Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Implementation of the Transmission Control Protocol(TCP). | |
7 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * 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> | |
5a0e3ad6 | 22 | #include <linux/gfp.h> |
1da177e4 LT |
23 | #include <net/tcp.h> |
24 | ||
ab32ea5d BH |
25 | int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES; |
26 | int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES; | |
27 | int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME; | |
28 | int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES; | |
29 | int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL; | |
30 | int sysctl_tcp_retries1 __read_mostly = TCP_RETR1; | |
31 | int sysctl_tcp_retries2 __read_mostly = TCP_RETR2; | |
32 | int sysctl_tcp_orphan_retries __read_mostly; | |
36e31b0a | 33 | int sysctl_tcp_thin_linear_timeouts __read_mostly; |
1da177e4 LT |
34 | |
35 | static void tcp_write_timer(unsigned long); | |
36 | static void tcp_delack_timer(unsigned long); | |
37 | static void tcp_keepalive_timer (unsigned long data); | |
38 | ||
463c84b9 ACM |
39 | void tcp_init_xmit_timers(struct sock *sk) |
40 | { | |
41 | inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer, | |
42 | &tcp_keepalive_timer); | |
43 | } | |
44 | ||
3f421baa ACM |
45 | EXPORT_SYMBOL(tcp_init_xmit_timers); |
46 | ||
1da177e4 LT |
47 | static void tcp_write_err(struct sock *sk) |
48 | { | |
49 | sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT; | |
50 | sk->sk_error_report(sk); | |
51 | ||
52 | tcp_done(sk); | |
de0744af | 53 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT); |
1da177e4 LT |
54 | } |
55 | ||
56 | /* Do not allow orphaned sockets to eat all our resources. | |
57 | * This is direct violation of TCP specs, but it is required | |
58 | * to prevent DoS attacks. It is called when a retransmission timeout | |
59 | * or zero probe timeout occurs on orphaned socket. | |
60 | * | |
caa20d9a | 61 | * Criteria is still not confirmed experimentally and may change. |
1da177e4 LT |
62 | * We kill the socket, if: |
63 | * 1. If number of orphaned sockets exceeds an administratively configured | |
64 | * limit. | |
65 | * 2. If we have strong memory pressure. | |
66 | */ | |
67 | static int tcp_out_of_resources(struct sock *sk, int do_reset) | |
68 | { | |
69 | struct tcp_sock *tp = tcp_sk(sk); | |
dd24c001 | 70 | int orphans = percpu_counter_read_positive(&tcp_orphan_count); |
1da177e4 | 71 | |
e905a9ed | 72 | /* If peer does not open window for long time, or did not transmit |
1da177e4 LT |
73 | * anything for long time, penalize it. */ |
74 | if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset) | |
75 | orphans <<= 1; | |
76 | ||
77 | /* If some dubious ICMP arrived, penalize even more. */ | |
78 | if (sk->sk_err_soft) | |
79 | orphans <<= 1; | |
80 | ||
e4fd5da3 | 81 | if (tcp_too_many_orphans(sk, orphans)) { |
1da177e4 LT |
82 | if (net_ratelimit()) |
83 | printk(KERN_INFO "Out of socket memory\n"); | |
84 | ||
85 | /* Catch exceptional cases, when connection requires reset. | |
86 | * 1. Last segment was sent recently. */ | |
87 | if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN || | |
88 | /* 2. Window is closed. */ | |
89 | (!tp->snd_wnd && !tp->packets_out)) | |
90 | do_reset = 1; | |
91 | if (do_reset) | |
92 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
93 | tcp_done(sk); | |
de0744af | 94 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
95 | return 1; |
96 | } | |
97 | return 0; | |
98 | } | |
99 | ||
100 | /* Calculate maximal number or retries on an orphaned socket. */ | |
101 | static int tcp_orphan_retries(struct sock *sk, int alive) | |
102 | { | |
103 | int retries = sysctl_tcp_orphan_retries; /* May be zero. */ | |
104 | ||
105 | /* We know from an ICMP that something is wrong. */ | |
106 | if (sk->sk_err_soft && !alive) | |
107 | retries = 0; | |
108 | ||
109 | /* However, if socket sent something recently, select some safe | |
110 | * number of retries. 8 corresponds to >100 seconds with minimal | |
111 | * RTO of 200msec. */ | |
112 | if (retries == 0 && alive) | |
113 | retries = 8; | |
114 | return retries; | |
115 | } | |
116 | ||
ce55dd36 ED |
117 | static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk) |
118 | { | |
ce55dd36 ED |
119 | /* Black hole detection */ |
120 | if (sysctl_tcp_mtu_probing) { | |
121 | if (!icsk->icsk_mtup.enabled) { | |
122 | icsk->icsk_mtup.enabled = 1; | |
123 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); | |
124 | } else { | |
125 | struct tcp_sock *tp = tcp_sk(sk); | |
829942c1 DM |
126 | int mss; |
127 | ||
8beb5c5f | 128 | mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1; |
ce55dd36 ED |
129 | mss = min(sysctl_tcp_base_mss, mss); |
130 | mss = max(mss, 68 - tp->tcp_header_len); | |
131 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); | |
132 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); | |
133 | } | |
134 | } | |
135 | } | |
136 | ||
2f7de571 | 137 | /* This function calculates a "timeout" which is equivalent to the timeout of a |
3ad2f3fb | 138 | * TCP connection after "boundary" unsuccessful, exponentially backed-off |
2f7de571 DL |
139 | * retransmissions with an initial RTO of TCP_RTO_MIN. |
140 | */ | |
141 | static bool retransmits_timed_out(struct sock *sk, | |
142 | unsigned int boundary) | |
143 | { | |
144 | unsigned int timeout, linear_backoff_thresh; | |
145 | unsigned int start_ts; | |
146 | ||
147 | if (!inet_csk(sk)->icsk_retransmits) | |
148 | return false; | |
149 | ||
150 | if (unlikely(!tcp_sk(sk)->retrans_stamp)) | |
151 | start_ts = TCP_SKB_CB(tcp_write_queue_head(sk))->when; | |
152 | else | |
153 | start_ts = tcp_sk(sk)->retrans_stamp; | |
154 | ||
155 | linear_backoff_thresh = ilog2(TCP_RTO_MAX/TCP_RTO_MIN); | |
156 | ||
157 | if (boundary <= linear_backoff_thresh) | |
158 | timeout = ((2 << boundary) - 1) * TCP_RTO_MIN; | |
159 | else | |
160 | timeout = ((2 << linear_backoff_thresh) - 1) * TCP_RTO_MIN + | |
161 | (boundary - linear_backoff_thresh) * TCP_RTO_MAX; | |
162 | ||
163 | return (tcp_time_stamp - start_ts) >= timeout; | |
164 | } | |
165 | ||
1da177e4 LT |
166 | /* A write timeout has occurred. Process the after effects. */ |
167 | static int tcp_write_timeout(struct sock *sk) | |
168 | { | |
5d424d5a | 169 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 170 | int retry_until; |
6fa12c85 | 171 | bool do_reset; |
1da177e4 LT |
172 | |
173 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
463c84b9 | 174 | if (icsk->icsk_retransmits) |
ea94ff3b | 175 | dst_negative_advice(&sk->sk_dst_cache, sk); |
463c84b9 | 176 | retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries; |
1da177e4 | 177 | } else { |
6fa12c85 | 178 | if (retransmits_timed_out(sk, sysctl_tcp_retries1)) { |
5d424d5a | 179 | /* Black hole detection */ |
ce55dd36 | 180 | tcp_mtu_probing(icsk, sk); |
1da177e4 | 181 | |
ea94ff3b | 182 | dst_negative_advice(&sk->sk_dst_cache, sk); |
1da177e4 LT |
183 | } |
184 | ||
185 | retry_until = sysctl_tcp_retries2; | |
186 | if (sock_flag(sk, SOCK_DEAD)) { | |
463c84b9 | 187 | const int alive = (icsk->icsk_rto < TCP_RTO_MAX); |
e905a9ed | 188 | |
1da177e4 | 189 | retry_until = tcp_orphan_retries(sk, alive); |
6fa12c85 DL |
190 | do_reset = alive || |
191 | !retransmits_timed_out(sk, retry_until); | |
1da177e4 | 192 | |
6fa12c85 | 193 | if (tcp_out_of_resources(sk, do_reset)) |
1da177e4 LT |
194 | return 1; |
195 | } | |
196 | } | |
197 | ||
6fa12c85 | 198 | if (retransmits_timed_out(sk, retry_until)) { |
1da177e4 LT |
199 | /* Has it gone just too far? */ |
200 | tcp_write_err(sk); | |
201 | return 1; | |
202 | } | |
203 | return 0; | |
204 | } | |
205 | ||
206 | static void tcp_delack_timer(unsigned long data) | |
207 | { | |
fd3f8c4c | 208 | struct sock *sk = (struct sock *)data; |
1da177e4 | 209 | struct tcp_sock *tp = tcp_sk(sk); |
463c84b9 | 210 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
211 | |
212 | bh_lock_sock(sk); | |
213 | if (sock_owned_by_user(sk)) { | |
214 | /* Try again later. */ | |
463c84b9 | 215 | icsk->icsk_ack.blocked = 1; |
de0744af | 216 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED); |
463c84b9 | 217 | sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN); |
1da177e4 LT |
218 | goto out_unlock; |
219 | } | |
220 | ||
9993e7d3 | 221 | sk_mem_reclaim_partial(sk); |
1da177e4 | 222 | |
463c84b9 | 223 | if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER)) |
1da177e4 LT |
224 | goto out; |
225 | ||
463c84b9 ACM |
226 | if (time_after(icsk->icsk_ack.timeout, jiffies)) { |
227 | sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout); | |
1da177e4 LT |
228 | goto out; |
229 | } | |
463c84b9 | 230 | icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER; |
1da177e4 | 231 | |
b03efcfb | 232 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
233 | struct sk_buff *skb; |
234 | ||
de0744af | 235 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED); |
1da177e4 LT |
236 | |
237 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 238 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
239 | |
240 | tp->ucopy.memory = 0; | |
241 | } | |
242 | ||
463c84b9 ACM |
243 | if (inet_csk_ack_scheduled(sk)) { |
244 | if (!icsk->icsk_ack.pingpong) { | |
1da177e4 | 245 | /* Delayed ACK missed: inflate ATO. */ |
463c84b9 | 246 | icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto); |
1da177e4 LT |
247 | } else { |
248 | /* Delayed ACK missed: leave pingpong mode and | |
249 | * deflate ATO. | |
250 | */ | |
463c84b9 ACM |
251 | icsk->icsk_ack.pingpong = 0; |
252 | icsk->icsk_ack.ato = TCP_ATO_MIN; | |
1da177e4 LT |
253 | } |
254 | tcp_send_ack(sk); | |
de0744af | 255 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS); |
1da177e4 LT |
256 | } |
257 | TCP_CHECK_TIMER(sk); | |
258 | ||
259 | out: | |
260 | if (tcp_memory_pressure) | |
3ab224be | 261 | sk_mem_reclaim(sk); |
1da177e4 LT |
262 | out_unlock: |
263 | bh_unlock_sock(sk); | |
264 | sock_put(sk); | |
265 | } | |
266 | ||
267 | static void tcp_probe_timer(struct sock *sk) | |
268 | { | |
6687e988 | 269 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
270 | struct tcp_sock *tp = tcp_sk(sk); |
271 | int max_probes; | |
272 | ||
fe067e8a | 273 | if (tp->packets_out || !tcp_send_head(sk)) { |
6687e988 | 274 | icsk->icsk_probes_out = 0; |
1da177e4 LT |
275 | return; |
276 | } | |
277 | ||
278 | /* *WARNING* RFC 1122 forbids this | |
279 | * | |
280 | * It doesn't AFAIK, because we kill the retransmit timer -AK | |
281 | * | |
282 | * FIXME: We ought not to do it, Solaris 2.5 actually has fixing | |
283 | * this behaviour in Solaris down as a bug fix. [AC] | |
284 | * | |
6687e988 | 285 | * Let me to explain. icsk_probes_out is zeroed by incoming ACKs |
1da177e4 LT |
286 | * even if they advertise zero window. Hence, connection is killed only |
287 | * if we received no ACKs for normal connection timeout. It is not killed | |
288 | * only because window stays zero for some time, window may be zero | |
289 | * until armageddon and even later. We are in full accordance | |
290 | * with RFCs, only probe timer combines both retransmission timeout | |
291 | * and probe timeout in one bottle. --ANK | |
292 | */ | |
293 | max_probes = sysctl_tcp_retries2; | |
294 | ||
295 | if (sock_flag(sk, SOCK_DEAD)) { | |
463c84b9 | 296 | const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX); |
e905a9ed | 297 | |
1da177e4 LT |
298 | max_probes = tcp_orphan_retries(sk, alive); |
299 | ||
6687e988 | 300 | if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes)) |
1da177e4 LT |
301 | return; |
302 | } | |
303 | ||
6687e988 | 304 | if (icsk->icsk_probes_out > max_probes) { |
1da177e4 LT |
305 | tcp_write_err(sk); |
306 | } else { | |
307 | /* Only send another probe if we didn't close things up. */ | |
308 | tcp_send_probe0(sk); | |
309 | } | |
310 | } | |
311 | ||
312 | /* | |
313 | * The TCP retransmit timer. | |
314 | */ | |
315 | ||
f1ecd5d9 | 316 | void tcp_retransmit_timer(struct sock *sk) |
1da177e4 LT |
317 | { |
318 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 319 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
320 | |
321 | if (!tp->packets_out) | |
322 | goto out; | |
323 | ||
547b792c | 324 | WARN_ON(tcp_write_queue_empty(sk)); |
1da177e4 LT |
325 | |
326 | if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) && | |
327 | !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) { | |
328 | /* Receiver dastardly shrinks window. Our retransmits | |
329 | * become zero probes, but we should not timeout this | |
330 | * connection. If the socket is an orphan, time it out, | |
331 | * we cannot allow such beasts to hang infinitely. | |
332 | */ | |
333 | #ifdef TCP_DEBUG | |
569508c9 YH |
334 | struct inet_sock *inet = inet_sk(sk); |
335 | if (sk->sk_family == AF_INET) { | |
6086ebca | 336 | LIMIT_NETDEBUG(KERN_DEBUG "TCP: Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
c720c7e8 ED |
337 | &inet->inet_daddr, ntohs(inet->inet_dport), |
338 | inet->inet_num, tp->snd_una, tp->snd_nxt); | |
1da177e4 | 339 | } |
569508c9 YH |
340 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
341 | else if (sk->sk_family == AF_INET6) { | |
342 | struct ipv6_pinfo *np = inet6_sk(sk); | |
6086ebca | 343 | LIMIT_NETDEBUG(KERN_DEBUG "TCP: Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
c720c7e8 ED |
344 | &np->daddr, ntohs(inet->inet_dport), |
345 | inet->inet_num, tp->snd_una, tp->snd_nxt); | |
569508c9 YH |
346 | } |
347 | #endif | |
1da177e4 LT |
348 | #endif |
349 | if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) { | |
350 | tcp_write_err(sk); | |
351 | goto out; | |
352 | } | |
353 | tcp_enter_loss(sk, 0); | |
fe067e8a | 354 | tcp_retransmit_skb(sk, tcp_write_queue_head(sk)); |
1da177e4 LT |
355 | __sk_dst_reset(sk); |
356 | goto out_reset_timer; | |
357 | } | |
358 | ||
359 | if (tcp_write_timeout(sk)) | |
360 | goto out; | |
361 | ||
463c84b9 | 362 | if (icsk->icsk_retransmits == 0) { |
40b215e5 PE |
363 | int mib_idx; |
364 | ||
bc079e9e IJ |
365 | if (icsk->icsk_ca_state == TCP_CA_Disorder) { |
366 | if (tcp_is_sack(tp)) | |
367 | mib_idx = LINUX_MIB_TCPSACKFAILURES; | |
368 | else | |
369 | mib_idx = LINUX_MIB_TCPRENOFAILURES; | |
370 | } else if (icsk->icsk_ca_state == TCP_CA_Recovery) { | |
371 | if (tcp_is_sack(tp)) | |
372 | mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL; | |
373 | else | |
374 | mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL; | |
6687e988 | 375 | } else if (icsk->icsk_ca_state == TCP_CA_Loss) { |
40b215e5 | 376 | mib_idx = LINUX_MIB_TCPLOSSFAILURES; |
1da177e4 | 377 | } else { |
40b215e5 | 378 | mib_idx = LINUX_MIB_TCPTIMEOUTS; |
1da177e4 | 379 | } |
de0744af | 380 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
1da177e4 LT |
381 | } |
382 | ||
383 | if (tcp_use_frto(sk)) { | |
384 | tcp_enter_frto(sk); | |
385 | } else { | |
386 | tcp_enter_loss(sk, 0); | |
387 | } | |
388 | ||
fe067e8a | 389 | if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) { |
1da177e4 LT |
390 | /* Retransmission failed because of local congestion, |
391 | * do not backoff. | |
392 | */ | |
463c84b9 ACM |
393 | if (!icsk->icsk_retransmits) |
394 | icsk->icsk_retransmits = 1; | |
395 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
3f421baa ACM |
396 | min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL), |
397 | TCP_RTO_MAX); | |
1da177e4 LT |
398 | goto out; |
399 | } | |
400 | ||
401 | /* Increase the timeout each time we retransmit. Note that | |
402 | * we do not increase the rtt estimate. rto is initialized | |
403 | * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests | |
404 | * that doubling rto each time is the least we can get away with. | |
405 | * In KA9Q, Karn uses this for the first few times, and then | |
406 | * goes to quadratic. netBSD doubles, but only goes up to *64, | |
407 | * and clamps at 1 to 64 sec afterwards. Note that 120 sec is | |
408 | * defined in the protocol as the maximum possible RTT. I guess | |
409 | * we'll have to use something other than TCP to talk to the | |
410 | * University of Mars. | |
411 | * | |
412 | * PAWS allows us longer timeouts and large windows, so once | |
413 | * implemented ftp to mars will work nicely. We will have to fix | |
414 | * the 120 second clamps though! | |
415 | */ | |
463c84b9 ACM |
416 | icsk->icsk_backoff++; |
417 | icsk->icsk_retransmits++; | |
1da177e4 LT |
418 | |
419 | out_reset_timer: | |
36e31b0a AP |
420 | /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is |
421 | * used to reset timer, set to 0. Recalculate 'icsk_rto' as this | |
422 | * might be increased if the stream oscillates between thin and thick, | |
423 | * thus the old value might already be too high compared to the value | |
424 | * set by 'tcp_set_rto' in tcp_input.c which resets the rto without | |
425 | * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating | |
426 | * exponential backoff behaviour to avoid continue hammering | |
427 | * linear-timeout retransmissions into a black hole | |
428 | */ | |
429 | if (sk->sk_state == TCP_ESTABLISHED && | |
430 | (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) && | |
431 | tcp_stream_is_thin(tp) && | |
432 | icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) { | |
433 | icsk->icsk_backoff = 0; | |
434 | icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX); | |
435 | } else { | |
436 | /* Use normal (exponential) backoff */ | |
437 | icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX); | |
438 | } | |
3f421baa | 439 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX); |
6fa12c85 | 440 | if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1)) |
1da177e4 LT |
441 | __sk_dst_reset(sk); |
442 | ||
443 | out:; | |
444 | } | |
445 | ||
446 | static void tcp_write_timer(unsigned long data) | |
447 | { | |
fd3f8c4c | 448 | struct sock *sk = (struct sock *)data; |
463c84b9 | 449 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
450 | int event; |
451 | ||
452 | bh_lock_sock(sk); | |
453 | if (sock_owned_by_user(sk)) { | |
454 | /* Try again later */ | |
463c84b9 | 455 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20)); |
1da177e4 LT |
456 | goto out_unlock; |
457 | } | |
458 | ||
463c84b9 | 459 | if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending) |
1da177e4 LT |
460 | goto out; |
461 | ||
463c84b9 ACM |
462 | if (time_after(icsk->icsk_timeout, jiffies)) { |
463 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout); | |
1da177e4 LT |
464 | goto out; |
465 | } | |
466 | ||
463c84b9 ACM |
467 | event = icsk->icsk_pending; |
468 | icsk->icsk_pending = 0; | |
1da177e4 LT |
469 | |
470 | switch (event) { | |
463c84b9 | 471 | case ICSK_TIME_RETRANS: |
1da177e4 LT |
472 | tcp_retransmit_timer(sk); |
473 | break; | |
463c84b9 | 474 | case ICSK_TIME_PROBE0: |
1da177e4 LT |
475 | tcp_probe_timer(sk); |
476 | break; | |
477 | } | |
478 | TCP_CHECK_TIMER(sk); | |
479 | ||
480 | out: | |
3ab224be | 481 | sk_mem_reclaim(sk); |
1da177e4 LT |
482 | out_unlock: |
483 | bh_unlock_sock(sk); | |
484 | sock_put(sk); | |
485 | } | |
486 | ||
295f7324 ACM |
487 | /* |
488 | * Timer for listening sockets | |
489 | */ | |
490 | ||
491 | static void tcp_synack_timer(struct sock *sk) | |
492 | { | |
a019d6fe ACM |
493 | inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL, |
494 | TCP_TIMEOUT_INIT, TCP_RTO_MAX); | |
1da177e4 LT |
495 | } |
496 | ||
72659ecc OP |
497 | void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req) |
498 | { | |
499 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS); | |
500 | } | |
501 | EXPORT_SYMBOL(tcp_syn_ack_timeout); | |
502 | ||
1da177e4 LT |
503 | void tcp_set_keepalive(struct sock *sk, int val) |
504 | { | |
505 | if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) | |
506 | return; | |
507 | ||
508 | if (val && !sock_flag(sk, SOCK_KEEPOPEN)) | |
463c84b9 | 509 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk))); |
1da177e4 | 510 | else if (!val) |
463c84b9 | 511 | inet_csk_delete_keepalive_timer(sk); |
1da177e4 LT |
512 | } |
513 | ||
514 | ||
515 | static void tcp_keepalive_timer (unsigned long data) | |
516 | { | |
517 | struct sock *sk = (struct sock *) data; | |
6687e988 | 518 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
519 | struct tcp_sock *tp = tcp_sk(sk); |
520 | __u32 elapsed; | |
521 | ||
522 | /* Only process if socket is not in use. */ | |
523 | bh_lock_sock(sk); | |
524 | if (sock_owned_by_user(sk)) { | |
e905a9ed | 525 | /* Try again later. */ |
463c84b9 | 526 | inet_csk_reset_keepalive_timer (sk, HZ/20); |
1da177e4 LT |
527 | goto out; |
528 | } | |
529 | ||
530 | if (sk->sk_state == TCP_LISTEN) { | |
531 | tcp_synack_timer(sk); | |
532 | goto out; | |
533 | } | |
534 | ||
535 | if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) { | |
536 | if (tp->linger2 >= 0) { | |
463c84b9 | 537 | const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN; |
1da177e4 LT |
538 | |
539 | if (tmo > 0) { | |
540 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); | |
541 | goto out; | |
542 | } | |
543 | } | |
544 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
545 | goto death; | |
546 | } | |
547 | ||
548 | if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE) | |
549 | goto out; | |
550 | ||
551 | elapsed = keepalive_time_when(tp); | |
552 | ||
553 | /* It is alive without keepalive 8) */ | |
fe067e8a | 554 | if (tp->packets_out || tcp_send_head(sk)) |
1da177e4 LT |
555 | goto resched; |
556 | ||
557 | elapsed = tcp_time_stamp - tp->rcv_tstamp; | |
558 | ||
559 | if (elapsed >= keepalive_time_when(tp)) { | |
df19a626 | 560 | if (icsk->icsk_probes_out >= keepalive_probes(tp)) { |
1da177e4 LT |
561 | tcp_send_active_reset(sk, GFP_ATOMIC); |
562 | tcp_write_err(sk); | |
563 | goto out; | |
564 | } | |
565 | if (tcp_write_wakeup(sk) <= 0) { | |
6687e988 | 566 | icsk->icsk_probes_out++; |
1da177e4 LT |
567 | elapsed = keepalive_intvl_when(tp); |
568 | } else { | |
569 | /* If keepalive was lost due to local congestion, | |
570 | * try harder. | |
571 | */ | |
572 | elapsed = TCP_RESOURCE_PROBE_INTERVAL; | |
573 | } | |
574 | } else { | |
575 | /* It is tp->rcv_tstamp + keepalive_time_when(tp) */ | |
576 | elapsed = keepalive_time_when(tp) - elapsed; | |
577 | } | |
578 | ||
579 | TCP_CHECK_TIMER(sk); | |
3ab224be | 580 | sk_mem_reclaim(sk); |
1da177e4 LT |
581 | |
582 | resched: | |
463c84b9 | 583 | inet_csk_reset_keepalive_timer (sk, elapsed); |
1da177e4 LT |
584 | goto out; |
585 | ||
e905a9ed | 586 | death: |
1da177e4 LT |
587 | tcp_done(sk); |
588 | ||
589 | out: | |
590 | bh_unlock_sock(sk); | |
591 | sock_put(sk); | |
592 | } |