I noticed that sss_dp_get_account_int_done() would free sdp_req but then the callback destructor would dereference sdp_req as well. When the destructor is called, sdp_req is already gone.
It looks that the callback is not needed after it has called the request it links to so I think we can just free it and let the destructor remove it from the list.
On Fri, 2011-12-16 at 10:31 +0100, Jakub Hrozek wrote:
I noticed that sss_dp_get_account_int_done() would free sdp_req but then the callback destructor would dereference sdp_req as well. When the destructor is called, sdp_req is already gone.
It looks that the callback is not needed after it has called the request it links to so I think we can just free it and let the destructor remove it from the list.
It took me a while to puzzle out what the problem was here. The issue is that the sdp_req is always freed just after the callbacks with the results are sent back. However, the callback contexts are not allocated beneath the sdp_req (because they need to be cancellable if the client disconnects). So what would happen is that they would hang around until the client eventually freed them, which would always happen *after* the sdp_req had been freed.
This has been working only because we were lucky and the memory for sdp_req was not being reclaimed.
Ack (and good catch!)
On Fri, 2011-12-16 at 14:12 -0500, Stephen Gallagher wrote:
On Fri, 2011-12-16 at 10:31 +0100, Jakub Hrozek wrote:
I noticed that sss_dp_get_account_int_done() would free sdp_req but then the callback destructor would dereference sdp_req as well. When the destructor is called, sdp_req is already gone.
It looks that the callback is not needed after it has called the request it links to so I think we can just free it and let the destructor remove it from the list.
It took me a while to puzzle out what the problem was here. The issue is that the sdp_req is always freed just after the callbacks with the results are sent back. However, the callback contexts are not allocated beneath the sdp_req (because they need to be cancellable if the client disconnects). So what would happen is that they would hang around until the client eventually freed them, which would always happen *after* the sdp_req had been freed.
This has been working only because we were lucky and the memory for sdp_req was not being reclaimed.
Ack (and good catch!)
Pushed to master.
sssd-devel@lists.fedorahosted.org
-
Jakub Hrozek -
Stephen Gallagher