.type = LTTNG_CONSUMER_UNKNOWN,
};
-/* timeout parameter, to control the polling thread grace period. */
-int consumer_poll_timeout = -1;
-
/*
* Flag to inform the polling thread to quit when all fd hung up. Updated by
* the consumer_thread_receive_fds when it notices that all fds has hung up.
volatile int consumer_quit;
/*
- * The following two hash tables are visible by all threads which are separated
- * in different source files.
- *
* Global hash table containing respectively metadata and data streams. The
* stream element in this ht should only be updated by the metadata poll thread
* for the metadata and the data poll thread for the data.
*/
-struct lttng_ht *metadata_ht;
-struct lttng_ht *data_ht;
+static struct lttng_ht *metadata_ht;
+static struct lttng_ht *data_ht;
/*
* Notify a thread pipe to poll back again. This usually means that some global
rcu_read_unlock();
}
+/*
+ * Return a channel object for the given key.
+ *
+ * RCU read side lock MUST be acquired before calling this function and
+ * protects the channel ptr.
+ */
static struct lttng_consumer_channel *consumer_find_channel(int key)
{
struct lttng_ht_iter iter;
return NULL;
}
- rcu_read_lock();
-
lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key),
&iter);
node = lttng_ht_iter_get_node_ulong(&iter);
channel = caa_container_of(node, struct lttng_consumer_channel, node);
}
- rcu_read_unlock();
-
return channel;
}
assert(relayd);
+ DBG("Cleaning up relayd sockets");
+
/* Save the net sequence index before destroying the object */
netidx = relayd->net_seq_idx;
goto free_stream;
}
- pthread_mutex_lock(&stream->lock);
pthread_mutex_lock(&consumer_data.lock);
+ pthread_mutex_lock(&stream->lock);
switch (consumer_data.type) {
case LTTNG_CONSUMER_KERNEL:
end:
consumer_data.need_update = 1;
- pthread_mutex_unlock(&consumer_data.lock);
pthread_mutex_unlock(&stream->lock);
+ pthread_mutex_unlock(&consumer_data.lock);
if (free_chan) {
consumer_del_channel(free_chan);
goto end;
}
+ rcu_read_lock();
+
/*
* Get stream's channel reference. Needed when adding the stream to the
* global hash table.
stream->path_name, stream->key, stream->shm_fd, stream->wait_fd,
(unsigned long long) stream->mmap_len, stream->out_fd,
stream->net_seq_idx, stream->session_id);
+
+ rcu_read_unlock();
return stream;
error:
+ rcu_read_unlock();
free(stream);
end:
return NULL;
DBG3("Adding consumer stream %d", stream->key);
pthread_mutex_lock(&consumer_data.lock);
+ pthread_mutex_lock(&stream->lock);
rcu_read_lock();
/* Steal stream identifier to avoid having streams with the same key */
consumer_data.need_update = 1;
rcu_read_unlock();
+ pthread_mutex_unlock(&stream->lock);
pthread_mutex_unlock(&consumer_data.lock);
return ret;
* changed where this function will be called back again.
*/
if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
- stream->endpoint_status) {
+ stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
continue;
}
DBG("Active FD %d", stream->wait_fd);
* core function for writing trace buffers to either the local filesystem or
* the network.
*
+ * It must be called with the stream lock held.
+ *
* Careful review MUST be put if any changes occur!
*
* Returns the number of bytes written
/* RCU lock for the relayd pointer */
rcu_read_lock();
- pthread_mutex_lock(&stream->lock);
-
/* Flag that the current stream if set for network streaming. */
if (stream->net_seq_idx != -1) {
relayd = consumer_find_relayd(stream->net_seq_idx);
if (relayd && stream->metadata_flag) {
pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
}
- pthread_mutex_unlock(&stream->lock);
rcu_read_unlock();
return written;
/*
* Splice the data from the ring buffer to the tracefile.
*
+ * It must be called with the stream lock held.
+ *
* Returns the number of bytes spliced.
*/
ssize_t lttng_consumer_on_read_subbuffer_splice(
/* RCU lock for the relayd pointer */
rcu_read_lock();
- pthread_mutex_lock(&stream->lock);
-
/* Flag that the current stream if set for network streaming. */
if (stream->net_seq_idx != -1) {
relayd = consumer_find_relayd(stream->net_seq_idx);
written = ret_splice;
}
/* Socket operation failed. We consider the relayd dead */
- if (errno == EBADF) {
+ if (errno == EBADF || errno == EPIPE) {
WARN("Remote relayd disconnected. Stopping");
relayd_hang_up = 1;
goto write_error;
if (relayd && stream->metadata_flag) {
pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
}
- pthread_mutex_unlock(&stream->lock);
rcu_read_unlock();
return written;
goto free_stream;
}
+ pthread_mutex_lock(&consumer_data.lock);
pthread_mutex_lock(&stream->lock);
- pthread_mutex_lock(&consumer_data.lock);
switch (consumer_data.type) {
case LTTNG_CONSUMER_KERNEL:
if (stream->mmap_base != NULL) {
}
end:
- pthread_mutex_unlock(&consumer_data.lock);
pthread_mutex_unlock(&stream->lock);
+ pthread_mutex_unlock(&consumer_data.lock);
if (free_chan) {
consumer_del_channel(free_chan);
{
int ret = 0;
struct consumer_relayd_sock_pair *relayd;
+ struct lttng_ht_iter iter;
+ struct lttng_ht_node_ulong *node;
assert(stream);
assert(ht);
DBG3("Adding metadata stream %d to hash table", stream->wait_fd);
pthread_mutex_lock(&consumer_data.lock);
+ pthread_mutex_lock(&stream->lock);
/*
* From here, refcounts are updated so be _careful_ when returning an error
*/
rcu_read_lock();
+
+ /*
+ * Lookup the stream just to make sure it does not exist in our internal
+ * state. This should NEVER happen.
+ */
+ lttng_ht_lookup(ht, (void *)((unsigned long) stream->wait_fd), &iter);
+ node = lttng_ht_iter_get_node_ulong(&iter);
+ assert(!node);
+
/* Find relayd and, if one is found, increment refcount. */
relayd = consumer_find_relayd(stream->net_seq_idx);
if (relayd != NULL) {
uatomic_dec(&stream->chan->nb_init_streams);
}
- /* Steal stream identifier to avoid having streams with the same key */
- consumer_steal_stream_key(stream->key, ht);
-
lttng_ht_add_unique_ulong(ht, &stream->node);
/*
rcu_read_unlock();
+ pthread_mutex_unlock(&stream->lock);
pthread_mutex_unlock(&consumer_data.lock);
return ret;
}
rcu_read_lock();
cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
/* Validate delete flag of the stream */
- if (!stream->endpoint_status) {
+ if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
continue;
}
/* Delete it right now */
rcu_read_lock();
cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
/* Validate delete flag of the stream */
- if (!stream->endpoint_status) {
+ if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
continue;
}
/*
* since their might be data to consume.
*/
lttng_poll_del(&events, ctx->consumer_metadata_pipe[0]);
- close(ctx->consumer_metadata_pipe[0]);
+ ret = close(ctx->consumer_metadata_pipe[0]);
+ if (ret < 0) {
+ PERROR("close metadata pipe");
+ }
continue;
} else if (revents & LPOLLIN) {
do {
/* poll on the array of fds */
restart:
DBG("polling on %d fd", nb_fd + 1);
- num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout);
+ num_rdy = poll(pollfd, nb_fd + 1, -1);
DBG("poll num_rdy : %d", num_rdy);
if (num_rdy == -1) {
/*
/* Take care of high priority channels first. */
for (i = 0; i < nb_fd; i++) {
+ if (local_stream[i] == NULL) {
+ continue;
+ }
if (pollfd[i].revents & POLLPRI) {
DBG("Urgent read on fd %d", pollfd[i].fd);
high_prio = 1;
if (len < 0 && len != -EAGAIN && len != -ENODATA) {
/* Clean the stream and free it. */
consumer_del_stream(local_stream[i], data_ht);
+ local_stream[i] = NULL;
} else if (len > 0) {
local_stream[i]->data_read = 1;
}
/* Take care of low priority channels. */
for (i = 0; i < nb_fd; i++) {
+ if (local_stream[i] == NULL) {
+ continue;
+ }
if ((pollfd[i].revents & POLLIN) ||
local_stream[i]->hangup_flush_done) {
DBG("Normal read on fd %d", pollfd[i].fd);
if (len < 0 && len != -EAGAIN && len != -ENODATA) {
/* Clean the stream and free it. */
consumer_del_stream(local_stream[i], data_ht);
+ local_stream[i] = NULL;
} else if (len > 0) {
local_stream[i]->data_read = 1;
}
/* Handle hangup and errors */
for (i = 0; i < nb_fd; i++) {
+ if (local_stream[i] == NULL) {
+ continue;
+ }
if (!local_stream[i]->hangup_flush_done
&& (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
&& (consumer_data.type == LTTNG_CONSUMER32_UST
|| consumer_data.type == LTTNG_CONSUMER64_UST)) {
DBG("fd %d is hup|err|nval. Attempting flush and read.",
- pollfd[i].fd);
+ pollfd[i].fd);
lttng_ustconsumer_on_stream_hangup(local_stream[i]);
/* Attempt read again, for the data we just flushed. */
local_stream[i]->data_read = 1;
DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
if (!local_stream[i]->data_read) {
consumer_del_stream(local_stream[i], data_ht);
+ local_stream[i] = NULL;
num_hup++;
}
} else if (pollfd[i].revents & POLLERR) {
ERR("Error returned in polling fd %d.", pollfd[i].fd);
if (!local_stream[i]->data_read) {
consumer_del_stream(local_stream[i], data_ht);
+ local_stream[i] = NULL;
num_hup++;
}
} else if (pollfd[i].revents & POLLNVAL) {
ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
if (!local_stream[i]->data_read) {
consumer_del_stream(local_stream[i], data_ht);
+ local_stream[i] = NULL;
num_hup++;
}
}
- local_stream[i]->data_read = 0;
+ if (local_stream[i] != NULL) {
+ local_stream[i]->data_read = 0;
+ }
}
}
end:
* only tracked fd in the poll set. The thread will take care of closing
* the read side.
*/
- close(ctx->consumer_metadata_pipe[1]);
+ ret = close(ctx->consumer_metadata_pipe[1]);
+ if (ret < 0) {
+ PERROR("close data pipe");
+ }
if (data_ht) {
destroy_data_stream_ht(data_ht);
*/
void *consumer_thread_sessiond_poll(void *data)
{
- int sock, client_socket, ret;
+ int sock = -1, client_socket, ret;
/*
* structure to poll for incoming data on communication socket avoids
* making blocking sockets.
goto end;
}
+ /* This socket is not useful anymore. */
+ ret = close(client_socket);
+ if (ret < 0) {
+ PERROR("close client_socket");
+ }
+ client_socket = -1;
+
/* update the polling structure to poll on the established socket */
consumer_sockpoll[1].fd = sock;
consumer_sockpoll[1].events = POLLIN | POLLPRI;
*/
consumer_quit = 1;
- /*
- * 2s of grace period, if no polling events occur during
- * this period, the polling thread will exit even if there
- * are still open FDs (should not happen, but safety mechanism).
- */
- consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT;
-
/*
* Notify the data poll thread to poll back again and test the
* consumer_quit state that we just set so to quit gracefully.
*/
notify_thread_pipe(ctx->consumer_data_pipe[1]);
+ /* Cleaning up possibly open sockets. */
+ if (sock >= 0) {
+ ret = close(sock);
+ if (ret < 0) {
+ PERROR("close sock sessiond poll");
+ }
+ }
+ if (client_socket >= 0) {
+ ret = close(sock);
+ if (ret < 0) {
+ PERROR("close client_socket sessiond poll");
+ }
+ }
+
rcu_unregister_thread();
return NULL;
}
ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
struct lttng_consumer_local_data *ctx)
{
+ ssize_t ret;
+
+ pthread_mutex_lock(&stream->lock);
+
switch (consumer_data.type) {
case LTTNG_CONSUMER_KERNEL:
- return lttng_kconsumer_read_subbuffer(stream, ctx);
+ ret = lttng_kconsumer_read_subbuffer(stream, ctx);
+ break;
case LTTNG_CONSUMER32_UST:
case LTTNG_CONSUMER64_UST:
- return lttng_ustconsumer_read_subbuffer(stream, ctx);
+ ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
+ break;
default:
ERR("Unknown consumer_data type");
assert(0);
- return -ENOSYS;
+ ret = -ENOSYS;
+ break;
}
+
+ pthread_mutex_unlock(&stream->lock);
+ return ret;
}
int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
struct lttng_consumer_local_data *ctx, int sock,
struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock)
{
- int fd, ret = -1;
+ int fd = -1, ret = -1;
struct consumer_relayd_sock_pair *relayd;
DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx);
if (ret != sizeof(fd)) {
lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
ret = -1;
+ fd = -1; /* Just in case it gets set with an invalid value. */
goto error;
}
/* Copy received lttcomm socket */
lttcomm_copy_sock(&relayd->control_sock, relayd_sock);
ret = lttcomm_create_sock(&relayd->control_sock);
+ /* Immediately try to close the created socket if valid. */
+ if (relayd->control_sock.fd >= 0) {
+ if (close(relayd->control_sock.fd)) {
+ PERROR("close relayd control socket");
+ }
+ }
+ /* Handle create_sock error. */
if (ret < 0) {
goto error;
}
- /* Close the created socket fd which is useless */
- close(relayd->control_sock.fd);
-
/* Assign new file descriptor */
relayd->control_sock.fd = fd;
break;
/* Copy received lttcomm socket */
lttcomm_copy_sock(&relayd->data_sock, relayd_sock);
ret = lttcomm_create_sock(&relayd->data_sock);
+ /* Immediately try to close the created socket if valid. */
+ if (relayd->data_sock.fd >= 0) {
+ if (close(relayd->data_sock.fd)) {
+ PERROR("close relayd data socket");
+ }
+ }
+ /* Handle create_sock error. */
if (ret < 0) {
goto error;
}
- /* Close the created socket fd which is useless */
- close(relayd->data_sock.fd);
-
/* Assign new file descriptor */
relayd->data_sock.fd = fd;
break;
add_relayd(relayd);
/* All good! */
- ret = 0;
+ return 0;
error:
+ /* Close received socket if valid. */
+ if (fd >= 0) {
+ if (close(fd)) {
+ PERROR("close received socket");
+ }
+ }
+ return ret;
+}
+
+/*
+ * Try to lock the stream mutex.
+ *
+ * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
+ */
+static int stream_try_lock(struct lttng_consumer_stream *stream)
+{
+ int ret;
+
+ assert(stream);
+
+ /*
+ * Try to lock the stream mutex. On failure, we know that the stream is
+ * being used else where hence there is data still being extracted.
+ */
+ ret = pthread_mutex_trylock(&stream->lock);
+ if (ret) {
+ /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
+ ret = 0;
+ goto end;
+ }
+
+ ret = 1;
+
+end:
return ret;
}
* Check if for a given session id there is still data needed to be extract
* from the buffers.
*
- * Return 1 if data is in fact available to be read or else 0.
+ * Return 1 if data is pending or else 0 meaning ready to be read.
*/
-int consumer_data_available(uint64_t id)
+int consumer_data_pending(uint64_t id)
{
int ret;
struct lttng_ht_iter iter;
struct lttng_ht *ht;
struct lttng_consumer_stream *stream;
struct consumer_relayd_sock_pair *relayd;
- int (*data_available)(struct lttng_consumer_stream *);
+ int (*data_pending)(struct lttng_consumer_stream *);
- DBG("Consumer data available command on session id %" PRIu64, id);
+ DBG("Consumer data pending command on session id %" PRIu64, id);
rcu_read_lock();
pthread_mutex_lock(&consumer_data.lock);
switch (consumer_data.type) {
case LTTNG_CONSUMER_KERNEL:
- data_available = lttng_kconsumer_data_available;
+ data_pending = lttng_kconsumer_data_pending;
break;
case LTTNG_CONSUMER32_UST:
case LTTNG_CONSUMER64_UST:
- data_available = lttng_ustconsumer_data_available;
+ data_pending = lttng_ustconsumer_data_pending;
break;
default:
ERR("Unknown consumer data type");
ht = consumer_data.stream_list_ht;
cds_lfht_for_each_entry_duplicate(ht->ht,
- ht->hash_fct((void *)((unsigned long) id), 0x42UL),
+ ht->hash_fct((void *)((unsigned long) id), lttng_ht_seed),
ht->match_fct, (void *)((unsigned long) id),
&iter.iter, stream, node_session_id.node) {
- /* Check the stream for data. */
- ret = data_available(stream);
- if (ret == 0) {
- goto data_not_available;
+ /* If this call fails, the stream is being used hence data pending. */
+ ret = stream_try_lock(stream);
+ if (!ret) {
+ goto data_not_pending;
+ }
+
+ /*
+ * A removed node from the hash table indicates that the stream has
+ * been deleted thus having a guarantee that the buffers are closed
+ * on the consumer side. However, data can still be transmitted
+ * over the network so don't skip the relayd check.
+ */
+ ret = cds_lfht_is_node_deleted(&stream->node.node);
+ if (!ret) {
+ /* Check the stream if there is data in the buffers. */
+ ret = data_pending(stream);
+ if (ret == 1) {
+ pthread_mutex_unlock(&stream->lock);
+ goto data_not_pending;
+ }
}
+ /* Relayd check */
if (stream->net_seq_idx != -1) {
relayd = consumer_find_relayd(stream->net_seq_idx);
- assert(relayd);
+ if (!relayd) {
+ /*
+ * At this point, if the relayd object is not available for the
+ * given stream, it is because the relayd is being cleaned up
+ * so every stream associated with it (for a session id value)
+ * are or will be marked for deletion hence no data pending.
+ */
+ pthread_mutex_unlock(&stream->lock);
+ goto data_not_pending;
+ }
- pthread_mutex_lock(&stream->lock);
pthread_mutex_lock(&relayd->ctrl_sock_mutex);
if (stream->metadata_flag) {
ret = relayd_quiescent_control(&relayd->control_sock);
} else {
- ret = relayd_data_available(&relayd->control_sock,
+ ret = relayd_data_pending(&relayd->control_sock,
stream->relayd_stream_id, stream->next_net_seq_num);
}
pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
- pthread_mutex_unlock(&stream->lock);
- if (ret == 0) {
- goto data_not_available;
+ if (ret == 1) {
+ pthread_mutex_unlock(&stream->lock);
+ goto data_not_pending;
}
}
+ pthread_mutex_unlock(&stream->lock);
}
/*
/* Data is available to be read by a viewer. */
pthread_mutex_unlock(&consumer_data.lock);
rcu_read_unlock();
- return 1;
+ return 0;
-data_not_available:
+data_not_pending:
/* Data is still being extracted from buffers. */
pthread_mutex_unlock(&consumer_data.lock);
rcu_read_unlock();
- return 0;
+ return 1;
}
projects / lttng-tools.git / blobdiff