]>
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> |
c125e80b | 27 | #include <net/sock_reuseport.h> |
3f421baa ACM |
28 | |
29 | #ifdef INET_CSK_DEBUG | |
30 | const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n"; | |
31 | EXPORT_SYMBOL(inet_csk_timer_bug_msg); | |
32 | #endif | |
33 | ||
0bbf87d8 | 34 | void inet_get_local_port_range(struct net *net, int *low, int *high) |
227b60f5 | 35 | { |
95c96174 ED |
36 | unsigned int seq; |
37 | ||
227b60f5 | 38 | do { |
c9d8f1a6 | 39 | seq = read_seqbegin(&net->ipv4.ip_local_ports.lock); |
227b60f5 | 40 | |
c9d8f1a6 CW |
41 | *low = net->ipv4.ip_local_ports.range[0]; |
42 | *high = net->ipv4.ip_local_ports.range[1]; | |
43 | } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq)); | |
227b60f5 SH |
44 | } |
45 | EXPORT_SYMBOL(inet_get_local_port_range); | |
3f421baa | 46 | |
971af18b | 47 | int inet_csk_bind_conflict(const struct sock *sk, |
0643ee4f TH |
48 | const struct inet_bind_bucket *tb, bool relax, |
49 | bool reuseport_ok) | |
3f421baa | 50 | { |
3f421baa | 51 | struct sock *sk2; |
0643ee4f TH |
52 | bool reuse = sk->sk_reuse; |
53 | bool reuseport = !!sk->sk_reuseport && reuseport_ok; | |
da5e3630 | 54 | kuid_t uid = sock_i_uid((struct sock *)sk); |
3f421baa | 55 | |
7477fd2e PE |
56 | /* |
57 | * Unlike other sk lookup places we do not check | |
58 | * for sk_net here, since _all_ the socks listed | |
59 | * in tb->owners list belong to the same net - the | |
60 | * one this bucket belongs to. | |
61 | */ | |
62 | ||
b67bfe0d | 63 | sk_for_each_bound(sk2, &tb->owners) { |
3f421baa ACM |
64 | if (sk != sk2 && |
65 | !inet_v6_ipv6only(sk2) && | |
66 | (!sk->sk_bound_dev_if || | |
67 | !sk2->sk_bound_dev_if || | |
68 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { | |
da5e3630 TH |
69 | if ((!reuse || !sk2->sk_reuse || |
70 | sk2->sk_state == TCP_LISTEN) && | |
71 | (!reuseport || !sk2->sk_reuseport || | |
c125e80b CG |
72 | rcu_access_pointer(sk->sk_reuseport_cb) || |
73 | (sk2->sk_state != TCP_TIME_WAIT && | |
da5e3630 | 74 | !uid_eq(uid, sock_i_uid(sk2))))) { |
50805466 ED |
75 | |
76 | if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr || | |
77 | sk2->sk_rcv_saddr == sk->sk_rcv_saddr) | |
3f421baa | 78 | break; |
8d238b25 | 79 | } |
aacd9289 AC |
80 | if (!relax && reuse && sk2->sk_reuse && |
81 | sk2->sk_state != TCP_LISTEN) { | |
aacd9289 | 82 | |
50805466 ED |
83 | if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr || |
84 | sk2->sk_rcv_saddr == sk->sk_rcv_saddr) | |
aacd9289 AC |
85 | break; |
86 | } | |
3f421baa ACM |
87 | } |
88 | } | |
b67bfe0d | 89 | return sk2 != NULL; |
3f421baa | 90 | } |
971af18b ACM |
91 | EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); |
92 | ||
3f421baa ACM |
93 | /* Obtain a reference to a local port for the given sock, |
94 | * if snum is zero it means select any available local port. | |
ea8add2b | 95 | * We try to allocate an odd port (and leave even ports for connect()) |
3f421baa | 96 | */ |
ab1e0a13 | 97 | int inet_csk_get_port(struct sock *sk, unsigned short snum) |
3f421baa | 98 | { |
ea8add2b ED |
99 | bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; |
100 | struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo; | |
101 | int ret = 1, attempts = 5, port = snum; | |
102 | int smallest_size = -1, smallest_port; | |
3f421baa | 103 | struct inet_bind_hashbucket *head; |
3b1e0a65 | 104 | struct net *net = sock_net(sk); |
ea8add2b ED |
105 | int i, low, high, attempt_half; |
106 | struct inet_bind_bucket *tb; | |
da5e3630 | 107 | kuid_t uid = sock_i_uid(sk); |
ea8add2b | 108 | u32 remaining, offset; |
0643ee4f | 109 | bool reuseport_ok = !!snum; |
3f421baa | 110 | |
ea8add2b ED |
111 | if (port) { |
112 | have_port: | |
113 | head = &hinfo->bhash[inet_bhashfn(net, port, | |
114 | hinfo->bhash_size)]; | |
115 | spin_lock_bh(&head->lock); | |
116 | inet_bind_bucket_for_each(tb, &head->chain) | |
117 | if (net_eq(ib_net(tb), net) && tb->port == port) | |
118 | goto tb_found; | |
227b60f5 | 119 | |
ea8add2b ED |
120 | goto tb_not_found; |
121 | } | |
a9d8f911 | 122 | again: |
ea8add2b ED |
123 | attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0; |
124 | other_half_scan: | |
125 | inet_get_local_port_range(net, &low, &high); | |
126 | high++; /* [32768, 60999] -> [32768, 61000[ */ | |
127 | if (high - low < 4) | |
128 | attempt_half = 0; | |
129 | if (attempt_half) { | |
130 | int half = low + (((high - low) >> 2) << 1); | |
131 | ||
132 | if (attempt_half == 1) | |
133 | high = half; | |
134 | else | |
135 | low = half; | |
136 | } | |
137 | remaining = high - low; | |
138 | if (likely(remaining > 1)) | |
139 | remaining &= ~1U; | |
3f421baa | 140 | |
ea8add2b ED |
141 | offset = prandom_u32() % remaining; |
142 | /* __inet_hash_connect() favors ports having @low parity | |
143 | * We do the opposite to not pollute connect() users. | |
144 | */ | |
145 | offset |= 1U; | |
146 | smallest_size = -1; | |
147 | smallest_port = low; /* avoid compiler warning */ | |
148 | ||
149 | other_parity_scan: | |
150 | port = low + offset; | |
151 | for (i = 0; i < remaining; i += 2, port += 2) { | |
152 | if (unlikely(port >= high)) | |
153 | port -= remaining; | |
154 | if (inet_is_local_reserved_port(net, port)) | |
155 | continue; | |
156 | head = &hinfo->bhash[inet_bhashfn(net, port, | |
157 | hinfo->bhash_size)]; | |
158 | spin_lock_bh(&head->lock); | |
159 | inet_bind_bucket_for_each(tb, &head->chain) | |
160 | if (net_eq(ib_net(tb), net) && tb->port == port) { | |
161 | if (((tb->fastreuse > 0 && reuse) || | |
162 | (tb->fastreuseport > 0 && | |
163 | sk->sk_reuseport && | |
164 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
165 | uid_eq(tb->fastuid, uid))) && | |
166 | (tb->num_owners < smallest_size || smallest_size == -1)) { | |
167 | smallest_size = tb->num_owners; | |
168 | smallest_port = port; | |
a9d8f911 | 169 | } |
0643ee4f TH |
170 | if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false, |
171 | reuseport_ok)) | |
ea8add2b ED |
172 | goto tb_found; |
173 | goto next_port; | |
946f9eb2 | 174 | } |
ea8add2b ED |
175 | goto tb_not_found; |
176 | next_port: | |
177 | spin_unlock_bh(&head->lock); | |
178 | cond_resched(); | |
179 | } | |
180 | ||
181 | if (smallest_size != -1) { | |
182 | port = smallest_port; | |
183 | goto have_port; | |
3f421baa | 184 | } |
ea8add2b ED |
185 | offset--; |
186 | if (!(offset & 1)) | |
187 | goto other_parity_scan; | |
188 | ||
189 | if (attempt_half == 1) { | |
190 | /* OK we now try the upper half of the range */ | |
191 | attempt_half = 2; | |
192 | goto other_half_scan; | |
193 | } | |
194 | return ret; | |
195 | ||
196 | tb_not_found: | |
197 | tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, | |
198 | net, head, port); | |
199 | if (!tb) | |
200 | goto fail_unlock; | |
3f421baa ACM |
201 | tb_found: |
202 | if (!hlist_empty(&tb->owners)) { | |
4a17fd52 PE |
203 | if (sk->sk_reuse == SK_FORCE_REUSE) |
204 | goto success; | |
205 | ||
ea8add2b | 206 | if (((tb->fastreuse > 0 && reuse) || |
da5e3630 | 207 | (tb->fastreuseport > 0 && |
e5fbfc1c | 208 | !rcu_access_pointer(sk->sk_reuseport_cb) && |
ea8add2b ED |
209 | sk->sk_reuseport && uid_eq(tb->fastuid, uid))) && |
210 | smallest_size == -1) | |
3f421baa | 211 | goto success; |
0643ee4f TH |
212 | if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true, |
213 | reuseport_ok)) { | |
ea8add2b ED |
214 | if ((reuse || |
215 | (tb->fastreuseport > 0 && | |
216 | sk->sk_reuseport && | |
217 | !rcu_access_pointer(sk->sk_reuseport_cb) && | |
218 | uid_eq(tb->fastuid, uid))) && | |
9af7e923 | 219 | !snum && smallest_size != -1 && --attempts >= 0) { |
ea8add2b ED |
220 | spin_unlock_bh(&head->lock); |
221 | goto again; | |
a9d8f911 | 222 | } |
ea8add2b | 223 | goto fail_unlock; |
3f421baa | 224 | } |
ea8add2b | 225 | if (!reuse) |
3f421baa | 226 | tb->fastreuse = 0; |
ea8add2b ED |
227 | if (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)) |
228 | tb->fastreuseport = 0; | |
229 | } else { | |
230 | tb->fastreuse = reuse; | |
da5e3630 TH |
231 | if (sk->sk_reuseport) { |
232 | tb->fastreuseport = 1; | |
233 | tb->fastuid = uid; | |
ea8add2b | 234 | } else { |
da5e3630 | 235 | tb->fastreuseport = 0; |
ea8add2b | 236 | } |
da5e3630 | 237 | } |
3f421baa ACM |
238 | success: |
239 | if (!inet_csk(sk)->icsk_bind_hash) | |
ea8add2b | 240 | inet_bind_hash(sk, tb, port); |
547b792c | 241 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
e905a9ed | 242 | ret = 0; |
3f421baa ACM |
243 | |
244 | fail_unlock: | |
ea8add2b | 245 | spin_unlock_bh(&head->lock); |
3f421baa ACM |
246 | return ret; |
247 | } | |
3f421baa ACM |
248 | EXPORT_SYMBOL_GPL(inet_csk_get_port); |
249 | ||
250 | /* | |
251 | * Wait for an incoming connection, avoid race conditions. This must be called | |
252 | * with the socket locked. | |
253 | */ | |
254 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) | |
255 | { | |
256 | struct inet_connection_sock *icsk = inet_csk(sk); | |
257 | DEFINE_WAIT(wait); | |
258 | int err; | |
259 | ||
260 | /* | |
261 | * True wake-one mechanism for incoming connections: only | |
262 | * one process gets woken up, not the 'whole herd'. | |
263 | * Since we do not 'race & poll' for established sockets | |
264 | * anymore, the common case will execute the loop only once. | |
265 | * | |
266 | * Subtle issue: "add_wait_queue_exclusive()" will be added | |
267 | * after any current non-exclusive waiters, and we know that | |
268 | * it will always _stay_ after any new non-exclusive waiters | |
269 | * because all non-exclusive waiters are added at the | |
270 | * beginning of the wait-queue. As such, it's ok to "drop" | |
271 | * our exclusiveness temporarily when we get woken up without | |
272 | * having to remove and re-insert us on the wait queue. | |
273 | */ | |
274 | for (;;) { | |
aa395145 | 275 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, |
3f421baa ACM |
276 | TASK_INTERRUPTIBLE); |
277 | release_sock(sk); | |
278 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
279 | timeo = schedule_timeout(timeo); | |
cb7cf8a3 | 280 | sched_annotate_sleep(); |
3f421baa ACM |
281 | lock_sock(sk); |
282 | err = 0; | |
283 | if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
284 | break; | |
285 | err = -EINVAL; | |
286 | if (sk->sk_state != TCP_LISTEN) | |
287 | break; | |
288 | err = sock_intr_errno(timeo); | |
289 | if (signal_pending(current)) | |
290 | break; | |
291 | err = -EAGAIN; | |
292 | if (!timeo) | |
293 | break; | |
294 | } | |
aa395145 | 295 | finish_wait(sk_sleep(sk), &wait); |
3f421baa ACM |
296 | return err; |
297 | } | |
298 | ||
299 | /* | |
300 | * This will accept the next outstanding connection. | |
301 | */ | |
302 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) | |
303 | { | |
304 | struct inet_connection_sock *icsk = inet_csk(sk); | |
8336886f | 305 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
8336886f | 306 | struct request_sock *req; |
e3d95ad7 | 307 | struct sock *newsk; |
3f421baa ACM |
308 | int error; |
309 | ||
310 | lock_sock(sk); | |
311 | ||
312 | /* We need to make sure that this socket is listening, | |
313 | * and that it has something pending. | |
314 | */ | |
315 | error = -EINVAL; | |
316 | if (sk->sk_state != TCP_LISTEN) | |
317 | goto out_err; | |
318 | ||
319 | /* Find already established connection */ | |
8336886f | 320 | if (reqsk_queue_empty(queue)) { |
3f421baa ACM |
321 | long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); |
322 | ||
323 | /* If this is a non blocking socket don't sleep */ | |
324 | error = -EAGAIN; | |
325 | if (!timeo) | |
326 | goto out_err; | |
327 | ||
328 | error = inet_csk_wait_for_connect(sk, timeo); | |
329 | if (error) | |
330 | goto out_err; | |
331 | } | |
fff1f300 | 332 | req = reqsk_queue_remove(queue, sk); |
8336886f JC |
333 | newsk = req->sk; |
334 | ||
e3d95ad7 | 335 | if (sk->sk_protocol == IPPROTO_TCP && |
0536fcc0 ED |
336 | tcp_rsk(req)->tfo_listener) { |
337 | spin_lock_bh(&queue->fastopenq.lock); | |
9439ce00 | 338 | if (tcp_rsk(req)->tfo_listener) { |
8336886f JC |
339 | /* We are still waiting for the final ACK from 3WHS |
340 | * so can't free req now. Instead, we set req->sk to | |
341 | * NULL to signify that the child socket is taken | |
342 | * so reqsk_fastopen_remove() will free the req | |
343 | * when 3WHS finishes (or is aborted). | |
344 | */ | |
345 | req->sk = NULL; | |
346 | req = NULL; | |
347 | } | |
0536fcc0 | 348 | spin_unlock_bh(&queue->fastopenq.lock); |
8336886f | 349 | } |
3f421baa ACM |
350 | out: |
351 | release_sock(sk); | |
8336886f | 352 | if (req) |
13854e5a | 353 | reqsk_put(req); |
3f421baa ACM |
354 | return newsk; |
355 | out_err: | |
356 | newsk = NULL; | |
8336886f | 357 | req = NULL; |
3f421baa ACM |
358 | *err = error; |
359 | goto out; | |
360 | } | |
3f421baa ACM |
361 | EXPORT_SYMBOL(inet_csk_accept); |
362 | ||
363 | /* | |
364 | * Using different timers for retransmit, delayed acks and probes | |
e905a9ed | 365 | * We may wish use just one timer maintaining a list of expire jiffies |
3f421baa ACM |
366 | * to optimize. |
367 | */ | |
368 | void inet_csk_init_xmit_timers(struct sock *sk, | |
369 | void (*retransmit_handler)(unsigned long), | |
370 | void (*delack_handler)(unsigned long), | |
371 | void (*keepalive_handler)(unsigned long)) | |
372 | { | |
373 | struct inet_connection_sock *icsk = inet_csk(sk); | |
374 | ||
b24b8a24 PE |
375 | setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, |
376 | (unsigned long)sk); | |
377 | setup_timer(&icsk->icsk_delack_timer, delack_handler, | |
378 | (unsigned long)sk); | |
379 | setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); | |
3f421baa ACM |
380 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
381 | } | |
3f421baa ACM |
382 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); |
383 | ||
384 | void inet_csk_clear_xmit_timers(struct sock *sk) | |
385 | { | |
386 | struct inet_connection_sock *icsk = inet_csk(sk); | |
387 | ||
388 | icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; | |
389 | ||
390 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
391 | sk_stop_timer(sk, &icsk->icsk_delack_timer); | |
392 | sk_stop_timer(sk, &sk->sk_timer); | |
393 | } | |
3f421baa ACM |
394 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); |
395 | ||
396 | void inet_csk_delete_keepalive_timer(struct sock *sk) | |
397 | { | |
398 | sk_stop_timer(sk, &sk->sk_timer); | |
399 | } | |
3f421baa ACM |
400 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); |
401 | ||
402 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) | |
403 | { | |
404 | sk_reset_timer(sk, &sk->sk_timer, jiffies + len); | |
405 | } | |
3f421baa ACM |
406 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); |
407 | ||
e5895bc6 | 408 | struct dst_entry *inet_csk_route_req(const struct sock *sk, |
6bd023f3 | 409 | struct flowi4 *fl4, |
ba3f7f04 | 410 | const struct request_sock *req) |
3f421baa | 411 | { |
3f421baa | 412 | const struct inet_request_sock *ireq = inet_rsk(req); |
8b929ab1 ED |
413 | struct net *net = read_pnet(&ireq->ireq_net); |
414 | struct ip_options_rcu *opt = ireq->opt; | |
415 | struct rtable *rt; | |
3f421baa | 416 | |
8b929ab1 | 417 | flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, |
e79d9bc7 | 418 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
8b929ab1 | 419 | sk->sk_protocol, inet_sk_flowi_flags(sk), |
634fb979 | 420 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
8b929ab1 | 421 | ireq->ir_loc_addr, ireq->ir_rmt_port, |
e2d118a1 | 422 | htons(ireq->ir_num), sk->sk_uid); |
6bd023f3 DM |
423 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); |
424 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe | 425 | if (IS_ERR(rt)) |
857a6e0a | 426 | goto no_route; |
155e8336 | 427 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
857a6e0a | 428 | goto route_err; |
d8d1f30b | 429 | return &rt->dst; |
857a6e0a IJ |
430 | |
431 | route_err: | |
432 | ip_rt_put(rt); | |
433 | no_route: | |
b45386ef | 434 | __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
857a6e0a | 435 | return NULL; |
3f421baa | 436 | } |
3f421baa ACM |
437 | EXPORT_SYMBOL_GPL(inet_csk_route_req); |
438 | ||
a2432c4f | 439 | struct dst_entry *inet_csk_route_child_sock(const struct sock *sk, |
77357a95 DM |
440 | struct sock *newsk, |
441 | const struct request_sock *req) | |
442 | { | |
443 | const struct inet_request_sock *ireq = inet_rsk(req); | |
8b929ab1 | 444 | struct net *net = read_pnet(&ireq->ireq_net); |
77357a95 | 445 | struct inet_sock *newinet = inet_sk(newsk); |
1a7b27c9 | 446 | struct ip_options_rcu *opt; |
77357a95 DM |
447 | struct flowi4 *fl4; |
448 | struct rtable *rt; | |
449 | ||
450 | fl4 = &newinet->cork.fl.u.ip4; | |
1a7b27c9 CP |
451 | |
452 | rcu_read_lock(); | |
453 | opt = rcu_dereference(newinet->inet_opt); | |
8b929ab1 | 454 | flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, |
77357a95 DM |
455 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
456 | sk->sk_protocol, inet_sk_flowi_flags(sk), | |
634fb979 | 457 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
8b929ab1 | 458 | ireq->ir_loc_addr, ireq->ir_rmt_port, |
e2d118a1 | 459 | htons(ireq->ir_num), sk->sk_uid); |
77357a95 DM |
460 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); |
461 | rt = ip_route_output_flow(net, fl4, sk); | |
462 | if (IS_ERR(rt)) | |
463 | goto no_route; | |
155e8336 | 464 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
77357a95 | 465 | goto route_err; |
1a7b27c9 | 466 | rcu_read_unlock(); |
77357a95 DM |
467 | return &rt->dst; |
468 | ||
469 | route_err: | |
470 | ip_rt_put(rt); | |
471 | no_route: | |
1a7b27c9 | 472 | rcu_read_unlock(); |
b45386ef | 473 | __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
77357a95 DM |
474 | return NULL; |
475 | } | |
476 | EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); | |
477 | ||
dfd56b8b | 478 | #if IS_ENABLED(CONFIG_IPV6) |
3f421baa ACM |
479 | #define AF_INET_FAMILY(fam) ((fam) == AF_INET) |
480 | #else | |
fa76ce73 | 481 | #define AF_INET_FAMILY(fam) true |
3f421baa ACM |
482 | #endif |
483 | ||
0c3d79bc JA |
484 | /* Decide when to expire the request and when to resend SYN-ACK */ |
485 | static inline void syn_ack_recalc(struct request_sock *req, const int thresh, | |
486 | const int max_retries, | |
487 | const u8 rskq_defer_accept, | |
488 | int *expire, int *resend) | |
489 | { | |
490 | if (!rskq_defer_accept) { | |
e6c022a4 | 491 | *expire = req->num_timeout >= thresh; |
0c3d79bc JA |
492 | *resend = 1; |
493 | return; | |
494 | } | |
e6c022a4 ED |
495 | *expire = req->num_timeout >= thresh && |
496 | (!inet_rsk(req)->acked || req->num_timeout >= max_retries); | |
0c3d79bc JA |
497 | /* |
498 | * Do not resend while waiting for data after ACK, | |
499 | * start to resend on end of deferring period to give | |
500 | * last chance for data or ACK to create established socket. | |
501 | */ | |
502 | *resend = !inet_rsk(req)->acked || | |
e6c022a4 | 503 | req->num_timeout >= rskq_defer_accept - 1; |
0c3d79bc JA |
504 | } |
505 | ||
1b70e977 | 506 | int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req) |
e6c022a4 | 507 | { |
1a2c6181 | 508 | int err = req->rsk_ops->rtx_syn_ack(parent, req); |
e6c022a4 ED |
509 | |
510 | if (!err) | |
511 | req->num_retrans++; | |
512 | return err; | |
513 | } | |
514 | EXPORT_SYMBOL(inet_rtx_syn_ack); | |
515 | ||
079096f1 | 516 | /* return true if req was found in the ehash table */ |
b357a364 ED |
517 | static bool reqsk_queue_unlink(struct request_sock_queue *queue, |
518 | struct request_sock *req) | |
519 | { | |
079096f1 | 520 | struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo; |
5e0724d0 | 521 | bool found = false; |
b357a364 | 522 | |
5e0724d0 ED |
523 | if (sk_hashed(req_to_sk(req))) { |
524 | spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash); | |
b357a364 | 525 | |
5e0724d0 ED |
526 | spin_lock(lock); |
527 | found = __sk_nulls_del_node_init_rcu(req_to_sk(req)); | |
528 | spin_unlock(lock); | |
529 | } | |
83fccfc3 | 530 | if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer)) |
b357a364 ED |
531 | reqsk_put(req); |
532 | return found; | |
533 | } | |
534 | ||
535 | void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req) | |
536 | { | |
537 | if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) { | |
538 | reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); | |
539 | reqsk_put(req); | |
540 | } | |
541 | } | |
542 | EXPORT_SYMBOL(inet_csk_reqsk_queue_drop); | |
543 | ||
f03f2e15 ED |
544 | void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req) |
545 | { | |
546 | inet_csk_reqsk_queue_drop(sk, req); | |
547 | reqsk_put(req); | |
548 | } | |
549 | EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put); | |
550 | ||
fa76ce73 | 551 | static void reqsk_timer_handler(unsigned long data) |
a019d6fe | 552 | { |
fa76ce73 ED |
553 | struct request_sock *req = (struct request_sock *)data; |
554 | struct sock *sk_listener = req->rsk_listener; | |
7c083ecb | 555 | struct net *net = sock_net(sk_listener); |
fa76ce73 | 556 | struct inet_connection_sock *icsk = inet_csk(sk_listener); |
a019d6fe | 557 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
2b41fab7 | 558 | int qlen, expire = 0, resend = 0; |
fa76ce73 | 559 | int max_retries, thresh; |
2b41fab7 | 560 | u8 defer_accept; |
a019d6fe | 561 | |
00fd38d9 | 562 | if (sk_state_load(sk_listener) != TCP_LISTEN) |
079096f1 | 563 | goto drop; |
a019d6fe | 564 | |
7c083ecb | 565 | max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries; |
fa76ce73 | 566 | thresh = max_retries; |
a019d6fe ACM |
567 | /* Normally all the openreqs are young and become mature |
568 | * (i.e. converted to established socket) for first timeout. | |
fd4f2cea | 569 | * If synack was not acknowledged for 1 second, it means |
a019d6fe ACM |
570 | * one of the following things: synack was lost, ack was lost, |
571 | * rtt is high or nobody planned to ack (i.e. synflood). | |
572 | * When server is a bit loaded, queue is populated with old | |
573 | * open requests, reducing effective size of queue. | |
574 | * When server is well loaded, queue size reduces to zero | |
575 | * after several minutes of work. It is not synflood, | |
576 | * it is normal operation. The solution is pruning | |
577 | * too old entries overriding normal timeout, when | |
578 | * situation becomes dangerous. | |
579 | * | |
580 | * Essentially, we reserve half of room for young | |
581 | * embrions; and abort old ones without pity, if old | |
582 | * ones are about to clog our table. | |
583 | */ | |
aac065c5 | 584 | qlen = reqsk_queue_len(queue); |
acb4a6bf | 585 | if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) { |
aac065c5 | 586 | int young = reqsk_queue_len_young(queue) << 1; |
a019d6fe ACM |
587 | |
588 | while (thresh > 2) { | |
2b41fab7 | 589 | if (qlen < young) |
a019d6fe ACM |
590 | break; |
591 | thresh--; | |
592 | young <<= 1; | |
593 | } | |
594 | } | |
2b41fab7 ED |
595 | defer_accept = READ_ONCE(queue->rskq_defer_accept); |
596 | if (defer_accept) | |
597 | max_retries = defer_accept; | |
598 | syn_ack_recalc(req, thresh, max_retries, defer_accept, | |
fa76ce73 | 599 | &expire, &resend); |
42cb80a2 | 600 | req->rsk_ops->syn_ack_timeout(req); |
fa76ce73 ED |
601 | if (!expire && |
602 | (!resend || | |
603 | !inet_rtx_syn_ack(sk_listener, req) || | |
604 | inet_rsk(req)->acked)) { | |
605 | unsigned long timeo; | |
606 | ||
607 | if (req->num_timeout++ == 0) | |
aac065c5 | 608 | atomic_dec(&queue->young); |
fa76ce73 | 609 | timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); |
f3438bc7 | 610 | mod_timer(&req->rsk_timer, jiffies + timeo); |
fa76ce73 ED |
611 | return; |
612 | } | |
079096f1 | 613 | drop: |
f03f2e15 | 614 | inet_csk_reqsk_queue_drop_and_put(sk_listener, req); |
fa76ce73 | 615 | } |
ec0a1966 | 616 | |
079096f1 ED |
617 | static void reqsk_queue_hash_req(struct request_sock *req, |
618 | unsigned long timeout) | |
fa76ce73 | 619 | { |
fa76ce73 ED |
620 | req->num_retrans = 0; |
621 | req->num_timeout = 0; | |
622 | req->sk = NULL; | |
a019d6fe | 623 | |
f3438bc7 TG |
624 | setup_pinned_timer(&req->rsk_timer, reqsk_timer_handler, |
625 | (unsigned long)req); | |
626 | mod_timer(&req->rsk_timer, jiffies + timeout); | |
29c68526 | 627 | |
079096f1 | 628 | inet_ehash_insert(req_to_sk(req), NULL); |
fa76ce73 ED |
629 | /* before letting lookups find us, make sure all req fields |
630 | * are committed to memory and refcnt initialized. | |
631 | */ | |
632 | smp_wmb(); | |
ca6fb065 | 633 | atomic_set(&req->rsk_refcnt, 2 + 1); |
079096f1 | 634 | } |
a019d6fe | 635 | |
079096f1 ED |
636 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, |
637 | unsigned long timeout) | |
638 | { | |
639 | reqsk_queue_hash_req(req, timeout); | |
640 | inet_csk_reqsk_queue_added(sk); | |
a019d6fe | 641 | } |
079096f1 | 642 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
a019d6fe | 643 | |
e56c57d0 ED |
644 | /** |
645 | * inet_csk_clone_lock - clone an inet socket, and lock its clone | |
646 | * @sk: the socket to clone | |
647 | * @req: request_sock | |
648 | * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) | |
649 | * | |
650 | * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) | |
651 | */ | |
652 | struct sock *inet_csk_clone_lock(const struct sock *sk, | |
653 | const struct request_sock *req, | |
654 | const gfp_t priority) | |
9f1d2604 | 655 | { |
e56c57d0 | 656 | struct sock *newsk = sk_clone_lock(sk, priority); |
9f1d2604 | 657 | |
00db4124 | 658 | if (newsk) { |
9f1d2604 ACM |
659 | struct inet_connection_sock *newicsk = inet_csk(newsk); |
660 | ||
661 | newsk->sk_state = TCP_SYN_RECV; | |
662 | newicsk->icsk_bind_hash = NULL; | |
663 | ||
634fb979 | 664 | inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port; |
b44084c2 ED |
665 | inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num; |
666 | inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num); | |
9f1d2604 ACM |
667 | newsk->sk_write_space = sk_stream_write_space; |
668 | ||
85017869 ED |
669 | /* listeners have SOCK_RCU_FREE, not the children */ |
670 | sock_reset_flag(newsk, SOCK_RCU_FREE); | |
671 | ||
d082b328 ED |
672 | inet_sk(newsk)->mc_list = NULL; |
673 | ||
84f39b08 | 674 | newsk->sk_mark = inet_rsk(req)->ir_mark; |
33cf7c90 ED |
675 | atomic64_set(&newsk->sk_cookie, |
676 | atomic64_read(&inet_rsk(req)->ir_cookie)); | |
84f39b08 | 677 | |
9f1d2604 | 678 | newicsk->icsk_retransmits = 0; |
6687e988 ACM |
679 | newicsk->icsk_backoff = 0; |
680 | newicsk->icsk_probes_out = 0; | |
9f1d2604 ACM |
681 | |
682 | /* Deinitialize accept_queue to trap illegal accesses. */ | |
683 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); | |
4237c75c VY |
684 | |
685 | security_inet_csk_clone(newsk, req); | |
9f1d2604 ACM |
686 | } |
687 | return newsk; | |
688 | } | |
e56c57d0 | 689 | EXPORT_SYMBOL_GPL(inet_csk_clone_lock); |
a019d6fe ACM |
690 | |
691 | /* | |
692 | * At this point, there should be no process reference to this | |
693 | * socket, and thus no user references at all. Therefore we | |
694 | * can assume the socket waitqueue is inactive and nobody will | |
695 | * try to jump onto it. | |
696 | */ | |
697 | void inet_csk_destroy_sock(struct sock *sk) | |
698 | { | |
547b792c IJ |
699 | WARN_ON(sk->sk_state != TCP_CLOSE); |
700 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); | |
a019d6fe ACM |
701 | |
702 | /* It cannot be in hash table! */ | |
547b792c | 703 | WARN_ON(!sk_unhashed(sk)); |
a019d6fe | 704 | |
c720c7e8 ED |
705 | /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ |
706 | WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); | |
a019d6fe ACM |
707 | |
708 | sk->sk_prot->destroy(sk); | |
709 | ||
710 | sk_stream_kill_queues(sk); | |
711 | ||
712 | xfrm_sk_free_policy(sk); | |
713 | ||
714 | sk_refcnt_debug_release(sk); | |
715 | ||
777c6ae5 | 716 | local_bh_disable(); |
dd24c001 | 717 | percpu_counter_dec(sk->sk_prot->orphan_count); |
777c6ae5 | 718 | local_bh_enable(); |
a019d6fe ACM |
719 | sock_put(sk); |
720 | } | |
a019d6fe ACM |
721 | EXPORT_SYMBOL(inet_csk_destroy_sock); |
722 | ||
e337e24d CP |
723 | /* This function allows to force a closure of a socket after the call to |
724 | * tcp/dccp_create_openreq_child(). | |
725 | */ | |
726 | void inet_csk_prepare_forced_close(struct sock *sk) | |
c10cb5fc | 727 | __releases(&sk->sk_lock.slock) |
e337e24d CP |
728 | { |
729 | /* sk_clone_lock locked the socket and set refcnt to 2 */ | |
730 | bh_unlock_sock(sk); | |
731 | sock_put(sk); | |
732 | ||
733 | /* The below has to be done to allow calling inet_csk_destroy_sock */ | |
734 | sock_set_flag(sk, SOCK_DEAD); | |
735 | percpu_counter_inc(sk->sk_prot->orphan_count); | |
736 | inet_sk(sk)->inet_num = 0; | |
737 | } | |
738 | EXPORT_SYMBOL(inet_csk_prepare_forced_close); | |
739 | ||
f985c65c | 740 | int inet_csk_listen_start(struct sock *sk, int backlog) |
a019d6fe | 741 | { |
a019d6fe | 742 | struct inet_connection_sock *icsk = inet_csk(sk); |
10cbc8f1 | 743 | struct inet_sock *inet = inet_sk(sk); |
086c653f | 744 | int err = -EADDRINUSE; |
a019d6fe | 745 | |
ef547f2a | 746 | reqsk_queue_alloc(&icsk->icsk_accept_queue); |
a019d6fe | 747 | |
f985c65c | 748 | sk->sk_max_ack_backlog = backlog; |
a019d6fe ACM |
749 | sk->sk_ack_backlog = 0; |
750 | inet_csk_delack_init(sk); | |
751 | ||
752 | /* There is race window here: we announce ourselves listening, | |
753 | * but this transition is still not validated by get_port(). | |
754 | * It is OK, because this socket enters to hash table only | |
755 | * after validation is complete. | |
756 | */ | |
00fd38d9 | 757 | sk_state_store(sk, TCP_LISTEN); |
c720c7e8 ED |
758 | if (!sk->sk_prot->get_port(sk, inet->inet_num)) { |
759 | inet->inet_sport = htons(inet->inet_num); | |
a019d6fe ACM |
760 | |
761 | sk_dst_reset(sk); | |
086c653f | 762 | err = sk->sk_prot->hash(sk); |
a019d6fe | 763 | |
086c653f CG |
764 | if (likely(!err)) |
765 | return 0; | |
a019d6fe ACM |
766 | } |
767 | ||
768 | sk->sk_state = TCP_CLOSE; | |
086c653f | 769 | return err; |
a019d6fe | 770 | } |
a019d6fe ACM |
771 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); |
772 | ||
ebb516af ED |
773 | static void inet_child_forget(struct sock *sk, struct request_sock *req, |
774 | struct sock *child) | |
775 | { | |
776 | sk->sk_prot->disconnect(child, O_NONBLOCK); | |
777 | ||
778 | sock_orphan(child); | |
779 | ||
780 | percpu_counter_inc(sk->sk_prot->orphan_count); | |
781 | ||
782 | if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) { | |
783 | BUG_ON(tcp_sk(child)->fastopen_rsk != req); | |
784 | BUG_ON(sk != req->rsk_listener); | |
785 | ||
786 | /* Paranoid, to prevent race condition if | |
787 | * an inbound pkt destined for child is | |
788 | * blocked by sock lock in tcp_v4_rcv(). | |
789 | * Also to satisfy an assertion in | |
790 | * tcp_v4_destroy_sock(). | |
791 | */ | |
792 | tcp_sk(child)->fastopen_rsk = NULL; | |
793 | } | |
794 | inet_csk_destroy_sock(child); | |
795 | reqsk_put(req); | |
796 | } | |
797 | ||
7716682c ED |
798 | struct sock *inet_csk_reqsk_queue_add(struct sock *sk, |
799 | struct request_sock *req, | |
800 | struct sock *child) | |
ebb516af ED |
801 | { |
802 | struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; | |
803 | ||
804 | spin_lock(&queue->rskq_lock); | |
805 | if (unlikely(sk->sk_state != TCP_LISTEN)) { | |
806 | inet_child_forget(sk, req, child); | |
7716682c | 807 | child = NULL; |
ebb516af ED |
808 | } else { |
809 | req->sk = child; | |
810 | req->dl_next = NULL; | |
811 | if (queue->rskq_accept_head == NULL) | |
812 | queue->rskq_accept_head = req; | |
813 | else | |
814 | queue->rskq_accept_tail->dl_next = req; | |
815 | queue->rskq_accept_tail = req; | |
816 | sk_acceptq_added(sk); | |
817 | } | |
818 | spin_unlock(&queue->rskq_lock); | |
7716682c | 819 | return child; |
ebb516af ED |
820 | } |
821 | EXPORT_SYMBOL(inet_csk_reqsk_queue_add); | |
822 | ||
5e0724d0 ED |
823 | struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child, |
824 | struct request_sock *req, bool own_req) | |
825 | { | |
826 | if (own_req) { | |
827 | inet_csk_reqsk_queue_drop(sk, req); | |
828 | reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); | |
7716682c ED |
829 | if (inet_csk_reqsk_queue_add(sk, req, child)) |
830 | return child; | |
5e0724d0 ED |
831 | } |
832 | /* Too bad, another child took ownership of the request, undo. */ | |
833 | bh_unlock_sock(child); | |
834 | sock_put(child); | |
835 | return NULL; | |
836 | } | |
837 | EXPORT_SYMBOL(inet_csk_complete_hashdance); | |
838 | ||
a019d6fe ACM |
839 | /* |
840 | * This routine closes sockets which have been at least partially | |
841 | * opened, but not yet accepted. | |
842 | */ | |
843 | void inet_csk_listen_stop(struct sock *sk) | |
844 | { | |
845 | struct inet_connection_sock *icsk = inet_csk(sk); | |
8336886f | 846 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
fff1f300 | 847 | struct request_sock *next, *req; |
a019d6fe ACM |
848 | |
849 | /* Following specs, it would be better either to send FIN | |
850 | * (and enter FIN-WAIT-1, it is normal close) | |
851 | * or to send active reset (abort). | |
852 | * Certainly, it is pretty dangerous while synflood, but it is | |
853 | * bad justification for our negligence 8) | |
854 | * To be honest, we are not able to make either | |
855 | * of the variants now. --ANK | |
856 | */ | |
fff1f300 | 857 | while ((req = reqsk_queue_remove(queue, sk)) != NULL) { |
a019d6fe ACM |
858 | struct sock *child = req->sk; |
859 | ||
a019d6fe ACM |
860 | local_bh_disable(); |
861 | bh_lock_sock(child); | |
547b792c | 862 | WARN_ON(sock_owned_by_user(child)); |
a019d6fe ACM |
863 | sock_hold(child); |
864 | ||
ebb516af | 865 | inet_child_forget(sk, req, child); |
a019d6fe ACM |
866 | bh_unlock_sock(child); |
867 | local_bh_enable(); | |
868 | sock_put(child); | |
869 | ||
92d6f176 | 870 | cond_resched(); |
a019d6fe | 871 | } |
0536fcc0 | 872 | if (queue->fastopenq.rskq_rst_head) { |
8336886f | 873 | /* Free all the reqs queued in rskq_rst_head. */ |
0536fcc0 | 874 | spin_lock_bh(&queue->fastopenq.lock); |
fff1f300 | 875 | req = queue->fastopenq.rskq_rst_head; |
0536fcc0 ED |
876 | queue->fastopenq.rskq_rst_head = NULL; |
877 | spin_unlock_bh(&queue->fastopenq.lock); | |
fff1f300 ED |
878 | while (req != NULL) { |
879 | next = req->dl_next; | |
13854e5a | 880 | reqsk_put(req); |
fff1f300 | 881 | req = next; |
8336886f JC |
882 | } |
883 | } | |
ebb516af | 884 | WARN_ON_ONCE(sk->sk_ack_backlog); |
a019d6fe | 885 | } |
a019d6fe | 886 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); |
af05dc93 ACM |
887 | |
888 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) | |
889 | { | |
890 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; | |
891 | const struct inet_sock *inet = inet_sk(sk); | |
892 | ||
893 | sin->sin_family = AF_INET; | |
c720c7e8 ED |
894 | sin->sin_addr.s_addr = inet->inet_daddr; |
895 | sin->sin_port = inet->inet_dport; | |
af05dc93 | 896 | } |
af05dc93 | 897 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); |
c4d93909 | 898 | |
dec73ff0 ACM |
899 | #ifdef CONFIG_COMPAT |
900 | int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, | |
901 | char __user *optval, int __user *optlen) | |
902 | { | |
dbeff12b | 903 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 | 904 | |
00db4124 | 905 | if (icsk->icsk_af_ops->compat_getsockopt) |
dec73ff0 ACM |
906 | return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, |
907 | optval, optlen); | |
908 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
909 | optval, optlen); | |
910 | } | |
dec73ff0 ACM |
911 | EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); |
912 | ||
913 | int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 914 | char __user *optval, unsigned int optlen) |
dec73ff0 | 915 | { |
dbeff12b | 916 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 | 917 | |
00db4124 | 918 | if (icsk->icsk_af_ops->compat_setsockopt) |
dec73ff0 ACM |
919 | return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, |
920 | optval, optlen); | |
921 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
922 | optval, optlen); | |
923 | } | |
dec73ff0 ACM |
924 | EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); |
925 | #endif | |
80d0a69f DM |
926 | |
927 | static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) | |
928 | { | |
5abf7f7e ED |
929 | const struct inet_sock *inet = inet_sk(sk); |
930 | const struct ip_options_rcu *inet_opt; | |
80d0a69f DM |
931 | __be32 daddr = inet->inet_daddr; |
932 | struct flowi4 *fl4; | |
933 | struct rtable *rt; | |
934 | ||
935 | rcu_read_lock(); | |
936 | inet_opt = rcu_dereference(inet->inet_opt); | |
937 | if (inet_opt && inet_opt->opt.srr) | |
938 | daddr = inet_opt->opt.faddr; | |
939 | fl4 = &fl->u.ip4; | |
940 | rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, | |
941 | inet->inet_saddr, inet->inet_dport, | |
942 | inet->inet_sport, sk->sk_protocol, | |
943 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if); | |
944 | if (IS_ERR(rt)) | |
945 | rt = NULL; | |
946 | if (rt) | |
947 | sk_setup_caps(sk, &rt->dst); | |
948 | rcu_read_unlock(); | |
949 | ||
950 | return &rt->dst; | |
951 | } | |
952 | ||
953 | struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) | |
954 | { | |
955 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
956 | struct inet_sock *inet = inet_sk(sk); | |
957 | ||
958 | if (!dst) { | |
959 | dst = inet_csk_rebuild_route(sk, &inet->cork.fl); | |
960 | if (!dst) | |
961 | goto out; | |
962 | } | |
6700c270 | 963 | dst->ops->update_pmtu(dst, sk, NULL, mtu); |
80d0a69f DM |
964 | |
965 | dst = __sk_dst_check(sk, 0); | |
966 | if (!dst) | |
967 | dst = inet_csk_rebuild_route(sk, &inet->cork.fl); | |
968 | out: | |
969 | return dst; | |
970 | } | |
971 | EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); |