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