src/bin/lttng-relayd/live.c

   1 /*
   2  * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
   3  *                      David Goulet <dgoulet@efficios.com>
   4  *               2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License, version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19
  20 #define _LGPL_SOURCE
  21 #include <getopt.h>
  22 #include <grp.h>
  23 #include <limits.h>
  24 #include <pthread.h>
  25 #include <signal.h>
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include <sys/mman.h>
  30 #include <sys/mount.h>
  31 #include <sys/resource.h>
  32 #include <sys/socket.h>
  33 #include <sys/stat.h>
  34 #include <sys/types.h>
  35 #include <sys/wait.h>
  36 #include <inttypes.h>
  37 #include <urcu/futex.h>
  38 #include <urcu/uatomic.h>
  39 #include <urcu/rculist.h>
  40 #include <unistd.h>
  41 #include <fcntl.h>
  42
  43 #include <lttng/lttng.h>
  44 #include <common/common.h>
  45 #include <common/compat/poll.h>
  46 #include <common/compat/socket.h>
  47 #include <common/compat/endian.h>
  48 #include <common/defaults.h>
  49 #include <common/futex.h>
  50 #include <common/index/index.h>
  51 #include <common/sessiond-comm/sessiond-comm.h>
  52 #include <common/sessiond-comm/inet.h>
  53 #include <common/sessiond-comm/relayd.h>
  54 #include <common/uri.h>
  55 #include <common/utils.h>
  56
  57 #include "cmd.h"
  58 #include "live.h"
  59 #include "lttng-relayd.h"
  60 #include "utils.h"
  61 #include "health-relayd.h"
  62 #include "testpoint.h"
  63 #include "viewer-stream.h"
  64 #include "stream.h"
  65 #include "session.h"
  66 #include "ctf-trace.h"
  67 #include "connection.h"
  68 #include "viewer-session.h"
  69
  70 #define SESSION_BUF_DEFAULT_COUNT       16
  71
  72 static struct lttng_uri *live_uri;
  73
  74 /*
  75  * This pipe is used to inform the worker thread that a command is queued and
  76  * ready to be processed.
  77  */
  78 static int live_conn_pipe[2] = { -1, -1 };
  79
  80 /* Shared between threads */
  81 static int live_dispatch_thread_exit;
  82
  83 static pthread_t live_listener_thread;
  84 static pthread_t live_dispatcher_thread;
  85 static pthread_t live_worker_thread;
  86
  87 /*
  88  * Relay command queue.
  89  *
  90  * The live_thread_listener and live_thread_dispatcher communicate with this
  91  * queue.
  92  */
  93 static struct relay_conn_queue viewer_conn_queue;
  94
  95 static uint64_t last_relay_viewer_session_id;
  96 static pthread_mutex_t last_relay_viewer_session_id_lock =
  97                 PTHREAD_MUTEX_INITIALIZER;
  98
  99 /*
 100  * Cleanup the daemon
 101  */
 102 static
 103 void cleanup_relayd_live(void)
 104 {
 105         DBG("Cleaning up");
 106
 107         free(live_uri);
 108 }
 109
 110 /*
 111  * Receive a request buffer using a given socket, destination allocated buffer
 112  * of length size.
 113  *
 114  * Return the size of the received message or else a negative value on error
 115  * with errno being set by recvmsg() syscall.
 116  */
 117 static
 118 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 119 {
 120         ssize_t ret;
 121
 122         ret = sock->ops->recvmsg(sock, buf, size, 0);
 123         if (ret < 0 || ret != size) {
 124                 if (ret == 0) {
 125                         /* Orderly shutdown. Not necessary to print an error. */
 126                         DBG("Socket %d did an orderly shutdown", sock->fd);
 127                 } else {
 128                         ERR("Relay failed to receive request.");
 129                 }
 130                 ret = -1;
 131         }
 132
 133         return ret;
 134 }
 135
 136 /*
 137  * Send a response buffer using a given socket, source allocated buffer of
 138  * length size.
 139  *
 140  * Return the size of the sent message or else a negative value on error with
 141  * errno being set by sendmsg() syscall.
 142  */
 143 static
 144 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 145 {
 146         ssize_t ret;
 147
 148         ret = sock->ops->sendmsg(sock, buf, size, 0);
 149         if (ret < 0) {
 150                 ERR("Relayd failed to send response.");
 151         }
 152
 153         return ret;
 154 }
 155
 156 /*
 157  * Atomically check if new streams got added in one of the sessions attached
 158  * and reset the flag to 0.
 159  *
 160  * Returns 1 if new streams got added, 0 if nothing changed, a negative value
 161  * on error.
 162  */
 163 static
 164 int check_new_streams(struct relay_connection *conn)
 165 {
 166         struct relay_session *session;
 167         unsigned long current_val;
 168         int ret = 0;
 169
 170         if (!conn->viewer_session) {
 171                 goto end;
 172         }
 173         rcu_read_lock();
 174         cds_list_for_each_entry_rcu(session,
 175                         &conn->viewer_session->session_list,
 176                         viewer_session_node) {
 177                 if (!session_get(session)) {
 178                         continue;
 179                 }
 180                 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
 181                 ret = current_val;
 182                 session_put(session);
 183                 if (ret == 1) {
 184                         goto end;
 185                 }
 186         }
 187 end:
 188         rcu_read_unlock();
 189         return ret;
 190 }
 191
 192 /*
 193  * Send viewer streams to the given socket. The ignore_sent_flag indicates if
 194  * this function should ignore the sent flag or not.
 195  *
 196  * Return 0 on success or else a negative value.
 197  */
 198 static
 199 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
 200                 uint64_t session_id, unsigned int ignore_sent_flag)
 201 {
 202         ssize_t ret;
 203         struct lttng_viewer_stream send_stream;
 204         struct lttng_ht_iter iter;
 205         struct relay_viewer_stream *vstream;
 206
 207         rcu_read_lock();
 208
 209         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
 210                         stream_n.node) {
 211                 struct ctf_trace *ctf_trace;
 212
 213                 health_code_update();
 214
 215                 if (!viewer_stream_get(vstream)) {
 216                         continue;
 217                 }
 218
 219                 pthread_mutex_lock(&vstream->stream->lock);
 220                 /* Ignore if not the same session. */
 221                 if (vstream->stream->trace->session->id != session_id ||
 222                                 (!ignore_sent_flag && vstream->sent_flag)) {
 223                         pthread_mutex_unlock(&vstream->stream->lock);
 224                         viewer_stream_put(vstream);
 225                         continue;
 226                 }
 227
 228                 ctf_trace = vstream->stream->trace;
 229                 send_stream.id = htobe64(vstream->stream->stream_handle);
 230                 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
 231                 send_stream.metadata_flag = htobe32(
 232                                 vstream->stream->is_metadata);
 233                 if (lttng_strncpy(send_stream.path_name, vstream->path_name,
 234                                 sizeof(send_stream.path_name))) {
 235                         pthread_mutex_unlock(&vstream->stream->lock);
 236                         viewer_stream_put(vstream);
 237                         ret = -1;       /* Error. */
 238                         goto end_unlock;
 239                 }
 240                 if (lttng_strncpy(send_stream.channel_name,
 241                                 vstream->channel_name,
 242                                 sizeof(send_stream.channel_name))) {
 243                         pthread_mutex_unlock(&vstream->stream->lock);
 244                         viewer_stream_put(vstream);
 245                         ret = -1;       /* Error. */
 246                         goto end_unlock;
 247                 }
 248
 249                 DBG("Sending stream %" PRIu64 " to viewer",
 250                                 vstream->stream->stream_handle);
 251                 vstream->sent_flag = 1;
 252                 pthread_mutex_unlock(&vstream->stream->lock);
 253
 254                 ret = send_response(sock, &send_stream, sizeof(send_stream));
 255                 viewer_stream_put(vstream);
 256                 if (ret < 0) {
 257                         goto end_unlock;
 258                 }
 259         }
 260
 261         ret = 0;
 262
 263 end_unlock:
 264         rcu_read_unlock();
 265         return ret;
 266 }
 267
 268 /*
 269  * Create every viewer stream possible for the given session with the seek
 270  * type. Three counters *can* be return which are in order the total amount of
 271  * viewer stream of the session, the number of unsent stream and the number of
 272  * stream created. Those counters can be NULL and thus will be ignored.
 273  *
 274  * session must be locked to ensure that we see either none or all initial
 275  * streams for a session, but no intermediate state..
 276  *
 277  * Return 0 on success or else a negative value.
 278  */
 279 static int make_viewer_streams(struct relay_session *session,
 280                 struct lttng_trace_chunk *viewer_trace_chunk,
 281                 enum lttng_viewer_seek seek_t,
 282                 uint32_t *nb_total,
 283                 uint32_t *nb_unsent,
 284                 uint32_t *nb_created,
 285                 bool *closed)
 286 {
 287         int ret;
 288         struct lttng_ht_iter iter;
 289         struct ctf_trace *ctf_trace;
 290
 291         assert(session);
 292         ASSERT_LOCKED(session->lock);
 293
 294         if (!viewer_trace_chunk) {
 295                 ERR("Internal error: viewer session associated with session \"%s\" has a NULL trace chunk",
 296                                 session->session_name);
 297                 ret = -1;
 298                 goto error;
 299         }
 300
 301         if (session->connection_closed) {
 302                 *closed = true;
 303         }
 304
 305         /*
 306          * Create viewer streams for relay streams that are ready to be
 307          * used for a the given session id only.
 308          */
 309         rcu_read_lock();
 310         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 311                         node.node) {
 312                 bool trace_has_metadata_stream = false;
 313                 struct relay_stream *stream;
 314
 315                 health_code_update();
 316
 317                 if (!ctf_trace_get(ctf_trace)) {
 318                         continue;
 319                 }
 320
 321                 /*
 322                  * Iterate over all the streams of the trace to see if we have a
 323                  * metadata stream.
 324                  */
 325                 cds_list_for_each_entry_rcu(
 326                                 stream, &ctf_trace->stream_list, stream_node)
 327                 {
 328                         if (stream->is_metadata) {
 329                                 trace_has_metadata_stream = true;
 330                                 break;
 331                         }
 332                 }
 333
 334                 /*
 335                  * If there is no metadata stream in this trace at the moment
 336                  * and we never sent one to the viewer, skip the trace. We
 337                  * accept that the viewer will not see this trace at all.
 338                  */
 339                 if (!trace_has_metadata_stream &&
 340                                 !ctf_trace->metadata_stream_sent_to_viewer) {
 341                         ctf_trace_put(ctf_trace);
 342                         continue;
 343                 }
 344
 345                 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
 346                         struct relay_viewer_stream *vstream;
 347
 348                         if (!stream_get(stream)) {
 349                                 continue;
 350                         }
 351                         /*
 352                          * stream published is protected by the session lock.
 353                          */
 354                         if (!stream->published) {
 355                                 goto next;
 356                         }
 357                         vstream = viewer_stream_get_by_id(stream->stream_handle);
 358                         if (!vstream) {
 359                                 /*
 360                                  * Save that we sent the metadata stream to the
 361                                  * viewer. So that we know what trace the viewer
 362                                  * is aware of.
 363                                  */
 364                                 if (stream->is_metadata) {
 365                                         ctf_trace->metadata_stream_sent_to_viewer =
 366                                                         true;
 367                                 }
 368                                 vstream = viewer_stream_create(stream,
 369                                                 viewer_trace_chunk, seek_t);
 370                                 if (!vstream) {
 371                                         ret = -1;
 372                                         ctf_trace_put(ctf_trace);
 373                                         stream_put(stream);
 374                                         goto error_unlock;
 375                                 }
 376
 377                                 if (nb_created) {
 378                                         /* Update number of created stream counter. */
 379                                         (*nb_created)++;
 380                                 }
 381                                 /*
 382                                  * Ensure a self-reference is preserved even
 383                                  * after we have put our local reference.
 384                                  */
 385                                 if (!viewer_stream_get(vstream)) {
 386                                         ERR("Unable to get self-reference on viewer stream, logic error.");
 387                                         abort();
 388                                 }
 389                         } else {
 390                                 if (!vstream->sent_flag && nb_unsent) {
 391                                         /* Update number of unsent stream counter. */
 392                                         (*nb_unsent)++;
 393                                 }
 394                         }
 395                         /* Update number of total stream counter. */
 396                         if (nb_total) {
 397                                 if (stream->is_metadata) {
 398                                         if (!stream->closed ||
 399                                                         stream->metadata_received > vstream->metadata_sent) {
 400                                                 (*nb_total)++;
 401                                         }
 402                                 } else {
 403                                         if (!stream->closed ||
 404                                                 !(((int64_t) (stream->prev_data_seq - stream->last_net_seq_num)) >= 0)) {
 405
 406                                                 (*nb_total)++;
 407                                         }
 408                                 }
 409                         }
 410                         /* Put local reference. */
 411                         viewer_stream_put(vstream);
 412                 next:
 413                         stream_put(stream);
 414                 }
 415                 ctf_trace_put(ctf_trace);
 416         }
 417
 418         ret = 0;
 419
 420 error_unlock:
 421         rcu_read_unlock();
 422 error:
 423         return ret;
 424 }
 425
 426 int relayd_live_stop(void)
 427 {
 428         /* Stop dispatch thread */
 429         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 430         futex_nto1_wake(&viewer_conn_queue.futex);
 431         return 0;
 432 }
 433
 434 /*
 435  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 436  */
 437 static
 438 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 439 {
 440         int ret;
 441
 442         if (events == NULL || size == 0) {
 443                 ret = -1;
 444                 goto error;
 445         }
 446
 447         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 448         if (ret < 0) {
 449                 goto error;
 450         }
 451
 452         /* Add quit pipe */
 453         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 454         if (ret < 0) {
 455                 goto error;
 456         }
 457
 458         return 0;
 459
 460 error:
 461         return ret;
 462 }
 463
 464 /*
 465  * Check if the thread quit pipe was triggered.
 466  *
 467  * Return 1 if it was triggered else 0;
 468  */
 469 static
 470 int check_thread_quit_pipe(int fd, uint32_t events)
 471 {
 472         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 473                 return 1;
 474         }
 475
 476         return 0;
 477 }
 478
 479 /*
 480  * Create and init socket from uri.
 481  */
 482 static
 483 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 484 {
 485         int ret;
 486         struct lttcomm_sock *sock = NULL;
 487
 488         sock = lttcomm_alloc_sock_from_uri(uri);
 489         if (sock == NULL) {
 490                 ERR("Allocating socket");
 491                 goto error;
 492         }
 493
 494         ret = lttcomm_create_sock(sock);
 495         if (ret < 0) {
 496                 goto error;
 497         }
 498         DBG("Listening on sock %d for lttng-live", sock->fd);
 499
 500         ret = sock->ops->bind(sock);
 501         if (ret < 0) {
 502                 PERROR("Failed to bind lttng-live socket");
 503                 goto error;
 504         }
 505
 506         ret = sock->ops->listen(sock, -1);
 507         if (ret < 0) {
 508                 goto error;
 509
 510         }
 511
 512         return sock;
 513
 514 error:
 515         if (sock) {
 516                 lttcomm_destroy_sock(sock);
 517         }
 518         return NULL;
 519 }
 520
 521 /*
 522  * This thread manages the listening for new connections on the network
 523  */
 524 static
 525 void *thread_listener(void *data)
 526 {
 527         int i, ret, pollfd, err = -1;
 528         uint32_t revents, nb_fd;
 529         struct lttng_poll_event events;
 530         struct lttcomm_sock *live_control_sock;
 531
 532         DBG("[thread] Relay live listener started");
 533
 534         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 535
 536         health_code_update();
 537
 538         live_control_sock = init_socket(live_uri);
 539         if (!live_control_sock) {
 540                 goto error_sock_control;
 541         }
 542
 543         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 544         ret = create_thread_poll_set(&events, 2);
 545         if (ret < 0) {
 546                 goto error_create_poll;
 547         }
 548
 549         /* Add the control socket */
 550         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 551         if (ret < 0) {
 552                 goto error_poll_add;
 553         }
 554
 555         lttng_relay_notify_ready();
 556
 557         if (testpoint(relayd_thread_live_listener)) {
 558                 goto error_testpoint;
 559         }
 560
 561         while (1) {
 562                 health_code_update();
 563
 564                 DBG("Listener accepting live viewers connections");
 565
 566 restart:
 567                 health_poll_entry();
 568                 ret = lttng_poll_wait(&events, -1);
 569                 health_poll_exit();
 570                 if (ret < 0) {
 571                         /*
 572                          * Restart interrupted system call.
 573                          */
 574                         if (errno == EINTR) {
 575                                 goto restart;
 576                         }
 577                         goto error;
 578                 }
 579                 nb_fd = ret;
 580
 581                 DBG("Relay new viewer connection received");
 582                 for (i = 0; i < nb_fd; i++) {
 583                         health_code_update();
 584
 585                         /* Fetch once the poll data */
 586                         revents = LTTNG_POLL_GETEV(&events, i);
 587                         pollfd = LTTNG_POLL_GETFD(&events, i);
 588
 589                         /* Thread quit pipe has been closed. Killing thread. */
 590                         ret = check_thread_quit_pipe(pollfd, revents);
 591                         if (ret) {
 592                                 err = 0;
 593                                 goto exit;
 594                         }
 595
 596                         if (revents & LPOLLIN) {
 597                                 /*
 598                                  * A new connection is requested, therefore a
 599                                  * viewer connection is allocated in this
 600                                  * thread, enqueued to a global queue and
 601                                  * dequeued (and freed) in the worker thread.
 602                                  */
 603                                 int val = 1;
 604                                 struct relay_connection *new_conn;
 605                                 struct lttcomm_sock *newsock;
 606
 607                                 newsock = live_control_sock->ops->accept(live_control_sock);
 608                                 if (!newsock) {
 609                                         PERROR("accepting control sock");
 610                                         goto error;
 611                                 }
 612                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 613
 614                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 615                                                 sizeof(val));
 616                                 if (ret < 0) {
 617                                         PERROR("setsockopt inet");
 618                                         lttcomm_destroy_sock(newsock);
 619                                         goto error;
 620                                 }
 621                                 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
 622                                 if (!new_conn) {
 623                                         lttcomm_destroy_sock(newsock);
 624                                         goto error;
 625                                 }
 626                                 /* Ownership assumed by the connection. */
 627                                 newsock = NULL;
 628
 629                                 /* Enqueue request for the dispatcher thread. */
 630                                 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
 631                                                  &new_conn->qnode);
 632
 633                                 /*
 634                                  * Wake the dispatch queue futex.
 635                                  * Implicit memory barrier with the
 636                                  * exchange in cds_wfcq_enqueue.
 637                                  */
 638                                 futex_nto1_wake(&viewer_conn_queue.futex);
 639                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 640                                 ERR("socket poll error");
 641                                 goto error;
 642                         } else {
 643                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
 644                                 goto error;
 645                         }
 646                 }
 647         }
 648
 649 exit:
 650 error:
 651 error_poll_add:
 652 error_testpoint:
 653         lttng_poll_clean(&events);
 654 error_create_poll:
 655         if (live_control_sock->fd >= 0) {
 656                 ret = live_control_sock->ops->close(live_control_sock);
 657                 if (ret) {
 658                         PERROR("close");
 659                 }
 660         }
 661         lttcomm_destroy_sock(live_control_sock);
 662 error_sock_control:
 663         if (err) {
 664                 health_error();
 665                 DBG("Live viewer listener thread exited with error");
 666         }
 667         health_unregister(health_relayd);
 668         DBG("Live viewer listener thread cleanup complete");
 669         if (lttng_relay_stop_threads()) {
 670                 ERR("Error stopping threads");
 671         }
 672         return NULL;
 673 }
 674
 675 /*
 676  * This thread manages the dispatching of the requests to worker threads
 677  */
 678 static
 679 void *thread_dispatcher(void *data)
 680 {
 681         int err = -1;
 682         ssize_t ret;
 683         struct cds_wfcq_node *node;
 684         struct relay_connection *conn = NULL;
 685
 686         DBG("[thread] Live viewer relay dispatcher started");
 687
 688         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 689
 690         if (testpoint(relayd_thread_live_dispatcher)) {
 691                 goto error_testpoint;
 692         }
 693
 694         health_code_update();
 695
 696         for (;;) {
 697                 health_code_update();
 698
 699                 /* Atomically prepare the queue futex */
 700                 futex_nto1_prepare(&viewer_conn_queue.futex);
 701
 702                 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 703                         break;
 704                 }
 705
 706                 do {
 707                         health_code_update();
 708
 709                         /* Dequeue commands */
 710                         node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
 711                                                          &viewer_conn_queue.tail);
 712                         if (node == NULL) {
 713                                 DBG("Woken up but nothing in the live-viewer "
 714                                                 "relay command queue");
 715                                 /* Continue thread execution */
 716                                 break;
 717                         }
 718                         conn = caa_container_of(node, struct relay_connection, qnode);
 719                         DBG("Dispatching viewer request waiting on sock %d",
 720                                         conn->sock->fd);
 721
 722                         /*
 723                          * Inform worker thread of the new request. This
 724                          * call is blocking so we can be assured that
 725                          * the data will be read at some point in time
 726                          * or wait to the end of the world :)
 727                          */
 728                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 729                         if (ret < 0) {
 730                                 PERROR("write conn pipe");
 731                                 connection_put(conn);
 732                                 goto error;
 733                         }
 734                 } while (node != NULL);
 735
 736                 /* Futex wait on queue. Blocking call on futex() */
 737                 health_poll_entry();
 738                 futex_nto1_wait(&viewer_conn_queue.futex);
 739                 health_poll_exit();
 740         }
 741
 742         /* Normal exit, no error */
 743         err = 0;
 744
 745 error:
 746 error_testpoint:
 747         if (err) {
 748                 health_error();
 749                 ERR("Health error occurred in %s", __func__);
 750         }
 751         health_unregister(health_relayd);
 752         DBG("Live viewer dispatch thread dying");
 753         if (lttng_relay_stop_threads()) {
 754                 ERR("Error stopping threads");
 755         }
 756         return NULL;
 757 }
 758
 759 /*
 760  * Establish connection with the viewer and check the versions.
 761  *
 762  * Return 0 on success or else negative value.
 763  */
 764 static
 765 int viewer_connect(struct relay_connection *conn)
 766 {
 767         int ret;
 768         struct lttng_viewer_connect reply, msg;
 769
 770         conn->version_check_done = 1;
 771
 772         health_code_update();
 773
 774         DBG("Viewer is establishing a connection to the relayd.");
 775
 776         ret = recv_request(conn->sock, &msg, sizeof(msg));
 777         if (ret < 0) {
 778                 goto end;
 779         }
 780
 781         health_code_update();
 782
 783         memset(&reply, 0, sizeof(reply));
 784         reply.major = RELAYD_VERSION_COMM_MAJOR;
 785         reply.minor = RELAYD_VERSION_COMM_MINOR;
 786
 787         /* Major versions must be the same */
 788         if (reply.major != be32toh(msg.major)) {
 789                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 790                                 reply.major, be32toh(msg.major));
 791                 ret = -1;
 792                 goto end;
 793         }
 794
 795         conn->major = reply.major;
 796         /* We adapt to the lowest compatible version */
 797         if (reply.minor <= be32toh(msg.minor)) {
 798                 conn->minor = reply.minor;
 799         } else {
 800                 conn->minor = be32toh(msg.minor);
 801         }
 802
 803         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 804                 conn->type = RELAY_VIEWER_COMMAND;
 805         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 806                 conn->type = RELAY_VIEWER_NOTIFICATION;
 807         } else {
 808                 ERR("Unknown connection type : %u", be32toh(msg.type));
 809                 ret = -1;
 810                 goto end;
 811         }
 812
 813         reply.major = htobe32(reply.major);
 814         reply.minor = htobe32(reply.minor);
 815         if (conn->type == RELAY_VIEWER_COMMAND) {
 816                 /*
 817                  * Increment outside of htobe64 macro, because the argument can
 818                  * be used more than once within the macro, and thus the
 819                  * operation may be undefined.
 820                  */
 821                 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
 822                 last_relay_viewer_session_id++;
 823                 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
 824                 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
 825         }
 826
 827         health_code_update();
 828
 829         ret = send_response(conn->sock, &reply, sizeof(reply));
 830         if (ret < 0) {
 831                 goto end;
 832         }
 833
 834         health_code_update();
 835
 836         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 837         ret = 0;
 838
 839 end:
 840         return ret;
 841 }
 842
 843 /*
 844  * Send the viewer the list of current sessions.
 845  * We need to create a copy of the hash table content because otherwise
 846  * we cannot assume the number of entries stays the same between getting
 847  * the number of HT elements and iteration over the HT.
 848  *
 849  * Return 0 on success or else a negative value.
 850  */
 851 static
 852 int viewer_list_sessions(struct relay_connection *conn)
 853 {
 854         int ret = 0;
 855         struct lttng_viewer_list_sessions session_list;
 856         struct lttng_ht_iter iter;
 857         struct relay_session *session;
 858         struct lttng_viewer_session *send_session_buf = NULL;
 859         uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
 860         uint32_t count = 0;
 861
 862         DBG("List sessions received");
 863
 864         send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
 865         if (!send_session_buf) {
 866                 return -1;
 867         }
 868
 869         rcu_read_lock();
 870         cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
 871                         session_n.node) {
 872                 struct lttng_viewer_session *send_session;
 873
 874                 health_code_update();
 875
 876                 pthread_mutex_lock(&session->lock);
 877                 if (session->connection_closed) {
 878                         /* Skip closed session */
 879                         goto next_session;
 880                 }
 881                 if (!session->current_trace_chunk) {
 882                         /*
 883                          * Skip un-attachable session. It is either
 884                          * being destroyed or has not had a trace
 885                          * chunk created against it yet.
 886                          */
 887                         goto next_session;
 888                 }
 889
 890                 if (count >= buf_count) {
 891                         struct lttng_viewer_session *newbuf;
 892                         uint32_t new_buf_count = buf_count << 1;
 893
 894                         newbuf = realloc(send_session_buf,
 895                                 new_buf_count * sizeof(*send_session_buf));
 896                         if (!newbuf) {
 897                                 ret = -1;
 898                                 goto break_loop;
 899                         }
 900                         send_session_buf = newbuf;
 901                         buf_count = new_buf_count;
 902                 }
 903                 send_session = &send_session_buf[count];
 904                 if (lttng_strncpy(send_session->session_name,
 905                                 session->session_name,
 906                                 sizeof(send_session->session_name))) {
 907                         ret = -1;
 908                         goto break_loop;
 909                 }
 910                 if (lttng_strncpy(send_session->hostname, session->hostname,
 911                                 sizeof(send_session->hostname))) {
 912                         ret = -1;
 913                         goto break_loop;
 914                 }
 915                 send_session->id = htobe64(session->id);
 916                 send_session->live_timer = htobe32(session->live_timer);
 917                 if (session->viewer_attached) {
 918                         send_session->clients = htobe32(1);
 919                 } else {
 920                         send_session->clients = htobe32(0);
 921                 }
 922                 send_session->streams = htobe32(session->stream_count);
 923                 count++;
 924         next_session:
 925                 pthread_mutex_unlock(&session->lock);
 926                 continue;
 927         break_loop:
 928                 pthread_mutex_unlock(&session->lock);
 929                 break;
 930         }
 931         rcu_read_unlock();
 932         if (ret < 0) {
 933                 goto end_free;
 934         }
 935
 936         session_list.sessions_count = htobe32(count);
 937
 938         health_code_update();
 939
 940         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 941         if (ret < 0) {
 942                 goto end_free;
 943         }
 944
 945         health_code_update();
 946
 947         ret = send_response(conn->sock, send_session_buf,
 948                         count * sizeof(*send_session_buf));
 949         if (ret < 0) {
 950                 goto end_free;
 951         }
 952         health_code_update();
 953
 954         ret = 0;
 955 end_free:
 956         free(send_session_buf);
 957         return ret;
 958 }
 959
 960 /*
 961  * Send the viewer the list of current streams.
 962  */
 963 static
 964 int viewer_get_new_streams(struct relay_connection *conn)
 965 {
 966         int ret, send_streams = 0;
 967         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
 968         struct lttng_viewer_new_streams_request request;
 969         struct lttng_viewer_new_streams_response response;
 970         struct relay_session *session = NULL;
 971         uint64_t session_id;
 972         bool closed = false;
 973
 974         assert(conn);
 975
 976         DBG("Get new streams received");
 977
 978         health_code_update();
 979
 980         /* Receive the request from the connected client. */
 981         ret = recv_request(conn->sock, &request, sizeof(request));
 982         if (ret < 0) {
 983                 goto error;
 984         }
 985         session_id = be64toh(request.session_id);
 986
 987         health_code_update();
 988
 989         memset(&response, 0, sizeof(response));
 990
 991         session = session_get_by_id(session_id);
 992         if (!session) {
 993                 DBG("Relay session %" PRIu64 " not found", session_id);
 994                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 995                 goto send_reply;
 996         }
 997
 998         if (!viewer_session_is_attached(conn->viewer_session, session)) {
 999                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
1000                 goto send_reply;
1001         }
1002
1003         pthread_mutex_lock(&session->lock);
1004         ret = make_viewer_streams(session,
1005                         conn->viewer_session->current_trace_chunk,
1006                         LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
1007                         &nb_created, &closed);
1008         if (ret < 0) {
1009                 goto error_unlock_session;
1010         }
1011         send_streams = 1;
1012         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
1013
1014         /* Only send back the newly created streams with the unsent ones. */
1015         nb_streams = nb_created + nb_unsent;
1016         response.streams_count = htobe32(nb_streams);
1017
1018         /*
1019          * If the session is closed, HUP when there are no more streams
1020          * with data.
1021          */
1022         if (closed && nb_total == 0) {
1023                 send_streams = 0;
1024                 response.streams_count = 0;
1025                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1026                 goto send_reply_unlock;
1027         }
1028 send_reply_unlock:
1029         pthread_mutex_unlock(&session->lock);
1030
1031 send_reply:
1032         health_code_update();
1033         ret = send_response(conn->sock, &response, sizeof(response));
1034         if (ret < 0) {
1035                 goto end_put_session;
1036         }
1037         health_code_update();
1038
1039         /*
1040          * Unknown or empty session, just return gracefully, the viewer
1041          * knows what is happening.
1042          */
1043         if (!send_streams || !nb_streams) {
1044                 ret = 0;
1045                 goto end_put_session;
1046         }
1047
1048         /*
1049          * Send stream and *DON'T* ignore the sent flag so every viewer
1050          * streams that were not sent from that point will be sent to
1051          * the viewer.
1052          */
1053         ret = send_viewer_streams(conn->sock, session_id, 0);
1054         if (ret < 0) {
1055                 goto end_put_session;
1056         }
1057
1058 end_put_session:
1059         if (session) {
1060                 session_put(session);
1061         }
1062 error:
1063         return ret;
1064 error_unlock_session:
1065         pthread_mutex_unlock(&session->lock);
1066         session_put(session);
1067         return ret;
1068 }
1069
1070 /*
1071  * Send the viewer the list of current sessions.
1072  */
1073 static
1074 int viewer_attach_session(struct relay_connection *conn)
1075 {
1076         int send_streams = 0;
1077         ssize_t ret;
1078         uint32_t nb_streams = 0;
1079         enum lttng_viewer_seek seek_type;
1080         struct lttng_viewer_attach_session_request request;
1081         struct lttng_viewer_attach_session_response response;
1082         struct relay_session *session = NULL;
1083         enum lttng_viewer_attach_return_code viewer_attach_status;
1084         bool closed = false;
1085         uint64_t session_id;
1086
1087         assert(conn);
1088
1089         health_code_update();
1090
1091         /* Receive the request from the connected client. */
1092         ret = recv_request(conn->sock, &request, sizeof(request));
1093         if (ret < 0) {
1094                 goto error;
1095         }
1096
1097         session_id = be64toh(request.session_id);
1098         health_code_update();
1099
1100         memset(&response, 0, sizeof(response));
1101
1102         if (!conn->viewer_session) {
1103                 DBG("Client trying to attach before creating a live viewer session");
1104                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1105                 goto send_reply;
1106         }
1107
1108         session = session_get_by_id(session_id);
1109         if (!session) {
1110                 DBG("Relay session %" PRIu64 " not found", session_id);
1111                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1112                 goto send_reply;
1113         }
1114         DBG("Attach session ID %" PRIu64 " received", session_id);
1115
1116         pthread_mutex_lock(&session->lock);
1117         if (!session->current_trace_chunk) {
1118                 /*
1119                  * Session is either being destroyed or it never had a trace
1120                  * chunk created against it.
1121                  */
1122                 DBG("Session requested by live client has no current trace chunk, returning unknown session");
1123                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1124                 goto send_reply;
1125         }
1126         if (session->live_timer == 0) {
1127                 DBG("Not live session");
1128                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1129                 goto send_reply;
1130         }
1131
1132         send_streams = 1;
1133         viewer_attach_status = viewer_session_attach(conn->viewer_session,
1134                         session);
1135         if (viewer_attach_status != LTTNG_VIEWER_ATTACH_OK) {
1136                 response.status = htobe32(viewer_attach_status);
1137                 goto send_reply;
1138         }
1139
1140         switch (be32toh(request.seek)) {
1141         case LTTNG_VIEWER_SEEK_BEGINNING:
1142         case LTTNG_VIEWER_SEEK_LAST:
1143                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1144                 seek_type = be32toh(request.seek);
1145                 break;
1146         default:
1147                 ERR("Wrong seek parameter");
1148                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1149                 send_streams = 0;
1150                 goto send_reply;
1151         }
1152
1153         ret = make_viewer_streams(session,
1154                         conn->viewer_session->current_trace_chunk, seek_type,
1155                         &nb_streams, NULL, NULL, &closed);
1156         if (ret < 0) {
1157                 goto end_put_session;
1158         }
1159         pthread_mutex_unlock(&session->lock);
1160         session_put(session);
1161         session = NULL;
1162
1163         response.streams_count = htobe32(nb_streams);
1164         /*
1165          * If the session is closed when the viewer is attaching, it
1166          * means some of the streams may have been concurrently removed,
1167          * so we don't allow the viewer to attach, even if there are
1168          * streams available.
1169          */
1170         if (closed) {
1171                 send_streams = 0;
1172                 response.streams_count = 0;
1173                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1174                 goto send_reply;
1175         }
1176
1177 send_reply:
1178         health_code_update();
1179         ret = send_response(conn->sock, &response, sizeof(response));
1180         if (ret < 0) {
1181                 goto end_put_session;
1182         }
1183         health_code_update();
1184
1185         /*
1186          * Unknown or empty session, just return gracefully, the viewer
1187          * knows what is happening.
1188          */
1189         if (!send_streams || !nb_streams) {
1190                 ret = 0;
1191                 goto end_put_session;
1192         }
1193
1194         /* Send stream and ignore the sent flag. */
1195         ret = send_viewer_streams(conn->sock, session_id, 1);
1196         if (ret < 0) {
1197                 goto end_put_session;
1198         }
1199
1200 end_put_session:
1201         if (session) {
1202                 pthread_mutex_unlock(&session->lock);
1203                 session_put(session);
1204         }
1205 error:
1206         return ret;
1207 }
1208
1209 /*
1210  * Open the index file if needed for the given vstream.
1211  *
1212  * If an index file is successfully opened, the vstream will set it as its
1213  * current index file.
1214  *
1215  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1216  *
1217  * Called with rstream lock held.
1218  */
1219 static int try_open_index(struct relay_viewer_stream *vstream,
1220                 struct relay_stream *rstream)
1221 {
1222         int ret = 0;
1223         const uint32_t connection_major = rstream->trace->session->major;
1224         const uint32_t connection_minor = rstream->trace->session->minor;
1225
1226         if (vstream->index_file) {
1227                 goto end;
1228         }
1229
1230         /*
1231          * First time, we open the index file and at least one index is ready.
1232          */
1233         if (rstream->index_received_seqcount == 0) {
1234                 ret = -ENOENT;
1235                 goto end;
1236         }
1237         vstream->index_file = lttng_index_file_create_from_trace_chunk_read_only(
1238                         vstream->stream_file.trace_chunk, rstream->path_name,
1239                         rstream->channel_name, rstream->tracefile_size,
1240                         vstream->current_tracefile_id,
1241                         lttng_to_index_major(connection_major, connection_minor),
1242                         lttng_to_index_minor(connection_major, connection_minor));
1243         if (!vstream->index_file) {
1244                 ret = -1;
1245         }
1246
1247 end:
1248         return ret;
1249 }
1250
1251 /*
1252  * Check the status of the index for the given stream. This function
1253  * updates the index structure if needed and can put (close) the vstream
1254  * in the HUP situation.
1255  *
1256  * Return 0 means that we can proceed with the index. A value of 1 means
1257  * that the index has been updated and is ready to be sent to the
1258  * client. A negative value indicates an error that can't be handled.
1259  *
1260  * Called with rstream lock held.
1261  */
1262 static int check_index_status(struct relay_viewer_stream *vstream,
1263                 struct relay_stream *rstream, struct ctf_trace *trace,
1264                 struct lttng_viewer_index *index)
1265 {
1266         int ret;
1267
1268         if ((trace->session->connection_closed || rstream->closed)
1269                         && rstream->index_received_seqcount
1270                                 == vstream->index_sent_seqcount) {
1271                 /*
1272                  * Last index sent and session connection or relay
1273                  * stream are closed.
1274                  */
1275                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1276                 goto hup;
1277         } else if (rstream->beacon_ts_end != -1ULL &&
1278                         rstream->index_received_seqcount
1279                                 == vstream->index_sent_seqcount) {
1280                 /*
1281                  * We've received a synchronization beacon and the last index
1282                  * available has been sent, the index for now is inactive.
1283                  *
1284                  * In this case, we have received a beacon which allows us to
1285                  * inform the client of a time interval during which we can
1286                  * guarantee that there are no events to read (and never will
1287                  * be).
1288                  */
1289                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1290                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1291                 index->stream_id = htobe64(rstream->ctf_stream_id);
1292                 goto index_ready;
1293         } else if (rstream->index_received_seqcount
1294                         == vstream->index_sent_seqcount) {
1295                 /*
1296                  * This checks whether received == sent seqcount. In
1297                  * this case, we have not received a beacon. Therefore,
1298                  * we can only ask the client to retry later.
1299                  */
1300                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1301                 goto index_ready;
1302         } else if (!tracefile_array_seq_in_file(rstream->tfa,
1303                         vstream->current_tracefile_id,
1304                         vstream->index_sent_seqcount)) {
1305                 /*
1306                  * The next index we want to send cannot be read either
1307                  * because we need to perform a rotation, or due to
1308                  * the producer having overwritten its trace file.
1309                  */
1310                 DBG("Viewer stream %" PRIu64 " rotation",
1311                                 vstream->stream->stream_handle);
1312                 ret = viewer_stream_rotate(vstream);
1313                 if (ret < 0) {
1314                         goto end;
1315                 } else if (ret == 1) {
1316                         /* EOF across entire stream. */
1317                         index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1318                         goto hup;
1319                 }
1320                 /*
1321                  * If we have been pushed due to overwrite, it
1322                  * necessarily means there is data that can be read in
1323                  * the stream. If we rotated because we reached the end
1324                  * of a tracefile, it means the following tracefile
1325                  * needs to contain at least one index, else we would
1326                  * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1327                  * viewer. The updated index_sent_seqcount needs to
1328                  * point to a readable index entry now.
1329                  *
1330                  * In the case where we "rotate" on a single file, we
1331                  * can end up in a case where the requested index is
1332                  * still unavailable.
1333                  */
1334                 if (rstream->tracefile_count == 1 &&
1335                                 !tracefile_array_seq_in_file(
1336                                         rstream->tfa,
1337                                         vstream->current_tracefile_id,
1338                                         vstream->index_sent_seqcount)) {
1339                         index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1340                         goto index_ready;
1341                 }
1342                 assert(tracefile_array_seq_in_file(rstream->tfa,
1343                                 vstream->current_tracefile_id,
1344                                 vstream->index_sent_seqcount));
1345         }
1346         /* ret == 0 means successful so we continue. */
1347         ret = 0;
1348 end:
1349         return ret;
1350
1351 hup:
1352         viewer_stream_put(vstream);
1353 index_ready:
1354         return 1;
1355 }
1356
1357 /*
1358  * Send the next index for a stream.
1359  *
1360  * Return 0 on success or else a negative value.
1361  */
1362 static
1363 int viewer_get_next_index(struct relay_connection *conn)
1364 {
1365         int ret;
1366         struct lttng_viewer_get_next_index request_index;
1367         struct lttng_viewer_index viewer_index;
1368         struct ctf_packet_index packet_index;
1369         struct relay_viewer_stream *vstream = NULL;
1370         struct relay_stream *rstream = NULL;
1371         struct ctf_trace *ctf_trace = NULL;
1372         struct relay_viewer_stream *metadata_viewer_stream = NULL;
1373
1374         assert(conn);
1375
1376         DBG("Viewer get next index");
1377
1378         memset(&viewer_index, 0, sizeof(viewer_index));
1379         health_code_update();
1380
1381         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1382         if (ret < 0) {
1383                 goto end;
1384         }
1385         health_code_update();
1386
1387         vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1388         if (!vstream) {
1389                 DBG("Client requested index of unknown stream id %" PRIu64,
1390                                 (uint64_t) be64toh(request_index.stream_id));
1391                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1392                 goto send_reply;
1393         }
1394
1395         /* Use back. ref. Protected by refcounts. */
1396         rstream = vstream->stream;
1397         ctf_trace = rstream->trace;
1398
1399         /* metadata_viewer_stream may be NULL. */
1400         metadata_viewer_stream =
1401                         ctf_trace_get_viewer_metadata_stream(ctf_trace);
1402
1403         pthread_mutex_lock(&rstream->lock);
1404
1405         /*
1406          * The viewer should not ask for index on metadata stream.
1407          */
1408         if (rstream->is_metadata) {
1409                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1410                 goto send_reply;
1411         }
1412
1413         /* Try to open an index if one is needed for that stream. */
1414         ret = try_open_index(vstream, rstream);
1415         if (ret < 0) {
1416                 if (ret == -ENOENT) {
1417                         /*
1418                          * The index is created only when the first data
1419                          * packet arrives, it might not be ready at the
1420                          * beginning of the session
1421                          */
1422                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1423                 } else {
1424                         /* Unhandled error. */
1425                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1426                 }
1427                 goto send_reply;
1428         }
1429
1430         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1431         if (ret < 0) {
1432                 goto error_put;
1433         } else if (ret == 1) {
1434                 /*
1435                  * We have no index to send and check_index_status has populated
1436                  * viewer_index's status.
1437                  */
1438                 goto send_reply;
1439         }
1440         /* At this point, ret is 0 thus we will be able to read the index. */
1441         assert(!ret);
1442
1443         /*
1444          * vstream->stream_fd may be NULL if it has been closed by
1445          * tracefile rotation, or if we are at the beginning of the
1446          * stream. We open the data stream file here to protect against
1447          * overwrite caused by tracefile rotation (in association with
1448          * unlink performed before overwrite).
1449          */
1450         if (!vstream->stream_file.fd) {
1451                 int fd;
1452                 char file_path[LTTNG_PATH_MAX];
1453                 enum lttng_trace_chunk_status status;
1454
1455                 ret = utils_stream_file_path(rstream->path_name,
1456                                 rstream->channel_name, rstream->tracefile_size,
1457                                 vstream->current_tracefile_id, NULL, file_path,
1458                                 sizeof(file_path));
1459                 if (ret < 0) {
1460                         goto error_put;
1461                 }
1462
1463                 status = lttng_trace_chunk_open_file(
1464                                 vstream->stream_file.trace_chunk,
1465                                 file_path, O_RDONLY, 0, &fd);
1466                 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1467                         PERROR("Failed to open trace file for viewer stream");
1468                         goto error_put;
1469                 }
1470                 vstream->stream_file.fd = stream_fd_create(fd);
1471                 if (!vstream->stream_file.fd) {
1472                         if (close(fd)) {
1473                                 PERROR("Failed to close viewer stream file");
1474                         }
1475                         goto error_put;
1476                 }
1477         }
1478
1479         ret = check_new_streams(conn);
1480         if (ret < 0) {
1481                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1482                 goto send_reply;
1483         } else if (ret == 1) {
1484                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1485         }
1486
1487         ret = lttng_index_file_read(vstream->index_file, &packet_index);
1488         if (ret) {
1489                 ERR("Relay error reading index file %d",
1490                                 vstream->index_file->fd);
1491                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1492                 goto send_reply;
1493         } else {
1494                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1495                 vstream->index_sent_seqcount++;
1496         }
1497
1498         /*
1499          * Indexes are stored in big endian, no need to switch before sending.
1500          */
1501         DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1502                 rstream->stream_handle,
1503                 (uint64_t) be64toh(packet_index.offset));
1504         viewer_index.offset = packet_index.offset;
1505         viewer_index.packet_size = packet_index.packet_size;
1506         viewer_index.content_size = packet_index.content_size;
1507         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1508         viewer_index.timestamp_end = packet_index.timestamp_end;
1509         viewer_index.events_discarded = packet_index.events_discarded;
1510         viewer_index.stream_id = packet_index.stream_id;
1511
1512 send_reply:
1513         if (rstream) {
1514                 pthread_mutex_unlock(&rstream->lock);
1515         }
1516
1517         if (metadata_viewer_stream) {
1518                 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1519                 DBG("get next index metadata check: recv %" PRIu64
1520                                 " sent %" PRIu64,
1521                         metadata_viewer_stream->stream->metadata_received,
1522                         metadata_viewer_stream->metadata_sent);
1523                 if (!metadata_viewer_stream->stream->metadata_received ||
1524                                 metadata_viewer_stream->stream->metadata_received >
1525                                         metadata_viewer_stream->metadata_sent) {
1526                         viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1527                 }
1528                 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1529         }
1530
1531         viewer_index.flags = htobe32(viewer_index.flags);
1532         health_code_update();
1533
1534         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1535         if (ret < 0) {
1536                 goto end;
1537         }
1538         health_code_update();
1539
1540         if (vstream) {
1541                 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1542                                 vstream->index_sent_seqcount,
1543                                 vstream->stream->stream_handle);
1544         }
1545 end:
1546         if (metadata_viewer_stream) {
1547                 viewer_stream_put(metadata_viewer_stream);
1548         }
1549         if (vstream) {
1550                 viewer_stream_put(vstream);
1551         }
1552         return ret;
1553
1554 error_put:
1555         pthread_mutex_unlock(&rstream->lock);
1556         if (metadata_viewer_stream) {
1557                 viewer_stream_put(metadata_viewer_stream);
1558         }
1559         viewer_stream_put(vstream);
1560         return ret;
1561 }
1562
1563 /*
1564  * Send the next index for a stream
1565  *
1566  * Return 0 on success or else a negative value.
1567  */
1568 static
1569 int viewer_get_packet(struct relay_connection *conn)
1570 {
1571         int ret;
1572         off_t lseek_ret;
1573         char *reply = NULL;
1574         struct lttng_viewer_get_packet get_packet_info;
1575         struct lttng_viewer_trace_packet reply_header;
1576         struct relay_viewer_stream *vstream = NULL;
1577         uint32_t reply_size = sizeof(reply_header);
1578         uint32_t packet_data_len = 0;
1579         ssize_t read_len;
1580
1581         DBG2("Relay get data packet");
1582
1583         health_code_update();
1584
1585         ret = recv_request(conn->sock, &get_packet_info,
1586                         sizeof(get_packet_info));
1587         if (ret < 0) {
1588                 goto end;
1589         }
1590         health_code_update();
1591
1592         /* From this point on, the error label can be reached. */
1593         memset(&reply_header, 0, sizeof(reply_header));
1594
1595         vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1596         if (!vstream) {
1597                 DBG("Client requested packet of unknown stream id %" PRIu64,
1598                                 (uint64_t) be64toh(get_packet_info.stream_id));
1599                 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1600                 goto send_reply_nolock;
1601         } else {
1602                 packet_data_len = be32toh(get_packet_info.len);
1603                 reply_size += packet_data_len;
1604         }
1605
1606         reply = zmalloc(reply_size);
1607         if (!reply) {
1608                 PERROR("packet reply zmalloc");
1609                 reply_size = sizeof(reply_header);
1610                 goto error;
1611         }
1612
1613         pthread_mutex_lock(&vstream->stream->lock);
1614         lseek_ret = lseek(vstream->stream_file.fd->fd,
1615                         be64toh(get_packet_info.offset), SEEK_SET);
1616         if (lseek_ret < 0) {
1617                 PERROR("lseek fd %d to offset %" PRIu64,
1618                                 vstream->stream_file.fd->fd,
1619                                 (uint64_t) be64toh(get_packet_info.offset));
1620                 goto error;
1621         }
1622         read_len = lttng_read(vstream->stream_file.fd->fd,
1623                         reply + sizeof(reply_header), packet_data_len);
1624         if (read_len < packet_data_len) {
1625                 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1626                                 vstream->stream_file.fd->fd,
1627                                 (uint64_t) be64toh(get_packet_info.offset));
1628                 goto error;
1629         }
1630         reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1631         reply_header.len = htobe32(packet_data_len);
1632         goto send_reply;
1633
1634 error:
1635         reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1636
1637 send_reply:
1638         if (vstream) {
1639                 pthread_mutex_unlock(&vstream->stream->lock);
1640         }
1641 send_reply_nolock:
1642
1643         health_code_update();
1644
1645         if (reply) {
1646                 memcpy(reply, &reply_header, sizeof(reply_header));
1647                 ret = send_response(conn->sock, reply, reply_size);
1648         } else {
1649                 /* No reply to send. */
1650                 ret = send_response(conn->sock, &reply_header,
1651                                 reply_size);
1652         }
1653
1654         health_code_update();
1655         if (ret < 0) {
1656                 PERROR("sendmsg of packet data failed");
1657                 goto end_free;
1658         }
1659
1660         DBG("Sent %u bytes for stream %" PRIu64, reply_size,
1661                         (uint64_t) be64toh(get_packet_info.stream_id));
1662
1663 end_free:
1664         free(reply);
1665 end:
1666         if (vstream) {
1667                 viewer_stream_put(vstream);
1668         }
1669         return ret;
1670 }
1671
1672 /*
1673  * Send the session's metadata
1674  *
1675  * Return 0 on success else a negative value.
1676  */
1677 static
1678 int viewer_get_metadata(struct relay_connection *conn)
1679 {
1680         int ret = 0;
1681         ssize_t read_len;
1682         uint64_t len = 0;
1683         char *data = NULL;
1684         struct lttng_viewer_get_metadata request;
1685         struct lttng_viewer_metadata_packet reply;
1686         struct relay_viewer_stream *vstream = NULL;
1687
1688         assert(conn);
1689
1690         DBG("Relay get metadata");
1691
1692         health_code_update();
1693
1694         ret = recv_request(conn->sock, &request, sizeof(request));
1695         if (ret < 0) {
1696                 goto end;
1697         }
1698         health_code_update();
1699
1700         memset(&reply, 0, sizeof(reply));
1701
1702         vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1703         if (!vstream) {
1704                 /*
1705                  * The metadata stream can be closed by a CLOSE command
1706                  * just before we attach. It can also be closed by
1707                  * per-pid tracing during tracing. Therefore, it is
1708                  * possible that we cannot find this viewer stream.
1709                  * Reply back to the client with an error if we cannot
1710                  * find it.
1711                  */
1712                 DBG("Client requested metadata of unknown stream id %" PRIu64,
1713                                 (uint64_t) be64toh(request.stream_id));
1714                 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1715                 goto send_reply;
1716         }
1717         pthread_mutex_lock(&vstream->stream->lock);
1718         if (!vstream->stream->is_metadata) {
1719                 ERR("Invalid metadata stream");
1720                 goto error;
1721         }
1722
1723         assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1724
1725         len = vstream->stream->metadata_received - vstream->metadata_sent;
1726         if (len == 0) {
1727                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1728                 goto send_reply;
1729         }
1730
1731         /* first time, we open the metadata file */
1732         if (!vstream->stream_file.fd) {
1733                 int fd;
1734                 char file_path[LTTNG_PATH_MAX];
1735                 enum lttng_trace_chunk_status status;
1736                 struct relay_stream *rstream = vstream->stream;
1737
1738                 ret = utils_stream_file_path(rstream->path_name,
1739                                 rstream->channel_name, rstream->tracefile_size,
1740                                 vstream->current_tracefile_id, NULL, file_path,
1741                                 sizeof(file_path));
1742                 if (ret < 0) {
1743                         goto error;
1744                 }
1745
1746                 status = lttng_trace_chunk_open_file(
1747                                 vstream->stream_file.trace_chunk,
1748                                 file_path, O_RDONLY, 0, &fd);
1749                 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1750                         PERROR("Failed to open metadata file for viewer stream");
1751                         goto error;
1752                 }
1753                 vstream->stream_file.fd = stream_fd_create(fd);
1754                 if (!vstream->stream_file.fd) {
1755                         if (close(fd)) {
1756                                 PERROR("Failed to close viewer metadata file");
1757                         }
1758                         goto error;
1759                 }
1760         }
1761
1762         reply.len = htobe64(len);
1763         data = zmalloc(len);
1764         if (!data) {
1765                 PERROR("viewer metadata zmalloc");
1766                 goto error;
1767         }
1768
1769         read_len = lttng_read(vstream->stream_file.fd->fd, data, len);
1770         if (read_len < len) {
1771                 PERROR("Relay reading metadata file");
1772                 goto error;
1773         }
1774         vstream->metadata_sent += read_len;
1775         if (vstream->metadata_sent == vstream->stream->metadata_received
1776                         && vstream->stream->closed) {
1777                 /* Release ownership for the viewer metadata stream. */
1778                 viewer_stream_put(vstream);
1779         }
1780
1781         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1782
1783         goto send_reply;
1784
1785 error:
1786         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1787
1788 send_reply:
1789         health_code_update();
1790         if (vstream) {
1791                 pthread_mutex_unlock(&vstream->stream->lock);
1792         }
1793         ret = send_response(conn->sock, &reply, sizeof(reply));
1794         if (ret < 0) {
1795                 goto end_free;
1796         }
1797         health_code_update();
1798
1799         if (len > 0) {
1800                 ret = send_response(conn->sock, data, len);
1801                 if (ret < 0) {
1802                         goto end_free;
1803                 }
1804         }
1805
1806         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1807                         (uint64_t) be64toh(request.stream_id));
1808
1809         DBG("Metadata sent");
1810
1811 end_free:
1812         free(data);
1813 end:
1814         if (vstream) {
1815                 viewer_stream_put(vstream);
1816         }
1817         return ret;
1818 }
1819
1820 /*
1821  * Create a viewer session.
1822  *
1823  * Return 0 on success or else a negative value.
1824  */
1825 static
1826 int viewer_create_session(struct relay_connection *conn)
1827 {
1828         int ret;
1829         struct lttng_viewer_create_session_response resp;
1830
1831         DBG("Viewer create session received");
1832
1833         memset(&resp, 0, sizeof(resp));
1834         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1835         conn->viewer_session = viewer_session_create();
1836         if (!conn->viewer_session) {
1837                 ERR("Allocation viewer session");
1838                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1839                 goto send_reply;
1840         }
1841
1842 send_reply:
1843         health_code_update();
1844         ret = send_response(conn->sock, &resp, sizeof(resp));
1845         if (ret < 0) {
1846                 goto end;
1847         }
1848         health_code_update();
1849         ret = 0;
1850
1851 end:
1852         return ret;
1853 }
1854
1855 /*
1856  * Detach a viewer session.
1857  *
1858  * Return 0 on success or else a negative value.
1859  */
1860 static
1861 int viewer_detach_session(struct relay_connection *conn)
1862 {
1863         int ret;
1864         struct lttng_viewer_detach_session_response response;
1865         struct lttng_viewer_detach_session_request request;
1866         struct relay_session *session = NULL;
1867         uint64_t viewer_session_to_close;
1868
1869         DBG("Viewer detach session received");
1870
1871         assert(conn);
1872
1873         health_code_update();
1874
1875         /* Receive the request from the connected client. */
1876         ret = recv_request(conn->sock, &request, sizeof(request));
1877         if (ret < 0) {
1878                 goto end;
1879         }
1880         viewer_session_to_close = be64toh(request.session_id);
1881
1882         if (!conn->viewer_session) {
1883                 DBG("Client trying to detach before creating a live viewer session");
1884                 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1885                 goto send_reply;
1886         }
1887
1888         health_code_update();
1889
1890         memset(&response, 0, sizeof(response));
1891         DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
1892
1893         session = session_get_by_id(be64toh(request.session_id));
1894         if (!session) {
1895                 DBG("Relay session %" PRIu64 " not found",
1896                                 (uint64_t) be64toh(request.session_id));
1897                 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
1898                 goto send_reply;
1899         }
1900
1901         ret = viewer_session_is_attached(conn->viewer_session, session);
1902         if (ret != 1) {
1903                 DBG("Not attached to this session");
1904                 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1905                 goto send_reply_put;
1906         }
1907
1908         viewer_session_close_one_session(conn->viewer_session, session);
1909         response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
1910         DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
1911
1912 send_reply_put:
1913         session_put(session);
1914
1915 send_reply:
1916         health_code_update();
1917         ret = send_response(conn->sock, &response, sizeof(response));
1918         if (ret < 0) {
1919                 goto end;
1920         }
1921         health_code_update();
1922         ret = 0;
1923
1924 end:
1925         return ret;
1926 }
1927
1928 /*
1929  * live_relay_unknown_command: send -1 if received unknown command
1930  */
1931 static
1932 void live_relay_unknown_command(struct relay_connection *conn)
1933 {
1934         struct lttcomm_relayd_generic_reply reply;
1935
1936         memset(&reply, 0, sizeof(reply));
1937         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1938         (void) send_response(conn->sock, &reply, sizeof(reply));
1939 }
1940
1941 /*
1942  * Process the commands received on the control socket
1943  */
1944 static
1945 int process_control(struct lttng_viewer_cmd *recv_hdr,
1946                 struct relay_connection *conn)
1947 {
1948         int ret = 0;
1949         uint32_t msg_value;
1950
1951         msg_value = be32toh(recv_hdr->cmd);
1952
1953         /*
1954          * Make sure we've done the version check before any command other then a
1955          * new client connection.
1956          */
1957         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1958                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1959                 ret = -1;
1960                 goto end;
1961         }
1962
1963         switch (msg_value) {
1964         case LTTNG_VIEWER_CONNECT:
1965                 ret = viewer_connect(conn);
1966                 break;
1967         case LTTNG_VIEWER_LIST_SESSIONS:
1968                 ret = viewer_list_sessions(conn);
1969                 break;
1970         case LTTNG_VIEWER_ATTACH_SESSION:
1971                 ret = viewer_attach_session(conn);
1972                 break;
1973         case LTTNG_VIEWER_GET_NEXT_INDEX:
1974                 ret = viewer_get_next_index(conn);
1975                 break;
1976         case LTTNG_VIEWER_GET_PACKET:
1977                 ret = viewer_get_packet(conn);
1978                 break;
1979         case LTTNG_VIEWER_GET_METADATA:
1980                 ret = viewer_get_metadata(conn);
1981                 break;
1982         case LTTNG_VIEWER_GET_NEW_STREAMS:
1983                 ret = viewer_get_new_streams(conn);
1984                 break;
1985         case LTTNG_VIEWER_CREATE_SESSION:
1986                 ret = viewer_create_session(conn);
1987                 break;
1988         case LTTNG_VIEWER_DETACH_SESSION:
1989                 ret = viewer_detach_session(conn);
1990                 break;
1991         default:
1992                 ERR("Received unknown viewer command (%u)",
1993                                 be32toh(recv_hdr->cmd));
1994                 live_relay_unknown_command(conn);
1995                 ret = -1;
1996                 goto end;
1997         }
1998
1999 end:
2000         return ret;
2001 }
2002
2003 static
2004 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2005 {
2006         int ret;
2007
2008         (void) lttng_poll_del(events, pollfd);
2009
2010         ret = close(pollfd);
2011         if (ret < 0) {
2012                 ERR("Closing pollfd %d", pollfd);
2013         }
2014 }
2015
2016 /*
2017  * This thread does the actual work
2018  */
2019 static
2020 void *thread_worker(void *data)
2021 {
2022         int ret, err = -1;
2023         uint32_t nb_fd;
2024         struct lttng_poll_event events;
2025         struct lttng_ht *viewer_connections_ht;
2026         struct lttng_ht_iter iter;
2027         struct lttng_viewer_cmd recv_hdr;
2028         struct relay_connection *destroy_conn;
2029
2030         DBG("[thread] Live viewer relay worker started");
2031
2032         rcu_register_thread();
2033
2034         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
2035
2036         if (testpoint(relayd_thread_live_worker)) {
2037                 goto error_testpoint;
2038         }
2039
2040         /* table of connections indexed on socket */
2041         viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2042         if (!viewer_connections_ht) {
2043                 goto viewer_connections_ht_error;
2044         }
2045
2046         ret = create_thread_poll_set(&events, 2);
2047         if (ret < 0) {
2048                 goto error_poll_create;
2049         }
2050
2051         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2052         if (ret < 0) {
2053                 goto error;
2054         }
2055
2056 restart:
2057         while (1) {
2058                 int i;
2059
2060                 health_code_update();
2061
2062                 /* Infinite blocking call, waiting for transmission */
2063                 DBG3("Relayd live viewer worker thread polling...");
2064                 health_poll_entry();
2065                 ret = lttng_poll_wait(&events, -1);
2066                 health_poll_exit();
2067                 if (ret < 0) {
2068                         /*
2069                          * Restart interrupted system call.
2070                          */
2071                         if (errno == EINTR) {
2072                                 goto restart;
2073                         }
2074                         goto error;
2075                 }
2076
2077                 nb_fd = ret;
2078
2079                 /*
2080                  * Process control. The control connection is prioritised so we don't
2081                  * starve it with high throughput tracing data on the data
2082                  * connection.
2083                  */
2084                 for (i = 0; i < nb_fd; i++) {
2085                         /* Fetch once the poll data */
2086                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2087                         int pollfd = LTTNG_POLL_GETFD(&events, i);
2088
2089                         health_code_update();
2090
2091                         /* Thread quit pipe has been closed. Killing thread. */
2092                         ret = check_thread_quit_pipe(pollfd, revents);
2093                         if (ret) {
2094                                 err = 0;
2095                                 goto exit;
2096                         }
2097
2098                         /* Inspect the relay conn pipe for new connection. */
2099                         if (pollfd == live_conn_pipe[0]) {
2100                                 if (revents & LPOLLIN) {
2101                                         struct relay_connection *conn;
2102
2103                                         ret = lttng_read(live_conn_pipe[0],
2104                                                         &conn, sizeof(conn));
2105                                         if (ret < 0) {
2106                                                 goto error;
2107                                         }
2108                                         ret = lttng_poll_add(&events,
2109                                                         conn->sock->fd,
2110                                                         LPOLLIN | LPOLLRDHUP);
2111                                         if (ret) {
2112                                                 ERR("Failed to add new live connection file descriptor to poll set");
2113                                                 goto error;
2114                                         }
2115                                         connection_ht_add(viewer_connections_ht, conn);
2116                                         DBG("Connection socket %d added to poll", conn->sock->fd);
2117                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2118                                         ERR("Relay live pipe error");
2119                                         goto error;
2120                                 } else {
2121                                         ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2122                                         goto error;
2123                                 }
2124                         } else {
2125                                 /* Connection activity. */
2126                                 struct relay_connection *conn;
2127
2128                                 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2129                                 if (!conn) {
2130                                         continue;
2131                                 }
2132
2133                                 if (revents & LPOLLIN) {
2134                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2135                                                         sizeof(recv_hdr), 0);
2136                                         if (ret <= 0) {
2137                                                 /* Connection closed. */
2138                                                 cleanup_connection_pollfd(&events, pollfd);
2139                                                 /* Put "create" ownership reference. */
2140                                                 connection_put(conn);
2141                                                 DBG("Viewer control conn closed with %d", pollfd);
2142                                         } else {
2143                                                 ret = process_control(&recv_hdr, conn);
2144                                                 if (ret < 0) {
2145                                                         /* Clear the session on error. */
2146                                                         cleanup_connection_pollfd(&events, pollfd);
2147                                                         /* Put "create" ownership reference. */
2148                                                         connection_put(conn);
2149                                                         DBG("Viewer connection closed with %d", pollfd);
2150                                                 }
2151                                         }
2152                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2153                                         cleanup_connection_pollfd(&events, pollfd);
2154                                         /* Put "create" ownership reference. */
2155                                         connection_put(conn);
2156                                 } else {
2157                                         ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2158                                         connection_put(conn);
2159                                         goto error;
2160                                 }
2161                                 /* Put local "get_by_sock" reference. */
2162                                 connection_put(conn);
2163                         }
2164                 }
2165         }
2166
2167 exit:
2168 error:
2169         lttng_poll_clean(&events);
2170
2171         /* Cleanup remaining connection object. */
2172         rcu_read_lock();
2173         cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2174                         destroy_conn,
2175                         sock_n.node) {
2176                 health_code_update();
2177                 connection_put(destroy_conn);
2178         }
2179         rcu_read_unlock();
2180 error_poll_create:
2181         lttng_ht_destroy(viewer_connections_ht);
2182 viewer_connections_ht_error:
2183         /* Close relay conn pipes */
2184         utils_close_pipe(live_conn_pipe);
2185         if (err) {
2186                 DBG("Viewer worker thread exited with error");
2187         }
2188         DBG("Viewer worker thread cleanup complete");
2189 error_testpoint:
2190         if (err) {
2191                 health_error();
2192                 ERR("Health error occurred in %s", __func__);
2193         }
2194         health_unregister(health_relayd);
2195         if (lttng_relay_stop_threads()) {
2196                 ERR("Error stopping threads");
2197         }
2198         rcu_unregister_thread();
2199         return NULL;
2200 }
2201
2202 /*
2203  * Create the relay command pipe to wake thread_manage_apps.
2204  * Closed in cleanup().
2205  */
2206 static int create_conn_pipe(void)
2207 {
2208         return utils_create_pipe_cloexec(live_conn_pipe);
2209 }
2210
2211 int relayd_live_join(void)
2212 {
2213         int ret, retval = 0;
2214         void *status;
2215
2216         ret = pthread_join(live_listener_thread, &status);
2217         if (ret) {
2218                 errno = ret;
2219                 PERROR("pthread_join live listener");
2220                 retval = -1;
2221         }
2222
2223         ret = pthread_join(live_worker_thread, &status);
2224         if (ret) {
2225                 errno = ret;
2226                 PERROR("pthread_join live worker");
2227                 retval = -1;
2228         }
2229
2230         ret = pthread_join(live_dispatcher_thread, &status);
2231         if (ret) {
2232                 errno = ret;
2233                 PERROR("pthread_join live dispatcher");
2234                 retval = -1;
2235         }
2236
2237         cleanup_relayd_live();
2238
2239         return retval;
2240 }
2241
2242 /*
2243  * main
2244  */
2245 int relayd_live_create(struct lttng_uri *uri)
2246 {
2247         int ret = 0, retval = 0;
2248         void *status;
2249         int is_root;
2250
2251         if (!uri) {
2252                 retval = -1;
2253                 goto exit_init_data;
2254         }
2255         live_uri = uri;
2256
2257         /* Check if daemon is UID = 0 */
2258         is_root = !getuid();
2259
2260         if (!is_root) {
2261                 if (live_uri->port < 1024) {
2262                         ERR("Need to be root to use ports < 1024");
2263                         retval = -1;
2264                         goto exit_init_data;
2265                 }
2266         }
2267
2268         /* Setup the thread apps communication pipe. */
2269         if (create_conn_pipe()) {
2270                 retval = -1;
2271                 goto exit_init_data;
2272         }
2273
2274         /* Init relay command queue. */
2275         cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2276
2277         /* Set up max poll set size */
2278         if (lttng_poll_set_max_size()) {
2279                 retval = -1;
2280                 goto exit_init_data;
2281         }
2282
2283         /* Setup the dispatcher thread */
2284         ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2285                         thread_dispatcher, (void *) NULL);
2286         if (ret) {
2287                 errno = ret;
2288                 PERROR("pthread_create viewer dispatcher");
2289                 retval = -1;
2290                 goto exit_dispatcher_thread;
2291         }
2292
2293         /* Setup the worker thread */
2294         ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2295                         thread_worker, NULL);
2296         if (ret) {
2297                 errno = ret;
2298                 PERROR("pthread_create viewer worker");
2299                 retval = -1;
2300                 goto exit_worker_thread;
2301         }
2302
2303         /* Setup the listener thread */
2304         ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2305                         thread_listener, (void *) NULL);
2306         if (ret) {
2307                 errno = ret;
2308                 PERROR("pthread_create viewer listener");
2309                 retval = -1;
2310                 goto exit_listener_thread;
2311         }
2312
2313         /*
2314          * All OK, started all threads.
2315          */
2316         return retval;
2317
2318         /*
2319          * Join on the live_listener_thread should anything be added after
2320          * the live_listener thread's creation.
2321          */
2322
2323 exit_listener_thread:
2324
2325         ret = pthread_join(live_worker_thread, &status);
2326         if (ret) {
2327                 errno = ret;
2328                 PERROR("pthread_join live worker");
2329                 retval = -1;
2330         }
2331 exit_worker_thread:
2332
2333         ret = pthread_join(live_dispatcher_thread, &status);
2334         if (ret) {
2335                 errno = ret;
2336                 PERROR("pthread_join live dispatcher");
2337                 retval = -1;
2338         }
2339 exit_dispatcher_thread:
2340
2341 exit_init_data:
2342         cleanup_relayd_live();
2343
2344         return retval;
2345 }