Re: locking crash using coda

["J. Hannken-Illjes" <hannken%eis.cs.tu-bs.de@localhost>]

> On 17 Jun 2016, at 05:12, Brett Lymn <blymn%internode.on.net@localhost> wrote:
> 
> 
> It looks like there is a problem with the locking when using coda with a
> -current updated a few hours ago.  If I try to list the /coda directory
> with venus (the userland daemon part of coda) then I get this panic:
> 
> panic: kernel diagnostic assertion "VOP_ISLOCKED(vp)" failed: file
> "../../../../ufs/ufs/ufs_readwrite.c", line 165
> fatal breakpoint trap in supervisor mode...
> 
> the handcrafted bt goes as follows:
> 
> ffs_bufrd()
> ufs_readdir()
> VOP_READDIR()
> coda_readdir()
> VOP_READDIR()
> vn_readdir()
> sys___getdents30()
> 
> Guessing that the coda readdir is not holding a lock it should when
> calling the underlying file store?

This assertion is from 2015/03/27, last change to coda was 2015/09/06.

Looks like coda nearly never locks vnodes before it calls vnode
operations using c_ovp.  This is a bug.

--
J. Hannken-Illjes - hannken%eis.cs.tu-bs.de@localhost - TU Braunschweig (Germany)