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