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);
33 * This struct holds the first and last local port number.
35 struct local_ports sysctl_local_ports __read_mostly = {
36 .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock),
37 .range = { 32768, 61000 },
40 unsigned long *sysctl_local_reserved_ports;
41 EXPORT_SYMBOL(sysctl_local_reserved_ports);
43 void inet_get_local_port_range(int *low, int *high)
48 seq = read_seqbegin(&sysctl_local_ports.lock);
50 *low = sysctl_local_ports.range[0];
51 *high = sysctl_local_ports.range[1];
52 } while (read_seqretry(&sysctl_local_ports.lock, seq));
54 EXPORT_SYMBOL(inet_get_local_port_range);
56 int inet_csk_bind_conflict(const struct sock *sk,
57 const struct inet_bind_bucket *tb, bool relax)
60 struct hlist_node *node;
61 int reuse = sk->sk_reuse;
64 * Unlike other sk lookup places we do not check
65 * for sk_net here, since _all_ the socks listed
66 * in tb->owners list belong to the same net - the
67 * one this bucket belongs to.
70 sk_for_each_bound(sk2, node, &tb->owners) {
72 !inet_v6_ipv6only(sk2) &&
73 (!sk->sk_bound_dev_if ||
74 !sk2->sk_bound_dev_if ||
75 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
76 if (!reuse || !sk2->sk_reuse ||
77 sk2->sk_state == TCP_LISTEN) {
78 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
79 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
80 sk2_rcv_saddr == sk_rcv_saddr(sk))
83 if (!relax && reuse && sk2->sk_reuse &&
84 sk2->sk_state != TCP_LISTEN) {
85 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
87 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
88 sk2_rcv_saddr == sk_rcv_saddr(sk))
95 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
97 /* Obtain a reference to a local port for the given sock,
98 * if snum is zero it means select any available local port.
100 int inet_csk_get_port(struct sock *sk, unsigned short snum)
102 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
103 struct inet_bind_hashbucket *head;
104 struct hlist_node *node;
105 struct inet_bind_bucket *tb;
106 int ret, attempts = 5;
107 struct net *net = sock_net(sk);
108 int smallest_size = -1, smallest_rover;
112 int remaining, rover, low, high;
115 inet_get_local_port_range(&low, &high);
116 remaining = (high - low) + 1;
117 smallest_rover = rover = net_random() % remaining + low;
121 if (inet_is_reserved_local_port(rover))
123 head = &hashinfo->bhash[inet_bhashfn(net, rover,
124 hashinfo->bhash_size)];
125 spin_lock(&head->lock);
126 inet_bind_bucket_for_each(tb, node, &head->chain)
127 if (net_eq(ib_net(tb), net) && tb->port == rover) {
128 if (tb->fastreuse > 0 &&
130 sk->sk_state != TCP_LISTEN &&
131 (tb->num_owners < smallest_size || smallest_size == -1)) {
132 smallest_size = tb->num_owners;
133 smallest_rover = rover;
134 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
135 !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
136 snum = smallest_rover;
140 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
148 spin_unlock(&head->lock);
152 } while (--remaining > 0);
154 /* Exhausted local port range during search? It is not
155 * possible for us to be holding one of the bind hash
156 * locks if this test triggers, because if 'remaining'
157 * drops to zero, we broke out of the do/while loop at
158 * the top level, not from the 'break;' statement.
161 if (remaining <= 0) {
162 if (smallest_size != -1) {
163 snum = smallest_rover;
168 /* OK, here is the one we will use. HEAD is
169 * non-NULL and we hold it's mutex.
174 head = &hashinfo->bhash[inet_bhashfn(net, snum,
175 hashinfo->bhash_size)];
176 spin_lock(&head->lock);
177 inet_bind_bucket_for_each(tb, node, &head->chain)
178 if (net_eq(ib_net(tb), net) && tb->port == snum)
184 if (!hlist_empty(&tb->owners)) {
185 if (sk->sk_reuse == SK_FORCE_REUSE)
188 if (tb->fastreuse > 0 &&
189 sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
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 smallest_size != -1 && --attempts >= 0) {
197 spin_unlock(&head->lock);
207 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
208 net, head, snum)) == NULL)
210 if (hlist_empty(&tb->owners)) {
211 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
215 } else if (tb->fastreuse &&
216 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
219 if (!inet_csk(sk)->icsk_bind_hash)
220 inet_bind_hash(sk, tb, snum);
221 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
225 spin_unlock(&head->lock);
230 EXPORT_SYMBOL_GPL(inet_csk_get_port);
233 * Wait for an incoming connection, avoid race conditions. This must be called
234 * with the socket locked.
236 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
238 struct inet_connection_sock *icsk = inet_csk(sk);
243 * True wake-one mechanism for incoming connections: only
244 * one process gets woken up, not the 'whole herd'.
245 * Since we do not 'race & poll' for established sockets
246 * anymore, the common case will execute the loop only once.
248 * Subtle issue: "add_wait_queue_exclusive()" will be added
249 * after any current non-exclusive waiters, and we know that
250 * it will always _stay_ after any new non-exclusive waiters
251 * because all non-exclusive waiters are added at the
252 * beginning of the wait-queue. As such, it's ok to "drop"
253 * our exclusiveness temporarily when we get woken up without
254 * having to remove and re-insert us on the wait queue.
257 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
260 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
261 timeo = schedule_timeout(timeo);
264 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
267 if (sk->sk_state != TCP_LISTEN)
269 err = sock_intr_errno(timeo);
270 if (signal_pending(current))
276 finish_wait(sk_sleep(sk), &wait);
281 * This will accept the next outstanding connection.
283 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
285 struct inet_connection_sock *icsk = inet_csk(sk);
291 /* We need to make sure that this socket is listening,
292 * and that it has something pending.
295 if (sk->sk_state != TCP_LISTEN)
298 /* Find already established connection */
299 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
300 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
302 /* If this is a non blocking socket don't sleep */
307 error = inet_csk_wait_for_connect(sk, timeo);
312 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
313 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
322 EXPORT_SYMBOL(inet_csk_accept);
325 * Using different timers for retransmit, delayed acks and probes
326 * We may wish use just one timer maintaining a list of expire jiffies
329 void inet_csk_init_xmit_timers(struct sock *sk,
330 void (*retransmit_handler)(unsigned long),
331 void (*delack_handler)(unsigned long),
332 void (*keepalive_handler)(unsigned long))
334 struct inet_connection_sock *icsk = inet_csk(sk);
336 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
338 setup_timer(&icsk->icsk_delack_timer, delack_handler,
340 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
341 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
343 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
345 void inet_csk_clear_xmit_timers(struct sock *sk)
347 struct inet_connection_sock *icsk = inet_csk(sk);
349 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
351 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
352 sk_stop_timer(sk, &icsk->icsk_delack_timer);
353 sk_stop_timer(sk, &sk->sk_timer);
355 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
357 void inet_csk_delete_keepalive_timer(struct sock *sk)
359 sk_stop_timer(sk, &sk->sk_timer);
361 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
363 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
365 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
367 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
369 struct dst_entry *inet_csk_route_req(struct sock *sk,
371 const struct request_sock *req,
375 const struct inet_request_sock *ireq = inet_rsk(req);
376 struct ip_options_rcu *opt = inet_rsk(req)->opt;
377 struct net *net = sock_net(sk);
378 int flags = inet_sk_flowi_flags(sk);
381 flags |= FLOWI_FLAG_RT_NOCACHE;
382 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
383 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
386 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
387 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
388 security_req_classify_flow(req, flowi4_to_flowi(fl4));
389 rt = ip_route_output_flow(net, fl4, sk);
392 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
399 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
402 EXPORT_SYMBOL_GPL(inet_csk_route_req);
404 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
406 const struct request_sock *req)
408 const struct inet_request_sock *ireq = inet_rsk(req);
409 struct inet_sock *newinet = inet_sk(newsk);
410 struct ip_options_rcu *opt = ireq->opt;
411 struct net *net = sock_net(sk);
415 fl4 = &newinet->cork.fl.u.ip4;
416 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
417 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
418 sk->sk_protocol, inet_sk_flowi_flags(sk),
419 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
420 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
421 security_req_classify_flow(req, flowi4_to_flowi(fl4));
422 rt = ip_route_output_flow(net, fl4, sk);
425 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
432 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
435 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
437 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
438 const u32 rnd, const u32 synq_hsize)
440 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
443 #if IS_ENABLED(CONFIG_IPV6)
444 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
446 #define AF_INET_FAMILY(fam) 1
449 struct request_sock *inet_csk_search_req(const struct sock *sk,
450 struct request_sock ***prevp,
451 const __be16 rport, const __be32 raddr,
454 const struct inet_connection_sock *icsk = inet_csk(sk);
455 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
456 struct request_sock *req, **prev;
458 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
459 lopt->nr_table_entries)];
460 (req = *prev) != NULL;
461 prev = &req->dl_next) {
462 const struct inet_request_sock *ireq = inet_rsk(req);
464 if (ireq->rmt_port == rport &&
465 ireq->rmt_addr == raddr &&
466 ireq->loc_addr == laddr &&
467 AF_INET_FAMILY(req->rsk_ops->family)) {
476 EXPORT_SYMBOL_GPL(inet_csk_search_req);
478 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
479 unsigned long timeout)
481 struct inet_connection_sock *icsk = inet_csk(sk);
482 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
483 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
484 lopt->hash_rnd, lopt->nr_table_entries);
486 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
487 inet_csk_reqsk_queue_added(sk, timeout);
489 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
491 /* Only thing we need from tcp.h */
492 extern int sysctl_tcp_synack_retries;
495 /* Decide when to expire the request and when to resend SYN-ACK */
496 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
497 const int max_retries,
498 const u8 rskq_defer_accept,
499 int *expire, int *resend)
501 if (!rskq_defer_accept) {
502 *expire = req->retrans >= thresh;
506 *expire = req->retrans >= thresh &&
507 (!inet_rsk(req)->acked || req->retrans >= max_retries);
509 * Do not resend while waiting for data after ACK,
510 * start to resend on end of deferring period to give
511 * last chance for data or ACK to create established socket.
513 *resend = !inet_rsk(req)->acked ||
514 req->retrans >= rskq_defer_accept - 1;
517 void inet_csk_reqsk_queue_prune(struct sock *parent,
518 const unsigned long interval,
519 const unsigned long timeout,
520 const unsigned long max_rto)
522 struct inet_connection_sock *icsk = inet_csk(parent);
523 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
524 struct listen_sock *lopt = queue->listen_opt;
525 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
526 int thresh = max_retries;
527 unsigned long now = jiffies;
528 struct request_sock **reqp, *req;
531 if (lopt == NULL || lopt->qlen == 0)
534 /* Normally all the openreqs are young and become mature
535 * (i.e. converted to established socket) for first timeout.
536 * If synack was not acknowledged for 1 second, it means
537 * one of the following things: synack was lost, ack was lost,
538 * rtt is high or nobody planned to ack (i.e. synflood).
539 * When server is a bit loaded, queue is populated with old
540 * open requests, reducing effective size of queue.
541 * When server is well loaded, queue size reduces to zero
542 * after several minutes of work. It is not synflood,
543 * it is normal operation. The solution is pruning
544 * too old entries overriding normal timeout, when
545 * situation becomes dangerous.
547 * Essentially, we reserve half of room for young
548 * embrions; and abort old ones without pity, if old
549 * ones are about to clog our table.
551 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
552 int young = (lopt->qlen_young<<1);
555 if (lopt->qlen < young)
562 if (queue->rskq_defer_accept)
563 max_retries = queue->rskq_defer_accept;
565 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
566 i = lopt->clock_hand;
569 reqp=&lopt->syn_table[i];
570 while ((req = *reqp) != NULL) {
571 if (time_after_eq(now, req->expires)) {
572 int expire = 0, resend = 0;
574 syn_ack_recalc(req, thresh, max_retries,
575 queue->rskq_defer_accept,
577 req->rsk_ops->syn_ack_timeout(parent, req);
580 !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
581 inet_rsk(req)->acked)) {
584 if (req->retrans++ == 0)
586 timeo = min((timeout << req->retrans), max_rto);
587 req->expires = now + timeo;
588 reqp = &req->dl_next;
592 /* Drop this request */
593 inet_csk_reqsk_queue_unlink(parent, req, reqp);
594 reqsk_queue_removed(queue, req);
598 reqp = &req->dl_next;
601 i = (i + 1) & (lopt->nr_table_entries - 1);
603 } while (--budget > 0);
605 lopt->clock_hand = i;
608 inet_csk_reset_keepalive_timer(parent, interval);
610 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
613 * inet_csk_clone_lock - clone an inet socket, and lock its clone
614 * @sk: the socket to clone
616 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
618 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
620 struct sock *inet_csk_clone_lock(const struct sock *sk,
621 const struct request_sock *req,
622 const gfp_t priority)
624 struct sock *newsk = sk_clone_lock(sk, priority);
627 struct inet_connection_sock *newicsk = inet_csk(newsk);
629 newsk->sk_state = TCP_SYN_RECV;
630 newicsk->icsk_bind_hash = NULL;
632 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
633 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
634 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
635 newsk->sk_write_space = sk_stream_write_space;
637 newicsk->icsk_retransmits = 0;
638 newicsk->icsk_backoff = 0;
639 newicsk->icsk_probes_out = 0;
641 /* Deinitialize accept_queue to trap illegal accesses. */
642 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
644 security_inet_csk_clone(newsk, req);
648 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
651 * At this point, there should be no process reference to this
652 * socket, and thus no user references at all. Therefore we
653 * can assume the socket waitqueue is inactive and nobody will
654 * try to jump onto it.
656 void inet_csk_destroy_sock(struct sock *sk)
658 WARN_ON(sk->sk_state != TCP_CLOSE);
659 WARN_ON(!sock_flag(sk, SOCK_DEAD));
661 /* It cannot be in hash table! */
662 WARN_ON(!sk_unhashed(sk));
664 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
665 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
667 sk->sk_prot->destroy(sk);
669 sk_stream_kill_queues(sk);
671 xfrm_sk_free_policy(sk);
673 sk_refcnt_debug_release(sk);
675 percpu_counter_dec(sk->sk_prot->orphan_count);
678 EXPORT_SYMBOL(inet_csk_destroy_sock);
680 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
682 struct inet_sock *inet = inet_sk(sk);
683 struct inet_connection_sock *icsk = inet_csk(sk);
684 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
689 sk->sk_max_ack_backlog = 0;
690 sk->sk_ack_backlog = 0;
691 inet_csk_delack_init(sk);
693 /* There is race window here: we announce ourselves listening,
694 * but this transition is still not validated by get_port().
695 * It is OK, because this socket enters to hash table only
696 * after validation is complete.
698 sk->sk_state = TCP_LISTEN;
699 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
700 inet->inet_sport = htons(inet->inet_num);
703 sk->sk_prot->hash(sk);
708 sk->sk_state = TCP_CLOSE;
709 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
712 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
715 * This routine closes sockets which have been at least partially
716 * opened, but not yet accepted.
718 void inet_csk_listen_stop(struct sock *sk)
720 struct inet_connection_sock *icsk = inet_csk(sk);
721 struct request_sock *acc_req;
722 struct request_sock *req;
724 inet_csk_delete_keepalive_timer(sk);
726 /* make all the listen_opt local to us */
727 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
729 /* Following specs, it would be better either to send FIN
730 * (and enter FIN-WAIT-1, it is normal close)
731 * or to send active reset (abort).
732 * Certainly, it is pretty dangerous while synflood, but it is
733 * bad justification for our negligence 8)
734 * To be honest, we are not able to make either
735 * of the variants now. --ANK
737 reqsk_queue_destroy(&icsk->icsk_accept_queue);
739 while ((req = acc_req) != NULL) {
740 struct sock *child = req->sk;
742 acc_req = req->dl_next;
746 WARN_ON(sock_owned_by_user(child));
749 sk->sk_prot->disconnect(child, O_NONBLOCK);
753 percpu_counter_inc(sk->sk_prot->orphan_count);
755 inet_csk_destroy_sock(child);
757 bh_unlock_sock(child);
761 sk_acceptq_removed(sk);
764 WARN_ON(sk->sk_ack_backlog);
766 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
768 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
770 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
771 const struct inet_sock *inet = inet_sk(sk);
773 sin->sin_family = AF_INET;
774 sin->sin_addr.s_addr = inet->inet_daddr;
775 sin->sin_port = inet->inet_dport;
777 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
780 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
781 char __user *optval, int __user *optlen)
783 const struct inet_connection_sock *icsk = inet_csk(sk);
785 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
786 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
788 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
791 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
793 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
794 char __user *optval, unsigned int optlen)
796 const struct inet_connection_sock *icsk = inet_csk(sk);
798 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
799 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
801 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
804 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
807 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
809 const struct inet_sock *inet = inet_sk(sk);
810 const struct ip_options_rcu *inet_opt;
811 __be32 daddr = inet->inet_daddr;
816 inet_opt = rcu_dereference(inet->inet_opt);
817 if (inet_opt && inet_opt->opt.srr)
818 daddr = inet_opt->opt.faddr;
820 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
821 inet->inet_saddr, inet->inet_dport,
822 inet->inet_sport, sk->sk_protocol,
823 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
827 sk_setup_caps(sk, &rt->dst);
833 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
835 struct dst_entry *dst = __sk_dst_check(sk, 0);
836 struct inet_sock *inet = inet_sk(sk);
839 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
843 dst->ops->update_pmtu(dst, sk, NULL, mtu);
845 dst = __sk_dst_check(sk, 0);
847 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
851 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);