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> | |
22 | #include <net/tcp.h> | |
23 | ||
ab32ea5d BH |
24 | int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES; |
25 | int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES; | |
26 | int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME; | |
27 | int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES; | |
28 | int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL; | |
29 | int sysctl_tcp_retries1 __read_mostly = TCP_RETR1; | |
30 | int sysctl_tcp_retries2 __read_mostly = TCP_RETR2; | |
31 | int sysctl_tcp_orphan_retries __read_mostly; | |
1da177e4 LT |
32 | |
33 | static void tcp_write_timer(unsigned long); | |
34 | static void tcp_delack_timer(unsigned long); | |
35 | static void tcp_keepalive_timer (unsigned long data); | |
36 | ||
463c84b9 ACM |
37 | void tcp_init_xmit_timers(struct sock *sk) |
38 | { | |
39 | inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer, | |
40 | &tcp_keepalive_timer); | |
41 | } | |
42 | ||
3f421baa ACM |
43 | EXPORT_SYMBOL(tcp_init_xmit_timers); |
44 | ||
1da177e4 LT |
45 | static 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); | |
de0744af | 51 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT); |
1da177e4 LT |
52 | } |
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 | * | |
caa20d9a | 59 | * Criteria is still not confirmed experimentally and may change. |
1da177e4 LT |
60 | * 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 | */ | |
65 | static 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 | ||
e905a9ed | 70 | /* If peer does not open window for long time, or did not transmit |
1da177e4 LT |
71 | * 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 | ||
e4fd5da3 | 79 | if (tcp_too_many_orphans(sk, orphans)) { |
1da177e4 LT |
80 | 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); | |
de0744af | 92 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
93 | return 1; |
94 | } | |
95 | return 0; | |
96 | } | |
97 | ||
98 | /* Calculate maximal number or retries on an orphaned socket. */ | |
99 | static 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 | ||
ce55dd36 ED |
115 | static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk) |
116 | { | |
ce55dd36 ED |
117 | /* 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); | |
829942c1 DM |
124 | int mss; |
125 | ||
8beb5c5f | 126 | mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1; |
ce55dd36 ED |
127 | 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 | ||
1da177e4 LT |
135 | /* A write timeout has occurred. Process the after effects. */ |
136 | static int tcp_write_timeout(struct sock *sk) | |
137 | { | |
5d424d5a | 138 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
139 | int retry_until; |
140 | ||
141 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
463c84b9 | 142 | if (icsk->icsk_retransmits) |
1da177e4 | 143 | dst_negative_advice(&sk->sk_dst_cache); |
463c84b9 | 144 | retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries; |
1da177e4 | 145 | } else { |
463c84b9 | 146 | if (icsk->icsk_retransmits >= sysctl_tcp_retries1) { |
5d424d5a | 147 | /* Black hole detection */ |
ce55dd36 | 148 | tcp_mtu_probing(icsk, sk); |
1da177e4 LT |
149 | |
150 | dst_negative_advice(&sk->sk_dst_cache); | |
151 | } | |
152 | ||
153 | retry_until = sysctl_tcp_retries2; | |
154 | if (sock_flag(sk, SOCK_DEAD)) { | |
463c84b9 | 155 | const int alive = (icsk->icsk_rto < TCP_RTO_MAX); |
e905a9ed | 156 | |
1da177e4 LT |
157 | retry_until = tcp_orphan_retries(sk, alive); |
158 | ||
463c84b9 | 159 | if (tcp_out_of_resources(sk, alive || icsk->icsk_retransmits < retry_until)) |
1da177e4 LT |
160 | return 1; |
161 | } | |
162 | } | |
163 | ||
463c84b9 | 164 | if (icsk->icsk_retransmits >= retry_until) { |
1da177e4 LT |
165 | /* Has it gone just too far? */ |
166 | tcp_write_err(sk); | |
167 | return 1; | |
168 | } | |
169 | return 0; | |
170 | } | |
171 | ||
172 | static void tcp_delack_timer(unsigned long data) | |
173 | { | |
174 | struct sock *sk = (struct sock*)data; | |
175 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 176 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
177 | |
178 | bh_lock_sock(sk); | |
179 | if (sock_owned_by_user(sk)) { | |
180 | /* Try again later. */ | |
463c84b9 | 181 | icsk->icsk_ack.blocked = 1; |
de0744af | 182 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED); |
463c84b9 | 183 | sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN); |
1da177e4 LT |
184 | goto out_unlock; |
185 | } | |
186 | ||
9993e7d3 | 187 | sk_mem_reclaim_partial(sk); |
1da177e4 | 188 | |
463c84b9 | 189 | if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER)) |
1da177e4 LT |
190 | goto out; |
191 | ||
463c84b9 ACM |
192 | if (time_after(icsk->icsk_ack.timeout, jiffies)) { |
193 | sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout); | |
1da177e4 LT |
194 | goto out; |
195 | } | |
463c84b9 | 196 | icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER; |
1da177e4 | 197 | |
b03efcfb | 198 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
199 | struct sk_buff *skb; |
200 | ||
de0744af | 201 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED); |
1da177e4 LT |
202 | |
203 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 204 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
205 | |
206 | tp->ucopy.memory = 0; | |
207 | } | |
208 | ||
463c84b9 ACM |
209 | if (inet_csk_ack_scheduled(sk)) { |
210 | if (!icsk->icsk_ack.pingpong) { | |
1da177e4 | 211 | /* Delayed ACK missed: inflate ATO. */ |
463c84b9 | 212 | icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto); |
1da177e4 LT |
213 | } else { |
214 | /* Delayed ACK missed: leave pingpong mode and | |
215 | * deflate ATO. | |
216 | */ | |
463c84b9 ACM |
217 | icsk->icsk_ack.pingpong = 0; |
218 | icsk->icsk_ack.ato = TCP_ATO_MIN; | |
1da177e4 LT |
219 | } |
220 | tcp_send_ack(sk); | |
de0744af | 221 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS); |
1da177e4 LT |
222 | } |
223 | TCP_CHECK_TIMER(sk); | |
224 | ||
225 | out: | |
226 | if (tcp_memory_pressure) | |
3ab224be | 227 | sk_mem_reclaim(sk); |
1da177e4 LT |
228 | out_unlock: |
229 | bh_unlock_sock(sk); | |
230 | sock_put(sk); | |
231 | } | |
232 | ||
233 | static void tcp_probe_timer(struct sock *sk) | |
234 | { | |
6687e988 | 235 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
236 | struct tcp_sock *tp = tcp_sk(sk); |
237 | int max_probes; | |
238 | ||
fe067e8a | 239 | if (tp->packets_out || !tcp_send_head(sk)) { |
6687e988 | 240 | icsk->icsk_probes_out = 0; |
1da177e4 LT |
241 | 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 | * | |
6687e988 | 251 | * Let me to explain. icsk_probes_out is zeroed by incoming ACKs |
1da177e4 LT |
252 | * 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)) { | |
463c84b9 | 262 | const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX); |
e905a9ed | 263 | |
1da177e4 LT |
264 | max_probes = tcp_orphan_retries(sk, alive); |
265 | ||
6687e988 | 266 | if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes)) |
1da177e4 LT |
267 | return; |
268 | } | |
269 | ||
6687e988 | 270 | if (icsk->icsk_probes_out > max_probes) { |
1da177e4 LT |
271 | 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 | ||
282 | static void tcp_retransmit_timer(struct sock *sk) | |
283 | { | |
284 | struct tcp_sock *tp = tcp_sk(sk); | |
463c84b9 | 285 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
286 | |
287 | if (!tp->packets_out) | |
288 | goto out; | |
289 | ||
547b792c | 290 | WARN_ON(tcp_write_queue_empty(sk)); |
1da177e4 LT |
291 | |
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 | |
569508c9 YH |
300 | struct inet_sock *inet = inet_sk(sk); |
301 | if (sk->sk_family == AF_INET) { | |
a7d632b6 | 302 | LIMIT_NETDEBUG(KERN_DEBUG "TCP: Treason uncloaked! Peer " NIPQUAD_FMT ":%u/%u shrinks window %u:%u. Repaired.\n", |
714e85be | 303 | NIPQUAD(inet->daddr), ntohs(inet->dport), |
1da177e4 LT |
304 | inet->num, tp->snd_una, tp->snd_nxt); |
305 | } | |
569508c9 YH |
306 | #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 | |
1da177e4 LT |
314 | #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); | |
fe067e8a | 320 | tcp_retransmit_skb(sk, tcp_write_queue_head(sk)); |
1da177e4 LT |
321 | __sk_dst_reset(sk); |
322 | goto out_reset_timer; | |
323 | } | |
324 | ||
325 | if (tcp_write_timeout(sk)) | |
326 | goto out; | |
327 | ||
463c84b9 | 328 | if (icsk->icsk_retransmits == 0) { |
40b215e5 PE |
329 | int mib_idx; |
330 | ||
6687e988 ACM |
331 | if (icsk->icsk_ca_state == TCP_CA_Disorder || |
332 | icsk->icsk_ca_state == TCP_CA_Recovery) { | |
e60402d0 | 333 | if (tcp_is_sack(tp)) { |
6687e988 | 334 | if (icsk->icsk_ca_state == TCP_CA_Recovery) |
40b215e5 | 335 | mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL; |
1da177e4 | 336 | else |
40b215e5 | 337 | mib_idx = LINUX_MIB_TCPSACKFAILURES; |
1da177e4 | 338 | } else { |
6687e988 | 339 | if (icsk->icsk_ca_state == TCP_CA_Recovery) |
40b215e5 | 340 | mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL; |
1da177e4 | 341 | else |
40b215e5 | 342 | mib_idx = LINUX_MIB_TCPRENOFAILURES; |
1da177e4 | 343 | } |
6687e988 | 344 | } else if (icsk->icsk_ca_state == TCP_CA_Loss) { |
40b215e5 | 345 | mib_idx = LINUX_MIB_TCPLOSSFAILURES; |
1da177e4 | 346 | } else { |
40b215e5 | 347 | mib_idx = LINUX_MIB_TCPTIMEOUTS; |
1da177e4 | 348 | } |
de0744af | 349 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
1da177e4 LT |
350 | } |
351 | ||
352 | if (tcp_use_frto(sk)) { | |
353 | tcp_enter_frto(sk); | |
354 | } else { | |
355 | tcp_enter_loss(sk, 0); | |
356 | } | |
357 | ||
fe067e8a | 358 | if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) { |
1da177e4 LT |
359 | /* Retransmission failed because of local congestion, |
360 | * do not backoff. | |
361 | */ | |
463c84b9 ACM |
362 | if (!icsk->icsk_retransmits) |
363 | icsk->icsk_retransmits = 1; | |
364 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
3f421baa ACM |
365 | min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL), |
366 | TCP_RTO_MAX); | |
1da177e4 LT |
367 | 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 | */ | |
463c84b9 ACM |
385 | icsk->icsk_backoff++; |
386 | icsk->icsk_retransmits++; | |
1da177e4 LT |
387 | |
388 | out_reset_timer: | |
463c84b9 | 389 | icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX); |
3f421baa | 390 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX); |
463c84b9 | 391 | if (icsk->icsk_retransmits > sysctl_tcp_retries1) |
1da177e4 LT |
392 | __sk_dst_reset(sk); |
393 | ||
394 | out:; | |
395 | } | |
396 | ||
397 | static void tcp_write_timer(unsigned long data) | |
398 | { | |
399 | struct sock *sk = (struct sock*)data; | |
463c84b9 | 400 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
401 | int event; |
402 | ||
403 | bh_lock_sock(sk); | |
404 | if (sock_owned_by_user(sk)) { | |
405 | /* Try again later */ | |
463c84b9 | 406 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20)); |
1da177e4 LT |
407 | goto out_unlock; |
408 | } | |
409 | ||
463c84b9 | 410 | if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending) |
1da177e4 LT |
411 | goto out; |
412 | ||
463c84b9 ACM |
413 | if (time_after(icsk->icsk_timeout, jiffies)) { |
414 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout); | |
1da177e4 LT |
415 | goto out; |
416 | } | |
417 | ||
463c84b9 ACM |
418 | event = icsk->icsk_pending; |
419 | icsk->icsk_pending = 0; | |
1da177e4 LT |
420 | |
421 | switch (event) { | |
463c84b9 | 422 | case ICSK_TIME_RETRANS: |
1da177e4 LT |
423 | tcp_retransmit_timer(sk); |
424 | break; | |
463c84b9 | 425 | case ICSK_TIME_PROBE0: |
1da177e4 LT |
426 | tcp_probe_timer(sk); |
427 | break; | |
428 | } | |
429 | TCP_CHECK_TIMER(sk); | |
430 | ||
431 | out: | |
3ab224be | 432 | sk_mem_reclaim(sk); |
1da177e4 LT |
433 | out_unlock: |
434 | bh_unlock_sock(sk); | |
435 | sock_put(sk); | |
436 | } | |
437 | ||
295f7324 ACM |
438 | /* |
439 | * Timer for listening sockets | |
440 | */ | |
441 | ||
442 | static void tcp_synack_timer(struct sock *sk) | |
443 | { | |
a019d6fe ACM |
444 | inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL, |
445 | TCP_TIMEOUT_INIT, TCP_RTO_MAX); | |
1da177e4 LT |
446 | } |
447 | ||
1da177e4 LT |
448 | void 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)) | |
463c84b9 | 454 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk))); |
1da177e4 | 455 | else if (!val) |
463c84b9 | 456 | inet_csk_delete_keepalive_timer(sk); |
1da177e4 LT |
457 | } |
458 | ||
459 | ||
460 | static void tcp_keepalive_timer (unsigned long data) | |
461 | { | |
462 | struct sock *sk = (struct sock *) data; | |
6687e988 | 463 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
464 | 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)) { | |
e905a9ed | 470 | /* Try again later. */ |
463c84b9 | 471 | inet_csk_reset_keepalive_timer (sk, HZ/20); |
1da177e4 LT |
472 | 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) { | |
463c84b9 | 482 | const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN; |
1da177e4 LT |
483 | |
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) */ | |
fe067e8a | 499 | if (tp->packets_out || tcp_send_head(sk)) |
1da177e4 LT |
500 | goto resched; |
501 | ||
502 | elapsed = tcp_time_stamp - tp->rcv_tstamp; | |
503 | ||
504 | if (elapsed >= keepalive_time_when(tp)) { | |
6687e988 ACM |
505 | if ((!tp->keepalive_probes && icsk->icsk_probes_out >= sysctl_tcp_keepalive_probes) || |
506 | (tp->keepalive_probes && icsk->icsk_probes_out >= tp->keepalive_probes)) { | |
1da177e4 LT |
507 | tcp_send_active_reset(sk, GFP_ATOMIC); |
508 | tcp_write_err(sk); | |
509 | goto out; | |
510 | } | |
511 | if (tcp_write_wakeup(sk) <= 0) { | |
6687e988 | 512 | icsk->icsk_probes_out++; |
1da177e4 LT |
513 | 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); | |
3ab224be | 526 | sk_mem_reclaim(sk); |
1da177e4 LT |
527 | |
528 | resched: | |
463c84b9 | 529 | inet_csk_reset_keepalive_timer (sk, elapsed); |
1da177e4 LT |
530 | goto out; |
531 | ||
e905a9ed | 532 | death: |
1da177e4 LT |
533 | tcp_done(sk); |
534 | ||
535 | out: | |
536 | bh_unlock_sock(sk); | |
537 | sock_put(sk); | |
538 | } |