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/sessiond-comm/sessiond-comm.h>
  49 #include <common/sessiond-comm/inet.h>
  50 #include <common/sessiond-comm/relayd.h>
  51 #include <common/uri.h>
  52 #include <common/utils.h>
  53
  54 #include "cmd.h"
  55 #include "live.h"
  56 #include "lttng-relayd.h"
  57 #include "lttng-viewer.h"
  58 #include "utils.h"
  59
  60 static struct lttng_uri *live_uri;
  61
  62 /*
  63  * Quit pipe for all threads. This permits a single cancellation point
  64  * for all threads when receiving an event on the pipe.
  65  */
  66 static int live_thread_quit_pipe[2] = { -1, -1 };
  67
  68 /*
  69  * This pipe is used to inform the worker thread that a command is queued and
  70  * ready to be processed.
  71  */
  72 static int live_relay_cmd_pipe[2] = { -1, -1 };
  73
  74 /* Shared between threads */
  75 static int live_dispatch_thread_exit;
  76
  77 static pthread_t live_listener_thread;
  78 static pthread_t live_dispatcher_thread;
  79 static pthread_t live_worker_thread;
  80
  81 /*
  82  * Relay command queue.
  83  *
  84  * The live_thread_listener and live_thread_dispatcher communicate with this
  85  * queue.
  86  */
  87 static struct relay_cmd_queue viewer_cmd_queue;
  88
  89 static uint64_t last_relay_viewer_session_id;
  90
  91 /*
  92  * Cleanup the daemon
  93  */
  94 static
  95 void cleanup(void)
  96 {
  97         DBG("Cleaning up");
  98
  99         /* Close thread quit pipes */
 100         utils_close_pipe(live_thread_quit_pipe);
 101         free(live_uri);
 102 }
 103
 104 /*
 105  * Write to writable pipe used to notify a thread.
 106  */
 107 static
 108 int notify_thread_pipe(int wpipe)
 109 {
 110         int ret;
 111
 112         do {
 113                 ret = write(wpipe, "!", 1);
 114         } while (ret < 0 && errno == EINTR);
 115         if (ret < 0 || ret != 1) {
 116                 PERROR("write poll pipe");
 117         }
 118
 119         return ret;
 120 }
 121
 122 /*
 123  * Stop all threads by closing the thread quit pipe.
 124  */
 125 static
 126 void stop_threads(void)
 127 {
 128         int ret;
 129
 130         /* Stopping all threads */
 131         DBG("Terminating all live threads");
 132         ret = notify_thread_pipe(live_thread_quit_pipe[1]);
 133         if (ret < 0) {
 134                 ERR("write error on thread quit pipe");
 135         }
 136
 137         /* Dispatch thread */
 138         CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
 139         futex_nto1_wake(&viewer_cmd_queue.futex);
 140 }
 141
 142 /*
 143  * Init thread quit pipe.
 144  *
 145  * Return -1 on error or 0 if all pipes are created.
 146  */
 147 static
 148 int init_thread_quit_pipe(void)
 149 {
 150         int ret;
 151
 152         ret = utils_create_pipe_cloexec(live_thread_quit_pipe);
 153
 154         return ret;
 155 }
 156
 157 /*
 158  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 159  */
 160 static
 161 int create_thread_poll_set(struct lttng_poll_event *events, int size)
 162 {
 163         int ret;
 164
 165         if (events == NULL || size == 0) {
 166                 ret = -1;
 167                 goto error;
 168         }
 169
 170         ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
 171         if (ret < 0) {
 172                 goto error;
 173         }
 174
 175         /* Add quit pipe */
 176         ret = lttng_poll_add(events, live_thread_quit_pipe[0], LPOLLIN);
 177         if (ret < 0) {
 178                 goto error;
 179         }
 180
 181         return 0;
 182
 183 error:
 184         return ret;
 185 }
 186
 187 /*
 188  * Check if the thread quit pipe was triggered.
 189  *
 190  * Return 1 if it was triggered else 0;
 191  */
 192 static
 193 int check_thread_quit_pipe(int fd, uint32_t events)
 194 {
 195         if (fd == live_thread_quit_pipe[0] && (events & LPOLLIN)) {
 196                 return 1;
 197         }
 198
 199         return 0;
 200 }
 201
 202 /*
 203  * Create and init socket from uri.
 204  */
 205 static
 206 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
 207 {
 208         int ret;
 209         struct lttcomm_sock *sock = NULL;
 210
 211         sock = lttcomm_alloc_sock_from_uri(uri);
 212         if (sock == NULL) {
 213                 ERR("Allocating socket");
 214                 goto error;
 215         }
 216
 217         ret = lttcomm_create_sock(sock);
 218         if (ret < 0) {
 219                 goto error;
 220         }
 221         DBG("Listening on sock %d for live", sock->fd);
 222
 223         ret = sock->ops->bind(sock);
 224         if (ret < 0) {
 225                 goto error;
 226         }
 227
 228         ret = sock->ops->listen(sock, -1);
 229         if (ret < 0) {
 230                 goto error;
 231
 232         }
 233
 234         return sock;
 235
 236 error:
 237         if (sock) {
 238                 lttcomm_destroy_sock(sock);
 239         }
 240         return NULL;
 241 }
 242
 243 /*
 244  * This thread manages the listening for new connections on the network
 245  */
 246 static
 247 void *thread_listener(void *data)
 248 {
 249         int i, ret, pollfd, err = -1;
 250         int val = 1;
 251         uint32_t revents, nb_fd;
 252         struct lttng_poll_event events;
 253         struct lttcomm_sock *live_control_sock;
 254
 255         DBG("[thread] Relay live listener started");
 256
 257         live_control_sock = init_socket(live_uri);
 258         if (!live_control_sock) {
 259                 goto error_sock_control;
 260         }
 261
 262         /*
 263          * Pass 3 as size here for the thread quit pipe, control and data socket.
 264          */
 265         ret = create_thread_poll_set(&events, 2);
 266         if (ret < 0) {
 267                 goto error_create_poll;
 268         }
 269
 270         /* Add the control socket */
 271         ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
 272         if (ret < 0) {
 273                 goto error_poll_add;
 274         }
 275
 276         while (1) {
 277                 DBG("Listener accepting live viewers connections");
 278
 279 restart:
 280                 ret = lttng_poll_wait(&events, -1);
 281                 if (ret < 0) {
 282                         /*
 283                          * Restart interrupted system call.
 284                          */
 285                         if (errno == EINTR) {
 286                                 goto restart;
 287                         }
 288                         goto error;
 289                 }
 290                 nb_fd = ret;
 291
 292                 DBG("Relay new viewer connection received");
 293                 for (i = 0; i < nb_fd; i++) {
 294                         /* Fetch once the poll data */
 295                         revents = LTTNG_POLL_GETEV(&events, i);
 296                         pollfd = LTTNG_POLL_GETFD(&events, i);
 297
 298                         /* Thread quit pipe has been closed. Killing thread. */
 299                         ret = check_thread_quit_pipe(pollfd, revents);
 300                         if (ret) {
 301                                 err = 0;
 302                                 goto exit;
 303                         }
 304
 305                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 306                                 ERR("socket poll error");
 307                                 goto error;
 308                         } else if (revents & LPOLLIN) {
 309                                 /*
 310                                  * Get allocated in this thread, enqueued to a global queue,
 311                                  * dequeued and freed in the worker thread.
 312                                  */
 313                                 struct relay_command *relay_cmd;
 314                                 struct lttcomm_sock *newsock;
 315
 316                                 relay_cmd = zmalloc(sizeof(*relay_cmd));
 317                                 if (!relay_cmd) {
 318                                         PERROR("relay command zmalloc");
 319                                         goto error;
 320                                 }
 321
 322                                 assert(pollfd == live_control_sock->fd);
 323                                 newsock = live_control_sock->ops->accept(live_control_sock);
 324                                 if (!newsock) {
 325                                         PERROR("accepting control sock");
 326                                         free(relay_cmd);
 327                                         goto error;
 328                                 }
 329                                 DBG("Relay viewer connection accepted socket %d", newsock->fd);
 330                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
 331                                                 sizeof(int));
 332                                 if (ret < 0) {
 333                                         PERROR("setsockopt inet");
 334                                         lttcomm_destroy_sock(newsock);
 335                                         free(relay_cmd);
 336                                         goto error;
 337                                 }
 338                                 relay_cmd->sock = newsock;
 339
 340                                 /*
 341                                  * Lock free enqueue the request.
 342                                  */
 343                                 cds_wfq_enqueue(&viewer_cmd_queue.queue, &relay_cmd->node);
 344
 345                                 /*
 346                                  * Wake the dispatch queue futex. Implicit memory
 347                                  * barrier with the exchange in cds_wfq_enqueue.
 348                                  */
 349                                 futex_nto1_wake(&viewer_cmd_queue.futex);
 350                         }
 351                 }
 352         }
 353
 354 exit:
 355 error:
 356 error_poll_add:
 357         lttng_poll_clean(&events);
 358 error_create_poll:
 359         if (live_control_sock->fd >= 0) {
 360                 ret = live_control_sock->ops->close(live_control_sock);
 361                 if (ret) {
 362                         PERROR("close");
 363                 }
 364         }
 365         lttcomm_destroy_sock(live_control_sock);
 366 error_sock_control:
 367         if (err) {
 368                 DBG("Live viewer listener thread exited with error");
 369         }
 370         DBG("Live viewer listener thread cleanup complete");
 371         stop_threads();
 372         return NULL;
 373 }
 374
 375 /*
 376  * This thread manages the dispatching of the requests to worker threads
 377  */
 378 static
 379 void *thread_dispatcher(void *data)
 380 {
 381         int ret;
 382         struct cds_wfq_node *node;
 383         struct relay_command *relay_cmd = NULL;
 384
 385         DBG("[thread] Live viewer relay dispatcher started");
 386
 387         while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
 388                 /* Atomically prepare the queue futex */
 389                 futex_nto1_prepare(&viewer_cmd_queue.futex);
 390
 391                 do {
 392                         /* Dequeue commands */
 393                         node = cds_wfq_dequeue_blocking(&viewer_cmd_queue.queue);
 394                         if (node == NULL) {
 395                                 DBG("Woken up but nothing in the live-viewer "
 396                                                 "relay command queue");
 397                                 /* Continue thread execution */
 398                                 break;
 399                         }
 400
 401                         relay_cmd = caa_container_of(node, struct relay_command, node);
 402                         DBG("Dispatching viewer request waiting on sock %d",
 403                                         relay_cmd->sock->fd);
 404
 405                         /*
 406                          * Inform worker thread of the new request. This call is blocking
 407                          * so we can be assured that the data will be read at some point in
 408                          * time or wait to the end of the world :)
 409                          */
 410                         do {
 411                                 ret = write(live_relay_cmd_pipe[1], relay_cmd,
 412                                                 sizeof(*relay_cmd));
 413                         } while (ret < 0 && errno == EINTR);
 414                         free(relay_cmd);
 415                         if (ret < 0 || ret != sizeof(struct relay_command)) {
 416                                 PERROR("write cmd pipe");
 417                                 goto error;
 418                         }
 419                 } while (node != NULL);
 420
 421                 /* Futex wait on queue. Blocking call on futex() */
 422                 futex_nto1_wait(&viewer_cmd_queue.futex);
 423         }
 424
 425 error:
 426         DBG("Live viewer dispatch thread dying");
 427         stop_threads();
 428         return NULL;
 429 }
 430
 431 /*
 432  * Establish connection with the viewer and check the versions.
 433  *
 434  * Return 0 on success or else negative value.
 435  */
 436 static
 437 int viewer_connect(struct relay_command *cmd)
 438 {
 439         int ret;
 440         struct lttng_viewer_connect reply, msg;
 441
 442         assert(cmd);
 443
 444         cmd->version_check_done = 1;
 445
 446         /* Get version from the other side. */
 447         ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
 448         if (ret < 0 || ret != sizeof(msg)) {
 449                 if (ret == 0) {
 450                         /* Orderly shutdown. Not necessary to print an error. */
 451                         DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
 452                 } else {
 453                         ERR("Relay failed to receive the version values.");
 454                 }
 455                 ret = -1;
 456                 goto end;
 457         }
 458
 459         reply.major = RELAYD_VERSION_COMM_MAJOR;
 460         reply.minor = RELAYD_VERSION_COMM_MINOR;
 461
 462         /* Major versions must be the same */
 463         if (reply.major != be32toh(msg.major)) {
 464                 DBG("Incompatible major versions (%u vs %u)", reply.major,
 465                                 be32toh(msg.major));
 466                 ret = 0;
 467                 goto end;
 468         }
 469
 470         cmd->major = reply.major;
 471         /* We adapt to the lowest compatible version */
 472         if (reply.minor <= be32toh(msg.minor)) {
 473                 cmd->minor = reply.minor;
 474         } else {
 475                 cmd->minor = be32toh(msg.minor);
 476         }
 477
 478         if (be32toh(msg.type) == VIEWER_CLIENT_COMMAND) {
 479                 cmd->type = RELAY_VIEWER_COMMAND;
 480         } else if (be32toh(msg.type) == VIEWER_CLIENT_NOTIFICATION) {
 481                 cmd->type = RELAY_VIEWER_NOTIFICATION;
 482         } else {
 483                 ERR("Unknown connection type : %u", be32toh(msg.type));
 484                 ret = -1;
 485                 goto end;
 486         }
 487
 488         reply.major = htobe32(reply.major);
 489         reply.minor = htobe32(reply.minor);
 490         if (cmd->type == RELAY_VIEWER_COMMAND) {
 491                 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
 492         }
 493         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
 494                         sizeof(struct lttng_viewer_connect), 0);
 495         if (ret < 0) {
 496                 ERR("Relay sending version");
 497         }
 498
 499         DBG("Version check done using protocol %u.%u", cmd->major, cmd->minor);
 500         ret = 0;
 501
 502 end:
 503         return ret;
 504 }
 505
 506 /*
 507  * Send the viewer the list of current sessions.
 508  *
 509  * Return 0 on success or else a negative value.
 510  */
 511 static
 512 int viewer_list_sessions(struct relay_command *cmd,
 513                 struct lttng_ht *sessions_ht)
 514 {
 515         int ret;
 516         struct lttng_viewer_list_sessions session_list;
 517         unsigned long count;
 518         long approx_before, approx_after;
 519         struct lttng_ht_node_ulong *node;
 520         struct lttng_ht_iter iter;
 521         struct lttng_viewer_session send_session;
 522         struct relay_session *session;
 523
 524         DBG("List sessions received");
 525
 526         if (cmd->version_check_done == 0) {
 527                 ERR("Trying to list sessions before version check");
 528                 ret = -1;
 529                 goto end_no_session;
 530         }
 531
 532         rcu_read_lock();
 533         cds_lfht_count_nodes(sessions_ht->ht, &approx_before, &count, &approx_after);
 534         session_list.sessions_count = htobe32(count);
 535
 536         ret = cmd->sock->ops->sendmsg(cmd->sock, &session_list,
 537                         sizeof(session_list), 0);
 538         if (ret < 0) {
 539                 ERR("Relay sending sessions list");
 540                 goto end_unlock;
 541         }
 542
 543         cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, node, node) {
 544                 node = lttng_ht_iter_get_node_ulong(&iter);
 545                 if (!node) {
 546                         goto end_unlock;
 547                 }
 548                 session = caa_container_of(node, struct relay_session, session_n);
 549
 550                 strncpy(send_session.session_name, session->session_name,
 551                                 sizeof(send_session.session_name));
 552                 strncpy(send_session.hostname, session->hostname,
 553                                 sizeof(send_session.hostname));
 554                 send_session.id = htobe64(session->id);
 555                 send_session.live_timer = htobe32(session->live_timer);
 556                 send_session.clients = htobe32(session->viewer_attached);
 557
 558                 ret = cmd->sock->ops->sendmsg(cmd->sock, &send_session,
 559                                 sizeof(send_session), 0);
 560                 if (ret < 0) {
 561                         ERR("Relay sending session info");
 562                         goto end_unlock;
 563                 }
 564         }
 565         rcu_read_unlock();
 566         ret = 0;
 567         goto end;
 568
 569 end_unlock:
 570         rcu_read_unlock();
 571
 572 end:
 573 end_no_session:
 574         return ret;
 575 }
 576
 577 /*
 578  * Allocate and init a new viewer_stream.
 579  *
 580  * Copies the values from the stream passed in parameter and insert the new
 581  * stream in the viewer_streams_ht.
 582  *
 583  * MUST be called with rcu_read_lock held.
 584  *
 585  * Returns 0 on success or a negative value on error.
 586  */
 587 static
 588 int init_viewer_stream(struct relay_stream *stream)
 589 {
 590         int ret;
 591         struct relay_viewer_stream *viewer_stream;
 592
 593         assert(stream);
 594
 595         viewer_stream = zmalloc(sizeof(*viewer_stream));
 596         if (!viewer_stream) {
 597                 PERROR("relay viewer stream zmalloc");
 598                 ret = -1;
 599                 goto error;
 600         }
 601
 602         viewer_stream->read_fd = -1;
 603         viewer_stream->index_read_fd = -1;
 604         viewer_stream->session_id = stream->session->id;
 605         viewer_stream->stream_handle = stream->stream_handle;
 606         viewer_stream->path_name = strndup(stream->path_name,
 607                         LTTNG_VIEWER_PATH_MAX);
 608         viewer_stream->channel_name = strndup(stream->channel_name,
 609                         LTTNG_VIEWER_NAME_MAX);
 610         viewer_stream->total_index_received = stream->total_index_received;
 611         viewer_stream->tracefile_size = stream->tracefile_size;
 612         viewer_stream->tracefile_count = stream->tracefile_count;
 613         viewer_stream->metadata_flag = stream->metadata_flag;
 614
 615         /*
 616          * This is to avoid a race between the initialization of this object and
 617          * the close of the given stream. If the stream is unable to find this
 618          * viewer stream when closing, this copy will at least take the latest
 619          * value.
 620          */
 621         viewer_stream->total_index_received = stream->total_index_received;
 622
 623         /*
 624          * The deletion of this ctf_trace object is only done in a call RCU of the
 625          * relay stream making it valid as long as we have the read side lock.
 626          */
 627         viewer_stream->ctf_trace = stream->ctf_trace;
 628         uatomic_inc(&viewer_stream->ctf_trace->refcount);
 629
 630         lttng_ht_node_init_u64(&viewer_stream->stream_n, stream->stream_handle);
 631         lttng_ht_add_unique_u64(viewer_streams_ht, &viewer_stream->stream_n);
 632
 633         ret = 0;
 634
 635 error:
 636         return ret;
 637 }
 638
 639 /*
 640  * Send the viewer the list of current sessions.
 641  */
 642 static
 643 int viewer_attach_session(struct relay_command *cmd,
 644                 struct lttng_ht *sessions_ht)
 645 {
 646         int ret, send_streams = 0, nb_streams = 0;
 647         struct lttng_viewer_attach_session_request request;
 648         struct lttng_viewer_attach_session_response response;
 649         struct lttng_viewer_stream send_stream;
 650         struct relay_stream *stream;
 651         struct relay_viewer_stream *viewer_stream;
 652         struct lttng_ht_node_ulong *node;
 653         struct lttng_ht_node_u64 *node64;
 654         struct lttng_ht_iter iter;
 655         struct relay_session *session;
 656
 657         assert(cmd);
 658         assert(sessions_ht);
 659
 660         DBG("Attach session received");
 661
 662         if (cmd->version_check_done == 0) {
 663                 ERR("Trying to attach session before version check");
 664                 ret = -1;
 665                 goto end_no_session;
 666         }
 667
 668         ret = cmd->sock->ops->recvmsg(cmd->sock, &request, sizeof(request), 0);
 669         if (ret < 0 || ret != sizeof(request)) {
 670                 if (ret == 0) {
 671                         /* Orderly shutdown. Not necessary to print an error. */
 672                         DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
 673                 } else {
 674                         ERR("Relay failed to receive the attach parameters.");
 675                 }
 676                 ret = -1;
 677                 goto error;
 678         }
 679
 680         rcu_read_lock();
 681         lttng_ht_lookup(sessions_ht,
 682                         (void *)((unsigned long) be64toh(request.session_id)), &iter);
 683         node = lttng_ht_iter_get_node_ulong(&iter);
 684         if (node == NULL) {
 685                 DBG("Relay session %" PRIu64 " not found",
 686                                 be64toh(request.session_id));
 687                 response.status = htobe32(VIEWER_ATTACH_UNK);
 688                 goto send_reply;
 689         }
 690
 691         session = caa_container_of(node, struct relay_session, session_n);
 692         if (cmd->session_id == session->id) {
 693                 /* Same viewer already attached, just send the stream list. */
 694                 send_streams = 1;
 695                 response.status = htobe32(VIEWER_ATTACH_OK);
 696         } else if (session->viewer_attached != 0) {
 697                 DBG("Already a viewer attached");
 698                 response.status = htobe32(VIEWER_ATTACH_ALREADY);
 699                 goto send_reply;
 700         } else if (session->live_timer == 0) {
 701                 DBG("Not live session");
 702                 response.status = htobe32(VIEWER_ATTACH_NOT_LIVE);
 703                 goto send_reply;
 704         } else {
 705                 session->viewer_attached++;
 706                 send_streams = 1;
 707                 response.status = htobe32(VIEWER_ATTACH_OK);
 708                 cmd->session_id = session->id;
 709                 cmd->session = session;
 710         }
 711
 712         switch (be32toh(request.seek)) {
 713         case VIEWER_SEEK_BEGINNING:
 714                 /* Default behaviour. */
 715                 break;
 716         case VIEWER_SEEK_LAST:
 717                 /* TODO */
 718                 break;
 719         default:
 720                 ERR("Wrong seek parameter");
 721                 response.status = htobe32(VIEWER_ATTACH_SEEK_ERR);
 722                 send_streams = 0;
 723                 goto send_reply;
 724         }
 725
 726         if (send_streams) {
 727                 /* We should only be there if we have a session to attach to. */
 728                 assert(session);
 729
 730                 /*
 731                  * Fill the viewer_streams_ht to count the number of streams
 732                  * ready to be sent and avoid concurrency issues on the
 733                  * relay_streams_ht and don't rely on a total session stream count.
 734                  */
 735                 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
 736                         struct relay_viewer_stream *vstream;
 737
 738                         node = lttng_ht_iter_get_node_ulong(&iter);
 739                         if (!node) {
 740                                 continue;
 741                         }
 742                         stream = caa_container_of(node, struct relay_stream, stream_n);
 743                         if (stream->session != cmd->session) {
 744                                 continue;
 745                         }
 746
 747                         /*
 748                          * Don't send streams with no ctf_trace, they are not ready to be
 749                          * read.
 750                          */
 751                         if (!stream->ctf_trace) {
 752                                 continue;
 753                         }
 754
 755                         vstream = live_find_viewer_stream_by_id(stream->stream_handle);
 756                         if (!vstream) {
 757                                 ret = init_viewer_stream(stream);
 758                                 if (ret < 0) {
 759                                         goto end_unlock;
 760                                 }
 761                         }
 762                         nb_streams++;
 763                 }
 764                 response.streams_count = htobe32(nb_streams);
 765         }
 766
 767 send_reply:
 768         ret = cmd->sock->ops->sendmsg(cmd->sock, &response, sizeof(response), 0);
 769         if (ret < 0) {
 770                 ERR("Relay sending viewer attach response");
 771                 goto end_unlock;
 772         }
 773
 774         /*
 775          * Unknown or busy session, just return gracefully, the viewer knows what
 776          * is happening.
 777          */
 778         if (!send_streams) {
 779                 ret = 0;
 780                 goto end_unlock;
 781         }
 782
 783         /* We should only be there if we have a session to attach to. */
 784         assert(session);
 785         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
 786                 node64 = lttng_ht_iter_get_node_u64(&iter);
 787                 if (!node64) {
 788                         continue;
 789                 }
 790                 viewer_stream = caa_container_of(node64, struct relay_viewer_stream,
 791                                 stream_n);
 792                 if (viewer_stream->session_id != cmd->session->id) {
 793                         continue;
 794                 }
 795
 796                 send_stream.id = htobe64(viewer_stream->stream_handle);
 797                 send_stream.ctf_trace_id = htobe64(viewer_stream->ctf_trace->id);
 798                 send_stream.metadata_flag = htobe32(viewer_stream->metadata_flag);
 799                 strncpy(send_stream.path_name, viewer_stream->path_name,
 800                                 sizeof(send_stream.path_name));
 801                 strncpy(send_stream.channel_name, viewer_stream->channel_name,
 802                                 sizeof(send_stream.channel_name));
 803
 804                 ret = cmd->sock->ops->sendmsg(cmd->sock, &send_stream,
 805                                 sizeof(send_stream), 0);
 806                 if (ret < 0) {
 807                         ERR("Relay sending stream %" PRIu64, viewer_stream->stream_handle);
 808                         goto end_unlock;
 809                 }
 810                 DBG("Sent stream %" PRIu64 " to viewer", viewer_stream->stream_handle);
 811         }
 812         ret = 0;
 813
 814 end_unlock:
 815         rcu_read_unlock();
 816 end_no_session:
 817 error:
 818         return ret;
 819 }
 820
 821 /*
 822  * Open index file using a given viewer stream.
 823  *
 824  * Return 0 on success or else a negative value.
 825  */
 826 static int open_index(struct relay_viewer_stream *stream)
 827 {
 828         int ret;
 829         char fullpath[PATH_MAX];
 830         struct lttng_packet_index_file_hdr hdr;
 831
 832         if (stream->tracefile_size > 0) {
 833                 /* For now we don't support on-disk ring buffer. */
 834                 ret = -1;
 835                 goto end;
 836         } else {
 837                 ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR
 838                                 "/%s" DEFAULT_INDEX_FILE_SUFFIX,
 839                                 stream->path_name, stream->channel_name);
 840                 if (ret < 0) {
 841                         PERROR("snprintf index path");
 842                         goto error;
 843                 }
 844         }
 845
 846         DBG("Opening index file %s in read only", fullpath);
 847         ret = open(fullpath, O_RDONLY);
 848         if (ret < 0) {
 849                 if (errno == ENOENT) {
 850                         ret = ENOENT;
 851                         goto error;
 852                 } else {
 853                         PERROR("opening index in read-only");
 854                 }
 855                 goto error;
 856         }
 857         stream->index_read_fd = ret;
 858         DBG("Opening index file %s in read only, (fd: %d)", fullpath, ret);
 859
 860         do {
 861                 ret = read(stream->index_read_fd, &hdr, sizeof(hdr));
 862         } while (ret < 0 && errno == EINTR);
 863         if (ret < 0) {
 864                 PERROR("Reading index header");
 865                 goto error;
 866         }
 867         if (strncmp(hdr.magic, INDEX_MAGIC, sizeof(hdr.magic)) != 0) {
 868                 ERR("Invalid header magic");
 869                 ret = -1;
 870                 goto error;
 871         }
 872         if (be32toh(hdr.index_major) != INDEX_MAJOR ||
 873                         be32toh(hdr.index_minor) != INDEX_MINOR) {
 874                 ERR("Invalid header version");
 875                 ret = -1;
 876                 goto error;
 877         }
 878         ret = 0;
 879
 880 error:
 881 end:
 882         return ret;
 883 }
 884
 885 /*
 886  * Get viewer stream from stream id.
 887  *
 888  * RCU read side lock MUST be acquired.
 889  */
 890 struct relay_viewer_stream *live_find_viewer_stream_by_id(uint64_t stream_id)
 891 {
 892         struct lttng_ht_node_u64 *node;
 893         struct lttng_ht_iter iter;
 894         struct relay_viewer_stream *stream = NULL;
 895
 896         lttng_ht_lookup(viewer_streams_ht, &stream_id, &iter);
 897         node = lttng_ht_iter_get_node_u64(&iter);
 898         if (node == NULL) {
 899                 DBG("Relay viewer stream %" PRIu64 " not found", stream_id);
 900                 goto end;
 901         }
 902         stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
 903
 904 end:
 905         return stream;
 906 }
 907
 908 /*
 909  * Send the next index for a stream.
 910  *
 911  * Return 0 on success or else a negative value.
 912  */
 913 static
 914 int viewer_get_next_index(struct relay_command *cmd,
 915                 struct lttng_ht *sessions_ht)
 916 {
 917         int ret;
 918         struct lttng_viewer_get_next_index request_index;
 919         struct lttng_viewer_index viewer_index;
 920         struct lttng_packet_index packet_index;
 921         struct relay_viewer_stream *vstream;
 922         struct relay_stream *rstream;
 923
 924         assert(cmd);
 925         assert(sessions_ht);
 926
 927         DBG("Viewer get next index");
 928
 929         if (cmd->version_check_done == 0) {
 930                 ERR("Trying to request index before version check");
 931                 ret = -1;
 932                 goto end_no_session;
 933         }
 934
 935         ret = cmd->sock->ops->recvmsg(cmd->sock, &request_index,
 936                         sizeof(request_index), 0);
 937         if (ret < 0 || ret != sizeof(request_index)) {
 938                 ret = -1;
 939                 ERR("Relay didn't receive the whole packet");
 940                 goto end;
 941         }
 942
 943         rcu_read_lock();
 944         vstream = live_find_viewer_stream_by_id(be64toh(request_index.stream_id));
 945         if (!vstream) {
 946                 ret = -1;
 947                 goto end_unlock;
 948         }
 949
 950         memset(&viewer_index, 0, sizeof(viewer_index));
 951
 952         /*
 953          * The viewer should not ask for index on metadata stream.
 954          */
 955         if (vstream->metadata_flag) {
 956                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
 957                 goto send_reply;
 958         }
 959
 960         /* First time, we open the index file */
 961         if (vstream->index_read_fd < 0) {
 962                 ret = open_index(vstream);
 963                 if (ret == ENOENT) {
 964                         /*
 965                          * The index is created only when the first data packet arrives, it
 966                          * might not be ready at the beginning of the session
 967                          */
 968                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
 969                         goto send_reply;
 970                 } else if (ret < 0) {
 971                         viewer_index.status = htobe32(VIEWER_INDEX_ERR);
 972                         goto send_reply;
 973                 }
 974         }
 975
 976         rstream = relay_stream_find_by_id(vstream->stream_handle);
 977         if (rstream) {
 978                 if (rstream->beacon_ts_end != -1ULL &&
 979                                 vstream->last_sent_index == rstream->total_index_received) {
 980                         viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE);
 981                         viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
 982                         goto send_reply;
 983                 }
 984
 985                 if (rstream->total_index_received <= vstream->last_sent_index) {
 986                         /* No new index to send, retry later. */
 987                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
 988                         goto send_reply;
 989                 }
 990         } else if (!rstream &&
 991                         vstream->total_index_received == vstream->last_sent_index) {
 992                 /* Last index sent and stream closed */
 993                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
 994                 goto send_reply;
 995         }
 996
 997         if (!vstream->ctf_trace->metadata_received ||
 998                         vstream->ctf_trace->metadata_received >
 999                         vstream->ctf_trace->metadata_sent) {
1000                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1001         }
1002
1003         do {
1004                 ret = read(vstream->index_read_fd, &packet_index,
1005                                 sizeof(packet_index));
1006         } while (ret < 0 && errno == EINTR);
1007         if (ret < sizeof(packet_index)) {
1008                 PERROR("Relay reading index file");
1009                 viewer_index.status = htobe32(VIEWER_INDEX_ERR);
1010         } else {
1011                 viewer_index.status = htobe32(VIEWER_INDEX_OK);
1012                 vstream->last_sent_index++;
1013         }
1014
1015         /*
1016          * Indexes are stored in big endian, no need to switch before sending.
1017          */
1018         viewer_index.offset = packet_index.offset;
1019         viewer_index.packet_size = packet_index.packet_size;
1020         viewer_index.content_size = packet_index.content_size;
1021         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1022         viewer_index.timestamp_end = packet_index.timestamp_end;
1023         viewer_index.events_discarded = packet_index.events_discarded;
1024         viewer_index.stream_id = packet_index.stream_id;
1025
1026 send_reply:
1027         viewer_index.flags = htobe32(viewer_index.flags);
1028         ret = cmd->sock->ops->sendmsg(cmd->sock, &viewer_index,
1029                         sizeof(viewer_index), 0);
1030         if (ret < 0) {
1031                 ERR("Relay index to viewer");
1032                 goto end_unlock;
1033         }
1034
1035         DBG("Index %" PRIu64 "for stream %" PRIu64 "sent",
1036                         vstream->last_sent_index, vstream->stream_handle);
1037
1038 end_unlock:
1039         rcu_read_unlock();
1040
1041 end_no_session:
1042 end:
1043         return ret;
1044 }
1045
1046 /*
1047  * Send the next index for a stream
1048  *
1049  * Return 0 on success or else a negative value.
1050  */
1051 static
1052 int viewer_get_packet(struct relay_command *cmd)
1053 {
1054         int ret, send_data = 0;
1055         char *data = NULL;
1056         uint32_t len = 0;
1057         ssize_t read_len;
1058         struct lttng_viewer_get_packet get_packet_info;
1059         struct lttng_viewer_trace_packet reply;
1060         struct relay_viewer_stream *stream;
1061
1062         assert(cmd);
1063
1064         DBG2("Relay get data packet");
1065
1066         if (cmd->version_check_done == 0) {
1067                 ERR("Trying to get packet before version check");
1068                 ret = -1;
1069                 goto end;
1070         }
1071
1072         ret = cmd->sock->ops->recvmsg(cmd->sock, &get_packet_info,
1073                         sizeof(get_packet_info), 0);
1074         if (ret < 0 || ret != sizeof(get_packet_info)) {
1075                 ret = -1;
1076                 ERR("Relay didn't receive the whole packet");
1077                 goto end;
1078         }
1079
1080         rcu_read_lock();
1081         stream = live_find_viewer_stream_by_id(be64toh(get_packet_info.stream_id));
1082         if (!stream) {
1083                 goto error;
1084         }
1085         assert(stream->ctf_trace);
1086
1087         /*
1088          * First time we read this stream, we need open the tracefile, we should
1089          * only arrive here if an index has already been sent to the viewer, so the
1090          * tracefile must exist, if it does not it is a fatal error.
1091          */
1092         if (stream->read_fd < 0) {
1093                 char fullpath[PATH_MAX];
1094
1095                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1096                                 stream->channel_name);
1097                 if (ret < 0) {
1098                         goto error;
1099                 }
1100                 ret = open(fullpath, O_RDONLY);
1101                 if (ret < 0) {
1102                         PERROR("Relay opening trace file");
1103                         goto error;
1104                 }
1105                 stream->read_fd = ret;
1106         }
1107
1108         memset(&reply, 0, sizeof(reply));
1109
1110         if (!stream->ctf_trace->metadata_received ||
1111                         stream->ctf_trace->metadata_received >
1112                         stream->ctf_trace->metadata_sent) {
1113                 reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1114                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1115
1116                 goto send_reply;
1117         }
1118
1119         len = be32toh(get_packet_info.len);
1120         data = zmalloc(len);
1121         if (!data) {
1122                 PERROR("relay data zmalloc");
1123                 goto error;
1124         }
1125
1126         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1127         if (ret < 0) {
1128                 PERROR("lseek");
1129                 goto error;
1130         }
1131         read_len = read(stream->read_fd, data, len);
1132         if (read_len < (ssize_t) len) {
1133                 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1134                                 stream->read_fd, be64toh(get_packet_info.offset));
1135                 goto error;
1136         }
1137         reply.status = htobe32(VIEWER_GET_PACKET_OK);
1138         reply.len = htobe32(len);
1139         send_data = 1;
1140         goto send_reply;
1141
1142 error:
1143         reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1144
1145 send_reply:
1146         reply.flags = htobe32(reply.flags);
1147         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1148         if (ret < 0) {
1149                 ERR("Relay data header to viewer");
1150                 goto end_unlock;
1151         }
1152
1153         if (send_data) {
1154                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1155                 if (ret < 0) {
1156                         ERR("Relay send data to viewer");
1157                         goto end_unlock;
1158                 }
1159         }
1160
1161         DBG("Sent %u bytes for stream %" PRIu64, len,
1162                         be64toh(get_packet_info.stream_id));
1163
1164 end_unlock:
1165         free(data);
1166         rcu_read_unlock();
1167
1168 end:
1169         return ret;
1170 }
1171
1172 /*
1173  * Send the session's metadata
1174  *
1175  * Return 0 on success else a negative value.
1176  */
1177 static
1178 int viewer_get_metadata(struct relay_command *cmd)
1179 {
1180         int ret = 0;
1181         ssize_t read_len;
1182         uint64_t len = 0;
1183         char *data = NULL;
1184         struct lttng_viewer_get_metadata request;
1185         struct lttng_viewer_metadata_packet reply;
1186         struct relay_viewer_stream *stream;
1187
1188         assert(cmd);
1189
1190         DBG("Relay get metadata");
1191
1192         if (cmd->version_check_done == 0) {
1193                 ERR("Trying to get metadata before version check");
1194                 ret = -1;
1195                 goto end;
1196         }
1197
1198         ret = cmd->sock->ops->recvmsg(cmd->sock, &request,
1199                         sizeof(request), 0);
1200         if (ret < 0 || ret != sizeof(request)) {
1201                 ret = -1;
1202                 ERR("Relay didn't receive the whole packet");
1203                 goto end;
1204         }
1205
1206         rcu_read_lock();
1207         stream = live_find_viewer_stream_by_id(be64toh(request.stream_id));
1208         if (!stream || !stream->metadata_flag) {
1209                 ERR("Invalid metadata stream");
1210                 goto error;
1211         }
1212         assert(stream->ctf_trace);
1213         assert(stream->ctf_trace->metadata_sent <=
1214                         stream->ctf_trace->metadata_received);
1215
1216         len = stream->ctf_trace->metadata_received -
1217                 stream->ctf_trace->metadata_sent;
1218         if (len == 0) {
1219                 reply.status = htobe32(VIEWER_NO_NEW_METADATA);
1220                 goto send_reply;
1221         }
1222
1223         /* first time, we open the metadata file */
1224         if (stream->read_fd < 0) {
1225                 char fullpath[PATH_MAX];
1226
1227                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1228                                 stream->channel_name);
1229                 if (ret < 0) {
1230                         goto error;
1231                 }
1232                 ret = open(fullpath, O_RDONLY);
1233                 if (ret < 0) {
1234                         PERROR("Relay opening metadata file");
1235                         goto error;
1236                 }
1237                 stream->read_fd = ret;
1238         }
1239
1240         reply.len = htobe64(len);
1241         data = zmalloc(len);
1242         if (!data) {
1243                 PERROR("viewer metadata zmalloc");
1244                 goto error;
1245         }
1246
1247         read_len = read(stream->read_fd, data, len);
1248         if (read_len < (ssize_t) len) {
1249                 PERROR("Relay reading metadata file");
1250                 goto error;
1251         }
1252         stream->ctf_trace->metadata_sent += read_len;
1253         reply.status = htobe32(VIEWER_METADATA_OK);
1254         goto send_reply;
1255
1256 error:
1257         reply.status = htobe32(VIEWER_METADATA_ERR);
1258
1259 send_reply:
1260         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1261         if (ret < 0) {
1262                 ERR("Relay data header to viewer");
1263                 goto end_unlock;
1264         }
1265
1266         if (len > 0) {
1267                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1268                 if (ret < 0) {
1269                         ERR("Relay send data to viewer");
1270                         goto end_unlock;
1271                 }
1272         }
1273
1274         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1275                         be64toh(request.stream_id));
1276
1277         DBG("Metadata sent");
1278
1279 end_unlock:
1280         free(data);
1281         rcu_read_unlock();
1282 end:
1283         return ret;
1284 }
1285
1286 /*
1287  * live_relay_unknown_command: send -1 if received unknown command
1288  */
1289 static
1290 void live_relay_unknown_command(struct relay_command *cmd)
1291 {
1292         struct lttcomm_relayd_generic_reply reply;
1293         int ret;
1294
1295         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1296         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1297                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1298         if (ret < 0) {
1299                 ERR("Relay sending unknown command");
1300         }
1301 }
1302
1303 /*
1304  * Process the commands received on the control socket
1305  */
1306 static
1307 int process_control(struct lttng_viewer_cmd *recv_hdr,
1308                 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1309 {
1310         int ret = 0;
1311
1312         switch (be32toh(recv_hdr->cmd)) {
1313         case VIEWER_CONNECT:
1314                 ret = viewer_connect(cmd);
1315                 break;
1316         case VIEWER_LIST_SESSIONS:
1317                 ret = viewer_list_sessions(cmd, sessions_ht);
1318                 break;
1319         case VIEWER_ATTACH_SESSION:
1320                 ret = viewer_attach_session(cmd, sessions_ht);
1321                 break;
1322         case VIEWER_GET_NEXT_INDEX:
1323                 ret = viewer_get_next_index(cmd, sessions_ht);
1324                 break;
1325         case VIEWER_GET_PACKET:
1326                 ret = viewer_get_packet(cmd);
1327                 break;
1328         case VIEWER_GET_METADATA:
1329                 ret = viewer_get_metadata(cmd);
1330                 break;
1331         default:
1332                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1333                 live_relay_unknown_command(cmd);
1334                 ret = -1;
1335                 goto end;
1336         }
1337
1338 end:
1339         return ret;
1340 }
1341
1342 static
1343 void cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1344 {
1345         int ret;
1346
1347         assert(events);
1348
1349         lttng_poll_del(events, pollfd);
1350
1351         ret = close(pollfd);
1352         if (ret < 0) {
1353                 ERR("Closing pollfd %d", pollfd);
1354         }
1355 }
1356
1357 /*
1358  * Create and add connection to the given hash table.
1359  *
1360  * Return poll add value or else -1 on error.
1361  */
1362 static
1363 int add_connection(int fd, struct lttng_poll_event *events,
1364                 struct lttng_ht *relay_connections_ht)
1365 {
1366         int ret;
1367         struct relay_command *relay_connection;
1368
1369         assert(events);
1370         assert(relay_connections_ht);
1371
1372         relay_connection = zmalloc(sizeof(struct relay_command));
1373         if (relay_connection == NULL) {
1374                 PERROR("Relay command zmalloc");
1375                 goto error;
1376         }
1377
1378         do {
1379                 ret = read(fd, relay_connection, sizeof(*relay_connection));
1380         } while (ret < 0 && errno == EINTR);
1381         if (ret < 0 || ret < sizeof(*relay_connection)) {
1382                 PERROR("read relay cmd pipe");
1383                 goto error_read;
1384         }
1385
1386         lttng_ht_node_init_ulong(&relay_connection->sock_n,
1387                         (unsigned long) relay_connection->sock->fd);
1388         rcu_read_lock();
1389         lttng_ht_add_unique_ulong(relay_connections_ht,
1390                         &relay_connection->sock_n);
1391         rcu_read_unlock();
1392
1393         return lttng_poll_add(events, relay_connection->sock->fd,
1394                         LPOLLIN | LPOLLRDHUP);
1395
1396 error_read:
1397         free(relay_connection);
1398 error:
1399         return -1;
1400 }
1401
1402 static
1403 void deferred_free_connection(struct rcu_head *head)
1404 {
1405         struct relay_command *relay_connection =
1406                 caa_container_of(head, struct relay_command, rcu_node);
1407
1408         if (relay_connection->session &&
1409                         relay_connection->session->viewer_attached > 0) {
1410                 relay_connection->session->viewer_attached--;
1411         }
1412         lttcomm_destroy_sock(relay_connection->sock);
1413         free(relay_connection);
1414 }
1415
1416 static
1417 void deferred_free_viewer_stream(struct rcu_head *head)
1418 {
1419         struct relay_viewer_stream *stream =
1420                 caa_container_of(head, struct relay_viewer_stream, rcu_node);
1421
1422         if (stream->ctf_trace) {
1423                 uatomic_dec(&stream->ctf_trace->refcount);
1424                 assert(uatomic_read(&stream->ctf_trace->refcount) >= 0);
1425                 if (uatomic_read(&stream->ctf_trace->refcount) == 0) {
1426                         DBG("Freeing ctf_trace %" PRIu64, stream->ctf_trace->id);
1427                         free(stream->ctf_trace);
1428                 }
1429         }
1430
1431         free(stream->path_name);
1432         free(stream->channel_name);
1433         free(stream);
1434 }
1435
1436 static
1437 void viewer_del_streams(uint64_t session_id)
1438 {
1439         int ret;
1440         struct relay_viewer_stream *stream;
1441         struct lttng_ht_node_u64 *node;
1442         struct lttng_ht_iter iter;
1443
1444         rcu_read_lock();
1445         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
1446                 node = lttng_ht_iter_get_node_u64(&iter);
1447                 if (!node) {
1448                         continue;
1449                 }
1450
1451                 stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
1452                 if (stream->session_id != session_id) {
1453                         continue;
1454                 }
1455
1456                 if (stream->read_fd > 0) {
1457                         ret = close(stream->read_fd);
1458                         if (ret < 0) {
1459                                 PERROR("close read_fd");
1460                         }
1461                 }
1462                 if (stream->index_read_fd > 0) {
1463                         ret = close(stream->index_read_fd);
1464                         if (ret < 0) {
1465                                 PERROR("close index_read_fd");
1466                         }
1467                 }
1468                 if (stream->metadata_flag && stream->ctf_trace) {
1469                         stream->ctf_trace->metadata_sent = 0;
1470                 }
1471                 ret = lttng_ht_del(viewer_streams_ht, &iter);
1472                 assert(!ret);
1473                 call_rcu(&stream->rcu_node, deferred_free_viewer_stream);
1474         }
1475         rcu_read_unlock();
1476 }
1477
1478 /*
1479  * Delete and free a connection.
1480  *
1481  * RCU read side lock MUST be acquired.
1482  */
1483 static
1484 void del_connection(struct lttng_ht *relay_connections_ht,
1485                 struct lttng_ht_iter *iter, struct relay_command *relay_connection)
1486 {
1487         int ret;
1488
1489         assert(relay_connections_ht);
1490         assert(iter);
1491         assert(relay_connection);
1492
1493         ret = lttng_ht_del(relay_connections_ht, iter);
1494         assert(!ret);
1495
1496         viewer_del_streams(relay_connection->session_id);
1497
1498         call_rcu(&relay_connection->rcu_node, deferred_free_connection);
1499 }
1500
1501 /*
1502  * This thread does the actual work
1503  */
1504 static
1505 void *thread_worker(void *data)
1506 {
1507         int ret, err = -1;
1508         uint32_t nb_fd;
1509         struct relay_command *relay_connection;
1510         struct lttng_poll_event events;
1511         struct lttng_ht *relay_connections_ht;
1512         struct lttng_ht_node_ulong *node;
1513         struct lttng_ht_iter iter;
1514         struct lttng_viewer_cmd recv_hdr;
1515         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1516         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1517
1518         DBG("[thread] Live viewer relay worker started");
1519
1520         rcu_register_thread();
1521
1522         /* table of connections indexed on socket */
1523         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1524         if (!relay_connections_ht) {
1525                 goto relay_connections_ht_error;
1526         }
1527
1528         ret = create_thread_poll_set(&events, 2);
1529         if (ret < 0) {
1530                 goto error_poll_create;
1531         }
1532
1533         ret = lttng_poll_add(&events, live_relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1534         if (ret < 0) {
1535                 goto error;
1536         }
1537
1538 restart:
1539         while (1) {
1540                 int i;
1541
1542                 /* Infinite blocking call, waiting for transmission */
1543                 DBG3("Relayd live viewer worker thread polling...");
1544                 ret = lttng_poll_wait(&events, -1);
1545                 if (ret < 0) {
1546                         /*
1547                          * Restart interrupted system call.
1548                          */
1549                         if (errno == EINTR) {
1550                                 goto restart;
1551                         }
1552                         goto error;
1553                 }
1554
1555                 nb_fd = ret;
1556
1557                 /*
1558                  * Process control. The control connection is prioritised so we don't
1559                  * starve it with high throughput tracing data on the data
1560                  * connection.
1561                  */
1562                 for (i = 0; i < nb_fd; i++) {
1563                         /* Fetch once the poll data */
1564                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1565                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1566
1567                         /* Thread quit pipe has been closed. Killing thread. */
1568                         ret = check_thread_quit_pipe(pollfd, revents);
1569                         if (ret) {
1570                                 err = 0;
1571                                 goto exit;
1572                         }
1573
1574                         /* Inspect the relay cmd pipe for new connection */
1575                         if (pollfd == live_relay_cmd_pipe[0]) {
1576                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1577                                         ERR("Relay live pipe error");
1578                                         goto error;
1579                                 } else if (revents & LPOLLIN) {
1580                                         DBG("Relay live viewer command received");
1581                                         ret = add_connection(live_relay_cmd_pipe[0],
1582                                                         &events, relay_connections_ht);
1583                                         if (ret < 0) {
1584                                                 goto error;
1585                                         }
1586                                 }
1587                         } else if (revents) {
1588                                 rcu_read_lock();
1589                                 lttng_ht_lookup(relay_connections_ht,
1590                                                 (void *)((unsigned long) pollfd), &iter);
1591                                 node = lttng_ht_iter_get_node_ulong(&iter);
1592                                 if (node == NULL) {
1593                                         DBG2("Relay viewer sock %d not found", pollfd);
1594                                         rcu_read_unlock();
1595                                         goto error;
1596                                 }
1597                                 relay_connection = caa_container_of(node, struct relay_command,
1598                                                 sock_n);
1599
1600                                 if (revents & (LPOLLERR)) {
1601                                         cleanup_poll_connection(&events, pollfd);
1602                                         del_connection(relay_connections_ht, &iter,
1603                                                         relay_connection);
1604                                 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1605                                         DBG("Viewer socket %d hung up", pollfd);
1606                                         cleanup_poll_connection(&events, pollfd);
1607                                         del_connection(relay_connections_ht, &iter,
1608                                                         relay_connection);
1609                                 } else if (revents & LPOLLIN) {
1610                                         ret = relay_connection->sock->ops->recvmsg(
1611                                                         relay_connection->sock, &recv_hdr,
1612                                                         sizeof(struct lttng_viewer_cmd),
1613                                                         0);
1614                                         /* connection closed */
1615                                         if (ret <= 0) {
1616                                                 cleanup_poll_connection(&events, pollfd);
1617                                                 del_connection( relay_connections_ht, &iter,
1618                                                                 relay_connection);
1619                                                 DBG("Viewer control connection closed with %d",
1620                                                                 pollfd);
1621                                         } else {
1622                                                 if (relay_connection->session) {
1623                                                         DBG2("Relay viewer worker receiving data for "
1624                                                                         "session: %" PRIu64,
1625                                                                         relay_connection->session->id);
1626                                                 }
1627                                                 ret = process_control(&recv_hdr, relay_connection,
1628                                                                 sessions_ht);
1629                                                 if (ret < 0) {
1630                                                         /* Clear the session on error. */
1631                                                         cleanup_poll_connection(&events, pollfd);
1632                                                         del_connection(relay_connections_ht, &iter,
1633                                                                         relay_connection);
1634                                                         DBG("Viewer connection closed with %d", pollfd);
1635                                                 }
1636                                         }
1637                                 }
1638                                 rcu_read_unlock();
1639                         }
1640                 }
1641         }
1642
1643 exit:
1644 error:
1645         lttng_poll_clean(&events);
1646
1647         /* empty the hash table and free the memory */
1648         rcu_read_lock();
1649         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
1650                 node = lttng_ht_iter_get_node_ulong(&iter);
1651                 if (!node) {
1652                         continue;
1653                 }
1654
1655                 relay_connection = caa_container_of(node, struct relay_command,
1656                                 sock_n);
1657                 del_connection(relay_connections_ht, &iter, relay_connection);
1658         }
1659         rcu_read_unlock();
1660 error_poll_create:
1661         lttng_ht_destroy(relay_connections_ht);
1662 relay_connections_ht_error:
1663         /* Close relay cmd pipes */
1664         utils_close_pipe(live_relay_cmd_pipe);
1665         if (err) {
1666                 DBG("Viewer worker thread exited with error");
1667         }
1668         DBG("Viewer worker thread cleanup complete");
1669         stop_threads();
1670         rcu_unregister_thread();
1671         return NULL;
1672 }
1673
1674 /*
1675  * Create the relay command pipe to wake thread_manage_apps.
1676  * Closed in cleanup().
1677  */
1678 static int create_relay_cmd_pipe(void)
1679 {
1680         int ret;
1681
1682         ret = utils_create_pipe_cloexec(live_relay_cmd_pipe);
1683
1684         return ret;
1685 }
1686
1687 void live_stop_threads()
1688 {
1689         int ret;
1690         void *status;
1691
1692         stop_threads();
1693
1694         ret = pthread_join(live_listener_thread, &status);
1695         if (ret != 0) {
1696                 PERROR("pthread_join live listener");
1697                 goto error;     /* join error, exit without cleanup */
1698         }
1699
1700         ret = pthread_join(live_worker_thread, &status);
1701         if (ret != 0) {
1702                 PERROR("pthread_join live worker");
1703                 goto error;     /* join error, exit without cleanup */
1704         }
1705
1706         ret = pthread_join(live_dispatcher_thread, &status);
1707         if (ret != 0) {
1708                 PERROR("pthread_join live dispatcher");
1709                 goto error;     /* join error, exit without cleanup */
1710         }
1711
1712         cleanup();
1713
1714 error:
1715         return;
1716 }
1717
1718 /*
1719  * main
1720  */
1721 int live_start_threads(struct lttng_uri *uri,
1722                 struct relay_local_data *relay_ctx)
1723 {
1724         int ret = 0;
1725         void *status;
1726         int is_root;
1727
1728         assert(uri);
1729         live_uri = uri;
1730
1731         /* Create thread quit pipe */
1732         if ((ret = init_thread_quit_pipe()) < 0) {
1733                 goto error;
1734         }
1735
1736         /* Check if daemon is UID = 0 */
1737         is_root = !getuid();
1738
1739         if (!is_root) {
1740                 if (live_uri->port < 1024) {
1741                         ERR("Need to be root to use ports < 1024");
1742                         ret = -1;
1743                         goto exit;
1744                 }
1745         }
1746
1747         /* Setup the thread apps communication pipe. */
1748         if ((ret = create_relay_cmd_pipe()) < 0) {
1749                 goto exit;
1750         }
1751
1752         /* Init relay command queue. */
1753         cds_wfq_init(&viewer_cmd_queue.queue);
1754
1755         /* Set up max poll set size */
1756         lttng_poll_set_max_size();
1757
1758         /* Setup the dispatcher thread */
1759         ret = pthread_create(&live_dispatcher_thread, NULL,
1760                         thread_dispatcher, (void *) NULL);
1761         if (ret != 0) {
1762                 PERROR("pthread_create viewer dispatcher");
1763                 goto exit_dispatcher;
1764         }
1765
1766         /* Setup the worker thread */
1767         ret = pthread_create(&live_worker_thread, NULL,
1768                         thread_worker, relay_ctx);
1769         if (ret != 0) {
1770                 PERROR("pthread_create viewer worker");
1771                 goto exit_worker;
1772         }
1773
1774         /* Setup the listener thread */
1775         ret = pthread_create(&live_listener_thread, NULL,
1776                         thread_listener, (void *) NULL);
1777         if (ret != 0) {
1778                 PERROR("pthread_create viewer listener");
1779                 goto exit_listener;
1780         }
1781
1782         ret = 0;
1783         goto end;
1784
1785 exit_listener:
1786         ret = pthread_join(live_listener_thread, &status);
1787         if (ret != 0) {
1788                 PERROR("pthread_join live listener");
1789                 goto error;     /* join error, exit without cleanup */
1790         }
1791
1792 exit_worker:
1793         ret = pthread_join(live_worker_thread, &status);
1794         if (ret != 0) {
1795                 PERROR("pthread_join live worker");
1796                 goto error;     /* join error, exit without cleanup */
1797         }
1798
1799 exit_dispatcher:
1800         ret = pthread_join(live_dispatcher_thread, &status);
1801         if (ret != 0) {
1802                 PERROR("pthread_join live dispatcher");
1803                 goto error;     /* join error, exit without cleanup */
1804         }
1805
1806 exit:
1807         cleanup();
1808
1809 end:
1810 error:
1811         return ret;
1812 }