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 & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 546                                 ERR("socket poll error");
 547                                 goto error;
 548                         } else if (revents & LPOLLIN) {
 549                                 /*
 550                                  * A new connection is requested, therefore a
 551                                  * viewer connection is allocated in this
 552                                  * thread, enqueued to a global queue and
 553                                  * dequeued (and freed) in the worker thread.
 554                                  */
 555                                 int val = 1;
 556                                 struct relay_connection *new_conn;
 557                                 struct lttcomm_sock *newsock;
 558
 559                                 newsock = live_control_sock->ops->accept(live_control_sock);
 560                                 if (!newsock) {
 561                                         PERROR("accepting control sock");
 562                                         goto error;
 563                                 }
 564                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 565
 566                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 567                                                 sizeof(val));
 568                                 if (ret < 0) {
 569                                         PERROR("setsockopt inet");
 570                                         lttcomm_destroy_sock(newsock);
 571                                         goto error;
 572                                 }
 573                                 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
 574                                 if (!new_conn) {
 575                                         lttcomm_destroy_sock(newsock);
 576                                         goto error;
 577                                 }
 578                                 /* Ownership assumed by the connection. */
 579                                 newsock = NULL;
 580
 581                                 /* Enqueue request for the dispatcher thread. */
 582                                 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
 583                                                  &new_conn->qnode);
 584
 585                                 /*
 586                                  * Wake the dispatch queue futex.
 587                                  * Implicit memory barrier with the
 588                                  * exchange in cds_wfcq_enqueue.
 589                                  */
 590                                 futex_nto1_wake(&viewer_conn_queue.futex);
 591                         }
 592                 }
 593         }
 594
 595 exit:
 596 error:
 597 error_poll_add:
 598 error_testpoint:
 599         lttng_poll_clean(&events);
 600 error_create_poll:
 601         if (live_control_sock->fd >= 0) {
 602                 ret = live_control_sock->ops->close(live_control_sock);
 603                 if (ret) {
 604                         PERROR("close");
 605                 }
 606         }
 607         lttcomm_destroy_sock(live_control_sock);
 608 error_sock_control:
 609         if (err) {
 610                 health_error();
 611                 DBG("Live viewer listener thread exited with error");
 612         }
 613         health_unregister(health_relayd);
 614         DBG("Live viewer listener thread cleanup complete");
 615         if (lttng_relay_stop_threads()) {
 616                 ERR("Error stopping threads");
 617         }
 618         return NULL;
 619 }
 620
 621 /*
 622  * This thread manages the dispatching of the requests to worker threads
 623  */
 624 static
 625 void *thread_dispatcher(void *data)
 626 {
 627         int err = -1;
 628         ssize_t ret;
 629         struct cds_wfcq_node *node;
 630         struct relay_connection *conn = NULL;
 631
 632         DBG("[thread] Live viewer relay dispatcher started");
 633
 634         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 635
 636         if (testpoint(relayd_thread_live_dispatcher)) {
 637                 goto error_testpoint;
 638         }
 639
 640         health_code_update();
 641
 642         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 643                 health_code_update();
 644
 645                 /* Atomically prepare the queue futex */
 646                 futex_nto1_prepare(&viewer_conn_queue.futex);
 647
 648                 do {
 649                         health_code_update();
 650
 651                         /* Dequeue commands */
 652                         node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
 653                                                          &viewer_conn_queue.tail);
 654                         if (node == NULL) {
 655                                 DBG("Woken up but nothing in the live-viewer "
 656                                                 "relay command queue");
 657                                 /* Continue thread execution */
 658                                 break;
 659                         }
 660                         conn = caa_container_of(node, struct relay_connection, qnode);
 661                         DBG("Dispatching viewer request waiting on sock %d",
 662                                         conn->sock->fd);
 663
 664                         /*
 665                          * Inform worker thread of the new request. This
 666                          * call is blocking so we can be assured that
 667                          * the data will be read at some point in time
 668                          * or wait to the end of the world :)
 669                          */
 670                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 671                         if (ret < 0) {
 672                                 PERROR("write conn pipe");
 673                                 connection_put(conn);
 674                                 goto error;
 675                         }
 676                 } while (node != NULL);
 677
 678                 /* Futex wait on queue. Blocking call on futex() */
 679                 health_poll_entry();
 680                 futex_nto1_wait(&viewer_conn_queue.futex);
 681                 health_poll_exit();
 682         }
 683
 684         /* Normal exit, no error */
 685         err = 0;
 686
 687 error:
 688 error_testpoint:
 689         if (err) {
 690                 health_error();
 691                 ERR("Health error occurred in %s", __func__);
 692         }
 693         health_unregister(health_relayd);
 694         DBG("Live viewer dispatch thread dying");
 695         if (lttng_relay_stop_threads()) {
 696                 ERR("Error stopping threads");
 697         }
 698         return NULL;
 699 }
 700
 701 /*
 702  * Establish connection with the viewer and check the versions.
 703  *
 704  * Return 0 on success or else negative value.
 705  */
 706 static
 707 int viewer_connect(struct relay_connection *conn)
 708 {
 709         int ret;
 710         struct lttng_viewer_connect reply, msg;
 711
 712         conn->version_check_done = 1;
 713
 714         health_code_update();
 715
 716         DBG("Viewer is establishing a connection to the relayd.");
 717
 718         ret = recv_request(conn->sock, &msg, sizeof(msg));
 719         if (ret < 0) {
 720                 goto end;
 721         }
 722
 723         health_code_update();
 724
 725         memset(&reply, 0, sizeof(reply));
 726         reply.major = RELAYD_VERSION_COMM_MAJOR;
 727         reply.minor = RELAYD_VERSION_COMM_MINOR;
 728
 729         /* Major versions must be the same */
 730         if (reply.major != be32toh(msg.major)) {
 731                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 732                                 reply.major, be32toh(msg.major));
 733                 ret = -1;
 734                 goto end;
 735         }
 736
 737         conn->major = reply.major;
 738         /* We adapt to the lowest compatible version */
 739         if (reply.minor <= be32toh(msg.minor)) {
 740                 conn->minor = reply.minor;
 741         } else {
 742                 conn->minor = be32toh(msg.minor);
 743         }
 744
 745         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 746                 conn->type = RELAY_VIEWER_COMMAND;
 747         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 748                 conn->type = RELAY_VIEWER_NOTIFICATION;
 749         } else {
 750                 ERR("Unknown connection type : %u", be32toh(msg.type));
 751                 ret = -1;
 752                 goto end;
 753         }
 754
 755         reply.major = htobe32(reply.major);
 756         reply.minor = htobe32(reply.minor);
 757         if (conn->type == RELAY_VIEWER_COMMAND) {
 758                 /*
 759                  * Increment outside of htobe64 macro, because the argument can
 760                  * be used more than once within the macro, and thus the
 761                  * operation may be undefined.
 762                  */
 763                 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
 764                 last_relay_viewer_session_id++;
 765                 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
 766                 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
 767         }
 768
 769         health_code_update();
 770
 771         ret = send_response(conn->sock, &reply, sizeof(reply));
 772         if (ret < 0) {
 773                 goto end;
 774         }
 775
 776         health_code_update();
 777
 778         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 779         ret = 0;
 780
 781 end:
 782         return ret;
 783 }
 784
 785 /*
 786  * Send the viewer the list of current sessions.
 787  * We need to create a copy of the hash table content because otherwise
 788  * we cannot assume the number of entries stays the same between getting
 789  * the number of HT elements and iteration over the HT.
 790  *
 791  * Return 0 on success or else a negative value.
 792  */
 793 static
 794 int viewer_list_sessions(struct relay_connection *conn)
 795 {
 796         int ret;
 797         struct lttng_viewer_list_sessions session_list;
 798         struct lttng_ht_iter iter;
 799         struct relay_session *session;
 800         struct lttng_viewer_session *send_session_buf = NULL;
 801         uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
 802         uint32_t count = 0;
 803
 804         DBG("List sessions received");
 805
 806         send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
 807         if (!send_session_buf) {
 808                 return -1;
 809         }
 810
 811         rcu_read_lock();
 812         cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
 813                         session_n.node) {
 814                 struct lttng_viewer_session *send_session;
 815
 816                 health_code_update();
 817
 818                 if (count >= buf_count) {
 819                         struct lttng_viewer_session *newbuf;
 820                         uint32_t new_buf_count = buf_count << 1;
 821
 822                         newbuf = realloc(send_session_buf,
 823                                 new_buf_count * sizeof(*send_session_buf));
 824                         if (!newbuf) {
 825                                 ret = -1;
 826                                 rcu_read_unlock();
 827                                 goto end_free;
 828                         }
 829                         send_session_buf = newbuf;
 830                         buf_count = new_buf_count;
 831                 }
 832                 send_session = &send_session_buf[count];
 833                 strncpy(send_session->session_name, session->session_name,
 834                                 sizeof(send_session->session_name));
 835                 strncpy(send_session->hostname, session->hostname,
 836                                 sizeof(send_session->hostname));
 837                 send_session->id = htobe64(session->id);
 838                 send_session->live_timer = htobe32(session->live_timer);
 839                 if (session->viewer_attached) {
 840                         send_session->clients = htobe32(1);
 841                 } else {
 842                         send_session->clients = htobe32(0);
 843                 }
 844                 send_session->streams = htobe32(session->stream_count);
 845                 count++;
 846         }
 847         rcu_read_unlock();
 848
 849         session_list.sessions_count = htobe32(count);
 850
 851         health_code_update();
 852
 853         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 854         if (ret < 0) {
 855                 goto end_free;
 856         }
 857
 858         health_code_update();
 859
 860         ret = send_response(conn->sock, send_session_buf,
 861                         count * sizeof(*send_session_buf));
 862         if (ret < 0) {
 863                 goto end_free;
 864         }
 865         health_code_update();
 866
 867         ret = 0;
 868 end_free:
 869         free(send_session_buf);
 870         return ret;
 871 }
 872
 873 /*
 874  * Send the viewer the list of current streams.
 875  */
 876 static
 877 int viewer_get_new_streams(struct relay_connection *conn)
 878 {
 879         int ret, send_streams = 0;
 880         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
 881         struct lttng_viewer_new_streams_request request;
 882         struct lttng_viewer_new_streams_response response;
 883         struct relay_session *session;
 884         uint64_t session_id;
 885         bool closed = false;
 886
 887         assert(conn);
 888
 889         DBG("Get new streams received");
 890
 891         health_code_update();
 892
 893         /* Receive the request from the connected client. */
 894         ret = recv_request(conn->sock, &request, sizeof(request));
 895         if (ret < 0) {
 896                 goto error;
 897         }
 898         session_id = be64toh(request.session_id);
 899
 900         health_code_update();
 901
 902         memset(&response, 0, sizeof(response));
 903
 904         session = session_get_by_id(session_id);
 905         if (!session) {
 906                 DBG("Relay session %" PRIu64 " not found", session_id);
 907                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 908                 goto send_reply;
 909         }
 910
 911         if (!viewer_session_is_attached(conn->viewer_session, session)) {
 912                 send_streams = 0;
 913                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 914                 goto send_reply;
 915         }
 916
 917         send_streams = 1;
 918         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 919
 920         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
 921                         &nb_created, &closed);
 922         if (ret < 0) {
 923                 goto end_put_session;
 924         }
 925         /* Only send back the newly created streams with the unsent ones. */
 926         nb_streams = nb_created + nb_unsent;
 927         response.streams_count = htobe32(nb_streams);
 928
 929         /*
 930          * If the session is closed, HUP when there are no more streams
 931          * with data.
 932          */
 933         if (closed && nb_total == 0) {
 934                 send_streams = 0;
 935                 response.streams_count = 0;
 936                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 937                 goto send_reply;
 938         }
 939
 940 send_reply:
 941         health_code_update();
 942         ret = send_response(conn->sock, &response, sizeof(response));
 943         if (ret < 0) {
 944                 goto end_put_session;
 945         }
 946         health_code_update();
 947
 948         /*
 949          * Unknown or empty session, just return gracefully, the viewer
 950          * knows what is happening.
 951          */
 952         if (!send_streams || !nb_streams) {
 953                 ret = 0;
 954                 goto end_put_session;
 955         }
 956
 957         /*
 958          * Send stream and *DON'T* ignore the sent flag so every viewer
 959          * streams that were not sent from that point will be sent to
 960          * the viewer.
 961          */
 962         ret = send_viewer_streams(conn->sock, session, 0);
 963         if (ret < 0) {
 964                 goto end_put_session;
 965         }
 966
 967 end_put_session:
 968         if (session) {
 969                 session_put(session);
 970         }
 971 error:
 972         return ret;
 973 }
 974
 975 /*
 976  * Send the viewer the list of current sessions.
 977  */
 978 static
 979 int viewer_attach_session(struct relay_connection *conn)
 980 {
 981         int send_streams = 0;
 982         ssize_t ret;
 983         uint32_t nb_streams = 0;
 984         enum lttng_viewer_seek seek_type;
 985         struct lttng_viewer_attach_session_request request;
 986         struct lttng_viewer_attach_session_response response;
 987         struct relay_session *session = NULL;
 988         bool closed = false;
 989
 990         assert(conn);
 991
 992         health_code_update();
 993
 994         /* Receive the request from the connected client. */
 995         ret = recv_request(conn->sock, &request, sizeof(request));
 996         if (ret < 0) {
 997                 goto error;
 998         }
 999
1000         health_code_update();
1001
1002         memset(&response, 0, sizeof(response));
1003
1004         if (!conn->viewer_session) {
1005                 DBG("Client trying to attach before creating a live viewer session");
1006                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1007                 goto send_reply;
1008         }
1009
1010         session = session_get_by_id(be64toh(request.session_id));
1011         if (!session) {
1012                 DBG("Relay session %" PRIu64 " not found",
1013                                 be64toh(request.session_id));
1014                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1015                 goto send_reply;
1016         }
1017         DBG("Attach session ID %" PRIu64 " received",
1018                 be64toh(request.session_id));
1019
1020         if (session->live_timer == 0) {
1021                 DBG("Not live session");
1022                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1023                 goto send_reply;
1024         }
1025
1026         send_streams = 1;
1027         ret = viewer_session_attach(conn->viewer_session, session);
1028         if (ret) {
1029                 DBG("Already a viewer attached");
1030                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1031                 goto send_reply;
1032         }
1033
1034         switch (be32toh(request.seek)) {
1035         case LTTNG_VIEWER_SEEK_BEGINNING:
1036         case LTTNG_VIEWER_SEEK_LAST:
1037                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1038                 seek_type = be32toh(request.seek);
1039                 break;
1040         default:
1041                 ERR("Wrong seek parameter");
1042                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1043                 send_streams = 0;
1044                 goto send_reply;
1045         }
1046
1047         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1048                         NULL, &closed);
1049         if (ret < 0) {
1050                 goto end_put_session;
1051         }
1052         response.streams_count = htobe32(nb_streams);
1053
1054         /*
1055          * If the session is closed when the viewer is attaching, it
1056          * means some of the streams may have been concurrently removed,
1057          * so we don't allow the viewer to attach, even if there are
1058          * streams available.
1059          */
1060         if (closed) {
1061                 send_streams = 0;
1062                 response.streams_count = 0;
1063                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1064                 goto send_reply;
1065         }
1066
1067 send_reply:
1068         health_code_update();
1069         ret = send_response(conn->sock, &response, sizeof(response));
1070         if (ret < 0) {
1071                 goto end_put_session;
1072         }
1073         health_code_update();
1074
1075         /*
1076          * Unknown or empty session, just return gracefully, the viewer
1077          * knows what is happening.
1078          */
1079         if (!send_streams || !nb_streams) {
1080                 ret = 0;
1081                 goto end_put_session;
1082         }
1083
1084         /* Send stream and ignore the sent flag. */
1085         ret = send_viewer_streams(conn->sock, session, 1);
1086         if (ret < 0) {
1087                 goto end_put_session;
1088         }
1089
1090 end_put_session:
1091         if (session) {
1092                 session_put(session);
1093         }
1094 error:
1095         return ret;
1096 }
1097
1098 /*
1099  * Open the index file if needed for the given vstream.
1100  *
1101  * If an index file is successfully opened, the vstream index_fd set with
1102  * it.
1103  *
1104  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1105  *
1106  * Called with rstream lock held.
1107  */
1108 static int try_open_index(struct relay_viewer_stream *vstream,
1109                 struct relay_stream *rstream)
1110 {
1111         int ret = 0;
1112
1113         if (vstream->index_fd) {
1114                 goto end;
1115         }
1116
1117         /*
1118          * First time, we open the index file and at least one index is ready.
1119          */
1120         if (rstream->index_received_seqcount == 0) {
1121                 ret = -ENOENT;
1122                 goto end;
1123         }
1124         ret = index_open(vstream->path_name, vstream->channel_name,
1125                         vstream->stream->tracefile_count,
1126                         vstream->current_tracefile_id);
1127         if (ret >= 0) {
1128                 vstream->index_fd = stream_fd_create(ret);
1129                 if (!vstream->index_fd) {
1130                         if (close(ret)) {
1131                                 PERROR("close");
1132                         }
1133                         ret = -1;
1134                 } else {
1135                         ret = 0;
1136                 }
1137                 goto end;
1138         }
1139
1140 end:
1141         return ret;
1142 }
1143
1144 /*
1145  * Check the status of the index for the given stream. This function
1146  * updates the index structure if needed and can put (close) the vstream
1147  * in the HUP situation.
1148  *
1149  * Return 0 means that we can proceed with the index. A value of 1 means
1150  * that the index has been updated and is ready to be sent to the
1151  * client. A negative value indicates an error that can't be handled.
1152  *
1153  * Called with rstream lock held.
1154  */
1155 static int check_index_status(struct relay_viewer_stream *vstream,
1156                 struct relay_stream *rstream, struct ctf_trace *trace,
1157                 struct lttng_viewer_index *index)
1158 {
1159         int ret;
1160
1161         if (trace->session->connection_closed
1162                         && rstream->index_received_seqcount
1163                                 == vstream->index_sent_seqcount) {
1164                 /* Last index sent and session connection is closed. */
1165                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1166                 goto hup;
1167         } else if (rstream->beacon_ts_end != -1ULL &&
1168                         rstream->index_received_seqcount
1169                                 == vstream->index_sent_seqcount) {
1170                 /*
1171                  * We've received a synchronization beacon and the last index
1172                  * available has been sent, the index for now is inactive.
1173                  *
1174                  * In this case, we have received a beacon which allows us to
1175                  * inform the client of a time interval during which we can
1176                  * guarantee that there are no events to read (and never will
1177                  * be).
1178                  */
1179                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1180                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1181                 index->stream_id = htobe64(rstream->ctf_stream_id);
1182                 goto index_ready;
1183         } else if (rstream->index_received_seqcount
1184                         == vstream->index_sent_seqcount) {
1185                 /*
1186                  * This checks whether received == sent seqcount. In
1187                  * this case, we have not received a beacon. Therefore,
1188                  * we can only ask the client to retry later.
1189                  */
1190                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1191                 goto index_ready;
1192         } else if (!tracefile_array_seq_in_file(rstream->tfa,
1193                         vstream->current_tracefile_id,
1194                         vstream->index_sent_seqcount)) {
1195                 /*
1196                  * The next index we want to send cannot be read either
1197                  * because we need to perform a rotation, or due to
1198                  * the producer having overwritten its trace file.
1199                  */
1200                 DBG("Viewer stream %" PRIu64 " rotation",
1201                                 vstream->stream->stream_handle);
1202                 ret = viewer_stream_rotate(vstream);
1203                 if (ret < 0) {
1204                         goto end;
1205                 } else if (ret == 1) {
1206                         /* EOF across entire stream. */
1207                         index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1208                         goto hup;
1209                 }
1210                 /*
1211                  * If we have been pushed due to overwrite, it
1212                  * necessarily means there is data that can be read in
1213                  * the stream. If we rotated because we reached the end
1214                  * of a tracefile, it means the following tracefile
1215                  * needs to contain at least one index, else we would
1216                  * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1217                  * viewer. The updated index_sent_seqcount needs to
1218                  * point to a readable index entry now.
1219                  *
1220                  * In the case where we "rotate" on a single file, we
1221                  * can end up in a case where the requested index is
1222                  * still unavailable.
1223                  */
1224                 if (rstream->tracefile_count == 1 &&
1225                                 !tracefile_array_seq_in_file(
1226                                         rstream->tfa,
1227                                         vstream->current_tracefile_id,
1228                                         vstream->index_sent_seqcount)) {
1229                         index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1230                         goto index_ready;
1231                 }
1232                 assert(tracefile_array_seq_in_file(rstream->tfa,
1233                                 vstream->current_tracefile_id,
1234                                 vstream->index_sent_seqcount));
1235         }
1236         /* ret == 0 means successful so we continue. */
1237         ret = 0;
1238 end:
1239         return ret;
1240
1241 hup:
1242         viewer_stream_put(vstream);
1243 index_ready:
1244         return 1;
1245 }
1246
1247 /*
1248  * Send the next index for a stream.
1249  *
1250  * Return 0 on success or else a negative value.
1251  */
1252 static
1253 int viewer_get_next_index(struct relay_connection *conn)
1254 {
1255         int ret;
1256         ssize_t read_ret;
1257         struct lttng_viewer_get_next_index request_index;
1258         struct lttng_viewer_index viewer_index;
1259         struct ctf_packet_index packet_index;
1260         struct relay_viewer_stream *vstream = NULL;
1261         struct relay_stream *rstream = NULL;
1262         struct ctf_trace *ctf_trace = NULL;
1263         struct relay_viewer_stream *metadata_viewer_stream = NULL;
1264
1265         assert(conn);
1266
1267         DBG("Viewer get next index");
1268
1269         memset(&viewer_index, 0, sizeof(viewer_index));
1270         health_code_update();
1271
1272         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1273         if (ret < 0) {
1274                 goto end;
1275         }
1276         health_code_update();
1277
1278         vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1279         if (!vstream) {
1280                 DBG("Client requested index of unknown stream id %" PRIu64,
1281                                 be64toh(request_index.stream_id));
1282                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1283                 goto send_reply;
1284         }
1285
1286         /* Use back. ref. Protected by refcounts. */
1287         rstream = vstream->stream;
1288         ctf_trace = rstream->trace;
1289
1290         /* metadata_viewer_stream may be NULL. */
1291         metadata_viewer_stream =
1292                         ctf_trace_get_viewer_metadata_stream(ctf_trace);
1293
1294         pthread_mutex_lock(&rstream->lock);
1295
1296         /*
1297          * The viewer should not ask for index on metadata stream.
1298          */
1299         if (rstream->is_metadata) {
1300                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1301                 goto send_reply;
1302         }
1303
1304         /* Try to open an index if one is needed for that stream. */
1305         ret = try_open_index(vstream, rstream);
1306         if (ret < 0) {
1307                 if (ret == -ENOENT) {
1308                         /*
1309                          * The index is created only when the first data
1310                          * packet arrives, it might not be ready at the
1311                          * beginning of the session
1312                          */
1313                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1314                 } else {
1315                         /* Unhandled error. */
1316                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1317                 }
1318                 goto send_reply;
1319         }
1320
1321         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1322         if (ret < 0) {
1323                 goto error_put;
1324         } else if (ret == 1) {
1325                 /*
1326                  * We have no index to send and check_index_status has populated
1327                  * viewer_index's status.
1328                  */
1329                 goto send_reply;
1330         }
1331         /* At this point, ret is 0 thus we will be able to read the index. */
1332         assert(!ret);
1333
1334         /*
1335          * vstream->stream_fd may be NULL if it has been closed by
1336          * tracefile rotation, or if we are at the beginning of the
1337          * stream. We open the data stream file here to protect against
1338          * overwrite caused by tracefile rotation (in association with
1339          * unlink performed before overwrite).
1340          */
1341         if (!vstream->stream_fd) {
1342                 char fullpath[PATH_MAX];
1343
1344                 if (vstream->stream->tracefile_count > 0) {
1345                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1346                                         vstream->path_name,
1347                                         vstream->channel_name,
1348                                         vstream->current_tracefile_id);
1349                 } else {
1350                         ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1351                                         vstream->path_name,
1352                                         vstream->channel_name);
1353                 }
1354                 if (ret < 0) {
1355                         goto error_put;
1356                 }
1357                 ret = open(fullpath, O_RDONLY);
1358                 if (ret < 0) {
1359                         PERROR("Relay opening trace file");
1360                         goto error_put;
1361                 }
1362                 vstream->stream_fd = stream_fd_create(ret);
1363                 if (!vstream->stream_fd) {
1364                         if (close(ret)) {
1365                                 PERROR("close");
1366                         }
1367                         goto error_put;
1368                 }
1369         }
1370
1371         ret = check_new_streams(conn);
1372         if (ret < 0) {
1373                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1374                 goto send_reply;
1375         } else if (ret == 1) {
1376                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1377         }
1378
1379         read_ret = lttng_read(vstream->index_fd->fd, &packet_index,
1380                         sizeof(packet_index));
1381         if (read_ret < sizeof(packet_index)) {
1382                 ERR("Relay reading index file %d returned %zd",
1383                         vstream->index_fd->fd, read_ret);
1384                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1385                 goto send_reply;
1386         } else {
1387                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1388                 vstream->index_sent_seqcount++;
1389         }
1390
1391         /*
1392          * Indexes are stored in big endian, no need to switch before sending.
1393          */
1394         DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1395                 rstream->stream_handle,
1396                 be64toh(packet_index.offset));
1397         viewer_index.offset = packet_index.offset;
1398         viewer_index.packet_size = packet_index.packet_size;
1399         viewer_index.content_size = packet_index.content_size;
1400         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1401         viewer_index.timestamp_end = packet_index.timestamp_end;
1402         viewer_index.events_discarded = packet_index.events_discarded;
1403         viewer_index.stream_id = packet_index.stream_id;
1404
1405 send_reply:
1406         if (rstream) {
1407                 pthread_mutex_unlock(&rstream->lock);
1408         }
1409
1410         if (metadata_viewer_stream) {
1411                 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1412                 DBG("get next index metadata check: recv %" PRIu64
1413                                 " sent %" PRIu64,
1414                         metadata_viewer_stream->stream->metadata_received,
1415                         metadata_viewer_stream->metadata_sent);
1416                 if (!metadata_viewer_stream->stream->metadata_received ||
1417                                 metadata_viewer_stream->stream->metadata_received >
1418                                         metadata_viewer_stream->metadata_sent) {
1419                         viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1420                 }
1421                 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1422         }
1423
1424         viewer_index.flags = htobe32(viewer_index.flags);
1425         health_code_update();
1426
1427         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1428         if (ret < 0) {
1429                 goto end;
1430         }
1431         health_code_update();
1432
1433         if (vstream) {
1434                 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1435                                 vstream->index_sent_seqcount,
1436                                 vstream->stream->stream_handle);
1437         }
1438 end:
1439         if (metadata_viewer_stream) {
1440                 viewer_stream_put(metadata_viewer_stream);
1441         }
1442         if (vstream) {
1443                 viewer_stream_put(vstream);
1444         }
1445         return ret;
1446
1447 error_put:
1448         pthread_mutex_unlock(&rstream->lock);
1449         if (metadata_viewer_stream) {
1450                 viewer_stream_put(metadata_viewer_stream);
1451         }
1452         viewer_stream_put(vstream);
1453         return ret;
1454 }
1455
1456 /*
1457  * Send the next index for a stream
1458  *
1459  * Return 0 on success or else a negative value.
1460  */
1461 static
1462 int viewer_get_packet(struct relay_connection *conn)
1463 {
1464         int ret, send_data = 0;
1465         char *data = NULL;
1466         uint32_t len = 0;
1467         ssize_t read_len;
1468         struct lttng_viewer_get_packet get_packet_info;
1469         struct lttng_viewer_trace_packet reply;
1470         struct relay_viewer_stream *vstream = NULL;
1471
1472         DBG2("Relay get data packet");
1473
1474         health_code_update();
1475
1476         ret = recv_request(conn->sock, &get_packet_info,
1477                         sizeof(get_packet_info));
1478         if (ret < 0) {
1479                 goto end;
1480         }
1481         health_code_update();
1482
1483         /* From this point on, the error label can be reached. */
1484         memset(&reply, 0, sizeof(reply));
1485
1486         vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1487         if (!vstream) {
1488                 DBG("Client requested packet of unknown stream id %" PRIu64,
1489                                 be64toh(get_packet_info.stream_id));
1490                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1491                 goto send_reply_nolock;
1492         }
1493
1494         pthread_mutex_lock(&vstream->stream->lock);
1495
1496         len = be32toh(get_packet_info.len);
1497         data = zmalloc(len);
1498         if (!data) {
1499                 PERROR("relay data zmalloc");
1500                 goto error;
1501         }
1502
1503         ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1504                         SEEK_SET);
1505         if (ret < 0) {
1506                 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1507                         be64toh(get_packet_info.offset));
1508                 goto error;
1509         }
1510         read_len = lttng_read(vstream->stream_fd->fd, data, len);
1511         if (read_len < len) {
1512                 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1513                                 vstream->stream_fd->fd,
1514                                 be64toh(get_packet_info.offset));
1515                 goto error;
1516         }
1517         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1518         reply.len = htobe32(len);
1519         send_data = 1;
1520         goto send_reply;
1521
1522 error:
1523         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1524
1525 send_reply:
1526         if (vstream) {
1527                 pthread_mutex_unlock(&vstream->stream->lock);
1528         }
1529 send_reply_nolock:
1530         reply.flags = htobe32(reply.flags);
1531
1532         health_code_update();
1533
1534         ret = send_response(conn->sock, &reply, sizeof(reply));
1535         if (ret < 0) {
1536                 goto end_free;
1537         }
1538         health_code_update();
1539
1540         if (send_data) {
1541                 health_code_update();
1542                 ret = send_response(conn->sock, data, len);
1543                 if (ret < 0) {
1544                         goto end_free;
1545                 }
1546                 health_code_update();
1547         }
1548
1549         DBG("Sent %u bytes for stream %" PRIu64, len,
1550                         be64toh(get_packet_info.stream_id));
1551
1552 end_free:
1553         free(data);
1554 end:
1555         if (vstream) {
1556                 viewer_stream_put(vstream);
1557         }
1558         return ret;
1559 }
1560
1561 /*
1562  * Send the session's metadata
1563  *
1564  * Return 0 on success else a negative value.
1565  */
1566 static
1567 int viewer_get_metadata(struct relay_connection *conn)
1568 {
1569         int ret = 0;
1570         ssize_t read_len;
1571         uint64_t len = 0;
1572         char *data = NULL;
1573         struct lttng_viewer_get_metadata request;
1574         struct lttng_viewer_metadata_packet reply;
1575         struct relay_viewer_stream *vstream = NULL;
1576
1577         assert(conn);
1578
1579         DBG("Relay get metadata");
1580
1581         health_code_update();
1582
1583         ret = recv_request(conn->sock, &request, sizeof(request));
1584         if (ret < 0) {
1585                 goto end;
1586         }
1587         health_code_update();
1588
1589         memset(&reply, 0, sizeof(reply));
1590
1591         vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1592         if (!vstream) {
1593                 /*
1594                  * The metadata stream can be closed by a CLOSE command
1595                  * just before we attach. It can also be closed by
1596                  * per-pid tracing during tracing. Therefore, it is
1597                  * possible that we cannot find this viewer stream.
1598                  * Reply back to the client with an error if we cannot
1599                  * find it.
1600                  */
1601                 DBG("Client requested metadata of unknown stream id %" PRIu64,
1602                                 be64toh(request.stream_id));
1603                 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1604                 goto send_reply;
1605         }
1606         pthread_mutex_lock(&vstream->stream->lock);
1607         if (!vstream->stream->is_metadata) {
1608                 ERR("Invalid metadata stream");
1609                 goto error;
1610         }
1611
1612         assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1613
1614         len = vstream->stream->metadata_received - vstream->metadata_sent;
1615         if (len == 0) {
1616                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1617                 goto send_reply;
1618         }
1619
1620         /* first time, we open the metadata file */
1621         if (!vstream->stream_fd) {
1622                 char fullpath[PATH_MAX];
1623
1624                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1625                                 vstream->channel_name);
1626                 if (ret < 0) {
1627                         goto error;
1628                 }
1629                 ret = open(fullpath, O_RDONLY);
1630                 if (ret < 0) {
1631                         PERROR("Relay opening metadata file");
1632                         goto error;
1633                 }
1634                 vstream->stream_fd = stream_fd_create(ret);
1635                 if (!vstream->stream_fd) {
1636                         if (close(ret)) {
1637                                 PERROR("close");
1638                         }
1639                         goto error;
1640                 }
1641         }
1642
1643         reply.len = htobe64(len);
1644         data = zmalloc(len);
1645         if (!data) {
1646                 PERROR("viewer metadata zmalloc");
1647                 goto error;
1648         }
1649
1650         read_len = lttng_read(vstream->stream_fd->fd, data, len);
1651         if (read_len < len) {
1652                 PERROR("Relay reading metadata file");
1653                 goto error;
1654         }
1655         vstream->metadata_sent += read_len;
1656         if (vstream->metadata_sent == vstream->stream->metadata_received
1657                         && vstream->stream->closed) {
1658                 /* Release ownership for the viewer metadata stream. */
1659                 viewer_stream_put(vstream);
1660         }
1661
1662         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1663
1664         goto send_reply;
1665
1666 error:
1667         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1668
1669 send_reply:
1670         health_code_update();
1671         if (vstream) {
1672                 pthread_mutex_unlock(&vstream->stream->lock);
1673         }
1674         ret = send_response(conn->sock, &reply, sizeof(reply));
1675         if (ret < 0) {
1676                 goto end_free;
1677         }
1678         health_code_update();
1679
1680         if (len > 0) {
1681                 ret = send_response(conn->sock, data, len);
1682                 if (ret < 0) {
1683                         goto end_free;
1684                 }
1685         }
1686
1687         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1688                         be64toh(request.stream_id));
1689
1690         DBG("Metadata sent");
1691
1692 end_free:
1693         free(data);
1694 end:
1695         if (vstream) {
1696                 viewer_stream_put(vstream);
1697         }
1698         return ret;
1699 }
1700
1701 /*
1702  * Create a viewer session.
1703  *
1704  * Return 0 on success or else a negative value.
1705  */
1706 static
1707 int viewer_create_session(struct relay_connection *conn)
1708 {
1709         int ret;
1710         struct lttng_viewer_create_session_response resp;
1711
1712         DBG("Viewer create session received");
1713
1714         memset(&resp, 0, sizeof(resp));
1715         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1716         conn->viewer_session = viewer_session_create();
1717         if (!conn->viewer_session) {
1718                 ERR("Allocation viewer session");
1719                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1720                 goto send_reply;
1721         }
1722
1723 send_reply:
1724         health_code_update();
1725         ret = send_response(conn->sock, &resp, sizeof(resp));
1726         if (ret < 0) {
1727                 goto end;
1728         }
1729         health_code_update();
1730         ret = 0;
1731
1732 end:
1733         return ret;
1734 }
1735
1736
1737 /*
1738  * live_relay_unknown_command: send -1 if received unknown command
1739  */
1740 static
1741 void live_relay_unknown_command(struct relay_connection *conn)
1742 {
1743         struct lttcomm_relayd_generic_reply reply;
1744
1745         memset(&reply, 0, sizeof(reply));
1746         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1747         (void) send_response(conn->sock, &reply, sizeof(reply));
1748 }
1749
1750 /*
1751  * Process the commands received on the control socket
1752  */
1753 static
1754 int process_control(struct lttng_viewer_cmd *recv_hdr,
1755                 struct relay_connection *conn)
1756 {
1757         int ret = 0;
1758         uint32_t msg_value;
1759
1760         msg_value = be32toh(recv_hdr->cmd);
1761
1762         /*
1763          * Make sure we've done the version check before any command other then a
1764          * new client connection.
1765          */
1766         if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1767                 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1768                 ret = -1;
1769                 goto end;
1770         }
1771
1772         switch (msg_value) {
1773         case LTTNG_VIEWER_CONNECT:
1774                 ret = viewer_connect(conn);
1775                 break;
1776         case LTTNG_VIEWER_LIST_SESSIONS:
1777                 ret = viewer_list_sessions(conn);
1778                 break;
1779         case LTTNG_VIEWER_ATTACH_SESSION:
1780                 ret = viewer_attach_session(conn);
1781                 break;
1782         case LTTNG_VIEWER_GET_NEXT_INDEX:
1783                 ret = viewer_get_next_index(conn);
1784                 break;
1785         case LTTNG_VIEWER_GET_PACKET:
1786                 ret = viewer_get_packet(conn);
1787                 break;
1788         case LTTNG_VIEWER_GET_METADATA:
1789                 ret = viewer_get_metadata(conn);
1790                 break;
1791         case LTTNG_VIEWER_GET_NEW_STREAMS:
1792                 ret = viewer_get_new_streams(conn);
1793                 break;
1794         case LTTNG_VIEWER_CREATE_SESSION:
1795                 ret = viewer_create_session(conn);
1796                 break;
1797         default:
1798                 ERR("Received unknown viewer command (%u)",
1799                                 be32toh(recv_hdr->cmd));
1800                 live_relay_unknown_command(conn);
1801                 ret = -1;
1802                 goto end;
1803         }
1804
1805 end:
1806         return ret;
1807 }
1808
1809 static
1810 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1811 {
1812         int ret;
1813
1814         (void) lttng_poll_del(events, pollfd);
1815
1816         ret = close(pollfd);
1817         if (ret < 0) {
1818                 ERR("Closing pollfd %d", pollfd);
1819         }
1820 }
1821
1822 /*
1823  * This thread does the actual work
1824  */
1825 static
1826 void *thread_worker(void *data)
1827 {
1828         int ret, err = -1;
1829         uint32_t nb_fd;
1830         struct lttng_poll_event events;
1831         struct lttng_ht *viewer_connections_ht;
1832         struct lttng_ht_iter iter;
1833         struct lttng_viewer_cmd recv_hdr;
1834         struct relay_connection *destroy_conn;
1835
1836         DBG("[thread] Live viewer relay worker started");
1837
1838         rcu_register_thread();
1839
1840         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1841
1842         if (testpoint(relayd_thread_live_worker)) {
1843                 goto error_testpoint;
1844         }
1845
1846         /* table of connections indexed on socket */
1847         viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1848         if (!viewer_connections_ht) {
1849                 goto viewer_connections_ht_error;
1850         }
1851
1852         ret = create_thread_poll_set(&events, 2);
1853         if (ret < 0) {
1854                 goto error_poll_create;
1855         }
1856
1857         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1858         if (ret < 0) {
1859                 goto error;
1860         }
1861
1862 restart:
1863         while (1) {
1864                 int i;
1865
1866                 health_code_update();
1867
1868                 /* Infinite blocking call, waiting for transmission */
1869                 DBG3("Relayd live viewer worker thread polling...");
1870                 health_poll_entry();
1871                 ret = lttng_poll_wait(&events, -1);
1872                 health_poll_exit();
1873                 if (ret < 0) {
1874                         /*
1875                          * Restart interrupted system call.
1876                          */
1877                         if (errno == EINTR) {
1878                                 goto restart;
1879                         }
1880                         goto error;
1881                 }
1882
1883                 nb_fd = ret;
1884
1885                 /*
1886                  * Process control. The control connection is prioritised so we don't
1887                  * starve it with high throughput tracing data on the data
1888                  * connection.
1889                  */
1890                 for (i = 0; i < nb_fd; i++) {
1891                         /* Fetch once the poll data */
1892                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1893                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1894
1895                         health_code_update();
1896
1897                         if (!revents) {
1898                                 /* No activity for this FD (poll implementation). */
1899                                 continue;
1900                         }
1901
1902                         /* Thread quit pipe has been closed. Killing thread. */
1903                         ret = check_thread_quit_pipe(pollfd, revents);
1904                         if (ret) {
1905                                 err = 0;
1906                                 goto exit;
1907                         }
1908
1909                         /* Inspect the relay conn pipe for new connection. */
1910                         if (pollfd == live_conn_pipe[0]) {
1911                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1912                                         ERR("Relay live pipe error");
1913                                         goto error;
1914                                 } else if (revents & LPOLLIN) {
1915                                         struct relay_connection *conn;
1916
1917                                         ret = lttng_read(live_conn_pipe[0],
1918                                                         &conn, sizeof(conn));
1919                                         if (ret < 0) {
1920                                                 goto error;
1921                                         }
1922                                         lttng_poll_add(&events, conn->sock->fd,
1923                                                         LPOLLIN | LPOLLRDHUP);
1924                                         connection_ht_add(viewer_connections_ht, conn);
1925                                         DBG("Connection socket %d added to poll", conn->sock->fd);
1926                                 }
1927                         } else {
1928                                 /* Connection activity. */
1929                                 struct relay_connection *conn;
1930
1931                                 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
1932                                 if (!conn) {
1933                                         continue;
1934                                 }
1935
1936                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1937                                         cleanup_connection_pollfd(&events, pollfd);
1938                                         /* Put "create" ownership reference. */
1939                                         connection_put(conn);
1940                                 } else if (revents & LPOLLIN) {
1941                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1942                                                         sizeof(recv_hdr), 0);
1943                                         if (ret <= 0) {
1944                                                 /* Connection closed. */
1945                                                 cleanup_connection_pollfd(&events, pollfd);
1946                                                 /* Put "create" ownership reference. */
1947                                                 connection_put(conn);
1948                                                 DBG("Viewer control conn closed with %d", pollfd);
1949                                         } else {
1950                                                 ret = process_control(&recv_hdr, conn);
1951                                                 if (ret < 0) {
1952                                                         /* Clear the session on error. */
1953                                                         cleanup_connection_pollfd(&events, pollfd);
1954                                                         /* Put "create" ownership reference. */
1955                                                         connection_put(conn);
1956                                                         DBG("Viewer connection closed with %d", pollfd);
1957                                                 }
1958                                         }
1959                                 }
1960                                 /* Put local "get_by_sock" reference. */
1961                                 connection_put(conn);
1962                         }
1963                 }
1964         }
1965
1966 exit:
1967 error:
1968         lttng_poll_clean(&events);
1969
1970         /* Cleanup reamaining connection object. */
1971         rcu_read_lock();
1972         cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
1973                         destroy_conn,
1974                         sock_n.node) {
1975                 health_code_update();
1976                 connection_put(destroy_conn);
1977         }
1978         rcu_read_unlock();
1979 error_poll_create:
1980         lttng_ht_destroy(viewer_connections_ht);
1981 viewer_connections_ht_error:
1982         /* Close relay conn pipes */
1983         utils_close_pipe(live_conn_pipe);
1984         if (err) {
1985                 DBG("Viewer worker thread exited with error");
1986         }
1987         DBG("Viewer worker thread cleanup complete");
1988 error_testpoint:
1989         if (err) {
1990                 health_error();
1991                 ERR("Health error occurred in %s", __func__);
1992         }
1993         health_unregister(health_relayd);
1994         if (lttng_relay_stop_threads()) {
1995                 ERR("Error stopping threads");
1996         }
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_conn_pipe(void)
2006 {
2007         return utils_create_pipe_cloexec(live_conn_pipe);
2008 }
2009
2010 int relayd_live_join(void)
2011 {
2012         int ret, retval = 0;
2013         void *status;
2014
2015         ret = pthread_join(live_listener_thread, &status);
2016         if (ret) {
2017                 errno = ret;
2018                 PERROR("pthread_join live listener");
2019                 retval = -1;
2020         }
2021
2022         ret = pthread_join(live_worker_thread, &status);
2023         if (ret) {
2024                 errno = ret;
2025                 PERROR("pthread_join live worker");
2026                 retval = -1;
2027         }
2028
2029         ret = pthread_join(live_dispatcher_thread, &status);
2030         if (ret) {
2031                 errno = ret;
2032                 PERROR("pthread_join live dispatcher");
2033                 retval = -1;
2034         }
2035
2036         cleanup_relayd_live();
2037
2038         return retval;
2039 }
2040
2041 /*
2042  * main
2043  */
2044 int relayd_live_create(struct lttng_uri *uri)
2045 {
2046         int ret = 0, retval = 0;
2047         void *status;
2048         int is_root;
2049
2050         if (!uri) {
2051                 retval = -1;
2052                 goto exit_init_data;
2053         }
2054         live_uri = uri;
2055
2056         /* Check if daemon is UID = 0 */
2057         is_root = !getuid();
2058
2059         if (!is_root) {
2060                 if (live_uri->port < 1024) {
2061                         ERR("Need to be root to use ports < 1024");
2062                         retval = -1;
2063                         goto exit_init_data;
2064                 }
2065         }
2066
2067         /* Setup the thread apps communication pipe. */
2068         if (create_conn_pipe()) {
2069                 retval = -1;
2070                 goto exit_init_data;
2071         }
2072
2073         /* Init relay command queue. */
2074         cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2075
2076         /* Set up max poll set size */
2077         if (lttng_poll_set_max_size()) {
2078                 retval = -1;
2079                 goto exit_init_data;
2080         }
2081
2082         /* Setup the dispatcher thread */
2083         ret = pthread_create(&live_dispatcher_thread, NULL,
2084                         thread_dispatcher, (void *) NULL);
2085         if (ret) {
2086                 errno = ret;
2087                 PERROR("pthread_create viewer dispatcher");
2088                 retval = -1;
2089                 goto exit_dispatcher_thread;
2090         }
2091
2092         /* Setup the worker thread */
2093         ret = pthread_create(&live_worker_thread, NULL,
2094                         thread_worker, NULL);
2095         if (ret) {
2096                 errno = ret;
2097                 PERROR("pthread_create viewer worker");
2098                 retval = -1;
2099                 goto exit_worker_thread;
2100         }
2101
2102         /* Setup the listener thread */
2103         ret = pthread_create(&live_listener_thread, NULL,
2104                         thread_listener, (void *) NULL);
2105         if (ret) {
2106                 errno = ret;
2107                 PERROR("pthread_create viewer listener");
2108                 retval = -1;
2109                 goto exit_listener_thread;
2110         }
2111
2112         /*
2113          * All OK, started all threads.
2114          */
2115         return retval;
2116
2117         /*
2118          * Join on the live_listener_thread should anything be added after
2119          * the live_listener thread's creation.
2120          */
2121
2122 exit_listener_thread:
2123
2124         ret = pthread_join(live_worker_thread, &status);
2125         if (ret) {
2126                 errno = ret;
2127                 PERROR("pthread_join live worker");
2128                 retval = -1;
2129         }
2130 exit_worker_thread:
2131
2132         ret = pthread_join(live_dispatcher_thread, &status);
2133         if (ret) {
2134                 errno = ret;
2135                 PERROR("pthread_join live dispatcher");
2136                 retval = -1;
2137         }
2138 exit_dispatcher_thread:
2139
2140 exit_init_data:
2141         cleanup_relayd_live();
2142
2143         return retval;
2144 }