blob: 328e0cf42b3c6e164c09da5d54b29566ff5c1480 [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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Corey Minyard <wf-rch!minyard@relay.EU.net>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14 * Linus Torvalds, <torvalds@cs.helsinki.fi>
15 * Alan Cox, <gw4pts@gw4pts.ampr.org>
16 * Matthew Dillon, <dillon@apollo.west.oic.com>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 * Jorge Cwik, <jorge@laser.satlink.net>
19 */
20
21#include <linux/module.h>
22#include <net/tcp.h>
23
Brian Haleyab32ea52006-09-22 14:15:41 -070024int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES;
25int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES;
26int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME;
27int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES;
28int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL;
29int sysctl_tcp_retries1 __read_mostly = TCP_RETR1;
30int sysctl_tcp_retries2 __read_mostly = TCP_RETR2;
31int sysctl_tcp_orphan_retries __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33static void tcp_write_timer(unsigned long);
34static void tcp_delack_timer(unsigned long);
35static void tcp_keepalive_timer (unsigned long data);
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037void tcp_init_xmit_timers(struct sock *sk)
38{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070039 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
40 &tcp_keepalive_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041}
42
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070043EXPORT_SYMBOL(tcp_init_xmit_timers);
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045static void tcp_write_err(struct sock *sk)
46{
47 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
48 sk->sk_error_report(sk);
49
50 tcp_done(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -070051 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052}
53
54/* Do not allow orphaned sockets to eat all our resources.
55 * This is direct violation of TCP specs, but it is required
56 * to prevent DoS attacks. It is called when a retransmission timeout
57 * or zero probe timeout occurs on orphaned socket.
58 *
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080059 * Criteria is still not confirmed experimentally and may change.
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * We kill the socket, if:
61 * 1. If number of orphaned sockets exceeds an administratively configured
62 * limit.
63 * 2. If we have strong memory pressure.
64 */
65static int tcp_out_of_resources(struct sock *sk, int do_reset)
66{
67 struct tcp_sock *tp = tcp_sk(sk);
68 int orphans = atomic_read(&tcp_orphan_count);
69
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090070 /* If peer does not open window for long time, or did not transmit
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * anything for long time, penalize it. */
72 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
73 orphans <<= 1;
74
75 /* If some dubious ICMP arrived, penalize even more. */
76 if (sk->sk_err_soft)
77 orphans <<= 1;
78
Pavel Emelianove4fd5da2007-05-29 13:19:18 -070079 if (tcp_too_many_orphans(sk, orphans)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 if (net_ratelimit())
81 printk(KERN_INFO "Out of socket memory\n");
82
83 /* Catch exceptional cases, when connection requires reset.
84 * 1. Last segment was sent recently. */
85 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
86 /* 2. Window is closed. */
87 (!tp->snd_wnd && !tp->packets_out))
88 do_reset = 1;
89 if (do_reset)
90 tcp_send_active_reset(sk, GFP_ATOMIC);
91 tcp_done(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -070092 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 return 1;
94 }
95 return 0;
96}
97
98/* Calculate maximal number or retries on an orphaned socket. */
99static int tcp_orphan_retries(struct sock *sk, int alive)
100{
101 int retries = sysctl_tcp_orphan_retries; /* May be zero. */
102
103 /* We know from an ICMP that something is wrong. */
104 if (sk->sk_err_soft && !alive)
105 retries = 0;
106
107 /* However, if socket sent something recently, select some safe
108 * number of retries. 8 corresponds to >100 seconds with minimal
109 * RTO of 200msec. */
110 if (retries == 0 && alive)
111 retries = 8;
112 return retries;
113}
114
Eric Dumazetce55dd32007-12-21 01:50:43 -0800115static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
116{
Eric Dumazetce55dd32007-12-21 01:50:43 -0800117 /* Black hole detection */
118 if (sysctl_tcp_mtu_probing) {
119 if (!icsk->icsk_mtup.enabled) {
120 icsk->icsk_mtup.enabled = 1;
121 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
122 } else {
123 struct tcp_sock *tp = tcp_sk(sk);
David S. Miller829942c2007-12-21 04:29:16 -0800124 int mss;
125
Eric Dumazet8beb5c52007-12-21 05:58:29 -0800126 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
Eric Dumazetce55dd32007-12-21 01:50:43 -0800127 mss = min(sysctl_tcp_base_mss, mss);
128 mss = max(mss, 68 - tp->tcp_header_len);
129 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
130 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
131 }
132 }
133}
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/* A write timeout has occurred. Process the after effects. */
136static int tcp_write_timeout(struct sock *sk)
137{
John Heffner5d424d52006-03-20 17:53:41 -0800138 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 int retry_until;
140
141 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700142 if (icsk->icsk_retransmits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 dst_negative_advice(&sk->sk_dst_cache);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700144 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700146 if (icsk->icsk_retransmits >= sysctl_tcp_retries1) {
John Heffner5d424d52006-03-20 17:53:41 -0800147 /* Black hole detection */
Eric Dumazetce55dd32007-12-21 01:50:43 -0800148 tcp_mtu_probing(icsk, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
150 dst_negative_advice(&sk->sk_dst_cache);
151 }
152
153 retry_until = sysctl_tcp_retries2;
154 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700155 const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 retry_until = tcp_orphan_retries(sk, alive);
158
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700159 if (tcp_out_of_resources(sk, alive || icsk->icsk_retransmits < retry_until))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 return 1;
161 }
162 }
163
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700164 if (icsk->icsk_retransmits >= retry_until) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 /* Has it gone just too far? */
166 tcp_write_err(sk);
167 return 1;
168 }
169 return 0;
170}
171
172static void tcp_delack_timer(unsigned long data)
173{
174 struct sock *sk = (struct sock*)data;
175 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700176 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178 bh_lock_sock(sk);
179 if (sock_owned_by_user(sk)) {
180 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700181 icsk->icsk_ack.blocked = 1;
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700182 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700183 sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 goto out_unlock;
185 }
186
David S. Miller9993e7d2008-01-10 21:56:38 -0800187 sk_mem_reclaim_partial(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700189 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 goto out;
191
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700192 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
193 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 goto out;
195 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700196 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
David S. Millerb03efcf2005-07-08 14:57:23 -0700198 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 struct sk_buff *skb;
200
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700201 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
204 sk->sk_backlog_rcv(sk, skb);
205
206 tp->ucopy.memory = 0;
207 }
208
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700209 if (inet_csk_ack_scheduled(sk)) {
210 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700212 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 } else {
214 /* Delayed ACK missed: leave pingpong mode and
215 * deflate ATO.
216 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700217 icsk->icsk_ack.pingpong = 0;
218 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 }
220 tcp_send_ack(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700221 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 }
223 TCP_CHECK_TIMER(sk);
224
225out:
226 if (tcp_memory_pressure)
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800227 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228out_unlock:
229 bh_unlock_sock(sk);
230 sock_put(sk);
231}
232
233static void tcp_probe_timer(struct sock *sk)
234{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300235 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 struct tcp_sock *tp = tcp_sk(sk);
237 int max_probes;
238
David S. Millerfe067e82007-03-07 12:12:44 -0800239 if (tp->packets_out || !tcp_send_head(sk)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300240 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 return;
242 }
243
244 /* *WARNING* RFC 1122 forbids this
245 *
246 * It doesn't AFAIK, because we kill the retransmit timer -AK
247 *
248 * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
249 * this behaviour in Solaris down as a bug fix. [AC]
250 *
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300251 * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 * even if they advertise zero window. Hence, connection is killed only
253 * if we received no ACKs for normal connection timeout. It is not killed
254 * only because window stays zero for some time, window may be zero
255 * until armageddon and even later. We are in full accordance
256 * with RFCs, only probe timer combines both retransmission timeout
257 * and probe timeout in one bottle. --ANK
258 */
259 max_probes = sysctl_tcp_retries2;
260
261 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700262 const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 max_probes = tcp_orphan_retries(sk, alive);
265
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300266 if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 return;
268 }
269
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300270 if (icsk->icsk_probes_out > max_probes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 tcp_write_err(sk);
272 } else {
273 /* Only send another probe if we didn't close things up. */
274 tcp_send_probe0(sk);
275 }
276}
277
278/*
279 * The TCP retransmit timer.
280 */
281
282static void tcp_retransmit_timer(struct sock *sk)
283{
284 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700285 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 if (!tp->packets_out)
288 goto out;
289
David S. Millerfe067e82007-03-07 12:12:44 -0800290 BUG_TRAP(!tcp_write_queue_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
292 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
293 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
294 /* Receiver dastardly shrinks window. Our retransmits
295 * become zero probes, but we should not timeout this
296 * connection. If the socket is an orphan, time it out,
297 * we cannot allow such beasts to hang infinitely.
298 */
299#ifdef TCP_DEBUG
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700300 struct inet_sock *inet = inet_sk(sk);
301 if (sk->sk_family == AF_INET) {
YOSHIFUJI Hideakia7d632b2008-04-14 04:09:00 -0700302 LIMIT_NETDEBUG(KERN_DEBUG "TCP: Treason uncloaked! Peer " NIPQUAD_FMT ":%u/%u shrinks window %u:%u. Repaired.\n",
Al Viro714e85b2006-11-14 20:51:49 -0800303 NIPQUAD(inet->daddr), ntohs(inet->dport),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 inet->num, tp->snd_una, tp->snd_nxt);
305 }
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700306#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
307 else if (sk->sk_family == AF_INET6) {
308 struct ipv6_pinfo *np = inet6_sk(sk);
309 LIMIT_NETDEBUG(KERN_DEBUG "TCP: Treason uncloaked! Peer " NIP6_FMT ":%u/%u shrinks window %u:%u. Repaired.\n",
310 NIP6(np->daddr), ntohs(inet->dport),
311 inet->num, tp->snd_una, tp->snd_nxt);
312 }
313#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314#endif
315 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
316 tcp_write_err(sk);
317 goto out;
318 }
319 tcp_enter_loss(sk, 0);
David S. Millerfe067e82007-03-07 12:12:44 -0800320 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 __sk_dst_reset(sk);
322 goto out_reset_timer;
323 }
324
325 if (tcp_write_timeout(sk))
326 goto out;
327
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700328 if (icsk->icsk_retransmits == 0) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700329 int mib_idx;
330
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300331 if (icsk->icsk_ca_state == TCP_CA_Disorder ||
332 icsk->icsk_ca_state == TCP_CA_Recovery) {
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300333 if (tcp_is_sack(tp)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300334 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700335 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 else
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700337 mib_idx = LINUX_MIB_TCPSACKFAILURES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 } else {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300339 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700340 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 else
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700342 mib_idx = LINUX_MIB_TCPRENOFAILURES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 }
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300344 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700345 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 } else {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700347 mib_idx = LINUX_MIB_TCPTIMEOUTS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 }
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700349 NET_INC_STATS_BH(sock_net(sk), mib_idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
351
352 if (tcp_use_frto(sk)) {
353 tcp_enter_frto(sk);
354 } else {
355 tcp_enter_loss(sk, 0);
356 }
357
David S. Millerfe067e82007-03-07 12:12:44 -0800358 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 /* Retransmission failed because of local congestion,
360 * do not backoff.
361 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700362 if (!icsk->icsk_retransmits)
363 icsk->icsk_retransmits = 1;
364 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700365 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
366 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 goto out;
368 }
369
370 /* Increase the timeout each time we retransmit. Note that
371 * we do not increase the rtt estimate. rto is initialized
372 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
373 * that doubling rto each time is the least we can get away with.
374 * In KA9Q, Karn uses this for the first few times, and then
375 * goes to quadratic. netBSD doubles, but only goes up to *64,
376 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
377 * defined in the protocol as the maximum possible RTT. I guess
378 * we'll have to use something other than TCP to talk to the
379 * University of Mars.
380 *
381 * PAWS allows us longer timeouts and large windows, so once
382 * implemented ftp to mars will work nicely. We will have to fix
383 * the 120 second clamps though!
384 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700385 icsk->icsk_backoff++;
386 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
388out_reset_timer:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700389 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700390 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700391 if (icsk->icsk_retransmits > sysctl_tcp_retries1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 __sk_dst_reset(sk);
393
394out:;
395}
396
397static void tcp_write_timer(unsigned long data)
398{
399 struct sock *sk = (struct sock*)data;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700400 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 int event;
402
403 bh_lock_sock(sk);
404 if (sock_owned_by_user(sk)) {
405 /* Try again later */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700406 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 goto out_unlock;
408 }
409
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700410 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 goto out;
412
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700413 if (time_after(icsk->icsk_timeout, jiffies)) {
414 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 goto out;
416 }
417
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700418 event = icsk->icsk_pending;
419 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421 switch (event) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700422 case ICSK_TIME_RETRANS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 tcp_retransmit_timer(sk);
424 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700425 case ICSK_TIME_PROBE0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 tcp_probe_timer(sk);
427 break;
428 }
429 TCP_CHECK_TIMER(sk);
430
431out:
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800432 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433out_unlock:
434 bh_unlock_sock(sk);
435 sock_put(sk);
436}
437
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700438/*
439 * Timer for listening sockets
440 */
441
442static void tcp_synack_timer(struct sock *sk)
443{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700444 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
445 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448void tcp_set_keepalive(struct sock *sk, int val)
449{
450 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
451 return;
452
453 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700454 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700456 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457}
458
459
460static void tcp_keepalive_timer (unsigned long data)
461{
462 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300463 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 struct tcp_sock *tp = tcp_sk(sk);
465 __u32 elapsed;
466
467 /* Only process if socket is not in use. */
468 bh_lock_sock(sk);
469 if (sock_owned_by_user(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900470 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700471 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 goto out;
473 }
474
475 if (sk->sk_state == TCP_LISTEN) {
476 tcp_synack_timer(sk);
477 goto out;
478 }
479
480 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
481 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700482 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 if (tmo > 0) {
485 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
486 goto out;
487 }
488 }
489 tcp_send_active_reset(sk, GFP_ATOMIC);
490 goto death;
491 }
492
493 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
494 goto out;
495
496 elapsed = keepalive_time_when(tp);
497
498 /* It is alive without keepalive 8) */
David S. Millerfe067e82007-03-07 12:12:44 -0800499 if (tp->packets_out || tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 goto resched;
501
502 elapsed = tcp_time_stamp - tp->rcv_tstamp;
503
504 if (elapsed >= keepalive_time_when(tp)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300505 if ((!tp->keepalive_probes && icsk->icsk_probes_out >= sysctl_tcp_keepalive_probes) ||
506 (tp->keepalive_probes && icsk->icsk_probes_out >= tp->keepalive_probes)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 tcp_send_active_reset(sk, GFP_ATOMIC);
508 tcp_write_err(sk);
509 goto out;
510 }
511 if (tcp_write_wakeup(sk) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300512 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 elapsed = keepalive_intvl_when(tp);
514 } else {
515 /* If keepalive was lost due to local congestion,
516 * try harder.
517 */
518 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
519 }
520 } else {
521 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
522 elapsed = keepalive_time_when(tp) - elapsed;
523 }
524
525 TCP_CHECK_TIMER(sk);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800526 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700529 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 goto out;
531
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900532death:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 tcp_done(sk);
534
535out:
536 bh_unlock_sock(sk);
537 sock_put(sk);
538}