1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 3 * INET An implementation of the TCP Authentication Option (TCP-AO). 4 * See RFC5925. 5 * 6 * Authors: Dmitry Safonov <dima@arista.com> 7 * Francesco Ruggeri <fruggeri@arista.com> 8 * Salam Noureddine <noureddine@arista.com> 9 */ 10#define pr_fmt(fmt) "TCP: " fmt 11 12#include <crypto/hash.h> 13#include <linux/inetdevice.h> 14#include <linux/tcp.h> 15 16#include <net/tcp.h> 17#include <net/ipv6.h> 18#include <net/icmp.h> 19 20DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_ao_needed, HZ); 21 22int tcp_ao_calc_traffic_key(struct tcp_ao_key *mkt, u8 *key, void *ctx, 23 unsigned int len, struct tcp_sigpool *hp) 24{ 25 struct scatterlist sg; 26 int ret; 27 28 if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp->req), 29 mkt->key, mkt->keylen)) 30 goto clear_hash; 31 32 ret = crypto_ahash_init(hp->req); 33 if (ret) 34 goto clear_hash; 35 36 sg_init_one(&sg, ctx, len); 37 ahash_request_set_crypt(hp->req, &sg, key, len); 38 crypto_ahash_update(hp->req); 39 40 ret = crypto_ahash_final(hp->req); 41 if (ret) 42 goto clear_hash; 43 44 return 0; 45clear_hash: 46 memset(key, 0, tcp_ao_digest_size(mkt)); 47 return 1; 48} 49 50bool tcp_ao_ignore_icmp(const struct sock *sk, int family, int type, int code) 51{ 52 bool ignore_icmp = false; 53 struct tcp_ao_info *ao; 54 55 if (!static_branch_unlikely(&tcp_ao_needed.key)) 56 return false; 57 58 /* RFC5925, 7.8: 59 * >> A TCP-AO implementation MUST default to ignore incoming ICMPv4 60 * messages of Type 3 (destination unreachable), Codes 2-4 (protocol 61 * unreachable, port unreachable, and fragmentation needed -- ���hard 62 * errors���), and ICMPv6 Type 1 (destination unreachable), Code 1 63 * (administratively prohibited) and Code 4 (port unreachable) intended 64 * for connections in synchronized states (ESTABLISHED, FIN-WAIT-1, FIN- 65 * WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, TIME-WAIT) that match MKTs. 66 */ 67 if (family == AF_INET) { 68 if (type != ICMP_DEST_UNREACH) 69 return false; 70 if (code < ICMP_PROT_UNREACH || code > ICMP_FRAG_NEEDED) 71 return false; 72 } else { 73 if (type != ICMPV6_DEST_UNREACH) 74 return false; 75 if (code != ICMPV6_ADM_PROHIBITED && code != ICMPV6_PORT_UNREACH) 76 return false; 77 } 78 79 rcu_read_lock(); 80 switch (sk->sk_state) { 81 case TCP_TIME_WAIT: 82 ao = rcu_dereference(tcp_twsk(sk)->ao_info); 83 break; 84 case TCP_SYN_SENT: 85 case TCP_SYN_RECV: 86 case TCP_LISTEN: 87 case TCP_NEW_SYN_RECV: 88 /* RFC5925 specifies to ignore ICMPs *only* on connections 89 * in synchronized states. 90 */ 91 rcu_read_unlock(); 92 return false; 93 default: 94 ao = rcu_dereference(tcp_sk(sk)->ao_info); 95 } 96 97 if (ao && !ao->accept_icmps) { 98 ignore_icmp = true; 99 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAODROPPEDICMPS); 100 atomic64_inc(&ao->counters.dropped_icmp); 101 } 102 rcu_read_unlock(); 103 104 return ignore_icmp; 105} 106 107/* Optimized version of tcp_ao_do_lookup(): only for sockets for which 108 * it's known that the keys in ao_info are matching peer's 109 * family/address/VRF/etc. 110 */ 111struct tcp_ao_key *tcp_ao_established_key(struct tcp_ao_info *ao, 112 int sndid, int rcvid) 113{ 114 struct tcp_ao_key *key; 115 116 hlist_for_each_entry_rcu(key, &ao->head, node) { 117 if ((sndid >= 0 && key->sndid != sndid) || 118 (rcvid >= 0 && key->rcvid != rcvid)) 119 continue; 120 return key; 121 } 122 123 return NULL; 124} 125 126static int ipv4_prefix_cmp(const struct in_addr *addr1, 127 const struct in_addr *addr2, 128 unsigned int prefixlen) 129{ 130 __be32 mask = inet_make_mask(prefixlen); 131 __be32 a1 = addr1->s_addr & mask; 132 __be32 a2 = addr2->s_addr & mask; 133 134 if (a1 == a2) 135 return 0; 136 return memcmp(&a1, &a2, sizeof(a1)); 137} 138 139static int __tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index, 140 const union tcp_ao_addr *addr, u8 prefixlen, 141 int family, int sndid, int rcvid) 142{ 143 if (sndid >= 0 && key->sndid != sndid) 144 return (key->sndid > sndid) ? 1 : -1; 145 if (rcvid >= 0 && key->rcvid != rcvid) 146 return (key->rcvid > rcvid) ? 1 : -1; 147 if (l3index >= 0 && (key->keyflags & TCP_AO_KEYF_IFINDEX)) { 148 if (key->l3index != l3index) 149 return (key->l3index > l3index) ? 1 : -1; 150 } 151 152 if (family == AF_UNSPEC) 153 return 0; 154 if (key->family != family) 155 return (key->family > family) ? 1 : -1; 156 157 if (family == AF_INET) { 158 if (ntohl(key->addr.a4.s_addr) == INADDR_ANY) 159 return 0; 160 if (ntohl(addr->a4.s_addr) == INADDR_ANY) 161 return 0; 162 return ipv4_prefix_cmp(&key->addr.a4, &addr->a4, prefixlen); 163#if IS_ENABLED(CONFIG_IPV6) 164 } else { 165 if (ipv6_addr_any(&key->addr.a6) || ipv6_addr_any(&addr->a6)) 166 return 0; 167 if (ipv6_prefix_equal(&key->addr.a6, &addr->a6, prefixlen)) 168 return 0; 169 return memcmp(&key->addr.a6, &addr->a6, sizeof(addr->a6)); 170#endif 171 } 172 return -1; 173} 174 175static int tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index, 176 const union tcp_ao_addr *addr, u8 prefixlen, 177 int family, int sndid, int rcvid) 178{ 179#if IS_ENABLED(CONFIG_IPV6) 180 if (family == AF_INET6 && ipv6_addr_v4mapped(&addr->a6)) { 181 __be32 addr4 = addr->a6.s6_addr32[3]; 182 183 return __tcp_ao_key_cmp(key, l3index, 184 (union tcp_ao_addr *)&addr4, 185 prefixlen, AF_INET, sndid, rcvid); 186 } 187#endif 188 return __tcp_ao_key_cmp(key, l3index, addr, 189 prefixlen, family, sndid, rcvid); 190} 191 192static struct tcp_ao_key *__tcp_ao_do_lookup(const struct sock *sk, int l3index, 193 const union tcp_ao_addr *addr, int family, u8 prefix, 194 int sndid, int rcvid) 195{ 196 struct tcp_ao_key *key; 197 struct tcp_ao_info *ao; 198 199 if (!static_branch_unlikely(&tcp_ao_needed.key)) 200 return NULL; 201 202 ao = rcu_dereference_check(tcp_sk(sk)->ao_info, 203 lockdep_sock_is_held(sk)); 204 if (!ao) 205 return NULL; 206 207 hlist_for_each_entry_rcu(key, &ao->head, node) { 208 u8 prefixlen = min(prefix, key->prefixlen); 209 210 if (!tcp_ao_key_cmp(key, l3index, addr, prefixlen, 211 family, sndid, rcvid)) 212 return key; 213 } 214 return NULL; 215} 216 217struct tcp_ao_key *tcp_ao_do_lookup(const struct sock *sk, int l3index, 218 const union tcp_ao_addr *addr, 219 int family, int sndid, int rcvid) 220{ 221 return __tcp_ao_do_lookup(sk, l3index, addr, family, U8_MAX, sndid, rcvid); 222} 223 224static struct tcp_ao_info *tcp_ao_alloc_info(gfp_t flags) 225{ 226 struct tcp_ao_info *ao; 227 228 ao = kzalloc(sizeof(*ao), flags); 229 if (!ao) 230 return NULL; 231 INIT_HLIST_HEAD(&ao->head); 232 refcount_set(&ao->refcnt, 1); 233 234 return ao; 235} 236 237static void tcp_ao_link_mkt(struct tcp_ao_info *ao, struct tcp_ao_key *mkt) 238{ 239 hlist_add_head_rcu(&mkt->node, &ao->head); 240} 241 242static struct tcp_ao_key *tcp_ao_copy_key(struct sock *sk, 243 struct tcp_ao_key *key) 244{ 245 struct tcp_ao_key *new_key; 246 247 new_key = sock_kmalloc(sk, tcp_ao_sizeof_key(key), 248 GFP_ATOMIC); 249 if (!new_key) 250 return NULL; 251 252 *new_key = *key; 253 INIT_HLIST_NODE(&new_key->node); 254 tcp_sigpool_get(new_key->tcp_sigpool_id); 255 atomic64_set(&new_key->pkt_good, 0); 256 atomic64_set(&new_key->pkt_bad, 0); 257 258 return new_key; 259} 260 261static void tcp_ao_key_free_rcu(struct rcu_head *head) 262{ 263 struct tcp_ao_key *key = container_of(head, struct tcp_ao_key, rcu); 264 265 tcp_sigpool_release(key->tcp_sigpool_id); 266 kfree_sensitive(key); 267} 268 269void tcp_ao_destroy_sock(struct sock *sk, bool twsk) 270{ 271 struct tcp_ao_info *ao; 272 struct tcp_ao_key *key; 273 struct hlist_node *n; 274 275 if (twsk) { 276 ao = rcu_dereference_protected(tcp_twsk(sk)->ao_info, 1); 277 tcp_twsk(sk)->ao_info = NULL; 278 } else { 279 ao = rcu_dereference_protected(tcp_sk(sk)->ao_info, 1); 280 tcp_sk(sk)->ao_info = NULL; 281 } 282 283 if (!ao || !refcount_dec_and_test(&ao->refcnt)) 284 return; 285 286 hlist_for_each_entry_safe(key, n, &ao->head, node) { 287 hlist_del_rcu(&key->node); 288 if (!twsk) 289 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc); 290 call_rcu(&key->rcu, tcp_ao_key_free_rcu); 291 } 292 293 kfree_rcu(ao, rcu); 294 static_branch_slow_dec_deferred(&tcp_ao_needed); 295} 296 297void tcp_ao_time_wait(struct tcp_timewait_sock *tcptw, struct tcp_sock *tp) 298{ 299 struct tcp_ao_info *ao_info = rcu_dereference_protected(tp->ao_info, 1); 300 301 if (ao_info) { 302 struct tcp_ao_key *key; 303 struct hlist_node *n; 304 int omem = 0; 305 306 hlist_for_each_entry_safe(key, n, &ao_info->head, node) { 307 omem += tcp_ao_sizeof_key(key); 308 } 309 310 refcount_inc(&ao_info->refcnt); 311 atomic_sub(omem, &(((struct sock *)tp)->sk_omem_alloc)); 312 rcu_assign_pointer(tcptw->ao_info, ao_info); 313 } else { 314 tcptw->ao_info = NULL; 315 } 316} 317 318/* 4 tuple and ISNs are expected in NBO */ 319static int tcp_v4_ao_calc_key(struct tcp_ao_key *mkt, u8 *key, 320 __be32 saddr, __be32 daddr, 321 __be16 sport, __be16 dport, 322 __be32 sisn, __be32 disn) 323{ 324 /* See RFC5926 3.1.1 */ 325 struct kdf_input_block { 326 u8 counter; 327 u8 label[6]; 328 struct tcp4_ao_context ctx; 329 __be16 outlen; 330 } __packed * tmp; 331 struct tcp_sigpool hp; 332 int err; 333 334 err = tcp_sigpool_start(mkt->tcp_sigpool_id, &hp); 335 if (err) 336 return err; 337 338 tmp = hp.scratch; 339 tmp->counter = 1; 340 memcpy(tmp->label, "TCP-AO", 6); 341 tmp->ctx.saddr = saddr; 342 tmp->ctx.daddr = daddr; 343 tmp->ctx.sport = sport; 344 tmp->ctx.dport = dport; 345 tmp->ctx.sisn = sisn; 346 tmp->ctx.disn = disn; 347 tmp->outlen = htons(tcp_ao_digest_size(mkt) * 8); /* in bits */ 348 349 err = tcp_ao_calc_traffic_key(mkt, key, tmp, sizeof(*tmp), &hp); 350 tcp_sigpool_end(&hp); 351 352 return err; 353} 354 355int tcp_v4_ao_calc_key_sk(struct tcp_ao_key *mkt, u8 *key, 356 const struct sock *sk, 357 __be32 sisn, __be32 disn, bool send) 358{ 359 if (send) 360 return tcp_v4_ao_calc_key(mkt, key, sk->sk_rcv_saddr, 361 sk->sk_daddr, htons(sk->sk_num), 362 sk->sk_dport, sisn, disn); 363 else 364 return tcp_v4_ao_calc_key(mkt, key, sk->sk_daddr, 365 sk->sk_rcv_saddr, sk->sk_dport, 366 htons(sk->sk_num), disn, sisn); 367} 368 369static int tcp_ao_calc_key_sk(struct tcp_ao_key *mkt, u8 *key, 370 const struct sock *sk, 371 __be32 sisn, __be32 disn, bool send) 372{ 373 if (mkt->family == AF_INET) 374 return tcp_v4_ao_calc_key_sk(mkt, key, sk, sisn, disn, send); 375#if IS_ENABLED(CONFIG_IPV6) 376 else if (mkt->family == AF_INET6) 377 return tcp_v6_ao_calc_key_sk(mkt, key, sk, sisn, disn, send); 378#endif 379 else 380 return -EOPNOTSUPP; 381} 382 383int tcp_v4_ao_calc_key_rsk(struct tcp_ao_key *mkt, u8 *key, 384 struct request_sock *req) 385{ 386 struct inet_request_sock *ireq = inet_rsk(req); 387 388 return tcp_v4_ao_calc_key(mkt, key, 389 ireq->ir_loc_addr, ireq->ir_rmt_addr, 390 htons(ireq->ir_num), ireq->ir_rmt_port, 391 htonl(tcp_rsk(req)->snt_isn), 392 htonl(tcp_rsk(req)->rcv_isn)); 393} 394 395static int tcp_v4_ao_calc_key_skb(struct tcp_ao_key *mkt, u8 *key, 396 const struct sk_buff *skb, 397 __be32 sisn, __be32 disn) 398{ 399 const struct iphdr *iph = ip_hdr(skb); 400 const struct tcphdr *th = tcp_hdr(skb); 401 402 return tcp_v4_ao_calc_key(mkt, key, iph->saddr, iph->daddr, 403 th->source, th->dest, sisn, disn); 404} 405 406static int tcp_ao_calc_key_skb(struct tcp_ao_key *mkt, u8 *key, 407 const struct sk_buff *skb, 408 __be32 sisn, __be32 disn, int family) 409{ 410 if (family == AF_INET) 411 return tcp_v4_ao_calc_key_skb(mkt, key, skb, sisn, disn); 412#if IS_ENABLED(CONFIG_IPV6) 413 else if (family == AF_INET6) 414 return tcp_v6_ao_calc_key_skb(mkt, key, skb, sisn, disn); 415#endif 416 return -EAFNOSUPPORT; 417} 418 419static int tcp_v4_ao_hash_pseudoheader(struct tcp_sigpool *hp, 420 __be32 daddr, __be32 saddr, 421 int nbytes) 422{ 423 struct tcp4_pseudohdr *bp; 424 struct scatterlist sg; 425 426 bp = hp->scratch; 427 bp->saddr = saddr; 428 bp->daddr = daddr; 429 bp->pad = 0; 430 bp->protocol = IPPROTO_TCP; 431 bp->len = cpu_to_be16(nbytes); 432 433 sg_init_one(&sg, bp, sizeof(*bp)); 434 ahash_request_set_crypt(hp->req, &sg, NULL, sizeof(*bp)); 435 return crypto_ahash_update(hp->req); 436} 437 438static int tcp_ao_hash_pseudoheader(unsigned short int family, 439 const struct sock *sk, 440 const struct sk_buff *skb, 441 struct tcp_sigpool *hp, int nbytes) 442{ 443 const struct tcphdr *th = tcp_hdr(skb); 444 445 /* TODO: Can we rely on checksum being zero to mean outbound pkt? */ 446 if (!th->check) { 447 if (family == AF_INET) 448 return tcp_v4_ao_hash_pseudoheader(hp, sk->sk_daddr, 449 sk->sk_rcv_saddr, skb->len); 450#if IS_ENABLED(CONFIG_IPV6) 451 else if (family == AF_INET6) 452 return tcp_v6_ao_hash_pseudoheader(hp, &sk->sk_v6_daddr, 453 &sk->sk_v6_rcv_saddr, skb->len); 454#endif 455 else 456 return -EAFNOSUPPORT; 457 } 458 459 if (family == AF_INET) { 460 const struct iphdr *iph = ip_hdr(skb); 461 462 return tcp_v4_ao_hash_pseudoheader(hp, iph->daddr, 463 iph->saddr, skb->len); 464#if IS_ENABLED(CONFIG_IPV6) 465 } else if (family == AF_INET6) { 466 const struct ipv6hdr *iph = ipv6_hdr(skb); 467 468 return tcp_v6_ao_hash_pseudoheader(hp, &iph->daddr, 469 &iph->saddr, skb->len); 470#endif 471 } 472 return -EAFNOSUPPORT; 473} 474 475u32 tcp_ao_compute_sne(u32 next_sne, u32 next_seq, u32 seq) 476{ 477 u32 sne = next_sne; 478 479 if (before(seq, next_seq)) { 480 if (seq > next_seq) 481 sne--; 482 } else { 483 if (seq < next_seq) 484 sne++; 485 } 486 487 return sne; 488} 489 490/* tcp_ao_hash_sne(struct tcp_sigpool *hp) 491 * @hp - used for hashing 492 * @sne - sne value 493 */ 494static int tcp_ao_hash_sne(struct tcp_sigpool *hp, u32 sne) 495{ 496 struct scatterlist sg; 497 __be32 *bp; 498 499 bp = (__be32 *)hp->scratch; 500 *bp = htonl(sne); 501 502 sg_init_one(&sg, bp, sizeof(*bp)); 503 ahash_request_set_crypt(hp->req, &sg, NULL, sizeof(*bp)); 504 return crypto_ahash_update(hp->req); 505} 506 507static int tcp_ao_hash_header(struct tcp_sigpool *hp, 508 const struct tcphdr *th, 509 bool exclude_options, u8 *hash, 510 int hash_offset, int hash_len) 511{ 512 struct scatterlist sg; 513 u8 *hdr = hp->scratch; 514 int err, len; 515 516 /* We are not allowed to change tcphdr, make a local copy */ 517 if (exclude_options) { 518 len = sizeof(*th) + sizeof(struct tcp_ao_hdr) + hash_len; 519 memcpy(hdr, th, sizeof(*th)); 520 memcpy(hdr + sizeof(*th), 521 (u8 *)th + hash_offset - sizeof(struct tcp_ao_hdr), 522 sizeof(struct tcp_ao_hdr)); 523 memset(hdr + sizeof(*th) + sizeof(struct tcp_ao_hdr), 524 0, hash_len); 525 ((struct tcphdr *)hdr)->check = 0; 526 } else { 527 len = th->doff << 2; 528 memcpy(hdr, th, len); 529 /* zero out tcp-ao hash */ 530 ((struct tcphdr *)hdr)->check = 0; 531 memset(hdr + hash_offset, 0, hash_len); 532 } 533 534 sg_init_one(&sg, hdr, len); 535 ahash_request_set_crypt(hp->req, &sg, NULL, len); 536 err = crypto_ahash_update(hp->req); 537 WARN_ON_ONCE(err != 0); 538 return err; 539} 540 541int tcp_ao_hash_hdr(unsigned short int family, char *ao_hash, 542 struct tcp_ao_key *key, const u8 *tkey, 543 const union tcp_ao_addr *daddr, 544 const union tcp_ao_addr *saddr, 545 const struct tcphdr *th, u32 sne) 546{ 547 int tkey_len = tcp_ao_digest_size(key); 548 int hash_offset = ao_hash - (char *)th; 549 struct tcp_sigpool hp; 550 void *hash_buf = NULL; 551 552 hash_buf = kmalloc(tkey_len, GFP_ATOMIC); 553 if (!hash_buf) 554 goto clear_hash_noput; 555 556 if (tcp_sigpool_start(key->tcp_sigpool_id, &hp)) 557 goto clear_hash_noput; 558 559 if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp.req), tkey, tkey_len)) 560 goto clear_hash; 561 562 if (crypto_ahash_init(hp.req)) 563 goto clear_hash; 564 565 if (tcp_ao_hash_sne(&hp, sne)) 566 goto clear_hash; 567 if (family == AF_INET) { 568 if (tcp_v4_ao_hash_pseudoheader(&hp, daddr->a4.s_addr, 569 saddr->a4.s_addr, th->doff * 4)) 570 goto clear_hash; 571#if IS_ENABLED(CONFIG_IPV6) 572 } else if (family == AF_INET6) { 573 if (tcp_v6_ao_hash_pseudoheader(&hp, &daddr->a6, 574 &saddr->a6, th->doff * 4)) 575 goto clear_hash; 576#endif 577 } else { 578 WARN_ON_ONCE(1); 579 goto clear_hash; 580 } 581 if (tcp_ao_hash_header(&hp, th, 582 !!(key->keyflags & TCP_AO_KEYF_EXCLUDE_OPT), 583 ao_hash, hash_offset, tcp_ao_maclen(key))) 584 goto clear_hash; 585 ahash_request_set_crypt(hp.req, NULL, hash_buf, 0); 586 if (crypto_ahash_final(hp.req)) 587 goto clear_hash; 588 589 memcpy(ao_hash, hash_buf, tcp_ao_maclen(key)); 590 tcp_sigpool_end(&hp); 591 kfree(hash_buf); 592 return 0; 593 594clear_hash: 595 tcp_sigpool_end(&hp); 596clear_hash_noput: 597 memset(ao_hash, 0, tcp_ao_maclen(key)); 598 kfree(hash_buf); 599 return 1; 600} 601 602int tcp_ao_hash_skb(unsigned short int family, 603 char *ao_hash, struct tcp_ao_key *key, 604 const struct sock *sk, const struct sk_buff *skb, 605 const u8 *tkey, int hash_offset, u32 sne) 606{ 607 const struct tcphdr *th = tcp_hdr(skb); 608 int tkey_len = tcp_ao_digest_size(key); 609 struct tcp_sigpool hp; 610 void *hash_buf = NULL; 611 612 hash_buf = kmalloc(tkey_len, GFP_ATOMIC); 613 if (!hash_buf) 614 goto clear_hash_noput; 615 616 if (tcp_sigpool_start(key->tcp_sigpool_id, &hp)) 617 goto clear_hash_noput; 618 619 if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp.req), tkey, tkey_len)) 620 goto clear_hash; 621 622 /* For now use sha1 by default. Depends on alg in tcp_ao_key */ 623 if (crypto_ahash_init(hp.req)) 624 goto clear_hash; 625 626 if (tcp_ao_hash_sne(&hp, sne)) 627 goto clear_hash; 628 if (tcp_ao_hash_pseudoheader(family, sk, skb, &hp, skb->len)) 629 goto clear_hash; 630 if (tcp_ao_hash_header(&hp, th, 631 !!(key->keyflags & TCP_AO_KEYF_EXCLUDE_OPT), 632 ao_hash, hash_offset, tcp_ao_maclen(key))) 633 goto clear_hash; 634 if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2)) 635 goto clear_hash; 636 ahash_request_set_crypt(hp.req, NULL, hash_buf, 0); 637 if (crypto_ahash_final(hp.req)) 638 goto clear_hash; 639 640 memcpy(ao_hash, hash_buf, tcp_ao_maclen(key)); 641 tcp_sigpool_end(&hp); 642 kfree(hash_buf); 643 return 0; 644 645clear_hash: 646 tcp_sigpool_end(&hp); 647clear_hash_noput: 648 memset(ao_hash, 0, tcp_ao_maclen(key)); 649 kfree(hash_buf); 650 return 1; 651} 652 653int tcp_v4_ao_hash_skb(char *ao_hash, struct tcp_ao_key *key, 654 const struct sock *sk, const struct sk_buff *skb, 655 const u8 *tkey, int hash_offset, u32 sne) 656{ 657 return tcp_ao_hash_skb(AF_INET, ao_hash, key, sk, skb, 658 tkey, hash_offset, sne); 659} 660 661int tcp_v4_ao_synack_hash(char *ao_hash, struct tcp_ao_key *ao_key, 662 struct request_sock *req, const struct sk_buff *skb, 663 int hash_offset, u32 sne) 664{ 665 void *hash_buf = NULL; 666 int err; 667 668 hash_buf = kmalloc(tcp_ao_digest_size(ao_key), GFP_ATOMIC); 669 if (!hash_buf) 670 return -ENOMEM; 671 672 err = tcp_v4_ao_calc_key_rsk(ao_key, hash_buf, req); 673 if (err) 674 goto out; 675 676 err = tcp_ao_hash_skb(AF_INET, ao_hash, ao_key, req_to_sk(req), skb, 677 hash_buf, hash_offset, sne); 678out: 679 kfree(hash_buf); 680 return err; 681} 682 683struct tcp_ao_key *tcp_v4_ao_lookup_rsk(const struct sock *sk, 684 struct request_sock *req, 685 int sndid, int rcvid) 686{ 687 struct inet_request_sock *ireq = inet_rsk(req); 688 union tcp_ao_addr *addr = (union tcp_ao_addr *)&ireq->ir_rmt_addr; 689 int l3index; 690 691 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif); 692 return tcp_ao_do_lookup(sk, l3index, addr, AF_INET, sndid, rcvid); 693} 694 695struct tcp_ao_key *tcp_v4_ao_lookup(const struct sock *sk, struct sock *addr_sk, 696 int sndid, int rcvid) 697{ 698 int l3index = l3mdev_master_ifindex_by_index(sock_net(sk), 699 addr_sk->sk_bound_dev_if); 700 union tcp_ao_addr *addr = (union tcp_ao_addr *)&addr_sk->sk_daddr; 701 702 return tcp_ao_do_lookup(sk, l3index, addr, AF_INET, sndid, rcvid); 703} 704 705int tcp_ao_prepare_reset(const struct sock *sk, struct sk_buff *skb, 706 const struct tcp_ao_hdr *aoh, int l3index, u32 seq, 707 struct tcp_ao_key **key, char **traffic_key, 708 bool *allocated_traffic_key, u8 *keyid, u32 *sne) 709{ 710 const struct tcphdr *th = tcp_hdr(skb); 711 struct tcp_ao_info *ao_info; 712 713 *allocated_traffic_key = false; 714 /* If there's no socket - than initial sisn/disn are unknown. 715 * Drop the segment. RFC5925 (7.7) advises to require graceful 716 * restart [RFC4724]. Alternatively, the RFC5925 advises to 717 * save/restore traffic keys before/after reboot. 718 * Linux TCP-AO support provides TCP_AO_ADD_KEY and TCP_AO_REPAIR 719 * options to restore a socket post-reboot. 720 */ 721 if (!sk) 722 return -ENOTCONN; 723 724 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV)) { 725 unsigned int family = READ_ONCE(sk->sk_family); 726 union tcp_ao_addr *addr; 727 __be32 disn, sisn; 728 729 if (sk->sk_state == TCP_NEW_SYN_RECV) { 730 struct request_sock *req = inet_reqsk(sk); 731 732 sisn = htonl(tcp_rsk(req)->rcv_isn); 733 disn = htonl(tcp_rsk(req)->snt_isn); 734 *sne = tcp_ao_compute_sne(0, tcp_rsk(req)->snt_isn, seq); 735 } else { 736 sisn = th->seq; 737 disn = 0; 738 } 739 if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) 740 addr = (union tcp_md5_addr *)&ipv6_hdr(skb)->saddr; 741 else 742 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr; 743#if IS_ENABLED(CONFIG_IPV6) 744 if (family == AF_INET6 && ipv6_addr_v4mapped(&sk->sk_v6_daddr)) 745 family = AF_INET; 746#endif 747 748 sk = sk_const_to_full_sk(sk); 749 ao_info = rcu_dereference(tcp_sk(sk)->ao_info); 750 if (!ao_info) 751 return -ENOENT; 752 *key = tcp_ao_do_lookup(sk, l3index, addr, family, 753 -1, aoh->rnext_keyid); 754 if (!*key) 755 return -ENOENT; 756 *traffic_key = kmalloc(tcp_ao_digest_size(*key), GFP_ATOMIC); 757 if (!*traffic_key) 758 return -ENOMEM; 759 *allocated_traffic_key = true; 760 if (tcp_ao_calc_key_skb(*key, *traffic_key, skb, 761 sisn, disn, family)) 762 return -1; 763 *keyid = (*key)->rcvid; 764 } else { 765 struct tcp_ao_key *rnext_key; 766 u32 snd_basis; 767 768 if (sk->sk_state == TCP_TIME_WAIT) { 769 ao_info = rcu_dereference(tcp_twsk(sk)->ao_info); 770 snd_basis = tcp_twsk(sk)->tw_snd_nxt; 771 } else { 772 ao_info = rcu_dereference(tcp_sk(sk)->ao_info); 773 snd_basis = tcp_sk(sk)->snd_una; 774 } 775 if (!ao_info) 776 return -ENOENT; 777 778 *key = tcp_ao_established_key(ao_info, aoh->rnext_keyid, -1); 779 if (!*key) 780 return -ENOENT; 781 *traffic_key = snd_other_key(*key); 782 rnext_key = READ_ONCE(ao_info->rnext_key); 783 *keyid = rnext_key->rcvid; 784 *sne = tcp_ao_compute_sne(READ_ONCE(ao_info->snd_sne), 785 snd_basis, seq); 786 } 787 return 0; 788} 789 790int tcp_ao_transmit_skb(struct sock *sk, struct sk_buff *skb, 791 struct tcp_ao_key *key, struct tcphdr *th, 792 __u8 *hash_location) 793{ 794 struct tcp_skb_cb *tcb = TCP_SKB_CB(skb); 795 struct tcp_sock *tp = tcp_sk(sk); 796 struct tcp_ao_info *ao; 797 void *tkey_buf = NULL; 798 u8 *traffic_key; 799 u32 sne; 800 801 ao = rcu_dereference_protected(tcp_sk(sk)->ao_info, 802 lockdep_sock_is_held(sk)); 803 traffic_key = snd_other_key(key); 804 if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) { 805 __be32 disn; 806 807 if (!(tcb->tcp_flags & TCPHDR_ACK)) { 808 disn = 0; 809 tkey_buf = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC); 810 if (!tkey_buf) 811 return -ENOMEM; 812 traffic_key = tkey_buf; 813 } else { 814 disn = ao->risn; 815 } 816 tp->af_specific->ao_calc_key_sk(key, traffic_key, 817 sk, ao->lisn, disn, true); 818 } 819 sne = tcp_ao_compute_sne(READ_ONCE(ao->snd_sne), READ_ONCE(tp->snd_una), 820 ntohl(th->seq)); 821 tp->af_specific->calc_ao_hash(hash_location, key, sk, skb, traffic_key, 822 hash_location - (u8 *)th, sne); 823 kfree(tkey_buf); 824 return 0; 825} 826 827static struct tcp_ao_key *tcp_ao_inbound_lookup(unsigned short int family, 828 const struct sock *sk, const struct sk_buff *skb, 829 int sndid, int rcvid, int l3index) 830{ 831 if (family == AF_INET) { 832 const struct iphdr *iph = ip_hdr(skb); 833 834 return tcp_ao_do_lookup(sk, l3index, 835 (union tcp_ao_addr *)&iph->saddr, 836 AF_INET, sndid, rcvid); 837 } else { 838 const struct ipv6hdr *iph = ipv6_hdr(skb); 839 840 return tcp_ao_do_lookup(sk, l3index, 841 (union tcp_ao_addr *)&iph->saddr, 842 AF_INET6, sndid, rcvid); 843 } 844} 845 846void tcp_ao_syncookie(struct sock *sk, const struct sk_buff *skb, 847 struct request_sock *req, unsigned short int family) 848{ 849 struct tcp_request_sock *treq = tcp_rsk(req); 850 const struct tcphdr *th = tcp_hdr(skb); 851 const struct tcp_ao_hdr *aoh; 852 struct tcp_ao_key *key; 853 int l3index; 854 855 /* treq->af_specific is used to perform TCP_AO lookup 856 * in tcp_create_openreq_child(). 857 */ 858#if IS_ENABLED(CONFIG_IPV6) 859 if (family == AF_INET6) 860 treq->af_specific = &tcp_request_sock_ipv6_ops; 861 else 862#endif 863 treq->af_specific = &tcp_request_sock_ipv4_ops; 864 865 treq->used_tcp_ao = false; 866 867 if (tcp_parse_auth_options(th, NULL, &aoh) || !aoh) 868 return; 869 870 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), inet_rsk(req)->ir_iif); 871 key = tcp_ao_inbound_lookup(family, sk, skb, -1, aoh->keyid, l3index); 872 if (!key) 873 /* Key not found, continue without TCP-AO */ 874 return; 875 876 treq->ao_rcv_next = aoh->keyid; 877 treq->ao_keyid = aoh->rnext_keyid; 878 treq->used_tcp_ao = true; 879} 880 881static enum skb_drop_reason 882tcp_ao_verify_hash(const struct sock *sk, const struct sk_buff *skb, 883 unsigned short int family, struct tcp_ao_info *info, 884 const struct tcp_ao_hdr *aoh, struct tcp_ao_key *key, 885 u8 *traffic_key, u8 *phash, u32 sne, int l3index) 886{ 887 u8 maclen = aoh->length - sizeof(struct tcp_ao_hdr); 888 const struct tcphdr *th = tcp_hdr(skb); 889 void *hash_buf = NULL; 890 891 if (maclen != tcp_ao_maclen(key)) { 892 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD); 893 atomic64_inc(&info->counters.pkt_bad); 894 atomic64_inc(&key->pkt_bad); 895 tcp_hash_fail("AO hash wrong length", family, skb, 896 "%u != %d L3index: %d", maclen, 897 tcp_ao_maclen(key), l3index); 898 return SKB_DROP_REASON_TCP_AOFAILURE; 899 } 900 901 hash_buf = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC); 902 if (!hash_buf) 903 return SKB_DROP_REASON_NOT_SPECIFIED; 904 905 /* XXX: make it per-AF callback? */ 906 tcp_ao_hash_skb(family, hash_buf, key, sk, skb, traffic_key, 907 (phash - (u8 *)th), sne); 908 if (memcmp(phash, hash_buf, maclen)) { 909 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD); 910 atomic64_inc(&info->counters.pkt_bad); 911 atomic64_inc(&key->pkt_bad); 912 tcp_hash_fail("AO hash mismatch", family, skb, 913 "L3index: %d", l3index); 914 kfree(hash_buf); 915 return SKB_DROP_REASON_TCP_AOFAILURE; 916 } 917 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOGOOD); 918 atomic64_inc(&info->counters.pkt_good); 919 atomic64_inc(&key->pkt_good); 920 kfree(hash_buf); 921 return SKB_NOT_DROPPED_YET; 922} 923 924enum skb_drop_reason 925tcp_inbound_ao_hash(struct sock *sk, const struct sk_buff *skb, 926 unsigned short int family, const struct request_sock *req, 927 int l3index, const struct tcp_ao_hdr *aoh) 928{ 929 const struct tcphdr *th = tcp_hdr(skb); 930 u8 *phash = (u8 *)(aoh + 1); /* hash goes just after the header */ 931 struct tcp_ao_info *info; 932 enum skb_drop_reason ret; 933 struct tcp_ao_key *key; 934 __be32 sisn, disn; 935 u8 *traffic_key; 936 u32 sne = 0; 937 938 info = rcu_dereference(tcp_sk(sk)->ao_info); 939 if (!info) { 940 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOKEYNOTFOUND); 941 tcp_hash_fail("AO key not found", family, skb, 942 "keyid: %u L3index: %d", aoh->keyid, l3index); 943 return SKB_DROP_REASON_TCP_AOUNEXPECTED; 944 } 945 946 if (unlikely(th->syn)) { 947 sisn = th->seq; 948 disn = 0; 949 } 950 951 /* Fast-path */ 952 if (likely((1 << sk->sk_state) & TCP_AO_ESTABLISHED)) { 953 enum skb_drop_reason err; 954 struct tcp_ao_key *current_key; 955 956 /* Check if this socket's rnext_key matches the keyid in the 957 * packet. If not we lookup the key based on the keyid 958 * matching the rcvid in the mkt. 959 */ 960 key = READ_ONCE(info->rnext_key); 961 if (key->rcvid != aoh->keyid) { 962 key = tcp_ao_established_key(info, -1, aoh->keyid); 963 if (!key) 964 goto key_not_found; 965 } 966 967 /* Delayed retransmitted SYN */ 968 if (unlikely(th->syn && !th->ack)) 969 goto verify_hash; 970 971 sne = tcp_ao_compute_sne(info->rcv_sne, tcp_sk(sk)->rcv_nxt, 972 ntohl(th->seq)); 973 /* Established socket, traffic key are cached */ 974 traffic_key = rcv_other_key(key); 975 err = tcp_ao_verify_hash(sk, skb, family, info, aoh, key, 976 traffic_key, phash, sne, l3index); 977 if (err) 978 return err; 979 current_key = READ_ONCE(info->current_key); 980 /* Key rotation: the peer asks us to use new key (RNext) */ 981 if (unlikely(aoh->rnext_keyid != current_key->sndid)) { 982 /* If the key is not found we do nothing. */ 983 key = tcp_ao_established_key(info, aoh->rnext_keyid, -1); 984 if (key) 985 /* pairs with tcp_ao_del_cmd */ 986 WRITE_ONCE(info->current_key, key); 987 } 988 return SKB_NOT_DROPPED_YET; 989 } 990 991 /* Lookup key based on peer address and keyid. 992 * current_key and rnext_key must not be used on tcp listen 993 * sockets as otherwise: 994 * - request sockets would race on those key pointers 995 * - tcp_ao_del_cmd() allows async key removal 996 */ 997 key = tcp_ao_inbound_lookup(family, sk, skb, -1, aoh->keyid, l3index); 998 if (!key) 999 goto key_not_found; 1000 1001 if (th->syn && !th->ack) 1002 goto verify_hash; 1003 1004 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV)) { 1005 /* Make the initial syn the likely case here */ 1006 if (unlikely(req)) { 1007 sne = tcp_ao_compute_sne(0, tcp_rsk(req)->rcv_isn, 1008 ntohl(th->seq)); 1009 sisn = htonl(tcp_rsk(req)->rcv_isn); 1010 disn = htonl(tcp_rsk(req)->snt_isn); 1011 } else if (unlikely(th->ack && !th->syn)) { 1012 /* Possible syncookie packet */ 1013 sisn = htonl(ntohl(th->seq) - 1); 1014 disn = htonl(ntohl(th->ack_seq) - 1); 1015 sne = tcp_ao_compute_sne(0, ntohl(sisn), 1016 ntohl(th->seq)); 1017 } else if (unlikely(!th->syn)) { 1018 /* no way to figure out initial sisn/disn - drop */ 1019 return SKB_DROP_REASON_TCP_FLAGS; 1020 } 1021 } else if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1022 disn = info->lisn; 1023 if (th->syn || th->rst) 1024 sisn = th->seq; 1025 else 1026 sisn = info->risn; 1027 } else { 1028 WARN_ONCE(1, "TCP-AO: Unexpected sk_state %d", sk->sk_state); 1029 return SKB_DROP_REASON_TCP_AOFAILURE; 1030 } 1031verify_hash: 1032 traffic_key = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC); 1033 if (!traffic_key) 1034 return SKB_DROP_REASON_NOT_SPECIFIED; 1035 tcp_ao_calc_key_skb(key, traffic_key, skb, sisn, disn, family); 1036 ret = tcp_ao_verify_hash(sk, skb, family, info, aoh, key, 1037 traffic_key, phash, sne, l3index); 1038 kfree(traffic_key); 1039 return ret; 1040 1041key_not_found: 1042 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOKEYNOTFOUND); 1043 atomic64_inc(&info->counters.key_not_found); 1044 tcp_hash_fail("Requested by the peer AO key id not found", 1045 family, skb, "L3index: %d", l3index); 1046 return SKB_DROP_REASON_TCP_AOKEYNOTFOUND; 1047} 1048 1049static int tcp_ao_cache_traffic_keys(const struct sock *sk, 1050 struct tcp_ao_info *ao, 1051 struct tcp_ao_key *ao_key) 1052{ 1053 u8 *traffic_key = snd_other_key(ao_key); 1054 int ret; 1055 1056 ret = tcp_ao_calc_key_sk(ao_key, traffic_key, sk, 1057 ao->lisn, ao->risn, true); 1058 if (ret) 1059 return ret; 1060 1061 traffic_key = rcv_other_key(ao_key); 1062 ret = tcp_ao_calc_key_sk(ao_key, traffic_key, sk, 1063 ao->lisn, ao->risn, false); 1064 return ret; 1065} 1066 1067void tcp_ao_connect_init(struct sock *sk) 1068{ 1069 struct tcp_sock *tp = tcp_sk(sk); 1070 struct tcp_ao_info *ao_info; 1071 struct hlist_node *next; 1072 union tcp_ao_addr *addr; 1073 struct tcp_ao_key *key; 1074 int family, l3index; 1075 1076 ao_info = rcu_dereference_protected(tp->ao_info, 1077 lockdep_sock_is_held(sk)); 1078 if (!ao_info) 1079 return; 1080 1081 /* Remove all keys that don't match the peer */ 1082 family = sk->sk_family; 1083 if (family == AF_INET) 1084 addr = (union tcp_ao_addr *)&sk->sk_daddr; 1085#if IS_ENABLED(CONFIG_IPV6) 1086 else if (family == AF_INET6) 1087 addr = (union tcp_ao_addr *)&sk->sk_v6_daddr; 1088#endif 1089 else 1090 return; 1091 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), 1092 sk->sk_bound_dev_if); 1093 1094 hlist_for_each_entry_safe(key, next, &ao_info->head, node) { 1095 if (!tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1)) 1096 continue; 1097 1098 if (key == ao_info->current_key) 1099 ao_info->current_key = NULL; 1100 if (key == ao_info->rnext_key) 1101 ao_info->rnext_key = NULL; 1102 hlist_del_rcu(&key->node); 1103 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc); 1104 call_rcu(&key->rcu, tcp_ao_key_free_rcu); 1105 } 1106 1107 key = tp->af_specific->ao_lookup(sk, sk, -1, -1); 1108 if (key) { 1109 /* if current_key or rnext_key were not provided, 1110 * use the first key matching the peer 1111 */ 1112 if (!ao_info->current_key) 1113 ao_info->current_key = key; 1114 if (!ao_info->rnext_key) 1115 ao_info->rnext_key = key; 1116 tp->tcp_header_len += tcp_ao_len_aligned(key); 1117 1118 ao_info->lisn = htonl(tp->write_seq); 1119 ao_info->snd_sne = 0; 1120 } else { 1121 /* Can't happen: tcp_connect() verifies that there's 1122 * at least one tcp-ao key that matches the remote peer. 1123 */ 1124 WARN_ON_ONCE(1); 1125 rcu_assign_pointer(tp->ao_info, NULL); 1126 kfree(ao_info); 1127 } 1128} 1129 1130void tcp_ao_established(struct sock *sk) 1131{ 1132 struct tcp_ao_info *ao; 1133 struct tcp_ao_key *key; 1134 1135 ao = rcu_dereference_protected(tcp_sk(sk)->ao_info, 1136 lockdep_sock_is_held(sk)); 1137 if (!ao) 1138 return; 1139 1140 hlist_for_each_entry_rcu(key, &ao->head, node) 1141 tcp_ao_cache_traffic_keys(sk, ao, key); 1142} 1143 1144void tcp_ao_finish_connect(struct sock *sk, struct sk_buff *skb) 1145{ 1146 struct tcp_ao_info *ao; 1147 struct tcp_ao_key *key; 1148 1149 ao = rcu_dereference_protected(tcp_sk(sk)->ao_info, 1150 lockdep_sock_is_held(sk)); 1151 if (!ao) 1152 return; 1153 1154 WRITE_ONCE(ao->risn, tcp_hdr(skb)->seq); 1155 ao->rcv_sne = 0; 1156 1157 hlist_for_each_entry_rcu(key, &ao->head, node) 1158 tcp_ao_cache_traffic_keys(sk, ao, key); 1159} 1160 1161int tcp_ao_copy_all_matching(const struct sock *sk, struct sock *newsk, 1162 struct request_sock *req, struct sk_buff *skb, 1163 int family) 1164{ 1165 struct tcp_ao_key *key, *new_key, *first_key; 1166 struct tcp_ao_info *new_ao, *ao; 1167 struct hlist_node *key_head; 1168 int l3index, ret = -ENOMEM; 1169 union tcp_ao_addr *addr; 1170 bool match = false; 1171 1172 ao = rcu_dereference(tcp_sk(sk)->ao_info); 1173 if (!ao) 1174 return 0; 1175 1176 /* New socket without TCP-AO on it */ 1177 if (!tcp_rsk_used_ao(req)) 1178 return 0; 1179 1180 new_ao = tcp_ao_alloc_info(GFP_ATOMIC); 1181 if (!new_ao) 1182 return -ENOMEM; 1183 new_ao->lisn = htonl(tcp_rsk(req)->snt_isn); 1184 new_ao->risn = htonl(tcp_rsk(req)->rcv_isn); 1185 new_ao->ao_required = ao->ao_required; 1186 new_ao->accept_icmps = ao->accept_icmps; 1187 1188 if (family == AF_INET) { 1189 addr = (union tcp_ao_addr *)&newsk->sk_daddr; 1190#if IS_ENABLED(CONFIG_IPV6) 1191 } else if (family == AF_INET6) { 1192 addr = (union tcp_ao_addr *)&newsk->sk_v6_daddr; 1193#endif 1194 } else { 1195 ret = -EAFNOSUPPORT; 1196 goto free_ao; 1197 } 1198 l3index = l3mdev_master_ifindex_by_index(sock_net(newsk), 1199 newsk->sk_bound_dev_if); 1200 1201 hlist_for_each_entry_rcu(key, &ao->head, node) { 1202 if (tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1)) 1203 continue; 1204 1205 new_key = tcp_ao_copy_key(newsk, key); 1206 if (!new_key) 1207 goto free_and_exit; 1208 1209 tcp_ao_cache_traffic_keys(newsk, new_ao, new_key); 1210 tcp_ao_link_mkt(new_ao, new_key); 1211 match = true; 1212 } 1213 1214 if (!match) { 1215 /* RFC5925 (7.4.1) specifies that the TCP-AO status 1216 * of a connection is determined on the initial SYN. 1217 * At this point the connection was TCP-AO enabled, so 1218 * it can't switch to being unsigned if peer's key 1219 * disappears on the listening socket. 1220 */ 1221 ret = -EKEYREJECTED; 1222 goto free_and_exit; 1223 } 1224 1225 if (!static_key_fast_inc_not_disabled(&tcp_ao_needed.key.key)) { 1226 ret = -EUSERS; 1227 goto free_and_exit; 1228 } 1229 1230 key_head = rcu_dereference(hlist_first_rcu(&new_ao->head)); 1231 first_key = hlist_entry_safe(key_head, struct tcp_ao_key, node); 1232 1233 key = tcp_ao_established_key(new_ao, tcp_rsk(req)->ao_keyid, -1); 1234 if (key) 1235 new_ao->current_key = key; 1236 else 1237 new_ao->current_key = first_key; 1238 1239 /* set rnext_key */ 1240 key = tcp_ao_established_key(new_ao, -1, tcp_rsk(req)->ao_rcv_next); 1241 if (key) 1242 new_ao->rnext_key = key; 1243 else 1244 new_ao->rnext_key = first_key; 1245 1246 sk_gso_disable(newsk); 1247 rcu_assign_pointer(tcp_sk(newsk)->ao_info, new_ao); 1248 1249 return 0; 1250 1251free_and_exit: 1252 hlist_for_each_entry_safe(key, key_head, &new_ao->head, node) { 1253 hlist_del(&key->node); 1254 tcp_sigpool_release(key->tcp_sigpool_id); 1255 atomic_sub(tcp_ao_sizeof_key(key), &newsk->sk_omem_alloc); 1256 kfree_sensitive(key); 1257 } 1258free_ao: 1259 kfree(new_ao); 1260 return ret; 1261} 1262 1263static bool tcp_ao_can_set_current_rnext(struct sock *sk) 1264{ 1265 /* There aren't current/rnext keys on TCP_LISTEN sockets */ 1266 if (sk->sk_state == TCP_LISTEN) 1267 return false; 1268 return true; 1269} 1270 1271static int tcp_ao_verify_ipv4(struct sock *sk, struct tcp_ao_add *cmd, 1272 union tcp_ao_addr **addr) 1273{ 1274 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd->addr; 1275 struct inet_sock *inet = inet_sk(sk); 1276 1277 if (sin->sin_family != AF_INET) 1278 return -EINVAL; 1279 1280 /* Currently matching is not performed on port (or port ranges) */ 1281 if (sin->sin_port != 0) 1282 return -EINVAL; 1283 1284 /* Check prefix and trailing 0's in addr */ 1285 if (cmd->prefix != 0) { 1286 __be32 mask; 1287 1288 if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY) 1289 return -EINVAL; 1290 if (cmd->prefix > 32) 1291 return -EINVAL; 1292 1293 mask = inet_make_mask(cmd->prefix); 1294 if (sin->sin_addr.s_addr & ~mask) 1295 return -EINVAL; 1296 1297 /* Check that MKT address is consistent with socket */ 1298 if (ntohl(inet->inet_daddr) != INADDR_ANY && 1299 (inet->inet_daddr & mask) != sin->sin_addr.s_addr) 1300 return -EINVAL; 1301 } else { 1302 if (ntohl(sin->sin_addr.s_addr) != INADDR_ANY) 1303 return -EINVAL; 1304 } 1305 1306 *addr = (union tcp_ao_addr *)&sin->sin_addr; 1307 return 0; 1308} 1309 1310static int tcp_ao_parse_crypto(struct tcp_ao_add *cmd, struct tcp_ao_key *key) 1311{ 1312 unsigned int syn_tcp_option_space; 1313 bool is_kdf_aes_128_cmac = false; 1314 struct crypto_ahash *tfm; 1315 struct tcp_sigpool hp; 1316 void *tmp_key = NULL; 1317 int err; 1318 1319 /* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */ 1320 if (!strcmp("cmac(aes128)", cmd->alg_name)) { 1321 strscpy(cmd->alg_name, "cmac(aes)", sizeof(cmd->alg_name)); 1322 is_kdf_aes_128_cmac = (cmd->keylen != 16); 1323 tmp_key = kmalloc(cmd->keylen, GFP_KERNEL); 1324 if (!tmp_key) 1325 return -ENOMEM; 1326 } 1327 1328 key->maclen = cmd->maclen ?: 12; /* 12 is the default in RFC5925 */ 1329 1330 /* Check: maclen + tcp-ao header <= (MAX_TCP_OPTION_SPACE - mss 1331 * - tstamp (including sackperm) 1332 * - wscale), 1333 * see tcp_syn_options(), tcp_synack_options(), commit 33ad798c924b. 1334 * 1335 * In order to allow D-SACK with TCP-AO, the header size should be: 1336 * (MAX_TCP_OPTION_SPACE - TCPOLEN_TSTAMP_ALIGNED 1337 * - TCPOLEN_SACK_BASE_ALIGNED 1338 * - 2 * TCPOLEN_SACK_PERBLOCK) = 8 (maclen = 4), 1339 * see tcp_established_options(). 1340 * 1341 * RFC5925, 2.2: 1342 * Typical MACs are 96-128 bits (12-16 bytes), but any length 1343 * that fits in the header of the segment being authenticated 1344 * is allowed. 1345 * 1346 * RFC5925, 7.6: 1347 * TCP-AO continues to consume 16 bytes in non-SYN segments, 1348 * leaving a total of 24 bytes for other options, of which 1349 * the timestamp consumes 10. This leaves 14 bytes, of which 10 1350 * are used for a single SACK block. When two SACK blocks are used, 1351 * such as to handle D-SACK, a smaller TCP-AO MAC would be required 1352 * to make room for the additional SACK block (i.e., to leave 18 1353 * bytes for the D-SACK variant of the SACK option) [RFC2883]. 1354 * Note that D-SACK is not supportable in TCP MD5 in the presence 1355 * of timestamps, because TCP MD5���s MAC length is fixed and too 1356 * large to leave sufficient option space. 1357 */ 1358 syn_tcp_option_space = MAX_TCP_OPTION_SPACE; 1359 syn_tcp_option_space -= TCPOLEN_MSS_ALIGNED; 1360 syn_tcp_option_space -= TCPOLEN_TSTAMP_ALIGNED; 1361 syn_tcp_option_space -= TCPOLEN_WSCALE_ALIGNED; 1362 if (tcp_ao_len_aligned(key) > syn_tcp_option_space) { 1363 err = -EMSGSIZE; 1364 goto err_kfree; 1365 } 1366 1367 key->keylen = cmd->keylen; 1368 memcpy(key->key, cmd->key, cmd->keylen); 1369 1370 err = tcp_sigpool_start(key->tcp_sigpool_id, &hp); 1371 if (err) 1372 goto err_kfree; 1373 1374 tfm = crypto_ahash_reqtfm(hp.req); 1375 if (is_kdf_aes_128_cmac) { 1376 void *scratch = hp.scratch; 1377 struct scatterlist sg; 1378 1379 memcpy(tmp_key, cmd->key, cmd->keylen); 1380 sg_init_one(&sg, tmp_key, cmd->keylen); 1381 1382 /* Using zero-key of 16 bytes as described in RFC5926 */ 1383 memset(scratch, 0, 16); 1384 err = crypto_ahash_setkey(tfm, scratch, 16); 1385 if (err) 1386 goto err_pool_end; 1387 1388 err = crypto_ahash_init(hp.req); 1389 if (err) 1390 goto err_pool_end; 1391 1392 ahash_request_set_crypt(hp.req, &sg, key->key, cmd->keylen); 1393 err = crypto_ahash_update(hp.req); 1394 if (err) 1395 goto err_pool_end; 1396 1397 err |= crypto_ahash_final(hp.req); 1398 if (err) 1399 goto err_pool_end; 1400 key->keylen = 16; 1401 } 1402 1403 err = crypto_ahash_setkey(tfm, key->key, key->keylen); 1404 if (err) 1405 goto err_pool_end; 1406 1407 tcp_sigpool_end(&hp); 1408 kfree_sensitive(tmp_key); 1409 1410 if (tcp_ao_maclen(key) > key->digest_size) 1411 return -EINVAL; 1412 1413 return 0; 1414 1415err_pool_end: 1416 tcp_sigpool_end(&hp); 1417err_kfree: 1418 kfree_sensitive(tmp_key); 1419 return err; 1420} 1421 1422#if IS_ENABLED(CONFIG_IPV6) 1423static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd, 1424 union tcp_ao_addr **paddr, 1425 unsigned short int *family) 1426{ 1427 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd->addr; 1428 struct in6_addr *addr = &sin6->sin6_addr; 1429 u8 prefix = cmd->prefix; 1430 1431 if (sin6->sin6_family != AF_INET6) 1432 return -EINVAL; 1433 1434 /* Currently matching is not performed on port (or port ranges) */ 1435 if (sin6->sin6_port != 0) 1436 return -EINVAL; 1437 1438 /* Check prefix and trailing 0's in addr */ 1439 if (cmd->prefix != 0 && ipv6_addr_v4mapped(addr)) { 1440 __be32 addr4 = addr->s6_addr32[3]; 1441 __be32 mask; 1442 1443 if (prefix > 32 || ntohl(addr4) == INADDR_ANY) 1444 return -EINVAL; 1445 1446 mask = inet_make_mask(prefix); 1447 if (addr4 & ~mask) 1448 return -EINVAL; 1449 1450 /* Check that MKT address is consistent with socket */ 1451 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 1452 __be32 daddr4 = sk->sk_v6_daddr.s6_addr32[3]; 1453 1454 if (!ipv6_addr_v4mapped(&sk->sk_v6_daddr)) 1455 return -EINVAL; 1456 if ((daddr4 & mask) != addr4) 1457 return -EINVAL; 1458 } 1459 1460 *paddr = (union tcp_ao_addr *)&addr->s6_addr32[3]; 1461 *family = AF_INET; 1462 return 0; 1463 } else if (cmd->prefix != 0) { 1464 struct in6_addr pfx; 1465 1466 if (ipv6_addr_any(addr) || prefix > 128) 1467 return -EINVAL; 1468 1469 ipv6_addr_prefix(&pfx, addr, prefix); 1470 if (ipv6_addr_cmp(&pfx, addr)) 1471 return -EINVAL; 1472 1473 /* Check that MKT address is consistent with socket */ 1474 if (!ipv6_addr_any(&sk->sk_v6_daddr) && 1475 !ipv6_prefix_equal(&sk->sk_v6_daddr, addr, prefix)) 1476 1477 return -EINVAL; 1478 } else { 1479 if (!ipv6_addr_any(addr)) 1480 return -EINVAL; 1481 } 1482 1483 *paddr = (union tcp_ao_addr *)addr; 1484 return 0; 1485} 1486#else 1487static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd, 1488 union tcp_ao_addr **paddr, 1489 unsigned short int *family) 1490{ 1491 return -EOPNOTSUPP; 1492} 1493#endif 1494 1495static struct tcp_ao_info *setsockopt_ao_info(struct sock *sk) 1496{ 1497 if (sk_fullsock(sk)) { 1498 return rcu_dereference_protected(tcp_sk(sk)->ao_info, 1499 lockdep_sock_is_held(sk)); 1500 } else if (sk->sk_state == TCP_TIME_WAIT) { 1501 return rcu_dereference_protected(tcp_twsk(sk)->ao_info, 1502 lockdep_sock_is_held(sk)); 1503 } 1504 return ERR_PTR(-ESOCKTNOSUPPORT); 1505} 1506 1507static struct tcp_ao_info *getsockopt_ao_info(struct sock *sk) 1508{ 1509 if (sk_fullsock(sk)) 1510 return rcu_dereference(tcp_sk(sk)->ao_info); 1511 else if (sk->sk_state == TCP_TIME_WAIT) 1512 return rcu_dereference(tcp_twsk(sk)->ao_info); 1513 1514 return ERR_PTR(-ESOCKTNOSUPPORT); 1515} 1516 1517#define TCP_AO_KEYF_ALL (TCP_AO_KEYF_IFINDEX | TCP_AO_KEYF_EXCLUDE_OPT) 1518#define TCP_AO_GET_KEYF_VALID (TCP_AO_KEYF_IFINDEX) 1519 1520static struct tcp_ao_key *tcp_ao_key_alloc(struct sock *sk, 1521 struct tcp_ao_add *cmd) 1522{ 1523 const char *algo = cmd->alg_name; 1524 unsigned int digest_size; 1525 struct crypto_ahash *tfm; 1526 struct tcp_ao_key *key; 1527 struct tcp_sigpool hp; 1528 int err, pool_id; 1529 size_t size; 1530 1531 /* Force null-termination of alg_name */ 1532 cmd->alg_name[ARRAY_SIZE(cmd->alg_name) - 1] = '\0'; 1533 1534 /* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */ 1535 if (!strcmp("cmac(aes128)", algo)) 1536 algo = "cmac(aes)"; 1537 1538 /* Full TCP header (th->doff << 2) should fit into scratch area, 1539 * see tcp_ao_hash_header(). 1540 */ 1541 pool_id = tcp_sigpool_alloc_ahash(algo, 60); 1542 if (pool_id < 0) 1543 return ERR_PTR(pool_id); 1544 1545 err = tcp_sigpool_start(pool_id, &hp); 1546 if (err) 1547 goto err_free_pool; 1548 1549 tfm = crypto_ahash_reqtfm(hp.req); 1550 digest_size = crypto_ahash_digestsize(tfm); 1551 tcp_sigpool_end(&hp); 1552 1553 size = sizeof(struct tcp_ao_key) + (digest_size << 1); 1554 key = sock_kmalloc(sk, size, GFP_KERNEL); 1555 if (!key) { 1556 err = -ENOMEM; 1557 goto err_free_pool; 1558 } 1559 1560 key->tcp_sigpool_id = pool_id; 1561 key->digest_size = digest_size; 1562 return key; 1563 1564err_free_pool: 1565 tcp_sigpool_release(pool_id); 1566 return ERR_PTR(err); 1567} 1568 1569static int tcp_ao_add_cmd(struct sock *sk, unsigned short int family, 1570 sockptr_t optval, int optlen) 1571{ 1572 struct tcp_ao_info *ao_info; 1573 union tcp_ao_addr *addr; 1574 struct tcp_ao_key *key; 1575 struct tcp_ao_add cmd; 1576 int ret, l3index = 0; 1577 bool first = false; 1578 1579 if (optlen < sizeof(cmd)) 1580 return -EINVAL; 1581 1582 ret = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen); 1583 if (ret) 1584 return ret; 1585 1586 if (cmd.keylen > TCP_AO_MAXKEYLEN) 1587 return -EINVAL; 1588 1589 if (cmd.reserved != 0 || cmd.reserved2 != 0) 1590 return -EINVAL; 1591 1592 if (family == AF_INET) 1593 ret = tcp_ao_verify_ipv4(sk, &cmd, &addr); 1594 else 1595 ret = tcp_ao_verify_ipv6(sk, &cmd, &addr, &family); 1596 if (ret) 1597 return ret; 1598 1599 if (cmd.keyflags & ~TCP_AO_KEYF_ALL) 1600 return -EINVAL; 1601 1602 if (cmd.set_current || cmd.set_rnext) { 1603 if (!tcp_ao_can_set_current_rnext(sk)) 1604 return -EINVAL; 1605 } 1606 1607 if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX)) 1608 return -EINVAL; 1609 1610 /* For cmd.tcp_ifindex = 0 the key will apply to the default VRF */ 1611 if (cmd.keyflags & TCP_AO_KEYF_IFINDEX && cmd.ifindex) { 1612 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 1613 struct net_device *dev; 1614 1615 rcu_read_lock(); 1616 dev = dev_get_by_index_rcu(sock_net(sk), cmd.ifindex); 1617 if (dev && netif_is_l3_master(dev)) 1618 l3index = dev->ifindex; 1619 rcu_read_unlock(); 1620 1621 if (!dev || !l3index) 1622 return -EINVAL; 1623 1624 if (!bound_dev_if || bound_dev_if != cmd.ifindex) { 1625 /* tcp_ao_established_key() doesn't expect having 1626 * non peer-matching key on an established TCP-AO 1627 * connection. 1628 */ 1629 if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) 1630 return -EINVAL; 1631 } 1632 1633 /* It's still possible to bind after adding keys or even 1634 * re-bind to a different dev (with CAP_NET_RAW). 1635 * So, no reason to return error here, rather try to be 1636 * nice and warn the user. 1637 */ 1638 if (bound_dev_if && bound_dev_if != cmd.ifindex) 1639 net_warn_ratelimited("AO key ifindex %d != sk bound ifindex %d\n", 1640 cmd.ifindex, bound_dev_if); 1641 } 1642 1643 /* Don't allow keys for peers that have a matching TCP-MD5 key */ 1644 if (cmd.keyflags & TCP_AO_KEYF_IFINDEX) { 1645 /* Non-_exact version of tcp_md5_do_lookup() will 1646 * as well match keys that aren't bound to a specific VRF 1647 * (that will make them match AO key with 1648 * sysctl_tcp_l3dev_accept = 1 1649 */ 1650 if (tcp_md5_do_lookup(sk, l3index, addr, family)) 1651 return -EKEYREJECTED; 1652 } else { 1653 if (tcp_md5_do_lookup_any_l3index(sk, addr, family)) 1654 return -EKEYREJECTED; 1655 } 1656 1657 ao_info = setsockopt_ao_info(sk); 1658 if (IS_ERR(ao_info)) 1659 return PTR_ERR(ao_info); 1660 1661 if (!ao_info) { 1662 ao_info = tcp_ao_alloc_info(GFP_KERNEL); 1663 if (!ao_info) 1664 return -ENOMEM; 1665 first = true; 1666 } else { 1667 /* Check that neither RecvID nor SendID match any 1668 * existing key for the peer, RFC5925 3.1: 1669 * > The IDs of MKTs MUST NOT overlap where their 1670 * > TCP connection identifiers overlap. 1671 */ 1672 if (__tcp_ao_do_lookup(sk, l3index, addr, family, cmd.prefix, -1, cmd.rcvid)) 1673 return -EEXIST; 1674 if (__tcp_ao_do_lookup(sk, l3index, addr, family, 1675 cmd.prefix, cmd.sndid, -1)) 1676 return -EEXIST; 1677 } 1678 1679 key = tcp_ao_key_alloc(sk, &cmd); 1680 if (IS_ERR(key)) { 1681 ret = PTR_ERR(key); 1682 goto err_free_ao; 1683 } 1684 1685 INIT_HLIST_NODE(&key->node); 1686 memcpy(&key->addr, addr, (family == AF_INET) ? sizeof(struct in_addr) : 1687 sizeof(struct in6_addr)); 1688 key->prefixlen = cmd.prefix; 1689 key->family = family; 1690 key->keyflags = cmd.keyflags; 1691 key->sndid = cmd.sndid; 1692 key->rcvid = cmd.rcvid; 1693 key->l3index = l3index; 1694 atomic64_set(&key->pkt_good, 0); 1695 atomic64_set(&key->pkt_bad, 0); 1696 1697 ret = tcp_ao_parse_crypto(&cmd, key); 1698 if (ret < 0) 1699 goto err_free_sock; 1700 1701 if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) { 1702 tcp_ao_cache_traffic_keys(sk, ao_info, key); 1703 if (first) { 1704 ao_info->current_key = key; 1705 ao_info->rnext_key = key; 1706 } 1707 } 1708 1709 tcp_ao_link_mkt(ao_info, key); 1710 if (first) { 1711 if (!static_branch_inc(&tcp_ao_needed.key)) { 1712 ret = -EUSERS; 1713 goto err_free_sock; 1714 } 1715 sk_gso_disable(sk); 1716 rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info); 1717 } 1718 1719 if (cmd.set_current) 1720 WRITE_ONCE(ao_info->current_key, key); 1721 if (cmd.set_rnext) 1722 WRITE_ONCE(ao_info->rnext_key, key); 1723 return 0; 1724 1725err_free_sock: 1726 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc); 1727 tcp_sigpool_release(key->tcp_sigpool_id); 1728 kfree_sensitive(key); 1729err_free_ao: 1730 if (first) 1731 kfree(ao_info); 1732 return ret; 1733} 1734 1735static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info, 1736 bool del_async, struct tcp_ao_key *key, 1737 struct tcp_ao_key *new_current, 1738 struct tcp_ao_key *new_rnext) 1739{ 1740 int err; 1741 1742 hlist_del_rcu(&key->node); 1743 1744 /* Support for async delete on listening sockets: as they don't 1745 * need current_key/rnext_key maintaining, we don't need to check 1746 * them and we can just free all resources in RCU fashion. 1747 */ 1748 if (del_async) { 1749 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc); 1750 call_rcu(&key->rcu, tcp_ao_key_free_rcu); 1751 return 0; 1752 } 1753 1754 /* At this moment another CPU could have looked this key up 1755 * while it was unlinked from the list. Wait for RCU grace period, 1756 * after which the key is off-list and can't be looked up again; 1757 * the rx path [just before RCU came] might have used it and set it 1758 * as current_key (very unlikely). 1759 * Free the key with next RCU grace period (in case it was 1760 * current_key before tcp_ao_current_rnext() might have 1761 * changed it in forced-delete). 1762 */ 1763 synchronize_rcu(); 1764 if (new_current) 1765 WRITE_ONCE(ao_info->current_key, new_current); 1766 if (new_rnext) 1767 WRITE_ONCE(ao_info->rnext_key, new_rnext); 1768 1769 if (unlikely(READ_ONCE(ao_info->current_key) == key || 1770 READ_ONCE(ao_info->rnext_key) == key)) { 1771 err = -EBUSY; 1772 goto add_key; 1773 } 1774 1775 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc); 1776 call_rcu(&key->rcu, tcp_ao_key_free_rcu); 1777 1778 return 0; 1779add_key: 1780 hlist_add_head_rcu(&key->node, &ao_info->head); 1781 return err; 1782} 1783 1784#define TCP_AO_DEL_KEYF_ALL (TCP_AO_KEYF_IFINDEX) 1785static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family, 1786 sockptr_t optval, int optlen) 1787{ 1788 struct tcp_ao_key *key, *new_current = NULL, *new_rnext = NULL; 1789 int err, addr_len, l3index = 0; 1790 struct tcp_ao_info *ao_info; 1791 union tcp_ao_addr *addr; 1792 struct tcp_ao_del cmd; 1793 __u8 prefix; 1794 u16 port; 1795 1796 if (optlen < sizeof(cmd)) 1797 return -EINVAL; 1798 1799 err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen); 1800 if (err) 1801 return err; 1802 1803 if (cmd.reserved != 0 || cmd.reserved2 != 0) 1804 return -EINVAL; 1805 1806 if (cmd.set_current || cmd.set_rnext) { 1807 if (!tcp_ao_can_set_current_rnext(sk)) 1808 return -EINVAL; 1809 } 1810 1811 if (cmd.keyflags & ~TCP_AO_DEL_KEYF_ALL) 1812 return -EINVAL; 1813 1814 /* No sanity check for TCP_AO_KEYF_IFINDEX as if a VRF 1815 * was destroyed, there still should be a way to delete keys, 1816 * that were bound to that l3intf. So, fail late at lookup stage 1817 * if there is no key for that ifindex. 1818 */ 1819 if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX)) 1820 return -EINVAL; 1821 1822 ao_info = setsockopt_ao_info(sk); 1823 if (IS_ERR(ao_info)) 1824 return PTR_ERR(ao_info); 1825 if (!ao_info) 1826 return -ENOENT; 1827 1828 /* For sockets in TCP_CLOSED it's possible set keys that aren't 1829 * matching the future peer (address/VRF/etc), 1830 * tcp_ao_connect_init() will choose a correct matching MKT 1831 * if there's any. 1832 */ 1833 if (cmd.set_current) { 1834 new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1); 1835 if (!new_current) 1836 return -ENOENT; 1837 } 1838 if (cmd.set_rnext) { 1839 new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext); 1840 if (!new_rnext) 1841 return -ENOENT; 1842 } 1843 if (cmd.del_async && sk->sk_state != TCP_LISTEN) 1844 return -EINVAL; 1845 1846 if (family == AF_INET) { 1847 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr; 1848 1849 addr = (union tcp_ao_addr *)&sin->sin_addr; 1850 addr_len = sizeof(struct in_addr); 1851 port = ntohs(sin->sin_port); 1852 } else { 1853 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.addr; 1854 struct in6_addr *addr6 = &sin6->sin6_addr; 1855 1856 if (ipv6_addr_v4mapped(addr6)) { 1857 addr = (union tcp_ao_addr *)&addr6->s6_addr32[3]; 1858 addr_len = sizeof(struct in_addr); 1859 family = AF_INET; 1860 } else { 1861 addr = (union tcp_ao_addr *)addr6; 1862 addr_len = sizeof(struct in6_addr); 1863 } 1864 port = ntohs(sin6->sin6_port); 1865 } 1866 prefix = cmd.prefix; 1867 1868 /* Currently matching is not performed on port (or port ranges) */ 1869 if (port != 0) 1870 return -EINVAL; 1871 1872 /* We could choose random present key here for current/rnext 1873 * but that's less predictable. Let's be strict and don't 1874 * allow removing a key that's in use. RFC5925 doesn't 1875 * specify how-to coordinate key removal, but says: 1876 * "It is presumed that an MKT affecting a particular 1877 * connection cannot be destroyed during an active connection" 1878 */ 1879 hlist_for_each_entry_rcu(key, &ao_info->head, node) { 1880 if (cmd.sndid != key->sndid || 1881 cmd.rcvid != key->rcvid) 1882 continue; 1883 1884 if (family != key->family || 1885 prefix != key->prefixlen || 1886 memcmp(addr, &key->addr, addr_len)) 1887 continue; 1888 1889 if ((cmd.keyflags & TCP_AO_KEYF_IFINDEX) != 1890 (key->keyflags & TCP_AO_KEYF_IFINDEX)) 1891 continue; 1892 1893 if (key->l3index != l3index) 1894 continue; 1895 1896 if (key == new_current || key == new_rnext) 1897 continue; 1898 1899 return tcp_ao_delete_key(sk, ao_info, cmd.del_async, key, 1900 new_current, new_rnext); 1901 } 1902 return -ENOENT; 1903} 1904 1905/* cmd.ao_required makes a socket TCP-AO only. 1906 * Don't allow any md5 keys for any l3intf on the socket together with it. 1907 * Restricting it early in setsockopt() removes a check for 1908 * ao_info->ao_required on inbound tcp segment fast-path. 1909 */ 1910static int tcp_ao_required_verify(struct sock *sk) 1911{ 1912#ifdef CONFIG_TCP_MD5SIG 1913 const struct tcp_md5sig_info *md5sig; 1914 1915 if (!static_branch_unlikely(&tcp_md5_needed.key)) 1916 return 0; 1917 1918 md5sig = rcu_dereference_check(tcp_sk(sk)->md5sig_info, 1919 lockdep_sock_is_held(sk)); 1920 if (!md5sig) 1921 return 0; 1922 1923 if (rcu_dereference_check(hlist_first_rcu(&md5sig->head), 1924 lockdep_sock_is_held(sk))) 1925 return 1; 1926#endif 1927 return 0; 1928} 1929 1930static int tcp_ao_info_cmd(struct sock *sk, unsigned short int family, 1931 sockptr_t optval, int optlen) 1932{ 1933 struct tcp_ao_key *new_current = NULL, *new_rnext = NULL; 1934 struct tcp_ao_info *ao_info; 1935 struct tcp_ao_info_opt cmd; 1936 bool first = false; 1937 int err; 1938 1939 if (optlen < sizeof(cmd)) 1940 return -EINVAL; 1941 1942 err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen); 1943 if (err) 1944 return err; 1945 1946 if (cmd.set_current || cmd.set_rnext) { 1947 if (!tcp_ao_can_set_current_rnext(sk)) 1948 return -EINVAL; 1949 } 1950 1951 if (cmd.reserved != 0 || cmd.reserved2 != 0) 1952 return -EINVAL; 1953 1954 ao_info = setsockopt_ao_info(sk); 1955 if (IS_ERR(ao_info)) 1956 return PTR_ERR(ao_info); 1957 if (!ao_info) { 1958 if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) 1959 return -EINVAL; 1960 ao_info = tcp_ao_alloc_info(GFP_KERNEL); 1961 if (!ao_info) 1962 return -ENOMEM; 1963 first = true; 1964 } 1965 1966 if (cmd.ao_required && tcp_ao_required_verify(sk)) 1967 return -EKEYREJECTED; 1968 1969 /* For sockets in TCP_CLOSED it's possible set keys that aren't 1970 * matching the future peer (address/port/VRF/etc), 1971 * tcp_ao_connect_init() will choose a correct matching MKT 1972 * if there's any. 1973 */ 1974 if (cmd.set_current) { 1975 new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1); 1976 if (!new_current) { 1977 err = -ENOENT; 1978 goto out; 1979 } 1980 } 1981 if (cmd.set_rnext) { 1982 new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext); 1983 if (!new_rnext) { 1984 err = -ENOENT; 1985 goto out; 1986 } 1987 } 1988 if (cmd.set_counters) { 1989 atomic64_set(&ao_info->counters.pkt_good, cmd.pkt_good); 1990 atomic64_set(&ao_info->counters.pkt_bad, cmd.pkt_bad); 1991 atomic64_set(&ao_info->counters.key_not_found, cmd.pkt_key_not_found); 1992 atomic64_set(&ao_info->counters.ao_required, cmd.pkt_ao_required); 1993 atomic64_set(&ao_info->counters.dropped_icmp, cmd.pkt_dropped_icmp); 1994 } 1995 1996 ao_info->ao_required = cmd.ao_required; 1997 ao_info->accept_icmps = cmd.accept_icmps; 1998 if (new_current) 1999 WRITE_ONCE(ao_info->current_key, new_current); 2000 if (new_rnext) 2001 WRITE_ONCE(ao_info->rnext_key, new_rnext); 2002 if (first) { 2003 if (!static_branch_inc(&tcp_ao_needed.key)) { 2004 err = -EUSERS; 2005 goto out; 2006 } 2007 sk_gso_disable(sk); 2008 rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info); 2009 } 2010 return 0; 2011out: 2012 if (first) 2013 kfree(ao_info); 2014 return err; 2015} 2016 2017int tcp_parse_ao(struct sock *sk, int cmd, unsigned short int family, 2018 sockptr_t optval, int optlen) 2019{ 2020 if (WARN_ON_ONCE(family != AF_INET && family != AF_INET6)) 2021 return -EAFNOSUPPORT; 2022 2023 switch (cmd) { 2024 case TCP_AO_ADD_KEY: 2025 return tcp_ao_add_cmd(sk, family, optval, optlen); 2026 case TCP_AO_DEL_KEY: 2027 return tcp_ao_del_cmd(sk, family, optval, optlen); 2028 case TCP_AO_INFO: 2029 return tcp_ao_info_cmd(sk, family, optval, optlen); 2030 default: 2031 WARN_ON_ONCE(1); 2032 return -EINVAL; 2033 } 2034} 2035 2036int tcp_v4_parse_ao(struct sock *sk, int cmd, sockptr_t optval, int optlen) 2037{ 2038 return tcp_parse_ao(sk, cmd, AF_INET, optval, optlen); 2039} 2040 2041/* tcp_ao_copy_mkts_to_user(ao_info, optval, optlen) 2042 * 2043 * @ao_info: struct tcp_ao_info on the socket that 2044 * socket getsockopt(TCP_AO_GET_KEYS) is executed on 2045 * @optval: pointer to array of tcp_ao_getsockopt structures in user space. 2046 * Must be != NULL. 2047 * @optlen: pointer to size of tcp_ao_getsockopt structure. 2048 * Must be != NULL. 2049 * 2050 * Return value: 0 on success, a negative error number otherwise. 2051 * 2052 * optval points to an array of tcp_ao_getsockopt structures in user space. 2053 * optval[0] is used as both input and output to getsockopt. It determines 2054 * which keys are returned by the kernel. 2055 * optval[0].nkeys is the size of the array in user space. On return it contains 2056 * the number of keys matching the search criteria. 2057 * If tcp_ao_getsockopt::get_all is set, then all keys in the socket are 2058 * returned, otherwise only keys matching <addr, prefix, sndid, rcvid> 2059 * in optval[0] are returned. 2060 * optlen is also used as both input and output. The user provides the size 2061 * of struct tcp_ao_getsockopt in user space, and the kernel returns the size 2062 * of the structure in kernel space. 2063 * The size of struct tcp_ao_getsockopt may differ between user and kernel. 2064 * There are three cases to consider: 2065 * * If usize == ksize, then keys are copied verbatim. 2066 * * If usize < ksize, then the userspace has passed an old struct to a 2067 * newer kernel. The rest of the trailing bytes in optval[0] 2068 * (ksize - usize) are interpreted as 0 by the kernel. 2069 * * If usize > ksize, then the userspace has passed a new struct to an 2070 * older kernel. The trailing bytes unknown to the kernel (usize - ksize) 2071 * are checked to ensure they are zeroed, otherwise -E2BIG is returned. 2072 * On return the kernel fills in min(usize, ksize) in each entry of the array. 2073 * The layout of the fields in the user and kernel structures is expected to 2074 * be the same (including in the 32bit vs 64bit case). 2075 */ 2076static int tcp_ao_copy_mkts_to_user(struct tcp_ao_info *ao_info, 2077 sockptr_t optval, sockptr_t optlen) 2078{ 2079 struct tcp_ao_getsockopt opt_in, opt_out; 2080 struct tcp_ao_key *key, *current_key; 2081 bool do_address_matching = true; 2082 union tcp_ao_addr *addr = NULL; 2083 int err, l3index, user_len; 2084 unsigned int max_keys; /* maximum number of keys to copy to user */ 2085 size_t out_offset = 0; 2086 size_t bytes_to_write; /* number of bytes to write to user level */ 2087 u32 matched_keys; /* keys from ao_info matched so far */ 2088 int optlen_out; 2089 __be16 port = 0; 2090 2091 if (copy_from_sockptr(&user_len, optlen, sizeof(int))) 2092 return -EFAULT; 2093 2094 if (user_len <= 0) 2095 return -EINVAL; 2096 2097 memset(&opt_in, 0, sizeof(struct tcp_ao_getsockopt)); 2098 err = copy_struct_from_sockptr(&opt_in, sizeof(opt_in), 2099 optval, user_len); 2100 if (err < 0) 2101 return err; 2102 2103 if (opt_in.pkt_good || opt_in.pkt_bad) 2104 return -EINVAL; 2105 if (opt_in.keyflags & ~TCP_AO_GET_KEYF_VALID) 2106 return -EINVAL; 2107 if (opt_in.ifindex && !(opt_in.keyflags & TCP_AO_KEYF_IFINDEX)) 2108 return -EINVAL; 2109 2110 if (opt_in.reserved != 0) 2111 return -EINVAL; 2112 2113 max_keys = opt_in.nkeys; 2114 l3index = (opt_in.keyflags & TCP_AO_KEYF_IFINDEX) ? opt_in.ifindex : -1; 2115 2116 if (opt_in.get_all || opt_in.is_current || opt_in.is_rnext) { 2117 if (opt_in.get_all && (opt_in.is_current || opt_in.is_rnext)) 2118 return -EINVAL; 2119 do_address_matching = false; 2120 } 2121 2122 switch (opt_in.addr.ss_family) { 2123 case AF_INET: { 2124 struct sockaddr_in *sin; 2125 __be32 mask; 2126 2127 sin = (struct sockaddr_in *)&opt_in.addr; 2128 port = sin->sin_port; 2129 addr = (union tcp_ao_addr *)&sin->sin_addr; 2130 2131 if (opt_in.prefix > 32) 2132 return -EINVAL; 2133 2134 if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY && 2135 opt_in.prefix != 0) 2136 return -EINVAL; 2137 2138 mask = inet_make_mask(opt_in.prefix); 2139 if (sin->sin_addr.s_addr & ~mask) 2140 return -EINVAL; 2141 2142 break; 2143 } 2144 case AF_INET6: { 2145 struct sockaddr_in6 *sin6; 2146 struct in6_addr *addr6; 2147 2148 sin6 = (struct sockaddr_in6 *)&opt_in.addr; 2149 addr = (union tcp_ao_addr *)&sin6->sin6_addr; 2150 addr6 = &sin6->sin6_addr; 2151 port = sin6->sin6_port; 2152 2153 /* We don't have to change family and @addr here if 2154 * ipv6_addr_v4mapped() like in key adding: 2155 * tcp_ao_key_cmp() does it. Do the sanity checks though. 2156 */ 2157 if (opt_in.prefix != 0) { 2158 if (ipv6_addr_v4mapped(addr6)) { 2159 __be32 mask, addr4 = addr6->s6_addr32[3]; 2160 2161 if (opt_in.prefix > 32 || 2162 ntohl(addr4) == INADDR_ANY) 2163 return -EINVAL; 2164 mask = inet_make_mask(opt_in.prefix); 2165 if (addr4 & ~mask) 2166 return -EINVAL; 2167 } else { 2168 struct in6_addr pfx; 2169 2170 if (ipv6_addr_any(addr6) || 2171 opt_in.prefix > 128) 2172 return -EINVAL; 2173 2174 ipv6_addr_prefix(&pfx, addr6, opt_in.prefix); 2175 if (ipv6_addr_cmp(&pfx, addr6)) 2176 return -EINVAL; 2177 } 2178 } else if (!ipv6_addr_any(addr6)) { 2179 return -EINVAL; 2180 } 2181 break; 2182 } 2183 case 0: 2184 if (!do_address_matching) 2185 break; 2186 fallthrough; 2187 default: 2188 return -EAFNOSUPPORT; 2189 } 2190 2191 if (!do_address_matching) { 2192 /* We could just ignore those, but let's do stricter checks */ 2193 if (addr || port) 2194 return -EINVAL; 2195 if (opt_in.prefix || opt_in.sndid || opt_in.rcvid) 2196 return -EINVAL; 2197 } 2198 2199 bytes_to_write = min_t(int, user_len, sizeof(struct tcp_ao_getsockopt)); 2200 matched_keys = 0; 2201 /* May change in RX, while we're dumping, pre-fetch it */ 2202 current_key = READ_ONCE(ao_info->current_key); 2203 2204 hlist_for_each_entry_rcu(key, &ao_info->head, node) { 2205 if (opt_in.get_all) 2206 goto match; 2207 2208 if (opt_in.is_current || opt_in.is_rnext) { 2209 if (opt_in.is_current && key == current_key) 2210 goto match; 2211 if (opt_in.is_rnext && key == ao_info->rnext_key) 2212 goto match; 2213 continue; 2214 } 2215 2216 if (tcp_ao_key_cmp(key, l3index, addr, opt_in.prefix, 2217 opt_in.addr.ss_family, 2218 opt_in.sndid, opt_in.rcvid) != 0) 2219 continue; 2220match: 2221 matched_keys++; 2222 if (matched_keys > max_keys) 2223 continue; 2224 2225 memset(&opt_out, 0, sizeof(struct tcp_ao_getsockopt)); 2226 2227 if (key->family == AF_INET) { 2228 struct sockaddr_in *sin_out = (struct sockaddr_in *)&opt_out.addr; 2229 2230 sin_out->sin_family = key->family; 2231 sin_out->sin_port = 0; 2232 memcpy(&sin_out->sin_addr, &key->addr, sizeof(struct in_addr)); 2233 } else { 2234 struct sockaddr_in6 *sin6_out = (struct sockaddr_in6 *)&opt_out.addr; 2235 2236 sin6_out->sin6_family = key->family; 2237 sin6_out->sin6_port = 0; 2238 memcpy(&sin6_out->sin6_addr, &key->addr, sizeof(struct in6_addr)); 2239 } 2240 opt_out.sndid = key->sndid; 2241 opt_out.rcvid = key->rcvid; 2242 opt_out.prefix = key->prefixlen; 2243 opt_out.keyflags = key->keyflags; 2244 opt_out.is_current = (key == current_key); 2245 opt_out.is_rnext = (key == ao_info->rnext_key); 2246 opt_out.nkeys = 0; 2247 opt_out.maclen = key->maclen; 2248 opt_out.keylen = key->keylen; 2249 opt_out.ifindex = key->l3index; 2250 opt_out.pkt_good = atomic64_read(&key->pkt_good); 2251 opt_out.pkt_bad = atomic64_read(&key->pkt_bad); 2252 memcpy(&opt_out.key, key->key, key->keylen); 2253 tcp_sigpool_algo(key->tcp_sigpool_id, opt_out.alg_name, 64); 2254 2255 /* Copy key to user */ 2256 if (copy_to_sockptr_offset(optval, out_offset, 2257 &opt_out, bytes_to_write)) 2258 return -EFAULT; 2259 out_offset += user_len; 2260 } 2261 2262 optlen_out = (int)sizeof(struct tcp_ao_getsockopt); 2263 if (copy_to_sockptr(optlen, &optlen_out, sizeof(int))) 2264 return -EFAULT; 2265 2266 out_offset = offsetof(struct tcp_ao_getsockopt, nkeys); 2267 if (copy_to_sockptr_offset(optval, out_offset, 2268 &matched_keys, sizeof(u32))) 2269 return -EFAULT; 2270 2271 return 0; 2272} 2273 2274int tcp_ao_get_mkts(struct sock *sk, sockptr_t optval, sockptr_t optlen) 2275{ 2276 struct tcp_ao_info *ao_info; 2277 2278 ao_info = setsockopt_ao_info(sk); 2279 if (IS_ERR(ao_info)) 2280 return PTR_ERR(ao_info); 2281 if (!ao_info) 2282 return -ENOENT; 2283 2284 return tcp_ao_copy_mkts_to_user(ao_info, optval, optlen); 2285} 2286 2287int tcp_ao_get_sock_info(struct sock *sk, sockptr_t optval, sockptr_t optlen) 2288{ 2289 struct tcp_ao_info_opt out, in = {}; 2290 struct tcp_ao_key *current_key; 2291 struct tcp_ao_info *ao; 2292 int err, len; 2293 2294 if (copy_from_sockptr(&len, optlen, sizeof(int))) 2295 return -EFAULT; 2296 2297 if (len <= 0) 2298 return -EINVAL; 2299 2300 /* Copying this "in" only to check ::reserved, ::reserved2, 2301 * that may be needed to extend (struct tcp_ao_info_opt) and 2302 * what getsockopt() provides in future. 2303 */ 2304 err = copy_struct_from_sockptr(&in, sizeof(in), optval, len); 2305 if (err) 2306 return err; 2307 2308 if (in.reserved != 0 || in.reserved2 != 0) 2309 return -EINVAL; 2310 2311 ao = setsockopt_ao_info(sk); 2312 if (IS_ERR(ao)) 2313 return PTR_ERR(ao); 2314 if (!ao) 2315 return -ENOENT; 2316 2317 memset(&out, 0, sizeof(out)); 2318 out.ao_required = ao->ao_required; 2319 out.accept_icmps = ao->accept_icmps; 2320 out.pkt_good = atomic64_read(&ao->counters.pkt_good); 2321 out.pkt_bad = atomic64_read(&ao->counters.pkt_bad); 2322 out.pkt_key_not_found = atomic64_read(&ao->counters.key_not_found); 2323 out.pkt_ao_required = atomic64_read(&ao->counters.ao_required); 2324 out.pkt_dropped_icmp = atomic64_read(&ao->counters.dropped_icmp); 2325 2326 current_key = READ_ONCE(ao->current_key); 2327 if (current_key) { 2328 out.set_current = 1; 2329 out.current_key = current_key->sndid; 2330 } 2331 if (ao->rnext_key) { 2332 out.set_rnext = 1; 2333 out.rnext = ao->rnext_key->rcvid; 2334 } 2335 2336 if (copy_to_sockptr(optval, &out, min_t(int, len, sizeof(out)))) 2337 return -EFAULT; 2338 2339 return 0; 2340} 2341 2342int tcp_ao_set_repair(struct sock *sk, sockptr_t optval, unsigned int optlen) 2343{ 2344 struct tcp_sock *tp = tcp_sk(sk); 2345 struct tcp_ao_repair cmd; 2346 struct tcp_ao_key *key; 2347 struct tcp_ao_info *ao; 2348 int err; 2349 2350 if (optlen < sizeof(cmd)) 2351 return -EINVAL; 2352 2353 err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen); 2354 if (err) 2355 return err; 2356 2357 if (!tp->repair) 2358 return -EPERM; 2359 2360 ao = setsockopt_ao_info(sk); 2361 if (IS_ERR(ao)) 2362 return PTR_ERR(ao); 2363 if (!ao) 2364 return -ENOENT; 2365 2366 WRITE_ONCE(ao->lisn, cmd.snt_isn); 2367 WRITE_ONCE(ao->risn, cmd.rcv_isn); 2368 WRITE_ONCE(ao->snd_sne, cmd.snd_sne); 2369 WRITE_ONCE(ao->rcv_sne, cmd.rcv_sne); 2370 2371 hlist_for_each_entry_rcu(key, &ao->head, node) 2372 tcp_ao_cache_traffic_keys(sk, ao, key); 2373 2374 return 0; 2375} 2376 2377int tcp_ao_get_repair(struct sock *sk, sockptr_t optval, sockptr_t optlen) 2378{ 2379 struct tcp_sock *tp = tcp_sk(sk); 2380 struct tcp_ao_repair opt; 2381 struct tcp_ao_info *ao; 2382 int len; 2383 2384 if (copy_from_sockptr(&len, optlen, sizeof(int))) 2385 return -EFAULT; 2386 2387 if (len <= 0) 2388 return -EINVAL; 2389 2390 if (!tp->repair) 2391 return -EPERM; 2392 2393 rcu_read_lock(); 2394 ao = getsockopt_ao_info(sk); 2395 if (IS_ERR_OR_NULL(ao)) { 2396 rcu_read_unlock(); 2397 return ao ? PTR_ERR(ao) : -ENOENT; 2398 } 2399 2400 opt.snt_isn = ao->lisn; 2401 opt.rcv_isn = ao->risn; 2402 opt.snd_sne = READ_ONCE(ao->snd_sne); 2403 opt.rcv_sne = READ_ONCE(ao->rcv_sne); 2404 rcu_read_unlock(); 2405 2406 if (copy_to_sockptr(optval, &opt, min_t(int, len, sizeof(opt)))) 2407 return -EFAULT; 2408 return 0; 2409} 2410