On Thu, 24 Feb 2022, Jan Kara wrote:
On Wed 23-02-22 09:48:26, Byron Stanoszek wrote:
On Wed, 23 Feb 2022, Dave Chinner wrote:
On Tue, Feb 22, 2022 at 11:04:08AM +0100, Jan Kara wrote:
Hello!
On Sun 20-02-22 12:13:04, Matthew Wilcox wrote:
Keeping reiserfs in the tree has certain costs. For example, I would
very much like to remove the 'flags' argument to ->write_begin. We have
the infrastructure in place to handle AOP_FLAG_NOFS differently, but
AOP_FLAG_CONT_EXPAND is still around, used only by reiserfs.
Looking over the patches to reiserfs over the past couple of years, there
are fixes for a few syzbot reports and treewide changes. There don't
seem to be any fixes for user-spotted bugs since 2019. Does reiserfs
still have a large install base that is just very happy with an old
stable filesystem? Or have all its users migrated to new and exciting
filesystems with active feature development?
We've removed support for senescent filesystems before (ext, xiafs), so
it's not unprecedented. But while I have a clear idea of the benefits to
other developers of removing reiserfs, I don't have enough information to
weigh the costs to users. Maybe they're happy with having 5.15 support
for their reiserfs filesystems and can migrate to another filesystem
before they upgrade their kernel after 5.15.
Another possibility beyond outright removal would be to trim the kernel
code down to read-only support for reiserfs. Most of the quirks of
reiserfs have to do with write support, so this could be a useful way
forward. Again, I don't have a clear picture of how people actually
use reiserfs, so I don't know whether it is useful or not.
NB: Please don't discuss the personalities involved. This is purely a
"we have old code using old APIs" discussion.
So from my distro experience installed userbase of reiserfs is pretty small
and shrinking. We still do build reiserfs in openSUSE / SLES kernels but
for enterprise offerings it is unsupported (for like 3-4 years) and the module
is not in the default kernel rpm anymore.
So clearly the filesystem is on the deprecation path, the question is
whether it is far enough to remove it from the kernel completely. Maybe
time to start deprecation by printing warnings when reiserfs gets mounted
and then if nobody yells for year or two, we'll go ahead and remove it?
Yup, I'd say we should deprecate it and add it to the removal
schedule. The less poorly tested legacy filesystem code we have to
maintain the better.
Along those lines, I think we really need to be more aggressive
about deprecating and removing filesystems that cannot (or will not)
be made y2038k compliant in the new future. We're getting to close
to the point where long term distro and/or product development life
cycles will overlap with y2038k, so we should be thinking of
deprecating and removing such filesystems before they end up in
products that will still be in use in 15 years time.
And just so everyone in the discussion is aware: XFS already has a
deprecation and removal schedule for the non-y2038k-compliant v4
filesystem format. It's officially deprecated right now, we'll stop
building kernels with v4 support enabled by default in 2025, and
we're removing the code that supports the v4 format entirely in
2030.
For what it's worth, I have a number of production servers still using
Reiserfs, which I regularly maintain by upgrading to the latest Linux kernel
annually (mostly to apply security patches). I figured this filesystem would
still be available for several more years, since it's not quite y2038k yet.
I originally installed Reiserfs on these systems as early as 2005 due to the
tail-packing feature, which saved space with many small files on older
harddrives. Since then, I witnessed the development of ext4, and then btrfs.
For a long time, these newer filesystems had occasional reports of instabilities
and lost data, and so I shied away from using them. Meanwhile, Reiserfs reached
a level of maturity and no longer had active development on it, except for the
occasional bugfix. I felt this was a filesystem I could trust going forward
(despite its relative slowness), even after popular Linux distributions
eventually dropped it from being installed by default.
I have only recently begun to use XFS on newer installs, only since the XFS
developers added bigtime support for y2038k. But for existing installs, I ask
that we keep Reiserfs supported in the kernel a little longer. Perhaps use the
same deprecation schedule that was picked for XFS v4 (roughly 10 years of
deprecation before eventual removal)?
Thanks for letting us know about your usage! Frankly the reality of
reiserfs is that it gets practically no development time and little
testing. Now this would not be a big problem on its own because what used
to work should keep working but the rest of the common filesystem
infrastructure keeps moving (e.g. with Matthew's page cache changes, new
mount API, ...) and so it can happen that with a lack of testing &
development reiserfs will break without us noticing. So I would not
consider reiserfs a particularly safe choice these days and rather consider
migration to some other filesystem. Originally I thought about 2 years
deprecation period but if 5 years make things significantly easier for you
(here I have to admit I don't have experience with maintaining larger fleet
of servers and how much effort it takes to move it to a different fs) we can
live with that I guess.
I understand the necessity to drop uncommon or unmaintained subsystems in order
to keep moving forward with code development. And thanks for the warning
regarding reiserfs not being a particularly safe choice these days.
For my specific use case, I can probably squeeze by with 2 years of
deprecation, but I would feel more comfortable with having 3. Thanks for your
consideration!
Regards,
-Byron
