Fix: perform relayd socket pair cleanup on control socket error

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

diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c
index e0d6ea496ade52af826cf02b0fc9fe0054b64895..63a2ab288c865e338582485199a7a268064bb4af 100644 (file)
--- a/src/common/consumer/consumer.c
+++ b/src/common/consumer/consumer.c
@@ -465,14 +465,13 @@ static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
  * If a local data context is available, notify the threads that the streams'
  * state have changed.
  */
-static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
-               struct lttng_consumer_local_data *ctx)
+void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd)
 {
        uint64_t netidx;
 
        assert(relayd);
 
-       DBG("Cleaning up relayd sockets");
+       DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx);
 
        /* Save the net sequence index before destroying the object */
        netidx = relayd->net_seq_idx;
@@ -492,10 +491,8 @@ static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
         * memory barrier ordering the updates of the end point status from the
         * read of this status which happens AFTER receiving this notify.
         */
-       if (ctx) {
-               notify_thread_lttng_pipe(ctx->consumer_data_pipe);
-               notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
-       }
+       notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe);
+       notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe);
 }
 
 /*
@@ -813,6 +810,8 @@ int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
                                stream->trace_archive_id);
                pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
                if (ret < 0) {
+                       ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+                       lttng_consumer_cleanup_relayd(relayd);
                        goto end;
                }
 
@@ -854,6 +853,8 @@ int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
                ret = relayd_streams_sent(&relayd->control_sock);
                pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
                if (ret < 0) {
+                       ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+                       lttng_consumer_cleanup_relayd(relayd);
                        goto end;
                }
        } else {
@@ -1721,7 +1722,8 @@ write_error:
         * cleanup the relayd object and all associated streams.
         */
        if (relayd && relayd_hang_up) {
-               cleanup_relayd(relayd, ctx);
+               ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+               lttng_consumer_cleanup_relayd(relayd);
        }
 
 end:
@@ -1947,7 +1949,8 @@ write_error:
         * cleanup the relayd object and all associated streams.
         */
        if (relayd && relayd_hang_up) {
-               cleanup_relayd(relayd, ctx);
+               ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+               lttng_consumer_cleanup_relayd(relayd);
                /* Skip splice error so the consumer does not fail */
                goto end;
        }
@@ -3646,6 +3649,7 @@ error:
         * Add relayd socket pair to consumer data hashtable. If object already
         * exists or on error, the function gracefully returns.
         */
+       relayd->ctx = ctx;
        add_relayd(relayd);
 
        /* All good! */
@@ -3772,6 +3776,8 @@ int consumer_data_pending(uint64_t id)
                pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
                if (ret < 0) {
                        /* Communication error thus the relayd so no data pending. */
+                       ERR("Relayd begin data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+                       lttng_consumer_cleanup_relayd(relayd);
                        goto data_not_pending;
                }
        }
@@ -3813,6 +3819,13 @@ int consumer_data_pending(uint64_t id)
                                                stream->relayd_stream_id,
                                                stream->next_net_seq_num - 1);
                        }
+                       if (ret < 0) {
+                               ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+                               lttng_consumer_cleanup_relayd(relayd);
+                               pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
+                               pthread_mutex_unlock(&stream->lock);
+                               goto data_not_pending;
+                       }
                        pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
                        if (ret == 1) {
                                pthread_mutex_unlock(&stream->lock);
@@ -3831,6 +3844,8 @@ int consumer_data_pending(uint64_t id)
                                relayd->relayd_session_id, &is_data_inflight);
                pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
                if (ret < 0) {
+                       ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+                       lttng_consumer_cleanup_relayd(relayd);
                        goto data_not_pending;
                }
                if (is_data_inflight) {