]>
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 | 35 | struct local_ports sysctl_local_ports __read_mostly = { |
c4dbe54e | 36 | .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock), |
3c689b73 ED |
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 | { | |
95c96174 ED |
45 | unsigned int seq; |
46 | ||
227b60f5 | 47 | do { |
3c689b73 | 48 | seq = read_seqbegin(&sysctl_local_ports.lock); |
227b60f5 | 49 | |
3c689b73 ED |
50 | *low = sysctl_local_ports.range[0]; |
51 | *high = sysctl_local_ports.range[1]; | |
52 | } while (read_seqretry(&sysctl_local_ports.lock, seq)); | |
227b60f5 SH |
53 | } |
54 | EXPORT_SYMBOL(inet_get_local_port_range); | |
3f421baa | 55 | |
971af18b | 56 | int inet_csk_bind_conflict(const struct sock *sk, |
aacd9289 | 57 | const struct inet_bind_bucket *tb, bool relax) |
3f421baa | 58 | { |
3f421baa ACM |
59 | struct sock *sk2; |
60 | struct hlist_node *node; | |
61 | int reuse = sk->sk_reuse; | |
62 | ||
7477fd2e PE |
63 | /* |
64 | * Unlike other sk lookup places we do not check | |
65 | * for sk_net here, since _all_ the socks listed | |
66 | * in tb->owners list belong to the same net - the | |
67 | * one this bucket belongs to. | |
68 | */ | |
69 | ||
3f421baa ACM |
70 | sk_for_each_bound(sk2, node, &tb->owners) { |
71 | if (sk != sk2 && | |
72 | !inet_v6_ipv6only(sk2) && | |
73 | (!sk->sk_bound_dev_if || | |
74 | !sk2->sk_bound_dev_if || | |
75 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { | |
76 | if (!reuse || !sk2->sk_reuse || | |
3e8c806a | 77 | sk2->sk_state == TCP_LISTEN) { |
68835aba ED |
78 | const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2); |
79 | if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) || | |
80 | sk2_rcv_saddr == sk_rcv_saddr(sk)) | |
3f421baa | 81 | break; |
8d238b25 | 82 | } |
aacd9289 AC |
83 | if (!relax && reuse && sk2->sk_reuse && |
84 | sk2->sk_state != TCP_LISTEN) { | |
85 | const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2); | |
86 | ||
87 | if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) || | |
88 | sk2_rcv_saddr == sk_rcv_saddr(sk)) | |
89 | break; | |
90 | } | |
3f421baa ACM |
91 | } |
92 | } | |
93 | return node != NULL; | |
94 | } | |
971af18b ACM |
95 | EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); |
96 | ||
3f421baa ACM |
97 | /* Obtain a reference to a local port for the given sock, |
98 | * if snum is zero it means select any available local port. | |
99 | */ | |
ab1e0a13 | 100 | int inet_csk_get_port(struct sock *sk, unsigned short snum) |
3f421baa | 101 | { |
39d8cda7 | 102 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3f421baa ACM |
103 | struct inet_bind_hashbucket *head; |
104 | struct hlist_node *node; | |
105 | struct inet_bind_bucket *tb; | |
a9d8f911 | 106 | int ret, attempts = 5; |
3b1e0a65 | 107 | struct net *net = sock_net(sk); |
a9d8f911 | 108 | int smallest_size = -1, smallest_rover; |
3f421baa ACM |
109 | |
110 | local_bh_disable(); | |
111 | if (!snum) { | |
227b60f5 SH |
112 | int remaining, rover, low, high; |
113 | ||
a9d8f911 | 114 | again: |
227b60f5 | 115 | inet_get_local_port_range(&low, &high); |
a25de534 | 116 | remaining = (high - low) + 1; |
a9d8f911 | 117 | smallest_rover = rover = net_random() % remaining + low; |
3f421baa | 118 | |
a9d8f911 | 119 | smallest_size = -1; |
3f421baa | 120 | do { |
e3826f1e AW |
121 | if (inet_is_reserved_local_port(rover)) |
122 | goto next_nolock; | |
7f635ab7 PE |
123 | head = &hashinfo->bhash[inet_bhashfn(net, rover, |
124 | hashinfo->bhash_size)]; | |
3f421baa ACM |
125 | spin_lock(&head->lock); |
126 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
09ad9bc7 | 127 | if (net_eq(ib_net(tb), net) && tb->port == rover) { |
a9d8f911 EP |
128 | if (tb->fastreuse > 0 && |
129 | sk->sk_reuse && | |
130 | sk->sk_state != TCP_LISTEN && | |
131 | (tb->num_owners < smallest_size || smallest_size == -1)) { | |
132 | smallest_size = tb->num_owners; | |
133 | smallest_rover = rover; | |
aacd9289 AC |
134 | if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 && |
135 | !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) { | |
8d238b25 | 136 | snum = smallest_rover; |
fddb7b57 | 137 | goto tb_found; |
a9d8f911 EP |
138 | } |
139 | } | |
aacd9289 | 140 | if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) { |
2b05ad33 | 141 | snum = rover; |
fddb7b57 | 142 | goto tb_found; |
2b05ad33 | 143 | } |
3f421baa | 144 | goto next; |
a9d8f911 | 145 | } |
3f421baa ACM |
146 | break; |
147 | next: | |
148 | spin_unlock(&head->lock); | |
e3826f1e | 149 | next_nolock: |
6df71634 SH |
150 | if (++rover > high) |
151 | rover = low; | |
3f421baa | 152 | } while (--remaining > 0); |
3f421baa ACM |
153 | |
154 | /* Exhausted local port range during search? It is not | |
155 | * possible for us to be holding one of the bind hash | |
156 | * locks if this test triggers, because if 'remaining' | |
157 | * drops to zero, we broke out of the do/while loop at | |
158 | * the top level, not from the 'break;' statement. | |
159 | */ | |
160 | ret = 1; | |
a9d8f911 EP |
161 | if (remaining <= 0) { |
162 | if (smallest_size != -1) { | |
163 | snum = smallest_rover; | |
164 | goto have_snum; | |
165 | } | |
3f421baa | 166 | goto fail; |
a9d8f911 | 167 | } |
3f421baa ACM |
168 | /* OK, here is the one we will use. HEAD is |
169 | * non-NULL and we hold it's mutex. | |
170 | */ | |
171 | snum = rover; | |
172 | } else { | |
a9d8f911 | 173 | have_snum: |
7f635ab7 PE |
174 | head = &hashinfo->bhash[inet_bhashfn(net, snum, |
175 | hashinfo->bhash_size)]; | |
3f421baa ACM |
176 | spin_lock(&head->lock); |
177 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
09ad9bc7 | 178 | if (net_eq(ib_net(tb), net) && tb->port == snum) |
3f421baa ACM |
179 | goto tb_found; |
180 | } | |
181 | tb = NULL; | |
182 | goto tb_not_found; | |
183 | tb_found: | |
184 | if (!hlist_empty(&tb->owners)) { | |
3f421baa | 185 | if (tb->fastreuse > 0 && |
a9d8f911 EP |
186 | sk->sk_reuse && sk->sk_state != TCP_LISTEN && |
187 | smallest_size == -1) { | |
3f421baa ACM |
188 | goto success; |
189 | } else { | |
190 | ret = 1; | |
aacd9289 | 191 | if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) { |
5add3009 SH |
192 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN && |
193 | smallest_size != -1 && --attempts >= 0) { | |
a9d8f911 EP |
194 | spin_unlock(&head->lock); |
195 | goto again; | |
196 | } | |
aacd9289 | 197 | |
3f421baa | 198 | goto fail_unlock; |
a9d8f911 | 199 | } |
3f421baa ACM |
200 | } |
201 | } | |
202 | tb_not_found: | |
203 | ret = 1; | |
941b1d22 PE |
204 | if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, |
205 | net, head, snum)) == NULL) | |
3f421baa ACM |
206 | goto fail_unlock; |
207 | if (hlist_empty(&tb->owners)) { | |
208 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) | |
209 | tb->fastreuse = 1; | |
210 | else | |
211 | tb->fastreuse = 0; | |
212 | } else if (tb->fastreuse && | |
213 | (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) | |
214 | tb->fastreuse = 0; | |
215 | success: | |
216 | if (!inet_csk(sk)->icsk_bind_hash) | |
217 | inet_bind_hash(sk, tb, snum); | |
547b792c | 218 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
e905a9ed | 219 | ret = 0; |
3f421baa ACM |
220 | |
221 | fail_unlock: | |
222 | spin_unlock(&head->lock); | |
223 | fail: | |
224 | local_bh_enable(); | |
225 | return ret; | |
226 | } | |
3f421baa ACM |
227 | EXPORT_SYMBOL_GPL(inet_csk_get_port); |
228 | ||
229 | /* | |
230 | * Wait for an incoming connection, avoid race conditions. This must be called | |
231 | * with the socket locked. | |
232 | */ | |
233 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) | |
234 | { | |
235 | struct inet_connection_sock *icsk = inet_csk(sk); | |
236 | DEFINE_WAIT(wait); | |
237 | int err; | |
238 | ||
239 | /* | |
240 | * True wake-one mechanism for incoming connections: only | |
241 | * one process gets woken up, not the 'whole herd'. | |
242 | * Since we do not 'race & poll' for established sockets | |
243 | * anymore, the common case will execute the loop only once. | |
244 | * | |
245 | * Subtle issue: "add_wait_queue_exclusive()" will be added | |
246 | * after any current non-exclusive waiters, and we know that | |
247 | * it will always _stay_ after any new non-exclusive waiters | |
248 | * because all non-exclusive waiters are added at the | |
249 | * beginning of the wait-queue. As such, it's ok to "drop" | |
250 | * our exclusiveness temporarily when we get woken up without | |
251 | * having to remove and re-insert us on the wait queue. | |
252 | */ | |
253 | for (;;) { | |
aa395145 | 254 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, |
3f421baa ACM |
255 | TASK_INTERRUPTIBLE); |
256 | release_sock(sk); | |
257 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
258 | timeo = schedule_timeout(timeo); | |
259 | lock_sock(sk); | |
260 | err = 0; | |
261 | if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
262 | break; | |
263 | err = -EINVAL; | |
264 | if (sk->sk_state != TCP_LISTEN) | |
265 | break; | |
266 | err = sock_intr_errno(timeo); | |
267 | if (signal_pending(current)) | |
268 | break; | |
269 | err = -EAGAIN; | |
270 | if (!timeo) | |
271 | break; | |
272 | } | |
aa395145 | 273 | finish_wait(sk_sleep(sk), &wait); |
3f421baa ACM |
274 | return err; |
275 | } | |
276 | ||
277 | /* | |
278 | * This will accept the next outstanding connection. | |
279 | */ | |
280 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) | |
281 | { | |
282 | struct inet_connection_sock *icsk = inet_csk(sk); | |
283 | struct sock *newsk; | |
284 | int error; | |
285 | ||
286 | lock_sock(sk); | |
287 | ||
288 | /* We need to make sure that this socket is listening, | |
289 | * and that it has something pending. | |
290 | */ | |
291 | error = -EINVAL; | |
292 | if (sk->sk_state != TCP_LISTEN) | |
293 | goto out_err; | |
294 | ||
295 | /* Find already established connection */ | |
296 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) { | |
297 | long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); | |
298 | ||
299 | /* If this is a non blocking socket don't sleep */ | |
300 | error = -EAGAIN; | |
301 | if (!timeo) | |
302 | goto out_err; | |
303 | ||
304 | error = inet_csk_wait_for_connect(sk, timeo); | |
305 | if (error) | |
306 | goto out_err; | |
307 | } | |
308 | ||
309 | newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk); | |
547b792c | 310 | WARN_ON(newsk->sk_state == TCP_SYN_RECV); |
3f421baa ACM |
311 | out: |
312 | release_sock(sk); | |
313 | return newsk; | |
314 | out_err: | |
315 | newsk = NULL; | |
316 | *err = error; | |
317 | goto out; | |
318 | } | |
3f421baa ACM |
319 | EXPORT_SYMBOL(inet_csk_accept); |
320 | ||
321 | /* | |
322 | * Using different timers for retransmit, delayed acks and probes | |
e905a9ed | 323 | * We may wish use just one timer maintaining a list of expire jiffies |
3f421baa ACM |
324 | * to optimize. |
325 | */ | |
326 | void inet_csk_init_xmit_timers(struct sock *sk, | |
327 | void (*retransmit_handler)(unsigned long), | |
328 | void (*delack_handler)(unsigned long), | |
329 | void (*keepalive_handler)(unsigned long)) | |
330 | { | |
331 | struct inet_connection_sock *icsk = inet_csk(sk); | |
332 | ||
b24b8a24 PE |
333 | setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, |
334 | (unsigned long)sk); | |
335 | setup_timer(&icsk->icsk_delack_timer, delack_handler, | |
336 | (unsigned long)sk); | |
337 | setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); | |
3f421baa ACM |
338 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
339 | } | |
3f421baa ACM |
340 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); |
341 | ||
342 | void inet_csk_clear_xmit_timers(struct sock *sk) | |
343 | { | |
344 | struct inet_connection_sock *icsk = inet_csk(sk); | |
345 | ||
346 | icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; | |
347 | ||
348 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
349 | sk_stop_timer(sk, &icsk->icsk_delack_timer); | |
350 | sk_stop_timer(sk, &sk->sk_timer); | |
351 | } | |
3f421baa ACM |
352 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); |
353 | ||
354 | void inet_csk_delete_keepalive_timer(struct sock *sk) | |
355 | { | |
356 | sk_stop_timer(sk, &sk->sk_timer); | |
357 | } | |
3f421baa ACM |
358 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); |
359 | ||
360 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) | |
361 | { | |
362 | sk_reset_timer(sk, &sk->sk_timer, jiffies + len); | |
363 | } | |
3f421baa ACM |
364 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); |
365 | ||
d9319100 | 366 | struct dst_entry *inet_csk_route_req(struct sock *sk, |
6bd023f3 | 367 | struct flowi4 *fl4, |
3f421baa ACM |
368 | const struct request_sock *req) |
369 | { | |
370 | struct rtable *rt; | |
371 | const struct inet_request_sock *ireq = inet_rsk(req); | |
f6d8bd05 | 372 | struct ip_options_rcu *opt = inet_rsk(req)->opt; |
84a3aa00 | 373 | struct net *net = sock_net(sk); |
3f421baa | 374 | |
6bd023f3 | 375 | flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, |
e79d9bc7 DM |
376 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
377 | sk->sk_protocol, inet_sk_flowi_flags(sk), | |
f6d8bd05 | 378 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr, |
e79d9bc7 | 379 | ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport); |
6bd023f3 DM |
380 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); |
381 | rt = ip_route_output_flow(net, fl4, sk); | |
b23dd4fe | 382 | if (IS_ERR(rt)) |
857a6e0a | 383 | goto no_route; |
6bd023f3 | 384 | if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway) |
857a6e0a | 385 | goto route_err; |
d8d1f30b | 386 | return &rt->dst; |
857a6e0a IJ |
387 | |
388 | route_err: | |
389 | ip_rt_put(rt); | |
390 | no_route: | |
391 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); | |
392 | return NULL; | |
3f421baa | 393 | } |
3f421baa ACM |
394 | EXPORT_SYMBOL_GPL(inet_csk_route_req); |
395 | ||
77357a95 DM |
396 | struct dst_entry *inet_csk_route_child_sock(struct sock *sk, |
397 | struct sock *newsk, | |
398 | const struct request_sock *req) | |
399 | { | |
400 | const struct inet_request_sock *ireq = inet_rsk(req); | |
401 | struct inet_sock *newinet = inet_sk(newsk); | |
402 | struct ip_options_rcu *opt = ireq->opt; | |
403 | struct net *net = sock_net(sk); | |
404 | struct flowi4 *fl4; | |
405 | struct rtable *rt; | |
406 | ||
407 | fl4 = &newinet->cork.fl.u.ip4; | |
408 | flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, | |
409 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, | |
410 | sk->sk_protocol, inet_sk_flowi_flags(sk), | |
411 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr, | |
412 | ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport); | |
413 | security_req_classify_flow(req, flowi4_to_flowi(fl4)); | |
414 | rt = ip_route_output_flow(net, fl4, sk); | |
415 | if (IS_ERR(rt)) | |
416 | goto no_route; | |
417 | if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway) | |
418 | goto route_err; | |
419 | return &rt->dst; | |
420 | ||
421 | route_err: | |
422 | ip_rt_put(rt); | |
423 | no_route: | |
424 | IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); | |
425 | return NULL; | |
426 | } | |
427 | EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); | |
428 | ||
6b72977b | 429 | static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport, |
72a3effa | 430 | const u32 rnd, const u32 synq_hsize) |
3f421baa | 431 | { |
6b72977b | 432 | return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1); |
3f421baa ACM |
433 | } |
434 | ||
dfd56b8b | 435 | #if IS_ENABLED(CONFIG_IPV6) |
3f421baa ACM |
436 | #define AF_INET_FAMILY(fam) ((fam) == AF_INET) |
437 | #else | |
438 | #define AF_INET_FAMILY(fam) 1 | |
439 | #endif | |
440 | ||
441 | struct request_sock *inet_csk_search_req(const struct sock *sk, | |
442 | struct request_sock ***prevp, | |
6b72977b | 443 | const __be16 rport, const __be32 raddr, |
7f25afbb | 444 | const __be32 laddr) |
3f421baa ACM |
445 | { |
446 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
447 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
448 | struct request_sock *req, **prev; | |
449 | ||
450 | for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd, | |
451 | lopt->nr_table_entries)]; | |
452 | (req = *prev) != NULL; | |
453 | prev = &req->dl_next) { | |
454 | const struct inet_request_sock *ireq = inet_rsk(req); | |
455 | ||
456 | if (ireq->rmt_port == rport && | |
457 | ireq->rmt_addr == raddr && | |
458 | ireq->loc_addr == laddr && | |
459 | AF_INET_FAMILY(req->rsk_ops->family)) { | |
547b792c | 460 | WARN_ON(req->sk); |
3f421baa ACM |
461 | *prevp = prev; |
462 | break; | |
463 | } | |
464 | } | |
465 | ||
466 | return req; | |
467 | } | |
3f421baa ACM |
468 | EXPORT_SYMBOL_GPL(inet_csk_search_req); |
469 | ||
470 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, | |
c2977c22 | 471 | unsigned long timeout) |
3f421baa ACM |
472 | { |
473 | struct inet_connection_sock *icsk = inet_csk(sk); | |
474 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
475 | const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port, | |
476 | lopt->hash_rnd, lopt->nr_table_entries); | |
477 | ||
478 | reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout); | |
479 | inet_csk_reqsk_queue_added(sk, timeout); | |
480 | } | |
4bc2f18b | 481 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
3f421baa | 482 | |
a019d6fe ACM |
483 | /* Only thing we need from tcp.h */ |
484 | extern int sysctl_tcp_synack_retries; | |
485 | ||
9f1d2604 | 486 | |
0c3d79bc JA |
487 | /* Decide when to expire the request and when to resend SYN-ACK */ |
488 | static inline void syn_ack_recalc(struct request_sock *req, const int thresh, | |
489 | const int max_retries, | |
490 | const u8 rskq_defer_accept, | |
491 | int *expire, int *resend) | |
492 | { | |
493 | if (!rskq_defer_accept) { | |
494 | *expire = req->retrans >= thresh; | |
495 | *resend = 1; | |
496 | return; | |
497 | } | |
498 | *expire = req->retrans >= thresh && | |
499 | (!inet_rsk(req)->acked || req->retrans >= max_retries); | |
500 | /* | |
501 | * Do not resend while waiting for data after ACK, | |
502 | * start to resend on end of deferring period to give | |
503 | * last chance for data or ACK to create established socket. | |
504 | */ | |
505 | *resend = !inet_rsk(req)->acked || | |
506 | req->retrans >= rskq_defer_accept - 1; | |
507 | } | |
508 | ||
a019d6fe ACM |
509 | void inet_csk_reqsk_queue_prune(struct sock *parent, |
510 | const unsigned long interval, | |
511 | const unsigned long timeout, | |
512 | const unsigned long max_rto) | |
513 | { | |
514 | struct inet_connection_sock *icsk = inet_csk(parent); | |
515 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; | |
516 | struct listen_sock *lopt = queue->listen_opt; | |
ec0a1966 DM |
517 | int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries; |
518 | int thresh = max_retries; | |
a019d6fe ACM |
519 | unsigned long now = jiffies; |
520 | struct request_sock **reqp, *req; | |
521 | int i, budget; | |
522 | ||
523 | if (lopt == NULL || lopt->qlen == 0) | |
524 | return; | |
525 | ||
526 | /* Normally all the openreqs are young and become mature | |
527 | * (i.e. converted to established socket) for first timeout. | |
fd4f2cea | 528 | * If synack was not acknowledged for 1 second, it means |
a019d6fe ACM |
529 | * one of the following things: synack was lost, ack was lost, |
530 | * rtt is high or nobody planned to ack (i.e. synflood). | |
531 | * When server is a bit loaded, queue is populated with old | |
532 | * open requests, reducing effective size of queue. | |
533 | * When server is well loaded, queue size reduces to zero | |
534 | * after several minutes of work. It is not synflood, | |
535 | * it is normal operation. The solution is pruning | |
536 | * too old entries overriding normal timeout, when | |
537 | * situation becomes dangerous. | |
538 | * | |
539 | * Essentially, we reserve half of room for young | |
540 | * embrions; and abort old ones without pity, if old | |
541 | * ones are about to clog our table. | |
542 | */ | |
543 | if (lopt->qlen>>(lopt->max_qlen_log-1)) { | |
544 | int young = (lopt->qlen_young<<1); | |
545 | ||
546 | while (thresh > 2) { | |
547 | if (lopt->qlen < young) | |
548 | break; | |
549 | thresh--; | |
550 | young <<= 1; | |
551 | } | |
552 | } | |
553 | ||
ec0a1966 DM |
554 | if (queue->rskq_defer_accept) |
555 | max_retries = queue->rskq_defer_accept; | |
556 | ||
a019d6fe ACM |
557 | budget = 2 * (lopt->nr_table_entries / (timeout / interval)); |
558 | i = lopt->clock_hand; | |
559 | ||
560 | do { | |
561 | reqp=&lopt->syn_table[i]; | |
562 | while ((req = *reqp) != NULL) { | |
563 | if (time_after_eq(now, req->expires)) { | |
0c3d79bc JA |
564 | int expire = 0, resend = 0; |
565 | ||
566 | syn_ack_recalc(req, thresh, max_retries, | |
567 | queue->rskq_defer_accept, | |
568 | &expire, &resend); | |
c72e1183 | 569 | req->rsk_ops->syn_ack_timeout(parent, req); |
0c3d79bc JA |
570 | if (!expire && |
571 | (!resend || | |
e6b4d113 | 572 | !req->rsk_ops->rtx_syn_ack(parent, req, NULL) || |
0c3d79bc | 573 | inet_rsk(req)->acked)) { |
a019d6fe ACM |
574 | unsigned long timeo; |
575 | ||
576 | if (req->retrans++ == 0) | |
577 | lopt->qlen_young--; | |
578 | timeo = min((timeout << req->retrans), max_rto); | |
579 | req->expires = now + timeo; | |
580 | reqp = &req->dl_next; | |
581 | continue; | |
582 | } | |
583 | ||
584 | /* Drop this request */ | |
585 | inet_csk_reqsk_queue_unlink(parent, req, reqp); | |
586 | reqsk_queue_removed(queue, req); | |
587 | reqsk_free(req); | |
588 | continue; | |
589 | } | |
590 | reqp = &req->dl_next; | |
591 | } | |
592 | ||
593 | i = (i + 1) & (lopt->nr_table_entries - 1); | |
594 | ||
595 | } while (--budget > 0); | |
596 | ||
597 | lopt->clock_hand = i; | |
598 | ||
599 | if (lopt->qlen) | |
600 | inet_csk_reset_keepalive_timer(parent, interval); | |
601 | } | |
a019d6fe ACM |
602 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune); |
603 | ||
e56c57d0 ED |
604 | /** |
605 | * inet_csk_clone_lock - clone an inet socket, and lock its clone | |
606 | * @sk: the socket to clone | |
607 | * @req: request_sock | |
608 | * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) | |
609 | * | |
610 | * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) | |
611 | */ | |
612 | struct sock *inet_csk_clone_lock(const struct sock *sk, | |
613 | const struct request_sock *req, | |
614 | const gfp_t priority) | |
9f1d2604 | 615 | { |
e56c57d0 | 616 | struct sock *newsk = sk_clone_lock(sk, priority); |
9f1d2604 ACM |
617 | |
618 | if (newsk != NULL) { | |
619 | struct inet_connection_sock *newicsk = inet_csk(newsk); | |
620 | ||
621 | newsk->sk_state = TCP_SYN_RECV; | |
622 | newicsk->icsk_bind_hash = NULL; | |
623 | ||
c720c7e8 ED |
624 | inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port; |
625 | inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port); | |
626 | inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port; | |
9f1d2604 ACM |
627 | newsk->sk_write_space = sk_stream_write_space; |
628 | ||
629 | newicsk->icsk_retransmits = 0; | |
6687e988 ACM |
630 | newicsk->icsk_backoff = 0; |
631 | newicsk->icsk_probes_out = 0; | |
9f1d2604 ACM |
632 | |
633 | /* Deinitialize accept_queue to trap illegal accesses. */ | |
634 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); | |
4237c75c VY |
635 | |
636 | security_inet_csk_clone(newsk, req); | |
9f1d2604 ACM |
637 | } |
638 | return newsk; | |
639 | } | |
e56c57d0 | 640 | EXPORT_SYMBOL_GPL(inet_csk_clone_lock); |
a019d6fe ACM |
641 | |
642 | /* | |
643 | * At this point, there should be no process reference to this | |
644 | * socket, and thus no user references at all. Therefore we | |
645 | * can assume the socket waitqueue is inactive and nobody will | |
646 | * try to jump onto it. | |
647 | */ | |
648 | void inet_csk_destroy_sock(struct sock *sk) | |
649 | { | |
547b792c IJ |
650 | WARN_ON(sk->sk_state != TCP_CLOSE); |
651 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); | |
a019d6fe ACM |
652 | |
653 | /* It cannot be in hash table! */ | |
547b792c | 654 | WARN_ON(!sk_unhashed(sk)); |
a019d6fe | 655 | |
c720c7e8 ED |
656 | /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ |
657 | WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); | |
a019d6fe ACM |
658 | |
659 | sk->sk_prot->destroy(sk); | |
660 | ||
661 | sk_stream_kill_queues(sk); | |
662 | ||
663 | xfrm_sk_free_policy(sk); | |
664 | ||
665 | sk_refcnt_debug_release(sk); | |
666 | ||
dd24c001 | 667 | percpu_counter_dec(sk->sk_prot->orphan_count); |
a019d6fe ACM |
668 | sock_put(sk); |
669 | } | |
a019d6fe ACM |
670 | EXPORT_SYMBOL(inet_csk_destroy_sock); |
671 | ||
672 | int inet_csk_listen_start(struct sock *sk, const int nr_table_entries) | |
673 | { | |
674 | struct inet_sock *inet = inet_sk(sk); | |
675 | struct inet_connection_sock *icsk = inet_csk(sk); | |
676 | int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries); | |
677 | ||
678 | if (rc != 0) | |
679 | return rc; | |
680 | ||
681 | sk->sk_max_ack_backlog = 0; | |
682 | sk->sk_ack_backlog = 0; | |
683 | inet_csk_delack_init(sk); | |
684 | ||
685 | /* There is race window here: we announce ourselves listening, | |
686 | * but this transition is still not validated by get_port(). | |
687 | * It is OK, because this socket enters to hash table only | |
688 | * after validation is complete. | |
689 | */ | |
690 | sk->sk_state = TCP_LISTEN; | |
c720c7e8 ED |
691 | if (!sk->sk_prot->get_port(sk, inet->inet_num)) { |
692 | inet->inet_sport = htons(inet->inet_num); | |
a019d6fe ACM |
693 | |
694 | sk_dst_reset(sk); | |
695 | sk->sk_prot->hash(sk); | |
696 | ||
697 | return 0; | |
698 | } | |
699 | ||
700 | sk->sk_state = TCP_CLOSE; | |
701 | __reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
702 | return -EADDRINUSE; | |
703 | } | |
a019d6fe ACM |
704 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); |
705 | ||
706 | /* | |
707 | * This routine closes sockets which have been at least partially | |
708 | * opened, but not yet accepted. | |
709 | */ | |
710 | void inet_csk_listen_stop(struct sock *sk) | |
711 | { | |
712 | struct inet_connection_sock *icsk = inet_csk(sk); | |
713 | struct request_sock *acc_req; | |
714 | struct request_sock *req; | |
715 | ||
716 | inet_csk_delete_keepalive_timer(sk); | |
717 | ||
718 | /* make all the listen_opt local to us */ | |
719 | acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue); | |
720 | ||
721 | /* Following specs, it would be better either to send FIN | |
722 | * (and enter FIN-WAIT-1, it is normal close) | |
723 | * or to send active reset (abort). | |
724 | * Certainly, it is pretty dangerous while synflood, but it is | |
725 | * bad justification for our negligence 8) | |
726 | * To be honest, we are not able to make either | |
727 | * of the variants now. --ANK | |
728 | */ | |
729 | reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
730 | ||
731 | while ((req = acc_req) != NULL) { | |
732 | struct sock *child = req->sk; | |
733 | ||
734 | acc_req = req->dl_next; | |
735 | ||
736 | local_bh_disable(); | |
737 | bh_lock_sock(child); | |
547b792c | 738 | WARN_ON(sock_owned_by_user(child)); |
a019d6fe ACM |
739 | sock_hold(child); |
740 | ||
741 | sk->sk_prot->disconnect(child, O_NONBLOCK); | |
742 | ||
743 | sock_orphan(child); | |
744 | ||
eb4dea58 HX |
745 | percpu_counter_inc(sk->sk_prot->orphan_count); |
746 | ||
a019d6fe ACM |
747 | inet_csk_destroy_sock(child); |
748 | ||
749 | bh_unlock_sock(child); | |
750 | local_bh_enable(); | |
751 | sock_put(child); | |
752 | ||
753 | sk_acceptq_removed(sk); | |
754 | __reqsk_free(req); | |
755 | } | |
547b792c | 756 | WARN_ON(sk->sk_ack_backlog); |
a019d6fe | 757 | } |
a019d6fe | 758 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); |
af05dc93 ACM |
759 | |
760 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) | |
761 | { | |
762 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; | |
763 | const struct inet_sock *inet = inet_sk(sk); | |
764 | ||
765 | sin->sin_family = AF_INET; | |
c720c7e8 ED |
766 | sin->sin_addr.s_addr = inet->inet_daddr; |
767 | sin->sin_port = inet->inet_dport; | |
af05dc93 | 768 | } |
af05dc93 | 769 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); |
c4d93909 | 770 | |
dec73ff0 ACM |
771 | #ifdef CONFIG_COMPAT |
772 | int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, | |
773 | char __user *optval, int __user *optlen) | |
774 | { | |
dbeff12b | 775 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
776 | |
777 | if (icsk->icsk_af_ops->compat_getsockopt != NULL) | |
778 | return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, | |
779 | optval, optlen); | |
780 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
781 | optval, optlen); | |
782 | } | |
dec73ff0 ACM |
783 | EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); |
784 | ||
785 | int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 786 | char __user *optval, unsigned int optlen) |
dec73ff0 | 787 | { |
dbeff12b | 788 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
789 | |
790 | if (icsk->icsk_af_ops->compat_setsockopt != NULL) | |
791 | return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, | |
792 | optval, optlen); | |
793 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
794 | optval, optlen); | |
795 | } | |
dec73ff0 ACM |
796 | EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); |
797 | #endif |