src/bin/lttng-relayd/main.c

   1 /*
   2  * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
   3  * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
   4  * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
   5  * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   6  *
   7  * SPDX-License-Identifier: GPL-2.0-only
   8  *
   9  */
  10
  11 #define _LGPL_SOURCE
  12 #include <getopt.h>
  13 #include <grp.h>
  14 #include <limits.h>
  15 #include <pthread.h>
  16 #include <signal.h>
  17 #include <stdio.h>
  18 #include <stdlib.h>
  19 #include <string.h>
  20 #include <sys/mman.h>
  21 #include <sys/mount.h>
  22 #include <sys/resource.h>
  23 #include <sys/socket.h>
  24 #include <sys/stat.h>
  25 #include <sys/types.h>
  26 #include <sys/wait.h>
  27 #include <sys/resource.h>
  28 #include <inttypes.h>
  29 #include <urcu/futex.h>
  30 #include <urcu/uatomic.h>
  31 #include <urcu/rculist.h>
  32 #include <unistd.h>
  33 #include <fcntl.h>
  34 #include <strings.h>
  35 #include <ctype.h>
  36
  37 #include <lttng/lttng.h>
  38 #include <common/common.h>
  39 #include <common/compat/poll.h>
  40 #include <common/compat/socket.h>
  41 #include <common/compat/endian.h>
  42 #include <common/compat/getenv.h>
  43 #include <common/defaults.h>
  44 #include <common/daemonize.h>
  45 #include <common/futex.h>
  46 #include <common/sessiond-comm/sessiond-comm.h>
  47 #include <common/sessiond-comm/inet.h>
  48 #include <common/sessiond-comm/relayd.h>
  49 #include <common/uri.h>
  50 #include <common/utils.h>
  51 #include <common/align.h>
  52 #include <common/config/session-config.h>
  53 #include <common/dynamic-buffer.h>
  54 #include <common/buffer-view.h>
  55 #include <common/string-utils/format.h>
  56 #include <common/fd-tracker/fd-tracker.h>
  57 #include <common/fd-tracker/utils.h>
  58
  59 #include "backward-compatibility-group-by.h"
  60 #include "cmd.h"
  61 #include "connection.h"
  62 #include "ctf-trace.h"
  63 #include "health-relayd.h"
  64 #include "index.h"
  65 #include "live.h"
  66 #include "lttng-relayd.h"
  67 #include "session.h"
  68 #include "sessiond-trace-chunks.h"
  69 #include "stream.h"
  70 #include "tcp_keep_alive.h"
  71 #include "testpoint.h"
  72 #include "tracefile-array.h"
  73 #include "utils.h"
  74 #include "version.h"
  75 #include "viewer-stream.h"
  76
  77 static const char *help_msg =
  78 #ifdef LTTNG_EMBED_HELP
  79 #include <lttng-relayd.8.h>
  80 #else
  81 NULL
  82 #endif
  83 ;
  84
  85 enum relay_connection_status {
  86         RELAY_CONNECTION_STATUS_OK,
  87         /* An error occurred while processing an event on the connection. */
  88         RELAY_CONNECTION_STATUS_ERROR,
  89         /* Connection closed/shutdown cleanly. */
  90         RELAY_CONNECTION_STATUS_CLOSED,
  91 };
  92
  93 /* command line options */
  94 char *opt_output_path, *opt_working_directory;
  95 static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
  96 enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
  97
  98 /*
  99  * We need to wait for listener and live listener threads, as well as
 100  * health check thread, before being ready to signal readiness.
 101  */
 102 #define NR_LTTNG_RELAY_READY    3
 103 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
 104
 105 /* Size of receive buffer. */
 106 #define RECV_DATA_BUFFER_SIZE           65536
 107
 108 static int recv_child_signal;   /* Set to 1 when a SIGUSR1 signal is received. */
 109 static pid_t child_ppid;        /* Internal parent PID use with daemonize. */
 110
 111 static struct lttng_uri *control_uri;
 112 static struct lttng_uri *data_uri;
 113 static struct lttng_uri *live_uri;
 114
 115 const char *progname;
 116
 117 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
 118 static int tracing_group_name_override;
 119
 120 const char * const config_section_name = "relayd";
 121
 122 /*
 123  * Quit pipe for all threads. This permits a single cancellation point
 124  * for all threads when receiving an event on the pipe.
 125  */
 126 int thread_quit_pipe[2] = { -1, -1 };
 127
 128 /*
 129  * This pipe is used to inform the worker thread that a command is queued and
 130  * ready to be processed.
 131  */
 132 static int relay_conn_pipe[2] = { -1, -1 };
 133
 134 /* Shared between threads */
 135 static int dispatch_thread_exit;
 136
 137 static pthread_t listener_thread;
 138 static pthread_t dispatcher_thread;
 139 static pthread_t worker_thread;
 140 static pthread_t health_thread;
 141
 142 /*
 143  * last_relay_stream_id_lock protects last_relay_stream_id increment
 144  * atomicity on 32-bit architectures.
 145  */
 146 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
 147 static uint64_t last_relay_stream_id;
 148
 149 /*
 150  * Relay command queue.
 151  *
 152  * The relay_thread_listener and relay_thread_dispatcher communicate with this
 153  * queue.
 154  */
 155 static struct relay_conn_queue relay_conn_queue;
 156
 157 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
 158 static unsigned int lttng_opt_fd_pool_size = -1;
 159
 160 /* Global relay stream hash table. */
 161 struct lttng_ht *relay_streams_ht;
 162
 163 /* Global relay viewer stream hash table. */
 164 struct lttng_ht *viewer_streams_ht;
 165
 166 /* Global relay sessions hash table. */
 167 struct lttng_ht *sessions_ht;
 168
 169 /* Relayd health monitoring */
 170 struct health_app *health_relayd;
 171
 172 struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
 173
 174 /* Global fd tracker. */
 175 struct fd_tracker *the_fd_tracker;
 176
 177 static struct option long_options[] = {
 178         { "control-port", 1, 0, 'C', },
 179         { "data-port", 1, 0, 'D', },
 180         { "live-port", 1, 0, 'L', },
 181         { "daemonize", 0, 0, 'd', },
 182         { "background", 0, 0, 'b', },
 183         { "group", 1, 0, 'g', },
 184         { "fd-pool-size", 1, 0, '\0', },
 185         { "help", 0, 0, 'h', },
 186         { "output", 1, 0, 'o', },
 187         { "verbose", 0, 0, 'v', },
 188         { "config", 1, 0, 'f' },
 189         { "version", 0, 0, 'V' },
 190         { "working-directory", 1, 0, 'w', },
 191         { "group-output-by-session", 0, 0, 's', },
 192         { "group-output-by-host", 0, 0, 'p', },
 193         { "disallow-clear", 0, 0, 'x' },
 194         { NULL, 0, 0, 0, },
 195 };
 196
 197 static const char *config_ignore_options[] = { "help", "config", "version" };
 198
 199 static void print_version(void) {
 200         fprintf(stdout, "%s\n", VERSION);
 201 }
 202
 203 static void relayd_config_log(void)
 204 {
 205         DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
 206                         GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
 207                         EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
 208         if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
 209                 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
 210         }
 211         if (EXTRA_VERSION_PATCHES[0] != '\0') {
 212                 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
 213         }
 214 }
 215
 216 /*
 217  * Take an option from the getopt output and set it in the right variable to be
 218  * used later.
 219  *
 220  * Return 0 on success else a negative value.
 221  */
 222 static int set_option(int opt, const char *arg, const char *optname)
 223 {
 224         int ret;
 225
 226         switch (opt) {
 227         case 0:
 228                 if (!strcmp(optname, "fd-pool-size")) {
 229                         unsigned long v;
 230
 231                         errno = 0;
 232                         v = strtoul(arg, NULL, 0);
 233                         if (errno != 0 || !isdigit(arg[0])) {
 234                                 ERR("Wrong value in --fd-pool-size parameter: %s", arg);
 235                                 ret = -1;
 236                                 goto end;
 237                         }
 238                         if (v >= UINT_MAX) {
 239                                 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg);
 240                                 ret = -1;
 241                                 goto end;
 242                         }
 243                         lttng_opt_fd_pool_size = (unsigned int) v;
 244                 } else {
 245                         fprintf(stderr, "unknown option %s", optname);
 246                         if (arg) {
 247                                 fprintf(stderr, " with arg %s\n", arg);
 248                         }
 249                 }
 250                 break;
 251         case 'C':
 252                 if (lttng_is_setuid_setgid()) {
 253                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 254                                 "-C, --control-port");
 255                 } else {
 256                         ret = uri_parse(arg, &control_uri);
 257                         if (ret < 0) {
 258                                 ERR("Invalid control URI specified");
 259                                 goto end;
 260                         }
 261                         if (control_uri->port == 0) {
 262                                 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
 263                         }
 264                 }
 265                 break;
 266         case 'D':
 267                 if (lttng_is_setuid_setgid()) {
 268                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 269                                 "-D, -data-port");
 270                 } else {
 271                         ret = uri_parse(arg, &data_uri);
 272                         if (ret < 0) {
 273                                 ERR("Invalid data URI specified");
 274                                 goto end;
 275                         }
 276                         if (data_uri->port == 0) {
 277                                 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
 278                         }
 279                 }
 280                 break;
 281         case 'L':
 282                 if (lttng_is_setuid_setgid()) {
 283                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 284                                 "-L, -live-port");
 285                 } else {
 286                         ret = uri_parse(arg, &live_uri);
 287                         if (ret < 0) {
 288                                 ERR("Invalid live URI specified");
 289                                 goto end;
 290                         }
 291                         if (live_uri->port == 0) {
 292                                 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
 293                         }
 294                 }
 295                 break;
 296         case 'd':
 297                 opt_daemon = 1;
 298                 break;
 299         case 'b':
 300                 opt_background = 1;
 301                 break;
 302         case 'g':
 303                 if (lttng_is_setuid_setgid()) {
 304                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 305                                 "-g, --group");
 306                 } else {
 307                         tracing_group_name = strdup(arg);
 308                         if (tracing_group_name == NULL) {
 309                                 ret = -errno;
 310                                 PERROR("strdup");
 311                                 goto end;
 312                         }
 313                         tracing_group_name_override = 1;
 314                 }
 315                 break;
 316         case 'h':
 317                 ret = utils_show_help(8, "lttng-relayd", help_msg);
 318                 if (ret) {
 319                         ERR("Cannot show --help for `lttng-relayd`");
 320                         perror("exec");
 321                 }
 322                 exit(EXIT_FAILURE);
 323         case 'V':
 324                 opt_print_version = 1;
 325                 break;
 326         case 'o':
 327                 if (lttng_is_setuid_setgid()) {
 328                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 329                                 "-o, --output");
 330                 } else {
 331                         ret = asprintf(&opt_output_path, "%s", arg);
 332                         if (ret < 0) {
 333                                 ret = -errno;
 334                                 PERROR("asprintf opt_output_path");
 335                                 goto end;
 336                         }
 337                 }
 338                 break;
 339         case 'w':
 340                 if (lttng_is_setuid_setgid()) {
 341                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 342                                 "-w, --working-directory");
 343                 } else {
 344                         ret = asprintf(&opt_working_directory, "%s", arg);
 345                         if (ret < 0) {
 346                                 ret = -errno;
 347                                 PERROR("asprintf opt_working_directory");
 348                                 goto end;
 349                         }
 350                 }
 351                 break;
 352
 353         case 'v':
 354                 /* Verbose level can increase using multiple -v */
 355                 if (arg) {
 356                         lttng_opt_verbose = config_parse_value(arg);
 357                 } else {
 358                         /* Only 3 level of verbosity (-vvv). */
 359                         if (lttng_opt_verbose < 3) {
 360                                 lttng_opt_verbose += 1;
 361                         }
 362                 }
 363                 break;
 364         case 's':
 365                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 366                         ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
 367                         exit(EXIT_FAILURE);
 368                 }
 369                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
 370                 break;
 371         case 'p':
 372                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 373                         ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
 374                         exit(EXIT_FAILURE);
 375                 }
 376                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 377                 break;
 378         case 'x':
 379                 /* Disallow clear */
 380                 opt_allow_clear = 0;
 381                 break;
 382         default:
 383                 /* Unknown option or other error.
 384                  * Error is printed by getopt, just return */
 385                 ret = -1;
 386                 goto end;
 387         }
 388
 389         /* All good. */
 390         ret = 0;
 391
 392 end:
 393         return ret;
 394 }
 395
 396 /*
 397  * config_entry_handler_cb used to handle options read from a config file.
 398  * See config_entry_handler_cb comment in common/config/session-config.h for the
 399  * return value conventions.
 400  */
 401 static int config_entry_handler(const struct config_entry *entry, void *unused)
 402 {
 403         int ret = 0, i;
 404
 405         if (!entry || !entry->name || !entry->value) {
 406                 ret = -EINVAL;
 407                 goto end;
 408         }
 409
 410         /* Check if the option is to be ignored */
 411         for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
 412                 if (!strcmp(entry->name, config_ignore_options[i])) {
 413                         goto end;
 414                 }
 415         }
 416
 417         for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
 418                 /* Ignore if entry name is not fully matched. */
 419                 if (strcmp(entry->name, long_options[i].name)) {
 420                         continue;
 421                 }
 422
 423                 /*
 424                  * If the option takes no argument on the command line,
 425                  * we have to check if the value is "true". We support
 426                  * non-zero numeric values, true, on and yes.
 427                  */
 428                 if (!long_options[i].has_arg) {
 429                         ret = config_parse_value(entry->value);
 430                         if (ret <= 0) {
 431                                 if (ret) {
 432                                         WARN("Invalid configuration value \"%s\" for option %s",
 433                                                         entry->value, entry->name);
 434                                 }
 435                                 /* False, skip boolean config option. */
 436                                 goto end;
 437                         }
 438                 }
 439
 440                 ret = set_option(long_options[i].val, entry->value, entry->name);
 441                 goto end;
 442         }
 443
 444         WARN("Unrecognized option \"%s\" in daemon configuration file.",
 445                         entry->name);
 446
 447 end:
 448         return ret;
 449 }
 450
 451 static int parse_env_options(void)
 452 {
 453         int ret = 0;
 454         char *value = NULL;
 455
 456         value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
 457         if (value) {
 458                 opt_working_directory = strdup(value);
 459                 if (!opt_working_directory) {
 460                         ERR("Failed to allocate working directory string (\"%s\")",
 461                                         value);
 462                         ret = -1;
 463                 }
 464         }
 465         return ret;
 466 }
 467
 468 static int set_fd_pool_size(void)
 469 {
 470         int ret = 0;
 471         struct rlimit rlimit;
 472
 473         ret = getrlimit(RLIMIT_NOFILE, &rlimit);
 474         if (ret) {
 475                 PERROR("Failed to get file descriptor limit");
 476                 ret = -1;
 477                 goto end;
 478         }
 479
 480         DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
 481                         (uint64_t) rlimit.rlim_cur,
 482                         (uint64_t) rlimit.rlim_max);
 483         if (lttng_opt_fd_pool_size == -1) {
 484                 /* Use default value (soft limit - reserve). */
 485                 if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 486                         ERR("The process' file number limit is too low (%" PRIu64 "). The process' file number limit must be set to at least %i.",
 487                                         (uint64_t) rlimit.rlim_cur, DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 488                         ret = -1;
 489                         goto end;
 490                 }
 491                 lttng_opt_fd_pool_size = rlimit.rlim_cur -
 492                                 DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 493                 goto end;
 494         }
 495
 496         if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 497                 ERR("File descriptor pool size must be set to at least %d",
 498                                 DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 499                 ret = -1;
 500                 goto end;
 501         }
 502
 503         if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
 504                 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64 ").",
 505                                 lttng_opt_fd_pool_size, (uint64_t) rlimit.rlim_cur);
 506                 ret = -1;
 507                 goto end;
 508         }
 509
 510         DBG("File descriptor pool size argument (%u) adjusted to %u to accomodate transient fd uses",
 511                         lttng_opt_fd_pool_size,
 512                         lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
 513         lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 514 end:
 515         return ret;
 516 }
 517
 518 static int set_options(int argc, char **argv)
 519 {
 520         int c, ret = 0, option_index = 0, retval = 0;
 521         int orig_optopt = optopt, orig_optind = optind;
 522         char *default_address, *optstring;
 523         const char *config_path = NULL;
 524
 525         optstring = utils_generate_optstring(long_options,
 526                         sizeof(long_options) / sizeof(struct option));
 527         if (!optstring) {
 528                 retval = -ENOMEM;
 529                 goto exit;
 530         }
 531
 532         /* Check for the --config option */
 533
 534         while ((c = getopt_long(argc, argv, optstring, long_options,
 535                                         &option_index)) != -1) {
 536                 if (c == '?') {
 537                         retval = -EINVAL;
 538                         goto exit;
 539                 } else if (c != 'f') {
 540                         continue;
 541                 }
 542
 543                 if (lttng_is_setuid_setgid()) {
 544                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 545                                 "-f, --config");
 546                 } else {
 547                         config_path = utils_expand_path(optarg);
 548                         if (!config_path) {
 549                                 ERR("Failed to resolve path: %s", optarg);
 550                         }
 551                 }
 552         }
 553
 554         ret = config_get_section_entries(config_path, config_section_name,
 555                         config_entry_handler, NULL);
 556         if (ret) {
 557                 if (ret > 0) {
 558                         ERR("Invalid configuration option at line %i", ret);
 559                 }
 560                 retval = -1;
 561                 goto exit;
 562         }
 563
 564         /* Reset getopt's global state */
 565         optopt = orig_optopt;
 566         optind = orig_optind;
 567         while (1) {
 568                 c = getopt_long(argc, argv, optstring, long_options, &option_index);
 569                 if (c == -1) {
 570                         break;
 571                 }
 572
 573                 ret = set_option(c, optarg, long_options[option_index].name);
 574                 if (ret < 0) {
 575                         retval = -1;
 576                         goto exit;
 577                 }
 578         }
 579
 580         /* assign default values */
 581         if (control_uri == NULL) {
 582                 ret = asprintf(&default_address,
 583                         "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
 584                         DEFAULT_NETWORK_CONTROL_PORT);
 585                 if (ret < 0) {
 586                         PERROR("asprintf default data address");
 587                         retval = -1;
 588                         goto exit;
 589                 }
 590
 591                 ret = uri_parse(default_address, &control_uri);
 592                 free(default_address);
 593                 if (ret < 0) {
 594                         ERR("Invalid control URI specified");
 595                         retval = -1;
 596                         goto exit;
 597                 }
 598         }
 599         if (data_uri == NULL) {
 600                 ret = asprintf(&default_address,
 601                         "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
 602                         DEFAULT_NETWORK_DATA_PORT);
 603                 if (ret < 0) {
 604                         PERROR("asprintf default data address");
 605                         retval = -1;
 606                         goto exit;
 607                 }
 608
 609                 ret = uri_parse(default_address, &data_uri);
 610                 free(default_address);
 611                 if (ret < 0) {
 612                         ERR("Invalid data URI specified");
 613                         retval = -1;
 614                         goto exit;
 615                 }
 616         }
 617         if (live_uri == NULL) {
 618                 ret = asprintf(&default_address,
 619                         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
 620                         DEFAULT_NETWORK_VIEWER_PORT);
 621                 if (ret < 0) {
 622                         PERROR("asprintf default viewer control address");
 623                         retval = -1;
 624                         goto exit;
 625                 }
 626
 627                 ret = uri_parse(default_address, &live_uri);
 628                 free(default_address);
 629                 if (ret < 0) {
 630                         ERR("Invalid viewer control URI specified");
 631                         retval = -1;
 632                         goto exit;
 633                 }
 634         }
 635         ret = set_fd_pool_size();
 636         if (ret) {
 637                 retval = -1;
 638                 goto exit;
 639         }
 640
 641         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 642                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 643         }
 644         if (opt_allow_clear) {
 645                 /* Check if env variable exists. */
 646                 const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 647                 if (value) {
 648                         ret = config_parse_value(value);
 649                         if (ret < 0) {
 650                                 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 651                                 retval = -1;
 652                                 goto exit;
 653                         }
 654                         opt_allow_clear = !ret;
 655                 }
 656         }
 657
 658 exit:
 659         free(optstring);
 660         return retval;
 661 }
 662
 663 static void print_global_objects(void)
 664 {
 665         print_viewer_streams();
 666         print_relay_streams();
 667         print_sessions();
 668 }
 669
 670 static int noop_close(void *data, int *fds)
 671 {
 672         return 0;
 673 }
 674
 675 static void untrack_stdio(void)
 676 {
 677         int fds[] = { fileno(stdout), fileno(stderr) };
 678
 679         /*
 680          * noop_close is used since we don't really want to close
 681          * the stdio output fds; we merely want to stop tracking them.
 682          */
 683         (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker,
 684                         fds, 2, noop_close, NULL);
 685 }
 686
 687 /*
 688  * Cleanup the daemon
 689  */
 690 static void relayd_cleanup(void)
 691 {
 692         print_global_objects();
 693
 694         DBG("Cleaning up");
 695
 696         if (viewer_streams_ht)
 697                 lttng_ht_destroy(viewer_streams_ht);
 698         if (relay_streams_ht)
 699                 lttng_ht_destroy(relay_streams_ht);
 700         if (sessions_ht)
 701                 lttng_ht_destroy(sessions_ht);
 702
 703         free(opt_output_path);
 704         free(opt_working_directory);
 705
 706         if (health_relayd) {
 707                 health_app_destroy(health_relayd);
 708         }
 709         /* Close thread quit pipes */
 710         if (health_quit_pipe[0] != -1) {
 711                 (void) fd_tracker_util_pipe_close(
 712                                 the_fd_tracker, health_quit_pipe);
 713         }
 714         if (thread_quit_pipe[0] != -1) {
 715                 (void) fd_tracker_util_pipe_close(
 716                                 the_fd_tracker, thread_quit_pipe);
 717         }
 718         if (sessiond_trace_chunk_registry) {
 719                 sessiond_trace_chunk_registry_destroy(
 720                                 sessiond_trace_chunk_registry);
 721         }
 722         if (the_fd_tracker) {
 723                 untrack_stdio();
 724                 /*
 725                  * fd_tracker_destroy() will log the contents of the fd-tracker
 726                  * if a leak is detected.
 727                  */
 728                 fd_tracker_destroy(the_fd_tracker);
 729         }
 730
 731         uri_free(control_uri);
 732         uri_free(data_uri);
 733         /* Live URI is freed in the live thread. */
 734
 735         if (tracing_group_name_override) {
 736                 free((void *) tracing_group_name);
 737         }
 738 }
 739
 740 /*
 741  * Write to writable pipe used to notify a thread.
 742  */
 743 static int notify_thread_pipe(int wpipe)
 744 {
 745         ssize_t ret;
 746
 747         ret = lttng_write(wpipe, "!", 1);
 748         if (ret < 1) {
 749                 PERROR("write poll pipe");
 750                 goto end;
 751         }
 752         ret = 0;
 753 end:
 754         return ret;
 755 }
 756
 757 static int notify_health_quit_pipe(int *pipe)
 758 {
 759         ssize_t ret;
 760
 761         ret = lttng_write(pipe[1], "4", 1);
 762         if (ret < 1) {
 763                 PERROR("write relay health quit");
 764                 goto end;
 765         }
 766         ret = 0;
 767 end:
 768         return ret;
 769 }
 770
 771 /*
 772  * Stop all relayd and relayd-live threads.
 773  */
 774 int lttng_relay_stop_threads(void)
 775 {
 776         int retval = 0;
 777
 778         /* Stopping all threads */
 779         DBG("Terminating all threads");
 780         if (notify_thread_pipe(thread_quit_pipe[1])) {
 781                 ERR("write error on thread quit pipe");
 782                 retval = -1;
 783         }
 784
 785         if (notify_health_quit_pipe(health_quit_pipe)) {
 786                 ERR("write error on health quit pipe");
 787         }
 788
 789         /* Dispatch thread */
 790         CMM_STORE_SHARED(dispatch_thread_exit, 1);
 791         futex_nto1_wake(&relay_conn_queue.futex);
 792
 793         if (relayd_live_stop()) {
 794                 ERR("Error stopping live threads");
 795                 retval = -1;
 796         }
 797         return retval;
 798 }
 799
 800 /*
 801  * Signal handler for the daemon
 802  *
 803  * Simply stop all worker threads, leaving main() return gracefully after
 804  * joining all threads and calling cleanup().
 805  */
 806 static void sighandler(int sig)
 807 {
 808         switch (sig) {
 809         case SIGINT:
 810                 DBG("SIGINT caught");
 811                 if (lttng_relay_stop_threads()) {
 812                         ERR("Error stopping threads");
 813                 }
 814                 break;
 815         case SIGTERM:
 816                 DBG("SIGTERM caught");
 817                 if (lttng_relay_stop_threads()) {
 818                         ERR("Error stopping threads");
 819                 }
 820                 break;
 821         case SIGUSR1:
 822                 CMM_STORE_SHARED(recv_child_signal, 1);
 823                 break;
 824         default:
 825                 break;
 826         }
 827 }
 828
 829 /*
 830  * Setup signal handler for :
 831  *              SIGINT, SIGTERM, SIGPIPE
 832  */
 833 static int set_signal_handler(void)
 834 {
 835         int ret = 0;
 836         struct sigaction sa;
 837         sigset_t sigset;
 838
 839         if ((ret = sigemptyset(&sigset)) < 0) {
 840                 PERROR("sigemptyset");
 841                 return ret;
 842         }
 843
 844         sa.sa_mask = sigset;
 845         sa.sa_flags = 0;
 846
 847         sa.sa_handler = sighandler;
 848         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
 849                 PERROR("sigaction");
 850                 return ret;
 851         }
 852
 853         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
 854                 PERROR("sigaction");
 855                 return ret;
 856         }
 857
 858         if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
 859                 PERROR("sigaction");
 860                 return ret;
 861         }
 862
 863         sa.sa_handler = SIG_IGN;
 864         if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
 865                 PERROR("sigaction");
 866                 return ret;
 867         }
 868
 869         DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
 870
 871         return ret;
 872 }
 873
 874 void lttng_relay_notify_ready(void)
 875 {
 876         /* Notify the parent of the fork() process that we are ready. */
 877         if (opt_daemon || opt_background) {
 878                 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
 879                         kill(child_ppid, SIGUSR1);
 880                 }
 881         }
 882 }
 883
 884 /*
 885  * Init thread quit pipe.
 886  *
 887  * Return -1 on error or 0 if all pipes are created.
 888  */
 889 static int init_thread_quit_pipe(void)
 890 {
 891         return fd_tracker_util_pipe_open_cloexec(
 892                         the_fd_tracker, "Quit pipe", thread_quit_pipe);
 893 }
 894
 895 /*
 896  * Init health quit pipe.
 897  *
 898  * Return -1 on error or 0 if all pipes are created.
 899  */
 900 static int init_health_quit_pipe(void)
 901 {
 902         return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
 903                         "Health quit pipe", health_quit_pipe);
 904 }
 905
 906 /*
 907  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 908  */
 909 static int create_named_thread_poll_set(struct lttng_poll_event *events,
 910                 int size, const char *name)
 911 {
 912         int ret;
 913
 914         if (events == NULL || size == 0) {
 915                 ret = -1;
 916                 goto error;
 917         }
 918
 919         ret = fd_tracker_util_poll_create(the_fd_tracker,
 920                         name, events, 1, LTTNG_CLOEXEC);
 921         if (ret) {
 922                 PERROR("Failed to create \"%s\" poll file descriptor", name);
 923                 goto error;
 924         }
 925
 926         /* Add quit pipe */
 927         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 928         if (ret < 0) {
 929                 goto error;
 930         }
 931
 932         return 0;
 933
 934 error:
 935         return ret;
 936 }
 937
 938 /*
 939  * Check if the thread quit pipe was triggered.
 940  *
 941  * Return 1 if it was triggered else 0;
 942  */
 943 static int check_thread_quit_pipe(int fd, uint32_t events)
 944 {
 945         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 946                 return 1;
 947         }
 948
 949         return 0;
 950 }
 951
 952 static int create_sock(void *data, int *out_fd)
 953 {
 954         int ret;
 955         struct lttcomm_sock *sock = data;
 956
 957         ret = lttcomm_create_sock(sock);
 958         if (ret < 0) {
 959                 goto end;
 960         }
 961
 962         *out_fd = sock->fd;
 963 end:
 964         return ret;
 965 }
 966
 967 static int close_sock(void *data, int *in_fd)
 968 {
 969         struct lttcomm_sock *sock = data;
 970
 971         return sock->ops->close(sock);
 972 }
 973
 974 static int accept_sock(void *data, int *out_fd)
 975 {
 976         int ret = 0;
 977         /* Socks is an array of in_sock, out_sock. */
 978         struct lttcomm_sock **socks = data;
 979         struct lttcomm_sock *in_sock = socks[0];
 980
 981         socks[1] = in_sock->ops->accept(in_sock);
 982         if (!socks[1]) {
 983                 ret = -1;
 984                 goto end;
 985         }
 986         *out_fd = socks[1]->fd;
 987 end:
 988         return ret;
 989 }
 990
 991 /*
 992  * Create and init socket from uri.
 993  */
 994 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri,
 995                 const char *name)
 996 {
 997         int ret, sock_fd;
 998         struct lttcomm_sock *sock = NULL;
 999         char uri_str[PATH_MAX];
1000         char *formated_name = NULL;
1001
1002         sock = lttcomm_alloc_sock_from_uri(uri);
1003         if (sock == NULL) {
1004                 ERR("Allocating socket");
1005                 goto error;
1006         }
1007
1008         /*
1009          * Don't fail to create the socket if the name can't be built as it is
1010          * only used for debugging purposes.
1011          */
1012         ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
1013         uri_str[sizeof(uri_str) - 1] = '\0';
1014         if (ret >= 0) {
1015                 ret = asprintf(&formated_name, "%s socket @ %s", name,
1016                                 uri_str);
1017                 if (ret < 0) {
1018                         formated_name = NULL;
1019                 }
1020         }
1021
1022         ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
1023                         (const char **) (formated_name ? &formated_name : NULL),
1024                         1, create_sock, sock);
1025         free(formated_name);
1026         DBG("Listening on %s socket %d", name, sock->fd);
1027
1028         ret = sock->ops->bind(sock);
1029         if (ret < 0) {
1030                 PERROR("Failed to bind socket");
1031                 goto error;
1032         }
1033
1034         ret = sock->ops->listen(sock, -1);
1035         if (ret < 0) {
1036                 goto error;
1037
1038         }
1039
1040         return sock;
1041
1042 error:
1043         if (sock) {
1044                 lttcomm_destroy_sock(sock);
1045         }
1046         return NULL;
1047 }
1048
1049 static
1050 struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1051                 const char *name)
1052 {
1053         int out_fd, ret;
1054         struct lttcomm_sock *socks[2] = { listening_sock, NULL };
1055         struct lttcomm_sock *new_sock = NULL;
1056
1057         ret = fd_tracker_open_unsuspendable_fd(
1058                         the_fd_tracker, &out_fd,
1059                         (const char **) &name,
1060                         1, accept_sock, &socks);
1061         if (ret) {
1062                 goto end;
1063         }
1064         new_sock = socks[1];
1065         DBG("%s accepted, socket %d", name, new_sock->fd);
1066 end:
1067         return new_sock;
1068 }
1069
1070 /*
1071  * This thread manages the listening for new connections on the network
1072  */
1073 static void *relay_thread_listener(void *data)
1074 {
1075         int i, ret, pollfd, err = -1;
1076         uint32_t revents, nb_fd;
1077         struct lttng_poll_event events;
1078         struct lttcomm_sock *control_sock, *data_sock;
1079
1080         DBG("[thread] Relay listener started");
1081
1082         rcu_register_thread();
1083         health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1084
1085         health_code_update();
1086
1087         control_sock = relay_socket_create(control_uri, "Control listener");
1088         if (!control_sock) {
1089                 goto error_sock_control;
1090         }
1091
1092         data_sock = relay_socket_create(data_uri, "Data listener");
1093         if (!data_sock) {
1094                 goto error_sock_relay;
1095         }
1096
1097         /*
1098          * Pass 3 as size here for the thread quit pipe, control and
1099          * data socket.
1100          */
1101         ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1102         if (ret < 0) {
1103                 goto error_create_poll;
1104         }
1105
1106         /* Add the control socket */
1107         ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1108         if (ret < 0) {
1109                 goto error_poll_add;
1110         }
1111
1112         /* Add the data socket */
1113         ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1114         if (ret < 0) {
1115                 goto error_poll_add;
1116         }
1117
1118         lttng_relay_notify_ready();
1119
1120         if (testpoint(relayd_thread_listener)) {
1121                 goto error_testpoint;
1122         }
1123
1124         while (1) {
1125                 health_code_update();
1126
1127                 DBG("Listener accepting connections");
1128
1129 restart:
1130                 health_poll_entry();
1131                 ret = lttng_poll_wait(&events, -1);
1132                 health_poll_exit();
1133                 if (ret < 0) {
1134                         /*
1135                          * Restart interrupted system call.
1136                          */
1137                         if (errno == EINTR) {
1138                                 goto restart;
1139                         }
1140                         goto error;
1141                 }
1142
1143                 nb_fd = ret;
1144
1145                 DBG("Relay new connection received");
1146                 for (i = 0; i < nb_fd; i++) {
1147                         health_code_update();
1148
1149                         /* Fetch once the poll data */
1150                         revents = LTTNG_POLL_GETEV(&events, i);
1151                         pollfd = LTTNG_POLL_GETFD(&events, i);
1152
1153                         /* Thread quit pipe has been closed. Killing thread. */
1154                         ret = check_thread_quit_pipe(pollfd, revents);
1155                         if (ret) {
1156                                 err = 0;
1157                                 goto exit;
1158                         }
1159
1160                         if (revents & LPOLLIN) {
1161                                 /*
1162                                  * A new connection is requested, therefore a
1163                                  * sessiond/consumerd connection is allocated in
1164                                  * this thread, enqueued to a global queue and
1165                                  * dequeued (and freed) in the worker thread.
1166                                  */
1167                                 int val = 1;
1168                                 struct relay_connection *new_conn;
1169                                 struct lttcomm_sock *newsock = NULL;
1170                                 enum connection_type type;
1171
1172                                 if (pollfd == data_sock->fd) {
1173                                         type = RELAY_DATA;
1174                                         newsock = accept_relayd_sock(data_sock,
1175                                                         "Data socket to relayd");
1176                                 } else {
1177                                         assert(pollfd == control_sock->fd);
1178                                         type = RELAY_CONTROL;
1179                                         newsock = accept_relayd_sock(control_sock,
1180                                                         "Control socket to relayd");
1181                                 }
1182                                 if (!newsock) {
1183                                         PERROR("accepting sock");
1184                                         goto error;
1185                                 }
1186
1187                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
1188                                                 sizeof(val));
1189                                 if (ret < 0) {
1190                                         PERROR("setsockopt inet");
1191                                         lttcomm_destroy_sock(newsock);
1192                                         goto error;
1193                                 }
1194
1195                                 ret = socket_apply_keep_alive_config(newsock->fd);
1196                                 if (ret < 0) {
1197                                         ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1198                                                         newsock->fd);
1199                                         lttcomm_destroy_sock(newsock);
1200                                         goto error;
1201                                 }
1202
1203                                 new_conn = connection_create(newsock, type);
1204                                 if (!new_conn) {
1205                                         lttcomm_destroy_sock(newsock);
1206                                         goto error;
1207                                 }
1208
1209                                 /* Enqueue request for the dispatcher thread. */
1210                                 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
1211                                                  &new_conn->qnode);
1212
1213                                 /*
1214                                  * Wake the dispatch queue futex.
1215                                  * Implicit memory barrier with the
1216                                  * exchange in cds_wfcq_enqueue.
1217                                  */
1218                                 futex_nto1_wake(&relay_conn_queue.futex);
1219                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1220                                 ERR("socket poll error");
1221                                 goto error;
1222                         } else {
1223                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1224                                 goto error;
1225                         }
1226                 }
1227         }
1228
1229 exit:
1230 error:
1231 error_poll_add:
1232 error_testpoint:
1233         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1234 error_create_poll:
1235         if (data_sock->fd >= 0) {
1236                 int data_sock_fd = data_sock->fd;
1237
1238                 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1239                                 &data_sock_fd, 1, close_sock,
1240                                 data_sock);
1241                 if (ret) {
1242                         PERROR("Failed to close the data listener socket file descriptor");
1243                 }
1244                 data_sock->fd = -1;
1245         }
1246         lttcomm_destroy_sock(data_sock);
1247 error_sock_relay:
1248         if (control_sock->fd >= 0) {
1249                 int control_sock_fd = control_sock->fd;
1250
1251                 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1252                                 &control_sock_fd, 1, close_sock,
1253                                 control_sock);
1254                 if (ret) {
1255                         PERROR("Failed to close the control listener socket file descriptor");
1256                 }
1257                 control_sock->fd = -1;
1258         }
1259         lttcomm_destroy_sock(control_sock);
1260 error_sock_control:
1261         if (err) {
1262                 health_error();
1263                 ERR("Health error occurred in %s", __func__);
1264         }
1265         health_unregister(health_relayd);
1266         rcu_unregister_thread();
1267         DBG("Relay listener thread cleanup complete");
1268         lttng_relay_stop_threads();
1269         return NULL;
1270 }
1271
1272 /*
1273  * This thread manages the dispatching of the requests to worker threads
1274  */
1275 static void *relay_thread_dispatcher(void *data)
1276 {
1277         int err = -1;
1278         ssize_t ret;
1279         struct cds_wfcq_node *node;
1280         struct relay_connection *new_conn = NULL;
1281
1282         DBG("[thread] Relay dispatcher started");
1283
1284         health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1285
1286         if (testpoint(relayd_thread_dispatcher)) {
1287                 goto error_testpoint;
1288         }
1289
1290         health_code_update();
1291
1292         for (;;) {
1293                 health_code_update();
1294
1295                 /* Atomically prepare the queue futex */
1296                 futex_nto1_prepare(&relay_conn_queue.futex);
1297
1298                 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1299                         break;
1300                 }
1301
1302                 do {
1303                         health_code_update();
1304
1305                         /* Dequeue commands */
1306                         node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1307                                                          &relay_conn_queue.tail);
1308                         if (node == NULL) {
1309                                 DBG("Woken up but nothing in the relay command queue");
1310                                 /* Continue thread execution */
1311                                 break;
1312                         }
1313                         new_conn = caa_container_of(node, struct relay_connection, qnode);
1314
1315                         DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1316
1317                         /*
1318                          * Inform worker thread of the new request. This
1319                          * call is blocking so we can be assured that
1320                          * the data will be read at some point in time
1321                          * or wait to the end of the world :)
1322                          */
1323                         ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1324                         if (ret < 0) {
1325                                 PERROR("write connection pipe");
1326                                 connection_put(new_conn);
1327                                 goto error;
1328                         }
1329                 } while (node != NULL);
1330
1331                 /* Futex wait on queue. Blocking call on futex() */
1332                 health_poll_entry();
1333                 futex_nto1_wait(&relay_conn_queue.futex);
1334                 health_poll_exit();
1335         }
1336
1337         /* Normal exit, no error */
1338         err = 0;
1339
1340 error:
1341 error_testpoint:
1342         if (err) {
1343                 health_error();
1344                 ERR("Health error occurred in %s", __func__);
1345         }
1346         health_unregister(health_relayd);
1347         DBG("Dispatch thread dying");
1348         lttng_relay_stop_threads();
1349         return NULL;
1350 }
1351
1352 static bool session_streams_have_index(const struct relay_session *session)
1353 {
1354         return session->minor >= 4 && !session->snapshot;
1355 }
1356
1357 /*
1358  * Handle the RELAYD_CREATE_SESSION command.
1359  *
1360  * On success, send back the session id or else return a negative value.
1361  */
1362 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr,
1363                 struct relay_connection *conn,
1364                 const struct lttng_buffer_view *payload)
1365 {
1366         int ret = 0;
1367         ssize_t send_ret;
1368         struct relay_session *session = NULL;
1369         struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1370         char session_name[LTTNG_NAME_MAX] = {};
1371         char hostname[LTTNG_HOST_NAME_MAX] = {};
1372         uint32_t live_timer = 0;
1373         bool snapshot = false;
1374         bool session_name_contains_creation_timestamp = false;
1375         /* Left nil for peers < 2.11. */
1376         char base_path[LTTNG_PATH_MAX] = {};
1377         lttng_uuid sessiond_uuid = {};
1378         LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1379         LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1380         LTTNG_OPTIONAL(time_t) creation_time = {};
1381         struct lttng_dynamic_buffer reply_payload;
1382
1383         lttng_dynamic_buffer_init(&reply_payload);
1384
1385         if (conn->minor < 4) {
1386                 /* From 2.1 to 2.3 */
1387                 ret = 0;
1388         } else if (conn->minor >= 4 && conn->minor < 11) {
1389                 /* From 2.4 to 2.10 */
1390                 ret = cmd_create_session_2_4(payload, session_name,
1391                         hostname, &live_timer, &snapshot);
1392         } else {
1393                 bool has_current_chunk;
1394                 uint64_t current_chunk_id_value;
1395                 time_t creation_time_value;
1396                 uint64_t id_sessiond_value;
1397
1398                 /* From 2.11 to ... */
1399                 ret = cmd_create_session_2_11(payload, session_name, hostname,
1400                                 base_path, &live_timer, &snapshot, &id_sessiond_value,
1401                                 sessiond_uuid, &has_current_chunk,
1402                                 &current_chunk_id_value, &creation_time_value,
1403                                 &session_name_contains_creation_timestamp);
1404                 if (lttng_uuid_is_nil(sessiond_uuid)) {
1405                         /* The nil UUID is reserved for pre-2.11 clients. */
1406                         ERR("Illegal nil UUID announced by peer in create session command");
1407                         ret = -1;
1408                         goto send_reply;
1409                 }
1410                 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1411                 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1412                 if (has_current_chunk) {
1413                         LTTNG_OPTIONAL_SET(&current_chunk_id,
1414                                         current_chunk_id_value);
1415                 }
1416         }
1417
1418         if (ret < 0) {
1419                 goto send_reply;
1420         }
1421
1422         session = session_create(session_name, hostname, base_path, live_timer,
1423                         snapshot, sessiond_uuid,
1424                         id_sessiond.is_set ? &id_sessiond.value : NULL,
1425                         current_chunk_id.is_set ? &current_chunk_id.value : NULL,
1426                         creation_time.is_set ? &creation_time.value : NULL,
1427                         conn->major, conn->minor,
1428                         session_name_contains_creation_timestamp);
1429         if (!session) {
1430                 ret = -1;
1431                 goto send_reply;
1432         }
1433         assert(!conn->session);
1434         conn->session = session;
1435         DBG("Created session %" PRIu64, session->id);
1436
1437         reply.generic.session_id = htobe64(session->id);
1438
1439 send_reply:
1440         if (ret < 0) {
1441                 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1442         } else {
1443                 reply.generic.ret_code = htobe32(LTTNG_OK);
1444         }
1445
1446         if (conn->minor < 11) {
1447                 /* From 2.1 to 2.10 */
1448                 ret = lttng_dynamic_buffer_append(&reply_payload,
1449                                 &reply.generic, sizeof(reply.generic));
1450                 if (ret) {
1451                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1452                         ret = -1;
1453                         goto end;
1454                 }
1455         } else {
1456                 const uint32_t output_path_length =
1457                                 session ? strlen(session->output_path) + 1 : 0;
1458
1459                 reply.output_path_length = htobe32(output_path_length);
1460                 ret = lttng_dynamic_buffer_append(
1461                                 &reply_payload, &reply, sizeof(reply));
1462                 if (ret) {
1463                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1464                         goto end;
1465                 }
1466
1467                 if (output_path_length) {
1468                         ret = lttng_dynamic_buffer_append(&reply_payload,
1469                                         session->output_path,
1470                                         output_path_length);
1471                         if (ret) {
1472                                 ERR("Failed to append \"create session\" command reply path to payload buffer");
1473                                 goto end;
1474                         }
1475                 }
1476         }
1477
1478         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data,
1479                         reply_payload.size, 0);
1480         if (send_ret < (ssize_t) reply_payload.size) {
1481                 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1482                                 reply_payload.size, send_ret);
1483                 ret = -1;
1484         }
1485 end:
1486         if (ret < 0 && session) {
1487                 session_put(session);
1488         }
1489         lttng_dynamic_buffer_reset(&reply_payload);
1490         return ret;
1491 }
1492
1493 /*
1494  * When we have received all the streams and the metadata for a channel,
1495  * we make them visible to the viewer threads.
1496  */
1497 static void publish_connection_local_streams(struct relay_connection *conn)
1498 {
1499         struct relay_stream *stream;
1500         struct relay_session *session = conn->session;
1501
1502         /*
1503          * We publish all streams belonging to a session atomically wrt
1504          * session lock.
1505          */
1506         pthread_mutex_lock(&session->lock);
1507         rcu_read_lock();
1508         cds_list_for_each_entry_rcu(stream, &session->recv_list,
1509                         recv_node) {
1510                 stream_publish(stream);
1511         }
1512         rcu_read_unlock();
1513
1514         /*
1515          * Inform the viewer that there are new streams in the session.
1516          */
1517         if (session->viewer_attached) {
1518                 uatomic_set(&session->new_streams, 1);
1519         }
1520         pthread_mutex_unlock(&session->lock);
1521 }
1522
1523 static int conform_channel_path(char *channel_path)
1524 {
1525         int ret = 0;
1526
1527         if (strstr("../", channel_path)) {
1528                 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1529                                 channel_path);
1530                 ret = -1;
1531                 goto end;
1532         }
1533
1534         if (*channel_path == '/') {
1535                 const size_t len = strlen(channel_path);
1536
1537                 /*
1538                  * Channel paths from peers prior to 2.11 are expressed as an
1539                  * absolute path that is, in reality, relative to the relay
1540                  * daemon's output directory. Remove the leading slash so it
1541                  * is correctly interpreted as a relative path later on.
1542                  *
1543                  * len (and not len - 1) is used to copy the trailing NULL.
1544                  */
1545                 bcopy(channel_path + 1, channel_path, len);
1546         }
1547 end:
1548         return ret;
1549 }
1550
1551 /*
1552  * relay_add_stream: allocate a new stream for a session
1553  */
1554 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1555                 struct relay_connection *conn,
1556                 const struct lttng_buffer_view *payload)
1557 {
1558         int ret;
1559         ssize_t send_ret;
1560         struct relay_session *session = conn->session;
1561         struct relay_stream *stream = NULL;
1562         struct lttcomm_relayd_status_stream reply;
1563         struct ctf_trace *trace = NULL;
1564         uint64_t stream_handle = -1ULL;
1565         char *path_name = NULL, *channel_name = NULL;
1566         uint64_t tracefile_size = 0, tracefile_count = 0;
1567         LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1568
1569         if (!session || !conn->version_check_done) {
1570                 ERR("Trying to add a stream before version check");
1571                 ret = -1;
1572                 goto end_no_session;
1573         }
1574
1575         if (session->minor == 1) {
1576                 /* For 2.1 */
1577                 ret = cmd_recv_stream_2_1(payload, &path_name,
1578                         &channel_name);
1579         } else if (session->minor > 1 && session->minor < 11) {
1580                 /* From 2.2 to 2.10 */
1581                 ret = cmd_recv_stream_2_2(payload, &path_name,
1582                         &channel_name, &tracefile_size, &tracefile_count);
1583         } else {
1584                 /* From 2.11 to ... */
1585                 ret = cmd_recv_stream_2_11(payload, &path_name,
1586                         &channel_name, &tracefile_size, &tracefile_count,
1587                         &stream_chunk_id.value);
1588                 stream_chunk_id.is_set = true;
1589         }
1590
1591         if (ret < 0) {
1592                 goto send_reply;
1593         }
1594
1595         if (conform_channel_path(path_name)) {
1596                 goto send_reply;
1597         }
1598
1599         /*
1600          * Backward compatibility for --group-output-by-session.
1601          * Prior to lttng 2.11, the complete path is passed by the stream.
1602          * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1603          * >=2.11 the chunk is responsible for the output path. When dealing
1604          * with producer < 2.11 the chunk output_path is the root output path
1605          * and the stream carries the complete path (path_name).
1606          * To support --group-output-by-session with older producer (<2.11), we
1607          * need to craft the path based on the stream path.
1608          */
1609         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1610                 if (conn->minor < 4) {
1611                         /*
1612                          * From 2.1 to 2.3, the session_name is not passed on
1613                          * the RELAYD_CREATE_SESSION command. The session name
1614                          * is necessary to detect the presence of a base_path
1615                          * inside the stream path. Without it we cannot perform
1616                          * a valid group-output-by-session transformation.
1617                          */
1618                         WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1619                              " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1620                                         session->id, path_name);
1621                 } else if (conn->minor >= 4 && conn->minor < 11) {
1622                         char *group_by_session_path_name;
1623
1624                         assert(session->session_name[0] != '\0');
1625
1626                         group_by_session_path_name =
1627                                         backward_compat_group_by_session(
1628                                                         path_name,
1629                                                         session->session_name);
1630                         if (!group_by_session_path_name) {
1631                                 ERR("Failed to apply group by session to stream of session %" PRIu64,
1632                                                 session->id);
1633                                 goto send_reply;
1634                         }
1635
1636                         DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1637                                         path_name, group_by_session_path_name);
1638
1639                         free(path_name);
1640                         path_name = group_by_session_path_name;
1641                 }
1642         }
1643
1644         trace = ctf_trace_get_by_path_or_create(session, path_name);
1645         if (!trace) {
1646                 goto send_reply;
1647         }
1648
1649         /* This stream here has one reference on the trace. */
1650         pthread_mutex_lock(&last_relay_stream_id_lock);
1651         stream_handle = ++last_relay_stream_id;
1652         pthread_mutex_unlock(&last_relay_stream_id_lock);
1653
1654         /* We pass ownership of path_name and channel_name. */
1655         stream = stream_create(trace, stream_handle, path_name,
1656                 channel_name, tracefile_size, tracefile_count);
1657         path_name = NULL;
1658         channel_name = NULL;
1659
1660         /*
1661          * Streams are the owners of their trace. Reference to trace is
1662          * kept within stream_create().
1663          */
1664         ctf_trace_put(trace);
1665
1666 send_reply:
1667         memset(&reply, 0, sizeof(reply));
1668         reply.handle = htobe64(stream_handle);
1669         if (!stream) {
1670                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1671         } else {
1672                 reply.ret_code = htobe32(LTTNG_OK);
1673         }
1674
1675         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1676                         sizeof(struct lttcomm_relayd_status_stream), 0);
1677         if (send_ret < (ssize_t) sizeof(reply)) {
1678                 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1679                                 send_ret);
1680                 ret = -1;
1681         }
1682
1683 end_no_session:
1684         free(path_name);
1685         free(channel_name);
1686         return ret;
1687 }
1688
1689 /*
1690  * relay_close_stream: close a specific stream
1691  */
1692 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1693                 struct relay_connection *conn,
1694                 const struct lttng_buffer_view *payload)
1695 {
1696         int ret;
1697         ssize_t send_ret;
1698         struct relay_session *session = conn->session;
1699         struct lttcomm_relayd_close_stream stream_info;
1700         struct lttcomm_relayd_generic_reply reply;
1701         struct relay_stream *stream;
1702
1703         DBG("Close stream received");
1704
1705         if (!session || !conn->version_check_done) {
1706                 ERR("Trying to close a stream before version check");
1707                 ret = -1;
1708                 goto end_no_session;
1709         }
1710
1711         if (payload->size < sizeof(stream_info)) {
1712                 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1713                                 sizeof(stream_info), payload->size);
1714                 ret = -1;
1715                 goto end_no_session;
1716         }
1717         memcpy(&stream_info, payload->data, sizeof(stream_info));
1718         stream_info.stream_id = be64toh(stream_info.stream_id);
1719         stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1720
1721         stream = stream_get_by_id(stream_info.stream_id);
1722         if (!stream) {
1723                 ret = -1;
1724                 goto end;
1725         }
1726
1727         /*
1728          * Set last_net_seq_num before the close flag. Required by data
1729          * pending check.
1730          */
1731         pthread_mutex_lock(&stream->lock);
1732         stream->last_net_seq_num = stream_info.last_net_seq_num;
1733         pthread_mutex_unlock(&stream->lock);
1734
1735         /*
1736          * This is one of the conditions which may trigger a stream close
1737          * with the others being:
1738          *     1) A close command is received for a stream
1739          *     2) The control connection owning the stream is closed
1740          *     3) We have received all of the stream's data _after_ a close
1741          *        request.
1742          */
1743         try_stream_close(stream);
1744         if (stream->is_metadata) {
1745                 struct relay_viewer_stream *vstream;
1746
1747                 vstream = viewer_stream_get_by_id(stream->stream_handle);
1748                 if (vstream) {
1749                         if (stream->no_new_metadata_notified) {
1750                                 /*
1751                                  * Since all the metadata has been sent to the
1752                                  * viewer and that we have a request to close
1753                                  * its stream, we can safely teardown the
1754                                  * corresponding metadata viewer stream.
1755                                  */
1756                                 viewer_stream_put(vstream);
1757                         }
1758                         /* Put local reference. */
1759                         viewer_stream_put(vstream);
1760                 }
1761         }
1762         stream_put(stream);
1763         ret = 0;
1764
1765 end:
1766         memset(&reply, 0, sizeof(reply));
1767         if (ret < 0) {
1768                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1769         } else {
1770                 reply.ret_code = htobe32(LTTNG_OK);
1771         }
1772         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1773                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1774         if (send_ret < (ssize_t) sizeof(reply)) {
1775                 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1776                                 send_ret);
1777                 ret = -1;
1778         }
1779
1780 end_no_session:
1781         return ret;
1782 }
1783
1784 /*
1785  * relay_reset_metadata: reset a metadata stream
1786  */
1787 static
1788 int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1789                 struct relay_connection *conn,
1790                 const struct lttng_buffer_view *payload)
1791 {
1792         int ret;
1793         ssize_t send_ret;
1794         struct relay_session *session = conn->session;
1795         struct lttcomm_relayd_reset_metadata stream_info;
1796         struct lttcomm_relayd_generic_reply reply;
1797         struct relay_stream *stream;
1798
1799         DBG("Reset metadata received");
1800
1801         if (!session || !conn->version_check_done) {
1802                 ERR("Trying to reset a metadata stream before version check");
1803                 ret = -1;
1804                 goto end_no_session;
1805         }
1806
1807         if (payload->size < sizeof(stream_info)) {
1808                 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1809                                 sizeof(stream_info), payload->size);
1810                 ret = -1;
1811                 goto end_no_session;
1812         }
1813         memcpy(&stream_info, payload->data, sizeof(stream_info));
1814         stream_info.stream_id = be64toh(stream_info.stream_id);
1815         stream_info.version = be64toh(stream_info.version);
1816
1817         DBG("Update metadata to version %" PRIu64, stream_info.version);
1818
1819         /* Unsupported for live sessions for now. */
1820         if (session->live_timer != 0) {
1821                 ret = -1;
1822                 goto end;
1823         }
1824
1825         stream = stream_get_by_id(stream_info.stream_id);
1826         if (!stream) {
1827                 ret = -1;
1828                 goto end;
1829         }
1830         pthread_mutex_lock(&stream->lock);
1831         if (!stream->is_metadata) {
1832                 ret = -1;
1833                 goto end_unlock;
1834         }
1835
1836         ret = stream_reset_file(stream);
1837         if (ret < 0) {
1838                 ERR("Failed to reset metadata stream %" PRIu64
1839                                 ": stream_path = %s, channel = %s",
1840                                 stream->stream_handle, stream->path_name,
1841                                 stream->channel_name);
1842                 goto end_unlock;
1843         }
1844 end_unlock:
1845         pthread_mutex_unlock(&stream->lock);
1846         stream_put(stream);
1847
1848 end:
1849         memset(&reply, 0, sizeof(reply));
1850         if (ret < 0) {
1851                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1852         } else {
1853                 reply.ret_code = htobe32(LTTNG_OK);
1854         }
1855         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1856                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1857         if (send_ret < (ssize_t) sizeof(reply)) {
1858                 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1859                                 send_ret);
1860                 ret = -1;
1861         }
1862
1863 end_no_session:
1864         return ret;
1865 }
1866
1867 /*
1868  * relay_unknown_command: send -1 if received unknown command
1869  */
1870 static void relay_unknown_command(struct relay_connection *conn)
1871 {
1872         struct lttcomm_relayd_generic_reply reply;
1873         ssize_t send_ret;
1874
1875         memset(&reply, 0, sizeof(reply));
1876         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1877         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1878         if (send_ret < sizeof(reply)) {
1879                 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1880         }
1881 }
1882
1883 /*
1884  * relay_start: send an acknowledgment to the client to tell if we are
1885  * ready to receive data. We are ready if a session is established.
1886  */
1887 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr,
1888                 struct relay_connection *conn,
1889                 const struct lttng_buffer_view *payload)
1890 {
1891         int ret = 0;
1892         ssize_t send_ret;
1893         struct lttcomm_relayd_generic_reply reply;
1894         struct relay_session *session = conn->session;
1895
1896         if (!session) {
1897                 DBG("Trying to start the streaming without a session established");
1898                 ret = htobe32(LTTNG_ERR_UNK);
1899         }
1900
1901         memset(&reply, 0, sizeof(reply));
1902         reply.ret_code = htobe32(LTTNG_OK);
1903         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1904                         sizeof(reply), 0);
1905         if (send_ret < (ssize_t) sizeof(reply)) {
1906                 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1907                                 send_ret);
1908                 ret = -1;
1909         }
1910
1911         return ret;
1912 }
1913
1914 /*
1915  * relay_recv_metadata: receive the metadata for the session.
1916  */
1917 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1918                 struct relay_connection *conn,
1919                 const struct lttng_buffer_view *payload)
1920 {
1921         int ret = 0;
1922         struct relay_session *session = conn->session;
1923         struct lttcomm_relayd_metadata_payload metadata_payload_header;
1924         struct relay_stream *metadata_stream;
1925         uint64_t metadata_payload_size;
1926         struct lttng_buffer_view packet_view;
1927
1928         if (!session) {
1929                 ERR("Metadata sent before version check");
1930                 ret = -1;
1931                 goto end;
1932         }
1933
1934         if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1935                 ERR("Incorrect data size");
1936                 ret = -1;
1937                 goto end;
1938         }
1939         metadata_payload_size = recv_hdr->data_size -
1940                         sizeof(struct lttcomm_relayd_metadata_payload);
1941
1942         memcpy(&metadata_payload_header, payload->data,
1943                         sizeof(metadata_payload_header));
1944         metadata_payload_header.stream_id = be64toh(
1945                         metadata_payload_header.stream_id);
1946         metadata_payload_header.padding_size = be32toh(
1947                         metadata_payload_header.padding_size);
1948
1949         metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1950         if (!metadata_stream) {
1951                 ret = -1;
1952                 goto end;
1953         }
1954
1955         packet_view = lttng_buffer_view_from_view(payload,
1956                         sizeof(metadata_payload_header), metadata_payload_size);
1957         if (!packet_view.data) {
1958                 ERR("Invalid metadata packet length announced by header");
1959                 ret = -1;
1960                 goto end_put;
1961         }
1962
1963         pthread_mutex_lock(&metadata_stream->lock);
1964         ret = stream_write(metadata_stream, &packet_view,
1965                         metadata_payload_header.padding_size);
1966         pthread_mutex_unlock(&metadata_stream->lock);
1967         if (ret){
1968                 ret = -1;
1969                 goto end_put;
1970         }
1971 end_put:
1972         stream_put(metadata_stream);
1973 end:
1974         return ret;
1975 }
1976
1977 /*
1978  * relay_send_version: send relayd version number
1979  */
1980 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
1981                 struct relay_connection *conn,
1982                 const struct lttng_buffer_view *payload)
1983 {
1984         int ret;
1985         ssize_t send_ret;
1986         struct lttcomm_relayd_version reply, msg;
1987         bool compatible = true;
1988
1989         conn->version_check_done = true;
1990
1991         /* Get version from the other side. */
1992         if (payload->size < sizeof(msg)) {
1993                 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1994                                 sizeof(msg), payload->size);
1995                 ret = -1;
1996                 goto end;
1997         }
1998
1999         memcpy(&msg, payload->data, sizeof(msg));
2000         msg.major = be32toh(msg.major);
2001         msg.minor = be32toh(msg.minor);
2002
2003         memset(&reply, 0, sizeof(reply));
2004         reply.major = RELAYD_VERSION_COMM_MAJOR;
2005         reply.minor = RELAYD_VERSION_COMM_MINOR;
2006
2007         /* Major versions must be the same */
2008         if (reply.major != msg.major) {
2009                 DBG("Incompatible major versions (%u vs %u), deleting session",
2010                                 reply.major, msg.major);
2011                 compatible = false;
2012         }
2013
2014         conn->major = reply.major;
2015         /* We adapt to the lowest compatible version */
2016         if (reply.minor <= msg.minor) {
2017                 conn->minor = reply.minor;
2018         } else {
2019                 conn->minor = msg.minor;
2020         }
2021
2022         reply.major = htobe32(reply.major);
2023         reply.minor = htobe32(reply.minor);
2024         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2025                         sizeof(reply), 0);
2026         if (send_ret < (ssize_t) sizeof(reply)) {
2027                 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2028                                 send_ret);
2029                 ret = -1;
2030                 goto end;
2031         } else {
2032                 ret = 0;
2033         }
2034
2035         if (!compatible) {
2036                 ret = -1;
2037                 goto end;
2038         }
2039
2040         DBG("Version check done using protocol %u.%u", conn->major,
2041                         conn->minor);
2042
2043 end:
2044         return ret;
2045 }
2046
2047 /*
2048  * Check for data pending for a given stream id from the session daemon.
2049  */
2050 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2051                 struct relay_connection *conn,
2052                 const struct lttng_buffer_view *payload)
2053 {
2054         struct relay_session *session = conn->session;
2055         struct lttcomm_relayd_data_pending msg;
2056         struct lttcomm_relayd_generic_reply reply;
2057         struct relay_stream *stream;
2058         ssize_t send_ret;
2059         int ret;
2060         uint64_t stream_seq;
2061
2062         DBG("Data pending command received");
2063
2064         if (!session || !conn->version_check_done) {
2065                 ERR("Trying to check for data before version check");
2066                 ret = -1;
2067                 goto end_no_session;
2068         }
2069
2070         if (payload->size < sizeof(msg)) {
2071                 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2072                                 sizeof(msg), payload->size);
2073                 ret = -1;
2074                 goto end_no_session;
2075         }
2076         memcpy(&msg, payload->data, sizeof(msg));
2077         msg.stream_id = be64toh(msg.stream_id);
2078         msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2079
2080         stream = stream_get_by_id(msg.stream_id);
2081         if (stream == NULL) {
2082                 ret = -1;
2083                 goto end;
2084         }
2085
2086         pthread_mutex_lock(&stream->lock);
2087
2088         if (session_streams_have_index(session)) {
2089                 /*
2090                  * Ensure that both the index and stream data have been
2091                  * flushed up to the requested point.
2092                  */
2093                 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2094         } else {
2095                 stream_seq = stream->prev_data_seq;
2096         }
2097         DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2098                         ", prev_index_seq %" PRIu64
2099                         ", and last_seq %" PRIu64, msg.stream_id,
2100                         stream->prev_data_seq, stream->prev_index_seq,
2101                         msg.last_net_seq_num);
2102
2103         /* Avoid wrapping issue */
2104         if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2105                 /* Data has in fact been written and is NOT pending */
2106                 ret = 0;
2107         } else {
2108                 /* Data still being streamed thus pending */
2109                 ret = 1;
2110         }
2111
2112         stream->data_pending_check_done = true;
2113         pthread_mutex_unlock(&stream->lock);
2114
2115         stream_put(stream);
2116 end:
2117
2118         memset(&reply, 0, sizeof(reply));
2119         reply.ret_code = htobe32(ret);
2120         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2121         if (send_ret < (ssize_t) sizeof(reply)) {
2122                 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2123                                 send_ret);
2124                 ret = -1;
2125         }
2126
2127 end_no_session:
2128         return ret;
2129 }
2130
2131 /*
2132  * Wait for the control socket to reach a quiescent state.
2133  *
2134  * Note that for now, when receiving this command from the session
2135  * daemon, this means that every subsequent commands or data received on
2136  * the control socket has been handled. So, this is why we simply return
2137  * OK here.
2138  */
2139 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
2140                 struct relay_connection *conn,
2141                 const struct lttng_buffer_view *payload)
2142 {
2143         int ret;
2144         ssize_t send_ret;
2145         struct relay_stream *stream;
2146         struct lttcomm_relayd_quiescent_control msg;
2147         struct lttcomm_relayd_generic_reply reply;
2148
2149         DBG("Checking quiescent state on control socket");
2150
2151         if (!conn->session || !conn->version_check_done) {
2152                 ERR("Trying to check for data before version check");
2153                 ret = -1;
2154                 goto end_no_session;
2155         }
2156
2157         if (payload->size < sizeof(msg)) {
2158                 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2159                                 sizeof(msg), payload->size);
2160                 ret = -1;
2161                 goto end_no_session;
2162         }
2163         memcpy(&msg, payload->data, sizeof(msg));
2164         msg.stream_id = be64toh(msg.stream_id);
2165
2166         stream = stream_get_by_id(msg.stream_id);
2167         if (!stream) {
2168                 goto reply;
2169         }
2170         pthread_mutex_lock(&stream->lock);
2171         stream->data_pending_check_done = true;
2172         pthread_mutex_unlock(&stream->lock);
2173
2174         DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2175         stream_put(stream);
2176 reply:
2177         memset(&reply, 0, sizeof(reply));
2178         reply.ret_code = htobe32(LTTNG_OK);
2179         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2180         if (send_ret < (ssize_t) sizeof(reply)) {
2181                 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2182                                 send_ret);
2183                 ret = -1;
2184         } else {
2185                 ret = 0;
2186         }
2187
2188 end_no_session:
2189         return ret;
2190 }
2191
2192 /*
2193  * Initialize a data pending command. This means that a consumer is about
2194  * to ask for data pending for each stream it holds. Simply iterate over
2195  * all streams of a session and set the data_pending_check_done flag.
2196  *
2197  * This command returns to the client a LTTNG_OK code.
2198  */
2199 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2200                 struct relay_connection *conn,
2201                 const struct lttng_buffer_view *payload)
2202 {
2203         int ret;
2204         ssize_t send_ret;
2205         struct lttng_ht_iter iter;
2206         struct lttcomm_relayd_begin_data_pending msg;
2207         struct lttcomm_relayd_generic_reply reply;
2208         struct relay_stream *stream;
2209
2210         assert(recv_hdr);
2211         assert(conn);
2212
2213         DBG("Init streams for data pending");
2214
2215         if (!conn->session || !conn->version_check_done) {
2216                 ERR("Trying to check for data before version check");
2217                 ret = -1;
2218                 goto end_no_session;
2219         }
2220
2221         if (payload->size < sizeof(msg)) {
2222                 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2223                                 sizeof(msg), payload->size);
2224                 ret = -1;
2225                 goto end_no_session;
2226         }
2227         memcpy(&msg, payload->data, sizeof(msg));
2228         msg.session_id = be64toh(msg.session_id);
2229
2230         /*
2231          * Iterate over all streams to set the begin data pending flag.
2232          * For now, the streams are indexed by stream handle so we have
2233          * to iterate over all streams to find the one associated with
2234          * the right session_id.
2235          */
2236         rcu_read_lock();
2237         cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2238                         node.node) {
2239                 if (!stream_get(stream)) {
2240                         continue;
2241                 }
2242                 if (stream->trace->session->id == msg.session_id) {
2243                         pthread_mutex_lock(&stream->lock);
2244                         stream->data_pending_check_done = false;
2245                         pthread_mutex_unlock(&stream->lock);
2246                         DBG("Set begin data pending flag to stream %" PRIu64,
2247                                         stream->stream_handle);
2248                 }
2249                 stream_put(stream);
2250         }
2251         rcu_read_unlock();
2252
2253         memset(&reply, 0, sizeof(reply));
2254         /* All good, send back reply. */
2255         reply.ret_code = htobe32(LTTNG_OK);
2256
2257         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2258         if (send_ret < (ssize_t) sizeof(reply)) {
2259                 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2260                         send_ret);
2261                 ret = -1;
2262         } else {
2263                 ret = 0;
2264         }
2265
2266 end_no_session:
2267         return ret;
2268 }
2269
2270 /*
2271  * End data pending command. This will check, for a given session id, if
2272  * each stream associated with it has its data_pending_check_done flag
2273  * set. If not, this means that the client lost track of the stream but
2274  * the data is still being streamed on our side. In this case, we inform
2275  * the client that data is in flight.
2276  *
2277  * Return to the client if there is data in flight or not with a ret_code.
2278  */
2279 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2280                 struct relay_connection *conn,
2281                 const struct lttng_buffer_view *payload)
2282 {
2283         int ret;
2284         ssize_t send_ret;
2285         struct lttng_ht_iter iter;
2286         struct lttcomm_relayd_end_data_pending msg;
2287         struct lttcomm_relayd_generic_reply reply;
2288         struct relay_stream *stream;
2289         uint32_t is_data_inflight = 0;
2290
2291         DBG("End data pending command");
2292
2293         if (!conn->session || !conn->version_check_done) {
2294                 ERR("Trying to check for data before version check");
2295                 ret = -1;
2296                 goto end_no_session;
2297         }
2298
2299         if (payload->size < sizeof(msg)) {
2300                 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2301                                 sizeof(msg), payload->size);
2302                 ret = -1;
2303                 goto end_no_session;
2304         }
2305         memcpy(&msg, payload->data, sizeof(msg));
2306         msg.session_id = be64toh(msg.session_id);
2307
2308         /*
2309          * Iterate over all streams to see if the begin data pending
2310          * flag is set.
2311          */
2312         rcu_read_lock();
2313         cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2314                         node.node) {
2315                 if (!stream_get(stream)) {
2316                         continue;
2317                 }
2318                 if (stream->trace->session->id != msg.session_id) {
2319                         stream_put(stream);
2320                         continue;
2321                 }
2322                 pthread_mutex_lock(&stream->lock);
2323                 if (!stream->data_pending_check_done) {
2324                         uint64_t stream_seq;
2325
2326                         if (session_streams_have_index(conn->session)) {
2327                                 /*
2328                                  * Ensure that both the index and stream data have been
2329                                  * flushed up to the requested point.
2330                                  */
2331                                 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2332                         } else {
2333                                 stream_seq = stream->prev_data_seq;
2334                         }
2335                         if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2336                                 is_data_inflight = 1;
2337                                 DBG("Data is still in flight for stream %" PRIu64,
2338                                                 stream->stream_handle);
2339                                 pthread_mutex_unlock(&stream->lock);
2340                                 stream_put(stream);
2341                                 break;
2342                         }
2343                 }
2344                 pthread_mutex_unlock(&stream->lock);
2345                 stream_put(stream);
2346         }
2347         rcu_read_unlock();
2348
2349         memset(&reply, 0, sizeof(reply));
2350         /* All good, send back reply. */
2351         reply.ret_code = htobe32(is_data_inflight);
2352
2353         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2354         if (send_ret < (ssize_t) sizeof(reply)) {
2355                 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2356                         send_ret);
2357                 ret = -1;
2358         } else {
2359                 ret = 0;
2360         }
2361
2362 end_no_session:
2363         return ret;
2364 }
2365
2366 /*
2367  * Receive an index for a specific stream.
2368  *
2369  * Return 0 on success else a negative value.
2370  */
2371 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
2372                 struct relay_connection *conn,
2373                 const struct lttng_buffer_view *payload)
2374 {
2375         int ret;
2376         ssize_t send_ret;
2377         struct relay_session *session = conn->session;
2378         struct lttcomm_relayd_index index_info;
2379         struct lttcomm_relayd_generic_reply reply;
2380         struct relay_stream *stream;
2381         size_t msg_len;
2382
2383         assert(conn);
2384
2385         DBG("Relay receiving index");
2386
2387         if (!session || !conn->version_check_done) {
2388                 ERR("Trying to close a stream before version check");
2389                 ret = -1;
2390                 goto end_no_session;
2391         }
2392
2393         msg_len = lttcomm_relayd_index_len(
2394                         lttng_to_index_major(conn->major, conn->minor),
2395                         lttng_to_index_minor(conn->major, conn->minor));
2396         if (payload->size < msg_len) {
2397                 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2398                                 msg_len, payload->size);
2399                 ret = -1;
2400                 goto end_no_session;
2401         }
2402         memcpy(&index_info, payload->data, msg_len);
2403         index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2404         index_info.net_seq_num = be64toh(index_info.net_seq_num);
2405         index_info.packet_size = be64toh(index_info.packet_size);
2406         index_info.content_size = be64toh(index_info.content_size);
2407         index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2408         index_info.timestamp_end = be64toh(index_info.timestamp_end);
2409         index_info.events_discarded = be64toh(index_info.events_discarded);
2410         index_info.stream_id = be64toh(index_info.stream_id);
2411
2412         if (conn->minor >= 8) {
2413                 index_info.stream_instance_id =
2414                                 be64toh(index_info.stream_instance_id);
2415                 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2416         } else {
2417                 index_info.stream_instance_id = -1ULL;
2418                 index_info.packet_seq_num = -1ULL;
2419         }
2420
2421         stream = stream_get_by_id(index_info.relay_stream_id);
2422         if (!stream) {
2423                 ERR("stream_get_by_id not found");
2424                 ret = -1;
2425                 goto end;
2426         }
2427
2428         pthread_mutex_lock(&stream->lock);
2429         ret = stream_add_index(stream, &index_info);
2430         pthread_mutex_unlock(&stream->lock);
2431         if (ret) {
2432                 goto end_stream_put;
2433         }
2434
2435 end_stream_put:
2436         stream_put(stream);
2437 end:
2438         memset(&reply, 0, sizeof(reply));
2439         if (ret < 0) {
2440                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2441         } else {
2442                 reply.ret_code = htobe32(LTTNG_OK);
2443         }
2444         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2445         if (send_ret < (ssize_t) sizeof(reply)) {
2446                 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2447                 ret = -1;
2448         }
2449
2450 end_no_session:
2451         return ret;
2452 }
2453
2454 /*
2455  * Receive the streams_sent message.
2456  *
2457  * Return 0 on success else a negative value.
2458  */
2459 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
2460                 struct relay_connection *conn,
2461                 const struct lttng_buffer_view *payload)
2462 {
2463         int ret;
2464         ssize_t send_ret;
2465         struct lttcomm_relayd_generic_reply reply;
2466
2467         assert(conn);
2468
2469         DBG("Relay receiving streams_sent");
2470
2471         if (!conn->session || !conn->version_check_done) {
2472                 ERR("Trying to close a stream before version check");
2473                 ret = -1;
2474                 goto end_no_session;
2475         }
2476
2477         /*
2478          * Publish every pending stream in the connection recv list which are
2479          * now ready to be used by the viewer.
2480          */
2481         publish_connection_local_streams(conn);
2482
2483         memset(&reply, 0, sizeof(reply));
2484         reply.ret_code = htobe32(LTTNG_OK);
2485         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2486         if (send_ret < (ssize_t) sizeof(reply)) {
2487                 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2488                         send_ret);
2489                 ret = -1;
2490         } else {
2491                 /* Success. */
2492                 ret = 0;
2493         }
2494
2495 end_no_session:
2496         return ret;
2497 }
2498
2499 /*
2500  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2501  * session rotation feature (not the tracefile rotation feature).
2502  */
2503 static int relay_rotate_session_streams(
2504                 const struct lttcomm_relayd_hdr *recv_hdr,
2505                 struct relay_connection *conn,
2506                 const struct lttng_buffer_view *payload)
2507 {
2508         int ret = 0;
2509         uint32_t i;
2510         ssize_t send_ret;
2511         enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2512         struct relay_session *session = conn->session;
2513         struct lttcomm_relayd_rotate_streams rotate_streams;
2514         struct lttcomm_relayd_generic_reply reply = {};
2515         struct relay_stream *stream = NULL;
2516         const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams);
2517         struct lttng_trace_chunk *next_trace_chunk = NULL;
2518         struct lttng_buffer_view stream_positions;
2519         char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2520         const char *chunk_id_str = "none";
2521
2522         if (!session || !conn->version_check_done) {
2523                 ERR("Trying to rotate a stream before version check");
2524                 ret = -1;
2525                 goto end_no_reply;
2526         }
2527
2528         if (session->major == 2 && session->minor < 11) {
2529                 ERR("Unsupported feature before 2.11");
2530                 ret = -1;
2531                 goto end_no_reply;
2532         }
2533
2534         if (payload->size < header_len) {
2535                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2536                                 header_len, payload->size);
2537                 ret = -1;
2538                 goto end_no_reply;
2539         }
2540
2541         memcpy(&rotate_streams, payload->data, header_len);
2542
2543         /* Convert header to host endianness. */
2544         rotate_streams = (typeof(rotate_streams)) {
2545                 .stream_count = be32toh(rotate_streams.stream_count),
2546                 .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) {
2547                         .is_set = !!rotate_streams.new_chunk_id.is_set,
2548                         .value = be64toh(rotate_streams.new_chunk_id.value),
2549                 }
2550         };
2551
2552         if (rotate_streams.new_chunk_id.is_set) {
2553                 /*
2554                  * Retrieve the trace chunk the stream must transition to. As
2555                  * per the protocol, this chunk should have been created
2556                  * before this command is received.
2557                  */
2558                 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2559                                 sessiond_trace_chunk_registry,
2560                                 session->sessiond_uuid, session->id,
2561                                 rotate_streams.new_chunk_id.value);
2562                 if (!next_trace_chunk) {
2563                         char uuid_str[LTTNG_UUID_STR_LEN];
2564
2565                         lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2566                         ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2567                                         ", trace_chunk_id = %" PRIu64,
2568                                         uuid_str, session->id,
2569                                         rotate_streams.new_chunk_id.value);
2570                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2571                         ret = -1;
2572                         goto end;
2573                 }
2574
2575                 ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64,
2576                                 rotate_streams.new_chunk_id.value);
2577                 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2578                         chunk_id_str = "formatting error";
2579                 } else {
2580                         chunk_id_str = chunk_id_buf;
2581                 }
2582         }
2583
2584         DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2585                         rotate_streams.stream_count, session->session_name,
2586                         chunk_id_str);
2587
2588         stream_positions = lttng_buffer_view_from_view(payload,
2589                         sizeof(rotate_streams), -1);
2590         if (!stream_positions.data ||
2591                         stream_positions.size <
2592                                         (rotate_streams.stream_count *
2593                                                         sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2594                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2595                 ret = -1;
2596                 goto end;
2597         }
2598
2599         for (i = 0; i < rotate_streams.stream_count; i++) {
2600                 struct lttcomm_relayd_stream_rotation_position *position_comm =
2601                                 &((typeof(position_comm)) stream_positions.data)[i];
2602                 const struct lttcomm_relayd_stream_rotation_position pos = {
2603                         .stream_id = be64toh(position_comm->stream_id),
2604                         .rotate_at_seq_num = be64toh(
2605                                         position_comm->rotate_at_seq_num),
2606                 };
2607
2608                 stream = stream_get_by_id(pos.stream_id);
2609                 if (!stream) {
2610                         reply_code = LTTNG_ERR_INVALID;
2611                         ret = -1;
2612                         goto end;
2613                 }
2614
2615                 pthread_mutex_lock(&stream->lock);
2616                 ret = stream_set_pending_rotation(stream, next_trace_chunk,
2617                                 pos.rotate_at_seq_num);
2618                 pthread_mutex_unlock(&stream->lock);
2619                 if (ret) {
2620                         reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2621                         goto end;
2622                 }
2623
2624                 stream_put(stream);
2625                 stream = NULL;
2626         }
2627
2628         reply_code = LTTNG_OK;
2629         ret = 0;
2630 end:
2631         if (stream) {
2632                 stream_put(stream);
2633         }
2634
2635         reply.ret_code = htobe32((uint32_t) reply_code);
2636         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2637                         sizeof(struct lttcomm_relayd_generic_reply), 0);
2638         if (send_ret < (ssize_t) sizeof(reply)) {
2639                 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2640                                 send_ret);
2641                 ret = -1;
2642         }
2643 end_no_reply:
2644         lttng_trace_chunk_put(next_trace_chunk);
2645         return ret;
2646 }
2647
2648
2649
2650 /*
2651  * relay_create_trace_chunk: create a new trace chunk
2652  */
2653 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2654                 struct relay_connection *conn,
2655                 const struct lttng_buffer_view *payload)
2656 {
2657         int ret = 0;
2658         ssize_t send_ret;
2659         struct relay_session *session = conn->session;
2660         struct lttcomm_relayd_create_trace_chunk *msg;
2661         struct lttcomm_relayd_generic_reply reply = {};
2662         struct lttng_buffer_view header_view;
2663         struct lttng_buffer_view chunk_name_view;
2664         struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
2665         enum lttng_error_code reply_code = LTTNG_OK;
2666         enum lttng_trace_chunk_status chunk_status;
2667         const char *new_path;
2668
2669         if (!session || !conn->version_check_done) {
2670                 ERR("Trying to create a trace chunk before version check");
2671                 ret = -1;
2672                 goto end_no_reply;
2673         }
2674
2675         if (session->major == 2 && session->minor < 11) {
2676                 ERR("Chunk creation command is unsupported before 2.11");
2677                 ret = -1;
2678                 goto end_no_reply;
2679         }
2680
2681         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2682         if (!header_view.data) {
2683                 ERR("Failed to receive payload of chunk creation command");
2684                 ret = -1;
2685                 goto end_no_reply;
2686         }
2687
2688         /* Convert to host endianness. */
2689         msg = (typeof(msg)) header_view.data;
2690         msg->chunk_id = be64toh(msg->chunk_id);
2691         msg->creation_timestamp = be64toh(msg->creation_timestamp);
2692         msg->override_name_length = be32toh(msg->override_name_length);
2693
2694         if (session->current_trace_chunk &&
2695                         !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2696                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2697                                         DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2698                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2699                         ERR("Failed to rename old chunk");
2700                         ret = -1;
2701                         reply_code = LTTNG_ERR_UNK;
2702                         goto end;
2703                 }
2704         }
2705         session->ongoing_rotation = true;
2706         if (!session->current_trace_chunk) {
2707                 if (!session->has_rotated) {
2708                         new_path = "";
2709                 } else {
2710                         new_path = NULL;
2711                 }
2712         } else {
2713                 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2714         }
2715         chunk = lttng_trace_chunk_create(
2716                         msg->chunk_id, msg->creation_timestamp, new_path);
2717         if (!chunk) {
2718                 ERR("Failed to create trace chunk in trace chunk creation command");
2719                 ret = -1;
2720                 reply_code = LTTNG_ERR_NOMEM;
2721                 goto end;
2722         }
2723         lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2724
2725         if (msg->override_name_length) {
2726                 const char *name;
2727
2728                 chunk_name_view = lttng_buffer_view_from_view(payload,
2729                                 sizeof(*msg),
2730                                 msg->override_name_length);
2731                 name = chunk_name_view.data;
2732                 if (!name || name[msg->override_name_length - 1]) {
2733                         ERR("Failed to receive payload of chunk creation command");
2734                         ret = -1;
2735                         reply_code = LTTNG_ERR_INVALID;
2736                         goto end;
2737                 }
2738
2739                 chunk_status = lttng_trace_chunk_override_name(
2740                                 chunk, chunk_name_view.data);
2741                 switch (chunk_status) {
2742                 case LTTNG_TRACE_CHUNK_STATUS_OK:
2743                         break;
2744                 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2745                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2746                         reply_code = LTTNG_ERR_INVALID;
2747                         ret = -1;
2748                         goto end;
2749                 default:
2750                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2751                         reply_code = LTTNG_ERR_UNK;
2752                         ret = -1;
2753                         goto end;
2754                 }
2755         }
2756
2757         chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2758         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2759                 reply_code = LTTNG_ERR_UNK;
2760                 ret = -1;
2761                 goto end;
2762         }
2763
2764         assert(conn->session->output_directory);
2765         chunk_status = lttng_trace_chunk_set_as_owner(chunk,
2766                         conn->session->output_directory);
2767         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2768                 reply_code = LTTNG_ERR_UNK;
2769                 ret = -1;
2770                 goto end;
2771         }
2772
2773         published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2774                         sessiond_trace_chunk_registry,
2775                         conn->session->sessiond_uuid,
2776                         conn->session->id,
2777                         chunk);
2778         if (!published_chunk) {
2779                 char uuid_str[LTTNG_UUID_STR_LEN];
2780
2781                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2782                 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
2783                                 uuid_str,
2784                                 conn->session->id,
2785                                 msg->chunk_id);
2786                 ret = -1;
2787                 reply_code = LTTNG_ERR_NOMEM;
2788                 goto end;
2789         }
2790
2791         pthread_mutex_lock(&conn->session->lock);
2792         if (conn->session->pending_closure_trace_chunk) {
2793                 /*
2794                  * Invalid; this means a second create_trace_chunk command was
2795                  * received before a close_trace_chunk.
2796                  */
2797                 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2798                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2799                 ret = -1;
2800                 goto end_unlock_session;
2801         }
2802         conn->session->pending_closure_trace_chunk =
2803                         conn->session->current_trace_chunk;
2804         conn->session->current_trace_chunk = published_chunk;
2805         published_chunk = NULL;
2806         if (!conn->session->pending_closure_trace_chunk) {
2807                 session->ongoing_rotation = false;
2808         }
2809 end_unlock_session:
2810         pthread_mutex_unlock(&conn->session->lock);
2811 end:
2812         reply.ret_code = htobe32((uint32_t) reply_code);
2813         send_ret = conn->sock->ops->sendmsg(conn->sock,
2814                         &reply,
2815                         sizeof(struct lttcomm_relayd_generic_reply),
2816                         0);
2817         if (send_ret < (ssize_t) sizeof(reply)) {
2818                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2819                                 send_ret);
2820                 ret = -1;
2821         }
2822 end_no_reply:
2823         lttng_trace_chunk_put(chunk);
2824         lttng_trace_chunk_put(published_chunk);
2825         return ret;
2826 }
2827
2828 /*
2829  * relay_close_trace_chunk: close a trace chunk
2830  */
2831 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2832                 struct relay_connection *conn,
2833                 const struct lttng_buffer_view *payload)
2834 {
2835         int ret = 0, buf_ret;
2836         ssize_t send_ret;
2837         struct relay_session *session = conn->session;
2838         struct lttcomm_relayd_close_trace_chunk *msg;
2839         struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2840         struct lttng_buffer_view header_view;
2841         struct lttng_trace_chunk *chunk = NULL;
2842         enum lttng_error_code reply_code = LTTNG_OK;
2843         enum lttng_trace_chunk_status chunk_status;
2844         uint64_t chunk_id;
2845         LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2846         time_t close_timestamp;
2847         char closed_trace_chunk_path[LTTNG_PATH_MAX];
2848         size_t path_length = 0;
2849         const char *chunk_name = NULL;
2850         struct lttng_dynamic_buffer reply_payload;
2851         const char *new_path;
2852
2853         lttng_dynamic_buffer_init(&reply_payload);
2854
2855         if (!session || !conn->version_check_done) {
2856                 ERR("Trying to close a trace chunk before version check");
2857                 ret = -1;
2858                 goto end_no_reply;
2859         }
2860
2861         if (session->major == 2 && session->minor < 11) {
2862                 ERR("Chunk close command is unsupported before 2.11");
2863                 ret = -1;
2864                 goto end_no_reply;
2865         }
2866
2867         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2868         if (!header_view.data) {
2869                 ERR("Failed to receive payload of chunk close command");
2870                 ret = -1;
2871                 goto end_no_reply;
2872         }
2873
2874         /* Convert to host endianness. */
2875         msg = (typeof(msg)) header_view.data;
2876         chunk_id = be64toh(msg->chunk_id);
2877         close_timestamp = (time_t) be64toh(msg->close_timestamp);
2878         close_command = (typeof(close_command)){
2879                 .value = be32toh(msg->close_command.value),
2880                 .is_set = msg->close_command.is_set,
2881         };
2882
2883         chunk = sessiond_trace_chunk_registry_get_chunk(
2884                         sessiond_trace_chunk_registry,
2885                         conn->session->sessiond_uuid,
2886                         conn->session->id,
2887                         chunk_id);
2888         if (!chunk) {
2889                 char uuid_str[LTTNG_UUID_STR_LEN];
2890
2891                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2892                 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
2893                                 uuid_str,
2894                                 conn->session->id,
2895                                 msg->chunk_id);
2896                 ret = -1;
2897                 reply_code = LTTNG_ERR_NOMEM;
2898                 goto end;
2899         }
2900
2901         pthread_mutex_lock(&session->lock);
2902         if (close_command.is_set &&
2903                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
2904                 /*
2905                  * Clear command. It is a protocol error to ask for a
2906                  * clear on a relay which does not allow it. Querying
2907                  * the configuration allows figuring out whether
2908                  * clearing is allowed before doing the clear.
2909                  */
2910                 if (!opt_allow_clear) {
2911                         ret = -1;
2912                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2913                         goto end_unlock_session;
2914                 }
2915         }
2916         if (session->pending_closure_trace_chunk &&
2917                         session->pending_closure_trace_chunk != chunk) {
2918                 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2919                                 session->session_name);
2920                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2921                 ret = -1;
2922                 goto end_unlock_session;
2923         }
2924
2925         if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
2926                         !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2927                 if (close_command.is_set &&
2928                                 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
2929                                 !session->has_rotated) {
2930                         /* New chunk stays in session output directory. */
2931                         new_path = "";
2932                 } else {
2933                         /* Use chunk name for new chunk. */
2934                         new_path = NULL;
2935                 }
2936                 /* Rename new chunk path. */
2937                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2938                                 new_path);
2939                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2940                         ret = -1;
2941                         goto end_unlock_session;
2942                 }
2943                 session->ongoing_rotation = false;
2944         }
2945         if ((!close_command.is_set ||
2946                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
2947                         !lttng_trace_chunk_get_name_overridden(chunk)) {
2948                 const char *old_path;
2949
2950                 if (!session->has_rotated) {
2951                         old_path = "";
2952                 } else {
2953                         old_path = NULL;
2954                 }
2955                 /* We need to move back the .tmp_old_chunk to its rightful place. */
2956                 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
2957                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2958                         ret = -1;
2959                         goto end_unlock_session;
2960                 }
2961         }
2962         chunk_status = lttng_trace_chunk_set_close_timestamp(
2963                         chunk, close_timestamp);
2964         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2965                 ERR("Failed to set trace chunk close timestamp");
2966                 ret = -1;
2967                 reply_code = LTTNG_ERR_UNK;
2968                 goto end_unlock_session;
2969         }
2970
2971         if (close_command.is_set) {
2972                 chunk_status = lttng_trace_chunk_set_close_command(
2973                                 chunk, close_command.value);
2974                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2975                         ret = -1;
2976                         reply_code = LTTNG_ERR_INVALID;
2977                         goto end_unlock_session;
2978                 }
2979         }
2980         chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL);
2981         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2982                 ERR("Failed to get chunk name");
2983                 ret = -1;
2984                 reply_code = LTTNG_ERR_UNK;
2985                 goto end_unlock_session;
2986         }
2987         if (!session->has_rotated && !session->snapshot) {
2988                 ret = lttng_strncpy(closed_trace_chunk_path,
2989                                 session->output_path,
2990                                 sizeof(closed_trace_chunk_path));
2991                 if (ret) {
2992                         ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
2993                                         strlen(session->output_path),
2994                                         sizeof(closed_trace_chunk_path));
2995                         reply_code = LTTNG_ERR_NOMEM;
2996                         ret = -1;
2997                         goto end_unlock_session;
2998                 }
2999         } else {
3000                 if (session->snapshot) {
3001                         ret = snprintf(closed_trace_chunk_path,
3002                                         sizeof(closed_trace_chunk_path),
3003                                         "%s/%s", session->output_path,
3004                                         chunk_name);
3005                 } else {
3006                         ret = snprintf(closed_trace_chunk_path,
3007                                         sizeof(closed_trace_chunk_path),
3008                                         "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3009                                         "/%s",
3010                                         session->output_path, chunk_name);
3011                 }
3012                 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3013                         ERR("Failed to format closed trace chunk resulting path");
3014                         reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3015                         ret = -1;
3016                         goto end_unlock_session;
3017                 }
3018         }
3019         if (close_command.is_set &&
3020                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3021                 session->has_rotated = true;
3022         }
3023         DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3024         path_length = strlen(closed_trace_chunk_path) + 1;
3025         if (path_length > UINT32_MAX) {
3026                 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3027                 ret = -1;
3028                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3029                 goto end_unlock_session;
3030         }
3031
3032         if (session->current_trace_chunk == chunk) {
3033                 /*
3034                  * After a trace chunk close command, no new streams
3035                  * referencing the chunk may be created. Hence, on the
3036                  * event that no new trace chunk have been created for
3037                  * the session, the reference to the current trace chunk
3038                  * is released in order to allow it to be reclaimed when
3039                  * the last stream releases its reference to it.
3040                  */
3041                 lttng_trace_chunk_put(session->current_trace_chunk);
3042                 session->current_trace_chunk = NULL;
3043         }
3044         lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3045         session->pending_closure_trace_chunk = NULL;
3046 end_unlock_session:
3047         pthread_mutex_unlock(&session->lock);
3048
3049 end:
3050         reply.generic.ret_code = htobe32((uint32_t) reply_code);
3051         reply.path_length = htobe32((uint32_t) path_length);
3052         buf_ret = lttng_dynamic_buffer_append(
3053                         &reply_payload, &reply, sizeof(reply));
3054         if (buf_ret) {
3055                 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3056                 goto end_no_reply;
3057         }
3058
3059         if (reply_code == LTTNG_OK) {
3060                 buf_ret = lttng_dynamic_buffer_append(&reply_payload,
3061                                 closed_trace_chunk_path, path_length);
3062                 if (buf_ret) {
3063                         ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3064                         goto end_no_reply;
3065                 }
3066         }
3067
3068         send_ret = conn->sock->ops->sendmsg(conn->sock,
3069                         reply_payload.data,
3070                         reply_payload.size,
3071                         0);
3072         if (send_ret < reply_payload.size) {
3073                 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3074                                 reply_payload.size, send_ret);
3075                 ret = -1;
3076                 goto end_no_reply;
3077         }
3078 end_no_reply:
3079         lttng_trace_chunk_put(chunk);
3080         lttng_dynamic_buffer_reset(&reply_payload);
3081         return ret;
3082 }
3083
3084 /*
3085  * relay_trace_chunk_exists: check if a trace chunk exists
3086  */
3087 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr,
3088                 struct relay_connection *conn,
3089                 const struct lttng_buffer_view *payload)
3090 {
3091         int ret = 0;
3092         ssize_t send_ret;
3093         struct relay_session *session = conn->session;
3094         struct lttcomm_relayd_trace_chunk_exists *msg;
3095         struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3096         struct lttng_buffer_view header_view;
3097         uint64_t chunk_id;
3098         bool chunk_exists;
3099
3100         if (!session || !conn->version_check_done) {
3101                 ERR("Trying to close a trace chunk before version check");
3102                 ret = -1;
3103                 goto end_no_reply;
3104         }
3105
3106         if (session->major == 2 && session->minor < 11) {
3107                 ERR("Chunk close command is unsupported before 2.11");
3108                 ret = -1;
3109                 goto end_no_reply;
3110         }
3111
3112         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3113         if (!header_view.data) {
3114                 ERR("Failed to receive payload of chunk close command");
3115                 ret = -1;
3116                 goto end_no_reply;
3117         }
3118
3119         /* Convert to host endianness. */
3120         msg = (typeof(msg)) header_view.data;
3121         chunk_id = be64toh(msg->chunk_id);
3122
3123         ret = sessiond_trace_chunk_registry_chunk_exists(
3124                         sessiond_trace_chunk_registry,
3125                         conn->session->sessiond_uuid,
3126                         conn->session->id,
3127                         chunk_id, &chunk_exists);
3128         /*
3129          * If ret is not 0, send the reply and report the error to the caller.
3130          * It is a protocol (or internal) error and the session/connection
3131          * should be torn down.
3132          */
3133         reply = (typeof(reply)){
3134                 .generic.ret_code = htobe32((uint32_t)
3135                         (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3136                 .trace_chunk_exists = ret == 0 ? chunk_exists : 0,
3137         };
3138         send_ret = conn->sock->ops->sendmsg(
3139                         conn->sock, &reply, sizeof(reply), 0);
3140         if (send_ret < (ssize_t) sizeof(reply)) {
3141                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3142                                 send_ret);
3143                 ret = -1;
3144         }
3145 end_no_reply:
3146         return ret;
3147 }
3148
3149 /*
3150  * relay_get_configuration: query whether feature is available
3151  */
3152 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr,
3153                 struct relay_connection *conn,
3154                 const struct lttng_buffer_view *payload)
3155 {
3156         int ret = 0;
3157         ssize_t send_ret;
3158         struct lttcomm_relayd_get_configuration *msg;
3159         struct lttcomm_relayd_get_configuration_reply reply = {};
3160         struct lttng_buffer_view header_view;
3161         uint64_t query_flags = 0;
3162         uint64_t result_flags = 0;
3163
3164         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3165         if (!header_view.data) {
3166                 ERR("Failed to receive payload of chunk close command");
3167                 ret = -1;
3168                 goto end_no_reply;
3169         }
3170
3171         /* Convert to host endianness. */
3172         msg = (typeof(msg)) header_view.data;
3173         query_flags = be64toh(msg->query_flags);
3174
3175         if (query_flags) {
3176                 ret = LTTNG_ERR_INVALID_PROTOCOL;
3177                 goto reply;
3178         }
3179         if (opt_allow_clear) {
3180                 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3181         }
3182         ret = 0;
3183 reply:
3184         reply = (typeof(reply)){
3185                 .generic.ret_code = htobe32((uint32_t)
3186                         (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3187                 .relayd_configuration_flags = htobe64(result_flags),
3188         };
3189         send_ret = conn->sock->ops->sendmsg(
3190                         conn->sock, &reply, sizeof(reply), 0);
3191         if (send_ret < (ssize_t) sizeof(reply)) {
3192                 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3193                                 send_ret);
3194                 ret = -1;
3195         }
3196 end_no_reply:
3197         return ret;
3198 }
3199
3200 #define DBG_CMD(cmd_name, conn) \
3201                 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3202
3203 static int relay_process_control_command(struct relay_connection *conn,
3204                 const struct lttcomm_relayd_hdr *header,
3205                 const struct lttng_buffer_view *payload)
3206 {
3207         int ret = 0;
3208
3209         switch (header->cmd) {
3210         case RELAYD_CREATE_SESSION:
3211                 DBG_CMD("RELAYD_CREATE_SESSION", conn);
3212                 ret = relay_create_session(header, conn, payload);
3213                 break;
3214         case RELAYD_ADD_STREAM:
3215                 DBG_CMD("RELAYD_ADD_STREAM", conn);
3216                 ret = relay_add_stream(header, conn, payload);
3217                 break;
3218         case RELAYD_START_DATA:
3219                 DBG_CMD("RELAYD_START_DATA", conn);
3220                 ret = relay_start(header, conn, payload);
3221                 break;
3222         case RELAYD_SEND_METADATA:
3223                 DBG_CMD("RELAYD_SEND_METADATA", conn);
3224                 ret = relay_recv_metadata(header, conn, payload);
3225                 break;
3226         case RELAYD_VERSION:
3227                 DBG_CMD("RELAYD_VERSION", conn);
3228                 ret = relay_send_version(header, conn, payload);
3229                 break;
3230         case RELAYD_CLOSE_STREAM:
3231                 DBG_CMD("RELAYD_CLOSE_STREAM", conn);
3232                 ret = relay_close_stream(header, conn, payload);
3233                 break;
3234         case RELAYD_DATA_PENDING:
3235                 DBG_CMD("RELAYD_DATA_PENDING", conn);
3236                 ret = relay_data_pending(header, conn, payload);
3237                 break;
3238         case RELAYD_QUIESCENT_CONTROL:
3239                 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
3240                 ret = relay_quiescent_control(header, conn, payload);
3241                 break;
3242         case RELAYD_BEGIN_DATA_PENDING:
3243                 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
3244                 ret = relay_begin_data_pending(header, conn, payload);
3245                 break;
3246         case RELAYD_END_DATA_PENDING:
3247                 DBG_CMD("RELAYD_END_DATA_PENDING", conn);
3248                 ret = relay_end_data_pending(header, conn, payload);
3249                 break;
3250         case RELAYD_SEND_INDEX:
3251                 DBG_CMD("RELAYD_SEND_INDEX", conn);
3252                 ret = relay_recv_index(header, conn, payload);
3253                 break;
3254         case RELAYD_STREAMS_SENT:
3255                 DBG_CMD("RELAYD_STREAMS_SENT", conn);
3256                 ret = relay_streams_sent(header, conn, payload);
3257                 break;
3258         case RELAYD_RESET_METADATA:
3259                 DBG_CMD("RELAYD_RESET_METADATA", conn);
3260                 ret = relay_reset_metadata(header, conn, payload);
3261                 break;
3262         case RELAYD_ROTATE_STREAMS:
3263                 DBG_CMD("RELAYD_ROTATE_STREAMS", conn);
3264                 ret = relay_rotate_session_streams(header, conn, payload);
3265                 break;
3266         case RELAYD_CREATE_TRACE_CHUNK:
3267                 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn);
3268                 ret = relay_create_trace_chunk(header, conn, payload);
3269                 break;
3270         case RELAYD_CLOSE_TRACE_CHUNK:
3271                 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn);
3272                 ret = relay_close_trace_chunk(header, conn, payload);
3273                 break;
3274         case RELAYD_TRACE_CHUNK_EXISTS:
3275                 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn);
3276                 ret = relay_trace_chunk_exists(header, conn, payload);
3277                 break;
3278         case RELAYD_GET_CONFIGURATION:
3279                 DBG_CMD("RELAYD_GET_CONFIGURATION", conn);
3280                 ret = relay_get_configuration(header, conn, payload);
3281                 break;
3282         case RELAYD_UPDATE_SYNC_INFO:
3283         default:
3284                 ERR("Received unknown command (%u)", header->cmd);
3285                 relay_unknown_command(conn);
3286                 ret = -1;
3287                 goto end;
3288         }
3289
3290 end:
3291         return ret;
3292 }
3293
3294 static enum relay_connection_status relay_process_control_receive_payload(
3295                 struct relay_connection *conn)
3296 {
3297         int ret = 0;
3298         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3299         struct lttng_dynamic_buffer *reception_buffer =
3300                         &conn->protocol.ctrl.reception_buffer;
3301         struct ctrl_connection_state_receive_payload *state =
3302                         &conn->protocol.ctrl.state.receive_payload;
3303         struct lttng_buffer_view payload_view;
3304
3305         if (state->left_to_receive == 0) {
3306                 /* Short-circuit for payload-less commands. */
3307                 goto reception_complete;
3308         }
3309
3310         ret = conn->sock->ops->recvmsg(conn->sock,
3311                         reception_buffer->data + state->received,
3312                         state->left_to_receive, MSG_DONTWAIT);
3313         if (ret < 0) {
3314                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3315                         PERROR("Unable to receive command payload on sock %d",
3316                                         conn->sock->fd);
3317                         status = RELAY_CONNECTION_STATUS_ERROR;
3318                 }
3319                 goto end;
3320         } else if (ret == 0) {
3321                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3322                 status = RELAY_CONNECTION_STATUS_CLOSED;
3323                 goto end;
3324         }
3325
3326         assert(ret > 0);
3327         assert(ret <= state->left_to_receive);
3328
3329         state->left_to_receive -= ret;
3330         state->received += ret;
3331
3332         if (state->left_to_receive > 0) {
3333                 /*
3334                  * Can't transition to the protocol's next state, wait to
3335                  * receive the rest of the header.
3336                  */
3337                 DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3338                                 state->received, state->left_to_receive,
3339                                 conn->sock->fd);
3340                 goto end;
3341         }
3342
3343 reception_complete:
3344         DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3345                         conn->sock->fd, state->received);
3346         /*
3347          * The payload required to process the command has been received.
3348          * A view to the reception buffer is forwarded to the various
3349          * commands and the state of the control is reset on success.
3350          *
3351          * Commands are responsible for sending their reply to the peer.
3352          */
3353         payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
3354                         0, -1);
3355         ret = relay_process_control_command(conn,
3356                         &state->header, &payload_view);
3357         if (ret < 0) {
3358                 status = RELAY_CONNECTION_STATUS_ERROR;
3359                 goto end;
3360         }
3361
3362         ret = connection_reset_protocol_state(conn);
3363         if (ret) {
3364                 status = RELAY_CONNECTION_STATUS_ERROR;
3365         }
3366 end:
3367         return status;
3368 }
3369
3370 static enum relay_connection_status relay_process_control_receive_header(
3371                 struct relay_connection *conn)
3372 {
3373         int ret = 0;
3374         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3375         struct lttcomm_relayd_hdr header;
3376         struct lttng_dynamic_buffer *reception_buffer =
3377                         &conn->protocol.ctrl.reception_buffer;
3378         struct ctrl_connection_state_receive_header *state =
3379                         &conn->protocol.ctrl.state.receive_header;
3380
3381         assert(state->left_to_receive != 0);
3382
3383         ret = conn->sock->ops->recvmsg(conn->sock,
3384                         reception_buffer->data + state->received,
3385                         state->left_to_receive, MSG_DONTWAIT);
3386         if (ret < 0) {
3387                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3388                         PERROR("Unable to receive control command header on sock %d",
3389                                         conn->sock->fd);
3390                         status = RELAY_CONNECTION_STATUS_ERROR;
3391                 }
3392                 goto end;
3393         } else if (ret == 0) {
3394                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3395                 status = RELAY_CONNECTION_STATUS_CLOSED;
3396                 goto end;
3397         }
3398
3399         assert(ret > 0);
3400         assert(ret <= state->left_to_receive);
3401
3402         state->left_to_receive -= ret;
3403         state->received += ret;
3404
3405         if (state->left_to_receive > 0) {
3406                 /*
3407                  * Can't transition to the protocol's next state, wait to
3408                  * receive the rest of the header.
3409                  */
3410                 DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3411                                 state->received, state->left_to_receive,
3412                                 conn->sock->fd);
3413                 goto end;
3414         }
3415
3416         /* Transition to next state: receiving the command's payload. */
3417         conn->protocol.ctrl.state_id =
3418                         CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3419         memcpy(&header, reception_buffer->data, sizeof(header));
3420         header.circuit_id = be64toh(header.circuit_id);
3421         header.data_size = be64toh(header.data_size);
3422         header.cmd = be32toh(header.cmd);
3423         header.cmd_version = be32toh(header.cmd_version);
3424         memcpy(&conn->protocol.ctrl.state.receive_payload.header,
3425                         &header, sizeof(header));
3426
3427         DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
3428                         conn->sock->fd, header.cmd, header.cmd_version,
3429                         header.data_size);
3430
3431         if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3432                 ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3433                                 header.data_size);
3434                 status = RELAY_CONNECTION_STATUS_ERROR;
3435                 goto end;
3436         }
3437
3438         conn->protocol.ctrl.state.receive_payload.left_to_receive =
3439                         header.data_size;
3440         conn->protocol.ctrl.state.receive_payload.received = 0;
3441         ret = lttng_dynamic_buffer_set_size(reception_buffer,
3442                         header.data_size);
3443         if (ret) {
3444                 status = RELAY_CONNECTION_STATUS_ERROR;
3445                 goto end;
3446         }
3447
3448         if (header.data_size == 0) {
3449                 /*
3450                  * Manually invoke the next state as the poll loop
3451                  * will not wake-up to allow us to proceed further.
3452                  */
3453                 status = relay_process_control_receive_payload(conn);
3454         }
3455 end:
3456         return status;
3457 }
3458
3459 /*
3460  * Process the commands received on the control socket
3461  */
3462 static enum relay_connection_status relay_process_control(
3463                 struct relay_connection *conn)
3464 {
3465         enum relay_connection_status status;
3466
3467         switch (conn->protocol.ctrl.state_id) {
3468         case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3469                 status = relay_process_control_receive_header(conn);
3470                 break;
3471         case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3472                 status = relay_process_control_receive_payload(conn);
3473                 break;
3474         default:
3475                 ERR("Unknown control connection protocol state encountered.");
3476                 abort();
3477         }
3478
3479         return status;
3480 }
3481
3482 static enum relay_connection_status relay_process_data_receive_header(
3483                 struct relay_connection *conn)
3484 {
3485         int ret;
3486         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3487         struct data_connection_state_receive_header *state =
3488                         &conn->protocol.data.state.receive_header;
3489         struct lttcomm_relayd_data_hdr header;
3490         struct relay_stream *stream;
3491
3492         assert(state->left_to_receive != 0);
3493
3494         ret = conn->sock->ops->recvmsg(conn->sock,
3495                         state->header_reception_buffer + state->received,
3496                         state->left_to_receive, MSG_DONTWAIT);
3497         if (ret < 0) {
3498                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3499                         PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3500                         status = RELAY_CONNECTION_STATUS_ERROR;
3501                 }
3502                 goto end;
3503         } else if (ret == 0) {
3504                 /* Orderly shutdown. Not necessary to print an error. */
3505                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3506                 status = RELAY_CONNECTION_STATUS_CLOSED;
3507                 goto end;
3508         }
3509
3510         assert(ret > 0);
3511         assert(ret <= state->left_to_receive);
3512
3513         state->left_to_receive -= ret;
3514         state->received += ret;
3515
3516         if (state->left_to_receive > 0) {
3517                 /*
3518                  * Can't transition to the protocol's next state, wait to
3519                  * receive the rest of the header.
3520                  */
3521                 DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3522                                 state->received, state->left_to_receive,
3523                                 conn->sock->fd);
3524                 goto end;
3525         }
3526
3527         /* Transition to next state: receiving the payload. */
3528         conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3529
3530         memcpy(&header, state->header_reception_buffer, sizeof(header));
3531         header.circuit_id = be64toh(header.circuit_id);
3532         header.stream_id = be64toh(header.stream_id);
3533         header.data_size = be32toh(header.data_size);
3534         header.net_seq_num = be64toh(header.net_seq_num);
3535         header.padding_size = be32toh(header.padding_size);
3536         memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3537
3538         conn->protocol.data.state.receive_payload.left_to_receive =
3539                         header.data_size;
3540         conn->protocol.data.state.receive_payload.received = 0;
3541         conn->protocol.data.state.receive_payload.rotate_index = false;
3542
3543         DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32,
3544                         conn->sock->fd, header.circuit_id,
3545                         header.stream_id, header.data_size,
3546                         header.net_seq_num, header.padding_size);
3547
3548         stream = stream_get_by_id(header.stream_id);
3549         if (!stream) {
3550                 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3551                                 header.stream_id);
3552                 /* Protocol error. */
3553                 status = RELAY_CONNECTION_STATUS_ERROR;
3554                 goto end;
3555         }
3556
3557         pthread_mutex_lock(&stream->lock);
3558         /* Prepare stream for the reception of a new packet. */
3559         ret = stream_init_packet(stream, header.data_size,
3560                         &conn->protocol.data.state.receive_payload.rotate_index);
3561         pthread_mutex_unlock(&stream->lock);
3562         if (ret) {
3563                 ERR("Failed to rotate stream output file");
3564                 status = RELAY_CONNECTION_STATUS_ERROR;
3565                 goto end_stream_unlock;
3566         }
3567
3568 end_stream_unlock:
3569         stream_put(stream);
3570 end:
3571         return status;
3572 }
3573
3574 static enum relay_connection_status relay_process_data_receive_payload(
3575                 struct relay_connection *conn)
3576 {
3577         int ret;
3578         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3579         struct relay_stream *stream;
3580         struct data_connection_state_receive_payload *state =
3581                         &conn->protocol.data.state.receive_payload;
3582         const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3583         char data_buffer[chunk_size];
3584         bool partial_recv = false;
3585         bool new_stream = false, close_requested = false, index_flushed = false;
3586         uint64_t left_to_receive = state->left_to_receive;
3587         struct relay_session *session;
3588
3589         DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
3590                         state->header.stream_id, state->header.net_seq_num,
3591                         state->received, left_to_receive);
3592
3593         stream = stream_get_by_id(state->header.stream_id);
3594         if (!stream) {
3595                 /* Protocol error. */
3596                 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3597                                 state->header.stream_id);
3598                 status = RELAY_CONNECTION_STATUS_ERROR;
3599                 goto end;
3600         }
3601
3602         pthread_mutex_lock(&stream->lock);
3603         session = stream->trace->session;
3604         if (!conn->session) {
3605                 ret = connection_set_session(conn, session);
3606                 if (ret) {
3607                         status = RELAY_CONNECTION_STATUS_ERROR;
3608                         goto end_stream_unlock;
3609                 }
3610         }
3611
3612         /*
3613          * The size of the "chunk" received on any iteration is bounded by:
3614          *   - the data left to receive,
3615          *   - the data immediately available on the socket,
3616          *   - the on-stack data buffer
3617          */
3618         while (left_to_receive > 0 && !partial_recv) {
3619                 size_t recv_size = min(left_to_receive, chunk_size);
3620                 struct lttng_buffer_view packet_chunk;
3621
3622                 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer,
3623                                 recv_size, MSG_DONTWAIT);
3624                 if (ret < 0) {
3625                         if (errno != EAGAIN && errno != EWOULDBLOCK) {
3626                                 PERROR("Socket %d error", conn->sock->fd);
3627                                 status = RELAY_CONNECTION_STATUS_ERROR;
3628                         }
3629                         goto end_stream_unlock;
3630                 } else if (ret == 0) {
3631                         /* No more data ready to be consumed on socket. */
3632                         DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3633                                         state->header.stream_id);
3634                         status = RELAY_CONNECTION_STATUS_CLOSED;
3635                         break;
3636                 } else if (ret < (int) recv_size) {
3637                         /*
3638                          * All the data available on the socket has been
3639                          * consumed.
3640                          */
3641                         partial_recv = true;
3642                         recv_size = ret;
3643                 }
3644
3645                 packet_chunk = lttng_buffer_view_init(data_buffer,
3646                                 0, recv_size);
3647                 assert(packet_chunk.data);
3648
3649                 ret = stream_write(stream, &packet_chunk, 0);
3650                 if (ret) {
3651                         ERR("Relay error writing data to file");
3652                         status = RELAY_CONNECTION_STATUS_ERROR;
3653                         goto end_stream_unlock;
3654                 }
3655
3656                 left_to_receive -= recv_size;
3657                 state->received += recv_size;
3658                 state->left_to_receive = left_to_receive;
3659         }
3660
3661         if (state->left_to_receive > 0) {
3662                 /*
3663                  * Did not receive all the data expected, wait for more data to
3664                  * become available on the socket.
3665                  */
3666                 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
3667                                 state->header.stream_id, state->received,
3668                                 state->left_to_receive);
3669                 goto end_stream_unlock;
3670         }
3671
3672         ret = stream_write(stream, NULL, state->header.padding_size);
3673         if (ret) {
3674                 status = RELAY_CONNECTION_STATUS_ERROR;
3675                 goto end_stream_unlock;
3676         }
3677
3678         if (session_streams_have_index(session)) {
3679                 ret = stream_update_index(stream, state->header.net_seq_num,
3680                                 state->rotate_index, &index_flushed,
3681                                 state->header.data_size + state->header.padding_size);
3682                 if (ret < 0) {
3683                         ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
3684                                         stream->stream_handle,
3685                                         state->header.net_seq_num, ret);
3686                         status = RELAY_CONNECTION_STATUS_ERROR;
3687                         goto end_stream_unlock;
3688                 }
3689         }
3690
3691         if (stream->prev_data_seq == -1ULL) {
3692                 new_stream = true;
3693         }
3694
3695         ret = stream_complete_packet(stream, state->header.data_size +
3696                         state->header.padding_size, state->header.net_seq_num,
3697                         index_flushed);
3698         if (ret) {
3699                 status = RELAY_CONNECTION_STATUS_ERROR;
3700                 goto end_stream_unlock;
3701         }
3702
3703         /*
3704          * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3705          * contents of *state which are aliased (union) to the same location as
3706          * the new state. Don't use it beyond this point.
3707          */
3708         connection_reset_protocol_state(conn);
3709         state = NULL;
3710
3711 end_stream_unlock:
3712         close_requested = stream->close_requested;
3713         pthread_mutex_unlock(&stream->lock);
3714         if (close_requested && left_to_receive == 0) {
3715                 try_stream_close(stream);
3716         }
3717
3718         if (new_stream) {
3719                 pthread_mutex_lock(&session->lock);
3720                 uatomic_set(&session->new_streams, 1);
3721                 pthread_mutex_unlock(&session->lock);
3722         }
3723
3724         stream_put(stream);
3725 end:
3726         return status;
3727 }
3728
3729 /*
3730  * relay_process_data: Process the data received on the data socket
3731  */
3732 static enum relay_connection_status relay_process_data(
3733                 struct relay_connection *conn)
3734 {
3735         enum relay_connection_status status;
3736
3737         switch (conn->protocol.data.state_id) {
3738         case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3739                 status = relay_process_data_receive_header(conn);
3740                 break;
3741         case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3742                 status = relay_process_data_receive_payload(conn);
3743                 break;
3744         default:
3745                 ERR("Unexpected data connection communication state.");
3746                 abort();
3747         }
3748
3749         return status;
3750 }
3751
3752 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3753 {
3754         int ret;
3755
3756         (void) lttng_poll_del(events, pollfd);
3757
3758         ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
3759                         fd_tracker_util_close_fd, NULL);
3760         if (ret < 0) {
3761                 ERR("Closing pollfd %d", pollfd);
3762         }
3763 }
3764
3765 static void relay_thread_close_connection(struct lttng_poll_event *events,
3766                 int pollfd, struct relay_connection *conn)
3767 {
3768         const char *type_str;
3769
3770         switch (conn->type) {
3771         case RELAY_DATA:
3772                 type_str = "Data";
3773                 break;
3774         case RELAY_CONTROL:
3775                 type_str = "Control";
3776                 break;
3777         case RELAY_VIEWER_COMMAND:
3778                 type_str = "Viewer Command";
3779                 break;
3780         case RELAY_VIEWER_NOTIFICATION:
3781                 type_str = "Viewer Notification";
3782                 break;
3783         default:
3784                 type_str = "Unknown";
3785         }
3786         cleanup_connection_pollfd(events, pollfd);
3787         connection_put(conn);
3788         DBG("%s connection closed with %d", type_str, pollfd);
3789 }
3790
3791 /*
3792  * This thread does the actual work
3793  */
3794 static void *relay_thread_worker(void *data)
3795 {
3796         int ret, err = -1, last_seen_data_fd = -1;
3797         uint32_t nb_fd;
3798         struct lttng_poll_event events;
3799         struct lttng_ht *relay_connections_ht;
3800         struct lttng_ht_iter iter;
3801         struct relay_connection *destroy_conn = NULL;
3802
3803         DBG("[thread] Relay worker started");
3804
3805         rcu_register_thread();
3806
3807         health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3808
3809         if (testpoint(relayd_thread_worker)) {
3810                 goto error_testpoint;
3811         }
3812
3813         health_code_update();
3814
3815         /* table of connections indexed on socket */
3816         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3817         if (!relay_connections_ht) {
3818                 goto relay_connections_ht_error;
3819         }
3820
3821         ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3822         if (ret < 0) {
3823                 goto error_poll_create;
3824         }
3825
3826         ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3827         if (ret < 0) {
3828                 goto error;
3829         }
3830
3831 restart:
3832         while (1) {
3833                 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3834
3835                 health_code_update();
3836
3837                 /* Infinite blocking call, waiting for transmission */
3838                 DBG3("Relayd worker thread polling...");
3839                 health_poll_entry();
3840                 ret = lttng_poll_wait(&events, -1);
3841                 health_poll_exit();
3842                 if (ret < 0) {
3843                         /*
3844                          * Restart interrupted system call.
3845                          */
3846                         if (errno == EINTR) {
3847                                 goto restart;
3848                         }
3849                         goto error;
3850                 }
3851
3852                 nb_fd = ret;
3853
3854                 /*
3855                  * Process control. The control connection is
3856                  * prioritized so we don't starve it with high
3857                  * throughput tracing data on the data connection.
3858                  */
3859                 for (i = 0; i < nb_fd; i++) {
3860                         /* Fetch once the poll data */
3861                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3862                         int pollfd = LTTNG_POLL_GETFD(&events, i);
3863
3864                         health_code_update();
3865
3866                         /* Thread quit pipe has been closed. Killing thread. */
3867                         ret = check_thread_quit_pipe(pollfd, revents);
3868                         if (ret) {
3869                                 err = 0;
3870                                 goto exit;
3871                         }
3872
3873                         /* Inspect the relay conn pipe for new connection */
3874                         if (pollfd == relay_conn_pipe[0]) {
3875                                 if (revents & LPOLLIN) {
3876                                         struct relay_connection *conn;
3877
3878                                         ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
3879                                         if (ret < 0) {
3880                                                 goto error;
3881                                         }
3882                                         ret = lttng_poll_add(&events,
3883                                                         conn->sock->fd,
3884                                                         LPOLLIN | LPOLLRDHUP);
3885                                         if (ret) {
3886                                                 ERR("Failed to add new connection file descriptor to poll set");
3887                                                 goto error;
3888                                         }
3889                                         connection_ht_add(relay_connections_ht, conn);
3890                                         DBG("Connection socket %d added", conn->sock->fd);
3891                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3892                                         ERR("Relay connection pipe error");
3893                                         goto error;
3894                                 } else {
3895                                         ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3896                                         goto error;
3897                                 }
3898                         } else {
3899                                 struct relay_connection *ctrl_conn;
3900
3901                                 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
3902                                 /* If not found, there is a synchronization issue. */
3903                                 assert(ctrl_conn);
3904
3905                                 if (ctrl_conn->type == RELAY_DATA) {
3906                                         if (revents & LPOLLIN) {
3907                                                 /*
3908                                                  * Flag the last seen data fd not deleted. It will be
3909                                                  * used as the last seen fd if any fd gets deleted in
3910                                                  * this first loop.
3911                                                  */
3912                                                 last_notdel_data_fd = pollfd;
3913                                         }
3914                                         goto put_ctrl_connection;
3915                                 }
3916                                 assert(ctrl_conn->type == RELAY_CONTROL);
3917
3918                                 if (revents & LPOLLIN) {
3919                                         enum relay_connection_status status;
3920
3921                                         status = relay_process_control(ctrl_conn);
3922                                         if (status != RELAY_CONNECTION_STATUS_OK) {
3923                                                 /*
3924                                                  * On socket error flag the session as aborted to force
3925                                                  * the cleanup of its stream otherwise it can leak
3926                                                  * during the lifetime of the relayd.
3927                                                  *
3928                                                  * This prevents situations in which streams can be
3929                                                  * left opened because an index was received, the
3930                                                  * control connection is closed, and the data
3931                                                  * connection is closed (uncleanly) before the packet's
3932                                                  * data provided.
3933                                                  *
3934                                                  * Since the control connection encountered an error,
3935                                                  * it is okay to be conservative and close the
3936                                                  * session right now as we can't rely on the protocol
3937                                                  * being respected anymore.
3938                                                  */
3939                                                 if (status == RELAY_CONNECTION_STATUS_ERROR) {
3940                                                         session_abort(ctrl_conn->session);
3941                                                 }
3942
3943                                                 /* Clear the connection on error or close. */
3944                                                 relay_thread_close_connection(&events,
3945                                                                 pollfd,
3946                                                                 ctrl_conn);
3947                                         }
3948                                         seen_control = 1;
3949                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3950                                         relay_thread_close_connection(&events,
3951                                                         pollfd, ctrl_conn);
3952                                         if (last_seen_data_fd == pollfd) {
3953                                                 last_seen_data_fd = last_notdel_data_fd;
3954                                         }
3955                                 } else {
3956                                         ERR("Unexpected poll events %u for control sock %d",
3957                                                         revents, pollfd);
3958                                         connection_put(ctrl_conn);
3959                                         goto error;
3960                                 }
3961                         put_ctrl_connection:
3962                                 connection_put(ctrl_conn);
3963                         }
3964                 }
3965
3966                 /*
3967                  * The last loop handled a control request, go back to poll to make
3968                  * sure we prioritise the control socket.
3969                  */
3970                 if (seen_control) {
3971                         continue;
3972                 }
3973
3974                 if (last_seen_data_fd >= 0) {
3975                         for (i = 0; i < nb_fd; i++) {
3976                                 int pollfd = LTTNG_POLL_GETFD(&events, i);
3977
3978                                 health_code_update();
3979
3980                                 if (last_seen_data_fd == pollfd) {
3981                                         idx = i;
3982                                         break;
3983                                 }
3984                         }
3985                 }
3986
3987                 /* Process data connection. */
3988                 for (i = idx + 1; i < nb_fd; i++) {
3989                         /* Fetch the poll data. */
3990                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3991                         int pollfd = LTTNG_POLL_GETFD(&events, i);
3992                         struct relay_connection *data_conn;
3993
3994                         health_code_update();
3995
3996                         if (!revents) {
3997                                 /* No activity for this FD (poll implementation). */
3998                                 continue;
3999                         }
4000
4001                         /* Skip the command pipe. It's handled in the first loop. */
4002                         if (pollfd == relay_conn_pipe[0]) {
4003                                 continue;
4004                         }
4005
4006                         data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4007                         if (!data_conn) {
4008                                 /* Skip it. Might be removed before. */
4009                                 continue;
4010                         }
4011                         if (data_conn->type == RELAY_CONTROL) {
4012                                 goto put_data_connection;
4013                         }
4014                         assert(data_conn->type == RELAY_DATA);
4015
4016                         if (revents & LPOLLIN) {
4017                                 enum relay_connection_status status;
4018
4019                                 status = relay_process_data(data_conn);
4020                                 /* Connection closed or error. */
4021                                 if (status != RELAY_CONNECTION_STATUS_OK) {
4022                                         /*
4023                                          * On socket error flag the session as aborted to force
4024                                          * the cleanup of its stream otherwise it can leak
4025                                          * during the lifetime of the relayd.
4026                                          *
4027                                          * This prevents situations in which streams can be
4028                                          * left opened because an index was received, the
4029                                          * control connection is closed, and the data
4030                                          * connection is closed (uncleanly) before the packet's
4031                                          * data provided.
4032                                          *
4033                                          * Since the data connection encountered an error,
4034                                          * it is okay to be conservative and close the
4035                                          * session right now as we can't rely on the protocol
4036                                          * being respected anymore.
4037                                          */
4038                                         if (status == RELAY_CONNECTION_STATUS_ERROR) {
4039                                                 session_abort(data_conn->session);
4040                                         }
4041                                         relay_thread_close_connection(&events, pollfd,
4042                                                         data_conn);
4043                                         /*
4044                                          * Every goto restart call sets the last seen fd where
4045                                          * here we don't really care since we gracefully
4046                                          * continue the loop after the connection is deleted.
4047                                          */
4048                                 } else {
4049                                         /* Keep last seen port. */
4050                                         last_seen_data_fd = pollfd;
4051                                         connection_put(data_conn);
4052                                         goto restart;
4053                                 }
4054                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4055                                 relay_thread_close_connection(&events, pollfd,
4056                                                 data_conn);
4057                         } else {
4058                                 ERR("Unknown poll events %u for data sock %d",
4059                                                 revents, pollfd);
4060                         }
4061                 put_data_connection:
4062                         connection_put(data_conn);
4063                 }
4064                 last_seen_data_fd = -1;
4065         }
4066
4067         /* Normal exit, no error */
4068         ret = 0;
4069
4070 exit:
4071 error:
4072         /* Cleanup remaining connection object. */
4073         rcu_read_lock();
4074         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
4075                         destroy_conn,
4076                         sock_n.node) {
4077                 health_code_update();
4078
4079                 session_abort(destroy_conn->session);
4080
4081                 /*
4082                  * No need to grab another ref, because we own
4083                  * destroy_conn.
4084                  */
4085                 relay_thread_close_connection(&events, destroy_conn->sock->fd,
4086                                 destroy_conn);
4087         }
4088         rcu_read_unlock();
4089
4090         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4091 error_poll_create:
4092         lttng_ht_destroy(relay_connections_ht);
4093 relay_connections_ht_error:
4094         /* Close relay conn pipes */
4095         (void) fd_tracker_util_pipe_close(the_fd_tracker,
4096                         relay_conn_pipe);
4097         if (err) {
4098                 DBG("Thread exited with error");
4099         }
4100         DBG("Worker thread cleanup complete");
4101 error_testpoint:
4102         if (err) {
4103                 health_error();
4104                 ERR("Health error occurred in %s", __func__);
4105         }
4106         health_unregister(health_relayd);
4107         rcu_unregister_thread();
4108         lttng_relay_stop_threads();
4109         return NULL;
4110 }
4111
4112 /*
4113  * Create the relay command pipe to wake thread_manage_apps.
4114  * Closed in cleanup().
4115  */
4116 static int create_relay_conn_pipe(void)
4117 {
4118         return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
4119                         "Relayd connection pipe", relay_conn_pipe);
4120 }
4121
4122 static int stdio_open(void *data, int *fds)
4123 {
4124         fds[0] = fileno(stdout);
4125         fds[1] = fileno(stderr);
4126         return 0;
4127 }
4128
4129 static int track_stdio(void)
4130 {
4131         int fds[2];
4132         const char *names[] = { "stdout", "stderr" };
4133
4134         return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds,
4135                         names, 2, stdio_open, NULL);
4136 }
4137
4138 /*
4139  * main
4140  */
4141 int main(int argc, char **argv)
4142 {
4143         bool thread_is_rcu_registered = false;
4144         int ret = 0, retval = 0;
4145         void *status;
4146         char *unlinked_file_directory_path = NULL, *output_path = NULL;
4147
4148         /* Parse environment variables */
4149         ret = parse_env_options();
4150         if (ret) {
4151                 retval = -1;
4152                 goto exit_options;
4153         }
4154
4155         /*
4156          * Parse arguments.
4157          * Command line arguments overwrite environment.
4158          */
4159         progname = argv[0];
4160         if (set_options(argc, argv)) {
4161                 retval = -1;
4162                 goto exit_options;
4163         }
4164
4165         if (set_signal_handler()) {
4166                 retval = -1;
4167                 goto exit_options;
4168         }
4169
4170         relayd_config_log();
4171
4172         if (opt_print_version) {
4173                 print_version();
4174                 retval = 0;
4175                 goto exit_options;
4176         }
4177
4178         ret = fclose(stdin);
4179         if (ret) {
4180                 PERROR("Failed to close stdin");
4181                 goto exit_options;
4182         }
4183
4184         DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4185
4186         /* Try to create directory if -o, --output is specified. */
4187         if (opt_output_path) {
4188                 if (*opt_output_path != '/') {
4189                         ERR("Please specify an absolute path for -o, --output PATH");
4190                         retval = -1;
4191                         goto exit_options;
4192                 }
4193
4194                 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
4195                                 -1, -1);
4196                 if (ret < 0) {
4197                         ERR("Unable to create %s", opt_output_path);
4198                         retval = -1;
4199                         goto exit_options;
4200                 }
4201         }
4202
4203         /* Daemonize */
4204         if (opt_daemon || opt_background) {
4205                 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
4206                         !opt_background);
4207                 if (ret < 0) {
4208                         retval = -1;
4209                         goto exit_options;
4210                 }
4211         }
4212
4213         if (opt_working_directory) {
4214                 ret = utils_change_working_directory(opt_working_directory);
4215                 if (ret) {
4216                         /* All errors are already logged. */
4217                         goto exit_options;
4218                 }
4219         }
4220
4221         sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4222         if (!sessiond_trace_chunk_registry) {
4223                 ERR("Failed to initialize session daemon trace chunk registry");
4224                 retval = -1;
4225                 goto exit_options;
4226         }
4227
4228         /*
4229          * The RCU thread registration (and use, through the fd-tracker's
4230          * creation) is done after the daemonization to allow us to not
4231          * deal with liburcu's fork() management as the call RCU needs to
4232          * be restored.
4233          */
4234         rcu_register_thread();
4235         thread_is_rcu_registered = true;
4236
4237         output_path = create_output_path("");
4238         if (!output_path) {
4239                 ERR("Failed to get output path");
4240                 retval = -1;
4241                 goto exit_options;
4242         }
4243         ret = asprintf(&unlinked_file_directory_path, "%s/%s", output_path,
4244                         DEFAULT_UNLINKED_FILES_DIRECTORY);
4245         free(output_path);
4246         if (ret < 0) {
4247                 ERR("Failed to format unlinked file directory path");
4248                 retval = -1;
4249                 goto exit_options;
4250         }
4251         the_fd_tracker = fd_tracker_create(
4252                         unlinked_file_directory_path, lttng_opt_fd_pool_size);
4253         free(unlinked_file_directory_path);
4254         if (!the_fd_tracker) {
4255                 retval = -1;
4256                 goto exit_options;
4257         }
4258
4259         ret = track_stdio();
4260         if (ret) {
4261                 retval = -1;
4262                 goto exit_options;
4263         }
4264
4265         /* Initialize thread health monitoring */
4266         health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4267         if (!health_relayd) {
4268                 PERROR("health_app_create error");
4269                 retval = -1;
4270                 goto exit_options;
4271         }
4272
4273         /* Create thread quit pipe */
4274         if (init_thread_quit_pipe()) {
4275                 retval = -1;
4276                 goto exit_options;
4277         }
4278
4279         /* Setup the thread apps communication pipe. */
4280         if (create_relay_conn_pipe()) {
4281                 retval = -1;
4282                 goto exit_options;
4283         }
4284
4285         /* Init relay command queue. */
4286         cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4287
4288         /* Initialize communication library */
4289         lttcomm_init();
4290         lttcomm_inet_init();
4291
4292         /* tables of sessions indexed by session ID */
4293         sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4294         if (!sessions_ht) {
4295                 retval = -1;
4296                 goto exit_options;
4297         }
4298
4299         /* tables of streams indexed by stream ID */
4300         relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4301         if (!relay_streams_ht) {
4302                 retval = -1;
4303                 goto exit_options;
4304         }
4305
4306         /* tables of streams indexed by stream ID */
4307         viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4308         if (!viewer_streams_ht) {
4309                 retval = -1;
4310                 goto exit_options;
4311         }
4312
4313         ret = init_health_quit_pipe();
4314         if (ret) {
4315                 retval = -1;
4316                 goto exit_options;
4317         }
4318
4319         /* Create thread to manage the client socket */
4320         ret = pthread_create(&health_thread, default_pthread_attr(),
4321                         thread_manage_health, (void *) NULL);
4322         if (ret) {
4323                 errno = ret;
4324                 PERROR("pthread_create health");
4325                 retval = -1;
4326                 goto exit_options;
4327         }
4328
4329         /* Setup the dispatcher thread */
4330         ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
4331                         relay_thread_dispatcher, (void *) NULL);
4332         if (ret) {
4333                 errno = ret;
4334                 PERROR("pthread_create dispatcher");
4335                 retval = -1;
4336                 goto exit_dispatcher_thread;
4337         }
4338
4339         /* Setup the worker thread */
4340         ret = pthread_create(&worker_thread, default_pthread_attr(),
4341                         relay_thread_worker, NULL);
4342         if (ret) {
4343                 errno = ret;
4344                 PERROR("pthread_create worker");
4345                 retval = -1;
4346                 goto exit_worker_thread;
4347         }
4348
4349         /* Setup the listener thread */
4350         ret = pthread_create(&listener_thread, default_pthread_attr(),
4351                         relay_thread_listener, (void *) NULL);
4352         if (ret) {
4353                 errno = ret;
4354                 PERROR("pthread_create listener");
4355                 retval = -1;
4356                 goto exit_listener_thread;
4357         }
4358
4359         ret = relayd_live_create(live_uri);
4360         if (ret) {
4361                 ERR("Starting live viewer threads");
4362                 retval = -1;
4363                 goto exit_live;
4364         }
4365
4366         /*
4367          * This is where we start awaiting program completion (e.g. through
4368          * signal that asks threads to teardown).
4369          */
4370
4371         ret = relayd_live_join();
4372         if (ret) {
4373                 retval = -1;
4374         }
4375 exit_live:
4376
4377         ret = pthread_join(listener_thread, &status);
4378         if (ret) {
4379                 errno = ret;
4380                 PERROR("pthread_join listener_thread");
4381                 retval = -1;
4382         }
4383
4384 exit_listener_thread:
4385         ret = pthread_join(worker_thread, &status);
4386         if (ret) {
4387                 errno = ret;
4388                 PERROR("pthread_join worker_thread");
4389                 retval = -1;
4390         }
4391
4392 exit_worker_thread:
4393         ret = pthread_join(dispatcher_thread, &status);
4394         if (ret) {
4395                 errno = ret;
4396                 PERROR("pthread_join dispatcher_thread");
4397                 retval = -1;
4398         }
4399 exit_dispatcher_thread:
4400
4401         ret = pthread_join(health_thread, &status);
4402         if (ret) {
4403                 errno = ret;
4404                 PERROR("pthread_join health_thread");
4405                 retval = -1;
4406         }
4407 exit_options:
4408         /*
4409          * Wait for all pending call_rcu work to complete before tearing
4410          * down data structures. call_rcu worker may be trying to
4411          * perform lookups in those structures.
4412          */
4413         rcu_barrier();
4414         relayd_cleanup();
4415
4416         /* Ensure all prior call_rcu are done. */
4417         rcu_barrier();
4418
4419         if (thread_is_rcu_registered) {
4420                 rcu_unregister_thread();
4421         }
4422
4423         if (!retval) {
4424                 exit(EXIT_SUCCESS);
4425         } else {
4426                 exit(EXIT_FAILURE);
4427         }
4428 }