Fixes #324
Signed-off-by: David Goulet <[email protected]>
---
src/bin/lttng-sessiond/consumer.h | 8 ++-
src/bin/lttng-sessiond/main.c | 116 +++++++++++++++++++++++++++++--------
2 files changed, 96 insertions(+), 28 deletions(-)
diff --git a/src/bin/lttng-sessiond/consumer.h
b/src/bin/lttng-sessiond/consumer.h
index 1337f32..a5437d8 100644
--- a/src/bin/lttng-sessiond/consumer.h
+++ b/src/bin/lttng-sessiond/consumer.h
@@ -18,8 +18,6 @@
#ifndef _CONSUMER_H
#define _CONSUMER_H
-#include <semaphore.h>
-
#include <common/consumer.h>
#include <common/hashtable/hashtable.h>
#include <lttng/lttng.h>
@@ -54,7 +52,11 @@ struct consumer_data {
enum lttng_consumer_type type;
pthread_t thread; /* Worker thread interacting with the consumer
*/
- sem_t sem;
+
+ /* Conditions used by the consumer thread to indicate readiness. */
+ pthread_cond_t cond;
+ pthread_condattr_t condattr;
+ pthread_mutex_t cond_mutex;
/* Mutex to control consumerd pid assignation */
pthread_mutex_t pid_mutex;
diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c
index 730ac65..df817c1 100644
--- a/src/bin/lttng-sessiond/main.c
+++ b/src/bin/lttng-sessiond/main.c
@@ -21,7 +21,6 @@
#include <grp.h>
#include <limits.h>
#include <pthread.h>
-#include <semaphore.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
@@ -90,6 +89,8 @@ static struct consumer_data kconsumer_data = {
.cmd_sock = -1,
.pid_mutex = PTHREAD_MUTEX_INITIALIZER,
.lock = PTHREAD_MUTEX_INITIALIZER,
+ .cond = PTHREAD_COND_INITIALIZER,
+ .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer64_data = {
.type = LTTNG_CONSUMER64_UST,
@@ -99,6 +100,8 @@ static struct consumer_data ustconsumer64_data = {
.cmd_sock = -1,
.pid_mutex = PTHREAD_MUTEX_INITIALIZER,
.lock = PTHREAD_MUTEX_INITIALIZER,
+ .cond = PTHREAD_COND_INITIALIZER,
+ .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer32_data = {
.type = LTTNG_CONSUMER32_UST,
@@ -108,6 +111,8 @@ static struct consumer_data ustconsumer32_data = {
.cmd_sock = -1,
.pid_mutex = PTHREAD_MUTEX_INITIALIZER,
.lock = PTHREAD_MUTEX_INITIALIZER,
+ .cond = PTHREAD_COND_INITIALIZER,
+ .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
/* Shared between threads */
@@ -213,6 +218,17 @@ enum consumerd_state {
static enum consumerd_state ust_consumerd_state;
static enum consumerd_state kernel_consumerd_state;
+/*
+ * This is a flag condition indicating that the consumer thread is ready and
+ * connected to the lttng-consumerd daemon.
+ *
+ * A value of 0 indicates that the thread is NOT ready. A value of 1 means that
+ * the thread consumer did connect successfully to the lttng-consumerd daemon.
+ * A negative value means that there is been an error and the thread as likely
+ * quit.
+ */
+static int consumer_thread_is_ready;
+
/* Used for the health monitoring of the session daemon. See health.h */
struct health_state health_thread_cmd;
struct health_state health_thread_app_manage;
@@ -789,6 +805,16 @@ error_poll_create:
}
/*
+ * Signal pthread condition of the consumer data that the thread.
+ */
+static void signal_consumer_condition(struct consumer_data *data)
+{
+ pthread_mutex_lock(&data->cond_mutex);
+ pthread_cond_signal(&data->cond);
+ pthread_mutex_unlock(&data->cond_mutex);
+}
+
+/*
* This thread manage the consumer error sent back to the session daemon.
*/
static void *thread_manage_consumer(void *data)
@@ -801,6 +827,9 @@ static void *thread_manage_consumer(void *data)
DBG("[thread] Manage consumer started");
+ /* Make sure we set the readiness flag to 0 because we are NOT ready */
+ consumer_thread_is_ready = 0;
+
health_code_update(&consumer_data->health);
ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
@@ -886,13 +915,16 @@ restart:
consumer_data->cmd_sock =
lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path);
if (consumer_data->cmd_sock < 0) {
- sem_post(&consumer_data->sem);
+ /* On error, signal condition and quit. */
+ consumer_thread_is_ready = -1;
+ signal_consumer_condition(consumer_data);
PERROR("consumer connect");
goto error;
}
- /* Signal condition to tell that the kconsumerd is ready */
- sem_post(&consumer_data->sem);
- DBG("consumer command socket ready");
+ /* Signal condition to tell that the consumerd is ready */
+ consumer_thread_is_ready = 1;
+ signal_consumer_condition(consumer_data);
+ DBG("Consumer command socket ready");
} else {
ERR("consumer error when waiting for SOCK_READY : %s",
lttcomm_get_readable_code(-code));
@@ -1446,16 +1478,33 @@ error_create_poll:
*/
static int spawn_consumer_thread(struct consumer_data *consumer_data)
{
- int ret;
+ int ret, clock_ret;
struct timespec timeout;
- timeout.tv_sec = DEFAULT_SEM_WAIT_TIMEOUT;
- timeout.tv_nsec = 0;
+ /* Setup pthread condition */
+ ret = pthread_condattr_init(&consumer_data->condattr);
+ if (ret != 0) {
+ errno = ret;
+ PERROR("pthread_condattr_init consumer data");
+ goto error;
+ }
- /* Setup semaphore */
- ret = sem_init(&consumer_data->sem, 0, 0);
- if (ret < 0) {
- PERROR("sem_init consumer semaphore");
+ /*
+ * Set the monotonic clock in order to make sure we DO NOT jump in time
+ * between the clock_gettime() call and the timedwait call. See bug #324
+ * for a more details and how we noticed it.
+ */
+ ret = pthread_condattr_setclock(&consumer_data->condattr,
CLOCK_MONOTONIC);
+ if (ret != 0) {
+ errno = ret;
+ PERROR("pthread_condattr_setclock consumer data");
+ goto error;
+ }
+
+ ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
+ if (ret != 0) {
+ errno = ret;
+ PERROR("pthread_cond_init consumer data");
goto error;
}
@@ -1467,31 +1516,48 @@ static int spawn_consumer_thread(struct consumer_data
*consumer_data)
goto error;
}
+ /* We are about to wait on a pthread condition */
+ pthread_mutex_lock(&consumer_data->cond_mutex);
+
/* Get time for sem_timedwait absolute timeout */
- ret = clock_gettime(CLOCK_REALTIME, &timeout);
- if (ret < 0) {
- PERROR("clock_gettime spawn consumer");
- /* Infinite wait for the kconsumerd thread to be ready */
- ret = sem_wait(&consumer_data->sem);
- } else {
- /* Normal timeout if the gettime was successful */
- timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
- ret = sem_timedwait(&consumer_data->sem, &timeout);
+ clock_ret = clock_gettime(CLOCK_MONOTONIC, &timeout);
+ while (!consumer_thread_is_ready) {
+ if (clock_ret < 0) {
+ PERROR("clock_gettime spawn consumer");
+ /* Infinite wait for the consumerd thread to be ready */
+ ret = pthread_cond_wait(&consumer_data->cond,
+ &consumer_data->cond_mutex);
+ } else {
+ /* Normal timeout if the gettime was successful */
+ timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;
+ ret = pthread_cond_timedwait(&consumer_data->cond,
+ &consumer_data->cond_mutex, &timeout);
+ }
}
- if (ret < 0) {
- if (errno == ETIMEDOUT) {
+ /*
+ * Reset back the flag so we could respawn a consumer thread at some
point
+ * and avoid skipping the above readiness wait period.
+ */
+ consumer_thread_is_ready = 0;
+
+ pthread_mutex_unlock(&consumer_data->cond_mutex);
+
+ if (ret != 0) {
+ errno = ret;
+ if (ret == ETIMEDOUT) {
/*
* Call has timed out so we kill the kconsumerd_thread
and return
* an error.
*/
- ERR("The consumer thread was never ready. Killing it");
+ ERR("Condition timed out. The consumer thread was never
ready."
+ " Killing it");
ret = pthread_cancel(consumer_data->thread);
if (ret < 0) {
PERROR("pthread_cancel consumer thread");
}
} else {
- PERROR("semaphore wait failed consumer thread");
+ PERROR("pthread_cond_wait failed consumer thread");
}
goto error;
}
--
1.7.10.4
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
