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

   1 /*
   2  * Copyright (C) 2013 - David Goulet <dgoulet@efficios.com>
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License, version 2 only, as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 51
  15  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _LGPL_SOURCE
  19 #include <assert.h>
  20
  21 #include <common/common.h>
  22 #include <common/sessiond-comm/sessiond-comm.h>
  23 #include <common/uri.h>
  24 #include <common/utils.h>
  25
  26 #include <common/compat/endian.h>
  27
  28 #include "fd-limit.h"
  29 #include "agent-thread.h"
  30 #include "agent.h"
  31 #include "lttng-sessiond.h"
  32 #include "session.h"
  33 #include "utils.h"
  34
  35 static int agent_tracing_enabled = -1;
  36
  37 /*
  38  * Note that there is not port here. It's set after this URI is parsed so we
  39  * can let the user define a custom one. However, localhost is ALWAYS the
  40  * default listening address.
  41  */
  42 static const char *default_reg_uri =
  43         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS;
  44
  45 /*
  46  * Update agent application using the given socket. This is done just after
  47  * registration was successful.
  48  *
  49  * This is a quite heavy call in terms of locking since the session list lock
  50  * AND session lock are acquired.
  51  */
  52 static void update_agent_app(struct agent_app *app)
  53 {
  54         struct ltt_session *session, *stmp;
  55         struct ltt_session_list *list;
  56
  57         list = session_get_list();
  58         assert(list);
  59
  60         session_lock_list();
  61         cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
  62                 session_lock(session);
  63                 if (session->ust_session) {
  64                         struct agent *agt;
  65
  66                         rcu_read_lock();
  67                         agt = trace_ust_find_agent(session->ust_session, app->domain);
  68                         if (agt) {
  69                                 agent_update(agt, app->sock->fd);
  70                         }
  71                         rcu_read_unlock();
  72                 }
  73                 session_unlock(session);
  74         }
  75         session_unlock_list();
  76 }
  77
  78 /*
  79  * Create and init socket from uri.
  80  */
  81 static struct lttcomm_sock *init_tcp_socket(void)
  82 {
  83         int ret;
  84         struct lttng_uri *uri = NULL;
  85         struct lttcomm_sock *sock = NULL;
  86         unsigned int port;
  87         bool bind_succeeded = false;
  88
  89         /*
  90          * This should never fail since the URI is hardcoded and the port is set
  91          * before this thread is launched.
  92          */
  93         ret = uri_parse(default_reg_uri, &uri);
  94         assert(ret);
  95         assert(config.agent_tcp_port.begin > 0);
  96         uri->port = config.agent_tcp_port.begin;
  97
  98         sock = lttcomm_alloc_sock_from_uri(uri);
  99         uri_free(uri);
 100         if (sock == NULL) {
 101                 ERR("[agent-thread] agent allocating TCP socket");
 102                 goto error;
 103         }
 104
 105         ret = lttcomm_create_sock(sock);
 106         if (ret < 0) {
 107                 goto error;
 108         }
 109
 110         for (port = config.agent_tcp_port.begin;
 111                         port <= config.agent_tcp_port.end; port++) {
 112                 ret = lttcomm_sock_set_port(sock, (uint16_t) port);
 113                 if (ret) {
 114                         ERR("[agent-thread] Failed to set port %u on socket",
 115                                         port);
 116                         goto error;
 117                 }
 118                 DBG3("[agent-thread] Trying to bind on port %u", port);
 119                 ret = sock->ops->bind(sock);
 120                 if (!ret) {
 121                         bind_succeeded = true;
 122                         break;
 123                 }
 124
 125                 if (errno == EADDRINUSE) {
 126                         DBG("Failed to bind to port %u since it is already in use",
 127                                         port);
 128                 } else {
 129                         PERROR("Failed to bind to port %u", port);
 130                         goto error;
 131                 }
 132         }
 133
 134         if (!bind_succeeded) {
 135                 if (config.agent_tcp_port.begin == config.agent_tcp_port.end) {
 136                         WARN("Another process is already using the agent port %i. "
 137                                         "Agent support will be deactivated.",
 138                                         config.agent_tcp_port.begin);
 139                         goto error;
 140                 } else {
 141                         WARN("All ports in the range [%i, %i] are already in use. "
 142                                         "Agent support will be deactivated.",
 143                                         config.agent_tcp_port.begin,
 144                                         config.agent_tcp_port.end);
 145                         goto error;
 146                 }
 147         }
 148
 149         ret = sock->ops->listen(sock, -1);
 150         if (ret < 0) {
 151                 goto error;
 152         }
 153
 154         DBG("[agent-thread] Listening on TCP port %u and socket %d",
 155                         port, sock->fd);
 156
 157         return sock;
 158
 159 error:
 160         if (sock) {
 161                 lttcomm_destroy_sock(sock);
 162         }
 163         return NULL;
 164 }
 165
 166 /*
 167  * Close and destroy the given TCP socket.
 168  */
 169 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 170 {
 171         int ret;
 172         uint16_t port;
 173
 174         assert(sock);
 175
 176         ret = lttcomm_sock_get_port(sock, &port);
 177         if (ret) {
 178                 ERR("[agent-thread] Failed to get port of agent TCP socket");
 179                 port = 0;
 180         }
 181
 182         DBG3("[agent-thread] Destroy TCP socket on port %" PRIu16,
 183                         port);
 184
 185         /* This will return gracefully if fd is invalid. */
 186         sock->ops->close(sock);
 187         lttcomm_destroy_sock(sock);
 188 }
 189
 190 /*
 191  * Handle a new agent registration using the reg socket. After that, a new
 192  * agent application is added to the global hash table and attach to an UST app
 193  * object. If r_app is not NULL, the created app is set to the pointer.
 194  *
 195  * Return the new FD created upon accept() on success or else a negative errno
 196  * value.
 197  */
 198 static int handle_registration(struct lttcomm_sock *reg_sock,
 199                 struct agent_app **r_app)
 200 {
 201         int ret;
 202         pid_t pid;
 203         uint32_t major_version, minor_version;
 204         ssize_t size;
 205         enum lttng_domain_type domain;
 206         struct agent_app *app;
 207         struct agent_register_msg msg;
 208         struct lttcomm_sock *new_sock;
 209
 210         assert(reg_sock);
 211
 212         new_sock = reg_sock->ops->accept(reg_sock);
 213         if (!new_sock) {
 214                 ret = -ENOTCONN;
 215                 goto error;
 216         }
 217
 218         size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
 219         if (size < sizeof(msg)) {
 220                 ret = -EINVAL;
 221                 goto error_socket;
 222         }
 223         domain = be32toh(msg.domain);
 224         pid = be32toh(msg.pid);
 225         major_version = be32toh(msg.major_version);
 226         minor_version = be32toh(msg.minor_version);
 227
 228         /* Test communication protocol version of the registring agent. */
 229         if (major_version != AGENT_MAJOR_VERSION) {
 230                 ret = -EINVAL;
 231                 goto error_socket;
 232         }
 233         if (minor_version != AGENT_MINOR_VERSION) {
 234                 ret = -EINVAL;
 235                 goto error_socket;
 236         }
 237
 238         DBG2("[agent-thread] New registration for pid %d domain %d on socket %d",
 239                         pid, domain, new_sock->fd);
 240
 241         app = agent_create_app(pid, domain, new_sock);
 242         if (!app) {
 243                 ret = -ENOMEM;
 244                 goto error_socket;
 245         }
 246
 247         /*
 248          * Add before assigning the socket value to the UST app so it can be found
 249          * concurrently.
 250          */
 251         agent_add_app(app);
 252
 253         /*
 254          * We don't need to attach the agent app to the app. If we ever do so, we
 255          * should consider both registration order of agent before app and app
 256          * before agent.
 257          */
 258
 259         if (r_app) {
 260                 *r_app = app;
 261         }
 262
 263         return new_sock->fd;
 264
 265 error_socket:
 266         new_sock->ops->close(new_sock);
 267         lttcomm_destroy_sock(new_sock);
 268 error:
 269         return ret;
 270 }
 271
 272 bool agent_tracing_is_enabled(void)
 273 {
 274         int enabled;
 275
 276         enabled = uatomic_read(&agent_tracing_enabled);
 277         assert(enabled != -1);
 278         return enabled == 1;
 279 }
 280
 281 /*
 282  * Write agent TCP port using the rundir.
 283  */
 284 static int write_agent_port(uint16_t port)
 285 {
 286         return utils_create_pid_file((pid_t) port,
 287                         config.agent_port_file_path.value);
 288 }
 289
 290 /*
 291  * This thread manage application notify communication.
 292  */
 293 void *agent_thread_manage_registration(void *data)
 294 {
 295         int i, ret, pollfd;
 296         uint32_t revents, nb_fd;
 297         struct lttng_poll_event events;
 298         struct lttcomm_sock *reg_sock;
 299
 300         DBG("[agent-thread] Manage agent application registration.");
 301
 302         rcu_register_thread();
 303         rcu_thread_online();
 304
 305         /* Agent initialization call MUST be called before starting the thread. */
 306         assert(agent_apps_ht_by_sock);
 307
 308         /* Create pollset with size 2, quit pipe and socket. */
 309         ret = sessiond_set_thread_pollset(&events, 2);
 310         if (ret < 0) {
 311                 sessiond_notify_ready();
 312                 goto error_poll_create;
 313         }
 314
 315         reg_sock = init_tcp_socket();
 316         if (reg_sock) {
 317                 uint16_t port;
 318
 319                 assert(lttcomm_sock_get_port(reg_sock, &port) == 0);
 320
 321                 ret = write_agent_port(port);
 322                 if (ret) {
 323                         ERR("[agent-thread] Failed to create agent port file: agent tracing will be unavailable");
 324                         /* Don't prevent the launch of the sessiond on error. */
 325                         sessiond_notify_ready();
 326                         goto error;
 327                 }
 328         } else {
 329                 /* Don't prevent the launch of the sessiond on error. */
 330                 sessiond_notify_ready();
 331                 goto error_tcp_socket;
 332         }
 333
 334         /*
 335          * Signal that the agent thread is ready. The command thread
 336          * may start to query whether or not agent tracing is enabled.
 337          */
 338         uatomic_set(&agent_tracing_enabled, 1);
 339         sessiond_notify_ready();
 340
 341         /* Add TCP socket to poll set. */
 342         ret = lttng_poll_add(&events, reg_sock->fd,
 343                         LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 344         if (ret < 0) {
 345                 goto error;
 346         }
 347
 348         while (1) {
 349                 DBG3("[agent-thread] Manage agent polling");
 350
 351                 /* Inifinite blocking call, waiting for transmission */
 352 restart:
 353                 ret = lttng_poll_wait(&events, -1);
 354                 DBG3("[agent-thread] Manage agent return from poll on %d fds",
 355                                 LTTNG_POLL_GETNB(&events));
 356                 if (ret < 0) {
 357                         /*
 358                          * Restart interrupted system call.
 359                          */
 360                         if (errno == EINTR) {
 361                                 goto restart;
 362                         }
 363                         goto error;
 364                 }
 365                 nb_fd = ret;
 366                 DBG3("[agent-thread] %d fd ready", nb_fd);
 367
 368                 for (i = 0; i < nb_fd; i++) {
 369                         /* Fetch once the poll data */
 370                         revents = LTTNG_POLL_GETEV(&events, i);
 371                         pollfd = LTTNG_POLL_GETFD(&events, i);
 372
 373                         if (!revents) {
 374                                 /* No activity for this FD (poll implementation). */
 375                                 continue;
 376                         }
 377
 378                         /* Thread quit pipe has been closed. Killing thread. */
 379                         ret = sessiond_check_thread_quit_pipe(pollfd, revents);
 380                         if (ret) {
 381                                 goto exit;
 382                         }
 383
 384                         if (revents & LPOLLIN) {
 385                                 int new_fd;
 386                                 struct agent_app *app = NULL;
 387
 388                                 assert(pollfd == reg_sock->fd);
 389                                 new_fd = handle_registration(reg_sock, &app);
 390                                 if (new_fd < 0) {
 391                                         continue;
 392                                 }
 393                                 /* Should not have a NULL app on success. */
 394                                 assert(app);
 395
 396                                 /*
 397                                  * Since this is a command socket (write then read),
 398                                  * only add poll error event to only detect shutdown.
 399                                  */
 400                                 ret = lttng_poll_add(&events, new_fd,
 401                                                 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 402                                 if (ret < 0) {
 403                                         agent_destroy_app_by_sock(new_fd);
 404                                         continue;
 405                                 }
 406
 407                                 /* Update newly registered app. */
 408                                 update_agent_app(app);
 409
 410                                 /* On failure, the poll will detect it and clean it up. */
 411                                 ret = agent_send_registration_done(app);
 412                                 if (ret < 0) {
 413                                         /* Removing from the poll set */
 414                                         ret = lttng_poll_del(&events, new_fd);
 415                                         if (ret < 0) {
 416                                                 goto error;
 417                                         }
 418                                         agent_destroy_app_by_sock(new_fd);
 419                                         continue;
 420                                 }
 421                         } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 422                                 /* Removing from the poll set */
 423                                 ret = lttng_poll_del(&events, pollfd);
 424                                 if (ret < 0) {
 425                                         goto error;
 426                                 }
 427                                 agent_destroy_app_by_sock(pollfd);
 428                         } else {
 429                                 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
 430                                 goto error;
 431                         }
 432                 }
 433         }
 434
 435 exit:
 436         /* Whatever happens, try to delete it and exit. */
 437         (void) lttng_poll_del(&events, reg_sock->fd);
 438 error:
 439         destroy_tcp_socket(reg_sock);
 440 error_tcp_socket:
 441         lttng_poll_clean(&events);
 442 error_poll_create:
 443         uatomic_set(&agent_tracing_enabled, 0);
 444         DBG("[agent-thread] is cleaning up and stopping.");
 445
 446         rcu_thread_offline();
 447         rcu_unregister_thread();
 448         return NULL;
 449 }