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 #define _LGPL_SOURCE
  20 #include <assert.h>
  21
  22 #include <common/common.h>
  23 #include <common/sessiond-comm/sessiond-comm.h>
  24 #include <common/uri.h>
  25 #include <common/utils.h>
  26
  27 #include <common/compat/endian.h>
  28
  29 #include "fd-limit.h"
  30 #include "agent-thread.h"
  31 #include "lttng-sessiond.h"
  32 #include "session.h"
  33 #include "utils.h"
  34
  35 /*
  36  * Note that there is not port here. It's set after this URI is parsed so we
  37  * can let the user define a custom one. However, localhost is ALWAYS the
  38  * default listening address.
  39  */
  40 static const char *default_reg_uri =
  41         "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS;
  42
  43 /*
  44  * Update agent application using the given socket. This is done just after
  45  * registration was successful.
  46  *
  47  * This is a quite heavy call in terms of locking since the session list lock
  48  * AND session lock are acquired.
  49  */
  50 static void update_agent_app(struct agent_app *app)
  51 {
  52         struct ltt_session *session, *stmp;
  53         struct ltt_session_list *list;
  54
  55         list = session_get_list();
  56         assert(list);
  57
  58         session_lock_list();
  59         cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
  60                 session_lock(session);
  61                 if (session->ust_session) {
  62                         struct agent *agt;
  63
  64                         rcu_read_lock();
  65                         agt = trace_ust_find_agent(session->ust_session, app->domain);
  66                         if (agt) {
  67                                 agent_update(agt, app->sock->fd);
  68                         }
  69                         rcu_read_unlock();
  70                 }
  71                 session_unlock(session);
  72         }
  73         session_unlock_list();
  74 }
  75
  76 /*
  77  * Destroy a agent application by socket.
  78  */
  79 static void destroy_agent_app(int sock)
  80 {
  81         struct agent_app *app;
  82
  83         assert(sock >= 0);
  84
  85         /*
  86          * Not finding an application is a very important error that should NEVER
  87          * happen. The hash table deletion is ONLY done through this call even on
  88          * thread cleanup.
  89          */
  90         rcu_read_lock();
  91         app = agent_find_app_by_sock(sock);
  92         assert(app);
  93
  94         /* RCU read side lock is assumed to be held by this function. */
  95         agent_delete_app(app);
  96
  97         /* The application is freed in a RCU call but the socket is closed here. */
  98         agent_destroy_app(app);
  99         rcu_read_unlock();
 100 }
 101
 102 /*
 103  * Cleanup remaining agent apps in the hash table. This should only be called in
 104  * the exit path of the thread.
 105  */
 106 static void clean_agent_apps_ht(void)
 107 {
 108         struct lttng_ht_node_ulong *node;
 109         struct lttng_ht_iter iter;
 110
 111         DBG3("[agent-thread] Cleaning agent apps ht");
 112
 113         rcu_read_lock();
 114         cds_lfht_for_each_entry(agent_apps_ht_by_sock->ht, &iter.iter, node, node) {
 115                 struct agent_app *app;
 116
 117                 app = caa_container_of(node, struct agent_app, node);
 118                 destroy_agent_app(app->sock->fd);
 119         }
 120         rcu_read_unlock();
 121 }
 122
 123 /*
 124  * Create and init socket from uri.
 125  */
 126 static struct lttcomm_sock *init_tcp_socket(void)
 127 {
 128         int ret;
 129         struct lttng_uri *uri = NULL;
 130         struct lttcomm_sock *sock = NULL;
 131
 132         /*
 133          * This should never fail since the URI is hardcoded and the port is set
 134          * before this thread is launched.
 135          */
 136         ret = uri_parse(default_reg_uri, &uri);
 137         assert(ret);
 138         assert(agent_tcp_port);
 139         uri->port = agent_tcp_port;
 140
 141         sock = lttcomm_alloc_sock_from_uri(uri);
 142         uri_free(uri);
 143         if (sock == NULL) {
 144                 ERR("[agent-thread] agent allocating TCP socket");
 145                 goto error;
 146         }
 147
 148         ret = lttcomm_create_sock(sock);
 149         if (ret < 0) {
 150                 goto error;
 151         }
 152
 153         ret = sock->ops->bind(sock);
 154         if (ret < 0) {
 155                 WARN("Another session daemon is using this agent port. Agent support "
 156                                 "will be deactivated to prevent interfering with the tracing.");
 157                 goto error;
 158         }
 159
 160         ret = sock->ops->listen(sock, -1);
 161         if (ret < 0) {
 162                 goto error;
 163         }
 164
 165         DBG("[agent-thread] Listening on TCP port %u and socket %d",
 166                         agent_tcp_port, sock->fd);
 167
 168         return sock;
 169
 170 error:
 171         if (sock) {
 172                 lttcomm_destroy_sock(sock);
 173         }
 174         return NULL;
 175 }
 176
 177 /*
 178  * Close and destroy the given TCP socket.
 179  */
 180 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 181 {
 182         assert(sock);
 183
 184         DBG3("[agent-thread] Destroy TCP socket on port %u", agent_tcp_port);
 185
 186         /* This will return gracefully if fd is invalid. */
 187         sock->ops->close(sock);
 188         lttcomm_destroy_sock(sock);
 189 }
 190
 191 /*
 192  * Handle a new agent registration using the reg socket. After that, a new
 193  * agent application is added to the global hash table and attach to an UST app
 194  * object. If r_app is not NULL, the created app is set to the pointer.
 195  *
 196  * Return the new FD created upon accept() on success or else a negative errno
 197  * value.
 198  */
 199 static int handle_registration(struct lttcomm_sock *reg_sock,
 200                 struct agent_app **r_app)
 201 {
 202         int ret;
 203         pid_t pid;
 204         uint32_t major_version, minor_version;
 205         ssize_t size;
 206         enum lttng_domain_type domain;
 207         struct agent_app *app;
 208         struct agent_register_msg msg;
 209         struct lttcomm_sock *new_sock;
 210
 211         assert(reg_sock);
 212
 213         new_sock = reg_sock->ops->accept(reg_sock);
 214         if (!new_sock) {
 215                 ret = -ENOTCONN;
 216                 goto error;
 217         }
 218
 219         size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
 220         if (size < sizeof(msg)) {
 221                 ret = -EINVAL;
 222                 goto error_socket;
 223         }
 224         domain = be32toh(msg.domain);
 225         pid = be32toh(msg.pid);
 226         major_version = be32toh(msg.major_version);
 227         minor_version = be32toh(msg.minor_version);
 228
 229         /* Test communication protocol version of the registring agent. */
 230         if (major_version != AGENT_MAJOR_VERSION) {
 231                 ret = -EINVAL;
 232                 goto error_socket;
 233         }
 234         if (minor_version != AGENT_MINOR_VERSION) {
 235                 ret = -EINVAL;
 236                 goto error_socket;
 237         }
 238
 239         DBG2("[agent-thread] New registration for pid %d domain %d on socket %d",
 240                         pid, domain, new_sock->fd);
 241
 242         app = agent_create_app(pid, domain, new_sock);
 243         if (!app) {
 244                 ret = -ENOMEM;
 245                 goto error_socket;
 246         }
 247
 248         /*
 249          * Add before assigning the socket value to the UST app so it can be found
 250          * concurrently.
 251          */
 252         agent_add_app(app);
 253
 254         /*
 255          * We don't need to attach the agent app to the app. If we ever do so, we
 256          * should consider both registration order of agent before app and app
 257          * before agent.
 258          */
 259
 260         if (r_app) {
 261                 *r_app = app;
 262         }
 263
 264         return new_sock->fd;
 265
 266 error_socket:
 267         new_sock->ops->close(new_sock);
 268         lttcomm_destroy_sock(new_sock);
 269 error:
 270         return ret;
 271 }
 272
 273 /*
 274  * This thread manage application notify communication.
 275  */
 276 void *agent_thread_manage_registration(void *data)
 277 {
 278         int i, ret, pollfd;
 279         uint32_t revents, nb_fd;
 280         struct lttng_poll_event events;
 281         struct lttcomm_sock *reg_sock;
 282
 283         DBG("[agent-thread] Manage agent application registration.");
 284
 285         rcu_register_thread();
 286         rcu_thread_online();
 287
 288         /* Agent initialization call MUST be called before starting the thread. */
 289         assert(agent_apps_ht_by_sock);
 290
 291         /* Create pollset with size 2, quit pipe and socket. */
 292         ret = sessiond_set_thread_pollset(&events, 2);
 293         if (ret < 0) {
 294                 goto error_poll_create;
 295         }
 296
 297         reg_sock = init_tcp_socket();
 298         if (!reg_sock) {
 299                 goto error_tcp_socket;
 300         }
 301
 302         /* Add create valid TCP socket to poll set. */
 303         ret = lttng_poll_add(&events, reg_sock->fd,
 304                         LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 305         if (ret < 0) {
 306                 goto error;
 307         }
 308
 309         while (1) {
 310                 DBG3("[agent-thread] Manage agent polling");
 311
 312                 /* Inifinite blocking call, waiting for transmission */
 313 restart:
 314                 ret = lttng_poll_wait(&events, -1);
 315                 DBG3("[agent-thread] Manage agent return from poll on %d fds",
 316                                 LTTNG_POLL_GETNB(&events));
 317                 if (ret < 0) {
 318                         /*
 319                          * Restart interrupted system call.
 320                          */
 321                         if (errno == EINTR) {
 322                                 goto restart;
 323                         }
 324                         goto error;
 325                 }
 326                 nb_fd = ret;
 327                 DBG3("[agent-thread] %d fd ready", nb_fd);
 328
 329                 for (i = 0; i < nb_fd; i++) {
 330                         /* Fetch once the poll data */
 331                         revents = LTTNG_POLL_GETEV(&events, i);
 332                         pollfd = LTTNG_POLL_GETFD(&events, i);
 333
 334                         if (!revents) {
 335                                 /* No activity for this FD (poll implementation). */
 336                                 continue;
 337                         }
 338
 339                         /* Thread quit pipe has been closed. Killing thread. */
 340                         ret = sessiond_check_thread_quit_pipe(pollfd, revents);
 341                         if (ret) {
 342                                 goto exit;
 343                         }
 344
 345                         /*
 346                          * Check first if this is a POLLERR since POLLIN is also included
 347                          * in an error value thus checking first.
 348                          */
 349                         if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
 350                                 /* Removing from the poll set */
 351                                 ret = lttng_poll_del(&events, pollfd);
 352                                 if (ret < 0) {
 353                                         goto error;
 354                                 }
 355
 356                                 destroy_agent_app(pollfd);
 357                         } else if (revents & (LPOLLIN)) {
 358                                 int new_fd;
 359                                 struct agent_app *app = NULL;
 360
 361                                 /* Pollin event of agent app socket should NEVER happen. */
 362                                 assert(pollfd == reg_sock->fd);
 363
 364                                 new_fd = handle_registration(reg_sock, &app);
 365                                 if (new_fd < 0) {
 366                                         WARN("[agent-thread] agent registration failed. Ignoring.");
 367                                         /* Somehow the communication failed. Just continue. */
 368                                         continue;
 369                                 }
 370                                 /* Should not have a NULL app on success. */
 371                                 assert(app);
 372
 373                                 /* Only add poll error event to only detect shutdown. */
 374                                 ret = lttng_poll_add(&events, new_fd,
 375                                                 LPOLLERR | LPOLLHUP | LPOLLRDHUP);
 376                                 if (ret < 0) {
 377                                         destroy_agent_app(new_fd);
 378                                         continue;
 379                                 }
 380
 381                                 /* Update newly registered app. */
 382                                 update_agent_app(app);
 383
 384                                 /* On failure, the poll will detect it and clean it up. */
 385                                 (void) agent_send_registration_done(app);
 386                         } else {
 387                                 ERR("Unknown poll events %u for sock %d", revents, pollfd);
 388                                 continue;
 389                         }
 390                 }
 391         }
 392
 393 exit:
 394         /* Whatever happens, try to delete it and exit. */
 395         (void) lttng_poll_del(&events, reg_sock->fd);
 396 error:
 397         destroy_tcp_socket(reg_sock);
 398 error_tcp_socket:
 399         lttng_poll_clean(&events);
 400 error_poll_create:
 401         DBG("[agent-thread] is cleaning up and stopping.");
 402
 403         if (agent_apps_ht_by_sock) {
 404                 clean_agent_apps_ht();
 405                 lttng_ht_destroy(agent_apps_ht_by_sock);
 406         }
 407
 408         rcu_thread_offline();
 409         rcu_unregister_thread();
 410         return NULL;
 411 }