]>
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 | |
60236fdd | 39 | * request_sock handling and moved |
1da177e4 LT |
40 | * most of it into the af independent code. |
41 | * Added tail drop and some other bugfixes. | |
caa20d9a | 42 | * Added new listen semantics. |
1da177e4 LT |
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> | |
304a1618 | 67 | #include <net/inet_hashtables.h> |
1da177e4 | 68 | #include <net/tcp.h> |
20380731 | 69 | #include <net/transp_v6.h> |
1da177e4 LT |
70 | #include <net/ipv6.h> |
71 | #include <net/inet_common.h> | |
72 | #include <net/xfrm.h> | |
73 | ||
74 | #include <linux/inet.h> | |
75 | #include <linux/ipv6.h> | |
76 | #include <linux/stddef.h> | |
77 | #include <linux/proc_fs.h> | |
78 | #include <linux/seq_file.h> | |
79 | ||
1da177e4 LT |
80 | int sysctl_tcp_tw_reuse; |
81 | int sysctl_tcp_low_latency; | |
82 | ||
83 | /* Check TCP sequence numbers in ICMP packets. */ | |
84 | #define ICMP_MIN_LENGTH 8 | |
85 | ||
86 | /* Socket used for sending RSTs */ | |
87 | static struct socket *tcp_socket; | |
88 | ||
89 | void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len, | |
90 | struct sk_buff *skb); | |
91 | ||
0f7ff927 ACM |
92 | struct inet_hashinfo __cacheline_aligned tcp_hashinfo = { |
93 | .lhash_lock = RW_LOCK_UNLOCKED, | |
94 | .lhash_users = ATOMIC_INIT(0), | |
95 | .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait), | |
1da177e4 LT |
96 | }; |
97 | ||
463c84b9 ACM |
98 | static int tcp_v4_get_port(struct sock *sk, unsigned short snum) |
99 | { | |
100 | return inet_csk_get_port(&tcp_hashinfo, sk, snum); | |
101 | } | |
102 | ||
1da177e4 LT |
103 | static void tcp_v4_hash(struct sock *sk) |
104 | { | |
81849d10 | 105 | inet_hash(&tcp_hashinfo, sk); |
1da177e4 LT |
106 | } |
107 | ||
108 | void tcp_unhash(struct sock *sk) | |
109 | { | |
81849d10 | 110 | inet_unhash(&tcp_hashinfo, sk); |
1da177e4 LT |
111 | } |
112 | ||
1da177e4 LT |
113 | static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb) |
114 | { | |
115 | return secure_tcp_sequence_number(skb->nh.iph->daddr, | |
116 | skb->nh.iph->saddr, | |
117 | skb->h.th->dest, | |
118 | skb->h.th->source); | |
119 | } | |
120 | ||
121 | /* called with local bh disabled */ | |
122 | static int __tcp_v4_check_established(struct sock *sk, __u16 lport, | |
8feaf0c0 | 123 | struct inet_timewait_sock **twp) |
1da177e4 LT |
124 | { |
125 | struct inet_sock *inet = inet_sk(sk); | |
126 | u32 daddr = inet->rcv_saddr; | |
127 | u32 saddr = inet->daddr; | |
128 | int dif = sk->sk_bound_dev_if; | |
8feaf0c0 ACM |
129 | INET_ADDR_COOKIE(acookie, saddr, daddr) |
130 | const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport); | |
81c3d547 ED |
131 | unsigned int hash = inet_ehashfn(daddr, lport, saddr, inet->dport); |
132 | struct inet_ehash_bucket *head = inet_ehash_bucket(&tcp_hashinfo, hash); | |
1da177e4 | 133 | struct sock *sk2; |
8feaf0c0 ACM |
134 | const struct hlist_node *node; |
135 | struct inet_timewait_sock *tw; | |
1da177e4 | 136 | |
81c3d547 | 137 | prefetch(head->chain.first); |
1da177e4 LT |
138 | write_lock(&head->lock); |
139 | ||
140 | /* Check TIME-WAIT sockets first. */ | |
6e04e021 | 141 | sk_for_each(sk2, node, &(head + tcp_hashinfo.ehash_size)->chain) { |
8feaf0c0 | 142 | tw = inet_twsk(sk2); |
1da177e4 | 143 | |
81c3d547 | 144 | if (INET_TW_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif)) { |
8feaf0c0 | 145 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sk2); |
1da177e4 LT |
146 | struct tcp_sock *tp = tcp_sk(sk); |
147 | ||
148 | /* With PAWS, it is safe from the viewpoint | |
149 | of data integrity. Even without PAWS it | |
150 | is safe provided sequence spaces do not | |
151 | overlap i.e. at data rates <= 80Mbit/sec. | |
152 | ||
153 | Actually, the idea is close to VJ's one, | |
154 | only timestamp cache is held not per host, | |
155 | but per port pair and TW bucket is used | |
156 | as state holder. | |
157 | ||
158 | If TW bucket has been already destroyed we | |
159 | fall back to VJ's scheme and use initial | |
160 | timestamp retrieved from peer table. | |
161 | */ | |
8feaf0c0 | 162 | if (tcptw->tw_ts_recent_stamp && |
1da177e4 LT |
163 | (!twp || (sysctl_tcp_tw_reuse && |
164 | xtime.tv_sec - | |
8feaf0c0 ACM |
165 | tcptw->tw_ts_recent_stamp > 1))) { |
166 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; | |
167 | if (tp->write_seq == 0) | |
1da177e4 | 168 | tp->write_seq = 1; |
8feaf0c0 ACM |
169 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; |
170 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
1da177e4 LT |
171 | sock_hold(sk2); |
172 | goto unique; | |
173 | } else | |
174 | goto not_unique; | |
175 | } | |
176 | } | |
177 | tw = NULL; | |
178 | ||
179 | /* And established part... */ | |
180 | sk_for_each(sk2, node, &head->chain) { | |
81c3d547 | 181 | if (INET_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif)) |
1da177e4 LT |
182 | goto not_unique; |
183 | } | |
184 | ||
185 | unique: | |
186 | /* Must record num and sport now. Otherwise we will see | |
187 | * in hash table socket with a funny identity. */ | |
188 | inet->num = lport; | |
189 | inet->sport = htons(lport); | |
81c3d547 | 190 | sk->sk_hash = hash; |
1da177e4 LT |
191 | BUG_TRAP(sk_unhashed(sk)); |
192 | __sk_add_node(sk, &head->chain); | |
193 | sock_prot_inc_use(sk->sk_prot); | |
194 | write_unlock(&head->lock); | |
195 | ||
196 | if (twp) { | |
197 | *twp = tw; | |
198 | NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED); | |
199 | } else if (tw) { | |
200 | /* Silly. Should hash-dance instead... */ | |
295ff7ed | 201 | inet_twsk_deschedule(tw, &tcp_death_row); |
1da177e4 LT |
202 | NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED); |
203 | ||
8feaf0c0 | 204 | inet_twsk_put(tw); |
1da177e4 LT |
205 | } |
206 | ||
207 | return 0; | |
208 | ||
209 | not_unique: | |
210 | write_unlock(&head->lock); | |
211 | return -EADDRNOTAVAIL; | |
212 | } | |
213 | ||
214 | static inline u32 connect_port_offset(const struct sock *sk) | |
215 | { | |
216 | const struct inet_sock *inet = inet_sk(sk); | |
217 | ||
218 | return secure_tcp_port_ephemeral(inet->rcv_saddr, inet->daddr, | |
219 | inet->dport); | |
220 | } | |
221 | ||
222 | /* | |
223 | * Bind a port for a connect operation and hash it. | |
224 | */ | |
225 | static inline int tcp_v4_hash_connect(struct sock *sk) | |
226 | { | |
0f7ff927 ACM |
227 | const unsigned short snum = inet_sk(sk)->num; |
228 | struct inet_bind_hashbucket *head; | |
229 | struct inet_bind_bucket *tb; | |
1da177e4 LT |
230 | int ret; |
231 | ||
232 | if (!snum) { | |
233 | int low = sysctl_local_port_range[0]; | |
234 | int high = sysctl_local_port_range[1]; | |
235 | int range = high - low; | |
236 | int i; | |
237 | int port; | |
238 | static u32 hint; | |
239 | u32 offset = hint + connect_port_offset(sk); | |
240 | struct hlist_node *node; | |
8feaf0c0 | 241 | struct inet_timewait_sock *tw = NULL; |
1da177e4 LT |
242 | |
243 | local_bh_disable(); | |
244 | for (i = 1; i <= range; i++) { | |
245 | port = low + (i + offset) % range; | |
6e04e021 | 246 | head = &tcp_hashinfo.bhash[inet_bhashfn(port, tcp_hashinfo.bhash_size)]; |
1da177e4 LT |
247 | spin_lock(&head->lock); |
248 | ||
249 | /* Does not bother with rcv_saddr checks, | |
250 | * because the established check is already | |
251 | * unique enough. | |
252 | */ | |
0f7ff927 | 253 | inet_bind_bucket_for_each(tb, node, &head->chain) { |
1da177e4 LT |
254 | if (tb->port == port) { |
255 | BUG_TRAP(!hlist_empty(&tb->owners)); | |
256 | if (tb->fastreuse >= 0) | |
257 | goto next_port; | |
258 | if (!__tcp_v4_check_established(sk, | |
259 | port, | |
260 | &tw)) | |
261 | goto ok; | |
262 | goto next_port; | |
263 | } | |
264 | } | |
265 | ||
6e04e021 | 266 | tb = inet_bind_bucket_create(tcp_hashinfo.bind_bucket_cachep, head, port); |
1da177e4 LT |
267 | if (!tb) { |
268 | spin_unlock(&head->lock); | |
269 | break; | |
270 | } | |
271 | tb->fastreuse = -1; | |
272 | goto ok; | |
273 | ||
274 | next_port: | |
275 | spin_unlock(&head->lock); | |
276 | } | |
277 | local_bh_enable(); | |
278 | ||
279 | return -EADDRNOTAVAIL; | |
280 | ||
281 | ok: | |
282 | hint += i; | |
283 | ||
284 | /* Head lock still held and bh's disabled */ | |
2d8c4ce5 | 285 | inet_bind_hash(sk, tb, port); |
1da177e4 LT |
286 | if (sk_unhashed(sk)) { |
287 | inet_sk(sk)->sport = htons(port); | |
f3f05f70 | 288 | __inet_hash(&tcp_hashinfo, sk, 0); |
1da177e4 LT |
289 | } |
290 | spin_unlock(&head->lock); | |
291 | ||
292 | if (tw) { | |
295ff7ed | 293 | inet_twsk_deschedule(tw, &tcp_death_row);; |
8feaf0c0 | 294 | inet_twsk_put(tw); |
1da177e4 LT |
295 | } |
296 | ||
297 | ret = 0; | |
298 | goto out; | |
299 | } | |
300 | ||
6e04e021 | 301 | head = &tcp_hashinfo.bhash[inet_bhashfn(snum, tcp_hashinfo.bhash_size)]; |
463c84b9 | 302 | tb = inet_csk(sk)->icsk_bind_hash; |
1da177e4 LT |
303 | spin_lock_bh(&head->lock); |
304 | if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) { | |
f3f05f70 | 305 | __inet_hash(&tcp_hashinfo, sk, 0); |
1da177e4 LT |
306 | spin_unlock_bh(&head->lock); |
307 | return 0; | |
308 | } else { | |
309 | spin_unlock(&head->lock); | |
310 | /* No definite answer... Walk to established hash table */ | |
311 | ret = __tcp_v4_check_established(sk, snum, NULL); | |
312 | out: | |
313 | local_bh_enable(); | |
314 | return ret; | |
315 | } | |
316 | } | |
317 | ||
318 | /* This will initiate an outgoing connection. */ | |
319 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
320 | { | |
321 | struct inet_sock *inet = inet_sk(sk); | |
322 | struct tcp_sock *tp = tcp_sk(sk); | |
323 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; | |
324 | struct rtable *rt; | |
325 | u32 daddr, nexthop; | |
326 | int tmp; | |
327 | int err; | |
328 | ||
329 | if (addr_len < sizeof(struct sockaddr_in)) | |
330 | return -EINVAL; | |
331 | ||
332 | if (usin->sin_family != AF_INET) | |
333 | return -EAFNOSUPPORT; | |
334 | ||
335 | nexthop = daddr = usin->sin_addr.s_addr; | |
336 | if (inet->opt && inet->opt->srr) { | |
337 | if (!daddr) | |
338 | return -EINVAL; | |
339 | nexthop = inet->opt->faddr; | |
340 | } | |
341 | ||
342 | tmp = ip_route_connect(&rt, nexthop, inet->saddr, | |
343 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, | |
344 | IPPROTO_TCP, | |
345 | inet->sport, usin->sin_port, sk); | |
346 | if (tmp < 0) | |
347 | return tmp; | |
348 | ||
349 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
350 | ip_rt_put(rt); | |
351 | return -ENETUNREACH; | |
352 | } | |
353 | ||
354 | if (!inet->opt || !inet->opt->srr) | |
355 | daddr = rt->rt_dst; | |
356 | ||
357 | if (!inet->saddr) | |
358 | inet->saddr = rt->rt_src; | |
359 | inet->rcv_saddr = inet->saddr; | |
360 | ||
361 | if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) { | |
362 | /* Reset inherited state */ | |
363 | tp->rx_opt.ts_recent = 0; | |
364 | tp->rx_opt.ts_recent_stamp = 0; | |
365 | tp->write_seq = 0; | |
366 | } | |
367 | ||
295ff7ed | 368 | if (tcp_death_row.sysctl_tw_recycle && |
1da177e4 LT |
369 | !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) { |
370 | struct inet_peer *peer = rt_get_peer(rt); | |
371 | ||
372 | /* VJ's idea. We save last timestamp seen from | |
373 | * the destination in peer table, when entering state TIME-WAIT | |
374 | * and initialize rx_opt.ts_recent from it, when trying new connection. | |
375 | */ | |
376 | ||
377 | if (peer && peer->tcp_ts_stamp + TCP_PAWS_MSL >= xtime.tv_sec) { | |
378 | tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp; | |
379 | tp->rx_opt.ts_recent = peer->tcp_ts; | |
380 | } | |
381 | } | |
382 | ||
383 | inet->dport = usin->sin_port; | |
384 | inet->daddr = daddr; | |
385 | ||
386 | tp->ext_header_len = 0; | |
387 | if (inet->opt) | |
388 | tp->ext_header_len = inet->opt->optlen; | |
389 | ||
390 | tp->rx_opt.mss_clamp = 536; | |
391 | ||
392 | /* Socket identity is still unknown (sport may be zero). | |
393 | * However we set state to SYN-SENT and not releasing socket | |
394 | * lock select source port, enter ourselves into the hash tables and | |
395 | * complete initialization after this. | |
396 | */ | |
397 | tcp_set_state(sk, TCP_SYN_SENT); | |
398 | err = tcp_v4_hash_connect(sk); | |
399 | if (err) | |
400 | goto failure; | |
401 | ||
402 | err = ip_route_newports(&rt, inet->sport, inet->dport, sk); | |
403 | if (err) | |
404 | goto failure; | |
405 | ||
406 | /* OK, now commit destination to socket. */ | |
6cbb0df7 | 407 | sk_setup_caps(sk, &rt->u.dst); |
1da177e4 LT |
408 | |
409 | if (!tp->write_seq) | |
410 | tp->write_seq = secure_tcp_sequence_number(inet->saddr, | |
411 | inet->daddr, | |
412 | inet->sport, | |
413 | usin->sin_port); | |
414 | ||
415 | inet->id = tp->write_seq ^ jiffies; | |
416 | ||
417 | err = tcp_connect(sk); | |
418 | rt = NULL; | |
419 | if (err) | |
420 | goto failure; | |
421 | ||
422 | return 0; | |
423 | ||
424 | failure: | |
425 | /* This unhashes the socket and releases the local port, if necessary. */ | |
426 | tcp_set_state(sk, TCP_CLOSE); | |
427 | ip_rt_put(rt); | |
428 | sk->sk_route_caps = 0; | |
429 | inet->dport = 0; | |
430 | return err; | |
431 | } | |
432 | ||
1da177e4 LT |
433 | /* |
434 | * This routine does path mtu discovery as defined in RFC1191. | |
435 | */ | |
436 | static inline void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, | |
437 | u32 mtu) | |
438 | { | |
439 | struct dst_entry *dst; | |
440 | struct inet_sock *inet = inet_sk(sk); | |
441 | struct tcp_sock *tp = tcp_sk(sk); | |
442 | ||
443 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs | |
444 | * send out by Linux are always <576bytes so they should go through | |
445 | * unfragmented). | |
446 | */ | |
447 | if (sk->sk_state == TCP_LISTEN) | |
448 | return; | |
449 | ||
450 | /* We don't check in the destentry if pmtu discovery is forbidden | |
451 | * on this route. We just assume that no packet_to_big packets | |
452 | * are send back when pmtu discovery is not active. | |
453 | * There is a small race when the user changes this flag in the | |
454 | * route, but I think that's acceptable. | |
455 | */ | |
456 | if ((dst = __sk_dst_check(sk, 0)) == NULL) | |
457 | return; | |
458 | ||
459 | dst->ops->update_pmtu(dst, mtu); | |
460 | ||
461 | /* Something is about to be wrong... Remember soft error | |
462 | * for the case, if this connection will not able to recover. | |
463 | */ | |
464 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
465 | sk->sk_err_soft = EMSGSIZE; | |
466 | ||
467 | mtu = dst_mtu(dst); | |
468 | ||
469 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
470 | tp->pmtu_cookie > mtu) { | |
471 | tcp_sync_mss(sk, mtu); | |
472 | ||
473 | /* Resend the TCP packet because it's | |
474 | * clear that the old packet has been | |
475 | * dropped. This is the new "fast" path mtu | |
476 | * discovery. | |
477 | */ | |
478 | tcp_simple_retransmit(sk); | |
479 | } /* else let the usual retransmit timer handle it */ | |
480 | } | |
481 | ||
482 | /* | |
483 | * This routine is called by the ICMP module when it gets some | |
484 | * sort of error condition. If err < 0 then the socket should | |
485 | * be closed and the error returned to the user. If err > 0 | |
486 | * it's just the icmp type << 8 | icmp code. After adjustment | |
487 | * header points to the first 8 bytes of the tcp header. We need | |
488 | * to find the appropriate port. | |
489 | * | |
490 | * The locking strategy used here is very "optimistic". When | |
491 | * someone else accesses the socket the ICMP is just dropped | |
492 | * and for some paths there is no check at all. | |
493 | * A more general error queue to queue errors for later handling | |
494 | * is probably better. | |
495 | * | |
496 | */ | |
497 | ||
498 | void tcp_v4_err(struct sk_buff *skb, u32 info) | |
499 | { | |
500 | struct iphdr *iph = (struct iphdr *)skb->data; | |
501 | struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2)); | |
502 | struct tcp_sock *tp; | |
503 | struct inet_sock *inet; | |
504 | int type = skb->h.icmph->type; | |
505 | int code = skb->h.icmph->code; | |
506 | struct sock *sk; | |
507 | __u32 seq; | |
508 | int err; | |
509 | ||
510 | if (skb->len < (iph->ihl << 2) + 8) { | |
511 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
512 | return; | |
513 | } | |
514 | ||
e48c414e | 515 | sk = inet_lookup(&tcp_hashinfo, iph->daddr, th->dest, iph->saddr, |
463c84b9 | 516 | th->source, inet_iif(skb)); |
1da177e4 LT |
517 | if (!sk) { |
518 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
519 | return; | |
520 | } | |
521 | if (sk->sk_state == TCP_TIME_WAIT) { | |
8feaf0c0 | 522 | inet_twsk_put((struct inet_timewait_sock *)sk); |
1da177e4 LT |
523 | return; |
524 | } | |
525 | ||
526 | bh_lock_sock(sk); | |
527 | /* If too many ICMPs get dropped on busy | |
528 | * servers this needs to be solved differently. | |
529 | */ | |
530 | if (sock_owned_by_user(sk)) | |
531 | NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS); | |
532 | ||
533 | if (sk->sk_state == TCP_CLOSE) | |
534 | goto out; | |
535 | ||
536 | tp = tcp_sk(sk); | |
537 | seq = ntohl(th->seq); | |
538 | if (sk->sk_state != TCP_LISTEN && | |
539 | !between(seq, tp->snd_una, tp->snd_nxt)) { | |
540 | NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS); | |
541 | goto out; | |
542 | } | |
543 | ||
544 | switch (type) { | |
545 | case ICMP_SOURCE_QUENCH: | |
546 | /* Just silently ignore these. */ | |
547 | goto out; | |
548 | case ICMP_PARAMETERPROB: | |
549 | err = EPROTO; | |
550 | break; | |
551 | case ICMP_DEST_UNREACH: | |
552 | if (code > NR_ICMP_UNREACH) | |
553 | goto out; | |
554 | ||
555 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
556 | if (!sock_owned_by_user(sk)) | |
557 | do_pmtu_discovery(sk, iph, info); | |
558 | goto out; | |
559 | } | |
560 | ||
561 | err = icmp_err_convert[code].errno; | |
562 | break; | |
563 | case ICMP_TIME_EXCEEDED: | |
564 | err = EHOSTUNREACH; | |
565 | break; | |
566 | default: | |
567 | goto out; | |
568 | } | |
569 | ||
570 | switch (sk->sk_state) { | |
60236fdd | 571 | struct request_sock *req, **prev; |
1da177e4 LT |
572 | case TCP_LISTEN: |
573 | if (sock_owned_by_user(sk)) | |
574 | goto out; | |
575 | ||
463c84b9 ACM |
576 | req = inet_csk_search_req(sk, &prev, th->dest, |
577 | iph->daddr, iph->saddr); | |
1da177e4 LT |
578 | if (!req) |
579 | goto out; | |
580 | ||
581 | /* ICMPs are not backlogged, hence we cannot get | |
582 | an established socket here. | |
583 | */ | |
584 | BUG_TRAP(!req->sk); | |
585 | ||
2e6599cb | 586 | if (seq != tcp_rsk(req)->snt_isn) { |
1da177e4 LT |
587 | NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS); |
588 | goto out; | |
589 | } | |
590 | ||
591 | /* | |
592 | * Still in SYN_RECV, just remove it silently. | |
593 | * There is no good way to pass the error to the newly | |
594 | * created socket, and POSIX does not want network | |
595 | * errors returned from accept(). | |
596 | */ | |
463c84b9 | 597 | inet_csk_reqsk_queue_drop(sk, req, prev); |
1da177e4 LT |
598 | goto out; |
599 | ||
600 | case TCP_SYN_SENT: | |
601 | case TCP_SYN_RECV: /* Cannot happen. | |
602 | It can f.e. if SYNs crossed. | |
603 | */ | |
604 | if (!sock_owned_by_user(sk)) { | |
605 | TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); | |
606 | sk->sk_err = err; | |
607 | ||
608 | sk->sk_error_report(sk); | |
609 | ||
610 | tcp_done(sk); | |
611 | } else { | |
612 | sk->sk_err_soft = err; | |
613 | } | |
614 | goto out; | |
615 | } | |
616 | ||
617 | /* If we've already connected we will keep trying | |
618 | * until we time out, or the user gives up. | |
619 | * | |
620 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
621 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
622 | * but it is obsoleted by pmtu discovery). | |
623 | * | |
624 | * Note, that in modern internet, where routing is unreliable | |
625 | * and in each dark corner broken firewalls sit, sending random | |
626 | * errors ordered by their masters even this two messages finally lose | |
627 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
628 | * | |
629 | * Now we are in compliance with RFCs. | |
630 | * --ANK (980905) | |
631 | */ | |
632 | ||
633 | inet = inet_sk(sk); | |
634 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
635 | sk->sk_err = err; | |
636 | sk->sk_error_report(sk); | |
637 | } else { /* Only an error on timeout */ | |
638 | sk->sk_err_soft = err; | |
639 | } | |
640 | ||
641 | out: | |
642 | bh_unlock_sock(sk); | |
643 | sock_put(sk); | |
644 | } | |
645 | ||
646 | /* This routine computes an IPv4 TCP checksum. */ | |
647 | void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len, | |
648 | struct sk_buff *skb) | |
649 | { | |
650 | struct inet_sock *inet = inet_sk(sk); | |
651 | ||
652 | if (skb->ip_summed == CHECKSUM_HW) { | |
653 | th->check = ~tcp_v4_check(th, len, inet->saddr, inet->daddr, 0); | |
654 | skb->csum = offsetof(struct tcphdr, check); | |
655 | } else { | |
656 | th->check = tcp_v4_check(th, len, inet->saddr, inet->daddr, | |
657 | csum_partial((char *)th, | |
658 | th->doff << 2, | |
659 | skb->csum)); | |
660 | } | |
661 | } | |
662 | ||
663 | /* | |
664 | * This routine will send an RST to the other tcp. | |
665 | * | |
666 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
667 | * for reset. | |
668 | * Answer: if a packet caused RST, it is not for a socket | |
669 | * existing in our system, if it is matched to a socket, | |
670 | * it is just duplicate segment or bug in other side's TCP. | |
671 | * So that we build reply only basing on parameters | |
672 | * arrived with segment. | |
673 | * Exception: precedence violation. We do not implement it in any case. | |
674 | */ | |
675 | ||
676 | static void tcp_v4_send_reset(struct sk_buff *skb) | |
677 | { | |
678 | struct tcphdr *th = skb->h.th; | |
679 | struct tcphdr rth; | |
680 | struct ip_reply_arg arg; | |
681 | ||
682 | /* Never send a reset in response to a reset. */ | |
683 | if (th->rst) | |
684 | return; | |
685 | ||
686 | if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL) | |
687 | return; | |
688 | ||
689 | /* Swap the send and the receive. */ | |
690 | memset(&rth, 0, sizeof(struct tcphdr)); | |
691 | rth.dest = th->source; | |
692 | rth.source = th->dest; | |
693 | rth.doff = sizeof(struct tcphdr) / 4; | |
694 | rth.rst = 1; | |
695 | ||
696 | if (th->ack) { | |
697 | rth.seq = th->ack_seq; | |
698 | } else { | |
699 | rth.ack = 1; | |
700 | rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
701 | skb->len - (th->doff << 2)); | |
702 | } | |
703 | ||
704 | memset(&arg, 0, sizeof arg); | |
705 | arg.iov[0].iov_base = (unsigned char *)&rth; | |
706 | arg.iov[0].iov_len = sizeof rth; | |
707 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, | |
708 | skb->nh.iph->saddr, /*XXX*/ | |
709 | sizeof(struct tcphdr), IPPROTO_TCP, 0); | |
710 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
711 | ||
712 | ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth); | |
713 | ||
714 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
715 | TCP_INC_STATS_BH(TCP_MIB_OUTRSTS); | |
716 | } | |
717 | ||
718 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
719 | outside socket context is ugly, certainly. What can I do? | |
720 | */ | |
721 | ||
722 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, | |
723 | u32 win, u32 ts) | |
724 | { | |
725 | struct tcphdr *th = skb->h.th; | |
726 | struct { | |
727 | struct tcphdr th; | |
728 | u32 tsopt[3]; | |
729 | } rep; | |
730 | struct ip_reply_arg arg; | |
731 | ||
732 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
733 | memset(&arg, 0, sizeof arg); | |
734 | ||
735 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
736 | arg.iov[0].iov_len = sizeof(rep.th); | |
737 | if (ts) { | |
738 | rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
739 | (TCPOPT_TIMESTAMP << 8) | | |
740 | TCPOLEN_TIMESTAMP); | |
741 | rep.tsopt[1] = htonl(tcp_time_stamp); | |
742 | rep.tsopt[2] = htonl(ts); | |
743 | arg.iov[0].iov_len = sizeof(rep); | |
744 | } | |
745 | ||
746 | /* Swap the send and the receive. */ | |
747 | rep.th.dest = th->source; | |
748 | rep.th.source = th->dest; | |
749 | rep.th.doff = arg.iov[0].iov_len / 4; | |
750 | rep.th.seq = htonl(seq); | |
751 | rep.th.ack_seq = htonl(ack); | |
752 | rep.th.ack = 1; | |
753 | rep.th.window = htons(win); | |
754 | ||
755 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, | |
756 | skb->nh.iph->saddr, /*XXX*/ | |
757 | arg.iov[0].iov_len, IPPROTO_TCP, 0); | |
758 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
759 | ||
760 | ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len); | |
761 | ||
762 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
763 | } | |
764 | ||
765 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
766 | { | |
8feaf0c0 ACM |
767 | struct inet_timewait_sock *tw = inet_twsk(sk); |
768 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sk); | |
1da177e4 | 769 | |
8feaf0c0 ACM |
770 | tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
771 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, tcptw->tw_ts_recent); | |
1da177e4 | 772 | |
8feaf0c0 | 773 | inet_twsk_put(tw); |
1da177e4 LT |
774 | } |
775 | ||
60236fdd | 776 | static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req) |
1da177e4 | 777 | { |
2e6599cb | 778 | tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, |
1da177e4 LT |
779 | req->ts_recent); |
780 | } | |
781 | ||
1da177e4 LT |
782 | /* |
783 | * Send a SYN-ACK after having received an ACK. | |
60236fdd | 784 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
785 | * socket. |
786 | */ | |
60236fdd | 787 | static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req, |
1da177e4 LT |
788 | struct dst_entry *dst) |
789 | { | |
2e6599cb | 790 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
791 | int err = -1; |
792 | struct sk_buff * skb; | |
793 | ||
794 | /* First, grab a route. */ | |
463c84b9 | 795 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
1da177e4 LT |
796 | goto out; |
797 | ||
798 | skb = tcp_make_synack(sk, dst, req); | |
799 | ||
800 | if (skb) { | |
801 | struct tcphdr *th = skb->h.th; | |
802 | ||
803 | th->check = tcp_v4_check(th, skb->len, | |
2e6599cb ACM |
804 | ireq->loc_addr, |
805 | ireq->rmt_addr, | |
1da177e4 LT |
806 | csum_partial((char *)th, skb->len, |
807 | skb->csum)); | |
808 | ||
2e6599cb ACM |
809 | err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, |
810 | ireq->rmt_addr, | |
811 | ireq->opt); | |
1da177e4 LT |
812 | if (err == NET_XMIT_CN) |
813 | err = 0; | |
814 | } | |
815 | ||
816 | out: | |
817 | dst_release(dst); | |
818 | return err; | |
819 | } | |
820 | ||
821 | /* | |
60236fdd | 822 | * IPv4 request_sock destructor. |
1da177e4 | 823 | */ |
60236fdd | 824 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 825 | { |
a51482bd | 826 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
827 | } |
828 | ||
829 | static inline void syn_flood_warning(struct sk_buff *skb) | |
830 | { | |
831 | static unsigned long warntime; | |
832 | ||
833 | if (time_after(jiffies, (warntime + HZ * 60))) { | |
834 | warntime = jiffies; | |
835 | printk(KERN_INFO | |
836 | "possible SYN flooding on port %d. Sending cookies.\n", | |
837 | ntohs(skb->h.th->dest)); | |
838 | } | |
839 | } | |
840 | ||
841 | /* | |
60236fdd | 842 | * Save and compile IPv4 options into the request_sock if needed. |
1da177e4 LT |
843 | */ |
844 | static inline struct ip_options *tcp_v4_save_options(struct sock *sk, | |
845 | struct sk_buff *skb) | |
846 | { | |
847 | struct ip_options *opt = &(IPCB(skb)->opt); | |
848 | struct ip_options *dopt = NULL; | |
849 | ||
850 | if (opt && opt->optlen) { | |
851 | int opt_size = optlength(opt); | |
852 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
853 | if (dopt) { | |
854 | if (ip_options_echo(dopt, skb)) { | |
855 | kfree(dopt); | |
856 | dopt = NULL; | |
857 | } | |
858 | } | |
859 | } | |
860 | return dopt; | |
861 | } | |
862 | ||
60236fdd | 863 | struct request_sock_ops tcp_request_sock_ops = { |
1da177e4 | 864 | .family = PF_INET, |
2e6599cb | 865 | .obj_size = sizeof(struct tcp_request_sock), |
1da177e4 | 866 | .rtx_syn_ack = tcp_v4_send_synack, |
60236fdd ACM |
867 | .send_ack = tcp_v4_reqsk_send_ack, |
868 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 LT |
869 | .send_reset = tcp_v4_send_reset, |
870 | }; | |
871 | ||
872 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) | |
873 | { | |
2e6599cb | 874 | struct inet_request_sock *ireq; |
1da177e4 | 875 | struct tcp_options_received tmp_opt; |
60236fdd | 876 | struct request_sock *req; |
1da177e4 LT |
877 | __u32 saddr = skb->nh.iph->saddr; |
878 | __u32 daddr = skb->nh.iph->daddr; | |
879 | __u32 isn = TCP_SKB_CB(skb)->when; | |
880 | struct dst_entry *dst = NULL; | |
881 | #ifdef CONFIG_SYN_COOKIES | |
882 | int want_cookie = 0; | |
883 | #else | |
884 | #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */ | |
885 | #endif | |
886 | ||
887 | /* Never answer to SYNs send to broadcast or multicast */ | |
888 | if (((struct rtable *)skb->dst)->rt_flags & | |
889 | (RTCF_BROADCAST | RTCF_MULTICAST)) | |
890 | goto drop; | |
891 | ||
892 | /* TW buckets are converted to open requests without | |
893 | * limitations, they conserve resources and peer is | |
894 | * evidently real one. | |
895 | */ | |
463c84b9 | 896 | if (inet_csk_reqsk_queue_is_full(sk) && !isn) { |
1da177e4 LT |
897 | #ifdef CONFIG_SYN_COOKIES |
898 | if (sysctl_tcp_syncookies) { | |
899 | want_cookie = 1; | |
900 | } else | |
901 | #endif | |
902 | goto drop; | |
903 | } | |
904 | ||
905 | /* Accept backlog is full. If we have already queued enough | |
906 | * of warm entries in syn queue, drop request. It is better than | |
907 | * clogging syn queue with openreqs with exponentially increasing | |
908 | * timeout. | |
909 | */ | |
463c84b9 | 910 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) |
1da177e4 LT |
911 | goto drop; |
912 | ||
60236fdd | 913 | req = reqsk_alloc(&tcp_request_sock_ops); |
1da177e4 LT |
914 | if (!req) |
915 | goto drop; | |
916 | ||
917 | tcp_clear_options(&tmp_opt); | |
918 | tmp_opt.mss_clamp = 536; | |
919 | tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss; | |
920 | ||
921 | tcp_parse_options(skb, &tmp_opt, 0); | |
922 | ||
923 | if (want_cookie) { | |
924 | tcp_clear_options(&tmp_opt); | |
925 | tmp_opt.saw_tstamp = 0; | |
926 | } | |
927 | ||
928 | if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) { | |
929 | /* Some OSes (unknown ones, but I see them on web server, which | |
930 | * contains information interesting only for windows' | |
931 | * users) do not send their stamp in SYN. It is easy case. | |
932 | * We simply do not advertise TS support. | |
933 | */ | |
934 | tmp_opt.saw_tstamp = 0; | |
935 | tmp_opt.tstamp_ok = 0; | |
936 | } | |
937 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; | |
938 | ||
939 | tcp_openreq_init(req, &tmp_opt, skb); | |
940 | ||
2e6599cb ACM |
941 | ireq = inet_rsk(req); |
942 | ireq->loc_addr = daddr; | |
943 | ireq->rmt_addr = saddr; | |
944 | ireq->opt = tcp_v4_save_options(sk, skb); | |
1da177e4 LT |
945 | if (!want_cookie) |
946 | TCP_ECN_create_request(req, skb->h.th); | |
947 | ||
948 | if (want_cookie) { | |
949 | #ifdef CONFIG_SYN_COOKIES | |
950 | syn_flood_warning(skb); | |
951 | #endif | |
952 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); | |
953 | } else if (!isn) { | |
954 | struct inet_peer *peer = NULL; | |
955 | ||
956 | /* VJ's idea. We save last timestamp seen | |
957 | * from the destination in peer table, when entering | |
958 | * state TIME-WAIT, and check against it before | |
959 | * accepting new connection request. | |
960 | * | |
961 | * If "isn" is not zero, this request hit alive | |
962 | * timewait bucket, so that all the necessary checks | |
963 | * are made in the function processing timewait state. | |
964 | */ | |
965 | if (tmp_opt.saw_tstamp && | |
295ff7ed | 966 | tcp_death_row.sysctl_tw_recycle && |
463c84b9 | 967 | (dst = inet_csk_route_req(sk, req)) != NULL && |
1da177e4 LT |
968 | (peer = rt_get_peer((struct rtable *)dst)) != NULL && |
969 | peer->v4daddr == saddr) { | |
970 | if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL && | |
971 | (s32)(peer->tcp_ts - req->ts_recent) > | |
972 | TCP_PAWS_WINDOW) { | |
973 | NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED); | |
974 | dst_release(dst); | |
975 | goto drop_and_free; | |
976 | } | |
977 | } | |
978 | /* Kill the following clause, if you dislike this way. */ | |
979 | else if (!sysctl_tcp_syncookies && | |
463c84b9 | 980 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
1da177e4 LT |
981 | (sysctl_max_syn_backlog >> 2)) && |
982 | (!peer || !peer->tcp_ts_stamp) && | |
983 | (!dst || !dst_metric(dst, RTAX_RTT))) { | |
984 | /* Without syncookies last quarter of | |
985 | * backlog is filled with destinations, | |
986 | * proven to be alive. | |
987 | * It means that we continue to communicate | |
988 | * to destinations, already remembered | |
989 | * to the moment of synflood. | |
990 | */ | |
64ce2073 PM |
991 | LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open " |
992 | "request from %u.%u.%u.%u/%u\n", | |
993 | NIPQUAD(saddr), | |
994 | ntohs(skb->h.th->source)); | |
1da177e4 LT |
995 | dst_release(dst); |
996 | goto drop_and_free; | |
997 | } | |
998 | ||
999 | isn = tcp_v4_init_sequence(sk, skb); | |
1000 | } | |
2e6599cb | 1001 | tcp_rsk(req)->snt_isn = isn; |
1da177e4 LT |
1002 | |
1003 | if (tcp_v4_send_synack(sk, req, dst)) | |
1004 | goto drop_and_free; | |
1005 | ||
1006 | if (want_cookie) { | |
60236fdd | 1007 | reqsk_free(req); |
1da177e4 | 1008 | } else { |
3f421baa | 1009 | inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
1da177e4 LT |
1010 | } |
1011 | return 0; | |
1012 | ||
1013 | drop_and_free: | |
60236fdd | 1014 | reqsk_free(req); |
1da177e4 LT |
1015 | drop: |
1016 | TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); | |
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | ||
1021 | /* | |
1022 | * The three way handshake has completed - we got a valid synack - | |
1023 | * now create the new socket. | |
1024 | */ | |
1025 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
60236fdd | 1026 | struct request_sock *req, |
1da177e4 LT |
1027 | struct dst_entry *dst) |
1028 | { | |
2e6599cb | 1029 | struct inet_request_sock *ireq; |
1da177e4 LT |
1030 | struct inet_sock *newinet; |
1031 | struct tcp_sock *newtp; | |
1032 | struct sock *newsk; | |
1033 | ||
1034 | if (sk_acceptq_is_full(sk)) | |
1035 | goto exit_overflow; | |
1036 | ||
463c84b9 | 1037 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
1da177e4 LT |
1038 | goto exit; |
1039 | ||
1040 | newsk = tcp_create_openreq_child(sk, req, skb); | |
1041 | if (!newsk) | |
1042 | goto exit; | |
1043 | ||
6cbb0df7 | 1044 | sk_setup_caps(newsk, dst); |
1da177e4 LT |
1045 | |
1046 | newtp = tcp_sk(newsk); | |
1047 | newinet = inet_sk(newsk); | |
2e6599cb ACM |
1048 | ireq = inet_rsk(req); |
1049 | newinet->daddr = ireq->rmt_addr; | |
1050 | newinet->rcv_saddr = ireq->loc_addr; | |
1051 | newinet->saddr = ireq->loc_addr; | |
1052 | newinet->opt = ireq->opt; | |
1053 | ireq->opt = NULL; | |
463c84b9 | 1054 | newinet->mc_index = inet_iif(skb); |
1da177e4 LT |
1055 | newinet->mc_ttl = skb->nh.iph->ttl; |
1056 | newtp->ext_header_len = 0; | |
1057 | if (newinet->opt) | |
1058 | newtp->ext_header_len = newinet->opt->optlen; | |
1059 | newinet->id = newtp->write_seq ^ jiffies; | |
1060 | ||
1061 | tcp_sync_mss(newsk, dst_mtu(dst)); | |
1062 | newtp->advmss = dst_metric(dst, RTAX_ADVMSS); | |
1063 | tcp_initialize_rcv_mss(newsk); | |
1064 | ||
f3f05f70 | 1065 | __inet_hash(&tcp_hashinfo, newsk, 0); |
2d8c4ce5 | 1066 | __inet_inherit_port(&tcp_hashinfo, sk, newsk); |
1da177e4 LT |
1067 | |
1068 | return newsk; | |
1069 | ||
1070 | exit_overflow: | |
1071 | NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS); | |
1072 | exit: | |
1073 | NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS); | |
1074 | dst_release(dst); | |
1075 | return NULL; | |
1076 | } | |
1077 | ||
1078 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1079 | { | |
1080 | struct tcphdr *th = skb->h.th; | |
1081 | struct iphdr *iph = skb->nh.iph; | |
1da177e4 | 1082 | struct sock *nsk; |
60236fdd | 1083 | struct request_sock **prev; |
1da177e4 | 1084 | /* Find possible connection requests. */ |
463c84b9 ACM |
1085 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, |
1086 | iph->saddr, iph->daddr); | |
1da177e4 LT |
1087 | if (req) |
1088 | return tcp_check_req(sk, skb, req, prev); | |
1089 | ||
e48c414e ACM |
1090 | nsk = __inet_lookup_established(&tcp_hashinfo, skb->nh.iph->saddr, |
1091 | th->source, skb->nh.iph->daddr, | |
463c84b9 | 1092 | ntohs(th->dest), inet_iif(skb)); |
1da177e4 LT |
1093 | |
1094 | if (nsk) { | |
1095 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1096 | bh_lock_sock(nsk); | |
1097 | return nsk; | |
1098 | } | |
8feaf0c0 | 1099 | inet_twsk_put((struct inet_timewait_sock *)nsk); |
1da177e4 LT |
1100 | return NULL; |
1101 | } | |
1102 | ||
1103 | #ifdef CONFIG_SYN_COOKIES | |
1104 | if (!th->rst && !th->syn && th->ack) | |
1105 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); | |
1106 | #endif | |
1107 | return sk; | |
1108 | } | |
1109 | ||
1110 | static int tcp_v4_checksum_init(struct sk_buff *skb) | |
1111 | { | |
1112 | if (skb->ip_summed == CHECKSUM_HW) { | |
1da177e4 | 1113 | if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr, |
fb286bb2 HX |
1114 | skb->nh.iph->daddr, skb->csum)) { |
1115 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1da177e4 | 1116 | return 0; |
fb286bb2 | 1117 | } |
1da177e4 | 1118 | } |
fb286bb2 HX |
1119 | |
1120 | skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr, skb->nh.iph->daddr, | |
1121 | skb->len, IPPROTO_TCP, 0); | |
1122 | ||
1da177e4 | 1123 | if (skb->len <= 76) { |
fb286bb2 | 1124 | return __skb_checksum_complete(skb); |
1da177e4 LT |
1125 | } |
1126 | return 0; | |
1127 | } | |
1128 | ||
1129 | ||
1130 | /* The socket must have it's spinlock held when we get | |
1131 | * here. | |
1132 | * | |
1133 | * We have a potential double-lock case here, so even when | |
1134 | * doing backlog processing we use the BH locking scheme. | |
1135 | * This is because we cannot sleep with the original spinlock | |
1136 | * held. | |
1137 | */ | |
1138 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1139 | { | |
1140 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ | |
1141 | TCP_CHECK_TIMER(sk); | |
1142 | if (tcp_rcv_established(sk, skb, skb->h.th, skb->len)) | |
1143 | goto reset; | |
1144 | TCP_CHECK_TIMER(sk); | |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb)) | |
1149 | goto csum_err; | |
1150 | ||
1151 | if (sk->sk_state == TCP_LISTEN) { | |
1152 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1153 | if (!nsk) | |
1154 | goto discard; | |
1155 | ||
1156 | if (nsk != sk) { | |
1157 | if (tcp_child_process(sk, nsk, skb)) | |
1158 | goto reset; | |
1159 | return 0; | |
1160 | } | |
1161 | } | |
1162 | ||
1163 | TCP_CHECK_TIMER(sk); | |
1164 | if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len)) | |
1165 | goto reset; | |
1166 | TCP_CHECK_TIMER(sk); | |
1167 | return 0; | |
1168 | ||
1169 | reset: | |
1170 | tcp_v4_send_reset(skb); | |
1171 | discard: | |
1172 | kfree_skb(skb); | |
1173 | /* Be careful here. If this function gets more complicated and | |
1174 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1175 | * might be destroyed here. This current version compiles correctly, | |
1176 | * but you have been warned. | |
1177 | */ | |
1178 | return 0; | |
1179 | ||
1180 | csum_err: | |
1181 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1182 | goto discard; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * From tcp_input.c | |
1187 | */ | |
1188 | ||
1189 | int tcp_v4_rcv(struct sk_buff *skb) | |
1190 | { | |
1191 | struct tcphdr *th; | |
1192 | struct sock *sk; | |
1193 | int ret; | |
1194 | ||
1195 | if (skb->pkt_type != PACKET_HOST) | |
1196 | goto discard_it; | |
1197 | ||
1198 | /* Count it even if it's bad */ | |
1199 | TCP_INC_STATS_BH(TCP_MIB_INSEGS); | |
1200 | ||
1201 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1202 | goto discard_it; | |
1203 | ||
1204 | th = skb->h.th; | |
1205 | ||
1206 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1207 | goto bad_packet; | |
1208 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1209 | goto discard_it; | |
1210 | ||
1211 | /* An explanation is required here, I think. | |
1212 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1213 | * provided case of th->doff==0 is eliminated. |
1da177e4 LT |
1214 | * So, we defer the checks. */ |
1215 | if ((skb->ip_summed != CHECKSUM_UNNECESSARY && | |
fb286bb2 | 1216 | tcp_v4_checksum_init(skb))) |
1da177e4 LT |
1217 | goto bad_packet; |
1218 | ||
1219 | th = skb->h.th; | |
1220 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); | |
1221 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1222 | skb->len - th->doff * 4); | |
1223 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1224 | TCP_SKB_CB(skb)->when = 0; | |
1225 | TCP_SKB_CB(skb)->flags = skb->nh.iph->tos; | |
1226 | TCP_SKB_CB(skb)->sacked = 0; | |
1227 | ||
e48c414e ACM |
1228 | sk = __inet_lookup(&tcp_hashinfo, skb->nh.iph->saddr, th->source, |
1229 | skb->nh.iph->daddr, ntohs(th->dest), | |
463c84b9 | 1230 | inet_iif(skb)); |
1da177e4 LT |
1231 | |
1232 | if (!sk) | |
1233 | goto no_tcp_socket; | |
1234 | ||
1235 | process: | |
1236 | if (sk->sk_state == TCP_TIME_WAIT) | |
1237 | goto do_time_wait; | |
1238 | ||
1239 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1240 | goto discard_and_relse; | |
1241 | ||
1242 | if (sk_filter(sk, skb, 0)) | |
1243 | goto discard_and_relse; | |
1244 | ||
1245 | skb->dev = NULL; | |
1246 | ||
1247 | bh_lock_sock(sk); | |
1248 | ret = 0; | |
1249 | if (!sock_owned_by_user(sk)) { | |
1250 | if (!tcp_prequeue(sk, skb)) | |
1251 | ret = tcp_v4_do_rcv(sk, skb); | |
1252 | } else | |
1253 | sk_add_backlog(sk, skb); | |
1254 | bh_unlock_sock(sk); | |
1255 | ||
1256 | sock_put(sk); | |
1257 | ||
1258 | return ret; | |
1259 | ||
1260 | no_tcp_socket: | |
1261 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1262 | goto discard_it; | |
1263 | ||
1264 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1265 | bad_packet: | |
1266 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1267 | } else { | |
1268 | tcp_v4_send_reset(skb); | |
1269 | } | |
1270 | ||
1271 | discard_it: | |
1272 | /* Discard frame. */ | |
1273 | kfree_skb(skb); | |
1274 | return 0; | |
1275 | ||
1276 | discard_and_relse: | |
1277 | sock_put(sk); | |
1278 | goto discard_it; | |
1279 | ||
1280 | do_time_wait: | |
1281 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
8feaf0c0 | 1282 | inet_twsk_put((struct inet_timewait_sock *) sk); |
1da177e4 LT |
1283 | goto discard_it; |
1284 | } | |
1285 | ||
1286 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1287 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
8feaf0c0 | 1288 | inet_twsk_put((struct inet_timewait_sock *) sk); |
1da177e4 LT |
1289 | goto discard_it; |
1290 | } | |
8feaf0c0 ACM |
1291 | switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk, |
1292 | skb, th)) { | |
1da177e4 | 1293 | case TCP_TW_SYN: { |
33b62231 ACM |
1294 | struct sock *sk2 = inet_lookup_listener(&tcp_hashinfo, |
1295 | skb->nh.iph->daddr, | |
1296 | ntohs(th->dest), | |
463c84b9 | 1297 | inet_iif(skb)); |
1da177e4 | 1298 | if (sk2) { |
295ff7ed ACM |
1299 | inet_twsk_deschedule((struct inet_timewait_sock *)sk, |
1300 | &tcp_death_row); | |
8feaf0c0 | 1301 | inet_twsk_put((struct inet_timewait_sock *)sk); |
1da177e4 LT |
1302 | sk = sk2; |
1303 | goto process; | |
1304 | } | |
1305 | /* Fall through to ACK */ | |
1306 | } | |
1307 | case TCP_TW_ACK: | |
1308 | tcp_v4_timewait_ack(sk, skb); | |
1309 | break; | |
1310 | case TCP_TW_RST: | |
1311 | goto no_tcp_socket; | |
1312 | case TCP_TW_SUCCESS:; | |
1313 | } | |
1314 | goto discard_it; | |
1315 | } | |
1316 | ||
1da177e4 LT |
1317 | static void v4_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr) |
1318 | { | |
1319 | struct sockaddr_in *sin = (struct sockaddr_in *) uaddr; | |
1320 | struct inet_sock *inet = inet_sk(sk); | |
1321 | ||
1322 | sin->sin_family = AF_INET; | |
1323 | sin->sin_addr.s_addr = inet->daddr; | |
1324 | sin->sin_port = inet->dport; | |
1325 | } | |
1326 | ||
1327 | /* VJ's idea. Save last timestamp seen from this destination | |
1328 | * and hold it at least for normal timewait interval to use for duplicate | |
1329 | * segment detection in subsequent connections, before they enter synchronized | |
1330 | * state. | |
1331 | */ | |
1332 | ||
1333 | int tcp_v4_remember_stamp(struct sock *sk) | |
1334 | { | |
1335 | struct inet_sock *inet = inet_sk(sk); | |
1336 | struct tcp_sock *tp = tcp_sk(sk); | |
1337 | struct rtable *rt = (struct rtable *)__sk_dst_get(sk); | |
1338 | struct inet_peer *peer = NULL; | |
1339 | int release_it = 0; | |
1340 | ||
1341 | if (!rt || rt->rt_dst != inet->daddr) { | |
1342 | peer = inet_getpeer(inet->daddr, 1); | |
1343 | release_it = 1; | |
1344 | } else { | |
1345 | if (!rt->peer) | |
1346 | rt_bind_peer(rt, 1); | |
1347 | peer = rt->peer; | |
1348 | } | |
1349 | ||
1350 | if (peer) { | |
1351 | if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 || | |
1352 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && | |
1353 | peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) { | |
1354 | peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp; | |
1355 | peer->tcp_ts = tp->rx_opt.ts_recent; | |
1356 | } | |
1357 | if (release_it) | |
1358 | inet_putpeer(peer); | |
1359 | return 1; | |
1360 | } | |
1361 | ||
1362 | return 0; | |
1363 | } | |
1364 | ||
8feaf0c0 | 1365 | int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw) |
1da177e4 | 1366 | { |
8feaf0c0 | 1367 | struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1); |
1da177e4 LT |
1368 | |
1369 | if (peer) { | |
8feaf0c0 ACM |
1370 | const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
1371 | ||
1372 | if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 || | |
1da177e4 | 1373 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && |
8feaf0c0 ACM |
1374 | peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) { |
1375 | peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp; | |
1376 | peer->tcp_ts = tcptw->tw_ts_recent; | |
1da177e4 LT |
1377 | } |
1378 | inet_putpeer(peer); | |
1379 | return 1; | |
1380 | } | |
1381 | ||
1382 | return 0; | |
1383 | } | |
1384 | ||
1385 | struct tcp_func ipv4_specific = { | |
1386 | .queue_xmit = ip_queue_xmit, | |
1387 | .send_check = tcp_v4_send_check, | |
32519f11 | 1388 | .rebuild_header = inet_sk_rebuild_header, |
1da177e4 LT |
1389 | .conn_request = tcp_v4_conn_request, |
1390 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
1391 | .remember_stamp = tcp_v4_remember_stamp, | |
1392 | .net_header_len = sizeof(struct iphdr), | |
1393 | .setsockopt = ip_setsockopt, | |
1394 | .getsockopt = ip_getsockopt, | |
1395 | .addr2sockaddr = v4_addr2sockaddr, | |
1396 | .sockaddr_len = sizeof(struct sockaddr_in), | |
1397 | }; | |
1398 | ||
1399 | /* NOTE: A lot of things set to zero explicitly by call to | |
1400 | * sk_alloc() so need not be done here. | |
1401 | */ | |
1402 | static int tcp_v4_init_sock(struct sock *sk) | |
1403 | { | |
6687e988 | 1404 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
1405 | struct tcp_sock *tp = tcp_sk(sk); |
1406 | ||
1407 | skb_queue_head_init(&tp->out_of_order_queue); | |
1408 | tcp_init_xmit_timers(sk); | |
1409 | tcp_prequeue_init(tp); | |
1410 | ||
6687e988 | 1411 | icsk->icsk_rto = TCP_TIMEOUT_INIT; |
1da177e4 LT |
1412 | tp->mdev = TCP_TIMEOUT_INIT; |
1413 | ||
1414 | /* So many TCP implementations out there (incorrectly) count the | |
1415 | * initial SYN frame in their delayed-ACK and congestion control | |
1416 | * algorithms that we must have the following bandaid to talk | |
1417 | * efficiently to them. -DaveM | |
1418 | */ | |
1419 | tp->snd_cwnd = 2; | |
1420 | ||
1421 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
1422 | * initialization of these values. | |
1423 | */ | |
1424 | tp->snd_ssthresh = 0x7fffffff; /* Infinity */ | |
1425 | tp->snd_cwnd_clamp = ~0; | |
c1b4a7e6 | 1426 | tp->mss_cache = 536; |
1da177e4 LT |
1427 | |
1428 | tp->reordering = sysctl_tcp_reordering; | |
6687e988 | 1429 | icsk->icsk_ca_ops = &tcp_init_congestion_ops; |
1da177e4 LT |
1430 | |
1431 | sk->sk_state = TCP_CLOSE; | |
1432 | ||
1433 | sk->sk_write_space = sk_stream_write_space; | |
1434 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
1435 | ||
1436 | tp->af_specific = &ipv4_specific; | |
1437 | ||
1438 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; | |
1439 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
1440 | ||
1441 | atomic_inc(&tcp_sockets_allocated); | |
1442 | ||
1443 | return 0; | |
1444 | } | |
1445 | ||
1446 | int tcp_v4_destroy_sock(struct sock *sk) | |
1447 | { | |
1448 | struct tcp_sock *tp = tcp_sk(sk); | |
1449 | ||
1450 | tcp_clear_xmit_timers(sk); | |
1451 | ||
6687e988 | 1452 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1453 | |
1da177e4 LT |
1454 | /* Cleanup up the write buffer. */ |
1455 | sk_stream_writequeue_purge(sk); | |
1456 | ||
1457 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
1458 | __skb_queue_purge(&tp->out_of_order_queue); | |
1459 | ||
1460 | /* Clean prequeue, it must be empty really */ | |
1461 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1462 | ||
1463 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1464 | if (inet_csk(sk)->icsk_bind_hash) |
2d8c4ce5 | 1465 | inet_put_port(&tcp_hashinfo, sk); |
1da177e4 LT |
1466 | |
1467 | /* | |
1468 | * If sendmsg cached page exists, toss it. | |
1469 | */ | |
1470 | if (sk->sk_sndmsg_page) { | |
1471 | __free_page(sk->sk_sndmsg_page); | |
1472 | sk->sk_sndmsg_page = NULL; | |
1473 | } | |
1474 | ||
1475 | atomic_dec(&tcp_sockets_allocated); | |
1476 | ||
1477 | return 0; | |
1478 | } | |
1479 | ||
1480 | EXPORT_SYMBOL(tcp_v4_destroy_sock); | |
1481 | ||
1482 | #ifdef CONFIG_PROC_FS | |
1483 | /* Proc filesystem TCP sock list dumping. */ | |
1484 | ||
8feaf0c0 | 1485 | static inline struct inet_timewait_sock *tw_head(struct hlist_head *head) |
1da177e4 LT |
1486 | { |
1487 | return hlist_empty(head) ? NULL : | |
8feaf0c0 | 1488 | list_entry(head->first, struct inet_timewait_sock, tw_node); |
1da177e4 LT |
1489 | } |
1490 | ||
8feaf0c0 | 1491 | static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw) |
1da177e4 LT |
1492 | { |
1493 | return tw->tw_node.next ? | |
1494 | hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; | |
1495 | } | |
1496 | ||
1497 | static void *listening_get_next(struct seq_file *seq, void *cur) | |
1498 | { | |
463c84b9 | 1499 | struct inet_connection_sock *icsk; |
1da177e4 LT |
1500 | struct hlist_node *node; |
1501 | struct sock *sk = cur; | |
1502 | struct tcp_iter_state* st = seq->private; | |
1503 | ||
1504 | if (!sk) { | |
1505 | st->bucket = 0; | |
6e04e021 | 1506 | sk = sk_head(&tcp_hashinfo.listening_hash[0]); |
1da177e4 LT |
1507 | goto get_sk; |
1508 | } | |
1509 | ||
1510 | ++st->num; | |
1511 | ||
1512 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
60236fdd | 1513 | struct request_sock *req = cur; |
1da177e4 | 1514 | |
463c84b9 | 1515 | icsk = inet_csk(st->syn_wait_sk); |
1da177e4 LT |
1516 | req = req->dl_next; |
1517 | while (1) { | |
1518 | while (req) { | |
60236fdd | 1519 | if (req->rsk_ops->family == st->family) { |
1da177e4 LT |
1520 | cur = req; |
1521 | goto out; | |
1522 | } | |
1523 | req = req->dl_next; | |
1524 | } | |
1525 | if (++st->sbucket >= TCP_SYNQ_HSIZE) | |
1526 | break; | |
1527 | get_req: | |
463c84b9 | 1528 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; |
1da177e4 LT |
1529 | } |
1530 | sk = sk_next(st->syn_wait_sk); | |
1531 | st->state = TCP_SEQ_STATE_LISTENING; | |
463c84b9 | 1532 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 1533 | } else { |
463c84b9 ACM |
1534 | icsk = inet_csk(sk); |
1535 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1536 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
1da177e4 | 1537 | goto start_req; |
463c84b9 | 1538 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 LT |
1539 | sk = sk_next(sk); |
1540 | } | |
1541 | get_sk: | |
1542 | sk_for_each_from(sk, node) { | |
1543 | if (sk->sk_family == st->family) { | |
1544 | cur = sk; | |
1545 | goto out; | |
1546 | } | |
463c84b9 ACM |
1547 | icsk = inet_csk(sk); |
1548 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1549 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
1da177e4 LT |
1550 | start_req: |
1551 | st->uid = sock_i_uid(sk); | |
1552 | st->syn_wait_sk = sk; | |
1553 | st->state = TCP_SEQ_STATE_OPENREQ; | |
1554 | st->sbucket = 0; | |
1555 | goto get_req; | |
1556 | } | |
463c84b9 | 1557 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 1558 | } |
0f7ff927 | 1559 | if (++st->bucket < INET_LHTABLE_SIZE) { |
6e04e021 | 1560 | sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]); |
1da177e4 LT |
1561 | goto get_sk; |
1562 | } | |
1563 | cur = NULL; | |
1564 | out: | |
1565 | return cur; | |
1566 | } | |
1567 | ||
1568 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
1569 | { | |
1570 | void *rc = listening_get_next(seq, NULL); | |
1571 | ||
1572 | while (rc && *pos) { | |
1573 | rc = listening_get_next(seq, rc); | |
1574 | --*pos; | |
1575 | } | |
1576 | return rc; | |
1577 | } | |
1578 | ||
1579 | static void *established_get_first(struct seq_file *seq) | |
1580 | { | |
1581 | struct tcp_iter_state* st = seq->private; | |
1582 | void *rc = NULL; | |
1583 | ||
6e04e021 | 1584 | for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) { |
1da177e4 LT |
1585 | struct sock *sk; |
1586 | struct hlist_node *node; | |
8feaf0c0 | 1587 | struct inet_timewait_sock *tw; |
1da177e4 LT |
1588 | |
1589 | /* We can reschedule _before_ having picked the target: */ | |
1590 | cond_resched_softirq(); | |
1591 | ||
6e04e021 ACM |
1592 | read_lock(&tcp_hashinfo.ehash[st->bucket].lock); |
1593 | sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { | |
1da177e4 LT |
1594 | if (sk->sk_family != st->family) { |
1595 | continue; | |
1596 | } | |
1597 | rc = sk; | |
1598 | goto out; | |
1599 | } | |
1600 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
8feaf0c0 ACM |
1601 | inet_twsk_for_each(tw, node, |
1602 | &tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain) { | |
1da177e4 LT |
1603 | if (tw->tw_family != st->family) { |
1604 | continue; | |
1605 | } | |
1606 | rc = tw; | |
1607 | goto out; | |
1608 | } | |
6e04e021 | 1609 | read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); |
1da177e4 LT |
1610 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
1611 | } | |
1612 | out: | |
1613 | return rc; | |
1614 | } | |
1615 | ||
1616 | static void *established_get_next(struct seq_file *seq, void *cur) | |
1617 | { | |
1618 | struct sock *sk = cur; | |
8feaf0c0 | 1619 | struct inet_timewait_sock *tw; |
1da177e4 LT |
1620 | struct hlist_node *node; |
1621 | struct tcp_iter_state* st = seq->private; | |
1622 | ||
1623 | ++st->num; | |
1624 | ||
1625 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { | |
1626 | tw = cur; | |
1627 | tw = tw_next(tw); | |
1628 | get_tw: | |
1629 | while (tw && tw->tw_family != st->family) { | |
1630 | tw = tw_next(tw); | |
1631 | } | |
1632 | if (tw) { | |
1633 | cur = tw; | |
1634 | goto out; | |
1635 | } | |
6e04e021 | 1636 | read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); |
1da177e4 LT |
1637 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
1638 | ||
1639 | /* We can reschedule between buckets: */ | |
1640 | cond_resched_softirq(); | |
1641 | ||
6e04e021 ACM |
1642 | if (++st->bucket < tcp_hashinfo.ehash_size) { |
1643 | read_lock(&tcp_hashinfo.ehash[st->bucket].lock); | |
1644 | sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain); | |
1da177e4 LT |
1645 | } else { |
1646 | cur = NULL; | |
1647 | goto out; | |
1648 | } | |
1649 | } else | |
1650 | sk = sk_next(sk); | |
1651 | ||
1652 | sk_for_each_from(sk, node) { | |
1653 | if (sk->sk_family == st->family) | |
1654 | goto found; | |
1655 | } | |
1656 | ||
1657 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
6e04e021 | 1658 | tw = tw_head(&tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain); |
1da177e4 LT |
1659 | goto get_tw; |
1660 | found: | |
1661 | cur = sk; | |
1662 | out: | |
1663 | return cur; | |
1664 | } | |
1665 | ||
1666 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
1667 | { | |
1668 | void *rc = established_get_first(seq); | |
1669 | ||
1670 | while (rc && pos) { | |
1671 | rc = established_get_next(seq, rc); | |
1672 | --pos; | |
1673 | } | |
1674 | return rc; | |
1675 | } | |
1676 | ||
1677 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
1678 | { | |
1679 | void *rc; | |
1680 | struct tcp_iter_state* st = seq->private; | |
1681 | ||
f3f05f70 | 1682 | inet_listen_lock(&tcp_hashinfo); |
1da177e4 LT |
1683 | st->state = TCP_SEQ_STATE_LISTENING; |
1684 | rc = listening_get_idx(seq, &pos); | |
1685 | ||
1686 | if (!rc) { | |
f3f05f70 | 1687 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
1688 | local_bh_disable(); |
1689 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
1690 | rc = established_get_idx(seq, pos); | |
1691 | } | |
1692 | ||
1693 | return rc; | |
1694 | } | |
1695 | ||
1696 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) | |
1697 | { | |
1698 | struct tcp_iter_state* st = seq->private; | |
1699 | st->state = TCP_SEQ_STATE_LISTENING; | |
1700 | st->num = 0; | |
1701 | return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
1702 | } | |
1703 | ||
1704 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1705 | { | |
1706 | void *rc = NULL; | |
1707 | struct tcp_iter_state* st; | |
1708 | ||
1709 | if (v == SEQ_START_TOKEN) { | |
1710 | rc = tcp_get_idx(seq, 0); | |
1711 | goto out; | |
1712 | } | |
1713 | st = seq->private; | |
1714 | ||
1715 | switch (st->state) { | |
1716 | case TCP_SEQ_STATE_OPENREQ: | |
1717 | case TCP_SEQ_STATE_LISTENING: | |
1718 | rc = listening_get_next(seq, v); | |
1719 | if (!rc) { | |
f3f05f70 | 1720 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
1721 | local_bh_disable(); |
1722 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
1723 | rc = established_get_first(seq); | |
1724 | } | |
1725 | break; | |
1726 | case TCP_SEQ_STATE_ESTABLISHED: | |
1727 | case TCP_SEQ_STATE_TIME_WAIT: | |
1728 | rc = established_get_next(seq, v); | |
1729 | break; | |
1730 | } | |
1731 | out: | |
1732 | ++*pos; | |
1733 | return rc; | |
1734 | } | |
1735 | ||
1736 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
1737 | { | |
1738 | struct tcp_iter_state* st = seq->private; | |
1739 | ||
1740 | switch (st->state) { | |
1741 | case TCP_SEQ_STATE_OPENREQ: | |
1742 | if (v) { | |
463c84b9 ACM |
1743 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); |
1744 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1da177e4 LT |
1745 | } |
1746 | case TCP_SEQ_STATE_LISTENING: | |
1747 | if (v != SEQ_START_TOKEN) | |
f3f05f70 | 1748 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
1749 | break; |
1750 | case TCP_SEQ_STATE_TIME_WAIT: | |
1751 | case TCP_SEQ_STATE_ESTABLISHED: | |
1752 | if (v) | |
6e04e021 | 1753 | read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); |
1da177e4 LT |
1754 | local_bh_enable(); |
1755 | break; | |
1756 | } | |
1757 | } | |
1758 | ||
1759 | static int tcp_seq_open(struct inode *inode, struct file *file) | |
1760 | { | |
1761 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; | |
1762 | struct seq_file *seq; | |
1763 | struct tcp_iter_state *s; | |
1764 | int rc; | |
1765 | ||
1766 | if (unlikely(afinfo == NULL)) | |
1767 | return -EINVAL; | |
1768 | ||
1769 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
1770 | if (!s) | |
1771 | return -ENOMEM; | |
1772 | memset(s, 0, sizeof(*s)); | |
1773 | s->family = afinfo->family; | |
1774 | s->seq_ops.start = tcp_seq_start; | |
1775 | s->seq_ops.next = tcp_seq_next; | |
1776 | s->seq_ops.show = afinfo->seq_show; | |
1777 | s->seq_ops.stop = tcp_seq_stop; | |
1778 | ||
1779 | rc = seq_open(file, &s->seq_ops); | |
1780 | if (rc) | |
1781 | goto out_kfree; | |
1782 | seq = file->private_data; | |
1783 | seq->private = s; | |
1784 | out: | |
1785 | return rc; | |
1786 | out_kfree: | |
1787 | kfree(s); | |
1788 | goto out; | |
1789 | } | |
1790 | ||
1791 | int tcp_proc_register(struct tcp_seq_afinfo *afinfo) | |
1792 | { | |
1793 | int rc = 0; | |
1794 | struct proc_dir_entry *p; | |
1795 | ||
1796 | if (!afinfo) | |
1797 | return -EINVAL; | |
1798 | afinfo->seq_fops->owner = afinfo->owner; | |
1799 | afinfo->seq_fops->open = tcp_seq_open; | |
1800 | afinfo->seq_fops->read = seq_read; | |
1801 | afinfo->seq_fops->llseek = seq_lseek; | |
1802 | afinfo->seq_fops->release = seq_release_private; | |
1803 | ||
1804 | p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops); | |
1805 | if (p) | |
1806 | p->data = afinfo; | |
1807 | else | |
1808 | rc = -ENOMEM; | |
1809 | return rc; | |
1810 | } | |
1811 | ||
1812 | void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo) | |
1813 | { | |
1814 | if (!afinfo) | |
1815 | return; | |
1816 | proc_net_remove(afinfo->name); | |
1817 | memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); | |
1818 | } | |
1819 | ||
60236fdd | 1820 | static void get_openreq4(struct sock *sk, struct request_sock *req, |
1da177e4 LT |
1821 | char *tmpbuf, int i, int uid) |
1822 | { | |
2e6599cb | 1823 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
1824 | int ttd = req->expires - jiffies; |
1825 | ||
1826 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" | |
1827 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p", | |
1828 | i, | |
2e6599cb | 1829 | ireq->loc_addr, |
1da177e4 | 1830 | ntohs(inet_sk(sk)->sport), |
2e6599cb ACM |
1831 | ireq->rmt_addr, |
1832 | ntohs(ireq->rmt_port), | |
1da177e4 LT |
1833 | TCP_SYN_RECV, |
1834 | 0, 0, /* could print option size, but that is af dependent. */ | |
1835 | 1, /* timers active (only the expire timer) */ | |
1836 | jiffies_to_clock_t(ttd), | |
1837 | req->retrans, | |
1838 | uid, | |
1839 | 0, /* non standard timer */ | |
1840 | 0, /* open_requests have no inode */ | |
1841 | atomic_read(&sk->sk_refcnt), | |
1842 | req); | |
1843 | } | |
1844 | ||
1845 | static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i) | |
1846 | { | |
1847 | int timer_active; | |
1848 | unsigned long timer_expires; | |
1849 | struct tcp_sock *tp = tcp_sk(sp); | |
463c84b9 | 1850 | const struct inet_connection_sock *icsk = inet_csk(sp); |
1da177e4 LT |
1851 | struct inet_sock *inet = inet_sk(sp); |
1852 | unsigned int dest = inet->daddr; | |
1853 | unsigned int src = inet->rcv_saddr; | |
1854 | __u16 destp = ntohs(inet->dport); | |
1855 | __u16 srcp = ntohs(inet->sport); | |
1856 | ||
463c84b9 | 1857 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) { |
1da177e4 | 1858 | timer_active = 1; |
463c84b9 ACM |
1859 | timer_expires = icsk->icsk_timeout; |
1860 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 1861 | timer_active = 4; |
463c84b9 | 1862 | timer_expires = icsk->icsk_timeout; |
1da177e4 LT |
1863 | } else if (timer_pending(&sp->sk_timer)) { |
1864 | timer_active = 2; | |
1865 | timer_expires = sp->sk_timer.expires; | |
1866 | } else { | |
1867 | timer_active = 0; | |
1868 | timer_expires = jiffies; | |
1869 | } | |
1870 | ||
1871 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " | |
1872 | "%08X %5d %8d %lu %d %p %u %u %u %u %d", | |
1873 | i, src, srcp, dest, destp, sp->sk_state, | |
1874 | tp->write_seq - tp->snd_una, tp->rcv_nxt - tp->copied_seq, | |
1875 | timer_active, | |
1876 | jiffies_to_clock_t(timer_expires - jiffies), | |
463c84b9 | 1877 | icsk->icsk_retransmits, |
1da177e4 | 1878 | sock_i_uid(sp), |
6687e988 | 1879 | icsk->icsk_probes_out, |
1da177e4 LT |
1880 | sock_i_ino(sp), |
1881 | atomic_read(&sp->sk_refcnt), sp, | |
463c84b9 ACM |
1882 | icsk->icsk_rto, |
1883 | icsk->icsk_ack.ato, | |
1884 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, | |
1da177e4 LT |
1885 | tp->snd_cwnd, |
1886 | tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh); | |
1887 | } | |
1888 | ||
8feaf0c0 | 1889 | static void get_timewait4_sock(struct inet_timewait_sock *tw, char *tmpbuf, int i) |
1da177e4 LT |
1890 | { |
1891 | unsigned int dest, src; | |
1892 | __u16 destp, srcp; | |
1893 | int ttd = tw->tw_ttd - jiffies; | |
1894 | ||
1895 | if (ttd < 0) | |
1896 | ttd = 0; | |
1897 | ||
1898 | dest = tw->tw_daddr; | |
1899 | src = tw->tw_rcv_saddr; | |
1900 | destp = ntohs(tw->tw_dport); | |
1901 | srcp = ntohs(tw->tw_sport); | |
1902 | ||
1903 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" | |
1904 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p", | |
1905 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, | |
1906 | 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, | |
1907 | atomic_read(&tw->tw_refcnt), tw); | |
1908 | } | |
1909 | ||
1910 | #define TMPSZ 150 | |
1911 | ||
1912 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
1913 | { | |
1914 | struct tcp_iter_state* st; | |
1915 | char tmpbuf[TMPSZ + 1]; | |
1916 | ||
1917 | if (v == SEQ_START_TOKEN) { | |
1918 | seq_printf(seq, "%-*s\n", TMPSZ - 1, | |
1919 | " sl local_address rem_address st tx_queue " | |
1920 | "rx_queue tr tm->when retrnsmt uid timeout " | |
1921 | "inode"); | |
1922 | goto out; | |
1923 | } | |
1924 | st = seq->private; | |
1925 | ||
1926 | switch (st->state) { | |
1927 | case TCP_SEQ_STATE_LISTENING: | |
1928 | case TCP_SEQ_STATE_ESTABLISHED: | |
1929 | get_tcp4_sock(v, tmpbuf, st->num); | |
1930 | break; | |
1931 | case TCP_SEQ_STATE_OPENREQ: | |
1932 | get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid); | |
1933 | break; | |
1934 | case TCP_SEQ_STATE_TIME_WAIT: | |
1935 | get_timewait4_sock(v, tmpbuf, st->num); | |
1936 | break; | |
1937 | } | |
1938 | seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf); | |
1939 | out: | |
1940 | return 0; | |
1941 | } | |
1942 | ||
1943 | static struct file_operations tcp4_seq_fops; | |
1944 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { | |
1945 | .owner = THIS_MODULE, | |
1946 | .name = "tcp", | |
1947 | .family = AF_INET, | |
1948 | .seq_show = tcp4_seq_show, | |
1949 | .seq_fops = &tcp4_seq_fops, | |
1950 | }; | |
1951 | ||
1952 | int __init tcp4_proc_init(void) | |
1953 | { | |
1954 | return tcp_proc_register(&tcp4_seq_afinfo); | |
1955 | } | |
1956 | ||
1957 | void tcp4_proc_exit(void) | |
1958 | { | |
1959 | tcp_proc_unregister(&tcp4_seq_afinfo); | |
1960 | } | |
1961 | #endif /* CONFIG_PROC_FS */ | |
1962 | ||
1963 | struct proto tcp_prot = { | |
1964 | .name = "TCP", | |
1965 | .owner = THIS_MODULE, | |
1966 | .close = tcp_close, | |
1967 | .connect = tcp_v4_connect, | |
1968 | .disconnect = tcp_disconnect, | |
463c84b9 | 1969 | .accept = inet_csk_accept, |
1da177e4 LT |
1970 | .ioctl = tcp_ioctl, |
1971 | .init = tcp_v4_init_sock, | |
1972 | .destroy = tcp_v4_destroy_sock, | |
1973 | .shutdown = tcp_shutdown, | |
1974 | .setsockopt = tcp_setsockopt, | |
1975 | .getsockopt = tcp_getsockopt, | |
1976 | .sendmsg = tcp_sendmsg, | |
1977 | .recvmsg = tcp_recvmsg, | |
1978 | .backlog_rcv = tcp_v4_do_rcv, | |
1979 | .hash = tcp_v4_hash, | |
1980 | .unhash = tcp_unhash, | |
1981 | .get_port = tcp_v4_get_port, | |
1982 | .enter_memory_pressure = tcp_enter_memory_pressure, | |
1983 | .sockets_allocated = &tcp_sockets_allocated, | |
0a5578cf | 1984 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
1985 | .memory_allocated = &tcp_memory_allocated, |
1986 | .memory_pressure = &tcp_memory_pressure, | |
1987 | .sysctl_mem = sysctl_tcp_mem, | |
1988 | .sysctl_wmem = sysctl_tcp_wmem, | |
1989 | .sysctl_rmem = sysctl_tcp_rmem, | |
1990 | .max_header = MAX_TCP_HEADER, | |
1991 | .obj_size = sizeof(struct tcp_sock), | |
8feaf0c0 | 1992 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), |
60236fdd | 1993 | .rsk_prot = &tcp_request_sock_ops, |
1da177e4 LT |
1994 | }; |
1995 | ||
1996 | ||
1997 | ||
1998 | void __init tcp_v4_init(struct net_proto_family *ops) | |
1999 | { | |
2000 | int err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_TCP, &tcp_socket); | |
2001 | if (err < 0) | |
2002 | panic("Failed to create the TCP control socket.\n"); | |
2003 | tcp_socket->sk->sk_allocation = GFP_ATOMIC; | |
2004 | inet_sk(tcp_socket->sk)->uc_ttl = -1; | |
2005 | ||
2006 | /* Unhash it so that IP input processing does not even | |
2007 | * see it, we do not wish this socket to see incoming | |
2008 | * packets. | |
2009 | */ | |
2010 | tcp_socket->sk->sk_prot->unhash(tcp_socket->sk); | |
2011 | } | |
2012 | ||
2013 | EXPORT_SYMBOL(ipv4_specific); | |
0f7ff927 | 2014 | EXPORT_SYMBOL(inet_bind_bucket_create); |
1da177e4 | 2015 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 2016 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 LT |
2017 | EXPORT_SYMBOL(tcp_unhash); |
2018 | EXPORT_SYMBOL(tcp_v4_conn_request); | |
2019 | EXPORT_SYMBOL(tcp_v4_connect); | |
2020 | EXPORT_SYMBOL(tcp_v4_do_rcv); | |
1da177e4 LT |
2021 | EXPORT_SYMBOL(tcp_v4_remember_stamp); |
2022 | EXPORT_SYMBOL(tcp_v4_send_check); | |
2023 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); | |
2024 | ||
2025 | #ifdef CONFIG_PROC_FS | |
2026 | EXPORT_SYMBOL(tcp_proc_register); | |
2027 | EXPORT_SYMBOL(tcp_proc_unregister); | |
2028 | #endif | |
2029 | EXPORT_SYMBOL(sysctl_local_port_range); | |
1da177e4 LT |
2030 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
2031 | EXPORT_SYMBOL(sysctl_tcp_tw_reuse); | |
2032 |