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 int inet_csk_bind_conflict(const struct sock *sk,
134 const struct inet_bind_bucket *tb,
135 bool relax, bool reuseport_ok)
138 bool reuseport_cb_ok;
139 bool reuse = sk->sk_reuse;
140 bool reuseport = !!sk->sk_reuseport;
141 struct sock_reuseport *reuseport_cb;
142 kuid_t uid = sock_i_uid((struct sock *)sk);
145 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
146 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
147 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
151 * Unlike other sk lookup places we do not check
152 * for sk_net here, since _all_ the socks listed
153 * in tb->owners list belong to the same net - the
154 * one this bucket belongs to.
157 sk_for_each_bound(sk2, &tb->owners) {
159 (!sk->sk_bound_dev_if ||
160 !sk2->sk_bound_dev_if ||
161 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
162 if (reuse && sk2->sk_reuse &&
163 sk2->sk_state != TCP_LISTEN) {
166 reuseport && sk2->sk_reuseport &&
168 (sk2->sk_state == TCP_TIME_WAIT ||
169 uid_eq(uid, sock_i_uid(sk2))))) &&
170 inet_rcv_saddr_equal(sk, sk2, true))
172 } else if (!reuseport_ok ||
173 !reuseport || !sk2->sk_reuseport ||
175 (sk2->sk_state != TCP_TIME_WAIT &&
176 !uid_eq(uid, sock_i_uid(sk2)))) {
177 if (inet_rcv_saddr_equal(sk, sk2, true))
186 * Find an open port number for the socket. Returns with the
187 * inet_bind_hashbucket lock held.
189 static struct inet_bind_hashbucket *
190 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
192 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
194 struct inet_bind_hashbucket *head;
195 struct net *net = sock_net(sk);
197 int i, low, high, attempt_half;
198 struct inet_bind_bucket *tb;
199 u32 remaining, offset;
202 l3mdev = inet_sk_bound_l3mdev(sk);
204 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
206 inet_get_local_port_range(net, &low, &high);
207 high++; /* [32768, 60999] -> [32768, 61000[ */
211 int half = low + (((high - low) >> 2) << 1);
213 if (attempt_half == 1)
218 remaining = high - low;
219 if (likely(remaining > 1))
222 offset = prandom_u32() % remaining;
223 /* __inet_hash_connect() favors ports having @low parity
224 * We do the opposite to not pollute connect() users.
230 for (i = 0; i < remaining; i += 2, port += 2) {
231 if (unlikely(port >= high))
233 if (inet_is_local_reserved_port(net, port))
235 head = &hinfo->bhash[inet_bhashfn(net, port,
237 spin_lock_bh(&head->lock);
238 inet_bind_bucket_for_each(tb, &head->chain)
239 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
241 if (!inet_csk_bind_conflict(sk, tb, relax, false))
248 spin_unlock_bh(&head->lock);
254 goto other_parity_scan;
256 if (attempt_half == 1) {
257 /* OK we now try the upper half of the range */
259 goto other_half_scan;
262 if (net->ipv4.sysctl_ip_autobind_reuse && !relax) {
263 /* We still have a chance to connect to different destinations */
265 goto ports_exhausted;
274 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
277 kuid_t uid = sock_i_uid(sk);
279 if (tb->fastreuseport <= 0)
281 if (!sk->sk_reuseport)
283 if (rcu_access_pointer(sk->sk_reuseport_cb))
285 if (!uid_eq(tb->fastuid, uid))
287 /* We only need to check the rcv_saddr if this tb was once marked
288 * without fastreuseport and then was reset, as we can only know that
289 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
292 if (tb->fastreuseport == FASTREUSEPORT_ANY)
294 #if IS_ENABLED(CONFIG_IPV6)
295 if (tb->fast_sk_family == AF_INET6)
296 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
301 ipv6_only_sock(sk), true, false);
303 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
304 ipv6_only_sock(sk), true, false);
307 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
310 kuid_t uid = sock_i_uid(sk);
311 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
313 if (hlist_empty(&tb->owners)) {
314 tb->fastreuse = reuse;
315 if (sk->sk_reuseport) {
316 tb->fastreuseport = FASTREUSEPORT_ANY;
318 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
319 tb->fast_ipv6_only = ipv6_only_sock(sk);
320 tb->fast_sk_family = sk->sk_family;
321 #if IS_ENABLED(CONFIG_IPV6)
322 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
325 tb->fastreuseport = 0;
330 if (sk->sk_reuseport) {
331 /* We didn't match or we don't have fastreuseport set on
332 * the tb, but we have sk_reuseport set on this socket
333 * and we know that there are no bind conflicts with
334 * this socket in this tb, so reset our tb's reuseport
335 * settings so that any subsequent sockets that match
336 * our current socket will be put on the fast path.
338 * If we reset we need to set FASTREUSEPORT_STRICT so we
339 * do extra checking for all subsequent sk_reuseport
342 if (!sk_reuseport_match(tb, sk)) {
343 tb->fastreuseport = FASTREUSEPORT_STRICT;
345 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
346 tb->fast_ipv6_only = ipv6_only_sock(sk);
347 tb->fast_sk_family = sk->sk_family;
348 #if IS_ENABLED(CONFIG_IPV6)
349 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
353 tb->fastreuseport = 0;
358 /* Obtain a reference to a local port for the given sock,
359 * if snum is zero it means select any available local port.
360 * We try to allocate an odd port (and leave even ports for connect())
362 int inet_csk_get_port(struct sock *sk, unsigned short snum)
364 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
365 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
366 int ret = 1, port = snum;
367 struct inet_bind_hashbucket *head;
368 struct net *net = sock_net(sk);
369 struct inet_bind_bucket *tb = NULL;
372 l3mdev = inet_sk_bound_l3mdev(sk);
375 head = inet_csk_find_open_port(sk, &tb, &port);
382 head = &hinfo->bhash[inet_bhashfn(net, port,
384 spin_lock_bh(&head->lock);
385 inet_bind_bucket_for_each(tb, &head->chain)
386 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
390 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
391 net, head, port, l3mdev);
395 if (!hlist_empty(&tb->owners)) {
396 if (sk->sk_reuse == SK_FORCE_REUSE)
399 if ((tb->fastreuse > 0 && reuse) ||
400 sk_reuseport_match(tb, sk))
402 if (inet_csk_bind_conflict(sk, tb, true, true))
406 inet_csk_update_fastreuse(tb, sk);
408 if (!inet_csk(sk)->icsk_bind_hash)
409 inet_bind_hash(sk, tb, port);
410 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
414 spin_unlock_bh(&head->lock);
417 EXPORT_SYMBOL_GPL(inet_csk_get_port);
420 * Wait for an incoming connection, avoid race conditions. This must be called
421 * with the socket locked.
423 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
425 struct inet_connection_sock *icsk = inet_csk(sk);
430 * True wake-one mechanism for incoming connections: only
431 * one process gets woken up, not the 'whole herd'.
432 * Since we do not 'race & poll' for established sockets
433 * anymore, the common case will execute the loop only once.
435 * Subtle issue: "add_wait_queue_exclusive()" will be added
436 * after any current non-exclusive waiters, and we know that
437 * it will always _stay_ after any new non-exclusive waiters
438 * because all non-exclusive waiters are added at the
439 * beginning of the wait-queue. As such, it's ok to "drop"
440 * our exclusiveness temporarily when we get woken up without
441 * having to remove and re-insert us on the wait queue.
444 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
447 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
448 timeo = schedule_timeout(timeo);
449 sched_annotate_sleep();
452 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
455 if (sk->sk_state != TCP_LISTEN)
457 err = sock_intr_errno(timeo);
458 if (signal_pending(current))
464 finish_wait(sk_sleep(sk), &wait);
469 * This will accept the next outstanding connection.
471 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
473 struct inet_connection_sock *icsk = inet_csk(sk);
474 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
475 struct request_sock *req;
481 /* We need to make sure that this socket is listening,
482 * and that it has something pending.
485 if (sk->sk_state != TCP_LISTEN)
488 /* Find already established connection */
489 if (reqsk_queue_empty(queue)) {
490 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
492 /* If this is a non blocking socket don't sleep */
497 error = inet_csk_wait_for_connect(sk, timeo);
501 req = reqsk_queue_remove(queue, sk);
504 if (sk->sk_protocol == IPPROTO_TCP &&
505 tcp_rsk(req)->tfo_listener) {
506 spin_lock_bh(&queue->fastopenq.lock);
507 if (tcp_rsk(req)->tfo_listener) {
508 /* We are still waiting for the final ACK from 3WHS
509 * so can't free req now. Instead, we set req->sk to
510 * NULL to signify that the child socket is taken
511 * so reqsk_fastopen_remove() will free the req
512 * when 3WHS finishes (or is aborted).
517 spin_unlock_bh(&queue->fastopenq.lock);
522 if (newsk && mem_cgroup_sockets_enabled) {
525 /* atomically get the memory usage, set and charge the
530 /* The socket has not been accepted yet, no need to look at
531 * newsk->sk_wmem_queued.
533 amt = sk_mem_pages(newsk->sk_forward_alloc +
534 atomic_read(&newsk->sk_rmem_alloc));
535 mem_cgroup_sk_alloc(newsk);
536 if (newsk->sk_memcg && amt)
537 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
538 GFP_KERNEL | __GFP_NOFAIL);
551 EXPORT_SYMBOL(inet_csk_accept);
554 * Using different timers for retransmit, delayed acks and probes
555 * We may wish use just one timer maintaining a list of expire jiffies
558 void inet_csk_init_xmit_timers(struct sock *sk,
559 void (*retransmit_handler)(struct timer_list *t),
560 void (*delack_handler)(struct timer_list *t),
561 void (*keepalive_handler)(struct timer_list *t))
563 struct inet_connection_sock *icsk = inet_csk(sk);
565 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
566 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
567 timer_setup(&sk->sk_timer, keepalive_handler, 0);
568 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
570 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
572 void inet_csk_clear_xmit_timers(struct sock *sk)
574 struct inet_connection_sock *icsk = inet_csk(sk);
576 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
578 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
579 sk_stop_timer(sk, &icsk->icsk_delack_timer);
580 sk_stop_timer(sk, &sk->sk_timer);
582 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
584 void inet_csk_delete_keepalive_timer(struct sock *sk)
586 sk_stop_timer(sk, &sk->sk_timer);
588 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
590 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
592 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
594 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
596 struct dst_entry *inet_csk_route_req(const struct sock *sk,
598 const struct request_sock *req)
600 const struct inet_request_sock *ireq = inet_rsk(req);
601 struct net *net = read_pnet(&ireq->ireq_net);
602 struct ip_options_rcu *opt;
606 opt = rcu_dereference(ireq->ireq_opt);
608 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
609 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
610 sk->sk_protocol, inet_sk_flowi_flags(sk),
611 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
612 ireq->ir_loc_addr, ireq->ir_rmt_port,
613 htons(ireq->ir_num), sk->sk_uid);
614 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
615 rt = ip_route_output_flow(net, fl4, sk);
618 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
627 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
630 EXPORT_SYMBOL_GPL(inet_csk_route_req);
632 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
634 const struct request_sock *req)
636 const struct inet_request_sock *ireq = inet_rsk(req);
637 struct net *net = read_pnet(&ireq->ireq_net);
638 struct inet_sock *newinet = inet_sk(newsk);
639 struct ip_options_rcu *opt;
643 opt = rcu_dereference(ireq->ireq_opt);
644 fl4 = &newinet->cork.fl.u.ip4;
646 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
647 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
648 sk->sk_protocol, inet_sk_flowi_flags(sk),
649 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
650 ireq->ir_loc_addr, ireq->ir_rmt_port,
651 htons(ireq->ir_num), sk->sk_uid);
652 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
653 rt = ip_route_output_flow(net, fl4, sk);
656 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
663 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
666 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
668 /* Decide when to expire the request and when to resend SYN-ACK */
669 static void syn_ack_recalc(struct request_sock *req,
670 const int max_syn_ack_retries,
671 const u8 rskq_defer_accept,
672 int *expire, int *resend)
674 if (!rskq_defer_accept) {
675 *expire = req->num_timeout >= max_syn_ack_retries;
679 *expire = req->num_timeout >= max_syn_ack_retries &&
680 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
681 /* Do not resend while waiting for data after ACK,
682 * start to resend on end of deferring period to give
683 * last chance for data or ACK to create established socket.
685 *resend = !inet_rsk(req)->acked ||
686 req->num_timeout >= rskq_defer_accept - 1;
689 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
691 int err = req->rsk_ops->rtx_syn_ack(parent, req);
697 EXPORT_SYMBOL(inet_rtx_syn_ack);
699 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
702 struct sock *req_sk, *nreq_sk;
703 struct request_sock *nreq;
705 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
707 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
709 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
714 req_sk = req_to_sk(req);
715 nreq_sk = req_to_sk(nreq);
717 memcpy(nreq_sk, req_sk,
718 offsetof(struct sock, sk_dontcopy_begin));
719 memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
720 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
722 sk_node_init(&nreq_sk->sk_node);
723 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
725 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
727 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
729 nreq->rsk_listener = sk;
731 /* We need not acquire fastopenq->lock
732 * because the child socket is locked in inet_csk_listen_stop().
734 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
735 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
740 static void reqsk_queue_migrated(struct request_sock_queue *queue,
741 const struct request_sock *req)
743 if (req->num_timeout == 0)
744 atomic_inc(&queue->young);
745 atomic_inc(&queue->qlen);
748 static void reqsk_migrate_reset(struct request_sock *req)
750 req->saved_syn = NULL;
751 #if IS_ENABLED(CONFIG_IPV6)
752 inet_rsk(req)->ipv6_opt = NULL;
753 inet_rsk(req)->pktopts = NULL;
755 inet_rsk(req)->ireq_opt = NULL;
759 /* return true if req was found in the ehash table */
760 static bool reqsk_queue_unlink(struct request_sock *req)
762 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
765 if (sk_hashed(req_to_sk(req))) {
766 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
769 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
772 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
777 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
779 bool unlinked = reqsk_queue_unlink(req);
782 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
787 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
789 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
791 inet_csk_reqsk_queue_drop(sk, req);
794 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
796 static void reqsk_timer_handler(struct timer_list *t)
798 struct request_sock *req = from_timer(req, t, rsk_timer);
799 struct request_sock *nreq = NULL, *oreq = req;
800 struct sock *sk_listener = req->rsk_listener;
801 struct inet_connection_sock *icsk;
802 struct request_sock_queue *queue;
804 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
806 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
809 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
813 nreq = inet_reqsk_clone(req, nsk);
817 /* The new timer for the cloned req can decrease the 2
818 * by calling inet_csk_reqsk_queue_drop_and_put(), so
819 * hold another count to prevent use-after-free and
820 * call reqsk_put() just before return.
822 refcount_set(&nreq->rsk_refcnt, 2 + 1);
823 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
824 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
830 icsk = inet_csk(sk_listener);
831 net = sock_net(sk_listener);
832 max_syn_ack_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
833 /* Normally all the openreqs are young and become mature
834 * (i.e. converted to established socket) for first timeout.
835 * If synack was not acknowledged for 1 second, it means
836 * one of the following things: synack was lost, ack was lost,
837 * rtt is high or nobody planned to ack (i.e. synflood).
838 * When server is a bit loaded, queue is populated with old
839 * open requests, reducing effective size of queue.
840 * When server is well loaded, queue size reduces to zero
841 * after several minutes of work. It is not synflood,
842 * it is normal operation. The solution is pruning
843 * too old entries overriding normal timeout, when
844 * situation becomes dangerous.
846 * Essentially, we reserve half of room for young
847 * embrions; and abort old ones without pity, if old
848 * ones are about to clog our table.
850 queue = &icsk->icsk_accept_queue;
851 qlen = reqsk_queue_len(queue);
852 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
853 int young = reqsk_queue_len_young(queue) << 1;
855 while (max_syn_ack_retries > 2) {
858 max_syn_ack_retries--;
862 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
864 req->rsk_ops->syn_ack_timeout(req);
867 !inet_rtx_syn_ack(sk_listener, req) ||
868 inet_rsk(req)->acked)) {
871 if (req->num_timeout++ == 0)
872 atomic_dec(&queue->young);
873 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
874 mod_timer(&req->rsk_timer, jiffies + timeo);
879 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
881 inet_csk_reqsk_queue_drop(sk_listener, nreq);
885 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
886 reqsk_migrate_reset(oreq);
887 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
894 /* Even if we can clone the req, we may need not retransmit any more
895 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
896 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
899 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
901 reqsk_migrate_reset(nreq);
902 reqsk_queue_removed(queue, nreq);
907 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
910 static void reqsk_queue_hash_req(struct request_sock *req,
911 unsigned long timeout)
913 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
914 mod_timer(&req->rsk_timer, jiffies + timeout);
916 inet_ehash_insert(req_to_sk(req), NULL, NULL);
917 /* before letting lookups find us, make sure all req fields
918 * are committed to memory and refcnt initialized.
921 refcount_set(&req->rsk_refcnt, 2 + 1);
924 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
925 unsigned long timeout)
927 reqsk_queue_hash_req(req, timeout);
928 inet_csk_reqsk_queue_added(sk);
930 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
932 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
933 const gfp_t priority)
935 struct inet_connection_sock *icsk = inet_csk(newsk);
937 if (!icsk->icsk_ulp_ops)
940 if (icsk->icsk_ulp_ops->clone)
941 icsk->icsk_ulp_ops->clone(req, newsk, priority);
945 * inet_csk_clone_lock - clone an inet socket, and lock its clone
946 * @sk: the socket to clone
948 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
950 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
952 struct sock *inet_csk_clone_lock(const struct sock *sk,
953 const struct request_sock *req,
954 const gfp_t priority)
956 struct sock *newsk = sk_clone_lock(sk, priority);
959 struct inet_connection_sock *newicsk = inet_csk(newsk);
961 inet_sk_set_state(newsk, TCP_SYN_RECV);
962 newicsk->icsk_bind_hash = NULL;
964 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
965 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
966 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
968 /* listeners have SOCK_RCU_FREE, not the children */
969 sock_reset_flag(newsk, SOCK_RCU_FREE);
971 inet_sk(newsk)->mc_list = NULL;
973 newsk->sk_mark = inet_rsk(req)->ir_mark;
974 atomic64_set(&newsk->sk_cookie,
975 atomic64_read(&inet_rsk(req)->ir_cookie));
977 newicsk->icsk_retransmits = 0;
978 newicsk->icsk_backoff = 0;
979 newicsk->icsk_probes_out = 0;
980 newicsk->icsk_probes_tstamp = 0;
982 /* Deinitialize accept_queue to trap illegal accesses. */
983 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
985 inet_clone_ulp(req, newsk, priority);
987 security_inet_csk_clone(newsk, req);
991 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
994 * At this point, there should be no process reference to this
995 * socket, and thus no user references at all. Therefore we
996 * can assume the socket waitqueue is inactive and nobody will
997 * try to jump onto it.
999 void inet_csk_destroy_sock(struct sock *sk)
1001 WARN_ON(sk->sk_state != TCP_CLOSE);
1002 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1004 /* It cannot be in hash table! */
1005 WARN_ON(!sk_unhashed(sk));
1007 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1008 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1010 sk->sk_prot->destroy(sk);
1012 sk_stream_kill_queues(sk);
1014 xfrm_sk_free_policy(sk);
1016 sk_refcnt_debug_release(sk);
1018 percpu_counter_dec(sk->sk_prot->orphan_count);
1022 EXPORT_SYMBOL(inet_csk_destroy_sock);
1024 /* This function allows to force a closure of a socket after the call to
1025 * tcp/dccp_create_openreq_child().
1027 void inet_csk_prepare_forced_close(struct sock *sk)
1028 __releases(&sk->sk_lock.slock)
1030 /* sk_clone_lock locked the socket and set refcnt to 2 */
1033 inet_csk_prepare_for_destroy_sock(sk);
1034 inet_sk(sk)->inet_num = 0;
1036 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1038 int inet_csk_listen_start(struct sock *sk, int backlog)
1040 struct inet_connection_sock *icsk = inet_csk(sk);
1041 struct inet_sock *inet = inet_sk(sk);
1042 int err = -EADDRINUSE;
1044 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1046 sk->sk_ack_backlog = 0;
1047 inet_csk_delack_init(sk);
1049 /* There is race window here: we announce ourselves listening,
1050 * but this transition is still not validated by get_port().
1051 * It is OK, because this socket enters to hash table only
1052 * after validation is complete.
1054 inet_sk_state_store(sk, TCP_LISTEN);
1055 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
1056 inet->inet_sport = htons(inet->inet_num);
1059 err = sk->sk_prot->hash(sk);
1065 inet_sk_set_state(sk, TCP_CLOSE);
1068 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1070 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1073 sk->sk_prot->disconnect(child, O_NONBLOCK);
1077 percpu_counter_inc(sk->sk_prot->orphan_count);
1079 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1080 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1081 BUG_ON(sk != req->rsk_listener);
1083 /* Paranoid, to prevent race condition if
1084 * an inbound pkt destined for child is
1085 * blocked by sock lock in tcp_v4_rcv().
1086 * Also to satisfy an assertion in
1087 * tcp_v4_destroy_sock().
1089 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1091 inet_csk_destroy_sock(child);
1094 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1095 struct request_sock *req,
1098 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1100 spin_lock(&queue->rskq_lock);
1101 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1102 inet_child_forget(sk, req, child);
1106 req->dl_next = NULL;
1107 if (queue->rskq_accept_head == NULL)
1108 WRITE_ONCE(queue->rskq_accept_head, req);
1110 queue->rskq_accept_tail->dl_next = req;
1111 queue->rskq_accept_tail = req;
1112 sk_acceptq_added(sk);
1114 spin_unlock(&queue->rskq_lock);
1117 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1119 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1120 struct request_sock *req, bool own_req)
1123 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1124 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1126 if (sk != req->rsk_listener) {
1127 /* another listening sk has been selected,
1128 * migrate the req to it.
1130 struct request_sock *nreq;
1132 /* hold a refcnt for the nreq->rsk_listener
1133 * which is assigned in inet_reqsk_clone()
1136 nreq = inet_reqsk_clone(req, sk);
1138 inet_child_forget(sk, req, child);
1142 refcount_set(&nreq->rsk_refcnt, 1);
1143 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1144 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1145 reqsk_migrate_reset(req);
1150 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1151 reqsk_migrate_reset(nreq);
1153 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1157 /* Too bad, another child took ownership of the request, undo. */
1159 bh_unlock_sock(child);
1163 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1166 * This routine closes sockets which have been at least partially
1167 * opened, but not yet accepted.
1169 void inet_csk_listen_stop(struct sock *sk)
1171 struct inet_connection_sock *icsk = inet_csk(sk);
1172 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1173 struct request_sock *next, *req;
1175 /* Following specs, it would be better either to send FIN
1176 * (and enter FIN-WAIT-1, it is normal close)
1177 * or to send active reset (abort).
1178 * Certainly, it is pretty dangerous while synflood, but it is
1179 * bad justification for our negligence 8)
1180 * To be honest, we are not able to make either
1181 * of the variants now. --ANK
1183 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1184 struct sock *child = req->sk, *nsk;
1185 struct request_sock *nreq;
1188 bh_lock_sock(child);
1189 WARN_ON(sock_owned_by_user(child));
1192 nsk = reuseport_migrate_sock(sk, child, NULL);
1194 nreq = inet_reqsk_clone(req, nsk);
1196 refcount_set(&nreq->rsk_refcnt, 1);
1198 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1199 __NET_INC_STATS(sock_net(nsk),
1200 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1201 reqsk_migrate_reset(req);
1203 __NET_INC_STATS(sock_net(nsk),
1204 LINUX_MIB_TCPMIGRATEREQFAILURE);
1205 reqsk_migrate_reset(nreq);
1209 /* inet_csk_reqsk_queue_add() has already
1210 * called inet_child_forget() on failure case.
1212 goto skip_child_forget;
1216 inet_child_forget(sk, req, child);
1219 bh_unlock_sock(child);
1225 if (queue->fastopenq.rskq_rst_head) {
1226 /* Free all the reqs queued in rskq_rst_head. */
1227 spin_lock_bh(&queue->fastopenq.lock);
1228 req = queue->fastopenq.rskq_rst_head;
1229 queue->fastopenq.rskq_rst_head = NULL;
1230 spin_unlock_bh(&queue->fastopenq.lock);
1231 while (req != NULL) {
1232 next = req->dl_next;
1237 WARN_ON_ONCE(sk->sk_ack_backlog);
1239 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1241 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1243 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1244 const struct inet_sock *inet = inet_sk(sk);
1246 sin->sin_family = AF_INET;
1247 sin->sin_addr.s_addr = inet->inet_daddr;
1248 sin->sin_port = inet->inet_dport;
1250 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1252 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1254 const struct inet_sock *inet = inet_sk(sk);
1255 const struct ip_options_rcu *inet_opt;
1256 __be32 daddr = inet->inet_daddr;
1261 inet_opt = rcu_dereference(inet->inet_opt);
1262 if (inet_opt && inet_opt->opt.srr)
1263 daddr = inet_opt->opt.faddr;
1265 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1266 inet->inet_saddr, inet->inet_dport,
1267 inet->inet_sport, sk->sk_protocol,
1268 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1272 sk_setup_caps(sk, &rt->dst);
1278 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1280 struct dst_entry *dst = __sk_dst_check(sk, 0);
1281 struct inet_sock *inet = inet_sk(sk);
1284 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1288 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1290 dst = __sk_dst_check(sk, 0);
1292 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1296 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);