Frank Filz

Yea, but what do we do for 4.x? I'm tempted to go ahead and remove cacheinode for 4.1 even though it's technically too late. But we either come up with...

OK, let's take advantage that none of the actively involved Ganesha vendors have picked up V4.x yet and allow some things into V4.1 that we would not normally take in...

I think the problem is that state_complete_grant doesn't hold an obj reference. If so, state_release_grant also needs to take an obj reference. HOWEVER, I think we have an obj refcount...

I've done a bit more looking. I don't see the refcount issue yet. The problem with state_complete_grant not taking an obj reference is that free_cookie can result in all obj...

The issue COULD instead be a lock entry refcount issue.

OK, so fsal_obj_handle_fini() has been called on the object (that will zero out all the fields that are zero above). I'm pretty sure this is a refcount issue and the...

I think this should fix the issue. https://review.gerrithub.io/c/ffilz/nfs-ganesha/+/541117

Ah good catch. I was going to add an object reference to the cookie also because I see an actual problem where we might reference sce_obj after all references have...

OK, I've done some auditing. There are definitely some problems with obj references. I think actually the references associated with lock_list should not be there. They originally were pin references...

OK, fixed up a bunch of refcount issues that I think will help. Please download the patch series with this as the top: https://review.gerrithub.io/c/ffilz/nfs-ganesha/+/541256/1 This will pull all the patches...