> > doesn't "goto kthread_end" leak vnx and bp?
> 
> Yes. I think at this point we should just wait for the buffer to complete I/O
> (that it, remove the VNF_VUNCONF check at this place, and let it
> complete the for (resid = bp->b_resid; resid; resid -= sz) loop).
> Does this sounds good ?

yes, the fix you've committed looks reasonable.  thanks.

YAMAMOTO Takashi