Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221201160724.2593341-1-cccheng@synology.com>
 <CAL3q7H5oet2P9XijTtzPo3joZWdoa3OuD9L-wK9nTEFya2PY8w@mail.gmail.com> <CAHuHWtnbVsS2pp5EySmZ_72fCrDqKJTAOkssa-D-X5wKoR9uWQ@mail.gmail.com>
In-Reply-To: <CAHuHWtnbVsS2pp5EySmZ_72fCrDqKJTAOkssa-D-X5wKoR9uWQ@mail.gmail.com>
From:   Filipe Manana <fdmanana@kernel.org>
Date:   Tue, 6 Dec 2022 11:14:56 +0000
Message-ID: <CAL3q7H5+aVYC-nQaX4_F=eRPpU8xRNEE73MHy-_fdoQFwNMDsA@mail.gmail.com>
Subject: Re: [PATCH] btrfs: refuse to remount read-write with unsupported
 compat-ro features
To:     Chung-Chiang Cheng <shepjeng@gmail.com>
Cc:     Chung-Chiang Cheng <cccheng@synology.com>, clm@fb.com,
        josef@toxicpanda.com, dsterba@suse.com,
        linux-btrfs@vger.kernel.org, linux-kernel@vger.kernel.org,
        kernel@cccheng.net, Johnny Chang <johnnyc@synology.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Dec 6, 2022 at 2:42 AM Chung-Chiang Cheng <shepjeng@gmail.com> wrote:
>
> On Mon, Dec 5, 2022 at 6:45 PM Filipe Manana <fdmanana@kernel.org> wrote:
> >
> > Wasn't this already done by the following commit?
> >
> > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=81d5d61454c365718655cfc87d8200c84e25d596
> >
> > Thanks.
> >
>
> Wow. I did not notice this commit doing the same job by Qu. But I have
> tested the latest linux-6.1 rc-7, and it's still able to mount a unsupported
> comat-ro btrfs as read-write via remount.
>
> It's caused by the follow-up commit d7f67ac9a928 ("btrfs: relax
> block-group-tree feature dependency checks"). This commit checks read-
> only with the current superblock, which will always pass in the situation
> remounting from read-only to read-write. It seems `btrfs_check_features()`
> cannot cover this scenario.
>
>         if (compat_ro_unsupp && !sb_rdonly(sb)) {
>                                 ^^^^^^^^^^^^^^

Yep, that's a bug.
btrfs_check_features() is called before the read only flag is updated
in the super block.

So the condition should be:

if (compat_ro_unsupp && sb_rdonly(sb) && we_want_to_transition_to_rw)

We need to pass the flags passed to btrfs_remount() to
btrfs_check_features() for that "we_want_to_transition_to_rw" check.

That seems to be what needs to be fixed.

Thanks.


>
> Thanks,
> C.C.Cheng