]>
Commit | Line | Data |
---|---|---|
3f421baa ACM |
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 | ||
3f421baa ACM |
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> | |
a019d6fe | 25 | #include <net/xfrm.h> |
fa76ce73 | 26 | #include <net/tcp.h> |
3f421baa ACM |
27 | |
28 | #ifdef INET_CSK_DEBUG | |
29 | const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n"; | |
30 | EXPORT_SYMBOL(inet_csk_timer_bug_msg); | |
31 | #endif | |
32 | ||
0bbf87d8 | 33 | void inet_get_local_port_range(struct net *net, int *low, int *high) |
227b60f5 | 34 | { |
95c96174 ED |
35 | unsigned int seq; |
36 | ||
227b60f5 | 37 | do { |
c9d8f1a6 | 38 | seq = read_seqbegin(&net->ipv4.ip_local_ports.lock); |
227b60f5 | 39 | |
c9d8f1a6 CW |
40 | *low = net->ipv4.ip_local_ports.range[0]; |
41 | *high = net->ipv4.ip_local_ports.range[1]; | |
42 | } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq)); | |
227b60f5 SH |
43 | } |
44 | EXPORT_SYMBOL(inet_get_local_port_range); | |
3f421baa | 45 | |
971af18b | 46 | int inet_csk_bind_conflict(const struct sock *sk, |
aacd9289 | 47 | const struct inet_bind_bucket *tb, bool relax) |
3f421baa | 48 | { |
3f421baa | 49 | struct sock *sk2; |
3f421baa | 50 | int reuse = sk->sk_reuse; |
da5e3630 TH |
51 | int reuseport = sk->sk_reuseport; |
52 | kuid_t uid = sock_i_uid((struct sock *)sk); | |
3f421baa | 53 | |
7477fd2e PE |
54 | /* |
55 | * Unlike other sk lookup places we do not check | |
56 | * for sk_net here, since _all_ the socks listed | |
57 | * in tb->owners list belong to the same net - the | |
58 | * one this bucket belongs to. | |
59 | */ | |
60 | ||
b67bfe0d | 61 | sk_for_each_bound(sk2, &tb->owners) { |
3f421baa ACM |
62 | if (sk != sk2 && |
63 | !inet_v6_ipv6only(sk2) && | |
64 | (!sk->sk_bound_dev_if || | |
65 | !sk2->sk_bound_dev_if || | |
66 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { | |
da5e3630 TH |
67 | if ((!reuse || !sk2->sk_reuse || |
68 | sk2->sk_state == TCP_LISTEN) && | |
69 | (!reuseport || !sk2->sk_reuseport || | |
70 | (sk2->sk_state != TCP_TIME_WAIT && | |
71 | !uid_eq(uid, sock_i_uid(sk2))))) { | |
50805466 ED |
72 | |
73 | if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr || | |
74 | sk2->sk_rcv_saddr == sk->sk_rcv_saddr) | |
3f421baa | 75 | break; |
8d238b25 | 76 | } |
aacd9289 AC |
77 | if (!relax && reuse && sk2->sk_reuse && |
78 | sk2->sk_state != TCP_LISTEN) { | |
aacd9289 | 79 | |
50805466 ED |
80 | if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr || |
81 | sk2->sk_rcv_saddr == sk->sk_rcv_saddr) | |
aacd9289 AC |
82 | break; |
83 | } | |
3f421baa ACM |
84 | } |
85 | } | |
b67bfe0d | 86 | return sk2 != NULL; |
3f421baa | 87 | } |
971af18b ACM |
88 | EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); |
89 | ||
3f421baa ACM |
90 | /* Obtain a reference to a local port for the given sock, |
91 | * if snum is zero it means select any available local port. | |
92 | */ | |
ab1e0a13 | 93 | int inet_csk_get_port(struct sock *sk, unsigned short snum) |
3f421baa | 94 | { |
39d8cda7 | 95 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3f421baa | 96 | struct inet_bind_hashbucket *head; |
3f421baa | 97 | struct inet_bind_bucket *tb; |
a9d8f911 | 98 | int ret, attempts = 5; |
3b1e0a65 | 99 | struct net *net = sock_net(sk); |
a9d8f911 | 100 | int smallest_size = -1, smallest_rover; |
da5e3630 | 101 | kuid_t uid = sock_i_uid(sk); |
3f421baa ACM |
102 | |
103 | local_bh_disable(); | |
104 | if (!snum) { | |
227b60f5 SH |
105 | int remaining, rover, low, high; |
106 | ||
a9d8f911 | 107 | again: |
0bbf87d8 | 108 | inet_get_local_port_range(net, &low, &high); |
a25de534 | 109 | remaining = (high - low) + 1; |
63862b5b | 110 | smallest_rover = rover = prandom_u32() % remaining + low; |
3f421baa | 111 | |
a9d8f911 | 112 | smallest_size = -1; |
3f421baa | 113 | do { |
122ff243 | 114 | if (inet_is_local_reserved_port(net, rover)) |
e3826f1e | 115 | goto next_nolock; |
7f635ab7 PE |
116 | head = &hashinfo->bhash[inet_bhashfn(net, rover, |
117 | hashinfo->bhash_size)]; | |
3f421baa | 118 | spin_lock(&head->lock); |
b67bfe0d | 119 | inet_bind_bucket_for_each(tb, &head->chain) |
09ad9bc7 | 120 | if (net_eq(ib_net(tb), net) && tb->port == rover) { |
da5e3630 TH |
121 | if (((tb->fastreuse > 0 && |
122 | sk->sk_reuse && | |
123 | sk->sk_state != TCP_LISTEN) || | |
124 | (tb->fastreuseport > 0 && | |
125 | sk->sk_reuseport && | |
126 | uid_eq(tb->fastuid, uid))) && | |
a9d8f911 EP |
127 | (tb->num_owners < smallest_size || smallest_size == -1)) { |
128 | smallest_size = tb->num_owners; | |
129 | smallest_rover = rover; | |
a9d8f911 | 130 | } |
aacd9289 | 131 | if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) { |
2b05ad33 | 132 | snum = rover; |
fddb7b57 | 133 | goto tb_found; |
2b05ad33 | 134 | } |
3f421baa | 135 | goto next; |
a9d8f911 | 136 | } |
3f421baa ACM |
137 | break; |
138 | next: | |
139 | spin_unlock(&head->lock); | |
e3826f1e | 140 | next_nolock: |
6df71634 SH |
141 | if (++rover > high) |
142 | rover = low; | |
3f421baa | 143 | } while (--remaining > 0); |
3f421baa ACM |
144 | |
145 | /* Exhausted local port range during search? It is not | |
146 | * possible for us to be holding one of the bind hash | |
147 | * locks if this test triggers, because if 'remaining' | |
148 | * drops to zero, we broke out of the do/while loop at | |
149 | * the top level, not from the 'break;' statement. | |
150 | */ | |
151 | ret = 1; | |
a9d8f911 EP |
152 | if (remaining <= 0) { |
153 | if (smallest_size != -1) { | |
154 | snum = smallest_rover; | |
155 | goto have_snum; | |
156 | } | |
3f421baa | 157 | goto fail; |
a9d8f911 | 158 | } |
3f421baa ACM |
159 | /* OK, here is the one we will use. HEAD is |
160 | * non-NULL and we hold it's mutex. | |
161 | */ | |
162 | snum = rover; | |
163 | } else { | |
a9d8f911 | 164 | have_snum: |
7f635ab7 PE |
165 | head = &hashinfo->bhash[inet_bhashfn(net, snum, |
166 | hashinfo->bhash_size)]; | |
3f421baa | 167 | spin_lock(&head->lock); |
b67bfe0d | 168 | inet_bind_bucket_for_each(tb, &head->chain) |
09ad9bc7 | 169 | if (net_eq(ib_net(tb), net) && tb->port == snum) |
3f421baa ACM |
170 | goto tb_found; |
171 | } | |
172 | tb = NULL; | |
173 | goto tb_not_found; | |
174 | tb_found: | |
175 | if (!hlist_empty(&tb->owners)) { | |
4a17fd52 PE |
176 | if (sk->sk_reuse == SK_FORCE_REUSE) |
177 | goto success; | |
178 | ||
da5e3630 TH |
179 | if (((tb->fastreuse > 0 && |
180 | sk->sk_reuse && sk->sk_state != TCP_LISTEN) || | |
181 | (tb->fastreuseport > 0 && | |
182 | sk->sk_reuseport && uid_eq(tb->fastuid, uid))) && | |
a9d8f911 | 183 | smallest_size == -1) { |
3f421baa ACM |
184 | goto success; |
185 | } else { | |
186 | ret = 1; | |
aacd9289 | 187 | if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) { |
da5e3630 | 188 | if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) || |
9c5e0c0b TH |
189 | (tb->fastreuseport > 0 && |
190 | sk->sk_reuseport && uid_eq(tb->fastuid, uid))) && | |
5add3009 | 191 | smallest_size != -1 && --attempts >= 0) { |
a9d8f911 EP |
192 | spin_unlock(&head->lock); |
193 | goto again; | |
194 | } | |
aacd9289 | 195 | |
3f421baa | 196 | goto fail_unlock; |
a9d8f911 | 197 | } |
3f421baa ACM |
198 | } |
199 | } | |
200 | tb_not_found: | |
201 | ret = 1; | |
941b1d22 PE |
202 | if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, |
203 | net, head, snum)) == NULL) | |
3f421baa ACM |
204 | goto fail_unlock; |
205 | if (hlist_empty(&tb->owners)) { | |
206 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) | |
207 | tb->fastreuse = 1; | |
208 | else | |
209 | tb->fastreuse = 0; | |
da5e3630 TH |
210 | if (sk->sk_reuseport) { |
211 | tb->fastreuseport = 1; | |
212 | tb->fastuid = uid; | |
9c5e0c0b | 213 | } else |
da5e3630 | 214 | tb->fastreuseport = 0; |
da5e3630 TH |
215 | } else { |
216 | if (tb->fastreuse && | |
217 | (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) | |
218 | tb->fastreuse = 0; | |
219 | if (tb->fastreuseport && | |
9c5e0c0b | 220 | (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid))) |
da5e3630 | 221 | tb->fastreuseport = 0; |
da5e3630 | 222 | } |
3f421baa ACM |
223 | success: |
224 | if (!inet_csk(sk)->icsk_bind_hash) | |
225 | inet_bind_hash(sk, tb, snum); | |
547b792c | 226 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
e905a9ed | 227 | ret = 0; |
3f421baa ACM |
228 | |
229 | fail_unlock: | |
230 | spin_unlock(&head->lock); | |
231 | fail: | |
232 | local_bh_enable(); | |
233 | return ret; | |
234 | } | |
3f421baa ACM |
235 | EXPORT_SYMBOL_GPL(inet_csk_get_port); |
236 | ||
237 | /* | |
238 | * Wait for an incoming connection, avoid race conditions. This must be called | |
239 | * with the socket locked. | |
240 | */ | |
241 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) | |
242 | { | |
243 | struct inet_connection_sock *icsk = inet_csk(sk); | |
244 | DEFINE_WAIT(wait); | |
245 | int err; | |
246 | ||
247 | /* | |
248 | * True wake-one mechanism for incoming connections: only | |
249 | * one process gets woken up, not the 'whole herd'. | |
250 | * Since we do not 'race & poll' for established sockets | |
251 | * anymore, the common case will execute the loop only once. | |
252 | * | |
253 | * Subtle issue: "add_wait_queue_exclusive()" will be added | |
254 | * after any current non-exclusive waiters, and we know that | |
255 | * it will always _stay_ after any new non-exclusive waiters | |
256 | * because all non-exclusive waiters are added at the | |
257 | * beginning of the wait-queue. As such, it's ok to "drop" | |
258 | * our exclusiveness temporarily when we get woken up without | |
259 | * having to remove and re-insert us on the wait queue. | |
260 | */ | |
261 | for (;;) { | |
aa395145 | 262 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, |
3f421baa ACM |
263 | TASK_INTERRUPTIBLE); |
264 | release_sock(sk); | |
265 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
266 | timeo = schedule_timeout(timeo); | |
cb7cf8a3 | 267 | sched_annotate_sleep(); |
3f421baa ACM |
268 | lock_sock(sk); |
269 | err = 0; | |
270 | if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
271 | break; | |
272 | err = -EINVAL; | |
273 | if (sk->sk_state != TCP_LISTEN) | |
274 | break; | |
275 | err = sock_intr_errno(timeo); | |
276 | if (signal_pending(current)) | |
277 | break; | |
278 | err = -EAGAIN; | |
279 | if (!timeo) | |
280 | break; | |
281 | } | |
aa395145 | 282 | finish_wait(sk_sleep(sk), &wait); |
3f421baa ACM |
283 | return err; |
284 | } | |
285 | ||
286 | /* | |
287 | * This will accept the next outstanding connection. | |
288 | */ | |
289 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) | |
290 | { | |
291 | struct inet_connection_sock *icsk = inet_csk(sk); | |
8336886f | 292 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
8336886f | 293 | struct request_sock *req; |
e3d95ad7 | 294 | struct sock *newsk; |
3f421baa ACM |
295 | int error; |
296 | ||
297 | lock_sock(sk); | |
298 | ||
299 | /* We need to make sure that this socket is listening, | |
300 | * and that it has something pending. | |
301 | */ | |
302 | error = -EINVAL; | |
303 | if (sk->sk_state != TCP_LISTEN) | |
304 | goto out_err; | |
305 | ||
306 | /* Find already established connection */ | |
8336886f | 307 | if (reqsk_queue_empty(queue)) { |
3f421baa ACM |
308 | long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); |
309 | ||
310 | /* If this is a non blocking socket don't sleep */ | |
311 | error = -EAGAIN; | |
312 | if (!timeo) | |
313 | goto out_err; | |
314 | ||
315 | error = inet_csk_wait_for_connect(sk, timeo); | |
316 | if (error) | |
317 | goto out_err; | |
318 | } | |
8336886f JC |
319 | req = reqsk_queue_remove(queue); |
320 | newsk = req->sk; | |
321 | ||
322 | sk_acceptq_removed(sk); | |
e3d95ad7 ED |
323 | if (sk->sk_protocol == IPPROTO_TCP && |
324 | tcp_rsk(req)->tfo_listener && | |
325 | queue->fastopenq) { | |
8336886f | 326 | spin_lock_bh(&queue->fastopenq->lock); |
9439ce00 | 327 | if (tcp_rsk(req)->tfo_listener) { |
8336886f JC |
328 | /* We are still waiting for the final ACK from 3WHS |
329 | * so can't free req now. Instead, we set req->sk to | |
330 | * NULL to signify that the child socket is taken | |
331 | * so reqsk_fastopen_remove() will free the req | |
332 | * when 3WHS finishes (or is aborted). | |
333 | */ | |
334 | req->sk = NULL; | |
335 | req = NULL; | |
336 | } | |
337 | spin_unlock_bh(&queue->fastopenq->lock); | |
338 | } | |
3f421baa ACM |
339 | out: |
340 | release_sock(sk); | |
8336886f | 341 | if (req) |
13854e5a | 342 | reqsk_put(req); |
3f421baa ACM |
343 | return newsk; |
344 | out_err: | |
345 | newsk = NULL; | |
8336886f | 346 | req = NULL; |
3f421baa ACM |
347 | *err = error; |
348 | goto out; | |
349 | } | |
3f421baa ACM |
350 | EXPORT_SYMBOL(inet_csk_accept); |
351 | ||
352 | /* | |
353 | * Using different timers for retransmit, delayed acks and probes | |
e905a9ed | 354 | * We may wish use just one timer maintaining a list of expire jiffies |
3f421baa ACM |
355 | * to optimize. |
356 | */ | |
357 | void inet_csk_init_xmit_timers(struct sock *sk, | |
358 | void (*retransmit_handler)(unsigned long), | |
359 | void (*delack_handler)(unsigned long), | |
360 | void (*keepalive_handler)(unsigned long)) | |
361 | { | |
362 | struct inet_connection_sock *icsk = inet_csk(sk); | |
363 | ||
b24b8a24 PE |
364 | setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, |
365 | (unsigned long)sk); | |
366 | setup_timer(&icsk->icsk_delack_timer, delack_handler, | |
367 | (unsigned long)sk); | |
368 | setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); | |
3f421baa ACM |
369 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
370 | } | |
3f421baa ACM |
371 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); |
372 | ||
373 | void inet_csk_clear_xmit_timers(struct sock *sk) | |
374 | { | |
375 | struct inet_connection_sock *icsk = inet_csk(sk); | |
376 | ||
377 | icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; | |
378 | ||
379 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
380 | sk_stop_timer(sk, &icsk->icsk_delack_timer); | |
381 | sk_stop_timer(sk, &sk->sk_timer); | |
382 | } | |
3f421baa ACM |
383 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); |
384 | ||
385 | void inet_csk_delete_keepalive_timer(struct sock *sk) | |
386 | { | |
387 | sk_stop_timer(sk, &sk->sk_timer); | |
388 | } | |
3f421baa ACM |
389 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); |
390 | ||
391 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) | |
392 | { | |
393 | sk_reset_timer(sk, &sk->sk_timer, jiffies + len); | |
394 | } | |
3f421baa ACM |
395 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); |
396 | ||
d9319100 | 397 | struct dst_entry *inet_csk_route_req(struct sock *sk, |
6bd023f3 | 398 | struct flowi4 *fl4, |
ba3f7f04 | 399 | const struct request_sock *req) |
3f421baa | 400 | { |
3f421baa | 401 | const struct inet_request_sock *ireq = inet_rsk(req); |
8b929ab1 ED |
402 | struct net *net = read_pnet(&ireq->ireq_net); |
403 | struct ip_options_rcu *opt = ireq->opt; | |
404 | struct rtable *rt; | |
3f421baa | 405 | |
8b929ab1 | 406 | flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, |
e79d9bc7 | 407 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
8b929ab1 | 408 | sk->sk_protocol, inet_sk_flowi_flags(sk), |
634fb979 | 409 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
8b929ab1 ED |
410 | ireq->ir_loc_addr, ireq->ir_rmt_port, |
411 | htons(ireq->ir_num)); | |
6bd023f3 DM |
412 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); |
413 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe | 414 | if (IS_ERR(rt)) |
857a6e0a | 415 | goto no_route; |
155e8336 | 416 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
857a6e0a | 417 | goto route_err; |
d8d1f30b | 418 | return &rt->dst; |
857a6e0a IJ |
419 | |
420 | route_err: | |
421 | ip_rt_put(rt); | |
422 | no_route: | |
423 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); | |
424 | return NULL; | |
3f421baa | 425 | } |
3f421baa ACM |
426 | EXPORT_SYMBOL_GPL(inet_csk_route_req); |
427 | ||
77357a95 DM |
428 | struct dst_entry *inet_csk_route_child_sock(struct sock *sk, |
429 | struct sock *newsk, | |
430 | const struct request_sock *req) | |
431 | { | |
432 | const struct inet_request_sock *ireq = inet_rsk(req); | |
8b929ab1 | 433 | struct net *net = read_pnet(&ireq->ireq_net); |
77357a95 | 434 | struct inet_sock *newinet = inet_sk(newsk); |
1a7b27c9 | 435 | struct ip_options_rcu *opt; |
77357a95 DM |
436 | struct flowi4 *fl4; |
437 | struct rtable *rt; | |
438 | ||
439 | fl4 = &newinet->cork.fl.u.ip4; | |
1a7b27c9 CP |
440 | |
441 | rcu_read_lock(); | |
442 | opt = rcu_dereference(newinet->inet_opt); | |
8b929ab1 | 443 | flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, |
77357a95 DM |
444 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
445 | sk->sk_protocol, inet_sk_flowi_flags(sk), | |
634fb979 | 446 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
8b929ab1 ED |
447 | ireq->ir_loc_addr, ireq->ir_rmt_port, |
448 | htons(ireq->ir_num)); | |
77357a95 DM |
449 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); |
450 | rt = ip_route_output_flow(net, fl4, sk); | |
451 | if (IS_ERR(rt)) | |
452 | goto no_route; | |
155e8336 | 453 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
77357a95 | 454 | goto route_err; |
1a7b27c9 | 455 | rcu_read_unlock(); |
77357a95 DM |
456 | return &rt->dst; |
457 | ||
458 | route_err: | |
459 | ip_rt_put(rt); | |
460 | no_route: | |
1a7b27c9 | 461 | rcu_read_unlock(); |
77357a95 DM |
462 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); |
463 | return NULL; | |
464 | } | |
465 | EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); | |
466 | ||
6b72977b | 467 | static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport, |
72a3effa | 468 | const u32 rnd, const u32 synq_hsize) |
3f421baa | 469 | { |
6b72977b | 470 | return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1); |
3f421baa ACM |
471 | } |
472 | ||
dfd56b8b | 473 | #if IS_ENABLED(CONFIG_IPV6) |
3f421baa ACM |
474 | #define AF_INET_FAMILY(fam) ((fam) == AF_INET) |
475 | #else | |
fa76ce73 | 476 | #define AF_INET_FAMILY(fam) true |
3f421baa ACM |
477 | #endif |
478 | ||
fa76ce73 ED |
479 | /* Note: this is temporary : |
480 | * req sock will no longer be in listener hash table | |
481 | */ | |
482 | struct request_sock *inet_csk_search_req(struct sock *sk, | |
483 | const __be16 rport, | |
484 | const __be32 raddr, | |
7f25afbb | 485 | const __be32 laddr) |
3f421baa | 486 | { |
fa76ce73 | 487 | struct inet_connection_sock *icsk = inet_csk(sk); |
3f421baa | 488 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; |
52452c54 | 489 | struct request_sock *req; |
fa76ce73 ED |
490 | u32 hash = inet_synq_hash(raddr, rport, lopt->hash_rnd, |
491 | lopt->nr_table_entries); | |
3f421baa | 492 | |
b2827053 | 493 | spin_lock(&icsk->icsk_accept_queue.syn_wait_lock); |
fa76ce73 | 494 | for (req = lopt->syn_table[hash]; req != NULL; req = req->dl_next) { |
3f421baa ACM |
495 | const struct inet_request_sock *ireq = inet_rsk(req); |
496 | ||
634fb979 ED |
497 | if (ireq->ir_rmt_port == rport && |
498 | ireq->ir_rmt_addr == raddr && | |
499 | ireq->ir_loc_addr == laddr && | |
3f421baa | 500 | AF_INET_FAMILY(req->rsk_ops->family)) { |
fa76ce73 | 501 | atomic_inc(&req->rsk_refcnt); |
547b792c | 502 | WARN_ON(req->sk); |
3f421baa ACM |
503 | break; |
504 | } | |
505 | } | |
b2827053 | 506 | spin_unlock(&icsk->icsk_accept_queue.syn_wait_lock); |
3f421baa ACM |
507 | |
508 | return req; | |
509 | } | |
3f421baa ACM |
510 | EXPORT_SYMBOL_GPL(inet_csk_search_req); |
511 | ||
512 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, | |
c2977c22 | 513 | unsigned long timeout) |
3f421baa ACM |
514 | { |
515 | struct inet_connection_sock *icsk = inet_csk(sk); | |
516 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
634fb979 ED |
517 | const u32 h = inet_synq_hash(inet_rsk(req)->ir_rmt_addr, |
518 | inet_rsk(req)->ir_rmt_port, | |
3f421baa ACM |
519 | lopt->hash_rnd, lopt->nr_table_entries); |
520 | ||
521 | reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout); | |
522 | inet_csk_reqsk_queue_added(sk, timeout); | |
523 | } | |
4bc2f18b | 524 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
3f421baa | 525 | |
a019d6fe ACM |
526 | /* Only thing we need from tcp.h */ |
527 | extern int sysctl_tcp_synack_retries; | |
528 | ||
9f1d2604 | 529 | |
0c3d79bc JA |
530 | /* Decide when to expire the request and when to resend SYN-ACK */ |
531 | static inline void syn_ack_recalc(struct request_sock *req, const int thresh, | |
532 | const int max_retries, | |
533 | const u8 rskq_defer_accept, | |
534 | int *expire, int *resend) | |
535 | { | |
536 | if (!rskq_defer_accept) { | |
e6c022a4 | 537 | *expire = req->num_timeout >= thresh; |
0c3d79bc JA |
538 | *resend = 1; |
539 | return; | |
540 | } | |
e6c022a4 ED |
541 | *expire = req->num_timeout >= thresh && |
542 | (!inet_rsk(req)->acked || req->num_timeout >= max_retries); | |
0c3d79bc JA |
543 | /* |
544 | * Do not resend while waiting for data after ACK, | |
545 | * start to resend on end of deferring period to give | |
546 | * last chance for data or ACK to create established socket. | |
547 | */ | |
548 | *resend = !inet_rsk(req)->acked || | |
e6c022a4 | 549 | req->num_timeout >= rskq_defer_accept - 1; |
0c3d79bc JA |
550 | } |
551 | ||
e6c022a4 ED |
552 | int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req) |
553 | { | |
1a2c6181 | 554 | int err = req->rsk_ops->rtx_syn_ack(parent, req); |
e6c022a4 ED |
555 | |
556 | if (!err) | |
557 | req->num_retrans++; | |
558 | return err; | |
559 | } | |
560 | EXPORT_SYMBOL(inet_rtx_syn_ack); | |
561 | ||
b357a364 ED |
562 | /* return true if req was found in the syn_table[] */ |
563 | static bool reqsk_queue_unlink(struct request_sock_queue *queue, | |
564 | struct request_sock *req) | |
565 | { | |
566 | struct listen_sock *lopt = queue->listen_opt; | |
567 | struct request_sock **prev; | |
568 | bool found = false; | |
569 | ||
570 | spin_lock(&queue->syn_wait_lock); | |
571 | ||
572 | for (prev = &lopt->syn_table[req->rsk_hash]; *prev != NULL; | |
573 | prev = &(*prev)->dl_next) { | |
574 | if (*prev == req) { | |
575 | *prev = req->dl_next; | |
576 | found = true; | |
577 | break; | |
578 | } | |
579 | } | |
580 | ||
581 | spin_unlock(&queue->syn_wait_lock); | |
582 | if (del_timer(&req->rsk_timer)) | |
583 | reqsk_put(req); | |
584 | return found; | |
585 | } | |
586 | ||
587 | void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req) | |
588 | { | |
589 | if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) { | |
590 | reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); | |
591 | reqsk_put(req); | |
592 | } | |
593 | } | |
594 | EXPORT_SYMBOL(inet_csk_reqsk_queue_drop); | |
595 | ||
fa76ce73 | 596 | static void reqsk_timer_handler(unsigned long data) |
a019d6fe | 597 | { |
fa76ce73 ED |
598 | struct request_sock *req = (struct request_sock *)data; |
599 | struct sock *sk_listener = req->rsk_listener; | |
600 | struct inet_connection_sock *icsk = inet_csk(sk_listener); | |
a019d6fe ACM |
601 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
602 | struct listen_sock *lopt = queue->listen_opt; | |
2b41fab7 | 603 | int qlen, expire = 0, resend = 0; |
fa76ce73 | 604 | int max_retries, thresh; |
2b41fab7 | 605 | u8 defer_accept; |
a019d6fe | 606 | |
fa76ce73 ED |
607 | if (sk_listener->sk_state != TCP_LISTEN || !lopt) { |
608 | reqsk_put(req); | |
a019d6fe | 609 | return; |
fa76ce73 | 610 | } |
a019d6fe | 611 | |
fa76ce73 ED |
612 | max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries; |
613 | thresh = max_retries; | |
a019d6fe ACM |
614 | /* Normally all the openreqs are young and become mature |
615 | * (i.e. converted to established socket) for first timeout. | |
fd4f2cea | 616 | * If synack was not acknowledged for 1 second, it means |
a019d6fe ACM |
617 | * one of the following things: synack was lost, ack was lost, |
618 | * rtt is high or nobody planned to ack (i.e. synflood). | |
619 | * When server is a bit loaded, queue is populated with old | |
620 | * open requests, reducing effective size of queue. | |
621 | * When server is well loaded, queue size reduces to zero | |
622 | * after several minutes of work. It is not synflood, | |
623 | * it is normal operation. The solution is pruning | |
624 | * too old entries overriding normal timeout, when | |
625 | * situation becomes dangerous. | |
626 | * | |
627 | * Essentially, we reserve half of room for young | |
628 | * embrions; and abort old ones without pity, if old | |
629 | * ones are about to clog our table. | |
630 | */ | |
2b41fab7 ED |
631 | qlen = listen_sock_qlen(lopt); |
632 | if (qlen >> (lopt->max_qlen_log - 1)) { | |
fa76ce73 | 633 | int young = listen_sock_young(lopt) << 1; |
a019d6fe ACM |
634 | |
635 | while (thresh > 2) { | |
2b41fab7 | 636 | if (qlen < young) |
a019d6fe ACM |
637 | break; |
638 | thresh--; | |
639 | young <<= 1; | |
640 | } | |
641 | } | |
2b41fab7 ED |
642 | defer_accept = READ_ONCE(queue->rskq_defer_accept); |
643 | if (defer_accept) | |
644 | max_retries = defer_accept; | |
645 | syn_ack_recalc(req, thresh, max_retries, defer_accept, | |
fa76ce73 | 646 | &expire, &resend); |
42cb80a2 | 647 | req->rsk_ops->syn_ack_timeout(req); |
fa76ce73 ED |
648 | if (!expire && |
649 | (!resend || | |
650 | !inet_rtx_syn_ack(sk_listener, req) || | |
651 | inet_rsk(req)->acked)) { | |
652 | unsigned long timeo; | |
653 | ||
654 | if (req->num_timeout++ == 0) | |
655 | atomic_inc(&lopt->young_dec); | |
656 | timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); | |
657 | mod_timer_pinned(&req->rsk_timer, jiffies + timeo); | |
658 | return; | |
659 | } | |
660 | inet_csk_reqsk_queue_drop(sk_listener, req); | |
661 | reqsk_put(req); | |
662 | } | |
ec0a1966 | 663 | |
fa76ce73 ED |
664 | void reqsk_queue_hash_req(struct request_sock_queue *queue, |
665 | u32 hash, struct request_sock *req, | |
666 | unsigned long timeout) | |
667 | { | |
668 | struct listen_sock *lopt = queue->listen_opt; | |
a019d6fe | 669 | |
fa76ce73 ED |
670 | req->num_retrans = 0; |
671 | req->num_timeout = 0; | |
672 | req->sk = NULL; | |
a019d6fe | 673 | |
fa76ce73 ED |
674 | /* before letting lookups find us, make sure all req fields |
675 | * are committed to memory and refcnt initialized. | |
676 | */ | |
677 | smp_wmb(); | |
678 | atomic_set(&req->rsk_refcnt, 2); | |
679 | setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req); | |
680 | req->rsk_hash = hash; | |
a019d6fe | 681 | |
b2827053 | 682 | spin_lock(&queue->syn_wait_lock); |
fa76ce73 ED |
683 | req->dl_next = lopt->syn_table[hash]; |
684 | lopt->syn_table[hash] = req; | |
b2827053 | 685 | spin_unlock(&queue->syn_wait_lock); |
a019d6fe | 686 | |
fa76ce73 | 687 | mod_timer_pinned(&req->rsk_timer, jiffies + timeout); |
a019d6fe | 688 | } |
fa76ce73 | 689 | EXPORT_SYMBOL(reqsk_queue_hash_req); |
a019d6fe | 690 | |
e56c57d0 ED |
691 | /** |
692 | * inet_csk_clone_lock - clone an inet socket, and lock its clone | |
693 | * @sk: the socket to clone | |
694 | * @req: request_sock | |
695 | * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) | |
696 | * | |
697 | * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) | |
698 | */ | |
699 | struct sock *inet_csk_clone_lock(const struct sock *sk, | |
700 | const struct request_sock *req, | |
701 | const gfp_t priority) | |
9f1d2604 | 702 | { |
e56c57d0 | 703 | struct sock *newsk = sk_clone_lock(sk, priority); |
9f1d2604 | 704 | |
00db4124 | 705 | if (newsk) { |
9f1d2604 ACM |
706 | struct inet_connection_sock *newicsk = inet_csk(newsk); |
707 | ||
708 | newsk->sk_state = TCP_SYN_RECV; | |
709 | newicsk->icsk_bind_hash = NULL; | |
710 | ||
634fb979 | 711 | inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port; |
b44084c2 ED |
712 | inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num; |
713 | inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num); | |
9f1d2604 ACM |
714 | newsk->sk_write_space = sk_stream_write_space; |
715 | ||
84f39b08 | 716 | newsk->sk_mark = inet_rsk(req)->ir_mark; |
33cf7c90 ED |
717 | atomic64_set(&newsk->sk_cookie, |
718 | atomic64_read(&inet_rsk(req)->ir_cookie)); | |
84f39b08 | 719 | |
9f1d2604 | 720 | newicsk->icsk_retransmits = 0; |
6687e988 ACM |
721 | newicsk->icsk_backoff = 0; |
722 | newicsk->icsk_probes_out = 0; | |
9f1d2604 ACM |
723 | |
724 | /* Deinitialize accept_queue to trap illegal accesses. */ | |
725 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); | |
4237c75c VY |
726 | |
727 | security_inet_csk_clone(newsk, req); | |
9f1d2604 ACM |
728 | } |
729 | return newsk; | |
730 | } | |
e56c57d0 | 731 | EXPORT_SYMBOL_GPL(inet_csk_clone_lock); |
a019d6fe ACM |
732 | |
733 | /* | |
734 | * At this point, there should be no process reference to this | |
735 | * socket, and thus no user references at all. Therefore we | |
736 | * can assume the socket waitqueue is inactive and nobody will | |
737 | * try to jump onto it. | |
738 | */ | |
739 | void inet_csk_destroy_sock(struct sock *sk) | |
740 | { | |
547b792c IJ |
741 | WARN_ON(sk->sk_state != TCP_CLOSE); |
742 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); | |
a019d6fe ACM |
743 | |
744 | /* It cannot be in hash table! */ | |
547b792c | 745 | WARN_ON(!sk_unhashed(sk)); |
a019d6fe | 746 | |
c720c7e8 ED |
747 | /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ |
748 | WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); | |
a019d6fe ACM |
749 | |
750 | sk->sk_prot->destroy(sk); | |
751 | ||
752 | sk_stream_kill_queues(sk); | |
753 | ||
754 | xfrm_sk_free_policy(sk); | |
755 | ||
756 | sk_refcnt_debug_release(sk); | |
757 | ||
dd24c001 | 758 | percpu_counter_dec(sk->sk_prot->orphan_count); |
a019d6fe ACM |
759 | sock_put(sk); |
760 | } | |
a019d6fe ACM |
761 | EXPORT_SYMBOL(inet_csk_destroy_sock); |
762 | ||
e337e24d CP |
763 | /* This function allows to force a closure of a socket after the call to |
764 | * tcp/dccp_create_openreq_child(). | |
765 | */ | |
766 | void inet_csk_prepare_forced_close(struct sock *sk) | |
c10cb5fc | 767 | __releases(&sk->sk_lock.slock) |
e337e24d CP |
768 | { |
769 | /* sk_clone_lock locked the socket and set refcnt to 2 */ | |
770 | bh_unlock_sock(sk); | |
771 | sock_put(sk); | |
772 | ||
773 | /* The below has to be done to allow calling inet_csk_destroy_sock */ | |
774 | sock_set_flag(sk, SOCK_DEAD); | |
775 | percpu_counter_inc(sk->sk_prot->orphan_count); | |
776 | inet_sk(sk)->inet_num = 0; | |
777 | } | |
778 | EXPORT_SYMBOL(inet_csk_prepare_forced_close); | |
779 | ||
a019d6fe ACM |
780 | int inet_csk_listen_start(struct sock *sk, const int nr_table_entries) |
781 | { | |
782 | struct inet_sock *inet = inet_sk(sk); | |
783 | struct inet_connection_sock *icsk = inet_csk(sk); | |
784 | int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries); | |
785 | ||
786 | if (rc != 0) | |
787 | return rc; | |
788 | ||
789 | sk->sk_max_ack_backlog = 0; | |
790 | sk->sk_ack_backlog = 0; | |
791 | inet_csk_delack_init(sk); | |
792 | ||
793 | /* There is race window here: we announce ourselves listening, | |
794 | * but this transition is still not validated by get_port(). | |
795 | * It is OK, because this socket enters to hash table only | |
796 | * after validation is complete. | |
797 | */ | |
798 | sk->sk_state = TCP_LISTEN; | |
c720c7e8 ED |
799 | if (!sk->sk_prot->get_port(sk, inet->inet_num)) { |
800 | inet->inet_sport = htons(inet->inet_num); | |
a019d6fe ACM |
801 | |
802 | sk_dst_reset(sk); | |
803 | sk->sk_prot->hash(sk); | |
804 | ||
805 | return 0; | |
806 | } | |
807 | ||
808 | sk->sk_state = TCP_CLOSE; | |
809 | __reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
810 | return -EADDRINUSE; | |
811 | } | |
a019d6fe ACM |
812 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); |
813 | ||
814 | /* | |
815 | * This routine closes sockets which have been at least partially | |
816 | * opened, but not yet accepted. | |
817 | */ | |
818 | void inet_csk_listen_stop(struct sock *sk) | |
819 | { | |
820 | struct inet_connection_sock *icsk = inet_csk(sk); | |
8336886f | 821 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
a019d6fe ACM |
822 | struct request_sock *acc_req; |
823 | struct request_sock *req; | |
824 | ||
a019d6fe | 825 | /* make all the listen_opt local to us */ |
8336886f | 826 | acc_req = reqsk_queue_yank_acceptq(queue); |
a019d6fe ACM |
827 | |
828 | /* Following specs, it would be better either to send FIN | |
829 | * (and enter FIN-WAIT-1, it is normal close) | |
830 | * or to send active reset (abort). | |
831 | * Certainly, it is pretty dangerous while synflood, but it is | |
832 | * bad justification for our negligence 8) | |
833 | * To be honest, we are not able to make either | |
834 | * of the variants now. --ANK | |
835 | */ | |
8336886f | 836 | reqsk_queue_destroy(queue); |
a019d6fe ACM |
837 | |
838 | while ((req = acc_req) != NULL) { | |
839 | struct sock *child = req->sk; | |
840 | ||
841 | acc_req = req->dl_next; | |
842 | ||
843 | local_bh_disable(); | |
844 | bh_lock_sock(child); | |
547b792c | 845 | WARN_ON(sock_owned_by_user(child)); |
a019d6fe ACM |
846 | sock_hold(child); |
847 | ||
848 | sk->sk_prot->disconnect(child, O_NONBLOCK); | |
849 | ||
850 | sock_orphan(child); | |
851 | ||
eb4dea58 HX |
852 | percpu_counter_inc(sk->sk_prot->orphan_count); |
853 | ||
9439ce00 | 854 | if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) { |
8336886f | 855 | BUG_ON(tcp_sk(child)->fastopen_rsk != req); |
9439ce00 | 856 | BUG_ON(sk != req->rsk_listener); |
8336886f JC |
857 | |
858 | /* Paranoid, to prevent race condition if | |
859 | * an inbound pkt destined for child is | |
860 | * blocked by sock lock in tcp_v4_rcv(). | |
861 | * Also to satisfy an assertion in | |
862 | * tcp_v4_destroy_sock(). | |
863 | */ | |
864 | tcp_sk(child)->fastopen_rsk = NULL; | |
8336886f | 865 | } |
a019d6fe ACM |
866 | inet_csk_destroy_sock(child); |
867 | ||
868 | bh_unlock_sock(child); | |
869 | local_bh_enable(); | |
870 | sock_put(child); | |
871 | ||
872 | sk_acceptq_removed(sk); | |
13854e5a | 873 | reqsk_put(req); |
a019d6fe | 874 | } |
00db4124 | 875 | if (queue->fastopenq) { |
8336886f JC |
876 | /* Free all the reqs queued in rskq_rst_head. */ |
877 | spin_lock_bh(&queue->fastopenq->lock); | |
878 | acc_req = queue->fastopenq->rskq_rst_head; | |
879 | queue->fastopenq->rskq_rst_head = NULL; | |
880 | spin_unlock_bh(&queue->fastopenq->lock); | |
881 | while ((req = acc_req) != NULL) { | |
882 | acc_req = req->dl_next; | |
13854e5a | 883 | reqsk_put(req); |
8336886f JC |
884 | } |
885 | } | |
547b792c | 886 | WARN_ON(sk->sk_ack_backlog); |
a019d6fe | 887 | } |
a019d6fe | 888 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); |
af05dc93 ACM |
889 | |
890 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) | |
891 | { | |
892 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; | |
893 | const struct inet_sock *inet = inet_sk(sk); | |
894 | ||
895 | sin->sin_family = AF_INET; | |
c720c7e8 ED |
896 | sin->sin_addr.s_addr = inet->inet_daddr; |
897 | sin->sin_port = inet->inet_dport; | |
af05dc93 | 898 | } |
af05dc93 | 899 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); |
c4d93909 | 900 | |
dec73ff0 ACM |
901 | #ifdef CONFIG_COMPAT |
902 | int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, | |
903 | char __user *optval, int __user *optlen) | |
904 | { | |
dbeff12b | 905 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 | 906 | |
00db4124 | 907 | if (icsk->icsk_af_ops->compat_getsockopt) |
dec73ff0 ACM |
908 | return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, |
909 | optval, optlen); | |
910 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
911 | optval, optlen); | |
912 | } | |
dec73ff0 ACM |
913 | EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); |
914 | ||
915 | int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 916 | char __user *optval, unsigned int optlen) |
dec73ff0 | 917 | { |
dbeff12b | 918 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 | 919 | |
00db4124 | 920 | if (icsk->icsk_af_ops->compat_setsockopt) |
dec73ff0 ACM |
921 | return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, |
922 | optval, optlen); | |
923 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
924 | optval, optlen); | |
925 | } | |
dec73ff0 ACM |
926 | EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); |
927 | #endif | |
80d0a69f DM |
928 | |
929 | static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) | |
930 | { | |
5abf7f7e ED |
931 | const struct inet_sock *inet = inet_sk(sk); |
932 | const struct ip_options_rcu *inet_opt; | |
80d0a69f DM |
933 | __be32 daddr = inet->inet_daddr; |
934 | struct flowi4 *fl4; | |
935 | struct rtable *rt; | |
936 | ||
937 | rcu_read_lock(); | |
938 | inet_opt = rcu_dereference(inet->inet_opt); | |
939 | if (inet_opt && inet_opt->opt.srr) | |
940 | daddr = inet_opt->opt.faddr; | |
941 | fl4 = &fl->u.ip4; | |
942 | rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, | |
943 | inet->inet_saddr, inet->inet_dport, | |
944 | inet->inet_sport, sk->sk_protocol, | |
945 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if); | |
946 | if (IS_ERR(rt)) | |
947 | rt = NULL; | |
948 | if (rt) | |
949 | sk_setup_caps(sk, &rt->dst); | |
950 | rcu_read_unlock(); | |
951 | ||
952 | return &rt->dst; | |
953 | } | |
954 | ||
955 | struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) | |
956 | { | |
957 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
958 | struct inet_sock *inet = inet_sk(sk); | |
959 | ||
960 | if (!dst) { | |
961 | dst = inet_csk_rebuild_route(sk, &inet->cork.fl); | |
962 | if (!dst) | |
963 | goto out; | |
964 | } | |
6700c270 | 965 | dst->ops->update_pmtu(dst, sk, NULL, mtu); |
80d0a69f DM |
966 | |
967 | dst = __sk_dst_check(sk, 0); | |
968 | if (!dst) | |
969 | dst = inet_csk_rebuild_route(sk, &inet->cork.fl); | |
970 | out: | |
971 | return dst; | |
972 | } | |
973 | EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); |