> I just committed a patch that fixes a fencepost error in the fdesc
> file system ...  and will likely also fix the nullfs problem someone
> reported

If you're refering to the "kern/337: rename can cause kernel panic on a nullfs"
problem that I submitted last week, it still occurs in exactly the same way in
a kernel compiled with "fdesc_vnops.c,v 1.15.2.1 1994/07/15 21:42:51".

I'm sure the problem is caused by the rename call passing a NULL vnode to the
nullfs.  Initially I wasn't sure whether it was a bug in the rename call in
that it shouldn't do that, or in the nullfs in that it should expect it.  But
now I'm tending towards the latter.  When renaming to a non-existent file, it
doesn't seem unreasonable that the rename system call will pass a NULL vnode
pointer for the target file to the lower layer.  Certainly the ufs_rename code
seems to expect that, so presumably the nullfs layer should too.

Duncan

------------------------------------------------------------------------------