Re: mount -o extattr (extended attributes)

[Steven Bellovin <smb%cs.columbia.edu@localhost>]

On Jun 16, 2011, at 3:22 29PM, Emmanuel Dreyfus wrote:

> Thor Lancelot Simon <tls%panix.com@localhost> wrote:
> 
>> But since the extended attribute database isn't validated by fsck, the
>> kernel can still suddenly crash at that point, no?
> 
> I did not observe that with -o extattr, at least yet. The EA code makes
> some efforts to sanity check what it finds. I think my problems were
> caused by reading file from a dirty filesystem before fsck.
> 
But can it happen?  Is the kernel code that deals with this database 
sufficiently robust that it *cannot* crash if it reads garbage?  Do
we need an fsckeattr program?  Is there the potential for semantic
errors in using this data if a piece of an attribute is missing
(perhaps because of the write ordering that Thor mentioned)?


                --Steve Bellovin, https://www.cs.columbia.edu/~smb