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 void *old_data; 3184 3185 /* skb_ensure_writable() is not needed here, as we're 3186 * already working on an uncloned skb. 3187 */ 3188 if (unlikely(!pskb_may_pull(skb, off + len))) 3189 return -ENOMEM; 3190 3191 old_data = skb->data; 3192 __skb_pull(skb, len); 3193 skb_postpull_rcsum(skb, old_data + off, len); 3194 memmove(skb->data, old_data, off); 3195 3196 return 0; 3197 } 3198 3199 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 3200 { 3201 bool trans_same = skb->transport_header == skb->network_header; 3202 int ret; 3203 3204 /* There's no need for __skb_push()/__skb_pull() pair to 3205 * get to the start of the mac header as we're guaranteed 3206 * to always start from here under eBPF. 3207 */ 3208 ret = bpf_skb_generic_push(skb, off, len); 3209 if (likely(!ret)) { 3210 skb->mac_header -= len; 3211 skb->network_header -= len; 3212 if (trans_same) 3213 skb->transport_header = skb->network_header; 3214 } 3215 3216 return ret; 3217 } 3218 3219 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 3220 { 3221 bool trans_same = skb->transport_header == skb->network_header; 3222 int ret; 3223 3224 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 3225 ret = bpf_skb_generic_pop(skb, off, len); 3226 if (likely(!ret)) { 3227 skb->mac_header += len; 3228 skb->network_header += len; 3229 if (trans_same) 3230 skb->transport_header = skb->network_header; 3231 } 3232 3233 return ret; 3234 } 3235 3236 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 3237 { 3238 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3239 u32 off = skb_mac_header_len(skb); 3240 int ret; 3241 3242 ret = skb_cow(skb, len_diff); 3243 if (unlikely(ret < 0)) 3244 return ret; 3245 3246 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3247 if (unlikely(ret < 0)) 3248 return ret; 3249 3250 if (skb_is_gso(skb)) { 3251 struct skb_shared_info *shinfo = skb_shinfo(skb); 3252 3253 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */ 3254 if (shinfo->gso_type & SKB_GSO_TCPV4) { 3255 shinfo->gso_type &= ~SKB_GSO_TCPV4; 3256 shinfo->gso_type |= SKB_GSO_TCPV6; 3257 } 3258 } 3259 3260 skb->protocol = htons(ETH_P_IPV6); 3261 skb_clear_hash(skb); 3262 3263 return 0; 3264 } 3265 3266 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 3267 { 3268 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3269 u32 off = skb_mac_header_len(skb); 3270 int ret; 3271 3272 ret = skb_unclone(skb, GFP_ATOMIC); 3273 if (unlikely(ret < 0)) 3274 return ret; 3275 3276 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3277 if (unlikely(ret < 0)) 3278 return ret; 3279 3280 if (skb_is_gso(skb)) { 3281 struct skb_shared_info *shinfo = skb_shinfo(skb); 3282 3283 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */ 3284 if (shinfo->gso_type & SKB_GSO_TCPV6) { 3285 shinfo->gso_type &= ~SKB_GSO_TCPV6; 3286 shinfo->gso_type |= SKB_GSO_TCPV4; 3287 } 3288 } 3289 3290 skb->protocol = htons(ETH_P_IP); 3291 skb_clear_hash(skb); 3292 3293 return 0; 3294 } 3295 3296 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 3297 { 3298 __be16 from_proto = skb->protocol; 3299 3300 if (from_proto == htons(ETH_P_IP) && 3301 to_proto == htons(ETH_P_IPV6)) 3302 return bpf_skb_proto_4_to_6(skb); 3303 3304 if (from_proto == htons(ETH_P_IPV6) && 3305 to_proto == htons(ETH_P_IP)) 3306 return bpf_skb_proto_6_to_4(skb); 3307 3308 return -ENOTSUPP; 3309 } 3310 3311 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 3312 u64, flags) 3313 { 3314 int ret; 3315 3316 if (unlikely(flags)) 3317 return -EINVAL; 3318 3319 /* General idea is that this helper does the basic groundwork 3320 * needed for changing the protocol, and eBPF program fills the 3321 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 3322 * and other helpers, rather than passing a raw buffer here. 3323 * 3324 * The rationale is to keep this minimal and without a need to 3325 * deal with raw packet data. F.e. even if we would pass buffers 3326 * here, the program still needs to call the bpf_lX_csum_replace() 3327 * helpers anyway. Plus, this way we keep also separation of 3328 * concerns, since f.e. bpf_skb_store_bytes() should only take 3329 * care of stores. 3330 * 3331 * Currently, additional options and extension header space are 3332 * not supported, but flags register is reserved so we can adapt 3333 * that. For offloads, we mark packet as dodgy, so that headers 3334 * need to be verified first. 3335 */ 3336 ret = bpf_skb_proto_xlat(skb, proto); 3337 bpf_compute_data_pointers(skb); 3338 return ret; 3339 } 3340 3341 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 3342 .func = bpf_skb_change_proto, 3343 .gpl_only = false, 3344 .ret_type = RET_INTEGER, 3345 .arg1_type = ARG_PTR_TO_CTX, 3346 .arg2_type = ARG_ANYTHING, 3347 .arg3_type = ARG_ANYTHING, 3348 }; 3349 3350 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 3351 { 3352 /* We only allow a restricted subset to be changed for now. */ 3353 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 3354 !skb_pkt_type_ok(pkt_type))) 3355 return -EINVAL; 3356 3357 skb->pkt_type = pkt_type; 3358 return 0; 3359 } 3360 3361 static const struct bpf_func_proto bpf_skb_change_type_proto = { 3362 .func = bpf_skb_change_type, 3363 .gpl_only = false, 3364 .ret_type = RET_INTEGER, 3365 .arg1_type = ARG_PTR_TO_CTX, 3366 .arg2_type = ARG_ANYTHING, 3367 }; 3368 3369 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3370 { 3371 switch (skb->protocol) { 3372 case htons(ETH_P_IP): 3373 return sizeof(struct iphdr); 3374 case htons(ETH_P_IPV6): 3375 return sizeof(struct ipv6hdr); 3376 default: 3377 return ~0U; 3378 } 3379 } 3380 3381 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3382 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3383 3384 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \ 3385 BPF_F_ADJ_ROOM_DECAP_L3_IPV6) 3386 3387 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3388 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3389 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3390 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3391 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \ 3392 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3393 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \ 3394 BPF_F_ADJ_ROOM_DECAP_L3_MASK) 3395 3396 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3397 u64 flags) 3398 { 3399 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3400 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3401 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3402 unsigned int gso_type = SKB_GSO_DODGY; 3403 int ret; 3404 3405 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3406 /* udp gso_size delineates datagrams, only allow if fixed */ 3407 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3408 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3409 return -ENOTSUPP; 3410 } 3411 3412 ret = skb_cow_head(skb, len_diff); 3413 if (unlikely(ret < 0)) 3414 return ret; 3415 3416 if (encap) { 3417 if (skb->protocol != htons(ETH_P_IP) && 3418 skb->protocol != htons(ETH_P_IPV6)) 3419 return -ENOTSUPP; 3420 3421 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3422 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3423 return -EINVAL; 3424 3425 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3426 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3427 return -EINVAL; 3428 3429 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH && 3430 inner_mac_len < ETH_HLEN) 3431 return -EINVAL; 3432 3433 if (skb->encapsulation) 3434 return -EALREADY; 3435 3436 mac_len = skb->network_header - skb->mac_header; 3437 inner_net = skb->network_header; 3438 if (inner_mac_len > len_diff) 3439 return -EINVAL; 3440 inner_trans = skb->transport_header; 3441 } 3442 3443 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3444 if (unlikely(ret < 0)) 3445 return ret; 3446 3447 if (encap) { 3448 skb->inner_mac_header = inner_net - inner_mac_len; 3449 skb->inner_network_header = inner_net; 3450 skb->inner_transport_header = inner_trans; 3451 3452 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH) 3453 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 3454 else 3455 skb_set_inner_protocol(skb, skb->protocol); 3456 3457 skb->encapsulation = 1; 3458 skb_set_network_header(skb, mac_len); 3459 3460 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3461 gso_type |= SKB_GSO_UDP_TUNNEL; 3462 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3463 gso_type |= SKB_GSO_GRE; 3464 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3465 gso_type |= SKB_GSO_IPXIP6; 3466 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3467 gso_type |= SKB_GSO_IPXIP4; 3468 3469 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3470 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3471 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3472 sizeof(struct ipv6hdr) : 3473 sizeof(struct iphdr); 3474 3475 skb_set_transport_header(skb, mac_len + nh_len); 3476 } 3477 3478 /* Match skb->protocol to new outer l3 protocol */ 3479 if (skb->protocol == htons(ETH_P_IP) && 3480 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3481 skb->protocol = htons(ETH_P_IPV6); 3482 else if (skb->protocol == htons(ETH_P_IPV6) && 3483 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3484 skb->protocol = htons(ETH_P_IP); 3485 } 3486 3487 if (skb_is_gso(skb)) { 3488 struct skb_shared_info *shinfo = skb_shinfo(skb); 3489 3490 /* Due to header grow, MSS needs to be downgraded. */ 3491 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3492 skb_decrease_gso_size(shinfo, len_diff); 3493 3494 /* Header must be checked, and gso_segs recomputed. */ 3495 shinfo->gso_type |= gso_type; 3496 shinfo->gso_segs = 0; 3497 } 3498 3499 return 0; 3500 } 3501 3502 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3503 u64 flags) 3504 { 3505 int ret; 3506 3507 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO | 3508 BPF_F_ADJ_ROOM_DECAP_L3_MASK | 3509 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3510 return -EINVAL; 3511 3512 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3513 /* udp gso_size delineates datagrams, only allow if fixed */ 3514 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3515 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3516 return -ENOTSUPP; 3517 } 3518 3519 ret = skb_unclone(skb, GFP_ATOMIC); 3520 if (unlikely(ret < 0)) 3521 return ret; 3522 3523 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3524 if (unlikely(ret < 0)) 3525 return ret; 3526 3527 /* Match skb->protocol to new outer l3 protocol */ 3528 if (skb->protocol == htons(ETH_P_IP) && 3529 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6) 3530 skb->protocol = htons(ETH_P_IPV6); 3531 else if (skb->protocol == htons(ETH_P_IPV6) && 3532 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4) 3533 skb->protocol = htons(ETH_P_IP); 3534 3535 if (skb_is_gso(skb)) { 3536 struct skb_shared_info *shinfo = skb_shinfo(skb); 3537 3538 /* Due to header shrink, MSS can be upgraded. */ 3539 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3540 skb_increase_gso_size(shinfo, len_diff); 3541 3542 /* Header must be checked, and gso_segs recomputed. */ 3543 shinfo->gso_type |= SKB_GSO_DODGY; 3544 shinfo->gso_segs = 0; 3545 } 3546 3547 return 0; 3548 } 3549 3550 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC 3551 3552 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3553 u32, mode, u64, flags) 3554 { 3555 u32 len_diff_abs = abs(len_diff); 3556 bool shrink = len_diff < 0; 3557 int ret = 0; 3558 3559 if (unlikely(flags || mode)) 3560 return -EINVAL; 3561 if (unlikely(len_diff_abs > 0xfffU)) 3562 return -EFAULT; 3563 3564 if (!shrink) { 3565 ret = skb_cow(skb, len_diff); 3566 if (unlikely(ret < 0)) 3567 return ret; 3568 __skb_push(skb, len_diff_abs); 3569 memset(skb->data, 0, len_diff_abs); 3570 } else { 3571 if (unlikely(!pskb_may_pull(skb, len_diff_abs))) 3572 return -ENOMEM; 3573 __skb_pull(skb, len_diff_abs); 3574 } 3575 if (tls_sw_has_ctx_rx(skb->sk)) { 3576 struct strp_msg *rxm = strp_msg(skb); 3577 3578 rxm->full_len += len_diff; 3579 } 3580 return ret; 3581 } 3582 3583 static const struct bpf_func_proto sk_skb_adjust_room_proto = { 3584 .func = sk_skb_adjust_room, 3585 .gpl_only = false, 3586 .ret_type = RET_INTEGER, 3587 .arg1_type = ARG_PTR_TO_CTX, 3588 .arg2_type = ARG_ANYTHING, 3589 .arg3_type = ARG_ANYTHING, 3590 .arg4_type = ARG_ANYTHING, 3591 }; 3592 3593 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3594 u32, mode, u64, flags) 3595 { 3596 u32 len_cur, len_diff_abs = abs(len_diff); 3597 u32 len_min = bpf_skb_net_base_len(skb); 3598 u32 len_max = BPF_SKB_MAX_LEN; 3599 __be16 proto = skb->protocol; 3600 bool shrink = len_diff < 0; 3601 u32 off; 3602 int ret; 3603 3604 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK | 3605 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3606 return -EINVAL; 3607 if (unlikely(len_diff_abs > 0xfffU)) 3608 return -EFAULT; 3609 if (unlikely(proto != htons(ETH_P_IP) && 3610 proto != htons(ETH_P_IPV6))) 3611 return -ENOTSUPP; 3612 3613 off = skb_mac_header_len(skb); 3614 switch (mode) { 3615 case BPF_ADJ_ROOM_NET: 3616 off += bpf_skb_net_base_len(skb); 3617 break; 3618 case BPF_ADJ_ROOM_MAC: 3619 break; 3620 default: 3621 return -ENOTSUPP; 3622 } 3623 3624 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) { 3625 if (!shrink) 3626 return -EINVAL; 3627 3628 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) { 3629 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4: 3630 len_min = sizeof(struct iphdr); 3631 break; 3632 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6: 3633 len_min = sizeof(struct ipv6hdr); 3634 break; 3635 default: 3636 return -EINVAL; 3637 } 3638 } 3639 3640 len_cur = skb->len - skb_network_offset(skb); 3641 if ((shrink && (len_diff_abs >= len_cur || 3642 len_cur - len_diff_abs < len_min)) || 3643 (!shrink && (skb->len + len_diff_abs > len_max && 3644 !skb_is_gso(skb)))) 3645 return -ENOTSUPP; 3646 3647 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3648 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3649 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET)) 3650 __skb_reset_checksum_unnecessary(skb); 3651 3652 bpf_compute_data_pointers(skb); 3653 return ret; 3654 } 3655 3656 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3657 .func = bpf_skb_adjust_room, 3658 .gpl_only = false, 3659 .ret_type = RET_INTEGER, 3660 .arg1_type = ARG_PTR_TO_CTX, 3661 .arg2_type = ARG_ANYTHING, 3662 .arg3_type = ARG_ANYTHING, 3663 .arg4_type = ARG_ANYTHING, 3664 }; 3665 3666 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3667 { 3668 u32 min_len = skb_network_offset(skb); 3669 3670 if (skb_transport_header_was_set(skb)) 3671 min_len = skb_transport_offset(skb); 3672 if (skb->ip_summed == CHECKSUM_PARTIAL) 3673 min_len = skb_checksum_start_offset(skb) + 3674 skb->csum_offset + sizeof(__sum16); 3675 return min_len; 3676 } 3677 3678 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3679 { 3680 unsigned int old_len = skb->len; 3681 int ret; 3682 3683 ret = __skb_grow_rcsum(skb, new_len); 3684 if (!ret) 3685 memset(skb->data + old_len, 0, new_len - old_len); 3686 return ret; 3687 } 3688 3689 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3690 { 3691 return __skb_trim_rcsum(skb, new_len); 3692 } 3693 3694 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3695 u64 flags) 3696 { 3697 u32 max_len = BPF_SKB_MAX_LEN; 3698 u32 min_len = __bpf_skb_min_len(skb); 3699 int ret; 3700 3701 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3702 return -EINVAL; 3703 if (skb->encapsulation) 3704 return -ENOTSUPP; 3705 3706 /* The basic idea of this helper is that it's performing the 3707 * needed work to either grow or trim an skb, and eBPF program 3708 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3709 * bpf_lX_csum_replace() and others rather than passing a raw 3710 * buffer here. This one is a slow path helper and intended 3711 * for replies with control messages. 3712 * 3713 * Like in bpf_skb_change_proto(), we want to keep this rather 3714 * minimal and without protocol specifics so that we are able 3715 * to separate concerns as in bpf_skb_store_bytes() should only 3716 * be the one responsible for writing buffers. 3717 * 3718 * It's really expected to be a slow path operation here for 3719 * control message replies, so we're implicitly linearizing, 3720 * uncloning and drop offloads from the skb by this. 3721 */ 3722 ret = __bpf_try_make_writable(skb, skb->len); 3723 if (!ret) { 3724 if (new_len > skb->len) 3725 ret = bpf_skb_grow_rcsum(skb, new_len); 3726 else if (new_len < skb->len) 3727 ret = bpf_skb_trim_rcsum(skb, new_len); 3728 if (!ret && skb_is_gso(skb)) 3729 skb_gso_reset(skb); 3730 } 3731 return ret; 3732 } 3733 3734 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3735 u64, flags) 3736 { 3737 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3738 3739 bpf_compute_data_pointers(skb); 3740 return ret; 3741 } 3742 3743 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3744 .func = bpf_skb_change_tail, 3745 .gpl_only = false, 3746 .ret_type = RET_INTEGER, 3747 .arg1_type = ARG_PTR_TO_CTX, 3748 .arg2_type = ARG_ANYTHING, 3749 .arg3_type = ARG_ANYTHING, 3750 }; 3751 3752 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3753 u64, flags) 3754 { 3755 return __bpf_skb_change_tail(skb, new_len, flags); 3756 } 3757 3758 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3759 .func = sk_skb_change_tail, 3760 .gpl_only = false, 3761 .ret_type = RET_INTEGER, 3762 .arg1_type = ARG_PTR_TO_CTX, 3763 .arg2_type = ARG_ANYTHING, 3764 .arg3_type = ARG_ANYTHING, 3765 }; 3766 3767 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3768 u64 flags) 3769 { 3770 u32 max_len = BPF_SKB_MAX_LEN; 3771 u32 new_len = skb->len + head_room; 3772 int ret; 3773 3774 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3775 new_len < skb->len)) 3776 return -EINVAL; 3777 3778 ret = skb_cow(skb, head_room); 3779 if (likely(!ret)) { 3780 /* Idea for this helper is that we currently only 3781 * allow to expand on mac header. This means that 3782 * skb->protocol network header, etc, stay as is. 3783 * Compared to bpf_skb_change_tail(), we're more 3784 * flexible due to not needing to linearize or 3785 * reset GSO. Intention for this helper is to be 3786 * used by an L3 skb that needs to push mac header 3787 * for redirection into L2 device. 3788 */ 3789 __skb_push(skb, head_room); 3790 memset(skb->data, 0, head_room); 3791 skb_reset_mac_header(skb); 3792 skb_reset_mac_len(skb); 3793 } 3794 3795 return ret; 3796 } 3797 3798 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3799 u64, flags) 3800 { 3801 int ret = __bpf_skb_change_head(skb, head_room, flags); 3802 3803 bpf_compute_data_pointers(skb); 3804 return ret; 3805 } 3806 3807 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3808 .func = bpf_skb_change_head, 3809 .gpl_only = false, 3810 .ret_type = RET_INTEGER, 3811 .arg1_type = ARG_PTR_TO_CTX, 3812 .arg2_type = ARG_ANYTHING, 3813 .arg3_type = ARG_ANYTHING, 3814 }; 3815 3816 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3817 u64, flags) 3818 { 3819 return __bpf_skb_change_head(skb, head_room, flags); 3820 } 3821 3822 static const struct bpf_func_proto sk_skb_change_head_proto = { 3823 .func = sk_skb_change_head, 3824 .gpl_only = false, 3825 .ret_type = RET_INTEGER, 3826 .arg1_type = ARG_PTR_TO_CTX, 3827 .arg2_type = ARG_ANYTHING, 3828 .arg3_type = ARG_ANYTHING, 3829 }; 3830 3831 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp) 3832 { 3833 return xdp_get_buff_len(xdp); 3834 } 3835 3836 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = { 3837 .func = bpf_xdp_get_buff_len, 3838 .gpl_only = false, 3839 .ret_type = RET_INTEGER, 3840 .arg1_type = ARG_PTR_TO_CTX, 3841 }; 3842 3843 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff) 3844 3845 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = { 3846 .func = bpf_xdp_get_buff_len, 3847 .gpl_only = false, 3848 .arg1_type = ARG_PTR_TO_BTF_ID, 3849 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0], 3850 }; 3851 3852 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3853 { 3854 return xdp_data_meta_unsupported(xdp) ? 0 : 3855 xdp->data - xdp->data_meta; 3856 } 3857 3858 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3859 { 3860 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3861 unsigned long metalen = xdp_get_metalen(xdp); 3862 void *data_start = xdp_frame_end + metalen; 3863 void *data = xdp->data + offset; 3864 3865 if (unlikely(data < data_start || 3866 data > xdp->data_end - ETH_HLEN)) 3867 return -EINVAL; 3868 3869 if (metalen) 3870 memmove(xdp->data_meta + offset, 3871 xdp->data_meta, metalen); 3872 xdp->data_meta += offset; 3873 xdp->data = data; 3874 3875 return 0; 3876 } 3877 3878 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3879 .func = bpf_xdp_adjust_head, 3880 .gpl_only = false, 3881 .ret_type = RET_INTEGER, 3882 .arg1_type = ARG_PTR_TO_CTX, 3883 .arg2_type = ARG_ANYTHING, 3884 }; 3885 3886 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 3887 void *buf, unsigned long len, bool flush) 3888 { 3889 unsigned long ptr_len, ptr_off = 0; 3890 skb_frag_t *next_frag, *end_frag; 3891 struct skb_shared_info *sinfo; 3892 void *src, *dst; 3893 u8 *ptr_buf; 3894 3895 if (likely(xdp->data_end - xdp->data >= off + len)) { 3896 src = flush ? buf : xdp->data + off; 3897 dst = flush ? xdp->data + off : buf; 3898 memcpy(dst, src, len); 3899 return; 3900 } 3901 3902 sinfo = xdp_get_shared_info_from_buff(xdp); 3903 end_frag = &sinfo->frags[sinfo->nr_frags]; 3904 next_frag = &sinfo->frags[0]; 3905 3906 ptr_len = xdp->data_end - xdp->data; 3907 ptr_buf = xdp->data; 3908 3909 while (true) { 3910 if (off < ptr_off + ptr_len) { 3911 unsigned long copy_off = off - ptr_off; 3912 unsigned long copy_len = min(len, ptr_len - copy_off); 3913 3914 src = flush ? buf : ptr_buf + copy_off; 3915 dst = flush ? ptr_buf + copy_off : buf; 3916 memcpy(dst, src, copy_len); 3917 3918 off += copy_len; 3919 len -= copy_len; 3920 buf += copy_len; 3921 } 3922 3923 if (!len || next_frag == end_frag) 3924 break; 3925 3926 ptr_off += ptr_len; 3927 ptr_buf = skb_frag_address(next_frag); 3928 ptr_len = skb_frag_size(next_frag); 3929 next_frag++; 3930 } 3931 } 3932 3933 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 3934 { 3935 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 3936 u32 size = xdp->data_end - xdp->data; 3937 void *addr = xdp->data; 3938 int i; 3939 3940 if (unlikely(offset > 0xffff || len > 0xffff)) 3941 return ERR_PTR(-EFAULT); 3942 3943 if (offset + len > xdp_get_buff_len(xdp)) 3944 return ERR_PTR(-EINVAL); 3945 3946 if (offset < size) /* linear area */ 3947 goto out; 3948 3949 offset -= size; 3950 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */ 3951 u32 frag_size = skb_frag_size(&sinfo->frags[i]); 3952 3953 if (offset < frag_size) { 3954 addr = skb_frag_address(&sinfo->frags[i]); 3955 size = frag_size; 3956 break; 3957 } 3958 offset -= frag_size; 3959 } 3960 out: 3961 return offset + len <= size ? addr + offset : NULL; 3962 } 3963 3964 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset, 3965 void *, buf, u32, len) 3966 { 3967 void *ptr; 3968 3969 ptr = bpf_xdp_pointer(xdp, offset, len); 3970 if (IS_ERR(ptr)) 3971 return PTR_ERR(ptr); 3972 3973 if (!ptr) 3974 bpf_xdp_copy_buf(xdp, offset, buf, len, false); 3975 else 3976 memcpy(buf, ptr, len); 3977 3978 return 0; 3979 } 3980 3981 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = { 3982 .func = bpf_xdp_load_bytes, 3983 .gpl_only = false, 3984 .ret_type = RET_INTEGER, 3985 .arg1_type = ARG_PTR_TO_CTX, 3986 .arg2_type = ARG_ANYTHING, 3987 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 3988 .arg4_type = ARG_CONST_SIZE, 3989 }; 3990 3991 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset, 3992 void *, buf, u32, len) 3993 { 3994 void *ptr; 3995 3996 ptr = bpf_xdp_pointer(xdp, offset, len); 3997 if (IS_ERR(ptr)) 3998 return PTR_ERR(ptr); 3999 4000 if (!ptr) 4001 bpf_xdp_copy_buf(xdp, offset, buf, len, true); 4002 else 4003 memcpy(ptr, buf, len); 4004 4005 return 0; 4006 } 4007 4008 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = { 4009 .func = bpf_xdp_store_bytes, 4010 .gpl_only = false, 4011 .ret_type = RET_INTEGER, 4012 .arg1_type = ARG_PTR_TO_CTX, 4013 .arg2_type = ARG_ANYTHING, 4014 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 4015 .arg4_type = ARG_CONST_SIZE, 4016 }; 4017 4018 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset) 4019 { 4020 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4021 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1]; 4022 struct xdp_rxq_info *rxq = xdp->rxq; 4023 unsigned int tailroom; 4024 4025 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz) 4026 return -EOPNOTSUPP; 4027 4028 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag); 4029 if (unlikely(offset > tailroom)) 4030 return -EINVAL; 4031 4032 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset); 4033 skb_frag_size_add(frag, offset); 4034 sinfo->xdp_frags_size += offset; 4035 4036 return 0; 4037 } 4038 4039 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset) 4040 { 4041 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4042 int i, n_frags_free = 0, len_free = 0; 4043 4044 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN)) 4045 return -EINVAL; 4046 4047 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) { 4048 skb_frag_t *frag = &sinfo->frags[i]; 4049 int shrink = min_t(int, offset, skb_frag_size(frag)); 4050 4051 len_free += shrink; 4052 offset -= shrink; 4053 4054 if (skb_frag_size(frag) == shrink) { 4055 struct page *page = skb_frag_page(frag); 4056 4057 __xdp_return(page_address(page), &xdp->rxq->mem, 4058 false, NULL); 4059 n_frags_free++; 4060 } else { 4061 skb_frag_size_sub(frag, shrink); 4062 break; 4063 } 4064 } 4065 sinfo->nr_frags -= n_frags_free; 4066 sinfo->xdp_frags_size -= len_free; 4067 4068 if (unlikely(!sinfo->nr_frags)) { 4069 xdp_buff_clear_frags_flag(xdp); 4070 xdp->data_end -= offset; 4071 } 4072 4073 return 0; 4074 } 4075 4076 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 4077 { 4078 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */ 4079 void *data_end = xdp->data_end + offset; 4080 4081 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */ 4082 if (offset < 0) 4083 return bpf_xdp_frags_shrink_tail(xdp, -offset); 4084 4085 return bpf_xdp_frags_increase_tail(xdp, offset); 4086 } 4087 4088 /* Notice that xdp_data_hard_end have reserved some tailroom */ 4089 if (unlikely(data_end > data_hard_end)) 4090 return -EINVAL; 4091 4092 /* ALL drivers MUST init xdp->frame_sz, chicken check below */ 4093 if (unlikely(xdp->frame_sz > PAGE_SIZE)) { 4094 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz); 4095 return -EINVAL; 4096 } 4097 4098 if (unlikely(data_end < xdp->data + ETH_HLEN)) 4099 return -EINVAL; 4100 4101 /* Clear memory area on grow, can contain uninit kernel memory */ 4102 if (offset > 0) 4103 memset(xdp->data_end, 0, offset); 4104 4105 xdp->data_end = data_end; 4106 4107 return 0; 4108 } 4109 4110 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 4111 .func = bpf_xdp_adjust_tail, 4112 .gpl_only = false, 4113 .ret_type = RET_INTEGER, 4114 .arg1_type = ARG_PTR_TO_CTX, 4115 .arg2_type = ARG_ANYTHING, 4116 }; 4117 4118 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 4119 { 4120 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 4121 void *meta = xdp->data_meta + offset; 4122 unsigned long metalen = xdp->data - meta; 4123 4124 if (xdp_data_meta_unsupported(xdp)) 4125 return -ENOTSUPP; 4126 if (unlikely(meta < xdp_frame_end || 4127 meta > xdp->data)) 4128 return -EINVAL; 4129 if (unlikely(xdp_metalen_invalid(metalen))) 4130 return -EACCES; 4131 4132 xdp->data_meta = meta; 4133 4134 return 0; 4135 } 4136 4137 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 4138 .func = bpf_xdp_adjust_meta, 4139 .gpl_only = false, 4140 .ret_type = RET_INTEGER, 4141 .arg1_type = ARG_PTR_TO_CTX, 4142 .arg2_type = ARG_ANYTHING, 4143 }; 4144 4145 /** 4146 * DOC: xdp redirect 4147 * 4148 * XDP_REDIRECT works by a three-step process, implemented in the functions 4149 * below: 4150 * 4151 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target 4152 * of the redirect and store it (along with some other metadata) in a per-CPU 4153 * struct bpf_redirect_info. 4154 * 4155 * 2. When the program returns the XDP_REDIRECT return code, the driver will 4156 * call xdp_do_redirect() which will use the information in struct 4157 * bpf_redirect_info to actually enqueue the frame into a map type-specific 4158 * bulk queue structure. 4159 * 4160 * 3. Before exiting its NAPI poll loop, the driver will call 4161 * xdp_do_flush(), which will flush all the different bulk queues, 4162 * thus completing the redirect. Note that xdp_do_flush() must be 4163 * called before napi_complete_done() in the driver, as the 4164 * XDP_REDIRECT logic relies on being inside a single NAPI instance 4165 * through to the xdp_do_flush() call for RCU protection of all 4166 * in-kernel data structures. 4167 */ 4168 /* 4169 * Pointers to the map entries will be kept around for this whole sequence of 4170 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in 4171 * the core code; instead, the RCU protection relies on everything happening 4172 * inside a single NAPI poll sequence, which means it's between a pair of calls 4173 * to local_bh_disable()/local_bh_enable(). 4174 * 4175 * The map entries are marked as __rcu and the map code makes sure to 4176 * dereference those pointers with rcu_dereference_check() in a way that works 4177 * for both sections that to hold an rcu_read_lock() and sections that are 4178 * called from NAPI without a separate rcu_read_lock(). The code below does not 4179 * use RCU annotations, but relies on those in the map code. 4180 */ 4181 void xdp_do_flush(void) 4182 { 4183 __dev_flush(); 4184 __cpu_map_flush(); 4185 __xsk_map_flush(); 4186 } 4187 EXPORT_SYMBOL_GPL(xdp_do_flush); 4188 4189 void bpf_clear_redirect_map(struct bpf_map *map) 4190 { 4191 struct bpf_redirect_info *ri; 4192 int cpu; 4193 4194 for_each_possible_cpu(cpu) { 4195 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 4196 /* Avoid polluting remote cacheline due to writes if 4197 * not needed. Once we pass this test, we need the 4198 * cmpxchg() to make sure it hasn't been changed in 4199 * the meantime by remote CPU. 4200 */ 4201 if (unlikely(READ_ONCE(ri->map) == map)) 4202 cmpxchg(&ri->map, map, NULL); 4203 } 4204 } 4205 4206 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 4207 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key); 4208 4209 u32 xdp_master_redirect(struct xdp_buff *xdp) 4210 { 4211 struct net_device *master, *slave; 4212 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4213 4214 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev); 4215 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp); 4216 if (slave && slave != xdp->rxq->dev) { 4217 /* The target device is different from the receiving device, so 4218 * redirect it to the new device. 4219 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled 4220 * drivers to unmap the packet from their rx ring. 4221 */ 4222 ri->tgt_index = slave->ifindex; 4223 ri->map_id = INT_MAX; 4224 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4225 return XDP_REDIRECT; 4226 } 4227 return XDP_TX; 4228 } 4229 EXPORT_SYMBOL_GPL(xdp_master_redirect); 4230 4231 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri, 4232 struct net_device *dev, 4233 struct xdp_buff *xdp, 4234 struct bpf_prog *xdp_prog) 4235 { 4236 enum bpf_map_type map_type = ri->map_type; 4237 void *fwd = ri->tgt_value; 4238 u32 map_id = ri->map_id; 4239 int err; 4240 4241 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4242 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4243 4244 err = __xsk_map_redirect(fwd, xdp); 4245 if (unlikely(err)) 4246 goto err; 4247 4248 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4249 return 0; 4250 err: 4251 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4252 return err; 4253 } 4254 4255 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri, 4256 struct net_device *dev, 4257 struct xdp_frame *xdpf, 4258 struct bpf_prog *xdp_prog) 4259 { 4260 enum bpf_map_type map_type = ri->map_type; 4261 void *fwd = ri->tgt_value; 4262 u32 map_id = ri->map_id; 4263 struct bpf_map *map; 4264 int err; 4265 4266 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4267 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4268 4269 if (unlikely(!xdpf)) { 4270 err = -EOVERFLOW; 4271 goto err; 4272 } 4273 4274 switch (map_type) { 4275 case BPF_MAP_TYPE_DEVMAP: 4276 fallthrough; 4277 case BPF_MAP_TYPE_DEVMAP_HASH: 4278 map = READ_ONCE(ri->map); 4279 if (unlikely(map)) { 4280 WRITE_ONCE(ri->map, NULL); 4281 err = dev_map_enqueue_multi(xdpf, dev, map, 4282 ri->flags & BPF_F_EXCLUDE_INGRESS); 4283 } else { 4284 err = dev_map_enqueue(fwd, xdpf, dev); 4285 } 4286 break; 4287 case BPF_MAP_TYPE_CPUMAP: 4288 err = cpu_map_enqueue(fwd, xdpf, dev); 4289 break; 4290 case BPF_MAP_TYPE_UNSPEC: 4291 if (map_id == INT_MAX) { 4292 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4293 if (unlikely(!fwd)) { 4294 err = -EINVAL; 4295 break; 4296 } 4297 err = dev_xdp_enqueue(fwd, xdpf, dev); 4298 break; 4299 } 4300 fallthrough; 4301 default: 4302 err = -EBADRQC; 4303 } 4304 4305 if (unlikely(err)) 4306 goto err; 4307 4308 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4309 return 0; 4310 err: 4311 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4312 return err; 4313 } 4314 4315 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 4316 struct bpf_prog *xdp_prog) 4317 { 4318 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4319 enum bpf_map_type map_type = ri->map_type; 4320 4321 if (map_type == BPF_MAP_TYPE_XSKMAP) { 4322 /* XDP_REDIRECT is not supported AF_XDP yet. */ 4323 if (unlikely(xdp_buff_has_frags(xdp))) 4324 return -EOPNOTSUPP; 4325 4326 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4327 } 4328 4329 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp), 4330 xdp_prog); 4331 } 4332 EXPORT_SYMBOL_GPL(xdp_do_redirect); 4333 4334 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp, 4335 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog) 4336 { 4337 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4338 enum bpf_map_type map_type = ri->map_type; 4339 4340 if (map_type == BPF_MAP_TYPE_XSKMAP) 4341 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4342 4343 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog); 4344 } 4345 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame); 4346 4347 static int xdp_do_generic_redirect_map(struct net_device *dev, 4348 struct sk_buff *skb, 4349 struct xdp_buff *xdp, 4350 struct bpf_prog *xdp_prog, 4351 void *fwd, 4352 enum bpf_map_type map_type, u32 map_id) 4353 { 4354 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4355 struct bpf_map *map; 4356 int err; 4357 4358 switch (map_type) { 4359 case BPF_MAP_TYPE_DEVMAP: 4360 fallthrough; 4361 case BPF_MAP_TYPE_DEVMAP_HASH: 4362 map = READ_ONCE(ri->map); 4363 if (unlikely(map)) { 4364 WRITE_ONCE(ri->map, NULL); 4365 err = dev_map_redirect_multi(dev, skb, xdp_prog, map, 4366 ri->flags & BPF_F_EXCLUDE_INGRESS); 4367 } else { 4368 err = dev_map_generic_redirect(fwd, skb, xdp_prog); 4369 } 4370 if (unlikely(err)) 4371 goto err; 4372 break; 4373 case BPF_MAP_TYPE_XSKMAP: 4374 err = xsk_generic_rcv(fwd, xdp); 4375 if (err) 4376 goto err; 4377 consume_skb(skb); 4378 break; 4379 case BPF_MAP_TYPE_CPUMAP: 4380 err = cpu_map_generic_redirect(fwd, skb); 4381 if (unlikely(err)) 4382 goto err; 4383 break; 4384 default: 4385 err = -EBADRQC; 4386 goto err; 4387 } 4388 4389 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4390 return 0; 4391 err: 4392 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4393 return err; 4394 } 4395 4396 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 4397 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 4398 { 4399 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4400 enum bpf_map_type map_type = ri->map_type; 4401 void *fwd = ri->tgt_value; 4402 u32 map_id = ri->map_id; 4403 int err; 4404 4405 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4406 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4407 4408 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) { 4409 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4410 if (unlikely(!fwd)) { 4411 err = -EINVAL; 4412 goto err; 4413 } 4414 4415 err = xdp_ok_fwd_dev(fwd, skb->len); 4416 if (unlikely(err)) 4417 goto err; 4418 4419 skb->dev = fwd; 4420 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index); 4421 generic_xdp_tx(skb, xdp_prog); 4422 return 0; 4423 } 4424 4425 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id); 4426 err: 4427 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err); 4428 return err; 4429 } 4430 4431 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 4432 { 4433 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4434 4435 if (unlikely(flags)) 4436 return XDP_ABORTED; 4437 4438 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated 4439 * by map_idr) is used for ifindex based XDP redirect. 4440 */ 4441 ri->tgt_index = ifindex; 4442 ri->map_id = INT_MAX; 4443 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4444 4445 return XDP_REDIRECT; 4446 } 4447 4448 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 4449 .func = bpf_xdp_redirect, 4450 .gpl_only = false, 4451 .ret_type = RET_INTEGER, 4452 .arg1_type = ARG_ANYTHING, 4453 .arg2_type = ARG_ANYTHING, 4454 }; 4455 4456 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key, 4457 u64, flags) 4458 { 4459 return map->ops->map_redirect(map, key, flags); 4460 } 4461 4462 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 4463 .func = bpf_xdp_redirect_map, 4464 .gpl_only = false, 4465 .ret_type = RET_INTEGER, 4466 .arg1_type = ARG_CONST_MAP_PTR, 4467 .arg2_type = ARG_ANYTHING, 4468 .arg3_type = ARG_ANYTHING, 4469 }; 4470 4471 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 4472 unsigned long off, unsigned long len) 4473 { 4474 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 4475 4476 if (unlikely(!ptr)) 4477 return len; 4478 if (ptr != dst_buff) 4479 memcpy(dst_buff, ptr, len); 4480 4481 return 0; 4482 } 4483 4484 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 4485 u64, flags, void *, meta, u64, meta_size) 4486 { 4487 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4488 4489 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4490 return -EINVAL; 4491 if (unlikely(!skb || skb_size > skb->len)) 4492 return -EFAULT; 4493 4494 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 4495 bpf_skb_copy); 4496 } 4497 4498 static const struct bpf_func_proto bpf_skb_event_output_proto = { 4499 .func = bpf_skb_event_output, 4500 .gpl_only = true, 4501 .ret_type = RET_INTEGER, 4502 .arg1_type = ARG_PTR_TO_CTX, 4503 .arg2_type = ARG_CONST_MAP_PTR, 4504 .arg3_type = ARG_ANYTHING, 4505 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4506 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4507 }; 4508 4509 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff) 4510 4511 const struct bpf_func_proto bpf_skb_output_proto = { 4512 .func = bpf_skb_event_output, 4513 .gpl_only = true, 4514 .ret_type = RET_INTEGER, 4515 .arg1_type = ARG_PTR_TO_BTF_ID, 4516 .arg1_btf_id = &bpf_skb_output_btf_ids[0], 4517 .arg2_type = ARG_CONST_MAP_PTR, 4518 .arg3_type = ARG_ANYTHING, 4519 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4520 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4521 }; 4522 4523 static unsigned short bpf_tunnel_key_af(u64 flags) 4524 { 4525 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 4526 } 4527 4528 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 4529 u32, size, u64, flags) 4530 { 4531 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4532 u8 compat[sizeof(struct bpf_tunnel_key)]; 4533 void *to_orig = to; 4534 int err; 4535 4536 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 | 4537 BPF_F_TUNINFO_FLAGS)))) { 4538 err = -EINVAL; 4539 goto err_clear; 4540 } 4541 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4542 err = -EPROTO; 4543 goto err_clear; 4544 } 4545 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4546 err = -EINVAL; 4547 switch (size) { 4548 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4549 case offsetof(struct bpf_tunnel_key, tunnel_label): 4550 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4551 goto set_compat; 4552 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4553 /* Fixup deprecated structure layouts here, so we have 4554 * a common path later on. 4555 */ 4556 if (ip_tunnel_info_af(info) != AF_INET) 4557 goto err_clear; 4558 set_compat: 4559 to = (struct bpf_tunnel_key *)compat; 4560 break; 4561 default: 4562 goto err_clear; 4563 } 4564 } 4565 4566 to->tunnel_id = be64_to_cpu(info->key.tun_id); 4567 to->tunnel_tos = info->key.tos; 4568 to->tunnel_ttl = info->key.ttl; 4569 if (flags & BPF_F_TUNINFO_FLAGS) 4570 to->tunnel_flags = info->key.tun_flags; 4571 else 4572 to->tunnel_ext = 0; 4573 4574 if (flags & BPF_F_TUNINFO_IPV6) { 4575 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 4576 sizeof(to->remote_ipv6)); 4577 memcpy(to->local_ipv6, &info->key.u.ipv6.dst, 4578 sizeof(to->local_ipv6)); 4579 to->tunnel_label = be32_to_cpu(info->key.label); 4580 } else { 4581 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 4582 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4583 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst); 4584 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3); 4585 to->tunnel_label = 0; 4586 } 4587 4588 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 4589 memcpy(to_orig, to, size); 4590 4591 return 0; 4592 err_clear: 4593 memset(to_orig, 0, size); 4594 return err; 4595 } 4596 4597 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 4598 .func = bpf_skb_get_tunnel_key, 4599 .gpl_only = false, 4600 .ret_type = RET_INTEGER, 4601 .arg1_type = ARG_PTR_TO_CTX, 4602 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4603 .arg3_type = ARG_CONST_SIZE, 4604 .arg4_type = ARG_ANYTHING, 4605 }; 4606 4607 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 4608 { 4609 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4610 int err; 4611 4612 if (unlikely(!info || 4613 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 4614 err = -ENOENT; 4615 goto err_clear; 4616 } 4617 if (unlikely(size < info->options_len)) { 4618 err = -ENOMEM; 4619 goto err_clear; 4620 } 4621 4622 ip_tunnel_info_opts_get(to, info); 4623 if (size > info->options_len) 4624 memset(to + info->options_len, 0, size - info->options_len); 4625 4626 return info->options_len; 4627 err_clear: 4628 memset(to, 0, size); 4629 return err; 4630 } 4631 4632 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 4633 .func = bpf_skb_get_tunnel_opt, 4634 .gpl_only = false, 4635 .ret_type = RET_INTEGER, 4636 .arg1_type = ARG_PTR_TO_CTX, 4637 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4638 .arg3_type = ARG_CONST_SIZE, 4639 }; 4640 4641 static struct metadata_dst __percpu *md_dst; 4642 4643 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 4644 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 4645 { 4646 struct metadata_dst *md = this_cpu_ptr(md_dst); 4647 u8 compat[sizeof(struct bpf_tunnel_key)]; 4648 struct ip_tunnel_info *info; 4649 4650 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 4651 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER | 4652 BPF_F_NO_TUNNEL_KEY))) 4653 return -EINVAL; 4654 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4655 switch (size) { 4656 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4657 case offsetof(struct bpf_tunnel_key, tunnel_label): 4658 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4659 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4660 /* Fixup deprecated structure layouts here, so we have 4661 * a common path later on. 4662 */ 4663 memcpy(compat, from, size); 4664 memset(compat + size, 0, sizeof(compat) - size); 4665 from = (const struct bpf_tunnel_key *) compat; 4666 break; 4667 default: 4668 return -EINVAL; 4669 } 4670 } 4671 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 4672 from->tunnel_ext)) 4673 return -EINVAL; 4674 4675 skb_dst_drop(skb); 4676 dst_hold((struct dst_entry *) md); 4677 skb_dst_set(skb, (struct dst_entry *) md); 4678 4679 info = &md->u.tun_info; 4680 memset(info, 0, sizeof(*info)); 4681 info->mode = IP_TUNNEL_INFO_TX; 4682 4683 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 4684 if (flags & BPF_F_DONT_FRAGMENT) 4685 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 4686 if (flags & BPF_F_ZERO_CSUM_TX) 4687 info->key.tun_flags &= ~TUNNEL_CSUM; 4688 if (flags & BPF_F_SEQ_NUMBER) 4689 info->key.tun_flags |= TUNNEL_SEQ; 4690 if (flags & BPF_F_NO_TUNNEL_KEY) 4691 info->key.tun_flags &= ~TUNNEL_KEY; 4692 4693 info->key.tun_id = cpu_to_be64(from->tunnel_id); 4694 info->key.tos = from->tunnel_tos; 4695 info->key.ttl = from->tunnel_ttl; 4696 4697 if (flags & BPF_F_TUNINFO_IPV6) { 4698 info->mode |= IP_TUNNEL_INFO_IPV6; 4699 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 4700 sizeof(from->remote_ipv6)); 4701 memcpy(&info->key.u.ipv6.src, from->local_ipv6, 4702 sizeof(from->local_ipv6)); 4703 info->key.label = cpu_to_be32(from->tunnel_label) & 4704 IPV6_FLOWLABEL_MASK; 4705 } else { 4706 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 4707 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4); 4708 info->key.flow_flags = FLOWI_FLAG_ANYSRC; 4709 } 4710 4711 return 0; 4712 } 4713 4714 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 4715 .func = bpf_skb_set_tunnel_key, 4716 .gpl_only = false, 4717 .ret_type = RET_INTEGER, 4718 .arg1_type = ARG_PTR_TO_CTX, 4719 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4720 .arg3_type = ARG_CONST_SIZE, 4721 .arg4_type = ARG_ANYTHING, 4722 }; 4723 4724 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4725 const u8 *, from, u32, size) 4726 { 4727 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4728 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4729 4730 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4731 return -EINVAL; 4732 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4733 return -ENOMEM; 4734 4735 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4736 4737 return 0; 4738 } 4739 4740 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4741 .func = bpf_skb_set_tunnel_opt, 4742 .gpl_only = false, 4743 .ret_type = RET_INTEGER, 4744 .arg1_type = ARG_PTR_TO_CTX, 4745 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4746 .arg3_type = ARG_CONST_SIZE, 4747 }; 4748 4749 static const struct bpf_func_proto * 4750 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4751 { 4752 if (!md_dst) { 4753 struct metadata_dst __percpu *tmp; 4754 4755 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4756 METADATA_IP_TUNNEL, 4757 GFP_KERNEL); 4758 if (!tmp) 4759 return NULL; 4760 if (cmpxchg(&md_dst, NULL, tmp)) 4761 metadata_dst_free_percpu(tmp); 4762 } 4763 4764 switch (which) { 4765 case BPF_FUNC_skb_set_tunnel_key: 4766 return &bpf_skb_set_tunnel_key_proto; 4767 case BPF_FUNC_skb_set_tunnel_opt: 4768 return &bpf_skb_set_tunnel_opt_proto; 4769 default: 4770 return NULL; 4771 } 4772 } 4773 4774 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4775 u32, idx) 4776 { 4777 struct bpf_array *array = container_of(map, struct bpf_array, map); 4778 struct cgroup *cgrp; 4779 struct sock *sk; 4780 4781 sk = skb_to_full_sk(skb); 4782 if (!sk || !sk_fullsock(sk)) 4783 return -ENOENT; 4784 if (unlikely(idx >= array->map.max_entries)) 4785 return -E2BIG; 4786 4787 cgrp = READ_ONCE(array->ptrs[idx]); 4788 if (unlikely(!cgrp)) 4789 return -EAGAIN; 4790 4791 return sk_under_cgroup_hierarchy(sk, cgrp); 4792 } 4793 4794 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4795 .func = bpf_skb_under_cgroup, 4796 .gpl_only = false, 4797 .ret_type = RET_INTEGER, 4798 .arg1_type = ARG_PTR_TO_CTX, 4799 .arg2_type = ARG_CONST_MAP_PTR, 4800 .arg3_type = ARG_ANYTHING, 4801 }; 4802 4803 #ifdef CONFIG_SOCK_CGROUP_DATA 4804 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 4805 { 4806 struct cgroup *cgrp; 4807 4808 sk = sk_to_full_sk(sk); 4809 if (!sk || !sk_fullsock(sk)) 4810 return 0; 4811 4812 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4813 return cgroup_id(cgrp); 4814 } 4815 4816 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4817 { 4818 return __bpf_sk_cgroup_id(skb->sk); 4819 } 4820 4821 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4822 .func = bpf_skb_cgroup_id, 4823 .gpl_only = false, 4824 .ret_type = RET_INTEGER, 4825 .arg1_type = ARG_PTR_TO_CTX, 4826 }; 4827 4828 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 4829 int ancestor_level) 4830 { 4831 struct cgroup *ancestor; 4832 struct cgroup *cgrp; 4833 4834 sk = sk_to_full_sk(sk); 4835 if (!sk || !sk_fullsock(sk)) 4836 return 0; 4837 4838 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4839 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4840 if (!ancestor) 4841 return 0; 4842 4843 return cgroup_id(ancestor); 4844 } 4845 4846 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4847 ancestor_level) 4848 { 4849 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 4850 } 4851 4852 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4853 .func = bpf_skb_ancestor_cgroup_id, 4854 .gpl_only = false, 4855 .ret_type = RET_INTEGER, 4856 .arg1_type = ARG_PTR_TO_CTX, 4857 .arg2_type = ARG_ANYTHING, 4858 }; 4859 4860 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 4861 { 4862 return __bpf_sk_cgroup_id(sk); 4863 } 4864 4865 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 4866 .func = bpf_sk_cgroup_id, 4867 .gpl_only = false, 4868 .ret_type = RET_INTEGER, 4869 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4870 }; 4871 4872 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 4873 { 4874 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 4875 } 4876 4877 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 4878 .func = bpf_sk_ancestor_cgroup_id, 4879 .gpl_only = false, 4880 .ret_type = RET_INTEGER, 4881 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4882 .arg2_type = ARG_ANYTHING, 4883 }; 4884 #endif 4885 4886 static unsigned long bpf_xdp_copy(void *dst, const void *ctx, 4887 unsigned long off, unsigned long len) 4888 { 4889 struct xdp_buff *xdp = (struct xdp_buff *)ctx; 4890 4891 bpf_xdp_copy_buf(xdp, off, dst, len, false); 4892 return 0; 4893 } 4894 4895 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4896 u64, flags, void *, meta, u64, meta_size) 4897 { 4898 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4899 4900 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4901 return -EINVAL; 4902 4903 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp))) 4904 return -EFAULT; 4905 4906 return bpf_event_output(map, flags, meta, meta_size, xdp, 4907 xdp_size, bpf_xdp_copy); 4908 } 4909 4910 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4911 .func = bpf_xdp_event_output, 4912 .gpl_only = true, 4913 .ret_type = RET_INTEGER, 4914 .arg1_type = ARG_PTR_TO_CTX, 4915 .arg2_type = ARG_CONST_MAP_PTR, 4916 .arg3_type = ARG_ANYTHING, 4917 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4918 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4919 }; 4920 4921 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 4922 4923 const struct bpf_func_proto bpf_xdp_output_proto = { 4924 .func = bpf_xdp_event_output, 4925 .gpl_only = true, 4926 .ret_type = RET_INTEGER, 4927 .arg1_type = ARG_PTR_TO_BTF_ID, 4928 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 4929 .arg2_type = ARG_CONST_MAP_PTR, 4930 .arg3_type = ARG_ANYTHING, 4931 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4932 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4933 }; 4934 4935 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4936 { 4937 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 4938 } 4939 4940 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4941 .func = bpf_get_socket_cookie, 4942 .gpl_only = false, 4943 .ret_type = RET_INTEGER, 4944 .arg1_type = ARG_PTR_TO_CTX, 4945 }; 4946 4947 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4948 { 4949 return __sock_gen_cookie(ctx->sk); 4950 } 4951 4952 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4953 .func = bpf_get_socket_cookie_sock_addr, 4954 .gpl_only = false, 4955 .ret_type = RET_INTEGER, 4956 .arg1_type = ARG_PTR_TO_CTX, 4957 }; 4958 4959 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4960 { 4961 return __sock_gen_cookie(ctx); 4962 } 4963 4964 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4965 .func = bpf_get_socket_cookie_sock, 4966 .gpl_only = false, 4967 .ret_type = RET_INTEGER, 4968 .arg1_type = ARG_PTR_TO_CTX, 4969 }; 4970 4971 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk) 4972 { 4973 return sk ? sock_gen_cookie(sk) : 0; 4974 } 4975 4976 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = { 4977 .func = bpf_get_socket_ptr_cookie, 4978 .gpl_only = false, 4979 .ret_type = RET_INTEGER, 4980 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4981 }; 4982 4983 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4984 { 4985 return __sock_gen_cookie(ctx->sk); 4986 } 4987 4988 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4989 .func = bpf_get_socket_cookie_sock_ops, 4990 .gpl_only = false, 4991 .ret_type = RET_INTEGER, 4992 .arg1_type = ARG_PTR_TO_CTX, 4993 }; 4994 4995 static u64 __bpf_get_netns_cookie(struct sock *sk) 4996 { 4997 const struct net *net = sk ? sock_net(sk) : &init_net; 4998 4999 return net->net_cookie; 5000 } 5001 5002 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 5003 { 5004 return __bpf_get_netns_cookie(ctx); 5005 } 5006 5007 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 5008 .func = bpf_get_netns_cookie_sock, 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_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 5015 { 5016 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5017 } 5018 5019 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 5020 .func = bpf_get_netns_cookie_sock_addr, 5021 .gpl_only = false, 5022 .ret_type = RET_INTEGER, 5023 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5024 }; 5025 5026 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 5027 { 5028 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5029 } 5030 5031 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = { 5032 .func = bpf_get_netns_cookie_sock_ops, 5033 .gpl_only = false, 5034 .ret_type = RET_INTEGER, 5035 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5036 }; 5037 5038 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx) 5039 { 5040 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5041 } 5042 5043 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = { 5044 .func = bpf_get_netns_cookie_sk_msg, 5045 .gpl_only = false, 5046 .ret_type = RET_INTEGER, 5047 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5048 }; 5049 5050 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 5051 { 5052 struct sock *sk = sk_to_full_sk(skb->sk); 5053 kuid_t kuid; 5054 5055 if (!sk || !sk_fullsock(sk)) 5056 return overflowuid; 5057 kuid = sock_net_uid(sock_net(sk), sk); 5058 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 5059 } 5060 5061 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 5062 .func = bpf_get_socket_uid, 5063 .gpl_only = false, 5064 .ret_type = RET_INTEGER, 5065 .arg1_type = ARG_PTR_TO_CTX, 5066 }; 5067 5068 static int sol_socket_sockopt(struct sock *sk, int optname, 5069 char *optval, int *optlen, 5070 bool getopt) 5071 { 5072 switch (optname) { 5073 case SO_REUSEADDR: 5074 case SO_SNDBUF: 5075 case SO_RCVBUF: 5076 case SO_KEEPALIVE: 5077 case SO_PRIORITY: 5078 case SO_REUSEPORT: 5079 case SO_RCVLOWAT: 5080 case SO_MARK: 5081 case SO_MAX_PACING_RATE: 5082 case SO_BINDTOIFINDEX: 5083 case SO_TXREHASH: 5084 if (*optlen != sizeof(int)) 5085 return -EINVAL; 5086 break; 5087 case SO_BINDTODEVICE: 5088 break; 5089 default: 5090 return -EINVAL; 5091 } 5092 5093 if (getopt) { 5094 if (optname == SO_BINDTODEVICE) 5095 return -EINVAL; 5096 return sk_getsockopt(sk, SOL_SOCKET, optname, 5097 KERNEL_SOCKPTR(optval), 5098 KERNEL_SOCKPTR(optlen)); 5099 } 5100 5101 return sk_setsockopt(sk, SOL_SOCKET, optname, 5102 KERNEL_SOCKPTR(optval), *optlen); 5103 } 5104 5105 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname, 5106 char *optval, int optlen) 5107 { 5108 struct tcp_sock *tp = tcp_sk(sk); 5109 unsigned long timeout; 5110 int val; 5111 5112 if (optlen != sizeof(int)) 5113 return -EINVAL; 5114 5115 val = *(int *)optval; 5116 5117 /* Only some options are supported */ 5118 switch (optname) { 5119 case TCP_BPF_IW: 5120 if (val <= 0 || tp->data_segs_out > tp->syn_data) 5121 return -EINVAL; 5122 tcp_snd_cwnd_set(tp, val); 5123 break; 5124 case TCP_BPF_SNDCWND_CLAMP: 5125 if (val <= 0) 5126 return -EINVAL; 5127 tp->snd_cwnd_clamp = val; 5128 tp->snd_ssthresh = val; 5129 break; 5130 case TCP_BPF_DELACK_MAX: 5131 timeout = usecs_to_jiffies(val); 5132 if (timeout > TCP_DELACK_MAX || 5133 timeout < TCP_TIMEOUT_MIN) 5134 return -EINVAL; 5135 inet_csk(sk)->icsk_delack_max = timeout; 5136 break; 5137 case TCP_BPF_RTO_MIN: 5138 timeout = usecs_to_jiffies(val); 5139 if (timeout > TCP_RTO_MIN || 5140 timeout < TCP_TIMEOUT_MIN) 5141 return -EINVAL; 5142 inet_csk(sk)->icsk_rto_min = timeout; 5143 break; 5144 default: 5145 return -EINVAL; 5146 } 5147 5148 return 0; 5149 } 5150 5151 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval, 5152 int *optlen, bool getopt) 5153 { 5154 struct tcp_sock *tp; 5155 int ret; 5156 5157 if (*optlen < 2) 5158 return -EINVAL; 5159 5160 if (getopt) { 5161 if (!inet_csk(sk)->icsk_ca_ops) 5162 return -EINVAL; 5163 /* BPF expects NULL-terminated tcp-cc string */ 5164 optval[--(*optlen)] = '\0'; 5165 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION, 5166 KERNEL_SOCKPTR(optval), 5167 KERNEL_SOCKPTR(optlen)); 5168 } 5169 5170 /* "cdg" is the only cc that alloc a ptr 5171 * in inet_csk_ca area. The bpf-tcp-cc may 5172 * overwrite this ptr after switching to cdg. 5173 */ 5174 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen)) 5175 return -ENOTSUPP; 5176 5177 /* It stops this looping 5178 * 5179 * .init => bpf_setsockopt(tcp_cc) => .init => 5180 * bpf_setsockopt(tcp_cc)" => .init => .... 5181 * 5182 * The second bpf_setsockopt(tcp_cc) is not allowed 5183 * in order to break the loop when both .init 5184 * are the same bpf prog. 5185 * 5186 * This applies even the second bpf_setsockopt(tcp_cc) 5187 * does not cause a loop. This limits only the first 5188 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to 5189 * pick a fallback cc (eg. peer does not support ECN) 5190 * and the second '.init' cannot fallback to 5191 * another. 5192 */ 5193 tp = tcp_sk(sk); 5194 if (tp->bpf_chg_cc_inprogress) 5195 return -EBUSY; 5196 5197 tp->bpf_chg_cc_inprogress = 1; 5198 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION, 5199 KERNEL_SOCKPTR(optval), *optlen); 5200 tp->bpf_chg_cc_inprogress = 0; 5201 return ret; 5202 } 5203 5204 static int sol_tcp_sockopt(struct sock *sk, int optname, 5205 char *optval, int *optlen, 5206 bool getopt) 5207 { 5208 if (sk->sk_protocol != IPPROTO_TCP) 5209 return -EINVAL; 5210 5211 switch (optname) { 5212 case TCP_NODELAY: 5213 case TCP_MAXSEG: 5214 case TCP_KEEPIDLE: 5215 case TCP_KEEPINTVL: 5216 case TCP_KEEPCNT: 5217 case TCP_SYNCNT: 5218 case TCP_WINDOW_CLAMP: 5219 case TCP_THIN_LINEAR_TIMEOUTS: 5220 case TCP_USER_TIMEOUT: 5221 case TCP_NOTSENT_LOWAT: 5222 case TCP_SAVE_SYN: 5223 if (*optlen != sizeof(int)) 5224 return -EINVAL; 5225 break; 5226 case TCP_CONGESTION: 5227 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt); 5228 case TCP_SAVED_SYN: 5229 if (*optlen < 1) 5230 return -EINVAL; 5231 break; 5232 default: 5233 if (getopt) 5234 return -EINVAL; 5235 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen); 5236 } 5237 5238 if (getopt) { 5239 if (optname == TCP_SAVED_SYN) { 5240 struct tcp_sock *tp = tcp_sk(sk); 5241 5242 if (!tp->saved_syn || 5243 *optlen > tcp_saved_syn_len(tp->saved_syn)) 5244 return -EINVAL; 5245 memcpy(optval, tp->saved_syn->data, *optlen); 5246 /* It cannot free tp->saved_syn here because it 5247 * does not know if the user space still needs it. 5248 */ 5249 return 0; 5250 } 5251 5252 return do_tcp_getsockopt(sk, SOL_TCP, optname, 5253 KERNEL_SOCKPTR(optval), 5254 KERNEL_SOCKPTR(optlen)); 5255 } 5256 5257 return do_tcp_setsockopt(sk, SOL_TCP, optname, 5258 KERNEL_SOCKPTR(optval), *optlen); 5259 } 5260 5261 static int sol_ip_sockopt(struct sock *sk, int optname, 5262 char *optval, int *optlen, 5263 bool getopt) 5264 { 5265 if (sk->sk_family != AF_INET) 5266 return -EINVAL; 5267 5268 switch (optname) { 5269 case IP_TOS: 5270 if (*optlen != sizeof(int)) 5271 return -EINVAL; 5272 break; 5273 default: 5274 return -EINVAL; 5275 } 5276 5277 if (getopt) 5278 return do_ip_getsockopt(sk, SOL_IP, optname, 5279 KERNEL_SOCKPTR(optval), 5280 KERNEL_SOCKPTR(optlen)); 5281 5282 return do_ip_setsockopt(sk, SOL_IP, optname, 5283 KERNEL_SOCKPTR(optval), *optlen); 5284 } 5285 5286 static int sol_ipv6_sockopt(struct sock *sk, int optname, 5287 char *optval, int *optlen, 5288 bool getopt) 5289 { 5290 if (sk->sk_family != AF_INET6) 5291 return -EINVAL; 5292 5293 switch (optname) { 5294 case IPV6_TCLASS: 5295 case IPV6_AUTOFLOWLABEL: 5296 if (*optlen != sizeof(int)) 5297 return -EINVAL; 5298 break; 5299 default: 5300 return -EINVAL; 5301 } 5302 5303 if (getopt) 5304 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname, 5305 KERNEL_SOCKPTR(optval), 5306 KERNEL_SOCKPTR(optlen)); 5307 5308 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname, 5309 KERNEL_SOCKPTR(optval), *optlen); 5310 } 5311 5312 static int __bpf_setsockopt(struct sock *sk, int level, int optname, 5313 char *optval, int optlen) 5314 { 5315 if (!sk_fullsock(sk)) 5316 return -EINVAL; 5317 5318 if (level == SOL_SOCKET) 5319 return sol_socket_sockopt(sk, optname, optval, &optlen, false); 5320 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5321 return sol_ip_sockopt(sk, optname, optval, &optlen, false); 5322 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5323 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false); 5324 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5325 return sol_tcp_sockopt(sk, optname, optval, &optlen, false); 5326 5327 return -EINVAL; 5328 } 5329 5330 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 5331 char *optval, int optlen) 5332 { 5333 if (sk_fullsock(sk)) 5334 sock_owned_by_me(sk); 5335 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5336 } 5337 5338 static int __bpf_getsockopt(struct sock *sk, int level, int optname, 5339 char *optval, int optlen) 5340 { 5341 int err, saved_optlen = optlen; 5342 5343 if (!sk_fullsock(sk)) { 5344 err = -EINVAL; 5345 goto done; 5346 } 5347 5348 if (level == SOL_SOCKET) 5349 err = sol_socket_sockopt(sk, optname, optval, &optlen, true); 5350 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5351 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true); 5352 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5353 err = sol_ip_sockopt(sk, optname, optval, &optlen, true); 5354 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5355 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true); 5356 else 5357 err = -EINVAL; 5358 5359 done: 5360 if (err) 5361 optlen = 0; 5362 if (optlen < saved_optlen) 5363 memset(optval + optlen, 0, saved_optlen - optlen); 5364 return err; 5365 } 5366 5367 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 5368 char *optval, int optlen) 5369 { 5370 if (sk_fullsock(sk)) 5371 sock_owned_by_me(sk); 5372 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5373 } 5374 5375 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, 5376 int, optname, char *, optval, int, optlen) 5377 { 5378 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5379 } 5380 5381 const struct bpf_func_proto bpf_sk_setsockopt_proto = { 5382 .func = bpf_sk_setsockopt, 5383 .gpl_only = false, 5384 .ret_type = RET_INTEGER, 5385 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5386 .arg2_type = ARG_ANYTHING, 5387 .arg3_type = ARG_ANYTHING, 5388 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5389 .arg5_type = ARG_CONST_SIZE, 5390 }; 5391 5392 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level, 5393 int, optname, char *, optval, int, optlen) 5394 { 5395 return _bpf_getsockopt(sk, level, optname, optval, optlen); 5396 } 5397 5398 const struct bpf_func_proto bpf_sk_getsockopt_proto = { 5399 .func = bpf_sk_getsockopt, 5400 .gpl_only = false, 5401 .ret_type = RET_INTEGER, 5402 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5403 .arg2_type = ARG_ANYTHING, 5404 .arg3_type = ARG_ANYTHING, 5405 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5406 .arg5_type = ARG_CONST_SIZE, 5407 }; 5408 5409 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level, 5410 int, optname, char *, optval, int, optlen) 5411 { 5412 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5413 } 5414 5415 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = { 5416 .func = bpf_unlocked_sk_setsockopt, 5417 .gpl_only = false, 5418 .ret_type = RET_INTEGER, 5419 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5420 .arg2_type = ARG_ANYTHING, 5421 .arg3_type = ARG_ANYTHING, 5422 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5423 .arg5_type = ARG_CONST_SIZE, 5424 }; 5425 5426 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level, 5427 int, optname, char *, optval, int, optlen) 5428 { 5429 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5430 } 5431 5432 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = { 5433 .func = bpf_unlocked_sk_getsockopt, 5434 .gpl_only = false, 5435 .ret_type = RET_INTEGER, 5436 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5437 .arg2_type = ARG_ANYTHING, 5438 .arg3_type = ARG_ANYTHING, 5439 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5440 .arg5_type = ARG_CONST_SIZE, 5441 }; 5442 5443 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 5444 int, level, int, optname, char *, optval, int, optlen) 5445 { 5446 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5447 } 5448 5449 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5450 .func = bpf_sock_addr_setsockopt, 5451 .gpl_only = false, 5452 .ret_type = RET_INTEGER, 5453 .arg1_type = ARG_PTR_TO_CTX, 5454 .arg2_type = ARG_ANYTHING, 5455 .arg3_type = ARG_ANYTHING, 5456 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5457 .arg5_type = ARG_CONST_SIZE, 5458 }; 5459 5460 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5461 int, level, int, optname, char *, optval, int, optlen) 5462 { 5463 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5464 } 5465 5466 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5467 .func = bpf_sock_addr_getsockopt, 5468 .gpl_only = false, 5469 .ret_type = RET_INTEGER, 5470 .arg1_type = ARG_PTR_TO_CTX, 5471 .arg2_type = ARG_ANYTHING, 5472 .arg3_type = ARG_ANYTHING, 5473 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5474 .arg5_type = ARG_CONST_SIZE, 5475 }; 5476 5477 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5478 int, level, int, optname, char *, optval, int, optlen) 5479 { 5480 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5481 } 5482 5483 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5484 .func = bpf_sock_ops_setsockopt, 5485 .gpl_only = false, 5486 .ret_type = RET_INTEGER, 5487 .arg1_type = ARG_PTR_TO_CTX, 5488 .arg2_type = ARG_ANYTHING, 5489 .arg3_type = ARG_ANYTHING, 5490 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5491 .arg5_type = ARG_CONST_SIZE, 5492 }; 5493 5494 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5495 int optname, const u8 **start) 5496 { 5497 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5498 const u8 *hdr_start; 5499 int ret; 5500 5501 if (syn_skb) { 5502 /* sk is a request_sock here */ 5503 5504 if (optname == TCP_BPF_SYN) { 5505 hdr_start = syn_skb->data; 5506 ret = tcp_hdrlen(syn_skb); 5507 } else if (optname == TCP_BPF_SYN_IP) { 5508 hdr_start = skb_network_header(syn_skb); 5509 ret = skb_network_header_len(syn_skb) + 5510 tcp_hdrlen(syn_skb); 5511 } else { 5512 /* optname == TCP_BPF_SYN_MAC */ 5513 hdr_start = skb_mac_header(syn_skb); 5514 ret = skb_mac_header_len(syn_skb) + 5515 skb_network_header_len(syn_skb) + 5516 tcp_hdrlen(syn_skb); 5517 } 5518 } else { 5519 struct sock *sk = bpf_sock->sk; 5520 struct saved_syn *saved_syn; 5521 5522 if (sk->sk_state == TCP_NEW_SYN_RECV) 5523 /* synack retransmit. bpf_sock->syn_skb will 5524 * not be available. It has to resort to 5525 * saved_syn (if it is saved). 5526 */ 5527 saved_syn = inet_reqsk(sk)->saved_syn; 5528 else 5529 saved_syn = tcp_sk(sk)->saved_syn; 5530 5531 if (!saved_syn) 5532 return -ENOENT; 5533 5534 if (optname == TCP_BPF_SYN) { 5535 hdr_start = saved_syn->data + 5536 saved_syn->mac_hdrlen + 5537 saved_syn->network_hdrlen; 5538 ret = saved_syn->tcp_hdrlen; 5539 } else if (optname == TCP_BPF_SYN_IP) { 5540 hdr_start = saved_syn->data + 5541 saved_syn->mac_hdrlen; 5542 ret = saved_syn->network_hdrlen + 5543 saved_syn->tcp_hdrlen; 5544 } else { 5545 /* optname == TCP_BPF_SYN_MAC */ 5546 5547 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5548 if (!saved_syn->mac_hdrlen) 5549 return -ENOENT; 5550 5551 hdr_start = saved_syn->data; 5552 ret = saved_syn->mac_hdrlen + 5553 saved_syn->network_hdrlen + 5554 saved_syn->tcp_hdrlen; 5555 } 5556 } 5557 5558 *start = hdr_start; 5559 return ret; 5560 } 5561 5562 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5563 int, level, int, optname, char *, optval, int, optlen) 5564 { 5565 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5566 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5567 int ret, copy_len = 0; 5568 const u8 *start; 5569 5570 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5571 if (ret > 0) { 5572 copy_len = ret; 5573 if (optlen < copy_len) { 5574 copy_len = optlen; 5575 ret = -ENOSPC; 5576 } 5577 5578 memcpy(optval, start, copy_len); 5579 } 5580 5581 /* Zero out unused buffer at the end */ 5582 memset(optval + copy_len, 0, optlen - copy_len); 5583 5584 return ret; 5585 } 5586 5587 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5588 } 5589 5590 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5591 .func = bpf_sock_ops_getsockopt, 5592 .gpl_only = false, 5593 .ret_type = RET_INTEGER, 5594 .arg1_type = ARG_PTR_TO_CTX, 5595 .arg2_type = ARG_ANYTHING, 5596 .arg3_type = ARG_ANYTHING, 5597 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5598 .arg5_type = ARG_CONST_SIZE, 5599 }; 5600 5601 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5602 int, argval) 5603 { 5604 struct sock *sk = bpf_sock->sk; 5605 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5606 5607 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5608 return -EINVAL; 5609 5610 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5611 5612 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5613 } 5614 5615 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5616 .func = bpf_sock_ops_cb_flags_set, 5617 .gpl_only = false, 5618 .ret_type = RET_INTEGER, 5619 .arg1_type = ARG_PTR_TO_CTX, 5620 .arg2_type = ARG_ANYTHING, 5621 }; 5622 5623 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5624 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5625 5626 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5627 int, addr_len) 5628 { 5629 #ifdef CONFIG_INET 5630 struct sock *sk = ctx->sk; 5631 u32 flags = BIND_FROM_BPF; 5632 int err; 5633 5634 err = -EINVAL; 5635 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5636 return err; 5637 if (addr->sa_family == AF_INET) { 5638 if (addr_len < sizeof(struct sockaddr_in)) 5639 return err; 5640 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5641 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5642 return __inet_bind(sk, addr, addr_len, flags); 5643 #if IS_ENABLED(CONFIG_IPV6) 5644 } else if (addr->sa_family == AF_INET6) { 5645 if (addr_len < SIN6_LEN_RFC2133) 5646 return err; 5647 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5648 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5649 /* ipv6_bpf_stub cannot be NULL, since it's called from 5650 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5651 */ 5652 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5653 #endif /* CONFIG_IPV6 */ 5654 } 5655 #endif /* CONFIG_INET */ 5656 5657 return -EAFNOSUPPORT; 5658 } 5659 5660 static const struct bpf_func_proto bpf_bind_proto = { 5661 .func = bpf_bind, 5662 .gpl_only = false, 5663 .ret_type = RET_INTEGER, 5664 .arg1_type = ARG_PTR_TO_CTX, 5665 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5666 .arg3_type = ARG_CONST_SIZE, 5667 }; 5668 5669 #ifdef CONFIG_XFRM 5670 5671 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \ 5672 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)) 5673 5674 struct metadata_dst __percpu *xfrm_bpf_md_dst; 5675 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst); 5676 5677 #endif 5678 5679 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5680 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5681 { 5682 const struct sec_path *sp = skb_sec_path(skb); 5683 const struct xfrm_state *x; 5684 5685 if (!sp || unlikely(index >= sp->len || flags)) 5686 goto err_clear; 5687 5688 x = sp->xvec[index]; 5689 5690 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5691 goto err_clear; 5692 5693 to->reqid = x->props.reqid; 5694 to->spi = x->id.spi; 5695 to->family = x->props.family; 5696 to->ext = 0; 5697 5698 if (to->family == AF_INET6) { 5699 memcpy(to->remote_ipv6, x->props.saddr.a6, 5700 sizeof(to->remote_ipv6)); 5701 } else { 5702 to->remote_ipv4 = x->props.saddr.a4; 5703 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5704 } 5705 5706 return 0; 5707 err_clear: 5708 memset(to, 0, size); 5709 return -EINVAL; 5710 } 5711 5712 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5713 .func = bpf_skb_get_xfrm_state, 5714 .gpl_only = false, 5715 .ret_type = RET_INTEGER, 5716 .arg1_type = ARG_PTR_TO_CTX, 5717 .arg2_type = ARG_ANYTHING, 5718 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5719 .arg4_type = ARG_CONST_SIZE, 5720 .arg5_type = ARG_ANYTHING, 5721 }; 5722 #endif 5723 5724 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5725 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu) 5726 { 5727 params->h_vlan_TCI = 0; 5728 params->h_vlan_proto = 0; 5729 if (mtu) 5730 params->mtu_result = mtu; /* union with tot_len */ 5731 5732 return 0; 5733 } 5734 #endif 5735 5736 #if IS_ENABLED(CONFIG_INET) 5737 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5738 u32 flags, bool check_mtu) 5739 { 5740 struct fib_nh_common *nhc; 5741 struct in_device *in_dev; 5742 struct neighbour *neigh; 5743 struct net_device *dev; 5744 struct fib_result res; 5745 struct flowi4 fl4; 5746 u32 mtu = 0; 5747 int err; 5748 5749 dev = dev_get_by_index_rcu(net, params->ifindex); 5750 if (unlikely(!dev)) 5751 return -ENODEV; 5752 5753 /* verify forwarding is enabled on this interface */ 5754 in_dev = __in_dev_get_rcu(dev); 5755 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5756 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5757 5758 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5759 fl4.flowi4_iif = 1; 5760 fl4.flowi4_oif = params->ifindex; 5761 } else { 5762 fl4.flowi4_iif = params->ifindex; 5763 fl4.flowi4_oif = 0; 5764 } 5765 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5766 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5767 fl4.flowi4_flags = 0; 5768 5769 fl4.flowi4_proto = params->l4_protocol; 5770 fl4.daddr = params->ipv4_dst; 5771 fl4.saddr = params->ipv4_src; 5772 fl4.fl4_sport = params->sport; 5773 fl4.fl4_dport = params->dport; 5774 fl4.flowi4_multipath_hash = 0; 5775 5776 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5777 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5778 struct fib_table *tb; 5779 5780 tb = fib_get_table(net, tbid); 5781 if (unlikely(!tb)) 5782 return BPF_FIB_LKUP_RET_NOT_FWDED; 5783 5784 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5785 } else { 5786 fl4.flowi4_mark = 0; 5787 fl4.flowi4_secid = 0; 5788 fl4.flowi4_tun_key.tun_id = 0; 5789 fl4.flowi4_uid = sock_net_uid(net, NULL); 5790 5791 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5792 } 5793 5794 if (err) { 5795 /* map fib lookup errors to RTN_ type */ 5796 if (err == -EINVAL) 5797 return BPF_FIB_LKUP_RET_BLACKHOLE; 5798 if (err == -EHOSTUNREACH) 5799 return BPF_FIB_LKUP_RET_UNREACHABLE; 5800 if (err == -EACCES) 5801 return BPF_FIB_LKUP_RET_PROHIBIT; 5802 5803 return BPF_FIB_LKUP_RET_NOT_FWDED; 5804 } 5805 5806 if (res.type != RTN_UNICAST) 5807 return BPF_FIB_LKUP_RET_NOT_FWDED; 5808 5809 if (fib_info_num_path(res.fi) > 1) 5810 fib_select_path(net, &res, &fl4, NULL); 5811 5812 if (check_mtu) { 5813 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5814 if (params->tot_len > mtu) { 5815 params->mtu_result = mtu; /* union with tot_len */ 5816 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5817 } 5818 } 5819 5820 nhc = res.nhc; 5821 5822 /* do not handle lwt encaps right now */ 5823 if (nhc->nhc_lwtstate) 5824 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5825 5826 dev = nhc->nhc_dev; 5827 5828 params->rt_metric = res.fi->fib_priority; 5829 params->ifindex = dev->ifindex; 5830 5831 /* xdp and cls_bpf programs are run in RCU-bh so 5832 * rcu_read_lock_bh is not needed here 5833 */ 5834 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5835 if (nhc->nhc_gw_family) 5836 params->ipv4_dst = nhc->nhc_gw.ipv4; 5837 } else { 5838 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5839 5840 params->family = AF_INET6; 5841 *dst = nhc->nhc_gw.ipv6; 5842 } 5843 5844 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH) 5845 goto set_fwd_params; 5846 5847 if (likely(nhc->nhc_gw_family != AF_INET6)) 5848 neigh = __ipv4_neigh_lookup_noref(dev, 5849 (__force u32)params->ipv4_dst); 5850 else 5851 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst); 5852 5853 if (!neigh || !(neigh->nud_state & NUD_VALID)) 5854 return BPF_FIB_LKUP_RET_NO_NEIGH; 5855 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5856 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5857 5858 set_fwd_params: 5859 return bpf_fib_set_fwd_params(params, mtu); 5860 } 5861 #endif 5862 5863 #if IS_ENABLED(CONFIG_IPV6) 5864 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5865 u32 flags, bool check_mtu) 5866 { 5867 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5868 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5869 struct fib6_result res = {}; 5870 struct neighbour *neigh; 5871 struct net_device *dev; 5872 struct inet6_dev *idev; 5873 struct flowi6 fl6; 5874 int strict = 0; 5875 int oif, err; 5876 u32 mtu = 0; 5877 5878 /* link local addresses are never forwarded */ 5879 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5880 return BPF_FIB_LKUP_RET_NOT_FWDED; 5881 5882 dev = dev_get_by_index_rcu(net, params->ifindex); 5883 if (unlikely(!dev)) 5884 return -ENODEV; 5885 5886 idev = __in6_dev_get_safely(dev); 5887 if (unlikely(!idev || !idev->cnf.forwarding)) 5888 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5889 5890 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5891 fl6.flowi6_iif = 1; 5892 oif = fl6.flowi6_oif = params->ifindex; 5893 } else { 5894 oif = fl6.flowi6_iif = params->ifindex; 5895 fl6.flowi6_oif = 0; 5896 strict = RT6_LOOKUP_F_HAS_SADDR; 5897 } 5898 fl6.flowlabel = params->flowinfo; 5899 fl6.flowi6_scope = 0; 5900 fl6.flowi6_flags = 0; 5901 fl6.mp_hash = 0; 5902 5903 fl6.flowi6_proto = params->l4_protocol; 5904 fl6.daddr = *dst; 5905 fl6.saddr = *src; 5906 fl6.fl6_sport = params->sport; 5907 fl6.fl6_dport = params->dport; 5908 5909 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5910 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5911 struct fib6_table *tb; 5912 5913 tb = ipv6_stub->fib6_get_table(net, tbid); 5914 if (unlikely(!tb)) 5915 return BPF_FIB_LKUP_RET_NOT_FWDED; 5916 5917 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5918 strict); 5919 } else { 5920 fl6.flowi6_mark = 0; 5921 fl6.flowi6_secid = 0; 5922 fl6.flowi6_tun_key.tun_id = 0; 5923 fl6.flowi6_uid = sock_net_uid(net, NULL); 5924 5925 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5926 } 5927 5928 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5929 res.f6i == net->ipv6.fib6_null_entry)) 5930 return BPF_FIB_LKUP_RET_NOT_FWDED; 5931 5932 switch (res.fib6_type) { 5933 /* only unicast is forwarded */ 5934 case RTN_UNICAST: 5935 break; 5936 case RTN_BLACKHOLE: 5937 return BPF_FIB_LKUP_RET_BLACKHOLE; 5938 case RTN_UNREACHABLE: 5939 return BPF_FIB_LKUP_RET_UNREACHABLE; 5940 case RTN_PROHIBIT: 5941 return BPF_FIB_LKUP_RET_PROHIBIT; 5942 default: 5943 return BPF_FIB_LKUP_RET_NOT_FWDED; 5944 } 5945 5946 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5947 fl6.flowi6_oif != 0, NULL, strict); 5948 5949 if (check_mtu) { 5950 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5951 if (params->tot_len > mtu) { 5952 params->mtu_result = mtu; /* union with tot_len */ 5953 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5954 } 5955 } 5956 5957 if (res.nh->fib_nh_lws) 5958 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5959 5960 if (res.nh->fib_nh_gw_family) 5961 *dst = res.nh->fib_nh_gw6; 5962 5963 dev = res.nh->fib_nh_dev; 5964 params->rt_metric = res.f6i->fib6_metric; 5965 params->ifindex = dev->ifindex; 5966 5967 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH) 5968 goto set_fwd_params; 5969 5970 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5971 * not needed here. 5972 */ 5973 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5974 if (!neigh || !(neigh->nud_state & NUD_VALID)) 5975 return BPF_FIB_LKUP_RET_NO_NEIGH; 5976 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5977 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5978 5979 set_fwd_params: 5980 return bpf_fib_set_fwd_params(params, mtu); 5981 } 5982 #endif 5983 5984 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \ 5985 BPF_FIB_LOOKUP_SKIP_NEIGH) 5986 5987 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 5988 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5989 { 5990 if (plen < sizeof(*params)) 5991 return -EINVAL; 5992 5993 if (flags & ~BPF_FIB_LOOKUP_MASK) 5994 return -EINVAL; 5995 5996 switch (params->family) { 5997 #if IS_ENABLED(CONFIG_INET) 5998 case AF_INET: 5999 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 6000 flags, true); 6001 #endif 6002 #if IS_ENABLED(CONFIG_IPV6) 6003 case AF_INET6: 6004 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 6005 flags, true); 6006 #endif 6007 } 6008 return -EAFNOSUPPORT; 6009 } 6010 6011 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 6012 .func = bpf_xdp_fib_lookup, 6013 .gpl_only = true, 6014 .ret_type = RET_INTEGER, 6015 .arg1_type = ARG_PTR_TO_CTX, 6016 .arg2_type = ARG_PTR_TO_MEM, 6017 .arg3_type = ARG_CONST_SIZE, 6018 .arg4_type = ARG_ANYTHING, 6019 }; 6020 6021 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 6022 struct bpf_fib_lookup *, params, int, plen, u32, flags) 6023 { 6024 struct net *net = dev_net(skb->dev); 6025 int rc = -EAFNOSUPPORT; 6026 bool check_mtu = false; 6027 6028 if (plen < sizeof(*params)) 6029 return -EINVAL; 6030 6031 if (flags & ~BPF_FIB_LOOKUP_MASK) 6032 return -EINVAL; 6033 6034 if (params->tot_len) 6035 check_mtu = true; 6036 6037 switch (params->family) { 6038 #if IS_ENABLED(CONFIG_INET) 6039 case AF_INET: 6040 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu); 6041 break; 6042 #endif 6043 #if IS_ENABLED(CONFIG_IPV6) 6044 case AF_INET6: 6045 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu); 6046 break; 6047 #endif 6048 } 6049 6050 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) { 6051 struct net_device *dev; 6052 6053 /* When tot_len isn't provided by user, check skb 6054 * against MTU of FIB lookup resulting net_device 6055 */ 6056 dev = dev_get_by_index_rcu(net, params->ifindex); 6057 if (!is_skb_forwardable(dev, skb)) 6058 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 6059 6060 params->mtu_result = dev->mtu; /* union with tot_len */ 6061 } 6062 6063 return rc; 6064 } 6065 6066 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 6067 .func = bpf_skb_fib_lookup, 6068 .gpl_only = true, 6069 .ret_type = RET_INTEGER, 6070 .arg1_type = ARG_PTR_TO_CTX, 6071 .arg2_type = ARG_PTR_TO_MEM, 6072 .arg3_type = ARG_CONST_SIZE, 6073 .arg4_type = ARG_ANYTHING, 6074 }; 6075 6076 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr, 6077 u32 ifindex) 6078 { 6079 struct net *netns = dev_net(dev_curr); 6080 6081 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */ 6082 if (ifindex == 0) 6083 return dev_curr; 6084 6085 return dev_get_by_index_rcu(netns, ifindex); 6086 } 6087 6088 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb, 6089 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6090 { 6091 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6092 struct net_device *dev = skb->dev; 6093 int skb_len, dev_len; 6094 int mtu; 6095 6096 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) 6097 return -EINVAL; 6098 6099 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) 6100 return -EINVAL; 6101 6102 dev = __dev_via_ifindex(dev, ifindex); 6103 if (unlikely(!dev)) 6104 return -ENODEV; 6105 6106 mtu = READ_ONCE(dev->mtu); 6107 6108 dev_len = mtu + dev->hard_header_len; 6109 6110 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6111 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len; 6112 6113 skb_len += len_diff; /* minus result pass check */ 6114 if (skb_len <= dev_len) { 6115 ret = BPF_MTU_CHK_RET_SUCCESS; 6116 goto out; 6117 } 6118 /* At this point, skb->len exceed MTU, but as it include length of all 6119 * segments, it can still be below MTU. The SKB can possibly get 6120 * re-segmented in transmit path (see validate_xmit_skb). Thus, user 6121 * must choose if segs are to be MTU checked. 6122 */ 6123 if (skb_is_gso(skb)) { 6124 ret = BPF_MTU_CHK_RET_SUCCESS; 6125 6126 if (flags & BPF_MTU_CHK_SEGS && 6127 !skb_gso_validate_network_len(skb, mtu)) 6128 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG; 6129 } 6130 out: 6131 /* BPF verifier guarantees valid pointer */ 6132 *mtu_len = mtu; 6133 6134 return ret; 6135 } 6136 6137 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp, 6138 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6139 { 6140 struct net_device *dev = xdp->rxq->dev; 6141 int xdp_len = xdp->data_end - xdp->data; 6142 int ret = BPF_MTU_CHK_RET_SUCCESS; 6143 int mtu, dev_len; 6144 6145 /* XDP variant doesn't support multi-buffer segment check (yet) */ 6146 if (unlikely(flags)) 6147 return -EINVAL; 6148 6149 dev = __dev_via_ifindex(dev, ifindex); 6150 if (unlikely(!dev)) 6151 return -ENODEV; 6152 6153 mtu = READ_ONCE(dev->mtu); 6154 6155 /* Add L2-header as dev MTU is L3 size */ 6156 dev_len = mtu + dev->hard_header_len; 6157 6158 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6159 if (*mtu_len) 6160 xdp_len = *mtu_len + dev->hard_header_len; 6161 6162 xdp_len += len_diff; /* minus result pass check */ 6163 if (xdp_len > dev_len) 6164 ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6165 6166 /* BPF verifier guarantees valid pointer */ 6167 *mtu_len = mtu; 6168 6169 return ret; 6170 } 6171 6172 static const struct bpf_func_proto bpf_skb_check_mtu_proto = { 6173 .func = bpf_skb_check_mtu, 6174 .gpl_only = true, 6175 .ret_type = RET_INTEGER, 6176 .arg1_type = ARG_PTR_TO_CTX, 6177 .arg2_type = ARG_ANYTHING, 6178 .arg3_type = ARG_PTR_TO_INT, 6179 .arg4_type = ARG_ANYTHING, 6180 .arg5_type = ARG_ANYTHING, 6181 }; 6182 6183 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = { 6184 .func = bpf_xdp_check_mtu, 6185 .gpl_only = true, 6186 .ret_type = RET_INTEGER, 6187 .arg1_type = ARG_PTR_TO_CTX, 6188 .arg2_type = ARG_ANYTHING, 6189 .arg3_type = ARG_PTR_TO_INT, 6190 .arg4_type = ARG_ANYTHING, 6191 .arg5_type = ARG_ANYTHING, 6192 }; 6193 6194 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6195 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 6196 { 6197 int err; 6198 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 6199 6200 if (!seg6_validate_srh(srh, len, false)) 6201 return -EINVAL; 6202 6203 switch (type) { 6204 case BPF_LWT_ENCAP_SEG6_INLINE: 6205 if (skb->protocol != htons(ETH_P_IPV6)) 6206 return -EBADMSG; 6207 6208 err = seg6_do_srh_inline(skb, srh); 6209 break; 6210 case BPF_LWT_ENCAP_SEG6: 6211 skb_reset_inner_headers(skb); 6212 skb->encapsulation = 1; 6213 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 6214 break; 6215 default: 6216 return -EINVAL; 6217 } 6218 6219 bpf_compute_data_pointers(skb); 6220 if (err) 6221 return err; 6222 6223 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 6224 6225 return seg6_lookup_nexthop(skb, NULL, 0); 6226 } 6227 #endif /* CONFIG_IPV6_SEG6_BPF */ 6228 6229 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6230 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 6231 bool ingress) 6232 { 6233 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 6234 } 6235 #endif 6236 6237 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 6238 u32, len) 6239 { 6240 switch (type) { 6241 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6242 case BPF_LWT_ENCAP_SEG6: 6243 case BPF_LWT_ENCAP_SEG6_INLINE: 6244 return bpf_push_seg6_encap(skb, type, hdr, len); 6245 #endif 6246 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6247 case BPF_LWT_ENCAP_IP: 6248 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 6249 #endif 6250 default: 6251 return -EINVAL; 6252 } 6253 } 6254 6255 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 6256 void *, hdr, u32, len) 6257 { 6258 switch (type) { 6259 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6260 case BPF_LWT_ENCAP_IP: 6261 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 6262 #endif 6263 default: 6264 return -EINVAL; 6265 } 6266 } 6267 6268 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 6269 .func = bpf_lwt_in_push_encap, 6270 .gpl_only = false, 6271 .ret_type = RET_INTEGER, 6272 .arg1_type = ARG_PTR_TO_CTX, 6273 .arg2_type = ARG_ANYTHING, 6274 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6275 .arg4_type = ARG_CONST_SIZE 6276 }; 6277 6278 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 6279 .func = bpf_lwt_xmit_push_encap, 6280 .gpl_only = false, 6281 .ret_type = RET_INTEGER, 6282 .arg1_type = ARG_PTR_TO_CTX, 6283 .arg2_type = ARG_ANYTHING, 6284 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6285 .arg4_type = ARG_CONST_SIZE 6286 }; 6287 6288 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6289 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 6290 const void *, from, u32, len) 6291 { 6292 struct seg6_bpf_srh_state *srh_state = 6293 this_cpu_ptr(&seg6_bpf_srh_states); 6294 struct ipv6_sr_hdr *srh = srh_state->srh; 6295 void *srh_tlvs, *srh_end, *ptr; 6296 int srhoff = 0; 6297 6298 if (srh == NULL) 6299 return -EINVAL; 6300 6301 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 6302 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 6303 6304 ptr = skb->data + offset; 6305 if (ptr >= srh_tlvs && ptr + len <= srh_end) 6306 srh_state->valid = false; 6307 else if (ptr < (void *)&srh->flags || 6308 ptr + len > (void *)&srh->segments) 6309 return -EFAULT; 6310 6311 if (unlikely(bpf_try_make_writable(skb, offset + len))) 6312 return -EFAULT; 6313 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6314 return -EINVAL; 6315 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6316 6317 memcpy(skb->data + offset, from, len); 6318 return 0; 6319 } 6320 6321 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 6322 .func = bpf_lwt_seg6_store_bytes, 6323 .gpl_only = false, 6324 .ret_type = RET_INTEGER, 6325 .arg1_type = ARG_PTR_TO_CTX, 6326 .arg2_type = ARG_ANYTHING, 6327 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6328 .arg4_type = ARG_CONST_SIZE 6329 }; 6330 6331 static void bpf_update_srh_state(struct sk_buff *skb) 6332 { 6333 struct seg6_bpf_srh_state *srh_state = 6334 this_cpu_ptr(&seg6_bpf_srh_states); 6335 int srhoff = 0; 6336 6337 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 6338 srh_state->srh = NULL; 6339 } else { 6340 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6341 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 6342 srh_state->valid = true; 6343 } 6344 } 6345 6346 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 6347 u32, action, void *, param, u32, param_len) 6348 { 6349 struct seg6_bpf_srh_state *srh_state = 6350 this_cpu_ptr(&seg6_bpf_srh_states); 6351 int hdroff = 0; 6352 int err; 6353 6354 switch (action) { 6355 case SEG6_LOCAL_ACTION_END_X: 6356 if (!seg6_bpf_has_valid_srh(skb)) 6357 return -EBADMSG; 6358 if (param_len != sizeof(struct in6_addr)) 6359 return -EINVAL; 6360 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 6361 case SEG6_LOCAL_ACTION_END_T: 6362 if (!seg6_bpf_has_valid_srh(skb)) 6363 return -EBADMSG; 6364 if (param_len != sizeof(int)) 6365 return -EINVAL; 6366 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6367 case SEG6_LOCAL_ACTION_END_DT6: 6368 if (!seg6_bpf_has_valid_srh(skb)) 6369 return -EBADMSG; 6370 if (param_len != sizeof(int)) 6371 return -EINVAL; 6372 6373 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 6374 return -EBADMSG; 6375 if (!pskb_pull(skb, hdroff)) 6376 return -EBADMSG; 6377 6378 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 6379 skb_reset_network_header(skb); 6380 skb_reset_transport_header(skb); 6381 skb->encapsulation = 0; 6382 6383 bpf_compute_data_pointers(skb); 6384 bpf_update_srh_state(skb); 6385 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6386 case SEG6_LOCAL_ACTION_END_B6: 6387 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6388 return -EBADMSG; 6389 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 6390 param, param_len); 6391 if (!err) 6392 bpf_update_srh_state(skb); 6393 6394 return err; 6395 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 6396 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6397 return -EBADMSG; 6398 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 6399 param, param_len); 6400 if (!err) 6401 bpf_update_srh_state(skb); 6402 6403 return err; 6404 default: 6405 return -EINVAL; 6406 } 6407 } 6408 6409 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 6410 .func = bpf_lwt_seg6_action, 6411 .gpl_only = false, 6412 .ret_type = RET_INTEGER, 6413 .arg1_type = ARG_PTR_TO_CTX, 6414 .arg2_type = ARG_ANYTHING, 6415 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6416 .arg4_type = ARG_CONST_SIZE 6417 }; 6418 6419 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 6420 s32, len) 6421 { 6422 struct seg6_bpf_srh_state *srh_state = 6423 this_cpu_ptr(&seg6_bpf_srh_states); 6424 struct ipv6_sr_hdr *srh = srh_state->srh; 6425 void *srh_end, *srh_tlvs, *ptr; 6426 struct ipv6hdr *hdr; 6427 int srhoff = 0; 6428 int ret; 6429 6430 if (unlikely(srh == NULL)) 6431 return -EINVAL; 6432 6433 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 6434 ((srh->first_segment + 1) << 4)); 6435 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 6436 srh_state->hdrlen); 6437 ptr = skb->data + offset; 6438 6439 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 6440 return -EFAULT; 6441 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 6442 return -EFAULT; 6443 6444 if (len > 0) { 6445 ret = skb_cow_head(skb, len); 6446 if (unlikely(ret < 0)) 6447 return ret; 6448 6449 ret = bpf_skb_net_hdr_push(skb, offset, len); 6450 } else { 6451 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 6452 } 6453 6454 bpf_compute_data_pointers(skb); 6455 if (unlikely(ret < 0)) 6456 return ret; 6457 6458 hdr = (struct ipv6hdr *)skb->data; 6459 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 6460 6461 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6462 return -EINVAL; 6463 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6464 srh_state->hdrlen += len; 6465 srh_state->valid = false; 6466 return 0; 6467 } 6468 6469 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 6470 .func = bpf_lwt_seg6_adjust_srh, 6471 .gpl_only = false, 6472 .ret_type = RET_INTEGER, 6473 .arg1_type = ARG_PTR_TO_CTX, 6474 .arg2_type = ARG_ANYTHING, 6475 .arg3_type = ARG_ANYTHING, 6476 }; 6477 #endif /* CONFIG_IPV6_SEG6_BPF */ 6478 6479 #ifdef CONFIG_INET 6480 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 6481 int dif, int sdif, u8 family, u8 proto) 6482 { 6483 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo; 6484 bool refcounted = false; 6485 struct sock *sk = NULL; 6486 6487 if (family == AF_INET) { 6488 __be32 src4 = tuple->ipv4.saddr; 6489 __be32 dst4 = tuple->ipv4.daddr; 6490 6491 if (proto == IPPROTO_TCP) 6492 sk = __inet_lookup(net, hinfo, NULL, 0, 6493 src4, tuple->ipv4.sport, 6494 dst4, tuple->ipv4.dport, 6495 dif, sdif, &refcounted); 6496 else 6497 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 6498 dst4, tuple->ipv4.dport, 6499 dif, sdif, net->ipv4.udp_table, NULL); 6500 #if IS_ENABLED(CONFIG_IPV6) 6501 } else { 6502 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 6503 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 6504 6505 if (proto == IPPROTO_TCP) 6506 sk = __inet6_lookup(net, hinfo, NULL, 0, 6507 src6, tuple->ipv6.sport, 6508 dst6, ntohs(tuple->ipv6.dport), 6509 dif, sdif, &refcounted); 6510 else if (likely(ipv6_bpf_stub)) 6511 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 6512 src6, tuple->ipv6.sport, 6513 dst6, tuple->ipv6.dport, 6514 dif, sdif, 6515 net->ipv4.udp_table, NULL); 6516 #endif 6517 } 6518 6519 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 6520 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6521 sk = NULL; 6522 } 6523 return sk; 6524 } 6525 6526 /* bpf_skc_lookup performs the core lookup for different types of sockets, 6527 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 6528 */ 6529 static struct sock * 6530 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6531 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6532 u64 flags) 6533 { 6534 struct sock *sk = NULL; 6535 struct net *net; 6536 u8 family; 6537 int sdif; 6538 6539 if (len == sizeof(tuple->ipv4)) 6540 family = AF_INET; 6541 else if (len == sizeof(tuple->ipv6)) 6542 family = AF_INET6; 6543 else 6544 return NULL; 6545 6546 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX))) 6547 goto out; 6548 6549 if (family == AF_INET) 6550 sdif = inet_sdif(skb); 6551 else 6552 sdif = inet6_sdif(skb); 6553 6554 if ((s32)netns_id < 0) { 6555 net = caller_net; 6556 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6557 } else { 6558 net = get_net_ns_by_id(caller_net, netns_id); 6559 if (unlikely(!net)) 6560 goto out; 6561 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6562 put_net(net); 6563 } 6564 6565 out: 6566 return sk; 6567 } 6568 6569 static struct sock * 6570 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6571 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6572 u64 flags) 6573 { 6574 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 6575 ifindex, proto, netns_id, flags); 6576 6577 if (sk) { 6578 struct sock *sk2 = sk_to_full_sk(sk); 6579 6580 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6581 * sock refcnt is decremented to prevent a request_sock leak. 6582 */ 6583 if (!sk_fullsock(sk2)) 6584 sk2 = NULL; 6585 if (sk2 != sk) { 6586 sock_gen_put(sk); 6587 /* Ensure there is no need to bump sk2 refcnt */ 6588 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6589 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6590 return NULL; 6591 } 6592 sk = sk2; 6593 } 6594 } 6595 6596 return sk; 6597 } 6598 6599 static struct sock * 6600 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6601 u8 proto, u64 netns_id, u64 flags) 6602 { 6603 struct net *caller_net; 6604 int ifindex; 6605 6606 if (skb->dev) { 6607 caller_net = dev_net(skb->dev); 6608 ifindex = skb->dev->ifindex; 6609 } else { 6610 caller_net = sock_net(skb->sk); 6611 ifindex = 0; 6612 } 6613 6614 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6615 netns_id, flags); 6616 } 6617 6618 static struct sock * 6619 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6620 u8 proto, u64 netns_id, u64 flags) 6621 { 6622 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6623 flags); 6624 6625 if (sk) { 6626 struct sock *sk2 = sk_to_full_sk(sk); 6627 6628 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6629 * sock refcnt is decremented to prevent a request_sock leak. 6630 */ 6631 if (!sk_fullsock(sk2)) 6632 sk2 = NULL; 6633 if (sk2 != sk) { 6634 sock_gen_put(sk); 6635 /* Ensure there is no need to bump sk2 refcnt */ 6636 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6637 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6638 return NULL; 6639 } 6640 sk = sk2; 6641 } 6642 } 6643 6644 return sk; 6645 } 6646 6647 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6648 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6649 { 6650 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6651 netns_id, flags); 6652 } 6653 6654 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6655 .func = bpf_skc_lookup_tcp, 6656 .gpl_only = false, 6657 .pkt_access = true, 6658 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6659 .arg1_type = ARG_PTR_TO_CTX, 6660 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6661 .arg3_type = ARG_CONST_SIZE, 6662 .arg4_type = ARG_ANYTHING, 6663 .arg5_type = ARG_ANYTHING, 6664 }; 6665 6666 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6667 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6668 { 6669 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6670 netns_id, flags); 6671 } 6672 6673 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6674 .func = bpf_sk_lookup_tcp, 6675 .gpl_only = false, 6676 .pkt_access = true, 6677 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6678 .arg1_type = ARG_PTR_TO_CTX, 6679 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6680 .arg3_type = ARG_CONST_SIZE, 6681 .arg4_type = ARG_ANYTHING, 6682 .arg5_type = ARG_ANYTHING, 6683 }; 6684 6685 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6686 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6687 { 6688 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6689 netns_id, flags); 6690 } 6691 6692 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6693 .func = bpf_sk_lookup_udp, 6694 .gpl_only = false, 6695 .pkt_access = true, 6696 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6697 .arg1_type = ARG_PTR_TO_CTX, 6698 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6699 .arg3_type = ARG_CONST_SIZE, 6700 .arg4_type = ARG_ANYTHING, 6701 .arg5_type = ARG_ANYTHING, 6702 }; 6703 6704 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6705 { 6706 if (sk && sk_is_refcounted(sk)) 6707 sock_gen_put(sk); 6708 return 0; 6709 } 6710 6711 static const struct bpf_func_proto bpf_sk_release_proto = { 6712 .func = bpf_sk_release, 6713 .gpl_only = false, 6714 .ret_type = RET_INTEGER, 6715 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE, 6716 }; 6717 6718 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6719 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6720 { 6721 struct net *caller_net = dev_net(ctx->rxq->dev); 6722 int ifindex = ctx->rxq->dev->ifindex; 6723 6724 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6725 ifindex, IPPROTO_UDP, netns_id, 6726 flags); 6727 } 6728 6729 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6730 .func = bpf_xdp_sk_lookup_udp, 6731 .gpl_only = false, 6732 .pkt_access = true, 6733 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6734 .arg1_type = ARG_PTR_TO_CTX, 6735 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6736 .arg3_type = ARG_CONST_SIZE, 6737 .arg4_type = ARG_ANYTHING, 6738 .arg5_type = ARG_ANYTHING, 6739 }; 6740 6741 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6742 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6743 { 6744 struct net *caller_net = dev_net(ctx->rxq->dev); 6745 int ifindex = ctx->rxq->dev->ifindex; 6746 6747 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6748 ifindex, IPPROTO_TCP, netns_id, 6749 flags); 6750 } 6751 6752 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6753 .func = bpf_xdp_skc_lookup_tcp, 6754 .gpl_only = false, 6755 .pkt_access = true, 6756 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6757 .arg1_type = ARG_PTR_TO_CTX, 6758 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6759 .arg3_type = ARG_CONST_SIZE, 6760 .arg4_type = ARG_ANYTHING, 6761 .arg5_type = ARG_ANYTHING, 6762 }; 6763 6764 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6765 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6766 { 6767 struct net *caller_net = dev_net(ctx->rxq->dev); 6768 int ifindex = ctx->rxq->dev->ifindex; 6769 6770 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6771 ifindex, IPPROTO_TCP, netns_id, 6772 flags); 6773 } 6774 6775 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6776 .func = bpf_xdp_sk_lookup_tcp, 6777 .gpl_only = false, 6778 .pkt_access = true, 6779 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6780 .arg1_type = ARG_PTR_TO_CTX, 6781 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6782 .arg3_type = ARG_CONST_SIZE, 6783 .arg4_type = ARG_ANYTHING, 6784 .arg5_type = ARG_ANYTHING, 6785 }; 6786 6787 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6788 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6789 { 6790 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6791 sock_net(ctx->sk), 0, 6792 IPPROTO_TCP, netns_id, flags); 6793 } 6794 6795 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6796 .func = bpf_sock_addr_skc_lookup_tcp, 6797 .gpl_only = false, 6798 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6799 .arg1_type = ARG_PTR_TO_CTX, 6800 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6801 .arg3_type = ARG_CONST_SIZE, 6802 .arg4_type = ARG_ANYTHING, 6803 .arg5_type = ARG_ANYTHING, 6804 }; 6805 6806 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6807 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6808 { 6809 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6810 sock_net(ctx->sk), 0, IPPROTO_TCP, 6811 netns_id, flags); 6812 } 6813 6814 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6815 .func = bpf_sock_addr_sk_lookup_tcp, 6816 .gpl_only = false, 6817 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6818 .arg1_type = ARG_PTR_TO_CTX, 6819 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6820 .arg3_type = ARG_CONST_SIZE, 6821 .arg4_type = ARG_ANYTHING, 6822 .arg5_type = ARG_ANYTHING, 6823 }; 6824 6825 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6826 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6827 { 6828 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6829 sock_net(ctx->sk), 0, IPPROTO_UDP, 6830 netns_id, flags); 6831 } 6832 6833 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6834 .func = bpf_sock_addr_sk_lookup_udp, 6835 .gpl_only = false, 6836 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6837 .arg1_type = ARG_PTR_TO_CTX, 6838 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6839 .arg3_type = ARG_CONST_SIZE, 6840 .arg4_type = ARG_ANYTHING, 6841 .arg5_type = ARG_ANYTHING, 6842 }; 6843 6844 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6845 struct bpf_insn_access_aux *info) 6846 { 6847 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6848 icsk_retransmits)) 6849 return false; 6850 6851 if (off % size != 0) 6852 return false; 6853 6854 switch (off) { 6855 case offsetof(struct bpf_tcp_sock, bytes_received): 6856 case offsetof(struct bpf_tcp_sock, bytes_acked): 6857 return size == sizeof(__u64); 6858 default: 6859 return size == sizeof(__u32); 6860 } 6861 } 6862 6863 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6864 const struct bpf_insn *si, 6865 struct bpf_insn *insn_buf, 6866 struct bpf_prog *prog, u32 *target_size) 6867 { 6868 struct bpf_insn *insn = insn_buf; 6869 6870 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6871 do { \ 6872 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6873 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6874 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6875 si->dst_reg, si->src_reg, \ 6876 offsetof(struct tcp_sock, FIELD)); \ 6877 } while (0) 6878 6879 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6880 do { \ 6881 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6882 FIELD) > \ 6883 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6884 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6885 struct inet_connection_sock, \ 6886 FIELD), \ 6887 si->dst_reg, si->src_reg, \ 6888 offsetof( \ 6889 struct inet_connection_sock, \ 6890 FIELD)); \ 6891 } while (0) 6892 6893 switch (si->off) { 6894 case offsetof(struct bpf_tcp_sock, rtt_min): 6895 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6896 sizeof(struct minmax)); 6897 BUILD_BUG_ON(sizeof(struct minmax) < 6898 sizeof(struct minmax_sample)); 6899 6900 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6901 offsetof(struct tcp_sock, rtt_min) + 6902 offsetof(struct minmax_sample, v)); 6903 break; 6904 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6905 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6906 break; 6907 case offsetof(struct bpf_tcp_sock, srtt_us): 6908 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6909 break; 6910 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6911 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6912 break; 6913 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6914 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6915 break; 6916 case offsetof(struct bpf_tcp_sock, snd_nxt): 6917 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6918 break; 6919 case offsetof(struct bpf_tcp_sock, snd_una): 6920 BPF_TCP_SOCK_GET_COMMON(snd_una); 6921 break; 6922 case offsetof(struct bpf_tcp_sock, mss_cache): 6923 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6924 break; 6925 case offsetof(struct bpf_tcp_sock, ecn_flags): 6926 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6927 break; 6928 case offsetof(struct bpf_tcp_sock, rate_delivered): 6929 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6930 break; 6931 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6932 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6933 break; 6934 case offsetof(struct bpf_tcp_sock, packets_out): 6935 BPF_TCP_SOCK_GET_COMMON(packets_out); 6936 break; 6937 case offsetof(struct bpf_tcp_sock, retrans_out): 6938 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6939 break; 6940 case offsetof(struct bpf_tcp_sock, total_retrans): 6941 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6942 break; 6943 case offsetof(struct bpf_tcp_sock, segs_in): 6944 BPF_TCP_SOCK_GET_COMMON(segs_in); 6945 break; 6946 case offsetof(struct bpf_tcp_sock, data_segs_in): 6947 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6948 break; 6949 case offsetof(struct bpf_tcp_sock, segs_out): 6950 BPF_TCP_SOCK_GET_COMMON(segs_out); 6951 break; 6952 case offsetof(struct bpf_tcp_sock, data_segs_out): 6953 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6954 break; 6955 case offsetof(struct bpf_tcp_sock, lost_out): 6956 BPF_TCP_SOCK_GET_COMMON(lost_out); 6957 break; 6958 case offsetof(struct bpf_tcp_sock, sacked_out): 6959 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6960 break; 6961 case offsetof(struct bpf_tcp_sock, bytes_received): 6962 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6963 break; 6964 case offsetof(struct bpf_tcp_sock, bytes_acked): 6965 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6966 break; 6967 case offsetof(struct bpf_tcp_sock, dsack_dups): 6968 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6969 break; 6970 case offsetof(struct bpf_tcp_sock, delivered): 6971 BPF_TCP_SOCK_GET_COMMON(delivered); 6972 break; 6973 case offsetof(struct bpf_tcp_sock, delivered_ce): 6974 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 6975 break; 6976 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 6977 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 6978 break; 6979 } 6980 6981 return insn - insn_buf; 6982 } 6983 6984 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 6985 { 6986 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 6987 return (unsigned long)sk; 6988 6989 return (unsigned long)NULL; 6990 } 6991 6992 const struct bpf_func_proto bpf_tcp_sock_proto = { 6993 .func = bpf_tcp_sock, 6994 .gpl_only = false, 6995 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 6996 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6997 }; 6998 6999 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 7000 { 7001 sk = sk_to_full_sk(sk); 7002 7003 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 7004 return (unsigned long)sk; 7005 7006 return (unsigned long)NULL; 7007 } 7008 7009 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 7010 .func = bpf_get_listener_sock, 7011 .gpl_only = false, 7012 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7013 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 7014 }; 7015 7016 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 7017 { 7018 unsigned int iphdr_len; 7019 7020 switch (skb_protocol(skb, true)) { 7021 case cpu_to_be16(ETH_P_IP): 7022 iphdr_len = sizeof(struct iphdr); 7023 break; 7024 case cpu_to_be16(ETH_P_IPV6): 7025 iphdr_len = sizeof(struct ipv6hdr); 7026 break; 7027 default: 7028 return 0; 7029 } 7030 7031 if (skb_headlen(skb) < iphdr_len) 7032 return 0; 7033 7034 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 7035 return 0; 7036 7037 return INET_ECN_set_ce(skb); 7038 } 7039 7040 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7041 struct bpf_insn_access_aux *info) 7042 { 7043 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 7044 return false; 7045 7046 if (off % size != 0) 7047 return false; 7048 7049 switch (off) { 7050 default: 7051 return size == sizeof(__u32); 7052 } 7053 } 7054 7055 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 7056 const struct bpf_insn *si, 7057 struct bpf_insn *insn_buf, 7058 struct bpf_prog *prog, u32 *target_size) 7059 { 7060 struct bpf_insn *insn = insn_buf; 7061 7062 #define BPF_XDP_SOCK_GET(FIELD) \ 7063 do { \ 7064 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 7065 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 7066 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 7067 si->dst_reg, si->src_reg, \ 7068 offsetof(struct xdp_sock, FIELD)); \ 7069 } while (0) 7070 7071 switch (si->off) { 7072 case offsetof(struct bpf_xdp_sock, queue_id): 7073 BPF_XDP_SOCK_GET(queue_id); 7074 break; 7075 } 7076 7077 return insn - insn_buf; 7078 } 7079 7080 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 7081 .func = bpf_skb_ecn_set_ce, 7082 .gpl_only = false, 7083 .ret_type = RET_INTEGER, 7084 .arg1_type = ARG_PTR_TO_CTX, 7085 }; 7086 7087 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7088 struct tcphdr *, th, u32, th_len) 7089 { 7090 #ifdef CONFIG_SYN_COOKIES 7091 u32 cookie; 7092 int ret; 7093 7094 if (unlikely(!sk || th_len < sizeof(*th))) 7095 return -EINVAL; 7096 7097 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 7098 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7099 return -EINVAL; 7100 7101 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7102 return -EINVAL; 7103 7104 if (!th->ack || th->rst || th->syn) 7105 return -ENOENT; 7106 7107 if (unlikely(iph_len < sizeof(struct iphdr))) 7108 return -EINVAL; 7109 7110 if (tcp_synq_no_recent_overflow(sk)) 7111 return -ENOENT; 7112 7113 cookie = ntohl(th->ack_seq) - 1; 7114 7115 /* Both struct iphdr and struct ipv6hdr have the version field at the 7116 * same offset so we can cast to the shorter header (struct iphdr). 7117 */ 7118 switch (((struct iphdr *)iph)->version) { 7119 case 4: 7120 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7121 return -EINVAL; 7122 7123 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 7124 break; 7125 7126 #if IS_BUILTIN(CONFIG_IPV6) 7127 case 6: 7128 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7129 return -EINVAL; 7130 7131 if (sk->sk_family != AF_INET6) 7132 return -EINVAL; 7133 7134 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 7135 break; 7136 #endif /* CONFIG_IPV6 */ 7137 7138 default: 7139 return -EPROTONOSUPPORT; 7140 } 7141 7142 if (ret > 0) 7143 return 0; 7144 7145 return -ENOENT; 7146 #else 7147 return -ENOTSUPP; 7148 #endif 7149 } 7150 7151 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 7152 .func = bpf_tcp_check_syncookie, 7153 .gpl_only = true, 7154 .pkt_access = true, 7155 .ret_type = RET_INTEGER, 7156 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7157 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7158 .arg3_type = ARG_CONST_SIZE, 7159 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7160 .arg5_type = ARG_CONST_SIZE, 7161 }; 7162 7163 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7164 struct tcphdr *, th, u32, th_len) 7165 { 7166 #ifdef CONFIG_SYN_COOKIES 7167 u32 cookie; 7168 u16 mss; 7169 7170 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 7171 return -EINVAL; 7172 7173 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7174 return -EINVAL; 7175 7176 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7177 return -ENOENT; 7178 7179 if (!th->syn || th->ack || th->fin || th->rst) 7180 return -EINVAL; 7181 7182 if (unlikely(iph_len < sizeof(struct iphdr))) 7183 return -EINVAL; 7184 7185 /* Both struct iphdr and struct ipv6hdr have the version field at the 7186 * same offset so we can cast to the shorter header (struct iphdr). 7187 */ 7188 switch (((struct iphdr *)iph)->version) { 7189 case 4: 7190 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7191 return -EINVAL; 7192 7193 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 7194 break; 7195 7196 #if IS_BUILTIN(CONFIG_IPV6) 7197 case 6: 7198 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7199 return -EINVAL; 7200 7201 if (sk->sk_family != AF_INET6) 7202 return -EINVAL; 7203 7204 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 7205 break; 7206 #endif /* CONFIG_IPV6 */ 7207 7208 default: 7209 return -EPROTONOSUPPORT; 7210 } 7211 if (mss == 0) 7212 return -ENOENT; 7213 7214 return cookie | ((u64)mss << 32); 7215 #else 7216 return -EOPNOTSUPP; 7217 #endif /* CONFIG_SYN_COOKIES */ 7218 } 7219 7220 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 7221 .func = bpf_tcp_gen_syncookie, 7222 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 7223 .pkt_access = true, 7224 .ret_type = RET_INTEGER, 7225 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7226 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7227 .arg3_type = ARG_CONST_SIZE, 7228 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7229 .arg5_type = ARG_CONST_SIZE, 7230 }; 7231 7232 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 7233 { 7234 if (!sk || flags != 0) 7235 return -EINVAL; 7236 if (!skb_at_tc_ingress(skb)) 7237 return -EOPNOTSUPP; 7238 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 7239 return -ENETUNREACH; 7240 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 7241 return -ESOCKTNOSUPPORT; 7242 if (sk_is_refcounted(sk) && 7243 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 7244 return -ENOENT; 7245 7246 skb_orphan(skb); 7247 skb->sk = sk; 7248 skb->destructor = sock_pfree; 7249 7250 return 0; 7251 } 7252 7253 static const struct bpf_func_proto bpf_sk_assign_proto = { 7254 .func = bpf_sk_assign, 7255 .gpl_only = false, 7256 .ret_type = RET_INTEGER, 7257 .arg1_type = ARG_PTR_TO_CTX, 7258 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7259 .arg3_type = ARG_ANYTHING, 7260 }; 7261 7262 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 7263 u8 search_kind, const u8 *magic, 7264 u8 magic_len, bool *eol) 7265 { 7266 u8 kind, kind_len; 7267 7268 *eol = false; 7269 7270 while (op < opend) { 7271 kind = op[0]; 7272 7273 if (kind == TCPOPT_EOL) { 7274 *eol = true; 7275 return ERR_PTR(-ENOMSG); 7276 } else if (kind == TCPOPT_NOP) { 7277 op++; 7278 continue; 7279 } 7280 7281 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 7282 /* Something is wrong in the received header. 7283 * Follow the TCP stack's tcp_parse_options() 7284 * and just bail here. 7285 */ 7286 return ERR_PTR(-EFAULT); 7287 7288 kind_len = op[1]; 7289 if (search_kind == kind) { 7290 if (!magic_len) 7291 return op; 7292 7293 if (magic_len > kind_len - 2) 7294 return ERR_PTR(-ENOMSG); 7295 7296 if (!memcmp(&op[2], magic, magic_len)) 7297 return op; 7298 } 7299 7300 op += kind_len; 7301 } 7302 7303 return ERR_PTR(-ENOMSG); 7304 } 7305 7306 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7307 void *, search_res, u32, len, u64, flags) 7308 { 7309 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 7310 const u8 *op, *opend, *magic, *search = search_res; 7311 u8 search_kind, search_len, copy_len, magic_len; 7312 int ret; 7313 7314 /* 2 byte is the minimal option len except TCPOPT_NOP and 7315 * TCPOPT_EOL which are useless for the bpf prog to learn 7316 * and this helper disallow loading them also. 7317 */ 7318 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 7319 return -EINVAL; 7320 7321 search_kind = search[0]; 7322 search_len = search[1]; 7323 7324 if (search_len > len || search_kind == TCPOPT_NOP || 7325 search_kind == TCPOPT_EOL) 7326 return -EINVAL; 7327 7328 if (search_kind == TCPOPT_EXP || search_kind == 253) { 7329 /* 16 or 32 bit magic. +2 for kind and kind length */ 7330 if (search_len != 4 && search_len != 6) 7331 return -EINVAL; 7332 magic = &search[2]; 7333 magic_len = search_len - 2; 7334 } else { 7335 if (search_len) 7336 return -EINVAL; 7337 magic = NULL; 7338 magic_len = 0; 7339 } 7340 7341 if (load_syn) { 7342 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 7343 if (ret < 0) 7344 return ret; 7345 7346 opend = op + ret; 7347 op += sizeof(struct tcphdr); 7348 } else { 7349 if (!bpf_sock->skb || 7350 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7351 /* This bpf_sock->op cannot call this helper */ 7352 return -EPERM; 7353 7354 opend = bpf_sock->skb_data_end; 7355 op = bpf_sock->skb->data + sizeof(struct tcphdr); 7356 } 7357 7358 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 7359 &eol); 7360 if (IS_ERR(op)) 7361 return PTR_ERR(op); 7362 7363 copy_len = op[1]; 7364 ret = copy_len; 7365 if (copy_len > len) { 7366 ret = -ENOSPC; 7367 copy_len = len; 7368 } 7369 7370 memcpy(search_res, op, copy_len); 7371 return ret; 7372 } 7373 7374 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 7375 .func = bpf_sock_ops_load_hdr_opt, 7376 .gpl_only = false, 7377 .ret_type = RET_INTEGER, 7378 .arg1_type = ARG_PTR_TO_CTX, 7379 .arg2_type = ARG_PTR_TO_MEM, 7380 .arg3_type = ARG_CONST_SIZE, 7381 .arg4_type = ARG_ANYTHING, 7382 }; 7383 7384 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7385 const void *, from, u32, len, u64, flags) 7386 { 7387 u8 new_kind, new_kind_len, magic_len = 0, *opend; 7388 const u8 *op, *new_op, *magic = NULL; 7389 struct sk_buff *skb; 7390 bool eol; 7391 7392 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 7393 return -EPERM; 7394 7395 if (len < 2 || flags) 7396 return -EINVAL; 7397 7398 new_op = from; 7399 new_kind = new_op[0]; 7400 new_kind_len = new_op[1]; 7401 7402 if (new_kind_len > len || new_kind == TCPOPT_NOP || 7403 new_kind == TCPOPT_EOL) 7404 return -EINVAL; 7405 7406 if (new_kind_len > bpf_sock->remaining_opt_len) 7407 return -ENOSPC; 7408 7409 /* 253 is another experimental kind */ 7410 if (new_kind == TCPOPT_EXP || new_kind == 253) { 7411 if (new_kind_len < 4) 7412 return -EINVAL; 7413 /* Match for the 2 byte magic also. 7414 * RFC 6994: the magic could be 2 or 4 bytes. 7415 * Hence, matching by 2 byte only is on the 7416 * conservative side but it is the right 7417 * thing to do for the 'search-for-duplication' 7418 * purpose. 7419 */ 7420 magic = &new_op[2]; 7421 magic_len = 2; 7422 } 7423 7424 /* Check for duplication */ 7425 skb = bpf_sock->skb; 7426 op = skb->data + sizeof(struct tcphdr); 7427 opend = bpf_sock->skb_data_end; 7428 7429 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 7430 &eol); 7431 if (!IS_ERR(op)) 7432 return -EEXIST; 7433 7434 if (PTR_ERR(op) != -ENOMSG) 7435 return PTR_ERR(op); 7436 7437 if (eol) 7438 /* The option has been ended. Treat it as no more 7439 * header option can be written. 7440 */ 7441 return -ENOSPC; 7442 7443 /* No duplication found. Store the header option. */ 7444 memcpy(opend, from, new_kind_len); 7445 7446 bpf_sock->remaining_opt_len -= new_kind_len; 7447 bpf_sock->skb_data_end += new_kind_len; 7448 7449 return 0; 7450 } 7451 7452 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 7453 .func = bpf_sock_ops_store_hdr_opt, 7454 .gpl_only = false, 7455 .ret_type = RET_INTEGER, 7456 .arg1_type = ARG_PTR_TO_CTX, 7457 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7458 .arg3_type = ARG_CONST_SIZE, 7459 .arg4_type = ARG_ANYTHING, 7460 }; 7461 7462 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7463 u32, len, u64, flags) 7464 { 7465 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7466 return -EPERM; 7467 7468 if (flags || len < 2) 7469 return -EINVAL; 7470 7471 if (len > bpf_sock->remaining_opt_len) 7472 return -ENOSPC; 7473 7474 bpf_sock->remaining_opt_len -= len; 7475 7476 return 0; 7477 } 7478 7479 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 7480 .func = bpf_sock_ops_reserve_hdr_opt, 7481 .gpl_only = false, 7482 .ret_type = RET_INTEGER, 7483 .arg1_type = ARG_PTR_TO_CTX, 7484 .arg2_type = ARG_ANYTHING, 7485 .arg3_type = ARG_ANYTHING, 7486 }; 7487 7488 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, 7489 u64, tstamp, u32, tstamp_type) 7490 { 7491 /* skb_clear_delivery_time() is done for inet protocol */ 7492 if (skb->protocol != htons(ETH_P_IP) && 7493 skb->protocol != htons(ETH_P_IPV6)) 7494 return -EOPNOTSUPP; 7495 7496 switch (tstamp_type) { 7497 case BPF_SKB_TSTAMP_DELIVERY_MONO: 7498 if (!tstamp) 7499 return -EINVAL; 7500 skb->tstamp = tstamp; 7501 skb->mono_delivery_time = 1; 7502 break; 7503 case BPF_SKB_TSTAMP_UNSPEC: 7504 if (tstamp) 7505 return -EINVAL; 7506 skb->tstamp = 0; 7507 skb->mono_delivery_time = 0; 7508 break; 7509 default: 7510 return -EINVAL; 7511 } 7512 7513 return 0; 7514 } 7515 7516 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = { 7517 .func = bpf_skb_set_tstamp, 7518 .gpl_only = false, 7519 .ret_type = RET_INTEGER, 7520 .arg1_type = ARG_PTR_TO_CTX, 7521 .arg2_type = ARG_ANYTHING, 7522 .arg3_type = ARG_ANYTHING, 7523 }; 7524 7525 #ifdef CONFIG_SYN_COOKIES 7526 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph, 7527 struct tcphdr *, th, u32, th_len) 7528 { 7529 u32 cookie; 7530 u16 mss; 7531 7532 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7533 return -EINVAL; 7534 7535 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT; 7536 cookie = __cookie_v4_init_sequence(iph, th, &mss); 7537 7538 return cookie | ((u64)mss << 32); 7539 } 7540 7541 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = { 7542 .func = bpf_tcp_raw_gen_syncookie_ipv4, 7543 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */ 7544 .pkt_access = true, 7545 .ret_type = RET_INTEGER, 7546 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7547 .arg1_size = sizeof(struct iphdr), 7548 .arg2_type = ARG_PTR_TO_MEM, 7549 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 7550 }; 7551 7552 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph, 7553 struct tcphdr *, th, u32, th_len) 7554 { 7555 #if IS_BUILTIN(CONFIG_IPV6) 7556 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 7557 sizeof(struct ipv6hdr); 7558 u32 cookie; 7559 u16 mss; 7560 7561 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7562 return -EINVAL; 7563 7564 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp; 7565 cookie = __cookie_v6_init_sequence(iph, th, &mss); 7566 7567 return cookie | ((u64)mss << 32); 7568 #else 7569 return -EPROTONOSUPPORT; 7570 #endif 7571 } 7572 7573 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = { 7574 .func = bpf_tcp_raw_gen_syncookie_ipv6, 7575 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */ 7576 .pkt_access = true, 7577 .ret_type = RET_INTEGER, 7578 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7579 .arg1_size = sizeof(struct ipv6hdr), 7580 .arg2_type = ARG_PTR_TO_MEM, 7581 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 7582 }; 7583 7584 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph, 7585 struct tcphdr *, th) 7586 { 7587 u32 cookie = ntohl(th->ack_seq) - 1; 7588 7589 if (__cookie_v4_check(iph, th, cookie) > 0) 7590 return 0; 7591 7592 return -EACCES; 7593 } 7594 7595 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = { 7596 .func = bpf_tcp_raw_check_syncookie_ipv4, 7597 .gpl_only = true, /* __cookie_v4_check is GPL */ 7598 .pkt_access = true, 7599 .ret_type = RET_INTEGER, 7600 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7601 .arg1_size = sizeof(struct iphdr), 7602 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7603 .arg2_size = sizeof(struct tcphdr), 7604 }; 7605 7606 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph, 7607 struct tcphdr *, th) 7608 { 7609 #if IS_BUILTIN(CONFIG_IPV6) 7610 u32 cookie = ntohl(th->ack_seq) - 1; 7611 7612 if (__cookie_v6_check(iph, th, cookie) > 0) 7613 return 0; 7614 7615 return -EACCES; 7616 #else 7617 return -EPROTONOSUPPORT; 7618 #endif 7619 } 7620 7621 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = { 7622 .func = bpf_tcp_raw_check_syncookie_ipv6, 7623 .gpl_only = true, /* __cookie_v6_check is GPL */ 7624 .pkt_access = true, 7625 .ret_type = RET_INTEGER, 7626 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7627 .arg1_size = sizeof(struct ipv6hdr), 7628 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7629 .arg2_size = sizeof(struct tcphdr), 7630 }; 7631 #endif /* CONFIG_SYN_COOKIES */ 7632 7633 #endif /* CONFIG_INET */ 7634 7635 bool bpf_helper_changes_pkt_data(void *func) 7636 { 7637 if (func == bpf_skb_vlan_push || 7638 func == bpf_skb_vlan_pop || 7639 func == bpf_skb_store_bytes || 7640 func == bpf_skb_change_proto || 7641 func == bpf_skb_change_head || 7642 func == sk_skb_change_head || 7643 func == bpf_skb_change_tail || 7644 func == sk_skb_change_tail || 7645 func == bpf_skb_adjust_room || 7646 func == sk_skb_adjust_room || 7647 func == bpf_skb_pull_data || 7648 func == sk_skb_pull_data || 7649 func == bpf_clone_redirect || 7650 func == bpf_l3_csum_replace || 7651 func == bpf_l4_csum_replace || 7652 func == bpf_xdp_adjust_head || 7653 func == bpf_xdp_adjust_meta || 7654 func == bpf_msg_pull_data || 7655 func == bpf_msg_push_data || 7656 func == bpf_msg_pop_data || 7657 func == bpf_xdp_adjust_tail || 7658 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7659 func == bpf_lwt_seg6_store_bytes || 7660 func == bpf_lwt_seg6_adjust_srh || 7661 func == bpf_lwt_seg6_action || 7662 #endif 7663 #ifdef CONFIG_INET 7664 func == bpf_sock_ops_store_hdr_opt || 7665 #endif 7666 func == bpf_lwt_in_push_encap || 7667 func == bpf_lwt_xmit_push_encap) 7668 return true; 7669 7670 return false; 7671 } 7672 7673 const struct bpf_func_proto bpf_event_output_data_proto __weak; 7674 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 7675 7676 static const struct bpf_func_proto * 7677 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7678 { 7679 const struct bpf_func_proto *func_proto; 7680 7681 func_proto = cgroup_common_func_proto(func_id, prog); 7682 if (func_proto) 7683 return func_proto; 7684 7685 func_proto = cgroup_current_func_proto(func_id, prog); 7686 if (func_proto) 7687 return func_proto; 7688 7689 switch (func_id) { 7690 case BPF_FUNC_get_socket_cookie: 7691 return &bpf_get_socket_cookie_sock_proto; 7692 case BPF_FUNC_get_netns_cookie: 7693 return &bpf_get_netns_cookie_sock_proto; 7694 case BPF_FUNC_perf_event_output: 7695 return &bpf_event_output_data_proto; 7696 case BPF_FUNC_sk_storage_get: 7697 return &bpf_sk_storage_get_cg_sock_proto; 7698 case BPF_FUNC_ktime_get_coarse_ns: 7699 return &bpf_ktime_get_coarse_ns_proto; 7700 default: 7701 return bpf_base_func_proto(func_id); 7702 } 7703 } 7704 7705 static const struct bpf_func_proto * 7706 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7707 { 7708 const struct bpf_func_proto *func_proto; 7709 7710 func_proto = cgroup_common_func_proto(func_id, prog); 7711 if (func_proto) 7712 return func_proto; 7713 7714 func_proto = cgroup_current_func_proto(func_id, prog); 7715 if (func_proto) 7716 return func_proto; 7717 7718 switch (func_id) { 7719 case BPF_FUNC_bind: 7720 switch (prog->expected_attach_type) { 7721 case BPF_CGROUP_INET4_CONNECT: 7722 case BPF_CGROUP_INET6_CONNECT: 7723 return &bpf_bind_proto; 7724 default: 7725 return NULL; 7726 } 7727 case BPF_FUNC_get_socket_cookie: 7728 return &bpf_get_socket_cookie_sock_addr_proto; 7729 case BPF_FUNC_get_netns_cookie: 7730 return &bpf_get_netns_cookie_sock_addr_proto; 7731 case BPF_FUNC_perf_event_output: 7732 return &bpf_event_output_data_proto; 7733 #ifdef CONFIG_INET 7734 case BPF_FUNC_sk_lookup_tcp: 7735 return &bpf_sock_addr_sk_lookup_tcp_proto; 7736 case BPF_FUNC_sk_lookup_udp: 7737 return &bpf_sock_addr_sk_lookup_udp_proto; 7738 case BPF_FUNC_sk_release: 7739 return &bpf_sk_release_proto; 7740 case BPF_FUNC_skc_lookup_tcp: 7741 return &bpf_sock_addr_skc_lookup_tcp_proto; 7742 #endif /* CONFIG_INET */ 7743 case BPF_FUNC_sk_storage_get: 7744 return &bpf_sk_storage_get_proto; 7745 case BPF_FUNC_sk_storage_delete: 7746 return &bpf_sk_storage_delete_proto; 7747 case BPF_FUNC_setsockopt: 7748 switch (prog->expected_attach_type) { 7749 case BPF_CGROUP_INET4_BIND: 7750 case BPF_CGROUP_INET6_BIND: 7751 case BPF_CGROUP_INET4_CONNECT: 7752 case BPF_CGROUP_INET6_CONNECT: 7753 case BPF_CGROUP_UDP4_RECVMSG: 7754 case BPF_CGROUP_UDP6_RECVMSG: 7755 case BPF_CGROUP_UDP4_SENDMSG: 7756 case BPF_CGROUP_UDP6_SENDMSG: 7757 case BPF_CGROUP_INET4_GETPEERNAME: 7758 case BPF_CGROUP_INET6_GETPEERNAME: 7759 case BPF_CGROUP_INET4_GETSOCKNAME: 7760 case BPF_CGROUP_INET6_GETSOCKNAME: 7761 return &bpf_sock_addr_setsockopt_proto; 7762 default: 7763 return NULL; 7764 } 7765 case BPF_FUNC_getsockopt: 7766 switch (prog->expected_attach_type) { 7767 case BPF_CGROUP_INET4_BIND: 7768 case BPF_CGROUP_INET6_BIND: 7769 case BPF_CGROUP_INET4_CONNECT: 7770 case BPF_CGROUP_INET6_CONNECT: 7771 case BPF_CGROUP_UDP4_RECVMSG: 7772 case BPF_CGROUP_UDP6_RECVMSG: 7773 case BPF_CGROUP_UDP4_SENDMSG: 7774 case BPF_CGROUP_UDP6_SENDMSG: 7775 case BPF_CGROUP_INET4_GETPEERNAME: 7776 case BPF_CGROUP_INET6_GETPEERNAME: 7777 case BPF_CGROUP_INET4_GETSOCKNAME: 7778 case BPF_CGROUP_INET6_GETSOCKNAME: 7779 return &bpf_sock_addr_getsockopt_proto; 7780 default: 7781 return NULL; 7782 } 7783 default: 7784 return bpf_sk_base_func_proto(func_id); 7785 } 7786 } 7787 7788 static const struct bpf_func_proto * 7789 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7790 { 7791 switch (func_id) { 7792 case BPF_FUNC_skb_load_bytes: 7793 return &bpf_skb_load_bytes_proto; 7794 case BPF_FUNC_skb_load_bytes_relative: 7795 return &bpf_skb_load_bytes_relative_proto; 7796 case BPF_FUNC_get_socket_cookie: 7797 return &bpf_get_socket_cookie_proto; 7798 case BPF_FUNC_get_socket_uid: 7799 return &bpf_get_socket_uid_proto; 7800 case BPF_FUNC_perf_event_output: 7801 return &bpf_skb_event_output_proto; 7802 default: 7803 return bpf_sk_base_func_proto(func_id); 7804 } 7805 } 7806 7807 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7808 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7809 7810 static const struct bpf_func_proto * 7811 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7812 { 7813 const struct bpf_func_proto *func_proto; 7814 7815 func_proto = cgroup_common_func_proto(func_id, prog); 7816 if (func_proto) 7817 return func_proto; 7818 7819 switch (func_id) { 7820 case BPF_FUNC_sk_fullsock: 7821 return &bpf_sk_fullsock_proto; 7822 case BPF_FUNC_sk_storage_get: 7823 return &bpf_sk_storage_get_proto; 7824 case BPF_FUNC_sk_storage_delete: 7825 return &bpf_sk_storage_delete_proto; 7826 case BPF_FUNC_perf_event_output: 7827 return &bpf_skb_event_output_proto; 7828 #ifdef CONFIG_SOCK_CGROUP_DATA 7829 case BPF_FUNC_skb_cgroup_id: 7830 return &bpf_skb_cgroup_id_proto; 7831 case BPF_FUNC_skb_ancestor_cgroup_id: 7832 return &bpf_skb_ancestor_cgroup_id_proto; 7833 case BPF_FUNC_sk_cgroup_id: 7834 return &bpf_sk_cgroup_id_proto; 7835 case BPF_FUNC_sk_ancestor_cgroup_id: 7836 return &bpf_sk_ancestor_cgroup_id_proto; 7837 #endif 7838 #ifdef CONFIG_INET 7839 case BPF_FUNC_sk_lookup_tcp: 7840 return &bpf_sk_lookup_tcp_proto; 7841 case BPF_FUNC_sk_lookup_udp: 7842 return &bpf_sk_lookup_udp_proto; 7843 case BPF_FUNC_sk_release: 7844 return &bpf_sk_release_proto; 7845 case BPF_FUNC_skc_lookup_tcp: 7846 return &bpf_skc_lookup_tcp_proto; 7847 case BPF_FUNC_tcp_sock: 7848 return &bpf_tcp_sock_proto; 7849 case BPF_FUNC_get_listener_sock: 7850 return &bpf_get_listener_sock_proto; 7851 case BPF_FUNC_skb_ecn_set_ce: 7852 return &bpf_skb_ecn_set_ce_proto; 7853 #endif 7854 default: 7855 return sk_filter_func_proto(func_id, prog); 7856 } 7857 } 7858 7859 static const struct bpf_func_proto * 7860 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7861 { 7862 switch (func_id) { 7863 case BPF_FUNC_skb_store_bytes: 7864 return &bpf_skb_store_bytes_proto; 7865 case BPF_FUNC_skb_load_bytes: 7866 return &bpf_skb_load_bytes_proto; 7867 case BPF_FUNC_skb_load_bytes_relative: 7868 return &bpf_skb_load_bytes_relative_proto; 7869 case BPF_FUNC_skb_pull_data: 7870 return &bpf_skb_pull_data_proto; 7871 case BPF_FUNC_csum_diff: 7872 return &bpf_csum_diff_proto; 7873 case BPF_FUNC_csum_update: 7874 return &bpf_csum_update_proto; 7875 case BPF_FUNC_csum_level: 7876 return &bpf_csum_level_proto; 7877 case BPF_FUNC_l3_csum_replace: 7878 return &bpf_l3_csum_replace_proto; 7879 case BPF_FUNC_l4_csum_replace: 7880 return &bpf_l4_csum_replace_proto; 7881 case BPF_FUNC_clone_redirect: 7882 return &bpf_clone_redirect_proto; 7883 case BPF_FUNC_get_cgroup_classid: 7884 return &bpf_get_cgroup_classid_proto; 7885 case BPF_FUNC_skb_vlan_push: 7886 return &bpf_skb_vlan_push_proto; 7887 case BPF_FUNC_skb_vlan_pop: 7888 return &bpf_skb_vlan_pop_proto; 7889 case BPF_FUNC_skb_change_proto: 7890 return &bpf_skb_change_proto_proto; 7891 case BPF_FUNC_skb_change_type: 7892 return &bpf_skb_change_type_proto; 7893 case BPF_FUNC_skb_adjust_room: 7894 return &bpf_skb_adjust_room_proto; 7895 case BPF_FUNC_skb_change_tail: 7896 return &bpf_skb_change_tail_proto; 7897 case BPF_FUNC_skb_change_head: 7898 return &bpf_skb_change_head_proto; 7899 case BPF_FUNC_skb_get_tunnel_key: 7900 return &bpf_skb_get_tunnel_key_proto; 7901 case BPF_FUNC_skb_set_tunnel_key: 7902 return bpf_get_skb_set_tunnel_proto(func_id); 7903 case BPF_FUNC_skb_get_tunnel_opt: 7904 return &bpf_skb_get_tunnel_opt_proto; 7905 case BPF_FUNC_skb_set_tunnel_opt: 7906 return bpf_get_skb_set_tunnel_proto(func_id); 7907 case BPF_FUNC_redirect: 7908 return &bpf_redirect_proto; 7909 case BPF_FUNC_redirect_neigh: 7910 return &bpf_redirect_neigh_proto; 7911 case BPF_FUNC_redirect_peer: 7912 return &bpf_redirect_peer_proto; 7913 case BPF_FUNC_get_route_realm: 7914 return &bpf_get_route_realm_proto; 7915 case BPF_FUNC_get_hash_recalc: 7916 return &bpf_get_hash_recalc_proto; 7917 case BPF_FUNC_set_hash_invalid: 7918 return &bpf_set_hash_invalid_proto; 7919 case BPF_FUNC_set_hash: 7920 return &bpf_set_hash_proto; 7921 case BPF_FUNC_perf_event_output: 7922 return &bpf_skb_event_output_proto; 7923 case BPF_FUNC_get_smp_processor_id: 7924 return &bpf_get_smp_processor_id_proto; 7925 case BPF_FUNC_skb_under_cgroup: 7926 return &bpf_skb_under_cgroup_proto; 7927 case BPF_FUNC_get_socket_cookie: 7928 return &bpf_get_socket_cookie_proto; 7929 case BPF_FUNC_get_socket_uid: 7930 return &bpf_get_socket_uid_proto; 7931 case BPF_FUNC_fib_lookup: 7932 return &bpf_skb_fib_lookup_proto; 7933 case BPF_FUNC_check_mtu: 7934 return &bpf_skb_check_mtu_proto; 7935 case BPF_FUNC_sk_fullsock: 7936 return &bpf_sk_fullsock_proto; 7937 case BPF_FUNC_sk_storage_get: 7938 return &bpf_sk_storage_get_proto; 7939 case BPF_FUNC_sk_storage_delete: 7940 return &bpf_sk_storage_delete_proto; 7941 #ifdef CONFIG_XFRM 7942 case BPF_FUNC_skb_get_xfrm_state: 7943 return &bpf_skb_get_xfrm_state_proto; 7944 #endif 7945 #ifdef CONFIG_CGROUP_NET_CLASSID 7946 case BPF_FUNC_skb_cgroup_classid: 7947 return &bpf_skb_cgroup_classid_proto; 7948 #endif 7949 #ifdef CONFIG_SOCK_CGROUP_DATA 7950 case BPF_FUNC_skb_cgroup_id: 7951 return &bpf_skb_cgroup_id_proto; 7952 case BPF_FUNC_skb_ancestor_cgroup_id: 7953 return &bpf_skb_ancestor_cgroup_id_proto; 7954 #endif 7955 #ifdef CONFIG_INET 7956 case BPF_FUNC_sk_lookup_tcp: 7957 return &bpf_sk_lookup_tcp_proto; 7958 case BPF_FUNC_sk_lookup_udp: 7959 return &bpf_sk_lookup_udp_proto; 7960 case BPF_FUNC_sk_release: 7961 return &bpf_sk_release_proto; 7962 case BPF_FUNC_tcp_sock: 7963 return &bpf_tcp_sock_proto; 7964 case BPF_FUNC_get_listener_sock: 7965 return &bpf_get_listener_sock_proto; 7966 case BPF_FUNC_skc_lookup_tcp: 7967 return &bpf_skc_lookup_tcp_proto; 7968 case BPF_FUNC_tcp_check_syncookie: 7969 return &bpf_tcp_check_syncookie_proto; 7970 case BPF_FUNC_skb_ecn_set_ce: 7971 return &bpf_skb_ecn_set_ce_proto; 7972 case BPF_FUNC_tcp_gen_syncookie: 7973 return &bpf_tcp_gen_syncookie_proto; 7974 case BPF_FUNC_sk_assign: 7975 return &bpf_sk_assign_proto; 7976 case BPF_FUNC_skb_set_tstamp: 7977 return &bpf_skb_set_tstamp_proto; 7978 #ifdef CONFIG_SYN_COOKIES 7979 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 7980 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 7981 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 7982 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 7983 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 7984 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 7985 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 7986 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 7987 #endif 7988 #endif 7989 default: 7990 return bpf_sk_base_func_proto(func_id); 7991 } 7992 } 7993 7994 static const struct bpf_func_proto * 7995 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7996 { 7997 switch (func_id) { 7998 case BPF_FUNC_perf_event_output: 7999 return &bpf_xdp_event_output_proto; 8000 case BPF_FUNC_get_smp_processor_id: 8001 return &bpf_get_smp_processor_id_proto; 8002 case BPF_FUNC_csum_diff: 8003 return &bpf_csum_diff_proto; 8004 case BPF_FUNC_xdp_adjust_head: 8005 return &bpf_xdp_adjust_head_proto; 8006 case BPF_FUNC_xdp_adjust_meta: 8007 return &bpf_xdp_adjust_meta_proto; 8008 case BPF_FUNC_redirect: 8009 return &bpf_xdp_redirect_proto; 8010 case BPF_FUNC_redirect_map: 8011 return &bpf_xdp_redirect_map_proto; 8012 case BPF_FUNC_xdp_adjust_tail: 8013 return &bpf_xdp_adjust_tail_proto; 8014 case BPF_FUNC_xdp_get_buff_len: 8015 return &bpf_xdp_get_buff_len_proto; 8016 case BPF_FUNC_xdp_load_bytes: 8017 return &bpf_xdp_load_bytes_proto; 8018 case BPF_FUNC_xdp_store_bytes: 8019 return &bpf_xdp_store_bytes_proto; 8020 case BPF_FUNC_fib_lookup: 8021 return &bpf_xdp_fib_lookup_proto; 8022 case BPF_FUNC_check_mtu: 8023 return &bpf_xdp_check_mtu_proto; 8024 #ifdef CONFIG_INET 8025 case BPF_FUNC_sk_lookup_udp: 8026 return &bpf_xdp_sk_lookup_udp_proto; 8027 case BPF_FUNC_sk_lookup_tcp: 8028 return &bpf_xdp_sk_lookup_tcp_proto; 8029 case BPF_FUNC_sk_release: 8030 return &bpf_sk_release_proto; 8031 case BPF_FUNC_skc_lookup_tcp: 8032 return &bpf_xdp_skc_lookup_tcp_proto; 8033 case BPF_FUNC_tcp_check_syncookie: 8034 return &bpf_tcp_check_syncookie_proto; 8035 case BPF_FUNC_tcp_gen_syncookie: 8036 return &bpf_tcp_gen_syncookie_proto; 8037 #ifdef CONFIG_SYN_COOKIES 8038 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 8039 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 8040 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 8041 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 8042 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 8043 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 8044 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 8045 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 8046 #endif 8047 #endif 8048 default: 8049 return bpf_sk_base_func_proto(func_id); 8050 } 8051 8052 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES) 8053 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The 8054 * kfuncs are defined in two different modules, and we want to be able 8055 * to use them interchangably with the same BTF type ID. Because modules 8056 * can't de-duplicate BTF IDs between each other, we need the type to be 8057 * referenced in the vmlinux BTF or the verifier will get confused about 8058 * the different types. So we add this dummy type reference which will 8059 * be included in vmlinux BTF, allowing both modules to refer to the 8060 * same type ID. 8061 */ 8062 BTF_TYPE_EMIT(struct nf_conn___init); 8063 #endif 8064 } 8065 8066 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 8067 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 8068 8069 static const struct bpf_func_proto * 8070 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8071 { 8072 const struct bpf_func_proto *func_proto; 8073 8074 func_proto = cgroup_common_func_proto(func_id, prog); 8075 if (func_proto) 8076 return func_proto; 8077 8078 switch (func_id) { 8079 case BPF_FUNC_setsockopt: 8080 return &bpf_sock_ops_setsockopt_proto; 8081 case BPF_FUNC_getsockopt: 8082 return &bpf_sock_ops_getsockopt_proto; 8083 case BPF_FUNC_sock_ops_cb_flags_set: 8084 return &bpf_sock_ops_cb_flags_set_proto; 8085 case BPF_FUNC_sock_map_update: 8086 return &bpf_sock_map_update_proto; 8087 case BPF_FUNC_sock_hash_update: 8088 return &bpf_sock_hash_update_proto; 8089 case BPF_FUNC_get_socket_cookie: 8090 return &bpf_get_socket_cookie_sock_ops_proto; 8091 case BPF_FUNC_perf_event_output: 8092 return &bpf_event_output_data_proto; 8093 case BPF_FUNC_sk_storage_get: 8094 return &bpf_sk_storage_get_proto; 8095 case BPF_FUNC_sk_storage_delete: 8096 return &bpf_sk_storage_delete_proto; 8097 case BPF_FUNC_get_netns_cookie: 8098 return &bpf_get_netns_cookie_sock_ops_proto; 8099 #ifdef CONFIG_INET 8100 case BPF_FUNC_load_hdr_opt: 8101 return &bpf_sock_ops_load_hdr_opt_proto; 8102 case BPF_FUNC_store_hdr_opt: 8103 return &bpf_sock_ops_store_hdr_opt_proto; 8104 case BPF_FUNC_reserve_hdr_opt: 8105 return &bpf_sock_ops_reserve_hdr_opt_proto; 8106 case BPF_FUNC_tcp_sock: 8107 return &bpf_tcp_sock_proto; 8108 #endif /* CONFIG_INET */ 8109 default: 8110 return bpf_sk_base_func_proto(func_id); 8111 } 8112 } 8113 8114 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 8115 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 8116 8117 static const struct bpf_func_proto * 8118 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8119 { 8120 switch (func_id) { 8121 case BPF_FUNC_msg_redirect_map: 8122 return &bpf_msg_redirect_map_proto; 8123 case BPF_FUNC_msg_redirect_hash: 8124 return &bpf_msg_redirect_hash_proto; 8125 case BPF_FUNC_msg_apply_bytes: 8126 return &bpf_msg_apply_bytes_proto; 8127 case BPF_FUNC_msg_cork_bytes: 8128 return &bpf_msg_cork_bytes_proto; 8129 case BPF_FUNC_msg_pull_data: 8130 return &bpf_msg_pull_data_proto; 8131 case BPF_FUNC_msg_push_data: 8132 return &bpf_msg_push_data_proto; 8133 case BPF_FUNC_msg_pop_data: 8134 return &bpf_msg_pop_data_proto; 8135 case BPF_FUNC_perf_event_output: 8136 return &bpf_event_output_data_proto; 8137 case BPF_FUNC_get_current_uid_gid: 8138 return &bpf_get_current_uid_gid_proto; 8139 case BPF_FUNC_get_current_pid_tgid: 8140 return &bpf_get_current_pid_tgid_proto; 8141 case BPF_FUNC_sk_storage_get: 8142 return &bpf_sk_storage_get_proto; 8143 case BPF_FUNC_sk_storage_delete: 8144 return &bpf_sk_storage_delete_proto; 8145 case BPF_FUNC_get_netns_cookie: 8146 return &bpf_get_netns_cookie_sk_msg_proto; 8147 #ifdef CONFIG_CGROUPS 8148 case BPF_FUNC_get_current_cgroup_id: 8149 return &bpf_get_current_cgroup_id_proto; 8150 case BPF_FUNC_get_current_ancestor_cgroup_id: 8151 return &bpf_get_current_ancestor_cgroup_id_proto; 8152 #endif 8153 #ifdef CONFIG_CGROUP_NET_CLASSID 8154 case BPF_FUNC_get_cgroup_classid: 8155 return &bpf_get_cgroup_classid_curr_proto; 8156 #endif 8157 default: 8158 return bpf_sk_base_func_proto(func_id); 8159 } 8160 } 8161 8162 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 8163 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 8164 8165 static const struct bpf_func_proto * 8166 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8167 { 8168 switch (func_id) { 8169 case BPF_FUNC_skb_store_bytes: 8170 return &bpf_skb_store_bytes_proto; 8171 case BPF_FUNC_skb_load_bytes: 8172 return &bpf_skb_load_bytes_proto; 8173 case BPF_FUNC_skb_pull_data: 8174 return &sk_skb_pull_data_proto; 8175 case BPF_FUNC_skb_change_tail: 8176 return &sk_skb_change_tail_proto; 8177 case BPF_FUNC_skb_change_head: 8178 return &sk_skb_change_head_proto; 8179 case BPF_FUNC_skb_adjust_room: 8180 return &sk_skb_adjust_room_proto; 8181 case BPF_FUNC_get_socket_cookie: 8182 return &bpf_get_socket_cookie_proto; 8183 case BPF_FUNC_get_socket_uid: 8184 return &bpf_get_socket_uid_proto; 8185 case BPF_FUNC_sk_redirect_map: 8186 return &bpf_sk_redirect_map_proto; 8187 case BPF_FUNC_sk_redirect_hash: 8188 return &bpf_sk_redirect_hash_proto; 8189 case BPF_FUNC_perf_event_output: 8190 return &bpf_skb_event_output_proto; 8191 #ifdef CONFIG_INET 8192 case BPF_FUNC_sk_lookup_tcp: 8193 return &bpf_sk_lookup_tcp_proto; 8194 case BPF_FUNC_sk_lookup_udp: 8195 return &bpf_sk_lookup_udp_proto; 8196 case BPF_FUNC_sk_release: 8197 return &bpf_sk_release_proto; 8198 case BPF_FUNC_skc_lookup_tcp: 8199 return &bpf_skc_lookup_tcp_proto; 8200 #endif 8201 default: 8202 return bpf_sk_base_func_proto(func_id); 8203 } 8204 } 8205 8206 static const struct bpf_func_proto * 8207 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8208 { 8209 switch (func_id) { 8210 case BPF_FUNC_skb_load_bytes: 8211 return &bpf_flow_dissector_load_bytes_proto; 8212 default: 8213 return bpf_sk_base_func_proto(func_id); 8214 } 8215 } 8216 8217 static const struct bpf_func_proto * 8218 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8219 { 8220 switch (func_id) { 8221 case BPF_FUNC_skb_load_bytes: 8222 return &bpf_skb_load_bytes_proto; 8223 case BPF_FUNC_skb_pull_data: 8224 return &bpf_skb_pull_data_proto; 8225 case BPF_FUNC_csum_diff: 8226 return &bpf_csum_diff_proto; 8227 case BPF_FUNC_get_cgroup_classid: 8228 return &bpf_get_cgroup_classid_proto; 8229 case BPF_FUNC_get_route_realm: 8230 return &bpf_get_route_realm_proto; 8231 case BPF_FUNC_get_hash_recalc: 8232 return &bpf_get_hash_recalc_proto; 8233 case BPF_FUNC_perf_event_output: 8234 return &bpf_skb_event_output_proto; 8235 case BPF_FUNC_get_smp_processor_id: 8236 return &bpf_get_smp_processor_id_proto; 8237 case BPF_FUNC_skb_under_cgroup: 8238 return &bpf_skb_under_cgroup_proto; 8239 default: 8240 return bpf_sk_base_func_proto(func_id); 8241 } 8242 } 8243 8244 static const struct bpf_func_proto * 8245 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8246 { 8247 switch (func_id) { 8248 case BPF_FUNC_lwt_push_encap: 8249 return &bpf_lwt_in_push_encap_proto; 8250 default: 8251 return lwt_out_func_proto(func_id, prog); 8252 } 8253 } 8254 8255 static const struct bpf_func_proto * 8256 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8257 { 8258 switch (func_id) { 8259 case BPF_FUNC_skb_get_tunnel_key: 8260 return &bpf_skb_get_tunnel_key_proto; 8261 case BPF_FUNC_skb_set_tunnel_key: 8262 return bpf_get_skb_set_tunnel_proto(func_id); 8263 case BPF_FUNC_skb_get_tunnel_opt: 8264 return &bpf_skb_get_tunnel_opt_proto; 8265 case BPF_FUNC_skb_set_tunnel_opt: 8266 return bpf_get_skb_set_tunnel_proto(func_id); 8267 case BPF_FUNC_redirect: 8268 return &bpf_redirect_proto; 8269 case BPF_FUNC_clone_redirect: 8270 return &bpf_clone_redirect_proto; 8271 case BPF_FUNC_skb_change_tail: 8272 return &bpf_skb_change_tail_proto; 8273 case BPF_FUNC_skb_change_head: 8274 return &bpf_skb_change_head_proto; 8275 case BPF_FUNC_skb_store_bytes: 8276 return &bpf_skb_store_bytes_proto; 8277 case BPF_FUNC_csum_update: 8278 return &bpf_csum_update_proto; 8279 case BPF_FUNC_csum_level: 8280 return &bpf_csum_level_proto; 8281 case BPF_FUNC_l3_csum_replace: 8282 return &bpf_l3_csum_replace_proto; 8283 case BPF_FUNC_l4_csum_replace: 8284 return &bpf_l4_csum_replace_proto; 8285 case BPF_FUNC_set_hash_invalid: 8286 return &bpf_set_hash_invalid_proto; 8287 case BPF_FUNC_lwt_push_encap: 8288 return &bpf_lwt_xmit_push_encap_proto; 8289 default: 8290 return lwt_out_func_proto(func_id, prog); 8291 } 8292 } 8293 8294 static const struct bpf_func_proto * 8295 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8296 { 8297 switch (func_id) { 8298 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 8299 case BPF_FUNC_lwt_seg6_store_bytes: 8300 return &bpf_lwt_seg6_store_bytes_proto; 8301 case BPF_FUNC_lwt_seg6_action: 8302 return &bpf_lwt_seg6_action_proto; 8303 case BPF_FUNC_lwt_seg6_adjust_srh: 8304 return &bpf_lwt_seg6_adjust_srh_proto; 8305 #endif 8306 default: 8307 return lwt_out_func_proto(func_id, prog); 8308 } 8309 } 8310 8311 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 8312 const struct bpf_prog *prog, 8313 struct bpf_insn_access_aux *info) 8314 { 8315 const int size_default = sizeof(__u32); 8316 8317 if (off < 0 || off >= sizeof(struct __sk_buff)) 8318 return false; 8319 8320 /* The verifier guarantees that size > 0. */ 8321 if (off % size != 0) 8322 return false; 8323 8324 switch (off) { 8325 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8326 if (off + size > offsetofend(struct __sk_buff, cb[4])) 8327 return false; 8328 break; 8329 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 8330 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 8331 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 8332 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 8333 case bpf_ctx_range(struct __sk_buff, data): 8334 case bpf_ctx_range(struct __sk_buff, data_meta): 8335 case bpf_ctx_range(struct __sk_buff, data_end): 8336 if (size != size_default) 8337 return false; 8338 break; 8339 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8340 return false; 8341 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8342 if (type == BPF_WRITE || size != sizeof(__u64)) 8343 return false; 8344 break; 8345 case bpf_ctx_range(struct __sk_buff, tstamp): 8346 if (size != sizeof(__u64)) 8347 return false; 8348 break; 8349 case offsetof(struct __sk_buff, sk): 8350 if (type == BPF_WRITE || size != sizeof(__u64)) 8351 return false; 8352 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 8353 break; 8354 case offsetof(struct __sk_buff, tstamp_type): 8355 return false; 8356 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1: 8357 /* Explicitly prohibit access to padding in __sk_buff. */ 8358 return false; 8359 default: 8360 /* Only narrow read access allowed for now. */ 8361 if (type == BPF_WRITE) { 8362 if (size != size_default) 8363 return false; 8364 } else { 8365 bpf_ctx_record_field_size(info, size_default); 8366 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8367 return false; 8368 } 8369 } 8370 8371 return true; 8372 } 8373 8374 static bool sk_filter_is_valid_access(int off, int size, 8375 enum bpf_access_type type, 8376 const struct bpf_prog *prog, 8377 struct bpf_insn_access_aux *info) 8378 { 8379 switch (off) { 8380 case bpf_ctx_range(struct __sk_buff, tc_classid): 8381 case bpf_ctx_range(struct __sk_buff, data): 8382 case bpf_ctx_range(struct __sk_buff, data_meta): 8383 case bpf_ctx_range(struct __sk_buff, data_end): 8384 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8385 case bpf_ctx_range(struct __sk_buff, tstamp): 8386 case bpf_ctx_range(struct __sk_buff, wire_len): 8387 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8388 return false; 8389 } 8390 8391 if (type == BPF_WRITE) { 8392 switch (off) { 8393 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8394 break; 8395 default: 8396 return false; 8397 } 8398 } 8399 8400 return bpf_skb_is_valid_access(off, size, type, prog, info); 8401 } 8402 8403 static bool cg_skb_is_valid_access(int off, int size, 8404 enum bpf_access_type type, 8405 const struct bpf_prog *prog, 8406 struct bpf_insn_access_aux *info) 8407 { 8408 switch (off) { 8409 case bpf_ctx_range(struct __sk_buff, tc_classid): 8410 case bpf_ctx_range(struct __sk_buff, data_meta): 8411 case bpf_ctx_range(struct __sk_buff, wire_len): 8412 return false; 8413 case bpf_ctx_range(struct __sk_buff, data): 8414 case bpf_ctx_range(struct __sk_buff, data_end): 8415 if (!bpf_capable()) 8416 return false; 8417 break; 8418 } 8419 8420 if (type == BPF_WRITE) { 8421 switch (off) { 8422 case bpf_ctx_range(struct __sk_buff, mark): 8423 case bpf_ctx_range(struct __sk_buff, priority): 8424 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8425 break; 8426 case bpf_ctx_range(struct __sk_buff, tstamp): 8427 if (!bpf_capable()) 8428 return false; 8429 break; 8430 default: 8431 return false; 8432 } 8433 } 8434 8435 switch (off) { 8436 case bpf_ctx_range(struct __sk_buff, data): 8437 info->reg_type = PTR_TO_PACKET; 8438 break; 8439 case bpf_ctx_range(struct __sk_buff, data_end): 8440 info->reg_type = PTR_TO_PACKET_END; 8441 break; 8442 } 8443 8444 return bpf_skb_is_valid_access(off, size, type, prog, info); 8445 } 8446 8447 static bool lwt_is_valid_access(int off, int size, 8448 enum bpf_access_type type, 8449 const struct bpf_prog *prog, 8450 struct bpf_insn_access_aux *info) 8451 { 8452 switch (off) { 8453 case bpf_ctx_range(struct __sk_buff, tc_classid): 8454 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8455 case bpf_ctx_range(struct __sk_buff, data_meta): 8456 case bpf_ctx_range(struct __sk_buff, tstamp): 8457 case bpf_ctx_range(struct __sk_buff, wire_len): 8458 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8459 return false; 8460 } 8461 8462 if (type == BPF_WRITE) { 8463 switch (off) { 8464 case bpf_ctx_range(struct __sk_buff, mark): 8465 case bpf_ctx_range(struct __sk_buff, priority): 8466 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8467 break; 8468 default: 8469 return false; 8470 } 8471 } 8472 8473 switch (off) { 8474 case bpf_ctx_range(struct __sk_buff, data): 8475 info->reg_type = PTR_TO_PACKET; 8476 break; 8477 case bpf_ctx_range(struct __sk_buff, data_end): 8478 info->reg_type = PTR_TO_PACKET_END; 8479 break; 8480 } 8481 8482 return bpf_skb_is_valid_access(off, size, type, prog, info); 8483 } 8484 8485 /* Attach type specific accesses */ 8486 static bool __sock_filter_check_attach_type(int off, 8487 enum bpf_access_type access_type, 8488 enum bpf_attach_type attach_type) 8489 { 8490 switch (off) { 8491 case offsetof(struct bpf_sock, bound_dev_if): 8492 case offsetof(struct bpf_sock, mark): 8493 case offsetof(struct bpf_sock, priority): 8494 switch (attach_type) { 8495 case BPF_CGROUP_INET_SOCK_CREATE: 8496 case BPF_CGROUP_INET_SOCK_RELEASE: 8497 goto full_access; 8498 default: 8499 return false; 8500 } 8501 case bpf_ctx_range(struct bpf_sock, src_ip4): 8502 switch (attach_type) { 8503 case BPF_CGROUP_INET4_POST_BIND: 8504 goto read_only; 8505 default: 8506 return false; 8507 } 8508 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8509 switch (attach_type) { 8510 case BPF_CGROUP_INET6_POST_BIND: 8511 goto read_only; 8512 default: 8513 return false; 8514 } 8515 case bpf_ctx_range(struct bpf_sock, src_port): 8516 switch (attach_type) { 8517 case BPF_CGROUP_INET4_POST_BIND: 8518 case BPF_CGROUP_INET6_POST_BIND: 8519 goto read_only; 8520 default: 8521 return false; 8522 } 8523 } 8524 read_only: 8525 return access_type == BPF_READ; 8526 full_access: 8527 return true; 8528 } 8529 8530 bool bpf_sock_common_is_valid_access(int off, int size, 8531 enum bpf_access_type type, 8532 struct bpf_insn_access_aux *info) 8533 { 8534 switch (off) { 8535 case bpf_ctx_range_till(struct bpf_sock, type, priority): 8536 return false; 8537 default: 8538 return bpf_sock_is_valid_access(off, size, type, info); 8539 } 8540 } 8541 8542 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 8543 struct bpf_insn_access_aux *info) 8544 { 8545 const int size_default = sizeof(__u32); 8546 int field_size; 8547 8548 if (off < 0 || off >= sizeof(struct bpf_sock)) 8549 return false; 8550 if (off % size != 0) 8551 return false; 8552 8553 switch (off) { 8554 case offsetof(struct bpf_sock, state): 8555 case offsetof(struct bpf_sock, family): 8556 case offsetof(struct bpf_sock, type): 8557 case offsetof(struct bpf_sock, protocol): 8558 case offsetof(struct bpf_sock, src_port): 8559 case offsetof(struct bpf_sock, rx_queue_mapping): 8560 case bpf_ctx_range(struct bpf_sock, src_ip4): 8561 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8562 case bpf_ctx_range(struct bpf_sock, dst_ip4): 8563 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 8564 bpf_ctx_record_field_size(info, size_default); 8565 return bpf_ctx_narrow_access_ok(off, size, size_default); 8566 case bpf_ctx_range(struct bpf_sock, dst_port): 8567 field_size = size == size_default ? 8568 size_default : sizeof_field(struct bpf_sock, dst_port); 8569 bpf_ctx_record_field_size(info, field_size); 8570 return bpf_ctx_narrow_access_ok(off, size, field_size); 8571 case offsetofend(struct bpf_sock, dst_port) ... 8572 offsetof(struct bpf_sock, dst_ip4) - 1: 8573 return false; 8574 } 8575 8576 return size == size_default; 8577 } 8578 8579 static bool sock_filter_is_valid_access(int off, int size, 8580 enum bpf_access_type type, 8581 const struct bpf_prog *prog, 8582 struct bpf_insn_access_aux *info) 8583 { 8584 if (!bpf_sock_is_valid_access(off, size, type, info)) 8585 return false; 8586 return __sock_filter_check_attach_type(off, type, 8587 prog->expected_attach_type); 8588 } 8589 8590 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 8591 const struct bpf_prog *prog) 8592 { 8593 /* Neither direct read nor direct write requires any preliminary 8594 * action. 8595 */ 8596 return 0; 8597 } 8598 8599 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 8600 const struct bpf_prog *prog, int drop_verdict) 8601 { 8602 struct bpf_insn *insn = insn_buf; 8603 8604 if (!direct_write) 8605 return 0; 8606 8607 /* if (!skb->cloned) 8608 * goto start; 8609 * 8610 * (Fast-path, otherwise approximation that we might be 8611 * a clone, do the rest in helper.) 8612 */ 8613 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET); 8614 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 8615 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 8616 8617 /* ret = bpf_skb_pull_data(skb, 0); */ 8618 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 8619 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 8620 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 8621 BPF_FUNC_skb_pull_data); 8622 /* if (!ret) 8623 * goto restore; 8624 * return TC_ACT_SHOT; 8625 */ 8626 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 8627 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 8628 *insn++ = BPF_EXIT_INSN(); 8629 8630 /* restore: */ 8631 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 8632 /* start: */ 8633 *insn++ = prog->insnsi[0]; 8634 8635 return insn - insn_buf; 8636 } 8637 8638 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 8639 struct bpf_insn *insn_buf) 8640 { 8641 bool indirect = BPF_MODE(orig->code) == BPF_IND; 8642 struct bpf_insn *insn = insn_buf; 8643 8644 if (!indirect) { 8645 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 8646 } else { 8647 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 8648 if (orig->imm) 8649 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 8650 } 8651 /* We're guaranteed here that CTX is in R6. */ 8652 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 8653 8654 switch (BPF_SIZE(orig->code)) { 8655 case BPF_B: 8656 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 8657 break; 8658 case BPF_H: 8659 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 8660 break; 8661 case BPF_W: 8662 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 8663 break; 8664 } 8665 8666 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 8667 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 8668 *insn++ = BPF_EXIT_INSN(); 8669 8670 return insn - insn_buf; 8671 } 8672 8673 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 8674 const struct bpf_prog *prog) 8675 { 8676 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 8677 } 8678 8679 static bool tc_cls_act_is_valid_access(int off, int size, 8680 enum bpf_access_type type, 8681 const struct bpf_prog *prog, 8682 struct bpf_insn_access_aux *info) 8683 { 8684 if (type == BPF_WRITE) { 8685 switch (off) { 8686 case bpf_ctx_range(struct __sk_buff, mark): 8687 case bpf_ctx_range(struct __sk_buff, tc_index): 8688 case bpf_ctx_range(struct __sk_buff, priority): 8689 case bpf_ctx_range(struct __sk_buff, tc_classid): 8690 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8691 case bpf_ctx_range(struct __sk_buff, tstamp): 8692 case bpf_ctx_range(struct __sk_buff, queue_mapping): 8693 break; 8694 default: 8695 return false; 8696 } 8697 } 8698 8699 switch (off) { 8700 case bpf_ctx_range(struct __sk_buff, data): 8701 info->reg_type = PTR_TO_PACKET; 8702 break; 8703 case bpf_ctx_range(struct __sk_buff, data_meta): 8704 info->reg_type = PTR_TO_PACKET_META; 8705 break; 8706 case bpf_ctx_range(struct __sk_buff, data_end): 8707 info->reg_type = PTR_TO_PACKET_END; 8708 break; 8709 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8710 return false; 8711 case offsetof(struct __sk_buff, tstamp_type): 8712 /* The convert_ctx_access() on reading and writing 8713 * __sk_buff->tstamp depends on whether the bpf prog 8714 * has used __sk_buff->tstamp_type or not. 8715 * Thus, we need to set prog->tstamp_type_access 8716 * earlier during is_valid_access() here. 8717 */ 8718 ((struct bpf_prog *)prog)->tstamp_type_access = 1; 8719 return size == sizeof(__u8); 8720 } 8721 8722 return bpf_skb_is_valid_access(off, size, type, prog, info); 8723 } 8724 8725 DEFINE_MUTEX(nf_conn_btf_access_lock); 8726 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock); 8727 8728 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 8729 const struct bpf_reg_state *reg, 8730 int off, int size, enum bpf_access_type atype, 8731 u32 *next_btf_id, enum bpf_type_flag *flag); 8732 EXPORT_SYMBOL_GPL(nfct_btf_struct_access); 8733 8734 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log, 8735 const struct bpf_reg_state *reg, 8736 int off, int size, enum bpf_access_type atype, 8737 u32 *next_btf_id, enum bpf_type_flag *flag) 8738 { 8739 int ret = -EACCES; 8740 8741 if (atype == BPF_READ) 8742 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8743 8744 mutex_lock(&nf_conn_btf_access_lock); 8745 if (nfct_btf_struct_access) 8746 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8747 mutex_unlock(&nf_conn_btf_access_lock); 8748 8749 return ret; 8750 } 8751 8752 static bool __is_valid_xdp_access(int off, int size) 8753 { 8754 if (off < 0 || off >= sizeof(struct xdp_md)) 8755 return false; 8756 if (off % size != 0) 8757 return false; 8758 if (size != sizeof(__u32)) 8759 return false; 8760 8761 return true; 8762 } 8763 8764 static bool xdp_is_valid_access(int off, int size, 8765 enum bpf_access_type type, 8766 const struct bpf_prog *prog, 8767 struct bpf_insn_access_aux *info) 8768 { 8769 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 8770 switch (off) { 8771 case offsetof(struct xdp_md, egress_ifindex): 8772 return false; 8773 } 8774 } 8775 8776 if (type == BPF_WRITE) { 8777 if (bpf_prog_is_offloaded(prog->aux)) { 8778 switch (off) { 8779 case offsetof(struct xdp_md, rx_queue_index): 8780 return __is_valid_xdp_access(off, size); 8781 } 8782 } 8783 return false; 8784 } 8785 8786 switch (off) { 8787 case offsetof(struct xdp_md, data): 8788 info->reg_type = PTR_TO_PACKET; 8789 break; 8790 case offsetof(struct xdp_md, data_meta): 8791 info->reg_type = PTR_TO_PACKET_META; 8792 break; 8793 case offsetof(struct xdp_md, data_end): 8794 info->reg_type = PTR_TO_PACKET_END; 8795 break; 8796 } 8797 8798 return __is_valid_xdp_access(off, size); 8799 } 8800 8801 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act) 8802 { 8803 const u32 act_max = XDP_REDIRECT; 8804 8805 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n", 8806 act > act_max ? "Illegal" : "Driver unsupported", 8807 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A"); 8808 } 8809 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 8810 8811 static int xdp_btf_struct_access(struct bpf_verifier_log *log, 8812 const struct bpf_reg_state *reg, 8813 int off, int size, enum bpf_access_type atype, 8814 u32 *next_btf_id, enum bpf_type_flag *flag) 8815 { 8816 int ret = -EACCES; 8817 8818 if (atype == BPF_READ) 8819 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8820 8821 mutex_lock(&nf_conn_btf_access_lock); 8822 if (nfct_btf_struct_access) 8823 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag); 8824 mutex_unlock(&nf_conn_btf_access_lock); 8825 8826 return ret; 8827 } 8828 8829 static bool sock_addr_is_valid_access(int off, int size, 8830 enum bpf_access_type type, 8831 const struct bpf_prog *prog, 8832 struct bpf_insn_access_aux *info) 8833 { 8834 const int size_default = sizeof(__u32); 8835 8836 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 8837 return false; 8838 if (off % size != 0) 8839 return false; 8840 8841 /* Disallow access to IPv6 fields from IPv4 contex and vise 8842 * versa. 8843 */ 8844 switch (off) { 8845 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8846 switch (prog->expected_attach_type) { 8847 case BPF_CGROUP_INET4_BIND: 8848 case BPF_CGROUP_INET4_CONNECT: 8849 case BPF_CGROUP_INET4_GETPEERNAME: 8850 case BPF_CGROUP_INET4_GETSOCKNAME: 8851 case BPF_CGROUP_UDP4_SENDMSG: 8852 case BPF_CGROUP_UDP4_RECVMSG: 8853 break; 8854 default: 8855 return false; 8856 } 8857 break; 8858 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8859 switch (prog->expected_attach_type) { 8860 case BPF_CGROUP_INET6_BIND: 8861 case BPF_CGROUP_INET6_CONNECT: 8862 case BPF_CGROUP_INET6_GETPEERNAME: 8863 case BPF_CGROUP_INET6_GETSOCKNAME: 8864 case BPF_CGROUP_UDP6_SENDMSG: 8865 case BPF_CGROUP_UDP6_RECVMSG: 8866 break; 8867 default: 8868 return false; 8869 } 8870 break; 8871 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8872 switch (prog->expected_attach_type) { 8873 case BPF_CGROUP_UDP4_SENDMSG: 8874 break; 8875 default: 8876 return false; 8877 } 8878 break; 8879 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8880 msg_src_ip6[3]): 8881 switch (prog->expected_attach_type) { 8882 case BPF_CGROUP_UDP6_SENDMSG: 8883 break; 8884 default: 8885 return false; 8886 } 8887 break; 8888 } 8889 8890 switch (off) { 8891 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8892 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8893 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8894 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8895 msg_src_ip6[3]): 8896 case bpf_ctx_range(struct bpf_sock_addr, user_port): 8897 if (type == BPF_READ) { 8898 bpf_ctx_record_field_size(info, size_default); 8899 8900 if (bpf_ctx_wide_access_ok(off, size, 8901 struct bpf_sock_addr, 8902 user_ip6)) 8903 return true; 8904 8905 if (bpf_ctx_wide_access_ok(off, size, 8906 struct bpf_sock_addr, 8907 msg_src_ip6)) 8908 return true; 8909 8910 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8911 return false; 8912 } else { 8913 if (bpf_ctx_wide_access_ok(off, size, 8914 struct bpf_sock_addr, 8915 user_ip6)) 8916 return true; 8917 8918 if (bpf_ctx_wide_access_ok(off, size, 8919 struct bpf_sock_addr, 8920 msg_src_ip6)) 8921 return true; 8922 8923 if (size != size_default) 8924 return false; 8925 } 8926 break; 8927 case offsetof(struct bpf_sock_addr, sk): 8928 if (type != BPF_READ) 8929 return false; 8930 if (size != sizeof(__u64)) 8931 return false; 8932 info->reg_type = PTR_TO_SOCKET; 8933 break; 8934 default: 8935 if (type == BPF_READ) { 8936 if (size != size_default) 8937 return false; 8938 } else { 8939 return false; 8940 } 8941 } 8942 8943 return true; 8944 } 8945 8946 static bool sock_ops_is_valid_access(int off, int size, 8947 enum bpf_access_type type, 8948 const struct bpf_prog *prog, 8949 struct bpf_insn_access_aux *info) 8950 { 8951 const int size_default = sizeof(__u32); 8952 8953 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8954 return false; 8955 8956 /* The verifier guarantees that size > 0. */ 8957 if (off % size != 0) 8958 return false; 8959 8960 if (type == BPF_WRITE) { 8961 switch (off) { 8962 case offsetof(struct bpf_sock_ops, reply): 8963 case offsetof(struct bpf_sock_ops, sk_txhash): 8964 if (size != size_default) 8965 return false; 8966 break; 8967 default: 8968 return false; 8969 } 8970 } else { 8971 switch (off) { 8972 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8973 bytes_acked): 8974 if (size != sizeof(__u64)) 8975 return false; 8976 break; 8977 case offsetof(struct bpf_sock_ops, sk): 8978 if (size != sizeof(__u64)) 8979 return false; 8980 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8981 break; 8982 case offsetof(struct bpf_sock_ops, skb_data): 8983 if (size != sizeof(__u64)) 8984 return false; 8985 info->reg_type = PTR_TO_PACKET; 8986 break; 8987 case offsetof(struct bpf_sock_ops, skb_data_end): 8988 if (size != sizeof(__u64)) 8989 return false; 8990 info->reg_type = PTR_TO_PACKET_END; 8991 break; 8992 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 8993 bpf_ctx_record_field_size(info, size_default); 8994 return bpf_ctx_narrow_access_ok(off, size, 8995 size_default); 8996 case offsetof(struct bpf_sock_ops, skb_hwtstamp): 8997 if (size != sizeof(__u64)) 8998 return false; 8999 break; 9000 default: 9001 if (size != size_default) 9002 return false; 9003 break; 9004 } 9005 } 9006 9007 return true; 9008 } 9009 9010 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 9011 const struct bpf_prog *prog) 9012 { 9013 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 9014 } 9015 9016 static bool sk_skb_is_valid_access(int off, int size, 9017 enum bpf_access_type type, 9018 const struct bpf_prog *prog, 9019 struct bpf_insn_access_aux *info) 9020 { 9021 switch (off) { 9022 case bpf_ctx_range(struct __sk_buff, tc_classid): 9023 case bpf_ctx_range(struct __sk_buff, data_meta): 9024 case bpf_ctx_range(struct __sk_buff, tstamp): 9025 case bpf_ctx_range(struct __sk_buff, wire_len): 9026 case bpf_ctx_range(struct __sk_buff, hwtstamp): 9027 return false; 9028 } 9029 9030 if (type == BPF_WRITE) { 9031 switch (off) { 9032 case bpf_ctx_range(struct __sk_buff, tc_index): 9033 case bpf_ctx_range(struct __sk_buff, priority): 9034 break; 9035 default: 9036 return false; 9037 } 9038 } 9039 9040 switch (off) { 9041 case bpf_ctx_range(struct __sk_buff, mark): 9042 return false; 9043 case bpf_ctx_range(struct __sk_buff, data): 9044 info->reg_type = PTR_TO_PACKET; 9045 break; 9046 case bpf_ctx_range(struct __sk_buff, data_end): 9047 info->reg_type = PTR_TO_PACKET_END; 9048 break; 9049 } 9050 9051 return bpf_skb_is_valid_access(off, size, type, prog, info); 9052 } 9053 9054 static bool sk_msg_is_valid_access(int off, int size, 9055 enum bpf_access_type type, 9056 const struct bpf_prog *prog, 9057 struct bpf_insn_access_aux *info) 9058 { 9059 if (type == BPF_WRITE) 9060 return false; 9061 9062 if (off % size != 0) 9063 return false; 9064 9065 switch (off) { 9066 case offsetof(struct sk_msg_md, data): 9067 info->reg_type = PTR_TO_PACKET; 9068 if (size != sizeof(__u64)) 9069 return false; 9070 break; 9071 case offsetof(struct sk_msg_md, data_end): 9072 info->reg_type = PTR_TO_PACKET_END; 9073 if (size != sizeof(__u64)) 9074 return false; 9075 break; 9076 case offsetof(struct sk_msg_md, sk): 9077 if (size != sizeof(__u64)) 9078 return false; 9079 info->reg_type = PTR_TO_SOCKET; 9080 break; 9081 case bpf_ctx_range(struct sk_msg_md, family): 9082 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 9083 case bpf_ctx_range(struct sk_msg_md, local_ip4): 9084 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 9085 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 9086 case bpf_ctx_range(struct sk_msg_md, remote_port): 9087 case bpf_ctx_range(struct sk_msg_md, local_port): 9088 case bpf_ctx_range(struct sk_msg_md, size): 9089 if (size != sizeof(__u32)) 9090 return false; 9091 break; 9092 default: 9093 return false; 9094 } 9095 return true; 9096 } 9097 9098 static bool flow_dissector_is_valid_access(int off, int size, 9099 enum bpf_access_type type, 9100 const struct bpf_prog *prog, 9101 struct bpf_insn_access_aux *info) 9102 { 9103 const int size_default = sizeof(__u32); 9104 9105 if (off < 0 || off >= sizeof(struct __sk_buff)) 9106 return false; 9107 9108 if (type == BPF_WRITE) 9109 return false; 9110 9111 switch (off) { 9112 case bpf_ctx_range(struct __sk_buff, data): 9113 if (size != size_default) 9114 return false; 9115 info->reg_type = PTR_TO_PACKET; 9116 return true; 9117 case bpf_ctx_range(struct __sk_buff, data_end): 9118 if (size != size_default) 9119 return false; 9120 info->reg_type = PTR_TO_PACKET_END; 9121 return true; 9122 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 9123 if (size != sizeof(__u64)) 9124 return false; 9125 info->reg_type = PTR_TO_FLOW_KEYS; 9126 return true; 9127 default: 9128 return false; 9129 } 9130 } 9131 9132 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 9133 const struct bpf_insn *si, 9134 struct bpf_insn *insn_buf, 9135 struct bpf_prog *prog, 9136 u32 *target_size) 9137 9138 { 9139 struct bpf_insn *insn = insn_buf; 9140 9141 switch (si->off) { 9142 case offsetof(struct __sk_buff, data): 9143 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 9144 si->dst_reg, si->src_reg, 9145 offsetof(struct bpf_flow_dissector, data)); 9146 break; 9147 9148 case offsetof(struct __sk_buff, data_end): 9149 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 9150 si->dst_reg, si->src_reg, 9151 offsetof(struct bpf_flow_dissector, data_end)); 9152 break; 9153 9154 case offsetof(struct __sk_buff, flow_keys): 9155 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 9156 si->dst_reg, si->src_reg, 9157 offsetof(struct bpf_flow_dissector, flow_keys)); 9158 break; 9159 } 9160 9161 return insn - insn_buf; 9162 } 9163 9164 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si, 9165 struct bpf_insn *insn) 9166 { 9167 __u8 value_reg = si->dst_reg; 9168 __u8 skb_reg = si->src_reg; 9169 /* AX is needed because src_reg and dst_reg could be the same */ 9170 __u8 tmp_reg = BPF_REG_AX; 9171 9172 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, 9173 PKT_VLAN_PRESENT_OFFSET); 9174 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, 9175 SKB_MONO_DELIVERY_TIME_MASK, 2); 9176 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); 9177 *insn++ = BPF_JMP_A(1); 9178 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO); 9179 9180 return insn; 9181 } 9182 9183 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg, 9184 struct bpf_insn *insn) 9185 { 9186 /* si->dst_reg = skb_shinfo(SKB); */ 9187 #ifdef NET_SKBUFF_DATA_USES_OFFSET 9188 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9189 BPF_REG_AX, skb_reg, 9190 offsetof(struct sk_buff, end)); 9191 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 9192 dst_reg, skb_reg, 9193 offsetof(struct sk_buff, head)); 9194 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX); 9195 #else 9196 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9197 dst_reg, skb_reg, 9198 offsetof(struct sk_buff, end)); 9199 #endif 9200 9201 return insn; 9202 } 9203 9204 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, 9205 const struct bpf_insn *si, 9206 struct bpf_insn *insn) 9207 { 9208 __u8 value_reg = si->dst_reg; 9209 __u8 skb_reg = si->src_reg; 9210 9211 #ifdef CONFIG_NET_CLS_ACT 9212 /* If the tstamp_type is read, 9213 * the bpf prog is aware the tstamp could have delivery time. 9214 * Thus, read skb->tstamp as is if tstamp_type_access is true. 9215 */ 9216 if (!prog->tstamp_type_access) { 9217 /* AX is needed because src_reg and dst_reg could be the same */ 9218 __u8 tmp_reg = BPF_REG_AX; 9219 9220 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 9221 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, 9222 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK); 9223 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg, 9224 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2); 9225 /* skb->tc_at_ingress && skb->mono_delivery_time, 9226 * read 0 as the (rcv) timestamp. 9227 */ 9228 *insn++ = BPF_MOV64_IMM(value_reg, 0); 9229 *insn++ = BPF_JMP_A(1); 9230 } 9231 #endif 9232 9233 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg, 9234 offsetof(struct sk_buff, tstamp)); 9235 return insn; 9236 } 9237 9238 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, 9239 const struct bpf_insn *si, 9240 struct bpf_insn *insn) 9241 { 9242 __u8 value_reg = si->src_reg; 9243 __u8 skb_reg = si->dst_reg; 9244 9245 #ifdef CONFIG_NET_CLS_ACT 9246 /* If the tstamp_type is read, 9247 * the bpf prog is aware the tstamp could have delivery time. 9248 * Thus, write skb->tstamp as is if tstamp_type_access is true. 9249 * Otherwise, writing at ingress will have to clear the 9250 * mono_delivery_time bit also. 9251 */ 9252 if (!prog->tstamp_type_access) { 9253 __u8 tmp_reg = BPF_REG_AX; 9254 9255 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 9256 /* Writing __sk_buff->tstamp as ingress, goto <clear> */ 9257 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); 9258 /* goto <store> */ 9259 *insn++ = BPF_JMP_A(2); 9260 /* <clear>: mono_delivery_time */ 9261 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); 9262 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET); 9263 } 9264 #endif 9265 9266 /* <store>: skb->tstamp = tstamp */ 9267 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg, 9268 offsetof(struct sk_buff, tstamp)); 9269 return insn; 9270 } 9271 9272 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 9273 const struct bpf_insn *si, 9274 struct bpf_insn *insn_buf, 9275 struct bpf_prog *prog, u32 *target_size) 9276 { 9277 struct bpf_insn *insn = insn_buf; 9278 int off; 9279 9280 switch (si->off) { 9281 case offsetof(struct __sk_buff, len): 9282 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9283 bpf_target_off(struct sk_buff, len, 4, 9284 target_size)); 9285 break; 9286 9287 case offsetof(struct __sk_buff, protocol): 9288 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9289 bpf_target_off(struct sk_buff, protocol, 2, 9290 target_size)); 9291 break; 9292 9293 case offsetof(struct __sk_buff, vlan_proto): 9294 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9295 bpf_target_off(struct sk_buff, vlan_proto, 2, 9296 target_size)); 9297 break; 9298 9299 case offsetof(struct __sk_buff, priority): 9300 if (type == BPF_WRITE) 9301 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9302 bpf_target_off(struct sk_buff, priority, 4, 9303 target_size)); 9304 else 9305 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9306 bpf_target_off(struct sk_buff, priority, 4, 9307 target_size)); 9308 break; 9309 9310 case offsetof(struct __sk_buff, ingress_ifindex): 9311 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9312 bpf_target_off(struct sk_buff, skb_iif, 4, 9313 target_size)); 9314 break; 9315 9316 case offsetof(struct __sk_buff, ifindex): 9317 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9318 si->dst_reg, si->src_reg, 9319 offsetof(struct sk_buff, dev)); 9320 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9321 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9322 bpf_target_off(struct net_device, ifindex, 4, 9323 target_size)); 9324 break; 9325 9326 case offsetof(struct __sk_buff, hash): 9327 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9328 bpf_target_off(struct sk_buff, hash, 4, 9329 target_size)); 9330 break; 9331 9332 case offsetof(struct __sk_buff, mark): 9333 if (type == BPF_WRITE) 9334 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9335 bpf_target_off(struct sk_buff, mark, 4, 9336 target_size)); 9337 else 9338 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9339 bpf_target_off(struct sk_buff, mark, 4, 9340 target_size)); 9341 break; 9342 9343 case offsetof(struct __sk_buff, pkt_type): 9344 *target_size = 1; 9345 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 9346 PKT_TYPE_OFFSET); 9347 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 9348 #ifdef __BIG_ENDIAN_BITFIELD 9349 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 9350 #endif 9351 break; 9352 9353 case offsetof(struct __sk_buff, queue_mapping): 9354 if (type == BPF_WRITE) { 9355 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 9356 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9357 bpf_target_off(struct sk_buff, 9358 queue_mapping, 9359 2, target_size)); 9360 } else { 9361 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9362 bpf_target_off(struct sk_buff, 9363 queue_mapping, 9364 2, target_size)); 9365 } 9366 break; 9367 9368 case offsetof(struct __sk_buff, vlan_present): 9369 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9370 bpf_target_off(struct sk_buff, 9371 vlan_all, 4, target_size)); 9372 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9373 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1); 9374 break; 9375 9376 case offsetof(struct __sk_buff, vlan_tci): 9377 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9378 bpf_target_off(struct sk_buff, vlan_tci, 2, 9379 target_size)); 9380 break; 9381 9382 case offsetof(struct __sk_buff, cb[0]) ... 9383 offsetofend(struct __sk_buff, cb[4]) - 1: 9384 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 9385 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 9386 offsetof(struct qdisc_skb_cb, data)) % 9387 sizeof(__u64)); 9388 9389 prog->cb_access = 1; 9390 off = si->off; 9391 off -= offsetof(struct __sk_buff, cb[0]); 9392 off += offsetof(struct sk_buff, cb); 9393 off += offsetof(struct qdisc_skb_cb, data); 9394 if (type == BPF_WRITE) 9395 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 9396 si->src_reg, off); 9397 else 9398 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 9399 si->src_reg, off); 9400 break; 9401 9402 case offsetof(struct __sk_buff, tc_classid): 9403 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 9404 9405 off = si->off; 9406 off -= offsetof(struct __sk_buff, tc_classid); 9407 off += offsetof(struct sk_buff, cb); 9408 off += offsetof(struct qdisc_skb_cb, tc_classid); 9409 *target_size = 2; 9410 if (type == BPF_WRITE) 9411 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 9412 si->src_reg, off); 9413 else 9414 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 9415 si->src_reg, off); 9416 break; 9417 9418 case offsetof(struct __sk_buff, data): 9419 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9420 si->dst_reg, si->src_reg, 9421 offsetof(struct sk_buff, data)); 9422 break; 9423 9424 case offsetof(struct __sk_buff, data_meta): 9425 off = si->off; 9426 off -= offsetof(struct __sk_buff, data_meta); 9427 off += offsetof(struct sk_buff, cb); 9428 off += offsetof(struct bpf_skb_data_end, data_meta); 9429 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9430 si->src_reg, off); 9431 break; 9432 9433 case offsetof(struct __sk_buff, data_end): 9434 off = si->off; 9435 off -= offsetof(struct __sk_buff, data_end); 9436 off += offsetof(struct sk_buff, cb); 9437 off += offsetof(struct bpf_skb_data_end, data_end); 9438 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9439 si->src_reg, off); 9440 break; 9441 9442 case offsetof(struct __sk_buff, tc_index): 9443 #ifdef CONFIG_NET_SCHED 9444 if (type == BPF_WRITE) 9445 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9446 bpf_target_off(struct sk_buff, tc_index, 2, 9447 target_size)); 9448 else 9449 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9450 bpf_target_off(struct sk_buff, tc_index, 2, 9451 target_size)); 9452 #else 9453 *target_size = 2; 9454 if (type == BPF_WRITE) 9455 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 9456 else 9457 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9458 #endif 9459 break; 9460 9461 case offsetof(struct __sk_buff, napi_id): 9462 #if defined(CONFIG_NET_RX_BUSY_POLL) 9463 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9464 bpf_target_off(struct sk_buff, napi_id, 4, 9465 target_size)); 9466 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 9467 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9468 #else 9469 *target_size = 4; 9470 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9471 #endif 9472 break; 9473 case offsetof(struct __sk_buff, family): 9474 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9475 9476 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9477 si->dst_reg, si->src_reg, 9478 offsetof(struct sk_buff, sk)); 9479 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9480 bpf_target_off(struct sock_common, 9481 skc_family, 9482 2, target_size)); 9483 break; 9484 case offsetof(struct __sk_buff, remote_ip4): 9485 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9486 9487 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9488 si->dst_reg, si->src_reg, 9489 offsetof(struct sk_buff, sk)); 9490 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9491 bpf_target_off(struct sock_common, 9492 skc_daddr, 9493 4, target_size)); 9494 break; 9495 case offsetof(struct __sk_buff, local_ip4): 9496 BUILD_BUG_ON(sizeof_field(struct sock_common, 9497 skc_rcv_saddr) != 4); 9498 9499 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9500 si->dst_reg, si->src_reg, 9501 offsetof(struct sk_buff, sk)); 9502 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9503 bpf_target_off(struct sock_common, 9504 skc_rcv_saddr, 9505 4, target_size)); 9506 break; 9507 case offsetof(struct __sk_buff, remote_ip6[0]) ... 9508 offsetof(struct __sk_buff, remote_ip6[3]): 9509 #if IS_ENABLED(CONFIG_IPV6) 9510 BUILD_BUG_ON(sizeof_field(struct sock_common, 9511 skc_v6_daddr.s6_addr32[0]) != 4); 9512 9513 off = si->off; 9514 off -= offsetof(struct __sk_buff, remote_ip6[0]); 9515 9516 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9517 si->dst_reg, si->src_reg, 9518 offsetof(struct sk_buff, sk)); 9519 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9520 offsetof(struct sock_common, 9521 skc_v6_daddr.s6_addr32[0]) + 9522 off); 9523 #else 9524 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9525 #endif 9526 break; 9527 case offsetof(struct __sk_buff, local_ip6[0]) ... 9528 offsetof(struct __sk_buff, local_ip6[3]): 9529 #if IS_ENABLED(CONFIG_IPV6) 9530 BUILD_BUG_ON(sizeof_field(struct sock_common, 9531 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9532 9533 off = si->off; 9534 off -= offsetof(struct __sk_buff, local_ip6[0]); 9535 9536 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9537 si->dst_reg, si->src_reg, 9538 offsetof(struct sk_buff, sk)); 9539 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9540 offsetof(struct sock_common, 9541 skc_v6_rcv_saddr.s6_addr32[0]) + 9542 off); 9543 #else 9544 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9545 #endif 9546 break; 9547 9548 case offsetof(struct __sk_buff, remote_port): 9549 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9550 9551 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9552 si->dst_reg, si->src_reg, 9553 offsetof(struct sk_buff, sk)); 9554 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9555 bpf_target_off(struct sock_common, 9556 skc_dport, 9557 2, target_size)); 9558 #ifndef __BIG_ENDIAN_BITFIELD 9559 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9560 #endif 9561 break; 9562 9563 case offsetof(struct __sk_buff, local_port): 9564 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9565 9566 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9567 si->dst_reg, si->src_reg, 9568 offsetof(struct sk_buff, sk)); 9569 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9570 bpf_target_off(struct sock_common, 9571 skc_num, 2, target_size)); 9572 break; 9573 9574 case offsetof(struct __sk_buff, tstamp): 9575 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 9576 9577 if (type == BPF_WRITE) 9578 insn = bpf_convert_tstamp_write(prog, si, insn); 9579 else 9580 insn = bpf_convert_tstamp_read(prog, si, insn); 9581 break; 9582 9583 case offsetof(struct __sk_buff, tstamp_type): 9584 insn = bpf_convert_tstamp_type_read(si, insn); 9585 break; 9586 9587 case offsetof(struct __sk_buff, gso_segs): 9588 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 9590 si->dst_reg, si->dst_reg, 9591 bpf_target_off(struct skb_shared_info, 9592 gso_segs, 2, 9593 target_size)); 9594 break; 9595 case offsetof(struct __sk_buff, gso_size): 9596 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9597 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 9598 si->dst_reg, si->dst_reg, 9599 bpf_target_off(struct skb_shared_info, 9600 gso_size, 2, 9601 target_size)); 9602 break; 9603 case offsetof(struct __sk_buff, wire_len): 9604 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 9605 9606 off = si->off; 9607 off -= offsetof(struct __sk_buff, wire_len); 9608 off += offsetof(struct sk_buff, cb); 9609 off += offsetof(struct qdisc_skb_cb, pkt_len); 9610 *target_size = 4; 9611 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 9612 break; 9613 9614 case offsetof(struct __sk_buff, sk): 9615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9616 si->dst_reg, si->src_reg, 9617 offsetof(struct sk_buff, sk)); 9618 break; 9619 case offsetof(struct __sk_buff, hwtstamp): 9620 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8); 9621 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0); 9622 9623 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9624 *insn++ = BPF_LDX_MEM(BPF_DW, 9625 si->dst_reg, si->dst_reg, 9626 bpf_target_off(struct skb_shared_info, 9627 hwtstamps, 8, 9628 target_size)); 9629 break; 9630 } 9631 9632 return insn - insn_buf; 9633 } 9634 9635 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 9636 const struct bpf_insn *si, 9637 struct bpf_insn *insn_buf, 9638 struct bpf_prog *prog, u32 *target_size) 9639 { 9640 struct bpf_insn *insn = insn_buf; 9641 int off; 9642 9643 switch (si->off) { 9644 case offsetof(struct bpf_sock, bound_dev_if): 9645 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 9646 9647 if (type == BPF_WRITE) 9648 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9649 offsetof(struct sock, sk_bound_dev_if)); 9650 else 9651 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9652 offsetof(struct sock, sk_bound_dev_if)); 9653 break; 9654 9655 case offsetof(struct bpf_sock, mark): 9656 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 9657 9658 if (type == BPF_WRITE) 9659 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9660 offsetof(struct sock, sk_mark)); 9661 else 9662 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9663 offsetof(struct sock, sk_mark)); 9664 break; 9665 9666 case offsetof(struct bpf_sock, priority): 9667 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 9668 9669 if (type == BPF_WRITE) 9670 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9671 offsetof(struct sock, sk_priority)); 9672 else 9673 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9674 offsetof(struct sock, sk_priority)); 9675 break; 9676 9677 case offsetof(struct bpf_sock, family): 9678 *insn++ = BPF_LDX_MEM( 9679 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 9680 si->dst_reg, si->src_reg, 9681 bpf_target_off(struct sock_common, 9682 skc_family, 9683 sizeof_field(struct sock_common, 9684 skc_family), 9685 target_size)); 9686 break; 9687 9688 case offsetof(struct bpf_sock, type): 9689 *insn++ = BPF_LDX_MEM( 9690 BPF_FIELD_SIZEOF(struct sock, sk_type), 9691 si->dst_reg, si->src_reg, 9692 bpf_target_off(struct sock, sk_type, 9693 sizeof_field(struct sock, sk_type), 9694 target_size)); 9695 break; 9696 9697 case offsetof(struct bpf_sock, protocol): 9698 *insn++ = BPF_LDX_MEM( 9699 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 9700 si->dst_reg, si->src_reg, 9701 bpf_target_off(struct sock, sk_protocol, 9702 sizeof_field(struct sock, sk_protocol), 9703 target_size)); 9704 break; 9705 9706 case offsetof(struct bpf_sock, src_ip4): 9707 *insn++ = BPF_LDX_MEM( 9708 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9709 bpf_target_off(struct sock_common, skc_rcv_saddr, 9710 sizeof_field(struct sock_common, 9711 skc_rcv_saddr), 9712 target_size)); 9713 break; 9714 9715 case offsetof(struct bpf_sock, dst_ip4): 9716 *insn++ = BPF_LDX_MEM( 9717 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9718 bpf_target_off(struct sock_common, skc_daddr, 9719 sizeof_field(struct sock_common, 9720 skc_daddr), 9721 target_size)); 9722 break; 9723 9724 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9725 #if IS_ENABLED(CONFIG_IPV6) 9726 off = si->off; 9727 off -= offsetof(struct bpf_sock, src_ip6[0]); 9728 *insn++ = BPF_LDX_MEM( 9729 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9730 bpf_target_off( 9731 struct sock_common, 9732 skc_v6_rcv_saddr.s6_addr32[0], 9733 sizeof_field(struct sock_common, 9734 skc_v6_rcv_saddr.s6_addr32[0]), 9735 target_size) + off); 9736 #else 9737 (void)off; 9738 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9739 #endif 9740 break; 9741 9742 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 9743 #if IS_ENABLED(CONFIG_IPV6) 9744 off = si->off; 9745 off -= offsetof(struct bpf_sock, dst_ip6[0]); 9746 *insn++ = BPF_LDX_MEM( 9747 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9748 bpf_target_off(struct sock_common, 9749 skc_v6_daddr.s6_addr32[0], 9750 sizeof_field(struct sock_common, 9751 skc_v6_daddr.s6_addr32[0]), 9752 target_size) + off); 9753 #else 9754 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9755 *target_size = 4; 9756 #endif 9757 break; 9758 9759 case offsetof(struct bpf_sock, src_port): 9760 *insn++ = BPF_LDX_MEM( 9761 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 9762 si->dst_reg, si->src_reg, 9763 bpf_target_off(struct sock_common, skc_num, 9764 sizeof_field(struct sock_common, 9765 skc_num), 9766 target_size)); 9767 break; 9768 9769 case offsetof(struct bpf_sock, dst_port): 9770 *insn++ = BPF_LDX_MEM( 9771 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 9772 si->dst_reg, si->src_reg, 9773 bpf_target_off(struct sock_common, skc_dport, 9774 sizeof_field(struct sock_common, 9775 skc_dport), 9776 target_size)); 9777 break; 9778 9779 case offsetof(struct bpf_sock, state): 9780 *insn++ = BPF_LDX_MEM( 9781 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 9782 si->dst_reg, si->src_reg, 9783 bpf_target_off(struct sock_common, skc_state, 9784 sizeof_field(struct sock_common, 9785 skc_state), 9786 target_size)); 9787 break; 9788 case offsetof(struct bpf_sock, rx_queue_mapping): 9789 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 9790 *insn++ = BPF_LDX_MEM( 9791 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 9792 si->dst_reg, si->src_reg, 9793 bpf_target_off(struct sock, sk_rx_queue_mapping, 9794 sizeof_field(struct sock, 9795 sk_rx_queue_mapping), 9796 target_size)); 9797 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 9798 1); 9799 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9800 #else 9801 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9802 *target_size = 2; 9803 #endif 9804 break; 9805 } 9806 9807 return insn - insn_buf; 9808 } 9809 9810 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 9811 const struct bpf_insn *si, 9812 struct bpf_insn *insn_buf, 9813 struct bpf_prog *prog, u32 *target_size) 9814 { 9815 struct bpf_insn *insn = insn_buf; 9816 9817 switch (si->off) { 9818 case offsetof(struct __sk_buff, ifindex): 9819 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9820 si->dst_reg, si->src_reg, 9821 offsetof(struct sk_buff, dev)); 9822 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9823 bpf_target_off(struct net_device, ifindex, 4, 9824 target_size)); 9825 break; 9826 default: 9827 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9828 target_size); 9829 } 9830 9831 return insn - insn_buf; 9832 } 9833 9834 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 9835 const struct bpf_insn *si, 9836 struct bpf_insn *insn_buf, 9837 struct bpf_prog *prog, u32 *target_size) 9838 { 9839 struct bpf_insn *insn = insn_buf; 9840 9841 switch (si->off) { 9842 case offsetof(struct xdp_md, data): 9843 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 9844 si->dst_reg, si->src_reg, 9845 offsetof(struct xdp_buff, data)); 9846 break; 9847 case offsetof(struct xdp_md, data_meta): 9848 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 9849 si->dst_reg, si->src_reg, 9850 offsetof(struct xdp_buff, data_meta)); 9851 break; 9852 case offsetof(struct xdp_md, data_end): 9853 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 9854 si->dst_reg, si->src_reg, 9855 offsetof(struct xdp_buff, data_end)); 9856 break; 9857 case offsetof(struct xdp_md, ingress_ifindex): 9858 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9859 si->dst_reg, si->src_reg, 9860 offsetof(struct xdp_buff, rxq)); 9861 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 9862 si->dst_reg, si->dst_reg, 9863 offsetof(struct xdp_rxq_info, dev)); 9864 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9865 offsetof(struct net_device, ifindex)); 9866 break; 9867 case offsetof(struct xdp_md, rx_queue_index): 9868 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9869 si->dst_reg, si->src_reg, 9870 offsetof(struct xdp_buff, rxq)); 9871 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9872 offsetof(struct xdp_rxq_info, 9873 queue_index)); 9874 break; 9875 case offsetof(struct xdp_md, egress_ifindex): 9876 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 9877 si->dst_reg, si->src_reg, 9878 offsetof(struct xdp_buff, txq)); 9879 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 9880 si->dst_reg, si->dst_reg, 9881 offsetof(struct xdp_txq_info, dev)); 9882 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9883 offsetof(struct net_device, ifindex)); 9884 break; 9885 } 9886 9887 return insn - insn_buf; 9888 } 9889 9890 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 9891 * context Structure, F is Field in context structure that contains a pointer 9892 * to Nested Structure of type NS that has the field NF. 9893 * 9894 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 9895 * sure that SIZE is not greater than actual size of S.F.NF. 9896 * 9897 * If offset OFF is provided, the load happens from that offset relative to 9898 * offset of NF. 9899 */ 9900 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 9901 do { \ 9902 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 9903 si->src_reg, offsetof(S, F)); \ 9904 *insn++ = BPF_LDX_MEM( \ 9905 SIZE, si->dst_reg, si->dst_reg, \ 9906 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9907 target_size) \ 9908 + OFF); \ 9909 } while (0) 9910 9911 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 9912 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 9913 BPF_FIELD_SIZEOF(NS, NF), 0) 9914 9915 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 9916 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 9917 * 9918 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 9919 * "register" since two registers available in convert_ctx_access are not 9920 * enough: we can't override neither SRC, since it contains value to store, nor 9921 * DST since it contains pointer to context that may be used by later 9922 * instructions. But we need a temporary place to save pointer to nested 9923 * structure whose field we want to store to. 9924 */ 9925 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 9926 do { \ 9927 int tmp_reg = BPF_REG_9; \ 9928 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9929 --tmp_reg; \ 9930 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9931 --tmp_reg; \ 9932 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 9933 offsetof(S, TF)); \ 9934 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 9935 si->dst_reg, offsetof(S, F)); \ 9936 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 9937 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9938 target_size) \ 9939 + OFF); \ 9940 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 9941 offsetof(S, TF)); \ 9942 } while (0) 9943 9944 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 9945 TF) \ 9946 do { \ 9947 if (type == BPF_WRITE) { \ 9948 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 9949 OFF, TF); \ 9950 } else { \ 9951 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 9952 S, NS, F, NF, SIZE, OFF); \ 9953 } \ 9954 } while (0) 9955 9956 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 9957 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 9958 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 9959 9960 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 9961 const struct bpf_insn *si, 9962 struct bpf_insn *insn_buf, 9963 struct bpf_prog *prog, u32 *target_size) 9964 { 9965 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 9966 struct bpf_insn *insn = insn_buf; 9967 9968 switch (si->off) { 9969 case offsetof(struct bpf_sock_addr, user_family): 9970 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9971 struct sockaddr, uaddr, sa_family); 9972 break; 9973 9974 case offsetof(struct bpf_sock_addr, user_ip4): 9975 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9976 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 9977 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 9978 break; 9979 9980 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 9981 off = si->off; 9982 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 9983 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9984 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9985 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 9986 tmp_reg); 9987 break; 9988 9989 case offsetof(struct bpf_sock_addr, user_port): 9990 /* To get port we need to know sa_family first and then treat 9991 * sockaddr as either sockaddr_in or sockaddr_in6. 9992 * Though we can simplify since port field has same offset and 9993 * size in both structures. 9994 * Here we check this invariant and use just one of the 9995 * structures if it's true. 9996 */ 9997 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 9998 offsetof(struct sockaddr_in6, sin6_port)); 9999 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 10000 sizeof_field(struct sockaddr_in6, sin6_port)); 10001 /* Account for sin6_port being smaller than user_port. */ 10002 port_size = min(port_size, BPF_LDST_BYTES(si)); 10003 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10004 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 10005 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 10006 break; 10007 10008 case offsetof(struct bpf_sock_addr, family): 10009 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10010 struct sock, sk, sk_family); 10011 break; 10012 10013 case offsetof(struct bpf_sock_addr, type): 10014 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10015 struct sock, sk, sk_type); 10016 break; 10017 10018 case offsetof(struct bpf_sock_addr, protocol): 10019 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10020 struct sock, sk, sk_protocol); 10021 break; 10022 10023 case offsetof(struct bpf_sock_addr, msg_src_ip4): 10024 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 10025 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10026 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 10027 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 10028 break; 10029 10030 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 10031 msg_src_ip6[3]): 10032 off = si->off; 10033 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 10034 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 10035 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10036 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 10037 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 10038 break; 10039 case offsetof(struct bpf_sock_addr, sk): 10040 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 10041 si->dst_reg, si->src_reg, 10042 offsetof(struct bpf_sock_addr_kern, sk)); 10043 break; 10044 } 10045 10046 return insn - insn_buf; 10047 } 10048 10049 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 10050 const struct bpf_insn *si, 10051 struct bpf_insn *insn_buf, 10052 struct bpf_prog *prog, 10053 u32 *target_size) 10054 { 10055 struct bpf_insn *insn = insn_buf; 10056 int off; 10057 10058 /* Helper macro for adding read access to tcp_sock or sock fields. */ 10059 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10060 do { \ 10061 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 10062 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10063 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10064 if (si->dst_reg == reg || si->src_reg == reg) \ 10065 reg--; \ 10066 if (si->dst_reg == reg || si->src_reg == reg) \ 10067 reg--; \ 10068 if (si->dst_reg == si->src_reg) { \ 10069 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10070 offsetof(struct bpf_sock_ops_kern, \ 10071 temp)); \ 10072 fullsock_reg = reg; \ 10073 jmp += 2; \ 10074 } \ 10075 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10076 struct bpf_sock_ops_kern, \ 10077 is_fullsock), \ 10078 fullsock_reg, si->src_reg, \ 10079 offsetof(struct bpf_sock_ops_kern, \ 10080 is_fullsock)); \ 10081 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10082 if (si->dst_reg == si->src_reg) \ 10083 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10084 offsetof(struct bpf_sock_ops_kern, \ 10085 temp)); \ 10086 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10087 struct bpf_sock_ops_kern, sk),\ 10088 si->dst_reg, si->src_reg, \ 10089 offsetof(struct bpf_sock_ops_kern, sk));\ 10090 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 10091 OBJ_FIELD), \ 10092 si->dst_reg, si->dst_reg, \ 10093 offsetof(OBJ, OBJ_FIELD)); \ 10094 if (si->dst_reg == si->src_reg) { \ 10095 *insn++ = BPF_JMP_A(1); \ 10096 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10097 offsetof(struct bpf_sock_ops_kern, \ 10098 temp)); \ 10099 } \ 10100 } while (0) 10101 10102 #define SOCK_OPS_GET_SK() \ 10103 do { \ 10104 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 10105 if (si->dst_reg == reg || si->src_reg == reg) \ 10106 reg--; \ 10107 if (si->dst_reg == reg || si->src_reg == reg) \ 10108 reg--; \ 10109 if (si->dst_reg == si->src_reg) { \ 10110 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10111 offsetof(struct bpf_sock_ops_kern, \ 10112 temp)); \ 10113 fullsock_reg = reg; \ 10114 jmp += 2; \ 10115 } \ 10116 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10117 struct bpf_sock_ops_kern, \ 10118 is_fullsock), \ 10119 fullsock_reg, si->src_reg, \ 10120 offsetof(struct bpf_sock_ops_kern, \ 10121 is_fullsock)); \ 10122 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10123 if (si->dst_reg == si->src_reg) \ 10124 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10125 offsetof(struct bpf_sock_ops_kern, \ 10126 temp)); \ 10127 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10128 struct bpf_sock_ops_kern, sk),\ 10129 si->dst_reg, si->src_reg, \ 10130 offsetof(struct bpf_sock_ops_kern, sk));\ 10131 if (si->dst_reg == si->src_reg) { \ 10132 *insn++ = BPF_JMP_A(1); \ 10133 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10134 offsetof(struct bpf_sock_ops_kern, \ 10135 temp)); \ 10136 } \ 10137 } while (0) 10138 10139 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 10140 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 10141 10142 /* Helper macro for adding write access to tcp_sock or sock fields. 10143 * The macro is called with two registers, dst_reg which contains a pointer 10144 * to ctx (context) and src_reg which contains the value that should be 10145 * stored. However, we need an additional register since we cannot overwrite 10146 * dst_reg because it may be used later in the program. 10147 * Instead we "borrow" one of the other register. We first save its value 10148 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 10149 * it at the end of the macro. 10150 */ 10151 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10152 do { \ 10153 int reg = BPF_REG_9; \ 10154 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10155 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10156 if (si->dst_reg == reg || si->src_reg == reg) \ 10157 reg--; \ 10158 if (si->dst_reg == reg || si->src_reg == reg) \ 10159 reg--; \ 10160 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 10161 offsetof(struct bpf_sock_ops_kern, \ 10162 temp)); \ 10163 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10164 struct bpf_sock_ops_kern, \ 10165 is_fullsock), \ 10166 reg, si->dst_reg, \ 10167 offsetof(struct bpf_sock_ops_kern, \ 10168 is_fullsock)); \ 10169 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 10170 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10171 struct bpf_sock_ops_kern, sk),\ 10172 reg, si->dst_reg, \ 10173 offsetof(struct bpf_sock_ops_kern, sk));\ 10174 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 10175 reg, si->src_reg, \ 10176 offsetof(OBJ, OBJ_FIELD)); \ 10177 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 10178 offsetof(struct bpf_sock_ops_kern, \ 10179 temp)); \ 10180 } while (0) 10181 10182 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 10183 do { \ 10184 if (TYPE == BPF_WRITE) \ 10185 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10186 else \ 10187 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10188 } while (0) 10189 10190 switch (si->off) { 10191 case offsetof(struct bpf_sock_ops, op): 10192 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10193 op), 10194 si->dst_reg, si->src_reg, 10195 offsetof(struct bpf_sock_ops_kern, op)); 10196 break; 10197 10198 case offsetof(struct bpf_sock_ops, replylong[0]) ... 10199 offsetof(struct bpf_sock_ops, replylong[3]): 10200 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 10201 sizeof_field(struct bpf_sock_ops_kern, reply)); 10202 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 10203 sizeof_field(struct bpf_sock_ops_kern, replylong)); 10204 off = si->off; 10205 off -= offsetof(struct bpf_sock_ops, replylong[0]); 10206 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 10207 if (type == BPF_WRITE) 10208 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 10209 off); 10210 else 10211 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10212 off); 10213 break; 10214 10215 case offsetof(struct bpf_sock_ops, family): 10216 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10217 10218 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10219 struct bpf_sock_ops_kern, sk), 10220 si->dst_reg, si->src_reg, 10221 offsetof(struct bpf_sock_ops_kern, sk)); 10222 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10223 offsetof(struct sock_common, skc_family)); 10224 break; 10225 10226 case offsetof(struct bpf_sock_ops, remote_ip4): 10227 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10228 10229 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10230 struct bpf_sock_ops_kern, sk), 10231 si->dst_reg, si->src_reg, 10232 offsetof(struct bpf_sock_ops_kern, sk)); 10233 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10234 offsetof(struct sock_common, skc_daddr)); 10235 break; 10236 10237 case offsetof(struct bpf_sock_ops, local_ip4): 10238 BUILD_BUG_ON(sizeof_field(struct sock_common, 10239 skc_rcv_saddr) != 4); 10240 10241 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10242 struct bpf_sock_ops_kern, sk), 10243 si->dst_reg, si->src_reg, 10244 offsetof(struct bpf_sock_ops_kern, sk)); 10245 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10246 offsetof(struct sock_common, 10247 skc_rcv_saddr)); 10248 break; 10249 10250 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 10251 offsetof(struct bpf_sock_ops, remote_ip6[3]): 10252 #if IS_ENABLED(CONFIG_IPV6) 10253 BUILD_BUG_ON(sizeof_field(struct sock_common, 10254 skc_v6_daddr.s6_addr32[0]) != 4); 10255 10256 off = si->off; 10257 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 10258 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10259 struct bpf_sock_ops_kern, sk), 10260 si->dst_reg, si->src_reg, 10261 offsetof(struct bpf_sock_ops_kern, sk)); 10262 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10263 offsetof(struct sock_common, 10264 skc_v6_daddr.s6_addr32[0]) + 10265 off); 10266 #else 10267 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10268 #endif 10269 break; 10270 10271 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 10272 offsetof(struct bpf_sock_ops, local_ip6[3]): 10273 #if IS_ENABLED(CONFIG_IPV6) 10274 BUILD_BUG_ON(sizeof_field(struct sock_common, 10275 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10276 10277 off = si->off; 10278 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 10279 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10280 struct bpf_sock_ops_kern, sk), 10281 si->dst_reg, si->src_reg, 10282 offsetof(struct bpf_sock_ops_kern, sk)); 10283 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10284 offsetof(struct sock_common, 10285 skc_v6_rcv_saddr.s6_addr32[0]) + 10286 off); 10287 #else 10288 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10289 #endif 10290 break; 10291 10292 case offsetof(struct bpf_sock_ops, remote_port): 10293 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10294 10295 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10296 struct bpf_sock_ops_kern, sk), 10297 si->dst_reg, si->src_reg, 10298 offsetof(struct bpf_sock_ops_kern, sk)); 10299 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10300 offsetof(struct sock_common, skc_dport)); 10301 #ifndef __BIG_ENDIAN_BITFIELD 10302 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10303 #endif 10304 break; 10305 10306 case offsetof(struct bpf_sock_ops, local_port): 10307 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10308 10309 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10310 struct bpf_sock_ops_kern, sk), 10311 si->dst_reg, si->src_reg, 10312 offsetof(struct bpf_sock_ops_kern, sk)); 10313 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10314 offsetof(struct sock_common, skc_num)); 10315 break; 10316 10317 case offsetof(struct bpf_sock_ops, is_fullsock): 10318 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10319 struct bpf_sock_ops_kern, 10320 is_fullsock), 10321 si->dst_reg, si->src_reg, 10322 offsetof(struct bpf_sock_ops_kern, 10323 is_fullsock)); 10324 break; 10325 10326 case offsetof(struct bpf_sock_ops, state): 10327 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 10328 10329 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10330 struct bpf_sock_ops_kern, sk), 10331 si->dst_reg, si->src_reg, 10332 offsetof(struct bpf_sock_ops_kern, sk)); 10333 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 10334 offsetof(struct sock_common, skc_state)); 10335 break; 10336 10337 case offsetof(struct bpf_sock_ops, rtt_min): 10338 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 10339 sizeof(struct minmax)); 10340 BUILD_BUG_ON(sizeof(struct minmax) < 10341 sizeof(struct minmax_sample)); 10342 10343 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10344 struct bpf_sock_ops_kern, sk), 10345 si->dst_reg, si->src_reg, 10346 offsetof(struct bpf_sock_ops_kern, sk)); 10347 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10348 offsetof(struct tcp_sock, rtt_min) + 10349 sizeof_field(struct minmax_sample, t)); 10350 break; 10351 10352 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 10353 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 10354 struct tcp_sock); 10355 break; 10356 10357 case offsetof(struct bpf_sock_ops, sk_txhash): 10358 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 10359 struct sock, type); 10360 break; 10361 case offsetof(struct bpf_sock_ops, snd_cwnd): 10362 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 10363 break; 10364 case offsetof(struct bpf_sock_ops, srtt_us): 10365 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 10366 break; 10367 case offsetof(struct bpf_sock_ops, snd_ssthresh): 10368 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 10369 break; 10370 case offsetof(struct bpf_sock_ops, rcv_nxt): 10371 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 10372 break; 10373 case offsetof(struct bpf_sock_ops, snd_nxt): 10374 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 10375 break; 10376 case offsetof(struct bpf_sock_ops, snd_una): 10377 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 10378 break; 10379 case offsetof(struct bpf_sock_ops, mss_cache): 10380 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 10381 break; 10382 case offsetof(struct bpf_sock_ops, ecn_flags): 10383 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 10384 break; 10385 case offsetof(struct bpf_sock_ops, rate_delivered): 10386 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 10387 break; 10388 case offsetof(struct bpf_sock_ops, rate_interval_us): 10389 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 10390 break; 10391 case offsetof(struct bpf_sock_ops, packets_out): 10392 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 10393 break; 10394 case offsetof(struct bpf_sock_ops, retrans_out): 10395 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 10396 break; 10397 case offsetof(struct bpf_sock_ops, total_retrans): 10398 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 10399 break; 10400 case offsetof(struct bpf_sock_ops, segs_in): 10401 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 10402 break; 10403 case offsetof(struct bpf_sock_ops, data_segs_in): 10404 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 10405 break; 10406 case offsetof(struct bpf_sock_ops, segs_out): 10407 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 10408 break; 10409 case offsetof(struct bpf_sock_ops, data_segs_out): 10410 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 10411 break; 10412 case offsetof(struct bpf_sock_ops, lost_out): 10413 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 10414 break; 10415 case offsetof(struct bpf_sock_ops, sacked_out): 10416 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 10417 break; 10418 case offsetof(struct bpf_sock_ops, bytes_received): 10419 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 10420 break; 10421 case offsetof(struct bpf_sock_ops, bytes_acked): 10422 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 10423 break; 10424 case offsetof(struct bpf_sock_ops, sk): 10425 SOCK_OPS_GET_SK(); 10426 break; 10427 case offsetof(struct bpf_sock_ops, skb_data_end): 10428 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10429 skb_data_end), 10430 si->dst_reg, si->src_reg, 10431 offsetof(struct bpf_sock_ops_kern, 10432 skb_data_end)); 10433 break; 10434 case offsetof(struct bpf_sock_ops, skb_data): 10435 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10436 skb), 10437 si->dst_reg, si->src_reg, 10438 offsetof(struct bpf_sock_ops_kern, 10439 skb)); 10440 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10442 si->dst_reg, si->dst_reg, 10443 offsetof(struct sk_buff, data)); 10444 break; 10445 case offsetof(struct bpf_sock_ops, skb_len): 10446 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10447 skb), 10448 si->dst_reg, si->src_reg, 10449 offsetof(struct bpf_sock_ops_kern, 10450 skb)); 10451 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10452 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10453 si->dst_reg, si->dst_reg, 10454 offsetof(struct sk_buff, len)); 10455 break; 10456 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 10457 off = offsetof(struct sk_buff, cb); 10458 off += offsetof(struct tcp_skb_cb, tcp_flags); 10459 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 10460 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10461 skb), 10462 si->dst_reg, si->src_reg, 10463 offsetof(struct bpf_sock_ops_kern, 10464 skb)); 10465 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10466 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 10467 tcp_flags), 10468 si->dst_reg, si->dst_reg, off); 10469 break; 10470 case offsetof(struct bpf_sock_ops, skb_hwtstamp): { 10471 struct bpf_insn *jmp_on_null_skb; 10472 10473 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10474 skb), 10475 si->dst_reg, si->src_reg, 10476 offsetof(struct bpf_sock_ops_kern, 10477 skb)); 10478 /* Reserve one insn to test skb == NULL */ 10479 jmp_on_null_skb = insn++; 10480 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn); 10481 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 10482 bpf_target_off(struct skb_shared_info, 10483 hwtstamps, 8, 10484 target_size)); 10485 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 10486 insn - jmp_on_null_skb - 1); 10487 break; 10488 } 10489 } 10490 return insn - insn_buf; 10491 } 10492 10493 /* data_end = skb->data + skb_headlen() */ 10494 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si, 10495 struct bpf_insn *insn) 10496 { 10497 int reg; 10498 int temp_reg_off = offsetof(struct sk_buff, cb) + 10499 offsetof(struct sk_skb_cb, temp_reg); 10500 10501 if (si->src_reg == si->dst_reg) { 10502 /* We need an extra register, choose and save a register. */ 10503 reg = BPF_REG_9; 10504 if (si->src_reg == reg || si->dst_reg == reg) 10505 reg--; 10506 if (si->src_reg == reg || si->dst_reg == reg) 10507 reg--; 10508 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off); 10509 } else { 10510 reg = si->dst_reg; 10511 } 10512 10513 /* reg = skb->data */ 10514 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10515 reg, si->src_reg, 10516 offsetof(struct sk_buff, data)); 10517 /* AX = skb->len */ 10518 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10519 BPF_REG_AX, si->src_reg, 10520 offsetof(struct sk_buff, len)); 10521 /* reg = skb->data + skb->len */ 10522 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX); 10523 /* AX = skb->data_len */ 10524 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len), 10525 BPF_REG_AX, si->src_reg, 10526 offsetof(struct sk_buff, data_len)); 10527 10528 /* reg = skb->data + skb->len - skb->data_len */ 10529 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX); 10530 10531 if (si->src_reg == si->dst_reg) { 10532 /* Restore the saved register */ 10533 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg); 10534 *insn++ = BPF_MOV64_REG(si->dst_reg, reg); 10535 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off); 10536 } 10537 10538 return insn; 10539 } 10540 10541 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 10542 const struct bpf_insn *si, 10543 struct bpf_insn *insn_buf, 10544 struct bpf_prog *prog, u32 *target_size) 10545 { 10546 struct bpf_insn *insn = insn_buf; 10547 int off; 10548 10549 switch (si->off) { 10550 case offsetof(struct __sk_buff, data_end): 10551 insn = bpf_convert_data_end_access(si, insn); 10552 break; 10553 case offsetof(struct __sk_buff, cb[0]) ... 10554 offsetofend(struct __sk_buff, cb[4]) - 1: 10555 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20); 10556 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 10557 offsetof(struct sk_skb_cb, data)) % 10558 sizeof(__u64)); 10559 10560 prog->cb_access = 1; 10561 off = si->off; 10562 off -= offsetof(struct __sk_buff, cb[0]); 10563 off += offsetof(struct sk_buff, cb); 10564 off += offsetof(struct sk_skb_cb, data); 10565 if (type == BPF_WRITE) 10566 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 10567 si->src_reg, off); 10568 else 10569 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 10570 si->src_reg, off); 10571 break; 10572 10573 10574 default: 10575 return bpf_convert_ctx_access(type, si, insn_buf, prog, 10576 target_size); 10577 } 10578 10579 return insn - insn_buf; 10580 } 10581 10582 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 10583 const struct bpf_insn *si, 10584 struct bpf_insn *insn_buf, 10585 struct bpf_prog *prog, u32 *target_size) 10586 { 10587 struct bpf_insn *insn = insn_buf; 10588 #if IS_ENABLED(CONFIG_IPV6) 10589 int off; 10590 #endif 10591 10592 /* convert ctx uses the fact sg element is first in struct */ 10593 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 10594 10595 switch (si->off) { 10596 case offsetof(struct sk_msg_md, data): 10597 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 10598 si->dst_reg, si->src_reg, 10599 offsetof(struct sk_msg, data)); 10600 break; 10601 case offsetof(struct sk_msg_md, data_end): 10602 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 10603 si->dst_reg, si->src_reg, 10604 offsetof(struct sk_msg, data_end)); 10605 break; 10606 case offsetof(struct sk_msg_md, family): 10607 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10608 10609 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10610 struct sk_msg, sk), 10611 si->dst_reg, si->src_reg, 10612 offsetof(struct sk_msg, sk)); 10613 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10614 offsetof(struct sock_common, skc_family)); 10615 break; 10616 10617 case offsetof(struct sk_msg_md, remote_ip4): 10618 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10619 10620 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10621 struct sk_msg, sk), 10622 si->dst_reg, si->src_reg, 10623 offsetof(struct sk_msg, sk)); 10624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10625 offsetof(struct sock_common, skc_daddr)); 10626 break; 10627 10628 case offsetof(struct sk_msg_md, local_ip4): 10629 BUILD_BUG_ON(sizeof_field(struct sock_common, 10630 skc_rcv_saddr) != 4); 10631 10632 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10633 struct sk_msg, sk), 10634 si->dst_reg, si->src_reg, 10635 offsetof(struct sk_msg, sk)); 10636 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10637 offsetof(struct sock_common, 10638 skc_rcv_saddr)); 10639 break; 10640 10641 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 10642 offsetof(struct sk_msg_md, remote_ip6[3]): 10643 #if IS_ENABLED(CONFIG_IPV6) 10644 BUILD_BUG_ON(sizeof_field(struct sock_common, 10645 skc_v6_daddr.s6_addr32[0]) != 4); 10646 10647 off = si->off; 10648 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 10649 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10650 struct sk_msg, sk), 10651 si->dst_reg, si->src_reg, 10652 offsetof(struct sk_msg, sk)); 10653 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10654 offsetof(struct sock_common, 10655 skc_v6_daddr.s6_addr32[0]) + 10656 off); 10657 #else 10658 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10659 #endif 10660 break; 10661 10662 case offsetof(struct sk_msg_md, local_ip6[0]) ... 10663 offsetof(struct sk_msg_md, local_ip6[3]): 10664 #if IS_ENABLED(CONFIG_IPV6) 10665 BUILD_BUG_ON(sizeof_field(struct sock_common, 10666 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10667 10668 off = si->off; 10669 off -= offsetof(struct sk_msg_md, local_ip6[0]); 10670 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10671 struct sk_msg, sk), 10672 si->dst_reg, si->src_reg, 10673 offsetof(struct sk_msg, sk)); 10674 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10675 offsetof(struct sock_common, 10676 skc_v6_rcv_saddr.s6_addr32[0]) + 10677 off); 10678 #else 10679 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10680 #endif 10681 break; 10682 10683 case offsetof(struct sk_msg_md, remote_port): 10684 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10685 10686 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10687 struct sk_msg, sk), 10688 si->dst_reg, si->src_reg, 10689 offsetof(struct sk_msg, sk)); 10690 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10691 offsetof(struct sock_common, skc_dport)); 10692 #ifndef __BIG_ENDIAN_BITFIELD 10693 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10694 #endif 10695 break; 10696 10697 case offsetof(struct sk_msg_md, local_port): 10698 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10699 10700 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10701 struct sk_msg, sk), 10702 si->dst_reg, si->src_reg, 10703 offsetof(struct sk_msg, sk)); 10704 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10705 offsetof(struct sock_common, skc_num)); 10706 break; 10707 10708 case offsetof(struct sk_msg_md, size): 10709 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 10710 si->dst_reg, si->src_reg, 10711 offsetof(struct sk_msg_sg, size)); 10712 break; 10713 10714 case offsetof(struct sk_msg_md, sk): 10715 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 10716 si->dst_reg, si->src_reg, 10717 offsetof(struct sk_msg, sk)); 10718 break; 10719 } 10720 10721 return insn - insn_buf; 10722 } 10723 10724 const struct bpf_verifier_ops sk_filter_verifier_ops = { 10725 .get_func_proto = sk_filter_func_proto, 10726 .is_valid_access = sk_filter_is_valid_access, 10727 .convert_ctx_access = bpf_convert_ctx_access, 10728 .gen_ld_abs = bpf_gen_ld_abs, 10729 }; 10730 10731 const struct bpf_prog_ops sk_filter_prog_ops = { 10732 .test_run = bpf_prog_test_run_skb, 10733 }; 10734 10735 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 10736 .get_func_proto = tc_cls_act_func_proto, 10737 .is_valid_access = tc_cls_act_is_valid_access, 10738 .convert_ctx_access = tc_cls_act_convert_ctx_access, 10739 .gen_prologue = tc_cls_act_prologue, 10740 .gen_ld_abs = bpf_gen_ld_abs, 10741 .btf_struct_access = tc_cls_act_btf_struct_access, 10742 }; 10743 10744 const struct bpf_prog_ops tc_cls_act_prog_ops = { 10745 .test_run = bpf_prog_test_run_skb, 10746 }; 10747 10748 const struct bpf_verifier_ops xdp_verifier_ops = { 10749 .get_func_proto = xdp_func_proto, 10750 .is_valid_access = xdp_is_valid_access, 10751 .convert_ctx_access = xdp_convert_ctx_access, 10752 .gen_prologue = bpf_noop_prologue, 10753 .btf_struct_access = xdp_btf_struct_access, 10754 }; 10755 10756 const struct bpf_prog_ops xdp_prog_ops = { 10757 .test_run = bpf_prog_test_run_xdp, 10758 }; 10759 10760 const struct bpf_verifier_ops cg_skb_verifier_ops = { 10761 .get_func_proto = cg_skb_func_proto, 10762 .is_valid_access = cg_skb_is_valid_access, 10763 .convert_ctx_access = bpf_convert_ctx_access, 10764 }; 10765 10766 const struct bpf_prog_ops cg_skb_prog_ops = { 10767 .test_run = bpf_prog_test_run_skb, 10768 }; 10769 10770 const struct bpf_verifier_ops lwt_in_verifier_ops = { 10771 .get_func_proto = lwt_in_func_proto, 10772 .is_valid_access = lwt_is_valid_access, 10773 .convert_ctx_access = bpf_convert_ctx_access, 10774 }; 10775 10776 const struct bpf_prog_ops lwt_in_prog_ops = { 10777 .test_run = bpf_prog_test_run_skb, 10778 }; 10779 10780 const struct bpf_verifier_ops lwt_out_verifier_ops = { 10781 .get_func_proto = lwt_out_func_proto, 10782 .is_valid_access = lwt_is_valid_access, 10783 .convert_ctx_access = bpf_convert_ctx_access, 10784 }; 10785 10786 const struct bpf_prog_ops lwt_out_prog_ops = { 10787 .test_run = bpf_prog_test_run_skb, 10788 }; 10789 10790 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 10791 .get_func_proto = lwt_xmit_func_proto, 10792 .is_valid_access = lwt_is_valid_access, 10793 .convert_ctx_access = bpf_convert_ctx_access, 10794 .gen_prologue = tc_cls_act_prologue, 10795 }; 10796 10797 const struct bpf_prog_ops lwt_xmit_prog_ops = { 10798 .test_run = bpf_prog_test_run_skb, 10799 }; 10800 10801 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 10802 .get_func_proto = lwt_seg6local_func_proto, 10803 .is_valid_access = lwt_is_valid_access, 10804 .convert_ctx_access = bpf_convert_ctx_access, 10805 }; 10806 10807 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 10808 .test_run = bpf_prog_test_run_skb, 10809 }; 10810 10811 const struct bpf_verifier_ops cg_sock_verifier_ops = { 10812 .get_func_proto = sock_filter_func_proto, 10813 .is_valid_access = sock_filter_is_valid_access, 10814 .convert_ctx_access = bpf_sock_convert_ctx_access, 10815 }; 10816 10817 const struct bpf_prog_ops cg_sock_prog_ops = { 10818 }; 10819 10820 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 10821 .get_func_proto = sock_addr_func_proto, 10822 .is_valid_access = sock_addr_is_valid_access, 10823 .convert_ctx_access = sock_addr_convert_ctx_access, 10824 }; 10825 10826 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 10827 }; 10828 10829 const struct bpf_verifier_ops sock_ops_verifier_ops = { 10830 .get_func_proto = sock_ops_func_proto, 10831 .is_valid_access = sock_ops_is_valid_access, 10832 .convert_ctx_access = sock_ops_convert_ctx_access, 10833 }; 10834 10835 const struct bpf_prog_ops sock_ops_prog_ops = { 10836 }; 10837 10838 const struct bpf_verifier_ops sk_skb_verifier_ops = { 10839 .get_func_proto = sk_skb_func_proto, 10840 .is_valid_access = sk_skb_is_valid_access, 10841 .convert_ctx_access = sk_skb_convert_ctx_access, 10842 .gen_prologue = sk_skb_prologue, 10843 }; 10844 10845 const struct bpf_prog_ops sk_skb_prog_ops = { 10846 }; 10847 10848 const struct bpf_verifier_ops sk_msg_verifier_ops = { 10849 .get_func_proto = sk_msg_func_proto, 10850 .is_valid_access = sk_msg_is_valid_access, 10851 .convert_ctx_access = sk_msg_convert_ctx_access, 10852 .gen_prologue = bpf_noop_prologue, 10853 }; 10854 10855 const struct bpf_prog_ops sk_msg_prog_ops = { 10856 }; 10857 10858 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 10859 .get_func_proto = flow_dissector_func_proto, 10860 .is_valid_access = flow_dissector_is_valid_access, 10861 .convert_ctx_access = flow_dissector_convert_ctx_access, 10862 }; 10863 10864 const struct bpf_prog_ops flow_dissector_prog_ops = { 10865 .test_run = bpf_prog_test_run_flow_dissector, 10866 }; 10867 10868 int sk_detach_filter(struct sock *sk) 10869 { 10870 int ret = -ENOENT; 10871 struct sk_filter *filter; 10872 10873 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 10874 return -EPERM; 10875 10876 filter = rcu_dereference_protected(sk->sk_filter, 10877 lockdep_sock_is_held(sk)); 10878 if (filter) { 10879 RCU_INIT_POINTER(sk->sk_filter, NULL); 10880 sk_filter_uncharge(sk, filter); 10881 ret = 0; 10882 } 10883 10884 return ret; 10885 } 10886 EXPORT_SYMBOL_GPL(sk_detach_filter); 10887 10888 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len) 10889 { 10890 struct sock_fprog_kern *fprog; 10891 struct sk_filter *filter; 10892 int ret = 0; 10893 10894 sockopt_lock_sock(sk); 10895 filter = rcu_dereference_protected(sk->sk_filter, 10896 lockdep_sock_is_held(sk)); 10897 if (!filter) 10898 goto out; 10899 10900 /* We're copying the filter that has been originally attached, 10901 * so no conversion/decode needed anymore. eBPF programs that 10902 * have no original program cannot be dumped through this. 10903 */ 10904 ret = -EACCES; 10905 fprog = filter->prog->orig_prog; 10906 if (!fprog) 10907 goto out; 10908 10909 ret = fprog->len; 10910 if (!len) 10911 /* User space only enquires number of filter blocks. */ 10912 goto out; 10913 10914 ret = -EINVAL; 10915 if (len < fprog->len) 10916 goto out; 10917 10918 ret = -EFAULT; 10919 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog))) 10920 goto out; 10921 10922 /* Instead of bytes, the API requests to return the number 10923 * of filter blocks. 10924 */ 10925 ret = fprog->len; 10926 out: 10927 sockopt_release_sock(sk); 10928 return ret; 10929 } 10930 10931 #ifdef CONFIG_INET 10932 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 10933 struct sock_reuseport *reuse, 10934 struct sock *sk, struct sk_buff *skb, 10935 struct sock *migrating_sk, 10936 u32 hash) 10937 { 10938 reuse_kern->skb = skb; 10939 reuse_kern->sk = sk; 10940 reuse_kern->selected_sk = NULL; 10941 reuse_kern->migrating_sk = migrating_sk; 10942 reuse_kern->data_end = skb->data + skb_headlen(skb); 10943 reuse_kern->hash = hash; 10944 reuse_kern->reuseport_id = reuse->reuseport_id; 10945 reuse_kern->bind_inany = reuse->bind_inany; 10946 } 10947 10948 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 10949 struct bpf_prog *prog, struct sk_buff *skb, 10950 struct sock *migrating_sk, 10951 u32 hash) 10952 { 10953 struct sk_reuseport_kern reuse_kern; 10954 enum sk_action action; 10955 10956 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash); 10957 action = bpf_prog_run(prog, &reuse_kern); 10958 10959 if (action == SK_PASS) 10960 return reuse_kern.selected_sk; 10961 else 10962 return ERR_PTR(-ECONNREFUSED); 10963 } 10964 10965 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 10966 struct bpf_map *, map, void *, key, u32, flags) 10967 { 10968 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 10969 struct sock_reuseport *reuse; 10970 struct sock *selected_sk; 10971 10972 selected_sk = map->ops->map_lookup_elem(map, key); 10973 if (!selected_sk) 10974 return -ENOENT; 10975 10976 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 10977 if (!reuse) { 10978 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 10979 if (sk_is_refcounted(selected_sk)) 10980 sock_put(selected_sk); 10981 10982 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 10983 * The only (!reuse) case here is - the sk has already been 10984 * unhashed (e.g. by close()), so treat it as -ENOENT. 10985 * 10986 * Other maps (e.g. sock_map) do not provide this guarantee and 10987 * the sk may never be in the reuseport group to begin with. 10988 */ 10989 return is_sockarray ? -ENOENT : -EINVAL; 10990 } 10991 10992 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 10993 struct sock *sk = reuse_kern->sk; 10994 10995 if (sk->sk_protocol != selected_sk->sk_protocol) 10996 return -EPROTOTYPE; 10997 else if (sk->sk_family != selected_sk->sk_family) 10998 return -EAFNOSUPPORT; 10999 11000 /* Catch all. Likely bound to a different sockaddr. */ 11001 return -EBADFD; 11002 } 11003 11004 reuse_kern->selected_sk = selected_sk; 11005 11006 return 0; 11007 } 11008 11009 static const struct bpf_func_proto sk_select_reuseport_proto = { 11010 .func = sk_select_reuseport, 11011 .gpl_only = false, 11012 .ret_type = RET_INTEGER, 11013 .arg1_type = ARG_PTR_TO_CTX, 11014 .arg2_type = ARG_CONST_MAP_PTR, 11015 .arg3_type = ARG_PTR_TO_MAP_KEY, 11016 .arg4_type = ARG_ANYTHING, 11017 }; 11018 11019 BPF_CALL_4(sk_reuseport_load_bytes, 11020 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 11021 void *, to, u32, len) 11022 { 11023 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 11024 } 11025 11026 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 11027 .func = sk_reuseport_load_bytes, 11028 .gpl_only = false, 11029 .ret_type = RET_INTEGER, 11030 .arg1_type = ARG_PTR_TO_CTX, 11031 .arg2_type = ARG_ANYTHING, 11032 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 11033 .arg4_type = ARG_CONST_SIZE, 11034 }; 11035 11036 BPF_CALL_5(sk_reuseport_load_bytes_relative, 11037 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 11038 void *, to, u32, len, u32, start_header) 11039 { 11040 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 11041 len, start_header); 11042 } 11043 11044 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 11045 .func = sk_reuseport_load_bytes_relative, 11046 .gpl_only = false, 11047 .ret_type = RET_INTEGER, 11048 .arg1_type = ARG_PTR_TO_CTX, 11049 .arg2_type = ARG_ANYTHING, 11050 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 11051 .arg4_type = ARG_CONST_SIZE, 11052 .arg5_type = ARG_ANYTHING, 11053 }; 11054 11055 static const struct bpf_func_proto * 11056 sk_reuseport_func_proto(enum bpf_func_id func_id, 11057 const struct bpf_prog *prog) 11058 { 11059 switch (func_id) { 11060 case BPF_FUNC_sk_select_reuseport: 11061 return &sk_select_reuseport_proto; 11062 case BPF_FUNC_skb_load_bytes: 11063 return &sk_reuseport_load_bytes_proto; 11064 case BPF_FUNC_skb_load_bytes_relative: 11065 return &sk_reuseport_load_bytes_relative_proto; 11066 case BPF_FUNC_get_socket_cookie: 11067 return &bpf_get_socket_ptr_cookie_proto; 11068 case BPF_FUNC_ktime_get_coarse_ns: 11069 return &bpf_ktime_get_coarse_ns_proto; 11070 default: 11071 return bpf_base_func_proto(func_id); 11072 } 11073 } 11074 11075 static bool 11076 sk_reuseport_is_valid_access(int off, int size, 11077 enum bpf_access_type type, 11078 const struct bpf_prog *prog, 11079 struct bpf_insn_access_aux *info) 11080 { 11081 const u32 size_default = sizeof(__u32); 11082 11083 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 11084 off % size || type != BPF_READ) 11085 return false; 11086 11087 switch (off) { 11088 case offsetof(struct sk_reuseport_md, data): 11089 info->reg_type = PTR_TO_PACKET; 11090 return size == sizeof(__u64); 11091 11092 case offsetof(struct sk_reuseport_md, data_end): 11093 info->reg_type = PTR_TO_PACKET_END; 11094 return size == sizeof(__u64); 11095 11096 case offsetof(struct sk_reuseport_md, hash): 11097 return size == size_default; 11098 11099 case offsetof(struct sk_reuseport_md, sk): 11100 info->reg_type = PTR_TO_SOCKET; 11101 return size == sizeof(__u64); 11102 11103 case offsetof(struct sk_reuseport_md, migrating_sk): 11104 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 11105 return size == sizeof(__u64); 11106 11107 /* Fields that allow narrowing */ 11108 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 11109 if (size < sizeof_field(struct sk_buff, protocol)) 11110 return false; 11111 fallthrough; 11112 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 11113 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 11114 case bpf_ctx_range(struct sk_reuseport_md, len): 11115 bpf_ctx_record_field_size(info, size_default); 11116 return bpf_ctx_narrow_access_ok(off, size, size_default); 11117 11118 default: 11119 return false; 11120 } 11121 } 11122 11123 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 11124 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 11125 si->dst_reg, si->src_reg, \ 11126 bpf_target_off(struct sk_reuseport_kern, F, \ 11127 sizeof_field(struct sk_reuseport_kern, F), \ 11128 target_size)); \ 11129 }) 11130 11131 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 11132 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11133 struct sk_buff, \ 11134 skb, \ 11135 SKB_FIELD) 11136 11137 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 11138 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11139 struct sock, \ 11140 sk, \ 11141 SK_FIELD) 11142 11143 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 11144 const struct bpf_insn *si, 11145 struct bpf_insn *insn_buf, 11146 struct bpf_prog *prog, 11147 u32 *target_size) 11148 { 11149 struct bpf_insn *insn = insn_buf; 11150 11151 switch (si->off) { 11152 case offsetof(struct sk_reuseport_md, data): 11153 SK_REUSEPORT_LOAD_SKB_FIELD(data); 11154 break; 11155 11156 case offsetof(struct sk_reuseport_md, len): 11157 SK_REUSEPORT_LOAD_SKB_FIELD(len); 11158 break; 11159 11160 case offsetof(struct sk_reuseport_md, eth_protocol): 11161 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 11162 break; 11163 11164 case offsetof(struct sk_reuseport_md, ip_protocol): 11165 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 11166 break; 11167 11168 case offsetof(struct sk_reuseport_md, data_end): 11169 SK_REUSEPORT_LOAD_FIELD(data_end); 11170 break; 11171 11172 case offsetof(struct sk_reuseport_md, hash): 11173 SK_REUSEPORT_LOAD_FIELD(hash); 11174 break; 11175 11176 case offsetof(struct sk_reuseport_md, bind_inany): 11177 SK_REUSEPORT_LOAD_FIELD(bind_inany); 11178 break; 11179 11180 case offsetof(struct sk_reuseport_md, sk): 11181 SK_REUSEPORT_LOAD_FIELD(sk); 11182 break; 11183 11184 case offsetof(struct sk_reuseport_md, migrating_sk): 11185 SK_REUSEPORT_LOAD_FIELD(migrating_sk); 11186 break; 11187 } 11188 11189 return insn - insn_buf; 11190 } 11191 11192 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 11193 .get_func_proto = sk_reuseport_func_proto, 11194 .is_valid_access = sk_reuseport_is_valid_access, 11195 .convert_ctx_access = sk_reuseport_convert_ctx_access, 11196 }; 11197 11198 const struct bpf_prog_ops sk_reuseport_prog_ops = { 11199 }; 11200 11201 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 11202 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 11203 11204 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 11205 struct sock *, sk, u64, flags) 11206 { 11207 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 11208 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 11209 return -EINVAL; 11210 if (unlikely(sk && sk_is_refcounted(sk))) 11211 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 11212 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN)) 11213 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */ 11214 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE)) 11215 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */ 11216 11217 /* Check if socket is suitable for packet L3/L4 protocol */ 11218 if (sk && sk->sk_protocol != ctx->protocol) 11219 return -EPROTOTYPE; 11220 if (sk && sk->sk_family != ctx->family && 11221 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 11222 return -EAFNOSUPPORT; 11223 11224 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 11225 return -EEXIST; 11226 11227 /* Select socket as lookup result */ 11228 ctx->selected_sk = sk; 11229 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 11230 return 0; 11231 } 11232 11233 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 11234 .func = bpf_sk_lookup_assign, 11235 .gpl_only = false, 11236 .ret_type = RET_INTEGER, 11237 .arg1_type = ARG_PTR_TO_CTX, 11238 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 11239 .arg3_type = ARG_ANYTHING, 11240 }; 11241 11242 static const struct bpf_func_proto * 11243 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 11244 { 11245 switch (func_id) { 11246 case BPF_FUNC_perf_event_output: 11247 return &bpf_event_output_data_proto; 11248 case BPF_FUNC_sk_assign: 11249 return &bpf_sk_lookup_assign_proto; 11250 case BPF_FUNC_sk_release: 11251 return &bpf_sk_release_proto; 11252 default: 11253 return bpf_sk_base_func_proto(func_id); 11254 } 11255 } 11256 11257 static bool sk_lookup_is_valid_access(int off, int size, 11258 enum bpf_access_type type, 11259 const struct bpf_prog *prog, 11260 struct bpf_insn_access_aux *info) 11261 { 11262 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 11263 return false; 11264 if (off % size != 0) 11265 return false; 11266 if (type != BPF_READ) 11267 return false; 11268 11269 switch (off) { 11270 case offsetof(struct bpf_sk_lookup, sk): 11271 info->reg_type = PTR_TO_SOCKET_OR_NULL; 11272 return size == sizeof(__u64); 11273 11274 case bpf_ctx_range(struct bpf_sk_lookup, family): 11275 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 11276 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 11277 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 11278 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 11279 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 11280 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 11281 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex): 11282 bpf_ctx_record_field_size(info, sizeof(__u32)); 11283 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 11284 11285 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 11286 /* Allow 4-byte access to 2-byte field for backward compatibility */ 11287 if (size == sizeof(__u32)) 11288 return true; 11289 bpf_ctx_record_field_size(info, sizeof(__be16)); 11290 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16)); 11291 11292 case offsetofend(struct bpf_sk_lookup, remote_port) ... 11293 offsetof(struct bpf_sk_lookup, local_ip4) - 1: 11294 /* Allow access to zero padding for backward compatibility */ 11295 bpf_ctx_record_field_size(info, sizeof(__u16)); 11296 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16)); 11297 11298 default: 11299 return false; 11300 } 11301 } 11302 11303 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 11304 const struct bpf_insn *si, 11305 struct bpf_insn *insn_buf, 11306 struct bpf_prog *prog, 11307 u32 *target_size) 11308 { 11309 struct bpf_insn *insn = insn_buf; 11310 11311 switch (si->off) { 11312 case offsetof(struct bpf_sk_lookup, sk): 11313 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11314 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 11315 break; 11316 11317 case offsetof(struct bpf_sk_lookup, family): 11318 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11319 bpf_target_off(struct bpf_sk_lookup_kern, 11320 family, 2, target_size)); 11321 break; 11322 11323 case offsetof(struct bpf_sk_lookup, protocol): 11324 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11325 bpf_target_off(struct bpf_sk_lookup_kern, 11326 protocol, 2, target_size)); 11327 break; 11328 11329 case offsetof(struct bpf_sk_lookup, remote_ip4): 11330 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11331 bpf_target_off(struct bpf_sk_lookup_kern, 11332 v4.saddr, 4, target_size)); 11333 break; 11334 11335 case offsetof(struct bpf_sk_lookup, local_ip4): 11336 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11337 bpf_target_off(struct bpf_sk_lookup_kern, 11338 v4.daddr, 4, target_size)); 11339 break; 11340 11341 case bpf_ctx_range_till(struct bpf_sk_lookup, 11342 remote_ip6[0], remote_ip6[3]): { 11343 #if IS_ENABLED(CONFIG_IPV6) 11344 int off = si->off; 11345 11346 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 11347 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11348 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11349 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 11350 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11351 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11352 #else 11353 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11354 #endif 11355 break; 11356 } 11357 case bpf_ctx_range_till(struct bpf_sk_lookup, 11358 local_ip6[0], local_ip6[3]): { 11359 #if IS_ENABLED(CONFIG_IPV6) 11360 int off = si->off; 11361 11362 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 11363 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11364 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11365 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 11366 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11367 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11368 #else 11369 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11370 #endif 11371 break; 11372 } 11373 case offsetof(struct bpf_sk_lookup, remote_port): 11374 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11375 bpf_target_off(struct bpf_sk_lookup_kern, 11376 sport, 2, target_size)); 11377 break; 11378 11379 case offsetofend(struct bpf_sk_lookup, remote_port): 11380 *target_size = 2; 11381 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11382 break; 11383 11384 case offsetof(struct bpf_sk_lookup, local_port): 11385 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11386 bpf_target_off(struct bpf_sk_lookup_kern, 11387 dport, 2, target_size)); 11388 break; 11389 11390 case offsetof(struct bpf_sk_lookup, ingress_ifindex): 11391 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11392 bpf_target_off(struct bpf_sk_lookup_kern, 11393 ingress_ifindex, 4, target_size)); 11394 break; 11395 } 11396 11397 return insn - insn_buf; 11398 } 11399 11400 const struct bpf_prog_ops sk_lookup_prog_ops = { 11401 .test_run = bpf_prog_test_run_sk_lookup, 11402 }; 11403 11404 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 11405 .get_func_proto = sk_lookup_func_proto, 11406 .is_valid_access = sk_lookup_is_valid_access, 11407 .convert_ctx_access = sk_lookup_convert_ctx_access, 11408 }; 11409 11410 #endif /* CONFIG_INET */ 11411 11412 DEFINE_BPF_DISPATCHER(xdp) 11413 11414 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 11415 { 11416 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 11417 } 11418 11419 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE) 11420 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 11421 BTF_SOCK_TYPE_xxx 11422 #undef BTF_SOCK_TYPE 11423 11424 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 11425 { 11426 /* tcp6_sock type is not generated in dwarf and hence btf, 11427 * trigger an explicit type generation here. 11428 */ 11429 BTF_TYPE_EMIT(struct tcp6_sock); 11430 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 11431 sk->sk_family == AF_INET6) 11432 return (unsigned long)sk; 11433 11434 return (unsigned long)NULL; 11435 } 11436 11437 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 11438 .func = bpf_skc_to_tcp6_sock, 11439 .gpl_only = false, 11440 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11441 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11442 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 11443 }; 11444 11445 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 11446 { 11447 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 11448 return (unsigned long)sk; 11449 11450 return (unsigned long)NULL; 11451 } 11452 11453 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 11454 .func = bpf_skc_to_tcp_sock, 11455 .gpl_only = false, 11456 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11457 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11458 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 11459 }; 11460 11461 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 11462 { 11463 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 11464 * generated if CONFIG_INET=n. Trigger an explicit generation here. 11465 */ 11466 BTF_TYPE_EMIT(struct inet_timewait_sock); 11467 BTF_TYPE_EMIT(struct tcp_timewait_sock); 11468 11469 #ifdef CONFIG_INET 11470 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 11471 return (unsigned long)sk; 11472 #endif 11473 11474 #if IS_BUILTIN(CONFIG_IPV6) 11475 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 11476 return (unsigned long)sk; 11477 #endif 11478 11479 return (unsigned long)NULL; 11480 } 11481 11482 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 11483 .func = bpf_skc_to_tcp_timewait_sock, 11484 .gpl_only = false, 11485 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11486 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11487 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 11488 }; 11489 11490 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 11491 { 11492 #ifdef CONFIG_INET 11493 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11494 return (unsigned long)sk; 11495 #endif 11496 11497 #if IS_BUILTIN(CONFIG_IPV6) 11498 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11499 return (unsigned long)sk; 11500 #endif 11501 11502 return (unsigned long)NULL; 11503 } 11504 11505 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 11506 .func = bpf_skc_to_tcp_request_sock, 11507 .gpl_only = false, 11508 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11509 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11510 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 11511 }; 11512 11513 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 11514 { 11515 /* udp6_sock type is not generated in dwarf and hence btf, 11516 * trigger an explicit type generation here. 11517 */ 11518 BTF_TYPE_EMIT(struct udp6_sock); 11519 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 11520 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 11521 return (unsigned long)sk; 11522 11523 return (unsigned long)NULL; 11524 } 11525 11526 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 11527 .func = bpf_skc_to_udp6_sock, 11528 .gpl_only = false, 11529 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11530 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11531 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 11532 }; 11533 11534 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk) 11535 { 11536 /* unix_sock type is not generated in dwarf and hence btf, 11537 * trigger an explicit type generation here. 11538 */ 11539 BTF_TYPE_EMIT(struct unix_sock); 11540 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX) 11541 return (unsigned long)sk; 11542 11543 return (unsigned long)NULL; 11544 } 11545 11546 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { 11547 .func = bpf_skc_to_unix_sock, 11548 .gpl_only = false, 11549 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11550 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11551 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX], 11552 }; 11553 11554 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) 11555 { 11556 BTF_TYPE_EMIT(struct mptcp_sock); 11557 return (unsigned long)bpf_mptcp_sock_from_subflow(sk); 11558 } 11559 11560 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { 11561 .func = bpf_skc_to_mptcp_sock, 11562 .gpl_only = false, 11563 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11564 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 11565 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP], 11566 }; 11567 11568 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 11569 { 11570 return (unsigned long)sock_from_file(file); 11571 } 11572 11573 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 11574 BTF_ID(struct, socket) 11575 BTF_ID(struct, file) 11576 11577 const struct bpf_func_proto bpf_sock_from_file_proto = { 11578 .func = bpf_sock_from_file, 11579 .gpl_only = false, 11580 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11581 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 11582 .arg1_type = ARG_PTR_TO_BTF_ID, 11583 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 11584 }; 11585 11586 static const struct bpf_func_proto * 11587 bpf_sk_base_func_proto(enum bpf_func_id func_id) 11588 { 11589 const struct bpf_func_proto *func; 11590 11591 switch (func_id) { 11592 case BPF_FUNC_skc_to_tcp6_sock: 11593 func = &bpf_skc_to_tcp6_sock_proto; 11594 break; 11595 case BPF_FUNC_skc_to_tcp_sock: 11596 func = &bpf_skc_to_tcp_sock_proto; 11597 break; 11598 case BPF_FUNC_skc_to_tcp_timewait_sock: 11599 func = &bpf_skc_to_tcp_timewait_sock_proto; 11600 break; 11601 case BPF_FUNC_skc_to_tcp_request_sock: 11602 func = &bpf_skc_to_tcp_request_sock_proto; 11603 break; 11604 case BPF_FUNC_skc_to_udp6_sock: 11605 func = &bpf_skc_to_udp6_sock_proto; 11606 break; 11607 case BPF_FUNC_skc_to_unix_sock: 11608 func = &bpf_skc_to_unix_sock_proto; 11609 break; 11610 case BPF_FUNC_skc_to_mptcp_sock: 11611 func = &bpf_skc_to_mptcp_sock_proto; 11612 break; 11613 case BPF_FUNC_ktime_get_coarse_ns: 11614 return &bpf_ktime_get_coarse_ns_proto; 11615 default: 11616 return bpf_base_func_proto(func_id); 11617 } 11618 11619 if (!perfmon_capable()) 11620 return NULL; 11621 11622 return func; 11623 } 11624