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;
121 * inet_sk_get_local_port_range - fetch ephemeral ports range
123 * @low: pointer to low port
124 * @high: pointer to high port
126 * Fetch netns port range (/proc/sys/net/ipv4/ip_local_port_range)
127 * Range can be overridden if socket got IP_LOCAL_PORT_RANGE option.
128 * Returns true if IP_LOCAL_PORT_RANGE was set on this socket.
130 bool inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
132 int lo, hi, sk_lo, sk_hi;
133 bool local_range = false;
136 inet_get_local_port_range(sock_net(sk), &lo, &hi);
138 sk_range = READ_ONCE(inet_sk(sk)->local_port_range);
139 if (unlikely(sk_range)) {
140 sk_lo = sk_range & 0xffff;
141 sk_hi = sk_range >> 16;
143 if (lo <= sk_lo && sk_lo <= hi)
145 if (lo <= sk_hi && sk_hi <= hi)
154 EXPORT_SYMBOL(inet_sk_get_local_port_range);
156 static bool inet_use_bhash2_on_bind(const struct sock *sk)
158 #if IS_ENABLED(CONFIG_IPV6)
159 if (sk->sk_family == AF_INET6) {
160 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
162 if (addr_type == IPV6_ADDR_ANY)
165 if (addr_type != IPV6_ADDR_MAPPED)
169 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
172 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
173 kuid_t sk_uid, bool relax,
174 bool reuseport_cb_ok, bool reuseport_ok)
181 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
183 if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
184 sk->sk_bound_dev_if == bound_dev_if2) {
185 if (sk->sk_reuse && sk2->sk_reuse &&
186 sk2->sk_state != TCP_LISTEN) {
187 if (!relax || (!reuseport_ok && sk->sk_reuseport &&
188 sk2->sk_reuseport && reuseport_cb_ok &&
189 (sk2->sk_state == TCP_TIME_WAIT ||
190 uid_eq(sk_uid, sock_i_uid(sk2)))))
192 } else if (!reuseport_ok || !sk->sk_reuseport ||
193 !sk2->sk_reuseport || !reuseport_cb_ok ||
194 (sk2->sk_state != TCP_TIME_WAIT &&
195 !uid_eq(sk_uid, sock_i_uid(sk2)))) {
202 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
203 kuid_t sk_uid, bool relax,
204 bool reuseport_cb_ok, bool reuseport_ok)
206 if (ipv6_only_sock(sk2)) {
207 if (sk->sk_family == AF_INET)
210 #if IS_ENABLED(CONFIG_IPV6)
211 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
216 return inet_bind_conflict(sk, sk2, sk_uid, relax,
217 reuseport_cb_ok, reuseport_ok);
220 static bool inet_bhash2_conflict(const struct sock *sk,
221 const struct inet_bind2_bucket *tb2,
223 bool relax, bool reuseport_cb_ok,
228 sk_for_each_bound(sk2, &tb2->owners) {
229 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
230 reuseport_cb_ok, reuseport_ok))
237 #define sk_for_each_bound_bhash(__sk, __tb2, __tb) \
238 hlist_for_each_entry(__tb2, &(__tb)->bhash2, bhash_node) \
239 sk_for_each_bound(sk2, &(__tb2)->owners)
241 /* This should be called only when the tb and tb2 hashbuckets' locks are held */
242 static int inet_csk_bind_conflict(const struct sock *sk,
243 const struct inet_bind_bucket *tb,
244 const struct inet_bind2_bucket *tb2, /* may be null */
245 bool relax, bool reuseport_ok)
247 kuid_t uid = sock_i_uid((struct sock *)sk);
248 struct sock_reuseport *reuseport_cb;
249 bool reuseport_cb_ok;
253 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
254 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
255 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
258 /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
259 * ipv4) should have been checked already. We need to do these two
260 * checks separately because their spinlocks have to be acquired/released
261 * independently of each other, to prevent possible deadlocks
263 if (inet_use_bhash2_on_bind(sk))
264 return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax,
265 reuseport_cb_ok, reuseport_ok);
267 /* Unlike other sk lookup places we do not check
268 * for sk_net here, since _all_ the socks listed
269 * in tb->owners and tb2->owners list belong
270 * to the same net - the one this bucket belongs to.
272 sk_for_each_bound_bhash(sk2, tb2, tb) {
273 if (!inet_bind_conflict(sk, sk2, uid, relax, reuseport_cb_ok, reuseport_ok))
276 if (inet_rcv_saddr_equal(sk, sk2, true))
283 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
284 * INADDR_ANY (if ipv4) socket.
286 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
287 * against concurrent binds on the port for addr any
289 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
290 bool relax, bool reuseport_ok)
292 kuid_t uid = sock_i_uid((struct sock *)sk);
293 const struct net *net = sock_net(sk);
294 struct sock_reuseport *reuseport_cb;
295 struct inet_bind_hashbucket *head2;
296 struct inet_bind2_bucket *tb2;
297 bool conflict = false;
298 bool reuseport_cb_ok;
301 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
302 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
303 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
306 head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
308 spin_lock(&head2->lock);
310 inet_bind_bucket_for_each(tb2, &head2->chain) {
311 if (!inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
314 if (!inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok, reuseport_ok))
321 spin_unlock(&head2->lock);
327 * Find an open port number for the socket. Returns with the
328 * inet_bind_hashbucket locks held if successful.
330 static struct inet_bind_hashbucket *
331 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
332 struct inet_bind2_bucket **tb2_ret,
333 struct inet_bind_hashbucket **head2_ret, int *port_ret)
335 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
336 int i, low, high, attempt_half, port, l3mdev;
337 struct inet_bind_hashbucket *head, *head2;
338 struct net *net = sock_net(sk);
339 struct inet_bind2_bucket *tb2;
340 struct inet_bind_bucket *tb;
341 u32 remaining, offset;
344 l3mdev = inet_sk_bound_l3mdev(sk);
346 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
348 inet_sk_get_local_port_range(sk, &low, &high);
349 high++; /* [32768, 60999] -> [32768, 61000[ */
353 int half = low + (((high - low) >> 2) << 1);
355 if (attempt_half == 1)
360 remaining = high - low;
361 if (likely(remaining > 1))
364 offset = get_random_u32_below(remaining);
365 /* __inet_hash_connect() favors ports having @low parity
366 * We do the opposite to not pollute connect() users.
372 for (i = 0; i < remaining; i += 2, port += 2) {
373 if (unlikely(port >= high))
375 if (inet_is_local_reserved_port(net, port))
377 head = &hinfo->bhash[inet_bhashfn(net, port,
379 spin_lock_bh(&head->lock);
380 if (inet_use_bhash2_on_bind(sk)) {
381 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
385 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
386 spin_lock(&head2->lock);
387 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
388 inet_bind_bucket_for_each(tb, &head->chain)
389 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
390 if (!inet_csk_bind_conflict(sk, tb, tb2,
393 spin_unlock(&head2->lock);
399 spin_unlock_bh(&head->lock);
405 goto other_parity_scan;
407 if (attempt_half == 1) {
408 /* OK we now try the upper half of the range */
410 goto other_half_scan;
413 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
414 /* We still have a chance to connect to different destinations */
416 goto ports_exhausted;
427 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
430 kuid_t uid = sock_i_uid(sk);
432 if (tb->fastreuseport <= 0)
434 if (!sk->sk_reuseport)
436 if (rcu_access_pointer(sk->sk_reuseport_cb))
438 if (!uid_eq(tb->fastuid, uid))
440 /* We only need to check the rcv_saddr if this tb was once marked
441 * without fastreuseport and then was reset, as we can only know that
442 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
445 if (tb->fastreuseport == FASTREUSEPORT_ANY)
447 #if IS_ENABLED(CONFIG_IPV6)
448 if (tb->fast_sk_family == AF_INET6)
449 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
454 ipv6_only_sock(sk), true, false);
456 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
457 ipv6_only_sock(sk), true, false);
460 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
463 kuid_t uid = sock_i_uid(sk);
464 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
466 if (hlist_empty(&tb->bhash2)) {
467 tb->fastreuse = reuse;
468 if (sk->sk_reuseport) {
469 tb->fastreuseport = FASTREUSEPORT_ANY;
471 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
472 tb->fast_ipv6_only = ipv6_only_sock(sk);
473 tb->fast_sk_family = sk->sk_family;
474 #if IS_ENABLED(CONFIG_IPV6)
475 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
478 tb->fastreuseport = 0;
483 if (sk->sk_reuseport) {
484 /* We didn't match or we don't have fastreuseport set on
485 * the tb, but we have sk_reuseport set on this socket
486 * and we know that there are no bind conflicts with
487 * this socket in this tb, so reset our tb's reuseport
488 * settings so that any subsequent sockets that match
489 * our current socket will be put on the fast path.
491 * If we reset we need to set FASTREUSEPORT_STRICT so we
492 * do extra checking for all subsequent sk_reuseport
495 if (!sk_reuseport_match(tb, sk)) {
496 tb->fastreuseport = FASTREUSEPORT_STRICT;
498 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
499 tb->fast_ipv6_only = ipv6_only_sock(sk);
500 tb->fast_sk_family = sk->sk_family;
501 #if IS_ENABLED(CONFIG_IPV6)
502 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
506 tb->fastreuseport = 0;
511 /* Obtain a reference to a local port for the given sock,
512 * if snum is zero it means select any available local port.
513 * We try to allocate an odd port (and leave even ports for connect())
515 int inet_csk_get_port(struct sock *sk, unsigned short snum)
517 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
518 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
519 bool found_port = false, check_bind_conflict = true;
520 bool bhash_created = false, bhash2_created = false;
521 int ret = -EADDRINUSE, port = snum, l3mdev;
522 struct inet_bind_hashbucket *head, *head2;
523 struct inet_bind2_bucket *tb2 = NULL;
524 struct inet_bind_bucket *tb = NULL;
525 bool head2_lock_acquired = false;
526 struct net *net = sock_net(sk);
528 l3mdev = inet_sk_bound_l3mdev(sk);
531 head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
535 head2_lock_acquired = true;
541 head = &hinfo->bhash[inet_bhashfn(net, port,
543 spin_lock_bh(&head->lock);
544 inet_bind_bucket_for_each(tb, &head->chain)
545 if (inet_bind_bucket_match(tb, net, port, l3mdev))
550 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
554 bhash_created = true;
558 if (!hlist_empty(&tb->bhash2)) {
559 if (sk->sk_reuse == SK_FORCE_REUSE ||
560 (tb->fastreuse > 0 && reuse) ||
561 sk_reuseport_match(tb, sk))
562 check_bind_conflict = false;
565 if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
566 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
570 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
571 spin_lock(&head2->lock);
572 head2_lock_acquired = true;
573 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
577 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
581 bhash2_created = true;
584 if (!found_port && check_bind_conflict) {
585 if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
590 inet_csk_update_fastreuse(tb, sk);
592 if (!inet_csk(sk)->icsk_bind_hash)
593 inet_bind_hash(sk, tb, tb2, port);
594 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
595 WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
601 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, tb2);
603 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
605 if (head2_lock_acquired)
606 spin_unlock(&head2->lock);
607 spin_unlock_bh(&head->lock);
610 EXPORT_SYMBOL_GPL(inet_csk_get_port);
613 * Wait for an incoming connection, avoid race conditions. This must be called
614 * with the socket locked.
616 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
618 struct inet_connection_sock *icsk = inet_csk(sk);
623 * True wake-one mechanism for incoming connections: only
624 * one process gets woken up, not the 'whole herd'.
625 * Since we do not 'race & poll' for established sockets
626 * anymore, the common case will execute the loop only once.
628 * Subtle issue: "add_wait_queue_exclusive()" will be added
629 * after any current non-exclusive waiters, and we know that
630 * it will always _stay_ after any new non-exclusive waiters
631 * because all non-exclusive waiters are added at the
632 * beginning of the wait-queue. As such, it's ok to "drop"
633 * our exclusiveness temporarily when we get woken up without
634 * having to remove and re-insert us on the wait queue.
637 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
640 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
641 timeo = schedule_timeout(timeo);
642 sched_annotate_sleep();
645 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
648 if (sk->sk_state != TCP_LISTEN)
650 err = sock_intr_errno(timeo);
651 if (signal_pending(current))
657 finish_wait(sk_sleep(sk), &wait);
662 * This will accept the next outstanding connection.
664 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
666 struct inet_connection_sock *icsk = inet_csk(sk);
667 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
668 struct request_sock *req;
674 /* We need to make sure that this socket is listening,
675 * and that it has something pending.
678 if (sk->sk_state != TCP_LISTEN)
681 /* Find already established connection */
682 if (reqsk_queue_empty(queue)) {
683 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
685 /* If this is a non blocking socket don't sleep */
690 error = inet_csk_wait_for_connect(sk, timeo);
694 req = reqsk_queue_remove(queue, sk);
697 if (sk->sk_protocol == IPPROTO_TCP &&
698 tcp_rsk(req)->tfo_listener) {
699 spin_lock_bh(&queue->fastopenq.lock);
700 if (tcp_rsk(req)->tfo_listener) {
701 /* We are still waiting for the final ACK from 3WHS
702 * so can't free req now. Instead, we set req->sk to
703 * NULL to signify that the child socket is taken
704 * so reqsk_fastopen_remove() will free the req
705 * when 3WHS finishes (or is aborted).
710 spin_unlock_bh(&queue->fastopenq.lock);
715 if (newsk && mem_cgroup_sockets_enabled) {
718 /* atomically get the memory usage, set and charge the
723 mem_cgroup_sk_alloc(newsk);
724 if (newsk->sk_memcg) {
725 /* The socket has not been accepted yet, no need
726 * to look at newsk->sk_wmem_queued.
728 amt = sk_mem_pages(newsk->sk_forward_alloc +
729 atomic_read(&newsk->sk_rmem_alloc));
733 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
734 GFP_KERNEL | __GFP_NOFAIL);
742 inet_init_csk_locks(newsk);
751 EXPORT_SYMBOL(inet_csk_accept);
754 * Using different timers for retransmit, delayed acks and probes
755 * We may wish use just one timer maintaining a list of expire jiffies
758 void inet_csk_init_xmit_timers(struct sock *sk,
759 void (*retransmit_handler)(struct timer_list *t),
760 void (*delack_handler)(struct timer_list *t),
761 void (*keepalive_handler)(struct timer_list *t))
763 struct inet_connection_sock *icsk = inet_csk(sk);
765 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
766 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
767 timer_setup(&sk->sk_timer, keepalive_handler, 0);
768 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
770 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
772 void inet_csk_clear_xmit_timers(struct sock *sk)
774 struct inet_connection_sock *icsk = inet_csk(sk);
776 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
778 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
779 sk_stop_timer(sk, &icsk->icsk_delack_timer);
780 sk_stop_timer(sk, &sk->sk_timer);
782 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
784 void inet_csk_clear_xmit_timers_sync(struct sock *sk)
786 struct inet_connection_sock *icsk = inet_csk(sk);
788 /* ongoing timer handlers need to acquire socket lock. */
789 sock_not_owned_by_me(sk);
791 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
793 sk_stop_timer_sync(sk, &icsk->icsk_retransmit_timer);
794 sk_stop_timer_sync(sk, &icsk->icsk_delack_timer);
795 sk_stop_timer_sync(sk, &sk->sk_timer);
798 void inet_csk_delete_keepalive_timer(struct sock *sk)
800 sk_stop_timer(sk, &sk->sk_timer);
802 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
804 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
806 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
808 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
810 struct dst_entry *inet_csk_route_req(const struct sock *sk,
812 const struct request_sock *req)
814 const struct inet_request_sock *ireq = inet_rsk(req);
815 struct net *net = read_pnet(&ireq->ireq_net);
816 struct ip_options_rcu *opt;
820 opt = rcu_dereference(ireq->ireq_opt);
822 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
823 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
824 sk->sk_protocol, inet_sk_flowi_flags(sk),
825 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
826 ireq->ir_loc_addr, ireq->ir_rmt_port,
827 htons(ireq->ir_num), sk->sk_uid);
828 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
829 rt = ip_route_output_flow(net, fl4, sk);
832 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
841 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
844 EXPORT_SYMBOL_GPL(inet_csk_route_req);
846 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
848 const struct request_sock *req)
850 const struct inet_request_sock *ireq = inet_rsk(req);
851 struct net *net = read_pnet(&ireq->ireq_net);
852 struct inet_sock *newinet = inet_sk(newsk);
853 struct ip_options_rcu *opt;
857 opt = rcu_dereference(ireq->ireq_opt);
858 fl4 = &newinet->cork.fl.u.ip4;
860 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
861 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
862 sk->sk_protocol, inet_sk_flowi_flags(sk),
863 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
864 ireq->ir_loc_addr, ireq->ir_rmt_port,
865 htons(ireq->ir_num), sk->sk_uid);
866 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
867 rt = ip_route_output_flow(net, fl4, sk);
870 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
877 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
880 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
882 /* Decide when to expire the request and when to resend SYN-ACK */
883 static void syn_ack_recalc(struct request_sock *req,
884 const int max_syn_ack_retries,
885 const u8 rskq_defer_accept,
886 int *expire, int *resend)
888 if (!rskq_defer_accept) {
889 *expire = req->num_timeout >= max_syn_ack_retries;
893 *expire = req->num_timeout >= max_syn_ack_retries &&
894 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
895 /* Do not resend while waiting for data after ACK,
896 * start to resend on end of deferring period to give
897 * last chance for data or ACK to create established socket.
899 *resend = !inet_rsk(req)->acked ||
900 req->num_timeout >= rskq_defer_accept - 1;
903 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
905 int err = req->rsk_ops->rtx_syn_ack(parent, req);
911 EXPORT_SYMBOL(inet_rtx_syn_ack);
913 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
916 struct sock *req_sk, *nreq_sk;
917 struct request_sock *nreq;
919 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
921 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
923 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
928 req_sk = req_to_sk(req);
929 nreq_sk = req_to_sk(nreq);
931 memcpy(nreq_sk, req_sk,
932 offsetof(struct sock, sk_dontcopy_begin));
933 memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
934 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
936 sk_node_init(&nreq_sk->sk_node);
937 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
938 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
939 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
941 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
943 nreq->rsk_listener = sk;
945 /* We need not acquire fastopenq->lock
946 * because the child socket is locked in inet_csk_listen_stop().
948 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
949 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
954 static void reqsk_queue_migrated(struct request_sock_queue *queue,
955 const struct request_sock *req)
957 if (req->num_timeout == 0)
958 atomic_inc(&queue->young);
959 atomic_inc(&queue->qlen);
962 static void reqsk_migrate_reset(struct request_sock *req)
964 req->saved_syn = NULL;
965 #if IS_ENABLED(CONFIG_IPV6)
966 inet_rsk(req)->ipv6_opt = NULL;
967 inet_rsk(req)->pktopts = NULL;
969 inet_rsk(req)->ireq_opt = NULL;
973 /* return true if req was found in the ehash table */
974 static bool reqsk_queue_unlink(struct request_sock *req)
976 struct sock *sk = req_to_sk(req);
980 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
981 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
984 found = __sk_nulls_del_node_init_rcu(sk);
987 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
992 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
994 bool unlinked = reqsk_queue_unlink(req);
997 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
1002 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
1004 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
1006 inet_csk_reqsk_queue_drop(sk, req);
1009 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
1011 static void reqsk_timer_handler(struct timer_list *t)
1013 struct request_sock *req = from_timer(req, t, rsk_timer);
1014 struct request_sock *nreq = NULL, *oreq = req;
1015 struct sock *sk_listener = req->rsk_listener;
1016 struct inet_connection_sock *icsk;
1017 struct request_sock_queue *queue;
1019 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
1021 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
1024 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1028 nreq = inet_reqsk_clone(req, nsk);
1032 /* The new timer for the cloned req can decrease the 2
1033 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1034 * hold another count to prevent use-after-free and
1035 * call reqsk_put() just before return.
1037 refcount_set(&nreq->rsk_refcnt, 2 + 1);
1038 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1039 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1045 icsk = inet_csk(sk_listener);
1046 net = sock_net(sk_listener);
1047 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1048 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1049 /* Normally all the openreqs are young and become mature
1050 * (i.e. converted to established socket) for first timeout.
1051 * If synack was not acknowledged for 1 second, it means
1052 * one of the following things: synack was lost, ack was lost,
1053 * rtt is high or nobody planned to ack (i.e. synflood).
1054 * When server is a bit loaded, queue is populated with old
1055 * open requests, reducing effective size of queue.
1056 * When server is well loaded, queue size reduces to zero
1057 * after several minutes of work. It is not synflood,
1058 * it is normal operation. The solution is pruning
1059 * too old entries overriding normal timeout, when
1060 * situation becomes dangerous.
1062 * Essentially, we reserve half of room for young
1063 * embrions; and abort old ones without pity, if old
1064 * ones are about to clog our table.
1066 queue = &icsk->icsk_accept_queue;
1067 qlen = reqsk_queue_len(queue);
1068 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1069 int young = reqsk_queue_len_young(queue) << 1;
1071 while (max_syn_ack_retries > 2) {
1074 max_syn_ack_retries--;
1078 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1080 req->rsk_ops->syn_ack_timeout(req);
1083 !inet_rtx_syn_ack(sk_listener, req) ||
1084 inet_rsk(req)->acked)) {
1085 if (req->num_timeout++ == 0)
1086 atomic_dec(&queue->young);
1087 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1092 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1094 inet_csk_reqsk_queue_drop(sk_listener, nreq);
1098 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1099 reqsk_migrate_reset(oreq);
1100 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1107 /* Even if we can clone the req, we may need not retransmit any more
1108 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1109 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1112 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1114 reqsk_migrate_reset(nreq);
1115 reqsk_queue_removed(queue, nreq);
1120 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1123 static void reqsk_queue_hash_req(struct request_sock *req,
1124 unsigned long timeout)
1126 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1127 mod_timer(&req->rsk_timer, jiffies + timeout);
1129 inet_ehash_insert(req_to_sk(req), NULL, NULL);
1130 /* before letting lookups find us, make sure all req fields
1131 * are committed to memory and refcnt initialized.
1134 refcount_set(&req->rsk_refcnt, 2 + 1);
1137 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1138 unsigned long timeout)
1140 reqsk_queue_hash_req(req, timeout);
1141 inet_csk_reqsk_queue_added(sk);
1143 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1145 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1146 const gfp_t priority)
1148 struct inet_connection_sock *icsk = inet_csk(newsk);
1150 if (!icsk->icsk_ulp_ops)
1153 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1157 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1158 * @sk: the socket to clone
1159 * @req: request_sock
1160 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1162 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1164 struct sock *inet_csk_clone_lock(const struct sock *sk,
1165 const struct request_sock *req,
1166 const gfp_t priority)
1168 struct sock *newsk = sk_clone_lock(sk, priority);
1171 struct inet_connection_sock *newicsk = inet_csk(newsk);
1173 inet_sk_set_state(newsk, TCP_SYN_RECV);
1174 newicsk->icsk_bind_hash = NULL;
1175 newicsk->icsk_bind2_hash = NULL;
1177 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1178 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1179 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1181 /* listeners have SOCK_RCU_FREE, not the children */
1182 sock_reset_flag(newsk, SOCK_RCU_FREE);
1184 inet_sk(newsk)->mc_list = NULL;
1186 newsk->sk_mark = inet_rsk(req)->ir_mark;
1187 atomic64_set(&newsk->sk_cookie,
1188 atomic64_read(&inet_rsk(req)->ir_cookie));
1190 newicsk->icsk_retransmits = 0;
1191 newicsk->icsk_backoff = 0;
1192 newicsk->icsk_probes_out = 0;
1193 newicsk->icsk_probes_tstamp = 0;
1195 /* Deinitialize accept_queue to trap illegal accesses. */
1196 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1198 inet_clone_ulp(req, newsk, priority);
1200 security_inet_csk_clone(newsk, req);
1204 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1207 * At this point, there should be no process reference to this
1208 * socket, and thus no user references at all. Therefore we
1209 * can assume the socket waitqueue is inactive and nobody will
1210 * try to jump onto it.
1212 void inet_csk_destroy_sock(struct sock *sk)
1214 WARN_ON(sk->sk_state != TCP_CLOSE);
1215 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1217 /* It cannot be in hash table! */
1218 WARN_ON(!sk_unhashed(sk));
1220 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1221 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1223 sk->sk_prot->destroy(sk);
1225 sk_stream_kill_queues(sk);
1227 xfrm_sk_free_policy(sk);
1229 this_cpu_dec(*sk->sk_prot->orphan_count);
1233 EXPORT_SYMBOL(inet_csk_destroy_sock);
1235 /* This function allows to force a closure of a socket after the call to
1236 * tcp/dccp_create_openreq_child().
1238 void inet_csk_prepare_forced_close(struct sock *sk)
1239 __releases(&sk->sk_lock.slock)
1241 /* sk_clone_lock locked the socket and set refcnt to 2 */
1244 inet_csk_prepare_for_destroy_sock(sk);
1245 inet_sk(sk)->inet_num = 0;
1247 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1249 static int inet_ulp_can_listen(const struct sock *sk)
1251 const struct inet_connection_sock *icsk = inet_csk(sk);
1253 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1259 int inet_csk_listen_start(struct sock *sk)
1261 struct inet_connection_sock *icsk = inet_csk(sk);
1262 struct inet_sock *inet = inet_sk(sk);
1265 err = inet_ulp_can_listen(sk);
1269 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1271 sk->sk_ack_backlog = 0;
1272 inet_csk_delack_init(sk);
1274 /* There is race window here: we announce ourselves listening,
1275 * but this transition is still not validated by get_port().
1276 * It is OK, because this socket enters to hash table only
1277 * after validation is complete.
1279 inet_sk_state_store(sk, TCP_LISTEN);
1280 err = sk->sk_prot->get_port(sk, inet->inet_num);
1282 inet->inet_sport = htons(inet->inet_num);
1285 err = sk->sk_prot->hash(sk);
1291 inet_sk_set_state(sk, TCP_CLOSE);
1294 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1296 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1299 sk->sk_prot->disconnect(child, O_NONBLOCK);
1303 this_cpu_inc(*sk->sk_prot->orphan_count);
1305 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1306 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1307 BUG_ON(sk != req->rsk_listener);
1309 /* Paranoid, to prevent race condition if
1310 * an inbound pkt destined for child is
1311 * blocked by sock lock in tcp_v4_rcv().
1312 * Also to satisfy an assertion in
1313 * tcp_v4_destroy_sock().
1315 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1317 inet_csk_destroy_sock(child);
1320 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1321 struct request_sock *req,
1324 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1326 spin_lock(&queue->rskq_lock);
1327 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1328 inet_child_forget(sk, req, child);
1332 req->dl_next = NULL;
1333 if (queue->rskq_accept_head == NULL)
1334 WRITE_ONCE(queue->rskq_accept_head, req);
1336 queue->rskq_accept_tail->dl_next = req;
1337 queue->rskq_accept_tail = req;
1338 sk_acceptq_added(sk);
1340 spin_unlock(&queue->rskq_lock);
1343 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1345 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1346 struct request_sock *req, bool own_req)
1349 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1350 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1352 if (sk != req->rsk_listener) {
1353 /* another listening sk has been selected,
1354 * migrate the req to it.
1356 struct request_sock *nreq;
1358 /* hold a refcnt for the nreq->rsk_listener
1359 * which is assigned in inet_reqsk_clone()
1362 nreq = inet_reqsk_clone(req, sk);
1364 inet_child_forget(sk, req, child);
1368 refcount_set(&nreq->rsk_refcnt, 1);
1369 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1370 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1371 reqsk_migrate_reset(req);
1376 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1377 reqsk_migrate_reset(nreq);
1379 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1383 /* Too bad, another child took ownership of the request, undo. */
1385 bh_unlock_sock(child);
1389 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1392 * This routine closes sockets which have been at least partially
1393 * opened, but not yet accepted.
1395 void inet_csk_listen_stop(struct sock *sk)
1397 struct inet_connection_sock *icsk = inet_csk(sk);
1398 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1399 struct request_sock *next, *req;
1401 /* Following specs, it would be better either to send FIN
1402 * (and enter FIN-WAIT-1, it is normal close)
1403 * or to send active reset (abort).
1404 * Certainly, it is pretty dangerous while synflood, but it is
1405 * bad justification for our negligence 8)
1406 * To be honest, we are not able to make either
1407 * of the variants now. --ANK
1409 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1410 struct sock *child = req->sk, *nsk;
1411 struct request_sock *nreq;
1414 bh_lock_sock(child);
1415 WARN_ON(sock_owned_by_user(child));
1418 nsk = reuseport_migrate_sock(sk, child, NULL);
1420 nreq = inet_reqsk_clone(req, nsk);
1422 refcount_set(&nreq->rsk_refcnt, 1);
1424 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1425 __NET_INC_STATS(sock_net(nsk),
1426 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1427 reqsk_migrate_reset(req);
1429 __NET_INC_STATS(sock_net(nsk),
1430 LINUX_MIB_TCPMIGRATEREQFAILURE);
1431 reqsk_migrate_reset(nreq);
1435 /* inet_csk_reqsk_queue_add() has already
1436 * called inet_child_forget() on failure case.
1438 goto skip_child_forget;
1442 inet_child_forget(sk, req, child);
1445 bh_unlock_sock(child);
1451 if (queue->fastopenq.rskq_rst_head) {
1452 /* Free all the reqs queued in rskq_rst_head. */
1453 spin_lock_bh(&queue->fastopenq.lock);
1454 req = queue->fastopenq.rskq_rst_head;
1455 queue->fastopenq.rskq_rst_head = NULL;
1456 spin_unlock_bh(&queue->fastopenq.lock);
1457 while (req != NULL) {
1458 next = req->dl_next;
1463 WARN_ON_ONCE(sk->sk_ack_backlog);
1465 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1467 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1469 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1470 const struct inet_sock *inet = inet_sk(sk);
1472 sin->sin_family = AF_INET;
1473 sin->sin_addr.s_addr = inet->inet_daddr;
1474 sin->sin_port = inet->inet_dport;
1476 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1478 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1480 const struct inet_sock *inet = inet_sk(sk);
1481 const struct ip_options_rcu *inet_opt;
1482 __be32 daddr = inet->inet_daddr;
1487 inet_opt = rcu_dereference(inet->inet_opt);
1488 if (inet_opt && inet_opt->opt.srr)
1489 daddr = inet_opt->opt.faddr;
1491 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1492 inet->inet_saddr, inet->inet_dport,
1493 inet->inet_sport, sk->sk_protocol,
1494 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1498 sk_setup_caps(sk, &rt->dst);
1504 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1506 struct dst_entry *dst = __sk_dst_check(sk, 0);
1507 struct inet_sock *inet = inet_sk(sk);
1510 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1514 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1516 dst = __sk_dst_check(sk, 0);
1518 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1522 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);