src/bin/lttng-relayd/main.c

   1 /*
   2  * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
   3  * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
   4  * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
   5  * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   6  *
   7  * SPDX-License-Identifier: GPL-2.0-only
   8  *
   9  */
  10
  11 #define _LGPL_SOURCE
  12 #include <getopt.h>
  13 #include <grp.h>
  14 #include <limits.h>
  15 #include <pthread.h>
  16 #include <signal.h>
  17 #include <stdio.h>
  18 #include <stdlib.h>
  19 #include <string.h>
  20 #include <sys/mman.h>
  21 #include <sys/mount.h>
  22 #include <sys/resource.h>
  23 #include <sys/socket.h>
  24 #include <sys/stat.h>
  25 #include <sys/types.h>
  26 #include <sys/wait.h>
  27 #include <sys/resource.h>
  28 #include <inttypes.h>
  29 #include <urcu/futex.h>
  30 #include <urcu/uatomic.h>
  31 #include <urcu/rculist.h>
  32 #include <unistd.h>
  33 #include <fcntl.h>
  34 #include <strings.h>
  35 #include <ctype.h>
  36
  37 #include <lttng/lttng.h>
  38 #include <common/common.h>
  39 #include <common/compat/poll.h>
  40 #include <common/compat/socket.h>
  41 #include <common/compat/endian.h>
  42 #include <common/compat/getenv.h>
  43 #include <common/defaults.h>
  44 #include <common/daemonize.h>
  45 #include <common/futex.h>
  46 #include <common/sessiond-comm/sessiond-comm.h>
  47 #include <common/sessiond-comm/inet.h>
  48 #include <common/sessiond-comm/relayd.h>
  49 #include <common/uri.h>
  50 #include <common/utils.h>
  51 #include <common/align.h>
  52 #include <common/config/session-config.h>
  53 #include <common/dynamic-buffer.h>
  54 #include <common/buffer-view.h>
  55 #include <common/string-utils/format.h>
  56 #include <common/fd-tracker/fd-tracker.h>
  57 #include <common/fd-tracker/utils.h>
  58
  59 #include "backward-compatibility-group-by.h"
  60 #include "cmd.h"
  61 #include "connection.h"
  62 #include "ctf-trace.h"
  63 #include "health-relayd.h"
  64 #include "index.h"
  65 #include "live.h"
  66 #include "lttng-relayd.h"
  67 #include "session.h"
  68 #include "sessiond-trace-chunks.h"
  69 #include "stream.h"
  70 #include "tcp_keep_alive.h"
  71 #include "testpoint.h"
  72 #include "tracefile-array.h"
  73 #include "utils.h"
  74 #include "version.h"
  75 #include "viewer-stream.h"
  76
  77 static const char *help_msg =
  78 #ifdef LTTNG_EMBED_HELP
  79 #include <lttng-relayd.8.h>
  80 #else
  81 NULL
  82 #endif
  83 ;
  84
  85 enum relay_connection_status {
  86         RELAY_CONNECTION_STATUS_OK,
  87         /* An error occurred while processing an event on the connection. */
  88         RELAY_CONNECTION_STATUS_ERROR,
  89         /* Connection closed/shutdown cleanly. */
  90         RELAY_CONNECTION_STATUS_CLOSED,
  91 };
  92
  93 /* command line options */
  94 char *opt_output_path, *opt_working_directory;
  95 static int opt_daemon, opt_background, opt_print_version, opt_allow_clear = 1;
  96 enum relay_group_output_by opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_UNKNOWN;
  97
  98 /*
  99  * We need to wait for listener and live listener threads, as well as
 100  * health check thread, before being ready to signal readiness.
 101  */
 102 #define NR_LTTNG_RELAY_READY    3
 103 static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
 104
 105 /* Size of receive buffer. */
 106 #define RECV_DATA_BUFFER_SIZE           65536
 107
 108 static int recv_child_signal;   /* Set to 1 when a SIGUSR1 signal is received. */
 109 static pid_t child_ppid;        /* Internal parent PID use with daemonize. */
 110
 111 static struct lttng_uri *control_uri;
 112 static struct lttng_uri *data_uri;
 113 static struct lttng_uri *live_uri;
 114
 115 const char *progname;
 116
 117 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
 118 static int tracing_group_name_override;
 119
 120 const char * const config_section_name = "relayd";
 121
 122 /*
 123  * Quit pipe for all threads. This permits a single cancellation point
 124  * for all threads when receiving an event on the pipe.
 125  */
 126 int thread_quit_pipe[2] = { -1, -1 };
 127
 128 /*
 129  * This pipe is used to inform the worker thread that a command is queued and
 130  * ready to be processed.
 131  */
 132 static int relay_conn_pipe[2] = { -1, -1 };
 133
 134 /* Shared between threads */
 135 static int dispatch_thread_exit;
 136
 137 static pthread_t listener_thread;
 138 static pthread_t dispatcher_thread;
 139 static pthread_t worker_thread;
 140 static pthread_t health_thread;
 141
 142 /*
 143  * last_relay_stream_id_lock protects last_relay_stream_id increment
 144  * atomicity on 32-bit architectures.
 145  */
 146 static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
 147 static uint64_t last_relay_stream_id;
 148
 149 /*
 150  * Relay command queue.
 151  *
 152  * The relay_thread_listener and relay_thread_dispatcher communicate with this
 153  * queue.
 154  */
 155 static struct relay_conn_queue relay_conn_queue;
 156
 157 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
 158 static unsigned int lttng_opt_fd_pool_size = -1;
 159
 160 /* Global relay stream hash table. */
 161 struct lttng_ht *relay_streams_ht;
 162
 163 /* Global relay viewer stream hash table. */
 164 struct lttng_ht *viewer_streams_ht;
 165
 166 /* Global relay sessions hash table. */
 167 struct lttng_ht *sessions_ht;
 168
 169 /* Relayd health monitoring */
 170 struct health_app *health_relayd;
 171
 172 struct sessiond_trace_chunk_registry *sessiond_trace_chunk_registry;
 173
 174 /* Global fd tracker. */
 175 struct fd_tracker *the_fd_tracker;
 176
 177 static struct option long_options[] = {
 178         { "control-port", 1, 0, 'C', },
 179         { "data-port", 1, 0, 'D', },
 180         { "live-port", 1, 0, 'L', },
 181         { "daemonize", 0, 0, 'd', },
 182         { "background", 0, 0, 'b', },
 183         { "group", 1, 0, 'g', },
 184         { "fd-pool-size", 1, 0, '\0', },
 185         { "help", 0, 0, 'h', },
 186         { "output", 1, 0, 'o', },
 187         { "verbose", 0, 0, 'v', },
 188         { "config", 1, 0, 'f' },
 189         { "version", 0, 0, 'V' },
 190         { "working-directory", 1, 0, 'w', },
 191         { "group-output-by-session", 0, 0, 's', },
 192         { "group-output-by-host", 0, 0, 'p', },
 193         { "disallow-clear", 0, 0, 'x' },
 194         { NULL, 0, 0, 0, },
 195 };
 196
 197 static const char *config_ignore_options[] = { "help", "config", "version" };
 198
 199 static void print_version(void) {
 200         fprintf(stdout, "%s\n", VERSION);
 201 }
 202
 203 static void relayd_config_log(void)
 204 {
 205         DBG("LTTng-relayd " VERSION " - " VERSION_NAME "%s%s",
 206                         GIT_VERSION[0] == '\0' ? "" : " - " GIT_VERSION,
 207                         EXTRA_VERSION_NAME[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME);
 208         if (EXTRA_VERSION_DESCRIPTION[0] != '\0') {
 209                 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION "\n");
 210         }
 211         if (EXTRA_VERSION_PATCHES[0] != '\0') {
 212                 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES "\n");
 213         }
 214 }
 215
 216 /*
 217  * Take an option from the getopt output and set it in the right variable to be
 218  * used later.
 219  *
 220  * Return 0 on success else a negative value.
 221  */
 222 static int set_option(int opt, const char *arg, const char *optname)
 223 {
 224         int ret;
 225
 226         switch (opt) {
 227         case 0:
 228                 if (!strcmp(optname, "fd-pool-size")) {
 229                         unsigned long v;
 230
 231                         errno = 0;
 232                         v = strtoul(arg, NULL, 0);
 233                         if (errno != 0 || !isdigit((unsigned char) arg[0])) {
 234                                 ERR("Wrong value in --fd-pool-size parameter: %s", arg);
 235                                 ret = -1;
 236                                 goto end;
 237                         }
 238                         if (v >= UINT_MAX) {
 239                                 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg);
 240                                 ret = -1;
 241                                 goto end;
 242                         }
 243                         lttng_opt_fd_pool_size = (unsigned int) v;
 244                 } else {
 245                         fprintf(stderr, "unknown option %s", optname);
 246                         if (arg) {
 247                                 fprintf(stderr, " with arg %s\n", arg);
 248                         }
 249                 }
 250                 break;
 251         case 'C':
 252                 if (lttng_is_setuid_setgid()) {
 253                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 254                                 "-C, --control-port");
 255                 } else {
 256                         ret = uri_parse(arg, &control_uri);
 257                         if (ret < 0) {
 258                                 ERR("Invalid control URI specified");
 259                                 goto end;
 260                         }
 261                         if (control_uri->port == 0) {
 262                                 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
 263                         }
 264                 }
 265                 break;
 266         case 'D':
 267                 if (lttng_is_setuid_setgid()) {
 268                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 269                                 "-D, -data-port");
 270                 } else {
 271                         ret = uri_parse(arg, &data_uri);
 272                         if (ret < 0) {
 273                                 ERR("Invalid data URI specified");
 274                                 goto end;
 275                         }
 276                         if (data_uri->port == 0) {
 277                                 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
 278                         }
 279                 }
 280                 break;
 281         case 'L':
 282                 if (lttng_is_setuid_setgid()) {
 283                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 284                                 "-L, -live-port");
 285                 } else {
 286                         ret = uri_parse(arg, &live_uri);
 287                         if (ret < 0) {
 288                                 ERR("Invalid live URI specified");
 289                                 goto end;
 290                         }
 291                         if (live_uri->port == 0) {
 292                                 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
 293                         }
 294                 }
 295                 break;
 296         case 'd':
 297                 opt_daemon = 1;
 298                 break;
 299         case 'b':
 300                 opt_background = 1;
 301                 break;
 302         case 'g':
 303                 if (lttng_is_setuid_setgid()) {
 304                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 305                                 "-g, --group");
 306                 } else {
 307                         tracing_group_name = strdup(arg);
 308                         if (tracing_group_name == NULL) {
 309                                 ret = -errno;
 310                                 PERROR("strdup");
 311                                 goto end;
 312                         }
 313                         tracing_group_name_override = 1;
 314                 }
 315                 break;
 316         case 'h':
 317                 ret = utils_show_help(8, "lttng-relayd", help_msg);
 318                 if (ret) {
 319                         ERR("Cannot show --help for `lttng-relayd`");
 320                         perror("exec");
 321                 }
 322                 exit(EXIT_FAILURE);
 323         case 'V':
 324                 opt_print_version = 1;
 325                 break;
 326         case 'o':
 327                 if (lttng_is_setuid_setgid()) {
 328                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 329                                 "-o, --output");
 330                 } else {
 331                         ret = asprintf(&opt_output_path, "%s", arg);
 332                         if (ret < 0) {
 333                                 ret = -errno;
 334                                 PERROR("asprintf opt_output_path");
 335                                 goto end;
 336                         }
 337                 }
 338                 break;
 339         case 'w':
 340                 if (lttng_is_setuid_setgid()) {
 341                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 342                                 "-w, --working-directory");
 343                 } else {
 344                         ret = asprintf(&opt_working_directory, "%s", arg);
 345                         if (ret < 0) {
 346                                 ret = -errno;
 347                                 PERROR("asprintf opt_working_directory");
 348                                 goto end;
 349                         }
 350                 }
 351                 break;
 352
 353         case 'v':
 354                 /* Verbose level can increase using multiple -v */
 355                 if (arg) {
 356                         lttng_opt_verbose = config_parse_value(arg);
 357                 } else {
 358                         /* Only 3 level of verbosity (-vvv). */
 359                         if (lttng_opt_verbose < 3) {
 360                                 lttng_opt_verbose += 1;
 361                         }
 362                 }
 363                 break;
 364         case 's':
 365                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 366                         ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
 367                         exit(EXIT_FAILURE);
 368                 }
 369                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_SESSION;
 370                 break;
 371         case 'p':
 372                 if (opt_group_output_by != RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 373                         ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
 374                         exit(EXIT_FAILURE);
 375                 }
 376                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 377                 break;
 378         case 'x':
 379                 /* Disallow clear */
 380                 opt_allow_clear = 0;
 381                 break;
 382         default:
 383                 /* Unknown option or other error.
 384                  * Error is printed by getopt, just return */
 385                 ret = -1;
 386                 goto end;
 387         }
 388
 389         /* All good. */
 390         ret = 0;
 391
 392 end:
 393         return ret;
 394 }
 395
 396 /*
 397  * config_entry_handler_cb used to handle options read from a config file.
 398  * See config_entry_handler_cb comment in common/config/session-config.h for the
 399  * return value conventions.
 400  */
 401 static int config_entry_handler(const struct config_entry *entry, void *unused)
 402 {
 403         int ret = 0, i;
 404
 405         if (!entry || !entry->name || !entry->value) {
 406                 ret = -EINVAL;
 407                 goto end;
 408         }
 409
 410         /* Check if the option is to be ignored */
 411         for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
 412                 if (!strcmp(entry->name, config_ignore_options[i])) {
 413                         goto end;
 414                 }
 415         }
 416
 417         for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
 418                 /* Ignore if entry name is not fully matched. */
 419                 if (strcmp(entry->name, long_options[i].name)) {
 420                         continue;
 421                 }
 422
 423                 /*
 424                  * If the option takes no argument on the command line,
 425                  * we have to check if the value is "true". We support
 426                  * non-zero numeric values, true, on and yes.
 427                  */
 428                 if (!long_options[i].has_arg) {
 429                         ret = config_parse_value(entry->value);
 430                         if (ret <= 0) {
 431                                 if (ret) {
 432                                         WARN("Invalid configuration value \"%s\" for option %s",
 433                                                         entry->value, entry->name);
 434                                 }
 435                                 /* False, skip boolean config option. */
 436                                 goto end;
 437                         }
 438                 }
 439
 440                 ret = set_option(long_options[i].val, entry->value, entry->name);
 441                 goto end;
 442         }
 443
 444         WARN("Unrecognized option \"%s\" in daemon configuration file.",
 445                         entry->name);
 446
 447 end:
 448         return ret;
 449 }
 450
 451 static int parse_env_options(void)
 452 {
 453         int ret = 0;
 454         char *value = NULL;
 455
 456         value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV);
 457         if (value) {
 458                 opt_working_directory = strdup(value);
 459                 if (!opt_working_directory) {
 460                         ERR("Failed to allocate working directory string (\"%s\")",
 461                                         value);
 462                         ret = -1;
 463                 }
 464         }
 465         return ret;
 466 }
 467
 468 static int set_fd_pool_size(void)
 469 {
 470         int ret = 0;
 471         struct rlimit rlimit;
 472
 473         ret = getrlimit(RLIMIT_NOFILE, &rlimit);
 474         if (ret) {
 475                 PERROR("Failed to get file descriptor limit");
 476                 ret = -1;
 477                 goto end;
 478         }
 479
 480         DBG("File descriptor count limits are %" PRIu64 " (soft) and %" PRIu64 " (hard)",
 481                         (uint64_t) rlimit.rlim_cur,
 482                         (uint64_t) rlimit.rlim_max);
 483         if (lttng_opt_fd_pool_size == -1) {
 484                 /* Use default value (soft limit - reserve). */
 485                 if (rlimit.rlim_cur < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 486                         ERR("The process' file number limit is too low (%" PRIu64 "). The process' file number limit must be set to at least %i.",
 487                                         (uint64_t) rlimit.rlim_cur, DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 488                         ret = -1;
 489                         goto end;
 490                 }
 491                 lttng_opt_fd_pool_size = rlimit.rlim_cur -
 492                                 DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 493                 goto end;
 494         }
 495
 496         if (lttng_opt_fd_pool_size < DEFAULT_RELAYD_MIN_FD_POOL_SIZE) {
 497                 ERR("File descriptor pool size must be set to at least %d",
 498                                 DEFAULT_RELAYD_MIN_FD_POOL_SIZE);
 499                 ret = -1;
 500                 goto end;
 501         }
 502
 503         if (lttng_opt_fd_pool_size > rlimit.rlim_cur) {
 504                 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64 ").",
 505                                 lttng_opt_fd_pool_size, (uint64_t) rlimit.rlim_cur);
 506                 ret = -1;
 507                 goto end;
 508         }
 509
 510         DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
 511                         lttng_opt_fd_pool_size,
 512                         lttng_opt_fd_pool_size - DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE);
 513         lttng_opt_fd_pool_size -= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE;
 514 end:
 515         return ret;
 516 }
 517
 518 static int set_options(int argc, char **argv)
 519 {
 520         int c, ret = 0, option_index = 0, retval = 0;
 521         int orig_optopt = optopt, orig_optind = optind;
 522         char *default_address, *optstring;
 523         const char *config_path = NULL;
 524
 525         optstring = utils_generate_optstring(long_options,
 526                         sizeof(long_options) / sizeof(struct option));
 527         if (!optstring) {
 528                 retval = -ENOMEM;
 529                 goto exit;
 530         }
 531
 532         /* Check for the --config option */
 533
 534         while ((c = getopt_long(argc, argv, optstring, long_options,
 535                                         &option_index)) != -1) {
 536                 if (c == '?') {
 537                         retval = -EINVAL;
 538                         goto exit;
 539                 } else if (c != 'f') {
 540                         continue;
 541                 }
 542
 543                 if (lttng_is_setuid_setgid()) {
 544                         WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 545                                 "-f, --config");
 546                 } else {
 547                         config_path = utils_expand_path(optarg);
 548                         if (!config_path) {
 549                                 ERR("Failed to resolve path: %s", optarg);
 550                         }
 551                 }
 552         }
 553
 554         ret = config_get_section_entries(config_path, config_section_name,
 555                         config_entry_handler, NULL);
 556         if (ret) {
 557                 if (ret > 0) {
 558                         ERR("Invalid configuration option at line %i", ret);
 559                 }
 560                 retval = -1;
 561                 goto exit;
 562         }
 563
 564         /* Reset getopt's global state */
 565         optopt = orig_optopt;
 566         optind = orig_optind;
 567         while (1) {
 568                 c = getopt_long(argc, argv, optstring, long_options, &option_index);
 569                 if (c == -1) {
 570                         break;
 571                 }
 572
 573                 ret = set_option(c, optarg, long_options[option_index].name);
 574                 if (ret < 0) {
 575                         retval = -1;
 576                         goto exit;
 577                 }
 578         }
 579
 580         /* assign default values */
 581         if (control_uri == NULL) {
 582                 ret = asprintf(&default_address,
 583                         "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
 584                         DEFAULT_NETWORK_CONTROL_PORT);
 585                 if (ret < 0) {
 586                         PERROR("asprintf default data address");
 587                         retval = -1;
 588                         goto exit;
 589                 }
 590
 591                 ret = uri_parse(default_address, &control_uri);
 592                 free(default_address);
 593                 if (ret < 0) {
 594                         ERR("Invalid control URI specified");
 595                         retval = -1;
 596                         goto exit;
 597                 }
 598         }
 599         if (data_uri == NULL) {
 600                 ret = asprintf(&default_address,
 601                         "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
 602                         DEFAULT_NETWORK_DATA_PORT);
 603                 if (ret < 0) {
 604                         PERROR("asprintf default data address");
 605                         retval = -1;
 606                         goto exit;
 607                 }
 608
 609                 ret = uri_parse(default_address, &data_uri);
 610                 free(default_address);
 611                 if (ret < 0) {
 612                         ERR("Invalid data URI specified");
 613                         retval = -1;
 614                         goto exit;
 615                 }
 616         }
 617         if (live_uri == NULL) {
 618                 ret = asprintf(&default_address,
 619                         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
 620                         DEFAULT_NETWORK_VIEWER_PORT);
 621                 if (ret < 0) {
 622                         PERROR("asprintf default viewer control address");
 623                         retval = -1;
 624                         goto exit;
 625                 }
 626
 627                 ret = uri_parse(default_address, &live_uri);
 628                 free(default_address);
 629                 if (ret < 0) {
 630                         ERR("Invalid viewer control URI specified");
 631                         retval = -1;
 632                         goto exit;
 633                 }
 634         }
 635         ret = set_fd_pool_size();
 636         if (ret) {
 637                 retval = -1;
 638                 goto exit;
 639         }
 640
 641         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
 642                 opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
 643         }
 644         if (opt_allow_clear) {
 645                 /* Check if env variable exists. */
 646                 const char *value = lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 647                 if (value) {
 648                         ret = config_parse_value(value);
 649                         if (ret < 0) {
 650                                 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV);
 651                                 retval = -1;
 652                                 goto exit;
 653                         }
 654                         opt_allow_clear = !ret;
 655                 }
 656         }
 657
 658 exit:
 659         free(optstring);
 660         return retval;
 661 }
 662
 663 static void print_global_objects(void)
 664 {
 665         print_viewer_streams();
 666         print_relay_streams();
 667         print_sessions();
 668 }
 669
 670 static int noop_close(void *data, int *fds)
 671 {
 672         return 0;
 673 }
 674
 675 static void untrack_stdio(void)
 676 {
 677         int fds[] = { fileno(stdout), fileno(stderr) };
 678
 679         /*
 680          * noop_close is used since we don't really want to close
 681          * the stdio output fds; we merely want to stop tracking them.
 682          */
 683         (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker,
 684                         fds, 2, noop_close, NULL);
 685 }
 686
 687 /*
 688  * Cleanup the daemon
 689  */
 690 static void relayd_cleanup(void)
 691 {
 692         print_global_objects();
 693
 694         DBG("Cleaning up");
 695
 696         if (viewer_streams_ht)
 697                 lttng_ht_destroy(viewer_streams_ht);
 698         if (relay_streams_ht)
 699                 lttng_ht_destroy(relay_streams_ht);
 700         if (sessions_ht)
 701                 lttng_ht_destroy(sessions_ht);
 702
 703         free(opt_output_path);
 704         free(opt_working_directory);
 705
 706         if (health_relayd) {
 707                 health_app_destroy(health_relayd);
 708         }
 709         /* Close thread quit pipes */
 710         if (health_quit_pipe[0] != -1) {
 711                 (void) fd_tracker_util_pipe_close(
 712                                 the_fd_tracker, health_quit_pipe);
 713         }
 714         if (thread_quit_pipe[0] != -1) {
 715                 (void) fd_tracker_util_pipe_close(
 716                                 the_fd_tracker, thread_quit_pipe);
 717         }
 718         if (sessiond_trace_chunk_registry) {
 719                 sessiond_trace_chunk_registry_destroy(
 720                                 sessiond_trace_chunk_registry);
 721         }
 722         if (the_fd_tracker) {
 723                 untrack_stdio();
 724                 /*
 725                  * fd_tracker_destroy() will log the contents of the fd-tracker
 726                  * if a leak is detected.
 727                  */
 728                 fd_tracker_destroy(the_fd_tracker);
 729         }
 730
 731         uri_free(control_uri);
 732         uri_free(data_uri);
 733         /* Live URI is freed in the live thread. */
 734
 735         if (tracing_group_name_override) {
 736                 free((void *) tracing_group_name);
 737         }
 738 }
 739
 740 /*
 741  * Write to writable pipe used to notify a thread.
 742  */
 743 static int notify_thread_pipe(int wpipe)
 744 {
 745         ssize_t ret;
 746
 747         ret = lttng_write(wpipe, "!", 1);
 748         if (ret < 1) {
 749                 PERROR("write poll pipe");
 750                 goto end;
 751         }
 752         ret = 0;
 753 end:
 754         return ret;
 755 }
 756
 757 static int notify_health_quit_pipe(int *pipe)
 758 {
 759         ssize_t ret;
 760
 761         ret = lttng_write(pipe[1], "4", 1);
 762         if (ret < 1) {
 763                 PERROR("write relay health quit");
 764                 goto end;
 765         }
 766         ret = 0;
 767 end:
 768         return ret;
 769 }
 770
 771 /*
 772  * Stop all relayd and relayd-live threads.
 773  */
 774 int lttng_relay_stop_threads(void)
 775 {
 776         int retval = 0;
 777
 778         /* Stopping all threads */
 779         DBG("Terminating all threads");
 780         if (notify_thread_pipe(thread_quit_pipe[1])) {
 781                 ERR("write error on thread quit pipe");
 782                 retval = -1;
 783         }
 784
 785         if (notify_health_quit_pipe(health_quit_pipe)) {
 786                 ERR("write error on health quit pipe");
 787         }
 788
 789         /* Dispatch thread */
 790         CMM_STORE_SHARED(dispatch_thread_exit, 1);
 791         futex_nto1_wake(&relay_conn_queue.futex);
 792
 793         if (relayd_live_stop()) {
 794                 ERR("Error stopping live threads");
 795                 retval = -1;
 796         }
 797         return retval;
 798 }
 799
 800 /*
 801  * Signal handler for the daemon
 802  *
 803  * Simply stop all worker threads, leaving main() return gracefully after
 804  * joining all threads and calling cleanup().
 805  */
 806 static void sighandler(int sig)
 807 {
 808         switch (sig) {
 809         case SIGINT:
 810                 DBG("SIGINT caught");
 811                 if (lttng_relay_stop_threads()) {
 812                         ERR("Error stopping threads");
 813                 }
 814                 break;
 815         case SIGTERM:
 816                 DBG("SIGTERM caught");
 817                 if (lttng_relay_stop_threads()) {
 818                         ERR("Error stopping threads");
 819                 }
 820                 break;
 821         case SIGUSR1:
 822                 CMM_STORE_SHARED(recv_child_signal, 1);
 823                 break;
 824         default:
 825                 break;
 826         }
 827 }
 828
 829 /*
 830  * Setup signal handler for :
 831  *              SIGINT, SIGTERM, SIGPIPE
 832  */
 833 static int set_signal_handler(void)
 834 {
 835         int ret = 0;
 836         struct sigaction sa;
 837         sigset_t sigset;
 838
 839         if ((ret = sigemptyset(&sigset)) < 0) {
 840                 PERROR("sigemptyset");
 841                 return ret;
 842         }
 843
 844         sa.sa_mask = sigset;
 845         sa.sa_flags = 0;
 846
 847         sa.sa_handler = sighandler;
 848         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
 849                 PERROR("sigaction");
 850                 return ret;
 851         }
 852
 853         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
 854                 PERROR("sigaction");
 855                 return ret;
 856         }
 857
 858         if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
 859                 PERROR("sigaction");
 860                 return ret;
 861         }
 862
 863         sa.sa_handler = SIG_IGN;
 864         if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
 865                 PERROR("sigaction");
 866                 return ret;
 867         }
 868
 869         DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
 870
 871         return ret;
 872 }
 873
 874 void lttng_relay_notify_ready(void)
 875 {
 876         /* Notify the parent of the fork() process that we are ready. */
 877         if (opt_daemon || opt_background) {
 878                 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
 879                         kill(child_ppid, SIGUSR1);
 880                 }
 881         }
 882 }
 883
 884 /*
 885  * Init thread quit pipe.
 886  *
 887  * Return -1 on error or 0 if all pipes are created.
 888  */
 889 static int init_thread_quit_pipe(void)
 890 {
 891         return fd_tracker_util_pipe_open_cloexec(
 892                         the_fd_tracker, "Quit pipe", thread_quit_pipe);
 893 }
 894
 895 /*
 896  * Init health quit pipe.
 897  *
 898  * Return -1 on error or 0 if all pipes are created.
 899  */
 900 static int init_health_quit_pipe(void)
 901 {
 902         return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
 903                         "Health quit pipe", health_quit_pipe);
 904 }
 905
 906 /*
 907  * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 908  */
 909 static int create_named_thread_poll_set(struct lttng_poll_event *events,
 910                 int size, const char *name)
 911 {
 912         int ret;
 913
 914         if (events == NULL || size == 0) {
 915                 ret = -1;
 916                 goto error;
 917         }
 918
 919         ret = fd_tracker_util_poll_create(the_fd_tracker,
 920                         name, events, 1, LTTNG_CLOEXEC);
 921         if (ret) {
 922                 PERROR("Failed to create \"%s\" poll file descriptor", name);
 923                 goto error;
 924         }
 925
 926         /* Add quit pipe */
 927         ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 928         if (ret < 0) {
 929                 goto error;
 930         }
 931
 932         return 0;
 933
 934 error:
 935         return ret;
 936 }
 937
 938 /*
 939  * Check if the thread quit pipe was triggered.
 940  *
 941  * Return 1 if it was triggered else 0;
 942  */
 943 static int check_thread_quit_pipe(int fd, uint32_t events)
 944 {
 945         if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
 946                 return 1;
 947         }
 948
 949         return 0;
 950 }
 951
 952 static int create_sock(void *data, int *out_fd)
 953 {
 954         int ret;
 955         struct lttcomm_sock *sock = data;
 956
 957         ret = lttcomm_create_sock(sock);
 958         if (ret < 0) {
 959                 goto end;
 960         }
 961
 962         *out_fd = sock->fd;
 963 end:
 964         return ret;
 965 }
 966
 967 static int close_sock(void *data, int *in_fd)
 968 {
 969         struct lttcomm_sock *sock = data;
 970
 971         return sock->ops->close(sock);
 972 }
 973
 974 static int accept_sock(void *data, int *out_fd)
 975 {
 976         int ret = 0;
 977         /* Socks is an array of in_sock, out_sock. */
 978         struct lttcomm_sock **socks = data;
 979         struct lttcomm_sock *in_sock = socks[0];
 980
 981         socks[1] = in_sock->ops->accept(in_sock);
 982         if (!socks[1]) {
 983                 ret = -1;
 984                 goto end;
 985         }
 986         *out_fd = socks[1]->fd;
 987 end:
 988         return ret;
 989 }
 990
 991 /*
 992  * Create and init socket from uri.
 993  */
 994 static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri,
 995                 const char *name)
 996 {
 997         int ret, sock_fd;
 998         struct lttcomm_sock *sock = NULL;
 999         char uri_str[PATH_MAX];
1000         char *formated_name = NULL;
1001
1002         sock = lttcomm_alloc_sock_from_uri(uri);
1003         if (sock == NULL) {
1004                 ERR("Allocating socket");
1005                 goto error;
1006         }
1007
1008         /*
1009          * Don't fail to create the socket if the name can't be built as it is
1010          * only used for debugging purposes.
1011          */
1012         ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
1013         uri_str[sizeof(uri_str) - 1] = '\0';
1014         if (ret >= 0) {
1015                 ret = asprintf(&formated_name, "%s socket @ %s", name,
1016                                 uri_str);
1017                 if (ret < 0) {
1018                         formated_name = NULL;
1019                 }
1020         }
1021
1022         ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
1023                         (const char **) (formated_name ? &formated_name : NULL),
1024                         1, create_sock, sock);
1025         if (ret) {
1026                 PERROR("Failed to open \"%s\" relay socket",
1027                                 formated_name ?: "Unknown");
1028                 goto error;
1029         }
1030         DBG("Listening on %s socket %d", name, sock->fd);
1031
1032         ret = sock->ops->bind(sock);
1033         if (ret < 0) {
1034                 PERROR("Failed to bind socket");
1035                 goto error;
1036         }
1037
1038         ret = sock->ops->listen(sock, -1);
1039         if (ret < 0) {
1040                 goto error;
1041
1042         }
1043
1044         free(formated_name);
1045         return sock;
1046
1047 error:
1048         if (sock) {
1049                 lttcomm_destroy_sock(sock);
1050         }
1051         free(formated_name);
1052         return NULL;
1053 }
1054
1055 static
1056 struct lttcomm_sock *accept_relayd_sock(struct lttcomm_sock *listening_sock,
1057                 const char *name)
1058 {
1059         int out_fd, ret;
1060         struct lttcomm_sock *socks[2] = { listening_sock, NULL };
1061         struct lttcomm_sock *new_sock = NULL;
1062
1063         ret = fd_tracker_open_unsuspendable_fd(
1064                         the_fd_tracker, &out_fd,
1065                         (const char **) &name,
1066                         1, accept_sock, &socks);
1067         if (ret) {
1068                 goto end;
1069         }
1070         new_sock = socks[1];
1071         DBG("%s accepted, socket %d", name, new_sock->fd);
1072 end:
1073         return new_sock;
1074 }
1075
1076 /*
1077  * This thread manages the listening for new connections on the network
1078  */
1079 static void *relay_thread_listener(void *data)
1080 {
1081         int i, ret, pollfd, err = -1;
1082         uint32_t revents, nb_fd;
1083         struct lttng_poll_event events;
1084         struct lttcomm_sock *control_sock, *data_sock;
1085
1086         DBG("[thread] Relay listener started");
1087
1088         rcu_register_thread();
1089         health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
1090
1091         health_code_update();
1092
1093         control_sock = relay_socket_create(control_uri, "Control listener");
1094         if (!control_sock) {
1095                 goto error_sock_control;
1096         }
1097
1098         data_sock = relay_socket_create(data_uri, "Data listener");
1099         if (!data_sock) {
1100                 goto error_sock_relay;
1101         }
1102
1103         /*
1104          * Pass 3 as size here for the thread quit pipe, control and
1105          * data socket.
1106          */
1107         ret = create_named_thread_poll_set(&events, 3, "Listener thread epoll");
1108         if (ret < 0) {
1109                 goto error_create_poll;
1110         }
1111
1112         /* Add the control socket */
1113         ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
1114         if (ret < 0) {
1115                 goto error_poll_add;
1116         }
1117
1118         /* Add the data socket */
1119         ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
1120         if (ret < 0) {
1121                 goto error_poll_add;
1122         }
1123
1124         lttng_relay_notify_ready();
1125
1126         if (testpoint(relayd_thread_listener)) {
1127                 goto error_testpoint;
1128         }
1129
1130         while (1) {
1131                 health_code_update();
1132
1133                 DBG("Listener accepting connections");
1134
1135 restart:
1136                 health_poll_entry();
1137                 ret = lttng_poll_wait(&events, -1);
1138                 health_poll_exit();
1139                 if (ret < 0) {
1140                         /*
1141                          * Restart interrupted system call.
1142                          */
1143                         if (errno == EINTR) {
1144                                 goto restart;
1145                         }
1146                         goto error;
1147                 }
1148
1149                 nb_fd = ret;
1150
1151                 DBG("Relay new connection received");
1152                 for (i = 0; i < nb_fd; i++) {
1153                         health_code_update();
1154
1155                         /* Fetch once the poll data */
1156                         revents = LTTNG_POLL_GETEV(&events, i);
1157                         pollfd = LTTNG_POLL_GETFD(&events, i);
1158
1159                         /* Thread quit pipe has been closed. Killing thread. */
1160                         ret = check_thread_quit_pipe(pollfd, revents);
1161                         if (ret) {
1162                                 err = 0;
1163                                 goto exit;
1164                         }
1165
1166                         if (revents & LPOLLIN) {
1167                                 /*
1168                                  * A new connection is requested, therefore a
1169                                  * sessiond/consumerd connection is allocated in
1170                                  * this thread, enqueued to a global queue and
1171                                  * dequeued (and freed) in the worker thread.
1172                                  */
1173                                 int val = 1;
1174                                 struct relay_connection *new_conn;
1175                                 struct lttcomm_sock *newsock = NULL;
1176                                 enum connection_type type;
1177
1178                                 if (pollfd == data_sock->fd) {
1179                                         type = RELAY_DATA;
1180                                         newsock = accept_relayd_sock(data_sock,
1181                                                         "Data socket to relayd");
1182                                 } else {
1183                                         assert(pollfd == control_sock->fd);
1184                                         type = RELAY_CONTROL;
1185                                         newsock = accept_relayd_sock(control_sock,
1186                                                         "Control socket to relayd");
1187                                 }
1188                                 if (!newsock) {
1189                                         PERROR("accepting sock");
1190                                         goto error;
1191                                 }
1192
1193                                 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
1194                                                 sizeof(val));
1195                                 if (ret < 0) {
1196                                         PERROR("setsockopt inet");
1197                                         lttcomm_destroy_sock(newsock);
1198                                         goto error;
1199                                 }
1200
1201                                 ret = socket_apply_keep_alive_config(newsock->fd);
1202                                 if (ret < 0) {
1203                                         ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1204                                                         newsock->fd);
1205                                         lttcomm_destroy_sock(newsock);
1206                                         goto error;
1207                                 }
1208
1209                                 new_conn = connection_create(newsock, type);
1210                                 if (!new_conn) {
1211                                         lttcomm_destroy_sock(newsock);
1212                                         goto error;
1213                                 }
1214
1215                                 /* Enqueue request for the dispatcher thread. */
1216                                 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
1217                                                  &new_conn->qnode);
1218
1219                                 /*
1220                                  * Wake the dispatch queue futex.
1221                                  * Implicit memory barrier with the
1222                                  * exchange in cds_wfcq_enqueue.
1223                                  */
1224                                 futex_nto1_wake(&relay_conn_queue.futex);
1225                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1226                                 ERR("socket poll error");
1227                                 goto error;
1228                         } else {
1229                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1230                                 goto error;
1231                         }
1232                 }
1233         }
1234
1235 exit:
1236 error:
1237 error_poll_add:
1238 error_testpoint:
1239         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
1240 error_create_poll:
1241         if (data_sock->fd >= 0) {
1242                 int data_sock_fd = data_sock->fd;
1243
1244                 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1245                                 &data_sock_fd, 1, close_sock,
1246                                 data_sock);
1247                 if (ret) {
1248                         PERROR("Failed to close the data listener socket file descriptor");
1249                 }
1250                 data_sock->fd = -1;
1251         }
1252         lttcomm_destroy_sock(data_sock);
1253 error_sock_relay:
1254         if (control_sock->fd >= 0) {
1255                 int control_sock_fd = control_sock->fd;
1256
1257                 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
1258                                 &control_sock_fd, 1, close_sock,
1259                                 control_sock);
1260                 if (ret) {
1261                         PERROR("Failed to close the control listener socket file descriptor");
1262                 }
1263                 control_sock->fd = -1;
1264         }
1265         lttcomm_destroy_sock(control_sock);
1266 error_sock_control:
1267         if (err) {
1268                 health_error();
1269                 ERR("Health error occurred in %s", __func__);
1270         }
1271         health_unregister(health_relayd);
1272         rcu_unregister_thread();
1273         DBG("Relay listener thread cleanup complete");
1274         lttng_relay_stop_threads();
1275         return NULL;
1276 }
1277
1278 /*
1279  * This thread manages the dispatching of the requests to worker threads
1280  */
1281 static void *relay_thread_dispatcher(void *data)
1282 {
1283         int err = -1;
1284         ssize_t ret;
1285         struct cds_wfcq_node *node;
1286         struct relay_connection *new_conn = NULL;
1287
1288         DBG("[thread] Relay dispatcher started");
1289
1290         health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
1291
1292         if (testpoint(relayd_thread_dispatcher)) {
1293                 goto error_testpoint;
1294         }
1295
1296         health_code_update();
1297
1298         for (;;) {
1299                 health_code_update();
1300
1301                 /* Atomically prepare the queue futex */
1302                 futex_nto1_prepare(&relay_conn_queue.futex);
1303
1304                 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1305                         break;
1306                 }
1307
1308                 do {
1309                         health_code_update();
1310
1311                         /* Dequeue commands */
1312                         node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1313                                                          &relay_conn_queue.tail);
1314                         if (node == NULL) {
1315                                 DBG("Woken up but nothing in the relay command queue");
1316                                 /* Continue thread execution */
1317                                 break;
1318                         }
1319                         new_conn = caa_container_of(node, struct relay_connection, qnode);
1320
1321                         DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
1322
1323                         /*
1324                          * Inform worker thread of the new request. This
1325                          * call is blocking so we can be assured that
1326                          * the data will be read at some point in time
1327                          * or wait to the end of the world :)
1328                          */
1329                         ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1330                         if (ret < 0) {
1331                                 PERROR("write connection pipe");
1332                                 connection_put(new_conn);
1333                                 goto error;
1334                         }
1335                 } while (node != NULL);
1336
1337                 /* Futex wait on queue. Blocking call on futex() */
1338                 health_poll_entry();
1339                 futex_nto1_wait(&relay_conn_queue.futex);
1340                 health_poll_exit();
1341         }
1342
1343         /* Normal exit, no error */
1344         err = 0;
1345
1346 error:
1347 error_testpoint:
1348         if (err) {
1349                 health_error();
1350                 ERR("Health error occurred in %s", __func__);
1351         }
1352         health_unregister(health_relayd);
1353         DBG("Dispatch thread dying");
1354         lttng_relay_stop_threads();
1355         return NULL;
1356 }
1357
1358 static bool session_streams_have_index(const struct relay_session *session)
1359 {
1360         return session->minor >= 4 && !session->snapshot;
1361 }
1362
1363 /*
1364  * Handle the RELAYD_CREATE_SESSION command.
1365  *
1366  * On success, send back the session id or else return a negative value.
1367  */
1368 static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr,
1369                 struct relay_connection *conn,
1370                 const struct lttng_buffer_view *payload)
1371 {
1372         int ret = 0;
1373         ssize_t send_ret;
1374         struct relay_session *session = NULL;
1375         struct lttcomm_relayd_create_session_reply_2_11 reply = {};
1376         char session_name[LTTNG_NAME_MAX] = {};
1377         char hostname[LTTNG_HOST_NAME_MAX] = {};
1378         uint32_t live_timer = 0;
1379         bool snapshot = false;
1380         bool session_name_contains_creation_timestamp = false;
1381         /* Left nil for peers < 2.11. */
1382         char base_path[LTTNG_PATH_MAX] = {};
1383         lttng_uuid sessiond_uuid = {};
1384         LTTNG_OPTIONAL(uint64_t) id_sessiond = {};
1385         LTTNG_OPTIONAL(uint64_t) current_chunk_id = {};
1386         LTTNG_OPTIONAL(time_t) creation_time = {};
1387         struct lttng_dynamic_buffer reply_payload;
1388
1389         lttng_dynamic_buffer_init(&reply_payload);
1390
1391         if (conn->minor < 4) {
1392                 /* From 2.1 to 2.3 */
1393                 ret = 0;
1394         } else if (conn->minor >= 4 && conn->minor < 11) {
1395                 /* From 2.4 to 2.10 */
1396                 ret = cmd_create_session_2_4(payload, session_name,
1397                         hostname, &live_timer, &snapshot);
1398         } else {
1399                 bool has_current_chunk;
1400                 uint64_t current_chunk_id_value;
1401                 time_t creation_time_value;
1402                 uint64_t id_sessiond_value;
1403
1404                 /* From 2.11 to ... */
1405                 ret = cmd_create_session_2_11(payload, session_name, hostname,
1406                                 base_path, &live_timer, &snapshot, &id_sessiond_value,
1407                                 sessiond_uuid, &has_current_chunk,
1408                                 &current_chunk_id_value, &creation_time_value,
1409                                 &session_name_contains_creation_timestamp);
1410                 if (lttng_uuid_is_nil(sessiond_uuid)) {
1411                         /* The nil UUID is reserved for pre-2.11 clients. */
1412                         ERR("Illegal nil UUID announced by peer in create session command");
1413                         ret = -1;
1414                         goto send_reply;
1415                 }
1416                 LTTNG_OPTIONAL_SET(&id_sessiond, id_sessiond_value);
1417                 LTTNG_OPTIONAL_SET(&creation_time, creation_time_value);
1418                 if (has_current_chunk) {
1419                         LTTNG_OPTIONAL_SET(&current_chunk_id,
1420                                         current_chunk_id_value);
1421                 }
1422         }
1423
1424         if (ret < 0) {
1425                 goto send_reply;
1426         }
1427
1428         session = session_create(session_name, hostname, base_path, live_timer,
1429                         snapshot, sessiond_uuid,
1430                         id_sessiond.is_set ? &id_sessiond.value : NULL,
1431                         current_chunk_id.is_set ? &current_chunk_id.value : NULL,
1432                         creation_time.is_set ? &creation_time.value : NULL,
1433                         conn->major, conn->minor,
1434                         session_name_contains_creation_timestamp);
1435         if (!session) {
1436                 ret = -1;
1437                 goto send_reply;
1438         }
1439         assert(!conn->session);
1440         conn->session = session;
1441         DBG("Created session %" PRIu64, session->id);
1442
1443         reply.generic.session_id = htobe64(session->id);
1444
1445 send_reply:
1446         if (ret < 0) {
1447                 reply.generic.ret_code = htobe32(LTTNG_ERR_FATAL);
1448         } else {
1449                 reply.generic.ret_code = htobe32(LTTNG_OK);
1450         }
1451
1452         if (conn->minor < 11) {
1453                 /* From 2.1 to 2.10 */
1454                 ret = lttng_dynamic_buffer_append(&reply_payload,
1455                                 &reply.generic, sizeof(reply.generic));
1456                 if (ret) {
1457                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1458                         ret = -1;
1459                         goto end;
1460                 }
1461         } else {
1462                 const uint32_t output_path_length =
1463                                 session ? strlen(session->output_path) + 1 : 0;
1464
1465                 reply.output_path_length = htobe32(output_path_length);
1466                 ret = lttng_dynamic_buffer_append(
1467                                 &reply_payload, &reply, sizeof(reply));
1468                 if (ret) {
1469                         ERR("Failed to append \"create session\" command reply header to payload buffer");
1470                         goto end;
1471                 }
1472
1473                 if (output_path_length) {
1474                         ret = lttng_dynamic_buffer_append(&reply_payload,
1475                                         session->output_path,
1476                                         output_path_length);
1477                         if (ret) {
1478                                 ERR("Failed to append \"create session\" command reply path to payload buffer");
1479                                 goto end;
1480                         }
1481                 }
1482         }
1483
1484         send_ret = conn->sock->ops->sendmsg(conn->sock, reply_payload.data,
1485                         reply_payload.size, 0);
1486         if (send_ret < (ssize_t) reply_payload.size) {
1487                 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1488                                 reply_payload.size, send_ret);
1489                 ret = -1;
1490         }
1491 end:
1492         if (ret < 0 && session) {
1493                 session_put(session);
1494         }
1495         lttng_dynamic_buffer_reset(&reply_payload);
1496         return ret;
1497 }
1498
1499 /*
1500  * When we have received all the streams and the metadata for a channel,
1501  * we make them visible to the viewer threads.
1502  */
1503 static void publish_connection_local_streams(struct relay_connection *conn)
1504 {
1505         struct relay_stream *stream;
1506         struct relay_session *session = conn->session;
1507
1508         /*
1509          * We publish all streams belonging to a session atomically wrt
1510          * session lock.
1511          */
1512         pthread_mutex_lock(&session->lock);
1513         rcu_read_lock();
1514         cds_list_for_each_entry_rcu(stream, &session->recv_list,
1515                         recv_node) {
1516                 stream_publish(stream);
1517         }
1518         rcu_read_unlock();
1519
1520         /*
1521          * Inform the viewer that there are new streams in the session.
1522          */
1523         if (session->viewer_attached) {
1524                 uatomic_set(&session->new_streams, 1);
1525         }
1526         pthread_mutex_unlock(&session->lock);
1527 }
1528
1529 static int conform_channel_path(char *channel_path)
1530 {
1531         int ret = 0;
1532
1533         if (strstr("../", channel_path)) {
1534                 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1535                                 channel_path);
1536                 ret = -1;
1537                 goto end;
1538         }
1539
1540         if (*channel_path == '/') {
1541                 const size_t len = strlen(channel_path);
1542
1543                 /*
1544                  * Channel paths from peers prior to 2.11 are expressed as an
1545                  * absolute path that is, in reality, relative to the relay
1546                  * daemon's output directory. Remove the leading slash so it
1547                  * is correctly interpreted as a relative path later on.
1548                  *
1549                  * len (and not len - 1) is used to copy the trailing NULL.
1550                  */
1551                 bcopy(channel_path + 1, channel_path, len);
1552         }
1553 end:
1554         return ret;
1555 }
1556
1557 /*
1558  * relay_add_stream: allocate a new stream for a session
1559  */
1560 static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1561                 struct relay_connection *conn,
1562                 const struct lttng_buffer_view *payload)
1563 {
1564         int ret;
1565         ssize_t send_ret;
1566         struct relay_session *session = conn->session;
1567         struct relay_stream *stream = NULL;
1568         struct lttcomm_relayd_status_stream reply;
1569         struct ctf_trace *trace = NULL;
1570         uint64_t stream_handle = -1ULL;
1571         char *path_name = NULL, *channel_name = NULL;
1572         uint64_t tracefile_size = 0, tracefile_count = 0;
1573         LTTNG_OPTIONAL(uint64_t) stream_chunk_id = {};
1574
1575         if (!session || !conn->version_check_done) {
1576                 ERR("Trying to add a stream before version check");
1577                 ret = -1;
1578                 goto end_no_session;
1579         }
1580
1581         if (session->minor == 1) {
1582                 /* For 2.1 */
1583                 ret = cmd_recv_stream_2_1(payload, &path_name,
1584                         &channel_name);
1585         } else if (session->minor > 1 && session->minor < 11) {
1586                 /* From 2.2 to 2.10 */
1587                 ret = cmd_recv_stream_2_2(payload, &path_name,
1588                         &channel_name, &tracefile_size, &tracefile_count);
1589         } else {
1590                 /* From 2.11 to ... */
1591                 ret = cmd_recv_stream_2_11(payload, &path_name,
1592                         &channel_name, &tracefile_size, &tracefile_count,
1593                         &stream_chunk_id.value);
1594                 stream_chunk_id.is_set = true;
1595         }
1596
1597         if (ret < 0) {
1598                 goto send_reply;
1599         }
1600
1601         if (conform_channel_path(path_name)) {
1602                 goto send_reply;
1603         }
1604
1605         /*
1606          * Backward compatibility for --group-output-by-session.
1607          * Prior to lttng 2.11, the complete path is passed by the stream.
1608          * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1609          * >=2.11 the chunk is responsible for the output path. When dealing
1610          * with producer < 2.11 the chunk output_path is the root output path
1611          * and the stream carries the complete path (path_name).
1612          * To support --group-output-by-session with older producer (<2.11), we
1613          * need to craft the path based on the stream path.
1614          */
1615         if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_SESSION) {
1616                 if (conn->minor < 4) {
1617                         /*
1618                          * From 2.1 to 2.3, the session_name is not passed on
1619                          * the RELAYD_CREATE_SESSION command. The session name
1620                          * is necessary to detect the presence of a base_path
1621                          * inside the stream path. Without it we cannot perform
1622                          * a valid group-output-by-session transformation.
1623                          */
1624                         WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1625                              " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1626                                         session->id, path_name);
1627                 } else if (conn->minor >= 4 && conn->minor < 11) {
1628                         char *group_by_session_path_name;
1629
1630                         assert(session->session_name[0] != '\0');
1631
1632                         group_by_session_path_name =
1633                                         backward_compat_group_by_session(
1634                                                         path_name,
1635                                                         session->session_name,
1636                                                         session->creation_time.value);
1637                         if (!group_by_session_path_name) {
1638                                 ERR("Failed to apply group by session to stream of session %" PRIu64,
1639                                                 session->id);
1640                                 goto send_reply;
1641                         }
1642
1643                         DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1644                                         path_name, group_by_session_path_name);
1645
1646                         free(path_name);
1647                         path_name = group_by_session_path_name;
1648                 }
1649         }
1650
1651         trace = ctf_trace_get_by_path_or_create(session, path_name);
1652         if (!trace) {
1653                 goto send_reply;
1654         }
1655
1656         /* This stream here has one reference on the trace. */
1657         pthread_mutex_lock(&last_relay_stream_id_lock);
1658         stream_handle = ++last_relay_stream_id;
1659         pthread_mutex_unlock(&last_relay_stream_id_lock);
1660
1661         /* We pass ownership of path_name and channel_name. */
1662         stream = stream_create(trace, stream_handle, path_name,
1663                 channel_name, tracefile_size, tracefile_count);
1664         path_name = NULL;
1665         channel_name = NULL;
1666
1667         /*
1668          * Streams are the owners of their trace. Reference to trace is
1669          * kept within stream_create().
1670          */
1671         ctf_trace_put(trace);
1672
1673 send_reply:
1674         memset(&reply, 0, sizeof(reply));
1675         reply.handle = htobe64(stream_handle);
1676         if (!stream) {
1677                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1678         } else {
1679                 reply.ret_code = htobe32(LTTNG_OK);
1680         }
1681
1682         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1683                         sizeof(struct lttcomm_relayd_status_stream), 0);
1684         if (send_ret < (ssize_t) sizeof(reply)) {
1685                 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1686                                 send_ret);
1687                 ret = -1;
1688         }
1689
1690 end_no_session:
1691         free(path_name);
1692         free(channel_name);
1693         return ret;
1694 }
1695
1696 /*
1697  * relay_close_stream: close a specific stream
1698  */
1699 static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1700                 struct relay_connection *conn,
1701                 const struct lttng_buffer_view *payload)
1702 {
1703         int ret;
1704         ssize_t send_ret;
1705         struct relay_session *session = conn->session;
1706         struct lttcomm_relayd_close_stream stream_info;
1707         struct lttcomm_relayd_generic_reply reply;
1708         struct relay_stream *stream;
1709
1710         DBG("Close stream received");
1711
1712         if (!session || !conn->version_check_done) {
1713                 ERR("Trying to close a stream before version check");
1714                 ret = -1;
1715                 goto end_no_session;
1716         }
1717
1718         if (payload->size < sizeof(stream_info)) {
1719                 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1720                                 sizeof(stream_info), payload->size);
1721                 ret = -1;
1722                 goto end_no_session;
1723         }
1724         memcpy(&stream_info, payload->data, sizeof(stream_info));
1725         stream_info.stream_id = be64toh(stream_info.stream_id);
1726         stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
1727
1728         stream = stream_get_by_id(stream_info.stream_id);
1729         if (!stream) {
1730                 ret = -1;
1731                 goto end;
1732         }
1733
1734         /*
1735          * Set last_net_seq_num before the close flag. Required by data
1736          * pending check.
1737          */
1738         pthread_mutex_lock(&stream->lock);
1739         stream->last_net_seq_num = stream_info.last_net_seq_num;
1740         pthread_mutex_unlock(&stream->lock);
1741
1742         /*
1743          * This is one of the conditions which may trigger a stream close
1744          * with the others being:
1745          *     1) A close command is received for a stream
1746          *     2) The control connection owning the stream is closed
1747          *     3) We have received all of the stream's data _after_ a close
1748          *        request.
1749          */
1750         try_stream_close(stream);
1751         if (stream->is_metadata) {
1752                 struct relay_viewer_stream *vstream;
1753
1754                 vstream = viewer_stream_get_by_id(stream->stream_handle);
1755                 if (vstream) {
1756                         if (stream->no_new_metadata_notified) {
1757                                 /*
1758                                  * Since all the metadata has been sent to the
1759                                  * viewer and that we have a request to close
1760                                  * its stream, we can safely teardown the
1761                                  * corresponding metadata viewer stream.
1762                                  */
1763                                 viewer_stream_put(vstream);
1764                         }
1765                         /* Put local reference. */
1766                         viewer_stream_put(vstream);
1767                 }
1768         }
1769         stream_put(stream);
1770         ret = 0;
1771
1772 end:
1773         memset(&reply, 0, sizeof(reply));
1774         if (ret < 0) {
1775                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1776         } else {
1777                 reply.ret_code = htobe32(LTTNG_OK);
1778         }
1779         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1780                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1781         if (send_ret < (ssize_t) sizeof(reply)) {
1782                 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1783                                 send_ret);
1784                 ret = -1;
1785         }
1786
1787 end_no_session:
1788         return ret;
1789 }
1790
1791 /*
1792  * relay_reset_metadata: reset a metadata stream
1793  */
1794 static
1795 int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1796                 struct relay_connection *conn,
1797                 const struct lttng_buffer_view *payload)
1798 {
1799         int ret;
1800         ssize_t send_ret;
1801         struct relay_session *session = conn->session;
1802         struct lttcomm_relayd_reset_metadata stream_info;
1803         struct lttcomm_relayd_generic_reply reply;
1804         struct relay_stream *stream;
1805
1806         DBG("Reset metadata received");
1807
1808         if (!session || !conn->version_check_done) {
1809                 ERR("Trying to reset a metadata stream before version check");
1810                 ret = -1;
1811                 goto end_no_session;
1812         }
1813
1814         if (payload->size < sizeof(stream_info)) {
1815                 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1816                                 sizeof(stream_info), payload->size);
1817                 ret = -1;
1818                 goto end_no_session;
1819         }
1820         memcpy(&stream_info, payload->data, sizeof(stream_info));
1821         stream_info.stream_id = be64toh(stream_info.stream_id);
1822         stream_info.version = be64toh(stream_info.version);
1823
1824         DBG("Update metadata to version %" PRIu64, stream_info.version);
1825
1826         /* Unsupported for live sessions for now. */
1827         if (session->live_timer != 0) {
1828                 ret = -1;
1829                 goto end;
1830         }
1831
1832         stream = stream_get_by_id(stream_info.stream_id);
1833         if (!stream) {
1834                 ret = -1;
1835                 goto end;
1836         }
1837         pthread_mutex_lock(&stream->lock);
1838         if (!stream->is_metadata) {
1839                 ret = -1;
1840                 goto end_unlock;
1841         }
1842
1843         ret = stream_reset_file(stream);
1844         if (ret < 0) {
1845                 ERR("Failed to reset metadata stream %" PRIu64
1846                                 ": stream_path = %s, channel = %s",
1847                                 stream->stream_handle, stream->path_name,
1848                                 stream->channel_name);
1849                 goto end_unlock;
1850         }
1851 end_unlock:
1852         pthread_mutex_unlock(&stream->lock);
1853         stream_put(stream);
1854
1855 end:
1856         memset(&reply, 0, sizeof(reply));
1857         if (ret < 0) {
1858                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1859         } else {
1860                 reply.ret_code = htobe32(LTTNG_OK);
1861         }
1862         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1863                         sizeof(struct lttcomm_relayd_generic_reply), 0);
1864         if (send_ret < (ssize_t) sizeof(reply)) {
1865                 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1866                                 send_ret);
1867                 ret = -1;
1868         }
1869
1870 end_no_session:
1871         return ret;
1872 }
1873
1874 /*
1875  * relay_unknown_command: send -1 if received unknown command
1876  */
1877 static void relay_unknown_command(struct relay_connection *conn)
1878 {
1879         struct lttcomm_relayd_generic_reply reply;
1880         ssize_t send_ret;
1881
1882         memset(&reply, 0, sizeof(reply));
1883         reply.ret_code = htobe32(LTTNG_ERR_UNK);
1884         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1885         if (send_ret < sizeof(reply)) {
1886                 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
1887         }
1888 }
1889
1890 /*
1891  * relay_start: send an acknowledgment to the client to tell if we are
1892  * ready to receive data. We are ready if a session is established.
1893  */
1894 static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr,
1895                 struct relay_connection *conn,
1896                 const struct lttng_buffer_view *payload)
1897 {
1898         int ret = 0;
1899         ssize_t send_ret;
1900         struct lttcomm_relayd_generic_reply reply;
1901         struct relay_session *session = conn->session;
1902
1903         if (!session) {
1904                 DBG("Trying to start the streaming without a session established");
1905                 ret = htobe32(LTTNG_ERR_UNK);
1906         }
1907
1908         memset(&reply, 0, sizeof(reply));
1909         reply.ret_code = htobe32(LTTNG_OK);
1910         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1911                         sizeof(reply), 0);
1912         if (send_ret < (ssize_t) sizeof(reply)) {
1913                 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1914                                 send_ret);
1915                 ret = -1;
1916         }
1917
1918         return ret;
1919 }
1920
1921 /*
1922  * relay_recv_metadata: receive the metadata for the session.
1923  */
1924 static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1925                 struct relay_connection *conn,
1926                 const struct lttng_buffer_view *payload)
1927 {
1928         int ret = 0;
1929         struct relay_session *session = conn->session;
1930         struct lttcomm_relayd_metadata_payload metadata_payload_header;
1931         struct relay_stream *metadata_stream;
1932         uint64_t metadata_payload_size;
1933         struct lttng_buffer_view packet_view;
1934
1935         if (!session) {
1936                 ERR("Metadata sent before version check");
1937                 ret = -1;
1938                 goto end;
1939         }
1940
1941         if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1942                 ERR("Incorrect data size");
1943                 ret = -1;
1944                 goto end;
1945         }
1946         metadata_payload_size = recv_hdr->data_size -
1947                         sizeof(struct lttcomm_relayd_metadata_payload);
1948
1949         memcpy(&metadata_payload_header, payload->data,
1950                         sizeof(metadata_payload_header));
1951         metadata_payload_header.stream_id = be64toh(
1952                         metadata_payload_header.stream_id);
1953         metadata_payload_header.padding_size = be32toh(
1954                         metadata_payload_header.padding_size);
1955
1956         metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
1957         if (!metadata_stream) {
1958                 ret = -1;
1959                 goto end;
1960         }
1961
1962         packet_view = lttng_buffer_view_from_view(payload,
1963                         sizeof(metadata_payload_header), metadata_payload_size);
1964         if (!packet_view.data) {
1965                 ERR("Invalid metadata packet length announced by header");
1966                 ret = -1;
1967                 goto end_put;
1968         }
1969
1970         pthread_mutex_lock(&metadata_stream->lock);
1971         ret = stream_write(metadata_stream, &packet_view,
1972                         metadata_payload_header.padding_size);
1973         pthread_mutex_unlock(&metadata_stream->lock);
1974         if (ret){
1975                 ret = -1;
1976                 goto end_put;
1977         }
1978 end_put:
1979         stream_put(metadata_stream);
1980 end:
1981         return ret;
1982 }
1983
1984 /*
1985  * relay_send_version: send relayd version number
1986  */
1987 static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
1988                 struct relay_connection *conn,
1989                 const struct lttng_buffer_view *payload)
1990 {
1991         int ret;
1992         ssize_t send_ret;
1993         struct lttcomm_relayd_version reply, msg;
1994         bool compatible = true;
1995
1996         conn->version_check_done = true;
1997
1998         /* Get version from the other side. */
1999         if (payload->size < sizeof(msg)) {
2000                 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
2001                                 sizeof(msg), payload->size);
2002                 ret = -1;
2003                 goto end;
2004         }
2005
2006         memcpy(&msg, payload->data, sizeof(msg));
2007         msg.major = be32toh(msg.major);
2008         msg.minor = be32toh(msg.minor);
2009
2010         memset(&reply, 0, sizeof(reply));
2011         reply.major = RELAYD_VERSION_COMM_MAJOR;
2012         reply.minor = RELAYD_VERSION_COMM_MINOR;
2013
2014         /* Major versions must be the same */
2015         if (reply.major != msg.major) {
2016                 DBG("Incompatible major versions (%u vs %u), deleting session",
2017                                 reply.major, msg.major);
2018                 compatible = false;
2019         }
2020
2021         conn->major = reply.major;
2022         /* We adapt to the lowest compatible version */
2023         if (reply.minor <= msg.minor) {
2024                 conn->minor = reply.minor;
2025         } else {
2026                 conn->minor = msg.minor;
2027         }
2028
2029         reply.major = htobe32(reply.major);
2030         reply.minor = htobe32(reply.minor);
2031         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2032                         sizeof(reply), 0);
2033         if (send_ret < (ssize_t) sizeof(reply)) {
2034                 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2035                                 send_ret);
2036                 ret = -1;
2037                 goto end;
2038         } else {
2039                 ret = 0;
2040         }
2041
2042         if (!compatible) {
2043                 ret = -1;
2044                 goto end;
2045         }
2046
2047         DBG("Version check done using protocol %u.%u", conn->major,
2048                         conn->minor);
2049
2050 end:
2051         return ret;
2052 }
2053
2054 /*
2055  * Check for data pending for a given stream id from the session daemon.
2056  */
2057 static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2058                 struct relay_connection *conn,
2059                 const struct lttng_buffer_view *payload)
2060 {
2061         struct relay_session *session = conn->session;
2062         struct lttcomm_relayd_data_pending msg;
2063         struct lttcomm_relayd_generic_reply reply;
2064         struct relay_stream *stream;
2065         ssize_t send_ret;
2066         int ret;
2067         uint64_t stream_seq;
2068
2069         DBG("Data pending command received");
2070
2071         if (!session || !conn->version_check_done) {
2072                 ERR("Trying to check for data before version check");
2073                 ret = -1;
2074                 goto end_no_session;
2075         }
2076
2077         if (payload->size < sizeof(msg)) {
2078                 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2079                                 sizeof(msg), payload->size);
2080                 ret = -1;
2081                 goto end_no_session;
2082         }
2083         memcpy(&msg, payload->data, sizeof(msg));
2084         msg.stream_id = be64toh(msg.stream_id);
2085         msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
2086
2087         stream = stream_get_by_id(msg.stream_id);
2088         if (stream == NULL) {
2089                 ret = -1;
2090                 goto end;
2091         }
2092
2093         pthread_mutex_lock(&stream->lock);
2094
2095         if (session_streams_have_index(session)) {
2096                 /*
2097                  * Ensure that both the index and stream data have been
2098                  * flushed up to the requested point.
2099                  */
2100                 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2101         } else {
2102                 stream_seq = stream->prev_data_seq;
2103         }
2104         DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
2105                         ", prev_index_seq %" PRIu64
2106                         ", and last_seq %" PRIu64, msg.stream_id,
2107                         stream->prev_data_seq, stream->prev_index_seq,
2108                         msg.last_net_seq_num);
2109
2110         /* Avoid wrapping issue */
2111         if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
2112                 /* Data has in fact been written and is NOT pending */
2113                 ret = 0;
2114         } else {
2115                 /* Data still being streamed thus pending */
2116                 ret = 1;
2117         }
2118
2119         stream->data_pending_check_done = true;
2120         pthread_mutex_unlock(&stream->lock);
2121
2122         stream_put(stream);
2123 end:
2124
2125         memset(&reply, 0, sizeof(reply));
2126         reply.ret_code = htobe32(ret);
2127         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2128         if (send_ret < (ssize_t) sizeof(reply)) {
2129                 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2130                                 send_ret);
2131                 ret = -1;
2132         }
2133
2134 end_no_session:
2135         return ret;
2136 }
2137
2138 /*
2139  * Wait for the control socket to reach a quiescent state.
2140  *
2141  * Note that for now, when receiving this command from the session
2142  * daemon, this means that every subsequent commands or data received on
2143  * the control socket has been handled. So, this is why we simply return
2144  * OK here.
2145  */
2146 static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
2147                 struct relay_connection *conn,
2148                 const struct lttng_buffer_view *payload)
2149 {
2150         int ret;
2151         ssize_t send_ret;
2152         struct relay_stream *stream;
2153         struct lttcomm_relayd_quiescent_control msg;
2154         struct lttcomm_relayd_generic_reply reply;
2155
2156         DBG("Checking quiescent state on control socket");
2157
2158         if (!conn->session || !conn->version_check_done) {
2159                 ERR("Trying to check for data before version check");
2160                 ret = -1;
2161                 goto end_no_session;
2162         }
2163
2164         if (payload->size < sizeof(msg)) {
2165                 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2166                                 sizeof(msg), payload->size);
2167                 ret = -1;
2168                 goto end_no_session;
2169         }
2170         memcpy(&msg, payload->data, sizeof(msg));
2171         msg.stream_id = be64toh(msg.stream_id);
2172
2173         stream = stream_get_by_id(msg.stream_id);
2174         if (!stream) {
2175                 goto reply;
2176         }
2177         pthread_mutex_lock(&stream->lock);
2178         stream->data_pending_check_done = true;
2179         pthread_mutex_unlock(&stream->lock);
2180
2181         DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
2182         stream_put(stream);
2183 reply:
2184         memset(&reply, 0, sizeof(reply));
2185         reply.ret_code = htobe32(LTTNG_OK);
2186         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2187         if (send_ret < (ssize_t) sizeof(reply)) {
2188                 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2189                                 send_ret);
2190                 ret = -1;
2191         } else {
2192                 ret = 0;
2193         }
2194
2195 end_no_session:
2196         return ret;
2197 }
2198
2199 /*
2200  * Initialize a data pending command. This means that a consumer is about
2201  * to ask for data pending for each stream it holds. Simply iterate over
2202  * all streams of a session and set the data_pending_check_done flag.
2203  *
2204  * This command returns to the client a LTTNG_OK code.
2205  */
2206 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2207                 struct relay_connection *conn,
2208                 const struct lttng_buffer_view *payload)
2209 {
2210         int ret;
2211         ssize_t send_ret;
2212         struct lttng_ht_iter iter;
2213         struct lttcomm_relayd_begin_data_pending msg;
2214         struct lttcomm_relayd_generic_reply reply;
2215         struct relay_stream *stream;
2216
2217         assert(recv_hdr);
2218         assert(conn);
2219
2220         DBG("Init streams for data pending");
2221
2222         if (!conn->session || !conn->version_check_done) {
2223                 ERR("Trying to check for data before version check");
2224                 ret = -1;
2225                 goto end_no_session;
2226         }
2227
2228         if (payload->size < sizeof(msg)) {
2229                 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2230                                 sizeof(msg), payload->size);
2231                 ret = -1;
2232                 goto end_no_session;
2233         }
2234         memcpy(&msg, payload->data, sizeof(msg));
2235         msg.session_id = be64toh(msg.session_id);
2236
2237         /*
2238          * Iterate over all streams to set the begin data pending flag.
2239          * For now, the streams are indexed by stream handle so we have
2240          * to iterate over all streams to find the one associated with
2241          * the right session_id.
2242          */
2243         rcu_read_lock();
2244         cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2245                         node.node) {
2246                 if (!stream_get(stream)) {
2247                         continue;
2248                 }
2249                 if (stream->trace->session->id == msg.session_id) {
2250                         pthread_mutex_lock(&stream->lock);
2251                         stream->data_pending_check_done = false;
2252                         pthread_mutex_unlock(&stream->lock);
2253                         DBG("Set begin data pending flag to stream %" PRIu64,
2254                                         stream->stream_handle);
2255                 }
2256                 stream_put(stream);
2257         }
2258         rcu_read_unlock();
2259
2260         memset(&reply, 0, sizeof(reply));
2261         /* All good, send back reply. */
2262         reply.ret_code = htobe32(LTTNG_OK);
2263
2264         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2265         if (send_ret < (ssize_t) sizeof(reply)) {
2266                 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2267                         send_ret);
2268                 ret = -1;
2269         } else {
2270                 ret = 0;
2271         }
2272
2273 end_no_session:
2274         return ret;
2275 }
2276
2277 /*
2278  * End data pending command. This will check, for a given session id, if
2279  * each stream associated with it has its data_pending_check_done flag
2280  * set. If not, this means that the client lost track of the stream but
2281  * the data is still being streamed on our side. In this case, we inform
2282  * the client that data is in flight.
2283  *
2284  * Return to the client if there is data in flight or not with a ret_code.
2285  */
2286 static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2287                 struct relay_connection *conn,
2288                 const struct lttng_buffer_view *payload)
2289 {
2290         int ret;
2291         ssize_t send_ret;
2292         struct lttng_ht_iter iter;
2293         struct lttcomm_relayd_end_data_pending msg;
2294         struct lttcomm_relayd_generic_reply reply;
2295         struct relay_stream *stream;
2296         uint32_t is_data_inflight = 0;
2297
2298         DBG("End data pending command");
2299
2300         if (!conn->session || !conn->version_check_done) {
2301                 ERR("Trying to check for data before version check");
2302                 ret = -1;
2303                 goto end_no_session;
2304         }
2305
2306         if (payload->size < sizeof(msg)) {
2307                 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2308                                 sizeof(msg), payload->size);
2309                 ret = -1;
2310                 goto end_no_session;
2311         }
2312         memcpy(&msg, payload->data, sizeof(msg));
2313         msg.session_id = be64toh(msg.session_id);
2314
2315         /*
2316          * Iterate over all streams to see if the begin data pending
2317          * flag is set.
2318          */
2319         rcu_read_lock();
2320         cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2321                         node.node) {
2322                 if (!stream_get(stream)) {
2323                         continue;
2324                 }
2325                 if (stream->trace->session->id != msg.session_id) {
2326                         stream_put(stream);
2327                         continue;
2328                 }
2329                 pthread_mutex_lock(&stream->lock);
2330                 if (!stream->data_pending_check_done) {
2331                         uint64_t stream_seq;
2332
2333                         if (session_streams_have_index(conn->session)) {
2334                                 /*
2335                                  * Ensure that both the index and stream data have been
2336                                  * flushed up to the requested point.
2337                                  */
2338                                 stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
2339                         } else {
2340                                 stream_seq = stream->prev_data_seq;
2341                         }
2342                         if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
2343                                 is_data_inflight = 1;
2344                                 DBG("Data is still in flight for stream %" PRIu64,
2345                                                 stream->stream_handle);
2346                                 pthread_mutex_unlock(&stream->lock);
2347                                 stream_put(stream);
2348                                 break;
2349                         }
2350                 }
2351                 pthread_mutex_unlock(&stream->lock);
2352                 stream_put(stream);
2353         }
2354         rcu_read_unlock();
2355
2356         memset(&reply, 0, sizeof(reply));
2357         /* All good, send back reply. */
2358         reply.ret_code = htobe32(is_data_inflight);
2359
2360         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2361         if (send_ret < (ssize_t) sizeof(reply)) {
2362                 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2363                         send_ret);
2364                 ret = -1;
2365         } else {
2366                 ret = 0;
2367         }
2368
2369 end_no_session:
2370         return ret;
2371 }
2372
2373 /*
2374  * Receive an index for a specific stream.
2375  *
2376  * Return 0 on success else a negative value.
2377  */
2378 static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
2379                 struct relay_connection *conn,
2380                 const struct lttng_buffer_view *payload)
2381 {
2382         int ret;
2383         ssize_t send_ret;
2384         struct relay_session *session = conn->session;
2385         struct lttcomm_relayd_index index_info;
2386         struct lttcomm_relayd_generic_reply reply;
2387         struct relay_stream *stream;
2388         size_t msg_len;
2389
2390         assert(conn);
2391
2392         DBG("Relay receiving index");
2393
2394         if (!session || !conn->version_check_done) {
2395                 ERR("Trying to close a stream before version check");
2396                 ret = -1;
2397                 goto end_no_session;
2398         }
2399
2400         msg_len = lttcomm_relayd_index_len(
2401                         lttng_to_index_major(conn->major, conn->minor),
2402                         lttng_to_index_minor(conn->major, conn->minor));
2403         if (payload->size < msg_len) {
2404                 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2405                                 msg_len, payload->size);
2406                 ret = -1;
2407                 goto end_no_session;
2408         }
2409         memcpy(&index_info, payload->data, msg_len);
2410         index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2411         index_info.net_seq_num = be64toh(index_info.net_seq_num);
2412         index_info.packet_size = be64toh(index_info.packet_size);
2413         index_info.content_size = be64toh(index_info.content_size);
2414         index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2415         index_info.timestamp_end = be64toh(index_info.timestamp_end);
2416         index_info.events_discarded = be64toh(index_info.events_discarded);
2417         index_info.stream_id = be64toh(index_info.stream_id);
2418
2419         if (conn->minor >= 8) {
2420                 index_info.stream_instance_id =
2421                                 be64toh(index_info.stream_instance_id);
2422                 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2423         } else {
2424                 index_info.stream_instance_id = -1ULL;
2425                 index_info.packet_seq_num = -1ULL;
2426         }
2427
2428         stream = stream_get_by_id(index_info.relay_stream_id);
2429         if (!stream) {
2430                 ERR("stream_get_by_id not found");
2431                 ret = -1;
2432                 goto end;
2433         }
2434
2435         pthread_mutex_lock(&stream->lock);
2436         ret = stream_add_index(stream, &index_info);
2437         pthread_mutex_unlock(&stream->lock);
2438         if (ret) {
2439                 goto end_stream_put;
2440         }
2441
2442 end_stream_put:
2443         stream_put(stream);
2444 end:
2445         memset(&reply, 0, sizeof(reply));
2446         if (ret < 0) {
2447                 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2448         } else {
2449                 reply.ret_code = htobe32(LTTNG_OK);
2450         }
2451         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2452         if (send_ret < (ssize_t) sizeof(reply)) {
2453                 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2454                 ret = -1;
2455         }
2456
2457 end_no_session:
2458         return ret;
2459 }
2460
2461 /*
2462  * Receive the streams_sent message.
2463  *
2464  * Return 0 on success else a negative value.
2465  */
2466 static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
2467                 struct relay_connection *conn,
2468                 const struct lttng_buffer_view *payload)
2469 {
2470         int ret;
2471         ssize_t send_ret;
2472         struct lttcomm_relayd_generic_reply reply;
2473
2474         assert(conn);
2475
2476         DBG("Relay receiving streams_sent");
2477
2478         if (!conn->session || !conn->version_check_done) {
2479                 ERR("Trying to close a stream before version check");
2480                 ret = -1;
2481                 goto end_no_session;
2482         }
2483
2484         /*
2485          * Publish every pending stream in the connection recv list which are
2486          * now ready to be used by the viewer.
2487          */
2488         publish_connection_local_streams(conn);
2489
2490         memset(&reply, 0, sizeof(reply));
2491         reply.ret_code = htobe32(LTTNG_OK);
2492         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2493         if (send_ret < (ssize_t) sizeof(reply)) {
2494                 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2495                         send_ret);
2496                 ret = -1;
2497         } else {
2498                 /* Success. */
2499                 ret = 0;
2500         }
2501
2502 end_no_session:
2503         return ret;
2504 }
2505
2506 static ssize_t relay_unpack_rotate_streams_header(
2507                 const struct lttng_buffer_view *payload,
2508                 struct lttcomm_relayd_rotate_streams *_rotate_streams)
2509 {
2510         struct lttcomm_relayd_rotate_streams rotate_streams;
2511         /*
2512          * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2513          * This is the smallest version of this structure, but it can be larger;
2514          * this variable is updated once the proper size of the structure is known.
2515          *
2516          * See comment at the declaration of this structure for more information.
2517          */
2518         ssize_t header_len = sizeof(struct lttcomm_relayd_rotate_streams_packed);
2519         size_t expected_payload_size_no_padding,
2520                 expected_payload_size_3_bytes_padding,
2521                 expected_payload_size_7_bytes_padding;
2522
2523         if (payload->size < header_len) {
2524                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2525                                 header_len, payload->size);
2526                 goto error;
2527         }
2528
2529         /*
2530          * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2531          * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2532          *
2533          * We start by "unpacking" `stream_count` to figure out the padding length
2534          * emited by our peer.
2535          */
2536         memcpy(&rotate_streams.stream_count, payload->data,
2537                         sizeof(rotate_streams.stream_count));
2538         rotate_streams = (typeof(rotate_streams)) {
2539                 .stream_count = be32toh(rotate_streams.stream_count),
2540         };
2541
2542         /*
2543          * Payload size expected given the possible padding lengths in
2544          * `struct lttcomm_relayd_rotate_streams`.
2545          */
2546         expected_payload_size_no_padding = (rotate_streams.stream_count *
2547                 sizeof(*rotate_streams.rotation_positions)) +
2548                 sizeof(struct lttcomm_relayd_rotate_streams_packed);
2549         expected_payload_size_3_bytes_padding = (rotate_streams.stream_count *
2550                 sizeof(*rotate_streams.rotation_positions)) +
2551                 sizeof(struct lttcomm_relayd_rotate_streams_3_bytes_padding);
2552         expected_payload_size_7_bytes_padding = (rotate_streams.stream_count *
2553                 sizeof(*rotate_streams.rotation_positions)) +
2554                 sizeof(struct lttcomm_relayd_rotate_streams_7_bytes_padding);
2555
2556         if (payload->size == expected_payload_size_no_padding) {
2557                 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams;
2558
2559                 /*
2560                  * This handles cases where someone might build with
2561                  * -fpack-struct or any other toolchain that wouldn't produce
2562                  * padding to align `value`.
2563                  */
2564                 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2565
2566                 header_len = sizeof(packed_rotate_streams);
2567                 memcpy(&packed_rotate_streams, payload->data, header_len);
2568
2569                 /* Unpack the packed structure to the natively-packed version. */
2570                 *_rotate_streams = (typeof(*_rotate_streams)) {
2571                         .stream_count = be32toh(packed_rotate_streams.stream_count),
2572                         .new_chunk_id = (typeof(_rotate_streams->new_chunk_id)) {
2573                                 .is_set = !!packed_rotate_streams.new_chunk_id.is_set,
2574                                 .value = be64toh(packed_rotate_streams.new_chunk_id.value),
2575                         }
2576                 };
2577         } else if (payload->size == expected_payload_size_3_bytes_padding) {
2578                 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams;
2579
2580                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2581
2582                 header_len = sizeof(padded_rotate_streams);
2583                 memcpy(&padded_rotate_streams, payload->data, header_len);
2584
2585                 /* Unpack the 3-byte padded structure to the natively-packed version. */
2586                 *_rotate_streams = (typeof(*_rotate_streams)) {
2587                         .stream_count = be32toh(padded_rotate_streams.stream_count),
2588                         .new_chunk_id = (typeof(_rotate_streams->new_chunk_id)) {
2589                                 .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2590                                 .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2591                         }
2592                 };
2593         } else if (payload->size == expected_payload_size_7_bytes_padding) {
2594                 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams;
2595
2596                 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2597
2598                 header_len = sizeof(padded_rotate_streams);
2599                 memcpy(&padded_rotate_streams, payload->data, header_len);
2600
2601                 /* Unpack the 7-byte padded structure to the natively-packed version. */
2602                 *_rotate_streams = (typeof(*_rotate_streams)) {
2603                         .stream_count = be32toh(padded_rotate_streams.stream_count),
2604                         .new_chunk_id = (typeof(_rotate_streams->new_chunk_id)) {
2605                                 .is_set = !!padded_rotate_streams.new_chunk_id.is_set,
2606                                 .value = be64toh(padded_rotate_streams.new_chunk_id.value),
2607                         }
2608                 };
2609
2610                 header_len = sizeof(padded_rotate_streams);
2611         } else {
2612                 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2613                                 expected_payload_size_no_padding,
2614                                 expected_payload_size_3_bytes_padding,
2615                                 expected_payload_size_7_bytes_padding,
2616                                 payload->size);
2617                 goto error;
2618         }
2619
2620         return header_len;
2621 error:
2622         return -1;
2623 }
2624
2625 /*
2626  * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2627  * session rotation feature (not the tracefile rotation feature).
2628  */
2629 static int relay_rotate_session_streams(
2630                 const struct lttcomm_relayd_hdr *recv_hdr,
2631                 struct relay_connection *conn,
2632                 const struct lttng_buffer_view *payload)
2633 {
2634         int ret = 0;
2635         uint32_t i;
2636         ssize_t send_ret;
2637         enum lttng_error_code reply_code = LTTNG_ERR_UNK;
2638         struct relay_session *session = conn->session;
2639         struct lttcomm_relayd_rotate_streams rotate_streams;
2640         struct lttcomm_relayd_generic_reply reply = {};
2641         struct relay_stream *stream = NULL;
2642         struct lttng_trace_chunk *next_trace_chunk = NULL;
2643         struct lttng_buffer_view stream_positions;
2644         char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
2645         const char *chunk_id_str = "none";
2646         ssize_t header_len;
2647
2648         if (!session || !conn->version_check_done) {
2649                 ERR("Trying to rotate a stream before version check");
2650                 ret = -1;
2651                 goto end_no_reply;
2652         }
2653
2654         if (session->major == 2 && session->minor < 11) {
2655                 ERR("Unsupported feature before 2.11");
2656                 ret = -1;
2657                 goto end_no_reply;
2658         }
2659
2660         header_len = relay_unpack_rotate_streams_header(payload, &rotate_streams);
2661         if (header_len < 0) {
2662                 ret = -1;
2663                 goto end_no_reply;
2664         }
2665
2666         if (rotate_streams.new_chunk_id.is_set) {
2667                 /*
2668                  * Retrieve the trace chunk the stream must transition to. As
2669                  * per the protocol, this chunk should have been created
2670                  * before this command is received.
2671                  */
2672                 next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
2673                                 sessiond_trace_chunk_registry,
2674                                 session->sessiond_uuid, session->id,
2675                                 rotate_streams.new_chunk_id.value);
2676                 if (!next_trace_chunk) {
2677                         char uuid_str[LTTNG_UUID_STR_LEN];
2678
2679                         lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
2680                         ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2681                                         ", trace_chunk_id = %" PRIu64,
2682                                         uuid_str, session->id,
2683                                         rotate_streams.new_chunk_id.value);
2684                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2685                         ret = -1;
2686                         goto end;
2687                 }
2688
2689                 ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64,
2690                                 rotate_streams.new_chunk_id.value);
2691                 if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
2692                         chunk_id_str = "formatting error";
2693                 } else {
2694                         chunk_id_str = chunk_id_buf;
2695                 }
2696         }
2697
2698         DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
2699                         rotate_streams.stream_count, session->session_name,
2700                         chunk_id_str);
2701
2702         stream_positions = lttng_buffer_view_from_view(payload,
2703                         header_len, -1);
2704         if (!stream_positions.data ||
2705                         stream_positions.size <
2706                                         (rotate_streams.stream_count *
2707                                                         sizeof(struct lttcomm_relayd_stream_rotation_position))) {
2708                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2709                 ret = -1;
2710                 goto end;
2711         }
2712
2713         for (i = 0; i < rotate_streams.stream_count; i++) {
2714                 struct lttcomm_relayd_stream_rotation_position *position_comm =
2715                                 &((typeof(position_comm)) stream_positions.data)[i];
2716                 const struct lttcomm_relayd_stream_rotation_position pos = {
2717                         .stream_id = be64toh(position_comm->stream_id),
2718                         .rotate_at_seq_num = be64toh(
2719                                         position_comm->rotate_at_seq_num),
2720                 };
2721
2722                 stream = stream_get_by_id(pos.stream_id);
2723                 if (!stream) {
2724                         reply_code = LTTNG_ERR_INVALID;
2725                         ret = -1;
2726                         goto end;
2727                 }
2728
2729                 pthread_mutex_lock(&stream->lock);
2730                 ret = stream_set_pending_rotation(stream, next_trace_chunk,
2731                                 pos.rotate_at_seq_num);
2732                 pthread_mutex_unlock(&stream->lock);
2733                 if (ret) {
2734                         reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
2735                         goto end;
2736                 }
2737
2738                 stream_put(stream);
2739                 stream = NULL;
2740         }
2741
2742         reply_code = LTTNG_OK;
2743         ret = 0;
2744 end:
2745         if (stream) {
2746                 stream_put(stream);
2747         }
2748
2749         reply.ret_code = htobe32((uint32_t) reply_code);
2750         send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2751                         sizeof(struct lttcomm_relayd_generic_reply), 0);
2752         if (send_ret < (ssize_t) sizeof(reply)) {
2753                 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2754                                 send_ret);
2755                 ret = -1;
2756         }
2757 end_no_reply:
2758         lttng_trace_chunk_put(next_trace_chunk);
2759         return ret;
2760 }
2761
2762 /*
2763  * relay_create_trace_chunk: create a new trace chunk
2764  */
2765 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2766                 struct relay_connection *conn,
2767                 const struct lttng_buffer_view *payload)
2768 {
2769         int ret = 0;
2770         ssize_t send_ret;
2771         struct relay_session *session = conn->session;
2772         struct lttcomm_relayd_create_trace_chunk *msg;
2773         struct lttcomm_relayd_generic_reply reply = {};
2774         struct lttng_buffer_view header_view;
2775         struct lttng_buffer_view chunk_name_view;
2776         struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
2777         enum lttng_error_code reply_code = LTTNG_OK;
2778         enum lttng_trace_chunk_status chunk_status;
2779         const char *new_path;
2780
2781         if (!session || !conn->version_check_done) {
2782                 ERR("Trying to create a trace chunk before version check");
2783                 ret = -1;
2784                 goto end_no_reply;
2785         }
2786
2787         if (session->major == 2 && session->minor < 11) {
2788                 ERR("Chunk creation command is unsupported before 2.11");
2789                 ret = -1;
2790                 goto end_no_reply;
2791         }
2792
2793         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2794         if (!header_view.data) {
2795                 ERR("Failed to receive payload of chunk creation command");
2796                 ret = -1;
2797                 goto end_no_reply;
2798         }
2799
2800         /* Convert to host endianness. */
2801         msg = (typeof(msg)) header_view.data;
2802         msg->chunk_id = be64toh(msg->chunk_id);
2803         msg->creation_timestamp = be64toh(msg->creation_timestamp);
2804         msg->override_name_length = be32toh(msg->override_name_length);
2805
2806         if (session->current_trace_chunk &&
2807                         !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
2808                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
2809                                         DEFAULT_CHUNK_TMP_OLD_DIRECTORY);
2810                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2811                         ERR("Failed to rename old chunk");
2812                         ret = -1;
2813                         reply_code = LTTNG_ERR_UNK;
2814                         goto end;
2815                 }
2816         }
2817         session->ongoing_rotation = true;
2818         if (!session->current_trace_chunk) {
2819                 if (!session->has_rotated) {
2820                         new_path = "";
2821                 } else {
2822                         new_path = NULL;
2823                 }
2824         } else {
2825                 new_path = DEFAULT_CHUNK_TMP_NEW_DIRECTORY;
2826         }
2827         chunk = lttng_trace_chunk_create(
2828                         msg->chunk_id, msg->creation_timestamp, new_path);
2829         if (!chunk) {
2830                 ERR("Failed to create trace chunk in trace chunk creation command");
2831                 ret = -1;
2832                 reply_code = LTTNG_ERR_NOMEM;
2833                 goto end;
2834         }
2835         lttng_trace_chunk_set_fd_tracker(chunk, the_fd_tracker);
2836
2837         if (msg->override_name_length) {
2838                 const char *name;
2839
2840                 chunk_name_view = lttng_buffer_view_from_view(payload,
2841                                 sizeof(*msg),
2842                                 msg->override_name_length);
2843                 name = chunk_name_view.data;
2844                 if (!name || name[msg->override_name_length - 1]) {
2845                         ERR("Failed to receive payload of chunk creation command");
2846                         ret = -1;
2847                         reply_code = LTTNG_ERR_INVALID;
2848                         goto end;
2849                 }
2850
2851                 chunk_status = lttng_trace_chunk_override_name(
2852                                 chunk, chunk_name_view.data);
2853                 switch (chunk_status) {
2854                 case LTTNG_TRACE_CHUNK_STATUS_OK:
2855                         break;
2856                 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
2857                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2858                         reply_code = LTTNG_ERR_INVALID;
2859                         ret = -1;
2860                         goto end;
2861                 default:
2862                         ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2863                         reply_code = LTTNG_ERR_UNK;
2864                         ret = -1;
2865                         goto end;
2866                 }
2867         }
2868
2869         chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
2870         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2871                 reply_code = LTTNG_ERR_UNK;
2872                 ret = -1;
2873                 goto end;
2874         }
2875
2876         assert(conn->session->output_directory);
2877         chunk_status = lttng_trace_chunk_set_as_owner(chunk,
2878                         conn->session->output_directory);
2879         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2880                 reply_code = LTTNG_ERR_UNK;
2881                 ret = -1;
2882                 goto end;
2883         }
2884
2885         published_chunk = sessiond_trace_chunk_registry_publish_chunk(
2886                         sessiond_trace_chunk_registry,
2887                         conn->session->sessiond_uuid,
2888                         conn->session->id,
2889                         chunk);
2890         if (!published_chunk) {
2891                 char uuid_str[LTTNG_UUID_STR_LEN];
2892
2893                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
2894                 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
2895                                 uuid_str,
2896                                 conn->session->id,
2897                                 msg->chunk_id);
2898                 ret = -1;
2899                 reply_code = LTTNG_ERR_NOMEM;
2900                 goto end;
2901         }
2902
2903         pthread_mutex_lock(&conn->session->lock);
2904         if (conn->session->pending_closure_trace_chunk) {
2905                 /*
2906                  * Invalid; this means a second create_trace_chunk command was
2907                  * received before a close_trace_chunk.
2908                  */
2909                 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2910                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
2911                 ret = -1;
2912                 goto end_unlock_session;
2913         }
2914         conn->session->pending_closure_trace_chunk =
2915                         conn->session->current_trace_chunk;
2916         conn->session->current_trace_chunk = published_chunk;
2917         published_chunk = NULL;
2918         if (!conn->session->pending_closure_trace_chunk) {
2919                 session->ongoing_rotation = false;
2920         }
2921 end_unlock_session:
2922         pthread_mutex_unlock(&conn->session->lock);
2923 end:
2924         reply.ret_code = htobe32((uint32_t) reply_code);
2925         send_ret = conn->sock->ops->sendmsg(conn->sock,
2926                         &reply,
2927                         sizeof(struct lttcomm_relayd_generic_reply),
2928                         0);
2929         if (send_ret < (ssize_t) sizeof(reply)) {
2930                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2931                                 send_ret);
2932                 ret = -1;
2933         }
2934 end_no_reply:
2935         lttng_trace_chunk_put(chunk);
2936         lttng_trace_chunk_put(published_chunk);
2937         return ret;
2938 }
2939
2940 /*
2941  * relay_close_trace_chunk: close a trace chunk
2942  */
2943 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
2944                 struct relay_connection *conn,
2945                 const struct lttng_buffer_view *payload)
2946 {
2947         int ret = 0, buf_ret;
2948         ssize_t send_ret;
2949         struct relay_session *session = conn->session;
2950         struct lttcomm_relayd_close_trace_chunk *msg;
2951         struct lttcomm_relayd_close_trace_chunk_reply reply = {};
2952         struct lttng_buffer_view header_view;
2953         struct lttng_trace_chunk *chunk = NULL;
2954         enum lttng_error_code reply_code = LTTNG_OK;
2955         enum lttng_trace_chunk_status chunk_status;
2956         uint64_t chunk_id;
2957         LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
2958         time_t close_timestamp;
2959         char closed_trace_chunk_path[LTTNG_PATH_MAX];
2960         size_t path_length = 0;
2961         const char *chunk_name = NULL;
2962         struct lttng_dynamic_buffer reply_payload;
2963         const char *new_path;
2964
2965         lttng_dynamic_buffer_init(&reply_payload);
2966
2967         if (!session || !conn->version_check_done) {
2968                 ERR("Trying to close a trace chunk before version check");
2969                 ret = -1;
2970                 goto end_no_reply;
2971         }
2972
2973         if (session->major == 2 && session->minor < 11) {
2974                 ERR("Chunk close command is unsupported before 2.11");
2975                 ret = -1;
2976                 goto end_no_reply;
2977         }
2978
2979         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
2980         if (!header_view.data) {
2981                 ERR("Failed to receive payload of chunk close command");
2982                 ret = -1;
2983                 goto end_no_reply;
2984         }
2985
2986         /* Convert to host endianness. */
2987         msg = (typeof(msg)) header_view.data;
2988         chunk_id = be64toh(msg->chunk_id);
2989         close_timestamp = (time_t) be64toh(msg->close_timestamp);
2990         close_command = (typeof(close_command)){
2991                 .value = be32toh(msg->close_command.value),
2992                 .is_set = msg->close_command.is_set,
2993         };
2994
2995         chunk = sessiond_trace_chunk_registry_get_chunk(
2996                         sessiond_trace_chunk_registry,
2997                         conn->session->sessiond_uuid,
2998                         conn->session->id,
2999                         chunk_id);
3000         if (!chunk) {
3001                 char uuid_str[LTTNG_UUID_STR_LEN];
3002
3003                 lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
3004                 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
3005                                 uuid_str,
3006                                 conn->session->id,
3007                                 msg->chunk_id);
3008                 ret = -1;
3009                 reply_code = LTTNG_ERR_NOMEM;
3010                 goto end;
3011         }
3012
3013         pthread_mutex_lock(&session->lock);
3014         if (close_command.is_set &&
3015                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
3016                 /*
3017                  * Clear command. It is a protocol error to ask for a
3018                  * clear on a relay which does not allow it. Querying
3019                  * the configuration allows figuring out whether
3020                  * clearing is allowed before doing the clear.
3021                  */
3022                 if (!opt_allow_clear) {
3023                         ret = -1;
3024                         reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3025                         goto end_unlock_session;
3026                 }
3027         }
3028         if (session->pending_closure_trace_chunk &&
3029                         session->pending_closure_trace_chunk != chunk) {
3030                 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3031                                 session->session_name);
3032                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3033                 ret = -1;
3034                 goto end_unlock_session;
3035         }
3036
3037         if (session->current_trace_chunk && session->current_trace_chunk != chunk &&
3038                         !lttng_trace_chunk_get_name_overridden(session->current_trace_chunk)) {
3039                 if (close_command.is_set &&
3040                                 close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE &&
3041                                 !session->has_rotated) {
3042                         /* New chunk stays in session output directory. */
3043                         new_path = "";
3044                 } else {
3045                         /* Use chunk name for new chunk. */
3046                         new_path = NULL;
3047                 }
3048                 /* Rename new chunk path. */
3049                 chunk_status = lttng_trace_chunk_rename_path(session->current_trace_chunk,
3050                                 new_path);
3051                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3052                         ret = -1;
3053                         goto end_unlock_session;
3054                 }
3055                 session->ongoing_rotation = false;
3056         }
3057         if ((!close_command.is_set ||
3058                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION) &&
3059                         !lttng_trace_chunk_get_name_overridden(chunk)) {
3060                 const char *old_path;
3061
3062                 if (!session->has_rotated) {
3063                         old_path = "";
3064                 } else {
3065                         old_path = NULL;
3066                 }
3067                 /* We need to move back the .tmp_old_chunk to its rightful place. */
3068                 chunk_status = lttng_trace_chunk_rename_path(chunk, old_path);
3069                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3070                         ret = -1;
3071                         goto end_unlock_session;
3072                 }
3073         }
3074         chunk_status = lttng_trace_chunk_set_close_timestamp(
3075                         chunk, close_timestamp);
3076         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3077                 ERR("Failed to set trace chunk close timestamp");
3078                 ret = -1;
3079                 reply_code = LTTNG_ERR_UNK;
3080                 goto end_unlock_session;
3081         }
3082
3083         if (close_command.is_set) {
3084                 chunk_status = lttng_trace_chunk_set_close_command(
3085                                 chunk, close_command.value);
3086                 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3087                         ret = -1;
3088                         reply_code = LTTNG_ERR_INVALID;
3089                         goto end_unlock_session;
3090                 }
3091         }
3092         chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL);
3093         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
3094                 ERR("Failed to get chunk name");
3095                 ret = -1;
3096                 reply_code = LTTNG_ERR_UNK;
3097                 goto end_unlock_session;
3098         }
3099         if (!session->has_rotated && !session->snapshot) {
3100                 ret = lttng_strncpy(closed_trace_chunk_path,
3101                                 session->output_path,
3102                                 sizeof(closed_trace_chunk_path));
3103                 if (ret) {
3104                         ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3105                                         strlen(session->output_path),
3106                                         sizeof(closed_trace_chunk_path));
3107                         reply_code = LTTNG_ERR_NOMEM;
3108                         ret = -1;
3109                         goto end_unlock_session;
3110                 }
3111         } else {
3112                 if (session->snapshot) {
3113                         ret = snprintf(closed_trace_chunk_path,
3114                                         sizeof(closed_trace_chunk_path),
3115                                         "%s/%s", session->output_path,
3116                                         chunk_name);
3117                 } else {
3118                         ret = snprintf(closed_trace_chunk_path,
3119                                         sizeof(closed_trace_chunk_path),
3120                                         "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3121                                         "/%s",
3122                                         session->output_path, chunk_name);
3123                 }
3124                 if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
3125                         ERR("Failed to format closed trace chunk resulting path");
3126                         reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
3127                         ret = -1;
3128                         goto end_unlock_session;
3129                 }
3130         }
3131         if (close_command.is_set &&
3132                         close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
3133                 session->has_rotated = true;
3134         }
3135         DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
3136         path_length = strlen(closed_trace_chunk_path) + 1;
3137         if (path_length > UINT32_MAX) {
3138                 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3139                 ret = -1;
3140                 reply_code = LTTNG_ERR_INVALID_PROTOCOL;
3141                 goto end_unlock_session;
3142         }
3143
3144         if (session->current_trace_chunk == chunk) {
3145                 /*
3146                  * After a trace chunk close command, no new streams
3147                  * referencing the chunk may be created. Hence, on the
3148                  * event that no new trace chunk have been created for
3149                  * the session, the reference to the current trace chunk
3150                  * is released in order to allow it to be reclaimed when
3151                  * the last stream releases its reference to it.
3152                  */
3153                 lttng_trace_chunk_put(session->current_trace_chunk);
3154                 session->current_trace_chunk = NULL;
3155         }
3156         lttng_trace_chunk_put(session->pending_closure_trace_chunk);
3157         session->pending_closure_trace_chunk = NULL;
3158 end_unlock_session:
3159         pthread_mutex_unlock(&session->lock);
3160
3161 end:
3162         reply.generic.ret_code = htobe32((uint32_t) reply_code);
3163         reply.path_length = htobe32((uint32_t) path_length);
3164         buf_ret = lttng_dynamic_buffer_append(
3165                         &reply_payload, &reply, sizeof(reply));
3166         if (buf_ret) {
3167                 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3168                 goto end_no_reply;
3169         }
3170
3171         if (reply_code == LTTNG_OK) {
3172                 buf_ret = lttng_dynamic_buffer_append(&reply_payload,
3173                                 closed_trace_chunk_path, path_length);
3174                 if (buf_ret) {
3175                         ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3176                         goto end_no_reply;
3177                 }
3178         }
3179
3180         send_ret = conn->sock->ops->sendmsg(conn->sock,
3181                         reply_payload.data,
3182                         reply_payload.size,
3183                         0);
3184         if (send_ret < reply_payload.size) {
3185                 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3186                                 reply_payload.size, send_ret);
3187                 ret = -1;
3188                 goto end_no_reply;
3189         }
3190 end_no_reply:
3191         lttng_trace_chunk_put(chunk);
3192         lttng_dynamic_buffer_reset(&reply_payload);
3193         return ret;
3194 }
3195
3196 /*
3197  * relay_trace_chunk_exists: check if a trace chunk exists
3198  */
3199 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr,
3200                 struct relay_connection *conn,
3201                 const struct lttng_buffer_view *payload)
3202 {
3203         int ret = 0;
3204         ssize_t send_ret;
3205         struct relay_session *session = conn->session;
3206         struct lttcomm_relayd_trace_chunk_exists *msg;
3207         struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
3208         struct lttng_buffer_view header_view;
3209         uint64_t chunk_id;
3210         bool chunk_exists;
3211
3212         if (!session || !conn->version_check_done) {
3213                 ERR("Trying to close a trace chunk before version check");
3214                 ret = -1;
3215                 goto end_no_reply;
3216         }
3217
3218         if (session->major == 2 && session->minor < 11) {
3219                 ERR("Chunk close command is unsupported before 2.11");
3220                 ret = -1;
3221                 goto end_no_reply;
3222         }
3223
3224         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3225         if (!header_view.data) {
3226                 ERR("Failed to receive payload of chunk close command");
3227                 ret = -1;
3228                 goto end_no_reply;
3229         }
3230
3231         /* Convert to host endianness. */
3232         msg = (typeof(msg)) header_view.data;
3233         chunk_id = be64toh(msg->chunk_id);
3234
3235         ret = sessiond_trace_chunk_registry_chunk_exists(
3236                         sessiond_trace_chunk_registry,
3237                         conn->session->sessiond_uuid,
3238                         conn->session->id,
3239                         chunk_id, &chunk_exists);
3240         /*
3241          * If ret is not 0, send the reply and report the error to the caller.
3242          * It is a protocol (or internal) error and the session/connection
3243          * should be torn down.
3244          */
3245         reply = (typeof(reply)){
3246                 .generic.ret_code = htobe32((uint32_t)
3247                         (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3248                 .trace_chunk_exists = ret == 0 ? chunk_exists : 0,
3249         };
3250         send_ret = conn->sock->ops->sendmsg(
3251                         conn->sock, &reply, sizeof(reply), 0);
3252         if (send_ret < (ssize_t) sizeof(reply)) {
3253                 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3254                                 send_ret);
3255                 ret = -1;
3256         }
3257 end_no_reply:
3258         return ret;
3259 }
3260
3261 /*
3262  * relay_get_configuration: query whether feature is available
3263  */
3264 static int relay_get_configuration(const struct lttcomm_relayd_hdr *recv_hdr,
3265                 struct relay_connection *conn,
3266                 const struct lttng_buffer_view *payload)
3267 {
3268         int ret = 0;
3269         ssize_t send_ret;
3270         struct lttcomm_relayd_get_configuration *msg;
3271         struct lttcomm_relayd_get_configuration_reply reply = {};
3272         struct lttng_buffer_view header_view;
3273         uint64_t query_flags = 0;
3274         uint64_t result_flags = 0;
3275
3276         header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
3277         if (!header_view.data) {
3278                 ERR("Failed to receive payload of chunk close command");
3279                 ret = -1;
3280                 goto end_no_reply;
3281         }
3282
3283         /* Convert to host endianness. */
3284         msg = (typeof(msg)) header_view.data;
3285         query_flags = be64toh(msg->query_flags);
3286
3287         if (query_flags) {
3288                 ret = LTTNG_ERR_INVALID_PROTOCOL;
3289                 goto reply;
3290         }
3291         if (opt_allow_clear) {
3292                 result_flags |= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED;
3293         }
3294         ret = 0;
3295 reply:
3296         reply = (typeof(reply)){
3297                 .generic.ret_code = htobe32((uint32_t)
3298                         (ret == 0 ? LTTNG_OK : LTTNG_ERR_INVALID_PROTOCOL)),
3299                 .relayd_configuration_flags = htobe64(result_flags),
3300         };
3301         send_ret = conn->sock->ops->sendmsg(
3302                         conn->sock, &reply, sizeof(reply), 0);
3303         if (send_ret < (ssize_t) sizeof(reply)) {
3304                 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3305                                 send_ret);
3306                 ret = -1;
3307         }
3308 end_no_reply:
3309         return ret;
3310 }
3311
3312 #define DBG_CMD(cmd_name, conn) \
3313                 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3314
3315 static int relay_process_control_command(struct relay_connection *conn,
3316                 const struct lttcomm_relayd_hdr *header,
3317                 const struct lttng_buffer_view *payload)
3318 {
3319         int ret = 0;
3320
3321         switch (header->cmd) {
3322         case RELAYD_CREATE_SESSION:
3323                 DBG_CMD("RELAYD_CREATE_SESSION", conn);
3324                 ret = relay_create_session(header, conn, payload);
3325                 break;
3326         case RELAYD_ADD_STREAM:
3327                 DBG_CMD("RELAYD_ADD_STREAM", conn);
3328                 ret = relay_add_stream(header, conn, payload);
3329                 break;
3330         case RELAYD_START_DATA:
3331                 DBG_CMD("RELAYD_START_DATA", conn);
3332                 ret = relay_start(header, conn, payload);
3333                 break;
3334         case RELAYD_SEND_METADATA:
3335                 DBG_CMD("RELAYD_SEND_METADATA", conn);
3336                 ret = relay_recv_metadata(header, conn, payload);
3337                 break;
3338         case RELAYD_VERSION:
3339                 DBG_CMD("RELAYD_VERSION", conn);
3340                 ret = relay_send_version(header, conn, payload);
3341                 break;
3342         case RELAYD_CLOSE_STREAM:
3343                 DBG_CMD("RELAYD_CLOSE_STREAM", conn);
3344                 ret = relay_close_stream(header, conn, payload);
3345                 break;
3346         case RELAYD_DATA_PENDING:
3347                 DBG_CMD("RELAYD_DATA_PENDING", conn);
3348                 ret = relay_data_pending(header, conn, payload);
3349                 break;
3350         case RELAYD_QUIESCENT_CONTROL:
3351                 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
3352                 ret = relay_quiescent_control(header, conn, payload);
3353                 break;
3354         case RELAYD_BEGIN_DATA_PENDING:
3355                 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
3356                 ret = relay_begin_data_pending(header, conn, payload);
3357                 break;
3358         case RELAYD_END_DATA_PENDING:
3359                 DBG_CMD("RELAYD_END_DATA_PENDING", conn);
3360                 ret = relay_end_data_pending(header, conn, payload);
3361                 break;
3362         case RELAYD_SEND_INDEX:
3363                 DBG_CMD("RELAYD_SEND_INDEX", conn);
3364                 ret = relay_recv_index(header, conn, payload);
3365                 break;
3366         case RELAYD_STREAMS_SENT:
3367                 DBG_CMD("RELAYD_STREAMS_SENT", conn);
3368                 ret = relay_streams_sent(header, conn, payload);
3369                 break;
3370         case RELAYD_RESET_METADATA:
3371                 DBG_CMD("RELAYD_RESET_METADATA", conn);
3372                 ret = relay_reset_metadata(header, conn, payload);
3373                 break;
3374         case RELAYD_ROTATE_STREAMS:
3375                 DBG_CMD("RELAYD_ROTATE_STREAMS", conn);
3376                 ret = relay_rotate_session_streams(header, conn, payload);
3377                 break;
3378         case RELAYD_CREATE_TRACE_CHUNK:
3379                 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn);
3380                 ret = relay_create_trace_chunk(header, conn, payload);
3381                 break;
3382         case RELAYD_CLOSE_TRACE_CHUNK:
3383                 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn);
3384                 ret = relay_close_trace_chunk(header, conn, payload);
3385                 break;
3386         case RELAYD_TRACE_CHUNK_EXISTS:
3387                 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn);
3388                 ret = relay_trace_chunk_exists(header, conn, payload);
3389                 break;
3390         case RELAYD_GET_CONFIGURATION:
3391                 DBG_CMD("RELAYD_GET_CONFIGURATION", conn);
3392                 ret = relay_get_configuration(header, conn, payload);
3393                 break;
3394         case RELAYD_UPDATE_SYNC_INFO:
3395         default:
3396                 ERR("Received unknown command (%u)", header->cmd);
3397                 relay_unknown_command(conn);
3398                 ret = -1;
3399                 goto end;
3400         }
3401
3402 end:
3403         return ret;
3404 }
3405
3406 static enum relay_connection_status relay_process_control_receive_payload(
3407                 struct relay_connection *conn)
3408 {
3409         int ret = 0;
3410         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3411         struct lttng_dynamic_buffer *reception_buffer =
3412                         &conn->protocol.ctrl.reception_buffer;
3413         struct ctrl_connection_state_receive_payload *state =
3414                         &conn->protocol.ctrl.state.receive_payload;
3415         struct lttng_buffer_view payload_view;
3416
3417         if (state->left_to_receive == 0) {
3418                 /* Short-circuit for payload-less commands. */
3419                 goto reception_complete;
3420         }
3421
3422         ret = conn->sock->ops->recvmsg(conn->sock,
3423                         reception_buffer->data + state->received,
3424                         state->left_to_receive, MSG_DONTWAIT);
3425         if (ret < 0) {
3426                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3427                         PERROR("Unable to receive command payload on sock %d",
3428                                         conn->sock->fd);
3429                         status = RELAY_CONNECTION_STATUS_ERROR;
3430                 }
3431                 goto end;
3432         } else if (ret == 0) {
3433                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3434                 status = RELAY_CONNECTION_STATUS_CLOSED;
3435                 goto end;
3436         }
3437
3438         assert(ret > 0);
3439         assert(ret <= state->left_to_receive);
3440
3441         state->left_to_receive -= ret;
3442         state->received += ret;
3443
3444         if (state->left_to_receive > 0) {
3445                 /*
3446                  * Can't transition to the protocol's next state, wait to
3447                  * receive the rest of the header.
3448                  */
3449                 DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3450                                 state->received, state->left_to_receive,
3451                                 conn->sock->fd);
3452                 goto end;
3453         }
3454
3455 reception_complete:
3456         DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3457                         conn->sock->fd, state->received);
3458         /*
3459          * The payload required to process the command has been received.
3460          * A view to the reception buffer is forwarded to the various
3461          * commands and the state of the control is reset on success.
3462          *
3463          * Commands are responsible for sending their reply to the peer.
3464          */
3465         payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
3466                         0, -1);
3467         ret = relay_process_control_command(conn,
3468                         &state->header, &payload_view);
3469         if (ret < 0) {
3470                 status = RELAY_CONNECTION_STATUS_ERROR;
3471                 goto end;
3472         }
3473
3474         ret = connection_reset_protocol_state(conn);
3475         if (ret) {
3476                 status = RELAY_CONNECTION_STATUS_ERROR;
3477         }
3478 end:
3479         return status;
3480 }
3481
3482 static enum relay_connection_status relay_process_control_receive_header(
3483                 struct relay_connection *conn)
3484 {
3485         int ret = 0;
3486         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3487         struct lttcomm_relayd_hdr header;
3488         struct lttng_dynamic_buffer *reception_buffer =
3489                         &conn->protocol.ctrl.reception_buffer;
3490         struct ctrl_connection_state_receive_header *state =
3491                         &conn->protocol.ctrl.state.receive_header;
3492
3493         assert(state->left_to_receive != 0);
3494
3495         ret = conn->sock->ops->recvmsg(conn->sock,
3496                         reception_buffer->data + state->received,
3497                         state->left_to_receive, MSG_DONTWAIT);
3498         if (ret < 0) {
3499                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3500                         PERROR("Unable to receive control command header on sock %d",
3501                                         conn->sock->fd);
3502                         status = RELAY_CONNECTION_STATUS_ERROR;
3503                 }
3504                 goto end;
3505         } else if (ret == 0) {
3506                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3507                 status = RELAY_CONNECTION_STATUS_CLOSED;
3508                 goto end;
3509         }
3510
3511         assert(ret > 0);
3512         assert(ret <= state->left_to_receive);
3513
3514         state->left_to_receive -= ret;
3515         state->received += ret;
3516
3517         if (state->left_to_receive > 0) {
3518                 /*
3519                  * Can't transition to the protocol's next state, wait to
3520                  * receive the rest of the header.
3521                  */
3522                 DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3523                                 state->received, state->left_to_receive,
3524                                 conn->sock->fd);
3525                 goto end;
3526         }
3527
3528         /* Transition to next state: receiving the command's payload. */
3529         conn->protocol.ctrl.state_id =
3530                         CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3531         memcpy(&header, reception_buffer->data, sizeof(header));
3532         header.circuit_id = be64toh(header.circuit_id);
3533         header.data_size = be64toh(header.data_size);
3534         header.cmd = be32toh(header.cmd);
3535         header.cmd_version = be32toh(header.cmd_version);
3536         memcpy(&conn->protocol.ctrl.state.receive_payload.header,
3537                         &header, sizeof(header));
3538
3539         DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
3540                         conn->sock->fd, header.cmd, header.cmd_version,
3541                         header.data_size);
3542
3543         if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
3544                 ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3545                                 header.data_size);
3546                 status = RELAY_CONNECTION_STATUS_ERROR;
3547                 goto end;
3548         }
3549
3550         conn->protocol.ctrl.state.receive_payload.left_to_receive =
3551                         header.data_size;
3552         conn->protocol.ctrl.state.receive_payload.received = 0;
3553         ret = lttng_dynamic_buffer_set_size(reception_buffer,
3554                         header.data_size);
3555         if (ret) {
3556                 status = RELAY_CONNECTION_STATUS_ERROR;
3557                 goto end;
3558         }
3559
3560         if (header.data_size == 0) {
3561                 /*
3562                  * Manually invoke the next state as the poll loop
3563                  * will not wake-up to allow us to proceed further.
3564                  */
3565                 status = relay_process_control_receive_payload(conn);
3566         }
3567 end:
3568         return status;
3569 }
3570
3571 /*
3572  * Process the commands received on the control socket
3573  */
3574 static enum relay_connection_status relay_process_control(
3575                 struct relay_connection *conn)
3576 {
3577         enum relay_connection_status status;
3578
3579         switch (conn->protocol.ctrl.state_id) {
3580         case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
3581                 status = relay_process_control_receive_header(conn);
3582                 break;
3583         case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
3584                 status = relay_process_control_receive_payload(conn);
3585                 break;
3586         default:
3587                 ERR("Unknown control connection protocol state encountered.");
3588                 abort();
3589         }
3590
3591         return status;
3592 }
3593
3594 static enum relay_connection_status relay_process_data_receive_header(
3595                 struct relay_connection *conn)
3596 {
3597         int ret;
3598         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3599         struct data_connection_state_receive_header *state =
3600                         &conn->protocol.data.state.receive_header;
3601         struct lttcomm_relayd_data_hdr header;
3602         struct relay_stream *stream;
3603
3604         assert(state->left_to_receive != 0);
3605
3606         ret = conn->sock->ops->recvmsg(conn->sock,
3607                         state->header_reception_buffer + state->received,
3608                         state->left_to_receive, MSG_DONTWAIT);
3609         if (ret < 0) {
3610                 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3611                         PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3612                         status = RELAY_CONNECTION_STATUS_ERROR;
3613                 }
3614                 goto end;
3615         } else if (ret == 0) {
3616                 /* Orderly shutdown. Not necessary to print an error. */
3617                 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
3618                 status = RELAY_CONNECTION_STATUS_CLOSED;
3619                 goto end;
3620         }
3621
3622         assert(ret > 0);
3623         assert(ret <= state->left_to_receive);
3624
3625         state->left_to_receive -= ret;
3626         state->received += ret;
3627
3628         if (state->left_to_receive > 0) {
3629                 /*
3630                  * Can't transition to the protocol's next state, wait to
3631                  * receive the rest of the header.
3632                  */
3633                 DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3634                                 state->received, state->left_to_receive,
3635                                 conn->sock->fd);
3636                 goto end;
3637         }
3638
3639         /* Transition to next state: receiving the payload. */
3640         conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
3641
3642         memcpy(&header, state->header_reception_buffer, sizeof(header));
3643         header.circuit_id = be64toh(header.circuit_id);
3644         header.stream_id = be64toh(header.stream_id);
3645         header.data_size = be32toh(header.data_size);
3646         header.net_seq_num = be64toh(header.net_seq_num);
3647         header.padding_size = be32toh(header.padding_size);
3648         memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3649
3650         conn->protocol.data.state.receive_payload.left_to_receive =
3651                         header.data_size;
3652         conn->protocol.data.state.receive_payload.received = 0;
3653         conn->protocol.data.state.receive_payload.rotate_index = false;
3654
3655         DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32,
3656                         conn->sock->fd, header.circuit_id,
3657                         header.stream_id, header.data_size,
3658                         header.net_seq_num, header.padding_size);
3659
3660         stream = stream_get_by_id(header.stream_id);
3661         if (!stream) {
3662                 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3663                                 header.stream_id);
3664                 /* Protocol error. */
3665                 status = RELAY_CONNECTION_STATUS_ERROR;
3666                 goto end;
3667         }
3668
3669         pthread_mutex_lock(&stream->lock);
3670         /* Prepare stream for the reception of a new packet. */
3671         ret = stream_init_packet(stream, header.data_size,
3672                         &conn->protocol.data.state.receive_payload.rotate_index);
3673         pthread_mutex_unlock(&stream->lock);
3674         if (ret) {
3675                 ERR("Failed to rotate stream output file");
3676                 status = RELAY_CONNECTION_STATUS_ERROR;
3677                 goto end_stream_unlock;
3678         }
3679
3680 end_stream_unlock:
3681         stream_put(stream);
3682 end:
3683         return status;
3684 }
3685
3686 static enum relay_connection_status relay_process_data_receive_payload(
3687                 struct relay_connection *conn)
3688 {
3689         int ret;
3690         enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
3691         struct relay_stream *stream;
3692         struct data_connection_state_receive_payload *state =
3693                         &conn->protocol.data.state.receive_payload;
3694         const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3695         char data_buffer[chunk_size];
3696         bool partial_recv = false;
3697         bool new_stream = false, close_requested = false, index_flushed = false;
3698         uint64_t left_to_receive = state->left_to_receive;
3699         struct relay_session *session;
3700
3701         DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
3702                         state->header.stream_id, state->header.net_seq_num,
3703                         state->received, left_to_receive);
3704
3705         stream = stream_get_by_id(state->header.stream_id);
3706         if (!stream) {
3707                 /* Protocol error. */
3708                 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
3709                                 state->header.stream_id);
3710                 status = RELAY_CONNECTION_STATUS_ERROR;
3711                 goto end;
3712         }
3713
3714         pthread_mutex_lock(&stream->lock);
3715         session = stream->trace->session;
3716         if (!conn->session) {
3717                 ret = connection_set_session(conn, session);
3718                 if (ret) {
3719                         status = RELAY_CONNECTION_STATUS_ERROR;
3720                         goto end_stream_unlock;
3721                 }
3722         }
3723
3724         /*
3725          * The size of the "chunk" received on any iteration is bounded by:
3726          *   - the data left to receive,
3727          *   - the data immediately available on the socket,
3728          *   - the on-stack data buffer
3729          */
3730         while (left_to_receive > 0 && !partial_recv) {
3731                 size_t recv_size = min(left_to_receive, chunk_size);
3732                 struct lttng_buffer_view packet_chunk;
3733
3734                 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer,
3735                                 recv_size, MSG_DONTWAIT);
3736                 if (ret < 0) {
3737                         if (errno != EAGAIN && errno != EWOULDBLOCK) {
3738                                 PERROR("Socket %d error", conn->sock->fd);
3739                                 status = RELAY_CONNECTION_STATUS_ERROR;
3740                         }
3741                         goto end_stream_unlock;
3742                 } else if (ret == 0) {
3743                         /* No more data ready to be consumed on socket. */
3744                         DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3745                                         state->header.stream_id);
3746                         status = RELAY_CONNECTION_STATUS_CLOSED;
3747                         break;
3748                 } else if (ret < (int) recv_size) {
3749                         /*
3750                          * All the data available on the socket has been
3751                          * consumed.
3752                          */
3753                         partial_recv = true;
3754                         recv_size = ret;
3755                 }
3756
3757                 packet_chunk = lttng_buffer_view_init(data_buffer,
3758                                 0, recv_size);
3759                 assert(packet_chunk.data);
3760
3761                 ret = stream_write(stream, &packet_chunk, 0);
3762                 if (ret) {
3763                         ERR("Relay error writing data to file");
3764                         status = RELAY_CONNECTION_STATUS_ERROR;
3765                         goto end_stream_unlock;
3766                 }
3767
3768                 left_to_receive -= recv_size;
3769                 state->received += recv_size;
3770                 state->left_to_receive = left_to_receive;
3771         }
3772
3773         if (state->left_to_receive > 0) {
3774                 /*
3775                  * Did not receive all the data expected, wait for more data to
3776                  * become available on the socket.
3777                  */
3778                 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
3779                                 state->header.stream_id, state->received,
3780                                 state->left_to_receive);
3781                 goto end_stream_unlock;
3782         }
3783
3784         ret = stream_write(stream, NULL, state->header.padding_size);
3785         if (ret) {
3786                 status = RELAY_CONNECTION_STATUS_ERROR;
3787                 goto end_stream_unlock;
3788         }
3789
3790         if (session_streams_have_index(session)) {
3791                 ret = stream_update_index(stream, state->header.net_seq_num,
3792                                 state->rotate_index, &index_flushed,
3793                                 state->header.data_size + state->header.padding_size);
3794                 if (ret < 0) {
3795                         ERR("Failed to update index: stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
3796                                         stream->stream_handle,
3797                                         state->header.net_seq_num, ret);
3798                         status = RELAY_CONNECTION_STATUS_ERROR;
3799                         goto end_stream_unlock;
3800                 }
3801         }
3802
3803         if (stream->prev_data_seq == -1ULL) {
3804                 new_stream = true;
3805         }
3806
3807         ret = stream_complete_packet(stream, state->header.data_size +
3808                         state->header.padding_size, state->header.net_seq_num,
3809                         index_flushed);
3810         if (ret) {
3811                 status = RELAY_CONNECTION_STATUS_ERROR;
3812                 goto end_stream_unlock;
3813         }
3814
3815         /*
3816          * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3817          * contents of *state which are aliased (union) to the same location as
3818          * the new state. Don't use it beyond this point.
3819          */
3820         connection_reset_protocol_state(conn);
3821         state = NULL;
3822
3823 end_stream_unlock:
3824         close_requested = stream->close_requested;
3825         pthread_mutex_unlock(&stream->lock);
3826         if (close_requested && left_to_receive == 0) {
3827                 try_stream_close(stream);
3828         }
3829
3830         if (new_stream) {
3831                 pthread_mutex_lock(&session->lock);
3832                 uatomic_set(&session->new_streams, 1);
3833                 pthread_mutex_unlock(&session->lock);
3834         }
3835
3836         stream_put(stream);
3837 end:
3838         return status;
3839 }
3840
3841 /*
3842  * relay_process_data: Process the data received on the data socket
3843  */
3844 static enum relay_connection_status relay_process_data(
3845                 struct relay_connection *conn)
3846 {
3847         enum relay_connection_status status;
3848
3849         switch (conn->protocol.data.state_id) {
3850         case DATA_CONNECTION_STATE_RECEIVE_HEADER:
3851                 status = relay_process_data_receive_header(conn);
3852                 break;
3853         case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
3854                 status = relay_process_data_receive_payload(conn);
3855                 break;
3856         default:
3857                 ERR("Unexpected data connection communication state.");
3858                 abort();
3859         }
3860
3861         return status;
3862 }
3863
3864 static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
3865 {
3866         int ret;
3867
3868         (void) lttng_poll_del(events, pollfd);
3869
3870         ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
3871                         fd_tracker_util_close_fd, NULL);
3872         if (ret < 0) {
3873                 ERR("Closing pollfd %d", pollfd);
3874         }
3875 }
3876
3877 static void relay_thread_close_connection(struct lttng_poll_event *events,
3878                 int pollfd, struct relay_connection *conn)
3879 {
3880         const char *type_str;
3881
3882         switch (conn->type) {
3883         case RELAY_DATA:
3884                 type_str = "Data";
3885                 break;
3886         case RELAY_CONTROL:
3887                 type_str = "Control";
3888                 break;
3889         case RELAY_VIEWER_COMMAND:
3890                 type_str = "Viewer Command";
3891                 break;
3892         case RELAY_VIEWER_NOTIFICATION:
3893                 type_str = "Viewer Notification";
3894                 break;
3895         default:
3896                 type_str = "Unknown";
3897         }
3898         cleanup_connection_pollfd(events, pollfd);
3899         connection_put(conn);
3900         DBG("%s connection closed with %d", type_str, pollfd);
3901 }
3902
3903 /*
3904  * This thread does the actual work
3905  */
3906 static void *relay_thread_worker(void *data)
3907 {
3908         int ret, err = -1, last_seen_data_fd = -1;
3909         uint32_t nb_fd;
3910         struct lttng_poll_event events;
3911         struct lttng_ht *relay_connections_ht;
3912         struct lttng_ht_iter iter;
3913         struct relay_connection *destroy_conn = NULL;
3914
3915         DBG("[thread] Relay worker started");
3916
3917         rcu_register_thread();
3918
3919         health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3920
3921         if (testpoint(relayd_thread_worker)) {
3922                 goto error_testpoint;
3923         }
3924
3925         health_code_update();
3926
3927         /* table of connections indexed on socket */
3928         relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
3929         if (!relay_connections_ht) {
3930                 goto relay_connections_ht_error;
3931         }
3932
3933         ret = create_named_thread_poll_set(&events, 2, "Worker thread epoll");
3934         if (ret < 0) {
3935                 goto error_poll_create;
3936         }
3937
3938         ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
3939         if (ret < 0) {
3940                 goto error;
3941         }
3942
3943 restart:
3944         while (1) {
3945                 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3946
3947                 health_code_update();
3948
3949                 /* Infinite blocking call, waiting for transmission */
3950                 DBG3("Relayd worker thread polling...");
3951                 health_poll_entry();
3952                 ret = lttng_poll_wait(&events, -1);
3953                 health_poll_exit();
3954                 if (ret < 0) {
3955                         /*
3956                          * Restart interrupted system call.
3957                          */
3958                         if (errno == EINTR) {
3959                                 goto restart;
3960                         }
3961                         goto error;
3962                 }
3963
3964                 nb_fd = ret;
3965
3966                 /*
3967                  * Process control. The control connection is
3968                  * prioritized so we don't starve it with high
3969                  * throughput tracing data on the data connection.
3970                  */
3971                 for (i = 0; i < nb_fd; i++) {
3972                         /* Fetch once the poll data */
3973                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3974                         int pollfd = LTTNG_POLL_GETFD(&events, i);
3975
3976                         health_code_update();
3977
3978                         /* Thread quit pipe has been closed. Killing thread. */
3979                         ret = check_thread_quit_pipe(pollfd, revents);
3980                         if (ret) {
3981                                 err = 0;
3982                                 goto exit;
3983                         }
3984
3985                         /* Inspect the relay conn pipe for new connection */
3986                         if (pollfd == relay_conn_pipe[0]) {
3987                                 if (revents & LPOLLIN) {
3988                                         struct relay_connection *conn;
3989
3990                                         ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
3991                                         if (ret < 0) {
3992                                                 goto error;
3993                                         }
3994                                         ret = lttng_poll_add(&events,
3995                                                         conn->sock->fd,
3996                                                         LPOLLIN | LPOLLRDHUP);
3997                                         if (ret) {
3998                                                 ERR("Failed to add new connection file descriptor to poll set");
3999                                                 goto error;
4000                                         }
4001                                         connection_ht_add(relay_connections_ht, conn);
4002                                         DBG("Connection socket %d added", conn->sock->fd);
4003                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4004                                         ERR("Relay connection pipe error");
4005                                         goto error;
4006                                 } else {
4007                                         ERR("Unexpected poll events %u for sock %d", revents, pollfd);
4008                                         goto error;
4009                                 }
4010                         } else {
4011                                 struct relay_connection *ctrl_conn;
4012
4013                                 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4014                                 /* If not found, there is a synchronization issue. */
4015                                 assert(ctrl_conn);
4016
4017                                 if (ctrl_conn->type == RELAY_DATA) {
4018                                         if (revents & LPOLLIN) {
4019                                                 /*
4020                                                  * Flag the last seen data fd not deleted. It will be
4021                                                  * used as the last seen fd if any fd gets deleted in
4022                                                  * this first loop.
4023                                                  */
4024                                                 last_notdel_data_fd = pollfd;
4025                                         }
4026                                         goto put_ctrl_connection;
4027                                 }
4028                                 assert(ctrl_conn->type == RELAY_CONTROL);
4029
4030                                 if (revents & LPOLLIN) {
4031                                         enum relay_connection_status status;
4032
4033                                         status = relay_process_control(ctrl_conn);
4034                                         if (status != RELAY_CONNECTION_STATUS_OK) {
4035                                                 /*
4036                                                  * On socket error flag the session as aborted to force
4037                                                  * the cleanup of its stream otherwise it can leak
4038                                                  * during the lifetime of the relayd.
4039                                                  *
4040                                                  * This prevents situations in which streams can be
4041                                                  * left opened because an index was received, the
4042                                                  * control connection is closed, and the data
4043                                                  * connection is closed (uncleanly) before the packet's
4044                                                  * data provided.
4045                                                  *
4046                                                  * Since the control connection encountered an error,
4047                                                  * it is okay to be conservative and close the
4048                                                  * session right now as we can't rely on the protocol
4049                                                  * being respected anymore.
4050                                                  */
4051                                                 if (status == RELAY_CONNECTION_STATUS_ERROR) {
4052                                                         session_abort(ctrl_conn->session);
4053                                                 }
4054
4055                                                 /* Clear the connection on error or close. */
4056                                                 relay_thread_close_connection(&events,
4057                                                                 pollfd,
4058                                                                 ctrl_conn);
4059                                         }
4060                                         seen_control = 1;
4061                                 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4062                                         relay_thread_close_connection(&events,
4063                                                         pollfd, ctrl_conn);
4064                                         if (last_seen_data_fd == pollfd) {
4065                                                 last_seen_data_fd = last_notdel_data_fd;
4066                                         }
4067                                 } else {
4068                                         ERR("Unexpected poll events %u for control sock %d",
4069                                                         revents, pollfd);
4070                                         connection_put(ctrl_conn);
4071                                         goto error;
4072                                 }
4073                         put_ctrl_connection:
4074                                 connection_put(ctrl_conn);
4075                         }
4076                 }
4077
4078                 /*
4079                  * The last loop handled a control request, go back to poll to make
4080                  * sure we prioritise the control socket.
4081                  */
4082                 if (seen_control) {
4083                         continue;
4084                 }
4085
4086                 if (last_seen_data_fd >= 0) {
4087                         for (i = 0; i < nb_fd; i++) {
4088                                 int pollfd = LTTNG_POLL_GETFD(&events, i);
4089
4090                                 health_code_update();
4091
4092                                 if (last_seen_data_fd == pollfd) {
4093                                         idx = i;
4094                                         break;
4095                                 }
4096                         }
4097                 }
4098
4099                 /* Process data connection. */
4100                 for (i = idx + 1; i < nb_fd; i++) {
4101                         /* Fetch the poll data. */
4102                         uint32_t revents = LTTNG_POLL_GETEV(&events, i);
4103                         int pollfd = LTTNG_POLL_GETFD(&events, i);
4104                         struct relay_connection *data_conn;
4105
4106                         health_code_update();
4107
4108                         if (!revents) {
4109                                 /* No activity for this FD (poll implementation). */
4110                                 continue;
4111                         }
4112
4113                         /* Skip the command pipe. It's handled in the first loop. */
4114                         if (pollfd == relay_conn_pipe[0]) {
4115                                 continue;
4116                         }
4117
4118                         data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
4119                         if (!data_conn) {
4120                                 /* Skip it. Might be removed before. */
4121                                 continue;
4122                         }
4123                         if (data_conn->type == RELAY_CONTROL) {
4124                                 goto put_data_connection;
4125                         }
4126                         assert(data_conn->type == RELAY_DATA);
4127
4128                         if (revents & LPOLLIN) {
4129                                 enum relay_connection_status status;
4130
4131                                 status = relay_process_data(data_conn);
4132                                 /* Connection closed or error. */
4133                                 if (status != RELAY_CONNECTION_STATUS_OK) {
4134                                         /*
4135                                          * On socket error flag the session as aborted to force
4136                                          * the cleanup of its stream otherwise it can leak
4137                                          * during the lifetime of the relayd.
4138                                          *
4139                                          * This prevents situations in which streams can be
4140                                          * left opened because an index was received, the
4141                                          * control connection is closed, and the data
4142                                          * connection is closed (uncleanly) before the packet's
4143                                          * data provided.
4144                                          *
4145                                          * Since the data connection encountered an error,
4146                                          * it is okay to be conservative and close the
4147                                          * session right now as we can't rely on the protocol
4148                                          * being respected anymore.
4149                                          */
4150                                         if (status == RELAY_CONNECTION_STATUS_ERROR) {
4151                                                 session_abort(data_conn->session);
4152                                         }
4153                                         relay_thread_close_connection(&events, pollfd,
4154                                                         data_conn);
4155                                         /*
4156                                          * Every goto restart call sets the last seen fd where
4157                                          * here we don't really care since we gracefully
4158                                          * continue the loop after the connection is deleted.
4159                                          */
4160                                 } else {
4161                                         /* Keep last seen port. */
4162                                         last_seen_data_fd = pollfd;
4163                                         connection_put(data_conn);
4164                                         goto restart;
4165                                 }
4166                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
4167                                 relay_thread_close_connection(&events, pollfd,
4168                                                 data_conn);
4169                         } else {
4170                                 ERR("Unknown poll events %u for data sock %d",
4171                                                 revents, pollfd);
4172                         }
4173                 put_data_connection:
4174                         connection_put(data_conn);
4175                 }
4176                 last_seen_data_fd = -1;
4177         }
4178
4179         /* Normal exit, no error */
4180         ret = 0;
4181
4182 exit:
4183 error:
4184         /* Cleanup remaining connection object. */
4185         rcu_read_lock();
4186         cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
4187                         destroy_conn,
4188                         sock_n.node) {
4189                 health_code_update();
4190
4191                 session_abort(destroy_conn->session);
4192
4193                 /*
4194                  * No need to grab another ref, because we own
4195                  * destroy_conn.
4196                  */
4197                 relay_thread_close_connection(&events, destroy_conn->sock->fd,
4198                                 destroy_conn);
4199         }
4200         rcu_read_unlock();
4201
4202         (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
4203 error_poll_create:
4204         lttng_ht_destroy(relay_connections_ht);
4205 relay_connections_ht_error:
4206         /* Close relay conn pipes */
4207         (void) fd_tracker_util_pipe_close(the_fd_tracker,
4208                         relay_conn_pipe);
4209         if (err) {
4210                 DBG("Thread exited with error");
4211         }
4212         DBG("Worker thread cleanup complete");
4213 error_testpoint:
4214         if (err) {
4215                 health_error();
4216                 ERR("Health error occurred in %s", __func__);
4217         }
4218         health_unregister(health_relayd);
4219         rcu_unregister_thread();
4220         lttng_relay_stop_threads();
4221         return NULL;
4222 }
4223
4224 /*
4225  * Create the relay command pipe to wake thread_manage_apps.
4226  * Closed in cleanup().
4227  */
4228 static int create_relay_conn_pipe(void)
4229 {
4230         return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
4231                         "Relayd connection pipe", relay_conn_pipe);
4232 }
4233
4234 static int stdio_open(void *data, int *fds)
4235 {
4236         fds[0] = fileno(stdout);
4237         fds[1] = fileno(stderr);
4238         return 0;
4239 }
4240
4241 static int track_stdio(void)
4242 {
4243         int fds[2];
4244         const char *names[] = { "stdout", "stderr" };
4245
4246         return fd_tracker_open_unsuspendable_fd(the_fd_tracker, fds,
4247                         names, 2, stdio_open, NULL);
4248 }
4249
4250 /*
4251  * main
4252  */
4253 int main(int argc, char **argv)
4254 {
4255         bool thread_is_rcu_registered = false;
4256         int ret = 0, retval = 0;
4257         void *status;
4258         char *unlinked_file_directory_path = NULL, *output_path = NULL;
4259
4260         /* Parse environment variables */
4261         ret = parse_env_options();
4262         if (ret) {
4263                 retval = -1;
4264                 goto exit_options;
4265         }
4266
4267         /*
4268          * Parse arguments.
4269          * Command line arguments overwrite environment.
4270          */
4271         progname = argv[0];
4272         if (set_options(argc, argv)) {
4273                 retval = -1;
4274                 goto exit_options;
4275         }
4276
4277         if (set_signal_handler()) {
4278                 retval = -1;
4279                 goto exit_options;
4280         }
4281
4282         relayd_config_log();
4283
4284         if (opt_print_version) {
4285                 print_version();
4286                 retval = 0;
4287                 goto exit_options;
4288         }
4289
4290         ret = fclose(stdin);
4291         if (ret) {
4292                 PERROR("Failed to close stdin");
4293                 goto exit_options;
4294         }
4295
4296         DBG("Clear command %s", opt_allow_clear ? "allowed" : "disallowed");
4297
4298         /* Try to create directory if -o, --output is specified. */
4299         if (opt_output_path) {
4300                 if (*opt_output_path != '/') {
4301                         ERR("Please specify an absolute path for -o, --output PATH");
4302                         retval = -1;
4303                         goto exit_options;
4304                 }
4305
4306                 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
4307                                 -1, -1);
4308                 if (ret < 0) {
4309                         ERR("Unable to create %s", opt_output_path);
4310                         retval = -1;
4311                         goto exit_options;
4312                 }
4313         }
4314
4315         /* Daemonize */
4316         if (opt_daemon || opt_background) {
4317                 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
4318                         !opt_background);
4319                 if (ret < 0) {
4320                         retval = -1;
4321                         goto exit_options;
4322                 }
4323         }
4324
4325         if (opt_working_directory) {
4326                 ret = utils_change_working_directory(opt_working_directory);
4327                 if (ret) {
4328                         /* All errors are already logged. */
4329                         goto exit_options;
4330                 }
4331         }
4332
4333         sessiond_trace_chunk_registry = sessiond_trace_chunk_registry_create();
4334         if (!sessiond_trace_chunk_registry) {
4335                 ERR("Failed to initialize session daemon trace chunk registry");
4336                 retval = -1;
4337                 goto exit_options;
4338         }
4339
4340         /*
4341          * The RCU thread registration (and use, through the fd-tracker's
4342          * creation) is done after the daemonization to allow us to not
4343          * deal with liburcu's fork() management as the call RCU needs to
4344          * be restored.
4345          */
4346         rcu_register_thread();
4347         thread_is_rcu_registered = true;
4348
4349         output_path = create_output_path("");
4350         if (!output_path) {
4351                 ERR("Failed to get output path");
4352                 retval = -1;
4353                 goto exit_options;
4354         }
4355         ret = asprintf(&unlinked_file_directory_path, "%s/%s", output_path,
4356                         DEFAULT_UNLINKED_FILES_DIRECTORY);
4357         free(output_path);
4358         if (ret < 0) {
4359                 ERR("Failed to format unlinked file directory path");
4360                 retval = -1;
4361                 goto exit_options;
4362         }
4363         the_fd_tracker = fd_tracker_create(
4364                         unlinked_file_directory_path, lttng_opt_fd_pool_size);
4365         free(unlinked_file_directory_path);
4366         if (!the_fd_tracker) {
4367                 retval = -1;
4368                 goto exit_options;
4369         }
4370
4371         ret = track_stdio();
4372         if (ret) {
4373                 retval = -1;
4374                 goto exit_options;
4375         }
4376
4377         /* Initialize thread health monitoring */
4378         health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
4379         if (!health_relayd) {
4380                 PERROR("health_app_create error");
4381                 retval = -1;
4382                 goto exit_options;
4383         }
4384
4385         /* Create thread quit pipe */
4386         if (init_thread_quit_pipe()) {
4387                 retval = -1;
4388                 goto exit_options;
4389         }
4390
4391         /* Setup the thread apps communication pipe. */
4392         if (create_relay_conn_pipe()) {
4393                 retval = -1;
4394                 goto exit_options;
4395         }
4396
4397         /* Init relay command queue. */
4398         cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
4399
4400         /* Initialize communication library */
4401         lttcomm_init();
4402         lttcomm_inet_init();
4403
4404         /* tables of sessions indexed by session ID */
4405         sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4406         if (!sessions_ht) {
4407                 retval = -1;
4408                 goto exit_options;
4409         }
4410
4411         /* tables of streams indexed by stream ID */
4412         relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4413         if (!relay_streams_ht) {
4414                 retval = -1;
4415                 goto exit_options;
4416         }
4417
4418         /* tables of streams indexed by stream ID */
4419         viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4420         if (!viewer_streams_ht) {
4421                 retval = -1;
4422                 goto exit_options;
4423         }
4424
4425         ret = init_health_quit_pipe();
4426         if (ret) {
4427                 retval = -1;
4428                 goto exit_options;
4429         }
4430
4431         /* Create thread to manage the client socket */
4432         ret = pthread_create(&health_thread, default_pthread_attr(),
4433                         thread_manage_health, (void *) NULL);
4434         if (ret) {
4435                 errno = ret;
4436                 PERROR("pthread_create health");
4437                 retval = -1;
4438                 goto exit_options;
4439         }
4440
4441         /* Setup the dispatcher thread */
4442         ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
4443                         relay_thread_dispatcher, (void *) NULL);
4444         if (ret) {
4445                 errno = ret;
4446                 PERROR("pthread_create dispatcher");
4447                 retval = -1;
4448                 goto exit_dispatcher_thread;
4449         }
4450
4451         /* Setup the worker thread */
4452         ret = pthread_create(&worker_thread, default_pthread_attr(),
4453                         relay_thread_worker, NULL);
4454         if (ret) {
4455                 errno = ret;
4456                 PERROR("pthread_create worker");
4457                 retval = -1;
4458                 goto exit_worker_thread;
4459         }
4460
4461         /* Setup the listener thread */
4462         ret = pthread_create(&listener_thread, default_pthread_attr(),
4463                         relay_thread_listener, (void *) NULL);
4464         if (ret) {
4465                 errno = ret;
4466                 PERROR("pthread_create listener");
4467                 retval = -1;
4468                 goto exit_listener_thread;
4469         }
4470
4471         ret = relayd_live_create(live_uri);
4472         if (ret) {
4473                 ERR("Starting live viewer threads");
4474                 retval = -1;
4475                 goto exit_live;
4476         }
4477
4478         /*
4479          * This is where we start awaiting program completion (e.g. through
4480          * signal that asks threads to teardown).
4481          */
4482
4483         ret = relayd_live_join();
4484         if (ret) {
4485                 retval = -1;
4486         }
4487 exit_live:
4488
4489         ret = pthread_join(listener_thread, &status);
4490         if (ret) {
4491                 errno = ret;
4492                 PERROR("pthread_join listener_thread");
4493                 retval = -1;
4494         }
4495
4496 exit_listener_thread:
4497         ret = pthread_join(worker_thread, &status);
4498         if (ret) {
4499                 errno = ret;
4500                 PERROR("pthread_join worker_thread");
4501                 retval = -1;
4502         }
4503
4504 exit_worker_thread:
4505         ret = pthread_join(dispatcher_thread, &status);
4506         if (ret) {
4507                 errno = ret;
4508                 PERROR("pthread_join dispatcher_thread");
4509                 retval = -1;
4510         }
4511 exit_dispatcher_thread:
4512
4513         ret = pthread_join(health_thread, &status);
4514         if (ret) {
4515                 errno = ret;
4516                 PERROR("pthread_join health_thread");
4517                 retval = -1;
4518         }
4519 exit_options:
4520         /*
4521          * Wait for all pending call_rcu work to complete before tearing
4522          * down data structures. call_rcu worker may be trying to
4523          * perform lookups in those structures.
4524          */
4525         rcu_barrier();
4526         relayd_cleanup();
4527
4528         /* Ensure all prior call_rcu are done. */
4529         rcu_barrier();
4530
4531         if (thread_is_rcu_registered) {
4532                 rcu_unregister_thread();
4533         }
4534
4535         if (!retval) {
4536                 exit(EXIT_SUCCESS);
4537         } else {
4538                 exit(EXIT_FAILURE);
4539         }
4540 }