blob: 319c8852644904370759c381ce124effdcb5421a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * IPv4 specific functions
9 *
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24/*
25 * Changes:
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -070037 * request_sock handling and moved
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080040 * Added new listen semantics.
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
45 * coma.
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
51 */
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Herbert Xueb4dea52008-12-29 23:04:08 -080054#include <linux/bottom_half.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/types.h>
56#include <linux/fcntl.h>
57#include <linux/module.h>
58#include <linux/random.h>
59#include <linux/cache.h>
60#include <linux/jhash.h>
61#include <linux/init.h>
62#include <linux/times.h>
63
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020064#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <net/icmp.h>
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -070066#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <net/tcp.h>
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -030068#include <net/transp_v6.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include <net/ipv6.h>
70#include <net/inet_common.h>
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -080071#include <net/timewait_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include <net/xfrm.h>
Chris Leech1a2449a2006-05-23 18:05:53 -070073#include <net/netdma.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
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
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080081#include <linux/crypto.h>
82#include <linux/scatterlist.h>
83
Brian Haleyab32ea52006-09-22 14:15:41 -070084int sysctl_tcp_tw_reuse __read_mostly;
85int sysctl_tcp_low_latency __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080088#ifdef CONFIG_TCP_MD5SIG
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -020089static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
90 __be32 addr);
Adam Langley49a72df2008-07-19 00:01:42 -070091static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
92 __be32 daddr, __be32 saddr, struct tcphdr *th);
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +090093#else
94static inline
95struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
96{
97 return NULL;
98}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080099#endif
100
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800101struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Gerrit Renkera94f7232006-11-10 14:06:49 -0800103static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700105 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
106 ip_hdr(skb)->saddr,
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700107 tcp_hdr(skb)->dest,
108 tcp_hdr(skb)->source);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109}
110
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800111int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
112{
113 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
114 struct tcp_sock *tp = tcp_sk(sk);
115
116 /* With PAWS, it is safe from the viewpoint
117 of data integrity. Even without PAWS it is safe provided sequence
118 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
119
120 Actually, the idea is close to VJ's one, only timestamp cache is
121 held not per host, but per port pair and TW bucket is used as state
122 holder.
123
124 If TW bucket has been already destroyed we fall back to VJ's scheme
125 and use initial timestamp retrieved from peer table.
126 */
127 if (tcptw->tw_ts_recent_stamp &&
128 (twp == NULL || (sysctl_tcp_tw_reuse &&
James Morris9d729f72007-03-04 16:12:44 -0800129 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800130 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
131 if (tp->write_seq == 0)
132 tp->write_seq = 1;
133 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
134 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
135 sock_hold(sktw);
136 return 1;
137 }
138
139 return 0;
140}
141
142EXPORT_SYMBOL_GPL(tcp_twsk_unique);
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144/* This will initiate an outgoing connection. */
145int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
146{
147 struct inet_sock *inet = inet_sk(sk);
148 struct tcp_sock *tp = tcp_sk(sk);
149 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
150 struct rtable *rt;
Al Virobada8ad2006-09-26 21:27:15 -0700151 __be32 daddr, nexthop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 int tmp;
153 int err;
154
155 if (addr_len < sizeof(struct sockaddr_in))
156 return -EINVAL;
157
158 if (usin->sin_family != AF_INET)
159 return -EAFNOSUPPORT;
160
161 nexthop = daddr = usin->sin_addr.s_addr;
162 if (inet->opt && inet->opt->srr) {
163 if (!daddr)
164 return -EINVAL;
165 nexthop = inet->opt->faddr;
166 }
167
168 tmp = ip_route_connect(&rt, nexthop, inet->saddr,
169 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
170 IPPROTO_TCP,
David S. Miller8eb90862007-02-08 02:09:21 -0800171 inet->sport, usin->sin_port, sk, 1);
Wei Dong584bdf82007-05-31 22:49:28 -0700172 if (tmp < 0) {
173 if (tmp == -ENETUNREACH)
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700174 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 return tmp;
Wei Dong584bdf82007-05-31 22:49:28 -0700176 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
179 ip_rt_put(rt);
180 return -ENETUNREACH;
181 }
182
183 if (!inet->opt || !inet->opt->srr)
184 daddr = rt->rt_dst;
185
186 if (!inet->saddr)
187 inet->saddr = rt->rt_src;
188 inet->rcv_saddr = inet->saddr;
189
190 if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) {
191 /* Reset inherited state */
192 tp->rx_opt.ts_recent = 0;
193 tp->rx_opt.ts_recent_stamp = 0;
194 tp->write_seq = 0;
195 }
196
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700197 if (tcp_death_row.sysctl_tw_recycle &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
199 struct inet_peer *peer = rt_get_peer(rt);
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200200 /*
201 * VJ's idea. We save last timestamp seen from
202 * the destination in peer table, when entering state
203 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
204 * when trying new connection.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200206 if (peer != NULL &&
James Morris9d729f72007-03-04 16:12:44 -0800207 peer->tcp_ts_stamp + TCP_PAWS_MSL >= get_seconds()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
209 tp->rx_opt.ts_recent = peer->tcp_ts;
210 }
211 }
212
213 inet->dport = usin->sin_port;
214 inet->daddr = daddr;
215
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800216 inet_csk(sk)->icsk_ext_hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 if (inet->opt)
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800218 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
220 tp->rx_opt.mss_clamp = 536;
221
222 /* Socket identity is still unknown (sport may be zero).
223 * However we set state to SYN-SENT and not releasing socket
224 * lock select source port, enter ourselves into the hash tables and
225 * complete initialization after this.
226 */
227 tcp_set_state(sk, TCP_SYN_SENT);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800228 err = inet_hash_connect(&tcp_death_row, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 if (err)
230 goto failure;
231
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200232 err = ip_route_newports(&rt, IPPROTO_TCP,
233 inet->sport, inet->dport, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 if (err)
235 goto failure;
236
237 /* OK, now commit destination to socket. */
Herbert Xubcd76112006-06-30 13:36:35 -0700238 sk->sk_gso_type = SKB_GSO_TCPV4;
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -0700239 sk_setup_caps(sk, &rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
241 if (!tp->write_seq)
242 tp->write_seq = secure_tcp_sequence_number(inet->saddr,
243 inet->daddr,
244 inet->sport,
245 usin->sin_port);
246
247 inet->id = tp->write_seq ^ jiffies;
248
249 err = tcp_connect(sk);
250 rt = NULL;
251 if (err)
252 goto failure;
253
254 return 0;
255
256failure:
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200257 /*
258 * This unhashes the socket and releases the local port,
259 * if necessary.
260 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 tcp_set_state(sk, TCP_CLOSE);
262 ip_rt_put(rt);
263 sk->sk_route_caps = 0;
264 inet->dport = 0;
265 return err;
266}
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268/*
269 * This routine does path mtu discovery as defined in RFC1191.
270 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800271static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
276 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
277 * send out by Linux are always <576bytes so they should go through
278 * unfragmented).
279 */
280 if (sk->sk_state == TCP_LISTEN)
281 return;
282
283 /* We don't check in the destentry if pmtu discovery is forbidden
284 * on this route. We just assume that no packet_to_big packets
285 * are send back when pmtu discovery is not active.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900286 * There is a small race when the user changes this flag in the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 * route, but I think that's acceptable.
288 */
289 if ((dst = __sk_dst_check(sk, 0)) == NULL)
290 return;
291
292 dst->ops->update_pmtu(dst, mtu);
293
294 /* Something is about to be wrong... Remember soft error
295 * for the case, if this connection will not able to recover.
296 */
297 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
298 sk->sk_err_soft = EMSGSIZE;
299
300 mtu = dst_mtu(dst);
301
302 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800303 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 tcp_sync_mss(sk, mtu);
305
306 /* Resend the TCP packet because it's
307 * clear that the old packet has been
308 * dropped. This is the new "fast" path mtu
309 * discovery.
310 */
311 tcp_simple_retransmit(sk);
312 } /* else let the usual retransmit timer handle it */
313}
314
315/*
316 * This routine is called by the ICMP module when it gets some
317 * sort of error condition. If err < 0 then the socket should
318 * be closed and the error returned to the user. If err > 0
319 * it's just the icmp type << 8 | icmp code. After adjustment
320 * header points to the first 8 bytes of the tcp header. We need
321 * to find the appropriate port.
322 *
323 * The locking strategy used here is very "optimistic". When
324 * someone else accesses the socket the ICMP is just dropped
325 * and for some paths there is no check at all.
326 * A more general error queue to queue errors for later handling
327 * is probably better.
328 *
329 */
330
331void tcp_v4_err(struct sk_buff *skb, u32 info)
332{
333 struct iphdr *iph = (struct iphdr *)skb->data;
334 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
335 struct tcp_sock *tp;
336 struct inet_sock *inet;
Arnaldo Carvalho de Melo88c76642007-03-13 14:43:18 -0300337 const int type = icmp_hdr(skb)->type;
338 const int code = icmp_hdr(skb)->code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 struct sock *sk;
340 __u32 seq;
341 int err;
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700342 struct net *net = dev_net(skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
344 if (skb->len < (iph->ihl << 2) + 8) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700345 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 return;
347 }
348
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700349 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
Pavel Emelyanovc67499c2008-01-31 05:06:40 -0800350 iph->saddr, th->source, inet_iif(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (!sk) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700352 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return;
354 }
355 if (sk->sk_state == TCP_TIME_WAIT) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -0700356 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return;
358 }
359
360 bh_lock_sock(sk);
361 /* If too many ICMPs get dropped on busy
362 * servers this needs to be solved differently.
363 */
364 if (sock_owned_by_user(sk))
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700365 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 if (sk->sk_state == TCP_CLOSE)
368 goto out;
369
370 tp = tcp_sk(sk);
371 seq = ntohl(th->seq);
372 if (sk->sk_state != TCP_LISTEN &&
373 !between(seq, tp->snd_una, tp->snd_nxt)) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700374 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 goto out;
376 }
377
378 switch (type) {
379 case ICMP_SOURCE_QUENCH:
380 /* Just silently ignore these. */
381 goto out;
382 case ICMP_PARAMETERPROB:
383 err = EPROTO;
384 break;
385 case ICMP_DEST_UNREACH:
386 if (code > NR_ICMP_UNREACH)
387 goto out;
388
389 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
390 if (!sock_owned_by_user(sk))
391 do_pmtu_discovery(sk, iph, info);
392 goto out;
393 }
394
395 err = icmp_err_convert[code].errno;
396 break;
397 case ICMP_TIME_EXCEEDED:
398 err = EHOSTUNREACH;
399 break;
400 default:
401 goto out;
402 }
403
404 switch (sk->sk_state) {
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700405 struct request_sock *req, **prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 case TCP_LISTEN:
407 if (sock_owned_by_user(sk))
408 goto out;
409
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700410 req = inet_csk_search_req(sk, &prev, th->dest,
411 iph->daddr, iph->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 if (!req)
413 goto out;
414
415 /* ICMPs are not backlogged, hence we cannot get
416 an established socket here.
417 */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700418 WARN_ON(req->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700420 if (seq != tcp_rsk(req)->snt_isn) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700421 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 goto out;
423 }
424
425 /*
426 * Still in SYN_RECV, just remove it silently.
427 * There is no good way to pass the error to the newly
428 * created socket, and POSIX does not want network
429 * errors returned from accept().
430 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700431 inet_csk_reqsk_queue_drop(sk, req, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 goto out;
433
434 case TCP_SYN_SENT:
435 case TCP_SYN_RECV: /* Cannot happen.
436 It can f.e. if SYNs crossed.
437 */
438 if (!sock_owned_by_user(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 sk->sk_err = err;
440
441 sk->sk_error_report(sk);
442
443 tcp_done(sk);
444 } else {
445 sk->sk_err_soft = err;
446 }
447 goto out;
448 }
449
450 /* If we've already connected we will keep trying
451 * until we time out, or the user gives up.
452 *
453 * rfc1122 4.2.3.9 allows to consider as hard errors
454 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
455 * but it is obsoleted by pmtu discovery).
456 *
457 * Note, that in modern internet, where routing is unreliable
458 * and in each dark corner broken firewalls sit, sending random
459 * errors ordered by their masters even this two messages finally lose
460 * their original sense (even Linux sends invalid PORT_UNREACHs)
461 *
462 * Now we are in compliance with RFCs.
463 * --ANK (980905)
464 */
465
466 inet = inet_sk(sk);
467 if (!sock_owned_by_user(sk) && inet->recverr) {
468 sk->sk_err = err;
469 sk->sk_error_report(sk);
470 } else { /* Only an error on timeout */
471 sk->sk_err_soft = err;
472 }
473
474out:
475 bh_unlock_sock(sk);
476 sock_put(sk);
477}
478
479/* This routine computes an IPv4 TCP checksum. */
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800480void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
482 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700483 struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Patrick McHardy84fa7932006-08-29 16:44:56 -0700485 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Frederik Deweerdtba7808e2007-02-04 20:15:27 -0800486 th->check = ~tcp_v4_check(len, inet->saddr,
487 inet->daddr, 0);
Herbert Xu663ead32007-04-09 11:59:07 -0700488 skb->csum_start = skb_transport_header(skb) - skb->head;
Al Viroff1dcad2006-11-20 18:07:29 -0800489 skb->csum_offset = offsetof(struct tcphdr, check);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 } else {
Frederik Deweerdtba7808e2007-02-04 20:15:27 -0800491 th->check = tcp_v4_check(len, inet->saddr, inet->daddr,
Joe Perches07f07572008-11-19 15:44:53 -0800492 csum_partial(th,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 th->doff << 2,
494 skb->csum));
495 }
496}
497
Herbert Xua430a432006-07-08 13:34:56 -0700498int tcp_v4_gso_send_check(struct sk_buff *skb)
499{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700500 const struct iphdr *iph;
Herbert Xua430a432006-07-08 13:34:56 -0700501 struct tcphdr *th;
502
503 if (!pskb_may_pull(skb, sizeof(*th)))
504 return -EINVAL;
505
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700506 iph = ip_hdr(skb);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700507 th = tcp_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -0700508
509 th->check = 0;
Frederik Deweerdtba7808e2007-02-04 20:15:27 -0800510 th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0);
Herbert Xu663ead32007-04-09 11:59:07 -0700511 skb->csum_start = skb_transport_header(skb) - skb->head;
Al Viroff1dcad2006-11-20 18:07:29 -0800512 skb->csum_offset = offsetof(struct tcphdr, check);
Patrick McHardy84fa7932006-08-29 16:44:56 -0700513 skb->ip_summed = CHECKSUM_PARTIAL;
Herbert Xua430a432006-07-08 13:34:56 -0700514 return 0;
515}
516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517/*
518 * This routine will send an RST to the other tcp.
519 *
520 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
521 * for reset.
522 * Answer: if a packet caused RST, it is not for a socket
523 * existing in our system, if it is matched to a socket,
524 * it is just duplicate segment or bug in other side's TCP.
525 * So that we build reply only basing on parameters
526 * arrived with segment.
527 * Exception: precedence violation. We do not implement it in any case.
528 */
529
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800530static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700532 struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800533 struct {
534 struct tcphdr th;
535#ifdef CONFIG_TCP_MD5SIG
Al Viro714e85b2006-11-14 20:51:49 -0800536 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800537#endif
538 } rep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 struct ip_reply_arg arg;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800540#ifdef CONFIG_TCP_MD5SIG
541 struct tcp_md5sig_key *key;
542#endif
Pavel Emelyanova86b1e32008-07-16 20:20:58 -0700543 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
545 /* Never send a reset in response to a reset. */
546 if (th->rst)
547 return;
548
Eric Dumazet511c3f92009-06-02 05:14:27 +0000549 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 return;
551
552 /* Swap the send and the receive. */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800553 memset(&rep, 0, sizeof(rep));
554 rep.th.dest = th->source;
555 rep.th.source = th->dest;
556 rep.th.doff = sizeof(struct tcphdr) / 4;
557 rep.th.rst = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559 if (th->ack) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800560 rep.th.seq = th->ack_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 } else {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800562 rep.th.ack = 1;
563 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
564 skb->len - (th->doff << 2));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
566
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200567 memset(&arg, 0, sizeof(arg));
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800568 arg.iov[0].iov_base = (unsigned char *)&rep;
569 arg.iov[0].iov_len = sizeof(rep.th);
570
571#ifdef CONFIG_TCP_MD5SIG
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700572 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800573 if (key) {
574 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
575 (TCPOPT_NOP << 16) |
576 (TCPOPT_MD5SIG << 8) |
577 TCPOLEN_MD5SIG);
578 /* Update length and the length the header thinks exists */
579 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
580 rep.th.doff = arg.iov[0].iov_len / 4;
581
Adam Langley49a72df2008-07-19 00:01:42 -0700582 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
Ilpo Järvinen78e645cb2008-10-09 14:37:47 -0700583 key, ip_hdr(skb)->saddr,
584 ip_hdr(skb)->daddr, &rep.th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800585 }
586#endif
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700587 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
588 ip_hdr(skb)->saddr, /* XXX */
Ilpo Järvinen52cd5752008-10-08 11:34:06 -0700589 arg.iov[0].iov_len, IPPROTO_TCP, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700591 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Pavel Emelyanova86b1e32008-07-16 20:20:58 -0700593 net = dev_net(skb->dst->dev);
594 ip_send_reply(net->ipv4.tcp_sock, skb,
Denis V. Lunev7feb49c2008-04-03 14:32:00 -0700595 &arg, arg.iov[0].iov_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700597 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
598 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
601/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
602 outside socket context is ugly, certainly. What can I do?
603 */
604
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900605static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
606 u32 win, u32 ts, int oif,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700607 struct tcp_md5sig_key *key,
608 int reply_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700610 struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 struct {
612 struct tcphdr th;
Al Viro714e85b2006-11-14 20:51:49 -0800613 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800614#ifdef CONFIG_TCP_MD5SIG
Al Viro714e85b2006-11-14 20:51:49 -0800615 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800616#endif
617 ];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 } rep;
619 struct ip_reply_arg arg;
Vitaliy Gusev4dd79722008-10-01 01:51:39 -0700620 struct net *net = dev_net(skb->dst->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622 memset(&rep.th, 0, sizeof(struct tcphdr));
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200623 memset(&arg, 0, sizeof(arg));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
625 arg.iov[0].iov_base = (unsigned char *)&rep;
626 arg.iov[0].iov_len = sizeof(rep.th);
627 if (ts) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800628 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
629 (TCPOPT_TIMESTAMP << 8) |
630 TCPOLEN_TIMESTAMP);
631 rep.opt[1] = htonl(tcp_time_stamp);
632 rep.opt[2] = htonl(ts);
Craig Schlentercb48cfe2007-01-09 00:11:15 -0800633 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635
636 /* Swap the send and the receive. */
637 rep.th.dest = th->source;
638 rep.th.source = th->dest;
639 rep.th.doff = arg.iov[0].iov_len / 4;
640 rep.th.seq = htonl(seq);
641 rep.th.ack_seq = htonl(ack);
642 rep.th.ack = 1;
643 rep.th.window = htons(win);
644
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800645#ifdef CONFIG_TCP_MD5SIG
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800646 if (key) {
647 int offset = (ts) ? 3 : 0;
648
649 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
650 (TCPOPT_NOP << 16) |
651 (TCPOPT_MD5SIG << 8) |
652 TCPOLEN_MD5SIG);
653 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
654 rep.th.doff = arg.iov[0].iov_len/4;
655
Adam Langley49a72df2008-07-19 00:01:42 -0700656 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
Adam Langley90b7e112008-07-31 20:49:48 -0700657 key, ip_hdr(skb)->saddr,
658 ip_hdr(skb)->daddr, &rep.th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800659 }
660#endif
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700661 arg.flags = reply_flags;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700662 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
663 ip_hdr(skb)->saddr, /* XXX */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 arg.iov[0].iov_len, IPPROTO_TCP, 0);
665 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900666 if (oif)
667 arg.bound_dev_if = oif;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Pavel Emelyanova86b1e32008-07-16 20:20:58 -0700669 ip_send_reply(net->ipv4.tcp_sock, skb,
Denis V. Lunev7feb49c2008-04-03 14:32:00 -0700670 &arg, arg.iov[0].iov_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700672 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
675static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
676{
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700677 struct inet_timewait_sock *tw = inet_twsk(sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800678 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900680 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200681 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900682 tcptw->tw_ts_recent,
683 tw->tw_bound_dev_if,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700684 tcp_twsk_md5_key(tcptw),
685 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900686 );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700688 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
Gui Jianfeng6edafaa2008-08-06 23:50:04 -0700691static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200692 struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900694 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800695 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900696 req->ts_recent,
697 0,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700698 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
699 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702/*
Kris Katterjohn9bf1d832008-02-17 22:29:19 -0800703 * Send a SYN-ACK after having received a SYN.
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700704 * This still operates on a request_sock only, not on a big
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 * socket.
706 */
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800707static int __tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
708 struct dst_entry *dst)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700710 const struct inet_request_sock *ireq = inet_rsk(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 int err = -1;
712 struct sk_buff * skb;
713
714 /* First, grab a route. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700715 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800716 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718 skb = tcp_make_synack(sk, dst, req);
719
720 if (skb) {
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700721 struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Frederik Deweerdtba7808e2007-02-04 20:15:27 -0800723 th->check = tcp_v4_check(skb->len,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700724 ireq->loc_addr,
725 ireq->rmt_addr,
Joe Perches07f07572008-11-19 15:44:53 -0800726 csum_partial(th, skb->len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 skb->csum));
728
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700729 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
730 ireq->rmt_addr,
731 ireq->opt);
Gerrit Renkerb9df3cb2006-11-14 11:21:36 -0200732 err = net_xmit_eval(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 }
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 dst_release(dst);
736 return err;
737}
738
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800739static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req)
740{
741 return __tcp_v4_send_synack(sk, req, NULL);
742}
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744/*
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700745 * IPv4 request_sock destructor.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700747static void tcp_v4_reqsk_destructor(struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Jesper Juhla51482b2005-11-08 09:41:34 -0800749 kfree(inet_rsk(req)->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
Arnaldo Carvalho de Melo80e40da2006-01-04 01:58:06 -0200752#ifdef CONFIG_SYN_COOKIES
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800753static void syn_flood_warning(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
755 static unsigned long warntime;
756
757 if (time_after(jiffies, (warntime + HZ * 60))) {
758 warntime = jiffies;
759 printk(KERN_INFO
760 "possible SYN flooding on port %d. Sending cookies.\n",
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700761 ntohs(tcp_hdr(skb)->dest));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763}
Arnaldo Carvalho de Melo80e40da2006-01-04 01:58:06 -0200764#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766/*
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700767 * Save and compile IPv4 options into the request_sock if needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800769static struct ip_options *tcp_v4_save_options(struct sock *sk,
770 struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
772 struct ip_options *opt = &(IPCB(skb)->opt);
773 struct ip_options *dopt = NULL;
774
775 if (opt && opt->optlen) {
776 int opt_size = optlength(opt);
777 dopt = kmalloc(opt_size, GFP_ATOMIC);
778 if (dopt) {
779 if (ip_options_echo(dopt, skb)) {
780 kfree(dopt);
781 dopt = NULL;
782 }
783 }
784 }
785 return dopt;
786}
787
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800788#ifdef CONFIG_TCP_MD5SIG
789/*
790 * RFC2385 MD5 checksumming requires a mapping of
791 * IP address->MD5 Key.
792 * We need to maintain these in the sk structure.
793 */
794
795/* Find the Key structure for an address. */
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200796static struct tcp_md5sig_key *
797 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800798{
799 struct tcp_sock *tp = tcp_sk(sk);
800 int i;
801
802 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
803 return NULL;
804 for (i = 0; i < tp->md5sig_info->entries4; i++) {
805 if (tp->md5sig_info->keys4[i].addr == addr)
David S. Millerf8ab18d2007-09-28 15:18:35 -0700806 return &tp->md5sig_info->keys4[i].base;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800807 }
808 return NULL;
809}
810
811struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
812 struct sock *addr_sk)
813{
814 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->daddr);
815}
816
817EXPORT_SYMBOL(tcp_v4_md5_lookup);
818
Adrian Bunkf5b99bc2006-11-30 17:22:29 -0800819static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
820 struct request_sock *req)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800821{
822 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
823}
824
825/* This can be called on a newly created socket, from other files */
826int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
827 u8 *newkey, u8 newkeylen)
828{
829 /* Add Key to the list */
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700830 struct tcp_md5sig_key *key;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800831 struct tcp_sock *tp = tcp_sk(sk);
832 struct tcp4_md5sig_key *keys;
833
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700834 key = tcp_v4_md5_do_lookup(sk, addr);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800835 if (key) {
836 /* Pre-existing entry - just update that one. */
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700837 kfree(key->key);
838 key->key = newkey;
839 key->keylen = newkeylen;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800840 } else {
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200841 struct tcp_md5sig_info *md5sig;
842
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800843 if (!tp->md5sig_info) {
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200844 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
845 GFP_ATOMIC);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800846 if (!tp->md5sig_info) {
847 kfree(newkey);
848 return -ENOMEM;
849 }
David S. Miller3d7dbea2007-06-12 14:36:42 -0700850 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800851 }
852 if (tcp_alloc_md5sig_pool() == NULL) {
853 kfree(newkey);
854 return -ENOMEM;
855 }
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200856 md5sig = tp->md5sig_info;
857
858 if (md5sig->alloced4 == md5sig->entries4) {
859 keys = kmalloc((sizeof(*keys) *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900860 (md5sig->entries4 + 1)), GFP_ATOMIC);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800861 if (!keys) {
862 kfree(newkey);
863 tcp_free_md5sig_pool();
864 return -ENOMEM;
865 }
866
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200867 if (md5sig->entries4)
868 memcpy(keys, md5sig->keys4,
869 sizeof(*keys) * md5sig->entries4);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800870
871 /* Free old key list, and reference new one */
YOSHIFUJI Hideakia80cc202007-11-20 17:30:06 -0800872 kfree(md5sig->keys4);
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200873 md5sig->keys4 = keys;
874 md5sig->alloced4++;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800875 }
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200876 md5sig->entries4++;
David S. Millerf8ab18d2007-09-28 15:18:35 -0700877 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
878 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
879 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800880 }
881 return 0;
882}
883
884EXPORT_SYMBOL(tcp_v4_md5_do_add);
885
886static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
887 u8 *newkey, u8 newkeylen)
888{
889 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->daddr,
890 newkey, newkeylen);
891}
892
893int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
894{
895 struct tcp_sock *tp = tcp_sk(sk);
896 int i;
897
898 for (i = 0; i < tp->md5sig_info->entries4; i++) {
899 if (tp->md5sig_info->keys4[i].addr == addr) {
900 /* Free the key */
David S. Millerf8ab18d2007-09-28 15:18:35 -0700901 kfree(tp->md5sig_info->keys4[i].base.key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800902 tp->md5sig_info->entries4--;
903
904 if (tp->md5sig_info->entries4 == 0) {
905 kfree(tp->md5sig_info->keys4);
906 tp->md5sig_info->keys4 = NULL;
Leigh Brown8228a18d2006-12-17 17:12:30 -0800907 tp->md5sig_info->alloced4 = 0;
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200908 } else if (tp->md5sig_info->entries4 != i) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800909 /* Need to do some manipulation */
YOSHIFUJI Hideaki354faf02007-11-20 17:30:31 -0800910 memmove(&tp->md5sig_info->keys4[i],
911 &tp->md5sig_info->keys4[i+1],
912 (tp->md5sig_info->entries4 - i) *
913 sizeof(struct tcp4_md5sig_key));
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800914 }
915 tcp_free_md5sig_pool();
916 return 0;
917 }
918 }
919 return -ENOENT;
920}
921
922EXPORT_SYMBOL(tcp_v4_md5_do_del);
923
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200924static void tcp_v4_clear_md5_list(struct sock *sk)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800925{
926 struct tcp_sock *tp = tcp_sk(sk);
927
928 /* Free each key, then the set of key keys,
929 * the crypto element, and then decrement our
930 * hold on the last resort crypto.
931 */
932 if (tp->md5sig_info->entries4) {
933 int i;
934 for (i = 0; i < tp->md5sig_info->entries4; i++)
David S. Millerf8ab18d2007-09-28 15:18:35 -0700935 kfree(tp->md5sig_info->keys4[i].base.key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800936 tp->md5sig_info->entries4 = 0;
937 tcp_free_md5sig_pool();
938 }
939 if (tp->md5sig_info->keys4) {
940 kfree(tp->md5sig_info->keys4);
941 tp->md5sig_info->keys4 = NULL;
942 tp->md5sig_info->alloced4 = 0;
943 }
944}
945
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200946static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
947 int optlen)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800948{
949 struct tcp_md5sig cmd;
950 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
951 u8 *newkey;
952
953 if (optlen < sizeof(cmd))
954 return -EINVAL;
955
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200956 if (copy_from_user(&cmd, optval, sizeof(cmd)))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800957 return -EFAULT;
958
959 if (sin->sin_family != AF_INET)
960 return -EINVAL;
961
962 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
963 if (!tcp_sk(sk)->md5sig_info)
964 return -ENOENT;
965 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
966 }
967
968 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
969 return -EINVAL;
970
971 if (!tcp_sk(sk)->md5sig_info) {
972 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200973 struct tcp_md5sig_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800974
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800975 if (!p)
976 return -EINVAL;
977
978 tp->md5sig_info = p;
David S. Miller3d7dbea2007-06-12 14:36:42 -0700979 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800980 }
981
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200982 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800983 if (!newkey)
984 return -ENOMEM;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800985 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
986 newkey, cmd.tcpm_keylen);
987}
988
Adam Langley49a72df2008-07-19 00:01:42 -0700989static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
990 __be32 daddr, __be32 saddr, int nbytes)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800991{
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800992 struct tcp4_pseudohdr *bp;
Adam Langley49a72df2008-07-19 00:01:42 -0700993 struct scatterlist sg;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800994
995 bp = &hp->md5_blk.ip4;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800996
997 /*
Adam Langley49a72df2008-07-19 00:01:42 -0700998 * 1. the TCP pseudo-header (in the order: source IP address,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800999 * destination IP address, zero-padded protocol number, and
1000 * segment length)
1001 */
1002 bp->saddr = saddr;
1003 bp->daddr = daddr;
1004 bp->pad = 0;
YOSHIFUJI Hideaki076fb722008-04-17 12:48:12 +09001005 bp->protocol = IPPROTO_TCP;
Adam Langley49a72df2008-07-19 00:01:42 -07001006 bp->len = cpu_to_be16(nbytes);
David S. Millerc7da57a2007-10-26 00:41:21 -07001007
Adam Langley49a72df2008-07-19 00:01:42 -07001008 sg_init_one(&sg, bp, sizeof(*bp));
1009 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1010}
1011
1012static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1013 __be32 daddr, __be32 saddr, struct tcphdr *th)
1014{
1015 struct tcp_md5sig_pool *hp;
1016 struct hash_desc *desc;
1017
1018 hp = tcp_get_md5sig_pool();
1019 if (!hp)
1020 goto clear_hash_noput;
1021 desc = &hp->md5_desc;
1022
1023 if (crypto_hash_init(desc))
1024 goto clear_hash;
1025 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1026 goto clear_hash;
1027 if (tcp_md5_hash_header(hp, th))
1028 goto clear_hash;
1029 if (tcp_md5_hash_key(hp, key))
1030 goto clear_hash;
1031 if (crypto_hash_final(desc, md5_hash))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001032 goto clear_hash;
1033
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001034 tcp_put_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001035 return 0;
Adam Langley49a72df2008-07-19 00:01:42 -07001036
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001037clear_hash:
1038 tcp_put_md5sig_pool();
1039clear_hash_noput:
1040 memset(md5_hash, 0, 16);
Adam Langley49a72df2008-07-19 00:01:42 -07001041 return 1;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001042}
1043
Adam Langley49a72df2008-07-19 00:01:42 -07001044int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1045 struct sock *sk, struct request_sock *req,
1046 struct sk_buff *skb)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001047{
Adam Langley49a72df2008-07-19 00:01:42 -07001048 struct tcp_md5sig_pool *hp;
1049 struct hash_desc *desc;
1050 struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001051 __be32 saddr, daddr;
1052
1053 if (sk) {
1054 saddr = inet_sk(sk)->saddr;
1055 daddr = inet_sk(sk)->daddr;
Adam Langley49a72df2008-07-19 00:01:42 -07001056 } else if (req) {
1057 saddr = inet_rsk(req)->loc_addr;
1058 daddr = inet_rsk(req)->rmt_addr;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001059 } else {
Adam Langley49a72df2008-07-19 00:01:42 -07001060 const struct iphdr *iph = ip_hdr(skb);
1061 saddr = iph->saddr;
1062 daddr = iph->daddr;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001063 }
Adam Langley49a72df2008-07-19 00:01:42 -07001064
1065 hp = tcp_get_md5sig_pool();
1066 if (!hp)
1067 goto clear_hash_noput;
1068 desc = &hp->md5_desc;
1069
1070 if (crypto_hash_init(desc))
1071 goto clear_hash;
1072
1073 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1074 goto clear_hash;
1075 if (tcp_md5_hash_header(hp, th))
1076 goto clear_hash;
1077 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1078 goto clear_hash;
1079 if (tcp_md5_hash_key(hp, key))
1080 goto clear_hash;
1081 if (crypto_hash_final(desc, md5_hash))
1082 goto clear_hash;
1083
1084 tcp_put_md5sig_pool();
1085 return 0;
1086
1087clear_hash:
1088 tcp_put_md5sig_pool();
1089clear_hash_noput:
1090 memset(md5_hash, 0, 16);
1091 return 1;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001092}
1093
Adam Langley49a72df2008-07-19 00:01:42 -07001094EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001095
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02001096static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001097{
1098 /*
1099 * This gets called for each TCP segment that arrives
1100 * so we want to be efficient.
1101 * We have 3 drop cases:
1102 * o No MD5 hash and one expected.
1103 * o MD5 hash and we're not expecting one.
1104 * o MD5 hash and its wrong.
1105 */
1106 __u8 *hash_location = NULL;
1107 struct tcp_md5sig_key *hash_expected;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001108 const struct iphdr *iph = ip_hdr(skb);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001109 struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001110 int genhash;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001111 unsigned char newhash[16];
1112
1113 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +09001114 hash_location = tcp_parse_md5sig_option(th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001115
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001116 /* We've parsed the options - do we have a hash? */
1117 if (!hash_expected && !hash_location)
1118 return 0;
1119
1120 if (hash_expected && !hash_location) {
David S. Miller785957d2008-07-30 03:03:15 -07001121 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001122 return 1;
1123 }
1124
1125 if (!hash_expected && hash_location) {
David S. Miller785957d2008-07-30 03:03:15 -07001126 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001127 return 1;
1128 }
1129
1130 /* Okay, so this is hash_expected and hash_location -
1131 * so we need to calculate the checksum.
1132 */
Adam Langley49a72df2008-07-19 00:01:42 -07001133 genhash = tcp_v4_md5_hash_skb(newhash,
1134 hash_expected,
1135 NULL, NULL, skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001136
1137 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1138 if (net_ratelimit()) {
Harvey Harrison673d57e2008-10-31 00:53:57 -07001139 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1140 &iph->saddr, ntohs(th->source),
1141 &iph->daddr, ntohs(th->dest),
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001142 genhash ? " tcp_v4_calc_md5_hash failed" : "");
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001143 }
1144 return 1;
1145 }
1146 return 0;
1147}
1148
1149#endif
1150
Eric Dumazet72a3eff2006-11-16 02:30:37 -08001151struct request_sock_ops tcp_request_sock_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 .family = PF_INET,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001153 .obj_size = sizeof(struct tcp_request_sock),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 .rtx_syn_ack = tcp_v4_send_synack,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001155 .send_ack = tcp_v4_reqsk_send_ack,
1156 .destructor = tcp_v4_reqsk_destructor,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 .send_reset = tcp_v4_send_reset,
1158};
1159
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001160#ifdef CONFIG_TCP_MD5SIG
Andrew Mortonb6332e62006-11-30 19:16:28 -08001161static struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001162 .md5_lookup = tcp_v4_reqsk_md5_lookup,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001163};
Andrew Mortonb6332e62006-11-30 19:16:28 -08001164#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001165
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08001166static struct timewait_sock_ops tcp_timewait_sock_ops = {
1167 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1168 .twsk_unique = tcp_twsk_unique,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001169 .twsk_destructor= tcp_twsk_destructor,
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08001170};
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1173{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001174 struct inet_request_sock *ireq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 struct tcp_options_received tmp_opt;
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001176 struct request_sock *req;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001177 __be32 saddr = ip_hdr(skb)->saddr;
1178 __be32 daddr = ip_hdr(skb)->daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 __u32 isn = TCP_SKB_CB(skb)->when;
1180 struct dst_entry *dst = NULL;
1181#ifdef CONFIG_SYN_COOKIES
1182 int want_cookie = 0;
1183#else
1184#define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1185#endif
1186
1187 /* Never answer to SYNs send to broadcast or multicast */
Eric Dumazet511c3f92009-06-02 05:14:27 +00001188 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 goto drop;
1190
1191 /* TW buckets are converted to open requests without
1192 * limitations, they conserve resources and peer is
1193 * evidently real one.
1194 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001195 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196#ifdef CONFIG_SYN_COOKIES
1197 if (sysctl_tcp_syncookies) {
1198 want_cookie = 1;
1199 } else
1200#endif
1201 goto drop;
1202 }
1203
1204 /* Accept backlog is full. If we have already queued enough
1205 * of warm entries in syn queue, drop request. It is better than
1206 * clogging syn queue with openreqs with exponentially increasing
1207 * timeout.
1208 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001209 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 goto drop;
1211
Arnaldo Carvalho de Meloce4a7d02008-06-10 12:39:35 -07001212 req = inet_reqsk_alloc(&tcp_request_sock_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 if (!req)
1214 goto drop;
1215
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001216#ifdef CONFIG_TCP_MD5SIG
1217 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1218#endif
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 tcp_clear_options(&tmp_opt);
1221 tmp_opt.mss_clamp = 536;
1222 tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss;
1223
1224 tcp_parse_options(skb, &tmp_opt, 0);
1225
Florian Westphal4dfc2812008-04-10 03:12:40 -07001226 if (want_cookie && !tmp_opt.saw_tstamp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 tcp_clear_options(&tmp_opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1230
1231 tcp_openreq_init(req, &tmp_opt, skb);
1232
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001233 ireq = inet_rsk(req);
1234 ireq->loc_addr = daddr;
1235 ireq->rmt_addr = saddr;
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -07001236 ireq->no_srccheck = inet_sk(sk)->transparent;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001237 ireq->opt = tcp_v4_save_options(sk, skb);
Paul Moore284904a2009-03-27 17:10:28 -04001238
1239 if (security_inet_conn_request(sk, skb, req))
1240 goto drop_and_free;
1241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (!want_cookie)
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001243 TCP_ECN_create_request(req, tcp_hdr(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
1245 if (want_cookie) {
1246#ifdef CONFIG_SYN_COOKIES
1247 syn_flood_warning(skb);
Florian Westphal4dfc2812008-04-10 03:12:40 -07001248 req->cookie_ts = tmp_opt.tstamp_ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249#endif
1250 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1251 } else if (!isn) {
1252 struct inet_peer *peer = NULL;
1253
1254 /* VJ's idea. We save last timestamp seen
1255 * from the destination in peer table, when entering
1256 * state TIME-WAIT, and check against it before
1257 * accepting new connection request.
1258 *
1259 * If "isn" is not zero, this request hit alive
1260 * timewait bucket, so that all the necessary checks
1261 * are made in the function processing timewait state.
1262 */
1263 if (tmp_opt.saw_tstamp &&
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07001264 tcp_death_row.sysctl_tw_recycle &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001265 (dst = inet_csk_route_req(sk, req)) != NULL &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1267 peer->v4daddr == saddr) {
James Morris9d729f72007-03-04 16:12:44 -08001268 if (get_seconds() < peer->tcp_ts_stamp + TCP_PAWS_MSL &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 (s32)(peer->tcp_ts - req->ts_recent) >
1270 TCP_PAWS_WINDOW) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001271 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001272 goto drop_and_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274 }
1275 /* Kill the following clause, if you dislike this way. */
1276 else if (!sysctl_tcp_syncookies &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001277 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 (sysctl_max_syn_backlog >> 2)) &&
1279 (!peer || !peer->tcp_ts_stamp) &&
1280 (!dst || !dst_metric(dst, RTAX_RTT))) {
1281 /* Without syncookies last quarter of
1282 * backlog is filled with destinations,
1283 * proven to be alive.
1284 * It means that we continue to communicate
1285 * to destinations, already remembered
1286 * to the moment of synflood.
1287 */
Harvey Harrison673d57e2008-10-31 00:53:57 -07001288 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1289 &saddr, ntohs(tcp_hdr(skb)->source));
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001290 goto drop_and_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 }
1292
Gerrit Renkera94f7232006-11-10 14:06:49 -08001293 isn = tcp_v4_init_sequence(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001295 tcp_rsk(req)->snt_isn = isn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001297 if (__tcp_v4_send_synack(sk, req, dst) || want_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 goto drop_and_free;
1299
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001300 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 return 0;
1302
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001303drop_and_release:
1304 dst_release(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305drop_and_free:
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001306 reqsk_free(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 return 0;
1309}
1310
1311
1312/*
1313 * The three way handshake has completed - we got a valid synack -
1314 * now create the new socket.
1315 */
1316struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001317 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 struct dst_entry *dst)
1319{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001320 struct inet_request_sock *ireq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 struct inet_sock *newinet;
1322 struct tcp_sock *newtp;
1323 struct sock *newsk;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001324#ifdef CONFIG_TCP_MD5SIG
1325 struct tcp_md5sig_key *key;
1326#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
1328 if (sk_acceptq_is_full(sk))
1329 goto exit_overflow;
1330
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001331 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 goto exit;
1333
1334 newsk = tcp_create_openreq_child(sk, req, skb);
1335 if (!newsk)
1336 goto exit;
1337
Herbert Xubcd76112006-06-30 13:36:35 -07001338 newsk->sk_gso_type = SKB_GSO_TCPV4;
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -07001339 sk_setup_caps(newsk, dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
1341 newtp = tcp_sk(newsk);
1342 newinet = inet_sk(newsk);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001343 ireq = inet_rsk(req);
1344 newinet->daddr = ireq->rmt_addr;
1345 newinet->rcv_saddr = ireq->loc_addr;
1346 newinet->saddr = ireq->loc_addr;
1347 newinet->opt = ireq->opt;
1348 ireq->opt = NULL;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001349 newinet->mc_index = inet_iif(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001350 newinet->mc_ttl = ip_hdr(skb)->ttl;
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08001351 inet_csk(newsk)->icsk_ext_hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 if (newinet->opt)
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08001353 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 newinet->id = newtp->write_seq ^ jiffies;
1355
John Heffner5d424d52006-03-20 17:53:41 -08001356 tcp_mtup_init(newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 tcp_sync_mss(newsk, dst_mtu(dst));
1358 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
Tom Quetchenbachf5fff5d2008-09-21 00:21:51 -07001359 if (tcp_sk(sk)->rx_opt.user_mss &&
1360 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1361 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1362
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 tcp_initialize_rcv_mss(newsk);
1364
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001365#ifdef CONFIG_TCP_MD5SIG
1366 /* Copy over the MD5 key from the original socket */
1367 if ((key = tcp_v4_md5_do_lookup(sk, newinet->daddr)) != NULL) {
1368 /*
1369 * We're using one, so create a matching key
1370 * on the newsk structure. If we fail to get
1371 * memory, then we end up not copying the key
1372 * across. Shucks.
1373 */
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -02001374 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1375 if (newkey != NULL)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001376 tcp_v4_md5_do_add(newsk, inet_sk(sk)->daddr,
1377 newkey, key->keylen);
Adam Langley49a72df2008-07-19 00:01:42 -07001378 newsk->sk_route_caps &= ~NETIF_F_GSO_MASK;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001379 }
1380#endif
1381
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001382 __inet_hash_nolisten(newsk);
1383 __inet_inherit_port(sk, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 return newsk;
1386
1387exit_overflow:
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001388 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389exit:
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001390 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 dst_release(dst);
1392 return NULL;
1393}
1394
1395static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1396{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001397 struct tcphdr *th = tcp_hdr(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001398 const struct iphdr *iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 struct sock *nsk;
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001400 struct request_sock **prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 /* Find possible connection requests. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001402 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1403 iph->saddr, iph->daddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 if (req)
1405 return tcp_check_req(sk, skb, req, prev);
1406
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001407 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
Pavel Emelyanovc67499c2008-01-31 05:06:40 -08001408 th->source, iph->daddr, th->dest, inet_iif(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
1410 if (nsk) {
1411 if (nsk->sk_state != TCP_TIME_WAIT) {
1412 bh_lock_sock(nsk);
1413 return nsk;
1414 }
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001415 inet_twsk_put(inet_twsk(nsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 return NULL;
1417 }
1418
1419#ifdef CONFIG_SYN_COOKIES
1420 if (!th->rst && !th->syn && th->ack)
1421 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1422#endif
1423 return sk;
1424}
1425
Al Virob51655b2006-11-14 21:40:42 -08001426static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001428 const struct iphdr *iph = ip_hdr(skb);
1429
Patrick McHardy84fa7932006-08-29 16:44:56 -07001430 if (skb->ip_summed == CHECKSUM_COMPLETE) {
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001431 if (!tcp_v4_check(skb->len, iph->saddr,
1432 iph->daddr, skb->csum)) {
Herbert Xufb286bb2005-11-10 13:01:24 -08001433 skb->ip_summed = CHECKSUM_UNNECESSARY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 return 0;
Herbert Xufb286bb2005-11-10 13:01:24 -08001435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 }
Herbert Xufb286bb2005-11-10 13:01:24 -08001437
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001438 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
Herbert Xufb286bb2005-11-10 13:01:24 -08001439 skb->len, IPPROTO_TCP, 0);
1440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 if (skb->len <= 76) {
Herbert Xufb286bb2005-11-10 13:01:24 -08001442 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
1444 return 0;
1445}
1446
1447
1448/* The socket must have it's spinlock held when we get
1449 * here.
1450 *
1451 * We have a potential double-lock case here, so even when
1452 * doing backlog processing we use the BH locking scheme.
1453 * This is because we cannot sleep with the original spinlock
1454 * held.
1455 */
1456int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1457{
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001458 struct sock *rsk;
1459#ifdef CONFIG_TCP_MD5SIG
1460 /*
1461 * We really want to reject the packet as early as possible
1462 * if:
1463 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1464 * o There is an MD5 option and we're not expecting one
1465 */
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02001466 if (tcp_v4_inbound_md5_hash(sk, skb))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001467 goto discard;
1468#endif
1469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1471 TCP_CHECK_TIMER(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001472 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001473 rsk = sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001475 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 TCP_CHECK_TIMER(sk);
1477 return 0;
1478 }
1479
Arnaldo Carvalho de Meloab6a5bb2007-03-18 17:43:48 -07001480 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 goto csum_err;
1482
1483 if (sk->sk_state == TCP_LISTEN) {
1484 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1485 if (!nsk)
1486 goto discard;
1487
1488 if (nsk != sk) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001489 if (tcp_child_process(sk, nsk, skb)) {
1490 rsk = nsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return 0;
1494 }
1495 }
1496
1497 TCP_CHECK_TIMER(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001498 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001499 rsk = sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001501 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 TCP_CHECK_TIMER(sk);
1503 return 0;
1504
1505reset:
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001506 tcp_v4_send_reset(rsk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507discard:
1508 kfree_skb(skb);
1509 /* Be careful here. If this function gets more complicated and
1510 * gcc suffers from register pressure on the x86, sk (in %ebx)
1511 * might be destroyed here. This current version compiles correctly,
1512 * but you have been warned.
1513 */
1514 return 0;
1515
1516csum_err:
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001517 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 goto discard;
1519}
1520
1521/*
1522 * From tcp_input.c
1523 */
1524
1525int tcp_v4_rcv(struct sk_buff *skb)
1526{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001527 const struct iphdr *iph;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 struct tcphdr *th;
1529 struct sock *sk;
1530 int ret;
Pavel Emelyanova86b1e32008-07-16 20:20:58 -07001531 struct net *net = dev_net(skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
1533 if (skb->pkt_type != PACKET_HOST)
1534 goto discard_it;
1535
1536 /* Count it even if it's bad */
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001537 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
1539 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1540 goto discard_it;
1541
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001542 th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
1544 if (th->doff < sizeof(struct tcphdr) / 4)
1545 goto bad_packet;
1546 if (!pskb_may_pull(skb, th->doff * 4))
1547 goto discard_it;
1548
1549 /* An explanation is required here, I think.
1550 * Packet length and doff are validated by header prediction,
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08001551 * provided case of th->doff==0 is eliminated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 * So, we defer the checks. */
Herbert Xu60476372007-04-09 11:59:39 -07001553 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 goto bad_packet;
1555
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001556 th = tcp_hdr(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001557 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1559 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1560 skb->len - th->doff * 4);
1561 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1562 TCP_SKB_CB(skb)->when = 0;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001563 TCP_SKB_CB(skb)->flags = iph->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 TCP_SKB_CB(skb)->sacked = 0;
1565
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -07001566 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 if (!sk)
1568 goto no_tcp_socket;
1569
1570process:
1571 if (sk->sk_state == TCP_TIME_WAIT)
1572 goto do_time_wait;
1573
1574 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1575 goto discard_and_relse;
Patrick McHardyb59c2702006-01-06 23:06:10 -08001576 nf_reset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07001578 if (sk_filter(sk, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 goto discard_and_relse;
1580
1581 skb->dev = NULL;
1582
Ingo Molnarc6366182006-07-03 00:25:13 -07001583 bh_lock_sock_nested(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 ret = 0;
1585 if (!sock_owned_by_user(sk)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001586#ifdef CONFIG_NET_DMA
1587 struct tcp_sock *tp = tcp_sk(sk);
1588 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
Dan Williamsf67b4592009-01-06 11:38:15 -07001589 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
Chris Leech1a2449a2006-05-23 18:05:53 -07001590 if (tp->ucopy.dma_chan)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 ret = tcp_v4_do_rcv(sk, skb);
Chris Leech1a2449a2006-05-23 18:05:53 -07001592 else
1593#endif
1594 {
1595 if (!tcp_prequeue(sk, skb))
Shan Weiae8d7f82009-05-05 01:01:29 +00001596 ret = tcp_v4_do_rcv(sk, skb);
Chris Leech1a2449a2006-05-23 18:05:53 -07001597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 } else
1599 sk_add_backlog(sk, skb);
1600 bh_unlock_sock(sk);
1601
1602 sock_put(sk);
1603
1604 return ret;
1605
1606no_tcp_socket:
1607 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1608 goto discard_it;
1609
1610 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1611bad_packet:
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001612 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 } else {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001614 tcp_v4_send_reset(NULL, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
1616
1617discard_it:
1618 /* Discard frame. */
1619 kfree_skb(skb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001620 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
1622discard_and_relse:
1623 sock_put(sk);
1624 goto discard_it;
1625
1626do_time_wait:
1627 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001628 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 goto discard_it;
1630 }
1631
1632 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001633 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001634 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 goto discard_it;
1636 }
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001637 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 case TCP_TW_SYN: {
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09001639 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
Pavel Emelyanovc67499c2008-01-31 05:06:40 -08001640 &tcp_hashinfo,
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001641 iph->daddr, th->dest,
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001642 inet_iif(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 if (sk2) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001644 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1645 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 sk = sk2;
1647 goto process;
1648 }
1649 /* Fall through to ACK */
1650 }
1651 case TCP_TW_ACK:
1652 tcp_v4_timewait_ack(sk, skb);
1653 break;
1654 case TCP_TW_RST:
1655 goto no_tcp_socket;
1656 case TCP_TW_SUCCESS:;
1657 }
1658 goto discard_it;
1659}
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661/* VJ's idea. Save last timestamp seen from this destination
1662 * and hold it at least for normal timewait interval to use for duplicate
1663 * segment detection in subsequent connections, before they enter synchronized
1664 * state.
1665 */
1666
1667int tcp_v4_remember_stamp(struct sock *sk)
1668{
1669 struct inet_sock *inet = inet_sk(sk);
1670 struct tcp_sock *tp = tcp_sk(sk);
1671 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1672 struct inet_peer *peer = NULL;
1673 int release_it = 0;
1674
1675 if (!rt || rt->rt_dst != inet->daddr) {
1676 peer = inet_getpeer(inet->daddr, 1);
1677 release_it = 1;
1678 } else {
1679 if (!rt->peer)
1680 rt_bind_peer(rt, 1);
1681 peer = rt->peer;
1682 }
1683
1684 if (peer) {
1685 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
James Morris9d729f72007-03-04 16:12:44 -08001686 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1688 peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1689 peer->tcp_ts = tp->rx_opt.ts_recent;
1690 }
1691 if (release_it)
1692 inet_putpeer(peer);
1693 return 1;
1694 }
1695
1696 return 0;
1697}
1698
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001699int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700{
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001701 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
1703 if (peer) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001704 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1705
1706 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
James Morris9d729f72007-03-04 16:12:44 -08001707 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001708 peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1709 peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1710 peer->tcp_ts = tcptw->tw_ts_recent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 }
1712 inet_putpeer(peer);
1713 return 1;
1714 }
1715
1716 return 0;
1717}
1718
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -08001719struct inet_connection_sock_af_ops ipv4_specific = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001720 .queue_xmit = ip_queue_xmit,
1721 .send_check = tcp_v4_send_check,
1722 .rebuild_header = inet_sk_rebuild_header,
1723 .conn_request = tcp_v4_conn_request,
1724 .syn_recv_sock = tcp_v4_syn_recv_sock,
1725 .remember_stamp = tcp_v4_remember_stamp,
1726 .net_header_len = sizeof(struct iphdr),
1727 .setsockopt = ip_setsockopt,
1728 .getsockopt = ip_getsockopt,
1729 .addr2sockaddr = inet_csk_addr2sockaddr,
1730 .sockaddr_len = sizeof(struct sockaddr_in),
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001731 .bind_conflict = inet_csk_bind_conflict,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001732#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001733 .compat_setsockopt = compat_ip_setsockopt,
1734 .compat_getsockopt = compat_ip_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001735#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736};
1737
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001738#ifdef CONFIG_TCP_MD5SIG
Andrew Mortonb6332e62006-11-30 19:16:28 -08001739static struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001740 .md5_lookup = tcp_v4_md5_lookup,
Adam Langley49a72df2008-07-19 00:01:42 -07001741 .calc_md5_hash = tcp_v4_md5_hash_skb,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001742 .md5_add = tcp_v4_md5_add_func,
1743 .md5_parse = tcp_v4_parse_md5_keys,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001744};
Andrew Mortonb6332e62006-11-30 19:16:28 -08001745#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001746
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747/* NOTE: A lot of things set to zero explicitly by call to
1748 * sk_alloc() so need not be done here.
1749 */
1750static int tcp_v4_init_sock(struct sock *sk)
1751{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001752 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 struct tcp_sock *tp = tcp_sk(sk);
1754
1755 skb_queue_head_init(&tp->out_of_order_queue);
1756 tcp_init_xmit_timers(sk);
1757 tcp_prequeue_init(tp);
1758
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001759 icsk->icsk_rto = TCP_TIMEOUT_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 tp->mdev = TCP_TIMEOUT_INIT;
1761
1762 /* So many TCP implementations out there (incorrectly) count the
1763 * initial SYN frame in their delayed-ACK and congestion control
1764 * algorithms that we must have the following bandaid to talk
1765 * efficiently to them. -DaveM
1766 */
1767 tp->snd_cwnd = 2;
1768
1769 /* See draft-stevens-tcpca-spec-01 for discussion of the
1770 * initialization of these values.
1771 */
1772 tp->snd_ssthresh = 0x7fffffff; /* Infinity */
1773 tp->snd_cwnd_clamp = ~0;
David S. Millerc1b4a7e2005-07-05 15:24:38 -07001774 tp->mss_cache = 536;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
1776 tp->reordering = sysctl_tcp_reordering;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001777 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
1779 sk->sk_state = TCP_CLOSE;
1780
1781 sk->sk_write_space = sk_stream_write_space;
1782 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1783
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -08001784 icsk->icsk_af_ops = &ipv4_specific;
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08001785 icsk->icsk_sync_mss = tcp_sync_mss;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001786#ifdef CONFIG_TCP_MD5SIG
1787 tp->af_specific = &tcp_sock_ipv4_specific;
1788#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1791 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1792
Herbert Xueb4dea52008-12-29 23:04:08 -08001793 local_bh_disable();
Eric Dumazet17483762008-11-25 21:16:35 -08001794 percpu_counter_inc(&tcp_sockets_allocated);
Herbert Xueb4dea52008-12-29 23:04:08 -08001795 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
1797 return 0;
1798}
1799
Brian Haley7d06b2e2008-06-14 17:04:49 -07001800void tcp_v4_destroy_sock(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801{
1802 struct tcp_sock *tp = tcp_sk(sk);
1803
1804 tcp_clear_xmit_timers(sk);
1805
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001806 tcp_cleanup_congestion_control(sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001807
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 /* Cleanup up the write buffer. */
David S. Millerfe067e82007-03-07 12:12:44 -08001809 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810
1811 /* Cleans up our, hopefully empty, out_of_order_queue. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001812 __skb_queue_purge(&tp->out_of_order_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001814#ifdef CONFIG_TCP_MD5SIG
1815 /* Clean up the MD5 key list, if any */
1816 if (tp->md5sig_info) {
1817 tcp_v4_clear_md5_list(sk);
1818 kfree(tp->md5sig_info);
1819 tp->md5sig_info = NULL;
1820 }
1821#endif
1822
Chris Leech1a2449a2006-05-23 18:05:53 -07001823#ifdef CONFIG_NET_DMA
1824 /* Cleans up our sk_async_wait_queue */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001825 __skb_queue_purge(&sk->sk_async_wait_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07001826#endif
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 /* Clean prequeue, it must be empty really */
1829 __skb_queue_purge(&tp->ucopy.prequeue);
1830
1831 /* Clean up a referenced TCP bind bucket. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001832 if (inet_csk(sk)->icsk_bind_hash)
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001833 inet_put_port(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
1835 /*
1836 * If sendmsg cached page exists, toss it.
1837 */
1838 if (sk->sk_sndmsg_page) {
1839 __free_page(sk->sk_sndmsg_page);
1840 sk->sk_sndmsg_page = NULL;
1841 }
1842
Eric Dumazet17483762008-11-25 21:16:35 -08001843 percpu_counter_dec(&tcp_sockets_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844}
1845
1846EXPORT_SYMBOL(tcp_v4_destroy_sock);
1847
1848#ifdef CONFIG_PROC_FS
1849/* Proc filesystem TCP sock list dumping. */
1850
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001851static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001853 return hlist_nulls_empty(head) ? NULL :
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001854 list_entry(head->first, struct inet_timewait_sock, tw_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855}
1856
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001857static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001859 return !is_a_nulls(tw->tw_node.next) ?
1860 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861}
1862
1863static void *listening_get_next(struct seq_file *seq, void *cur)
1864{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001865 struct inet_connection_sock *icsk;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001866 struct hlist_nulls_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 struct sock *sk = cur;
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001868 struct inet_listen_hashbucket *ilb;
Jianjun Kong5799de02008-11-03 02:49:10 -08001869 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07001870 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871
1872 if (!sk) {
1873 st->bucket = 0;
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001874 ilb = &tcp_hashinfo.listening_hash[0];
1875 spin_lock_bh(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001876 sk = sk_nulls_head(&ilb->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 goto get_sk;
1878 }
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001879 ilb = &tcp_hashinfo.listening_hash[st->bucket];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 ++st->num;
1881
1882 if (st->state == TCP_SEQ_STATE_OPENREQ) {
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001883 struct request_sock *req = cur;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
Eric Dumazet72a3eff2006-11-16 02:30:37 -08001885 icsk = inet_csk(st->syn_wait_sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 req = req->dl_next;
1887 while (1) {
1888 while (req) {
Daniel Lezcanobdccc4c2008-07-19 00:15:13 -07001889 if (req->rsk_ops->family == st->family) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 cur = req;
1891 goto out;
1892 }
1893 req = req->dl_next;
1894 }
Eric Dumazet72a3eff2006-11-16 02:30:37 -08001895 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 break;
1897get_req:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001898 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 }
1900 sk = sk_next(st->syn_wait_sk);
1901 st->state = TCP_SEQ_STATE_LISTENING;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001902 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 } else {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001904 icsk = inet_csk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001905 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1906 if (reqsk_queue_len(&icsk->icsk_accept_queue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 goto start_req;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001908 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 sk = sk_next(sk);
1910 }
1911get_sk:
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001912 sk_nulls_for_each_from(sk, node) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001913 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 cur = sk;
1915 goto out;
1916 }
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001917 icsk = inet_csk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001918 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1919 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920start_req:
1921 st->uid = sock_i_uid(sk);
1922 st->syn_wait_sk = sk;
1923 st->state = TCP_SEQ_STATE_OPENREQ;
1924 st->sbucket = 0;
1925 goto get_req;
1926 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001927 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 }
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001929 spin_unlock_bh(&ilb->lock);
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07001930 if (++st->bucket < INET_LHTABLE_SIZE) {
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001931 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1932 spin_lock_bh(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001933 sk = sk_nulls_head(&ilb->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 goto get_sk;
1935 }
1936 cur = NULL;
1937out:
1938 return cur;
1939}
1940
1941static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1942{
1943 void *rc = listening_get_next(seq, NULL);
1944
1945 while (rc && *pos) {
1946 rc = listening_get_next(seq, rc);
1947 --*pos;
1948 }
1949 return rc;
1950}
1951
Andi Kleen6eac5602008-08-28 01:08:02 -07001952static inline int empty_bucket(struct tcp_iter_state *st)
1953{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001954 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
1955 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
Andi Kleen6eac5602008-08-28 01:08:02 -07001956}
1957
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958static void *established_get_first(struct seq_file *seq)
1959{
Jianjun Kong5799de02008-11-03 02:49:10 -08001960 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07001961 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 void *rc = NULL;
1963
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -07001964 for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001966 struct hlist_nulls_node *node;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001967 struct inet_timewait_sock *tw;
Eric Dumazet9db66bd2008-11-20 20:39:09 -08001968 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
Andi Kleen6eac5602008-08-28 01:08:02 -07001970 /* Lockless fast path for the common case of empty buckets */
1971 if (empty_bucket(st))
1972 continue;
1973
Eric Dumazet9db66bd2008-11-20 20:39:09 -08001974 spin_lock_bh(lock);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001975 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
Daniel Lezcanof40c8172008-03-21 04:13:54 -07001976 if (sk->sk_family != st->family ||
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001977 !net_eq(sock_net(sk), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 continue;
1979 }
1980 rc = sk;
1981 goto out;
1982 }
1983 st->state = TCP_SEQ_STATE_TIME_WAIT;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001984 inet_twsk_for_each(tw, node,
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08001985 &tcp_hashinfo.ehash[st->bucket].twchain) {
Pavel Emelyanov28518fc2008-03-21 15:52:00 -07001986 if (tw->tw_family != st->family ||
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001987 !net_eq(twsk_net(tw), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 continue;
1989 }
1990 rc = tw;
1991 goto out;
1992 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -08001993 spin_unlock_bh(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 st->state = TCP_SEQ_STATE_ESTABLISHED;
1995 }
1996out:
1997 return rc;
1998}
1999
2000static void *established_get_next(struct seq_file *seq, void *cur)
2001{
2002 struct sock *sk = cur;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002003 struct inet_timewait_sock *tw;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002004 struct hlist_nulls_node *node;
Jianjun Kong5799de02008-11-03 02:49:10 -08002005 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07002006 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
2008 ++st->num;
2009
2010 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2011 tw = cur;
2012 tw = tw_next(tw);
2013get_tw:
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002014 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 tw = tw_next(tw);
2016 }
2017 if (tw) {
2018 cur = tw;
2019 goto out;
2020 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002021 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 st->state = TCP_SEQ_STATE_ESTABLISHED;
2023
Andi Kleen6eac5602008-08-28 01:08:02 -07002024 /* Look for next non empty bucket */
2025 while (++st->bucket < tcp_hashinfo.ehash_size &&
2026 empty_bucket(st))
2027 ;
2028 if (st->bucket >= tcp_hashinfo.ehash_size)
2029 return NULL;
2030
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002031 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002032 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 } else
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002034 sk = sk_nulls_next(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002036 sk_nulls_for_each_from(sk, node) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002037 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 goto found;
2039 }
2040
2041 st->state = TCP_SEQ_STATE_TIME_WAIT;
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002042 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 goto get_tw;
2044found:
2045 cur = sk;
2046out:
2047 return cur;
2048}
2049
2050static void *established_get_idx(struct seq_file *seq, loff_t pos)
2051{
2052 void *rc = established_get_first(seq);
2053
2054 while (rc && pos) {
2055 rc = established_get_next(seq, rc);
2056 --pos;
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02002057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 return rc;
2059}
2060
2061static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2062{
2063 void *rc;
Jianjun Kong5799de02008-11-03 02:49:10 -08002064 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 st->state = TCP_SEQ_STATE_LISTENING;
2067 rc = listening_get_idx(seq, &pos);
2068
2069 if (!rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 st->state = TCP_SEQ_STATE_ESTABLISHED;
2071 rc = established_get_idx(seq, pos);
2072 }
2073
2074 return rc;
2075}
2076
2077static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2078{
Jianjun Kong5799de02008-11-03 02:49:10 -08002079 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 st->state = TCP_SEQ_STATE_LISTENING;
2081 st->num = 0;
2082 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2083}
2084
2085static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2086{
2087 void *rc = NULL;
Jianjun Kong5799de02008-11-03 02:49:10 -08002088 struct tcp_iter_state *st;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
2090 if (v == SEQ_START_TOKEN) {
2091 rc = tcp_get_idx(seq, 0);
2092 goto out;
2093 }
2094 st = seq->private;
2095
2096 switch (st->state) {
2097 case TCP_SEQ_STATE_OPENREQ:
2098 case TCP_SEQ_STATE_LISTENING:
2099 rc = listening_get_next(seq, v);
2100 if (!rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 st->state = TCP_SEQ_STATE_ESTABLISHED;
2102 rc = established_get_first(seq);
2103 }
2104 break;
2105 case TCP_SEQ_STATE_ESTABLISHED:
2106 case TCP_SEQ_STATE_TIME_WAIT:
2107 rc = established_get_next(seq, v);
2108 break;
2109 }
2110out:
2111 ++*pos;
2112 return rc;
2113}
2114
2115static void tcp_seq_stop(struct seq_file *seq, void *v)
2116{
Jianjun Kong5799de02008-11-03 02:49:10 -08002117 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
2119 switch (st->state) {
2120 case TCP_SEQ_STATE_OPENREQ:
2121 if (v) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002122 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2123 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 }
2125 case TCP_SEQ_STATE_LISTENING:
2126 if (v != SEQ_START_TOKEN)
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002127 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 break;
2129 case TCP_SEQ_STATE_TIME_WAIT:
2130 case TCP_SEQ_STATE_ESTABLISHED:
2131 if (v)
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002132 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 break;
2134 }
2135}
2136
2137static int tcp_seq_open(struct inode *inode, struct file *file)
2138{
2139 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 struct tcp_iter_state *s;
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002141 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002143 err = seq_open_net(inode, file, &afinfo->seq_ops,
2144 sizeof(struct tcp_iter_state));
2145 if (err < 0)
2146 return err;
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002147
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002148 s = ((struct seq_file *)file->private_data)->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 s->family = afinfo->family;
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002150 return 0;
2151}
2152
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002153int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154{
2155 int rc = 0;
2156 struct proc_dir_entry *p;
2157
Denis V. Lunev68fcadd2008-04-13 22:13:30 -07002158 afinfo->seq_fops.open = tcp_seq_open;
2159 afinfo->seq_fops.read = seq_read;
2160 afinfo->seq_fops.llseek = seq_lseek;
2161 afinfo->seq_fops.release = seq_release_net;
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02002162
Denis V. Lunev9427c4b2008-04-13 22:12:13 -07002163 afinfo->seq_ops.start = tcp_seq_start;
2164 afinfo->seq_ops.next = tcp_seq_next;
2165 afinfo->seq_ops.stop = tcp_seq_stop;
2166
Denis V. Lunev84841c32008-05-02 04:10:08 -07002167 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2168 &afinfo->seq_fops, afinfo);
2169 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 rc = -ENOMEM;
2171 return rc;
2172}
2173
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002174void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175{
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002176 proc_net_remove(net, afinfo->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177}
2178
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07002179static void get_openreq4(struct sock *sk, struct request_sock *req,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002180 struct seq_file *f, int i, int uid, int *len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002182 const struct inet_request_sock *ireq = inet_rsk(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 int ttd = req->expires - jiffies;
2184
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002185 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2186 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 i,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002188 ireq->loc_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 ntohs(inet_sk(sk)->sport),
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002190 ireq->rmt_addr,
2191 ntohs(ireq->rmt_port),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 TCP_SYN_RECV,
2193 0, 0, /* could print option size, but that is af dependent. */
2194 1, /* timers active (only the expire timer) */
2195 jiffies_to_clock_t(ttd),
2196 req->retrans,
2197 uid,
2198 0, /* non standard timer */
2199 0, /* open_requests have no inode */
2200 atomic_read(&sk->sk_refcnt),
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002201 req,
2202 len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203}
2204
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002205static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206{
2207 int timer_active;
2208 unsigned long timer_expires;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002209 struct tcp_sock *tp = tcp_sk(sk);
2210 const struct inet_connection_sock *icsk = inet_csk(sk);
2211 struct inet_sock *inet = inet_sk(sk);
Al Viro714e85b2006-11-14 20:51:49 -08002212 __be32 dest = inet->daddr;
2213 __be32 src = inet->rcv_saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 __u16 destp = ntohs(inet->dport);
2215 __u16 srcp = ntohs(inet->sport);
2216
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002217 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 timer_active = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002219 timer_expires = icsk->icsk_timeout;
2220 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 timer_active = 4;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002222 timer_expires = icsk->icsk_timeout;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002223 } else if (timer_pending(&sk->sk_timer)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 timer_active = 2;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002225 timer_expires = sk->sk_timer.expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 } else {
2227 timer_active = 0;
2228 timer_expires = jiffies;
2229 }
2230
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002231 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
Stephen Hemminger7be87352008-06-27 20:00:19 -07002232 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002233 i, src, srcp, dest, destp, sk->sk_state,
Sridhar Samudrala47da8ee2006-06-27 13:29:00 -07002234 tp->write_seq - tp->snd_una,
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002235 sk->sk_state == TCP_LISTEN ? sk->sk_ack_backlog :
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02002236 (tp->rcv_nxt - tp->copied_seq),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 timer_active,
2238 jiffies_to_clock_t(timer_expires - jiffies),
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002239 icsk->icsk_retransmits,
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002240 sock_i_uid(sk),
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002241 icsk->icsk_probes_out,
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002242 sock_i_ino(sk),
2243 atomic_read(&sk->sk_refcnt), sk,
Stephen Hemminger7be87352008-06-27 20:00:19 -07002244 jiffies_to_clock_t(icsk->icsk_rto),
2245 jiffies_to_clock_t(icsk->icsk_ack.ato),
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002246 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 tp->snd_cwnd,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002248 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh,
2249 len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250}
2251
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02002252static void get_timewait4_sock(struct inet_timewait_sock *tw,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002253 struct seq_file *f, int i, int *len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254{
Al Viro23f33c22006-09-27 18:43:50 -07002255 __be32 dest, src;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 __u16 destp, srcp;
2257 int ttd = tw->tw_ttd - jiffies;
2258
2259 if (ttd < 0)
2260 ttd = 0;
2261
2262 dest = tw->tw_daddr;
2263 src = tw->tw_rcv_saddr;
2264 destp = ntohs(tw->tw_dport);
2265 srcp = ntohs(tw->tw_sport);
2266
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002267 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2268 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2270 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002271 atomic_read(&tw->tw_refcnt), tw, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272}
2273
2274#define TMPSZ 150
2275
2276static int tcp4_seq_show(struct seq_file *seq, void *v)
2277{
Jianjun Kong5799de02008-11-03 02:49:10 -08002278 struct tcp_iter_state *st;
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002279 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
2281 if (v == SEQ_START_TOKEN) {
2282 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2283 " sl local_address rem_address st tx_queue "
2284 "rx_queue tr tm->when retrnsmt uid timeout "
2285 "inode");
2286 goto out;
2287 }
2288 st = seq->private;
2289
2290 switch (st->state) {
2291 case TCP_SEQ_STATE_LISTENING:
2292 case TCP_SEQ_STATE_ESTABLISHED:
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002293 get_tcp4_sock(v, seq, st->num, &len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 break;
2295 case TCP_SEQ_STATE_OPENREQ:
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002296 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 break;
2298 case TCP_SEQ_STATE_TIME_WAIT:
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002299 get_timewait4_sock(v, seq, st->num, &len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 break;
2301 }
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002302 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303out:
2304 return 0;
2305}
2306
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307static struct tcp_seq_afinfo tcp4_seq_afinfo = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 .name = "tcp",
2309 .family = AF_INET,
Denis V. Lunev5f4472c2008-04-13 22:13:53 -07002310 .seq_fops = {
2311 .owner = THIS_MODULE,
2312 },
Denis V. Lunev9427c4b2008-04-13 22:12:13 -07002313 .seq_ops = {
2314 .show = tcp4_seq_show,
2315 },
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316};
2317
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002318static int tcp4_proc_init_net(struct net *net)
2319{
2320 return tcp_proc_register(net, &tcp4_seq_afinfo);
2321}
2322
2323static void tcp4_proc_exit_net(struct net *net)
2324{
2325 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2326}
2327
2328static struct pernet_operations tcp4_net_ops = {
2329 .init = tcp4_proc_init_net,
2330 .exit = tcp4_proc_exit_net,
2331};
2332
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333int __init tcp4_proc_init(void)
2334{
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002335 return register_pernet_subsys(&tcp4_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336}
2337
2338void tcp4_proc_exit(void)
2339{
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002340 unregister_pernet_subsys(&tcp4_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341}
2342#endif /* CONFIG_PROC_FS */
2343
Herbert Xubf296b12008-12-15 23:43:36 -08002344struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2345{
Herbert Xu36e7b1b2009-04-27 05:44:45 -07002346 struct iphdr *iph = skb_gro_network_header(skb);
Herbert Xubf296b12008-12-15 23:43:36 -08002347
2348 switch (skb->ip_summed) {
2349 case CHECKSUM_COMPLETE:
Herbert Xu86911732009-01-29 14:19:50 +00002350 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
Herbert Xubf296b12008-12-15 23:43:36 -08002351 skb->csum)) {
2352 skb->ip_summed = CHECKSUM_UNNECESSARY;
2353 break;
2354 }
2355
2356 /* fall through */
2357 case CHECKSUM_NONE:
2358 NAPI_GRO_CB(skb)->flush = 1;
2359 return NULL;
2360 }
2361
2362 return tcp_gro_receive(head, skb);
2363}
2364EXPORT_SYMBOL(tcp4_gro_receive);
2365
2366int tcp4_gro_complete(struct sk_buff *skb)
2367{
2368 struct iphdr *iph = ip_hdr(skb);
2369 struct tcphdr *th = tcp_hdr(skb);
2370
2371 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2372 iph->saddr, iph->daddr, 0);
2373 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2374
2375 return tcp_gro_complete(skb);
2376}
2377EXPORT_SYMBOL(tcp4_gro_complete);
2378
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379struct proto tcp_prot = {
2380 .name = "TCP",
2381 .owner = THIS_MODULE,
2382 .close = tcp_close,
2383 .connect = tcp_v4_connect,
2384 .disconnect = tcp_disconnect,
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002385 .accept = inet_csk_accept,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 .ioctl = tcp_ioctl,
2387 .init = tcp_v4_init_sock,
2388 .destroy = tcp_v4_destroy_sock,
2389 .shutdown = tcp_shutdown,
2390 .setsockopt = tcp_setsockopt,
2391 .getsockopt = tcp_getsockopt,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 .recvmsg = tcp_recvmsg,
2393 .backlog_rcv = tcp_v4_do_rcv,
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08002394 .hash = inet_hash,
2395 .unhash = inet_unhash,
2396 .get_port = inet_csk_get_port,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 .enter_memory_pressure = tcp_enter_memory_pressure,
2398 .sockets_allocated = &tcp_sockets_allocated,
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002399 .orphan_count = &tcp_orphan_count,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 .memory_allocated = &tcp_memory_allocated,
2401 .memory_pressure = &tcp_memory_pressure,
2402 .sysctl_mem = sysctl_tcp_mem,
2403 .sysctl_wmem = sysctl_tcp_wmem,
2404 .sysctl_rmem = sysctl_tcp_rmem,
2405 .max_header = MAX_TCP_HEADER,
2406 .obj_size = sizeof(struct tcp_sock),
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002407 .slab_flags = SLAB_DESTROY_BY_RCU,
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08002408 .twsk_prot = &tcp_timewait_sock_ops,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07002409 .rsk_prot = &tcp_request_sock_ops,
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -07002410 .h.hashinfo = &tcp_hashinfo,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002411#ifdef CONFIG_COMPAT
2412 .compat_setsockopt = compat_tcp_setsockopt,
2413 .compat_getsockopt = compat_tcp_getsockopt,
2414#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415};
2416
Denis V. Lunev046ee902008-04-03 14:31:33 -07002417
2418static int __net_init tcp_sk_init(struct net *net)
2419{
2420 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2421 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2422}
2423
2424static void __net_exit tcp_sk_exit(struct net *net)
2425{
2426 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
Daniel Lezcanod3154922008-09-08 13:17:27 -07002427 inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET);
Denis V. Lunev046ee902008-04-03 14:31:33 -07002428}
2429
2430static struct pernet_operations __net_initdata tcp_sk_ops = {
2431 .init = tcp_sk_init,
2432 .exit = tcp_sk_exit,
2433};
2434
Denis V. Lunev9b0f9762008-02-29 11:13:15 -08002435void __init tcp_v4_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436{
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002437 inet_hashinfo_init(&tcp_hashinfo);
Eric W. Biederman6a1b3052009-02-22 00:10:18 -08002438 if (register_pernet_subsys(&tcp_sk_ops))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 panic("Failed to create the TCP control socket.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440}
2441
2442EXPORT_SYMBOL(ipv4_specific);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443EXPORT_SYMBOL(tcp_hashinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444EXPORT_SYMBOL(tcp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445EXPORT_SYMBOL(tcp_v4_conn_request);
2446EXPORT_SYMBOL(tcp_v4_connect);
2447EXPORT_SYMBOL(tcp_v4_do_rcv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448EXPORT_SYMBOL(tcp_v4_remember_stamp);
2449EXPORT_SYMBOL(tcp_v4_send_check);
2450EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2451
2452#ifdef CONFIG_PROC_FS
2453EXPORT_SYMBOL(tcp_proc_register);
2454EXPORT_SYMBOL(tcp_proc_unregister);
2455#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456EXPORT_SYMBOL(sysctl_tcp_low_latency);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457