Re: [PATCH v2] generic/062: filter setfattr --restore symlink-safety warning

On Mon, Jul 06, 2026 at 09:13:26PM +0200, Avinesh Kumar wrote:
> From: Avinesh Kumar <[email protected]>
> 
> attr 2.6.0 (CVE-2026-54371 fix)[0] makes "setfattr --restore" warn when
> used without -P and the dump contains a symlink, which generic/062 does.
> The warning leaks into the compared output via the setfattr() wrapper
> and fails the test. Adding -P isn't portable (older attr doesn't have it),
> so filter the warning in the wrapper instead.
> 
> [0] https://cgit.git.savannah.nongnu.org/cgit/attr.git/commit/?id=3fb06b9ba314d37035d0877e6de313de754f1ac8

For generic/062, modifying the g/062 itself seems fine and logical, as
_extend_test_bed() explicitly introduces complex symlink paths for
testing purposes.

Have you run xfstests on XFS with attr 2.6.0? There are quite a few
setfattr --restore invocations in common/populate. If those also trigger
the new unsafe restore warnings, we might want to consider introducing
a helper function, such as _setfattr or _setfattr_restore to handle to
-h and -P things.

Reviewed-by: Zorro Lang <[email protected]>

> 
> Signed-off-by: Avinesh Kumar <[email protected]>
> ---
>  tests/generic/062 | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/tests/generic/062 b/tests/generic/062
> index 89659040..ddf0a478 100755
> --- a/tests/generic/062
> +++ b/tests/generic/062
> @@ -30,7 +30,12 @@ getfattr()
>  
>  setfattr()
>  {
> -    $SETFATTR_PROG $@ 2>&1 | _filter_scratch
> +    # attr >= 2.6.0 (CVE-2026-54371 fix) warns that "setfattr --restore" without
> +    # -P/--physical is unsafe because it can traverse symlinks. Older attr does
> +    # not accept -P, so just filter the warning to stay version-agnostic.
> +    $SETFATTR_PROG $@ 2>&1 | \
> +        sed -e '/^Warning: option --restore=file is unsafe without option/d' | \
> +        _filter_scratch
>  }
>  
>  _create_test_bed()
> -- 
> 2.54.0
>