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 | ||
1da177e4 LT |
55 | |
56 | #include <linux/types.h> | |
57 | #include <linux/fcntl.h> | |
58 | #include <linux/module.h> | |
59 | #include <linux/random.h> | |
60 | #include <linux/cache.h> | |
61 | #include <linux/jhash.h> | |
62 | #include <linux/init.h> | |
63 | #include <linux/times.h> | |
64 | ||
65 | #include <net/icmp.h> | |
304a1618 | 66 | #include <net/inet_hashtables.h> |
1da177e4 | 67 | #include <net/tcp.h> |
20380731 | 68 | #include <net/transp_v6.h> |
1da177e4 LT |
69 | #include <net/ipv6.h> |
70 | #include <net/inet_common.h> | |
6d6ee43e | 71 | #include <net/timewait_sock.h> |
1da177e4 | 72 | #include <net/xfrm.h> |
1a2449a8 | 73 | #include <net/netdma.h> |
1da177e4 LT |
74 | |
75 | #include <linux/inet.h> | |
76 | #include <linux/ipv6.h> | |
77 | #include <linux/stddef.h> | |
78 | #include <linux/proc_fs.h> | |
79 | #include <linux/seq_file.h> | |
80 | ||
ab32ea5d BH |
81 | int sysctl_tcp_tw_reuse __read_mostly; |
82 | int sysctl_tcp_low_latency __read_mostly; | |
1da177e4 LT |
83 | |
84 | /* Check TCP sequence numbers in ICMP packets. */ | |
85 | #define ICMP_MIN_LENGTH 8 | |
86 | ||
87 | /* Socket used for sending RSTs */ | |
88 | static struct socket *tcp_socket; | |
89 | ||
8292a17a | 90 | void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb); |
1da177e4 | 91 | |
0f7ff927 | 92 | struct inet_hashinfo __cacheline_aligned tcp_hashinfo = { |
e4d91918 | 93 | .lhash_lock = __RW_LOCK_UNLOCKED(tcp_hashinfo.lhash_lock), |
0f7ff927 ACM |
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 | { | |
971af18b ACM |
100 | return inet_csk_get_port(&tcp_hashinfo, sk, snum, |
101 | inet_csk_bind_conflict); | |
463c84b9 ACM |
102 | } |
103 | ||
1da177e4 LT |
104 | static void tcp_v4_hash(struct sock *sk) |
105 | { | |
81849d10 | 106 | inet_hash(&tcp_hashinfo, sk); |
1da177e4 LT |
107 | } |
108 | ||
109 | void tcp_unhash(struct sock *sk) | |
110 | { | |
81849d10 | 111 | inet_unhash(&tcp_hashinfo, sk); |
1da177e4 LT |
112 | } |
113 | ||
1da177e4 LT |
114 | static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb) |
115 | { | |
116 | return secure_tcp_sequence_number(skb->nh.iph->daddr, | |
117 | skb->nh.iph->saddr, | |
118 | skb->h.th->dest, | |
119 | skb->h.th->source); | |
120 | } | |
121 | ||
6d6ee43e ACM |
122 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
123 | { | |
124 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
125 | struct tcp_sock *tp = tcp_sk(sk); | |
126 | ||
127 | /* With PAWS, it is safe from the viewpoint | |
128 | of data integrity. Even without PAWS it is safe provided sequence | |
129 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
130 | ||
131 | Actually, the idea is close to VJ's one, only timestamp cache is | |
132 | held not per host, but per port pair and TW bucket is used as state | |
133 | holder. | |
134 | ||
135 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
136 | and use initial timestamp retrieved from peer table. | |
137 | */ | |
138 | if (tcptw->tw_ts_recent_stamp && | |
139 | (twp == NULL || (sysctl_tcp_tw_reuse && | |
140 | xtime.tv_sec - tcptw->tw_ts_recent_stamp > 1))) { | |
141 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; | |
142 | if (tp->write_seq == 0) | |
143 | tp->write_seq = 1; | |
144 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
145 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
146 | sock_hold(sktw); | |
147 | return 1; | |
148 | } | |
149 | ||
150 | return 0; | |
151 | } | |
152 | ||
153 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); | |
154 | ||
1da177e4 LT |
155 | /* This will initiate an outgoing connection. */ |
156 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
157 | { | |
158 | struct inet_sock *inet = inet_sk(sk); | |
159 | struct tcp_sock *tp = tcp_sk(sk); | |
160 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; | |
161 | struct rtable *rt; | |
bada8adc | 162 | __be32 daddr, nexthop; |
1da177e4 LT |
163 | int tmp; |
164 | int err; | |
165 | ||
166 | if (addr_len < sizeof(struct sockaddr_in)) | |
167 | return -EINVAL; | |
168 | ||
169 | if (usin->sin_family != AF_INET) | |
170 | return -EAFNOSUPPORT; | |
171 | ||
172 | nexthop = daddr = usin->sin_addr.s_addr; | |
173 | if (inet->opt && inet->opt->srr) { | |
174 | if (!daddr) | |
175 | return -EINVAL; | |
176 | nexthop = inet->opt->faddr; | |
177 | } | |
178 | ||
179 | tmp = ip_route_connect(&rt, nexthop, inet->saddr, | |
180 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, | |
181 | IPPROTO_TCP, | |
182 | inet->sport, usin->sin_port, sk); | |
183 | if (tmp < 0) | |
184 | return tmp; | |
185 | ||
186 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
187 | ip_rt_put(rt); | |
188 | return -ENETUNREACH; | |
189 | } | |
190 | ||
191 | if (!inet->opt || !inet->opt->srr) | |
192 | daddr = rt->rt_dst; | |
193 | ||
194 | if (!inet->saddr) | |
195 | inet->saddr = rt->rt_src; | |
196 | inet->rcv_saddr = inet->saddr; | |
197 | ||
198 | if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) { | |
199 | /* Reset inherited state */ | |
200 | tp->rx_opt.ts_recent = 0; | |
201 | tp->rx_opt.ts_recent_stamp = 0; | |
202 | tp->write_seq = 0; | |
203 | } | |
204 | ||
295ff7ed | 205 | if (tcp_death_row.sysctl_tw_recycle && |
1da177e4 LT |
206 | !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) { |
207 | struct inet_peer *peer = rt_get_peer(rt); | |
208 | ||
209 | /* VJ's idea. We save last timestamp seen from | |
210 | * the destination in peer table, when entering state TIME-WAIT | |
211 | * and initialize rx_opt.ts_recent from it, when trying new connection. | |
212 | */ | |
213 | ||
214 | if (peer && peer->tcp_ts_stamp + TCP_PAWS_MSL >= xtime.tv_sec) { | |
215 | tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp; | |
216 | tp->rx_opt.ts_recent = peer->tcp_ts; | |
217 | } | |
218 | } | |
219 | ||
220 | inet->dport = usin->sin_port; | |
221 | inet->daddr = daddr; | |
222 | ||
d83d8461 | 223 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
1da177e4 | 224 | if (inet->opt) |
d83d8461 | 225 | inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen; |
1da177e4 LT |
226 | |
227 | tp->rx_opt.mss_clamp = 536; | |
228 | ||
229 | /* Socket identity is still unknown (sport may be zero). | |
230 | * However we set state to SYN-SENT and not releasing socket | |
231 | * lock select source port, enter ourselves into the hash tables and | |
232 | * complete initialization after this. | |
233 | */ | |
234 | tcp_set_state(sk, TCP_SYN_SENT); | |
a7f5e7f1 | 235 | err = inet_hash_connect(&tcp_death_row, sk); |
1da177e4 LT |
236 | if (err) |
237 | goto failure; | |
238 | ||
5d39a795 | 239 | err = ip_route_newports(&rt, IPPROTO_TCP, inet->sport, inet->dport, sk); |
1da177e4 LT |
240 | if (err) |
241 | goto failure; | |
242 | ||
243 | /* OK, now commit destination to socket. */ | |
bcd76111 | 244 | sk->sk_gso_type = SKB_GSO_TCPV4; |
6cbb0df7 | 245 | sk_setup_caps(sk, &rt->u.dst); |
1da177e4 LT |
246 | |
247 | if (!tp->write_seq) | |
248 | tp->write_seq = secure_tcp_sequence_number(inet->saddr, | |
249 | inet->daddr, | |
250 | inet->sport, | |
251 | usin->sin_port); | |
252 | ||
253 | inet->id = tp->write_seq ^ jiffies; | |
254 | ||
255 | err = tcp_connect(sk); | |
256 | rt = NULL; | |
257 | if (err) | |
258 | goto failure; | |
259 | ||
260 | return 0; | |
261 | ||
262 | failure: | |
263 | /* This unhashes the socket and releases the local port, if necessary. */ | |
264 | tcp_set_state(sk, TCP_CLOSE); | |
265 | ip_rt_put(rt); | |
266 | sk->sk_route_caps = 0; | |
267 | inet->dport = 0; | |
268 | return err; | |
269 | } | |
270 | ||
1da177e4 LT |
271 | /* |
272 | * This routine does path mtu discovery as defined in RFC1191. | |
273 | */ | |
40efc6fa | 274 | static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu) |
1da177e4 LT |
275 | { |
276 | struct dst_entry *dst; | |
277 | struct inet_sock *inet = inet_sk(sk); | |
1da177e4 LT |
278 | |
279 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs | |
280 | * send out by Linux are always <576bytes so they should go through | |
281 | * unfragmented). | |
282 | */ | |
283 | if (sk->sk_state == TCP_LISTEN) | |
284 | return; | |
285 | ||
286 | /* We don't check in the destentry if pmtu discovery is forbidden | |
287 | * on this route. We just assume that no packet_to_big packets | |
288 | * are send back when pmtu discovery is not active. | |
289 | * There is a small race when the user changes this flag in the | |
290 | * route, but I think that's acceptable. | |
291 | */ | |
292 | if ((dst = __sk_dst_check(sk, 0)) == NULL) | |
293 | return; | |
294 | ||
295 | dst->ops->update_pmtu(dst, mtu); | |
296 | ||
297 | /* Something is about to be wrong... Remember soft error | |
298 | * for the case, if this connection will not able to recover. | |
299 | */ | |
300 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
301 | sk->sk_err_soft = EMSGSIZE; | |
302 | ||
303 | mtu = dst_mtu(dst); | |
304 | ||
305 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
d83d8461 | 306 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
307 | tcp_sync_mss(sk, mtu); |
308 | ||
309 | /* Resend the TCP packet because it's | |
310 | * clear that the old packet has been | |
311 | * dropped. This is the new "fast" path mtu | |
312 | * discovery. | |
313 | */ | |
314 | tcp_simple_retransmit(sk); | |
315 | } /* else let the usual retransmit timer handle it */ | |
316 | } | |
317 | ||
318 | /* | |
319 | * This routine is called by the ICMP module when it gets some | |
320 | * sort of error condition. If err < 0 then the socket should | |
321 | * be closed and the error returned to the user. If err > 0 | |
322 | * it's just the icmp type << 8 | icmp code. After adjustment | |
323 | * header points to the first 8 bytes of the tcp header. We need | |
324 | * to find the appropriate port. | |
325 | * | |
326 | * The locking strategy used here is very "optimistic". When | |
327 | * someone else accesses the socket the ICMP is just dropped | |
328 | * and for some paths there is no check at all. | |
329 | * A more general error queue to queue errors for later handling | |
330 | * is probably better. | |
331 | * | |
332 | */ | |
333 | ||
334 | void tcp_v4_err(struct sk_buff *skb, u32 info) | |
335 | { | |
336 | struct iphdr *iph = (struct iphdr *)skb->data; | |
337 | struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2)); | |
338 | struct tcp_sock *tp; | |
339 | struct inet_sock *inet; | |
340 | int type = skb->h.icmph->type; | |
341 | int code = skb->h.icmph->code; | |
342 | struct sock *sk; | |
343 | __u32 seq; | |
344 | int err; | |
345 | ||
346 | if (skb->len < (iph->ihl << 2) + 8) { | |
347 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
348 | return; | |
349 | } | |
350 | ||
e48c414e | 351 | sk = inet_lookup(&tcp_hashinfo, iph->daddr, th->dest, iph->saddr, |
463c84b9 | 352 | th->source, inet_iif(skb)); |
1da177e4 LT |
353 | if (!sk) { |
354 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
355 | return; | |
356 | } | |
357 | if (sk->sk_state == TCP_TIME_WAIT) { | |
8feaf0c0 | 358 | inet_twsk_put((struct inet_timewait_sock *)sk); |
1da177e4 LT |
359 | return; |
360 | } | |
361 | ||
362 | bh_lock_sock(sk); | |
363 | /* If too many ICMPs get dropped on busy | |
364 | * servers this needs to be solved differently. | |
365 | */ | |
366 | if (sock_owned_by_user(sk)) | |
367 | NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS); | |
368 | ||
369 | if (sk->sk_state == TCP_CLOSE) | |
370 | goto out; | |
371 | ||
372 | tp = tcp_sk(sk); | |
373 | seq = ntohl(th->seq); | |
374 | if (sk->sk_state != TCP_LISTEN && | |
375 | !between(seq, tp->snd_una, tp->snd_nxt)) { | |
376 | NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS); | |
377 | goto out; | |
378 | } | |
379 | ||
380 | switch (type) { | |
381 | case ICMP_SOURCE_QUENCH: | |
382 | /* Just silently ignore these. */ | |
383 | goto out; | |
384 | case ICMP_PARAMETERPROB: | |
385 | err = EPROTO; | |
386 | break; | |
387 | case ICMP_DEST_UNREACH: | |
388 | if (code > NR_ICMP_UNREACH) | |
389 | goto out; | |
390 | ||
391 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
392 | if (!sock_owned_by_user(sk)) | |
393 | do_pmtu_discovery(sk, iph, info); | |
394 | goto out; | |
395 | } | |
396 | ||
397 | err = icmp_err_convert[code].errno; | |
398 | break; | |
399 | case ICMP_TIME_EXCEEDED: | |
400 | err = EHOSTUNREACH; | |
401 | break; | |
402 | default: | |
403 | goto out; | |
404 | } | |
405 | ||
406 | switch (sk->sk_state) { | |
60236fdd | 407 | struct request_sock *req, **prev; |
1da177e4 LT |
408 | case TCP_LISTEN: |
409 | if (sock_owned_by_user(sk)) | |
410 | goto out; | |
411 | ||
463c84b9 ACM |
412 | req = inet_csk_search_req(sk, &prev, th->dest, |
413 | iph->daddr, iph->saddr); | |
1da177e4 LT |
414 | if (!req) |
415 | goto out; | |
416 | ||
417 | /* ICMPs are not backlogged, hence we cannot get | |
418 | an established socket here. | |
419 | */ | |
420 | BUG_TRAP(!req->sk); | |
421 | ||
2e6599cb | 422 | if (seq != tcp_rsk(req)->snt_isn) { |
1da177e4 LT |
423 | NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS); |
424 | goto out; | |
425 | } | |
426 | ||
427 | /* | |
428 | * Still in SYN_RECV, just remove it silently. | |
429 | * There is no good way to pass the error to the newly | |
430 | * created socket, and POSIX does not want network | |
431 | * errors returned from accept(). | |
432 | */ | |
463c84b9 | 433 | inet_csk_reqsk_queue_drop(sk, req, prev); |
1da177e4 LT |
434 | goto out; |
435 | ||
436 | case TCP_SYN_SENT: | |
437 | case TCP_SYN_RECV: /* Cannot happen. | |
438 | It can f.e. if SYNs crossed. | |
439 | */ | |
440 | if (!sock_owned_by_user(sk)) { | |
1da177e4 LT |
441 | sk->sk_err = err; |
442 | ||
443 | sk->sk_error_report(sk); | |
444 | ||
445 | tcp_done(sk); | |
446 | } else { | |
447 | sk->sk_err_soft = err; | |
448 | } | |
449 | goto out; | |
450 | } | |
451 | ||
452 | /* If we've already connected we will keep trying | |
453 | * until we time out, or the user gives up. | |
454 | * | |
455 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
456 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
457 | * but it is obsoleted by pmtu discovery). | |
458 | * | |
459 | * Note, that in modern internet, where routing is unreliable | |
460 | * and in each dark corner broken firewalls sit, sending random | |
461 | * errors ordered by their masters even this two messages finally lose | |
462 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
463 | * | |
464 | * Now we are in compliance with RFCs. | |
465 | * --ANK (980905) | |
466 | */ | |
467 | ||
468 | inet = inet_sk(sk); | |
469 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
470 | sk->sk_err = err; | |
471 | sk->sk_error_report(sk); | |
472 | } else { /* Only an error on timeout */ | |
473 | sk->sk_err_soft = err; | |
474 | } | |
475 | ||
476 | out: | |
477 | bh_unlock_sock(sk); | |
478 | sock_put(sk); | |
479 | } | |
480 | ||
481 | /* This routine computes an IPv4 TCP checksum. */ | |
8292a17a | 482 | void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb) |
1da177e4 LT |
483 | { |
484 | struct inet_sock *inet = inet_sk(sk); | |
8292a17a | 485 | struct tcphdr *th = skb->h.th; |
1da177e4 | 486 | |
84fa7933 | 487 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1da177e4 LT |
488 | th->check = ~tcp_v4_check(th, len, inet->saddr, inet->daddr, 0); |
489 | skb->csum = offsetof(struct tcphdr, check); | |
490 | } else { | |
491 | th->check = tcp_v4_check(th, len, inet->saddr, inet->daddr, | |
492 | csum_partial((char *)th, | |
493 | th->doff << 2, | |
494 | skb->csum)); | |
495 | } | |
496 | } | |
497 | ||
a430a43d HX |
498 | int tcp_v4_gso_send_check(struct sk_buff *skb) |
499 | { | |
500 | struct iphdr *iph; | |
501 | struct tcphdr *th; | |
502 | ||
503 | if (!pskb_may_pull(skb, sizeof(*th))) | |
504 | return -EINVAL; | |
505 | ||
506 | iph = skb->nh.iph; | |
507 | th = skb->h.th; | |
508 | ||
509 | th->check = 0; | |
510 | th->check = ~tcp_v4_check(th, skb->len, iph->saddr, iph->daddr, 0); | |
511 | skb->csum = offsetof(struct tcphdr, check); | |
84fa7933 | 512 | skb->ip_summed = CHECKSUM_PARTIAL; |
a430a43d HX |
513 | return 0; |
514 | } | |
515 | ||
1da177e4 LT |
516 | /* |
517 | * This routine will send an RST to the other tcp. | |
518 | * | |
519 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
520 | * for reset. | |
521 | * Answer: if a packet caused RST, it is not for a socket | |
522 | * existing in our system, if it is matched to a socket, | |
523 | * it is just duplicate segment or bug in other side's TCP. | |
524 | * So that we build reply only basing on parameters | |
525 | * arrived with segment. | |
526 | * Exception: precedence violation. We do not implement it in any case. | |
527 | */ | |
528 | ||
529 | static void tcp_v4_send_reset(struct sk_buff *skb) | |
530 | { | |
531 | struct tcphdr *th = skb->h.th; | |
532 | struct tcphdr rth; | |
533 | struct ip_reply_arg arg; | |
534 | ||
535 | /* Never send a reset in response to a reset. */ | |
536 | if (th->rst) | |
537 | return; | |
538 | ||
539 | if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL) | |
540 | return; | |
541 | ||
542 | /* Swap the send and the receive. */ | |
543 | memset(&rth, 0, sizeof(struct tcphdr)); | |
544 | rth.dest = th->source; | |
545 | rth.source = th->dest; | |
546 | rth.doff = sizeof(struct tcphdr) / 4; | |
547 | rth.rst = 1; | |
548 | ||
549 | if (th->ack) { | |
550 | rth.seq = th->ack_seq; | |
551 | } else { | |
552 | rth.ack = 1; | |
553 | rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
554 | skb->len - (th->doff << 2)); | |
555 | } | |
556 | ||
557 | memset(&arg, 0, sizeof arg); | |
558 | arg.iov[0].iov_base = (unsigned char *)&rth; | |
559 | arg.iov[0].iov_len = sizeof rth; | |
560 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, | |
561 | skb->nh.iph->saddr, /*XXX*/ | |
562 | sizeof(struct tcphdr), IPPROTO_TCP, 0); | |
563 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
564 | ||
565 | ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth); | |
566 | ||
567 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
568 | TCP_INC_STATS_BH(TCP_MIB_OUTRSTS); | |
569 | } | |
570 | ||
571 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
572 | outside socket context is ugly, certainly. What can I do? | |
573 | */ | |
574 | ||
575 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, | |
576 | u32 win, u32 ts) | |
577 | { | |
578 | struct tcphdr *th = skb->h.th; | |
579 | struct { | |
580 | struct tcphdr th; | |
581 | u32 tsopt[3]; | |
582 | } rep; | |
583 | struct ip_reply_arg arg; | |
584 | ||
585 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
586 | memset(&arg, 0, sizeof arg); | |
587 | ||
588 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
589 | arg.iov[0].iov_len = sizeof(rep.th); | |
590 | if (ts) { | |
591 | rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
592 | (TCPOPT_TIMESTAMP << 8) | | |
593 | TCPOLEN_TIMESTAMP); | |
594 | rep.tsopt[1] = htonl(tcp_time_stamp); | |
595 | rep.tsopt[2] = htonl(ts); | |
596 | arg.iov[0].iov_len = sizeof(rep); | |
597 | } | |
598 | ||
599 | /* Swap the send and the receive. */ | |
600 | rep.th.dest = th->source; | |
601 | rep.th.source = th->dest; | |
602 | rep.th.doff = arg.iov[0].iov_len / 4; | |
603 | rep.th.seq = htonl(seq); | |
604 | rep.th.ack_seq = htonl(ack); | |
605 | rep.th.ack = 1; | |
606 | rep.th.window = htons(win); | |
607 | ||
608 | arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, | |
609 | skb->nh.iph->saddr, /*XXX*/ | |
610 | arg.iov[0].iov_len, IPPROTO_TCP, 0); | |
611 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
612 | ||
613 | ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len); | |
614 | ||
615 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
616 | } | |
617 | ||
618 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
619 | { | |
8feaf0c0 ACM |
620 | struct inet_timewait_sock *tw = inet_twsk(sk); |
621 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sk); | |
1da177e4 | 622 | |
8feaf0c0 ACM |
623 | tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
624 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, tcptw->tw_ts_recent); | |
1da177e4 | 625 | |
8feaf0c0 | 626 | inet_twsk_put(tw); |
1da177e4 LT |
627 | } |
628 | ||
60236fdd | 629 | static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req) |
1da177e4 | 630 | { |
2e6599cb | 631 | tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, |
1da177e4 LT |
632 | req->ts_recent); |
633 | } | |
634 | ||
1da177e4 LT |
635 | /* |
636 | * Send a SYN-ACK after having received an ACK. | |
60236fdd | 637 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
638 | * socket. |
639 | */ | |
60236fdd | 640 | static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req, |
1da177e4 LT |
641 | struct dst_entry *dst) |
642 | { | |
2e6599cb | 643 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
644 | int err = -1; |
645 | struct sk_buff * skb; | |
646 | ||
647 | /* First, grab a route. */ | |
463c84b9 | 648 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
1da177e4 LT |
649 | goto out; |
650 | ||
651 | skb = tcp_make_synack(sk, dst, req); | |
652 | ||
653 | if (skb) { | |
654 | struct tcphdr *th = skb->h.th; | |
655 | ||
656 | th->check = tcp_v4_check(th, skb->len, | |
2e6599cb ACM |
657 | ireq->loc_addr, |
658 | ireq->rmt_addr, | |
1da177e4 LT |
659 | csum_partial((char *)th, skb->len, |
660 | skb->csum)); | |
661 | ||
2e6599cb ACM |
662 | err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, |
663 | ireq->rmt_addr, | |
664 | ireq->opt); | |
1da177e4 LT |
665 | if (err == NET_XMIT_CN) |
666 | err = 0; | |
667 | } | |
668 | ||
669 | out: | |
670 | dst_release(dst); | |
671 | return err; | |
672 | } | |
673 | ||
674 | /* | |
60236fdd | 675 | * IPv4 request_sock destructor. |
1da177e4 | 676 | */ |
60236fdd | 677 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 678 | { |
a51482bd | 679 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
680 | } |
681 | ||
80e40daa | 682 | #ifdef CONFIG_SYN_COOKIES |
40efc6fa | 683 | static void syn_flood_warning(struct sk_buff *skb) |
1da177e4 LT |
684 | { |
685 | static unsigned long warntime; | |
686 | ||
687 | if (time_after(jiffies, (warntime + HZ * 60))) { | |
688 | warntime = jiffies; | |
689 | printk(KERN_INFO | |
690 | "possible SYN flooding on port %d. Sending cookies.\n", | |
691 | ntohs(skb->h.th->dest)); | |
692 | } | |
693 | } | |
80e40daa | 694 | #endif |
1da177e4 LT |
695 | |
696 | /* | |
60236fdd | 697 | * Save and compile IPv4 options into the request_sock if needed. |
1da177e4 | 698 | */ |
40efc6fa SH |
699 | static struct ip_options *tcp_v4_save_options(struct sock *sk, |
700 | struct sk_buff *skb) | |
1da177e4 LT |
701 | { |
702 | struct ip_options *opt = &(IPCB(skb)->opt); | |
703 | struct ip_options *dopt = NULL; | |
704 | ||
705 | if (opt && opt->optlen) { | |
706 | int opt_size = optlength(opt); | |
707 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
708 | if (dopt) { | |
709 | if (ip_options_echo(dopt, skb)) { | |
710 | kfree(dopt); | |
711 | dopt = NULL; | |
712 | } | |
713 | } | |
714 | } | |
715 | return dopt; | |
716 | } | |
717 | ||
60236fdd | 718 | struct request_sock_ops tcp_request_sock_ops = { |
1da177e4 | 719 | .family = PF_INET, |
2e6599cb | 720 | .obj_size = sizeof(struct tcp_request_sock), |
1da177e4 | 721 | .rtx_syn_ack = tcp_v4_send_synack, |
60236fdd ACM |
722 | .send_ack = tcp_v4_reqsk_send_ack, |
723 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 LT |
724 | .send_reset = tcp_v4_send_reset, |
725 | }; | |
726 | ||
6d6ee43e ACM |
727 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
728 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
729 | .twsk_unique = tcp_twsk_unique, | |
730 | }; | |
731 | ||
1da177e4 LT |
732 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
733 | { | |
2e6599cb | 734 | struct inet_request_sock *ireq; |
1da177e4 | 735 | struct tcp_options_received tmp_opt; |
60236fdd | 736 | struct request_sock *req; |
1da177e4 LT |
737 | __u32 saddr = skb->nh.iph->saddr; |
738 | __u32 daddr = skb->nh.iph->daddr; | |
739 | __u32 isn = TCP_SKB_CB(skb)->when; | |
740 | struct dst_entry *dst = NULL; | |
741 | #ifdef CONFIG_SYN_COOKIES | |
742 | int want_cookie = 0; | |
743 | #else | |
744 | #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */ | |
745 | #endif | |
746 | ||
747 | /* Never answer to SYNs send to broadcast or multicast */ | |
748 | if (((struct rtable *)skb->dst)->rt_flags & | |
749 | (RTCF_BROADCAST | RTCF_MULTICAST)) | |
750 | goto drop; | |
751 | ||
752 | /* TW buckets are converted to open requests without | |
753 | * limitations, they conserve resources and peer is | |
754 | * evidently real one. | |
755 | */ | |
463c84b9 | 756 | if (inet_csk_reqsk_queue_is_full(sk) && !isn) { |
1da177e4 LT |
757 | #ifdef CONFIG_SYN_COOKIES |
758 | if (sysctl_tcp_syncookies) { | |
759 | want_cookie = 1; | |
760 | } else | |
761 | #endif | |
762 | goto drop; | |
763 | } | |
764 | ||
765 | /* Accept backlog is full. If we have already queued enough | |
766 | * of warm entries in syn queue, drop request. It is better than | |
767 | * clogging syn queue with openreqs with exponentially increasing | |
768 | * timeout. | |
769 | */ | |
463c84b9 | 770 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) |
1da177e4 LT |
771 | goto drop; |
772 | ||
60236fdd | 773 | req = reqsk_alloc(&tcp_request_sock_ops); |
1da177e4 LT |
774 | if (!req) |
775 | goto drop; | |
776 | ||
777 | tcp_clear_options(&tmp_opt); | |
778 | tmp_opt.mss_clamp = 536; | |
779 | tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss; | |
780 | ||
781 | tcp_parse_options(skb, &tmp_opt, 0); | |
782 | ||
783 | if (want_cookie) { | |
784 | tcp_clear_options(&tmp_opt); | |
785 | tmp_opt.saw_tstamp = 0; | |
786 | } | |
787 | ||
788 | if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) { | |
789 | /* Some OSes (unknown ones, but I see them on web server, which | |
790 | * contains information interesting only for windows' | |
791 | * users) do not send their stamp in SYN. It is easy case. | |
792 | * We simply do not advertise TS support. | |
793 | */ | |
794 | tmp_opt.saw_tstamp = 0; | |
795 | tmp_opt.tstamp_ok = 0; | |
796 | } | |
797 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; | |
798 | ||
799 | tcp_openreq_init(req, &tmp_opt, skb); | |
800 | ||
4237c75c VY |
801 | if (security_inet_conn_request(sk, skb, req)) |
802 | goto drop_and_free; | |
803 | ||
2e6599cb ACM |
804 | ireq = inet_rsk(req); |
805 | ireq->loc_addr = daddr; | |
806 | ireq->rmt_addr = saddr; | |
807 | ireq->opt = tcp_v4_save_options(sk, skb); | |
1da177e4 LT |
808 | if (!want_cookie) |
809 | TCP_ECN_create_request(req, skb->h.th); | |
810 | ||
811 | if (want_cookie) { | |
812 | #ifdef CONFIG_SYN_COOKIES | |
813 | syn_flood_warning(skb); | |
814 | #endif | |
815 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); | |
816 | } else if (!isn) { | |
817 | struct inet_peer *peer = NULL; | |
818 | ||
819 | /* VJ's idea. We save last timestamp seen | |
820 | * from the destination in peer table, when entering | |
821 | * state TIME-WAIT, and check against it before | |
822 | * accepting new connection request. | |
823 | * | |
824 | * If "isn" is not zero, this request hit alive | |
825 | * timewait bucket, so that all the necessary checks | |
826 | * are made in the function processing timewait state. | |
827 | */ | |
828 | if (tmp_opt.saw_tstamp && | |
295ff7ed | 829 | tcp_death_row.sysctl_tw_recycle && |
463c84b9 | 830 | (dst = inet_csk_route_req(sk, req)) != NULL && |
1da177e4 LT |
831 | (peer = rt_get_peer((struct rtable *)dst)) != NULL && |
832 | peer->v4daddr == saddr) { | |
833 | if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL && | |
834 | (s32)(peer->tcp_ts - req->ts_recent) > | |
835 | TCP_PAWS_WINDOW) { | |
836 | NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED); | |
837 | dst_release(dst); | |
838 | goto drop_and_free; | |
839 | } | |
840 | } | |
841 | /* Kill the following clause, if you dislike this way. */ | |
842 | else if (!sysctl_tcp_syncookies && | |
463c84b9 | 843 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
1da177e4 LT |
844 | (sysctl_max_syn_backlog >> 2)) && |
845 | (!peer || !peer->tcp_ts_stamp) && | |
846 | (!dst || !dst_metric(dst, RTAX_RTT))) { | |
847 | /* Without syncookies last quarter of | |
848 | * backlog is filled with destinations, | |
849 | * proven to be alive. | |
850 | * It means that we continue to communicate | |
851 | * to destinations, already remembered | |
852 | * to the moment of synflood. | |
853 | */ | |
64ce2073 PM |
854 | LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open " |
855 | "request from %u.%u.%u.%u/%u\n", | |
856 | NIPQUAD(saddr), | |
857 | ntohs(skb->h.th->source)); | |
1da177e4 LT |
858 | dst_release(dst); |
859 | goto drop_and_free; | |
860 | } | |
861 | ||
862 | isn = tcp_v4_init_sequence(sk, skb); | |
863 | } | |
2e6599cb | 864 | tcp_rsk(req)->snt_isn = isn; |
1da177e4 LT |
865 | |
866 | if (tcp_v4_send_synack(sk, req, dst)) | |
867 | goto drop_and_free; | |
868 | ||
869 | if (want_cookie) { | |
60236fdd | 870 | reqsk_free(req); |
1da177e4 | 871 | } else { |
3f421baa | 872 | inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
1da177e4 LT |
873 | } |
874 | return 0; | |
875 | ||
876 | drop_and_free: | |
60236fdd | 877 | reqsk_free(req); |
1da177e4 | 878 | drop: |
1da177e4 LT |
879 | return 0; |
880 | } | |
881 | ||
882 | ||
883 | /* | |
884 | * The three way handshake has completed - we got a valid synack - | |
885 | * now create the new socket. | |
886 | */ | |
887 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
60236fdd | 888 | struct request_sock *req, |
1da177e4 LT |
889 | struct dst_entry *dst) |
890 | { | |
2e6599cb | 891 | struct inet_request_sock *ireq; |
1da177e4 LT |
892 | struct inet_sock *newinet; |
893 | struct tcp_sock *newtp; | |
894 | struct sock *newsk; | |
895 | ||
896 | if (sk_acceptq_is_full(sk)) | |
897 | goto exit_overflow; | |
898 | ||
463c84b9 | 899 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
1da177e4 LT |
900 | goto exit; |
901 | ||
902 | newsk = tcp_create_openreq_child(sk, req, skb); | |
903 | if (!newsk) | |
904 | goto exit; | |
905 | ||
bcd76111 | 906 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
6cbb0df7 | 907 | sk_setup_caps(newsk, dst); |
1da177e4 LT |
908 | |
909 | newtp = tcp_sk(newsk); | |
910 | newinet = inet_sk(newsk); | |
2e6599cb ACM |
911 | ireq = inet_rsk(req); |
912 | newinet->daddr = ireq->rmt_addr; | |
913 | newinet->rcv_saddr = ireq->loc_addr; | |
914 | newinet->saddr = ireq->loc_addr; | |
915 | newinet->opt = ireq->opt; | |
916 | ireq->opt = NULL; | |
463c84b9 | 917 | newinet->mc_index = inet_iif(skb); |
1da177e4 | 918 | newinet->mc_ttl = skb->nh.iph->ttl; |
d83d8461 | 919 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
1da177e4 | 920 | if (newinet->opt) |
d83d8461 | 921 | inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen; |
1da177e4 LT |
922 | newinet->id = newtp->write_seq ^ jiffies; |
923 | ||
5d424d5a | 924 | tcp_mtup_init(newsk); |
1da177e4 LT |
925 | tcp_sync_mss(newsk, dst_mtu(dst)); |
926 | newtp->advmss = dst_metric(dst, RTAX_ADVMSS); | |
927 | tcp_initialize_rcv_mss(newsk); | |
928 | ||
f3f05f70 | 929 | __inet_hash(&tcp_hashinfo, newsk, 0); |
2d8c4ce5 | 930 | __inet_inherit_port(&tcp_hashinfo, sk, newsk); |
1da177e4 LT |
931 | |
932 | return newsk; | |
933 | ||
934 | exit_overflow: | |
935 | NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS); | |
936 | exit: | |
937 | NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS); | |
938 | dst_release(dst); | |
939 | return NULL; | |
940 | } | |
941 | ||
942 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
943 | { | |
944 | struct tcphdr *th = skb->h.th; | |
945 | struct iphdr *iph = skb->nh.iph; | |
1da177e4 | 946 | struct sock *nsk; |
60236fdd | 947 | struct request_sock **prev; |
1da177e4 | 948 | /* Find possible connection requests. */ |
463c84b9 ACM |
949 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, |
950 | iph->saddr, iph->daddr); | |
1da177e4 LT |
951 | if (req) |
952 | return tcp_check_req(sk, skb, req, prev); | |
953 | ||
8f491069 HX |
954 | nsk = inet_lookup_established(&tcp_hashinfo, skb->nh.iph->saddr, |
955 | th->source, skb->nh.iph->daddr, | |
956 | th->dest, inet_iif(skb)); | |
1da177e4 LT |
957 | |
958 | if (nsk) { | |
959 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
960 | bh_lock_sock(nsk); | |
961 | return nsk; | |
962 | } | |
8feaf0c0 | 963 | inet_twsk_put((struct inet_timewait_sock *)nsk); |
1da177e4 LT |
964 | return NULL; |
965 | } | |
966 | ||
967 | #ifdef CONFIG_SYN_COOKIES | |
968 | if (!th->rst && !th->syn && th->ack) | |
969 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); | |
970 | #endif | |
971 | return sk; | |
972 | } | |
973 | ||
974 | static int tcp_v4_checksum_init(struct sk_buff *skb) | |
975 | { | |
84fa7933 | 976 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
1da177e4 | 977 | if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr, |
fb286bb2 HX |
978 | skb->nh.iph->daddr, skb->csum)) { |
979 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1da177e4 | 980 | return 0; |
fb286bb2 | 981 | } |
1da177e4 | 982 | } |
fb286bb2 HX |
983 | |
984 | skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr, skb->nh.iph->daddr, | |
985 | skb->len, IPPROTO_TCP, 0); | |
986 | ||
1da177e4 | 987 | if (skb->len <= 76) { |
fb286bb2 | 988 | return __skb_checksum_complete(skb); |
1da177e4 LT |
989 | } |
990 | return 0; | |
991 | } | |
992 | ||
993 | ||
994 | /* The socket must have it's spinlock held when we get | |
995 | * here. | |
996 | * | |
997 | * We have a potential double-lock case here, so even when | |
998 | * doing backlog processing we use the BH locking scheme. | |
999 | * This is because we cannot sleep with the original spinlock | |
1000 | * held. | |
1001 | */ | |
1002 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1003 | { | |
1004 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ | |
1005 | TCP_CHECK_TIMER(sk); | |
1006 | if (tcp_rcv_established(sk, skb, skb->h.th, skb->len)) | |
1007 | goto reset; | |
1008 | TCP_CHECK_TIMER(sk); | |
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb)) | |
1013 | goto csum_err; | |
1014 | ||
1015 | if (sk->sk_state == TCP_LISTEN) { | |
1016 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1017 | if (!nsk) | |
1018 | goto discard; | |
1019 | ||
1020 | if (nsk != sk) { | |
1021 | if (tcp_child_process(sk, nsk, skb)) | |
1022 | goto reset; | |
1023 | return 0; | |
1024 | } | |
1025 | } | |
1026 | ||
1027 | TCP_CHECK_TIMER(sk); | |
1028 | if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len)) | |
1029 | goto reset; | |
1030 | TCP_CHECK_TIMER(sk); | |
1031 | return 0; | |
1032 | ||
1033 | reset: | |
1034 | tcp_v4_send_reset(skb); | |
1035 | discard: | |
1036 | kfree_skb(skb); | |
1037 | /* Be careful here. If this function gets more complicated and | |
1038 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1039 | * might be destroyed here. This current version compiles correctly, | |
1040 | * but you have been warned. | |
1041 | */ | |
1042 | return 0; | |
1043 | ||
1044 | csum_err: | |
1045 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1046 | goto discard; | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * From tcp_input.c | |
1051 | */ | |
1052 | ||
1053 | int tcp_v4_rcv(struct sk_buff *skb) | |
1054 | { | |
1055 | struct tcphdr *th; | |
1056 | struct sock *sk; | |
1057 | int ret; | |
1058 | ||
1059 | if (skb->pkt_type != PACKET_HOST) | |
1060 | goto discard_it; | |
1061 | ||
1062 | /* Count it even if it's bad */ | |
1063 | TCP_INC_STATS_BH(TCP_MIB_INSEGS); | |
1064 | ||
1065 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1066 | goto discard_it; | |
1067 | ||
1068 | th = skb->h.th; | |
1069 | ||
1070 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1071 | goto bad_packet; | |
1072 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1073 | goto discard_it; | |
1074 | ||
1075 | /* An explanation is required here, I think. | |
1076 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1077 | * provided case of th->doff==0 is eliminated. |
1da177e4 LT |
1078 | * So, we defer the checks. */ |
1079 | if ((skb->ip_summed != CHECKSUM_UNNECESSARY && | |
fb286bb2 | 1080 | tcp_v4_checksum_init(skb))) |
1da177e4 LT |
1081 | goto bad_packet; |
1082 | ||
1083 | th = skb->h.th; | |
1084 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); | |
1085 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1086 | skb->len - th->doff * 4); | |
1087 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1088 | TCP_SKB_CB(skb)->when = 0; | |
1089 | TCP_SKB_CB(skb)->flags = skb->nh.iph->tos; | |
1090 | TCP_SKB_CB(skb)->sacked = 0; | |
1091 | ||
e48c414e | 1092 | sk = __inet_lookup(&tcp_hashinfo, skb->nh.iph->saddr, th->source, |
8f491069 | 1093 | skb->nh.iph->daddr, th->dest, |
463c84b9 | 1094 | inet_iif(skb)); |
1da177e4 LT |
1095 | |
1096 | if (!sk) | |
1097 | goto no_tcp_socket; | |
1098 | ||
1099 | process: | |
1100 | if (sk->sk_state == TCP_TIME_WAIT) | |
1101 | goto do_time_wait; | |
1102 | ||
1103 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1104 | goto discard_and_relse; | |
b59c2701 | 1105 | nf_reset(skb); |
1da177e4 | 1106 | |
fda9ef5d | 1107 | if (sk_filter(sk, skb)) |
1da177e4 LT |
1108 | goto discard_and_relse; |
1109 | ||
1110 | skb->dev = NULL; | |
1111 | ||
c6366184 | 1112 | bh_lock_sock_nested(sk); |
1da177e4 LT |
1113 | ret = 0; |
1114 | if (!sock_owned_by_user(sk)) { | |
1a2449a8 CL |
1115 | #ifdef CONFIG_NET_DMA |
1116 | struct tcp_sock *tp = tcp_sk(sk); | |
1117 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
1118 | tp->ucopy.dma_chan = get_softnet_dma(); | |
1119 | if (tp->ucopy.dma_chan) | |
1da177e4 | 1120 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 CL |
1121 | else |
1122 | #endif | |
1123 | { | |
1124 | if (!tcp_prequeue(sk, skb)) | |
1125 | ret = tcp_v4_do_rcv(sk, skb); | |
1126 | } | |
1da177e4 LT |
1127 | } else |
1128 | sk_add_backlog(sk, skb); | |
1129 | bh_unlock_sock(sk); | |
1130 | ||
1131 | sock_put(sk); | |
1132 | ||
1133 | return ret; | |
1134 | ||
1135 | no_tcp_socket: | |
1136 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1137 | goto discard_it; | |
1138 | ||
1139 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1140 | bad_packet: | |
1141 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1142 | } else { | |
1143 | tcp_v4_send_reset(skb); | |
1144 | } | |
1145 | ||
1146 | discard_it: | |
1147 | /* Discard frame. */ | |
1148 | kfree_skb(skb); | |
1149 | return 0; | |
1150 | ||
1151 | discard_and_relse: | |
1152 | sock_put(sk); | |
1153 | goto discard_it; | |
1154 | ||
1155 | do_time_wait: | |
1156 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
8feaf0c0 | 1157 | inet_twsk_put((struct inet_timewait_sock *) sk); |
1da177e4 LT |
1158 | goto discard_it; |
1159 | } | |
1160 | ||
1161 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1162 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
8feaf0c0 | 1163 | inet_twsk_put((struct inet_timewait_sock *) sk); |
1da177e4 LT |
1164 | goto discard_it; |
1165 | } | |
8feaf0c0 ACM |
1166 | switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk, |
1167 | skb, th)) { | |
1da177e4 | 1168 | case TCP_TW_SYN: { |
33b62231 ACM |
1169 | struct sock *sk2 = inet_lookup_listener(&tcp_hashinfo, |
1170 | skb->nh.iph->daddr, | |
8f491069 | 1171 | th->dest, |
463c84b9 | 1172 | inet_iif(skb)); |
1da177e4 | 1173 | if (sk2) { |
295ff7ed ACM |
1174 | inet_twsk_deschedule((struct inet_timewait_sock *)sk, |
1175 | &tcp_death_row); | |
8feaf0c0 | 1176 | inet_twsk_put((struct inet_timewait_sock *)sk); |
1da177e4 LT |
1177 | sk = sk2; |
1178 | goto process; | |
1179 | } | |
1180 | /* Fall through to ACK */ | |
1181 | } | |
1182 | case TCP_TW_ACK: | |
1183 | tcp_v4_timewait_ack(sk, skb); | |
1184 | break; | |
1185 | case TCP_TW_RST: | |
1186 | goto no_tcp_socket; | |
1187 | case TCP_TW_SUCCESS:; | |
1188 | } | |
1189 | goto discard_it; | |
1190 | } | |
1191 | ||
1da177e4 LT |
1192 | /* VJ's idea. Save last timestamp seen from this destination |
1193 | * and hold it at least for normal timewait interval to use for duplicate | |
1194 | * segment detection in subsequent connections, before they enter synchronized | |
1195 | * state. | |
1196 | */ | |
1197 | ||
1198 | int tcp_v4_remember_stamp(struct sock *sk) | |
1199 | { | |
1200 | struct inet_sock *inet = inet_sk(sk); | |
1201 | struct tcp_sock *tp = tcp_sk(sk); | |
1202 | struct rtable *rt = (struct rtable *)__sk_dst_get(sk); | |
1203 | struct inet_peer *peer = NULL; | |
1204 | int release_it = 0; | |
1205 | ||
1206 | if (!rt || rt->rt_dst != inet->daddr) { | |
1207 | peer = inet_getpeer(inet->daddr, 1); | |
1208 | release_it = 1; | |
1209 | } else { | |
1210 | if (!rt->peer) | |
1211 | rt_bind_peer(rt, 1); | |
1212 | peer = rt->peer; | |
1213 | } | |
1214 | ||
1215 | if (peer) { | |
1216 | if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 || | |
1217 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && | |
1218 | peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) { | |
1219 | peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp; | |
1220 | peer->tcp_ts = tp->rx_opt.ts_recent; | |
1221 | } | |
1222 | if (release_it) | |
1223 | inet_putpeer(peer); | |
1224 | return 1; | |
1225 | } | |
1226 | ||
1227 | return 0; | |
1228 | } | |
1229 | ||
8feaf0c0 | 1230 | int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw) |
1da177e4 | 1231 | { |
8feaf0c0 | 1232 | struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1); |
1da177e4 LT |
1233 | |
1234 | if (peer) { | |
8feaf0c0 ACM |
1235 | const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
1236 | ||
1237 | if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 || | |
1da177e4 | 1238 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec && |
8feaf0c0 ACM |
1239 | peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) { |
1240 | peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp; | |
1241 | peer->tcp_ts = tcptw->tw_ts_recent; | |
1da177e4 LT |
1242 | } |
1243 | inet_putpeer(peer); | |
1244 | return 1; | |
1245 | } | |
1246 | ||
1247 | return 0; | |
1248 | } | |
1249 | ||
8292a17a | 1250 | struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
1251 | .queue_xmit = ip_queue_xmit, |
1252 | .send_check = tcp_v4_send_check, | |
1253 | .rebuild_header = inet_sk_rebuild_header, | |
1254 | .conn_request = tcp_v4_conn_request, | |
1255 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
1256 | .remember_stamp = tcp_v4_remember_stamp, | |
1257 | .net_header_len = sizeof(struct iphdr), | |
1258 | .setsockopt = ip_setsockopt, | |
1259 | .getsockopt = ip_getsockopt, | |
1260 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
1261 | .sockaddr_len = sizeof(struct sockaddr_in), | |
3fdadf7d | 1262 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
1263 | .compat_setsockopt = compat_ip_setsockopt, |
1264 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 1265 | #endif |
1da177e4 LT |
1266 | }; |
1267 | ||
1268 | /* NOTE: A lot of things set to zero explicitly by call to | |
1269 | * sk_alloc() so need not be done here. | |
1270 | */ | |
1271 | static int tcp_v4_init_sock(struct sock *sk) | |
1272 | { | |
6687e988 | 1273 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
1274 | struct tcp_sock *tp = tcp_sk(sk); |
1275 | ||
1276 | skb_queue_head_init(&tp->out_of_order_queue); | |
1277 | tcp_init_xmit_timers(sk); | |
1278 | tcp_prequeue_init(tp); | |
1279 | ||
6687e988 | 1280 | icsk->icsk_rto = TCP_TIMEOUT_INIT; |
1da177e4 LT |
1281 | tp->mdev = TCP_TIMEOUT_INIT; |
1282 | ||
1283 | /* So many TCP implementations out there (incorrectly) count the | |
1284 | * initial SYN frame in their delayed-ACK and congestion control | |
1285 | * algorithms that we must have the following bandaid to talk | |
1286 | * efficiently to them. -DaveM | |
1287 | */ | |
1288 | tp->snd_cwnd = 2; | |
1289 | ||
1290 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
1291 | * initialization of these values. | |
1292 | */ | |
1293 | tp->snd_ssthresh = 0x7fffffff; /* Infinity */ | |
1294 | tp->snd_cwnd_clamp = ~0; | |
c1b4a7e6 | 1295 | tp->mss_cache = 536; |
1da177e4 LT |
1296 | |
1297 | tp->reordering = sysctl_tcp_reordering; | |
6687e988 | 1298 | icsk->icsk_ca_ops = &tcp_init_congestion_ops; |
1da177e4 LT |
1299 | |
1300 | sk->sk_state = TCP_CLOSE; | |
1301 | ||
1302 | sk->sk_write_space = sk_stream_write_space; | |
1303 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
1304 | ||
8292a17a | 1305 | icsk->icsk_af_ops = &ipv4_specific; |
d83d8461 | 1306 | icsk->icsk_sync_mss = tcp_sync_mss; |
1da177e4 LT |
1307 | |
1308 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; | |
1309 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
1310 | ||
1311 | atomic_inc(&tcp_sockets_allocated); | |
1312 | ||
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | int tcp_v4_destroy_sock(struct sock *sk) | |
1317 | { | |
1318 | struct tcp_sock *tp = tcp_sk(sk); | |
1319 | ||
1320 | tcp_clear_xmit_timers(sk); | |
1321 | ||
6687e988 | 1322 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1323 | |
1da177e4 LT |
1324 | /* Cleanup up the write buffer. */ |
1325 | sk_stream_writequeue_purge(sk); | |
1326 | ||
1327 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
1328 | __skb_queue_purge(&tp->out_of_order_queue); | |
1329 | ||
1a2449a8 CL |
1330 | #ifdef CONFIG_NET_DMA |
1331 | /* Cleans up our sk_async_wait_queue */ | |
1332 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
1333 | #endif | |
1334 | ||
1da177e4 LT |
1335 | /* Clean prequeue, it must be empty really */ |
1336 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1337 | ||
1338 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1339 | if (inet_csk(sk)->icsk_bind_hash) |
2d8c4ce5 | 1340 | inet_put_port(&tcp_hashinfo, sk); |
1da177e4 LT |
1341 | |
1342 | /* | |
1343 | * If sendmsg cached page exists, toss it. | |
1344 | */ | |
1345 | if (sk->sk_sndmsg_page) { | |
1346 | __free_page(sk->sk_sndmsg_page); | |
1347 | sk->sk_sndmsg_page = NULL; | |
1348 | } | |
1349 | ||
1350 | atomic_dec(&tcp_sockets_allocated); | |
1351 | ||
1352 | return 0; | |
1353 | } | |
1354 | ||
1355 | EXPORT_SYMBOL(tcp_v4_destroy_sock); | |
1356 | ||
1357 | #ifdef CONFIG_PROC_FS | |
1358 | /* Proc filesystem TCP sock list dumping. */ | |
1359 | ||
8feaf0c0 | 1360 | static inline struct inet_timewait_sock *tw_head(struct hlist_head *head) |
1da177e4 LT |
1361 | { |
1362 | return hlist_empty(head) ? NULL : | |
8feaf0c0 | 1363 | list_entry(head->first, struct inet_timewait_sock, tw_node); |
1da177e4 LT |
1364 | } |
1365 | ||
8feaf0c0 | 1366 | static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw) |
1da177e4 LT |
1367 | { |
1368 | return tw->tw_node.next ? | |
1369 | hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; | |
1370 | } | |
1371 | ||
1372 | static void *listening_get_next(struct seq_file *seq, void *cur) | |
1373 | { | |
463c84b9 | 1374 | struct inet_connection_sock *icsk; |
1da177e4 LT |
1375 | struct hlist_node *node; |
1376 | struct sock *sk = cur; | |
1377 | struct tcp_iter_state* st = seq->private; | |
1378 | ||
1379 | if (!sk) { | |
1380 | st->bucket = 0; | |
6e04e021 | 1381 | sk = sk_head(&tcp_hashinfo.listening_hash[0]); |
1da177e4 LT |
1382 | goto get_sk; |
1383 | } | |
1384 | ||
1385 | ++st->num; | |
1386 | ||
1387 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
60236fdd | 1388 | struct request_sock *req = cur; |
1da177e4 | 1389 | |
463c84b9 | 1390 | icsk = inet_csk(st->syn_wait_sk); |
1da177e4 LT |
1391 | req = req->dl_next; |
1392 | while (1) { | |
1393 | while (req) { | |
60236fdd | 1394 | if (req->rsk_ops->family == st->family) { |
1da177e4 LT |
1395 | cur = req; |
1396 | goto out; | |
1397 | } | |
1398 | req = req->dl_next; | |
1399 | } | |
1400 | if (++st->sbucket >= TCP_SYNQ_HSIZE) | |
1401 | break; | |
1402 | get_req: | |
463c84b9 | 1403 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; |
1da177e4 LT |
1404 | } |
1405 | sk = sk_next(st->syn_wait_sk); | |
1406 | st->state = TCP_SEQ_STATE_LISTENING; | |
463c84b9 | 1407 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 1408 | } else { |
463c84b9 ACM |
1409 | icsk = inet_csk(sk); |
1410 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1411 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
1da177e4 | 1412 | goto start_req; |
463c84b9 | 1413 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 LT |
1414 | sk = sk_next(sk); |
1415 | } | |
1416 | get_sk: | |
1417 | sk_for_each_from(sk, node) { | |
1418 | if (sk->sk_family == st->family) { | |
1419 | cur = sk; | |
1420 | goto out; | |
1421 | } | |
463c84b9 ACM |
1422 | icsk = inet_csk(sk); |
1423 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1424 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
1da177e4 LT |
1425 | start_req: |
1426 | st->uid = sock_i_uid(sk); | |
1427 | st->syn_wait_sk = sk; | |
1428 | st->state = TCP_SEQ_STATE_OPENREQ; | |
1429 | st->sbucket = 0; | |
1430 | goto get_req; | |
1431 | } | |
463c84b9 | 1432 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 1433 | } |
0f7ff927 | 1434 | if (++st->bucket < INET_LHTABLE_SIZE) { |
6e04e021 | 1435 | sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]); |
1da177e4 LT |
1436 | goto get_sk; |
1437 | } | |
1438 | cur = NULL; | |
1439 | out: | |
1440 | return cur; | |
1441 | } | |
1442 | ||
1443 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
1444 | { | |
1445 | void *rc = listening_get_next(seq, NULL); | |
1446 | ||
1447 | while (rc && *pos) { | |
1448 | rc = listening_get_next(seq, rc); | |
1449 | --*pos; | |
1450 | } | |
1451 | return rc; | |
1452 | } | |
1453 | ||
1454 | static void *established_get_first(struct seq_file *seq) | |
1455 | { | |
1456 | struct tcp_iter_state* st = seq->private; | |
1457 | void *rc = NULL; | |
1458 | ||
6e04e021 | 1459 | for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) { |
1da177e4 LT |
1460 | struct sock *sk; |
1461 | struct hlist_node *node; | |
8feaf0c0 | 1462 | struct inet_timewait_sock *tw; |
1da177e4 LT |
1463 | |
1464 | /* We can reschedule _before_ having picked the target: */ | |
1465 | cond_resched_softirq(); | |
1466 | ||
6e04e021 ACM |
1467 | read_lock(&tcp_hashinfo.ehash[st->bucket].lock); |
1468 | sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { | |
1da177e4 LT |
1469 | if (sk->sk_family != st->family) { |
1470 | continue; | |
1471 | } | |
1472 | rc = sk; | |
1473 | goto out; | |
1474 | } | |
1475 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
8feaf0c0 ACM |
1476 | inet_twsk_for_each(tw, node, |
1477 | &tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain) { | |
1da177e4 LT |
1478 | if (tw->tw_family != st->family) { |
1479 | continue; | |
1480 | } | |
1481 | rc = tw; | |
1482 | goto out; | |
1483 | } | |
6e04e021 | 1484 | read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); |
1da177e4 LT |
1485 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
1486 | } | |
1487 | out: | |
1488 | return rc; | |
1489 | } | |
1490 | ||
1491 | static void *established_get_next(struct seq_file *seq, void *cur) | |
1492 | { | |
1493 | struct sock *sk = cur; | |
8feaf0c0 | 1494 | struct inet_timewait_sock *tw; |
1da177e4 LT |
1495 | struct hlist_node *node; |
1496 | struct tcp_iter_state* st = seq->private; | |
1497 | ||
1498 | ++st->num; | |
1499 | ||
1500 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { | |
1501 | tw = cur; | |
1502 | tw = tw_next(tw); | |
1503 | get_tw: | |
1504 | while (tw && tw->tw_family != st->family) { | |
1505 | tw = tw_next(tw); | |
1506 | } | |
1507 | if (tw) { | |
1508 | cur = tw; | |
1509 | goto out; | |
1510 | } | |
6e04e021 | 1511 | read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); |
1da177e4 LT |
1512 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
1513 | ||
1514 | /* We can reschedule between buckets: */ | |
1515 | cond_resched_softirq(); | |
1516 | ||
6e04e021 ACM |
1517 | if (++st->bucket < tcp_hashinfo.ehash_size) { |
1518 | read_lock(&tcp_hashinfo.ehash[st->bucket].lock); | |
1519 | sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain); | |
1da177e4 LT |
1520 | } else { |
1521 | cur = NULL; | |
1522 | goto out; | |
1523 | } | |
1524 | } else | |
1525 | sk = sk_next(sk); | |
1526 | ||
1527 | sk_for_each_from(sk, node) { | |
1528 | if (sk->sk_family == st->family) | |
1529 | goto found; | |
1530 | } | |
1531 | ||
1532 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
6e04e021 | 1533 | tw = tw_head(&tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain); |
1da177e4 LT |
1534 | goto get_tw; |
1535 | found: | |
1536 | cur = sk; | |
1537 | out: | |
1538 | return cur; | |
1539 | } | |
1540 | ||
1541 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
1542 | { | |
1543 | void *rc = established_get_first(seq); | |
1544 | ||
1545 | while (rc && pos) { | |
1546 | rc = established_get_next(seq, rc); | |
1547 | --pos; | |
1548 | } | |
1549 | return rc; | |
1550 | } | |
1551 | ||
1552 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
1553 | { | |
1554 | void *rc; | |
1555 | struct tcp_iter_state* st = seq->private; | |
1556 | ||
f3f05f70 | 1557 | inet_listen_lock(&tcp_hashinfo); |
1da177e4 LT |
1558 | st->state = TCP_SEQ_STATE_LISTENING; |
1559 | rc = listening_get_idx(seq, &pos); | |
1560 | ||
1561 | if (!rc) { | |
f3f05f70 | 1562 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
1563 | local_bh_disable(); |
1564 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
1565 | rc = established_get_idx(seq, pos); | |
1566 | } | |
1567 | ||
1568 | return rc; | |
1569 | } | |
1570 | ||
1571 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) | |
1572 | { | |
1573 | struct tcp_iter_state* st = seq->private; | |
1574 | st->state = TCP_SEQ_STATE_LISTENING; | |
1575 | st->num = 0; | |
1576 | return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
1577 | } | |
1578 | ||
1579 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
1580 | { | |
1581 | void *rc = NULL; | |
1582 | struct tcp_iter_state* st; | |
1583 | ||
1584 | if (v == SEQ_START_TOKEN) { | |
1585 | rc = tcp_get_idx(seq, 0); | |
1586 | goto out; | |
1587 | } | |
1588 | st = seq->private; | |
1589 | ||
1590 | switch (st->state) { | |
1591 | case TCP_SEQ_STATE_OPENREQ: | |
1592 | case TCP_SEQ_STATE_LISTENING: | |
1593 | rc = listening_get_next(seq, v); | |
1594 | if (!rc) { | |
f3f05f70 | 1595 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
1596 | local_bh_disable(); |
1597 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
1598 | rc = established_get_first(seq); | |
1599 | } | |
1600 | break; | |
1601 | case TCP_SEQ_STATE_ESTABLISHED: | |
1602 | case TCP_SEQ_STATE_TIME_WAIT: | |
1603 | rc = established_get_next(seq, v); | |
1604 | break; | |
1605 | } | |
1606 | out: | |
1607 | ++*pos; | |
1608 | return rc; | |
1609 | } | |
1610 | ||
1611 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
1612 | { | |
1613 | struct tcp_iter_state* st = seq->private; | |
1614 | ||
1615 | switch (st->state) { | |
1616 | case TCP_SEQ_STATE_OPENREQ: | |
1617 | if (v) { | |
463c84b9 ACM |
1618 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); |
1619 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1da177e4 LT |
1620 | } |
1621 | case TCP_SEQ_STATE_LISTENING: | |
1622 | if (v != SEQ_START_TOKEN) | |
f3f05f70 | 1623 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
1624 | break; |
1625 | case TCP_SEQ_STATE_TIME_WAIT: | |
1626 | case TCP_SEQ_STATE_ESTABLISHED: | |
1627 | if (v) | |
6e04e021 | 1628 | read_unlock(&tcp_hashinfo.ehash[st->bucket].lock); |
1da177e4 LT |
1629 | local_bh_enable(); |
1630 | break; | |
1631 | } | |
1632 | } | |
1633 | ||
1634 | static int tcp_seq_open(struct inode *inode, struct file *file) | |
1635 | { | |
1636 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; | |
1637 | struct seq_file *seq; | |
1638 | struct tcp_iter_state *s; | |
1639 | int rc; | |
1640 | ||
1641 | if (unlikely(afinfo == NULL)) | |
1642 | return -EINVAL; | |
1643 | ||
0da974f4 | 1644 | s = kzalloc(sizeof(*s), GFP_KERNEL); |
1da177e4 LT |
1645 | if (!s) |
1646 | return -ENOMEM; | |
1da177e4 LT |
1647 | s->family = afinfo->family; |
1648 | s->seq_ops.start = tcp_seq_start; | |
1649 | s->seq_ops.next = tcp_seq_next; | |
1650 | s->seq_ops.show = afinfo->seq_show; | |
1651 | s->seq_ops.stop = tcp_seq_stop; | |
1652 | ||
1653 | rc = seq_open(file, &s->seq_ops); | |
1654 | if (rc) | |
1655 | goto out_kfree; | |
1656 | seq = file->private_data; | |
1657 | seq->private = s; | |
1658 | out: | |
1659 | return rc; | |
1660 | out_kfree: | |
1661 | kfree(s); | |
1662 | goto out; | |
1663 | } | |
1664 | ||
1665 | int tcp_proc_register(struct tcp_seq_afinfo *afinfo) | |
1666 | { | |
1667 | int rc = 0; | |
1668 | struct proc_dir_entry *p; | |
1669 | ||
1670 | if (!afinfo) | |
1671 | return -EINVAL; | |
1672 | afinfo->seq_fops->owner = afinfo->owner; | |
1673 | afinfo->seq_fops->open = tcp_seq_open; | |
1674 | afinfo->seq_fops->read = seq_read; | |
1675 | afinfo->seq_fops->llseek = seq_lseek; | |
1676 | afinfo->seq_fops->release = seq_release_private; | |
1677 | ||
1678 | p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops); | |
1679 | if (p) | |
1680 | p->data = afinfo; | |
1681 | else | |
1682 | rc = -ENOMEM; | |
1683 | return rc; | |
1684 | } | |
1685 | ||
1686 | void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo) | |
1687 | { | |
1688 | if (!afinfo) | |
1689 | return; | |
1690 | proc_net_remove(afinfo->name); | |
1691 | memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); | |
1692 | } | |
1693 | ||
60236fdd | 1694 | static void get_openreq4(struct sock *sk, struct request_sock *req, |
1da177e4 LT |
1695 | char *tmpbuf, int i, int uid) |
1696 | { | |
2e6599cb | 1697 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
1698 | int ttd = req->expires - jiffies; |
1699 | ||
1700 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" | |
1701 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p", | |
1702 | i, | |
2e6599cb | 1703 | ireq->loc_addr, |
1da177e4 | 1704 | ntohs(inet_sk(sk)->sport), |
2e6599cb ACM |
1705 | ireq->rmt_addr, |
1706 | ntohs(ireq->rmt_port), | |
1da177e4 LT |
1707 | TCP_SYN_RECV, |
1708 | 0, 0, /* could print option size, but that is af dependent. */ | |
1709 | 1, /* timers active (only the expire timer) */ | |
1710 | jiffies_to_clock_t(ttd), | |
1711 | req->retrans, | |
1712 | uid, | |
1713 | 0, /* non standard timer */ | |
1714 | 0, /* open_requests have no inode */ | |
1715 | atomic_read(&sk->sk_refcnt), | |
1716 | req); | |
1717 | } | |
1718 | ||
1719 | static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i) | |
1720 | { | |
1721 | int timer_active; | |
1722 | unsigned long timer_expires; | |
1723 | struct tcp_sock *tp = tcp_sk(sp); | |
463c84b9 | 1724 | const struct inet_connection_sock *icsk = inet_csk(sp); |
1da177e4 LT |
1725 | struct inet_sock *inet = inet_sk(sp); |
1726 | unsigned int dest = inet->daddr; | |
1727 | unsigned int src = inet->rcv_saddr; | |
1728 | __u16 destp = ntohs(inet->dport); | |
1729 | __u16 srcp = ntohs(inet->sport); | |
1730 | ||
463c84b9 | 1731 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) { |
1da177e4 | 1732 | timer_active = 1; |
463c84b9 ACM |
1733 | timer_expires = icsk->icsk_timeout; |
1734 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 1735 | timer_active = 4; |
463c84b9 | 1736 | timer_expires = icsk->icsk_timeout; |
1da177e4 LT |
1737 | } else if (timer_pending(&sp->sk_timer)) { |
1738 | timer_active = 2; | |
1739 | timer_expires = sp->sk_timer.expires; | |
1740 | } else { | |
1741 | timer_active = 0; | |
1742 | timer_expires = jiffies; | |
1743 | } | |
1744 | ||
1745 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " | |
1746 | "%08X %5d %8d %lu %d %p %u %u %u %u %d", | |
1747 | i, src, srcp, dest, destp, sp->sk_state, | |
47da8ee6 SS |
1748 | tp->write_seq - tp->snd_una, |
1749 | (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq), | |
1da177e4 LT |
1750 | timer_active, |
1751 | jiffies_to_clock_t(timer_expires - jiffies), | |
463c84b9 | 1752 | icsk->icsk_retransmits, |
1da177e4 | 1753 | sock_i_uid(sp), |
6687e988 | 1754 | icsk->icsk_probes_out, |
1da177e4 LT |
1755 | sock_i_ino(sp), |
1756 | atomic_read(&sp->sk_refcnt), sp, | |
463c84b9 ACM |
1757 | icsk->icsk_rto, |
1758 | icsk->icsk_ack.ato, | |
1759 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, | |
1da177e4 LT |
1760 | tp->snd_cwnd, |
1761 | tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh); | |
1762 | } | |
1763 | ||
8feaf0c0 | 1764 | static void get_timewait4_sock(struct inet_timewait_sock *tw, char *tmpbuf, int i) |
1da177e4 LT |
1765 | { |
1766 | unsigned int dest, src; | |
1767 | __u16 destp, srcp; | |
1768 | int ttd = tw->tw_ttd - jiffies; | |
1769 | ||
1770 | if (ttd < 0) | |
1771 | ttd = 0; | |
1772 | ||
1773 | dest = tw->tw_daddr; | |
1774 | src = tw->tw_rcv_saddr; | |
1775 | destp = ntohs(tw->tw_dport); | |
1776 | srcp = ntohs(tw->tw_sport); | |
1777 | ||
1778 | sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" | |
1779 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p", | |
1780 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, | |
1781 | 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, | |
1782 | atomic_read(&tw->tw_refcnt), tw); | |
1783 | } | |
1784 | ||
1785 | #define TMPSZ 150 | |
1786 | ||
1787 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
1788 | { | |
1789 | struct tcp_iter_state* st; | |
1790 | char tmpbuf[TMPSZ + 1]; | |
1791 | ||
1792 | if (v == SEQ_START_TOKEN) { | |
1793 | seq_printf(seq, "%-*s\n", TMPSZ - 1, | |
1794 | " sl local_address rem_address st tx_queue " | |
1795 | "rx_queue tr tm->when retrnsmt uid timeout " | |
1796 | "inode"); | |
1797 | goto out; | |
1798 | } | |
1799 | st = seq->private; | |
1800 | ||
1801 | switch (st->state) { | |
1802 | case TCP_SEQ_STATE_LISTENING: | |
1803 | case TCP_SEQ_STATE_ESTABLISHED: | |
1804 | get_tcp4_sock(v, tmpbuf, st->num); | |
1805 | break; | |
1806 | case TCP_SEQ_STATE_OPENREQ: | |
1807 | get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid); | |
1808 | break; | |
1809 | case TCP_SEQ_STATE_TIME_WAIT: | |
1810 | get_timewait4_sock(v, tmpbuf, st->num); | |
1811 | break; | |
1812 | } | |
1813 | seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf); | |
1814 | out: | |
1815 | return 0; | |
1816 | } | |
1817 | ||
1818 | static struct file_operations tcp4_seq_fops; | |
1819 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { | |
1820 | .owner = THIS_MODULE, | |
1821 | .name = "tcp", | |
1822 | .family = AF_INET, | |
1823 | .seq_show = tcp4_seq_show, | |
1824 | .seq_fops = &tcp4_seq_fops, | |
1825 | }; | |
1826 | ||
1827 | int __init tcp4_proc_init(void) | |
1828 | { | |
1829 | return tcp_proc_register(&tcp4_seq_afinfo); | |
1830 | } | |
1831 | ||
1832 | void tcp4_proc_exit(void) | |
1833 | { | |
1834 | tcp_proc_unregister(&tcp4_seq_afinfo); | |
1835 | } | |
1836 | #endif /* CONFIG_PROC_FS */ | |
1837 | ||
1838 | struct proto tcp_prot = { | |
1839 | .name = "TCP", | |
1840 | .owner = THIS_MODULE, | |
1841 | .close = tcp_close, | |
1842 | .connect = tcp_v4_connect, | |
1843 | .disconnect = tcp_disconnect, | |
463c84b9 | 1844 | .accept = inet_csk_accept, |
1da177e4 LT |
1845 | .ioctl = tcp_ioctl, |
1846 | .init = tcp_v4_init_sock, | |
1847 | .destroy = tcp_v4_destroy_sock, | |
1848 | .shutdown = tcp_shutdown, | |
1849 | .setsockopt = tcp_setsockopt, | |
1850 | .getsockopt = tcp_getsockopt, | |
1851 | .sendmsg = tcp_sendmsg, | |
1852 | .recvmsg = tcp_recvmsg, | |
1853 | .backlog_rcv = tcp_v4_do_rcv, | |
1854 | .hash = tcp_v4_hash, | |
1855 | .unhash = tcp_unhash, | |
1856 | .get_port = tcp_v4_get_port, | |
1857 | .enter_memory_pressure = tcp_enter_memory_pressure, | |
1858 | .sockets_allocated = &tcp_sockets_allocated, | |
0a5578cf | 1859 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
1860 | .memory_allocated = &tcp_memory_allocated, |
1861 | .memory_pressure = &tcp_memory_pressure, | |
1862 | .sysctl_mem = sysctl_tcp_mem, | |
1863 | .sysctl_wmem = sysctl_tcp_wmem, | |
1864 | .sysctl_rmem = sysctl_tcp_rmem, | |
1865 | .max_header = MAX_TCP_HEADER, | |
1866 | .obj_size = sizeof(struct tcp_sock), | |
6d6ee43e | 1867 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 1868 | .rsk_prot = &tcp_request_sock_ops, |
543d9cfe ACM |
1869 | #ifdef CONFIG_COMPAT |
1870 | .compat_setsockopt = compat_tcp_setsockopt, | |
1871 | .compat_getsockopt = compat_tcp_getsockopt, | |
1872 | #endif | |
1da177e4 LT |
1873 | }; |
1874 | ||
1da177e4 LT |
1875 | void __init tcp_v4_init(struct net_proto_family *ops) |
1876 | { | |
c4d93909 | 1877 | if (inet_csk_ctl_sock_create(&tcp_socket, PF_INET, SOCK_RAW, IPPROTO_TCP) < 0) |
1da177e4 | 1878 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 LT |
1879 | } |
1880 | ||
1881 | EXPORT_SYMBOL(ipv4_specific); | |
1da177e4 | 1882 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 1883 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 LT |
1884 | EXPORT_SYMBOL(tcp_unhash); |
1885 | EXPORT_SYMBOL(tcp_v4_conn_request); | |
1886 | EXPORT_SYMBOL(tcp_v4_connect); | |
1887 | EXPORT_SYMBOL(tcp_v4_do_rcv); | |
1da177e4 LT |
1888 | EXPORT_SYMBOL(tcp_v4_remember_stamp); |
1889 | EXPORT_SYMBOL(tcp_v4_send_check); | |
1890 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); | |
1891 | ||
1892 | #ifdef CONFIG_PROC_FS | |
1893 | EXPORT_SYMBOL(tcp_proc_register); | |
1894 | EXPORT_SYMBOL(tcp_proc_unregister); | |
1895 | #endif | |
1896 | EXPORT_SYMBOL(sysctl_local_port_range); | |
1da177e4 | 1897 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
1da177e4 | 1898 |