On Tue, May 02, 2023 at 06:23:51PM +0000, Chuck Lever III wrote:

quoted

On May 2, 2023, at 12:50 PM, Chuck Lever III [off-list ref] wrote:

quoted

On May 2, 2023, at 9:49 AM, Chuck Lever III [off-list ref] wrote:

quoted

On May 2, 2023, at 9:36 AM, Christian Brauner [off-list ref] wrote:

We've aligned setgid behavior over multiple kernel releases. The details
can be found in commit cf619f891971 ("Merge tag 'fs.ovl.setgid.v6.2' of
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/idmapping") and
commit 426b4ca2d6a5 ("Merge tag 'fs.setgid.v6.0' of
git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux").
Consistent setgid stripping behavior is now encapsulated in the
setattr_should_drop_sgid() helper which is used by all filesystems that
strip setgid bits outside of vfs proper. Usually ATTR_KILL_SGID is
raised in e.g., chown_common() and is subject to the
setattr_should_drop_sgid() check to determine whether the setgid bit can
be retained. Since nfsd is raising ATTR_KILL_SGID unconditionally it
will cause notify_change() to strip it even if the caller had the
necessary privileges to retain it. Ensure that nfsd only raises
ATR_KILL_SGID if the caller lacks the necessary privileges to retain the
setgid bit.

Without this patch the setgid stripping tests in LTP will fail:

quoted

As you can see, the problem is S_ISGID (0002000) was dropped on a
non-group-executable file while chown was invoked by super-user, while

[...]

quoted

fchown02.c:66: TFAIL: testfile2: wrong mode permissions 0100700, expected 0102700

[...]

quoted

chown02.c:57: TFAIL: testfile2: wrong mode permissions 0100700, expected 0102700

With this patch all tests pass.

Reported-by: Sherry Yang <redacted>
Signed-off-by: Christian Brauner <brauner@kernel.org>

There are some similar fstests failures this fix might address.

I've applied this patch to the nfsd-fixes tree for broader
testing.

ERROR: modpost: "setattr_should_drop_sgid" [fs/nfsd/nfsd.ko] undefined!

Did I apply this patch to the wrong kernel?

setattr_should_drop_sgid() is not available to callers built as
modules. It needs an EXPORT_SYMBOL or _GPL.

Hey Chuck,

Thanks for taking a look!
The required export is part of
commit 4f704d9a835 ("nfs: use vfs setgid helper")
which is in current mainline as of Monday this week which was part of my
v6.4/vfs.misc PR that was merged.

So this should all work fine on mainline. Seems I didn't use --base
which is why that info was missing.

Thanks!
Christian