Fix: consumer: snapshot error return code
[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.c
index a5dcc663d68c67f4b465ecabe70004f682842f3f..b890416cac1d96db4fe481c27f9f395e4a14707d 100644 (file)
@@ -60,7 +60,11 @@ int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
        int infd = stream->wait_fd;
 
        ret = kernctl_snapshot(infd);
-       if (ret != 0) {
+       /*
+        * -EAGAIN is not an error, it just means that there is no data to
+        *  be read.
+        */
+       if (ret != 0 && ret != -EAGAIN) {
                PERROR("Getting sub-buffer snapshot.");
        }
 
@@ -120,28 +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);
@@ -187,14 +185,6 @@ int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
                        DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
                                        path, stream->name, stream->key);
                }
-               if (relayd_id != -1ULL) {
-                       ret = consumer_send_relayd_streams_sent(relayd_id);
-                       if (ret < 0) {
-                               ERR("sending streams sent to relayd");
-                               goto end_unlock;
-                       }
-                       channel->streams_sent_to_relayd = true;
-               }
 
                ret = kernctl_buffer_flush_empty(stream->wait_fd);
                if (ret < 0) {
@@ -337,15 +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);
@@ -355,15 +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) {
@@ -373,7 +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,
@@ -381,7 +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;
        }
@@ -389,12 +375,13 @@ int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
        do {
                health_code_update();
 
-               ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
+               ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx, NULL);
                if (ret_read < 0) {
                        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;
@@ -419,11 +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;
 }
@@ -648,7 +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:
@@ -662,6 +650,8 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 
                new_stream->chan = channel;
                new_stream->wait_fd = fd;
+               consumer_stream_update_channel_attributes(new_stream,
+                               channel);
                switch (channel->output) {
                case CONSUMER_CHANNEL_SPLICE:
                        new_stream->output = LTTNG_EVENT_SPLICE;
@@ -748,26 +738,14 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 
                /* Get the right pipe where the stream will be sent. */
                if (new_stream->metadata_flag) {
-                       ret = consumer_add_metadata_stream(new_stream);
-                       if (ret) {
-                               ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
-                                               new_stream->key);
-                               consumer_stream_free(new_stream);
-                               goto end_nosignal;
-                       }
+                       consumer_add_metadata_stream(new_stream);
                        stream_pipe = ctx->consumer_metadata_pipe;
                } else {
-                       ret = consumer_add_data_stream(new_stream);
-                       if (ret) {
-                               ERR("Consumer add stream %" PRIu64 " failed. Continuing",
-                                               new_stream->key);
-                               consumer_stream_free(new_stream);
-                               goto end_nosignal;
-                       }
+                       consumer_add_data_stream(new_stream);
                        stream_pipe = ctx->consumer_data_pipe;
                }
 
-               /* Vitible to other threads */
+               /* Visible to other threads */
                new_stream->globally_visible = 1;
 
                health_code_update();
@@ -785,8 +763,8 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                        goto end_nosignal;
                }
 
-               DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
-                               new_stream->name, fd, new_stream->relayd_stream_id);
+               DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
+                               new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id);
                break;
        }
        case LTTNG_CONSUMER_STREAMS_SENT:
@@ -910,26 +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);
@@ -1092,6 +1078,185 @@ int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
                }
                break;
        }
+       case LTTNG_CONSUMER_ROTATE_CHANNEL:
+       {
+               struct lttng_consumer_channel *channel;
+               uint64_t key = msg.u.rotate_channel.key;
+
+               DBG("Consumer rotate channel %" PRIu64, key);
+
+               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;
+                       }
+
+                       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 (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:
+       {
+               DBG("Consumer rename session %" PRIu64 " after rotation, old path = \"%s\", new path = \"%s\"",
+                               msg.u.rotate_rename.session_id,
+                               msg.u.rotate_rename.old_path,
+                               msg.u.rotate_rename.new_path);
+               ret = lttng_consumer_rotate_rename(msg.u.rotate_rename.old_path,
+                               msg.u.rotate_rename.new_path,
+                               msg.u.rotate_rename.uid,
+                               msg.u.rotate_rename.gid,
+                               msg.u.rotate_rename.relayd_id);
+               if (ret < 0) {
+                       ERR("Rotate rename failed");
+                       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();
+
+               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_MKDIR:
+       {
+               DBG("Consumer mkdir %s in session %" PRIu64,
+                               msg.u.mkdir.path,
+                               msg.u.mkdir.session_id);
+               ret = lttng_consumer_mkdir(msg.u.mkdir.path,
+                               msg.u.mkdir.uid,
+                               msg.u.mkdir.gid,
+                               msg.u.mkdir.relayd_id);
+               if (ret < 0) {
+                       ERR("consumer mkdir failed");
+                       ret_code = LTTCOMM_CONSUMERD_MKDIR_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;
+       }
        default:
                goto end_nosignal;
        }
@@ -1331,16 +1496,30 @@ end:
  * Consume data on a file descriptor and write it on a trace file.
  */
 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
-               struct lttng_consumer_local_data *ctx)
+               struct lttng_consumer_local_data *ctx, bool *rotated)
 {
        unsigned long len, subbuf_size, padding;
-       int err, write_index = 1;
+       int err, write_index = 1, rotation_ret;
        ssize_t ret = 0;
        int infd = stream->wait_fd;
        struct ctf_packet_index index;
 
        DBG("In read_subbuffer (infd : %d)", infd);
 
+       /*
+        * If the stream was flagged to be ready for rotation before we extract the
+        * next packet, rotate it now.
+        */
+       if (stream->rotate_ready) {
+               DBG("Rotate stream before extracting data");
+               rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
+               if (rotation_ret < 0) {
+                       ERR("Stream rotation error");
+                       ret = -1;
+                       goto error;
+               }
+       }
+
        /* Get the next subbuffer */
        err = kernctl_get_next_subbuf(infd);
        if (err != 0) {
@@ -1353,7 +1532,7 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                DBG("Reserving sub buffer failed (everything is normal, "
                                "it is due to concurrency)");
                ret = err;
-               goto end;
+               goto error;
        }
 
        /* Get the full subbuffer size including padding */
@@ -1369,10 +1548,10 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                                PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
                        }
                        ret = err;
-                       goto end;
+                       goto error;
                }
                ret = err;
-               goto end;
+               goto error;
        }
 
        if (!stream->metadata_flag) {
@@ -1387,9 +1566,9 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                                        PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
                                }
                                ret = err;
-                               goto end;
+                               goto error;
                        }
-                       goto end;
+                       goto error;
                }
                ret = update_stream_stats(stream);
                if (ret < 0) {
@@ -1402,9 +1581,9 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                                        PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
                                }
                                ret = err;
-                               goto end;
+                               goto error;
                        }
-                       goto end;
+                       goto error;
                }
        } else {
                write_index = 0;
@@ -1419,9 +1598,9 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                                        PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
                                }
                                ret = err;
-                               goto end;
+                               goto error;
                        }
-                       goto end;
+                       goto error;
                }
        }
 
@@ -1466,10 +1645,10 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                                        PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
                                }
                                ret = err;
-                               goto end;
+                               goto error;
                        }
                        ret = err;
-                       goto end;
+                       goto error;
                }
 
                /* Make sure the tracer is not gone mad on us! */
@@ -1512,12 +1691,12 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                        PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
                }
                ret = err;
-               goto end;
+               goto error;
        }
 
        /* Write index if needed. */
        if (!write_index) {
-               goto end;
+               goto rotate;
        }
 
        if (stream->chan->live_timer_interval && !stream->metadata_flag) {
@@ -1541,16 +1720,35 @@ ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
                        pthread_mutex_unlock(&stream->metadata_timer_lock);
                }
                if (err < 0) {
-                       goto end;
+                       goto error;
                }
        }
 
        err = consumer_stream_write_index(stream, &index);
        if (err < 0) {
-               goto end;
+               goto error;
        }
 
-end:
+rotate:
+       /*
+        * After extracting the packet, we check if the stream is now ready to be
+        * rotated and perform the action immediately.
+        */
+       rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
+       if (rotation_ret == 1) {
+               rotation_ret = lttng_consumer_rotate_stream(ctx, stream, rotated);
+               if (rotation_ret < 0) {
+                       ERR("Stream rotation error");
+                       ret = -1;
+                       goto error;
+               }
+       } else if (rotation_ret < 0) {
+               ERR("Checking if stream is ready to rotate");
+               ret = -1;
+               goto error;
+       }
+
+error:
        return ret;
 }
 
This page took 0.031189 seconds and 4 git commands to generate.