Fix: leaked chunk reference in lttng_consumer_create_trace_chunk

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

diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c
index 82be751c40ef82e83e47e10e785caa0d131cffbe..e1426f116ccc939c5f343343993ff4d70d3ea288 100644 (file)
--- a/src/common/consumer/consumer.c
+++ b/src/common/consumer/consumer.c
@@ -1408,6 +1408,7 @@ void lttng_consumer_cleanup(void)
 {
        struct lttng_ht_iter iter;
        struct lttng_consumer_channel *channel;
+       unsigned int trace_chunks_left;
 
        rcu_read_lock();
 
@@ -1432,6 +1433,27 @@ void lttng_consumer_cleanup(void)
         */
        lttng_ht_destroy(consumer_data.stream_list_ht);
 
+       /*
+        * Trace chunks in the registry may still exist if the session
+        * daemon has encountered an internal error and could not
+        * tear down its sessions and/or trace chunks properly.
+        *
+        * Release the session daemon's implicit reference to any remaining
+        * trace chunk and print an error if any trace chunk was found. Note
+        * that there are _no_ legitimate cases for trace chunks to be left,
+        * it is a leak. However, it can happen following a crash of the
+        * session daemon and not emptying the registry would cause an assertion
+        * to hit.
+        */
+       trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk(
+                       consumer_data.chunk_registry);
+       if (trace_chunks_left) {
+               ERR("%u trace chunks are leaked by lttng-consumerd. "
+                               "This can be caused by an internal error of the session daemon.",
+                               trace_chunks_left);
+       }
+       /* Run all callbacks freeing each chunk. */
+       rcu_barrier();
        lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry);
 }
 
@@ -4439,7 +4461,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
 {
        int ret;
        enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
-       struct lttng_trace_chunk *created_chunk, *published_chunk;
+       struct lttng_trace_chunk *created_chunk = NULL, *published_chunk = NULL;
        enum lttng_trace_chunk_status chunk_status;
        char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)];
        char creation_timestamp_buffer[ISO8601_STR_LEN];
@@ -4494,7 +4516,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
        if (!created_chunk) {
                ERR("Failed to create trace chunk");
                ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
-               goto end;
+               goto error;
        }
 
        if (chunk_override_name) {
@@ -4502,7 +4524,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
                                chunk_override_name);
                if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
-                       goto end;
+                       goto error;
                }
        }
 
@@ -4512,7 +4534,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
                if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        ERR("Failed to set trace chunk credentials");
                        ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
-                       goto end;
+                       goto error;
                }
                /*
                 * The consumer daemon has no ownership of the chunk output
@@ -4523,7 +4545,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
                if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        ERR("Failed to set trace chunk's directory handle");
                        ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
-                       goto end;
+                       goto error;
                }
        }
 
@@ -4535,7 +4557,7 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
        if (!published_chunk) {
                ERR("Failed to publish trace chunk");
                ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
-               goto end;
+               goto error;
        }
 
        rcu_read_lock();
@@ -4558,11 +4580,13 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
                         * channels.
                         */
                        enum lttcomm_return_code close_ret;
+                       char path[LTTNG_PATH_MAX];
 
                        DBG("Failed to set new trace chunk on existing channels, rolling back");
                        close_ret = lttng_consumer_close_trace_chunk(relayd_id,
                                        session_id, chunk_id,
-                                       chunk_creation_timestamp, NULL);
+                                       chunk_creation_timestamp, NULL,
+                                       path);
                        if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) {
                                ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64,
                                                session_id, chunk_id);
@@ -4588,12 +4612,13 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
 
                if (!relayd || ret) {
                        enum lttcomm_return_code close_ret;
+                       char path[LTTNG_PATH_MAX];
 
                        close_ret = lttng_consumer_close_trace_chunk(relayd_id,
                                        session_id,
                                        chunk_id,
                                        chunk_creation_timestamp,
-                                       NULL);
+                                       NULL, path);
                        if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) {
                                ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64,
                                                session_id,
@@ -4601,21 +4626,23 @@ enum lttcomm_return_code lttng_consumer_create_trace_chunk(
                        }
 
                        ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
-                       goto error;
+                       goto error_unlock;
                }
        }
-error:
+error_unlock:
        rcu_read_unlock();
+error:
        /* Release the reference returned by the "publish" operation. */
        lttng_trace_chunk_put(published_chunk);
-end:
+       lttng_trace_chunk_put(created_chunk);
        return ret_code;
 }
 
 enum lttcomm_return_code lttng_consumer_close_trace_chunk(
                const uint64_t *relayd_id, uint64_t session_id,
                uint64_t chunk_id, time_t chunk_close_timestamp,
-               const enum lttng_trace_chunk_command_type *close_command)
+               const enum lttng_trace_chunk_command_type *close_command,
+               char *path)
 {
        enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
        struct lttng_trace_chunk *chunk;
@@ -4713,7 +4740,8 @@ enum lttcomm_return_code lttng_consumer_close_trace_chunk(
                if (relayd) {
                        pthread_mutex_lock(&relayd->ctrl_sock_mutex);
                        ret = relayd_close_trace_chunk(
-                                       &relayd->control_sock, chunk);
+                                       &relayd->control_sock, chunk,
+                                       path);
                        pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
                } else {
                        ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64,