src/bin/lttng-relayd/live.c

   1 /*
   2  * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
   3  *                      David Goulet <dgoulet@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _GNU_SOURCE
  20 #include <getopt.h>
  21 #include <grp.h>
  22 #include <limits.h>
  23 #include <pthread.h>
  24 #include <signal.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <sys/mman.h>
  29 #include <sys/mount.h>
  30 #include <sys/resource.h>
  31 #include <sys/socket.h>
  32 #include <sys/stat.h>
  33 #include <sys/types.h>
  34 #include <sys/wait.h>
  35 #include <inttypes.h>
  36 #include <urcu/futex.h>
  37 #include <urcu/uatomic.h>
  38 #include <unistd.h>
  39 #include <fcntl.h>
  40 #include <config.h>
  41
  42 #include <lttng/lttng.h>
  43 #include <common/common.h>
  44 #include <common/compat/poll.h>
  45 #include <common/compat/socket.h>
  46 #include <common/defaults.h>
  47 #include <common/futex.h>
  48 #include <common/sessiond-comm/sessiond-comm.h>
  49 #include <common/sessiond-comm/inet.h>
  50 #include <common/sessiond-comm/relayd.h>
  51 #include <common/uri.h>
  52 #include <common/utils.h>
  53
  54 #include "cmd.h"
  55 #include "live.h"
  56 #include "lttng-relayd.h"
  57 #include "lttng-viewer.h"
  58 #include "utils.h"
  59 #include "health-relayd.h"
  60
  61 static struct lttng_uri *live_uri;
  62
  63 /*
  64  * Quit pipe for all threads. This permits a single cancellation point
  65  * for all threads when receiving an event on the pipe.
  66  */
  67 static int live_thread_quit_pipe[2] = { -1, -1 };
  68
  69 /*
  70  * This pipe is used to inform the worker thread that a command is queued and
  71  * ready to be processed.
  72  */
  73 static int live_relay_cmd_pipe[2] = { -1, -1 };
  74
  75 /* Shared between threads */
  76 static int live_dispatch_thread_exit;
  77
  78 static pthread_t live_listener_thread;
  79 static pthread_t live_dispatcher_thread;
  80 static pthread_t live_worker_thread;
  81
  82 /*
  83  * Relay command queue.
  84  *
  85  * The live_thread_listener and live_thread_dispatcher communicate with this
  86  * queue.
  87  */
  88 static struct relay_cmd_queue viewer_cmd_queue;
  89
  90 static uint64_t last_relay_viewer_session_id;
  91
  92 /*
  93  * Cleanup the daemon
  94  */
  95 static
  96 void cleanup(void)
  97 {
  98         DBG("Cleaning up");
  99
 100         free(live_uri);
 101 }
 102
 103 /*
 104  * Write to writable pipe used to notify a thread.
 105  */
 106 static
 107 int notify_thread_pipe(int wpipe)
 108 {
 109         ssize_t ret;
 110
 111         ret = lttng_write(wpipe, "!", 1);
 112         if (ret < 1) {
 113                 PERROR("write poll pipe");
 114         }
 115
 116         return (int) ret;
 117 }
 118
 119 /*
 120  * Stop all threads by closing the thread quit pipe.
 121  */
 122 static
 123 void stop_threads(void)
 124 {
 125         int ret;
 126
 127         /* Stopping all threads */
 128         DBG("Terminating all live threads");
 129         ret = notify_thread_pipe(live_thread_quit_pipe[1]);
 130         if (ret < 0) {
 131                 ERR("write error on thread quit pipe");
 132         }
 133
 134         /* Dispatch thread */
 135         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 136         futex_nto1_wake(&viewer_cmd_queue.futex);
 137 }
 138
 139 /*
 140  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 141  */
 142 static
 143 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 144 {
 145         int ret;
 146
 147         if (events == NULL || size == 0) {
 148                 ret = -1;
 149                 goto error;
 150         }
 151
 152         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 153         if (ret < 0) {
 154                 goto error;
 155         }
 156
 157         /* Add quit pipe */
 158         ret = lttng_poll_add(events, live_thread_quit_pipe[0], LPOLLIN);
 159         if (ret < 0) {
 160                 goto error;
 161         }
 162
 163         return 0;
 164
 165 error:
 166         return ret;
 167 }
 168
 169 /*
 170  * Check if the thread quit pipe was triggered.
 171  *
 172  * Return 1 if it was triggered else 0;
 173  */
 174 static
 175 int check_thread_quit_pipe(int fd, uint32_t events)
 176 {
 177         if (fd == live_thread_quit_pipe[0] && (events & LPOLLIN)) {
 178                 return 1;
 179         }
 180
 181         return 0;
 182 }
 183
 184 /*
 185  * Create and init socket from uri.
 186  */
 187 static
 188 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 189 {
 190         int ret;
 191         struct lttcomm_sock *sock = NULL;
 192
 193         sock = lttcomm_alloc_sock_from_uri(uri);
 194         if (sock == NULL) {
 195                 ERR("Allocating socket");
 196                 goto error;
 197         }
 198
 199         ret = lttcomm_create_sock(sock);
 200         if (ret < 0) {
 201                 goto error;
 202         }
 203         DBG("Listening on sock %d for live", sock->fd);
 204
 205         ret = sock->ops->bind(sock);
 206         if (ret < 0) {
 207                 goto error;
 208         }
 209
 210         ret = sock->ops->listen(sock, -1);
 211         if (ret < 0) {
 212                 goto error;
 213
 214         }
 215
 216         return sock;
 217
 218 error:
 219         if (sock) {
 220                 lttcomm_destroy_sock(sock);
 221         }
 222         return NULL;
 223 }
 224
 225 /*
 226  * This thread manages the listening for new connections on the network
 227  */
 228 static
 229 void *thread_listener(void *data)
 230 {
 231         int i, ret, pollfd, err = -1;
 232         int val = 1;
 233         uint32_t revents, nb_fd;
 234         struct lttng_poll_event events;
 235         struct lttcomm_sock *live_control_sock;
 236
 237         DBG("[thread] Relay live listener started");
 238
 239         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 240
 241         health_code_update();
 242
 243         live_control_sock = init_socket(live_uri);
 244         if (!live_control_sock) {
 245                 goto error_sock_control;
 246         }
 247
 248         /*
 249          * Pass 3 as size here for the thread quit pipe, control and data socket.
 250          */
 251         ret = create_thread_poll_set(&events, 2);
 252         if (ret < 0) {
 253                 goto error_create_poll;
 254         }
 255
 256         /* Add the control socket */
 257         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 258         if (ret < 0) {
 259                 goto error_poll_add;
 260         }
 261
 262         while (1) {
 263                 health_code_update();
 264
 265                 DBG("Listener accepting live viewers connections");
 266
 267 restart:
 268                 health_poll_entry();
 269                 ret = lttng_poll_wait(&events, -1);
 270                 health_poll_exit();
 271                 if (ret < 0) {
 272                         /*
 273                          * Restart interrupted system call.
 274                          */
 275                         if (errno == EINTR) {
 276                                 goto restart;
 277                         }
 278                         goto error;
 279                 }
 280                 nb_fd = ret;
 281
 282                 DBG("Relay new viewer connection received");
 283                 for (i = 0; i < nb_fd; i++) {
 284                         health_code_update();
 285
 286                         /* Fetch once the poll data */
 287                         revents = LTTNG_POLL_GETEV(&events, i);
 288                         pollfd = LTTNG_POLL_GETFD(&events, i);
 289
 290                         /* Thread quit pipe has been closed. Killing thread. */
 291                         ret = check_thread_quit_pipe(pollfd, revents);
 292                         if (ret) {
 293                                 err = 0;
 294                                 goto exit;
 295                         }
 296
 297                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 298                                 ERR("socket poll error");
 299                                 goto error;
 300                         } else if (revents & LPOLLIN) {
 301                                 /*
 302                                  * Get allocated in this thread, enqueued to a global queue,
 303                                  * dequeued and freed in the worker thread.
 304                                  */
 305                                 struct relay_command *relay_cmd;
 306                                 struct lttcomm_sock *newsock;
 307
 308                                 relay_cmd = zmalloc(sizeof(*relay_cmd));
 309                                 if (!relay_cmd) {
 310                                         PERROR("relay command zmalloc");
 311                                         goto error;
 312                                 }
 313
 314                                 assert(pollfd == live_control_sock->fd);
 315                                 newsock = live_control_sock->ops->accept(live_control_sock);
 316                                 if (!newsock) {
 317                                         PERROR("accepting control sock");
 318                                         free(relay_cmd);
 319                                         goto error;
 320                                 }
 321                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 322                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 323                                                 sizeof(int));
 324                                 if (ret < 0) {
 325                                         PERROR("setsockopt inet");
 326                                         lttcomm_destroy_sock(newsock);
 327                                         free(relay_cmd);
 328                                         goto error;
 329                                 }
 330                                 relay_cmd->sock = newsock;
 331
 332                                 /*
 333                                  * Lock free enqueue the request.
 334                                  */
 335                                 cds_wfq_enqueue(&viewer_cmd_queue.queue, &relay_cmd->node);
 336
 337                                 /*
 338                                  * Wake the dispatch queue futex. Implicit memory
 339                                  * barrier with the exchange in cds_wfq_enqueue.
 340                                  */
 341                                 futex_nto1_wake(&viewer_cmd_queue.futex);
 342                         }
 343                 }
 344         }
 345
 346 exit:
 347 error:
 348 error_poll_add:
 349         lttng_poll_clean(&events);
 350 error_create_poll:
 351         if (live_control_sock->fd >= 0) {
 352                 ret = live_control_sock->ops->close(live_control_sock);
 353                 if (ret) {
 354                         PERROR("close");
 355                 }
 356         }
 357         lttcomm_destroy_sock(live_control_sock);
 358 error_sock_control:
 359         if (err) {
 360                 health_error();
 361                 DBG("Live viewer listener thread exited with error");
 362         }
 363         health_unregister(health_relayd);
 364         DBG("Live viewer listener thread cleanup complete");
 365         stop_threads();
 366         return NULL;
 367 }
 368
 369 /*
 370  * This thread manages the dispatching of the requests to worker threads
 371  */
 372 static
 373 void *thread_dispatcher(void *data)
 374 {
 375         int err = -1;
 376         ssize_t ret;
 377         struct cds_wfq_node *node;
 378         struct relay_command *relay_cmd = NULL;
 379
 380         DBG("[thread] Live viewer relay dispatcher started");
 381
 382         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 383
 384         health_code_update();
 385
 386         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 387                 health_code_update();
 388
 389                 /* Atomically prepare the queue futex */
 390                 futex_nto1_prepare(&viewer_cmd_queue.futex);
 391
 392                 do {
 393                         health_code_update();
 394
 395                         /* Dequeue commands */
 396                         node = cds_wfq_dequeue_blocking(&viewer_cmd_queue.queue);
 397                         if (node == NULL) {
 398                                 DBG("Woken up but nothing in the live-viewer "
 399                                                 "relay command queue");
 400                                 /* Continue thread execution */
 401                                 break;
 402                         }
 403
 404                         relay_cmd = caa_container_of(node, struct relay_command, node);
 405                         DBG("Dispatching viewer request waiting on sock %d",
 406                                         relay_cmd->sock->fd);
 407
 408                         /*
 409                          * Inform worker thread of the new request. This call is blocking
 410                          * so we can be assured that the data will be read at some point in
 411                          * time or wait to the end of the world :)
 412                          */
 413                         ret = lttng_write(live_relay_cmd_pipe[1], relay_cmd,
 414                                         sizeof(*relay_cmd));
 415                         free(relay_cmd);
 416                         if (ret < sizeof(struct relay_command)) {
 417                                 PERROR("write cmd pipe");
 418                                 goto error;
 419                         }
 420                 } while (node != NULL);
 421
 422                 /* Futex wait on queue. Blocking call on futex() */
 423                 health_poll_entry();
 424                 futex_nto1_wait(&viewer_cmd_queue.futex);
 425                 health_poll_exit();
 426         }
 427
 428         /* Normal exit, no error */
 429         err = 0;
 430
 431 error:
 432         if (err) {
 433                 health_error();
 434                 ERR("Health error occurred in %s", __func__);
 435         }
 436         health_unregister(health_relayd);
 437         DBG("Live viewer dispatch thread dying");
 438         stop_threads();
 439         return NULL;
 440 }
 441
 442 /*
 443  * Establish connection with the viewer and check the versions.
 444  *
 445  * Return 0 on success or else negative value.
 446  */
 447 static
 448 int viewer_connect(struct relay_command *cmd)
 449 {
 450         int ret;
 451         struct lttng_viewer_connect reply, msg;
 452
 453         assert(cmd);
 454
 455         cmd->version_check_done = 1;
 456
 457         health_code_update();
 458
 459         /* Get version from the other side. */
 460         ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
 461         if (ret < 0 || ret != sizeof(msg)) {
 462                 if (ret == 0) {
 463                         /* Orderly shutdown. Not necessary to print an error. */
 464                         DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
 465                 } else {
 466                         ERR("Relay failed to receive the version values.");
 467                 }
 468                 ret = -1;
 469                 goto end;
 470         }
 471
 472         health_code_update();
 473
 474         reply.major = RELAYD_VERSION_COMM_MAJOR;
 475         reply.minor = RELAYD_VERSION_COMM_MINOR;
 476
 477         /* Major versions must be the same */
 478         if (reply.major != be32toh(msg.major)) {
 479                 DBG("Incompatible major versions (%u vs %u)", reply.major,
 480                                 be32toh(msg.major));
 481                 ret = 0;
 482                 goto end;
 483         }
 484
 485         cmd->major = reply.major;
 486         /* We adapt to the lowest compatible version */
 487         if (reply.minor <= be32toh(msg.minor)) {
 488                 cmd->minor = reply.minor;
 489         } else {
 490                 cmd->minor = be32toh(msg.minor);
 491         }
 492
 493         if (be32toh(msg.type) == VIEWER_CLIENT_COMMAND) {
 494                 cmd->type = RELAY_VIEWER_COMMAND;
 495         } else if (be32toh(msg.type) == VIEWER_CLIENT_NOTIFICATION) {
 496                 cmd->type = RELAY_VIEWER_NOTIFICATION;
 497         } else {
 498                 ERR("Unknown connection type : %u", be32toh(msg.type));
 499                 ret = -1;
 500                 goto end;
 501         }
 502
 503         reply.major = htobe32(reply.major);
 504         reply.minor = htobe32(reply.minor);
 505         if (cmd->type == RELAY_VIEWER_COMMAND) {
 506                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 507         }
 508
 509         health_code_update();
 510
 511         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
 512                         sizeof(struct lttng_viewer_connect), 0);
 513         if (ret < 0) {
 514                 ERR("Relay sending version");
 515         }
 516
 517         health_code_update();
 518
 519         DBG("Version check done using protocol %u.%u", cmd->major, cmd->minor);
 520         ret = 0;
 521
 522 end:
 523         return ret;
 524 }
 525
 526 /*
 527  * Send the viewer the list of current sessions.
 528  *
 529  * Return 0 on success or else a negative value.
 530  */
 531 static
 532 int viewer_list_sessions(struct relay_command *cmd,
 533                 struct lttng_ht *sessions_ht)
 534 {
 535         int ret;
 536         struct lttng_viewer_list_sessions session_list;
 537         unsigned long count;
 538         long approx_before, approx_after;
 539         struct lttng_ht_node_ulong *node;
 540         struct lttng_ht_iter iter;
 541         struct lttng_viewer_session send_session;
 542         struct relay_session *session;
 543
 544         DBG("List sessions received");
 545
 546         if (cmd->version_check_done == 0) {
 547                 ERR("Trying to list sessions before version check");
 548                 ret = -1;
 549                 goto end_no_session;
 550         }
 551
 552         rcu_read_lock();
 553         cds_lfht_count_nodes(sessions_ht->ht, &approx_before, &count, &approx_after);
 554         session_list.sessions_count = htobe32(count);
 555
 556         health_code_update();
 557
 558         ret = cmd->sock->ops->sendmsg(cmd->sock, &session_list,
 559                         sizeof(session_list), 0);
 560         if (ret < 0) {
 561                 ERR("Relay sending sessions list");
 562                 goto end_unlock;
 563         }
 564
 565         health_code_update();
 566
 567         cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, node, node) {
 568                 health_code_update();
 569
 570                 node = lttng_ht_iter_get_node_ulong(&iter);
 571                 if (!node) {
 572                         goto end_unlock;
 573                 }
 574                 session = caa_container_of(node, struct relay_session, session_n);
 575
 576                 strncpy(send_session.session_name, session->session_name,
 577                                 sizeof(send_session.session_name));
 578                 strncpy(send_session.hostname, session->hostname,
 579                                 sizeof(send_session.hostname));
 580                 send_session.id = htobe64(session->id);
 581                 send_session.live_timer = htobe32(session->live_timer);
 582                 send_session.clients = htobe32(session->viewer_attached);
 583                 send_session.streams = htobe32(session->stream_count);
 584
 585                 health_code_update();
 586
 587                 ret = cmd->sock->ops->sendmsg(cmd->sock, &send_session,
 588                                 sizeof(send_session), 0);
 589                 if (ret < 0) {
 590                         ERR("Relay sending session info");
 591                         goto end_unlock;
 592                 }
 593         }
 594         health_code_update();
 595
 596         rcu_read_unlock();
 597         ret = 0;
 598         goto end;
 599
 600 end_unlock:
 601         rcu_read_unlock();
 602
 603 end:
 604 end_no_session:
 605         return ret;
 606 }
 607
 608 /*
 609  * Open index file using a given viewer stream.
 610  *
 611  * Return 0 on success or else a negative value.
 612  */
 613 static int open_index(struct relay_viewer_stream *stream)
 614 {
 615         int ret;
 616         char fullpath[PATH_MAX];
 617         struct lttng_packet_index_file_hdr hdr;
 618
 619         if (stream->tracefile_count > 0) {
 620                 ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s_%"
 621                                 PRIu64 DEFAULT_INDEX_FILE_SUFFIX, stream->path_name,
 622                                 stream->channel_name, stream->tracefile_count_current);
 623         } else {
 624                 ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s"
 625                                 DEFAULT_INDEX_FILE_SUFFIX, stream->path_name,
 626                                 stream->channel_name);
 627         }
 628         if (ret < 0) {
 629                 PERROR("snprintf index path");
 630                 goto error;
 631         }
 632
 633         DBG("Opening index file %s in read only", fullpath);
 634         ret = open(fullpath, O_RDONLY);
 635         if (ret < 0) {
 636                 if (errno == ENOENT) {
 637                         ret = -ENOENT;
 638                         goto error;
 639                 } else {
 640                         PERROR("opening index in read-only");
 641                 }
 642                 goto error;
 643         }
 644         stream->index_read_fd = ret;
 645         DBG("Opening index file %s in read only, (fd: %d)", fullpath, ret);
 646
 647         ret = lttng_read(stream->index_read_fd, &hdr, sizeof(hdr));
 648         if (ret < sizeof(hdr)) {
 649                 PERROR("Reading index header");
 650                 goto error;
 651         }
 652         if (strncmp(hdr.magic, INDEX_MAGIC, sizeof(hdr.magic)) != 0) {
 653                 ERR("Invalid header magic");
 654                 ret = -1;
 655                 goto error;
 656         }
 657         if (be32toh(hdr.index_major) != INDEX_MAJOR ||
 658                         be32toh(hdr.index_minor) != INDEX_MINOR) {
 659                 ERR("Invalid header version");
 660                 ret = -1;
 661                 goto error;
 662         }
 663         ret = 0;
 664
 665 error:
 666         return ret;
 667 }
 668
 669 /*
 670  * Allocate and init a new viewer_stream.
 671  *
 672  * Copies the values from the stream passed in parameter and insert the new
 673  * stream in the viewer_streams_ht.
 674  *
 675  * MUST be called with rcu_read_lock held.
 676  *
 677  * Returns 0 on success or a negative value on error.
 678  */
 679 static
 680 int init_viewer_stream(struct relay_stream *stream, int seek_last)
 681 {
 682         int ret;
 683         struct relay_viewer_stream *viewer_stream;
 684
 685         assert(stream);
 686
 687         viewer_stream = zmalloc(sizeof(*viewer_stream));
 688         if (!viewer_stream) {
 689                 PERROR("relay viewer stream zmalloc");
 690                 ret = -1;
 691                 goto error;
 692         }
 693         viewer_stream->session_id = stream->session->id;
 694         viewer_stream->stream_handle = stream->stream_handle;
 695         viewer_stream->path_name = strndup(stream->path_name,
 696                         LTTNG_VIEWER_PATH_MAX);
 697         viewer_stream->channel_name = strndup(stream->channel_name,
 698                         LTTNG_VIEWER_NAME_MAX);
 699         viewer_stream->tracefile_count = stream->tracefile_count;
 700         viewer_stream->metadata_flag = stream->metadata_flag;
 701         if (seek_last) {
 702                 viewer_stream->tracefile_count_current =
 703                         stream->tracefile_count_current;
 704         } else {
 705                 viewer_stream->tracefile_count_current =
 706                         stream->oldest_tracefile_id;
 707         }
 708
 709         /*
 710          * The deletion of this ctf_trace object is only done in a call RCU of the
 711          * relay stream making it valid as long as we have the read side lock.
 712          */
 713         viewer_stream->ctf_trace = stream->ctf_trace;
 714         uatomic_inc(&viewer_stream->ctf_trace->refcount);
 715
 716         lttng_ht_node_init_u64(&viewer_stream->stream_n, stream->stream_handle);
 717         lttng_ht_add_unique_u64(viewer_streams_ht, &viewer_stream->stream_n);
 718
 719         viewer_stream->index_read_fd = -1;
 720         viewer_stream->read_fd = -1;
 721
 722         /*
 723          * This is to avoid a race between the initialization of this object and
 724          * the close of the given stream. If the stream is unable to find this
 725          * viewer stream when closing, this copy will at least take the latest
 726          * value.
 727          * We also need that for the seek_last.
 728          */
 729         viewer_stream->total_index_received = stream->total_index_received;
 730
 731         /*
 732          * If we never received an index for the current stream, delay
 733          * the opening of the index, otherwise open it right now.
 734          */
 735         if (viewer_stream->tracefile_count_current ==
 736                         stream->tracefile_count_current &&
 737                         viewer_stream->total_index_received == 0) {
 738                 viewer_stream->index_read_fd = -1;
 739         } else {
 740                 ret = open_index(viewer_stream);
 741                 if (ret < 0) {
 742                         goto error;
 743                 }
 744         }
 745
 746         if (seek_last && viewer_stream->index_read_fd > 0) {
 747                 ret = lseek(viewer_stream->index_read_fd,
 748                                 viewer_stream->total_index_received *
 749                                         sizeof(struct lttng_packet_index),
 750                                 SEEK_CUR);
 751                 if (ret < 0) {
 752                         goto error;
 753                 }
 754                 viewer_stream->last_sent_index =
 755                         viewer_stream->total_index_received;
 756         }
 757
 758         ret = 0;
 759
 760 error:
 761         return ret;
 762 }
 763
 764 /*
 765  * Rotate a stream to the next tracefile.
 766  *
 767  * Returns 0 on success, a negative value on error.
 768  */
 769 static
 770 int rotate_viewer_stream(struct relay_viewer_stream *viewer_stream,
 771                 struct relay_stream *stream)
 772 {
 773         int ret;
 774         uint64_t tracefile_id;
 775
 776         assert(viewer_stream);
 777
 778         tracefile_id = (viewer_stream->tracefile_count_current + 1) %
 779                 viewer_stream->tracefile_count;
 780
 781         if (stream) {
 782                 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
 783         }
 784         /*
 785          * The writer and the reader are not working in the same
 786          * tracefile, we can read up to EOF, we don't care about the
 787          * total_index_received.
 788          */
 789         if (!stream || (stream->tracefile_count_current != tracefile_id)) {
 790                 viewer_stream->close_write_flag = 1;
 791         } else {
 792                 /*
 793                  * We are opening a file that is still open in write, make
 794                  * sure we limit our reading to the number of indexes
 795                  * received.
 796                  */
 797                 viewer_stream->close_write_flag = 0;
 798                 if (stream) {
 799                         viewer_stream->total_index_received =
 800                                 stream->total_index_received;
 801                 }
 802         }
 803         viewer_stream->tracefile_count_current = tracefile_id;
 804
 805         ret = close(viewer_stream->index_read_fd);
 806         if (ret < 0) {
 807                 PERROR("close index file %d",
 808                                 viewer_stream->index_read_fd);
 809         }
 810         viewer_stream->index_read_fd = -1;
 811         ret = close(viewer_stream->read_fd);
 812         if (ret < 0) {
 813                 PERROR("close tracefile %d",
 814                                 viewer_stream->read_fd);
 815         }
 816         viewer_stream->read_fd = -1;
 817
 818         pthread_mutex_lock(&viewer_stream->overwrite_lock);
 819         viewer_stream->abort_flag = 0;
 820         pthread_mutex_unlock(&viewer_stream->overwrite_lock);
 821
 822         viewer_stream->index_read_fd = -1;
 823         viewer_stream->read_fd = -1;
 824
 825         if (stream) {
 826                 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
 827         }
 828         ret = open_index(viewer_stream);
 829         if (ret < 0) {
 830                 goto error;
 831         }
 832
 833         ret = 0;
 834
 835 error:
 836         return ret;
 837 }
 838
 839 /*
 840  * Send the viewer the list of current sessions.
 841  */
 842 static
 843 int viewer_attach_session(struct relay_command *cmd,
 844                 struct lttng_ht *sessions_ht)
 845 {
 846         int ret, send_streams = 0, nb_streams = 0;
 847         struct lttng_viewer_attach_session_request request;
 848         struct lttng_viewer_attach_session_response response;
 849         struct lttng_viewer_stream send_stream;
 850         struct relay_stream *stream;
 851         struct relay_viewer_stream *viewer_stream;
 852         struct lttng_ht_node_ulong *node;
 853         struct lttng_ht_node_u64 *node64;
 854         struct lttng_ht_iter iter;
 855         struct relay_session *session;
 856         int seek_last = 0;
 857
 858         assert(cmd);
 859         assert(sessions_ht);
 860
 861         DBG("Attach session received");
 862
 863         if (cmd->version_check_done == 0) {
 864                 ERR("Trying to attach session before version check");
 865                 ret = -1;
 866                 goto end_no_session;
 867         }
 868
 869         health_code_update();
 870
 871         ret = cmd->sock->ops->recvmsg(cmd->sock, &request, sizeof(request), 0);
 872         if (ret < 0 || ret != sizeof(request)) {
 873                 if (ret == 0) {
 874                         /* Orderly shutdown. Not necessary to print an error. */
 875                         DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
 876                 } else {
 877                         ERR("Relay failed to receive the attach parameters.");
 878                 }
 879                 ret = -1;
 880                 goto error;
 881         }
 882
 883         health_code_update();
 884
 885         rcu_read_lock();
 886         lttng_ht_lookup(sessions_ht,
 887                         (void *)((unsigned long) be64toh(request.session_id)), &iter);
 888         node = lttng_ht_iter_get_node_ulong(&iter);
 889         if (node == NULL) {
 890                 DBG("Relay session %" PRIu64 " not found",
 891                                 be64toh(request.session_id));
 892                 response.status = htobe32(VIEWER_ATTACH_UNK);
 893                 goto send_reply;
 894         }
 895
 896         session = caa_container_of(node, struct relay_session, session_n);
 897         if (cmd->session_id == session->id) {
 898                 /* Same viewer already attached, just send the stream list. */
 899                 send_streams = 1;
 900                 response.status = htobe32(VIEWER_ATTACH_OK);
 901         } else if (session->viewer_attached != 0) {
 902                 DBG("Already a viewer attached");
 903                 response.status = htobe32(VIEWER_ATTACH_ALREADY);
 904                 goto send_reply;
 905         } else if (session->live_timer == 0) {
 906                 DBG("Not live session");
 907                 response.status = htobe32(VIEWER_ATTACH_NOT_LIVE);
 908                 goto send_reply;
 909         } else {
 910                 session->viewer_attached++;
 911                 send_streams = 1;
 912                 response.status = htobe32(VIEWER_ATTACH_OK);
 913                 cmd->session_id = session->id;
 914                 cmd->session = session;
 915         }
 916
 917         switch (be32toh(request.seek)) {
 918         case VIEWER_SEEK_BEGINNING:
 919                 /* Default behaviour. */
 920                 break;
 921         case VIEWER_SEEK_LAST:
 922                 seek_last = 1;
 923                 break;
 924         default:
 925                 ERR("Wrong seek parameter");
 926                 response.status = htobe32(VIEWER_ATTACH_SEEK_ERR);
 927                 send_streams = 0;
 928                 goto send_reply;
 929         }
 930
 931         if (send_streams) {
 932                 /* We should only be there if we have a session to attach to. */
 933                 assert(session);
 934
 935                 /*
 936                  * Fill the viewer_streams_ht to count the number of streams
 937                  * ready to be sent and avoid concurrency issues on the
 938                  * relay_streams_ht and don't rely on a total session stream count.
 939                  */
 940                 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
 941                         struct relay_viewer_stream *vstream;
 942
 943                         health_code_update();
 944
 945                         node = lttng_ht_iter_get_node_ulong(&iter);
 946                         if (!node) {
 947                                 continue;
 948                         }
 949                         stream = caa_container_of(node, struct relay_stream, stream_n);
 950                         if (stream->session != cmd->session) {
 951                                 continue;
 952                         }
 953
 954                         /*
 955                          * Don't send streams with no ctf_trace, they are not ready to be
 956                          * read.
 957                          */
 958                         if (!stream->ctf_trace) {
 959                                 continue;
 960                         }
 961
 962                         vstream = live_find_viewer_stream_by_id(stream->stream_handle);
 963                         if (!vstream) {
 964                                 ret = init_viewer_stream(stream, seek_last);
 965                                 if (ret < 0) {
 966                                         goto end_unlock;
 967                                 }
 968                         }
 969                         nb_streams++;
 970                 }
 971                 response.streams_count = htobe32(nb_streams);
 972         }
 973
 974 send_reply:
 975         health_code_update();
 976         ret = cmd->sock->ops->sendmsg(cmd->sock, &response, sizeof(response), 0);
 977         if (ret < 0) {
 978                 ERR("Relay sending viewer attach response");
 979                 goto end_unlock;
 980         }
 981         health_code_update();
 982
 983         /*
 984          * Unknown or busy session, just return gracefully, the viewer knows what
 985          * is happening.
 986          */
 987         if (!send_streams) {
 988                 ret = 0;
 989                 goto end_unlock;
 990         }
 991
 992         /* We should only be there if we have a session to attach to. */
 993         assert(session);
 994         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
 995                 health_code_update();
 996
 997                 node64 = lttng_ht_iter_get_node_u64(&iter);
 998                 if (!node64) {
 999                         continue;
1000                 }
1001                 viewer_stream = caa_container_of(node64, struct relay_viewer_stream,
1002                                 stream_n);
1003                 if (viewer_stream->session_id != cmd->session->id) {
1004                         continue;
1005                 }
1006
1007                 send_stream.id = htobe64(viewer_stream->stream_handle);
1008                 send_stream.ctf_trace_id = htobe64(viewer_stream->ctf_trace->id);
1009                 send_stream.metadata_flag = htobe32(viewer_stream->metadata_flag);
1010                 strncpy(send_stream.path_name, viewer_stream->path_name,
1011                                 sizeof(send_stream.path_name));
1012                 strncpy(send_stream.channel_name, viewer_stream->channel_name,
1013                                 sizeof(send_stream.channel_name));
1014
1015                 ret = cmd->sock->ops->sendmsg(cmd->sock, &send_stream,
1016                                 sizeof(send_stream), 0);
1017                 if (ret < 0) {
1018                         ERR("Relay sending stream %" PRIu64, viewer_stream->stream_handle);
1019                         goto end_unlock;
1020                 }
1021                 DBG("Sent stream %" PRIu64 " to viewer", viewer_stream->stream_handle);
1022         }
1023         ret = 0;
1024
1025 end_unlock:
1026         rcu_read_unlock();
1027 end_no_session:
1028 error:
1029         return ret;
1030 }
1031
1032 /*
1033  * Get viewer stream from stream id.
1034  *
1035  * RCU read side lock MUST be acquired.
1036  */
1037 struct relay_viewer_stream *live_find_viewer_stream_by_id(uint64_t stream_id)
1038 {
1039         struct lttng_ht_node_u64 *node;
1040         struct lttng_ht_iter iter;
1041         struct relay_viewer_stream *stream = NULL;
1042
1043         lttng_ht_lookup(viewer_streams_ht, &stream_id, &iter);
1044         node = lttng_ht_iter_get_node_u64(&iter);
1045         if (node == NULL) {
1046                 DBG("Relay viewer stream %" PRIu64 " not found", stream_id);
1047                 goto end;
1048         }
1049         stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
1050
1051 end:
1052         return stream;
1053 }
1054
1055 /*
1056  * Send the next index for a stream.
1057  *
1058  * Return 0 on success or else a negative value.
1059  */
1060 static
1061 int viewer_get_next_index(struct relay_command *cmd,
1062                 struct lttng_ht *sessions_ht)
1063 {
1064         int ret;
1065         struct lttng_viewer_get_next_index request_index;
1066         struct lttng_viewer_index viewer_index;
1067         struct lttng_packet_index packet_index;
1068         struct relay_viewer_stream *vstream;
1069         struct relay_stream *rstream;
1070
1071         assert(cmd);
1072         assert(sessions_ht);
1073
1074         DBG("Viewer get next index");
1075
1076         if (cmd->version_check_done == 0) {
1077                 ERR("Trying to request index before version check");
1078                 ret = -1;
1079                 goto end_no_session;
1080         }
1081
1082         health_code_update();
1083         ret = cmd->sock->ops->recvmsg(cmd->sock, &request_index,
1084                         sizeof(request_index), 0);
1085         if (ret < 0 || ret != sizeof(request_index)) {
1086                 ret = -1;
1087                 ERR("Relay didn't receive the whole packet");
1088                 goto end;
1089         }
1090         health_code_update();
1091
1092         rcu_read_lock();
1093         vstream = live_find_viewer_stream_by_id(be64toh(request_index.stream_id));
1094         if (!vstream) {
1095                 ret = -1;
1096                 goto end_unlock;
1097         }
1098
1099         memset(&viewer_index, 0, sizeof(viewer_index));
1100
1101         /*
1102          * The viewer should not ask for index on metadata stream.
1103          */
1104         if (vstream->metadata_flag) {
1105                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
1106                 goto send_reply;
1107         }
1108
1109         /* First time, we open the index file */
1110         if (vstream->index_read_fd < 0) {
1111                 ret = open_index(vstream);
1112                 if (ret == -ENOENT) {
1113                         /*
1114                          * The index is created only when the first data packet arrives, it
1115                          * might not be ready at the beginning of the session
1116                          */
1117                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1118                         goto send_reply;
1119                 } else if (ret < 0) {
1120                         viewer_index.status = htobe32(VIEWER_INDEX_ERR);
1121                         goto send_reply;
1122                 }
1123         }
1124
1125         rstream = relay_stream_find_by_id(vstream->stream_handle);
1126         if (rstream) {
1127                 if (vstream->abort_flag) {
1128                         /* Rotate on abort (overwrite). */
1129                         DBG("Viewer rotate because of overwrite");
1130                         ret = rotate_viewer_stream(vstream, rstream);
1131                         if (ret < 0) {
1132                                 goto end_unlock;
1133                         }
1134                 }
1135                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1136                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1137                         if (rstream->beacon_ts_end != -1ULL &&
1138                                 vstream->last_sent_index == rstream->total_index_received) {
1139                                 viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE);
1140                                 viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
1141                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1142                                 goto send_reply;
1143                         /*
1144                          * Reader and writer are working in the same tracefile, so we care
1145                          * about the number of index received and sent. Otherwise, we read
1146                          * up to EOF.
1147                          */
1148                         } else if (rstream->total_index_received <= vstream->last_sent_index
1149                                         && !vstream->close_write_flag) {
1150                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1151                                 /* No new index to send, retry later. */
1152                                 viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1153                                 goto send_reply;
1154                         }
1155                 }
1156                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1157         } else if (!rstream && vstream->close_write_flag &&
1158                         vstream->total_index_received == vstream->last_sent_index) {
1159                 /* Last index sent and current tracefile closed in write */
1160                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
1161                 goto send_reply;
1162         } else {
1163                 vstream->close_write_flag = 1;
1164         }
1165
1166         if (!vstream->ctf_trace->metadata_received ||
1167                         vstream->ctf_trace->metadata_received >
1168                         vstream->ctf_trace->metadata_sent) {
1169                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1170         }
1171
1172         pthread_mutex_lock(&vstream->overwrite_lock);
1173         if (vstream->abort_flag) {
1174                 /*
1175                  * The file is being overwritten by the writer, we cannot
1176                  * use it.
1177                  */
1178                 viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1179                 pthread_mutex_unlock(&vstream->overwrite_lock);
1180                 ret = rotate_viewer_stream(vstream, rstream);
1181                 if (ret < 0) {
1182                         goto end_unlock;
1183                 }
1184                 goto send_reply;
1185         }
1186         ret = lttng_read(vstream->index_read_fd, &packet_index,
1187                         sizeof(packet_index));
1188         pthread_mutex_unlock(&vstream->overwrite_lock);
1189         if (ret < sizeof(packet_index)) {
1190                 /*
1191                  * The tracefile is closed in write, so we read up to EOF.
1192                  */
1193                 if (vstream->close_write_flag == 1) {
1194                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1195                         /* Rotate on normal EOF */
1196                         ret = rotate_viewer_stream(vstream, rstream);
1197                         if (ret < 0) {
1198                                 goto end_unlock;
1199                         }
1200                 } else {
1201                         PERROR("Relay reading index file %d",
1202                                         vstream->index_read_fd);
1203                         viewer_index.status = htobe32(VIEWER_INDEX_ERR);
1204                 }
1205                 goto send_reply;
1206         } else {
1207                 viewer_index.status = htobe32(VIEWER_INDEX_OK);
1208                 vstream->last_sent_index++;
1209         }
1210
1211         /*
1212          * Indexes are stored in big endian, no need to switch before sending.
1213          */
1214         viewer_index.offset = packet_index.offset;
1215         viewer_index.packet_size = packet_index.packet_size;
1216         viewer_index.content_size = packet_index.content_size;
1217         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1218         viewer_index.timestamp_end = packet_index.timestamp_end;
1219         viewer_index.events_discarded = packet_index.events_discarded;
1220         viewer_index.stream_id = packet_index.stream_id;
1221
1222 send_reply:
1223         viewer_index.flags = htobe32(viewer_index.flags);
1224         health_code_update();
1225         ret = cmd->sock->ops->sendmsg(cmd->sock, &viewer_index,
1226                         sizeof(viewer_index), 0);
1227         if (ret < 0) {
1228                 ERR("Relay index to viewer");
1229                 goto end_unlock;
1230         }
1231         health_code_update();
1232
1233         DBG("Index %" PRIu64 "for stream %" PRIu64 "sent",
1234                         vstream->last_sent_index, vstream->stream_handle);
1235
1236 end_unlock:
1237         rcu_read_unlock();
1238
1239 end_no_session:
1240 end:
1241         return ret;
1242 }
1243
1244 /*
1245  * Send the next index for a stream
1246  *
1247  * Return 0 on success or else a negative value.
1248  */
1249 static
1250 int viewer_get_packet(struct relay_command *cmd)
1251 {
1252         int ret, send_data = 0;
1253         char *data = NULL;
1254         uint32_t len = 0;
1255         ssize_t read_len;
1256         struct lttng_viewer_get_packet get_packet_info;
1257         struct lttng_viewer_trace_packet reply;
1258         struct relay_viewer_stream *stream;
1259
1260         assert(cmd);
1261
1262         DBG2("Relay get data packet");
1263
1264         if (cmd->version_check_done == 0) {
1265                 ERR("Trying to get packet before version check");
1266                 ret = -1;
1267                 goto end;
1268         }
1269
1270         health_code_update();
1271         ret = cmd->sock->ops->recvmsg(cmd->sock, &get_packet_info,
1272                         sizeof(get_packet_info), 0);
1273         if (ret < 0 || ret != sizeof(get_packet_info)) {
1274                 ret = -1;
1275                 ERR("Relay didn't receive the whole packet");
1276                 goto end;
1277         }
1278         health_code_update();
1279
1280         /* From this point on, the error label can be reached. */
1281         memset(&reply, 0, sizeof(reply));
1282
1283         rcu_read_lock();
1284         stream = live_find_viewer_stream_by_id(be64toh(get_packet_info.stream_id));
1285         if (!stream) {
1286                 goto error;
1287         }
1288         assert(stream->ctf_trace);
1289
1290         /*
1291          * First time we read this stream, we need open the tracefile, we should
1292          * only arrive here if an index has already been sent to the viewer, so the
1293          * tracefile must exist, if it does not it is a fatal error.
1294          */
1295         if (stream->read_fd < 0) {
1296                 char fullpath[PATH_MAX];
1297
1298                 if (stream->tracefile_count > 0) {
1299                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1300                                         stream->channel_name,
1301                                         stream->tracefile_count_current);
1302                 } else {
1303                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1304                                         stream->channel_name);
1305                 }
1306                 if (ret < 0) {
1307                         goto error;
1308                 }
1309                 ret = open(fullpath, O_RDONLY);
1310                 if (ret < 0) {
1311                         PERROR("Relay opening trace file");
1312                         goto error;
1313                 }
1314                 stream->read_fd = ret;
1315         }
1316
1317         if (!stream->ctf_trace->metadata_received ||
1318                         stream->ctf_trace->metadata_received >
1319                         stream->ctf_trace->metadata_sent) {
1320                 reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1321                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1322                 goto send_reply;
1323         }
1324
1325         len = be32toh(get_packet_info.len);
1326         data = zmalloc(len);
1327         if (!data) {
1328                 PERROR("relay data zmalloc");
1329                 goto error;
1330         }
1331
1332         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1333         if (ret < 0) {
1334                 /*
1335                  * If the read fd was closed by the streaming side, the
1336                  * abort_flag will be set to 1, otherwise it is an error.
1337                  */
1338                 if (stream->abort_flag == 0) {
1339                         PERROR("lseek");
1340                         goto error;
1341                 }
1342                 reply.status = htobe32(VIEWER_GET_PACKET_EOF);
1343                 goto send_reply;
1344         }
1345         read_len = lttng_read(stream->read_fd, data, len);
1346         if (read_len < len) {
1347                 /*
1348                  * If the read fd was closed by the streaming side, the
1349                  * abort_flag will be set to 1, otherwise it is an error.
1350                  */
1351                 if (stream->abort_flag == 0) {
1352                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1353                                         stream->read_fd,
1354                                         be64toh(get_packet_info.offset));
1355                         goto error;
1356                 } else {
1357                         reply.status = htobe32(VIEWER_GET_PACKET_EOF);
1358                         goto send_reply;
1359                 }
1360         }
1361         reply.status = htobe32(VIEWER_GET_PACKET_OK);
1362         reply.len = htobe32(len);
1363         send_data = 1;
1364         goto send_reply;
1365
1366 error:
1367         reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1368
1369 send_reply:
1370         reply.flags = htobe32(reply.flags);
1371
1372         health_code_update();
1373         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1374         if (ret < 0) {
1375                 ERR("Relay data header to viewer");
1376                 goto end_unlock;
1377         }
1378         health_code_update();
1379
1380         if (send_data) {
1381                 health_code_update();
1382                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1383                 if (ret < 0) {
1384                         ERR("Relay send data to viewer");
1385                         goto end_unlock;
1386                 }
1387                 health_code_update();
1388         }
1389
1390         DBG("Sent %u bytes for stream %" PRIu64, len,
1391                         be64toh(get_packet_info.stream_id));
1392
1393 end_unlock:
1394         free(data);
1395         rcu_read_unlock();
1396
1397 end:
1398         return ret;
1399 }
1400
1401 /*
1402  * Send the session's metadata
1403  *
1404  * Return 0 on success else a negative value.
1405  */
1406 static
1407 int viewer_get_metadata(struct relay_command *cmd)
1408 {
1409         int ret = 0;
1410         ssize_t read_len;
1411         uint64_t len = 0;
1412         char *data = NULL;
1413         struct lttng_viewer_get_metadata request;
1414         struct lttng_viewer_metadata_packet reply;
1415         struct relay_viewer_stream *stream;
1416
1417         assert(cmd);
1418
1419         DBG("Relay get metadata");
1420
1421         if (cmd->version_check_done == 0) {
1422                 ERR("Trying to get metadata before version check");
1423                 ret = -1;
1424                 goto end;
1425         }
1426
1427         health_code_update();
1428         ret = cmd->sock->ops->recvmsg(cmd->sock, &request,
1429                         sizeof(request), 0);
1430         if (ret < 0 || ret != sizeof(request)) {
1431                 ret = -1;
1432                 ERR("Relay didn't receive the whole packet");
1433                 goto end;
1434         }
1435         health_code_update();
1436
1437         rcu_read_lock();
1438         stream = live_find_viewer_stream_by_id(be64toh(request.stream_id));
1439         if (!stream || !stream->metadata_flag) {
1440                 ERR("Invalid metadata stream");
1441                 goto error;
1442         }
1443         assert(stream->ctf_trace);
1444         assert(stream->ctf_trace->metadata_sent <=
1445                         stream->ctf_trace->metadata_received);
1446
1447         len = stream->ctf_trace->metadata_received -
1448                 stream->ctf_trace->metadata_sent;
1449         if (len == 0) {
1450                 reply.status = htobe32(VIEWER_NO_NEW_METADATA);
1451                 goto send_reply;
1452         }
1453
1454         /* first time, we open the metadata file */
1455         if (stream->read_fd < 0) {
1456                 char fullpath[PATH_MAX];
1457
1458                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1459                                 stream->channel_name);
1460                 if (ret < 0) {
1461                         goto error;
1462                 }
1463                 ret = open(fullpath, O_RDONLY);
1464                 if (ret < 0) {
1465                         PERROR("Relay opening metadata file");
1466                         goto error;
1467                 }
1468                 stream->read_fd = ret;
1469         }
1470
1471         reply.len = htobe64(len);
1472         data = zmalloc(len);
1473         if (!data) {
1474                 PERROR("viewer metadata zmalloc");
1475                 goto error;
1476         }
1477
1478         read_len = lttng_read(stream->read_fd, data, len);
1479         if (read_len < len) {
1480                 PERROR("Relay reading metadata file");
1481                 goto error;
1482         }
1483         stream->ctf_trace->metadata_sent += read_len;
1484         reply.status = htobe32(VIEWER_METADATA_OK);
1485         goto send_reply;
1486
1487 error:
1488         reply.status = htobe32(VIEWER_METADATA_ERR);
1489
1490 send_reply:
1491         health_code_update();
1492         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1493         if (ret < 0) {
1494                 ERR("Relay data header to viewer");
1495                 goto end_unlock;
1496         }
1497         health_code_update();
1498
1499         if (len > 0) {
1500                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1501                 if (ret < 0) {
1502                         ERR("Relay send data to viewer");
1503                         goto end_unlock;
1504                 }
1505         }
1506
1507         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1508                         be64toh(request.stream_id));
1509
1510         DBG("Metadata sent");
1511
1512 end_unlock:
1513         free(data);
1514         rcu_read_unlock();
1515 end:
1516         return ret;
1517 }
1518
1519 /*
1520  * live_relay_unknown_command: send -1 if received unknown command
1521  */
1522 static
1523 void live_relay_unknown_command(struct relay_command *cmd)
1524 {
1525         struct lttcomm_relayd_generic_reply reply;
1526         int ret;
1527
1528         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1529         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1530                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1531         if (ret < 0) {
1532                 ERR("Relay sending unknown command");
1533         }
1534 }
1535
1536 /*
1537  * Process the commands received on the control socket
1538  */
1539 static
1540 int process_control(struct lttng_viewer_cmd *recv_hdr,
1541                 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1542 {
1543         int ret = 0;
1544
1545         switch (be32toh(recv_hdr->cmd)) {
1546         case VIEWER_CONNECT:
1547                 ret = viewer_connect(cmd);
1548                 break;
1549         case VIEWER_LIST_SESSIONS:
1550                 ret = viewer_list_sessions(cmd, sessions_ht);
1551                 break;
1552         case VIEWER_ATTACH_SESSION:
1553                 ret = viewer_attach_session(cmd, sessions_ht);
1554                 break;
1555         case VIEWER_GET_NEXT_INDEX:
1556                 ret = viewer_get_next_index(cmd, sessions_ht);
1557                 break;
1558         case VIEWER_GET_PACKET:
1559                 ret = viewer_get_packet(cmd);
1560                 break;
1561         case VIEWER_GET_METADATA:
1562                 ret = viewer_get_metadata(cmd);
1563                 break;
1564         default:
1565                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1566                 live_relay_unknown_command(cmd);
1567                 ret = -1;
1568                 goto end;
1569         }
1570
1571 end:
1572         return ret;
1573 }
1574
1575 static
1576 void cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1577 {
1578         int ret;
1579
1580         assert(events);
1581
1582         lttng_poll_del(events, pollfd);
1583
1584         ret = close(pollfd);
1585         if (ret < 0) {
1586                 ERR("Closing pollfd %d", pollfd);
1587         }
1588 }
1589
1590 /*
1591  * Create and add connection to the given hash table.
1592  *
1593  * Return poll add value or else -1 on error.
1594  */
1595 static
1596 int add_connection(int fd, struct lttng_poll_event *events,
1597                 struct lttng_ht *relay_connections_ht)
1598 {
1599         int ret;
1600         struct relay_command *relay_connection;
1601
1602         assert(events);
1603         assert(relay_connections_ht);
1604
1605         relay_connection = zmalloc(sizeof(struct relay_command));
1606         if (relay_connection == NULL) {
1607                 PERROR("Relay command zmalloc");
1608                 goto error;
1609         }
1610
1611         ret = lttng_read(fd, relay_connection, sizeof(*relay_connection));
1612         if (ret < sizeof(*relay_connection)) {
1613                 PERROR("read relay cmd pipe");
1614                 goto error_read;
1615         }
1616
1617         lttng_ht_node_init_ulong(&relay_connection->sock_n,
1618                         (unsigned long) relay_connection->sock->fd);
1619         rcu_read_lock();
1620         lttng_ht_add_unique_ulong(relay_connections_ht,
1621                         &relay_connection->sock_n);
1622         rcu_read_unlock();
1623
1624         return lttng_poll_add(events, relay_connection->sock->fd,
1625                         LPOLLIN | LPOLLRDHUP);
1626
1627 error_read:
1628         free(relay_connection);
1629 error:
1630         return -1;
1631 }
1632
1633 static
1634 void deferred_free_connection(struct rcu_head *head)
1635 {
1636         struct relay_command *relay_connection =
1637                 caa_container_of(head, struct relay_command, rcu_node);
1638
1639         if (relay_connection->session &&
1640                         relay_connection->session->viewer_attached > 0) {
1641                 relay_connection->session->viewer_attached--;
1642         }
1643         lttcomm_destroy_sock(relay_connection->sock);
1644         free(relay_connection);
1645 }
1646
1647 static
1648 void deferred_free_viewer_stream(struct rcu_head *head)
1649 {
1650         struct relay_viewer_stream *stream =
1651                 caa_container_of(head, struct relay_viewer_stream, rcu_node);
1652
1653         if (stream->ctf_trace) {
1654                 uatomic_dec(&stream->ctf_trace->refcount);
1655                 assert(uatomic_read(&stream->ctf_trace->refcount) >= 0);
1656                 if (uatomic_read(&stream->ctf_trace->refcount) == 0) {
1657                         DBG("Freeing ctf_trace %" PRIu64, stream->ctf_trace->id);
1658                         free(stream->ctf_trace);
1659                 }
1660         }
1661
1662         free(stream->path_name);
1663         free(stream->channel_name);
1664         free(stream);
1665 }
1666
1667 static
1668 void viewer_del_streams(uint64_t session_id)
1669 {
1670         int ret;
1671         struct relay_viewer_stream *stream;
1672         struct lttng_ht_node_u64 *node;
1673         struct lttng_ht_iter iter;
1674
1675         rcu_read_lock();
1676         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
1677                 health_code_update();
1678
1679                 node = lttng_ht_iter_get_node_u64(&iter);
1680                 if (!node) {
1681                         continue;
1682                 }
1683
1684                 stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
1685                 if (stream->session_id != session_id) {
1686                         continue;
1687                 }
1688
1689                 if (stream->read_fd >= 0) {
1690                         ret = close(stream->read_fd);
1691                         if (ret < 0) {
1692                                 PERROR("close read_fd");
1693                         }
1694                 }
1695                 if (stream->index_read_fd >= 0) {
1696                         ret = close(stream->index_read_fd);
1697                         if (ret < 0) {
1698                                 PERROR("close index_read_fd");
1699                         }
1700                 }
1701                 if (stream->metadata_flag && stream->ctf_trace) {
1702                         stream->ctf_trace->metadata_sent = 0;
1703                 }
1704                 ret = lttng_ht_del(viewer_streams_ht, &iter);
1705                 assert(!ret);
1706                 call_rcu(&stream->rcu_node, deferred_free_viewer_stream);
1707         }
1708         rcu_read_unlock();
1709 }
1710
1711 /*
1712  * Delete and free a connection.
1713  *
1714  * RCU read side lock MUST be acquired.
1715  */
1716 static
1717 void del_connection(struct lttng_ht *relay_connections_ht,
1718                 struct lttng_ht_iter *iter, struct relay_command *relay_connection)
1719 {
1720         int ret;
1721
1722         assert(relay_connections_ht);
1723         assert(iter);
1724         assert(relay_connection);
1725
1726         ret = lttng_ht_del(relay_connections_ht, iter);
1727         assert(!ret);
1728
1729         viewer_del_streams(relay_connection->session_id);
1730
1731         call_rcu(&relay_connection->rcu_node, deferred_free_connection);
1732 }
1733
1734 /*
1735  * This thread does the actual work
1736  */
1737 static
1738 void *thread_worker(void *data)
1739 {
1740         int ret, err = -1;
1741         uint32_t nb_fd;
1742         struct relay_command *relay_connection;
1743         struct lttng_poll_event events;
1744         struct lttng_ht *relay_connections_ht;
1745         struct lttng_ht_node_ulong *node;
1746         struct lttng_ht_iter iter;
1747         struct lttng_viewer_cmd recv_hdr;
1748         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1749         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1750
1751         DBG("[thread] Live viewer relay worker started");
1752
1753         rcu_register_thread();
1754
1755         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1756
1757         /* table of connections indexed on socket */
1758         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1759         if (!relay_connections_ht) {
1760                 goto relay_connections_ht_error;
1761         }
1762
1763         ret = create_thread_poll_set(&events, 2);
1764         if (ret < 0) {
1765                 goto error_poll_create;
1766         }
1767
1768         ret = lttng_poll_add(&events, live_relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1769         if (ret < 0) {
1770                 goto error;
1771         }
1772
1773 restart:
1774         while (1) {
1775                 int i;
1776
1777                 health_code_update();
1778
1779                 /* Infinite blocking call, waiting for transmission */
1780                 DBG3("Relayd live viewer worker thread polling...");
1781                 health_poll_entry();
1782                 ret = lttng_poll_wait(&events, -1);
1783                 health_poll_exit();
1784                 if (ret < 0) {
1785                         /*
1786                          * Restart interrupted system call.
1787                          */
1788                         if (errno == EINTR) {
1789                                 goto restart;
1790                         }
1791                         goto error;
1792                 }
1793
1794                 nb_fd = ret;
1795
1796                 /*
1797                  * Process control. The control connection is prioritised so we don't
1798                  * starve it with high throughput tracing data on the data
1799                  * connection.
1800                  */
1801                 for (i = 0; i < nb_fd; i++) {
1802                         /* Fetch once the poll data */
1803                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1804                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1805
1806                         health_code_update();
1807
1808                         /* Thread quit pipe has been closed. Killing thread. */
1809                         ret = check_thread_quit_pipe(pollfd, revents);
1810                         if (ret) {
1811                                 err = 0;
1812                                 goto exit;
1813                         }
1814
1815                         /* Inspect the relay cmd pipe for new connection */
1816                         if (pollfd == live_relay_cmd_pipe[0]) {
1817                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1818                                         ERR("Relay live pipe error");
1819                                         goto error;
1820                                 } else if (revents & LPOLLIN) {
1821                                         DBG("Relay live viewer command received");
1822                                         ret = add_connection(live_relay_cmd_pipe[0],
1823                                                         &events, relay_connections_ht);
1824                                         if (ret < 0) {
1825                                                 goto error;
1826                                         }
1827                                 }
1828                         } else if (revents) {
1829                                 rcu_read_lock();
1830                                 lttng_ht_lookup(relay_connections_ht,
1831                                                 (void *)((unsigned long) pollfd), &iter);
1832                                 node = lttng_ht_iter_get_node_ulong(&iter);
1833                                 if (node == NULL) {
1834                                         DBG2("Relay viewer sock %d not found", pollfd);
1835                                         rcu_read_unlock();
1836                                         goto error;
1837                                 }
1838                                 relay_connection = caa_container_of(node, struct relay_command,
1839                                                 sock_n);
1840
1841                                 if (revents & (LPOLLERR)) {
1842                                         cleanup_poll_connection(&events, pollfd);
1843                                         del_connection(relay_connections_ht, &iter,
1844                                                         relay_connection);
1845                                 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1846                                         DBG("Viewer socket %d hung up", pollfd);
1847                                         cleanup_poll_connection(&events, pollfd);
1848                                         del_connection(relay_connections_ht, &iter,
1849                                                         relay_connection);
1850                                 } else if (revents & LPOLLIN) {
1851                                         ret = relay_connection->sock->ops->recvmsg(
1852                                                         relay_connection->sock, &recv_hdr,
1853                                                         sizeof(struct lttng_viewer_cmd),
1854                                                         0);
1855                                         /* connection closed */
1856                                         if (ret <= 0) {
1857                                                 cleanup_poll_connection(&events, pollfd);
1858                                                 del_connection(relay_connections_ht, &iter,
1859                                                                 relay_connection);
1860                                                 DBG("Viewer control connection closed with %d",
1861                                                                 pollfd);
1862                                         } else {
1863                                                 if (relay_connection->session) {
1864                                                         DBG2("Relay viewer worker receiving data for "
1865                                                                         "session: %" PRIu64,
1866                                                                         relay_connection->session->id);
1867                                                 }
1868                                                 ret = process_control(&recv_hdr, relay_connection,
1869                                                                 sessions_ht);
1870                                                 if (ret < 0) {
1871                                                         /* Clear the session on error. */
1872                                                         cleanup_poll_connection(&events, pollfd);
1873                                                         del_connection(relay_connections_ht, &iter,
1874                                                                         relay_connection);
1875                                                         DBG("Viewer connection closed with %d", pollfd);
1876                                                 }
1877                                         }
1878                                 }
1879                                 rcu_read_unlock();
1880                         }
1881                 }
1882         }
1883
1884 exit:
1885 error:
1886         lttng_poll_clean(&events);
1887
1888         /* empty the hash table and free the memory */
1889         rcu_read_lock();
1890         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
1891                 health_code_update();
1892
1893                 node = lttng_ht_iter_get_node_ulong(&iter);
1894                 if (!node) {
1895                         continue;
1896                 }
1897
1898                 relay_connection = caa_container_of(node, struct relay_command,
1899                                 sock_n);
1900                 del_connection(relay_connections_ht, &iter, relay_connection);
1901         }
1902         rcu_read_unlock();
1903 error_poll_create:
1904         lttng_ht_destroy(relay_connections_ht);
1905 relay_connections_ht_error:
1906         /* Close relay cmd pipes */
1907         utils_close_pipe(live_relay_cmd_pipe);
1908         if (err) {
1909                 DBG("Viewer worker thread exited with error");
1910         }
1911         DBG("Viewer worker thread cleanup complete");
1912         if (err) {
1913                 health_error();
1914                 ERR("Health error occurred in %s", __func__);
1915         }
1916         health_unregister(health_relayd);
1917         stop_threads();
1918         rcu_unregister_thread();
1919         return NULL;
1920 }
1921
1922 /*
1923  * Create the relay command pipe to wake thread_manage_apps.
1924  * Closed in cleanup().
1925  */
1926 static int create_relay_cmd_pipe(void)
1927 {
1928         int ret;
1929
1930         ret = utils_create_pipe_cloexec(live_relay_cmd_pipe);
1931
1932         return ret;
1933 }
1934
1935 void live_stop_threads(void)
1936 {
1937         int ret;
1938         void *status;
1939
1940         stop_threads();
1941
1942         ret = pthread_join(live_listener_thread, &status);
1943         if (ret != 0) {
1944                 PERROR("pthread_join live listener");
1945                 goto error;     /* join error, exit without cleanup */
1946         }
1947
1948         ret = pthread_join(live_worker_thread, &status);
1949         if (ret != 0) {
1950                 PERROR("pthread_join live worker");
1951                 goto error;     /* join error, exit without cleanup */
1952         }
1953
1954         ret = pthread_join(live_dispatcher_thread, &status);
1955         if (ret != 0) {
1956                 PERROR("pthread_join live dispatcher");
1957                 goto error;     /* join error, exit without cleanup */
1958         }
1959
1960         cleanup();
1961
1962 error:
1963         return;
1964 }
1965
1966 /*
1967  * main
1968  */
1969 int live_start_threads(struct lttng_uri *uri,
1970                 struct relay_local_data *relay_ctx, int quit_pipe[2])
1971 {
1972         int ret = 0;
1973         void *status;
1974         int is_root;
1975
1976         assert(uri);
1977         live_uri = uri;
1978
1979         live_thread_quit_pipe[0] = quit_pipe[0];
1980         live_thread_quit_pipe[1] = quit_pipe[1];
1981
1982         /* Check if daemon is UID = 0 */
1983         is_root = !getuid();
1984
1985         if (!is_root) {
1986                 if (live_uri->port < 1024) {
1987                         ERR("Need to be root to use ports < 1024");
1988                         ret = -1;
1989                         goto exit;
1990                 }
1991         }
1992
1993         /* Setup the thread apps communication pipe. */
1994         if ((ret = create_relay_cmd_pipe()) < 0) {
1995                 goto exit;
1996         }
1997
1998         /* Init relay command queue. */
1999         cds_wfq_init(&viewer_cmd_queue.queue);
2000
2001         /* Set up max poll set size */
2002         lttng_poll_set_max_size();
2003
2004         /* Setup the dispatcher thread */
2005         ret = pthread_create(&live_dispatcher_thread, NULL,
2006                         thread_dispatcher, (void *) NULL);
2007         if (ret != 0) {
2008                 PERROR("pthread_create viewer dispatcher");
2009                 goto exit_dispatcher;
2010         }
2011
2012         /* Setup the worker thread */
2013         ret = pthread_create(&live_worker_thread, NULL,
2014                         thread_worker, relay_ctx);
2015         if (ret != 0) {
2016                 PERROR("pthread_create viewer worker");
2017                 goto exit_worker;
2018         }
2019
2020         /* Setup the listener thread */
2021         ret = pthread_create(&live_listener_thread, NULL,
2022                         thread_listener, (void *) NULL);
2023         if (ret != 0) {
2024                 PERROR("pthread_create viewer listener");
2025                 goto exit_listener;
2026         }
2027
2028         ret = 0;
2029         goto end;
2030
2031 exit_listener:
2032         ret = pthread_join(live_listener_thread, &status);
2033         if (ret != 0) {
2034                 PERROR("pthread_join live listener");
2035                 goto error;     /* join error, exit without cleanup */
2036         }
2037
2038 exit_worker:
2039         ret = pthread_join(live_worker_thread, &status);
2040         if (ret != 0) {
2041                 PERROR("pthread_join live worker");
2042                 goto error;     /* join error, exit without cleanup */
2043         }
2044
2045 exit_dispatcher:
2046         ret = pthread_join(live_dispatcher_thread, &status);
2047         if (ret != 0) {
2048                 PERROR("pthread_join live dispatcher");
2049                 goto error;     /* join error, exit without cleanup */
2050         }
2051
2052 exit:
2053         cleanup();
2054
2055 end:
2056 error:
2057         return ret;
2058 }