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) {
162 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
163 if ((!sk->sk_bound_dev_if ||
165 sk->sk_bound_dev_if == bound_dev_if2)) {
166 if (reuse && sk2->sk_reuse &&
167 sk2->sk_state != TCP_LISTEN) {
170 reuseport && sk2->sk_reuseport &&
172 (sk2->sk_state == TCP_TIME_WAIT ||
173 uid_eq(uid, sock_i_uid(sk2))))) &&
174 inet_rcv_saddr_equal(sk, sk2, true))
176 } else if (!reuseport_ok ||
177 !reuseport || !sk2->sk_reuseport ||
179 (sk2->sk_state != TCP_TIME_WAIT &&
180 !uid_eq(uid, sock_i_uid(sk2)))) {
181 if (inet_rcv_saddr_equal(sk, sk2, true))
190 * Find an open port number for the socket. Returns with the
191 * inet_bind_hashbucket lock held.
193 static struct inet_bind_hashbucket *
194 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
196 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
198 struct inet_bind_hashbucket *head;
199 struct net *net = sock_net(sk);
201 int i, low, high, attempt_half;
202 struct inet_bind_bucket *tb;
203 u32 remaining, offset;
206 l3mdev = inet_sk_bound_l3mdev(sk);
208 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
210 inet_get_local_port_range(net, &low, &high);
211 high++; /* [32768, 60999] -> [32768, 61000[ */
215 int half = low + (((high - low) >> 2) << 1);
217 if (attempt_half == 1)
222 remaining = high - low;
223 if (likely(remaining > 1))
226 offset = prandom_u32() % remaining;
227 /* __inet_hash_connect() favors ports having @low parity
228 * We do the opposite to not pollute connect() users.
234 for (i = 0; i < remaining; i += 2, port += 2) {
235 if (unlikely(port >= high))
237 if (inet_is_local_reserved_port(net, port))
239 head = &hinfo->bhash[inet_bhashfn(net, port,
241 spin_lock_bh(&head->lock);
242 inet_bind_bucket_for_each(tb, &head->chain)
243 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
245 if (!inet_csk_bind_conflict(sk, tb, relax, false))
252 spin_unlock_bh(&head->lock);
258 goto other_parity_scan;
260 if (attempt_half == 1) {
261 /* OK we now try the upper half of the range */
263 goto other_half_scan;
266 if (net->ipv4.sysctl_ip_autobind_reuse && !relax) {
267 /* We still have a chance to connect to different destinations */
269 goto ports_exhausted;
278 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
281 kuid_t uid = sock_i_uid(sk);
283 if (tb->fastreuseport <= 0)
285 if (!sk->sk_reuseport)
287 if (rcu_access_pointer(sk->sk_reuseport_cb))
289 if (!uid_eq(tb->fastuid, uid))
291 /* We only need to check the rcv_saddr if this tb was once marked
292 * without fastreuseport and then was reset, as we can only know that
293 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
296 if (tb->fastreuseport == FASTREUSEPORT_ANY)
298 #if IS_ENABLED(CONFIG_IPV6)
299 if (tb->fast_sk_family == AF_INET6)
300 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
305 ipv6_only_sock(sk), true, false);
307 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
308 ipv6_only_sock(sk), true, false);
311 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
314 kuid_t uid = sock_i_uid(sk);
315 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
317 if (hlist_empty(&tb->owners)) {
318 tb->fastreuse = reuse;
319 if (sk->sk_reuseport) {
320 tb->fastreuseport = FASTREUSEPORT_ANY;
322 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
323 tb->fast_ipv6_only = ipv6_only_sock(sk);
324 tb->fast_sk_family = sk->sk_family;
325 #if IS_ENABLED(CONFIG_IPV6)
326 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
329 tb->fastreuseport = 0;
334 if (sk->sk_reuseport) {
335 /* We didn't match or we don't have fastreuseport set on
336 * the tb, but we have sk_reuseport set on this socket
337 * and we know that there are no bind conflicts with
338 * this socket in this tb, so reset our tb's reuseport
339 * settings so that any subsequent sockets that match
340 * our current socket will be put on the fast path.
342 * If we reset we need to set FASTREUSEPORT_STRICT so we
343 * do extra checking for all subsequent sk_reuseport
346 if (!sk_reuseport_match(tb, sk)) {
347 tb->fastreuseport = FASTREUSEPORT_STRICT;
349 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
350 tb->fast_ipv6_only = ipv6_only_sock(sk);
351 tb->fast_sk_family = sk->sk_family;
352 #if IS_ENABLED(CONFIG_IPV6)
353 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
357 tb->fastreuseport = 0;
362 /* Obtain a reference to a local port for the given sock,
363 * if snum is zero it means select any available local port.
364 * We try to allocate an odd port (and leave even ports for connect())
366 int inet_csk_get_port(struct sock *sk, unsigned short snum)
368 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
369 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
370 int ret = 1, port = snum;
371 struct inet_bind_hashbucket *head;
372 struct net *net = sock_net(sk);
373 struct inet_bind_bucket *tb = NULL;
376 l3mdev = inet_sk_bound_l3mdev(sk);
379 head = inet_csk_find_open_port(sk, &tb, &port);
386 head = &hinfo->bhash[inet_bhashfn(net, port,
388 spin_lock_bh(&head->lock);
389 inet_bind_bucket_for_each(tb, &head->chain)
390 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
394 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
395 net, head, port, l3mdev);
399 if (!hlist_empty(&tb->owners)) {
400 if (sk->sk_reuse == SK_FORCE_REUSE)
403 if ((tb->fastreuse > 0 && reuse) ||
404 sk_reuseport_match(tb, sk))
406 if (inet_csk_bind_conflict(sk, tb, true, true))
410 inet_csk_update_fastreuse(tb, sk);
412 if (!inet_csk(sk)->icsk_bind_hash)
413 inet_bind_hash(sk, tb, port);
414 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
418 spin_unlock_bh(&head->lock);
421 EXPORT_SYMBOL_GPL(inet_csk_get_port);
424 * Wait for an incoming connection, avoid race conditions. This must be called
425 * with the socket locked.
427 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
429 struct inet_connection_sock *icsk = inet_csk(sk);
434 * True wake-one mechanism for incoming connections: only
435 * one process gets woken up, not the 'whole herd'.
436 * Since we do not 'race & poll' for established sockets
437 * anymore, the common case will execute the loop only once.
439 * Subtle issue: "add_wait_queue_exclusive()" will be added
440 * after any current non-exclusive waiters, and we know that
441 * it will always _stay_ after any new non-exclusive waiters
442 * because all non-exclusive waiters are added at the
443 * beginning of the wait-queue. As such, it's ok to "drop"
444 * our exclusiveness temporarily when we get woken up without
445 * having to remove and re-insert us on the wait queue.
448 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
451 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
452 timeo = schedule_timeout(timeo);
453 sched_annotate_sleep();
456 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
459 if (sk->sk_state != TCP_LISTEN)
461 err = sock_intr_errno(timeo);
462 if (signal_pending(current))
468 finish_wait(sk_sleep(sk), &wait);
473 * This will accept the next outstanding connection.
475 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
477 struct inet_connection_sock *icsk = inet_csk(sk);
478 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
479 struct request_sock *req;
485 /* We need to make sure that this socket is listening,
486 * and that it has something pending.
489 if (sk->sk_state != TCP_LISTEN)
492 /* Find already established connection */
493 if (reqsk_queue_empty(queue)) {
494 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
496 /* If this is a non blocking socket don't sleep */
501 error = inet_csk_wait_for_connect(sk, timeo);
505 req = reqsk_queue_remove(queue, sk);
508 if (sk->sk_protocol == IPPROTO_TCP &&
509 tcp_rsk(req)->tfo_listener) {
510 spin_lock_bh(&queue->fastopenq.lock);
511 if (tcp_rsk(req)->tfo_listener) {
512 /* We are still waiting for the final ACK from 3WHS
513 * so can't free req now. Instead, we set req->sk to
514 * NULL to signify that the child socket is taken
515 * so reqsk_fastopen_remove() will free the req
516 * when 3WHS finishes (or is aborted).
521 spin_unlock_bh(&queue->fastopenq.lock);
526 if (newsk && mem_cgroup_sockets_enabled) {
529 /* atomically get the memory usage, set and charge the
534 /* The socket has not been accepted yet, no need to look at
535 * newsk->sk_wmem_queued.
537 amt = sk_mem_pages(newsk->sk_forward_alloc +
538 atomic_read(&newsk->sk_rmem_alloc));
539 mem_cgroup_sk_alloc(newsk);
540 if (newsk->sk_memcg && amt)
541 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
542 GFP_KERNEL | __GFP_NOFAIL);
555 EXPORT_SYMBOL(inet_csk_accept);
558 * Using different timers for retransmit, delayed acks and probes
559 * We may wish use just one timer maintaining a list of expire jiffies
562 void inet_csk_init_xmit_timers(struct sock *sk,
563 void (*retransmit_handler)(struct timer_list *t),
564 void (*delack_handler)(struct timer_list *t),
565 void (*keepalive_handler)(struct timer_list *t))
567 struct inet_connection_sock *icsk = inet_csk(sk);
569 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
570 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
571 timer_setup(&sk->sk_timer, keepalive_handler, 0);
572 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
574 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
576 void inet_csk_clear_xmit_timers(struct sock *sk)
578 struct inet_connection_sock *icsk = inet_csk(sk);
580 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
582 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
583 sk_stop_timer(sk, &icsk->icsk_delack_timer);
584 sk_stop_timer(sk, &sk->sk_timer);
586 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
588 void inet_csk_delete_keepalive_timer(struct sock *sk)
590 sk_stop_timer(sk, &sk->sk_timer);
592 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
594 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
596 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
598 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
600 struct dst_entry *inet_csk_route_req(const struct sock *sk,
602 const struct request_sock *req)
604 const struct inet_request_sock *ireq = inet_rsk(req);
605 struct net *net = read_pnet(&ireq->ireq_net);
606 struct ip_options_rcu *opt;
610 opt = rcu_dereference(ireq->ireq_opt);
612 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
613 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
614 sk->sk_protocol, inet_sk_flowi_flags(sk),
615 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
616 ireq->ir_loc_addr, ireq->ir_rmt_port,
617 htons(ireq->ir_num), sk->sk_uid);
618 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
619 rt = ip_route_output_flow(net, fl4, sk);
622 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
631 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
634 EXPORT_SYMBOL_GPL(inet_csk_route_req);
636 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
638 const struct request_sock *req)
640 const struct inet_request_sock *ireq = inet_rsk(req);
641 struct net *net = read_pnet(&ireq->ireq_net);
642 struct inet_sock *newinet = inet_sk(newsk);
643 struct ip_options_rcu *opt;
647 opt = rcu_dereference(ireq->ireq_opt);
648 fl4 = &newinet->cork.fl.u.ip4;
650 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
651 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
652 sk->sk_protocol, inet_sk_flowi_flags(sk),
653 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
654 ireq->ir_loc_addr, ireq->ir_rmt_port,
655 htons(ireq->ir_num), sk->sk_uid);
656 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
657 rt = ip_route_output_flow(net, fl4, sk);
660 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
667 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
670 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
672 /* Decide when to expire the request and when to resend SYN-ACK */
673 static void syn_ack_recalc(struct request_sock *req,
674 const int max_syn_ack_retries,
675 const u8 rskq_defer_accept,
676 int *expire, int *resend)
678 if (!rskq_defer_accept) {
679 *expire = req->num_timeout >= max_syn_ack_retries;
683 *expire = req->num_timeout >= max_syn_ack_retries &&
684 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
685 /* Do not resend while waiting for data after ACK,
686 * start to resend on end of deferring period to give
687 * last chance for data or ACK to create established socket.
689 *resend = !inet_rsk(req)->acked ||
690 req->num_timeout >= rskq_defer_accept - 1;
693 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
695 int err = req->rsk_ops->rtx_syn_ack(parent, req);
701 EXPORT_SYMBOL(inet_rtx_syn_ack);
703 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
706 struct sock *req_sk, *nreq_sk;
707 struct request_sock *nreq;
709 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
711 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
713 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
718 req_sk = req_to_sk(req);
719 nreq_sk = req_to_sk(nreq);
721 memcpy(nreq_sk, req_sk,
722 offsetof(struct sock, sk_dontcopy_begin));
723 memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
724 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
726 sk_node_init(&nreq_sk->sk_node);
727 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
728 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
729 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
731 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
733 nreq->rsk_listener = sk;
735 /* We need not acquire fastopenq->lock
736 * because the child socket is locked in inet_csk_listen_stop().
738 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
739 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
744 static void reqsk_queue_migrated(struct request_sock_queue *queue,
745 const struct request_sock *req)
747 if (req->num_timeout == 0)
748 atomic_inc(&queue->young);
749 atomic_inc(&queue->qlen);
752 static void reqsk_migrate_reset(struct request_sock *req)
754 req->saved_syn = NULL;
755 #if IS_ENABLED(CONFIG_IPV6)
756 inet_rsk(req)->ipv6_opt = NULL;
757 inet_rsk(req)->pktopts = NULL;
759 inet_rsk(req)->ireq_opt = NULL;
763 /* return true if req was found in the ehash table */
764 static bool reqsk_queue_unlink(struct request_sock *req)
766 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
769 if (sk_hashed(req_to_sk(req))) {
770 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
773 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
776 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
781 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
783 bool unlinked = reqsk_queue_unlink(req);
786 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
791 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
793 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
795 inet_csk_reqsk_queue_drop(sk, req);
798 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
800 static void reqsk_timer_handler(struct timer_list *t)
802 struct request_sock *req = from_timer(req, t, rsk_timer);
803 struct request_sock *nreq = NULL, *oreq = req;
804 struct sock *sk_listener = req->rsk_listener;
805 struct inet_connection_sock *icsk;
806 struct request_sock_queue *queue;
808 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
810 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
813 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
817 nreq = inet_reqsk_clone(req, nsk);
821 /* The new timer for the cloned req can decrease the 2
822 * by calling inet_csk_reqsk_queue_drop_and_put(), so
823 * hold another count to prevent use-after-free and
824 * call reqsk_put() just before return.
826 refcount_set(&nreq->rsk_refcnt, 2 + 1);
827 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
828 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
834 icsk = inet_csk(sk_listener);
835 net = sock_net(sk_listener);
836 max_syn_ack_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
837 /* Normally all the openreqs are young and become mature
838 * (i.e. converted to established socket) for first timeout.
839 * If synack was not acknowledged for 1 second, it means
840 * one of the following things: synack was lost, ack was lost,
841 * rtt is high or nobody planned to ack (i.e. synflood).
842 * When server is a bit loaded, queue is populated with old
843 * open requests, reducing effective size of queue.
844 * When server is well loaded, queue size reduces to zero
845 * after several minutes of work. It is not synflood,
846 * it is normal operation. The solution is pruning
847 * too old entries overriding normal timeout, when
848 * situation becomes dangerous.
850 * Essentially, we reserve half of room for young
851 * embrions; and abort old ones without pity, if old
852 * ones are about to clog our table.
854 queue = &icsk->icsk_accept_queue;
855 qlen = reqsk_queue_len(queue);
856 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
857 int young = reqsk_queue_len_young(queue) << 1;
859 while (max_syn_ack_retries > 2) {
862 max_syn_ack_retries--;
866 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
868 req->rsk_ops->syn_ack_timeout(req);
871 !inet_rtx_syn_ack(sk_listener, req) ||
872 inet_rsk(req)->acked)) {
873 if (req->num_timeout++ == 0)
874 atomic_dec(&queue->young);
875 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
880 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
882 inet_csk_reqsk_queue_drop(sk_listener, nreq);
886 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
887 reqsk_migrate_reset(oreq);
888 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
895 /* Even if we can clone the req, we may need not retransmit any more
896 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
897 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
900 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
902 reqsk_migrate_reset(nreq);
903 reqsk_queue_removed(queue, nreq);
908 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
911 static void reqsk_queue_hash_req(struct request_sock *req,
912 unsigned long timeout)
914 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
915 mod_timer(&req->rsk_timer, jiffies + timeout);
917 inet_ehash_insert(req_to_sk(req), NULL, NULL);
918 /* before letting lookups find us, make sure all req fields
919 * are committed to memory and refcnt initialized.
922 refcount_set(&req->rsk_refcnt, 2 + 1);
925 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
926 unsigned long timeout)
928 reqsk_queue_hash_req(req, timeout);
929 inet_csk_reqsk_queue_added(sk);
931 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
933 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
934 const gfp_t priority)
936 struct inet_connection_sock *icsk = inet_csk(newsk);
938 if (!icsk->icsk_ulp_ops)
941 if (icsk->icsk_ulp_ops->clone)
942 icsk->icsk_ulp_ops->clone(req, newsk, priority);
946 * inet_csk_clone_lock - clone an inet socket, and lock its clone
947 * @sk: the socket to clone
949 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
951 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
953 struct sock *inet_csk_clone_lock(const struct sock *sk,
954 const struct request_sock *req,
955 const gfp_t priority)
957 struct sock *newsk = sk_clone_lock(sk, priority);
960 struct inet_connection_sock *newicsk = inet_csk(newsk);
962 inet_sk_set_state(newsk, TCP_SYN_RECV);
963 newicsk->icsk_bind_hash = NULL;
965 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
966 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
967 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
969 /* listeners have SOCK_RCU_FREE, not the children */
970 sock_reset_flag(newsk, SOCK_RCU_FREE);
972 inet_sk(newsk)->mc_list = NULL;
974 newsk->sk_mark = inet_rsk(req)->ir_mark;
975 atomic64_set(&newsk->sk_cookie,
976 atomic64_read(&inet_rsk(req)->ir_cookie));
978 newicsk->icsk_retransmits = 0;
979 newicsk->icsk_backoff = 0;
980 newicsk->icsk_probes_out = 0;
981 newicsk->icsk_probes_tstamp = 0;
983 /* Deinitialize accept_queue to trap illegal accesses. */
984 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
986 inet_clone_ulp(req, newsk, priority);
988 security_inet_csk_clone(newsk, req);
992 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
995 * At this point, there should be no process reference to this
996 * socket, and thus no user references at all. Therefore we
997 * can assume the socket waitqueue is inactive and nobody will
998 * try to jump onto it.
1000 void inet_csk_destroy_sock(struct sock *sk)
1002 WARN_ON(sk->sk_state != TCP_CLOSE);
1003 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1005 /* It cannot be in hash table! */
1006 WARN_ON(!sk_unhashed(sk));
1008 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1009 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1011 sk->sk_prot->destroy(sk);
1013 sk_stream_kill_queues(sk);
1015 xfrm_sk_free_policy(sk);
1017 sk_refcnt_debug_release(sk);
1019 this_cpu_dec(*sk->sk_prot->orphan_count);
1023 EXPORT_SYMBOL(inet_csk_destroy_sock);
1025 /* This function allows to force a closure of a socket after the call to
1026 * tcp/dccp_create_openreq_child().
1028 void inet_csk_prepare_forced_close(struct sock *sk)
1029 __releases(&sk->sk_lock.slock)
1031 /* sk_clone_lock locked the socket and set refcnt to 2 */
1034 inet_csk_prepare_for_destroy_sock(sk);
1035 inet_sk(sk)->inet_num = 0;
1037 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1039 int inet_csk_listen_start(struct sock *sk)
1041 struct inet_connection_sock *icsk = inet_csk(sk);
1042 struct inet_sock *inet = inet_sk(sk);
1043 int err = -EADDRINUSE;
1045 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1047 sk->sk_ack_backlog = 0;
1048 inet_csk_delack_init(sk);
1050 if (sk->sk_txrehash == SOCK_TXREHASH_DEFAULT)
1051 sk->sk_txrehash = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1053 /* There is race window here: we announce ourselves listening,
1054 * but this transition is still not validated by get_port().
1055 * It is OK, because this socket enters to hash table only
1056 * after validation is complete.
1058 inet_sk_state_store(sk, TCP_LISTEN);
1059 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
1060 inet->inet_sport = htons(inet->inet_num);
1063 err = sk->sk_prot->hash(sk);
1069 inet_sk_set_state(sk, TCP_CLOSE);
1072 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1074 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1077 sk->sk_prot->disconnect(child, O_NONBLOCK);
1081 this_cpu_inc(*sk->sk_prot->orphan_count);
1083 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1084 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1085 BUG_ON(sk != req->rsk_listener);
1087 /* Paranoid, to prevent race condition if
1088 * an inbound pkt destined for child is
1089 * blocked by sock lock in tcp_v4_rcv().
1090 * Also to satisfy an assertion in
1091 * tcp_v4_destroy_sock().
1093 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1095 inet_csk_destroy_sock(child);
1098 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1099 struct request_sock *req,
1102 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1104 spin_lock(&queue->rskq_lock);
1105 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1106 inet_child_forget(sk, req, child);
1110 req->dl_next = NULL;
1111 if (queue->rskq_accept_head == NULL)
1112 WRITE_ONCE(queue->rskq_accept_head, req);
1114 queue->rskq_accept_tail->dl_next = req;
1115 queue->rskq_accept_tail = req;
1116 sk_acceptq_added(sk);
1118 spin_unlock(&queue->rskq_lock);
1121 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1123 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1124 struct request_sock *req, bool own_req)
1127 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1128 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1130 if (sk != req->rsk_listener) {
1131 /* another listening sk has been selected,
1132 * migrate the req to it.
1134 struct request_sock *nreq;
1136 /* hold a refcnt for the nreq->rsk_listener
1137 * which is assigned in inet_reqsk_clone()
1140 nreq = inet_reqsk_clone(req, sk);
1142 inet_child_forget(sk, req, child);
1146 refcount_set(&nreq->rsk_refcnt, 1);
1147 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1148 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1149 reqsk_migrate_reset(req);
1154 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1155 reqsk_migrate_reset(nreq);
1157 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1161 /* Too bad, another child took ownership of the request, undo. */
1163 bh_unlock_sock(child);
1167 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1170 * This routine closes sockets which have been at least partially
1171 * opened, but not yet accepted.
1173 void inet_csk_listen_stop(struct sock *sk)
1175 struct inet_connection_sock *icsk = inet_csk(sk);
1176 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1177 struct request_sock *next, *req;
1179 /* Following specs, it would be better either to send FIN
1180 * (and enter FIN-WAIT-1, it is normal close)
1181 * or to send active reset (abort).
1182 * Certainly, it is pretty dangerous while synflood, but it is
1183 * bad justification for our negligence 8)
1184 * To be honest, we are not able to make either
1185 * of the variants now. --ANK
1187 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1188 struct sock *child = req->sk, *nsk;
1189 struct request_sock *nreq;
1192 bh_lock_sock(child);
1193 WARN_ON(sock_owned_by_user(child));
1196 nsk = reuseport_migrate_sock(sk, child, NULL);
1198 nreq = inet_reqsk_clone(req, nsk);
1200 refcount_set(&nreq->rsk_refcnt, 1);
1202 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1203 __NET_INC_STATS(sock_net(nsk),
1204 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1205 reqsk_migrate_reset(req);
1207 __NET_INC_STATS(sock_net(nsk),
1208 LINUX_MIB_TCPMIGRATEREQFAILURE);
1209 reqsk_migrate_reset(nreq);
1213 /* inet_csk_reqsk_queue_add() has already
1214 * called inet_child_forget() on failure case.
1216 goto skip_child_forget;
1220 inet_child_forget(sk, req, child);
1223 bh_unlock_sock(child);
1229 if (queue->fastopenq.rskq_rst_head) {
1230 /* Free all the reqs queued in rskq_rst_head. */
1231 spin_lock_bh(&queue->fastopenq.lock);
1232 req = queue->fastopenq.rskq_rst_head;
1233 queue->fastopenq.rskq_rst_head = NULL;
1234 spin_unlock_bh(&queue->fastopenq.lock);
1235 while (req != NULL) {
1236 next = req->dl_next;
1241 WARN_ON_ONCE(sk->sk_ack_backlog);
1243 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1245 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1247 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1248 const struct inet_sock *inet = inet_sk(sk);
1250 sin->sin_family = AF_INET;
1251 sin->sin_addr.s_addr = inet->inet_daddr;
1252 sin->sin_port = inet->inet_dport;
1254 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1256 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1258 const struct inet_sock *inet = inet_sk(sk);
1259 const struct ip_options_rcu *inet_opt;
1260 __be32 daddr = inet->inet_daddr;
1265 inet_opt = rcu_dereference(inet->inet_opt);
1266 if (inet_opt && inet_opt->opt.srr)
1267 daddr = inet_opt->opt.faddr;
1269 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1270 inet->inet_saddr, inet->inet_dport,
1271 inet->inet_sport, sk->sk_protocol,
1272 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1276 sk_setup_caps(sk, &rt->dst);
1282 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1284 struct dst_entry *dst = __sk_dst_check(sk, 0);
1285 struct inet_sock *inet = inet_sk(sk);
1288 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1292 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1294 dst = __sk_dst_check(sk, 0);
1296 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1300 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);