When a PJSIP endpoint is configured with set_var invoking a dialplan
function (e.g. PJSIP_HEADER(add,...)), chan_pjsip_new() calls
pbx_builtin_setvar_helper() while holding the channel lock.
For function-style variables, this dispatches to ast_func_write()
which, in the case of PJSIP_HEADER, calls
ast_sip_push_task_wait_serializer() -- blocking synchronously while
the channel lock is held.
If a concurrent operation (ARI, AMI, rtp_check_timeout) traverses
the channels container via ast_channel_get_by_name(), it acquires
the container lock then tries to lock individual channels in the
iteration callback (by_uniqueid_cb/by_name_cb). When the serializer
thread also needs the container lock, a circular dependency forms:
channel_lock -> serializer_wait -> container_lock -> channel_lock
This causes a complete Asterisk freeze. In the observed case, 36
threads were blocked on the container lock until res_freeze_check
triggered SIGABRT after its 30-second timeout.
Unlock the channel before iterating endpoint channel_vars so that
dialplan functions can block without holding the channel lock. Re-lock
the channel for ast_channel_stage_snapshot_done() so the batched
snapshot is published under lock and captures the full channel state
including the variables set during the loop.
Fixes: #1872
ast_channel_zone_set(chan, zone);
}
+ ast_channel_stage_snapshot_done(chan);
+ ast_channel_unlock(chan);
+
for (var = session->endpoint->channel_vars; var; var = var->next) {
char buf[512];
pbx_builtin_setvar_helper(chan, var->name, ast_get_encoded_str(
var->value, buf, sizeof(buf)));
}
- ast_channel_stage_snapshot_done(chan);
- ast_channel_unlock(chan);
-
set_channel_on_rtp_instance(session, ast_channel_uniqueid(chan));
SCOPE_EXIT_RTN_VALUE(chan);