Fix: consumer: snapshot error return code
[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.c
index 7c01bc772a008aa38a05a399c7acadf4849fd0a2..b890416cac1d96db4fe481c27f9f395e4a14707d 100644 (file)
@@ -124,28 +124,22 @@ int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
 
 /*
  * Take a snapshot of all the stream of a channel
+ * RCU read-side lock must be held across this function to ensure existence of
+ * channel.
  *
  * Returns 0 on success, < 0 on error
  */
-int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
-               uint64_t relayd_id, uint64_t nb_packets_per_stream,
+int lttng_kconsumer_snapshot_channel(struct lttng_consumer_channel *channel,
+               uint64_t key, char *path, uint64_t relayd_id, uint64_t nb_packets_per_stream,
                struct lttng_consumer_local_data *ctx)
 {
        int ret;
-       struct lttng_consumer_channel *channel;
        struct lttng_consumer_stream *stream;
 
        DBG("Kernel consumer snapshot channel %" PRIu64, key);
 
        rcu_read_lock();
 
-       channel = consumer_find_channel(key);
-       if (!channel) {
-               ERR("No channel found for key %" PRIu64, key);
-               ret = -1;
-               goto end;
-       }
-
        /* Splice is not supported yet for channel snapshot. */
        if (channel->output != CONSUMER_CHANNEL_MMAP) {
                ERR("Unsupported output %d", channel->output);
@@ -333,15 +327,17 @@ end:
 
 /*
  * Read the whole metadata available for a snapshot.
+ * RCU read-side lock must be held across this function to ensure existence of
+ * metadata_channel.
  *
  * Returns 0 on success, < 0 on error
  */
-int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
-               uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
+static int lttng_kconsumer_snapshot_metadata(struct lttng_consumer_channel *metadata_channel,
+               uint64_t key, char *path, uint64_t relayd_id,
+               struct lttng_consumer_local_data *ctx)
 {
        int ret, use_relayd = 0;
        ssize_t ret_read;
-       struct lttng_consumer_channel *metadata_channel;
        struct lttng_consumer_stream *metadata_stream;
 
        assert(ctx);
@@ -351,15 +347,9 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 
        rcu_read_lock();
 
-       metadata_channel = consumer_find_channel(key);
-       if (!metadata_channel) {
-               ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
-               ret = -1;
-               goto error;
-       }
-
        metadata_stream = metadata_channel->metadata_stream;
        assert(metadata_stream);
+       pthread_mutex_lock(&metadata_stream->lock);
 
        /* Flag once that we have a valid relayd for the stream. */
        if (relayd_id != (uint64_t) -1ULL) {
@@ -369,7 +359,7 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
        if (use_relayd) {
                ret = consumer_send_relayd_stream(metadata_stream, path);
                if (ret < 0) {
-                       goto error;
+                       goto error_snapshot;
                }
        } else {
                ret = utils_create_stream_file(path, metadata_stream->name,
@@ -377,7 +367,7 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
                                metadata_stream->tracefile_count_current,
                                metadata_stream->uid, metadata_stream->gid, NULL);
                if (ret < 0) {
-                       goto error;
+                       goto error_snapshot;
                }
                metadata_stream->out_fd = ret;
        }
@@ -390,7 +380,8 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
                        if (ret_read != -EAGAIN) {
                                ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
                                                ret_read);
-                               goto error;
+                               ret = ret_read;
+                               goto error_snapshot;
                        }
                        /* ret_read is negative at this point so we will exit the loop. */
                        continue;
@@ -415,11 +406,11 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
        }
 
        ret = 0;
-
+error_snapshot:
+       pthread_mutex_unlock(&metadata_stream->lock);
        cds_list_del(&metadata_stream->send_node);
        consumer_stream_destroy(metadata_stream, NULL);
        metadata_channel->metadata_stream = NULL;
-error:
        rcu_read_unlock();
        return ret;
 }
@@ -644,7 +635,8 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                                msg.u.stream.cpu,
                                &alloc_ret,
                                channel->type,
-                               channel->monitor);
+                               channel->monitor,
+                               msg.u.stream.trace_archive_id);
                if (new_stream == NULL) {
                        switch (alloc_ret) {
                        case -ENOMEM:
@@ -896,26 +888,34 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
        }
        case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
        {
-               if (msg.u.snapshot_channel.metadata == 1) {
-                       ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
-                                       msg.u.snapshot_channel.pathname,
-                                       msg.u.snapshot_channel.relayd_id, ctx);
-                       if (ret < 0) {
-                               ERR("Snapshot metadata failed");
-                               ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
-                       }
+               struct lttng_consumer_channel *channel;
+               uint64_t key = msg.u.snapshot_channel.key;
+
+               channel = consumer_find_channel(key);
+               if (!channel) {
+                       ERR("Channel %" PRIu64 " not found", key);
+                       ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
                } else {
-                       ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
-                                       msg.u.snapshot_channel.pathname,
-                                       msg.u.snapshot_channel.relayd_id,
-                                       msg.u.snapshot_channel.nb_packets_per_stream,
-                                       ctx);
-                       if (ret < 0) {
-                               ERR("Snapshot channel failed");
-                               ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+                       if (msg.u.snapshot_channel.metadata == 1) {
+                               ret = lttng_kconsumer_snapshot_metadata(channel, key,
+                                               msg.u.snapshot_channel.pathname,
+                                               msg.u.snapshot_channel.relayd_id, ctx);
+                               if (ret < 0) {
+                                       ERR("Snapshot metadata failed");
+                                       ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
+                               }
+                       } else {
+                               ret = lttng_kconsumer_snapshot_channel(channel, key,
+                                               msg.u.snapshot_channel.pathname,
+                                               msg.u.snapshot_channel.relayd_id,
+                                               msg.u.snapshot_channel.nb_packets_per_stream,
+                                               ctx);
+                               if (ret < 0) {
+                                       ERR("Snapshot channel failed");
+                                       ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
+                               }
                        }
                }
-
                health_code_update();
 
                ret = consumer_send_status_msg(sock, ret_code);
@@ -1078,45 +1078,48 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                }
                break;
        }
-       case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE:
+       case LTTNG_CONSUMER_ROTATE_CHANNEL:
        {
-               int channel_rotate_pipe;
-               int flags;
-
-               ret_code = LTTCOMM_CONSUMERD_SUCCESS;
-               /* Successfully received the command's type. */
-               ret = consumer_send_status_msg(sock, ret_code);
-               if (ret < 0) {
-                       goto error_fatal;
-               }
+               struct lttng_consumer_channel *channel;
+               uint64_t key = msg.u.rotate_channel.key;
 
-               ret = lttcomm_recv_fds_unix_sock(sock, &channel_rotate_pipe, 1);
-               if (ret != (ssize_t) sizeof(channel_rotate_pipe)) {
-                       ERR("Failed to receive channel rotate pipe");
-                       goto error_fatal;
-               }
+               DBG("Consumer rotate channel %" PRIu64, key);
 
-               DBG("Received channel rotate pipe (%d)", channel_rotate_pipe);
-               ctx->channel_rotate_pipe = channel_rotate_pipe;
-               /* Set the pipe as non-blocking. */
-               ret = fcntl(channel_rotate_pipe, F_GETFL, 0);
-               if (ret == -1) {
-                       PERROR("fcntl get flags of the channel rotate pipe");
-                       goto error_fatal;
-               }
-               flags = ret;
+               channel = consumer_find_channel(key);
+               if (!channel) {
+                       ERR("Channel %" PRIu64 " not found", key);
+                       ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+               } else {
+                       /*
+                        * Sample the rotate position of all the streams in this channel.
+                        */
+                       ret = lttng_consumer_rotate_channel(channel, key,
+                                       msg.u.rotate_channel.pathname,
+                                       msg.u.rotate_channel.relayd_id,
+                                       msg.u.rotate_channel.metadata,
+                                       msg.u.rotate_channel.new_chunk_id,
+                                       ctx);
+                       if (ret < 0) {
+                               ERR("Rotate channel failed");
+                               ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
+                       }
 
-               ret = fcntl(channel_rotate_pipe, F_SETFL, flags | O_NONBLOCK);
-               if (ret == -1) {
-                       PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
-                       goto error_fatal;
+                       health_code_update();
                }
-               DBG("Channel rotate pipe set as non-blocking");
-               ret_code = LTTCOMM_CONSUMERD_SUCCESS;
                ret = consumer_send_status_msg(sock, ret_code);
                if (ret < 0) {
-                       goto error_fatal;
+                       /* Somehow, the session daemon is not responding anymore. */
+                       goto end_nosignal;
                }
+               if (channel) {
+                       /* Rotate the streams that are ready right now. */
+                       ret = lttng_consumer_rotate_ready_streams(
+                                       channel, key, ctx);
+                       if (ret < 0) {
+                               ERR("Rotate ready streams failed");
+                       }
+               }
+
                break;
        }
        case LTTNG_CONSUMER_ROTATE_RENAME:
@@ -1132,7 +1135,77 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                                msg.u.rotate_rename.relayd_id);
                if (ret < 0) {
                        ERR("Rotate rename failed");
-                       ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+                       ret_code = LTTCOMM_CONSUMERD_ROTATE_RENAME_FAILED;
+               }
+
+               health_code_update();
+
+               ret = consumer_send_status_msg(sock, ret_code);
+               if (ret < 0) {
+                       /* Somehow, the session daemon is not responding anymore. */
+                       goto end_nosignal;
+               }
+               break;
+       }
+       case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_LOCAL:
+       {
+               int pending;
+               uint32_t pending_reply;
+
+               DBG("Perform local check of pending rotation for session id %" PRIu64,
+                               msg.u.check_rotation_pending_local.session_id);
+               pending = lttng_consumer_check_rotation_pending_local(
+                               msg.u.check_rotation_pending_local.session_id,
+                               msg.u.check_rotation_pending_local.chunk_id);
+               if (pending < 0) {
+                       ERR("Local rotation pending check failed with code %i", pending);
+                       ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_LOCAL_FAILED;
+               } else {
+                       pending_reply = !!pending;
+               }
+
+               health_code_update();
+
+               ret = consumer_send_status_msg(sock, ret_code);
+               if (ret < 0) {
+                       /* Somehow, the session daemon is not responding anymore. */
+                       goto end_nosignal;
+               }
+
+               if (pending < 0) {
+                       /*
+                        * An error occured while running the command;
+                        * don't send the 'pending' flag as the sessiond
+                        * will not read it.
+                        */
+                       break;
+               }
+
+               /* Send back returned value to session daemon */
+               ret = lttcomm_send_unix_sock(sock, &pending_reply,
+                               sizeof(pending_reply));
+               if (ret < 0) {
+                       PERROR("Failed to send rotation pending return code");
+                       goto error_fatal;
+               }
+               break;
+       }
+       case LTTNG_CONSUMER_CHECK_ROTATION_PENDING_RELAY:
+       {
+               int pending;
+               uint32_t pending_reply;
+
+               DBG("Perform relayd check of pending rotation for session id %" PRIu64,
+                               msg.u.check_rotation_pending_relay.session_id);
+               pending = lttng_consumer_check_rotation_pending_relay(
+                               msg.u.check_rotation_pending_relay.session_id,
+                               msg.u.check_rotation_pending_relay.relayd_id,
+                               msg.u.check_rotation_pending_relay.chunk_id);
+               if (pending < 0) {
+                       ERR("Relayd rotation pending check failed with code %i", pending);
+                       ret_code = LTTCOMM_CONSUMERD_ROTATION_PENDING_RELAY_FAILED;
+               } else {
+                       pending_reply = !!pending;
                }
 
                health_code_update();
@@ -1142,6 +1215,23 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                        /* Somehow, the session daemon is not responding anymore. */
                        goto end_nosignal;
                }
+
+               if (pending < 0) {
+                       /*
+                        * An error occured while running the command;
+                        * don't send the 'pending' flag as the sessiond
+                        * will not read it.
+                        */
+                       break;
+               }
+
+               /* Send back returned value to session daemon */
+               ret = lttcomm_send_unix_sock(sock, &pending_reply,
+                               sizeof(pending_reply));
+               if (ret < 0) {
+                       PERROR("Failed to send rotation pending return code");
+                       goto error_fatal;
+               }
                break;
        }
        case LTTNG_CONSUMER_MKDIR:
@@ -1155,7 +1245,7 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                                msg.u.mkdir.relayd_id);
                if (ret < 0) {
                        ERR("consumer mkdir failed");
-                       ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+                       ret_code = LTTCOMM_CONSUMERD_MKDIR_FAILED;
                }
 
                health_code_update();
This page took 0.027181 seconds and 4 git commands to generate.