src/bin/lttng-relayd/main.cpp

   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 "backward-compatibility-group-by.hpp"
  13 #include "cmd.hpp"
  14 #include "connection.hpp"
  15 #include "ctf-trace.hpp"
  16 #include "health-relayd.hpp"
  17 #include "index.hpp"
  18 #include "live.hpp"
  19 #include "lttng-relayd.hpp"
  20 #include "session.hpp"
  21 #include "sessiond-trace-chunks.hpp"
  22 #include "stream.hpp"
  23 #include "tcp_keep_alive.hpp"
  24 #include "testpoint.hpp"
  25 #include "tracefile-array.hpp"
  26 #include "utils.hpp"
  27 #include "version.hpp"
  28 #include "viewer-stream.hpp"
  29
  30 #include <common/align.hpp>
  31 #include <common/buffer-view.hpp>
  32 #include <common/common.hpp>
  33 #include <common/compat/endian.hpp>
  34 #include <common/compat/getenv.hpp>
  35 #include <common/compat/poll.hpp>
  36 #include <common/compat/socket.hpp>
  37 #include <common/daemonize.hpp>
  38 #include <common/defaults.hpp>
  39 #include <common/dynamic-buffer.hpp>
  40 #include <common/fd-tracker/fd-tracker.hpp>
  41 #include <common/fd-tracker/utils.hpp>
  42 #include <common/futex.hpp>
  43 #include <common/ini-config/ini-config.hpp>
  44 #include <common/path.hpp>
  45 #include <common/sessiond-comm/inet.hpp>
  46 #include <common/sessiond-comm/relayd.hpp>
  47 #include <common/sessiond-comm/sessiond-comm.hpp>
  48 #include <common/string-utils/format.hpp>
  49 #include <common/urcu.hpp>
  50 #include <common/uri.hpp>
  51 #include <common/utils.hpp>
  52
  53 #include <lttng/lttng.h>
  54
  55 #include <algorithm>
  56 #include <ctype.h>
  57 #include <fcntl.h>
  58 #include <getopt.h>
  59 #include <grp.h>
  60 #include <inttypes.h>
  61 #include <limits.h>
  62 #include <pthread.h>
  63 #include <signal.h>
  64 #include <stdio.h>
  65 #include <stdlib.h>
  66 #include <string.h>
  67 #include <strings.h>
  68 #include <sys/mman.h>
  69 #include <sys/mount.h>
  70 #include <sys/resource.h>
  71 #include <sys/socket.h>
  72 #include <sys/stat.h>
  73 #include <sys/types.h>
  74 #include <sys/wait.h>
  75 #include <unistd.h>
  76 #include <urcu/futex.h>
  77 #include <urcu/rculist.h>
  78 #include <urcu/uatomic.h>
  79
  80 static const char *help_msg =
  81 #ifdef LTTNG_EMBED_HELP
  82 #include <lttng-relayd.8.h>
  83 #else
  84         nullptr
  85 #endif
  86         ;
  87
  88 enum relay_connection_status {
  89         RELAY_CONNECTION_STATUS_OK,
  90         /* An error occurred while processing an event on the connection. */
  91         RELAY_CONNECTION_STATUS_ERROR,
  92         /* Connection closed/shutdown cleanly. */
  93         RELAY_CONNECTION_STATUS_CLOSED,
  94 };
  95
  96 /* command line options */
  97 char *opt_output_path, *opt_working_directory;
  98 static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
  99 enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
 100
 101 /* Argument variables */
 102 int lttng_opt_quiet; /* not static in error.h */
 103 int lttng_opt_verbose; /* not static in error.h */
 104 int lttng_opt_mi; /* not static in error.h */
 105
 106 /*
 107  * We need to wait for listener and live listener threads, as well as
 108  * health check thread, before being ready to signal readiness.
 109  */
 110 #define NR_LTTNG_RELAY_READY 3
 111 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
 112
 113 /* Size of receive buffer. */
 114 #define RECV_DATA_BUFFER_SIZE 65536
 115
 116 static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
 117 static pid_t child_ppid; /* Internal parent PID use with daemonize. */
 118
 119 static struct lttng_uri *control_uri;
 120 static struct lttng_uri *data_uri;
 121 static struct lttng_uri *live_uri;
 122
 123 const char *progname;
 124
 125 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
 126 static int tracing_group_name_override;
 127
 128 const char *const config_section_name = "relayd";
 129
 130 /*
 131  * This pipe is used to inform the worker thread that a command is queued and
 132  * ready to be processed.
 133  */
 134 static int relay_conn_pipe[2] = { -1, -1 };
 135
 136 /* Shared between threads */
 137 static int dispatch_thread_exit;
 138
 139 static pthread_t listener_thread;
 140 static pthread_t dispatcher_thread;
 141 static pthread_t worker_thread;
 142 static pthread_t health_thread;
 143
 144 /*
 145  * last_relay_stream_id_lock protects last_relay_stream_id increment
 146  * atomicity on 32-bit architectures.
 147  */
 148 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
 149 static uint64_t last_relay_stream_id;
 150
 151 /*
 152  * Relay command queue.
 153  *
 154  * The relay_thread_listener and relay_thread_dispatcher communicate with this
 155  * queue.
 156  */
 157 static struct relay_conn_queue relay_conn_queue;
 158
 159 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
 160 static unsigned int lttng_opt_fd_pool_size = -1;
 161
 162 /* Global relay stream hash table. */
 163 struct lttng_ht *relay_streams_ht;
 164
 165 /* Global relay viewer stream hash table. */
 166 struct lttng_ht *viewer_streams_ht;
 167
 168 /* Global relay sessions hash table. */
 169 struct lttng_ht *sessions_ht;
 170
 171 /* Relayd health monitoring */
 172 struct health_app *health_relayd;
 173
 174 struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
 175
 176 /* Global fd tracker. */
 177 struct fd_tracker *the_fd_tracker;
 178
 179 static struct option long_options[] = {
 180         {
 181                 "control-port",
 182                 1,
 183                 nullptr,
 184                 'C',
 185         },
 186         {
 187                 "data-port",
 188                 1,
 189                 nullptr,
 190                 'D',
 191         },
 192         {
 193                 "live-port",
 194                 1,
 195                 nullptr,
 196                 'L',
 197         },
 198         {
 199                 "daemonize",
 200                 0,
 201                 nullptr,
 202                 'd',
 203         },
 204         {
 205                 "background",
 206                 0,
 207                 nullptr,
 208                 'b',
 209         },
 210         {
 211                 "group",
 212                 1,
 213                 nullptr,
 214                 'g',
 215         },
 216         {
 217                 "fd-pool-size",
 218                 1,
 219                 nullptr,
 220                 '\0',
 221         },
 222         {
 223                 "help",
 224                 0,
 225                 nullptr,
 226                 'h',
 227         },
 228         {
 229                 "output",
 230                 1,
 231                 nullptr,
 232                 'o',
 233         },
 234         {
 235                 "verbose",
 236                 0,
 237                 nullptr,
 238                 'v',
 239         },
 240         { "config", 1, nullptr, 'f' },
 241         { "version", 0, nullptr, 'V' },
 242         {
 243                 "working-directory",
 244                 1,
 245                 nullptr,
 246                 'w',
 247         },
 248         {
 249                 "group-output-by-session",
 250                 0,
 251                 nullptr,
 252                 's',
 253         },
 254         {
 255                 "group-output-by-host",
 256                 0,
 257                 nullptr,
 258                 'p',
 259         },
 260         { "disallow-clear", 0, nullptr, 'x' },
 261         {
 262                 nullptr,
 263                 0,
 264                 nullptr,
 265                 0,
 266         },
 267 };
 268
 269 static const char *config_ignore_options[] = { "help", "config", "version" };
 270
 271 static void print_version()
 272 {
 273         fprintf(stdout, "%s\n", VERSION);
 274 }
 275
 276 static void relayd_config_log()
 277 {
 278         DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
 279             GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
 280             EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
 281         if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
 282                 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
 283         }
 284         if (EXTRA_VERSION_PATCHES[0] != '\0') {
 285                 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
 286         }
 287 }
 288
 289 /*
 290  * Take an option from the getopt output and set it in the right variable to be
 291  * used later.
 292  *
 293  * Return 0 on success else a negative value.
 294  */
 295 static int set_option(int opt, const char *arg, const char *optname)
 296 {
 297         int ret;
 298
 299         switch (opt) {
 300         case 0:
 301                 if (!strcmp(optname, "fd-pool-size")) {
 302                         unsigned long v;
 303
 304                         errno = 0;
 305                         v = strtoul(arg, nullptr, 0);
 306                         if (errno != 0 || !isdigit((unsigned char) arg[0])) {
 307                                 ERR("Wrong value in --fd-pool-size parameter: %s", arg);
 308                                 ret = -1;
 309                                 goto end;
 310                         }
 311                         if (v >= UINT_MAX) {
 312                                 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s",
 313                                     arg);
 314                                 ret = -1;
 315                                 goto end;
 316                         }
 317                         lttng_opt_fd_pool_size = (unsigned int) v;
 318                 } else {
 319                         fprintf(stderr, "unknown option %s", optname);
 320                         if (arg) {
 321                                 fprintf(stderr, " with arg %s\n", arg);
 322                         }
 323                 }
 324                 break;
 325         case 'C':
 326                 if (lttng_is_setuid_setgid()) {
 327                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 328                              "-C, --control-port");
 329                 } else {
 330                         ret = uri_parse(arg, &control_uri);
 331                         if (ret < 0) {
 332                                 ERR("Invalid control URI specified");
 333                                 goto end;
 334                         }
 335                         if (control_uri->port == 0) {
 336                                 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
 337                         }
 338                 }
 339                 break;
 340         case 'D':
 341                 if (lttng_is_setuid_setgid()) {
 342                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 343                              "-D, -data-port");
 344                 } else {
 345                         ret = uri_parse(arg, &data_uri);
 346                         if (ret < 0) {
 347                                 ERR("Invalid data URI specified");
 348                                 goto end;
 349                         }
 350                         if (data_uri->port == 0) {
 351                                 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
 352                         }
 353                 }
 354                 break;
 355         case 'L':
 356                 if (lttng_is_setuid_setgid()) {
 357                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 358                              "-L, -live-port");
 359                 } else {
 360                         ret = uri_parse(arg, &live_uri);
 361                         if (ret < 0) {
 362                                 ERR("Invalid live URI specified");
 363                                 goto end;
 364                         }
 365                         if (live_uri->port == 0) {
 366                                 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
 367                         }
 368                 }
 369                 break;
 370         case 'd':
 371                 opt_daemon = 1;
 372                 break;
 373         case 'b':
 374                 opt_background = 1;
 375                 break;
 376         case 'g':
 377                 if (lttng_is_setuid_setgid()) {
 378                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 379                              "-g, --group");
 380                 } else {
 381                         tracing_group_name = strdup(arg);
 382                         if (tracing_group_name == nullptr) {
 383                                 ret = -errno;
 384                                 PERROR("strdup");
 385                                 goto end;
 386                         }
 387                         tracing_group_name_override = 1;
 388                 }
 389                 break;
 390         case 'h':
 391                 ret = utils_show_help(8, "lttng-relayd", help_msg);
 392                 if (ret) {
 393                         ERR("Cannot show --help for `lttng-relayd`");
 394                         perror("exec");
 395                 }
 396                 exit(EXIT_FAILURE);
 397         case 'V':
 398                 opt_print_version = 1;
 399                 break;
 400         case 'o':
 401                 if (lttng_is_setuid_setgid()) {
 402                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 403                              "-o, --output");
 404                 } else {
 405                         ret = asprintf(&opt_output_path, "%s", arg);
 406                         if (ret < 0) {
 407                                 ret = -errno;
 408                                 PERROR("asprintf opt_output_path");
 409                                 goto end;
 410                         }
 411                 }
 412                 break;
 413         case 'w':
 414                 if (lttng_is_setuid_setgid()) {
 415                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 416                              "-w, --working-directory");
 417                 } else {
 418                         ret = asprintf(&opt_working_directory, "%s", arg);
 419                         if (ret < 0) {
 420                                 ret = -errno;
 421                                 PERROR("asprintf opt_working_directory");
 422                                 goto end;
 423                         }
 424                 }
 425                 break;
 426
 427         case 'v':
 428                 /* Verbose level can increase using multiple -v */
 429                 if (arg) {
 430                         lttng_opt_verbose = config_parse_value(arg);
 431                 } else {
 432                         /* Only 3 level of verbosity (-vvv). */
 433                         if (lttng_opt_verbose < 3) {
 434                                 lttng_opt_verbose += 1;
 435                         }
 436                 }
 437                 break;
 438         case 's':
 439                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 440                         ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
 441                         exit(EXIT_FAILURE);
 442                 }
 443                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
 444                 break;
 445         case 'p':
 446                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 447                         ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
 448                         exit(EXIT_FAILURE);
 449                 }
 450                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 451                 break;
 452         case 'x':
 453                 /* Disallow clear */
 454                 opt_allow_clear = 0;
 455                 break;
 456         default:
 457                 /* Unknown option or other error.
 458                  * Error is printed by getopt, just return */
 459                 ret = -1;
 460                 goto end;
 461         }
 462
 463         /* All good. */
 464         ret = 0;
 465
 466 end:
 467         return ret;
 468 }
 469
 470 /*
 471  * config_entry_handler_cb used to handle options read from a config file.
 472  * See config_entry_handler_cb comment in common/config/session-config.h for the
 473  * return value conventions.
 474  */
 475 static int config_entry_handler(const struct config_entry *entry,
 476                                 void *unused __attribute__((unused)))
 477 {
 478         int ret = 0, i;
 479
 480         if (!entry || !entry->name || !entry->value) {
 481                 ret = -EINVAL;
 482                 goto end;
 483         }
 484
 485         /* Check if the option is to be ignored */
 486         for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
 487                 if (!strcmp(entry->name, config_ignore_options[i])) {
 488                         goto end;
 489                 }
 490         }
 491
 492         for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
 493                 /* Ignore if entry name is not fully matched. */
 494                 if (strcmp(entry->name, long_options[i].name) != 0) {
 495                         continue;
 496                 }
 497
 498                 /*
 499                  * If the option takes no argument on the command line,
 500                  * we have to check if the value is "true". We support
 501                  * non-zero numeric values, true, on and yes.
 502                  */
 503                 if (!long_options[i].has_arg) {
 504                         ret = config_parse_value(entry->value);
 505                         if (ret <= 0) {
 506                                 if (ret) {
 507                                         WARN("Invalid configuration value \"%s\" for option %s",
 508                                              entry->value,
 509                                              entry->name);
 510                                 }
 511                                 /* False, skip boolean config option. */
 512                                 goto end;
 513                         }
 514                 }
 515
 516                 ret = set_option(long_options[i].val, entry->value, entry->name);
 517                 goto end;
 518         }
 519
 520         WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name);
 521
 522 end:
 523         return ret;
 524 }
 525
 526 static int parse_env_options()
 527 {
 528         int ret = 0;
 529         char *value = nullptr;
 530
 531         value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
 532         if (value) {
 533                 opt_working_directory = strdup(value);
 534                 if (!opt_working_directory) {
 535                         ERR("Failed to allocate working directory string (\"%s\")", value);
 536                         ret = -1;
 537                 }
 538         }
 539         return ret;
 540 }
 541
 542 static int set_fd_pool_size()
 543 {
 544         int ret = 0;
 545         struct rlimit rlimit;
 546
 547         ret = getrlimit(RLIMIT_NOFILE, &rlimit);
 548         if (ret) {
 549                 PERROR("Failed to get file descriptor limit");
 550                 ret = -1;
 551                 goto end;
 552         }
 553
 554         DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
 555             (uint64_t) rlimit.rlim_cur,
 556             (uint64_t) rlimit.rlim_max);
 557         if (lttng_opt_fd_pool_size == -1) {
 558                 /* Use default value (soft limit - reserve). */
 559                 if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 560                         ERR("The process' file number limit is too low (%" PRIu64
 561                             "). The process' file number limit must be set to at least %i.",
 562                             (uint64_t) rlimit.rlim_cur,
 563                             DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 564                         ret = -1;
 565                         goto end;
 566                 }
 567                 lttng_opt_fd_pool_size = rlimit.rlim_cur - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 568                 goto end;
 569         }
 570
 571         if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 572                 ERR("File descriptor pool size must be set to at least %d",
 573                     DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 574                 ret = -1;
 575                 goto end;
 576         }
 577
 578         if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
 579                 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
 580                     ").",
 581                     lttng_opt_fd_pool_size,
 582                     (uint64_t) rlimit.rlim_cur);
 583                 ret = -1;
 584                 goto end;
 585         }
 586
 587         DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
 588             lttng_opt_fd_pool_size,
 589             lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
 590         lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 591 end:
 592         return ret;
 593 }
 594
 595 static int set_options(int argc, char **argv)
 596 {
 597         int c, ret = 0, option_index = 0, retval = 0;
 598         int orig_optopt = optopt, orig_optind = optind;
 599         char *default_address, *optstring;
 600         char *config_path = nullptr;
 601
 602         optstring = utils_generate_optstring(long_options,
 603                                              sizeof(long_options) / sizeof(struct option));
 604         if (!optstring) {
 605                 retval = -ENOMEM;
 606                 goto exit;
 607         }
 608
 609         /* Check for the --config option */
 610
 611         while ((c = getopt_long(argc, argv, optstring, long_options, &option_index)) != -1) {
 612                 if (c == '?') {
 613                         retval = -EINVAL;
 614                         goto exit;
 615                 } else if (c != 'f') {
 616                         continue;
 617                 }
 618
 619                 if (lttng_is_setuid_setgid()) {
 620                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 621                              "-f, --config");
 622                 } else {
 623                         free(config_path);
 624                         config_path = utils_expand_path(optarg);
 625                         if (!config_path) {
 626                                 ERR("Failed to resolve path: %s", optarg);
 627                         }
 628                 }
 629         }
 630
 631         ret = config_get_section_entries(
 632                 config_path, config_section_name, config_entry_handler, nullptr);
 633         if (ret) {
 634                 if (ret > 0) {
 635                         ERR("Invalid configuration option at line %i", ret);
 636                 }
 637                 retval = -1;
 638                 goto exit;
 639         }
 640
 641         /* Reset getopt's global state */
 642         optopt = orig_optopt;
 643         optind = orig_optind;
 644         while (true) {
 645                 c = getopt_long(argc, argv, optstring, long_options, &option_index);
 646                 if (c == -1) {
 647                         break;
 648                 }
 649
 650                 ret = set_option(c, optarg, long_options[option_index].name);
 651                 if (ret < 0) {
 652                         retval = -1;
 653                         goto exit;
 654                 }
 655         }
 656
 657         /* assign default values */
 658         if (control_uri == nullptr) {
 659                 ret = asprintf(&default_address,
 660                                "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
 661                                DEFAULT_NETWORK_CONTROL_PORT);
 662                 if (ret < 0) {
 663                         PERROR("asprintf default data address");
 664                         retval = -1;
 665                         goto exit;
 666                 }
 667
 668                 ret = uri_parse(default_address, &control_uri);
 669                 free(default_address);
 670                 if (ret < 0) {
 671                         ERR("Invalid control URI specified");
 672                         retval = -1;
 673                         goto exit;
 674                 }
 675         }
 676         if (data_uri == nullptr) {
 677                 ret = asprintf(&default_address,
 678                                "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
 679                                DEFAULT_NETWORK_DATA_PORT);
 680                 if (ret < 0) {
 681                         PERROR("asprintf default data address");
 682                         retval = -1;
 683                         goto exit;
 684                 }
 685
 686                 ret = uri_parse(default_address, &data_uri);
 687                 free(default_address);
 688                 if (ret < 0) {
 689                         ERR("Invalid data URI specified");
 690                         retval = -1;
 691                         goto exit;
 692                 }
 693         }
 694         if (live_uri == nullptr) {
 695                 ret = asprintf(&default_address,
 696                                "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
 697                                DEFAULT_NETWORK_VIEWER_PORT);
 698                 if (ret < 0) {
 699                         PERROR("asprintf default viewer control address");
 700                         retval = -1;
 701                         goto exit;
 702                 }
 703
 704                 ret = uri_parse(default_address, &live_uri);
 705                 free(default_address);
 706                 if (ret < 0) {
 707                         ERR("Invalid viewer control URI specified");
 708                         retval = -1;
 709                         goto exit;
 710                 }
 711         }
 712         ret = set_fd_pool_size();
 713         if (ret) {
 714                 retval = -1;
 715                 goto exit;
 716         }
 717
 718         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 719                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 720         }
 721         if (opt_allow_clear) {
 722                 /* Check if env variable exists. */
 723                 const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 724                 if (value) {
 725                         ret = config_parse_value(value);
 726                         if (ret < 0) {
 727                                 ERR("Invalid value for %s specified",
 728                                     DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 729                                 retval = -1;
 730                                 goto exit;
 731                         }
 732                         opt_allow_clear = !ret;
 733                 }
 734         }
 735
 736 exit:
 737         free(config_path);
 738         free(optstring);
 739         return retval;
 740 }
 741
 742 static void print_global_objects()
 743 {
 744         print_viewer_streams();
 745         print_relay_streams();
 746         print_sessions();
 747 }
 748
 749 static int noop_close(void *data __attribute__((unused)), int *fds __attribute__((unused)))
 750 {
 751         return 0;
 752 }
 753
 754 static void untrack_stdio()
 755 {
 756         int fds[] = { fileno(stdout), fileno(stderr) };
 757
 758         /*
 759          * noop_close is used since we don't really want to close
 760          * the stdio output fds; we merely want to stop tracking them.
 761          */
 762         (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker, fds, 2, noop_close, nullptr);
 763 }
 764
 765 /*
 766  * Cleanup the daemon
 767  */
 768 static void relayd_cleanup()
 769 {
 770         print_global_objects();
 771
 772         DBG("Cleaning up");
 773
 774         if (viewer_streams_ht)
 775                 lttng_ht_destroy(viewer_streams_ht);
 776         if (relay_streams_ht)
 777                 lttng_ht_destroy(relay_streams_ht);
 778         if (sessions_ht)
 779                 lttng_ht_destroy(sessions_ht);
 780
 781         free(opt_output_path);
 782         free(opt_working_directory);
 783
 784         if (health_relayd) {
 785                 health_app_destroy(health_relayd);
 786         }
 787         /* Close thread quit pipes */
 788         if (health_quit_pipe[0] != -1) {
 789                 (void) fd_tracker_util_pipe_close(the_fd_tracker, health_quit_pipe);
 790         }
 791         relayd_close_thread_quit_pipe();
 792         if (sessiond_trace_chunk_registry) {
 793                 sessiond_trace_chunk_registry_destroy(sessiond_trace_chunk_registry);
 794         }
 795         if (the_fd_tracker) {
 796                 untrack_stdio();
 797                 /*
 798                  * fd_tracker_destroy() will log the contents of the fd-tracker
 799                  * if a leak is detected.
 800                  */
 801                 fd_tracker_destroy(the_fd_tracker);
 802         }
 803
 804         uri_free(control_uri);
 805         uri_free(data_uri);
 806         /* Live URI is freed in the live thread. */
 807
 808         if (tracing_group_name_override) {
 809                 free((void *) tracing_group_name);
 810         }
 811 }
 812
 813 static int notify_health_quit_pipe(int *pipe)
 814 {
 815         ssize_t ret;
 816
 817         ret = lttng_write(pipe[1], "4", 1);
 818         if (ret < 1) {
 819                 PERROR("write relay health quit");
 820                 goto end;
 821         }
 822         ret = 0;
 823 end:
 824         return ret;
 825 }
 826
 827 /*
 828  * Stop all relayd and relayd-live threads.
 829  */
 830 int lttng_relay_stop_threads()
 831 {
 832         int retval = 0;
 833
 834         /* Stopping all threads */
 835         DBG("Terminating all threads");
 836         if (relayd_notify_thread_quit_pipe()) {
 837                 ERR("write error on thread quit pipe");
 838                 retval = -1;
 839         }
 840
 841         if (notify_health_quit_pipe(health_quit_pipe)) {
 842                 ERR("write error on health quit pipe");
 843         }
 844
 845         /* Dispatch thread */
 846         CMM_STORE_SHARED(dispatch_thread_exit, 1);
 847         futex_nto1_wake(&relay_conn_queue.futex);
 848
 849         if (relayd_live_stop()) {
 850                 ERR("Error stopping live threads");
 851                 retval = -1;
 852         }
 853         return retval;
 854 }
 855
 856 /*
 857  * Signal handler for the daemon
 858  *
 859  * Simply stop all worker threads, leaving main() return gracefully after
 860  * joining all threads and calling cleanup().
 861  */
 862 static void sighandler(int sig)
 863 {
 864         switch (sig) {
 865         case SIGINT:
 866                 DBG("SIGINT caught");
 867                 if (lttng_relay_stop_threads()) {
 868                         ERR("Error stopping threads");
 869                 }
 870                 break;
 871         case SIGTERM:
 872                 DBG("SIGTERM caught");
 873                 if (lttng_relay_stop_threads()) {
 874                         ERR("Error stopping threads");
 875                 }
 876                 break;
 877         case SIGUSR1:
 878                 CMM_STORE_SHARED(recv_child_signal, 1);
 879                 break;
 880         default:
 881                 break;
 882         }
 883 }
 884
 885 /*
 886  * Setup signal handler for :
 887  *              SIGINT, SIGTERM, SIGPIPE
 888  */
 889 static int set_signal_handler()
 890 {
 891         int ret = 0;
 892         struct sigaction sa;
 893         sigset_t sigset;
 894
 895         if ((ret = sigemptyset(&sigset)) < 0) {
 896                 PERROR("sigemptyset");
 897                 return ret;
 898         }
 899
 900         sa.sa_mask = sigset;
 901         sa.sa_flags = 0;
 902
 903         sa.sa_handler = sighandler;
 904         if ((ret = sigaction(SIGTERM, &sa, nullptr)) < 0) {
 905                 PERROR("sigaction");
 906                 return ret;
 907         }
 908
 909         if ((ret = sigaction(SIGINT, &sa, nullptr)) < 0) {
 910                 PERROR("sigaction");
 911                 return ret;
 912         }
 913
 914         if ((ret = sigaction(SIGUSR1, &sa, nullptr)) < 0) {
 915                 PERROR("sigaction");
 916                 return ret;
 917         }
 918
 919         sa.sa_handler = SIG_IGN;
 920         if ((ret = sigaction(SIGPIPE, &sa, nullptr)) < 0) {
 921                 PERROR("sigaction");
 922                 return ret;
 923         }
 924
 925         DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
 926
 927         return ret;
 928 }
 929
 930 void lttng_relay_notify_ready()
 931 {
 932         /* Notify the parent of the fork() process that we are ready. */
 933         if (opt_daemon || opt_background) {
 934                 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
 935                         kill(child_ppid, SIGUSR1);
 936                 }
 937         }
 938 }
 939
 940 /*
 941  * Init health quit pipe.
 942  *
 943  * Return -1 on error or 0 if all pipes are created.
 944  */
 945 static int init_health_quit_pipe()
 946 {
 947         return fd_tracker_util_pipe_open_cloexec(
 948                 the_fd_tracker, "Health quit pipe", health_quit_pipe);
 949 }
 950
 951 static int create_sock(void *data, int *out_fd)
 952 {
 953         int ret;
 954         struct lttcomm_sock *sock = (lttcomm_sock *) data;
 955
 956         ret = lttcomm_create_sock(sock);
 957         if (ret < 0) {
 958                 goto end;
 959         }
 960
 961         *out_fd = sock->fd;
 962 end:
 963         return ret;
 964 }
 965
 966 static int close_sock(void *data, int *in_fd __attribute__((unused)))
 967 {
 968         struct lttcomm_sock *sock = (lttcomm_sock *) data;
 969
 970         return sock->ops->close(sock);
 971 }
 972
 973 static int accept_sock(void *data, int *out_fd)
 974 {
 975         int ret = 0;
 976         /* Socks is an array of in_sock, out_sock. */
 977         struct lttcomm_sock **socks = (lttcomm_sock **) data;
 978         struct lttcomm_sock *in_sock = socks[0];
 979
 980         socks[1] = in_sock->ops->accept(in_sock);
 981         if (!socks[1]) {
 982                 ret = -1;
 983                 goto end;
 984         }
 985         *out_fd = socks[1]->fd;
 986 end:
 987         return ret;
 988 }
 989
 990 /*
 991  * Create and init socket from uri.
 992  */
 993 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri, const char *name)
 994 {
 995         int ret, sock_fd;
 996         struct lttcomm_sock *sock = nullptr;
 997         char uri_str[PATH_MAX];
 998         char *formated_name = nullptr;
 999
1000         sock = lttcomm_alloc_sock_from_uri(uri);
1001         if (sock == nullptr) {
1002                 ERR("Allocating socket");
1003                 goto error;
1004         }
1005
1006         /*
1007          * Don't fail to create the socket if the name can't be built as it is
1008          * only used for debugging purposes.
1009          */
1010         ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
1011         uri_str[sizeof(uri_str) - 1] = '\0';
1012         if (ret >= 0) {
1013                 ret = asprintf(&formated_name, "%s socket @ %s", name, uri_str);
1014                 if (ret < 0) {
1015                         formated_name = nullptr;
1016                 }
1017         }
1018
1019         ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker,
1020                                                &sock_fd,
1021                                                (const char **) (formated_name ? &formated_name :
1022                                                                                 nullptr),
1023                                                1,
1024                                                create_sock,
1025                                                sock);
1026         if (ret) {
1027                 PERROR("Failed to open \"%s\" relay socket", 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         free(formated_name);
1044         return sock;
1045
1046 error:
1047         if (sock) {
1048                 lttcomm_destroy_sock(sock);
1049         }
1050         free(formated_name);
1051         return nullptr;
1052 }
1053
1054 static struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1055                                                const char *name)
1056 {
1057         int out_fd, ret;
1058         struct lttcomm_sock *socks[2] = { listening_sock, nullptr };
1059         struct lttcomm_sock *new_sock = nullptr;
1060
1061         ret = fd_tracker_open_unsuspendable_fd(
1062                 the_fd_tracker, &out_fd, (const char **) &name, 1, accept_sock, &socks);
1063         if (ret) {
1064                 goto end;
1065         }
1066         new_sock = socks[1];
1067         DBG("%s accepted, socket %d", name, new_sock->fd);
1068 end:
1069         return new_sock;
1070 }
1071
1072 /*
1073  * This thread manages the listening for new connections on the network
1074  */
1075 static void *relay_thread_listener(void *data __attribute__((unused)))
1076 {
1077         int i, ret, err = -1;
1078         uint32_t nb_fd;
1079         struct lttng_poll_event events;
1080         struct lttcomm_sock *control_sock, *data_sock;
1081
1082         DBG("[thread] Relay listener started");
1083
1084         rcu_register_thread();
1085         health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1086
1087         health_code_update();
1088
1089         control_sock = relay_socket_create(control_uri, "Control listener");
1090         if (!control_sock) {
1091                 goto error_sock_control;
1092         }
1093
1094         data_sock = relay_socket_create(data_uri, "Data listener");
1095         if (!data_sock) {
1096                 goto error_sock_relay;
1097         }
1098
1099         /*
1100          * Pass 3 as size here for the thread quit pipe, control and
1101          * data socket.
1102          */
1103         ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1104         if (ret < 0) {
1105                 goto error_create_poll;
1106         }
1107
1108         /* Add the control socket */
1109         ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1110         if (ret < 0) {
1111                 goto error_poll_add;
1112         }
1113
1114         /* Add the data socket */
1115         ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1116         if (ret < 0) {
1117                 goto error_poll_add;
1118         }
1119
1120         lttng_relay_notify_ready();
1121
1122         if (testpoint(relayd_thread_listener)) {
1123                 goto error_testpoint;
1124         }
1125
1126         while (true) {
1127                 health_code_update();
1128
1129                 DBG("Listener accepting connections");
1130
1131         restart:
1132                 health_poll_entry();
1133                 ret = lttng_poll_wait(&events, -1);
1134                 health_poll_exit();
1135                 if (ret < 0) {
1136                         /*
1137                          * Restart interrupted system call.
1138                          */
1139                         if (errno == EINTR) {
1140                                 goto restart;
1141                         }
1142                         goto error;
1143                 }
1144
1145                 nb_fd = ret;
1146
1147                 DBG("Relay new connection received");
1148                 for (i = 0; i < nb_fd; i++) {
1149                         /* Fetch once the poll data */
1150                         const auto revents = LTTNG_POLL_GETEV(&events, i);
1151                         const auto pollfd = LTTNG_POLL_GETFD(&events, i);
1152
1153                         health_code_update();
1154
1155                         /* Activity on thread quit pipe, exiting. */
1156                         if (relayd_is_thread_quit_pipe(pollfd)) {
1157                                 DBG("Activity on thread quit pipe");
1158                                 err = 0;
1159                                 goto exit;
1160                         }
1161
1162                         if (revents & LPOLLIN) {
1163                                 /*
1164                                  * A new connection is requested, therefore a
1165                                  * sessiond/consumerd connection is allocated in
1166                                  * this thread, enqueued to a global queue and
1167                                  * dequeued (and freed) in the worker thread.
1168                                  */
1169                                 int val = 1;
1170                                 struct relay_connection *new_conn;
1171                                 struct lttcomm_sock *newsock = nullptr;
1172                                 enum connection_type type;
1173
1174                                 if (pollfd == data_sock->fd) {
1175                                         type = RELAY_DATA;
1176                                         newsock = accept_relayd_sock(data_sock,
1177                                                                      "Data socket to relayd");
1178                                 } else {
1179                                         LTTNG_ASSERT(pollfd == control_sock->fd);
1180                                         type = RELAY_CONTROL;
1181                                         newsock = accept_relayd_sock(control_sock,
1182                                                                      "Control socket to relayd");
1183                                 }
1184                                 if (!newsock) {
1185                                         PERROR("accepting sock");
1186                                         goto error;
1187                                 }
1188
1189                                 ret = setsockopt(
1190                                         newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, sizeof(val));
1191                                 if (ret < 0) {
1192                                         PERROR("setsockopt inet");
1193                                         lttcomm_destroy_sock(newsock);
1194                                         goto error;
1195                                 }
1196
1197                                 ret = socket_apply_keep_alive_config(newsock->fd);
1198                                 if (ret < 0) {
1199                                         ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1200                                             newsock->fd);
1201                                         lttcomm_destroy_sock(newsock);
1202                                         goto error;
1203                                 }
1204
1205                                 new_conn = connection_create(newsock, type);
1206                                 if (!new_conn) {
1207                                         lttcomm_destroy_sock(newsock);
1208                                         goto error;
1209                                 }
1210
1211                                 /* Enqueue request for the dispatcher thread. */
1212                                 cds_wfcq_head_ptr_t head;
1213                                 head.h = &relay_conn_queue.head;
1214                                 cds_wfcq_enqueue(head, &relay_conn_queue.tail, &new_conn->qnode);
1215
1216                                 /*
1217                                  * Wake the dispatch queue futex.
1218                                  * Implicit memory barrier with the
1219                                  * exchange in cds_wfcq_enqueue.
1220                                  */
1221                                 futex_nto1_wake(&relay_conn_queue.futex);
1222                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1223                                 ERR("socket poll error");
1224                                 goto error;
1225                         } else {
1226                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1227                                 goto error;
1228                         }
1229                 }
1230         }
1231
1232 exit:
1233 error:
1234 error_poll_add:
1235 error_testpoint:
1236         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1237 error_create_poll:
1238         if (data_sock->fd >= 0) {
1239                 int data_sock_fd = data_sock->fd;
1240
1241                 ret = fd_tracker_close_unsuspendable_fd(
1242                         the_fd_tracker, &data_sock_fd, 1, close_sock, data_sock);
1243                 if (ret) {
1244                         PERROR("Failed to close the data listener socket file descriptor");
1245                 }
1246                 data_sock->fd = -1;
1247         }
1248         lttcomm_destroy_sock(data_sock);
1249 error_sock_relay:
1250         if (control_sock->fd >= 0) {
1251                 int control_sock_fd = control_sock->fd;
1252
1253                 ret = fd_tracker_close_unsuspendable_fd(
1254                         the_fd_tracker, &control_sock_fd, 1, close_sock, control_sock);
1255                 if (ret) {
1256                         PERROR("Failed to close the control listener socket file descriptor");
1257                 }
1258                 control_sock->fd = -1;
1259         }
1260         lttcomm_destroy_sock(control_sock);
1261 error_sock_control:
1262         if (err) {
1263                 health_error();
1264                 ERR("Health error occurred in %s", __func__);
1265         }
1266         health_unregister(health_relayd);
1267         rcu_unregister_thread();
1268         DBG("Relay listener thread cleanup complete");
1269         lttng_relay_stop_threads();
1270         return nullptr;
1271 }
1272
1273 /*
1274  * This thread manages the dispatching of the requests to worker threads
1275  */
1276 static void *relay_thread_dispatcher(void *data __attribute__((unused)))
1277 {
1278         int err = -1;
1279         ssize_t ret;
1280         struct cds_wfcq_node *node;
1281         struct relay_connection *new_conn = nullptr;
1282
1283         DBG("[thread] Relay dispatcher started");
1284
1285         health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1286
1287         if (testpoint(relayd_thread_dispatcher)) {
1288                 goto error_testpoint;
1289         }
1290
1291         health_code_update();
1292
1293         for (;;) {
1294                 health_code_update();
1295
1296                 /* Atomically prepare the queue futex */
1297                 futex_nto1_prepare(&relay_conn_queue.futex);
1298
1299                 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1300                         break;
1301                 }
1302
1303                 do {
1304                         health_code_update();
1305
1306                         /* Dequeue commands */
1307                         node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1308                                                          &relay_conn_queue.tail);
1309                         if (node == nullptr) {
1310                                 DBG("Woken up but nothing in the relay command queue");
1311                                 /* Continue thread execution */
1312                                 break;
1313                         }
1314                         new_conn = lttng::utils::container_of(node, &relay_connection::qnode);
1315
1316                         DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1317
1318                         /*
1319                          * Inform worker thread of the new request. This
1320                          * call is blocking so we can be assured that
1321                          * the data will be read at some point in time
1322                          * or wait to the end of the world :)
1323                          */
1324                         ret = lttng_write(
1325                                 relay_conn_pipe[1], &new_conn, sizeof(new_conn)); /* NOLINT
1326                                                                                      sizeof
1327                                                                                      used
1328                                                                                      on a
1329                                                                                      pointer.
1330                                                                                    */
1331                         if (ret < 0) {
1332                                 PERROR("write connection pipe");
1333                                 connection_put(new_conn);
1334                                 goto error;
1335                         }
1336                 } while (node != nullptr);
1337
1338                 /* Futex wait on queue. Blocking call on futex() */
1339                 health_poll_entry();
1340                 futex_nto1_wait(&relay_conn_queue.futex);
1341                 health_poll_exit();
1342         }
1343
1344         /* Normal exit, no error */
1345         err = 0;
1346
1347 error:
1348 error_testpoint:
1349         if (err) {
1350                 health_error();
1351                 ERR("Health error occurred in %s", __func__);
1352         }
1353         health_unregister(health_relayd);
1354         DBG("Dispatch thread dying");
1355         lttng_relay_stop_threads();
1356         return nullptr;
1357 }
1358
1359 static bool session_streams_have_index(const struct relay_session *session)
1360 {
1361         return session->minor >= 4 && !session->snapshot;
1362 }
1363
1364 /*
1365  * Handle the RELAYD_CREATE_SESSION command.
1366  *
1367  * On success, send back the session id or else return a negative value.
1368  */
1369 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1370                                 struct relay_connection *conn,
1371                                 const struct lttng_buffer_view *payload)
1372 {
1373         int ret = 0;
1374         ssize_t send_ret;
1375         struct relay_session *session = nullptr;
1376         struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1377         char session_name[LTTNG_NAME_MAX] = {};
1378         char hostname[LTTNG_HOST_NAME_MAX] = {};
1379         uint32_t live_timer = 0;
1380         bool snapshot = false;
1381         bool session_name_contains_creation_timestamp = false;
1382         /* Left nil for peers < 2.11. */
1383         char base_path[LTTNG_PATH_MAX] = {};
1384         lttng_uuid sessiond_uuid = {};
1385         LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1386         LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1387         LTTNG_OPTIONAL(time_t) creation_time = {};
1388         struct lttng_dynamic_buffer reply_payload;
1389
1390         lttng_dynamic_buffer_init(&reply_payload);
1391
1392         if (conn->minor < 4) {
1393                 /* From 2.1 to 2.3 */
1394                 ret = 0;
1395         } else if (conn->minor >= 4 && conn->minor < 11) {
1396                 /* From 2.4 to 2.10 */
1397                 ret = cmd_create_session_2_4(
1398                         payload, session_name, hostname, &live_timer, &snapshot);
1399         } else {
1400                 bool has_current_chunk;
1401                 uint64_t current_chunk_id_value;
1402                 time_t creation_time_value;
1403                 uint64_t id_sessiond_value;
1404
1405                 /* From 2.11 to ... */
1406                 ret = cmd_create_session_2_11(payload,
1407                                               session_name,
1408                                               hostname,
1409                                               base_path,
1410                                               &live_timer,
1411                                               &snapshot,
1412                                               &id_sessiond_value,
1413                                               sessiond_uuid,
1414                                               &has_current_chunk,
1415                                               &current_chunk_id_value,
1416                                               &creation_time_value,
1417                                               &session_name_contains_creation_timestamp);
1418                 if (lttng_uuid_is_nil(sessiond_uuid)) {
1419                         /* The nil UUID is reserved for pre-2.11 clients. */
1420                         ERR("Illegal nil UUID announced by peer in create session command");
1421                         ret = -1;
1422                         goto send_reply;
1423                 }
1424                 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1425                 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1426                 if (has_current_chunk) {
1427                         LTTNG_OPTIONAL_SET(&current_chunk_id, current_chunk_id_value);
1428                 }
1429         }
1430
1431         if (ret < 0) {
1432                 goto send_reply;
1433         }
1434
1435         session = session_create(session_name,
1436                                  hostname,
1437                                  base_path,
1438                                  live_timer,
1439                                  snapshot,
1440                                  sessiond_uuid,
1441                                  id_sessiond.is_set ? &id_sessiond.value : nullptr,
1442                                  current_chunk_id.is_set ? &current_chunk_id.value : nullptr,
1443                                  creation_time.is_set ? &creation_time.value : nullptr,
1444                                  conn->major,
1445                                  conn->minor,
1446                                  session_name_contains_creation_timestamp);
1447         if (!session) {
1448                 ret = -1;
1449                 goto send_reply;
1450         }
1451         LTTNG_ASSERT(!conn->session);
1452         conn->session = session;
1453         DBG("Created session %" PRIu64, session->id);
1454
1455         reply.generic.session_id = htobe64(session->id);
1456
1457 send_reply:
1458         if (ret < 0) {
1459                 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1460         } else {
1461                 reply.generic.ret_code = htobe32(LTTNG_OK);
1462         }
1463
1464         if (conn->minor < 11) {
1465                 /* From 2.1 to 2.10 */
1466                 ret = lttng_dynamic_buffer_append(
1467                         &reply_payload, &reply.generic, sizeof(reply.generic));
1468                 if (ret) {
1469                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1470                         ret = -1;
1471                         goto end;
1472                 }
1473         } else {
1474                 const uint32_t output_path_length = session ? strlen(session->output_path) + 1 : 0;
1475
1476                 reply.output_path_length = htobe32(output_path_length);
1477                 ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
1478                 if (ret) {
1479                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1480                         goto end;
1481                 }
1482
1483                 if (output_path_length) {
1484                         ret = lttng_dynamic_buffer_append(
1485                                 &reply_payload, session->output_path, output_path_length);
1486                         if (ret) {
1487                                 ERR("Failed to append \"create session\" command reply path to payload buffer");
1488                                 goto end;
1489                         }
1490                 }
1491         }
1492
1493         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
1494         if (send_ret < (ssize_t) reply_payload.size) {
1495                 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1496                     reply_payload.size,
1497                     send_ret);
1498                 ret = -1;
1499         }
1500 end:
1501         if (ret < 0 && session) {
1502                 session_put(session);
1503         }
1504         lttng_dynamic_buffer_reset(&reply_payload);
1505         return ret;
1506 }
1507
1508 /*
1509  * When we have received all the streams and the metadata for a channel,
1510  * we make them visible to the viewer threads.
1511  */
1512 static void publish_connection_local_streams(struct relay_connection *conn)
1513 {
1514         struct relay_stream *stream;
1515         struct relay_session *session = conn->session;
1516
1517         /*
1518          * We publish all streams belonging to a session atomically wrt
1519          * session lock.
1520          */
1521         pthread_mutex_lock(&session->lock);
1522         lttng::urcu::read_lock_guard read_lock;
1523         cds_list_for_each_entry_rcu(stream, &session->recv_list, recv_node)
1524         {
1525                 stream_publish(stream);
1526         }
1527
1528         /*
1529          * Inform the viewer that there are new streams in the session.
1530          */
1531         if (session->viewer_attached) {
1532                 uatomic_set(&session->new_streams, 1);
1533         }
1534
1535         pthread_mutex_unlock(&session->lock);
1536 }
1537
1538 static int conform_channel_path(char *channel_path)
1539 {
1540         int ret = 0;
1541
1542         if (strstr("../", channel_path)) {
1543                 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1544                     channel_path);
1545                 ret = -1;
1546                 goto end;
1547         }
1548
1549         if (*channel_path == '/') {
1550                 const size_t len = strlen(channel_path);
1551
1552                 /*
1553                  * Channel paths from peers prior to 2.11 are expressed as an
1554                  * absolute path that is, in reality, relative to the relay
1555                  * daemon's output directory. Remove the leading slash so it
1556                  * is correctly interpreted as a relative path later on.
1557                  *
1558                  * len (and not len - 1) is used to copy the trailing NULL.
1559                  */
1560                 bcopy(channel_path + 1, channel_path, len);
1561         }
1562 end:
1563         return ret;
1564 }
1565
1566 /*
1567  * relay_add_stream: allocate a new stream for a session
1568  */
1569 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1570                             struct relay_connection *conn,
1571                             const struct lttng_buffer_view *payload)
1572 {
1573         int ret;
1574         ssize_t send_ret;
1575         struct relay_session *session = conn->session;
1576         struct relay_stream *stream = nullptr;
1577         struct lttcomm_relayd_status_stream reply;
1578         struct ctf_trace *trace = nullptr;
1579         uint64_t stream_handle = -1ULL;
1580         char *path_name = nullptr, *channel_name = nullptr;
1581         uint64_t tracefile_size = 0, tracefile_count = 0;
1582         LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1583
1584         if (!session || !conn->version_check_done) {
1585                 ERR("Trying to add a stream before version check");
1586                 ret = -1;
1587                 goto end_no_session;
1588         }
1589
1590         if (session->minor == 1) {
1591                 /* For 2.1 */
1592                 ret = cmd_recv_stream_2_1(payload, &path_name, &channel_name);
1593         } else if (session->minor > 1 && session->minor < 11) {
1594                 /* From 2.2 to 2.10 */
1595                 ret = cmd_recv_stream_2_2(
1596                         payload, &path_name, &channel_name, &tracefile_size, &tracefile_count);
1597         } else {
1598                 /* From 2.11 to ... */
1599                 ret = cmd_recv_stream_2_11(payload,
1600                                            &path_name,
1601                                            &channel_name,
1602                                            &tracefile_size,
1603                                            &tracefile_count,
1604                                            &stream_chunk_id.value);
1605                 stream_chunk_id.is_set = true;
1606         }
1607
1608         if (ret < 0) {
1609                 goto send_reply;
1610         }
1611
1612         if (conform_channel_path(path_name)) {
1613                 goto send_reply;
1614         }
1615
1616         /*
1617          * Backward compatibility for --group-output-by-session.
1618          * Prior to lttng 2.11, the complete path is passed by the stream.
1619          * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1620          * >=2.11 the chunk is responsible for the output path. When dealing
1621          * with producer < 2.11 the chunk output_path is the root output path
1622          * and the stream carries the complete path (path_name).
1623          * To support --group-output-by-session with older producer (<2.11), we
1624          * need to craft the path based on the stream path.
1625          */
1626         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1627                 if (conn->minor < 4) {
1628                         /*
1629                          * From 2.1 to 2.3, the session_name is not passed on
1630                          * the RELAYD_CREATE_SESSION command. The session name
1631                          * is necessary to detect the presence of a base_path
1632                          * inside the stream path. Without it we cannot perform
1633                          * a valid group-output-by-session transformation.
1634                          */
1635                         WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1636                              " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1637                              session->id,
1638                              path_name);
1639                 } else if (conn->minor >= 4 && conn->minor < 11) {
1640                         char *group_by_session_path_name;
1641
1642                         LTTNG_ASSERT(session->session_name[0] != '\0');
1643
1644                         group_by_session_path_name = backward_compat_group_by_session(
1645                                 path_name, session->session_name, session->creation_time.value);
1646                         if (!group_by_session_path_name) {
1647                                 ERR("Failed to apply group by session to stream of session %" PRIu64,
1648                                     session->id);
1649                                 goto send_reply;
1650                         }
1651
1652                         DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1653                             path_name,
1654                             group_by_session_path_name);
1655
1656                         free(path_name);
1657                         path_name = group_by_session_path_name;
1658                 }
1659         }
1660
1661         trace = ctf_trace_get_by_path_or_create(session, path_name);
1662         if (!trace) {
1663                 goto send_reply;
1664         }
1665
1666         /* This stream here has one reference on the trace. */
1667         pthread_mutex_lock(&last_relay_stream_id_lock);
1668         stream_handle = ++last_relay_stream_id;
1669         pthread_mutex_unlock(&last_relay_stream_id_lock);
1670
1671         /* We pass ownership of path_name and channel_name. */
1672         stream = stream_create(
1673                 trace, stream_handle, path_name, channel_name, tracefile_size, tracefile_count);
1674         path_name = nullptr;
1675         channel_name = nullptr;
1676
1677         /*
1678          * Streams are the owners of their trace. Reference to trace is
1679          * kept within stream_create().
1680          */
1681         ctf_trace_put(trace);
1682
1683 send_reply:
1684         memset(&reply, 0, sizeof(reply));
1685         reply.handle = htobe64(stream_handle);
1686         if (!stream) {
1687                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1688         } else {
1689                 reply.ret_code = htobe32(LTTNG_OK);
1690         }
1691
1692         send_ret = conn->sock->ops->sendmsg(
1693                 conn->sock, &reply, sizeof(struct lttcomm_relayd_status_stream), 0);
1694         if (send_ret < (ssize_t) sizeof(reply)) {
1695                 ERR("Failed to send \"add stream\" command reply (ret = %zd)", send_ret);
1696                 ret = -1;
1697         }
1698
1699 end_no_session:
1700         free(path_name);
1701         free(channel_name);
1702         return ret;
1703 }
1704
1705 /*
1706  * relay_close_stream: close a specific stream
1707  */
1708 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1709                               struct relay_connection *conn,
1710                               const struct lttng_buffer_view *payload)
1711 {
1712         int ret;
1713         ssize_t send_ret;
1714         struct relay_session *session = conn->session;
1715         struct lttcomm_relayd_close_stream stream_info;
1716         struct lttcomm_relayd_generic_reply reply;
1717         struct relay_stream *stream;
1718
1719         DBG("Close stream received");
1720
1721         if (!session || !conn->version_check_done) {
1722                 ERR("Trying to close a stream before version check");
1723                 ret = -1;
1724                 goto end_no_session;
1725         }
1726
1727         if (payload->size < sizeof(stream_info)) {
1728                 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1729                     sizeof(stream_info),
1730                     payload->size);
1731                 ret = -1;
1732                 goto end_no_session;
1733         }
1734         memcpy(&stream_info, payload->data, sizeof(stream_info));
1735         stream_info.stream_id = be64toh(stream_info.stream_id);
1736         stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1737
1738         stream = stream_get_by_id(stream_info.stream_id);
1739         if (!stream) {
1740                 ret = -1;
1741                 goto end;
1742         }
1743
1744         /*
1745          * Set last_net_seq_num before the close flag. Required by data
1746          * pending check.
1747          */
1748         pthread_mutex_lock(&stream->lock);
1749         stream->last_net_seq_num = stream_info.last_net_seq_num;
1750         pthread_mutex_unlock(&stream->lock);
1751
1752         /*
1753          * This is one of the conditions which may trigger a stream close
1754          * with the others being:
1755          *     1) A close command is received for a stream
1756          *     2) The control connection owning the stream is closed
1757          *     3) We have received all of the stream's data _after_ a close
1758          *        request.
1759          */
1760         try_stream_close(stream);
1761         stream_put(stream);
1762         ret = 0;
1763
1764 end:
1765         memset(&reply, 0, sizeof(reply));
1766         if (ret < 0) {
1767                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1768         } else {
1769                 reply.ret_code = htobe32(LTTNG_OK);
1770         }
1771         send_ret = conn->sock->ops->sendmsg(
1772                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1773         if (send_ret < (ssize_t) sizeof(reply)) {
1774                 ERR("Failed to send \"close stream\" command reply (ret = %zd)", send_ret);
1775                 ret = -1;
1776         }
1777
1778 end_no_session:
1779         return ret;
1780 }
1781
1782 /*
1783  * relay_reset_metadata: reset a metadata stream
1784  */
1785 static int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1786                                 struct relay_connection *conn,
1787                                 const struct lttng_buffer_view *payload)
1788 {
1789         int ret;
1790         ssize_t send_ret;
1791         struct relay_session *session = conn->session;
1792         struct lttcomm_relayd_reset_metadata stream_info;
1793         struct lttcomm_relayd_generic_reply reply;
1794         struct relay_stream *stream;
1795
1796         DBG("Reset metadata received");
1797
1798         if (!session || !conn->version_check_done) {
1799                 ERR("Trying to reset a metadata stream before version check");
1800                 ret = -1;
1801                 goto end_no_session;
1802         }
1803
1804         if (payload->size < sizeof(stream_info)) {
1805                 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1806                     sizeof(stream_info),
1807                     payload->size);
1808                 ret = -1;
1809                 goto end_no_session;
1810         }
1811         memcpy(&stream_info, payload->data, sizeof(stream_info));
1812         stream_info.stream_id = be64toh(stream_info.stream_id);
1813         stream_info.version = be64toh(stream_info.version);
1814
1815         DBG("Update metadata to version %" PRIu64, stream_info.version);
1816
1817         /* Unsupported for live sessions for now. */
1818         if (session->live_timer != 0) {
1819                 ret = -1;
1820                 goto end;
1821         }
1822
1823         stream = stream_get_by_id(stream_info.stream_id);
1824         if (!stream) {
1825                 ret = -1;
1826                 goto end;
1827         }
1828         pthread_mutex_lock(&stream->lock);
1829         if (!stream->is_metadata) {
1830                 ret = -1;
1831                 goto end_unlock;
1832         }
1833
1834         ret = stream_reset_file(stream);
1835         if (ret < 0) {
1836                 ERR("Failed to reset metadata stream %" PRIu64 ": stream_path = %s, channel = %s",
1837                     stream->stream_handle,
1838                     stream->path_name,
1839                     stream->channel_name);
1840                 goto end_unlock;
1841         }
1842 end_unlock:
1843         pthread_mutex_unlock(&stream->lock);
1844         stream_put(stream);
1845
1846 end:
1847         memset(&reply, 0, sizeof(reply));
1848         if (ret < 0) {
1849                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1850         } else {
1851                 reply.ret_code = htobe32(LTTNG_OK);
1852         }
1853         send_ret = conn->sock->ops->sendmsg(
1854                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
1855         if (send_ret < (ssize_t) sizeof(reply)) {
1856                 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)", send_ret);
1857                 ret = -1;
1858         }
1859
1860 end_no_session:
1861         return ret;
1862 }
1863
1864 /*
1865  * relay_unknown_command: send -1 if received unknown command
1866  */
1867 static void relay_unknown_command(struct relay_connection *conn)
1868 {
1869         struct lttcomm_relayd_generic_reply reply;
1870         ssize_t send_ret;
1871
1872         memset(&reply, 0, sizeof(reply));
1873         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1874         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1875         if (send_ret < sizeof(reply)) {
1876                 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1877         }
1878 }
1879
1880 /*
1881  * relay_start: send an acknowledgment to the client to tell if we are
1882  * ready to receive data. We are ready if a session is established.
1883  */
1884 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1885                        struct relay_connection *conn,
1886                        const struct lttng_buffer_view *payload __attribute__((unused)))
1887 {
1888         int ret = 0;
1889         ssize_t send_ret;
1890         struct lttcomm_relayd_generic_reply reply;
1891         struct relay_session *session = conn->session;
1892
1893         if (!session) {
1894                 DBG("Trying to start the streaming without a session established");
1895                 ret = htobe32(LTTNG_ERR_UNK);
1896         }
1897
1898         memset(&reply, 0, sizeof(reply));
1899         reply.ret_code = htobe32(LTTNG_OK);
1900         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1901         if (send_ret < (ssize_t) sizeof(reply)) {
1902                 ERR("Failed to send \"relay_start\" command reply (ret = %zd)", send_ret);
1903                 ret = -1;
1904         }
1905
1906         return ret;
1907 }
1908
1909 /*
1910  * relay_recv_metadata: receive the metadata for the session.
1911  */
1912 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1913                                struct relay_connection *conn,
1914                                const struct lttng_buffer_view *payload)
1915 {
1916         int ret = 0;
1917         struct relay_session *session = conn->session;
1918         struct lttcomm_relayd_metadata_payload metadata_payload_header;
1919         struct relay_stream *metadata_stream;
1920         uint64_t metadata_payload_size;
1921         struct lttng_buffer_view packet_view;
1922
1923         if (!session) {
1924                 ERR("Metadata sent before version check");
1925                 ret = -1;
1926                 goto end;
1927         }
1928
1929         if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1930                 ERR("Incorrect data size");
1931                 ret = -1;
1932                 goto end;
1933         }
1934         metadata_payload_size =
1935                 recv_hdr->data_size - sizeof(struct lttcomm_relayd_metadata_payload);
1936
1937         memcpy(&metadata_payload_header, payload->data, sizeof(metadata_payload_header));
1938         metadata_payload_header.stream_id = be64toh(metadata_payload_header.stream_id);
1939         metadata_payload_header.padding_size = be32toh(metadata_payload_header.padding_size);
1940
1941         metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1942         if (!metadata_stream) {
1943                 ret = -1;
1944                 goto end;
1945         }
1946
1947         packet_view = lttng_buffer_view_from_view(
1948                 payload, sizeof(metadata_payload_header), metadata_payload_size);
1949         if (!lttng_buffer_view_is_valid(&packet_view)) {
1950                 ERR("Invalid metadata packet length announced by header");
1951                 ret = -1;
1952                 goto end_put;
1953         }
1954
1955         pthread_mutex_lock(&metadata_stream->lock);
1956         ret = stream_write(metadata_stream, &packet_view, metadata_payload_header.padding_size);
1957         pthread_mutex_unlock(&metadata_stream->lock);
1958         if (ret) {
1959                 ret = -1;
1960                 goto end_put;
1961         }
1962 end_put:
1963         stream_put(metadata_stream);
1964 end:
1965         return ret;
1966 }
1967
1968 /*
1969  * relay_send_version: send relayd version number
1970  */
1971 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
1972                               struct relay_connection *conn,
1973                               const struct lttng_buffer_view *payload)
1974 {
1975         int ret;
1976         ssize_t send_ret;
1977         struct lttcomm_relayd_version reply, msg;
1978         bool compatible = true;
1979
1980         conn->version_check_done = true;
1981
1982         /* Get version from the other side. */
1983         if (payload->size < sizeof(msg)) {
1984                 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1985                     sizeof(msg),
1986                     payload->size);
1987                 ret = -1;
1988                 goto end;
1989         }
1990
1991         memcpy(&msg, payload->data, sizeof(msg));
1992         msg.major = be32toh(msg.major);
1993         msg.minor = be32toh(msg.minor);
1994
1995         memset(&reply, 0, sizeof(reply));
1996         reply.major = RELAYD_VERSION_COMM_MAJOR;
1997         reply.minor = RELAYD_VERSION_COMM_MINOR;
1998
1999         /* Major versions must be the same */
2000         if (reply.major != msg.major) {
2001                 DBG("Incompatible major versions (%u vs %u), deleting session",
2002                     reply.major,
2003                     msg.major);
2004                 compatible = false;
2005         }
2006
2007         conn->major = reply.major;
2008         /* We adapt to the lowest compatible version */
2009         if (reply.minor <= msg.minor) {
2010                 conn->minor = reply.minor;
2011         } else {
2012                 conn->minor = msg.minor;
2013         }
2014
2015         reply.major = htobe32(reply.major);
2016         reply.minor = htobe32(reply.minor);
2017         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2018         if (send_ret < (ssize_t) sizeof(reply)) {
2019                 ERR("Failed to send \"send version\" command reply (ret = %zd)", send_ret);
2020                 ret = -1;
2021                 goto end;
2022         } else {
2023                 ret = 0;
2024         }
2025
2026         if (!compatible) {
2027                 ret = -1;
2028                 goto end;
2029         }
2030
2031         DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
2032
2033 end:
2034         return ret;
2035 }
2036
2037 /*
2038  * Check for data pending for a given stream id from the session daemon.
2039  */
2040 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2041                               struct relay_connection *conn,
2042                               const struct lttng_buffer_view *payload)
2043 {
2044         struct relay_session *session = conn->session;
2045         struct lttcomm_relayd_data_pending msg;
2046         struct lttcomm_relayd_generic_reply reply;
2047         struct relay_stream *stream;
2048         ssize_t send_ret;
2049         int ret;
2050         uint64_t stream_seq;
2051
2052         DBG("Data pending command received");
2053
2054         if (!session || !conn->version_check_done) {
2055                 ERR("Trying to check for data before version check");
2056                 ret = -1;
2057                 goto end_no_session;
2058         }
2059
2060         if (payload->size < sizeof(msg)) {
2061                 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2062                     sizeof(msg),
2063                     payload->size);
2064                 ret = -1;
2065                 goto end_no_session;
2066         }
2067         memcpy(&msg, payload->data, sizeof(msg));
2068         msg.stream_id = be64toh(msg.stream_id);
2069         msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2070
2071         stream = stream_get_by_id(msg.stream_id);
2072         if (stream == nullptr) {
2073                 ret = -1;
2074                 goto end;
2075         }
2076
2077         pthread_mutex_lock(&stream->lock);
2078
2079         if (session_streams_have_index(session)) {
2080                 /*
2081                  * Ensure that both the index and stream data have been
2082                  * flushed up to the requested point.
2083                  */
2084                 stream_seq = std::min(stream->prev_data_seq, stream->prev_index_seq);
2085         } else {
2086                 stream_seq = stream->prev_data_seq;
2087         }
2088         DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2089             ", prev_index_seq %" PRIu64 ", and last_seq %" PRIu64,
2090             msg.stream_id,
2091             stream->prev_data_seq,
2092             stream->prev_index_seq,
2093             msg.last_net_seq_num);
2094
2095         /* Avoid wrapping issue */
2096         if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2097                 /* Data has in fact been written and is NOT pending */
2098                 ret = 0;
2099         } else {
2100                 /* Data still being streamed thus pending */
2101                 ret = 1;
2102         }
2103
2104         stream->data_pending_check_done = true;
2105         pthread_mutex_unlock(&stream->lock);
2106
2107         stream_put(stream);
2108 end:
2109
2110         memset(&reply, 0, sizeof(reply));
2111         reply.ret_code = htobe32(ret);
2112         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2113         if (send_ret < (ssize_t) sizeof(reply)) {
2114                 ERR("Failed to send \"data pending\" command reply (ret = %zd)", send_ret);
2115                 ret = -1;
2116         }
2117
2118 end_no_session:
2119         return ret;
2120 }
2121
2122 /*
2123  * Wait for the control socket to reach a quiescent state.
2124  *
2125  * Note that for now, when receiving this command from the session
2126  * daemon, this means that every subsequent commands or data received on
2127  * the control socket has been handled. So, this is why we simply return
2128  * OK here.
2129  */
2130 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr
2131                                    __attribute__((unused)),
2132                                    struct relay_connection *conn,
2133                                    const struct lttng_buffer_view *payload)
2134 {
2135         int ret;
2136         ssize_t send_ret;
2137         struct relay_stream *stream;
2138         struct lttcomm_relayd_quiescent_control msg;
2139         struct lttcomm_relayd_generic_reply reply;
2140
2141         DBG("Checking quiescent state on control socket");
2142
2143         if (!conn->session || !conn->version_check_done) {
2144                 ERR("Trying to check for data before version check");
2145                 ret = -1;
2146                 goto end_no_session;
2147         }
2148
2149         if (payload->size < sizeof(msg)) {
2150                 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2151                     sizeof(msg),
2152                     payload->size);
2153                 ret = -1;
2154                 goto end_no_session;
2155         }
2156         memcpy(&msg, payload->data, sizeof(msg));
2157         msg.stream_id = be64toh(msg.stream_id);
2158
2159         stream = stream_get_by_id(msg.stream_id);
2160         if (!stream) {
2161                 goto reply;
2162         }
2163         pthread_mutex_lock(&stream->lock);
2164         stream->data_pending_check_done = true;
2165         pthread_mutex_unlock(&stream->lock);
2166
2167         DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2168         stream_put(stream);
2169 reply:
2170         memset(&reply, 0, sizeof(reply));
2171         reply.ret_code = htobe32(LTTNG_OK);
2172         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2173         if (send_ret < (ssize_t) sizeof(reply)) {
2174                 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)", send_ret);
2175                 ret = -1;
2176         } else {
2177                 ret = 0;
2178         }
2179
2180 end_no_session:
2181         return ret;
2182 }
2183
2184 /*
2185  * Initialize a data pending command. This means that a consumer is about
2186  * to ask for data pending for each stream it holds. Simply iterate over
2187  * all streams of a session and set the data_pending_check_done flag.
2188  *
2189  * This command returns to the client a LTTNG_OK code.
2190  */
2191 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2192                                     struct relay_connection *conn,
2193                                     const struct lttng_buffer_view *payload)
2194 {
2195         int ret;
2196         ssize_t send_ret;
2197         struct lttng_ht_iter iter;
2198         struct lttcomm_relayd_begin_data_pending msg;
2199         struct lttcomm_relayd_generic_reply reply;
2200         struct relay_stream *stream;
2201
2202         LTTNG_ASSERT(recv_hdr);
2203         LTTNG_ASSERT(conn);
2204
2205         DBG("Init streams for data pending");
2206
2207         if (!conn->session || !conn->version_check_done) {
2208                 ERR("Trying to check for data before version check");
2209                 ret = -1;
2210                 goto end_no_session;
2211         }
2212
2213         if (payload->size < sizeof(msg)) {
2214                 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2215                     sizeof(msg),
2216                     payload->size);
2217                 ret = -1;
2218                 goto end_no_session;
2219         }
2220         memcpy(&msg, payload->data, sizeof(msg));
2221         msg.session_id = be64toh(msg.session_id);
2222
2223         /*
2224          * Iterate over all streams to set the begin data pending flag.
2225          * For now, the streams are indexed by stream handle so we have
2226          * to iterate over all streams to find the one associated with
2227          * the right session_id.
2228          */
2229         {
2230                 lttng::urcu::read_lock_guard read_lock;
2231
2232                 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2233                         if (!stream_get(stream)) {
2234                                 continue;
2235                         }
2236
2237                         if (stream->trace->session->id == msg.session_id) {
2238                                 pthread_mutex_lock(&stream->lock);
2239                                 stream->data_pending_check_done = false;
2240                                 pthread_mutex_unlock(&stream->lock);
2241                                 DBG("Set begin data pending flag to stream %" PRIu64,
2242                                     stream->stream_handle);
2243                         }
2244
2245                         stream_put(stream);
2246                 }
2247         }
2248
2249         memset(&reply, 0, sizeof(reply));
2250         /* All good, send back reply. */
2251         reply.ret_code = htobe32(LTTNG_OK);
2252
2253         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2254         if (send_ret < (ssize_t) sizeof(reply)) {
2255                 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)", send_ret);
2256                 ret = -1;
2257         } else {
2258                 ret = 0;
2259         }
2260
2261 end_no_session:
2262         return ret;
2263 }
2264
2265 /*
2266  * End data pending command. This will check, for a given session id, if
2267  * each stream associated with it has its data_pending_check_done flag
2268  * set. If not, this means that the client lost track of the stream but
2269  * the data is still being streamed on our side. In this case, we inform
2270  * the client that data is in flight.
2271  *
2272  * Return to the client if there is data in flight or not with a ret_code.
2273  */
2274 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2275                                   struct relay_connection *conn,
2276                                   const struct lttng_buffer_view *payload)
2277 {
2278         int ret;
2279         ssize_t send_ret;
2280         struct lttng_ht_iter iter;
2281         struct lttcomm_relayd_end_data_pending msg;
2282         struct lttcomm_relayd_generic_reply reply;
2283         struct relay_stream *stream;
2284         uint32_t is_data_inflight = 0;
2285
2286         DBG("End data pending command");
2287
2288         if (!conn->session || !conn->version_check_done) {
2289                 ERR("Trying to check for data before version check");
2290                 ret = -1;
2291                 goto end_no_session;
2292         }
2293
2294         if (payload->size < sizeof(msg)) {
2295                 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2296                     sizeof(msg),
2297                     payload->size);
2298                 ret = -1;
2299                 goto end_no_session;
2300         }
2301         memcpy(&msg, payload->data, sizeof(msg));
2302         msg.session_id = be64toh(msg.session_id);
2303
2304         /*
2305          * Iterate over all streams to see if the begin data pending
2306          * flag is set.
2307          */
2308         {
2309                 lttng::urcu::read_lock_guard read_lock;
2310
2311                 cds_lfht_for_each_entry (relay_streams_ht->ht, &iter.iter, stream, node.node) {
2312                         if (!stream_get(stream)) {
2313                                 continue;
2314                         }
2315
2316                         if (stream->trace->session->id != msg.session_id) {
2317                                 stream_put(stream);
2318                                 continue;
2319                         }
2320
2321                         pthread_mutex_lock(&stream->lock);
2322                         if (!stream->data_pending_check_done) {
2323                                 uint64_t stream_seq;
2324
2325                                 if (session_streams_have_index(conn->session)) {
2326                                         /*
2327                                          * Ensure that both the index and stream data have been
2328                                          * flushed up to the requested point.
2329                                          */
2330                                         stream_seq = std::min(stream->prev_data_seq,
2331                                                               stream->prev_index_seq);
2332                                 } else {
2333                                         stream_seq = stream->prev_data_seq;
2334                                 }
2335
2336                                 if (!stream->closed ||
2337                                     !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2338                                         is_data_inflight = 1;
2339                                         DBG("Data is still in flight for stream %" PRIu64,
2340                                             stream->stream_handle);
2341                                         pthread_mutex_unlock(&stream->lock);
2342                                         stream_put(stream);
2343                                         break;
2344                                 }
2345                         }
2346
2347                         pthread_mutex_unlock(&stream->lock);
2348                         stream_put(stream);
2349                 }
2350         }
2351
2352         memset(&reply, 0, sizeof(reply));
2353         /* All good, send back reply. */
2354         reply.ret_code = htobe32(is_data_inflight);
2355
2356         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2357         if (send_ret < (ssize_t) sizeof(reply)) {
2358                 ERR("Failed to send \"end data pending\" command reply (ret = %zd)", send_ret);
2359                 ret = -1;
2360         } else {
2361                 ret = 0;
2362         }
2363
2364 end_no_session:
2365         return ret;
2366 }
2367
2368 /*
2369  * Receive an index for a specific stream.
2370  *
2371  * Return 0 on success else a negative value.
2372  */
2373 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2374                             struct relay_connection *conn,
2375                             const struct lttng_buffer_view *payload)
2376 {
2377         int ret;
2378         ssize_t send_ret;
2379         struct relay_session *session = conn->session;
2380         struct lttcomm_relayd_index index_info;
2381         struct lttcomm_relayd_generic_reply reply;
2382         struct relay_stream *stream;
2383         size_t msg_len;
2384
2385         LTTNG_ASSERT(conn);
2386
2387         DBG("Relay receiving index");
2388
2389         if (!session || !conn->version_check_done) {
2390                 ERR("Trying to close a stream before version check");
2391                 ret = -1;
2392                 goto end_no_session;
2393         }
2394
2395         msg_len = lttcomm_relayd_index_len(lttng_to_index_major(conn->major, conn->minor),
2396                                            lttng_to_index_minor(conn->major, conn->minor));
2397         if (payload->size < msg_len) {
2398                 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2399                     msg_len,
2400                     payload->size);
2401                 ret = -1;
2402                 goto end_no_session;
2403         }
2404         memcpy(&index_info, payload->data, msg_len);
2405         index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2406         index_info.net_seq_num = be64toh(index_info.net_seq_num);
2407         index_info.packet_size = be64toh(index_info.packet_size);
2408         index_info.content_size = be64toh(index_info.content_size);
2409         index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2410         index_info.timestamp_end = be64toh(index_info.timestamp_end);
2411         index_info.events_discarded = be64toh(index_info.events_discarded);
2412         index_info.stream_id = be64toh(index_info.stream_id);
2413
2414         if (conn->minor >= 8) {
2415                 index_info.stream_instance_id = be64toh(index_info.stream_instance_id);
2416                 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2417         } else {
2418                 index_info.stream_instance_id = -1ULL;
2419                 index_info.packet_seq_num = -1ULL;
2420         }
2421
2422         stream = stream_get_by_id(index_info.relay_stream_id);
2423         if (!stream) {
2424                 ERR("stream_get_by_id not found");
2425                 ret = -1;
2426                 goto end;
2427         }
2428
2429         pthread_mutex_lock(&stream->lock);
2430         ret = stream_add_index(stream, &index_info);
2431         pthread_mutex_unlock(&stream->lock);
2432         if (ret) {
2433                 goto end_stream_put;
2434         }
2435
2436 end_stream_put:
2437         stream_put(stream);
2438 end:
2439         memset(&reply, 0, sizeof(reply));
2440         if (ret < 0) {
2441                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2442         } else {
2443                 reply.ret_code = htobe32(LTTNG_OK);
2444         }
2445         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2446         if (send_ret < (ssize_t) sizeof(reply)) {
2447                 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2448                 ret = -1;
2449         }
2450
2451 end_no_session:
2452         return ret;
2453 }
2454
2455 /*
2456  * Receive the streams_sent message.
2457  *
2458  * Return 0 on success else a negative value.
2459  */
2460 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr __attribute__((unused)),
2461                               struct relay_connection *conn,
2462                               const struct lttng_buffer_view *payload __attribute__((unused)))
2463 {
2464         int ret;
2465         ssize_t send_ret;
2466         struct lttcomm_relayd_generic_reply reply;
2467
2468         LTTNG_ASSERT(conn);
2469
2470         DBG("Relay receiving streams_sent");
2471
2472         if (!conn->session || !conn->version_check_done) {
2473                 ERR("Trying to close a stream before version check");
2474                 ret = -1;
2475                 goto end_no_session;
2476         }
2477
2478         /*
2479          * Publish every pending stream in the connection recv list which are
2480          * now ready to be used by the viewer.
2481          */
2482         publish_connection_local_streams(conn);
2483
2484         memset(&reply, 0, sizeof(reply));
2485         reply.ret_code = htobe32(LTTNG_OK);
2486         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2487         if (send_ret < (ssize_t) sizeof(reply)) {
2488                 ERR("Failed to send \"streams sent\" command reply (ret = %zd)", send_ret);
2489                 ret = -1;
2490         } else {
2491                 /* Success. */
2492                 ret = 0;
2493         }
2494
2495 end_no_session:
2496         return ret;
2497 }
2498
2499 static ssize_t
2500 relay_unpack_rotate_streams_header(const struct lttng_buffer_view *payload,
2501                                    struct lttcomm_relayd_rotate_streams *_rotate_streams)
2502 {
2503         struct lttcomm_relayd_rotate_streams rotate_streams;
2504         /*
2505          * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2506          * This is the smallest version of this structure, but it can be larger;
2507          * this variable is updated once the proper size of the structure is known.
2508          *
2509          * See comment at the declaration of this structure for more information.
2510          */
2511         ssize_t header_len = sizeof(struct lttcomm_relayd_rotate_streams_packed);
2512         size_t expected_payload_size_no_padding, expected_payload_size_3_bytes_padding,
2513                 expected_payload_size_7_bytes_padding;
2514
2515         if (payload->size < header_len) {
2516                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2517                     header_len,
2518                     payload->size);
2519                 goto error;
2520         }
2521
2522         /*
2523          * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2524          * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2525          *
2526          * We start by "unpacking" `stream_count` to figure out the padding length
2527          * emited by our peer.
2528          */
2529         {
2530                 decltype(rotate_streams.stream_count) stream_count;
2531
2532                 memcpy(&stream_count, payload->data, sizeof(stream_count));
2533                 rotate_streams.stream_count = be32toh(stream_count);
2534         }
2535
2536         rotate_streams.new_chunk_id = LTTNG_OPTIONAL_INIT_UNSET;
2537
2538         /*
2539          * Payload size expected given the possible padding lengths in
2540          * `struct lttcomm_relayd_rotate_streams`.
2541          */
2542         expected_payload_size_no_padding =
2543                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2544                 sizeof(lttcomm_relayd_rotate_streams_packed);
2545         expected_payload_size_3_bytes_padding =
2546                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2547                 sizeof(lttcomm_relayd_rotate_streams_3_bytes_padding);
2548         expected_payload_size_7_bytes_padding =
2549                 (rotate_streams.stream_count * sizeof(*rotate_streams.rotation_positions)) +
2550                 sizeof(lttcomm_relayd_rotate_streams_7_bytes_padding);
2551
2552         if (payload->size == expected_payload_size_no_padding) {
2553                 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams;
2554
2555                 /*
2556                  * This handles cases where someone might build with
2557                  * -fpack-struct or any other toolchain that wouldn't produce
2558                  * padding to align `value`.
2559                  */
2560                 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2561
2562                 header_len = sizeof(packed_rotate_streams);
2563                 memcpy(&packed_rotate_streams, payload->data, header_len);
2564
2565                 /* Unpack the packed structure to the natively-packed version. */
2566                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2567                         .is_set = !!packed_rotate_streams.new_chunk_id.is_set,
2568                         .value = be64toh(packed_rotate_streams.new_chunk_id.value),
2569                 };
2570                 _rotate_streams->stream_count = be32toh(packed_rotate_streams.stream_count);
2571         } else if (payload->size == expected_payload_size_3_bytes_padding) {
2572                 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams;
2573
2574                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2575
2576                 header_len = sizeof(padded_rotate_streams);
2577                 memcpy(&padded_rotate_streams, payload->data, header_len);
2578
2579                 /* Unpack the 3-byte padded structure to the natively-packed version. */
2580                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2581                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2582                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2583                 };
2584                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2585         } else if (payload->size == expected_payload_size_7_bytes_padding) {
2586                 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams;
2587
2588                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2589
2590                 header_len = sizeof(padded_rotate_streams);
2591                 memcpy(&padded_rotate_streams, payload->data, header_len);
2592
2593                 /* Unpack the 7-byte padded structure to the natively-packed version. */
2594                 _rotate_streams->new_chunk_id = (typeof(_rotate_streams->new_chunk_id)){
2595                         .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2596                         .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2597                 };
2598                 _rotate_streams->stream_count = be32toh(padded_rotate_streams.stream_count);
2599
2600                 header_len = sizeof(padded_rotate_streams);
2601         } else {
2602                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2603                     expected_payload_size_no_padding,
2604                     expected_payload_size_3_bytes_padding,
2605                     expected_payload_size_7_bytes_padding,
2606                     payload->size);
2607                 goto error;
2608         }
2609
2610         return header_len;
2611 error:
2612         return -1;
2613 }
2614
2615 /*
2616  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2617  * session rotation feature (not the tracefile rotation feature).
2618  */
2619 static int relay_rotate_session_streams(const struct lttcomm_relayd_hdr *recv_hdr
2620                                         __attribute__((unused)),
2621                                         struct relay_connection *conn,
2622                                         const struct lttng_buffer_view *payload)
2623 {
2624         int ret = 0;
2625         uint32_t i;
2626         ssize_t send_ret;
2627         enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2628         struct relay_session *session = conn->session;
2629         struct lttcomm_relayd_rotate_streams rotate_streams;
2630         struct lttcomm_relayd_generic_reply reply = {};
2631         struct relay_stream *stream = nullptr;
2632         struct lttng_trace_chunk *next_trace_chunk = nullptr;
2633         struct lttng_buffer_view stream_positions;
2634         char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2635         const char *chunk_id_str = "none";
2636         ssize_t header_len;
2637
2638         if (!session || !conn->version_check_done) {
2639                 ERR("Trying to rotate a stream before version check");
2640                 ret = -1;
2641                 goto end_no_reply;
2642         }
2643
2644         if (session->major == 2 && session->minor < 11) {
2645                 ERR("Unsupported feature before 2.11");
2646                 ret = -1;
2647                 goto end_no_reply;
2648         }
2649
2650         header_len = relay_unpack_rotate_streams_header(payload, &rotate_streams);
2651         if (header_len < 0) {
2652                 ret = -1;
2653                 goto end_no_reply;
2654         }
2655
2656         if (rotate_streams.new_chunk_id.is_set) {
2657                 /*
2658                  * Retrieve the trace chunk the stream must transition to. As
2659                  * per the protocol, this chunk should have been created
2660                  * before this command is received.
2661                  */
2662                 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2663                         sessiond_trace_chunk_registry,
2664                         session->sessiond_uuid,
2665                         conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2666                                                             conn->session->id,
2667                         rotate_streams.new_chunk_id.value);
2668                 if (!next_trace_chunk) {
2669                         char uuid_str[LTTNG_UUID_STR_LEN];
2670
2671                         lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2672                         ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2673                             ", trace_chunk_id = %" PRIu64,
2674                             uuid_str,
2675                             session->id,
2676                             rotate_streams.new_chunk_id.value);
2677                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2678                         ret = -1;
2679                         goto end;
2680                 }
2681
2682                 ret = snprintf(chunk_id_buf,
2683                                sizeof(chunk_id_buf),
2684                                "%" PRIu64,
2685                                rotate_streams.new_chunk_id.value);
2686                 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2687                         chunk_id_str = "formatting error";
2688                 } else {
2689                         chunk_id_str = chunk_id_buf;
2690                 }
2691         }
2692
2693         DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2694             rotate_streams.stream_count,
2695             session->session_name,
2696             chunk_id_str);
2697
2698         stream_positions = lttng_buffer_view_from_view(payload, header_len, -1);
2699         if (!stream_positions.data ||
2700             stream_positions.size < (rotate_streams.stream_count *
2701                                      sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2702                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2703                 ret = -1;
2704                 goto end;
2705         }
2706
2707         for (i = 0; i < rotate_streams.stream_count; i++) {
2708                 struct lttcomm_relayd_stream_rotation_position *position_comm =
2709                         &((typeof(position_comm)) stream_positions.data)[i];
2710                 const struct lttcomm_relayd_stream_rotation_position pos = {
2711                         .stream_id = be64toh(position_comm->stream_id),
2712                         .rotate_at_seq_num = be64toh(position_comm->rotate_at_seq_num),
2713                 };
2714
2715                 stream = stream_get_by_id(pos.stream_id);
2716                 if (!stream) {
2717                         reply_code = LTTNG_ERR_INVALID;
2718                         ret = -1;
2719                         goto end;
2720                 }
2721
2722                 pthread_mutex_lock(&stream->lock);
2723                 ret = stream_set_pending_rotation(stream, next_trace_chunk, pos.rotate_at_seq_num);
2724                 pthread_mutex_unlock(&stream->lock);
2725                 if (ret) {
2726                         reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2727                         goto end;
2728                 }
2729
2730                 stream_put(stream);
2731                 stream = nullptr;
2732         }
2733
2734         reply_code = LTTNG_OK;
2735         ret = 0;
2736 end:
2737         if (stream) {
2738                 stream_put(stream);
2739         }
2740
2741         reply.ret_code = htobe32((uint32_t) reply_code);
2742         send_ret = conn->sock->ops->sendmsg(
2743                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2744         if (send_ret < (ssize_t) sizeof(reply)) {
2745                 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)", send_ret);
2746                 ret = -1;
2747         }
2748 end_no_reply:
2749         lttng_trace_chunk_put(next_trace_chunk);
2750         return ret;
2751 }
2752
2753 /*
2754  * relay_create_trace_chunk: create a new trace chunk
2755  */
2756 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2757                                     __attribute__((unused)),
2758                                     struct relay_connection *conn,
2759                                     const struct lttng_buffer_view *payload)
2760 {
2761         int ret = 0;
2762         ssize_t send_ret;
2763         struct relay_session *session = conn->session;
2764         struct lttcomm_relayd_create_trace_chunk *msg;
2765         struct lttcomm_relayd_generic_reply reply = {};
2766         struct lttng_buffer_view header_view;
2767         struct lttng_trace_chunk *chunk = nullptr, *published_chunk = nullptr;
2768         enum lttng_error_code reply_code = LTTNG_OK;
2769         enum lttng_trace_chunk_status chunk_status;
2770         const char *new_path;
2771
2772         if (!session || !conn->version_check_done) {
2773                 ERR("Trying to create a trace chunk before version check");
2774                 ret = -1;
2775                 goto end_no_reply;
2776         }
2777
2778         if (session->major == 2 && session->minor < 11) {
2779                 ERR("Chunk creation command is unsupported before 2.11");
2780                 ret = -1;
2781                 goto end_no_reply;
2782         }
2783
2784         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2785         if (!lttng_buffer_view_is_valid(&header_view)) {
2786                 ERR("Failed to receive payload of chunk creation command");
2787                 ret = -1;
2788                 goto end_no_reply;
2789         }
2790
2791         /* Convert to host endianness. */
2792         msg = (typeof(msg)) header_view.data;
2793         msg->chunk_id = be64toh(msg->chunk_id);
2794         msg->creation_timestamp = be64toh(msg->creation_timestamp);
2795         msg->override_name_length = be32toh(msg->override_name_length);
2796
2797         pthread_mutex_lock(&conn->session->lock);
2798         session->ongoing_rotation = true;
2799         if (session->current_trace_chunk &&
2800             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2801                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2802                                                              DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2803                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2804                         ERR("Failed to rename old chunk");
2805                         ret = -1;
2806                         reply_code = LTTNG_ERR_UNK;
2807                         goto end;
2808                 }
2809         }
2810         if (!session->current_trace_chunk) {
2811                 if (!session->has_rotated) {
2812                         new_path = "";
2813                 } else {
2814                         new_path = nullptr;
2815                 }
2816         } else {
2817                 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2818         }
2819         chunk = lttng_trace_chunk_create(msg->chunk_id, msg->creation_timestamp, new_path);
2820         if (!chunk) {
2821                 ERR("Failed to create trace chunk in trace chunk creation command");
2822                 ret = -1;
2823                 reply_code = LTTNG_ERR_NOMEM;
2824                 goto end;
2825         }
2826         lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2827
2828         if (msg->override_name_length) {
2829                 const char *name;
2830                 const struct lttng_buffer_view chunk_name_view = lttng_buffer_view_from_view(
2831                         payload, sizeof(*msg), msg->override_name_length);
2832
2833                 if (!lttng_buffer_view_is_valid(&chunk_name_view)) {
2834                         ERR("Invalid payload of chunk creation command (protocol error): buffer too short for expected name length");
2835                         ret = -1;
2836                         reply_code = LTTNG_ERR_INVALID;
2837                         goto end;
2838                 }
2839
2840                 name = chunk_name_view.data;
2841                 if (name[msg->override_name_length - 1]) {
2842                         ERR("Invalid payload of chunk creation command (protocol error): name is not null-terminated");
2843                         ret = -1;
2844                         reply_code = LTTNG_ERR_INVALID;
2845                         goto end;
2846                 }
2847
2848                 chunk_status = lttng_trace_chunk_override_name(chunk, chunk_name_view.data);
2849                 switch (chunk_status) {
2850                 case LTTNG_TRACE_CHUNK_STATUS_OK:
2851                         break;
2852                 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2853                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2854                         reply_code = LTTNG_ERR_INVALID;
2855                         ret = -1;
2856                         goto end;
2857                 default:
2858                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2859                         reply_code = LTTNG_ERR_UNK;
2860                         ret = -1;
2861                         goto end;
2862                 }
2863         }
2864
2865         chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2866         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2867                 reply_code = LTTNG_ERR_UNK;
2868                 ret = -1;
2869                 goto end;
2870         }
2871
2872         LTTNG_ASSERT(conn->session->output_directory);
2873         chunk_status = lttng_trace_chunk_set_as_owner(chunk, conn->session->output_directory);
2874         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2875                 reply_code = LTTNG_ERR_UNK;
2876                 ret = -1;
2877                 goto end;
2878         }
2879
2880         published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2881                 sessiond_trace_chunk_registry,
2882                 conn->session->sessiond_uuid,
2883                 conn->session->id_sessiond.is_set ? conn->session->id_sessiond.value :
2884                                                     conn->session->id,
2885                 chunk);
2886         if (!published_chunk) {
2887                 char uuid_str[LTTNG_UUID_STR_LEN];
2888
2889                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2890                 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
2891                     ", chunk_id = %" PRIu64,
2892                     uuid_str,
2893                     conn->session->id,
2894                     msg->chunk_id);
2895                 ret = -1;
2896                 reply_code = LTTNG_ERR_NOMEM;
2897                 goto end;
2898         }
2899
2900         if (conn->session->pending_closure_trace_chunk) {
2901                 /*
2902                  * Invalid; this means a second create_trace_chunk command was
2903                  * received before a close_trace_chunk.
2904                  */
2905                 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2906                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2907                 ret = -1;
2908                 goto end;
2909         }
2910         conn->session->pending_closure_trace_chunk = conn->session->current_trace_chunk;
2911         conn->session->current_trace_chunk = published_chunk;
2912         published_chunk = nullptr;
2913         if (!conn->session->pending_closure_trace_chunk) {
2914                 session->ongoing_rotation = false;
2915         }
2916 end:
2917         pthread_mutex_unlock(&conn->session->lock);
2918         reply.ret_code = htobe32((uint32_t) reply_code);
2919         send_ret = conn->sock->ops->sendmsg(
2920                 conn->sock, &reply, sizeof(struct lttcomm_relayd_generic_reply), 0);
2921         if (send_ret < (ssize_t) sizeof(reply)) {
2922                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
2923                 ret = -1;
2924         }
2925 end_no_reply:
2926         lttng_trace_chunk_put(chunk);
2927         lttng_trace_chunk_put(published_chunk);
2928         return ret;
2929 }
2930
2931 /*
2932  * relay_close_trace_chunk: close a trace chunk
2933  */
2934 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr
2935                                    __attribute__((unused)),
2936                                    struct relay_connection *conn,
2937                                    const struct lttng_buffer_view *payload)
2938 {
2939         int ret = 0, buf_ret;
2940         ssize_t send_ret;
2941         struct relay_session *session = conn->session;
2942         struct lttcomm_relayd_close_trace_chunk *msg;
2943         struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2944         struct lttng_buffer_view header_view;
2945         struct lttng_trace_chunk *chunk = nullptr;
2946         enum lttng_error_code reply_code = LTTNG_OK;
2947         enum lttng_trace_chunk_status chunk_status;
2948         uint64_t chunk_id;
2949         LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2950         time_t close_timestamp;
2951         char closed_trace_chunk_path[LTTNG_PATH_MAX];
2952         size_t path_length = 0;
2953         const char *chunk_name = nullptr;
2954         struct lttng_dynamic_buffer reply_payload;
2955         const char *new_path;
2956
2957         lttng_dynamic_buffer_init(&reply_payload);
2958
2959         if (!session || !conn->version_check_done) {
2960                 ERR("Trying to close a trace chunk before version check");
2961                 ret = -1;
2962                 goto end_no_reply;
2963         }
2964
2965         if (session->major == 2 && session->minor < 11) {
2966                 ERR("Chunk close command is unsupported before 2.11");
2967                 ret = -1;
2968                 goto end_no_reply;
2969         }
2970
2971         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2972         if (!lttng_buffer_view_is_valid(&header_view)) {
2973                 ERR("Failed to receive payload of chunk close command");
2974                 ret = -1;
2975                 goto end_no_reply;
2976         }
2977
2978         /* Convert to host endianness. */
2979         msg = (typeof(msg)) header_view.data;
2980         chunk_id = be64toh(msg->chunk_id);
2981         close_timestamp = (time_t) be64toh(msg->close_timestamp);
2982         close_command.value = (lttng_trace_chunk_command_type) be32toh(msg->close_command.value);
2983         close_command.is_set = msg->close_command.is_set;
2984
2985         chunk = sessiond_trace_chunk_registry_get_chunk(sessiond_trace_chunk_registry,
2986                                                         conn->session->sessiond_uuid,
2987                                                         conn->session->id_sessiond.is_set ?
2988                                                                 conn->session->id_sessiond.value :
2989                                                                 conn->session->id,
2990                                                         chunk_id);
2991         if (!chunk) {
2992                 char uuid_str[LTTNG_UUID_STR_LEN];
2993
2994                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2995                 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
2996                     ", chunk_id = %" PRIu64,
2997                     uuid_str,
2998                     conn->session->id,
2999                     msg->chunk_id);
3000                 ret = -1;
3001                 reply_code = LTTNG_ERR_NOMEM;
3002                 goto end;
3003         }
3004
3005         pthread_mutex_lock(&session->lock);
3006         if (close_command.is_set && close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
3007                 /*
3008                  * Clear command. It is a protocol error to ask for a
3009                  * clear on a relay which does not allow it. Querying
3010                  * the configuration allows figuring out whether
3011                  * clearing is allowed before doing the clear.
3012                  */
3013                 if (!opt_allow_clear) {
3014                         ret = -1;
3015                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3016                         goto end_unlock_session;
3017                 }
3018         }
3019         if (session->pending_closure_trace_chunk && session->pending_closure_trace_chunk != chunk) {
3020                 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3021                     session->session_name);
3022                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3023                 ret = -1;
3024                 goto end_unlock_session;
3025         }
3026
3027         if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
3028             !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
3029                 if (close_command.is_set &&
3030                     close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
3031                     !session->has_rotated) {
3032                         /* New chunk stays in session output directory. */
3033                         new_path = "";
3034                 } else {
3035                         /* Use chunk name for new chunk. */
3036                         new_path = nullptr;
3037                 }
3038                 /* Rename new chunk path. */
3039                 chunk_status =
3040                         lttng_trace_chunk_rename_path(session->current_trace_chunk, new_path);
3041                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3042                         ret = -1;
3043                         goto end_unlock_session;
3044                 }
3045                 session->ongoing_rotation = false;
3046         }
3047         if ((!close_command.is_set ||
3048              close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
3049             !lttng_trace_chunk_get_name_overridden(chunk)) {
3050                 const char *old_path;
3051
3052                 if (!session->has_rotated) {
3053                         old_path = "";
3054                 } else {
3055                         old_path = nullptr;
3056                 }
3057                 /* We need to move back the .tmp_old_chunk to its rightful place. */
3058                 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
3059                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3060                         ret = -1;
3061                         goto end_unlock_session;
3062                 }
3063         }
3064         chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, close_timestamp);
3065         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3066                 ERR("Failed to set trace chunk close timestamp");
3067                 ret = -1;
3068                 reply_code = LTTNG_ERR_UNK;
3069                 goto end_unlock_session;
3070         }
3071
3072         if (close_command.is_set) {
3073                 chunk_status = lttng_trace_chunk_set_close_command(chunk, close_command.value);
3074                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3075                         ret = -1;
3076                         reply_code = LTTNG_ERR_INVALID;
3077                         goto end_unlock_session;
3078                 }
3079         }
3080         chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, nullptr);
3081         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3082                 ERR("Failed to get chunk name");
3083                 ret = -1;
3084                 reply_code = LTTNG_ERR_UNK;
3085                 goto end_unlock_session;
3086         }
3087         if (!session->has_rotated && !session->snapshot) {
3088                 ret = lttng_strncpy(closed_trace_chunk_path,
3089                                     session->output_path,
3090                                     sizeof(closed_trace_chunk_path));
3091                 if (ret) {
3092                         ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3093                             strlen(session->output_path),
3094                             sizeof(closed_trace_chunk_path));
3095                         reply_code = LTTNG_ERR_NOMEM;
3096                         ret = -1;
3097                         goto end_unlock_session;
3098                 }
3099         } else {
3100                 if (session->snapshot) {
3101                         ret = snprintf(closed_trace_chunk_path,
3102                                        sizeof(closed_trace_chunk_path),
3103                                        "%s/%s",
3104                                        session->output_path,
3105                                        chunk_name);
3106                 } else {
3107                         ret = snprintf(closed_trace_chunk_path,
3108                                        sizeof(closed_trace_chunk_path),
3109                                        "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY "/%s",
3110                                        session->output_path,
3111                                        chunk_name);
3112                 }
3113                 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3114                         ERR("Failed to format closed trace chunk resulting path");
3115                         reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3116                         ret = -1;
3117                         goto end_unlock_session;
3118                 }
3119         }
3120         if (close_command.is_set &&
3121             close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3122                 session->has_rotated = true;
3123         }
3124         DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3125         path_length = strlen(closed_trace_chunk_path) + 1;
3126         if (path_length > UINT32_MAX) {
3127                 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3128                 ret = -1;
3129                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3130                 goto end_unlock_session;
3131         }
3132
3133         if (session->current_trace_chunk == chunk) {
3134                 /*
3135                  * After a trace chunk close command, no new streams
3136                  * referencing the chunk may be created. Hence, on the
3137                  * event that no new trace chunk have been created for
3138                  * the session, the reference to the current trace chunk
3139                  * is released in order to allow it to be reclaimed when
3140                  * the last stream releases its reference to it.
3141                  */
3142                 lttng_trace_chunk_put(session->current_trace_chunk);
3143                 session->current_trace_chunk = nullptr;
3144         }
3145         lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3146         session->pending_closure_trace_chunk = nullptr;
3147 end_unlock_session:
3148         pthread_mutex_unlock(&session->lock);
3149
3150 end:
3151         reply.generic.ret_code = htobe32((uint32_t) reply_code);
3152         reply.path_length = htobe32((uint32_t) path_length);
3153         buf_ret = lttng_dynamic_buffer_append(&reply_payload, &reply, sizeof(reply));
3154         if (buf_ret) {
3155                 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3156                 goto end_no_reply;
3157         }
3158
3159         if (reply_code == LTTNG_OK) {
3160                 buf_ret = lttng_dynamic_buffer_append(
3161                         &reply_payload, closed_trace_chunk_path, path_length);
3162                 if (buf_ret) {
3163                         ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3164                         goto end_no_reply;
3165                 }
3166         }
3167
3168         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data, reply_payload.size, 0);
3169         if (send_ret < reply_payload.size) {
3170                 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3171                     reply_payload.size,
3172                     send_ret);
3173                 ret = -1;
3174                 goto end_no_reply;
3175         }
3176 end_no_reply:
3177         lttng_trace_chunk_put(chunk);
3178         lttng_dynamic_buffer_reset(&reply_payload);
3179         return ret;
3180 }
3181
3182 /*
3183  * relay_trace_chunk_exists: check if a trace chunk exists
3184  */
3185 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr
3186                                     __attribute__((unused)),
3187                                     struct relay_connection *conn,
3188                                     const struct lttng_buffer_view *payload)
3189 {
3190         int ret = 0;
3191         ssize_t send_ret;
3192         struct relay_session *session = conn->session;
3193         struct lttcomm_relayd_trace_chunk_exists *msg;
3194         struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3195         struct lttng_buffer_view header_view;
3196         uint64_t chunk_id;
3197         bool chunk_exists;
3198
3199         if (!session || !conn->version_check_done) {
3200                 ERR("Trying to check for the presence of a trace chunk before version check");
3201                 ret = -1;
3202                 goto end_no_reply;
3203         }
3204
3205         if (session->major == 2 && session->minor < 11) {
3206                 ERR("Chunk exists command is unsupported before 2.11");
3207                 ret = -1;
3208                 goto end_no_reply;
3209         }
3210
3211         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3212         if (!lttng_buffer_view_is_valid(&header_view)) {
3213                 ERR("Failed to receive payload of chunk exists command");
3214                 ret = -1;
3215                 goto end_no_reply;
3216         }
3217
3218         /* Convert to host endianness. */
3219         msg = (typeof(msg)) header_view.data;
3220         chunk_id = be64toh(msg->chunk_id);
3221
3222         ret = sessiond_trace_chunk_registry_chunk_exists(sessiond_trace_chunk_registry,
3223                                                          conn->session->sessiond_uuid,
3224                                                          conn->session->id,
3225                                                          chunk_id,
3226                                                          &chunk_exists);
3227         /*
3228          * If ret is not 0, send the reply and report the error to the caller.
3229          * It is a protocol (or internal) error and the session/connection
3230          * should be torn down.
3231          */
3232         reply.generic.ret_code =
3233                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3234         reply.trace_chunk_exists = ret == 0 ? chunk_exists : 0;
3235
3236         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3237         if (send_ret < (ssize_t) sizeof(reply)) {
3238                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)", send_ret);
3239                 ret = -1;
3240         }
3241 end_no_reply:
3242         return ret;
3243 }
3244
3245 /*
3246  * relay_get_configuration: query whether feature is available
3247  */
3248 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr
3249                                    __attribute__((unused)),
3250                                    struct relay_connection *conn,
3251                                    const struct lttng_buffer_view *payload)
3252 {
3253         int ret = 0;
3254         ssize_t send_ret;
3255         struct lttcomm_relayd_get_configuration *msg;
3256         struct lttcomm_relayd_get_configuration_reply reply = {};
3257         struct lttng_buffer_view header_view;
3258         uint64_t query_flags = 0;
3259         uint64_t result_flags = 0;
3260
3261         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3262         if (!lttng_buffer_view_is_valid(&header_view)) {
3263                 ERR("Failed to receive payload of chunk close command");
3264                 ret = -1;
3265                 goto end_no_reply;
3266         }
3267
3268         /* Convert to host endianness. */
3269         msg = (typeof(msg)) header_view.data;
3270         query_flags = be64toh(msg->query_flags);
3271
3272         if (query_flags) {
3273                 ret = LTTNG_ERR_INVALID_PROTOCOL;
3274                 goto reply;
3275         }
3276         if (opt_allow_clear) {
3277                 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3278         }
3279         ret = 0;
3280 reply:
3281         reply.generic.ret_code =
3282                 htobe32((uint32_t) (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL));
3283         reply.relayd_configuration_flags = htobe64(result_flags);
3284
3285         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
3286         if (send_ret < (ssize_t) sizeof(reply)) {
3287                 ERR("Failed to send \"get configuration\" command reply (ret = %zd)", send_ret);
3288                 ret = -1;
3289         }
3290 end_no_reply:
3291         return ret;
3292 }
3293
3294 static int relay_process_control_command(struct relay_connection *conn,
3295                                          const struct lttcomm_relayd_hdr *header,
3296                                          const struct lttng_buffer_view *payload)
3297 {
3298         int ret = 0;
3299
3300         DBG3("Processing \"%s\" command for socket %i",
3301              lttcomm_relayd_command_str((lttcomm_relayd_command) header->cmd),
3302              conn->sock->fd);
3303         switch (header->cmd) {
3304         case RELAYD_CREATE_SESSION:
3305                 ret = relay_create_session(header, conn, payload);
3306                 break;
3307         case RELAYD_ADD_STREAM:
3308                 ret = relay_add_stream(header, conn, payload);
3309                 break;
3310         case RELAYD_START_DATA:
3311                 ret = relay_start(header, conn, payload);
3312                 break;
3313         case RELAYD_SEND_METADATA:
3314                 ret = relay_recv_metadata(header, conn, payload);
3315                 break;
3316         case RELAYD_VERSION:
3317                 ret = relay_send_version(header, conn, payload);
3318                 break;
3319         case RELAYD_CLOSE_STREAM:
3320                 ret = relay_close_stream(header, conn, payload);
3321                 break;
3322         case RELAYD_DATA_PENDING:
3323                 ret = relay_data_pending(header, conn, payload);
3324                 break;
3325         case RELAYD_QUIESCENT_CONTROL:
3326                 ret = relay_quiescent_control(header, conn, payload);
3327                 break;
3328         case RELAYD_BEGIN_DATA_PENDING:
3329                 ret = relay_begin_data_pending(header, conn, payload);
3330                 break;
3331         case RELAYD_END_DATA_PENDING:
3332                 ret = relay_end_data_pending(header, conn, payload);
3333                 break;
3334         case RELAYD_SEND_INDEX:
3335                 ret = relay_recv_index(header, conn, payload);
3336                 break;
3337         case RELAYD_STREAMS_SENT:
3338                 ret = relay_streams_sent(header, conn, payload);
3339                 break;
3340         case RELAYD_RESET_METADATA:
3341                 ret = relay_reset_metadata(header, conn, payload);
3342                 break;
3343         case RELAYD_ROTATE_STREAMS:
3344                 ret = relay_rotate_session_streams(header, conn, payload);
3345                 break;
3346         case RELAYD_CREATE_TRACE_CHUNK:
3347                 ret = relay_create_trace_chunk(header, conn, payload);
3348                 break;
3349         case RELAYD_CLOSE_TRACE_CHUNK:
3350                 ret = relay_close_trace_chunk(header, conn, payload);
3351                 break;
3352         case RELAYD_TRACE_CHUNK_EXISTS:
3353                 ret = relay_trace_chunk_exists(header, conn, payload);
3354                 break;
3355         case RELAYD_GET_CONFIGURATION:
3356                 ret = relay_get_configuration(header, conn, payload);
3357                 break;
3358         case RELAYD_UPDATE_SYNC_INFO:
3359         default:
3360                 ERR("Received unknown command (%u)", header->cmd);
3361                 relay_unknown_command(conn);
3362                 ret = -1;
3363                 goto end;
3364         }
3365
3366 end:
3367         return ret;
3368 }
3369
3370 static enum relay_connection_status
3371 relay_process_control_receive_payload(struct relay_connection *conn)
3372 {
3373         int ret = 0;
3374         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3375         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3376         struct ctrl_connection_state_receive_payload *state =
3377                 &conn->protocol.ctrl.state.receive_payload;
3378         struct lttng_buffer_view payload_view;
3379
3380         if (state->left_to_receive == 0) {
3381                 /* Short-circuit for payload-less commands. */
3382                 goto reception_complete;
3383         }
3384
3385         ret = conn->sock->ops->recvmsg(conn->sock,
3386                                        reception_buffer->data + state->received,
3387                                        state->left_to_receive,
3388                                        MSG_DONTWAIT);
3389         if (ret < 0) {
3390                 DIAGNOSTIC_PUSH
3391                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3392                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3393                         DIAGNOSTIC_POP
3394                         PERROR("Unable to receive command payload on sock %d", conn->sock->fd);
3395                         status = RELAY_CONNECTION_STATUS_ERROR;
3396                 }
3397                 goto end;
3398         } else if (ret == 0) {
3399                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3400                 status = RELAY_CONNECTION_STATUS_CLOSED;
3401                 goto end;
3402         }
3403
3404         LTTNG_ASSERT(ret > 0);
3405         LTTNG_ASSERT(ret <= state->left_to_receive);
3406
3407         state->left_to_receive -= ret;
3408         state->received += ret;
3409
3410         if (state->left_to_receive > 0) {
3411                 /*
3412                  * Can't transition to the protocol's next state, wait to
3413                  * receive the rest of the header.
3414                  */
3415                 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
3416                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3417                      state->received,
3418                      state->left_to_receive,
3419                      conn->sock->fd);
3420                 goto end;
3421         }
3422
3423 reception_complete:
3424         DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3425             conn->sock->fd,
3426             state->received);
3427         /*
3428          * The payload required to process the command has been received.
3429          * A view to the reception buffer is forwarded to the various
3430          * commands and the state of the control is reset on success.
3431          *
3432          * Commands are responsible for sending their reply to the peer.
3433          */
3434         payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer, 0, -1);
3435         ret = relay_process_control_command(conn, &state->header, &payload_view);
3436         if (ret < 0) {
3437                 status = RELAY_CONNECTION_STATUS_ERROR;
3438                 goto end;
3439         }
3440
3441         ret = connection_reset_protocol_state(conn);
3442         if (ret) {
3443                 status = RELAY_CONNECTION_STATUS_ERROR;
3444         }
3445 end:
3446         return status;
3447 }
3448
3449 static enum relay_connection_status
3450 relay_process_control_receive_header(struct relay_connection *conn)
3451 {
3452         int ret = 0;
3453         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3454         struct lttcomm_relayd_hdr header;
3455         struct lttng_dynamic_buffer *reception_buffer = &conn->protocol.ctrl.reception_buffer;
3456         struct ctrl_connection_state_receive_header *state =
3457                 &conn->protocol.ctrl.state.receive_header;
3458
3459         LTTNG_ASSERT(state->left_to_receive != 0);
3460
3461         ret = conn->sock->ops->recvmsg(conn->sock,
3462                                        reception_buffer->data + state->received,
3463                                        state->left_to_receive,
3464                                        MSG_DONTWAIT);
3465         if (ret < 0) {
3466                 DIAGNOSTIC_PUSH
3467                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3468                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3469                         DIAGNOSTIC_POP
3470                         PERROR("Unable to receive control command header on sock %d",
3471                                conn->sock->fd);
3472                         status = RELAY_CONNECTION_STATUS_ERROR;
3473                 }
3474                 goto end;
3475         } else if (ret == 0) {
3476                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3477                 status = RELAY_CONNECTION_STATUS_CLOSED;
3478                 goto end;
3479         }
3480
3481         LTTNG_ASSERT(ret > 0);
3482         LTTNG_ASSERT(ret <= state->left_to_receive);
3483
3484         state->left_to_receive -= ret;
3485         state->received += ret;
3486
3487         if (state->left_to_receive > 0) {
3488                 /*
3489                  * Can't transition to the protocol's next state, wait to
3490                  * receive the rest of the header.
3491                  */
3492                 DBG3("Partial reception of control connection protocol header (received %" PRIu64
3493                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3494                      state->received,
3495                      state->left_to_receive,
3496                      conn->sock->fd);
3497                 goto end;
3498         }
3499
3500         /* Transition to next state: receiving the command's payload. */
3501         conn->protocol.ctrl.state_id = CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3502         memcpy(&header, reception_buffer->data, sizeof(header));
3503         header.circuit_id = be64toh(header.circuit_id);
3504         header.data_size = be64toh(header.data_size);
3505         header.cmd = be32toh(header.cmd);
3506         header.cmd_version = be32toh(header.cmd_version);
3507         memcpy(&conn->protocol.ctrl.state.receive_payload.header, &header, sizeof(header));
3508
3509         DBG("Done receiving control command header: fd = %i, cmd = %s, cmd_version = %" PRIu32
3510             ", payload size = %" PRIu64 " bytes",
3511             conn->sock->fd,
3512             lttcomm_relayd_command_str((enum lttcomm_relayd_command) header.cmd),
3513             header.cmd_version,
3514             header.data_size);
3515
3516         if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3517                 ERR("Command header indicates a payload (%" PRIu64
3518                     " bytes) that exceeds the maximal payload size allowed on a control connection.",
3519                     header.data_size);
3520                 status = RELAY_CONNECTION_STATUS_ERROR;
3521                 goto end;
3522         }
3523
3524         conn->protocol.ctrl.state.receive_payload.left_to_receive = header.data_size;
3525         conn->protocol.ctrl.state.receive_payload.received = 0;
3526         ret = lttng_dynamic_buffer_set_size(reception_buffer, header.data_size);
3527         if (ret) {
3528                 status = RELAY_CONNECTION_STATUS_ERROR;
3529                 goto end;
3530         }
3531
3532         if (header.data_size == 0) {
3533                 /*
3534                  * Manually invoke the next state as the poll loop
3535                  * will not wake-up to allow us to proceed further.
3536                  */
3537                 status = relay_process_control_receive_payload(conn);
3538         }
3539 end:
3540         return status;
3541 }
3542
3543 /*
3544  * Process the commands received on the control socket
3545  */
3546 static enum relay_connection_status relay_process_control(struct relay_connection *conn)
3547 {
3548         enum relay_connection_status status;
3549
3550         switch (conn->protocol.ctrl.state_id) {
3551         case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3552                 status = relay_process_control_receive_header(conn);
3553                 break;
3554         case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3555                 status = relay_process_control_receive_payload(conn);
3556                 break;
3557         default:
3558                 ERR("Unknown control connection protocol state encountered.");
3559                 abort();
3560         }
3561
3562         return status;
3563 }
3564
3565 static enum relay_connection_status relay_process_data_receive_header(struct relay_connection *conn)
3566 {
3567         int ret;
3568         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3569         struct data_connection_state_receive_header *state =
3570                 &conn->protocol.data.state.receive_header;
3571         struct lttcomm_relayd_data_hdr header;
3572         struct relay_stream *stream;
3573
3574         LTTNG_ASSERT(state->left_to_receive != 0);
3575
3576         ret = conn->sock->ops->recvmsg(conn->sock,
3577                                        state->header_reception_buffer + state->received,
3578                                        state->left_to_receive,
3579                                        MSG_DONTWAIT);
3580         if (ret < 0) {
3581                 DIAGNOSTIC_PUSH
3582                 DIAGNOSTIC_IGNORE_LOGICAL_OP
3583                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3584                         DIAGNOSTIC_POP
3585                         PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3586                         status = RELAY_CONNECTION_STATUS_ERROR;
3587                 }
3588                 goto end;
3589         } else if (ret == 0) {
3590                 /* Orderly shutdown. Not necessary to print an error. */
3591                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3592                 status = RELAY_CONNECTION_STATUS_CLOSED;
3593                 goto end;
3594         }
3595
3596         LTTNG_ASSERT(ret > 0);
3597         LTTNG_ASSERT(ret <= state->left_to_receive);
3598
3599         state->left_to_receive -= ret;
3600         state->received += ret;
3601
3602         if (state->left_to_receive > 0) {
3603                 /*
3604                  * Can't transition to the protocol's next state, wait to
3605                  * receive the rest of the header.
3606                  */
3607                 DBG3("Partial reception of data connection header (received %" PRIu64
3608                      " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3609                      state->received,
3610                      state->left_to_receive,
3611                      conn->sock->fd);
3612                 goto end;
3613         }
3614
3615         /* Transition to next state: receiving the payload. */
3616         conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3617
3618         memcpy(&header, state->header_reception_buffer, sizeof(header));
3619         header.circuit_id = be64toh(header.circuit_id);
3620         header.stream_id = be64toh(header.stream_id);
3621         header.data_size = be32toh(header.data_size);
3622         header.net_seq_num = be64toh(header.net_seq_num);
3623         header.padding_size = be32toh(header.padding_size);
3624         memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3625
3626         conn->protocol.data.state.receive_payload.left_to_receive = header.data_size;
3627         conn->protocol.data.state.receive_payload.received = 0;
3628         conn->protocol.data.state.receive_payload.rotate_index = false;
3629
3630         DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
3631             ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64
3632             ", padding_size = %" PRIu32,
3633             conn->sock->fd,
3634             header.circuit_id,
3635             header.stream_id,
3636             header.data_size,
3637             header.net_seq_num,
3638             header.padding_size);
3639
3640         stream = stream_get_by_id(header.stream_id);
3641         if (!stream) {
3642                 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3643                     header.stream_id);
3644                 /* Protocol error. */
3645                 status = RELAY_CONNECTION_STATUS_ERROR;
3646                 goto end;
3647         }
3648
3649         pthread_mutex_lock(&stream->lock);
3650         /* Prepare stream for the reception of a new packet. */
3651         ret = stream_init_packet(
3652                 stream, header.data_size, &conn->protocol.data.state.receive_payload.rotate_index);
3653         pthread_mutex_unlock(&stream->lock);
3654         if (ret) {
3655                 ERR("Failed to rotate stream output file");
3656                 status = RELAY_CONNECTION_STATUS_ERROR;
3657                 goto end_stream_unlock;
3658         }
3659
3660 end_stream_unlock:
3661         stream_put(stream);
3662 end:
3663         return status;
3664 }
3665
3666 static enum relay_connection_status
3667 relay_process_data_receive_payload(struct relay_connection *conn)
3668 {
3669         int ret;
3670         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3671         struct relay_stream *stream;
3672         struct data_connection_state_receive_payload *state =
3673                 &conn->protocol.data.state.receive_payload;
3674         const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3675         char data_buffer[chunk_size];
3676         bool partial_recv = false;
3677         bool new_stream = false, close_requested = false, index_flushed = false;
3678         uint64_t left_to_receive = state->left_to_receive;
3679         struct relay_session *session;
3680
3681         DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64
3682              " bytes received, %" PRIu64 " bytes left to receive",
3683              state->header.stream_id,
3684              state->header.net_seq_num,
3685              state->received,
3686              left_to_receive);
3687
3688         stream = stream_get_by_id(state->header.stream_id);
3689         if (!stream) {
3690                 /* Protocol error. */
3691                 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3692                     state->header.stream_id);
3693                 status = RELAY_CONNECTION_STATUS_ERROR;
3694                 goto end;
3695         }
3696
3697         pthread_mutex_lock(&stream->lock);
3698         session = stream->trace->session;
3699         if (!conn->session) {
3700                 ret = connection_set_session(conn, session);
3701                 if (ret) {
3702                         status = RELAY_CONNECTION_STATUS_ERROR;
3703                         goto end_stream_unlock;
3704                 }
3705         }
3706
3707         /*
3708          * The size of the "chunk" received on any iteration is bounded by:
3709          *   - the data left to receive,
3710          *   - the data immediately available on the socket,
3711          *   - the on-stack data buffer
3712          */
3713         while (left_to_receive > 0 && !partial_recv) {
3714                 size_t recv_size = std::min<uint64_t>(left_to_receive, chunk_size);
3715                 struct lttng_buffer_view packet_chunk;
3716
3717                 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, recv_size, MSG_DONTWAIT);
3718                 if (ret < 0) {
3719                         DIAGNOSTIC_PUSH
3720                         DIAGNOSTIC_IGNORE_LOGICAL_OP
3721                         if (errno != EAGAIN && errno != EWOULDBLOCK) {
3722                                 DIAGNOSTIC_POP
3723                                 PERROR("Socket %d error", conn->sock->fd);
3724                                 status = RELAY_CONNECTION_STATUS_ERROR;
3725                         }
3726                         goto end_stream_unlock;
3727                 } else if (ret == 0) {
3728                         /* No more data ready to be consumed on socket. */
3729                         DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3730                              state->header.stream_id);
3731                         status = RELAY_CONNECTION_STATUS_CLOSED;
3732                         break;
3733                 } else if (ret < (int) recv_size) {
3734                         /*
3735                          * All the data available on the socket has been
3736                          * consumed.
3737                          */
3738                         partial_recv = true;
3739                         recv_size = ret;
3740                 }
3741
3742                 packet_chunk = lttng_buffer_view_init(data_buffer, 0, recv_size);
3743                 LTTNG_ASSERT(packet_chunk.data);
3744
3745                 ret = stream_write(stream, &packet_chunk, 0);
3746                 if (ret) {
3747                         ERR("Relay error writing data to file");
3748                         status = RELAY_CONNECTION_STATUS_ERROR;
3749                         goto end_stream_unlock;
3750                 }
3751
3752                 left_to_receive -= recv_size;
3753                 state->received += recv_size;
3754                 state->left_to_receive = left_to_receive;
3755         }
3756
3757         if (state->left_to_receive > 0) {
3758                 /*
3759                  * Did not receive all the data expected, wait for more data to
3760                  * become available on the socket.
3761                  */
3762                 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64
3763                      " bytes received, %" PRIu64 " bytes left to receive",
3764                      state->header.stream_id,
3765                      state->received,
3766                      state->left_to_receive);
3767                 goto end_stream_unlock;
3768         }
3769
3770         ret = stream_write(stream, nullptr, state->header.padding_size);
3771         if (ret) {
3772                 status = RELAY_CONNECTION_STATUS_ERROR;
3773                 goto end_stream_unlock;
3774         }
3775
3776         if (session_streams_have_index(session)) {
3777                 ret = stream_update_index(stream,
3778                                           state->header.net_seq_num,
3779                                           state->rotate_index,
3780                                           &index_flushed,
3781                                           state->header.data_size + state->header.padding_size);
3782                 if (ret < 0) {
3783                         ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64
3784                             " ret %d",
3785                             stream->stream_handle,
3786                             state->header.net_seq_num,
3787                             ret);
3788                         status = RELAY_CONNECTION_STATUS_ERROR;
3789                         goto end_stream_unlock;
3790                 }
3791         }
3792
3793         if (stream->prev_data_seq == -1ULL) {
3794                 new_stream = true;
3795         }
3796
3797         ret = stream_complete_packet(stream,
3798                                      state->header.data_size + state->header.padding_size,
3799                                      state->header.net_seq_num,
3800                                      index_flushed);
3801         if (ret) {
3802                 status = RELAY_CONNECTION_STATUS_ERROR;
3803                 goto end_stream_unlock;
3804         }
3805
3806         /*
3807          * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3808          * contents of *state which are aliased (union) to the same location as
3809          * the new state. Don't use it beyond this point.
3810          */
3811         connection_reset_protocol_state(conn);
3812         state = nullptr;
3813
3814 end_stream_unlock:
3815         close_requested = stream->close_requested;
3816         pthread_mutex_unlock(&stream->lock);
3817         if (close_requested && left_to_receive == 0) {
3818                 try_stream_close(stream);
3819         }
3820
3821         if (new_stream) {
3822                 pthread_mutex_lock(&session->lock);
3823                 uatomic_set(&session->new_streams, 1);
3824                 pthread_mutex_unlock(&session->lock);
3825         }
3826
3827         stream_put(stream);
3828 end:
3829         return status;
3830 }
3831
3832 /*
3833  * relay_process_data: Process the data received on the data socket
3834  */
3835 static enum relay_connection_status relay_process_data(struct relay_connection *conn)
3836 {
3837         enum relay_connection_status status;
3838
3839         switch (conn->protocol.data.state_id) {
3840         case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3841                 status = relay_process_data_receive_header(conn);
3842                 break;
3843         case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3844                 status = relay_process_data_receive_payload(conn);
3845                 break;
3846         default:
3847                 ERR("Unexpected data connection communication state.");
3848                 abort();
3849         }
3850
3851         return status;
3852 }
3853
3854 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3855 {
3856         int ret;
3857
3858         (void) lttng_poll_del(events, pollfd);
3859
3860         ret = fd_tracker_close_unsuspendable_fd(
3861                 the_fd_tracker, &pollfd, 1, fd_tracker_util_close_fd, nullptr);
3862         if (ret < 0) {
3863                 ERR("Closing pollfd %d", pollfd);
3864         }
3865 }
3866
3867 static void relay_thread_close_connection(struct lttng_poll_event *events,
3868                                           int pollfd,
3869                                           struct relay_connection *conn)
3870 {
3871         const char *type_str;
3872
3873         switch (conn->type) {
3874         case RELAY_DATA:
3875                 type_str = "Data";
3876                 break;
3877         case RELAY_CONTROL:
3878                 type_str = "Control";
3879                 break;
3880         case RELAY_VIEWER_COMMAND:
3881                 type_str = "Viewer Command";
3882                 break;
3883         case RELAY_VIEWER_NOTIFICATION:
3884                 type_str = "Viewer Notification";
3885                 break;
3886         default:
3887                 type_str = "Unknown";
3888         }
3889         cleanup_connection_pollfd(events, pollfd);
3890         connection_put(conn);
3891         DBG("%s connection closed with %d", type_str, pollfd);
3892 }
3893
3894 /*
3895  * This thread does the actual work
3896  */
3897 static void *relay_thread_worker(void *data __attribute__((unused)))
3898 {
3899         int ret, err = -1, last_seen_data_fd = -1;
3900         uint32_t nb_fd;
3901         struct lttng_poll_event events;
3902         struct lttng_ht *relay_connections_ht;
3903         struct lttng_ht_iter iter;
3904         struct relay_connection *destroy_conn = nullptr;
3905
3906         DBG("[thread] Relay worker started");
3907
3908         rcu_register_thread();
3909
3910         health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3911
3912         if (testpoint(relayd_thread_worker)) {
3913                 goto error_testpoint;
3914         }
3915
3916         health_code_update();
3917
3918         /* table of connections indexed on socket */
3919         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3920         if (!relay_connections_ht) {
3921                 goto relay_connections_ht_error;
3922         }
3923
3924         ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3925         if (ret < 0) {
3926                 goto error_poll_create;
3927         }
3928
3929         ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3930         if (ret < 0) {
3931                 goto error;
3932         }
3933
3934 restart:
3935         while (true) {
3936                 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3937
3938                 health_code_update();
3939
3940                 /* Infinite blocking call, waiting for transmission */
3941                 DBG3("Relayd worker thread polling...");
3942                 health_poll_entry();
3943                 ret = lttng_poll_wait(&events, -1);
3944                 health_poll_exit();
3945                 if (ret < 0) {
3946                         /*
3947                          * Restart interrupted system call.
3948                          */
3949                         if (errno == EINTR) {
3950                                 goto restart;
3951                         }
3952                         goto error;
3953                 }
3954
3955                 nb_fd = ret;
3956
3957                 /*
3958                  * Process control. The control connection is
3959                  * prioritized so we don't starve it with high
3960                  * throughput tracing data on the data connection.
3961                  */
3962                 for (i = 0; i < nb_fd; i++) {
3963                         /* Fetch once the poll data */
3964                         const auto revents = LTTNG_POLL_GETEV(&events, i);
3965                         const auto pollfd = LTTNG_POLL_GETFD(&events, i);
3966
3967                         health_code_update();
3968
3969                         /* Activity on thread quit pipe, exiting. */
3970                         if (relayd_is_thread_quit_pipe(pollfd)) {
3971                                 DBG("Activity on thread quit pipe");
3972                                 err = 0;
3973                                 goto exit;
3974                         }
3975
3976                         /* Inspect the relay conn pipe for new connection */
3977                         if (pollfd == relay_conn_pipe[0]) {
3978                                 if (revents & LPOLLIN) {
3979                                         struct relay_connection *conn;
3980
3981                                         ret = lttng_read(relay_conn_pipe[0],
3982                                                          &conn,
3983                                                          sizeof(conn)); /* NOLINT sizeof used on a
3984                                                                            pointer. */
3985                                         if (ret < 0) {
3986                                                 goto error;
3987                                         }
3988                                         ret = lttng_poll_add(
3989                                                 &events, conn->sock->fd, LPOLLIN | LPOLLRDHUP);
3990                                         if (ret) {
3991                                                 ERR("Failed to add new connection file descriptor to poll set");
3992                                                 goto error;
3993                                         }
3994                                         connection_ht_add(relay_connections_ht, conn);
3995                                         DBG("Connection socket %d added", conn->sock->fd);
3996                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3997                                         ERR("Relay connection pipe error");
3998                                         goto error;
3999                                 } else {
4000                                         ERR("Unexpected poll events %u for sock %d",
4001                                             revents,
4002                                             pollfd);
4003                                         goto error;
4004                                 }
4005                         } else {
4006                                 struct relay_connection *ctrl_conn;
4007
4008                                 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4009                                 /* If not found, there is a synchronization issue. */
4010                                 LTTNG_ASSERT(ctrl_conn);
4011
4012                                 if (ctrl_conn->type == RELAY_DATA) {
4013                                         if (revents & LPOLLIN) {
4014                                                 /*
4015                                                  * Flag the last seen data fd not deleted. It will
4016                                                  * be used as the last seen fd if any fd gets
4017                                                  * deleted in this first loop.
4018                                                  */
4019                                                 last_notdel_data_fd = pollfd;
4020                                         }
4021                                         goto put_ctrl_connection;
4022                                 }
4023                                 LTTNG_ASSERT(ctrl_conn->type == RELAY_CONTROL);
4024
4025                                 if (revents & LPOLLIN) {
4026                                         enum relay_connection_status status;
4027
4028                                         status = relay_process_control(ctrl_conn);
4029                                         if (status != RELAY_CONNECTION_STATUS_OK) {
4030                                                 /*
4031                                                  * On socket error flag the session as aborted to
4032                                                  * force the cleanup of its stream otherwise it can
4033                                                  * leak during the lifetime of the relayd.
4034                                                  *
4035                                                  * This prevents situations in which streams can be
4036                                                  * left opened because an index was received, the
4037                                                  * control connection is closed, and the data
4038                                                  * connection is closed (uncleanly) before the
4039                                                  * packet's data provided.
4040                                                  *
4041                                                  * Since the control connection encountered an
4042                                                  * error, it is okay to be conservative and close
4043                                                  * the session right now as we can't rely on the
4044                                                  * protocol being respected anymore.
4045                                                  */
4046                                                 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4047                                                         session_abort(ctrl_conn->session);
4048                                                 }
4049
4050                                                 /* Clear the connection on error or close. */
4051                                                 relay_thread_close_connection(
4052                                                         &events, pollfd, ctrl_conn);
4053                                         }
4054                                         seen_control = 1;
4055                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4056                                         relay_thread_close_connection(&events, pollfd, ctrl_conn);
4057                                         if (last_seen_data_fd == pollfd) {
4058                                                 last_seen_data_fd = last_notdel_data_fd;
4059                                         }
4060                                 } else {
4061                                         ERR("Unexpected poll events %u for control sock %d",
4062                                             revents,
4063                                             pollfd);
4064                                         connection_put(ctrl_conn);
4065                                         goto error;
4066                                 }
4067                         put_ctrl_connection:
4068                                 connection_put(ctrl_conn);
4069                         }
4070                 }
4071
4072                 /*
4073                  * The last loop handled a control request, go back to poll to make
4074                  * sure we prioritise the control socket.
4075                  */
4076                 if (seen_control) {
4077                         continue;
4078                 }
4079
4080                 if (last_seen_data_fd >= 0) {
4081                         for (i = 0; i < nb_fd; i++) {
4082                                 int pollfd = LTTNG_POLL_GETFD(&events, i);
4083
4084                                 health_code_update();
4085
4086                                 if (last_seen_data_fd == pollfd) {
4087                                         idx = i;
4088                                         break;
4089                                 }
4090                         }
4091                 }
4092
4093                 /* Process data connection. */
4094                 for (i = idx + 1; i < nb_fd; i++) {
4095                         /* Fetch the poll data. */
4096                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
4097                         int pollfd = LTTNG_POLL_GETFD(&events, i);
4098                         struct relay_connection *data_conn;
4099
4100                         health_code_update();
4101
4102                         if (!revents) {
4103                                 /* No activity for this FD (poll implementation). */
4104                                 continue;
4105                         }
4106
4107                         /* Skip the command pipe. It's handled in the first loop. */
4108                         if (pollfd == relay_conn_pipe[0]) {
4109                                 continue;
4110                         }
4111
4112                         data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4113                         if (!data_conn) {
4114                                 /* Skip it. Might be removed before. */
4115                                 continue;
4116                         }
4117                         if (data_conn->type == RELAY_CONTROL) {
4118                                 goto put_data_connection;
4119                         }
4120                         LTTNG_ASSERT(data_conn->type == RELAY_DATA);
4121
4122                         if (revents & LPOLLIN) {
4123                                 enum relay_connection_status status;
4124
4125                                 status = relay_process_data(data_conn);
4126                                 /* Connection closed or error. */
4127                                 if (status != RELAY_CONNECTION_STATUS_OK) {
4128                                         /*
4129                                          * On socket error flag the session as aborted to force
4130                                          * the cleanup of its stream otherwise it can leak
4131                                          * during the lifetime of the relayd.
4132                                          *
4133                                          * This prevents situations in which streams can be
4134                                          * left opened because an index was received, the
4135                                          * control connection is closed, and the data
4136                                          * connection is closed (uncleanly) before the packet's
4137                                          * data provided.
4138                                          *
4139                                          * Since the data connection encountered an error,
4140                                          * it is okay to be conservative and close the
4141                                          * session right now as we can't rely on the protocol
4142                                          * being respected anymore.
4143                                          */
4144                                         if (status == RELAY_CONNECTION_STATUS_ERROR) {
4145                                                 session_abort(data_conn->session);
4146                                         }
4147                                         relay_thread_close_connection(&events, pollfd, data_conn);
4148                                         /*
4149                                          * Every goto restart call sets the last seen fd where
4150                                          * here we don't really care since we gracefully
4151                                          * continue the loop after the connection is deleted.
4152                                          */
4153                                 } else {
4154                                         /* Keep last seen port. */
4155                                         last_seen_data_fd = pollfd;
4156                                         connection_put(data_conn);
4157                                         goto restart;
4158                                 }
4159                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4160                                 relay_thread_close_connection(&events, pollfd, data_conn);
4161                         } else {
4162                                 ERR("Unknown poll events %u for data sock %d", revents, pollfd);
4163                         }
4164                 put_data_connection:
4165                         connection_put(data_conn);
4166                 }
4167                 last_seen_data_fd = -1;
4168         }
4169
4170         /* Normal exit, no error */
4171         ret = 0;
4172
4173 exit:
4174 error:
4175         /* Cleanup remaining connection object. */
4176         {
4177                 lttng::urcu::read_lock_guard read_lock;
4178
4179                 cds_lfht_for_each_entry (
4180                         relay_connections_ht->ht, &iter.iter, destroy_conn, sock_n.node) {
4181                         health_code_update();
4182
4183                         session_abort(destroy_conn->session);
4184
4185                         /*
4186                          * No need to grab another ref, because we own
4187                          * destroy_conn.
4188                          */
4189                         relay_thread_close_connection(
4190                                 &events, destroy_conn->sock->fd, destroy_conn);
4191                 }
4192         }
4193
4194         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4195 error_poll_create:
4196         lttng_ht_destroy(relay_connections_ht);
4197 relay_connections_ht_error:
4198         /* Close relay conn pipes */
4199         (void) fd_tracker_util_pipe_close(the_fd_tracker, relay_conn_pipe);
4200         if (err) {
4201                 DBG("Thread exited with error");
4202         }
4203         DBG("Worker thread cleanup complete");
4204 error_testpoint:
4205         if (err) {
4206                 health_error();
4207                 ERR("Health error occurred in %s", __func__);
4208         }
4209         health_unregister(health_relayd);
4210         rcu_unregister_thread();
4211         lttng_relay_stop_threads();
4212         return nullptr;
4213 }
4214
4215 /*
4216  * Create the relay command pipe to wake thread_manage_apps.
4217  * Closed in cleanup().
4218  */
4219 static int create_relay_conn_pipe()
4220 {
4221         return fd_tracker_util_pipe_open_cloexec(
4222                 the_fd_tracker, "Relayd connection pipe", relay_conn_pipe);
4223 }
4224
4225 static int stdio_open(void *data __attribute__((unused)), int *fds)
4226 {
4227         fds[0] = fileno(stdout);
4228         fds[1] = fileno(stderr);
4229         return 0;
4230 }
4231
4232 static int track_stdio()
4233 {
4234         int fds[2];
4235         const char *names[] = { "stdout", "stderr" };
4236
4237         return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds, names, 2, stdio_open, nullptr);
4238 }
4239
4240 /*
4241  * main
4242  */
4243 int main(int argc, char **argv)
4244 {
4245         bool thread_is_rcu_registered = false;
4246         int ret = 0, retval = 0;
4247         void *status;
4248         char *unlinked_file_directory_path = nullptr, *output_path = nullptr;
4249
4250         /* Parse environment variables */
4251         ret = parse_env_options();
4252         if (ret) {
4253                 retval = -1;
4254                 goto exit_options;
4255         }
4256
4257         /*
4258          * Parse arguments.
4259          * Command line arguments overwrite environment.
4260          */
4261         progname = argv[0];
4262         if (set_options(argc, argv)) {
4263                 retval = -1;
4264                 goto exit_options;
4265         }
4266
4267         if (set_signal_handler()) {
4268                 retval = -1;
4269                 goto exit_options;
4270         }
4271
4272         relayd_config_log();
4273
4274         if (opt_print_version) {
4275                 print_version();
4276                 retval = 0;
4277                 goto exit_options;
4278         }
4279
4280         ret = fclose(stdin);
4281         if (ret) {
4282                 PERROR("Failed to close stdin");
4283                 goto exit_options;
4284         }
4285
4286         DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4287
4288         /* Try to create directory if -o, --output is specified. */
4289         if (opt_output_path) {
4290                 if (*opt_output_path != '/') {
4291                         ERR("Please specify an absolute path for -o, --output PATH");
4292                         retval = -1;
4293                         goto exit_options;
4294                 }
4295
4296                 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG, -1, -1);
4297                 if (ret < 0) {
4298                         ERR("Unable to create %s", opt_output_path);
4299                         retval = -1;
4300                         goto exit_options;
4301                 }
4302         }
4303
4304         /* Daemonize */
4305         if (opt_daemon || opt_background) {
4306                 ret = lttng_daemonize(&child_ppid, &recv_child_signal, !opt_background);
4307                 if (ret < 0) {
4308                         retval = -1;
4309                         goto exit_options;
4310                 }
4311         }
4312
4313         if (opt_working_directory) {
4314                 ret = utils_change_working_directory(opt_working_directory);
4315                 if (ret) {
4316                         /* All errors are already logged. */
4317                         goto exit_options;
4318                 }
4319         }
4320
4321         sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4322         if (!sessiond_trace_chunk_registry) {
4323                 ERR("Failed to initialize session daemon trace chunk registry");
4324                 retval = -1;
4325                 goto exit_options;
4326         }
4327
4328         /*
4329          * The RCU thread registration (and use, through the fd-tracker's
4330          * creation) is done after the daemonization to allow us to not
4331          * deal with liburcu's fork() management as the call RCU needs to
4332          * be restored.
4333          */
4334         rcu_register_thread();
4335         thread_is_rcu_registered = true;
4336
4337         output_path = create_output_path("");
4338         if (!output_path) {
4339                 ERR("Failed to get output path");
4340                 retval = -1;
4341                 goto exit_options;
4342         }
4343         ret = asprintf(&unlinked_file_directory_path,
4344                        "%s/%s",
4345                        output_path,
4346                        DEFAULT_UNLINKED_FILES_DIRECTORY);
4347         free(output_path);
4348         if (ret < 0) {
4349                 ERR("Failed to format unlinked file directory path");
4350                 retval = -1;
4351                 goto exit_options;
4352         }
4353         the_fd_tracker = fd_tracker_create(unlinked_file_directory_path, lttng_opt_fd_pool_size);
4354         free(unlinked_file_directory_path);
4355         if (!the_fd_tracker) {
4356                 retval = -1;
4357                 goto exit_options;
4358         }
4359
4360         ret = track_stdio();
4361         if (ret) {
4362                 retval = -1;
4363                 goto exit_options;
4364         }
4365
4366         /* Initialize thread health monitoring */
4367         health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4368         if (!health_relayd) {
4369                 PERROR("health_app_create error");
4370                 retval = -1;
4371                 goto exit_options;
4372         }
4373
4374         /* Create thread quit pipe */
4375         if (relayd_init_thread_quit_pipe()) {
4376                 retval = -1;
4377                 goto exit_options;
4378         }
4379
4380         /* Setup the thread apps communication pipe. */
4381         if (create_relay_conn_pipe()) {
4382                 retval = -1;
4383                 goto exit_options;
4384         }
4385
4386         /* Init relay command queue. */
4387         cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4388
4389         /* Initialize communication library */
4390         lttcomm_init();
4391         lttcomm_inet_init();
4392
4393         /* tables of sessions indexed by session ID */
4394         sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4395         if (!sessions_ht) {
4396                 retval = -1;
4397                 goto exit_options;
4398         }
4399
4400         /* tables of streams indexed by stream ID */
4401         relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4402         if (!relay_streams_ht) {
4403                 retval = -1;
4404                 goto exit_options;
4405         }
4406
4407         /* tables of streams indexed by stream ID */
4408         viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4409         if (!viewer_streams_ht) {
4410                 retval = -1;
4411                 goto exit_options;
4412         }
4413
4414         ret = init_health_quit_pipe();
4415         if (ret) {
4416                 retval = -1;
4417                 goto exit_options;
4418         }
4419
4420         /* Create thread to manage the client socket */
4421         ret = pthread_create(&health_thread,
4422                              default_pthread_attr(),
4423                              thread_manage_health_relayd,
4424                              (void *) nullptr);
4425         if (ret) {
4426                 errno = ret;
4427                 PERROR("pthread_create health");
4428                 retval = -1;
4429                 goto exit_options;
4430         }
4431
4432         /* Setup the dispatcher thread */
4433         ret = pthread_create(&dispatcher_thread,
4434                              default_pthread_attr(),
4435                              relay_thread_dispatcher,
4436                              (void *) nullptr);
4437         if (ret) {
4438                 errno = ret;
4439                 PERROR("pthread_create dispatcher");
4440                 retval = -1;
4441                 goto exit_dispatcher_thread;
4442         }
4443
4444         /* Setup the worker thread */
4445         ret = pthread_create(&worker_thread, default_pthread_attr(), relay_thread_worker, nullptr);
4446         if (ret) {
4447                 errno = ret;
4448                 PERROR("pthread_create worker");
4449                 retval = -1;
4450                 goto exit_worker_thread;
4451         }
4452
4453         /* Setup the listener thread */
4454         ret = pthread_create(
4455                 &listener_thread, default_pthread_attr(), relay_thread_listener, (void *) nullptr);
4456         if (ret) {
4457                 errno = ret;
4458                 PERROR("pthread_create listener");
4459                 retval = -1;
4460                 goto exit_listener_thread;
4461         }
4462
4463         ret = relayd_live_create(live_uri);
4464         if (ret) {
4465                 ERR("Starting live viewer threads");
4466                 retval = -1;
4467                 goto exit_live;
4468         }
4469
4470         /*
4471          * This is where we start awaiting program completion (e.g. through
4472          * signal that asks threads to teardown).
4473          */
4474
4475         ret = relayd_live_join();
4476         if (ret) {
4477                 retval = -1;
4478         }
4479 exit_live:
4480
4481         ret = pthread_join(listener_thread, &status);
4482         if (ret) {
4483                 errno = ret;
4484                 PERROR("pthread_join listener_thread");
4485                 retval = -1;
4486         }
4487
4488 exit_listener_thread:
4489         ret = pthread_join(worker_thread, &status);
4490         if (ret) {
4491                 errno = ret;
4492                 PERROR("pthread_join worker_thread");
4493                 retval = -1;
4494         }
4495
4496 exit_worker_thread:
4497         ret = pthread_join(dispatcher_thread, &status);
4498         if (ret) {
4499                 errno = ret;
4500                 PERROR("pthread_join dispatcher_thread");
4501                 retval = -1;
4502         }
4503 exit_dispatcher_thread:
4504
4505         ret = pthread_join(health_thread, &status);
4506         if (ret) {
4507                 errno = ret;
4508                 PERROR("pthread_join health_thread");
4509                 retval = -1;
4510         }
4511 exit_options:
4512         /*
4513          * Wait for all pending call_rcu work to complete before tearing
4514          * down data structures. call_rcu worker may be trying to
4515          * perform lookups in those structures.
4516          */
4517         rcu_barrier();
4518         relayd_cleanup();
4519
4520         /* Ensure all prior call_rcu are done. */
4521         rcu_barrier();
4522
4523         if (thread_is_rcu_registered) {
4524                 rcu_unregister_thread();
4525         }
4526
4527         if (!retval) {
4528                 exit(EXIT_SUCCESS);
4529         } else {
4530                 exit(EXIT_FAILURE);
4531         }
4532 }