555 .func = bpf_sock_ops_getsockopt, 5556 .gpl_only = false, 5557 .ret_type = RET_INTEGER, 5558 .arg1_type = ARG_PTR_TO_CTX, 5559 .arg2_type = ARG_ANYTHING, 5560 .arg3_type = ARG_ANYTHING, 5561 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5562 .arg5_type = ARG_CONST_SIZE, 5563 }; 5564 5565 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5566 int, argval) 5567 { 5568 struct sock *sk = bpf_sock->sk; 5569 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5570 5571 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5572 return -EINVAL; 5573 5574 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5575 5576 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5577 } 5578 5579 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5580 .func = bpf_sock_ops_cb_flags_set, 5581 .gpl_only = false, 5582 .ret_type = RET_INTEGER, 5583 .arg1_type = ARG_PTR_TO_CTX, 5584 .arg2_type = ARG_ANYTHING, 5585 }; 5586 5587 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5588 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5589 5590 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5591 int, addr_len) 5592 { 5593 #ifdef CONFIG_INET 5594 struct sock *sk = ctx->sk; 5595 u32 flags = BIND_FROM_BPF; 5596 int err; 5597 5598 err = -EINVAL; 5599 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5600 return err; 5601 if (addr->sa_family == AF_INET) { 5602 if (addr_len < sizeof(struct sockaddr_in)) 5603 return err; 5604 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5605 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5606 return __inet_bind(sk, addr, addr_len, flags); 5607 #if IS_ENABLED(CONFIG_IPV6) 5608 } else if (addr->sa_family == AF_INET6) { 5609 if (addr_len < SIN6_LEN_RFC2133) 5610 return err; 5611 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5612 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5613 /* ipv6_bpf_stub cannot be NULL, since it's called from 5614 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5615 */ 5616 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5617 #endif /* CONFIG_IPV6 */ 5618 } 5619 #endif /* CONFIG_INET */ 5620 5621 return -EAFNOSUPPORT; 5622 } 5623 5624 static const struct bpf_func_proto bpf_bind_proto = { 5625 .func = bpf_bind, 5626 .gpl_only = false, 5627 .ret_type = RET_INTEGER, 5628 .arg1_type = ARG_PTR_TO_CTX, 5629 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5630 .arg3_type = ARG_CONST_SIZE, 5631 }; 5632 5633 #ifdef CONFIG_XFRM 5634 5635 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \ 5636 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)) 5637 5638 struct metadata_dst __percpu *xfrm_bpf_md_dst; 5639 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst); 5640 5641 #endif 5642 5643 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5644 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5645 { 5646 const struct sec_path *sp = skb_sec_path(skb); 5647 const struct xfrm_state *x; 5648 5649 if (!sp || unlikely(index >= sp->len || flags)) 5650 goto err_clear; 5651 5652 x = sp->xvec[index]; 5653 5654 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5655 goto err_clear; 5656 5657 to->reqid = x->props.reqid; 5658 to->spi = x->id.spi; 5659 to->family = x->props.family; 5660 to->ext = 0; 5661 5662 if (to->family == AF_INET6) { 5663 memcpy(to->remote_ipv6, x->props.saddr.a6, 5664 sizeof(to->remote_ipv6)); 5665 } else { 5666 to->remote_ipv4 = x->props.saddr.a4; 5667 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5668 } 5669 5670 return 0; 5671 err_clear: 5672 memset(to, 0, size); 5673 return -EINVAL; 5674 } 5675 5676 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5677 .func = bpf_skb_get_xfrm_state, 5678 .gpl_only = false, 5679 .ret_type = RET_INTEGER, 5680 .arg1_type = ARG_PTR_TO_CTX, 5681 .arg2_type = ARG_ANYTHING, 5682 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5683 .arg4_type = ARG_CONST_SIZE, 5684 .arg5_type = ARG_ANYTHING, 5685 }; 5686 #endif 5687 5688 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5689 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 5690 const struct neighbour *neigh, 5691 const struct net_device *dev, u32 mtu) 5692 { 5693 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5694 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5695 params->h_vlan_TCI = 0; 5696 params->h_vlan_proto = 0; 5697 if (mtu) 5698 params->mtu_result = mtu; /* union with tot_len */ 5699 5700 return 0; 5701 } 5702 #endif 5703 5704 #if IS_ENABLED(CONFIG_INET) 5705 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5706 u32 flags, bool check_mtu) 5707 { 5708 struct fib_nh_common *nhc; 5709 struct in_device *in_dev; 5710 struct neighbour *neigh; 5711 struct net_device *dev; 5712 struct fib_result res; 5713 struct flowi4 fl4; 5714 u32 mtu = 0; 5715 int err; 5716 5717 dev = dev_get_by_index_rcu(net, params->ifindex); 5718 if (unlikely(!dev)) 5719 return -ENODEV; 5720 5721 /* verify forwarding is enabled on this interface */ 5722 in_dev = __in_dev_get_rcu(dev); 5723 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5724 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5725 5726 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5727 fl4.flowi4_iif = 1; 5728 fl4.flowi4_oif = params->ifindex; 5729 } else { 5730 fl4.flowi4_iif = params->ifindex; 5731 fl4.flowi4_oif = 0; 5732 } 5733 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5734 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5735 fl4.flowi4_flags = 0; 5736 5737 fl4.flowi4_proto = params->l4_protocol; 5738 fl4.daddr = params->ipv4_dst; 5739 fl4.saddr = params->ipv4_src; 5740 fl4.fl4_sport = params->sport; 5741 fl4.fl4_dport = params->dport; 5742 fl4.flowi4_multipath_hash = 0; 5743 5744 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5745 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5746 struct fib_table *tb; 5747 5748 tb = fib_get_table(net, tbid); 5749 if (unlikely(!tb)) 5750 return BPF_FIB_LKUP_RET_NOT_FWDED; 5751 5752 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5753 } else { 5754 fl4.flowi4_mark = 0; 5755 fl4.flowi4_secid = 0; 5756 fl4.flowi4_tun_key.tun_id = 0; 5757 fl4.flowi4_uid = sock_net_uid(net, NULL); 5758 5759 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5760 } 5761 5762 if (err) { 5763 /* map fib lookup errors to RTN_ type */ 5764 if (err == -EINVAL) 5765 return BPF_FIB_LKUP_RET_BLACKHOLE; 5766 if (err == -EHOSTUNREACH) 5767 return BPF_FIB_LKUP_RET_UNREACHABLE; 5768 if (err == -EACCES) 5769 return BPF_FIB_LKUP_RET_PROHIBIT; 5770 5771 return BPF_FIB_LKUP_RET_NOT_FWDED; 5772 } 5773 5774 if (res.type != RTN_UNICAST) 5775 return BPF_FIB_LKUP_RET_NOT_FWDED; 5776 5777 if (fib_info_num_path(res.fi) > 1) 5778 fib_select_path(net, &res, &fl4, NULL); 5779 5780 if (check_mtu) { 5781 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5782 if (params->tot_len > mtu) { 5783 params->mtu_result = mtu; /* union with tot_len */ 5784 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5785 } 5786 } 5787 5788 nhc = res.nhc; 5789 5790 /* do not handle lwt encaps right now */ 5791 if (nhc->nhc_lwtstate) 5792 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5793 5794 dev = nhc->nhc_dev; 5795 5796 params->rt_metric = res.fi->fib_priority; 5797 params->ifindex = dev->ifindex; 5798 5799 /* xdp and cls_bpf programs are run in RCU-bh so 5800 * rcu_read_lock_bh is not needed here 5801 */ 5802 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5803 if (nhc->nhc_gw_family) 5804 params->ipv4_dst = nhc->nhc_gw.ipv4; 5805 5806 neigh = __ipv4_neigh_lookup_noref(dev, 5807 (__force u32)params->ipv4_dst); 5808 } else { 5809 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5810 5811 params->family = AF_INET6; 5812 *dst = nhc->nhc_gw.ipv6; 5813 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5814 } 5815 5816 if (!neigh) 5817 return BPF_FIB_LKUP_RET_NO_NEIGH; 5818 5819 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5820 } 5821 #endif 5822 5823 #if IS_ENABLED(CONFIG_IPV6) 5824 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5825 u32 flags, bool check_mtu) 5826 { 5827 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5828 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5829 struct fib6_result res = {}; 5830 struct neighbour *neigh; 5831 struct net_device *dev; 5832 struct inet6_dev *idev; 5833 struct flowi6 fl6; 5834 int strict = 0; 5835 int oif, err; 5836 u32 mtu = 0; 5837 5838 /* link local addresses are never forwarded */ 5839 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5840 return BPF_FIB_LKUP_RET_NOT_FWDED; 5841 5842 dev = dev_get_by_index_rcu(net, params->ifindex); 5843 if (unlikely(!dev)) 5844 return -ENODEV; 5845 5846 idev = __in6_dev_get_safely(dev); 5847 if (unlikely(!idev || !idev->cnf.forwarding)) 5848 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5849 5850 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5851 fl6.flowi6_iif = 1; 5852 oif = fl6.flowi6_oif = params->ifindex; 5853 } else { 5854 oif = fl6.flowi6_iif = params->ifindex; 5855 fl6.flowi6_oif = 0; 5856 strict = RT6_LOOKUP_F_HAS_SADDR; 5857 } 5858 fl6.flowlabel = params->flowinfo; 5859 fl6.flowi6_scope = 0; 5860 fl6.flowi6_flags = 0; 5861 fl6.mp_hash = 0; 5862 5863 fl6.flowi6_proto = params->l4_protocol; 5864 fl6.daddr = *dst; 5865 fl6.saddr = *src; 5866 fl6.fl6_sport = params->sport; 5867 fl6.fl6_dport = params->dport; 5868 5869 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5870 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5871 struct fib6_table *tb; 5872 5873 tb = ipv6_stub->fib6_get_table(net, tbid); 5874 if (unlikely(!tb)) 5875 return BPF_FIB_LKUP_RET_NOT_FWDED; 5876 5877 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5878 strict); 5879 } else { 5880 fl6.flowi6_mark = 0; 5881 fl6.flowi6_secid = 0; 5882 fl6.flowi6_tun_key.tun_id = 0; 5883 fl6.flowi6_uid = sock_net_uid(net, NULL); 5884 5885 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5886 } 5887 5888 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5889 res.f6i == net->ipv6.fib6_null_entry)) 5890 return BPF_FIB_LKUP_RET_NOT_FWDED; 5891 5892 switch (res.fib6_type) { 5893 /* only unicast is forwarded */ 5894 case RTN_UNICAST: 5895 break; 5896 case RTN_BLACKHOLE: 5897 return BPF_FIB_LKUP_RET_BLACKHOLE; 5898 case RTN_UNREACHABLE: 5899 return BPF_FIB_LKUP_RET_UNREACHABLE; 5900 case RTN_PROHIBIT: 5901 return BPF_FIB_LKUP_RET_PROHIBIT; 5902 default: 5903 return BPF_FIB_LKUP_RET_NOT_FWDED; 5904 } 5905 5906 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5907 fl6.flowi6_oif != 0, NULL, strict); 5908 5909 if (check_mtu) { 5910 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5911 if (params->tot_len > mtu) { 5912 params->mtu_result = mtu; /* union with tot_len */ 5913 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5914 } 5915 } 5916 5917 if (res.nh->fib_nh_lws) 5918 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5919 5920 if (res.nh->fib_nh_gw_family) 5921 *dst = res.nh->fib_nh_gw6; 5922 5923 dev = res.nh->fib_nh_dev; 5924 params->rt_metric = res.f6i->fib6_metric; 5925 params->ifindex = dev->ifindex; 5926 5927 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5928 * not needed here. 5929 */ 5930 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5931 if (!neigh) 5932 return BPF_FIB_LKUP_RET_NO_NEIGH; 5933 5934 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5935 } 5936 #endif 5937 5938 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 5939 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5940 { 5941 if (plen < sizeof(*params)) 5942 return -EINVAL; 5943 5944 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5945 return -EINVAL; 5946 5947 switch (params->family) { 5948 #if IS_ENABLED(CONFIG_INET) 5949 case AF_INET: 5950 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 5951 flags, true); 5952 #endif 5953 #if IS_ENABLED(CONFIG_IPV6) 5954 case AF_INET6: 5955 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 5956 flags, true); 5957 #endif 5958 } 5959 return -EAFNOSUPPORT; 5960 } 5961 5962 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 5963 .func = bpf_xdp_fib_lookup, 5964 .gpl_only = true, 5965 .ret_type = RET_INTEGER, 5966 .arg1_type = ARG_PTR_TO_CTX, 5967 .arg2_type = ARG_PTR_TO_MEM, 5968 .arg3_type = ARG_CONST_SIZE, 5969 .arg4_type = ARG_ANYTHING, 5970 }; 5971 5972 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 5973 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5974 { 5975 struct net *net = dev_net(skb->dev); 5976 int rc = -EAFNOSUPPORT; 5977 bool check_mtu = false; 5978 5979 if (plen < sizeof(*params)) 5980 return -EINVAL; 5981 5982 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5983 return -EINVAL; 5984 5985 if (params->tot_len) 5986 check_mtu = true; 5987 5988 switch (params->family) { 5989 #if IS_ENABLED(CONFIG_INET) 5990 case AF_INET: 5991 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu); 5992 break; 5993 #endif 5994 #if IS_ENABLED(CONFIG_IPV6) 5995 case AF_INET6: 5996 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu); 5997 break; 5998 #endif 5999 } 6000 6001 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) { 6002 struct net_device *dev; 6003 6004 /* When tot_len isn't provided by user, check skb 6005 * against MTU of FIB lookup resulting net_device 6006 */ 6007 dev = dev_get_by_index_rcu(net, params->ifindex); 6008 if (!is_skb_forwardable(dev, skb)) 6009 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 6010 6011 params->mtu_result = dev->mtu; /* union with tot_len */ 6012 } 6013 6014 return rc; 6015 } 6016 6017 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 6018 .func = bpf_skb_fib_lookup, 6019 .gpl_only = true, 6020 .ret_type = RET_INTEGER, 6021 .arg1_type = ARG_PTR_TO_CTX, 6022 .arg2_type = ARG_PTR_TO_MEM, 6023 .arg3_type = ARG_CONST_SIZE, 6024 .arg4_type = ARG_ANYTHING, 6025 }; 6026 6027 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr, 6028 u32 ifindex) 6029 { 6030 struct net *netns = dev_net(dev_curr); 6031 6032 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */ 6033 if (ifindex == 0) 6034 return dev_curr; 6035 6036 return dev_get_by_index_rcu(netns, ifindex); 6037 } 6038 6039 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb, 6040 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6041 { 6042 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6043 struct net_device *dev = skb->dev; 6044 int skb_len, dev_len; 6045 int mtu; 6046 6047 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) 6048 return -EINVAL; 6049 6050 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) 6051 return -EINVAL; 6052 6053 dev = __dev_via_ifindex(dev, ifindex); 6054 if (unlikely(!dev)) 6055 return -ENODEV; 6056 6057 mtu = READ_ONCE(dev->mtu); 6058 6059 dev_len = mtu + dev->hard_header_len; 6060 6061 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6062 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len; 6063 6064 skb_len += len_diff; /* minus result pass check */ 6065 if (skb_len <= dev_len) { 6066 ret = BPF_MTU_CHK_RET_SUCCESS; 6067 goto out; 6068 } 6069 /* At this point, skb->len exceed MTU, but as it include length of all 6070 * segments, it can still be below MTU. The SKB can possibly get 6071 * re-segmented in transmit path (see validate_xmit_skb). Thus, user 6072 * must choose if segs are to be MTU checked. 6073 */ 6074 if (skb_is_gso(skb)) { 6075 ret = BPF_MTU_CHK_RET_SUCCESS; 6076 6077 if (flags & BPF_MTU_CHK_SEGS && 6078 !skb_gso_validate_network_len(skb, mtu)) 6079 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG; 6080 } 6081 out: 6082 /* BPF verifier guarantees valid pointer */ 6083 *mtu_len = mtu; 6084 6085 return ret; 6086 } 6087 6088 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp, 6089 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6090 { 6091 struct net_device *dev = xdp->rxq->dev; 6092 int xdp_len = xdp->data_end - xdp->data; 6093 int ret = BPF_MTU_CHK_RET_SUCCESS; 6094 int mtu, dev_len; 6095 6096 /* XDP variant doesn't support multi-buffer segment check (yet) */ 6097 if (unlikely(flags)) 6098 return -EINVAL; 6099 6100 dev = __dev_via_ifindex(dev, ifindex); 6101 if (unlikely(!dev)) 6102 return -ENODEV; 6103 6104 mtu = READ_ONCE(dev->mtu); 6105 6106 /* Add L2-header as dev MTU is L3 size */ 6107 dev_len = mtu + dev->hard_header_len; 6108 6109 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6110 if (*mtu_len) 6111 xdp_len = *mtu_len + dev->hard_header_len; 6112 6113 xdp_len += len_diff; /* minus result pass check */ 6114 if (xdp_len > dev_len) 6115 ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6116 6117 /* BPF verifier guarantees valid pointer */ 6118 *mtu_len = mtu; 6119 6120 return ret; 6121 } 6122 6123 static const struct bpf_func_proto bpf_skb_check_mtu_proto = { 6124 .func = bpf_skb_check_mtu, 6125 .gpl_only = true, 6126 .ret_type = RET_INTEGER, 6127 .arg1_type = ARG_PTR_TO_CTX, 6128 .arg2_type = ARG_ANYTHING, 6129 .arg3_type = ARG_PTR_TO_INT, 6130 .arg4_type = ARG_ANYTHING, 6131 .arg5_type = ARG_ANYTHING, 6132 }; 6133 6134 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = { 6135 .func = bpf_xdp_check_mtu, 6136 .gpl_only = true, 6137 .ret_type = RET_INTEGER, 6138 .arg1_type = ARG_PTR_TO_CTX, 6139 .arg2_type = ARG_ANYTHING, 6140 .arg3_type = ARG_PTR_TO_INT, 6141 .arg4_type = ARG_ANYTHING, 6142 .arg5_type = ARG_ANYTHING, 6143 }; 6144 6145 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6146 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 6147 { 6148 int err; 6149 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 6150 6151 if (!seg6_validate_srh(srh, len, false)) 6152 return -EINVAL; 6153 6154 switch (type) { 6155 case BPF_LWT_ENCAP_SEG6_INLINE: 6156 if (skb->protocol != htons(ETH_P_IPV6)) 6157 return -EBADMSG; 6158 6159 err = seg6_do_srh_inline(skb, srh); 6160 break; 6161 case BPF_LWT_ENCAP_SEG6: 6162 skb_reset_inner_headers(skb); 6163 skb->encapsulation = 1; 6164 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 6165 break; 6166 default: 6167 return -EINVAL; 6168 } 6169 6170 bpf_compute_data_pointers(skb); 6171 if (err) 6172 return err; 6173 6174 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 6175 6176 return seg6_lookup_nexthop(skb, NULL, 0); 6177 } 6178 #endif /* CONFIG_IPV6_SEG6_BPF */ 6179 6180 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6181 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 6182 bool ingress) 6183 { 6184 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 6185 } 6186 #endif 6187 6188 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 6189 u32, len) 6190 { 6191 switch (type) { 6192 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6193 case BPF_LWT_ENCAP_SEG6: 6194 case BPF_LWT_ENCAP_SEG6_INLINE: 6195 return bpf_push_seg6_encap(skb, type, hdr, len); 6196 #endif 6197 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6198 case BPF_LWT_ENCAP_IP: 6199 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 6200 #endif 6201 default: 6202 return -EINVAL; 6203 } 6204 } 6205 6206 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 6207 void *, hdr, u32, len) 6208 { 6209 switch (type) { 6210 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6211 case BPF_LWT_ENCAP_IP: 6212 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 6213 #endif 6214 default: 6215 return -EINVAL; 6216 } 6217 } 6218 6219 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 6220 .func = bpf_lwt_in_push_encap, 6221 .gpl_only = false, 6222 .ret_type = RET_INTEGER, 6223 .arg1_type = ARG_PTR_TO_CTX, 6224 .arg2_type = ARG_ANYTHING, 6225 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6226 .arg4_type = ARG_CONST_SIZE 6227 }; 6228 6229 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 6230 .func = bpf_lwt_xmit_push_encap, 6231 .gpl_only = false, 6232 .ret_type = RET_INTEGER, 6233 .arg1_type = ARG_PTR_TO_CTX, 6234 .arg2_type = ARG_ANYTHING, 6235 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6236 .arg4_type = ARG_CONST_SIZE 6237 }; 6238 6239 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6240 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 6241 const void *, from, u32, len) 6242 { 6243 struct seg6_bpf_srh_state *srh_state = 6244 this_cpu_ptr(&seg6_bpf_srh_states); 6245 struct ipv6_sr_hdr *srh = srh_state->srh; 6246 void *srh_tlvs, *srh_end, *ptr; 6247 int srhoff = 0; 6248 6249 if (srh == NULL) 6250 return -EINVAL; 6251 6252 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 6253 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 6254 6255 ptr = skb->data + offset; 6256 if (ptr >= srh_tlvs && ptr + len <= srh_end) 6257 srh_state->valid = false; 6258 else if (ptr < (void *)&srh->flags || 6259 ptr + len > (void *)&srh->segments) 6260 return -EFAULT; 6261 6262 if (unlikely(bpf_try_make_writable(skb, offset + len))) 6263 return -EFAULT; 6264 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6265 return -EINVAL; 6266 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6267 6268 memcpy(skb->data + offset, from, len); 6269 return 0; 6270 } 6271 6272 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 6273 .func = bpf_lwt_seg6_store_bytes, 6274 .gpl_only = false, 6275 .ret_type = RET_INTEGER, 6276 .arg1_type = ARG_PTR_TO_CTX, 6277 .arg2_type = ARG_ANYTHING, 6278 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6279 .arg4_type = ARG_CONST_SIZE 6280 }; 6281 6282 static void bpf_update_srh_state(struct sk_buff *skb) 6283 { 6284 struct seg6_bpf_srh_state *srh_state = 6285 this_cpu_ptr(&seg6_bpf_srh_states); 6286 int srhoff = 0; 6287 6288 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 6289 srh_state->srh = NULL; 6290 } else { 6291 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6292 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 6293 srh_state->valid = true; 6294 } 6295 } 6296 6297 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 6298 u32, action, void *, param, u32, param_len) 6299 { 6300 struct seg6_bpf_srh_state *srh_state = 6301 this_cpu_ptr(&seg6_bpf_srh_states); 6302 int hdroff = 0; 6303 int err; 6304 6305 switch (action) { 6306 case SEG6_LOCAL_ACTION_END_X: 6307 if (!seg6_bpf_has_valid_srh(skb)) 6308 return -EBADMSG; 6309 if (param_len != sizeof(struct in6_addr)) 6310 return -EINVAL; 6311 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 6312 case SEG6_LOCAL_ACTION_END_T: 6313 if (!seg6_bpf_has_valid_srh(skb)) 6314 return -EBADMSG; 6315 if (param_len != sizeof(int)) 6316 return -EINVAL; 6317 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6318 case SEG6_LOCAL_ACTION_END_DT6: 6319 if (!seg6_bpf_has_valid_srh(skb)) 6320 return -EBADMSG; 6321 if (param_len != sizeof(int)) 6322 return -EINVAL; 6323 6324 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 6325 return -EBADMSG; 6326 if (!pskb_pull(skb, hdroff)) 6327 return -EBADMSG; 6328 6329 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 6330 skb_reset_network_header(skb); 6331 skb_reset_transport_header(skb); 6332 skb->encapsulation = 0; 6333 6334 bpf_compute_data_pointers(skb); 6335 bpf_update_srh_state(skb); 6336 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6337 case SEG6_LOCAL_ACTION_END_B6: 6338 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6339 return -EBADMSG; 6340 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 6341 param, param_len); 6342 if (!err) 6343 bpf_update_srh_state(skb); 6344 6345 return err; 6346 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 6347 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6348 return -EBADMSG; 6349 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 6350 param, param_len); 6351 if (!err) 6352 bpf_update_srh_state(skb); 6353 6354 return err; 6355 default: 6356 return -EINVAL; 6357 } 6358 } 6359 6360 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 6361 .func = bpf_lwt_seg6_action, 6362 .gpl_only = false, 6363 .ret_type = RET_INTEGER, 6364 .arg1_type = ARG_PTR_TO_CTX, 6365 .arg2_type = ARG_ANYTHING, 6366 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6367 .arg4_type = ARG_CONST_SIZE 6368 }; 6369 6370 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 6371 s32, len) 6372 { 6373 struct seg6_bpf_srh_state *srh_state = 6374 this_cpu_ptr(&seg6_bpf_srh_states); 6375 struct ipv6_sr_hdr *srh = srh_state->srh; 6376 void *srh_end, *srh_tlvs, *ptr; 6377 struct ipv6hdr *hdr; 6378 int srhoff = 0; 6379 int ret; 6380 6381 if (unlikely(srh == NULL)) 6382 return -EINVAL; 6383 6384 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 6385 ((srh->first_segment + 1) << 4)); 6386 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 6387 srh_state->hdrlen); 6388 ptr = skb->data + offset; 6389 6390 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 6391 return -EFAULT; 6392 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 6393 return -EFAULT; 6394 6395 if (len > 0) { 6396 ret = skb_cow_head(skb, len); 6397 if (unlikely(ret < 0)) 6398 return ret; 6399 6400 ret = bpf_skb_net_hdr_push(skb, offset, len); 6401 } else { 6402 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 6403 } 6404 6405 bpf_compute_data_pointers(skb); 6406 if (unlikely(ret < 0)) 6407 return ret; 6408 6409 hdr = (struct ipv6hdr *)skb->data; 6410 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 6411 6412 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6413 return -EINVAL; 6414 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6415 srh_state->hdrlen += len; 6416 srh_state->valid = false; 6417 return 0; 6418 } 6419 6420 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 6421 .func = bpf_lwt_seg6_adjust_srh, 6422 .gpl_only = false, 6423 .ret_type = RET_INTEGER, 6424 .arg1_type = ARG_PTR_TO_CTX, 6425 .arg2_type = ARG_ANYTHING, 6426 .arg3_type = ARG_ANYTHING, 6427 }; 6428 #endif /* CONFIG_IPV6_SEG6_BPF */ 6429 6430 #ifdef CONFIG_INET 6431 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 6432 int dif, int sdif, u8 family, u8 proto) 6433 { 6434 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo; 6435 bool refcounted = false; 6436 struct sock *sk = NULL; 6437 6438 if (family == AF_INET) { 6439 __be32 src4 = tuple->ipv4.saddr; 6440 __be32 dst4 = tuple->ipv4.daddr; 6441 6442 if (proto == IPPROTO_TCP) 6443 sk = __inet_lookup(net, hinfo, NULL, 0, 6444 src4, tuple->ipv4.sport, 6445 dst4, tuple->ipv4.dport, 6446 dif, sdif, &refcounted); 6447 else 6448 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 6449 dst4, tuple->ipv4.dport, 6450 dif, sdif, net->ipv4.udp_table, NULL); 6451 #if IS_ENABLED(CONFIG_IPV6) 6452 } else { 6453 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 6454 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 6455 6456 if (proto == IPPROTO_TCP) 6457 sk = __inet6_lookup(net, hinfo, NULL, 0, 6458 src6, tuple->ipv6.sport, 6459 dst6, ntohs(tuple->ipv6.dport), 6460 dif, sdif, &refcounted); 6461 else if (likely(ipv6_bpf_stub)) 6462 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 6463 src6, tuple->ipv6.sport, 6464 dst6, tuple->ipv6.dport, 6465 dif, sdif, 6466 net->ipv4.udp_table, NULL); 6467 #endif 6468 } 6469 6470 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 6471 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6472 sk = NULL; 6473 } 6474 return sk; 6475 } 6476 6477 /* bpf_skc_lookup performs the core lookup for different types of sockets, 6478 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 6479 */ 6480 static struct sock * 6481 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6482 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6483 u64 flags) 6484 { 6485 struct sock *sk = NULL; 6486 struct net *net; 6487 u8 family; 6488 int sdif; 6489 6490 if (len == sizeof(tuple->ipv4)) 6491 family = AF_INET; 6492 else if (len == sizeof(tuple->ipv6)) 6493 family = AF_INET6; 6494 else 6495 return NULL; 6496 6497 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX))) 6498 goto out; 6499 6500 if (family == AF_INET) 6501 sdif = inet_sdif(skb); 6502 else 6503 sdif = inet6_sdif(skb); 6504 6505 if ((s32)netns_id < 0) { 6506 net = caller_net; 6507 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6508 } else { 6509 net = get_net_ns_by_id(caller_net, netns_id); 6510 if (unlikely(!net)) 6511 goto out; 6512 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6513 put_net(net); 6514 } 6515 6516 out: 6517 return sk; 6518 } 6519 6520 static struct sock * 6521 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6522 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6523 u64 flags) 6524 { 6525 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 6526 ifindex, proto, netns_id, flags); 6527 6528 if (sk) { 6529 struct sock *sk2 = sk_to_full_sk(sk); 6530 6531 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6532 * sock refcnt is decremented to prevent a request_sock leak. 6533 */ 6534 if (!sk_fullsock(sk2)) 6535 sk2 = NULL; 6536 if (sk2 != sk) { 6537 sock_gen_put(sk); 6538 /* Ensure there is no need to bump sk2 refcnt */ 6539 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6540 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6541 return NULL; 6542 } 6543 sk = sk2; 6544 } 6545 } 6546 6547 return sk; 6548 } 6549 6550 static struct sock * 6551 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6552 u8 proto, u64 netns_id, u64 flags) 6553 { 6554 struct net *caller_net; 6555 int ifindex; 6556 6557 if (skb->dev) { 6558 caller_net = dev_net(skb->dev); 6559 ifindex = skb->dev->ifindex; 6560 } else { 6561 caller_net = sock_net(skb->sk); 6562 ifindex = 0; 6563 } 6564 6565 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6566 netns_id, flags); 6567 } 6568 6569 static struct sock * 6570 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6571 u8 proto, u64 netns_id, u64 flags) 6572 { 6573 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6574 flags); 6575 6576 if (sk) { 6577 struct sock *sk2 = sk_to_full_sk(sk); 6578 6579 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6580 * sock refcnt is decremented to prevent a request_sock leak. 6581 */ 6582 if (!sk_fullsock(sk2)) 6583 sk2 = NULL; 6584 if (sk2 != sk) { 6585 sock_gen_put(sk); 6586 /* Ensure there is no need to bump sk2 refcnt */ 6587 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6588 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6589 return NULL; 6590 } 6591 sk = sk2; 6592 } 6593 } 6594 6595 return sk; 6596 } 6597 6598 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6599 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6600 { 6601 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6602 netns_id, flags); 6603 } 6604 6605 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6606 .func = bpf_skc_lookup_tcp, 6607 .gpl_only = false, 6608 .pkt_access = true, 6609 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6610 .arg1_type = ARG_PTR_TO_CTX, 6611 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6612 .arg3_type = ARG_CONST_SIZE, 6613 .arg4_type = ARG_ANYTHING, 6614 .arg5_type = ARG_ANYTHING, 6615 }; 6616 6617 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6618 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6619 { 6620 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6621 netns_id, flags); 6622 } 6623 6624 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6625 .func = bpf_sk_lookup_tcp, 6626 .gpl_only = false, 6627 .pkt_access = true, 6628 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6629 .arg1_type = ARG_PTR_TO_CTX, 6630 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6631 .arg3_type = ARG_CONST_SIZE, 6632 .arg4_type = ARG_ANYTHING, 6633 .arg5_type = ARG_ANYTHING, 6634 }; 6635 6636 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6637 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6638 { 6639 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6640 netns_id, flags); 6641 } 6642 6643 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6644 .func = bpf_sk_lookup_udp, 6645 .gpl_only = false, 6646 .pkt_access = true, 6647 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6648 .arg1_type = ARG_PTR_TO_CTX, 6649 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6650 .arg3_type = ARG_CONST_SIZE, 6651 .arg4_type = ARG_ANYTHING, 6652 .arg5_type = ARG_ANYTHING, 6653 }; 6654 6655 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6656 { 6657 if (sk && sk_is_refcounted(sk)) 6658 sock_gen_put(sk); 6659 return 0; 6660 } 6661 6662 static const struct bpf_func_proto bpf_sk_release_proto = { 6663 .func = bpf_sk_release, 6664 .gpl_only = false, 6665 .ret_type = RET_INTEGER, 6666 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE, 6667 }; 6668 6669 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6670 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6671 { 6672 struct net *caller_net = dev_net(ctx->rxq->dev); 6673 int ifindex = ctx->rxq->dev->ifindex; 6674 6675 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6676 ifindex, IPPROTO_UDP, netns_id, 6677 flags); 6678 } 6679 6680 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6681 .func = bpf_xdp_sk_lookup_udp, 6682 .gpl_only = false, 6683 .pkt_access = true, 6684 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6685 .arg1_type = ARG_PTR_TO_CTX, 6686 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6687 .arg3_type = ARG_CONST_SIZE, 6688 .arg4_type = ARG_ANYTHING, 6689 .arg5_type = ARG_ANYTHING, 6690 }; 6691 6692 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6693 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6694 { 6695 struct net *caller_net = dev_net(ctx->rxq->dev); 6696 int ifindex = ctx->rxq->dev->ifindex; 6697 6698 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6699 ifindex, IPPROTO_TCP, netns_id, 6700 flags); 6701 } 6702 6703 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6704 .func = bpf_xdp_skc_lookup_tcp, 6705 .gpl_only = false, 6706 .pkt_access = true, 6707 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6708 .arg1_type = ARG_PTR_TO_CTX, 6709 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6710 .arg3_type = ARG_CONST_SIZE, 6711 .arg4_type = ARG_ANYTHING, 6712 .arg5_type = ARG_ANYTHING, 6713 }; 6714 6715 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6716 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6717 { 6718 struct net *caller_net = dev_net(ctx->rxq->dev); 6719 int ifindex = ctx->rxq->dev->ifindex; 6720 6721 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6722 ifindex, IPPROTO_TCP, netns_id, 6723 flags); 6724 } 6725 6726 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6727 .func = bpf_xdp_sk_lookup_tcp, 6728 .gpl_only = false, 6729 .pkt_access = true, 6730 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6731 .arg1_type = ARG_PTR_TO_CTX, 6732 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6733 .arg3_type = ARG_CONST_SIZE, 6734 .arg4_type = ARG_ANYTHING, 6735 .arg5_type = ARG_ANYTHING, 6736 }; 6737 6738 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6739 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6740 { 6741 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6742 sock_net(ctx->sk), 0, 6743 IPPROTO_TCP, netns_id, flags); 6744 } 6745 6746 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6747 .func = bpf_sock_addr_skc_lookup_tcp, 6748 .gpl_only = false, 6749 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6750 .arg1_type = ARG_PTR_TO_CTX, 6751 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6752 .arg3_type = ARG_CONST_SIZE, 6753 .arg4_type = ARG_ANYTHING, 6754 .arg5_type = ARG_ANYTHING, 6755 }; 6756 6757 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6758 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6759 { 6760 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6761 sock_net(ctx->sk), 0, IPPROTO_TCP, 6762 netns_id, flags); 6763 } 6764 6765 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6766 .func = bpf_sock_addr_sk_lookup_tcp, 6767 .gpl_only = false, 6768 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6769 .arg1_type = ARG_PTR_TO_CTX, 6770 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6771 .arg3_type = ARG_CONST_SIZE, 6772 .arg4_type = ARG_ANYTHING, 6773 .arg5_type = ARG_ANYTHING, 6774 }; 6775 6776 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6777 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6778 { 6779 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6780 sock_net(ctx->sk), 0, IPPROTO_UDP, 6781 netns_id, flags); 6782 } 6783 6784 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6785 .func = bpf_sock_addr_sk_lookup_udp, 6786 .gpl_only = false, 6787 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6788 .arg1_type = ARG_PTR_TO_CTX, 6789 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6790 .arg3_type = ARG_CONST_SIZE, 6791 .arg4_type = ARG_ANYTHING, 6792 .arg5_type = ARG_ANYTHING, 6793 }; 6794 6795 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6796 struct bpf_insn_access_aux *info) 6797 { 6798 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6799 icsk_retransmits)) 6800 return false; 6801 6802 if (off % size != 0) 6803 return false; 6804 6805 switch (off) { 6806 case offsetof(struct bpf_tcp_sock, bytes_received): 6807 case offsetof(struct bpf_tcp_sock, bytes_acked): 6808 return size == sizeof(__u64); 6809 default: 6810 return size == sizeof(__u32); 6811 } 6812 } 6813 6814 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6815 const struct bpf_insn *si, 6816 struct bpf_insn *insn_buf, 6817 struct bpf_prog *prog, u32 *target_size) 6818 { 6819 struct bpf_insn *insn = insn_buf; 6820 6821 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6822 do { \ 6823 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6824 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6825 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6826 si->dst_reg, si->src_reg, \ 6827 offsetof(struct tcp_sock, FIELD)); \ 6828 } while (0) 6829 6830 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6831 do { \ 6832 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6833 FIELD) > \ 6834 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6835 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6836 struct inet_connection_sock, \ 6837 FIELD), \ 6838 si->dst_reg, si->src_reg, \ 6839 offsetof( \ 6840 struct inet_connection_sock, \ 6841 FIELD)); \ 6842 } while (0) 6843 6844 if (insn > insn_buf) 6845 return insn - insn_buf; 6846 6847 switch (si->off) { 6848 case offsetof(struct bpf_tcp_sock, rtt_min): 6849 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6850 sizeof(struct minmax)); 6851 BUILD_BUG_ON(sizeof(struct minmax) < 6852 sizeof(struct minmax_sample)); 6853 6854 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6855 offsetof(struct tcp_sock, rtt_min) + 6856 offsetof(struct minmax_sample, v)); 6857 break; 6858 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6859 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6860 break; 6861 case offsetof(struct bpf_tcp_sock, srtt_us): 6862 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6863 break; 6864 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6865 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6866 break; 6867 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6868 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6869 break; 6870 case offsetof(struct bpf_tcp_sock, snd_nxt): 6871 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6872 break; 6873 case offsetof(struct bpf_tcp_sock, snd_una): 6874 BPF_TCP_SOCK_GET_COMMON(snd_una); 6875 break; 6876 case offsetof(struct bpf_tcp_sock, mss_cache): 6877 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6878 break; 6879 case offsetof(struct bpf_tcp_sock, ecn_flags): 6880 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6881 break; 6882 case offsetof(struct bpf_tcp_sock, rate_delivered): 6883 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6884 break; 6885 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6886 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6887 break; 6888 case offsetof(struct bpf_tcp_sock, packets_out): 6889 BPF_TCP_SOCK_GET_COMMON(packets_out); 6890 break; 6891 case offsetof(struct bpf_tcp_sock, retrans_out): 6892 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6893 break; 6894 case offsetof(struct bpf_tcp_sock, total_retrans): 6895 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6896 break; 6897 case offsetof(struct bpf_tcp_sock, segs_in): 6898 BPF_TCP_SOCK_GET_COMMON(segs_in); 6899 break; 6900 case offsetof(struct bpf_tcp_sock, data_segs_in): 6901 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6902 break; 6903 case offsetof(struct bpf_tcp_sock, segs_out): 6904 BPF_TCP_SOCK_GET_COMMON(segs_out); 6905 break; 6906 case offsetof(struct bpf_tcp_sock, data_segs_out): 6907 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6908 break; 6909 case offsetof(struct bpf_tcp_sock, lost_out): 6910 BPF_TCP_SOCK_GET_COMMON(lost_out); 6911 break; 6912 case offsetof(struct bpf_tcp_sock, sacked_out): 6913 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6914 break; 6915 case offsetof(struct bpf_tcp_sock, bytes_received): 6916 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6917 break; 6918 case offsetof(struct bpf_tcp_sock, bytes_acked): 6919 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6920 break; 6921 case offsetof(struct bpf_tcp_sock, dsack_dups): 6922 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6923 break; 6924 case offsetof(struct bpf_tcp_sock, delivered): 6925 BPF_TCP_SOCK_GET_COMMON(delivered); 6926 break; 6927 case offsetof(struct bpf_tcp_sock, delivered_ce): 6928 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 6929 break; 6930 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 6931 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 6932 break; 6933 } 6934 6935 return insn - insn_buf; 6936 } 6937 6938 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 6939 { 6940 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 6941 return (unsigned long)sk; 6942 6943 return (unsigned long)NULL; 6944 } 6945 6946 const struct bpf_func_proto bpf_tcp_sock_proto = { 6947 .func = bpf_tcp_sock, 6948 .gpl_only = false, 6949 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 6950 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6951 }; 6952 6953 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 6954 { 6955 sk = sk_to_full_sk(sk); 6956 6957 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 6958 return (unsigned long)sk; 6959 6960 return (unsigned long)NULL; 6961 } 6962 6963 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 6964 .func = bpf_get_listener_sock, 6965 .gpl_only = false, 6966 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6967 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6968 }; 6969 6970 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 6971 { 6972 unsigned int iphdr_len; 6973 6974 switch (skb_protocol(skb, true)) { 6975 case cpu_to_be16(ETH_P_IP): 6976 iphdr_len = sizeof(struct iphdr); 6977 break; 6978 case cpu_to_be16(ETH_P_IPV6): 6979 iphdr_len = sizeof(struct ipv6hdr); 6980 break; 6981 default: 6982 return 0; 6983 } 6984 6985 if (skb_headlen(skb) < iphdr_len) 6986 return 0; 6987 6988 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 6989 return 0; 6990 6991 return INET_ECN_set_ce(skb); 6992 } 6993 6994 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6995 struct bpf_insn_access_aux *info) 6996 { 6997 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 6998 return false; 6999 7000 if (off % size != 0) 7001 return false; 7002 7003 switch (off) { 7004 default: 7005 return size == sizeof(__u32); 7006 } 7007 } 7008 7009 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 7010 const struct bpf_insn *si, 7011 struct bpf_insn *insn_buf, 7012 struct bpf_prog *prog, u32 *target_size) 7013 { 7014 struct bpf_insn *insn = insn_buf; 7015 7016 #define BPF_XDP_SOCK_GET(FIELD) \ 7017 do { \ 7018 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 7019 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 7020 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 7021 si->dst_reg, si->src_reg, \ 7022 offsetof(struct xdp_sock, FIELD)); \ 7023 } while (0) 7024 7025 switch (si->off) { 7026 case offsetof(struct bpf_xdp_sock, queue_id): 7027 BPF_XDP_SOCK_GET(queue_id); 7028 break; 7029 } 7030 7031 return insn - insn_buf; 7032 } 7033 7034 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 7035 .func = bpf_skb_ecn_set_ce, 7036 .gpl_only = false, 7037 .ret_type = RET_INTEGER, 7038 .arg1_type = ARG_PTR_TO_CTX, 7039 }; 7040 7041 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7042 struct tcphdr *, th, u32, th_len) 7043 { 7044 #ifdef CONFIG_SYN_COOKIES 7045 u32 cookie; 7046 int ret; 7047 7048 if (unlikely(!sk || th_len < sizeof(*th))) 7049 return -EINVAL; 7050 7051 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 7052 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7053 return -EINVAL; 7054 7055 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7056 return -EINVAL; 7057 7058 if (!th->ack || th->rst || th->syn) 7059 return -ENOENT; 7060 7061 if (unlikely(iph_len < sizeof(struct iphdr))) 7062 return -EINVAL; 7063 7064 if (tcp_synq_no_recent_overflow(sk)) 7065 return -ENOENT; 7066 7067 cookie = ntohl(th->ack_seq) - 1; 7068 7069 /* Both struct iphdr and struct ipv6hdr have the version field at the 7070 * same offset so we can cast to the shorter header (struct iphdr). 7071 */ 7072 switch (((struct iphdr *)iph)->version) { 7073 case 4: 7074 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7075 return -EINVAL; 7076 7077 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 7078 break; 7079 7080 #if IS_BUILTIN(CONFIG_IPV6) 7081 case 6: 7082 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7083 return -EINVAL; 7084 7085 if (sk->sk_family != AF_INET6) 7086 return -EINVAL; 7087 7088 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 7089 break; 7090 #endif /* CONFIG_IPV6 */ 7091 7092 default: 7093 return -EPROTONOSUPPORT; 7094 } 7095 7096 if (ret > 0) 7097 return 0; 7098 7099 return -ENOENT; 7100 #else 7101 return -ENOTSUPP; 7102 #endif 7103 } 7104 7105 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 7106 .func = bpf_tcp_check_syncookie, 7107 .gpl_only = true, 7108 .pkt_access = true, 7109 .ret_type = RET_INTEGER, 7110 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7111 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7112 .arg3_type = ARG_CONST_SIZE, 7113 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7114 .arg5_type = ARG_CONST_SIZE, 7115 }; 7116 7117 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7118 struct tcphdr *, th, u32, th_len) 7119 { 7120 #ifdef CONFIG_SYN_COOKIES 7121 u32 cookie; 7122 u16 mss; 7123 7124 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 7125 return -EINVAL; 7126 7127 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7128 return -EINVAL; 7129 7130 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7131 return -ENOENT; 7132 7133 if (!th->syn || th->ack || th->fin || th->rst) 7134 return -EINVAL; 7135 7136 if (unlikely(iph_len < sizeof(struct iphdr))) 7137 return -EINVAL; 7138 7139 /* Both struct iphdr and struct ipv6hdr have the version field at the 7140 * same offset so we can cast to the shorter header (struct iphdr). 7141 */ 7142 switch (((struct iphdr *)iph)->version) { 7143 case 4: 7144 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7145 return -EINVAL; 7146 7147 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 7148 break; 7149 7150 #if IS_BUILTIN(CONFIG_IPV6) 7151 case 6: 7152 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7153 return -EINVAL; 7154 7155 if (sk->sk_family != AF_INET6) 7156 return -EINVAL; 7157 7158 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 7159 break; 7160 #endif /* CONFIG_IPV6 */ 7161 7162 default: 7163 return -EPROTONOSUPPORT; 7164 } 7165 if (mss == 0) 7166 return -ENOENT; 7167 7168 return cookie | ((u64)mss << 32); 7169 #else 7170 return -EOPNOTSUPP; 7171 #endif /* CONFIG_SYN_COOKIES */ 7172 } 7173 7174 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 7175 .func = bpf_tcp_gen_syncookie, 7176 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 7177 .pkt_access = true, 7178 .ret_type = RET_INTEGER, 7179 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7180 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7181 .arg3_type = ARG_CONST_SIZE, 7182 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7183 .arg5_type = ARG_CONST_SIZE, 7184 }; 7185 7186 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 7187 { 7188 if (!sk || flags != 0) 7189 return -EINVAL; 7190 if (!skb_at_tc_ingress(skb)) 7191 return -EOPNOTSUPP; 7192 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 7193 return -ENETUNREACH; 7194 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 7195 return -ESOCKTNOSUPPORT; 7196 if (sk_is_refcounted(sk) && 7197 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 7198 return -ENOENT; 7199 7200 skb_orphan(skb); 7201 skb->sk = sk; 7202 skb->destructor = sock_pfree; 7203 7204 return 0; 7205 } 7206 7207 static const struct bpf_func_proto bpf_sk_assign_proto = { 7208 .func = bpf_sk_assign, 7209 .gpl_only = false, 7210 .ret_type = RET_INTEGER, 7211 .arg1_type = ARG_PTR_TO_CTX, 7212 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7213 .arg3_type = ARG_ANYTHING, 7214 }; 7215 7216 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 7217 u8 search_kind, const u8 *magic, 7218 u8 magic_len, bool *eol) 7219 { 7220 u8 kind, kind_len; 7221 7222 *eol = false; 7223 7224 while (op < opend) { 7225 kind = op[0]; 7226 7227 if (kind == TCPOPT_EOL) { 7228 *eol = true; 7229 return ERR_PTR(-ENOMSG); 7230 } else if (kind == TCPOPT_NOP) { 7231 op++; 7232 continue; 7233 } 7234 7235 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 7236 /* Something is wrong in the received header. 7237 * Follow the TCP stack's tcp_parse_options() 7238 * and just bail here. 7239 */ 7240 return ERR_PTR(-EFAULT); 7241 7242 kind_len = op[1]; 7243 if (search_kind == kind) { 7244 if (!magic_len) 7245 return op; 7246 7247 if (magic_len > kind_len - 2) 7248 return ERR_PTR(-ENOMSG); 7249 7250 if (!memcmp(&op[2], magic, magic_len)) 7251 return op; 7252 } 7253 7254 op += kind_len; 7255 } 7256 7257 return ERR_PTR(-ENOMSG); 7258 } 7259 7260 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7261 void *, search_res, u32, len, u64, flags) 7262 { 7263 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 7264 const u8 *op, *opend, *magic, *search = search_res; 7265 u8 search_kind, search_len, copy_len, magic_len; 7266 int ret; 7267 7268 /* 2 byte is the minimal option len except TCPOPT_NOP and 7269 * TCPOPT_EOL which are useless for the bpf prog to learn 7270 * and this helper disallow loading them also. 7271 */ 7272 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 7273 return -EINVAL; 7274 7275 search_kind = search[0]; 7276 search_len = search[1]; 7277 7278 if (search_len > len || search_kind == TCPOPT_NOP || 7279 search_kind == TCPOPT_EOL) 7280 return -EINVAL; 7281 7282 if (search_kind == TCPOPT_EXP || search_kind == 253) { 7283 /* 16 or 32 bit magic. +2 for kind and kind length */ 7284 if (search_len != 4 && search_len != 6) 7285 return -EINVAL; 7286 magic = &search[2]; 7287 magic_len = search_len - 2; 7288 } else { 7289 if (search_len) 7290 return -EINVAL; 7291 magic = NULL; 7292 magic_len = 0; 7293 } 7294 7295 if (load_syn) { 7296 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 7297 if (ret < 0) 7298 return ret; 7299 7300 opend = op + ret; 7301 op += sizeof(struct tcphdr); 7302 } else { 7303 if (!bpf_sock->skb || 7304 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7305 /* This bpf_sock->op cannot call this helper */ 7306 return -EPERM; 7307 7308 opend = bpf_sock->skb_data_end; 7309 op = bpf_sock->skb->data + sizeof(struct tcphdr); 7310 } 7311 7312 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 7313 &eol); 7314 if (IS_ERR(op)) 7315 return PTR_ERR(op); 7316 7317 copy_len = op[1]; 7318 ret = copy_len; 7319 if (copy_len > len) { 7320 ret = -ENOSPC; 7321 copy_len = len; 7322 } 7323 7324 memcpy(search_res, op, copy_len); 7325 return ret; 7326 } 7327 7328 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 7329 .func = bpf_sock_ops_load_hdr_opt, 7330 .gpl_only = false, 7331 .ret_type = RET_INTEGER, 7332 .arg1_type = ARG_PTR_TO_CTX, 7333 .arg2_type = ARG_PTR_TO_MEM, 7334 .arg3_type = ARG_CONST_SIZE, 7335 .arg4_type = ARG_ANYTHING, 7336 }; 7337 7338 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7339 const void *, from, u32, len, u64, flags) 7340 { 7341 u8 new_kind, new_kind_len, magic_len = 0, *opend; 7342 const u8 *op, *new_op, *magic = NULL; 7343 struct sk_buff *skb; 7344 bool eol; 7345 7346 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 7347 return -EPERM; 7348 7349 if (len < 2 || flags) 7350 return -EINVAL; 7351 7352 new_op = from; 7353 new_kind = new_op[0]; 7354 new_kind_len = new_op[1]; 7355 7356 if (new_kind_len > len || new_kind == TCPOPT_NOP || 7357 new_kind == TCPOPT_EOL) 7358 return -EINVAL; 7359 7360 if (new_kind_len > bpf_sock->remaining_opt_len) 7361 return -ENOSPC; 7362 7363 /* 253 is another experimental kind */ 7364 if (new_kind == TCPOPT_EXP || new_kind == 253) { 7365 if (new_kind_len < 4) 7366 return -EINVAL; 7367 /* Match for the 2 byte magic also. 7368 * RFC 6994: the magic could be 2 or 4 bytes. 7369 * Hence, matching by 2 byte only is on the 7370 * conservative side but it is the right 7371 * thing to do for the 'search-for-duplication' 7372 * purpose. 7373 */ 7374 magic = &new_op[2]; 7375 magic_len = 2; 7376 } 7377 7378 /* Check for duplication */ 7379 skb = bpf_sock->skb; 7380 op = skb->data + sizeof(struct tcphdr); 7381 opend = bpf_sock->skb_data_end; 7382 7383 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 7384 &eol); 7385 if (!IS_ERR(op)) 7386 return -EEXIST; 7387 7388 if (PTR_ERR(op) != -ENOMSG) 7389 return PTR_ERR(op); 7390 7391 if (eol) 7392 /* The option has been ended. Treat it as no more 7393 * header option can be written. 7394 */ 7395 return -ENOSPC; 7396 7397 /* No duplication found. Store the header option. */ 7398 memcpy(opend, from, new_kind_len); 7399 7400 bpf_sock->remaining_opt_len -= new_kind_len; 7401 bpf_sock->skb_data_end += new_kind_len; 7402 7403 return 0; 7404 } 7405 7406 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 7407 .func = bpf_sock_ops_store_hdr_opt, 7408 .gpl_only = false, 7409 .ret_type = RET_INTEGER, 7410 .arg1_type = ARG_PTR_TO_CTX, 7411 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7412 .arg3_type = ARG_CONST_SIZE, 7413 .arg4_type = ARG_ANYTHING, 7414 }; 7415 7416 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7417 u32, len, u64, flags) 7418 { 7419 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7420 return -EPERM; 7421 7422 if (flags || len < 2) 7423 return -EINVAL; 7424 7425 if (len > bpf_sock->remaining_opt_len) 7426 return -ENOSPC; 7427 7428 bpf_sock->remaining_opt_len -= len; 7429 7430 return 0; 7431 } 7432 7433 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 7434 .func = bpf_sock_ops_reserve_hdr_opt, 7435 .gpl_only = false, 7436 .ret_type = RET_INTEGER, 7437 .arg1_type = ARG_PTR_TO_CTX, 7438 .arg2_type = ARG_ANYTHING, 7439 .arg3_type = ARG_ANYTHING, 7440 }; 7441 7442 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, 7443 u64, tstamp, u32, tstamp_type) 7444 { 7445 /* skb_clear_delivery_time() is done for inet protocol */ 7446 if (skb->protocol != htons(ETH_P_IP) && 7447 skb->protocol != htons(ETH_P_IPV6)) 7448 return -EOPNOTSUPP; 7449 7450 switch (tstamp_type) { 7451 case BPF_SKB_TSTAMP_DELIVERY_MONO: 7452 if (!tstamp) 7453 return -EINVAL; 7454 skb->tstamp = tstamp; 7455 skb->mono_delivery_time = 1; 7456 break; 7457 case BPF_SKB_TSTAMP_UNSPEC: 7458 if (tstamp) 7459 return -EINVAL; 7460 skb->tstamp = 0; 7461 skb->mono_delivery_time = 0; 7462 break; 7463 default: 7464 return -EINVAL; 7465 } 7466 7467 return 0; 7468 } 7469 7470 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = { 7471 .func = bpf_skb_set_tstamp, 7472 .gpl_only = false, 7473 .ret_type = RET_INTEGER, 7474 .arg1_type = ARG_PTR_TO_CTX, 7475 .arg2_type = ARG_ANYTHING, 7476 .arg3_type = ARG_ANYTHING, 7477 }; 7478 7479 #ifdef CONFIG_SYN_COOKIES 7480 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph, 7481 struct tcphdr *, th, u32, th_len) 7482 { 7483 u32 cookie; 7484 u16 mss; 7485 7486 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7487 return -EINVAL; 7488 7489 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT; 7490 cookie = __cookie_v4_init_sequence(iph, th, &mss); 7491 7492 return cookie | ((u64)mss << 32); 7493 } 7494 7495 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = { 7496 .func = bpf_tcp_raw_gen_syncookie_ipv4, 7497 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */ 7498 .pkt_access = true, 7499 .ret_type = RET_INTEGER, 7500 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7501 .arg1_size = sizeof(struct iphdr), 7502 .arg2_type = ARG_PTR_TO_MEM, 7503 .arg3_type = ARG_CONST_SIZE, 7504 }; 7505 7506 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph, 7507 struct tcphdr *, th, u32, th_len) 7508 { 7509 #if IS_BUILTIN(CONFIG_IPV6) 7510 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 7511 sizeof(struct ipv6hdr); 7512 u32 cookie; 7513 u16 mss; 7514 7515 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7516 return -EINVAL; 7517 7518 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp; 7519 cookie = __cookie_v6_init_sequence(iph, th, &mss); 7520 7521 return cookie | ((u64)mss << 32); 7522 #else 7523 return -EPROTONOSUPPORT; 7524 #endif 7525 } 7526 7527 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = { 7528 .func = bpf_tcp_raw_gen_syncookie_ipv6, 7529 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */ 7530 .pkt_access = true, 7531 .ret_type = RET_INTEGER, 7532 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7533 .arg1_size = sizeof(struct ipv6hdr), 7534 .arg2_type = ARG_PTR_TO_MEM, 7535 .arg3_type = ARG_CONST_SIZE, 7536 }; 7537 7538 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph, 7539 struct tcphdr *, th) 7540 { 7541 u32 cookie = ntohl(th->ack_seq) - 1; 7542 7543 if (__cookie_v4_check(iph, th, cookie) > 0) 7544 return 0; 7545 7546 return -EACCES; 7547 } 7548 7549 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = { 7550 .func = bpf_tcp_raw_check_syncookie_ipv4, 7551 .gpl_only = true, /* __cookie_v4_check is GPL */ 7552 .pkt_access = true, 7553 .ret_type = RET_INTEGER, 7554 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7555 .arg1_size = sizeof(struct iphdr), 7556 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7557 .arg2_size = sizeof(struct tcphdr), 7558 }; 7559 7560 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph, 7561 struct tcphdr *, th) 7562 { 7563 #if IS_BUILTIN(CONFIG_IPV6) 7564 u32 cookie = ntohl(th->ack_seq) - 1; 7565 7566 if (__cookie_v6_check(iph, th, cookie) > 0) 7567 return 0; 7568 7569 return -EACCES; 7570 #else 7571 return -EPROTONOSUPPORT; 7572 #endif 7573 } 7574 7575 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = { 7576 .func = bpf_tcp_raw_check_syncookie_ipv6, 7577 .gpl_only = true, /* __cookie_v6_check is GPL */ 7578 .pkt_access = true, 7579 .ret_type = RET_INTEGER, 7580 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7581 .arg1_size = sizeof(struct ipv6hdr), 7582 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7583 .arg2_size = sizeof(struct tcphdr), 7584 }; 7585 #endif /* CONFIG_SYN_COOKIES */ 7586 7587 #endif /* CONFIG_INET */ 7588 7589 bool bpf_helper_changes_pkt_data(void *func) 7590 { 7591 if (func == bpf_skb_vlan_push || 7592 func == bpf_skb_vlan_pop || 7593 func == bpf_skb_store_bytes || 7594 func == bpf_skb_change_proto || 7595 func == bpf_skb_change_head || 7596 func == sk_skb_change_head || 7597 func == bpf_skb_change_tail || 7598 func == sk_skb_change_tail || 7599 func == bpf_skb_adjust_room || 7600 func == sk_skb_adjust_room || 7601 func == bpf_skb_pull_data || 7602 func == sk_skb_pull_data || 7603 func == bpf_clone_redirect || 7604 func == bpf_l3_csum_replace || 7605 func == bpf_l4_csum_replace || 7606 func == bpf_xdp_adjust_head || 7607 func == bpf_xdp_adjust_meta || 7608 func == bpf_msg_pull_data || 7609 func == bpf_msg_push_data || 7610 func == bpf_msg_pop_data || 7611 func == bpf_xdp_adjust_tail || 7612 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7613 func == bpf_lwt_seg6_store_bytes || 7614 func == bpf_lwt_seg6_adjust_srh || 7615 func == bpf_lwt_seg6_action || 7616 #endif 7617 #ifdef CONFIG_INET 7618 func == bpf_sock_ops_store_hdr_opt || 7619 #endif 7620 func == bpf_lwt_in_push_encap || 7621 func == bpf_lwt_xmit_push_encap) 7622 return true; 7623 7624 return false; 7625 } 7626 7627 const struct bpf_func_proto bpf_event_output_data_proto __weak; 7628 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 7629 7630 static const struct bpf_func_proto * 7631 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7632 { 7633 const struct bpf_func_proto *func_proto; 7634 7635 func_proto = cgroup_common_func_proto(func_id, prog); 7636 if (func_proto) 7637 return func_proto; 7638 7639 func_proto = cgroup_current_func_proto(func_id, prog); 7640 if (func_proto) 7641 return func_proto; 7642 7643 switch (func_id) { 7644 case BPF_FUNC_get_socket_cookie: 7645 return &bpf_get_socket_cookie_sock_proto; 7646 case BPF_FUNC_get_netns_cookie: 7647 return &bpf_get_netns_cookie_sock_proto; 7648 case BPF_FUNC_perf_event_output: 7649 return &bpf_event_output_data_proto; 7650 case BPF_FUNC_sk_storage_get: 7651 return &bpf_sk_storage_get_cg_sock_proto; 7652 case BPF_FUNC_ktime_get_coarse_ns: 7653 return &bpf_ktime_get_coarse_ns_proto; 7654 default: 7655 return bpf_base_func_proto(func_id); 7656 } 7657 } 7658 7659 static const struct bpf_func_proto * 7660 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7661 { 7662 const struct bpf_func_proto *func_proto; 7663 7664 func_proto = cgroup_common_func_proto(func_id, prog); 7665 if (func_proto) 7666 return func_proto; 7667 7668 func_proto = cgroup_current_func_proto(func_id, prog); 7669 if (func_proto) 7670 return func_proto; 7671 7672 switch (func_id) { 7673 case BPF_FUNC_bind: 7674 switch (prog->expected_attach_type) { 7675 case BPF_CGROUP_INET4_CONNECT: 7676 case BPF_CGROUP_INET6_CONNECT: 7677 return &bpf_bind_proto; 7678 default: 7679 return NULL; 7680 } 7681 case BPF_FUNC_get_socket_cookie: 7682 return &bpf_get_socket_cookie_sock_addr_proto; 7683 case BPF_FUNC_get_netns_cookie: 7684 return &bpf_get_netns_cookie_sock_addr_proto; 7685 case BPF_FUNC_perf_event_output: 7686 return &bpf_event_output_data_proto; 7687 #ifdef CONFIG_INET 7688 case BPF_FUNC_sk_lookup_tcp: 7689 return &bpf_sock_addr_sk_lookup_tcp_proto; 7690 case BPF_FUNC_sk_lookup_udp: 7691 return &bpf_sock_addr_sk_lookup_udp_proto; 7692 case BPF_FUNC_sk_release: 7693 return &bpf_sk_release_proto; 7694 case BPF_FUNC_skc_lookup_tcp: 7695 return &bpf_sock_addr_skc_lookup_tcp_proto; 7696 #endif /* CONFIG_INET */ 7697 case BPF_FUNC_sk_storage_get: 7698 return &bpf_sk_storage_get_proto; 7699 case BPF_FUNC_sk_storage_delete: 7700 return &bpf_sk_storage_delete_proto; 7701 case BPF_FUNC_setsockopt: 7702 switch (prog->expected_attach_type) { 7703 case BPF_CGROUP_INET4_BIND: 7704 case BPF_CGROUP_INET6_BIND: 7705 case BPF_CGROUP_INET4_CONNECT: 7706 case BPF_CGROUP_INET6_CONNECT: 7707 case BPF_CGROUP_UDP4_RECVMSG: 7708 case BPF_CGROUP_UDP6_RECVMSG: 7709 case BPF_CGROUP_UDP4_SENDMSG: 7710 case BPF_CGROUP_UDP6_SENDMSG: 7711 case BPF_CGROUP_INET4_GETPEERNAME: 7712 case BPF_CGROUP_INET6_GETPEERNAME: 7713 case BPF_CGROUP_INET4_GETSOCKNAME: 7714 case BPF_CGROUP_INET6_GETSOCKNAME: 7715 return &bpf_sock_addr_setsockopt_proto; 7716 default: 7717 return NULL; 7718 } 7719 case BPF_FUNC_getsockopt: 7720 switch (prog->expected_attach_type) { 7721 case BPF_CGROUP_INET4_BIND: 7722 case BPF_CGROUP_INET6_BIND: 7723 case BPF_CGROUP_INET4_CONNECT: 7724 case BPF_CGROUP_INET6_CONNECT: 7725 case BPF_CGROUP_UDP4_RECVMSG: 7726 case BPF_CGROUP_UDP6_RECVMSG: 7727 case BPF_CGROUP_UDP4_SENDMSG: 7728 case BPF_CGROUP_UDP6_SENDMSG: 7729 case BPF_CGROUP_INET4_GETPEERNAME: 7730 case BPF_CGROUP_INET6_GETPEERNAME: 7731 case BPF_CGROUP_INET4_GETSOCKNAME: 7732 case BPF_CGROUP_INET6_GETSOCKNAME: 7733 return &bpf_sock_addr_getsockopt_proto; 7734 default: 7735 return NULL; 7736 } 7737 default: 7738 return bpf_sk_base_func_proto(func_id); 7739 } 7740 } 7741 7742 static const struct bpf_func_proto * 7743 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7744 { 7745 switch (func_id) { 7746 case BPF_FUNC_skb_load_bytes: 7747 return &bpf_skb_load_bytes_proto; 7748 case BPF_FUNC_skb_load_bytes_relative: 7749 return &bpf_skb_load_bytes_relative_proto; 7750 case BPF_FUNC_get_socket_cookie: 7751 return &bpf_get_socket_cookie_proto; 7752 case BPF_FUNC_get_socket_uid: 7753 return &bpf_get_socket_uid_proto; 7754 case BPF_FUNC_perf_event_output: 7755 return &bpf_skb_event_output_proto; 7756 default: 7757 return bpf_sk_base_func_proto(func_id); 7758 } 7759 } 7760 7761 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7762 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7763 7764 static const struct bpf_func_proto * 7765 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7766 { 7767 const struct bpf_func_proto *func_proto; 7768 7769 func_proto = cgroup_common_func_proto(func_id, prog); 7770 if (func_proto) 7771 return func_proto; 7772 7773 switch (func_id) { 7774 case BPF_FUNC_sk_fullsock: 7775 return &bpf_sk_fullsock_proto; 7776 case BPF_FUNC_sk_storage_get: 7777 return &bpf_sk_storage_get_proto; 7778 case BPF_FUNC_sk_storage_delete: 7779 return &bpf_sk_storage_delete_proto; 7780 case BPF_FUNC_perf_event_output: 7781 return &bpf_skb_event_output_proto; 7782 #ifdef CONFIG_SOCK_CGROUP_DATA 7783 case BPF_FUNC_skb_cgroup_id: 7784 return &bpf_skb_cgroup_id_proto; 7785 case BPF_FUNC_skb_ancestor_cgroup_id: 7786 return &bpf_skb_ancestor_cgroup_id_proto; 7787 case BPF_FUNC_sk_cgroup_id: 7788 return &bpf_sk_cgroup_id_proto; 7789 case BPF_FUNC_sk_ancestor_cgroup_id: 7790 return &bpf_sk_ancestor_cgroup_id_proto; 7791 #endif 7792 #ifdef CONFIG_INET 7793 case BPF_FUNC_sk_lookup_tcp: 7794 return &bpf_sk_lookup_tcp_proto; 7795 case BPF_FUNC_sk_lookup_udp: 7796 return &bpf_sk_lookup_udp_proto; 7797 case BPF_FUNC_sk_release: 7798 return &bpf_sk_release_proto; 7799 case BPF_FUNC_skc_lookup_tcp: 7800 return &bpf_skc_lookup_tcp_proto; 7801 case BPF_FUNC_tcp_sock: 7802 return &bpf_tcp_sock_proto; 7803 case BPF_FUNC_get_listener_sock: 7804 return &bpf_get_listener_sock_proto; 7805 case BPF_FUNC_skb_ecn_set_ce: 7806 return &bpf_skb_ecn_set_ce_proto; 7807 #endif 7808 default: 7809 return sk_filter_func_proto(func_id, prog); 7810 } 7811 } 7812 7813 static const struct bpf_func_proto * 7814 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7815 { 7816 switch (func_id) { 7817 case BPF_FUNC_skb_store_bytes: 7818 return &bpf_skb_store_bytes_proto; 7819 case BPF_FUNC_skb_load_bytes: 7820 return &bpf_skb_load_bytes_proto; 7821 case BPF_FUNC_skb_load_bytes_relative: 7822 return &bpf_skb_load_bytes_relative_proto; 7823 case BPF_FUNC_skb_pull_data: 7824 return &bpf_skb_pull_data_proto; 7825 case BPF_FUNC_csum_diff: 7826 return &bpf_csum_diff_proto; 7827 case BPF_FUNC_csum_update: 7828 return &bpf_csum_update_proto; 7829 case BPF_FUNC_csum_level: 7830 return &bpf_csum_level_proto; 7831 case BPF_FUNC_l3_csum_replace: 7832 return &bpf_l3_csum_replace_proto; 7833 case BPF_FUNC_l4_csum_replace: 7834 return &bpf_l4_csum_replace_proto; 7835 case BPF_FUNC_clone_redirect: 7836 return &bpf_clone_redirect_proto; 7837 case BPF_FUNC_get_cgroup_classid: 7838 return &bpf_get_cgroup_classid_proto; 7839 case BPF_FUNC_skb_vlan_push: 7840 return &bpf_skb_vlan_push_proto; 7841 case BPF_FUNC_skb_vlan_pop: 7842 return &bpf_skb_vlan_pop_proto; 7843 case BPF_FUNC_skb_change_proto: 7844 return &bpf_skb_change_proto_proto; 7845 case BPF_FUNC_skb_change_type: 7846 return &bpf_skb_change_type_proto; 7847 case BPF_FUNC_skb_adjust_room: 7848 return &bpf_skb_adjust_room_proto; 7849 case BPF_FUNC_skb_change_tail: 7850 return &bpf_skb_change_tail_proto; 7851 case BPF_FUNC_skb_change_head: 7852 return &bpf_skb_change_head_proto; 7853 case BPF_FUNC_skb_get_tunnel_key: 7854 return &bpf_skb_get_tunnel_key_proto; 7855 case BPF_FUNC_skb_set_tunnel_key: 7856 return bpf_get_skb_set_tunnel_proto(func_id); 7857 case BPF_FUNC_skb_get_tunnel_opt: 7858 return &bpf_skb_get_tunnel_opt_proto; 7859 case BPF_FUNC_skb_set_tunnel_opt: 7860 return bpf_get_skb_set_tunnel_proto(func_id); 7861 case BPF_FUNC_redirect: 7862 return &bpf_redirect_proto; 7863 case BPF_FUNC_redirect_neigh: 7864 return &bpf_redirect_neigh_proto; 7865 case BPF_FUNC_redirect_peer: 7866 return &bpf_redirect_peer_proto; 7867 case BPF_FUNC_get_route_realm: 7868 return &bpf_get_route_realm_proto; 7869 case BPF_FUNC_get_hash_recalc: 7870 return &bpf_get_hash_recalc_proto; 7871 case BPF_FUNC_set_hash_invalid: 7872 return &bpf_set_hash_invalid_proto; 7873 case BPF_FUNC_set_hash: 7874 return &bpf_set_hash_proto; 7875 case BPF_FUNC_perf_event_output: 7876 return &bpf_skb_event_output_proto; 7877 case BPF_FUNC_get_smp_processor_id: 7878 return &bpf_get_smp_processor_id_proto; 7879 case BPF_FUNC_skb_under_cgroup: 7880 return &bpf_skb_under_cgroup_proto; 7881 case BPF_FUNC_get_socket_cookie: 7882 return &bpf_get_socket_cookie_proto; 7883 case BPF_FUNC_get_socket_uid: 7884 return &bpf_get_socket_uid_proto; 7885 case BPF_FUNC_fib_lookup: 7886 return &bpf_skb_fib_lookup_proto; 7887 case BPF_FUNC_check_mtu: 7888 return &bpf_skb_check_mtu_proto; 7889 case BPF_FUNC_sk_fullsock: 7890 return &bpf_sk_fullsock_proto; 7891 case BPF_FUNC_sk_storage_get: 7892 return &bpf_sk_storage_get_proto; 7893 case BPF_FUNC_sk_storage_delete: 7894 return &bpf_sk_storage_delete_proto; 7895 #ifdef CONFIG_XFRM 7896 case BPF_FUNC_skb_get_xfrm_state: 7897 return &bpf_skb_get_xfrm_state_proto; 7898 #endif 7899 #ifdef CONFIG_CGROUP_NET_CLASSID 7900 case BPF_FUNC_skb_cgroup_classid: 7901 return &bpf_skb_cgroup_classid_proto; 7902 #endif 7903 #ifdef CONFIG_SOCK_CGROUP_DATA 7904 case BPF_FUNC_skb_cgroup_id: 7905 return &bpf_skb_cgroup_id_proto; 7906 case BPF_FUNC_skb_ancestor_cgroup_id: 7907 return &bpf_skb_ancestor_cgroup_id_proto; 7908 #endif 7909 #ifdef CONFIG_INET 7910 case BPF_FUNC_sk_lookup_tcp: 7911 return &bpf_sk_lookup_tcp_proto; 7912 case BPF_FUNC_sk_lookup_udp: 7913 return &bpf_sk_lookup_udp_proto; 7914 case BPF_FUNC_sk_release: 7915 return &bpf_sk_release_proto; 7916 case BPF_FUNC_tcp_sock: 7917 return &bpf_tcp_sock_proto; 7918 case BPF_FUNC_get_listener_sock: 7919 return &bpf_get_listener_sock_proto; 7920 case BPF_FUNC_skc_lookup_tcp: 7921 return &bpf_skc_lookup_tcp_proto; 7922 case BPF_FUNC_tcp_check_syncookie: 7923 return &bpf_tcp_check_syncookie_proto; 7924 case BPF_FUNC_skb_ecn_set_ce: 7925 return &bpf_skb_ecn_set_ce_proto; 7926 case BPF_FUNC_tcp_gen_syncookie: 7927 return &bpf_tcp_gen_syncookie_proto; 7928 case BPF_FUNC_sk_assign: 7929 return &bpf_sk_assign_proto; 7930 case BPF_FUNC_skb_set_tstamp: 7931 return &bpf_skb_set_tstamp_proto; 7932 #ifdef CONFIG_SYN_COOKIES 7933 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 7934 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 7935 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 7936 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 7937 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 7938 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 7939 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 7940 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 7941 #endif 7942 #endif 7943 default: 7944 return bpf_sk_base_func_proto(func_id); 7945 } 7946 } 7947 7948 static const struct bpf_func_proto * 7949 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7950 { 7951 switch (func_id) { 7952 case BPF_FUNC_perf_event_output: 7953 return &bpf_xdp_event_output_proto; 7954 case BPF_FUNC_get_smp_processor_id: 7955 return &bpf_get_smp_processor_id_proto; 7956 case BPF_FUNC_csum_diff: 7957 return &bpf_csum_diff_proto; 7958 case BPF_FUNC_xdp_adjust_head: 7959 return &bpf_xdp_adjust_head_proto; 7960 case BPF_FUNC_xdp_adjust_meta: 7961 return &bpf_xdp_adjust_meta_proto; 7962 case BPF_FUNC_redirect: 7963 return &bpf_xdp_redirect_proto; 7964 case BPF_FUNC_redirect_map: 7965 return &bpf_xdp_redirect_map_proto; 7966 case BPF_FUNC_xdp_adjust_tail: 7967 return &bpf_xdp_adjust_tail_proto; 7968 case BPF_FUNC_xdp_get_buff_len: 7969 return &bpf_xdp_get_buff_len_proto; 7970 case BPF_FUNC_xdp_load_bytes: 7971 return &bpf_xdp_load_bytes_proto; 7972 case BPF_FUNC_xdp_store_bytes: 7973 return &bpf_xdp_store_bytes_proto; 7974 case BPF_FUNC_fib_lookup: 7975 return &bpf_xdp_fib_lookup_proto; 7976 case BPF_FUNC_check_mtu: 7977 return &bpf_xdp_check_mtu_proto; 7978 #ifdef CONFIG_INET 7979 case BPF_FUNC_sk_lookup_udp: 7980 return &bpf_xdp_sk_lookup_udp_proto; 7981 case BPF_FUNC_sk_lookup_tcp: 7982 return &bpf_xdp_sk_lookup_tcp_proto; 7983 case BPF_FUNC_sk_release: 7984 return &bpf_sk_release_proto; 7985 case BPF_FUNC_skc_lookup_tcp: 7986 return &bpf_xdp_skc_lookup_tcp_proto; 7987 case BPF_FUNC_tcp_check_syncookie: 7988 return &bpf_tcp_check_syncookie_proto; 7989 case BPF_FUNC_tcp_gen_syncookie: 7990 return &bpf_tcp_gen_syncookie_proto; 7991 #ifdef CONFIG_SYN_COOKIES 7992 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 7993 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 7994 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 7995 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 7996 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 7997 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 7998 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 7999 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 8000 #endif 8001 #endif 8002 default: 8003 return bpf_sk_base_func_proto(func_id); 8004 } 8005 8006 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES) 8007 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The 8008 * kfuncs are defined in two different modules, and we want to be able 8009 * to use them interchangably with the same BTF type ID. Because modules 8010 * can't de-duplicate BTF IDs between each other, we need the type to be 8011 * referenced in the vmlinux BTF or the verifier will get confused about 8012 * the different types. So we add this dummy type reference which will 8013 * be included in vmlinux BTF, allowing both modules to refer to the 8014 * same type ID. 8015 */ 8016 BTF_TYPE_EMIT(struct nf_conn___init); 8017 #endif 8018 } 8019 8020 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 8021 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 8022 8023 static const struct bpf_func_proto * 8024 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8025 { 8026 const struct bpf_func_proto *func_proto; 8027 8028 func_proto = cgroup_common_func_proto(func_id, prog); 8029 if (func_proto) 8030 return func_proto; 8031 8032 switch (func_id) { 8033 case BPF_FUNC_setsockopt: 8034 return &bpf_sock_ops_setsockopt_proto; 8035 case BPF_FUNC_getsockopt: 8036 return &bpf_sock_ops_getsockopt_proto; 8037 case BPF_FUNC_sock_ops_cb_flags_set: 8038 return &bpf_sock_ops_cb_flags_set_proto; 8039 case BPF_FUNC_sock_map_update: 8040 return &bpf_sock_map_update_proto; 8041 case BPF_FUNC_sock_hash_update: 8042 return &bpf_sock_hash_update_proto; 8043 case BPF_FUNC_get_socket_cookie: 8044 return &bpf_get_socket_cookie_sock_ops_proto; 8045 case BPF_FUNC_perf_event_output: 8046 return &bpf_event_output_data_proto; 8047 case BPF_FUNC_sk_storage_get: 8048 return &bpf_sk_storage_get_proto; 8049 case BPF_FUNC_sk_storage_delete: 8050 return &bpf_sk_storage_delete_proto; 8051 case BPF_FUNC_get_netns_cookie: 8052 return &bpf_get_netns_cookie_sock_ops_proto; 8053 #ifdef CONFIG_INET 8054 case BPF_FUNC_load_hdr_opt: 8055 return &bpf_sock_ops_load_hdr_opt_proto; 8056 case BPF_FUNC_store_hdr_opt: 8057 return &bpf_sock_ops_store_hdr_opt_proto; 8058 case BPF_FUNC_reserve_hdr_opt: 8059 return &bpf_sock_ops_reserve_hdr_opt_proto; 8060 case BPF_FUNC_tcp_sock: 8061 return &bpf_tcp_sock_proto; 8062 #endif /* CONFIG_INET */ 8063 default: 8064 return bpf_sk_base_func_proto(func_id); 8065 } 8066 } 8067 8068 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 8069 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 8070 8071 static const struct bpf_func_proto * 8072 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8073 { 8074 switch (func_id) { 8075 case BPF_FUNC_msg_redirect_map: 8076 return &bpf_msg_redirect_map_proto; 8077 case BPF_FUNC_msg_redirect_hash: 8078 return &bpf_msg_redirect_hash_proto; 8079 case BPF_FUNC_msg_apply_bytes: 8080 return &bpf_msg_apply_bytes_proto; 8081 case BPF_FUNC_msg_cork_bytes: 8082 return &bpf_msg_cork_bytes_proto; 8083 case BPF_FUNC_msg_pull_data: 8084 return &bpf_msg_pull_data_proto; 8085 case BPF_FUNC_msg_push_data: 8086 return &bpf_msg_push_data_proto; 8087 case BPF_FUNC_msg_pop_data: 8088 return &bpf_msg_pop_data_proto; 8089 case BPF_FUNC_perf_event_output: 8090 return &bpf_event_output_data_proto; 8091 case BPF_FUNC_get_current_uid_gid: 8092 return &bpf_get_current_uid_gid_proto; 8093 case BPF_FUNC_get_current_pid_tgid: 8094 return &bpf_get_current_pid_tgid_proto; 8095 case BPF_FUNC_sk_storage_get: 8096 return &bpf_sk_storage_get_proto; 8097 case BPF_FUNC_sk_storage_delete: 8098 return &bpf_sk_storage_delete_proto; 8099 case BPF_FUNC_get_netns_cookie: 8100 return &bpf_get_netns_cookie_sk_msg_proto; 8101 #ifdef CONFIG_CGROUPS 8102 case BPF_FUNC_get_current_cgroup_id: 8103 return &bpf_get_current_cgroup_id_proto; 8104 case BPF_FUNC_get_current_ancestor_cgroup_id: 8105 return &bpf_get_current_ancestor_cgroup_id_proto; 8106 #endif 8107 #ifdef CONFIG_CGROUP_NET_CLASSID 8108 case BPF_FUNC_get_cgroup_classid: 8109 return &bpf_get_cgroup_classid_curr_proto; 8110 #endif 8111 default: 8112 return bpf_sk_base_func_proto(func_id); 8113 } 8114 } 8115 8116 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 8117 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 8118 8119 static const struct bpf_func_proto * 8120 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8121 { 8122 switch (func_id) { 8123 case BPF_FUNC_skb_store_bytes: 8124 return &bpf_skb_store_bytes_proto; 8125 case BPF_FUNC_skb_load_bytes: 8126 return &bpf_skb_load_bytes_proto; 8127 case BPF_FUNC_skb_pull_data: 8128 return &sk_skb_pull_data_proto; 8129 case BPF_FUNC_skb_change_tail: 8130 return &sk_skb_change_tail_proto; 8131 case BPF_FUNC_skb_change_head: 8132 return &sk_skb_change_head_proto; 8133 case BPF_FUNC_skb_adjust_room: 8134 return &sk_skb_adjust_room_proto; 8135 case BPF_FUNC_get_socket_cookie: 8136 return &bpf_get_socket_cookie_proto; 8137 case BPF_FUNC_get_socket_uid: 8138 return &bpf_get_socket_uid_proto; 8139 case BPF_FUNC_sk_redirect_map: 8140 return &bpf_sk_redirect_map_proto; 8141 case BPF_FUNC_sk_redirect_hash: 8142 return &bpf_sk_redirect_hash_proto; 8143 case BPF_FUNC_perf_event_output: 8144 return &bpf_skb_event_output_proto; 8145 #ifdef CONFIG_INET 8146 case BPF_FUNC_sk_lookup_tcp: 8147 return &bpf_sk_lookup_tcp_proto; 8148 case BPF_FUNC_sk_lookup_udp: 8149 return &bpf_sk_lookup_udp_proto; 8150 case BPF_FUNC_sk_release: 8151 return &bpf_sk_release_proto; 8152 case BPF_FUNC_skc_lookup_tcp: 8153 return &bpf_skc_lookup_tcp_proto; 8154 #endif 8155 default: 8156 return bpf_sk_base_func_proto(func_id); 8157 } 8158 } 8159 8160 static const struct bpf_func_proto * 8161 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8162 { 8163 switch (func_id) { 8164 case BPF_FUNC_skb_load_bytes: 8165 return &bpf_flow_dissector_load_bytes_proto; 8166 default: 8167 return bpf_sk_base_func_proto(func_id); 8168 } 8169 } 8170 8171 static const struct bpf_func_proto * 8172 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8173 { 8174 switch (func_id) { 8175 case BPF_FUNC_skb_load_bytes: 8176 return &bpf_skb_load_bytes_proto; 8177 case BPF_FUNC_skb_pull_data: 8178 return &bpf_skb_pull_data_proto; 8179 case BPF_FUNC_csum_diff: 8180 return &bpf_csum_diff_proto; 8181 case BPF_FUNC_get_cgroup_classid: 8182 return &bpf_get_cgroup_classid_proto; 8183 case BPF_FUNC_get_route_realm: 8184 return &bpf_get_route_realm_proto; 8185 case BPF_FUNC_get_hash_recalc: 8186 return &bpf_get_hash_recalc_proto; 8187 case BPF_FUNC_perf_event_output: 8188 return &bpf_skb_event_output_proto; 8189 case BPF_FUNC_get_smp_processor_id: 8190 return &bpf_get_smp_processor_id_proto; 8191 case BPF_FUNC_skb_under_cgroup: 8192 return &bpf_skb_under_cgroup_proto; 8193 default: 8194 return bpf_sk_base_func_proto(func_id); 8195 } 8196 } 8197 8198 static const struct bpf_func_proto * 8199 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8200 { 8201 switch (func_id) { 8202 case BPF_FUNC_lwt_push_encap: 8203 return &bpf_lwt_in_push_encap_proto; 8204 default: 8205 return lwt_out_func_proto(func_id, prog); 8206 } 8207 } 8208 8209 static const struct bpf_func_proto * 8210 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8211 { 8212 switch (func_id) { 8213 case BPF_FUNC_skb_get_tunnel_key: 8214 return &bpf_skb_get_tunnel_key_proto; 8215 case BPF_FUNC_skb_set_tunnel_key: 8216 return bpf_get_skb_set_tunnel_proto(func_id); 8217 case BPF_FUNC_skb_get_tunnel_opt: 8218 return &bpf_skb_get_tunnel_opt_proto; 8219 case BPF_FUNC_skb_set_tunnel_opt: 8220 return bpf_get_skb_set_tunnel_proto(func_id); 8221 case BPF_FUNC_redirect: 8222 return &bpf_redirect_proto; 8223 case BPF_FUNC_clone_redirect: 8224 return &bpf_clone_redirect_proto; 8225 case BPF_FUNC_skb_change_tail: 8226 return &bpf_skb_change_tail_proto; 8227 case BPF_FUNC_skb_change_head: 8228 return &bpf_skb_change_head_proto; 8229 case BPF_FUNC_skb_store_bytes: 8230 return &bpf_skb_store_bytes_proto; 8231 case BPF_FUNC_csum_update: 8232 return &bpf_csum_update_proto; 8233 case BPF_FUNC_csum_level: 8234 return &bpf_csum_level_proto; 8235 case BPF_FUNC_l3_csum_replace: 8236 return &bpf_l3_csum_replace_proto; 8237 case BPF_FUNC_l4_csum_replace: 8238 return &bpf_l4_csum_replace_proto; 8239 case BPF_FUNC_set_hash_invalid: 8240 return &bpf_set_hash_invalid_proto; 8241 case BPF_FUNC_lwt_push_encap: 8242 return &bpf_lwt_xmit_push_encap_proto; 8243 default: 8244 return lwt_out_func_proto(func_id, prog); 8245 } 8246 } 8247 8248 static const struct bpf_func_proto * 8249 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8250 { 8251 switch (func_id) { 8252 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 8253 case BPF_FUNC_lwt_seg6_store_bytes: 8254 return &bpf_lwt_seg6_store_bytes_proto; 8255 case BPF_FUNC_lwt_seg6_action: 8256 return &bpf_lwt_seg6_action_proto; 8257 case BPF_FUNC_lwt_seg6_adjust_srh: 8258 return &bpf_lwt_seg6_adjust_srh_proto; 8259 #endif 8260 default: 8261 return lwt_out_func_proto(func_id, prog); 8262 } 8263 } 8264 8265 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 8266 const struct bpf_prog *prog, 8267 struct bpf_insn_access_aux *info) 8268 { 8269 const int size_default = sizeof(__u32); 8270 8271 if (off < 0 || off >= sizeof(struct __sk_buff)) 8272 return false; 8273 8274 /* The verifier guarantees that size > 0. */ 8275 if (off % size != 0) 8276 return false; 8277 8278 switch (off) { 8279 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8280 if (off + size > offsetofend(struct __sk_buff, cb[4])) 8281 return false; 8282 break; 8283 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 8284 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 8285 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 8286 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 8287 case bpf_ctx_range(struct __sk_buff, data): 8288 case bpf_ctx_range(struct __sk_buff, data_meta): 8289 case bpf_ctx_range(struct __sk_buff, data_end): 8290 if (size != size_default) 8291 return false; 8292 break; 8293 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8294 return false; 8295 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8296 if (type == BPF_WRITE || size != sizeof(__u64)) 8297 return false; 8298 break; 8299 case bpf_ctx_range(struct __sk_buff, tstamp): 8300 if (size != sizeof(__u64)) 8301 return false; 8302 break; 8303 case offsetof(struct __sk_buff, sk): 8304 if (type == BPF_WRITE || size != sizeof(__u64)) 8305 return false; 8306 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 8307 break; 8308 case offsetof(struct __sk_buff, tstamp_type): 8309 return false; 8310 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1: 8311 /* Explicitly prohibit access to padding in __sk_buff. */ 8312 return false; 8313 default: 8314 /* Only narrow read access allowed for now. */ 8315 if (type == BPF_WRITE) { 8316 if (size != size_default) 8317 return false; 8318 } else { 8319 bpf_ctx_record_field_size(info, size_default); 8320 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8321 return false; 8322 } 8323 } 8324 8325 return true; 8326 } 8327 8328 static bool sk_filter_is_valid_access(int off, int size, 8329 enum bpf_access_type type, 8330 const struct bpf_prog *prog, 8331 struct bpf_insn_access_aux *info) 8332 { 8333 switch (off) { 8334 case bpf_ctx_range(struct __sk_buff, tc_classid): 8335 case bpf_ctx_range(struct __sk_buff, data): 8336 case bpf_ctx_range(struct __sk_buff, data_meta): 8337 case bpf_ctx_range(struct __sk_buff, data_end): 8338 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8339 case bpf_ctx_range(struct __sk_buff, tstamp): 8340 case bpf_ctx_range(struct __sk_buff, wire_len): 8341 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8342 return false; 8343 } 8344 8345 if (type == BPF_WRITE) { 8346 switch (off) { 8347 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8348 break; 8349 default: 8350 return false; 8351 } 8352 } 8353 8354 return bpf_skb_is_valid_access(off, size, type, prog, info); 8355 } 8356 8357 static bool cg_skb_is_valid_access(int off, int size, 8358 enum bpf_access_type type, 8359 const struct bpf_prog *prog, 8360 struct bpf_insn_access_aux *info) 8361 { 8362 switch (off) { 8363 case bpf_ctx_range(struct __sk_buff, tc_classid): 8364 case bpf_ctx_range(struct __sk_buff, data_meta): 8365 case bpf_ctx_range(struct __sk_buff, wire_len): 8366 return false; 8367 case bpf_ctx_range(struct __sk_buff, data): 8368 case bpf_ctx_range(struct __sk_buff, data_end): 8369 if (!bpf_capable()) 8370 return false; 8371 break; 8372 } 8373 8374 if (type == BPF_WRITE) { 8375 switch (off) { 8376 case bpf_ctx_range(struct __sk_buff, mark): 8377 case bpf_ctx_range(struct __sk_buff, priority): 8378 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8379 break; 8380 case bpf_ctx_range(struct __sk_buff, tstamp): 8381 if (!bpf_capable()) 8382 return false; 8383 break; 8384 default: 8385 return false; 8386 } 8387 } 8388 8389 switch (off) { 8390 case bpf_ctx_range(struct __sk_buff, data): 8391 info->reg_type = PTR_TO_PACKET; 8392 break; 8393 case bpf_ctx_range(struct __sk_buff, data_end): 8394 info->reg_type = PTR_TO_PACKET_END; 8395 break; 8396 } 8397 8398 return bpf_skb_is_valid_access(off, size, type, prog, info); 8399 } 8400 8401 static bool lwt_is_valid_access(int off, int size, 8402 enum bpf_access_type type, 8403 const struct bpf_prog *prog, 8404 struct bpf_insn_access_aux *info) 8405 { 8406 switch (off) { 8407 case bpf_ctx_range(struct __sk_buff, tc_classid): 8408 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8409 case bpf_ctx_range(struct __sk_buff, data_meta): 8410 case bpf_ctx_range(struct __sk_buff, tstamp): 8411 case bpf_ctx_range(struct __sk_buff, wire_len): 8412 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8413 return false; 8414 } 8415 8416 if (type == BPF_WRITE) { 8417 switch (off) { 8418 case bpf_ctx_range(struct __sk_buff, mark): 8419 case bpf_ctx_range(struct __sk_buff, priority): 8420 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8421 break; 8422 default: 8423 return false; 8424 } 8425 } 8426 8427 switch (off) { 8428 case bpf_ctx_range(struct __sk_buff, data): 8429 info->reg_type = PTR_TO_PACKET; 8430 break; 8431 case bpf_ctx_range(struct __sk_buff, data_end): 8432 info->reg_type = PTR_TO_PACKET_END; 8433 break; 8434 } 8435 8436 return bpf_skb_is_valid_access(off, size, type, prog, info); 8437 } 8438 8439 /* Attach type specific accesses */ 8440 static bool __sock_filter_check_attach_type(int off, 8441 enum bpf_access_type access_type, 8442 enum bpf_attach_type attach_type) 8443 { 8444 switch (off) { 8445 case offsetof(struct bpf_sock, bound_dev_if): 8446 case offsetof(struct bpf_sock, mark): 8447 case offsetof(struct bpf_sock, priority): 8448 switch (attach_type) { 8449 case BPF_CGROUP_INET_SOCK_CREATE: 8450 case BPF_CGROUP_INET_SOCK_RELEASE: 8451 goto full_access; 8452 default: 8453 return false; 8454 } 8455 case bpf_ctx_range(struct bpf_sock, src_ip4): 8456 switch (attach_type) { 8457 case BPF_CGROUP_INET4_POST_BIND: 8458 goto read_only; 8459 default: 8460 return false; 8461 } 8462 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8463 switch (attach_type) { 8464 case BPF_CGROUP_INET6_POST_BIND: 8465 goto read_only; 8466 default: 8467 return false; 8468 } 8469 case bpf_ctx_range(struct bpf_sock, src_port): 8470 switch (attach_type) { 8471 case BPF_CGROUP_INET4_POST_BIND: 8472 case BPF_CGROUP_INET6_POST_BIND: 8473 goto read_only; 8474 default: 8475 return false; 8476 } 8477 } 8478 read_only: 8479 return access_type == BPF_READ; 8480 full_access: 8481 return true; 8482 } 8483 8484 bool bpf_sock_common_is_valid_access(int off, int size, 8485 enum bpf_access_type type, 8486 struct bpf_insn_access_aux *info) 8487 { 8488 switch (off) { 8489 case bpf_ctx_range_till(struct bpf_sock, type, priority): 8490 return false; 8491 default: 8492 return bpf_sock_is_valid_access(off, size, type, info); 8493 } 8494 } 8495 8496 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 8497 struct bpf_insn_access_aux *info) 8498 { 8499 const int size_default = sizeof(__u32); 8500 int field_size; 8501 8502 if (off < 0 || off >= sizeof(struct bpf_sock)) 8503 return false; 8504 if (off % size != 0) 8505 return false; 8506 8507 switch (off) { 8508 case offsetof(struct bpf_sock, state): 8509 case offsetof(struct bpf_sock, family): 8510 case offsetof(struct bpf_sock, type): 8511 case offsetof(struct bpf_sock, protocol): 8512 case offsetof(struct bpf_sock, src_port): 8513 case offsetof(struct bpf_sock, rx_queue_mapping): 8514 case bpf_ctx_range(struct bpf_sock, src_ip4): 8515 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8516 case bpf_ctx_range(struct bpf_sock, dst_ip4): 8517 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 8518 bpf_ctx_record_field_size(info, size_default); 8519 return bpf_ctx_narrow_access_ok(off, size, size_default); 8520 case bpf_ctx_range(struct bpf_sock, dst_port): 8521 field_size = size == size_default ? 8522 size_default : sizeof_field(struct bpf_sock, dst_port); 8523 bpf_ctx_record_field_size(info, field_size); 8524 return bpf_ctx_narrow_access_ok(off, size, field_size); 8525 case offsetofend(struct bpf_sock, dst_port) ... 8526 offsetof(struct bpf_sock, dst_ip4) - 1: 8527 return false; 8528 } 8529 8530 return size == size_default; 8531 } 8532 8533 static bool sock_filter_is_valid_access(int off, int size, 8534 enum bpf_access_type type, 8535 const struct bpf_prog *prog, 8536 struct bpf_insn_access_aux *info) 8537 { 8538 if (!bpf_sock_is_valid_access(off, size, type, info)) 8539 return false; 8540 return __sock_filter_check_attach_type(off, type, 8541 prog->expected_attach_type); 8542 } 8543 8544 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 8545 const struct bpf_prog *prog) 8546 { 8547 /* Neither direct read nor direct write requires any preliminary 8548 * action. 8549 */ 8550 return 0; 8551 } 8552 8553 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 8554 const struct bpf_prog *prog, int drop_verdict) 8555 { 8556 struct bpf_insn *insn = insn_buf; 8557 8558 if (!direct_write) 8559 return 0; 8560 8561 /* if (!skb->cloned) 8562 * goto start; 8563 * 8564 * (Fast-path, otherwise approximation that we might be 8565 * a clone, do the rest in helper.) 8566 */ 8567 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET); 8568 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 8569 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 8570 8571 /* ret = bpf_skb_pull_data(skb, 0); */ 8572 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 8573 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 8574 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 8575 BPF_FUNC_skb_pull_data); 8576 /* if (!ret) 8577 * goto restore; 8578 * return TC_ACT_SHOT; 8579 */ 8580 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 8581 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 8582 *insn++ = BPF_EXIT_INSN(); 8583 8584 /* restore: */ 8585 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 8586 /* start: */ 8587 *insn++ = prog->insnsi[0]; 8588 8589 return insn - insn_buf; 8590 } 8591 8592 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 8593 struct bpf_insn *insn_buf) 8594 { 8595 bool indirect = BPF_MODE(orig->code) == BPF_IND; 8596 struct bpf_insn *insn = insn_buf; 8597 8598 if (!indirect) { 8599 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 8600 } else { 8601 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 8602 if (orig->imm) 8603 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 8604 } 8605 /* We're guaranteed here that CTX is in R6. */ 8606 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 8607 8608 switch (BPF_SIZE(orig->code)) { 8609 case BPF_B: 8610 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 8611 break; 8612 case BPF_H: 8613 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 8614 break; 8615 case BPF_W: 8616 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 8617 break; 8618 } 8619 8620 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 8621 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 8622 *insn++ = BPF_EXIT_INSN(); 8623 8624 return insn - insn_buf; 8625 } 8626 8627 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 8628 const struct bpf_prog *prog) 8629 { 8630 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 8631 } 8632 8633 static bool tc_cls_act_is_valid_access(int off, int size, 8634 enum bpf_access_type type, 8635 const struct bpf_prog *prog, 8636 struct bpf_insn_access_aux *info) 8637 { 8638 if (type == BPF_WRITE) { 8639 switch (off) { 8640 case bpf_ctx_range(struct __sk_buff, mark): 8641 case bpf_ctx_range(struct __sk_buff, tc_index): 8642 case bpf_ctx_range(struct __sk_buff, priority): 8643 case bpf_ctx_range(struct __sk_buff, tc_classid): 8644 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8645 case bpf_ctx_range(struct __sk_buff, tstamp): 8646 case bpf_ctx_range(struct __sk_buff, queue_mapping): 8647 break; 8648 default: 8649 return false; 8650 } 8651 } 8652 8653 switch (off) { 8654 case bpf_ctx_range(struct __sk_buff, data): 8655 info->reg_type = PTR_TO_PACKET; 8656 break; 8657 case bpf_ctx_range(struct __sk_buff, data_meta): 8658 info->reg_type = PTR_TO_PACKET_META; 8659 break; 8660 case bpf_ctx_range(struct __sk_buff, data_end): 8661 info->reg_type = PTR_TO_PACKET_END; 8662 break; 8663 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8664 return false; 8665 case offsetof(struct __sk_buff, tstamp_type): 8666 /* The convert_ctx_access() on reading and writing 8667 * __sk_buff->tstamp depends on whether the bpf prog 8668 * has used __sk_buff->tstamp_type or not. 8669 * Thus, we need to set prog->tstamp_type_access 8670 * earlier during is_valid_access() here. 8671 */ 8672 ((struct bpf_prog *)prog)->tstamp_type_access = 1; 8673 return size == sizeof(__u8); 8674 } 8675 8676 return bpf_skb_is_valid_access(off, size, type, prog, info); 8677 } 8678 8679 DEFINE_MUTEX(nf_conn_btf_access_lock); 8680 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock); 8681 8682 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 8683 const struct bpf_reg_state *reg, 8684 int off, int size, enum bpf_access_type atype, 8685 u32 *next_btf_id, enum bpf_type_flag *flag); 8686 EXPORT_SYMBOL_GPL(nfct_btf_struct_access); 8687 8688 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log, 8689 const struct bpf_reg_state *reg, 8690 int off, int size, enum bpf_access_type atype, 8691 u32 *next_btf_id, enum bpf_type_flag *flag) 8692 { 8693 int ret = -EACCES; 8694 8695 if (atype == BPF_READ) 8696 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8697 8698 mutex_lock(&nf_conn_btf_access_lock); 8699 if (nfct_btf_struct_access) 8700 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8701 mutex_unlock(&nf_conn_btf_access_lock); 8702 8703 return ret; 8704 } 8705 8706 static bool __is_valid_xdp_access(int off, int size) 8707 { 8708 if (off < 0 || off >= sizeof(struct xdp_md)) 8709 return false; 8710 if (off % size != 0) 8711 return false; 8712 if (size != sizeof(__u32)) 8713 return false; 8714 8715 return true; 8716 } 8717 8718 static bool xdp_is_valid_access(int off, int size, 8719 enum bpf_access_type type, 8720 const struct bpf_prog *prog, 8721 struct bpf_insn_access_aux *info) 8722 { 8723 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 8724 switch (off) { 8725 case offsetof(struct xdp_md, egress_ifindex): 8726 return false; 8727 } 8728 } 8729 8730 if (type == BPF_WRITE) { 8731 if (bpf_prog_is_dev_bound(prog->aux)) { 8732 switch (off) { 8733 case offsetof(struct xdp_md, rx_queue_index): 8734 return __is_valid_xdp_access(off, size); 8735 } 8736 } 8737 return false; 8738 } 8739 8740 switch (off) { 8741 case offsetof(struct xdp_md, data): 8742 info->reg_type = PTR_TO_PACKET; 8743 break; 8744 case offsetof(struct xdp_md, data_meta): 8745 info->reg_type = PTR_TO_PACKET_META; 8746 break; 8747 case offsetof(struct xdp_md, data_end): 8748 info->reg_type = PTR_TO_PACKET_END; 8749 break; 8750 } 8751 8752 return __is_valid_xdp_access(off, size); 8753 } 8754 8755 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act) 8756 { 8757 const u32 act_max = XDP_REDIRECT; 8758 8759 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n", 8760 act > act_max ? "Illegal" : "Driver unsupported", 8761 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A"); 8762 } 8763 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 8764 8765 static int xdp_btf_struct_access(struct bpf_verifier_log *log, 8766 const struct bpf_reg_state *reg, 8767 int off, int size, enum bpf_access_type atype, 8768 u32 *next_btf_id, enum bpf_type_flag *flag) 8769 { 8770 int ret = -EACCES; 8771 8772 if (atype == BPF_READ) 8773 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8774 8775 mutex_lock(&nf_conn_btf_access_lock); 8776 if (nfct_btf_struct_access) 8777 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8778 mutex_unlock(&nf_conn_btf_access_lock); 8779 8780 return ret; 8781 } 8782 8783 static bool sock_addr_is_valid_access(int off, int size, 8784 enum bpf_access_type type, 8785 const struct bpf_prog *prog, 8786 struct bpf_insn_access_aux *info) 8787 { 8788 const int size_default = sizeof(__u32); 8789 8790 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 8791 return false; 8792 if (off % size != 0) 8793 return false; 8794 8795 /* Disallow access to IPv6 fields from IPv4 contex and vise 8796 * versa. 8797 */ 8798 switch (off) { 8799 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8800 switch (prog->expected_attach_type) { 8801 case BPF_CGROUP_INET4_BIND: 8802 case BPF_CGROUP_INET4_CONNECT: 8803 case BPF_CGROUP_INET4_GETPEERNAME: 8804 case BPF_CGROUP_INET4_GETSOCKNAME: 8805 case BPF_CGROUP_UDP4_SENDMSG: 8806 case BPF_CGROUP_UDP4_RECVMSG: 8807 break; 8808 default: 8809 return false; 8810 } 8811 break; 8812 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8813 switch (prog->expected_attach_type) { 8814 case BPF_CGROUP_INET6_BIND: 8815 case BPF_CGROUP_INET6_CONNECT: 8816 case BPF_CGROUP_INET6_GETPEERNAME: 8817 case BPF_CGROUP_INET6_GETSOCKNAME: 8818 case BPF_CGROUP_UDP6_SENDMSG: 8819 case BPF_CGROUP_UDP6_RECVMSG: 8820 break; 8821 default: 8822 return false; 8823 } 8824 break; 8825 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8826 switch (prog->expected_attach_type) { 8827 case BPF_CGROUP_UDP4_SENDMSG: 8828 break; 8829 default: 8830 return false; 8831 } 8832 break; 8833 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8834 msg_src_ip6[3]): 8835 switch (prog->expected_attach_type) { 8836 case BPF_CGROUP_UDP6_SENDMSG: 8837 break; 8838 default: 8839 return false; 8840 } 8841 break; 8842 } 8843 8844 switch (off) { 8845 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8846 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8847 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8848 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8849 msg_src_ip6[3]): 8850 case bpf_ctx_range(struct bpf_sock_addr, user_port): 8851 if (type == BPF_READ) { 8852 bpf_ctx_record_field_size(info, size_default); 8853 8854 if (bpf_ctx_wide_access_ok(off, size, 8855 struct bpf_sock_addr, 8856 user_ip6)) 8857 return true; 8858 8859 if (bpf_ctx_wide_access_ok(off, size, 8860 struct bpf_sock_addr, 8861 msg_src_ip6)) 8862 return true; 8863 8864 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8865 return false; 8866 } else { 8867 if (bpf_ctx_wide_access_ok(off, size, 8868 struct bpf_sock_addr, 8869 user_ip6)) 8870 return true; 8871 8872 if (bpf_ctx_wide_access_ok(off, size, 8873 struct bpf_sock_addr, 8874 msg_src_ip6)) 8875 return true; 8876 8877 if (size != size_default) 8878 return false; 8879 } 8880 break; 8881 case offsetof(struct bpf_sock_addr, sk): 8882 if (type != BPF_READ) 8883 return false; 8884 if (size != sizeof(__u64)) 8885 return false; 8886 info->reg_type = PTR_TO_SOCKET; 8887 break; 8888 default: 8889 if (type == BPF_READ) { 8890 if (size != size_default) 8891 return false; 8892 } else { 8893 return false; 8894 } 8895 } 8896 8897 return true; 8898 } 8899 8900 static bool sock_ops_is_valid_access(int off, int size, 8901 enum bpf_access_type type, 8902 const struct bpf_prog *prog, 8903 struct bpf_insn_access_aux *info) 8904 { 8905 const int size_default = sizeof(__u32); 8906 8907 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8908 return false; 8909 8910 /* The verifier guarantees that size > 0. */ 8911 if (off % size != 0) 8912 return false; 8913 8914 if (type == BPF_WRITE) { 8915 switch (off) { 8916 case offsetof(struct bpf_sock_ops, reply): 8917 case offsetof(struct bpf_sock_ops, sk_txhash): 8918 if (size != size_default) 8919 return false; 8920 break; 8921 default: 8922 return false; 8923 } 8924 } else { 8925 switch (off) { 8926 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8927 bytes_acked): 8928 if (size != sizeof(__u64)) 8929 return false; 8930 break; 8931 case offsetof(struct bpf_sock_ops, sk): 8932 if (size != sizeof(__u64)) 8933 return false; 8934 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8935 break; 8936 case offsetof(struct bpf_sock_ops, skb_data): 8937 if (size != sizeof(__u64)) 8938 return false; 8939 info->reg_type = PTR_TO_PACKET; 8940 break; 8941 case offsetof(struct bpf_sock_ops, skb_data_end): 8942 if (size != sizeof(__u64)) 8943 return false; 8944 info->reg_type = PTR_TO_PACKET_END; 8945 break; 8946 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 8947 bpf_ctx_record_field_size(info, size_default); 8948 return bpf_ctx_narrow_access_ok(off, size, 8949 size_default); 8950 case offsetof(struct bpf_sock_ops, skb_hwtstamp): 8951 if (size != sizeof(__u64)) 8952 return false; 8953 break; 8954 default: 8955 if (size != size_default) 8956 return false; 8957 break; 8958 } 8959 } 8960 8961 return true; 8962 } 8963 8964 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 8965 const struct bpf_prog *prog) 8966 { 8967 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 8968 } 8969 8970 static bool sk_skb_is_valid_access(int off, int size, 8971 enum bpf_access_type type, 8972 const struct bpf_prog *prog, 8973 struct bpf_insn_access_aux *info) 8974 { 8975 switch (off) { 8976 case bpf_ctx_range(struct __sk_buff, tc_classid): 8977 case bpf_ctx_range(struct __sk_buff, data_meta): 8978 case bpf_ctx_range(struct __sk_buff, tstamp): 8979 case bpf_ctx_range(struct __sk_buff, wire_len): 8980 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8981 return false; 8982 } 8983 8984 if (type == BPF_WRITE) { 8985 switch (off) { 8986 case bpf_ctx_range(struct __sk_buff, tc_index): 8987 case bpf_ctx_range(struct __sk_buff, priority): 8988 break; 8989 default: 8990 return false; 8991 } 8992 } 8993 8994 switch (off) { 8995 case bpf_ctx_range(struct __sk_buff, mark): 8996 return false; 8997 case bpf_ctx_range(struct __sk_buff, data): 8998 info->reg_type = PTR_TO_PACKET; 8999 break; 9000 case bpf_ctx_range(struct __sk_buff, data_end): 9001 info->reg_type = PTR_TO_PACKET_END; 9002 break; 9003 } 9004 9005 return bpf_skb_is_valid_access(off, size, type, prog, info); 9006 } 9007 9008 static bool sk_msg_is_valid_access(int off, int size, 9009 enum bpf_access_type type, 9010 const struct bpf_prog *prog, 9011 struct bpf_insn_access_aux *info) 9012 { 9013 if (type == BPF_WRITE) 9014 return false; 9015 9016 if (off % size != 0) 9017 return false; 9018 9019 switch (off) { 9020 case offsetof(struct sk_msg_md, data): 9021 info->reg_type = PTR_TO_PACKET; 9022 if (size != sizeof(__u64)) 9023 return false; 9024 break; 9025 case offsetof(struct sk_msg_md, data_end): 9026 info->reg_type = PTR_TO_PACKET_END; 9027 if (size != sizeof(__u64)) 9028 return false; 9029 break; 9030 case offsetof(struct sk_msg_md, sk): 9031 if (size != sizeof(__u64)) 9032 return false; 9033 info->reg_type = PTR_TO_SOCKET; 9034 break; 9035 case bpf_ctx_range(struct sk_msg_md, family): 9036 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 9037 case bpf_ctx_range(struct sk_msg_md, local_ip4): 9038 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 9039 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 9040 case bpf_ctx_range(struct sk_msg_md, remote_port): 9041 case bpf_ctx_range(struct sk_msg_md, local_port): 9042 case bpf_ctx_range(struct sk_msg_md, size): 9043 if (size != sizeof(__u32)) 9044 return false; 9045 break; 9046 default: 9047 return false; 9048 } 9049 return true; 9050 } 9051 9052 static bool flow_dissector_is_valid_access(int off, int size, 9053 enum bpf_access_type type, 9054 const struct bpf_prog *prog, 9055 struct bpf_insn_access_aux *info) 9056 { 9057 const int size_default = sizeof(__u32); 9058 9059 if (off < 0 || off >= sizeof(struct __sk_buff)) 9060 return false; 9061 9062 if (type == BPF_WRITE) 9063 return false; 9064 9065 switch (off) { 9066 case bpf_ctx_range(struct __sk_buff, data): 9067 if (size != size_default) 9068 return false; 9069 info->reg_type = PTR_TO_PACKET; 9070 return true; 9071 case bpf_ctx_range(struct __sk_buff, data_end): 9072 if (size != size_default) 9073 return false; 9074 info->reg_type = PTR_TO_PACKET_END; 9075 return true; 9076 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 9077 if (size != sizeof(__u64)) 9078 return false; 9079 info->reg_type = PTR_TO_FLOW_KEYS; 9080 return true; 9081 default: 9082 return false; 9083 } 9084 } 9085 9086 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 9087 const struct bpf_insn *si, 9088 struct bpf_insn *insn_buf, 9089 struct bpf_prog *prog, 9090 u32 *target_size) 9091 9092 { 9093 struct bpf_insn *insn = insn_buf; 9094 9095 switch (si->off) { 9096 case offsetof(struct __sk_buff, data): 9097 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 9098 si->dst_reg, si->src_reg, 9099 offsetof(struct bpf_flow_dissector, data)); 9100 break; 9101 9102 case offsetof(struct __sk_buff, data_end): 9103 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 9104 si->dst_reg, si->src_reg, 9105 offsetof(struct bpf_flow_dissector, data_end)); 9106 break; 9107 9108 case offsetof(struct __sk_buff, flow_keys): 9109 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 9110 si->dst_reg, si->src_reg, 9111 offsetof(struct bpf_flow_dissector, flow_keys)); 9112 break; 9113 } 9114 9115 return insn - insn_buf; 9116 } 9117 9118 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si, 9119 struct bpf_insn *insn) 9120 { 9121 __u8 value_reg = si->dst_reg; 9122 __u8 skb_reg = si->src_reg; 9123 /* AX is needed because src_reg and dst_reg could be the same */ 9124 __u8 tmp_reg = BPF_REG_AX; 9125 9126 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, 9127 PKT_VLAN_PRESENT_OFFSET); 9128 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, 9129 SKB_MONO_DELIVERY_TIME_MASK, 2); 9130 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); 9131 *insn++ = BPF_JMP_A(1); 9132 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO); 9133 9134 return insn; 9135 } 9136 9137 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg, 9138 struct bpf_insn *insn) 9139 { 9140 /* si->dst_reg = skb_shinfo(SKB); */ 9141 #ifdef NET_SKBUFF_DATA_USES_OFFSET 9142 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9143 BPF_REG_AX, skb_reg, 9144 offsetof(struct sk_buff, end)); 9145 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 9146 dst_reg, skb_reg, 9147 offsetof(struct sk_buff, head)); 9148 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX); 9149 #else 9150 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9151 dst_reg, skb_reg, 9152 offsetof(struct sk_buff, end)); 9153 #endif 9154 9155 return insn; 9156 } 9157 9158 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, 9159 const struct bpf_insn *si, 9160 struct bpf_insn *insn) 9161 { 9162 __u8 value_reg = si->dst_reg; 9163 __u8 skb_reg = si->src_reg; 9164 9165 #ifdef CONFIG_NET_CLS_ACT 9166 /* If the tstamp_type is read, 9167 * the bpf prog is aware the tstamp could have delivery time. 9168 * Thus, read skb->tstamp as is if tstamp_type_access is true. 9169 */ 9170 if (!prog->tstamp_type_access) { 9171 /* AX is needed because src_reg and dst_reg could be the same */ 9172 __u8 tmp_reg = BPF_REG_AX; 9173 9174 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 9175 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, 9176 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK); 9177 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg, 9178 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2); 9179 /* skb->tc_at_ingress && skb->mono_delivery_time, 9180 * read 0 as the (rcv) timestamp. 9181 */ 9182 *insn++ = BPF_MOV64_IMM(value_reg, 0); 9183 *insn++ = BPF_JMP_A(1); 9184 } 9185 #endif 9186 9187 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg, 9188 offsetof(struct sk_buff, tstamp)); 9189 return insn; 9190 } 9191 9192 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, 9193 const struct bpf_insn *si, 9194 struct bpf_insn *insn) 9195 { 9196 __u8 value_reg = si->src_reg; 9197 __u8 skb_reg = si->dst_reg; 9198 9199 #ifdef CONFIG_NET_CLS_ACT 9200 /* If the tstamp_type is read, 9201 * the bpf prog is aware the tstamp could have delivery time. 9202 * Thus, write skb->tstamp as is if tstamp_type_access is true. 9203 * Otherwise, writing at ingress will have to clear the 9204 * mono_delivery_time bit also. 9205 */ 9206 if (!prog->tstamp_type_access) { 9207 __u8 tmp_reg = BPF_REG_AX; 9208 9209 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 9210 /* Writing __sk_buff->tstamp as ingress, goto <clear> */ 9211 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); 9212 /* goto <store> */ 9213 *insn++ = BPF_JMP_A(2); 9214 /* <clear>: mono_delivery_time */ 9215 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); 9216 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET); 9217 } 9218 #endif 9219 9220 /* <store>: skb->tstamp = tstamp */ 9221 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg, 9222 offsetof(struct sk_buff, tstamp)); 9223 return insn; 9224 } 9225 9226 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 9227 const struct bpf_insn *si, 9228 struct bpf_insn *insn_buf, 9229 struct bpf_prog *prog, u32 *target_size) 9230 { 9231 struct bpf_insn *insn = insn_buf; 9232 int off; 9233 9234 switch (si->off) { 9235 case offsetof(struct __sk_buff, len): 9236 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9237 bpf_target_off(struct sk_buff, len, 4, 9238 target_size)); 9239 break; 9240 9241 case offsetof(struct __sk_buff, protocol): 9242 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9243 bpf_target_off(struct sk_buff, protocol, 2, 9244 target_size)); 9245 break; 9246 9247 case offsetof(struct __sk_buff, vlan_proto): 9248 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9249 bpf_target_off(struct sk_buff, vlan_proto, 2, 9250 target_size)); 9251 break; 9252 9253 case offsetof(struct __sk_buff, priority): 9254 if (type == BPF_WRITE) 9255 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9256 bpf_target_off(struct sk_buff, priority, 4, 9257 target_size)); 9258 else 9259 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9260 bpf_target_off(struct sk_buff, priority, 4, 9261 target_size)); 9262 break; 9263 9264 case offsetof(struct __sk_buff, ingress_ifindex): 9265 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9266 bpf_target_off(struct sk_buff, skb_iif, 4, 9267 target_size)); 9268 break; 9269 9270 case offsetof(struct __sk_buff, ifindex): 9271 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9272 si->dst_reg, si->src_reg, 9273 offsetof(struct sk_buff, dev)); 9274 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9275 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9276 bpf_target_off(struct net_device, ifindex, 4, 9277 target_size)); 9278 break; 9279 9280 case offsetof(struct __sk_buff, hash): 9281 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9282 bpf_target_off(struct sk_buff, hash, 4, 9283 target_size)); 9284 break; 9285 9286 case offsetof(struct __sk_buff, mark): 9287 if (type == BPF_WRITE) 9288 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9289 bpf_target_off(struct sk_buff, mark, 4, 9290 target_size)); 9291 else 9292 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9293 bpf_target_off(struct sk_buff, mark, 4, 9294 target_size)); 9295 break; 9296 9297 case offsetof(struct __sk_buff, pkt_type): 9298 *target_size = 1; 9299 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 9300 PKT_TYPE_OFFSET); 9301 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 9302 #ifdef __BIG_ENDIAN_BITFIELD 9303 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 9304 #endif 9305 break; 9306 9307 case offsetof(struct __sk_buff, queue_mapping): 9308 if (type == BPF_WRITE) { 9309 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 9310 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9311 bpf_target_off(struct sk_buff, 9312 queue_mapping, 9313 2, target_size)); 9314 } else { 9315 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9316 bpf_target_off(struct sk_buff, 9317 queue_mapping, 9318 2, target_size)); 9319 } 9320 break; 9321 9322 case offsetof(struct __sk_buff, vlan_present): 9323 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9324 bpf_target_off(struct sk_buff, 9325 vlan_all, 4, target_size)); 9326 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9327 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1); 9328 break; 9329 9330 case offsetof(struct __sk_buff, vlan_tci): 9331 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9332 bpf_target_off(struct sk_buff, vlan_tci, 2, 9333 target_size)); 9334 break; 9335 9336 case offsetof(struct __sk_buff, cb[0]) ... 9337 offsetofend(struct __sk_buff, cb[4]) - 1: 9338 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 9339 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 9340 offsetof(struct qdisc_skb_cb, data)) % 9341 sizeof(__u64)); 9342 9343 prog->cb_access = 1; 9344 off = si->off; 9345 off -= offsetof(struct __sk_buff, cb[0]); 9346 off += offsetof(struct sk_buff, cb); 9347 off += offsetof(struct qdisc_skb_cb, data); 9348 if (type == BPF_WRITE) 9349 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 9350 si->src_reg, off); 9351 else 9352 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 9353 si->src_reg, off); 9354 break; 9355 9356 case offsetof(struct __sk_buff, tc_classid): 9357 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 9358 9359 off = si->off; 9360 off -= offsetof(struct __sk_buff, tc_classid); 9361 off += offsetof(struct sk_buff, cb); 9362 off += offsetof(struct qdisc_skb_cb, tc_classid); 9363 *target_size = 2; 9364 if (type == BPF_WRITE) 9365 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 9366 si->src_reg, off); 9367 else 9368 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 9369 si->src_reg, off); 9370 break; 9371 9372 case offsetof(struct __sk_buff, data): 9373 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9374 si->dst_reg, si->src_reg, 9375 offsetof(struct sk_buff, data)); 9376 break; 9377 9378 case offsetof(struct __sk_buff, data_meta): 9379 off = si->off; 9380 off -= offsetof(struct __sk_buff, data_meta); 9381 off += offsetof(struct sk_buff, cb); 9382 off += offsetof(struct bpf_skb_data_end, data_meta); 9383 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9384 si->src_reg, off); 9385 break; 9386 9387 case offsetof(struct __sk_buff, data_end): 9388 off = si->off; 9389 off -= offsetof(struct __sk_buff, data_end); 9390 off += offsetof(struct sk_buff, cb); 9391 off += offsetof(struct bpf_skb_data_end, data_end); 9392 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9393 si->src_reg, off); 9394 break; 9395 9396 case offsetof(struct __sk_buff, tc_index): 9397 #ifdef CONFIG_NET_SCHED 9398 if (type == BPF_WRITE) 9399 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9400 bpf_target_off(struct sk_buff, tc_index, 2, 9401 target_size)); 9402 else 9403 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9404 bpf_target_off(struct sk_buff, tc_index, 2, 9405 target_size)); 9406 #else 9407 *target_size = 2; 9408 if (type == BPF_WRITE) 9409 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 9410 else 9411 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9412 #endif 9413 break; 9414 9415 case offsetof(struct __sk_buff, napi_id): 9416 #if defined(CONFIG_NET_RX_BUSY_POLL) 9417 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9418 bpf_target_off(struct sk_buff, napi_id, 4, 9419 target_size)); 9420 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 9421 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9422 #else 9423 *target_size = 4; 9424 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9425 #endif 9426 break; 9427 case offsetof(struct __sk_buff, family): 9428 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9429 9430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9431 si->dst_reg, si->src_reg, 9432 offsetof(struct sk_buff, sk)); 9433 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9434 bpf_target_off(struct sock_common, 9435 skc_family, 9436 2, target_size)); 9437 break; 9438 case offsetof(struct __sk_buff, remote_ip4): 9439 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9440 9441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9442 si->dst_reg, si->src_reg, 9443 offsetof(struct sk_buff, sk)); 9444 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9445 bpf_target_off(struct sock_common, 9446 skc_daddr, 9447 4, target_size)); 9448 break; 9449 case offsetof(struct __sk_buff, local_ip4): 9450 BUILD_BUG_ON(sizeof_field(struct sock_common, 9451 skc_rcv_saddr) != 4); 9452 9453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9454 si->dst_reg, si->src_reg, 9455 offsetof(struct sk_buff, sk)); 9456 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9457 bpf_target_off(struct sock_common, 9458 skc_rcv_saddr, 9459 4, target_size)); 9460 break; 9461 case offsetof(struct __sk_buff, remote_ip6[0]) ... 9462 offsetof(struct __sk_buff, remote_ip6[3]): 9463 #if IS_ENABLED(CONFIG_IPV6) 9464 BUILD_BUG_ON(sizeof_field(struct sock_common, 9465 skc_v6_daddr.s6_addr32[0]) != 4); 9466 9467 off = si->off; 9468 off -= offsetof(struct __sk_buff, remote_ip6[0]); 9469 9470 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9471 si->dst_reg, si->src_reg, 9472 offsetof(struct sk_buff, sk)); 9473 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9474 offsetof(struct sock_common, 9475 skc_v6_daddr.s6_addr32[0]) + 9476 off); 9477 #else 9478 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9479 #endif 9480 break; 9481 case offsetof(struct __sk_buff, local_ip6[0]) ... 9482 offsetof(struct __sk_buff, local_ip6[3]): 9483 #if IS_ENABLED(CONFIG_IPV6) 9484 BUILD_BUG_ON(sizeof_field(struct sock_common, 9485 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9486 9487 off = si->off; 9488 off -= offsetof(struct __sk_buff, local_ip6[0]); 9489 9490 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9491 si->dst_reg, si->src_reg, 9492 offsetof(struct sk_buff, sk)); 9493 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9494 offsetof(struct sock_common, 9495 skc_v6_rcv_saddr.s6_addr32[0]) + 9496 off); 9497 #else 9498 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9499 #endif 9500 break; 9501 9502 case offsetof(struct __sk_buff, remote_port): 9503 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9504 9505 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9506 si->dst_reg, si->src_reg, 9507 offsetof(struct sk_buff, sk)); 9508 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9509 bpf_target_off(struct sock_common, 9510 skc_dport, 9511 2, target_size)); 9512 #ifndef __BIG_ENDIAN_BITFIELD 9513 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9514 #endif 9515 break; 9516 9517 case offsetof(struct __sk_buff, local_port): 9518 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9519 9520 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9521 si->dst_reg, si->src_reg, 9522 offsetof(struct sk_buff, sk)); 9523 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9524 bpf_target_off(struct sock_common, 9525 skc_num, 2, target_size)); 9526 break; 9527 9528 case offsetof(struct __sk_buff, tstamp): 9529 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 9530 9531 if (type == BPF_WRITE) 9532 insn = bpf_convert_tstamp_write(prog, si, insn); 9533 else 9534 insn = bpf_convert_tstamp_read(prog, si, insn); 9535 break; 9536 9537 case offsetof(struct __sk_buff, tstamp_type): 9538 insn = bpf_convert_tstamp_type_read(si, insn); 9539 break; 9540 9541 case offsetof(struct __sk_buff, gso_segs): 9542 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9543 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 9544 si->dst_reg, si->dst_reg, 9545 bpf_target_off(struct skb_shared_info, 9546 gso_segs, 2, 9547 target_size)); 9548 break; 9549 case offsetof(struct __sk_buff, gso_size): 9550 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9551 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 9552 si->dst_reg, si->dst_reg, 9553 bpf_target_off(struct skb_shared_info, 9554 gso_size, 2, 9555 target_size)); 9556 break; 9557 case offsetof(struct __sk_buff, wire_len): 9558 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 9559 9560 off = si->off; 9561 off -= offsetof(struct __sk_buff, wire_len); 9562 off += offsetof(struct sk_buff, cb); 9563 off += offsetof(struct qdisc_skb_cb, pkt_len); 9564 *target_size = 4; 9565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 9566 break; 9567 9568 case offsetof(struct __sk_buff, sk): 9569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9570 si->dst_reg, si->src_reg, 9571 offsetof(struct sk_buff, sk)); 9572 break; 9573 case offsetof(struct __sk_buff, hwtstamp): 9574 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8); 9575 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0); 9576 9577 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9578 *insn++ = BPF_LDX_MEM(BPF_DW, 9579 si->dst_reg, si->dst_reg, 9580 bpf_target_off(struct skb_shared_info, 9581 hwtstamps, 8, 9582 target_size)); 9583 break; 9584 } 9585 9586 return insn - insn_buf; 9587 } 9588 9589 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 9590 const struct bpf_insn *si, 9591 struct bpf_insn *insn_buf, 9592 struct bpf_prog *prog, u32 *target_size) 9593 { 9594 struct bpf_insn *insn = insn_buf; 9595 int off; 9596 9597 switch (si->off) { 9598 case offsetof(struct bpf_sock, bound_dev_if): 9599 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 9600 9601 if (type == BPF_WRITE) 9602 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9603 offsetof(struct sock, sk_bound_dev_if)); 9604 else 9605 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9606 offsetof(struct sock, sk_bound_dev_if)); 9607 break; 9608 9609 case offsetof(struct bpf_sock, mark): 9610 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 9611 9612 if (type == BPF_WRITE) 9613 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9614 offsetof(struct sock, sk_mark)); 9615 else 9616 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9617 offsetof(struct sock, sk_mark)); 9618 break; 9619 9620 case offsetof(struct bpf_sock, priority): 9621 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 9622 9623 if (type == BPF_WRITE) 9624 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9625 offsetof(struct sock, sk_priority)); 9626 else 9627 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9628 offsetof(struct sock, sk_priority)); 9629 break; 9630 9631 case offsetof(struct bpf_sock, family): 9632 *insn++ = BPF_LDX_MEM( 9633 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 9634 si->dst_reg, si->src_reg, 9635 bpf_target_off(struct sock_common, 9636 skc_family, 9637 sizeof_field(struct sock_common, 9638 skc_family), 9639 target_size)); 9640 break; 9641 9642 case offsetof(struct bpf_sock, type): 9643 *insn++ = BPF_LDX_MEM( 9644 BPF_FIELD_SIZEOF(struct sock, sk_type), 9645 si->dst_reg, si->src_reg, 9646 bpf_target_off(struct sock, sk_type, 9647 sizeof_field(struct sock, sk_type), 9648 target_size)); 9649 break; 9650 9651 case offsetof(struct bpf_sock, protocol): 9652 *insn++ = BPF_LDX_MEM( 9653 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 9654 si->dst_reg, si->src_reg, 9655 bpf_target_off(struct sock, sk_protocol, 9656 sizeof_field(struct sock, sk_protocol), 9657 target_size)); 9658 break; 9659 9660 case offsetof(struct bpf_sock, src_ip4): 9661 *insn++ = BPF_LDX_MEM( 9662 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9663 bpf_target_off(struct sock_common, skc_rcv_saddr, 9664 sizeof_field(struct sock_common, 9665 skc_rcv_saddr), 9666 target_size)); 9667 break; 9668 9669 case offsetof(struct bpf_sock, dst_ip4): 9670 *insn++ = BPF_LDX_MEM( 9671 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9672 bpf_target_off(struct sock_common, skc_daddr, 9673 sizeof_field(struct sock_common, 9674 skc_daddr), 9675 target_size)); 9676 break; 9677 9678 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9679 #if IS_ENABLED(CONFIG_IPV6) 9680 off = si->off; 9681 off -= offsetof(struct bpf_sock, src_ip6[0]); 9682 *insn++ = BPF_LDX_MEM( 9683 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9684 bpf_target_off( 9685 struct sock_common, 9686 skc_v6_rcv_saddr.s6_addr32[0], 9687 sizeof_field(struct sock_common, 9688 skc_v6_rcv_saddr.s6_addr32[0]), 9689 target_size) + off); 9690 #else 9691 (void)off; 9692 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9693 #endif 9694 break; 9695 9696 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 9697 #if IS_ENABLED(CONFIG_IPV6) 9698 off = si->off; 9699 off -= offsetof(struct bpf_sock, dst_ip6[0]); 9700 *insn++ = BPF_LDX_MEM( 9701 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9702 bpf_target_off(struct sock_common, 9703 skc_v6_daddr.s6_addr32[0], 9704 sizeof_field(struct sock_common, 9705 skc_v6_daddr.s6_addr32[0]), 9706 target_size) + off); 9707 #else 9708 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9709 *target_size = 4; 9710 #endif 9711 break; 9712 9713 case offsetof(struct bpf_sock, src_port): 9714 *insn++ = BPF_LDX_MEM( 9715 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 9716 si->dst_reg, si->src_reg, 9717 bpf_target_off(struct sock_common, skc_num, 9718 sizeof_field(struct sock_common, 9719 skc_num), 9720 target_size)); 9721 break; 9722 9723 case offsetof(struct bpf_sock, dst_port): 9724 *insn++ = BPF_LDX_MEM( 9725 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 9726 si->dst_reg, si->src_reg, 9727 bpf_target_off(struct sock_common, skc_dport, 9728 sizeof_field(struct sock_common, 9729 skc_dport), 9730 target_size)); 9731 break; 9732 9733 case offsetof(struct bpf_sock, state): 9734 *insn++ = BPF_LDX_MEM( 9735 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 9736 si->dst_reg, si->src_reg, 9737 bpf_target_off(struct sock_common, skc_state, 9738 sizeof_field(struct sock_common, 9739 skc_state), 9740 target_size)); 9741 break; 9742 case offsetof(struct bpf_sock, rx_queue_mapping): 9743 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 9744 *insn++ = BPF_LDX_MEM( 9745 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 9746 si->dst_reg, si->src_reg, 9747 bpf_target_off(struct sock, sk_rx_queue_mapping, 9748 sizeof_field(struct sock, 9749 sk_rx_queue_mapping), 9750 target_size)); 9751 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 9752 1); 9753 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9754 #else 9755 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9756 *target_size = 2; 9757 #endif 9758 break; 9759 } 9760 9761 return insn - insn_buf; 9762 } 9763 9764 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 9765 const struct bpf_insn *si, 9766 struct bpf_insn *insn_buf, 9767 struct bpf_prog *prog, u32 *target_size) 9768 { 9769 struct bpf_insn *insn = insn_buf; 9770 9771 switch (si->off) { 9772 case offsetof(struct __sk_buff, ifindex): 9773 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9774 si->dst_reg, si->src_reg, 9775 offsetof(struct sk_buff, dev)); 9776 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9777 bpf_target_off(struct net_device, ifindex, 4, 9778 target_size)); 9779 break; 9780 default: 9781 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9782 target_size); 9783 } 9784 9785 return insn - insn_buf; 9786 } 9787 9788 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 9789 const struct bpf_insn *si, 9790 struct bpf_insn *insn_buf, 9791 struct bpf_prog *prog, u32 *target_size) 9792 { 9793 struct bpf_insn *insn = insn_buf; 9794 9795 switch (si->off) { 9796 case offsetof(struct xdp_md, data): 9797 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 9798 si->dst_reg, si->src_reg, 9799 offsetof(struct xdp_buff, data)); 9800 break; 9801 case offsetof(struct xdp_md, data_meta): 9802 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 9803 si->dst_reg, si->src_reg, 9804 offsetof(struct xdp_buff, data_meta)); 9805 break; 9806 case offsetof(struct xdp_md, data_end): 9807 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 9808 si->dst_reg, si->src_reg, 9809 offsetof(struct xdp_buff, data_end)); 9810 break; 9811 case offsetof(struct xdp_md, ingress_ifindex): 9812 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9813 si->dst_reg, si->src_reg, 9814 offsetof(struct xdp_buff, rxq)); 9815 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 9816 si->dst_reg, si->dst_reg, 9817 offsetof(struct xdp_rxq_info, dev)); 9818 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9819 offsetof(struct net_device, ifindex)); 9820 break; 9821 case offsetof(struct xdp_md, rx_queue_index): 9822 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9823 si->dst_reg, si->src_reg, 9824 offsetof(struct xdp_buff, rxq)); 9825 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9826 offsetof(struct xdp_rxq_info, 9827 queue_index)); 9828 break; 9829 case offsetof(struct xdp_md, egress_ifindex): 9830 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 9831 si->dst_reg, si->src_reg, 9832 offsetof(struct xdp_buff, txq)); 9833 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 9834 si->dst_reg, si->dst_reg, 9835 offsetof(struct xdp_txq_info, dev)); 9836 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9837 offsetof(struct net_device, ifindex)); 9838 break; 9839 } 9840 9841 return insn - insn_buf; 9842 } 9843 9844 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 9845 * context Structure, F is Field in context structure that contains a pointer 9846 * to Nested Structure of type NS that has the field NF. 9847 * 9848 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 9849 * sure that SIZE is not greater than actual size of S.F.NF. 9850 * 9851 * If offset OFF is provided, the load happens from that offset relative to 9852 * offset of NF. 9853 */ 9854 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 9855 do { \ 9856 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 9857 si->src_reg, offsetof(S, F)); \ 9858 *insn++ = BPF_LDX_MEM( \ 9859 SIZE, si->dst_reg, si->dst_reg, \ 9860 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9861 target_size) \ 9862 + OFF); \ 9863 } while (0) 9864 9865 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 9866 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 9867 BPF_FIELD_SIZEOF(NS, NF), 0) 9868 9869 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 9870 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 9871 * 9872 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 9873 * "register" since two registers available in convert_ctx_access are not 9874 * enough: we can't override neither SRC, since it contains value to store, nor 9875 * DST since it contains pointer to context that may be used by later 9876 * instructions. But we need a temporary place to save pointer to nested 9877 * structure whose field we want to store to. 9878 */ 9879 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 9880 do { \ 9881 int tmp_reg = BPF_REG_9; \ 9882 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9883 --tmp_reg; \ 9884 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9885 --tmp_reg; \ 9886 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 9887 offsetof(S, TF)); \ 9888 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 9889 si->dst_reg, offsetof(S, F)); \ 9890 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 9891 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9892 target_size) \ 9893 + OFF); \ 9894 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 9895 offsetof(S, TF)); \ 9896 } while (0) 9897 9898 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 9899 TF) \ 9900 do { \ 9901 if (type == BPF_WRITE) { \ 9902 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 9903 OFF, TF); \ 9904 } else { \ 9905 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 9906 S, NS, F, NF, SIZE, OFF); \ 9907 } \ 9908 } while (0) 9909 9910 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 9911 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 9912 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 9913 9914 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 9915 const struct bpf_insn *si, 9916 struct bpf_insn *insn_buf, 9917 struct bpf_prog *prog, u32 *target_size) 9918 { 9919 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 9920 struct bpf_insn *insn = insn_buf; 9921 9922 switch (si->off) { 9923 case offsetof(struct bpf_sock_addr, user_family): 9924 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9925 struct sockaddr, uaddr, sa_family); 9926 break; 9927 9928 case offsetof(struct bpf_sock_addr, user_ip4): 9929 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9930 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 9931 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 9932 break; 9933 9934 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 9935 off = si->off; 9936 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 9937 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9938 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9939 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 9940 tmp_reg); 9941 break; 9942 9943 case offsetof(struct bpf_sock_addr, user_port): 9944 /* To get port we need to know sa_family first and then treat 9945 * sockaddr as either sockaddr_in or sockaddr_in6. 9946 * Though we can simplify since port field has same offset and 9947 * size in both structures. 9948 * Here we check this invariant and use just one of the 9949 * structures if it's true. 9950 */ 9951 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 9952 offsetof(struct sockaddr_in6, sin6_port)); 9953 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 9954 sizeof_field(struct sockaddr_in6, sin6_port)); 9955 /* Account for sin6_port being smaller than user_port. */ 9956 port_size = min(port_size, BPF_LDST_BYTES(si)); 9957 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9958 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9959 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 9960 break; 9961 9962 case offsetof(struct bpf_sock_addr, family): 9963 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9964 struct sock, sk, sk_family); 9965 break; 9966 9967 case offsetof(struct bpf_sock_addr, type): 9968 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9969 struct sock, sk, sk_type); 9970 break; 9971 9972 case offsetof(struct bpf_sock_addr, protocol): 9973 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9974 struct sock, sk, sk_protocol); 9975 break; 9976 9977 case offsetof(struct bpf_sock_addr, msg_src_ip4): 9978 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 9979 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9980 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 9981 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 9982 break; 9983 9984 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9985 msg_src_ip6[3]): 9986 off = si->off; 9987 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 9988 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 9989 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9990 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 9991 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 9992 break; 9993 case offsetof(struct bpf_sock_addr, sk): 9994 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 9995 si->dst_reg, si->src_reg, 9996 offsetof(struct bpf_sock_addr_kern, sk)); 9997 break; 9998 } 9999 10000 return insn - insn_buf; 10001 } 10002 10003 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 10004 const struct bpf_insn *si, 10005 struct bpf_insn *insn_buf, 10006 struct bpf_prog *prog, 10007 u32 *target_size) 10008 { 10009 struct bpf_insn *insn = insn_buf; 10010 int off; 10011 10012 /* Helper macro for adding read access to tcp_sock or sock fields. */ 10013 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10014 do { \ 10015 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 10016 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10017 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10018 if (si->dst_reg == reg || si->src_reg == reg) \ 10019 reg--; \ 10020 if (si->dst_reg == reg || si->src_reg == reg) \ 10021 reg--; \ 10022 if (si->dst_reg == si->src_reg) { \ 10023 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10024 offsetof(struct bpf_sock_ops_kern, \ 10025 temp)); \ 10026 fullsock_reg = reg; \ 10027 jmp += 2; \ 10028 } \ 10029 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10030 struct bpf_sock_ops_kern, \ 10031 is_fullsock), \ 10032 fullsock_reg, si->src_reg, \ 10033 offsetof(struct bpf_sock_ops_kern, \ 10034 is_fullsock)); \ 10035 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10036 if (si->dst_reg == si->src_reg) \ 10037 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10038 offsetof(struct bpf_sock_ops_kern, \ 10039 temp)); \ 10040 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10041 struct bpf_sock_ops_kern, sk),\ 10042 si->dst_reg, si->src_reg, \ 10043 offsetof(struct bpf_sock_ops_kern, sk));\ 10044 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 10045 OBJ_FIELD), \ 10046 si->dst_reg, si->dst_reg, \ 10047 offsetof(OBJ, OBJ_FIELD)); \ 10048 if (si->dst_reg == si->src_reg) { \ 10049 *insn++ = BPF_JMP_A(1); \ 10050 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10051 offsetof(struct bpf_sock_ops_kern, \ 10052 temp)); \ 10053 } \ 10054 } while (0) 10055 10056 #define SOCK_OPS_GET_SK() \ 10057 do { \ 10058 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 10059 if (si->dst_reg == reg || si->src_reg == reg) \ 10060 reg--; \ 10061 if (si->dst_reg == reg || si->src_reg == reg) \ 10062 reg--; \ 10063 if (si->dst_reg == si->src_reg) { \ 10064 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10065 offsetof(struct bpf_sock_ops_kern, \ 10066 temp)); \ 10067 fullsock_reg = reg; \ 10068 jmp += 2; \ 10069 } \ 10070 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10071 struct bpf_sock_ops_kern, \ 10072 is_fullsock), \ 10073 fullsock_reg, si->src_reg, \ 10074 offsetof(struct bpf_sock_ops_kern, \ 10075 is_fullsock)); \ 10076 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10077 if (si->dst_reg == si->src_reg) \ 10078 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10079 offsetof(struct bpf_sock_ops_kern, \ 10080 temp)); \ 10081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10082 struct bpf_sock_ops_kern, sk),\ 10083 si->dst_reg, si->src_reg, \ 10084 offsetof(struct bpf_sock_ops_kern, sk));\ 10085 if (si->dst_reg == si->src_reg) { \ 10086 *insn++ = BPF_JMP_A(1); \ 10087 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10088 offsetof(struct bpf_sock_ops_kern, \ 10089 temp)); \ 10090 } \ 10091 } while (0) 10092 10093 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 10094 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 10095 10096 /* Helper macro for adding write access to tcp_sock or sock fields. 10097 * The macro is called with two registers, dst_reg which contains a pointer 10098 * to ctx (context) and src_reg which contains the value that should be 10099 * stored. However, we need an additional register since we cannot overwrite 10100 * dst_reg because it may be used later in the program. 10101 * Instead we "borrow" one of the other register. We first save its value 10102 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 10103 * it at the end of the macro. 10104 */ 10105 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10106 do { \ 10107 int reg = BPF_REG_9; \ 10108 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10109 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10110 if (si->dst_reg == reg || si->src_reg == reg) \ 10111 reg--; \ 10112 if (si->dst_reg == reg || si->src_reg == reg) \ 10113 reg--; \ 10114 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 10115 offsetof(struct bpf_sock_ops_kern, \ 10116 temp)); \ 10117 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10118 struct bpf_sock_ops_kern, \ 10119 is_fullsock), \ 10120 reg, si->dst_reg, \ 10121 offsetof(struct bpf_sock_ops_kern, \ 10122 is_fullsock)); \ 10123 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 10124 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10125 struct bpf_sock_ops_kern, sk),\ 10126 reg, si->dst_reg, \ 10127 offsetof(struct bpf_sock_ops_kern, sk));\ 10128 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 10129 reg, si->src_reg, \ 10130 offsetof(OBJ, OBJ_FIELD)); \ 10131 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 10132 offsetof(struct bpf_sock_ops_kern, \ 10133 temp)); \ 10134 } while (0) 10135 10136 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 10137 do { \ 10138 if (TYPE == BPF_WRITE) \ 10139 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10140 else \ 10141 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10142 } while (0) 10143 10144 if (insn > insn_buf) 10145 return insn - insn_buf; 10146 10147 switch (si->off) { 10148 case offsetof(struct bpf_sock_ops, op): 10149 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10150 op), 10151 si->dst_reg, si->src_reg, 10152 offsetof(struct bpf_sock_ops_kern, op)); 10153 break; 10154 10155 case offsetof(struct bpf_sock_ops, replylong[0]) ... 10156 offsetof(struct bpf_sock_ops, replylong[3]): 10157 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 10158 sizeof_field(struct bpf_sock_ops_kern, reply)); 10159 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 10160 sizeof_field(struct bpf_sock_ops_kern, replylong)); 10161 off = si->off; 10162 off -= offsetof(struct bpf_sock_ops, replylong[0]); 10163 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 10164 if (type == BPF_WRITE) 10165 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 10166 off); 10167 else 10168 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10169 off); 10170 break; 10171 10172 case offsetof(struct bpf_sock_ops, family): 10173 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10174 10175 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10176 struct bpf_sock_ops_kern, sk), 10177 si->dst_reg, si->src_reg, 10178 offsetof(struct bpf_sock_ops_kern, sk)); 10179 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10180 offsetof(struct sock_common, skc_family)); 10181 break; 10182 10183 case offsetof(struct bpf_sock_ops, remote_ip4): 10184 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10185 10186 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10187 struct bpf_sock_ops_kern, sk), 10188 si->dst_reg, si->src_reg, 10189 offsetof(struct bpf_sock_ops_kern, sk)); 10190 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10191 offsetof(struct sock_common, skc_daddr)); 10192 break; 10193 10194 case offsetof(struct bpf_sock_ops, local_ip4): 10195 BUILD_BUG_ON(sizeof_field(struct sock_common, 10196 skc_rcv_saddr) != 4); 10197 10198 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10199 struct bpf_sock_ops_kern, sk), 10200 si->dst_reg, si->src_reg, 10201 offsetof(struct bpf_sock_ops_kern, sk)); 10202 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10203 offsetof(struct sock_common, 10204 skc_rcv_saddr)); 10205 break; 10206 10207 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 10208 offsetof(struct bpf_sock_ops, remote_ip6[3]): 10209 #if IS_ENABLED(CONFIG_IPV6) 10210 BUILD_BUG_ON(sizeof_field(struct sock_common, 10211 skc_v6_daddr.s6_addr32[0]) != 4); 10212 10213 off = si->off; 10214 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 10215 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10216 struct bpf_sock_ops_kern, sk), 10217 si->dst_reg, si->src_reg, 10218 offsetof(struct bpf_sock_ops_kern, sk)); 10219 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10220 offsetof(struct sock_common, 10221 skc_v6_daddr.s6_addr32[0]) + 10222 off); 10223 #else 10224 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10225 #endif 10226 break; 10227 10228 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 10229 offsetof(struct bpf_sock_ops, local_ip6[3]): 10230 #if IS_ENABLED(CONFIG_IPV6) 10231 BUILD_BUG_ON(sizeof_field(struct sock_common, 10232 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10233 10234 off = si->off; 10235 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 10236 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10237 struct bpf_sock_ops_kern, sk), 10238 si->dst_reg, si->src_reg, 10239 offsetof(struct bpf_sock_ops_kern, sk)); 10240 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10241 offsetof(struct sock_common, 10242 skc_v6_rcv_saddr.s6_addr32[0]) + 10243 off); 10244 #else 10245 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10246 #endif 10247 break; 10248 10249 case offsetof(struct bpf_sock_ops, remote_port): 10250 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10251 10252 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10253 struct bpf_sock_ops_kern, sk), 10254 si->dst_reg, si->src_reg, 10255 offsetof(struct bpf_sock_ops_kern, sk)); 10256 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10257 offsetof(struct sock_common, skc_dport)); 10258 #ifndef __BIG_ENDIAN_BITFIELD 10259 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10260 #endif 10261 break; 10262 10263 case offsetof(struct bpf_sock_ops, local_port): 10264 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10265 10266 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10267 struct bpf_sock_ops_kern, sk), 10268 si->dst_reg, si->src_reg, 10269 offsetof(struct bpf_sock_ops_kern, sk)); 10270 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10271 offsetof(struct sock_common, skc_num)); 10272 break; 10273 10274 case offsetof(struct bpf_sock_ops, is_fullsock): 10275 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10276 struct bpf_sock_ops_kern, 10277 is_fullsock), 10278 si->dst_reg, si->src_reg, 10279 offsetof(struct bpf_sock_ops_kern, 10280 is_fullsock)); 10281 break; 10282 10283 case offsetof(struct bpf_sock_ops, state): 10284 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 10285 10286 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10287 struct bpf_sock_ops_kern, sk), 10288 si->dst_reg, si->src_reg, 10289 offsetof(struct bpf_sock_ops_kern, sk)); 10290 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 10291 offsetof(struct sock_common, skc_state)); 10292 break; 10293 10294 case offsetof(struct bpf_sock_ops, rtt_min): 10295 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 10296 sizeof(struct minmax)); 10297 BUILD_BUG_ON(sizeof(struct minmax) < 10298 sizeof(struct minmax_sample)); 10299 10300 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10301 struct bpf_sock_ops_kern, sk), 10302 si->dst_reg, si->src_reg, 10303 offsetof(struct bpf_sock_ops_kern, sk)); 10304 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10305 offsetof(struct tcp_sock, rtt_min) + 10306 sizeof_field(struct minmax_sample, t)); 10307 break; 10308 10309 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 10310 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 10311 struct tcp_sock); 10312 break; 10313 10314 case offsetof(struct bpf_sock_ops, sk_txhash): 10315 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 10316 struct sock, type); 10317 break; 10318 case offsetof(struct bpf_sock_ops, snd_cwnd): 10319 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 10320 break; 10321 case offsetof(struct bpf_sock_ops, srtt_us): 10322 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 10323 break; 10324 case offsetof(struct bpf_sock_ops, snd_ssthresh): 10325 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 10326 break; 10327 case offsetof(struct bpf_sock_ops, rcv_nxt): 10328 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 10329 break; 10330 case offsetof(struct bpf_sock_ops, snd_nxt): 10331 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 10332 break; 10333 case offsetof(struct bpf_sock_ops, snd_una): 10334 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 10335 break; 10336 case offsetof(struct bpf_sock_ops, mss_cache): 10337 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 10338 break; 10339 case offsetof(struct bpf_sock_ops, ecn_flags): 10340 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 10341 break; 10342 case offsetof(struct bpf_sock_ops, rate_delivered): 10343 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 10344 break; 10345 case offsetof(struct bpf_sock_ops, rate_interval_us): 10346 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 10347 break; 10348 case offsetof(struct bpf_sock_ops, packets_out): 10349 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 10350 break; 10351 case offsetof(struct bpf_sock_ops, retrans_out): 10352 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 10353 break; 10354 case offsetof(struct bpf_sock_ops, total_retrans): 10355 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 10356 break; 10357 case offsetof(struct bpf_sock_ops, segs_in): 10358 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 10359 break; 10360 case offsetof(struct bpf_sock_ops, data_segs_in): 10361 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 10362 break; 10363 case offsetof(struct bpf_sock_ops, segs_out): 10364 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 10365 break; 10366 case offsetof(struct bpf_sock_ops, data_segs_out): 10367 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 10368 break; 10369 case offsetof(struct bpf_sock_ops, lost_out): 10370 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 10371 break; 10372 case offsetof(struct bpf_sock_ops, sacked_out): 10373 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 10374 break; 10375 case offsetof(struct bpf_sock_ops, bytes_received): 10376 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 10377 break; 10378 case offsetof(struct bpf_sock_ops, bytes_acked): 10379 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 10380 break; 10381 case offsetof(struct bpf_sock_ops, sk): 10382 SOCK_OPS_GET_SK(); 10383 break; 10384 case offsetof(struct bpf_sock_ops, skb_data_end): 10385 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10386 skb_data_end), 10387 si->dst_reg, si->src_reg, 10388 offsetof(struct bpf_sock_ops_kern, 10389 skb_data_end)); 10390 break; 10391 case offsetof(struct bpf_sock_ops, skb_data): 10392 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10393 skb), 10394 si->dst_reg, si->src_reg, 10395 offsetof(struct bpf_sock_ops_kern, 10396 skb)); 10397 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10398 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10399 si->dst_reg, si->dst_reg, 10400 offsetof(struct sk_buff, data)); 10401 break; 10402 case offsetof(struct bpf_sock_ops, skb_len): 10403 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10404 skb), 10405 si->dst_reg, si->src_reg, 10406 offsetof(struct bpf_sock_ops_kern, 10407 skb)); 10408 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10409 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10410 si->dst_reg, si->dst_reg, 10411 offsetof(struct sk_buff, len)); 10412 break; 10413 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 10414 off = offsetof(struct sk_buff, cb); 10415 off += offsetof(struct tcp_skb_cb, tcp_flags); 10416 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 10417 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10418 skb), 10419 si->dst_reg, si->src_reg, 10420 offsetof(struct bpf_sock_ops_kern, 10421 skb)); 10422 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10423 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 10424 tcp_flags), 10425 si->dst_reg, si->dst_reg, off); 10426 break; 10427 case offsetof(struct bpf_sock_ops, skb_hwtstamp): { 10428 struct bpf_insn *jmp_on_null_skb; 10429 10430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10431 skb), 10432 si->dst_reg, si->src_reg, 10433 offsetof(struct bpf_sock_ops_kern, 10434 skb)); 10435 /* Reserve one insn to test skb == NULL */ 10436 jmp_on_null_skb = insn++; 10437 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn); 10438 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 10439 bpf_target_off(struct skb_shared_info, 10440 hwtstamps, 8, 10441 target_size)); 10442 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 10443 insn - jmp_on_null_skb - 1); 10444 break; 10445 } 10446 } 10447 return insn - insn_buf; 10448 } 10449 10450 /* data_end = skb->data + skb_headlen() */ 10451 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si, 10452 struct bpf_insn *insn) 10453 { 10454 int reg; 10455 int temp_reg_off = offsetof(struct sk_buff, cb) + 10456 offsetof(struct sk_skb_cb, temp_reg); 10457 10458 if (si->src_reg == si->dst_reg) { 10459 /* We need an extra register, choose and save a register. */ 10460 reg = BPF_REG_9; 10461 if (si->src_reg == reg || si->dst_reg == reg) 10462 reg--; 10463 if (si->src_reg == reg || si->dst_reg == reg) 10464 reg--; 10465 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off); 10466 } else { 10467 reg = si->dst_reg; 10468 } 10469 10470 /* reg = skb->data */ 10471 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10472 reg, si->src_reg, 10473 offsetof(struct sk_buff, data)); 10474 /* AX = skb->len */ 10475 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10476 BPF_REG_AX, si->src_reg, 10477 offsetof(struct sk_buff, len)); 10478 /* reg = skb->data + skb->len */ 10479 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX); 10480 /* AX = skb->data_len */ 10481 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len), 10482 BPF_REG_AX, si->src_reg, 10483 offsetof(struct sk_buff, data_len)); 10484 10485 /* reg = skb->data + skb->len - skb->data_len */ 10486 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX); 10487 10488 if (si->src_reg == si->dst_reg) { 10489 /* Restore the saved register */ 10490 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg); 10491 *insn++ = BPF_MOV64_REG(si->dst_reg, reg); 10492 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off); 10493 } 10494 10495 return insn; 10496 } 10497 10498 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 10499 const struct bpf_insn *si, 10500 struct bpf_insn *insn_buf, 10501 struct bpf_prog *prog, u32 *target_size) 10502 { 10503 struct bpf_insn *insn = insn_buf; 10504 int off; 10505 10506 switch (si->off) { 10507 case offsetof(struct __sk_buff, data_end): 10508 insn = bpf_convert_data_end_access(si, insn); 10509 break; 10510 case offsetof(struct __sk_buff, cb[0]) ... 10511 offsetofend(struct __sk_buff, cb[4]) - 1: 10512 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20); 10513 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 10514 offsetof(struct sk_skb_cb, data)) % 10515 sizeof(__u64)); 10516 10517 prog->cb_access = 1; 10518 off = si->off; 10519 off -= offsetof(struct __sk_buff, cb[0]); 10520 off += offsetof(struct sk_buff, cb); 10521 off += offsetof(struct sk_skb_cb, data); 10522 if (type == BPF_WRITE) 10523 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 10524 si->src_reg, off); 10525 else 10526 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 10527 si->src_reg, off); 10528 break; 10529 10530 10531 default: 10532 return bpf_convert_ctx_access(type, si, insn_buf, prog, 10533 target_size); 10534 } 10535 10536 return insn - insn_buf; 10537 } 10538 10539 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 10540 const struct bpf_insn *si, 10541 struct bpf_insn *insn_buf, 10542 struct bpf_prog *prog, u32 *target_size) 10543 { 10544 struct bpf_insn *insn = insn_buf; 10545 #if IS_ENABLED(CONFIG_IPV6) 10546 int off; 10547 #endif 10548 10549 /* convert ctx uses the fact sg element is first in struct */ 10550 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 10551 10552 switch (si->off) { 10553 case offsetof(struct sk_msg_md, data): 10554 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 10555 si->dst_reg, si->src_reg, 10556 offsetof(struct sk_msg, data)); 10557 break; 10558 case offsetof(struct sk_msg_md, data_end): 10559 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 10560 si->dst_reg, si->src_reg, 10561 offsetof(struct sk_msg, data_end)); 10562 break; 10563 case offsetof(struct sk_msg_md, family): 10564 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10565 10566 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10567 struct sk_msg, sk), 10568 si->dst_reg, si->src_reg, 10569 offsetof(struct sk_msg, sk)); 10570 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10571 offsetof(struct sock_common, skc_family)); 10572 break; 10573 10574 case offsetof(struct sk_msg_md, remote_ip4): 10575 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10576 10577 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10578 struct sk_msg, sk), 10579 si->dst_reg, si->src_reg, 10580 offsetof(struct sk_msg, sk)); 10581 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10582 offsetof(struct sock_common, skc_daddr)); 10583 break; 10584 10585 case offsetof(struct sk_msg_md, local_ip4): 10586 BUILD_BUG_ON(sizeof_field(struct sock_common, 10587 skc_rcv_saddr) != 4); 10588 10589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10590 struct sk_msg, sk), 10591 si->dst_reg, si->src_reg, 10592 offsetof(struct sk_msg, sk)); 10593 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10594 offsetof(struct sock_common, 10595 skc_rcv_saddr)); 10596 break; 10597 10598 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 10599 offsetof(struct sk_msg_md, remote_ip6[3]): 10600 #if IS_ENABLED(CONFIG_IPV6) 10601 BUILD_BUG_ON(sizeof_field(struct sock_common, 10602 skc_v6_daddr.s6_addr32[0]) != 4); 10603 10604 off = si->off; 10605 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 10606 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10607 struct sk_msg, sk), 10608 si->dst_reg, si->src_reg, 10609 offsetof(struct sk_msg, sk)); 10610 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10611 offsetof(struct sock_common, 10612 skc_v6_daddr.s6_addr32[0]) + 10613 off); 10614 #else 10615 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10616 #endif 10617 break; 10618 10619 case offsetof(struct sk_msg_md, local_ip6[0]) ... 10620 offsetof(struct sk_msg_md, local_ip6[3]): 10621 #if IS_ENABLED(CONFIG_IPV6) 10622 BUILD_BUG_ON(sizeof_field(struct sock_common, 10623 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10624 10625 off = si->off; 10626 off -= offsetof(struct sk_msg_md, local_ip6[0]); 10627 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10628 struct sk_msg, sk), 10629 si->dst_reg, si->src_reg, 10630 offsetof(struct sk_msg, sk)); 10631 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10632 offsetof(struct sock_common, 10633 skc_v6_rcv_saddr.s6_addr32[0]) + 10634 off); 10635 #else 10636 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10637 #endif 10638 break; 10639 10640 case offsetof(struct sk_msg_md, remote_port): 10641 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10642 10643 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10644 struct sk_msg, sk), 10645 si->dst_reg, si->src_reg, 10646 offsetof(struct sk_msg, sk)); 10647 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10648 offsetof(struct sock_common, skc_dport)); 10649 #ifndef __BIG_ENDIAN_BITFIELD 10650 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10651 #endif 10652 break; 10653 10654 case offsetof(struct sk_msg_md, local_port): 10655 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10656 10657 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10658 struct sk_msg, sk), 10659 si->dst_reg, si->src_reg, 10660 offsetof(struct sk_msg, sk)); 10661 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10662 offsetof(struct sock_common, skc_num)); 10663 break; 10664 10665 case offsetof(struct sk_msg_md, size): 10666 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 10667 si->dst_reg, si->src_reg, 10668 offsetof(struct sk_msg_sg, size)); 10669 break; 10670 10671 case offsetof(struct sk_msg_md, sk): 10672 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 10673 si->dst_reg, si->src_reg, 10674 offsetof(struct sk_msg, sk)); 10675 break; 10676 } 10677 10678 return insn - insn_buf; 10679 } 10680 10681 const struct bpf_verifier_ops sk_filter_verifier_ops = { 10682 .get_func_proto = sk_filter_func_proto, 10683 .is_valid_access = sk_filter_is_valid_access, 10684 .convert_ctx_access = bpf_convert_ctx_access, 10685 .gen_ld_abs = bpf_gen_ld_abs, 10686 }; 10687 10688 const struct bpf_prog_ops sk_filter_prog_ops = { 10689 .test_run = bpf_prog_test_run_skb, 10690 }; 10691 10692 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 10693 .get_func_proto = tc_cls_act_func_proto, 10694 .is_valid_access = tc_cls_act_is_valid_access, 10695 .convert_ctx_access = tc_cls_act_convert_ctx_access, 10696 .gen_prologue = tc_cls_act_prologue, 10697 .gen_ld_abs = bpf_gen_ld_abs, 10698 .btf_struct_access = tc_cls_act_btf_struct_access, 10699 }; 10700 10701 const struct bpf_prog_ops tc_cls_act_prog_ops = { 10702 .test_run = bpf_prog_test_run_skb, 10703 }; 10704 10705 const struct bpf_verifier_ops xdp_verifier_ops = { 10706 .get_func_proto = xdp_func_proto, 10707 .is_valid_access = xdp_is_valid_access, 10708 .convert_ctx_access = xdp_convert_ctx_access, 10709 .gen_prologue = bpf_noop_prologue, 10710 .btf_struct_access = xdp_btf_struct_access, 10711 }; 10712 10713 const struct bpf_prog_ops xdp_prog_ops = { 10714 .test_run = bpf_prog_test_run_xdp, 10715 }; 10716 10717 const struct bpf_verifier_ops cg_skb_verifier_ops = { 10718 .get_func_proto = cg_skb_func_proto, 10719 .is_valid_access = cg_skb_is_valid_access, 10720 .convert_ctx_access = bpf_convert_ctx_access, 10721 }; 10722 10723 const struct bpf_prog_ops cg_skb_prog_ops = { 10724 .test_run = bpf_prog_test_run_skb, 10725 }; 10726 10727 const struct bpf_verifier_ops lwt_in_verifier_ops = { 10728 .get_func_proto = lwt_in_func_proto, 10729 .is_valid_access = lwt_is_valid_access, 10730 .convert_ctx_access = bpf_convert_ctx_access, 10731 }; 10732 10733 const struct bpf_prog_ops lwt_in_prog_ops = { 10734 .test_run = bpf_prog_test_run_skb, 10735 }; 10736 10737 const struct bpf_verifier_ops lwt_out_verifier_ops = { 10738 .get_func_proto = lwt_out_func_proto, 10739 .is_valid_access = lwt_is_valid_access, 10740 .convert_ctx_access = bpf_convert_ctx_access, 10741 }; 10742 10743 const struct bpf_prog_ops lwt_out_prog_ops = { 10744 .test_run = bpf_prog_test_run_skb, 10745 }; 10746 10747 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 10748 .get_func_proto = lwt_xmit_func_proto, 10749 .is_valid_access = lwt_is_valid_access, 10750 .convert_ctx_access = bpf_convert_ctx_access, 10751 .gen_prologue = tc_cls_act_prologue, 10752 }; 10753 10754 const struct bpf_prog_ops lwt_xmit_prog_ops = { 10755 .test_run = bpf_prog_test_run_skb, 10756 }; 10757 10758 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 10759 .get_func_proto = lwt_seg6local_func_proto, 10760 .is_valid_access = lwt_is_valid_access, 10761 .convert_ctx_access = bpf_convert_ctx_access, 10762 }; 10763 10764 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 10765 .test_run = bpf_prog_test_run_skb, 10766 }; 10767 10768 const struct bpf_verifier_ops cg_sock_verifier_ops = { 10769 .get_func_proto = sock_filter_func_proto, 10770 .is_valid_access = sock_filter_is_valid_access, 10771 .convert_ctx_access = bpf_sock_convert_ctx_access, 10772 }; 10773 10774 const struct bpf_prog_ops cg_sock_prog_ops = { 10775 }; 10776 10777 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 10778 .get_func_proto = sock_addr_func_proto, 10779 .is_valid_access = sock_addr_is_valid_access, 10780 .convert_ctx_access = sock_addr_convert_ctx_access, 10781 }; 10782 10783 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 10784 }; 10785 10786 const struct bpf_verifier_ops sock_ops_verifier_ops = { 10787 .get_func_proto = sock_ops_func_proto, 10788 .is_valid_access = sock_ops_is_valid_access, 10789 .convert_ctx_access = sock_ops_convert_ctx_access, 10790 }; 10791 10792 const struct bpf_prog_ops sock_ops_prog_ops = { 10793 }; 10794 10795 const struct bpf_verifier_ops sk_skb_verifier_ops = { 10796 .get_func_proto = sk_skb_func_proto, 10797 .is_valid_access = sk_skb_is_valid_access, 10798 .convert_ctx_access = sk_skb_convert_ctx_access, 10799 .gen_prologue = sk_skb_prologue, 10800 }; 10801 10802 const struct bpf_prog_ops sk_skb_prog_ops = { 10803 }; 10804 10805 const struct bpf_verifier_ops sk_msg_verifier_ops = { 10806 .get_func_proto = sk_msg_func_proto, 10807 .is_valid_access = sk_msg_is_valid_access, 10808 .convert_ctx_access = sk_msg_convert_ctx_access, 10809 .gen_prologue = bpf_noop_prologue, 10810 }; 10811 10812 const struct bpf_prog_ops sk_msg_prog_ops = { 10813 }; 10814 10815 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 10816 .get_func_proto = flow_dissector_func_proto, 10817 .is_valid_access = flow_dissector_is_valid_access, 10818 .convert_ctx_access = flow_dissector_convert_ctx_access, 10819 }; 10820 10821 const struct bpf_prog_ops flow_dissector_prog_ops = { 10822 .test_run = bpf_prog_test_run_flow_dissector, 10823 }; 10824 10825 int sk_detach_filter(struct sock *sk) 10826 { 10827 int ret = -ENOENT; 10828 struct sk_filter *filter; 10829 10830 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 10831 return -EPERM; 10832 10833 filter = rcu_dereference_protected(sk->sk_filter, 10834 lockdep_sock_is_held(sk)); 10835 if (filter) { 10836 RCU_INIT_POINTER(sk->sk_filter, NULL); 10837 sk_filter_uncharge(sk, filter); 10838 ret = 0; 10839 } 10840 10841 return ret; 10842 } 10843 EXPORT_SYMBOL_GPL(sk_detach_filter); 10844 10845 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len) 10846 { 10847 struct sock_fprog_kern *fprog; 10848 struct sk_filter *filter; 10849 int ret = 0; 10850 10851 sockopt_lock_sock(sk); 10852 filter = rcu_dereference_protected(sk->sk_filter, 10853 lockdep_sock_is_held(sk)); 10854 if (!filter) 10855 goto out; 10856 10857 /* We're copying the filter that has been originally attached, 10858 * so no conversion/decode needed anymore. eBPF programs that 10859 * have no original program cannot be dumped through this. 10860 */ 10861 ret = -EACCES; 10862 fprog = filter->prog->orig_prog; 10863 if (!fprog) 10864 goto out; 10865 10866 ret = fprog->len; 10867 if (!len) 10868 /* User space only enquires number of filter blocks. */ 10869 goto out; 10870 10871 ret = -EINVAL; 10872 if (len < fprog->len) 10873 goto out; 10874 10875 ret = -EFAULT; 10876 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog))) 10877 goto out; 10878 10879 /* Instead of bytes, the API requests to return the number 10880 * of filter blocks. 10881 */ 10882 ret = fprog->len; 10883 out: 10884 sockopt_release_sock(sk); 10885 return ret; 10886 } 10887 10888 #ifdef CONFIG_INET 10889 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 10890 struct sock_reuseport *reuse, 10891 struct sock *sk, struct sk_buff *skb, 10892 struct sock *migrating_sk, 10893 u32 hash) 10894 { 10895 reuse_kern->skb = skb; 10896 reuse_kern->sk = sk; 10897 reuse_kern->selected_sk = NULL; 10898 reuse_kern->migrating_sk = migrating_sk; 10899 reuse_kern->data_end = skb->data + skb_headlen(skb); 10900 reuse_kern->hash = hash; 10901 reuse_kern->reuseport_id = reuse->reuseport_id; 10902 reuse_kern->bind_inany = reuse->bind_inany; 10903 } 10904 10905 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 10906 struct bpf_prog *prog, struct sk_buff *skb, 10907 struct sock *migrating_sk, 10908 u32 hash) 10909 { 10910 struct sk_reuseport_kern reuse_kern; 10911 enum sk_action action; 10912 10913 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash); 10914 action = bpf_prog_run(prog, &reuse_kern); 10915 10916 if (action == SK_PASS) 10917 return reuse_kern.selected_sk; 10918 else 10919 return ERR_PTR(-ECONNREFUSED); 10920 } 10921 10922 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 10923 struct bpf_map *, map, void *, key, u32, flags) 10924 { 10925 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 10926 struct sock_reuseport *reuse; 10927 struct sock *selected_sk; 10928 10929 selected_sk = map->ops->map_lookup_elem(map, key); 10930 if (!selected_sk) 10931 return -ENOENT; 10932 10933 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 10934 if (!reuse) { 10935 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 10936 if (sk_is_refcounted(selected_sk)) 10937 sock_put(selected_sk); 10938 10939 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 10940 * The only (!reuse) case here is - the sk has already been 10941 * unhashed (e.g. by close()), so treat it as -ENOENT. 10942 * 10943 * Other maps (e.g. sock_map) do not provide this guarantee and 10944 * the sk may never be in the reuseport group to begin with. 10945 */ 10946 return is_sockarray ? -ENOENT : -EINVAL; 10947 } 10948 10949 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 10950 struct sock *sk = reuse_kern->sk; 10951 10952 if (sk->sk_protocol != selected_sk->sk_protocol) 10953 return -EPROTOTYPE; 10954 else if (sk->sk_family != selected_sk->sk_family) 10955 return -EAFNOSUPPORT; 10956 10957 /* Catch all. Likely bound to a different sockaddr. */ 10958 return -EBADFD; 10959 } 10960 10961 reuse_kern->selected_sk = selected_sk; 10962 10963 return 0; 10964 } 10965 10966 static const struct bpf_func_proto sk_select_reuseport_proto = { 10967 .func = sk_select_reuseport, 10968 .gpl_only = false, 10969 .ret_type = RET_INTEGER, 10970 .arg1_type = ARG_PTR_TO_CTX, 10971 .arg2_type = ARG_CONST_MAP_PTR, 10972 .arg3_type = ARG_PTR_TO_MAP_KEY, 10973 .arg4_type = ARG_ANYTHING, 10974 }; 10975 10976 BPF_CALL_4(sk_reuseport_load_bytes, 10977 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10978 void *, to, u32, len) 10979 { 10980 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 10981 } 10982 10983 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 10984 .func = sk_reuseport_load_bytes, 10985 .gpl_only = false, 10986 .ret_type = RET_INTEGER, 10987 .arg1_type = ARG_PTR_TO_CTX, 10988 .arg2_type = ARG_ANYTHING, 10989 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 10990 .arg4_type = ARG_CONST_SIZE, 10991 }; 10992 10993 BPF_CALL_5(sk_reuseport_load_bytes_relative, 10994 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10995 void *, to, u32, len, u32, start_header) 10996 { 10997 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 10998 len, start_header); 10999 } 11000 11001 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 11002 .func = sk_reuseport_load_bytes_relative, 11003 .gpl_only = false, 11004 .ret_type = RET_INTEGER, 11005 .arg1_type = ARG_PTR_TO_CTX, 11006 .arg2_type = ARG_ANYTHING, 11007 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 11008 .arg4_type = ARG_CONST_SIZE, 11009 .arg5_type = ARG_ANYTHING, 11010 }; 11011 11012 static const struct bpf_func_proto * 11013 sk_reuseport_func_proto(enum bpf_func_id func_id, 11014 const struct bpf_prog *prog) 11015 { 11016 switch (func_id) { 11017 case BPF_FUNC_sk_select_reuseport: 11018 return &sk_select_reuseport_proto; 11019 case BPF_FUNC_skb_load_bytes: 11020 return &sk_reuseport_load_bytes_proto; 11021 case BPF_FUNC_skb_load_bytes_relative: 11022 return &sk_reuseport_load_bytes_relative_proto; 11023 case BPF_FUNC_get_socket_cookie: 11024 return &bpf_get_socket_ptr_cookie_proto; 11025 case BPF_FUNC_ktime_get_coarse_ns: 11026 return &bpf_ktime_get_coarse_ns_proto; 11027 default: 11028 return bpf_base_func_proto(func_id); 11029 } 11030 } 11031 11032 static bool 11033 sk_reuseport_is_valid_access(int off, int size, 11034 enum bpf_access_type type, 11035 const struct bpf_prog *prog, 11036 struct bpf_insn_access_aux *info) 11037 { 11038 const u32 size_default = sizeof(__u32); 11039 11040 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 11041 off % size || type != BPF_READ) 11042 return false; 11043 11044 switch (off) { 11045 case offsetof(struct sk_reuseport_md, data): 11046 info->reg_type = PTR_TO_PACKET; 11047 return size == sizeof(__u64); 11048 11049 case offsetof(struct sk_reuseport_md, data_end): 11050 info->reg_type = PTR_TO_PACKET_END; 11051 return size == sizeof(__u64); 11052 11053 case offsetof(struct sk_reuseport_md, hash): 11054 return size == size_default; 11055 11056 case offsetof(struct sk_reuseport_md, sk): 11057 info->reg_type = PTR_TO_SOCKET; 11058 return size == sizeof(__u64); 11059 11060 case offsetof(struct sk_reuseport_md, migrating_sk): 11061 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 11062 return size == sizeof(__u64); 11063 11064 /* Fields that allow narrowing */ 11065 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 11066 if (size < sizeof_field(struct sk_buff, protocol)) 11067 return false; 11068 fallthrough; 11069 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 11070 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 11071 case bpf_ctx_range(struct sk_reuseport_md, len): 11072 bpf_ctx_record_field_size(info, size_default); 11073 return bpf_ctx_narrow_access_ok(off, size, size_default); 11074 11075 default: 11076 return false; 11077 } 11078 } 11079 11080 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 11081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 11082 si->dst_reg, si->src_reg, \ 11083 bpf_target_off(struct sk_reuseport_kern, F, \ 11084 sizeof_field(struct sk_reuseport_kern, F), \ 11085 target_size)); \ 11086 }) 11087 11088 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 11089 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11090 struct sk_buff, \ 11091 skb, \ 11092 SKB_FIELD) 11093 11094 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 11095 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11096 struct sock, \ 11097 sk, \ 11098 SK_FIELD) 11099 11100 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 11101 const struct bpf_insn *si, 11102 struct bpf_insn *insn_buf, 11103 struct bpf_prog *prog, 11104 u32 *target_size) 11105 { 11106 struct bpf_insn *insn = insn_buf; 11107 11108 switch (si->off) { 11109 case offsetof(struct sk_reuseport_md, data): 11110 SK_REUSEPORT_LOAD_SKB_FIELD(data); 11111 break; 11112 11113 case offsetof(struct sk_reuseport_md, len): 11114 SK_REUSEPORT_LOAD_SKB_FIELD(len); 11115 break; 11116 11117 case offsetof(struct sk_reuseport_md, eth_protocol): 11118 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 11119 break; 11120 11121 case offsetof(struct sk_reuseport_md, ip_protocol): 11122 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 11123 break; 11124 11125 case offsetof(struct sk_reuseport_md, data_end): 11126 SK_REUSEPORT_LOAD_FIELD(data_end); 11127 break; 11128 11129 case offsetof(struct sk_reuseport_md, hash): 11130 SK_REUSEPORT_LOAD_FIELD(hash); 11131 break; 11132 11133 case offsetof(struct sk_reuseport_md, bind_inany): 11134 SK_REUSEPORT_LOAD_FIELD(bind_inany); 11135 break; 11136 11137 case offsetof(struct sk_reuseport_md, sk): 11138 SK_REUSEPORT_LOAD_FIELD(sk); 11139 break; 11140 11141 case offsetof(struct sk_reuseport_md, migrating_sk): 11142 SK_REUSEPORT_LOAD_FIELD(migrating_sk); 11143 break; 11144 } 11145 11146 return insn - insn_buf; 11147 } 11148 11149 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 11150 .get_func_proto = sk_reuseport_func_proto, 11151 .is_valid_access = sk_reuseport_is_valid_access, 11152 .convert_ctx_access = sk_reuseport_convert_ctx_access, 11153 }; 11154 11155 const struct bpf_prog_ops sk_reuseport_prog_ops = { 11156 }; 11157 11158 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 11159 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 11160 11161 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 11162 struct sock *, sk, u64, flags) 11163 { 11164 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 11165 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 11166 return -EINVAL; 11167 if (unlikely(sk && sk_is_refcounted(sk))) 11168 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 11169 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN)) 11170 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */ 11171 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE)) 11172 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */ 11173 11174 /* Check if socket is suitable for packet L3/L4 protocol */ 11175 if (sk && sk->sk_protocol != ctx->protocol) 11176 return -EPROTOTYPE; 11177 if (sk && sk->sk_family != ctx->family && 11178 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 11179 return -EAFNOSUPPORT; 11180 11181 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 11182 return -EEXIST; 11183 11184 /* Select socket as lookup result */ 11185 ctx->selected_sk = sk; 11186 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 11187 return 0; 11188 } 11189 11190 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 11191 .func = bpf_sk_lookup_assign, 11192 .gpl_only = false, 11193 .ret_type = RET_INTEGER, 11194 .arg1_type = ARG_PTR_TO_CTX, 11195 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 11196 .arg3_type = ARG_ANYTHING, 11197 }; 11198 11199 static const struct bpf_func_proto * 11200 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 11201 { 11202 switch (func_id) { 11203 case BPF_FUNC_perf_event_output: 11204 return &bpf_event_output_data_proto; 11205 case BPF_FUNC_sk_assign: 11206 return &bpf_sk_lookup_assign_proto; 11207 case BPF_FUNC_sk_release: 11208 return &bpf_sk_release_proto; 11209 default: 11210 return bpf_sk_base_func_proto(func_id); 11211 } 11212 } 11213 11214 static bool sk_lookup_is_valid_access(int off, int size, 11215 enum bpf_access_type type, 11216 const struct bpf_prog *prog, 11217 struct bpf_insn_access_aux *info) 11218 { 11219 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 11220 return false; 11221 if (off % size != 0) 11222 return false; 11223 if (type != BPF_READ) 11224 return false; 11225 11226 switch (off) { 11227 case offsetof(struct bpf_sk_lookup, sk): 11228 info->reg_type = PTR_TO_SOCKET_OR_NULL; 11229 return size == sizeof(__u64); 11230 11231 case bpf_ctx_range(struct bpf_sk_lookup, family): 11232 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 11233 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 11234 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 11235 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 11236 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 11237 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 11238 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex): 11239 bpf_ctx_record_field_size(info, sizeof(__u32)); 11240 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 11241 11242 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 11243 /* Allow 4-byte access to 2-byte field for backward compatibility */ 11244 if (size == sizeof(__u32)) 11245 return true; 11246 bpf_ctx_record_field_size(info, sizeof(__be16)); 11247 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16)); 11248 11249 case offsetofend(struct bpf_sk_lookup, remote_port) ... 11250 offsetof(struct bpf_sk_lookup, local_ip4) - 1: 11251 /* Allow access to zero padding for backward compatibility */ 11252 bpf_ctx_record_field_size(info, sizeof(__u16)); 11253 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16)); 11254 11255 default: 11256 return false; 11257 } 11258 } 11259 11260 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 11261 const struct bpf_insn *si, 11262 struct bpf_insn *insn_buf, 11263 struct bpf_prog *prog, 11264 u32 *target_size) 11265 { 11266 struct bpf_insn *insn = insn_buf; 11267 11268 switch (si->off) { 11269 case offsetof(struct bpf_sk_lookup, sk): 11270 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11271 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 11272 break; 11273 11274 case offsetof(struct bpf_sk_lookup, family): 11275 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11276 bpf_target_off(struct bpf_sk_lookup_kern, 11277 family, 2, target_size)); 11278 break; 11279 11280 case offsetof(struct bpf_sk_lookup, protocol): 11281 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11282 bpf_target_off(struct bpf_sk_lookup_kern, 11283 protocol, 2, target_size)); 11284 break; 11285 11286 case offsetof(struct bpf_sk_lookup, remote_ip4): 11287 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11288 bpf_target_off(struct bpf_sk_lookup_kern, 11289 v4.saddr, 4, target_size)); 11290 break; 11291 11292 case offsetof(struct bpf_sk_lookup, local_ip4): 11293 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11294 bpf_target_off(struct bpf_sk_lookup_kern, 11295 v4.daddr, 4, target_size)); 11296 break; 11297 11298 case bpf_ctx_range_till(struct bpf_sk_lookup, 11299 remote_ip6[0], remote_ip6[3]): { 11300 #if IS_ENABLED(CONFIG_IPV6) 11301 int off = si->off; 11302 11303 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 11304 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11305 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11306 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 11307 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11308 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11309 #else 11310 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11311 #endif 11312 break; 11313 } 11314 case bpf_ctx_range_till(struct bpf_sk_lookup, 11315 local_ip6[0], local_ip6[3]): { 11316 #if IS_ENABLED(CONFIG_IPV6) 11317 int off = si->off; 11318 11319 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 11320 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11321 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11322 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 11323 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11324 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11325 #else 11326 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11327 #endif 11328 break; 11329 } 11330 case offsetof(struct bpf_sk_lookup, remote_port): 11331 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11332 bpf_target_off(struct bpf_sk_lookup_kern, 11333 sport, 2, target_size)); 11334 break; 11335 11336 case offsetofend(struct bpf_sk_lookup, remote_port): 11337 *target_size = 2; 11338 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11339 break; 11340 11341 case offsetof(struct bpf_sk_lookup, local_port): 11342 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11343 bpf_target_off(struct bpf_sk_lookup_kern, 11344 dport, 2, target_size)); 11345 break; 11346 11347 case offsetof(struct bpf_sk_lookup, ingress_ifindex): 11348 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11349 bpf_target_off(struct bpf_sk_lookup_kern, 11350 ingress_ifindex, 4, target_size)); 11351 break; 11352 } 11353 11354 return insn - insn_buf; 11355 } 11356 11357 const struct bpf_prog_ops sk_lookup_prog_ops = { 11358 .test_run = bpf_prog_test_run_sk_lookup, 11359 }; 11360 11361 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 11362 .get_func_proto = sk_lookup_func_proto, 11363 .is_valid_access = sk_lookup_is_valid_access, 11364 .convert_ctx_access = sk_lookup_convert_ctx_access, 11365 }; 11366 11367 #endif /* CONFIG_INET */ 11368 11369 DEFINE_BPF_DISPATCHER(xdp) 11370 11371 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 11372 { 11373 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 11374 } 11375 11376 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE) 11377 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 11378 BTF_SOCK_TYPE_xxx 11379 #undef BTF_SOCK_TYPE 11380 11381 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 11382 { 11383 /* tcp6_sock type is not generated in dwarf and hence btf, 11384 * trigger an explicit type generation here. 11385 */ 11386 BTF_TYPE_EMIT(struct tcp6_sock); 11387 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 11388 sk->sk_family == AF_INET6) 11389 return (unsigned long)sk; 11390 11391 return (unsigned long)NULL; 11392 } 11393 11394 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 11395 .func = bpf_skc_to_tcp6_sock, 11396 .gpl_only = false, 11397 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11398 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11399 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 11400 }; 11401 11402 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 11403 { 11404 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 11405 return (unsigned long)sk; 11406 11407 return (unsigned long)NULL; 11408 } 11409 11410 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 11411 .func = bpf_skc_to_tcp_sock, 11412 .gpl_only = false, 11413 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11414 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11415 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 11416 }; 11417 11418 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 11419 { 11420 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 11421 * generated if CONFIG_INET=n. Trigger an explicit generation here. 11422 */ 11423 BTF_TYPE_EMIT(struct inet_timewait_sock); 11424 BTF_TYPE_EMIT(struct tcp_timewait_sock); 11425 11426 #ifdef CONFIG_INET 11427 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 11428 return (unsigned long)sk; 11429 #endif 11430 11431 #if IS_BUILTIN(CONFIG_IPV6) 11432 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 11433 return (unsigned long)sk; 11434 #endif 11435 11436 return (unsigned long)NULL; 11437 } 11438 11439 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 11440 .func = bpf_skc_to_tcp_timewait_sock, 11441 .gpl_only = false, 11442 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11443 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11444 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 11445 }; 11446 11447 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 11448 { 11449 #ifdef CONFIG_INET 11450 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11451 return (unsigned long)sk; 11452 #endif 11453 11454 #if IS_BUILTIN(CONFIG_IPV6) 11455 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11456 return (unsigned long)sk; 11457 #endif 11458 11459 return (unsigned long)NULL; 11460 } 11461 11462 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 11463 .func = bpf_skc_to_tcp_request_sock, 11464 .gpl_only = false, 11465 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11466 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11467 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 11468 }; 11469 11470 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 11471 { 11472 /* udp6_sock type is not generated in dwarf and hence btf, 11473 * trigger an explicit type generation here. 11474 */ 11475 BTF_TYPE_EMIT(struct udp6_sock); 11476 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 11477 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 11478 return (unsigned long)sk; 11479 11480 return (unsigned long)NULL; 11481 } 11482 11483 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 11484 .func = bpf_skc_to_udp6_sock, 11485 .gpl_only = false, 11486 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11487 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11488 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 11489 }; 11490 11491 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk) 11492 { 11493 /* unix_sock type is not generated in dwarf and hence btf, 11494 * trigger an explicit type generation here. 11495 */ 11496 BTF_TYPE_EMIT(struct unix_sock); 11497 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX) 11498 return (unsigned long)sk; 11499 11500 return (unsigned long)NULL; 11501 } 11502 11503 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { 11504 .func = bpf_skc_to_unix_sock, 11505 .gpl_only = false, 11506 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11507 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11508 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX], 11509 }; 11510 11511 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) 11512 { 11513 BTF_TYPE_EMIT(struct mptcp_sock); 11514 return (unsigned long)bpf_mptcp_sock_from_subflow(sk); 11515 } 11516 11517 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { 11518 .func = bpf_skc_to_mptcp_sock, 11519 .gpl_only = false, 11520 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11521 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 11522 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP], 11523 }; 11524 11525 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 11526 { 11527 return (unsigned long)sock_from_file(file); 11528 } 11529 11530 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 11531 BTF_ID(struct, socket) 11532 BTF_ID(struct, file) 11533 11534 const struct bpf_func_proto bpf_sock_from_file_proto = { 11535 .func = bpf_sock_from_file, 11536 .gpl_only = false, 11537 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11538 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 11539 .arg1_type = ARG_PTR_TO_BTF_ID, 11540 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 11541 }; 11542 11543 static const struct bpf_func_proto * 11544 bpf_sk_base_func_proto(enum bpf_func_id func_id) 11545 { 11546 const struct bpf_func_proto *func; 11547 11548 switch (func_id) { 11549 case BPF_FUNC_skc_to_tcp6_sock: 11550 func = &bpf_skc_to_tcp6_sock_proto; 11551 break; 11552 case BPF_FUNC_skc_to_tcp_sock: 11553 func = &bpf_skc_to_tcp_sock_proto; 11554 break; 11555 case BPF_FUNC_skc_to_tcp_timewait_sock: 11556 func = &bpf_skc_to_tcp_timewait_sock_proto; 11557 break; 11558 case BPF_FUNC_skc_to_tcp_request_sock: 11559 func = &bpf_skc_to_tcp_request_sock_proto; 11560 break; 11561 case BPF_FUNC_skc_to_udp6_sock: 11562 func = &bpf_skc_to_udp6_sock_proto; 11563 break; 11564 case BPF_FUNC_skc_to_unix_sock: 11565 func = &bpf_skc_to_unix_sock_proto; 11566 break; 11567 case BPF_FUNC_skc_to_mptcp_sock: 11568 func = &bpf_skc_to_mptcp_sock_proto; 11569 break; 11570 case BPF_FUNC_ktime_get_coarse_ns: 11571 return &bpf_ktime_get_coarse_ns_proto; 11572 default: 11573 return bpf_base_func_proto(func_id); 11574 } 11575 11576 if (!perfmon_capable()) 11577 return NULL; 11578 11579 return func; 11580 } 11581