src/bin/lttng-relayd/live.c

   1 /*
   2  * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
   3  *                      David Goulet <dgoulet@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _GNU_SOURCE
  20 #include <getopt.h>
  21 #include <grp.h>
  22 #include <limits.h>
  23 #include <pthread.h>
  24 #include <signal.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <string.h>
  28 #include <sys/mman.h>
  29 #include <sys/mount.h>
  30 #include <sys/resource.h>
  31 #include <sys/socket.h>
  32 #include <sys/stat.h>
  33 #include <sys/types.h>
  34 #include <sys/wait.h>
  35 #include <inttypes.h>
  36 #include <urcu/futex.h>
  37 #include <urcu/uatomic.h>
  38 #include <unistd.h>
  39 #include <fcntl.h>
  40 #include <config.h>
  41
  42 #include <lttng/lttng.h>
  43 #include <common/common.h>
  44 #include <common/compat/poll.h>
  45 #include <common/compat/socket.h>
  46 #include <common/defaults.h>
  47 #include <common/futex.h>
  48 #include <common/index/index.h>
  49 #include <common/sessiond-comm/sessiond-comm.h>
  50 #include <common/sessiond-comm/inet.h>
  51 #include <common/sessiond-comm/relayd.h>
  52 #include <common/uri.h>
  53 #include <common/utils.h>
  54
  55 #include "cmd.h"
  56 #include "live.h"
  57 #include "lttng-relayd.h"
  58 #include "utils.h"
  59 #include "health-relayd.h"
  60 #include "testpoint.h"
  61 #include "viewer-stream.h"
  62 #include "stream.h"
  63 #include "session.h"
  64 #include "ctf-trace.h"
  65 #include "connection.h"
  66
  67 static struct lttng_uri *live_uri;
  68
  69 /*
  70  * This pipe is used to inform the worker thread that a command is queued and
  71  * ready to be processed.
  72  */
  73 static int live_conn_pipe[2] = { -1, -1 };
  74
  75 /* Shared between threads */
  76 static int live_dispatch_thread_exit;
  77
  78 static pthread_t live_listener_thread;
  79 static pthread_t live_dispatcher_thread;
  80 static pthread_t live_worker_thread;
  81
  82 /*
  83  * Relay command queue.
  84  *
  85  * The live_thread_listener and live_thread_dispatcher communicate with this
  86  * queue.
  87  */
  88 static struct relay_conn_queue viewer_conn_queue;
  89
  90 static uint64_t last_relay_viewer_session_id;
  91
  92 /*
  93  * Cleanup the daemon
  94  */
  95 static
  96 void cleanup(void)
  97 {
  98         DBG("Cleaning up");
  99
 100         free(live_uri);
 101 }
 102
 103 /*
 104  * Receive a request buffer using a given socket, destination allocated buffer
 105  * of length size.
 106  *
 107  * Return the size of the received message or else a negative value on error
 108  * with errno being set by recvmsg() syscall.
 109  */
 110 static
 111 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
 112 {
 113         ssize_t ret;
 114
 115         assert(sock);
 116         assert(buf);
 117
 118         ret = sock->ops->recvmsg(sock, buf, size, 0);
 119         if (ret < 0 || ret != size) {
 120                 if (ret == 0) {
 121                         /* Orderly shutdown. Not necessary to print an error. */
 122                         DBG("Socket %d did an orderly shutdown", sock->fd);
 123                 } else {
 124                         ERR("Relay failed to receive request.");
 125                 }
 126                 ret = -1;
 127         }
 128
 129         return ret;
 130 }
 131
 132 /*
 133  * Send a response buffer using a given socket, source allocated buffer of
 134  * length size.
 135  *
 136  * Return the size of the sent message or else a negative value on error with
 137  * errno being set by sendmsg() syscall.
 138  */
 139 static
 140 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
 141 {
 142         ssize_t ret;
 143
 144         assert(sock);
 145         assert(buf);
 146
 147         ret = sock->ops->sendmsg(sock, buf, size, 0);
 148         if (ret < 0) {
 149                 ERR("Relayd failed to send response.");
 150         }
 151
 152         return ret;
 153 }
 154
 155 /*
 156  * Atomically check if new streams got added in one of the sessions attached
 157  * and reset the flag to 0.
 158  *
 159  * Returns 1 if new streams got added, 0 if nothing changed, a negative value
 160  * on error.
 161  */
 162 static
 163 int check_new_streams(struct relay_connection *conn)
 164 {
 165         struct relay_session *session;
 166         unsigned long current_val;
 167         int ret = 0;
 168
 169         if (!conn->viewer_session) {
 170                 goto end;
 171         }
 172         cds_list_for_each_entry(session,
 173                         &conn->viewer_session->sessions_head,
 174                         viewer_session_list) {
 175                 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
 176                 ret = current_val;
 177                 if (ret == 1) {
 178                         goto end;
 179                 }
 180         }
 181
 182 end:
 183         return ret;
 184 }
 185
 186 /*
 187  * Send viewer streams to the given socket. The ignore_sent_flag indicates if
 188  * this function should ignore the sent flag or not.
 189  *
 190  * Return 0 on success or else a negative value.
 191  */
 192 static
 193 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
 194                 struct relay_session *session, unsigned int ignore_sent_flag)
 195 {
 196         ssize_t ret;
 197         struct lttng_viewer_stream send_stream;
 198         struct lttng_ht_iter iter;
 199         struct relay_viewer_stream *vstream;
 200
 201         assert(session);
 202
 203         rcu_read_lock();
 204
 205         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
 206                         stream_n.node) {
 207                 struct ctf_trace *ctf_trace;
 208
 209                 health_code_update();
 210
 211                 /* Ignore if not the same session. */
 212                 if (vstream->session_id != session->id ||
 213                                 (!ignore_sent_flag && vstream->sent_flag)) {
 214                         continue;
 215                 }
 216
 217                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 218                                 vstream->path_name);
 219                 assert(ctf_trace);
 220
 221                 send_stream.id = htobe64(vstream->stream_handle);
 222                 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
 223                 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
 224                 strncpy(send_stream.path_name, vstream->path_name,
 225                                 sizeof(send_stream.path_name));
 226                 strncpy(send_stream.channel_name, vstream->channel_name,
 227                                 sizeof(send_stream.channel_name));
 228
 229                 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
 230                 ret = send_response(sock, &send_stream, sizeof(send_stream));
 231                 if (ret < 0) {
 232                         goto end_unlock;
 233                 }
 234                 vstream->sent_flag = 1;
 235         }
 236
 237         ret = 0;
 238
 239 end_unlock:
 240         rcu_read_unlock();
 241         return ret;
 242 }
 243
 244 /*
 245  * Create every viewer stream possible for the given session with the seek
 246  * type. Three counters *can* be return which are in order the total amount of
 247  * viewer stream of the session, the number of unsent stream and the number of
 248  * stream created. Those counters can be NULL and thus will be ignored.
 249  *
 250  * Return 0 on success or else a negative value.
 251  */
 252 static
 253 int make_viewer_streams(struct relay_session *session,
 254                 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
 255                 uint32_t *nb_created)
 256 {
 257         int ret;
 258         struct lttng_ht_iter iter;
 259         struct ctf_trace *ctf_trace;
 260
 261         assert(session);
 262
 263         /*
 264          * This is to make sure we create viewer streams for a full received
 265          * channel. For instance, if we have 8 streams for a channel that are
 266          * concurrently being flagged ready, we can end up creating just a subset
 267          * of the 8 streams (the ones that are flagged). This lock avoids this
 268          * limbo state.
 269          */
 270         pthread_mutex_lock(&session->viewer_ready_lock);
 271
 272         /*
 273          * Create viewer streams for relay streams that are ready to be used for a
 274          * the given session id only.
 275          */
 276         rcu_read_lock();
 277         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 278                         node.node) {
 279                 struct relay_stream *stream;
 280
 281                 health_code_update();
 282
 283                 if (ctf_trace->invalid_flag) {
 284                         continue;
 285                 }
 286
 287                 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
 288                         struct relay_viewer_stream *vstream;
 289
 290                         if (!stream->viewer_ready) {
 291                                 continue;
 292                         }
 293
 294                         vstream = viewer_stream_find_by_id(stream->stream_handle);
 295                         if (!vstream) {
 296                                 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
 297                                 if (!vstream) {
 298                                         ret = -1;
 299                                         goto error_unlock;
 300                                 }
 301                                 /* Acquire reference to ctf_trace. */
 302                                 ctf_trace_get_ref(ctf_trace);
 303
 304                                 if (nb_created) {
 305                                         /* Update number of created stream counter. */
 306                                         (*nb_created)++;
 307                                 }
 308                         } else if (!vstream->sent_flag && nb_unsent) {
 309                                 /* Update number of unsent stream counter. */
 310                                 (*nb_unsent)++;
 311                         }
 312                         /* Update number of total stream counter. */
 313                         if (nb_total) {
 314                                 (*nb_total)++;
 315                         }
 316                 }
 317         }
 318
 319         ret = 0;
 320
 321 error_unlock:
 322         rcu_read_unlock();
 323         pthread_mutex_unlock(&session->viewer_ready_lock);
 324         return ret;
 325 }
 326
 327 /*
 328  * Write to writable pipe used to notify a thread.
 329  */
 330 static
 331 int notify_thread_pipe(int wpipe)
 332 {
 333         ssize_t ret;
 334
 335         ret = lttng_write(wpipe, "!", 1);
 336         if (ret < 1) {
 337                 PERROR("write poll pipe");
 338         }
 339
 340         return (int) ret;
 341 }
 342
 343 /*
 344  * Stop all threads by closing the thread quit pipe.
 345  */
 346 static
 347 void stop_threads(void)
 348 {
 349         int ret;
 350
 351         /* Stopping all threads */
 352         DBG("Terminating all live threads");
 353         ret = notify_thread_pipe(thread_quit_pipe[1]);
 354         if (ret < 0) {
 355                 ERR("write error on thread quit pipe");
 356         }
 357
 358         /* Dispatch thread */
 359         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 360         futex_nto1_wake(&viewer_conn_queue.futex);
 361 }
 362
 363 /*
 364  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 365  */
 366 static
 367 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 368 {
 369         int ret;
 370
 371         if (events == NULL || size == 0) {
 372                 ret = -1;
 373                 goto error;
 374         }
 375
 376         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 377         if (ret < 0) {
 378                 goto error;
 379         }
 380
 381         /* Add quit pipe */
 382         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 383         if (ret < 0) {
 384                 goto error;
 385         }
 386
 387         return 0;
 388
 389 error:
 390         return ret;
 391 }
 392
 393 /*
 394  * Check if the thread quit pipe was triggered.
 395  *
 396  * Return 1 if it was triggered else 0;
 397  */
 398 static
 399 int check_thread_quit_pipe(int fd, uint32_t events)
 400 {
 401         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 402                 return 1;
 403         }
 404
 405         return 0;
 406 }
 407
 408 /*
 409  * Create and init socket from uri.
 410  */
 411 static
 412 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 413 {
 414         int ret;
 415         struct lttcomm_sock *sock = NULL;
 416
 417         sock = lttcomm_alloc_sock_from_uri(uri);
 418         if (sock == NULL) {
 419                 ERR("Allocating socket");
 420                 goto error;
 421         }
 422
 423         ret = lttcomm_create_sock(sock);
 424         if (ret < 0) {
 425                 goto error;
 426         }
 427         DBG("Listening on sock %d for live", sock->fd);
 428
 429         ret = sock->ops->bind(sock);
 430         if (ret < 0) {
 431                 goto error;
 432         }
 433
 434         ret = sock->ops->listen(sock, -1);
 435         if (ret < 0) {
 436                 goto error;
 437
 438         }
 439
 440         return sock;
 441
 442 error:
 443         if (sock) {
 444                 lttcomm_destroy_sock(sock);
 445         }
 446         return NULL;
 447 }
 448
 449 /*
 450  * This thread manages the listening for new connections on the network
 451  */
 452 static
 453 void *thread_listener(void *data)
 454 {
 455         int i, ret, pollfd, err = -1;
 456         uint32_t revents, nb_fd;
 457         struct lttng_poll_event events;
 458         struct lttcomm_sock *live_control_sock;
 459
 460         DBG("[thread] Relay live listener started");
 461
 462         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
 463
 464         health_code_update();
 465
 466         live_control_sock = init_socket(live_uri);
 467         if (!live_control_sock) {
 468                 goto error_sock_control;
 469         }
 470
 471         /* Pass 2 as size here for the thread quit pipe and control sockets. */
 472         ret = create_thread_poll_set(&events, 2);
 473         if (ret < 0) {
 474                 goto error_create_poll;
 475         }
 476
 477         /* Add the control socket */
 478         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 479         if (ret < 0) {
 480                 goto error_poll_add;
 481         }
 482
 483         lttng_relay_notify_ready();
 484
 485         if (testpoint(relayd_thread_live_listener)) {
 486                 goto error_testpoint;
 487         }
 488
 489         while (1) {
 490                 health_code_update();
 491
 492                 DBG("Listener accepting live viewers connections");
 493
 494 restart:
 495                 health_poll_entry();
 496                 ret = lttng_poll_wait(&events, -1);
 497                 health_poll_exit();
 498                 if (ret < 0) {
 499                         /*
 500                          * Restart interrupted system call.
 501                          */
 502                         if (errno == EINTR) {
 503                                 goto restart;
 504                         }
 505                         goto error;
 506                 }
 507                 nb_fd = ret;
 508
 509                 DBG("Relay new viewer connection received");
 510                 for (i = 0; i < nb_fd; i++) {
 511                         health_code_update();
 512
 513                         /* Fetch once the poll data */
 514                         revents = LTTNG_POLL_GETEV(&events, i);
 515                         pollfd = LTTNG_POLL_GETFD(&events, i);
 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                                  * Get allocated in this thread, enqueued to a global queue,
 530                                  * dequeued and freed in the worker thread.
 531                                  */
 532                                 int val = 1;
 533                                 struct relay_connection *new_conn;
 534                                 struct lttcomm_sock *newsock;
 535
 536                                 new_conn = connection_create();
 537                                 if (!new_conn) {
 538                                         goto error;
 539                                 }
 540
 541                                 newsock = live_control_sock->ops->accept(live_control_sock);
 542                                 if (!newsock) {
 543                                         PERROR("accepting control sock");
 544                                         connection_free(new_conn);
 545                                         goto error;
 546                                 }
 547                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 548
 549                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 550                                                 sizeof(val));
 551                                 if (ret < 0) {
 552                                         PERROR("setsockopt inet");
 553                                         lttcomm_destroy_sock(newsock);
 554                                         connection_free(new_conn);
 555                                         goto error;
 556                                 }
 557                                 new_conn->sock = newsock;
 558
 559                                 /* Enqueue request for the dispatcher thread. */
 560                                 cds_wfq_enqueue(&viewer_conn_queue.queue, &new_conn->qnode);
 561
 562                                 /*
 563                                  * Wake the dispatch queue futex. Implicit memory barrier with
 564                                  * the exchange in cds_wfq_enqueue.
 565                                  */
 566                                 futex_nto1_wake(&viewer_conn_queue.futex);
 567                         }
 568                 }
 569         }
 570
 571 exit:
 572 error:
 573 error_poll_add:
 574 error_testpoint:
 575         lttng_poll_clean(&events);
 576 error_create_poll:
 577         if (live_control_sock->fd >= 0) {
 578                 ret = live_control_sock->ops->close(live_control_sock);
 579                 if (ret) {
 580                         PERROR("close");
 581                 }
 582         }
 583         lttcomm_destroy_sock(live_control_sock);
 584 error_sock_control:
 585         if (err) {
 586                 health_error();
 587                 DBG("Live viewer listener thread exited with error");
 588         }
 589         health_unregister(health_relayd);
 590         DBG("Live viewer listener thread cleanup complete");
 591         stop_threads();
 592         return NULL;
 593 }
 594
 595 /*
 596  * This thread manages the dispatching of the requests to worker threads
 597  */
 598 static
 599 void *thread_dispatcher(void *data)
 600 {
 601         int err = -1;
 602         ssize_t ret;
 603         struct cds_wfq_node *node;
 604         struct relay_connection *conn = NULL;
 605
 606         DBG("[thread] Live viewer relay dispatcher started");
 607
 608         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
 609
 610         if (testpoint(relayd_thread_live_dispatcher)) {
 611                 goto error_testpoint;
 612         }
 613
 614         health_code_update();
 615
 616         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 617                 health_code_update();
 618
 619                 /* Atomically prepare the queue futex */
 620                 futex_nto1_prepare(&viewer_conn_queue.futex);
 621
 622                 do {
 623                         health_code_update();
 624
 625                         /* Dequeue commands */
 626                         node = cds_wfq_dequeue_blocking(&viewer_conn_queue.queue);
 627                         if (node == NULL) {
 628                                 DBG("Woken up but nothing in the live-viewer "
 629                                                 "relay command queue");
 630                                 /* Continue thread execution */
 631                                 break;
 632                         }
 633                         conn = caa_container_of(node, struct relay_connection, qnode);
 634                         DBG("Dispatching viewer request waiting on sock %d",
 635                                         conn->sock->fd);
 636
 637                         /*
 638                          * Inform worker thread of the new request. This call is blocking
 639                          * so we can be assured that the data will be read at some point in
 640                          * time or wait to the end of the world :)
 641                          */
 642                         ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
 643                         if (ret < 0) {
 644                                 PERROR("write conn pipe");
 645                                 connection_destroy(conn);
 646                                 goto error;
 647                         }
 648                 } while (node != NULL);
 649
 650                 /* Futex wait on queue. Blocking call on futex() */
 651                 health_poll_entry();
 652                 futex_nto1_wait(&viewer_conn_queue.futex);
 653                 health_poll_exit();
 654         }
 655
 656         /* Normal exit, no error */
 657         err = 0;
 658
 659 error:
 660 error_testpoint:
 661         if (err) {
 662                 health_error();
 663                 ERR("Health error occurred in %s", __func__);
 664         }
 665         health_unregister(health_relayd);
 666         DBG("Live viewer dispatch thread dying");
 667         stop_threads();
 668         return NULL;
 669 }
 670
 671 /*
 672  * Establish connection with the viewer and check the versions.
 673  *
 674  * Return 0 on success or else negative value.
 675  */
 676 static
 677 int viewer_connect(struct relay_connection *conn)
 678 {
 679         int ret;
 680         struct lttng_viewer_connect reply, msg;
 681
 682         assert(conn);
 683
 684         conn->version_check_done = 1;
 685
 686         health_code_update();
 687
 688         DBG("Viewer is establishing a connection to the relayd.");
 689
 690         ret = recv_request(conn->sock, &msg, sizeof(msg));
 691         if (ret < 0) {
 692                 goto end;
 693         }
 694
 695         health_code_update();
 696
 697         memset(&reply, 0, sizeof(reply));
 698         reply.major = RELAYD_VERSION_COMM_MAJOR;
 699         reply.minor = RELAYD_VERSION_COMM_MINOR;
 700
 701         /* Major versions must be the same */
 702         if (reply.major != be32toh(msg.major)) {
 703                 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
 704                                 reply.major, be32toh(msg.major));
 705                 ret = -1;
 706                 goto end;
 707         }
 708
 709         conn->major = reply.major;
 710         /* We adapt to the lowest compatible version */
 711         if (reply.minor <= be32toh(msg.minor)) {
 712                 conn->minor = reply.minor;
 713         } else {
 714                 conn->minor = be32toh(msg.minor);
 715         }
 716
 717         if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
 718                 conn->type = RELAY_VIEWER_COMMAND;
 719         } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
 720                 conn->type = RELAY_VIEWER_NOTIFICATION;
 721         } else {
 722                 ERR("Unknown connection type : %u", be32toh(msg.type));
 723                 ret = -1;
 724                 goto end;
 725         }
 726
 727         reply.major = htobe32(reply.major);
 728         reply.minor = htobe32(reply.minor);
 729         if (conn->type == RELAY_VIEWER_COMMAND) {
 730                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 731         }
 732
 733         health_code_update();
 734
 735         ret = send_response(conn->sock, &reply, sizeof(reply));
 736         if (ret < 0) {
 737                 goto end;
 738         }
 739
 740         health_code_update();
 741
 742         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
 743         ret = 0;
 744
 745 end:
 746         return ret;
 747 }
 748
 749 /*
 750  * Send the viewer the list of current sessions.
 751  *
 752  * Return 0 on success or else a negative value.
 753  */
 754 static
 755 int viewer_list_sessions(struct relay_connection *conn)
 756 {
 757         int ret;
 758         struct lttng_viewer_list_sessions session_list;
 759         unsigned long count;
 760         long approx_before, approx_after;
 761         struct lttng_ht_iter iter;
 762         struct lttng_viewer_session send_session;
 763         struct relay_session *session;
 764
 765         DBG("List sessions received");
 766
 767         rcu_read_lock();
 768         cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
 769                         &approx_after);
 770         session_list.sessions_count = htobe32(count);
 771
 772         health_code_update();
 773
 774         ret = send_response(conn->sock, &session_list, sizeof(session_list));
 775         if (ret < 0) {
 776                 goto end_unlock;
 777         }
 778
 779         health_code_update();
 780
 781         cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
 782                         session_n.node) {
 783                 health_code_update();
 784
 785                 strncpy(send_session.session_name, session->session_name,
 786                                 sizeof(send_session.session_name));
 787                 strncpy(send_session.hostname, session->hostname,
 788                                 sizeof(send_session.hostname));
 789                 send_session.id = htobe64(session->id);
 790                 send_session.live_timer = htobe32(session->live_timer);
 791                 send_session.clients = htobe32(session->viewer_refcount);
 792                 send_session.streams = htobe32(session->stream_count);
 793
 794                 health_code_update();
 795
 796                 ret = send_response(conn->sock, &send_session, sizeof(send_session));
 797                 if (ret < 0) {
 798                         goto end_unlock;
 799                 }
 800         }
 801         health_code_update();
 802
 803         rcu_read_unlock();
 804         ret = 0;
 805         goto end;
 806
 807 end_unlock:
 808         rcu_read_unlock();
 809
 810 end:
 811         return ret;
 812 }
 813
 814 /*
 815  * Check if a connection is attached to a session.
 816  * Return 1 if attached, 0 if not attached, a negative value on error.
 817  */
 818 static
 819 int session_attached(struct relay_connection *conn, uint64_t session_id)
 820 {
 821         struct relay_session *session;
 822         int found = 0;
 823
 824         if (!conn->viewer_session) {
 825                 goto end;
 826         }
 827         cds_list_for_each_entry(session,
 828                         &conn->viewer_session->sessions_head,
 829                         viewer_session_list) {
 830                 if (session->id == session_id) {
 831                         found = 1;
 832                         goto end;
 833                 }
 834         }
 835
 836 end:
 837         return found;
 838 }
 839
 840 /*
 841  * Delete all streams for a specific session ID.
 842  */
 843 static void destroy_viewer_streams_by_session(struct relay_session *session)
 844 {
 845         struct relay_viewer_stream *stream;
 846         struct lttng_ht_iter iter;
 847
 848         assert(session);
 849
 850         rcu_read_lock();
 851         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
 852                         stream_n.node) {
 853                 struct ctf_trace *ctf_trace;
 854
 855                 health_code_update();
 856                 if (stream->session_id != session->id) {
 857                         continue;
 858                 }
 859
 860                 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
 861                                 stream->path_name);
 862                 assert(ctf_trace);
 863
 864                 viewer_stream_delete(stream);
 865
 866                 if (stream->metadata_flag) {
 867                         ctf_trace->metadata_sent = 0;
 868                         ctf_trace->viewer_metadata_stream = NULL;
 869                 }
 870
 871                 viewer_stream_destroy(ctf_trace, stream);
 872         }
 873         rcu_read_unlock();
 874 }
 875
 876 static void try_destroy_streams(struct relay_session *session)
 877 {
 878         struct ctf_trace *ctf_trace;
 879         struct lttng_ht_iter iter;
 880
 881         assert(session);
 882
 883         cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
 884                         node.node) {
 885                 /* Attempt to destroy the ctf trace of that session. */
 886                 ctf_trace_try_destroy(session, ctf_trace);
 887         }
 888 }
 889
 890 /*
 891  * Cleanup a session.
 892  */
 893 static void cleanup_session(struct relay_connection *conn,
 894                 struct relay_session *session)
 895 {
 896         /*
 897          * Very important that this is done before destroying the session so we
 898          * can put back every viewer stream reference from the ctf_trace.
 899          */
 900         destroy_viewer_streams_by_session(session);
 901         try_destroy_streams(session);
 902         cds_list_del(&session->viewer_session_list);
 903         session_viewer_try_destroy(conn->sessions_ht, session);
 904 }
 905
 906 /*
 907  * Send the viewer the list of current sessions.
 908  */
 909 static
 910 int viewer_get_new_streams(struct relay_connection *conn)
 911 {
 912         int ret, send_streams = 0;
 913         uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
 914         struct lttng_viewer_new_streams_request request;
 915         struct lttng_viewer_new_streams_response response;
 916         struct relay_session *session;
 917         uint64_t session_id;
 918
 919         assert(conn);
 920
 921         DBG("Get new streams received");
 922
 923         health_code_update();
 924
 925         /* Receive the request from the connected client. */
 926         ret = recv_request(conn->sock, &request, sizeof(request));
 927         if (ret < 0) {
 928                 goto error;
 929         }
 930         session_id = be64toh(request.session_id);
 931
 932         health_code_update();
 933
 934         rcu_read_lock();
 935         session = session_find_by_id(conn->sessions_ht, session_id);
 936         if (!session) {
 937                 DBG("Relay session %" PRIu64 " not found", session_id);
 938                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 939                 goto send_reply;
 940         }
 941
 942         if (!session_attached(conn, session_id)) {
 943                 send_streams = 0;
 944                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
 945                 goto send_reply;
 946         }
 947
 948         send_streams = 1;
 949         response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
 950
 951         ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
 952                         &nb_created);
 953         if (ret < 0) {
 954                 goto end_unlock;
 955         }
 956         /* Only send back the newly created streams with the unsent ones. */
 957         nb_streams = nb_created + nb_unsent;
 958         response.streams_count = htobe32(nb_streams);
 959
 960         /*
 961          * If the session is closed and we have no new streams to send,
 962          * it means that the viewer has already received the whole trace
 963          * for this session and should now close it.
 964          */
 965         if (nb_streams == 0 && session->close_flag) {
 966                 send_streams = 0;
 967                 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
 968                 /*
 969                  * Remove the session from the attached list of the connection
 970                  * and try to destroy it.
 971                  */
 972                 cds_list_del(&session->viewer_session_list);
 973                 cleanup_session(conn, session);
 974                 goto send_reply;
 975         }
 976
 977 send_reply:
 978         health_code_update();
 979         ret = send_response(conn->sock, &response, sizeof(response));
 980         if (ret < 0) {
 981                 goto end_unlock;
 982         }
 983         health_code_update();
 984
 985         /*
 986          * Unknown or empty session, just return gracefully, the viewer knows what
 987          * is happening.
 988          */
 989         if (!send_streams || !nb_streams) {
 990                 ret = 0;
 991                 goto end_unlock;
 992         }
 993
 994         /*
 995          * Send stream and *DON'T* ignore the sent flag so every viewer streams
 996          * that were not sent from that point will be sent to the viewer.
 997          */
 998         ret = send_viewer_streams(conn->sock, session, 0);
 999         if (ret < 0) {
1000                 goto end_unlock;
1001         }
1002
1003 end_unlock:
1004         rcu_read_unlock();
1005 error:
1006         return ret;
1007 }
1008
1009 /*
1010  * Send the viewer the list of current sessions.
1011  */
1012 static
1013 int viewer_attach_session(struct relay_connection *conn)
1014 {
1015         int send_streams = 0;
1016         ssize_t ret;
1017         uint32_t nb_streams = 0;
1018         enum lttng_viewer_seek seek_type;
1019         struct lttng_viewer_attach_session_request request;
1020         struct lttng_viewer_attach_session_response response;
1021         struct relay_session *session;
1022
1023         assert(conn);
1024
1025         health_code_update();
1026
1027         /* Receive the request from the connected client. */
1028         ret = recv_request(conn->sock, &request, sizeof(request));
1029         if (ret < 0) {
1030                 goto error;
1031         }
1032
1033         health_code_update();
1034
1035         if (!conn->viewer_session) {
1036                 DBG("Client trying to attach before creating a live viewer session");
1037                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1038                 goto send_reply;
1039         }
1040
1041         rcu_read_lock();
1042         session = session_find_by_id(conn->sessions_ht,
1043                         be64toh(request.session_id));
1044         if (!session) {
1045                 DBG("Relay session %" PRIu64 " not found",
1046                                 be64toh(request.session_id));
1047                 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1048                 goto send_reply;
1049         }
1050         session_viewer_attach(session);
1051         DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1052
1053         if (uatomic_read(&session->viewer_refcount) > 1) {
1054                 DBG("Already a viewer attached");
1055                 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1056                 session_viewer_detach(session);
1057                 goto send_reply;
1058         } else if (session->live_timer == 0) {
1059                 DBG("Not live session");
1060                 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1061                 goto send_reply;
1062         } else {
1063                 send_streams = 1;
1064                 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1065                 cds_list_add(&session->viewer_session_list,
1066                                 &conn->viewer_session->sessions_head);
1067         }
1068
1069         switch (be32toh(request.seek)) {
1070         case LTTNG_VIEWER_SEEK_BEGINNING:
1071         case LTTNG_VIEWER_SEEK_LAST:
1072                 seek_type = be32toh(request.seek);
1073                 break;
1074         default:
1075                 ERR("Wrong seek parameter");
1076                 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1077                 send_streams = 0;
1078                 goto send_reply;
1079         }
1080
1081         ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1082         if (ret < 0) {
1083                 goto end_unlock;
1084         }
1085         response.streams_count = htobe32(nb_streams);
1086
1087 send_reply:
1088         health_code_update();
1089         ret = send_response(conn->sock, &response, sizeof(response));
1090         if (ret < 0) {
1091                 goto end_unlock;
1092         }
1093         health_code_update();
1094
1095         /*
1096          * Unknown or empty session, just return gracefully, the viewer knows what
1097          * is happening.
1098          */
1099         if (!send_streams || !nb_streams) {
1100                 ret = 0;
1101                 goto end_unlock;
1102         }
1103
1104         /* Send stream and ignore the sent flag. */
1105         ret = send_viewer_streams(conn->sock, session, 1);
1106         if (ret < 0) {
1107                 goto end_unlock;
1108         }
1109
1110 end_unlock:
1111         rcu_read_unlock();
1112 error:
1113         return ret;
1114 }
1115
1116 /*
1117  * Open the index file if needed for the given vstream.
1118  *
1119  * If an index file is successfully opened, the index_read_fd of the stream is
1120  * set with it.
1121  *
1122  * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1123  */
1124 static int try_open_index(struct relay_viewer_stream *vstream,
1125                 struct relay_stream *rstream)
1126 {
1127         int ret = 0;
1128
1129         assert(vstream);
1130         assert(rstream);
1131
1132         if (vstream->index_read_fd >= 0) {
1133                 goto end;
1134         }
1135
1136         /*
1137          * First time, we open the index file and at least one index is ready.  The
1138          * race between the read and write of the total_index_received is
1139          * acceptable here since the client will be notified to simply come back
1140          * and get the next index.
1141          */
1142         if (rstream->total_index_received <= 0) {
1143                 ret = -ENOENT;
1144                 goto end;
1145         }
1146         ret = index_open(vstream->path_name, vstream->channel_name,
1147                         vstream->tracefile_count, vstream->tracefile_count_current);
1148         if (ret >= 0) {
1149                 vstream->index_read_fd = ret;
1150                 ret = 0;
1151                 goto end;
1152         }
1153
1154 end:
1155         return ret;
1156 }
1157
1158 /*
1159  * Check the status of the index for the given stream. This function updates
1160  * the index structure if needed and can destroy the vstream also for the HUP
1161  * situation.
1162  *
1163  * Return 0 means that we can proceed with the index. A value of 1 means that
1164  * the index has been updated and is ready to be send to the client. A negative
1165  * value indicates an error that can't be handled.
1166  */
1167 static int check_index_status(struct relay_viewer_stream *vstream,
1168                 struct relay_stream *rstream, struct ctf_trace *trace,
1169                 struct lttng_viewer_index *index)
1170 {
1171         int ret;
1172
1173         assert(vstream);
1174         assert(rstream);
1175         assert(index);
1176         assert(trace);
1177
1178         if (!rstream->close_flag) {
1179                 /* Rotate on abort (overwrite). */
1180                 if (vstream->abort_flag) {
1181                         DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1182                                         vstream->stream_handle);
1183                         ret = viewer_stream_rotate(vstream, rstream);
1184                         if (ret < 0) {
1185                                 goto error;
1186                         } else if (ret == 1) {
1187                                 /* EOF */
1188                                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1189                                 goto hup;
1190                         }
1191                         /* ret == 0 means successful so we continue. */
1192                 }
1193
1194                 /* Check if we are in the same trace file at this point. */
1195                 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1196                         if (rstream->beacon_ts_end != -1ULL &&
1197                                         vstream->last_sent_index == rstream->total_index_received) {
1198                                 /*
1199                                  * We've received a synchronization beacon and the last index
1200                                  * available has been sent, the index for now is inactive.
1201                                  */
1202                                 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1203                                 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1204                                 goto index_ready;
1205                         } else if (rstream->total_index_received <= vstream->last_sent_index
1206                                         && !vstream->close_write_flag) {
1207                                 /*
1208                                  * Reader and writer are working in the same tracefile, so we care
1209                                  * about the number of index received and sent. Otherwise, we read
1210                                  * up to EOF.
1211                                  */
1212                                 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1213                                 goto index_ready;
1214                         }
1215                 }
1216                 /* Nothing to do with the index, continue with it. */
1217                 ret = 0;
1218         } else if (rstream->close_flag && vstream->close_write_flag &&
1219                         vstream->total_index_received == vstream->last_sent_index) {
1220                 /* Last index sent and current tracefile closed in write */
1221                 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1222                 goto hup;
1223         } else {
1224                 vstream->close_write_flag = 1;
1225                 ret = 0;
1226         }
1227
1228 error:
1229         return ret;
1230
1231 hup:
1232         viewer_stream_delete(vstream);
1233         viewer_stream_destroy(trace, vstream);
1234 index_ready:
1235         return 1;
1236 }
1237
1238 /*
1239  * Send the next index for a stream.
1240  *
1241  * Return 0 on success or else a negative value.
1242  */
1243 static
1244 int viewer_get_next_index(struct relay_connection *conn)
1245 {
1246         int ret;
1247         ssize_t read_ret;
1248         struct lttng_viewer_get_next_index request_index;
1249         struct lttng_viewer_index viewer_index;
1250         struct ctf_packet_index packet_index;
1251         struct relay_viewer_stream *vstream;
1252         struct relay_stream *rstream;
1253         struct ctf_trace *ctf_trace;
1254         struct relay_session *session;
1255
1256         assert(conn);
1257
1258         DBG("Viewer get next index");
1259
1260         health_code_update();
1261
1262         ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1263         if (ret < 0) {
1264                 goto end;
1265         }
1266         health_code_update();
1267
1268         rcu_read_lock();
1269         vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1270         if (!vstream) {
1271                 ret = -1;
1272                 goto end_unlock;
1273         }
1274
1275         session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1276         if (!session) {
1277                 ret = -1;
1278                 goto end_unlock;
1279         }
1280
1281         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1282         assert(ctf_trace);
1283
1284         memset(&viewer_index, 0, sizeof(viewer_index));
1285
1286         /*
1287          * The viewer should not ask for index on metadata stream.
1288          */
1289         if (vstream->metadata_flag) {
1290                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1291                 goto send_reply;
1292         }
1293
1294         rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1295         assert(rstream);
1296
1297         /* Try to open an index if one is needed for that stream. */
1298         ret = try_open_index(vstream, rstream);
1299         if (ret < 0) {
1300                 if (ret == -ENOENT) {
1301                         /*
1302                          * The index is created only when the first data packet arrives, it
1303                          * might not be ready at the beginning of the session
1304                          */
1305                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1306                 } else {
1307                         /* Unhandled error. */
1308                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1309                 }
1310                 goto send_reply;
1311         }
1312
1313         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1314         ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1315         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1316         if (ret < 0) {
1317                 goto end;
1318         } else if (ret == 1) {
1319                 /*
1320                  * This means the viewer index data structure has been populated by the
1321                  * check call thus we now send back the reply to the client.
1322                  */
1323                 goto send_reply;
1324         }
1325         /* At this point, ret MUST be 0 thus we continue with the get. */
1326         assert(!ret);
1327
1328         if (!ctf_trace->metadata_received ||
1329                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1330                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1331         }
1332
1333         ret = check_new_streams(conn);
1334         if (ret < 0) {
1335                 goto end_unlock;
1336         } else if (ret == 1) {
1337                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1338         }
1339
1340         pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1341         pthread_mutex_lock(&vstream->overwrite_lock);
1342         if (vstream->abort_flag) {
1343                 /* The file is being overwritten by the writer, we cannot use it. */
1344                 pthread_mutex_unlock(&vstream->overwrite_lock);
1345                 ret = viewer_stream_rotate(vstream, rstream);
1346                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1347                 if (ret < 0) {
1348                         goto end_unlock;
1349                 } else if (ret == 1) {
1350                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1351                         viewer_stream_delete(vstream);
1352                         viewer_stream_destroy(ctf_trace, vstream);
1353                 } else {
1354                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1355                 }
1356                 goto send_reply;
1357         }
1358
1359         read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1360                         sizeof(packet_index));
1361         pthread_mutex_unlock(&vstream->overwrite_lock);
1362         pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1363         if (read_ret < 0) {
1364                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1365                 viewer_stream_delete(vstream);
1366                 viewer_stream_destroy(ctf_trace, vstream);
1367                 goto send_reply;
1368         } else if (read_ret < sizeof(packet_index)) {
1369                 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1370                 if (vstream->close_write_flag) {
1371                         ret = viewer_stream_rotate(vstream, rstream);
1372                         if (ret < 0) {
1373                                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1374                                 goto end_unlock;
1375                         } else if (ret == 1) {
1376                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1377                                 viewer_stream_delete(vstream);
1378                                 viewer_stream_destroy(ctf_trace, vstream);
1379                         } else {
1380                                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1381                         }
1382                 } else {
1383                         ERR("Relay reading index file %d", vstream->index_read_fd);
1384                         viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1385                 }
1386                 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1387                 goto send_reply;
1388         } else {
1389                 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1390                 vstream->last_sent_index++;
1391         }
1392
1393         /*
1394          * Indexes are stored in big endian, no need to switch before sending.
1395          */
1396         viewer_index.offset = packet_index.offset;
1397         viewer_index.packet_size = packet_index.packet_size;
1398         viewer_index.content_size = packet_index.content_size;
1399         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1400         viewer_index.timestamp_end = packet_index.timestamp_end;
1401         viewer_index.events_discarded = packet_index.events_discarded;
1402         viewer_index.stream_id = packet_index.stream_id;
1403
1404 send_reply:
1405         viewer_index.flags = htobe32(viewer_index.flags);
1406         health_code_update();
1407
1408         ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1409         if (ret < 0) {
1410                 goto end_unlock;
1411         }
1412         health_code_update();
1413
1414         DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1415                         vstream->last_sent_index, vstream->stream_handle);
1416
1417 end_unlock:
1418         rcu_read_unlock();
1419
1420 end:
1421         return ret;
1422 }
1423
1424 /*
1425  * Send the next index for a stream
1426  *
1427  * Return 0 on success or else a negative value.
1428  */
1429 static
1430 int viewer_get_packet(struct relay_connection *conn)
1431 {
1432         int ret, send_data = 0;
1433         char *data = NULL;
1434         uint32_t len = 0;
1435         ssize_t read_len;
1436         struct lttng_viewer_get_packet get_packet_info;
1437         struct lttng_viewer_trace_packet reply;
1438         struct relay_viewer_stream *stream;
1439         struct relay_session *session;
1440         struct ctf_trace *ctf_trace;
1441
1442         assert(conn);
1443
1444         DBG2("Relay get data packet");
1445
1446         health_code_update();
1447
1448         ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1449         if (ret < 0) {
1450                 goto end;
1451         }
1452         health_code_update();
1453
1454         /* From this point on, the error label can be reached. */
1455         memset(&reply, 0, sizeof(reply));
1456
1457         rcu_read_lock();
1458         stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1459         if (!stream) {
1460                 goto error;
1461         }
1462
1463         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1464         if (!session) {
1465                 ret = -1;
1466                 goto error;
1467         }
1468
1469         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1470                         stream->path_name);
1471         assert(ctf_trace);
1472
1473         /*
1474          * First time we read this stream, we need open the tracefile, we should
1475          * only arrive here if an index has already been sent to the viewer, so the
1476          * tracefile must exist, if it does not it is a fatal error.
1477          */
1478         if (stream->read_fd < 0) {
1479                 char fullpath[PATH_MAX];
1480
1481                 if (stream->tracefile_count > 0) {
1482                         ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1483                                         stream->channel_name,
1484                                         stream->tracefile_count_current);
1485                 } else {
1486                         ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1487                                         stream->channel_name);
1488                 }
1489                 if (ret < 0) {
1490                         goto error;
1491                 }
1492                 ret = open(fullpath, O_RDONLY);
1493                 if (ret < 0) {
1494                         PERROR("Relay opening trace file");
1495                         goto error;
1496                 }
1497                 stream->read_fd = ret;
1498         }
1499
1500         if (!ctf_trace->metadata_received ||
1501                         ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1502                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1503                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1504                 goto send_reply;
1505         }
1506
1507         ret = check_new_streams(conn);
1508         if (ret < 0) {
1509                 goto end_unlock;
1510         } else if (ret == 1) {
1511                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1512                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1513                 goto send_reply;
1514         }
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(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1524         if (ret < 0) {
1525                 /*
1526                  * If the read fd was closed by the streaming side, the
1527                  * abort_flag will be set to 1, otherwise it is an error.
1528                  */
1529                 if (stream->abort_flag == 0) {
1530                         PERROR("lseek");
1531                         goto error;
1532                 }
1533                 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1534                 goto send_reply;
1535         }
1536         read_len = lttng_read(stream->read_fd, data, len);
1537         if (read_len < len) {
1538                 /*
1539                  * If the read fd was closed by the streaming side, the
1540                  * abort_flag will be set to 1, otherwise it is an error.
1541                  */
1542                 if (stream->abort_flag == 0) {
1543                         PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1544                                         stream->read_fd,
1545                                         be64toh(get_packet_info.offset));
1546                         goto error;
1547                 } else {
1548                         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1549                         goto send_reply;
1550                 }
1551         }
1552         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1553         reply.len = htobe32(len);
1554         send_data = 1;
1555         goto send_reply;
1556
1557 error:
1558         reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1559
1560 send_reply:
1561         reply.flags = htobe32(reply.flags);
1562
1563         health_code_update();
1564
1565         ret = send_response(conn->sock, &reply, sizeof(reply));
1566         if (ret < 0) {
1567                 goto end_unlock;
1568         }
1569         health_code_update();
1570
1571         if (send_data) {
1572                 health_code_update();
1573                 ret = send_response(conn->sock, data, len);
1574                 if (ret < 0) {
1575                         goto end_unlock;
1576                 }
1577                 health_code_update();
1578         }
1579
1580         DBG("Sent %u bytes for stream %" PRIu64, len,
1581                         be64toh(get_packet_info.stream_id));
1582
1583 end_unlock:
1584         free(data);
1585         rcu_read_unlock();
1586
1587 end:
1588         return ret;
1589 }
1590
1591 /*
1592  * Send the session's metadata
1593  *
1594  * Return 0 on success else a negative value.
1595  */
1596 static
1597 int viewer_get_metadata(struct relay_connection *conn)
1598 {
1599         int ret = 0;
1600         ssize_t read_len;
1601         uint64_t len = 0;
1602         char *data = NULL;
1603         struct lttng_viewer_get_metadata request;
1604         struct lttng_viewer_metadata_packet reply;
1605         struct relay_viewer_stream *stream;
1606         struct ctf_trace *ctf_trace;
1607         struct relay_session *session;
1608
1609         assert(conn);
1610
1611         DBG("Relay get metadata");
1612
1613         health_code_update();
1614
1615         ret = recv_request(conn->sock, &request, sizeof(request));
1616         if (ret < 0) {
1617                 goto end;
1618         }
1619         health_code_update();
1620
1621         rcu_read_lock();
1622         stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1623         if (!stream || !stream->metadata_flag) {
1624                 ERR("Invalid metadata stream");
1625                 goto error;
1626         }
1627
1628         session = session_find_by_id(conn->sessions_ht, stream->session_id);
1629         if (!session) {
1630                 ret = -1;
1631                 goto error;
1632         }
1633
1634         ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1635                         stream->path_name);
1636         assert(ctf_trace);
1637         assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1638
1639         len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1640         if (len == 0) {
1641                 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1642                 goto send_reply;
1643         }
1644
1645         /* first time, we open the metadata file */
1646         if (stream->read_fd < 0) {
1647                 char fullpath[PATH_MAX];
1648
1649                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1650                                 stream->channel_name);
1651                 if (ret < 0) {
1652                         goto error;
1653                 }
1654                 ret = open(fullpath, O_RDONLY);
1655                 if (ret < 0) {
1656                         PERROR("Relay opening metadata file");
1657                         goto error;
1658                 }
1659                 stream->read_fd = ret;
1660         }
1661
1662         reply.len = htobe64(len);
1663         data = zmalloc(len);
1664         if (!data) {
1665                 PERROR("viewer metadata zmalloc");
1666                 goto error;
1667         }
1668
1669         read_len = lttng_read(stream->read_fd, data, len);
1670         if (read_len < len) {
1671                 PERROR("Relay reading metadata file");
1672                 goto error;
1673         }
1674         ctf_trace->metadata_sent += read_len;
1675         reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1676         goto send_reply;
1677
1678 error:
1679         reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1680
1681 send_reply:
1682         health_code_update();
1683         ret = send_response(conn->sock, &reply, sizeof(reply));
1684         if (ret < 0) {
1685                 goto end_unlock;
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_unlock;
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_unlock:
1702         free(data);
1703         rcu_read_unlock();
1704 end:
1705         return ret;
1706 }
1707
1708 /*
1709  * Create a viewer session.
1710  *
1711  * Return 0 on success or else a negative value.
1712  */
1713 static
1714 int viewer_create_session(struct relay_connection *conn)
1715 {
1716         int ret;
1717         struct lttng_viewer_create_session_response resp;
1718
1719         DBG("Viewer create session received");
1720
1721         resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1722         conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1723         if (!conn->viewer_session) {
1724                 ERR("Allocation viewer session");
1725                 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1726                 goto send_reply;
1727         }
1728         CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1729
1730 send_reply:
1731         health_code_update();
1732         ret = send_response(conn->sock, &resp, sizeof(resp));
1733         if (ret < 0) {
1734                 goto end;
1735         }
1736         health_code_update();
1737         ret = 0;
1738
1739 end:
1740         return ret;
1741 }
1742
1743
1744 /*
1745  * live_relay_unknown_command: send -1 if received unknown command
1746  */
1747 static
1748 void live_relay_unknown_command(struct relay_connection *conn)
1749 {
1750         struct lttcomm_relayd_generic_reply reply;
1751
1752         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1753         (void) send_response(conn->sock, &reply, sizeof(reply));
1754 }
1755
1756 /*
1757  * Process the commands received on the control socket
1758  */
1759 static
1760 int process_control(struct lttng_viewer_cmd *recv_hdr,
1761                 struct relay_connection *conn)
1762 {
1763         int ret = 0;
1764         uint32_t msg_value;
1765
1766         assert(recv_hdr);
1767         assert(conn);
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)", be32toh(recv_hdr->cmd));
1808                 live_relay_unknown_command(conn);
1809                 ret = -1;
1810                 goto end;
1811         }
1812
1813 end:
1814         return ret;
1815 }
1816
1817 static
1818 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1819 {
1820         int ret;
1821
1822         assert(events);
1823
1824         (void) lttng_poll_del(events, pollfd);
1825
1826         ret = close(pollfd);
1827         if (ret < 0) {
1828                 ERR("Closing pollfd %d", pollfd);
1829         }
1830 }
1831
1832 /*
1833  * Delete and destroy a connection.
1834  *
1835  * RCU read side lock MUST be acquired.
1836  */
1837 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1838                 struct relay_connection *conn)
1839 {
1840         struct relay_session *session, *tmp_session;
1841
1842         assert(relay_connections_ht);
1843         assert(conn);
1844
1845         connection_delete(relay_connections_ht, conn);
1846
1847         if (!conn->viewer_session) {
1848                 goto end;
1849         }
1850
1851         rcu_read_lock();
1852         cds_list_for_each_entry_safe(session, tmp_session,
1853                         &conn->viewer_session->sessions_head,
1854                         viewer_session_list) {
1855                 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1856                 cleanup_session(conn, session);
1857         }
1858         rcu_read_unlock();
1859
1860 end:
1861         connection_destroy(conn);
1862 }
1863
1864 /*
1865  * This thread does the actual work
1866  */
1867 static
1868 void *thread_worker(void *data)
1869 {
1870         int ret, err = -1;
1871         uint32_t nb_fd;
1872         struct relay_connection *conn;
1873         struct lttng_poll_event events;
1874         struct lttng_ht *relay_connections_ht;
1875         struct lttng_ht_iter iter;
1876         struct lttng_viewer_cmd recv_hdr;
1877         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1878         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1879
1880         DBG("[thread] Live viewer relay worker started");
1881
1882         rcu_register_thread();
1883
1884         health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1885
1886         if (testpoint(relayd_thread_live_worker)) {
1887                 goto error_testpoint;
1888         }
1889
1890         /* table of connections indexed on socket */
1891         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1892         if (!relay_connections_ht) {
1893                 goto relay_connections_ht_error;
1894         }
1895
1896         ret = create_thread_poll_set(&events, 2);
1897         if (ret < 0) {
1898                 goto error_poll_create;
1899         }
1900
1901         ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1902         if (ret < 0) {
1903                 goto error;
1904         }
1905
1906 restart:
1907         while (1) {
1908                 int i;
1909
1910                 health_code_update();
1911
1912                 /* Infinite blocking call, waiting for transmission */
1913                 DBG3("Relayd live viewer worker thread polling...");
1914                 health_poll_entry();
1915                 ret = lttng_poll_wait(&events, -1);
1916                 health_poll_exit();
1917                 if (ret < 0) {
1918                         /*
1919                          * Restart interrupted system call.
1920                          */
1921                         if (errno == EINTR) {
1922                                 goto restart;
1923                         }
1924                         goto error;
1925                 }
1926
1927                 nb_fd = ret;
1928
1929                 /*
1930                  * Process control. The control connection is prioritised so we don't
1931                  * starve it with high throughput tracing data on the data
1932                  * connection.
1933                  */
1934                 for (i = 0; i < nb_fd; i++) {
1935                         /* Fetch once the poll data */
1936                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1937                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1938
1939                         health_code_update();
1940
1941                         /* Thread quit pipe has been closed. Killing thread. */
1942                         ret = check_thread_quit_pipe(pollfd, revents);
1943                         if (ret) {
1944                                 err = 0;
1945                                 goto exit;
1946                         }
1947
1948                         /* Inspect the relay conn pipe for new connection */
1949                         if (pollfd == live_conn_pipe[0]) {
1950                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1951                                         ERR("Relay live pipe error");
1952                                         goto error;
1953                                 } else if (revents & LPOLLIN) {
1954                                         ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1955                                         if (ret < 0) {
1956                                                 goto error;
1957                                         }
1958                                         conn->sessions_ht = sessions_ht;
1959                                         connection_init(conn);
1960                                         lttng_poll_add(&events, conn->sock->fd,
1961                                                         LPOLLIN | LPOLLRDHUP);
1962                                         rcu_read_lock();
1963                                         lttng_ht_add_unique_ulong(relay_connections_ht,
1964                                                         &conn->sock_n);
1965                                         rcu_read_unlock();
1966                                         DBG("Connection socket %d added", conn->sock->fd);
1967                                 }
1968                         } else {
1969                                 rcu_read_lock();
1970                                 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1971                                 /* If not found, there is a synchronization issue. */
1972                                 assert(conn);
1973
1974                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1975                                         cleanup_connection_pollfd(&events, pollfd);
1976                                         destroy_connection(relay_connections_ht, conn);
1977                                 } else if (revents & LPOLLIN) {
1978                                         ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1979                                                         sizeof(recv_hdr), 0);
1980                                         if (ret <= 0) {
1981                                                 /* Connection closed */
1982                                                 cleanup_connection_pollfd(&events, pollfd);
1983                                                 destroy_connection(relay_connections_ht, conn);
1984                                                 DBG("Viewer control conn closed with %d", pollfd);
1985                                         } else {
1986                                                 ret = process_control(&recv_hdr, conn);
1987                                                 if (ret < 0) {
1988                                                         /* Clear the session on error. */
1989                                                         cleanup_connection_pollfd(&events, pollfd);
1990                                                         destroy_connection(relay_connections_ht, conn);
1991                                                         DBG("Viewer connection closed with %d", pollfd);
1992                                                 }
1993                                         }
1994                                 }
1995                                 rcu_read_unlock();
1996                         }
1997                 }
1998         }
1999
2000 exit:
2001 error:
2002         lttng_poll_clean(&events);
2003
2004         /* Cleanup reamaining connection object. */
2005         rcu_read_lock();
2006         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2007                         sock_n.node) {
2008                 health_code_update();
2009                 destroy_connection(relay_connections_ht, conn);
2010         }
2011         rcu_read_unlock();
2012 error_poll_create:
2013         lttng_ht_destroy(relay_connections_ht);
2014 relay_connections_ht_error:
2015         /* Close relay conn pipes */
2016         utils_close_pipe(live_conn_pipe);
2017         if (err) {
2018                 DBG("Viewer worker thread exited with error");
2019         }
2020         DBG("Viewer worker thread cleanup complete");
2021 error_testpoint:
2022         if (err) {
2023                 health_error();
2024                 ERR("Health error occurred in %s", __func__);
2025         }
2026         health_unregister(health_relayd);
2027         stop_threads();
2028         rcu_unregister_thread();
2029         return NULL;
2030 }
2031
2032 /*
2033  * Create the relay command pipe to wake thread_manage_apps.
2034  * Closed in cleanup().
2035  */
2036 static int create_conn_pipe(void)
2037 {
2038         int ret;
2039
2040         ret = utils_create_pipe_cloexec(live_conn_pipe);
2041
2042         return ret;
2043 }
2044
2045 void live_stop_threads(void)
2046 {
2047         int ret;
2048         void *status;
2049
2050         stop_threads();
2051
2052         ret = pthread_join(live_listener_thread, &status);
2053         if (ret != 0) {
2054                 PERROR("pthread_join live listener");
2055                 goto error;     /* join error, exit without cleanup */
2056         }
2057
2058         ret = pthread_join(live_worker_thread, &status);
2059         if (ret != 0) {
2060                 PERROR("pthread_join live worker");
2061                 goto error;     /* join error, exit without cleanup */
2062         }
2063
2064         ret = pthread_join(live_dispatcher_thread, &status);
2065         if (ret != 0) {
2066                 PERROR("pthread_join live dispatcher");
2067                 goto error;     /* join error, exit without cleanup */
2068         }
2069
2070         cleanup();
2071
2072 error:
2073         return;
2074 }
2075
2076 /*
2077  * main
2078  */
2079 int live_start_threads(struct lttng_uri *uri,
2080                 struct relay_local_data *relay_ctx)
2081 {
2082         int ret = 0;
2083         void *status;
2084         int is_root;
2085
2086         assert(uri);
2087         live_uri = uri;
2088
2089         /* Check if daemon is UID = 0 */
2090         is_root = !getuid();
2091
2092         if (!is_root) {
2093                 if (live_uri->port < 1024) {
2094                         ERR("Need to be root to use ports < 1024");
2095                         ret = -1;
2096                         goto exit;
2097                 }
2098         }
2099
2100         /* Setup the thread apps communication pipe. */
2101         if ((ret = create_conn_pipe()) < 0) {
2102                 goto exit;
2103         }
2104
2105         /* Init relay command queue. */
2106         cds_wfq_init(&viewer_conn_queue.queue);
2107
2108         /* Set up max poll set size */
2109         lttng_poll_set_max_size();
2110
2111         /* Setup the dispatcher thread */
2112         ret = pthread_create(&live_dispatcher_thread, NULL,
2113                         thread_dispatcher, (void *) NULL);
2114         if (ret != 0) {
2115                 PERROR("pthread_create viewer dispatcher");
2116                 goto exit_dispatcher;
2117         }
2118
2119         /* Setup the worker thread */
2120         ret = pthread_create(&live_worker_thread, NULL,
2121                         thread_worker, relay_ctx);
2122         if (ret != 0) {
2123                 PERROR("pthread_create viewer worker");
2124                 goto exit_worker;
2125         }
2126
2127         /* Setup the listener thread */
2128         ret = pthread_create(&live_listener_thread, NULL,
2129                         thread_listener, (void *) NULL);
2130         if (ret != 0) {
2131                 PERROR("pthread_create viewer listener");
2132                 goto exit_listener;
2133         }
2134
2135         ret = 0;
2136         goto end;
2137
2138 exit_listener:
2139         ret = pthread_join(live_listener_thread, &status);
2140         if (ret != 0) {
2141                 PERROR("pthread_join live listener");
2142                 goto error;     /* join error, exit without cleanup */
2143         }
2144
2145 exit_worker:
2146         ret = pthread_join(live_worker_thread, &status);
2147         if (ret != 0) {
2148                 PERROR("pthread_join live worker");
2149                 goto error;     /* join error, exit without cleanup */
2150         }
2151
2152 exit_dispatcher:
2153         ret = pthread_join(live_dispatcher_thread, &status);
2154         if (ret != 0) {
2155                 PERROR("pthread_join live dispatcher");
2156                 goto error;     /* join error, exit without cleanup */
2157         }
2158
2159 exit:
2160         cleanup();
2161
2162 end:
2163 error:
2164         return ret;
2165 }