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 | LPOLLERR);
 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         lttng_relay_notify_ready();
 263
 264         while (1) {
 265                 health_code_update();
 266
 267                 DBG("Listener accepting live viewers connections");
 268
 269 restart:
 270                 health_poll_entry();
 271                 ret = lttng_poll_wait(&events, -1);
 272                 health_poll_exit();
 273                 if (ret < 0) {
 274                         /*
 275                          * Restart interrupted system call.
 276                          */
 277                         if (errno == EINTR) {
 278                                 goto restart;
 279                         }
 280                         goto error;
 281                 }
 282                 nb_fd = ret;
 283
 284                 DBG("Relay new viewer connection received");
 285                 for (i = 0; i < nb_fd; i++) {
 286                         health_code_update();
 287
 288                         /* Fetch once the poll data */
 289                         revents = LTTNG_POLL_GETEV(&events, i);
 290                         pollfd = LTTNG_POLL_GETFD(&events, i);
 291
 292                         /* Thread quit pipe has been closed. Killing thread. */
 293                         ret = check_thread_quit_pipe(pollfd, revents);
 294                         if (ret) {
 295                                 err = 0;
 296                                 goto exit;
 297                         }
 298
 299                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 300                                 ERR("socket poll error");
 301                                 goto error;
 302                         } else if (revents & LPOLLIN) {
 303                                 /*
 304                                  * Get allocated in this thread, enqueued to a global queue,
 305                                  * dequeued and freed in the worker thread.
 306                                  */
 307                                 struct relay_command *relay_cmd;
 308                                 struct lttcomm_sock *newsock;
 309
 310                                 relay_cmd = zmalloc(sizeof(*relay_cmd));
 311                                 if (!relay_cmd) {
 312                                         PERROR("relay command zmalloc");
 313                                         goto error;
 314                                 }
 315
 316                                 assert(pollfd == live_control_sock->fd);
 317                                 newsock = live_control_sock->ops->accept(live_control_sock);
 318                                 if (!newsock) {
 319                                         PERROR("accepting control sock");
 320                                         free(relay_cmd);
 321                                         goto error;
 322                                 }
 323                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 324                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 325                                                 sizeof(int));
 326                                 if (ret < 0) {
 327                                         PERROR("setsockopt inet");
 328                                         lttcomm_destroy_sock(newsock);
 329                                         free(relay_cmd);
 330                                         goto error;
 331                                 }
 332                                 relay_cmd->sock = newsock;
 333
 334                                 /*
 335                                  * Lock free enqueue the request.
 336                                  */
 337                                 cds_wfq_enqueue(&viewer_cmd_queue.queue, &relay_cmd->node);
 338
 339                                 /*
 340                                  * Wake the dispatch queue futex. Implicit memory
 341                                  * barrier with the exchange in cds_wfq_enqueue.
 342                                  */
 343                                 futex_nto1_wake(&viewer_cmd_queue.futex);
 344                         }
 345                 }
 346         }
 347
 348 exit:
 349 error:
 350 error_poll_add:
 351         lttng_poll_clean(&events);
 352 error_create_poll:
 353         if (live_control_sock->fd >= 0) {
 354                 ret = live_control_sock->ops->close(live_control_sock);
 355                 if (ret) {
 356                         PERROR("close");
 357                 }
 358         }
 359         lttcomm_destroy_sock(live_control_sock);
 360 error_sock_control:
 361         if (err) {
 362                 health_error();
 363                 DBG("Live viewer listener thread exited with error");
 364         }
 365         health_unregister(health_relayd);
 366         DBG("Live viewer listener thread cleanup complete");
 367         stop_threads();
 368         return NULL;
 369 }
 370
 371 /*
 372  * This thread manages the dispatching of the requests to worker threads
 373  */
 374 static
 375 void *thread_dispatcher(void *data)
 376 {
 377         int err = -1;
 378         ssize_t ret;
 379         struct cds_wfq_node *node;
 380         struct relay_command *relay_cmd = NULL;
 381
 382         DBG("[thread] Live viewer relay dispatcher started");
 383
 384         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 385
 386         health_code_update();
 387
 388         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 389                 health_code_update();
 390
 391                 /* Atomically prepare the queue futex */
 392                 futex_nto1_prepare(&viewer_cmd_queue.futex);
 393
 394                 do {
 395                         health_code_update();
 396
 397                         /* Dequeue commands */
 398                         node = cds_wfq_dequeue_blocking(&viewer_cmd_queue.queue);
 399                         if (node == NULL) {
 400                                 DBG("Woken up but nothing in the live-viewer "
 401                                                 "relay command queue");
 402                                 /* Continue thread execution */
 403                                 break;
 404                         }
 405
 406                         relay_cmd = caa_container_of(node, struct relay_command, node);
 407                         DBG("Dispatching viewer request waiting on sock %d",
 408                                         relay_cmd->sock->fd);
 409
 410                         /*
 411                          * Inform worker thread of the new request. This call is blocking
 412                          * so we can be assured that the data will be read at some point in
 413                          * time or wait to the end of the world :)
 414                          */
 415                         ret = lttng_write(live_relay_cmd_pipe[1], relay_cmd,
 416                                         sizeof(*relay_cmd));
 417                         free(relay_cmd);
 418                         if (ret < sizeof(struct relay_command)) {
 419                                 PERROR("write cmd pipe");
 420                                 goto error;
 421                         }
 422                 } while (node != NULL);
 423
 424                 /* Futex wait on queue. Blocking call on futex() */
 425                 health_poll_entry();
 426                 futex_nto1_wait(&viewer_cmd_queue.futex);
 427                 health_poll_exit();
 428         }
 429
 430         /* Normal exit, no error */
 431         err = 0;
 432
 433 error:
 434         if (err) {
 435                 health_error();
 436                 ERR("Health error occurred in %s", __func__);
 437         }
 438         health_unregister(health_relayd);
 439         DBG("Live viewer dispatch thread dying");
 440         stop_threads();
 441         return NULL;
 442 }
 443
 444 /*
 445  * Establish connection with the viewer and check the versions.
 446  *
 447  * Return 0 on success or else negative value.
 448  */
 449 static
 450 int viewer_connect(struct relay_command *cmd)
 451 {
 452         int ret;
 453         struct lttng_viewer_connect reply, msg;
 454
 455         assert(cmd);
 456
 457         cmd->version_check_done = 1;
 458
 459         health_code_update();
 460
 461         /* Get version from the other side. */
 462         ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
 463         if (ret < 0 || ret != sizeof(msg)) {
 464                 if (ret == 0) {
 465                         /* Orderly shutdown. Not necessary to print an error. */
 466                         DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
 467                 } else {
 468                         ERR("Relay failed to receive the version values.");
 469                 }
 470                 ret = -1;
 471                 goto end;
 472         }
 473
 474         health_code_update();
 475
 476         reply.major = RELAYD_VERSION_COMM_MAJOR;
 477         reply.minor = RELAYD_VERSION_COMM_MINOR;
 478
 479         /* Major versions must be the same */
 480         if (reply.major != be32toh(msg.major)) {
 481                 DBG("Incompatible major versions (%u vs %u)", reply.major,
 482                                 be32toh(msg.major));
 483                 ret = -1;
 484                 goto end;
 485         }
 486
 487         cmd->major = reply.major;
 488         /* We adapt to the lowest compatible version */
 489         if (reply.minor <= be32toh(msg.minor)) {
 490                 cmd->minor = reply.minor;
 491         } else {
 492                 cmd->minor = be32toh(msg.minor);
 493         }
 494
 495         if (be32toh(msg.type) == VIEWER_CLIENT_COMMAND) {
 496                 cmd->type = RELAY_VIEWER_COMMAND;
 497         } else if (be32toh(msg.type) == VIEWER_CLIENT_NOTIFICATION) {
 498                 cmd->type = RELAY_VIEWER_NOTIFICATION;
 499         } else {
 500                 ERR("Unknown connection type : %u", be32toh(msg.type));
 501                 ret = -1;
 502                 goto end;
 503         }
 504
 505         reply.major = htobe32(reply.major);
 506         reply.minor = htobe32(reply.minor);
 507         if (cmd->type == RELAY_VIEWER_COMMAND) {
 508                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 509         }
 510
 511         health_code_update();
 512
 513         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
 514                         sizeof(struct lttng_viewer_connect), 0);
 515         if (ret < 0) {
 516                 ERR("Relay sending version");
 517         }
 518
 519         health_code_update();
 520
 521         DBG("Version check done using protocol %u.%u", cmd->major, cmd->minor);
 522         ret = 0;
 523
 524 end:
 525         return ret;
 526 }
 527
 528 /*
 529  * Send the viewer the list of current sessions.
 530  *
 531  * Return 0 on success or else a negative value.
 532  */
 533 static
 534 int viewer_list_sessions(struct relay_command *cmd,
 535                 struct lttng_ht *sessions_ht)
 536 {
 537         int ret;
 538         struct lttng_viewer_list_sessions session_list;
 539         unsigned long count;
 540         long approx_before, approx_after;
 541         struct lttng_ht_node_ulong *node;
 542         struct lttng_ht_iter iter;
 543         struct lttng_viewer_session send_session;
 544         struct relay_session *session;
 545
 546         DBG("List sessions received");
 547
 548         if (cmd->version_check_done == 0) {
 549                 ERR("Trying to list sessions before version check");
 550                 ret = -1;
 551                 goto end_no_session;
 552         }
 553
 554         rcu_read_lock();
 555         cds_lfht_count_nodes(sessions_ht->ht, &approx_before, &count, &approx_after);
 556         session_list.sessions_count = htobe32(count);
 557
 558         health_code_update();
 559
 560         ret = cmd->sock->ops->sendmsg(cmd->sock, &session_list,
 561                         sizeof(session_list), 0);
 562         if (ret < 0) {
 563                 ERR("Relay sending sessions list");
 564                 goto end_unlock;
 565         }
 566
 567         health_code_update();
 568
 569         cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, node, node) {
 570                 health_code_update();
 571
 572                 node = lttng_ht_iter_get_node_ulong(&iter);
 573                 if (!node) {
 574                         goto end_unlock;
 575                 }
 576                 session = caa_container_of(node, struct relay_session, session_n);
 577
 578                 strncpy(send_session.session_name, session->session_name,
 579                                 sizeof(send_session.session_name));
 580                 strncpy(send_session.hostname, session->hostname,
 581                                 sizeof(send_session.hostname));
 582                 send_session.id = htobe64(session->id);
 583                 send_session.live_timer = htobe32(session->live_timer);
 584                 send_session.clients = htobe32(session->viewer_attached);
 585                 send_session.streams = htobe32(session->stream_count);
 586
 587                 health_code_update();
 588
 589                 ret = cmd->sock->ops->sendmsg(cmd->sock, &send_session,
 590                                 sizeof(send_session), 0);
 591                 if (ret < 0) {
 592                         ERR("Relay sending session info");
 593                         goto end_unlock;
 594                 }
 595         }
 596         health_code_update();
 597
 598         rcu_read_unlock();
 599         ret = 0;
 600         goto end;
 601
 602 end_unlock:
 603         rcu_read_unlock();
 604
 605 end:
 606 end_no_session:
 607         return ret;
 608 }
 609
 610 /*
 611  * Open index file using a given viewer stream.
 612  *
 613  * Return 0 on success or else a negative value.
 614  */
 615 static int open_index(struct relay_viewer_stream *stream)
 616 {
 617         int ret;
 618         char fullpath[PATH_MAX];
 619         struct ctf_packet_index_file_hdr hdr;
 620
 621         if (stream->tracefile_count > 0) {
 622                 ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s_%"
 623                                 PRIu64 DEFAULT_INDEX_FILE_SUFFIX, stream->path_name,
 624                                 stream->channel_name, stream->tracefile_count_current);
 625         } else {
 626                 ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR "/%s"
 627                                 DEFAULT_INDEX_FILE_SUFFIX, stream->path_name,
 628                                 stream->channel_name);
 629         }
 630         if (ret < 0) {
 631                 PERROR("snprintf index path");
 632                 goto error;
 633         }
 634
 635         DBG("Opening index file %s in read only", fullpath);
 636         ret = open(fullpath, O_RDONLY);
 637         if (ret < 0) {
 638                 if (errno == ENOENT) {
 639                         ret = -ENOENT;
 640                         goto error;
 641                 } else {
 642                         PERROR("opening index in read-only");
 643                 }
 644                 goto error;
 645         }
 646         stream->index_read_fd = ret;
 647         DBG("Opening index file %s in read only, (fd: %d)", fullpath, ret);
 648
 649         ret = lttng_read(stream->index_read_fd, &hdr, sizeof(hdr));
 650         if (ret < sizeof(hdr)) {
 651                 PERROR("Reading index header");
 652                 goto error;
 653         }
 654         if (be32toh(hdr.magic) != CTF_INDEX_MAGIC) {
 655                 ERR("Invalid header magic");
 656                 ret = -1;
 657                 goto error;
 658         }
 659         if (be32toh(hdr.index_major) != CTF_INDEX_MAJOR ||
 660                         be32toh(hdr.index_minor) != CTF_INDEX_MINOR) {
 661                 ERR("Invalid header version");
 662                 ret = -1;
 663                 goto error;
 664         }
 665         ret = 0;
 666
 667 error:
 668         return ret;
 669 }
 670
 671 /*
 672  * Allocate and init a new viewer_stream.
 673  *
 674  * Copies the values from the stream passed in parameter and insert the new
 675  * stream in the viewer_streams_ht.
 676  *
 677  * MUST be called with rcu_read_lock held.
 678  *
 679  * Returns 0 on success or a negative value on error.
 680  */
 681 static
 682 int init_viewer_stream(struct relay_stream *stream, int seek_last)
 683 {
 684         int ret;
 685         struct relay_viewer_stream *viewer_stream;
 686
 687         assert(stream);
 688
 689         viewer_stream = zmalloc(sizeof(*viewer_stream));
 690         if (!viewer_stream) {
 691                 PERROR("relay viewer stream zmalloc");
 692                 ret = -1;
 693                 goto error;
 694         }
 695         viewer_stream->session_id = stream->session->id;
 696         viewer_stream->stream_handle = stream->stream_handle;
 697         viewer_stream->path_name = strndup(stream->path_name,
 698                         LTTNG_VIEWER_PATH_MAX);
 699         viewer_stream->channel_name = strndup(stream->channel_name,
 700                         LTTNG_VIEWER_NAME_MAX);
 701         viewer_stream->tracefile_count = stream->tracefile_count;
 702         viewer_stream->metadata_flag = stream->metadata_flag;
 703         viewer_stream->tracefile_count_last = -1ULL;
 704         if (seek_last) {
 705                 viewer_stream->tracefile_count_current =
 706                         stream->tracefile_count_current;
 707         } else {
 708                 viewer_stream->tracefile_count_current =
 709                         stream->oldest_tracefile_id;
 710         }
 711
 712         viewer_stream->ctf_trace = stream->ctf_trace;
 713         if (viewer_stream->metadata_flag) {
 714                 viewer_stream->ctf_trace->viewer_metadata_stream =
 715                         viewer_stream;
 716         }
 717         uatomic_inc(&viewer_stream->ctf_trace->refcount);
 718
 719         lttng_ht_node_init_u64(&viewer_stream->stream_n, stream->stream_handle);
 720         lttng_ht_add_unique_u64(viewer_streams_ht, &viewer_stream->stream_n);
 721
 722         viewer_stream->index_read_fd = -1;
 723         viewer_stream->read_fd = -1;
 724
 725         /*
 726          * This is to avoid a race between the initialization of this object and
 727          * the close of the given stream. If the stream is unable to find this
 728          * viewer stream when closing, this copy will at least take the latest
 729          * value.
 730          * We also need that for the seek_last.
 731          */
 732         viewer_stream->total_index_received = stream->total_index_received;
 733
 734         /*
 735          * If we never received an index for the current stream, delay
 736          * the opening of the index, otherwise open it right now.
 737          */
 738         if (viewer_stream->tracefile_count_current ==
 739                         stream->tracefile_count_current &&
 740                         viewer_stream->total_index_received == 0) {
 741                 viewer_stream->index_read_fd = -1;
 742         } else {
 743                 ret = open_index(viewer_stream);
 744                 if (ret < 0) {
 745                         goto error;
 746                 }
 747         }
 748
 749         if (seek_last && viewer_stream->index_read_fd > 0) {
 750                 ret = lseek(viewer_stream->index_read_fd,
 751                                 viewer_stream->total_index_received *
 752                                         sizeof(struct ctf_packet_index),
 753                                 SEEK_CUR);
 754                 if (ret < 0) {
 755                         goto error;
 756                 }
 757                 viewer_stream->last_sent_index =
 758                         viewer_stream->total_index_received;
 759         }
 760
 761         ret = 0;
 762
 763 error:
 764         return ret;
 765 }
 766
 767 /*
 768  * Rotate a stream to the next tracefile.
 769  *
 770  * Returns 0 on success, 1 on EOF, a negative value on error.
 771  */
 772 static
 773 int rotate_viewer_stream(struct relay_viewer_stream *viewer_stream,
 774                 struct relay_stream *stream)
 775 {
 776         int ret;
 777         uint64_t tracefile_id;
 778
 779         assert(viewer_stream);
 780
 781         tracefile_id = (viewer_stream->tracefile_count_current + 1) %
 782                 viewer_stream->tracefile_count;
 783         /*
 784          * Detect the last tracefile to open.
 785          */
 786         if (viewer_stream->tracefile_count_last != -1ULL &&
 787                         viewer_stream->tracefile_count_last ==
 788                         viewer_stream->tracefile_count_current) {
 789                 ret = 1;
 790                 goto end;
 791         }
 792
 793         if (stream) {
 794                 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
 795         }
 796         /*
 797          * The writer and the reader are not working in the same
 798          * tracefile, we can read up to EOF, we don't care about the
 799          * total_index_received.
 800          */
 801         if (!stream || (stream->tracefile_count_current != tracefile_id)) {
 802                 viewer_stream->close_write_flag = 1;
 803         } else {
 804                 /*
 805                  * We are opening a file that is still open in write, make
 806                  * sure we limit our reading to the number of indexes
 807                  * received.
 808                  */
 809                 viewer_stream->close_write_flag = 0;
 810                 if (stream) {
 811                         viewer_stream->total_index_received =
 812                                 stream->total_index_received;
 813                 }
 814         }
 815         viewer_stream->tracefile_count_current = tracefile_id;
 816
 817         ret = close(viewer_stream->index_read_fd);
 818         if (ret < 0) {
 819                 PERROR("close index file %d",
 820                                 viewer_stream->index_read_fd);
 821         }
 822         viewer_stream->index_read_fd = -1;
 823         ret = close(viewer_stream->read_fd);
 824         if (ret < 0) {
 825                 PERROR("close tracefile %d",
 826                                 viewer_stream->read_fd);
 827         }
 828         viewer_stream->read_fd = -1;
 829
 830         pthread_mutex_lock(&viewer_stream->overwrite_lock);
 831         viewer_stream->abort_flag = 0;
 832         pthread_mutex_unlock(&viewer_stream->overwrite_lock);
 833
 834         viewer_stream->index_read_fd = -1;
 835         viewer_stream->read_fd = -1;
 836
 837         if (stream) {
 838                 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
 839         }
 840         ret = open_index(viewer_stream);
 841         if (ret < 0) {
 842                 goto error;
 843         }
 844
 845         ret = 0;
 846
 847 end:
 848 error:
 849         return ret;
 850 }
 851
 852 /*
 853  * Send the viewer the list of current sessions.
 854  */
 855 static
 856 int viewer_attach_session(struct relay_command *cmd,
 857                 struct lttng_ht *sessions_ht)
 858 {
 859         int ret, send_streams = 0;
 860         uint32_t nb_streams = 0, nb_streams_ready = 0;
 861         struct lttng_viewer_attach_session_request request;
 862         struct lttng_viewer_attach_session_response response;
 863         struct lttng_viewer_stream send_stream;
 864         struct relay_stream *stream;
 865         struct relay_viewer_stream *viewer_stream;
 866         struct lttng_ht_node_ulong *node;
 867         struct lttng_ht_node_u64 *node64;
 868         struct lttng_ht_iter iter;
 869         struct relay_session *session;
 870         int seek_last = 0;
 871
 872         assert(cmd);
 873         assert(sessions_ht);
 874
 875         DBG("Attach session received");
 876
 877         if (cmd->version_check_done == 0) {
 878                 ERR("Trying to attach session before version check");
 879                 ret = -1;
 880                 goto end_no_session;
 881         }
 882
 883         health_code_update();
 884
 885         ret = cmd->sock->ops->recvmsg(cmd->sock, &request, sizeof(request), 0);
 886         if (ret < 0 || ret != sizeof(request)) {
 887                 if (ret == 0) {
 888                         /* Orderly shutdown. Not necessary to print an error. */
 889                         DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
 890                 } else {
 891                         ERR("Relay failed to receive the attach parameters.");
 892                 }
 893                 ret = -1;
 894                 goto error;
 895         }
 896
 897         health_code_update();
 898
 899         rcu_read_lock();
 900         lttng_ht_lookup(sessions_ht,
 901                         (void *)((unsigned long) be64toh(request.session_id)), &iter);
 902         node = lttng_ht_iter_get_node_ulong(&iter);
 903         if (node == NULL) {
 904                 DBG("Relay session %" PRIu64 " not found",
 905                                 be64toh(request.session_id));
 906                 response.status = htobe32(VIEWER_ATTACH_UNK);
 907                 goto send_reply;
 908         }
 909
 910         session = caa_container_of(node, struct relay_session, session_n);
 911         if (cmd->session_id == session->id) {
 912                 /* Same viewer already attached, just send the stream list. */
 913                 send_streams = 1;
 914                 response.status = htobe32(VIEWER_ATTACH_OK);
 915         } else if (session->viewer_attached != 0) {
 916                 DBG("Already a viewer attached");
 917                 response.status = htobe32(VIEWER_ATTACH_ALREADY);
 918                 goto send_reply;
 919         } else if (session->live_timer == 0) {
 920                 DBG("Not live session");
 921                 response.status = htobe32(VIEWER_ATTACH_NOT_LIVE);
 922                 goto send_reply;
 923         } else {
 924                 session->viewer_attached++;
 925                 send_streams = 1;
 926                 response.status = htobe32(VIEWER_ATTACH_OK);
 927                 cmd->session_id = session->id;
 928                 cmd->session = session;
 929         }
 930
 931         switch (be32toh(request.seek)) {
 932         case VIEWER_SEEK_BEGINNING:
 933                 /* Default behaviour. */
 934                 break;
 935         case VIEWER_SEEK_LAST:
 936                 seek_last = 1;
 937                 break;
 938         default:
 939                 ERR("Wrong seek parameter");
 940                 response.status = htobe32(VIEWER_ATTACH_SEEK_ERR);
 941                 send_streams = 0;
 942                 goto send_reply;
 943         }
 944
 945         if (send_streams) {
 946                 /* We should only be there if we have a session to attach to. */
 947                 assert(session);
 948
 949                 /*
 950                  * Fill the viewer_streams_ht to count the number of streams
 951                  * ready to be sent and avoid concurrency issues on the
 952                  * relay_streams_ht and don't rely on a total session stream count.
 953                  */
 954                 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
 955                         struct relay_viewer_stream *vstream;
 956
 957                         health_code_update();
 958
 959                         node = lttng_ht_iter_get_node_ulong(&iter);
 960                         if (!node) {
 961                                 continue;
 962                         }
 963                         stream = caa_container_of(node, struct relay_stream, stream_n);
 964                         if (stream->session != cmd->session) {
 965                                 continue;
 966                         }
 967                         nb_streams++;
 968
 969                         /*
 970                          * Don't send streams with no ctf_trace, they are not
 971                          * ready to be read.
 972                          */
 973                         if (!stream->ctf_trace || !stream->viewer_ready) {
 974                                 continue;
 975                         }
 976                         nb_streams_ready++;
 977
 978                         vstream = live_find_viewer_stream_by_id(stream->stream_handle);
 979                         if (!vstream) {
 980                                 ret = init_viewer_stream(stream, seek_last);
 981                                 if (ret < 0) {
 982                                         goto end_unlock;
 983                                 }
 984                         }
 985                 }
 986
 987                 /* We must have the same amount of existing stream and ready stream. */
 988                 if (nb_streams != nb_streams_ready) {
 989                         nb_streams = 0;
 990                 }
 991                 response.streams_count = htobe32(nb_streams);
 992         }
 993
 994 send_reply:
 995         health_code_update();
 996         ret = cmd->sock->ops->sendmsg(cmd->sock, &response, sizeof(response), 0);
 997         if (ret < 0) {
 998                 ERR("Relay sending viewer attach response");
 999                 goto end_unlock;
1000         }
1001         health_code_update();
1002
1003         /*
1004          * Unknown or empty session, just return gracefully, the viewer knows what
1005          * is happening.
1006          */
1007         if (!send_streams || !nb_streams) {
1008                 ret = 0;
1009                 goto end_unlock;
1010         }
1011
1012         /* We should only be there if we have a session to attach to. */
1013         assert(session);
1014         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
1015                 health_code_update();
1016
1017                 node64 = lttng_ht_iter_get_node_u64(&iter);
1018                 if (!node64) {
1019                         continue;
1020                 }
1021                 viewer_stream = caa_container_of(node64, struct relay_viewer_stream,
1022                                 stream_n);
1023                 if (viewer_stream->session_id != cmd->session->id) {
1024                         continue;
1025                 }
1026
1027                 send_stream.id = htobe64(viewer_stream->stream_handle);
1028                 send_stream.ctf_trace_id = htobe64(viewer_stream->ctf_trace->id);
1029                 send_stream.metadata_flag = htobe32(viewer_stream->metadata_flag);
1030                 strncpy(send_stream.path_name, viewer_stream->path_name,
1031                                 sizeof(send_stream.path_name));
1032                 strncpy(send_stream.channel_name, viewer_stream->channel_name,
1033                                 sizeof(send_stream.channel_name));
1034
1035                 ret = cmd->sock->ops->sendmsg(cmd->sock, &send_stream,
1036                                 sizeof(send_stream), 0);
1037                 if (ret < 0) {
1038                         ERR("Relay sending stream %" PRIu64, viewer_stream->stream_handle);
1039                         goto end_unlock;
1040                 }
1041                 DBG("Sent stream %" PRIu64 " to viewer", viewer_stream->stream_handle);
1042         }
1043         ret = 0;
1044
1045 end_unlock:
1046         rcu_read_unlock();
1047 end_no_session:
1048 error:
1049         return ret;
1050 }
1051
1052 /*
1053  * Get viewer stream from stream id.
1054  *
1055  * RCU read side lock MUST be acquired.
1056  */
1057 struct relay_viewer_stream *live_find_viewer_stream_by_id(uint64_t stream_id)
1058 {
1059         struct lttng_ht_node_u64 *node;
1060         struct lttng_ht_iter iter;
1061         struct relay_viewer_stream *stream = NULL;
1062
1063         lttng_ht_lookup(viewer_streams_ht, &stream_id, &iter);
1064         node = lttng_ht_iter_get_node_u64(&iter);
1065         if (node == NULL) {
1066                 DBG("Relay viewer stream %" PRIu64 " not found", stream_id);
1067                 goto end;
1068         }
1069         stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
1070
1071 end:
1072         return stream;
1073 }
1074
1075 static
1076 void deferred_free_viewer_stream(struct rcu_head *head)
1077 {
1078         struct relay_viewer_stream *stream =
1079                 caa_container_of(head, struct relay_viewer_stream, rcu_node);
1080
1081         free(stream->path_name);
1082         free(stream->channel_name);
1083         free(stream);
1084 }
1085
1086 static
1087 void delete_viewer_stream(struct relay_viewer_stream *vstream)
1088 {
1089         int delret;
1090         struct lttng_ht_iter iter;
1091
1092         iter.iter.node = &vstream->stream_n.node;
1093         delret = lttng_ht_del(viewer_streams_ht, &iter);
1094         assert(!delret);
1095 }
1096
1097 static
1098 void destroy_viewer_stream(struct relay_viewer_stream *vstream)
1099 {
1100         unsigned long ret_ref;
1101         int ret;
1102
1103         assert(vstream);
1104         ret_ref = uatomic_add_return(&vstream->ctf_trace->refcount, -1);
1105         assert(ret_ref >= 0);
1106
1107         if (vstream->read_fd >= 0) {
1108                 ret = close(vstream->read_fd);
1109                 if (ret < 0) {
1110                         PERROR("close read_fd");
1111                 }
1112         }
1113         if (vstream->index_read_fd >= 0) {
1114                 ret = close(vstream->index_read_fd);
1115                 if (ret < 0) {
1116                         PERROR("close index_read_fd");
1117                 }
1118         }
1119
1120         /*
1121          * If the only stream left in the HT is the metadata stream,
1122          * we need to remove it because we won't detect a EOF for this
1123          * stream.
1124          */
1125         if (ret_ref == 1 && vstream->ctf_trace->viewer_metadata_stream) {
1126                 delete_viewer_stream(vstream->ctf_trace->viewer_metadata_stream);
1127                 destroy_viewer_stream(vstream->ctf_trace->viewer_metadata_stream);
1128                 vstream->ctf_trace->metadata_stream = NULL;
1129                 DBG("Freeing ctf_trace %" PRIu64, vstream->ctf_trace->id);
1130                 /*
1131                  * The streaming-side is already closed and we can't receive a new
1132                  * stream concurrently at this point (since the session is being
1133                  * destroyed), so when we detect the refcount equals 0, we are the
1134                  * only owners of the ctf_trace and we can free it ourself.
1135                  */
1136                 free(vstream->ctf_trace);
1137         }
1138
1139         call_rcu(&vstream->rcu_node, deferred_free_viewer_stream);
1140 }
1141
1142 /*
1143  * Send the next index for a stream.
1144  *
1145  * Return 0 on success or else a negative value.
1146  */
1147 static
1148 int viewer_get_next_index(struct relay_command *cmd,
1149                 struct lttng_ht *sessions_ht)
1150 {
1151         int ret;
1152         struct lttng_viewer_get_next_index request_index;
1153         struct lttng_viewer_index viewer_index;
1154         struct ctf_packet_index packet_index;
1155         struct relay_viewer_stream *vstream;
1156         struct relay_stream *rstream;
1157
1158         assert(cmd);
1159         assert(sessions_ht);
1160
1161         DBG("Viewer get next index");
1162
1163         if (cmd->version_check_done == 0) {
1164                 ERR("Trying to request index before version check");
1165                 ret = -1;
1166                 goto end_no_session;
1167         }
1168
1169         health_code_update();
1170         ret = cmd->sock->ops->recvmsg(cmd->sock, &request_index,
1171                         sizeof(request_index), 0);
1172         if (ret < 0 || ret != sizeof(request_index)) {
1173                 ret = -1;
1174                 ERR("Relay didn't receive the whole packet");
1175                 goto end;
1176         }
1177         health_code_update();
1178
1179         rcu_read_lock();
1180         vstream = live_find_viewer_stream_by_id(be64toh(request_index.stream_id));
1181         if (!vstream) {
1182                 ret = -1;
1183                 goto end_unlock;
1184         }
1185
1186         memset(&viewer_index, 0, sizeof(viewer_index));
1187
1188         /*
1189          * The viewer should not ask for index on metadata stream.
1190          */
1191         if (vstream->metadata_flag) {
1192                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
1193                 goto send_reply;
1194         }
1195
1196         /* First time, we open the index file */
1197         if (vstream->index_read_fd < 0) {
1198                 ret = open_index(vstream);
1199                 if (ret == -ENOENT) {
1200                         /*
1201                          * The index is created only when the first data packet arrives, it
1202                          * might not be ready at the beginning of the session
1203                          */
1204                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1205                         goto send_reply;
1206                 } else if (ret < 0) {
1207                         viewer_index.status = htobe32(VIEWER_INDEX_ERR);
1208                         goto send_reply;
1209                 }
1210         }
1211
1212         rstream = relay_stream_find_by_id(vstream->stream_handle);
1213         if (rstream) {
1214                 if (vstream->abort_flag) {
1215                         /* Rotate on abort (overwrite). */
1216                         DBG("Viewer rotate because of overwrite");
1217                         ret = rotate_viewer_stream(vstream, rstream);
1218                         if (ret < 0) {
1219                                 goto end_unlock;
1220                         } else if (ret == 1) {
1221                                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
1222                                 delete_viewer_stream(vstream);
1223                                 destroy_viewer_stream(vstream);
1224                                 goto send_reply;
1225                         }
1226                 }
1227                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1228                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1229                         if (rstream->beacon_ts_end != -1ULL &&
1230                                 vstream->last_sent_index == rstream->total_index_received) {
1231                                 viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE);
1232                                 viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
1233                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1234                                 goto send_reply;
1235                         /*
1236                          * Reader and writer are working in the same tracefile, so we care
1237                          * about the number of index received and sent. Otherwise, we read
1238                          * up to EOF.
1239                          */
1240                         } else if (rstream->total_index_received <= vstream->last_sent_index
1241                                         && !vstream->close_write_flag) {
1242                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1243                                 /* No new index to send, retry later. */
1244                                 viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1245                                 goto send_reply;
1246                         }
1247                 }
1248                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1249         } else if (!rstream && vstream->close_write_flag &&
1250                         vstream->total_index_received == vstream->last_sent_index) {
1251                 /* Last index sent and current tracefile closed in write */
1252                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
1253                 delete_viewer_stream(vstream);
1254                 destroy_viewer_stream(vstream);
1255                 goto send_reply;
1256         } else {
1257                 vstream->close_write_flag = 1;
1258         }
1259
1260         if (!vstream->ctf_trace->metadata_received ||
1261                         vstream->ctf_trace->metadata_received >
1262                         vstream->ctf_trace->metadata_sent) {
1263                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1264         }
1265
1266         pthread_mutex_lock(&vstream->overwrite_lock);
1267         if (vstream->abort_flag) {
1268                 /*
1269                  * The file is being overwritten by the writer, we cannot
1270                  * use it.
1271                  */
1272                 viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1273                 pthread_mutex_unlock(&vstream->overwrite_lock);
1274                 ret = rotate_viewer_stream(vstream, rstream);
1275                 if (ret < 0) {
1276                         goto end_unlock;
1277                 } else if (ret == 1) {
1278                         viewer_index.status = htobe32(VIEWER_INDEX_HUP);
1279                         delete_viewer_stream(vstream);
1280                         destroy_viewer_stream(vstream);
1281                         goto send_reply;
1282                 }
1283                 goto send_reply;
1284         }
1285         ret = lttng_read(vstream->index_read_fd, &packet_index,
1286                         sizeof(packet_index));
1287         pthread_mutex_unlock(&vstream->overwrite_lock);
1288         if (ret < sizeof(packet_index)) {
1289                 /*
1290                  * The tracefile is closed in write, so we read up to EOF.
1291                  */
1292                 if (vstream->close_write_flag == 1) {
1293                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
1294                         /* Rotate on normal EOF */
1295                         ret = rotate_viewer_stream(vstream, rstream);
1296                         if (ret < 0) {
1297                                 goto end_unlock;
1298                         } else if (ret == 1) {
1299                                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
1300                                 delete_viewer_stream(vstream);
1301                                 destroy_viewer_stream(vstream);
1302                                 goto send_reply;
1303                         }
1304                 } else {
1305                         PERROR("Relay reading index file %d",
1306                                         vstream->index_read_fd);
1307                         viewer_index.status = htobe32(VIEWER_INDEX_ERR);
1308                 }
1309                 goto send_reply;
1310         } else {
1311                 viewer_index.status = htobe32(VIEWER_INDEX_OK);
1312                 vstream->last_sent_index++;
1313         }
1314
1315         /*
1316          * Indexes are stored in big endian, no need to switch before sending.
1317          */
1318         viewer_index.offset = packet_index.offset;
1319         viewer_index.packet_size = packet_index.packet_size;
1320         viewer_index.content_size = packet_index.content_size;
1321         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1322         viewer_index.timestamp_end = packet_index.timestamp_end;
1323         viewer_index.events_discarded = packet_index.events_discarded;
1324         viewer_index.stream_id = packet_index.stream_id;
1325
1326 send_reply:
1327         viewer_index.flags = htobe32(viewer_index.flags);
1328         health_code_update();
1329         ret = cmd->sock->ops->sendmsg(cmd->sock, &viewer_index,
1330                         sizeof(viewer_index), 0);
1331         if (ret < 0) {
1332                 ERR("Relay index to viewer");
1333                 goto end_unlock;
1334         }
1335         health_code_update();
1336
1337         DBG("Index %" PRIu64 "for stream %" PRIu64 "sent",
1338                         vstream->last_sent_index, vstream->stream_handle);
1339
1340 end_unlock:
1341         rcu_read_unlock();
1342
1343 end_no_session:
1344 end:
1345         return ret;
1346 }
1347
1348 /*
1349  * Send the next index for a stream
1350  *
1351  * Return 0 on success or else a negative value.
1352  */
1353 static
1354 int viewer_get_packet(struct relay_command *cmd)
1355 {
1356         int ret, send_data = 0;
1357         char *data = NULL;
1358         uint32_t len = 0;
1359         ssize_t read_len;
1360         struct lttng_viewer_get_packet get_packet_info;
1361         struct lttng_viewer_trace_packet reply;
1362         struct relay_viewer_stream *stream;
1363
1364         assert(cmd);
1365
1366         DBG2("Relay get data packet");
1367
1368         if (cmd->version_check_done == 0) {
1369                 ERR("Trying to get packet before version check");
1370                 ret = -1;
1371                 goto end;
1372         }
1373
1374         health_code_update();
1375         ret = cmd->sock->ops->recvmsg(cmd->sock, &get_packet_info,
1376                         sizeof(get_packet_info), 0);
1377         if (ret < 0 || ret != sizeof(get_packet_info)) {
1378                 ret = -1;
1379                 ERR("Relay didn't receive the whole packet");
1380                 goto end;
1381         }
1382         health_code_update();
1383
1384         /* From this point on, the error label can be reached. */
1385         memset(&reply, 0, sizeof(reply));
1386
1387         rcu_read_lock();
1388         stream = live_find_viewer_stream_by_id(be64toh(get_packet_info.stream_id));
1389         if (!stream) {
1390                 goto error;
1391         }
1392         assert(stream->ctf_trace);
1393
1394         /*
1395          * First time we read this stream, we need open the tracefile, we should
1396          * only arrive here if an index has already been sent to the viewer, so the
1397          * tracefile must exist, if it does not it is a fatal error.
1398          */
1399         if (stream->read_fd < 0) {
1400                 char fullpath[PATH_MAX];
1401
1402                 if (stream->tracefile_count > 0) {
1403                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1404                                         stream->channel_name,
1405                                         stream->tracefile_count_current);
1406                 } else {
1407                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1408                                         stream->channel_name);
1409                 }
1410                 if (ret < 0) {
1411                         goto error;
1412                 }
1413                 ret = open(fullpath, O_RDONLY);
1414                 if (ret < 0) {
1415                         PERROR("Relay opening trace file");
1416                         goto error;
1417                 }
1418                 stream->read_fd = ret;
1419         }
1420
1421         if (!stream->ctf_trace->metadata_received ||
1422                         stream->ctf_trace->metadata_received >
1423                         stream->ctf_trace->metadata_sent) {
1424                 reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1425                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1426                 goto send_reply;
1427         }
1428
1429         len = be32toh(get_packet_info.len);
1430         data = zmalloc(len);
1431         if (!data) {
1432                 PERROR("relay data zmalloc");
1433                 goto error;
1434         }
1435
1436         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1437         if (ret < 0) {
1438                 /*
1439                  * If the read fd was closed by the streaming side, the
1440                  * abort_flag will be set to 1, otherwise it is an error.
1441                  */
1442                 if (stream->abort_flag == 0) {
1443                         PERROR("lseek");
1444                         goto error;
1445                 }
1446                 reply.status = htobe32(VIEWER_GET_PACKET_EOF);
1447                 goto send_reply;
1448         }
1449         read_len = lttng_read(stream->read_fd, data, len);
1450         if (read_len < len) {
1451                 /*
1452                  * If the read fd was closed by the streaming side, the
1453                  * abort_flag will be set to 1, otherwise it is an error.
1454                  */
1455                 if (stream->abort_flag == 0) {
1456                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1457                                         stream->read_fd,
1458                                         be64toh(get_packet_info.offset));
1459                         goto error;
1460                 } else {
1461                         reply.status = htobe32(VIEWER_GET_PACKET_EOF);
1462                         goto send_reply;
1463                 }
1464         }
1465         reply.status = htobe32(VIEWER_GET_PACKET_OK);
1466         reply.len = htobe32(len);
1467         send_data = 1;
1468         goto send_reply;
1469
1470 error:
1471         reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1472
1473 send_reply:
1474         reply.flags = htobe32(reply.flags);
1475
1476         health_code_update();
1477         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1478         if (ret < 0) {
1479                 ERR("Relay data header to viewer");
1480                 goto end_unlock;
1481         }
1482         health_code_update();
1483
1484         if (send_data) {
1485                 health_code_update();
1486                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1487                 if (ret < 0) {
1488                         ERR("Relay send data to viewer");
1489                         goto end_unlock;
1490                 }
1491                 health_code_update();
1492         }
1493
1494         DBG("Sent %u bytes for stream %" PRIu64, len,
1495                         be64toh(get_packet_info.stream_id));
1496
1497 end_unlock:
1498         free(data);
1499         rcu_read_unlock();
1500
1501 end:
1502         return ret;
1503 }
1504
1505 /*
1506  * Send the session's metadata
1507  *
1508  * Return 0 on success else a negative value.
1509  */
1510 static
1511 int viewer_get_metadata(struct relay_command *cmd)
1512 {
1513         int ret = 0;
1514         ssize_t read_len;
1515         uint64_t len = 0;
1516         char *data = NULL;
1517         struct lttng_viewer_get_metadata request;
1518         struct lttng_viewer_metadata_packet reply;
1519         struct relay_viewer_stream *stream;
1520
1521         assert(cmd);
1522
1523         DBG("Relay get metadata");
1524
1525         if (cmd->version_check_done == 0) {
1526                 ERR("Trying to get metadata before version check");
1527                 ret = -1;
1528                 goto end;
1529         }
1530
1531         health_code_update();
1532         ret = cmd->sock->ops->recvmsg(cmd->sock, &request,
1533                         sizeof(request), 0);
1534         if (ret < 0 || ret != sizeof(request)) {
1535                 ret = -1;
1536                 ERR("Relay didn't receive the whole packet");
1537                 goto end;
1538         }
1539         health_code_update();
1540
1541         rcu_read_lock();
1542         stream = live_find_viewer_stream_by_id(be64toh(request.stream_id));
1543         if (!stream || !stream->metadata_flag) {
1544                 ERR("Invalid metadata stream");
1545                 goto error;
1546         }
1547         assert(stream->ctf_trace);
1548         assert(stream->ctf_trace->metadata_sent <=
1549                         stream->ctf_trace->metadata_received);
1550
1551         len = stream->ctf_trace->metadata_received -
1552                 stream->ctf_trace->metadata_sent;
1553         if (len == 0) {
1554                 reply.status = htobe32(VIEWER_NO_NEW_METADATA);
1555                 goto send_reply;
1556         }
1557
1558         /* first time, we open the metadata file */
1559         if (stream->read_fd < 0) {
1560                 char fullpath[PATH_MAX];
1561
1562                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1563                                 stream->channel_name);
1564                 if (ret < 0) {
1565                         goto error;
1566                 }
1567                 ret = open(fullpath, O_RDONLY);
1568                 if (ret < 0) {
1569                         PERROR("Relay opening metadata file");
1570                         goto error;
1571                 }
1572                 stream->read_fd = ret;
1573         }
1574
1575         reply.len = htobe64(len);
1576         data = zmalloc(len);
1577         if (!data) {
1578                 PERROR("viewer metadata zmalloc");
1579                 goto error;
1580         }
1581
1582         read_len = lttng_read(stream->read_fd, data, len);
1583         if (read_len < len) {
1584                 PERROR("Relay reading metadata file");
1585                 goto error;
1586         }
1587         stream->ctf_trace->metadata_sent += read_len;
1588         reply.status = htobe32(VIEWER_METADATA_OK);
1589         goto send_reply;
1590
1591 error:
1592         reply.status = htobe32(VIEWER_METADATA_ERR);
1593
1594 send_reply:
1595         health_code_update();
1596         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1597         if (ret < 0) {
1598                 ERR("Relay data header to viewer");
1599                 goto end_unlock;
1600         }
1601         health_code_update();
1602
1603         if (len > 0) {
1604                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1605                 if (ret < 0) {
1606                         ERR("Relay send data to viewer");
1607                         goto end_unlock;
1608                 }
1609         }
1610
1611         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1612                         be64toh(request.stream_id));
1613
1614         DBG("Metadata sent");
1615
1616 end_unlock:
1617         free(data);
1618         rcu_read_unlock();
1619 end:
1620         return ret;
1621 }
1622
1623 /*
1624  * live_relay_unknown_command: send -1 if received unknown command
1625  */
1626 static
1627 void live_relay_unknown_command(struct relay_command *cmd)
1628 {
1629         struct lttcomm_relayd_generic_reply reply;
1630         int ret;
1631
1632         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1633         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1634                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1635         if (ret < 0) {
1636                 ERR("Relay sending unknown command");
1637         }
1638 }
1639
1640 /*
1641  * Process the commands received on the control socket
1642  */
1643 static
1644 int process_control(struct lttng_viewer_cmd *recv_hdr,
1645                 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1646 {
1647         int ret = 0;
1648
1649         switch (be32toh(recv_hdr->cmd)) {
1650         case VIEWER_CONNECT:
1651                 ret = viewer_connect(cmd);
1652                 break;
1653         case VIEWER_LIST_SESSIONS:
1654                 ret = viewer_list_sessions(cmd, sessions_ht);
1655                 break;
1656         case VIEWER_ATTACH_SESSION:
1657                 ret = viewer_attach_session(cmd, sessions_ht);
1658                 break;
1659         case VIEWER_GET_NEXT_INDEX:
1660                 ret = viewer_get_next_index(cmd, sessions_ht);
1661                 break;
1662         case VIEWER_GET_PACKET:
1663                 ret = viewer_get_packet(cmd);
1664                 break;
1665         case VIEWER_GET_METADATA:
1666                 ret = viewer_get_metadata(cmd);
1667                 break;
1668         default:
1669                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1670                 live_relay_unknown_command(cmd);
1671                 ret = -1;
1672                 goto end;
1673         }
1674
1675 end:
1676         return ret;
1677 }
1678
1679 static
1680 void cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1681 {
1682         int ret;
1683
1684         assert(events);
1685
1686         lttng_poll_del(events, pollfd);
1687
1688         ret = close(pollfd);
1689         if (ret < 0) {
1690                 ERR("Closing pollfd %d", pollfd);
1691         }
1692 }
1693
1694 /*
1695  * Create and add connection to the given hash table.
1696  *
1697  * Return poll add value or else -1 on error.
1698  */
1699 static
1700 int add_connection(int fd, struct lttng_poll_event *events,
1701                 struct lttng_ht *relay_connections_ht)
1702 {
1703         int ret;
1704         struct relay_command *relay_connection;
1705
1706         assert(events);
1707         assert(relay_connections_ht);
1708
1709         relay_connection = zmalloc(sizeof(struct relay_command));
1710         if (relay_connection == NULL) {
1711                 PERROR("Relay command zmalloc");
1712                 goto error;
1713         }
1714
1715         ret = lttng_read(fd, relay_connection, sizeof(*relay_connection));
1716         if (ret < sizeof(*relay_connection)) {
1717                 PERROR("read relay cmd pipe");
1718                 goto error_read;
1719         }
1720
1721         lttng_ht_node_init_ulong(&relay_connection->sock_n,
1722                         (unsigned long) relay_connection->sock->fd);
1723         rcu_read_lock();
1724         lttng_ht_add_unique_ulong(relay_connections_ht,
1725                         &relay_connection->sock_n);
1726         rcu_read_unlock();
1727
1728         return lttng_poll_add(events, relay_connection->sock->fd,
1729                         LPOLLIN | LPOLLRDHUP);
1730
1731 error_read:
1732         free(relay_connection);
1733 error:
1734         return -1;
1735 }
1736
1737 static
1738 void deferred_free_connection(struct rcu_head *head)
1739 {
1740         struct relay_command *relay_connection =
1741                 caa_container_of(head, struct relay_command, rcu_node);
1742
1743         if (relay_connection->session &&
1744                         relay_connection->session->viewer_attached > 0) {
1745                 relay_connection->session->viewer_attached--;
1746         }
1747         lttcomm_destroy_sock(relay_connection->sock);
1748         free(relay_connection);
1749 }
1750
1751 /*
1752  * Delete all streams for a specific session ID.
1753  */
1754 static
1755 void viewer_del_streams(uint64_t session_id)
1756 {
1757         struct relay_viewer_stream *stream;
1758         struct lttng_ht_iter iter;
1759
1760         rcu_read_lock();
1761         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
1762                         stream_n.node) {
1763                 health_code_update();
1764
1765                 if (stream->session_id != session_id) {
1766                         continue;
1767                 }
1768
1769                 delete_viewer_stream(stream);
1770                 assert(stream->ctf_trace);
1771
1772                 if (stream->metadata_flag) {
1773                         /*
1774                          * The metadata viewer stream is destroyed once the refcount on the
1775                          * ctf trace goes to 0 in the destroy stream function thus there is
1776                          * no explicit call to that function here.
1777                          */
1778                         stream->ctf_trace->metadata_sent = 0;
1779                         stream->ctf_trace->viewer_metadata_stream = NULL;
1780                 } else {
1781                         destroy_viewer_stream(stream);
1782                 }
1783         }
1784         rcu_read_unlock();
1785 }
1786
1787 /*
1788  * Delete and free a connection.
1789  *
1790  * RCU read side lock MUST be acquired.
1791  */
1792 static
1793 void del_connection(struct lttng_ht *relay_connections_ht,
1794                 struct lttng_ht_iter *iter, struct relay_command *relay_connection)
1795 {
1796         int ret;
1797
1798         assert(relay_connections_ht);
1799         assert(iter);
1800         assert(relay_connection);
1801
1802         DBG("Cleaning connection of session ID %" PRIu64,
1803                         relay_connection->session_id);
1804
1805         ret = lttng_ht_del(relay_connections_ht, iter);
1806         assert(!ret);
1807
1808         viewer_del_streams(relay_connection->session_id);
1809
1810         call_rcu(&relay_connection->rcu_node, deferred_free_connection);
1811 }
1812
1813 /*
1814  * This thread does the actual work
1815  */
1816 static
1817 void *thread_worker(void *data)
1818 {
1819         int ret, err = -1;
1820         uint32_t nb_fd;
1821         struct relay_command *relay_connection;
1822         struct lttng_poll_event events;
1823         struct lttng_ht *relay_connections_ht;
1824         struct lttng_ht_node_ulong *node;
1825         struct lttng_ht_iter iter;
1826         struct lttng_viewer_cmd recv_hdr;
1827         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1828         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1829
1830         DBG("[thread] Live viewer relay worker started");
1831
1832         rcu_register_thread();
1833
1834         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1835
1836         /* table of connections indexed on socket */
1837         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1838         if (!relay_connections_ht) {
1839                 goto relay_connections_ht_error;
1840         }
1841
1842         ret = create_thread_poll_set(&events, 2);
1843         if (ret < 0) {
1844                 goto error_poll_create;
1845         }
1846
1847         ret = lttng_poll_add(&events, live_relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1848         if (ret < 0) {
1849                 goto error;
1850         }
1851
1852 restart:
1853         while (1) {
1854                 int i;
1855
1856                 health_code_update();
1857
1858                 /* Infinite blocking call, waiting for transmission */
1859                 DBG3("Relayd live viewer worker thread polling...");
1860                 health_poll_entry();
1861                 ret = lttng_poll_wait(&events, -1);
1862                 health_poll_exit();
1863                 if (ret < 0) {
1864                         /*
1865                          * Restart interrupted system call.
1866                          */
1867                         if (errno == EINTR) {
1868                                 goto restart;
1869                         }
1870                         goto error;
1871                 }
1872
1873                 nb_fd = ret;
1874
1875                 /*
1876                  * Process control. The control connection is prioritised so we don't
1877                  * starve it with high throughput tracing data on the data
1878                  * connection.
1879                  */
1880                 for (i = 0; i < nb_fd; i++) {
1881                         /* Fetch once the poll data */
1882                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1883                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1884
1885                         health_code_update();
1886
1887                         /* Thread quit pipe has been closed. Killing thread. */
1888                         ret = check_thread_quit_pipe(pollfd, revents);
1889                         if (ret) {
1890                                 err = 0;
1891                                 goto exit;
1892                         }
1893
1894                         /* Inspect the relay cmd pipe for new connection */
1895                         if (pollfd == live_relay_cmd_pipe[0]) {
1896                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1897                                         ERR("Relay live pipe error");
1898                                         goto error;
1899                                 } else if (revents & LPOLLIN) {
1900                                         DBG("Relay live viewer command received");
1901                                         ret = add_connection(live_relay_cmd_pipe[0],
1902                                                         &events, relay_connections_ht);
1903                                         if (ret < 0) {
1904                                                 goto error;
1905                                         }
1906                                 }
1907                         } else if (revents) {
1908                                 rcu_read_lock();
1909                                 lttng_ht_lookup(relay_connections_ht,
1910                                                 (void *)((unsigned long) pollfd), &iter);
1911                                 node = lttng_ht_iter_get_node_ulong(&iter);
1912                                 if (node == NULL) {
1913                                         DBG2("Relay viewer sock %d not found", pollfd);
1914                                         rcu_read_unlock();
1915                                         goto error;
1916                                 }
1917                                 relay_connection = caa_container_of(node, struct relay_command,
1918                                                 sock_n);
1919
1920                                 if (revents & (LPOLLERR)) {
1921                                         cleanup_poll_connection(&events, pollfd);
1922                                         del_connection(relay_connections_ht, &iter,
1923                                                         relay_connection);
1924                                 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1925                                         DBG("Viewer socket %d hung up", pollfd);
1926                                         cleanup_poll_connection(&events, pollfd);
1927                                         del_connection(relay_connections_ht, &iter,
1928                                                         relay_connection);
1929                                 } else if (revents & LPOLLIN) {
1930                                         ret = relay_connection->sock->ops->recvmsg(
1931                                                         relay_connection->sock, &recv_hdr,
1932                                                         sizeof(struct lttng_viewer_cmd),
1933                                                         0);
1934                                         /* connection closed */
1935                                         if (ret <= 0) {
1936                                                 cleanup_poll_connection(&events, pollfd);
1937                                                 del_connection(relay_connections_ht, &iter,
1938                                                                 relay_connection);
1939                                                 DBG("Viewer control connection closed with %d",
1940                                                                 pollfd);
1941                                         } else {
1942                                                 if (relay_connection->session) {
1943                                                         DBG2("Relay viewer worker receiving data for "
1944                                                                         "session: %" PRIu64,
1945                                                                         relay_connection->session->id);
1946                                                 }
1947                                                 ret = process_control(&recv_hdr, relay_connection,
1948                                                                 sessions_ht);
1949                                                 if (ret < 0) {
1950                                                         /* Clear the session on error. */
1951                                                         cleanup_poll_connection(&events, pollfd);
1952                                                         del_connection(relay_connections_ht, &iter,
1953                                                                         relay_connection);
1954                                                         DBG("Viewer connection closed with %d", pollfd);
1955                                                 }
1956                                         }
1957                                 }
1958                                 rcu_read_unlock();
1959                         }
1960                 }
1961         }
1962
1963 exit:
1964 error:
1965         lttng_poll_clean(&events);
1966
1967         /* empty the hash table and free the memory */
1968         rcu_read_lock();
1969         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
1970                 health_code_update();
1971
1972                 node = lttng_ht_iter_get_node_ulong(&iter);
1973                 if (!node) {
1974                         continue;
1975                 }
1976
1977                 relay_connection = caa_container_of(node, struct relay_command,
1978                                 sock_n);
1979                 del_connection(relay_connections_ht, &iter, relay_connection);
1980         }
1981         rcu_read_unlock();
1982 error_poll_create:
1983         lttng_ht_destroy(relay_connections_ht);
1984 relay_connections_ht_error:
1985         /* Close relay cmd pipes */
1986         utils_close_pipe(live_relay_cmd_pipe);
1987         if (err) {
1988                 DBG("Viewer worker thread exited with error");
1989         }
1990         DBG("Viewer worker thread cleanup complete");
1991         if (err) {
1992                 health_error();
1993                 ERR("Health error occurred in %s", __func__);
1994         }
1995         health_unregister(health_relayd);
1996         stop_threads();
1997         rcu_unregister_thread();
1998         return NULL;
1999 }
2000
2001 /*
2002  * Create the relay command pipe to wake thread_manage_apps.
2003  * Closed in cleanup().
2004  */
2005 static int create_relay_cmd_pipe(void)
2006 {
2007         int ret;
2008
2009         ret = utils_create_pipe_cloexec(live_relay_cmd_pipe);
2010
2011         return ret;
2012 }
2013
2014 void live_stop_threads(void)
2015 {
2016         int ret;
2017         void *status;
2018
2019         stop_threads();
2020
2021         ret = pthread_join(live_listener_thread, &status);
2022         if (ret != 0) {
2023                 PERROR("pthread_join live listener");
2024                 goto error;     /* join error, exit without cleanup */
2025         }
2026
2027         ret = pthread_join(live_worker_thread, &status);
2028         if (ret != 0) {
2029                 PERROR("pthread_join live worker");
2030                 goto error;     /* join error, exit without cleanup */
2031         }
2032
2033         ret = pthread_join(live_dispatcher_thread, &status);
2034         if (ret != 0) {
2035                 PERROR("pthread_join live dispatcher");
2036                 goto error;     /* join error, exit without cleanup */
2037         }
2038
2039         cleanup();
2040
2041 error:
2042         return;
2043 }
2044
2045 /*
2046  * main
2047  */
2048 int live_start_threads(struct lttng_uri *uri,
2049                 struct relay_local_data *relay_ctx, int quit_pipe[2])
2050 {
2051         int ret = 0;
2052         void *status;
2053         int is_root;
2054
2055         assert(uri);
2056         live_uri = uri;
2057
2058         live_thread_quit_pipe[0] = quit_pipe[0];
2059         live_thread_quit_pipe[1] = quit_pipe[1];
2060
2061         /* Check if daemon is UID = 0 */
2062         is_root = !getuid();
2063
2064         if (!is_root) {
2065                 if (live_uri->port < 1024) {
2066                         ERR("Need to be root to use ports < 1024");
2067                         ret = -1;
2068                         goto exit;
2069                 }
2070         }
2071
2072         /* Setup the thread apps communication pipe. */
2073         if ((ret = create_relay_cmd_pipe()) < 0) {
2074                 goto exit;
2075         }
2076
2077         /* Init relay command queue. */
2078         cds_wfq_init(&viewer_cmd_queue.queue);
2079
2080         /* Set up max poll set size */
2081         lttng_poll_set_max_size();
2082
2083         /* Setup the dispatcher thread */
2084         ret = pthread_create(&live_dispatcher_thread, NULL,
2085                         thread_dispatcher, (void *) NULL);
2086         if (ret != 0) {
2087                 PERROR("pthread_create viewer dispatcher");
2088                 goto exit_dispatcher;
2089         }
2090
2091         /* Setup the worker thread */
2092         ret = pthread_create(&live_worker_thread, NULL,
2093                         thread_worker, relay_ctx);
2094         if (ret != 0) {
2095                 PERROR("pthread_create viewer worker");
2096                 goto exit_worker;
2097         }
2098
2099         /* Setup the listener thread */
2100         ret = pthread_create(&live_listener_thread, NULL,
2101                         thread_listener, (void *) NULL);
2102         if (ret != 0) {
2103                 PERROR("pthread_create viewer listener");
2104                 goto exit_listener;
2105         }
2106
2107         ret = 0;
2108         goto end;
2109
2110 exit_listener:
2111         ret = pthread_join(live_listener_thread, &status);
2112         if (ret != 0) {
2113                 PERROR("pthread_join live listener");
2114                 goto error;     /* join error, exit without cleanup */
2115         }
2116
2117 exit_worker:
2118         ret = pthread_join(live_worker_thread, &status);
2119         if (ret != 0) {
2120                 PERROR("pthread_join live worker");
2121                 goto error;     /* join error, exit without cleanup */
2122         }
2123
2124 exit_dispatcher:
2125         ret = pthread_join(live_dispatcher_thread, &status);
2126         if (ret != 0) {
2127                 PERROR("pthread_join live dispatcher");
2128                 goto error;     /* join error, exit without cleanup */
2129         }
2130
2131 exit:
2132         cleanup();
2133
2134 end:
2135 error:
2136         return ret;
2137 }