2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Support for INET connection oriented protocols.
8 * Authors: See the TCP sources
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or(at your option) any later version.
16 #include <linux/module.h>
17 #include <linux/jhash.h>
19 #include <net/inet_connection_sock.h>
20 #include <net/inet_hashtables.h>
21 #include <net/inet_timewait_sock.h>
23 #include <net/route.h>
24 #include <net/tcp_states.h>
28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
32 unsigned long *sysctl_local_reserved_ports;
33 EXPORT_SYMBOL(sysctl_local_reserved_ports);
35 void inet_get_local_port_range(struct net *net, int *low, int *high)
40 seq = read_seqbegin(&net->ipv4.sysctl_local_ports.lock);
42 *low = net->ipv4.sysctl_local_ports.range[0];
43 *high = net->ipv4.sysctl_local_ports.range[1];
44 } while (read_seqretry(&net->ipv4.sysctl_local_ports.lock, seq));
46 EXPORT_SYMBOL(inet_get_local_port_range);
48 int inet_csk_bind_conflict(const struct sock *sk,
49 const struct inet_bind_bucket *tb, bool relax)
52 int reuse = sk->sk_reuse;
53 int reuseport = sk->sk_reuseport;
54 kuid_t uid = sock_i_uid((struct sock *)sk);
57 * Unlike other sk lookup places we do not check
58 * for sk_net here, since _all_ the socks listed
59 * in tb->owners list belong to the same net - the
60 * one this bucket belongs to.
63 sk_for_each_bound(sk2, &tb->owners) {
65 !inet_v6_ipv6only(sk2) &&
66 (!sk->sk_bound_dev_if ||
67 !sk2->sk_bound_dev_if ||
68 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
69 if ((!reuse || !sk2->sk_reuse ||
70 sk2->sk_state == TCP_LISTEN) &&
71 (!reuseport || !sk2->sk_reuseport ||
72 (sk2->sk_state != TCP_TIME_WAIT &&
73 !uid_eq(uid, sock_i_uid(sk2))))) {
75 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
76 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
79 if (!relax && reuse && sk2->sk_reuse &&
80 sk2->sk_state != TCP_LISTEN) {
82 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
83 sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
90 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
92 /* Obtain a reference to a local port for the given sock,
93 * if snum is zero it means select any available local port.
95 int inet_csk_get_port(struct sock *sk, unsigned short snum)
97 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
98 struct inet_bind_hashbucket *head;
99 struct inet_bind_bucket *tb;
100 int ret, attempts = 5;
101 struct net *net = sock_net(sk);
102 int smallest_size = -1, smallest_rover;
103 kuid_t uid = sock_i_uid(sk);
107 int remaining, rover, low, high;
110 inet_get_local_port_range(net, &low, &high);
111 remaining = (high - low) + 1;
112 smallest_rover = rover = net_random() % remaining + low;
116 if (inet_is_reserved_local_port(rover))
118 head = &hashinfo->bhash[inet_bhashfn(net, rover,
119 hashinfo->bhash_size)];
120 spin_lock(&head->lock);
121 inet_bind_bucket_for_each(tb, &head->chain)
122 if (net_eq(ib_net(tb), net) && tb->port == rover) {
123 if (((tb->fastreuse > 0 &&
125 sk->sk_state != TCP_LISTEN) ||
126 (tb->fastreuseport > 0 &&
128 uid_eq(tb->fastuid, uid))) &&
129 (tb->num_owners < smallest_size || smallest_size == -1)) {
130 smallest_size = tb->num_owners;
131 smallest_rover = rover;
132 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
133 !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
134 snum = smallest_rover;
138 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
146 spin_unlock(&head->lock);
150 } while (--remaining > 0);
152 /* Exhausted local port range during search? It is not
153 * possible for us to be holding one of the bind hash
154 * locks if this test triggers, because if 'remaining'
155 * drops to zero, we broke out of the do/while loop at
156 * the top level, not from the 'break;' statement.
159 if (remaining <= 0) {
160 if (smallest_size != -1) {
161 snum = smallest_rover;
166 /* OK, here is the one we will use. HEAD is
167 * non-NULL and we hold it's mutex.
172 head = &hashinfo->bhash[inet_bhashfn(net, snum,
173 hashinfo->bhash_size)];
174 spin_lock(&head->lock);
175 inet_bind_bucket_for_each(tb, &head->chain)
176 if (net_eq(ib_net(tb), net) && tb->port == snum)
182 if (!hlist_empty(&tb->owners)) {
183 if (sk->sk_reuse == SK_FORCE_REUSE)
186 if (((tb->fastreuse > 0 &&
187 sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
188 (tb->fastreuseport > 0 &&
189 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
190 smallest_size == -1) {
194 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
195 if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||
196 (tb->fastreuseport > 0 &&
197 sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
198 smallest_size != -1 && --attempts >= 0) {
199 spin_unlock(&head->lock);
209 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
210 net, head, snum)) == NULL)
212 if (hlist_empty(&tb->owners)) {
213 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
217 if (sk->sk_reuseport) {
218 tb->fastreuseport = 1;
221 tb->fastreuseport = 0;
224 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
226 if (tb->fastreuseport &&
227 (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))
228 tb->fastreuseport = 0;
231 if (!inet_csk(sk)->icsk_bind_hash)
232 inet_bind_hash(sk, tb, snum);
233 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
237 spin_unlock(&head->lock);
242 EXPORT_SYMBOL_GPL(inet_csk_get_port);
245 * Wait for an incoming connection, avoid race conditions. This must be called
246 * with the socket locked.
248 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
250 struct inet_connection_sock *icsk = inet_csk(sk);
255 * True wake-one mechanism for incoming connections: only
256 * one process gets woken up, not the 'whole herd'.
257 * Since we do not 'race & poll' for established sockets
258 * anymore, the common case will execute the loop only once.
260 * Subtle issue: "add_wait_queue_exclusive()" will be added
261 * after any current non-exclusive waiters, and we know that
262 * it will always _stay_ after any new non-exclusive waiters
263 * because all non-exclusive waiters are added at the
264 * beginning of the wait-queue. As such, it's ok to "drop"
265 * our exclusiveness temporarily when we get woken up without
266 * having to remove and re-insert us on the wait queue.
269 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
272 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
273 timeo = schedule_timeout(timeo);
276 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
279 if (sk->sk_state != TCP_LISTEN)
281 err = sock_intr_errno(timeo);
282 if (signal_pending(current))
288 finish_wait(sk_sleep(sk), &wait);
293 * This will accept the next outstanding connection.
295 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
297 struct inet_connection_sock *icsk = inet_csk(sk);
298 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
300 struct request_sock *req;
305 /* We need to make sure that this socket is listening,
306 * and that it has something pending.
309 if (sk->sk_state != TCP_LISTEN)
312 /* Find already established connection */
313 if (reqsk_queue_empty(queue)) {
314 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
316 /* If this is a non blocking socket don't sleep */
321 error = inet_csk_wait_for_connect(sk, timeo);
325 req = reqsk_queue_remove(queue);
328 sk_acceptq_removed(sk);
329 if (sk->sk_protocol == IPPROTO_TCP && queue->fastopenq != NULL) {
330 spin_lock_bh(&queue->fastopenq->lock);
331 if (tcp_rsk(req)->listener) {
332 /* We are still waiting for the final ACK from 3WHS
333 * so can't free req now. Instead, we set req->sk to
334 * NULL to signify that the child socket is taken
335 * so reqsk_fastopen_remove() will free the req
336 * when 3WHS finishes (or is aborted).
341 spin_unlock_bh(&queue->fastopenq->lock);
354 EXPORT_SYMBOL(inet_csk_accept);
357 * Using different timers for retransmit, delayed acks and probes
358 * We may wish use just one timer maintaining a list of expire jiffies
361 void inet_csk_init_xmit_timers(struct sock *sk,
362 void (*retransmit_handler)(unsigned long),
363 void (*delack_handler)(unsigned long),
364 void (*keepalive_handler)(unsigned long))
366 struct inet_connection_sock *icsk = inet_csk(sk);
368 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
370 setup_timer(&icsk->icsk_delack_timer, delack_handler,
372 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
373 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
375 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
377 void inet_csk_clear_xmit_timers(struct sock *sk)
379 struct inet_connection_sock *icsk = inet_csk(sk);
381 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
383 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
384 sk_stop_timer(sk, &icsk->icsk_delack_timer);
385 sk_stop_timer(sk, &sk->sk_timer);
387 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
389 void inet_csk_delete_keepalive_timer(struct sock *sk)
391 sk_stop_timer(sk, &sk->sk_timer);
393 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
395 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
397 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
399 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
401 struct dst_entry *inet_csk_route_req(struct sock *sk,
403 const struct request_sock *req)
406 const struct inet_request_sock *ireq = inet_rsk(req);
407 struct ip_options_rcu *opt = inet_rsk(req)->opt;
408 struct net *net = sock_net(sk);
409 int flags = inet_sk_flowi_flags(sk);
411 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
412 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
415 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
416 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
417 security_req_classify_flow(req, flowi4_to_flowi(fl4));
418 rt = ip_route_output_flow(net, fl4, sk);
421 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
428 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
431 EXPORT_SYMBOL_GPL(inet_csk_route_req);
433 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
435 const struct request_sock *req)
437 const struct inet_request_sock *ireq = inet_rsk(req);
438 struct inet_sock *newinet = inet_sk(newsk);
439 struct ip_options_rcu *opt;
440 struct net *net = sock_net(sk);
444 fl4 = &newinet->cork.fl.u.ip4;
447 opt = rcu_dereference(newinet->inet_opt);
448 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
449 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
450 sk->sk_protocol, inet_sk_flowi_flags(sk),
451 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
452 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
453 security_req_classify_flow(req, flowi4_to_flowi(fl4));
454 rt = ip_route_output_flow(net, fl4, sk);
457 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
466 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
469 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
471 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
472 const u32 rnd, const u32 synq_hsize)
474 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
477 #if IS_ENABLED(CONFIG_IPV6)
478 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
480 #define AF_INET_FAMILY(fam) 1
483 struct request_sock *inet_csk_search_req(const struct sock *sk,
484 struct request_sock ***prevp,
485 const __be16 rport, const __be32 raddr,
488 const struct inet_connection_sock *icsk = inet_csk(sk);
489 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
490 struct request_sock *req, **prev;
492 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
493 lopt->nr_table_entries)];
494 (req = *prev) != NULL;
495 prev = &req->dl_next) {
496 const struct inet_request_sock *ireq = inet_rsk(req);
498 if (ireq->rmt_port == rport &&
499 ireq->rmt_addr == raddr &&
500 ireq->loc_addr == laddr &&
501 AF_INET_FAMILY(req->rsk_ops->family)) {
510 EXPORT_SYMBOL_GPL(inet_csk_search_req);
512 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
513 unsigned long timeout)
515 struct inet_connection_sock *icsk = inet_csk(sk);
516 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
517 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
518 lopt->hash_rnd, lopt->nr_table_entries);
520 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
521 inet_csk_reqsk_queue_added(sk, timeout);
523 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
525 /* Only thing we need from tcp.h */
526 extern int sysctl_tcp_synack_retries;
529 /* Decide when to expire the request and when to resend SYN-ACK */
530 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
531 const int max_retries,
532 const u8 rskq_defer_accept,
533 int *expire, int *resend)
535 if (!rskq_defer_accept) {
536 *expire = req->num_timeout >= thresh;
540 *expire = req->num_timeout >= thresh &&
541 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
543 * Do not resend while waiting for data after ACK,
544 * start to resend on end of deferring period to give
545 * last chance for data or ACK to create established socket.
547 *resend = !inet_rsk(req)->acked ||
548 req->num_timeout >= rskq_defer_accept - 1;
551 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req)
553 int err = req->rsk_ops->rtx_syn_ack(parent, req);
559 EXPORT_SYMBOL(inet_rtx_syn_ack);
561 void inet_csk_reqsk_queue_prune(struct sock *parent,
562 const unsigned long interval,
563 const unsigned long timeout,
564 const unsigned long max_rto)
566 struct inet_connection_sock *icsk = inet_csk(parent);
567 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
568 struct listen_sock *lopt = queue->listen_opt;
569 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
570 int thresh = max_retries;
571 unsigned long now = jiffies;
572 struct request_sock **reqp, *req;
575 if (lopt == NULL || lopt->qlen == 0)
578 /* Normally all the openreqs are young and become mature
579 * (i.e. converted to established socket) for first timeout.
580 * If synack was not acknowledged for 1 second, it means
581 * one of the following things: synack was lost, ack was lost,
582 * rtt is high or nobody planned to ack (i.e. synflood).
583 * When server is a bit loaded, queue is populated with old
584 * open requests, reducing effective size of queue.
585 * When server is well loaded, queue size reduces to zero
586 * after several minutes of work. It is not synflood,
587 * it is normal operation. The solution is pruning
588 * too old entries overriding normal timeout, when
589 * situation becomes dangerous.
591 * Essentially, we reserve half of room for young
592 * embrions; and abort old ones without pity, if old
593 * ones are about to clog our table.
595 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
596 int young = (lopt->qlen_young<<1);
599 if (lopt->qlen < young)
606 if (queue->rskq_defer_accept)
607 max_retries = queue->rskq_defer_accept;
609 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
610 i = lopt->clock_hand;
613 reqp=&lopt->syn_table[i];
614 while ((req = *reqp) != NULL) {
615 if (time_after_eq(now, req->expires)) {
616 int expire = 0, resend = 0;
618 syn_ack_recalc(req, thresh, max_retries,
619 queue->rskq_defer_accept,
621 req->rsk_ops->syn_ack_timeout(parent, req);
624 !inet_rtx_syn_ack(parent, req) ||
625 inet_rsk(req)->acked)) {
628 if (req->num_timeout++ == 0)
630 timeo = min(timeout << req->num_timeout,
632 req->expires = now + timeo;
633 reqp = &req->dl_next;
637 /* Drop this request */
638 inet_csk_reqsk_queue_unlink(parent, req, reqp);
639 reqsk_queue_removed(queue, req);
643 reqp = &req->dl_next;
646 i = (i + 1) & (lopt->nr_table_entries - 1);
648 } while (--budget > 0);
650 lopt->clock_hand = i;
653 inet_csk_reset_keepalive_timer(parent, interval);
655 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
658 * inet_csk_clone_lock - clone an inet socket, and lock its clone
659 * @sk: the socket to clone
661 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
663 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
665 struct sock *inet_csk_clone_lock(const struct sock *sk,
666 const struct request_sock *req,
667 const gfp_t priority)
669 struct sock *newsk = sk_clone_lock(sk, priority);
672 struct inet_connection_sock *newicsk = inet_csk(newsk);
674 newsk->sk_state = TCP_SYN_RECV;
675 newicsk->icsk_bind_hash = NULL;
677 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
678 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
679 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
680 newsk->sk_write_space = sk_stream_write_space;
682 newicsk->icsk_retransmits = 0;
683 newicsk->icsk_backoff = 0;
684 newicsk->icsk_probes_out = 0;
686 /* Deinitialize accept_queue to trap illegal accesses. */
687 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
689 security_inet_csk_clone(newsk, req);
693 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
696 * At this point, there should be no process reference to this
697 * socket, and thus no user references at all. Therefore we
698 * can assume the socket waitqueue is inactive and nobody will
699 * try to jump onto it.
701 void inet_csk_destroy_sock(struct sock *sk)
703 WARN_ON(sk->sk_state != TCP_CLOSE);
704 WARN_ON(!sock_flag(sk, SOCK_DEAD));
706 /* It cannot be in hash table! */
707 WARN_ON(!sk_unhashed(sk));
709 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
710 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
712 sk->sk_prot->destroy(sk);
714 sk_stream_kill_queues(sk);
716 xfrm_sk_free_policy(sk);
718 sk_refcnt_debug_release(sk);
720 percpu_counter_dec(sk->sk_prot->orphan_count);
723 EXPORT_SYMBOL(inet_csk_destroy_sock);
725 /* This function allows to force a closure of a socket after the call to
726 * tcp/dccp_create_openreq_child().
728 void inet_csk_prepare_forced_close(struct sock *sk)
729 __releases(&sk->sk_lock.slock)
731 /* sk_clone_lock locked the socket and set refcnt to 2 */
735 /* The below has to be done to allow calling inet_csk_destroy_sock */
736 sock_set_flag(sk, SOCK_DEAD);
737 percpu_counter_inc(sk->sk_prot->orphan_count);
738 inet_sk(sk)->inet_num = 0;
740 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
742 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
744 struct inet_sock *inet = inet_sk(sk);
745 struct inet_connection_sock *icsk = inet_csk(sk);
746 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
751 sk->sk_max_ack_backlog = 0;
752 sk->sk_ack_backlog = 0;
753 inet_csk_delack_init(sk);
755 /* There is race window here: we announce ourselves listening,
756 * but this transition is still not validated by get_port().
757 * It is OK, because this socket enters to hash table only
758 * after validation is complete.
760 sk->sk_state = TCP_LISTEN;
761 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
762 inet->inet_sport = htons(inet->inet_num);
765 sk->sk_prot->hash(sk);
770 sk->sk_state = TCP_CLOSE;
771 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
774 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
777 * This routine closes sockets which have been at least partially
778 * opened, but not yet accepted.
780 void inet_csk_listen_stop(struct sock *sk)
782 struct inet_connection_sock *icsk = inet_csk(sk);
783 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
784 struct request_sock *acc_req;
785 struct request_sock *req;
787 inet_csk_delete_keepalive_timer(sk);
789 /* make all the listen_opt local to us */
790 acc_req = reqsk_queue_yank_acceptq(queue);
792 /* Following specs, it would be better either to send FIN
793 * (and enter FIN-WAIT-1, it is normal close)
794 * or to send active reset (abort).
795 * Certainly, it is pretty dangerous while synflood, but it is
796 * bad justification for our negligence 8)
797 * To be honest, we are not able to make either
798 * of the variants now. --ANK
800 reqsk_queue_destroy(queue);
802 while ((req = acc_req) != NULL) {
803 struct sock *child = req->sk;
805 acc_req = req->dl_next;
809 WARN_ON(sock_owned_by_user(child));
812 sk->sk_prot->disconnect(child, O_NONBLOCK);
816 percpu_counter_inc(sk->sk_prot->orphan_count);
818 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->listener) {
819 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
820 BUG_ON(sk != tcp_rsk(req)->listener);
822 /* Paranoid, to prevent race condition if
823 * an inbound pkt destined for child is
824 * blocked by sock lock in tcp_v4_rcv().
825 * Also to satisfy an assertion in
826 * tcp_v4_destroy_sock().
828 tcp_sk(child)->fastopen_rsk = NULL;
831 inet_csk_destroy_sock(child);
833 bh_unlock_sock(child);
837 sk_acceptq_removed(sk);
840 if (queue->fastopenq != NULL) {
841 /* Free all the reqs queued in rskq_rst_head. */
842 spin_lock_bh(&queue->fastopenq->lock);
843 acc_req = queue->fastopenq->rskq_rst_head;
844 queue->fastopenq->rskq_rst_head = NULL;
845 spin_unlock_bh(&queue->fastopenq->lock);
846 while ((req = acc_req) != NULL) {
847 acc_req = req->dl_next;
851 WARN_ON(sk->sk_ack_backlog);
853 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
855 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
857 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
858 const struct inet_sock *inet = inet_sk(sk);
860 sin->sin_family = AF_INET;
861 sin->sin_addr.s_addr = inet->inet_daddr;
862 sin->sin_port = inet->inet_dport;
864 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
867 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
868 char __user *optval, int __user *optlen)
870 const struct inet_connection_sock *icsk = inet_csk(sk);
872 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
873 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
875 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
878 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
880 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
881 char __user *optval, unsigned int optlen)
883 const struct inet_connection_sock *icsk = inet_csk(sk);
885 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
886 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
888 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
891 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
894 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
896 const struct inet_sock *inet = inet_sk(sk);
897 const struct ip_options_rcu *inet_opt;
898 __be32 daddr = inet->inet_daddr;
903 inet_opt = rcu_dereference(inet->inet_opt);
904 if (inet_opt && inet_opt->opt.srr)
905 daddr = inet_opt->opt.faddr;
907 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
908 inet->inet_saddr, inet->inet_dport,
909 inet->inet_sport, sk->sk_protocol,
910 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
914 sk_setup_caps(sk, &rt->dst);
920 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
922 struct dst_entry *dst = __sk_dst_check(sk, 0);
923 struct inet_sock *inet = inet_sk(sk);
926 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
930 dst->ops->update_pmtu(dst, sk, NULL, mtu);
932 dst = __sk_dst_check(sk, 0);
934 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
938 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);