2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Support for INET connection oriented protocols.
8 * Authors: See the TCP sources
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or(at your option) any later version.
16 #include <linux/module.h>
17 #include <linux/jhash.h>
19 #include <net/inet_connection_sock.h>
20 #include <net/inet_hashtables.h>
21 #include <net/inet_timewait_sock.h>
23 #include <net/route.h>
24 #include <net/tcp_states.h>
27 #include <net/sock_reuseport.h>
28 #include <net/addrconf.h>
31 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
32 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
35 #if IS_ENABLED(CONFIG_IPV6)
36 /* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
37 * only, and any IPv4 addresses if not IPv6 only
38 * match_wildcard == false: addresses must be exactly the same, i.e.
39 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
40 * and 0.0.0.0 equals to 0.0.0.0 only
42 static int ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
43 const struct in6_addr *sk2_rcv_saddr6,
44 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
45 bool sk1_ipv6only, bool sk2_ipv6only,
48 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
49 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
51 /* if both are mapped, treat as IPv4 */
52 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
54 if (sk1_rcv_saddr == sk2_rcv_saddr)
56 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
57 return match_wildcard;
62 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
65 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
66 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
69 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
70 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
74 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
81 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
82 * match_wildcard == false: addresses must be exactly the same, i.e.
83 * 0.0.0.0 only equals to 0.0.0.0
85 static int ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
86 bool sk2_ipv6only, bool match_wildcard)
89 if (sk1_rcv_saddr == sk2_rcv_saddr)
91 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
92 return match_wildcard;
97 int inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
100 #if IS_ENABLED(CONFIG_IPV6)
101 if (sk->sk_family == AF_INET6)
102 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
103 inet6_rcv_saddr(sk2),
110 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
111 ipv6_only_sock(sk2), match_wildcard);
113 EXPORT_SYMBOL(inet_rcv_saddr_equal);
115 void inet_get_local_port_range(struct net *net, int *low, int *high)
120 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
122 *low = net->ipv4.ip_local_ports.range[0];
123 *high = net->ipv4.ip_local_ports.range[1];
124 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
126 EXPORT_SYMBOL(inet_get_local_port_range);
128 static int inet_csk_bind_conflict(const struct sock *sk,
129 const struct inet_bind_bucket *tb,
130 bool relax, bool reuseport_ok)
133 bool reuse = sk->sk_reuse;
134 bool reuseport = !!sk->sk_reuseport && reuseport_ok;
135 kuid_t uid = sock_i_uid((struct sock *)sk);
138 * Unlike other sk lookup places we do not check
139 * for sk_net here, since _all_ the socks listed
140 * in tb->owners list belong to the same net - the
141 * one this bucket belongs to.
144 sk_for_each_bound(sk2, &tb->owners) {
146 (!sk->sk_bound_dev_if ||
147 !sk2->sk_bound_dev_if ||
148 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
149 if ((!reuse || !sk2->sk_reuse ||
150 sk2->sk_state == TCP_LISTEN) &&
151 (!reuseport || !sk2->sk_reuseport ||
152 rcu_access_pointer(sk->sk_reuseport_cb) ||
153 (sk2->sk_state != TCP_TIME_WAIT &&
154 !uid_eq(uid, sock_i_uid(sk2))))) {
155 if (inet_rcv_saddr_equal(sk, sk2, true))
158 if (!relax && reuse && sk2->sk_reuse &&
159 sk2->sk_state != TCP_LISTEN) {
160 if (inet_rcv_saddr_equal(sk, sk2, true))
169 * Find an open port number for the socket. Returns with the
170 * inet_bind_hashbucket lock held.
172 static struct inet_bind_hashbucket *
173 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
175 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
177 struct inet_bind_hashbucket *head;
178 struct net *net = sock_net(sk);
179 int i, low, high, attempt_half;
180 struct inet_bind_bucket *tb;
181 u32 remaining, offset;
183 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
185 inet_get_local_port_range(net, &low, &high);
186 high++; /* [32768, 60999] -> [32768, 61000[ */
190 int half = low + (((high - low) >> 2) << 1);
192 if (attempt_half == 1)
197 remaining = high - low;
198 if (likely(remaining > 1))
201 offset = prandom_u32() % remaining;
202 /* __inet_hash_connect() favors ports having @low parity
203 * We do the opposite to not pollute connect() users.
209 for (i = 0; i < remaining; i += 2, port += 2) {
210 if (unlikely(port >= high))
212 if (inet_is_local_reserved_port(net, port))
214 head = &hinfo->bhash[inet_bhashfn(net, port,
216 spin_lock_bh(&head->lock);
217 inet_bind_bucket_for_each(tb, &head->chain)
218 if (net_eq(ib_net(tb), net) && tb->port == port) {
219 if (!inet_csk_bind_conflict(sk, tb, false, false))
226 spin_unlock_bh(&head->lock);
232 goto other_parity_scan;
234 if (attempt_half == 1) {
235 /* OK we now try the upper half of the range */
237 goto other_half_scan;
246 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
249 kuid_t uid = sock_i_uid(sk);
251 if (tb->fastreuseport <= 0)
253 if (!sk->sk_reuseport)
255 if (rcu_access_pointer(sk->sk_reuseport_cb))
257 if (!uid_eq(tb->fastuid, uid))
259 /* We only need to check the rcv_saddr if this tb was once marked
260 * without fastreuseport and then was reset, as we can only know that
261 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
264 if (tb->fastreuseport == FASTREUSEPORT_ANY)
266 #if IS_ENABLED(CONFIG_IPV6)
267 if (tb->fast_sk_family == AF_INET6)
268 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
273 ipv6_only_sock(sk), true);
275 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
276 ipv6_only_sock(sk), true);
279 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
282 kuid_t uid = sock_i_uid(sk);
283 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
285 if (hlist_empty(&tb->owners)) {
286 tb->fastreuse = reuse;
287 if (sk->sk_reuseport) {
288 tb->fastreuseport = FASTREUSEPORT_ANY;
290 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
291 tb->fast_ipv6_only = ipv6_only_sock(sk);
292 tb->fast_sk_family = sk->sk_family;
293 #if IS_ENABLED(CONFIG_IPV6)
294 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
297 tb->fastreuseport = 0;
302 if (sk->sk_reuseport) {
303 /* We didn't match or we don't have fastreuseport set on
304 * the tb, but we have sk_reuseport set on this socket
305 * and we know that there are no bind conflicts with
306 * this socket in this tb, so reset our tb's reuseport
307 * settings so that any subsequent sockets that match
308 * our current socket will be put on the fast path.
310 * If we reset we need to set FASTREUSEPORT_STRICT so we
311 * do extra checking for all subsequent sk_reuseport
314 if (!sk_reuseport_match(tb, sk)) {
315 tb->fastreuseport = FASTREUSEPORT_STRICT;
317 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
318 tb->fast_ipv6_only = ipv6_only_sock(sk);
319 tb->fast_sk_family = sk->sk_family;
320 #if IS_ENABLED(CONFIG_IPV6)
321 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
325 tb->fastreuseport = 0;
330 /* Obtain a reference to a local port for the given sock,
331 * if snum is zero it means select any available local port.
332 * We try to allocate an odd port (and leave even ports for connect())
334 int inet_csk_get_port(struct sock *sk, unsigned short snum)
336 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
337 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
338 int ret = 1, port = snum;
339 struct inet_bind_hashbucket *head;
340 struct net *net = sock_net(sk);
341 struct inet_bind_bucket *tb = NULL;
344 head = inet_csk_find_open_port(sk, &tb, &port);
351 head = &hinfo->bhash[inet_bhashfn(net, port,
353 spin_lock_bh(&head->lock);
354 inet_bind_bucket_for_each(tb, &head->chain)
355 if (net_eq(ib_net(tb), net) && tb->port == port)
358 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
363 if (!hlist_empty(&tb->owners)) {
364 if (sk->sk_reuse == SK_FORCE_REUSE)
367 if ((tb->fastreuse > 0 && reuse) ||
368 sk_reuseport_match(tb, sk))
370 if (inet_csk_bind_conflict(sk, tb, true, true))
374 inet_csk_update_fastreuse(tb, sk);
376 if (!inet_csk(sk)->icsk_bind_hash)
377 inet_bind_hash(sk, tb, port);
378 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
382 spin_unlock_bh(&head->lock);
385 EXPORT_SYMBOL_GPL(inet_csk_get_port);
388 * Wait for an incoming connection, avoid race conditions. This must be called
389 * with the socket locked.
391 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
393 struct inet_connection_sock *icsk = inet_csk(sk);
398 * True wake-one mechanism for incoming connections: only
399 * one process gets woken up, not the 'whole herd'.
400 * Since we do not 'race & poll' for established sockets
401 * anymore, the common case will execute the loop only once.
403 * Subtle issue: "add_wait_queue_exclusive()" will be added
404 * after any current non-exclusive waiters, and we know that
405 * it will always _stay_ after any new non-exclusive waiters
406 * because all non-exclusive waiters are added at the
407 * beginning of the wait-queue. As such, it's ok to "drop"
408 * our exclusiveness temporarily when we get woken up without
409 * having to remove and re-insert us on the wait queue.
412 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
415 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
416 timeo = schedule_timeout(timeo);
417 sched_annotate_sleep();
420 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
423 if (sk->sk_state != TCP_LISTEN)
425 err = sock_intr_errno(timeo);
426 if (signal_pending(current))
432 finish_wait(sk_sleep(sk), &wait);
437 * This will accept the next outstanding connection.
439 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
441 struct inet_connection_sock *icsk = inet_csk(sk);
442 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
443 struct request_sock *req;
449 /* We need to make sure that this socket is listening,
450 * and that it has something pending.
453 if (sk->sk_state != TCP_LISTEN)
456 /* Find already established connection */
457 if (reqsk_queue_empty(queue)) {
458 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
460 /* If this is a non blocking socket don't sleep */
465 error = inet_csk_wait_for_connect(sk, timeo);
469 req = reqsk_queue_remove(queue, sk);
472 if (sk->sk_protocol == IPPROTO_TCP &&
473 tcp_rsk(req)->tfo_listener) {
474 spin_lock_bh(&queue->fastopenq.lock);
475 if (tcp_rsk(req)->tfo_listener) {
476 /* We are still waiting for the final ACK from 3WHS
477 * so can't free req now. Instead, we set req->sk to
478 * NULL to signify that the child socket is taken
479 * so reqsk_fastopen_remove() will free the req
480 * when 3WHS finishes (or is aborted).
485 spin_unlock_bh(&queue->fastopenq.lock);
490 if (newsk && mem_cgroup_sockets_enabled) {
493 /* atomically get the memory usage, set and charge the
498 /* The socket has not been accepted yet, no need to look at
499 * newsk->sk_wmem_queued.
501 amt = sk_mem_pages(newsk->sk_forward_alloc +
502 atomic_read(&newsk->sk_rmem_alloc));
503 mem_cgroup_sk_alloc(newsk);
504 if (newsk->sk_memcg && amt)
505 mem_cgroup_charge_skmem(newsk->sk_memcg, amt);
518 EXPORT_SYMBOL(inet_csk_accept);
521 * Using different timers for retransmit, delayed acks and probes
522 * We may wish use just one timer maintaining a list of expire jiffies
525 void inet_csk_init_xmit_timers(struct sock *sk,
526 void (*retransmit_handler)(unsigned long),
527 void (*delack_handler)(unsigned long),
528 void (*keepalive_handler)(unsigned long))
530 struct inet_connection_sock *icsk = inet_csk(sk);
532 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
534 setup_timer(&icsk->icsk_delack_timer, delack_handler,
536 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
537 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
539 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
541 void inet_csk_clear_xmit_timers(struct sock *sk)
543 struct inet_connection_sock *icsk = inet_csk(sk);
545 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
547 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
548 sk_stop_timer(sk, &icsk->icsk_delack_timer);
549 sk_stop_timer(sk, &sk->sk_timer);
551 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
553 void inet_csk_delete_keepalive_timer(struct sock *sk)
555 sk_stop_timer(sk, &sk->sk_timer);
557 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
559 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
561 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
563 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
565 struct dst_entry *inet_csk_route_req(const struct sock *sk,
567 const struct request_sock *req)
569 const struct inet_request_sock *ireq = inet_rsk(req);
570 struct net *net = read_pnet(&ireq->ireq_net);
571 struct ip_options_rcu *opt;
575 opt = rcu_dereference(ireq->ireq_opt);
577 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
578 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
579 sk->sk_protocol, inet_sk_flowi_flags(sk),
580 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
581 ireq->ir_loc_addr, ireq->ir_rmt_port,
582 htons(ireq->ir_num), sk->sk_uid);
583 security_req_classify_flow(req, flowi4_to_flowi(fl4));
584 rt = ip_route_output_flow(net, fl4, sk);
587 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
596 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
599 EXPORT_SYMBOL_GPL(inet_csk_route_req);
601 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
603 const struct request_sock *req)
605 const struct inet_request_sock *ireq = inet_rsk(req);
606 struct net *net = read_pnet(&ireq->ireq_net);
607 struct inet_sock *newinet = inet_sk(newsk);
608 struct ip_options_rcu *opt;
612 opt = rcu_dereference(ireq->ireq_opt);
613 fl4 = &newinet->cork.fl.u.ip4;
615 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
616 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
617 sk->sk_protocol, inet_sk_flowi_flags(sk),
618 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
619 ireq->ir_loc_addr, ireq->ir_rmt_port,
620 htons(ireq->ir_num), sk->sk_uid);
621 security_req_classify_flow(req, flowi4_to_flowi(fl4));
622 rt = ip_route_output_flow(net, fl4, sk);
625 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
632 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
635 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
637 #if IS_ENABLED(CONFIG_IPV6)
638 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
640 #define AF_INET_FAMILY(fam) true
643 /* Decide when to expire the request and when to resend SYN-ACK */
644 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
645 const int max_retries,
646 const u8 rskq_defer_accept,
647 int *expire, int *resend)
649 if (!rskq_defer_accept) {
650 *expire = req->num_timeout >= thresh;
654 *expire = req->num_timeout >= thresh &&
655 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
657 * Do not resend while waiting for data after ACK,
658 * start to resend on end of deferring period to give
659 * last chance for data or ACK to create established socket.
661 *resend = !inet_rsk(req)->acked ||
662 req->num_timeout >= rskq_defer_accept - 1;
665 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
667 int err = req->rsk_ops->rtx_syn_ack(parent, req);
673 EXPORT_SYMBOL(inet_rtx_syn_ack);
675 /* return true if req was found in the ehash table */
676 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
677 struct request_sock *req)
679 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
682 if (sk_hashed(req_to_sk(req))) {
683 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
686 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
689 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
694 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
696 if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
697 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
701 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
703 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
705 inet_csk_reqsk_queue_drop(sk, req);
708 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
710 static void reqsk_timer_handler(unsigned long data)
712 struct request_sock *req = (struct request_sock *)data;
713 struct sock *sk_listener = req->rsk_listener;
714 struct net *net = sock_net(sk_listener);
715 struct inet_connection_sock *icsk = inet_csk(sk_listener);
716 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
717 int qlen, expire = 0, resend = 0;
718 int max_retries, thresh;
721 if (sk_state_load(sk_listener) != TCP_LISTEN)
724 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
725 thresh = max_retries;
726 /* Normally all the openreqs are young and become mature
727 * (i.e. converted to established socket) for first timeout.
728 * If synack was not acknowledged for 1 second, it means
729 * one of the following things: synack was lost, ack was lost,
730 * rtt is high or nobody planned to ack (i.e. synflood).
731 * When server is a bit loaded, queue is populated with old
732 * open requests, reducing effective size of queue.
733 * When server is well loaded, queue size reduces to zero
734 * after several minutes of work. It is not synflood,
735 * it is normal operation. The solution is pruning
736 * too old entries overriding normal timeout, when
737 * situation becomes dangerous.
739 * Essentially, we reserve half of room for young
740 * embrions; and abort old ones without pity, if old
741 * ones are about to clog our table.
743 qlen = reqsk_queue_len(queue);
744 if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
745 int young = reqsk_queue_len_young(queue) << 1;
754 defer_accept = READ_ONCE(queue->rskq_defer_accept);
756 max_retries = defer_accept;
757 syn_ack_recalc(req, thresh, max_retries, defer_accept,
759 req->rsk_ops->syn_ack_timeout(req);
762 !inet_rtx_syn_ack(sk_listener, req) ||
763 inet_rsk(req)->acked)) {
766 if (req->num_timeout++ == 0)
767 atomic_dec(&queue->young);
768 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
769 mod_timer(&req->rsk_timer, jiffies + timeo);
773 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
776 static void reqsk_queue_hash_req(struct request_sock *req,
777 unsigned long timeout)
779 req->num_retrans = 0;
780 req->num_timeout = 0;
783 setup_pinned_timer(&req->rsk_timer, reqsk_timer_handler,
785 mod_timer(&req->rsk_timer, jiffies + timeout);
787 inet_ehash_insert(req_to_sk(req), NULL, NULL);
788 /* before letting lookups find us, make sure all req fields
789 * are committed to memory and refcnt initialized.
792 refcount_set(&req->rsk_refcnt, 2 + 1);
795 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
796 unsigned long timeout)
798 reqsk_queue_hash_req(req, timeout);
799 inet_csk_reqsk_queue_added(sk);
801 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
804 * inet_csk_clone_lock - clone an inet socket, and lock its clone
805 * @sk: the socket to clone
807 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
809 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
811 struct sock *inet_csk_clone_lock(const struct sock *sk,
812 const struct request_sock *req,
813 const gfp_t priority)
815 struct sock *newsk = sk_clone_lock(sk, priority);
818 struct inet_connection_sock *newicsk = inet_csk(newsk);
820 newsk->sk_state = TCP_SYN_RECV;
821 newicsk->icsk_bind_hash = NULL;
823 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
824 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
825 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
827 /* listeners have SOCK_RCU_FREE, not the children */
828 sock_reset_flag(newsk, SOCK_RCU_FREE);
830 inet_sk(newsk)->mc_list = NULL;
832 newsk->sk_mark = inet_rsk(req)->ir_mark;
833 atomic64_set(&newsk->sk_cookie,
834 atomic64_read(&inet_rsk(req)->ir_cookie));
836 newicsk->icsk_retransmits = 0;
837 newicsk->icsk_backoff = 0;
838 newicsk->icsk_probes_out = 0;
840 /* Deinitialize accept_queue to trap illegal accesses. */
841 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
843 security_inet_csk_clone(newsk, req);
847 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
850 * At this point, there should be no process reference to this
851 * socket, and thus no user references at all. Therefore we
852 * can assume the socket waitqueue is inactive and nobody will
853 * try to jump onto it.
855 void inet_csk_destroy_sock(struct sock *sk)
857 WARN_ON(sk->sk_state != TCP_CLOSE);
858 WARN_ON(!sock_flag(sk, SOCK_DEAD));
860 /* It cannot be in hash table! */
861 WARN_ON(!sk_unhashed(sk));
863 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
864 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
866 sk->sk_prot->destroy(sk);
868 sk_stream_kill_queues(sk);
870 xfrm_sk_free_policy(sk);
872 sk_refcnt_debug_release(sk);
874 percpu_counter_dec(sk->sk_prot->orphan_count);
878 EXPORT_SYMBOL(inet_csk_destroy_sock);
880 /* This function allows to force a closure of a socket after the call to
881 * tcp/dccp_create_openreq_child().
883 void inet_csk_prepare_forced_close(struct sock *sk)
884 __releases(&sk->sk_lock.slock)
886 /* sk_clone_lock locked the socket and set refcnt to 2 */
890 /* The below has to be done to allow calling inet_csk_destroy_sock */
891 sock_set_flag(sk, SOCK_DEAD);
892 percpu_counter_inc(sk->sk_prot->orphan_count);
893 inet_sk(sk)->inet_num = 0;
895 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
897 static int inet_ulp_can_listen(const struct sock *sk)
899 const struct inet_connection_sock *icsk = inet_csk(sk);
901 if (icsk->icsk_ulp_ops)
907 int inet_csk_listen_start(struct sock *sk, int backlog)
909 struct inet_connection_sock *icsk = inet_csk(sk);
910 struct inet_sock *inet = inet_sk(sk);
913 err = inet_ulp_can_listen(sk);
917 reqsk_queue_alloc(&icsk->icsk_accept_queue);
919 sk->sk_max_ack_backlog = backlog;
920 sk->sk_ack_backlog = 0;
921 inet_csk_delack_init(sk);
923 /* There is race window here: we announce ourselves listening,
924 * but this transition is still not validated by get_port().
925 * It is OK, because this socket enters to hash table only
926 * after validation is complete.
929 sk_state_store(sk, TCP_LISTEN);
930 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
931 inet->inet_sport = htons(inet->inet_num);
934 err = sk->sk_prot->hash(sk);
940 sk->sk_state = TCP_CLOSE;
943 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
945 static void inet_child_forget(struct sock *sk, struct request_sock *req,
948 sk->sk_prot->disconnect(child, O_NONBLOCK);
952 percpu_counter_inc(sk->sk_prot->orphan_count);
954 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
955 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
956 BUG_ON(sk != req->rsk_listener);
958 /* Paranoid, to prevent race condition if
959 * an inbound pkt destined for child is
960 * blocked by sock lock in tcp_v4_rcv().
961 * Also to satisfy an assertion in
962 * tcp_v4_destroy_sock().
964 tcp_sk(child)->fastopen_rsk = NULL;
966 inet_csk_destroy_sock(child);
969 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
970 struct request_sock *req,
973 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
975 spin_lock(&queue->rskq_lock);
976 if (unlikely(sk->sk_state != TCP_LISTEN)) {
977 inet_child_forget(sk, req, child);
982 if (queue->rskq_accept_head == NULL)
983 WRITE_ONCE(queue->rskq_accept_head, req);
985 queue->rskq_accept_tail->dl_next = req;
986 queue->rskq_accept_tail = req;
987 sk_acceptq_added(sk);
989 spin_unlock(&queue->rskq_lock);
992 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
994 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
995 struct request_sock *req, bool own_req)
998 inet_csk_reqsk_queue_drop(sk, req);
999 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
1000 if (inet_csk_reqsk_queue_add(sk, req, child))
1003 /* Too bad, another child took ownership of the request, undo. */
1004 bh_unlock_sock(child);
1008 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1011 * This routine closes sockets which have been at least partially
1012 * opened, but not yet accepted.
1014 void inet_csk_listen_stop(struct sock *sk)
1016 struct inet_connection_sock *icsk = inet_csk(sk);
1017 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1018 struct request_sock *next, *req;
1020 /* Following specs, it would be better either to send FIN
1021 * (and enter FIN-WAIT-1, it is normal close)
1022 * or to send active reset (abort).
1023 * Certainly, it is pretty dangerous while synflood, but it is
1024 * bad justification for our negligence 8)
1025 * To be honest, we are not able to make either
1026 * of the variants now. --ANK
1028 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1029 struct sock *child = req->sk;
1032 bh_lock_sock(child);
1033 WARN_ON(sock_owned_by_user(child));
1036 inet_child_forget(sk, req, child);
1038 bh_unlock_sock(child);
1044 if (queue->fastopenq.rskq_rst_head) {
1045 /* Free all the reqs queued in rskq_rst_head. */
1046 spin_lock_bh(&queue->fastopenq.lock);
1047 req = queue->fastopenq.rskq_rst_head;
1048 queue->fastopenq.rskq_rst_head = NULL;
1049 spin_unlock_bh(&queue->fastopenq.lock);
1050 while (req != NULL) {
1051 next = req->dl_next;
1056 WARN_ON_ONCE(sk->sk_ack_backlog);
1058 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1060 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1062 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1063 const struct inet_sock *inet = inet_sk(sk);
1065 sin->sin_family = AF_INET;
1066 sin->sin_addr.s_addr = inet->inet_daddr;
1067 sin->sin_port = inet->inet_dport;
1069 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1071 #ifdef CONFIG_COMPAT
1072 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1073 char __user *optval, int __user *optlen)
1075 const struct inet_connection_sock *icsk = inet_csk(sk);
1077 if (icsk->icsk_af_ops->compat_getsockopt)
1078 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1080 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1083 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1085 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1086 char __user *optval, unsigned int optlen)
1088 const struct inet_connection_sock *icsk = inet_csk(sk);
1090 if (icsk->icsk_af_ops->compat_setsockopt)
1091 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1093 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1096 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1099 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1101 const struct inet_sock *inet = inet_sk(sk);
1102 const struct ip_options_rcu *inet_opt;
1103 __be32 daddr = inet->inet_daddr;
1108 inet_opt = rcu_dereference(inet->inet_opt);
1109 if (inet_opt && inet_opt->opt.srr)
1110 daddr = inet_opt->opt.faddr;
1112 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1113 inet->inet_saddr, inet->inet_dport,
1114 inet->inet_sport, sk->sk_protocol,
1115 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1119 sk_setup_caps(sk, &rt->dst);
1125 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1127 struct dst_entry *dst = __sk_dst_check(sk, 0);
1128 struct inet_sock *inet = inet_sk(sk);
1131 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1135 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1137 dst = __sk_dst_check(sk, 0);
1139 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1143 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);