Fix: consumerd metadata channel/cache/timer races
[lttng-tools.git] / src / common / ust-consumer / ust-consumer.c
index 0edb513f89c15263b2244983dcbf22626f95c734..031a7cb261a52fa8d6b0c4c53be291be5896b973 100644 (file)
@@ -581,6 +581,7 @@ static int flush_channel(uint64_t chan_key)
 
        DBG("UST consumer flush channel key %" PRIu64, chan_key);
 
+       rcu_read_lock();
        channel = consumer_find_channel(chan_key);
        if (!channel) {
                ERR("UST consumer flush channel %" PRIu64 " not found", chan_key);
@@ -591,26 +592,25 @@ static int flush_channel(uint64_t chan_key)
        ht = consumer_data.stream_per_chan_id_ht;
 
        /* For each stream of the channel id, flush it. */
-       rcu_read_lock();
        cds_lfht_for_each_entry_duplicate(ht->ht,
                        ht->hash_fct(&channel->key, lttng_ht_seed), ht->match_fct,
                        &channel->key, &iter.iter, stream, node_channel_id.node) {
                        ustctl_flush_buffer(stream->ustream, 1);
        }
-       rcu_read_unlock();
-
 error:
+       rcu_read_unlock();
        return ret;
 }
 
 /*
  * Close metadata stream wakeup_fd using the given key to retrieve the channel.
+ * RCU read side lock MUST be acquired before calling this function.
  *
  * Return 0 on success else an LTTng error code.
  */
 static int close_metadata(uint64_t chan_key)
 {
-       int ret;
+       int ret = 0;
        struct lttng_consumer_channel *channel;
 
        DBG("UST consumer close metadata key %" PRIu64, chan_key);
@@ -622,18 +622,22 @@ static int close_metadata(uint64_t chan_key)
                goto error;
        }
 
-       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);
-               ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
-               goto error;
-       }
-       if (channel->switch_timer_enabled == 1) {
-               DBG("Deleting timer on metadata channel");
-               consumer_timer_switch_stop(channel);
+       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);
+               }
+               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);
+                       ret = LTTCOMM_CONSUMERD_ERROR_METADATA;
+                       goto error_unlock;
+               }
        }
-       consumer_metadata_cache_destroy(channel);
 
+error_unlock:
+       pthread_mutex_unlock(&consumer_data.lock);
 error:
        return ret;
 }
@@ -883,6 +887,12 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                case LTTNG_UST_CHAN_PER_CPU:
                        channel->type = CONSUMER_CHANNEL_TYPE_DATA;
                        attr.type = LTTNG_UST_CHAN_PER_CPU;
+                       /*
+                        * Set refcount to 1 for owner. Below, we will
+                        * pass ownership to the
+                        * consumer_thread_channel_poll() thread.
+                        */
+                       channel->refcount = 1;
                        break;
                case LTTNG_UST_CHAN_METADATA:
                        channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
@@ -898,17 +908,34 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                        goto end_channel_error;
                }
 
+               if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
+                       ret = consumer_metadata_cache_allocate(channel);
+                       if (ret < 0) {
+                               ERR("Allocating metadata cache");
+                               goto end_channel_error;
+                       }
+                       consumer_timer_switch_start(channel, attr.switch_timer_interval);
+                       attr.switch_timer_interval = 0;
+               }
+
                /*
                 * Add the channel to the internal state AFTER all streams were created
                 * and successfully sent to session daemon. This way, all streams must
                 * be ready before this channel is visible to the threads.
+                * If add_channel succeeds, ownership of the channel is
+                * passed to consumer_thread_channel_poll().
                 */
                ret = add_channel(channel, ctx);
                if (ret < 0) {
+                       if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
+                               if (channel->switch_timer_enabled == 1) {
+                                       consumer_timer_switch_stop(channel);
+                               }
+                               consumer_metadata_cache_destroy(channel);
+                       }
                        goto end_channel_error;
                }
 
-
                /*
                 * Channel and streams are now created. Inform the session daemon that
                 * everything went well and should wait to receive the channel and
@@ -923,16 +950,6 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                        goto end_nosignal;
                }
 
-               if (msg.u.ask_channel.type == LTTNG_UST_CHAN_METADATA) {
-                       ret = consumer_metadata_cache_allocate(channel);
-                       if (ret < 0) {
-                               ERR("Allocating metadata cache");
-                               goto end_channel_error;
-                       }
-                       consumer_timer_switch_start(channel, attr.switch_timer_interval);
-                       attr.switch_timer_interval = 0;
-               }
-
                break;
        }
        case LTTNG_CONSUMER_GET_CHANNEL:
@@ -990,17 +1007,13 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
        case LTTNG_CONSUMER_DESTROY_CHANNEL:
        {
                uint64_t key = msg.u.destroy_channel.key;
-               struct lttng_consumer_channel *channel;
-
-               channel = consumer_find_channel(key);
-               if (!channel) {
-                       ERR("UST consumer get channel key %" PRIu64 " not found", key);
-                       ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
-                       goto end_msg_sessiond;
-               }
-
-               destroy_channel(channel);
 
+               /*
+                * Only called if streams have not been sent to stream
+                * manager thread. However, channel has been sent to
+                * channel manager thread.
+                */
+               notify_thread_del_channel(ctx, key);
                goto end_msg_sessiond;
        }
        case LTTNG_CONSUMER_CLOSE_METADATA:
@@ -1040,6 +1053,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                if (!channel) {
                        ERR("UST consumer push metadata %" PRIu64 " not found", key);
                        ret_code = LTTNG_ERR_UST_CHAN_NOT_FOUND;
+                       goto end_msg_sessiond;
                }
 
                /* Tell session daemon we are ready to receive the metadata. */
@@ -1188,6 +1202,10 @@ void lttng_ustconsumer_del_channel(struct lttng_consumer_channel *chan)
        assert(chan);
        assert(chan->uchan);
 
+       if (chan->switch_timer_enabled == 1) {
+               consumer_timer_switch_stop(chan);
+       }
+       consumer_metadata_cache_destroy(chan);
        ustctl_destroy_channel(chan->uchan);
 }
 
@@ -1196,6 +1214,9 @@ void lttng_ustconsumer_del_stream(struct lttng_consumer_stream *stream)
        assert(stream);
        assert(stream->ustream);
 
+       if (stream->chan->switch_timer_enabled == 1) {
+               consumer_timer_switch_stop(stream->chan);
+       }
        ustctl_destroy_stream(stream->ustream);
 }
 
This page took 0.02671 seconds and 5 git commands to generate.