Date:   Thu, 19 May 2022 18:40:33 +0800
From:   Zorro Lang <zlang@...hat.com>
To:     Lukas Czerner <lczerner@...hat.com>
Cc:     Eric Biggers <ebiggers@...nel.org>, fstests@...r.kernel.org,
        linux-fscrypt@...r.kernel.org, linux-ext4@...r.kernel.org
Subject: Re: [xfstests PATCH 0/2] update test_dummy_encryption testing in
 ext4/053

On Thu, May 19, 2022 at 10:33:22AM +0200, Lukas Czerner wrote:
> On Thu, May 19, 2022 at 12:47:01PM +0800, Zorro Lang wrote:
> > On Wed, May 18, 2022 at 03:01:08PM -0700, Eric Biggers wrote:
> > > Zorro, can you fix your email configuration?  Your emails have a
> > > Mail-Followup-To header that excludes you, so replying doesn't work correctly;
> > > I had to manually fix the recipients list.  If you're using mutt, you need to
> > > add 'set followup_to = no' to your muttrc.
> > 
> > Oh, I didn't notice that, I use neomutt, it might enable the followup_to by
> > default. OK, I've set followup_to = no, and restart my neomutt. Hope it helps:)
> > 
> > > 
> > > On Thu, May 19, 2022 at 02:16:07AM +0800, Zorro Lang wrote:
> > > > > > 
> > > > > > And I saw some discussion under this patchset, and no any RVB, so I'm wondering
> > > > > > if you are still working/changing on it?
> > > > > > 
> > > > > 
> > > > > I might add a check for kernel version >= 5.19 in patch 1.  Otherwise I'm not
> > > > > planning any more changes.
> > > > 
> > > > Actually I don't think the kernel version check (in fstests) is a good method. Better
> > > > to check a behavior/feature directly likes those "_require_*" functions.
> > > > 
> > > > Why ext4/053 need >=5.12 or even >=5.19, what features restrict that? If some
> > > > features testing might break the garden image (.out file), we can refer to
> > > > _link_out_file(). Or even split this case to several small cases, make ext4/053
> > > > only test old stable behaviors. Then use other cases to test new features,
> > > > and use _require_$feature_you_test for them (avoid the kernel version
> > > > restriction).
> > > 
> > > This has been discussed earlier in this thread as well as on the patch that
> > > added ext4/053 originally.  ext4/053 has been gated on version >= 5.12 since the
> > > beginning.  Kernel version checks are certainly bad in general, but ext4/053 is
> > > a very nit-picky test intended to detect if anything changed, where a change
> > > does not necessarily mean a bug.  So maybe the kernel version check makes sense
> > 
> > Even on old RHEL-8 system (with a variant of kernel 3.10), the ext4/053 fails
> > as [1]. Most of mount options test passed, only a few options (inlinecrypt,
> > test_dummy_encryption, prefetch_block_bitmaps, dioread_lock) might not be
> > supported.
> 
> No it does not. On RHEL-8 system the test will not run because of kernel
> version test. It will be skipped.

Yes, it will be skipped, I just ran it by removing that "kernel_gte 5.12" line :)

> 
> > 
> > I think it's not necessary to mix all old and new ext4 mount options test into
> > one single test cause. If it's too complicated, we can move some functions into
> > common/ext4 (or others you like), split ext4/053 to several cases. Let ext4/053
> > test stable enough mount option (supported from an old enough kernel). Then let
> > other newer mount options in different single cases.
> > 
> > For example, make those CONFIG_FS_ENCRYPTION* tests into a seperated case,
> > and add something likes require_(fs_encryption?), and src/feature.c can be
> > used too. Then about dioread_lock and prefetch_block_bitmaps things, we can
> > deal with them specially, or split them out from ext4/053. I even don't mind
> > if you test ext2 and ext3/4 in separate case.
> 
> Sure, but why to split it? It all should be stable enough, it's user
> facing interface, that's the whole point of the test. I certainly see
> the benefit of having the test for all ext4 mount option in one test -
> it's faster and it's easier to see what's there. I would be agains
> splitting it.

OK, although you can have a 'group name' to help to run all ext4 mount options
regression test, but as I said: "as it's an ext4 specific testing, I respect
the opinion from ext4 list particularly", so I won't touch this case, if you
against :)

> 
> As it is now there is only one kernel_gte() check to avoid testing the
> entire history. With any new mount option as a separate test we would
> still need kernel_gte test to avoid failing on kernels that don't have
> the mount option. At least until kernel gains ability to list supported
> mount options it's the only test we have.
> 
> On the other hand I do see some value in making a new test for a new
> mount option. But I don't have a strong opinion about that.
> 
> As for the original topic of the discussion, as I said in previous
> reply, maybe the right solution here is to treat the change as a bug fix
> which is arguably is and let it fail on old behavior.
> 
> Thanks!
> -Lukas
> 
> > 
> > That's my personal opinion, I can try to help to do that after merging this
> > patchset, if ext4 forks agree and would like to give me some supports
> > (review and Q&A). Anyway, as it's an ext4 specific testing, I respect the
> > opinion from ext4 list particularly.
> > 
> > [1]
> > +SHOULD FAIL remounting ext2 "commit=7" (remount unexpectedly succeeded) FAILED
> > +mounting ext2 "test_dummy_encryption=v1" (failed mount) FAILED
> > +mounting ext2 "test_dummy_encryption=v2" (failed mount) FAILED
> > +mounting ext2 "test_dummy_encryption=v3" (failed mount) FAILED
> > +mounting ext2 "inlinecrypt" (failed mount) FAILED
> > +mounting ext2 "prefetch_block_bitmaps" (failed mount) FAILED
> > +mounting ext2 "no_prefetch_block_bitmaps" (failed mount) FAILED
> > +mounting ext3 "test_dummy_encryption=v1" (failed mount) FAILED
> > +mounting ext3 "test_dummy_encryption=v2" (failed mount) FAILED
> > +mounting ext3 "test_dummy_encryption=v3" (failed mount) FAILED
> > +mounting ext3 "inlinecrypt" (failed mount) FAILED
> > +mounting ext3 "prefetch_block_bitmaps" (failed mount) FAILED
> > +mounting ext3 "no_prefetch_block_bitmaps" (failed mount) FAILED
> > +mounting ext4 "nodioread_nolock" (failed mount) FAILED
> > +mounting ext4 "dioread_lock" checking "nodioread_nolock" (not found) FAILED
> > +mounting ext4 "test_dummy_encryption=v1" (failed mount) FAILED
> > +mounting ext4 "test_dummy_encryption=v2" (failed mount) FAILED
> > +mounting ext4 "test_dummy_encryption=v3" (failed mount) FAILED
> > +mounting ext4 "inlinecrypt" (failed mount) FAILED
> > +mounting ext4 "prefetch_block_bitmaps" (failed mount) FAILED
> > +mounting ext4 "no_prefetch_block_bitmaps" (failed mount) FAILED
> > 
> > > there.  Lukas, any thoughts about the issues you encountered when running
> > > ext4/053 on older kernels?
> > > 
> > > If you don't want a >= 5.19 version check for the test_dummy_encryption test
> > > case as well, then I'd rather treat the kernel patch
> > > "ext4: only allow test_dummy_encryption when supported" as a bug fix and
> > > backport it to the LTS kernels.  The patch is fixing the mount option to work
> > > the way it should have worked originally.  Either that or we just remove the
> > > test_dummy_encryption test case as Ted suggested.
> > 
> > Oh, I'd not like to push anyone to do more jobs:) And there're many Linux
> > distributions with their downstream kernel, not only LTS kernel project.
> > So I don't mean to make fstests' cases support the oldest existing kernel
> > version, just hope some common cases try not *only* work for the latest
> > one, if they have the chance :)
> > 
> > Thanks,
> > Zorro
> > 
> > > 
> > > - Eric
> > > 
> > 
> 

Powered by blists - more mailing lists