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(const 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 void inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
135 const struct inet_sock *inet = inet_sk(sk);
136 const struct net *net = sock_net(sk);
137 int lo, hi, sk_lo, sk_hi;
139 inet_get_local_port_range(net, &lo, &hi);
141 sk_lo = inet->local_port_range.lo;
142 sk_hi = inet->local_port_range.hi;
144 if (unlikely(lo <= sk_lo && sk_lo <= hi))
146 if (unlikely(lo <= sk_hi && sk_hi <= hi))
152 EXPORT_SYMBOL(inet_sk_get_local_port_range);
154 static bool inet_use_bhash2_on_bind(const struct sock *sk)
156 #if IS_ENABLED(CONFIG_IPV6)
157 if (sk->sk_family == AF_INET6) {
158 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
160 return addr_type != IPV6_ADDR_ANY &&
161 addr_type != IPV6_ADDR_MAPPED;
164 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
167 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
168 kuid_t sk_uid, bool relax,
169 bool reuseport_cb_ok, bool reuseport_ok)
176 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
178 if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
179 sk->sk_bound_dev_if == bound_dev_if2) {
180 if (sk->sk_reuse && sk2->sk_reuse &&
181 sk2->sk_state != TCP_LISTEN) {
182 if (!relax || (!reuseport_ok && sk->sk_reuseport &&
183 sk2->sk_reuseport && reuseport_cb_ok &&
184 (sk2->sk_state == TCP_TIME_WAIT ||
185 uid_eq(sk_uid, sock_i_uid(sk2)))))
187 } else if (!reuseport_ok || !sk->sk_reuseport ||
188 !sk2->sk_reuseport || !reuseport_cb_ok ||
189 (sk2->sk_state != TCP_TIME_WAIT &&
190 !uid_eq(sk_uid, sock_i_uid(sk2)))) {
197 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
198 kuid_t sk_uid, bool relax,
199 bool reuseport_cb_ok, bool reuseport_ok)
201 if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
204 return inet_bind_conflict(sk, sk2, sk_uid, relax,
205 reuseport_cb_ok, reuseport_ok);
208 static bool inet_bhash2_conflict(const struct sock *sk,
209 const struct inet_bind2_bucket *tb2,
211 bool relax, bool reuseport_cb_ok,
214 struct inet_timewait_sock *tw2;
217 sk_for_each_bound_bhash2(sk2, &tb2->owners) {
218 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
219 reuseport_cb_ok, reuseport_ok))
223 twsk_for_each_bound_bhash2(tw2, &tb2->deathrow) {
224 sk2 = (struct sock *)tw2;
226 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
227 reuseport_cb_ok, reuseport_ok))
234 /* This should be called only when the tb and tb2 hashbuckets' locks are held */
235 static int inet_csk_bind_conflict(const struct sock *sk,
236 const struct inet_bind_bucket *tb,
237 const struct inet_bind2_bucket *tb2, /* may be null */
238 bool relax, bool reuseport_ok)
240 bool reuseport_cb_ok;
241 struct sock_reuseport *reuseport_cb;
242 kuid_t uid = sock_i_uid((struct sock *)sk);
245 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
246 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
247 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
251 * Unlike other sk lookup places we do not check
252 * for sk_net here, since _all_ the socks listed
253 * in tb->owners and tb2->owners list belong
254 * to the same net - the one this bucket belongs to.
257 if (!inet_use_bhash2_on_bind(sk)) {
260 sk_for_each_bound(sk2, &tb->owners)
261 if (inet_bind_conflict(sk, sk2, uid, relax,
262 reuseport_cb_ok, reuseport_ok) &&
263 inet_rcv_saddr_equal(sk, sk2, true))
269 /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
270 * ipv4) should have been checked already. We need to do these two
271 * checks separately because their spinlocks have to be acquired/released
272 * independently of each other, to prevent possible deadlocks
274 return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
278 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
279 * INADDR_ANY (if ipv4) socket.
281 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
282 * against concurrent binds on the port for addr any
284 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
285 bool relax, bool reuseport_ok)
287 kuid_t uid = sock_i_uid((struct sock *)sk);
288 const struct net *net = sock_net(sk);
289 struct sock_reuseport *reuseport_cb;
290 struct inet_bind_hashbucket *head2;
291 struct inet_bind2_bucket *tb2;
292 bool reuseport_cb_ok;
295 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
296 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
297 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
300 head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
302 spin_lock(&head2->lock);
304 inet_bind_bucket_for_each(tb2, &head2->chain)
305 if (inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
308 if (tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
310 spin_unlock(&head2->lock);
314 spin_unlock(&head2->lock);
319 * Find an open port number for the socket. Returns with the
320 * inet_bind_hashbucket locks held if successful.
322 static struct inet_bind_hashbucket *
323 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
324 struct inet_bind2_bucket **tb2_ret,
325 struct inet_bind_hashbucket **head2_ret, int *port_ret)
327 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
328 int i, low, high, attempt_half, port, l3mdev;
329 struct inet_bind_hashbucket *head, *head2;
330 struct net *net = sock_net(sk);
331 struct inet_bind2_bucket *tb2;
332 struct inet_bind_bucket *tb;
333 u32 remaining, offset;
336 l3mdev = inet_sk_bound_l3mdev(sk);
338 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
340 inet_sk_get_local_port_range(sk, &low, &high);
341 high++; /* [32768, 60999] -> [32768, 61000[ */
345 int half = low + (((high - low) >> 2) << 1);
347 if (attempt_half == 1)
352 remaining = high - low;
353 if (likely(remaining > 1))
356 offset = get_random_u32_below(remaining);
357 /* __inet_hash_connect() favors ports having @low parity
358 * We do the opposite to not pollute connect() users.
364 for (i = 0; i < remaining; i += 2, port += 2) {
365 if (unlikely(port >= high))
367 if (inet_is_local_reserved_port(net, port))
369 head = &hinfo->bhash[inet_bhashfn(net, port,
371 spin_lock_bh(&head->lock);
372 if (inet_use_bhash2_on_bind(sk)) {
373 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
377 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
378 spin_lock(&head2->lock);
379 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
380 inet_bind_bucket_for_each(tb, &head->chain)
381 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
382 if (!inet_csk_bind_conflict(sk, tb, tb2,
385 spin_unlock(&head2->lock);
391 spin_unlock_bh(&head->lock);
397 goto other_parity_scan;
399 if (attempt_half == 1) {
400 /* OK we now try the upper half of the range */
402 goto other_half_scan;
405 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
406 /* We still have a chance to connect to different destinations */
408 goto ports_exhausted;
419 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
422 kuid_t uid = sock_i_uid(sk);
424 if (tb->fastreuseport <= 0)
426 if (!sk->sk_reuseport)
428 if (rcu_access_pointer(sk->sk_reuseport_cb))
430 if (!uid_eq(tb->fastuid, uid))
432 /* We only need to check the rcv_saddr if this tb was once marked
433 * without fastreuseport and then was reset, as we can only know that
434 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
437 if (tb->fastreuseport == FASTREUSEPORT_ANY)
439 #if IS_ENABLED(CONFIG_IPV6)
440 if (tb->fast_sk_family == AF_INET6)
441 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
446 ipv6_only_sock(sk), true, false);
448 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
449 ipv6_only_sock(sk), true, false);
452 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
455 kuid_t uid = sock_i_uid(sk);
456 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
458 if (hlist_empty(&tb->owners)) {
459 tb->fastreuse = reuse;
460 if (sk->sk_reuseport) {
461 tb->fastreuseport = FASTREUSEPORT_ANY;
463 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
464 tb->fast_ipv6_only = ipv6_only_sock(sk);
465 tb->fast_sk_family = sk->sk_family;
466 #if IS_ENABLED(CONFIG_IPV6)
467 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
470 tb->fastreuseport = 0;
475 if (sk->sk_reuseport) {
476 /* We didn't match or we don't have fastreuseport set on
477 * the tb, but we have sk_reuseport set on this socket
478 * and we know that there are no bind conflicts with
479 * this socket in this tb, so reset our tb's reuseport
480 * settings so that any subsequent sockets that match
481 * our current socket will be put on the fast path.
483 * If we reset we need to set FASTREUSEPORT_STRICT so we
484 * do extra checking for all subsequent sk_reuseport
487 if (!sk_reuseport_match(tb, sk)) {
488 tb->fastreuseport = FASTREUSEPORT_STRICT;
490 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
491 tb->fast_ipv6_only = ipv6_only_sock(sk);
492 tb->fast_sk_family = sk->sk_family;
493 #if IS_ENABLED(CONFIG_IPV6)
494 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
498 tb->fastreuseport = 0;
503 /* Obtain a reference to a local port for the given sock,
504 * if snum is zero it means select any available local port.
505 * We try to allocate an odd port (and leave even ports for connect())
507 int inet_csk_get_port(struct sock *sk, unsigned short snum)
509 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
510 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
511 bool found_port = false, check_bind_conflict = true;
512 bool bhash_created = false, bhash2_created = false;
513 int ret = -EADDRINUSE, port = snum, l3mdev;
514 struct inet_bind_hashbucket *head, *head2;
515 struct inet_bind2_bucket *tb2 = NULL;
516 struct inet_bind_bucket *tb = NULL;
517 bool head2_lock_acquired = false;
518 struct net *net = sock_net(sk);
520 l3mdev = inet_sk_bound_l3mdev(sk);
523 head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
527 head2_lock_acquired = true;
533 head = &hinfo->bhash[inet_bhashfn(net, port,
535 spin_lock_bh(&head->lock);
536 inet_bind_bucket_for_each(tb, &head->chain)
537 if (inet_bind_bucket_match(tb, net, port, l3mdev))
542 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
546 bhash_created = true;
550 if (!hlist_empty(&tb->owners)) {
551 if (sk->sk_reuse == SK_FORCE_REUSE ||
552 (tb->fastreuse > 0 && reuse) ||
553 sk_reuseport_match(tb, sk))
554 check_bind_conflict = false;
557 if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
558 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
562 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
563 spin_lock(&head2->lock);
564 head2_lock_acquired = true;
565 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
569 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
570 net, head2, port, l3mdev, sk);
573 bhash2_created = true;
576 if (!found_port && check_bind_conflict) {
577 if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
582 inet_csk_update_fastreuse(tb, sk);
584 if (!inet_csk(sk)->icsk_bind_hash)
585 inet_bind_hash(sk, tb, tb2, port);
586 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
587 WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
593 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
595 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep,
598 if (head2_lock_acquired)
599 spin_unlock(&head2->lock);
600 spin_unlock_bh(&head->lock);
603 EXPORT_SYMBOL_GPL(inet_csk_get_port);
606 * Wait for an incoming connection, avoid race conditions. This must be called
607 * with the socket locked.
609 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
611 struct inet_connection_sock *icsk = inet_csk(sk);
616 * True wake-one mechanism for incoming connections: only
617 * one process gets woken up, not the 'whole herd'.
618 * Since we do not 'race & poll' for established sockets
619 * anymore, the common case will execute the loop only once.
621 * Subtle issue: "add_wait_queue_exclusive()" will be added
622 * after any current non-exclusive waiters, and we know that
623 * it will always _stay_ after any new non-exclusive waiters
624 * because all non-exclusive waiters are added at the
625 * beginning of the wait-queue. As such, it's ok to "drop"
626 * our exclusiveness temporarily when we get woken up without
627 * having to remove and re-insert us on the wait queue.
630 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
633 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
634 timeo = schedule_timeout(timeo);
635 sched_annotate_sleep();
638 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
641 if (sk->sk_state != TCP_LISTEN)
643 err = sock_intr_errno(timeo);
644 if (signal_pending(current))
650 finish_wait(sk_sleep(sk), &wait);
655 * This will accept the next outstanding connection.
657 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
659 struct inet_connection_sock *icsk = inet_csk(sk);
660 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
661 struct request_sock *req;
667 /* We need to make sure that this socket is listening,
668 * and that it has something pending.
671 if (sk->sk_state != TCP_LISTEN)
674 /* Find already established connection */
675 if (reqsk_queue_empty(queue)) {
676 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
678 /* If this is a non blocking socket don't sleep */
683 error = inet_csk_wait_for_connect(sk, timeo);
687 req = reqsk_queue_remove(queue, sk);
690 if (sk->sk_protocol == IPPROTO_TCP &&
691 tcp_rsk(req)->tfo_listener) {
692 spin_lock_bh(&queue->fastopenq.lock);
693 if (tcp_rsk(req)->tfo_listener) {
694 /* We are still waiting for the final ACK from 3WHS
695 * so can't free req now. Instead, we set req->sk to
696 * NULL to signify that the child socket is taken
697 * so reqsk_fastopen_remove() will free the req
698 * when 3WHS finishes (or is aborted).
703 spin_unlock_bh(&queue->fastopenq.lock);
708 if (newsk && mem_cgroup_sockets_enabled) {
711 /* atomically get the memory usage, set and charge the
716 mem_cgroup_sk_alloc(newsk);
717 if (newsk->sk_memcg) {
718 /* The socket has not been accepted yet, no need
719 * to look at newsk->sk_wmem_queued.
721 amt = sk_mem_pages(newsk->sk_forward_alloc +
722 atomic_read(&newsk->sk_rmem_alloc));
726 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
727 GFP_KERNEL | __GFP_NOFAIL);
740 EXPORT_SYMBOL(inet_csk_accept);
743 * Using different timers for retransmit, delayed acks and probes
744 * We may wish use just one timer maintaining a list of expire jiffies
747 void inet_csk_init_xmit_timers(struct sock *sk,
748 void (*retransmit_handler)(struct timer_list *t),
749 void (*delack_handler)(struct timer_list *t),
750 void (*keepalive_handler)(struct timer_list *t))
752 struct inet_connection_sock *icsk = inet_csk(sk);
754 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
755 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
756 timer_setup(&sk->sk_timer, keepalive_handler, 0);
757 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
759 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
761 void inet_csk_clear_xmit_timers(struct sock *sk)
763 struct inet_connection_sock *icsk = inet_csk(sk);
765 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
767 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
768 sk_stop_timer(sk, &icsk->icsk_delack_timer);
769 sk_stop_timer(sk, &sk->sk_timer);
771 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
773 void inet_csk_delete_keepalive_timer(struct sock *sk)
775 sk_stop_timer(sk, &sk->sk_timer);
777 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
779 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
781 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
783 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
785 struct dst_entry *inet_csk_route_req(const struct sock *sk,
787 const struct request_sock *req)
789 const struct inet_request_sock *ireq = inet_rsk(req);
790 struct net *net = read_pnet(&ireq->ireq_net);
791 struct ip_options_rcu *opt;
795 opt = rcu_dereference(ireq->ireq_opt);
797 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
798 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
799 sk->sk_protocol, inet_sk_flowi_flags(sk),
800 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
801 ireq->ir_loc_addr, ireq->ir_rmt_port,
802 htons(ireq->ir_num), sk->sk_uid);
803 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
804 rt = ip_route_output_flow(net, fl4, sk);
807 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
816 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
819 EXPORT_SYMBOL_GPL(inet_csk_route_req);
821 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
823 const struct request_sock *req)
825 const struct inet_request_sock *ireq = inet_rsk(req);
826 struct net *net = read_pnet(&ireq->ireq_net);
827 struct inet_sock *newinet = inet_sk(newsk);
828 struct ip_options_rcu *opt;
832 opt = rcu_dereference(ireq->ireq_opt);
833 fl4 = &newinet->cork.fl.u.ip4;
835 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
836 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
837 sk->sk_protocol, inet_sk_flowi_flags(sk),
838 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
839 ireq->ir_loc_addr, ireq->ir_rmt_port,
840 htons(ireq->ir_num), sk->sk_uid);
841 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
842 rt = ip_route_output_flow(net, fl4, sk);
845 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
852 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
855 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
857 /* Decide when to expire the request and when to resend SYN-ACK */
858 static void syn_ack_recalc(struct request_sock *req,
859 const int max_syn_ack_retries,
860 const u8 rskq_defer_accept,
861 int *expire, int *resend)
863 if (!rskq_defer_accept) {
864 *expire = req->num_timeout >= max_syn_ack_retries;
868 *expire = req->num_timeout >= max_syn_ack_retries &&
869 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
870 /* Do not resend while waiting for data after ACK,
871 * start to resend on end of deferring period to give
872 * last chance for data or ACK to create established socket.
874 *resend = !inet_rsk(req)->acked ||
875 req->num_timeout >= rskq_defer_accept - 1;
878 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
880 int err = req->rsk_ops->rtx_syn_ack(parent, req);
886 EXPORT_SYMBOL(inet_rtx_syn_ack);
888 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
891 struct sock *req_sk, *nreq_sk;
892 struct request_sock *nreq;
894 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
896 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
898 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
903 req_sk = req_to_sk(req);
904 nreq_sk = req_to_sk(nreq);
906 memcpy(nreq_sk, req_sk,
907 offsetof(struct sock, sk_dontcopy_begin));
908 memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
909 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
911 sk_node_init(&nreq_sk->sk_node);
912 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
913 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
914 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
916 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
918 nreq->rsk_listener = sk;
920 /* We need not acquire fastopenq->lock
921 * because the child socket is locked in inet_csk_listen_stop().
923 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
924 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
929 static void reqsk_queue_migrated(struct request_sock_queue *queue,
930 const struct request_sock *req)
932 if (req->num_timeout == 0)
933 atomic_inc(&queue->young);
934 atomic_inc(&queue->qlen);
937 static void reqsk_migrate_reset(struct request_sock *req)
939 req->saved_syn = NULL;
940 #if IS_ENABLED(CONFIG_IPV6)
941 inet_rsk(req)->ipv6_opt = NULL;
942 inet_rsk(req)->pktopts = NULL;
944 inet_rsk(req)->ireq_opt = NULL;
948 /* return true if req was found in the ehash table */
949 static bool reqsk_queue_unlink(struct request_sock *req)
951 struct sock *sk = req_to_sk(req);
955 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
956 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
959 found = __sk_nulls_del_node_init_rcu(sk);
962 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
967 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
969 bool unlinked = reqsk_queue_unlink(req);
972 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
977 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
979 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
981 inet_csk_reqsk_queue_drop(sk, req);
984 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
986 static void reqsk_timer_handler(struct timer_list *t)
988 struct request_sock *req = from_timer(req, t, rsk_timer);
989 struct request_sock *nreq = NULL, *oreq = req;
990 struct sock *sk_listener = req->rsk_listener;
991 struct inet_connection_sock *icsk;
992 struct request_sock_queue *queue;
994 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
996 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
999 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1003 nreq = inet_reqsk_clone(req, nsk);
1007 /* The new timer for the cloned req can decrease the 2
1008 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1009 * hold another count to prevent use-after-free and
1010 * call reqsk_put() just before return.
1012 refcount_set(&nreq->rsk_refcnt, 2 + 1);
1013 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1014 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1020 icsk = inet_csk(sk_listener);
1021 net = sock_net(sk_listener);
1022 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1023 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1024 /* Normally all the openreqs are young and become mature
1025 * (i.e. converted to established socket) for first timeout.
1026 * If synack was not acknowledged for 1 second, it means
1027 * one of the following things: synack was lost, ack was lost,
1028 * rtt is high or nobody planned to ack (i.e. synflood).
1029 * When server is a bit loaded, queue is populated with old
1030 * open requests, reducing effective size of queue.
1031 * When server is well loaded, queue size reduces to zero
1032 * after several minutes of work. It is not synflood,
1033 * it is normal operation. The solution is pruning
1034 * too old entries overriding normal timeout, when
1035 * situation becomes dangerous.
1037 * Essentially, we reserve half of room for young
1038 * embrions; and abort old ones without pity, if old
1039 * ones are about to clog our table.
1041 queue = &icsk->icsk_accept_queue;
1042 qlen = reqsk_queue_len(queue);
1043 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1044 int young = reqsk_queue_len_young(queue) << 1;
1046 while (max_syn_ack_retries > 2) {
1049 max_syn_ack_retries--;
1053 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1055 req->rsk_ops->syn_ack_timeout(req);
1058 !inet_rtx_syn_ack(sk_listener, req) ||
1059 inet_rsk(req)->acked)) {
1060 if (req->num_timeout++ == 0)
1061 atomic_dec(&queue->young);
1062 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1067 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1069 inet_csk_reqsk_queue_drop(sk_listener, nreq);
1073 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1074 reqsk_migrate_reset(oreq);
1075 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1082 /* Even if we can clone the req, we may need not retransmit any more
1083 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1084 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1087 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1089 reqsk_migrate_reset(nreq);
1090 reqsk_queue_removed(queue, nreq);
1095 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1098 static void reqsk_queue_hash_req(struct request_sock *req,
1099 unsigned long timeout)
1101 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1102 mod_timer(&req->rsk_timer, jiffies + timeout);
1104 inet_ehash_insert(req_to_sk(req), NULL, NULL);
1105 /* before letting lookups find us, make sure all req fields
1106 * are committed to memory and refcnt initialized.
1109 refcount_set(&req->rsk_refcnt, 2 + 1);
1112 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1113 unsigned long timeout)
1115 reqsk_queue_hash_req(req, timeout);
1116 inet_csk_reqsk_queue_added(sk);
1118 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1120 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1121 const gfp_t priority)
1123 struct inet_connection_sock *icsk = inet_csk(newsk);
1125 if (!icsk->icsk_ulp_ops)
1128 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1132 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1133 * @sk: the socket to clone
1134 * @req: request_sock
1135 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1137 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1139 struct sock *inet_csk_clone_lock(const struct sock *sk,
1140 const struct request_sock *req,
1141 const gfp_t priority)
1143 struct sock *newsk = sk_clone_lock(sk, priority);
1146 struct inet_connection_sock *newicsk = inet_csk(newsk);
1148 inet_sk_set_state(newsk, TCP_SYN_RECV);
1149 newicsk->icsk_bind_hash = NULL;
1150 newicsk->icsk_bind2_hash = NULL;
1152 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1153 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1154 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1156 /* listeners have SOCK_RCU_FREE, not the children */
1157 sock_reset_flag(newsk, SOCK_RCU_FREE);
1159 inet_sk(newsk)->mc_list = NULL;
1161 newsk->sk_mark = inet_rsk(req)->ir_mark;
1162 atomic64_set(&newsk->sk_cookie,
1163 atomic64_read(&inet_rsk(req)->ir_cookie));
1165 newicsk->icsk_retransmits = 0;
1166 newicsk->icsk_backoff = 0;
1167 newicsk->icsk_probes_out = 0;
1168 newicsk->icsk_probes_tstamp = 0;
1170 /* Deinitialize accept_queue to trap illegal accesses. */
1171 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1173 inet_clone_ulp(req, newsk, priority);
1175 security_inet_csk_clone(newsk, req);
1179 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1182 * At this point, there should be no process reference to this
1183 * socket, and thus no user references at all. Therefore we
1184 * can assume the socket waitqueue is inactive and nobody will
1185 * try to jump onto it.
1187 void inet_csk_destroy_sock(struct sock *sk)
1189 WARN_ON(sk->sk_state != TCP_CLOSE);
1190 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1192 /* It cannot be in hash table! */
1193 WARN_ON(!sk_unhashed(sk));
1195 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1196 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1198 sk->sk_prot->destroy(sk);
1200 sk_stream_kill_queues(sk);
1202 xfrm_sk_free_policy(sk);
1204 this_cpu_dec(*sk->sk_prot->orphan_count);
1208 EXPORT_SYMBOL(inet_csk_destroy_sock);
1210 /* This function allows to force a closure of a socket after the call to
1211 * tcp/dccp_create_openreq_child().
1213 void inet_csk_prepare_forced_close(struct sock *sk)
1214 __releases(&sk->sk_lock.slock)
1216 /* sk_clone_lock locked the socket and set refcnt to 2 */
1219 inet_csk_prepare_for_destroy_sock(sk);
1220 inet_sk(sk)->inet_num = 0;
1222 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1224 static int inet_ulp_can_listen(const struct sock *sk)
1226 const struct inet_connection_sock *icsk = inet_csk(sk);
1228 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1234 int inet_csk_listen_start(struct sock *sk)
1236 struct inet_connection_sock *icsk = inet_csk(sk);
1237 struct inet_sock *inet = inet_sk(sk);
1240 err = inet_ulp_can_listen(sk);
1244 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1246 sk->sk_ack_backlog = 0;
1247 inet_csk_delack_init(sk);
1249 /* There is race window here: we announce ourselves listening,
1250 * but this transition is still not validated by get_port().
1251 * It is OK, because this socket enters to hash table only
1252 * after validation is complete.
1254 inet_sk_state_store(sk, TCP_LISTEN);
1255 err = sk->sk_prot->get_port(sk, inet->inet_num);
1257 inet->inet_sport = htons(inet->inet_num);
1260 err = sk->sk_prot->hash(sk);
1266 inet_sk_set_state(sk, TCP_CLOSE);
1269 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1271 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1274 sk->sk_prot->disconnect(child, O_NONBLOCK);
1278 this_cpu_inc(*sk->sk_prot->orphan_count);
1280 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1281 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1282 BUG_ON(sk != req->rsk_listener);
1284 /* Paranoid, to prevent race condition if
1285 * an inbound pkt destined for child is
1286 * blocked by sock lock in tcp_v4_rcv().
1287 * Also to satisfy an assertion in
1288 * tcp_v4_destroy_sock().
1290 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1292 inet_csk_destroy_sock(child);
1295 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1296 struct request_sock *req,
1299 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1301 spin_lock(&queue->rskq_lock);
1302 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1303 inet_child_forget(sk, req, child);
1307 req->dl_next = NULL;
1308 if (queue->rskq_accept_head == NULL)
1309 WRITE_ONCE(queue->rskq_accept_head, req);
1311 queue->rskq_accept_tail->dl_next = req;
1312 queue->rskq_accept_tail = req;
1313 sk_acceptq_added(sk);
1315 spin_unlock(&queue->rskq_lock);
1318 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1320 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1321 struct request_sock *req, bool own_req)
1324 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1325 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1327 if (sk != req->rsk_listener) {
1328 /* another listening sk has been selected,
1329 * migrate the req to it.
1331 struct request_sock *nreq;
1333 /* hold a refcnt for the nreq->rsk_listener
1334 * which is assigned in inet_reqsk_clone()
1337 nreq = inet_reqsk_clone(req, sk);
1339 inet_child_forget(sk, req, child);
1343 refcount_set(&nreq->rsk_refcnt, 1);
1344 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1345 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1346 reqsk_migrate_reset(req);
1351 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1352 reqsk_migrate_reset(nreq);
1354 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1358 /* Too bad, another child took ownership of the request, undo. */
1360 bh_unlock_sock(child);
1364 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1367 * This routine closes sockets which have been at least partially
1368 * opened, but not yet accepted.
1370 void inet_csk_listen_stop(struct sock *sk)
1372 struct inet_connection_sock *icsk = inet_csk(sk);
1373 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1374 struct request_sock *next, *req;
1376 /* Following specs, it would be better either to send FIN
1377 * (and enter FIN-WAIT-1, it is normal close)
1378 * or to send active reset (abort).
1379 * Certainly, it is pretty dangerous while synflood, but it is
1380 * bad justification for our negligence 8)
1381 * To be honest, we are not able to make either
1382 * of the variants now. --ANK
1384 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1385 struct sock *child = req->sk, *nsk;
1386 struct request_sock *nreq;
1389 bh_lock_sock(child);
1390 WARN_ON(sock_owned_by_user(child));
1393 nsk = reuseport_migrate_sock(sk, child, NULL);
1395 nreq = inet_reqsk_clone(req, nsk);
1397 refcount_set(&nreq->rsk_refcnt, 1);
1399 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1400 __NET_INC_STATS(sock_net(nsk),
1401 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1402 reqsk_migrate_reset(req);
1404 __NET_INC_STATS(sock_net(nsk),
1405 LINUX_MIB_TCPMIGRATEREQFAILURE);
1406 reqsk_migrate_reset(nreq);
1410 /* inet_csk_reqsk_queue_add() has already
1411 * called inet_child_forget() on failure case.
1413 goto skip_child_forget;
1417 inet_child_forget(sk, req, child);
1420 bh_unlock_sock(child);
1426 if (queue->fastopenq.rskq_rst_head) {
1427 /* Free all the reqs queued in rskq_rst_head. */
1428 spin_lock_bh(&queue->fastopenq.lock);
1429 req = queue->fastopenq.rskq_rst_head;
1430 queue->fastopenq.rskq_rst_head = NULL;
1431 spin_unlock_bh(&queue->fastopenq.lock);
1432 while (req != NULL) {
1433 next = req->dl_next;
1438 WARN_ON_ONCE(sk->sk_ack_backlog);
1440 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1442 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1444 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1445 const struct inet_sock *inet = inet_sk(sk);
1447 sin->sin_family = AF_INET;
1448 sin->sin_addr.s_addr = inet->inet_daddr;
1449 sin->sin_port = inet->inet_dport;
1451 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1453 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1455 const struct inet_sock *inet = inet_sk(sk);
1456 const struct ip_options_rcu *inet_opt;
1457 __be32 daddr = inet->inet_daddr;
1462 inet_opt = rcu_dereference(inet->inet_opt);
1463 if (inet_opt && inet_opt->opt.srr)
1464 daddr = inet_opt->opt.faddr;
1466 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1467 inet->inet_saddr, inet->inet_dport,
1468 inet->inet_sport, sk->sk_protocol,
1469 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1473 sk_setup_caps(sk, &rt->dst);
1479 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1481 struct dst_entry *dst = __sk_dst_check(sk, 0);
1482 struct inet_sock *inet = inet_sk(sk);
1485 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1489 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1491 dst = __sk_dst_check(sk, 0);
1493 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1497 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);