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 * Support for INET connection oriented protocols.
9 * Authors: See the TCP sources
12 #include <linux/module.h>
13 #include <linux/jhash.h>
15 #include <net/inet_connection_sock.h>
16 #include <net/inet_hashtables.h>
17 #include <net/inet_timewait_sock.h>
19 #include <net/route.h>
20 #include <net/tcp_states.h>
23 #include <net/sock_reuseport.h>
24 #include <net/addrconf.h>
26 #if IS_ENABLED(CONFIG_IPV6)
27 /* match_sk*_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses
28 * if IPv6 only, and any IPv4 addresses
30 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
31 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
32 * and 0.0.0.0 equals to 0.0.0.0 only
34 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
35 const struct in6_addr *sk2_rcv_saddr6,
36 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
37 bool sk1_ipv6only, bool sk2_ipv6only,
38 bool match_sk1_wildcard,
39 bool match_sk2_wildcard)
41 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
42 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
44 /* if both are mapped, treat as IPv4 */
45 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
47 if (sk1_rcv_saddr == sk2_rcv_saddr)
49 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
50 (match_sk2_wildcard && !sk2_rcv_saddr);
55 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
58 if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard &&
59 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
62 if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard &&
63 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
67 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
74 /* match_sk*_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
75 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
76 * 0.0.0.0 only equals to 0.0.0.0
78 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
79 bool sk2_ipv6only, bool match_sk1_wildcard,
80 bool match_sk2_wildcard)
83 if (sk1_rcv_saddr == sk2_rcv_saddr)
85 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
86 (match_sk2_wildcard && !sk2_rcv_saddr);
91 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
94 #if IS_ENABLED(CONFIG_IPV6)
95 if (sk->sk_family == AF_INET6)
96 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
105 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
106 ipv6_only_sock(sk2), match_wildcard,
109 EXPORT_SYMBOL(inet_rcv_saddr_equal);
111 bool inet_rcv_saddr_any(const struct sock *sk)
113 #if IS_ENABLED(CONFIG_IPV6)
114 if (sk->sk_family == AF_INET6)
115 return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
117 return !sk->sk_rcv_saddr;
120 void inet_get_local_port_range(struct net *net, int *low, int *high)
125 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
127 *low = net->ipv4.ip_local_ports.range[0];
128 *high = net->ipv4.ip_local_ports.range[1];
129 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
131 EXPORT_SYMBOL(inet_get_local_port_range);
133 static bool inet_use_bhash2_on_bind(const struct sock *sk)
135 #if IS_ENABLED(CONFIG_IPV6)
136 if (sk->sk_family == AF_INET6) {
137 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
139 return addr_type != IPV6_ADDR_ANY &&
140 addr_type != IPV6_ADDR_MAPPED;
143 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
146 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
147 kuid_t sk_uid, bool relax,
148 bool reuseport_cb_ok, bool reuseport_ok)
155 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
157 if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
158 sk->sk_bound_dev_if == bound_dev_if2) {
159 if (sk->sk_reuse && sk2->sk_reuse &&
160 sk2->sk_state != TCP_LISTEN) {
161 if (!relax || (!reuseport_ok && sk->sk_reuseport &&
162 sk2->sk_reuseport && reuseport_cb_ok &&
163 (sk2->sk_state == TCP_TIME_WAIT ||
164 uid_eq(sk_uid, sock_i_uid(sk2)))))
166 } else if (!reuseport_ok || !sk->sk_reuseport ||
167 !sk2->sk_reuseport || !reuseport_cb_ok ||
168 (sk2->sk_state != TCP_TIME_WAIT &&
169 !uid_eq(sk_uid, sock_i_uid(sk2)))) {
176 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
177 kuid_t sk_uid, bool relax,
178 bool reuseport_cb_ok, bool reuseport_ok)
180 if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
183 return inet_bind_conflict(sk, sk2, sk_uid, relax,
184 reuseport_cb_ok, reuseport_ok);
187 static bool inet_bhash2_conflict(const struct sock *sk,
188 const struct inet_bind2_bucket *tb2,
190 bool relax, bool reuseport_cb_ok,
193 struct inet_timewait_sock *tw2;
196 sk_for_each_bound_bhash2(sk2, &tb2->owners) {
197 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
198 reuseport_cb_ok, reuseport_ok))
202 twsk_for_each_bound_bhash2(tw2, &tb2->deathrow) {
203 sk2 = (struct sock *)tw2;
205 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
206 reuseport_cb_ok, reuseport_ok))
213 /* This should be called only when the tb and tb2 hashbuckets' locks are held */
214 static int inet_csk_bind_conflict(const struct sock *sk,
215 const struct inet_bind_bucket *tb,
216 const struct inet_bind2_bucket *tb2, /* may be null */
217 bool relax, bool reuseport_ok)
219 bool reuseport_cb_ok;
220 struct sock_reuseport *reuseport_cb;
221 kuid_t uid = sock_i_uid((struct sock *)sk);
224 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
225 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
226 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
230 * Unlike other sk lookup places we do not check
231 * for sk_net here, since _all_ the socks listed
232 * in tb->owners and tb2->owners list belong
233 * to the same net - the one this bucket belongs to.
236 if (!inet_use_bhash2_on_bind(sk)) {
239 sk_for_each_bound(sk2, &tb->owners)
240 if (inet_bind_conflict(sk, sk2, uid, relax,
241 reuseport_cb_ok, reuseport_ok) &&
242 inet_rcv_saddr_equal(sk, sk2, true))
248 /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
249 * ipv4) should have been checked already. We need to do these two
250 * checks separately because their spinlocks have to be acquired/released
251 * independently of each other, to prevent possible deadlocks
253 return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
257 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
258 * INADDR_ANY (if ipv4) socket.
260 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
261 * against concurrent binds on the port for addr any
263 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
264 bool relax, bool reuseport_ok)
266 kuid_t uid = sock_i_uid((struct sock *)sk);
267 const struct net *net = sock_net(sk);
268 struct sock_reuseport *reuseport_cb;
269 struct inet_bind_hashbucket *head2;
270 struct inet_bind2_bucket *tb2;
271 bool reuseport_cb_ok;
274 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
275 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
276 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
279 head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
281 spin_lock(&head2->lock);
283 inet_bind_bucket_for_each(tb2, &head2->chain)
284 if (inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
287 if (tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
289 spin_unlock(&head2->lock);
293 spin_unlock(&head2->lock);
298 * Find an open port number for the socket. Returns with the
299 * inet_bind_hashbucket locks held if successful.
301 static struct inet_bind_hashbucket *
302 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
303 struct inet_bind2_bucket **tb2_ret,
304 struct inet_bind_hashbucket **head2_ret, int *port_ret)
306 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
307 int i, low, high, attempt_half, port, l3mdev;
308 struct inet_bind_hashbucket *head, *head2;
309 struct net *net = sock_net(sk);
310 struct inet_bind2_bucket *tb2;
311 struct inet_bind_bucket *tb;
312 u32 remaining, offset;
315 l3mdev = inet_sk_bound_l3mdev(sk);
317 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
319 inet_get_local_port_range(net, &low, &high);
320 high++; /* [32768, 60999] -> [32768, 61000[ */
324 int half = low + (((high - low) >> 2) << 1);
326 if (attempt_half == 1)
331 remaining = high - low;
332 if (likely(remaining > 1))
335 offset = get_random_u32_below(remaining);
336 /* __inet_hash_connect() favors ports having @low parity
337 * We do the opposite to not pollute connect() users.
343 for (i = 0; i < remaining; i += 2, port += 2) {
344 if (unlikely(port >= high))
346 if (inet_is_local_reserved_port(net, port))
348 head = &hinfo->bhash[inet_bhashfn(net, port,
350 spin_lock_bh(&head->lock);
351 if (inet_use_bhash2_on_bind(sk)) {
352 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
356 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
357 spin_lock(&head2->lock);
358 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
359 inet_bind_bucket_for_each(tb, &head->chain)
360 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
361 if (!inet_csk_bind_conflict(sk, tb, tb2,
364 spin_unlock(&head2->lock);
370 spin_unlock_bh(&head->lock);
376 goto other_parity_scan;
378 if (attempt_half == 1) {
379 /* OK we now try the upper half of the range */
381 goto other_half_scan;
384 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
385 /* We still have a chance to connect to different destinations */
387 goto ports_exhausted;
398 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
401 kuid_t uid = sock_i_uid(sk);
403 if (tb->fastreuseport <= 0)
405 if (!sk->sk_reuseport)
407 if (rcu_access_pointer(sk->sk_reuseport_cb))
409 if (!uid_eq(tb->fastuid, uid))
411 /* We only need to check the rcv_saddr if this tb was once marked
412 * without fastreuseport and then was reset, as we can only know that
413 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
416 if (tb->fastreuseport == FASTREUSEPORT_ANY)
418 #if IS_ENABLED(CONFIG_IPV6)
419 if (tb->fast_sk_family == AF_INET6)
420 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
425 ipv6_only_sock(sk), true, false);
427 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
428 ipv6_only_sock(sk), true, false);
431 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
434 kuid_t uid = sock_i_uid(sk);
435 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
437 if (hlist_empty(&tb->owners)) {
438 tb->fastreuse = reuse;
439 if (sk->sk_reuseport) {
440 tb->fastreuseport = FASTREUSEPORT_ANY;
442 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
443 tb->fast_ipv6_only = ipv6_only_sock(sk);
444 tb->fast_sk_family = sk->sk_family;
445 #if IS_ENABLED(CONFIG_IPV6)
446 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
449 tb->fastreuseport = 0;
454 if (sk->sk_reuseport) {
455 /* We didn't match or we don't have fastreuseport set on
456 * the tb, but we have sk_reuseport set on this socket
457 * and we know that there are no bind conflicts with
458 * this socket in this tb, so reset our tb's reuseport
459 * settings so that any subsequent sockets that match
460 * our current socket will be put on the fast path.
462 * If we reset we need to set FASTREUSEPORT_STRICT so we
463 * do extra checking for all subsequent sk_reuseport
466 if (!sk_reuseport_match(tb, sk)) {
467 tb->fastreuseport = FASTREUSEPORT_STRICT;
469 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
470 tb->fast_ipv6_only = ipv6_only_sock(sk);
471 tb->fast_sk_family = sk->sk_family;
472 #if IS_ENABLED(CONFIG_IPV6)
473 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
477 tb->fastreuseport = 0;
482 /* Obtain a reference to a local port for the given sock,
483 * if snum is zero it means select any available local port.
484 * We try to allocate an odd port (and leave even ports for connect())
486 int inet_csk_get_port(struct sock *sk, unsigned short snum)
488 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
489 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
490 bool found_port = false, check_bind_conflict = true;
491 bool bhash_created = false, bhash2_created = false;
492 int ret = -EADDRINUSE, port = snum, l3mdev;
493 struct inet_bind_hashbucket *head, *head2;
494 struct inet_bind2_bucket *tb2 = NULL;
495 struct inet_bind_bucket *tb = NULL;
496 bool head2_lock_acquired = false;
497 struct net *net = sock_net(sk);
499 l3mdev = inet_sk_bound_l3mdev(sk);
502 head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
506 head2_lock_acquired = true;
512 head = &hinfo->bhash[inet_bhashfn(net, port,
514 spin_lock_bh(&head->lock);
515 inet_bind_bucket_for_each(tb, &head->chain)
516 if (inet_bind_bucket_match(tb, net, port, l3mdev))
521 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
525 bhash_created = true;
529 if (!hlist_empty(&tb->owners)) {
530 if (sk->sk_reuse == SK_FORCE_REUSE ||
531 (tb->fastreuse > 0 && reuse) ||
532 sk_reuseport_match(tb, sk))
533 check_bind_conflict = false;
536 if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
537 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
541 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
542 spin_lock(&head2->lock);
543 head2_lock_acquired = true;
544 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
548 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
549 net, head2, port, l3mdev, sk);
552 bhash2_created = true;
555 if (!found_port && check_bind_conflict) {
556 if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
561 inet_csk_update_fastreuse(tb, sk);
563 if (!inet_csk(sk)->icsk_bind_hash)
564 inet_bind_hash(sk, tb, tb2, port);
565 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
566 WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
572 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
574 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep,
577 if (head2_lock_acquired)
578 spin_unlock(&head2->lock);
579 spin_unlock_bh(&head->lock);
582 EXPORT_SYMBOL_GPL(inet_csk_get_port);
585 * Wait for an incoming connection, avoid race conditions. This must be called
586 * with the socket locked.
588 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
590 struct inet_connection_sock *icsk = inet_csk(sk);
595 * True wake-one mechanism for incoming connections: only
596 * one process gets woken up, not the 'whole herd'.
597 * Since we do not 'race & poll' for established sockets
598 * anymore, the common case will execute the loop only once.
600 * Subtle issue: "add_wait_queue_exclusive()" will be added
601 * after any current non-exclusive waiters, and we know that
602 * it will always _stay_ after any new non-exclusive waiters
603 * because all non-exclusive waiters are added at the
604 * beginning of the wait-queue. As such, it's ok to "drop"
605 * our exclusiveness temporarily when we get woken up without
606 * having to remove and re-insert us on the wait queue.
609 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
612 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
613 timeo = schedule_timeout(timeo);
614 sched_annotate_sleep();
617 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
620 if (sk->sk_state != TCP_LISTEN)
622 err = sock_intr_errno(timeo);
623 if (signal_pending(current))
629 finish_wait(sk_sleep(sk), &wait);
634 * This will accept the next outstanding connection.
636 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
638 struct inet_connection_sock *icsk = inet_csk(sk);
639 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
640 struct request_sock *req;
646 /* We need to make sure that this socket is listening,
647 * and that it has something pending.
650 if (sk->sk_state != TCP_LISTEN)
653 /* Find already established connection */
654 if (reqsk_queue_empty(queue)) {
655 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
657 /* If this is a non blocking socket don't sleep */
662 error = inet_csk_wait_for_connect(sk, timeo);
666 req = reqsk_queue_remove(queue, sk);
669 if (sk->sk_protocol == IPPROTO_TCP &&
670 tcp_rsk(req)->tfo_listener) {
671 spin_lock_bh(&queue->fastopenq.lock);
672 if (tcp_rsk(req)->tfo_listener) {
673 /* We are still waiting for the final ACK from 3WHS
674 * so can't free req now. Instead, we set req->sk to
675 * NULL to signify that the child socket is taken
676 * so reqsk_fastopen_remove() will free the req
677 * when 3WHS finishes (or is aborted).
682 spin_unlock_bh(&queue->fastopenq.lock);
687 if (newsk && mem_cgroup_sockets_enabled) {
690 /* atomically get the memory usage, set and charge the
695 /* The socket has not been accepted yet, no need to look at
696 * newsk->sk_wmem_queued.
698 amt = sk_mem_pages(newsk->sk_forward_alloc +
699 atomic_read(&newsk->sk_rmem_alloc));
700 mem_cgroup_sk_alloc(newsk);
701 if (newsk->sk_memcg && amt)
702 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
703 GFP_KERNEL | __GFP_NOFAIL);
716 EXPORT_SYMBOL(inet_csk_accept);
719 * Using different timers for retransmit, delayed acks and probes
720 * We may wish use just one timer maintaining a list of expire jiffies
723 void inet_csk_init_xmit_timers(struct sock *sk,
724 void (*retransmit_handler)(struct timer_list *t),
725 void (*delack_handler)(struct timer_list *t),
726 void (*keepalive_handler)(struct timer_list *t))
728 struct inet_connection_sock *icsk = inet_csk(sk);
730 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
731 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
732 timer_setup(&sk->sk_timer, keepalive_handler, 0);
733 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
735 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
737 void inet_csk_clear_xmit_timers(struct sock *sk)
739 struct inet_connection_sock *icsk = inet_csk(sk);
741 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
743 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
744 sk_stop_timer(sk, &icsk->icsk_delack_timer);
745 sk_stop_timer(sk, &sk->sk_timer);
747 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
749 void inet_csk_delete_keepalive_timer(struct sock *sk)
751 sk_stop_timer(sk, &sk->sk_timer);
753 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
755 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
757 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
759 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
761 struct dst_entry *inet_csk_route_req(const struct sock *sk,
763 const struct request_sock *req)
765 const struct inet_request_sock *ireq = inet_rsk(req);
766 struct net *net = read_pnet(&ireq->ireq_net);
767 struct ip_options_rcu *opt;
771 opt = rcu_dereference(ireq->ireq_opt);
773 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
774 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
775 sk->sk_protocol, inet_sk_flowi_flags(sk),
776 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
777 ireq->ir_loc_addr, ireq->ir_rmt_port,
778 htons(ireq->ir_num), sk->sk_uid);
779 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
780 rt = ip_route_output_flow(net, fl4, sk);
783 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
792 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
795 EXPORT_SYMBOL_GPL(inet_csk_route_req);
797 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
799 const struct request_sock *req)
801 const struct inet_request_sock *ireq = inet_rsk(req);
802 struct net *net = read_pnet(&ireq->ireq_net);
803 struct inet_sock *newinet = inet_sk(newsk);
804 struct ip_options_rcu *opt;
808 opt = rcu_dereference(ireq->ireq_opt);
809 fl4 = &newinet->cork.fl.u.ip4;
811 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
812 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
813 sk->sk_protocol, inet_sk_flowi_flags(sk),
814 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
815 ireq->ir_loc_addr, ireq->ir_rmt_port,
816 htons(ireq->ir_num), sk->sk_uid);
817 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
818 rt = ip_route_output_flow(net, fl4, sk);
821 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
828 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
831 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
833 /* Decide when to expire the request and when to resend SYN-ACK */
834 static void syn_ack_recalc(struct request_sock *req,
835 const int max_syn_ack_retries,
836 const u8 rskq_defer_accept,
837 int *expire, int *resend)
839 if (!rskq_defer_accept) {
840 *expire = req->num_timeout >= max_syn_ack_retries;
844 *expire = req->num_timeout >= max_syn_ack_retries &&
845 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
846 /* Do not resend while waiting for data after ACK,
847 * start to resend on end of deferring period to give
848 * last chance for data or ACK to create established socket.
850 *resend = !inet_rsk(req)->acked ||
851 req->num_timeout >= rskq_defer_accept - 1;
854 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
856 int err = req->rsk_ops->rtx_syn_ack(parent, req);
862 EXPORT_SYMBOL(inet_rtx_syn_ack);
864 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
867 struct sock *req_sk, *nreq_sk;
868 struct request_sock *nreq;
870 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
872 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
874 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
879 req_sk = req_to_sk(req);
880 nreq_sk = req_to_sk(nreq);
882 memcpy(nreq_sk, req_sk,
883 offsetof(struct sock, sk_dontcopy_begin));
884 memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
885 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
887 sk_node_init(&nreq_sk->sk_node);
888 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
889 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
890 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
892 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
894 nreq->rsk_listener = sk;
896 /* We need not acquire fastopenq->lock
897 * because the child socket is locked in inet_csk_listen_stop().
899 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
900 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
905 static void reqsk_queue_migrated(struct request_sock_queue *queue,
906 const struct request_sock *req)
908 if (req->num_timeout == 0)
909 atomic_inc(&queue->young);
910 atomic_inc(&queue->qlen);
913 static void reqsk_migrate_reset(struct request_sock *req)
915 req->saved_syn = NULL;
916 #if IS_ENABLED(CONFIG_IPV6)
917 inet_rsk(req)->ipv6_opt = NULL;
918 inet_rsk(req)->pktopts = NULL;
920 inet_rsk(req)->ireq_opt = NULL;
924 /* return true if req was found in the ehash table */
925 static bool reqsk_queue_unlink(struct request_sock *req)
927 struct sock *sk = req_to_sk(req);
931 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
932 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
935 found = __sk_nulls_del_node_init_rcu(sk);
938 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
943 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
945 bool unlinked = reqsk_queue_unlink(req);
948 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
953 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
955 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
957 inet_csk_reqsk_queue_drop(sk, req);
960 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
962 static void reqsk_timer_handler(struct timer_list *t)
964 struct request_sock *req = from_timer(req, t, rsk_timer);
965 struct request_sock *nreq = NULL, *oreq = req;
966 struct sock *sk_listener = req->rsk_listener;
967 struct inet_connection_sock *icsk;
968 struct request_sock_queue *queue;
970 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
972 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
975 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
979 nreq = inet_reqsk_clone(req, nsk);
983 /* The new timer for the cloned req can decrease the 2
984 * by calling inet_csk_reqsk_queue_drop_and_put(), so
985 * hold another count to prevent use-after-free and
986 * call reqsk_put() just before return.
988 refcount_set(&nreq->rsk_refcnt, 2 + 1);
989 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
990 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
996 icsk = inet_csk(sk_listener);
997 net = sock_net(sk_listener);
998 max_syn_ack_retries = icsk->icsk_syn_retries ? :
999 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1000 /* Normally all the openreqs are young and become mature
1001 * (i.e. converted to established socket) for first timeout.
1002 * If synack was not acknowledged for 1 second, it means
1003 * one of the following things: synack was lost, ack was lost,
1004 * rtt is high or nobody planned to ack (i.e. synflood).
1005 * When server is a bit loaded, queue is populated with old
1006 * open requests, reducing effective size of queue.
1007 * When server is well loaded, queue size reduces to zero
1008 * after several minutes of work. It is not synflood,
1009 * it is normal operation. The solution is pruning
1010 * too old entries overriding normal timeout, when
1011 * situation becomes dangerous.
1013 * Essentially, we reserve half of room for young
1014 * embrions; and abort old ones without pity, if old
1015 * ones are about to clog our table.
1017 queue = &icsk->icsk_accept_queue;
1018 qlen = reqsk_queue_len(queue);
1019 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1020 int young = reqsk_queue_len_young(queue) << 1;
1022 while (max_syn_ack_retries > 2) {
1025 max_syn_ack_retries--;
1029 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1031 req->rsk_ops->syn_ack_timeout(req);
1034 !inet_rtx_syn_ack(sk_listener, req) ||
1035 inet_rsk(req)->acked)) {
1036 if (req->num_timeout++ == 0)
1037 atomic_dec(&queue->young);
1038 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1043 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1045 inet_csk_reqsk_queue_drop(sk_listener, nreq);
1049 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1050 reqsk_migrate_reset(oreq);
1051 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1058 /* Even if we can clone the req, we may need not retransmit any more
1059 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1060 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1063 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1065 reqsk_migrate_reset(nreq);
1066 reqsk_queue_removed(queue, nreq);
1071 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1074 static void reqsk_queue_hash_req(struct request_sock *req,
1075 unsigned long timeout)
1077 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1078 mod_timer(&req->rsk_timer, jiffies + timeout);
1080 inet_ehash_insert(req_to_sk(req), NULL, NULL);
1081 /* before letting lookups find us, make sure all req fields
1082 * are committed to memory and refcnt initialized.
1085 refcount_set(&req->rsk_refcnt, 2 + 1);
1088 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1089 unsigned long timeout)
1091 reqsk_queue_hash_req(req, timeout);
1092 inet_csk_reqsk_queue_added(sk);
1094 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1096 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1097 const gfp_t priority)
1099 struct inet_connection_sock *icsk = inet_csk(newsk);
1101 if (!icsk->icsk_ulp_ops)
1104 if (icsk->icsk_ulp_ops->clone)
1105 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1109 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1110 * @sk: the socket to clone
1111 * @req: request_sock
1112 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1114 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1116 struct sock *inet_csk_clone_lock(const struct sock *sk,
1117 const struct request_sock *req,
1118 const gfp_t priority)
1120 struct sock *newsk = sk_clone_lock(sk, priority);
1123 struct inet_connection_sock *newicsk = inet_csk(newsk);
1125 inet_sk_set_state(newsk, TCP_SYN_RECV);
1126 newicsk->icsk_bind_hash = NULL;
1127 newicsk->icsk_bind2_hash = NULL;
1129 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1130 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1131 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1133 /* listeners have SOCK_RCU_FREE, not the children */
1134 sock_reset_flag(newsk, SOCK_RCU_FREE);
1136 inet_sk(newsk)->mc_list = NULL;
1138 newsk->sk_mark = inet_rsk(req)->ir_mark;
1139 atomic64_set(&newsk->sk_cookie,
1140 atomic64_read(&inet_rsk(req)->ir_cookie));
1142 newicsk->icsk_retransmits = 0;
1143 newicsk->icsk_backoff = 0;
1144 newicsk->icsk_probes_out = 0;
1145 newicsk->icsk_probes_tstamp = 0;
1147 /* Deinitialize accept_queue to trap illegal accesses. */
1148 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1150 inet_clone_ulp(req, newsk, priority);
1152 security_inet_csk_clone(newsk, req);
1156 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1159 * At this point, there should be no process reference to this
1160 * socket, and thus no user references at all. Therefore we
1161 * can assume the socket waitqueue is inactive and nobody will
1162 * try to jump onto it.
1164 void inet_csk_destroy_sock(struct sock *sk)
1166 WARN_ON(sk->sk_state != TCP_CLOSE);
1167 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1169 /* It cannot be in hash table! */
1170 WARN_ON(!sk_unhashed(sk));
1172 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1173 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1175 sk->sk_prot->destroy(sk);
1177 sk_stream_kill_queues(sk);
1179 xfrm_sk_free_policy(sk);
1181 sk_refcnt_debug_release(sk);
1183 this_cpu_dec(*sk->sk_prot->orphan_count);
1187 EXPORT_SYMBOL(inet_csk_destroy_sock);
1189 /* This function allows to force a closure of a socket after the call to
1190 * tcp/dccp_create_openreq_child().
1192 void inet_csk_prepare_forced_close(struct sock *sk)
1193 __releases(&sk->sk_lock.slock)
1195 /* sk_clone_lock locked the socket and set refcnt to 2 */
1198 inet_csk_prepare_for_destroy_sock(sk);
1199 inet_sk(sk)->inet_num = 0;
1201 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1203 int inet_csk_listen_start(struct sock *sk)
1205 struct inet_connection_sock *icsk = inet_csk(sk);
1206 struct inet_sock *inet = inet_sk(sk);
1209 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1211 sk->sk_ack_backlog = 0;
1212 inet_csk_delack_init(sk);
1214 if (sk->sk_txrehash == SOCK_TXREHASH_DEFAULT)
1215 sk->sk_txrehash = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1217 /* There is race window here: we announce ourselves listening,
1218 * but this transition is still not validated by get_port().
1219 * It is OK, because this socket enters to hash table only
1220 * after validation is complete.
1222 inet_sk_state_store(sk, TCP_LISTEN);
1223 err = sk->sk_prot->get_port(sk, inet->inet_num);
1225 inet->inet_sport = htons(inet->inet_num);
1228 err = sk->sk_prot->hash(sk);
1234 inet_sk_set_state(sk, TCP_CLOSE);
1237 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1239 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1242 sk->sk_prot->disconnect(child, O_NONBLOCK);
1246 this_cpu_inc(*sk->sk_prot->orphan_count);
1248 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1249 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1250 BUG_ON(sk != req->rsk_listener);
1252 /* Paranoid, to prevent race condition if
1253 * an inbound pkt destined for child is
1254 * blocked by sock lock in tcp_v4_rcv().
1255 * Also to satisfy an assertion in
1256 * tcp_v4_destroy_sock().
1258 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1260 inet_csk_destroy_sock(child);
1263 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1264 struct request_sock *req,
1267 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1269 spin_lock(&queue->rskq_lock);
1270 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1271 inet_child_forget(sk, req, child);
1275 req->dl_next = NULL;
1276 if (queue->rskq_accept_head == NULL)
1277 WRITE_ONCE(queue->rskq_accept_head, req);
1279 queue->rskq_accept_tail->dl_next = req;
1280 queue->rskq_accept_tail = req;
1281 sk_acceptq_added(sk);
1283 spin_unlock(&queue->rskq_lock);
1286 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1288 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1289 struct request_sock *req, bool own_req)
1292 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1293 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1295 if (sk != req->rsk_listener) {
1296 /* another listening sk has been selected,
1297 * migrate the req to it.
1299 struct request_sock *nreq;
1301 /* hold a refcnt for the nreq->rsk_listener
1302 * which is assigned in inet_reqsk_clone()
1305 nreq = inet_reqsk_clone(req, sk);
1307 inet_child_forget(sk, req, child);
1311 refcount_set(&nreq->rsk_refcnt, 1);
1312 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1313 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1314 reqsk_migrate_reset(req);
1319 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1320 reqsk_migrate_reset(nreq);
1322 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1326 /* Too bad, another child took ownership of the request, undo. */
1328 bh_unlock_sock(child);
1332 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1335 * This routine closes sockets which have been at least partially
1336 * opened, but not yet accepted.
1338 void inet_csk_listen_stop(struct sock *sk)
1340 struct inet_connection_sock *icsk = inet_csk(sk);
1341 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1342 struct request_sock *next, *req;
1344 /* Following specs, it would be better either to send FIN
1345 * (and enter FIN-WAIT-1, it is normal close)
1346 * or to send active reset (abort).
1347 * Certainly, it is pretty dangerous while synflood, but it is
1348 * bad justification for our negligence 8)
1349 * To be honest, we are not able to make either
1350 * of the variants now. --ANK
1352 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1353 struct sock *child = req->sk, *nsk;
1354 struct request_sock *nreq;
1357 bh_lock_sock(child);
1358 WARN_ON(sock_owned_by_user(child));
1361 nsk = reuseport_migrate_sock(sk, child, NULL);
1363 nreq = inet_reqsk_clone(req, nsk);
1365 refcount_set(&nreq->rsk_refcnt, 1);
1367 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1368 __NET_INC_STATS(sock_net(nsk),
1369 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1370 reqsk_migrate_reset(req);
1372 __NET_INC_STATS(sock_net(nsk),
1373 LINUX_MIB_TCPMIGRATEREQFAILURE);
1374 reqsk_migrate_reset(nreq);
1378 /* inet_csk_reqsk_queue_add() has already
1379 * called inet_child_forget() on failure case.
1381 goto skip_child_forget;
1385 inet_child_forget(sk, req, child);
1388 bh_unlock_sock(child);
1394 if (queue->fastopenq.rskq_rst_head) {
1395 /* Free all the reqs queued in rskq_rst_head. */
1396 spin_lock_bh(&queue->fastopenq.lock);
1397 req = queue->fastopenq.rskq_rst_head;
1398 queue->fastopenq.rskq_rst_head = NULL;
1399 spin_unlock_bh(&queue->fastopenq.lock);
1400 while (req != NULL) {
1401 next = req->dl_next;
1406 WARN_ON_ONCE(sk->sk_ack_backlog);
1408 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1410 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1412 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1413 const struct inet_sock *inet = inet_sk(sk);
1415 sin->sin_family = AF_INET;
1416 sin->sin_addr.s_addr = inet->inet_daddr;
1417 sin->sin_port = inet->inet_dport;
1419 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1421 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1423 const struct inet_sock *inet = inet_sk(sk);
1424 const struct ip_options_rcu *inet_opt;
1425 __be32 daddr = inet->inet_daddr;
1430 inet_opt = rcu_dereference(inet->inet_opt);
1431 if (inet_opt && inet_opt->opt.srr)
1432 daddr = inet_opt->opt.faddr;
1434 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1435 inet->inet_saddr, inet->inet_dport,
1436 inet->inet_sport, sk->sk_protocol,
1437 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1441 sk_setup_caps(sk, &rt->dst);
1447 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1449 struct dst_entry *dst = __sk_dst_check(sk, 0);
1450 struct inet_sock *inet = inet_sk(sk);
1453 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1457 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1459 dst = __sk_dst_check(sk, 0);
1461 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1465 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);