Fix: unprivilieged sessiond agent port clashes with root sessiond
[lttng-tools.git] / src / bin / lttng-sessiond / agent-thread.c
index d7d9c6c349412ba659c3d3720e950b18ac3b8389..f8456b428d49882945c2089bf12e3c4121c7420b 100644 (file)
@@ -15,7 +15,6 @@
  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  */
 
-#define _GNU_SOURCE
 #define _LGPL_SOURCE
 #include <assert.h>
 
 
 #include "fd-limit.h"
 #include "agent-thread.h"
+#include "agent.h"
 #include "lttng-sessiond.h"
 #include "session.h"
 #include "utils.h"
 
+static int agent_tracing_enabled = -1;
+
 /*
  * Note that there is not port here. It's set after this URI is parsed so we
  * can let the user define a custom one. However, localhost is ALWAYS the
@@ -73,53 +75,6 @@ static void update_agent_app(struct agent_app *app)
        session_unlock_list();
 }
 
-/*
- * Destroy a agent application by socket.
- */
-static void destroy_agent_app(int sock)
-{
-       struct agent_app *app;
-
-       assert(sock >= 0);
-
-       /*
-        * Not finding an application is a very important error that should NEVER
-        * happen. The hash table deletion is ONLY done through this call even on
-        * thread cleanup.
-        */
-       rcu_read_lock();
-       app = agent_find_app_by_sock(sock);
-       assert(app);
-       rcu_read_unlock();
-
-       /* RCU read side lock is taken in this function call. */
-       agent_delete_app(app);
-
-       /* The application is freed in a RCU call but the socket is closed here. */
-       agent_destroy_app(app);
-}
-
-/*
- * Cleanup remaining agent apps in the hash table. This should only be called in
- * the exit path of the thread.
- */
-static void clean_agent_apps_ht(void)
-{
-       struct lttng_ht_node_ulong *node;
-       struct lttng_ht_iter iter;
-
-       DBG3("[agent-thread] Cleaning agent apps ht");
-
-       rcu_read_lock();
-       cds_lfht_for_each_entry(agent_apps_ht_by_sock->ht, &iter.iter, node, node) {
-               struct agent_app *app;
-
-               app = caa_container_of(node, struct agent_app, node);
-               destroy_agent_app(app->sock->fd);
-       }
-       rcu_read_unlock();
-}
-
 /*
  * Create and init socket from uri.
  */
@@ -128,6 +83,8 @@ static struct lttcomm_sock *init_tcp_socket(void)
        int ret;
        struct lttng_uri *uri = NULL;
        struct lttcomm_sock *sock = NULL;
+       unsigned int port;
+       bool bind_succeeded = false;
 
        /*
         * This should never fail since the URI is hardcoded and the port is set
@@ -135,8 +92,8 @@ static struct lttcomm_sock *init_tcp_socket(void)
         */
        ret = uri_parse(default_reg_uri, &uri);
        assert(ret);
-       assert(agent_tcp_port);
-       uri->port = agent_tcp_port;
+       assert(config.agent_tcp_port.begin > 0);
+       uri->port = config.agent_tcp_port.begin;
 
        sock = lttcomm_alloc_sock_from_uri(uri);
        uri_free(uri);
@@ -150,11 +107,43 @@ static struct lttcomm_sock *init_tcp_socket(void)
                goto error;
        }
 
-       ret = sock->ops->bind(sock);
-       if (ret < 0) {
-               WARN("Another session daemon is using this agent port. Agent support "
-                               "will be deactivated to prevent interfering with the tracing.");
-               goto error;
+       for (port = config.agent_tcp_port.begin;
+                       port <= config.agent_tcp_port.end; port++) {
+               ret = lttcomm_sock_set_port(sock, (uint16_t) port);
+               if (ret) {
+                       ERR("[agent-thread] Failed to set port %u on socket",
+                                       port);
+                       goto error;
+               }
+               DBG3("[agent-thread] Trying to bind on port %u", port);
+               ret = sock->ops->bind(sock);
+               if (!ret) {
+                       bind_succeeded = true;
+                       break;
+               }
+
+               if (errno == EADDRINUSE) {
+                       DBG("Failed to bind to port %u since it is already in use",
+                                       port);
+               } else {
+                       PERROR("Failed to bind to port %u", port);
+                       goto error;
+               }
+       }
+
+       if (!bind_succeeded) {
+               if (config.agent_tcp_port.begin == config.agent_tcp_port.end) {
+                       WARN("Another process is already using the agent port %i. "
+                                       "Agent support will be deactivated.",
+                                       config.agent_tcp_port.begin);
+                       goto error;
+               } else {
+                       WARN("All ports in the range [%i, %i] are already in use. "
+                                       "Agent support will be deactivated.",
+                                       config.agent_tcp_port.begin,
+                                       config.agent_tcp_port.end);
+                       goto error;
+               }
        }
 
        ret = sock->ops->listen(sock, -1);
@@ -163,7 +152,7 @@ static struct lttcomm_sock *init_tcp_socket(void)
        }
 
        DBG("[agent-thread] Listening on TCP port %u and socket %d",
-                       agent_tcp_port, sock->fd);
+                       port, sock->fd);
 
        return sock;
 
@@ -179,9 +168,19 @@ error:
  */
 static void destroy_tcp_socket(struct lttcomm_sock *sock)
 {
+       int ret;
+       uint16_t port;
+
        assert(sock);
 
-       DBG3("[agent-thread] Destroy TCP socket on port %u", agent_tcp_port);
+       ret = lttcomm_sock_get_port(sock, &port);
+       if (ret) {
+               ERR("[agent-thread] Failed to get port of agent TCP socket");
+               port = 0;
+       }
+
+       DBG3("[agent-thread] Destroy TCP socket on port %" PRIu16,
+                       port);
 
        /* This will return gracefully if fd is invalid. */
        sock->ops->close(sock);
@@ -270,6 +269,24 @@ error:
        return ret;
 }
 
+bool agent_tracing_is_enabled(void)
+{
+       int enabled;
+
+       enabled = uatomic_read(&agent_tracing_enabled);
+       assert(enabled != -1);
+       return enabled == 1;
+}
+
+/*
+ * Write agent TCP port using the rundir.
+ */
+static int write_agent_port(uint16_t port)
+{
+       return utils_create_pid_file((pid_t) port,
+                       config.agent_port_file_path.value);
+}
+
 /*
  * This thread manage application notify communication.
  */
@@ -295,11 +312,32 @@ void *agent_thread_manage_registration(void *data)
        }
 
        reg_sock = init_tcp_socket();
-       if (!reg_sock) {
+       if (reg_sock) {
+               uint16_t port;
+
+               assert(lttcomm_sock_get_port(reg_sock, &port) == 0);
+
+               ret = write_agent_port(port);
+               if (ret) {
+                       ERR("[agent-thread] Failed to create agent port file: agent tracing will be unavailable");
+                       /* Don't prevent the launch of the sessiond on error. */
+                       sessiond_notify_ready();
+                       goto error;
+               }
+       } else {
+               /* Don't prevent the launch of the sessiond on error. */
+               sessiond_notify_ready();
                goto error_tcp_socket;
        }
 
-       /* Add create valid TCP socket to poll set. */
+       /*
+        * Signal that the agent thread is ready. The command thread
+        * may start to query whether or not agent tracing is enabled.
+        */
+       uatomic_set(&agent_tracing_enabled, 1);
+       sessiond_notify_ready();
+
+       /* Add TCP socket to poll set. */
        ret = lttng_poll_add(&events, reg_sock->fd,
                        LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
        if (ret < 0) {
@@ -307,12 +345,13 @@ void *agent_thread_manage_registration(void *data)
        }
 
        while (1) {
-               DBG3("[agent-thread] Manage agent polling on %d fds",
-                               LTTNG_POLL_GETNB(&events));
+               DBG3("[agent-thread] Manage agent polling");
 
                /* Inifinite blocking call, waiting for transmission */
 restart:
                ret = lttng_poll_wait(&events, -1);
+               DBG3("[agent-thread] Manage agent return from poll on %d fds",
+                               LTTNG_POLL_GETNB(&events));
                if (ret < 0) {
                        /*
                         * Restart interrupted system call.
@@ -330,45 +369,37 @@ restart:
                        revents = LTTNG_POLL_GETEV(&events, i);
                        pollfd = LTTNG_POLL_GETFD(&events, i);
 
+                       if (!revents) {
+                               /* No activity for this FD (poll implementation). */
+                               continue;
+                       }
+
                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = sessiond_check_thread_quit_pipe(pollfd, revents);
                        if (ret) {
                                goto exit;
                        }
 
-                       /*
-                        * Check first if this is a POLLERR since POLLIN is also included
-                        * in an error value thus checking first.
-                        */
-                       if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
-                               /* Removing from the poll set */
-                               ret = lttng_poll_del(&events, pollfd);
-                               if (ret < 0) {
-                                       goto error;
-                               }
-
-                               destroy_agent_app(pollfd);
-                       } else if (revents & (LPOLLIN)) {
+                       if (revents & LPOLLIN) {
                                int new_fd;
                                struct agent_app *app = NULL;
 
-                               /* Pollin event of agent app socket should NEVER happen. */
                                assert(pollfd == reg_sock->fd);
-
                                new_fd = handle_registration(reg_sock, &app);
                                if (new_fd < 0) {
-                                       WARN("[agent-thread] agent registration failed. Ignoring.");
-                                       /* Somehow the communication failed. Just continue. */
                                        continue;
                                }
                                /* Should not have a NULL app on success. */
                                assert(app);
 
-                               /* Only add poll error event to only detect shutdown. */
+                               /*
+                                * Since this is a command socket (write then read),
+                                * only add poll error event to only detect shutdown.
+                                */
                                ret = lttng_poll_add(&events, new_fd,
                                                LPOLLERR | LPOLLHUP | LPOLLRDHUP);
                                if (ret < 0) {
-                                       destroy_agent_app(new_fd);
+                                       agent_destroy_app_by_sock(new_fd);
                                        continue;
                                }
 
@@ -376,10 +407,26 @@ restart:
                                update_agent_app(app);
 
                                /* On failure, the poll will detect it and clean it up. */
-                               (void) agent_send_registration_done(app);
+                               ret = agent_send_registration_done(app);
+                               if (ret < 0) {
+                                       /* Removing from the poll set */
+                                       ret = lttng_poll_del(&events, new_fd);
+                                       if (ret < 0) {
+                                               goto error;
+                                       }
+                                       agent_destroy_app_by_sock(new_fd);
+                                       continue;
+                               }
+                       } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
+                               /* Removing from the poll set */
+                               ret = lttng_poll_del(&events, pollfd);
+                               if (ret < 0) {
+                                       goto error;
+                               }
+                               agent_destroy_app_by_sock(pollfd);
                        } else {
-                               ERR("Unknown poll events %u for sock %d", revents, pollfd);
-                               continue;
+                               ERR("Unexpected poll events %u for sock %d", revents, pollfd);
+                               goto error;
                        }
                }
        }
@@ -392,13 +439,9 @@ error:
 error_tcp_socket:
        lttng_poll_clean(&events);
 error_poll_create:
+       uatomic_set(&agent_tracing_enabled, 0);
        DBG("[agent-thread] is cleaning up and stopping.");
 
-       if (agent_apps_ht_by_sock) {
-               clean_agent_apps_ht();
-               lttng_ht_destroy(agent_apps_ht_by_sock);
-       }
-
        rcu_thread_offline();
        rcu_unregister_thread();
        return NULL;
This page took 0.027204 seconds and 4 git commands to generate.