src/bin/lttng-relayd/live.c

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