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