src/bin/lttng-consumerd/lttng-consumerd.c

   1 /*
   2  * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
   3  *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _LGPL_SOURCE
  20 #include <fcntl.h>
  21 #include <getopt.h>
  22 #include <grp.h>
  23 #include <limits.h>
  24 #include <pthread.h>
  25 #include <signal.h>
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include <sys/ipc.h>
  30 #include <sys/resource.h>
  31 #include <sys/shm.h>
  32 #include <sys/socket.h>
  33 #include <sys/stat.h>
  34 #include <sys/types.h>
  35 #include <urcu/list.h>
  36 #include <poll.h>
  37 #include <unistd.h>
  38 #include <sys/mman.h>
  39 #include <assert.h>
  40 #include <urcu/compiler.h>
  41 #include <ulimit.h>
  42
  43 #include <common/defaults.h>
  44 #include <common/common.h>
  45 #include <common/consumer/consumer.h>
  46 #include <common/consumer/consumer-timer.h>
  47 #include <common/compat/poll.h>
  48 #include <common/compat/getenv.h>
  49 #include <common/sessiond-comm/sessiond-comm.h>
  50 #include <common/utils.h>
  51
  52 #include "lttng-consumerd.h"
  53 #include "health-consumerd.h"
  54
  55 /* threads (channel handling, poll, metadata, sessiond) */
  56
  57 static pthread_t channel_thread, data_thread, metadata_thread,
  58                 sessiond_thread, metadata_timer_thread, health_thread;
  59
  60 /* to count the number of times the user pressed ctrl+c */
  61 static int sigintcount = 0;
  62
  63 /* Argument variables */
  64 int lttng_opt_quiet;    /* not static in error.h */
  65 int lttng_opt_verbose;  /* not static in error.h */
  66 int lttng_opt_mi;       /* not static in error.h */
  67
  68 static int opt_daemon;
  69 static const char *progname;
  70 static char command_sock_path[PATH_MAX]; /* Global command socket path */
  71 static char error_sock_path[PATH_MAX]; /* Global error path */
  72 static enum lttng_consumer_type opt_type = LTTNG_CONSUMER_KERNEL;
  73
  74 /* the liblttngconsumerd context */
  75 static struct lttng_consumer_local_data *ctx;
  76
  77 /* Consumerd health monitoring */
  78 struct health_app *health_consumerd;
  79
  80 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
  81
  82 int lttng_consumer_ready = NR_LTTNG_CONSUMER_READY;
  83
  84 enum lttng_consumer_type lttng_consumer_get_type(void)
  85 {
  86         if (!ctx) {
  87                 return LTTNG_CONSUMER_UNKNOWN;
  88         }
  89         return ctx->type;
  90 }
  91
  92 /*
  93  * Signal handler for the daemon
  94  */
  95 static void sighandler(int sig)
  96 {
  97         if (sig == SIGINT && sigintcount++ == 0) {
  98                 DBG("ignoring first SIGINT");
  99                 return;
 100         }
 101
 102         if (ctx) {
 103                 lttng_consumer_should_exit(ctx);
 104         }
 105 }
 106
 107 /*
 108  * Setup signal handler for :
 109  *      SIGINT, SIGTERM, SIGPIPE
 110  */
 111 static int set_signal_handler(void)
 112 {
 113         int ret = 0;
 114         struct sigaction sa;
 115         sigset_t sigset;
 116
 117         if ((ret = sigemptyset(&sigset)) < 0) {
 118                 PERROR("sigemptyset");
 119                 return ret;
 120         }
 121
 122         sa.sa_mask = sigset;
 123         sa.sa_flags = 0;
 124
 125         sa.sa_handler = sighandler;
 126         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
 127                 PERROR("sigaction");
 128                 return ret;
 129         }
 130
 131         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
 132                 PERROR("sigaction");
 133                 return ret;
 134         }
 135
 136         sa.sa_handler = SIG_IGN;
 137         if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
 138                 PERROR("sigaction");
 139                 return ret;
 140         }
 141
 142         return ret;
 143 }
 144
 145 /*
 146  * Usage function on stream file.
 147  */
 148 static void usage(FILE *fp)
 149 {
 150         fprintf(fp, "Usage: %s OPTIONS\n\nOptions:\n", progname);
 151         fprintf(fp, "  -h, --help                         "
 152                         "Display this usage.\n");
 153         fprintf(fp, "  -c, --consumerd-cmd-sock PATH      "
 154                         "Specify path for the command socket\n");
 155         fprintf(fp, "  -e, --consumerd-err-sock PATH      "
 156                         "Specify path for the error socket\n");
 157         fprintf(fp, "  -d, --daemonize                    "
 158                         "Start as a daemon.\n");
 159         fprintf(fp, "  -q, --quiet                        "
 160                         "No output at all.\n");
 161         fprintf(fp, "  -v, --verbose                      "
 162                         "Verbose mode. Activate DBG() macro.\n");
 163         fprintf(fp, "  -V, --version                      "
 164                         "Show version number.\n");
 165         fprintf(fp, "  -g, --group NAME                   "
 166                         "Specify the tracing group name. (default: tracing)\n");
 167         fprintf(fp, "  -k, --kernel                       "
 168                         "Consumer kernel buffers (default).\n");
 169         fprintf(fp, "  -u, --ust                          "
 170                         "Consumer UST buffers.%s\n",
 171 #ifdef HAVE_LIBLTTNG_UST_CTL
 172                         ""
 173 #else
 174                         " (support not compiled in)"
 175 #endif
 176                         );
 177 }
 178
 179 /*
 180  * daemon argument parsing
 181  */
 182 static int parse_args(int argc, char **argv)
 183 {
 184         int c, ret = 0;
 185
 186         static struct option long_options[] = {
 187                 { "consumerd-cmd-sock", 1, 0, 'c' },
 188                 { "consumerd-err-sock", 1, 0, 'e' },
 189                 { "daemonize", 0, 0, 'd' },
 190                 { "group", 1, 0, 'g' },
 191                 { "help", 0, 0, 'h' },
 192                 { "quiet", 0, 0, 'q' },
 193                 { "verbose", 0, 0, 'v' },
 194                 { "version", 0, 0, 'V' },
 195                 { "kernel", 0, 0, 'k' },
 196 #ifdef HAVE_LIBLTTNG_UST_CTL
 197                 { "ust", 0, 0, 'u' },
 198 #endif
 199                 { NULL, 0, 0, 0 }
 200         };
 201
 202         while (1) {
 203                 int option_index = 0;
 204                 c = getopt_long(argc, argv, "dhqvVku" "c:e:g:",
 205                                 long_options, &option_index);
 206                 if (c == -1) {
 207                         break;
 208                 }
 209
 210                 switch (c) {
 211                 case 0:
 212                         fprintf(stderr, "option %s",
 213                                 long_options[option_index].name);
 214                         if (optarg) {
 215                                 fprintf(stderr, " with arg %s\n", optarg);
 216                                 ret = -1;
 217                                 goto end;
 218                         }
 219                         break;
 220                 case 'c':
 221                         if (lttng_is_setuid_setgid()) {
 222                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 223                                         "-c, --consumerd-cmd-sock");
 224                         } else {
 225                                 snprintf(command_sock_path, PATH_MAX, "%s", optarg);
 226                         }
 227                         break;
 228                 case 'e':
 229                         if (lttng_is_setuid_setgid()) {
 230                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 231                                         "-e, --consumerd-err-sock");
 232                         } else {
 233                                 snprintf(error_sock_path, PATH_MAX, "%s", optarg);
 234                         }
 235                         break;
 236                 case 'd':
 237                         opt_daemon = 1;
 238                         break;
 239                 case 'g':
 240                         if (lttng_is_setuid_setgid()) {
 241                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 242                                         "-g, --group");
 243                         } else {
 244                                 tracing_group_name = optarg;
 245                         }
 246                         break;
 247                 case 'h':
 248                         usage(stdout);
 249                         exit(EXIT_SUCCESS);
 250                 case 'q':
 251                         lttng_opt_quiet = 1;
 252                         break;
 253                 case 'v':
 254                         lttng_opt_verbose = 1;
 255                         break;
 256                 case 'V':
 257                         fprintf(stdout, "%s\n", VERSION);
 258                         exit(EXIT_SUCCESS);
 259                 case 'k':
 260                         opt_type = LTTNG_CONSUMER_KERNEL;
 261                         break;
 262 #ifdef HAVE_LIBLTTNG_UST_CTL
 263                 case 'u':
 264 # if (CAA_BITS_PER_LONG == 64)
 265                         opt_type = LTTNG_CONSUMER64_UST;
 266 # elif (CAA_BITS_PER_LONG == 32)
 267                         opt_type = LTTNG_CONSUMER32_UST;
 268 # else
 269 #  error "Unknown bitness"
 270 # endif
 271                         break;
 272 #endif
 273                 default:
 274                         usage(stderr);
 275                         ret = -1;
 276                         goto end;
 277                 }
 278         }
 279 end:
 280         return ret;
 281 }
 282
 283 /*
 284  * Set open files limit to unlimited. This daemon can open a large number of
 285  * file descriptors in order to consumer multiple kernel traces.
 286  */
 287 static void set_ulimit(void)
 288 {
 289         int ret;
 290         struct rlimit lim;
 291
 292         /* The kernel does not allowed an infinite limit for open files */
 293         lim.rlim_cur = 65535;
 294         lim.rlim_max = 65535;
 295
 296         ret = setrlimit(RLIMIT_NOFILE, &lim);
 297         if (ret < 0) {
 298                 PERROR("failed to set open files limit");
 299         }
 300 }
 301
 302 /*
 303  * main
 304  */
 305 int main(int argc, char **argv)
 306 {
 307         int ret = 0, retval = 0;
 308         void *status;
 309         struct lttng_consumer_local_data *tmp_ctx;
 310
 311         if (set_signal_handler()) {
 312                 retval = -1;
 313                 goto exit_set_signal_handler;
 314         }
 315
 316         /* Parse arguments */
 317         progname = argv[0];
 318         if (parse_args(argc, argv)) {
 319                 retval = -1;
 320                 goto exit_options;
 321         }
 322
 323         /* Daemonize */
 324         if (opt_daemon) {
 325                 int i;
 326
 327                 /*
 328                  * fork
 329                  * child: setsid, close FD 0, 1, 2, chdir /
 330                  * parent: exit (if fork is successful)
 331                  */
 332                 ret = daemon(0, 0);
 333                 if (ret < 0) {
 334                         PERROR("daemon");
 335                         retval = -1;
 336                         goto exit_options;
 337                 }
 338                 /*
 339                  * We are in the child. Make sure all other file
 340                  * descriptors are closed, in case we are called with
 341                  * more opened file descriptors than the standard ones.
 342                  */
 343                 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
 344                         (void) close(i);
 345                 }
 346         }
 347
 348         /*
 349          * Starting from here, we can create threads. This needs to be after
 350          * lttng_daemonize due to RCU.
 351          */
 352
 353         health_consumerd = health_app_create(NR_HEALTH_CONSUMERD_TYPES);
 354         if (!health_consumerd) {
 355                 retval = -1;
 356                 goto exit_health_consumerd_cleanup;
 357         }
 358
 359         /* Set up max poll set size */
 360         if (lttng_poll_set_max_size()) {
 361                 retval = -1;
 362                 goto exit_init_data;
 363         }
 364
 365         if (*command_sock_path == '\0') {
 366                 switch (opt_type) {
 367                 case LTTNG_CONSUMER_KERNEL:
 368                         ret = snprintf(command_sock_path, PATH_MAX,
 369                                         DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
 370                                         DEFAULT_LTTNG_RUNDIR);
 371                         if (ret < 0) {
 372                                 retval = -1;
 373                                 goto exit_init_data;
 374                         }
 375                         break;
 376                 case LTTNG_CONSUMER64_UST:
 377                         ret = snprintf(command_sock_path, PATH_MAX,
 378                                         DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
 379                                         DEFAULT_LTTNG_RUNDIR);
 380                         if (ret < 0) {
 381                                 retval = -1;
 382                                 goto exit_init_data;
 383                         }
 384                         break;
 385                 case LTTNG_CONSUMER32_UST:
 386                         ret = snprintf(command_sock_path, PATH_MAX,
 387                                         DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
 388                                         DEFAULT_LTTNG_RUNDIR);
 389                         if (ret < 0) {
 390                                 retval = -1;
 391                                 goto exit_init_data;
 392                         }
 393                         break;
 394                 default:
 395                         ERR("Unknown consumerd type");
 396                         retval = -1;
 397                         goto exit_init_data;
 398                 }
 399         }
 400
 401         /* Init */
 402         if (lttng_consumer_init()) {
 403                 retval = -1;
 404                 goto exit_init_data;
 405         }
 406
 407         /* Initialize communication library */
 408         lttcomm_init();
 409         /* Initialize TCP timeout values */
 410         lttcomm_inet_init();
 411
 412         if (!getuid()) {
 413                 /* Set limit for open files */
 414                 set_ulimit();
 415         }
 416
 417         if (run_as_create_worker(argv[0]) < 0) {
 418                 goto exit_init_data;
 419         }
 420
 421         /* create the consumer instance with and assign the callbacks */
 422         ctx = lttng_consumer_create(opt_type, lttng_consumer_read_subbuffer,
 423                 NULL, lttng_consumer_on_recv_stream, NULL);
 424         if (!ctx) {
 425                 retval = -1;
 426                 goto exit_init_data;
 427         }
 428
 429         lttng_consumer_set_command_sock_path(ctx, command_sock_path);
 430         if (*error_sock_path == '\0') {
 431                 switch (opt_type) {
 432                 case LTTNG_CONSUMER_KERNEL:
 433                         ret = snprintf(error_sock_path, PATH_MAX,
 434                                         DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
 435                                         DEFAULT_LTTNG_RUNDIR);
 436                         if (ret < 0) {
 437                                 retval = -1;
 438                                 goto exit_init_data;
 439                         }
 440                         break;
 441                 case LTTNG_CONSUMER64_UST:
 442                         ret = snprintf(error_sock_path, PATH_MAX,
 443                                         DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
 444                                         DEFAULT_LTTNG_RUNDIR);
 445                         if (ret < 0) {
 446                                 retval = -1;
 447                                 goto exit_init_data;
 448                         }
 449                         break;
 450                 case LTTNG_CONSUMER32_UST:
 451                         ret = snprintf(error_sock_path, PATH_MAX,
 452                                         DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
 453                                         DEFAULT_LTTNG_RUNDIR);
 454                         if (ret < 0) {
 455                                 retval = -1;
 456                                 goto exit_init_data;
 457                         }
 458                         break;
 459                 default:
 460                         ERR("Unknown consumerd type");
 461                         retval = -1;
 462                         goto exit_init_data;
 463                 }
 464         }
 465
 466         /* Connect to the socket created by lttng-sessiond to report errors */
 467         DBG("Connecting to error socket %s", error_sock_path);
 468         ret = lttcomm_connect_unix_sock(error_sock_path);
 469         /*
 470          * Not a fatal error, but all communication with lttng-sessiond will
 471          * fail.
 472          */
 473         if (ret < 0) {
 474                 WARN("Cannot connect to error socket (is lttng-sessiond started?)");
 475         }
 476         lttng_consumer_set_error_sock(ctx, ret);
 477
 478         /*
 479          * Block RT signals used for UST periodical metadata flush and the live
 480          * timer in main, and create a dedicated thread to handle these signals.
 481          */
 482         if (consumer_signal_init()) {
 483                 retval = -1;
 484                 goto exit_init_data;
 485         }
 486
 487         ctx->type = opt_type;
 488
 489         if (utils_create_pipe(health_quit_pipe)) {
 490                 retval = -1;
 491                 goto exit_health_pipe;
 492         }
 493
 494         /* Create thread to manage the client socket */
 495         ret = pthread_create(&health_thread, NULL,
 496                         thread_manage_health, (void *) NULL);
 497         if (ret) {
 498                 errno = ret;
 499                 PERROR("pthread_create health");
 500                 retval = -1;
 501                 goto exit_health_thread;
 502         }
 503
 504         /*
 505          * Wait for health thread to be initialized before letting the
 506          * sessiond thread reply to the sessiond that we are ready.
 507          */
 508         while (uatomic_read(&lttng_consumer_ready)) {
 509                 usleep(100000);
 510         }
 511         cmm_smp_mb();   /* Read ready before following operations */
 512
 513         /* Create thread to manage channels */
 514         ret = pthread_create(&channel_thread, NULL,
 515                         consumer_thread_channel_poll,
 516                         (void *) ctx);
 517         if (ret) {
 518                 errno = ret;
 519                 PERROR("pthread_create");
 520                 retval = -1;
 521                 goto exit_channel_thread;
 522         }
 523
 524         /* Create thread to manage the polling/writing of trace metadata */
 525         ret = pthread_create(&metadata_thread, NULL,
 526                         consumer_thread_metadata_poll,
 527                         (void *) ctx);
 528         if (ret) {
 529                 errno = ret;
 530                 PERROR("pthread_create");
 531                 retval = -1;
 532                 goto exit_metadata_thread;
 533         }
 534
 535         /* Create thread to manage the polling/writing of trace data */
 536         ret = pthread_create(&data_thread, NULL, consumer_thread_data_poll,
 537                         (void *) ctx);
 538         if (ret) {
 539                 errno = ret;
 540                 PERROR("pthread_create");
 541                 retval = -1;
 542                 goto exit_data_thread;
 543         }
 544
 545         /* Create the thread to manage the receive of fd */
 546         ret = pthread_create(&sessiond_thread, NULL,
 547                         consumer_thread_sessiond_poll,
 548                         (void *) ctx);
 549         if (ret) {
 550                 errno = ret;
 551                 PERROR("pthread_create");
 552                 retval = -1;
 553                 goto exit_sessiond_thread;
 554         }
 555
 556         /*
 557          * Create the thread to manage the UST metadata periodic timer and
 558          * live timer.
 559          */
 560         ret = pthread_create(&metadata_timer_thread, NULL,
 561                         consumer_timer_thread, (void *) ctx);
 562         if (ret) {
 563                 errno = ret;
 564                 PERROR("pthread_create");
 565                 retval = -1;
 566                 goto exit_metadata_timer_thread;
 567         }
 568
 569         ret = pthread_detach(metadata_timer_thread);
 570         if (ret) {
 571                 errno = ret;
 572                 PERROR("pthread_detach");
 573                 retval = -1;
 574                 goto exit_metadata_timer_detach;
 575         }
 576
 577         /*
 578          * This is where we start awaiting program completion (e.g. through
 579          * signal that asks threads to teardown.
 580          */
 581
 582 exit_metadata_timer_detach:
 583 exit_metadata_timer_thread:
 584         ret = pthread_join(sessiond_thread, &status);
 585         if (ret) {
 586                 errno = ret;
 587                 PERROR("pthread_join sessiond_thread");
 588                 retval = -1;
 589         }
 590 exit_sessiond_thread:
 591
 592         ret = pthread_join(data_thread, &status);
 593         if (ret) {
 594                 errno = ret;
 595                 PERROR("pthread_join data_thread");
 596                 retval = -1;
 597         }
 598 exit_data_thread:
 599
 600         ret = pthread_join(metadata_thread, &status);
 601         if (ret) {
 602                 errno = ret;
 603                 PERROR("pthread_join metadata_thread");
 604                 retval = -1;
 605         }
 606 exit_metadata_thread:
 607
 608         ret = pthread_join(channel_thread, &status);
 609         if (ret) {
 610                 errno = ret;
 611                 PERROR("pthread_join channel_thread");
 612                 retval = -1;
 613         }
 614 exit_channel_thread:
 615
 616         ret = pthread_join(health_thread, &status);
 617         if (ret) {
 618                 errno = ret;
 619                 PERROR("pthread_join health_thread");
 620                 retval = -1;
 621         }
 622 exit_health_thread:
 623
 624         utils_close_pipe(health_quit_pipe);
 625 exit_health_pipe:
 626
 627 exit_init_data:
 628         tmp_ctx = ctx;
 629         ctx = NULL;
 630         cmm_barrier();  /* Clear ctx for signal handler. */
 631         /*
 632          * Wait for all pending call_rcu work to complete before tearing
 633          * down data structures. call_rcu worker may be trying to
 634          * perform lookups in those structures.
 635          */
 636         rcu_barrier();
 637         lttng_consumer_destroy(tmp_ctx);
 638         lttng_consumer_cleanup();
 639
 640         if (health_consumerd) {
 641                 health_app_destroy(health_consumerd);
 642         }
 643         /* Ensure all prior call_rcu are done. */
 644         rcu_barrier();
 645
 646         run_as_destroy_worker();
 647
 648 exit_health_consumerd_cleanup:
 649 exit_options:
 650 exit_set_signal_handler:
 651
 652         if (!retval) {
 653                 exit(EXIT_SUCCESS);
 654         } else {
 655                 exit(EXIT_FAILURE);
 656         }
 657 }