If an exit has already been scheduled we should not schedule it again.
Otherwise, the exit signal is never emitted if the peer reschedules the
exit before the timeout occurs.
schedule_exit() now only takes the context as argument. The signal is
hard coded to SIGTERM, and the interval is read directly from the
context options.
Furthermore, schedule_exit() now returns a bool signifying whether an
exit was scheduled; false if exit is already scheduled. The call sites
are updated accordingly. A notable difference is that management is only
notified *once* when an exit is scheduled - we no longer notify
management on redundant exit.
This patch was assigned a CVE number after already reviewed and ACKed,
because it was discovered that a misbehaving client can use the (now
fixed) server behaviour to avoid being disconnected by means of a
managment interface "client-kill" command - the security issue here is
"client can circumvent security policy set by management interface".
This only affects previously authenticated clients, and only management
client-kill, so normal renegotion / AUTH_FAIL ("your session ends") is not
affected.
CVE: 2024-28882
Change-Id: I9457f005f4ba970502e6b667d9dc4299a588d661
Signed-off-by: Reynir Björnsson <reynir@reynir.dk>
Acked-by: Arne Schwabe <arne-openvpn@rfc2549.org>
Message-Id: <
20240516120434.23499-1-gert@greenie.muc.de>
URL: https://www.mail-archive.com/openvpn-devel@lists.sourceforge.net/msg28679.html
Signed-off-by: Gert Doering <gert@greenie.muc.de>
}
/*
- * Schedule a signal n_seconds from now.
+ * Schedule a SIGTERM signal c->options.scheduled_exit_interval seconds from now.
*/
-void
-schedule_exit(struct context *c, const int n_seconds, const int signal)
+bool
+schedule_exit(struct context *c)
{
+ const int n_seconds = c->options.scheduled_exit_interval;
+ /* don't reschedule if already scheduled. */
+ if (event_timeout_defined(&c->c2.scheduled_exit))
+ {
+ return false;
+ }
tls_set_single_session(c->c2.tls_multi);
update_time();
reset_coarse_timers(c);
event_timeout_init(&c->c2.scheduled_exit, n_seconds, now);
- c->c2.scheduled_exit_signal = signal;
+ c->c2.scheduled_exit_signal = SIGTERM;
msg(D_SCHED_EXIT, "Delayed exit in %d seconds", n_seconds);
+ return true;
}
/*
void process_ip_header(struct context *c, unsigned int flags, struct buffer *buf);
-void schedule_exit(struct context *c, const int n_seconds, const int signal);
+bool schedule_exit(struct context *c);
static inline struct link_socket_info *
get_link_socket_info(struct context *c)
* */
if (c->options.mode == MODE_SERVER)
{
- schedule_exit(c, c->options.scheduled_exit_interval, SIGTERM);
+ if (!schedule_exit(c))
+ {
+ /* Return early when we don't need to notify management */
+ return;
+ }
}
else
{
void
send_auth_failed(struct context *c, const char *client_reason)
{
- if (event_timeout_defined(&c->c2.scheduled_exit))
+ if (!schedule_exit(c))
{
msg(D_TLS_DEBUG, "exit already scheduled for context");
return;
static const char auth_failed[] = "AUTH_FAILED";
size_t len;
- schedule_exit(c, c->options.scheduled_exit_interval, SIGTERM);
-
len = (client_reason ? strlen(client_reason)+1 : 0) + sizeof(auth_failed);
if (len > PUSH_BUNDLE_SIZE)
{
void
send_restart(struct context *c, const char *kill_msg)
{
- schedule_exit(c, c->options.scheduled_exit_interval, SIGTERM);
+ schedule_exit(c);
send_control_channel_string(c, kill_msg ? kill_msg : "RESTART", D_PUSH);
}