1 /*
2 * net/dccp/timer.c
3 *
4 * An implementation of the DCCP protocol
5 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13 #include <linux/dccp.h>
14 #include <linux/skbuff.h>
15 #include <linux/export.h>
16
17 #include "dccp.h"
18
19 /* sysctl variables governing numbers of retransmission attempts */
20 int sysctl_dccp_request_retries __read_mostly = TCP_SYN_RETRIES;
21 int sysctl_dccp_retries1 __read_mostly = TCP_RETR1;
22 int sysctl_dccp_retries2 __read_mostly = TCP_RETR2;
23
dccp_write_err(struct sock * sk)24 static void dccp_write_err(struct sock *sk)
25 {
26 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
27 sk->sk_error_report(sk);
28
29 dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
30 dccp_done(sk);
31 DCCP_INC_STATS_BH(DCCP_MIB_ABORTONTIMEOUT);
32 }
33
34 /* A write timeout has occurred. Process the after effects. */
dccp_write_timeout(struct sock * sk)35 static int dccp_write_timeout(struct sock *sk)
36 {
37 const struct inet_connection_sock *icsk = inet_csk(sk);
38 int retry_until;
39
40 if (sk->sk_state == DCCP_REQUESTING || sk->sk_state == DCCP_PARTOPEN) {
41 if (icsk->icsk_retransmits != 0)
42 dst_negative_advice(sk);
43 retry_until = icsk->icsk_syn_retries ?
44 : sysctl_dccp_request_retries;
45 } else {
46 if (icsk->icsk_retransmits >= sysctl_dccp_retries1) {
47 /* NOTE. draft-ietf-tcpimpl-pmtud-01.txt requires pmtu
48 black hole detection. :-(
49
50 It is place to make it. It is not made. I do not want
51 to make it. It is disguisting. It does not work in any
52 case. Let me to cite the same draft, which requires for
53 us to implement this:
54
55 "The one security concern raised by this memo is that ICMP black holes
56 are often caused by over-zealous security administrators who block
57 all ICMP messages. It is vitally important that those who design and
58 deploy security systems understand the impact of strict filtering on
59 upper-layer protocols. The safest web site in the world is worthless
60 if most TCP implementations cannot transfer data from it. It would
61 be far nicer to have all of the black holes fixed rather than fixing
62 all of the TCP implementations."
63
64 Golden words :-).
65 */
66
67 dst_negative_advice(sk);
68 }
69
70 retry_until = sysctl_dccp_retries2;
71 /*
72 * FIXME: see tcp_write_timout and tcp_out_of_resources
73 */
74 }
75
76 if (icsk->icsk_retransmits >= retry_until) {
77 /* Has it gone just too far? */
78 dccp_write_err(sk);
79 return 1;
80 }
81 return 0;
82 }
83
84 /*
85 * The DCCP retransmit timer.
86 */
dccp_retransmit_timer(struct sock * sk)87 static void dccp_retransmit_timer(struct sock *sk)
88 {
89 struct inet_connection_sock *icsk = inet_csk(sk);
90
91 /*
92 * More than than 4MSL (8 minutes) has passed, a RESET(aborted) was
93 * sent, no need to retransmit, this sock is dead.
94 */
95 if (dccp_write_timeout(sk))
96 return;
97
98 /*
99 * We want to know the number of packets retransmitted, not the
100 * total number of retransmissions of clones of original packets.
101 */
102 if (icsk->icsk_retransmits == 0)
103 DCCP_INC_STATS_BH(DCCP_MIB_TIMEOUTS);
104
105 if (dccp_retransmit_skb(sk) != 0) {
106 /*
107 * Retransmission failed because of local congestion,
108 * do not backoff.
109 */
110 if (--icsk->icsk_retransmits == 0)
111 icsk->icsk_retransmits = 1;
112 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
113 min(icsk->icsk_rto,
114 TCP_RESOURCE_PROBE_INTERVAL),
115 DCCP_RTO_MAX);
116 return;
117 }
118
119 icsk->icsk_backoff++;
120
121 icsk->icsk_rto = min(icsk->icsk_rto << 1, DCCP_RTO_MAX);
122 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto,
123 DCCP_RTO_MAX);
124 if (icsk->icsk_retransmits > sysctl_dccp_retries1)
125 __sk_dst_reset(sk);
126 }
127
dccp_write_timer(unsigned long data)128 static void dccp_write_timer(unsigned long data)
129 {
130 struct sock *sk = (struct sock *)data;
131 struct inet_connection_sock *icsk = inet_csk(sk);
132 int event = 0;
133
134 bh_lock_sock(sk);
135 if (sock_owned_by_user(sk)) {
136 /* Try again later */
137 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
138 jiffies + (HZ / 20));
139 goto out;
140 }
141
142 if (sk->sk_state == DCCP_CLOSED || !icsk->icsk_pending)
143 goto out;
144
145 if (time_after(icsk->icsk_timeout, jiffies)) {
146 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
147 icsk->icsk_timeout);
148 goto out;
149 }
150
151 event = icsk->icsk_pending;
152 icsk->icsk_pending = 0;
153
154 switch (event) {
155 case ICSK_TIME_RETRANS:
156 dccp_retransmit_timer(sk);
157 break;
158 }
159 out:
160 bh_unlock_sock(sk);
161 sock_put(sk);
162 }
163
164 /*
165 * Timer for listening sockets
166 */
dccp_response_timer(struct sock * sk)167 static void dccp_response_timer(struct sock *sk)
168 {
169 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL, DCCP_TIMEOUT_INIT,
170 DCCP_RTO_MAX);
171 }
172
dccp_keepalive_timer(unsigned long data)173 static void dccp_keepalive_timer(unsigned long data)
174 {
175 struct sock *sk = (struct sock *)data;
176
177 /* Only process if socket is not in use. */
178 bh_lock_sock(sk);
179 if (sock_owned_by_user(sk)) {
180 /* Try again later. */
181 inet_csk_reset_keepalive_timer(sk, HZ / 20);
182 goto out;
183 }
184
185 if (sk->sk_state == DCCP_LISTEN) {
186 dccp_response_timer(sk);
187 goto out;
188 }
189 out:
190 bh_unlock_sock(sk);
191 sock_put(sk);
192 }
193
194 /* This is the same as tcp_delack_timer, sans prequeue & mem_reclaim stuff */
dccp_delack_timer(unsigned long data)195 static void dccp_delack_timer(unsigned long data)
196 {
197 struct sock *sk = (struct sock *)data;
198 struct inet_connection_sock *icsk = inet_csk(sk);
199
200 bh_lock_sock(sk);
201 if (sock_owned_by_user(sk)) {
202 /* Try again later. */
203 icsk->icsk_ack.blocked = 1;
204 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
205 sk_reset_timer(sk, &icsk->icsk_delack_timer,
206 jiffies + TCP_DELACK_MIN);
207 goto out;
208 }
209
210 if (sk->sk_state == DCCP_CLOSED ||
211 !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
212 goto out;
213 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
214 sk_reset_timer(sk, &icsk->icsk_delack_timer,
215 icsk->icsk_ack.timeout);
216 goto out;
217 }
218
219 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
220
221 if (inet_csk_ack_scheduled(sk)) {
222 if (!icsk->icsk_ack.pingpong) {
223 /* Delayed ACK missed: inflate ATO. */
224 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1,
225 icsk->icsk_rto);
226 } else {
227 /* Delayed ACK missed: leave pingpong mode and
228 * deflate ATO.
229 */
230 icsk->icsk_ack.pingpong = 0;
231 icsk->icsk_ack.ato = TCP_ATO_MIN;
232 }
233 dccp_send_ack(sk);
234 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
235 }
236 out:
237 bh_unlock_sock(sk);
238 sock_put(sk);
239 }
240
241 /**
242 * dccp_write_xmitlet - Workhorse for CCID packet dequeueing interface
243 * See the comments above %ccid_dequeueing_decision for supported modes.
244 */
dccp_write_xmitlet(unsigned long data)245 static void dccp_write_xmitlet(unsigned long data)
246 {
247 struct sock *sk = (struct sock *)data;
248
249 bh_lock_sock(sk);
250 if (sock_owned_by_user(sk))
251 sk_reset_timer(sk, &dccp_sk(sk)->dccps_xmit_timer, jiffies + 1);
252 else
253 dccp_write_xmit(sk);
254 bh_unlock_sock(sk);
255 }
256
dccp_write_xmit_timer(unsigned long data)257 static void dccp_write_xmit_timer(unsigned long data)
258 {
259 dccp_write_xmitlet(data);
260 sock_put((struct sock *)data);
261 }
262
dccp_init_xmit_timers(struct sock * sk)263 void dccp_init_xmit_timers(struct sock *sk)
264 {
265 struct dccp_sock *dp = dccp_sk(sk);
266
267 tasklet_init(&dp->dccps_xmitlet, dccp_write_xmitlet, (unsigned long)sk);
268 setup_timer(&dp->dccps_xmit_timer, dccp_write_xmit_timer,
269 (unsigned long)sk);
270 inet_csk_init_xmit_timers(sk, &dccp_write_timer, &dccp_delack_timer,
271 &dccp_keepalive_timer);
272 }
273
274 static ktime_t dccp_timestamp_seed;
275 /**
276 * dccp_timestamp - 10s of microseconds time source
277 * Returns the number of 10s of microseconds since loading DCCP. This is native
278 * DCCP time difference format (RFC 4340, sec. 13).
279 * Please note: This will wrap around about circa every 11.9 hours.
280 */
dccp_timestamp(void)281 u32 dccp_timestamp(void)
282 {
283 s64 delta = ktime_us_delta(ktime_get_real(), dccp_timestamp_seed);
284
285 do_div(delta, 10);
286 return delta;
287 }
288 EXPORT_SYMBOL_GPL(dccp_timestamp);
289
dccp_timestamping_init(void)290 void __init dccp_timestamping_init(void)
291 {
292 dccp_timestamp_seed = ktime_get_real();
293 }
294