Loading...
Note: File does not exist in v5.9.
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 int err, len = th->doff << 2;
513 struct scatterlist sg;
514 u8 *hdr = hp->scratch;
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 union tcp_ao_addr *addr;
1072 struct tcp_ao_key *key;
1073 int family, l3index;
1074
1075 ao_info = rcu_dereference_protected(tp->ao_info,
1076 lockdep_sock_is_held(sk));
1077 if (!ao_info)
1078 return;
1079
1080 /* Remove all keys that don't match the peer */
1081 family = sk->sk_family;
1082 if (family == AF_INET)
1083 addr = (union tcp_ao_addr *)&sk->sk_daddr;
1084#if IS_ENABLED(CONFIG_IPV6)
1085 else if (family == AF_INET6)
1086 addr = (union tcp_ao_addr *)&sk->sk_v6_daddr;
1087#endif
1088 else
1089 return;
1090 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1091 sk->sk_bound_dev_if);
1092
1093 hlist_for_each_entry_rcu(key, &ao_info->head, node) {
1094 if (!tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1))
1095 continue;
1096
1097 if (key == ao_info->current_key)
1098 ao_info->current_key = NULL;
1099 if (key == ao_info->rnext_key)
1100 ao_info->rnext_key = NULL;
1101 hlist_del_rcu(&key->node);
1102 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1103 call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1104 }
1105
1106 key = tp->af_specific->ao_lookup(sk, sk, -1, -1);
1107 if (key) {
1108 /* if current_key or rnext_key were not provided,
1109 * use the first key matching the peer
1110 */
1111 if (!ao_info->current_key)
1112 ao_info->current_key = key;
1113 if (!ao_info->rnext_key)
1114 ao_info->rnext_key = key;
1115 tp->tcp_header_len += tcp_ao_len_aligned(key);
1116
1117 ao_info->lisn = htonl(tp->write_seq);
1118 ao_info->snd_sne = 0;
1119 } else {
1120 /* Can't happen: tcp_connect() verifies that there's
1121 * at least one tcp-ao key that matches the remote peer.
1122 */
1123 WARN_ON_ONCE(1);
1124 rcu_assign_pointer(tp->ao_info, NULL);
1125 kfree(ao_info);
1126 }
1127}
1128
1129void tcp_ao_established(struct sock *sk)
1130{
1131 struct tcp_ao_info *ao;
1132 struct tcp_ao_key *key;
1133
1134 ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
1135 lockdep_sock_is_held(sk));
1136 if (!ao)
1137 return;
1138
1139 hlist_for_each_entry_rcu(key, &ao->head, node)
1140 tcp_ao_cache_traffic_keys(sk, ao, key);
1141}
1142
1143void tcp_ao_finish_connect(struct sock *sk, struct sk_buff *skb)
1144{
1145 struct tcp_ao_info *ao;
1146 struct tcp_ao_key *key;
1147
1148 ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
1149 lockdep_sock_is_held(sk));
1150 if (!ao)
1151 return;
1152
1153 WRITE_ONCE(ao->risn, tcp_hdr(skb)->seq);
1154 ao->rcv_sne = 0;
1155
1156 hlist_for_each_entry_rcu(key, &ao->head, node)
1157 tcp_ao_cache_traffic_keys(sk, ao, key);
1158}
1159
1160int tcp_ao_copy_all_matching(const struct sock *sk, struct sock *newsk,
1161 struct request_sock *req, struct sk_buff *skb,
1162 int family)
1163{
1164 struct tcp_ao_key *key, *new_key, *first_key;
1165 struct tcp_ao_info *new_ao, *ao;
1166 struct hlist_node *key_head;
1167 int l3index, ret = -ENOMEM;
1168 union tcp_ao_addr *addr;
1169 bool match = false;
1170
1171 ao = rcu_dereference(tcp_sk(sk)->ao_info);
1172 if (!ao)
1173 return 0;
1174
1175 /* New socket without TCP-AO on it */
1176 if (!tcp_rsk_used_ao(req))
1177 return 0;
1178
1179 new_ao = tcp_ao_alloc_info(GFP_ATOMIC);
1180 if (!new_ao)
1181 return -ENOMEM;
1182 new_ao->lisn = htonl(tcp_rsk(req)->snt_isn);
1183 new_ao->risn = htonl(tcp_rsk(req)->rcv_isn);
1184 new_ao->ao_required = ao->ao_required;
1185 new_ao->accept_icmps = ao->accept_icmps;
1186
1187 if (family == AF_INET) {
1188 addr = (union tcp_ao_addr *)&newsk->sk_daddr;
1189#if IS_ENABLED(CONFIG_IPV6)
1190 } else if (family == AF_INET6) {
1191 addr = (union tcp_ao_addr *)&newsk->sk_v6_daddr;
1192#endif
1193 } else {
1194 ret = -EAFNOSUPPORT;
1195 goto free_ao;
1196 }
1197 l3index = l3mdev_master_ifindex_by_index(sock_net(newsk),
1198 newsk->sk_bound_dev_if);
1199
1200 hlist_for_each_entry_rcu(key, &ao->head, node) {
1201 if (tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1))
1202 continue;
1203
1204 new_key = tcp_ao_copy_key(newsk, key);
1205 if (!new_key)
1206 goto free_and_exit;
1207
1208 tcp_ao_cache_traffic_keys(newsk, new_ao, new_key);
1209 tcp_ao_link_mkt(new_ao, new_key);
1210 match = true;
1211 }
1212
1213 if (!match) {
1214 /* RFC5925 (7.4.1) specifies that the TCP-AO status
1215 * of a connection is determined on the initial SYN.
1216 * At this point the connection was TCP-AO enabled, so
1217 * it can't switch to being unsigned if peer's key
1218 * disappears on the listening socket.
1219 */
1220 ret = -EKEYREJECTED;
1221 goto free_and_exit;
1222 }
1223
1224 if (!static_key_fast_inc_not_disabled(&tcp_ao_needed.key.key)) {
1225 ret = -EUSERS;
1226 goto free_and_exit;
1227 }
1228
1229 key_head = rcu_dereference(hlist_first_rcu(&new_ao->head));
1230 first_key = hlist_entry_safe(key_head, struct tcp_ao_key, node);
1231
1232 key = tcp_ao_established_key(new_ao, tcp_rsk(req)->ao_keyid, -1);
1233 if (key)
1234 new_ao->current_key = key;
1235 else
1236 new_ao->current_key = first_key;
1237
1238 /* set rnext_key */
1239 key = tcp_ao_established_key(new_ao, -1, tcp_rsk(req)->ao_rcv_next);
1240 if (key)
1241 new_ao->rnext_key = key;
1242 else
1243 new_ao->rnext_key = first_key;
1244
1245 sk_gso_disable(newsk);
1246 rcu_assign_pointer(tcp_sk(newsk)->ao_info, new_ao);
1247
1248 return 0;
1249
1250free_and_exit:
1251 hlist_for_each_entry_safe(key, key_head, &new_ao->head, node) {
1252 hlist_del(&key->node);
1253 tcp_sigpool_release(key->tcp_sigpool_id);
1254 atomic_sub(tcp_ao_sizeof_key(key), &newsk->sk_omem_alloc);
1255 kfree_sensitive(key);
1256 }
1257free_ao:
1258 kfree(new_ao);
1259 return ret;
1260}
1261
1262static bool tcp_ao_can_set_current_rnext(struct sock *sk)
1263{
1264 /* There aren't current/rnext keys on TCP_LISTEN sockets */
1265 if (sk->sk_state == TCP_LISTEN)
1266 return false;
1267 return true;
1268}
1269
1270static int tcp_ao_verify_ipv4(struct sock *sk, struct tcp_ao_add *cmd,
1271 union tcp_ao_addr **addr)
1272{
1273 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd->addr;
1274 struct inet_sock *inet = inet_sk(sk);
1275
1276 if (sin->sin_family != AF_INET)
1277 return -EINVAL;
1278
1279 /* Currently matching is not performed on port (or port ranges) */
1280 if (sin->sin_port != 0)
1281 return -EINVAL;
1282
1283 /* Check prefix and trailing 0's in addr */
1284 if (cmd->prefix != 0) {
1285 __be32 mask;
1286
1287 if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY)
1288 return -EINVAL;
1289 if (cmd->prefix > 32)
1290 return -EINVAL;
1291
1292 mask = inet_make_mask(cmd->prefix);
1293 if (sin->sin_addr.s_addr & ~mask)
1294 return -EINVAL;
1295
1296 /* Check that MKT address is consistent with socket */
1297 if (ntohl(inet->inet_daddr) != INADDR_ANY &&
1298 (inet->inet_daddr & mask) != sin->sin_addr.s_addr)
1299 return -EINVAL;
1300 } else {
1301 if (ntohl(sin->sin_addr.s_addr) != INADDR_ANY)
1302 return -EINVAL;
1303 }
1304
1305 *addr = (union tcp_ao_addr *)&sin->sin_addr;
1306 return 0;
1307}
1308
1309static int tcp_ao_parse_crypto(struct tcp_ao_add *cmd, struct tcp_ao_key *key)
1310{
1311 unsigned int syn_tcp_option_space;
1312 bool is_kdf_aes_128_cmac = false;
1313 struct crypto_ahash *tfm;
1314 struct tcp_sigpool hp;
1315 void *tmp_key = NULL;
1316 int err;
1317
1318 /* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1319 if (!strcmp("cmac(aes128)", cmd->alg_name)) {
1320 strscpy(cmd->alg_name, "cmac(aes)", sizeof(cmd->alg_name));
1321 is_kdf_aes_128_cmac = (cmd->keylen != 16);
1322 tmp_key = kmalloc(cmd->keylen, GFP_KERNEL);
1323 if (!tmp_key)
1324 return -ENOMEM;
1325 }
1326
1327 key->maclen = cmd->maclen ?: 12; /* 12 is the default in RFC5925 */
1328
1329 /* Check: maclen + tcp-ao header <= (MAX_TCP_OPTION_SPACE - mss
1330 * - tstamp (including sackperm)
1331 * - wscale),
1332 * see tcp_syn_options(), tcp_synack_options(), commit 33ad798c924b.
1333 *
1334 * In order to allow D-SACK with TCP-AO, the header size should be:
1335 * (MAX_TCP_OPTION_SPACE - TCPOLEN_TSTAMP_ALIGNED
1336 * - TCPOLEN_SACK_BASE_ALIGNED
1337 * - 2 * TCPOLEN_SACK_PERBLOCK) = 8 (maclen = 4),
1338 * see tcp_established_options().
1339 *
1340 * RFC5925, 2.2:
1341 * Typical MACs are 96-128 bits (12-16 bytes), but any length
1342 * that fits in the header of the segment being authenticated
1343 * is allowed.
1344 *
1345 * RFC5925, 7.6:
1346 * TCP-AO continues to consume 16 bytes in non-SYN segments,
1347 * leaving a total of 24 bytes for other options, of which
1348 * the timestamp consumes 10. This leaves 14 bytes, of which 10
1349 * are used for a single SACK block. When two SACK blocks are used,
1350 * such as to handle D-SACK, a smaller TCP-AO MAC would be required
1351 * to make room for the additional SACK block (i.e., to leave 18
1352 * bytes for the D-SACK variant of the SACK option) [RFC2883].
1353 * Note that D-SACK is not supportable in TCP MD5 in the presence
1354 * of timestamps, because TCP MD5’s MAC length is fixed and too
1355 * large to leave sufficient option space.
1356 */
1357 syn_tcp_option_space = MAX_TCP_OPTION_SPACE;
1358 syn_tcp_option_space -= TCPOLEN_MSS_ALIGNED;
1359 syn_tcp_option_space -= TCPOLEN_TSTAMP_ALIGNED;
1360 syn_tcp_option_space -= TCPOLEN_WSCALE_ALIGNED;
1361 if (tcp_ao_len_aligned(key) > syn_tcp_option_space) {
1362 err = -EMSGSIZE;
1363 goto err_kfree;
1364 }
1365
1366 key->keylen = cmd->keylen;
1367 memcpy(key->key, cmd->key, cmd->keylen);
1368
1369 err = tcp_sigpool_start(key->tcp_sigpool_id, &hp);
1370 if (err)
1371 goto err_kfree;
1372
1373 tfm = crypto_ahash_reqtfm(hp.req);
1374 if (is_kdf_aes_128_cmac) {
1375 void *scratch = hp.scratch;
1376 struct scatterlist sg;
1377
1378 memcpy(tmp_key, cmd->key, cmd->keylen);
1379 sg_init_one(&sg, tmp_key, cmd->keylen);
1380
1381 /* Using zero-key of 16 bytes as described in RFC5926 */
1382 memset(scratch, 0, 16);
1383 err = crypto_ahash_setkey(tfm, scratch, 16);
1384 if (err)
1385 goto err_pool_end;
1386
1387 err = crypto_ahash_init(hp.req);
1388 if (err)
1389 goto err_pool_end;
1390
1391 ahash_request_set_crypt(hp.req, &sg, key->key, cmd->keylen);
1392 err = crypto_ahash_update(hp.req);
1393 if (err)
1394 goto err_pool_end;
1395
1396 err |= crypto_ahash_final(hp.req);
1397 if (err)
1398 goto err_pool_end;
1399 key->keylen = 16;
1400 }
1401
1402 err = crypto_ahash_setkey(tfm, key->key, key->keylen);
1403 if (err)
1404 goto err_pool_end;
1405
1406 tcp_sigpool_end(&hp);
1407 kfree_sensitive(tmp_key);
1408
1409 if (tcp_ao_maclen(key) > key->digest_size)
1410 return -EINVAL;
1411
1412 return 0;
1413
1414err_pool_end:
1415 tcp_sigpool_end(&hp);
1416err_kfree:
1417 kfree_sensitive(tmp_key);
1418 return err;
1419}
1420
1421#if IS_ENABLED(CONFIG_IPV6)
1422static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1423 union tcp_ao_addr **paddr,
1424 unsigned short int *family)
1425{
1426 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd->addr;
1427 struct in6_addr *addr = &sin6->sin6_addr;
1428 u8 prefix = cmd->prefix;
1429
1430 if (sin6->sin6_family != AF_INET6)
1431 return -EINVAL;
1432
1433 /* Currently matching is not performed on port (or port ranges) */
1434 if (sin6->sin6_port != 0)
1435 return -EINVAL;
1436
1437 /* Check prefix and trailing 0's in addr */
1438 if (cmd->prefix != 0 && ipv6_addr_v4mapped(addr)) {
1439 __be32 addr4 = addr->s6_addr32[3];
1440 __be32 mask;
1441
1442 if (prefix > 32 || ntohl(addr4) == INADDR_ANY)
1443 return -EINVAL;
1444
1445 mask = inet_make_mask(prefix);
1446 if (addr4 & ~mask)
1447 return -EINVAL;
1448
1449 /* Check that MKT address is consistent with socket */
1450 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
1451 __be32 daddr4 = sk->sk_v6_daddr.s6_addr32[3];
1452
1453 if (!ipv6_addr_v4mapped(&sk->sk_v6_daddr))
1454 return -EINVAL;
1455 if ((daddr4 & mask) != addr4)
1456 return -EINVAL;
1457 }
1458
1459 *paddr = (union tcp_ao_addr *)&addr->s6_addr32[3];
1460 *family = AF_INET;
1461 return 0;
1462 } else if (cmd->prefix != 0) {
1463 struct in6_addr pfx;
1464
1465 if (ipv6_addr_any(addr) || prefix > 128)
1466 return -EINVAL;
1467
1468 ipv6_addr_prefix(&pfx, addr, prefix);
1469 if (ipv6_addr_cmp(&pfx, addr))
1470 return -EINVAL;
1471
1472 /* Check that MKT address is consistent with socket */
1473 if (!ipv6_addr_any(&sk->sk_v6_daddr) &&
1474 !ipv6_prefix_equal(&sk->sk_v6_daddr, addr, prefix))
1475
1476 return -EINVAL;
1477 } else {
1478 if (!ipv6_addr_any(addr))
1479 return -EINVAL;
1480 }
1481
1482 *paddr = (union tcp_ao_addr *)addr;
1483 return 0;
1484}
1485#else
1486static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1487 union tcp_ao_addr **paddr,
1488 unsigned short int *family)
1489{
1490 return -EOPNOTSUPP;
1491}
1492#endif
1493
1494static struct tcp_ao_info *setsockopt_ao_info(struct sock *sk)
1495{
1496 if (sk_fullsock(sk)) {
1497 return rcu_dereference_protected(tcp_sk(sk)->ao_info,
1498 lockdep_sock_is_held(sk));
1499 } else if (sk->sk_state == TCP_TIME_WAIT) {
1500 return rcu_dereference_protected(tcp_twsk(sk)->ao_info,
1501 lockdep_sock_is_held(sk));
1502 }
1503 return ERR_PTR(-ESOCKTNOSUPPORT);
1504}
1505
1506static struct tcp_ao_info *getsockopt_ao_info(struct sock *sk)
1507{
1508 if (sk_fullsock(sk))
1509 return rcu_dereference(tcp_sk(sk)->ao_info);
1510 else if (sk->sk_state == TCP_TIME_WAIT)
1511 return rcu_dereference(tcp_twsk(sk)->ao_info);
1512
1513 return ERR_PTR(-ESOCKTNOSUPPORT);
1514}
1515
1516#define TCP_AO_KEYF_ALL (TCP_AO_KEYF_IFINDEX | TCP_AO_KEYF_EXCLUDE_OPT)
1517#define TCP_AO_GET_KEYF_VALID (TCP_AO_KEYF_IFINDEX)
1518
1519static struct tcp_ao_key *tcp_ao_key_alloc(struct sock *sk,
1520 struct tcp_ao_add *cmd)
1521{
1522 const char *algo = cmd->alg_name;
1523 unsigned int digest_size;
1524 struct crypto_ahash *tfm;
1525 struct tcp_ao_key *key;
1526 struct tcp_sigpool hp;
1527 int err, pool_id;
1528 size_t size;
1529
1530 /* Force null-termination of alg_name */
1531 cmd->alg_name[ARRAY_SIZE(cmd->alg_name) - 1] = '\0';
1532
1533 /* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1534 if (!strcmp("cmac(aes128)", algo))
1535 algo = "cmac(aes)";
1536
1537 /* Full TCP header (th->doff << 2) should fit into scratch area,
1538 * see tcp_ao_hash_header().
1539 */
1540 pool_id = tcp_sigpool_alloc_ahash(algo, 60);
1541 if (pool_id < 0)
1542 return ERR_PTR(pool_id);
1543
1544 err = tcp_sigpool_start(pool_id, &hp);
1545 if (err)
1546 goto err_free_pool;
1547
1548 tfm = crypto_ahash_reqtfm(hp.req);
1549 digest_size = crypto_ahash_digestsize(tfm);
1550 tcp_sigpool_end(&hp);
1551
1552 size = sizeof(struct tcp_ao_key) + (digest_size << 1);
1553 key = sock_kmalloc(sk, size, GFP_KERNEL);
1554 if (!key) {
1555 err = -ENOMEM;
1556 goto err_free_pool;
1557 }
1558
1559 key->tcp_sigpool_id = pool_id;
1560 key->digest_size = digest_size;
1561 return key;
1562
1563err_free_pool:
1564 tcp_sigpool_release(pool_id);
1565 return ERR_PTR(err);
1566}
1567
1568static int tcp_ao_add_cmd(struct sock *sk, unsigned short int family,
1569 sockptr_t optval, int optlen)
1570{
1571 struct tcp_ao_info *ao_info;
1572 union tcp_ao_addr *addr;
1573 struct tcp_ao_key *key;
1574 struct tcp_ao_add cmd;
1575 int ret, l3index = 0;
1576 bool first = false;
1577
1578 if (optlen < sizeof(cmd))
1579 return -EINVAL;
1580
1581 ret = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1582 if (ret)
1583 return ret;
1584
1585 if (cmd.keylen > TCP_AO_MAXKEYLEN)
1586 return -EINVAL;
1587
1588 if (cmd.reserved != 0 || cmd.reserved2 != 0)
1589 return -EINVAL;
1590
1591 if (family == AF_INET)
1592 ret = tcp_ao_verify_ipv4(sk, &cmd, &addr);
1593 else
1594 ret = tcp_ao_verify_ipv6(sk, &cmd, &addr, &family);
1595 if (ret)
1596 return ret;
1597
1598 if (cmd.keyflags & ~TCP_AO_KEYF_ALL)
1599 return -EINVAL;
1600
1601 if (cmd.set_current || cmd.set_rnext) {
1602 if (!tcp_ao_can_set_current_rnext(sk))
1603 return -EINVAL;
1604 }
1605
1606 if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1607 return -EINVAL;
1608
1609 /* For cmd.tcp_ifindex = 0 the key will apply to the default VRF */
1610 if (cmd.keyflags & TCP_AO_KEYF_IFINDEX && cmd.ifindex) {
1611 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
1612 struct net_device *dev;
1613
1614 rcu_read_lock();
1615 dev = dev_get_by_index_rcu(sock_net(sk), cmd.ifindex);
1616 if (dev && netif_is_l3_master(dev))
1617 l3index = dev->ifindex;
1618 rcu_read_unlock();
1619
1620 if (!dev || !l3index)
1621 return -EINVAL;
1622
1623 if (!bound_dev_if || bound_dev_if != cmd.ifindex) {
1624 /* tcp_ao_established_key() doesn't expect having
1625 * non peer-matching key on an established TCP-AO
1626 * connection.
1627 */
1628 if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1629 return -EINVAL;
1630 }
1631
1632 /* It's still possible to bind after adding keys or even
1633 * re-bind to a different dev (with CAP_NET_RAW).
1634 * So, no reason to return error here, rather try to be
1635 * nice and warn the user.
1636 */
1637 if (bound_dev_if && bound_dev_if != cmd.ifindex)
1638 net_warn_ratelimited("AO key ifindex %d != sk bound ifindex %d\n",
1639 cmd.ifindex, bound_dev_if);
1640 }
1641
1642 /* Don't allow keys for peers that have a matching TCP-MD5 key */
1643 if (cmd.keyflags & TCP_AO_KEYF_IFINDEX) {
1644 /* Non-_exact version of tcp_md5_do_lookup() will
1645 * as well match keys that aren't bound to a specific VRF
1646 * (that will make them match AO key with
1647 * sysctl_tcp_l3dev_accept = 1
1648 */
1649 if (tcp_md5_do_lookup(sk, l3index, addr, family))
1650 return -EKEYREJECTED;
1651 } else {
1652 if (tcp_md5_do_lookup_any_l3index(sk, addr, family))
1653 return -EKEYREJECTED;
1654 }
1655
1656 ao_info = setsockopt_ao_info(sk);
1657 if (IS_ERR(ao_info))
1658 return PTR_ERR(ao_info);
1659
1660 if (!ao_info) {
1661 ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1662 if (!ao_info)
1663 return -ENOMEM;
1664 first = true;
1665 } else {
1666 /* Check that neither RecvID nor SendID match any
1667 * existing key for the peer, RFC5925 3.1:
1668 * > The IDs of MKTs MUST NOT overlap where their
1669 * > TCP connection identifiers overlap.
1670 */
1671 if (__tcp_ao_do_lookup(sk, l3index, addr, family, cmd.prefix, -1, cmd.rcvid))
1672 return -EEXIST;
1673 if (__tcp_ao_do_lookup(sk, l3index, addr, family,
1674 cmd.prefix, cmd.sndid, -1))
1675 return -EEXIST;
1676 }
1677
1678 key = tcp_ao_key_alloc(sk, &cmd);
1679 if (IS_ERR(key)) {
1680 ret = PTR_ERR(key);
1681 goto err_free_ao;
1682 }
1683
1684 INIT_HLIST_NODE(&key->node);
1685 memcpy(&key->addr, addr, (family == AF_INET) ? sizeof(struct in_addr) :
1686 sizeof(struct in6_addr));
1687 key->prefixlen = cmd.prefix;
1688 key->family = family;
1689 key->keyflags = cmd.keyflags;
1690 key->sndid = cmd.sndid;
1691 key->rcvid = cmd.rcvid;
1692 key->l3index = l3index;
1693 atomic64_set(&key->pkt_good, 0);
1694 atomic64_set(&key->pkt_bad, 0);
1695
1696 ret = tcp_ao_parse_crypto(&cmd, key);
1697 if (ret < 0)
1698 goto err_free_sock;
1699
1700 if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) {
1701 tcp_ao_cache_traffic_keys(sk, ao_info, key);
1702 if (first) {
1703 ao_info->current_key = key;
1704 ao_info->rnext_key = key;
1705 }
1706 }
1707
1708 tcp_ao_link_mkt(ao_info, key);
1709 if (first) {
1710 if (!static_branch_inc(&tcp_ao_needed.key)) {
1711 ret = -EUSERS;
1712 goto err_free_sock;
1713 }
1714 sk_gso_disable(sk);
1715 rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
1716 }
1717
1718 if (cmd.set_current)
1719 WRITE_ONCE(ao_info->current_key, key);
1720 if (cmd.set_rnext)
1721 WRITE_ONCE(ao_info->rnext_key, key);
1722 return 0;
1723
1724err_free_sock:
1725 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1726 tcp_sigpool_release(key->tcp_sigpool_id);
1727 kfree_sensitive(key);
1728err_free_ao:
1729 if (first)
1730 kfree(ao_info);
1731 return ret;
1732}
1733
1734static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info,
1735 bool del_async, struct tcp_ao_key *key,
1736 struct tcp_ao_key *new_current,
1737 struct tcp_ao_key *new_rnext)
1738{
1739 int err;
1740
1741 hlist_del_rcu(&key->node);
1742
1743 /* Support for async delete on listening sockets: as they don't
1744 * need current_key/rnext_key maintaining, we don't need to check
1745 * them and we can just free all resources in RCU fashion.
1746 */
1747 if (del_async) {
1748 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1749 call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1750 return 0;
1751 }
1752
1753 /* At this moment another CPU could have looked this key up
1754 * while it was unlinked from the list. Wait for RCU grace period,
1755 * after which the key is off-list and can't be looked up again;
1756 * the rx path [just before RCU came] might have used it and set it
1757 * as current_key (very unlikely).
1758 * Free the key with next RCU grace period (in case it was
1759 * current_key before tcp_ao_current_rnext() might have
1760 * changed it in forced-delete).
1761 */
1762 synchronize_rcu();
1763 if (new_current)
1764 WRITE_ONCE(ao_info->current_key, new_current);
1765 if (new_rnext)
1766 WRITE_ONCE(ao_info->rnext_key, new_rnext);
1767
1768 if (unlikely(READ_ONCE(ao_info->current_key) == key ||
1769 READ_ONCE(ao_info->rnext_key) == key)) {
1770 err = -EBUSY;
1771 goto add_key;
1772 }
1773
1774 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1775 call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1776
1777 return 0;
1778add_key:
1779 hlist_add_head_rcu(&key->node, &ao_info->head);
1780 return err;
1781}
1782
1783#define TCP_AO_DEL_KEYF_ALL (TCP_AO_KEYF_IFINDEX)
1784static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family,
1785 sockptr_t optval, int optlen)
1786{
1787 struct tcp_ao_key *key, *new_current = NULL, *new_rnext = NULL;
1788 int err, addr_len, l3index = 0;
1789 struct tcp_ao_info *ao_info;
1790 union tcp_ao_addr *addr;
1791 struct tcp_ao_del cmd;
1792 __u8 prefix;
1793 u16 port;
1794
1795 if (optlen < sizeof(cmd))
1796 return -EINVAL;
1797
1798 err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1799 if (err)
1800 return err;
1801
1802 if (cmd.reserved != 0 || cmd.reserved2 != 0)
1803 return -EINVAL;
1804
1805 if (cmd.set_current || cmd.set_rnext) {
1806 if (!tcp_ao_can_set_current_rnext(sk))
1807 return -EINVAL;
1808 }
1809
1810 if (cmd.keyflags & ~TCP_AO_DEL_KEYF_ALL)
1811 return -EINVAL;
1812
1813 /* No sanity check for TCP_AO_KEYF_IFINDEX as if a VRF
1814 * was destroyed, there still should be a way to delete keys,
1815 * that were bound to that l3intf. So, fail late at lookup stage
1816 * if there is no key for that ifindex.
1817 */
1818 if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1819 return -EINVAL;
1820
1821 ao_info = setsockopt_ao_info(sk);
1822 if (IS_ERR(ao_info))
1823 return PTR_ERR(ao_info);
1824 if (!ao_info)
1825 return -ENOENT;
1826
1827 /* For sockets in TCP_CLOSED it's possible set keys that aren't
1828 * matching the future peer (address/VRF/etc),
1829 * tcp_ao_connect_init() will choose a correct matching MKT
1830 * if there's any.
1831 */
1832 if (cmd.set_current) {
1833 new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1);
1834 if (!new_current)
1835 return -ENOENT;
1836 }
1837 if (cmd.set_rnext) {
1838 new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext);
1839 if (!new_rnext)
1840 return -ENOENT;
1841 }
1842 if (cmd.del_async && sk->sk_state != TCP_LISTEN)
1843 return -EINVAL;
1844
1845 if (family == AF_INET) {
1846 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr;
1847
1848 addr = (union tcp_ao_addr *)&sin->sin_addr;
1849 addr_len = sizeof(struct in_addr);
1850 port = ntohs(sin->sin_port);
1851 } else {
1852 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.addr;
1853 struct in6_addr *addr6 = &sin6->sin6_addr;
1854
1855 if (ipv6_addr_v4mapped(addr6)) {
1856 addr = (union tcp_ao_addr *)&addr6->s6_addr32[3];
1857 addr_len = sizeof(struct in_addr);
1858 family = AF_INET;
1859 } else {
1860 addr = (union tcp_ao_addr *)addr6;
1861 addr_len = sizeof(struct in6_addr);
1862 }
1863 port = ntohs(sin6->sin6_port);
1864 }
1865 prefix = cmd.prefix;
1866
1867 /* Currently matching is not performed on port (or port ranges) */
1868 if (port != 0)
1869 return -EINVAL;
1870
1871 /* We could choose random present key here for current/rnext
1872 * but that's less predictable. Let's be strict and don't
1873 * allow removing a key that's in use. RFC5925 doesn't
1874 * specify how-to coordinate key removal, but says:
1875 * "It is presumed that an MKT affecting a particular
1876 * connection cannot be destroyed during an active connection"
1877 */
1878 hlist_for_each_entry_rcu(key, &ao_info->head, node) {
1879 if (cmd.sndid != key->sndid ||
1880 cmd.rcvid != key->rcvid)
1881 continue;
1882
1883 if (family != key->family ||
1884 prefix != key->prefixlen ||
1885 memcmp(addr, &key->addr, addr_len))
1886 continue;
1887
1888 if ((cmd.keyflags & TCP_AO_KEYF_IFINDEX) !=
1889 (key->keyflags & TCP_AO_KEYF_IFINDEX))
1890 continue;
1891
1892 if (key->l3index != l3index)
1893 continue;
1894
1895 if (key == new_current || key == new_rnext)
1896 continue;
1897
1898 return tcp_ao_delete_key(sk, ao_info, cmd.del_async, key,
1899 new_current, new_rnext);
1900 }
1901 return -ENOENT;
1902}
1903
1904/* cmd.ao_required makes a socket TCP-AO only.
1905 * Don't allow any md5 keys for any l3intf on the socket together with it.
1906 * Restricting it early in setsockopt() removes a check for
1907 * ao_info->ao_required on inbound tcp segment fast-path.
1908 */
1909static int tcp_ao_required_verify(struct sock *sk)
1910{
1911#ifdef CONFIG_TCP_MD5SIG
1912 const struct tcp_md5sig_info *md5sig;
1913
1914 if (!static_branch_unlikely(&tcp_md5_needed.key))
1915 return 0;
1916
1917 md5sig = rcu_dereference_check(tcp_sk(sk)->md5sig_info,
1918 lockdep_sock_is_held(sk));
1919 if (!md5sig)
1920 return 0;
1921
1922 if (rcu_dereference_check(hlist_first_rcu(&md5sig->head),
1923 lockdep_sock_is_held(sk)))
1924 return 1;
1925#endif
1926 return 0;
1927}
1928
1929static int tcp_ao_info_cmd(struct sock *sk, unsigned short int family,
1930 sockptr_t optval, int optlen)
1931{
1932 struct tcp_ao_key *new_current = NULL, *new_rnext = NULL;
1933 struct tcp_ao_info *ao_info;
1934 struct tcp_ao_info_opt cmd;
1935 bool first = false;
1936 int err;
1937
1938 if (optlen < sizeof(cmd))
1939 return -EINVAL;
1940
1941 err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1942 if (err)
1943 return err;
1944
1945 if (cmd.set_current || cmd.set_rnext) {
1946 if (!tcp_ao_can_set_current_rnext(sk))
1947 return -EINVAL;
1948 }
1949
1950 if (cmd.reserved != 0 || cmd.reserved2 != 0)
1951 return -EINVAL;
1952
1953 ao_info = setsockopt_ao_info(sk);
1954 if (IS_ERR(ao_info))
1955 return PTR_ERR(ao_info);
1956 if (!ao_info) {
1957 if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1958 return -EINVAL;
1959 ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1960 if (!ao_info)
1961 return -ENOMEM;
1962 first = true;
1963 }
1964
1965 if (cmd.ao_required && tcp_ao_required_verify(sk))
1966 return -EKEYREJECTED;
1967
1968 /* For sockets in TCP_CLOSED it's possible set keys that aren't
1969 * matching the future peer (address/port/VRF/etc),
1970 * tcp_ao_connect_init() will choose a correct matching MKT
1971 * if there's any.
1972 */
1973 if (cmd.set_current) {
1974 new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1);
1975 if (!new_current) {
1976 err = -ENOENT;
1977 goto out;
1978 }
1979 }
1980 if (cmd.set_rnext) {
1981 new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext);
1982 if (!new_rnext) {
1983 err = -ENOENT;
1984 goto out;
1985 }
1986 }
1987 if (cmd.set_counters) {
1988 atomic64_set(&ao_info->counters.pkt_good, cmd.pkt_good);
1989 atomic64_set(&ao_info->counters.pkt_bad, cmd.pkt_bad);
1990 atomic64_set(&ao_info->counters.key_not_found, cmd.pkt_key_not_found);
1991 atomic64_set(&ao_info->counters.ao_required, cmd.pkt_ao_required);
1992 atomic64_set(&ao_info->counters.dropped_icmp, cmd.pkt_dropped_icmp);
1993 }
1994
1995 ao_info->ao_required = cmd.ao_required;
1996 ao_info->accept_icmps = cmd.accept_icmps;
1997 if (new_current)
1998 WRITE_ONCE(ao_info->current_key, new_current);
1999 if (new_rnext)
2000 WRITE_ONCE(ao_info->rnext_key, new_rnext);
2001 if (first) {
2002 if (!static_branch_inc(&tcp_ao_needed.key)) {
2003 err = -EUSERS;
2004 goto out;
2005 }
2006 sk_gso_disable(sk);
2007 rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
2008 }
2009 return 0;
2010out:
2011 if (first)
2012 kfree(ao_info);
2013 return err;
2014}
2015
2016int tcp_parse_ao(struct sock *sk, int cmd, unsigned short int family,
2017 sockptr_t optval, int optlen)
2018{
2019 if (WARN_ON_ONCE(family != AF_INET && family != AF_INET6))
2020 return -EAFNOSUPPORT;
2021
2022 switch (cmd) {
2023 case TCP_AO_ADD_KEY:
2024 return tcp_ao_add_cmd(sk, family, optval, optlen);
2025 case TCP_AO_DEL_KEY:
2026 return tcp_ao_del_cmd(sk, family, optval, optlen);
2027 case TCP_AO_INFO:
2028 return tcp_ao_info_cmd(sk, family, optval, optlen);
2029 default:
2030 WARN_ON_ONCE(1);
2031 return -EINVAL;
2032 }
2033}
2034
2035int tcp_v4_parse_ao(struct sock *sk, int cmd, sockptr_t optval, int optlen)
2036{
2037 return tcp_parse_ao(sk, cmd, AF_INET, optval, optlen);
2038}
2039
2040/* tcp_ao_copy_mkts_to_user(ao_info, optval, optlen)
2041 *
2042 * @ao_info: struct tcp_ao_info on the socket that
2043 * socket getsockopt(TCP_AO_GET_KEYS) is executed on
2044 * @optval: pointer to array of tcp_ao_getsockopt structures in user space.
2045 * Must be != NULL.
2046 * @optlen: pointer to size of tcp_ao_getsockopt structure.
2047 * Must be != NULL.
2048 *
2049 * Return value: 0 on success, a negative error number otherwise.
2050 *
2051 * optval points to an array of tcp_ao_getsockopt structures in user space.
2052 * optval[0] is used as both input and output to getsockopt. It determines
2053 * which keys are returned by the kernel.
2054 * optval[0].nkeys is the size of the array in user space. On return it contains
2055 * the number of keys matching the search criteria.
2056 * If tcp_ao_getsockopt::get_all is set, then all keys in the socket are
2057 * returned, otherwise only keys matching <addr, prefix, sndid, rcvid>
2058 * in optval[0] are returned.
2059 * optlen is also used as both input and output. The user provides the size
2060 * of struct tcp_ao_getsockopt in user space, and the kernel returns the size
2061 * of the structure in kernel space.
2062 * The size of struct tcp_ao_getsockopt may differ between user and kernel.
2063 * There are three cases to consider:
2064 * * If usize == ksize, then keys are copied verbatim.
2065 * * If usize < ksize, then the userspace has passed an old struct to a
2066 * newer kernel. The rest of the trailing bytes in optval[0]
2067 * (ksize - usize) are interpreted as 0 by the kernel.
2068 * * If usize > ksize, then the userspace has passed a new struct to an
2069 * older kernel. The trailing bytes unknown to the kernel (usize - ksize)
2070 * are checked to ensure they are zeroed, otherwise -E2BIG is returned.
2071 * On return the kernel fills in min(usize, ksize) in each entry of the array.
2072 * The layout of the fields in the user and kernel structures is expected to
2073 * be the same (including in the 32bit vs 64bit case).
2074 */
2075static int tcp_ao_copy_mkts_to_user(struct tcp_ao_info *ao_info,
2076 sockptr_t optval, sockptr_t optlen)
2077{
2078 struct tcp_ao_getsockopt opt_in, opt_out;
2079 struct tcp_ao_key *key, *current_key;
2080 bool do_address_matching = true;
2081 union tcp_ao_addr *addr = NULL;
2082 int err, l3index, user_len;
2083 unsigned int max_keys; /* maximum number of keys to copy to user */
2084 size_t out_offset = 0;
2085 size_t bytes_to_write; /* number of bytes to write to user level */
2086 u32 matched_keys; /* keys from ao_info matched so far */
2087 int optlen_out;
2088 __be16 port = 0;
2089
2090 if (copy_from_sockptr(&user_len, optlen, sizeof(int)))
2091 return -EFAULT;
2092
2093 if (user_len <= 0)
2094 return -EINVAL;
2095
2096 memset(&opt_in, 0, sizeof(struct tcp_ao_getsockopt));
2097 err = copy_struct_from_sockptr(&opt_in, sizeof(opt_in),
2098 optval, user_len);
2099 if (err < 0)
2100 return err;
2101
2102 if (opt_in.pkt_good || opt_in.pkt_bad)
2103 return -EINVAL;
2104 if (opt_in.keyflags & ~TCP_AO_GET_KEYF_VALID)
2105 return -EINVAL;
2106 if (opt_in.ifindex && !(opt_in.keyflags & TCP_AO_KEYF_IFINDEX))
2107 return -EINVAL;
2108
2109 if (opt_in.reserved != 0)
2110 return -EINVAL;
2111
2112 max_keys = opt_in.nkeys;
2113 l3index = (opt_in.keyflags & TCP_AO_KEYF_IFINDEX) ? opt_in.ifindex : -1;
2114
2115 if (opt_in.get_all || opt_in.is_current || opt_in.is_rnext) {
2116 if (opt_in.get_all && (opt_in.is_current || opt_in.is_rnext))
2117 return -EINVAL;
2118 do_address_matching = false;
2119 }
2120
2121 switch (opt_in.addr.ss_family) {
2122 case AF_INET: {
2123 struct sockaddr_in *sin;
2124 __be32 mask;
2125
2126 sin = (struct sockaddr_in *)&opt_in.addr;
2127 port = sin->sin_port;
2128 addr = (union tcp_ao_addr *)&sin->sin_addr;
2129
2130 if (opt_in.prefix > 32)
2131 return -EINVAL;
2132
2133 if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY &&
2134 opt_in.prefix != 0)
2135 return -EINVAL;
2136
2137 mask = inet_make_mask(opt_in.prefix);
2138 if (sin->sin_addr.s_addr & ~mask)
2139 return -EINVAL;
2140
2141 break;
2142 }
2143 case AF_INET6: {
2144 struct sockaddr_in6 *sin6;
2145 struct in6_addr *addr6;
2146
2147 sin6 = (struct sockaddr_in6 *)&opt_in.addr;
2148 addr = (union tcp_ao_addr *)&sin6->sin6_addr;
2149 addr6 = &sin6->sin6_addr;
2150 port = sin6->sin6_port;
2151
2152 /* We don't have to change family and @addr here if
2153 * ipv6_addr_v4mapped() like in key adding:
2154 * tcp_ao_key_cmp() does it. Do the sanity checks though.
2155 */
2156 if (opt_in.prefix != 0) {
2157 if (ipv6_addr_v4mapped(addr6)) {
2158 __be32 mask, addr4 = addr6->s6_addr32[3];
2159
2160 if (opt_in.prefix > 32 ||
2161 ntohl(addr4) == INADDR_ANY)
2162 return -EINVAL;
2163 mask = inet_make_mask(opt_in.prefix);
2164 if (addr4 & ~mask)
2165 return -EINVAL;
2166 } else {
2167 struct in6_addr pfx;
2168
2169 if (ipv6_addr_any(addr6) ||
2170 opt_in.prefix > 128)
2171 return -EINVAL;
2172
2173 ipv6_addr_prefix(&pfx, addr6, opt_in.prefix);
2174 if (ipv6_addr_cmp(&pfx, addr6))
2175 return -EINVAL;
2176 }
2177 } else if (!ipv6_addr_any(addr6)) {
2178 return -EINVAL;
2179 }
2180 break;
2181 }
2182 case 0:
2183 if (!do_address_matching)
2184 break;
2185 fallthrough;
2186 default:
2187 return -EAFNOSUPPORT;
2188 }
2189
2190 if (!do_address_matching) {
2191 /* We could just ignore those, but let's do stricter checks */
2192 if (addr || port)
2193 return -EINVAL;
2194 if (opt_in.prefix || opt_in.sndid || opt_in.rcvid)
2195 return -EINVAL;
2196 }
2197
2198 bytes_to_write = min_t(int, user_len, sizeof(struct tcp_ao_getsockopt));
2199 matched_keys = 0;
2200 /* May change in RX, while we're dumping, pre-fetch it */
2201 current_key = READ_ONCE(ao_info->current_key);
2202
2203 hlist_for_each_entry_rcu(key, &ao_info->head, node) {
2204 if (opt_in.get_all)
2205 goto match;
2206
2207 if (opt_in.is_current || opt_in.is_rnext) {
2208 if (opt_in.is_current && key == current_key)
2209 goto match;
2210 if (opt_in.is_rnext && key == ao_info->rnext_key)
2211 goto match;
2212 continue;
2213 }
2214
2215 if (tcp_ao_key_cmp(key, l3index, addr, opt_in.prefix,
2216 opt_in.addr.ss_family,
2217 opt_in.sndid, opt_in.rcvid) != 0)
2218 continue;
2219match:
2220 matched_keys++;
2221 if (matched_keys > max_keys)
2222 continue;
2223
2224 memset(&opt_out, 0, sizeof(struct tcp_ao_getsockopt));
2225
2226 if (key->family == AF_INET) {
2227 struct sockaddr_in *sin_out = (struct sockaddr_in *)&opt_out.addr;
2228
2229 sin_out->sin_family = key->family;
2230 sin_out->sin_port = 0;
2231 memcpy(&sin_out->sin_addr, &key->addr, sizeof(struct in_addr));
2232 } else {
2233 struct sockaddr_in6 *sin6_out = (struct sockaddr_in6 *)&opt_out.addr;
2234
2235 sin6_out->sin6_family = key->family;
2236 sin6_out->sin6_port = 0;
2237 memcpy(&sin6_out->sin6_addr, &key->addr, sizeof(struct in6_addr));
2238 }
2239 opt_out.sndid = key->sndid;
2240 opt_out.rcvid = key->rcvid;
2241 opt_out.prefix = key->prefixlen;
2242 opt_out.keyflags = key->keyflags;
2243 opt_out.is_current = (key == current_key);
2244 opt_out.is_rnext = (key == ao_info->rnext_key);
2245 opt_out.nkeys = 0;
2246 opt_out.maclen = key->maclen;
2247 opt_out.keylen = key->keylen;
2248 opt_out.ifindex = key->l3index;
2249 opt_out.pkt_good = atomic64_read(&key->pkt_good);
2250 opt_out.pkt_bad = atomic64_read(&key->pkt_bad);
2251 memcpy(&opt_out.key, key->key, key->keylen);
2252 tcp_sigpool_algo(key->tcp_sigpool_id, opt_out.alg_name, 64);
2253
2254 /* Copy key to user */
2255 if (copy_to_sockptr_offset(optval, out_offset,
2256 &opt_out, bytes_to_write))
2257 return -EFAULT;
2258 out_offset += user_len;
2259 }
2260
2261 optlen_out = (int)sizeof(struct tcp_ao_getsockopt);
2262 if (copy_to_sockptr(optlen, &optlen_out, sizeof(int)))
2263 return -EFAULT;
2264
2265 out_offset = offsetof(struct tcp_ao_getsockopt, nkeys);
2266 if (copy_to_sockptr_offset(optval, out_offset,
2267 &matched_keys, sizeof(u32)))
2268 return -EFAULT;
2269
2270 return 0;
2271}
2272
2273int tcp_ao_get_mkts(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2274{
2275 struct tcp_ao_info *ao_info;
2276
2277 ao_info = setsockopt_ao_info(sk);
2278 if (IS_ERR(ao_info))
2279 return PTR_ERR(ao_info);
2280 if (!ao_info)
2281 return -ENOENT;
2282
2283 return tcp_ao_copy_mkts_to_user(ao_info, optval, optlen);
2284}
2285
2286int tcp_ao_get_sock_info(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2287{
2288 struct tcp_ao_info_opt out, in = {};
2289 struct tcp_ao_key *current_key;
2290 struct tcp_ao_info *ao;
2291 int err, len;
2292
2293 if (copy_from_sockptr(&len, optlen, sizeof(int)))
2294 return -EFAULT;
2295
2296 if (len <= 0)
2297 return -EINVAL;
2298
2299 /* Copying this "in" only to check ::reserved, ::reserved2,
2300 * that may be needed to extend (struct tcp_ao_info_opt) and
2301 * what getsockopt() provides in future.
2302 */
2303 err = copy_struct_from_sockptr(&in, sizeof(in), optval, len);
2304 if (err)
2305 return err;
2306
2307 if (in.reserved != 0 || in.reserved2 != 0)
2308 return -EINVAL;
2309
2310 ao = setsockopt_ao_info(sk);
2311 if (IS_ERR(ao))
2312 return PTR_ERR(ao);
2313 if (!ao)
2314 return -ENOENT;
2315
2316 memset(&out, 0, sizeof(out));
2317 out.ao_required = ao->ao_required;
2318 out.accept_icmps = ao->accept_icmps;
2319 out.pkt_good = atomic64_read(&ao->counters.pkt_good);
2320 out.pkt_bad = atomic64_read(&ao->counters.pkt_bad);
2321 out.pkt_key_not_found = atomic64_read(&ao->counters.key_not_found);
2322 out.pkt_ao_required = atomic64_read(&ao->counters.ao_required);
2323 out.pkt_dropped_icmp = atomic64_read(&ao->counters.dropped_icmp);
2324
2325 current_key = READ_ONCE(ao->current_key);
2326 if (current_key) {
2327 out.set_current = 1;
2328 out.current_key = current_key->sndid;
2329 }
2330 if (ao->rnext_key) {
2331 out.set_rnext = 1;
2332 out.rnext = ao->rnext_key->rcvid;
2333 }
2334
2335 if (copy_to_sockptr(optval, &out, min_t(int, len, sizeof(out))))
2336 return -EFAULT;
2337
2338 return 0;
2339}
2340
2341int tcp_ao_set_repair(struct sock *sk, sockptr_t optval, unsigned int optlen)
2342{
2343 struct tcp_sock *tp = tcp_sk(sk);
2344 struct tcp_ao_repair cmd;
2345 struct tcp_ao_key *key;
2346 struct tcp_ao_info *ao;
2347 int err;
2348
2349 if (optlen < sizeof(cmd))
2350 return -EINVAL;
2351
2352 err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
2353 if (err)
2354 return err;
2355
2356 if (!tp->repair)
2357 return -EPERM;
2358
2359 ao = setsockopt_ao_info(sk);
2360 if (IS_ERR(ao))
2361 return PTR_ERR(ao);
2362 if (!ao)
2363 return -ENOENT;
2364
2365 WRITE_ONCE(ao->lisn, cmd.snt_isn);
2366 WRITE_ONCE(ao->risn, cmd.rcv_isn);
2367 WRITE_ONCE(ao->snd_sne, cmd.snd_sne);
2368 WRITE_ONCE(ao->rcv_sne, cmd.rcv_sne);
2369
2370 hlist_for_each_entry_rcu(key, &ao->head, node)
2371 tcp_ao_cache_traffic_keys(sk, ao, key);
2372
2373 return 0;
2374}
2375
2376int tcp_ao_get_repair(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2377{
2378 struct tcp_sock *tp = tcp_sk(sk);
2379 struct tcp_ao_repair opt;
2380 struct tcp_ao_info *ao;
2381 int len;
2382
2383 if (copy_from_sockptr(&len, optlen, sizeof(int)))
2384 return -EFAULT;
2385
2386 if (len <= 0)
2387 return -EINVAL;
2388
2389 if (!tp->repair)
2390 return -EPERM;
2391
2392 rcu_read_lock();
2393 ao = getsockopt_ao_info(sk);
2394 if (IS_ERR_OR_NULL(ao)) {
2395 rcu_read_unlock();
2396 return ao ? PTR_ERR(ao) : -ENOENT;
2397 }
2398
2399 opt.snt_isn = ao->lisn;
2400 opt.rcv_isn = ao->risn;
2401 opt.snd_sne = READ_ONCE(ao->snd_sne);
2402 opt.rcv_sne = READ_ONCE(ao->rcv_sne);
2403 rcu_read_unlock();
2404
2405 if (copy_to_sockptr(optval, &opt, min_t(int, len, sizeof(opt))))
2406 return -EFAULT;
2407 return 0;
2408}