Re: [PATCH] NFSD: Checking whether kill_suid by should_remove_suid()

[Kinglong Mee <kinglongmee@gmail.com>]

On 2014/4/18 21:02, J. Bruce Fields wrote:
> On Sun, Apr 13, 2014 at 11:11:39PM +0800, Kinglong Mee wrote:
>> As local filesystem, writing data to the file by non-owner will
>> clears the SUID+SGID, owner will not.
>
> Are you sure about this?  (Do you have a test case that fails?)
>
> I don't see an owner check in should_remove_suid.
>
> And I think that an nfsd thread will always have CAP_FSETID set (see
> cap_raise_nfsd_set and the definition of CAP_NFSD_SET), so that
> should_remove_suid() will always return 0.

You are right, should_remove_suid always return 0, nfsd will never call
kill_suid(). Coincidentally, that's the fix for bug of root clears the
SUID/SGID after writing data.

The right fix should drops the kill_suid(), because vfs_writev() have
do it correctly. I have push a new patch.

thanks,
Kinglong Mee