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 | ||
36e31b0a | 25 | int sysctl_tcp_thin_linear_timeouts __read_mostly; |
1da177e4 | 26 | |
1da177e4 LT |
27 | static void tcp_write_err(struct sock *sk) |
28 | { | |
29 | sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT; | |
30 | sk->sk_error_report(sk); | |
31 | ||
32 | tcp_done(sk); | |
de0744af | 33 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT); |
1da177e4 LT |
34 | } |
35 | ||
36 | /* Do not allow orphaned sockets to eat all our resources. | |
37 | * This is direct violation of TCP specs, but it is required | |
38 | * to prevent DoS attacks. It is called when a retransmission timeout | |
39 | * or zero probe timeout occurs on orphaned socket. | |
40 | * | |
caa20d9a | 41 | * Criteria is still not confirmed experimentally and may change. |
1da177e4 LT |
42 | * We kill the socket, if: |
43 | * 1. If number of orphaned sockets exceeds an administratively configured | |
44 | * limit. | |
45 | * 2. If we have strong memory pressure. | |
46 | */ | |
b248230c | 47 | static int tcp_out_of_resources(struct sock *sk, bool do_reset) |
1da177e4 LT |
48 | { |
49 | struct tcp_sock *tp = tcp_sk(sk); | |
ad1af0fe | 50 | int shift = 0; |
1da177e4 | 51 | |
e905a9ed | 52 | /* If peer does not open window for long time, or did not transmit |
1da177e4 LT |
53 | * anything for long time, penalize it. */ |
54 | if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset) | |
ad1af0fe | 55 | shift++; |
1da177e4 LT |
56 | |
57 | /* If some dubious ICMP arrived, penalize even more. */ | |
58 | if (sk->sk_err_soft) | |
ad1af0fe | 59 | shift++; |
1da177e4 | 60 | |
efcdbf24 | 61 | if (tcp_check_oom(sk, shift)) { |
1da177e4 LT |
62 | /* Catch exceptional cases, when connection requires reset. |
63 | * 1. Last segment was sent recently. */ | |
64 | if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN || | |
65 | /* 2. Window is closed. */ | |
66 | (!tp->snd_wnd && !tp->packets_out)) | |
b248230c | 67 | do_reset = true; |
1da177e4 LT |
68 | if (do_reset) |
69 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
70 | tcp_done(sk); | |
de0744af | 71 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
72 | return 1; |
73 | } | |
74 | return 0; | |
75 | } | |
76 | ||
77 | /* Calculate maximal number or retries on an orphaned socket. */ | |
7533ce30 | 78 | static int tcp_orphan_retries(struct sock *sk, bool alive) |
1da177e4 | 79 | { |
c402d9be | 80 | int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */ |
1da177e4 LT |
81 | |
82 | /* We know from an ICMP that something is wrong. */ | |
83 | if (sk->sk_err_soft && !alive) | |
84 | retries = 0; | |
85 | ||
86 | /* However, if socket sent something recently, select some safe | |
87 | * number of retries. 8 corresponds to >100 seconds with minimal | |
88 | * RTO of 200msec. */ | |
89 | if (retries == 0 && alive) | |
90 | retries = 8; | |
91 | return retries; | |
92 | } | |
93 | ||
ce55dd36 ED |
94 | static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk) |
95 | { | |
b0f9ca53 FD |
96 | struct net *net = sock_net(sk); |
97 | ||
ce55dd36 | 98 | /* Black hole detection */ |
b0f9ca53 | 99 | if (net->ipv4.sysctl_tcp_mtu_probing) { |
ce55dd36 ED |
100 | if (!icsk->icsk_mtup.enabled) { |
101 | icsk->icsk_mtup.enabled = 1; | |
05cbc0db | 102 | icsk->icsk_mtup.probe_timestamp = tcp_time_stamp; |
ce55dd36 ED |
103 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
104 | } else { | |
b0f9ca53 | 105 | struct net *net = sock_net(sk); |
ce55dd36 | 106 | struct tcp_sock *tp = tcp_sk(sk); |
829942c1 DM |
107 | int mss; |
108 | ||
8beb5c5f | 109 | mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1; |
b0f9ca53 | 110 | mss = min(net->ipv4.sysctl_tcp_base_mss, mss); |
ce55dd36 ED |
111 | mss = max(mss, 68 - tp->tcp_header_len); |
112 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); | |
113 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); | |
114 | } | |
115 | } | |
116 | } | |
117 | ||
2f7de571 | 118 | /* This function calculates a "timeout" which is equivalent to the timeout of a |
3ad2f3fb | 119 | * TCP connection after "boundary" unsuccessful, exponentially backed-off |
4d22f7d3 DL |
120 | * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if |
121 | * syn_set flag is set. | |
2f7de571 DL |
122 | */ |
123 | static bool retransmits_timed_out(struct sock *sk, | |
dca43c75 | 124 | unsigned int boundary, |
21a180cd | 125 | unsigned int timeout, |
4d22f7d3 | 126 | bool syn_set) |
2f7de571 | 127 | { |
dca43c75 | 128 | unsigned int linear_backoff_thresh, start_ts; |
4d22f7d3 | 129 | unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN; |
2f7de571 DL |
130 | |
131 | if (!inet_csk(sk)->icsk_retransmits) | |
132 | return false; | |
133 | ||
7faee5c0 ED |
134 | start_ts = tcp_sk(sk)->retrans_stamp; |
135 | if (unlikely(!start_ts)) | |
136 | start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk)); | |
2f7de571 | 137 | |
dca43c75 | 138 | if (likely(timeout == 0)) { |
21a180cd | 139 | linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base); |
2f7de571 | 140 | |
dca43c75 | 141 | if (boundary <= linear_backoff_thresh) |
21a180cd | 142 | timeout = ((2 << boundary) - 1) * rto_base; |
dca43c75 | 143 | else |
21a180cd | 144 | timeout = ((2 << linear_backoff_thresh) - 1) * rto_base + |
dca43c75 JC |
145 | (boundary - linear_backoff_thresh) * TCP_RTO_MAX; |
146 | } | |
2f7de571 DL |
147 | return (tcp_time_stamp - start_ts) >= timeout; |
148 | } | |
149 | ||
1da177e4 LT |
150 | /* A write timeout has occurred. Process the after effects. */ |
151 | static int tcp_write_timeout(struct sock *sk) | |
152 | { | |
5d424d5a | 153 | struct inet_connection_sock *icsk = inet_csk(sk); |
c968601d | 154 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 155 | struct net *net = sock_net(sk); |
1da177e4 | 156 | int retry_until; |
3db1cd5c | 157 | bool do_reset, syn_set = false; |
1da177e4 LT |
158 | |
159 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
c968601d | 160 | if (icsk->icsk_retransmits) { |
b6c6712a | 161 | dst_negative_advice(sk); |
c968601d | 162 | if (tp->syn_fastopen || tp->syn_data) |
2646c831 | 163 | tcp_fastopen_cache_set(sk, 0, NULL, true, 0); |
dd52bc2b | 164 | if (tp->syn_data && icsk->icsk_retransmits == 1) |
f19c29e3 YC |
165 | NET_INC_STATS_BH(sock_net(sk), |
166 | LINUX_MIB_TCPFASTOPENACTIVEFAIL); | |
c968601d | 167 | } |
6fa25166 | 168 | retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
3db1cd5c | 169 | syn_set = true; |
1da177e4 | 170 | } else { |
ae5c3f40 | 171 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0, 0)) { |
0e45f4da YC |
172 | /* Some middle-boxes may black-hole Fast Open _after_ |
173 | * the handshake. Therefore we conservatively disable | |
174 | * Fast Open on this path on recurring timeouts with | |
175 | * few or zero bytes acked after Fast Open. | |
176 | */ | |
177 | if (tp->syn_data_acked && | |
178 | tp->bytes_acked <= tp->rx_opt.mss_clamp) { | |
179 | tcp_fastopen_cache_set(sk, 0, NULL, true, 0); | |
ae5c3f40 | 180 | if (icsk->icsk_retransmits == net->ipv4.sysctl_tcp_retries1) |
0e45f4da YC |
181 | NET_INC_STATS_BH(sock_net(sk), |
182 | LINUX_MIB_TCPFASTOPENACTIVEFAIL); | |
183 | } | |
5d424d5a | 184 | /* Black hole detection */ |
ce55dd36 | 185 | tcp_mtu_probing(icsk, sk); |
1da177e4 | 186 | |
b6c6712a | 187 | dst_negative_advice(sk); |
1da177e4 LT |
188 | } |
189 | ||
c6214a97 | 190 | retry_until = net->ipv4.sysctl_tcp_retries2; |
1da177e4 | 191 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 192 | const bool alive = icsk->icsk_rto < TCP_RTO_MAX; |
e905a9ed | 193 | |
1da177e4 | 194 | retry_until = tcp_orphan_retries(sk, alive); |
6fa12c85 | 195 | do_reset = alive || |
21a180cd | 196 | !retransmits_timed_out(sk, retry_until, 0, 0); |
1da177e4 | 197 | |
6fa12c85 | 198 | if (tcp_out_of_resources(sk, do_reset)) |
1da177e4 LT |
199 | return 1; |
200 | } | |
201 | } | |
202 | ||
dca43c75 | 203 | if (retransmits_timed_out(sk, retry_until, |
21a180cd | 204 | syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) { |
1da177e4 LT |
205 | /* Has it gone just too far? */ |
206 | tcp_write_err(sk); | |
207 | return 1; | |
208 | } | |
209 | return 0; | |
210 | } | |
211 | ||
6f458dfb | 212 | void tcp_delack_timer_handler(struct sock *sk) |
1da177e4 | 213 | { |
1da177e4 | 214 | struct tcp_sock *tp = tcp_sk(sk); |
463c84b9 | 215 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 216 | |
9993e7d3 | 217 | sk_mem_reclaim_partial(sk); |
1da177e4 | 218 | |
463c84b9 | 219 | if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER)) |
1da177e4 LT |
220 | goto out; |
221 | ||
463c84b9 ACM |
222 | if (time_after(icsk->icsk_ack.timeout, jiffies)) { |
223 | sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout); | |
1da177e4 LT |
224 | goto out; |
225 | } | |
463c84b9 | 226 | icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER; |
1da177e4 | 227 | |
b03efcfb | 228 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
229 | struct sk_buff *skb; |
230 | ||
de0744af | 231 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED); |
1da177e4 LT |
232 | |
233 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 234 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
235 | |
236 | tp->ucopy.memory = 0; | |
237 | } | |
238 | ||
463c84b9 ACM |
239 | if (inet_csk_ack_scheduled(sk)) { |
240 | if (!icsk->icsk_ack.pingpong) { | |
1da177e4 | 241 | /* Delayed ACK missed: inflate ATO. */ |
463c84b9 | 242 | icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto); |
1da177e4 LT |
243 | } else { |
244 | /* Delayed ACK missed: leave pingpong mode and | |
245 | * deflate ATO. | |
246 | */ | |
463c84b9 ACM |
247 | icsk->icsk_ack.pingpong = 0; |
248 | icsk->icsk_ack.ato = TCP_ATO_MIN; | |
1da177e4 LT |
249 | } |
250 | tcp_send_ack(sk); | |
de0744af | 251 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS); |
1da177e4 | 252 | } |
1da177e4 LT |
253 | |
254 | out: | |
b8da51eb | 255 | if (tcp_under_memory_pressure(sk)) |
3ab224be | 256 | sk_mem_reclaim(sk); |
6f458dfb ED |
257 | } |
258 | ||
259 | static void tcp_delack_timer(unsigned long data) | |
260 | { | |
261 | struct sock *sk = (struct sock *)data; | |
262 | ||
263 | bh_lock_sock(sk); | |
264 | if (!sock_owned_by_user(sk)) { | |
265 | tcp_delack_timer_handler(sk); | |
266 | } else { | |
267 | inet_csk(sk)->icsk_ack.blocked = 1; | |
268 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED); | |
269 | /* deleguate our work to tcp_release_cb() */ | |
144d56e9 ED |
270 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags)) |
271 | sock_hold(sk); | |
6f458dfb | 272 | } |
1da177e4 LT |
273 | bh_unlock_sock(sk); |
274 | sock_put(sk); | |
275 | } | |
276 | ||
277 | static void tcp_probe_timer(struct sock *sk) | |
278 | { | |
6687e988 | 279 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
280 | struct tcp_sock *tp = tcp_sk(sk); |
281 | int max_probes; | |
b248230c | 282 | u32 start_ts; |
1da177e4 | 283 | |
fe067e8a | 284 | if (tp->packets_out || !tcp_send_head(sk)) { |
6687e988 | 285 | icsk->icsk_probes_out = 0; |
1da177e4 LT |
286 | return; |
287 | } | |
288 | ||
b248230c YC |
289 | /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as |
290 | * long as the receiver continues to respond probes. We support this by | |
291 | * default and reset icsk_probes_out with incoming ACKs. But if the | |
292 | * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we | |
293 | * kill the socket when the retry count and the time exceeds the | |
294 | * corresponding system limit. We also implement similar policy when | |
295 | * we use RTO to probe window in tcp_retransmit_timer(). | |
1da177e4 | 296 | */ |
b248230c YC |
297 | start_ts = tcp_skb_timestamp(tcp_send_head(sk)); |
298 | if (!start_ts) | |
299 | skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp); | |
300 | else if (icsk->icsk_user_timeout && | |
301 | (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout) | |
302 | goto abort; | |
1da177e4 | 303 | |
c6214a97 | 304 | max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2; |
1da177e4 | 305 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 306 | const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX; |
e905a9ed | 307 | |
1da177e4 | 308 | max_probes = tcp_orphan_retries(sk, alive); |
b248230c YC |
309 | if (!alive && icsk->icsk_backoff >= max_probes) |
310 | goto abort; | |
311 | if (tcp_out_of_resources(sk, true)) | |
1da177e4 LT |
312 | return; |
313 | } | |
314 | ||
6687e988 | 315 | if (icsk->icsk_probes_out > max_probes) { |
b248230c | 316 | abort: tcp_write_err(sk); |
1da177e4 LT |
317 | } else { |
318 | /* Only send another probe if we didn't close things up. */ | |
319 | tcp_send_probe0(sk); | |
320 | } | |
321 | } | |
322 | ||
8336886f JC |
323 | /* |
324 | * Timer for Fast Open socket to retransmit SYNACK. Note that the | |
325 | * sk here is the child socket, not the parent (listener) socket. | |
326 | */ | |
327 | static void tcp_fastopen_synack_timer(struct sock *sk) | |
328 | { | |
329 | struct inet_connection_sock *icsk = inet_csk(sk); | |
330 | int max_retries = icsk->icsk_syn_retries ? : | |
7c083ecb | 331 | sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */ |
8336886f JC |
332 | struct request_sock *req; |
333 | ||
334 | req = tcp_sk(sk)->fastopen_rsk; | |
42cb80a2 | 335 | req->rsk_ops->syn_ack_timeout(req); |
8336886f | 336 | |
e6c022a4 | 337 | if (req->num_timeout >= max_retries) { |
8336886f JC |
338 | tcp_write_err(sk); |
339 | return; | |
340 | } | |
341 | /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error | |
342 | * returned from rtx_syn_ack() to make it more persistent like | |
343 | * regular retransmit because if the child socket has been accepted | |
344 | * it's not good to give up too easily. | |
345 | */ | |
e6c022a4 ED |
346 | inet_rtx_syn_ack(sk, req); |
347 | req->num_timeout++; | |
8336886f | 348 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
e6c022a4 | 349 | TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); |
8336886f JC |
350 | } |
351 | ||
1da177e4 LT |
352 | /* |
353 | * The TCP retransmit timer. | |
354 | */ | |
355 | ||
f1ecd5d9 | 356 | void tcp_retransmit_timer(struct sock *sk) |
1da177e4 LT |
357 | { |
358 | struct tcp_sock *tp = tcp_sk(sk); | |
ae5c3f40 | 359 | struct net *net = sock_net(sk); |
463c84b9 | 360 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 361 | |
8336886f | 362 | if (tp->fastopen_rsk) { |
37561f68 JC |
363 | WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && |
364 | sk->sk_state != TCP_FIN_WAIT1); | |
8336886f JC |
365 | tcp_fastopen_synack_timer(sk); |
366 | /* Before we receive ACK to our SYN-ACK don't retransmit | |
367 | * anything else (e.g., data or FIN segments). | |
368 | */ | |
369 | return; | |
370 | } | |
1da177e4 LT |
371 | if (!tp->packets_out) |
372 | goto out; | |
373 | ||
547b792c | 374 | WARN_ON(tcp_write_queue_empty(sk)); |
1da177e4 | 375 | |
9b717a8d ND |
376 | tp->tlp_high_seq = 0; |
377 | ||
1da177e4 LT |
378 | if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) && |
379 | !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) { | |
380 | /* Receiver dastardly shrinks window. Our retransmits | |
381 | * become zero probes, but we should not timeout this | |
382 | * connection. If the socket is an orphan, time it out, | |
383 | * we cannot allow such beasts to hang infinitely. | |
384 | */ | |
569508c9 YH |
385 | struct inet_sock *inet = inet_sk(sk); |
386 | if (sk->sk_family == AF_INET) { | |
ba7a46f1 JP |
387 | net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
388 | &inet->inet_daddr, | |
389 | ntohs(inet->inet_dport), | |
390 | inet->inet_num, | |
391 | tp->snd_una, tp->snd_nxt); | |
1da177e4 | 392 | } |
dfd56b8b | 393 | #if IS_ENABLED(CONFIG_IPV6) |
569508c9 | 394 | else if (sk->sk_family == AF_INET6) { |
ba7a46f1 JP |
395 | net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
396 | &sk->sk_v6_daddr, | |
397 | ntohs(inet->inet_dport), | |
398 | inet->inet_num, | |
399 | tp->snd_una, tp->snd_nxt); | |
569508c9 | 400 | } |
1da177e4 LT |
401 | #endif |
402 | if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) { | |
403 | tcp_write_err(sk); | |
404 | goto out; | |
405 | } | |
5ae344c9 | 406 | tcp_enter_loss(sk); |
fe067e8a | 407 | tcp_retransmit_skb(sk, tcp_write_queue_head(sk)); |
1da177e4 LT |
408 | __sk_dst_reset(sk); |
409 | goto out_reset_timer; | |
410 | } | |
411 | ||
412 | if (tcp_write_timeout(sk)) | |
413 | goto out; | |
414 | ||
463c84b9 | 415 | if (icsk->icsk_retransmits == 0) { |
40b215e5 PE |
416 | int mib_idx; |
417 | ||
c60ce4e2 | 418 | if (icsk->icsk_ca_state == TCP_CA_Recovery) { |
bc079e9e IJ |
419 | if (tcp_is_sack(tp)) |
420 | mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL; | |
421 | else | |
422 | mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL; | |
6687e988 | 423 | } else if (icsk->icsk_ca_state == TCP_CA_Loss) { |
40b215e5 | 424 | mib_idx = LINUX_MIB_TCPLOSSFAILURES; |
c60ce4e2 IJ |
425 | } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) || |
426 | tp->sacked_out) { | |
427 | if (tcp_is_sack(tp)) | |
428 | mib_idx = LINUX_MIB_TCPSACKFAILURES; | |
429 | else | |
430 | mib_idx = LINUX_MIB_TCPRENOFAILURES; | |
1da177e4 | 431 | } else { |
40b215e5 | 432 | mib_idx = LINUX_MIB_TCPTIMEOUTS; |
1da177e4 | 433 | } |
de0744af | 434 | NET_INC_STATS_BH(sock_net(sk), mib_idx); |
1da177e4 LT |
435 | } |
436 | ||
5ae344c9 | 437 | tcp_enter_loss(sk); |
1da177e4 | 438 | |
fe067e8a | 439 | if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) { |
1da177e4 LT |
440 | /* Retransmission failed because of local congestion, |
441 | * do not backoff. | |
442 | */ | |
463c84b9 ACM |
443 | if (!icsk->icsk_retransmits) |
444 | icsk->icsk_retransmits = 1; | |
445 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
3f421baa ACM |
446 | min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL), |
447 | TCP_RTO_MAX); | |
1da177e4 LT |
448 | goto out; |
449 | } | |
450 | ||
451 | /* Increase the timeout each time we retransmit. Note that | |
452 | * we do not increase the rtt estimate. rto is initialized | |
453 | * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests | |
454 | * that doubling rto each time is the least we can get away with. | |
455 | * In KA9Q, Karn uses this for the first few times, and then | |
456 | * goes to quadratic. netBSD doubles, but only goes up to *64, | |
457 | * and clamps at 1 to 64 sec afterwards. Note that 120 sec is | |
458 | * defined in the protocol as the maximum possible RTT. I guess | |
459 | * we'll have to use something other than TCP to talk to the | |
460 | * University of Mars. | |
461 | * | |
462 | * PAWS allows us longer timeouts and large windows, so once | |
463 | * implemented ftp to mars will work nicely. We will have to fix | |
464 | * the 120 second clamps though! | |
465 | */ | |
463c84b9 ACM |
466 | icsk->icsk_backoff++; |
467 | icsk->icsk_retransmits++; | |
1da177e4 LT |
468 | |
469 | out_reset_timer: | |
36e31b0a AP |
470 | /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is |
471 | * used to reset timer, set to 0. Recalculate 'icsk_rto' as this | |
472 | * might be increased if the stream oscillates between thin and thick, | |
473 | * thus the old value might already be too high compared to the value | |
474 | * set by 'tcp_set_rto' in tcp_input.c which resets the rto without | |
475 | * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating | |
476 | * exponential backoff behaviour to avoid continue hammering | |
477 | * linear-timeout retransmissions into a black hole | |
478 | */ | |
479 | if (sk->sk_state == TCP_ESTABLISHED && | |
480 | (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) && | |
481 | tcp_stream_is_thin(tp) && | |
482 | icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) { | |
483 | icsk->icsk_backoff = 0; | |
484 | icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX); | |
485 | } else { | |
486 | /* Use normal (exponential) backoff */ | |
487 | icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX); | |
488 | } | |
3f421baa | 489 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX); |
ae5c3f40 | 490 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0, 0)) |
1da177e4 LT |
491 | __sk_dst_reset(sk); |
492 | ||
493 | out:; | |
494 | } | |
495 | ||
6f458dfb | 496 | void tcp_write_timer_handler(struct sock *sk) |
1da177e4 | 497 | { |
463c84b9 | 498 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
499 | int event; |
500 | ||
463c84b9 | 501 | if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending) |
1da177e4 LT |
502 | goto out; |
503 | ||
463c84b9 ACM |
504 | if (time_after(icsk->icsk_timeout, jiffies)) { |
505 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout); | |
1da177e4 LT |
506 | goto out; |
507 | } | |
508 | ||
463c84b9 | 509 | event = icsk->icsk_pending; |
1da177e4 LT |
510 | |
511 | switch (event) { | |
6ba8a3b1 ND |
512 | case ICSK_TIME_EARLY_RETRANS: |
513 | tcp_resume_early_retransmit(sk); | |
514 | break; | |
515 | case ICSK_TIME_LOSS_PROBE: | |
516 | tcp_send_loss_probe(sk); | |
517 | break; | |
463c84b9 | 518 | case ICSK_TIME_RETRANS: |
6ba8a3b1 | 519 | icsk->icsk_pending = 0; |
1da177e4 LT |
520 | tcp_retransmit_timer(sk); |
521 | break; | |
463c84b9 | 522 | case ICSK_TIME_PROBE0: |
6ba8a3b1 | 523 | icsk->icsk_pending = 0; |
1da177e4 LT |
524 | tcp_probe_timer(sk); |
525 | break; | |
526 | } | |
1da177e4 LT |
527 | |
528 | out: | |
3ab224be | 529 | sk_mem_reclaim(sk); |
6f458dfb ED |
530 | } |
531 | ||
532 | static void tcp_write_timer(unsigned long data) | |
533 | { | |
534 | struct sock *sk = (struct sock *)data; | |
535 | ||
536 | bh_lock_sock(sk); | |
537 | if (!sock_owned_by_user(sk)) { | |
538 | tcp_write_timer_handler(sk); | |
539 | } else { | |
540 | /* deleguate our work to tcp_release_cb() */ | |
144d56e9 ED |
541 | if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags)) |
542 | sock_hold(sk); | |
6f458dfb | 543 | } |
1da177e4 LT |
544 | bh_unlock_sock(sk); |
545 | sock_put(sk); | |
546 | } | |
547 | ||
42cb80a2 | 548 | void tcp_syn_ack_timeout(const struct request_sock *req) |
72659ecc | 549 | { |
42cb80a2 ED |
550 | struct net *net = read_pnet(&inet_rsk(req)->ireq_net); |
551 | ||
552 | NET_INC_STATS_BH(net, LINUX_MIB_TCPTIMEOUTS); | |
72659ecc OP |
553 | } |
554 | EXPORT_SYMBOL(tcp_syn_ack_timeout); | |
555 | ||
1da177e4 LT |
556 | void tcp_set_keepalive(struct sock *sk, int val) |
557 | { | |
558 | if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) | |
559 | return; | |
560 | ||
561 | if (val && !sock_flag(sk, SOCK_KEEPOPEN)) | |
463c84b9 | 562 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk))); |
1da177e4 | 563 | else if (!val) |
463c84b9 | 564 | inet_csk_delete_keepalive_timer(sk); |
1da177e4 LT |
565 | } |
566 | ||
567 | ||
568 | static void tcp_keepalive_timer (unsigned long data) | |
569 | { | |
570 | struct sock *sk = (struct sock *) data; | |
6687e988 | 571 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 572 | struct tcp_sock *tp = tcp_sk(sk); |
6c37e5de | 573 | u32 elapsed; |
1da177e4 LT |
574 | |
575 | /* Only process if socket is not in use. */ | |
576 | bh_lock_sock(sk); | |
577 | if (sock_owned_by_user(sk)) { | |
e905a9ed | 578 | /* Try again later. */ |
463c84b9 | 579 | inet_csk_reset_keepalive_timer (sk, HZ/20); |
1da177e4 LT |
580 | goto out; |
581 | } | |
582 | ||
583 | if (sk->sk_state == TCP_LISTEN) { | |
fa76ce73 | 584 | pr_err("Hmm... keepalive on a LISTEN ???\n"); |
1da177e4 LT |
585 | goto out; |
586 | } | |
587 | ||
588 | if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) { | |
589 | if (tp->linger2 >= 0) { | |
463c84b9 | 590 | const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN; |
1da177e4 LT |
591 | |
592 | if (tmo > 0) { | |
593 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); | |
594 | goto out; | |
595 | } | |
596 | } | |
597 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
598 | goto death; | |
599 | } | |
600 | ||
601 | if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE) | |
602 | goto out; | |
603 | ||
604 | elapsed = keepalive_time_when(tp); | |
605 | ||
606 | /* It is alive without keepalive 8) */ | |
fe067e8a | 607 | if (tp->packets_out || tcp_send_head(sk)) |
1da177e4 LT |
608 | goto resched; |
609 | ||
6c37e5de | 610 | elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
611 | |
612 | if (elapsed >= keepalive_time_when(tp)) { | |
dca43c75 JC |
613 | /* If the TCP_USER_TIMEOUT option is enabled, use that |
614 | * to determine when to timeout instead. | |
615 | */ | |
616 | if ((icsk->icsk_user_timeout != 0 && | |
617 | elapsed >= icsk->icsk_user_timeout && | |
618 | icsk->icsk_probes_out > 0) || | |
619 | (icsk->icsk_user_timeout == 0 && | |
620 | icsk->icsk_probes_out >= keepalive_probes(tp))) { | |
1da177e4 LT |
621 | tcp_send_active_reset(sk, GFP_ATOMIC); |
622 | tcp_write_err(sk); | |
623 | goto out; | |
624 | } | |
e520af48 | 625 | if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) { |
6687e988 | 626 | icsk->icsk_probes_out++; |
1da177e4 LT |
627 | elapsed = keepalive_intvl_when(tp); |
628 | } else { | |
629 | /* If keepalive was lost due to local congestion, | |
630 | * try harder. | |
631 | */ | |
632 | elapsed = TCP_RESOURCE_PROBE_INTERVAL; | |
633 | } | |
634 | } else { | |
635 | /* It is tp->rcv_tstamp + keepalive_time_when(tp) */ | |
636 | elapsed = keepalive_time_when(tp) - elapsed; | |
637 | } | |
638 | ||
3ab224be | 639 | sk_mem_reclaim(sk); |
1da177e4 LT |
640 | |
641 | resched: | |
463c84b9 | 642 | inet_csk_reset_keepalive_timer (sk, elapsed); |
1da177e4 LT |
643 | goto out; |
644 | ||
e905a9ed | 645 | death: |
1da177e4 LT |
646 | tcp_done(sk); |
647 | ||
648 | out: | |
649 | bh_unlock_sock(sk); | |
650 | sock_put(sk); | |
651 | } | |
6f458dfb ED |
652 | |
653 | void tcp_init_xmit_timers(struct sock *sk) | |
654 | { | |
655 | inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer, | |
656 | &tcp_keepalive_timer); | |
657 | } |