Received-SPF: pass (google.com: domain of linux-kernel+bounces-104055-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Fri, 15 Mar 2024 00:41:09 -0400
From: Kent Overstreet <kent.overstreet@linux.dev>
To: linux-bcachefs@vger.kernel.org, linux-fsdevel@vger.kernel.org, 
	linux-kernel@vger.kernel.org
Subject: bcachefs: do not run 6.7: upgrade to 6.8 immediately if you have a
 multi device fs
Message-ID: <muwlfryvafsskt2l2hgv3szwzjfn7cswmmnoka6zlpz2bxj6lh@ugceww4kv3jr>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

there's a bug in 6.7 with filesystems that are mid upgrade and then get
downgraded not getting marked in the superblock as downgraded, and this
translates to a really horrific bug in splitbrain detection when the old
version isn't updating member sequence nmubers and you go back to the
new version - this results in every device being kicked out of the fs.

and our backports are not being picked up by the stable team, so - do
not run 6.7, switch to 6.8 immediately, running 6.7 with new -tools will
trigger it.

if you are affected:

 - 6.9 (once Linus merges) will have a new no_splitbrain_check option,
   which runs the splitbrain checks in dry mode and won't kick your
   devices out

 - we have new repair code landing soon that can recover from
   missing/unreadable btree roots by scanning the entire device(s) for
   btree nodes (which, fortunately, we have sufficient metadata in btree
   node headers to do safely; reiserfs famously did not). i've seen some
   crazy corruption resulting from this, but it might still be
   recoverable