xref: /linux/net/ipv4/tcp_ao.c (revision 07fdad3a93756b872da7b53647715c48d0f4a2d0)
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 #include <trace/events/tcp.h>
20 
21 DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_ao_needed, HZ);
22 
23 int tcp_ao_calc_traffic_key(struct tcp_ao_key *mkt, u8 *key, void *ctx,
24 			    unsigned int len, struct tcp_sigpool *hp)
25 {
26 	struct scatterlist sg;
27 	int ret;
28 
29 	if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp->req),
30 				mkt->key, mkt->keylen))
31 		goto clear_hash;
32 
33 	ret = crypto_ahash_init(hp->req);
34 	if (ret)
35 		goto clear_hash;
36 
37 	sg_init_one(&sg, ctx, len);
38 	ahash_request_set_crypt(hp->req, &sg, key, len);
39 	crypto_ahash_update(hp->req);
40 
41 	ret = crypto_ahash_final(hp->req);
42 	if (ret)
43 		goto clear_hash;
44 
45 	return 0;
46 clear_hash:
47 	memset(key, 0, tcp_ao_digest_size(mkt));
48 	return 1;
49 }
50 
51 bool tcp_ao_ignore_icmp(const struct sock *sk, int family, int type, int code)
52 {
53 	bool ignore_icmp = false;
54 	struct tcp_ao_info *ao;
55 
56 	if (!static_branch_unlikely(&tcp_ao_needed.key))
57 		return false;
58 
59 	/* RFC5925, 7.8:
60 	 * >> A TCP-AO implementation MUST default to ignore incoming ICMPv4
61 	 * messages of Type 3 (destination unreachable), Codes 2-4 (protocol
62 	 * unreachable, port unreachable, and fragmentation needed -- ’hard
63 	 * errors’), and ICMPv6 Type 1 (destination unreachable), Code 1
64 	 * (administratively prohibited) and Code 4 (port unreachable) intended
65 	 * for connections in synchronized states (ESTABLISHED, FIN-WAIT-1, FIN-
66 	 * WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, TIME-WAIT) that match MKTs.
67 	 */
68 	if (family == AF_INET) {
69 		if (type != ICMP_DEST_UNREACH)
70 			return false;
71 		if (code < ICMP_PROT_UNREACH || code > ICMP_FRAG_NEEDED)
72 			return false;
73 	} else {
74 		if (type != ICMPV6_DEST_UNREACH)
75 			return false;
76 		if (code != ICMPV6_ADM_PROHIBITED && code != ICMPV6_PORT_UNREACH)
77 			return false;
78 	}
79 
80 	rcu_read_lock();
81 	switch (sk->sk_state) {
82 	case TCP_TIME_WAIT:
83 		ao = rcu_dereference(tcp_twsk(sk)->ao_info);
84 		break;
85 	case TCP_SYN_SENT:
86 	case TCP_SYN_RECV:
87 	case TCP_LISTEN:
88 	case TCP_NEW_SYN_RECV:
89 		/* RFC5925 specifies to ignore ICMPs *only* on connections
90 		 * in synchronized states.
91 		 */
92 		rcu_read_unlock();
93 		return false;
94 	default:
95 		ao = rcu_dereference(tcp_sk(sk)->ao_info);
96 	}
97 
98 	if (ao && !ao->accept_icmps) {
99 		ignore_icmp = true;
100 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAODROPPEDICMPS);
101 		atomic64_inc(&ao->counters.dropped_icmp);
102 	}
103 	rcu_read_unlock();
104 
105 	return ignore_icmp;
106 }
107 
108 /* Optimized version of tcp_ao_do_lookup(): only for sockets for which
109  * it's known that the keys in ao_info are matching peer's
110  * family/address/VRF/etc.
111  */
112 struct tcp_ao_key *tcp_ao_established_key(const struct sock *sk,
113 					  struct tcp_ao_info *ao,
114 					  int sndid, int rcvid)
115 {
116 	struct tcp_ao_key *key;
117 
118 	hlist_for_each_entry_rcu(key, &ao->head, node, lockdep_sock_is_held(sk)) {
119 		if ((sndid >= 0 && key->sndid != sndid) ||
120 		    (rcvid >= 0 && key->rcvid != rcvid))
121 			continue;
122 		return key;
123 	}
124 
125 	return NULL;
126 }
127 
128 static int ipv4_prefix_cmp(const struct in_addr *addr1,
129 			   const struct in_addr *addr2,
130 			   unsigned int prefixlen)
131 {
132 	__be32 mask = inet_make_mask(prefixlen);
133 	__be32 a1 = addr1->s_addr & mask;
134 	__be32 a2 = addr2->s_addr & mask;
135 
136 	if (a1 == a2)
137 		return 0;
138 	return memcmp(&a1, &a2, sizeof(a1));
139 }
140 
141 static int __tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index,
142 			    const union tcp_ao_addr *addr, u8 prefixlen,
143 			    int family, int sndid, int rcvid)
144 {
145 	if (sndid >= 0 && key->sndid != sndid)
146 		return (key->sndid > sndid) ? 1 : -1;
147 	if (rcvid >= 0 && key->rcvid != rcvid)
148 		return (key->rcvid > rcvid) ? 1 : -1;
149 	if (l3index >= 0 && (key->keyflags & TCP_AO_KEYF_IFINDEX)) {
150 		if (key->l3index != l3index)
151 			return (key->l3index > l3index) ? 1 : -1;
152 	}
153 
154 	if (family == AF_UNSPEC)
155 		return 0;
156 	if (key->family != family)
157 		return (key->family > family) ? 1 : -1;
158 
159 	if (family == AF_INET) {
160 		if (ntohl(key->addr.a4.s_addr) == INADDR_ANY)
161 			return 0;
162 		if (ntohl(addr->a4.s_addr) == INADDR_ANY)
163 			return 0;
164 		return ipv4_prefix_cmp(&key->addr.a4, &addr->a4, prefixlen);
165 #if IS_ENABLED(CONFIG_IPV6)
166 	} else {
167 		if (ipv6_addr_any(&key->addr.a6) || ipv6_addr_any(&addr->a6))
168 			return 0;
169 		if (ipv6_prefix_equal(&key->addr.a6, &addr->a6, prefixlen))
170 			return 0;
171 		return memcmp(&key->addr.a6, &addr->a6, sizeof(addr->a6));
172 #endif
173 	}
174 	return -1;
175 }
176 
177 static int tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index,
178 			  const union tcp_ao_addr *addr, u8 prefixlen,
179 			  int family, int sndid, int rcvid)
180 {
181 #if IS_ENABLED(CONFIG_IPV6)
182 	if (family == AF_INET6 && ipv6_addr_v4mapped(&addr->a6)) {
183 		__be32 addr4 = addr->a6.s6_addr32[3];
184 
185 		return __tcp_ao_key_cmp(key, l3index,
186 					(union tcp_ao_addr *)&addr4,
187 					prefixlen, AF_INET, sndid, rcvid);
188 	}
189 #endif
190 	return __tcp_ao_key_cmp(key, l3index, addr,
191 				prefixlen, family, sndid, rcvid);
192 }
193 
194 static struct tcp_ao_key *__tcp_ao_do_lookup(const struct sock *sk, int l3index,
195 		const union tcp_ao_addr *addr, int family, u8 prefix,
196 		int sndid, int rcvid)
197 {
198 	struct tcp_ao_key *key;
199 	struct tcp_ao_info *ao;
200 
201 	if (!static_branch_unlikely(&tcp_ao_needed.key))
202 		return NULL;
203 
204 	ao = rcu_dereference_check(tcp_sk(sk)->ao_info,
205 				   lockdep_sock_is_held(sk));
206 	if (!ao)
207 		return NULL;
208 
209 	hlist_for_each_entry_rcu(key, &ao->head, node, lockdep_sock_is_held(sk)) {
210 		u8 prefixlen = min(prefix, key->prefixlen);
211 
212 		if (!tcp_ao_key_cmp(key, l3index, addr, prefixlen,
213 				    family, sndid, rcvid))
214 			return key;
215 	}
216 	return NULL;
217 }
218 
219 struct tcp_ao_key *tcp_ao_do_lookup(const struct sock *sk, int l3index,
220 				    const union tcp_ao_addr *addr,
221 				    int family, int sndid, int rcvid)
222 {
223 	return __tcp_ao_do_lookup(sk, l3index, addr, family, U8_MAX, sndid, rcvid);
224 }
225 
226 static struct tcp_ao_info *tcp_ao_alloc_info(gfp_t flags)
227 {
228 	struct tcp_ao_info *ao;
229 
230 	ao = kzalloc(sizeof(*ao), flags);
231 	if (!ao)
232 		return NULL;
233 	INIT_HLIST_HEAD(&ao->head);
234 	refcount_set(&ao->refcnt, 1);
235 
236 	return ao;
237 }
238 
239 static void tcp_ao_link_mkt(struct tcp_ao_info *ao, struct tcp_ao_key *mkt)
240 {
241 	hlist_add_head_rcu(&mkt->node, &ao->head);
242 }
243 
244 static struct tcp_ao_key *tcp_ao_copy_key(struct sock *sk,
245 					  struct tcp_ao_key *key)
246 {
247 	struct tcp_ao_key *new_key;
248 
249 	new_key = sock_kmalloc(sk, tcp_ao_sizeof_key(key),
250 			       GFP_ATOMIC);
251 	if (!new_key)
252 		return NULL;
253 
254 	*new_key = *key;
255 	INIT_HLIST_NODE(&new_key->node);
256 	tcp_sigpool_get(new_key->tcp_sigpool_id);
257 	atomic64_set(&new_key->pkt_good, 0);
258 	atomic64_set(&new_key->pkt_bad, 0);
259 
260 	return new_key;
261 }
262 
263 static void tcp_ao_key_free_rcu(struct rcu_head *head)
264 {
265 	struct tcp_ao_key *key = container_of(head, struct tcp_ao_key, rcu);
266 
267 	tcp_sigpool_release(key->tcp_sigpool_id);
268 	kfree_sensitive(key);
269 }
270 
271 static void tcp_ao_info_free(struct tcp_ao_info *ao)
272 {
273 	struct tcp_ao_key *key;
274 	struct hlist_node *n;
275 
276 	hlist_for_each_entry_safe(key, n, &ao->head, node) {
277 		hlist_del(&key->node);
278 		tcp_sigpool_release(key->tcp_sigpool_id);
279 		kfree_sensitive(key);
280 	}
281 	kfree(ao);
282 	static_branch_slow_dec_deferred(&tcp_ao_needed);
283 }
284 
285 static void tcp_ao_sk_omem_free(struct sock *sk, struct tcp_ao_info *ao)
286 {
287 	size_t total_ao_sk_mem = 0;
288 	struct tcp_ao_key *key;
289 
290 	hlist_for_each_entry(key,  &ao->head, node)
291 		total_ao_sk_mem += tcp_ao_sizeof_key(key);
292 	atomic_sub(total_ao_sk_mem, &sk->sk_omem_alloc);
293 }
294 
295 void tcp_ao_destroy_sock(struct sock *sk, bool twsk)
296 {
297 	struct tcp_ao_info *ao;
298 
299 	if (twsk) {
300 		ao = rcu_dereference_protected(tcp_twsk(sk)->ao_info, 1);
301 		rcu_assign_pointer(tcp_twsk(sk)->ao_info, NULL);
302 	} else {
303 		ao = rcu_dereference_protected(tcp_sk(sk)->ao_info, 1);
304 		rcu_assign_pointer(tcp_sk(sk)->ao_info, NULL);
305 	}
306 
307 	if (!ao || !refcount_dec_and_test(&ao->refcnt))
308 		return;
309 
310 	if (!twsk)
311 		tcp_ao_sk_omem_free(sk, ao);
312 	tcp_ao_info_free(ao);
313 }
314 
315 void tcp_ao_time_wait(struct tcp_timewait_sock *tcptw, struct tcp_sock *tp)
316 {
317 	struct tcp_ao_info *ao_info = rcu_dereference_protected(tp->ao_info, 1);
318 
319 	if (ao_info) {
320 		struct tcp_ao_key *key;
321 		struct hlist_node *n;
322 		int omem = 0;
323 
324 		hlist_for_each_entry_safe(key, n, &ao_info->head, node) {
325 			omem += tcp_ao_sizeof_key(key);
326 		}
327 
328 		refcount_inc(&ao_info->refcnt);
329 		atomic_sub(omem, &(((struct sock *)tp)->sk_omem_alloc));
330 		rcu_assign_pointer(tcptw->ao_info, ao_info);
331 	} else {
332 		tcptw->ao_info = NULL;
333 	}
334 }
335 
336 /* 4 tuple and ISNs are expected in NBO */
337 static int tcp_v4_ao_calc_key(struct tcp_ao_key *mkt, u8 *key,
338 			      __be32 saddr, __be32 daddr,
339 			      __be16 sport, __be16 dport,
340 			      __be32 sisn,  __be32 disn)
341 {
342 	/* See RFC5926 3.1.1 */
343 	struct kdf_input_block {
344 		u8                      counter;
345 		u8                      label[6];
346 		struct tcp4_ao_context	ctx;
347 		__be16                  outlen;
348 	} __packed * tmp;
349 	struct tcp_sigpool hp;
350 	int err;
351 
352 	err = tcp_sigpool_start(mkt->tcp_sigpool_id, &hp);
353 	if (err)
354 		return err;
355 
356 	tmp = hp.scratch;
357 	tmp->counter	= 1;
358 	memcpy(tmp->label, "TCP-AO", 6);
359 	tmp->ctx.saddr	= saddr;
360 	tmp->ctx.daddr	= daddr;
361 	tmp->ctx.sport	= sport;
362 	tmp->ctx.dport	= dport;
363 	tmp->ctx.sisn	= sisn;
364 	tmp->ctx.disn	= disn;
365 	tmp->outlen	= htons(tcp_ao_digest_size(mkt) * 8); /* in bits */
366 
367 	err = tcp_ao_calc_traffic_key(mkt, key, tmp, sizeof(*tmp), &hp);
368 	tcp_sigpool_end(&hp);
369 
370 	return err;
371 }
372 
373 int tcp_v4_ao_calc_key_sk(struct tcp_ao_key *mkt, u8 *key,
374 			  const struct sock *sk,
375 			  __be32 sisn, __be32 disn, bool send)
376 {
377 	if (send)
378 		return tcp_v4_ao_calc_key(mkt, key, sk->sk_rcv_saddr,
379 					  sk->sk_daddr, htons(sk->sk_num),
380 					  sk->sk_dport, sisn, disn);
381 	else
382 		return tcp_v4_ao_calc_key(mkt, key, sk->sk_daddr,
383 					  sk->sk_rcv_saddr, sk->sk_dport,
384 					  htons(sk->sk_num), disn, sisn);
385 }
386 
387 static int tcp_ao_calc_key_sk(struct tcp_ao_key *mkt, u8 *key,
388 			      const struct sock *sk,
389 			      __be32 sisn, __be32 disn, bool send)
390 {
391 	if (mkt->family == AF_INET)
392 		return tcp_v4_ao_calc_key_sk(mkt, key, sk, sisn, disn, send);
393 #if IS_ENABLED(CONFIG_IPV6)
394 	else if (mkt->family == AF_INET6)
395 		return tcp_v6_ao_calc_key_sk(mkt, key, sk, sisn, disn, send);
396 #endif
397 	else
398 		return -EOPNOTSUPP;
399 }
400 
401 int tcp_v4_ao_calc_key_rsk(struct tcp_ao_key *mkt, u8 *key,
402 			   struct request_sock *req)
403 {
404 	struct inet_request_sock *ireq = inet_rsk(req);
405 
406 	return tcp_v4_ao_calc_key(mkt, key,
407 				  ireq->ir_loc_addr, ireq->ir_rmt_addr,
408 				  htons(ireq->ir_num), ireq->ir_rmt_port,
409 				  htonl(tcp_rsk(req)->snt_isn),
410 				  htonl(tcp_rsk(req)->rcv_isn));
411 }
412 
413 static int tcp_v4_ao_calc_key_skb(struct tcp_ao_key *mkt, u8 *key,
414 				  const struct sk_buff *skb,
415 				  __be32 sisn, __be32 disn)
416 {
417 	const struct iphdr *iph = ip_hdr(skb);
418 	const struct tcphdr *th = tcp_hdr(skb);
419 
420 	return tcp_v4_ao_calc_key(mkt, key, iph->saddr, iph->daddr,
421 				  th->source, th->dest, sisn, disn);
422 }
423 
424 static int tcp_ao_calc_key_skb(struct tcp_ao_key *mkt, u8 *key,
425 			       const struct sk_buff *skb,
426 			       __be32 sisn, __be32 disn, int family)
427 {
428 	if (family == AF_INET)
429 		return tcp_v4_ao_calc_key_skb(mkt, key, skb, sisn, disn);
430 #if IS_ENABLED(CONFIG_IPV6)
431 	else if (family == AF_INET6)
432 		return tcp_v6_ao_calc_key_skb(mkt, key, skb, sisn, disn);
433 #endif
434 	return -EAFNOSUPPORT;
435 }
436 
437 static int tcp_v4_ao_hash_pseudoheader(struct tcp_sigpool *hp,
438 				       __be32 daddr, __be32 saddr,
439 				       int nbytes)
440 {
441 	struct tcp4_pseudohdr *bp;
442 	struct scatterlist sg;
443 
444 	bp = hp->scratch;
445 	bp->saddr = saddr;
446 	bp->daddr = daddr;
447 	bp->pad = 0;
448 	bp->protocol = IPPROTO_TCP;
449 	bp->len = cpu_to_be16(nbytes);
450 
451 	sg_init_one(&sg, bp, sizeof(*bp));
452 	ahash_request_set_crypt(hp->req, &sg, NULL, sizeof(*bp));
453 	return crypto_ahash_update(hp->req);
454 }
455 
456 static int tcp_ao_hash_pseudoheader(unsigned short int family,
457 				    const struct sock *sk,
458 				    const struct sk_buff *skb,
459 				    struct tcp_sigpool *hp, int nbytes)
460 {
461 	const struct tcphdr *th = tcp_hdr(skb);
462 
463 	/* TODO: Can we rely on checksum being zero to mean outbound pkt? */
464 	if (!th->check) {
465 		if (family == AF_INET)
466 			return tcp_v4_ao_hash_pseudoheader(hp, sk->sk_daddr,
467 					sk->sk_rcv_saddr, skb->len);
468 #if IS_ENABLED(CONFIG_IPV6)
469 		else if (family == AF_INET6)
470 			return tcp_v6_ao_hash_pseudoheader(hp, &sk->sk_v6_daddr,
471 					&sk->sk_v6_rcv_saddr, skb->len);
472 #endif
473 		else
474 			return -EAFNOSUPPORT;
475 	}
476 
477 	if (family == AF_INET) {
478 		const struct iphdr *iph = ip_hdr(skb);
479 
480 		return tcp_v4_ao_hash_pseudoheader(hp, iph->daddr,
481 				iph->saddr, skb->len);
482 #if IS_ENABLED(CONFIG_IPV6)
483 	} else if (family == AF_INET6) {
484 		const struct ipv6hdr *iph = ipv6_hdr(skb);
485 
486 		return tcp_v6_ao_hash_pseudoheader(hp, &iph->daddr,
487 				&iph->saddr, skb->len);
488 #endif
489 	}
490 	return -EAFNOSUPPORT;
491 }
492 
493 u32 tcp_ao_compute_sne(u32 next_sne, u32 next_seq, u32 seq)
494 {
495 	u32 sne = next_sne;
496 
497 	if (before(seq, next_seq)) {
498 		if (seq > next_seq)
499 			sne--;
500 	} else {
501 		if (seq < next_seq)
502 			sne++;
503 	}
504 
505 	return sne;
506 }
507 
508 /* tcp_ao_hash_sne(struct tcp_sigpool *hp)
509  * @hp	- used for hashing
510  * @sne - sne value
511  */
512 static int tcp_ao_hash_sne(struct tcp_sigpool *hp, u32 sne)
513 {
514 	struct scatterlist sg;
515 	__be32 *bp;
516 
517 	bp = (__be32 *)hp->scratch;
518 	*bp = htonl(sne);
519 
520 	sg_init_one(&sg, bp, sizeof(*bp));
521 	ahash_request_set_crypt(hp->req, &sg, NULL, sizeof(*bp));
522 	return crypto_ahash_update(hp->req);
523 }
524 
525 static int tcp_ao_hash_header(struct tcp_sigpool *hp,
526 			      const struct tcphdr *th,
527 			      bool exclude_options, u8 *hash,
528 			      int hash_offset, int hash_len)
529 {
530 	struct scatterlist sg;
531 	u8 *hdr = hp->scratch;
532 	int err, len;
533 
534 	/* We are not allowed to change tcphdr, make a local copy */
535 	if (exclude_options) {
536 		len = sizeof(*th) + sizeof(struct tcp_ao_hdr) + hash_len;
537 		memcpy(hdr, th, sizeof(*th));
538 		memcpy(hdr + sizeof(*th),
539 		       (u8 *)th + hash_offset - sizeof(struct tcp_ao_hdr),
540 		       sizeof(struct tcp_ao_hdr));
541 		memset(hdr + sizeof(*th) + sizeof(struct tcp_ao_hdr),
542 		       0, hash_len);
543 		((struct tcphdr *)hdr)->check = 0;
544 	} else {
545 		len = th->doff << 2;
546 		memcpy(hdr, th, len);
547 		/* zero out tcp-ao hash */
548 		((struct tcphdr *)hdr)->check = 0;
549 		memset(hdr + hash_offset, 0, hash_len);
550 	}
551 
552 	sg_init_one(&sg, hdr, len);
553 	ahash_request_set_crypt(hp->req, &sg, NULL, len);
554 	err = crypto_ahash_update(hp->req);
555 	WARN_ON_ONCE(err != 0);
556 	return err;
557 }
558 
559 int tcp_ao_hash_hdr(unsigned short int family, char *ao_hash,
560 		    struct tcp_ao_key *key, const u8 *tkey,
561 		    const union tcp_ao_addr *daddr,
562 		    const union tcp_ao_addr *saddr,
563 		    const struct tcphdr *th, u32 sne)
564 {
565 	int tkey_len = tcp_ao_digest_size(key);
566 	int hash_offset = ao_hash - (char *)th;
567 	struct tcp_sigpool hp;
568 	void *hash_buf = NULL;
569 
570 	hash_buf = kmalloc(tkey_len, GFP_ATOMIC);
571 	if (!hash_buf)
572 		goto clear_hash_noput;
573 
574 	if (tcp_sigpool_start(key->tcp_sigpool_id, &hp))
575 		goto clear_hash_noput;
576 
577 	if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp.req), tkey, tkey_len))
578 		goto clear_hash;
579 
580 	if (crypto_ahash_init(hp.req))
581 		goto clear_hash;
582 
583 	if (tcp_ao_hash_sne(&hp, sne))
584 		goto clear_hash;
585 	if (family == AF_INET) {
586 		if (tcp_v4_ao_hash_pseudoheader(&hp, daddr->a4.s_addr,
587 						saddr->a4.s_addr, th->doff * 4))
588 			goto clear_hash;
589 #if IS_ENABLED(CONFIG_IPV6)
590 	} else if (family == AF_INET6) {
591 		if (tcp_v6_ao_hash_pseudoheader(&hp, &daddr->a6,
592 						&saddr->a6, th->doff * 4))
593 			goto clear_hash;
594 #endif
595 	} else {
596 		WARN_ON_ONCE(1);
597 		goto clear_hash;
598 	}
599 	if (tcp_ao_hash_header(&hp, th,
600 			       !!(key->keyflags & TCP_AO_KEYF_EXCLUDE_OPT),
601 			       ao_hash, hash_offset, tcp_ao_maclen(key)))
602 		goto clear_hash;
603 	ahash_request_set_crypt(hp.req, NULL, hash_buf, 0);
604 	if (crypto_ahash_final(hp.req))
605 		goto clear_hash;
606 
607 	memcpy(ao_hash, hash_buf, tcp_ao_maclen(key));
608 	tcp_sigpool_end(&hp);
609 	kfree(hash_buf);
610 	return 0;
611 
612 clear_hash:
613 	tcp_sigpool_end(&hp);
614 clear_hash_noput:
615 	memset(ao_hash, 0, tcp_ao_maclen(key));
616 	kfree(hash_buf);
617 	return 1;
618 }
619 
620 int tcp_ao_hash_skb(unsigned short int family,
621 		    char *ao_hash, struct tcp_ao_key *key,
622 		    const struct sock *sk, const struct sk_buff *skb,
623 		    const u8 *tkey, int hash_offset, u32 sne)
624 {
625 	const struct tcphdr *th = tcp_hdr(skb);
626 	int tkey_len = tcp_ao_digest_size(key);
627 	struct tcp_sigpool hp;
628 	void *hash_buf = NULL;
629 
630 	hash_buf = kmalloc(tkey_len, GFP_ATOMIC);
631 	if (!hash_buf)
632 		goto clear_hash_noput;
633 
634 	if (tcp_sigpool_start(key->tcp_sigpool_id, &hp))
635 		goto clear_hash_noput;
636 
637 	if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp.req), tkey, tkey_len))
638 		goto clear_hash;
639 
640 	/* For now use sha1 by default. Depends on alg in tcp_ao_key */
641 	if (crypto_ahash_init(hp.req))
642 		goto clear_hash;
643 
644 	if (tcp_ao_hash_sne(&hp, sne))
645 		goto clear_hash;
646 	if (tcp_ao_hash_pseudoheader(family, sk, skb, &hp, skb->len))
647 		goto clear_hash;
648 	if (tcp_ao_hash_header(&hp, th,
649 			       !!(key->keyflags & TCP_AO_KEYF_EXCLUDE_OPT),
650 			       ao_hash, hash_offset, tcp_ao_maclen(key)))
651 		goto clear_hash;
652 	if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
653 		goto clear_hash;
654 	ahash_request_set_crypt(hp.req, NULL, hash_buf, 0);
655 	if (crypto_ahash_final(hp.req))
656 		goto clear_hash;
657 
658 	memcpy(ao_hash, hash_buf, tcp_ao_maclen(key));
659 	tcp_sigpool_end(&hp);
660 	kfree(hash_buf);
661 	return 0;
662 
663 clear_hash:
664 	tcp_sigpool_end(&hp);
665 clear_hash_noput:
666 	memset(ao_hash, 0, tcp_ao_maclen(key));
667 	kfree(hash_buf);
668 	return 1;
669 }
670 
671 int tcp_v4_ao_hash_skb(char *ao_hash, struct tcp_ao_key *key,
672 		       const struct sock *sk, const struct sk_buff *skb,
673 		       const u8 *tkey, int hash_offset, u32 sne)
674 {
675 	return tcp_ao_hash_skb(AF_INET, ao_hash, key, sk, skb,
676 			       tkey, hash_offset, sne);
677 }
678 
679 int tcp_v4_ao_synack_hash(char *ao_hash, struct tcp_ao_key *ao_key,
680 			  struct request_sock *req, const struct sk_buff *skb,
681 			  int hash_offset, u32 sne)
682 {
683 	void *hash_buf = NULL;
684 	int err;
685 
686 	hash_buf = kmalloc(tcp_ao_digest_size(ao_key), GFP_ATOMIC);
687 	if (!hash_buf)
688 		return -ENOMEM;
689 
690 	err = tcp_v4_ao_calc_key_rsk(ao_key, hash_buf, req);
691 	if (err)
692 		goto out;
693 
694 	err = tcp_ao_hash_skb(AF_INET, ao_hash, ao_key, req_to_sk(req), skb,
695 			      hash_buf, hash_offset, sne);
696 out:
697 	kfree(hash_buf);
698 	return err;
699 }
700 
701 struct tcp_ao_key *tcp_v4_ao_lookup_rsk(const struct sock *sk,
702 					struct request_sock *req,
703 					int sndid, int rcvid)
704 {
705 	struct inet_request_sock *ireq = inet_rsk(req);
706 	union tcp_ao_addr *addr = (union tcp_ao_addr *)&ireq->ir_rmt_addr;
707 	int l3index;
708 
709 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
710 	return tcp_ao_do_lookup(sk, l3index, addr, AF_INET, sndid, rcvid);
711 }
712 
713 struct tcp_ao_key *tcp_v4_ao_lookup(const struct sock *sk, struct sock *addr_sk,
714 				    int sndid, int rcvid)
715 {
716 	int l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
717 						     addr_sk->sk_bound_dev_if);
718 	union tcp_ao_addr *addr = (union tcp_ao_addr *)&addr_sk->sk_daddr;
719 
720 	return tcp_ao_do_lookup(sk, l3index, addr, AF_INET, sndid, rcvid);
721 }
722 
723 int tcp_ao_prepare_reset(const struct sock *sk, struct sk_buff *skb,
724 			 const struct tcp_ao_hdr *aoh, int l3index, u32 seq,
725 			 struct tcp_ao_key **key, char **traffic_key,
726 			 bool *allocated_traffic_key, u8 *keyid, u32 *sne)
727 {
728 	const struct tcphdr *th = tcp_hdr(skb);
729 	struct tcp_ao_info *ao_info;
730 
731 	*allocated_traffic_key = false;
732 	/* If there's no socket - than initial sisn/disn are unknown.
733 	 * Drop the segment. RFC5925 (7.7) advises to require graceful
734 	 * restart [RFC4724]. Alternatively, the RFC5925 advises to
735 	 * save/restore traffic keys before/after reboot.
736 	 * Linux TCP-AO support provides TCP_AO_ADD_KEY and TCP_AO_REPAIR
737 	 * options to restore a socket post-reboot.
738 	 */
739 	if (!sk)
740 		return -ENOTCONN;
741 
742 	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV)) {
743 		unsigned int family = READ_ONCE(sk->sk_family);
744 		union tcp_ao_addr *addr;
745 		__be32 disn, sisn;
746 
747 		if (sk->sk_state == TCP_NEW_SYN_RECV) {
748 			struct request_sock *req = inet_reqsk(sk);
749 
750 			sisn = htonl(tcp_rsk(req)->rcv_isn);
751 			disn = htonl(tcp_rsk(req)->snt_isn);
752 			*sne = tcp_ao_compute_sne(0, tcp_rsk(req)->snt_isn, seq);
753 		} else {
754 			sisn = th->seq;
755 			disn = 0;
756 		}
757 		if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
758 			addr = (union tcp_md5_addr *)&ipv6_hdr(skb)->saddr;
759 		else
760 			addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
761 #if IS_ENABLED(CONFIG_IPV6)
762 		if (family == AF_INET6 && ipv6_addr_v4mapped(&sk->sk_v6_daddr))
763 			family = AF_INET;
764 #endif
765 
766 		sk = sk_const_to_full_sk(sk);
767 		ao_info = rcu_dereference(tcp_sk(sk)->ao_info);
768 		if (!ao_info)
769 			return -ENOENT;
770 		*key = tcp_ao_do_lookup(sk, l3index, addr, family,
771 					-1, aoh->rnext_keyid);
772 		if (!*key)
773 			return -ENOENT;
774 		*traffic_key = kmalloc(tcp_ao_digest_size(*key), GFP_ATOMIC);
775 		if (!*traffic_key)
776 			return -ENOMEM;
777 		*allocated_traffic_key = true;
778 		if (tcp_ao_calc_key_skb(*key, *traffic_key, skb,
779 					sisn, disn, family))
780 			return -1;
781 		*keyid = (*key)->rcvid;
782 	} else {
783 		struct tcp_ao_key *rnext_key;
784 		u32 snd_basis;
785 
786 		if (sk->sk_state == TCP_TIME_WAIT) {
787 			ao_info = rcu_dereference(tcp_twsk(sk)->ao_info);
788 			snd_basis = tcp_twsk(sk)->tw_snd_nxt;
789 		} else {
790 			ao_info = rcu_dereference(tcp_sk(sk)->ao_info);
791 			snd_basis = tcp_sk(sk)->snd_una;
792 		}
793 		if (!ao_info)
794 			return -ENOENT;
795 
796 		*key = tcp_ao_established_key(sk, ao_info, aoh->rnext_keyid, -1);
797 		if (!*key)
798 			return -ENOENT;
799 		*traffic_key = snd_other_key(*key);
800 		rnext_key = READ_ONCE(ao_info->rnext_key);
801 		*keyid = rnext_key->rcvid;
802 		*sne = tcp_ao_compute_sne(READ_ONCE(ao_info->snd_sne),
803 					  snd_basis, seq);
804 	}
805 	return 0;
806 }
807 
808 int tcp_ao_transmit_skb(struct sock *sk, struct sk_buff *skb,
809 			struct tcp_ao_key *key, struct tcphdr *th,
810 			__u8 *hash_location)
811 {
812 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
813 	struct tcp_sock *tp = tcp_sk(sk);
814 	struct tcp_ao_info *ao;
815 	void *tkey_buf = NULL;
816 	u8 *traffic_key;
817 	u32 sne;
818 
819 	ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
820 				       lockdep_sock_is_held(sk));
821 	traffic_key = snd_other_key(key);
822 	if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
823 		__be32 disn;
824 
825 		if (!(tcb->tcp_flags & TCPHDR_ACK)) {
826 			disn = 0;
827 			tkey_buf = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC);
828 			if (!tkey_buf)
829 				return -ENOMEM;
830 			traffic_key = tkey_buf;
831 		} else {
832 			disn = ao->risn;
833 		}
834 		tp->af_specific->ao_calc_key_sk(key, traffic_key,
835 						sk, ao->lisn, disn, true);
836 	}
837 	sne = tcp_ao_compute_sne(READ_ONCE(ao->snd_sne), READ_ONCE(tp->snd_una),
838 				 ntohl(th->seq));
839 	tp->af_specific->calc_ao_hash(hash_location, key, sk, skb, traffic_key,
840 				      hash_location - (u8 *)th, sne);
841 	kfree(tkey_buf);
842 	return 0;
843 }
844 
845 static struct tcp_ao_key *tcp_ao_inbound_lookup(unsigned short int family,
846 		const struct sock *sk, const struct sk_buff *skb,
847 		int sndid, int rcvid, int l3index)
848 {
849 	if (family == AF_INET) {
850 		const struct iphdr *iph = ip_hdr(skb);
851 
852 		return tcp_ao_do_lookup(sk, l3index,
853 					(union tcp_ao_addr *)&iph->saddr,
854 					AF_INET, sndid, rcvid);
855 	} else {
856 		const struct ipv6hdr *iph = ipv6_hdr(skb);
857 
858 		return tcp_ao_do_lookup(sk, l3index,
859 					(union tcp_ao_addr *)&iph->saddr,
860 					AF_INET6, sndid, rcvid);
861 	}
862 }
863 
864 void tcp_ao_syncookie(struct sock *sk, const struct sk_buff *skb,
865 		      struct request_sock *req, unsigned short int family)
866 {
867 	struct tcp_request_sock *treq = tcp_rsk(req);
868 	const struct tcphdr *th = tcp_hdr(skb);
869 	const struct tcp_ao_hdr *aoh;
870 	struct tcp_ao_key *key;
871 	int l3index;
872 
873 	/* treq->af_specific is used to perform TCP_AO lookup
874 	 * in tcp_create_openreq_child().
875 	 */
876 #if IS_ENABLED(CONFIG_IPV6)
877 	if (family == AF_INET6)
878 		treq->af_specific = &tcp_request_sock_ipv6_ops;
879 	else
880 #endif
881 		treq->af_specific = &tcp_request_sock_ipv4_ops;
882 
883 	treq->used_tcp_ao = false;
884 
885 	if (tcp_parse_auth_options(th, NULL, &aoh) || !aoh)
886 		return;
887 
888 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), inet_rsk(req)->ir_iif);
889 	key = tcp_ao_inbound_lookup(family, sk, skb, -1, aoh->keyid, l3index);
890 	if (!key)
891 		/* Key not found, continue without TCP-AO */
892 		return;
893 
894 	treq->ao_rcv_next = aoh->keyid;
895 	treq->ao_keyid = aoh->rnext_keyid;
896 	treq->used_tcp_ao = true;
897 }
898 
899 static enum skb_drop_reason
900 tcp_ao_verify_hash(const struct sock *sk, const struct sk_buff *skb,
901 		   unsigned short int family, struct tcp_ao_info *info,
902 		   const struct tcp_ao_hdr *aoh, struct tcp_ao_key *key,
903 		   u8 *traffic_key, u8 *phash, u32 sne, int l3index)
904 {
905 	const struct tcphdr *th = tcp_hdr(skb);
906 	u8 maclen = tcp_ao_hdr_maclen(aoh);
907 	void *hash_buf = NULL;
908 
909 	if (maclen != tcp_ao_maclen(key)) {
910 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD);
911 		atomic64_inc(&info->counters.pkt_bad);
912 		atomic64_inc(&key->pkt_bad);
913 		trace_tcp_ao_wrong_maclen(sk, skb, aoh->keyid,
914 					  aoh->rnext_keyid, maclen);
915 		return SKB_DROP_REASON_TCP_AOFAILURE;
916 	}
917 
918 	hash_buf = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC);
919 	if (!hash_buf)
920 		return SKB_DROP_REASON_NOT_SPECIFIED;
921 
922 	/* XXX: make it per-AF callback? */
923 	tcp_ao_hash_skb(family, hash_buf, key, sk, skb, traffic_key,
924 			(phash - (u8 *)th), sne);
925 	if (memcmp(phash, hash_buf, maclen)) {
926 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD);
927 		atomic64_inc(&info->counters.pkt_bad);
928 		atomic64_inc(&key->pkt_bad);
929 		trace_tcp_ao_mismatch(sk, skb, aoh->keyid,
930 				      aoh->rnext_keyid, maclen);
931 		kfree(hash_buf);
932 		return SKB_DROP_REASON_TCP_AOFAILURE;
933 	}
934 	NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOGOOD);
935 	atomic64_inc(&info->counters.pkt_good);
936 	atomic64_inc(&key->pkt_good);
937 	kfree(hash_buf);
938 	return SKB_NOT_DROPPED_YET;
939 }
940 
941 enum skb_drop_reason
942 tcp_inbound_ao_hash(struct sock *sk, const struct sk_buff *skb,
943 		    unsigned short int family, const struct request_sock *req,
944 		    int l3index, const struct tcp_ao_hdr *aoh)
945 {
946 	const struct tcphdr *th = tcp_hdr(skb);
947 	u8 maclen = tcp_ao_hdr_maclen(aoh);
948 	u8 *phash = (u8 *)(aoh + 1); /* hash goes just after the header */
949 	struct tcp_ao_info *info;
950 	enum skb_drop_reason ret;
951 	struct tcp_ao_key *key;
952 	__be32 sisn, disn;
953 	u8 *traffic_key;
954 	int state;
955 	u32 sne = 0;
956 
957 	info = rcu_dereference(tcp_sk(sk)->ao_info);
958 	if (!info) {
959 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOKEYNOTFOUND);
960 		trace_tcp_ao_key_not_found(sk, skb, aoh->keyid,
961 					   aoh->rnext_keyid, maclen);
962 		return SKB_DROP_REASON_TCP_AOUNEXPECTED;
963 	}
964 
965 	if (unlikely(th->syn)) {
966 		sisn = th->seq;
967 		disn = 0;
968 	}
969 
970 	state = READ_ONCE(sk->sk_state);
971 	/* Fast-path */
972 	if (likely((1 << state) & TCP_AO_ESTABLISHED)) {
973 		enum skb_drop_reason err;
974 		struct tcp_ao_key *current_key;
975 
976 		/* Check if this socket's rnext_key matches the keyid in the
977 		 * packet. If not we lookup the key based on the keyid
978 		 * matching the rcvid in the mkt.
979 		 */
980 		key = READ_ONCE(info->rnext_key);
981 		if (key->rcvid != aoh->keyid) {
982 			key = tcp_ao_established_key(sk, info, -1, aoh->keyid);
983 			if (!key)
984 				goto key_not_found;
985 		}
986 
987 		/* Delayed retransmitted SYN */
988 		if (unlikely(th->syn && !th->ack))
989 			goto verify_hash;
990 
991 		sne = tcp_ao_compute_sne(info->rcv_sne, tcp_sk(sk)->rcv_nxt,
992 					 ntohl(th->seq));
993 		/* Established socket, traffic key are cached */
994 		traffic_key = rcv_other_key(key);
995 		err = tcp_ao_verify_hash(sk, skb, family, info, aoh, key,
996 					 traffic_key, phash, sne, l3index);
997 		if (err)
998 			return err;
999 		current_key = READ_ONCE(info->current_key);
1000 		/* Key rotation: the peer asks us to use new key (RNext) */
1001 		if (unlikely(aoh->rnext_keyid != current_key->sndid)) {
1002 			trace_tcp_ao_rnext_request(sk, skb, current_key->sndid,
1003 						   aoh->rnext_keyid,
1004 						   tcp_ao_hdr_maclen(aoh));
1005 			/* If the key is not found we do nothing. */
1006 			key = tcp_ao_established_key(sk, info, aoh->rnext_keyid, -1);
1007 			if (key)
1008 				/* pairs with tcp_ao_del_cmd */
1009 				WRITE_ONCE(info->current_key, key);
1010 		}
1011 		return SKB_NOT_DROPPED_YET;
1012 	}
1013 
1014 	if (unlikely(state == TCP_CLOSE))
1015 		return SKB_DROP_REASON_TCP_CLOSE;
1016 
1017 	/* Lookup key based on peer address and keyid.
1018 	 * current_key and rnext_key must not be used on tcp listen
1019 	 * sockets as otherwise:
1020 	 * - request sockets would race on those key pointers
1021 	 * - tcp_ao_del_cmd() allows async key removal
1022 	 */
1023 	key = tcp_ao_inbound_lookup(family, sk, skb, -1, aoh->keyid, l3index);
1024 	if (!key)
1025 		goto key_not_found;
1026 
1027 	if (th->syn && !th->ack)
1028 		goto verify_hash;
1029 
1030 	if ((1 << state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV)) {
1031 		/* Make the initial syn the likely case here */
1032 		if (unlikely(req)) {
1033 			sne = tcp_ao_compute_sne(0, tcp_rsk(req)->rcv_isn,
1034 						 ntohl(th->seq));
1035 			sisn = htonl(tcp_rsk(req)->rcv_isn);
1036 			disn = htonl(tcp_rsk(req)->snt_isn);
1037 		} else if (unlikely(th->ack && !th->syn)) {
1038 			/* Possible syncookie packet */
1039 			sisn = htonl(ntohl(th->seq) - 1);
1040 			disn = htonl(ntohl(th->ack_seq) - 1);
1041 			sne = tcp_ao_compute_sne(0, ntohl(sisn),
1042 						 ntohl(th->seq));
1043 		} else if (unlikely(!th->syn)) {
1044 			/* no way to figure out initial sisn/disn - drop */
1045 			return SKB_DROP_REASON_TCP_FLAGS;
1046 		}
1047 	} else if ((1 << state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1048 		disn = info->lisn;
1049 		if (th->syn || th->rst)
1050 			sisn = th->seq;
1051 		else
1052 			sisn = info->risn;
1053 	} else {
1054 		WARN_ONCE(1, "TCP-AO: Unexpected sk_state %d", state);
1055 		return SKB_DROP_REASON_TCP_AOFAILURE;
1056 	}
1057 verify_hash:
1058 	traffic_key = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC);
1059 	if (!traffic_key)
1060 		return SKB_DROP_REASON_NOT_SPECIFIED;
1061 	tcp_ao_calc_key_skb(key, traffic_key, skb, sisn, disn, family);
1062 	ret = tcp_ao_verify_hash(sk, skb, family, info, aoh, key,
1063 				 traffic_key, phash, sne, l3index);
1064 	kfree(traffic_key);
1065 	return ret;
1066 
1067 key_not_found:
1068 	NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOKEYNOTFOUND);
1069 	atomic64_inc(&info->counters.key_not_found);
1070 	trace_tcp_ao_key_not_found(sk, skb, aoh->keyid,
1071 				   aoh->rnext_keyid, maclen);
1072 	return SKB_DROP_REASON_TCP_AOKEYNOTFOUND;
1073 }
1074 
1075 static int tcp_ao_cache_traffic_keys(const struct sock *sk,
1076 				     struct tcp_ao_info *ao,
1077 				     struct tcp_ao_key *ao_key)
1078 {
1079 	u8 *traffic_key = snd_other_key(ao_key);
1080 	int ret;
1081 
1082 	ret = tcp_ao_calc_key_sk(ao_key, traffic_key, sk,
1083 				 ao->lisn, ao->risn, true);
1084 	if (ret)
1085 		return ret;
1086 
1087 	traffic_key = rcv_other_key(ao_key);
1088 	ret = tcp_ao_calc_key_sk(ao_key, traffic_key, sk,
1089 				 ao->lisn, ao->risn, false);
1090 	return ret;
1091 }
1092 
1093 void tcp_ao_connect_init(struct sock *sk)
1094 {
1095 	struct tcp_sock *tp = tcp_sk(sk);
1096 	struct tcp_ao_info *ao_info;
1097 	struct hlist_node *next;
1098 	union tcp_ao_addr *addr;
1099 	struct tcp_ao_key *key;
1100 	int family, l3index;
1101 
1102 	ao_info = rcu_dereference_protected(tp->ao_info,
1103 					    lockdep_sock_is_held(sk));
1104 	if (!ao_info)
1105 		return;
1106 
1107 	/* Remove all keys that don't match the peer */
1108 	family = sk->sk_family;
1109 	if (family == AF_INET)
1110 		addr = (union tcp_ao_addr *)&sk->sk_daddr;
1111 #if IS_ENABLED(CONFIG_IPV6)
1112 	else if (family == AF_INET6)
1113 		addr = (union tcp_ao_addr *)&sk->sk_v6_daddr;
1114 #endif
1115 	else
1116 		return;
1117 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1118 						 sk->sk_bound_dev_if);
1119 
1120 	hlist_for_each_entry_safe(key, next, &ao_info->head, node) {
1121 		if (!tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1))
1122 			continue;
1123 
1124 		if (key == ao_info->current_key)
1125 			ao_info->current_key = NULL;
1126 		if (key == ao_info->rnext_key)
1127 			ao_info->rnext_key = NULL;
1128 		hlist_del_rcu(&key->node);
1129 		atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1130 		call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1131 	}
1132 
1133 	key = tp->af_specific->ao_lookup(sk, sk, -1, -1);
1134 	if (key) {
1135 		/* if current_key or rnext_key were not provided,
1136 		 * use the first key matching the peer
1137 		 */
1138 		if (!ao_info->current_key)
1139 			ao_info->current_key = key;
1140 		if (!ao_info->rnext_key)
1141 			ao_info->rnext_key = key;
1142 		tp->tcp_header_len += tcp_ao_len_aligned(key);
1143 
1144 		ao_info->lisn = htonl(tp->write_seq);
1145 		ao_info->snd_sne = 0;
1146 	} else {
1147 		/* Can't happen: tcp_connect() verifies that there's
1148 		 * at least one tcp-ao key that matches the remote peer.
1149 		 */
1150 		WARN_ON_ONCE(1);
1151 		rcu_assign_pointer(tp->ao_info, NULL);
1152 		kfree(ao_info);
1153 	}
1154 }
1155 
1156 void tcp_ao_established(struct sock *sk)
1157 {
1158 	struct tcp_ao_info *ao;
1159 	struct tcp_ao_key *key;
1160 
1161 	ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
1162 				       lockdep_sock_is_held(sk));
1163 	if (!ao)
1164 		return;
1165 
1166 	hlist_for_each_entry_rcu(key, &ao->head, node, lockdep_sock_is_held(sk))
1167 		tcp_ao_cache_traffic_keys(sk, ao, key);
1168 }
1169 
1170 void tcp_ao_finish_connect(struct sock *sk, struct sk_buff *skb)
1171 {
1172 	struct tcp_ao_info *ao;
1173 	struct tcp_ao_key *key;
1174 
1175 	ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
1176 				       lockdep_sock_is_held(sk));
1177 	if (!ao)
1178 		return;
1179 
1180 	/* sk with TCP_REPAIR_ON does not have skb in tcp_finish_connect */
1181 	if (skb)
1182 		WRITE_ONCE(ao->risn, tcp_hdr(skb)->seq);
1183 	ao->rcv_sne = 0;
1184 
1185 	hlist_for_each_entry_rcu(key, &ao->head, node, lockdep_sock_is_held(sk))
1186 		tcp_ao_cache_traffic_keys(sk, ao, key);
1187 }
1188 
1189 int tcp_ao_copy_all_matching(const struct sock *sk, struct sock *newsk,
1190 			     struct request_sock *req, struct sk_buff *skb,
1191 			     int family)
1192 {
1193 	struct tcp_ao_key *key, *new_key, *first_key;
1194 	struct tcp_ao_info *new_ao, *ao;
1195 	struct hlist_node *key_head;
1196 	int l3index, ret = -ENOMEM;
1197 	union tcp_ao_addr *addr;
1198 	bool match = false;
1199 
1200 	ao = rcu_dereference(tcp_sk(sk)->ao_info);
1201 	if (!ao)
1202 		return 0;
1203 
1204 	/* New socket without TCP-AO on it */
1205 	if (!tcp_rsk_used_ao(req))
1206 		return 0;
1207 
1208 	new_ao = tcp_ao_alloc_info(GFP_ATOMIC);
1209 	if (!new_ao)
1210 		return -ENOMEM;
1211 	new_ao->lisn = htonl(tcp_rsk(req)->snt_isn);
1212 	new_ao->risn = htonl(tcp_rsk(req)->rcv_isn);
1213 	new_ao->ao_required = ao->ao_required;
1214 	new_ao->accept_icmps = ao->accept_icmps;
1215 
1216 	if (family == AF_INET) {
1217 		addr = (union tcp_ao_addr *)&newsk->sk_daddr;
1218 #if IS_ENABLED(CONFIG_IPV6)
1219 	} else if (family == AF_INET6) {
1220 		addr = (union tcp_ao_addr *)&newsk->sk_v6_daddr;
1221 #endif
1222 	} else {
1223 		ret = -EAFNOSUPPORT;
1224 		goto free_ao;
1225 	}
1226 	l3index = l3mdev_master_ifindex_by_index(sock_net(newsk),
1227 						 newsk->sk_bound_dev_if);
1228 
1229 	hlist_for_each_entry_rcu(key, &ao->head, node) {
1230 		if (tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1))
1231 			continue;
1232 
1233 		new_key = tcp_ao_copy_key(newsk, key);
1234 		if (!new_key)
1235 			goto free_and_exit;
1236 
1237 		tcp_ao_cache_traffic_keys(newsk, new_ao, new_key);
1238 		tcp_ao_link_mkt(new_ao, new_key);
1239 		match = true;
1240 	}
1241 
1242 	if (!match) {
1243 		/* RFC5925 (7.4.1) specifies that the TCP-AO status
1244 		 * of a connection is determined on the initial SYN.
1245 		 * At this point the connection was TCP-AO enabled, so
1246 		 * it can't switch to being unsigned if peer's key
1247 		 * disappears on the listening socket.
1248 		 */
1249 		ret = -EKEYREJECTED;
1250 		goto free_and_exit;
1251 	}
1252 
1253 	if (!static_key_fast_inc_not_disabled(&tcp_ao_needed.key.key)) {
1254 		ret = -EUSERS;
1255 		goto free_and_exit;
1256 	}
1257 
1258 	key_head = rcu_dereference(hlist_first_rcu(&new_ao->head));
1259 	first_key = hlist_entry_safe(key_head, struct tcp_ao_key, node);
1260 
1261 	key = tcp_ao_established_key(req_to_sk(req), new_ao, tcp_rsk(req)->ao_keyid, -1);
1262 	if (key)
1263 		new_ao->current_key = key;
1264 	else
1265 		new_ao->current_key = first_key;
1266 
1267 	/* set rnext_key */
1268 	key = tcp_ao_established_key(req_to_sk(req), new_ao, -1, tcp_rsk(req)->ao_rcv_next);
1269 	if (key)
1270 		new_ao->rnext_key = key;
1271 	else
1272 		new_ao->rnext_key = first_key;
1273 
1274 	sk_gso_disable(newsk);
1275 	rcu_assign_pointer(tcp_sk(newsk)->ao_info, new_ao);
1276 
1277 	return 0;
1278 
1279 free_and_exit:
1280 	hlist_for_each_entry_safe(key, key_head, &new_ao->head, node) {
1281 		hlist_del(&key->node);
1282 		tcp_sigpool_release(key->tcp_sigpool_id);
1283 		atomic_sub(tcp_ao_sizeof_key(key), &newsk->sk_omem_alloc);
1284 		kfree_sensitive(key);
1285 	}
1286 free_ao:
1287 	kfree(new_ao);
1288 	return ret;
1289 }
1290 
1291 static bool tcp_ao_can_set_current_rnext(struct sock *sk)
1292 {
1293 	/* There aren't current/rnext keys on TCP_LISTEN sockets */
1294 	if (sk->sk_state == TCP_LISTEN)
1295 		return false;
1296 	return true;
1297 }
1298 
1299 static int tcp_ao_verify_ipv4(struct sock *sk, struct tcp_ao_add *cmd,
1300 			      union tcp_ao_addr **addr)
1301 {
1302 	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd->addr;
1303 	struct inet_sock *inet = inet_sk(sk);
1304 
1305 	if (sin->sin_family != AF_INET)
1306 		return -EINVAL;
1307 
1308 	/* Currently matching is not performed on port (or port ranges) */
1309 	if (sin->sin_port != 0)
1310 		return -EINVAL;
1311 
1312 	/* Check prefix and trailing 0's in addr */
1313 	if (cmd->prefix != 0) {
1314 		__be32 mask;
1315 
1316 		if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY)
1317 			return -EINVAL;
1318 		if (cmd->prefix > 32)
1319 			return -EINVAL;
1320 
1321 		mask = inet_make_mask(cmd->prefix);
1322 		if (sin->sin_addr.s_addr & ~mask)
1323 			return -EINVAL;
1324 
1325 		/* Check that MKT address is consistent with socket */
1326 		if (ntohl(inet->inet_daddr) != INADDR_ANY &&
1327 		    (inet->inet_daddr & mask) != sin->sin_addr.s_addr)
1328 			return -EINVAL;
1329 	} else {
1330 		if (ntohl(sin->sin_addr.s_addr) != INADDR_ANY)
1331 			return -EINVAL;
1332 	}
1333 
1334 	*addr = (union tcp_ao_addr *)&sin->sin_addr;
1335 	return 0;
1336 }
1337 
1338 static int tcp_ao_parse_crypto(struct tcp_ao_add *cmd, struct tcp_ao_key *key)
1339 {
1340 	unsigned int syn_tcp_option_space;
1341 	bool is_kdf_aes_128_cmac = false;
1342 	struct crypto_ahash *tfm;
1343 	struct tcp_sigpool hp;
1344 	void *tmp_key = NULL;
1345 	int err;
1346 
1347 	/* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1348 	if (!strcmp("cmac(aes128)", cmd->alg_name)) {
1349 		strscpy(cmd->alg_name, "cmac(aes)", sizeof(cmd->alg_name));
1350 		is_kdf_aes_128_cmac = (cmd->keylen != 16);
1351 		tmp_key = kmalloc(cmd->keylen, GFP_KERNEL);
1352 		if (!tmp_key)
1353 			return -ENOMEM;
1354 	}
1355 
1356 	key->maclen = cmd->maclen ?: 12; /* 12 is the default in RFC5925 */
1357 
1358 	/* Check: maclen + tcp-ao header <= (MAX_TCP_OPTION_SPACE - mss
1359 	 *					- tstamp (including sackperm)
1360 	 *					- wscale),
1361 	 * see tcp_syn_options(), tcp_synack_options(), commit 33ad798c924b.
1362 	 *
1363 	 * In order to allow D-SACK with TCP-AO, the header size should be:
1364 	 * (MAX_TCP_OPTION_SPACE - TCPOLEN_TSTAMP_ALIGNED
1365 	 *			- TCPOLEN_SACK_BASE_ALIGNED
1366 	 *			- 2 * TCPOLEN_SACK_PERBLOCK) = 8 (maclen = 4),
1367 	 * see tcp_established_options().
1368 	 *
1369 	 * RFC5925, 2.2:
1370 	 * Typical MACs are 96-128 bits (12-16 bytes), but any length
1371 	 * that fits in the header of the segment being authenticated
1372 	 * is allowed.
1373 	 *
1374 	 * RFC5925, 7.6:
1375 	 * TCP-AO continues to consume 16 bytes in non-SYN segments,
1376 	 * leaving a total of 24 bytes for other options, of which
1377 	 * the timestamp consumes 10.  This leaves 14 bytes, of which 10
1378 	 * are used for a single SACK block. When two SACK blocks are used,
1379 	 * such as to handle D-SACK, a smaller TCP-AO MAC would be required
1380 	 * to make room for the additional SACK block (i.e., to leave 18
1381 	 * bytes for the D-SACK variant of the SACK option) [RFC2883].
1382 	 * Note that D-SACK is not supportable in TCP MD5 in the presence
1383 	 * of timestamps, because TCP MD5’s MAC length is fixed and too
1384 	 * large to leave sufficient option space.
1385 	 */
1386 	syn_tcp_option_space = MAX_TCP_OPTION_SPACE;
1387 	syn_tcp_option_space -= TCPOLEN_MSS_ALIGNED;
1388 	syn_tcp_option_space -= TCPOLEN_TSTAMP_ALIGNED;
1389 	syn_tcp_option_space -= TCPOLEN_WSCALE_ALIGNED;
1390 	if (tcp_ao_len_aligned(key) > syn_tcp_option_space) {
1391 		err = -EMSGSIZE;
1392 		goto err_kfree;
1393 	}
1394 
1395 	key->keylen = cmd->keylen;
1396 	memcpy(key->key, cmd->key, cmd->keylen);
1397 
1398 	err = tcp_sigpool_start(key->tcp_sigpool_id, &hp);
1399 	if (err)
1400 		goto err_kfree;
1401 
1402 	tfm = crypto_ahash_reqtfm(hp.req);
1403 	if (is_kdf_aes_128_cmac) {
1404 		void *scratch = hp.scratch;
1405 		struct scatterlist sg;
1406 
1407 		memcpy(tmp_key, cmd->key, cmd->keylen);
1408 		sg_init_one(&sg, tmp_key, cmd->keylen);
1409 
1410 		/* Using zero-key of 16 bytes as described in RFC5926 */
1411 		memset(scratch, 0, 16);
1412 		err = crypto_ahash_setkey(tfm, scratch, 16);
1413 		if (err)
1414 			goto err_pool_end;
1415 
1416 		err = crypto_ahash_init(hp.req);
1417 		if (err)
1418 			goto err_pool_end;
1419 
1420 		ahash_request_set_crypt(hp.req, &sg, key->key, cmd->keylen);
1421 		err = crypto_ahash_update(hp.req);
1422 		if (err)
1423 			goto err_pool_end;
1424 
1425 		err |= crypto_ahash_final(hp.req);
1426 		if (err)
1427 			goto err_pool_end;
1428 		key->keylen = 16;
1429 	}
1430 
1431 	err = crypto_ahash_setkey(tfm, key->key, key->keylen);
1432 	if (err)
1433 		goto err_pool_end;
1434 
1435 	tcp_sigpool_end(&hp);
1436 	kfree_sensitive(tmp_key);
1437 
1438 	if (tcp_ao_maclen(key) > key->digest_size)
1439 		return -EINVAL;
1440 
1441 	return 0;
1442 
1443 err_pool_end:
1444 	tcp_sigpool_end(&hp);
1445 err_kfree:
1446 	kfree_sensitive(tmp_key);
1447 	return err;
1448 }
1449 
1450 #if IS_ENABLED(CONFIG_IPV6)
1451 static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1452 			      union tcp_ao_addr **paddr,
1453 			      unsigned short int *family)
1454 {
1455 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd->addr;
1456 	struct in6_addr *addr = &sin6->sin6_addr;
1457 	u8 prefix = cmd->prefix;
1458 
1459 	if (sin6->sin6_family != AF_INET6)
1460 		return -EINVAL;
1461 
1462 	/* Currently matching is not performed on port (or port ranges) */
1463 	if (sin6->sin6_port != 0)
1464 		return -EINVAL;
1465 
1466 	/* Check prefix and trailing 0's in addr */
1467 	if (cmd->prefix != 0 && ipv6_addr_v4mapped(addr)) {
1468 		__be32 addr4 = addr->s6_addr32[3];
1469 		__be32 mask;
1470 
1471 		if (prefix > 32 || ntohl(addr4) == INADDR_ANY)
1472 			return -EINVAL;
1473 
1474 		mask = inet_make_mask(prefix);
1475 		if (addr4 & ~mask)
1476 			return -EINVAL;
1477 
1478 		/* Check that MKT address is consistent with socket */
1479 		if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
1480 			__be32 daddr4 = sk->sk_v6_daddr.s6_addr32[3];
1481 
1482 			if (!ipv6_addr_v4mapped(&sk->sk_v6_daddr))
1483 				return -EINVAL;
1484 			if ((daddr4 & mask) != addr4)
1485 				return -EINVAL;
1486 		}
1487 
1488 		*paddr = (union tcp_ao_addr *)&addr->s6_addr32[3];
1489 		*family = AF_INET;
1490 		return 0;
1491 	} else if (cmd->prefix != 0) {
1492 		struct in6_addr pfx;
1493 
1494 		if (ipv6_addr_any(addr) || prefix > 128)
1495 			return -EINVAL;
1496 
1497 		ipv6_addr_prefix(&pfx, addr, prefix);
1498 		if (ipv6_addr_cmp(&pfx, addr))
1499 			return -EINVAL;
1500 
1501 		/* Check that MKT address is consistent with socket */
1502 		if (!ipv6_addr_any(&sk->sk_v6_daddr) &&
1503 		    !ipv6_prefix_equal(&sk->sk_v6_daddr, addr, prefix))
1504 
1505 			return -EINVAL;
1506 	} else {
1507 		if (!ipv6_addr_any(addr))
1508 			return -EINVAL;
1509 	}
1510 
1511 	*paddr = (union tcp_ao_addr *)addr;
1512 	return 0;
1513 }
1514 #else
1515 static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1516 			      union tcp_ao_addr **paddr,
1517 			      unsigned short int *family)
1518 {
1519 	return -EOPNOTSUPP;
1520 }
1521 #endif
1522 
1523 static struct tcp_ao_info *setsockopt_ao_info(struct sock *sk)
1524 {
1525 	if (sk_fullsock(sk)) {
1526 		return rcu_dereference_protected(tcp_sk(sk)->ao_info,
1527 						 lockdep_sock_is_held(sk));
1528 	} else if (sk->sk_state == TCP_TIME_WAIT) {
1529 		return rcu_dereference_protected(tcp_twsk(sk)->ao_info,
1530 						 lockdep_sock_is_held(sk));
1531 	}
1532 	return ERR_PTR(-ESOCKTNOSUPPORT);
1533 }
1534 
1535 static struct tcp_ao_info *getsockopt_ao_info(struct sock *sk)
1536 {
1537 	if (sk_fullsock(sk))
1538 		return rcu_dereference(tcp_sk(sk)->ao_info);
1539 	else if (sk->sk_state == TCP_TIME_WAIT)
1540 		return rcu_dereference(tcp_twsk(sk)->ao_info);
1541 
1542 	return ERR_PTR(-ESOCKTNOSUPPORT);
1543 }
1544 
1545 #define TCP_AO_KEYF_ALL (TCP_AO_KEYF_IFINDEX | TCP_AO_KEYF_EXCLUDE_OPT)
1546 #define TCP_AO_GET_KEYF_VALID	(TCP_AO_KEYF_IFINDEX)
1547 
1548 static struct tcp_ao_key *tcp_ao_key_alloc(struct sock *sk,
1549 					   struct tcp_ao_add *cmd)
1550 {
1551 	const char *algo = cmd->alg_name;
1552 	unsigned int digest_size;
1553 	struct crypto_ahash *tfm;
1554 	struct tcp_ao_key *key;
1555 	struct tcp_sigpool hp;
1556 	int err, pool_id;
1557 	size_t size;
1558 
1559 	/* Force null-termination of alg_name */
1560 	cmd->alg_name[ARRAY_SIZE(cmd->alg_name) - 1] = '\0';
1561 
1562 	/* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1563 	if (!strcmp("cmac(aes128)", algo))
1564 		algo = "cmac(aes)";
1565 
1566 	/* Full TCP header (th->doff << 2) should fit into scratch area,
1567 	 * see tcp_ao_hash_header().
1568 	 */
1569 	pool_id = tcp_sigpool_alloc_ahash(algo, 60);
1570 	if (pool_id < 0)
1571 		return ERR_PTR(pool_id);
1572 
1573 	err = tcp_sigpool_start(pool_id, &hp);
1574 	if (err)
1575 		goto err_free_pool;
1576 
1577 	tfm = crypto_ahash_reqtfm(hp.req);
1578 	digest_size = crypto_ahash_digestsize(tfm);
1579 	tcp_sigpool_end(&hp);
1580 
1581 	size = sizeof(struct tcp_ao_key) + (digest_size << 1);
1582 	key = sock_kmalloc(sk, size, GFP_KERNEL);
1583 	if (!key) {
1584 		err = -ENOMEM;
1585 		goto err_free_pool;
1586 	}
1587 
1588 	key->tcp_sigpool_id = pool_id;
1589 	key->digest_size = digest_size;
1590 	return key;
1591 
1592 err_free_pool:
1593 	tcp_sigpool_release(pool_id);
1594 	return ERR_PTR(err);
1595 }
1596 
1597 static int tcp_ao_add_cmd(struct sock *sk, unsigned short int family,
1598 			  sockptr_t optval, int optlen)
1599 {
1600 	struct tcp_ao_info *ao_info;
1601 	union tcp_ao_addr *addr;
1602 	struct tcp_ao_key *key;
1603 	struct tcp_ao_add cmd;
1604 	int ret, l3index = 0;
1605 	bool first = false;
1606 
1607 	if (optlen < sizeof(cmd))
1608 		return -EINVAL;
1609 
1610 	ret = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1611 	if (ret)
1612 		return ret;
1613 
1614 	if (cmd.keylen > TCP_AO_MAXKEYLEN)
1615 		return -EINVAL;
1616 
1617 	if (cmd.reserved != 0 || cmd.reserved2 != 0)
1618 		return -EINVAL;
1619 
1620 	if (family == AF_INET)
1621 		ret = tcp_ao_verify_ipv4(sk, &cmd, &addr);
1622 	else
1623 		ret = tcp_ao_verify_ipv6(sk, &cmd, &addr, &family);
1624 	if (ret)
1625 		return ret;
1626 
1627 	if (cmd.keyflags & ~TCP_AO_KEYF_ALL)
1628 		return -EINVAL;
1629 
1630 	if (cmd.set_current || cmd.set_rnext) {
1631 		if (!tcp_ao_can_set_current_rnext(sk))
1632 			return -EINVAL;
1633 	}
1634 
1635 	if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1636 		return -EINVAL;
1637 
1638 	/* For cmd.tcp_ifindex = 0 the key will apply to the default VRF */
1639 	if (cmd.keyflags & TCP_AO_KEYF_IFINDEX && cmd.ifindex) {
1640 		int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
1641 		struct net_device *dev;
1642 
1643 		rcu_read_lock();
1644 		dev = dev_get_by_index_rcu(sock_net(sk), cmd.ifindex);
1645 		if (dev && netif_is_l3_master(dev))
1646 			l3index = dev->ifindex;
1647 		rcu_read_unlock();
1648 
1649 		if (!dev || !l3index)
1650 			return -EINVAL;
1651 
1652 		if (!bound_dev_if || bound_dev_if != cmd.ifindex) {
1653 			/* tcp_ao_established_key() doesn't expect having
1654 			 * non peer-matching key on an established TCP-AO
1655 			 * connection.
1656 			 */
1657 			if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1658 				return -EINVAL;
1659 		}
1660 
1661 		/* It's still possible to bind after adding keys or even
1662 		 * re-bind to a different dev (with CAP_NET_RAW).
1663 		 * So, no reason to return error here, rather try to be
1664 		 * nice and warn the user.
1665 		 */
1666 		if (bound_dev_if && bound_dev_if != cmd.ifindex)
1667 			net_warn_ratelimited("AO key ifindex %d != sk bound ifindex %d\n",
1668 					     cmd.ifindex, bound_dev_if);
1669 	}
1670 
1671 	/* Don't allow keys for peers that have a matching TCP-MD5 key */
1672 	if (cmd.keyflags & TCP_AO_KEYF_IFINDEX) {
1673 		/* Non-_exact version of tcp_md5_do_lookup() will
1674 		 * as well match keys that aren't bound to a specific VRF
1675 		 * (that will make them match AO key with
1676 		 * sysctl_tcp_l3dev_accept = 1
1677 		 */
1678 		if (tcp_md5_do_lookup(sk, l3index, addr, family))
1679 			return -EKEYREJECTED;
1680 	} else {
1681 		if (tcp_md5_do_lookup_any_l3index(sk, addr, family))
1682 			return -EKEYREJECTED;
1683 	}
1684 
1685 	ao_info = setsockopt_ao_info(sk);
1686 	if (IS_ERR(ao_info))
1687 		return PTR_ERR(ao_info);
1688 
1689 	if (!ao_info) {
1690 		ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1691 		if (!ao_info)
1692 			return -ENOMEM;
1693 		first = true;
1694 	} else {
1695 		/* Check that neither RecvID nor SendID match any
1696 		 * existing key for the peer, RFC5925 3.1:
1697 		 * > The IDs of MKTs MUST NOT overlap where their
1698 		 * > TCP connection identifiers overlap.
1699 		 */
1700 		if (__tcp_ao_do_lookup(sk, l3index, addr, family, cmd.prefix, -1, cmd.rcvid))
1701 			return -EEXIST;
1702 		if (__tcp_ao_do_lookup(sk, l3index, addr, family,
1703 				       cmd.prefix, cmd.sndid, -1))
1704 			return -EEXIST;
1705 	}
1706 
1707 	key = tcp_ao_key_alloc(sk, &cmd);
1708 	if (IS_ERR(key)) {
1709 		ret = PTR_ERR(key);
1710 		goto err_free_ao;
1711 	}
1712 
1713 	INIT_HLIST_NODE(&key->node);
1714 	memcpy(&key->addr, addr, (family == AF_INET) ? sizeof(struct in_addr) :
1715 						       sizeof(struct in6_addr));
1716 	key->prefixlen	= cmd.prefix;
1717 	key->family	= family;
1718 	key->keyflags	= cmd.keyflags;
1719 	key->sndid	= cmd.sndid;
1720 	key->rcvid	= cmd.rcvid;
1721 	key->l3index	= l3index;
1722 	atomic64_set(&key->pkt_good, 0);
1723 	atomic64_set(&key->pkt_bad, 0);
1724 
1725 	ret = tcp_ao_parse_crypto(&cmd, key);
1726 	if (ret < 0)
1727 		goto err_free_sock;
1728 
1729 	if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) {
1730 		tcp_ao_cache_traffic_keys(sk, ao_info, key);
1731 		if (first) {
1732 			ao_info->current_key = key;
1733 			ao_info->rnext_key = key;
1734 		}
1735 	}
1736 
1737 	tcp_ao_link_mkt(ao_info, key);
1738 	if (first) {
1739 		if (!static_branch_inc(&tcp_ao_needed.key)) {
1740 			ret = -EUSERS;
1741 			goto err_free_sock;
1742 		}
1743 		sk_gso_disable(sk);
1744 		rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
1745 	}
1746 
1747 	if (cmd.set_current)
1748 		WRITE_ONCE(ao_info->current_key, key);
1749 	if (cmd.set_rnext)
1750 		WRITE_ONCE(ao_info->rnext_key, key);
1751 	return 0;
1752 
1753 err_free_sock:
1754 	atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1755 	tcp_sigpool_release(key->tcp_sigpool_id);
1756 	kfree_sensitive(key);
1757 err_free_ao:
1758 	if (first)
1759 		kfree(ao_info);
1760 	return ret;
1761 }
1762 
1763 static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info,
1764 			     bool del_async, struct tcp_ao_key *key,
1765 			     struct tcp_ao_key *new_current,
1766 			     struct tcp_ao_key *new_rnext)
1767 {
1768 	int err;
1769 
1770 	hlist_del_rcu(&key->node);
1771 
1772 	/* Support for async delete on listening sockets: as they don't
1773 	 * need current_key/rnext_key maintaining, we don't need to check
1774 	 * them and we can just free all resources in RCU fashion.
1775 	 */
1776 	if (del_async) {
1777 		atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1778 		call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1779 		return 0;
1780 	}
1781 
1782 	/* At this moment another CPU could have looked this key up
1783 	 * while it was unlinked from the list. Wait for RCU grace period,
1784 	 * after which the key is off-list and can't be looked up again;
1785 	 * the rx path [just before RCU came] might have used it and set it
1786 	 * as current_key (very unlikely).
1787 	 * Free the key with next RCU grace period (in case it was
1788 	 * current_key before tcp_ao_current_rnext() might have
1789 	 * changed it in forced-delete).
1790 	 */
1791 	synchronize_rcu();
1792 	if (new_current)
1793 		WRITE_ONCE(ao_info->current_key, new_current);
1794 	if (new_rnext)
1795 		WRITE_ONCE(ao_info->rnext_key, new_rnext);
1796 
1797 	if (unlikely(READ_ONCE(ao_info->current_key) == key ||
1798 		     READ_ONCE(ao_info->rnext_key) == key)) {
1799 		err = -EBUSY;
1800 		goto add_key;
1801 	}
1802 
1803 	atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1804 	call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1805 
1806 	return 0;
1807 add_key:
1808 	hlist_add_head_rcu(&key->node, &ao_info->head);
1809 	return err;
1810 }
1811 
1812 #define TCP_AO_DEL_KEYF_ALL (TCP_AO_KEYF_IFINDEX)
1813 static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family,
1814 			  sockptr_t optval, int optlen)
1815 {
1816 	struct tcp_ao_key *key, *new_current = NULL, *new_rnext = NULL;
1817 	int err, addr_len, l3index = 0;
1818 	struct tcp_ao_info *ao_info;
1819 	union tcp_ao_addr *addr;
1820 	struct tcp_ao_del cmd;
1821 	__u8 prefix;
1822 	u16 port;
1823 
1824 	if (optlen < sizeof(cmd))
1825 		return -EINVAL;
1826 
1827 	err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1828 	if (err)
1829 		return err;
1830 
1831 	if (cmd.reserved != 0 || cmd.reserved2 != 0)
1832 		return -EINVAL;
1833 
1834 	if (cmd.set_current || cmd.set_rnext) {
1835 		if (!tcp_ao_can_set_current_rnext(sk))
1836 			return -EINVAL;
1837 	}
1838 
1839 	if (cmd.keyflags & ~TCP_AO_DEL_KEYF_ALL)
1840 		return -EINVAL;
1841 
1842 	/* No sanity check for TCP_AO_KEYF_IFINDEX as if a VRF
1843 	 * was destroyed, there still should be a way to delete keys,
1844 	 * that were bound to that l3intf. So, fail late at lookup stage
1845 	 * if there is no key for that ifindex.
1846 	 */
1847 	if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1848 		return -EINVAL;
1849 
1850 	ao_info = setsockopt_ao_info(sk);
1851 	if (IS_ERR(ao_info))
1852 		return PTR_ERR(ao_info);
1853 	if (!ao_info)
1854 		return -ENOENT;
1855 
1856 	/* For sockets in TCP_CLOSED it's possible set keys that aren't
1857 	 * matching the future peer (address/VRF/etc),
1858 	 * tcp_ao_connect_init() will choose a correct matching MKT
1859 	 * if there's any.
1860 	 */
1861 	if (cmd.set_current) {
1862 		new_current = tcp_ao_established_key(sk, ao_info, cmd.current_key, -1);
1863 		if (!new_current)
1864 			return -ENOENT;
1865 	}
1866 	if (cmd.set_rnext) {
1867 		new_rnext = tcp_ao_established_key(sk, ao_info, -1, cmd.rnext);
1868 		if (!new_rnext)
1869 			return -ENOENT;
1870 	}
1871 	if (cmd.del_async && sk->sk_state != TCP_LISTEN)
1872 		return -EINVAL;
1873 
1874 	if (family == AF_INET) {
1875 		struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr;
1876 
1877 		addr = (union tcp_ao_addr *)&sin->sin_addr;
1878 		addr_len = sizeof(struct in_addr);
1879 		port = ntohs(sin->sin_port);
1880 	} else {
1881 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.addr;
1882 		struct in6_addr *addr6 = &sin6->sin6_addr;
1883 
1884 		if (ipv6_addr_v4mapped(addr6)) {
1885 			addr = (union tcp_ao_addr *)&addr6->s6_addr32[3];
1886 			addr_len = sizeof(struct in_addr);
1887 			family = AF_INET;
1888 		} else {
1889 			addr = (union tcp_ao_addr *)addr6;
1890 			addr_len = sizeof(struct in6_addr);
1891 		}
1892 		port = ntohs(sin6->sin6_port);
1893 	}
1894 	prefix = cmd.prefix;
1895 
1896 	/* Currently matching is not performed on port (or port ranges) */
1897 	if (port != 0)
1898 		return -EINVAL;
1899 
1900 	/* We could choose random present key here for current/rnext
1901 	 * but that's less predictable. Let's be strict and don't
1902 	 * allow removing a key that's in use. RFC5925 doesn't
1903 	 * specify how-to coordinate key removal, but says:
1904 	 * "It is presumed that an MKT affecting a particular
1905 	 * connection cannot be destroyed during an active connection"
1906 	 */
1907 	hlist_for_each_entry_rcu(key, &ao_info->head, node,
1908 				 lockdep_sock_is_held(sk)) {
1909 		if (cmd.sndid != key->sndid ||
1910 		    cmd.rcvid != key->rcvid)
1911 			continue;
1912 
1913 		if (family != key->family ||
1914 		    prefix != key->prefixlen ||
1915 		    memcmp(addr, &key->addr, addr_len))
1916 			continue;
1917 
1918 		if ((cmd.keyflags & TCP_AO_KEYF_IFINDEX) !=
1919 		    (key->keyflags & TCP_AO_KEYF_IFINDEX))
1920 			continue;
1921 
1922 		if (key->l3index != l3index)
1923 			continue;
1924 
1925 		if (key == new_current || key == new_rnext)
1926 			continue;
1927 
1928 		return tcp_ao_delete_key(sk, ao_info, cmd.del_async, key,
1929 					 new_current, new_rnext);
1930 	}
1931 	return -ENOENT;
1932 }
1933 
1934 /* cmd.ao_required makes a socket TCP-AO only.
1935  * Don't allow any md5 keys for any l3intf on the socket together with it.
1936  * Restricting it early in setsockopt() removes a check for
1937  * ao_info->ao_required on inbound tcp segment fast-path.
1938  */
1939 static int tcp_ao_required_verify(struct sock *sk)
1940 {
1941 #ifdef CONFIG_TCP_MD5SIG
1942 	const struct tcp_md5sig_info *md5sig;
1943 
1944 	if (!static_branch_unlikely(&tcp_md5_needed.key))
1945 		return 0;
1946 
1947 	md5sig = rcu_dereference_check(tcp_sk(sk)->md5sig_info,
1948 				       lockdep_sock_is_held(sk));
1949 	if (!md5sig)
1950 		return 0;
1951 
1952 	if (rcu_dereference_check(hlist_first_rcu(&md5sig->head),
1953 				  lockdep_sock_is_held(sk)))
1954 		return 1;
1955 #endif
1956 	return 0;
1957 }
1958 
1959 static int tcp_ao_info_cmd(struct sock *sk, unsigned short int family,
1960 			   sockptr_t optval, int optlen)
1961 {
1962 	struct tcp_ao_key *new_current = NULL, *new_rnext = NULL;
1963 	struct tcp_ao_info *ao_info;
1964 	struct tcp_ao_info_opt cmd;
1965 	bool first = false;
1966 	int err;
1967 
1968 	if (optlen < sizeof(cmd))
1969 		return -EINVAL;
1970 
1971 	err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1972 	if (err)
1973 		return err;
1974 
1975 	if (cmd.set_current || cmd.set_rnext) {
1976 		if (!tcp_ao_can_set_current_rnext(sk))
1977 			return -EINVAL;
1978 	}
1979 
1980 	if (cmd.reserved != 0 || cmd.reserved2 != 0)
1981 		return -EINVAL;
1982 
1983 	ao_info = setsockopt_ao_info(sk);
1984 	if (IS_ERR(ao_info))
1985 		return PTR_ERR(ao_info);
1986 	if (!ao_info) {
1987 		if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1988 			return -EINVAL;
1989 		ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1990 		if (!ao_info)
1991 			return -ENOMEM;
1992 		first = true;
1993 	}
1994 
1995 	if (cmd.ao_required && tcp_ao_required_verify(sk)) {
1996 		err = -EKEYREJECTED;
1997 		goto out;
1998 	}
1999 
2000 	/* For sockets in TCP_CLOSED it's possible set keys that aren't
2001 	 * matching the future peer (address/port/VRF/etc),
2002 	 * tcp_ao_connect_init() will choose a correct matching MKT
2003 	 * if there's any.
2004 	 */
2005 	if (cmd.set_current) {
2006 		new_current = tcp_ao_established_key(sk, ao_info, cmd.current_key, -1);
2007 		if (!new_current) {
2008 			err = -ENOENT;
2009 			goto out;
2010 		}
2011 	}
2012 	if (cmd.set_rnext) {
2013 		new_rnext = tcp_ao_established_key(sk, ao_info, -1, cmd.rnext);
2014 		if (!new_rnext) {
2015 			err = -ENOENT;
2016 			goto out;
2017 		}
2018 	}
2019 	if (cmd.set_counters) {
2020 		atomic64_set(&ao_info->counters.pkt_good, cmd.pkt_good);
2021 		atomic64_set(&ao_info->counters.pkt_bad, cmd.pkt_bad);
2022 		atomic64_set(&ao_info->counters.key_not_found, cmd.pkt_key_not_found);
2023 		atomic64_set(&ao_info->counters.ao_required, cmd.pkt_ao_required);
2024 		atomic64_set(&ao_info->counters.dropped_icmp, cmd.pkt_dropped_icmp);
2025 	}
2026 
2027 	ao_info->ao_required = cmd.ao_required;
2028 	ao_info->accept_icmps = cmd.accept_icmps;
2029 	if (new_current)
2030 		WRITE_ONCE(ao_info->current_key, new_current);
2031 	if (new_rnext)
2032 		WRITE_ONCE(ao_info->rnext_key, new_rnext);
2033 	if (first) {
2034 		if (!static_branch_inc(&tcp_ao_needed.key)) {
2035 			err = -EUSERS;
2036 			goto out;
2037 		}
2038 		sk_gso_disable(sk);
2039 		rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
2040 	}
2041 	return 0;
2042 out:
2043 	if (first)
2044 		kfree(ao_info);
2045 	return err;
2046 }
2047 
2048 int tcp_parse_ao(struct sock *sk, int cmd, unsigned short int family,
2049 		 sockptr_t optval, int optlen)
2050 {
2051 	if (WARN_ON_ONCE(family != AF_INET && family != AF_INET6))
2052 		return -EAFNOSUPPORT;
2053 
2054 	switch (cmd) {
2055 	case TCP_AO_ADD_KEY:
2056 		return tcp_ao_add_cmd(sk, family, optval, optlen);
2057 	case TCP_AO_DEL_KEY:
2058 		return tcp_ao_del_cmd(sk, family, optval, optlen);
2059 	case TCP_AO_INFO:
2060 		return tcp_ao_info_cmd(sk, family, optval, optlen);
2061 	default:
2062 		WARN_ON_ONCE(1);
2063 		return -EINVAL;
2064 	}
2065 }
2066 
2067 int tcp_v4_parse_ao(struct sock *sk, int cmd, sockptr_t optval, int optlen)
2068 {
2069 	return tcp_parse_ao(sk, cmd, AF_INET, optval, optlen);
2070 }
2071 
2072 /* tcp_ao_copy_mkts_to_user(ao_info, optval, optlen)
2073  *
2074  * @ao_info:	struct tcp_ao_info on the socket that
2075  *		socket getsockopt(TCP_AO_GET_KEYS) is executed on
2076  * @optval:	pointer to array of tcp_ao_getsockopt structures in user space.
2077  *		Must be != NULL.
2078  * @optlen:	pointer to size of tcp_ao_getsockopt structure.
2079  *		Must be != NULL.
2080  *
2081  * Return value: 0 on success, a negative error number otherwise.
2082  *
2083  * optval points to an array of tcp_ao_getsockopt structures in user space.
2084  * optval[0] is used as both input and output to getsockopt. It determines
2085  * which keys are returned by the kernel.
2086  * optval[0].nkeys is the size of the array in user space. On return it contains
2087  * the number of keys matching the search criteria.
2088  * If tcp_ao_getsockopt::get_all is set, then all keys in the socket are
2089  * returned, otherwise only keys matching <addr, prefix, sndid, rcvid>
2090  * in optval[0] are returned.
2091  * optlen is also used as both input and output. The user provides the size
2092  * of struct tcp_ao_getsockopt in user space, and the kernel returns the size
2093  * of the structure in kernel space.
2094  * The size of struct tcp_ao_getsockopt may differ between user and kernel.
2095  * There are three cases to consider:
2096  *  * If usize == ksize, then keys are copied verbatim.
2097  *  * If usize < ksize, then the userspace has passed an old struct to a
2098  *    newer kernel. The rest of the trailing bytes in optval[0]
2099  *    (ksize - usize) are interpreted as 0 by the kernel.
2100  *  * If usize > ksize, then the userspace has passed a new struct to an
2101  *    older kernel. The trailing bytes unknown to the kernel (usize - ksize)
2102  *    are checked to ensure they are zeroed, otherwise -E2BIG is returned.
2103  * On return the kernel fills in min(usize, ksize) in each entry of the array.
2104  * The layout of the fields in the user and kernel structures is expected to
2105  * be the same (including in the 32bit vs 64bit case).
2106  */
2107 static int tcp_ao_copy_mkts_to_user(const struct sock *sk,
2108 				    struct tcp_ao_info *ao_info,
2109 				    sockptr_t optval, sockptr_t optlen)
2110 {
2111 	struct tcp_ao_getsockopt opt_in, opt_out;
2112 	struct tcp_ao_key *key, *current_key;
2113 	bool do_address_matching = true;
2114 	union tcp_ao_addr *addr = NULL;
2115 	int err, l3index, user_len;
2116 	unsigned int max_keys;	/* maximum number of keys to copy to user */
2117 	size_t out_offset = 0;
2118 	size_t bytes_to_write;	/* number of bytes to write to user level */
2119 	u32 matched_keys;	/* keys from ao_info matched so far */
2120 	int optlen_out;
2121 	__be16 port = 0;
2122 
2123 	if (copy_from_sockptr(&user_len, optlen, sizeof(int)))
2124 		return -EFAULT;
2125 
2126 	if (user_len <= 0)
2127 		return -EINVAL;
2128 
2129 	memset(&opt_in, 0, sizeof(struct tcp_ao_getsockopt));
2130 	err = copy_struct_from_sockptr(&opt_in, sizeof(opt_in),
2131 				       optval, user_len);
2132 	if (err < 0)
2133 		return err;
2134 
2135 	if (opt_in.pkt_good || opt_in.pkt_bad)
2136 		return -EINVAL;
2137 	if (opt_in.keyflags & ~TCP_AO_GET_KEYF_VALID)
2138 		return -EINVAL;
2139 	if (opt_in.ifindex && !(opt_in.keyflags & TCP_AO_KEYF_IFINDEX))
2140 		return -EINVAL;
2141 
2142 	if (opt_in.reserved != 0)
2143 		return -EINVAL;
2144 
2145 	max_keys = opt_in.nkeys;
2146 	l3index = (opt_in.keyflags & TCP_AO_KEYF_IFINDEX) ? opt_in.ifindex : -1;
2147 
2148 	if (opt_in.get_all || opt_in.is_current || opt_in.is_rnext) {
2149 		if (opt_in.get_all && (opt_in.is_current || opt_in.is_rnext))
2150 			return -EINVAL;
2151 		do_address_matching = false;
2152 	}
2153 
2154 	switch (opt_in.addr.ss_family) {
2155 	case AF_INET: {
2156 		struct sockaddr_in *sin;
2157 		__be32 mask;
2158 
2159 		sin = (struct sockaddr_in *)&opt_in.addr;
2160 		port = sin->sin_port;
2161 		addr = (union tcp_ao_addr *)&sin->sin_addr;
2162 
2163 		if (opt_in.prefix > 32)
2164 			return -EINVAL;
2165 
2166 		if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY &&
2167 		    opt_in.prefix != 0)
2168 			return -EINVAL;
2169 
2170 		mask = inet_make_mask(opt_in.prefix);
2171 		if (sin->sin_addr.s_addr & ~mask)
2172 			return -EINVAL;
2173 
2174 		break;
2175 	}
2176 	case AF_INET6: {
2177 		struct sockaddr_in6 *sin6;
2178 		struct in6_addr *addr6;
2179 
2180 		sin6 = (struct sockaddr_in6 *)&opt_in.addr;
2181 		addr = (union tcp_ao_addr *)&sin6->sin6_addr;
2182 		addr6 = &sin6->sin6_addr;
2183 		port = sin6->sin6_port;
2184 
2185 		/* We don't have to change family and @addr here if
2186 		 * ipv6_addr_v4mapped() like in key adding:
2187 		 * tcp_ao_key_cmp() does it. Do the sanity checks though.
2188 		 */
2189 		if (opt_in.prefix != 0) {
2190 			if (ipv6_addr_v4mapped(addr6)) {
2191 				__be32 mask, addr4 = addr6->s6_addr32[3];
2192 
2193 				if (opt_in.prefix > 32 ||
2194 				    ntohl(addr4) == INADDR_ANY)
2195 					return -EINVAL;
2196 				mask = inet_make_mask(opt_in.prefix);
2197 				if (addr4 & ~mask)
2198 					return -EINVAL;
2199 			} else {
2200 				struct in6_addr pfx;
2201 
2202 				if (ipv6_addr_any(addr6) ||
2203 				    opt_in.prefix > 128)
2204 					return -EINVAL;
2205 
2206 				ipv6_addr_prefix(&pfx, addr6, opt_in.prefix);
2207 				if (ipv6_addr_cmp(&pfx, addr6))
2208 					return -EINVAL;
2209 			}
2210 		} else if (!ipv6_addr_any(addr6)) {
2211 			return -EINVAL;
2212 		}
2213 		break;
2214 	}
2215 	case 0:
2216 		if (!do_address_matching)
2217 			break;
2218 		fallthrough;
2219 	default:
2220 		return -EAFNOSUPPORT;
2221 	}
2222 
2223 	if (!do_address_matching) {
2224 		/* We could just ignore those, but let's do stricter checks */
2225 		if (addr || port)
2226 			return -EINVAL;
2227 		if (opt_in.prefix || opt_in.sndid || opt_in.rcvid)
2228 			return -EINVAL;
2229 	}
2230 
2231 	bytes_to_write = min_t(int, user_len, sizeof(struct tcp_ao_getsockopt));
2232 	matched_keys = 0;
2233 	/* May change in RX, while we're dumping, pre-fetch it */
2234 	current_key = READ_ONCE(ao_info->current_key);
2235 
2236 	hlist_for_each_entry_rcu(key, &ao_info->head, node,
2237 				 lockdep_sock_is_held(sk)) {
2238 		if (opt_in.get_all)
2239 			goto match;
2240 
2241 		if (opt_in.is_current || opt_in.is_rnext) {
2242 			if (opt_in.is_current && key == current_key)
2243 				goto match;
2244 			if (opt_in.is_rnext && key == ao_info->rnext_key)
2245 				goto match;
2246 			continue;
2247 		}
2248 
2249 		if (tcp_ao_key_cmp(key, l3index, addr, opt_in.prefix,
2250 				   opt_in.addr.ss_family,
2251 				   opt_in.sndid, opt_in.rcvid) != 0)
2252 			continue;
2253 match:
2254 		matched_keys++;
2255 		if (matched_keys > max_keys)
2256 			continue;
2257 
2258 		memset(&opt_out, 0, sizeof(struct tcp_ao_getsockopt));
2259 
2260 		if (key->family == AF_INET) {
2261 			struct sockaddr_in *sin_out = (struct sockaddr_in *)&opt_out.addr;
2262 
2263 			sin_out->sin_family = key->family;
2264 			sin_out->sin_port = 0;
2265 			memcpy(&sin_out->sin_addr, &key->addr, sizeof(struct in_addr));
2266 		} else {
2267 			struct sockaddr_in6 *sin6_out = (struct sockaddr_in6 *)&opt_out.addr;
2268 
2269 			sin6_out->sin6_family = key->family;
2270 			sin6_out->sin6_port = 0;
2271 			memcpy(&sin6_out->sin6_addr, &key->addr, sizeof(struct in6_addr));
2272 		}
2273 		opt_out.sndid = key->sndid;
2274 		opt_out.rcvid = key->rcvid;
2275 		opt_out.prefix = key->prefixlen;
2276 		opt_out.keyflags = key->keyflags;
2277 		opt_out.is_current = (key == current_key);
2278 		opt_out.is_rnext = (key == ao_info->rnext_key);
2279 		opt_out.nkeys = 0;
2280 		opt_out.maclen = key->maclen;
2281 		opt_out.keylen = key->keylen;
2282 		opt_out.ifindex = key->l3index;
2283 		opt_out.pkt_good = atomic64_read(&key->pkt_good);
2284 		opt_out.pkt_bad = atomic64_read(&key->pkt_bad);
2285 		memcpy(&opt_out.key, key->key, key->keylen);
2286 		tcp_sigpool_algo(key->tcp_sigpool_id, opt_out.alg_name, 64);
2287 
2288 		/* Copy key to user */
2289 		if (copy_to_sockptr_offset(optval, out_offset,
2290 					   &opt_out, bytes_to_write))
2291 			return -EFAULT;
2292 		out_offset += user_len;
2293 	}
2294 
2295 	optlen_out = (int)sizeof(struct tcp_ao_getsockopt);
2296 	if (copy_to_sockptr(optlen, &optlen_out, sizeof(int)))
2297 		return -EFAULT;
2298 
2299 	out_offset = offsetof(struct tcp_ao_getsockopt, nkeys);
2300 	if (copy_to_sockptr_offset(optval, out_offset,
2301 				   &matched_keys, sizeof(u32)))
2302 		return -EFAULT;
2303 
2304 	return 0;
2305 }
2306 
2307 int tcp_ao_get_mkts(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2308 {
2309 	struct tcp_ao_info *ao_info;
2310 
2311 	ao_info = setsockopt_ao_info(sk);
2312 	if (IS_ERR(ao_info))
2313 		return PTR_ERR(ao_info);
2314 	if (!ao_info)
2315 		return -ENOENT;
2316 
2317 	return tcp_ao_copy_mkts_to_user(sk, ao_info, optval, optlen);
2318 }
2319 
2320 int tcp_ao_get_sock_info(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2321 {
2322 	struct tcp_ao_info_opt out, in = {};
2323 	struct tcp_ao_key *current_key;
2324 	struct tcp_ao_info *ao;
2325 	int err, len;
2326 
2327 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
2328 		return -EFAULT;
2329 
2330 	if (len <= 0)
2331 		return -EINVAL;
2332 
2333 	/* Copying this "in" only to check ::reserved, ::reserved2,
2334 	 * that may be needed to extend (struct tcp_ao_info_opt) and
2335 	 * what getsockopt() provides in future.
2336 	 */
2337 	err = copy_struct_from_sockptr(&in, sizeof(in), optval, len);
2338 	if (err)
2339 		return err;
2340 
2341 	if (in.reserved != 0 || in.reserved2 != 0)
2342 		return -EINVAL;
2343 
2344 	ao = setsockopt_ao_info(sk);
2345 	if (IS_ERR(ao))
2346 		return PTR_ERR(ao);
2347 	if (!ao)
2348 		return -ENOENT;
2349 
2350 	memset(&out, 0, sizeof(out));
2351 	out.ao_required		= ao->ao_required;
2352 	out.accept_icmps	= ao->accept_icmps;
2353 	out.pkt_good		= atomic64_read(&ao->counters.pkt_good);
2354 	out.pkt_bad		= atomic64_read(&ao->counters.pkt_bad);
2355 	out.pkt_key_not_found	= atomic64_read(&ao->counters.key_not_found);
2356 	out.pkt_ao_required	= atomic64_read(&ao->counters.ao_required);
2357 	out.pkt_dropped_icmp	= atomic64_read(&ao->counters.dropped_icmp);
2358 
2359 	current_key = READ_ONCE(ao->current_key);
2360 	if (current_key) {
2361 		out.set_current = 1;
2362 		out.current_key = current_key->sndid;
2363 	}
2364 	if (ao->rnext_key) {
2365 		out.set_rnext = 1;
2366 		out.rnext = ao->rnext_key->rcvid;
2367 	}
2368 
2369 	if (copy_to_sockptr(optval, &out, min_t(int, len, sizeof(out))))
2370 		return -EFAULT;
2371 
2372 	return 0;
2373 }
2374 
2375 int tcp_ao_set_repair(struct sock *sk, sockptr_t optval, unsigned int optlen)
2376 {
2377 	struct tcp_sock *tp = tcp_sk(sk);
2378 	struct tcp_ao_repair cmd;
2379 	struct tcp_ao_key *key;
2380 	struct tcp_ao_info *ao;
2381 	int err;
2382 
2383 	if (optlen < sizeof(cmd))
2384 		return -EINVAL;
2385 
2386 	err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
2387 	if (err)
2388 		return err;
2389 
2390 	if (!tp->repair)
2391 		return -EPERM;
2392 
2393 	ao = setsockopt_ao_info(sk);
2394 	if (IS_ERR(ao))
2395 		return PTR_ERR(ao);
2396 	if (!ao)
2397 		return -ENOENT;
2398 
2399 	WRITE_ONCE(ao->lisn, cmd.snt_isn);
2400 	WRITE_ONCE(ao->risn, cmd.rcv_isn);
2401 	WRITE_ONCE(ao->snd_sne, cmd.snd_sne);
2402 	WRITE_ONCE(ao->rcv_sne, cmd.rcv_sne);
2403 
2404 	hlist_for_each_entry_rcu(key, &ao->head, node, lockdep_sock_is_held(sk))
2405 		tcp_ao_cache_traffic_keys(sk, ao, key);
2406 
2407 	return 0;
2408 }
2409 
2410 int tcp_ao_get_repair(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2411 {
2412 	struct tcp_sock *tp = tcp_sk(sk);
2413 	struct tcp_ao_repair opt;
2414 	struct tcp_ao_info *ao;
2415 	int len;
2416 
2417 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
2418 		return -EFAULT;
2419 
2420 	if (len <= 0)
2421 		return -EINVAL;
2422 
2423 	if (!tp->repair)
2424 		return -EPERM;
2425 
2426 	rcu_read_lock();
2427 	ao = getsockopt_ao_info(sk);
2428 	if (IS_ERR_OR_NULL(ao)) {
2429 		rcu_read_unlock();
2430 		return ao ? PTR_ERR(ao) : -ENOENT;
2431 	}
2432 
2433 	opt.snt_isn	= ao->lisn;
2434 	opt.rcv_isn	= ao->risn;
2435 	opt.snd_sne	= READ_ONCE(ao->snd_sne);
2436 	opt.rcv_sne	= READ_ONCE(ao->rcv_sne);
2437 	rcu_read_unlock();
2438 
2439 	if (copy_to_sockptr(optval, &opt, min_t(int, len, sizeof(opt))))
2440 		return -EFAULT;
2441 	return 0;
2442 }
2443