[PATCH] Subject: [PATCH] xfs: fix sgid inheritance for subdirectories inheriting default acls [V3]

[Dave Chinner]

On Mon, Jul 08, 2013 at 04:11:21PM -0500, Ben Myers wrote:
> On Fri, Jun 21, 2013 at 02:45:53PM -0300, Carlos Maiolino wrote:
> > XFS removes sgid bits of subdirectories under a directory containing a default
> > acl.
> > 
> > When a default acl is set, it implies xfs to call xfs_setattr_nonsize() in its
> > code path. Such function is shared among mkdir and chmod system calls, and
> > does some checks unneeded by mkdir (calling inode_change_ok()). Such checks
> > remove sgid bit from the inode after it has been granted.
> > 
> > With this patch, we extend the meaning of XFS_ATTR_NOACL flag to avoid these
> > checks when acls are being inherited (thanks hch).
> > 
> > Also, xfs_setattr_mode, doesn't need to re-check for group id and capabilities
> > permissions, this only implies in another try to remove sgid bit from the
> > directories. Such check is already done either on inode_change_ok() or
> > xfs_setattr_nonsize().
> > 
> > Changelog:
> > 
> > V2: Extends the meaning of XFS_ATTR_NOACL instead of wrap the tests into another
> >     function
> > 
> > V3: Remove S_ISDIR check in xfs_setattr_nonsize() from the patch
> > 
> > Signed-off-by: Carlos Maiolino <cmaiolino at redhat.com>
> >  
> > -	if (mp->m_flags & XFS_MOUNT_RDONLY)
> > -		return XFS_ERROR(EROFS);
> > +	/* If acls are being inherited, we already have this checked */
> > +	if (!(flags & XFS_ATTR_NOACL)) {
> > +		if (mp->m_flags & XFS_MOUNT_RDONLY)
> > +			return XFS_ERROR(EROFS);
> >  
> > -	if (XFS_FORCED_SHUTDOWN(mp))
> > -		return XFS_ERROR(EIO);
> > +		if (XFS_FORCED_SHUTDOWN(mp))
> > +			return XFS_ERROR(EIO);
> >  
> > -	error = -inode_change_ok(inode, iattr);
> > -	if (error)
> > -		return XFS_ERROR(error);
> > +		error = -inode_change_ok(inode, iattr);
> > +		if (error)
> > +			return XFS_ERROR(error);
> > +	}
> 
> I'm not so sure about this change yet.  Looks like the two relevant callers are:
> 
> .set - xattr_handler
>   xfs_xattr_acl_set
>     xfs_set_mode
>       xfs_setattr_nonsize(..., XFS_ATTR_NOACL);
> 
> and
> 
> xfs_vn_mknod
>   xfs_inherit_acl
>     xfs_set_mode
>       xfs_setattr_nonsize(..., XFS_ATTR_NOACL);
> 
> I suggest moving the forced shutdown and readonly checks outside of the
> XFS_ATTR_NOACL conditional.  I'm not seeing those checks in xfs_attr_acl_set or
> xfs_vn_mknod and it won't hurt to be careful.

In both cases, the read-only checks are done at much higher layers
and so we don't ever get to xfs_setattr_nonsize() through these
paths with a read-only filesystem. Shutdown doesn't really matter -
the transaction commit will fail if the filesystem is shut down...

> It also seems like inode_change_ok might have some other checks that are
> necessary to determine whether it is ok to update the mode and ctime here.  A
> call to inode_owner_or_capable as is done in inode_change_ok would cover this
> possibility.

The inode permission checks are already done by xfs_xattr_acl_set():

	if ((current_fsuid() != inode->i_uid) && !capable(CAP_FOWNER))
		return -EPERM;

and in the case of xfs_inherit_acl() the user has just created the
file so they - by definition - have permission to inherit the ACL
and modify the mode of the inode they just created.

So there is no need for changes to inode_change_ok() here.

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com