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