Fix: consumer add channel return value was positive on error

[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 6acaacd63c942d1092ad9cdc056713f28a704462..780a601f67e74b70b3632a6d489067d82e1359ef 100644 (file)
--- a/src/common/ust-consumer/ust-consumer.c
+++ b/src/common/ust-consumer/ust-consumer.c
@@ -624,7 +624,13 @@ static int close_metadata(uint64_t chan_key)
 
        channel = consumer_find_channel(chan_key);
        if (!channel) {
-               ERR("UST consumer close metadata %" PRIu64 " not found", chan_key);
+               /*
+                * This is possible if the metadata thread has issue a delete because
+                * the endpoint point of the stream hung up. There is no way the
+                * session daemon can know about it thus use a DBG instead of an actual
+                * error.
+                */
+               DBG("UST consumer close metadata %" PRIu64 " not found", chan_key);
                ret = LTTNG_ERR_UST_CHAN_NOT_FOUND;
                goto error;
        }
@@ -751,6 +757,14 @@ int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
        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);
@@ -788,6 +802,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                 * The ret value might 0 meaning an orderly shutdown but this is ok
                 * since the caller handles this.
                 */
+               if (ret > 0) {
+                       ret = -1;
+               }
                return ret;
        }
        if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
@@ -863,6 +880,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                                sizeof(is_data_pending));
                if (ret < 0) {
                        DBG("Error when sending the data pending ret code: %d", ret);
+                       goto error_fatal;
                }
 
                /*
@@ -969,10 +987,9 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                ret = consumer_send_status_channel(sock, channel);
                if (ret < 0) {
                        /*
-                        * There is probably a problem on the socket so the poll will get
-                        * it and clean everything up.
+                        * There is probably a problem on the socket.
                         */
-                       goto end_nosignal;
+                       goto error_fatal;
                }
 
                break;
@@ -994,7 +1011,7 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                ret = consumer_send_status_msg(sock, LTTNG_OK);
                if (ret < 0) {
                        /* Somehow, the session daemon is not responding anymore. */
-                       goto end_nosignal;
+                       goto error_fatal;
                }
 
                /* Send everything to sessiond. */
@@ -1090,14 +1107,14 @@ int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 
                /* Wait for more data. */
                if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
-                       goto end_nosignal;
+                       goto error_fatal;
                }
 
                ret = lttng_ustconsumer_recv_metadata(sock, key, offset,
                                len, channel);
                if (ret < 0) {
                        /* error receiving from sessiond */
-                       goto end_nosignal;
+                       goto error_fatal;
                } else {
                        ret_code = ret;
                        goto end_msg_sessiond;
@@ -1132,7 +1149,10 @@ end_msg_sessiond:
         * the caller because the session daemon socket management is done
         * elsewhere. Returning a negative code or 0 will shutdown the consumer.
         */
-       (void) consumer_send_status_msg(sock, ret_code);
+       ret = consumer_send_status_msg(sock, ret_code);
+       if (ret < 0) {
+               goto error_fatal;
+       }
        rcu_read_unlock();
        return 1;
 end_channel_error: