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

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@redhat.com>
Reviewed-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>

fs/xfs/xfs_iops.c

@@ -467,9 +467,6 @@ xfs_setattr_mode(
 	ASSERT(tp);
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 
-	if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
-		mode &= ~S_ISGID;
-
 	ip->i_d.di_mode &= S_IFMT;
 	ip->i_d.di_mode |= mode & ~S_IFMT;
 
@@ -495,6 +492,8 @@ xfs_setattr_nonsize(
 
 	trace_xfs_setattr(ip);
 
+	/* 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);
 
@@ -504,6 +503,7 @@ xfs_setattr_nonsize(
 		error = -inode_change_ok(inode, iattr);
 		if (error)
 			return XFS_ERROR(error);
+	}
 
 	ASSERT((mask & ATTR_SIZE) == 0);