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> |
3f421baa ACM |
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 | /* | |
3c689b73 | 33 | * This struct holds the first and last local port number. |
3f421baa | 34 | */ |
3c689b73 ED |
35 | struct local_ports sysctl_local_ports __read_mostly = { |
36 | .lock = SEQLOCK_UNLOCKED, | |
37 | .range = { 32768, 61000 }, | |
38 | }; | |
227b60f5 | 39 | |
e3826f1e AW |
40 | unsigned long *sysctl_local_reserved_ports; |
41 | EXPORT_SYMBOL(sysctl_local_reserved_ports); | |
42 | ||
227b60f5 SH |
43 | void inet_get_local_port_range(int *low, int *high) |
44 | { | |
45 | unsigned seq; | |
46 | do { | |
3c689b73 | 47 | seq = read_seqbegin(&sysctl_local_ports.lock); |
227b60f5 | 48 | |
3c689b73 ED |
49 | *low = sysctl_local_ports.range[0]; |
50 | *high = sysctl_local_ports.range[1]; | |
51 | } while (read_seqretry(&sysctl_local_ports.lock, seq)); | |
227b60f5 SH |
52 | } |
53 | EXPORT_SYMBOL(inet_get_local_port_range); | |
3f421baa | 54 | |
971af18b ACM |
55 | int inet_csk_bind_conflict(const struct sock *sk, |
56 | const struct inet_bind_bucket *tb) | |
3f421baa | 57 | { |
3f421baa ACM |
58 | struct sock *sk2; |
59 | struct hlist_node *node; | |
60 | int reuse = sk->sk_reuse; | |
61 | ||
7477fd2e PE |
62 | /* |
63 | * Unlike other sk lookup places we do not check | |
64 | * for sk_net here, since _all_ the socks listed | |
65 | * in tb->owners list belong to the same net - the | |
66 | * one this bucket belongs to. | |
67 | */ | |
68 | ||
3f421baa ACM |
69 | sk_for_each_bound(sk2, node, &tb->owners) { |
70 | if (sk != sk2 && | |
71 | !inet_v6_ipv6only(sk2) && | |
72 | (!sk->sk_bound_dev_if || | |
73 | !sk2->sk_bound_dev_if || | |
74 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { | |
75 | if (!reuse || !sk2->sk_reuse || | |
3e8c806a | 76 | sk2->sk_state == TCP_LISTEN) { |
68835aba ED |
77 | const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2); |
78 | if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) || | |
79 | sk2_rcv_saddr == sk_rcv_saddr(sk)) | |
3f421baa | 80 | break; |
8d238b25 | 81 | } |
3f421baa ACM |
82 | } |
83 | } | |
84 | return node != NULL; | |
85 | } | |
971af18b ACM |
86 | EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); |
87 | ||
3f421baa ACM |
88 | /* Obtain a reference to a local port for the given sock, |
89 | * if snum is zero it means select any available local port. | |
90 | */ | |
ab1e0a13 | 91 | int inet_csk_get_port(struct sock *sk, unsigned short snum) |
3f421baa | 92 | { |
39d8cda7 | 93 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3f421baa ACM |
94 | struct inet_bind_hashbucket *head; |
95 | struct hlist_node *node; | |
96 | struct inet_bind_bucket *tb; | |
a9d8f911 | 97 | int ret, attempts = 5; |
3b1e0a65 | 98 | struct net *net = sock_net(sk); |
a9d8f911 | 99 | int smallest_size = -1, smallest_rover; |
3f421baa ACM |
100 | |
101 | local_bh_disable(); | |
102 | if (!snum) { | |
227b60f5 SH |
103 | int remaining, rover, low, high; |
104 | ||
a9d8f911 | 105 | again: |
227b60f5 | 106 | inet_get_local_port_range(&low, &high); |
a25de534 | 107 | remaining = (high - low) + 1; |
a9d8f911 | 108 | smallest_rover = rover = net_random() % remaining + low; |
3f421baa | 109 | |
a9d8f911 | 110 | smallest_size = -1; |
3f421baa | 111 | do { |
e3826f1e AW |
112 | if (inet_is_reserved_local_port(rover)) |
113 | goto next_nolock; | |
7f635ab7 PE |
114 | head = &hashinfo->bhash[inet_bhashfn(net, rover, |
115 | hashinfo->bhash_size)]; | |
3f421baa ACM |
116 | spin_lock(&head->lock); |
117 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
09ad9bc7 | 118 | if (net_eq(ib_net(tb), net) && tb->port == rover) { |
a9d8f911 EP |
119 | if (tb->fastreuse > 0 && |
120 | sk->sk_reuse && | |
121 | sk->sk_state != TCP_LISTEN && | |
122 | (tb->num_owners < smallest_size || smallest_size == -1)) { | |
123 | smallest_size = tb->num_owners; | |
124 | smallest_rover = rover; | |
3e8c806a | 125 | if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) { |
8d238b25 DM |
126 | spin_unlock(&head->lock); |
127 | snum = smallest_rover; | |
128 | goto have_snum; | |
a9d8f911 EP |
129 | } |
130 | } | |
3f421baa | 131 | goto next; |
a9d8f911 | 132 | } |
3f421baa ACM |
133 | break; |
134 | next: | |
135 | spin_unlock(&head->lock); | |
e3826f1e | 136 | next_nolock: |
6df71634 SH |
137 | if (++rover > high) |
138 | rover = low; | |
3f421baa | 139 | } while (--remaining > 0); |
3f421baa ACM |
140 | |
141 | /* Exhausted local port range during search? It is not | |
142 | * possible for us to be holding one of the bind hash | |
143 | * locks if this test triggers, because if 'remaining' | |
144 | * drops to zero, we broke out of the do/while loop at | |
145 | * the top level, not from the 'break;' statement. | |
146 | */ | |
147 | ret = 1; | |
a9d8f911 EP |
148 | if (remaining <= 0) { |
149 | if (smallest_size != -1) { | |
150 | snum = smallest_rover; | |
151 | goto have_snum; | |
152 | } | |
3f421baa | 153 | goto fail; |
a9d8f911 | 154 | } |
3f421baa ACM |
155 | /* OK, here is the one we will use. HEAD is |
156 | * non-NULL and we hold it's mutex. | |
157 | */ | |
158 | snum = rover; | |
159 | } else { | |
a9d8f911 | 160 | have_snum: |
7f635ab7 PE |
161 | head = &hashinfo->bhash[inet_bhashfn(net, snum, |
162 | hashinfo->bhash_size)]; | |
3f421baa ACM |
163 | spin_lock(&head->lock); |
164 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
09ad9bc7 | 165 | if (net_eq(ib_net(tb), net) && tb->port == snum) |
3f421baa ACM |
166 | goto tb_found; |
167 | } | |
168 | tb = NULL; | |
169 | goto tb_not_found; | |
170 | tb_found: | |
171 | if (!hlist_empty(&tb->owners)) { | |
3f421baa | 172 | if (tb->fastreuse > 0 && |
a9d8f911 EP |
173 | sk->sk_reuse && sk->sk_state != TCP_LISTEN && |
174 | smallest_size == -1) { | |
3f421baa ACM |
175 | goto success; |
176 | } else { | |
177 | ret = 1; | |
a9d8f911 | 178 | if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) { |
5add3009 SH |
179 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN && |
180 | smallest_size != -1 && --attempts >= 0) { | |
a9d8f911 EP |
181 | spin_unlock(&head->lock); |
182 | goto again; | |
183 | } | |
3f421baa | 184 | goto fail_unlock; |
a9d8f911 | 185 | } |
3f421baa ACM |
186 | } |
187 | } | |
188 | tb_not_found: | |
189 | ret = 1; | |
941b1d22 PE |
190 | if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, |
191 | net, head, snum)) == NULL) | |
3f421baa ACM |
192 | goto fail_unlock; |
193 | if (hlist_empty(&tb->owners)) { | |
194 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) | |
195 | tb->fastreuse = 1; | |
196 | else | |
197 | tb->fastreuse = 0; | |
198 | } else if (tb->fastreuse && | |
199 | (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) | |
200 | tb->fastreuse = 0; | |
201 | success: | |
202 | if (!inet_csk(sk)->icsk_bind_hash) | |
203 | inet_bind_hash(sk, tb, snum); | |
547b792c | 204 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
e905a9ed | 205 | ret = 0; |
3f421baa ACM |
206 | |
207 | fail_unlock: | |
208 | spin_unlock(&head->lock); | |
209 | fail: | |
210 | local_bh_enable(); | |
211 | return ret; | |
212 | } | |
3f421baa ACM |
213 | EXPORT_SYMBOL_GPL(inet_csk_get_port); |
214 | ||
215 | /* | |
216 | * Wait for an incoming connection, avoid race conditions. This must be called | |
217 | * with the socket locked. | |
218 | */ | |
219 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) | |
220 | { | |
221 | struct inet_connection_sock *icsk = inet_csk(sk); | |
222 | DEFINE_WAIT(wait); | |
223 | int err; | |
224 | ||
225 | /* | |
226 | * True wake-one mechanism for incoming connections: only | |
227 | * one process gets woken up, not the 'whole herd'. | |
228 | * Since we do not 'race & poll' for established sockets | |
229 | * anymore, the common case will execute the loop only once. | |
230 | * | |
231 | * Subtle issue: "add_wait_queue_exclusive()" will be added | |
232 | * after any current non-exclusive waiters, and we know that | |
233 | * it will always _stay_ after any new non-exclusive waiters | |
234 | * because all non-exclusive waiters are added at the | |
235 | * beginning of the wait-queue. As such, it's ok to "drop" | |
236 | * our exclusiveness temporarily when we get woken up without | |
237 | * having to remove and re-insert us on the wait queue. | |
238 | */ | |
239 | for (;;) { | |
aa395145 | 240 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, |
3f421baa ACM |
241 | TASK_INTERRUPTIBLE); |
242 | release_sock(sk); | |
243 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
244 | timeo = schedule_timeout(timeo); | |
245 | lock_sock(sk); | |
246 | err = 0; | |
247 | if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
248 | break; | |
249 | err = -EINVAL; | |
250 | if (sk->sk_state != TCP_LISTEN) | |
251 | break; | |
252 | err = sock_intr_errno(timeo); | |
253 | if (signal_pending(current)) | |
254 | break; | |
255 | err = -EAGAIN; | |
256 | if (!timeo) | |
257 | break; | |
258 | } | |
aa395145 | 259 | finish_wait(sk_sleep(sk), &wait); |
3f421baa ACM |
260 | return err; |
261 | } | |
262 | ||
263 | /* | |
264 | * This will accept the next outstanding connection. | |
265 | */ | |
266 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) | |
267 | { | |
268 | struct inet_connection_sock *icsk = inet_csk(sk); | |
269 | struct sock *newsk; | |
270 | int error; | |
271 | ||
272 | lock_sock(sk); | |
273 | ||
274 | /* We need to make sure that this socket is listening, | |
275 | * and that it has something pending. | |
276 | */ | |
277 | error = -EINVAL; | |
278 | if (sk->sk_state != TCP_LISTEN) | |
279 | goto out_err; | |
280 | ||
281 | /* Find already established connection */ | |
282 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) { | |
283 | long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); | |
284 | ||
285 | /* If this is a non blocking socket don't sleep */ | |
286 | error = -EAGAIN; | |
287 | if (!timeo) | |
288 | goto out_err; | |
289 | ||
290 | error = inet_csk_wait_for_connect(sk, timeo); | |
291 | if (error) | |
292 | goto out_err; | |
293 | } | |
294 | ||
295 | newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk); | |
547b792c | 296 | WARN_ON(newsk->sk_state == TCP_SYN_RECV); |
3f421baa ACM |
297 | out: |
298 | release_sock(sk); | |
299 | return newsk; | |
300 | out_err: | |
301 | newsk = NULL; | |
302 | *err = error; | |
303 | goto out; | |
304 | } | |
3f421baa ACM |
305 | EXPORT_SYMBOL(inet_csk_accept); |
306 | ||
307 | /* | |
308 | * Using different timers for retransmit, delayed acks and probes | |
e905a9ed | 309 | * We may wish use just one timer maintaining a list of expire jiffies |
3f421baa ACM |
310 | * to optimize. |
311 | */ | |
312 | void inet_csk_init_xmit_timers(struct sock *sk, | |
313 | void (*retransmit_handler)(unsigned long), | |
314 | void (*delack_handler)(unsigned long), | |
315 | void (*keepalive_handler)(unsigned long)) | |
316 | { | |
317 | struct inet_connection_sock *icsk = inet_csk(sk); | |
318 | ||
b24b8a24 PE |
319 | setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, |
320 | (unsigned long)sk); | |
321 | setup_timer(&icsk->icsk_delack_timer, delack_handler, | |
322 | (unsigned long)sk); | |
323 | setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); | |
3f421baa ACM |
324 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
325 | } | |
3f421baa ACM |
326 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); |
327 | ||
328 | void inet_csk_clear_xmit_timers(struct sock *sk) | |
329 | { | |
330 | struct inet_connection_sock *icsk = inet_csk(sk); | |
331 | ||
332 | icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; | |
333 | ||
334 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
335 | sk_stop_timer(sk, &icsk->icsk_delack_timer); | |
336 | sk_stop_timer(sk, &sk->sk_timer); | |
337 | } | |
3f421baa ACM |
338 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); |
339 | ||
340 | void inet_csk_delete_keepalive_timer(struct sock *sk) | |
341 | { | |
342 | sk_stop_timer(sk, &sk->sk_timer); | |
343 | } | |
3f421baa ACM |
344 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); |
345 | ||
346 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) | |
347 | { | |
348 | sk_reset_timer(sk, &sk->sk_timer, jiffies + len); | |
349 | } | |
3f421baa ACM |
350 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); |
351 | ||
d9319100 | 352 | struct dst_entry *inet_csk_route_req(struct sock *sk, |
6bd023f3 | 353 | struct flowi4 *fl4, |
3f421baa ACM |
354 | const struct request_sock *req) |
355 | { | |
356 | struct rtable *rt; | |
357 | const struct inet_request_sock *ireq = inet_rsk(req); | |
f6d8bd05 | 358 | struct ip_options_rcu *opt = inet_rsk(req)->opt; |
84a3aa00 | 359 | struct net *net = sock_net(sk); |
3f421baa | 360 | |
6bd023f3 | 361 | flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, |
e79d9bc7 DM |
362 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
363 | sk->sk_protocol, inet_sk_flowi_flags(sk), | |
f6d8bd05 | 364 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr, |
e79d9bc7 | 365 | ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport); |
6bd023f3 DM |
366 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); |
367 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe | 368 | if (IS_ERR(rt)) |
857a6e0a | 369 | goto no_route; |
6bd023f3 | 370 | if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway) |
857a6e0a | 371 | goto route_err; |
d8d1f30b | 372 | return &rt->dst; |
857a6e0a IJ |
373 | |
374 | route_err: | |
375 | ip_rt_put(rt); | |
376 | no_route: | |
377 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); | |
378 | return NULL; | |
3f421baa | 379 | } |
3f421baa ACM |
380 | EXPORT_SYMBOL_GPL(inet_csk_route_req); |
381 | ||
77357a95 DM |
382 | struct dst_entry *inet_csk_route_child_sock(struct sock *sk, |
383 | struct sock *newsk, | |
384 | const struct request_sock *req) | |
385 | { | |
386 | const struct inet_request_sock *ireq = inet_rsk(req); | |
387 | struct inet_sock *newinet = inet_sk(newsk); | |
388 | struct ip_options_rcu *opt = ireq->opt; | |
389 | struct net *net = sock_net(sk); | |
390 | struct flowi4 *fl4; | |
391 | struct rtable *rt; | |
392 | ||
393 | fl4 = &newinet->cork.fl.u.ip4; | |
394 | flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, | |
395 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, | |
396 | sk->sk_protocol, inet_sk_flowi_flags(sk), | |
397 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr, | |
398 | ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport); | |
399 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); | |
400 | rt = ip_route_output_flow(net, fl4, sk); | |
401 | if (IS_ERR(rt)) | |
402 | goto no_route; | |
403 | if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway) | |
404 | goto route_err; | |
405 | return &rt->dst; | |
406 | ||
407 | route_err: | |
408 | ip_rt_put(rt); | |
409 | no_route: | |
410 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); | |
411 | return NULL; | |
412 | } | |
413 | EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); | |
414 | ||
6b72977b | 415 | static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport, |
72a3effa | 416 | const u32 rnd, const u32 synq_hsize) |
3f421baa | 417 | { |
6b72977b | 418 | return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1); |
3f421baa ACM |
419 | } |
420 | ||
421 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
422 | #define AF_INET_FAMILY(fam) ((fam) == AF_INET) | |
423 | #else | |
424 | #define AF_INET_FAMILY(fam) 1 | |
425 | #endif | |
426 | ||
427 | struct request_sock *inet_csk_search_req(const struct sock *sk, | |
428 | struct request_sock ***prevp, | |
6b72977b | 429 | const __be16 rport, const __be32 raddr, |
7f25afbb | 430 | const __be32 laddr) |
3f421baa ACM |
431 | { |
432 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
433 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
434 | struct request_sock *req, **prev; | |
435 | ||
436 | for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd, | |
437 | lopt->nr_table_entries)]; | |
438 | (req = *prev) != NULL; | |
439 | prev = &req->dl_next) { | |
440 | const struct inet_request_sock *ireq = inet_rsk(req); | |
441 | ||
442 | if (ireq->rmt_port == rport && | |
443 | ireq->rmt_addr == raddr && | |
444 | ireq->loc_addr == laddr && | |
445 | AF_INET_FAMILY(req->rsk_ops->family)) { | |
547b792c | 446 | WARN_ON(req->sk); |
3f421baa ACM |
447 | *prevp = prev; |
448 | break; | |
449 | } | |
450 | } | |
451 | ||
452 | return req; | |
453 | } | |
3f421baa ACM |
454 | EXPORT_SYMBOL_GPL(inet_csk_search_req); |
455 | ||
456 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, | |
c2977c22 | 457 | unsigned long timeout) |
3f421baa ACM |
458 | { |
459 | struct inet_connection_sock *icsk = inet_csk(sk); | |
460 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
461 | const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port, | |
462 | lopt->hash_rnd, lopt->nr_table_entries); | |
463 | ||
464 | reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout); | |
465 | inet_csk_reqsk_queue_added(sk, timeout); | |
466 | } | |
4bc2f18b | 467 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
3f421baa | 468 | |
a019d6fe ACM |
469 | /* Only thing we need from tcp.h */ |
470 | extern int sysctl_tcp_synack_retries; | |
471 | ||
9f1d2604 | 472 | |
0c3d79bc JA |
473 | /* Decide when to expire the request and when to resend SYN-ACK */ |
474 | static inline void syn_ack_recalc(struct request_sock *req, const int thresh, | |
475 | const int max_retries, | |
476 | const u8 rskq_defer_accept, | |
477 | int *expire, int *resend) | |
478 | { | |
479 | if (!rskq_defer_accept) { | |
480 | *expire = req->retrans >= thresh; | |
481 | *resend = 1; | |
482 | return; | |
483 | } | |
484 | *expire = req->retrans >= thresh && | |
485 | (!inet_rsk(req)->acked || req->retrans >= max_retries); | |
486 | /* | |
487 | * Do not resend while waiting for data after ACK, | |
488 | * start to resend on end of deferring period to give | |
489 | * last chance for data or ACK to create established socket. | |
490 | */ | |
491 | *resend = !inet_rsk(req)->acked || | |
492 | req->retrans >= rskq_defer_accept - 1; | |
493 | } | |
494 | ||
a019d6fe ACM |
495 | void inet_csk_reqsk_queue_prune(struct sock *parent, |
496 | const unsigned long interval, | |
497 | const unsigned long timeout, | |
498 | const unsigned long max_rto) | |
499 | { | |
500 | struct inet_connection_sock *icsk = inet_csk(parent); | |
501 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; | |
502 | struct listen_sock *lopt = queue->listen_opt; | |
ec0a1966 DM |
503 | int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries; |
504 | int thresh = max_retries; | |
a019d6fe ACM |
505 | unsigned long now = jiffies; |
506 | struct request_sock **reqp, *req; | |
507 | int i, budget; | |
508 | ||
509 | if (lopt == NULL || lopt->qlen == 0) | |
510 | return; | |
511 | ||
512 | /* Normally all the openreqs are young and become mature | |
513 | * (i.e. converted to established socket) for first timeout. | |
514 | * If synack was not acknowledged for 3 seconds, it means | |
515 | * one of the following things: synack was lost, ack was lost, | |
516 | * rtt is high or nobody planned to ack (i.e. synflood). | |
517 | * When server is a bit loaded, queue is populated with old | |
518 | * open requests, reducing effective size of queue. | |
519 | * When server is well loaded, queue size reduces to zero | |
520 | * after several minutes of work. It is not synflood, | |
521 | * it is normal operation. The solution is pruning | |
522 | * too old entries overriding normal timeout, when | |
523 | * situation becomes dangerous. | |
524 | * | |
525 | * Essentially, we reserve half of room for young | |
526 | * embrions; and abort old ones without pity, if old | |
527 | * ones are about to clog our table. | |
528 | */ | |
529 | if (lopt->qlen>>(lopt->max_qlen_log-1)) { | |
530 | int young = (lopt->qlen_young<<1); | |
531 | ||
532 | while (thresh > 2) { | |
533 | if (lopt->qlen < young) | |
534 | break; | |
535 | thresh--; | |
536 | young <<= 1; | |
537 | } | |
538 | } | |
539 | ||
ec0a1966 DM |
540 | if (queue->rskq_defer_accept) |
541 | max_retries = queue->rskq_defer_accept; | |
542 | ||
a019d6fe ACM |
543 | budget = 2 * (lopt->nr_table_entries / (timeout / interval)); |
544 | i = lopt->clock_hand; | |
545 | ||
546 | do { | |
547 | reqp=&lopt->syn_table[i]; | |
548 | while ((req = *reqp) != NULL) { | |
549 | if (time_after_eq(now, req->expires)) { | |
0c3d79bc JA |
550 | int expire = 0, resend = 0; |
551 | ||
552 | syn_ack_recalc(req, thresh, max_retries, | |
553 | queue->rskq_defer_accept, | |
554 | &expire, &resend); | |
72659ecc OP |
555 | if (req->rsk_ops->syn_ack_timeout) |
556 | req->rsk_ops->syn_ack_timeout(parent, req); | |
0c3d79bc JA |
557 | if (!expire && |
558 | (!resend || | |
e6b4d113 | 559 | !req->rsk_ops->rtx_syn_ack(parent, req, NULL) || |
0c3d79bc | 560 | inet_rsk(req)->acked)) { |
a019d6fe ACM |
561 | unsigned long timeo; |
562 | ||
563 | if (req->retrans++ == 0) | |
564 | lopt->qlen_young--; | |
565 | timeo = min((timeout << req->retrans), max_rto); | |
566 | req->expires = now + timeo; | |
567 | reqp = &req->dl_next; | |
568 | continue; | |
569 | } | |
570 | ||
571 | /* Drop this request */ | |
572 | inet_csk_reqsk_queue_unlink(parent, req, reqp); | |
573 | reqsk_queue_removed(queue, req); | |
574 | reqsk_free(req); | |
575 | continue; | |
576 | } | |
577 | reqp = &req->dl_next; | |
578 | } | |
579 | ||
580 | i = (i + 1) & (lopt->nr_table_entries - 1); | |
581 | ||
582 | } while (--budget > 0); | |
583 | ||
584 | lopt->clock_hand = i; | |
585 | ||
586 | if (lopt->qlen) | |
587 | inet_csk_reset_keepalive_timer(parent, interval); | |
588 | } | |
a019d6fe ACM |
589 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune); |
590 | ||
9f1d2604 | 591 | struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req, |
dd0fc66f | 592 | const gfp_t priority) |
9f1d2604 ACM |
593 | { |
594 | struct sock *newsk = sk_clone(sk, priority); | |
595 | ||
596 | if (newsk != NULL) { | |
597 | struct inet_connection_sock *newicsk = inet_csk(newsk); | |
598 | ||
599 | newsk->sk_state = TCP_SYN_RECV; | |
600 | newicsk->icsk_bind_hash = NULL; | |
601 | ||
c720c7e8 ED |
602 | inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port; |
603 | inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port); | |
604 | inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port; | |
9f1d2604 ACM |
605 | newsk->sk_write_space = sk_stream_write_space; |
606 | ||
607 | newicsk->icsk_retransmits = 0; | |
6687e988 ACM |
608 | newicsk->icsk_backoff = 0; |
609 | newicsk->icsk_probes_out = 0; | |
9f1d2604 ACM |
610 | |
611 | /* Deinitialize accept_queue to trap illegal accesses. */ | |
612 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); | |
4237c75c VY |
613 | |
614 | security_inet_csk_clone(newsk, req); | |
9f1d2604 ACM |
615 | } |
616 | return newsk; | |
617 | } | |
9f1d2604 | 618 | EXPORT_SYMBOL_GPL(inet_csk_clone); |
a019d6fe ACM |
619 | |
620 | /* | |
621 | * At this point, there should be no process reference to this | |
622 | * socket, and thus no user references at all. Therefore we | |
623 | * can assume the socket waitqueue is inactive and nobody will | |
624 | * try to jump onto it. | |
625 | */ | |
626 | void inet_csk_destroy_sock(struct sock *sk) | |
627 | { | |
547b792c IJ |
628 | WARN_ON(sk->sk_state != TCP_CLOSE); |
629 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); | |
a019d6fe ACM |
630 | |
631 | /* It cannot be in hash table! */ | |
547b792c | 632 | WARN_ON(!sk_unhashed(sk)); |
a019d6fe | 633 | |
c720c7e8 ED |
634 | /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ |
635 | WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); | |
a019d6fe ACM |
636 | |
637 | sk->sk_prot->destroy(sk); | |
638 | ||
639 | sk_stream_kill_queues(sk); | |
640 | ||
641 | xfrm_sk_free_policy(sk); | |
642 | ||
643 | sk_refcnt_debug_release(sk); | |
644 | ||
dd24c001 | 645 | percpu_counter_dec(sk->sk_prot->orphan_count); |
a019d6fe ACM |
646 | sock_put(sk); |
647 | } | |
a019d6fe ACM |
648 | EXPORT_SYMBOL(inet_csk_destroy_sock); |
649 | ||
650 | int inet_csk_listen_start(struct sock *sk, const int nr_table_entries) | |
651 | { | |
652 | struct inet_sock *inet = inet_sk(sk); | |
653 | struct inet_connection_sock *icsk = inet_csk(sk); | |
654 | int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries); | |
655 | ||
656 | if (rc != 0) | |
657 | return rc; | |
658 | ||
659 | sk->sk_max_ack_backlog = 0; | |
660 | sk->sk_ack_backlog = 0; | |
661 | inet_csk_delack_init(sk); | |
662 | ||
663 | /* There is race window here: we announce ourselves listening, | |
664 | * but this transition is still not validated by get_port(). | |
665 | * It is OK, because this socket enters to hash table only | |
666 | * after validation is complete. | |
667 | */ | |
668 | sk->sk_state = TCP_LISTEN; | |
c720c7e8 ED |
669 | if (!sk->sk_prot->get_port(sk, inet->inet_num)) { |
670 | inet->inet_sport = htons(inet->inet_num); | |
a019d6fe ACM |
671 | |
672 | sk_dst_reset(sk); | |
673 | sk->sk_prot->hash(sk); | |
674 | ||
675 | return 0; | |
676 | } | |
677 | ||
678 | sk->sk_state = TCP_CLOSE; | |
679 | __reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
680 | return -EADDRINUSE; | |
681 | } | |
a019d6fe ACM |
682 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); |
683 | ||
684 | /* | |
685 | * This routine closes sockets which have been at least partially | |
686 | * opened, but not yet accepted. | |
687 | */ | |
688 | void inet_csk_listen_stop(struct sock *sk) | |
689 | { | |
690 | struct inet_connection_sock *icsk = inet_csk(sk); | |
691 | struct request_sock *acc_req; | |
692 | struct request_sock *req; | |
693 | ||
694 | inet_csk_delete_keepalive_timer(sk); | |
695 | ||
696 | /* make all the listen_opt local to us */ | |
697 | acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue); | |
698 | ||
699 | /* Following specs, it would be better either to send FIN | |
700 | * (and enter FIN-WAIT-1, it is normal close) | |
701 | * or to send active reset (abort). | |
702 | * Certainly, it is pretty dangerous while synflood, but it is | |
703 | * bad justification for our negligence 8) | |
704 | * To be honest, we are not able to make either | |
705 | * of the variants now. --ANK | |
706 | */ | |
707 | reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
708 | ||
709 | while ((req = acc_req) != NULL) { | |
710 | struct sock *child = req->sk; | |
711 | ||
712 | acc_req = req->dl_next; | |
713 | ||
714 | local_bh_disable(); | |
715 | bh_lock_sock(child); | |
547b792c | 716 | WARN_ON(sock_owned_by_user(child)); |
a019d6fe ACM |
717 | sock_hold(child); |
718 | ||
719 | sk->sk_prot->disconnect(child, O_NONBLOCK); | |
720 | ||
721 | sock_orphan(child); | |
722 | ||
eb4dea58 HX |
723 | percpu_counter_inc(sk->sk_prot->orphan_count); |
724 | ||
a019d6fe ACM |
725 | inet_csk_destroy_sock(child); |
726 | ||
727 | bh_unlock_sock(child); | |
728 | local_bh_enable(); | |
729 | sock_put(child); | |
730 | ||
731 | sk_acceptq_removed(sk); | |
732 | __reqsk_free(req); | |
733 | } | |
547b792c | 734 | WARN_ON(sk->sk_ack_backlog); |
a019d6fe | 735 | } |
a019d6fe | 736 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); |
af05dc93 ACM |
737 | |
738 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) | |
739 | { | |
740 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; | |
741 | const struct inet_sock *inet = inet_sk(sk); | |
742 | ||
743 | sin->sin_family = AF_INET; | |
c720c7e8 ED |
744 | sin->sin_addr.s_addr = inet->inet_daddr; |
745 | sin->sin_port = inet->inet_dport; | |
af05dc93 | 746 | } |
af05dc93 | 747 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); |
c4d93909 | 748 | |
dec73ff0 ACM |
749 | #ifdef CONFIG_COMPAT |
750 | int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, | |
751 | char __user *optval, int __user *optlen) | |
752 | { | |
dbeff12b | 753 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
754 | |
755 | if (icsk->icsk_af_ops->compat_getsockopt != NULL) | |
756 | return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, | |
757 | optval, optlen); | |
758 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
759 | optval, optlen); | |
760 | } | |
dec73ff0 ACM |
761 | EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); |
762 | ||
763 | int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 764 | char __user *optval, unsigned int optlen) |
dec73ff0 | 765 | { |
dbeff12b | 766 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
767 | |
768 | if (icsk->icsk_af_ops->compat_setsockopt != NULL) | |
769 | return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, | |
770 | optval, optlen); | |
771 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
772 | optval, optlen); | |
773 | } | |
dec73ff0 ACM |
774 | EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); |
775 | #endif |