sig_analog: Fix channel leak when mwimonitor is enabled.
When mwimonitor=yes is enabled for an FXO port,
the do_monitor thread will launch mwi_thread if it thinks
there could be MWI on an FXO channel, due to the noise
threshold being satisfied. This, in turns, calls
analog_ss_thread_start in sig_analog. However, unlike
all other instances where __analog_ss_thread is called
in sig_analog, this call path does not properly set
pvt->ss_astchan to the Asterisk channel, which means
that the Asterisk channel is NULL when __analog_ss_thread
starts executing. As a result, the thread exits and the
channel is never properly cleaned up by calling ast_hangup.
This caused issues with do_monitor on incoming calls,
as it would think the channel was still owned even while
receiving events, leading to an infinite barrage of
warning messages; additionally, the channel would persist
improperly.
To fix this, the assignment is added to the call path
where it is missing (which is only used for mwi_thread).
A warning message is also added since previously there
was no indication that __analog_ss_thread was exiting
abnormally. This resolves both the channel leak and the
condition that led to the warning messages.
Resolves: #458
(cherry picked from commit c930230a73
)
This commit is contained in:
parent
72d631b7bd
commit
4e774da45a
|
@ -1764,10 +1764,7 @@ static void *__analog_ss_thread(void *data)
|
|||
|
||||
ast_debug(1, "%s %d\n", __FUNCTION__, p->channel);
|
||||
|
||||
if (!chan) {
|
||||
/* What happened to the channel? */
|
||||
goto quit;
|
||||
}
|
||||
ast_assert(chan != NULL);
|
||||
|
||||
if ((callid = ast_channel_callid(chan))) {
|
||||
ast_callid_threadassoc_add(callid);
|
||||
|
@ -2756,6 +2753,7 @@ int analog_ss_thread_start(struct analog_pvt *p, struct ast_channel *chan)
|
|||
{
|
||||
pthread_t threadid;
|
||||
|
||||
p->ss_astchan = chan;
|
||||
return ast_pthread_create_detached(&threadid, NULL, __analog_ss_thread, p);
|
||||
}
|
||||
|
||||
|
|
Loading…
Reference in New Issue