Fix: skip metadata flushed check if write failed

[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c

diff --git a/src/common/ust-consumer/ust-consumer.c b/src/common/ust-consumer/ust-consumer.c
index e0280f1489e2e8defa25e1311023f4f518e4d666..f7ffb0febf5b7bcdc338b30975ae7e3379724935 100644 (file)
--- a/src/common/ust-consumer/ust-consumer.c
+++ b/src/common/ust-consumer/ust-consumer.c
@@ -491,10 +491,6 @@ static int ask_channel(struct lttng_consumer_local_data *ctx, int sock,
 
        channel->wait_fd = ustctl_channel_get_wait_fd(channel->uchan);
 
-       if (ret < 0) {
-               goto error;
-       }
-
        /* Open all streams for this channel. */
        ret = create_ust_streams(channel, ctx);
        if (ret < 0) {
@@ -558,6 +554,11 @@ int lttng_ustconsumer_push_metadata(struct lttng_consumer_channel *metadata,
 
        DBG("UST consumer writing metadata to channel %s", metadata->name);
 
+       if (!metadata->metadata_stream) {
+               ret = 0;
+               goto error;
+       }
+
        assert(target_offset <= metadata->metadata_cache->max_offset);
        ret = ustctl_write_metadata_to_channel(metadata->uchan,
                        metadata_str + target_offset, len);
@@ -629,11 +630,17 @@ static int close_metadata(uint64_t chan_key)
        }
 
        pthread_mutex_lock(&consumer_data.lock);
-       if (!cds_lfht_is_node_deleted(&channel->node.node)) {
-               if (channel->switch_timer_enabled == 1) {
-                       DBG("Deleting timer on metadata channel");
-                       consumer_timer_switch_stop(channel);
-               }
+
+       if (cds_lfht_is_node_deleted(&channel->node.node)) {
+               goto error_unlock;
+       }
+
+       if (channel->switch_timer_enabled == 1) {
+               DBG("Deleting timer on metadata channel");
+               consumer_timer_switch_stop(channel);
+       }
+
+       if (channel->metadata_stream) {
                ret = ustctl_stream_close_wakeup_fd(channel->metadata_stream->ustream);
                if (ret < 0) {
                        ERR("UST consumer unable to close fd of metadata (ret: %d)", ret);
@@ -728,13 +735,33 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
                goto end_free;
        }
 
+       /*
+        * XXX: The consumer data lock is acquired before calling metadata cache
+        * write which calls push metadata that MUST be protected by the consumer
+        * lock in order to be able to check the validity of the metadata stream of
+        * the channel.
+        *
+        * Note that this will be subject to change to better fine grained locking
+        * and ultimately try to get rid of this global consumer data lock.
+        */
+       pthread_mutex_lock(&consumer_data.lock);
+
        pthread_mutex_lock(&channel->metadata_cache->lock);
        ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
        if (ret < 0) {
                /* Unable to handle metadata. Notify session daemon. */
                ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
+               /*
+                * Skip metadata flush on write error since the offset and len might
+                * not have been updated which could create an infinite loop below when
+                * waiting for the metadata cache to be flushed.
+                */
+               pthread_mutex_unlock(&channel->metadata_cache->lock);
+               pthread_mutex_unlock(&consumer_data.lock);
+               goto end_free;
        }
        pthread_mutex_unlock(&channel->metadata_cache->lock);
+       pthread_mutex_unlock(&consumer_data.lock);
 
        while (consumer_metadata_cache_flushed(channel, offset + len)) {
                DBG("Waiting for metadata to be flushed");