Btrfs support coming to AlmaLinux 10.1
The AlmaLinux project has announced that the upcoming 10.1 release will include support for Btrfs:
Btrfs support encompasses both kernel and userspace enablement, and it is now possible to install AlmaLinux OS with a Btrfs filesystem from the very beginning. Initial enablement was scoped to the installer and storage management stack, and broader support within the AlmaLinux software collection for Btrfs features is forthcoming.
Btrfs support in AlmaLinux OS did not happen in isolation. This was proposed and scoped in RFC 0005, and has been built upon prior efforts by the Fedora Btrfs SIG in Fedora Linux and the CentOS Hyperscale SIG in CentOS Stream.
AlmaLinux OS is designed to be binary compatible with Red Hat Enterprise Linux (RHEL); Btrfs, however, has never been supported in RHEL. A technology preview of Btrfs in RHEL 6 and 7 ended with the filesystem being dropped from RHEL 8 and onward. AlmaLinux OS 10.1 is currently in beta.
Posted Oct 23, 2025 15:01 UTC (Thu)
by rsidd (subscriber, #2582)
[Link] (42 responses)
Yes, I know the legal status debate. I find Ubuntu's position completely logical and nobody has challenged it in nearly a decade: OpenZFS is not a derived work of Linux (it works on FreeBSD and IllumOS) and obviously Linux is not a derived work of OpenZFS. Loading the kernel module is "mere aggregation". A decade after Ubuntu laid the legal argument out, it's time it was accepted.
This is philosophically, though maybe not legally, different from the loadable proprietary binary module debate. OpenZFS is free software. It is ironic that FreeBSD considers it free enough to include by default, though it is more restrictive than the BSD licence; while most Linux distros don't include it because of the GPL interaction worry, though it is less restrictive than the GPL.
Anyway, how does an enterprise distro like RHEL bring its filesystem into the 21st century? Maybe support bcachefs and risk the maintainer's human interaction issues?
Posted Oct 23, 2025 15:05 UTC (Thu)
by bobolopolis (subscriber, #119051)
[Link] (1 responses)
No idea how well it works, I haven't had time to really play with it. We're mostly using btrfs and have been happy with it for years.
Posted Oct 23, 2025 22:02 UTC (Thu)
by jhoblitt (subscriber, #77733)
[Link]
Posted Oct 23, 2025 17:49 UTC (Thu)
by raven667 (subscriber, #5198)
[Link]
Posted Oct 23, 2025 18:35 UTC (Thu)
by koverstreet (✭ supporter ✭, #4296)
[Link]
Things have gotten a lot calmer since parting ways from the kernel :)
I've found interacting with the distros directly to be drastically less contentious...
Posted Oct 23, 2025 18:37 UTC (Thu)
by tuna (guest, #44480)
[Link] (21 responses)
It is accepted until Oracle's lawyers change their mind and sue IBM and all their customers.
Posted Oct 24, 2025 9:48 UTC (Fri)
by paulj (subscriber, #341)
[Link] (7 responses)
Posted Oct 24, 2025 11:45 UTC (Fri)
by khim (subscriber, #9252)
[Link] (6 responses)
Oracle has code in kernel, too and it may, in a very persuading terms claim that if it wanted these two projects to be used together then it would have released them on compatible terms. It haven't don't so —because it obviously expected that people who wanted ZFS would either buy Solaris or use *BSD (which helps people to deal with non-Linux OSes). I have no idea how plausibly that approach would look in court and I don't want to try to find out. Maybe with some other company I would have risked that, but with Oracle… no.
Posted Oct 24, 2025 12:50 UTC (Fri)
by paulj (subscriber, #341)
[Link] (5 responses)
Note carefully: The Linux kernel copyright holders could collectively fix this by making it clear they are fine with Linux being combined with ZFS. They havn't done so (to echo your words about Oracle). Admittedly, it would be a lot easier for Oracle to grant a GPL licence for ZFS, than getting all the Linux copyright holders to come together...
AFAICT Oracle have never distributed ZFS for Oracle Linux, however it is possible to install ZFS on Oracle Linux from third party sources. Though, even if they did, the issue is not that Oracle alone could sue but /any Linux copyright holder/.
Posted Oct 24, 2025 13:47 UTC (Fri)
by khim (subscriber, #9252)
[Link] (4 responses)
You are forgetting one important thing: Oracle is a copyright holder of upstream Linux, too. And because Oracle is copyright owner for both Linux and ZFS it has a pretty high chance of arguing that it intented to keep ZFS out of Linux: after all the only thing it needs to make it available under GPL terms is one press-release. That's why getting all the Linux copyright holders to come together is not needed: Oracle is in that group and it can make ZFS available for Linux pretty trivially, why try anything else, if Oracle agreement is needed anyway?
Posted Oct 24, 2025 14:26 UTC (Fri)
by rsidd (subscriber, #2582)
[Link]
So, in terms of "intent", the question is not what Oracle wants today, but what the CDDL permits free software users to do with OpenZFS which is a continuation from the last free-software version of ZFS.
FreeBSD has decided, years ago, that they can include OpenZFS in their OS. Ubuntu has decided, years ago, that they can do the same as a loadable kernel module. I'm not sure Oracle's actions permit them to selectively claim the Linux version is disallowed based on "intent" when their intent has been exhibited to far exceed this particular interdiction.
Posted Oct 24, 2025 14:45 UTC (Fri)
by paulj (subscriber, #341)
[Link]
> this isn't a concern about Oracle particularly, but _any_ kernel copyright holder -
The "Not just X, but any Y" construct naturally implies that X is a Y. That Oracle is a Linux copyright holder, and that it is Linux copyright holders from whom there is any legal peril, is the entire thrust of my comment. ;)
Posted Oct 24, 2025 15:19 UTC (Fri)
by koverstreet (✭ supporter ✭, #4296)
[Link]
I think for that argument to stick they'd need to actually make some sort of statement. Given that they employee a bunch of people who work on the kernel, there's no way they could plausibly say after the fact "oops, we didn't want that!" - they could very easily make themselves heard if they wanted to.
Posted Oct 24, 2025 18:09 UTC (Fri)
by Cyberax (✭ supporter ✭, #52523)
[Link]
The next month, in a lawsuit filed by RedHat: "Your Honor, the defendant admits that the ZFS module that they include in their Unbreakable Linux can not be treated under the 'mere aggregation' clause of GPL. Please see the copies of the court documents filed by Oracle".
Posted Oct 24, 2025 10:07 UTC (Fri)
by anselm (subscriber, #2796)
[Link] (12 responses)
Because that was such an unqualified success when “The SCO Group” did it (over JFS, among other things) …
Posted Oct 24, 2025 11:47 UTC (Fri)
by khim (subscriber, #9252)
[Link] (1 responses)
In case of The SCO Group JFS was third-party code with no obvious ancestry toward anything copyright “The SCO Group”. Here we are, literally, talking about two products from the exact same, very litigous, company.
Posted Oct 24, 2025 12:25 UTC (Fri)
by anselm (subscriber, #2796)
[Link]
IIRC, The SCO Group's contention vis-à-vis JFS was (in a nutshell) that JFS was used in AIX, which was a derivative of Unix, to which The SCO Group claimed to hold the copyright. Hence, they said, IBM should not have been allowed to contribute JFS to Linux under the GPL (certainly not without seeking TSG's permission). As it turned out, the JFS implementation in Linux is derived from IBM's JFS implementation for OS/2, which is independent of the one used in AIX, but that was just one of the problems with TSG's legal omnishambles.
Oracle's claim to ownership of ZFS is a lot more clear-cut, but – legitimate differences of opinion with respect to its license vs. the GPL notwithstanding – in my non-professional opinion, the fact that Oracle has let Canonical include ZFS with Ubuntu for a decade or so without intervening isn't going to increase the likelihood of their prevailing in a court of law if they ever do decide to take the plunge after all. Also there is always the risk of losing, which would presumably encourage every other Linux distributor to reconsider including ZFS with their offerings, too (right now the uncertainty seems to put most people off the idea). Therefore it seems unlikely to me that Oracle wants to embark on this type of expensive and long-running PR nightmare with a very uncertain outcome anytime soon.
Posted Oct 24, 2025 18:15 UTC (Fri)
by tuna (guest, #44480)
[Link] (9 responses)
Posted Oct 24, 2025 22:56 UTC (Fri)
by anselm (subscriber, #2796)
[Link] (8 responses)
Not as much as it cost The SCO Group. Hint: IBM is still around but The SCO Group isn't.
Sure, but anyone can sue anyone else about anything, so sometimes you can't help it. IBM, after all, has pretty deep pockets and as such presents a juicy target for nuisance-type lawsuits. OTOH this means they can afford good lawyers when they need them, so that's something of a two-edged sword. One popular hypothesis is that The SCO Group sued IBM hoping that IBM would just buy the company outright in order to make the lawsuit go away with the least fuss, but as we know that's not how it turned out in the end.
Posted Oct 27, 2025 19:54 UTC (Mon)
by tuna (guest, #44480)
[Link] (7 responses)
Posted Oct 27, 2025 20:15 UTC (Mon)
by pizza (subscriber, #46)
[Link]
It's been said that Oracle is a legal department with a database; this was famously captured by Manu Cornet's unofficial org chart:
https://bonkersworld.net/organizational-charts
Posted Oct 28, 2025 0:46 UTC (Tue)
by anselm (subscriber, #2796)
[Link] (5 responses)
Red Hat does it by not distributing ZFS. (There are probably other reasons for that, too, and I wouldn't presume to speak to Red Hat's motivations here.) Which is not to say that if Oracle was desperate to sue Red Hat/IBM they might not find something else to sue about instead, but I think it's unlikely.
At this point Oracle would probably find it difficult to (successfully) sue Canonical for distributing ZFS with the Ubuntu kernel. They would have to claim standing as copyright holders in the Linux kernel, not ZFS, in the first place, and it would be a challenge to explain to the court why they sat back and did nothing at all for a decade or so while Canonical was being quite up-front about what they were doing with ZFS during all that time.
Posted Oct 28, 2025 10:17 UTC (Tue)
by Wol (subscriber, #4433)
[Link] (4 responses)
Not impossible, but it's a lot more expensive (you can easily find yourself paying your victim's legal costs - well most of them), and the English legal system is much much easier to short-circuit, as in "please can we have a ruling on the evidence so far, we want a summary dismissal". There's not all this gathering evidence up front for the entire trial. They tend to work their way down the claim-sheet collecting evidence and ruling on each claim in turn, so if you can get the initial claim dismissed, the trial would typically collapse there and then.
And while I gather it's almost never happened, the defendant has the option of filing to have a nuisance plaintiff declared a "vexatious litigant". Which pretty much bars them from the legal system. They need the permission of a Judge before they're even allowed to file a suit, which means they have to prove they have a reasonable case, on their own dime, before their target even officially gets told of the suit ...
Cheers,
Posted Oct 28, 2025 13:37 UTC (Tue)
by draco (subscriber, #1792)
[Link] (3 responses)
Further, it's my impression that it's simply ritual that the defense attorney in US criminal trials will move to dismiss the trial for lack of evidence. I gather that rarely succeeds, but it's allowed to if the facts are right.
But the fact that a jury generally must be empaneled to decide points of fact does make it unlikely for a trial with multiple claims to be split into multiple smaller trials like it sounds like you can do in the UK. I don't think it's disallowed in the US (I've seen judges suddenly start trial proceedings in sanctions or show cause order proceedings when a jury isn't required, so it seems the rules of procedure are flexible enough) but I've never seen a jury trial split that way.
Also, it's possible to bar nuisance litigants from filing claims in US courts without permission from the judge. I think it happened to the Prenda Law lawyers (in addition to eventually disbarring them).
Posted Oct 28, 2025 15:31 UTC (Tue)
by paulj (subscriber, #341)
[Link] (2 responses)
https://www.justice.gov.uk/courts/procedure-rules/civil/p...
I guess US systems have something similar?
Posted Oct 28, 2025 16:04 UTC (Tue)
by draco (subscriber, #1792)
[Link] (1 responses)
You can't really talk about a single US law system either, since there's a federal system and 50 different state systems (plus slightly different rules for criminal and civil). But there's a lot of similarity between them regardless.
I don't know how to answer your other question. I know parties are always encouraged to settle or find resolution via mediation, but as far as I know, that's largely driven by court costs, legal fees, court schedules, and fear of the jury more than any formal rules. That said, there's a lot of steps before the trial starts, and I think there are quite a few legal standards and precedents that may serve a similar function. After all, all court systems need to contend with high demand for their services. People are people, after all!
Posted Oct 29, 2025 8:51 UTC (Wed)
by taladar (subscriber, #68407)
[Link]
Posted Oct 23, 2025 19:11 UTC (Thu)
by IanKelling (subscriber, #89418)
[Link] (2 responses)
If you know it, you have not made an accurate or a good summary of it here.
> I find Ubuntu
Other people do not agree with you or Ubuntu.
Posted Oct 31, 2025 6:34 UTC (Fri)
by linuxrocks123 (subscriber, #34648)
[Link] (1 responses)
Yes, he did.
> Other people do not agree with you or Ubuntu.
Those people are wrong, which is why you were unable to say anything substantive about the reasons for disagreement. ZFS on Linux is no more of a problem than proprietary NVidia drivers on Linux, which are clearly fine, so Red Hat could obviously do DKMS instead of pre-built if they were skittish about distributing the .ko files directly.
They're not doing that because they're not skittish; they're paranoid.
I don't have a dog in this fight. I use Slackware and ext4 with the journaling disabled, libeatmydata in ld.so.preload, and these sysctl.conf settings:
vm.dirty_ratio = 90
My filesystem is screaming fast.
Posted Oct 31, 2025 10:40 UTC (Fri)
by taladar (subscriber, #68407)
[Link]
Posted Oct 24, 2025 7:21 UTC (Fri)
by taladar (subscriber, #68407)
[Link]
Posted Oct 24, 2025 10:12 UTC (Fri)
by nim-nim (subscriber, #34454)
[Link]
Red Hat customers really care about data loss (that includes software raid when economics dictate deployment on cheap nodes without hardware raid) and care mildly about new features such as COW and snapshots (that require expensive specialists to make use of). Data loss on plain old basic filesystems is not sexy but in the Enterprise space that saves you a lot of embarrassing and *expensive* incidents (a post-mortem with review of god knows how many historical mistakes and piles of technical debt is not cheap).
Posted Oct 28, 2025 13:33 UTC (Tue)
by claudex (subscriber, #92510)
[Link] (10 responses)
You can snapshot, compress and deduplicate with LVM.
Posted Oct 29, 2025 8:52 UTC (Wed)
by taladar (subscriber, #68407)
[Link] (9 responses)
Posted Oct 29, 2025 18:20 UTC (Wed)
by mbunkus (subscriber, #87248)
[Link] (8 responses)
Posted Oct 30, 2025 8:57 UTC (Thu)
by claudex (subscriber, #92510)
[Link] (7 responses)
You don't need to with LVM thin
Posted Oct 30, 2025 9:09 UTC (Thu)
by mbunkus (subscriber, #87248)
[Link] (6 responses)
1. Create partition table (not necessarily needed if it's a second disk, I'm aware) with one partition spanning the whole drive, type LVM
How does this setup handle the underlying storage (the VG) getting full due to snapshots occupying at least some part of the VG? With snapshot-capable file systems the file system itself is aware of all allocated space & can just calmly reject further allocations with ENOSPACE. But how does ext4 or XFS cope with the underlying storage suddenly being full even though the device (the LV in this case) is supposed to be 4 TB large?
Posted Oct 31, 2025 18:58 UTC (Fri)
by raven667 (subscriber, #5198)
[Link] (5 responses)
Posted Oct 31, 2025 19:10 UTC (Fri)
by mbunkus (subscriber, #87248)
[Link] (4 responses)
With a file system on LVM thin+LVM snapshots an application cannot rely on `df` & the corresponding syscalls anymore, as the file system doesn't know about how much space the VG actually still has.
Furthermore I'd be interested in how a file systems without snapshot support would react to the kernel reporting ENOSPACE while writing to a block device that the file system thinks should have free space left. Is this a kernel panic? Is this handled gracefully?
With file systems that support internal snapshots such as btrfs but also zfs (and I bet bcachefs) you really just allocate the whole VG to btrfs & let btrfs deal with how much space there is.
> you'd want to estimate how much space the snapshots are likely to take and factor that into your allocations
That's what I thought, too, but what claudex claimed was _not_ needed with LVM thin+LVM snapshots.
Posted Oct 31, 2025 19:29 UTC (Fri)
by raven667 (subscriber, #5198)
[Link] (3 responses)
Right, you have to look at the free space in the Volume Group and manage that, which is a different command (`vgscan` or `vgdisplay` IIRC), but that's just a small technical difference, it doesn't make the space usage *unmanageable*
> but what claudex claimed was _not_ needed with LVM thin+LVM snapshots.
That seems risky, but what do I know? I think in most common cases where you aren't bumping up against the limits this would work fine, but as soon as you start really filling the filesystem, or have very large snapshots, you are going to run into conflicts and surprise ENOSPACE is no fun at all. Even in the best case, regardless of the reliability of the underlying filesystem, something is going to break in my experience. Neither thin provisioning nor btrfs/zfs make more space than exists, so the free space has to be managed one way or another regardless of which flavor you choose, layering lvm/filesystem or integrated filesystem with volume management.
I haven't used Stratis from Redhat either but as I understand it that puts some automation and tools around the stack to make this kind of management straightforward and reliable.
Posted Nov 1, 2025 0:38 UTC (Sat)
by intelfx (subscriber, #130118)
[Link] (2 responses)
It does make free space unmanageable by normal means.
It's not a "small technical difference", it breaks the tool *everyone* knows to use, as well as the the API the tool relies on. Now if you need to retrieve the free space programmatically, it's suddenly not enough to call `df -P` or `lsblk -f` or `findmnt` do a stat() on the mountpoint — now you have to maintain a bespoke code path and resolve numerous levels of indirection yourself (find the filesystem corresponding to the mountpoint, find the LV corresponding to the filesystem, find the VG...).
This *is* a big deal for programmability and maintainability of the resulting system, and for mental load of anyone who comes into contact with this system.
Posted Nov 1, 2025 11:58 UTC (Sat)
by cesarb (subscriber, #6266)
[Link] (1 responses)
Can you do all that as non-root (honest question, I don't know)? At work, I've written code which detects when disk space is running low, and dynamically limits the size of an on-disk cache to leave at least 5% of free space in the disk (that cache also has a configurable maximum size, detecting the available disk space is an extra layer of defense in case something else is filling the disk); that code runs as a normal user, or sometimes within a container.
(Also, containers are an issue, they obviously don't have access to do all that; the traditional disk space API works fine in them.)
Posted Nov 1, 2025 17:59 UTC (Sat)
by raven667 (subscriber, #5198)
[Link]
No, managing LVM is a root-only activity, but I think we are talking about very different use cases and audiences here (also responding to intelfx), I'm talking about professional sysadmins running fleets of servers, or manufacturers writing firmware for products who use LVM and snapshots (or btrfs and snapshots) as part of their system management, not enthusiasts with their own highly bespoke setup or j-random-application managing its own storage needs but not managing the whole system.
I haven't used LVM snapshots, but I've been using LVM since it was available in Linux (and briefly EVMS from IBM before that, anyone remember that one?) and I've used VMware VMFS with thin provisioned VMs and snapshots, which I imagine has the same overall problems and constraints and so the experience would be relevant, ill-advised (or forgotten) use of snapshots can get you into trouble in VMware as well, changing the IO access pattern of the VM and filling up a tight volume unintentionally, and use of thin provisioning doesn't mean you can oversubscribe the storage forever without running into eventual consequences, it just gives you extra breathing room in the common case where not all storage is used.
I think of thin provisioning like a quota, it allows you to oversubscribe your storage but puts a quota on each individual usage of the storage, if all the users of the storage tried to use their max quota at the same time you'd be in trouble (because over-subscription) but no single (or small number) of users hitting their quota will break the storage for anyone else.
I didn't realize RHEL had never supported Btrfs. XFS is a fine filesystem but lacks features that have been stable in other filesystems for well over a decade. By which I don't mean Btrfs (I don't know) but, in particular, ZFS. I have been using ZFS on desktop and laptop for 10+ years and zero issues. It in fact once helped me recover from a failing disk in time to back up with no data loss. And it supports a bunch of things XFS doesn't; snapshots in particular are a lifesaver.
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
1) Canonical is based in a different jurisdiction than Redhat so their legal analysis might be different
2) Canonical is still personally run by the founder while Redhat is run more impersonally, and their organizational tolerance for risk might be substantially different.
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
It is accepted until Oracle's lawyers change their mind and sue IBM and all their customers.
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
How much do you think that law suit cost IBM?
Even better than winning is not getting sued in the first place.
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
You can do your best to avoid it, especially if you are potentially dealing with Oracle.
RHEL's default is XFS?
Wol
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
vm.dirty_background_ratio = 50
vm.dirty_expire_centisecs = 360000
vm.dirty_writeback_centisecs = 60000
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
RHEL's default is XFS?
LVM thin & snapshots & edge cases
2. Create VG with that partition
3. Create _thin_ LV with 100%FREE extents = 4 TB size ??
4. Create file system on LV
5. Create LVM snapshot whenever I need one
LVM thin & snapshots & edge cases
LVM thin & snapshots & edge cases
LVM thin & snapshots & edge cases
LVM thin & snapshots & edge cases
LVM thin & snapshots & edge cases
LVM thin & snapshots & edge cases