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>
27 #include <net/sock_reuseport.h>
28 #include <net/addrconf.h>
30 #if IS_ENABLED(CONFIG_IPV6)
31 /* match_sk*_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses
32 * if IPv6 only, and any IPv4 addresses
34 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
35 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
36 * and 0.0.0.0 equals to 0.0.0.0 only
38 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
39 const struct in6_addr *sk2_rcv_saddr6,
40 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
41 bool sk1_ipv6only, bool sk2_ipv6only,
42 bool match_sk1_wildcard,
43 bool match_sk2_wildcard)
45 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
46 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
48 /* if both are mapped, treat as IPv4 */
49 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
51 if (sk1_rcv_saddr == sk2_rcv_saddr)
53 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
54 (match_sk2_wildcard && !sk2_rcv_saddr);
59 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
62 if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard &&
63 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
66 if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard &&
67 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
71 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
78 /* match_sk*_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
79 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
80 * 0.0.0.0 only equals to 0.0.0.0
82 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
83 bool sk2_ipv6only, bool match_sk1_wildcard,
84 bool match_sk2_wildcard)
87 if (sk1_rcv_saddr == sk2_rcv_saddr)
89 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
90 (match_sk2_wildcard && !sk2_rcv_saddr);
95 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
98 #if IS_ENABLED(CONFIG_IPV6)
99 if (sk->sk_family == AF_INET6)
100 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
101 inet6_rcv_saddr(sk2),
109 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
110 ipv6_only_sock(sk2), match_wildcard,
113 EXPORT_SYMBOL(inet_rcv_saddr_equal);
115 bool inet_rcv_saddr_any(const struct sock *sk)
117 #if IS_ENABLED(CONFIG_IPV6)
118 if (sk->sk_family == AF_INET6)
119 return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
121 return !sk->sk_rcv_saddr;
124 void inet_get_local_port_range(struct net *net, int *low, int *high)
129 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
131 *low = net->ipv4.ip_local_ports.range[0];
132 *high = net->ipv4.ip_local_ports.range[1];
133 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
135 EXPORT_SYMBOL(inet_get_local_port_range);
137 static int inet_csk_bind_conflict(const struct sock *sk,
138 const struct inet_bind_bucket *tb,
139 bool relax, bool reuseport_ok)
142 bool reuse = sk->sk_reuse;
143 bool reuseport = !!sk->sk_reuseport && reuseport_ok;
144 kuid_t uid = sock_i_uid((struct sock *)sk);
147 * Unlike other sk lookup places we do not check
148 * for sk_net here, since _all_ the socks listed
149 * in tb->owners list belong to the same net - the
150 * one this bucket belongs to.
153 sk_for_each_bound(sk2, &tb->owners) {
155 (!sk->sk_bound_dev_if ||
156 !sk2->sk_bound_dev_if ||
157 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
158 if ((!reuse || !sk2->sk_reuse ||
159 sk2->sk_state == TCP_LISTEN) &&
160 (!reuseport || !sk2->sk_reuseport ||
161 rcu_access_pointer(sk->sk_reuseport_cb) ||
162 (sk2->sk_state != TCP_TIME_WAIT &&
163 !uid_eq(uid, sock_i_uid(sk2))))) {
164 if (inet_rcv_saddr_equal(sk, sk2, true))
167 if (!relax && reuse && sk2->sk_reuse &&
168 sk2->sk_state != TCP_LISTEN) {
169 if (inet_rcv_saddr_equal(sk, sk2, true))
178 * Find an open port number for the socket. Returns with the
179 * inet_bind_hashbucket lock held.
181 static struct inet_bind_hashbucket *
182 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
184 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
186 struct inet_bind_hashbucket *head;
187 struct net *net = sock_net(sk);
188 int i, low, high, attempt_half;
189 struct inet_bind_bucket *tb;
190 u32 remaining, offset;
192 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
194 inet_get_local_port_range(net, &low, &high);
195 high++; /* [32768, 60999] -> [32768, 61000[ */
199 int half = low + (((high - low) >> 2) << 1);
201 if (attempt_half == 1)
206 remaining = high - low;
207 if (likely(remaining > 1))
210 offset = prandom_u32() % remaining;
211 /* __inet_hash_connect() favors ports having @low parity
212 * We do the opposite to not pollute connect() users.
218 for (i = 0; i < remaining; i += 2, port += 2) {
219 if (unlikely(port >= high))
221 if (inet_is_local_reserved_port(net, port))
223 head = &hinfo->bhash[inet_bhashfn(net, port,
225 spin_lock_bh(&head->lock);
226 inet_bind_bucket_for_each(tb, &head->chain)
227 if (net_eq(ib_net(tb), net) && tb->port == port) {
228 if (!inet_csk_bind_conflict(sk, tb, false, false))
235 spin_unlock_bh(&head->lock);
241 goto other_parity_scan;
243 if (attempt_half == 1) {
244 /* OK we now try the upper half of the range */
246 goto other_half_scan;
255 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
258 kuid_t uid = sock_i_uid(sk);
260 if (tb->fastreuseport <= 0)
262 if (!sk->sk_reuseport)
264 if (rcu_access_pointer(sk->sk_reuseport_cb))
266 if (!uid_eq(tb->fastuid, uid))
268 /* We only need to check the rcv_saddr if this tb was once marked
269 * without fastreuseport and then was reset, as we can only know that
270 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
273 if (tb->fastreuseport == FASTREUSEPORT_ANY)
275 #if IS_ENABLED(CONFIG_IPV6)
276 if (tb->fast_sk_family == AF_INET6)
277 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
282 ipv6_only_sock(sk), true, false);
284 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
285 ipv6_only_sock(sk), true, false);
288 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
291 kuid_t uid = sock_i_uid(sk);
292 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
294 if (hlist_empty(&tb->owners)) {
295 tb->fastreuse = reuse;
296 if (sk->sk_reuseport) {
297 tb->fastreuseport = FASTREUSEPORT_ANY;
299 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
300 tb->fast_ipv6_only = ipv6_only_sock(sk);
301 tb->fast_sk_family = sk->sk_family;
302 #if IS_ENABLED(CONFIG_IPV6)
303 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
306 tb->fastreuseport = 0;
311 if (sk->sk_reuseport) {
312 /* We didn't match or we don't have fastreuseport set on
313 * the tb, but we have sk_reuseport set on this socket
314 * and we know that there are no bind conflicts with
315 * this socket in this tb, so reset our tb's reuseport
316 * settings so that any subsequent sockets that match
317 * our current socket will be put on the fast path.
319 * If we reset we need to set FASTREUSEPORT_STRICT so we
320 * do extra checking for all subsequent sk_reuseport
323 if (!sk_reuseport_match(tb, sk)) {
324 tb->fastreuseport = FASTREUSEPORT_STRICT;
326 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
327 tb->fast_ipv6_only = ipv6_only_sock(sk);
328 tb->fast_sk_family = sk->sk_family;
329 #if IS_ENABLED(CONFIG_IPV6)
330 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
334 tb->fastreuseport = 0;
339 /* Obtain a reference to a local port for the given sock,
340 * if snum is zero it means select any available local port.
341 * We try to allocate an odd port (and leave even ports for connect())
343 int inet_csk_get_port(struct sock *sk, unsigned short snum)
345 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
346 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
347 int ret = 1, port = snum;
348 struct inet_bind_hashbucket *head;
349 struct net *net = sock_net(sk);
350 struct inet_bind_bucket *tb = NULL;
353 head = inet_csk_find_open_port(sk, &tb, &port);
360 head = &hinfo->bhash[inet_bhashfn(net, port,
362 spin_lock_bh(&head->lock);
363 inet_bind_bucket_for_each(tb, &head->chain)
364 if (net_eq(ib_net(tb), net) && tb->port == port)
367 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
372 if (!hlist_empty(&tb->owners)) {
373 if (sk->sk_reuse == SK_FORCE_REUSE)
376 if ((tb->fastreuse > 0 && reuse) ||
377 sk_reuseport_match(tb, sk))
379 if (inet_csk_bind_conflict(sk, tb, true, true))
383 inet_csk_update_fastreuse(tb, sk);
385 if (!inet_csk(sk)->icsk_bind_hash)
386 inet_bind_hash(sk, tb, port);
387 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
391 spin_unlock_bh(&head->lock);
394 EXPORT_SYMBOL_GPL(inet_csk_get_port);
397 * Wait for an incoming connection, avoid race conditions. This must be called
398 * with the socket locked.
400 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
402 struct inet_connection_sock *icsk = inet_csk(sk);
407 * True wake-one mechanism for incoming connections: only
408 * one process gets woken up, not the 'whole herd'.
409 * Since we do not 'race & poll' for established sockets
410 * anymore, the common case will execute the loop only once.
412 * Subtle issue: "add_wait_queue_exclusive()" will be added
413 * after any current non-exclusive waiters, and we know that
414 * it will always _stay_ after any new non-exclusive waiters
415 * because all non-exclusive waiters are added at the
416 * beginning of the wait-queue. As such, it's ok to "drop"
417 * our exclusiveness temporarily when we get woken up without
418 * having to remove and re-insert us on the wait queue.
421 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
424 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
425 timeo = schedule_timeout(timeo);
426 sched_annotate_sleep();
429 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
432 if (sk->sk_state != TCP_LISTEN)
434 err = sock_intr_errno(timeo);
435 if (signal_pending(current))
441 finish_wait(sk_sleep(sk), &wait);
446 * This will accept the next outstanding connection.
448 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
450 struct inet_connection_sock *icsk = inet_csk(sk);
451 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
452 struct request_sock *req;
458 /* We need to make sure that this socket is listening,
459 * and that it has something pending.
462 if (sk->sk_state != TCP_LISTEN)
465 /* Find already established connection */
466 if (reqsk_queue_empty(queue)) {
467 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
469 /* If this is a non blocking socket don't sleep */
474 error = inet_csk_wait_for_connect(sk, timeo);
478 req = reqsk_queue_remove(queue, sk);
481 if (sk->sk_protocol == IPPROTO_TCP &&
482 tcp_rsk(req)->tfo_listener) {
483 spin_lock_bh(&queue->fastopenq.lock);
484 if (tcp_rsk(req)->tfo_listener) {
485 /* We are still waiting for the final ACK from 3WHS
486 * so can't free req now. Instead, we set req->sk to
487 * NULL to signify that the child socket is taken
488 * so reqsk_fastopen_remove() will free the req
489 * when 3WHS finishes (or is aborted).
494 spin_unlock_bh(&queue->fastopenq.lock);
499 if (newsk && mem_cgroup_sockets_enabled) {
502 /* atomically get the memory usage, set and charge the
507 /* The socket has not been accepted yet, no need to look at
508 * newsk->sk_wmem_queued.
510 amt = sk_mem_pages(newsk->sk_forward_alloc +
511 atomic_read(&newsk->sk_rmem_alloc));
512 mem_cgroup_sk_alloc(newsk);
513 if (newsk->sk_memcg && amt)
514 mem_cgroup_charge_skmem(newsk->sk_memcg, amt);
527 EXPORT_SYMBOL(inet_csk_accept);
530 * Using different timers for retransmit, delayed acks and probes
531 * We may wish use just one timer maintaining a list of expire jiffies
534 void inet_csk_init_xmit_timers(struct sock *sk,
535 void (*retransmit_handler)(struct timer_list *t),
536 void (*delack_handler)(struct timer_list *t),
537 void (*keepalive_handler)(struct timer_list *t))
539 struct inet_connection_sock *icsk = inet_csk(sk);
541 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
542 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
543 timer_setup(&sk->sk_timer, keepalive_handler, 0);
544 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
546 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
548 void inet_csk_clear_xmit_timers(struct sock *sk)
550 struct inet_connection_sock *icsk = inet_csk(sk);
552 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
554 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
555 sk_stop_timer(sk, &icsk->icsk_delack_timer);
556 sk_stop_timer(sk, &sk->sk_timer);
558 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
560 void inet_csk_delete_keepalive_timer(struct sock *sk)
562 sk_stop_timer(sk, &sk->sk_timer);
564 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
566 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
568 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
570 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
572 struct dst_entry *inet_csk_route_req(const struct sock *sk,
574 const struct request_sock *req)
576 const struct inet_request_sock *ireq = inet_rsk(req);
577 struct net *net = read_pnet(&ireq->ireq_net);
578 struct ip_options_rcu *opt;
582 opt = rcu_dereference(ireq->ireq_opt);
584 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
585 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
586 sk->sk_protocol, inet_sk_flowi_flags(sk),
587 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
588 ireq->ir_loc_addr, ireq->ir_rmt_port,
589 htons(ireq->ir_num), sk->sk_uid);
590 security_req_classify_flow(req, flowi4_to_flowi(fl4));
591 rt = ip_route_output_flow(net, fl4, sk);
594 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
603 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
606 EXPORT_SYMBOL_GPL(inet_csk_route_req);
608 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
610 const struct request_sock *req)
612 const struct inet_request_sock *ireq = inet_rsk(req);
613 struct net *net = read_pnet(&ireq->ireq_net);
614 struct inet_sock *newinet = inet_sk(newsk);
615 struct ip_options_rcu *opt;
619 opt = rcu_dereference(ireq->ireq_opt);
620 fl4 = &newinet->cork.fl.u.ip4;
622 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
623 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
624 sk->sk_protocol, inet_sk_flowi_flags(sk),
625 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
626 ireq->ir_loc_addr, ireq->ir_rmt_port,
627 htons(ireq->ir_num), sk->sk_uid);
628 security_req_classify_flow(req, flowi4_to_flowi(fl4));
629 rt = ip_route_output_flow(net, fl4, sk);
632 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
639 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
642 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
644 #if IS_ENABLED(CONFIG_IPV6)
645 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
647 #define AF_INET_FAMILY(fam) true
650 /* Decide when to expire the request and when to resend SYN-ACK */
651 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
652 const int max_retries,
653 const u8 rskq_defer_accept,
654 int *expire, int *resend)
656 if (!rskq_defer_accept) {
657 *expire = req->num_timeout >= thresh;
661 *expire = req->num_timeout >= thresh &&
662 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
664 * Do not resend while waiting for data after ACK,
665 * start to resend on end of deferring period to give
666 * last chance for data or ACK to create established socket.
668 *resend = !inet_rsk(req)->acked ||
669 req->num_timeout >= rskq_defer_accept - 1;
672 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
674 int err = req->rsk_ops->rtx_syn_ack(parent, req);
680 EXPORT_SYMBOL(inet_rtx_syn_ack);
682 /* return true if req was found in the ehash table */
683 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
684 struct request_sock *req)
686 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
689 if (sk_hashed(req_to_sk(req))) {
690 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
693 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
696 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
701 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
703 bool unlinked = reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req);
706 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
711 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
713 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
715 inet_csk_reqsk_queue_drop(sk, req);
718 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
720 static void reqsk_timer_handler(struct timer_list *t)
722 struct request_sock *req = from_timer(req, t, rsk_timer);
723 struct sock *sk_listener = req->rsk_listener;
724 struct net *net = sock_net(sk_listener);
725 struct inet_connection_sock *icsk = inet_csk(sk_listener);
726 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
727 int qlen, expire = 0, resend = 0;
728 int max_retries, thresh;
731 if (inet_sk_state_load(sk_listener) != TCP_LISTEN)
734 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
735 thresh = max_retries;
736 /* Normally all the openreqs are young and become mature
737 * (i.e. converted to established socket) for first timeout.
738 * If synack was not acknowledged for 1 second, it means
739 * one of the following things: synack was lost, ack was lost,
740 * rtt is high or nobody planned to ack (i.e. synflood).
741 * When server is a bit loaded, queue is populated with old
742 * open requests, reducing effective size of queue.
743 * When server is well loaded, queue size reduces to zero
744 * after several minutes of work. It is not synflood,
745 * it is normal operation. The solution is pruning
746 * too old entries overriding normal timeout, when
747 * situation becomes dangerous.
749 * Essentially, we reserve half of room for young
750 * embrions; and abort old ones without pity, if old
751 * ones are about to clog our table.
753 qlen = reqsk_queue_len(queue);
754 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
755 int young = reqsk_queue_len_young(queue) << 1;
764 defer_accept = READ_ONCE(queue->rskq_defer_accept);
766 max_retries = defer_accept;
767 syn_ack_recalc(req, thresh, max_retries, defer_accept,
769 req->rsk_ops->syn_ack_timeout(req);
772 !inet_rtx_syn_ack(sk_listener, req) ||
773 inet_rsk(req)->acked)) {
776 if (req->num_timeout++ == 0)
777 atomic_dec(&queue->young);
778 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
779 mod_timer(&req->rsk_timer, jiffies + timeo);
783 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
786 static void reqsk_queue_hash_req(struct request_sock *req,
787 unsigned long timeout)
789 req->num_retrans = 0;
790 req->num_timeout = 0;
793 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
794 mod_timer(&req->rsk_timer, jiffies + timeout);
796 inet_ehash_insert(req_to_sk(req), NULL);
797 /* before letting lookups find us, make sure all req fields
798 * are committed to memory and refcnt initialized.
801 refcount_set(&req->rsk_refcnt, 2 + 1);
804 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
805 unsigned long timeout)
807 reqsk_queue_hash_req(req, timeout);
808 inet_csk_reqsk_queue_added(sk);
810 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
813 * inet_csk_clone_lock - clone an inet socket, and lock its clone
814 * @sk: the socket to clone
816 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
818 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
820 struct sock *inet_csk_clone_lock(const struct sock *sk,
821 const struct request_sock *req,
822 const gfp_t priority)
824 struct sock *newsk = sk_clone_lock(sk, priority);
827 struct inet_connection_sock *newicsk = inet_csk(newsk);
829 inet_sk_set_state(newsk, TCP_SYN_RECV);
830 newicsk->icsk_bind_hash = NULL;
832 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
833 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
834 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
836 /* listeners have SOCK_RCU_FREE, not the children */
837 sock_reset_flag(newsk, SOCK_RCU_FREE);
839 inet_sk(newsk)->mc_list = NULL;
841 newsk->sk_mark = inet_rsk(req)->ir_mark;
842 atomic64_set(&newsk->sk_cookie,
843 atomic64_read(&inet_rsk(req)->ir_cookie));
845 newicsk->icsk_retransmits = 0;
846 newicsk->icsk_backoff = 0;
847 newicsk->icsk_probes_out = 0;
849 /* Deinitialize accept_queue to trap illegal accesses. */
850 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
852 security_inet_csk_clone(newsk, req);
856 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
859 * At this point, there should be no process reference to this
860 * socket, and thus no user references at all. Therefore we
861 * can assume the socket waitqueue is inactive and nobody will
862 * try to jump onto it.
864 void inet_csk_destroy_sock(struct sock *sk)
866 WARN_ON(sk->sk_state != TCP_CLOSE);
867 WARN_ON(!sock_flag(sk, SOCK_DEAD));
869 /* It cannot be in hash table! */
870 WARN_ON(!sk_unhashed(sk));
872 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
873 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
875 sk->sk_prot->destroy(sk);
877 sk_stream_kill_queues(sk);
879 xfrm_sk_free_policy(sk);
881 sk_refcnt_debug_release(sk);
883 percpu_counter_dec(sk->sk_prot->orphan_count);
887 EXPORT_SYMBOL(inet_csk_destroy_sock);
889 /* This function allows to force a closure of a socket after the call to
890 * tcp/dccp_create_openreq_child().
892 void inet_csk_prepare_forced_close(struct sock *sk)
893 __releases(&sk->sk_lock.slock)
895 /* sk_clone_lock locked the socket and set refcnt to 2 */
899 /* The below has to be done to allow calling inet_csk_destroy_sock */
900 sock_set_flag(sk, SOCK_DEAD);
901 percpu_counter_inc(sk->sk_prot->orphan_count);
902 inet_sk(sk)->inet_num = 0;
904 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
906 int inet_csk_listen_start(struct sock *sk, int backlog)
908 struct inet_connection_sock *icsk = inet_csk(sk);
909 struct inet_sock *inet = inet_sk(sk);
910 int err = -EADDRINUSE;
912 reqsk_queue_alloc(&icsk->icsk_accept_queue);
914 sk->sk_max_ack_backlog = backlog;
915 sk->sk_ack_backlog = 0;
916 inet_csk_delack_init(sk);
918 /* There is race window here: we announce ourselves listening,
919 * but this transition is still not validated by get_port().
920 * It is OK, because this socket enters to hash table only
921 * after validation is complete.
923 inet_sk_state_store(sk, TCP_LISTEN);
924 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
925 inet->inet_sport = htons(inet->inet_num);
928 err = sk->sk_prot->hash(sk);
934 inet_sk_set_state(sk, TCP_CLOSE);
937 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
939 static void inet_child_forget(struct sock *sk, struct request_sock *req,
942 sk->sk_prot->disconnect(child, O_NONBLOCK);
946 percpu_counter_inc(sk->sk_prot->orphan_count);
948 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
949 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
950 BUG_ON(sk != req->rsk_listener);
952 /* Paranoid, to prevent race condition if
953 * an inbound pkt destined for child is
954 * blocked by sock lock in tcp_v4_rcv().
955 * Also to satisfy an assertion in
956 * tcp_v4_destroy_sock().
958 tcp_sk(child)->fastopen_rsk = NULL;
960 inet_csk_destroy_sock(child);
963 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
964 struct request_sock *req,
967 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
969 spin_lock(&queue->rskq_lock);
970 if (unlikely(sk->sk_state != TCP_LISTEN)) {
971 inet_child_forget(sk, req, child);
976 if (queue->rskq_accept_head == NULL)
977 WRITE_ONCE(queue->rskq_accept_head, req);
979 queue->rskq_accept_tail->dl_next = req;
980 queue->rskq_accept_tail = req;
981 sk_acceptq_added(sk);
983 spin_unlock(&queue->rskq_lock);
986 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
988 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
989 struct request_sock *req, bool own_req)
992 inet_csk_reqsk_queue_drop(sk, req);
993 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
994 if (inet_csk_reqsk_queue_add(sk, req, child))
997 /* Too bad, another child took ownership of the request, undo. */
998 bh_unlock_sock(child);
1002 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1005 * This routine closes sockets which have been at least partially
1006 * opened, but not yet accepted.
1008 void inet_csk_listen_stop(struct sock *sk)
1010 struct inet_connection_sock *icsk = inet_csk(sk);
1011 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1012 struct request_sock *next, *req;
1014 /* Following specs, it would be better either to send FIN
1015 * (and enter FIN-WAIT-1, it is normal close)
1016 * or to send active reset (abort).
1017 * Certainly, it is pretty dangerous while synflood, but it is
1018 * bad justification for our negligence 8)
1019 * To be honest, we are not able to make either
1020 * of the variants now. --ANK
1022 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1023 struct sock *child = req->sk;
1026 bh_lock_sock(child);
1027 WARN_ON(sock_owned_by_user(child));
1030 inet_child_forget(sk, req, child);
1032 bh_unlock_sock(child);
1038 if (queue->fastopenq.rskq_rst_head) {
1039 /* Free all the reqs queued in rskq_rst_head. */
1040 spin_lock_bh(&queue->fastopenq.lock);
1041 req = queue->fastopenq.rskq_rst_head;
1042 queue->fastopenq.rskq_rst_head = NULL;
1043 spin_unlock_bh(&queue->fastopenq.lock);
1044 while (req != NULL) {
1045 next = req->dl_next;
1050 WARN_ON_ONCE(sk->sk_ack_backlog);
1052 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1054 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1056 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1057 const struct inet_sock *inet = inet_sk(sk);
1059 sin->sin_family = AF_INET;
1060 sin->sin_addr.s_addr = inet->inet_daddr;
1061 sin->sin_port = inet->inet_dport;
1063 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1065 #ifdef CONFIG_COMPAT
1066 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1067 char __user *optval, int __user *optlen)
1069 const struct inet_connection_sock *icsk = inet_csk(sk);
1071 if (icsk->icsk_af_ops->compat_getsockopt)
1072 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1074 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1077 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1079 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1080 char __user *optval, unsigned int optlen)
1082 const struct inet_connection_sock *icsk = inet_csk(sk);
1084 if (icsk->icsk_af_ops->compat_setsockopt)
1085 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1087 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1090 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1093 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1095 const struct inet_sock *inet = inet_sk(sk);
1096 const struct ip_options_rcu *inet_opt;
1097 __be32 daddr = inet->inet_daddr;
1102 inet_opt = rcu_dereference(inet->inet_opt);
1103 if (inet_opt && inet_opt->opt.srr)
1104 daddr = inet_opt->opt.faddr;
1106 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1107 inet->inet_saddr, inet->inet_dport,
1108 inet->inet_sport, sk->sk_protocol,
1109 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1113 sk_setup_caps(sk, &rt->dst);
1119 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1121 struct dst_entry *dst = __sk_dst_check(sk, 0);
1122 struct inet_sock *inet = inet_sk(sk);
1125 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1129 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1131 dst = __sk_dst_check(sk, 0);
1133 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1137 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);