ipv4: introduce rt_uses_gateway
[linux-3.10.git] / net / ipv4 / inet_connection_sock.c
1 /*
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.
5  *
6  *              Support for INET connection oriented protocols.
7  *
8  * Authors:     See the TCP sources
9  *
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.
14  */
15
16 #include <linux/module.h>
17 #include <linux/jhash.h>
18
19 #include <net/inet_connection_sock.h>
20 #include <net/inet_hashtables.h>
21 #include <net/inet_timewait_sock.h>
22 #include <net/ip.h>
23 #include <net/route.h>
24 #include <net/tcp_states.h>
25 #include <net/xfrm.h>
26
27 #ifdef INET_CSK_DEBUG
28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
30 #endif
31
32 /*
33  * This struct holds the first and last local port number.
34  */
35 struct local_ports sysctl_local_ports __read_mostly = {
36         .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock),
37         .range = { 32768, 61000 },
38 };
39
40 unsigned long *sysctl_local_reserved_ports;
41 EXPORT_SYMBOL(sysctl_local_reserved_ports);
42
43 void inet_get_local_port_range(int *low, int *high)
44 {
45         unsigned int seq;
46
47         do {
48                 seq = read_seqbegin(&sysctl_local_ports.lock);
49
50                 *low = sysctl_local_ports.range[0];
51                 *high = sysctl_local_ports.range[1];
52         } while (read_seqretry(&sysctl_local_ports.lock, seq));
53 }
54 EXPORT_SYMBOL(inet_get_local_port_range);
55
56 int inet_csk_bind_conflict(const struct sock *sk,
57                            const struct inet_bind_bucket *tb, bool relax)
58 {
59         struct sock *sk2;
60         struct hlist_node *node;
61         int reuse = sk->sk_reuse;
62
63         /*
64          * Unlike other sk lookup places we do not check
65          * for sk_net here, since _all_ the socks listed
66          * in tb->owners list belong to the same net - the
67          * one this bucket belongs to.
68          */
69
70         sk_for_each_bound(sk2, node, &tb->owners) {
71                 if (sk != sk2 &&
72                     !inet_v6_ipv6only(sk2) &&
73                     (!sk->sk_bound_dev_if ||
74                      !sk2->sk_bound_dev_if ||
75                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
76                         if (!reuse || !sk2->sk_reuse ||
77                             sk2->sk_state == TCP_LISTEN) {
78                                 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
79                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
80                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
81                                         break;
82                         }
83                         if (!relax && reuse && sk2->sk_reuse &&
84                             sk2->sk_state != TCP_LISTEN) {
85                                 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
86
87                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
88                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
89                                         break;
90                         }
91                 }
92         }
93         return node != NULL;
94 }
95 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
96
97 /* Obtain a reference to a local port for the given sock,
98  * if snum is zero it means select any available local port.
99  */
100 int inet_csk_get_port(struct sock *sk, unsigned short snum)
101 {
102         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
103         struct inet_bind_hashbucket *head;
104         struct hlist_node *node;
105         struct inet_bind_bucket *tb;
106         int ret, attempts = 5;
107         struct net *net = sock_net(sk);
108         int smallest_size = -1, smallest_rover;
109
110         local_bh_disable();
111         if (!snum) {
112                 int remaining, rover, low, high;
113
114 again:
115                 inet_get_local_port_range(&low, &high);
116                 remaining = (high - low) + 1;
117                 smallest_rover = rover = net_random() % remaining + low;
118
119                 smallest_size = -1;
120                 do {
121                         if (inet_is_reserved_local_port(rover))
122                                 goto next_nolock;
123                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
124                                         hashinfo->bhash_size)];
125                         spin_lock(&head->lock);
126                         inet_bind_bucket_for_each(tb, node, &head->chain)
127                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
128                                         if (tb->fastreuse > 0 &&
129                                             sk->sk_reuse &&
130                                             sk->sk_state != TCP_LISTEN &&
131                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
132                                                 smallest_size = tb->num_owners;
133                                                 smallest_rover = rover;
134                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
135                                                     !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
136                                                         snum = smallest_rover;
137                                                         goto tb_found;
138                                                 }
139                                         }
140                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
141                                                 snum = rover;
142                                                 goto tb_found;
143                                         }
144                                         goto next;
145                                 }
146                         break;
147                 next:
148                         spin_unlock(&head->lock);
149                 next_nolock:
150                         if (++rover > high)
151                                 rover = low;
152                 } while (--remaining > 0);
153
154                 /* Exhausted local port range during search?  It is not
155                  * possible for us to be holding one of the bind hash
156                  * locks if this test triggers, because if 'remaining'
157                  * drops to zero, we broke out of the do/while loop at
158                  * the top level, not from the 'break;' statement.
159                  */
160                 ret = 1;
161                 if (remaining <= 0) {
162                         if (smallest_size != -1) {
163                                 snum = smallest_rover;
164                                 goto have_snum;
165                         }
166                         goto fail;
167                 }
168                 /* OK, here is the one we will use.  HEAD is
169                  * non-NULL and we hold it's mutex.
170                  */
171                 snum = rover;
172         } else {
173 have_snum:
174                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
175                                 hashinfo->bhash_size)];
176                 spin_lock(&head->lock);
177                 inet_bind_bucket_for_each(tb, node, &head->chain)
178                         if (net_eq(ib_net(tb), net) && tb->port == snum)
179                                 goto tb_found;
180         }
181         tb = NULL;
182         goto tb_not_found;
183 tb_found:
184         if (!hlist_empty(&tb->owners)) {
185                 if (sk->sk_reuse == SK_FORCE_REUSE)
186                         goto success;
187
188                 if (tb->fastreuse > 0 &&
189                     sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
190                     smallest_size == -1) {
191                         goto success;
192                 } else {
193                         ret = 1;
194                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
195                                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
196                                     smallest_size != -1 && --attempts >= 0) {
197                                         spin_unlock(&head->lock);
198                                         goto again;
199                                 }
200
201                                 goto fail_unlock;
202                         }
203                 }
204         }
205 tb_not_found:
206         ret = 1;
207         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
208                                         net, head, snum)) == NULL)
209                 goto fail_unlock;
210         if (hlist_empty(&tb->owners)) {
211                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
212                         tb->fastreuse = 1;
213                 else
214                         tb->fastreuse = 0;
215         } else if (tb->fastreuse &&
216                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
217                 tb->fastreuse = 0;
218 success:
219         if (!inet_csk(sk)->icsk_bind_hash)
220                 inet_bind_hash(sk, tb, snum);
221         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
222         ret = 0;
223
224 fail_unlock:
225         spin_unlock(&head->lock);
226 fail:
227         local_bh_enable();
228         return ret;
229 }
230 EXPORT_SYMBOL_GPL(inet_csk_get_port);
231
232 /*
233  * Wait for an incoming connection, avoid race conditions. This must be called
234  * with the socket locked.
235  */
236 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
237 {
238         struct inet_connection_sock *icsk = inet_csk(sk);
239         DEFINE_WAIT(wait);
240         int err;
241
242         /*
243          * True wake-one mechanism for incoming connections: only
244          * one process gets woken up, not the 'whole herd'.
245          * Since we do not 'race & poll' for established sockets
246          * anymore, the common case will execute the loop only once.
247          *
248          * Subtle issue: "add_wait_queue_exclusive()" will be added
249          * after any current non-exclusive waiters, and we know that
250          * it will always _stay_ after any new non-exclusive waiters
251          * because all non-exclusive waiters are added at the
252          * beginning of the wait-queue. As such, it's ok to "drop"
253          * our exclusiveness temporarily when we get woken up without
254          * having to remove and re-insert us on the wait queue.
255          */
256         for (;;) {
257                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
258                                           TASK_INTERRUPTIBLE);
259                 release_sock(sk);
260                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
261                         timeo = schedule_timeout(timeo);
262                 lock_sock(sk);
263                 err = 0;
264                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
265                         break;
266                 err = -EINVAL;
267                 if (sk->sk_state != TCP_LISTEN)
268                         break;
269                 err = sock_intr_errno(timeo);
270                 if (signal_pending(current))
271                         break;
272                 err = -EAGAIN;
273                 if (!timeo)
274                         break;
275         }
276         finish_wait(sk_sleep(sk), &wait);
277         return err;
278 }
279
280 /*
281  * This will accept the next outstanding connection.
282  */
283 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
284 {
285         struct inet_connection_sock *icsk = inet_csk(sk);
286         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
287         struct sock *newsk;
288         struct request_sock *req;
289         int error;
290
291         lock_sock(sk);
292
293         /* We need to make sure that this socket is listening,
294          * and that it has something pending.
295          */
296         error = -EINVAL;
297         if (sk->sk_state != TCP_LISTEN)
298                 goto out_err;
299
300         /* Find already established connection */
301         if (reqsk_queue_empty(queue)) {
302                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
303
304                 /* If this is a non blocking socket don't sleep */
305                 error = -EAGAIN;
306                 if (!timeo)
307                         goto out_err;
308
309                 error = inet_csk_wait_for_connect(sk, timeo);
310                 if (error)
311                         goto out_err;
312         }
313         req = reqsk_queue_remove(queue);
314         newsk = req->sk;
315
316         sk_acceptq_removed(sk);
317         if (sk->sk_protocol == IPPROTO_TCP && queue->fastopenq != NULL) {
318                 spin_lock_bh(&queue->fastopenq->lock);
319                 if (tcp_rsk(req)->listener) {
320                         /* We are still waiting for the final ACK from 3WHS
321                          * so can't free req now. Instead, we set req->sk to
322                          * NULL to signify that the child socket is taken
323                          * so reqsk_fastopen_remove() will free the req
324                          * when 3WHS finishes (or is aborted).
325                          */
326                         req->sk = NULL;
327                         req = NULL;
328                 }
329                 spin_unlock_bh(&queue->fastopenq->lock);
330         }
331 out:
332         release_sock(sk);
333         if (req)
334                 __reqsk_free(req);
335         return newsk;
336 out_err:
337         newsk = NULL;
338         req = NULL;
339         *err = error;
340         goto out;
341 }
342 EXPORT_SYMBOL(inet_csk_accept);
343
344 /*
345  * Using different timers for retransmit, delayed acks and probes
346  * We may wish use just one timer maintaining a list of expire jiffies
347  * to optimize.
348  */
349 void inet_csk_init_xmit_timers(struct sock *sk,
350                                void (*retransmit_handler)(unsigned long),
351                                void (*delack_handler)(unsigned long),
352                                void (*keepalive_handler)(unsigned long))
353 {
354         struct inet_connection_sock *icsk = inet_csk(sk);
355
356         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
357                         (unsigned long)sk);
358         setup_timer(&icsk->icsk_delack_timer, delack_handler,
359                         (unsigned long)sk);
360         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
361         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
362 }
363 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
364
365 void inet_csk_clear_xmit_timers(struct sock *sk)
366 {
367         struct inet_connection_sock *icsk = inet_csk(sk);
368
369         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
370
371         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
372         sk_stop_timer(sk, &icsk->icsk_delack_timer);
373         sk_stop_timer(sk, &sk->sk_timer);
374 }
375 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
376
377 void inet_csk_delete_keepalive_timer(struct sock *sk)
378 {
379         sk_stop_timer(sk, &sk->sk_timer);
380 }
381 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
382
383 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
384 {
385         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
386 }
387 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
388
389 struct dst_entry *inet_csk_route_req(struct sock *sk,
390                                      struct flowi4 *fl4,
391                                      const struct request_sock *req)
392 {
393         struct rtable *rt;
394         const struct inet_request_sock *ireq = inet_rsk(req);
395         struct ip_options_rcu *opt = inet_rsk(req)->opt;
396         struct net *net = sock_net(sk);
397         int flags = inet_sk_flowi_flags(sk);
398
399         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
400                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
401                            sk->sk_protocol,
402                            flags,
403                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
404                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
405         security_req_classify_flow(req, flowi4_to_flowi(fl4));
406         rt = ip_route_output_flow(net, fl4, sk);
407         if (IS_ERR(rt))
408                 goto no_route;
409         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
410                 goto route_err;
411         return &rt->dst;
412
413 route_err:
414         ip_rt_put(rt);
415 no_route:
416         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
417         return NULL;
418 }
419 EXPORT_SYMBOL_GPL(inet_csk_route_req);
420
421 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
422                                             struct sock *newsk,
423                                             const struct request_sock *req)
424 {
425         const struct inet_request_sock *ireq = inet_rsk(req);
426         struct inet_sock *newinet = inet_sk(newsk);
427         struct ip_options_rcu *opt;
428         struct net *net = sock_net(sk);
429         struct flowi4 *fl4;
430         struct rtable *rt;
431
432         fl4 = &newinet->cork.fl.u.ip4;
433
434         rcu_read_lock();
435         opt = rcu_dereference(newinet->inet_opt);
436         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
437                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
438                            sk->sk_protocol, inet_sk_flowi_flags(sk),
439                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
440                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
441         security_req_classify_flow(req, flowi4_to_flowi(fl4));
442         rt = ip_route_output_flow(net, fl4, sk);
443         if (IS_ERR(rt))
444                 goto no_route;
445         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
446                 goto route_err;
447         rcu_read_unlock();
448         return &rt->dst;
449
450 route_err:
451         ip_rt_put(rt);
452 no_route:
453         rcu_read_unlock();
454         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
455         return NULL;
456 }
457 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
458
459 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
460                                  const u32 rnd, const u32 synq_hsize)
461 {
462         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
463 }
464
465 #if IS_ENABLED(CONFIG_IPV6)
466 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
467 #else
468 #define AF_INET_FAMILY(fam) 1
469 #endif
470
471 struct request_sock *inet_csk_search_req(const struct sock *sk,
472                                          struct request_sock ***prevp,
473                                          const __be16 rport, const __be32 raddr,
474                                          const __be32 laddr)
475 {
476         const struct inet_connection_sock *icsk = inet_csk(sk);
477         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
478         struct request_sock *req, **prev;
479
480         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
481                                                     lopt->nr_table_entries)];
482              (req = *prev) != NULL;
483              prev = &req->dl_next) {
484                 const struct inet_request_sock *ireq = inet_rsk(req);
485
486                 if (ireq->rmt_port == rport &&
487                     ireq->rmt_addr == raddr &&
488                     ireq->loc_addr == laddr &&
489                     AF_INET_FAMILY(req->rsk_ops->family)) {
490                         WARN_ON(req->sk);
491                         *prevp = prev;
492                         break;
493                 }
494         }
495
496         return req;
497 }
498 EXPORT_SYMBOL_GPL(inet_csk_search_req);
499
500 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
501                                    unsigned long timeout)
502 {
503         struct inet_connection_sock *icsk = inet_csk(sk);
504         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
505         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
506                                      lopt->hash_rnd, lopt->nr_table_entries);
507
508         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
509         inet_csk_reqsk_queue_added(sk, timeout);
510 }
511 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
512
513 /* Only thing we need from tcp.h */
514 extern int sysctl_tcp_synack_retries;
515
516
517 /* Decide when to expire the request and when to resend SYN-ACK */
518 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
519                                   const int max_retries,
520                                   const u8 rskq_defer_accept,
521                                   int *expire, int *resend)
522 {
523         if (!rskq_defer_accept) {
524                 *expire = req->retrans >= thresh;
525                 *resend = 1;
526                 return;
527         }
528         *expire = req->retrans >= thresh &&
529                   (!inet_rsk(req)->acked || req->retrans >= max_retries);
530         /*
531          * Do not resend while waiting for data after ACK,
532          * start to resend on end of deferring period to give
533          * last chance for data or ACK to create established socket.
534          */
535         *resend = !inet_rsk(req)->acked ||
536                   req->retrans >= rskq_defer_accept - 1;
537 }
538
539 void inet_csk_reqsk_queue_prune(struct sock *parent,
540                                 const unsigned long interval,
541                                 const unsigned long timeout,
542                                 const unsigned long max_rto)
543 {
544         struct inet_connection_sock *icsk = inet_csk(parent);
545         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
546         struct listen_sock *lopt = queue->listen_opt;
547         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
548         int thresh = max_retries;
549         unsigned long now = jiffies;
550         struct request_sock **reqp, *req;
551         int i, budget;
552
553         if (lopt == NULL || lopt->qlen == 0)
554                 return;
555
556         /* Normally all the openreqs are young and become mature
557          * (i.e. converted to established socket) for first timeout.
558          * If synack was not acknowledged for 1 second, it means
559          * one of the following things: synack was lost, ack was lost,
560          * rtt is high or nobody planned to ack (i.e. synflood).
561          * When server is a bit loaded, queue is populated with old
562          * open requests, reducing effective size of queue.
563          * When server is well loaded, queue size reduces to zero
564          * after several minutes of work. It is not synflood,
565          * it is normal operation. The solution is pruning
566          * too old entries overriding normal timeout, when
567          * situation becomes dangerous.
568          *
569          * Essentially, we reserve half of room for young
570          * embrions; and abort old ones without pity, if old
571          * ones are about to clog our table.
572          */
573         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
574                 int young = (lopt->qlen_young<<1);
575
576                 while (thresh > 2) {
577                         if (lopt->qlen < young)
578                                 break;
579                         thresh--;
580                         young <<= 1;
581                 }
582         }
583
584         if (queue->rskq_defer_accept)
585                 max_retries = queue->rskq_defer_accept;
586
587         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
588         i = lopt->clock_hand;
589
590         do {
591                 reqp=&lopt->syn_table[i];
592                 while ((req = *reqp) != NULL) {
593                         if (time_after_eq(now, req->expires)) {
594                                 int expire = 0, resend = 0;
595
596                                 syn_ack_recalc(req, thresh, max_retries,
597                                                queue->rskq_defer_accept,
598                                                &expire, &resend);
599                                 req->rsk_ops->syn_ack_timeout(parent, req);
600                                 if (!expire &&
601                                     (!resend ||
602                                      !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
603                                      inet_rsk(req)->acked)) {
604                                         unsigned long timeo;
605
606                                         if (req->retrans++ == 0)
607                                                 lopt->qlen_young--;
608                                         timeo = min((timeout << req->retrans), max_rto);
609                                         req->expires = now + timeo;
610                                         reqp = &req->dl_next;
611                                         continue;
612                                 }
613
614                                 /* Drop this request */
615                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
616                                 reqsk_queue_removed(queue, req);
617                                 reqsk_free(req);
618                                 continue;
619                         }
620                         reqp = &req->dl_next;
621                 }
622
623                 i = (i + 1) & (lopt->nr_table_entries - 1);
624
625         } while (--budget > 0);
626
627         lopt->clock_hand = i;
628
629         if (lopt->qlen)
630                 inet_csk_reset_keepalive_timer(parent, interval);
631 }
632 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
633
634 /**
635  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
636  *      @sk: the socket to clone
637  *      @req: request_sock
638  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
639  *
640  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
641  */
642 struct sock *inet_csk_clone_lock(const struct sock *sk,
643                                  const struct request_sock *req,
644                                  const gfp_t priority)
645 {
646         struct sock *newsk = sk_clone_lock(sk, priority);
647
648         if (newsk != NULL) {
649                 struct inet_connection_sock *newicsk = inet_csk(newsk);
650
651                 newsk->sk_state = TCP_SYN_RECV;
652                 newicsk->icsk_bind_hash = NULL;
653
654                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
655                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
656                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
657                 newsk->sk_write_space = sk_stream_write_space;
658
659                 newicsk->icsk_retransmits = 0;
660                 newicsk->icsk_backoff     = 0;
661                 newicsk->icsk_probes_out  = 0;
662
663                 /* Deinitialize accept_queue to trap illegal accesses. */
664                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
665
666                 security_inet_csk_clone(newsk, req);
667         }
668         return newsk;
669 }
670 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
671
672 /*
673  * At this point, there should be no process reference to this
674  * socket, and thus no user references at all.  Therefore we
675  * can assume the socket waitqueue is inactive and nobody will
676  * try to jump onto it.
677  */
678 void inet_csk_destroy_sock(struct sock *sk)
679 {
680         WARN_ON(sk->sk_state != TCP_CLOSE);
681         WARN_ON(!sock_flag(sk, SOCK_DEAD));
682
683         /* It cannot be in hash table! */
684         WARN_ON(!sk_unhashed(sk));
685
686         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
687         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
688
689         sk->sk_prot->destroy(sk);
690
691         sk_stream_kill_queues(sk);
692
693         xfrm_sk_free_policy(sk);
694
695         sk_refcnt_debug_release(sk);
696
697         percpu_counter_dec(sk->sk_prot->orphan_count);
698         sock_put(sk);
699 }
700 EXPORT_SYMBOL(inet_csk_destroy_sock);
701
702 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
703 {
704         struct inet_sock *inet = inet_sk(sk);
705         struct inet_connection_sock *icsk = inet_csk(sk);
706         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
707
708         if (rc != 0)
709                 return rc;
710
711         sk->sk_max_ack_backlog = 0;
712         sk->sk_ack_backlog = 0;
713         inet_csk_delack_init(sk);
714
715         /* There is race window here: we announce ourselves listening,
716          * but this transition is still not validated by get_port().
717          * It is OK, because this socket enters to hash table only
718          * after validation is complete.
719          */
720         sk->sk_state = TCP_LISTEN;
721         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
722                 inet->inet_sport = htons(inet->inet_num);
723
724                 sk_dst_reset(sk);
725                 sk->sk_prot->hash(sk);
726
727                 return 0;
728         }
729
730         sk->sk_state = TCP_CLOSE;
731         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
732         return -EADDRINUSE;
733 }
734 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
735
736 /*
737  *      This routine closes sockets which have been at least partially
738  *      opened, but not yet accepted.
739  */
740 void inet_csk_listen_stop(struct sock *sk)
741 {
742         struct inet_connection_sock *icsk = inet_csk(sk);
743         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
744         struct request_sock *acc_req;
745         struct request_sock *req;
746
747         inet_csk_delete_keepalive_timer(sk);
748
749         /* make all the listen_opt local to us */
750         acc_req = reqsk_queue_yank_acceptq(queue);
751
752         /* Following specs, it would be better either to send FIN
753          * (and enter FIN-WAIT-1, it is normal close)
754          * or to send active reset (abort).
755          * Certainly, it is pretty dangerous while synflood, but it is
756          * bad justification for our negligence 8)
757          * To be honest, we are not able to make either
758          * of the variants now.                 --ANK
759          */
760         reqsk_queue_destroy(queue);
761
762         while ((req = acc_req) != NULL) {
763                 struct sock *child = req->sk;
764
765                 acc_req = req->dl_next;
766
767                 local_bh_disable();
768                 bh_lock_sock(child);
769                 WARN_ON(sock_owned_by_user(child));
770                 sock_hold(child);
771
772                 sk->sk_prot->disconnect(child, O_NONBLOCK);
773
774                 sock_orphan(child);
775
776                 percpu_counter_inc(sk->sk_prot->orphan_count);
777
778                 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->listener) {
779                         BUG_ON(tcp_sk(child)->fastopen_rsk != req);
780                         BUG_ON(sk != tcp_rsk(req)->listener);
781
782                         /* Paranoid, to prevent race condition if
783                          * an inbound pkt destined for child is
784                          * blocked by sock lock in tcp_v4_rcv().
785                          * Also to satisfy an assertion in
786                          * tcp_v4_destroy_sock().
787                          */
788                         tcp_sk(child)->fastopen_rsk = NULL;
789                         sock_put(sk);
790                 }
791                 inet_csk_destroy_sock(child);
792
793                 bh_unlock_sock(child);
794                 local_bh_enable();
795                 sock_put(child);
796
797                 sk_acceptq_removed(sk);
798                 __reqsk_free(req);
799         }
800         if (queue->fastopenq != NULL) {
801                 /* Free all the reqs queued in rskq_rst_head. */
802                 spin_lock_bh(&queue->fastopenq->lock);
803                 acc_req = queue->fastopenq->rskq_rst_head;
804                 queue->fastopenq->rskq_rst_head = NULL;
805                 spin_unlock_bh(&queue->fastopenq->lock);
806                 while ((req = acc_req) != NULL) {
807                         acc_req = req->dl_next;
808                         __reqsk_free(req);
809                 }
810         }
811         WARN_ON(sk->sk_ack_backlog);
812 }
813 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
814
815 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
816 {
817         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
818         const struct inet_sock *inet = inet_sk(sk);
819
820         sin->sin_family         = AF_INET;
821         sin->sin_addr.s_addr    = inet->inet_daddr;
822         sin->sin_port           = inet->inet_dport;
823 }
824 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
825
826 #ifdef CONFIG_COMPAT
827 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
828                                char __user *optval, int __user *optlen)
829 {
830         const struct inet_connection_sock *icsk = inet_csk(sk);
831
832         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
833                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
834                                                             optval, optlen);
835         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
836                                              optval, optlen);
837 }
838 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
839
840 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
841                                char __user *optval, unsigned int optlen)
842 {
843         const struct inet_connection_sock *icsk = inet_csk(sk);
844
845         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
846                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
847                                                             optval, optlen);
848         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
849                                              optval, optlen);
850 }
851 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
852 #endif
853
854 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
855 {
856         const struct inet_sock *inet = inet_sk(sk);
857         const struct ip_options_rcu *inet_opt;
858         __be32 daddr = inet->inet_daddr;
859         struct flowi4 *fl4;
860         struct rtable *rt;
861
862         rcu_read_lock();
863         inet_opt = rcu_dereference(inet->inet_opt);
864         if (inet_opt && inet_opt->opt.srr)
865                 daddr = inet_opt->opt.faddr;
866         fl4 = &fl->u.ip4;
867         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
868                                    inet->inet_saddr, inet->inet_dport,
869                                    inet->inet_sport, sk->sk_protocol,
870                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
871         if (IS_ERR(rt))
872                 rt = NULL;
873         if (rt)
874                 sk_setup_caps(sk, &rt->dst);
875         rcu_read_unlock();
876
877         return &rt->dst;
878 }
879
880 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
881 {
882         struct dst_entry *dst = __sk_dst_check(sk, 0);
883         struct inet_sock *inet = inet_sk(sk);
884
885         if (!dst) {
886                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
887                 if (!dst)
888                         goto out;
889         }
890         dst->ops->update_pmtu(dst, sk, NULL, mtu);
891
892         dst = __sk_dst_check(sk, 0);
893         if (!dst)
894                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
895 out:
896         return dst;
897 }
898 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);