/*
- * Copyright (C) 2013 - David Goulet <dgoulet@efficios.com>
+ * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
*
- * This program is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License, version 2 only, as
- * published by the Free Software Foundation.
+ * SPDX-License-Identifier: GPL-2.0-only
*
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
- * more details.
- *
- * You should have received a copy of the GNU General Public License along with
- * this program; if not, write to the Free Software Foundation, Inc., 51
- * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#define _LGPL_SOURCE
#include "utils.h"
#include "thread.h"
+struct thread_notifiers {
+ struct lttng_pipe *quit_pipe;
+ sem_t ready;
+};
+
+struct agent_app_id {
+ pid_t pid;
+ enum lttng_domain_type domain;
+};
+
+struct agent_protocol_version {
+ unsigned int major, minor;
+};
+
static int agent_tracing_enabled = -1;
/*
* Update agent application using the given socket. This is done just after
* registration was successful.
*
- * This is a quite heavy call in terms of locking since the session list lock
- * AND session lock are acquired.
+ * This will acquire the various sessions' lock; none must be held by the
+ * caller.
+ * The caller must hold the session list lock.
*/
-static void update_agent_app(struct agent_app *app)
+static void update_agent_app(const struct agent_app *app)
{
struct ltt_session *session, *stmp;
struct ltt_session_list *list;
list = session_get_list();
assert(list);
- session_lock_list();
cds_list_for_each_entry_safe(session, stmp, &list->head, list) {
if (!session_get(session)) {
continue;
session_lock(session);
if (session->ust_session) {
- struct agent *agt;
+ const struct agent *agt;
rcu_read_lock();
agt = trace_ust_find_agent(session->ust_session, app->domain);
if (agt) {
- agent_update(agt, app->sock->fd);
+ agent_update(agt, app);
}
rcu_read_unlock();
}
session_unlock(session);
session_put(session);
}
- session_unlock_list();
}
/*
lttcomm_destroy_sock(sock);
}
+static const char *domain_type_str(enum lttng_domain_type domain_type)
+{
+ switch (domain_type) {
+ case LTTNG_DOMAIN_NONE:
+ return "none";
+ case LTTNG_DOMAIN_KERNEL:
+ return "kernel";
+ case LTTNG_DOMAIN_UST:
+ return "ust";
+ case LTTNG_DOMAIN_JUL:
+ return "jul";
+ case LTTNG_DOMAIN_LOG4J:
+ return "log4j";
+ case LTTNG_DOMAIN_PYTHON:
+ return "python";
+ default:
+ return "unknown";
+ }
+}
+
+static bool is_agent_protocol_version_supported(
+ const struct agent_protocol_version *version)
+{
+ const bool is_supported = version->major == AGENT_MAJOR_VERSION &&
+ version->minor == AGENT_MINOR_VERSION;
+
+ if (!is_supported) {
+ WARN("Refusing agent connection: unsupported protocol version %ui.%ui, expected %i.%i",
+ version->major, version->minor,
+ AGENT_MAJOR_VERSION, AGENT_MINOR_VERSION);
+ }
+
+ return is_supported;
+}
+
/*
- * Handle a new agent registration using the reg socket. After that, a new
- * agent application is added to the global hash table and attach to an UST app
- * object. If r_app is not NULL, the created app is set to the pointer.
+ * Handle a new agent connection on the registration socket.
*
- * Return the new FD created upon accept() on success or else a negative errno
- * value.
+ * Returns 0 on success, or else a negative errno value.
+ * On success, the resulting socket is returned through `agent_app_socket`
+ * and the application's reported id is updated through `agent_app_id`.
*/
-static int handle_registration(struct lttcomm_sock *reg_sock,
- struct agent_app **r_app)
+static int accept_agent_connection(
+ struct lttcomm_sock *reg_sock,
+ struct agent_app_id *agent_app_id,
+ struct lttcomm_sock **agent_app_socket)
{
int ret;
- pid_t pid;
- uint32_t major_version, minor_version;
+ struct agent_protocol_version agent_version;
ssize_t size;
- enum lttng_domain_type domain;
- struct agent_app *app;
struct agent_register_msg msg;
struct lttcomm_sock *new_sock;
new_sock = reg_sock->ops->accept(reg_sock);
if (!new_sock) {
ret = -ENOTCONN;
- goto error;
+ goto end;
}
size = new_sock->ops->recvmsg(new_sock, &msg, sizeof(msg), 0);
if (size < sizeof(msg)) {
+ if (size < 0) {
+ PERROR("Failed to register new agent application");
+ } else if (size != 0) {
+ ERR("Failed to register new agent application: invalid registration message length: expected length = %zu, message length = %zd",
+ sizeof(msg), size);
+ } else {
+ DBG("Failed to register new agent application: connection closed");
+ }
ret = -EINVAL;
- goto error_socket;
+ goto error_close_socket;
}
- domain = be32toh(msg.domain);
- pid = be32toh(msg.pid);
- major_version = be32toh(msg.major_version);
- minor_version = be32toh(msg.minor_version);
- /* Test communication protocol version of the registring agent. */
- if (major_version != AGENT_MAJOR_VERSION) {
- ret = -EINVAL;
- goto error_socket;
- }
- if (minor_version != AGENT_MINOR_VERSION) {
+ agent_version = (struct agent_protocol_version) {
+ be32toh(msg.major_version),
+ be32toh(msg.minor_version),
+ };
+
+ /* Test communication protocol version of the registering agent. */
+ if (!is_agent_protocol_version_supported(&agent_version)) {
ret = -EINVAL;
- goto error_socket;
+ goto error_close_socket;
}
- DBG2("[agent-thread] New registration for pid %d domain %d on socket %d",
- pid, domain, new_sock->fd);
+ *agent_app_id = (struct agent_app_id) {
+ .domain = (enum lttng_domain_type) be32toh(msg.domain),
+ .pid = (pid_t) be32toh(msg.pid),
+ };
- app = agent_create_app(pid, domain, new_sock);
- if (!app) {
- ret = -ENOMEM;
- goto error_socket;
- }
+ DBG2("New registration for agent application: pid = %ld, domain = %s, socket fd = %d",
+ (long) agent_app_id->pid,
+ domain_type_str(agent_app_id->domain), new_sock->fd);
- /*
- * Add before assigning the socket value to the UST app so it can be found
- * concurrently.
- */
- agent_add_app(app);
-
- /*
- * We don't need to attach the agent app to the app. If we ever do so, we
- * should consider both registration order of agent before app and app
- * before agent.
- */
+ *agent_app_socket = new_sock;
+ new_sock = NULL;
+ ret = 0;
+ goto end;
- if (r_app) {
- *r_app = app;
- }
-
- return new_sock->fd;
-
-error_socket:
+error_close_socket:
new_sock->ops->close(new_sock);
lttcomm_destroy_sock(new_sock);
-error:
+end:
return ret;
}
config.agent_port_file_path.value);
}
+static
+void mark_thread_as_ready(struct thread_notifiers *notifiers)
+{
+ DBG("Marking agent management thread as ready");
+ sem_post(¬ifiers->ready);
+}
+
+static
+void wait_until_thread_is_ready(struct thread_notifiers *notifiers)
+{
+ DBG("Waiting for agent management thread to be ready");
+ sem_wait(¬ifiers->ready);
+ DBG("Agent management thread is ready");
+}
+
/*
* This thread manage application notify communication.
*/
uint32_t revents, nb_fd;
struct lttng_poll_event events;
struct lttcomm_sock *reg_sock;
- struct lttng_pipe *quit_pipe = data;
- const int quit_pipe_read_fd = lttng_pipe_get_readfd(quit_pipe);
+ struct thread_notifiers *notifiers = data;
+ const int quit_pipe_read_fd = lttng_pipe_get_readfd(
+ notifiers->quit_pipe);
DBG("[agent-thread] Manage agent application registration.");
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();
+ mark_thread_as_ready(notifiers);
goto error;
}
} else {
/* Don't prevent the launch of the sessiond on error. */
- sessiond_notify_ready();
+ mark_thread_as_ready(notifiers);
goto error_tcp_socket;
}
* may start to query whether or not agent tracing is enabled.
*/
uatomic_set(&agent_tracing_enabled, 1);
- sessiond_notify_ready();
+ mark_thread_as_ready(notifiers);
- /* Add TCP socket to poll set. */
+ /* Add TCP socket to the poll set. */
ret = lttng_poll_add(&events, reg_sock->fd,
LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP);
if (ret < 0) {
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. */
if (pollfd == quit_pipe_read_fd) {
goto exit;
}
+ /* Activity on the registration socket. */
if (revents & LPOLLIN) {
- int new_fd;
- struct agent_app *app = NULL;
+ struct agent_app_id new_app_id;
+ struct agent_app *new_app = NULL;
+ struct lttcomm_sock *new_app_socket;
+ int new_app_socket_fd;
assert(pollfd == reg_sock->fd);
- new_fd = handle_registration(reg_sock, &app);
- if (new_fd < 0) {
+
+ ret = accept_agent_connection(
+ reg_sock, &new_app_id, &new_app_socket);
+ if (ret < 0) {
+ /* Errors are already logged. */
+ continue;
+ }
+
+ /*
+ * new_app_socket's ownership has been
+ * transferred to the new agent app.
+ */
+ new_app = agent_create_app(new_app_id.pid,
+ new_app_id.domain,
+ new_app_socket);
+ if (!new_app) {
+ new_app_socket->ops->close(
+ new_app_socket);
continue;
}
- /* Should not have a NULL app on success. */
- assert(app);
+ new_app_socket_fd = new_app_socket->fd;
+ new_app_socket = NULL;
/*
- * Since this is a command socket (write then read),
- * 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,
+ ret = lttng_poll_add(&events, new_app_socket_fd,
LPOLLERR | LPOLLHUP | LPOLLRDHUP);
if (ret < 0) {
- agent_destroy_app_by_sock(new_fd);
+ agent_destroy_app(new_app);
continue;
}
- /* Update newly registered app. */
- update_agent_app(app);
+ /*
+ * Prevent sessions from being modified while
+ * the agent application's configuration is
+ * updated.
+ */
+ session_lock_list();
- /* On failure, the poll will detect it and clean it up. */
- ret = agent_send_registration_done(app);
+ /*
+ * Update the newly registered applications's
+ * configuration.
+ */
+ update_agent_app(new_app);
+
+ ret = agent_send_registration_done(new_app);
if (ret < 0) {
- /* Removing from the poll set */
- ret = lttng_poll_del(&events, new_fd);
+ agent_destroy_app(new_app);
+ /* Removing from the poll set. */
+ ret = lttng_poll_del(&events,
+ new_app_socket_fd);
if (ret < 0) {
+ session_unlock_list();
goto error;
}
- agent_destroy_app_by_sock(new_fd);
continue;
}
+
+ /* Publish the new agent app. */
+ agent_add_app(new_app);
+
+ session_unlock_list();
} else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
/* Removing from the poll set */
ret = lttng_poll_del(&events, pollfd);
static bool shutdown_agent_management_thread(void *data)
{
- struct lttng_pipe *quit_pipe = data;
- const int write_fd = lttng_pipe_get_writefd(quit_pipe);
+ struct thread_notifiers *notifiers = data;
+ const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe);
return notify_thread_pipe(write_fd) == 1;
}
static void cleanup_agent_management_thread(void *data)
{
- struct lttng_pipe *quit_pipe = data;
+ struct thread_notifiers *notifiers = data;
- lttng_pipe_destroy(quit_pipe);
+ lttng_pipe_destroy(notifiers->quit_pipe);
+ sem_destroy(¬ifiers->ready);
+ free(notifiers);
}
bool launch_agent_management_thread(void)
{
- struct lttng_pipe *quit_pipe;
+ struct thread_notifiers *notifiers;
struct lttng_thread *thread;
- quit_pipe = lttng_pipe_open(FD_CLOEXEC);
- if (!quit_pipe) {
+ notifiers = zmalloc(sizeof(*notifiers));
+ if (!notifiers) {
+ goto error_alloc;
+ }
+
+ sem_init(¬ifiers->ready, 0, 0);
+ notifiers->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
+ if (!notifiers->quit_pipe) {
goto error;
}
thread = lttng_thread_create("Agent management",
thread_agent_management,
shutdown_agent_management_thread,
cleanup_agent_management_thread,
- quit_pipe);
+ notifiers);
if (!thread) {
goto error;
}
-
+ wait_until_thread_is_ready(notifiers);
lttng_thread_put(thread);
return true;
error:
- cleanup_agent_management_thread(quit_pipe);
+ cleanup_agent_management_thread(notifiers);
+error_alloc:
return false;
}