src/bin/lttng-sessiond/agent-thread.c

   1 /*
   2  * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
   3  *
   4  * SPDX-License-Identifier: GPL-2.0-only
   5  *
   6  */
   7
   8 #define _LGPL_SOURCE
   9 #include <assert.h>
  10
  11 #include <common/common.h>
  12 #include <common/sessiond-comm/sessiond-comm.h>
  13 #include <common/uri.h>
  14 #include <common/utils.h>
  15
  16 #include <common/compat/endian.h>
  17
  18 #include "fd-limit.h"
  19 #include "agent-thread.h"
  20 #include "agent.h"
  21 #include "lttng-sessiond.h"
  22 #include "session.h"
  23 #include "utils.h"
  24 #include "thread.h"
  25
  26 struct thread_notifiers {
  27         struct lttng_pipe *quit_pipe;
  28         sem_t ready;
  29 };
  30
  31 struct agent_app_id {
  32         pid_t pid;
  33         enum lttng_domain_type domain;
  34 };
  35
  36 struct agent_protocol_version {
  37         unsigned int major, minor;
  38 };
  39
  40 static int agent_tracing_enabled = -1;
  41
  42 /*
  43  * Note that there is not port here. It's set after this URI is parsed so we
  44  * can let the user define a custom one. However, localhost is ALWAYS the
  45  * default listening address.
  46  */
  47 static const char *default_reg_uri =
  48         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS;
  49
  50 /*
  51  * Update agent application using the given socket. This is done just after
  52  * registration was successful.
  53  *
  54  * This will acquire the various sessions' lock; none must be held by the
  55  * caller.
  56  * The caller must hold the session list lock.
  57  */
  58 static void update_agent_app(const struct agent_app *app)
  59 {
  60         struct ltt_session *session, *stmp;
  61         struct ltt_session_list *list;
  62
  63         list = session_get_list();
  64         assert(list);
  65
  66         cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
  67                 if (!session_get(session)) {
  68                         continue;
  69                 }
  70
  71                 session_lock(session);
  72                 if (session->ust_session) {
  73                         const struct agent *agt;
  74
  75                         rcu_read_lock();
  76                         agt = trace_ust_find_agent(session->ust_session, app->domain);
  77                         if (agt) {
  78                                 agent_update(agt, app);
  79                         }
  80                         rcu_read_unlock();
  81                 }
  82                 session_unlock(session);
  83                 session_put(session);
  84         }
  85 }
  86
  87 /*
  88  * Create and init socket from uri.
  89  */
  90 static struct lttcomm_sock *init_tcp_socket(void)
  91 {
  92         int ret;
  93         struct lttng_uri *uri = NULL;
  94         struct lttcomm_sock *sock = NULL;
  95         unsigned int port;
  96         bool bind_succeeded = false;
  97
  98         /*
  99          * This should never fail since the URI is hardcoded and the port is set
 100          * before this thread is launched.
 101          */
 102         ret = uri_parse(default_reg_uri, &uri);
 103         assert(ret);
 104         assert(config.agent_tcp_port.begin > 0);
 105         uri->port = config.agent_tcp_port.begin;
 106
 107         sock = lttcomm_alloc_sock_from_uri(uri);
 108         uri_free(uri);
 109         if (sock == NULL) {
 110                 ERR("[agent-thread] agent allocating TCP socket");
 111                 goto error;
 112         }
 113
 114         ret = lttcomm_create_sock(sock);
 115         if (ret < 0) {
 116                 goto error;
 117         }
 118
 119         for (port = config.agent_tcp_port.begin;
 120                         port <= config.agent_tcp_port.end; port++) {
 121                 ret = lttcomm_sock_set_port(sock, (uint16_t) port);
 122                 if (ret) {
 123                         ERR("[agent-thread] Failed to set port %u on socket",
 124                                         port);
 125                         goto error;
 126                 }
 127                 DBG3("[agent-thread] Trying to bind on port %u", port);
 128                 ret = sock->ops->bind(sock);
 129                 if (!ret) {
 130                         bind_succeeded = true;
 131                         break;
 132                 }
 133
 134                 if (errno == EADDRINUSE) {
 135                         DBG("Failed to bind to port %u since it is already in use",
 136                                         port);
 137                 } else {
 138                         PERROR("Failed to bind to port %u", port);
 139                         goto error;
 140                 }
 141         }
 142
 143         if (!bind_succeeded) {
 144                 if (config.agent_tcp_port.begin == config.agent_tcp_port.end) {
 145                         WARN("Another process is already using the agent port %i. "
 146                                         "Agent support will be deactivated.",
 147                                         config.agent_tcp_port.begin);
 148                         goto error;
 149                 } else {
 150                         WARN("All ports in the range [%i, %i] are already in use. "
 151                                         "Agent support will be deactivated.",
 152                                         config.agent_tcp_port.begin,
 153                                         config.agent_tcp_port.end);
 154                         goto error;
 155                 }
 156         }
 157
 158         ret = sock->ops->listen(sock, -1);
 159         if (ret < 0) {
 160                 goto error;
 161         }
 162
 163         DBG("[agent-thread] Listening on TCP port %u and socket %d",
 164                         port, sock->fd);
 165
 166         return sock;
 167
 168 error:
 169         if (sock) {
 170                 lttcomm_destroy_sock(sock);
 171         }
 172         return NULL;
 173 }
 174
 175 /*
 176  * Close and destroy the given TCP socket.
 177  */
 178 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 179 {
 180         int ret;
 181         uint16_t port;
 182
 183         assert(sock);
 184
 185         ret = lttcomm_sock_get_port(sock, &port);
 186         if (ret) {
 187                 ERR("[agent-thread] Failed to get port of agent TCP socket");
 188                 port = 0;
 189         }
 190
 191         DBG3("[agent-thread] Destroy TCP socket on port %" PRIu16,
 192                         port);
 193
 194         /* This will return gracefully if fd is invalid. */
 195         sock->ops->close(sock);
 196         lttcomm_destroy_sock(sock);
 197 }
 198
 199 static const char *domain_type_str(enum lttng_domain_type domain_type)
 200 {
 201         switch (domain_type) {
 202         case LTTNG_DOMAIN_NONE:
 203                 return "none";
 204         case LTTNG_DOMAIN_KERNEL:
 205                 return "kernel";
 206         case LTTNG_DOMAIN_UST:
 207                 return "ust";
 208         case LTTNG_DOMAIN_JUL:
 209                 return "jul";
 210         case LTTNG_DOMAIN_LOG4J:
 211                 return "log4j";
 212         case LTTNG_DOMAIN_PYTHON:
 213                 return "python";
 214         default:
 215                 return "unknown";
 216         }
 217 }
 218
 219 static bool is_agent_protocol_version_supported(
 220                 const struct agent_protocol_version *version)
 221 {
 222         const bool is_supported = version->major == AGENT_MAJOR_VERSION &&
 223                         version->minor == AGENT_MINOR_VERSION;
 224
 225         if (!is_supported) {
 226                 WARN("Refusing agent connection: unsupported protocol version %ui.%ui, expected %i.%i",
 227                                 version->major, version->minor,
 228                                 AGENT_MAJOR_VERSION, AGENT_MINOR_VERSION);
 229         }
 230
 231         return is_supported;
 232 }
 233
 234 /*
 235  * Handle a new agent connection on the registration socket.
 236  *
 237  * Returns 0 on success, or else a negative errno value.
 238  * On success, the resulting socket is returned through `agent_app_socket`
 239  * and the application's reported id is updated through `agent_app_id`.
 240  */
 241 static int accept_agent_connection(
 242                 struct lttcomm_sock *reg_sock,
 243                 struct agent_app_id *agent_app_id,
 244                 struct lttcomm_sock **agent_app_socket)
 245 {
 246         int ret;
 247         struct agent_protocol_version agent_version;
 248         ssize_t size;
 249         struct agent_register_msg msg;
 250         struct lttcomm_sock *new_sock;
 251
 252         assert(reg_sock);
 253
 254         new_sock = reg_sock->ops->accept(reg_sock);
 255         if (!new_sock) {
 256                 ret = -ENOTCONN;
 257                 goto end;
 258         }
 259
 260         size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
 261         if (size < sizeof(msg)) {
 262                 if (size < 0) {
 263                         PERROR("Failed to register new agent application");
 264                 } else if (size != 0) {
 265                         ERR("Failed to register new agent application: invalid registration message length: expected length = %zu, message length = %zd",
 266                                         sizeof(msg), size);
 267                 } else {
 268                         DBG("Failed to register new agent application: connection closed");
 269                 }
 270                 ret = -EINVAL;
 271                 goto error_close_socket;
 272         }
 273
 274         agent_version = (struct agent_protocol_version) {
 275                 be32toh(msg.major_version),
 276                 be32toh(msg.minor_version),
 277         };
 278
 279         /* Test communication protocol version of the registering agent. */
 280         if (!is_agent_protocol_version_supported(&agent_version)) {
 281                 ret = -EINVAL;
 282                 goto error_close_socket;
 283         }
 284
 285         *agent_app_id = (struct agent_app_id) {
 286                 .domain = (enum lttng_domain_type) be32toh(msg.domain),
 287                 .pid = (pid_t) be32toh(msg.pid),
 288         };
 289
 290         DBG2("New registration for agent application: pid = %ld, domain = %s, socket fd = %d",
 291                         (long) agent_app_id->pid,
 292                         domain_type_str(agent_app_id->domain), new_sock->fd);
 293
 294         *agent_app_socket = new_sock;
 295         new_sock = NULL;
 296         ret = 0;
 297         goto end;
 298
 299 error_close_socket:
 300         new_sock->ops->close(new_sock);
 301         lttcomm_destroy_sock(new_sock);
 302 end:
 303         return ret;
 304 }
 305
 306 bool agent_tracing_is_enabled(void)
 307 {
 308         int enabled;
 309
 310         enabled = uatomic_read(&agent_tracing_enabled);
 311         assert(enabled != -1);
 312         return enabled == 1;
 313 }
 314
 315 /*
 316  * Write agent TCP port using the rundir.
 317  */
 318 static int write_agent_port(uint16_t port)
 319 {
 320         return utils_create_pid_file((pid_t) port,
 321                         config.agent_port_file_path.value);
 322 }
 323
 324 static
 325 void mark_thread_as_ready(struct thread_notifiers *notifiers)
 326 {
 327         DBG("Marking agent management thread as ready");
 328         sem_post(&notifiers->ready);
 329 }
 330
 331 static
 332 void wait_until_thread_is_ready(struct thread_notifiers *notifiers)
 333 {
 334         DBG("Waiting for agent management thread to be ready");
 335         sem_wait(&notifiers->ready);
 336         DBG("Agent management thread is ready");
 337 }
 338
 339 /*
 340  * This thread manage application notify communication.
 341  */
 342 static void *thread_agent_management(void *data)
 343 {
 344         int i, ret, pollfd;
 345         uint32_t revents, nb_fd;
 346         struct lttng_poll_event events;
 347         struct lttcomm_sock *reg_sock;
 348         struct thread_notifiers *notifiers = data;
 349         const int quit_pipe_read_fd = lttng_pipe_get_readfd(
 350                         notifiers->quit_pipe);
 351
 352         DBG("[agent-thread] Manage agent application registration.");
 353
 354         rcu_register_thread();
 355         rcu_thread_online();
 356
 357         /* Agent initialization call MUST be called before starting the thread. */
 358         assert(agent_apps_ht_by_sock);
 359
 360         /* Create pollset with size 2, quit pipe and registration socket. */
 361         ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
 362         if (ret < 0) {
 363                 goto error_poll_create;
 364         }
 365
 366         ret = lttng_poll_add(&events, quit_pipe_read_fd,
 367                         LPOLLIN | LPOLLERR);
 368         if (ret < 0) {
 369                 goto error_tcp_socket;
 370         }
 371
 372         reg_sock = init_tcp_socket();
 373         if (reg_sock) {
 374                 uint16_t port;
 375
 376                 assert(lttcomm_sock_get_port(reg_sock, &port) == 0);
 377
 378                 ret = write_agent_port(port);
 379                 if (ret) {
 380                         ERR("[agent-thread] Failed to create agent port file: agent tracing will be unavailable");
 381                         /* Don't prevent the launch of the sessiond on error. */
 382                         mark_thread_as_ready(notifiers);
 383                         goto error;
 384                 }
 385         } else {
 386                 /* Don't prevent the launch of the sessiond on error. */
 387                 mark_thread_as_ready(notifiers);
 388                 goto error_tcp_socket;
 389         }
 390
 391         /*
 392          * Signal that the agent thread is ready. The command thread
 393          * may start to query whether or not agent tracing is enabled.
 394          */
 395         uatomic_set(&agent_tracing_enabled, 1);
 396         mark_thread_as_ready(notifiers);
 397
 398         /* Add TCP socket to the poll set. */
 399         ret = lttng_poll_add(&events, reg_sock->fd,
 400                         LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 401         if (ret < 0) {
 402                 goto error;
 403         }
 404
 405         while (1) {
 406                 DBG3("[agent-thread] Manage agent polling");
 407
 408                 /* Inifinite blocking call, waiting for transmission */
 409 restart:
 410                 ret = lttng_poll_wait(&events, -1);
 411                 DBG3("[agent-thread] Manage agent return from poll on %d fds",
 412                                 LTTNG_POLL_GETNB(&events));
 413                 if (ret < 0) {
 414                         /*
 415                          * Restart interrupted system call.
 416                          */
 417                         if (errno == EINTR) {
 418                                 goto restart;
 419                         }
 420                         goto error;
 421                 }
 422                 nb_fd = ret;
 423                 DBG3("[agent-thread] %d fd ready", nb_fd);
 424
 425                 for (i = 0; i < nb_fd; i++) {
 426                         /* Fetch once the poll data */
 427                         revents = LTTNG_POLL_GETEV(&events, i);
 428                         pollfd = LTTNG_POLL_GETFD(&events, i);
 429
 430                         /* Thread quit pipe has been closed. Killing thread. */
 431                         if (pollfd == quit_pipe_read_fd) {
 432                                 goto exit;
 433                         }
 434
 435                         /* Activity on the registration socket. */
 436                         if (revents & LPOLLIN) {
 437                                 struct agent_app_id new_app_id;
 438                                 struct agent_app *new_app = NULL;
 439                                 struct lttcomm_sock *new_app_socket;
 440                                 int new_app_socket_fd;
 441
 442                                 assert(pollfd == reg_sock->fd);
 443
 444                                 ret = accept_agent_connection(
 445                                         reg_sock, &new_app_id, &new_app_socket);
 446                                 if (ret < 0) {
 447                                         /* Errors are already logged. */
 448                                         continue;
 449                                 }
 450
 451                                 /*
 452                                  * new_app_socket's ownership has been
 453                                  * transferred to the new agent app.
 454                                  */
 455                                 new_app = agent_create_app(new_app_id.pid,
 456                                                 new_app_id.domain,
 457                                                 new_app_socket);
 458                                 if (!new_app) {
 459                                         new_app_socket->ops->close(
 460                                                         new_app_socket);
 461                                         continue;
 462                                 }
 463                                 new_app_socket_fd = new_app_socket->fd;
 464                                 new_app_socket = NULL;
 465
 466                                 /*
 467                                  * Since this is a command socket (write then
 468                                  * read), only add poll error event to only
 469                                  * detect shutdown.
 470                                  */
 471                                 ret = lttng_poll_add(&events, new_app_socket_fd,
 472                                                 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 473                                 if (ret < 0) {
 474                                         agent_destroy_app(new_app);
 475                                         continue;
 476                                 }
 477
 478                                 /*
 479                                  * Prevent sessions from being modified while
 480                                  * the agent application's configuration is
 481                                  * updated.
 482                                  */
 483                                 session_lock_list();
 484
 485                                 /*
 486                                  * Update the newly registered applications's
 487                                  * configuration.
 488                                  */
 489                                 update_agent_app(new_app);
 490
 491                                 ret = agent_send_registration_done(new_app);
 492                                 if (ret < 0) {
 493                                         agent_destroy_app(new_app);
 494                                         /* Removing from the poll set. */
 495                                         ret = lttng_poll_del(&events,
 496                                                         new_app_socket_fd);
 497                                         if (ret < 0) {
 498                                                 session_unlock_list();
 499                                                 goto error;
 500                                         }
 501                                         continue;
 502                                 }
 503
 504                                 /* Publish the new agent app. */
 505                                 agent_add_app(new_app);
 506
 507                                 session_unlock_list();
 508                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 509                                 /* Removing from the poll set */
 510                                 ret = lttng_poll_del(&events, pollfd);
 511                                 if (ret < 0) {
 512                                         goto error;
 513                                 }
 514                                 agent_destroy_app_by_sock(pollfd);
 515                         } else {
 516                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
 517                                 goto error;
 518                         }
 519                 }
 520         }
 521
 522 exit:
 523         /* Whatever happens, try to delete it and exit. */
 524         (void) lttng_poll_del(&events, reg_sock->fd);
 525 error:
 526         destroy_tcp_socket(reg_sock);
 527 error_tcp_socket:
 528         lttng_poll_clean(&events);
 529 error_poll_create:
 530         uatomic_set(&agent_tracing_enabled, 0);
 531         DBG("[agent-thread] Cleaning up and stopping.");
 532         rcu_thread_offline();
 533         rcu_unregister_thread();
 534         return NULL;
 535 }
 536
 537 static bool shutdown_agent_management_thread(void *data)
 538 {
 539         struct thread_notifiers *notifiers = data;
 540         const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe);
 541
 542         return notify_thread_pipe(write_fd) == 1;
 543 }
 544
 545 static void cleanup_agent_management_thread(void *data)
 546 {
 547         struct thread_notifiers *notifiers = data;
 548
 549         lttng_pipe_destroy(notifiers->quit_pipe);
 550         sem_destroy(&notifiers->ready);
 551         free(notifiers);
 552 }
 553
 554 bool launch_agent_management_thread(void)
 555 {
 556         struct thread_notifiers *notifiers;
 557         struct lttng_thread *thread;
 558
 559         notifiers = zmalloc(sizeof(*notifiers));
 560         if (!notifiers) {
 561                 goto error_alloc;
 562         }
 563
 564         sem_init(&notifiers->ready, 0, 0);
 565         notifiers->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
 566         if (!notifiers->quit_pipe) {
 567                 goto error;
 568         }
 569         thread = lttng_thread_create("Agent management",
 570                         thread_agent_management,
 571                         shutdown_agent_management_thread,
 572                         cleanup_agent_management_thread,
 573                         notifiers);
 574         if (!thread) {
 575                 goto error;
 576         }
 577         wait_until_thread_is_ready(notifiers);
 578         lttng_thread_put(thread);
 579         return true;
 580 error:
 581         cleanup_agent_management_thread(notifiers);
 582 error_alloc:
 583         return false;
 584 }