[lttng-tools.git] / src / bin / lttng-sessiond / manage-kernel.c

/*
 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
 *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 *               2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License, version 2 only,
 * as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License along
 * with this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
 */

#include <common/pipe.h>
#include <common/utils.h>

#include "manage-kernel.h"
#include "testpoint.h"
#include "health-sessiond.h"
#include "utils.h"
#include "thread.h"
#include "kernel.h"
#include "kernel-consumer.h"

struct thread_notifiers {
	struct lttng_pipe *quit_pipe;
	int kernel_poll_pipe_read_fd;
};

/*
 * Update the kernel poll set of all channel fd available over all tracing
 * session. Add the wakeup pipe at the end of the set.
 */
static int update_kernel_poll(struct lttng_poll_event *events)
{
	int ret;
	struct ltt_kernel_channel *channel;
	struct ltt_session *session;
	const struct ltt_session_list *session_list = session_get_list();

	DBG("Updating kernel poll set");

	session_lock_list();
	cds_list_for_each_entry(session, &session_list->head, list) {
		if (!session_get(session)) {
			continue;
		}
		session_lock(session);
		if (session->kernel_session == NULL) {
			session_unlock(session);
			session_put(session);
			continue;
		}

		cds_list_for_each_entry(channel,
				&session->kernel_session->channel_list.head, list) {
			/* Add channel fd to the kernel poll set */
			ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM);
			if (ret < 0) {
				session_unlock(session);
				session_put(session);
				goto error;
			}
			DBG("Channel fd %d added to kernel set", channel->fd);
		}
		session_unlock(session);
		session_put(session);
	}
	session_unlock_list();

	return 0;

error:
	session_unlock_list();
	return -1;
}

/*
 * Find the channel fd from 'fd' over all tracing session. When found, check
 * for new channel stream and send those stream fds to the kernel consumer.
 *
 * Useful for CPU hotplug feature.
 */
static int update_kernel_stream(int fd)
{
	int ret = 0;
	struct ltt_session *session;
	struct ltt_kernel_session *ksess;
	struct ltt_kernel_channel *channel;
	const struct ltt_session_list *session_list = session_get_list();

	DBG("Updating kernel streams for channel fd %d", fd);

	session_lock_list();
	cds_list_for_each_entry(session, &session_list->head, list) {
		if (!session_get(session)) {
			continue;
		}
		session_lock(session);
		if (session->kernel_session == NULL) {
			session_unlock(session);
			session_put(session);
			continue;
		}
		ksess = session->kernel_session;

		cds_list_for_each_entry(channel,
				&ksess->channel_list.head, list) {
			struct lttng_ht_iter iter;
			struct consumer_socket *socket;

			if (channel->fd != fd) {
				continue;
			}
			DBG("Channel found, updating kernel streams");
			ret = kernel_open_channel_stream(channel);
			if (ret < 0) {
				goto error;
			}
			/* Update the stream global counter */
			ksess->stream_count_global += ret;

			/*
			 * Have we already sent fds to the consumer? If yes, it
			 * means that tracing is started so it is safe to send
			 * our updated stream fds.
			 */
			if (ksess->consumer_fds_sent != 1
					|| ksess->consumer == NULL) {
				ret = -1;
				goto error;
			}

			rcu_read_lock();
			cds_lfht_for_each_entry(ksess->consumer->socks->ht,
					&iter.iter, socket, node.node) {
				pthread_mutex_lock(socket->lock);
				ret = kernel_consumer_send_channel_streams(socket,
						channel, ksess,
						session->output_traces ? 1 : 0);
				pthread_mutex_unlock(socket->lock);
				if (ret < 0) {
					rcu_read_unlock();
					goto error;
				}
			}
			rcu_read_unlock();
		}
		session_unlock(session);
		session_put(session);
	}
	session_unlock_list();
	return ret;

error:
	session_unlock(session);
	session_put(session);
	session_unlock_list();
	return ret;
}

/*
 * This thread manage event coming from the kernel.
 *
 * Features supported in this thread:
 *    -) CPU Hotplug
 */
static void *thread_kernel_management(void *data)
{
	int ret, i, pollfd, update_poll_flag = 1, err = -1;
	uint32_t revents, nb_fd;
	char tmp;
	struct lttng_poll_event events;
	struct thread_notifiers *notifiers = data;
	const int quit_pipe_read_fd = lttng_pipe_get_readfd(notifiers->quit_pipe);

	DBG("[thread] Thread manage kernel started");

	health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);

	/*
	 * This first step of the while is to clean this structure which could free
	 * non NULL pointers so initialize it before the loop.
	 */
	lttng_poll_init(&events);

	if (testpoint(sessiond_thread_manage_kernel)) {
		goto error_testpoint;
	}

	health_code_update();

	if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
		goto error_testpoint;
	}

	while (1) {
		health_code_update();

		if (update_poll_flag == 1) {
			/* Clean events object. We are about to populate it again. */
			lttng_poll_clean(&events);

			ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
			if (ret < 0) {
				goto error_poll_create;
			}

			ret = lttng_poll_add(&events,
					notifiers->kernel_poll_pipe_read_fd,
					LPOLLIN);
			if (ret < 0) {
				goto error;
			}

			ret = lttng_poll_add(&events,
					quit_pipe_read_fd,
					LPOLLIN);
			if (ret < 0) {
				goto error;
			}

			/* This will add the available kernel channel if any. */
			ret = update_kernel_poll(&events);
			if (ret < 0) {
				goto error;
			}
			update_poll_flag = 0;
		}

		DBG("Thread kernel polling");

		/* Poll infinite value of time */
	restart:
		health_poll_entry();
		ret = lttng_poll_wait(&events, -1);
		DBG("Thread kernel return from poll on %d fds",
				LTTNG_POLL_GETNB(&events));
		health_poll_exit();
		if (ret < 0) {
			/*
			 * Restart interrupted system call.
			 */
			if (errno == EINTR) {
				goto restart;
			}
			goto error;
		} else if (ret == 0) {
			/* Should not happen since timeout is infinite */
			ERR("Return value of poll is 0 with an infinite timeout.\n"
				"This should not have happened! Continuing...");
			continue;
		}

		nb_fd = ret;

		for (i = 0; i < nb_fd; i++) {
			/* Fetch once the poll data */
			revents = LTTNG_POLL_GETEV(&events, i);
			pollfd = LTTNG_POLL_GETFD(&events, i);

			health_code_update();

			if (!revents) {
				/* No activity for this FD (poll implementation). */
				continue;
			}

			if (pollfd == quit_pipe_read_fd) {
				err = 0;
				goto exit;
			}

			/* Check for data on kernel pipe */
			if (revents & LPOLLIN) {
				if (pollfd == notifiers->kernel_poll_pipe_read_fd) {
					(void) lttng_read(notifiers->kernel_poll_pipe_read_fd,
						&tmp, 1);
					/*
					 * Ret value is useless here, if this pipe gets any actions an
					 * update is required anyway.
					 */
					update_poll_flag = 1;
					continue;
				} else {
					/*
					 * New CPU detected by the kernel. Adding kernel stream to
					 * kernel session and updating the kernel consumer
					 */
					ret = update_kernel_stream(pollfd);
					if (ret < 0) {
						continue;
					}
					break;
				}
			} else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
				update_poll_flag = 1;
				continue;
			} else {
				ERR("Unexpected poll events %u for sock %d", revents, pollfd);
				goto error;
			}
		}
	}

exit:
error:
	lttng_poll_clean(&events);
error_poll_create:
error_testpoint:
	if (err) {
		health_error();
		ERR("Health error occurred in %s", __func__);
		WARN("Kernel thread died unexpectedly. "
				"Kernel tracing can continue but CPU hotplug is disabled.");
	}
	health_unregister(health_sessiond);
	DBG("Kernel thread dying");
	return NULL;
}

static bool shutdown_kernel_management_thread(void *data)
{
	struct thread_notifiers *notifiers = data;
	const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe);

	return notify_thread_pipe(write_fd) == 1;
}

static void cleanup_kernel_management_thread(void *data)
{
	struct thread_notifiers *notifiers = data;

	lttng_pipe_destroy(notifiers->quit_pipe);
	free(notifiers);
}

bool launch_kernel_management_thread(int kernel_poll_pipe_read_fd)
{
	struct lttng_pipe *quit_pipe;
	struct thread_notifiers *notifiers = NULL;
	struct lttng_thread *thread;

	quit_pipe = lttng_pipe_open(FD_CLOEXEC);
	if (!quit_pipe) {
		goto error;
	}

	notifiers = zmalloc(sizeof(*notifiers));
	if (!notifiers) {
		goto error;
	}
	notifiers->quit_pipe = quit_pipe;
	notifiers->kernel_poll_pipe_read_fd = kernel_poll_pipe_read_fd;

	thread = lttng_thread_create("Kernel management",
			thread_kernel_management,
			shutdown_kernel_management_thread,
			cleanup_kernel_management_thread,
			notifiers);
	if (!thread) {
		goto error;
	}
	lttng_thread_put(thread);
	return true;
error:
	cleanup_kernel_management_thread(notifiers);
	return false;
}
Commit	Line	Data
5b093681 JG	1	/*
	2	* Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
	3	* Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
	4	* 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
	5	*
	6	* This program is free software; you can redistribute it and/or modify
	7	* it under the terms of the GNU General Public License, version 2 only,
	8	* as published by the Free Software Foundation.
	9	*
	10	* This program is distributed in the hope that it will be useful,
	11	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	12	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	13	* GNU General Public License for more details.
	14	*
	15	* You should have received a copy of the GNU General Public License along
	16	* with this program; if not, write to the Free Software Foundation, Inc.,
	17	* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
	18	*/
	19
	20	#include <common/pipe.h>
	21	#include <common/utils.h>
	22
	23	#include "manage-kernel.h"
	24	#include "testpoint.h"
	25	#include "health-sessiond.h"
	26	#include "utils.h"
	27	#include "thread.h"
	28	#include "kernel.h"
	29	#include "kernel-consumer.h"
	30
	31	struct thread_notifiers {
	32	struct lttng_pipe *quit_pipe;
	33	int kernel_poll_pipe_read_fd;
	34	};
	35
	36	/*
	37	* Update the kernel poll set of all channel fd available over all tracing
	38	* session. Add the wakeup pipe at the end of the set.
	39	*/
	40	static int update_kernel_poll(struct lttng_poll_event *events)
	41	{
	42	int ret;
	43	struct ltt_kernel_channel *channel;
	44	struct ltt_session *session;
	45	const struct ltt_session_list *session_list = session_get_list();
	46
	47	DBG("Updating kernel poll set");
	48
	49	session_lock_list();
	50	cds_list_for_each_entry(session, &session_list->head, list) {
	51	if (!session_get(session)) {
	52	continue;
	53	}
	54	session_lock(session);
	55	if (session->kernel_session == NULL) {
	56	session_unlock(session);
	57	session_put(session);
	58	continue;
	59	}
	60
	61	cds_list_for_each_entry(channel,
	62	&session->kernel_session->channel_list.head, list) {
	63	/* Add channel fd to the kernel poll set */
	64	ret = lttng_poll_add(events, channel->fd, LPOLLIN \| LPOLLRDNORM);
65	if (ret < 0) {
66	session_unlock(session);
67	session_put(session);
68	goto error;
69	}
70	DBG("Channel fd %d added to kernel set", channel->fd);
71	}
72	session_unlock(session);
0318876b	73	session_put(session);
5b093681 JG	74	}
	75	session_unlock_list();
	76
	77	return 0;
	78
	79	error:
	80	session_unlock_list();
	81	return -1;
	82	}
	83
	84	/*
	85	* Find the channel fd from 'fd' over all tracing session. When found, check
	86	* for new channel stream and send those stream fds to the kernel consumer.
	87	*
	88	* Useful for CPU hotplug feature.
	89	*/
	90	static int update_kernel_stream(int fd)
	91	{
	92	int ret = 0;
	93	struct ltt_session *session;
	94	struct ltt_kernel_session *ksess;
	95	struct ltt_kernel_channel *channel;
	96	const struct ltt_session_list *session_list = session_get_list();
	97
	98	DBG("Updating kernel streams for channel fd %d", fd);
	99
	100	session_lock_list();
	101	cds_list_for_each_entry(session, &session_list->head, list) {
	102	if (!session_get(session)) {
	103	continue;
	104	}
	105	session_lock(session);
	106	if (session->kernel_session == NULL) {
	107	session_unlock(session);
	108	session_put(session);
	109	continue;
	110	}
	111	ksess = session->kernel_session;
	112
	113	cds_list_for_each_entry(channel,
	114	&ksess->channel_list.head, list) {
	115	struct lttng_ht_iter iter;
	116	struct consumer_socket *socket;
	117
	118	if (channel->fd != fd) {
	119	continue;
	120	}
	121	DBG("Channel found, updating kernel streams");
	122	ret = kernel_open_channel_stream(channel);
	123	if (ret < 0) {
	124	goto error;
	125	}
	126	/* Update the stream global counter */
	127	ksess->stream_count_global += ret;
	128
	129	/*
	130	* Have we already sent fds to the consumer? If yes, it
	131	* means that tracing is started so it is safe to send
	132	* our updated stream fds.
	133	*/
	134	if (ksess->consumer_fds_sent != 1
	135	\|\| ksess->consumer == NULL) {
	136	ret = -1;
	137	goto error;
138	}
139
140	rcu_read_lock();
141	cds_lfht_for_each_entry(ksess->consumer->socks->ht,
142	&iter.iter, socket, node.node) {
143	pthread_mutex_lock(socket->lock);
144	ret = kernel_consumer_send_channel_streams(socket,
145	channel, ksess,
146	session->output_traces ? 1 : 0);
147	pthread_mutex_unlock(socket->lock);
148	if (ret < 0) {
149	rcu_read_unlock();
150	goto error;
151	}
152	}
153	rcu_read_unlock();
154	}
155	session_unlock(session);
156	session_put(session);
157	}
158	session_unlock_list();
159	return ret;
160
161	error:
162	session_unlock(session);
163	session_put(session);
164	session_unlock_list();
165	return ret;
166	}
167
168	/*
169	* This thread manage event coming from the kernel.
170	*
171	* Features supported in this thread:
172	* -) CPU Hotplug
173	*/
174	static void thread_kernel_management(void data)
175	{
176	int ret, i, pollfd, update_poll_flag = 1, err = -1;
177	uint32_t revents, nb_fd;
178	char tmp;
179	struct lttng_poll_event events;
180	struct thread_notifiers *notifiers = data;
181	const int quit_pipe_read_fd = lttng_pipe_get_readfd(notifiers->quit_pipe);
182
183	DBG("[thread] Thread manage kernel started");
184
185	health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL);
186
187	/*
188	* This first step of the while is to clean this structure which could free
189	* non NULL pointers so initialize it before the loop.
190	*/
191	lttng_poll_init(&events);
192
193	if (testpoint(sessiond_thread_manage_kernel)) {
194	goto error_testpoint;
195	}
196
197	health_code_update();
198
199	if (testpoint(sessiond_thread_manage_kernel_before_loop)) {
200	goto error_testpoint;
201	}
202
203	while (1) {
204	health_code_update();
205
206	if (update_poll_flag == 1) {
207	/* Clean events object. We are about to populate it again. */
208	lttng_poll_clean(&events);
209
210	ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
211	if (ret < 0) {
212	goto error_poll_create;
213	}
214
215	ret = lttng_poll_add(&events,
216	notifiers->kernel_poll_pipe_read_fd,
217	LPOLLIN);
218	if (ret < 0) {
219	goto error;
220	}
221
222	ret = lttng_poll_add(&events,
223	quit_pipe_read_fd,
224	LPOLLIN);
225	if (ret < 0) {
226	goto error;
227	}
228
229	/* This will add the available kernel channel if any. */
230	ret = update_kernel_poll(&events);
231	if (ret < 0) {
232	goto error;
233	}
234	update_poll_flag = 0;
235	}
236
237	DBG("Thread kernel polling");
238
239	/* Poll infinite value of time */
240	restart:
241	health_poll_entry();
242	ret = lttng_poll_wait(&events, -1);
243	DBG("Thread kernel return from poll on %d fds",
244	LTTNG_POLL_GETNB(&events));
245	health_poll_exit();
246	if (ret < 0) {
247	/*
248	* Restart interrupted system call.
249	*/
250	if (errno == EINTR) {
251	goto restart;
252	}
253	goto error;
254	} else if (ret == 0) {
255	/* Should not happen since timeout is infinite */
256	ERR("Return value of poll is 0 with an infinite timeout.\n"
257	"This should not have happened! Continuing...");
258	continue;
259	}
260
261	nb_fd = ret;
262
263	for (i = 0; i < nb_fd; i++) {
264	/* Fetch once the poll data */
265	revents = LTTNG_POLL_GETEV(&events, i);
266	pollfd = LTTNG_POLL_GETFD(&events, i);
267
268	health_code_update();
269
270	if (!revents) {
271	/* No activity for this FD (poll implementation). */
272	continue;
273	}
274
275	if (pollfd == quit_pipe_read_fd) {
276	err = 0;
277	goto exit;
278	}
279
280	/* Check for data on kernel pipe */
281	if (revents & LPOLLIN) {
282	if (pollfd == notifiers->kernel_poll_pipe_read_fd) {
283	(void) lttng_read(notifiers->kernel_poll_pipe_read_fd,
284	&tmp, 1);
285	/*
286	* Ret value is useless here, if this pipe gets any actions an
287	* update is required anyway.
288	*/
289	update_poll_flag = 1;
290	continue;
291	} else {
292	/*
293	* New CPU detected by the kernel. Adding kernel stream to
294	* kernel session and updating the kernel consumer
295	*/
296	ret = update_kernel_stream(pollfd);
297	if (ret < 0) {
298	continue;
299	}
300	break;
301	}
302	} else if (revents & (LPOLLERR \| LPOLLHUP \| LPOLLRDHUP)) {
303	update_poll_flag = 1;
304	continue;
305	} else {
306	ERR("Unexpected poll events %u for sock %d", revents, pollfd);
307	goto error;
308	}
309	}
310	}
311
312	exit:
313	error:
314	lttng_poll_clean(&events);
315	error_poll_create:
316	error_testpoint:
317	if (err) {
318	health_error();
319	ERR("Health error occurred in %s", __func__);
320	WARN("Kernel thread died unexpectedly. "
321	"Kernel tracing can continue but CPU hotplug is disabled.");
322	}
323	health_unregister(health_sessiond);
324	DBG("Kernel thread dying");
325	return NULL;
326	}
327
328	static bool shutdown_kernel_management_thread(void *data)
329	{
330	struct thread_notifiers *notifiers = data;
331	const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe);
332
333	return notify_thread_pipe(write_fd) == 1;
334	}
335
336	static void cleanup_kernel_management_thread(void *data)
337	{
338	struct thread_notifiers *notifiers = data;
339
340	lttng_pipe_destroy(notifiers->quit_pipe);
341	free(notifiers);
342	}
343
344	bool launch_kernel_management_thread(int kernel_poll_pipe_read_fd)
345	{
346	struct lttng_pipe *quit_pipe;
347	struct thread_notifiers *notifiers = NULL;
348	struct lttng_thread *thread;
349
350	quit_pipe = lttng_pipe_open(FD_CLOEXEC);
351	if (!quit_pipe) {
352	goto error;
353	}
354
355	notifiers = zmalloc(sizeof(*notifiers));
356	if (!notifiers) {
357	goto error;
358	}
359	notifiers->quit_pipe = quit_pipe;
360	notifiers->kernel_poll_pipe_read_fd = kernel_poll_pipe_read_fd;
361
362	thread = lttng_thread_create("Kernel management",
363	thread_kernel_management,
364	shutdown_kernel_management_thread,
365	cleanup_kernel_management_thread,
366	notifiers);
367	if (!thread) {
368	goto error;
369	}
370	lttng_thread_put(thread);
371	return true;
372	error:
373	cleanup_kernel_management_thread(notifiers);
374	return false;
375	}