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 | ||
344db93a EC |
43 | u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when) |
44 | { | |
45 | struct inet_connection_sock *icsk = inet_csk(sk); | |
46 | u32 remaining; | |
47 | s32 elapsed; | |
48 | ||
49 | if (!icsk->icsk_user_timeout || !icsk->icsk_probes_tstamp) | |
50 | return when; | |
51 | ||
52 | elapsed = tcp_jiffies32 - icsk->icsk_probes_tstamp; | |
53 | if (unlikely(elapsed < 0)) | |
54 | elapsed = 0; | |
55 | remaining = msecs_to_jiffies(icsk->icsk_user_timeout) - elapsed; | |
56 | remaining = max_t(u32, remaining, TCP_TIMEOUT_MIN); | |
57 | ||
58 | return min_t(u32, remaining, when); | |
59 | } | |
60 | ||
c380d37e RS |
61 | /** |
62 | * tcp_write_err() - close socket and save error info | |
63 | * @sk: The socket the error has appeared on. | |
64 | * | |
65 | * Returns: Nothing (void) | |
66 | */ | |
67 | ||
1da177e4 LT |
68 | static void tcp_write_err(struct sock *sk) |
69 | { | |
70 | sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT; | |
e3ae2365 | 71 | sk_error_report(sk); |
1da177e4 | 72 | |
e05836ac | 73 | tcp_write_queue_purge(sk); |
1da177e4 | 74 | tcp_done(sk); |
02a1d6e7 | 75 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT); |
1da177e4 LT |
76 | } |
77 | ||
c380d37e RS |
78 | /** |
79 | * tcp_out_of_resources() - Close socket if out of resources | |
80 | * @sk: pointer to current socket | |
81 | * @do_reset: send a last packet with reset flag | |
1da177e4 | 82 | * |
c380d37e RS |
83 | * Do not allow orphaned sockets to eat all our resources. |
84 | * This is direct violation of TCP specs, but it is required | |
85 | * to prevent DoS attacks. It is called when a retransmission timeout | |
86 | * or zero probe timeout occurs on orphaned socket. | |
87 | * | |
4ee806d5 DS |
88 | * Also close if our net namespace is exiting; in that case there is no |
89 | * hope of ever communicating again since all netns interfaces are already | |
90 | * down (or about to be down), and we need to release our dst references, | |
91 | * which have been moved to the netns loopback interface, so the namespace | |
92 | * can finish exiting. This condition is only possible if we are a kernel | |
93 | * socket, as those do not hold references to the namespace. | |
94 | * | |
c380d37e RS |
95 | * Criteria is still not confirmed experimentally and may change. |
96 | * We kill the socket, if: | |
97 | * 1. If number of orphaned sockets exceeds an administratively configured | |
98 | * limit. | |
99 | * 2. If we have strong memory pressure. | |
4ee806d5 | 100 | * 3. If our net namespace is exiting. |
1da177e4 | 101 | */ |
b248230c | 102 | static int tcp_out_of_resources(struct sock *sk, bool do_reset) |
1da177e4 LT |
103 | { |
104 | struct tcp_sock *tp = tcp_sk(sk); | |
ad1af0fe | 105 | int shift = 0; |
1da177e4 | 106 | |
e905a9ed | 107 | /* If peer does not open window for long time, or did not transmit |
1da177e4 | 108 | * anything for long time, penalize it. */ |
d635fbe2 | 109 | if ((s32)(tcp_jiffies32 - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset) |
ad1af0fe | 110 | shift++; |
1da177e4 LT |
111 | |
112 | /* If some dubious ICMP arrived, penalize even more. */ | |
113 | if (sk->sk_err_soft) | |
ad1af0fe | 114 | shift++; |
1da177e4 | 115 | |
efcdbf24 | 116 | if (tcp_check_oom(sk, shift)) { |
1da177e4 LT |
117 | /* Catch exceptional cases, when connection requires reset. |
118 | * 1. Last segment was sent recently. */ | |
d635fbe2 | 119 | if ((s32)(tcp_jiffies32 - tp->lsndtime) <= TCP_TIMEWAIT_LEN || |
1da177e4 LT |
120 | /* 2. Window is closed. */ |
121 | (!tp->snd_wnd && !tp->packets_out)) | |
b248230c | 122 | do_reset = true; |
1da177e4 LT |
123 | if (do_reset) |
124 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
125 | tcp_done(sk); | |
02a1d6e7 | 126 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY); |
1da177e4 LT |
127 | return 1; |
128 | } | |
4ee806d5 DS |
129 | |
130 | if (!check_net(sock_net(sk))) { | |
131 | /* Not possible to send reset; just close */ | |
132 | tcp_done(sk); | |
133 | return 1; | |
134 | } | |
135 | ||
1da177e4 LT |
136 | return 0; |
137 | } | |
138 | ||
c380d37e RS |
139 | /** |
140 | * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket | |
141 | * @sk: Pointer to the current socket. | |
142 | * @alive: bool, socket alive state | |
143 | */ | |
7533ce30 | 144 | static int tcp_orphan_retries(struct sock *sk, bool alive) |
1da177e4 | 145 | { |
c402d9be | 146 | int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */ |
1da177e4 LT |
147 | |
148 | /* We know from an ICMP that something is wrong. */ | |
149 | if (sk->sk_err_soft && !alive) | |
150 | retries = 0; | |
151 | ||
152 | /* However, if socket sent something recently, select some safe | |
153 | * number of retries. 8 corresponds to >100 seconds with minimal | |
154 | * RTO of 200msec. */ | |
155 | if (retries == 0 && alive) | |
156 | retries = 8; | |
157 | return retries; | |
158 | } | |
159 | ||
ce55dd36 ED |
160 | static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk) |
161 | { | |
d0f36847 ED |
162 | const struct net *net = sock_net(sk); |
163 | int mss; | |
b0f9ca53 | 164 | |
ce55dd36 | 165 | /* Black hole detection */ |
f47d00e0 | 166 | if (!READ_ONCE(net->ipv4.sysctl_tcp_mtu_probing)) |
d0f36847 ED |
167 | return; |
168 | ||
169 | if (!icsk->icsk_mtup.enabled) { | |
170 | icsk->icsk_mtup.enabled = 1; | |
171 | icsk->icsk_mtup.probe_timestamp = tcp_jiffies32; | |
172 | } else { | |
173 | mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1; | |
88d78bc0 | 174 | mss = min(READ_ONCE(net->ipv4.sysctl_tcp_base_mss), mss); |
c04b79b6 | 175 | mss = max(mss, net->ipv4.sysctl_tcp_mtu_probe_floor); |
967c05ae | 176 | mss = max(mss, net->ipv4.sysctl_tcp_min_snd_mss); |
d0f36847 | 177 | icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss); |
ce55dd36 | 178 | } |
d0f36847 | 179 | tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); |
ce55dd36 ED |
180 | } |
181 | ||
01a523b0 YC |
182 | static unsigned int tcp_model_timeout(struct sock *sk, |
183 | unsigned int boundary, | |
184 | unsigned int rto_base) | |
185 | { | |
186 | unsigned int linear_backoff_thresh, timeout; | |
187 | ||
188 | linear_backoff_thresh = ilog2(TCP_RTO_MAX / rto_base); | |
189 | if (boundary <= linear_backoff_thresh) | |
190 | timeout = ((2 << boundary) - 1) * rto_base; | |
191 | else | |
192 | timeout = ((2 << linear_backoff_thresh) - 1) * rto_base + | |
193 | (boundary - linear_backoff_thresh) * TCP_RTO_MAX; | |
194 | return jiffies_to_msecs(timeout); | |
195 | } | |
c380d37e RS |
196 | /** |
197 | * retransmits_timed_out() - returns true if this connection has timed out | |
198 | * @sk: The current socket | |
199 | * @boundary: max number of retransmissions | |
200 | * @timeout: A custom timeout value. | |
201 | * If set to 0 the default timeout is calculated and used. | |
202 | * Using TCP_RTO_MIN and the number of unsuccessful retransmits. | |
c380d37e RS |
203 | * |
204 | * The default "timeout" value this function can calculate and use | |
205 | * is equivalent to the timeout of a TCP Connection | |
206 | * after "boundary" unsuccessful, exponentially backed-off | |
ce682ef6 | 207 | * retransmissions with an initial RTO of TCP_RTO_MIN. |
2f7de571 DL |
208 | */ |
209 | static bool retransmits_timed_out(struct sock *sk, | |
dca43c75 | 210 | unsigned int boundary, |
ce682ef6 | 211 | unsigned int timeout) |
2f7de571 | 212 | { |
01a523b0 | 213 | unsigned int start_ts; |
2f7de571 DL |
214 | |
215 | if (!inet_csk(sk)->icsk_retransmits) | |
216 | return false; | |
217 | ||
7ae18975 | 218 | start_ts = tcp_sk(sk)->retrans_stamp; |
3256a2d6 ED |
219 | if (likely(timeout == 0)) { |
220 | unsigned int rto_base = TCP_RTO_MIN; | |
221 | ||
222 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) | |
223 | rto_base = tcp_timeout_init(sk); | |
224 | timeout = tcp_model_timeout(sk, boundary, rto_base); | |
225 | } | |
01a523b0 | 226 | |
9efdda4e | 227 | return (s32)(tcp_time_stamp(tcp_sk(sk)) - start_ts - timeout) >= 0; |
2f7de571 DL |
228 | } |
229 | ||
1da177e4 LT |
230 | /* A write timeout has occurred. Process the after effects. */ |
231 | static int tcp_write_timeout(struct sock *sk) | |
232 | { | |
5d424d5a | 233 | struct inet_connection_sock *icsk = inet_csk(sk); |
c968601d | 234 | struct tcp_sock *tp = tcp_sk(sk); |
6fa25166 | 235 | struct net *net = sock_net(sk); |
a41e8a88 | 236 | bool expired = false, do_reset; |
1da177e4 LT |
237 | int retry_until; |
238 | ||
239 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
9c30ae83 YC |
240 | if (icsk->icsk_retransmits) |
241 | __dst_negative_advice(sk); | |
6fa25166 | 242 | retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries; |
ce682ef6 | 243 | expired = icsk->icsk_retransmits >= retry_until; |
1da177e4 | 244 | } else { |
ce682ef6 | 245 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0)) { |
5d424d5a | 246 | /* Black hole detection */ |
ce55dd36 | 247 | tcp_mtu_probing(icsk, sk); |
1da177e4 | 248 | |
9c30ae83 | 249 | __dst_negative_advice(sk); |
1da177e4 LT |
250 | } |
251 | ||
c6214a97 | 252 | retry_until = net->ipv4.sysctl_tcp_retries2; |
1da177e4 | 253 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 254 | const bool alive = icsk->icsk_rto < TCP_RTO_MAX; |
e905a9ed | 255 | |
1da177e4 | 256 | retry_until = tcp_orphan_retries(sk, alive); |
6fa12c85 | 257 | do_reset = alive || |
ce682ef6 | 258 | !retransmits_timed_out(sk, retry_until, 0); |
1da177e4 | 259 | |
6fa12c85 | 260 | if (tcp_out_of_resources(sk, do_reset)) |
1da177e4 LT |
261 | return 1; |
262 | } | |
a41e8a88 ED |
263 | } |
264 | if (!expired) | |
ce682ef6 ED |
265 | expired = retransmits_timed_out(sk, retry_until, |
266 | icsk->icsk_user_timeout); | |
7268586b | 267 | tcp_fastopen_active_detect_blackhole(sk, expired); |
f89013f6 LB |
268 | |
269 | if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RTO_CB_FLAG)) | |
270 | tcp_call_bpf_3arg(sk, BPF_SOCK_OPS_RTO_CB, | |
271 | icsk->icsk_retransmits, | |
272 | icsk->icsk_rto, (int)expired); | |
273 | ||
ce682ef6 | 274 | if (expired) { |
1da177e4 LT |
275 | /* Has it gone just too far? */ |
276 | tcp_write_err(sk); | |
277 | return 1; | |
278 | } | |
f89013f6 | 279 | |
9c30ae83 YC |
280 | if (sk_rethink_txhash(sk)) { |
281 | tp->timeout_rehash++; | |
282 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTREHASH); | |
283 | } | |
284 | ||
1da177e4 LT |
285 | return 0; |
286 | } | |
287 | ||
c10d9310 | 288 | /* Called with BH disabled */ |
6f458dfb | 289 | void tcp_delack_timer_handler(struct sock *sk) |
1da177e4 | 290 | { |
463c84b9 | 291 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 292 | |
9993e7d3 | 293 | sk_mem_reclaim_partial(sk); |
1da177e4 | 294 | |
02b2faaf ED |
295 | if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) || |
296 | !(icsk->icsk_ack.pending & ICSK_ACK_TIMER)) | |
1da177e4 LT |
297 | goto out; |
298 | ||
463c84b9 ACM |
299 | if (time_after(icsk->icsk_ack.timeout, jiffies)) { |
300 | sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout); | |
1da177e4 LT |
301 | goto out; |
302 | } | |
463c84b9 | 303 | icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER; |
1da177e4 | 304 | |
463c84b9 | 305 | if (inet_csk_ack_scheduled(sk)) { |
31954cd8 | 306 | if (!inet_csk_in_pingpong_mode(sk)) { |
1da177e4 | 307 | /* Delayed ACK missed: inflate ATO. */ |
463c84b9 | 308 | icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto); |
1da177e4 LT |
309 | } else { |
310 | /* Delayed ACK missed: leave pingpong mode and | |
311 | * deflate ATO. | |
312 | */ | |
31954cd8 | 313 | inet_csk_exit_pingpong_mode(sk); |
463c84b9 | 314 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
1da177e4 | 315 | } |
4688eb7c | 316 | tcp_mstamp_refresh(tcp_sk(sk)); |
1da177e4 | 317 | tcp_send_ack(sk); |
02a1d6e7 | 318 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS); |
1da177e4 | 319 | } |
1da177e4 LT |
320 | |
321 | out: | |
b8da51eb | 322 | if (tcp_under_memory_pressure(sk)) |
3ab224be | 323 | sk_mem_reclaim(sk); |
6f458dfb ED |
324 | } |
325 | ||
c380d37e RS |
326 | |
327 | /** | |
328 | * tcp_delack_timer() - The TCP delayed ACK timeout handler | |
3628e3cb | 329 | * @t: Pointer to the timer. (gets casted to struct sock *) |
c380d37e RS |
330 | * |
331 | * This function gets (indirectly) called when the kernel timer for a TCP packet | |
332 | * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work. | |
333 | * | |
334 | * Returns: Nothing (void) | |
335 | */ | |
59f379f9 | 336 | static void tcp_delack_timer(struct timer_list *t) |
6f458dfb | 337 | { |
59f379f9 KC |
338 | struct inet_connection_sock *icsk = |
339 | from_timer(icsk, t, icsk_delack_timer); | |
340 | struct sock *sk = &icsk->icsk_inet.sk; | |
6f458dfb ED |
341 | |
342 | bh_lock_sock(sk); | |
343 | if (!sock_owned_by_user(sk)) { | |
344 | tcp_delack_timer_handler(sk); | |
345 | } else { | |
02a1d6e7 | 346 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED); |
6f458dfb | 347 | /* deleguate our work to tcp_release_cb() */ |
7aa5470c | 348 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 | 349 | sock_hold(sk); |
6f458dfb | 350 | } |
1da177e4 LT |
351 | bh_unlock_sock(sk); |
352 | sock_put(sk); | |
353 | } | |
354 | ||
355 | static void tcp_probe_timer(struct sock *sk) | |
356 | { | |
6687e988 | 357 | struct inet_connection_sock *icsk = inet_csk(sk); |
75c119af | 358 | struct sk_buff *skb = tcp_send_head(sk); |
1da177e4 LT |
359 | struct tcp_sock *tp = tcp_sk(sk); |
360 | int max_probes; | |
361 | ||
75c119af | 362 | if (tp->packets_out || !skb) { |
6687e988 | 363 | icsk->icsk_probes_out = 0; |
9d9b1ee0 | 364 | icsk->icsk_probes_tstamp = 0; |
1da177e4 LT |
365 | return; |
366 | } | |
367 | ||
b248230c YC |
368 | /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as |
369 | * long as the receiver continues to respond probes. We support this by | |
370 | * default and reset icsk_probes_out with incoming ACKs. But if the | |
371 | * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we | |
372 | * kill the socket when the retry count and the time exceeds the | |
373 | * corresponding system limit. We also implement similar policy when | |
374 | * we use RTO to probe window in tcp_retransmit_timer(). | |
1da177e4 | 375 | */ |
9d9b1ee0 EC |
376 | if (!icsk->icsk_probes_tstamp) |
377 | icsk->icsk_probes_tstamp = tcp_jiffies32; | |
378 | else if (icsk->icsk_user_timeout && | |
379 | (s32)(tcp_jiffies32 - icsk->icsk_probes_tstamp) >= | |
380 | msecs_to_jiffies(icsk->icsk_user_timeout)) | |
381 | goto abort; | |
1da177e4 | 382 | |
c6214a97 | 383 | max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2; |
1da177e4 | 384 | if (sock_flag(sk, SOCK_DEAD)) { |
7533ce30 | 385 | const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX; |
e905a9ed | 386 | |
1da177e4 | 387 | max_probes = tcp_orphan_retries(sk, alive); |
b248230c YC |
388 | if (!alive && icsk->icsk_backoff >= max_probes) |
389 | goto abort; | |
390 | if (tcp_out_of_resources(sk, true)) | |
1da177e4 LT |
391 | return; |
392 | } | |
393 | ||
3976535a | 394 | if (icsk->icsk_probes_out >= max_probes) { |
b248230c | 395 | abort: tcp_write_err(sk); |
1da177e4 LT |
396 | } else { |
397 | /* Only send another probe if we didn't close things up. */ | |
398 | tcp_send_probe0(sk); | |
399 | } | |
400 | } | |
401 | ||
8336886f JC |
402 | /* |
403 | * Timer for Fast Open socket to retransmit SYNACK. Note that the | |
404 | * sk here is the child socket, not the parent (listener) socket. | |
405 | */ | |
d983ea6f | 406 | static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req) |
8336886f JC |
407 | { |
408 | struct inet_connection_sock *icsk = inet_csk(sk); | |
409 | int max_retries = icsk->icsk_syn_retries ? : | |
7c083ecb | 410 | sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */ |
c7d13c8f | 411 | struct tcp_sock *tp = tcp_sk(sk); |
8336886f | 412 | |
42cb80a2 | 413 | req->rsk_ops->syn_ack_timeout(req); |
8336886f | 414 | |
e6c022a4 | 415 | if (req->num_timeout >= max_retries) { |
8336886f JC |
416 | tcp_write_err(sk); |
417 | return; | |
418 | } | |
8c3cfe19 YC |
419 | /* Lower cwnd after certain SYNACK timeout like tcp_init_transfer() */ |
420 | if (icsk->icsk_retransmits == 1) | |
421 | tcp_enter_loss(sk); | |
8336886f JC |
422 | /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error |
423 | * returned from rtx_syn_ack() to make it more persistent like | |
424 | * regular retransmit because if the child socket has been accepted | |
425 | * it's not good to give up too easily. | |
426 | */ | |
e6c022a4 ED |
427 | inet_rtx_syn_ack(sk, req); |
428 | req->num_timeout++; | |
7e32b443 | 429 | icsk->icsk_retransmits++; |
c7d13c8f YC |
430 | if (!tp->retrans_stamp) |
431 | tp->retrans_stamp = tcp_time_stamp(tp); | |
8336886f | 432 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
e6c022a4 | 433 | TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); |
8336886f JC |
434 | } |
435 | ||
1da177e4 | 436 | |
c380d37e RS |
437 | /** |
438 | * tcp_retransmit_timer() - The TCP retransmit timeout handler | |
439 | * @sk: Pointer to the current socket. | |
440 | * | |
441 | * This function gets called when the kernel timer for a TCP packet | |
442 | * of this socket expires. | |
443 | * | |
974d8f86 | 444 | * It handles retransmission, timer adjustment and other necessary measures. |
c380d37e RS |
445 | * |
446 | * Returns: Nothing (void) | |
447 | */ | |
f1ecd5d9 | 448 | void tcp_retransmit_timer(struct sock *sk) |
1da177e4 LT |
449 | { |
450 | struct tcp_sock *tp = tcp_sk(sk); | |
ae5c3f40 | 451 | struct net *net = sock_net(sk); |
463c84b9 | 452 | struct inet_connection_sock *icsk = inet_csk(sk); |
d983ea6f | 453 | struct request_sock *req; |
0d580fbd | 454 | struct sk_buff *skb; |
1da177e4 | 455 | |
d983ea6f ED |
456 | req = rcu_dereference_protected(tp->fastopen_rsk, |
457 | lockdep_sock_is_held(sk)); | |
458 | if (req) { | |
37561f68 JC |
459 | WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && |
460 | sk->sk_state != TCP_FIN_WAIT1); | |
d983ea6f | 461 | tcp_fastopen_synack_timer(sk, req); |
8336886f JC |
462 | /* Before we receive ACK to our SYN-ACK don't retransmit |
463 | * anything else (e.g., data or FIN segments). | |
464 | */ | |
465 | return; | |
466 | } | |
0d580fbd ED |
467 | |
468 | if (!tp->packets_out) | |
469 | return; | |
470 | ||
471 | skb = tcp_rtx_queue_head(sk); | |
472 | if (WARN_ON_ONCE(!skb)) | |
88f8598d | 473 | return; |
1da177e4 | 474 | |
9b717a8d ND |
475 | tp->tlp_high_seq = 0; |
476 | ||
1da177e4 LT |
477 | if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) && |
478 | !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) { | |
479 | /* Receiver dastardly shrinks window. Our retransmits | |
480 | * become zero probes, but we should not timeout this | |
481 | * connection. If the socket is an orphan, time it out, | |
482 | * we cannot allow such beasts to hang infinitely. | |
483 | */ | |
569508c9 YH |
484 | struct inet_sock *inet = inet_sk(sk); |
485 | if (sk->sk_family == AF_INET) { | |
ba7a46f1 JP |
486 | net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
487 | &inet->inet_daddr, | |
488 | ntohs(inet->inet_dport), | |
489 | inet->inet_num, | |
490 | tp->snd_una, tp->snd_nxt); | |
1da177e4 | 491 | } |
dfd56b8b | 492 | #if IS_ENABLED(CONFIG_IPV6) |
569508c9 | 493 | else if (sk->sk_family == AF_INET6) { |
ba7a46f1 JP |
494 | net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n", |
495 | &sk->sk_v6_daddr, | |
496 | ntohs(inet->inet_dport), | |
497 | inet->inet_num, | |
498 | tp->snd_una, tp->snd_nxt); | |
569508c9 | 499 | } |
1da177e4 | 500 | #endif |
70eabf0e | 501 | if (tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX) { |
1da177e4 LT |
502 | tcp_write_err(sk); |
503 | goto out; | |
504 | } | |
5ae344c9 | 505 | tcp_enter_loss(sk); |
0d580fbd | 506 | tcp_retransmit_skb(sk, skb, 1); |
1da177e4 LT |
507 | __sk_dst_reset(sk); |
508 | goto out_reset_timer; | |
509 | } | |
510 | ||
e1561fe2 | 511 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTS); |
1da177e4 LT |
512 | if (tcp_write_timeout(sk)) |
513 | goto out; | |
514 | ||
463c84b9 | 515 | if (icsk->icsk_retransmits == 0) { |
e1561fe2 | 516 | int mib_idx = 0; |
40b215e5 | 517 | |
c60ce4e2 | 518 | if (icsk->icsk_ca_state == TCP_CA_Recovery) { |
bc079e9e IJ |
519 | if (tcp_is_sack(tp)) |
520 | mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL; | |
521 | else | |
522 | mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL; | |
6687e988 | 523 | } else if (icsk->icsk_ca_state == TCP_CA_Loss) { |
40b215e5 | 524 | mib_idx = LINUX_MIB_TCPLOSSFAILURES; |
c60ce4e2 IJ |
525 | } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) || |
526 | tp->sacked_out) { | |
527 | if (tcp_is_sack(tp)) | |
528 | mib_idx = LINUX_MIB_TCPSACKFAILURES; | |
529 | else | |
530 | mib_idx = LINUX_MIB_TCPRENOFAILURES; | |
1da177e4 | 531 | } |
e1561fe2 YC |
532 | if (mib_idx) |
533 | __NET_INC_STATS(sock_net(sk), mib_idx); | |
1da177e4 LT |
534 | } |
535 | ||
5ae344c9 | 536 | tcp_enter_loss(sk); |
1da177e4 | 537 | |
590d2026 | 538 | icsk->icsk_retransmits++; |
75c119af | 539 | if (tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1) > 0) { |
1da177e4 | 540 | /* Retransmission failed because of local congestion, |
590d2026 | 541 | * Let senders fight for local resources conservatively. |
1da177e4 | 542 | */ |
463c84b9 | 543 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
590d2026 | 544 | TCP_RESOURCE_PROBE_INTERVAL, |
3f421baa | 545 | TCP_RTO_MAX); |
1da177e4 LT |
546 | goto out; |
547 | } | |
548 | ||
549 | /* Increase the timeout each time we retransmit. Note that | |
550 | * we do not increase the rtt estimate. rto is initialized | |
551 | * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests | |
552 | * that doubling rto each time is the least we can get away with. | |
553 | * In KA9Q, Karn uses this for the first few times, and then | |
554 | * goes to quadratic. netBSD doubles, but only goes up to *64, | |
555 | * and clamps at 1 to 64 sec afterwards. Note that 120 sec is | |
556 | * defined in the protocol as the maximum possible RTT. I guess | |
557 | * we'll have to use something other than TCP to talk to the | |
558 | * University of Mars. | |
559 | * | |
560 | * PAWS allows us longer timeouts and large windows, so once | |
561 | * implemented ftp to mars will work nicely. We will have to fix | |
562 | * the 120 second clamps though! | |
563 | */ | |
463c84b9 | 564 | icsk->icsk_backoff++; |
1da177e4 LT |
565 | |
566 | out_reset_timer: | |
36e31b0a AP |
567 | /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is |
568 | * used to reset timer, set to 0. Recalculate 'icsk_rto' as this | |
569 | * might be increased if the stream oscillates between thin and thick, | |
570 | * thus the old value might already be too high compared to the value | |
571 | * set by 'tcp_set_rto' in tcp_input.c which resets the rto without | |
572 | * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating | |
573 | * exponential backoff behaviour to avoid continue hammering | |
574 | * linear-timeout retransmissions into a black hole | |
575 | */ | |
576 | if (sk->sk_state == TCP_ESTABLISHED && | |
2c04ac8a | 577 | (tp->thin_lto || net->ipv4.sysctl_tcp_thin_linear_timeouts) && |
36e31b0a AP |
578 | tcp_stream_is_thin(tp) && |
579 | icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) { | |
580 | icsk->icsk_backoff = 0; | |
581 | icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX); | |
582 | } else { | |
583 | /* Use normal (exponential) backoff */ | |
584 | icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX); | |
585 | } | |
b701a99e JM |
586 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
587 | tcp_clamp_rto_to_user_timeout(sk), TCP_RTO_MAX); | |
ce682ef6 | 588 | if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0)) |
1da177e4 LT |
589 | __sk_dst_reset(sk); |
590 | ||
591 | out:; | |
592 | } | |
593 | ||
c380d37e RS |
594 | /* Called with bottom-half processing disabled. |
595 | Called by tcp_write_timer() */ | |
6f458dfb | 596 | void tcp_write_timer_handler(struct sock *sk) |
1da177e4 | 597 | { |
463c84b9 | 598 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
599 | int event; |
600 | ||
02b2faaf ED |
601 | if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) || |
602 | !icsk->icsk_pending) | |
1da177e4 LT |
603 | goto out; |
604 | ||
463c84b9 ACM |
605 | if (time_after(icsk->icsk_timeout, jiffies)) { |
606 | sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout); | |
1da177e4 LT |
607 | goto out; |
608 | } | |
609 | ||
9a568de4 | 610 | tcp_mstamp_refresh(tcp_sk(sk)); |
463c84b9 | 611 | event = icsk->icsk_pending; |
1da177e4 LT |
612 | |
613 | switch (event) { | |
57dde7f7 YC |
614 | case ICSK_TIME_REO_TIMEOUT: |
615 | tcp_rack_reo_timeout(sk); | |
616 | break; | |
6ba8a3b1 ND |
617 | case ICSK_TIME_LOSS_PROBE: |
618 | tcp_send_loss_probe(sk); | |
619 | break; | |
463c84b9 | 620 | case ICSK_TIME_RETRANS: |
6ba8a3b1 | 621 | icsk->icsk_pending = 0; |
1da177e4 LT |
622 | tcp_retransmit_timer(sk); |
623 | break; | |
463c84b9 | 624 | case ICSK_TIME_PROBE0: |
6ba8a3b1 | 625 | icsk->icsk_pending = 0; |
1da177e4 LT |
626 | tcp_probe_timer(sk); |
627 | break; | |
628 | } | |
1da177e4 LT |
629 | |
630 | out: | |
3ab224be | 631 | sk_mem_reclaim(sk); |
6f458dfb ED |
632 | } |
633 | ||
59f379f9 | 634 | static void tcp_write_timer(struct timer_list *t) |
6f458dfb | 635 | { |
59f379f9 KC |
636 | struct inet_connection_sock *icsk = |
637 | from_timer(icsk, t, icsk_retransmit_timer); | |
638 | struct sock *sk = &icsk->icsk_inet.sk; | |
6f458dfb ED |
639 | |
640 | bh_lock_sock(sk); | |
641 | if (!sock_owned_by_user(sk)) { | |
642 | tcp_write_timer_handler(sk); | |
643 | } else { | |
c380d37e | 644 | /* delegate our work to tcp_release_cb() */ |
7aa5470c | 645 | if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 | 646 | sock_hold(sk); |
6f458dfb | 647 | } |
1da177e4 LT |
648 | bh_unlock_sock(sk); |
649 | sock_put(sk); | |
650 | } | |
651 | ||
42cb80a2 | 652 | void tcp_syn_ack_timeout(const struct request_sock *req) |
72659ecc | 653 | { |
42cb80a2 ED |
654 | struct net *net = read_pnet(&inet_rsk(req)->ireq_net); |
655 | ||
02a1d6e7 | 656 | __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS); |
72659ecc OP |
657 | } |
658 | EXPORT_SYMBOL(tcp_syn_ack_timeout); | |
659 | ||
1da177e4 LT |
660 | void tcp_set_keepalive(struct sock *sk, int val) |
661 | { | |
662 | if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) | |
663 | return; | |
664 | ||
665 | if (val && !sock_flag(sk, SOCK_KEEPOPEN)) | |
463c84b9 | 666 | inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk))); |
1da177e4 | 667 | else if (!val) |
463c84b9 | 668 | inet_csk_delete_keepalive_timer(sk); |
1da177e4 | 669 | } |
4b9d07a4 | 670 | EXPORT_SYMBOL_GPL(tcp_set_keepalive); |
1da177e4 LT |
671 | |
672 | ||
59f379f9 | 673 | static void tcp_keepalive_timer (struct timer_list *t) |
1da177e4 | 674 | { |
59f379f9 | 675 | struct sock *sk = from_timer(sk, t, sk_timer); |
6687e988 | 676 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 677 | struct tcp_sock *tp = tcp_sk(sk); |
6c37e5de | 678 | u32 elapsed; |
1da177e4 LT |
679 | |
680 | /* Only process if socket is not in use. */ | |
681 | bh_lock_sock(sk); | |
682 | if (sock_owned_by_user(sk)) { | |
e905a9ed | 683 | /* Try again later. */ |
463c84b9 | 684 | inet_csk_reset_keepalive_timer (sk, HZ/20); |
1da177e4 LT |
685 | goto out; |
686 | } | |
687 | ||
688 | if (sk->sk_state == TCP_LISTEN) { | |
fa76ce73 | 689 | pr_err("Hmm... keepalive on a LISTEN ???\n"); |
1da177e4 LT |
690 | goto out; |
691 | } | |
692 | ||
4688eb7c | 693 | tcp_mstamp_refresh(tp); |
1da177e4 LT |
694 | if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) { |
695 | if (tp->linger2 >= 0) { | |
463c84b9 | 696 | const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN; |
1da177e4 LT |
697 | |
698 | if (tmo > 0) { | |
699 | tcp_time_wait(sk, TCP_FIN_WAIT2, tmo); | |
700 | goto out; | |
701 | } | |
702 | } | |
703 | tcp_send_active_reset(sk, GFP_ATOMIC); | |
704 | goto death; | |
705 | } | |
706 | ||
2dda6400 ED |
707 | if (!sock_flag(sk, SOCK_KEEPOPEN) || |
708 | ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT))) | |
1da177e4 LT |
709 | goto out; |
710 | ||
711 | elapsed = keepalive_time_when(tp); | |
712 | ||
713 | /* It is alive without keepalive 8) */ | |
75c119af | 714 | if (tp->packets_out || !tcp_write_queue_empty(sk)) |
1da177e4 LT |
715 | goto resched; |
716 | ||
6c37e5de | 717 | elapsed = keepalive_time_elapsed(tp); |
1da177e4 LT |
718 | |
719 | if (elapsed >= keepalive_time_when(tp)) { | |
dca43c75 JC |
720 | /* If the TCP_USER_TIMEOUT option is enabled, use that |
721 | * to determine when to timeout instead. | |
722 | */ | |
723 | if ((icsk->icsk_user_timeout != 0 && | |
9bcc66e1 | 724 | elapsed >= msecs_to_jiffies(icsk->icsk_user_timeout) && |
dca43c75 JC |
725 | icsk->icsk_probes_out > 0) || |
726 | (icsk->icsk_user_timeout == 0 && | |
727 | icsk->icsk_probes_out >= keepalive_probes(tp))) { | |
1da177e4 LT |
728 | tcp_send_active_reset(sk, GFP_ATOMIC); |
729 | tcp_write_err(sk); | |
730 | goto out; | |
731 | } | |
e520af48 | 732 | if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) { |
6687e988 | 733 | icsk->icsk_probes_out++; |
1da177e4 LT |
734 | elapsed = keepalive_intvl_when(tp); |
735 | } else { | |
736 | /* If keepalive was lost due to local congestion, | |
737 | * try harder. | |
738 | */ | |
739 | elapsed = TCP_RESOURCE_PROBE_INTERVAL; | |
740 | } | |
741 | } else { | |
742 | /* It is tp->rcv_tstamp + keepalive_time_when(tp) */ | |
743 | elapsed = keepalive_time_when(tp) - elapsed; | |
744 | } | |
745 | ||
3ab224be | 746 | sk_mem_reclaim(sk); |
1da177e4 LT |
747 | |
748 | resched: | |
463c84b9 | 749 | inet_csk_reset_keepalive_timer (sk, elapsed); |
1da177e4 LT |
750 | goto out; |
751 | ||
e905a9ed | 752 | death: |
1da177e4 LT |
753 | tcp_done(sk); |
754 | ||
755 | out: | |
756 | bh_unlock_sock(sk); | |
757 | sock_put(sk); | |
758 | } | |
6f458dfb | 759 | |
5d9f4262 ED |
760 | static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer) |
761 | { | |
762 | struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer); | |
763 | struct sock *sk = (struct sock *)tp; | |
764 | ||
765 | bh_lock_sock(sk); | |
766 | if (!sock_owned_by_user(sk)) { | |
2b195850 ED |
767 | if (tp->compressed_ack) { |
768 | /* Since we have to send one ack finally, | |
974d8f86 | 769 | * subtract one from tp->compressed_ack to keep |
2b195850 ED |
770 | * LINUX_MIB_TCPACKCOMPRESSED accurate. |
771 | */ | |
772 | tp->compressed_ack--; | |
5d9f4262 | 773 | tcp_send_ack(sk); |
2b195850 | 774 | } |
5d9f4262 ED |
775 | } else { |
776 | if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, | |
777 | &sk->sk_tsq_flags)) | |
778 | sock_hold(sk); | |
779 | } | |
780 | bh_unlock_sock(sk); | |
781 | ||
782 | sock_put(sk); | |
783 | ||
784 | return HRTIMER_NORESTART; | |
785 | } | |
786 | ||
6f458dfb ED |
787 | void tcp_init_xmit_timers(struct sock *sk) |
788 | { | |
789 | inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer, | |
790 | &tcp_keepalive_timer); | |
fb420d5d | 791 | hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC, |
73a6bab5 | 792 | HRTIMER_MODE_ABS_PINNED_SOFT); |
218af599 | 793 | tcp_sk(sk)->pacing_timer.function = tcp_pace_kick; |
5d9f4262 ED |
794 | |
795 | hrtimer_init(&tcp_sk(sk)->compressed_ack_timer, CLOCK_MONOTONIC, | |
796 | HRTIMER_MODE_REL_PINNED_SOFT); | |
797 | tcp_sk(sk)->compressed_ack_timer.function = tcp_compressed_ack_kick; | |
6f458dfb | 798 | } |