xref: /linux/net/ipv4/tcp_ao.c (revision 55d0969c451159cff86949b38c39171cab962069)
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(struct tcp_ao_info *ao,
113 					  int sndid, int rcvid)
114 {
115 	struct tcp_ao_key *key;
116 
117 	hlist_for_each_entry_rcu(key, &ao->head, node) {
118 		if ((sndid >= 0 && key->sndid != sndid) ||
119 		    (rcvid >= 0 && key->rcvid != rcvid))
120 			continue;
121 		return key;
122 	}
123 
124 	return NULL;
125 }
126 
127 static int ipv4_prefix_cmp(const struct in_addr *addr1,
128 			   const struct in_addr *addr2,
129 			   unsigned int prefixlen)
130 {
131 	__be32 mask = inet_make_mask(prefixlen);
132 	__be32 a1 = addr1->s_addr & mask;
133 	__be32 a2 = addr2->s_addr & mask;
134 
135 	if (a1 == a2)
136 		return 0;
137 	return memcmp(&a1, &a2, sizeof(a1));
138 }
139 
140 static int __tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index,
141 			    const union tcp_ao_addr *addr, u8 prefixlen,
142 			    int family, int sndid, int rcvid)
143 {
144 	if (sndid >= 0 && key->sndid != sndid)
145 		return (key->sndid > sndid) ? 1 : -1;
146 	if (rcvid >= 0 && key->rcvid != rcvid)
147 		return (key->rcvid > rcvid) ? 1 : -1;
148 	if (l3index >= 0 && (key->keyflags & TCP_AO_KEYF_IFINDEX)) {
149 		if (key->l3index != l3index)
150 			return (key->l3index > l3index) ? 1 : -1;
151 	}
152 
153 	if (family == AF_UNSPEC)
154 		return 0;
155 	if (key->family != family)
156 		return (key->family > family) ? 1 : -1;
157 
158 	if (family == AF_INET) {
159 		if (ntohl(key->addr.a4.s_addr) == INADDR_ANY)
160 			return 0;
161 		if (ntohl(addr->a4.s_addr) == INADDR_ANY)
162 			return 0;
163 		return ipv4_prefix_cmp(&key->addr.a4, &addr->a4, prefixlen);
164 #if IS_ENABLED(CONFIG_IPV6)
165 	} else {
166 		if (ipv6_addr_any(&key->addr.a6) || ipv6_addr_any(&addr->a6))
167 			return 0;
168 		if (ipv6_prefix_equal(&key->addr.a6, &addr->a6, prefixlen))
169 			return 0;
170 		return memcmp(&key->addr.a6, &addr->a6, sizeof(addr->a6));
171 #endif
172 	}
173 	return -1;
174 }
175 
176 static int tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index,
177 			  const union tcp_ao_addr *addr, u8 prefixlen,
178 			  int family, int sndid, int rcvid)
179 {
180 #if IS_ENABLED(CONFIG_IPV6)
181 	if (family == AF_INET6 && ipv6_addr_v4mapped(&addr->a6)) {
182 		__be32 addr4 = addr->a6.s6_addr32[3];
183 
184 		return __tcp_ao_key_cmp(key, l3index,
185 					(union tcp_ao_addr *)&addr4,
186 					prefixlen, AF_INET, sndid, rcvid);
187 	}
188 #endif
189 	return __tcp_ao_key_cmp(key, l3index, addr,
190 				prefixlen, family, sndid, rcvid);
191 }
192 
193 static struct tcp_ao_key *__tcp_ao_do_lookup(const struct sock *sk, int l3index,
194 		const union tcp_ao_addr *addr, int family, u8 prefix,
195 		int sndid, int rcvid)
196 {
197 	struct tcp_ao_key *key;
198 	struct tcp_ao_info *ao;
199 
200 	if (!static_branch_unlikely(&tcp_ao_needed.key))
201 		return NULL;
202 
203 	ao = rcu_dereference_check(tcp_sk(sk)->ao_info,
204 				   lockdep_sock_is_held(sk));
205 	if (!ao)
206 		return NULL;
207 
208 	hlist_for_each_entry_rcu(key, &ao->head, node) {
209 		u8 prefixlen = min(prefix, key->prefixlen);
210 
211 		if (!tcp_ao_key_cmp(key, l3index, addr, prefixlen,
212 				    family, sndid, rcvid))
213 			return key;
214 	}
215 	return NULL;
216 }
217 
218 struct tcp_ao_key *tcp_ao_do_lookup(const struct sock *sk, int l3index,
219 				    const union tcp_ao_addr *addr,
220 				    int family, int sndid, int rcvid)
221 {
222 	return __tcp_ao_do_lookup(sk, l3index, addr, family, U8_MAX, sndid, rcvid);
223 }
224 
225 static struct tcp_ao_info *tcp_ao_alloc_info(gfp_t flags)
226 {
227 	struct tcp_ao_info *ao;
228 
229 	ao = kzalloc(sizeof(*ao), flags);
230 	if (!ao)
231 		return NULL;
232 	INIT_HLIST_HEAD(&ao->head);
233 	refcount_set(&ao->refcnt, 1);
234 
235 	return ao;
236 }
237 
238 static void tcp_ao_link_mkt(struct tcp_ao_info *ao, struct tcp_ao_key *mkt)
239 {
240 	hlist_add_head_rcu(&mkt->node, &ao->head);
241 }
242 
243 static struct tcp_ao_key *tcp_ao_copy_key(struct sock *sk,
244 					  struct tcp_ao_key *key)
245 {
246 	struct tcp_ao_key *new_key;
247 
248 	new_key = sock_kmalloc(sk, tcp_ao_sizeof_key(key),
249 			       GFP_ATOMIC);
250 	if (!new_key)
251 		return NULL;
252 
253 	*new_key = *key;
254 	INIT_HLIST_NODE(&new_key->node);
255 	tcp_sigpool_get(new_key->tcp_sigpool_id);
256 	atomic64_set(&new_key->pkt_good, 0);
257 	atomic64_set(&new_key->pkt_bad, 0);
258 
259 	return new_key;
260 }
261 
262 static void tcp_ao_key_free_rcu(struct rcu_head *head)
263 {
264 	struct tcp_ao_key *key = container_of(head, struct tcp_ao_key, rcu);
265 
266 	tcp_sigpool_release(key->tcp_sigpool_id);
267 	kfree_sensitive(key);
268 }
269 
270 static void tcp_ao_info_free_rcu(struct rcu_head *head)
271 {
272 	struct tcp_ao_info *ao = container_of(head, struct tcp_ao_info, rcu);
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 	call_rcu(&ao->rcu, tcp_ao_info_free_rcu);
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(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(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(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)
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 	WRITE_ONCE(ao->risn, tcp_hdr(skb)->seq);
1181 	ao->rcv_sne = 0;
1182 
1183 	hlist_for_each_entry_rcu(key, &ao->head, node)
1184 		tcp_ao_cache_traffic_keys(sk, ao, key);
1185 }
1186 
1187 int tcp_ao_copy_all_matching(const struct sock *sk, struct sock *newsk,
1188 			     struct request_sock *req, struct sk_buff *skb,
1189 			     int family)
1190 {
1191 	struct tcp_ao_key *key, *new_key, *first_key;
1192 	struct tcp_ao_info *new_ao, *ao;
1193 	struct hlist_node *key_head;
1194 	int l3index, ret = -ENOMEM;
1195 	union tcp_ao_addr *addr;
1196 	bool match = false;
1197 
1198 	ao = rcu_dereference(tcp_sk(sk)->ao_info);
1199 	if (!ao)
1200 		return 0;
1201 
1202 	/* New socket without TCP-AO on it */
1203 	if (!tcp_rsk_used_ao(req))
1204 		return 0;
1205 
1206 	new_ao = tcp_ao_alloc_info(GFP_ATOMIC);
1207 	if (!new_ao)
1208 		return -ENOMEM;
1209 	new_ao->lisn = htonl(tcp_rsk(req)->snt_isn);
1210 	new_ao->risn = htonl(tcp_rsk(req)->rcv_isn);
1211 	new_ao->ao_required = ao->ao_required;
1212 	new_ao->accept_icmps = ao->accept_icmps;
1213 
1214 	if (family == AF_INET) {
1215 		addr = (union tcp_ao_addr *)&newsk->sk_daddr;
1216 #if IS_ENABLED(CONFIG_IPV6)
1217 	} else if (family == AF_INET6) {
1218 		addr = (union tcp_ao_addr *)&newsk->sk_v6_daddr;
1219 #endif
1220 	} else {
1221 		ret = -EAFNOSUPPORT;
1222 		goto free_ao;
1223 	}
1224 	l3index = l3mdev_master_ifindex_by_index(sock_net(newsk),
1225 						 newsk->sk_bound_dev_if);
1226 
1227 	hlist_for_each_entry_rcu(key, &ao->head, node) {
1228 		if (tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1))
1229 			continue;
1230 
1231 		new_key = tcp_ao_copy_key(newsk, key);
1232 		if (!new_key)
1233 			goto free_and_exit;
1234 
1235 		tcp_ao_cache_traffic_keys(newsk, new_ao, new_key);
1236 		tcp_ao_link_mkt(new_ao, new_key);
1237 		match = true;
1238 	}
1239 
1240 	if (!match) {
1241 		/* RFC5925 (7.4.1) specifies that the TCP-AO status
1242 		 * of a connection is determined on the initial SYN.
1243 		 * At this point the connection was TCP-AO enabled, so
1244 		 * it can't switch to being unsigned if peer's key
1245 		 * disappears on the listening socket.
1246 		 */
1247 		ret = -EKEYREJECTED;
1248 		goto free_and_exit;
1249 	}
1250 
1251 	if (!static_key_fast_inc_not_disabled(&tcp_ao_needed.key.key)) {
1252 		ret = -EUSERS;
1253 		goto free_and_exit;
1254 	}
1255 
1256 	key_head = rcu_dereference(hlist_first_rcu(&new_ao->head));
1257 	first_key = hlist_entry_safe(key_head, struct tcp_ao_key, node);
1258 
1259 	key = tcp_ao_established_key(new_ao, tcp_rsk(req)->ao_keyid, -1);
1260 	if (key)
1261 		new_ao->current_key = key;
1262 	else
1263 		new_ao->current_key = first_key;
1264 
1265 	/* set rnext_key */
1266 	key = tcp_ao_established_key(new_ao, -1, tcp_rsk(req)->ao_rcv_next);
1267 	if (key)
1268 		new_ao->rnext_key = key;
1269 	else
1270 		new_ao->rnext_key = first_key;
1271 
1272 	sk_gso_disable(newsk);
1273 	rcu_assign_pointer(tcp_sk(newsk)->ao_info, new_ao);
1274 
1275 	return 0;
1276 
1277 free_and_exit:
1278 	hlist_for_each_entry_safe(key, key_head, &new_ao->head, node) {
1279 		hlist_del(&key->node);
1280 		tcp_sigpool_release(key->tcp_sigpool_id);
1281 		atomic_sub(tcp_ao_sizeof_key(key), &newsk->sk_omem_alloc);
1282 		kfree_sensitive(key);
1283 	}
1284 free_ao:
1285 	kfree(new_ao);
1286 	return ret;
1287 }
1288 
1289 static bool tcp_ao_can_set_current_rnext(struct sock *sk)
1290 {
1291 	/* There aren't current/rnext keys on TCP_LISTEN sockets */
1292 	if (sk->sk_state == TCP_LISTEN)
1293 		return false;
1294 	return true;
1295 }
1296 
1297 static int tcp_ao_verify_ipv4(struct sock *sk, struct tcp_ao_add *cmd,
1298 			      union tcp_ao_addr **addr)
1299 {
1300 	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd->addr;
1301 	struct inet_sock *inet = inet_sk(sk);
1302 
1303 	if (sin->sin_family != AF_INET)
1304 		return -EINVAL;
1305 
1306 	/* Currently matching is not performed on port (or port ranges) */
1307 	if (sin->sin_port != 0)
1308 		return -EINVAL;
1309 
1310 	/* Check prefix and trailing 0's in addr */
1311 	if (cmd->prefix != 0) {
1312 		__be32 mask;
1313 
1314 		if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY)
1315 			return -EINVAL;
1316 		if (cmd->prefix > 32)
1317 			return -EINVAL;
1318 
1319 		mask = inet_make_mask(cmd->prefix);
1320 		if (sin->sin_addr.s_addr & ~mask)
1321 			return -EINVAL;
1322 
1323 		/* Check that MKT address is consistent with socket */
1324 		if (ntohl(inet->inet_daddr) != INADDR_ANY &&
1325 		    (inet->inet_daddr & mask) != sin->sin_addr.s_addr)
1326 			return -EINVAL;
1327 	} else {
1328 		if (ntohl(sin->sin_addr.s_addr) != INADDR_ANY)
1329 			return -EINVAL;
1330 	}
1331 
1332 	*addr = (union tcp_ao_addr *)&sin->sin_addr;
1333 	return 0;
1334 }
1335 
1336 static int tcp_ao_parse_crypto(struct tcp_ao_add *cmd, struct tcp_ao_key *key)
1337 {
1338 	unsigned int syn_tcp_option_space;
1339 	bool is_kdf_aes_128_cmac = false;
1340 	struct crypto_ahash *tfm;
1341 	struct tcp_sigpool hp;
1342 	void *tmp_key = NULL;
1343 	int err;
1344 
1345 	/* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1346 	if (!strcmp("cmac(aes128)", cmd->alg_name)) {
1347 		strscpy(cmd->alg_name, "cmac(aes)", sizeof(cmd->alg_name));
1348 		is_kdf_aes_128_cmac = (cmd->keylen != 16);
1349 		tmp_key = kmalloc(cmd->keylen, GFP_KERNEL);
1350 		if (!tmp_key)
1351 			return -ENOMEM;
1352 	}
1353 
1354 	key->maclen = cmd->maclen ?: 12; /* 12 is the default in RFC5925 */
1355 
1356 	/* Check: maclen + tcp-ao header <= (MAX_TCP_OPTION_SPACE - mss
1357 	 *					- tstamp (including sackperm)
1358 	 *					- wscale),
1359 	 * see tcp_syn_options(), tcp_synack_options(), commit 33ad798c924b.
1360 	 *
1361 	 * In order to allow D-SACK with TCP-AO, the header size should be:
1362 	 * (MAX_TCP_OPTION_SPACE - TCPOLEN_TSTAMP_ALIGNED
1363 	 *			- TCPOLEN_SACK_BASE_ALIGNED
1364 	 *			- 2 * TCPOLEN_SACK_PERBLOCK) = 8 (maclen = 4),
1365 	 * see tcp_established_options().
1366 	 *
1367 	 * RFC5925, 2.2:
1368 	 * Typical MACs are 96-128 bits (12-16 bytes), but any length
1369 	 * that fits in the header of the segment being authenticated
1370 	 * is allowed.
1371 	 *
1372 	 * RFC5925, 7.6:
1373 	 * TCP-AO continues to consume 16 bytes in non-SYN segments,
1374 	 * leaving a total of 24 bytes for other options, of which
1375 	 * the timestamp consumes 10.  This leaves 14 bytes, of which 10
1376 	 * are used for a single SACK block. When two SACK blocks are used,
1377 	 * such as to handle D-SACK, a smaller TCP-AO MAC would be required
1378 	 * to make room for the additional SACK block (i.e., to leave 18
1379 	 * bytes for the D-SACK variant of the SACK option) [RFC2883].
1380 	 * Note that D-SACK is not supportable in TCP MD5 in the presence
1381 	 * of timestamps, because TCP MD5’s MAC length is fixed and too
1382 	 * large to leave sufficient option space.
1383 	 */
1384 	syn_tcp_option_space = MAX_TCP_OPTION_SPACE;
1385 	syn_tcp_option_space -= TCPOLEN_MSS_ALIGNED;
1386 	syn_tcp_option_space -= TCPOLEN_TSTAMP_ALIGNED;
1387 	syn_tcp_option_space -= TCPOLEN_WSCALE_ALIGNED;
1388 	if (tcp_ao_len_aligned(key) > syn_tcp_option_space) {
1389 		err = -EMSGSIZE;
1390 		goto err_kfree;
1391 	}
1392 
1393 	key->keylen = cmd->keylen;
1394 	memcpy(key->key, cmd->key, cmd->keylen);
1395 
1396 	err = tcp_sigpool_start(key->tcp_sigpool_id, &hp);
1397 	if (err)
1398 		goto err_kfree;
1399 
1400 	tfm = crypto_ahash_reqtfm(hp.req);
1401 	if (is_kdf_aes_128_cmac) {
1402 		void *scratch = hp.scratch;
1403 		struct scatterlist sg;
1404 
1405 		memcpy(tmp_key, cmd->key, cmd->keylen);
1406 		sg_init_one(&sg, tmp_key, cmd->keylen);
1407 
1408 		/* Using zero-key of 16 bytes as described in RFC5926 */
1409 		memset(scratch, 0, 16);
1410 		err = crypto_ahash_setkey(tfm, scratch, 16);
1411 		if (err)
1412 			goto err_pool_end;
1413 
1414 		err = crypto_ahash_init(hp.req);
1415 		if (err)
1416 			goto err_pool_end;
1417 
1418 		ahash_request_set_crypt(hp.req, &sg, key->key, cmd->keylen);
1419 		err = crypto_ahash_update(hp.req);
1420 		if (err)
1421 			goto err_pool_end;
1422 
1423 		err |= crypto_ahash_final(hp.req);
1424 		if (err)
1425 			goto err_pool_end;
1426 		key->keylen = 16;
1427 	}
1428 
1429 	err = crypto_ahash_setkey(tfm, key->key, key->keylen);
1430 	if (err)
1431 		goto err_pool_end;
1432 
1433 	tcp_sigpool_end(&hp);
1434 	kfree_sensitive(tmp_key);
1435 
1436 	if (tcp_ao_maclen(key) > key->digest_size)
1437 		return -EINVAL;
1438 
1439 	return 0;
1440 
1441 err_pool_end:
1442 	tcp_sigpool_end(&hp);
1443 err_kfree:
1444 	kfree_sensitive(tmp_key);
1445 	return err;
1446 }
1447 
1448 #if IS_ENABLED(CONFIG_IPV6)
1449 static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1450 			      union tcp_ao_addr **paddr,
1451 			      unsigned short int *family)
1452 {
1453 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd->addr;
1454 	struct in6_addr *addr = &sin6->sin6_addr;
1455 	u8 prefix = cmd->prefix;
1456 
1457 	if (sin6->sin6_family != AF_INET6)
1458 		return -EINVAL;
1459 
1460 	/* Currently matching is not performed on port (or port ranges) */
1461 	if (sin6->sin6_port != 0)
1462 		return -EINVAL;
1463 
1464 	/* Check prefix and trailing 0's in addr */
1465 	if (cmd->prefix != 0 && ipv6_addr_v4mapped(addr)) {
1466 		__be32 addr4 = addr->s6_addr32[3];
1467 		__be32 mask;
1468 
1469 		if (prefix > 32 || ntohl(addr4) == INADDR_ANY)
1470 			return -EINVAL;
1471 
1472 		mask = inet_make_mask(prefix);
1473 		if (addr4 & ~mask)
1474 			return -EINVAL;
1475 
1476 		/* Check that MKT address is consistent with socket */
1477 		if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
1478 			__be32 daddr4 = sk->sk_v6_daddr.s6_addr32[3];
1479 
1480 			if (!ipv6_addr_v4mapped(&sk->sk_v6_daddr))
1481 				return -EINVAL;
1482 			if ((daddr4 & mask) != addr4)
1483 				return -EINVAL;
1484 		}
1485 
1486 		*paddr = (union tcp_ao_addr *)&addr->s6_addr32[3];
1487 		*family = AF_INET;
1488 		return 0;
1489 	} else if (cmd->prefix != 0) {
1490 		struct in6_addr pfx;
1491 
1492 		if (ipv6_addr_any(addr) || prefix > 128)
1493 			return -EINVAL;
1494 
1495 		ipv6_addr_prefix(&pfx, addr, prefix);
1496 		if (ipv6_addr_cmp(&pfx, addr))
1497 			return -EINVAL;
1498 
1499 		/* Check that MKT address is consistent with socket */
1500 		if (!ipv6_addr_any(&sk->sk_v6_daddr) &&
1501 		    !ipv6_prefix_equal(&sk->sk_v6_daddr, addr, prefix))
1502 
1503 			return -EINVAL;
1504 	} else {
1505 		if (!ipv6_addr_any(addr))
1506 			return -EINVAL;
1507 	}
1508 
1509 	*paddr = (union tcp_ao_addr *)addr;
1510 	return 0;
1511 }
1512 #else
1513 static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1514 			      union tcp_ao_addr **paddr,
1515 			      unsigned short int *family)
1516 {
1517 	return -EOPNOTSUPP;
1518 }
1519 #endif
1520 
1521 static struct tcp_ao_info *setsockopt_ao_info(struct sock *sk)
1522 {
1523 	if (sk_fullsock(sk)) {
1524 		return rcu_dereference_protected(tcp_sk(sk)->ao_info,
1525 						 lockdep_sock_is_held(sk));
1526 	} else if (sk->sk_state == TCP_TIME_WAIT) {
1527 		return rcu_dereference_protected(tcp_twsk(sk)->ao_info,
1528 						 lockdep_sock_is_held(sk));
1529 	}
1530 	return ERR_PTR(-ESOCKTNOSUPPORT);
1531 }
1532 
1533 static struct tcp_ao_info *getsockopt_ao_info(struct sock *sk)
1534 {
1535 	if (sk_fullsock(sk))
1536 		return rcu_dereference(tcp_sk(sk)->ao_info);
1537 	else if (sk->sk_state == TCP_TIME_WAIT)
1538 		return rcu_dereference(tcp_twsk(sk)->ao_info);
1539 
1540 	return ERR_PTR(-ESOCKTNOSUPPORT);
1541 }
1542 
1543 #define TCP_AO_KEYF_ALL (TCP_AO_KEYF_IFINDEX | TCP_AO_KEYF_EXCLUDE_OPT)
1544 #define TCP_AO_GET_KEYF_VALID	(TCP_AO_KEYF_IFINDEX)
1545 
1546 static struct tcp_ao_key *tcp_ao_key_alloc(struct sock *sk,
1547 					   struct tcp_ao_add *cmd)
1548 {
1549 	const char *algo = cmd->alg_name;
1550 	unsigned int digest_size;
1551 	struct crypto_ahash *tfm;
1552 	struct tcp_ao_key *key;
1553 	struct tcp_sigpool hp;
1554 	int err, pool_id;
1555 	size_t size;
1556 
1557 	/* Force null-termination of alg_name */
1558 	cmd->alg_name[ARRAY_SIZE(cmd->alg_name) - 1] = '\0';
1559 
1560 	/* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1561 	if (!strcmp("cmac(aes128)", algo))
1562 		algo = "cmac(aes)";
1563 
1564 	/* Full TCP header (th->doff << 2) should fit into scratch area,
1565 	 * see tcp_ao_hash_header().
1566 	 */
1567 	pool_id = tcp_sigpool_alloc_ahash(algo, 60);
1568 	if (pool_id < 0)
1569 		return ERR_PTR(pool_id);
1570 
1571 	err = tcp_sigpool_start(pool_id, &hp);
1572 	if (err)
1573 		goto err_free_pool;
1574 
1575 	tfm = crypto_ahash_reqtfm(hp.req);
1576 	digest_size = crypto_ahash_digestsize(tfm);
1577 	tcp_sigpool_end(&hp);
1578 
1579 	size = sizeof(struct tcp_ao_key) + (digest_size << 1);
1580 	key = sock_kmalloc(sk, size, GFP_KERNEL);
1581 	if (!key) {
1582 		err = -ENOMEM;
1583 		goto err_free_pool;
1584 	}
1585 
1586 	key->tcp_sigpool_id = pool_id;
1587 	key->digest_size = digest_size;
1588 	return key;
1589 
1590 err_free_pool:
1591 	tcp_sigpool_release(pool_id);
1592 	return ERR_PTR(err);
1593 }
1594 
1595 static int tcp_ao_add_cmd(struct sock *sk, unsigned short int family,
1596 			  sockptr_t optval, int optlen)
1597 {
1598 	struct tcp_ao_info *ao_info;
1599 	union tcp_ao_addr *addr;
1600 	struct tcp_ao_key *key;
1601 	struct tcp_ao_add cmd;
1602 	int ret, l3index = 0;
1603 	bool first = false;
1604 
1605 	if (optlen < sizeof(cmd))
1606 		return -EINVAL;
1607 
1608 	ret = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1609 	if (ret)
1610 		return ret;
1611 
1612 	if (cmd.keylen > TCP_AO_MAXKEYLEN)
1613 		return -EINVAL;
1614 
1615 	if (cmd.reserved != 0 || cmd.reserved2 != 0)
1616 		return -EINVAL;
1617 
1618 	if (family == AF_INET)
1619 		ret = tcp_ao_verify_ipv4(sk, &cmd, &addr);
1620 	else
1621 		ret = tcp_ao_verify_ipv6(sk, &cmd, &addr, &family);
1622 	if (ret)
1623 		return ret;
1624 
1625 	if (cmd.keyflags & ~TCP_AO_KEYF_ALL)
1626 		return -EINVAL;
1627 
1628 	if (cmd.set_current || cmd.set_rnext) {
1629 		if (!tcp_ao_can_set_current_rnext(sk))
1630 			return -EINVAL;
1631 	}
1632 
1633 	if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1634 		return -EINVAL;
1635 
1636 	/* For cmd.tcp_ifindex = 0 the key will apply to the default VRF */
1637 	if (cmd.keyflags & TCP_AO_KEYF_IFINDEX && cmd.ifindex) {
1638 		int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
1639 		struct net_device *dev;
1640 
1641 		rcu_read_lock();
1642 		dev = dev_get_by_index_rcu(sock_net(sk), cmd.ifindex);
1643 		if (dev && netif_is_l3_master(dev))
1644 			l3index = dev->ifindex;
1645 		rcu_read_unlock();
1646 
1647 		if (!dev || !l3index)
1648 			return -EINVAL;
1649 
1650 		if (!bound_dev_if || bound_dev_if != cmd.ifindex) {
1651 			/* tcp_ao_established_key() doesn't expect having
1652 			 * non peer-matching key on an established TCP-AO
1653 			 * connection.
1654 			 */
1655 			if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1656 				return -EINVAL;
1657 		}
1658 
1659 		/* It's still possible to bind after adding keys or even
1660 		 * re-bind to a different dev (with CAP_NET_RAW).
1661 		 * So, no reason to return error here, rather try to be
1662 		 * nice and warn the user.
1663 		 */
1664 		if (bound_dev_if && bound_dev_if != cmd.ifindex)
1665 			net_warn_ratelimited("AO key ifindex %d != sk bound ifindex %d\n",
1666 					     cmd.ifindex, bound_dev_if);
1667 	}
1668 
1669 	/* Don't allow keys for peers that have a matching TCP-MD5 key */
1670 	if (cmd.keyflags & TCP_AO_KEYF_IFINDEX) {
1671 		/* Non-_exact version of tcp_md5_do_lookup() will
1672 		 * as well match keys that aren't bound to a specific VRF
1673 		 * (that will make them match AO key with
1674 		 * sysctl_tcp_l3dev_accept = 1
1675 		 */
1676 		if (tcp_md5_do_lookup(sk, l3index, addr, family))
1677 			return -EKEYREJECTED;
1678 	} else {
1679 		if (tcp_md5_do_lookup_any_l3index(sk, addr, family))
1680 			return -EKEYREJECTED;
1681 	}
1682 
1683 	ao_info = setsockopt_ao_info(sk);
1684 	if (IS_ERR(ao_info))
1685 		return PTR_ERR(ao_info);
1686 
1687 	if (!ao_info) {
1688 		ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1689 		if (!ao_info)
1690 			return -ENOMEM;
1691 		first = true;
1692 	} else {
1693 		/* Check that neither RecvID nor SendID match any
1694 		 * existing key for the peer, RFC5925 3.1:
1695 		 * > The IDs of MKTs MUST NOT overlap where their
1696 		 * > TCP connection identifiers overlap.
1697 		 */
1698 		if (__tcp_ao_do_lookup(sk, l3index, addr, family, cmd.prefix, -1, cmd.rcvid))
1699 			return -EEXIST;
1700 		if (__tcp_ao_do_lookup(sk, l3index, addr, family,
1701 				       cmd.prefix, cmd.sndid, -1))
1702 			return -EEXIST;
1703 	}
1704 
1705 	key = tcp_ao_key_alloc(sk, &cmd);
1706 	if (IS_ERR(key)) {
1707 		ret = PTR_ERR(key);
1708 		goto err_free_ao;
1709 	}
1710 
1711 	INIT_HLIST_NODE(&key->node);
1712 	memcpy(&key->addr, addr, (family == AF_INET) ? sizeof(struct in_addr) :
1713 						       sizeof(struct in6_addr));
1714 	key->prefixlen	= cmd.prefix;
1715 	key->family	= family;
1716 	key->keyflags	= cmd.keyflags;
1717 	key->sndid	= cmd.sndid;
1718 	key->rcvid	= cmd.rcvid;
1719 	key->l3index	= l3index;
1720 	atomic64_set(&key->pkt_good, 0);
1721 	atomic64_set(&key->pkt_bad, 0);
1722 
1723 	ret = tcp_ao_parse_crypto(&cmd, key);
1724 	if (ret < 0)
1725 		goto err_free_sock;
1726 
1727 	if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) {
1728 		tcp_ao_cache_traffic_keys(sk, ao_info, key);
1729 		if (first) {
1730 			ao_info->current_key = key;
1731 			ao_info->rnext_key = key;
1732 		}
1733 	}
1734 
1735 	tcp_ao_link_mkt(ao_info, key);
1736 	if (first) {
1737 		if (!static_branch_inc(&tcp_ao_needed.key)) {
1738 			ret = -EUSERS;
1739 			goto err_free_sock;
1740 		}
1741 		sk_gso_disable(sk);
1742 		rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
1743 	}
1744 
1745 	if (cmd.set_current)
1746 		WRITE_ONCE(ao_info->current_key, key);
1747 	if (cmd.set_rnext)
1748 		WRITE_ONCE(ao_info->rnext_key, key);
1749 	return 0;
1750 
1751 err_free_sock:
1752 	atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1753 	tcp_sigpool_release(key->tcp_sigpool_id);
1754 	kfree_sensitive(key);
1755 err_free_ao:
1756 	if (first)
1757 		kfree(ao_info);
1758 	return ret;
1759 }
1760 
1761 static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info,
1762 			     bool del_async, struct tcp_ao_key *key,
1763 			     struct tcp_ao_key *new_current,
1764 			     struct tcp_ao_key *new_rnext)
1765 {
1766 	int err;
1767 
1768 	hlist_del_rcu(&key->node);
1769 
1770 	/* Support for async delete on listening sockets: as they don't
1771 	 * need current_key/rnext_key maintaining, we don't need to check
1772 	 * them and we can just free all resources in RCU fashion.
1773 	 */
1774 	if (del_async) {
1775 		atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1776 		call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1777 		return 0;
1778 	}
1779 
1780 	/* At this moment another CPU could have looked this key up
1781 	 * while it was unlinked from the list. Wait for RCU grace period,
1782 	 * after which the key is off-list and can't be looked up again;
1783 	 * the rx path [just before RCU came] might have used it and set it
1784 	 * as current_key (very unlikely).
1785 	 * Free the key with next RCU grace period (in case it was
1786 	 * current_key before tcp_ao_current_rnext() might have
1787 	 * changed it in forced-delete).
1788 	 */
1789 	synchronize_rcu();
1790 	if (new_current)
1791 		WRITE_ONCE(ao_info->current_key, new_current);
1792 	if (new_rnext)
1793 		WRITE_ONCE(ao_info->rnext_key, new_rnext);
1794 
1795 	if (unlikely(READ_ONCE(ao_info->current_key) == key ||
1796 		     READ_ONCE(ao_info->rnext_key) == key)) {
1797 		err = -EBUSY;
1798 		goto add_key;
1799 	}
1800 
1801 	atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1802 	call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1803 
1804 	return 0;
1805 add_key:
1806 	hlist_add_head_rcu(&key->node, &ao_info->head);
1807 	return err;
1808 }
1809 
1810 #define TCP_AO_DEL_KEYF_ALL (TCP_AO_KEYF_IFINDEX)
1811 static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family,
1812 			  sockptr_t optval, int optlen)
1813 {
1814 	struct tcp_ao_key *key, *new_current = NULL, *new_rnext = NULL;
1815 	int err, addr_len, l3index = 0;
1816 	struct tcp_ao_info *ao_info;
1817 	union tcp_ao_addr *addr;
1818 	struct tcp_ao_del cmd;
1819 	__u8 prefix;
1820 	u16 port;
1821 
1822 	if (optlen < sizeof(cmd))
1823 		return -EINVAL;
1824 
1825 	err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1826 	if (err)
1827 		return err;
1828 
1829 	if (cmd.reserved != 0 || cmd.reserved2 != 0)
1830 		return -EINVAL;
1831 
1832 	if (cmd.set_current || cmd.set_rnext) {
1833 		if (!tcp_ao_can_set_current_rnext(sk))
1834 			return -EINVAL;
1835 	}
1836 
1837 	if (cmd.keyflags & ~TCP_AO_DEL_KEYF_ALL)
1838 		return -EINVAL;
1839 
1840 	/* No sanity check for TCP_AO_KEYF_IFINDEX as if a VRF
1841 	 * was destroyed, there still should be a way to delete keys,
1842 	 * that were bound to that l3intf. So, fail late at lookup stage
1843 	 * if there is no key for that ifindex.
1844 	 */
1845 	if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1846 		return -EINVAL;
1847 
1848 	ao_info = setsockopt_ao_info(sk);
1849 	if (IS_ERR(ao_info))
1850 		return PTR_ERR(ao_info);
1851 	if (!ao_info)
1852 		return -ENOENT;
1853 
1854 	/* For sockets in TCP_CLOSED it's possible set keys that aren't
1855 	 * matching the future peer (address/VRF/etc),
1856 	 * tcp_ao_connect_init() will choose a correct matching MKT
1857 	 * if there's any.
1858 	 */
1859 	if (cmd.set_current) {
1860 		new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1);
1861 		if (!new_current)
1862 			return -ENOENT;
1863 	}
1864 	if (cmd.set_rnext) {
1865 		new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext);
1866 		if (!new_rnext)
1867 			return -ENOENT;
1868 	}
1869 	if (cmd.del_async && sk->sk_state != TCP_LISTEN)
1870 		return -EINVAL;
1871 
1872 	if (family == AF_INET) {
1873 		struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr;
1874 
1875 		addr = (union tcp_ao_addr *)&sin->sin_addr;
1876 		addr_len = sizeof(struct in_addr);
1877 		port = ntohs(sin->sin_port);
1878 	} else {
1879 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.addr;
1880 		struct in6_addr *addr6 = &sin6->sin6_addr;
1881 
1882 		if (ipv6_addr_v4mapped(addr6)) {
1883 			addr = (union tcp_ao_addr *)&addr6->s6_addr32[3];
1884 			addr_len = sizeof(struct in_addr);
1885 			family = AF_INET;
1886 		} else {
1887 			addr = (union tcp_ao_addr *)addr6;
1888 			addr_len = sizeof(struct in6_addr);
1889 		}
1890 		port = ntohs(sin6->sin6_port);
1891 	}
1892 	prefix = cmd.prefix;
1893 
1894 	/* Currently matching is not performed on port (or port ranges) */
1895 	if (port != 0)
1896 		return -EINVAL;
1897 
1898 	/* We could choose random present key here for current/rnext
1899 	 * but that's less predictable. Let's be strict and don't
1900 	 * allow removing a key that's in use. RFC5925 doesn't
1901 	 * specify how-to coordinate key removal, but says:
1902 	 * "It is presumed that an MKT affecting a particular
1903 	 * connection cannot be destroyed during an active connection"
1904 	 */
1905 	hlist_for_each_entry_rcu(key, &ao_info->head, node) {
1906 		if (cmd.sndid != key->sndid ||
1907 		    cmd.rcvid != key->rcvid)
1908 			continue;
1909 
1910 		if (family != key->family ||
1911 		    prefix != key->prefixlen ||
1912 		    memcmp(addr, &key->addr, addr_len))
1913 			continue;
1914 
1915 		if ((cmd.keyflags & TCP_AO_KEYF_IFINDEX) !=
1916 		    (key->keyflags & TCP_AO_KEYF_IFINDEX))
1917 			continue;
1918 
1919 		if (key->l3index != l3index)
1920 			continue;
1921 
1922 		if (key == new_current || key == new_rnext)
1923 			continue;
1924 
1925 		return tcp_ao_delete_key(sk, ao_info, cmd.del_async, key,
1926 					 new_current, new_rnext);
1927 	}
1928 	return -ENOENT;
1929 }
1930 
1931 /* cmd.ao_required makes a socket TCP-AO only.
1932  * Don't allow any md5 keys for any l3intf on the socket together with it.
1933  * Restricting it early in setsockopt() removes a check for
1934  * ao_info->ao_required on inbound tcp segment fast-path.
1935  */
1936 static int tcp_ao_required_verify(struct sock *sk)
1937 {
1938 #ifdef CONFIG_TCP_MD5SIG
1939 	const struct tcp_md5sig_info *md5sig;
1940 
1941 	if (!static_branch_unlikely(&tcp_md5_needed.key))
1942 		return 0;
1943 
1944 	md5sig = rcu_dereference_check(tcp_sk(sk)->md5sig_info,
1945 				       lockdep_sock_is_held(sk));
1946 	if (!md5sig)
1947 		return 0;
1948 
1949 	if (rcu_dereference_check(hlist_first_rcu(&md5sig->head),
1950 				  lockdep_sock_is_held(sk)))
1951 		return 1;
1952 #endif
1953 	return 0;
1954 }
1955 
1956 static int tcp_ao_info_cmd(struct sock *sk, unsigned short int family,
1957 			   sockptr_t optval, int optlen)
1958 {
1959 	struct tcp_ao_key *new_current = NULL, *new_rnext = NULL;
1960 	struct tcp_ao_info *ao_info;
1961 	struct tcp_ao_info_opt cmd;
1962 	bool first = false;
1963 	int err;
1964 
1965 	if (optlen < sizeof(cmd))
1966 		return -EINVAL;
1967 
1968 	err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1969 	if (err)
1970 		return err;
1971 
1972 	if (cmd.set_current || cmd.set_rnext) {
1973 		if (!tcp_ao_can_set_current_rnext(sk))
1974 			return -EINVAL;
1975 	}
1976 
1977 	if (cmd.reserved != 0 || cmd.reserved2 != 0)
1978 		return -EINVAL;
1979 
1980 	ao_info = setsockopt_ao_info(sk);
1981 	if (IS_ERR(ao_info))
1982 		return PTR_ERR(ao_info);
1983 	if (!ao_info) {
1984 		if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1985 			return -EINVAL;
1986 		ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1987 		if (!ao_info)
1988 			return -ENOMEM;
1989 		first = true;
1990 	}
1991 
1992 	if (cmd.ao_required && tcp_ao_required_verify(sk)) {
1993 		err = -EKEYREJECTED;
1994 		goto out;
1995 	}
1996 
1997 	/* For sockets in TCP_CLOSED it's possible set keys that aren't
1998 	 * matching the future peer (address/port/VRF/etc),
1999 	 * tcp_ao_connect_init() will choose a correct matching MKT
2000 	 * if there's any.
2001 	 */
2002 	if (cmd.set_current) {
2003 		new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1);
2004 		if (!new_current) {
2005 			err = -ENOENT;
2006 			goto out;
2007 		}
2008 	}
2009 	if (cmd.set_rnext) {
2010 		new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext);
2011 		if (!new_rnext) {
2012 			err = -ENOENT;
2013 			goto out;
2014 		}
2015 	}
2016 	if (cmd.set_counters) {
2017 		atomic64_set(&ao_info->counters.pkt_good, cmd.pkt_good);
2018 		atomic64_set(&ao_info->counters.pkt_bad, cmd.pkt_bad);
2019 		atomic64_set(&ao_info->counters.key_not_found, cmd.pkt_key_not_found);
2020 		atomic64_set(&ao_info->counters.ao_required, cmd.pkt_ao_required);
2021 		atomic64_set(&ao_info->counters.dropped_icmp, cmd.pkt_dropped_icmp);
2022 	}
2023 
2024 	ao_info->ao_required = cmd.ao_required;
2025 	ao_info->accept_icmps = cmd.accept_icmps;
2026 	if (new_current)
2027 		WRITE_ONCE(ao_info->current_key, new_current);
2028 	if (new_rnext)
2029 		WRITE_ONCE(ao_info->rnext_key, new_rnext);
2030 	if (first) {
2031 		if (!static_branch_inc(&tcp_ao_needed.key)) {
2032 			err = -EUSERS;
2033 			goto out;
2034 		}
2035 		sk_gso_disable(sk);
2036 		rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
2037 	}
2038 	return 0;
2039 out:
2040 	if (first)
2041 		kfree(ao_info);
2042 	return err;
2043 }
2044 
2045 int tcp_parse_ao(struct sock *sk, int cmd, unsigned short int family,
2046 		 sockptr_t optval, int optlen)
2047 {
2048 	if (WARN_ON_ONCE(family != AF_INET && family != AF_INET6))
2049 		return -EAFNOSUPPORT;
2050 
2051 	switch (cmd) {
2052 	case TCP_AO_ADD_KEY:
2053 		return tcp_ao_add_cmd(sk, family, optval, optlen);
2054 	case TCP_AO_DEL_KEY:
2055 		return tcp_ao_del_cmd(sk, family, optval, optlen);
2056 	case TCP_AO_INFO:
2057 		return tcp_ao_info_cmd(sk, family, optval, optlen);
2058 	default:
2059 		WARN_ON_ONCE(1);
2060 		return -EINVAL;
2061 	}
2062 }
2063 
2064 int tcp_v4_parse_ao(struct sock *sk, int cmd, sockptr_t optval, int optlen)
2065 {
2066 	return tcp_parse_ao(sk, cmd, AF_INET, optval, optlen);
2067 }
2068 
2069 /* tcp_ao_copy_mkts_to_user(ao_info, optval, optlen)
2070  *
2071  * @ao_info:	struct tcp_ao_info on the socket that
2072  *		socket getsockopt(TCP_AO_GET_KEYS) is executed on
2073  * @optval:	pointer to array of tcp_ao_getsockopt structures in user space.
2074  *		Must be != NULL.
2075  * @optlen:	pointer to size of tcp_ao_getsockopt structure.
2076  *		Must be != NULL.
2077  *
2078  * Return value: 0 on success, a negative error number otherwise.
2079  *
2080  * optval points to an array of tcp_ao_getsockopt structures in user space.
2081  * optval[0] is used as both input and output to getsockopt. It determines
2082  * which keys are returned by the kernel.
2083  * optval[0].nkeys is the size of the array in user space. On return it contains
2084  * the number of keys matching the search criteria.
2085  * If tcp_ao_getsockopt::get_all is set, then all keys in the socket are
2086  * returned, otherwise only keys matching <addr, prefix, sndid, rcvid>
2087  * in optval[0] are returned.
2088  * optlen is also used as both input and output. The user provides the size
2089  * of struct tcp_ao_getsockopt in user space, and the kernel returns the size
2090  * of the structure in kernel space.
2091  * The size of struct tcp_ao_getsockopt may differ between user and kernel.
2092  * There are three cases to consider:
2093  *  * If usize == ksize, then keys are copied verbatim.
2094  *  * If usize < ksize, then the userspace has passed an old struct to a
2095  *    newer kernel. The rest of the trailing bytes in optval[0]
2096  *    (ksize - usize) are interpreted as 0 by the kernel.
2097  *  * If usize > ksize, then the userspace has passed a new struct to an
2098  *    older kernel. The trailing bytes unknown to the kernel (usize - ksize)
2099  *    are checked to ensure they are zeroed, otherwise -E2BIG is returned.
2100  * On return the kernel fills in min(usize, ksize) in each entry of the array.
2101  * The layout of the fields in the user and kernel structures is expected to
2102  * be the same (including in the 32bit vs 64bit case).
2103  */
2104 static int tcp_ao_copy_mkts_to_user(struct tcp_ao_info *ao_info,
2105 				    sockptr_t optval, sockptr_t optlen)
2106 {
2107 	struct tcp_ao_getsockopt opt_in, opt_out;
2108 	struct tcp_ao_key *key, *current_key;
2109 	bool do_address_matching = true;
2110 	union tcp_ao_addr *addr = NULL;
2111 	int err, l3index, user_len;
2112 	unsigned int max_keys;	/* maximum number of keys to copy to user */
2113 	size_t out_offset = 0;
2114 	size_t bytes_to_write;	/* number of bytes to write to user level */
2115 	u32 matched_keys;	/* keys from ao_info matched so far */
2116 	int optlen_out;
2117 	__be16 port = 0;
2118 
2119 	if (copy_from_sockptr(&user_len, optlen, sizeof(int)))
2120 		return -EFAULT;
2121 
2122 	if (user_len <= 0)
2123 		return -EINVAL;
2124 
2125 	memset(&opt_in, 0, sizeof(struct tcp_ao_getsockopt));
2126 	err = copy_struct_from_sockptr(&opt_in, sizeof(opt_in),
2127 				       optval, user_len);
2128 	if (err < 0)
2129 		return err;
2130 
2131 	if (opt_in.pkt_good || opt_in.pkt_bad)
2132 		return -EINVAL;
2133 	if (opt_in.keyflags & ~TCP_AO_GET_KEYF_VALID)
2134 		return -EINVAL;
2135 	if (opt_in.ifindex && !(opt_in.keyflags & TCP_AO_KEYF_IFINDEX))
2136 		return -EINVAL;
2137 
2138 	if (opt_in.reserved != 0)
2139 		return -EINVAL;
2140 
2141 	max_keys = opt_in.nkeys;
2142 	l3index = (opt_in.keyflags & TCP_AO_KEYF_IFINDEX) ? opt_in.ifindex : -1;
2143 
2144 	if (opt_in.get_all || opt_in.is_current || opt_in.is_rnext) {
2145 		if (opt_in.get_all && (opt_in.is_current || opt_in.is_rnext))
2146 			return -EINVAL;
2147 		do_address_matching = false;
2148 	}
2149 
2150 	switch (opt_in.addr.ss_family) {
2151 	case AF_INET: {
2152 		struct sockaddr_in *sin;
2153 		__be32 mask;
2154 
2155 		sin = (struct sockaddr_in *)&opt_in.addr;
2156 		port = sin->sin_port;
2157 		addr = (union tcp_ao_addr *)&sin->sin_addr;
2158 
2159 		if (opt_in.prefix > 32)
2160 			return -EINVAL;
2161 
2162 		if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY &&
2163 		    opt_in.prefix != 0)
2164 			return -EINVAL;
2165 
2166 		mask = inet_make_mask(opt_in.prefix);
2167 		if (sin->sin_addr.s_addr & ~mask)
2168 			return -EINVAL;
2169 
2170 		break;
2171 	}
2172 	case AF_INET6: {
2173 		struct sockaddr_in6 *sin6;
2174 		struct in6_addr *addr6;
2175 
2176 		sin6 = (struct sockaddr_in6 *)&opt_in.addr;
2177 		addr = (union tcp_ao_addr *)&sin6->sin6_addr;
2178 		addr6 = &sin6->sin6_addr;
2179 		port = sin6->sin6_port;
2180 
2181 		/* We don't have to change family and @addr here if
2182 		 * ipv6_addr_v4mapped() like in key adding:
2183 		 * tcp_ao_key_cmp() does it. Do the sanity checks though.
2184 		 */
2185 		if (opt_in.prefix != 0) {
2186 			if (ipv6_addr_v4mapped(addr6)) {
2187 				__be32 mask, addr4 = addr6->s6_addr32[3];
2188 
2189 				if (opt_in.prefix > 32 ||
2190 				    ntohl(addr4) == INADDR_ANY)
2191 					return -EINVAL;
2192 				mask = inet_make_mask(opt_in.prefix);
2193 				if (addr4 & ~mask)
2194 					return -EINVAL;
2195 			} else {
2196 				struct in6_addr pfx;
2197 
2198 				if (ipv6_addr_any(addr6) ||
2199 				    opt_in.prefix > 128)
2200 					return -EINVAL;
2201 
2202 				ipv6_addr_prefix(&pfx, addr6, opt_in.prefix);
2203 				if (ipv6_addr_cmp(&pfx, addr6))
2204 					return -EINVAL;
2205 			}
2206 		} else if (!ipv6_addr_any(addr6)) {
2207 			return -EINVAL;
2208 		}
2209 		break;
2210 	}
2211 	case 0:
2212 		if (!do_address_matching)
2213 			break;
2214 		fallthrough;
2215 	default:
2216 		return -EAFNOSUPPORT;
2217 	}
2218 
2219 	if (!do_address_matching) {
2220 		/* We could just ignore those, but let's do stricter checks */
2221 		if (addr || port)
2222 			return -EINVAL;
2223 		if (opt_in.prefix || opt_in.sndid || opt_in.rcvid)
2224 			return -EINVAL;
2225 	}
2226 
2227 	bytes_to_write = min_t(int, user_len, sizeof(struct tcp_ao_getsockopt));
2228 	matched_keys = 0;
2229 	/* May change in RX, while we're dumping, pre-fetch it */
2230 	current_key = READ_ONCE(ao_info->current_key);
2231 
2232 	hlist_for_each_entry_rcu(key, &ao_info->head, node) {
2233 		if (opt_in.get_all)
2234 			goto match;
2235 
2236 		if (opt_in.is_current || opt_in.is_rnext) {
2237 			if (opt_in.is_current && key == current_key)
2238 				goto match;
2239 			if (opt_in.is_rnext && key == ao_info->rnext_key)
2240 				goto match;
2241 			continue;
2242 		}
2243 
2244 		if (tcp_ao_key_cmp(key, l3index, addr, opt_in.prefix,
2245 				   opt_in.addr.ss_family,
2246 				   opt_in.sndid, opt_in.rcvid) != 0)
2247 			continue;
2248 match:
2249 		matched_keys++;
2250 		if (matched_keys > max_keys)
2251 			continue;
2252 
2253 		memset(&opt_out, 0, sizeof(struct tcp_ao_getsockopt));
2254 
2255 		if (key->family == AF_INET) {
2256 			struct sockaddr_in *sin_out = (struct sockaddr_in *)&opt_out.addr;
2257 
2258 			sin_out->sin_family = key->family;
2259 			sin_out->sin_port = 0;
2260 			memcpy(&sin_out->sin_addr, &key->addr, sizeof(struct in_addr));
2261 		} else {
2262 			struct sockaddr_in6 *sin6_out = (struct sockaddr_in6 *)&opt_out.addr;
2263 
2264 			sin6_out->sin6_family = key->family;
2265 			sin6_out->sin6_port = 0;
2266 			memcpy(&sin6_out->sin6_addr, &key->addr, sizeof(struct in6_addr));
2267 		}
2268 		opt_out.sndid = key->sndid;
2269 		opt_out.rcvid = key->rcvid;
2270 		opt_out.prefix = key->prefixlen;
2271 		opt_out.keyflags = key->keyflags;
2272 		opt_out.is_current = (key == current_key);
2273 		opt_out.is_rnext = (key == ao_info->rnext_key);
2274 		opt_out.nkeys = 0;
2275 		opt_out.maclen = key->maclen;
2276 		opt_out.keylen = key->keylen;
2277 		opt_out.ifindex = key->l3index;
2278 		opt_out.pkt_good = atomic64_read(&key->pkt_good);
2279 		opt_out.pkt_bad = atomic64_read(&key->pkt_bad);
2280 		memcpy(&opt_out.key, key->key, key->keylen);
2281 		tcp_sigpool_algo(key->tcp_sigpool_id, opt_out.alg_name, 64);
2282 
2283 		/* Copy key to user */
2284 		if (copy_to_sockptr_offset(optval, out_offset,
2285 					   &opt_out, bytes_to_write))
2286 			return -EFAULT;
2287 		out_offset += user_len;
2288 	}
2289 
2290 	optlen_out = (int)sizeof(struct tcp_ao_getsockopt);
2291 	if (copy_to_sockptr(optlen, &optlen_out, sizeof(int)))
2292 		return -EFAULT;
2293 
2294 	out_offset = offsetof(struct tcp_ao_getsockopt, nkeys);
2295 	if (copy_to_sockptr_offset(optval, out_offset,
2296 				   &matched_keys, sizeof(u32)))
2297 		return -EFAULT;
2298 
2299 	return 0;
2300 }
2301 
2302 int tcp_ao_get_mkts(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2303 {
2304 	struct tcp_ao_info *ao_info;
2305 
2306 	ao_info = setsockopt_ao_info(sk);
2307 	if (IS_ERR(ao_info))
2308 		return PTR_ERR(ao_info);
2309 	if (!ao_info)
2310 		return -ENOENT;
2311 
2312 	return tcp_ao_copy_mkts_to_user(ao_info, optval, optlen);
2313 }
2314 
2315 int tcp_ao_get_sock_info(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2316 {
2317 	struct tcp_ao_info_opt out, in = {};
2318 	struct tcp_ao_key *current_key;
2319 	struct tcp_ao_info *ao;
2320 	int err, len;
2321 
2322 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
2323 		return -EFAULT;
2324 
2325 	if (len <= 0)
2326 		return -EINVAL;
2327 
2328 	/* Copying this "in" only to check ::reserved, ::reserved2,
2329 	 * that may be needed to extend (struct tcp_ao_info_opt) and
2330 	 * what getsockopt() provides in future.
2331 	 */
2332 	err = copy_struct_from_sockptr(&in, sizeof(in), optval, len);
2333 	if (err)
2334 		return err;
2335 
2336 	if (in.reserved != 0 || in.reserved2 != 0)
2337 		return -EINVAL;
2338 
2339 	ao = setsockopt_ao_info(sk);
2340 	if (IS_ERR(ao))
2341 		return PTR_ERR(ao);
2342 	if (!ao)
2343 		return -ENOENT;
2344 
2345 	memset(&out, 0, sizeof(out));
2346 	out.ao_required		= ao->ao_required;
2347 	out.accept_icmps	= ao->accept_icmps;
2348 	out.pkt_good		= atomic64_read(&ao->counters.pkt_good);
2349 	out.pkt_bad		= atomic64_read(&ao->counters.pkt_bad);
2350 	out.pkt_key_not_found	= atomic64_read(&ao->counters.key_not_found);
2351 	out.pkt_ao_required	= atomic64_read(&ao->counters.ao_required);
2352 	out.pkt_dropped_icmp	= atomic64_read(&ao->counters.dropped_icmp);
2353 
2354 	current_key = READ_ONCE(ao->current_key);
2355 	if (current_key) {
2356 		out.set_current = 1;
2357 		out.current_key = current_key->sndid;
2358 	}
2359 	if (ao->rnext_key) {
2360 		out.set_rnext = 1;
2361 		out.rnext = ao->rnext_key->rcvid;
2362 	}
2363 
2364 	if (copy_to_sockptr(optval, &out, min_t(int, len, sizeof(out))))
2365 		return -EFAULT;
2366 
2367 	return 0;
2368 }
2369 
2370 int tcp_ao_set_repair(struct sock *sk, sockptr_t optval, unsigned int optlen)
2371 {
2372 	struct tcp_sock *tp = tcp_sk(sk);
2373 	struct tcp_ao_repair cmd;
2374 	struct tcp_ao_key *key;
2375 	struct tcp_ao_info *ao;
2376 	int err;
2377 
2378 	if (optlen < sizeof(cmd))
2379 		return -EINVAL;
2380 
2381 	err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
2382 	if (err)
2383 		return err;
2384 
2385 	if (!tp->repair)
2386 		return -EPERM;
2387 
2388 	ao = setsockopt_ao_info(sk);
2389 	if (IS_ERR(ao))
2390 		return PTR_ERR(ao);
2391 	if (!ao)
2392 		return -ENOENT;
2393 
2394 	WRITE_ONCE(ao->lisn, cmd.snt_isn);
2395 	WRITE_ONCE(ao->risn, cmd.rcv_isn);
2396 	WRITE_ONCE(ao->snd_sne, cmd.snd_sne);
2397 	WRITE_ONCE(ao->rcv_sne, cmd.rcv_sne);
2398 
2399 	hlist_for_each_entry_rcu(key, &ao->head, node)
2400 		tcp_ao_cache_traffic_keys(sk, ao, key);
2401 
2402 	return 0;
2403 }
2404 
2405 int tcp_ao_get_repair(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2406 {
2407 	struct tcp_sock *tp = tcp_sk(sk);
2408 	struct tcp_ao_repair opt;
2409 	struct tcp_ao_info *ao;
2410 	int len;
2411 
2412 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
2413 		return -EFAULT;
2414 
2415 	if (len <= 0)
2416 		return -EINVAL;
2417 
2418 	if (!tp->repair)
2419 		return -EPERM;
2420 
2421 	rcu_read_lock();
2422 	ao = getsockopt_ao_info(sk);
2423 	if (IS_ERR_OR_NULL(ao)) {
2424 		rcu_read_unlock();
2425 		return ao ? PTR_ERR(ao) : -ENOENT;
2426 	}
2427 
2428 	opt.snt_isn	= ao->lisn;
2429 	opt.rcv_isn	= ao->risn;
2430 	opt.snd_sne	= READ_ONCE(ao->snd_sne);
2431 	opt.rcv_sne	= READ_ONCE(ao->rcv_sne);
2432 	rcu_read_unlock();
2433 
2434 	if (copy_to_sockptr(optval, &opt, min_t(int, len, sizeof(opt))))
2435 		return -EFAULT;
2436 	return 0;
2437 }
2438