Re: vfs 3

"This is patch 3. Fix locking for root vnode in kernfs and procfs.
Remove a bogus VOP_UNLOCK from kernfs. This is meaningless without real locks, but it's less painful to fix this first. It also fixes a memory leak in procfs_umount. Only affects kernfs and procfs."

There may be room for improvement with this patch.

I think it's good to keep a pointer to the root vnode. But if we always ref it (procfs) we have to rele before calling flushing, which complicates the error path. kernfs does this in a better way I think, but the vnode might get reclaimed on us. The old way was to regenerate the vnode for each call of root(). Maybe we should stick with that, but make sure it's locked? I was trying to ensure two of the same vnode aren't made.

Patch as provided works well enough, but it's something to consider.

--


"The brave men who died in Vietnam, more than 100% of which were


black, were the ultimate sacrifice."
      - M. Barry, Mayor of Washington, DC