Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * Implementation of the Transmission Control Protocol(TCP). | |
8 | * | |
02c30a84 | 9 | * Authors: Ross Biro |
1da177e4 LT |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
12 | * Corey Minyard <wf-rch!minyard@relay.EU.net> | |
13 | * Florian La Roche, <flla@stud.uni-sb.de> | |
14 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> | |
15 | * Linus Torvalds, <torvalds@cs.helsinki.fi> | |
16 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
17 | * Matthew Dillon, <dillon@apollo.west.oic.com> | |
18 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
19 | * Jorge Cwik, <jorge@laser.satlink.net> | |
20 | */ | |
21 | ||
22 | #include <linux/module.h> | |
5a0e3ad6 | 23 | #include <linux/gfp.h> |
1da177e4 LT |
24 | #include <net/tcp.h> |
25 | ||
b701a99e JM |
26 | static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk) |
27 | { | |
28 | struct inet_connection_sock *icsk = inet_csk(sk); | |
29 | u32 elapsed, start_ts; | |
9efdda4e | 30 | s32 remaining; |
b701a99e | 31 | |
7ae18975 YC |
32 | start_ts = tcp_sk(sk)->retrans_stamp; |
33 | if (!icsk->icsk_user_timeout) | |
b701a99e JM |
34 | return icsk->icsk_rto; |
35 | elapsed = tcp_time_stamp(tcp_sk(sk)) - start_ts; | |
9efdda4e ED |
36 | remaining = icsk->icsk_user_timeout - elapsed; |
37 | if (remaining <= 0) | |
b701a99e | 38 | return 1; /* user timeout has passed; fire ASAP */ |
9efdda4e ED |
39 | |
40 | return min_t(u32, icsk->icsk_rto, msecs_to_jiffies(remaining)); | |
b701a99e JM |
41 | } |
42 | ||
c380d37e RS |
43 | /** |
44 | * tcp_write_err() - close socket and save error info | |
45 | * @sk: The socket the error has appeared on. | |
46 | * | |
47 | * Returns: Nothing (void) | |
48 | */ | |
49 | ||
1da177e4 LT |
50 | static void tcp_write_err(struct sock *sk) |
51 | { | |
52 | sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT; | |
53 | sk->sk_error_report(sk); | |
54 | ||
e05836ac | 55 | tcp_write_queue_purge(sk); |
1da177e4 | 56 | tcp_done(sk); |
02a1d6e7 | 57 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT); |
1da177e4 LT |
58 | } |
59 | ||
c380d37e RS |
60 | /** |
61 | * tcp_out_of_resources() - Close socket if out of resources | |
62 | * @sk: pointer to current socket | |
63 | * @do_reset: send a last packet with reset flag | |
1da177e4 | 64 | * |
c380d37e RS |
65 | * Do not allow orphaned sockets to eat all our resources. |
66 | * This is direct violation of TCP specs, but it is required | |
67 | * to prevent DoS attacks. It is called when a retransmission timeout | |
68 | * or zero probe timeout occurs on orphaned socket. | |
69 | * | |
4ee806d5 DS |
70 | * Also close if our net namespace is exiting; in that case there is no |
71 | * hope of ever communicating again since all netns interfaces are already | |
72 | * down (or about to be down), and we need to release our dst references, | |
73 | * which have been moved to the netns loopback interface, so the namespace | |
74 | * can finish exiting. This condition is only possible if we are a kernel | |
75 | * socket, as those do not hold references to the namespace. | |
76 | * | |
c380d37e RS |
77 | * Criteria is still not confirmed experimentally and may change. |
78 | * We kill the socket, if: | |
79 | * 1. If number of orphaned sockets exceeds an administratively configured | |
80 | * limit. | |
81 | * 2. If we have strong memory pressure. | |
4ee806d5 | 82 | * 3. If our net namespace is exiting. |
1da177e4 | 83 | */ |
b248230c | 84 | static int tcp_out_of_resources(struct sock *sk, bool do_reset) |
1da177e4 LT |
85 | { |
86 | struct tcp_sock *tp = tcp_sk(sk); | |
ad1af0fe | 87 | int shift = 0; |
1da177e4 | 88 | |
e905a9ed | 89 | /* If peer does not open window for long time, or did not transmit |
1da177e4 | 90 | * anything for long time, penalize it. */ |
d635fbe2 | 91 | if ((s32)(tcp_jiffies32 - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset) |
ad1af0fe | 92 | shift++; |
1da177e4 LT |
93 | |
94 | /* If some dubious ICMP arrived, penalize even more. */ | |
95 | if (sk->sk_err_soft) | |
ad1af0fe | 96 | shift++; |
1da177e4 | 97 | |
efcdbf24 | 98 | if (tcp_check_oom(sk, shift)) { |
1da177e4 LT |
99 | /* Catch exceptional cases, when connection requires reset. |
100 | * 1. Last segment was sent recently. */ | |
d635fbe2 | 101 | if ((s32)(tcp_jiffies32 - tp->lsndtime) <= TCP_TIMEWAIT_LEN || |
1da177e4 LT |
102 | /* 2. Window is closed. */ |
103 | (!tp->snd_wnd && !tp->packets_out)) | |
b248230c | 104 | do_reset = true; |
1da177e4 LT |
105 | if (do_reset) |
106 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
107 | tcp_done(sk); | |
02a1d6e7 | 108 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
109 | return 1; |
110 | } | |
4ee806d5 DS |
111 | |
112 | if (!check_net(sock_net(sk))) { | |
113 | /* Not possible to send reset; just close */ | |
114 | tcp_done(sk); | |
115 | return 1; | |
116 | } | |
117 | ||
1da177e4 LT |
118 | return 0; |
119 | } | |
120 | ||
c380d37e RS |
121 | /** |
122 | * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket | |
123 | * @sk: Pointer to the current socket. | |
124 | * @alive: bool, socket alive state | |
125 | */ | |
7533ce30 | 126 | static int tcp_orphan_retries(struct sock *sk, bool alive) |
1da177e4 | 127 | { |
c402d9be | 128 | int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */ |
1da177e4 LT |
129 | |
130 | /* We know from an ICMP that something is wrong. */ | |
131 | if (sk->sk_err_soft && !alive) | |
132 | retries = 0; | |
133 | ||
134 | /* However, if socket sent something recently, select some safe | |
135 | * number of retries. 8 corresponds to >100 seconds with minimal | |
136 | * RTO of 200msec. */ | |
137 | if (retries == 0 && alive) | |
138 | retries = 8; | |
139 | return retries; | |
140 | } | |
141 | ||
ce55dd36 ED |
142 | static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk) |
143 | { | |
d0f36847 ED |
144 | const struct net *net = sock_net(sk); |
145 | int mss; | |
b0f9ca53 | 146 | |
ce55dd36 | 147 | /* Black hole detection */ |
d0f36847 ED |
148 | if (!net->ipv4.sysctl_tcp_mtu_probing) |
149 | return; | |
150 | ||
151 | if (!icsk->icsk_mtup.enabled) { | |
152 | icsk->icsk_mtup.enabled = 1; | |
153 | icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; | |
154 | } else { | |
155 | mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1; | |
156 | mss = min(net->ipv4.sysctl_tcp_base_mss, mss); | |
c04b79b6 | 157 | mss = max(mss, net->ipv4.sysctl_tcp_mtu_probe_floor); |
967c05ae | 158 | mss = max(mss, net->ipv4.sysctl_tcp_min_snd_mss); |
d0f36847 | 159 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); |
ce55dd36 | 160 | } |
d0f36847 | 161 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
ce55dd36 ED |
162 | } |
163 | ||
01a523b0 YC |
164 | static unsigned int tcp_model_timeout(struct sock *sk, |
165 | unsigned int boundary, | |
166 | unsigned int rto_base) | |
167 | { | |
168 | unsigned int linear_backoff_thresh, timeout; | |
169 | ||
170 | linear_backoff_thresh = ilog2(TCP_RTO_MAX / rto_base); | |
171 | if (boundary <= linear_backoff_thresh) | |
172 | timeout = ((2 << boundary) - 1) * rto_base; | |
173 | else | |
174 | timeout = ((2 << linear_backoff_thresh) - 1) * rto_base + | |
175 | (boundary - linear_backoff_thresh) * TCP_RTO_MAX; | |
176 | return jiffies_to_msecs(timeout); | |
177 | } | |
c380d37e RS |
178 | /** |
179 | * retransmits_timed_out() - returns true if this connection has timed out | |
180 | * @sk: The current socket | |
181 | * @boundary: max number of retransmissions | |
182 | * @timeout: A custom timeout value. | |
183 | * If set to 0 the default timeout is calculated and used. | |
184 | * Using TCP_RTO_MIN and the number of unsuccessful retransmits. | |
c380d37e RS |
185 | * |
186 | * The default "timeout" value this function can calculate and use | |
187 | * is equivalent to the timeout of a TCP Connection | |
188 | * after "boundary" unsuccessful, exponentially backed-off | |
ce682ef6 | 189 | * retransmissions with an initial RTO of TCP_RTO_MIN. |
2f7de571 DL |
190 | */ |
191 | static bool retransmits_timed_out(struct sock *sk, | |
dca43c75 | 192 | unsigned int boundary, |
ce682ef6 | 193 | unsigned int timeout) |
2f7de571 | 194 | { |
01a523b0 | 195 | unsigned int start_ts; |
2f7de571 DL |
196 | |
197 | if (!inet_csk(sk)->icsk_retransmits) | |
198 | return false; | |
199 | ||
7ae18975 | 200 | start_ts = tcp_sk(sk)->retrans_stamp; |
01a523b0 YC |
201 | if (likely(timeout == 0)) |
202 | timeout = tcp_model_timeout(sk, boundary, TCP_RTO_MIN); | |
203 | ||
9efdda4e | 204 | return (s32)(tcp_time_stamp(tcp_sk(sk)) - start_ts - timeout) >= 0; |
2f7de571 DL |
205 | } |
206 | ||
1da177e4 LT |
207 | /* A write timeout has occurred. Process the after effects. */ |
208 | static int tcp_write_timeout(struct sock *sk) | |
209 | { | |
5d424d5a | 210 | struct inet_connection_sock *icsk = inet_csk(sk); |
c968601d | 211 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 212 | struct net *net = sock_net(sk); |
a41e8a88 | 213 | bool expired = false, do_reset; |
1da177e4 LT |
214 | int retry_until; |
215 | ||
216 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
c968601d | 217 | if (icsk->icsk_retransmits) { |
b6c6712a | 218 | dst_negative_advice(sk); |
c5715b8f | 219 | } else { |
3acf3ec3 | 220 | sk_rethink_txhash(sk); |
c968601d | 221 | } |
6fa25166 | 222 | retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
ce682ef6 | 223 | expired = icsk->icsk_retransmits >= retry_until; |
1da177e4 | 224 | } else { |
ce682ef6 | 225 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0)) { |
5d424d5a | 226 | /* Black hole detection */ |
ce55dd36 | 227 | tcp_mtu_probing(icsk, sk); |
1da177e4 | 228 | |
b6c6712a | 229 | dst_negative_advice(sk); |
3acf3ec3 LB |
230 | } else { |
231 | sk_rethink_txhash(sk); | |
1da177e4 LT |
232 | } |
233 | ||
c6214a97 | 234 | retry_until = net->ipv4.sysctl_tcp_retries2; |
1da177e4 | 235 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 236 | const bool alive = icsk->icsk_rto < TCP_RTO_MAX; |
e905a9ed | 237 | |
1da177e4 | 238 | retry_until = tcp_orphan_retries(sk, alive); |
6fa12c85 | 239 | do_reset = alive || |
ce682ef6 | 240 | !retransmits_timed_out(sk, retry_until, 0); |
1da177e4 | 241 | |
6fa12c85 | 242 | if (tcp_out_of_resources(sk, do_reset)) |
1da177e4 LT |
243 | return 1; |
244 | } | |
a41e8a88 ED |
245 | } |
246 | if (!expired) | |
ce682ef6 ED |
247 | expired = retransmits_timed_out(sk, retry_until, |
248 | icsk->icsk_user_timeout); | |
7268586b | 249 | tcp_fastopen_active_detect_blackhole(sk, expired); |
f89013f6 LB |
250 | |
251 | if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RTO_CB_FLAG)) | |
252 | tcp_call_bpf_3arg(sk, BPF_SOCK_OPS_RTO_CB, | |
253 | icsk->icsk_retransmits, | |
254 | icsk->icsk_rto, (int)expired); | |
255 | ||
ce682ef6 | 256 | if (expired) { |
1da177e4 LT |
257 | /* Has it gone just too far? */ |
258 | tcp_write_err(sk); | |
259 | return 1; | |
260 | } | |
f89013f6 | 261 | |
1da177e4 LT |
262 | return 0; |
263 | } | |
264 | ||
c10d9310 | 265 | /* Called with BH disabled */ |
6f458dfb | 266 | void tcp_delack_timer_handler(struct sock *sk) |
1da177e4 | 267 | { |
463c84b9 | 268 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 269 | |
9993e7d3 | 270 | sk_mem_reclaim_partial(sk); |
1da177e4 | 271 | |
02b2faaf ED |
272 | if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) || |
273 | !(icsk->icsk_ack.pending & ICSK_ACK_TIMER)) | |
1da177e4 LT |
274 | goto out; |
275 | ||
463c84b9 ACM |
276 | if (time_after(icsk->icsk_ack.timeout, jiffies)) { |
277 | sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout); | |
1da177e4 LT |
278 | goto out; |
279 | } | |
463c84b9 | 280 | icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER; |
1da177e4 | 281 | |
463c84b9 | 282 | if (inet_csk_ack_scheduled(sk)) { |
31954cd8 | 283 | if (!inet_csk_in_pingpong_mode(sk)) { |
1da177e4 | 284 | /* Delayed ACK missed: inflate ATO. */ |
463c84b9 | 285 | icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto); |
1da177e4 LT |
286 | } else { |
287 | /* Delayed ACK missed: leave pingpong mode and | |
288 | * deflate ATO. | |
289 | */ | |
31954cd8 | 290 | inet_csk_exit_pingpong_mode(sk); |
463c84b9 | 291 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
1da177e4 | 292 | } |
4688eb7c | 293 | tcp_mstamp_refresh(tcp_sk(sk)); |
1da177e4 | 294 | tcp_send_ack(sk); |
02a1d6e7 | 295 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS); |
1da177e4 | 296 | } |
1da177e4 LT |
297 | |
298 | out: | |
b8da51eb | 299 | if (tcp_under_memory_pressure(sk)) |
3ab224be | 300 | sk_mem_reclaim(sk); |
6f458dfb ED |
301 | } |
302 | ||
c380d37e RS |
303 | |
304 | /** | |
305 | * tcp_delack_timer() - The TCP delayed ACK timeout handler | |
306 | * @data: Pointer to the current socket. (gets casted to struct sock *) | |
307 | * | |
308 | * This function gets (indirectly) called when the kernel timer for a TCP packet | |
309 | * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work. | |
310 | * | |
311 | * Returns: Nothing (void) | |
312 | */ | |
59f379f9 | 313 | static void tcp_delack_timer(struct timer_list *t) |
6f458dfb | 314 | { |
59f379f9 KC |
315 | struct inet_connection_sock *icsk = |
316 | from_timer(icsk, t, icsk_delack_timer); | |
317 | struct sock *sk = &icsk->icsk_inet.sk; | |
6f458dfb ED |
318 | |
319 | bh_lock_sock(sk); | |
320 | if (!sock_owned_by_user(sk)) { | |
321 | tcp_delack_timer_handler(sk); | |
322 | } else { | |
59f379f9 | 323 | icsk->icsk_ack.blocked = 1; |
02a1d6e7 | 324 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED); |
6f458dfb | 325 | /* deleguate our work to tcp_release_cb() */ |
7aa5470c | 326 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 | 327 | sock_hold(sk); |
6f458dfb | 328 | } |
1da177e4 LT |
329 | bh_unlock_sock(sk); |
330 | sock_put(sk); | |
331 | } | |
332 | ||
333 | static void tcp_probe_timer(struct sock *sk) | |
334 | { | |
6687e988 | 335 | struct inet_connection_sock *icsk = inet_csk(sk); |
75c119af | 336 | struct sk_buff *skb = tcp_send_head(sk); |
1da177e4 LT |
337 | struct tcp_sock *tp = tcp_sk(sk); |
338 | int max_probes; | |
339 | ||
75c119af | 340 | if (tp->packets_out || !skb) { |
6687e988 | 341 | icsk->icsk_probes_out = 0; |
1da177e4 LT |
342 | return; |
343 | } | |
344 | ||
b248230c YC |
345 | /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as |
346 | * long as the receiver continues to respond probes. We support this by | |
347 | * default and reset icsk_probes_out with incoming ACKs. But if the | |
348 | * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we | |
349 | * kill the socket when the retry count and the time exceeds the | |
350 | * corresponding system limit. We also implement similar policy when | |
351 | * we use RTO to probe window in tcp_retransmit_timer(). | |
1da177e4 | 352 | */ |
9721e709 YC |
353 | if (icsk->icsk_user_timeout) { |
354 | u32 elapsed = tcp_model_timeout(sk, icsk->icsk_probes_out, | |
355 | tcp_probe0_base(sk)); | |
356 | ||
357 | if (elapsed >= icsk->icsk_user_timeout) | |
358 | goto abort; | |
359 | } | |
1da177e4 | 360 | |
c6214a97 | 361 | max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2; |
1da177e4 | 362 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 363 | const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX; |
e905a9ed | 364 | |
1da177e4 | 365 | max_probes = tcp_orphan_retries(sk, alive); |
b248230c YC |
366 | if (!alive && icsk->icsk_backoff >= max_probes) |
367 | goto abort; | |
368 | if (tcp_out_of_resources(sk, true)) | |
1da177e4 LT |
369 | return; |
370 | } | |
371 | ||
3976535a | 372 | if (icsk->icsk_probes_out >= max_probes) { |
b248230c | 373 | abort: tcp_write_err(sk); |
1da177e4 LT |
374 | } else { |
375 | /* Only send another probe if we didn't close things up. */ | |
376 | tcp_send_probe0(sk); | |
377 | } | |
378 | } | |
379 | ||
8336886f JC |
380 | /* |
381 | * Timer for Fast Open socket to retransmit SYNACK. Note that the | |
382 | * sk here is the child socket, not the parent (listener) socket. | |
383 | */ | |
384 | static void tcp_fastopen_synack_timer(struct sock *sk) | |
385 | { | |
386 | struct inet_connection_sock *icsk = inet_csk(sk); | |
387 | int max_retries = icsk->icsk_syn_retries ? : | |
7c083ecb | 388 | sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */ |
c7d13c8f | 389 | struct tcp_sock *tp = tcp_sk(sk); |
8336886f JC |
390 | struct request_sock *req; |
391 | ||
392 | req = tcp_sk(sk)->fastopen_rsk; | |
42cb80a2 | 393 | req->rsk_ops->syn_ack_timeout(req); |
8336886f | 394 | |
e6c022a4 | 395 | if (req->num_timeout >= max_retries) { |
8336886f JC |
396 | tcp_write_err(sk); |
397 | return; | |
398 | } | |
8c3cfe19 YC |
399 | /* Lower cwnd after certain SYNACK timeout like tcp_init_transfer() */ |
400 | if (icsk->icsk_retransmits == 1) | |
401 | tcp_enter_loss(sk); | |
8336886f JC |
402 | /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error |
403 | * returned from rtx_syn_ack() to make it more persistent like | |
404 | * regular retransmit because if the child socket has been accepted | |
405 | * it's not good to give up too easily. | |
406 | */ | |
e6c022a4 ED |
407 | inet_rtx_syn_ack(sk, req); |
408 | req->num_timeout++; | |
7e32b443 | 409 | icsk->icsk_retransmits++; |
c7d13c8f YC |
410 | if (!tp->retrans_stamp) |
411 | tp->retrans_stamp = tcp_time_stamp(tp); | |
8336886f | 412 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
e6c022a4 | 413 | TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); |
8336886f JC |
414 | } |
415 | ||
1da177e4 | 416 | |
c380d37e RS |
417 | /** |
418 | * tcp_retransmit_timer() - The TCP retransmit timeout handler | |
419 | * @sk: Pointer to the current socket. | |
420 | * | |
421 | * This function gets called when the kernel timer for a TCP packet | |
422 | * of this socket expires. | |
423 | * | |
424 | * It handles retransmission, timer adjustment and other necesarry measures. | |
425 | * | |
426 | * Returns: Nothing (void) | |
427 | */ | |
f1ecd5d9 | 428 | void tcp_retransmit_timer(struct sock *sk) |
1da177e4 LT |
429 | { |
430 | struct tcp_sock *tp = tcp_sk(sk); | |
ae5c3f40 | 431 | struct net *net = sock_net(sk); |
463c84b9 | 432 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 433 | |
8336886f | 434 | if (tp->fastopen_rsk) { |
37561f68 JC |
435 | WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && |
436 | sk->sk_state != TCP_FIN_WAIT1); | |
8336886f JC |
437 | tcp_fastopen_synack_timer(sk); |
438 | /* Before we receive ACK to our SYN-ACK don't retransmit | |
439 | * anything else (e.g., data or FIN segments). | |
440 | */ | |
441 | return; | |
442 | } | |
88f8598d YC |
443 | if (!tp->packets_out || WARN_ON_ONCE(tcp_rtx_queue_empty(sk))) |
444 | return; | |
1da177e4 | 445 | |
9b717a8d ND |
446 | tp->tlp_high_seq = 0; |
447 | ||
1da177e4 LT |
448 | if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) && |
449 | !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) { | |
450 | /* Receiver dastardly shrinks window. Our retransmits | |
451 | * become zero probes, but we should not timeout this | |
452 | * connection. If the socket is an orphan, time it out, | |
453 | * we cannot allow such beasts to hang infinitely. | |
454 | */ | |
569508c9 YH |
455 | struct inet_sock *inet = inet_sk(sk); |
456 | if (sk->sk_family == AF_INET) { | |
ba7a46f1 JP |
457 | net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
458 | &inet->inet_daddr, | |
459 | ntohs(inet->inet_dport), | |
460 | inet->inet_num, | |
461 | tp->snd_una, tp->snd_nxt); | |
1da177e4 | 462 | } |
dfd56b8b | 463 | #if IS_ENABLED(CONFIG_IPV6) |
569508c9 | 464 | else if (sk->sk_family == AF_INET6) { |
ba7a46f1 JP |
465 | net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
466 | &sk->sk_v6_daddr, | |
467 | ntohs(inet->inet_dport), | |
468 | inet->inet_num, | |
469 | tp->snd_una, tp->snd_nxt); | |
569508c9 | 470 | } |
1da177e4 | 471 | #endif |
70eabf0e | 472 | if (tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX) { |
1da177e4 LT |
473 | tcp_write_err(sk); |
474 | goto out; | |
475 | } | |
5ae344c9 | 476 | tcp_enter_loss(sk); |
75c119af | 477 | tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1); |
1da177e4 LT |
478 | __sk_dst_reset(sk); |
479 | goto out_reset_timer; | |
480 | } | |
481 | ||
e1561fe2 | 482 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTS); |
1da177e4 LT |
483 | if (tcp_write_timeout(sk)) |
484 | goto out; | |
485 | ||
463c84b9 | 486 | if (icsk->icsk_retransmits == 0) { |
e1561fe2 | 487 | int mib_idx = 0; |
40b215e5 | 488 | |
c60ce4e2 | 489 | if (icsk->icsk_ca_state == TCP_CA_Recovery) { |
bc079e9e IJ |
490 | if (tcp_is_sack(tp)) |
491 | mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL; | |
492 | else | |
493 | mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL; | |
6687e988 | 494 | } else if (icsk->icsk_ca_state == TCP_CA_Loss) { |
40b215e5 | 495 | mib_idx = LINUX_MIB_TCPLOSSFAILURES; |
c60ce4e2 IJ |
496 | } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) || |
497 | tp->sacked_out) { | |
498 | if (tcp_is_sack(tp)) | |
499 | mib_idx = LINUX_MIB_TCPSACKFAILURES; | |
500 | else | |
501 | mib_idx = LINUX_MIB_TCPRENOFAILURES; | |
1da177e4 | 502 | } |
e1561fe2 YC |
503 | if (mib_idx) |
504 | __NET_INC_STATS(sock_net(sk), mib_idx); | |
1da177e4 LT |
505 | } |
506 | ||
5ae344c9 | 507 | tcp_enter_loss(sk); |
1da177e4 | 508 | |
590d2026 | 509 | icsk->icsk_retransmits++; |
75c119af | 510 | if (tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1) > 0) { |
1da177e4 | 511 | /* Retransmission failed because of local congestion, |
590d2026 | 512 | * Let senders fight for local resources conservatively. |
1da177e4 | 513 | */ |
463c84b9 | 514 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
590d2026 | 515 | TCP_RESOURCE_PROBE_INTERVAL, |
3f421baa | 516 | TCP_RTO_MAX); |
1da177e4 LT |
517 | goto out; |
518 | } | |
519 | ||
520 | /* Increase the timeout each time we retransmit. Note that | |
521 | * we do not increase the rtt estimate. rto is initialized | |
522 | * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests | |
523 | * that doubling rto each time is the least we can get away with. | |
524 | * In KA9Q, Karn uses this for the first few times, and then | |
525 | * goes to quadratic. netBSD doubles, but only goes up to *64, | |
526 | * and clamps at 1 to 64 sec afterwards. Note that 120 sec is | |
527 | * defined in the protocol as the maximum possible RTT. I guess | |
528 | * we'll have to use something other than TCP to talk to the | |
529 | * University of Mars. | |
530 | * | |
531 | * PAWS allows us longer timeouts and large windows, so once | |
532 | * implemented ftp to mars will work nicely. We will have to fix | |
533 | * the 120 second clamps though! | |
534 | */ | |
463c84b9 | 535 | icsk->icsk_backoff++; |
1da177e4 LT |
536 | |
537 | out_reset_timer: | |
36e31b0a AP |
538 | /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is |
539 | * used to reset timer, set to 0. Recalculate 'icsk_rto' as this | |
540 | * might be increased if the stream oscillates between thin and thick, | |
541 | * thus the old value might already be too high compared to the value | |
542 | * set by 'tcp_set_rto' in tcp_input.c which resets the rto without | |
543 | * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating | |
544 | * exponential backoff behaviour to avoid continue hammering | |
545 | * linear-timeout retransmissions into a black hole | |
546 | */ | |
547 | if (sk->sk_state == TCP_ESTABLISHED && | |
2c04ac8a | 548 | (tp->thin_lto || net->ipv4.sysctl_tcp_thin_linear_timeouts) && |
36e31b0a AP |
549 | tcp_stream_is_thin(tp) && |
550 | icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) { | |
551 | icsk->icsk_backoff = 0; | |
552 | icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX); | |
553 | } else { | |
554 | /* Use normal (exponential) backoff */ | |
555 | icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX); | |
556 | } | |
b701a99e JM |
557 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
558 | tcp_clamp_rto_to_user_timeout(sk), TCP_RTO_MAX); | |
ce682ef6 | 559 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0)) |
1da177e4 LT |
560 | __sk_dst_reset(sk); |
561 | ||
562 | out:; | |
563 | } | |
564 | ||
c380d37e RS |
565 | /* Called with bottom-half processing disabled. |
566 | Called by tcp_write_timer() */ | |
6f458dfb | 567 | void tcp_write_timer_handler(struct sock *sk) |
1da177e4 | 568 | { |
463c84b9 | 569 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
570 | int event; |
571 | ||
02b2faaf ED |
572 | if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) || |
573 | !icsk->icsk_pending) | |
1da177e4 LT |
574 | goto out; |
575 | ||
463c84b9 ACM |
576 | if (time_after(icsk->icsk_timeout, jiffies)) { |
577 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout); | |
1da177e4 LT |
578 | goto out; |
579 | } | |
580 | ||
9a568de4 | 581 | tcp_mstamp_refresh(tcp_sk(sk)); |
463c84b9 | 582 | event = icsk->icsk_pending; |
1da177e4 LT |
583 | |
584 | switch (event) { | |
57dde7f7 YC |
585 | case ICSK_TIME_REO_TIMEOUT: |
586 | tcp_rack_reo_timeout(sk); | |
587 | break; | |
6ba8a3b1 ND |
588 | case ICSK_TIME_LOSS_PROBE: |
589 | tcp_send_loss_probe(sk); | |
590 | break; | |
463c84b9 | 591 | case ICSK_TIME_RETRANS: |
6ba8a3b1 | 592 | icsk->icsk_pending = 0; |
1da177e4 LT |
593 | tcp_retransmit_timer(sk); |
594 | break; | |
463c84b9 | 595 | case ICSK_TIME_PROBE0: |
6ba8a3b1 | 596 | icsk->icsk_pending = 0; |
1da177e4 LT |
597 | tcp_probe_timer(sk); |
598 | break; | |
599 | } | |
1da177e4 LT |
600 | |
601 | out: | |
3ab224be | 602 | sk_mem_reclaim(sk); |
6f458dfb ED |
603 | } |
604 | ||
59f379f9 | 605 | static void tcp_write_timer(struct timer_list *t) |
6f458dfb | 606 | { |
59f379f9 KC |
607 | struct inet_connection_sock *icsk = |
608 | from_timer(icsk, t, icsk_retransmit_timer); | |
609 | struct sock *sk = &icsk->icsk_inet.sk; | |
6f458dfb ED |
610 | |
611 | bh_lock_sock(sk); | |
612 | if (!sock_owned_by_user(sk)) { | |
613 | tcp_write_timer_handler(sk); | |
614 | } else { | |
c380d37e | 615 | /* delegate our work to tcp_release_cb() */ |
7aa5470c | 616 | if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 | 617 | sock_hold(sk); |
6f458dfb | 618 | } |
1da177e4 LT |
619 | bh_unlock_sock(sk); |
620 | sock_put(sk); | |
621 | } | |
622 | ||
42cb80a2 | 623 | void tcp_syn_ack_timeout(const struct request_sock *req) |
72659ecc | 624 | { |
42cb80a2 ED |
625 | struct net *net = read_pnet(&inet_rsk(req)->ireq_net); |
626 | ||
02a1d6e7 | 627 | __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS); |
72659ecc OP |
628 | } |
629 | EXPORT_SYMBOL(tcp_syn_ack_timeout); | |
630 | ||
1da177e4 LT |
631 | void tcp_set_keepalive(struct sock *sk, int val) |
632 | { | |
633 | if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) | |
634 | return; | |
635 | ||
636 | if (val && !sock_flag(sk, SOCK_KEEPOPEN)) | |
463c84b9 | 637 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk))); |
1da177e4 | 638 | else if (!val) |
463c84b9 | 639 | inet_csk_delete_keepalive_timer(sk); |
1da177e4 | 640 | } |
4b9d07a4 | 641 | EXPORT_SYMBOL_GPL(tcp_set_keepalive); |
1da177e4 LT |
642 | |
643 | ||
59f379f9 | 644 | static void tcp_keepalive_timer (struct timer_list *t) |
1da177e4 | 645 | { |
59f379f9 | 646 | struct sock *sk = from_timer(sk, t, sk_timer); |
6687e988 | 647 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 648 | struct tcp_sock *tp = tcp_sk(sk); |
6c37e5de | 649 | u32 elapsed; |
1da177e4 LT |
650 | |
651 | /* Only process if socket is not in use. */ | |
652 | bh_lock_sock(sk); | |
653 | if (sock_owned_by_user(sk)) { | |
e905a9ed | 654 | /* Try again later. */ |
463c84b9 | 655 | inet_csk_reset_keepalive_timer (sk, HZ/20); |
1da177e4 LT |
656 | goto out; |
657 | } | |
658 | ||
659 | if (sk->sk_state == TCP_LISTEN) { | |
fa76ce73 | 660 | pr_err("Hmm... keepalive on a LISTEN ???\n"); |
1da177e4 LT |
661 | goto out; |
662 | } | |
663 | ||
4688eb7c | 664 | tcp_mstamp_refresh(tp); |
1da177e4 LT |
665 | if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) { |
666 | if (tp->linger2 >= 0) { | |
463c84b9 | 667 | const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN; |
1da177e4 LT |
668 | |
669 | if (tmo > 0) { | |
670 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); | |
671 | goto out; | |
672 | } | |
673 | } | |
674 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
675 | goto death; | |
676 | } | |
677 | ||
2dda6400 ED |
678 | if (!sock_flag(sk, SOCK_KEEPOPEN) || |
679 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT))) | |
1da177e4 LT |
680 | goto out; |
681 | ||
682 | elapsed = keepalive_time_when(tp); | |
683 | ||
684 | /* It is alive without keepalive 8) */ | |
75c119af | 685 | if (tp->packets_out || !tcp_write_queue_empty(sk)) |
1da177e4 LT |
686 | goto resched; |
687 | ||
6c37e5de | 688 | elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
689 | |
690 | if (elapsed >= keepalive_time_when(tp)) { | |
dca43c75 JC |
691 | /* If the TCP_USER_TIMEOUT option is enabled, use that |
692 | * to determine when to timeout instead. | |
693 | */ | |
694 | if ((icsk->icsk_user_timeout != 0 && | |
9bcc66e1 | 695 | elapsed >= msecs_to_jiffies(icsk->icsk_user_timeout) && |
dca43c75 JC |
696 | icsk->icsk_probes_out > 0) || |
697 | (icsk->icsk_user_timeout == 0 && | |
698 | icsk->icsk_probes_out >= keepalive_probes(tp))) { | |
1da177e4 LT |
699 | tcp_send_active_reset(sk, GFP_ATOMIC); |
700 | tcp_write_err(sk); | |
701 | goto out; | |
702 | } | |
e520af48 | 703 | if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) { |
6687e988 | 704 | icsk->icsk_probes_out++; |
1da177e4 LT |
705 | elapsed = keepalive_intvl_when(tp); |
706 | } else { | |
707 | /* If keepalive was lost due to local congestion, | |
708 | * try harder. | |
709 | */ | |
710 | elapsed = TCP_RESOURCE_PROBE_INTERVAL; | |
711 | } | |
712 | } else { | |
713 | /* It is tp->rcv_tstamp + keepalive_time_when(tp) */ | |
714 | elapsed = keepalive_time_when(tp) - elapsed; | |
715 | } | |
716 | ||
3ab224be | 717 | sk_mem_reclaim(sk); |
1da177e4 LT |
718 | |
719 | resched: | |
463c84b9 | 720 | inet_csk_reset_keepalive_timer (sk, elapsed); |
1da177e4 LT |
721 | goto out; |
722 | ||
e905a9ed | 723 | death: |
1da177e4 LT |
724 | tcp_done(sk); |
725 | ||
726 | out: | |
727 | bh_unlock_sock(sk); | |
728 | sock_put(sk); | |
729 | } | |
6f458dfb | 730 | |
5d9f4262 ED |
731 | static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer) |
732 | { | |
733 | struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer); | |
734 | struct sock *sk = (struct sock *)tp; | |
735 | ||
736 | bh_lock_sock(sk); | |
737 | if (!sock_owned_by_user(sk)) { | |
86de5921 | 738 | if (tp->compressed_ack > TCP_FASTRETRANS_THRESH) |
5d9f4262 ED |
739 | tcp_send_ack(sk); |
740 | } else { | |
741 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, | |
742 | &sk->sk_tsq_flags)) | |
743 | sock_hold(sk); | |
744 | } | |
745 | bh_unlock_sock(sk); | |
746 | ||
747 | sock_put(sk); | |
748 | ||
749 | return HRTIMER_NORESTART; | |
750 | } | |
751 | ||
6f458dfb ED |
752 | void tcp_init_xmit_timers(struct sock *sk) |
753 | { | |
754 | inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer, | |
755 | &tcp_keepalive_timer); | |
fb420d5d | 756 | hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC, |
73a6bab5 | 757 | HRTIMER_MODE_ABS_PINNED_SOFT); |
218af599 | 758 | tcp_sk(sk)->pacing_timer.function = tcp_pace_kick; |
5d9f4262 ED |
759 | |
760 | hrtimer_init(&tcp_sk(sk)->compressed_ack_timer, CLOCK_MONOTONIC, | |
761 | HRTIMER_MODE_REL_PINNED_SOFT); | |
762 | tcp_sk(sk)->compressed_ack_timer.function = tcp_compressed_ack_kick; | |
6f458dfb | 763 | } |