X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Flive.c;h=2a73556dffa0c22c33df136ee53eb39bcc5af362;hp=bafdf35ce90034a90e8ee0e9eaa1a8996103de9e;hb=b0d240a2e2204087ff1634f0bd265660c0582f33;hpb=9237e6a108fdba7acc014f739d0569565552bdec diff --git a/src/bin/lttng-relayd/live.c b/src/bin/lttng-relayd/live.c index bafdf35ce..2a73556df 100644 --- a/src/bin/lttng-relayd/live.c +++ b/src/bin/lttng-relayd/live.c @@ -17,7 +17,6 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE #define _LGPL_SOURCE #include #include @@ -40,7 +39,6 @@ #include #include #include -#include #include #include @@ -199,7 +197,7 @@ end: */ static ssize_t send_viewer_streams(struct lttcomm_sock *sock, - struct relay_session *session, unsigned int ignore_sent_flag) + uint64_t session_id, unsigned int ignore_sent_flag) { ssize_t ret; struct lttng_viewer_stream send_stream; @@ -220,7 +218,7 @@ ssize_t send_viewer_streams(struct lttcomm_sock *sock, pthread_mutex_lock(&vstream->stream->lock); /* Ignore if not the same session. */ - if (vstream->stream->trace->session->id != session->id || + if (vstream->stream->trace->session->id != session_id || (!ignore_sent_flag && vstream->sent_flag)) { pthread_mutex_unlock(&vstream->stream->lock); viewer_stream_put(vstream); @@ -232,10 +230,21 @@ ssize_t send_viewer_streams(struct lttcomm_sock *sock, send_stream.ctf_trace_id = htobe64(ctf_trace->id); send_stream.metadata_flag = htobe32( vstream->stream->is_metadata); - strncpy(send_stream.path_name, vstream->path_name, - sizeof(send_stream.path_name)); - strncpy(send_stream.channel_name, vstream->channel_name, - sizeof(send_stream.channel_name)); + if (lttng_strncpy(send_stream.path_name, vstream->path_name, + sizeof(send_stream.path_name))) { + pthread_mutex_unlock(&vstream->stream->lock); + viewer_stream_put(vstream); + ret = -1; /* Error. */ + goto end_unlock; + } + if (lttng_strncpy(send_stream.channel_name, + vstream->channel_name, + sizeof(send_stream.channel_name))) { + pthread_mutex_unlock(&vstream->stream->lock); + viewer_stream_put(vstream); + ret = -1; /* Error. */ + goto end_unlock; + } DBG("Sending stream %" PRIu64 " to viewer", vstream->stream->stream_handle); @@ -262,24 +271,32 @@ end_unlock: * viewer stream of the session, the number of unsent stream and the number of * stream created. Those counters can be NULL and thus will be ignored. * + * session must be locked to ensure that we see either none or all initial + * streams for a session, but no intermediate state.. + * * Return 0 on success or else a negative value. */ -static -int make_viewer_streams(struct relay_session *session, - enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent, - uint32_t *nb_created, bool *closed) +static int make_viewer_streams(struct relay_session *session, + struct lttng_trace_chunk *viewer_trace_chunk, + enum lttng_viewer_seek seek_t, + uint32_t *nb_total, + uint32_t *nb_unsent, + uint32_t *nb_created, + bool *closed) { int ret; struct lttng_ht_iter iter; struct ctf_trace *ctf_trace; assert(session); + ASSERT_LOCKED(session->lock); - /* - * Hold the session lock to ensure that we see either none or - * all initial streams for a session, but no intermediate state. - */ - pthread_mutex_lock(&session->lock); + if (!viewer_trace_chunk) { + ERR("Internal error: viewer session associated with session \"%s\" has a NULL trace chunk", + session->session_name); + ret = -1; + goto error; + } if (session->connection_closed) { *closed = true; @@ -292,6 +309,7 @@ int make_viewer_streams(struct relay_session *session, rcu_read_lock(); cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace, node.node) { + bool trace_has_metadata_stream = false; struct relay_stream *stream; health_code_update(); @@ -300,6 +318,30 @@ int make_viewer_streams(struct relay_session *session, continue; } + /* + * Iterate over all the streams of the trace to see if we have a + * metadata stream. + */ + cds_list_for_each_entry_rcu( + stream, &ctf_trace->stream_list, stream_node) + { + if (stream->is_metadata) { + trace_has_metadata_stream = true; + break; + } + } + + /* + * If there is no metadata stream in this trace at the moment + * and we never sent one to the viewer, skip the trace. We + * accept that the viewer will not see this trace at all. + */ + if (!trace_has_metadata_stream && + !ctf_trace->metadata_stream_sent_to_viewer) { + ctf_trace_put(ctf_trace); + continue; + } + cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) { struct relay_viewer_stream *vstream; @@ -307,27 +349,24 @@ int make_viewer_streams(struct relay_session *session, continue; } /* - * stream published is protected by the session - * lock. + * stream published is protected by the session lock. */ if (!stream->published) { goto next; } - /* - * Stream has no data, don't consider it yet. - */ - if (stream->is_metadata) { - if (!stream->metadata_received) { - goto next; - } - } else { - if (stream->prev_seq == -1ULL) { - goto next; - } - } vstream = viewer_stream_get_by_id(stream->stream_handle); if (!vstream) { - vstream = viewer_stream_create(stream, seek_t); + /* + * Save that we sent the metadata stream to the + * viewer. So that we know what trace the viewer + * is aware of. + */ + if (stream->is_metadata) { + ctf_trace->metadata_stream_sent_to_viewer = + true; + } + vstream = viewer_stream_create(stream, + viewer_trace_chunk, seek_t); if (!vstream) { ret = -1; ctf_trace_put(ctf_trace); @@ -343,7 +382,10 @@ int make_viewer_streams(struct relay_session *session, * Ensure a self-reference is preserved even * after we have put our local reference. */ - viewer_stream_get(vstream); + if (!viewer_stream_get(vstream)) { + ERR("Unable to get self-reference on viewer stream, logic error."); + abort(); + } } else { if (!vstream->sent_flag && nb_unsent) { /* Update number of unsent stream counter. */ @@ -359,7 +401,7 @@ int make_viewer_streams(struct relay_session *session, } } else { if (!stream->closed || - !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) { + !(((int64_t) (stream->prev_data_seq - stream->last_net_seq_num)) >= 0)) { (*nb_total)++; } @@ -377,7 +419,7 @@ int make_viewer_streams(struct relay_session *session, error_unlock: rcu_read_unlock(); - pthread_mutex_unlock(&session->lock); +error: return ret; } @@ -453,10 +495,11 @@ struct lttcomm_sock *init_socket(struct lttng_uri *uri) if (ret < 0) { goto error; } - DBG("Listening on sock %d for live", sock->fd); + DBG("Listening on sock %d for lttng-live", sock->fd); ret = sock->ops->bind(sock); if (ret < 0) { + PERROR("Failed to bind lttng-live socket"); goto error; } @@ -543,11 +586,6 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -555,10 +593,7 @@ restart: goto exit; } - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("socket poll error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { /* * A new connection is requested, therefore a * viewer connection is allocated in this @@ -601,6 +636,12 @@ restart: * exchange in cds_wfcq_enqueue. */ futex_nto1_wake(&viewer_conn_queue.futex); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("socket poll error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } } @@ -652,12 +693,16 @@ void *thread_dispatcher(void *data) health_code_update(); - while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) { + for (;;) { health_code_update(); /* Atomically prepare the queue futex */ futex_nto1_prepare(&viewer_conn_queue.futex); + if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) { + break; + } + do { health_code_update(); @@ -806,7 +851,7 @@ end: static int viewer_list_sessions(struct relay_connection *conn) { - int ret; + int ret = 0; struct lttng_viewer_list_sessions session_list; struct lttng_ht_iter iter; struct relay_session *session; @@ -828,6 +873,20 @@ int viewer_list_sessions(struct relay_connection *conn) health_code_update(); + pthread_mutex_lock(&session->lock); + if (session->connection_closed) { + /* Skip closed session */ + goto next_session; + } + if (!session->current_trace_chunk) { + /* + * Skip un-attachable session. It is either + * being destroyed or has not had a trace + * chunk created against it yet. + */ + goto next_session; + } + if (count >= buf_count) { struct lttng_viewer_session *newbuf; uint32_t new_buf_count = buf_count << 1; @@ -836,17 +895,23 @@ int viewer_list_sessions(struct relay_connection *conn) new_buf_count * sizeof(*send_session_buf)); if (!newbuf) { ret = -1; - rcu_read_unlock(); - goto end_free; + goto break_loop; } send_session_buf = newbuf; buf_count = new_buf_count; } send_session = &send_session_buf[count]; - strncpy(send_session->session_name, session->session_name, - sizeof(send_session->session_name)); - strncpy(send_session->hostname, session->hostname, - sizeof(send_session->hostname)); + if (lttng_strncpy(send_session->session_name, + session->session_name, + sizeof(send_session->session_name))) { + ret = -1; + goto break_loop; + } + if (lttng_strncpy(send_session->hostname, session->hostname, + sizeof(send_session->hostname))) { + ret = -1; + goto break_loop; + } send_session->id = htobe64(session->id); send_session->live_timer = htobe32(session->live_timer); if (session->viewer_attached) { @@ -856,8 +921,17 @@ int viewer_list_sessions(struct relay_connection *conn) } send_session->streams = htobe32(session->stream_count); count++; + next_session: + pthread_mutex_unlock(&session->lock); + continue; + break_loop: + pthread_mutex_unlock(&session->lock); + break; } rcu_read_unlock(); + if (ret < 0) { + goto end_free; + } session_list.sessions_count = htobe32(count); @@ -893,7 +967,7 @@ int viewer_get_new_streams(struct relay_connection *conn) uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0; struct lttng_viewer_new_streams_request request; struct lttng_viewer_new_streams_response response; - struct relay_session *session; + struct relay_session *session = NULL; uint64_t session_id; bool closed = false; @@ -922,19 +996,21 @@ int viewer_get_new_streams(struct relay_connection *conn) } if (!viewer_session_is_attached(conn->viewer_session, session)) { - send_streams = 0; response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR); goto send_reply; } - send_streams = 1; - response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK); - - ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent, + pthread_mutex_lock(&session->lock); + ret = make_viewer_streams(session, + conn->viewer_session->current_trace_chunk, + LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent, &nb_created, &closed); if (ret < 0) { - goto end_put_session; + goto error_unlock_session; } + send_streams = 1; + response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK); + /* Only send back the newly created streams with the unsent ones. */ nb_streams = nb_created + nb_unsent; response.streams_count = htobe32(nb_streams); @@ -947,8 +1023,10 @@ int viewer_get_new_streams(struct relay_connection *conn) send_streams = 0; response.streams_count = 0; response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP); - goto send_reply; + goto send_reply_unlock; } +send_reply_unlock: + pthread_mutex_unlock(&session->lock); send_reply: health_code_update(); @@ -972,7 +1050,7 @@ send_reply: * streams that were not sent from that point will be sent to * the viewer. */ - ret = send_viewer_streams(conn->sock, session, 0); + ret = send_viewer_streams(conn->sock, session_id, 0); if (ret < 0) { goto end_put_session; } @@ -983,6 +1061,10 @@ end_put_session: } error: return ret; +error_unlock_session: + pthread_mutex_unlock(&session->lock); + session_put(session); + return ret; } /* @@ -998,7 +1080,9 @@ int viewer_attach_session(struct relay_connection *conn) struct lttng_viewer_attach_session_request request; struct lttng_viewer_attach_session_response response; struct relay_session *session = NULL; + enum lttng_viewer_attach_return_code viewer_attach_status; bool closed = false; + uint64_t session_id; assert(conn); @@ -1010,6 +1094,7 @@ int viewer_attach_session(struct relay_connection *conn) goto error; } + session_id = be64toh(request.session_id); health_code_update(); memset(&response, 0, sizeof(response)); @@ -1020,16 +1105,24 @@ int viewer_attach_session(struct relay_connection *conn) goto send_reply; } - session = session_get_by_id(be64toh(request.session_id)); + session = session_get_by_id(session_id); if (!session) { - DBG("Relay session %" PRIu64 " not found", - be64toh(request.session_id)); + DBG("Relay session %" PRIu64 " not found", session_id); response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK); goto send_reply; } - DBG("Attach session ID %" PRIu64 " received", - be64toh(request.session_id)); + DBG("Attach session ID %" PRIu64 " received", session_id); + pthread_mutex_lock(&session->lock); + if (!session->current_trace_chunk) { + /* + * Session is either being destroyed or it never had a trace + * chunk created against it. + */ + DBG("Session requested by live client has no current trace chunk, returning unknown session"); + response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK); + goto send_reply; + } if (session->live_timer == 0) { DBG("Not live session"); response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE); @@ -1037,10 +1130,10 @@ int viewer_attach_session(struct relay_connection *conn) } send_streams = 1; - ret = viewer_session_attach(conn->viewer_session, session); - if (ret) { - DBG("Already a viewer attached"); - response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY); + viewer_attach_status = viewer_session_attach(conn->viewer_session, + session); + if (viewer_attach_status != LTTNG_VIEWER_ATTACH_OK) { + response.status = htobe32(viewer_attach_status); goto send_reply; } @@ -1057,13 +1150,17 @@ int viewer_attach_session(struct relay_connection *conn) goto send_reply; } - ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, - NULL, &closed); + ret = make_viewer_streams(session, + conn->viewer_session->current_trace_chunk, seek_type, + &nb_streams, NULL, NULL, &closed); if (ret < 0) { goto end_put_session; } - response.streams_count = htobe32(nb_streams); + pthread_mutex_unlock(&session->lock); + session_put(session); + session = NULL; + response.streams_count = htobe32(nb_streams); /* * If the session is closed when the viewer is attaching, it * means some of the streams may have been concurrently removed, @@ -1073,7 +1170,7 @@ int viewer_attach_session(struct relay_connection *conn) if (closed) { send_streams = 0; response.streams_count = 0; - response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP); + response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK); goto send_reply; } @@ -1095,13 +1192,14 @@ send_reply: } /* Send stream and ignore the sent flag. */ - ret = send_viewer_streams(conn->sock, session, 1); + ret = send_viewer_streams(conn->sock, session_id, 1); if (ret < 0) { goto end_put_session; } end_put_session: if (session) { + pthread_mutex_unlock(&session->lock); session_put(session); } error: @@ -1111,8 +1209,8 @@ error: /* * Open the index file if needed for the given vstream. * - * If an index file is successfully opened, the vstream index_fd set with - * it. + * If an index file is successfully opened, the vstream will set it as its + * current index file. * * Return 0 on success, a negative value on error (-ENOENT if not ready yet). * @@ -1122,32 +1220,34 @@ static int try_open_index(struct relay_viewer_stream *vstream, struct relay_stream *rstream) { int ret = 0; + const uint32_t connection_major = rstream->trace->session->major; + const uint32_t connection_minor = rstream->trace->session->minor; + enum lttng_trace_chunk_status chunk_status; - if (vstream->index_fd) { + if (vstream->index_file) { goto end; } /* * First time, we open the index file and at least one index is ready. */ - if (rstream->total_index_received == 0) { + if (rstream->index_received_seqcount == 0) { ret = -ENOENT; goto end; } - ret = index_open(vstream->path_name, vstream->channel_name, - vstream->stream->tracefile_count, - vstream->current_tracefile_id); - if (ret >= 0) { - vstream->index_fd = stream_fd_create(ret); - if (!vstream->index_fd) { - if (close(ret)) { - PERROR("close"); - } - ret = -1; + chunk_status = lttng_index_file_create_from_trace_chunk_read_only( + vstream->stream_file.trace_chunk, rstream->path_name, + rstream->channel_name, rstream->tracefile_size, + vstream->current_tracefile_id, + lttng_to_index_major(connection_major, connection_minor), + lttng_to_index_minor(connection_major, connection_minor), + true, &vstream->index_file); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) { + ret = -ENOENT; } else { - ret = 0; + ret = -1; } - goto end; } end: @@ -1171,15 +1271,26 @@ static int check_index_status(struct relay_viewer_stream *vstream, { int ret; - if (trace->session->connection_closed - && rstream->total_index_received - == vstream->last_sent_index) { - /* Last index sent and session connection is closed. */ + DBG("Check index status: index_received_seqcount %" PRIu64 " " + "index_sent_seqcount %" PRIu64 " " + "for stream %" PRIu64, + rstream->index_received_seqcount, + vstream->index_sent_seqcount, + vstream->stream->stream_handle); + if ((trace->session->connection_closed || rstream->closed) + && rstream->index_received_seqcount + == vstream->index_sent_seqcount) { + /* + * Last index sent and session connection or relay + * stream are closed. + */ index->status = htobe32(LTTNG_VIEWER_INDEX_HUP); goto hup; } else if (rstream->beacon_ts_end != -1ULL && - rstream->total_index_received - == vstream->last_sent_index) { + (rstream->index_received_seqcount == 0 || + (vstream->index_sent_seqcount != 0 && + rstream->index_received_seqcount + <= vstream->index_sent_seqcount))) { /* * We've received a synchronization beacon and the last index * available has been sent, the index for now is inactive. @@ -1188,80 +1299,87 @@ static int check_index_status(struct relay_viewer_stream *vstream, * inform the client of a time interval during which we can * guarantee that there are no events to read (and never will * be). + * + * The sent seqcount can grow higher than receive seqcount on + * clear because the rotation performed by clear will push + * the index_sent_seqcount ahead (see + * viewer_stream_sync_tracefile_array_tail) and skip over + * packet sequence numbers. */ index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE); index->timestamp_end = htobe64(rstream->beacon_ts_end); index->stream_id = htobe64(rstream->ctf_stream_id); + DBG("Check index status: inactive with beacon, for stream %" PRIu64, + vstream->stream->stream_handle); goto index_ready; - } else if (rstream->total_index_received <= vstream->last_sent_index) { + } else if (rstream->index_received_seqcount == 0 || + (vstream->index_sent_seqcount != 0 && + rstream->index_received_seqcount + <= vstream->index_sent_seqcount)) { /* - * This actually checks the case where recv == last_sent. - * In this case, we have not received a beacon. Therefore, we - * can only ask the client to retry later. + * This checks whether received <= sent seqcount. In + * this case, we have not received a beacon. Therefore, + * we can only ask the client to retry later. + * + * The sent seqcount can grow higher than receive seqcount on + * clear because the rotation performed by clear will push + * the index_sent_seqcount ahead (see + * viewer_stream_sync_tracefile_array_tail) and skip over + * packet sequence numbers. */ index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + DBG("Check index status: retry for stream %" PRIu64, + vstream->stream->stream_handle); goto index_ready; - } else if (!viewer_stream_is_tracefile_seq_readable(vstream, - vstream->current_tracefile_seq)) { + } else if (!tracefile_array_seq_in_file(rstream->tfa, + vstream->current_tracefile_id, + vstream->index_sent_seqcount)) { /* - * The producer has overwritten our current file. We - * need to rotate. + * The next index we want to send cannot be read either + * because we need to perform a rotation, or due to + * the producer having overwritten its trace file. */ - DBG("Viewer stream %" PRIu64 " rotation due to overwrite", + DBG("Viewer stream %" PRIu64 " rotation", vstream->stream->stream_handle); ret = viewer_stream_rotate(vstream); - if (ret < 0) { - goto end; - } else if (ret == 1) { + if (ret == 1) { /* EOF across entire stream. */ index->status = htobe32(LTTNG_VIEWER_INDEX_HUP); goto hup; } - assert(viewer_stream_is_tracefile_seq_readable(vstream, - vstream->current_tracefile_seq)); - /* ret == 0 means successful so we continue. */ - ret = 0; - } else { - ssize_t read_ret; - char tmp[1]; - /* - * Use EOF on current index file to find out when we - * need to rotate. + * If we have been pushed due to overwrite, it + * necessarily means there is data that can be read in + * the stream. If we rotated because we reached the end + * of a tracefile, it means the following tracefile + * needs to contain at least one index, else we would + * have already returned LTTNG_VIEWER_INDEX_RETRY to the + * viewer. The updated index_sent_seqcount needs to + * point to a readable index entry now. + * + * In the case where we "rotate" on a single file, we + * can end up in a case where the requested index is + * still unavailable. */ - read_ret = lttng_read(vstream->index_fd->fd, tmp, 1); - if (read_ret == 1) { - off_t seek_ret; - - /* There is still data to read. Rewind position. */ - seek_ret = lseek(vstream->index_fd->fd, -1, SEEK_CUR); - if (seek_ret < 0) { - ret = -1; - goto end; - } - ret = 0; - } else if (read_ret == 0) { - /* EOF. We need to rotate. */ - DBG("Viewer stream %" PRIu64 " rotation due to EOF", - vstream->stream->stream_handle); - ret = viewer_stream_rotate(vstream); - if (ret < 0) { - goto end; - } else if (ret == 1) { - /* EOF across entire stream. */ - index->status = htobe32(LTTNG_VIEWER_INDEX_HUP); - goto hup; - } - assert(viewer_stream_is_tracefile_seq_readable(vstream, - vstream->current_tracefile_seq)); - /* ret == 0 means successful so we continue. */ - ret = 0; - } else { - /* Error reading index. */ - ret = -1; + if (rstream->tracefile_count == 1 && + !tracefile_array_seq_in_file( + rstream->tfa, + vstream->current_tracefile_id, + vstream->index_sent_seqcount)) { + index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + DBG("Check index status: retry: " + "tracefile array sequence number %" PRIu64 + " not in file for stream %" PRIu64, + vstream->index_sent_seqcount, + vstream->stream->stream_handle); + goto index_ready; } + assert(tracefile_array_seq_in_file(rstream->tfa, + vstream->current_tracefile_id, + vstream->index_sent_seqcount)); } -end: + /* ret == 0 means successful so we continue. */ + ret = 0; return ret; hup: @@ -1279,7 +1397,6 @@ static int viewer_get_next_index(struct relay_connection *conn) { int ret; - ssize_t read_ret; struct lttng_viewer_get_next_index request_index; struct lttng_viewer_index viewer_index; struct ctf_packet_index packet_index; @@ -1303,6 +1420,8 @@ int viewer_get_next_index(struct relay_connection *conn) vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id)); if (!vstream) { + DBG("Client requested index of unknown stream id %" PRIu64, + (uint64_t) be64toh(request_index.stream_id)); viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); goto send_reply; } @@ -1325,21 +1444,70 @@ int viewer_get_next_index(struct relay_connection *conn) goto send_reply; } - /* Try to open an index if one is needed for that stream. */ - ret = try_open_index(vstream, rstream); - if (ret < 0) { - if (ret == -ENOENT) { - /* - * The index is created only when the first data - * packet arrives, it might not be ready at the - * beginning of the session - */ - viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); - } else { - /* Unhandled error. */ + if (rstream->ongoing_rotation.is_set) { + /* Rotation is ongoing, try again later. */ + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto send_reply; + } + + if (rstream->trace->session->ongoing_rotation) { + /* Rotation is ongoing, try again later. */ + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto send_reply; + } + + if (rstream->trace_chunk) { + uint64_t rchunk_id, vchunk_id; + + /* + * If the relay stream is not yet closed, ensure the viewer + * chunk matches the relay chunk after clear. + */ + if (lttng_trace_chunk_get_id(rstream->trace_chunk, + &rchunk_id) != LTTNG_TRACE_CHUNK_STATUS_OK) { viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; } - goto send_reply; + if (lttng_trace_chunk_get_id( + conn->viewer_session->current_trace_chunk, + &vchunk_id) != LTTNG_TRACE_CHUNK_STATUS_OK) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; + } + + if (rchunk_id != vchunk_id) { + DBG("Relay and viewer chunk ids differ: " + "rchunk_id %" PRIu64 " vchunk_id %" PRIu64, + rchunk_id, vchunk_id); + + lttng_trace_chunk_put( + conn->viewer_session->current_trace_chunk); + conn->viewer_session->current_trace_chunk = NULL; + ret = viewer_session_set_trace_chunk_copy( + conn->viewer_session, + rstream->trace_chunk); + if (ret) { + viewer_index.status = + htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; + } + } + } + if (conn->viewer_session->current_trace_chunk != + vstream->stream_file.trace_chunk) { + bool acquired_reference; + + DBG("Viewer session and viewer stream chunk differ: " + "vsession chunk %p vstream chunk %p", + conn->viewer_session->current_trace_chunk, + vstream->stream_file.trace_chunk); + lttng_trace_chunk_put(vstream->stream_file.trace_chunk); + acquired_reference = lttng_trace_chunk_get(conn->viewer_session->current_trace_chunk); + assert(acquired_reference); + vstream->stream_file.trace_chunk = + conn->viewer_session->current_trace_chunk; + viewer_stream_sync_tracefile_array_tail(vstream); + viewer_stream_close_files(vstream); } ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index); @@ -1355,6 +1523,22 @@ int viewer_get_next_index(struct relay_connection *conn) /* At this point, ret is 0 thus we will be able to read the index. */ assert(!ret); + /* Try to open an index if one is needed for that stream. */ + ret = try_open_index(vstream, rstream); + if (ret == -ENOENT) { + if (rstream->closed) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP); + goto send_reply; + } else { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY); + goto send_reply; + } + } + if (ret < 0) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); + goto send_reply; + } + /* * vstream->stream_fd may be NULL if it has been closed by * tracefile rotation, or if we are at the beginning of the @@ -1362,31 +1546,40 @@ int viewer_get_next_index(struct relay_connection *conn) * overwrite caused by tracefile rotation (in association with * unlink performed before overwrite). */ - if (!vstream->stream_fd) { - char fullpath[PATH_MAX]; - - if (vstream->stream->tracefile_count > 0) { - ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, - vstream->path_name, - vstream->channel_name, - vstream->current_tracefile_id); - } else { - ret = snprintf(fullpath, PATH_MAX, "%s/%s", - vstream->path_name, - vstream->channel_name); - } + if (!vstream->stream_file.fd) { + int fd; + char file_path[LTTNG_PATH_MAX]; + enum lttng_trace_chunk_status status; + + ret = utils_stream_file_path(rstream->path_name, + rstream->channel_name, rstream->tracefile_size, + vstream->current_tracefile_id, NULL, file_path, + sizeof(file_path)); if (ret < 0) { goto error_put; } - ret = open(fullpath, O_RDONLY); - if (ret < 0) { - PERROR("Relay opening trace file"); + + /* + * It is possible the the file we are trying to open is + * missing if the stream has been closed (application exits with + * per-pid buffers) and a clear command has been performed. + */ + status = lttng_trace_chunk_open_file( + vstream->stream_file.trace_chunk, + file_path, O_RDONLY, 0, &fd, true); + if (status != LTTNG_TRACE_CHUNK_STATUS_OK) { + if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE && + rstream->closed) { + viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP); + goto send_reply; + } + PERROR("Failed to open trace file for viewer stream"); goto error_put; } - vstream->stream_fd = stream_fd_create(ret); - if (!vstream->stream_fd) { - if (close(ret)) { - PERROR("close"); + vstream->stream_file.fd = stream_fd_create(fd); + if (!vstream->stream_file.fd) { + if (close(fd)) { + PERROR("Failed to close viewer stream file"); } goto error_put; } @@ -1400,16 +1593,15 @@ int viewer_get_next_index(struct relay_connection *conn) viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM; } - read_ret = lttng_read(vstream->index_fd->fd, &packet_index, - sizeof(packet_index)); - if (read_ret < sizeof(packet_index)) { - ERR("Relay reading index file %d returned %zd", - vstream->index_fd->fd, read_ret); + ret = lttng_index_file_read(vstream->index_file, &packet_index); + if (ret) { + ERR("Relay error reading index file %d", + vstream->index_file->fd); viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR); goto send_reply; } else { viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK); - vstream->last_sent_index++; + vstream->index_sent_seqcount++; } /* @@ -1417,7 +1609,7 @@ int viewer_get_next_index(struct relay_connection *conn) */ DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64, rstream->stream_handle, - be64toh(packet_index.offset)); + (uint64_t) be64toh(packet_index.offset)); viewer_index.offset = packet_index.offset; viewer_index.packet_size = packet_index.packet_size; viewer_index.content_size = packet_index.content_size; @@ -1456,7 +1648,7 @@ send_reply: if (vstream) { DBG("Index %" PRIu64 " for stream %" PRIu64 " sent", - vstream->last_sent_index, + vstream->index_sent_seqcount, vstream->stream->stream_handle); } end: @@ -1485,13 +1677,15 @@ error_put: static int viewer_get_packet(struct relay_connection *conn) { - int ret, send_data = 0; - char *data = NULL; - uint32_t len = 0; - ssize_t read_len; + int ret; + off_t lseek_ret; + char *reply = NULL; struct lttng_viewer_get_packet get_packet_info; - struct lttng_viewer_trace_packet reply; + struct lttng_viewer_trace_packet reply_header; struct relay_viewer_stream *vstream = NULL; + uint32_t reply_size = sizeof(reply_header); + uint32_t packet_data_len = 0; + ssize_t read_len; DBG2("Relay get data packet"); @@ -1505,74 +1699,78 @@ int viewer_get_packet(struct relay_connection *conn) health_code_update(); /* From this point on, the error label can be reached. */ - memset(&reply, 0, sizeof(reply)); + memset(&reply_header, 0, sizeof(reply_header)); vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id)); if (!vstream) { - reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR); + DBG("Client requested packet of unknown stream id %" PRIu64, + (uint64_t) be64toh(get_packet_info.stream_id)); + reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR); goto send_reply_nolock; + } else { + packet_data_len = be32toh(get_packet_info.len); + reply_size += packet_data_len; } - pthread_mutex_lock(&vstream->stream->lock); - - len = be32toh(get_packet_info.len); - data = zmalloc(len); - if (!data) { - PERROR("relay data zmalloc"); + reply = zmalloc(reply_size); + if (!reply) { + PERROR("packet reply zmalloc"); + reply_size = sizeof(reply_header); goto error; } - ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset), - SEEK_SET); - if (ret < 0) { - PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd, - be64toh(get_packet_info.offset)); + pthread_mutex_lock(&vstream->stream->lock); + lseek_ret = lseek(vstream->stream_file.fd->fd, + be64toh(get_packet_info.offset), SEEK_SET); + if (lseek_ret < 0) { + PERROR("lseek fd %d to offset %" PRIu64, + vstream->stream_file.fd->fd, + (uint64_t) be64toh(get_packet_info.offset)); goto error; } - read_len = lttng_read(vstream->stream_fd->fd, data, len); - if (read_len < len) { + read_len = lttng_read(vstream->stream_file.fd->fd, + reply + sizeof(reply_header), packet_data_len); + if (read_len < packet_data_len) { PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64, - vstream->stream_fd->fd, - be64toh(get_packet_info.offset)); + vstream->stream_file.fd->fd, + (uint64_t) be64toh(get_packet_info.offset)); goto error; } - reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK); - reply.len = htobe32(len); - send_data = 1; + reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK); + reply_header.len = htobe32(packet_data_len); goto send_reply; error: - reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR); + reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR); send_reply: if (vstream) { pthread_mutex_unlock(&vstream->stream->lock); } send_reply_nolock: - reply.flags = htobe32(reply.flags); health_code_update(); - ret = send_response(conn->sock, &reply, sizeof(reply)); - if (ret < 0) { - goto end_free; + if (reply) { + memcpy(reply, &reply_header, sizeof(reply_header)); + ret = send_response(conn->sock, reply, reply_size); + } else { + /* No reply to send. */ + ret = send_response(conn->sock, &reply_header, + reply_size); } - health_code_update(); - if (send_data) { - health_code_update(); - ret = send_response(conn->sock, data, len); - if (ret < 0) { - goto end_free; - } - health_code_update(); + health_code_update(); + if (ret < 0) { + PERROR("sendmsg of packet data failed"); + goto end_free; } - DBG("Sent %u bytes for stream %" PRIu64, len, - be64toh(get_packet_info.stream_id)); + DBG("Sent %u bytes for stream %" PRIu64, reply_size, + (uint64_t) be64toh(get_packet_info.stream_id)); end_free: - free(data); + free(reply); end: if (vstream) { viewer_stream_put(vstream); @@ -1620,6 +1818,8 @@ int viewer_get_metadata(struct relay_connection *conn) * Reply back to the client with an error if we cannot * find it. */ + DBG("Client requested metadata of unknown stream id %" PRIu64, + (uint64_t) be64toh(request.stream_id)); reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR); goto send_reply; } @@ -1629,32 +1829,71 @@ int viewer_get_metadata(struct relay_connection *conn) goto error; } - assert(vstream->metadata_sent <= vstream->stream->metadata_received); - - len = vstream->stream->metadata_received - vstream->metadata_sent; - if (len == 0) { + if (vstream->metadata_sent >= vstream->stream->metadata_received) { + /* + * The live viewers expect to receive a NO_NEW_METADATA + * status before a stream disappears, otherwise they abort the + * entire live connection when receiving an error status. + * + * Clear feature resets the metadata_sent to 0 until the + * same metadata is received again. + */ reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA); + /* + * The live viewer considers a closed 0 byte metadata stream as + * an error. + */ + if (vstream->metadata_sent > 0) { + vstream->stream->no_new_metadata_notified = true; + if (vstream->stream->closed) { + /* Release ownership for the viewer metadata stream. */ + viewer_stream_put(vstream); + } + } goto send_reply; } - /* first time, we open the metadata file */ - if (!vstream->stream_fd) { - char fullpath[PATH_MAX]; + len = vstream->stream->metadata_received - vstream->metadata_sent; - ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name, - vstream->channel_name); + /* first time, we open the metadata file */ + if (!vstream->stream_file.fd) { + int fd; + char file_path[LTTNG_PATH_MAX]; + enum lttng_trace_chunk_status status; + struct relay_stream *rstream = vstream->stream; + + ret = utils_stream_file_path(rstream->path_name, + rstream->channel_name, rstream->tracefile_size, + vstream->current_tracefile_id, NULL, file_path, + sizeof(file_path)); if (ret < 0) { goto error; } - ret = open(fullpath, O_RDONLY); - if (ret < 0) { - PERROR("Relay opening metadata file"); + + /* + * It is possible the the metadata file we are trying to open is + * missing if the stream has been closed (application exits with + * per-pid buffers) and a clear command has been performed. + */ + status = lttng_trace_chunk_open_file( + vstream->stream_file.trace_chunk, + file_path, O_RDONLY, 0, &fd, true); + if (status != LTTNG_TRACE_CHUNK_STATUS_OK) { + if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) { + reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA); + len = 0; + if (vstream->stream->closed) { + viewer_stream_put(vstream); + } + goto send_reply; + } + PERROR("Failed to open metadata file for viewer stream"); goto error; } - vstream->stream_fd = stream_fd_create(ret); - if (!vstream->stream_fd) { - if (close(ret)) { - PERROR("close"); + vstream->stream_file.fd = stream_fd_create(fd); + if (!vstream->stream_file.fd) { + if (close(fd)) { + PERROR("Failed to close viewer metadata file"); } goto error; } @@ -1667,18 +1906,12 @@ int viewer_get_metadata(struct relay_connection *conn) goto error; } - read_len = lttng_read(vstream->stream_fd->fd, data, len); + read_len = lttng_read(vstream->stream_file.fd->fd, data, len); if (read_len < len) { PERROR("Relay reading metadata file"); goto error; } vstream->metadata_sent += read_len; - if (vstream->metadata_sent == vstream->stream->metadata_received - && vstream->stream->closed) { - /* Release ownership for the viewer metadata stream. */ - viewer_stream_put(vstream); - } - reply.status = htobe32(LTTNG_VIEWER_METADATA_OK); goto send_reply; @@ -1705,7 +1938,7 @@ send_reply: } DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len, - be64toh(request.stream_id)); + (uint64_t) be64toh(request.stream_id)); DBG("Metadata sent"); @@ -1753,6 +1986,78 @@ end: return ret; } +/* + * Detach a viewer session. + * + * Return 0 on success or else a negative value. + */ +static +int viewer_detach_session(struct relay_connection *conn) +{ + int ret; + struct lttng_viewer_detach_session_response response; + struct lttng_viewer_detach_session_request request; + struct relay_session *session = NULL; + uint64_t viewer_session_to_close; + + DBG("Viewer detach session received"); + + assert(conn); + + health_code_update(); + + /* Receive the request from the connected client. */ + ret = recv_request(conn->sock, &request, sizeof(request)); + if (ret < 0) { + goto end; + } + viewer_session_to_close = be64toh(request.session_id); + + if (!conn->viewer_session) { + DBG("Client trying to detach before creating a live viewer session"); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR); + goto send_reply; + } + + health_code_update(); + + memset(&response, 0, sizeof(response)); + DBG("Detaching from session ID %" PRIu64, viewer_session_to_close); + + session = session_get_by_id(be64toh(request.session_id)); + if (!session) { + DBG("Relay session %" PRIu64 " not found", + (uint64_t) be64toh(request.session_id)); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK); + goto send_reply; + } + + ret = viewer_session_is_attached(conn->viewer_session, session); + if (ret != 1) { + DBG("Not attached to this session"); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR); + goto send_reply_put; + } + + viewer_session_close_one_session(conn->viewer_session, session); + response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK); + DBG("Session %" PRIu64 " detached.", viewer_session_to_close); + +send_reply_put: + session_put(session); + +send_reply: + health_code_update(); + ret = send_response(conn->sock, &response, sizeof(response)); + if (ret < 0) { + goto end; + } + health_code_update(); + ret = 0; + +end: + return ret; +} /* * live_relay_unknown_command: send -1 if received unknown command @@ -1814,6 +2119,9 @@ int process_control(struct lttng_viewer_cmd *recv_hdr, case LTTNG_VIEWER_CREATE_SESSION: ret = viewer_create_session(conn); break; + case LTTNG_VIEWER_DETACH_SESSION: + ret = viewer_detach_session(conn); + break; default: ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd)); @@ -1914,11 +2222,6 @@ restart: health_code_update(); - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1928,10 +2231,7 @@ restart: /* Inspect the relay conn pipe for new connection. */ if (pollfd == live_conn_pipe[0]) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Relay live pipe error"); - goto error; - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { struct relay_connection *conn; ret = lttng_read(live_conn_pipe[0], @@ -1939,10 +2239,21 @@ restart: if (ret < 0) { goto error; } - lttng_poll_add(&events, conn->sock->fd, + ret = lttng_poll_add(&events, + conn->sock->fd, LPOLLIN | LPOLLRDHUP); + if (ret) { + ERR("Failed to add new live connection file descriptor to poll set"); + goto error; + } connection_ht_add(viewer_connections_ht, conn); DBG("Connection socket %d added to poll", conn->sock->fd); + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Relay live pipe error"); + goto error; + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + goto error; } } else { /* Connection activity. */ @@ -1953,11 +2264,7 @@ restart: continue; } - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - cleanup_connection_pollfd(&events, pollfd); - /* Put "create" ownership reference. */ - connection_put(conn); - } else if (revents & LPOLLIN) { + if (revents & LPOLLIN) { ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr, sizeof(recv_hdr), 0); if (ret <= 0) { @@ -1976,6 +2283,14 @@ restart: DBG("Viewer connection closed with %d", pollfd); } } + } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + cleanup_connection_pollfd(&events, pollfd); + /* Put "create" ownership reference. */ + connection_put(conn); + } else { + ERR("Unexpected poll events %u for sock %d", revents, pollfd); + connection_put(conn); + goto error; } /* Put local "get_by_sock" reference. */ connection_put(conn); @@ -1987,7 +2302,7 @@ exit: error: lttng_poll_clean(&events); - /* Cleanup reamaining connection object. */ + /* Cleanup remaining connection object. */ rcu_read_lock(); cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter, destroy_conn, @@ -2100,7 +2415,7 @@ int relayd_live_create(struct lttng_uri *uri) } /* Setup the dispatcher thread */ - ret = pthread_create(&live_dispatcher_thread, NULL, + ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(), thread_dispatcher, (void *) NULL); if (ret) { errno = ret; @@ -2110,7 +2425,7 @@ int relayd_live_create(struct lttng_uri *uri) } /* Setup the worker thread */ - ret = pthread_create(&live_worker_thread, NULL, + ret = pthread_create(&live_worker_thread, default_pthread_attr(), thread_worker, NULL); if (ret) { errno = ret; @@ -2120,7 +2435,7 @@ int relayd_live_create(struct lttng_uri *uri) } /* Setup the listener thread */ - ret = pthread_create(&live_listener_thread, NULL, + ret = pthread_create(&live_listener_thread, default_pthread_attr(), thread_listener, (void *) NULL); if (ret) { errno = ret;