Fix: metadata push -EPIPE should be recoverable
[lttng-tools.git] / src / bin / lttng-sessiond / ust-app.c
index b33378f00a36e1982c20fabe6f7db689d2486378..6f032da112baae7a9a97af061f4388bc621b6d6d 100644 (file)
@@ -431,6 +431,9 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan,
  * Must be called with the registry lock held.
  *
  * On success, return the len of metadata pushed or else a negative value.
+ * Returning a -EPIPE return value means we could not send the metadata,
+ * but it can be caused by recoverable errors (e.g. the application has
+ * terminated concurrently).
  */
 ssize_t ust_app_push_metadata(struct ust_registry_session *registry,
                struct consumer_socket *socket, int send_zero_data)
@@ -454,9 +457,10 @@ ssize_t ust_app_push_metadata(struct ust_registry_session *registry,
        /*
         * On a push metadata error either the consumer is dead or the
         * metadata channel has been destroyed because its endpoint
-        * might have died (e.g: relayd). If so, the metadata closed
-        * flag is set to 1 so we deny pushing metadata again which is
-        * not valid anymore on the consumer side.
+        * might have died (e.g: relayd), or because the application has
+        * exited. If so, the metadata closed flag is set to 1 so we
+        * deny pushing metadata again which is not valid anymore on the
+        * consumer side.
         */
        if (registry->metadata_closed) {
                return -EPIPE;
@@ -547,6 +551,9 @@ error_push:
  * of socket throughout this function.
  *
  * Return 0 on success else a negative error.
+ * Returning a -EPIPE return value means we could not send the metadata,
+ * but it can be caused by recoverable errors (e.g. the application has
+ * terminated concurrently).
  */
 static int push_metadata(struct ust_registry_session *registry,
                struct consumer_output *consumer)
@@ -1593,6 +1600,7 @@ static void shadow_copy_session(struct ust_app_session *ua_sess,
        struct tm *timeinfo;
        char datetime[16];
        int ret;
+       char tmp_shm_path[PATH_MAX];
 
        /* Get date and time for unique app path */
        time(&rawtime);
@@ -1636,6 +1644,38 @@ static void shadow_copy_session(struct ust_app_session *ua_sess,
                goto error;
        }
 
+       strncpy(ua_sess->root_shm_path, usess->root_shm_path,
+               sizeof(ua_sess->root_shm_path));
+       ua_sess->root_shm_path[sizeof(ua_sess->root_shm_path) - 1] = '\0';
+       strncpy(ua_sess->shm_path, usess->shm_path,
+               sizeof(ua_sess->shm_path));
+       ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0';
+       if (ua_sess->shm_path[0]) {
+               switch (ua_sess->buffer_type) {
+               case LTTNG_BUFFER_PER_PID:
+                       ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path),
+                                       DEFAULT_UST_TRACE_PID_PATH "/%s-%d-%s",
+                                       app->name, app->pid, datetime);
+                       break;
+               case LTTNG_BUFFER_PER_UID:
+                       ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path),
+                                       DEFAULT_UST_TRACE_UID_PATH,
+                                       app->uid, app->bits_per_long);
+                       break;
+               default:
+                       assert(0);
+                       goto error;
+               }
+               if (ret < 0) {
+                       PERROR("sprintf UST shadow copy session");
+                       assert(0);
+                       goto error;
+               }
+               strncat(ua_sess->shm_path, tmp_shm_path,
+                       sizeof(ua_sess->shm_path) - strlen(ua_sess->shm_path) - 1);
+               ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0';
+       }
+
        /* Iterate over all channels in global domain. */
        cds_lfht_for_each_entry(usess->domain_global.channels->ht, &iter.iter,
                        uchan, node.node) {
@@ -1727,7 +1767,8 @@ static int setup_buffer_reg_pid(struct ust_app_session *ua_sess,
                 * This is the create channel path meaning that if there is NO
                 * registry available, we have to create one for this session.
                 */
-               ret = buffer_reg_pid_create(ua_sess->id, &reg_pid);
+               ret = buffer_reg_pid_create(ua_sess->id, &reg_pid,
+                       ua_sess->root_shm_path, ua_sess->shm_path);
                if (ret < 0) {
                        goto error;
                }
@@ -1741,7 +1782,9 @@ static int setup_buffer_reg_pid(struct ust_app_session *ua_sess,
                        app->uint16_t_alignment, app->uint32_t_alignment,
                        app->uint64_t_alignment, app->long_alignment,
                        app->byte_order, app->version.major,
-                       app->version.minor);
+                       app->version.minor, reg_pid->root_shm_path,
+                       reg_pid->shm_path,
+                       ua_sess->euid, ua_sess->egid);
        if (ret < 0) {
                /*
                 * reg_pid->registry->reg.ust is NULL upon error, so we need to
@@ -1774,6 +1817,7 @@ error:
  * Return 0 on success or else a negative value.
  */
 static int setup_buffer_reg_uid(struct ltt_ust_session *usess,
+               struct ust_app_session *ua_sess,
                struct ust_app *app, struct buffer_reg_uid **regp)
 {
        int ret = 0;
@@ -1791,7 +1835,8 @@ static int setup_buffer_reg_uid(struct ltt_ust_session *usess,
                 * registry available, we have to create one for this session.
                 */
                ret = buffer_reg_uid_create(usess->id, app->bits_per_long, app->uid,
-                               LTTNG_DOMAIN_UST, &reg_uid);
+                               LTTNG_DOMAIN_UST, &reg_uid,
+                               ua_sess->root_shm_path, ua_sess->shm_path);
                if (ret < 0) {
                        goto error;
                }
@@ -1805,7 +1850,8 @@ static int setup_buffer_reg_uid(struct ltt_ust_session *usess,
                        app->uint16_t_alignment, app->uint32_t_alignment,
                        app->uint64_t_alignment, app->long_alignment,
                        app->byte_order, app->version.major,
-                       app->version.minor);
+                       app->version.minor, reg_uid->root_shm_path,
+                       reg_uid->shm_path, usess->uid, usess->gid);
        if (ret < 0) {
                /*
                 * reg_uid->registry->reg.ust is NULL upon error, so we need to
@@ -1880,7 +1926,7 @@ static int create_ust_app_session(struct ltt_ust_session *usess,
                break;
        case LTTNG_BUFFER_PER_UID:
                /* Look for a global registry. If none exists, create one. */
-               ret = setup_buffer_reg_uid(usess, app, NULL);
+               ret = setup_buffer_reg_uid(usess, ua_sess, app, NULL);
                if (ret < 0) {
                        delete_ust_app_session(-1, ua_sess, app);
                        goto error;
@@ -3687,6 +3733,11 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess,
                         */
                        continue;
                }
+               if (!trace_ust_pid_tracker_lookup(usess, app->pid)) {
+                       /* Skip. */
+                       continue;
+               }
+
                /*
                 * Create session on the tracer side and add it to app session HT. Note
                 * that if session exist, it will simply return a pointer to the ust
@@ -4185,7 +4236,6 @@ int ust_app_flush_session(struct ltt_ust_session *usess)
                break;
        }
 
-end_no_session:
        rcu_read_unlock();
        health_code_update();
        return ret;
@@ -4315,46 +4365,26 @@ int ust_app_destroy_trace_all(struct ltt_ust_session *usess)
        return 0;
 }
 
-/*
- * Add channels/events from UST global domain to registered apps at sock.
- */
-void ust_app_global_update(struct ltt_ust_session *usess, int sock)
+static
+void ust_app_global_create(struct ltt_ust_session *usess, struct ust_app *app)
 {
        int ret = 0;
        struct lttng_ht_iter iter, uiter;
-       struct ust_app *app;
        struct ust_app_session *ua_sess = NULL;
        struct ust_app_channel *ua_chan;
        struct ust_app_event *ua_event;
        struct ust_app_ctx *ua_ctx;
+       int is_created = 0;
 
-       assert(usess);
-       assert(sock >= 0);
-
-       DBG2("UST app global update for app sock %d for session id %" PRIu64, sock,
-                       usess->id);
-
-       rcu_read_lock();
-
-       app = ust_app_find_by_sock(sock);
-       if (app == NULL) {
-               /*
-                * Application can be unregistered before so this is possible hence
-                * simply stopping the update.
-                */
-               DBG3("UST app update failed to find app sock %d", sock);
-               goto error;
-       }
-
-       if (!app->compatible) {
-               goto error;
-       }
-
-       ret = create_ust_app_session(usess, app, &ua_sess, NULL);
+       ret = create_ust_app_session(usess, app, &ua_sess, &is_created);
        if (ret < 0) {
                /* Tracer is probably gone or ENOMEM. */
                goto error;
        }
+       if (!is_created) {
+               /* App session already created. */
+               goto end;
+       }
        assert(ua_sess);
 
        pthread_mutex_lock(&ua_sess->lock);
@@ -4408,9 +4438,8 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock)
 
                DBG2("UST trace started for app pid %d", app->pid);
        }
-
+end:
        /* Everything went well at this point. */
-       rcu_read_unlock();
        return;
 
 error_unlock:
@@ -4419,10 +4448,60 @@ error:
        if (ua_sess) {
                destroy_app_session(app, ua_sess);
        }
-       rcu_read_unlock();
        return;
 }
 
+static
+void ust_app_global_destroy(struct ltt_ust_session *usess, struct ust_app *app)
+{
+       struct ust_app_session *ua_sess;
+
+       ua_sess = lookup_session_by_app(usess, app);
+       if (ua_sess == NULL) {
+               return;
+       }
+       destroy_app_session(app, ua_sess);
+}
+
+/*
+ * Add channels/events from UST global domain to registered apps at sock.
+ *
+ * Called with session lock held.
+ * Called with RCU read-side lock held.
+ */
+void ust_app_global_update(struct ltt_ust_session *usess, struct ust_app *app)
+{
+       assert(usess);
+
+       DBG2("UST app global update for app sock %d for session id %" PRIu64,
+                       app->sock, usess->id);
+
+       if (!app->compatible) {
+               return;
+       }
+
+       if (trace_ust_pid_tracker_lookup(usess, app->pid)) {
+               ust_app_global_create(usess, app);
+       } else {
+               ust_app_global_destroy(usess, app);
+       }
+}
+
+/*
+ * Called with session lock held.
+ */
+void ust_app_global_update_all(struct ltt_ust_session *usess)
+{
+       struct lttng_ht_iter iter;
+       struct ust_app *app;
+
+       rcu_read_lock();
+       cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) {
+               ust_app_global_update(usess, app);
+       }
+       rcu_read_unlock();
+}
+
 /*
  * Add context to a specific channel for global UST domain.
  */
This page took 0.028775 seconds and 4 git commands to generate.