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;
121 * inet_sk_get_local_port_range - fetch ephemeral ports range
123 * @low: pointer to low port
124 * @high: pointer to high port
126 * Fetch netns port range (/proc/sys/net/ipv4/ip_local_port_range)
127 * Range can be overridden if socket got IP_LOCAL_PORT_RANGE option.
128 * Returns true if IP_LOCAL_PORT_RANGE was set on this socket.
130 bool inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
132 int lo, hi, sk_lo, sk_hi;
133 bool local_range = false;
136 inet_get_local_port_range(sock_net(sk), &lo, &hi);
138 sk_range = READ_ONCE(inet_sk(sk)->local_port_range);
139 if (unlikely(sk_range)) {
140 sk_lo = sk_range & 0xffff;
141 sk_hi = sk_range >> 16;
143 if (lo <= sk_lo && sk_lo <= hi)
145 if (lo <= sk_hi && sk_hi <= hi)
154 EXPORT_SYMBOL(inet_sk_get_local_port_range);
156 static bool inet_use_bhash2_on_bind(const struct sock *sk)
158 #if IS_ENABLED(CONFIG_IPV6)
159 if (sk->sk_family == AF_INET6) {
160 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
162 if (addr_type == IPV6_ADDR_ANY)
165 if (addr_type != IPV6_ADDR_MAPPED)
169 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
172 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
173 kuid_t sk_uid, bool relax,
174 bool reuseport_cb_ok, bool reuseport_ok)
181 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
183 if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
184 sk->sk_bound_dev_if == bound_dev_if2) {
185 if (sk->sk_reuse && sk2->sk_reuse &&
186 sk2->sk_state != TCP_LISTEN) {
187 if (!relax || (!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)))))
192 } else if (!reuseport_ok || !sk->sk_reuseport ||
193 !sk2->sk_reuseport || !reuseport_cb_ok ||
194 (sk2->sk_state != TCP_TIME_WAIT &&
195 !uid_eq(sk_uid, sock_i_uid(sk2)))) {
202 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
203 kuid_t sk_uid, bool relax,
204 bool reuseport_cb_ok, bool reuseport_ok)
206 if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
209 return inet_bind_conflict(sk, sk2, sk_uid, relax,
210 reuseport_cb_ok, reuseport_ok);
213 static bool inet_bhash2_conflict(const struct sock *sk,
214 const struct inet_bind2_bucket *tb2,
216 bool relax, bool reuseport_cb_ok,
221 sk_for_each_bound(sk2, &tb2->owners) {
222 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
223 reuseport_cb_ok, reuseport_ok))
230 #define sk_for_each_bound_bhash(__sk, __tb2, __tb) \
231 hlist_for_each_entry(__tb2, &(__tb)->bhash2, bhash_node) \
232 sk_for_each_bound(sk2, &(__tb2)->owners)
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 kuid_t uid = sock_i_uid((struct sock *)sk);
241 struct sock_reuseport *reuseport_cb;
242 bool reuseport_cb_ok;
246 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
247 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
248 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
251 /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
252 * ipv4) should have been checked already. We need to do these two
253 * checks separately because their spinlocks have to be acquired/released
254 * independently of each other, to prevent possible deadlocks
256 if (inet_use_bhash2_on_bind(sk))
257 return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax,
258 reuseport_cb_ok, reuseport_ok);
260 /* Unlike other sk lookup places we do not check
261 * for sk_net here, since _all_ the socks listed
262 * in tb->owners and tb2->owners list belong
263 * to the same net - the one this bucket belongs to.
265 sk_for_each_bound_bhash(sk2, tb2, tb) {
266 if (!inet_bind_conflict(sk, sk2, uid, relax, reuseport_cb_ok, reuseport_ok))
269 if (inet_rcv_saddr_equal(sk, sk2, true))
276 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
277 * INADDR_ANY (if ipv4) socket.
279 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
280 * against concurrent binds on the port for addr any
282 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
283 bool relax, bool reuseport_ok)
285 kuid_t uid = sock_i_uid((struct sock *)sk);
286 const struct net *net = sock_net(sk);
287 struct sock_reuseport *reuseport_cb;
288 struct inet_bind_hashbucket *head2;
289 struct inet_bind2_bucket *tb2;
290 bool reuseport_cb_ok;
293 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
294 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
295 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
298 head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
300 spin_lock(&head2->lock);
302 inet_bind_bucket_for_each(tb2, &head2->chain)
303 if (inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
306 if (tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
308 spin_unlock(&head2->lock);
312 spin_unlock(&head2->lock);
317 * Find an open port number for the socket. Returns with the
318 * inet_bind_hashbucket locks held if successful.
320 static struct inet_bind_hashbucket *
321 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
322 struct inet_bind2_bucket **tb2_ret,
323 struct inet_bind_hashbucket **head2_ret, int *port_ret)
325 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
326 int i, low, high, attempt_half, port, l3mdev;
327 struct inet_bind_hashbucket *head, *head2;
328 struct net *net = sock_net(sk);
329 struct inet_bind2_bucket *tb2;
330 struct inet_bind_bucket *tb;
331 u32 remaining, offset;
334 l3mdev = inet_sk_bound_l3mdev(sk);
336 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
338 inet_sk_get_local_port_range(sk, &low, &high);
339 high++; /* [32768, 60999] -> [32768, 61000[ */
343 int half = low + (((high - low) >> 2) << 1);
345 if (attempt_half == 1)
350 remaining = high - low;
351 if (likely(remaining > 1))
354 offset = get_random_u32_below(remaining);
355 /* __inet_hash_connect() favors ports having @low parity
356 * We do the opposite to not pollute connect() users.
362 for (i = 0; i < remaining; i += 2, port += 2) {
363 if (unlikely(port >= high))
365 if (inet_is_local_reserved_port(net, port))
367 head = &hinfo->bhash[inet_bhashfn(net, port,
369 spin_lock_bh(&head->lock);
370 if (inet_use_bhash2_on_bind(sk)) {
371 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
375 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
376 spin_lock(&head2->lock);
377 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
378 inet_bind_bucket_for_each(tb, &head->chain)
379 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
380 if (!inet_csk_bind_conflict(sk, tb, tb2,
383 spin_unlock(&head2->lock);
389 spin_unlock_bh(&head->lock);
395 goto other_parity_scan;
397 if (attempt_half == 1) {
398 /* OK we now try the upper half of the range */
400 goto other_half_scan;
403 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
404 /* We still have a chance to connect to different destinations */
406 goto ports_exhausted;
417 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
420 kuid_t uid = sock_i_uid(sk);
422 if (tb->fastreuseport <= 0)
424 if (!sk->sk_reuseport)
426 if (rcu_access_pointer(sk->sk_reuseport_cb))
428 if (!uid_eq(tb->fastuid, uid))
430 /* We only need to check the rcv_saddr if this tb was once marked
431 * without fastreuseport and then was reset, as we can only know that
432 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
435 if (tb->fastreuseport == FASTREUSEPORT_ANY)
437 #if IS_ENABLED(CONFIG_IPV6)
438 if (tb->fast_sk_family == AF_INET6)
439 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
444 ipv6_only_sock(sk), true, false);
446 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
447 ipv6_only_sock(sk), true, false);
450 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
453 kuid_t uid = sock_i_uid(sk);
454 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
456 if (hlist_empty(&tb->bhash2)) {
457 tb->fastreuse = reuse;
458 if (sk->sk_reuseport) {
459 tb->fastreuseport = FASTREUSEPORT_ANY;
461 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
462 tb->fast_ipv6_only = ipv6_only_sock(sk);
463 tb->fast_sk_family = sk->sk_family;
464 #if IS_ENABLED(CONFIG_IPV6)
465 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
468 tb->fastreuseport = 0;
473 if (sk->sk_reuseport) {
474 /* We didn't match or we don't have fastreuseport set on
475 * the tb, but we have sk_reuseport set on this socket
476 * and we know that there are no bind conflicts with
477 * this socket in this tb, so reset our tb's reuseport
478 * settings so that any subsequent sockets that match
479 * our current socket will be put on the fast path.
481 * If we reset we need to set FASTREUSEPORT_STRICT so we
482 * do extra checking for all subsequent sk_reuseport
485 if (!sk_reuseport_match(tb, sk)) {
486 tb->fastreuseport = FASTREUSEPORT_STRICT;
488 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
489 tb->fast_ipv6_only = ipv6_only_sock(sk);
490 tb->fast_sk_family = sk->sk_family;
491 #if IS_ENABLED(CONFIG_IPV6)
492 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
496 tb->fastreuseport = 0;
501 /* Obtain a reference to a local port for the given sock,
502 * if snum is zero it means select any available local port.
503 * We try to allocate an odd port (and leave even ports for connect())
505 int inet_csk_get_port(struct sock *sk, unsigned short snum)
507 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
508 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
509 bool found_port = false, check_bind_conflict = true;
510 bool bhash_created = false, bhash2_created = false;
511 int ret = -EADDRINUSE, port = snum, l3mdev;
512 struct inet_bind_hashbucket *head, *head2;
513 struct inet_bind2_bucket *tb2 = NULL;
514 struct inet_bind_bucket *tb = NULL;
515 bool head2_lock_acquired = false;
516 struct net *net = sock_net(sk);
518 l3mdev = inet_sk_bound_l3mdev(sk);
521 head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
525 head2_lock_acquired = true;
531 head = &hinfo->bhash[inet_bhashfn(net, port,
533 spin_lock_bh(&head->lock);
534 inet_bind_bucket_for_each(tb, &head->chain)
535 if (inet_bind_bucket_match(tb, net, port, l3mdev))
540 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
544 bhash_created = true;
548 if (!hlist_empty(&tb->bhash2)) {
549 if (sk->sk_reuse == SK_FORCE_REUSE ||
550 (tb->fastreuse > 0 && reuse) ||
551 sk_reuseport_match(tb, sk))
552 check_bind_conflict = false;
555 if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
556 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
560 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
561 spin_lock(&head2->lock);
562 head2_lock_acquired = true;
563 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
567 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
571 bhash2_created = true;
574 if (!found_port && check_bind_conflict) {
575 if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
580 inet_csk_update_fastreuse(tb, sk);
582 if (!inet_csk(sk)->icsk_bind_hash)
583 inet_bind_hash(sk, tb, tb2, port);
584 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
585 WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
591 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, tb2);
593 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
595 if (head2_lock_acquired)
596 spin_unlock(&head2->lock);
597 spin_unlock_bh(&head->lock);
600 EXPORT_SYMBOL_GPL(inet_csk_get_port);
603 * Wait for an incoming connection, avoid race conditions. This must be called
604 * with the socket locked.
606 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
608 struct inet_connection_sock *icsk = inet_csk(sk);
613 * True wake-one mechanism for incoming connections: only
614 * one process gets woken up, not the 'whole herd'.
615 * Since we do not 'race & poll' for established sockets
616 * anymore, the common case will execute the loop only once.
618 * Subtle issue: "add_wait_queue_exclusive()" will be added
619 * after any current non-exclusive waiters, and we know that
620 * it will always _stay_ after any new non-exclusive waiters
621 * because all non-exclusive waiters are added at the
622 * beginning of the wait-queue. As such, it's ok to "drop"
623 * our exclusiveness temporarily when we get woken up without
624 * having to remove and re-insert us on the wait queue.
627 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
630 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
631 timeo = schedule_timeout(timeo);
632 sched_annotate_sleep();
635 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
638 if (sk->sk_state != TCP_LISTEN)
640 err = sock_intr_errno(timeo);
641 if (signal_pending(current))
647 finish_wait(sk_sleep(sk), &wait);
652 * This will accept the next outstanding connection.
654 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
656 struct inet_connection_sock *icsk = inet_csk(sk);
657 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
658 struct request_sock *req;
664 /* We need to make sure that this socket is listening,
665 * and that it has something pending.
668 if (sk->sk_state != TCP_LISTEN)
671 /* Find already established connection */
672 if (reqsk_queue_empty(queue)) {
673 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
675 /* If this is a non blocking socket don't sleep */
680 error = inet_csk_wait_for_connect(sk, timeo);
684 req = reqsk_queue_remove(queue, sk);
687 if (sk->sk_protocol == IPPROTO_TCP &&
688 tcp_rsk(req)->tfo_listener) {
689 spin_lock_bh(&queue->fastopenq.lock);
690 if (tcp_rsk(req)->tfo_listener) {
691 /* We are still waiting for the final ACK from 3WHS
692 * so can't free req now. Instead, we set req->sk to
693 * NULL to signify that the child socket is taken
694 * so reqsk_fastopen_remove() will free the req
695 * when 3WHS finishes (or is aborted).
700 spin_unlock_bh(&queue->fastopenq.lock);
705 if (newsk && mem_cgroup_sockets_enabled) {
708 /* atomically get the memory usage, set and charge the
713 mem_cgroup_sk_alloc(newsk);
714 if (newsk->sk_memcg) {
715 /* The socket has not been accepted yet, no need
716 * to look at newsk->sk_wmem_queued.
718 amt = sk_mem_pages(newsk->sk_forward_alloc +
719 atomic_read(&newsk->sk_rmem_alloc));
723 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
724 GFP_KERNEL | __GFP_NOFAIL);
737 EXPORT_SYMBOL(inet_csk_accept);
740 * Using different timers for retransmit, delayed acks and probes
741 * We may wish use just one timer maintaining a list of expire jiffies
744 void inet_csk_init_xmit_timers(struct sock *sk,
745 void (*retransmit_handler)(struct timer_list *t),
746 void (*delack_handler)(struct timer_list *t),
747 void (*keepalive_handler)(struct timer_list *t))
749 struct inet_connection_sock *icsk = inet_csk(sk);
751 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
752 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
753 timer_setup(&sk->sk_timer, keepalive_handler, 0);
754 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
756 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
758 void inet_csk_clear_xmit_timers(struct sock *sk)
760 struct inet_connection_sock *icsk = inet_csk(sk);
762 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
764 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
765 sk_stop_timer(sk, &icsk->icsk_delack_timer);
766 sk_stop_timer(sk, &sk->sk_timer);
768 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
770 void inet_csk_delete_keepalive_timer(struct sock *sk)
772 sk_stop_timer(sk, &sk->sk_timer);
774 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
776 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
778 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
780 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
782 struct dst_entry *inet_csk_route_req(const struct sock *sk,
784 const struct request_sock *req)
786 const struct inet_request_sock *ireq = inet_rsk(req);
787 struct net *net = read_pnet(&ireq->ireq_net);
788 struct ip_options_rcu *opt;
792 opt = rcu_dereference(ireq->ireq_opt);
794 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
795 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
796 sk->sk_protocol, inet_sk_flowi_flags(sk),
797 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
798 ireq->ir_loc_addr, ireq->ir_rmt_port,
799 htons(ireq->ir_num), sk->sk_uid);
800 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
801 rt = ip_route_output_flow(net, fl4, sk);
804 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
813 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
816 EXPORT_SYMBOL_GPL(inet_csk_route_req);
818 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
820 const struct request_sock *req)
822 const struct inet_request_sock *ireq = inet_rsk(req);
823 struct net *net = read_pnet(&ireq->ireq_net);
824 struct inet_sock *newinet = inet_sk(newsk);
825 struct ip_options_rcu *opt;
829 opt = rcu_dereference(ireq->ireq_opt);
830 fl4 = &newinet->cork.fl.u.ip4;
832 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
833 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
834 sk->sk_protocol, inet_sk_flowi_flags(sk),
835 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
836 ireq->ir_loc_addr, ireq->ir_rmt_port,
837 htons(ireq->ir_num), sk->sk_uid);
838 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
839 rt = ip_route_output_flow(net, fl4, sk);
842 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
849 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
852 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
854 /* Decide when to expire the request and when to resend SYN-ACK */
855 static void syn_ack_recalc(struct request_sock *req,
856 const int max_syn_ack_retries,
857 const u8 rskq_defer_accept,
858 int *expire, int *resend)
860 if (!rskq_defer_accept) {
861 *expire = req->num_timeout >= max_syn_ack_retries;
865 *expire = req->num_timeout >= max_syn_ack_retries &&
866 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
867 /* Do not resend while waiting for data after ACK,
868 * start to resend on end of deferring period to give
869 * last chance for data or ACK to create established socket.
871 *resend = !inet_rsk(req)->acked ||
872 req->num_timeout >= rskq_defer_accept - 1;
875 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
877 int err = req->rsk_ops->rtx_syn_ack(parent, req);
883 EXPORT_SYMBOL(inet_rtx_syn_ack);
885 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
888 struct sock *req_sk, *nreq_sk;
889 struct request_sock *nreq;
891 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
893 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
895 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
900 req_sk = req_to_sk(req);
901 nreq_sk = req_to_sk(nreq);
903 memcpy(nreq_sk, req_sk,
904 offsetof(struct sock, sk_dontcopy_begin));
905 memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
906 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
908 sk_node_init(&nreq_sk->sk_node);
909 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
910 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
911 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
913 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
915 nreq->rsk_listener = sk;
917 /* We need not acquire fastopenq->lock
918 * because the child socket is locked in inet_csk_listen_stop().
920 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
921 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
926 static void reqsk_queue_migrated(struct request_sock_queue *queue,
927 const struct request_sock *req)
929 if (req->num_timeout == 0)
930 atomic_inc(&queue->young);
931 atomic_inc(&queue->qlen);
934 static void reqsk_migrate_reset(struct request_sock *req)
936 req->saved_syn = NULL;
937 #if IS_ENABLED(CONFIG_IPV6)
938 inet_rsk(req)->ipv6_opt = NULL;
939 inet_rsk(req)->pktopts = NULL;
941 inet_rsk(req)->ireq_opt = NULL;
945 /* return true if req was found in the ehash table */
946 static bool reqsk_queue_unlink(struct request_sock *req)
948 struct sock *sk = req_to_sk(req);
952 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
953 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
956 found = __sk_nulls_del_node_init_rcu(sk);
959 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
964 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
966 bool unlinked = reqsk_queue_unlink(req);
969 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
974 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
976 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
978 inet_csk_reqsk_queue_drop(sk, req);
981 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
983 static void reqsk_timer_handler(struct timer_list *t)
985 struct request_sock *req = from_timer(req, t, rsk_timer);
986 struct request_sock *nreq = NULL, *oreq = req;
987 struct sock *sk_listener = req->rsk_listener;
988 struct inet_connection_sock *icsk;
989 struct request_sock_queue *queue;
991 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
993 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
996 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1000 nreq = inet_reqsk_clone(req, nsk);
1004 /* The new timer for the cloned req can decrease the 2
1005 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1006 * hold another count to prevent use-after-free and
1007 * call reqsk_put() just before return.
1009 refcount_set(&nreq->rsk_refcnt, 2 + 1);
1010 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1011 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1017 icsk = inet_csk(sk_listener);
1018 net = sock_net(sk_listener);
1019 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1020 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1021 /* Normally all the openreqs are young and become mature
1022 * (i.e. converted to established socket) for first timeout.
1023 * If synack was not acknowledged for 1 second, it means
1024 * one of the following things: synack was lost, ack was lost,
1025 * rtt is high or nobody planned to ack (i.e. synflood).
1026 * When server is a bit loaded, queue is populated with old
1027 * open requests, reducing effective size of queue.
1028 * When server is well loaded, queue size reduces to zero
1029 * after several minutes of work. It is not synflood,
1030 * it is normal operation. The solution is pruning
1031 * too old entries overriding normal timeout, when
1032 * situation becomes dangerous.
1034 * Essentially, we reserve half of room for young
1035 * embrions; and abort old ones without pity, if old
1036 * ones are about to clog our table.
1038 queue = &icsk->icsk_accept_queue;
1039 qlen = reqsk_queue_len(queue);
1040 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1041 int young = reqsk_queue_len_young(queue) << 1;
1043 while (max_syn_ack_retries > 2) {
1046 max_syn_ack_retries--;
1050 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1052 req->rsk_ops->syn_ack_timeout(req);
1055 !inet_rtx_syn_ack(sk_listener, req) ||
1056 inet_rsk(req)->acked)) {
1057 if (req->num_timeout++ == 0)
1058 atomic_dec(&queue->young);
1059 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1064 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1066 inet_csk_reqsk_queue_drop(sk_listener, nreq);
1070 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1071 reqsk_migrate_reset(oreq);
1072 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1079 /* Even if we can clone the req, we may need not retransmit any more
1080 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1081 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1084 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1086 reqsk_migrate_reset(nreq);
1087 reqsk_queue_removed(queue, nreq);
1092 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1095 static void reqsk_queue_hash_req(struct request_sock *req,
1096 unsigned long timeout)
1098 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1099 mod_timer(&req->rsk_timer, jiffies + timeout);
1101 inet_ehash_insert(req_to_sk(req), NULL, NULL);
1102 /* before letting lookups find us, make sure all req fields
1103 * are committed to memory and refcnt initialized.
1106 refcount_set(&req->rsk_refcnt, 2 + 1);
1109 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1110 unsigned long timeout)
1112 reqsk_queue_hash_req(req, timeout);
1113 inet_csk_reqsk_queue_added(sk);
1115 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1117 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1118 const gfp_t priority)
1120 struct inet_connection_sock *icsk = inet_csk(newsk);
1122 if (!icsk->icsk_ulp_ops)
1125 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1129 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1130 * @sk: the socket to clone
1131 * @req: request_sock
1132 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1134 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1136 struct sock *inet_csk_clone_lock(const struct sock *sk,
1137 const struct request_sock *req,
1138 const gfp_t priority)
1140 struct sock *newsk = sk_clone_lock(sk, priority);
1143 struct inet_connection_sock *newicsk = inet_csk(newsk);
1145 inet_sk_set_state(newsk, TCP_SYN_RECV);
1146 newicsk->icsk_bind_hash = NULL;
1147 newicsk->icsk_bind2_hash = NULL;
1149 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1150 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1151 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1153 /* listeners have SOCK_RCU_FREE, not the children */
1154 sock_reset_flag(newsk, SOCK_RCU_FREE);
1156 inet_sk(newsk)->mc_list = NULL;
1158 newsk->sk_mark = inet_rsk(req)->ir_mark;
1159 atomic64_set(&newsk->sk_cookie,
1160 atomic64_read(&inet_rsk(req)->ir_cookie));
1162 newicsk->icsk_retransmits = 0;
1163 newicsk->icsk_backoff = 0;
1164 newicsk->icsk_probes_out = 0;
1165 newicsk->icsk_probes_tstamp = 0;
1167 /* Deinitialize accept_queue to trap illegal accesses. */
1168 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1170 inet_clone_ulp(req, newsk, priority);
1172 security_inet_csk_clone(newsk, req);
1176 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1179 * At this point, there should be no process reference to this
1180 * socket, and thus no user references at all. Therefore we
1181 * can assume the socket waitqueue is inactive and nobody will
1182 * try to jump onto it.
1184 void inet_csk_destroy_sock(struct sock *sk)
1186 WARN_ON(sk->sk_state != TCP_CLOSE);
1187 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1189 /* It cannot be in hash table! */
1190 WARN_ON(!sk_unhashed(sk));
1192 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1193 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1195 sk->sk_prot->destroy(sk);
1197 sk_stream_kill_queues(sk);
1199 xfrm_sk_free_policy(sk);
1201 this_cpu_dec(*sk->sk_prot->orphan_count);
1205 EXPORT_SYMBOL(inet_csk_destroy_sock);
1207 /* This function allows to force a closure of a socket after the call to
1208 * tcp/dccp_create_openreq_child().
1210 void inet_csk_prepare_forced_close(struct sock *sk)
1211 __releases(&sk->sk_lock.slock)
1213 /* sk_clone_lock locked the socket and set refcnt to 2 */
1216 inet_csk_prepare_for_destroy_sock(sk);
1217 inet_sk(sk)->inet_num = 0;
1219 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1221 static int inet_ulp_can_listen(const struct sock *sk)
1223 const struct inet_connection_sock *icsk = inet_csk(sk);
1225 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1231 int inet_csk_listen_start(struct sock *sk)
1233 struct inet_connection_sock *icsk = inet_csk(sk);
1234 struct inet_sock *inet = inet_sk(sk);
1237 err = inet_ulp_can_listen(sk);
1241 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1243 sk->sk_ack_backlog = 0;
1244 inet_csk_delack_init(sk);
1246 /* There is race window here: we announce ourselves listening,
1247 * but this transition is still not validated by get_port().
1248 * It is OK, because this socket enters to hash table only
1249 * after validation is complete.
1251 inet_sk_state_store(sk, TCP_LISTEN);
1252 err = sk->sk_prot->get_port(sk, inet->inet_num);
1254 inet->inet_sport = htons(inet->inet_num);
1257 err = sk->sk_prot->hash(sk);
1263 inet_sk_set_state(sk, TCP_CLOSE);
1266 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1268 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1271 sk->sk_prot->disconnect(child, O_NONBLOCK);
1275 this_cpu_inc(*sk->sk_prot->orphan_count);
1277 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1278 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1279 BUG_ON(sk != req->rsk_listener);
1281 /* Paranoid, to prevent race condition if
1282 * an inbound pkt destined for child is
1283 * blocked by sock lock in tcp_v4_rcv().
1284 * Also to satisfy an assertion in
1285 * tcp_v4_destroy_sock().
1287 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1289 inet_csk_destroy_sock(child);
1292 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1293 struct request_sock *req,
1296 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1298 spin_lock(&queue->rskq_lock);
1299 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1300 inet_child_forget(sk, req, child);
1304 req->dl_next = NULL;
1305 if (queue->rskq_accept_head == NULL)
1306 WRITE_ONCE(queue->rskq_accept_head, req);
1308 queue->rskq_accept_tail->dl_next = req;
1309 queue->rskq_accept_tail = req;
1310 sk_acceptq_added(sk);
1312 spin_unlock(&queue->rskq_lock);
1315 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1317 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1318 struct request_sock *req, bool own_req)
1321 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1322 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1324 if (sk != req->rsk_listener) {
1325 /* another listening sk has been selected,
1326 * migrate the req to it.
1328 struct request_sock *nreq;
1330 /* hold a refcnt for the nreq->rsk_listener
1331 * which is assigned in inet_reqsk_clone()
1334 nreq = inet_reqsk_clone(req, sk);
1336 inet_child_forget(sk, req, child);
1340 refcount_set(&nreq->rsk_refcnt, 1);
1341 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1342 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1343 reqsk_migrate_reset(req);
1348 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1349 reqsk_migrate_reset(nreq);
1351 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1355 /* Too bad, another child took ownership of the request, undo. */
1357 bh_unlock_sock(child);
1361 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1364 * This routine closes sockets which have been at least partially
1365 * opened, but not yet accepted.
1367 void inet_csk_listen_stop(struct sock *sk)
1369 struct inet_connection_sock *icsk = inet_csk(sk);
1370 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1371 struct request_sock *next, *req;
1373 /* Following specs, it would be better either to send FIN
1374 * (and enter FIN-WAIT-1, it is normal close)
1375 * or to send active reset (abort).
1376 * Certainly, it is pretty dangerous while synflood, but it is
1377 * bad justification for our negligence 8)
1378 * To be honest, we are not able to make either
1379 * of the variants now. --ANK
1381 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1382 struct sock *child = req->sk, *nsk;
1383 struct request_sock *nreq;
1386 bh_lock_sock(child);
1387 WARN_ON(sock_owned_by_user(child));
1390 nsk = reuseport_migrate_sock(sk, child, NULL);
1392 nreq = inet_reqsk_clone(req, nsk);
1394 refcount_set(&nreq->rsk_refcnt, 1);
1396 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1397 __NET_INC_STATS(sock_net(nsk),
1398 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1399 reqsk_migrate_reset(req);
1401 __NET_INC_STATS(sock_net(nsk),
1402 LINUX_MIB_TCPMIGRATEREQFAILURE);
1403 reqsk_migrate_reset(nreq);
1407 /* inet_csk_reqsk_queue_add() has already
1408 * called inet_child_forget() on failure case.
1410 goto skip_child_forget;
1414 inet_child_forget(sk, req, child);
1417 bh_unlock_sock(child);
1423 if (queue->fastopenq.rskq_rst_head) {
1424 /* Free all the reqs queued in rskq_rst_head. */
1425 spin_lock_bh(&queue->fastopenq.lock);
1426 req = queue->fastopenq.rskq_rst_head;
1427 queue->fastopenq.rskq_rst_head = NULL;
1428 spin_unlock_bh(&queue->fastopenq.lock);
1429 while (req != NULL) {
1430 next = req->dl_next;
1435 WARN_ON_ONCE(sk->sk_ack_backlog);
1437 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1439 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1441 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1442 const struct inet_sock *inet = inet_sk(sk);
1444 sin->sin_family = AF_INET;
1445 sin->sin_addr.s_addr = inet->inet_daddr;
1446 sin->sin_port = inet->inet_dport;
1448 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1450 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1452 const struct inet_sock *inet = inet_sk(sk);
1453 const struct ip_options_rcu *inet_opt;
1454 __be32 daddr = inet->inet_daddr;
1459 inet_opt = rcu_dereference(inet->inet_opt);
1460 if (inet_opt && inet_opt->opt.srr)
1461 daddr = inet_opt->opt.faddr;
1463 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1464 inet->inet_saddr, inet->inet_dport,
1465 inet->inet_sport, sk->sk_protocol,
1466 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1470 sk_setup_caps(sk, &rt->dst);
1476 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1478 struct dst_entry *dst = __sk_dst_check(sk, 0);
1479 struct inet_sock *inet = inet_sk(sk);
1482 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1486 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1488 dst = __sk_dst_check(sk, 0);
1490 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1494 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);