Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * Implementation of the Transmission Control Protocol(TCP). | |
7 | * | |
8 | * Version: $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $ | |
9 | * | |
10 | * IPv4 specific functions | |
11 | * | |
12 | * | |
13 | * code split from: | |
14 | * linux/ipv4/tcp.c | |
15 | * linux/ipv4/tcp_input.c | |
16 | * linux/ipv4/tcp_output.c | |
17 | * | |
18 | * See tcp.c for author information | |
19 | * | |
20 | * This program is free software; you can redistribute it and/or | |
21 | * modify it under the terms of the GNU General Public License | |
22 | * as published by the Free Software Foundation; either version | |
23 | * 2 of the License, or (at your option) any later version. | |
24 | */ | |
25 | ||
26 | /* | |
27 | * Changes: | |
28 | * David S. Miller : New socket lookup architecture. | |
29 | * This code is dedicated to John Dyson. | |
30 | * David S. Miller : Change semantics of established hash, | |
31 | * half is devoted to TIME_WAIT sockets | |
32 | * and the rest go in the other half. | |
33 | * Andi Kleen : Add support for syncookies and fixed | |
34 | * some bugs: ip options weren't passed to | |
35 | * the TCP layer, missed a check for an | |
36 | * ACK bit. | |
37 | * Andi Kleen : Implemented fast path mtu discovery. | |
38 | * Fixed many serious bugs in the | |
39 | * open_request handling and moved | |
40 | * most of it into the af independent code. | |
41 | * Added tail drop and some other bugfixes. | |
42 | * Added new listen sematics. | |
43 | * Mike McLagan : Routing by source | |
44 | * Juan Jose Ciarlante: ip_dynaddr bits | |
45 | * Andi Kleen: various fixes. | |
46 | * Vitaly E. Lavrov : Transparent proxy revived after year | |
47 | * coma. | |
48 | * Andi Kleen : Fix new listen. | |
49 | * Andi Kleen : Fix accept error reporting. | |
50 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
51 | * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind | |
52 | * a single port at the same time. | |
53 | */ | |
54 | ||
55 | #include <linux/config.h> | |
56 | ||
57 | #include <linux/types.h> | |
58 | #include <linux/fcntl.h> | |
59 | #include <linux/module.h> | |
60 | #include <linux/random.h> | |
61 | #include <linux/cache.h> | |
62 | #include <linux/jhash.h> | |
63 | #include <linux/init.h> | |
64 | #include <linux/times.h> | |
65 | ||
66 | #include <net/icmp.h> | |
67 | #include <net/tcp.h> | |
68 | #include <net/ipv6.h> | |
69 | #include <net/inet_common.h> | |
70 | #include <net/xfrm.h> | |
71 | ||
72 | #include <linux/inet.h> | |
73 | #include <linux/ipv6.h> | |
74 | #include <linux/stddef.h> | |
75 | #include <linux/proc_fs.h> | |
76 | #include <linux/seq_file.h> | |
77 | ||
78 | extern int sysctl_ip_dynaddr; | |
79 | int sysctl_tcp_tw_reuse; | |
80 | int sysctl_tcp_low_latency; | |
81 | ||
82 | /* Check TCP sequence numbers in ICMP packets. */ | |
83 | #define ICMP_MIN_LENGTH 8 | |
84 | ||
85 | /* Socket used for sending RSTs */ | |
86 | static struct socket *tcp_socket; | |
87 | ||
88 | void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len, | |
89 | struct sk_buff *skb); | |
90 | ||
91 | struct tcp_hashinfo __cacheline_aligned tcp_hashinfo = { | |
92 | .__tcp_lhash_lock = RW_LOCK_UNLOCKED, | |
93 | .__tcp_lhash_users = ATOMIC_INIT(0), | |
94 | .__tcp_lhash_wait | |
95 | = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.__tcp_lhash_wait), | |
96 | .__tcp_portalloc_lock = SPIN_LOCK_UNLOCKED | |
97 | }; | |
98 | ||
99 | /* | |
100 | * This array holds the first and last local port number. | |
101 | * For high-usage systems, use sysctl to change this to | |
102 | * 32768-61000 | |
103 | */ | |
104 | int sysctl_local_port_range[2] = { 1024, 4999 }; | |
105 | int tcp_port_rover = 1024 - 1; | |
106 | ||
107 | static __inline__ int tcp_hashfn(__u32 laddr, __u16 lport, | |
108 | __u32 faddr, __u16 fport) | |
109 | { | |
110 | int h = (laddr ^ lport) ^ (faddr ^ fport); | |
111 | h ^= h >> 16; | |
112 | h ^= h >> 8; | |
113 | return h & (tcp_ehash_size - 1); | |
114 | } | |
115 | ||
116 | static __inline__ int tcp_sk_hashfn(struct sock *sk) | |
117 | { | |
118 | struct inet_sock *inet = inet_sk(sk); | |
119 | __u32 laddr = inet->rcv_saddr; | |
120 | __u16 lport = inet->num; | |
121 | __u32 faddr = inet->daddr; | |
122 | __u16 fport = inet->dport; | |
123 | ||
124 | return tcp_hashfn(laddr, lport, faddr, fport); | |
125 | } | |
126 | ||
127 | /* Allocate and initialize a new TCP local port bind bucket. | |
128 | * The bindhash mutex for snum's hash chain must be held here. | |
129 | */ | |
130 | struct tcp_bind_bucket *tcp_bucket_create(struct tcp_bind_hashbucket *head, | |
131 | unsigned short snum) | |
132 | { | |
133 | struct tcp_bind_bucket *tb = kmem_cache_alloc(tcp_bucket_cachep, | |
134 | SLAB_ATOMIC); | |
135 | if (tb) { | |
136 | tb->port = snum; | |
137 | tb->fastreuse = 0; | |
138 | INIT_HLIST_HEAD(&tb->owners); | |
139 | hlist_add_head(&tb->node, &head->chain); | |
140 | } | |
141 | return tb; | |
142 | } | |
143 | ||
144 | /* Caller must hold hashbucket lock for this tb with local BH disabled */ | |
145 | void tcp_bucket_destroy(struct tcp_bind_bucket *tb) | |
146 | { | |
147 | if (hlist_empty(&tb->owners)) { | |
148 | __hlist_del(&tb->node); | |
149 | kmem_cache_free(tcp_bucket_cachep, tb); | |
150 | } | |
151 | } | |
152 | ||
153 | /* Caller must disable local BH processing. */ | |
154 | static __inline__ void __tcp_inherit_port(struct sock *sk, struct sock *child) | |
155 | { | |
156 | struct tcp_bind_hashbucket *head = | |
157 | &tcp_bhash[tcp_bhashfn(inet_sk(child)->num)]; | |
158 | struct tcp_bind_bucket *tb; | |
159 | ||
160 | spin_lock(&head->lock); | |
161 | tb = tcp_sk(sk)->bind_hash; | |
162 | sk_add_bind_node(child, &tb->owners); | |
163 | tcp_sk(child)->bind_hash = tb; | |
164 | spin_unlock(&head->lock); | |
165 | } | |
166 | ||
167 | inline void tcp_inherit_port(struct sock *sk, struct sock *child) | |
168 | { | |
169 | local_bh_disable(); | |
170 | __tcp_inherit_port(sk, child); | |
171 | local_bh_enable(); | |
172 | } | |
173 | ||
174 | void tcp_bind_hash(struct sock *sk, struct tcp_bind_bucket *tb, | |
175 | unsigned short snum) | |
176 | { | |
177 | inet_sk(sk)->num = snum; | |
178 | sk_add_bind_node(sk, &tb->owners); | |
179 | tcp_sk(sk)->bind_hash = tb; | |
180 | } | |
181 | ||
182 | static inline int tcp_bind_conflict(struct sock *sk, struct tcp_bind_bucket *tb) | |
183 | { | |
184 | const u32 sk_rcv_saddr = tcp_v4_rcv_saddr(sk); | |
185 | struct sock *sk2; | |
186 | struct hlist_node *node; | |
187 | int reuse = sk->sk_reuse; | |
188 | ||
189 | sk_for_each_bound(sk2, node, &tb->owners) { | |
190 | if (sk != sk2 && | |
191 | !tcp_v6_ipv6only(sk2) && | |
192 | (!sk->sk_bound_dev_if || | |
193 | !sk2->sk_bound_dev_if || | |
194 | sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { | |
195 | if (!reuse || !sk2->sk_reuse || | |
196 | sk2->sk_state == TCP_LISTEN) { | |
197 | const u32 sk2_rcv_saddr = tcp_v4_rcv_saddr(sk2); | |
198 | if (!sk2_rcv_saddr || !sk_rcv_saddr || | |
199 | sk2_rcv_saddr == sk_rcv_saddr) | |
200 | break; | |
201 | } | |
202 | } | |
203 | } | |
204 | return node != NULL; | |
205 | } | |
206 | ||
207 | /* Obtain a reference to a local port for the given sock, | |
208 | * if snum is zero it means select any available local port. | |
209 | */ | |
210 | static int tcp_v4_get_port(struct sock *sk, unsigned short snum) | |
211 | { | |
212 | struct tcp_bind_hashbucket *head; | |
213 | struct hlist_node *node; | |
214 | struct tcp_bind_bucket *tb; | |
215 | int ret; | |
216 | ||
217 | local_bh_disable(); | |
218 | if (!snum) { | |
219 | int low = sysctl_local_port_range[0]; | |
220 | int high = sysctl_local_port_range[1]; | |
221 | int remaining = (high - low) + 1; | |
222 | int rover; | |
223 | ||
224 | spin_lock(&tcp_portalloc_lock); | |
225 | rover = tcp_port_rover; | |
226 | do { | |
227 | rover++; | |
228 | if (rover < low || rover > high) | |
229 | rover = low; | |
230 | head = &tcp_bhash[tcp_bhashfn(rover)]; | |
231 | spin_lock(&head->lock); | |
232 | tb_for_each(tb, node, &head->chain) | |
233 | if (tb->port == rover) | |
234 | goto next; | |
235 | break; | |
236 | next: | |
237 | spin_unlock(&head->lock); | |
238 | } while (--remaining > 0); | |
239 | tcp_port_rover = rover; | |
240 | spin_unlock(&tcp_portalloc_lock); | |
241 | ||
242 | /* Exhausted local port range during search? */ | |
243 | ret = 1; | |
244 | if (remaining <= 0) | |
245 | goto fail; | |
246 | ||
247 | /* OK, here is the one we will use. HEAD is | |
248 | * non-NULL and we hold it's mutex. | |
249 | */ | |
250 | snum = rover; | |
251 | } else { | |
252 | head = &tcp_bhash[tcp_bhashfn(snum)]; | |
253 | spin_lock(&head->lock); | |
254 | tb_for_each(tb, node, &head->chain) | |
255 | if (tb->port == snum) | |
256 | goto tb_found; | |
257 | } | |
258 | tb = NULL; | |
259 | goto tb_not_found; | |
260 | tb_found: | |
261 | if (!hlist_empty(&tb->owners)) { | |
262 | if (sk->sk_reuse > 1) | |
263 | goto success; | |
264 | if (tb->fastreuse > 0 && | |
265 | sk->sk_reuse && sk->sk_state != TCP_LISTEN) { | |
266 | goto success; | |
267 | } else { | |
268 | ret = 1; | |
269 | if (tcp_bind_conflict(sk, tb)) | |
270 | goto fail_unlock; | |
271 | } | |
272 | } | |
273 | tb_not_found: | |
274 | ret = 1; | |
275 | if (!tb && (tb = tcp_bucket_create(head, snum)) == NULL) | |
276 | goto fail_unlock; | |
277 | if (hlist_empty(&tb->owners)) { | |
278 | if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) | |
279 | tb->fastreuse = 1; | |
280 | else | |
281 | tb->fastreuse = 0; | |
282 | } else if (tb->fastreuse && | |
283 | (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) | |
284 | tb->fastreuse = 0; | |
285 | success: | |
286 | if (!tcp_sk(sk)->bind_hash) | |
287 | tcp_bind_hash(sk, tb, snum); | |
288 | BUG_TRAP(tcp_sk(sk)->bind_hash == tb); | |
289 | ret = 0; | |
290 | ||
291 | fail_unlock: | |
292 | spin_unlock(&head->lock); | |
293 | fail: | |
294 | local_bh_enable(); | |
295 | return ret; | |
296 | } | |
297 | ||
298 | /* Get rid of any references to a local port held by the | |
299 | * given sock. | |
300 | */ | |
301 | static void __tcp_put_port(struct sock *sk) | |
302 | { | |
303 | struct inet_sock *inet = inet_sk(sk); | |
304 | struct tcp_bind_hashbucket *head = &tcp_bhash[tcp_bhashfn(inet->num)]; | |
305 | struct tcp_bind_bucket *tb; | |
306 | ||
307 | spin_lock(&head->lock); | |
308 | tb = tcp_sk(sk)->bind_hash; | |
309 | __sk_del_bind_node(sk); | |
310 | tcp_sk(sk)->bind_hash = NULL; | |
311 | inet->num = 0; | |
312 | tcp_bucket_destroy(tb); | |
313 | spin_unlock(&head->lock); | |
314 | } | |
315 | ||
316 | void tcp_put_port(struct sock *sk) | |
317 | { | |
318 | local_bh_disable(); | |
319 | __tcp_put_port(sk); | |
320 | local_bh_enable(); | |
321 | } | |
322 | ||
323 | /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP. | |
324 | * Look, when several writers sleep and reader wakes them up, all but one | |
325 | * immediately hit write lock and grab all the cpus. Exclusive sleep solves | |
326 | * this, _but_ remember, it adds useless work on UP machines (wake up each | |
327 | * exclusive lock release). It should be ifdefed really. | |
328 | */ | |
329 | ||
330 | void tcp_listen_wlock(void) | |
331 | { | |
332 | write_lock(&tcp_lhash_lock); | |
333 | ||
334 | if (atomic_read(&tcp_lhash_users)) { | |
335 | DEFINE_WAIT(wait); | |
336 | ||
337 | for (;;) { | |
338 | prepare_to_wait_exclusive(&tcp_lhash_wait, | |
339 | &wait, TASK_UNINTERRUPTIBLE); | |
340 | if (!atomic_read(&tcp_lhash_users)) | |
341 | break; | |
342 | write_unlock_bh(&tcp_lhash_lock); | |
343 | schedule(); | |
344 | write_lock_bh(&tcp_lhash_lock); | |
345 | } | |
346 | ||
347 | finish_wait(&tcp_lhash_wait, &wait); | |
348 | } | |
349 | } | |
350 | ||
351 | static __inline__ void __tcp_v4_hash(struct sock *sk, const int listen_possible) | |
352 | { | |
353 | struct hlist_head *list; | |
354 | rwlock_t *lock; | |
355 | ||
356 | BUG_TRAP(sk_unhashed(sk)); | |
357 | if (listen_possible && sk->sk_state == TCP_LISTEN) { | |
358 | list = &tcp_listening_hash[tcp_sk_listen_hashfn(sk)]; | |
359 | lock = &tcp_lhash_lock; | |
360 | tcp_listen_wlock(); | |
361 | } else { | |
362 | list = &tcp_ehash[(sk->sk_hashent = tcp_sk_hashfn(sk))].chain; | |
363 | lock = &tcp_ehash[sk->sk_hashent].lock; | |
364 | write_lock(lock); | |
365 | } | |
366 | __sk_add_node(sk, list); | |
367 | sock_prot_inc_use(sk->sk_prot); | |
368 | write_unlock(lock); | |
369 | if (listen_possible && sk->sk_state == TCP_LISTEN) | |
370 | wake_up(&tcp_lhash_wait); | |
371 | } | |
372 | ||
373 | static void tcp_v4_hash(struct sock *sk) | |
374 | { | |
375 | if (sk->sk_state != TCP_CLOSE) { | |
376 | local_bh_disable(); | |
377 | __tcp_v4_hash(sk, 1); | |
378 | local_bh_enable(); | |
379 | } | |
380 | } | |
381 | ||
382 | void tcp_unhash(struct sock *sk) | |
383 | { | |
384 | rwlock_t *lock; | |
385 | ||
386 | if (sk_unhashed(sk)) | |
387 | goto ende; | |
388 | ||
389 | if (sk->sk_state == TCP_LISTEN) { | |
390 | local_bh_disable(); | |
391 | tcp_listen_wlock(); | |
392 | lock = &tcp_lhash_lock; | |
393 | } else { | |
394 | struct tcp_ehash_bucket *head = &tcp_ehash[sk->sk_hashent]; | |
395 | lock = &head->lock; | |
396 | write_lock_bh(&head->lock); | |
397 | } | |
398 | ||
399 | if (__sk_del_node_init(sk)) | |
400 | sock_prot_dec_use(sk->sk_prot); | |
401 | write_unlock_bh(lock); | |
402 | ||
403 | ende: | |
404 | if (sk->sk_state == TCP_LISTEN) | |
405 | wake_up(&tcp_lhash_wait); | |
406 | } | |
407 | ||
408 | /* Don't inline this cruft. Here are some nice properties to | |
409 | * exploit here. The BSD API does not allow a listening TCP | |
410 | * to specify the remote port nor the remote address for the | |
411 | * connection. So always assume those are both wildcarded | |
412 | * during the search since they can never be otherwise. | |
413 | */ | |
414 | static struct sock *__tcp_v4_lookup_listener(struct hlist_head *head, u32 daddr, | |
415 | unsigned short hnum, int dif) | |
416 | { | |
417 | struct sock *result = NULL, *sk; | |
418 | struct hlist_node *node; | |
419 | int score, hiscore; | |
420 | ||
421 | hiscore=-1; | |
422 | sk_for_each(sk, node, head) { | |
423 | struct inet_sock *inet = inet_sk(sk); | |
424 | ||
425 | if (inet->num == hnum && !ipv6_only_sock(sk)) { | |
426 | __u32 rcv_saddr = inet->rcv_saddr; | |
427 | ||
428 | score = (sk->sk_family == PF_INET ? 1 : 0); | |
429 | if (rcv_saddr) { | |
430 | if (rcv_saddr != daddr) | |
431 | continue; | |
432 | score+=2; | |
433 | } | |
434 | if (sk->sk_bound_dev_if) { | |
435 | if (sk->sk_bound_dev_if != dif) | |
436 | continue; | |
437 | score+=2; | |
438 | } | |
439 | if (score == 5) | |
440 | return sk; | |
441 | if (score > hiscore) { | |
442 | hiscore = score; | |
443 | result = sk; | |
444 | } | |
445 | } | |
446 | } | |
447 | return result; | |
448 | } | |
449 | ||
450 | /* Optimize the common listener case. */ | |
451 | static inline struct sock *tcp_v4_lookup_listener(u32 daddr, | |
452 | unsigned short hnum, int dif) | |
453 | { | |
454 | struct sock *sk = NULL; | |
455 | struct hlist_head *head; | |
456 | ||
457 | read_lock(&tcp_lhash_lock); | |
458 | head = &tcp_listening_hash[tcp_lhashfn(hnum)]; | |
459 | if (!hlist_empty(head)) { | |
460 | struct inet_sock *inet = inet_sk((sk = __sk_head(head))); | |
461 | ||
462 | if (inet->num == hnum && !sk->sk_node.next && | |
463 | (!inet->rcv_saddr || inet->rcv_saddr == daddr) && | |
464 | (sk->sk_family == PF_INET || !ipv6_only_sock(sk)) && | |
465 | !sk->sk_bound_dev_if) | |
466 | goto sherry_cache; | |
467 | sk = __tcp_v4_lookup_listener(head, daddr, hnum, dif); | |
468 | } | |
469 | if (sk) { | |
470 | sherry_cache: | |
471 | sock_hold(sk); | |
472 | } | |
473 | read_unlock(&tcp_lhash_lock); | |
474 | return sk; | |
475 | } | |
476 | ||
477 | /* Sockets in TCP_CLOSE state are _always_ taken out of the hash, so | |
478 | * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM | |
479 | * | |
480 | * Local BH must be disabled here. | |
481 | */ | |
482 | ||
483 | static inline struct sock *__tcp_v4_lookup_established(u32 saddr, u16 sport, | |
484 | u32 daddr, u16 hnum, | |
485 | int dif) | |
486 | { | |
487 | struct tcp_ehash_bucket *head; | |
488 | TCP_V4_ADDR_COOKIE(acookie, saddr, daddr) | |
489 | __u32 ports = TCP_COMBINED_PORTS(sport, hnum); | |
490 | struct sock *sk; | |
491 | struct hlist_node *node; | |
492 | /* Optimize here for direct hit, only listening connections can | |
493 | * have wildcards anyways. | |
494 | */ | |
495 | int hash = tcp_hashfn(daddr, hnum, saddr, sport); | |
496 | head = &tcp_ehash[hash]; | |
497 | read_lock(&head->lock); | |
498 | sk_for_each(sk, node, &head->chain) { | |
499 | if (TCP_IPV4_MATCH(sk, acookie, saddr, daddr, ports, dif)) | |
500 | goto hit; /* You sunk my battleship! */ | |
501 | } | |
502 | ||
503 | /* Must check for a TIME_WAIT'er before going to listener hash. */ | |
504 | sk_for_each(sk, node, &(head + tcp_ehash_size)->chain) { | |
505 | if (TCP_IPV4_TW_MATCH(sk, acookie, saddr, daddr, ports, dif)) | |
506 | goto hit; | |
507 | } | |
508 | sk = NULL; | |
509 | out: | |
510 | read_unlock(&head->lock); | |
511 | return sk; | |
512 | hit: | |
513 | sock_hold(sk); | |
514 | goto out; | |
515 | } | |
516 | ||
517 | static inline struct sock *__tcp_v4_lookup(u32 saddr, u16 sport, | |
518 | u32 daddr, u16 hnum, int dif) | |
519 | { | |
520 | struct sock *sk = __tcp_v4_lookup_established(saddr, sport, | |
521 | daddr, hnum, dif); | |
522 | ||
523 | return sk ? : tcp_v4_lookup_listener(daddr, hnum, dif); | |
524 | } | |
525 | ||
526 | inline struct sock *tcp_v4_lookup(u32 saddr, u16 sport, u32 daddr, | |
527 | u16 dport, int dif) | |
528 | { | |
529 | struct sock *sk; | |
530 | ||
531 | local_bh_disable(); | |
532 | sk = __tcp_v4_lookup(saddr, sport, daddr, ntohs(dport), dif); | |
533 | local_bh_enable(); | |
534 | ||
535 | return sk; | |
536 | } | |
537 | ||
538 | EXPORT_SYMBOL_GPL(tcp_v4_lookup); | |
539 | ||
540 | static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb) | |
541 | { | |
542 | return secure_tcp_sequence_number(skb->nh.iph->daddr, | |
543 | skb->nh.iph->saddr, | |
544 | skb->h.th->dest, | |
545 | skb->h.th->source); | |
546 | } | |
547 | ||
548 | /* called with local bh disabled */ | |
549 | static int __tcp_v4_check_established(struct sock *sk, __u16 lport, | |
550 | struct tcp_tw_bucket **twp) | |
551 | { | |
552 | struct inet_sock *inet = inet_sk(sk); | |
553 | u32 daddr = inet->rcv_saddr; | |
554 | u32 saddr = inet->daddr; | |
555 | int dif = sk->sk_bound_dev_if; | |
556 | TCP_V4_ADDR_COOKIE(acookie, saddr, daddr) | |
557 | __u32 ports = TCP_COMBINED_PORTS(inet->dport, lport); | |
558 | int hash = tcp_hashfn(daddr, lport, saddr, inet->dport); | |
559 | struct tcp_ehash_bucket *head = &tcp_ehash[hash]; | |
560 | struct sock *sk2; | |
561 | struct hlist_node *node; | |
562 | struct tcp_tw_bucket *tw; | |
563 | ||
564 | write_lock(&head->lock); | |
565 | ||
566 | /* Check TIME-WAIT sockets first. */ | |
567 | sk_for_each(sk2, node, &(head + tcp_ehash_size)->chain) { | |
568 | tw = (struct tcp_tw_bucket *)sk2; | |
569 | ||
570 | if (TCP_IPV4_TW_MATCH(sk2, acookie, saddr, daddr, ports, dif)) { | |
571 | struct tcp_sock *tp = tcp_sk(sk); | |
572 | ||
573 | /* With PAWS, it is safe from the viewpoint | |
574 | of data integrity. Even without PAWS it | |
575 | is safe provided sequence spaces do not | |
576 | overlap i.e. at data rates <= 80Mbit/sec. | |
577 | ||
578 | Actually, the idea is close to VJ's one, | |
579 | only timestamp cache is held not per host, | |
580 | but per port pair and TW bucket is used | |
581 | as state holder. | |
582 | ||
583 | If TW bucket has been already destroyed we | |
584 | fall back to VJ's scheme and use initial | |
585 | timestamp retrieved from peer table. | |
586 | */ | |
587 | if (tw->tw_ts_recent_stamp && | |
588 | (!twp || (sysctl_tcp_tw_reuse && | |
589 | xtime.tv_sec - | |
590 | tw->tw_ts_recent_stamp > 1))) { | |
591 | if ((tp->write_seq = | |
592 | tw->tw_snd_nxt + 65535 + 2) == 0) | |
593 | tp->write_seq = 1; | |
594 | tp->rx_opt.ts_recent = tw->tw_ts_recent; | |
595 | tp->rx_opt.ts_recent_stamp = tw->tw_ts_recent_stamp; | |
596 | sock_hold(sk2); | |
597 | goto unique; | |
598 | } else | |
599 | goto not_unique; | |
600 | } | |
601 | } | |
602 | tw = NULL; | |
603 | ||
604 | /* And established part... */ | |
605 | sk_for_each(sk2, node, &head->chain) { | |
606 | if (TCP_IPV4_MATCH(sk2, acookie, saddr, daddr, ports, dif)) | |
607 | goto not_unique; | |
608 | } | |
609 | ||
610 | unique: | |
611 | /* Must record num and sport now. Otherwise we will see | |
612 | * in hash table socket with a funny identity. */ | |
613 | inet->num = lport; | |
614 | inet->sport = htons(lport); | |
615 | sk->sk_hashent = hash; | |
616 | BUG_TRAP(sk_unhashed(sk)); | |
617 | __sk_add_node(sk, &head->chain); | |
618 | sock_prot_inc_use(sk->sk_prot); | |
619 | write_unlock(&head->lock); | |
620 | ||
621 | if (twp) { | |
622 | *twp = tw; | |
623 | NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED); | |
624 | } else if (tw) { | |
625 | /* Silly. Should hash-dance instead... */ | |
626 | tcp_tw_deschedule(tw); | |
627 | NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED); | |
628 | ||
629 | tcp_tw_put(tw); | |
630 | } | |
631 | ||
632 | return 0; | |
633 | ||
634 | not_unique: | |
635 | write_unlock(&head->lock); | |
636 | return -EADDRNOTAVAIL; | |
637 | } | |
638 | ||
639 | static inline u32 connect_port_offset(const struct sock *sk) | |
640 | { | |
641 | const struct inet_sock *inet = inet_sk(sk); | |
642 | ||
643 | return secure_tcp_port_ephemeral(inet->rcv_saddr, inet->daddr, | |
644 | inet->dport); | |
645 | } | |
646 | ||
647 | /* | |
648 | * Bind a port for a connect operation and hash it. | |
649 | */ | |
650 | static inline int tcp_v4_hash_connect(struct sock *sk) | |
651 | { | |
652 | unsigned short snum = inet_sk(sk)->num; | |
653 | struct tcp_bind_hashbucket *head; | |
654 | struct tcp_bind_bucket *tb; | |
655 | int ret; | |
656 | ||
657 | if (!snum) { | |
658 | int low = sysctl_local_port_range[0]; | |
659 | int high = sysctl_local_port_range[1]; | |
660 | int range = high - low; | |
661 | int i; | |
662 | int port; | |
663 | static u32 hint; | |
664 | u32 offset = hint + connect_port_offset(sk); | |
665 | struct hlist_node *node; | |
666 | struct tcp_tw_bucket *tw = NULL; | |
667 | ||
668 | local_bh_disable(); | |
669 | for (i = 1; i <= range; i++) { | |
670 | port = low + (i + offset) % range; | |
671 | head = &tcp_bhash[tcp_bhashfn(port)]; | |
672 | spin_lock(&head->lock); | |
673 | ||
674 | /* Does not bother with rcv_saddr checks, | |
675 | * because the established check is already | |
676 | * unique enough. | |
677 | */ | |
678 | tb_for_each(tb, node, &head->chain) { | |
679 | if (tb->port == port) { | |
680 | BUG_TRAP(!hlist_empty(&tb->owners)); | |
681 | if (tb->fastreuse >= 0) | |
682 | goto next_port; | |
683 | if (!__tcp_v4_check_established(sk, | |
684 | port, | |
685 | &tw)) | |
686 | goto ok; | |
687 | goto next_port; | |
688 | } | |
689 | } | |
690 | ||
691 | tb = tcp_bucket_create(head, port); | |
692 | if (!tb) { | |
693 | spin_unlock(&head->lock); | |
694 | break; | |
695 | } | |
696 | tb->fastreuse = -1; | |
697 | goto ok; | |
698 | ||
699 | next_port: | |
700 | spin_unlock(&head->lock); | |
701 | } | |
702 | local_bh_enable(); | |
703 | ||
704 | return -EADDRNOTAVAIL; | |
705 | ||
706 | ok: | |
707 | hint += i; | |
708 | ||
709 | /* Head lock still held and bh's disabled */ | |
710 | tcp_bind_hash(sk, tb, port); | |
711 | if (sk_unhashed(sk)) { | |
712 | inet_sk(sk)->sport = htons(port); | |
713 | __tcp_v4_hash(sk, 0); | |
714 | } | |
715 | spin_unlock(&head->lock); | |
716 | ||
717 | if (tw) { | |
718 | tcp_tw_deschedule(tw); | |
719 | tcp_tw_put(tw); | |
720 | } | |
721 | ||
722 | ret = 0; | |
723 | goto out; | |
724 | } | |
725 | ||
726 | head = &tcp_bhash[tcp_bhashfn(snum)]; | |
727 | tb = tcp_sk(sk)->bind_hash; | |
728 | spin_lock_bh(&head->lock); | |
729 | if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) { | |
730 | __tcp_v4_hash(sk, 0); | |
731 | spin_unlock_bh(&head->lock); | |
732 | return 0; | |
733 | } else { | |
734 | spin_unlock(&head->lock); | |
735 | /* No definite answer... Walk to established hash table */ | |
736 | ret = __tcp_v4_check_established(sk, snum, NULL); | |
737 | out: | |
738 | local_bh_enable(); | |
739 | return ret; | |
740 | } | |
741 | } | |
742 | ||
743 | /* This will initiate an outgoing connection. */ | |
744 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
745 | { | |
746 | struct inet_sock *inet = inet_sk(sk); | |
747 | struct tcp_sock *tp = tcp_sk(sk); | |
748 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; | |
749 | struct rtable *rt; | |
750 | u32 daddr, nexthop; | |
751 | int tmp; | |
752 | int err; | |
753 | ||
754 | if (addr_len < sizeof(struct sockaddr_in)) | |
755 | return -EINVAL; | |
756 | ||
757 | if (usin->sin_family != AF_INET) | |
758 | return -EAFNOSUPPORT; | |
759 | ||
760 | nexthop = daddr = usin->sin_addr.s_addr; | |
761 | if (inet->opt && inet->opt->srr) { | |
762 | if (!daddr) | |
763 | return -EINVAL; | |
764 | nexthop = inet->opt->faddr; | |
765 | } | |
766 | ||
767 | tmp = ip_route_connect(&rt, nexthop, inet->saddr, | |
768 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, | |
769 | IPPROTO_TCP, | |
770 | inet->sport, usin->sin_port, sk); | |
771 | if (tmp < 0) | |
772 | return tmp; | |
773 | ||
774 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
775 | ip_rt_put(rt); | |
776 | return -ENETUNREACH; | |
777 | } | |
778 | ||
779 | if (!inet->opt || !inet->opt->srr) | |
780 | daddr = rt->rt_dst; | |
781 | ||
782 | if (!inet->saddr) | |
783 | inet->saddr = rt->rt_src; | |
784 | inet->rcv_saddr = inet->saddr; | |
785 | ||
786 | if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) { | |
787 | /* Reset inherited state */ | |
788 | tp->rx_opt.ts_recent = 0; | |
789 | tp->rx_opt.ts_recent_stamp = 0; | |
790 | tp->write_seq = 0; | |
791 | } | |
792 | ||
793 | if (sysctl_tcp_tw_recycle && | |
794 | !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) { | |
795 | struct inet_peer *peer = rt_get_peer(rt); | |
796 | ||
797 | /* VJ's idea. We save last timestamp seen from | |
798 | * the destination in peer table, when entering state TIME-WAIT | |
799 | * and initialize rx_opt.ts_recent from it, when trying new connection. | |
800 | */ | |
801 | ||
802 | if (peer && peer->tcp_ts_stamp + TCP_PAWS_MSL >= xtime.tv_sec) { | |
803 | tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp; | |
804 | tp->rx_opt.ts_recent = peer->tcp_ts; | |
805 | } | |
806 | } | |
807 | ||
808 | inet->dport = usin->sin_port; | |
809 | inet->daddr = daddr; | |
810 | ||
811 | tp->ext_header_len = 0; | |
812 | if (inet->opt) | |
813 | tp->ext_header_len = inet->opt->optlen; | |
814 | ||
815 | tp->rx_opt.mss_clamp = 536; | |
816 | ||
817 | /* Socket identity is still unknown (sport may be zero). | |
818 | * However we set state to SYN-SENT and not releasing socket | |
819 | * lock select source port, enter ourselves into the hash tables and | |
820 | * complete initialization after this. | |
821 | */ | |
822 | tcp_set_state(sk, TCP_SYN_SENT); | |
823 | err = tcp_v4_hash_connect(sk); | |
824 | if (err) | |
825 | goto failure; | |
826 | ||
827 | err = ip_route_newports(&rt, inet->sport, inet->dport, sk); | |
828 | if (err) | |
829 | goto failure; | |
830 | ||
831 | /* OK, now commit destination to socket. */ | |
832 | __sk_dst_set(sk, &rt->u.dst); | |
833 | tcp_v4_setup_caps(sk, &rt->u.dst); | |
834 | ||
835 | if (!tp->write_seq) | |
836 | tp->write_seq = secure_tcp_sequence_number(inet->saddr, | |
837 | inet->daddr, | |
838 | inet->sport, | |
839 | usin->sin_port); | |
840 | ||
841 | inet->id = tp->write_seq ^ jiffies; | |
842 | ||
843 | err = tcp_connect(sk); | |
844 | rt = NULL; | |
845 | if (err) | |
846 | goto failure; | |
847 | ||
848 | return 0; | |
849 | ||
850 | failure: | |
851 | /* This unhashes the socket and releases the local port, if necessary. */ | |
852 | tcp_set_state(sk, TCP_CLOSE); | |
853 | ip_rt_put(rt); | |
854 | sk->sk_route_caps = 0; | |
855 | inet->dport = 0; | |
856 | return err; | |
857 | } | |
858 | ||
859 | static __inline__ int tcp_v4_iif(struct sk_buff *skb) | |
860 | { | |
861 | return ((struct rtable *)skb->dst)->rt_iif; | |
862 | } | |
863 | ||
864 | static __inline__ u32 tcp_v4_synq_hash(u32 raddr, u16 rport, u32 rnd) | |
865 | { | |
866 | return (jhash_2words(raddr, (u32) rport, rnd) & (TCP_SYNQ_HSIZE - 1)); | |
867 | } | |
868 | ||
869 | static struct open_request *tcp_v4_search_req(struct tcp_sock *tp, | |
870 | struct open_request ***prevp, | |
871 | __u16 rport, | |
872 | __u32 raddr, __u32 laddr) | |
873 | { | |
874 | struct tcp_listen_opt *lopt = tp->listen_opt; | |
875 | struct open_request *req, **prev; | |
876 | ||
877 | for (prev = &lopt->syn_table[tcp_v4_synq_hash(raddr, rport, lopt->hash_rnd)]; | |
878 | (req = *prev) != NULL; | |
879 | prev = &req->dl_next) { | |
880 | if (req->rmt_port == rport && | |
881 | req->af.v4_req.rmt_addr == raddr && | |
882 | req->af.v4_req.loc_addr == laddr && | |
883 | TCP_INET_FAMILY(req->class->family)) { | |
884 | BUG_TRAP(!req->sk); | |
885 | *prevp = prev; | |
886 | break; | |
887 | } | |
888 | } | |
889 | ||
890 | return req; | |
891 | } | |
892 | ||
893 | static void tcp_v4_synq_add(struct sock *sk, struct open_request *req) | |
894 | { | |
895 | struct tcp_sock *tp = tcp_sk(sk); | |
896 | struct tcp_listen_opt *lopt = tp->listen_opt; | |
897 | u32 h = tcp_v4_synq_hash(req->af.v4_req.rmt_addr, req->rmt_port, lopt->hash_rnd); | |
898 | ||
899 | req->expires = jiffies + TCP_TIMEOUT_INIT; | |
900 | req->retrans = 0; | |
901 | req->sk = NULL; | |
902 | req->dl_next = lopt->syn_table[h]; | |
903 | ||
904 | write_lock(&tp->syn_wait_lock); | |
905 | lopt->syn_table[h] = req; | |
906 | write_unlock(&tp->syn_wait_lock); | |
907 | ||
908 | tcp_synq_added(sk); | |
909 | } | |
910 | ||
911 | ||
912 | /* | |
913 | * This routine does path mtu discovery as defined in RFC1191. | |
914 | */ | |
915 | static inline void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, | |
916 | u32 mtu) | |
917 | { | |
918 | struct dst_entry *dst; | |
919 | struct inet_sock *inet = inet_sk(sk); | |
920 | struct tcp_sock *tp = tcp_sk(sk); | |
921 | ||
922 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs | |
923 | * send out by Linux are always <576bytes so they should go through | |
924 | * unfragmented). | |
925 | */ | |
926 | if (sk->sk_state == TCP_LISTEN) | |
927 | return; | |
928 | ||
929 | /* We don't check in the destentry if pmtu discovery is forbidden | |
930 | * on this route. We just assume that no packet_to_big packets | |
931 | * are send back when pmtu discovery is not active. | |
932 | * There is a small race when the user changes this flag in the | |
933 | * route, but I think that's acceptable. | |
934 | */ | |
935 | if ((dst = __sk_dst_check(sk, 0)) == NULL) | |
936 | return; | |
937 | ||
938 | dst->ops->update_pmtu(dst, mtu); | |
939 | ||
940 | /* Something is about to be wrong... Remember soft error | |
941 | * for the case, if this connection will not able to recover. | |
942 | */ | |
943 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
944 | sk->sk_err_soft = EMSGSIZE; | |
945 | ||
946 | mtu = dst_mtu(dst); | |
947 | ||
948 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
949 | tp->pmtu_cookie > mtu) { | |
950 | tcp_sync_mss(sk, mtu); | |
951 | ||
952 | /* Resend the TCP packet because it's | |
953 | * clear that the old packet has been | |
954 | * dropped. This is the new "fast" path mtu | |
955 | * discovery. | |
956 | */ | |
957 | tcp_simple_retransmit(sk); | |
958 | } /* else let the usual retransmit timer handle it */ | |
959 | } | |
960 | ||
961 | /* | |
962 | * This routine is called by the ICMP module when it gets some | |
963 | * sort of error condition. If err < 0 then the socket should | |
964 | * be closed and the error returned to the user. If err > 0 | |
965 | * it's just the icmp type << 8 | icmp code. After adjustment | |
966 | * header points to the first 8 bytes of the tcp header. We need | |
967 | * to find the appropriate port. | |
968 | * | |
969 | * The locking strategy used here is very "optimistic". When | |
970 | * someone else accesses the socket the ICMP is just dropped | |
971 | * and for some paths there is no check at all. | |
972 | * A more general error queue to queue errors for later handling | |
973 | * is probably better. | |
974 | * | |
975 | */ | |
976 | ||
977 | void tcp_v4_err(struct sk_buff *skb, u32 info) | |
978 | { | |
979 | struct iphdr *iph = (struct iphdr *)skb->data; | |
980 | struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2)); | |
981 | struct tcp_sock *tp; | |
982 | struct inet_sock *inet; | |
983 | int type = skb->h.icmph->type; | |
984 | int code = skb->h.icmph->code; | |
985 | struct sock *sk; | |
986 | __u32 seq; | |
987 | int err; | |
988 | ||
989 | if (skb->len < (iph->ihl << 2) + 8) { | |
990 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
991 | return; | |
992 | } | |
993 | ||
994 | sk = tcp_v4_lookup(iph->daddr, th->dest, iph->saddr, | |
995 | th->source, tcp_v4_iif(skb)); | |
996 | if (!sk) { | |
997 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
998 | return; | |
999 | } | |
1000 | if (sk->sk_state == TCP_TIME_WAIT) { | |
1001 | tcp_tw_put((struct tcp_tw_bucket *)sk); | |
1002 | return; | |
1003 | } | |
1004 | ||
1005 | bh_lock_sock(sk); | |
1006 | /* If too many ICMPs get dropped on busy | |
1007 | * servers this needs to be solved differently. | |
1008 | */ | |
1009 | if (sock_owned_by_user(sk)) | |
1010 | NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS); | |
1011 | ||
1012 | if (sk->sk_state == TCP_CLOSE) | |
1013 | goto out; | |
1014 | ||
1015 | tp = tcp_sk(sk); | |
1016 | seq = ntohl(th->seq); | |
1017 | if (sk->sk_state != TCP_LISTEN && | |
1018 | !between(seq, tp->snd_una, tp->snd_nxt)) { | |
1019 | NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS); | |
1020 | goto out; | |
1021 | } | |
1022 | ||
1023 | switch (type) { | |
1024 | case ICMP_SOURCE_QUENCH: | |
1025 | /* Just silently ignore these. */ | |
1026 | goto out; | |
1027 | case ICMP_PARAMETERPROB: | |
1028 | err = EPROTO; | |
1029 | break; | |
1030 | case ICMP_DEST_UNREACH: | |
1031 | if (code > NR_ICMP_UNREACH) | |
1032 | goto out; | |
1033 | ||
1034 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
1035 | if (!sock_owned_by_user(sk)) | |
1036 | do_pmtu_discovery(sk, iph, info); | |
1037 | goto out; | |
1038 | } | |
1039 | ||
1040 | err = icmp_err_convert[code].errno; | |
1041 | break; | |
1042 | case ICMP_TIME_EXCEEDED: | |
1043 | err = EHOSTUNREACH; | |
1044 | break; | |
1045 | default: | |
1046 | goto out; | |
1047 | } | |
1048 | ||
1049 | switch (sk->sk_state) { | |
1050 | struct open_request *req, **prev; | |
1051 | case TCP_LISTEN: | |
1052 | if (sock_owned_by_user(sk)) | |
1053 | goto out; | |
1054 | ||
1055 | req = tcp_v4_search_req(tp, &prev, th->dest, | |
1056 | iph->daddr, iph->saddr); | |
1057 | if (!req) | |
1058 | goto out; | |
1059 | ||
1060 | /* ICMPs are not backlogged, hence we cannot get | |
1061 | an established socket here. | |
1062 | */ | |
1063 | BUG_TRAP(!req->sk); | |
1064 | ||
1065 | if (seq != req->snt_isn) { | |
1066 | NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS); | |
1067 | goto out; | |
1068 | } | |
1069 | ||
1070 | /* | |
1071 | * Still in SYN_RECV, just remove it silently. | |
1072 | * There is no good way to pass the error to the newly | |
1073 | * created socket, and POSIX does not want network | |
1074 | * errors returned from accept(). | |
1075 | */ | |
1076 | tcp_synq_drop(sk, req, prev); | |
1077 | goto out; | |
1078 | ||
1079 | case TCP_SYN_SENT: | |
1080 | case TCP_SYN_RECV: /* Cannot happen. | |
1081 | It can f.e. if SYNs crossed. | |
1082 | */ | |
1083 | if (!sock_owned_by_user(sk)) { | |
1084 | TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); | |
1085 | sk->sk_err = err; | |
1086 | ||
1087 | sk->sk_error_report(sk); | |
1088 | ||
1089 | tcp_done(sk); | |
1090 | } else { | |
1091 | sk->sk_err_soft = err; | |
1092 | } | |
1093 | goto out; | |
1094 | } | |
1095 | ||
1096 | /* If we've already connected we will keep trying | |
1097 | * until we time out, or the user gives up. | |
1098 | * | |
1099 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
1100 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
1101 | * but it is obsoleted by pmtu discovery). | |
1102 | * | |
1103 | * Note, that in modern internet, where routing is unreliable | |
1104 | * and in each dark corner broken firewalls sit, sending random | |
1105 | * errors ordered by their masters even this two messages finally lose | |
1106 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
1107 | * | |
1108 | * Now we are in compliance with RFCs. | |
1109 | * --ANK (980905) | |
1110 | */ | |
1111 | ||
1112 | inet = inet_sk(sk); | |
1113 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
1114 | sk->sk_err = err; | |
1115 | sk->sk_error_report(sk); | |
1116 | } else { /* Only an error on timeout */ | |
1117 | sk->sk_err_soft = err; | |
1118 | } | |
1119 | ||
1120 | out: | |
1121 | bh_unlock_sock(sk); | |
1122 | sock_put(sk); | |
1123 | } | |
1124 | ||
1125 | /* This routine computes an IPv4 TCP checksum. */ | |
1126 | void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len, | |
1127 | struct sk_buff *skb) | |
1128 | { | |
1129 | struct inet_sock *inet = inet_sk(sk); | |
1130 | ||
1131 | if (skb->ip_summed == CHECKSUM_HW) { | |
1132 | th->check = ~tcp_v4_check(th, len, inet->saddr, inet->daddr, 0); | |
1133 | skb->csum = offsetof(struct tcphdr, check); | |
1134 | } else { | |
1135 | th->check = tcp_v4_check(th, len, inet->saddr, inet->daddr, | |
1136 | csum_partial((char *)th, | |
1137 | th->doff << 2, | |
1138 | skb->csum)); | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * This routine will send an RST to the other tcp. | |
1144 | * | |
1145 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
1146 | * for reset. | |
1147 | * Answer: if a packet caused RST, it is not for a socket | |
1148 | * existing in our system, if it is matched to a socket, | |
1149 | * it is just duplicate segment or bug in other side's TCP. | |
1150 | * So that we build reply only basing on parameters | |
1151 | * arrived with segment. | |
1152 | * Exception: precedence violation. We do not implement it in any case. | |
1153 | */ | |
1154 | ||
1155 | static void tcp_v4_send_reset(struct sk_buff *skb) | |
1156 | { | |
1157 | struct tcphdr *th = skb->h.th; | |
1158 | struct tcphdr rth; | |
1159 | struct ip_reply_arg arg; | |
1160 | ||
1161 | /* Never send a reset in response to a reset. */ | |
1162 | if (th->rst) | |
1163 | return; | |
1164 | ||
1165 | if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL) | |
1166 | return; | |
1167 | ||
1168 | /* Swap the send and the receive. */ | |
1169 | memset(&rth, 0, sizeof(struct tcphdr)); | |
1170 | rth.dest = th->source; | |
1171 | rth.source = th->dest; | |
1172 | rth.doff = sizeof(struct tcphdr) / 4; | |
1173 | rth.rst = 1; | |
1174 | ||
1175 | if (th->ack) { | |
1176 | rth.seq = th->ack_seq; | |
1177 | } else { | |
1178 | rth.ack = 1; | |
1179 | rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
1180 | skb->len - (th->doff << 2)); | |
1181 | } | |
1182 | ||
1183 | memset(&arg, 0, sizeof arg); | |
1184 | arg.iov[0].iov_base = (unsigned char *)&rth; | |
1185 | arg.iov[0].iov_len = sizeof rth; | |
1186 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, | |
1187 | skb->nh.iph->saddr, /*XXX*/ | |
1188 | sizeof(struct tcphdr), IPPROTO_TCP, 0); | |
1189 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
1190 | ||
1191 | ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth); | |
1192 | ||
1193 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
1194 | TCP_INC_STATS_BH(TCP_MIB_OUTRSTS); | |
1195 | } | |
1196 | ||
1197 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
1198 | outside socket context is ugly, certainly. What can I do? | |
1199 | */ | |
1200 | ||
1201 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, | |
1202 | u32 win, u32 ts) | |
1203 | { | |
1204 | struct tcphdr *th = skb->h.th; | |
1205 | struct { | |
1206 | struct tcphdr th; | |
1207 | u32 tsopt[3]; | |
1208 | } rep; | |
1209 | struct ip_reply_arg arg; | |
1210 | ||
1211 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
1212 | memset(&arg, 0, sizeof arg); | |
1213 | ||
1214 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
1215 | arg.iov[0].iov_len = sizeof(rep.th); | |
1216 | if (ts) { | |
1217 | rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
1218 | (TCPOPT_TIMESTAMP << 8) | | |
1219 | TCPOLEN_TIMESTAMP); | |
1220 | rep.tsopt[1] = htonl(tcp_time_stamp); | |
1221 | rep.tsopt[2] = htonl(ts); | |
1222 | arg.iov[0].iov_len = sizeof(rep); | |
1223 | } | |
1224 | ||
1225 | /* Swap the send and the receive. */ | |
1226 | rep.th.dest = th->source; | |
1227 | rep.th.source = th->dest; | |
1228 | rep.th.doff = arg.iov[0].iov_len / 4; | |
1229 | rep.th.seq = htonl(seq); | |
1230 | rep.th.ack_seq = htonl(ack); | |
1231 | rep.th.ack = 1; | |
1232 | rep.th.window = htons(win); | |
1233 | ||
1234 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, | |
1235 | skb->nh.iph->saddr, /*XXX*/ | |
1236 | arg.iov[0].iov_len, IPPROTO_TCP, 0); | |
1237 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
1238 | ||
1239 | ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len); | |
1240 | ||
1241 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
1242 | } | |
1243 | ||
1244 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
1245 | { | |
1246 | struct tcp_tw_bucket *tw = (struct tcp_tw_bucket *)sk; | |
1247 | ||
1248 | tcp_v4_send_ack(skb, tw->tw_snd_nxt, tw->tw_rcv_nxt, | |
1249 | tw->tw_rcv_wnd >> tw->tw_rcv_wscale, tw->tw_ts_recent); | |
1250 | ||
1251 | tcp_tw_put(tw); | |
1252 | } | |
1253 | ||
1254 | static void tcp_v4_or_send_ack(struct sk_buff *skb, struct open_request *req) | |
1255 | { | |
1256 | tcp_v4_send_ack(skb, req->snt_isn + 1, req->rcv_isn + 1, req->rcv_wnd, | |
1257 | req->ts_recent); | |
1258 | } | |
1259 | ||
1260 | static struct dst_entry* tcp_v4_route_req(struct sock *sk, | |
1261 | struct open_request *req) | |
1262 | { | |
1263 | struct rtable *rt; | |
1264 | struct ip_options *opt = req->af.v4_req.opt; | |
1265 | struct flowi fl = { .oif = sk->sk_bound_dev_if, | |
1266 | .nl_u = { .ip4_u = | |
1267 | { .daddr = ((opt && opt->srr) ? | |
1268 | opt->faddr : | |
1269 | req->af.v4_req.rmt_addr), | |
1270 | .saddr = req->af.v4_req.loc_addr, | |
1271 | .tos = RT_CONN_FLAGS(sk) } }, | |
1272 | .proto = IPPROTO_TCP, | |
1273 | .uli_u = { .ports = | |
1274 | { .sport = inet_sk(sk)->sport, | |
1275 | .dport = req->rmt_port } } }; | |
1276 | ||
1277 | if (ip_route_output_flow(&rt, &fl, sk, 0)) { | |
1278 | IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); | |
1279 | return NULL; | |
1280 | } | |
1281 | if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) { | |
1282 | ip_rt_put(rt); | |
1283 | IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); | |
1284 | return NULL; | |
1285 | } | |
1286 | return &rt->u.dst; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * Send a SYN-ACK after having received an ACK. | |
1291 | * This still operates on a open_request only, not on a big | |
1292 | * socket. | |
1293 | */ | |
1294 | static int tcp_v4_send_synack(struct sock *sk, struct open_request *req, | |
1295 | struct dst_entry *dst) | |
1296 | { | |
1297 | int err = -1; | |
1298 | struct sk_buff * skb; | |
1299 | ||
1300 | /* First, grab a route. */ | |
1301 | if (!dst && (dst = tcp_v4_route_req(sk, req)) == NULL) | |
1302 | goto out; | |
1303 | ||
1304 | skb = tcp_make_synack(sk, dst, req); | |
1305 | ||
1306 | if (skb) { | |
1307 | struct tcphdr *th = skb->h.th; | |
1308 | ||
1309 | th->check = tcp_v4_check(th, skb->len, | |
1310 | req->af.v4_req.loc_addr, | |
1311 | req->af.v4_req.rmt_addr, | |
1312 | csum_partial((char *)th, skb->len, | |
1313 | skb->csum)); | |
1314 | ||
1315 | err = ip_build_and_send_pkt(skb, sk, req->af.v4_req.loc_addr, | |
1316 | req->af.v4_req.rmt_addr, | |
1317 | req->af.v4_req.opt); | |
1318 | if (err == NET_XMIT_CN) | |
1319 | err = 0; | |
1320 | } | |
1321 | ||
1322 | out: | |
1323 | dst_release(dst); | |
1324 | return err; | |
1325 | } | |
1326 | ||
1327 | /* | |
1328 | * IPv4 open_request destructor. | |
1329 | */ | |
1330 | static void tcp_v4_or_free(struct open_request *req) | |
1331 | { | |
1332 | if (req->af.v4_req.opt) | |
1333 | kfree(req->af.v4_req.opt); | |
1334 | } | |
1335 | ||
1336 | static inline void syn_flood_warning(struct sk_buff *skb) | |
1337 | { | |
1338 | static unsigned long warntime; | |
1339 | ||
1340 | if (time_after(jiffies, (warntime + HZ * 60))) { | |
1341 | warntime = jiffies; | |
1342 | printk(KERN_INFO | |
1343 | "possible SYN flooding on port %d. Sending cookies.\n", | |
1344 | ntohs(skb->h.th->dest)); | |
1345 | } | |
1346 | } | |
1347 | ||
1348 | /* | |
1349 | * Save and compile IPv4 options into the open_request if needed. | |
1350 | */ | |
1351 | static inline struct ip_options *tcp_v4_save_options(struct sock *sk, | |
1352 | struct sk_buff *skb) | |
1353 | { | |
1354 | struct ip_options *opt = &(IPCB(skb)->opt); | |
1355 | struct ip_options *dopt = NULL; | |
1356 | ||
1357 | if (opt && opt->optlen) { | |
1358 | int opt_size = optlength(opt); | |
1359 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
1360 | if (dopt) { | |
1361 | if (ip_options_echo(dopt, skb)) { | |
1362 | kfree(dopt); | |
1363 | dopt = NULL; | |
1364 | } | |
1365 | } | |
1366 | } | |
1367 | return dopt; | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Maximum number of SYN_RECV sockets in queue per LISTEN socket. | |
1372 | * One SYN_RECV socket costs about 80bytes on a 32bit machine. | |
1373 | * It would be better to replace it with a global counter for all sockets | |
1374 | * but then some measure against one socket starving all other sockets | |
1375 | * would be needed. | |
1376 | * | |
1377 | * It was 128 by default. Experiments with real servers show, that | |
1378 | * it is absolutely not enough even at 100conn/sec. 256 cures most | |
1379 | * of problems. This value is adjusted to 128 for very small machines | |
1380 | * (<=32Mb of memory) and to 1024 on normal or better ones (>=256Mb). | |
1381 | * Further increasing requires to change hash table size. | |
1382 | */ | |
1383 | int sysctl_max_syn_backlog = 256; | |
1384 | ||
1385 | struct or_calltable or_ipv4 = { | |
1386 | .family = PF_INET, | |
1387 | .rtx_syn_ack = tcp_v4_send_synack, | |
1388 | .send_ack = tcp_v4_or_send_ack, | |
1389 | .destructor = tcp_v4_or_free, | |
1390 | .send_reset = tcp_v4_send_reset, | |
1391 | }; | |
1392 | ||
1393 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) | |
1394 | { | |
1395 | struct tcp_options_received tmp_opt; | |
1396 | struct open_request *req; | |
1397 | __u32 saddr = skb->nh.iph->saddr; | |
1398 | __u32 daddr = skb->nh.iph->daddr; | |
1399 | __u32 isn = TCP_SKB_CB(skb)->when; | |
1400 | struct dst_entry *dst = NULL; | |
1401 | #ifdef CONFIG_SYN_COOKIES | |
1402 | int want_cookie = 0; | |
1403 | #else | |
1404 | #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */ | |
1405 | #endif | |
1406 | ||
1407 | /* Never answer to SYNs send to broadcast or multicast */ | |
1408 | if (((struct rtable *)skb->dst)->rt_flags & | |
1409 | (RTCF_BROADCAST | RTCF_MULTICAST)) | |
1410 | goto drop; | |
1411 | ||
1412 | /* TW buckets are converted to open requests without | |
1413 | * limitations, they conserve resources and peer is | |
1414 | * evidently real one. | |
1415 | */ | |
1416 | if (tcp_synq_is_full(sk) && !isn) { | |
1417 | #ifdef CONFIG_SYN_COOKIES | |
1418 | if (sysctl_tcp_syncookies) { | |
1419 | want_cookie = 1; | |
1420 | } else | |
1421 | #endif | |
1422 | goto drop; | |
1423 | } | |
1424 | ||
1425 | /* Accept backlog is full. If we have already queued enough | |
1426 | * of warm entries in syn queue, drop request. It is better than | |
1427 | * clogging syn queue with openreqs with exponentially increasing | |
1428 | * timeout. | |
1429 | */ | |
1430 | if (sk_acceptq_is_full(sk) && tcp_synq_young(sk) > 1) | |
1431 | goto drop; | |
1432 | ||
1433 | req = tcp_openreq_alloc(); | |
1434 | if (!req) | |
1435 | goto drop; | |
1436 | ||
1437 | tcp_clear_options(&tmp_opt); | |
1438 | tmp_opt.mss_clamp = 536; | |
1439 | tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss; | |
1440 | ||
1441 | tcp_parse_options(skb, &tmp_opt, 0); | |
1442 | ||
1443 | if (want_cookie) { | |
1444 | tcp_clear_options(&tmp_opt); | |
1445 | tmp_opt.saw_tstamp = 0; | |
1446 | } | |
1447 | ||
1448 | if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) { | |
1449 | /* Some OSes (unknown ones, but I see them on web server, which | |
1450 | * contains information interesting only for windows' | |
1451 | * users) do not send their stamp in SYN. It is easy case. | |
1452 | * We simply do not advertise TS support. | |
1453 | */ | |
1454 | tmp_opt.saw_tstamp = 0; | |
1455 | tmp_opt.tstamp_ok = 0; | |
1456 | } | |
1457 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; | |
1458 | ||
1459 | tcp_openreq_init(req, &tmp_opt, skb); | |
1460 | ||
1461 | req->af.v4_req.loc_addr = daddr; | |
1462 | req->af.v4_req.rmt_addr = saddr; | |
1463 | req->af.v4_req.opt = tcp_v4_save_options(sk, skb); | |
1464 | req->class = &or_ipv4; | |
1465 | if (!want_cookie) | |
1466 | TCP_ECN_create_request(req, skb->h.th); | |
1467 | ||
1468 | if (want_cookie) { | |
1469 | #ifdef CONFIG_SYN_COOKIES | |
1470 | syn_flood_warning(skb); | |
1471 | #endif | |
1472 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); | |
1473 | } else if (!isn) { | |
1474 | struct inet_peer *peer = NULL; | |
1475 | ||
1476 | /* VJ's idea. We save last timestamp seen | |
1477 | * from the destination in peer table, when entering | |
1478 | * state TIME-WAIT, and check against it before | |
1479 | * accepting new connection request. | |
1480 | * | |
1481 | * If "isn" is not zero, this request hit alive | |
1482 | * timewait bucket, so that all the necessary checks | |
1483 | * are made in the function processing timewait state. | |
1484 | */ | |
1485 | if (tmp_opt.saw_tstamp && | |
1486 | sysctl_tcp_tw_recycle && | |
1487 | (dst = tcp_v4_route_req(sk, req)) != NULL && | |
1488 | (peer = rt_get_peer((struct rtable *)dst)) != NULL && | |
1489 | peer->v4daddr == saddr) { | |
1490 | if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL && | |
1491 | (s32)(peer->tcp_ts - req->ts_recent) > | |
1492 | TCP_PAWS_WINDOW) { | |
1493 | NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED); | |
1494 | dst_release(dst); | |
1495 | goto drop_and_free; | |
1496 | } | |
1497 | } | |
1498 | /* Kill the following clause, if you dislike this way. */ | |
1499 | else if (!sysctl_tcp_syncookies && | |
1500 | (sysctl_max_syn_backlog - tcp_synq_len(sk) < | |
1501 | (sysctl_max_syn_backlog >> 2)) && | |
1502 | (!peer || !peer->tcp_ts_stamp) && | |
1503 | (!dst || !dst_metric(dst, RTAX_RTT))) { | |
1504 | /* Without syncookies last quarter of | |
1505 | * backlog is filled with destinations, | |
1506 | * proven to be alive. | |
1507 | * It means that we continue to communicate | |
1508 | * to destinations, already remembered | |
1509 | * to the moment of synflood. | |
1510 | */ | |
1511 | NETDEBUG(if (net_ratelimit()) \ | |
1512 | printk(KERN_DEBUG "TCP: drop open " | |
1513 | "request from %u.%u." | |
1514 | "%u.%u/%u\n", \ | |
1515 | NIPQUAD(saddr), | |
1516 | ntohs(skb->h.th->source))); | |
1517 | dst_release(dst); | |
1518 | goto drop_and_free; | |
1519 | } | |
1520 | ||
1521 | isn = tcp_v4_init_sequence(sk, skb); | |
1522 | } | |
1523 | req->snt_isn = isn; | |
1524 | ||
1525 | if (tcp_v4_send_synack(sk, req, dst)) | |
1526 | goto drop_and_free; | |
1527 | ||
1528 | if (want_cookie) { | |
1529 | tcp_openreq_free(req); | |
1530 | } else { | |
1531 | tcp_v4_synq_add(sk, req); | |
1532 | } | |
1533 | return 0; | |
1534 | ||
1535 | drop_and_free: | |
1536 | tcp_openreq_free(req); | |
1537 | drop: | |
1538 | TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); | |
1539 | return 0; | |
1540 | } | |
1541 | ||
1542 | ||
1543 | /* | |
1544 | * The three way handshake has completed - we got a valid synack - | |
1545 | * now create the new socket. | |
1546 | */ | |
1547 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
1548 | struct open_request *req, | |
1549 | struct dst_entry *dst) | |
1550 | { | |
1551 | struct inet_sock *newinet; | |
1552 | struct tcp_sock *newtp; | |
1553 | struct sock *newsk; | |
1554 | ||
1555 | if (sk_acceptq_is_full(sk)) | |
1556 | goto exit_overflow; | |
1557 | ||
1558 | if (!dst && (dst = tcp_v4_route_req(sk, req)) == NULL) | |
1559 | goto exit; | |
1560 | ||
1561 | newsk = tcp_create_openreq_child(sk, req, skb); | |
1562 | if (!newsk) | |
1563 | goto exit; | |
1564 | ||
1565 | newsk->sk_dst_cache = dst; | |
1566 | tcp_v4_setup_caps(newsk, dst); | |
1567 | ||
1568 | newtp = tcp_sk(newsk); | |
1569 | newinet = inet_sk(newsk); | |
1570 | newinet->daddr = req->af.v4_req.rmt_addr; | |
1571 | newinet->rcv_saddr = req->af.v4_req.loc_addr; | |
1572 | newinet->saddr = req->af.v4_req.loc_addr; | |
1573 | newinet->opt = req->af.v4_req.opt; | |
1574 | req->af.v4_req.opt = NULL; | |
1575 | newinet->mc_index = tcp_v4_iif(skb); | |
1576 | newinet->mc_ttl = skb->nh.iph->ttl; | |
1577 | newtp->ext_header_len = 0; | |
1578 | if (newinet->opt) | |
1579 | newtp->ext_header_len = newinet->opt->optlen; | |
1580 | newinet->id = newtp->write_seq ^ jiffies; | |
1581 | ||
1582 | tcp_sync_mss(newsk, dst_mtu(dst)); | |
1583 | newtp->advmss = dst_metric(dst, RTAX_ADVMSS); | |
1584 | tcp_initialize_rcv_mss(newsk); | |
1585 | ||
1586 | __tcp_v4_hash(newsk, 0); | |
1587 | __tcp_inherit_port(sk, newsk); | |
1588 | ||
1589 | return newsk; | |
1590 | ||
1591 | exit_overflow: | |
1592 | NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS); | |
1593 | exit: | |
1594 | NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS); | |
1595 | dst_release(dst); | |
1596 | return NULL; | |
1597 | } | |
1598 | ||
1599 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1600 | { | |
1601 | struct tcphdr *th = skb->h.th; | |
1602 | struct iphdr *iph = skb->nh.iph; | |
1603 | struct tcp_sock *tp = tcp_sk(sk); | |
1604 | struct sock *nsk; | |
1605 | struct open_request **prev; | |
1606 | /* Find possible connection requests. */ | |
1607 | struct open_request *req = tcp_v4_search_req(tp, &prev, th->source, | |
1608 | iph->saddr, iph->daddr); | |
1609 | if (req) | |
1610 | return tcp_check_req(sk, skb, req, prev); | |
1611 | ||
1612 | nsk = __tcp_v4_lookup_established(skb->nh.iph->saddr, | |
1613 | th->source, | |
1614 | skb->nh.iph->daddr, | |
1615 | ntohs(th->dest), | |
1616 | tcp_v4_iif(skb)); | |
1617 | ||
1618 | if (nsk) { | |
1619 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1620 | bh_lock_sock(nsk); | |
1621 | return nsk; | |
1622 | } | |
1623 | tcp_tw_put((struct tcp_tw_bucket *)nsk); | |
1624 | return NULL; | |
1625 | } | |
1626 | ||
1627 | #ifdef CONFIG_SYN_COOKIES | |
1628 | if (!th->rst && !th->syn && th->ack) | |
1629 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); | |
1630 | #endif | |
1631 | return sk; | |
1632 | } | |
1633 | ||
1634 | static int tcp_v4_checksum_init(struct sk_buff *skb) | |
1635 | { | |
1636 | if (skb->ip_summed == CHECKSUM_HW) { | |
1637 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1638 | if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr, | |
1639 | skb->nh.iph->daddr, skb->csum)) | |
1640 | return 0; | |
1641 | ||
1642 | NETDEBUG(if (net_ratelimit()) | |
1643 | printk(KERN_DEBUG "hw tcp v4 csum failed\n")); | |
1644 | skb->ip_summed = CHECKSUM_NONE; | |
1645 | } | |
1646 | if (skb->len <= 76) { | |
1647 | if (tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr, | |
1648 | skb->nh.iph->daddr, | |
1649 | skb_checksum(skb, 0, skb->len, 0))) | |
1650 | return -1; | |
1651 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1652 | } else { | |
1653 | skb->csum = ~tcp_v4_check(skb->h.th, skb->len, | |
1654 | skb->nh.iph->saddr, | |
1655 | skb->nh.iph->daddr, 0); | |
1656 | } | |
1657 | return 0; | |
1658 | } | |
1659 | ||
1660 | ||
1661 | /* The socket must have it's spinlock held when we get | |
1662 | * here. | |
1663 | * | |
1664 | * We have a potential double-lock case here, so even when | |
1665 | * doing backlog processing we use the BH locking scheme. | |
1666 | * This is because we cannot sleep with the original spinlock | |
1667 | * held. | |
1668 | */ | |
1669 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1670 | { | |
1671 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ | |
1672 | TCP_CHECK_TIMER(sk); | |
1673 | if (tcp_rcv_established(sk, skb, skb->h.th, skb->len)) | |
1674 | goto reset; | |
1675 | TCP_CHECK_TIMER(sk); | |
1676 | return 0; | |
1677 | } | |
1678 | ||
1679 | if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb)) | |
1680 | goto csum_err; | |
1681 | ||
1682 | if (sk->sk_state == TCP_LISTEN) { | |
1683 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1684 | if (!nsk) | |
1685 | goto discard; | |
1686 | ||
1687 | if (nsk != sk) { | |
1688 | if (tcp_child_process(sk, nsk, skb)) | |
1689 | goto reset; | |
1690 | return 0; | |
1691 | } | |
1692 | } | |
1693 | ||
1694 | TCP_CHECK_TIMER(sk); | |
1695 | if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len)) | |
1696 | goto reset; | |
1697 | TCP_CHECK_TIMER(sk); | |
1698 | return 0; | |
1699 | ||
1700 | reset: | |
1701 | tcp_v4_send_reset(skb); | |
1702 | discard: | |
1703 | kfree_skb(skb); | |
1704 | /* Be careful here. If this function gets more complicated and | |
1705 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1706 | * might be destroyed here. This current version compiles correctly, | |
1707 | * but you have been warned. | |
1708 | */ | |
1709 | return 0; | |
1710 | ||
1711 | csum_err: | |
1712 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1713 | goto discard; | |
1714 | } | |
1715 | ||
1716 | /* | |
1717 | * From tcp_input.c | |
1718 | */ | |
1719 | ||
1720 | int tcp_v4_rcv(struct sk_buff *skb) | |
1721 | { | |
1722 | struct tcphdr *th; | |
1723 | struct sock *sk; | |
1724 | int ret; | |
1725 | ||
1726 | if (skb->pkt_type != PACKET_HOST) | |
1727 | goto discard_it; | |
1728 | ||
1729 | /* Count it even if it's bad */ | |
1730 | TCP_INC_STATS_BH(TCP_MIB_INSEGS); | |
1731 | ||
1732 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1733 | goto discard_it; | |
1734 | ||
1735 | th = skb->h.th; | |
1736 | ||
1737 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1738 | goto bad_packet; | |
1739 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1740 | goto discard_it; | |
1741 | ||
1742 | /* An explanation is required here, I think. | |
1743 | * Packet length and doff are validated by header prediction, | |
1744 | * provided case of th->doff==0 is elimineted. | |
1745 | * So, we defer the checks. */ | |
1746 | if ((skb->ip_summed != CHECKSUM_UNNECESSARY && | |
1747 | tcp_v4_checksum_init(skb) < 0)) | |
1748 | goto bad_packet; | |
1749 | ||
1750 | th = skb->h.th; | |
1751 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); | |
1752 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1753 | skb->len - th->doff * 4); | |
1754 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1755 | TCP_SKB_CB(skb)->when = 0; | |
1756 | TCP_SKB_CB(skb)->flags = skb->nh.iph->tos; | |
1757 | TCP_SKB_CB(skb)->sacked = 0; | |
1758 | ||
1759 | sk = __tcp_v4_lookup(skb->nh.iph->saddr, th->source, | |
1760 | skb->nh.iph->daddr, ntohs(th->dest), | |
1761 | tcp_v4_iif(skb)); | |
1762 | ||
1763 | if (!sk) | |
1764 | goto no_tcp_socket; | |
1765 | ||
1766 | process: | |
1767 | if (sk->sk_state == TCP_TIME_WAIT) | |
1768 | goto do_time_wait; | |
1769 | ||
1770 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1771 | goto discard_and_relse; | |
1772 | ||
1773 | if (sk_filter(sk, skb, 0)) | |
1774 | goto discard_and_relse; | |
1775 | ||
1776 | skb->dev = NULL; | |
1777 | ||
1778 | bh_lock_sock(sk); | |
1779 | ret = 0; | |
1780 | if (!sock_owned_by_user(sk)) { | |
1781 | if (!tcp_prequeue(sk, skb)) | |
1782 | ret = tcp_v4_do_rcv(sk, skb); | |
1783 | } else | |
1784 | sk_add_backlog(sk, skb); | |
1785 | bh_unlock_sock(sk); | |
1786 | ||
1787 | sock_put(sk); | |
1788 | ||
1789 | return ret; | |
1790 | ||
1791 | no_tcp_socket: | |
1792 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1793 | goto discard_it; | |
1794 | ||
1795 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1796 | bad_packet: | |
1797 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1798 | } else { | |
1799 | tcp_v4_send_reset(skb); | |
1800 | } | |
1801 | ||
1802 | discard_it: | |
1803 | /* Discard frame. */ | |
1804 | kfree_skb(skb); | |
1805 | return 0; | |
1806 | ||
1807 | discard_and_relse: | |
1808 | sock_put(sk); | |
1809 | goto discard_it; | |
1810 | ||
1811 | do_time_wait: | |
1812 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
1813 | tcp_tw_put((struct tcp_tw_bucket *) sk); | |
1814 | goto discard_it; | |
1815 | } | |
1816 | ||
1817 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1818 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1819 | tcp_tw_put((struct tcp_tw_bucket *) sk); | |
1820 | goto discard_it; | |
1821 | } | |
1822 | switch (tcp_timewait_state_process((struct tcp_tw_bucket *)sk, | |
1823 | skb, th, skb->len)) { | |
1824 | case TCP_TW_SYN: { | |
1825 | struct sock *sk2 = tcp_v4_lookup_listener(skb->nh.iph->daddr, | |
1826 | ntohs(th->dest), | |
1827 | tcp_v4_iif(skb)); | |
1828 | if (sk2) { | |
1829 | tcp_tw_deschedule((struct tcp_tw_bucket *)sk); | |
1830 | tcp_tw_put((struct tcp_tw_bucket *)sk); | |
1831 | sk = sk2; | |
1832 | goto process; | |
1833 | } | |
1834 | /* Fall through to ACK */ | |
1835 | } | |
1836 | case TCP_TW_ACK: | |
1837 | tcp_v4_timewait_ack(sk, skb); | |
1838 | break; | |
1839 | case TCP_TW_RST: | |
1840 | goto no_tcp_socket; | |
1841 | case TCP_TW_SUCCESS:; | |
1842 | } | |
1843 | goto discard_it; | |
1844 | } | |
1845 | ||
1846 | /* With per-bucket locks this operation is not-atomic, so that | |
1847 | * this version is not worse. | |
1848 | */ | |
1849 | static void __tcp_v4_rehash(struct sock *sk) | |
1850 | { | |
1851 | sk->sk_prot->unhash(sk); | |
1852 | sk->sk_prot->hash(sk); | |
1853 | } | |
1854 | ||
1855 | static int tcp_v4_reselect_saddr(struct sock *sk) | |
1856 | { | |
1857 | struct inet_sock *inet = inet_sk(sk); | |
1858 | int err; | |
1859 | struct rtable *rt; | |
1860 | __u32 old_saddr = inet->saddr; | |
1861 | __u32 new_saddr; | |
1862 | __u32 daddr = inet->daddr; | |
1863 | ||
1864 | if (inet->opt && inet->opt->srr) | |
1865 | daddr = inet->opt->faddr; | |
1866 | ||
1867 | /* Query new route. */ | |
1868 | err = ip_route_connect(&rt, daddr, 0, | |
1869 | RT_CONN_FLAGS(sk), | |
1870 | sk->sk_bound_dev_if, | |
1871 | IPPROTO_TCP, | |
1872 | inet->sport, inet->dport, sk); | |
1873 | if (err) | |
1874 | return err; | |
1875 | ||
1876 | __sk_dst_set(sk, &rt->u.dst); | |
1877 | tcp_v4_setup_caps(sk, &rt->u.dst); | |
1878 | ||
1879 | new_saddr = rt->rt_src; | |
1880 | ||
1881 | if (new_saddr == old_saddr) | |
1882 | return 0; | |
1883 | ||
1884 | if (sysctl_ip_dynaddr > 1) { | |
1885 | printk(KERN_INFO "tcp_v4_rebuild_header(): shifting inet->" | |
1886 | "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n", | |
1887 | NIPQUAD(old_saddr), | |
1888 | NIPQUAD(new_saddr)); | |
1889 | } | |
1890 | ||
1891 | inet->saddr = new_saddr; | |
1892 | inet->rcv_saddr = new_saddr; | |
1893 | ||
1894 | /* XXX The only one ugly spot where we need to | |
1895 | * XXX really change the sockets identity after | |
1896 | * XXX it has entered the hashes. -DaveM | |
1897 | * | |
1898 | * Besides that, it does not check for connection | |
1899 | * uniqueness. Wait for troubles. | |
1900 | */ | |
1901 | __tcp_v4_rehash(sk); | |
1902 | return 0; | |
1903 | } | |
1904 | ||
1905 | int tcp_v4_rebuild_header(struct sock *sk) | |
1906 | { | |
1907 | struct inet_sock *inet = inet_sk(sk); | |
1908 | struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); | |
1909 | u32 daddr; | |
1910 | int err; | |
1911 | ||
1912 | /* Route is OK, nothing to do. */ | |
1913 | if (rt) | |
1914 | return 0; | |
1915 | ||
1916 | /* Reroute. */ | |
1917 | daddr = inet->daddr; | |
1918 | if (inet->opt && inet->opt->srr) | |
1919 | daddr = inet->opt->faddr; | |
1920 | ||
1921 | { | |
1922 | struct flowi fl = { .oif = sk->sk_bound_dev_if, | |
1923 | .nl_u = { .ip4_u = | |
1924 | { .daddr = daddr, | |
1925 | .saddr = inet->saddr, | |
1926 | .tos = RT_CONN_FLAGS(sk) } }, | |
1927 | .proto = IPPROTO_TCP, | |
1928 | .uli_u = { .ports = | |
1929 | { .sport = inet->sport, | |
1930 | .dport = inet->dport } } }; | |
1931 | ||
1932 | err = ip_route_output_flow(&rt, &fl, sk, 0); | |
1933 | } | |
1934 | if (!err) { | |
1935 | __sk_dst_set(sk, &rt->u.dst); | |
1936 | tcp_v4_setup_caps(sk, &rt->u.dst); | |
1937 | return 0; | |
1938 | } | |
1939 | ||
1940 | /* Routing failed... */ | |
1941 | sk->sk_route_caps = 0; | |
1942 | ||
1943 | if (!sysctl_ip_dynaddr || | |
1944 | sk->sk_state != TCP_SYN_SENT || | |
1945 | (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || | |
1946 | (err = tcp_v4_reselect_saddr(sk)) != 0) | |
1947 | sk->sk_err_soft = -err; | |
1948 | ||
1949 | return err; | |
1950 | } | |
1951 | ||
1952 | static void v4_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr) | |
1953 | { | |
1954 | struct sockaddr_in *sin = (struct sockaddr_in *) uaddr; | |
1955 | struct inet_sock *inet = inet_sk(sk); | |
1956 | ||
1957 | sin->sin_family = AF_INET; | |
1958 | sin->sin_addr.s_addr = inet->daddr; | |
1959 | sin->sin_port = inet->dport; | |
1960 | } | |
1961 | ||
1962 | /* VJ's idea. Save last timestamp seen from this destination | |
1963 | * and hold it at least for normal timewait interval to use for duplicate | |
1964 | * segment detection in subsequent connections, before they enter synchronized | |
1965 | * state. | |
1966 | */ | |
1967 | ||
1968 | int tcp_v4_remember_stamp(struct sock *sk) | |
1969 | { | |
1970 | struct inet_sock *inet = inet_sk(sk); | |
1971 | struct tcp_sock *tp = tcp_sk(sk); | |
1972 | struct rtable *rt = (struct rtable *)__sk_dst_get(sk); | |
1973 | struct inet_peer *peer = NULL; | |
1974 | int release_it = 0; | |
1975 | ||
1976 | if (!rt || rt->rt_dst != inet->daddr) { | |
1977 | peer = inet_getpeer(inet->daddr, 1); | |
1978 | release_it = 1; | |
1979 | } else { | |
1980 | if (!rt->peer) | |
1981 | rt_bind_peer(rt, 1); | |
1982 | peer = rt->peer; | |
1983 | } | |
1984 | ||
1985 | if (peer) { | |
1986 | if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 || | |
1987 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && | |
1988 | peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) { | |
1989 | peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp; | |
1990 | peer->tcp_ts = tp->rx_opt.ts_recent; | |
1991 | } | |
1992 | if (release_it) | |
1993 | inet_putpeer(peer); | |
1994 | return 1; | |
1995 | } | |
1996 | ||
1997 | return 0; | |
1998 | } | |
1999 | ||
2000 | int tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw) | |
2001 | { | |
2002 | struct inet_peer *peer = NULL; | |
2003 | ||
2004 | peer = inet_getpeer(tw->tw_daddr, 1); | |
2005 | ||
2006 | if (peer) { | |
2007 | if ((s32)(peer->tcp_ts - tw->tw_ts_recent) <= 0 || | |
2008 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && | |
2009 | peer->tcp_ts_stamp <= tw->tw_ts_recent_stamp)) { | |
2010 | peer->tcp_ts_stamp = tw->tw_ts_recent_stamp; | |
2011 | peer->tcp_ts = tw->tw_ts_recent; | |
2012 | } | |
2013 | inet_putpeer(peer); | |
2014 | return 1; | |
2015 | } | |
2016 | ||
2017 | return 0; | |
2018 | } | |
2019 | ||
2020 | struct tcp_func ipv4_specific = { | |
2021 | .queue_xmit = ip_queue_xmit, | |
2022 | .send_check = tcp_v4_send_check, | |
2023 | .rebuild_header = tcp_v4_rebuild_header, | |
2024 | .conn_request = tcp_v4_conn_request, | |
2025 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
2026 | .remember_stamp = tcp_v4_remember_stamp, | |
2027 | .net_header_len = sizeof(struct iphdr), | |
2028 | .setsockopt = ip_setsockopt, | |
2029 | .getsockopt = ip_getsockopt, | |
2030 | .addr2sockaddr = v4_addr2sockaddr, | |
2031 | .sockaddr_len = sizeof(struct sockaddr_in), | |
2032 | }; | |
2033 | ||
2034 | /* NOTE: A lot of things set to zero explicitly by call to | |
2035 | * sk_alloc() so need not be done here. | |
2036 | */ | |
2037 | static int tcp_v4_init_sock(struct sock *sk) | |
2038 | { | |
2039 | struct tcp_sock *tp = tcp_sk(sk); | |
2040 | ||
2041 | skb_queue_head_init(&tp->out_of_order_queue); | |
2042 | tcp_init_xmit_timers(sk); | |
2043 | tcp_prequeue_init(tp); | |
2044 | ||
2045 | tp->rto = TCP_TIMEOUT_INIT; | |
2046 | tp->mdev = TCP_TIMEOUT_INIT; | |
2047 | ||
2048 | /* So many TCP implementations out there (incorrectly) count the | |
2049 | * initial SYN frame in their delayed-ACK and congestion control | |
2050 | * algorithms that we must have the following bandaid to talk | |
2051 | * efficiently to them. -DaveM | |
2052 | */ | |
2053 | tp->snd_cwnd = 2; | |
2054 | ||
2055 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
2056 | * initialization of these values. | |
2057 | */ | |
2058 | tp->snd_ssthresh = 0x7fffffff; /* Infinity */ | |
2059 | tp->snd_cwnd_clamp = ~0; | |
2060 | tp->mss_cache_std = tp->mss_cache = 536; | |
2061 | ||
2062 | tp->reordering = sysctl_tcp_reordering; | |
2063 | ||
2064 | sk->sk_state = TCP_CLOSE; | |
2065 | ||
2066 | sk->sk_write_space = sk_stream_write_space; | |
2067 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
2068 | ||
2069 | tp->af_specific = &ipv4_specific; | |
2070 | ||
2071 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; | |
2072 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
2073 | ||
2074 | atomic_inc(&tcp_sockets_allocated); | |
2075 | ||
2076 | return 0; | |
2077 | } | |
2078 | ||
2079 | int tcp_v4_destroy_sock(struct sock *sk) | |
2080 | { | |
2081 | struct tcp_sock *tp = tcp_sk(sk); | |
2082 | ||
2083 | tcp_clear_xmit_timers(sk); | |
2084 | ||
2085 | /* Cleanup up the write buffer. */ | |
2086 | sk_stream_writequeue_purge(sk); | |
2087 | ||
2088 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
2089 | __skb_queue_purge(&tp->out_of_order_queue); | |
2090 | ||
2091 | /* Clean prequeue, it must be empty really */ | |
2092 | __skb_queue_purge(&tp->ucopy.prequeue); | |
2093 | ||
2094 | /* Clean up a referenced TCP bind bucket. */ | |
2095 | if (tp->bind_hash) | |
2096 | tcp_put_port(sk); | |
2097 | ||
2098 | /* | |
2099 | * If sendmsg cached page exists, toss it. | |
2100 | */ | |
2101 | if (sk->sk_sndmsg_page) { | |
2102 | __free_page(sk->sk_sndmsg_page); | |
2103 | sk->sk_sndmsg_page = NULL; | |
2104 | } | |
2105 | ||
2106 | atomic_dec(&tcp_sockets_allocated); | |
2107 | ||
2108 | return 0; | |
2109 | } | |
2110 | ||
2111 | EXPORT_SYMBOL(tcp_v4_destroy_sock); | |
2112 | ||
2113 | #ifdef CONFIG_PROC_FS | |
2114 | /* Proc filesystem TCP sock list dumping. */ | |
2115 | ||
2116 | static inline struct tcp_tw_bucket *tw_head(struct hlist_head *head) | |
2117 | { | |
2118 | return hlist_empty(head) ? NULL : | |
2119 | list_entry(head->first, struct tcp_tw_bucket, tw_node); | |
2120 | } | |
2121 | ||
2122 | static inline struct tcp_tw_bucket *tw_next(struct tcp_tw_bucket *tw) | |
2123 | { | |
2124 | return tw->tw_node.next ? | |
2125 | hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; | |
2126 | } | |
2127 | ||
2128 | static void *listening_get_next(struct seq_file *seq, void *cur) | |
2129 | { | |
2130 | struct tcp_sock *tp; | |
2131 | struct hlist_node *node; | |
2132 | struct sock *sk = cur; | |
2133 | struct tcp_iter_state* st = seq->private; | |
2134 | ||
2135 | if (!sk) { | |
2136 | st->bucket = 0; | |
2137 | sk = sk_head(&tcp_listening_hash[0]); | |
2138 | goto get_sk; | |
2139 | } | |
2140 | ||
2141 | ++st->num; | |
2142 | ||
2143 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
2144 | struct open_request *req = cur; | |
2145 | ||
2146 | tp = tcp_sk(st->syn_wait_sk); | |
2147 | req = req->dl_next; | |
2148 | while (1) { | |
2149 | while (req) { | |
2150 | if (req->class->family == st->family) { | |
2151 | cur = req; | |
2152 | goto out; | |
2153 | } | |
2154 | req = req->dl_next; | |
2155 | } | |
2156 | if (++st->sbucket >= TCP_SYNQ_HSIZE) | |
2157 | break; | |
2158 | get_req: | |
2159 | req = tp->listen_opt->syn_table[st->sbucket]; | |
2160 | } | |
2161 | sk = sk_next(st->syn_wait_sk); | |
2162 | st->state = TCP_SEQ_STATE_LISTENING; | |
2163 | read_unlock_bh(&tp->syn_wait_lock); | |
2164 | } else { | |
2165 | tp = tcp_sk(sk); | |
2166 | read_lock_bh(&tp->syn_wait_lock); | |
2167 | if (tp->listen_opt && tp->listen_opt->qlen) | |
2168 | goto start_req; | |
2169 | read_unlock_bh(&tp->syn_wait_lock); | |
2170 | sk = sk_next(sk); | |
2171 | } | |
2172 | get_sk: | |
2173 | sk_for_each_from(sk, node) { | |
2174 | if (sk->sk_family == st->family) { | |
2175 | cur = sk; | |
2176 | goto out; | |
2177 | } | |
2178 | tp = tcp_sk(sk); | |
2179 | read_lock_bh(&tp->syn_wait_lock); | |
2180 | if (tp->listen_opt && tp->listen_opt->qlen) { | |
2181 | start_req: | |
2182 | st->uid = sock_i_uid(sk); | |
2183 | st->syn_wait_sk = sk; | |
2184 | st->state = TCP_SEQ_STATE_OPENREQ; | |
2185 | st->sbucket = 0; | |
2186 | goto get_req; | |
2187 | } | |
2188 | read_unlock_bh(&tp->syn_wait_lock); | |
2189 | } | |
2190 | if (++st->bucket < TCP_LHTABLE_SIZE) { | |
2191 | sk = sk_head(&tcp_listening_hash[st->bucket]); | |
2192 | goto get_sk; | |
2193 | } | |
2194 | cur = NULL; | |
2195 | out: | |
2196 | return cur; | |
2197 | } | |
2198 | ||
2199 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
2200 | { | |
2201 | void *rc = listening_get_next(seq, NULL); | |
2202 | ||
2203 | while (rc && *pos) { | |
2204 | rc = listening_get_next(seq, rc); | |
2205 | --*pos; | |
2206 | } | |
2207 | return rc; | |
2208 | } | |
2209 | ||
2210 | static void *established_get_first(struct seq_file *seq) | |
2211 | { | |
2212 | struct tcp_iter_state* st = seq->private; | |
2213 | void *rc = NULL; | |
2214 | ||
2215 | for (st->bucket = 0; st->bucket < tcp_ehash_size; ++st->bucket) { | |
2216 | struct sock *sk; | |
2217 | struct hlist_node *node; | |
2218 | struct tcp_tw_bucket *tw; | |
2219 | ||
2220 | /* We can reschedule _before_ having picked the target: */ | |
2221 | cond_resched_softirq(); | |
2222 | ||
2223 | read_lock(&tcp_ehash[st->bucket].lock); | |
2224 | sk_for_each(sk, node, &tcp_ehash[st->bucket].chain) { | |
2225 | if (sk->sk_family != st->family) { | |
2226 | continue; | |
2227 | } | |
2228 | rc = sk; | |
2229 | goto out; | |
2230 | } | |
2231 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
2232 | tw_for_each(tw, node, | |
2233 | &tcp_ehash[st->bucket + tcp_ehash_size].chain) { | |
2234 | if (tw->tw_family != st->family) { | |
2235 | continue; | |
2236 | } | |
2237 | rc = tw; | |
2238 | goto out; | |
2239 | } | |
2240 | read_unlock(&tcp_ehash[st->bucket].lock); | |
2241 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2242 | } | |
2243 | out: | |
2244 | return rc; | |
2245 | } | |
2246 | ||
2247 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2248 | { | |
2249 | struct sock *sk = cur; | |
2250 | struct tcp_tw_bucket *tw; | |
2251 | struct hlist_node *node; | |
2252 | struct tcp_iter_state* st = seq->private; | |
2253 | ||
2254 | ++st->num; | |
2255 | ||
2256 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { | |
2257 | tw = cur; | |
2258 | tw = tw_next(tw); | |
2259 | get_tw: | |
2260 | while (tw && tw->tw_family != st->family) { | |
2261 | tw = tw_next(tw); | |
2262 | } | |
2263 | if (tw) { | |
2264 | cur = tw; | |
2265 | goto out; | |
2266 | } | |
2267 | read_unlock(&tcp_ehash[st->bucket].lock); | |
2268 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2269 | ||
2270 | /* We can reschedule between buckets: */ | |
2271 | cond_resched_softirq(); | |
2272 | ||
2273 | if (++st->bucket < tcp_ehash_size) { | |
2274 | read_lock(&tcp_ehash[st->bucket].lock); | |
2275 | sk = sk_head(&tcp_ehash[st->bucket].chain); | |
2276 | } else { | |
2277 | cur = NULL; | |
2278 | goto out; | |
2279 | } | |
2280 | } else | |
2281 | sk = sk_next(sk); | |
2282 | ||
2283 | sk_for_each_from(sk, node) { | |
2284 | if (sk->sk_family == st->family) | |
2285 | goto found; | |
2286 | } | |
2287 | ||
2288 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
2289 | tw = tw_head(&tcp_ehash[st->bucket + tcp_ehash_size].chain); | |
2290 | goto get_tw; | |
2291 | found: | |
2292 | cur = sk; | |
2293 | out: | |
2294 | return cur; | |
2295 | } | |
2296 | ||
2297 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2298 | { | |
2299 | void *rc = established_get_first(seq); | |
2300 | ||
2301 | while (rc && pos) { | |
2302 | rc = established_get_next(seq, rc); | |
2303 | --pos; | |
2304 | } | |
2305 | return rc; | |
2306 | } | |
2307 | ||
2308 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2309 | { | |
2310 | void *rc; | |
2311 | struct tcp_iter_state* st = seq->private; | |
2312 | ||
2313 | tcp_listen_lock(); | |
2314 | st->state = TCP_SEQ_STATE_LISTENING; | |
2315 | rc = listening_get_idx(seq, &pos); | |
2316 | ||
2317 | if (!rc) { | |
2318 | tcp_listen_unlock(); | |
2319 | local_bh_disable(); | |
2320 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2321 | rc = established_get_idx(seq, pos); | |
2322 | } | |
2323 | ||
2324 | return rc; | |
2325 | } | |
2326 | ||
2327 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) | |
2328 | { | |
2329 | struct tcp_iter_state* st = seq->private; | |
2330 | st->state = TCP_SEQ_STATE_LISTENING; | |
2331 | st->num = 0; | |
2332 | return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2333 | } | |
2334 | ||
2335 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2336 | { | |
2337 | void *rc = NULL; | |
2338 | struct tcp_iter_state* st; | |
2339 | ||
2340 | if (v == SEQ_START_TOKEN) { | |
2341 | rc = tcp_get_idx(seq, 0); | |
2342 | goto out; | |
2343 | } | |
2344 | st = seq->private; | |
2345 | ||
2346 | switch (st->state) { | |
2347 | case TCP_SEQ_STATE_OPENREQ: | |
2348 | case TCP_SEQ_STATE_LISTENING: | |
2349 | rc = listening_get_next(seq, v); | |
2350 | if (!rc) { | |
2351 | tcp_listen_unlock(); | |
2352 | local_bh_disable(); | |
2353 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2354 | rc = established_get_first(seq); | |
2355 | } | |
2356 | break; | |
2357 | case TCP_SEQ_STATE_ESTABLISHED: | |
2358 | case TCP_SEQ_STATE_TIME_WAIT: | |
2359 | rc = established_get_next(seq, v); | |
2360 | break; | |
2361 | } | |
2362 | out: | |
2363 | ++*pos; | |
2364 | return rc; | |
2365 | } | |
2366 | ||
2367 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2368 | { | |
2369 | struct tcp_iter_state* st = seq->private; | |
2370 | ||
2371 | switch (st->state) { | |
2372 | case TCP_SEQ_STATE_OPENREQ: | |
2373 | if (v) { | |
2374 | struct tcp_sock *tp = tcp_sk(st->syn_wait_sk); | |
2375 | read_unlock_bh(&tp->syn_wait_lock); | |
2376 | } | |
2377 | case TCP_SEQ_STATE_LISTENING: | |
2378 | if (v != SEQ_START_TOKEN) | |
2379 | tcp_listen_unlock(); | |
2380 | break; | |
2381 | case TCP_SEQ_STATE_TIME_WAIT: | |
2382 | case TCP_SEQ_STATE_ESTABLISHED: | |
2383 | if (v) | |
2384 | read_unlock(&tcp_ehash[st->bucket].lock); | |
2385 | local_bh_enable(); | |
2386 | break; | |
2387 | } | |
2388 | } | |
2389 | ||
2390 | static int tcp_seq_open(struct inode *inode, struct file *file) | |
2391 | { | |
2392 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; | |
2393 | struct seq_file *seq; | |
2394 | struct tcp_iter_state *s; | |
2395 | int rc; | |
2396 | ||
2397 | if (unlikely(afinfo == NULL)) | |
2398 | return -EINVAL; | |
2399 | ||
2400 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
2401 | if (!s) | |
2402 | return -ENOMEM; | |
2403 | memset(s, 0, sizeof(*s)); | |
2404 | s->family = afinfo->family; | |
2405 | s->seq_ops.start = tcp_seq_start; | |
2406 | s->seq_ops.next = tcp_seq_next; | |
2407 | s->seq_ops.show = afinfo->seq_show; | |
2408 | s->seq_ops.stop = tcp_seq_stop; | |
2409 | ||
2410 | rc = seq_open(file, &s->seq_ops); | |
2411 | if (rc) | |
2412 | goto out_kfree; | |
2413 | seq = file->private_data; | |
2414 | seq->private = s; | |
2415 | out: | |
2416 | return rc; | |
2417 | out_kfree: | |
2418 | kfree(s); | |
2419 | goto out; | |
2420 | } | |
2421 | ||
2422 | int tcp_proc_register(struct tcp_seq_afinfo *afinfo) | |
2423 | { | |
2424 | int rc = 0; | |
2425 | struct proc_dir_entry *p; | |
2426 | ||
2427 | if (!afinfo) | |
2428 | return -EINVAL; | |
2429 | afinfo->seq_fops->owner = afinfo->owner; | |
2430 | afinfo->seq_fops->open = tcp_seq_open; | |
2431 | afinfo->seq_fops->read = seq_read; | |
2432 | afinfo->seq_fops->llseek = seq_lseek; | |
2433 | afinfo->seq_fops->release = seq_release_private; | |
2434 | ||
2435 | p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops); | |
2436 | if (p) | |
2437 | p->data = afinfo; | |
2438 | else | |
2439 | rc = -ENOMEM; | |
2440 | return rc; | |
2441 | } | |
2442 | ||
2443 | void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo) | |
2444 | { | |
2445 | if (!afinfo) | |
2446 | return; | |
2447 | proc_net_remove(afinfo->name); | |
2448 | memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); | |
2449 | } | |
2450 | ||
2451 | static void get_openreq4(struct sock *sk, struct open_request *req, | |
2452 | char *tmpbuf, int i, int uid) | |
2453 | { | |
2454 | int ttd = req->expires - jiffies; | |
2455 | ||
2456 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" | |
2457 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p", | |
2458 | i, | |
2459 | req->af.v4_req.loc_addr, | |
2460 | ntohs(inet_sk(sk)->sport), | |
2461 | req->af.v4_req.rmt_addr, | |
2462 | ntohs(req->rmt_port), | |
2463 | TCP_SYN_RECV, | |
2464 | 0, 0, /* could print option size, but that is af dependent. */ | |
2465 | 1, /* timers active (only the expire timer) */ | |
2466 | jiffies_to_clock_t(ttd), | |
2467 | req->retrans, | |
2468 | uid, | |
2469 | 0, /* non standard timer */ | |
2470 | 0, /* open_requests have no inode */ | |
2471 | atomic_read(&sk->sk_refcnt), | |
2472 | req); | |
2473 | } | |
2474 | ||
2475 | static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i) | |
2476 | { | |
2477 | int timer_active; | |
2478 | unsigned long timer_expires; | |
2479 | struct tcp_sock *tp = tcp_sk(sp); | |
2480 | struct inet_sock *inet = inet_sk(sp); | |
2481 | unsigned int dest = inet->daddr; | |
2482 | unsigned int src = inet->rcv_saddr; | |
2483 | __u16 destp = ntohs(inet->dport); | |
2484 | __u16 srcp = ntohs(inet->sport); | |
2485 | ||
2486 | if (tp->pending == TCP_TIME_RETRANS) { | |
2487 | timer_active = 1; | |
2488 | timer_expires = tp->timeout; | |
2489 | } else if (tp->pending == TCP_TIME_PROBE0) { | |
2490 | timer_active = 4; | |
2491 | timer_expires = tp->timeout; | |
2492 | } else if (timer_pending(&sp->sk_timer)) { | |
2493 | timer_active = 2; | |
2494 | timer_expires = sp->sk_timer.expires; | |
2495 | } else { | |
2496 | timer_active = 0; | |
2497 | timer_expires = jiffies; | |
2498 | } | |
2499 | ||
2500 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " | |
2501 | "%08X %5d %8d %lu %d %p %u %u %u %u %d", | |
2502 | i, src, srcp, dest, destp, sp->sk_state, | |
2503 | tp->write_seq - tp->snd_una, tp->rcv_nxt - tp->copied_seq, | |
2504 | timer_active, | |
2505 | jiffies_to_clock_t(timer_expires - jiffies), | |
2506 | tp->retransmits, | |
2507 | sock_i_uid(sp), | |
2508 | tp->probes_out, | |
2509 | sock_i_ino(sp), | |
2510 | atomic_read(&sp->sk_refcnt), sp, | |
2511 | tp->rto, tp->ack.ato, (tp->ack.quick << 1) | tp->ack.pingpong, | |
2512 | tp->snd_cwnd, | |
2513 | tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh); | |
2514 | } | |
2515 | ||
2516 | static void get_timewait4_sock(struct tcp_tw_bucket *tw, char *tmpbuf, int i) | |
2517 | { | |
2518 | unsigned int dest, src; | |
2519 | __u16 destp, srcp; | |
2520 | int ttd = tw->tw_ttd - jiffies; | |
2521 | ||
2522 | if (ttd < 0) | |
2523 | ttd = 0; | |
2524 | ||
2525 | dest = tw->tw_daddr; | |
2526 | src = tw->tw_rcv_saddr; | |
2527 | destp = ntohs(tw->tw_dport); | |
2528 | srcp = ntohs(tw->tw_sport); | |
2529 | ||
2530 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" | |
2531 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p", | |
2532 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, | |
2533 | 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, | |
2534 | atomic_read(&tw->tw_refcnt), tw); | |
2535 | } | |
2536 | ||
2537 | #define TMPSZ 150 | |
2538 | ||
2539 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2540 | { | |
2541 | struct tcp_iter_state* st; | |
2542 | char tmpbuf[TMPSZ + 1]; | |
2543 | ||
2544 | if (v == SEQ_START_TOKEN) { | |
2545 | seq_printf(seq, "%-*s\n", TMPSZ - 1, | |
2546 | " sl local_address rem_address st tx_queue " | |
2547 | "rx_queue tr tm->when retrnsmt uid timeout " | |
2548 | "inode"); | |
2549 | goto out; | |
2550 | } | |
2551 | st = seq->private; | |
2552 | ||
2553 | switch (st->state) { | |
2554 | case TCP_SEQ_STATE_LISTENING: | |
2555 | case TCP_SEQ_STATE_ESTABLISHED: | |
2556 | get_tcp4_sock(v, tmpbuf, st->num); | |
2557 | break; | |
2558 | case TCP_SEQ_STATE_OPENREQ: | |
2559 | get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid); | |
2560 | break; | |
2561 | case TCP_SEQ_STATE_TIME_WAIT: | |
2562 | get_timewait4_sock(v, tmpbuf, st->num); | |
2563 | break; | |
2564 | } | |
2565 | seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf); | |
2566 | out: | |
2567 | return 0; | |
2568 | } | |
2569 | ||
2570 | static struct file_operations tcp4_seq_fops; | |
2571 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { | |
2572 | .owner = THIS_MODULE, | |
2573 | .name = "tcp", | |
2574 | .family = AF_INET, | |
2575 | .seq_show = tcp4_seq_show, | |
2576 | .seq_fops = &tcp4_seq_fops, | |
2577 | }; | |
2578 | ||
2579 | int __init tcp4_proc_init(void) | |
2580 | { | |
2581 | return tcp_proc_register(&tcp4_seq_afinfo); | |
2582 | } | |
2583 | ||
2584 | void tcp4_proc_exit(void) | |
2585 | { | |
2586 | tcp_proc_unregister(&tcp4_seq_afinfo); | |
2587 | } | |
2588 | #endif /* CONFIG_PROC_FS */ | |
2589 | ||
2590 | struct proto tcp_prot = { | |
2591 | .name = "TCP", | |
2592 | .owner = THIS_MODULE, | |
2593 | .close = tcp_close, | |
2594 | .connect = tcp_v4_connect, | |
2595 | .disconnect = tcp_disconnect, | |
2596 | .accept = tcp_accept, | |
2597 | .ioctl = tcp_ioctl, | |
2598 | .init = tcp_v4_init_sock, | |
2599 | .destroy = tcp_v4_destroy_sock, | |
2600 | .shutdown = tcp_shutdown, | |
2601 | .setsockopt = tcp_setsockopt, | |
2602 | .getsockopt = tcp_getsockopt, | |
2603 | .sendmsg = tcp_sendmsg, | |
2604 | .recvmsg = tcp_recvmsg, | |
2605 | .backlog_rcv = tcp_v4_do_rcv, | |
2606 | .hash = tcp_v4_hash, | |
2607 | .unhash = tcp_unhash, | |
2608 | .get_port = tcp_v4_get_port, | |
2609 | .enter_memory_pressure = tcp_enter_memory_pressure, | |
2610 | .sockets_allocated = &tcp_sockets_allocated, | |
2611 | .memory_allocated = &tcp_memory_allocated, | |
2612 | .memory_pressure = &tcp_memory_pressure, | |
2613 | .sysctl_mem = sysctl_tcp_mem, | |
2614 | .sysctl_wmem = sysctl_tcp_wmem, | |
2615 | .sysctl_rmem = sysctl_tcp_rmem, | |
2616 | .max_header = MAX_TCP_HEADER, | |
2617 | .obj_size = sizeof(struct tcp_sock), | |
2618 | }; | |
2619 | ||
2620 | ||
2621 | ||
2622 | void __init tcp_v4_init(struct net_proto_family *ops) | |
2623 | { | |
2624 | int err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_TCP, &tcp_socket); | |
2625 | if (err < 0) | |
2626 | panic("Failed to create the TCP control socket.\n"); | |
2627 | tcp_socket->sk->sk_allocation = GFP_ATOMIC; | |
2628 | inet_sk(tcp_socket->sk)->uc_ttl = -1; | |
2629 | ||
2630 | /* Unhash it so that IP input processing does not even | |
2631 | * see it, we do not wish this socket to see incoming | |
2632 | * packets. | |
2633 | */ | |
2634 | tcp_socket->sk->sk_prot->unhash(tcp_socket->sk); | |
2635 | } | |
2636 | ||
2637 | EXPORT_SYMBOL(ipv4_specific); | |
2638 | EXPORT_SYMBOL(tcp_bind_hash); | |
2639 | EXPORT_SYMBOL(tcp_bucket_create); | |
2640 | EXPORT_SYMBOL(tcp_hashinfo); | |
2641 | EXPORT_SYMBOL(tcp_inherit_port); | |
2642 | EXPORT_SYMBOL(tcp_listen_wlock); | |
2643 | EXPORT_SYMBOL(tcp_port_rover); | |
2644 | EXPORT_SYMBOL(tcp_prot); | |
2645 | EXPORT_SYMBOL(tcp_put_port); | |
2646 | EXPORT_SYMBOL(tcp_unhash); | |
2647 | EXPORT_SYMBOL(tcp_v4_conn_request); | |
2648 | EXPORT_SYMBOL(tcp_v4_connect); | |
2649 | EXPORT_SYMBOL(tcp_v4_do_rcv); | |
2650 | EXPORT_SYMBOL(tcp_v4_rebuild_header); | |
2651 | EXPORT_SYMBOL(tcp_v4_remember_stamp); | |
2652 | EXPORT_SYMBOL(tcp_v4_send_check); | |
2653 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); | |
2654 | ||
2655 | #ifdef CONFIG_PROC_FS | |
2656 | EXPORT_SYMBOL(tcp_proc_register); | |
2657 | EXPORT_SYMBOL(tcp_proc_unregister); | |
2658 | #endif | |
2659 | EXPORT_SYMBOL(sysctl_local_port_range); | |
2660 | EXPORT_SYMBOL(sysctl_max_syn_backlog); | |
2661 | EXPORT_SYMBOL(sysctl_tcp_low_latency); | |
2662 | EXPORT_SYMBOL(sysctl_tcp_tw_reuse); | |
2663 |