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 == session) {
 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 = session;
 709         }
 710
 711         switch (be32toh(request.seek)) {
 712         case VIEWER_SEEK_BEGINNING:
 713                 /* Default behaviour. */
 714                 break;
 715         case VIEWER_SEEK_LAST:
 716                 /* TODO */
 717                 break;
 718         default:
 719                 ERR("Wrong seek parameter");
 720                 response.status = htobe32(VIEWER_ATTACH_SEEK_ERR);
 721                 send_streams = 0;
 722                 goto send_reply;
 723         }
 724
 725         if (send_streams) {
 726                 /* We should only be there if we have a session to attach to. */
 727                 assert(session);
 728
 729                 /*
 730                  * Fill the viewer_streams_ht to count the number of streams
 731                  * ready to be sent and avoid concurrency issues on the
 732                  * relay_streams_ht and don't rely on a total session stream count.
 733                  */
 734                 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
 735                         struct relay_viewer_stream *vstream;
 736
 737                         node = lttng_ht_iter_get_node_ulong(&iter);
 738                         if (!node) {
 739                                 continue;
 740                         }
 741                         stream = caa_container_of(node, struct relay_stream, stream_n);
 742                         if (stream->session != cmd->session) {
 743                                 continue;
 744                         }
 745
 746                         /*
 747                          * Don't send streams with no ctf_trace, they are not ready to be
 748                          * read.
 749                          */
 750                         if (!stream->ctf_trace) {
 751                                 continue;
 752                         }
 753
 754                         vstream = live_find_viewer_stream_by_id(stream->stream_handle);
 755                         if (!vstream) {
 756                                 ret = init_viewer_stream(stream);
 757                                 if (ret < 0) {
 758                                         goto end_unlock;
 759                                 }
 760                         }
 761                         nb_streams++;
 762                 }
 763                 response.streams_count = htobe32(nb_streams);
 764         }
 765
 766 send_reply:
 767         ret = cmd->sock->ops->sendmsg(cmd->sock, &response, sizeof(response), 0);
 768         if (ret < 0) {
 769                 ERR("Relay sending viewer attach response");
 770                 goto end_unlock;
 771         }
 772
 773         /*
 774          * Unknown or busy session, just return gracefully, the viewer knows what
 775          * is happening.
 776          */
 777         if (!send_streams) {
 778                 ret = 0;
 779                 goto end_unlock;
 780         }
 781
 782         /* We should only be there if we have a session to attach to. */
 783         assert(session);
 784         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
 785                 node64 = lttng_ht_iter_get_node_u64(&iter);
 786                 if (!node64) {
 787                         continue;
 788                 }
 789                 viewer_stream = caa_container_of(node64, struct relay_viewer_stream,
 790                                 stream_n);
 791                 if (viewer_stream->session_id != cmd->session->id) {
 792                         continue;
 793                 }
 794
 795                 send_stream.id = htobe64(viewer_stream->stream_handle);
 796                 send_stream.ctf_trace_id = htobe64(viewer_stream->ctf_trace->id);
 797                 send_stream.metadata_flag = htobe32(viewer_stream->metadata_flag);
 798                 strncpy(send_stream.path_name, viewer_stream->path_name,
 799                                 sizeof(send_stream.path_name));
 800                 strncpy(send_stream.channel_name, viewer_stream->channel_name,
 801                                 sizeof(send_stream.channel_name));
 802
 803                 ret = cmd->sock->ops->sendmsg(cmd->sock, &send_stream,
 804                                 sizeof(send_stream), 0);
 805                 if (ret < 0) {
 806                         ERR("Relay sending stream %" PRIu64, viewer_stream->stream_handle);
 807                         goto end_unlock;
 808                 }
 809                 DBG("Sent stream %" PRIu64 " to viewer", viewer_stream->stream_handle);
 810         }
 811         ret = 0;
 812
 813 end_unlock:
 814         rcu_read_unlock();
 815 end_no_session:
 816 error:
 817         return ret;
 818 }
 819
 820 /*
 821  * Open index file using a given viewer stream.
 822  *
 823  * Return 0 on success or else a negative value.
 824  */
 825 static int open_index(struct relay_viewer_stream *stream)
 826 {
 827         int ret;
 828         char fullpath[PATH_MAX];
 829         struct lttng_packet_index_file_hdr hdr;
 830
 831         if (stream->tracefile_size > 0) {
 832                 /* For now we don't support on-disk ring buffer. */
 833                 ret = -1;
 834                 goto end;
 835         } else {
 836                 ret = snprintf(fullpath, sizeof(fullpath), "%s/" DEFAULT_INDEX_DIR
 837                                 "/%s" DEFAULT_INDEX_FILE_SUFFIX,
 838                                 stream->path_name, stream->channel_name);
 839                 if (ret < 0) {
 840                         PERROR("snprintf index path");
 841                         goto error;
 842                 }
 843         }
 844
 845         DBG("Opening index file %s in read only", fullpath);
 846         ret = open(fullpath, O_RDONLY);
 847         if (ret < 0) {
 848                 if (errno == ENOENT) {
 849                         ret = ENOENT;
 850                         goto error;
 851                 } else {
 852                         PERROR("opening index in read-only");
 853                 }
 854                 goto error;
 855         }
 856         stream->index_read_fd = ret;
 857         DBG("Opening index file %s in read only, (fd: %d)", fullpath, ret);
 858
 859         do {
 860                 ret = read(stream->index_read_fd, &hdr, sizeof(hdr));
 861         } while (ret < 0 && errno == EINTR);
 862         if (ret < 0) {
 863                 PERROR("Reading index header");
 864                 goto error;
 865         }
 866         if (strncmp(hdr.magic, INDEX_MAGIC, sizeof(hdr.magic)) != 0) {
 867                 ERR("Invalid header magic");
 868                 ret = -1;
 869                 goto error;
 870         }
 871         if (be32toh(hdr.index_major) != INDEX_MAJOR ||
 872                         be32toh(hdr.index_minor) != INDEX_MINOR) {
 873                 ERR("Invalid header version");
 874                 ret = -1;
 875                 goto error;
 876         }
 877         ret = 0;
 878
 879 error:
 880 end:
 881         return ret;
 882 }
 883
 884 /*
 885  * Get viewer stream from stream id.
 886  *
 887  * RCU read side lock MUST be acquired.
 888  */
 889 struct relay_viewer_stream *live_find_viewer_stream_by_id(uint64_t stream_id)
 890 {
 891         struct lttng_ht_node_u64 *node;
 892         struct lttng_ht_iter iter;
 893         struct relay_viewer_stream *stream = NULL;
 894
 895         lttng_ht_lookup(viewer_streams_ht, &stream_id, &iter);
 896         node = lttng_ht_iter_get_node_u64(&iter);
 897         if (node == NULL) {
 898                 DBG("Relay viewer stream %" PRIu64 " not found", stream_id);
 899                 goto end;
 900         }
 901         stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
 902
 903 end:
 904         return stream;
 905 }
 906
 907 /*
 908  * Send the next index for a stream.
 909  *
 910  * Return 0 on success or else a negative value.
 911  */
 912 static
 913 int viewer_get_next_index(struct relay_command *cmd,
 914                 struct lttng_ht *sessions_ht)
 915 {
 916         int ret;
 917         struct lttng_viewer_get_next_index request_index;
 918         struct lttng_viewer_index viewer_index;
 919         struct lttng_packet_index packet_index;
 920         struct relay_viewer_stream *vstream;
 921         struct relay_stream *rstream;
 922
 923         assert(cmd);
 924         assert(sessions_ht);
 925
 926         DBG("Viewer get next index");
 927
 928         if (cmd->version_check_done == 0) {
 929                 ERR("Trying to request index before version check");
 930                 ret = -1;
 931                 goto end_no_session;
 932         }
 933
 934         ret = cmd->sock->ops->recvmsg(cmd->sock, &request_index,
 935                         sizeof(request_index), 0);
 936         if (ret < 0 || ret != sizeof(request_index)) {
 937                 ret = -1;
 938                 ERR("Relay didn't receive the whole packet");
 939                 goto end;
 940         }
 941
 942         rcu_read_lock();
 943         vstream = live_find_viewer_stream_by_id(be64toh(request_index.stream_id));
 944         if (!vstream) {
 945                 ret = -1;
 946                 goto end_unlock;
 947         }
 948
 949         memset(&viewer_index, 0, sizeof(viewer_index));
 950
 951         /*
 952          * The viewer should not ask for index on metadata stream.
 953          */
 954         if (vstream->metadata_flag) {
 955                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
 956                 goto send_reply;
 957         }
 958
 959         /* First time, we open the index file */
 960         if (vstream->index_read_fd < 0) {
 961                 ret = open_index(vstream);
 962                 if (ret == ENOENT) {
 963                         /*
 964                          * The index is created only when the first data packet arrives, it
 965                          * might not be ready at the beginning of the session
 966                          */
 967                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
 968                         goto send_reply;
 969                 } else if (ret < 0) {
 970                         viewer_index.status = htobe32(VIEWER_INDEX_ERR);
 971                         goto send_reply;
 972                 }
 973         }
 974
 975         rstream = relay_stream_find_by_id(vstream->stream_handle);
 976         if (rstream) {
 977                 if (rstream->beacon_ts_end != -1ULL &&
 978                                 vstream->last_sent_index == rstream->total_index_received) {
 979                         viewer_index.status = htobe32(VIEWER_INDEX_INACTIVE);
 980                         viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
 981                         goto send_reply;
 982                 }
 983
 984                 if (rstream->total_index_received <= vstream->last_sent_index) {
 985                         /* No new index to send, retry later. */
 986                         viewer_index.status = htobe32(VIEWER_INDEX_RETRY);
 987                         goto send_reply;
 988                 }
 989         } else if (!rstream &&
 990                         vstream->total_index_received == vstream->last_sent_index) {
 991                 /* Last index sent and stream closed */
 992                 viewer_index.status = htobe32(VIEWER_INDEX_HUP);
 993                 goto send_reply;
 994         }
 995
 996         if (!vstream->ctf_trace->metadata_received ||
 997                         vstream->ctf_trace->metadata_received >
 998                         vstream->ctf_trace->metadata_sent) {
 999                 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1000         }
1001
1002         do {
1003                 ret = read(vstream->index_read_fd, &packet_index,
1004                                 sizeof(packet_index));
1005         } while (ret < 0 && errno == EINTR);
1006         if (ret < sizeof(packet_index)) {
1007                 PERROR("Relay reading index file");
1008                 viewer_index.status = htobe32(VIEWER_INDEX_ERR);
1009         } else {
1010                 viewer_index.status = htobe32(VIEWER_INDEX_OK);
1011                 vstream->last_sent_index++;
1012         }
1013
1014         /*
1015          * Indexes are stored in big endian, no need to switch before sending.
1016          */
1017         viewer_index.offset = packet_index.offset;
1018         viewer_index.packet_size = packet_index.packet_size;
1019         viewer_index.content_size = packet_index.content_size;
1020         viewer_index.timestamp_begin = packet_index.timestamp_begin;
1021         viewer_index.timestamp_end = packet_index.timestamp_end;
1022         viewer_index.events_discarded = packet_index.events_discarded;
1023         viewer_index.stream_id = packet_index.stream_id;
1024
1025 send_reply:
1026         viewer_index.flags = htobe32(viewer_index.flags);
1027         ret = cmd->sock->ops->sendmsg(cmd->sock, &viewer_index,
1028                         sizeof(viewer_index), 0);
1029         if (ret < 0) {
1030                 ERR("Relay index to viewer");
1031                 goto end_unlock;
1032         }
1033
1034         DBG("Index %" PRIu64 "for stream %" PRIu64 "sent",
1035                         vstream->last_sent_index, vstream->stream_handle);
1036
1037 end_unlock:
1038         rcu_read_unlock();
1039
1040 end_no_session:
1041 end:
1042         return ret;
1043 }
1044
1045 /*
1046  * Send the next index for a stream
1047  *
1048  * Return 0 on success or else a negative value.
1049  */
1050 static
1051 int viewer_get_packet(struct relay_command *cmd)
1052 {
1053         int ret, send_data = 0;
1054         char *data = NULL;
1055         uint32_t len = 0;
1056         ssize_t read_len;
1057         struct lttng_viewer_get_packet get_packet_info;
1058         struct lttng_viewer_trace_packet reply;
1059         struct relay_viewer_stream *stream;
1060
1061         assert(cmd);
1062
1063         DBG2("Relay get data packet");
1064
1065         if (cmd->version_check_done == 0) {
1066                 ERR("Trying to get packet before version check");
1067                 ret = -1;
1068                 goto end;
1069         }
1070
1071         ret = cmd->sock->ops->recvmsg(cmd->sock, &get_packet_info,
1072                         sizeof(get_packet_info), 0);
1073         if (ret < 0 || ret != sizeof(get_packet_info)) {
1074                 ret = -1;
1075                 ERR("Relay didn't receive the whole packet");
1076                 goto end;
1077         }
1078
1079         rcu_read_lock();
1080         stream = live_find_viewer_stream_by_id(be64toh(get_packet_info.stream_id));
1081         if (!stream) {
1082                 goto error;
1083         }
1084         assert(stream->ctf_trace);
1085
1086         /*
1087          * First time we read this stream, we need open the tracefile, we should
1088          * only arrive here if an index has already been sent to the viewer, so the
1089          * tracefile must exist, if it does not it is a fatal error.
1090          */
1091         if (stream->read_fd < 0) {
1092                 char fullpath[PATH_MAX];
1093
1094                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1095                                 stream->channel_name);
1096                 if (ret < 0) {
1097                         goto error;
1098                 }
1099                 ret = open(fullpath, O_RDONLY);
1100                 if (ret < 0) {
1101                         PERROR("Relay opening trace file");
1102                         goto error;
1103                 }
1104                 stream->read_fd = ret;
1105         }
1106
1107         memset(&reply, 0, sizeof(reply));
1108
1109         if (!stream->ctf_trace->metadata_received ||
1110                         stream->ctf_trace->metadata_received >
1111                         stream->ctf_trace->metadata_sent) {
1112                 reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1113                 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1114
1115                 goto send_reply;
1116         }
1117
1118         len = be32toh(get_packet_info.len);
1119         data = zmalloc(len);
1120         if (!data) {
1121                 PERROR("relay data zmalloc");
1122                 goto error;
1123         }
1124
1125         ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1126         if (ret < 0) {
1127                 PERROR("lseek");
1128                 goto error;
1129         }
1130         read_len = read(stream->read_fd, data, len);
1131         if (read_len < (ssize_t) len) {
1132                 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1133                                 stream->read_fd, be64toh(get_packet_info.offset));
1134                 goto error;
1135         }
1136         reply.status = htobe32(VIEWER_GET_PACKET_OK);
1137         reply.len = htobe32(len);
1138         send_data = 1;
1139         goto send_reply;
1140
1141 error:
1142         reply.status = htobe32(VIEWER_GET_PACKET_ERR);
1143
1144 send_reply:
1145         reply.flags = htobe32(reply.flags);
1146         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1147         if (ret < 0) {
1148                 ERR("Relay data header to viewer");
1149                 goto end_unlock;
1150         }
1151
1152         if (send_data) {
1153                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1154                 if (ret < 0) {
1155                         ERR("Relay send data to viewer");
1156                         goto end_unlock;
1157                 }
1158         }
1159
1160         DBG("Sent %u bytes for stream %" PRIu64, len,
1161                         be64toh(get_packet_info.stream_id));
1162
1163 end_unlock:
1164         free(data);
1165         rcu_read_unlock();
1166
1167 end:
1168         return ret;
1169 }
1170
1171 /*
1172  * Send the session's metadata
1173  *
1174  * Return 0 on success else a negative value.
1175  */
1176 static
1177 int viewer_get_metadata(struct relay_command *cmd)
1178 {
1179         int ret = 0;
1180         ssize_t read_len;
1181         uint64_t len = 0;
1182         char *data = NULL;
1183         struct lttng_viewer_get_metadata request;
1184         struct lttng_viewer_metadata_packet reply;
1185         struct relay_viewer_stream *stream;
1186
1187         assert(cmd);
1188
1189         DBG("Relay get metadata");
1190
1191         if (cmd->version_check_done == 0) {
1192                 ERR("Trying to get metadata before version check");
1193                 ret = -1;
1194                 goto end;
1195         }
1196
1197         ret = cmd->sock->ops->recvmsg(cmd->sock, &request,
1198                         sizeof(request), 0);
1199         if (ret < 0 || ret != sizeof(request)) {
1200                 ret = -1;
1201                 ERR("Relay didn't receive the whole packet");
1202                 goto end;
1203         }
1204
1205         rcu_read_lock();
1206         stream = live_find_viewer_stream_by_id(be64toh(request.stream_id));
1207         if (!stream || !stream->metadata_flag) {
1208                 ERR("Invalid metadata stream");
1209                 goto error;
1210         }
1211         assert(stream->ctf_trace);
1212         assert(stream->ctf_trace->metadata_sent <=
1213                         stream->ctf_trace->metadata_received);
1214
1215         len = stream->ctf_trace->metadata_received -
1216                 stream->ctf_trace->metadata_sent;
1217         if (len == 0) {
1218                 reply.status = htobe32(VIEWER_NO_NEW_METADATA);
1219                 goto send_reply;
1220         }
1221
1222         /* first time, we open the metadata file */
1223         if (stream->read_fd < 0) {
1224                 char fullpath[PATH_MAX];
1225
1226                 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1227                                 stream->channel_name);
1228                 if (ret < 0) {
1229                         goto error;
1230                 }
1231                 ret = open(fullpath, O_RDONLY);
1232                 if (ret < 0) {
1233                         PERROR("Relay opening metadata file");
1234                         goto error;
1235                 }
1236                 stream->read_fd = ret;
1237         }
1238
1239         reply.len = htobe64(len);
1240         data = zmalloc(len);
1241         if (!data) {
1242                 PERROR("viewer metadata zmalloc");
1243                 goto error;
1244         }
1245
1246         read_len = read(stream->read_fd, data, len);
1247         if (read_len < (ssize_t) len) {
1248                 PERROR("Relay reading metadata file");
1249                 goto error;
1250         }
1251         stream->ctf_trace->metadata_sent += read_len;
1252         reply.status = htobe32(VIEWER_METADATA_OK);
1253         goto send_reply;
1254
1255 error:
1256         reply.status = htobe32(VIEWER_METADATA_ERR);
1257
1258 send_reply:
1259         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1260         if (ret < 0) {
1261                 ERR("Relay data header to viewer");
1262                 goto end_unlock;
1263         }
1264
1265         if (len > 0) {
1266                 ret = cmd->sock->ops->sendmsg(cmd->sock, data, len, 0);
1267                 if (ret < 0) {
1268                         ERR("Relay send data to viewer");
1269                         goto end_unlock;
1270                 }
1271         }
1272
1273         DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1274                         be64toh(request.stream_id));
1275
1276         DBG("Metadata sent");
1277
1278 end_unlock:
1279         free(data);
1280         rcu_read_unlock();
1281 end:
1282         return ret;
1283 }
1284
1285 /*
1286  * live_relay_unknown_command: send -1 if received unknown command
1287  */
1288 static
1289 void live_relay_unknown_command(struct relay_command *cmd)
1290 {
1291         struct lttcomm_relayd_generic_reply reply;
1292         int ret;
1293
1294         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1295         ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1296                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1297         if (ret < 0) {
1298                 ERR("Relay sending unknown command");
1299         }
1300 }
1301
1302 /*
1303  * Process the commands received on the control socket
1304  */
1305 static
1306 int process_control(struct lttng_viewer_cmd *recv_hdr,
1307                 struct relay_command *cmd, struct lttng_ht *sessions_ht)
1308 {
1309         int ret = 0;
1310
1311         switch (be32toh(recv_hdr->cmd)) {
1312         case VIEWER_CONNECT:
1313                 ret = viewer_connect(cmd);
1314                 break;
1315         case VIEWER_LIST_SESSIONS:
1316                 ret = viewer_list_sessions(cmd, sessions_ht);
1317                 break;
1318         case VIEWER_ATTACH_SESSION:
1319                 ret = viewer_attach_session(cmd, sessions_ht);
1320                 break;
1321         case VIEWER_GET_NEXT_INDEX:
1322                 ret = viewer_get_next_index(cmd, sessions_ht);
1323                 break;
1324         case VIEWER_GET_PACKET:
1325                 ret = viewer_get_packet(cmd);
1326                 break;
1327         case VIEWER_GET_METADATA:
1328                 ret = viewer_get_metadata(cmd);
1329                 break;
1330         default:
1331                 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1332                 live_relay_unknown_command(cmd);
1333                 ret = -1;
1334                 goto end;
1335         }
1336
1337 end:
1338         return ret;
1339 }
1340
1341 static
1342 void cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
1343 {
1344         int ret;
1345
1346         assert(events);
1347
1348         lttng_poll_del(events, pollfd);
1349
1350         ret = close(pollfd);
1351         if (ret < 0) {
1352                 ERR("Closing pollfd %d", pollfd);
1353         }
1354 }
1355
1356 /*
1357  * Create and add connection to the given hash table.
1358  *
1359  * Return poll add value or else -1 on error.
1360  */
1361 static
1362 int add_connection(int fd, struct lttng_poll_event *events,
1363                 struct lttng_ht *relay_connections_ht)
1364 {
1365         int ret;
1366         struct relay_command *relay_connection;
1367
1368         assert(events);
1369         assert(relay_connections_ht);
1370
1371         relay_connection = zmalloc(sizeof(struct relay_command));
1372         if (relay_connection == NULL) {
1373                 PERROR("Relay command zmalloc");
1374                 goto error;
1375         }
1376
1377         do {
1378                 ret = read(fd, relay_connection, sizeof(*relay_connection));
1379         } while (ret < 0 && errno == EINTR);
1380         if (ret < 0 || ret < sizeof(*relay_connection)) {
1381                 PERROR("read relay cmd pipe");
1382                 goto error_read;
1383         }
1384
1385         lttng_ht_node_init_ulong(&relay_connection->sock_n,
1386                         (unsigned long) relay_connection->sock->fd);
1387         rcu_read_lock();
1388         lttng_ht_add_unique_ulong(relay_connections_ht,
1389                         &relay_connection->sock_n);
1390         rcu_read_unlock();
1391
1392         return lttng_poll_add(events, relay_connection->sock->fd,
1393                         LPOLLIN | LPOLLRDHUP);
1394
1395 error_read:
1396         free(relay_connection);
1397 error:
1398         return -1;
1399 }
1400
1401 static
1402 void deferred_free_connection(struct rcu_head *head)
1403 {
1404         struct relay_command *relay_connection =
1405                 caa_container_of(head, struct relay_command, rcu_node);
1406
1407         if (relay_connection->session &&
1408                         relay_connection->session->viewer_attached > 0) {
1409                 relay_connection->session->viewer_attached--;
1410         }
1411         lttcomm_destroy_sock(relay_connection->sock);
1412         free(relay_connection);
1413 }
1414
1415 static
1416 void deferred_free_viewer_stream(struct rcu_head *head)
1417 {
1418         struct relay_viewer_stream *stream =
1419                 caa_container_of(head, struct relay_viewer_stream, rcu_node);
1420
1421         if (stream->ctf_trace) {
1422                 uatomic_dec(&stream->ctf_trace->refcount);
1423                 assert(uatomic_read(&stream->ctf_trace->refcount) >= 0);
1424                 if (uatomic_read(&stream->ctf_trace->refcount) == 0) {
1425                         DBG("Freeing ctf_trace %" PRIu64, stream->ctf_trace->id);
1426                         free(stream->ctf_trace);
1427                 }
1428         }
1429
1430         free(stream->path_name);
1431         free(stream->channel_name);
1432         free(stream);
1433 }
1434
1435 static
1436 void viewer_del_streams(struct relay_session *session)
1437 {
1438         int ret;
1439         struct relay_viewer_stream *stream;
1440         struct lttng_ht_node_u64 *node;
1441         struct lttng_ht_iter iter;
1442
1443         assert(session);
1444
1445         rcu_read_lock();
1446         cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, node, node) {
1447                 node = lttng_ht_iter_get_node_u64(&iter);
1448                 if (!node) {
1449                         continue;
1450                 }
1451
1452                 stream = caa_container_of(node, struct relay_viewer_stream, stream_n);
1453                 if (stream->session_id != session->id) {
1454                         continue;
1455                 }
1456
1457                 if (stream->read_fd > 0) {
1458                         ret = close(stream->read_fd);
1459                         if (ret < 0) {
1460                                 PERROR("close read_fd");
1461                         }
1462                 }
1463                 if (stream->index_read_fd > 0) {
1464                         ret = close(stream->index_read_fd);
1465                         if (ret < 0) {
1466                                 PERROR("close index_read_fd");
1467                         }
1468                 }
1469                 if (stream->metadata_flag && stream->ctf_trace) {
1470                         stream->ctf_trace->metadata_sent = 0;
1471                 }
1472                 ret = lttng_ht_del(viewer_streams_ht, &iter);
1473                 assert(!ret);
1474                 call_rcu(&stream->rcu_node, deferred_free_viewer_stream);
1475         }
1476         rcu_read_unlock();
1477 }
1478
1479 /*
1480  * Delete and free a connection.
1481  *
1482  * RCU read side lock MUST be acquired.
1483  */
1484 static
1485 void del_connection(struct lttng_ht *relay_connections_ht,
1486                 struct lttng_ht_iter *iter, struct relay_command *relay_connection)
1487 {
1488         int ret;
1489
1490         assert(relay_connections_ht);
1491         assert(iter);
1492         assert(relay_connection);
1493
1494         ret = lttng_ht_del(relay_connections_ht, iter);
1495         assert(!ret);
1496
1497         if (relay_connection->session) {
1498                 viewer_del_streams(relay_connection->session);
1499         }
1500
1501         call_rcu(&relay_connection->rcu_node, deferred_free_connection);
1502 }
1503
1504 /*
1505  * This thread does the actual work
1506  */
1507 static
1508 void *thread_worker(void *data)
1509 {
1510         int ret, err = -1;
1511         uint32_t nb_fd;
1512         struct relay_command *relay_connection;
1513         struct lttng_poll_event events;
1514         struct lttng_ht *relay_connections_ht;
1515         struct lttng_ht_node_ulong *node;
1516         struct lttng_ht_iter iter;
1517         struct lttng_viewer_cmd recv_hdr;
1518         struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1519         struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1520
1521         DBG("[thread] Live viewer relay worker started");
1522
1523         rcu_register_thread();
1524
1525         /* table of connections indexed on socket */
1526         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1527         if (!relay_connections_ht) {
1528                 goto relay_connections_ht_error;
1529         }
1530
1531         ret = create_thread_poll_set(&events, 2);
1532         if (ret < 0) {
1533                 goto error_poll_create;
1534         }
1535
1536         ret = lttng_poll_add(&events, live_relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1537         if (ret < 0) {
1538                 goto error;
1539         }
1540
1541 restart:
1542         while (1) {
1543                 int i;
1544
1545                 /* Infinite blocking call, waiting for transmission */
1546                 DBG3("Relayd live viewer worker thread polling...");
1547                 ret = lttng_poll_wait(&events, -1);
1548                 if (ret < 0) {
1549                         /*
1550                          * Restart interrupted system call.
1551                          */
1552                         if (errno == EINTR) {
1553                                 goto restart;
1554                         }
1555                         goto error;
1556                 }
1557
1558                 nb_fd = ret;
1559
1560                 /*
1561                  * Process control. The control connection is prioritised so we don't
1562                  * starve it with high throughput tracing data on the data
1563                  * connection.
1564                  */
1565                 for (i = 0; i < nb_fd; i++) {
1566                         /* Fetch once the poll data */
1567                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1568                         int pollfd = LTTNG_POLL_GETFD(&events, i);
1569
1570                         /* Thread quit pipe has been closed. Killing thread. */
1571                         ret = check_thread_quit_pipe(pollfd, revents);
1572                         if (ret) {
1573                                 err = 0;
1574                                 goto exit;
1575                         }
1576
1577                         /* Inspect the relay cmd pipe for new connection */
1578                         if (pollfd == live_relay_cmd_pipe[0]) {
1579                                 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1580                                         ERR("Relay live pipe error");
1581                                         goto error;
1582                                 } else if (revents & LPOLLIN) {
1583                                         DBG("Relay live viewer command received");
1584                                         ret = add_connection(live_relay_cmd_pipe[0],
1585                                                         &events, relay_connections_ht);
1586                                         if (ret < 0) {
1587                                                 goto error;
1588                                         }
1589                                 }
1590                         } else if (revents) {
1591                                 rcu_read_lock();
1592                                 lttng_ht_lookup(relay_connections_ht,
1593                                                 (void *)((unsigned long) pollfd), &iter);
1594                                 node = lttng_ht_iter_get_node_ulong(&iter);
1595                                 if (node == NULL) {
1596                                         DBG2("Relay viewer sock %d not found", pollfd);
1597                                         rcu_read_unlock();
1598                                         goto error;
1599                                 }
1600                                 relay_connection = caa_container_of(node, struct relay_command,
1601                                                 sock_n);
1602
1603                                 if (revents & (LPOLLERR)) {
1604                                         ERR("VIEWER POLL ERROR");
1605                                         cleanup_poll_connection(&events, pollfd);
1606                                         del_connection(relay_connections_ht, &iter,
1607                                                         relay_connection);
1608                                 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1609                                         DBG("Viewer socket %d hung up", pollfd);
1610                                         cleanup_poll_connection(&events, pollfd);
1611                                         del_connection(relay_connections_ht, &iter,
1612                                                         relay_connection);
1613                                 } else if (revents & LPOLLIN) {
1614                                         ret = relay_connection->sock->ops->recvmsg(
1615                                                         relay_connection->sock, &recv_hdr,
1616                                                         sizeof(struct lttng_viewer_cmd),
1617                                                         0);
1618                                         /* connection closed */
1619                                         if (ret <= 0) {
1620                                                 cleanup_poll_connection(&events, pollfd);
1621                                                 del_connection( relay_connections_ht, &iter,
1622                                                                 relay_connection);
1623                                                 DBG("Viewer control connection closed with %d",
1624                                                                 pollfd);
1625                                         } else {
1626                                                 if (relay_connection->session) {
1627                                                         DBG2("Relay viewer worker receiving data for "
1628                                                                         "session: %" PRIu64,
1629                                                                         relay_connection->session->id);
1630                                                 }
1631                                                 ret = process_control(&recv_hdr, relay_connection,
1632                                                                 sessions_ht);
1633                                                 if (ret < 0) {
1634                                                         /* Clear the session on error. */
1635                                                         cleanup_poll_connection(&events, pollfd);
1636                                                         del_connection(relay_connections_ht, &iter,
1637                                                                         relay_connection);
1638                                                         DBG("Viewer connection closed with %d", pollfd);
1639                                                 }
1640                                         }
1641                                 }
1642                                 rcu_read_unlock();
1643                         }
1644                 }
1645         }
1646
1647 exit:
1648 error:
1649         lttng_poll_clean(&events);
1650
1651         /* empty the hash table and free the memory */
1652         rcu_read_lock();
1653         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
1654                 node = lttng_ht_iter_get_node_ulong(&iter);
1655                 if (!node) {
1656                         continue;
1657                 }
1658
1659                 relay_connection = caa_container_of(node, struct relay_command,
1660                                 sock_n);
1661                 del_connection(relay_connections_ht, &iter, relay_connection);
1662         }
1663         rcu_read_unlock();
1664 error_poll_create:
1665         lttng_ht_destroy(relay_connections_ht);
1666 relay_connections_ht_error:
1667         /* Close relay cmd pipes */
1668         utils_close_pipe(live_relay_cmd_pipe);
1669         if (err) {
1670                 DBG("Viewer worker thread exited with error");
1671         }
1672         DBG("Viewer worker thread cleanup complete");
1673         stop_threads();
1674         rcu_unregister_thread();
1675         return NULL;
1676 }
1677
1678 /*
1679  * Create the relay command pipe to wake thread_manage_apps.
1680  * Closed in cleanup().
1681  */
1682 static int create_relay_cmd_pipe(void)
1683 {
1684         int ret;
1685
1686         ret = utils_create_pipe_cloexec(live_relay_cmd_pipe);
1687
1688         return ret;
1689 }
1690
1691 void live_stop_threads()
1692 {
1693         int ret;
1694         void *status;
1695
1696         stop_threads();
1697
1698         ret = pthread_join(live_listener_thread, &status);
1699         if (ret != 0) {
1700                 PERROR("pthread_join live listener");
1701                 goto error;     /* join error, exit without cleanup */
1702         }
1703
1704         ret = pthread_join(live_worker_thread, &status);
1705         if (ret != 0) {
1706                 PERROR("pthread_join live worker");
1707                 goto error;     /* join error, exit without cleanup */
1708         }
1709
1710         ret = pthread_join(live_dispatcher_thread, &status);
1711         if (ret != 0) {
1712                 PERROR("pthread_join live dispatcher");
1713                 goto error;     /* join error, exit without cleanup */
1714         }
1715
1716         cleanup();
1717
1718 error:
1719         return;
1720 }
1721
1722 /*
1723  * main
1724  */
1725 int live_start_threads(struct lttng_uri *uri,
1726                 struct relay_local_data *relay_ctx)
1727 {
1728         int ret = 0;
1729         void *status;
1730         int is_root;
1731
1732         assert(uri);
1733         live_uri = uri;
1734
1735         /* Create thread quit pipe */
1736         if ((ret = init_thread_quit_pipe()) < 0) {
1737                 goto error;
1738         }
1739
1740         /* Check if daemon is UID = 0 */
1741         is_root = !getuid();
1742
1743         if (!is_root) {
1744                 if (live_uri->port < 1024) {
1745                         ERR("Need to be root to use ports < 1024");
1746                         ret = -1;
1747                         goto exit;
1748                 }
1749         }
1750
1751         /* Setup the thread apps communication pipe. */
1752         if ((ret = create_relay_cmd_pipe()) < 0) {
1753                 goto exit;
1754         }
1755
1756         /* Init relay command queue. */
1757         cds_wfq_init(&viewer_cmd_queue.queue);
1758
1759         /* Set up max poll set size */
1760         lttng_poll_set_max_size();
1761
1762         /* Setup the dispatcher thread */
1763         ret = pthread_create(&live_dispatcher_thread, NULL,
1764                         thread_dispatcher, (void *) NULL);
1765         if (ret != 0) {
1766                 PERROR("pthread_create viewer dispatcher");
1767                 goto exit_dispatcher;
1768         }
1769
1770         /* Setup the worker thread */
1771         ret = pthread_create(&live_worker_thread, NULL,
1772                         thread_worker, relay_ctx);
1773         if (ret != 0) {
1774                 PERROR("pthread_create viewer worker");
1775                 goto exit_worker;
1776         }
1777
1778         /* Setup the listener thread */
1779         ret = pthread_create(&live_listener_thread, NULL,
1780                         thread_listener, (void *) NULL);
1781         if (ret != 0) {
1782                 PERROR("pthread_create viewer listener");
1783                 goto exit_listener;
1784         }
1785
1786         ret = 0;
1787         goto end;
1788
1789 exit_listener:
1790         ret = pthread_join(live_listener_thread, &status);
1791         if (ret != 0) {
1792                 PERROR("pthread_join live listener");
1793                 goto error;     /* join error, exit without cleanup */
1794         }
1795
1796 exit_worker:
1797         ret = pthread_join(live_worker_thread, &status);
1798         if (ret != 0) {
1799                 PERROR("pthread_join live worker");
1800                 goto error;     /* join error, exit without cleanup */
1801         }
1802
1803 exit_dispatcher:
1804         ret = pthread_join(live_dispatcher_thread, &status);
1805         if (ret != 0) {
1806                 PERROR("pthread_join live dispatcher");
1807                 goto error;     /* join error, exit without cleanup */
1808         }
1809
1810 exit:
1811         cleanup();
1812
1813 end:
1814 error:
1815         return ret;
1816 }