Re: locking patches for ufs_rename

[David Holland <dholland-tech%netbsd.org@localhost>]

On Wed, Sep 30, 2009 at 05:21:05PM +0300, Antti Kantee wrote:
 > > > Did any of these kind of changes make it into cvs already?  I
 > > > got a report offline about a panic in -current tmpfs_rename
 > > > where tvp was supplied but wasn't locked (the IMPLIES KASSERT
 > > > at the top of the routine).
 > > 
 > > No, and I haven't gone near tmpfs recently either. Is it reproduceable?
 > 
 > Apparently not, which made it all the more weirder.  Is there any chance
 > the new namei code could contain the same mistake under some conditions?
 > (yea, I know, an unlikely shot in the dark, but ..)

Well, given that the vnode in question goes straight from namei with
LOCKLEAF into VOP_RENAME, it would practically have to be in namei,
but I haven't changed anything that should make that possible. The
vnode is returned locked from vget, and namei doesn't do anything
except *unlock* it if LOCKLEAF *isn't* set, so there isn't much
opportunity to have broken it in that direction.

-- 
David A. Holland
dholland%netbsd.org@localhost