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 _GNU_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 "lttng-sessiond.h"
  31 #include "session.h"
  32 #include "utils.h"
  33
  34 /*
  35  * Note that there is not port here. It's set after this URI is parsed so we
  36  * can let the user define a custom one. However, localhost is ALWAYS the
  37  * default listening address.
  38  */
  39 static const char *default_reg_uri =
  40         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS;
  41
  42 /*
  43  * Update agent application using the given socket. This is done just after
  44  * registration was successful.
  45  *
  46  * This is a quite heavy call in terms of locking since the session list lock
  47  * AND session lock are acquired.
  48  */
  49 static void update_agent_app(struct agent_app *app)
  50 {
  51         struct ltt_session *session, *stmp;
  52         struct ltt_session_list *list;
  53
  54         list = session_get_list();
  55         assert(list);
  56
  57         session_lock_list();
  58         cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
  59                 session_lock(session);
  60                 if (session->ust_session) {
  61                         struct agent *agt;
  62
  63                         rcu_read_lock();
  64                         agt = trace_ust_find_agent(session->ust_session, app->domain);
  65                         if (agt) {
  66                                 agent_update(agt, app->sock->fd);
  67                         }
  68                         rcu_read_unlock();
  69                 }
  70                 session_unlock(session);
  71         }
  72         session_unlock_list();
  73 }
  74
  75 /*
  76  * Destroy a agent application by socket.
  77  */
  78 static void destroy_agent_app(int sock)
  79 {
  80         struct agent_app *app;
  81
  82         assert(sock >= 0);
  83
  84         /*
  85          * Not finding an application is a very important error that should NEVER
  86          * happen. The hash table deletion is ONLY done through this call even on
  87          * thread cleanup.
  88          */
  89         rcu_read_lock();
  90         app = agent_find_app_by_sock(sock);
  91         assert(app);
  92
  93         /* RCU read side lock is assumed to be held by this function. */
  94         agent_delete_app(app);
  95
  96         /* The application is freed in a RCU call but the socket is closed here. */
  97         agent_destroy_app(app);
  98         rcu_read_unlock();
  99 }
 100
 101 /*
 102  * Cleanup remaining agent apps in the hash table. This should only be called in
 103  * the exit path of the thread.
 104  */
 105 static void clean_agent_apps_ht(void)
 106 {
 107         struct lttng_ht_node_ulong *node;
 108         struct lttng_ht_iter iter;
 109
 110         DBG3("[agent-thread] Cleaning agent apps ht");
 111
 112         rcu_read_lock();
 113         cds_lfht_for_each_entry(agent_apps_ht_by_sock->ht, &iter.iter, node, node) {
 114                 struct agent_app *app;
 115
 116                 app = caa_container_of(node, struct agent_app, node);
 117                 destroy_agent_app(app->sock->fd);
 118         }
 119         rcu_read_unlock();
 120 }
 121
 122 /*
 123  * Create and init socket from uri.
 124  */
 125 static struct lttcomm_sock *init_tcp_socket(void)
 126 {
 127         int ret;
 128         struct lttng_uri *uri = NULL;
 129         struct lttcomm_sock *sock = NULL;
 130
 131         /*
 132          * This should never fail since the URI is hardcoded and the port is set
 133          * before this thread is launched.
 134          */
 135         ret = uri_parse(default_reg_uri, &uri);
 136         assert(ret);
 137         assert(agent_tcp_port);
 138         uri->port = agent_tcp_port;
 139
 140         sock = lttcomm_alloc_sock_from_uri(uri);
 141         uri_free(uri);
 142         if (sock == NULL) {
 143                 ERR("[agent-thread] agent allocating TCP socket");
 144                 goto error;
 145         }
 146
 147         ret = lttcomm_create_sock(sock);
 148         if (ret < 0) {
 149                 goto error;
 150         }
 151
 152         ret = sock->ops->bind(sock);
 153         if (ret < 0) {
 154                 WARN("Another session daemon is using this agent port. Agent support "
 155                                 "will be deactivated to prevent interfering with the tracing.");
 156                 goto error;
 157         }
 158
 159         ret = sock->ops->listen(sock, -1);
 160         if (ret < 0) {
 161                 goto error;
 162         }
 163
 164         DBG("[agent-thread] Listening on TCP port %u and socket %d",
 165                         agent_tcp_port, sock->fd);
 166
 167         return sock;
 168
 169 error:
 170         if (sock) {
 171                 lttcomm_destroy_sock(sock);
 172         }
 173         return NULL;
 174 }
 175
 176 /*
 177  * Close and destroy the given TCP socket.
 178  */
 179 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 180 {
 181         assert(sock);
 182
 183         DBG3("[agent-thread] Destroy TCP socket on port %u", agent_tcp_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 /*
 273  * This thread manage application notify communication.
 274  */
 275 void *agent_thread_manage_registration(void *data)
 276 {
 277         int i, ret, pollfd;
 278         uint32_t revents, nb_fd;
 279         struct lttng_poll_event events;
 280         struct lttcomm_sock *reg_sock;
 281
 282         DBG("[agent-thread] Manage agent application registration.");
 283
 284         rcu_register_thread();
 285         rcu_thread_online();
 286
 287         /* Agent initialization call MUST be called before starting the thread. */
 288         assert(agent_apps_ht_by_sock);
 289
 290         /* Create pollset with size 2, quit pipe and socket. */
 291         ret = sessiond_set_thread_pollset(&events, 2);
 292         if (ret < 0) {
 293                 goto error_poll_create;
 294         }
 295
 296         reg_sock = init_tcp_socket();
 297         if (!reg_sock) {
 298                 goto error_tcp_socket;
 299         }
 300
 301         /* Add create valid TCP socket to poll set. */
 302         ret = lttng_poll_add(&events, reg_sock->fd,
 303                         LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 304         if (ret < 0) {
 305                 goto error;
 306         }
 307
 308         while (1) {
 309                 DBG3("[agent-thread] Manage agent polling");
 310
 311                 /* Inifinite blocking call, waiting for transmission */
 312 restart:
 313                 ret = lttng_poll_wait(&events, -1);
 314                 DBG3("[agent-thread] Manage agent return from poll on %d fds",
 315                                 LTTNG_POLL_GETNB(&events));
 316                 if (ret < 0) {
 317                         /*
 318                          * Restart interrupted system call.
 319                          */
 320                         if (errno == EINTR) {
 321                                 goto restart;
 322                         }
 323                         goto error;
 324                 }
 325                 nb_fd = ret;
 326                 DBG3("[agent-thread] %d fd ready", nb_fd);
 327
 328                 for (i = 0; i < nb_fd; i++) {
 329                         /* Fetch once the poll data */
 330                         revents = LTTNG_POLL_GETEV(&events, i);
 331                         pollfd = LTTNG_POLL_GETFD(&events, i);
 332
 333                         if (!revents) {
 334                                 /* No activity for this FD (poll implementation). */
 335                                 continue;
 336                         }
 337
 338                         /* Thread quit pipe has been closed. Killing thread. */
 339                         ret = sessiond_check_thread_quit_pipe(pollfd, revents);
 340                         if (ret) {
 341                                 goto exit;
 342                         }
 343
 344                         /*
 345                          * Check first if this is a POLLERR since POLLIN is also included
 346                          * in an error value thus checking first.
 347                          */
 348                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 349                                 /* Removing from the poll set */
 350                                 ret = lttng_poll_del(&events, pollfd);
 351                                 if (ret < 0) {
 352                                         goto error;
 353                                 }
 354
 355                                 destroy_agent_app(pollfd);
 356                         } else if (revents & (LPOLLIN)) {
 357                                 int new_fd;
 358                                 struct agent_app *app = NULL;
 359
 360                                 /* Pollin event of agent app socket should NEVER happen. */
 361                                 assert(pollfd == reg_sock->fd);
 362
 363                                 new_fd = handle_registration(reg_sock, &app);
 364                                 if (new_fd < 0) {
 365                                         WARN("[agent-thread] agent registration failed. Ignoring.");
 366                                         /* Somehow the communication failed. Just continue. */
 367                                         continue;
 368                                 }
 369                                 /* Should not have a NULL app on success. */
 370                                 assert(app);
 371
 372                                 /* Only add poll error event to only detect shutdown. */
 373                                 ret = lttng_poll_add(&events, new_fd,
 374                                                 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 375                                 if (ret < 0) {
 376                                         destroy_agent_app(new_fd);
 377                                         continue;
 378                                 }
 379
 380                                 /* Update newly registered app. */
 381                                 update_agent_app(app);
 382
 383                                 /* On failure, the poll will detect it and clean it up. */
 384                                 (void) agent_send_registration_done(app);
 385                         } else {
 386                                 ERR("Unknown poll events %u for sock %d", revents, pollfd);
 387                                 continue;
 388                         }
 389                 }
 390         }
 391
 392 exit:
 393         /* Whatever happens, try to delete it and exit. */
 394         (void) lttng_poll_del(&events, reg_sock->fd);
 395 error:
 396         destroy_tcp_socket(reg_sock);
 397 error_tcp_socket:
 398         lttng_poll_clean(&events);
 399 error_poll_create:
 400         DBG("[agent-thread] is cleaning up and stopping.");
 401
 402         if (agent_apps_ht_by_sock) {
 403                 clean_agent_apps_ht();
 404                 lttng_ht_destroy(agent_apps_ht_by_sock);
 405         }
 406
 407         rcu_thread_offline();
 408         rcu_unregister_thread();
 409         return NULL;
 410 }