]>
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 SH |
39 | |
40 | void inet_get_local_port_range(int *low, int *high) | |
41 | { | |
42 | unsigned seq; | |
43 | do { | |
3c689b73 | 44 | seq = read_seqbegin(&sysctl_local_ports.lock); |
227b60f5 | 45 | |
3c689b73 ED |
46 | *low = sysctl_local_ports.range[0]; |
47 | *high = sysctl_local_ports.range[1]; | |
48 | } while (read_seqretry(&sysctl_local_ports.lock, seq)); | |
227b60f5 SH |
49 | } |
50 | EXPORT_SYMBOL(inet_get_local_port_range); | |
3f421baa | 51 | |
971af18b ACM |
52 | int inet_csk_bind_conflict(const struct sock *sk, |
53 | const struct inet_bind_bucket *tb) | |
3f421baa | 54 | { |
82103232 | 55 | const __be32 sk_rcv_saddr = inet_rcv_saddr(sk); |
3f421baa ACM |
56 | struct sock *sk2; |
57 | struct hlist_node *node; | |
58 | int reuse = sk->sk_reuse; | |
59 | ||
7477fd2e PE |
60 | /* |
61 | * Unlike other sk lookup places we do not check | |
62 | * for sk_net here, since _all_ the socks listed | |
63 | * in tb->owners list belong to the same net - the | |
64 | * one this bucket belongs to. | |
65 | */ | |
66 | ||
3f421baa ACM |
67 | sk_for_each_bound(sk2, node, &tb->owners) { |
68 | if (sk != sk2 && | |
69 | !inet_v6_ipv6only(sk2) && | |
70 | (!sk->sk_bound_dev_if || | |
71 | !sk2->sk_bound_dev_if || | |
72 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { | |
fda48a0d ED |
73 | const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2); |
74 | ||
3f421baa ACM |
75 | if (!reuse || !sk2->sk_reuse || |
76 | sk2->sk_state == TCP_LISTEN) { | |
3f421baa ACM |
77 | if (!sk2_rcv_saddr || !sk_rcv_saddr || |
78 | sk2_rcv_saddr == sk_rcv_saddr) | |
79 | break; | |
fda48a0d ED |
80 | } else if (reuse && sk2->sk_reuse && |
81 | sk2_rcv_saddr && | |
82 | sk2_rcv_saddr == sk_rcv_saddr) | |
83 | break; | |
3f421baa ACM |
84 | } |
85 | } | |
86 | return node != NULL; | |
87 | } | |
88 | ||
971af18b ACM |
89 | EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); |
90 | ||
3f421baa ACM |
91 | /* Obtain a reference to a local port for the given sock, |
92 | * if snum is zero it means select any available local port. | |
93 | */ | |
ab1e0a13 | 94 | int inet_csk_get_port(struct sock *sk, unsigned short snum) |
3f421baa | 95 | { |
39d8cda7 | 96 | struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; |
3f421baa ACM |
97 | struct inet_bind_hashbucket *head; |
98 | struct hlist_node *node; | |
99 | struct inet_bind_bucket *tb; | |
a9d8f911 | 100 | int ret, attempts = 5; |
3b1e0a65 | 101 | struct net *net = sock_net(sk); |
a9d8f911 | 102 | int smallest_size = -1, smallest_rover; |
3f421baa ACM |
103 | |
104 | local_bh_disable(); | |
105 | if (!snum) { | |
227b60f5 SH |
106 | int remaining, rover, low, high; |
107 | ||
a9d8f911 | 108 | again: |
227b60f5 | 109 | inet_get_local_port_range(&low, &high); |
a25de534 | 110 | remaining = (high - low) + 1; |
a9d8f911 | 111 | smallest_rover = rover = net_random() % remaining + low; |
3f421baa | 112 | |
a9d8f911 | 113 | smallest_size = -1; |
3f421baa | 114 | do { |
7f635ab7 PE |
115 | head = &hashinfo->bhash[inet_bhashfn(net, rover, |
116 | hashinfo->bhash_size)]; | |
3f421baa ACM |
117 | spin_lock(&head->lock); |
118 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
09ad9bc7 | 119 | if (net_eq(ib_net(tb), net) && tb->port == rover) { |
a9d8f911 EP |
120 | if (tb->fastreuse > 0 && |
121 | sk->sk_reuse && | |
122 | sk->sk_state != TCP_LISTEN && | |
123 | (tb->num_owners < smallest_size || smallest_size == -1)) { | |
124 | smallest_size = tb->num_owners; | |
125 | smallest_rover = rover; | |
24dd1fa1 | 126 | if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) { |
fda48a0d ED |
127 | if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) { |
128 | spin_unlock(&head->lock); | |
129 | snum = smallest_rover; | |
130 | goto have_snum; | |
131 | } | |
a9d8f911 EP |
132 | } |
133 | } | |
3f421baa | 134 | goto next; |
a9d8f911 | 135 | } |
3f421baa ACM |
136 | break; |
137 | next: | |
138 | spin_unlock(&head->lock); | |
6df71634 SH |
139 | if (++rover > high) |
140 | rover = low; | |
3f421baa | 141 | } while (--remaining > 0); |
3f421baa ACM |
142 | |
143 | /* Exhausted local port range during search? It is not | |
144 | * possible for us to be holding one of the bind hash | |
145 | * locks if this test triggers, because if 'remaining' | |
146 | * drops to zero, we broke out of the do/while loop at | |
147 | * the top level, not from the 'break;' statement. | |
148 | */ | |
149 | ret = 1; | |
a9d8f911 EP |
150 | if (remaining <= 0) { |
151 | if (smallest_size != -1) { | |
152 | snum = smallest_rover; | |
153 | goto have_snum; | |
154 | } | |
3f421baa | 155 | goto fail; |
a9d8f911 | 156 | } |
3f421baa ACM |
157 | /* OK, here is the one we will use. HEAD is |
158 | * non-NULL and we hold it's mutex. | |
159 | */ | |
160 | snum = rover; | |
161 | } else { | |
a9d8f911 | 162 | have_snum: |
7f635ab7 PE |
163 | head = &hashinfo->bhash[inet_bhashfn(net, snum, |
164 | hashinfo->bhash_size)]; | |
3f421baa ACM |
165 | spin_lock(&head->lock); |
166 | inet_bind_bucket_for_each(tb, node, &head->chain) | |
09ad9bc7 | 167 | if (net_eq(ib_net(tb), net) && tb->port == snum) |
3f421baa ACM |
168 | goto tb_found; |
169 | } | |
170 | tb = NULL; | |
171 | goto tb_not_found; | |
172 | tb_found: | |
173 | if (!hlist_empty(&tb->owners)) { | |
3f421baa | 174 | if (tb->fastreuse > 0 && |
a9d8f911 EP |
175 | sk->sk_reuse && sk->sk_state != TCP_LISTEN && |
176 | smallest_size == -1) { | |
3f421baa ACM |
177 | goto success; |
178 | } else { | |
179 | ret = 1; | |
a9d8f911 | 180 | if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) { |
5add3009 SH |
181 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN && |
182 | smallest_size != -1 && --attempts >= 0) { | |
a9d8f911 EP |
183 | spin_unlock(&head->lock); |
184 | goto again; | |
185 | } | |
3f421baa | 186 | goto fail_unlock; |
a9d8f911 | 187 | } |
3f421baa ACM |
188 | } |
189 | } | |
190 | tb_not_found: | |
191 | ret = 1; | |
941b1d22 PE |
192 | if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, |
193 | net, head, snum)) == NULL) | |
3f421baa ACM |
194 | goto fail_unlock; |
195 | if (hlist_empty(&tb->owners)) { | |
196 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) | |
197 | tb->fastreuse = 1; | |
198 | else | |
199 | tb->fastreuse = 0; | |
200 | } else if (tb->fastreuse && | |
201 | (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) | |
202 | tb->fastreuse = 0; | |
203 | success: | |
204 | if (!inet_csk(sk)->icsk_bind_hash) | |
205 | inet_bind_hash(sk, tb, snum); | |
547b792c | 206 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
e905a9ed | 207 | ret = 0; |
3f421baa ACM |
208 | |
209 | fail_unlock: | |
210 | spin_unlock(&head->lock); | |
211 | fail: | |
212 | local_bh_enable(); | |
213 | return ret; | |
214 | } | |
215 | ||
216 | EXPORT_SYMBOL_GPL(inet_csk_get_port); | |
217 | ||
218 | /* | |
219 | * Wait for an incoming connection, avoid race conditions. This must be called | |
220 | * with the socket locked. | |
221 | */ | |
222 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) | |
223 | { | |
224 | struct inet_connection_sock *icsk = inet_csk(sk); | |
225 | DEFINE_WAIT(wait); | |
226 | int err; | |
227 | ||
228 | /* | |
229 | * True wake-one mechanism for incoming connections: only | |
230 | * one process gets woken up, not the 'whole herd'. | |
231 | * Since we do not 'race & poll' for established sockets | |
232 | * anymore, the common case will execute the loop only once. | |
233 | * | |
234 | * Subtle issue: "add_wait_queue_exclusive()" will be added | |
235 | * after any current non-exclusive waiters, and we know that | |
236 | * it will always _stay_ after any new non-exclusive waiters | |
237 | * because all non-exclusive waiters are added at the | |
238 | * beginning of the wait-queue. As such, it's ok to "drop" | |
239 | * our exclusiveness temporarily when we get woken up without | |
240 | * having to remove and re-insert us on the wait queue. | |
241 | */ | |
242 | for (;;) { | |
243 | prepare_to_wait_exclusive(sk->sk_sleep, &wait, | |
244 | TASK_INTERRUPTIBLE); | |
245 | release_sock(sk); | |
246 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
247 | timeo = schedule_timeout(timeo); | |
248 | lock_sock(sk); | |
249 | err = 0; | |
250 | if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
251 | break; | |
252 | err = -EINVAL; | |
253 | if (sk->sk_state != TCP_LISTEN) | |
254 | break; | |
255 | err = sock_intr_errno(timeo); | |
256 | if (signal_pending(current)) | |
257 | break; | |
258 | err = -EAGAIN; | |
259 | if (!timeo) | |
260 | break; | |
261 | } | |
262 | finish_wait(sk->sk_sleep, &wait); | |
263 | return err; | |
264 | } | |
265 | ||
266 | /* | |
267 | * This will accept the next outstanding connection. | |
268 | */ | |
269 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) | |
270 | { | |
271 | struct inet_connection_sock *icsk = inet_csk(sk); | |
272 | struct sock *newsk; | |
273 | int error; | |
274 | ||
275 | lock_sock(sk); | |
276 | ||
277 | /* We need to make sure that this socket is listening, | |
278 | * and that it has something pending. | |
279 | */ | |
280 | error = -EINVAL; | |
281 | if (sk->sk_state != TCP_LISTEN) | |
282 | goto out_err; | |
283 | ||
284 | /* Find already established connection */ | |
285 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) { | |
286 | long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); | |
287 | ||
288 | /* If this is a non blocking socket don't sleep */ | |
289 | error = -EAGAIN; | |
290 | if (!timeo) | |
291 | goto out_err; | |
292 | ||
293 | error = inet_csk_wait_for_connect(sk, timeo); | |
294 | if (error) | |
295 | goto out_err; | |
296 | } | |
297 | ||
298 | newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk); | |
547b792c | 299 | WARN_ON(newsk->sk_state == TCP_SYN_RECV); |
3f421baa ACM |
300 | out: |
301 | release_sock(sk); | |
302 | return newsk; | |
303 | out_err: | |
304 | newsk = NULL; | |
305 | *err = error; | |
306 | goto out; | |
307 | } | |
308 | ||
309 | EXPORT_SYMBOL(inet_csk_accept); | |
310 | ||
311 | /* | |
312 | * Using different timers for retransmit, delayed acks and probes | |
e905a9ed | 313 | * We may wish use just one timer maintaining a list of expire jiffies |
3f421baa ACM |
314 | * to optimize. |
315 | */ | |
316 | void inet_csk_init_xmit_timers(struct sock *sk, | |
317 | void (*retransmit_handler)(unsigned long), | |
318 | void (*delack_handler)(unsigned long), | |
319 | void (*keepalive_handler)(unsigned long)) | |
320 | { | |
321 | struct inet_connection_sock *icsk = inet_csk(sk); | |
322 | ||
b24b8a24 PE |
323 | setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, |
324 | (unsigned long)sk); | |
325 | setup_timer(&icsk->icsk_delack_timer, delack_handler, | |
326 | (unsigned long)sk); | |
327 | setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); | |
3f421baa ACM |
328 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
329 | } | |
330 | ||
331 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); | |
332 | ||
333 | void inet_csk_clear_xmit_timers(struct sock *sk) | |
334 | { | |
335 | struct inet_connection_sock *icsk = inet_csk(sk); | |
336 | ||
337 | icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; | |
338 | ||
339 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
340 | sk_stop_timer(sk, &icsk->icsk_delack_timer); | |
341 | sk_stop_timer(sk, &sk->sk_timer); | |
342 | } | |
343 | ||
344 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); | |
345 | ||
346 | void inet_csk_delete_keepalive_timer(struct sock *sk) | |
347 | { | |
348 | sk_stop_timer(sk, &sk->sk_timer); | |
349 | } | |
350 | ||
351 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); | |
352 | ||
353 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) | |
354 | { | |
355 | sk_reset_timer(sk, &sk->sk_timer, jiffies + len); | |
356 | } | |
357 | ||
358 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); | |
359 | ||
d9319100 | 360 | struct dst_entry *inet_csk_route_req(struct sock *sk, |
3f421baa ACM |
361 | const struct request_sock *req) |
362 | { | |
363 | struct rtable *rt; | |
364 | const struct inet_request_sock *ireq = inet_rsk(req); | |
365 | struct ip_options *opt = inet_rsk(req)->opt; | |
366 | struct flowi fl = { .oif = sk->sk_bound_dev_if, | |
ffce9082 | 367 | .mark = sk->sk_mark, |
3f421baa ACM |
368 | .nl_u = { .ip4_u = |
369 | { .daddr = ((opt && opt->srr) ? | |
370 | opt->faddr : | |
371 | ireq->rmt_addr), | |
372 | .saddr = ireq->loc_addr, | |
373 | .tos = RT_CONN_FLAGS(sk) } }, | |
374 | .proto = sk->sk_protocol, | |
86b08d86 | 375 | .flags = inet_sk_flowi_flags(sk), |
3f421baa | 376 | .uli_u = { .ports = |
c720c7e8 | 377 | { .sport = inet_sk(sk)->inet_sport, |
3f421baa | 378 | .dport = ireq->rmt_port } } }; |
84a3aa00 | 379 | struct net *net = sock_net(sk); |
3f421baa | 380 | |
4237c75c | 381 | security_req_classify_flow(req, &fl); |
857a6e0a IJ |
382 | if (ip_route_output_flow(net, &rt, &fl, sk, 0)) |
383 | goto no_route; | |
384 | if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) | |
385 | goto route_err; | |
3f421baa | 386 | return &rt->u.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 ACM |
393 | } |
394 | ||
395 | EXPORT_SYMBOL_GPL(inet_csk_route_req); | |
396 | ||
6b72977b | 397 | static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport, |
72a3effa | 398 | const u32 rnd, const u32 synq_hsize) |
3f421baa | 399 | { |
6b72977b | 400 | return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1); |
3f421baa ACM |
401 | } |
402 | ||
403 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
404 | #define AF_INET_FAMILY(fam) ((fam) == AF_INET) | |
405 | #else | |
406 | #define AF_INET_FAMILY(fam) 1 | |
407 | #endif | |
408 | ||
409 | struct request_sock *inet_csk_search_req(const struct sock *sk, | |
410 | struct request_sock ***prevp, | |
6b72977b | 411 | const __be16 rport, const __be32 raddr, |
7f25afbb | 412 | const __be32 laddr) |
3f421baa ACM |
413 | { |
414 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
415 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
416 | struct request_sock *req, **prev; | |
417 | ||
418 | for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd, | |
419 | lopt->nr_table_entries)]; | |
420 | (req = *prev) != NULL; | |
421 | prev = &req->dl_next) { | |
422 | const struct inet_request_sock *ireq = inet_rsk(req); | |
423 | ||
424 | if (ireq->rmt_port == rport && | |
425 | ireq->rmt_addr == raddr && | |
426 | ireq->loc_addr == laddr && | |
427 | AF_INET_FAMILY(req->rsk_ops->family)) { | |
547b792c | 428 | WARN_ON(req->sk); |
3f421baa ACM |
429 | *prevp = prev; |
430 | break; | |
431 | } | |
432 | } | |
433 | ||
434 | return req; | |
435 | } | |
436 | ||
437 | EXPORT_SYMBOL_GPL(inet_csk_search_req); | |
438 | ||
439 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, | |
c2977c22 | 440 | unsigned long timeout) |
3f421baa ACM |
441 | { |
442 | struct inet_connection_sock *icsk = inet_csk(sk); | |
443 | struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; | |
444 | const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port, | |
445 | lopt->hash_rnd, lopt->nr_table_entries); | |
446 | ||
447 | reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout); | |
448 | inet_csk_reqsk_queue_added(sk, timeout); | |
449 | } | |
450 | ||
a019d6fe ACM |
451 | /* Only thing we need from tcp.h */ |
452 | extern int sysctl_tcp_synack_retries; | |
453 | ||
3f421baa | 454 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
9f1d2604 | 455 | |
0c3d79bc JA |
456 | /* Decide when to expire the request and when to resend SYN-ACK */ |
457 | static inline void syn_ack_recalc(struct request_sock *req, const int thresh, | |
458 | const int max_retries, | |
459 | const u8 rskq_defer_accept, | |
460 | int *expire, int *resend) | |
461 | { | |
462 | if (!rskq_defer_accept) { | |
463 | *expire = req->retrans >= thresh; | |
464 | *resend = 1; | |
465 | return; | |
466 | } | |
467 | *expire = req->retrans >= thresh && | |
468 | (!inet_rsk(req)->acked || req->retrans >= max_retries); | |
469 | /* | |
470 | * Do not resend while waiting for data after ACK, | |
471 | * start to resend on end of deferring period to give | |
472 | * last chance for data or ACK to create established socket. | |
473 | */ | |
474 | *resend = !inet_rsk(req)->acked || | |
475 | req->retrans >= rskq_defer_accept - 1; | |
476 | } | |
477 | ||
a019d6fe ACM |
478 | void inet_csk_reqsk_queue_prune(struct sock *parent, |
479 | const unsigned long interval, | |
480 | const unsigned long timeout, | |
481 | const unsigned long max_rto) | |
482 | { | |
483 | struct inet_connection_sock *icsk = inet_csk(parent); | |
484 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; | |
485 | struct listen_sock *lopt = queue->listen_opt; | |
ec0a1966 DM |
486 | int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries; |
487 | int thresh = max_retries; | |
a019d6fe ACM |
488 | unsigned long now = jiffies; |
489 | struct request_sock **reqp, *req; | |
490 | int i, budget; | |
491 | ||
492 | if (lopt == NULL || lopt->qlen == 0) | |
493 | return; | |
494 | ||
495 | /* Normally all the openreqs are young and become mature | |
496 | * (i.e. converted to established socket) for first timeout. | |
497 | * If synack was not acknowledged for 3 seconds, it means | |
498 | * one of the following things: synack was lost, ack was lost, | |
499 | * rtt is high or nobody planned to ack (i.e. synflood). | |
500 | * When server is a bit loaded, queue is populated with old | |
501 | * open requests, reducing effective size of queue. | |
502 | * When server is well loaded, queue size reduces to zero | |
503 | * after several minutes of work. It is not synflood, | |
504 | * it is normal operation. The solution is pruning | |
505 | * too old entries overriding normal timeout, when | |
506 | * situation becomes dangerous. | |
507 | * | |
508 | * Essentially, we reserve half of room for young | |
509 | * embrions; and abort old ones without pity, if old | |
510 | * ones are about to clog our table. | |
511 | */ | |
512 | if (lopt->qlen>>(lopt->max_qlen_log-1)) { | |
513 | int young = (lopt->qlen_young<<1); | |
514 | ||
515 | while (thresh > 2) { | |
516 | if (lopt->qlen < young) | |
517 | break; | |
518 | thresh--; | |
519 | young <<= 1; | |
520 | } | |
521 | } | |
522 | ||
ec0a1966 DM |
523 | if (queue->rskq_defer_accept) |
524 | max_retries = queue->rskq_defer_accept; | |
525 | ||
a019d6fe ACM |
526 | budget = 2 * (lopt->nr_table_entries / (timeout / interval)); |
527 | i = lopt->clock_hand; | |
528 | ||
529 | do { | |
530 | reqp=&lopt->syn_table[i]; | |
531 | while ((req = *reqp) != NULL) { | |
532 | if (time_after_eq(now, req->expires)) { | |
0c3d79bc JA |
533 | int expire = 0, resend = 0; |
534 | ||
535 | syn_ack_recalc(req, thresh, max_retries, | |
536 | queue->rskq_defer_accept, | |
537 | &expire, &resend); | |
72659ecc OP |
538 | if (req->rsk_ops->syn_ack_timeout) |
539 | req->rsk_ops->syn_ack_timeout(parent, req); | |
0c3d79bc JA |
540 | if (!expire && |
541 | (!resend || | |
e6b4d113 | 542 | !req->rsk_ops->rtx_syn_ack(parent, req, NULL) || |
0c3d79bc | 543 | inet_rsk(req)->acked)) { |
a019d6fe ACM |
544 | unsigned long timeo; |
545 | ||
546 | if (req->retrans++ == 0) | |
547 | lopt->qlen_young--; | |
548 | timeo = min((timeout << req->retrans), max_rto); | |
549 | req->expires = now + timeo; | |
550 | reqp = &req->dl_next; | |
551 | continue; | |
552 | } | |
553 | ||
554 | /* Drop this request */ | |
555 | inet_csk_reqsk_queue_unlink(parent, req, reqp); | |
556 | reqsk_queue_removed(queue, req); | |
557 | reqsk_free(req); | |
558 | continue; | |
559 | } | |
560 | reqp = &req->dl_next; | |
561 | } | |
562 | ||
563 | i = (i + 1) & (lopt->nr_table_entries - 1); | |
564 | ||
565 | } while (--budget > 0); | |
566 | ||
567 | lopt->clock_hand = i; | |
568 | ||
569 | if (lopt->qlen) | |
570 | inet_csk_reset_keepalive_timer(parent, interval); | |
571 | } | |
572 | ||
573 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune); | |
574 | ||
9f1d2604 | 575 | struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req, |
dd0fc66f | 576 | const gfp_t priority) |
9f1d2604 ACM |
577 | { |
578 | struct sock *newsk = sk_clone(sk, priority); | |
579 | ||
580 | if (newsk != NULL) { | |
581 | struct inet_connection_sock *newicsk = inet_csk(newsk); | |
582 | ||
583 | newsk->sk_state = TCP_SYN_RECV; | |
584 | newicsk->icsk_bind_hash = NULL; | |
585 | ||
c720c7e8 ED |
586 | inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port; |
587 | inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port); | |
588 | inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port; | |
9f1d2604 ACM |
589 | newsk->sk_write_space = sk_stream_write_space; |
590 | ||
591 | newicsk->icsk_retransmits = 0; | |
6687e988 ACM |
592 | newicsk->icsk_backoff = 0; |
593 | newicsk->icsk_probes_out = 0; | |
9f1d2604 ACM |
594 | |
595 | /* Deinitialize accept_queue to trap illegal accesses. */ | |
596 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); | |
4237c75c VY |
597 | |
598 | security_inet_csk_clone(newsk, req); | |
9f1d2604 ACM |
599 | } |
600 | return newsk; | |
601 | } | |
602 | ||
603 | EXPORT_SYMBOL_GPL(inet_csk_clone); | |
a019d6fe ACM |
604 | |
605 | /* | |
606 | * At this point, there should be no process reference to this | |
607 | * socket, and thus no user references at all. Therefore we | |
608 | * can assume the socket waitqueue is inactive and nobody will | |
609 | * try to jump onto it. | |
610 | */ | |
611 | void inet_csk_destroy_sock(struct sock *sk) | |
612 | { | |
547b792c IJ |
613 | WARN_ON(sk->sk_state != TCP_CLOSE); |
614 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); | |
a019d6fe ACM |
615 | |
616 | /* It cannot be in hash table! */ | |
547b792c | 617 | WARN_ON(!sk_unhashed(sk)); |
a019d6fe | 618 | |
c720c7e8 ED |
619 | /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ |
620 | WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); | |
a019d6fe ACM |
621 | |
622 | sk->sk_prot->destroy(sk); | |
623 | ||
624 | sk_stream_kill_queues(sk); | |
625 | ||
626 | xfrm_sk_free_policy(sk); | |
627 | ||
628 | sk_refcnt_debug_release(sk); | |
629 | ||
dd24c001 | 630 | percpu_counter_dec(sk->sk_prot->orphan_count); |
a019d6fe ACM |
631 | sock_put(sk); |
632 | } | |
633 | ||
634 | EXPORT_SYMBOL(inet_csk_destroy_sock); | |
635 | ||
636 | int inet_csk_listen_start(struct sock *sk, const int nr_table_entries) | |
637 | { | |
638 | struct inet_sock *inet = inet_sk(sk); | |
639 | struct inet_connection_sock *icsk = inet_csk(sk); | |
640 | int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries); | |
641 | ||
642 | if (rc != 0) | |
643 | return rc; | |
644 | ||
645 | sk->sk_max_ack_backlog = 0; | |
646 | sk->sk_ack_backlog = 0; | |
647 | inet_csk_delack_init(sk); | |
648 | ||
649 | /* There is race window here: we announce ourselves listening, | |
650 | * but this transition is still not validated by get_port(). | |
651 | * It is OK, because this socket enters to hash table only | |
652 | * after validation is complete. | |
653 | */ | |
654 | sk->sk_state = TCP_LISTEN; | |
c720c7e8 ED |
655 | if (!sk->sk_prot->get_port(sk, inet->inet_num)) { |
656 | inet->inet_sport = htons(inet->inet_num); | |
a019d6fe ACM |
657 | |
658 | sk_dst_reset(sk); | |
659 | sk->sk_prot->hash(sk); | |
660 | ||
661 | return 0; | |
662 | } | |
663 | ||
664 | sk->sk_state = TCP_CLOSE; | |
665 | __reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
666 | return -EADDRINUSE; | |
667 | } | |
668 | ||
669 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); | |
670 | ||
671 | /* | |
672 | * This routine closes sockets which have been at least partially | |
673 | * opened, but not yet accepted. | |
674 | */ | |
675 | void inet_csk_listen_stop(struct sock *sk) | |
676 | { | |
677 | struct inet_connection_sock *icsk = inet_csk(sk); | |
678 | struct request_sock *acc_req; | |
679 | struct request_sock *req; | |
680 | ||
681 | inet_csk_delete_keepalive_timer(sk); | |
682 | ||
683 | /* make all the listen_opt local to us */ | |
684 | acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue); | |
685 | ||
686 | /* Following specs, it would be better either to send FIN | |
687 | * (and enter FIN-WAIT-1, it is normal close) | |
688 | * or to send active reset (abort). | |
689 | * Certainly, it is pretty dangerous while synflood, but it is | |
690 | * bad justification for our negligence 8) | |
691 | * To be honest, we are not able to make either | |
692 | * of the variants now. --ANK | |
693 | */ | |
694 | reqsk_queue_destroy(&icsk->icsk_accept_queue); | |
695 | ||
696 | while ((req = acc_req) != NULL) { | |
697 | struct sock *child = req->sk; | |
698 | ||
699 | acc_req = req->dl_next; | |
700 | ||
701 | local_bh_disable(); | |
702 | bh_lock_sock(child); | |
547b792c | 703 | WARN_ON(sock_owned_by_user(child)); |
a019d6fe ACM |
704 | sock_hold(child); |
705 | ||
706 | sk->sk_prot->disconnect(child, O_NONBLOCK); | |
707 | ||
708 | sock_orphan(child); | |
709 | ||
eb4dea58 HX |
710 | percpu_counter_inc(sk->sk_prot->orphan_count); |
711 | ||
a019d6fe ACM |
712 | inet_csk_destroy_sock(child); |
713 | ||
714 | bh_unlock_sock(child); | |
715 | local_bh_enable(); | |
716 | sock_put(child); | |
717 | ||
718 | sk_acceptq_removed(sk); | |
719 | __reqsk_free(req); | |
720 | } | |
547b792c | 721 | WARN_ON(sk->sk_ack_backlog); |
a019d6fe ACM |
722 | } |
723 | ||
724 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); | |
af05dc93 ACM |
725 | |
726 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) | |
727 | { | |
728 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; | |
729 | const struct inet_sock *inet = inet_sk(sk); | |
730 | ||
731 | sin->sin_family = AF_INET; | |
c720c7e8 ED |
732 | sin->sin_addr.s_addr = inet->inet_daddr; |
733 | sin->sin_port = inet->inet_dport; | |
af05dc93 ACM |
734 | } |
735 | ||
736 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); | |
c4d93909 | 737 | |
dec73ff0 ACM |
738 | #ifdef CONFIG_COMPAT |
739 | int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, | |
740 | char __user *optval, int __user *optlen) | |
741 | { | |
dbeff12b | 742 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
743 | |
744 | if (icsk->icsk_af_ops->compat_getsockopt != NULL) | |
745 | return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, | |
746 | optval, optlen); | |
747 | return icsk->icsk_af_ops->getsockopt(sk, level, optname, | |
748 | optval, optlen); | |
749 | } | |
750 | ||
751 | EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); | |
752 | ||
753 | int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, | |
b7058842 | 754 | char __user *optval, unsigned int optlen) |
dec73ff0 | 755 | { |
dbeff12b | 756 | const struct inet_connection_sock *icsk = inet_csk(sk); |
dec73ff0 ACM |
757 | |
758 | if (icsk->icsk_af_ops->compat_setsockopt != NULL) | |
759 | return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, | |
760 | optval, optlen); | |
761 | return icsk->icsk_af_ops->setsockopt(sk, level, optname, | |
762 | optval, optlen); | |
763 | } | |
764 | ||
765 | EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); | |
766 | #endif |