1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic INET transport hashtables
9 * Authors: Lotsa people, from code originally in tcp
12 #include <linux/module.h>
13 #include <linux/random.h>
14 #include <linux/sched.h>
15 #include <linux/slab.h>
16 #include <linux/wait.h>
17 #include <linux/vmalloc.h>
18 #include <linux/memblock.h>
20 #include <net/addrconf.h>
21 #include <net/inet_connection_sock.h>
22 #include <net/inet_hashtables.h>
23 #if IS_ENABLED(CONFIG_IPV6)
24 #include <net/inet6_hashtables.h>
26 #include <net/secure_seq.h>
29 #include <net/sock_reuseport.h>
31 static u32 inet_ehashfn(const struct net *net, const __be32 laddr,
32 const __u16 lport, const __be32 faddr,
35 static u32 inet_ehash_secret __read_mostly;
37 net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
39 return __inet_ehashfn(laddr, lport, faddr, fport,
40 inet_ehash_secret + net_hash_mix(net));
43 /* This function handles inet_sock, but also timewait and request sockets
46 static u32 sk_ehashfn(const struct sock *sk)
48 #if IS_ENABLED(CONFIG_IPV6)
49 if (sk->sk_family == AF_INET6 &&
50 !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
51 return inet6_ehashfn(sock_net(sk),
52 &sk->sk_v6_rcv_saddr, sk->sk_num,
53 &sk->sk_v6_daddr, sk->sk_dport);
55 return inet_ehashfn(sock_net(sk),
56 sk->sk_rcv_saddr, sk->sk_num,
57 sk->sk_daddr, sk->sk_dport);
61 * Allocate and initialize a new local port bind bucket.
62 * The bindhash mutex for snum's hash chain must be held here.
64 struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
66 struct inet_bind_hashbucket *head,
67 const unsigned short snum,
70 struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
73 write_pnet(&tb->ib_net, net);
77 tb->fastreuseport = 0;
78 INIT_HLIST_HEAD(&tb->owners);
79 hlist_add_head(&tb->node, &head->chain);
85 * Caller must hold hashbucket lock for this tb with local BH disabled
87 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
89 if (hlist_empty(&tb->owners)) {
90 __hlist_del(&tb->node);
91 kmem_cache_free(cachep, tb);
95 bool inet_bind_bucket_match(const struct inet_bind_bucket *tb, const struct net *net,
96 unsigned short port, int l3mdev)
98 return net_eq(ib_net(tb), net) && tb->port == port &&
102 static void inet_bind2_bucket_init(struct inet_bind2_bucket *tb,
104 struct inet_bind_hashbucket *head,
105 unsigned short port, int l3mdev,
106 const struct sock *sk)
108 write_pnet(&tb->ib_net, net);
111 #if IS_ENABLED(CONFIG_IPV6)
112 tb->family = sk->sk_family;
113 if (sk->sk_family == AF_INET6)
114 tb->v6_rcv_saddr = sk->sk_v6_rcv_saddr;
117 tb->rcv_saddr = sk->sk_rcv_saddr;
118 INIT_HLIST_HEAD(&tb->owners);
119 INIT_HLIST_HEAD(&tb->deathrow);
120 hlist_add_head(&tb->node, &head->chain);
123 struct inet_bind2_bucket *inet_bind2_bucket_create(struct kmem_cache *cachep,
125 struct inet_bind_hashbucket *head,
128 const struct sock *sk)
130 struct inet_bind2_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
133 inet_bind2_bucket_init(tb, net, head, port, l3mdev, sk);
138 /* Caller must hold hashbucket lock for this tb with local BH disabled */
139 void inet_bind2_bucket_destroy(struct kmem_cache *cachep, struct inet_bind2_bucket *tb)
141 if (hlist_empty(&tb->owners) && hlist_empty(&tb->deathrow)) {
142 __hlist_del(&tb->node);
143 kmem_cache_free(cachep, tb);
147 static bool inet_bind2_bucket_addr_match(const struct inet_bind2_bucket *tb2,
148 const struct sock *sk)
150 #if IS_ENABLED(CONFIG_IPV6)
151 if (sk->sk_family != tb2->family)
154 if (sk->sk_family == AF_INET6)
155 return ipv6_addr_equal(&tb2->v6_rcv_saddr,
156 &sk->sk_v6_rcv_saddr);
158 return tb2->rcv_saddr == sk->sk_rcv_saddr;
161 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
162 struct inet_bind2_bucket *tb2, unsigned short port)
164 inet_sk(sk)->inet_num = port;
165 sk_add_bind_node(sk, &tb->owners);
166 inet_csk(sk)->icsk_bind_hash = tb;
167 sk_add_bind2_node(sk, &tb2->owners);
168 inet_csk(sk)->icsk_bind2_hash = tb2;
172 * Get rid of any references to a local port held by the given sock.
174 static void __inet_put_port(struct sock *sk)
176 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
177 struct inet_bind_hashbucket *head, *head2;
178 struct net *net = sock_net(sk);
179 struct inet_bind_bucket *tb;
182 bhash = inet_bhashfn(net, inet_sk(sk)->inet_num, hashinfo->bhash_size);
183 head = &hashinfo->bhash[bhash];
184 head2 = inet_bhashfn_portaddr(hashinfo, sk, net, inet_sk(sk)->inet_num);
186 spin_lock(&head->lock);
187 tb = inet_csk(sk)->icsk_bind_hash;
188 __sk_del_bind_node(sk);
189 inet_csk(sk)->icsk_bind_hash = NULL;
190 inet_sk(sk)->inet_num = 0;
191 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
193 spin_lock(&head2->lock);
194 if (inet_csk(sk)->icsk_bind2_hash) {
195 struct inet_bind2_bucket *tb2 = inet_csk(sk)->icsk_bind2_hash;
197 __sk_del_bind2_node(sk);
198 inet_csk(sk)->icsk_bind2_hash = NULL;
199 inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
201 spin_unlock(&head2->lock);
203 spin_unlock(&head->lock);
206 void inet_put_port(struct sock *sk)
212 EXPORT_SYMBOL(inet_put_port);
214 int __inet_inherit_port(const struct sock *sk, struct sock *child)
216 struct inet_hashinfo *table = tcp_or_dccp_get_hashinfo(sk);
217 unsigned short port = inet_sk(child)->inet_num;
218 struct inet_bind_hashbucket *head, *head2;
219 bool created_inet_bind_bucket = false;
220 struct net *net = sock_net(sk);
221 bool update_fastreuse = false;
222 struct inet_bind2_bucket *tb2;
223 struct inet_bind_bucket *tb;
226 bhash = inet_bhashfn(net, port, table->bhash_size);
227 head = &table->bhash[bhash];
228 head2 = inet_bhashfn_portaddr(table, child, net, port);
230 spin_lock(&head->lock);
231 spin_lock(&head2->lock);
232 tb = inet_csk(sk)->icsk_bind_hash;
233 tb2 = inet_csk(sk)->icsk_bind2_hash;
234 if (unlikely(!tb || !tb2)) {
235 spin_unlock(&head2->lock);
236 spin_unlock(&head->lock);
239 if (tb->port != port) {
240 l3mdev = inet_sk_bound_l3mdev(sk);
242 /* NOTE: using tproxy and redirecting skbs to a proxy
243 * on a different listener port breaks the assumption
244 * that the listener socket's icsk_bind_hash is the same
245 * as that of the child socket. We have to look up or
246 * create a new bind bucket for the child here. */
247 inet_bind_bucket_for_each(tb, &head->chain) {
248 if (inet_bind_bucket_match(tb, net, port, l3mdev))
252 tb = inet_bind_bucket_create(table->bind_bucket_cachep,
253 net, head, port, l3mdev);
255 spin_unlock(&head2->lock);
256 spin_unlock(&head->lock);
259 created_inet_bind_bucket = true;
261 update_fastreuse = true;
264 } else if (!inet_bind2_bucket_addr_match(tb2, child)) {
265 l3mdev = inet_sk_bound_l3mdev(sk);
268 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, child);
270 tb2 = inet_bind2_bucket_create(table->bind2_bucket_cachep,
277 if (update_fastreuse)
278 inet_csk_update_fastreuse(tb, child);
279 inet_bind_hash(child, tb, tb2, port);
280 spin_unlock(&head2->lock);
281 spin_unlock(&head->lock);
286 if (created_inet_bind_bucket)
287 inet_bind_bucket_destroy(table->bind_bucket_cachep, tb);
288 spin_unlock(&head2->lock);
289 spin_unlock(&head->lock);
292 EXPORT_SYMBOL_GPL(__inet_inherit_port);
294 static struct inet_listen_hashbucket *
295 inet_lhash2_bucket_sk(struct inet_hashinfo *h, struct sock *sk)
299 #if IS_ENABLED(CONFIG_IPV6)
300 if (sk->sk_family == AF_INET6)
301 hash = ipv6_portaddr_hash(sock_net(sk),
302 &sk->sk_v6_rcv_saddr,
303 inet_sk(sk)->inet_num);
306 hash = ipv4_portaddr_hash(sock_net(sk),
307 inet_sk(sk)->inet_rcv_saddr,
308 inet_sk(sk)->inet_num);
309 return inet_lhash2_bucket(h, hash);
312 static inline int compute_score(struct sock *sk, struct net *net,
313 const unsigned short hnum, const __be32 daddr,
314 const int dif, const int sdif)
318 if (net_eq(sock_net(sk), net) && sk->sk_num == hnum &&
319 !ipv6_only_sock(sk)) {
320 if (sk->sk_rcv_saddr != daddr)
323 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
325 score = sk->sk_bound_dev_if ? 2 : 1;
327 if (sk->sk_family == PF_INET)
329 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
335 static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
336 struct sk_buff *skb, int doff,
337 __be32 saddr, __be16 sport,
338 __be32 daddr, unsigned short hnum)
340 struct sock *reuse_sk = NULL;
343 if (sk->sk_reuseport) {
344 phash = inet_ehashfn(net, daddr, hnum, saddr, sport);
345 reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
351 * Here are some nice properties to exploit here. The BSD API
352 * does not allow a listening sock to specify the remote port nor the
353 * remote address for the connection. So always assume those are both
354 * wildcarded during the search since they can never be otherwise.
357 /* called with rcu_read_lock() : No refcount taken on the socket */
358 static struct sock *inet_lhash2_lookup(struct net *net,
359 struct inet_listen_hashbucket *ilb2,
360 struct sk_buff *skb, int doff,
361 const __be32 saddr, __be16 sport,
362 const __be32 daddr, const unsigned short hnum,
363 const int dif, const int sdif)
365 struct sock *sk, *result = NULL;
366 struct hlist_nulls_node *node;
367 int score, hiscore = 0;
369 sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
370 score = compute_score(sk, net, hnum, daddr, dif, sdif);
371 if (score > hiscore) {
372 result = lookup_reuseport(net, sk, skb, doff,
373 saddr, sport, daddr, hnum);
385 static inline struct sock *inet_lookup_run_bpf(struct net *net,
386 struct inet_hashinfo *hashinfo,
387 struct sk_buff *skb, int doff,
388 __be32 saddr, __be16 sport,
389 __be32 daddr, u16 hnum, const int dif)
391 struct sock *sk, *reuse_sk;
394 if (hashinfo != net->ipv4.tcp_death_row.hashinfo)
395 return NULL; /* only TCP is supported */
397 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_TCP, saddr, sport,
398 daddr, hnum, dif, &sk);
399 if (no_reuseport || IS_ERR_OR_NULL(sk))
402 reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
408 struct sock *__inet_lookup_listener(struct net *net,
409 struct inet_hashinfo *hashinfo,
410 struct sk_buff *skb, int doff,
411 const __be32 saddr, __be16 sport,
412 const __be32 daddr, const unsigned short hnum,
413 const int dif, const int sdif)
415 struct inet_listen_hashbucket *ilb2;
416 struct sock *result = NULL;
419 /* Lookup redirect from BPF */
420 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
421 result = inet_lookup_run_bpf(net, hashinfo, skb, doff,
422 saddr, sport, daddr, hnum, dif);
427 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
428 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
430 result = inet_lhash2_lookup(net, ilb2, skb, doff,
431 saddr, sport, daddr, hnum,
436 /* Lookup lhash2 with INADDR_ANY */
437 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
438 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
440 result = inet_lhash2_lookup(net, ilb2, skb, doff,
441 saddr, sport, htonl(INADDR_ANY), hnum,
448 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
450 /* All sockets share common refcount, but have different destructors */
451 void sock_gen_put(struct sock *sk)
453 if (!refcount_dec_and_test(&sk->sk_refcnt))
456 if (sk->sk_state == TCP_TIME_WAIT)
457 inet_twsk_free(inet_twsk(sk));
458 else if (sk->sk_state == TCP_NEW_SYN_RECV)
459 reqsk_free(inet_reqsk(sk));
463 EXPORT_SYMBOL_GPL(sock_gen_put);
465 void sock_edemux(struct sk_buff *skb)
467 sock_gen_put(skb->sk);
469 EXPORT_SYMBOL(sock_edemux);
471 struct sock *__inet_lookup_established(struct net *net,
472 struct inet_hashinfo *hashinfo,
473 const __be32 saddr, const __be16 sport,
474 const __be32 daddr, const u16 hnum,
475 const int dif, const int sdif)
477 INET_ADDR_COOKIE(acookie, saddr, daddr);
478 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
480 const struct hlist_nulls_node *node;
481 /* Optimize here for direct hit, only listening connections can
482 * have wildcards anyways.
484 unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
485 unsigned int slot = hash & hashinfo->ehash_mask;
486 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
489 sk_nulls_for_each_rcu(sk, node, &head->chain) {
490 if (sk->sk_hash != hash)
492 if (likely(inet_match(net, sk, acookie, ports, dif, sdif))) {
493 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
495 if (unlikely(!inet_match(net, sk, acookie,
496 ports, dif, sdif))) {
504 * if the nulls value we got at the end of this lookup is
505 * not the expected one, we must restart lookup.
506 * We probably met an item that was moved to another chain.
508 if (get_nulls_value(node) != slot)
515 EXPORT_SYMBOL_GPL(__inet_lookup_established);
517 /* called with local bh disabled */
518 static int __inet_check_established(struct inet_timewait_death_row *death_row,
519 struct sock *sk, __u16 lport,
520 struct inet_timewait_sock **twp)
522 struct inet_hashinfo *hinfo = death_row->hashinfo;
523 struct inet_sock *inet = inet_sk(sk);
524 __be32 daddr = inet->inet_rcv_saddr;
525 __be32 saddr = inet->inet_daddr;
526 int dif = sk->sk_bound_dev_if;
527 struct net *net = sock_net(sk);
528 int sdif = l3mdev_master_ifindex_by_index(net, dif);
529 INET_ADDR_COOKIE(acookie, saddr, daddr);
530 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
531 unsigned int hash = inet_ehashfn(net, daddr, lport,
532 saddr, inet->inet_dport);
533 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
534 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
536 const struct hlist_nulls_node *node;
537 struct inet_timewait_sock *tw = NULL;
541 sk_nulls_for_each(sk2, node, &head->chain) {
542 if (sk2->sk_hash != hash)
545 if (likely(inet_match(net, sk2, acookie, ports, dif, sdif))) {
546 if (sk2->sk_state == TCP_TIME_WAIT) {
548 if (twsk_unique(sk, sk2, twp))
555 /* Must record num and sport now. Otherwise we will see
556 * in hash table socket with a funny identity.
558 inet->inet_num = lport;
559 inet->inet_sport = htons(lport);
561 WARN_ON(!sk_unhashed(sk));
562 __sk_nulls_add_node_rcu(sk, &head->chain);
564 sk_nulls_del_node_init_rcu((struct sock *)tw);
565 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
568 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
573 /* Silly. Should hash-dance instead... */
574 inet_twsk_deschedule_put(tw);
580 return -EADDRNOTAVAIL;
583 static u64 inet_sk_port_offset(const struct sock *sk)
585 const struct inet_sock *inet = inet_sk(sk);
587 return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
592 /* Searches for an exsiting socket in the ehash bucket list.
593 * Returns true if found, false otherwise.
595 static bool inet_ehash_lookup_by_sk(struct sock *sk,
596 struct hlist_nulls_head *list)
598 const __portpair ports = INET_COMBINED_PORTS(sk->sk_dport, sk->sk_num);
599 const int sdif = sk->sk_bound_dev_if;
600 const int dif = sk->sk_bound_dev_if;
601 const struct hlist_nulls_node *node;
602 struct net *net = sock_net(sk);
605 INET_ADDR_COOKIE(acookie, sk->sk_daddr, sk->sk_rcv_saddr);
607 sk_nulls_for_each_rcu(esk, node, list) {
608 if (esk->sk_hash != sk->sk_hash)
610 if (sk->sk_family == AF_INET) {
611 if (unlikely(inet_match(net, esk, acookie,
612 ports, dif, sdif))) {
616 #if IS_ENABLED(CONFIG_IPV6)
617 else if (sk->sk_family == AF_INET6) {
618 if (unlikely(inet6_match(net, esk,
620 &sk->sk_v6_rcv_saddr,
621 ports, dif, sdif))) {
630 /* Insert a socket into ehash, and eventually remove another one
631 * (The another one can be a SYN_RECV or TIMEWAIT)
632 * If an existing socket already exists, socket sk is not inserted,
633 * and sets found_dup_sk parameter to true.
635 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk)
637 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
638 struct inet_ehash_bucket *head;
639 struct hlist_nulls_head *list;
643 WARN_ON_ONCE(!sk_unhashed(sk));
645 sk->sk_hash = sk_ehashfn(sk);
646 head = inet_ehash_bucket(hashinfo, sk->sk_hash);
648 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
652 WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
653 ret = sk_hashed(osk);
655 /* Before deleting the node, we insert a new one to make
656 * sure that the look-up-sk process would not miss either
657 * of them and that at least one node would exist in ehash
658 * table all the time. Otherwise there's a tiny chance
659 * that lookup process could find nothing in ehash table.
661 __sk_nulls_add_node_tail_rcu(sk, list);
662 sk_nulls_del_node_init_rcu(osk);
667 *found_dup_sk = inet_ehash_lookup_by_sk(sk, list);
673 __sk_nulls_add_node_rcu(sk, list);
681 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk, bool *found_dup_sk)
683 bool ok = inet_ehash_insert(sk, osk, found_dup_sk);
686 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
688 this_cpu_inc(*sk->sk_prot->orphan_count);
689 inet_sk_set_state(sk, TCP_CLOSE);
690 sock_set_flag(sk, SOCK_DEAD);
691 inet_csk_destroy_sock(sk);
695 EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
697 static int inet_reuseport_add_sock(struct sock *sk,
698 struct inet_listen_hashbucket *ilb)
700 struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
701 const struct hlist_nulls_node *node;
703 kuid_t uid = sock_i_uid(sk);
705 sk_nulls_for_each_rcu(sk2, node, &ilb->nulls_head) {
707 sk2->sk_family == sk->sk_family &&
708 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
709 sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
710 inet_csk(sk2)->icsk_bind_hash == tb &&
711 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
712 inet_rcv_saddr_equal(sk, sk2, false))
713 return reuseport_add_sock(sk, sk2,
714 inet_rcv_saddr_any(sk));
717 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
720 int __inet_hash(struct sock *sk, struct sock *osk)
722 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
723 struct inet_listen_hashbucket *ilb2;
726 if (sk->sk_state != TCP_LISTEN) {
728 inet_ehash_nolisten(sk, osk, NULL);
732 WARN_ON(!sk_unhashed(sk));
733 ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
735 spin_lock(&ilb2->lock);
736 if (sk->sk_reuseport) {
737 err = inet_reuseport_add_sock(sk, ilb2);
741 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
742 sk->sk_family == AF_INET6)
743 __sk_nulls_add_node_tail_rcu(sk, &ilb2->nulls_head);
745 __sk_nulls_add_node_rcu(sk, &ilb2->nulls_head);
746 sock_set_flag(sk, SOCK_RCU_FREE);
747 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
749 spin_unlock(&ilb2->lock);
753 EXPORT_SYMBOL(__inet_hash);
755 int inet_hash(struct sock *sk)
759 if (sk->sk_state != TCP_CLOSE)
760 err = __inet_hash(sk, NULL);
764 EXPORT_SYMBOL_GPL(inet_hash);
766 void inet_unhash(struct sock *sk)
768 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
773 if (sk->sk_state == TCP_LISTEN) {
774 struct inet_listen_hashbucket *ilb2;
776 ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
777 /* Don't disable bottom halves while acquiring the lock to
778 * avoid circular locking dependency on PREEMPT_RT.
780 spin_lock(&ilb2->lock);
781 if (sk_unhashed(sk)) {
782 spin_unlock(&ilb2->lock);
786 if (rcu_access_pointer(sk->sk_reuseport_cb))
787 reuseport_stop_listen_sock(sk);
789 __sk_nulls_del_node_init_rcu(sk);
790 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
791 spin_unlock(&ilb2->lock);
793 spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
796 if (sk_unhashed(sk)) {
797 spin_unlock_bh(lock);
800 __sk_nulls_del_node_init_rcu(sk);
801 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
802 spin_unlock_bh(lock);
805 EXPORT_SYMBOL_GPL(inet_unhash);
807 static bool inet_bind2_bucket_match(const struct inet_bind2_bucket *tb,
808 const struct net *net, unsigned short port,
809 int l3mdev, const struct sock *sk)
811 #if IS_ENABLED(CONFIG_IPV6)
812 if (sk->sk_family != tb->family)
815 if (sk->sk_family == AF_INET6)
816 return net_eq(ib2_net(tb), net) && tb->port == port &&
817 tb->l3mdev == l3mdev &&
818 ipv6_addr_equal(&tb->v6_rcv_saddr, &sk->sk_v6_rcv_saddr);
821 return net_eq(ib2_net(tb), net) && tb->port == port &&
822 tb->l3mdev == l3mdev && tb->rcv_saddr == sk->sk_rcv_saddr;
825 bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb, const struct net *net,
826 unsigned short port, int l3mdev, const struct sock *sk)
828 #if IS_ENABLED(CONFIG_IPV6)
829 if (sk->sk_family != tb->family) {
830 if (sk->sk_family == AF_INET)
831 return net_eq(ib2_net(tb), net) && tb->port == port &&
832 tb->l3mdev == l3mdev &&
833 ipv6_addr_any(&tb->v6_rcv_saddr);
838 if (sk->sk_family == AF_INET6)
839 return net_eq(ib2_net(tb), net) && tb->port == port &&
840 tb->l3mdev == l3mdev &&
841 ipv6_addr_any(&tb->v6_rcv_saddr);
844 return net_eq(ib2_net(tb), net) && tb->port == port &&
845 tb->l3mdev == l3mdev && tb->rcv_saddr == 0;
848 /* The socket's bhash2 hashbucket spinlock must be held when this is called */
849 struct inet_bind2_bucket *
850 inet_bind2_bucket_find(const struct inet_bind_hashbucket *head, const struct net *net,
851 unsigned short port, int l3mdev, const struct sock *sk)
853 struct inet_bind2_bucket *bhash2 = NULL;
855 inet_bind_bucket_for_each(bhash2, &head->chain)
856 if (inet_bind2_bucket_match(bhash2, net, port, l3mdev, sk))
862 struct inet_bind_hashbucket *
863 inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port)
865 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
868 #if IS_ENABLED(CONFIG_IPV6)
869 if (sk->sk_family == AF_INET6)
870 hash = ipv6_portaddr_hash(net, &in6addr_any, port);
873 hash = ipv4_portaddr_hash(net, 0, port);
875 return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
878 static void inet_update_saddr(struct sock *sk, void *saddr, int family)
880 if (family == AF_INET) {
881 inet_sk(sk)->inet_saddr = *(__be32 *)saddr;
882 sk_rcv_saddr_set(sk, inet_sk(sk)->inet_saddr);
884 #if IS_ENABLED(CONFIG_IPV6)
886 sk->sk_v6_rcv_saddr = *(struct in6_addr *)saddr;
891 static int __inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family, bool reset)
893 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
894 struct inet_bind_hashbucket *head, *head2;
895 struct inet_bind2_bucket *tb2, *new_tb2;
896 int l3mdev = inet_sk_bound_l3mdev(sk);
897 int port = inet_sk(sk)->inet_num;
898 struct net *net = sock_net(sk);
901 if (!inet_csk(sk)->icsk_bind2_hash) {
902 /* Not bind()ed before. */
904 inet_reset_saddr(sk);
906 inet_update_saddr(sk, saddr, family);
911 /* Allocate a bind2 bucket ahead of time to avoid permanently putting
912 * the bhash2 table in an inconsistent state if a new tb2 bucket
915 new_tb2 = kmem_cache_alloc(hinfo->bind2_bucket_cachep, GFP_ATOMIC);
918 /* The (INADDR_ANY, port) bucket might have already
919 * been freed, then we cannot fixup icsk_bind2_hash,
920 * so we give up and unlink sk from bhash/bhash2 not
921 * to leave inconsistency in bhash2.
924 inet_reset_saddr(sk);
930 bhash = inet_bhashfn(net, port, hinfo->bhash_size);
931 head = &hinfo->bhash[bhash];
932 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
934 /* If we change saddr locklessly, another thread
935 * iterating over bhash might see corrupted address.
937 spin_lock_bh(&head->lock);
939 spin_lock(&head2->lock);
940 __sk_del_bind2_node(sk);
941 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, inet_csk(sk)->icsk_bind2_hash);
942 spin_unlock(&head2->lock);
945 inet_reset_saddr(sk);
947 inet_update_saddr(sk, saddr, family);
949 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
951 spin_lock(&head2->lock);
952 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
955 inet_bind2_bucket_init(tb2, net, head2, port, l3mdev, sk);
957 sk_add_bind2_node(sk, &tb2->owners);
958 inet_csk(sk)->icsk_bind2_hash = tb2;
959 spin_unlock(&head2->lock);
961 spin_unlock_bh(&head->lock);
964 kmem_cache_free(hinfo->bind2_bucket_cachep, new_tb2);
969 int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family)
971 return __inet_bhash2_update_saddr(sk, saddr, family, false);
973 EXPORT_SYMBOL_GPL(inet_bhash2_update_saddr);
975 void inet_bhash2_reset_saddr(struct sock *sk)
977 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
978 __inet_bhash2_update_saddr(sk, NULL, 0, true);
980 EXPORT_SYMBOL_GPL(inet_bhash2_reset_saddr);
982 /* RFC 6056 3.3.4. Algorithm 4: Double-Hash Port Selection Algorithm
983 * Note that we use 32bit integers (vs RFC 'short integers')
984 * because 2^16 is not a multiple of num_ephemeral and this
985 * property might be used by clever attacker.
987 * RFC claims using TABLE_LENGTH=10 buckets gives an improvement, though
988 * attacks were since demonstrated, thus we use 65536 by default instead
989 * to really give more isolation and privacy, at the expense of 256kB
992 #define INET_TABLE_PERTURB_SIZE (1 << CONFIG_INET_TABLE_PERTURB_ORDER)
993 static u32 *table_perturb;
995 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
996 struct sock *sk, u64 port_offset,
997 int (*check_established)(struct inet_timewait_death_row *,
998 struct sock *, __u16, struct inet_timewait_sock **))
1000 struct inet_hashinfo *hinfo = death_row->hashinfo;
1001 struct inet_bind_hashbucket *head, *head2;
1002 struct inet_timewait_sock *tw = NULL;
1003 int port = inet_sk(sk)->inet_num;
1004 struct net *net = sock_net(sk);
1005 struct inet_bind2_bucket *tb2;
1006 struct inet_bind_bucket *tb;
1007 bool tb_created = false;
1008 u32 remaining, offset;
1009 int ret, i, low, high;
1015 ret = check_established(death_row, sk, port, NULL);
1020 l3mdev = inet_sk_bound_l3mdev(sk);
1022 inet_sk_get_local_port_range(sk, &low, &high);
1023 high++; /* [32768, 60999] -> [32768, 61000[ */
1024 remaining = high - low;
1025 if (likely(remaining > 1))
1028 get_random_sleepable_once(table_perturb,
1029 INET_TABLE_PERTURB_SIZE * sizeof(*table_perturb));
1030 index = port_offset & (INET_TABLE_PERTURB_SIZE - 1);
1032 offset = READ_ONCE(table_perturb[index]) + (port_offset >> 32);
1033 offset %= remaining;
1035 /* In first pass we try ports of @low parity.
1036 * inet_csk_get_port() does the opposite choice.
1040 port = low + offset;
1041 for (i = 0; i < remaining; i += 2, port += 2) {
1042 if (unlikely(port >= high))
1044 if (inet_is_local_reserved_port(net, port))
1046 head = &hinfo->bhash[inet_bhashfn(net, port,
1047 hinfo->bhash_size)];
1048 spin_lock_bh(&head->lock);
1050 /* Does not bother with rcv_saddr checks, because
1051 * the established check is already unique enough.
1053 inet_bind_bucket_for_each(tb, &head->chain) {
1054 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
1055 if (tb->fastreuse >= 0 ||
1056 tb->fastreuseport >= 0)
1058 WARN_ON(hlist_empty(&tb->owners));
1059 if (!check_established(death_row, sk,
1066 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
1067 net, head, port, l3mdev);
1069 spin_unlock_bh(&head->lock);
1074 tb->fastreuseport = -1;
1077 spin_unlock_bh(&head->lock);
1082 if ((offset & 1) && remaining > 1)
1083 goto other_parity_scan;
1085 return -EADDRNOTAVAIL;
1088 /* Find the corresponding tb2 bucket since we need to
1089 * add the socket to the bhash2 table as well
1091 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
1092 spin_lock(&head2->lock);
1094 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
1096 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep, net,
1097 head2, port, l3mdev, sk);
1102 /* Here we want to add a little bit of randomness to the next source
1103 * port that will be chosen. We use a max() with a random here so that
1104 * on low contention the randomness is maximal and on high contention
1105 * it may be inexistent.
1107 i = max_t(int, i, get_random_u32_below(8) * 2);
1108 WRITE_ONCE(table_perturb[index], READ_ONCE(table_perturb[index]) + i + 2);
1110 /* Head lock still held and bh's disabled */
1111 inet_bind_hash(sk, tb, tb2, port);
1113 if (sk_unhashed(sk)) {
1114 inet_sk(sk)->inet_sport = htons(port);
1115 inet_ehash_nolisten(sk, (struct sock *)tw, NULL);
1118 inet_twsk_bind_unhash(tw, hinfo);
1120 spin_unlock(&head2->lock);
1121 spin_unlock(&head->lock);
1124 inet_twsk_deschedule_put(tw);
1129 spin_unlock(&head2->lock);
1131 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
1132 spin_unlock_bh(&head->lock);
1137 * Bind a port for a connect operation and hash it.
1139 int inet_hash_connect(struct inet_timewait_death_row *death_row,
1142 u64 port_offset = 0;
1144 if (!inet_sk(sk)->inet_num)
1145 port_offset = inet_sk_port_offset(sk);
1146 return __inet_hash_connect(death_row, sk, port_offset,
1147 __inet_check_established);
1149 EXPORT_SYMBOL_GPL(inet_hash_connect);
1151 static void init_hashinfo_lhash2(struct inet_hashinfo *h)
1155 for (i = 0; i <= h->lhash2_mask; i++) {
1156 spin_lock_init(&h->lhash2[i].lock);
1157 INIT_HLIST_NULLS_HEAD(&h->lhash2[i].nulls_head,
1158 i + LISTENING_NULLS_BASE);
1162 void __init inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
1163 unsigned long numentries, int scale,
1164 unsigned long low_limit,
1165 unsigned long high_limit)
1167 h->lhash2 = alloc_large_system_hash(name,
1176 init_hashinfo_lhash2(h);
1178 /* this one is used for source ports of outgoing connections */
1179 table_perturb = alloc_large_system_hash("Table-perturb",
1180 sizeof(*table_perturb),
1181 INET_TABLE_PERTURB_SIZE,
1183 INET_TABLE_PERTURB_SIZE,
1184 INET_TABLE_PERTURB_SIZE);
1187 int inet_hashinfo2_init_mod(struct inet_hashinfo *h)
1189 h->lhash2 = kmalloc_array(INET_LHTABLE_SIZE, sizeof(*h->lhash2), GFP_KERNEL);
1193 h->lhash2_mask = INET_LHTABLE_SIZE - 1;
1194 /* INET_LHTABLE_SIZE must be a power of 2 */
1195 BUG_ON(INET_LHTABLE_SIZE & h->lhash2_mask);
1197 init_hashinfo_lhash2(h);
1200 EXPORT_SYMBOL_GPL(inet_hashinfo2_init_mod);
1202 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
1204 unsigned int locksz = sizeof(spinlock_t);
1205 unsigned int i, nblocks = 1;
1208 /* allocate 2 cache lines or at least one spinlock per cpu */
1209 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
1210 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
1212 /* no more locks than number of hash buckets */
1213 nblocks = min(nblocks, hashinfo->ehash_mask + 1);
1215 hashinfo->ehash_locks = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
1216 if (!hashinfo->ehash_locks)
1219 for (i = 0; i < nblocks; i++)
1220 spin_lock_init(&hashinfo->ehash_locks[i]);
1222 hashinfo->ehash_locks_mask = nblocks - 1;
1225 EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);
1227 struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
1228 unsigned int ehash_entries)
1230 struct inet_hashinfo *new_hashinfo;
1233 new_hashinfo = kmemdup(hashinfo, sizeof(*hashinfo), GFP_KERNEL);
1237 new_hashinfo->ehash = vmalloc_huge(ehash_entries * sizeof(struct inet_ehash_bucket),
1238 GFP_KERNEL_ACCOUNT);
1239 if (!new_hashinfo->ehash)
1242 new_hashinfo->ehash_mask = ehash_entries - 1;
1244 if (inet_ehash_locks_alloc(new_hashinfo))
1247 for (i = 0; i < ehash_entries; i++)
1248 INIT_HLIST_NULLS_HEAD(&new_hashinfo->ehash[i].chain, i);
1250 new_hashinfo->pernet = true;
1252 return new_hashinfo;
1255 vfree(new_hashinfo->ehash);
1257 kfree(new_hashinfo);
1261 EXPORT_SYMBOL_GPL(inet_pernet_hashinfo_alloc);
1263 void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo)
1265 if (!hashinfo->pernet)
1268 inet_ehash_locks_free(hashinfo);
1269 vfree(hashinfo->ehash);
1272 EXPORT_SYMBOL_GPL(inet_pernet_hashinfo_free);