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((unsigned char) 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 accommodates 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         if (ret) {
1026                 PERROR("Failed to open \"%s\" relay socket",
1027                                 formated_name ?: "Unknown");
1028                 goto error;
1029         }
1030         DBG("Listening on %s socket %d", name, sock->fd);
1031
1032         ret = sock->ops->bind(sock);
1033         if (ret < 0) {
1034                 PERROR("Failed to bind socket");
1035                 goto error;
1036         }
1037
1038         ret = sock->ops->listen(sock, -1);
1039         if (ret < 0) {
1040                 goto error;
1041
1042         }
1043
1044         free(formated_name);
1045         return sock;
1046
1047 error:
1048         if (sock) {
1049                 lttcomm_destroy_sock(sock);
1050         }
1051         free(formated_name);
1052         return NULL;
1053 }
1054
1055 static
1056 struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1057                 const char *name)
1058 {
1059         int out_fd, ret;
1060         struct lttcomm_sock *socks[2] = { listening_sock, NULL };
1061         struct lttcomm_sock *new_sock = NULL;
1062
1063         ret = fd_tracker_open_unsuspendable_fd(
1064                         the_fd_tracker, &out_fd,
1065                         (const char **) &name,
1066                         1, accept_sock, &socks);
1067         if (ret) {
1068                 goto end;
1069         }
1070         new_sock = socks[1];
1071         DBG("%s accepted, socket %d", name, new_sock->fd);
1072 end:
1073         return new_sock;
1074 }
1075
1076 /*
1077  * This thread manages the listening for new connections on the network
1078  */
1079 static void *relay_thread_listener(void *data)
1080 {
1081         int i, ret, pollfd, err = -1;
1082         uint32_t revents, nb_fd;
1083         struct lttng_poll_event events;
1084         struct lttcomm_sock *control_sock, *data_sock;
1085
1086         DBG("[thread] Relay listener started");
1087
1088         rcu_register_thread();
1089         health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1090
1091         health_code_update();
1092
1093         control_sock = relay_socket_create(control_uri, "Control listener");
1094         if (!control_sock) {
1095                 goto error_sock_control;
1096         }
1097
1098         data_sock = relay_socket_create(data_uri, "Data listener");
1099         if (!data_sock) {
1100                 goto error_sock_relay;
1101         }
1102
1103         /*
1104          * Pass 3 as size here for the thread quit pipe, control and
1105          * data socket.
1106          */
1107         ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1108         if (ret < 0) {
1109                 goto error_create_poll;
1110         }
1111
1112         /* Add the control socket */
1113         ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1114         if (ret < 0) {
1115                 goto error_poll_add;
1116         }
1117
1118         /* Add the data socket */
1119         ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1120         if (ret < 0) {
1121                 goto error_poll_add;
1122         }
1123
1124         lttng_relay_notify_ready();
1125
1126         if (testpoint(relayd_thread_listener)) {
1127                 goto error_testpoint;
1128         }
1129
1130         while (1) {
1131                 health_code_update();
1132
1133                 DBG("Listener accepting connections");
1134
1135 restart:
1136                 health_poll_entry();
1137                 ret = lttng_poll_wait(&events, -1);
1138                 health_poll_exit();
1139                 if (ret < 0) {
1140                         /*
1141                          * Restart interrupted system call.
1142                          */
1143                         if (errno == EINTR) {
1144                                 goto restart;
1145                         }
1146                         goto error;
1147                 }
1148
1149                 nb_fd = ret;
1150
1151                 DBG("Relay new connection received");
1152                 for (i = 0; i < nb_fd; i++) {
1153                         health_code_update();
1154
1155                         /* Fetch once the poll data */
1156                         revents = LTTNG_POLL_GETEV(&events, i);
1157                         pollfd = LTTNG_POLL_GETFD(&events, i);
1158
1159                         /* Thread quit pipe has been closed. Killing thread. */
1160                         ret = check_thread_quit_pipe(pollfd, revents);
1161                         if (ret) {
1162                                 err = 0;
1163                                 goto exit;
1164                         }
1165
1166                         if (revents & LPOLLIN) {
1167                                 /*
1168                                  * A new connection is requested, therefore a
1169                                  * sessiond/consumerd connection is allocated in
1170                                  * this thread, enqueued to a global queue and
1171                                  * dequeued (and freed) in the worker thread.
1172                                  */
1173                                 int val = 1;
1174                                 struct relay_connection *new_conn;
1175                                 struct lttcomm_sock *newsock = NULL;
1176                                 enum connection_type type;
1177
1178                                 if (pollfd == data_sock->fd) {
1179                                         type = RELAY_DATA;
1180                                         newsock = accept_relayd_sock(data_sock,
1181                                                         "Data socket to relayd");
1182                                 } else {
1183                                         assert(pollfd == control_sock->fd);
1184                                         type = RELAY_CONTROL;
1185                                         newsock = accept_relayd_sock(control_sock,
1186                                                         "Control socket to relayd");
1187                                 }
1188                                 if (!newsock) {
1189                                         PERROR("accepting sock");
1190                                         goto error;
1191                                 }
1192
1193                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
1194                                                 sizeof(val));
1195                                 if (ret < 0) {
1196                                         PERROR("setsockopt inet");
1197                                         lttcomm_destroy_sock(newsock);
1198                                         goto error;
1199                                 }
1200
1201                                 ret = socket_apply_keep_alive_config(newsock->fd);
1202                                 if (ret < 0) {
1203                                         ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1204                                                         newsock->fd);
1205                                         lttcomm_destroy_sock(newsock);
1206                                         goto error;
1207                                 }
1208
1209                                 new_conn = connection_create(newsock, type);
1210                                 if (!new_conn) {
1211                                         lttcomm_destroy_sock(newsock);
1212                                         goto error;
1213                                 }
1214
1215                                 /* Enqueue request for the dispatcher thread. */
1216                                 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
1217                                                  &new_conn->qnode);
1218
1219                                 /*
1220                                  * Wake the dispatch queue futex.
1221                                  * Implicit memory barrier with the
1222                                  * exchange in cds_wfcq_enqueue.
1223                                  */
1224                                 futex_nto1_wake(&relay_conn_queue.futex);
1225                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1226                                 ERR("socket poll error");
1227                                 goto error;
1228                         } else {
1229                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1230                                 goto error;
1231                         }
1232                 }
1233         }
1234
1235 exit:
1236 error:
1237 error_poll_add:
1238 error_testpoint:
1239         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1240 error_create_poll:
1241         if (data_sock->fd >= 0) {
1242                 int data_sock_fd = data_sock->fd;
1243
1244                 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1245                                 &data_sock_fd, 1, close_sock,
1246                                 data_sock);
1247                 if (ret) {
1248                         PERROR("Failed to close the data listener socket file descriptor");
1249                 }
1250                 data_sock->fd = -1;
1251         }
1252         lttcomm_destroy_sock(data_sock);
1253 error_sock_relay:
1254         if (control_sock->fd >= 0) {
1255                 int control_sock_fd = control_sock->fd;
1256
1257                 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1258                                 &control_sock_fd, 1, close_sock,
1259                                 control_sock);
1260                 if (ret) {
1261                         PERROR("Failed to close the control listener socket file descriptor");
1262                 }
1263                 control_sock->fd = -1;
1264         }
1265         lttcomm_destroy_sock(control_sock);
1266 error_sock_control:
1267         if (err) {
1268                 health_error();
1269                 ERR("Health error occurred in %s", __func__);
1270         }
1271         health_unregister(health_relayd);
1272         rcu_unregister_thread();
1273         DBG("Relay listener thread cleanup complete");
1274         lttng_relay_stop_threads();
1275         return NULL;
1276 }
1277
1278 /*
1279  * This thread manages the dispatching of the requests to worker threads
1280  */
1281 static void *relay_thread_dispatcher(void *data)
1282 {
1283         int err = -1;
1284         ssize_t ret;
1285         struct cds_wfcq_node *node;
1286         struct relay_connection *new_conn = NULL;
1287
1288         DBG("[thread] Relay dispatcher started");
1289
1290         health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1291
1292         if (testpoint(relayd_thread_dispatcher)) {
1293                 goto error_testpoint;
1294         }
1295
1296         health_code_update();
1297
1298         for (;;) {
1299                 health_code_update();
1300
1301                 /* Atomically prepare the queue futex */
1302                 futex_nto1_prepare(&relay_conn_queue.futex);
1303
1304                 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1305                         break;
1306                 }
1307
1308                 do {
1309                         health_code_update();
1310
1311                         /* Dequeue commands */
1312                         node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1313                                                          &relay_conn_queue.tail);
1314                         if (node == NULL) {
1315                                 DBG("Woken up but nothing in the relay command queue");
1316                                 /* Continue thread execution */
1317                                 break;
1318                         }
1319                         new_conn = caa_container_of(node, struct relay_connection, qnode);
1320
1321                         DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1322
1323                         /*
1324                          * Inform worker thread of the new request. This
1325                          * call is blocking so we can be assured that
1326                          * the data will be read at some point in time
1327                          * or wait to the end of the world :)
1328                          */
1329                         ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1330                         if (ret < 0) {
1331                                 PERROR("write connection pipe");
1332                                 connection_put(new_conn);
1333                                 goto error;
1334                         }
1335                 } while (node != NULL);
1336
1337                 /* Futex wait on queue. Blocking call on futex() */
1338                 health_poll_entry();
1339                 futex_nto1_wait(&relay_conn_queue.futex);
1340                 health_poll_exit();
1341         }
1342
1343         /* Normal exit, no error */
1344         err = 0;
1345
1346 error:
1347 error_testpoint:
1348         if (err) {
1349                 health_error();
1350                 ERR("Health error occurred in %s", __func__);
1351         }
1352         health_unregister(health_relayd);
1353         DBG("Dispatch thread dying");
1354         lttng_relay_stop_threads();
1355         return NULL;
1356 }
1357
1358 static bool session_streams_have_index(const struct relay_session *session)
1359 {
1360         return session->minor >= 4 && !session->snapshot;
1361 }
1362
1363 /*
1364  * Handle the RELAYD_CREATE_SESSION command.
1365  *
1366  * On success, send back the session id or else return a negative value.
1367  */
1368 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr,
1369                 struct relay_connection *conn,
1370                 const struct lttng_buffer_view *payload)
1371 {
1372         int ret = 0;
1373         ssize_t send_ret;
1374         struct relay_session *session = NULL;
1375         struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1376         char session_name[LTTNG_NAME_MAX] = {};
1377         char hostname[LTTNG_HOST_NAME_MAX] = {};
1378         uint32_t live_timer = 0;
1379         bool snapshot = false;
1380         bool session_name_contains_creation_timestamp = false;
1381         /* Left nil for peers < 2.11. */
1382         char base_path[LTTNG_PATH_MAX] = {};
1383         lttng_uuid sessiond_uuid = {};
1384         LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1385         LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1386         LTTNG_OPTIONAL(time_t) creation_time = {};
1387         struct lttng_dynamic_buffer reply_payload;
1388
1389         lttng_dynamic_buffer_init(&reply_payload);
1390
1391         if (conn->minor < 4) {
1392                 /* From 2.1 to 2.3 */
1393                 ret = 0;
1394         } else if (conn->minor >= 4 && conn->minor < 11) {
1395                 /* From 2.4 to 2.10 */
1396                 ret = cmd_create_session_2_4(payload, session_name,
1397                         hostname, &live_timer, &snapshot);
1398         } else {
1399                 bool has_current_chunk;
1400                 uint64_t current_chunk_id_value;
1401                 time_t creation_time_value;
1402                 uint64_t id_sessiond_value;
1403
1404                 /* From 2.11 to ... */
1405                 ret = cmd_create_session_2_11(payload, session_name, hostname,
1406                                 base_path, &live_timer, &snapshot, &id_sessiond_value,
1407                                 sessiond_uuid, &has_current_chunk,
1408                                 &current_chunk_id_value, &creation_time_value,
1409                                 &session_name_contains_creation_timestamp);
1410                 if (lttng_uuid_is_nil(sessiond_uuid)) {
1411                         /* The nil UUID is reserved for pre-2.11 clients. */
1412                         ERR("Illegal nil UUID announced by peer in create session command");
1413                         ret = -1;
1414                         goto send_reply;
1415                 }
1416                 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1417                 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1418                 if (has_current_chunk) {
1419                         LTTNG_OPTIONAL_SET(&current_chunk_id,
1420                                         current_chunk_id_value);
1421                 }
1422         }
1423
1424         if (ret < 0) {
1425                 goto send_reply;
1426         }
1427
1428         session = session_create(session_name, hostname, base_path, live_timer,
1429                         snapshot, sessiond_uuid,
1430                         id_sessiond.is_set ? &id_sessiond.value : NULL,
1431                         current_chunk_id.is_set ? &current_chunk_id.value : NULL,
1432                         creation_time.is_set ? &creation_time.value : NULL,
1433                         conn->major, conn->minor,
1434                         session_name_contains_creation_timestamp);
1435         if (!session) {
1436                 ret = -1;
1437                 goto send_reply;
1438         }
1439         assert(!conn->session);
1440         conn->session = session;
1441         DBG("Created session %" PRIu64, session->id);
1442
1443         reply.generic.session_id = htobe64(session->id);
1444
1445 send_reply:
1446         if (ret < 0) {
1447                 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1448         } else {
1449                 reply.generic.ret_code = htobe32(LTTNG_OK);
1450         }
1451
1452         if (conn->minor < 11) {
1453                 /* From 2.1 to 2.10 */
1454                 ret = lttng_dynamic_buffer_append(&reply_payload,
1455                                 &reply.generic, sizeof(reply.generic));
1456                 if (ret) {
1457                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1458                         ret = -1;
1459                         goto end;
1460                 }
1461         } else {
1462                 const uint32_t output_path_length =
1463                                 session ? strlen(session->output_path) + 1 : 0;
1464
1465                 reply.output_path_length = htobe32(output_path_length);
1466                 ret = lttng_dynamic_buffer_append(
1467                                 &reply_payload, &reply, sizeof(reply));
1468                 if (ret) {
1469                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1470                         goto end;
1471                 }
1472
1473                 if (output_path_length) {
1474                         ret = lttng_dynamic_buffer_append(&reply_payload,
1475                                         session->output_path,
1476                                         output_path_length);
1477                         if (ret) {
1478                                 ERR("Failed to append \"create session\" command reply path to payload buffer");
1479                                 goto end;
1480                         }
1481                 }
1482         }
1483
1484         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data,
1485                         reply_payload.size, 0);
1486         if (send_ret < (ssize_t) reply_payload.size) {
1487                 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1488                                 reply_payload.size, send_ret);
1489                 ret = -1;
1490         }
1491 end:
1492         if (ret < 0 && session) {
1493                 session_put(session);
1494         }
1495         lttng_dynamic_buffer_reset(&reply_payload);
1496         return ret;
1497 }
1498
1499 /*
1500  * When we have received all the streams and the metadata for a channel,
1501  * we make them visible to the viewer threads.
1502  */
1503 static void publish_connection_local_streams(struct relay_connection *conn)
1504 {
1505         struct relay_stream *stream;
1506         struct relay_session *session = conn->session;
1507
1508         /*
1509          * We publish all streams belonging to a session atomically wrt
1510          * session lock.
1511          */
1512         pthread_mutex_lock(&session->lock);
1513         rcu_read_lock();
1514         cds_list_for_each_entry_rcu(stream, &session->recv_list,
1515                         recv_node) {
1516                 stream_publish(stream);
1517         }
1518         rcu_read_unlock();
1519
1520         /*
1521          * Inform the viewer that there are new streams in the session.
1522          */
1523         if (session->viewer_attached) {
1524                 uatomic_set(&session->new_streams, 1);
1525         }
1526         pthread_mutex_unlock(&session->lock);
1527 }
1528
1529 static int conform_channel_path(char *channel_path)
1530 {
1531         int ret = 0;
1532
1533         if (strstr("../", channel_path)) {
1534                 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1535                                 channel_path);
1536                 ret = -1;
1537                 goto end;
1538         }
1539
1540         if (*channel_path == '/') {
1541                 const size_t len = strlen(channel_path);
1542
1543                 /*
1544                  * Channel paths from peers prior to 2.11 are expressed as an
1545                  * absolute path that is, in reality, relative to the relay
1546                  * daemon's output directory. Remove the leading slash so it
1547                  * is correctly interpreted as a relative path later on.
1548                  *
1549                  * len (and not len - 1) is used to copy the trailing NULL.
1550                  */
1551                 bcopy(channel_path + 1, channel_path, len);
1552         }
1553 end:
1554         return ret;
1555 }
1556
1557 /*
1558  * relay_add_stream: allocate a new stream for a session
1559  */
1560 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1561                 struct relay_connection *conn,
1562                 const struct lttng_buffer_view *payload)
1563 {
1564         int ret;
1565         ssize_t send_ret;
1566         struct relay_session *session = conn->session;
1567         struct relay_stream *stream = NULL;
1568         struct lttcomm_relayd_status_stream reply;
1569         struct ctf_trace *trace = NULL;
1570         uint64_t stream_handle = -1ULL;
1571         char *path_name = NULL, *channel_name = NULL;
1572         uint64_t tracefile_size = 0, tracefile_count = 0;
1573         LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1574
1575         if (!session || !conn->version_check_done) {
1576                 ERR("Trying to add a stream before version check");
1577                 ret = -1;
1578                 goto end_no_session;
1579         }
1580
1581         if (session->minor == 1) {
1582                 /* For 2.1 */
1583                 ret = cmd_recv_stream_2_1(payload, &path_name,
1584                         &channel_name);
1585         } else if (session->minor > 1 && session->minor < 11) {
1586                 /* From 2.2 to 2.10 */
1587                 ret = cmd_recv_stream_2_2(payload, &path_name,
1588                         &channel_name, &tracefile_size, &tracefile_count);
1589         } else {
1590                 /* From 2.11 to ... */
1591                 ret = cmd_recv_stream_2_11(payload, &path_name,
1592                         &channel_name, &tracefile_size, &tracefile_count,
1593                         &stream_chunk_id.value);
1594                 stream_chunk_id.is_set = true;
1595         }
1596
1597         if (ret < 0) {
1598                 goto send_reply;
1599         }
1600
1601         if (conform_channel_path(path_name)) {
1602                 goto send_reply;
1603         }
1604
1605         /*
1606          * Backward compatibility for --group-output-by-session.
1607          * Prior to lttng 2.11, the complete path is passed by the stream.
1608          * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1609          * >=2.11 the chunk is responsible for the output path. When dealing
1610          * with producer < 2.11 the chunk output_path is the root output path
1611          * and the stream carries the complete path (path_name).
1612          * To support --group-output-by-session with older producer (<2.11), we
1613          * need to craft the path based on the stream path.
1614          */
1615         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1616                 if (conn->minor < 4) {
1617                         /*
1618                          * From 2.1 to 2.3, the session_name is not passed on
1619                          * the RELAYD_CREATE_SESSION command. The session name
1620                          * is necessary to detect the presence of a base_path
1621                          * inside the stream path. Without it we cannot perform
1622                          * a valid group-output-by-session transformation.
1623                          */
1624                         WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1625                              " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1626                                         session->id, path_name);
1627                 } else if (conn->minor >= 4 && conn->minor < 11) {
1628                         char *group_by_session_path_name;
1629
1630                         assert(session->session_name[0] != '\0');
1631
1632                         group_by_session_path_name =
1633                                         backward_compat_group_by_session(
1634                                                         path_name,
1635                                                         session->session_name,
1636                                                         session->creation_time.value);
1637                         if (!group_by_session_path_name) {
1638                                 ERR("Failed to apply group by session to stream of session %" PRIu64,
1639                                                 session->id);
1640                                 goto send_reply;
1641                         }
1642
1643                         DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1644                                         path_name, group_by_session_path_name);
1645
1646                         free(path_name);
1647                         path_name = group_by_session_path_name;
1648                 }
1649         }
1650
1651         trace = ctf_trace_get_by_path_or_create(session, path_name);
1652         if (!trace) {
1653                 goto send_reply;
1654         }
1655
1656         /* This stream here has one reference on the trace. */
1657         pthread_mutex_lock(&last_relay_stream_id_lock);
1658         stream_handle = ++last_relay_stream_id;
1659         pthread_mutex_unlock(&last_relay_stream_id_lock);
1660
1661         /* We pass ownership of path_name and channel_name. */
1662         stream = stream_create(trace, stream_handle, path_name,
1663                 channel_name, tracefile_size, tracefile_count);
1664         path_name = NULL;
1665         channel_name = NULL;
1666
1667         /*
1668          * Streams are the owners of their trace. Reference to trace is
1669          * kept within stream_create().
1670          */
1671         ctf_trace_put(trace);
1672
1673 send_reply:
1674         memset(&reply, 0, sizeof(reply));
1675         reply.handle = htobe64(stream_handle);
1676         if (!stream) {
1677                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1678         } else {
1679                 reply.ret_code = htobe32(LTTNG_OK);
1680         }
1681
1682         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1683                         sizeof(struct lttcomm_relayd_status_stream), 0);
1684         if (send_ret < (ssize_t) sizeof(reply)) {
1685                 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1686                                 send_ret);
1687                 ret = -1;
1688         }
1689
1690 end_no_session:
1691         free(path_name);
1692         free(channel_name);
1693         return ret;
1694 }
1695
1696 /*
1697  * relay_close_stream: close a specific stream
1698  */
1699 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1700                 struct relay_connection *conn,
1701                 const struct lttng_buffer_view *payload)
1702 {
1703         int ret;
1704         ssize_t send_ret;
1705         struct relay_session *session = conn->session;
1706         struct lttcomm_relayd_close_stream stream_info;
1707         struct lttcomm_relayd_generic_reply reply;
1708         struct relay_stream *stream;
1709
1710         DBG("Close stream received");
1711
1712         if (!session || !conn->version_check_done) {
1713                 ERR("Trying to close a stream before version check");
1714                 ret = -1;
1715                 goto end_no_session;
1716         }
1717
1718         if (payload->size < sizeof(stream_info)) {
1719                 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1720                                 sizeof(stream_info), payload->size);
1721                 ret = -1;
1722                 goto end_no_session;
1723         }
1724         memcpy(&stream_info, payload->data, sizeof(stream_info));
1725         stream_info.stream_id = be64toh(stream_info.stream_id);
1726         stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1727
1728         stream = stream_get_by_id(stream_info.stream_id);
1729         if (!stream) {
1730                 ret = -1;
1731                 goto end;
1732         }
1733
1734         /*
1735          * Set last_net_seq_num before the close flag. Required by data
1736          * pending check.
1737          */
1738         pthread_mutex_lock(&stream->lock);
1739         stream->last_net_seq_num = stream_info.last_net_seq_num;
1740         pthread_mutex_unlock(&stream->lock);
1741
1742         /*
1743          * This is one of the conditions which may trigger a stream close
1744          * with the others being:
1745          *     1) A close command is received for a stream
1746          *     2) The control connection owning the stream is closed
1747          *     3) We have received all of the stream's data _after_ a close
1748          *        request.
1749          */
1750         try_stream_close(stream);
1751         if (stream->is_metadata) {
1752                 struct relay_viewer_stream *vstream;
1753
1754                 vstream = viewer_stream_get_by_id(stream->stream_handle);
1755                 if (vstream) {
1756                         if (stream->no_new_metadata_notified) {
1757                                 /*
1758                                  * Since all the metadata has been sent to the
1759                                  * viewer and that we have a request to close
1760                                  * its stream, we can safely teardown the
1761                                  * corresponding metadata viewer stream.
1762                                  */
1763                                 viewer_stream_put(vstream);
1764                         }
1765                         /* Put local reference. */
1766                         viewer_stream_put(vstream);
1767                 }
1768         }
1769         stream_put(stream);
1770         ret = 0;
1771
1772 end:
1773         memset(&reply, 0, sizeof(reply));
1774         if (ret < 0) {
1775                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1776         } else {
1777                 reply.ret_code = htobe32(LTTNG_OK);
1778         }
1779         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1780                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1781         if (send_ret < (ssize_t) sizeof(reply)) {
1782                 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1783                                 send_ret);
1784                 ret = -1;
1785         }
1786
1787 end_no_session:
1788         return ret;
1789 }
1790
1791 /*
1792  * relay_reset_metadata: reset a metadata stream
1793  */
1794 static
1795 int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1796                 struct relay_connection *conn,
1797                 const struct lttng_buffer_view *payload)
1798 {
1799         int ret;
1800         ssize_t send_ret;
1801         struct relay_session *session = conn->session;
1802         struct lttcomm_relayd_reset_metadata stream_info;
1803         struct lttcomm_relayd_generic_reply reply;
1804         struct relay_stream *stream;
1805
1806         DBG("Reset metadata received");
1807
1808         if (!session || !conn->version_check_done) {
1809                 ERR("Trying to reset a metadata stream before version check");
1810                 ret = -1;
1811                 goto end_no_session;
1812         }
1813
1814         if (payload->size < sizeof(stream_info)) {
1815                 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1816                                 sizeof(stream_info), payload->size);
1817                 ret = -1;
1818                 goto end_no_session;
1819         }
1820         memcpy(&stream_info, payload->data, sizeof(stream_info));
1821         stream_info.stream_id = be64toh(stream_info.stream_id);
1822         stream_info.version = be64toh(stream_info.version);
1823
1824         DBG("Update metadata to version %" PRIu64, stream_info.version);
1825
1826         /* Unsupported for live sessions for now. */
1827         if (session->live_timer != 0) {
1828                 ret = -1;
1829                 goto end;
1830         }
1831
1832         stream = stream_get_by_id(stream_info.stream_id);
1833         if (!stream) {
1834                 ret = -1;
1835                 goto end;
1836         }
1837         pthread_mutex_lock(&stream->lock);
1838         if (!stream->is_metadata) {
1839                 ret = -1;
1840                 goto end_unlock;
1841         }
1842
1843         ret = stream_reset_file(stream);
1844         if (ret < 0) {
1845                 ERR("Failed to reset metadata stream %" PRIu64
1846                                 ": stream_path = %s, channel = %s",
1847                                 stream->stream_handle, stream->path_name,
1848                                 stream->channel_name);
1849                 goto end_unlock;
1850         }
1851 end_unlock:
1852         pthread_mutex_unlock(&stream->lock);
1853         stream_put(stream);
1854
1855 end:
1856         memset(&reply, 0, sizeof(reply));
1857         if (ret < 0) {
1858                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1859         } else {
1860                 reply.ret_code = htobe32(LTTNG_OK);
1861         }
1862         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1863                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1864         if (send_ret < (ssize_t) sizeof(reply)) {
1865                 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1866                                 send_ret);
1867                 ret = -1;
1868         }
1869
1870 end_no_session:
1871         return ret;
1872 }
1873
1874 /*
1875  * relay_unknown_command: send -1 if received unknown command
1876  */
1877 static void relay_unknown_command(struct relay_connection *conn)
1878 {
1879         struct lttcomm_relayd_generic_reply reply;
1880         ssize_t send_ret;
1881
1882         memset(&reply, 0, sizeof(reply));
1883         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1884         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1885         if (send_ret < sizeof(reply)) {
1886                 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1887         }
1888 }
1889
1890 /*
1891  * relay_start: send an acknowledgment to the client to tell if we are
1892  * ready to receive data. We are ready if a session is established.
1893  */
1894 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr,
1895                 struct relay_connection *conn,
1896                 const struct lttng_buffer_view *payload)
1897 {
1898         int ret = 0;
1899         ssize_t send_ret;
1900         struct lttcomm_relayd_generic_reply reply;
1901         struct relay_session *session = conn->session;
1902
1903         if (!session) {
1904                 DBG("Trying to start the streaming without a session established");
1905                 ret = htobe32(LTTNG_ERR_UNK);
1906         }
1907
1908         memset(&reply, 0, sizeof(reply));
1909         reply.ret_code = htobe32(LTTNG_OK);
1910         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1911                         sizeof(reply), 0);
1912         if (send_ret < (ssize_t) sizeof(reply)) {
1913                 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1914                                 send_ret);
1915                 ret = -1;
1916         }
1917
1918         return ret;
1919 }
1920
1921 /*
1922  * relay_recv_metadata: receive the metadata for the session.
1923  */
1924 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1925                 struct relay_connection *conn,
1926                 const struct lttng_buffer_view *payload)
1927 {
1928         int ret = 0;
1929         struct relay_session *session = conn->session;
1930         struct lttcomm_relayd_metadata_payload metadata_payload_header;
1931         struct relay_stream *metadata_stream;
1932         uint64_t metadata_payload_size;
1933         struct lttng_buffer_view packet_view;
1934
1935         if (!session) {
1936                 ERR("Metadata sent before version check");
1937                 ret = -1;
1938                 goto end;
1939         }
1940
1941         if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1942                 ERR("Incorrect data size");
1943                 ret = -1;
1944                 goto end;
1945         }
1946         metadata_payload_size = recv_hdr->data_size -
1947                         sizeof(struct lttcomm_relayd_metadata_payload);
1948
1949         memcpy(&metadata_payload_header, payload->data,
1950                         sizeof(metadata_payload_header));
1951         metadata_payload_header.stream_id = be64toh(
1952                         metadata_payload_header.stream_id);
1953         metadata_payload_header.padding_size = be32toh(
1954                         metadata_payload_header.padding_size);
1955
1956         metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1957         if (!metadata_stream) {
1958                 ret = -1;
1959                 goto end;
1960         }
1961
1962         packet_view = lttng_buffer_view_from_view(payload,
1963                         sizeof(metadata_payload_header), metadata_payload_size);
1964         if (!lttng_buffer_view_is_valid(&packet_view)) {
1965                 ERR("Invalid metadata packet length announced by header");
1966                 ret = -1;
1967                 goto end_put;
1968         }
1969
1970         pthread_mutex_lock(&metadata_stream->lock);
1971         ret = stream_write(metadata_stream, &packet_view,
1972                         metadata_payload_header.padding_size);
1973         pthread_mutex_unlock(&metadata_stream->lock);
1974         if (ret){
1975                 ret = -1;
1976                 goto end_put;
1977         }
1978 end_put:
1979         stream_put(metadata_stream);
1980 end:
1981         return ret;
1982 }
1983
1984 /*
1985  * relay_send_version: send relayd version number
1986  */
1987 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
1988                 struct relay_connection *conn,
1989                 const struct lttng_buffer_view *payload)
1990 {
1991         int ret;
1992         ssize_t send_ret;
1993         struct lttcomm_relayd_version reply, msg;
1994         bool compatible = true;
1995
1996         conn->version_check_done = true;
1997
1998         /* Get version from the other side. */
1999         if (payload->size < sizeof(msg)) {
2000                 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
2001                                 sizeof(msg), payload->size);
2002                 ret = -1;
2003                 goto end;
2004         }
2005
2006         memcpy(&msg, payload->data, sizeof(msg));
2007         msg.major = be32toh(msg.major);
2008         msg.minor = be32toh(msg.minor);
2009
2010         memset(&reply, 0, sizeof(reply));
2011         reply.major = RELAYD_VERSION_COMM_MAJOR;
2012         reply.minor = RELAYD_VERSION_COMM_MINOR;
2013
2014         /* Major versions must be the same */
2015         if (reply.major != msg.major) {
2016                 DBG("Incompatible major versions (%u vs %u), deleting session",
2017                                 reply.major, msg.major);
2018                 compatible = false;
2019         }
2020
2021         conn->major = reply.major;
2022         /* We adapt to the lowest compatible version */
2023         if (reply.minor <= msg.minor) {
2024                 conn->minor = reply.minor;
2025         } else {
2026                 conn->minor = msg.minor;
2027         }
2028
2029         reply.major = htobe32(reply.major);
2030         reply.minor = htobe32(reply.minor);
2031         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2032                         sizeof(reply), 0);
2033         if (send_ret < (ssize_t) sizeof(reply)) {
2034                 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2035                                 send_ret);
2036                 ret = -1;
2037                 goto end;
2038         } else {
2039                 ret = 0;
2040         }
2041
2042         if (!compatible) {
2043                 ret = -1;
2044                 goto end;
2045         }
2046
2047         DBG("Version check done using protocol %u.%u", conn->major,
2048                         conn->minor);
2049
2050 end:
2051         return ret;
2052 }
2053
2054 /*
2055  * Check for data pending for a given stream id from the session daemon.
2056  */
2057 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2058                 struct relay_connection *conn,
2059                 const struct lttng_buffer_view *payload)
2060 {
2061         struct relay_session *session = conn->session;
2062         struct lttcomm_relayd_data_pending msg;
2063         struct lttcomm_relayd_generic_reply reply;
2064         struct relay_stream *stream;
2065         ssize_t send_ret;
2066         int ret;
2067         uint64_t stream_seq;
2068
2069         DBG("Data pending command received");
2070
2071         if (!session || !conn->version_check_done) {
2072                 ERR("Trying to check for data before version check");
2073                 ret = -1;
2074                 goto end_no_session;
2075         }
2076
2077         if (payload->size < sizeof(msg)) {
2078                 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2079                                 sizeof(msg), payload->size);
2080                 ret = -1;
2081                 goto end_no_session;
2082         }
2083         memcpy(&msg, payload->data, sizeof(msg));
2084         msg.stream_id = be64toh(msg.stream_id);
2085         msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2086
2087         stream = stream_get_by_id(msg.stream_id);
2088         if (stream == NULL) {
2089                 ret = -1;
2090                 goto end;
2091         }
2092
2093         pthread_mutex_lock(&stream->lock);
2094
2095         if (session_streams_have_index(session)) {
2096                 /*
2097                  * Ensure that both the index and stream data have been
2098                  * flushed up to the requested point.
2099                  */
2100                 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2101         } else {
2102                 stream_seq = stream->prev_data_seq;
2103         }
2104         DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2105                         ", prev_index_seq %" PRIu64
2106                         ", and last_seq %" PRIu64, msg.stream_id,
2107                         stream->prev_data_seq, stream->prev_index_seq,
2108                         msg.last_net_seq_num);
2109
2110         /* Avoid wrapping issue */
2111         if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2112                 /* Data has in fact been written and is NOT pending */
2113                 ret = 0;
2114         } else {
2115                 /* Data still being streamed thus pending */
2116                 ret = 1;
2117         }
2118
2119         stream->data_pending_check_done = true;
2120         pthread_mutex_unlock(&stream->lock);
2121
2122         stream_put(stream);
2123 end:
2124
2125         memset(&reply, 0, sizeof(reply));
2126         reply.ret_code = htobe32(ret);
2127         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2128         if (send_ret < (ssize_t) sizeof(reply)) {
2129                 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2130                                 send_ret);
2131                 ret = -1;
2132         }
2133
2134 end_no_session:
2135         return ret;
2136 }
2137
2138 /*
2139  * Wait for the control socket to reach a quiescent state.
2140  *
2141  * Note that for now, when receiving this command from the session
2142  * daemon, this means that every subsequent commands or data received on
2143  * the control socket has been handled. So, this is why we simply return
2144  * OK here.
2145  */
2146 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
2147                 struct relay_connection *conn,
2148                 const struct lttng_buffer_view *payload)
2149 {
2150         int ret;
2151         ssize_t send_ret;
2152         struct relay_stream *stream;
2153         struct lttcomm_relayd_quiescent_control msg;
2154         struct lttcomm_relayd_generic_reply reply;
2155
2156         DBG("Checking quiescent state on control socket");
2157
2158         if (!conn->session || !conn->version_check_done) {
2159                 ERR("Trying to check for data before version check");
2160                 ret = -1;
2161                 goto end_no_session;
2162         }
2163
2164         if (payload->size < sizeof(msg)) {
2165                 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2166                                 sizeof(msg), payload->size);
2167                 ret = -1;
2168                 goto end_no_session;
2169         }
2170         memcpy(&msg, payload->data, sizeof(msg));
2171         msg.stream_id = be64toh(msg.stream_id);
2172
2173         stream = stream_get_by_id(msg.stream_id);
2174         if (!stream) {
2175                 goto reply;
2176         }
2177         pthread_mutex_lock(&stream->lock);
2178         stream->data_pending_check_done = true;
2179         pthread_mutex_unlock(&stream->lock);
2180
2181         DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2182         stream_put(stream);
2183 reply:
2184         memset(&reply, 0, sizeof(reply));
2185         reply.ret_code = htobe32(LTTNG_OK);
2186         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2187         if (send_ret < (ssize_t) sizeof(reply)) {
2188                 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2189                                 send_ret);
2190                 ret = -1;
2191         } else {
2192                 ret = 0;
2193         }
2194
2195 end_no_session:
2196         return ret;
2197 }
2198
2199 /*
2200  * Initialize a data pending command. This means that a consumer is about
2201  * to ask for data pending for each stream it holds. Simply iterate over
2202  * all streams of a session and set the data_pending_check_done flag.
2203  *
2204  * This command returns to the client a LTTNG_OK code.
2205  */
2206 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2207                 struct relay_connection *conn,
2208                 const struct lttng_buffer_view *payload)
2209 {
2210         int ret;
2211         ssize_t send_ret;
2212         struct lttng_ht_iter iter;
2213         struct lttcomm_relayd_begin_data_pending msg;
2214         struct lttcomm_relayd_generic_reply reply;
2215         struct relay_stream *stream;
2216
2217         assert(recv_hdr);
2218         assert(conn);
2219
2220         DBG("Init streams for data pending");
2221
2222         if (!conn->session || !conn->version_check_done) {
2223                 ERR("Trying to check for data before version check");
2224                 ret = -1;
2225                 goto end_no_session;
2226         }
2227
2228         if (payload->size < sizeof(msg)) {
2229                 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2230                                 sizeof(msg), payload->size);
2231                 ret = -1;
2232                 goto end_no_session;
2233         }
2234         memcpy(&msg, payload->data, sizeof(msg));
2235         msg.session_id = be64toh(msg.session_id);
2236
2237         /*
2238          * Iterate over all streams to set the begin data pending flag.
2239          * For now, the streams are indexed by stream handle so we have
2240          * to iterate over all streams to find the one associated with
2241          * the right session_id.
2242          */
2243         rcu_read_lock();
2244         cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2245                         node.node) {
2246                 if (!stream_get(stream)) {
2247                         continue;
2248                 }
2249                 if (stream->trace->session->id == msg.session_id) {
2250                         pthread_mutex_lock(&stream->lock);
2251                         stream->data_pending_check_done = false;
2252                         pthread_mutex_unlock(&stream->lock);
2253                         DBG("Set begin data pending flag to stream %" PRIu64,
2254                                         stream->stream_handle);
2255                 }
2256                 stream_put(stream);
2257         }
2258         rcu_read_unlock();
2259
2260         memset(&reply, 0, sizeof(reply));
2261         /* All good, send back reply. */
2262         reply.ret_code = htobe32(LTTNG_OK);
2263
2264         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2265         if (send_ret < (ssize_t) sizeof(reply)) {
2266                 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2267                         send_ret);
2268                 ret = -1;
2269         } else {
2270                 ret = 0;
2271         }
2272
2273 end_no_session:
2274         return ret;
2275 }
2276
2277 /*
2278  * End data pending command. This will check, for a given session id, if
2279  * each stream associated with it has its data_pending_check_done flag
2280  * set. If not, this means that the client lost track of the stream but
2281  * the data is still being streamed on our side. In this case, we inform
2282  * the client that data is in flight.
2283  *
2284  * Return to the client if there is data in flight or not with a ret_code.
2285  */
2286 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2287                 struct relay_connection *conn,
2288                 const struct lttng_buffer_view *payload)
2289 {
2290         int ret;
2291         ssize_t send_ret;
2292         struct lttng_ht_iter iter;
2293         struct lttcomm_relayd_end_data_pending msg;
2294         struct lttcomm_relayd_generic_reply reply;
2295         struct relay_stream *stream;
2296         uint32_t is_data_inflight = 0;
2297
2298         DBG("End data pending command");
2299
2300         if (!conn->session || !conn->version_check_done) {
2301                 ERR("Trying to check for data before version check");
2302                 ret = -1;
2303                 goto end_no_session;
2304         }
2305
2306         if (payload->size < sizeof(msg)) {
2307                 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2308                                 sizeof(msg), payload->size);
2309                 ret = -1;
2310                 goto end_no_session;
2311         }
2312         memcpy(&msg, payload->data, sizeof(msg));
2313         msg.session_id = be64toh(msg.session_id);
2314
2315         /*
2316          * Iterate over all streams to see if the begin data pending
2317          * flag is set.
2318          */
2319         rcu_read_lock();
2320         cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2321                         node.node) {
2322                 if (!stream_get(stream)) {
2323                         continue;
2324                 }
2325                 if (stream->trace->session->id != msg.session_id) {
2326                         stream_put(stream);
2327                         continue;
2328                 }
2329                 pthread_mutex_lock(&stream->lock);
2330                 if (!stream->data_pending_check_done) {
2331                         uint64_t stream_seq;
2332
2333                         if (session_streams_have_index(conn->session)) {
2334                                 /*
2335                                  * Ensure that both the index and stream data have been
2336                                  * flushed up to the requested point.
2337                                  */
2338                                 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2339                         } else {
2340                                 stream_seq = stream->prev_data_seq;
2341                         }
2342                         if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2343                                 is_data_inflight = 1;
2344                                 DBG("Data is still in flight for stream %" PRIu64,
2345                                                 stream->stream_handle);
2346                                 pthread_mutex_unlock(&stream->lock);
2347                                 stream_put(stream);
2348                                 break;
2349                         }
2350                 }
2351                 pthread_mutex_unlock(&stream->lock);
2352                 stream_put(stream);
2353         }
2354         rcu_read_unlock();
2355
2356         memset(&reply, 0, sizeof(reply));
2357         /* All good, send back reply. */
2358         reply.ret_code = htobe32(is_data_inflight);
2359
2360         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2361         if (send_ret < (ssize_t) sizeof(reply)) {
2362                 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2363                         send_ret);
2364                 ret = -1;
2365         } else {
2366                 ret = 0;
2367         }
2368
2369 end_no_session:
2370         return ret;
2371 }
2372
2373 /*
2374  * Receive an index for a specific stream.
2375  *
2376  * Return 0 on success else a negative value.
2377  */
2378 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
2379                 struct relay_connection *conn,
2380                 const struct lttng_buffer_view *payload)
2381 {
2382         int ret;
2383         ssize_t send_ret;
2384         struct relay_session *session = conn->session;
2385         struct lttcomm_relayd_index index_info;
2386         struct lttcomm_relayd_generic_reply reply;
2387         struct relay_stream *stream;
2388         size_t msg_len;
2389
2390         assert(conn);
2391
2392         DBG("Relay receiving index");
2393
2394         if (!session || !conn->version_check_done) {
2395                 ERR("Trying to close a stream before version check");
2396                 ret = -1;
2397                 goto end_no_session;
2398         }
2399
2400         msg_len = lttcomm_relayd_index_len(
2401                         lttng_to_index_major(conn->major, conn->minor),
2402                         lttng_to_index_minor(conn->major, conn->minor));
2403         if (payload->size < msg_len) {
2404                 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2405                                 msg_len, payload->size);
2406                 ret = -1;
2407                 goto end_no_session;
2408         }
2409         memcpy(&index_info, payload->data, msg_len);
2410         index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2411         index_info.net_seq_num = be64toh(index_info.net_seq_num);
2412         index_info.packet_size = be64toh(index_info.packet_size);
2413         index_info.content_size = be64toh(index_info.content_size);
2414         index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2415         index_info.timestamp_end = be64toh(index_info.timestamp_end);
2416         index_info.events_discarded = be64toh(index_info.events_discarded);
2417         index_info.stream_id = be64toh(index_info.stream_id);
2418
2419         if (conn->minor >= 8) {
2420                 index_info.stream_instance_id =
2421                                 be64toh(index_info.stream_instance_id);
2422                 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2423         } else {
2424                 index_info.stream_instance_id = -1ULL;
2425                 index_info.packet_seq_num = -1ULL;
2426         }
2427
2428         stream = stream_get_by_id(index_info.relay_stream_id);
2429         if (!stream) {
2430                 ERR("stream_get_by_id not found");
2431                 ret = -1;
2432                 goto end;
2433         }
2434
2435         pthread_mutex_lock(&stream->lock);
2436         ret = stream_add_index(stream, &index_info);
2437         pthread_mutex_unlock(&stream->lock);
2438         if (ret) {
2439                 goto end_stream_put;
2440         }
2441
2442 end_stream_put:
2443         stream_put(stream);
2444 end:
2445         memset(&reply, 0, sizeof(reply));
2446         if (ret < 0) {
2447                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2448         } else {
2449                 reply.ret_code = htobe32(LTTNG_OK);
2450         }
2451         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2452         if (send_ret < (ssize_t) sizeof(reply)) {
2453                 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2454                 ret = -1;
2455         }
2456
2457 end_no_session:
2458         return ret;
2459 }
2460
2461 /*
2462  * Receive the streams_sent message.
2463  *
2464  * Return 0 on success else a negative value.
2465  */
2466 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
2467                 struct relay_connection *conn,
2468                 const struct lttng_buffer_view *payload)
2469 {
2470         int ret;
2471         ssize_t send_ret;
2472         struct lttcomm_relayd_generic_reply reply;
2473
2474         assert(conn);
2475
2476         DBG("Relay receiving streams_sent");
2477
2478         if (!conn->session || !conn->version_check_done) {
2479                 ERR("Trying to close a stream before version check");
2480                 ret = -1;
2481                 goto end_no_session;
2482         }
2483
2484         /*
2485          * Publish every pending stream in the connection recv list which are
2486          * now ready to be used by the viewer.
2487          */
2488         publish_connection_local_streams(conn);
2489
2490         memset(&reply, 0, sizeof(reply));
2491         reply.ret_code = htobe32(LTTNG_OK);
2492         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2493         if (send_ret < (ssize_t) sizeof(reply)) {
2494                 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2495                         send_ret);
2496                 ret = -1;
2497         } else {
2498                 /* Success. */
2499                 ret = 0;
2500         }
2501
2502 end_no_session:
2503         return ret;
2504 }
2505
2506 /*
2507  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2508  * session rotation feature (not the tracefile rotation feature).
2509  */
2510 static int relay_rotate_session_streams(
2511                 const struct lttcomm_relayd_hdr *recv_hdr,
2512                 struct relay_connection *conn,
2513                 const struct lttng_buffer_view *payload)
2514 {
2515         int ret = 0;
2516         uint32_t i;
2517         ssize_t send_ret;
2518         enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2519         struct relay_session *session = conn->session;
2520         struct lttcomm_relayd_rotate_streams rotate_streams;
2521         struct lttcomm_relayd_generic_reply reply = {};
2522         struct relay_stream *stream = NULL;
2523         const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams);
2524         struct lttng_trace_chunk *next_trace_chunk = NULL;
2525         struct lttng_buffer_view stream_positions;
2526         char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2527         const char *chunk_id_str = "none";
2528
2529         if (!session || !conn->version_check_done) {
2530                 ERR("Trying to rotate a stream before version check");
2531                 ret = -1;
2532                 goto end_no_reply;
2533         }
2534
2535         if (session->major == 2 && session->minor < 11) {
2536                 ERR("Unsupported feature before 2.11");
2537                 ret = -1;
2538                 goto end_no_reply;
2539         }
2540
2541         if (payload->size < header_len) {
2542                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2543                                 header_len, payload->size);
2544                 ret = -1;
2545                 goto end_no_reply;
2546         }
2547
2548         memcpy(&rotate_streams, payload->data, header_len);
2549
2550         /* Convert header to host endianness. */
2551         rotate_streams = (typeof(rotate_streams)) {
2552                 .stream_count = be32toh(rotate_streams.stream_count),
2553                 .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) {
2554                         .is_set = !!rotate_streams.new_chunk_id.is_set,
2555                         .value = be64toh(rotate_streams.new_chunk_id.value),
2556                 }
2557         };
2558
2559         if (rotate_streams.new_chunk_id.is_set) {
2560                 /*
2561                  * Retrieve the trace chunk the stream must transition to. As
2562                  * per the protocol, this chunk should have been created
2563                  * before this command is received.
2564                  */
2565                 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2566                                 sessiond_trace_chunk_registry,
2567                                 session->sessiond_uuid, session->id,
2568                                 rotate_streams.new_chunk_id.value);
2569                 if (!next_trace_chunk) {
2570                         char uuid_str[LTTNG_UUID_STR_LEN];
2571
2572                         lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2573                         ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2574                                         ", trace_chunk_id = %" PRIu64,
2575                                         uuid_str, session->id,
2576                                         rotate_streams.new_chunk_id.value);
2577                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2578                         ret = -1;
2579                         goto end;
2580                 }
2581
2582                 ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64,
2583                                 rotate_streams.new_chunk_id.value);
2584                 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2585                         chunk_id_str = "formatting error";
2586                 } else {
2587                         chunk_id_str = chunk_id_buf;
2588                 }
2589         }
2590
2591         DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2592                         rotate_streams.stream_count, session->session_name,
2593                         chunk_id_str);
2594
2595         stream_positions = lttng_buffer_view_from_view(payload,
2596                         sizeof(rotate_streams), -1);
2597         if (!stream_positions.data ||
2598                         stream_positions.size <
2599                                         (rotate_streams.stream_count *
2600                                                         sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2601                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2602                 ret = -1;
2603                 goto end;
2604         }
2605
2606         for (i = 0; i < rotate_streams.stream_count; i++) {
2607                 struct lttcomm_relayd_stream_rotation_position *position_comm =
2608                                 &((typeof(position_comm)) stream_positions.data)[i];
2609                 const struct lttcomm_relayd_stream_rotation_position pos = {
2610                         .stream_id = be64toh(position_comm->stream_id),
2611                         .rotate_at_seq_num = be64toh(
2612                                         position_comm->rotate_at_seq_num),
2613                 };
2614
2615                 stream = stream_get_by_id(pos.stream_id);
2616                 if (!stream) {
2617                         reply_code = LTTNG_ERR_INVALID;
2618                         ret = -1;
2619                         goto end;
2620                 }
2621
2622                 pthread_mutex_lock(&stream->lock);
2623                 ret = stream_set_pending_rotation(stream, next_trace_chunk,
2624                                 pos.rotate_at_seq_num);
2625                 pthread_mutex_unlock(&stream->lock);
2626                 if (ret) {
2627                         reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2628                         goto end;
2629                 }
2630
2631                 stream_put(stream);
2632                 stream = NULL;
2633         }
2634
2635         reply_code = LTTNG_OK;
2636         ret = 0;
2637 end:
2638         if (stream) {
2639                 stream_put(stream);
2640         }
2641
2642         reply.ret_code = htobe32((uint32_t) reply_code);
2643         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2644                         sizeof(struct lttcomm_relayd_generic_reply), 0);
2645         if (send_ret < (ssize_t) sizeof(reply)) {
2646                 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2647                                 send_ret);
2648                 ret = -1;
2649         }
2650 end_no_reply:
2651         lttng_trace_chunk_put(next_trace_chunk);
2652         return ret;
2653 }
2654
2655
2656
2657 /*
2658  * relay_create_trace_chunk: create a new trace chunk
2659  */
2660 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2661                 struct relay_connection *conn,
2662                 const struct lttng_buffer_view *payload)
2663 {
2664         int ret = 0;
2665         ssize_t send_ret;
2666         struct relay_session *session = conn->session;
2667         struct lttcomm_relayd_create_trace_chunk *msg;
2668         struct lttcomm_relayd_generic_reply reply = {};
2669         struct lttng_buffer_view header_view;
2670         struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
2671         enum lttng_error_code reply_code = LTTNG_OK;
2672         enum lttng_trace_chunk_status chunk_status;
2673         const char *new_path;
2674
2675         if (!session || !conn->version_check_done) {
2676                 ERR("Trying to create a trace chunk before version check");
2677                 ret = -1;
2678                 goto end_no_reply;
2679         }
2680
2681         if (session->major == 2 && session->minor < 11) {
2682                 ERR("Chunk creation command is unsupported before 2.11");
2683                 ret = -1;
2684                 goto end_no_reply;
2685         }
2686
2687         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2688         if (!lttng_buffer_view_is_valid(&header_view)) {
2689                 ERR("Failed to receive payload of chunk creation command");
2690                 ret = -1;
2691                 goto end_no_reply;
2692         }
2693
2694         /* Convert to host endianness. */
2695         msg = (typeof(msg)) header_view.data;
2696         msg->chunk_id = be64toh(msg->chunk_id);
2697         msg->creation_timestamp = be64toh(msg->creation_timestamp);
2698         msg->override_name_length = be32toh(msg->override_name_length);
2699
2700         if (session->current_trace_chunk &&
2701                         !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2702                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2703                                         DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2704                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2705                         ERR("Failed to rename old chunk");
2706                         ret = -1;
2707                         reply_code = LTTNG_ERR_UNK;
2708                         goto end;
2709                 }
2710         }
2711         session->ongoing_rotation = true;
2712         if (!session->current_trace_chunk) {
2713                 if (!session->has_rotated) {
2714                         new_path = "";
2715                 } else {
2716                         new_path = NULL;
2717                 }
2718         } else {
2719                 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2720         }
2721         chunk = lttng_trace_chunk_create(
2722                         msg->chunk_id, msg->creation_timestamp, new_path);
2723         if (!chunk) {
2724                 ERR("Failed to create trace chunk in trace chunk creation command");
2725                 ret = -1;
2726                 reply_code = LTTNG_ERR_NOMEM;
2727                 goto end;
2728         }
2729         lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2730
2731         if (msg->override_name_length) {
2732                 const char *name;
2733                 const struct lttng_buffer_view chunk_name_view =
2734                                 lttng_buffer_view_from_view(payload,
2735                                                 sizeof(*msg),
2736                                                 msg->override_name_length);
2737
2738                 if (!lttng_buffer_view_is_valid(&chunk_name_view)) {
2739                         ERR("Invalid payload of chunk creation command (protocol error): buffer too short for expected name length");
2740                         ret = -1;
2741                         reply_code = LTTNG_ERR_INVALID;
2742                         goto end;
2743                 }
2744
2745                 name = chunk_name_view.data;
2746                 if (name[msg->override_name_length - 1]) {
2747                         ERR("Invalid payload of chunk creation command (protocol error): name is not null-terminated");
2748                         ret = -1;
2749                         reply_code = LTTNG_ERR_INVALID;
2750                         goto end;
2751                 }
2752
2753                 chunk_status = lttng_trace_chunk_override_name(
2754                                 chunk, chunk_name_view.data);
2755                 switch (chunk_status) {
2756                 case LTTNG_TRACE_CHUNK_STATUS_OK:
2757                         break;
2758                 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2759                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2760                         reply_code = LTTNG_ERR_INVALID;
2761                         ret = -1;
2762                         goto end;
2763                 default:
2764                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2765                         reply_code = LTTNG_ERR_UNK;
2766                         ret = -1;
2767                         goto end;
2768                 }
2769         }
2770
2771         chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2772         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2773                 reply_code = LTTNG_ERR_UNK;
2774                 ret = -1;
2775                 goto end;
2776         }
2777
2778         assert(conn->session->output_directory);
2779         chunk_status = lttng_trace_chunk_set_as_owner(chunk,
2780                         conn->session->output_directory);
2781         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2782                 reply_code = LTTNG_ERR_UNK;
2783                 ret = -1;
2784                 goto end;
2785         }
2786
2787         published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2788                         sessiond_trace_chunk_registry,
2789                         conn->session->sessiond_uuid,
2790                         conn->session->id,
2791                         chunk);
2792         if (!published_chunk) {
2793                 char uuid_str[LTTNG_UUID_STR_LEN];
2794
2795                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2796                 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
2797                                 uuid_str,
2798                                 conn->session->id,
2799                                 msg->chunk_id);
2800                 ret = -1;
2801                 reply_code = LTTNG_ERR_NOMEM;
2802                 goto end;
2803         }
2804
2805         pthread_mutex_lock(&conn->session->lock);
2806         if (conn->session->pending_closure_trace_chunk) {
2807                 /*
2808                  * Invalid; this means a second create_trace_chunk command was
2809                  * received before a close_trace_chunk.
2810                  */
2811                 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2812                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2813                 ret = -1;
2814                 goto end_unlock_session;
2815         }
2816         conn->session->pending_closure_trace_chunk =
2817                         conn->session->current_trace_chunk;
2818         conn->session->current_trace_chunk = published_chunk;
2819         published_chunk = NULL;
2820         if (!conn->session->pending_closure_trace_chunk) {
2821                 session->ongoing_rotation = false;
2822         }
2823 end_unlock_session:
2824         pthread_mutex_unlock(&conn->session->lock);
2825 end:
2826         reply.ret_code = htobe32((uint32_t) reply_code);
2827         send_ret = conn->sock->ops->sendmsg(conn->sock,
2828                         &reply,
2829                         sizeof(struct lttcomm_relayd_generic_reply),
2830                         0);
2831         if (send_ret < (ssize_t) sizeof(reply)) {
2832                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2833                                 send_ret);
2834                 ret = -1;
2835         }
2836 end_no_reply:
2837         lttng_trace_chunk_put(chunk);
2838         lttng_trace_chunk_put(published_chunk);
2839         return ret;
2840 }
2841
2842 /*
2843  * relay_close_trace_chunk: close a trace chunk
2844  */
2845 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2846                 struct relay_connection *conn,
2847                 const struct lttng_buffer_view *payload)
2848 {
2849         int ret = 0, buf_ret;
2850         ssize_t send_ret;
2851         struct relay_session *session = conn->session;
2852         struct lttcomm_relayd_close_trace_chunk *msg;
2853         struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2854         struct lttng_buffer_view header_view;
2855         struct lttng_trace_chunk *chunk = NULL;
2856         enum lttng_error_code reply_code = LTTNG_OK;
2857         enum lttng_trace_chunk_status chunk_status;
2858         uint64_t chunk_id;
2859         LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2860         time_t close_timestamp;
2861         char closed_trace_chunk_path[LTTNG_PATH_MAX];
2862         size_t path_length = 0;
2863         const char *chunk_name = NULL;
2864         struct lttng_dynamic_buffer reply_payload;
2865         const char *new_path;
2866
2867         lttng_dynamic_buffer_init(&reply_payload);
2868
2869         if (!session || !conn->version_check_done) {
2870                 ERR("Trying to close a trace chunk before version check");
2871                 ret = -1;
2872                 goto end_no_reply;
2873         }
2874
2875         if (session->major == 2 && session->minor < 11) {
2876                 ERR("Chunk close command is unsupported before 2.11");
2877                 ret = -1;
2878                 goto end_no_reply;
2879         }
2880
2881         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2882         if (!lttng_buffer_view_is_valid(&header_view)) {
2883                 ERR("Failed to receive payload of chunk close command");
2884                 ret = -1;
2885                 goto end_no_reply;
2886         }
2887
2888         /* Convert to host endianness. */
2889         msg = (typeof(msg)) header_view.data;
2890         chunk_id = be64toh(msg->chunk_id);
2891         close_timestamp = (time_t) be64toh(msg->close_timestamp);
2892         close_command = (typeof(close_command)){
2893                 .value = be32toh(msg->close_command.value),
2894                 .is_set = msg->close_command.is_set,
2895         };
2896
2897         chunk = sessiond_trace_chunk_registry_get_chunk(
2898                         sessiond_trace_chunk_registry,
2899                         conn->session->sessiond_uuid,
2900                         conn->session->id,
2901                         chunk_id);
2902         if (!chunk) {
2903                 char uuid_str[LTTNG_UUID_STR_LEN];
2904
2905                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2906                 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
2907                                 uuid_str,
2908                                 conn->session->id,
2909                                 msg->chunk_id);
2910                 ret = -1;
2911                 reply_code = LTTNG_ERR_NOMEM;
2912                 goto end;
2913         }
2914
2915         pthread_mutex_lock(&session->lock);
2916         if (close_command.is_set &&
2917                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
2918                 /*
2919                  * Clear command. It is a protocol error to ask for a
2920                  * clear on a relay which does not allow it. Querying
2921                  * the configuration allows figuring out whether
2922                  * clearing is allowed before doing the clear.
2923                  */
2924                 if (!opt_allow_clear) {
2925                         ret = -1;
2926                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2927                         goto end_unlock_session;
2928                 }
2929         }
2930         if (session->pending_closure_trace_chunk &&
2931                         session->pending_closure_trace_chunk != chunk) {
2932                 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
2933                                 session->session_name);
2934                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2935                 ret = -1;
2936                 goto end_unlock_session;
2937         }
2938
2939         if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
2940                         !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2941                 if (close_command.is_set &&
2942                                 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
2943                                 !session->has_rotated) {
2944                         /* New chunk stays in session output directory. */
2945                         new_path = "";
2946                 } else {
2947                         /* Use chunk name for new chunk. */
2948                         new_path = NULL;
2949                 }
2950                 /* Rename new chunk path. */
2951                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2952                                 new_path);
2953                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2954                         ret = -1;
2955                         goto end_unlock_session;
2956                 }
2957                 session->ongoing_rotation = false;
2958         }
2959         if ((!close_command.is_set ||
2960                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
2961                         !lttng_trace_chunk_get_name_overridden(chunk)) {
2962                 const char *old_path;
2963
2964                 if (!session->has_rotated) {
2965                         old_path = "";
2966                 } else {
2967                         old_path = NULL;
2968                 }
2969                 /* We need to move back the .tmp_old_chunk to its rightful place. */
2970                 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
2971                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2972                         ret = -1;
2973                         goto end_unlock_session;
2974                 }
2975         }
2976         chunk_status = lttng_trace_chunk_set_close_timestamp(
2977                         chunk, close_timestamp);
2978         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2979                 ERR("Failed to set trace chunk close timestamp");
2980                 ret = -1;
2981                 reply_code = LTTNG_ERR_UNK;
2982                 goto end_unlock_session;
2983         }
2984
2985         if (close_command.is_set) {
2986                 chunk_status = lttng_trace_chunk_set_close_command(
2987                                 chunk, close_command.value);
2988                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2989                         ret = -1;
2990                         reply_code = LTTNG_ERR_INVALID;
2991                         goto end_unlock_session;
2992                 }
2993         }
2994         chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL);
2995         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2996                 ERR("Failed to get chunk name");
2997                 ret = -1;
2998                 reply_code = LTTNG_ERR_UNK;
2999                 goto end_unlock_session;
3000         }
3001         if (!session->has_rotated && !session->snapshot) {
3002                 ret = lttng_strncpy(closed_trace_chunk_path,
3003                                 session->output_path,
3004                                 sizeof(closed_trace_chunk_path));
3005                 if (ret) {
3006                         ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3007                                         strlen(session->output_path),
3008                                         sizeof(closed_trace_chunk_path));
3009                         reply_code = LTTNG_ERR_NOMEM;
3010                         ret = -1;
3011                         goto end_unlock_session;
3012                 }
3013         } else {
3014                 if (session->snapshot) {
3015                         ret = snprintf(closed_trace_chunk_path,
3016                                         sizeof(closed_trace_chunk_path),
3017                                         "%s/%s", session->output_path,
3018                                         chunk_name);
3019                 } else {
3020                         ret = snprintf(closed_trace_chunk_path,
3021                                         sizeof(closed_trace_chunk_path),
3022                                         "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3023                                         "/%s",
3024                                         session->output_path, chunk_name);
3025                 }
3026                 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3027                         ERR("Failed to format closed trace chunk resulting path");
3028                         reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3029                         ret = -1;
3030                         goto end_unlock_session;
3031                 }
3032         }
3033         if (close_command.is_set &&
3034                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3035                 session->has_rotated = true;
3036         }
3037         DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3038         path_length = strlen(closed_trace_chunk_path) + 1;
3039         if (path_length > UINT32_MAX) {
3040                 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3041                 ret = -1;
3042                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3043                 goto end_unlock_session;
3044         }
3045
3046         if (session->current_trace_chunk == chunk) {
3047                 /*
3048                  * After a trace chunk close command, no new streams
3049                  * referencing the chunk may be created. Hence, on the
3050                  * event that no new trace chunk have been created for
3051                  * the session, the reference to the current trace chunk
3052                  * is released in order to allow it to be reclaimed when
3053                  * the last stream releases its reference to it.
3054                  */
3055                 lttng_trace_chunk_put(session->current_trace_chunk);
3056                 session->current_trace_chunk = NULL;
3057         }
3058         lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3059         session->pending_closure_trace_chunk = NULL;
3060 end_unlock_session:
3061         pthread_mutex_unlock(&session->lock);
3062
3063 end:
3064         reply.generic.ret_code = htobe32((uint32_t) reply_code);
3065         reply.path_length = htobe32((uint32_t) path_length);
3066         buf_ret = lttng_dynamic_buffer_append(
3067                         &reply_payload, &reply, sizeof(reply));
3068         if (buf_ret) {
3069                 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3070                 goto end_no_reply;
3071         }
3072
3073         if (reply_code == LTTNG_OK) {
3074                 buf_ret = lttng_dynamic_buffer_append(&reply_payload,
3075                                 closed_trace_chunk_path, path_length);
3076                 if (buf_ret) {
3077                         ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3078                         goto end_no_reply;
3079                 }
3080         }
3081
3082         send_ret = conn->sock->ops->sendmsg(conn->sock,
3083                         reply_payload.data,
3084                         reply_payload.size,
3085                         0);
3086         if (send_ret < reply_payload.size) {
3087                 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3088                                 reply_payload.size, send_ret);
3089                 ret = -1;
3090                 goto end_no_reply;
3091         }
3092 end_no_reply:
3093         lttng_trace_chunk_put(chunk);
3094         lttng_dynamic_buffer_reset(&reply_payload);
3095         return ret;
3096 }
3097
3098 /*
3099  * relay_trace_chunk_exists: check if a trace chunk exists
3100  */
3101 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr,
3102                 struct relay_connection *conn,
3103                 const struct lttng_buffer_view *payload)
3104 {
3105         int ret = 0;
3106         ssize_t send_ret;
3107         struct relay_session *session = conn->session;
3108         struct lttcomm_relayd_trace_chunk_exists *msg;
3109         struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3110         struct lttng_buffer_view header_view;
3111         uint64_t chunk_id;
3112         bool chunk_exists;
3113
3114         if (!session || !conn->version_check_done) {
3115                 ERR("Trying to check for the existance of a trace chunk before version check");
3116                 ret = -1;
3117                 goto end_no_reply;
3118         }
3119
3120         if (session->major == 2 && session->minor < 11) {
3121                 ERR("Chunk exists command is unsupported before 2.11");
3122                 ret = -1;
3123                 goto end_no_reply;
3124         }
3125
3126         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3127         if (!lttng_buffer_view_is_valid(&header_view)) {
3128                 ERR("Failed to receive payload of chunk exists command");
3129                 ret = -1;
3130                 goto end_no_reply;
3131         }
3132
3133         /* Convert to host endianness. */
3134         msg = (typeof(msg)) header_view.data;
3135         chunk_id = be64toh(msg->chunk_id);
3136
3137         ret = sessiond_trace_chunk_registry_chunk_exists(
3138                         sessiond_trace_chunk_registry,
3139                         conn->session->sessiond_uuid,
3140                         conn->session->id,
3141                         chunk_id, &chunk_exists);
3142         /*
3143          * If ret is not 0, send the reply and report the error to the caller.
3144          * It is a protocol (or internal) error and the session/connection
3145          * should be torn down.
3146          */
3147         reply = (typeof(reply)){
3148                 .generic.ret_code = htobe32((uint32_t)
3149                         (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3150                 .trace_chunk_exists = ret == 0 ? chunk_exists : 0,
3151         };
3152         send_ret = conn->sock->ops->sendmsg(
3153                         conn->sock, &reply, sizeof(reply), 0);
3154         if (send_ret < (ssize_t) sizeof(reply)) {
3155                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3156                                 send_ret);
3157                 ret = -1;
3158         }
3159 end_no_reply:
3160         return ret;
3161 }
3162
3163 /*
3164  * relay_get_configuration: query whether feature is available
3165  */
3166 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr,
3167                 struct relay_connection *conn,
3168                 const struct lttng_buffer_view *payload)
3169 {
3170         int ret = 0;
3171         ssize_t send_ret;
3172         struct lttcomm_relayd_get_configuration *msg;
3173         struct lttcomm_relayd_get_configuration_reply reply = {};
3174         struct lttng_buffer_view header_view;
3175         uint64_t query_flags = 0;
3176         uint64_t result_flags = 0;
3177
3178         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3179         if (!lttng_buffer_view_is_valid(&header_view)) {
3180                 ERR("Failed to receive payload of chunk close command");
3181                 ret = -1;
3182                 goto end_no_reply;
3183         }
3184
3185         /* Convert to host endianness. */
3186         msg = (typeof(msg)) header_view.data;
3187         query_flags = be64toh(msg->query_flags);
3188
3189         if (query_flags) {
3190                 ret = LTTNG_ERR_INVALID_PROTOCOL;
3191                 goto reply;
3192         }
3193         if (opt_allow_clear) {
3194                 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3195         }
3196         ret = 0;
3197 reply:
3198         reply = (typeof(reply)){
3199                 .generic.ret_code = htobe32((uint32_t)
3200                         (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3201                 .relayd_configuration_flags = htobe64(result_flags),
3202         };
3203         send_ret = conn->sock->ops->sendmsg(
3204                         conn->sock, &reply, sizeof(reply), 0);
3205         if (send_ret < (ssize_t) sizeof(reply)) {
3206                 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3207                                 send_ret);
3208                 ret = -1;
3209         }
3210 end_no_reply:
3211         return ret;
3212 }
3213
3214 #define DBG_CMD(cmd_name, conn) \
3215                 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3216
3217 static int relay_process_control_command(struct relay_connection *conn,
3218                 const struct lttcomm_relayd_hdr *header,
3219                 const struct lttng_buffer_view *payload)
3220 {
3221         int ret = 0;
3222
3223         switch (header->cmd) {
3224         case RELAYD_CREATE_SESSION:
3225                 DBG_CMD("RELAYD_CREATE_SESSION", conn);
3226                 ret = relay_create_session(header, conn, payload);
3227                 break;
3228         case RELAYD_ADD_STREAM:
3229                 DBG_CMD("RELAYD_ADD_STREAM", conn);
3230                 ret = relay_add_stream(header, conn, payload);
3231                 break;
3232         case RELAYD_START_DATA:
3233                 DBG_CMD("RELAYD_START_DATA", conn);
3234                 ret = relay_start(header, conn, payload);
3235                 break;
3236         case RELAYD_SEND_METADATA:
3237                 DBG_CMD("RELAYD_SEND_METADATA", conn);
3238                 ret = relay_recv_metadata(header, conn, payload);
3239                 break;
3240         case RELAYD_VERSION:
3241                 DBG_CMD("RELAYD_VERSION", conn);
3242                 ret = relay_send_version(header, conn, payload);
3243                 break;
3244         case RELAYD_CLOSE_STREAM:
3245                 DBG_CMD("RELAYD_CLOSE_STREAM", conn);
3246                 ret = relay_close_stream(header, conn, payload);
3247                 break;
3248         case RELAYD_DATA_PENDING:
3249                 DBG_CMD("RELAYD_DATA_PENDING", conn);
3250                 ret = relay_data_pending(header, conn, payload);
3251                 break;
3252         case RELAYD_QUIESCENT_CONTROL:
3253                 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
3254                 ret = relay_quiescent_control(header, conn, payload);
3255                 break;
3256         case RELAYD_BEGIN_DATA_PENDING:
3257                 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
3258                 ret = relay_begin_data_pending(header, conn, payload);
3259                 break;
3260         case RELAYD_END_DATA_PENDING:
3261                 DBG_CMD("RELAYD_END_DATA_PENDING", conn);
3262                 ret = relay_end_data_pending(header, conn, payload);
3263                 break;
3264         case RELAYD_SEND_INDEX:
3265                 DBG_CMD("RELAYD_SEND_INDEX", conn);
3266                 ret = relay_recv_index(header, conn, payload);
3267                 break;
3268         case RELAYD_STREAMS_SENT:
3269                 DBG_CMD("RELAYD_STREAMS_SENT", conn);
3270                 ret = relay_streams_sent(header, conn, payload);
3271                 break;
3272         case RELAYD_RESET_METADATA:
3273                 DBG_CMD("RELAYD_RESET_METADATA", conn);
3274                 ret = relay_reset_metadata(header, conn, payload);
3275                 break;
3276         case RELAYD_ROTATE_STREAMS:
3277                 DBG_CMD("RELAYD_ROTATE_STREAMS", conn);
3278                 ret = relay_rotate_session_streams(header, conn, payload);
3279                 break;
3280         case RELAYD_CREATE_TRACE_CHUNK:
3281                 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn);
3282                 ret = relay_create_trace_chunk(header, conn, payload);
3283                 break;
3284         case RELAYD_CLOSE_TRACE_CHUNK:
3285                 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn);
3286                 ret = relay_close_trace_chunk(header, conn, payload);
3287                 break;
3288         case RELAYD_TRACE_CHUNK_EXISTS:
3289                 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn);
3290                 ret = relay_trace_chunk_exists(header, conn, payload);
3291                 break;
3292         case RELAYD_GET_CONFIGURATION:
3293                 DBG_CMD("RELAYD_GET_CONFIGURATION", conn);
3294                 ret = relay_get_configuration(header, conn, payload);
3295                 break;
3296         case RELAYD_UPDATE_SYNC_INFO:
3297         default:
3298                 ERR("Received unknown command (%u)", header->cmd);
3299                 relay_unknown_command(conn);
3300                 ret = -1;
3301                 goto end;
3302         }
3303
3304 end:
3305         return ret;
3306 }
3307
3308 static enum relay_connection_status relay_process_control_receive_payload(
3309                 struct relay_connection *conn)
3310 {
3311         int ret = 0;
3312         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3313         struct lttng_dynamic_buffer *reception_buffer =
3314                         &conn->protocol.ctrl.reception_buffer;
3315         struct ctrl_connection_state_receive_payload *state =
3316                         &conn->protocol.ctrl.state.receive_payload;
3317         struct lttng_buffer_view payload_view;
3318
3319         if (state->left_to_receive == 0) {
3320                 /* Short-circuit for payload-less commands. */
3321                 goto reception_complete;
3322         }
3323
3324         ret = conn->sock->ops->recvmsg(conn->sock,
3325                         reception_buffer->data + state->received,
3326                         state->left_to_receive, MSG_DONTWAIT);
3327         if (ret < 0) {
3328                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3329                         PERROR("Unable to receive command payload on sock %d",
3330                                         conn->sock->fd);
3331                         status = RELAY_CONNECTION_STATUS_ERROR;
3332                 }
3333                 goto end;
3334         } else if (ret == 0) {
3335                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3336                 status = RELAY_CONNECTION_STATUS_CLOSED;
3337                 goto end;
3338         }
3339
3340         assert(ret > 0);
3341         assert(ret <= state->left_to_receive);
3342
3343         state->left_to_receive -= ret;
3344         state->received += ret;
3345
3346         if (state->left_to_receive > 0) {
3347                 /*
3348                  * Can't transition to the protocol's next state, wait to
3349                  * receive the rest of the header.
3350                  */
3351                 DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3352                                 state->received, state->left_to_receive,
3353                                 conn->sock->fd);
3354                 goto end;
3355         }
3356
3357 reception_complete:
3358         DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3359                         conn->sock->fd, state->received);
3360         /*
3361          * The payload required to process the command has been received.
3362          * A view to the reception buffer is forwarded to the various
3363          * commands and the state of the control is reset on success.
3364          *
3365          * Commands are responsible for sending their reply to the peer.
3366          */
3367         payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
3368                         0, -1);
3369         ret = relay_process_control_command(conn,
3370                         &state->header, &payload_view);
3371         if (ret < 0) {
3372                 status = RELAY_CONNECTION_STATUS_ERROR;
3373                 goto end;
3374         }
3375
3376         ret = connection_reset_protocol_state(conn);
3377         if (ret) {
3378                 status = RELAY_CONNECTION_STATUS_ERROR;
3379         }
3380 end:
3381         return status;
3382 }
3383
3384 static enum relay_connection_status relay_process_control_receive_header(
3385                 struct relay_connection *conn)
3386 {
3387         int ret = 0;
3388         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3389         struct lttcomm_relayd_hdr header;
3390         struct lttng_dynamic_buffer *reception_buffer =
3391                         &conn->protocol.ctrl.reception_buffer;
3392         struct ctrl_connection_state_receive_header *state =
3393                         &conn->protocol.ctrl.state.receive_header;
3394
3395         assert(state->left_to_receive != 0);
3396
3397         ret = conn->sock->ops->recvmsg(conn->sock,
3398                         reception_buffer->data + state->received,
3399                         state->left_to_receive, MSG_DONTWAIT);
3400         if (ret < 0) {
3401                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3402                         PERROR("Unable to receive control command header on sock %d",
3403                                         conn->sock->fd);
3404                         status = RELAY_CONNECTION_STATUS_ERROR;
3405                 }
3406                 goto end;
3407         } else if (ret == 0) {
3408                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3409                 status = RELAY_CONNECTION_STATUS_CLOSED;
3410                 goto end;
3411         }
3412
3413         assert(ret > 0);
3414         assert(ret <= state->left_to_receive);
3415
3416         state->left_to_receive -= ret;
3417         state->received += ret;
3418
3419         if (state->left_to_receive > 0) {
3420                 /*
3421                  * Can't transition to the protocol's next state, wait to
3422                  * receive the rest of the header.
3423                  */
3424                 DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3425                                 state->received, state->left_to_receive,
3426                                 conn->sock->fd);
3427                 goto end;
3428         }
3429
3430         /* Transition to next state: receiving the command's payload. */
3431         conn->protocol.ctrl.state_id =
3432                         CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3433         memcpy(&header, reception_buffer->data, sizeof(header));
3434         header.circuit_id = be64toh(header.circuit_id);
3435         header.data_size = be64toh(header.data_size);
3436         header.cmd = be32toh(header.cmd);
3437         header.cmd_version = be32toh(header.cmd_version);
3438         memcpy(&conn->protocol.ctrl.state.receive_payload.header,
3439                         &header, sizeof(header));
3440
3441         DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
3442                         conn->sock->fd, header.cmd, header.cmd_version,
3443                         header.data_size);
3444
3445         if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3446                 ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3447                                 header.data_size);
3448                 status = RELAY_CONNECTION_STATUS_ERROR;
3449                 goto end;
3450         }
3451
3452         conn->protocol.ctrl.state.receive_payload.left_to_receive =
3453                         header.data_size;
3454         conn->protocol.ctrl.state.receive_payload.received = 0;
3455         ret = lttng_dynamic_buffer_set_size(reception_buffer,
3456                         header.data_size);
3457         if (ret) {
3458                 status = RELAY_CONNECTION_STATUS_ERROR;
3459                 goto end;
3460         }
3461
3462         if (header.data_size == 0) {
3463                 /*
3464                  * Manually invoke the next state as the poll loop
3465                  * will not wake-up to allow us to proceed further.
3466                  */
3467                 status = relay_process_control_receive_payload(conn);
3468         }
3469 end:
3470         return status;
3471 }
3472
3473 /*
3474  * Process the commands received on the control socket
3475  */
3476 static enum relay_connection_status relay_process_control(
3477                 struct relay_connection *conn)
3478 {
3479         enum relay_connection_status status;
3480
3481         switch (conn->protocol.ctrl.state_id) {
3482         case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3483                 status = relay_process_control_receive_header(conn);
3484                 break;
3485         case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3486                 status = relay_process_control_receive_payload(conn);
3487                 break;
3488         default:
3489                 ERR("Unknown control connection protocol state encountered.");
3490                 abort();
3491         }
3492
3493         return status;
3494 }
3495
3496 static enum relay_connection_status relay_process_data_receive_header(
3497                 struct relay_connection *conn)
3498 {
3499         int ret;
3500         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3501         struct data_connection_state_receive_header *state =
3502                         &conn->protocol.data.state.receive_header;
3503         struct lttcomm_relayd_data_hdr header;
3504         struct relay_stream *stream;
3505
3506         assert(state->left_to_receive != 0);
3507
3508         ret = conn->sock->ops->recvmsg(conn->sock,
3509                         state->header_reception_buffer + state->received,
3510                         state->left_to_receive, MSG_DONTWAIT);
3511         if (ret < 0) {
3512                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3513                         PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3514                         status = RELAY_CONNECTION_STATUS_ERROR;
3515                 }
3516                 goto end;
3517         } else if (ret == 0) {
3518                 /* Orderly shutdown. Not necessary to print an error. */
3519                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3520                 status = RELAY_CONNECTION_STATUS_CLOSED;
3521                 goto end;
3522         }
3523
3524         assert(ret > 0);
3525         assert(ret <= state->left_to_receive);
3526
3527         state->left_to_receive -= ret;
3528         state->received += ret;
3529
3530         if (state->left_to_receive > 0) {
3531                 /*
3532                  * Can't transition to the protocol's next state, wait to
3533                  * receive the rest of the header.
3534                  */
3535                 DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3536                                 state->received, state->left_to_receive,
3537                                 conn->sock->fd);
3538                 goto end;
3539         }
3540
3541         /* Transition to next state: receiving the payload. */
3542         conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3543
3544         memcpy(&header, state->header_reception_buffer, sizeof(header));
3545         header.circuit_id = be64toh(header.circuit_id);
3546         header.stream_id = be64toh(header.stream_id);
3547         header.data_size = be32toh(header.data_size);
3548         header.net_seq_num = be64toh(header.net_seq_num);
3549         header.padding_size = be32toh(header.padding_size);
3550         memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3551
3552         conn->protocol.data.state.receive_payload.left_to_receive =
3553                         header.data_size;
3554         conn->protocol.data.state.receive_payload.received = 0;
3555         conn->protocol.data.state.receive_payload.rotate_index = false;
3556
3557         DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32,
3558                         conn->sock->fd, header.circuit_id,
3559                         header.stream_id, header.data_size,
3560                         header.net_seq_num, header.padding_size);
3561
3562         stream = stream_get_by_id(header.stream_id);
3563         if (!stream) {
3564                 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3565                                 header.stream_id);
3566                 /* Protocol error. */
3567                 status = RELAY_CONNECTION_STATUS_ERROR;
3568                 goto end;
3569         }
3570
3571         pthread_mutex_lock(&stream->lock);
3572         /* Prepare stream for the reception of a new packet. */
3573         ret = stream_init_packet(stream, header.data_size,
3574                         &conn->protocol.data.state.receive_payload.rotate_index);
3575         pthread_mutex_unlock(&stream->lock);
3576         if (ret) {
3577                 ERR("Failed to rotate stream output file");
3578                 status = RELAY_CONNECTION_STATUS_ERROR;
3579                 goto end_stream_unlock;
3580         }
3581
3582 end_stream_unlock:
3583         stream_put(stream);
3584 end:
3585         return status;
3586 }
3587
3588 static enum relay_connection_status relay_process_data_receive_payload(
3589                 struct relay_connection *conn)
3590 {
3591         int ret;
3592         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3593         struct relay_stream *stream;
3594         struct data_connection_state_receive_payload *state =
3595                         &conn->protocol.data.state.receive_payload;
3596         const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3597         char data_buffer[chunk_size];
3598         bool partial_recv = false;
3599         bool new_stream = false, close_requested = false, index_flushed = false;
3600         uint64_t left_to_receive = state->left_to_receive;
3601         struct relay_session *session;
3602
3603         DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
3604                         state->header.stream_id, state->header.net_seq_num,
3605                         state->received, left_to_receive);
3606
3607         stream = stream_get_by_id(state->header.stream_id);
3608         if (!stream) {
3609                 /* Protocol error. */
3610                 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3611                                 state->header.stream_id);
3612                 status = RELAY_CONNECTION_STATUS_ERROR;
3613                 goto end;
3614         }
3615
3616         pthread_mutex_lock(&stream->lock);
3617         session = stream->trace->session;
3618         if (!conn->session) {
3619                 ret = connection_set_session(conn, session);
3620                 if (ret) {
3621                         status = RELAY_CONNECTION_STATUS_ERROR;
3622                         goto end_stream_unlock;
3623                 }
3624         }
3625
3626         /*
3627          * The size of the "chunk" received on any iteration is bounded by:
3628          *   - the data left to receive,
3629          *   - the data immediately available on the socket,
3630          *   - the on-stack data buffer
3631          */
3632         while (left_to_receive > 0 && !partial_recv) {
3633                 size_t recv_size = min(left_to_receive, chunk_size);
3634                 struct lttng_buffer_view packet_chunk;
3635
3636                 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer,
3637                                 recv_size, MSG_DONTWAIT);
3638                 if (ret < 0) {
3639                         if (errno != EAGAIN && errno != EWOULDBLOCK) {
3640                                 PERROR("Socket %d error", conn->sock->fd);
3641                                 status = RELAY_CONNECTION_STATUS_ERROR;
3642                         }
3643                         goto end_stream_unlock;
3644                 } else if (ret == 0) {
3645                         /* No more data ready to be consumed on socket. */
3646                         DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3647                                         state->header.stream_id);
3648                         status = RELAY_CONNECTION_STATUS_CLOSED;
3649                         break;
3650                 } else if (ret < (int) recv_size) {
3651                         /*
3652                          * All the data available on the socket has been
3653                          * consumed.
3654                          */
3655                         partial_recv = true;
3656                         recv_size = ret;
3657                 }
3658
3659                 packet_chunk = lttng_buffer_view_init(data_buffer,
3660                                 0, recv_size);
3661                 assert(packet_chunk.data);
3662
3663                 ret = stream_write(stream, &packet_chunk, 0);
3664                 if (ret) {
3665                         ERR("Relay error writing data to file");
3666                         status = RELAY_CONNECTION_STATUS_ERROR;
3667                         goto end_stream_unlock;
3668                 }
3669
3670                 left_to_receive -= recv_size;
3671                 state->received += recv_size;
3672                 state->left_to_receive = left_to_receive;
3673         }
3674
3675         if (state->left_to_receive > 0) {
3676                 /*
3677                  * Did not receive all the data expected, wait for more data to
3678                  * become available on the socket.
3679                  */
3680                 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
3681                                 state->header.stream_id, state->received,
3682                                 state->left_to_receive);
3683                 goto end_stream_unlock;
3684         }
3685
3686         ret = stream_write(stream, NULL, state->header.padding_size);
3687         if (ret) {
3688                 status = RELAY_CONNECTION_STATUS_ERROR;
3689                 goto end_stream_unlock;
3690         }
3691
3692         if (session_streams_have_index(session)) {
3693                 ret = stream_update_index(stream, state->header.net_seq_num,
3694                                 state->rotate_index, &index_flushed,
3695                                 state->header.data_size + state->header.padding_size);
3696                 if (ret < 0) {
3697                         ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
3698                                         stream->stream_handle,
3699                                         state->header.net_seq_num, ret);
3700                         status = RELAY_CONNECTION_STATUS_ERROR;
3701                         goto end_stream_unlock;
3702                 }
3703         }
3704
3705         if (stream->prev_data_seq == -1ULL) {
3706                 new_stream = true;
3707         }
3708
3709         ret = stream_complete_packet(stream, state->header.data_size +
3710                         state->header.padding_size, state->header.net_seq_num,
3711                         index_flushed);
3712         if (ret) {
3713                 status = RELAY_CONNECTION_STATUS_ERROR;
3714                 goto end_stream_unlock;
3715         }
3716
3717         /*
3718          * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3719          * contents of *state which are aliased (union) to the same location as
3720          * the new state. Don't use it beyond this point.
3721          */
3722         connection_reset_protocol_state(conn);
3723         state = NULL;
3724
3725 end_stream_unlock:
3726         close_requested = stream->close_requested;
3727         pthread_mutex_unlock(&stream->lock);
3728         if (close_requested && left_to_receive == 0) {
3729                 try_stream_close(stream);
3730         }
3731
3732         if (new_stream) {
3733                 pthread_mutex_lock(&session->lock);
3734                 uatomic_set(&session->new_streams, 1);
3735                 pthread_mutex_unlock(&session->lock);
3736         }
3737
3738         stream_put(stream);
3739 end:
3740         return status;
3741 }
3742
3743 /*
3744  * relay_process_data: Process the data received on the data socket
3745  */
3746 static enum relay_connection_status relay_process_data(
3747                 struct relay_connection *conn)
3748 {
3749         enum relay_connection_status status;
3750
3751         switch (conn->protocol.data.state_id) {
3752         case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3753                 status = relay_process_data_receive_header(conn);
3754                 break;
3755         case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3756                 status = relay_process_data_receive_payload(conn);
3757                 break;
3758         default:
3759                 ERR("Unexpected data connection communication state.");
3760                 abort();
3761         }
3762
3763         return status;
3764 }
3765
3766 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3767 {
3768         int ret;
3769
3770         (void) lttng_poll_del(events, pollfd);
3771
3772         ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
3773                         fd_tracker_util_close_fd, NULL);
3774         if (ret < 0) {
3775                 ERR("Closing pollfd %d", pollfd);
3776         }
3777 }
3778
3779 static void relay_thread_close_connection(struct lttng_poll_event *events,
3780                 int pollfd, struct relay_connection *conn)
3781 {
3782         const char *type_str;
3783
3784         switch (conn->type) {
3785         case RELAY_DATA:
3786                 type_str = "Data";
3787                 break;
3788         case RELAY_CONTROL:
3789                 type_str = "Control";
3790                 break;
3791         case RELAY_VIEWER_COMMAND:
3792                 type_str = "Viewer Command";
3793                 break;
3794         case RELAY_VIEWER_NOTIFICATION:
3795                 type_str = "Viewer Notification";
3796                 break;
3797         default:
3798                 type_str = "Unknown";
3799         }
3800         cleanup_connection_pollfd(events, pollfd);
3801         connection_put(conn);
3802         DBG("%s connection closed with %d", type_str, pollfd);
3803 }
3804
3805 /*
3806  * This thread does the actual work
3807  */
3808 static void *relay_thread_worker(void *data)
3809 {
3810         int ret, err = -1, last_seen_data_fd = -1;
3811         uint32_t nb_fd;
3812         struct lttng_poll_event events;
3813         struct lttng_ht *relay_connections_ht;
3814         struct lttng_ht_iter iter;
3815         struct relay_connection *destroy_conn = NULL;
3816
3817         DBG("[thread] Relay worker started");
3818
3819         rcu_register_thread();
3820
3821         health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3822
3823         if (testpoint(relayd_thread_worker)) {
3824                 goto error_testpoint;
3825         }
3826
3827         health_code_update();
3828
3829         /* table of connections indexed on socket */
3830         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3831         if (!relay_connections_ht) {
3832                 goto relay_connections_ht_error;
3833         }
3834
3835         ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3836         if (ret < 0) {
3837                 goto error_poll_create;
3838         }
3839
3840         ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3841         if (ret < 0) {
3842                 goto error;
3843         }
3844
3845 restart:
3846         while (1) {
3847                 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3848
3849                 health_code_update();
3850
3851                 /* Infinite blocking call, waiting for transmission */
3852                 DBG3("Relayd worker thread polling...");
3853                 health_poll_entry();
3854                 ret = lttng_poll_wait(&events, -1);
3855                 health_poll_exit();
3856                 if (ret < 0) {
3857                         /*
3858                          * Restart interrupted system call.
3859                          */
3860                         if (errno == EINTR) {
3861                                 goto restart;
3862                         }
3863                         goto error;
3864                 }
3865
3866                 nb_fd = ret;
3867
3868                 /*
3869                  * Process control. The control connection is
3870                  * prioritized so we don't starve it with high
3871                  * throughput tracing data on the data connection.
3872                  */
3873                 for (i = 0; i < nb_fd; i++) {
3874                         /* Fetch once the poll data */
3875                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3876                         int pollfd = LTTNG_POLL_GETFD(&events, i);
3877
3878                         health_code_update();
3879
3880                         /* Thread quit pipe has been closed. Killing thread. */
3881                         ret = check_thread_quit_pipe(pollfd, revents);
3882                         if (ret) {
3883                                 err = 0;
3884                                 goto exit;
3885                         }
3886
3887                         /* Inspect the relay conn pipe for new connection */
3888                         if (pollfd == relay_conn_pipe[0]) {
3889                                 if (revents & LPOLLIN) {
3890                                         struct relay_connection *conn;
3891
3892                                         ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
3893                                         if (ret < 0) {
3894                                                 goto error;
3895                                         }
3896                                         ret = lttng_poll_add(&events,
3897                                                         conn->sock->fd,
3898                                                         LPOLLIN | LPOLLRDHUP);
3899                                         if (ret) {
3900                                                 ERR("Failed to add new connection file descriptor to poll set");
3901                                                 goto error;
3902                                         }
3903                                         connection_ht_add(relay_connections_ht, conn);
3904                                         DBG("Connection socket %d added", conn->sock->fd);
3905                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3906                                         ERR("Relay connection pipe error");
3907                                         goto error;
3908                                 } else {
3909                                         ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3910                                         goto error;
3911                                 }
3912                         } else {
3913                                 struct relay_connection *ctrl_conn;
3914
3915                                 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
3916                                 /* If not found, there is a synchronization issue. */
3917                                 assert(ctrl_conn);
3918
3919                                 if (ctrl_conn->type == RELAY_DATA) {
3920                                         if (revents & LPOLLIN) {
3921                                                 /*
3922                                                  * Flag the last seen data fd not deleted. It will be
3923                                                  * used as the last seen fd if any fd gets deleted in
3924                                                  * this first loop.
3925                                                  */
3926                                                 last_notdel_data_fd = pollfd;
3927                                         }
3928                                         goto put_ctrl_connection;
3929                                 }
3930                                 assert(ctrl_conn->type == RELAY_CONTROL);
3931
3932                                 if (revents & LPOLLIN) {
3933                                         enum relay_connection_status status;
3934
3935                                         status = relay_process_control(ctrl_conn);
3936                                         if (status != RELAY_CONNECTION_STATUS_OK) {
3937                                                 /*
3938                                                  * On socket error flag the session as aborted to force
3939                                                  * the cleanup of its stream otherwise it can leak
3940                                                  * during the lifetime of the relayd.
3941                                                  *
3942                                                  * This prevents situations in which streams can be
3943                                                  * left opened because an index was received, the
3944                                                  * control connection is closed, and the data
3945                                                  * connection is closed (uncleanly) before the packet's
3946                                                  * data provided.
3947                                                  *
3948                                                  * Since the control connection encountered an error,
3949                                                  * it is okay to be conservative and close the
3950                                                  * session right now as we can't rely on the protocol
3951                                                  * being respected anymore.
3952                                                  */
3953                                                 if (status == RELAY_CONNECTION_STATUS_ERROR) {
3954                                                         session_abort(ctrl_conn->session);
3955                                                 }
3956
3957                                                 /* Clear the connection on error or close. */
3958                                                 relay_thread_close_connection(&events,
3959                                                                 pollfd,
3960                                                                 ctrl_conn);
3961                                         }
3962                                         seen_control = 1;
3963                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3964                                         relay_thread_close_connection(&events,
3965                                                         pollfd, ctrl_conn);
3966                                         if (last_seen_data_fd == pollfd) {
3967                                                 last_seen_data_fd = last_notdel_data_fd;
3968                                         }
3969                                 } else {
3970                                         ERR("Unexpected poll events %u for control sock %d",
3971                                                         revents, pollfd);
3972                                         connection_put(ctrl_conn);
3973                                         goto error;
3974                                 }
3975                         put_ctrl_connection:
3976                                 connection_put(ctrl_conn);
3977                         }
3978                 }
3979
3980                 /*
3981                  * The last loop handled a control request, go back to poll to make
3982                  * sure we prioritise the control socket.
3983                  */
3984                 if (seen_control) {
3985                         continue;
3986                 }
3987
3988                 if (last_seen_data_fd >= 0) {
3989                         for (i = 0; i < nb_fd; i++) {
3990                                 int pollfd = LTTNG_POLL_GETFD(&events, i);
3991
3992                                 health_code_update();
3993
3994                                 if (last_seen_data_fd == pollfd) {
3995                                         idx = i;
3996                                         break;
3997                                 }
3998                         }
3999                 }
4000
4001                 /* Process data connection. */
4002                 for (i = idx + 1; i < nb_fd; i++) {
4003                         /* Fetch the poll data. */
4004                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
4005                         int pollfd = LTTNG_POLL_GETFD(&events, i);
4006                         struct relay_connection *data_conn;
4007
4008                         health_code_update();
4009
4010                         if (!revents) {
4011                                 /* No activity for this FD (poll implementation). */
4012                                 continue;
4013                         }
4014
4015                         /* Skip the command pipe. It's handled in the first loop. */
4016                         if (pollfd == relay_conn_pipe[0]) {
4017                                 continue;
4018                         }
4019
4020                         data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4021                         if (!data_conn) {
4022                                 /* Skip it. Might be removed before. */
4023                                 continue;
4024                         }
4025                         if (data_conn->type == RELAY_CONTROL) {
4026                                 goto put_data_connection;
4027                         }
4028                         assert(data_conn->type == RELAY_DATA);
4029
4030                         if (revents & LPOLLIN) {
4031                                 enum relay_connection_status status;
4032
4033                                 status = relay_process_data(data_conn);
4034                                 /* Connection closed or error. */
4035                                 if (status != RELAY_CONNECTION_STATUS_OK) {
4036                                         /*
4037                                          * On socket error flag the session as aborted to force
4038                                          * the cleanup of its stream otherwise it can leak
4039                                          * during the lifetime of the relayd.
4040                                          *
4041                                          * This prevents situations in which streams can be
4042                                          * left opened because an index was received, the
4043                                          * control connection is closed, and the data
4044                                          * connection is closed (uncleanly) before the packet's
4045                                          * data provided.
4046                                          *
4047                                          * Since the data connection encountered an error,
4048                                          * it is okay to be conservative and close the
4049                                          * session right now as we can't rely on the protocol
4050                                          * being respected anymore.
4051                                          */
4052                                         if (status == RELAY_CONNECTION_STATUS_ERROR) {
4053                                                 session_abort(data_conn->session);
4054                                         }
4055                                         relay_thread_close_connection(&events, pollfd,
4056                                                         data_conn);
4057                                         /*
4058                                          * Every goto restart call sets the last seen fd where
4059                                          * here we don't really care since we gracefully
4060                                          * continue the loop after the connection is deleted.
4061                                          */
4062                                 } else {
4063                                         /* Keep last seen port. */
4064                                         last_seen_data_fd = pollfd;
4065                                         connection_put(data_conn);
4066                                         goto restart;
4067                                 }
4068                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4069                                 relay_thread_close_connection(&events, pollfd,
4070                                                 data_conn);
4071                         } else {
4072                                 ERR("Unknown poll events %u for data sock %d",
4073                                                 revents, pollfd);
4074                         }
4075                 put_data_connection:
4076                         connection_put(data_conn);
4077                 }
4078                 last_seen_data_fd = -1;
4079         }
4080
4081         /* Normal exit, no error */
4082         ret = 0;
4083
4084 exit:
4085 error:
4086         /* Cleanup remaining connection object. */
4087         rcu_read_lock();
4088         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
4089                         destroy_conn,
4090                         sock_n.node) {
4091                 health_code_update();
4092
4093                 session_abort(destroy_conn->session);
4094
4095                 /*
4096                  * No need to grab another ref, because we own
4097                  * destroy_conn.
4098                  */
4099                 relay_thread_close_connection(&events, destroy_conn->sock->fd,
4100                                 destroy_conn);
4101         }
4102         rcu_read_unlock();
4103
4104         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4105 error_poll_create:
4106         lttng_ht_destroy(relay_connections_ht);
4107 relay_connections_ht_error:
4108         /* Close relay conn pipes */
4109         (void) fd_tracker_util_pipe_close(the_fd_tracker,
4110                         relay_conn_pipe);
4111         if (err) {
4112                 DBG("Thread exited with error");
4113         }
4114         DBG("Worker thread cleanup complete");
4115 error_testpoint:
4116         if (err) {
4117                 health_error();
4118                 ERR("Health error occurred in %s", __func__);
4119         }
4120         health_unregister(health_relayd);
4121         rcu_unregister_thread();
4122         lttng_relay_stop_threads();
4123         return NULL;
4124 }
4125
4126 /*
4127  * Create the relay command pipe to wake thread_manage_apps.
4128  * Closed in cleanup().
4129  */
4130 static int create_relay_conn_pipe(void)
4131 {
4132         return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
4133                         "Relayd connection pipe", relay_conn_pipe);
4134 }
4135
4136 static int stdio_open(void *data, int *fds)
4137 {
4138         fds[0] = fileno(stdout);
4139         fds[1] = fileno(stderr);
4140         return 0;
4141 }
4142
4143 static int track_stdio(void)
4144 {
4145         int fds[2];
4146         const char *names[] = { "stdout", "stderr" };
4147
4148         return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds,
4149                         names, 2, stdio_open, NULL);
4150 }
4151
4152 /*
4153  * main
4154  */
4155 int main(int argc, char **argv)
4156 {
4157         bool thread_is_rcu_registered = false;
4158         int ret = 0, retval = 0;
4159         void *status;
4160         char *unlinked_file_directory_path = NULL, *output_path = NULL;
4161
4162         /* Parse environment variables */
4163         ret = parse_env_options();
4164         if (ret) {
4165                 retval = -1;
4166                 goto exit_options;
4167         }
4168
4169         /*
4170          * Parse arguments.
4171          * Command line arguments overwrite environment.
4172          */
4173         progname = argv[0];
4174         if (set_options(argc, argv)) {
4175                 retval = -1;
4176                 goto exit_options;
4177         }
4178
4179         if (set_signal_handler()) {
4180                 retval = -1;
4181                 goto exit_options;
4182         }
4183
4184         relayd_config_log();
4185
4186         if (opt_print_version) {
4187                 print_version();
4188                 retval = 0;
4189                 goto exit_options;
4190         }
4191
4192         ret = fclose(stdin);
4193         if (ret) {
4194                 PERROR("Failed to close stdin");
4195                 goto exit_options;
4196         }
4197
4198         DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4199
4200         /* Try to create directory if -o, --output is specified. */
4201         if (opt_output_path) {
4202                 if (*opt_output_path != '/') {
4203                         ERR("Please specify an absolute path for -o, --output PATH");
4204                         retval = -1;
4205                         goto exit_options;
4206                 }
4207
4208                 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
4209                                 -1, -1);
4210                 if (ret < 0) {
4211                         ERR("Unable to create %s", opt_output_path);
4212                         retval = -1;
4213                         goto exit_options;
4214                 }
4215         }
4216
4217         /* Daemonize */
4218         if (opt_daemon || opt_background) {
4219                 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
4220                         !opt_background);
4221                 if (ret < 0) {
4222                         retval = -1;
4223                         goto exit_options;
4224                 }
4225         }
4226
4227         if (opt_working_directory) {
4228                 ret = utils_change_working_directory(opt_working_directory);
4229                 if (ret) {
4230                         /* All errors are already logged. */
4231                         goto exit_options;
4232                 }
4233         }
4234
4235         sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4236         if (!sessiond_trace_chunk_registry) {
4237                 ERR("Failed to initialize session daemon trace chunk registry");
4238                 retval = -1;
4239                 goto exit_options;
4240         }
4241
4242         /*
4243          * The RCU thread registration (and use, through the fd-tracker's
4244          * creation) is done after the daemonization to allow us to not
4245          * deal with liburcu's fork() management as the call RCU needs to
4246          * be restored.
4247          */
4248         rcu_register_thread();
4249         thread_is_rcu_registered = true;
4250
4251         output_path = create_output_path("");
4252         if (!output_path) {
4253                 ERR("Failed to get output path");
4254                 retval = -1;
4255                 goto exit_options;
4256         }
4257         ret = asprintf(&unlinked_file_directory_path, "%s/%s", output_path,
4258                         DEFAULT_UNLINKED_FILES_DIRECTORY);
4259         free(output_path);
4260         if (ret < 0) {
4261                 ERR("Failed to format unlinked file directory path");
4262                 retval = -1;
4263                 goto exit_options;
4264         }
4265         the_fd_tracker = fd_tracker_create(
4266                         unlinked_file_directory_path, lttng_opt_fd_pool_size);
4267         free(unlinked_file_directory_path);
4268         if (!the_fd_tracker) {
4269                 retval = -1;
4270                 goto exit_options;
4271         }
4272
4273         ret = track_stdio();
4274         if (ret) {
4275                 retval = -1;
4276                 goto exit_options;
4277         }
4278
4279         /* Initialize thread health monitoring */
4280         health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4281         if (!health_relayd) {
4282                 PERROR("health_app_create error");
4283                 retval = -1;
4284                 goto exit_options;
4285         }
4286
4287         /* Create thread quit pipe */
4288         if (init_thread_quit_pipe()) {
4289                 retval = -1;
4290                 goto exit_options;
4291         }
4292
4293         /* Setup the thread apps communication pipe. */
4294         if (create_relay_conn_pipe()) {
4295                 retval = -1;
4296                 goto exit_options;
4297         }
4298
4299         /* Init relay command queue. */
4300         cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4301
4302         /* Initialize communication library */
4303         lttcomm_init();
4304         lttcomm_inet_init();
4305
4306         /* tables of sessions indexed by session ID */
4307         sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4308         if (!sessions_ht) {
4309                 retval = -1;
4310                 goto exit_options;
4311         }
4312
4313         /* tables of streams indexed by stream ID */
4314         relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4315         if (!relay_streams_ht) {
4316                 retval = -1;
4317                 goto exit_options;
4318         }
4319
4320         /* tables of streams indexed by stream ID */
4321         viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4322         if (!viewer_streams_ht) {
4323                 retval = -1;
4324                 goto exit_options;
4325         }
4326
4327         ret = init_health_quit_pipe();
4328         if (ret) {
4329                 retval = -1;
4330                 goto exit_options;
4331         }
4332
4333         /* Create thread to manage the client socket */
4334         ret = pthread_create(&health_thread, default_pthread_attr(),
4335                         thread_manage_health, (void *) NULL);
4336         if (ret) {
4337                 errno = ret;
4338                 PERROR("pthread_create health");
4339                 retval = -1;
4340                 goto exit_options;
4341         }
4342
4343         /* Setup the dispatcher thread */
4344         ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
4345                         relay_thread_dispatcher, (void *) NULL);
4346         if (ret) {
4347                 errno = ret;
4348                 PERROR("pthread_create dispatcher");
4349                 retval = -1;
4350                 goto exit_dispatcher_thread;
4351         }
4352
4353         /* Setup the worker thread */
4354         ret = pthread_create(&worker_thread, default_pthread_attr(),
4355                         relay_thread_worker, NULL);
4356         if (ret) {
4357                 errno = ret;
4358                 PERROR("pthread_create worker");
4359                 retval = -1;
4360                 goto exit_worker_thread;
4361         }
4362
4363         /* Setup the listener thread */
4364         ret = pthread_create(&listener_thread, default_pthread_attr(),
4365                         relay_thread_listener, (void *) NULL);
4366         if (ret) {
4367                 errno = ret;
4368                 PERROR("pthread_create listener");
4369                 retval = -1;
4370                 goto exit_listener_thread;
4371         }
4372
4373         ret = relayd_live_create(live_uri);
4374         if (ret) {
4375                 ERR("Starting live viewer threads");
4376                 retval = -1;
4377                 goto exit_live;
4378         }
4379
4380         /*
4381          * This is where we start awaiting program completion (e.g. through
4382          * signal that asks threads to teardown).
4383          */
4384
4385         ret = relayd_live_join();
4386         if (ret) {
4387                 retval = -1;
4388         }
4389 exit_live:
4390
4391         ret = pthread_join(listener_thread, &status);
4392         if (ret) {
4393                 errno = ret;
4394                 PERROR("pthread_join listener_thread");
4395                 retval = -1;
4396         }
4397
4398 exit_listener_thread:
4399         ret = pthread_join(worker_thread, &status);
4400         if (ret) {
4401                 errno = ret;
4402                 PERROR("pthread_join worker_thread");
4403                 retval = -1;
4404         }
4405
4406 exit_worker_thread:
4407         ret = pthread_join(dispatcher_thread, &status);
4408         if (ret) {
4409                 errno = ret;
4410                 PERROR("pthread_join dispatcher_thread");
4411                 retval = -1;
4412         }
4413 exit_dispatcher_thread:
4414
4415         ret = pthread_join(health_thread, &status);
4416         if (ret) {
4417                 errno = ret;
4418                 PERROR("pthread_join health_thread");
4419                 retval = -1;
4420         }
4421 exit_options:
4422         /*
4423          * Wait for all pending call_rcu work to complete before tearing
4424          * down data structures. call_rcu worker may be trying to
4425          * perform lookups in those structures.
4426          */
4427         rcu_barrier();
4428         relayd_cleanup();
4429
4430         /* Ensure all prior call_rcu are done. */
4431         rcu_barrier();
4432
4433         if (thread_is_rcu_registered) {
4434                 rcu_unregister_thread();
4435         }
4436
4437         if (!retval) {
4438                 exit(EXIT_SUCCESS);
4439         } else {
4440                 exit(EXIT_FAILURE);
4441         }
4442 }