blob: 020766292bb01009201ec11e60bb0e214854414c [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090063#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020065#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include <net/icmp.h>
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -070067#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <net/tcp.h>
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -030069#include <net/transp_v6.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <net/ipv6.h>
71#include <net/inet_common.h>
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -080072#include <net/timewait_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#include <net/xfrm.h>
Chris Leech1a2449a2006-05-23 18:05:53 -070074#include <net/netdma.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76#include <linux/inet.h>
77#include <linux/ipv6.h>
78#include <linux/stddef.h>
79#include <linux/proc_fs.h>
80#include <linux/seq_file.h>
81
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080082#include <linux/crypto.h>
83#include <linux/scatterlist.h>
84
Brian Haleyab32ea52006-09-22 14:15:41 -070085int sysctl_tcp_tw_reuse __read_mostly;
86int sysctl_tcp_low_latency __read_mostly;
Eric Dumazet4bc2f182010-07-09 21:22:10 +000087EXPORT_SYMBOL(sysctl_tcp_low_latency);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080090#ifdef CONFIG_TCP_MD5SIG
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -020091static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
92 __be32 addr);
Adam Langley49a72df2008-07-19 00:01:42 -070093static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
94 __be32 daddr, __be32 saddr, struct tcphdr *th);
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +090095#else
96static inline
97struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
98{
99 return NULL;
100}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800101#endif
102
Eric Dumazet5caea4e2008-11-20 00:40:07 -0800103struct inet_hashinfo tcp_hashinfo;
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000104EXPORT_SYMBOL(tcp_hashinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Gerrit Renkera94f7232006-11-10 14:06:49 -0800106static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700108 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
109 ip_hdr(skb)->saddr,
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700110 tcp_hdr(skb)->dest,
111 tcp_hdr(skb)->source);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800114int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
115{
116 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
117 struct tcp_sock *tp = tcp_sk(sk);
118
119 /* With PAWS, it is safe from the viewpoint
120 of data integrity. Even without PAWS it is safe provided sequence
121 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
122
123 Actually, the idea is close to VJ's one, only timestamp cache is
124 held not per host, but per port pair and TW bucket is used as state
125 holder.
126
127 If TW bucket has been already destroyed we fall back to VJ's scheme
128 and use initial timestamp retrieved from peer table.
129 */
130 if (tcptw->tw_ts_recent_stamp &&
131 (twp == NULL || (sysctl_tcp_tw_reuse &&
James Morris9d729f72007-03-04 16:12:44 -0800132 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800133 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
134 if (tp->write_seq == 0)
135 tp->write_seq = 1;
136 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
137 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
138 sock_hold(sktw);
139 return 1;
140 }
141
142 return 0;
143}
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800144EXPORT_SYMBOL_GPL(tcp_twsk_unique);
145
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146/* This will initiate an outgoing connection. */
147int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
148{
149 struct inet_sock *inet = inet_sk(sk);
150 struct tcp_sock *tp = tcp_sk(sk);
151 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
152 struct rtable *rt;
Al Virobada8ad2006-09-26 21:27:15 -0700153 __be32 daddr, nexthop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 int tmp;
155 int err;
156
157 if (addr_len < sizeof(struct sockaddr_in))
158 return -EINVAL;
159
160 if (usin->sin_family != AF_INET)
161 return -EAFNOSUPPORT;
162
163 nexthop = daddr = usin->sin_addr.s_addr;
164 if (inet->opt && inet->opt->srr) {
165 if (!daddr)
166 return -EINVAL;
167 nexthop = inet->opt->faddr;
168 }
169
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000170 tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
172 IPPROTO_TCP,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000173 inet->inet_sport, usin->sin_port, sk, 1);
Wei Dong584bdf82007-05-31 22:49:28 -0700174 if (tmp < 0) {
175 if (tmp == -ENETUNREACH)
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700176 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 return tmp;
Wei Dong584bdf82007-05-31 22:49:28 -0700178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
180 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
181 ip_rt_put(rt);
182 return -ENETUNREACH;
183 }
184
185 if (!inet->opt || !inet->opt->srr)
186 daddr = rt->rt_dst;
187
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000188 if (!inet->inet_saddr)
189 inet->inet_saddr = rt->rt_src;
190 inet->inet_rcv_saddr = inet->inet_saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000192 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 /* Reset inherited state */
194 tp->rx_opt.ts_recent = 0;
195 tp->rx_opt.ts_recent_stamp = 0;
196 tp->write_seq = 0;
197 }
198
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700199 if (tcp_death_row.sysctl_tw_recycle &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
201 struct inet_peer *peer = rt_get_peer(rt);
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200202 /*
203 * VJ's idea. We save last timestamp seen from
204 * the destination in peer table, when entering state
205 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
206 * when trying new connection.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 */
Eric Dumazet317fe0e2010-06-16 04:52:13 +0000208 if (peer) {
209 inet_peer_refcheck(peer);
210 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
211 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
212 tp->rx_opt.ts_recent = peer->tcp_ts;
213 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
215 }
216
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000217 inet->inet_dport = usin->sin_port;
218 inet->inet_daddr = daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800220 inet_csk(sk)->icsk_ext_hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 if (inet->opt)
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800222 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
William Allen Simpsonbee7ca92009-11-10 09:51:18 +0000224 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
226 /* Socket identity is still unknown (sport may be zero).
227 * However we set state to SYN-SENT and not releasing socket
228 * lock select source port, enter ourselves into the hash tables and
229 * complete initialization after this.
230 */
231 tcp_set_state(sk, TCP_SYN_SENT);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800232 err = inet_hash_connect(&tcp_death_row, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (err)
234 goto failure;
235
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200236 err = ip_route_newports(&rt, IPPROTO_TCP,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000237 inet->inet_sport, inet->inet_dport, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 if (err)
239 goto failure;
240
241 /* OK, now commit destination to socket. */
Herbert Xubcd76112006-06-30 13:36:35 -0700242 sk->sk_gso_type = SKB_GSO_TCPV4;
Changli Gaod8d1f302010-06-10 23:31:35 -0700243 sk_setup_caps(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 if (!tp->write_seq)
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000246 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
247 inet->inet_daddr,
248 inet->inet_sport,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 usin->sin_port);
250
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000251 inet->inet_id = tp->write_seq ^ jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
253 err = tcp_connect(sk);
254 rt = NULL;
255 if (err)
256 goto failure;
257
258 return 0;
259
260failure:
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200261 /*
262 * This unhashes the socket and releases the local port,
263 * if necessary.
264 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 tcp_set_state(sk, TCP_CLOSE);
266 ip_rt_put(rt);
267 sk->sk_route_caps = 0;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000268 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return err;
270}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000271EXPORT_SYMBOL(tcp_v4_connect);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273/*
274 * This routine does path mtu discovery as defined in RFC1191.
275 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800276static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
278 struct dst_entry *dst;
279 struct inet_sock *inet = inet_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
282 * send out by Linux are always <576bytes so they should go through
283 * unfragmented).
284 */
285 if (sk->sk_state == TCP_LISTEN)
286 return;
287
288 /* We don't check in the destentry if pmtu discovery is forbidden
289 * on this route. We just assume that no packet_to_big packets
290 * are send back when pmtu discovery is not active.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900291 * There is a small race when the user changes this flag in the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 * route, but I think that's acceptable.
293 */
294 if ((dst = __sk_dst_check(sk, 0)) == NULL)
295 return;
296
297 dst->ops->update_pmtu(dst, mtu);
298
299 /* Something is about to be wrong... Remember soft error
300 * for the case, if this connection will not able to recover.
301 */
302 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
303 sk->sk_err_soft = EMSGSIZE;
304
305 mtu = dst_mtu(dst);
306
307 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800308 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 tcp_sync_mss(sk, mtu);
310
311 /* Resend the TCP packet because it's
312 * clear that the old packet has been
313 * dropped. This is the new "fast" path mtu
314 * discovery.
315 */
316 tcp_simple_retransmit(sk);
317 } /* else let the usual retransmit timer handle it */
318}
319
320/*
321 * This routine is called by the ICMP module when it gets some
322 * sort of error condition. If err < 0 then the socket should
323 * be closed and the error returned to the user. If err > 0
324 * it's just the icmp type << 8 | icmp code. After adjustment
325 * header points to the first 8 bytes of the tcp header. We need
326 * to find the appropriate port.
327 *
328 * The locking strategy used here is very "optimistic". When
329 * someone else accesses the socket the ICMP is just dropped
330 * and for some paths there is no check at all.
331 * A more general error queue to queue errors for later handling
332 * is probably better.
333 *
334 */
335
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000336void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000338 struct iphdr *iph = (struct iphdr *)icmp_skb->data;
339 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000340 struct inet_connection_sock *icsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 struct tcp_sock *tp;
342 struct inet_sock *inet;
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000343 const int type = icmp_hdr(icmp_skb)->type;
344 const int code = icmp_hdr(icmp_skb)->code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 struct sock *sk;
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000346 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 __u32 seq;
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000348 __u32 remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 int err;
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000350 struct net *net = dev_net(icmp_skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000352 if (icmp_skb->len < (iph->ihl << 2) + 8) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700353 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 return;
355 }
356
Pavel Emelyanovfd54d712008-07-14 23:01:40 -0700357 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000358 iph->saddr, th->source, inet_iif(icmp_skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 if (!sk) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700360 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 return;
362 }
363 if (sk->sk_state == TCP_TIME_WAIT) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -0700364 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 return;
366 }
367
368 bh_lock_sock(sk);
369 /* If too many ICMPs get dropped on busy
370 * servers this needs to be solved differently.
371 */
372 if (sock_owned_by_user(sk))
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700373 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375 if (sk->sk_state == TCP_CLOSE)
376 goto out;
377
stephen hemminger97e3ecd12010-03-18 11:27:32 +0000378 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
379 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
380 goto out;
381 }
382
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000383 icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 tp = tcp_sk(sk);
385 seq = ntohl(th->seq);
386 if (sk->sk_state != TCP_LISTEN &&
387 !between(seq, tp->snd_una, tp->snd_nxt)) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700388 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 goto out;
390 }
391
392 switch (type) {
393 case ICMP_SOURCE_QUENCH:
394 /* Just silently ignore these. */
395 goto out;
396 case ICMP_PARAMETERPROB:
397 err = EPROTO;
398 break;
399 case ICMP_DEST_UNREACH:
400 if (code > NR_ICMP_UNREACH)
401 goto out;
402
403 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
404 if (!sock_owned_by_user(sk))
405 do_pmtu_discovery(sk, iph, info);
406 goto out;
407 }
408
409 err = icmp_err_convert[code].errno;
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000410 /* check if icmp_skb allows revert of backoff
411 * (see draft-zimmermann-tcp-lcd) */
412 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
413 break;
414 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
415 !icsk->icsk_backoff)
416 break;
417
418 icsk->icsk_backoff--;
419 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
420 icsk->icsk_backoff;
421 tcp_bound_rto(sk);
422
423 skb = tcp_write_queue_head(sk);
424 BUG_ON(!skb);
425
426 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
427 tcp_time_stamp - TCP_SKB_CB(skb)->when);
428
429 if (remaining) {
430 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
431 remaining, TCP_RTO_MAX);
432 } else if (sock_owned_by_user(sk)) {
433 /* RTO revert clocked out retransmission,
434 * but socket is locked. Will defer. */
435 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
436 HZ/20, TCP_RTO_MAX);
437 } else {
438 /* RTO revert clocked out retransmission.
439 * Will retransmit now */
440 tcp_retransmit_timer(sk);
441 }
442
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 break;
444 case ICMP_TIME_EXCEEDED:
445 err = EHOSTUNREACH;
446 break;
447 default:
448 goto out;
449 }
450
451 switch (sk->sk_state) {
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700452 struct request_sock *req, **prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 case TCP_LISTEN:
454 if (sock_owned_by_user(sk))
455 goto out;
456
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700457 req = inet_csk_search_req(sk, &prev, th->dest,
458 iph->daddr, iph->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (!req)
460 goto out;
461
462 /* ICMPs are not backlogged, hence we cannot get
463 an established socket here.
464 */
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700465 WARN_ON(req->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700467 if (seq != tcp_rsk(req)->snt_isn) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700468 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 goto out;
470 }
471
472 /*
473 * Still in SYN_RECV, just remove it silently.
474 * There is no good way to pass the error to the newly
475 * created socket, and POSIX does not want network
476 * errors returned from accept().
477 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700478 inet_csk_reqsk_queue_drop(sk, req, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 goto out;
480
481 case TCP_SYN_SENT:
482 case TCP_SYN_RECV: /* Cannot happen.
483 It can f.e. if SYNs crossed.
484 */
485 if (!sock_owned_by_user(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 sk->sk_err = err;
487
488 sk->sk_error_report(sk);
489
490 tcp_done(sk);
491 } else {
492 sk->sk_err_soft = err;
493 }
494 goto out;
495 }
496
497 /* If we've already connected we will keep trying
498 * until we time out, or the user gives up.
499 *
500 * rfc1122 4.2.3.9 allows to consider as hard errors
501 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
502 * but it is obsoleted by pmtu discovery).
503 *
504 * Note, that in modern internet, where routing is unreliable
505 * and in each dark corner broken firewalls sit, sending random
506 * errors ordered by their masters even this two messages finally lose
507 * their original sense (even Linux sends invalid PORT_UNREACHs)
508 *
509 * Now we are in compliance with RFCs.
510 * --ANK (980905)
511 */
512
513 inet = inet_sk(sk);
514 if (!sock_owned_by_user(sk) && inet->recverr) {
515 sk->sk_err = err;
516 sk->sk_error_report(sk);
517 } else { /* Only an error on timeout */
518 sk->sk_err_soft = err;
519 }
520
521out:
522 bh_unlock_sock(sk);
523 sock_put(sk);
524}
525
Herbert Xu419f9f82010-04-11 02:15:53 +0000526static void __tcp_v4_send_check(struct sk_buff *skb,
527 __be32 saddr, __be32 daddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700529 struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Patrick McHardy84fa7932006-08-29 16:44:56 -0700531 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Herbert Xu419f9f82010-04-11 02:15:53 +0000532 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
Herbert Xu663ead32007-04-09 11:59:07 -0700533 skb->csum_start = skb_transport_header(skb) - skb->head;
Al Viroff1dcad2006-11-20 18:07:29 -0800534 skb->csum_offset = offsetof(struct tcphdr, check);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 } else {
Herbert Xu419f9f82010-04-11 02:15:53 +0000536 th->check = tcp_v4_check(skb->len, saddr, daddr,
Joe Perches07f07572008-11-19 15:44:53 -0800537 csum_partial(th,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 th->doff << 2,
539 skb->csum));
540 }
541}
542
Herbert Xu419f9f82010-04-11 02:15:53 +0000543/* This routine computes an IPv4 TCP checksum. */
Herbert Xubb296242010-04-11 02:15:55 +0000544void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
Herbert Xu419f9f82010-04-11 02:15:53 +0000545{
546 struct inet_sock *inet = inet_sk(sk);
547
548 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
549}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000550EXPORT_SYMBOL(tcp_v4_send_check);
Herbert Xu419f9f82010-04-11 02:15:53 +0000551
Herbert Xua430a432006-07-08 13:34:56 -0700552int tcp_v4_gso_send_check(struct sk_buff *skb)
553{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700554 const struct iphdr *iph;
Herbert Xua430a432006-07-08 13:34:56 -0700555 struct tcphdr *th;
556
557 if (!pskb_may_pull(skb, sizeof(*th)))
558 return -EINVAL;
559
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700560 iph = ip_hdr(skb);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700561 th = tcp_hdr(skb);
Herbert Xua430a432006-07-08 13:34:56 -0700562
563 th->check = 0;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700564 skb->ip_summed = CHECKSUM_PARTIAL;
Herbert Xu419f9f82010-04-11 02:15:53 +0000565 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
Herbert Xua430a432006-07-08 13:34:56 -0700566 return 0;
567}
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569/*
570 * This routine will send an RST to the other tcp.
571 *
572 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
573 * for reset.
574 * Answer: if a packet caused RST, it is not for a socket
575 * existing in our system, if it is matched to a socket,
576 * it is just duplicate segment or bug in other side's TCP.
577 * So that we build reply only basing on parameters
578 * arrived with segment.
579 * Exception: precedence violation. We do not implement it in any case.
580 */
581
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800582static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700584 struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800585 struct {
586 struct tcphdr th;
587#ifdef CONFIG_TCP_MD5SIG
Al Viro714e85b2006-11-14 20:51:49 -0800588 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800589#endif
590 } rep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 struct ip_reply_arg arg;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800592#ifdef CONFIG_TCP_MD5SIG
593 struct tcp_md5sig_key *key;
594#endif
Pavel Emelyanova86b1e32008-07-16 20:20:58 -0700595 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 /* Never send a reset in response to a reset. */
598 if (th->rst)
599 return;
600
Eric Dumazet511c3f92009-06-02 05:14:27 +0000601 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return;
603
604 /* Swap the send and the receive. */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800605 memset(&rep, 0, sizeof(rep));
606 rep.th.dest = th->source;
607 rep.th.source = th->dest;
608 rep.th.doff = sizeof(struct tcphdr) / 4;
609 rep.th.rst = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
611 if (th->ack) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800612 rep.th.seq = th->ack_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 } else {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800614 rep.th.ack = 1;
615 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
616 skb->len - (th->doff << 2));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200619 memset(&arg, 0, sizeof(arg));
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800620 arg.iov[0].iov_base = (unsigned char *)&rep;
621 arg.iov[0].iov_len = sizeof(rep.th);
622
623#ifdef CONFIG_TCP_MD5SIG
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700624 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800625 if (key) {
626 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
627 (TCPOPT_NOP << 16) |
628 (TCPOPT_MD5SIG << 8) |
629 TCPOLEN_MD5SIG);
630 /* Update length and the length the header thinks exists */
631 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
632 rep.th.doff = arg.iov[0].iov_len / 4;
633
Adam Langley49a72df2008-07-19 00:01:42 -0700634 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
Ilpo Järvinen78e645cb2008-10-09 14:37:47 -0700635 key, ip_hdr(skb)->saddr,
636 ip_hdr(skb)->daddr, &rep.th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800637 }
638#endif
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700639 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
640 ip_hdr(skb)->saddr, /* XXX */
Ilpo Järvinen52cd5752008-10-08 11:34:06 -0700641 arg.iov[0].iov_len, IPPROTO_TCP, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700643 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Eric Dumazetadf30902009-06-02 05:19:30 +0000645 net = dev_net(skb_dst(skb)->dev);
Pavel Emelyanova86b1e32008-07-16 20:20:58 -0700646 ip_send_reply(net->ipv4.tcp_sock, skb,
Denis V. Lunev7feb49c2008-04-03 14:32:00 -0700647 &arg, arg.iov[0].iov_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700649 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
650 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651}
652
653/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
654 outside socket context is ugly, certainly. What can I do?
655 */
656
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900657static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
658 u32 win, u32 ts, int oif,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700659 struct tcp_md5sig_key *key,
660 int reply_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700662 struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 struct {
664 struct tcphdr th;
Al Viro714e85b2006-11-14 20:51:49 -0800665 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800666#ifdef CONFIG_TCP_MD5SIG
Al Viro714e85b2006-11-14 20:51:49 -0800667 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800668#endif
669 ];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 } rep;
671 struct ip_reply_arg arg;
Eric Dumazetadf30902009-06-02 05:19:30 +0000672 struct net *net = dev_net(skb_dst(skb)->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
674 memset(&rep.th, 0, sizeof(struct tcphdr));
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200675 memset(&arg, 0, sizeof(arg));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 arg.iov[0].iov_base = (unsigned char *)&rep;
678 arg.iov[0].iov_len = sizeof(rep.th);
679 if (ts) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800680 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
681 (TCPOPT_TIMESTAMP << 8) |
682 TCPOLEN_TIMESTAMP);
683 rep.opt[1] = htonl(tcp_time_stamp);
684 rep.opt[2] = htonl(ts);
Craig Schlentercb48cfe2007-01-09 00:11:15 -0800685 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 }
687
688 /* Swap the send and the receive. */
689 rep.th.dest = th->source;
690 rep.th.source = th->dest;
691 rep.th.doff = arg.iov[0].iov_len / 4;
692 rep.th.seq = htonl(seq);
693 rep.th.ack_seq = htonl(ack);
694 rep.th.ack = 1;
695 rep.th.window = htons(win);
696
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800697#ifdef CONFIG_TCP_MD5SIG
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800698 if (key) {
699 int offset = (ts) ? 3 : 0;
700
701 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
702 (TCPOPT_NOP << 16) |
703 (TCPOPT_MD5SIG << 8) |
704 TCPOLEN_MD5SIG);
705 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
706 rep.th.doff = arg.iov[0].iov_len/4;
707
Adam Langley49a72df2008-07-19 00:01:42 -0700708 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
Adam Langley90b7e112008-07-31 20:49:48 -0700709 key, ip_hdr(skb)->saddr,
710 ip_hdr(skb)->daddr, &rep.th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800711 }
712#endif
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700713 arg.flags = reply_flags;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700714 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
715 ip_hdr(skb)->saddr, /* XXX */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 arg.iov[0].iov_len, IPPROTO_TCP, 0);
717 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900718 if (oif)
719 arg.bound_dev_if = oif;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
Pavel Emelyanova86b1e32008-07-16 20:20:58 -0700721 ip_send_reply(net->ipv4.tcp_sock, skb,
Denis V. Lunev7feb49c2008-04-03 14:32:00 -0700722 &arg, arg.iov[0].iov_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700724 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
727static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
728{
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700729 struct inet_timewait_sock *tw = inet_twsk(sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800730 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900732 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200733 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900734 tcptw->tw_ts_recent,
735 tw->tw_bound_dev_if,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700736 tcp_twsk_md5_key(tcptw),
737 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900738 );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700740 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741}
742
Gui Jianfeng6edafaa2008-08-06 23:50:04 -0700743static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200744 struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900746 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800747 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900748 req->ts_recent,
749 0,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700750 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
751 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754/*
Kris Katterjohn9bf1d832008-02-17 22:29:19 -0800755 * Send a SYN-ACK after having received a SYN.
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700756 * This still operates on a request_sock only, not on a big
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 * socket.
758 */
Octavian Purdila72659ec2010-01-17 19:09:39 -0800759static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
760 struct request_sock *req,
761 struct request_values *rvp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700763 const struct inet_request_sock *ireq = inet_rsk(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 int err = -1;
765 struct sk_buff * skb;
766
767 /* First, grab a route. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700768 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800769 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
William Allen Simpsone6b4d112009-12-02 18:07:39 +0000771 skb = tcp_make_synack(sk, dst, req, rvp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
773 if (skb) {
Herbert Xu419f9f82010-04-11 02:15:53 +0000774 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700776 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
777 ireq->rmt_addr,
778 ireq->opt);
Gerrit Renkerb9df3cb2006-11-14 11:21:36 -0200779 err = net_xmit_eval(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 }
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 dst_release(dst);
783 return err;
784}
785
Octavian Purdila72659ec2010-01-17 19:09:39 -0800786static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
William Allen Simpsone6b4d112009-12-02 18:07:39 +0000787 struct request_values *rvp)
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800788{
Octavian Purdila72659ec2010-01-17 19:09:39 -0800789 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
790 return tcp_v4_send_synack(sk, NULL, req, rvp);
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800791}
792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793/*
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700794 * IPv4 request_sock destructor.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 */
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700796static void tcp_v4_reqsk_destructor(struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Jesper Juhla51482b2005-11-08 09:41:34 -0800798 kfree(inet_rsk(req)->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799}
800
Florian Westphal2a1d4bd2010-06-03 00:43:12 +0000801static void syn_flood_warning(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Florian Westphal2a1d4bd2010-06-03 00:43:12 +0000803 const char *msg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Florian Westphal2a1d4bd2010-06-03 00:43:12 +0000805#ifdef CONFIG_SYN_COOKIES
806 if (sysctl_tcp_syncookies)
807 msg = "Sending cookies";
808 else
Arnaldo Carvalho de Melo80e40da2006-01-04 01:58:06 -0200809#endif
Florian Westphal2a1d4bd2010-06-03 00:43:12 +0000810 msg = "Dropping request";
811
812 pr_info("TCP: Possible SYN flooding on port %d. %s.\n",
813 ntohs(tcp_hdr(skb)->dest), msg);
814}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816/*
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700817 * Save and compile IPv4 options into the request_sock if needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800819static struct ip_options *tcp_v4_save_options(struct sock *sk,
820 struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
822 struct ip_options *opt = &(IPCB(skb)->opt);
823 struct ip_options *dopt = NULL;
824
825 if (opt && opt->optlen) {
826 int opt_size = optlength(opt);
827 dopt = kmalloc(opt_size, GFP_ATOMIC);
828 if (dopt) {
829 if (ip_options_echo(dopt, skb)) {
830 kfree(dopt);
831 dopt = NULL;
832 }
833 }
834 }
835 return dopt;
836}
837
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800838#ifdef CONFIG_TCP_MD5SIG
839/*
840 * RFC2385 MD5 checksumming requires a mapping of
841 * IP address->MD5 Key.
842 * We need to maintain these in the sk structure.
843 */
844
845/* Find the Key structure for an address. */
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200846static struct tcp_md5sig_key *
847 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800848{
849 struct tcp_sock *tp = tcp_sk(sk);
850 int i;
851
852 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
853 return NULL;
854 for (i = 0; i < tp->md5sig_info->entries4; i++) {
855 if (tp->md5sig_info->keys4[i].addr == addr)
David S. Millerf8ab18d2007-09-28 15:18:35 -0700856 return &tp->md5sig_info->keys4[i].base;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800857 }
858 return NULL;
859}
860
861struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
862 struct sock *addr_sk)
863{
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000864 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800865}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800866EXPORT_SYMBOL(tcp_v4_md5_lookup);
867
Adrian Bunkf5b99bc2006-11-30 17:22:29 -0800868static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
869 struct request_sock *req)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800870{
871 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
872}
873
874/* This can be called on a newly created socket, from other files */
875int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
876 u8 *newkey, u8 newkeylen)
877{
878 /* Add Key to the list */
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700879 struct tcp_md5sig_key *key;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800880 struct tcp_sock *tp = tcp_sk(sk);
881 struct tcp4_md5sig_key *keys;
882
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700883 key = tcp_v4_md5_do_lookup(sk, addr);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800884 if (key) {
885 /* Pre-existing entry - just update that one. */
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700886 kfree(key->key);
887 key->key = newkey;
888 key->keylen = newkeylen;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800889 } else {
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200890 struct tcp_md5sig_info *md5sig;
891
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800892 if (!tp->md5sig_info) {
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200893 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
894 GFP_ATOMIC);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800895 if (!tp->md5sig_info) {
896 kfree(newkey);
897 return -ENOMEM;
898 }
Eric Dumazeta4654192010-05-16 00:36:33 -0700899 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800900 }
Wu Fengguangaa133072009-09-02 23:45:45 -0700901 if (tcp_alloc_md5sig_pool(sk) == NULL) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800902 kfree(newkey);
903 return -ENOMEM;
904 }
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200905 md5sig = tp->md5sig_info;
906
907 if (md5sig->alloced4 == md5sig->entries4) {
908 keys = kmalloc((sizeof(*keys) *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900909 (md5sig->entries4 + 1)), GFP_ATOMIC);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800910 if (!keys) {
911 kfree(newkey);
912 tcp_free_md5sig_pool();
913 return -ENOMEM;
914 }
915
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200916 if (md5sig->entries4)
917 memcpy(keys, md5sig->keys4,
918 sizeof(*keys) * md5sig->entries4);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800919
920 /* Free old key list, and reference new one */
YOSHIFUJI Hideakia80cc202007-11-20 17:30:06 -0800921 kfree(md5sig->keys4);
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200922 md5sig->keys4 = keys;
923 md5sig->alloced4++;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800924 }
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -0200925 md5sig->entries4++;
David S. Millerf8ab18d2007-09-28 15:18:35 -0700926 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
927 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
928 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800929 }
930 return 0;
931}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800932EXPORT_SYMBOL(tcp_v4_md5_do_add);
933
934static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
935 u8 *newkey, u8 newkeylen)
936{
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000937 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800938 newkey, newkeylen);
939}
940
941int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
942{
943 struct tcp_sock *tp = tcp_sk(sk);
944 int i;
945
946 for (i = 0; i < tp->md5sig_info->entries4; i++) {
947 if (tp->md5sig_info->keys4[i].addr == addr) {
948 /* Free the key */
David S. Millerf8ab18d2007-09-28 15:18:35 -0700949 kfree(tp->md5sig_info->keys4[i].base.key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800950 tp->md5sig_info->entries4--;
951
952 if (tp->md5sig_info->entries4 == 0) {
953 kfree(tp->md5sig_info->keys4);
954 tp->md5sig_info->keys4 = NULL;
Leigh Brown8228a18d2006-12-17 17:12:30 -0800955 tp->md5sig_info->alloced4 = 0;
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200956 } else if (tp->md5sig_info->entries4 != i) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800957 /* Need to do some manipulation */
YOSHIFUJI Hideaki354faf02007-11-20 17:30:31 -0800958 memmove(&tp->md5sig_info->keys4[i],
959 &tp->md5sig_info->keys4[i+1],
960 (tp->md5sig_info->entries4 - i) *
961 sizeof(struct tcp4_md5sig_key));
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800962 }
963 tcp_free_md5sig_pool();
964 return 0;
965 }
966 }
967 return -ENOENT;
968}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800969EXPORT_SYMBOL(tcp_v4_md5_do_del);
970
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200971static void tcp_v4_clear_md5_list(struct sock *sk)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800972{
973 struct tcp_sock *tp = tcp_sk(sk);
974
975 /* Free each key, then the set of key keys,
976 * the crypto element, and then decrement our
977 * hold on the last resort crypto.
978 */
979 if (tp->md5sig_info->entries4) {
980 int i;
981 for (i = 0; i < tp->md5sig_info->entries4; i++)
David S. Millerf8ab18d2007-09-28 15:18:35 -0700982 kfree(tp->md5sig_info->keys4[i].base.key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800983 tp->md5sig_info->entries4 = 0;
984 tcp_free_md5sig_pool();
985 }
986 if (tp->md5sig_info->keys4) {
987 kfree(tp->md5sig_info->keys4);
988 tp->md5sig_info->keys4 = NULL;
989 tp->md5sig_info->alloced4 = 0;
990 }
991}
992
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200993static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
994 int optlen)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800995{
996 struct tcp_md5sig cmd;
997 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
998 u8 *newkey;
999
1000 if (optlen < sizeof(cmd))
1001 return -EINVAL;
1002
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02001003 if (copy_from_user(&cmd, optval, sizeof(cmd)))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001004 return -EFAULT;
1005
1006 if (sin->sin_family != AF_INET)
1007 return -EINVAL;
1008
1009 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1010 if (!tcp_sk(sk)->md5sig_info)
1011 return -ENOENT;
1012 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1013 }
1014
1015 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1016 return -EINVAL;
1017
1018 if (!tcp_sk(sk)->md5sig_info) {
1019 struct tcp_sock *tp = tcp_sk(sk);
Wu Fengguangaa133072009-09-02 23:45:45 -07001020 struct tcp_md5sig_info *p;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001021
Wu Fengguangaa133072009-09-02 23:45:45 -07001022 p = kzalloc(sizeof(*p), sk->sk_allocation);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001023 if (!p)
1024 return -EINVAL;
1025
1026 tp->md5sig_info = p;
Eric Dumazeta4654192010-05-16 00:36:33 -07001027 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001028 }
1029
Wu Fengguangaa133072009-09-02 23:45:45 -07001030 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001031 if (!newkey)
1032 return -ENOMEM;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001033 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1034 newkey, cmd.tcpm_keylen);
1035}
1036
Adam Langley49a72df2008-07-19 00:01:42 -07001037static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1038 __be32 daddr, __be32 saddr, int nbytes)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001039{
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001040 struct tcp4_pseudohdr *bp;
Adam Langley49a72df2008-07-19 00:01:42 -07001041 struct scatterlist sg;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001042
1043 bp = &hp->md5_blk.ip4;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001044
1045 /*
Adam Langley49a72df2008-07-19 00:01:42 -07001046 * 1. the TCP pseudo-header (in the order: source IP address,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001047 * destination IP address, zero-padded protocol number, and
1048 * segment length)
1049 */
1050 bp->saddr = saddr;
1051 bp->daddr = daddr;
1052 bp->pad = 0;
YOSHIFUJI Hideaki076fb722008-04-17 12:48:12 +09001053 bp->protocol = IPPROTO_TCP;
Adam Langley49a72df2008-07-19 00:01:42 -07001054 bp->len = cpu_to_be16(nbytes);
David S. Millerc7da57a2007-10-26 00:41:21 -07001055
Adam Langley49a72df2008-07-19 00:01:42 -07001056 sg_init_one(&sg, bp, sizeof(*bp));
1057 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1058}
1059
1060static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1061 __be32 daddr, __be32 saddr, struct tcphdr *th)
1062{
1063 struct tcp_md5sig_pool *hp;
1064 struct hash_desc *desc;
1065
1066 hp = tcp_get_md5sig_pool();
1067 if (!hp)
1068 goto clear_hash_noput;
1069 desc = &hp->md5_desc;
1070
1071 if (crypto_hash_init(desc))
1072 goto clear_hash;
1073 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1074 goto clear_hash;
1075 if (tcp_md5_hash_header(hp, th))
1076 goto clear_hash;
1077 if (tcp_md5_hash_key(hp, key))
1078 goto clear_hash;
1079 if (crypto_hash_final(desc, md5_hash))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001080 goto clear_hash;
1081
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001082 tcp_put_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001083 return 0;
Adam Langley49a72df2008-07-19 00:01:42 -07001084
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001085clear_hash:
1086 tcp_put_md5sig_pool();
1087clear_hash_noput:
1088 memset(md5_hash, 0, 16);
Adam Langley49a72df2008-07-19 00:01:42 -07001089 return 1;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001090}
1091
Adam Langley49a72df2008-07-19 00:01:42 -07001092int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1093 struct sock *sk, struct request_sock *req,
1094 struct sk_buff *skb)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001095{
Adam Langley49a72df2008-07-19 00:01:42 -07001096 struct tcp_md5sig_pool *hp;
1097 struct hash_desc *desc;
1098 struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001099 __be32 saddr, daddr;
1100
1101 if (sk) {
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001102 saddr = inet_sk(sk)->inet_saddr;
1103 daddr = inet_sk(sk)->inet_daddr;
Adam Langley49a72df2008-07-19 00:01:42 -07001104 } else if (req) {
1105 saddr = inet_rsk(req)->loc_addr;
1106 daddr = inet_rsk(req)->rmt_addr;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001107 } else {
Adam Langley49a72df2008-07-19 00:01:42 -07001108 const struct iphdr *iph = ip_hdr(skb);
1109 saddr = iph->saddr;
1110 daddr = iph->daddr;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001111 }
Adam Langley49a72df2008-07-19 00:01:42 -07001112
1113 hp = tcp_get_md5sig_pool();
1114 if (!hp)
1115 goto clear_hash_noput;
1116 desc = &hp->md5_desc;
1117
1118 if (crypto_hash_init(desc))
1119 goto clear_hash;
1120
1121 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1122 goto clear_hash;
1123 if (tcp_md5_hash_header(hp, th))
1124 goto clear_hash;
1125 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1126 goto clear_hash;
1127 if (tcp_md5_hash_key(hp, key))
1128 goto clear_hash;
1129 if (crypto_hash_final(desc, md5_hash))
1130 goto clear_hash;
1131
1132 tcp_put_md5sig_pool();
1133 return 0;
1134
1135clear_hash:
1136 tcp_put_md5sig_pool();
1137clear_hash_noput:
1138 memset(md5_hash, 0, 16);
1139 return 1;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001140}
Adam Langley49a72df2008-07-19 00:01:42 -07001141EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001142
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02001143static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001144{
1145 /*
1146 * This gets called for each TCP segment that arrives
1147 * so we want to be efficient.
1148 * We have 3 drop cases:
1149 * o No MD5 hash and one expected.
1150 * o MD5 hash and we're not expecting one.
1151 * o MD5 hash and its wrong.
1152 */
1153 __u8 *hash_location = NULL;
1154 struct tcp_md5sig_key *hash_expected;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001155 const struct iphdr *iph = ip_hdr(skb);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001156 struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001157 int genhash;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001158 unsigned char newhash[16];
1159
1160 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +09001161 hash_location = tcp_parse_md5sig_option(th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001162
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001163 /* We've parsed the options - do we have a hash? */
1164 if (!hash_expected && !hash_location)
1165 return 0;
1166
1167 if (hash_expected && !hash_location) {
David S. Miller785957d2008-07-30 03:03:15 -07001168 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001169 return 1;
1170 }
1171
1172 if (!hash_expected && hash_location) {
David S. Miller785957d2008-07-30 03:03:15 -07001173 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001174 return 1;
1175 }
1176
1177 /* Okay, so this is hash_expected and hash_location -
1178 * so we need to calculate the checksum.
1179 */
Adam Langley49a72df2008-07-19 00:01:42 -07001180 genhash = tcp_v4_md5_hash_skb(newhash,
1181 hash_expected,
1182 NULL, NULL, skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001183
1184 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1185 if (net_ratelimit()) {
Harvey Harrison673d57e2008-10-31 00:53:57 -07001186 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1187 &iph->saddr, ntohs(th->source),
1188 &iph->daddr, ntohs(th->dest),
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001189 genhash ? " tcp_v4_calc_md5_hash failed" : "");
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001190 }
1191 return 1;
1192 }
1193 return 0;
1194}
1195
1196#endif
1197
Eric Dumazet72a3eff2006-11-16 02:30:37 -08001198struct request_sock_ops tcp_request_sock_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 .family = PF_INET,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001200 .obj_size = sizeof(struct tcp_request_sock),
Octavian Purdila72659ec2010-01-17 19:09:39 -08001201 .rtx_syn_ack = tcp_v4_rtx_synack,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001202 .send_ack = tcp_v4_reqsk_send_ack,
1203 .destructor = tcp_v4_reqsk_destructor,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 .send_reset = tcp_v4_send_reset,
Octavian Purdila72659ec2010-01-17 19:09:39 -08001205 .syn_ack_timeout = tcp_syn_ack_timeout,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206};
1207
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001208#ifdef CONFIG_TCP_MD5SIG
Stephen Hemmingerb2e4b3d2009-09-01 19:25:03 +00001209static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001210 .md5_lookup = tcp_v4_reqsk_md5_lookup,
John Dykstrae3afe7b2009-07-16 05:04:51 +00001211 .calc_md5_hash = tcp_v4_md5_hash_skb,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001212};
Andrew Mortonb6332e62006-11-30 19:16:28 -08001213#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001214
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08001215static struct timewait_sock_ops tcp_timewait_sock_ops = {
1216 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1217 .twsk_unique = tcp_twsk_unique,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001218 .twsk_destructor= tcp_twsk_destructor,
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08001219};
1220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1222{
William Allen Simpson4957faade2009-12-02 18:25:27 +00001223 struct tcp_extend_values tmp_ext;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 struct tcp_options_received tmp_opt;
William Allen Simpson4957faade2009-12-02 18:25:27 +00001225 u8 *hash_location;
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001226 struct request_sock *req;
William Allen Simpsone6b4d112009-12-02 18:07:39 +00001227 struct inet_request_sock *ireq;
William Allen Simpson4957faade2009-12-02 18:25:27 +00001228 struct tcp_sock *tp = tcp_sk(sk);
William Allen Simpsone6b4d112009-12-02 18:07:39 +00001229 struct dst_entry *dst = NULL;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001230 __be32 saddr = ip_hdr(skb)->saddr;
1231 __be32 daddr = ip_hdr(skb)->daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 __u32 isn = TCP_SKB_CB(skb)->when;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233#ifdef CONFIG_SYN_COOKIES
1234 int want_cookie = 0;
1235#else
1236#define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1237#endif
1238
1239 /* Never answer to SYNs send to broadcast or multicast */
Eric Dumazet511c3f92009-06-02 05:14:27 +00001240 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 goto drop;
1242
1243 /* TW buckets are converted to open requests without
1244 * limitations, they conserve resources and peer is
1245 * evidently real one.
1246 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001247 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
Florian Westphal2a1d4bd2010-06-03 00:43:12 +00001248 if (net_ratelimit())
1249 syn_flood_warning(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250#ifdef CONFIG_SYN_COOKIES
1251 if (sysctl_tcp_syncookies) {
1252 want_cookie = 1;
1253 } else
1254#endif
1255 goto drop;
1256 }
1257
1258 /* Accept backlog is full. If we have already queued enough
1259 * of warm entries in syn queue, drop request. It is better than
1260 * clogging syn queue with openreqs with exponentially increasing
1261 * timeout.
1262 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001263 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 goto drop;
1265
Arnaldo Carvalho de Meloce4a7d02008-06-10 12:39:35 -07001266 req = inet_reqsk_alloc(&tcp_request_sock_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (!req)
1268 goto drop;
1269
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001270#ifdef CONFIG_TCP_MD5SIG
1271 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1272#endif
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 tcp_clear_options(&tmp_opt);
William Allen Simpsonbee7ca92009-11-10 09:51:18 +00001275 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
William Allen Simpson4957faade2009-12-02 18:25:27 +00001276 tmp_opt.user_mss = tp->rx_opt.user_mss;
David S. Millerbb5b7c12009-12-15 20:56:42 -08001277 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
William Allen Simpson4957faade2009-12-02 18:25:27 +00001279 if (tmp_opt.cookie_plus > 0 &&
1280 tmp_opt.saw_tstamp &&
1281 !tp->rx_opt.cookie_out_never &&
1282 (sysctl_tcp_cookie_size > 0 ||
1283 (tp->cookie_values != NULL &&
1284 tp->cookie_values->cookie_desired > 0))) {
1285 u8 *c;
1286 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1287 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1288
1289 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1290 goto drop_and_release;
1291
1292 /* Secret recipe starts with IP addresses */
Eric Dumazet0eae88f2010-04-20 19:06:52 -07001293 *mess++ ^= (__force u32)daddr;
1294 *mess++ ^= (__force u32)saddr;
William Allen Simpson4957faade2009-12-02 18:25:27 +00001295
1296 /* plus variable length Initiator Cookie */
1297 c = (u8 *)mess;
1298 while (l-- > 0)
1299 *c++ ^= *hash_location++;
1300
1301#ifdef CONFIG_SYN_COOKIES
1302 want_cookie = 0; /* not our kind of cookie */
1303#endif
1304 tmp_ext.cookie_out_never = 0; /* false */
1305 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1306 } else if (!tp->rx_opt.cookie_in_always) {
1307 /* redundant indications, but ensure initialization. */
1308 tmp_ext.cookie_out_never = 1; /* true */
1309 tmp_ext.cookie_plus = 0;
1310 } else {
1311 goto drop_and_release;
1312 }
1313 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Florian Westphal4dfc2812008-04-10 03:12:40 -07001315 if (want_cookie && !tmp_opt.saw_tstamp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 tcp_clear_options(&tmp_opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 tcp_openreq_init(req, &tmp_opt, skb);
1320
David S. Millerbb5b7c12009-12-15 20:56:42 -08001321 ireq = inet_rsk(req);
1322 ireq->loc_addr = daddr;
1323 ireq->rmt_addr = saddr;
1324 ireq->no_srccheck = inet_sk(sk)->transparent;
1325 ireq->opt = tcp_v4_save_options(sk, skb);
1326
Paul Moore284904a2009-03-27 17:10:28 -04001327 if (security_inet_conn_request(sk, skb, req))
David S. Millerbb5b7c12009-12-15 20:56:42 -08001328 goto drop_and_free;
Paul Moore284904a2009-03-27 17:10:28 -04001329
Florian Westphal172d69e2010-06-21 11:48:45 +00001330 if (!want_cookie || tmp_opt.tstamp_ok)
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001331 TCP_ECN_create_request(req, tcp_hdr(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 if (want_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
Florian Westphal172d69e2010-06-21 11:48:45 +00001335 req->cookie_ts = tmp_opt.tstamp_ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 } else if (!isn) {
1337 struct inet_peer *peer = NULL;
1338
1339 /* VJ's idea. We save last timestamp seen
1340 * from the destination in peer table, when entering
1341 * state TIME-WAIT, and check against it before
1342 * accepting new connection request.
1343 *
1344 * If "isn" is not zero, this request hit alive
1345 * timewait bucket, so that all the necessary checks
1346 * are made in the function processing timewait state.
1347 */
1348 if (tmp_opt.saw_tstamp &&
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -07001349 tcp_death_row.sysctl_tw_recycle &&
David S. Millerbb5b7c12009-12-15 20:56:42 -08001350 (dst = inet_csk_route_req(sk, req)) != NULL &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1352 peer->v4daddr == saddr) {
Eric Dumazet317fe0e2010-06-16 04:52:13 +00001353 inet_peer_refcheck(peer);
Eric Dumazet2c1409a2009-11-12 09:33:09 +00001354 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 (s32)(peer->tcp_ts - req->ts_recent) >
1356 TCP_PAWS_WINDOW) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001357 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001358 goto drop_and_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 }
1360 }
1361 /* Kill the following clause, if you dislike this way. */
1362 else if (!sysctl_tcp_syncookies &&
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001363 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 (sysctl_max_syn_backlog >> 2)) &&
1365 (!peer || !peer->tcp_ts_stamp) &&
1366 (!dst || !dst_metric(dst, RTAX_RTT))) {
1367 /* Without syncookies last quarter of
1368 * backlog is filled with destinations,
1369 * proven to be alive.
1370 * It means that we continue to communicate
1371 * to destinations, already remembered
1372 * to the moment of synflood.
1373 */
Harvey Harrison673d57e2008-10-31 00:53:57 -07001374 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1375 &saddr, ntohs(tcp_hdr(skb)->source));
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001376 goto drop_and_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
1378
Gerrit Renkera94f7232006-11-10 14:06:49 -08001379 isn = tcp_v4_init_sequence(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001381 tcp_rsk(req)->snt_isn = isn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Octavian Purdila72659ec2010-01-17 19:09:39 -08001383 if (tcp_v4_send_synack(sk, dst, req,
1384 (struct request_values *)&tmp_ext) ||
William Allen Simpson4957faade2009-12-02 18:25:27 +00001385 want_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 goto drop_and_free;
1387
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001388 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 return 0;
1390
Denis V. Lunev7cd04fa2008-03-03 11:59:32 -08001391drop_and_release:
1392 dst_release(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393drop_and_free:
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001394 reqsk_free(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 return 0;
1397}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001398EXPORT_SYMBOL(tcp_v4_conn_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
1400
1401/*
1402 * The three way handshake has completed - we got a valid synack -
1403 * now create the new socket.
1404 */
1405struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001406 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 struct dst_entry *dst)
1408{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001409 struct inet_request_sock *ireq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 struct inet_sock *newinet;
1411 struct tcp_sock *newtp;
1412 struct sock *newsk;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001413#ifdef CONFIG_TCP_MD5SIG
1414 struct tcp_md5sig_key *key;
1415#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
1417 if (sk_acceptq_is_full(sk))
1418 goto exit_overflow;
1419
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001420 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 goto exit;
1422
1423 newsk = tcp_create_openreq_child(sk, req, skb);
1424 if (!newsk)
1425 goto exit;
1426
Herbert Xubcd76112006-06-30 13:36:35 -07001427 newsk->sk_gso_type = SKB_GSO_TCPV4;
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -07001428 sk_setup_caps(newsk, dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 newtp = tcp_sk(newsk);
1431 newinet = inet_sk(newsk);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001432 ireq = inet_rsk(req);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001433 newinet->inet_daddr = ireq->rmt_addr;
1434 newinet->inet_rcv_saddr = ireq->loc_addr;
1435 newinet->inet_saddr = ireq->loc_addr;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001436 newinet->opt = ireq->opt;
1437 ireq->opt = NULL;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001438 newinet->mc_index = inet_iif(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001439 newinet->mc_ttl = ip_hdr(skb)->ttl;
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08001440 inet_csk(newsk)->icsk_ext_hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 if (newinet->opt)
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08001442 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001443 newinet->inet_id = newtp->write_seq ^ jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
John Heffner5d424d52006-03-20 17:53:41 -08001445 tcp_mtup_init(newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 tcp_sync_mss(newsk, dst_mtu(dst));
1447 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
Tom Quetchenbachf5fff5d2008-09-21 00:21:51 -07001448 if (tcp_sk(sk)->rx_opt.user_mss &&
1449 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1450 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 tcp_initialize_rcv_mss(newsk);
1453
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001454#ifdef CONFIG_TCP_MD5SIG
1455 /* Copy over the MD5 key from the original socket */
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001456 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1457 if (key != NULL) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001458 /*
1459 * We're using one, so create a matching key
1460 * on the newsk structure. If we fail to get
1461 * memory, then we end up not copying the key
1462 * across. Shucks.
1463 */
Arnaldo Carvalho de Melof6685932006-11-17 11:06:01 -02001464 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1465 if (newkey != NULL)
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001466 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001467 newkey, key->keylen);
Eric Dumazeta4654192010-05-16 00:36:33 -07001468 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001469 }
1470#endif
1471
Eric Dumazet9327f702009-12-04 03:46:54 +00001472 __inet_hash_nolisten(newsk, NULL);
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001473 __inet_inherit_port(sk, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
1475 return newsk;
1476
1477exit_overflow:
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001478 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479exit:
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001480 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 dst_release(dst);
1482 return NULL;
1483}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001484EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
1486static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1487{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001488 struct tcphdr *th = tcp_hdr(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001489 const struct iphdr *iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 struct sock *nsk;
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001491 struct request_sock **prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 /* Find possible connection requests. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001493 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1494 iph->saddr, iph->daddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (req)
1496 return tcp_check_req(sk, skb, req, prev);
1497
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001498 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
Pavel Emelyanovc67499c2008-01-31 05:06:40 -08001499 th->source, iph->daddr, th->dest, inet_iif(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
1501 if (nsk) {
1502 if (nsk->sk_state != TCP_TIME_WAIT) {
1503 bh_lock_sock(nsk);
1504 return nsk;
1505 }
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001506 inet_twsk_put(inet_twsk(nsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 return NULL;
1508 }
1509
1510#ifdef CONFIG_SYN_COOKIES
Florian Westphalaf9b4732010-06-03 00:43:44 +00001511 if (!th->syn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1513#endif
1514 return sk;
1515}
1516
Al Virob51655b2006-11-14 21:40:42 -08001517static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001519 const struct iphdr *iph = ip_hdr(skb);
1520
Patrick McHardy84fa7932006-08-29 16:44:56 -07001521 if (skb->ip_summed == CHECKSUM_COMPLETE) {
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001522 if (!tcp_v4_check(skb->len, iph->saddr,
1523 iph->daddr, skb->csum)) {
Herbert Xufb286bb2005-11-10 13:01:24 -08001524 skb->ip_summed = CHECKSUM_UNNECESSARY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 return 0;
Herbert Xufb286bb2005-11-10 13:01:24 -08001526 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
Herbert Xufb286bb2005-11-10 13:01:24 -08001528
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001529 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
Herbert Xufb286bb2005-11-10 13:01:24 -08001530 skb->len, IPPROTO_TCP, 0);
1531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 if (skb->len <= 76) {
Herbert Xufb286bb2005-11-10 13:01:24 -08001533 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 }
1535 return 0;
1536}
1537
1538
1539/* The socket must have it's spinlock held when we get
1540 * here.
1541 *
1542 * We have a potential double-lock case here, so even when
1543 * doing backlog processing we use the BH locking scheme.
1544 * This is because we cannot sleep with the original spinlock
1545 * held.
1546 */
1547int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1548{
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001549 struct sock *rsk;
1550#ifdef CONFIG_TCP_MD5SIG
1551 /*
1552 * We really want to reject the packet as early as possible
1553 * if:
1554 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1555 * o There is an MD5 option and we're not expecting one
1556 */
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02001557 if (tcp_v4_inbound_md5_hash(sk, skb))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001558 goto discard;
1559#endif
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
Eric Dumazetca551582010-06-03 09:03:58 +00001562 sock_rps_save_rxhash(sk, skb->rxhash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 TCP_CHECK_TIMER(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001564 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001565 rsk = sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 TCP_CHECK_TIMER(sk);
1569 return 0;
1570 }
1571
Arnaldo Carvalho de Meloab6a5bb2007-03-18 17:43:48 -07001572 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 goto csum_err;
1574
1575 if (sk->sk_state == TCP_LISTEN) {
1576 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1577 if (!nsk)
1578 goto discard;
1579
1580 if (nsk != sk) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001581 if (tcp_child_process(sk, nsk, skb)) {
1582 rsk = nsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 return 0;
1586 }
Eric Dumazetca551582010-06-03 09:03:58 +00001587 } else
1588 sock_rps_save_rxhash(sk, skb->rxhash);
1589
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 TCP_CHECK_TIMER(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001592 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001593 rsk = sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 TCP_CHECK_TIMER(sk);
1597 return 0;
1598
1599reset:
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001600 tcp_v4_send_reset(rsk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601discard:
1602 kfree_skb(skb);
1603 /* Be careful here. If this function gets more complicated and
1604 * gcc suffers from register pressure on the x86, sk (in %ebx)
1605 * might be destroyed here. This current version compiles correctly,
1606 * but you have been warned.
1607 */
1608 return 0;
1609
1610csum_err:
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001611 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 goto discard;
1613}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001614EXPORT_SYMBOL(tcp_v4_do_rcv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615
1616/*
1617 * From tcp_input.c
1618 */
1619
1620int tcp_v4_rcv(struct sk_buff *skb)
1621{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001622 const struct iphdr *iph;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 struct tcphdr *th;
1624 struct sock *sk;
1625 int ret;
Pavel Emelyanova86b1e32008-07-16 20:20:58 -07001626 struct net *net = dev_net(skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628 if (skb->pkt_type != PACKET_HOST)
1629 goto discard_it;
1630
1631 /* Count it even if it's bad */
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001632 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
1634 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1635 goto discard_it;
1636
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001637 th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
1639 if (th->doff < sizeof(struct tcphdr) / 4)
1640 goto bad_packet;
1641 if (!pskb_may_pull(skb, th->doff * 4))
1642 goto discard_it;
1643
1644 /* An explanation is required here, I think.
1645 * Packet length and doff are validated by header prediction,
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08001646 * provided case of th->doff==0 is eliminated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 * So, we defer the checks. */
Herbert Xu60476372007-04-09 11:59:39 -07001648 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 goto bad_packet;
1650
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001651 th = tcp_hdr(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001652 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1654 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1655 skb->len - th->doff * 4);
1656 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1657 TCP_SKB_CB(skb)->when = 0;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001658 TCP_SKB_CB(skb)->flags = iph->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 TCP_SKB_CB(skb)->sacked = 0;
1660
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -07001661 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 if (!sk)
1663 goto no_tcp_socket;
1664
Eric Dumazetbb134d52010-03-09 05:55:56 +00001665process:
1666 if (sk->sk_state == TCP_TIME_WAIT)
1667 goto do_time_wait;
1668
Eric Dumazet6cce09f2010-03-07 23:21:57 +00001669 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1670 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
Stephen Hemmingerd218d112010-01-11 16:28:01 -08001671 goto discard_and_relse;
Eric Dumazet6cce09f2010-03-07 23:21:57 +00001672 }
Stephen Hemmingerd218d112010-01-11 16:28:01 -08001673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1675 goto discard_and_relse;
Patrick McHardyb59c2702006-01-06 23:06:10 -08001676 nf_reset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07001678 if (sk_filter(sk, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 goto discard_and_relse;
1680
1681 skb->dev = NULL;
1682
Ingo Molnarc6366182006-07-03 00:25:13 -07001683 bh_lock_sock_nested(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 ret = 0;
1685 if (!sock_owned_by_user(sk)) {
Chris Leech1a2449a2006-05-23 18:05:53 -07001686#ifdef CONFIG_NET_DMA
1687 struct tcp_sock *tp = tcp_sk(sk);
1688 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
Dan Williamsf67b4592009-01-06 11:38:15 -07001689 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
Chris Leech1a2449a2006-05-23 18:05:53 -07001690 if (tp->ucopy.dma_chan)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 ret = tcp_v4_do_rcv(sk, skb);
Chris Leech1a2449a2006-05-23 18:05:53 -07001692 else
1693#endif
1694 {
1695 if (!tcp_prequeue(sk, skb))
Shan Weiae8d7f82009-05-05 01:01:29 +00001696 ret = tcp_v4_do_rcv(sk, skb);
Chris Leech1a2449a2006-05-23 18:05:53 -07001697 }
Eric Dumazet6cce09f2010-03-07 23:21:57 +00001698 } else if (unlikely(sk_add_backlog(sk, skb))) {
Zhu Yi6b03a532010-03-04 18:01:41 +00001699 bh_unlock_sock(sk);
Eric Dumazet6cce09f2010-03-07 23:21:57 +00001700 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
Zhu Yi6b03a532010-03-04 18:01:41 +00001701 goto discard_and_relse;
1702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 bh_unlock_sock(sk);
1704
1705 sock_put(sk);
1706
1707 return ret;
1708
1709no_tcp_socket:
1710 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1711 goto discard_it;
1712
1713 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1714bad_packet:
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001715 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 } else {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001717 tcp_v4_send_reset(NULL, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719
1720discard_it:
1721 /* Discard frame. */
1722 kfree_skb(skb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001723 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725discard_and_relse:
1726 sock_put(sk);
1727 goto discard_it;
1728
1729do_time_wait:
1730 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001731 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 goto discard_it;
1733 }
1734
1735 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001736 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001737 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 goto discard_it;
1739 }
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001740 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 case TCP_TW_SYN: {
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09001742 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
Pavel Emelyanovc67499c2008-01-31 05:06:40 -08001743 &tcp_hashinfo,
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001744 iph->daddr, th->dest,
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001745 inet_iif(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if (sk2) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001747 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1748 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 sk = sk2;
1750 goto process;
1751 }
1752 /* Fall through to ACK */
1753 }
1754 case TCP_TW_ACK:
1755 tcp_v4_timewait_ack(sk, skb);
1756 break;
1757 case TCP_TW_RST:
1758 goto no_tcp_socket;
1759 case TCP_TW_SUCCESS:;
1760 }
1761 goto discard_it;
1762}
1763
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764/* VJ's idea. Save last timestamp seen from this destination
1765 * and hold it at least for normal timewait interval to use for duplicate
1766 * segment detection in subsequent connections, before they enter synchronized
1767 * state.
1768 */
1769
1770int tcp_v4_remember_stamp(struct sock *sk)
1771{
1772 struct inet_sock *inet = inet_sk(sk);
1773 struct tcp_sock *tp = tcp_sk(sk);
1774 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1775 struct inet_peer *peer = NULL;
1776 int release_it = 0;
1777
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001778 if (!rt || rt->rt_dst != inet->inet_daddr) {
1779 peer = inet_getpeer(inet->inet_daddr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 release_it = 1;
1781 } else {
1782 if (!rt->peer)
1783 rt_bind_peer(rt, 1);
1784 peer = rt->peer;
1785 }
1786
1787 if (peer) {
1788 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
Eric Dumazet2c1409a2009-11-12 09:33:09 +00001789 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1790 peer->tcp_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
1791 peer->tcp_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 peer->tcp_ts = tp->rx_opt.ts_recent;
1793 }
1794 if (release_it)
1795 inet_putpeer(peer);
1796 return 1;
1797 }
1798
1799 return 0;
1800}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001801EXPORT_SYMBOL(tcp_v4_remember_stamp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001803int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804{
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001805 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807 if (peer) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001808 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1809
1810 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
Eric Dumazet2c1409a2009-11-12 09:33:09 +00001811 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1812 peer->tcp_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
1813 peer->tcp_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001814 peer->tcp_ts = tcptw->tw_ts_recent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 }
1816 inet_putpeer(peer);
1817 return 1;
1818 }
1819
1820 return 0;
1821}
1822
Stephen Hemminger3b401a82009-09-01 19:25:04 +00001823const struct inet_connection_sock_af_ops ipv4_specific = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001824 .queue_xmit = ip_queue_xmit,
1825 .send_check = tcp_v4_send_check,
1826 .rebuild_header = inet_sk_rebuild_header,
1827 .conn_request = tcp_v4_conn_request,
1828 .syn_recv_sock = tcp_v4_syn_recv_sock,
1829 .remember_stamp = tcp_v4_remember_stamp,
1830 .net_header_len = sizeof(struct iphdr),
1831 .setsockopt = ip_setsockopt,
1832 .getsockopt = ip_getsockopt,
1833 .addr2sockaddr = inet_csk_addr2sockaddr,
1834 .sockaddr_len = sizeof(struct sockaddr_in),
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001835 .bind_conflict = inet_csk_bind_conflict,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001836#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001837 .compat_setsockopt = compat_ip_setsockopt,
1838 .compat_getsockopt = compat_ip_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001839#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840};
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001841EXPORT_SYMBOL(ipv4_specific);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001843#ifdef CONFIG_TCP_MD5SIG
Stephen Hemmingerb2e4b3d2009-09-01 19:25:03 +00001844static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001845 .md5_lookup = tcp_v4_md5_lookup,
Adam Langley49a72df2008-07-19 00:01:42 -07001846 .calc_md5_hash = tcp_v4_md5_hash_skb,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001847 .md5_add = tcp_v4_md5_add_func,
1848 .md5_parse = tcp_v4_parse_md5_keys,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001849};
Andrew Mortonb6332e62006-11-30 19:16:28 -08001850#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852/* NOTE: A lot of things set to zero explicitly by call to
1853 * sk_alloc() so need not be done here.
1854 */
1855static int tcp_v4_init_sock(struct sock *sk)
1856{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001857 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 struct tcp_sock *tp = tcp_sk(sk);
1859
1860 skb_queue_head_init(&tp->out_of_order_queue);
1861 tcp_init_xmit_timers(sk);
1862 tcp_prequeue_init(tp);
1863
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001864 icsk->icsk_rto = TCP_TIMEOUT_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 tp->mdev = TCP_TIMEOUT_INIT;
1866
1867 /* So many TCP implementations out there (incorrectly) count the
1868 * initial SYN frame in their delayed-ACK and congestion control
1869 * algorithms that we must have the following bandaid to talk
1870 * efficiently to them. -DaveM
1871 */
1872 tp->snd_cwnd = 2;
1873
1874 /* See draft-stevens-tcpca-spec-01 for discussion of the
1875 * initialization of these values.
1876 */
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001877 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 tp->snd_cwnd_clamp = ~0;
William Allen Simpsonbee7ca92009-11-10 09:51:18 +00001879 tp->mss_cache = TCP_MSS_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
1881 tp->reordering = sysctl_tcp_reordering;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001882 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
1884 sk->sk_state = TCP_CLOSE;
1885
1886 sk->sk_write_space = sk_stream_write_space;
1887 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1888
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -08001889 icsk->icsk_af_ops = &ipv4_specific;
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08001890 icsk->icsk_sync_mss = tcp_sync_mss;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001891#ifdef CONFIG_TCP_MD5SIG
1892 tp->af_specific = &tcp_sock_ipv4_specific;
1893#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
William Allen Simpson435cf552009-12-02 18:17:05 +00001895 /* TCP Cookie Transactions */
1896 if (sysctl_tcp_cookie_size > 0) {
1897 /* Default, cookies without s_data_payload. */
1898 tp->cookie_values =
1899 kzalloc(sizeof(*tp->cookie_values),
1900 sk->sk_allocation);
1901 if (tp->cookie_values != NULL)
1902 kref_init(&tp->cookie_values->kref);
1903 }
1904 /* Presumed zeroed, in order of appearance:
1905 * cookie_in_always, cookie_out_never,
1906 * s_data_constant, s_data_in, s_data_out
1907 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1909 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1910
Herbert Xueb4dea52008-12-29 23:04:08 -08001911 local_bh_disable();
Eric Dumazet17483762008-11-25 21:16:35 -08001912 percpu_counter_inc(&tcp_sockets_allocated);
Herbert Xueb4dea52008-12-29 23:04:08 -08001913 local_bh_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
1915 return 0;
1916}
1917
Brian Haley7d06b2e2008-06-14 17:04:49 -07001918void tcp_v4_destroy_sock(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919{
1920 struct tcp_sock *tp = tcp_sk(sk);
1921
1922 tcp_clear_xmit_timers(sk);
1923
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001924 tcp_cleanup_congestion_control(sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 /* Cleanup up the write buffer. */
David S. Millerfe067e82007-03-07 12:12:44 -08001927 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
1929 /* Cleans up our, hopefully empty, out_of_order_queue. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001930 __skb_queue_purge(&tp->out_of_order_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001932#ifdef CONFIG_TCP_MD5SIG
1933 /* Clean up the MD5 key list, if any */
1934 if (tp->md5sig_info) {
1935 tcp_v4_clear_md5_list(sk);
1936 kfree(tp->md5sig_info);
1937 tp->md5sig_info = NULL;
1938 }
1939#endif
1940
Chris Leech1a2449a2006-05-23 18:05:53 -07001941#ifdef CONFIG_NET_DMA
1942 /* Cleans up our sk_async_wait_queue */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001943 __skb_queue_purge(&sk->sk_async_wait_queue);
Chris Leech1a2449a2006-05-23 18:05:53 -07001944#endif
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 /* Clean prequeue, it must be empty really */
1947 __skb_queue_purge(&tp->ucopy.prequeue);
1948
1949 /* Clean up a referenced TCP bind bucket. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001950 if (inet_csk(sk)->icsk_bind_hash)
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001951 inet_put_port(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /*
1954 * If sendmsg cached page exists, toss it.
1955 */
1956 if (sk->sk_sndmsg_page) {
1957 __free_page(sk->sk_sndmsg_page);
1958 sk->sk_sndmsg_page = NULL;
1959 }
1960
William Allen Simpson435cf552009-12-02 18:17:05 +00001961 /* TCP Cookie Transactions */
1962 if (tp->cookie_values != NULL) {
1963 kref_put(&tp->cookie_values->kref,
1964 tcp_cookie_values_release);
1965 tp->cookie_values = NULL;
1966 }
1967
Eric Dumazet17483762008-11-25 21:16:35 -08001968 percpu_counter_dec(&tcp_sockets_allocated);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970EXPORT_SYMBOL(tcp_v4_destroy_sock);
1971
1972#ifdef CONFIG_PROC_FS
1973/* Proc filesystem TCP sock list dumping. */
1974
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001975static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001977 return hlist_nulls_empty(head) ? NULL :
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001978 list_entry(head->first, struct inet_timewait_sock, tw_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979}
1980
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07001981static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001983 return !is_a_nulls(tw->tw_node.next) ?
1984 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985}
1986
Tom Herberta8b690f2010-06-07 00:43:42 -07001987/*
1988 * Get next listener socket follow cur. If cur is NULL, get first socket
1989 * starting from bucket given in st->bucket; when st->bucket is zero the
1990 * very first socket in the hash table is returned.
1991 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992static void *listening_get_next(struct seq_file *seq, void *cur)
1993{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001994 struct inet_connection_sock *icsk;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001995 struct hlist_nulls_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 struct sock *sk = cur;
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001997 struct inet_listen_hashbucket *ilb;
Jianjun Kong5799de02008-11-03 02:49:10 -08001998 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07001999 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
2001 if (!sk) {
Tom Herberta8b690f2010-06-07 00:43:42 -07002002 ilb = &tcp_hashinfo.listening_hash[st->bucket];
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002003 spin_lock_bh(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08002004 sk = sk_nulls_head(&ilb->head);
Tom Herberta8b690f2010-06-07 00:43:42 -07002005 st->offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 goto get_sk;
2007 }
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002008 ilb = &tcp_hashinfo.listening_hash[st->bucket];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 ++st->num;
Tom Herberta8b690f2010-06-07 00:43:42 -07002010 ++st->offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
2012 if (st->state == TCP_SEQ_STATE_OPENREQ) {
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07002013 struct request_sock *req = cur;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Eric Dumazet72a3eff2006-11-16 02:30:37 -08002015 icsk = inet_csk(st->syn_wait_sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 req = req->dl_next;
2017 while (1) {
2018 while (req) {
Daniel Lezcanobdccc4c2008-07-19 00:15:13 -07002019 if (req->rsk_ops->family == st->family) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 cur = req;
2021 goto out;
2022 }
2023 req = req->dl_next;
2024 }
Tom Herberta8b690f2010-06-07 00:43:42 -07002025 st->offset = 0;
Eric Dumazet72a3eff2006-11-16 02:30:37 -08002026 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 break;
2028get_req:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002029 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 }
2031 sk = sk_next(st->syn_wait_sk);
2032 st->state = TCP_SEQ_STATE_LISTENING;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002033 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 } else {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002035 icsk = inet_csk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002036 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2037 if (reqsk_queue_len(&icsk->icsk_accept_queue))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 goto start_req;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002039 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 sk = sk_next(sk);
2041 }
2042get_sk:
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08002043 sk_nulls_for_each_from(sk, node) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002044 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 cur = sk;
2046 goto out;
2047 }
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09002048 icsk = inet_csk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002049 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2050 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051start_req:
2052 st->uid = sock_i_uid(sk);
2053 st->syn_wait_sk = sk;
2054 st->state = TCP_SEQ_STATE_OPENREQ;
2055 st->sbucket = 0;
2056 goto get_req;
2057 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002058 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 }
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002060 spin_unlock_bh(&ilb->lock);
Tom Herberta8b690f2010-06-07 00:43:42 -07002061 st->offset = 0;
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07002062 if (++st->bucket < INET_LHTABLE_SIZE) {
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002063 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2064 spin_lock_bh(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08002065 sk = sk_nulls_head(&ilb->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 goto get_sk;
2067 }
2068 cur = NULL;
2069out:
2070 return cur;
2071}
2072
2073static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2074{
Tom Herberta8b690f2010-06-07 00:43:42 -07002075 struct tcp_iter_state *st = seq->private;
2076 void *rc;
2077
2078 st->bucket = 0;
2079 st->offset = 0;
2080 rc = listening_get_next(seq, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
2082 while (rc && *pos) {
2083 rc = listening_get_next(seq, rc);
2084 --*pos;
2085 }
2086 return rc;
2087}
2088
Andi Kleen6eac5602008-08-28 01:08:02 -07002089static inline int empty_bucket(struct tcp_iter_state *st)
2090{
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002091 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2092 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
Andi Kleen6eac5602008-08-28 01:08:02 -07002093}
2094
Tom Herberta8b690f2010-06-07 00:43:42 -07002095/*
2096 * Get first established socket starting from bucket given in st->bucket.
2097 * If st->bucket is zero, the very first socket in the hash is returned.
2098 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099static void *established_get_first(struct seq_file *seq)
2100{
Jianjun Kong5799de02008-11-03 02:49:10 -08002101 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07002102 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 void *rc = NULL;
2104
Tom Herberta8b690f2010-06-07 00:43:42 -07002105 st->offset = 0;
2106 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002108 struct hlist_nulls_node *node;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002109 struct inet_timewait_sock *tw;
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002110 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Andi Kleen6eac5602008-08-28 01:08:02 -07002112 /* Lockless fast path for the common case of empty buckets */
2113 if (empty_bucket(st))
2114 continue;
2115
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002116 spin_lock_bh(lock);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002117 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002118 if (sk->sk_family != st->family ||
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002119 !net_eq(sock_net(sk), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 continue;
2121 }
2122 rc = sk;
2123 goto out;
2124 }
2125 st->state = TCP_SEQ_STATE_TIME_WAIT;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002126 inet_twsk_for_each(tw, node,
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002127 &tcp_hashinfo.ehash[st->bucket].twchain) {
Pavel Emelyanov28518fc2008-03-21 15:52:00 -07002128 if (tw->tw_family != st->family ||
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002129 !net_eq(twsk_net(tw), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 continue;
2131 }
2132 rc = tw;
2133 goto out;
2134 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002135 spin_unlock_bh(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 st->state = TCP_SEQ_STATE_ESTABLISHED;
2137 }
2138out:
2139 return rc;
2140}
2141
2142static void *established_get_next(struct seq_file *seq, void *cur)
2143{
2144 struct sock *sk = cur;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -07002145 struct inet_timewait_sock *tw;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002146 struct hlist_nulls_node *node;
Jianjun Kong5799de02008-11-03 02:49:10 -08002147 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07002148 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
2150 ++st->num;
Tom Herberta8b690f2010-06-07 00:43:42 -07002151 ++st->offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
2153 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2154 tw = cur;
2155 tw = tw_next(tw);
2156get_tw:
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002157 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 tw = tw_next(tw);
2159 }
2160 if (tw) {
2161 cur = tw;
2162 goto out;
2163 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002164 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 st->state = TCP_SEQ_STATE_ESTABLISHED;
2166
Andi Kleen6eac5602008-08-28 01:08:02 -07002167 /* Look for next non empty bucket */
Tom Herberta8b690f2010-06-07 00:43:42 -07002168 st->offset = 0;
Eric Dumazetf373b532009-10-09 00:16:19 +00002169 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
Andi Kleen6eac5602008-08-28 01:08:02 -07002170 empty_bucket(st))
2171 ;
Eric Dumazetf373b532009-10-09 00:16:19 +00002172 if (st->bucket > tcp_hashinfo.ehash_mask)
Andi Kleen6eac5602008-08-28 01:08:02 -07002173 return NULL;
2174
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002175 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002176 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 } else
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002178 sk = sk_nulls_next(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002180 sk_nulls_for_each_from(sk, node) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09002181 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 goto found;
2183 }
2184
2185 st->state = TCP_SEQ_STATE_TIME_WAIT;
Eric Dumazetdbca9b2752007-02-08 14:16:46 -08002186 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 goto get_tw;
2188found:
2189 cur = sk;
2190out:
2191 return cur;
2192}
2193
2194static void *established_get_idx(struct seq_file *seq, loff_t pos)
2195{
Tom Herberta8b690f2010-06-07 00:43:42 -07002196 struct tcp_iter_state *st = seq->private;
2197 void *rc;
2198
2199 st->bucket = 0;
2200 rc = established_get_first(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201
2202 while (rc && pos) {
2203 rc = established_get_next(seq, rc);
2204 --pos;
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02002205 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 return rc;
2207}
2208
2209static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2210{
2211 void *rc;
Jianjun Kong5799de02008-11-03 02:49:10 -08002212 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 st->state = TCP_SEQ_STATE_LISTENING;
2215 rc = listening_get_idx(seq, &pos);
2216
2217 if (!rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 st->state = TCP_SEQ_STATE_ESTABLISHED;
2219 rc = established_get_idx(seq, pos);
2220 }
2221
2222 return rc;
2223}
2224
Tom Herberta8b690f2010-06-07 00:43:42 -07002225static void *tcp_seek_last_pos(struct seq_file *seq)
2226{
2227 struct tcp_iter_state *st = seq->private;
2228 int offset = st->offset;
2229 int orig_num = st->num;
2230 void *rc = NULL;
2231
2232 switch (st->state) {
2233 case TCP_SEQ_STATE_OPENREQ:
2234 case TCP_SEQ_STATE_LISTENING:
2235 if (st->bucket >= INET_LHTABLE_SIZE)
2236 break;
2237 st->state = TCP_SEQ_STATE_LISTENING;
2238 rc = listening_get_next(seq, NULL);
2239 while (offset-- && rc)
2240 rc = listening_get_next(seq, rc);
2241 if (rc)
2242 break;
2243 st->bucket = 0;
2244 /* Fallthrough */
2245 case TCP_SEQ_STATE_ESTABLISHED:
2246 case TCP_SEQ_STATE_TIME_WAIT:
2247 st->state = TCP_SEQ_STATE_ESTABLISHED;
2248 if (st->bucket > tcp_hashinfo.ehash_mask)
2249 break;
2250 rc = established_get_first(seq);
2251 while (offset-- && rc)
2252 rc = established_get_next(seq, rc);
2253 }
2254
2255 st->num = orig_num;
2256
2257 return rc;
2258}
2259
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2261{
Jianjun Kong5799de02008-11-03 02:49:10 -08002262 struct tcp_iter_state *st = seq->private;
Tom Herberta8b690f2010-06-07 00:43:42 -07002263 void *rc;
2264
2265 if (*pos && *pos == st->last_pos) {
2266 rc = tcp_seek_last_pos(seq);
2267 if (rc)
2268 goto out;
2269 }
2270
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 st->state = TCP_SEQ_STATE_LISTENING;
2272 st->num = 0;
Tom Herberta8b690f2010-06-07 00:43:42 -07002273 st->bucket = 0;
2274 st->offset = 0;
2275 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2276
2277out:
2278 st->last_pos = *pos;
2279 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280}
2281
2282static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2283{
Tom Herberta8b690f2010-06-07 00:43:42 -07002284 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 void *rc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287 if (v == SEQ_START_TOKEN) {
2288 rc = tcp_get_idx(seq, 0);
2289 goto out;
2290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
2292 switch (st->state) {
2293 case TCP_SEQ_STATE_OPENREQ:
2294 case TCP_SEQ_STATE_LISTENING:
2295 rc = listening_get_next(seq, v);
2296 if (!rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 st->state = TCP_SEQ_STATE_ESTABLISHED;
Tom Herberta8b690f2010-06-07 00:43:42 -07002298 st->bucket = 0;
2299 st->offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 rc = established_get_first(seq);
2301 }
2302 break;
2303 case TCP_SEQ_STATE_ESTABLISHED:
2304 case TCP_SEQ_STATE_TIME_WAIT:
2305 rc = established_get_next(seq, v);
2306 break;
2307 }
2308out:
2309 ++*pos;
Tom Herberta8b690f2010-06-07 00:43:42 -07002310 st->last_pos = *pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 return rc;
2312}
2313
2314static void tcp_seq_stop(struct seq_file *seq, void *v)
2315{
Jianjun Kong5799de02008-11-03 02:49:10 -08002316 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
2318 switch (st->state) {
2319 case TCP_SEQ_STATE_OPENREQ:
2320 if (v) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002321 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2322 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 }
2324 case TCP_SEQ_STATE_LISTENING:
2325 if (v != SEQ_START_TOKEN)
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002326 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 break;
2328 case TCP_SEQ_STATE_TIME_WAIT:
2329 case TCP_SEQ_STATE_ESTABLISHED:
2330 if (v)
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002331 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 break;
2333 }
2334}
2335
2336static int tcp_seq_open(struct inode *inode, struct file *file)
2337{
2338 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 struct tcp_iter_state *s;
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002340 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002342 err = seq_open_net(inode, file, &afinfo->seq_ops,
2343 sizeof(struct tcp_iter_state));
2344 if (err < 0)
2345 return err;
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002346
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002347 s = ((struct seq_file *)file->private_data)->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 s->family = afinfo->family;
Tom Herberta8b690f2010-06-07 00:43:42 -07002349 s->last_pos = 0;
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002350 return 0;
2351}
2352
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002353int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354{
2355 int rc = 0;
2356 struct proc_dir_entry *p;
2357
Denis V. Lunev68fcadd2008-04-13 22:13:30 -07002358 afinfo->seq_fops.open = tcp_seq_open;
2359 afinfo->seq_fops.read = seq_read;
2360 afinfo->seq_fops.llseek = seq_lseek;
2361 afinfo->seq_fops.release = seq_release_net;
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02002362
Denis V. Lunev9427c4b2008-04-13 22:12:13 -07002363 afinfo->seq_ops.start = tcp_seq_start;
2364 afinfo->seq_ops.next = tcp_seq_next;
2365 afinfo->seq_ops.stop = tcp_seq_stop;
2366
Denis V. Lunev84841c32008-05-02 04:10:08 -07002367 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2368 &afinfo->seq_fops, afinfo);
2369 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 rc = -ENOMEM;
2371 return rc;
2372}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002373EXPORT_SYMBOL(tcp_proc_register);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002375void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376{
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002377 proc_net_remove(net, afinfo->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002379EXPORT_SYMBOL(tcp_proc_unregister);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07002381static void get_openreq4(struct sock *sk, struct request_sock *req,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002382 struct seq_file *f, int i, int uid, int *len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002384 const struct inet_request_sock *ireq = inet_rsk(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 int ttd = req->expires - jiffies;
2386
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002387 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2388 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 i,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002390 ireq->loc_addr,
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002391 ntohs(inet_sk(sk)->inet_sport),
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002392 ireq->rmt_addr,
2393 ntohs(ireq->rmt_port),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 TCP_SYN_RECV,
2395 0, 0, /* could print option size, but that is af dependent. */
2396 1, /* timers active (only the expire timer) */
2397 jiffies_to_clock_t(ttd),
2398 req->retrans,
2399 uid,
2400 0, /* non standard timer */
2401 0, /* open_requests have no inode */
2402 atomic_read(&sk->sk_refcnt),
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002403 req,
2404 len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405}
2406
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002407static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
2409 int timer_active;
2410 unsigned long timer_expires;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002411 struct tcp_sock *tp = tcp_sk(sk);
2412 const struct inet_connection_sock *icsk = inet_csk(sk);
2413 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002414 __be32 dest = inet->inet_daddr;
2415 __be32 src = inet->inet_rcv_saddr;
2416 __u16 destp = ntohs(inet->inet_dport);
2417 __u16 srcp = ntohs(inet->inet_sport);
Eric Dumazet49d09002009-12-03 16:06:13 -08002418 int rx_queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002420 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 timer_active = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002422 timer_expires = icsk->icsk_timeout;
2423 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 timer_active = 4;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002425 timer_expires = icsk->icsk_timeout;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002426 } else if (timer_pending(&sk->sk_timer)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 timer_active = 2;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002428 timer_expires = sk->sk_timer.expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 } else {
2430 timer_active = 0;
2431 timer_expires = jiffies;
2432 }
2433
Eric Dumazet49d09002009-12-03 16:06:13 -08002434 if (sk->sk_state == TCP_LISTEN)
2435 rx_queue = sk->sk_ack_backlog;
2436 else
2437 /*
2438 * because we dont lock socket, we might find a transient negative value
2439 */
2440 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2441
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002442 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
Stephen Hemminger7be87352008-06-27 20:00:19 -07002443 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002444 i, src, srcp, dest, destp, sk->sk_state,
Sridhar Samudrala47da8ee2006-06-27 13:29:00 -07002445 tp->write_seq - tp->snd_una,
Eric Dumazet49d09002009-12-03 16:06:13 -08002446 rx_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 timer_active,
2448 jiffies_to_clock_t(timer_expires - jiffies),
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002449 icsk->icsk_retransmits,
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002450 sock_i_uid(sk),
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002451 icsk->icsk_probes_out,
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002452 sock_i_ino(sk),
2453 atomic_read(&sk->sk_refcnt), sk,
Stephen Hemminger7be87352008-06-27 20:00:19 -07002454 jiffies_to_clock_t(icsk->icsk_rto),
2455 jiffies_to_clock_t(icsk->icsk_ack.ato),
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002456 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 tp->snd_cwnd,
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07002458 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002459 len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460}
2461
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02002462static void get_timewait4_sock(struct inet_timewait_sock *tw,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002463 struct seq_file *f, int i, int *len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464{
Al Viro23f33c22006-09-27 18:43:50 -07002465 __be32 dest, src;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 __u16 destp, srcp;
2467 int ttd = tw->tw_ttd - jiffies;
2468
2469 if (ttd < 0)
2470 ttd = 0;
2471
2472 dest = tw->tw_daddr;
2473 src = tw->tw_rcv_saddr;
2474 destp = ntohs(tw->tw_dport);
2475 srcp = ntohs(tw->tw_sport);
2476
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002477 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2478 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2480 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002481 atomic_read(&tw->tw_refcnt), tw, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482}
2483
2484#define TMPSZ 150
2485
2486static int tcp4_seq_show(struct seq_file *seq, void *v)
2487{
Jianjun Kong5799de02008-11-03 02:49:10 -08002488 struct tcp_iter_state *st;
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002489 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490
2491 if (v == SEQ_START_TOKEN) {
2492 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2493 " sl local_address rem_address st tx_queue "
2494 "rx_queue tr tm->when retrnsmt uid timeout "
2495 "inode");
2496 goto out;
2497 }
2498 st = seq->private;
2499
2500 switch (st->state) {
2501 case TCP_SEQ_STATE_LISTENING:
2502 case TCP_SEQ_STATE_ESTABLISHED:
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002503 get_tcp4_sock(v, seq, st->num, &len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 break;
2505 case TCP_SEQ_STATE_OPENREQ:
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002506 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 break;
2508 case TCP_SEQ_STATE_TIME_WAIT:
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002509 get_timewait4_sock(v, seq, st->num, &len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 break;
2511 }
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002512 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513out:
2514 return 0;
2515}
2516
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517static struct tcp_seq_afinfo tcp4_seq_afinfo = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 .name = "tcp",
2519 .family = AF_INET,
Denis V. Lunev5f4472c2008-04-13 22:13:53 -07002520 .seq_fops = {
2521 .owner = THIS_MODULE,
2522 },
Denis V. Lunev9427c4b2008-04-13 22:12:13 -07002523 .seq_ops = {
2524 .show = tcp4_seq_show,
2525 },
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526};
2527
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002528static int __net_init tcp4_proc_init_net(struct net *net)
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002529{
2530 return tcp_proc_register(net, &tcp4_seq_afinfo);
2531}
2532
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002533static void __net_exit tcp4_proc_exit_net(struct net *net)
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002534{
2535 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2536}
2537
2538static struct pernet_operations tcp4_net_ops = {
2539 .init = tcp4_proc_init_net,
2540 .exit = tcp4_proc_exit_net,
2541};
2542
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543int __init tcp4_proc_init(void)
2544{
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002545 return register_pernet_subsys(&tcp4_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546}
2547
2548void tcp4_proc_exit(void)
2549{
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002550 unregister_pernet_subsys(&tcp4_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551}
2552#endif /* CONFIG_PROC_FS */
2553
Herbert Xubf296b12008-12-15 23:43:36 -08002554struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2555{
Herbert Xu36e7b1b2009-04-27 05:44:45 -07002556 struct iphdr *iph = skb_gro_network_header(skb);
Herbert Xubf296b12008-12-15 23:43:36 -08002557
2558 switch (skb->ip_summed) {
2559 case CHECKSUM_COMPLETE:
Herbert Xu86911732009-01-29 14:19:50 +00002560 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
Herbert Xubf296b12008-12-15 23:43:36 -08002561 skb->csum)) {
2562 skb->ip_summed = CHECKSUM_UNNECESSARY;
2563 break;
2564 }
2565
2566 /* fall through */
2567 case CHECKSUM_NONE:
2568 NAPI_GRO_CB(skb)->flush = 1;
2569 return NULL;
2570 }
2571
2572 return tcp_gro_receive(head, skb);
2573}
2574EXPORT_SYMBOL(tcp4_gro_receive);
2575
2576int tcp4_gro_complete(struct sk_buff *skb)
2577{
2578 struct iphdr *iph = ip_hdr(skb);
2579 struct tcphdr *th = tcp_hdr(skb);
2580
2581 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2582 iph->saddr, iph->daddr, 0);
2583 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2584
2585 return tcp_gro_complete(skb);
2586}
2587EXPORT_SYMBOL(tcp4_gro_complete);
2588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589struct proto tcp_prot = {
2590 .name = "TCP",
2591 .owner = THIS_MODULE,
2592 .close = tcp_close,
2593 .connect = tcp_v4_connect,
2594 .disconnect = tcp_disconnect,
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002595 .accept = inet_csk_accept,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 .ioctl = tcp_ioctl,
2597 .init = tcp_v4_init_sock,
2598 .destroy = tcp_v4_destroy_sock,
2599 .shutdown = tcp_shutdown,
2600 .setsockopt = tcp_setsockopt,
2601 .getsockopt = tcp_getsockopt,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 .recvmsg = tcp_recvmsg,
Changli Gao7ba42912010-07-10 20:41:55 +00002603 .sendmsg = tcp_sendmsg,
2604 .sendpage = tcp_sendpage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 .backlog_rcv = tcp_v4_do_rcv,
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08002606 .hash = inet_hash,
2607 .unhash = inet_unhash,
2608 .get_port = inet_csk_get_port,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 .enter_memory_pressure = tcp_enter_memory_pressure,
2610 .sockets_allocated = &tcp_sockets_allocated,
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002611 .orphan_count = &tcp_orphan_count,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 .memory_allocated = &tcp_memory_allocated,
2613 .memory_pressure = &tcp_memory_pressure,
2614 .sysctl_mem = sysctl_tcp_mem,
2615 .sysctl_wmem = sysctl_tcp_wmem,
2616 .sysctl_rmem = sysctl_tcp_rmem,
2617 .max_header = MAX_TCP_HEADER,
2618 .obj_size = sizeof(struct tcp_sock),
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002619 .slab_flags = SLAB_DESTROY_BY_RCU,
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08002620 .twsk_prot = &tcp_timewait_sock_ops,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07002621 .rsk_prot = &tcp_request_sock_ops,
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -07002622 .h.hashinfo = &tcp_hashinfo,
Changli Gao7ba42912010-07-10 20:41:55 +00002623 .no_autobind = true,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002624#ifdef CONFIG_COMPAT
2625 .compat_setsockopt = compat_tcp_setsockopt,
2626 .compat_getsockopt = compat_tcp_getsockopt,
2627#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628};
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002629EXPORT_SYMBOL(tcp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
Denis V. Lunev046ee902008-04-03 14:31:33 -07002631
2632static int __net_init tcp_sk_init(struct net *net)
2633{
2634 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2635 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2636}
2637
2638static void __net_exit tcp_sk_exit(struct net *net)
2639{
2640 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
Eric W. Biedermanb099ce22009-12-03 02:29:09 +00002641}
2642
2643static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2644{
2645 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
Denis V. Lunev046ee902008-04-03 14:31:33 -07002646}
2647
2648static struct pernet_operations __net_initdata tcp_sk_ops = {
Eric W. Biedermanb099ce22009-12-03 02:29:09 +00002649 .init = tcp_sk_init,
2650 .exit = tcp_sk_exit,
2651 .exit_batch = tcp_sk_exit_batch,
Denis V. Lunev046ee902008-04-03 14:31:33 -07002652};
2653
Denis V. Lunev9b0f9762008-02-29 11:13:15 -08002654void __init tcp_v4_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655{
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002656 inet_hashinfo_init(&tcp_hashinfo);
Eric W. Biederman6a1b3052009-02-22 00:10:18 -08002657 if (register_pernet_subsys(&tcp_sk_ops))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 panic("Failed to create the TCP control socket.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659}