Re: devfs: panic on vrele() in nresolve / status of devfs

[Alex Hornung <ahornung@xxxxxxxxx>]

Oh dear, please disregard my question. I missed that vp->v_ops is a
struct vop_ops ** and not struct vop_ops *. 

On Sat, 2009-05-30 at 13:04 +0100, Alex Hornung wrote:
> Hi all,
> 
> I'm running again into problems related to referencing and locking of
> vnodes. 
> This time it's in nresolve on vrele(vp) where vp is a newly allocated
> vnode (or recycled if possible). vrele() causes a page fault when the
> vnode is a device node. 
> It doesn't cause any problems as far as I've seen with any other type
> (directory, symlink). I can't quite understand what could be causing
> this.
> 
> devfs_nresolve() can be found in devfs_vnops.c and devfs_allocv() can be
> found in devfs_core.c. If anyone can offer me his insight on what could
> be causing this problem I'd appreciate it.
> 
> My current code is at:
> http://gitweb.dragonflybsd.org/~alexh/dragonfly.git/tree/aae02cbf0350ca50b64b381042f7eea01a44c1bc:/sys/vfs/devfs
> 
> 
> About devfs status:
> While I'm still dealing with the above panic and the code isn't cleaned
> up and there might still be some locking issues, devfs already creates
> devices dynamically, provides an automated cleanup for empty directories
> and aliases (links) to devices that have been removed, and allows to
> create symlinks. Right now I'm working on getting the interface to the
> device itself to work (hence the above panic which occurs doing a
> cat /mnt/null > test  or for that matter any other device node).
> 
> 
> Thank you in advance,
> 
> Alex Hornung
>