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 | |
c380d37e RS |
27 | /** |
28 | * tcp_write_err() - close socket and save error info | |
29 | * @sk: The socket the error has appeared on. | |
30 | * | |
31 | * Returns: Nothing (void) | |
32 | */ | |
33 | ||
1da177e4 LT |
34 | static void tcp_write_err(struct sock *sk) |
35 | { | |
36 | sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT; | |
37 | sk->sk_error_report(sk); | |
38 | ||
39 | tcp_done(sk); | |
02a1d6e7 | 40 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT); |
1da177e4 LT |
41 | } |
42 | ||
c380d37e RS |
43 | /** |
44 | * tcp_out_of_resources() - Close socket if out of resources | |
45 | * @sk: pointer to current socket | |
46 | * @do_reset: send a last packet with reset flag | |
1da177e4 | 47 | * |
c380d37e RS |
48 | * Do not allow orphaned sockets to eat all our resources. |
49 | * This is direct violation of TCP specs, but it is required | |
50 | * to prevent DoS attacks. It is called when a retransmission timeout | |
51 | * or zero probe timeout occurs on orphaned socket. | |
52 | * | |
53 | * Criteria is still not confirmed experimentally and may change. | |
54 | * We kill the socket, if: | |
55 | * 1. If number of orphaned sockets exceeds an administratively configured | |
56 | * limit. | |
57 | * 2. If we have strong memory pressure. | |
1da177e4 | 58 | */ |
b248230c | 59 | static int tcp_out_of_resources(struct sock *sk, bool do_reset) |
1da177e4 LT |
60 | { |
61 | struct tcp_sock *tp = tcp_sk(sk); | |
ad1af0fe | 62 | int shift = 0; |
1da177e4 | 63 | |
e905a9ed | 64 | /* If peer does not open window for long time, or did not transmit |
1da177e4 LT |
65 | * anything for long time, penalize it. */ |
66 | if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset) | |
ad1af0fe | 67 | shift++; |
1da177e4 LT |
68 | |
69 | /* If some dubious ICMP arrived, penalize even more. */ | |
70 | if (sk->sk_err_soft) | |
ad1af0fe | 71 | shift++; |
1da177e4 | 72 | |
efcdbf24 | 73 | if (tcp_check_oom(sk, shift)) { |
1da177e4 LT |
74 | /* Catch exceptional cases, when connection requires reset. |
75 | * 1. Last segment was sent recently. */ | |
76 | if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN || | |
77 | /* 2. Window is closed. */ | |
78 | (!tp->snd_wnd && !tp->packets_out)) | |
b248230c | 79 | do_reset = true; |
1da177e4 LT |
80 | if (do_reset) |
81 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
82 | tcp_done(sk); | |
02a1d6e7 | 83 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
84 | return 1; |
85 | } | |
86 | return 0; | |
87 | } | |
88 | ||
c380d37e RS |
89 | /** |
90 | * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket | |
91 | * @sk: Pointer to the current socket. | |
92 | * @alive: bool, socket alive state | |
93 | */ | |
7533ce30 | 94 | static int tcp_orphan_retries(struct sock *sk, bool alive) |
1da177e4 | 95 | { |
c402d9be | 96 | int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */ |
1da177e4 LT |
97 | |
98 | /* We know from an ICMP that something is wrong. */ | |
99 | if (sk->sk_err_soft && !alive) | |
100 | retries = 0; | |
101 | ||
102 | /* However, if socket sent something recently, select some safe | |
103 | * number of retries. 8 corresponds to >100 seconds with minimal | |
104 | * RTO of 200msec. */ | |
105 | if (retries == 0 && alive) | |
106 | retries = 8; | |
107 | return retries; | |
108 | } | |
109 | ||
ce55dd36 ED |
110 | static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk) |
111 | { | |
b0f9ca53 FD |
112 | struct net *net = sock_net(sk); |
113 | ||
ce55dd36 | 114 | /* Black hole detection */ |
b0f9ca53 | 115 | if (net->ipv4.sysctl_tcp_mtu_probing) { |
ce55dd36 ED |
116 | if (!icsk->icsk_mtup.enabled) { |
117 | icsk->icsk_mtup.enabled = 1; | |
05cbc0db | 118 | icsk->icsk_mtup.probe_timestamp = tcp_time_stamp; |
ce55dd36 ED |
119 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
120 | } else { | |
b0f9ca53 | 121 | struct net *net = sock_net(sk); |
ce55dd36 | 122 | struct tcp_sock *tp = tcp_sk(sk); |
829942c1 DM |
123 | int mss; |
124 | ||
8beb5c5f | 125 | mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1; |
b0f9ca53 | 126 | mss = min(net->ipv4.sysctl_tcp_base_mss, mss); |
ce55dd36 ED |
127 | mss = max(mss, 68 - tp->tcp_header_len); |
128 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); | |
129 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); | |
130 | } | |
131 | } | |
132 | } | |
133 | ||
c380d37e RS |
134 | |
135 | /** | |
136 | * retransmits_timed_out() - returns true if this connection has timed out | |
137 | * @sk: The current socket | |
138 | * @boundary: max number of retransmissions | |
139 | * @timeout: A custom timeout value. | |
140 | * If set to 0 the default timeout is calculated and used. | |
141 | * Using TCP_RTO_MIN and the number of unsuccessful retransmits. | |
142 | * @syn_set: true if the SYN Bit was set. | |
143 | * | |
144 | * The default "timeout" value this function can calculate and use | |
145 | * is equivalent to the timeout of a TCP Connection | |
146 | * after "boundary" unsuccessful, exponentially backed-off | |
4d22f7d3 DL |
147 | * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if |
148 | * syn_set flag is set. | |
c380d37e | 149 | * |
2f7de571 DL |
150 | */ |
151 | static bool retransmits_timed_out(struct sock *sk, | |
dca43c75 | 152 | unsigned int boundary, |
21a180cd | 153 | unsigned int timeout, |
4d22f7d3 | 154 | bool syn_set) |
2f7de571 | 155 | { |
dca43c75 | 156 | unsigned int linear_backoff_thresh, start_ts; |
4d22f7d3 | 157 | unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN; |
2f7de571 DL |
158 | |
159 | if (!inet_csk(sk)->icsk_retransmits) | |
160 | return false; | |
161 | ||
7faee5c0 ED |
162 | start_ts = tcp_sk(sk)->retrans_stamp; |
163 | if (unlikely(!start_ts)) | |
164 | start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk)); | |
2f7de571 | 165 | |
dca43c75 | 166 | if (likely(timeout == 0)) { |
21a180cd | 167 | linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base); |
2f7de571 | 168 | |
dca43c75 | 169 | if (boundary <= linear_backoff_thresh) |
21a180cd | 170 | timeout = ((2 << boundary) - 1) * rto_base; |
dca43c75 | 171 | else |
21a180cd | 172 | timeout = ((2 << linear_backoff_thresh) - 1) * rto_base + |
dca43c75 JC |
173 | (boundary - linear_backoff_thresh) * TCP_RTO_MAX; |
174 | } | |
2f7de571 DL |
175 | return (tcp_time_stamp - start_ts) >= timeout; |
176 | } | |
177 | ||
1da177e4 LT |
178 | /* A write timeout has occurred. Process the after effects. */ |
179 | static int tcp_write_timeout(struct sock *sk) | |
180 | { | |
5d424d5a | 181 | struct inet_connection_sock *icsk = inet_csk(sk); |
c968601d | 182 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 183 | struct net *net = sock_net(sk); |
1da177e4 | 184 | int retry_until; |
3db1cd5c | 185 | bool do_reset, syn_set = false; |
1da177e4 LT |
186 | |
187 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
c968601d | 188 | if (icsk->icsk_retransmits) { |
b6c6712a | 189 | dst_negative_advice(sk); |
c968601d | 190 | if (tp->syn_fastopen || tp->syn_data) |
2646c831 | 191 | tcp_fastopen_cache_set(sk, 0, NULL, true, 0); |
dd52bc2b | 192 | if (tp->syn_data && icsk->icsk_retransmits == 1) |
c10d9310 ED |
193 | NET_INC_STATS(sock_net(sk), |
194 | LINUX_MIB_TCPFASTOPENACTIVEFAIL); | |
3acf3ec3 LB |
195 | } else if (!tp->syn_data && !tp->syn_fastopen) { |
196 | sk_rethink_txhash(sk); | |
c968601d | 197 | } |
6fa25166 | 198 | retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
3db1cd5c | 199 | syn_set = true; |
1da177e4 | 200 | } else { |
ae5c3f40 | 201 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0, 0)) { |
0e45f4da YC |
202 | /* Some middle-boxes may black-hole Fast Open _after_ |
203 | * the handshake. Therefore we conservatively disable | |
204 | * Fast Open on this path on recurring timeouts with | |
205 | * few or zero bytes acked after Fast Open. | |
206 | */ | |
207 | if (tp->syn_data_acked && | |
208 | tp->bytes_acked <= tp->rx_opt.mss_clamp) { | |
209 | tcp_fastopen_cache_set(sk, 0, NULL, true, 0); | |
ae5c3f40 | 210 | if (icsk->icsk_retransmits == net->ipv4.sysctl_tcp_retries1) |
c10d9310 ED |
211 | NET_INC_STATS(sock_net(sk), |
212 | LINUX_MIB_TCPFASTOPENACTIVEFAIL); | |
0e45f4da | 213 | } |
5d424d5a | 214 | /* Black hole detection */ |
ce55dd36 | 215 | tcp_mtu_probing(icsk, sk); |
1da177e4 | 216 | |
b6c6712a | 217 | dst_negative_advice(sk); |
3acf3ec3 LB |
218 | } else { |
219 | sk_rethink_txhash(sk); | |
1da177e4 LT |
220 | } |
221 | ||
c6214a97 | 222 | retry_until = net->ipv4.sysctl_tcp_retries2; |
1da177e4 | 223 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 224 | const bool alive = icsk->icsk_rto < TCP_RTO_MAX; |
e905a9ed | 225 | |
1da177e4 | 226 | retry_until = tcp_orphan_retries(sk, alive); |
6fa12c85 | 227 | do_reset = alive || |
21a180cd | 228 | !retransmits_timed_out(sk, retry_until, 0, 0); |
1da177e4 | 229 | |
6fa12c85 | 230 | if (tcp_out_of_resources(sk, do_reset)) |
1da177e4 LT |
231 | return 1; |
232 | } | |
233 | } | |
234 | ||
dca43c75 | 235 | if (retransmits_timed_out(sk, retry_until, |
21a180cd | 236 | syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) { |
1da177e4 LT |
237 | /* Has it gone just too far? */ |
238 | tcp_write_err(sk); | |
239 | return 1; | |
240 | } | |
241 | return 0; | |
242 | } | |
243 | ||
c10d9310 | 244 | /* Called with BH disabled */ |
6f458dfb | 245 | void tcp_delack_timer_handler(struct sock *sk) |
1da177e4 | 246 | { |
1da177e4 | 247 | struct tcp_sock *tp = tcp_sk(sk); |
463c84b9 | 248 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 249 | |
9993e7d3 | 250 | sk_mem_reclaim_partial(sk); |
1da177e4 | 251 | |
463c84b9 | 252 | if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER)) |
1da177e4 LT |
253 | goto out; |
254 | ||
463c84b9 ACM |
255 | if (time_after(icsk->icsk_ack.timeout, jiffies)) { |
256 | sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout); | |
1da177e4 LT |
257 | goto out; |
258 | } | |
463c84b9 | 259 | icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER; |
1da177e4 | 260 | |
b03efcfb | 261 | if (!skb_queue_empty(&tp->ucopy.prequeue)) { |
1da177e4 LT |
262 | struct sk_buff *skb; |
263 | ||
02a1d6e7 | 264 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED); |
1da177e4 LT |
265 | |
266 | while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) | |
c57943a1 | 267 | sk_backlog_rcv(sk, skb); |
1da177e4 LT |
268 | |
269 | tp->ucopy.memory = 0; | |
270 | } | |
271 | ||
463c84b9 ACM |
272 | if (inet_csk_ack_scheduled(sk)) { |
273 | if (!icsk->icsk_ack.pingpong) { | |
1da177e4 | 274 | /* Delayed ACK missed: inflate ATO. */ |
463c84b9 | 275 | icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto); |
1da177e4 LT |
276 | } else { |
277 | /* Delayed ACK missed: leave pingpong mode and | |
278 | * deflate ATO. | |
279 | */ | |
463c84b9 ACM |
280 | icsk->icsk_ack.pingpong = 0; |
281 | icsk->icsk_ack.ato = TCP_ATO_MIN; | |
1da177e4 LT |
282 | } |
283 | tcp_send_ack(sk); | |
02a1d6e7 | 284 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS); |
1da177e4 | 285 | } |
1da177e4 LT |
286 | |
287 | out: | |
b8da51eb | 288 | if (tcp_under_memory_pressure(sk)) |
3ab224be | 289 | sk_mem_reclaim(sk); |
6f458dfb ED |
290 | } |
291 | ||
c380d37e RS |
292 | |
293 | /** | |
294 | * tcp_delack_timer() - The TCP delayed ACK timeout handler | |
295 | * @data: Pointer to the current socket. (gets casted to struct sock *) | |
296 | * | |
297 | * This function gets (indirectly) called when the kernel timer for a TCP packet | |
298 | * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work. | |
299 | * | |
300 | * Returns: Nothing (void) | |
301 | */ | |
6f458dfb ED |
302 | static void tcp_delack_timer(unsigned long data) |
303 | { | |
304 | struct sock *sk = (struct sock *)data; | |
305 | ||
306 | bh_lock_sock(sk); | |
307 | if (!sock_owned_by_user(sk)) { | |
308 | tcp_delack_timer_handler(sk); | |
309 | } else { | |
310 | inet_csk(sk)->icsk_ack.blocked = 1; | |
02a1d6e7 | 311 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED); |
6f458dfb | 312 | /* deleguate our work to tcp_release_cb() */ |
7aa5470c | 313 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 | 314 | sock_hold(sk); |
6f458dfb | 315 | } |
1da177e4 LT |
316 | bh_unlock_sock(sk); |
317 | sock_put(sk); | |
318 | } | |
319 | ||
320 | static void tcp_probe_timer(struct sock *sk) | |
321 | { | |
6687e988 | 322 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
323 | struct tcp_sock *tp = tcp_sk(sk); |
324 | int max_probes; | |
b248230c | 325 | u32 start_ts; |
1da177e4 | 326 | |
fe067e8a | 327 | if (tp->packets_out || !tcp_send_head(sk)) { |
6687e988 | 328 | icsk->icsk_probes_out = 0; |
1da177e4 LT |
329 | return; |
330 | } | |
331 | ||
b248230c YC |
332 | /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as |
333 | * long as the receiver continues to respond probes. We support this by | |
334 | * default and reset icsk_probes_out with incoming ACKs. But if the | |
335 | * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we | |
336 | * kill the socket when the retry count and the time exceeds the | |
337 | * corresponding system limit. We also implement similar policy when | |
338 | * we use RTO to probe window in tcp_retransmit_timer(). | |
1da177e4 | 339 | */ |
b248230c YC |
340 | start_ts = tcp_skb_timestamp(tcp_send_head(sk)); |
341 | if (!start_ts) | |
342 | skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp); | |
343 | else if (icsk->icsk_user_timeout && | |
344 | (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout) | |
345 | goto abort; | |
1da177e4 | 346 | |
c6214a97 | 347 | max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2; |
1da177e4 | 348 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 349 | const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX; |
e905a9ed | 350 | |
1da177e4 | 351 | max_probes = tcp_orphan_retries(sk, alive); |
b248230c YC |
352 | if (!alive && icsk->icsk_backoff >= max_probes) |
353 | goto abort; | |
354 | if (tcp_out_of_resources(sk, true)) | |
1da177e4 LT |
355 | return; |
356 | } | |
357 | ||
6687e988 | 358 | if (icsk->icsk_probes_out > max_probes) { |
b248230c | 359 | abort: tcp_write_err(sk); |
1da177e4 LT |
360 | } else { |
361 | /* Only send another probe if we didn't close things up. */ | |
362 | tcp_send_probe0(sk); | |
363 | } | |
364 | } | |
365 | ||
8336886f JC |
366 | /* |
367 | * Timer for Fast Open socket to retransmit SYNACK. Note that the | |
368 | * sk here is the child socket, not the parent (listener) socket. | |
369 | */ | |
370 | static void tcp_fastopen_synack_timer(struct sock *sk) | |
371 | { | |
372 | struct inet_connection_sock *icsk = inet_csk(sk); | |
373 | int max_retries = icsk->icsk_syn_retries ? : | |
7c083ecb | 374 | sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */ |
8336886f JC |
375 | struct request_sock *req; |
376 | ||
377 | req = tcp_sk(sk)->fastopen_rsk; | |
42cb80a2 | 378 | req->rsk_ops->syn_ack_timeout(req); |
8336886f | 379 | |
e6c022a4 | 380 | if (req->num_timeout >= max_retries) { |
8336886f JC |
381 | tcp_write_err(sk); |
382 | return; | |
383 | } | |
384 | /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error | |
385 | * returned from rtx_syn_ack() to make it more persistent like | |
386 | * regular retransmit because if the child socket has been accepted | |
387 | * it's not good to give up too easily. | |
388 | */ | |
e6c022a4 ED |
389 | inet_rtx_syn_ack(sk, req); |
390 | req->num_timeout++; | |
7e32b443 | 391 | icsk->icsk_retransmits++; |
8336886f | 392 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
e6c022a4 | 393 | TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); |
8336886f JC |
394 | } |
395 | ||
1da177e4 | 396 | |
c380d37e RS |
397 | /** |
398 | * tcp_retransmit_timer() - The TCP retransmit timeout handler | |
399 | * @sk: Pointer to the current socket. | |
400 | * | |
401 | * This function gets called when the kernel timer for a TCP packet | |
402 | * of this socket expires. | |
403 | * | |
404 | * It handles retransmission, timer adjustment and other necesarry measures. | |
405 | * | |
406 | * Returns: Nothing (void) | |
407 | */ | |
f1ecd5d9 | 408 | void tcp_retransmit_timer(struct sock *sk) |
1da177e4 LT |
409 | { |
410 | struct tcp_sock *tp = tcp_sk(sk); | |
ae5c3f40 | 411 | struct net *net = sock_net(sk); |
463c84b9 | 412 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 413 | |
8336886f | 414 | if (tp->fastopen_rsk) { |
37561f68 JC |
415 | WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && |
416 | sk->sk_state != TCP_FIN_WAIT1); | |
8336886f JC |
417 | tcp_fastopen_synack_timer(sk); |
418 | /* Before we receive ACK to our SYN-ACK don't retransmit | |
419 | * anything else (e.g., data or FIN segments). | |
420 | */ | |
421 | return; | |
422 | } | |
1da177e4 LT |
423 | if (!tp->packets_out) |
424 | goto out; | |
425 | ||
547b792c | 426 | WARN_ON(tcp_write_queue_empty(sk)); |
1da177e4 | 427 | |
9b717a8d ND |
428 | tp->tlp_high_seq = 0; |
429 | ||
1da177e4 LT |
430 | if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) && |
431 | !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) { | |
432 | /* Receiver dastardly shrinks window. Our retransmits | |
433 | * become zero probes, but we should not timeout this | |
434 | * connection. If the socket is an orphan, time it out, | |
435 | * we cannot allow such beasts to hang infinitely. | |
436 | */ | |
569508c9 YH |
437 | struct inet_sock *inet = inet_sk(sk); |
438 | if (sk->sk_family == AF_INET) { | |
ba7a46f1 JP |
439 | net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
440 | &inet->inet_daddr, | |
441 | ntohs(inet->inet_dport), | |
442 | inet->inet_num, | |
443 | tp->snd_una, tp->snd_nxt); | |
1da177e4 | 444 | } |
dfd56b8b | 445 | #if IS_ENABLED(CONFIG_IPV6) |
569508c9 | 446 | else if (sk->sk_family == AF_INET6) { |
ba7a46f1 JP |
447 | net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
448 | &sk->sk_v6_daddr, | |
449 | ntohs(inet->inet_dport), | |
450 | inet->inet_num, | |
451 | tp->snd_una, tp->snd_nxt); | |
569508c9 | 452 | } |
1da177e4 LT |
453 | #endif |
454 | if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) { | |
455 | tcp_write_err(sk); | |
456 | goto out; | |
457 | } | |
5ae344c9 | 458 | tcp_enter_loss(sk); |
10d3be56 | 459 | tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1); |
1da177e4 LT |
460 | __sk_dst_reset(sk); |
461 | goto out_reset_timer; | |
462 | } | |
463 | ||
464 | if (tcp_write_timeout(sk)) | |
465 | goto out; | |
466 | ||
463c84b9 | 467 | if (icsk->icsk_retransmits == 0) { |
40b215e5 PE |
468 | int mib_idx; |
469 | ||
c60ce4e2 | 470 | if (icsk->icsk_ca_state == TCP_CA_Recovery) { |
bc079e9e IJ |
471 | if (tcp_is_sack(tp)) |
472 | mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL; | |
473 | else | |
474 | mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL; | |
6687e988 | 475 | } else if (icsk->icsk_ca_state == TCP_CA_Loss) { |
40b215e5 | 476 | mib_idx = LINUX_MIB_TCPLOSSFAILURES; |
c60ce4e2 IJ |
477 | } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) || |
478 | tp->sacked_out) { | |
479 | if (tcp_is_sack(tp)) | |
480 | mib_idx = LINUX_MIB_TCPSACKFAILURES; | |
481 | else | |
482 | mib_idx = LINUX_MIB_TCPRENOFAILURES; | |
1da177e4 | 483 | } else { |
40b215e5 | 484 | mib_idx = LINUX_MIB_TCPTIMEOUTS; |
1da177e4 | 485 | } |
02a1d6e7 | 486 | __NET_INC_STATS(sock_net(sk), mib_idx); |
1da177e4 LT |
487 | } |
488 | ||
5ae344c9 | 489 | tcp_enter_loss(sk); |
1da177e4 | 490 | |
10d3be56 | 491 | if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1) > 0) { |
1da177e4 LT |
492 | /* Retransmission failed because of local congestion, |
493 | * do not backoff. | |
494 | */ | |
463c84b9 ACM |
495 | if (!icsk->icsk_retransmits) |
496 | icsk->icsk_retransmits = 1; | |
497 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
3f421baa ACM |
498 | min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL), |
499 | TCP_RTO_MAX); | |
1da177e4 LT |
500 | goto out; |
501 | } | |
502 | ||
503 | /* Increase the timeout each time we retransmit. Note that | |
504 | * we do not increase the rtt estimate. rto is initialized | |
505 | * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests | |
506 | * that doubling rto each time is the least we can get away with. | |
507 | * In KA9Q, Karn uses this for the first few times, and then | |
508 | * goes to quadratic. netBSD doubles, but only goes up to *64, | |
509 | * and clamps at 1 to 64 sec afterwards. Note that 120 sec is | |
510 | * defined in the protocol as the maximum possible RTT. I guess | |
511 | * we'll have to use something other than TCP to talk to the | |
512 | * University of Mars. | |
513 | * | |
514 | * PAWS allows us longer timeouts and large windows, so once | |
515 | * implemented ftp to mars will work nicely. We will have to fix | |
516 | * the 120 second clamps though! | |
517 | */ | |
463c84b9 ACM |
518 | icsk->icsk_backoff++; |
519 | icsk->icsk_retransmits++; | |
1da177e4 LT |
520 | |
521 | out_reset_timer: | |
36e31b0a AP |
522 | /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is |
523 | * used to reset timer, set to 0. Recalculate 'icsk_rto' as this | |
524 | * might be increased if the stream oscillates between thin and thick, | |
525 | * thus the old value might already be too high compared to the value | |
526 | * set by 'tcp_set_rto' in tcp_input.c which resets the rto without | |
527 | * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating | |
528 | * exponential backoff behaviour to avoid continue hammering | |
529 | * linear-timeout retransmissions into a black hole | |
530 | */ | |
531 | if (sk->sk_state == TCP_ESTABLISHED && | |
532 | (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) && | |
533 | tcp_stream_is_thin(tp) && | |
534 | icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) { | |
535 | icsk->icsk_backoff = 0; | |
536 | icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX); | |
537 | } else { | |
538 | /* Use normal (exponential) backoff */ | |
539 | icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX); | |
540 | } | |
3f421baa | 541 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX); |
ae5c3f40 | 542 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0, 0)) |
1da177e4 LT |
543 | __sk_dst_reset(sk); |
544 | ||
545 | out:; | |
546 | } | |
547 | ||
c380d37e RS |
548 | /* Called with bottom-half processing disabled. |
549 | Called by tcp_write_timer() */ | |
6f458dfb | 550 | void tcp_write_timer_handler(struct sock *sk) |
1da177e4 | 551 | { |
463c84b9 | 552 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
553 | int event; |
554 | ||
463c84b9 | 555 | if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending) |
1da177e4 LT |
556 | goto out; |
557 | ||
463c84b9 ACM |
558 | if (time_after(icsk->icsk_timeout, jiffies)) { |
559 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout); | |
1da177e4 LT |
560 | goto out; |
561 | } | |
562 | ||
463c84b9 | 563 | event = icsk->icsk_pending; |
1da177e4 LT |
564 | |
565 | switch (event) { | |
6ba8a3b1 ND |
566 | case ICSK_TIME_EARLY_RETRANS: |
567 | tcp_resume_early_retransmit(sk); | |
568 | break; | |
569 | case ICSK_TIME_LOSS_PROBE: | |
570 | tcp_send_loss_probe(sk); | |
571 | break; | |
463c84b9 | 572 | case ICSK_TIME_RETRANS: |
6ba8a3b1 | 573 | icsk->icsk_pending = 0; |
1da177e4 LT |
574 | tcp_retransmit_timer(sk); |
575 | break; | |
463c84b9 | 576 | case ICSK_TIME_PROBE0: |
6ba8a3b1 | 577 | icsk->icsk_pending = 0; |
1da177e4 LT |
578 | tcp_probe_timer(sk); |
579 | break; | |
580 | } | |
1da177e4 LT |
581 | |
582 | out: | |
3ab224be | 583 | sk_mem_reclaim(sk); |
6f458dfb ED |
584 | } |
585 | ||
586 | static void tcp_write_timer(unsigned long data) | |
587 | { | |
588 | struct sock *sk = (struct sock *)data; | |
589 | ||
590 | bh_lock_sock(sk); | |
591 | if (!sock_owned_by_user(sk)) { | |
592 | tcp_write_timer_handler(sk); | |
593 | } else { | |
c380d37e | 594 | /* delegate our work to tcp_release_cb() */ |
7aa5470c | 595 | if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 | 596 | sock_hold(sk); |
6f458dfb | 597 | } |
1da177e4 LT |
598 | bh_unlock_sock(sk); |
599 | sock_put(sk); | |
600 | } | |
601 | ||
42cb80a2 | 602 | void tcp_syn_ack_timeout(const struct request_sock *req) |
72659ecc | 603 | { |
42cb80a2 ED |
604 | struct net *net = read_pnet(&inet_rsk(req)->ireq_net); |
605 | ||
02a1d6e7 | 606 | __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS); |
72659ecc OP |
607 | } |
608 | EXPORT_SYMBOL(tcp_syn_ack_timeout); | |
609 | ||
1da177e4 LT |
610 | void tcp_set_keepalive(struct sock *sk, int val) |
611 | { | |
612 | if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) | |
613 | return; | |
614 | ||
615 | if (val && !sock_flag(sk, SOCK_KEEPOPEN)) | |
463c84b9 | 616 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk))); |
1da177e4 | 617 | else if (!val) |
463c84b9 | 618 | inet_csk_delete_keepalive_timer(sk); |
1da177e4 LT |
619 | } |
620 | ||
621 | ||
622 | static void tcp_keepalive_timer (unsigned long data) | |
623 | { | |
624 | struct sock *sk = (struct sock *) data; | |
6687e988 | 625 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 626 | struct tcp_sock *tp = tcp_sk(sk); |
6c37e5de | 627 | u32 elapsed; |
1da177e4 LT |
628 | |
629 | /* Only process if socket is not in use. */ | |
630 | bh_lock_sock(sk); | |
631 | if (sock_owned_by_user(sk)) { | |
e905a9ed | 632 | /* Try again later. */ |
463c84b9 | 633 | inet_csk_reset_keepalive_timer (sk, HZ/20); |
1da177e4 LT |
634 | goto out; |
635 | } | |
636 | ||
637 | if (sk->sk_state == TCP_LISTEN) { | |
fa76ce73 | 638 | pr_err("Hmm... keepalive on a LISTEN ???\n"); |
1da177e4 LT |
639 | goto out; |
640 | } | |
641 | ||
642 | if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) { | |
643 | if (tp->linger2 >= 0) { | |
463c84b9 | 644 | const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN; |
1da177e4 LT |
645 | |
646 | if (tmo > 0) { | |
647 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); | |
648 | goto out; | |
649 | } | |
650 | } | |
651 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
652 | goto death; | |
653 | } | |
654 | ||
655 | if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE) | |
656 | goto out; | |
657 | ||
658 | elapsed = keepalive_time_when(tp); | |
659 | ||
660 | /* It is alive without keepalive 8) */ | |
fe067e8a | 661 | if (tp->packets_out || tcp_send_head(sk)) |
1da177e4 LT |
662 | goto resched; |
663 | ||
6c37e5de | 664 | elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
665 | |
666 | if (elapsed >= keepalive_time_when(tp)) { | |
dca43c75 JC |
667 | /* If the TCP_USER_TIMEOUT option is enabled, use that |
668 | * to determine when to timeout instead. | |
669 | */ | |
670 | if ((icsk->icsk_user_timeout != 0 && | |
671 | elapsed >= icsk->icsk_user_timeout && | |
672 | icsk->icsk_probes_out > 0) || | |
673 | (icsk->icsk_user_timeout == 0 && | |
674 | icsk->icsk_probes_out >= keepalive_probes(tp))) { | |
1da177e4 LT |
675 | tcp_send_active_reset(sk, GFP_ATOMIC); |
676 | tcp_write_err(sk); | |
677 | goto out; | |
678 | } | |
e520af48 | 679 | if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) { |
6687e988 | 680 | icsk->icsk_probes_out++; |
1da177e4 LT |
681 | elapsed = keepalive_intvl_when(tp); |
682 | } else { | |
683 | /* If keepalive was lost due to local congestion, | |
684 | * try harder. | |
685 | */ | |
686 | elapsed = TCP_RESOURCE_PROBE_INTERVAL; | |
687 | } | |
688 | } else { | |
689 | /* It is tp->rcv_tstamp + keepalive_time_when(tp) */ | |
690 | elapsed = keepalive_time_when(tp) - elapsed; | |
691 | } | |
692 | ||
3ab224be | 693 | sk_mem_reclaim(sk); |
1da177e4 LT |
694 | |
695 | resched: | |
463c84b9 | 696 | inet_csk_reset_keepalive_timer (sk, elapsed); |
1da177e4 LT |
697 | goto out; |
698 | ||
e905a9ed | 699 | death: |
1da177e4 LT |
700 | tcp_done(sk); |
701 | ||
702 | out: | |
703 | bh_unlock_sock(sk); | |
704 | sock_put(sk); | |
705 | } | |
6f458dfb ED |
706 | |
707 | void tcp_init_xmit_timers(struct sock *sk) | |
708 | { | |
709 | inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer, | |
710 | &tcp_keepalive_timer); | |
711 | } |