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_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
28 * only, and any IPv4 addresses if not IPv6 only
29 * match_wildcard == false: addresses must be exactly the same, i.e.
30 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
31 * and 0.0.0.0 equals to 0.0.0.0 only
33 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
34 const struct in6_addr *sk2_rcv_saddr6,
35 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
36 bool sk1_ipv6only, bool sk2_ipv6only,
39 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
40 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
42 /* if both are mapped, treat as IPv4 */
43 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
45 if (sk1_rcv_saddr == sk2_rcv_saddr)
47 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
48 return match_wildcard;
53 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
56 if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
57 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
60 if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
61 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
65 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
72 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
73 * match_wildcard == false: addresses must be exactly the same, i.e.
74 * 0.0.0.0 only equals to 0.0.0.0
76 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
77 bool sk2_ipv6only, bool match_wildcard)
80 if (sk1_rcv_saddr == sk2_rcv_saddr)
82 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
83 return match_wildcard;
88 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
91 #if IS_ENABLED(CONFIG_IPV6)
92 if (sk->sk_family == AF_INET6)
93 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
101 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
102 ipv6_only_sock(sk2), match_wildcard);
104 EXPORT_SYMBOL(inet_rcv_saddr_equal);
106 bool inet_rcv_saddr_any(const struct sock *sk)
108 #if IS_ENABLED(CONFIG_IPV6)
109 if (sk->sk_family == AF_INET6)
110 return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
112 return !sk->sk_rcv_saddr;
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;
184 l3mdev = inet_sk_bound_l3mdev(sk);
185 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
187 inet_get_local_port_range(net, &low, &high);
188 high++; /* [32768, 60999] -> [32768, 61000[ */
192 int half = low + (((high - low) >> 2) << 1);
194 if (attempt_half == 1)
199 remaining = high - low;
200 if (likely(remaining > 1))
203 offset = prandom_u32() % remaining;
204 /* __inet_hash_connect() favors ports having @low parity
205 * We do the opposite to not pollute connect() users.
211 for (i = 0; i < remaining; i += 2, port += 2) {
212 if (unlikely(port >= high))
214 if (inet_is_local_reserved_port(net, port))
216 head = &hinfo->bhash[inet_bhashfn(net, port,
218 spin_lock_bh(&head->lock);
219 inet_bind_bucket_for_each(tb, &head->chain)
220 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
222 if (!inet_csk_bind_conflict(sk, tb, false, false))
229 spin_unlock_bh(&head->lock);
235 goto other_parity_scan;
237 if (attempt_half == 1) {
238 /* OK we now try the upper half of the range */
240 goto other_half_scan;
249 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
252 kuid_t uid = sock_i_uid(sk);
254 if (tb->fastreuseport <= 0)
256 if (!sk->sk_reuseport)
258 if (rcu_access_pointer(sk->sk_reuseport_cb))
260 if (!uid_eq(tb->fastuid, uid))
262 /* We only need to check the rcv_saddr if this tb was once marked
263 * without fastreuseport and then was reset, as we can only know that
264 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
267 if (tb->fastreuseport == FASTREUSEPORT_ANY)
269 #if IS_ENABLED(CONFIG_IPV6)
270 if (tb->fast_sk_family == AF_INET6)
271 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
276 ipv6_only_sock(sk), true);
278 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
279 ipv6_only_sock(sk), true);
282 /* Obtain a reference to a local port for the given sock,
283 * if snum is zero it means select any available local port.
284 * We try to allocate an odd port (and leave even ports for connect())
286 int inet_csk_get_port(struct sock *sk, unsigned short snum)
288 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
289 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
290 int ret = 1, port = snum;
291 struct inet_bind_hashbucket *head;
292 struct net *net = sock_net(sk);
293 struct inet_bind_bucket *tb = NULL;
294 kuid_t uid = sock_i_uid(sk);
297 l3mdev = inet_sk_bound_l3mdev(sk);
300 head = inet_csk_find_open_port(sk, &tb, &port);
307 head = &hinfo->bhash[inet_bhashfn(net, port,
309 spin_lock_bh(&head->lock);
310 inet_bind_bucket_for_each(tb, &head->chain)
311 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
315 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
316 net, head, port, l3mdev);
320 if (!hlist_empty(&tb->owners)) {
321 if (sk->sk_reuse == SK_FORCE_REUSE)
324 if ((tb->fastreuse > 0 && reuse) ||
325 sk_reuseport_match(tb, sk))
327 if (inet_csk_bind_conflict(sk, tb, true, true))
331 if (hlist_empty(&tb->owners)) {
332 tb->fastreuse = reuse;
333 if (sk->sk_reuseport) {
334 tb->fastreuseport = FASTREUSEPORT_ANY;
336 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
337 tb->fast_ipv6_only = ipv6_only_sock(sk);
338 tb->fast_sk_family = sk->sk_family;
339 #if IS_ENABLED(CONFIG_IPV6)
340 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
343 tb->fastreuseport = 0;
348 if (sk->sk_reuseport) {
349 /* We didn't match or we don't have fastreuseport set on
350 * the tb, but we have sk_reuseport set on this socket
351 * and we know that there are no bind conflicts with
352 * this socket in this tb, so reset our tb's reuseport
353 * settings so that any subsequent sockets that match
354 * our current socket will be put on the fast path.
356 * If we reset we need to set FASTREUSEPORT_STRICT so we
357 * do extra checking for all subsequent sk_reuseport
360 if (!sk_reuseport_match(tb, sk)) {
361 tb->fastreuseport = FASTREUSEPORT_STRICT;
363 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
364 tb->fast_ipv6_only = ipv6_only_sock(sk);
365 tb->fast_sk_family = sk->sk_family;
366 #if IS_ENABLED(CONFIG_IPV6)
367 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
371 tb->fastreuseport = 0;
374 if (!inet_csk(sk)->icsk_bind_hash)
375 inet_bind_hash(sk, tb, port);
376 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
380 spin_unlock_bh(&head->lock);
383 EXPORT_SYMBOL_GPL(inet_csk_get_port);
386 * Wait for an incoming connection, avoid race conditions. This must be called
387 * with the socket locked.
389 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
391 struct inet_connection_sock *icsk = inet_csk(sk);
396 * True wake-one mechanism for incoming connections: only
397 * one process gets woken up, not the 'whole herd'.
398 * Since we do not 'race & poll' for established sockets
399 * anymore, the common case will execute the loop only once.
401 * Subtle issue: "add_wait_queue_exclusive()" will be added
402 * after any current non-exclusive waiters, and we know that
403 * it will always _stay_ after any new non-exclusive waiters
404 * because all non-exclusive waiters are added at the
405 * beginning of the wait-queue. As such, it's ok to "drop"
406 * our exclusiveness temporarily when we get woken up without
407 * having to remove and re-insert us on the wait queue.
410 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
413 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
414 timeo = schedule_timeout(timeo);
415 sched_annotate_sleep();
418 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
421 if (sk->sk_state != TCP_LISTEN)
423 err = sock_intr_errno(timeo);
424 if (signal_pending(current))
430 finish_wait(sk_sleep(sk), &wait);
435 * This will accept the next outstanding connection.
437 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
439 struct inet_connection_sock *icsk = inet_csk(sk);
440 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
441 struct request_sock *req;
447 /* We need to make sure that this socket is listening,
448 * and that it has something pending.
451 if (sk->sk_state != TCP_LISTEN)
454 /* Find already established connection */
455 if (reqsk_queue_empty(queue)) {
456 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
458 /* If this is a non blocking socket don't sleep */
463 error = inet_csk_wait_for_connect(sk, timeo);
467 req = reqsk_queue_remove(queue, sk);
470 if (sk->sk_protocol == IPPROTO_TCP &&
471 tcp_rsk(req)->tfo_listener) {
472 spin_lock_bh(&queue->fastopenq.lock);
473 if (tcp_rsk(req)->tfo_listener) {
474 /* We are still waiting for the final ACK from 3WHS
475 * so can't free req now. Instead, we set req->sk to
476 * NULL to signify that the child socket is taken
477 * so reqsk_fastopen_remove() will free the req
478 * when 3WHS finishes (or is aborted).
483 spin_unlock_bh(&queue->fastopenq.lock);
488 if (newsk && mem_cgroup_sockets_enabled) {
491 /* atomically get the memory usage, set and charge the
496 /* The socket has not been accepted yet, no need to look at
497 * newsk->sk_wmem_queued.
499 amt = sk_mem_pages(newsk->sk_forward_alloc +
500 atomic_read(&newsk->sk_rmem_alloc));
501 mem_cgroup_sk_alloc(newsk);
502 if (newsk->sk_memcg && amt)
503 mem_cgroup_charge_skmem(newsk->sk_memcg, amt);
516 EXPORT_SYMBOL(inet_csk_accept);
519 * Using different timers for retransmit, delayed acks and probes
520 * We may wish use just one timer maintaining a list of expire jiffies
523 void inet_csk_init_xmit_timers(struct sock *sk,
524 void (*retransmit_handler)(struct timer_list *t),
525 void (*delack_handler)(struct timer_list *t),
526 void (*keepalive_handler)(struct timer_list *t))
528 struct inet_connection_sock *icsk = inet_csk(sk);
530 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
531 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
532 timer_setup(&sk->sk_timer, keepalive_handler, 0);
533 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
535 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
537 void inet_csk_clear_xmit_timers(struct sock *sk)
539 struct inet_connection_sock *icsk = inet_csk(sk);
541 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
543 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
544 sk_stop_timer(sk, &icsk->icsk_delack_timer);
545 sk_stop_timer(sk, &sk->sk_timer);
547 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
549 void inet_csk_delete_keepalive_timer(struct sock *sk)
551 sk_stop_timer(sk, &sk->sk_timer);
553 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
555 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
557 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
559 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
561 struct dst_entry *inet_csk_route_req(const struct sock *sk,
563 const struct request_sock *req)
565 const struct inet_request_sock *ireq = inet_rsk(req);
566 struct net *net = read_pnet(&ireq->ireq_net);
567 struct ip_options_rcu *opt;
571 opt = rcu_dereference(ireq->ireq_opt);
573 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
574 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
575 sk->sk_protocol, inet_sk_flowi_flags(sk),
576 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
577 ireq->ir_loc_addr, ireq->ir_rmt_port,
578 htons(ireq->ir_num), sk->sk_uid);
579 security_req_classify_flow(req, flowi4_to_flowi(fl4));
580 rt = ip_route_output_flow(net, fl4, sk);
583 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
592 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
595 EXPORT_SYMBOL_GPL(inet_csk_route_req);
597 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
599 const struct request_sock *req)
601 const struct inet_request_sock *ireq = inet_rsk(req);
602 struct net *net = read_pnet(&ireq->ireq_net);
603 struct inet_sock *newinet = inet_sk(newsk);
604 struct ip_options_rcu *opt;
608 opt = rcu_dereference(ireq->ireq_opt);
609 fl4 = &newinet->cork.fl.u.ip4;
611 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
612 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
613 sk->sk_protocol, inet_sk_flowi_flags(sk),
614 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
615 ireq->ir_loc_addr, ireq->ir_rmt_port,
616 htons(ireq->ir_num), sk->sk_uid);
617 security_req_classify_flow(req, flowi4_to_flowi(fl4));
618 rt = ip_route_output_flow(net, fl4, sk);
621 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
628 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
631 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
633 /* Decide when to expire the request and when to resend SYN-ACK */
634 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
635 const int max_retries,
636 const u8 rskq_defer_accept,
637 int *expire, int *resend)
639 if (!rskq_defer_accept) {
640 *expire = req->num_timeout >= thresh;
644 *expire = req->num_timeout >= thresh &&
645 (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
647 * Do not resend while waiting for data after ACK,
648 * start to resend on end of deferring period to give
649 * last chance for data or ACK to create established socket.
651 *resend = !inet_rsk(req)->acked ||
652 req->num_timeout >= rskq_defer_accept - 1;
655 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
657 int err = req->rsk_ops->rtx_syn_ack(parent, req);
663 EXPORT_SYMBOL(inet_rtx_syn_ack);
665 /* return true if req was found in the ehash table */
666 static bool reqsk_queue_unlink(struct request_sock *req)
668 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
671 if (sk_hashed(req_to_sk(req))) {
672 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
675 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
678 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
683 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
685 if (reqsk_queue_unlink(req)) {
686 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
690 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
692 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
694 inet_csk_reqsk_queue_drop(sk, req);
697 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
699 static void reqsk_timer_handler(struct timer_list *t)
701 struct request_sock *req = from_timer(req, t, rsk_timer);
702 struct sock *sk_listener = req->rsk_listener;
703 struct net *net = sock_net(sk_listener);
704 struct inet_connection_sock *icsk = inet_csk(sk_listener);
705 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
706 int qlen, expire = 0, resend = 0;
707 int max_retries, thresh;
710 if (inet_sk_state_load(sk_listener) != TCP_LISTEN)
713 max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
714 thresh = max_retries;
715 /* Normally all the openreqs are young and become mature
716 * (i.e. converted to established socket) for first timeout.
717 * If synack was not acknowledged for 1 second, it means
718 * one of the following things: synack was lost, ack was lost,
719 * rtt is high or nobody planned to ack (i.e. synflood).
720 * When server is a bit loaded, queue is populated with old
721 * open requests, reducing effective size of queue.
722 * When server is well loaded, queue size reduces to zero
723 * after several minutes of work. It is not synflood,
724 * it is normal operation. The solution is pruning
725 * too old entries overriding normal timeout, when
726 * situation becomes dangerous.
728 * Essentially, we reserve half of room for young
729 * embrions; and abort old ones without pity, if old
730 * ones are about to clog our table.
732 qlen = reqsk_queue_len(queue);
733 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
734 int young = reqsk_queue_len_young(queue) << 1;
743 defer_accept = READ_ONCE(queue->rskq_defer_accept);
745 max_retries = defer_accept;
746 syn_ack_recalc(req, thresh, max_retries, defer_accept,
748 req->rsk_ops->syn_ack_timeout(req);
751 !inet_rtx_syn_ack(sk_listener, req) ||
752 inet_rsk(req)->acked)) {
755 if (req->num_timeout++ == 0)
756 atomic_dec(&queue->young);
757 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
758 mod_timer(&req->rsk_timer, jiffies + timeo);
762 inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
765 static void reqsk_queue_hash_req(struct request_sock *req,
766 unsigned long timeout)
768 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
769 mod_timer(&req->rsk_timer, jiffies + timeout);
771 inet_ehash_insert(req_to_sk(req), NULL);
772 /* before letting lookups find us, make sure all req fields
773 * are committed to memory and refcnt initialized.
776 refcount_set(&req->rsk_refcnt, 2 + 1);
779 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
780 unsigned long timeout)
782 reqsk_queue_hash_req(req, timeout);
783 inet_csk_reqsk_queue_added(sk);
785 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
787 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
788 const gfp_t priority)
790 struct inet_connection_sock *icsk = inet_csk(newsk);
792 if (!icsk->icsk_ulp_ops)
795 if (icsk->icsk_ulp_ops->clone)
796 icsk->icsk_ulp_ops->clone(req, newsk, priority);
800 * inet_csk_clone_lock - clone an inet socket, and lock its clone
801 * @sk: the socket to clone
803 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
805 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
807 struct sock *inet_csk_clone_lock(const struct sock *sk,
808 const struct request_sock *req,
809 const gfp_t priority)
811 struct sock *newsk = sk_clone_lock(sk, priority);
814 struct inet_connection_sock *newicsk = inet_csk(newsk);
816 inet_sk_set_state(newsk, TCP_SYN_RECV);
817 newicsk->icsk_bind_hash = NULL;
819 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
820 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
821 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
823 /* listeners have SOCK_RCU_FREE, not the children */
824 sock_reset_flag(newsk, SOCK_RCU_FREE);
826 inet_sk(newsk)->mc_list = NULL;
828 newsk->sk_mark = inet_rsk(req)->ir_mark;
829 atomic64_set(&newsk->sk_cookie,
830 atomic64_read(&inet_rsk(req)->ir_cookie));
832 newicsk->icsk_retransmits = 0;
833 newicsk->icsk_backoff = 0;
834 newicsk->icsk_probes_out = 0;
836 /* Deinitialize accept_queue to trap illegal accesses. */
837 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
839 inet_clone_ulp(req, newsk, priority);
841 security_inet_csk_clone(newsk, req);
845 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
848 * At this point, there should be no process reference to this
849 * socket, and thus no user references at all. Therefore we
850 * can assume the socket waitqueue is inactive and nobody will
851 * try to jump onto it.
853 void inet_csk_destroy_sock(struct sock *sk)
855 WARN_ON(sk->sk_state != TCP_CLOSE);
856 WARN_ON(!sock_flag(sk, SOCK_DEAD));
858 /* It cannot be in hash table! */
859 WARN_ON(!sk_unhashed(sk));
861 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
862 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
864 sk->sk_prot->destroy(sk);
866 sk_stream_kill_queues(sk);
868 xfrm_sk_free_policy(sk);
870 sk_refcnt_debug_release(sk);
872 percpu_counter_dec(sk->sk_prot->orphan_count);
876 EXPORT_SYMBOL(inet_csk_destroy_sock);
878 /* This function allows to force a closure of a socket after the call to
879 * tcp/dccp_create_openreq_child().
881 void inet_csk_prepare_forced_close(struct sock *sk)
882 __releases(&sk->sk_lock.slock)
884 /* sk_clone_lock locked the socket and set refcnt to 2 */
888 /* The below has to be done to allow calling inet_csk_destroy_sock */
889 sock_set_flag(sk, SOCK_DEAD);
890 percpu_counter_inc(sk->sk_prot->orphan_count);
891 inet_sk(sk)->inet_num = 0;
893 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
895 int inet_csk_listen_start(struct sock *sk, int backlog)
897 struct inet_connection_sock *icsk = inet_csk(sk);
898 struct inet_sock *inet = inet_sk(sk);
899 int err = -EADDRINUSE;
901 reqsk_queue_alloc(&icsk->icsk_accept_queue);
903 sk->sk_ack_backlog = 0;
904 inet_csk_delack_init(sk);
906 /* There is race window here: we announce ourselves listening,
907 * but this transition is still not validated by get_port().
908 * It is OK, because this socket enters to hash table only
909 * after validation is complete.
911 inet_sk_state_store(sk, TCP_LISTEN);
912 if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
913 inet->inet_sport = htons(inet->inet_num);
916 err = sk->sk_prot->hash(sk);
922 inet_sk_set_state(sk, TCP_CLOSE);
925 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
927 static void inet_child_forget(struct sock *sk, struct request_sock *req,
930 sk->sk_prot->disconnect(child, O_NONBLOCK);
934 percpu_counter_inc(sk->sk_prot->orphan_count);
936 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
937 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
938 BUG_ON(sk != req->rsk_listener);
940 /* Paranoid, to prevent race condition if
941 * an inbound pkt destined for child is
942 * blocked by sock lock in tcp_v4_rcv().
943 * Also to satisfy an assertion in
944 * tcp_v4_destroy_sock().
946 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
948 inet_csk_destroy_sock(child);
951 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
952 struct request_sock *req,
955 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
957 spin_lock(&queue->rskq_lock);
958 if (unlikely(sk->sk_state != TCP_LISTEN)) {
959 inet_child_forget(sk, req, child);
964 if (queue->rskq_accept_head == NULL)
965 WRITE_ONCE(queue->rskq_accept_head, req);
967 queue->rskq_accept_tail->dl_next = req;
968 queue->rskq_accept_tail = req;
969 sk_acceptq_added(sk);
971 spin_unlock(&queue->rskq_lock);
974 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
976 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
977 struct request_sock *req, bool own_req)
980 inet_csk_reqsk_queue_drop(sk, req);
981 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
982 if (inet_csk_reqsk_queue_add(sk, req, child))
985 /* Too bad, another child took ownership of the request, undo. */
986 bh_unlock_sock(child);
990 EXPORT_SYMBOL(inet_csk_complete_hashdance);
993 * This routine closes sockets which have been at least partially
994 * opened, but not yet accepted.
996 void inet_csk_listen_stop(struct sock *sk)
998 struct inet_connection_sock *icsk = inet_csk(sk);
999 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1000 struct request_sock *next, *req;
1002 /* Following specs, it would be better either to send FIN
1003 * (and enter FIN-WAIT-1, it is normal close)
1004 * or to send active reset (abort).
1005 * Certainly, it is pretty dangerous while synflood, but it is
1006 * bad justification for our negligence 8)
1007 * To be honest, we are not able to make either
1008 * of the variants now. --ANK
1010 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1011 struct sock *child = req->sk;
1014 bh_lock_sock(child);
1015 WARN_ON(sock_owned_by_user(child));
1018 inet_child_forget(sk, req, child);
1020 bh_unlock_sock(child);
1026 if (queue->fastopenq.rskq_rst_head) {
1027 /* Free all the reqs queued in rskq_rst_head. */
1028 spin_lock_bh(&queue->fastopenq.lock);
1029 req = queue->fastopenq.rskq_rst_head;
1030 queue->fastopenq.rskq_rst_head = NULL;
1031 spin_unlock_bh(&queue->fastopenq.lock);
1032 while (req != NULL) {
1033 next = req->dl_next;
1038 WARN_ON_ONCE(sk->sk_ack_backlog);
1040 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1042 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1044 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1045 const struct inet_sock *inet = inet_sk(sk);
1047 sin->sin_family = AF_INET;
1048 sin->sin_addr.s_addr = inet->inet_daddr;
1049 sin->sin_port = inet->inet_dport;
1051 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1053 #ifdef CONFIG_COMPAT
1054 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1055 char __user *optval, int __user *optlen)
1057 const struct inet_connection_sock *icsk = inet_csk(sk);
1059 if (icsk->icsk_af_ops->compat_getsockopt)
1060 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1062 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1065 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1067 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1068 char __user *optval, unsigned int optlen)
1070 const struct inet_connection_sock *icsk = inet_csk(sk);
1072 if (icsk->icsk_af_ops->compat_setsockopt)
1073 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1075 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1078 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1081 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1083 const struct inet_sock *inet = inet_sk(sk);
1084 const struct ip_options_rcu *inet_opt;
1085 __be32 daddr = inet->inet_daddr;
1090 inet_opt = rcu_dereference(inet->inet_opt);
1091 if (inet_opt && inet_opt->opt.srr)
1092 daddr = inet_opt->opt.faddr;
1094 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1095 inet->inet_saddr, inet->inet_dport,
1096 inet->inet_sport, sk->sk_protocol,
1097 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1101 sk_setup_caps(sk, &rt->dst);
1107 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1109 struct dst_entry *dst = __sk_dst_check(sk, 0);
1110 struct inet_sock *inet = inet_sk(sk);
1113 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1117 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1119 dst = __sk_dst_check(sk, 0);
1121 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1125 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);