Fix: rotation may never complete in per-PID buffering mode
[lttng-tools.git] / src / bin / lttng-sessiond / sessiond-timer.c
CommitLineData
d086f507
JD
1/*
2 * Copyright (C) 2017 - Julien Desfossez <jdesfossez@efficios.com>
92816cc3 3 * Copyright (C) 2018 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
d086f507
JD
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License, version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 51
16 * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19#define _LGPL_SOURCE
20#include <assert.h>
21#include <inttypes.h>
22#include <signal.h>
23
24#include "sessiond-timer.h"
25#include "health-sessiond.h"
26#include "rotation-thread.h"
27
92816cc3
JG
28#define LTTNG_SESSIOND_SIG_QS SIGRTMIN + 10
29#define LTTNG_SESSIOND_SIG_EXIT SIGRTMIN + 11
30#define LTTNG_SESSIOND_SIG_PENDING_ROTATION_CHECK SIGRTMIN + 12
31#define LTTNG_SESSIOND_SIG_SCHEDULED_ROTATION SIGRTMIN + 13
32
33#define UINT_TO_PTR(value) \
34 ({ \
35 assert(value <= UINTPTR_MAX); \
36 (void *) (uintptr_t) value; \
37 })
38#define PTR_TO_UINT(ptr) ((uintptr_t) ptr)
39
40/*
41 * Handle timer teardown race wrt memory free of private data by sessiond
42 * signals are handled by a single thread, which permits a synchronization
43 * point between handling of each signal. Internal lock ensures mutual
44 * exclusion.
45 */
d086f507 46static
92816cc3
JG
47struct timer_signal_data {
48 /* Thread managing signals. */
49 pthread_t tid;
50 int qs_done;
51 pthread_mutex_t lock;
52} timer_signal = {
d086f507
JD
53 .tid = 0,
54 .qs_done = 0,
55 .lock = PTHREAD_MUTEX_INITIALIZER,
56};
57
58/*
59 * Set custom signal mask to current thread.
60 */
61static
62void setmask(sigset_t *mask)
63{
64 int ret;
65
66 ret = sigemptyset(mask);
67 if (ret) {
68 PERROR("sigemptyset");
69 }
92816cc3 70 ret = sigaddset(mask, LTTNG_SESSIOND_SIG_QS);
d086f507
JD
71 if (ret) {
72 PERROR("sigaddset teardown");
73 }
74 ret = sigaddset(mask, LTTNG_SESSIOND_SIG_EXIT);
75 if (ret) {
76 PERROR("sigaddset exit");
77 }
92816cc3 78 ret = sigaddset(mask, LTTNG_SESSIOND_SIG_PENDING_ROTATION_CHECK);
d88744a4 79 if (ret) {
92816cc3 80 PERROR("sigaddset pending rotation check");
d88744a4 81 }
92816cc3 82 ret = sigaddset(mask, LTTNG_SESSIOND_SIG_SCHEDULED_ROTATION);
259c2674 83 if (ret) {
92816cc3 84 PERROR("sigaddset scheduled rotation");
259c2674 85 }
d086f507
JD
86}
87
88/*
92816cc3 89 * This is the same function as timer_signal_thread_qs, when it
d086f507
JD
90 * returns, it means that no timer signr is currently pending or being handled
91 * by the timer thread. This cannot be called from the timer thread.
92 */
93static
92816cc3 94void timer_signal_thread_qs(unsigned int signr)
d086f507
JD
95{
96 sigset_t pending_set;
97 int ret;
98
99 /*
100 * We need to be the only thread interacting with the thread
101 * that manages signals for teardown synchronization.
102 */
103 pthread_mutex_lock(&timer_signal.lock);
104
105 /* Ensure we don't have any signal queued for this session. */
106 for (;;) {
107 ret = sigemptyset(&pending_set);
108 if (ret == -1) {
109 PERROR("sigemptyset");
110 }
111 ret = sigpending(&pending_set);
112 if (ret == -1) {
113 PERROR("sigpending");
114 }
115 if (!sigismember(&pending_set, signr)) {
116 break;
117 }
118 caa_cpu_relax();
119 }
120
121 /*
122 * From this point, no new signal handler will be fired that would try to
123 * access "session". However, we still need to wait for any currently
124 * executing handler to complete.
125 */
126 cmm_smp_mb();
127 CMM_STORE_SHARED(timer_signal.qs_done, 0);
128 cmm_smp_mb();
129
130 /*
92816cc3 131 * Kill with LTTNG_SESSIOND_SIG_QS, so signal management thread
d086f507
JD
132 * wakes up.
133 */
92816cc3 134 kill(getpid(), LTTNG_SESSIOND_SIG_QS);
d086f507
JD
135
136 while (!CMM_LOAD_SHARED(timer_signal.qs_done)) {
137 caa_cpu_relax();
138 }
139 cmm_smp_mb();
140
141 pthread_mutex_unlock(&timer_signal.lock);
142}
143
144/*
145 * Start a timer on a session that will fire at a given interval
146 * (timer_interval_us) and fire a given signal (signal).
147 *
148 * Returns a negative value on error, 0 if a timer was created, and
149 * a positive value if no timer was created (not an error).
150 */
151static
92816cc3 152int timer_start(timer_t *timer_id, uint64_t session_id,
d086f507
JD
153 unsigned int timer_interval_us, int signal, bool one_shot)
154{
155 int ret = 0, delete_ret;
156 struct sigevent sev;
157 struct itimerspec its;
158
d086f507
JD
159 sev.sigev_notify = SIGEV_SIGNAL;
160 sev.sigev_signo = signal;
92816cc3
JG
161 sev.sigev_value.sival_ptr = UINT_TO_PTR(session_id);
162 ret = timer_create(CLOCK_MONOTONIC, &sev, timer_id);
d086f507
JD
163 if (ret == -1) {
164 PERROR("timer_create");
165 goto end;
166 }
167
168 its.it_value.tv_sec = timer_interval_us / 1000000;
169 its.it_value.tv_nsec = (timer_interval_us % 1000000) * 1000;
170 if (one_shot) {
171 its.it_interval.tv_sec = 0;
172 its.it_interval.tv_nsec = 0;
173 } else {
174 its.it_interval.tv_sec = its.it_value.tv_sec;
175 its.it_interval.tv_nsec = its.it_value.tv_nsec;
176 }
177
178 ret = timer_settime(*timer_id, 0, &its, NULL);
179 if (ret == -1) {
180 PERROR("timer_settime");
181 goto error_destroy_timer;
182 }
183 goto end;
184
185error_destroy_timer:
186 delete_ret = timer_delete(*timer_id);
187 if (delete_ret == -1) {
188 PERROR("timer_delete");
189 }
190
191end:
192 return ret;
193}
194
195static
92816cc3 196int timer_stop(timer_t *timer_id, int signal)
d086f507
JD
197{
198 int ret = 0;
199
200 ret = timer_delete(*timer_id);
201 if (ret == -1) {
202 PERROR("timer_delete");
203 goto end;
204 }
205
92816cc3 206 timer_signal_thread_qs(signal);
d086f507
JD
207 *timer_id = 0;
208end:
209 return ret;
210}
211
92816cc3 212int timer_session_rotation_pending_check_start(struct ltt_session *session,
d88744a4
JD
213 unsigned int interval_us)
214{
215 int ret;
216
92816cc3
JG
217 DBG("Enabling session rotation pending check timer on session %" PRIu64,
218 session->id);
d88744a4
JD
219 /*
220 * We arm this timer in a one-shot mode so we don't have to disable it
92816cc3
JG
221 * explicitly (which could deadlock if the timer thread is blocked
222 * writing in the rotation_timer_pipe).
223 *
d88744a4 224 * Instead, we re-arm it if needed after the rotation_pending check as
92816cc3
JG
225 * returned. Also, this timer is usually only needed once, so there is
226 * no need to go through the whole signal teardown scheme everytime.
d88744a4 227 */
92816cc3
JG
228 ret = timer_start(&session->rotation_pending_check_timer,
229 session->id, interval_us,
230 LTTNG_SESSIOND_SIG_PENDING_ROTATION_CHECK,
d88744a4
JD
231 /* one-shot */ true);
232 if (ret == 0) {
92816cc3 233 session->rotation_pending_check_timer_enabled = true;
d88744a4
JD
234 }
235
236 return ret;
237}
238
239/*
92816cc3 240 * Call with session and session_list locks held.
d88744a4 241 */
92816cc3 242int timer_session_rotation_pending_check_stop(struct ltt_session *session)
d88744a4
JD
243{
244 int ret;
245
246 assert(session);
247
92816cc3
JG
248 DBG("Disabling session rotation pending check timer on session %" PRIu64,
249 session->id);
250 ret = timer_stop(&session->rotation_pending_check_timer,
251 LTTNG_SESSIOND_SIG_PENDING_ROTATION_CHECK);
d88744a4 252 if (ret == -1) {
92816cc3 253 ERR("Failed to stop rotate_pending_check timer");
259c2674 254 } else {
92816cc3 255 session->rotation_pending_check_timer_enabled = false;
259c2674
JD
256 }
257 return ret;
258}
259
92816cc3
JG
260/*
261 * Call with session and session_list locks held.
262 */
263int timer_session_rotation_schedule_timer_start(struct ltt_session *session,
259c2674
JD
264 unsigned int interval_us)
265{
266 int ret;
267
92816cc3 268 DBG("Enabling scheduled rotation timer on session \"%s\" (%ui µs)", session->name,
259c2674 269 interval_us);
92816cc3
JG
270 ret = timer_start(&session->rotation_schedule_timer, session->id,
271 interval_us, LTTNG_SESSIOND_SIG_SCHEDULED_ROTATION,
272 /* one-shot */ false);
259c2674
JD
273 if (ret < 0) {
274 goto end;
275 }
92816cc3 276 session->rotation_schedule_timer_enabled = true;
259c2674
JD
277end:
278 return ret;
279}
280
281/*
92816cc3 282 * Call with session and session_list locks held.
259c2674 283 */
92816cc3 284int timer_session_rotation_schedule_timer_stop(struct ltt_session *session)
259c2674
JD
285{
286 int ret = 0;
287
288 assert(session);
289
92816cc3 290 if (!session->rotation_schedule_timer_enabled) {
259c2674
JD
291 goto end;
292 }
293
92816cc3
JG
294 DBG("Disabling scheduled rotation timer on session %s", session->name);
295 ret = timer_stop(&session->rotation_schedule_timer,
296 LTTNG_SESSIOND_SIG_SCHEDULED_ROTATION);
259c2674 297 if (ret < 0) {
92816cc3 298 ERR("Failed to stop scheduled rotation timer of session \"%s\"",
259c2674
JD
299 session->name);
300 goto end;
d88744a4
JD
301 }
302
92816cc3 303 session->rotation_schedule_timer_enabled = false;
259c2674
JD
304 ret = 0;
305end:
306 return ret;
d88744a4
JD
307}
308
d086f507
JD
309/*
310 * Block the RT signals for the entire process. It must be called from the
311 * sessiond main before creating the threads
312 */
92816cc3 313int timer_signal_init(void)
d086f507
JD
314{
315 int ret;
316 sigset_t mask;
317
318 /* Block signal for entire process, so only our thread processes it. */
319 setmask(&mask);
320 ret = pthread_sigmask(SIG_BLOCK, &mask, NULL);
321 if (ret) {
322 errno = ret;
323 PERROR("pthread_sigmask");
324 return -1;
325 }
326 return 0;
327}
328
329/*
330 * This thread is the sighandler for the timer signals.
331 */
92816cc3 332void *timer_thread_func(void *data)
d086f507
JD
333{
334 int signr;
335 sigset_t mask;
336 siginfo_t info;
337 struct timer_thread_parameters *ctx = data;
338
339 rcu_register_thread();
340 rcu_thread_online();
341
342 health_register(health_sessiond, HEALTH_SESSIOND_TYPE_TIMER);
d086f507
JD
343 health_code_update();
344
345 /* Only self thread will receive signal mask. */
346 setmask(&mask);
347 CMM_STORE_SHARED(timer_signal.tid, pthread_self());
348
349 while (1) {
350 health_code_update();
351
352 health_poll_entry();
353 signr = sigwaitinfo(&mask, &info);
354 health_poll_exit();
355
356 /*
357 * NOTE: cascading conditions are used instead of a switch case
358 * since the use of SIGRTMIN in the definition of the signals'
359 * values prevents the reduction to an integer constant.
360 */
361 if (signr == -1) {
362 if (errno != EINTR) {
363 PERROR("sigwaitinfo");
364 }
365 continue;
92816cc3 366 } else if (signr == LTTNG_SESSIOND_SIG_QS) {
d086f507
JD
367 cmm_smp_mb();
368 CMM_STORE_SHARED(timer_signal.qs_done, 1);
369 cmm_smp_mb();
d086f507
JD
370 } else if (signr == LTTNG_SESSIOND_SIG_EXIT) {
371 goto end;
92816cc3
JG
372 } else if (signr == LTTNG_SESSIOND_SIG_PENDING_ROTATION_CHECK) {
373 rotation_thread_enqueue_job(ctx->rotation_thread_job_queue,
374 ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION,
375 /* session_id */ PTR_TO_UINT(info.si_value.sival_ptr));
376 } else if (signr == LTTNG_SESSIOND_SIG_SCHEDULED_ROTATION) {
377 rotation_thread_enqueue_job(ctx->rotation_thread_job_queue,
378 ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION,
379 /* session_id */ PTR_TO_UINT(info.si_value.sival_ptr));
d086f507
JD
380 } else {
381 ERR("Unexpected signal %d\n", info.si_signo);
382 }
383 }
384
385end:
386 DBG("[timer-thread] Exit");
387 health_unregister(health_sessiond);
388 rcu_thread_offline();
389 rcu_unregister_thread();
390 return NULL;
391}
92816cc3
JG
392
393void timer_exit(void)
394{
395 kill(getpid(), LTTNG_SESSIOND_SIG_EXIT);
396}
This page took 0.069432 seconds and 5 git commands to generate.