LWN: Comments on "The OpenZFS project launches"

ZFS Design

cortana — Mon, 30 Sep 2013 15:28:33 +0000

OTOH you know that when you are installing a new kernel you should wait for the out-of-tree modules to be updated.

I can't speak for the distribution you're using, but in Debian the kernel packages have an 'ABI number' embedded within them, so when you upgrade the kernel to apply e.g., a security fix you know you won't have to rebuild anything, let alone have to worry that you can't rebuild due to an API change.

ZFS Design

cortana — Mon, 30 Sep 2013 15:23:08 +0000

Not sure why installers can't install DKMS packages. In Debian the base 'dkms' support package depends on the stuff necessary to build modules, so installing foo-dkms will pull in the compiler and kernel headers automatically.

IIRC you can install Debian inside VirtualBox and you'll end up with a system including virtualbox-guest-dkms without any user intervention.

ZFS Design

peschmae — Sat, 28 Sep 2013 17:49:42 +0000

Having used both OpenAFS and the binary Nvidia drivers - which both are built by aforementioneds system - I can safely say that those two are the things that break the most regularly when upgrading my system.

They do so for a variety of reasons, starting with "Kernel too new and not supported yet". The situation seems to have improved recently, but its still not as reliable as modules that just *are* in the kernel.

I would hesitate format / with a filesystem that is due to break at the next upgrade...

The OpenZFS project launches

ssam — Wed, 25 Sep 2013 14:46:13 +0000

But someone could re-implement ZFS and licence the code however they wanted. Then it would only be patent fears holding back its inclusion. (There would not be much at all in the linux kernel if every patent fear was upheld).

The OpenZFS project launches

jospoortvliet — Mon, 23 Sep 2013 12:02:08 +0000

Link: http://news.opensuse.org/2013/09/19/opensuse-13-1-beta-is...

The OpenZFS project launches

jospoortvliet — Mon, 23 Sep 2013 12:01:55 +0000

In an effort to push btrfs forward, the openSUSE 13.1 beta testing work will focus on btrfs. The SUSE btrfs team is discussing the disabling of unsafe features - with those disabled btrfs should be perfectly safe.

ZFS Design

Rudd-O — Mon, 23 Sep 2013 01:49:15 +0000

https://github.com/Rudd-O/zfs-fedora-installer

I use it in all my systems.

The OpenZFS project launches

job — Fri, 20 Sep 2013 17:18:27 +0000

There are so many examples of this it's not funny anymore. Except for the ones gus3 mentioned there OpenSolaris and Hudson which were very high profile open source projects before Oracle.

One can only wonder why Oracle just doesn't let them loose instead of repeatedly stabbing them in the back until there is no community left. There can hardly be any money to be made from this behaviour.

ZFS Design

Lennie — Fri, 20 Sep 2013 12:02:20 +0000

Do installers install DKMS packages ? I'm not aware of any that do (would surprise me if they did, I guess in theory it could do that in a chroot maybe).

So if binaries can't be distributed and the installer doesn't support DKMS packages then the result will be it will never be the default filesystem.

ZFS Design

cortana — Fri, 20 Sep 2013 11:49:52 +0000

Not sure about binaries, but there exist systems (DKMS) that build modules during package installation, so from the end-user's point of view this doesn't matter. Debian has a forthcoming zfs-linux package that I'll be using as soon as it's available.

ZFS Design

Lennie — Fri, 20 Sep 2013 10:42:33 +0000

Well, there is one feature ZFS will never have... being included in the mainline Linux kernel. ;-)

That has mostly been solved by ZoL, but the real question is: can it be included with Linux distributions ? I don't think I've seen that yet.

If not, then it can't ever be the default filesystem on Linux either.

ZFS Design

Rudd-O — Fri, 20 Sep 2013 10:07:29 +0000

We're waiting for block pointer rewrite, then ZFS will have feature parity with btrfs (and btrfs still won't have feature parity with ZFS). That is starting to sound much more plausible with this OpenZFS announcement agglomerating hundreds of developers working on it.

The OpenZFS project launches

drag — Thu, 19 Sep 2013 16:46:47 +0000

I would expect that it's viable if you compile it yourself and they don't ship any GPL-originated code with their code.

That is as long as their code is not derivative of Linux kernel code then the GPL cannot apply because that is outside the scope of copyright owned by the various Linux kernel devs.

Presumably once it's compiled it's going to pull in code from the kernel it's compiled against and then it would be a derivative product and falls under the scope of the GPL license. The GPL license, of course, itself says that it's restrictions only apply to distributed code and nothing that you use yourself. Therefore it would be legal to combine CDDL code and GPL (and produce a derivative of both) code as long as you don't distribute it since GPL doesn't restrict in this sort of usage.

The key here is 'derivative'. This is a legal term with specific definitions that is in the USA copyright code. It's a core concept to the copyright law itself. What is and what is not derivative is ultimately up to a court to decide. It's actually very arbitrary and don't expect it to always make logical sense so there may be some unexpected 'gotchas' and is why people pay money for lawyers to help interpret (which I am not one of them).

But ultimately, as long as the OpenZFS are fairly careful, then it's perfectly legal to distribute the code that you need to build your own modules. This is similar to the situation with the Nvidia binary driver.

The OpenAFS file system drivers for Linux have similar legal situation also. It's distributed under the IBM Public License which the FSF claims is incompatible with the GPL.

From FSF's license comparison website: 'This is a free software license. Unfortunately, it has a choice of law clause which makes it incompatible with the GNU GPL.'

Pretty much the same boat as the CDDL for the subject under discussion. OpenAFS support is a pretty normal feature that distributions support.

ie:

% yum search openafs |grep $(uname -r)
kmod-openafs-3.10.11-200.fc19.x86_64.x86_64 : openafs kernel module(s) for
: 3.10.11-200.fc19.x86_64

Although you'd have to look at the code of OpenZFS vs OpenAFS and talk to a lawyer to be fairly sure. Some kernel modules will violate the GPL even if they were in source code format. It's a very case-by-case situation.

ZFS Design

nye — Thu, 19 Sep 2013 10:43:42 +0000

>Reduced performance due to data fragmentation

There's been some work done recently that might have some bearing on this. The fragmentation itself won't be reduced (really it's intrinsic to the design of a log-structured filesystem), but peformance when free space is running low should be much improved. See http://permalink.gmane.org/gmane.os.illumos.zfs/2565 - there's been some more discussion on the topic, but you'll have to dig around the list to find it.

>Not allowing to shrink a filesystem

This, along with actual defragmentation, would require block pointer rewrite functionality. This is a blocker for a lot of features, and has been talked about for years without any visible progress. Someone (might have been Matt Ahrens) mentioned a while back that they had been working on BP rewrite at Oracle, shortly before all the ZFS developers of note left the company. Reading between the lines, I suspect this might be taken as an indication that those most qualified to do that work might be concerned about legal repercussions if they did so (but this is purely speculation on my part).

Regardless, there's recently been some noise at Nexenta about implementing BP rewrite (see http://blog.gmane.org/gmane.os.illumos.zfs/month=20130701 for example), so you never know.

ZFS Design

Lennie — Thu, 19 Sep 2013 09:28:09 +0000

AFAIK ZFS still can't shrink filesystems.

I've not seen any indication anywhere that says anything else.

ZFS Design

abacus — Thu, 19 Sep 2013 07:14:46 +0000

Does anyone perhaps know what the current status is of the following ZFS aspects ?

Reduced performance due to data fragmentation. See e.g. Bart Sjerps, ZFS and Database fragmentation, February 2013 (http://bartsjerps.wordpress.com/2013/02/26/zfs-ora-database-fragmentation/).
Not allowing to shrink a filesystem. See e.g. Valerie Aurora, A short history of BTRFS, July 2009 (http://lwn.net/Articles/342892/).

The OpenZFS project launches

Blaisorblade — Thu, 19 Sep 2013 05:54:10 +0000

The project claims that they can distribute modules, just as anybody can distribute binary modules:
http://zfsonlinux.org/faq.html#WhatAboutTheLicensingIssue

How viable is that? Is there still a point in using the presumably slower ZFS-FUSE?

The OpenZFS project launches

tuomasjjrasanen — Wed, 18 Sep 2013 18:05:56 +0000

Yep. The golden touch of Oracle, probably not as legendary as the golden touch of Midas but definitely as devastating. Everything Oracle touches turns into a shimmering pool of pee.

The OpenZFS project launches

pizza — Wed, 18 Sep 2013 13:07:08 +0000

> We've been hearing that for five years now. Currently it's nowhere near feature-complete, progress is slow, and people are regularly losing data. I have a reasonably high degree of confidence that there will *never* be a day that it catches up to ZFS.

Unfortunately, I have to agree with you. It's been "almost ready now" for longer than I can remember. Meanwhile, my first attempt at a btrfs filesystem (on a small, light use SSD in a media center type of system), managed to corrupt itself beyond repair under normal operating conditions.

(The system had never had an unclean shutdown either, so it's not like I can blame a power outage)

The OpenZFS project launches

drag — Wed, 18 Sep 2013 12:00:56 +0000

The CDDL license is a free software license that is incompatible with the GPL.

There is not much more to say about the subject then that.

It's not the first license to do that and it's not going to be the last. The only people that can do anything about it are either the Linux kernel developers or Oracle and that can only happen by them choosing to change the license.

The OpenZFS project launches

drag — Wed, 18 Sep 2013 11:58:33 +0000

So far Oracle has shown zero interest in ZFS on Linux. I see no reason why that would change any time soon.

The OpenZFS project launches

nye — Wed, 18 Sep 2013 11:36:03 +0000

>This project is probably irrelevant for Linux, since there's no way to get ZFS accepted into the mainline kernel. The main benefactors of this project are *BSDs and various OpenSolaris forks.

On the contrary, this is likely to be particularly beneficial to Linux, as it represents ZFS as a standalone project, rather than a part of illumos. One of their goals is a greater separation of platform-specific parts, and an emphasis on improved portability and co-operation between the different ports of ZFS, which should reduce the work required to keep zfsonlinux up to date with upstream. (See http://open-zfs.org/wiki/Reduce_code_differences and http://open-zfs.org/wiki/Platform_code_differences)

>Anyway, btrfs is clearly a replacement for ZFS on Linux. It doesn't have all the ZFS features right now, but it'll probably get them in the near future.

We've been hearing that for five years now. Currently it's nowhere near feature-complete, progress is slow, and people are regularly losing data. I have a reasonably high degree of confidence that there will *never* be a day that it catches up to ZFS.

The OpenZFS project launches

hummassa — Wed, 18 Sep 2013 11:15:57 +0000

Usually all it takes is the supplier being bought by Oracle.

The OpenZFS project launches

sabroad — Wed, 18 Sep 2013 10:54:37 +0000

>> The exodus from Java and MySQL
>
> Care to explain? Has Oracle stopped shipping Java and MySQL?

exodus: a situation in which many people leave a place at the same time

A supplier doesn't need to stop supply for customers to leave.

The OpenZFS project launches

cyperpunks — Wed, 18 Sep 2013 10:40:24 +0000

> The exodus from Java and MySQL

Care to explain? Has Oracle stopped shipping Java and MySQL?

The OpenZFS project launches

paulj — Wed, 18 Sep 2013 09:46:27 +0000

I don't know, maybe. However, that's different from being sued over ZFS patents - the risk the OP was talking about, and the risk I was responding to. That risk doesn't exist if you use/distribute the Oracle ZFS code under the licence Sun^WOracle widely grants for it.

Re suing for CDDL/GPL incompatibility - any copyright holder could do that. That's nothing specific to Oracle. Indeed, Oracle possibly have less ability to sue on those grounds than others, as Ben pointed out.

Further, the incompatibilities between the CDDL and GPL are not terribly great. Philosophically, they are very close in intent. They differ mostly in technical details, e.g. the CDDL being explicit about patent rights and attempting to implement "patent MAD" (Sun having had bad experiences with being sued itself by patent holders), which the GPL (at that time, v2 was the latest) was very weak or quiet on. It'd be interesting to hear expert opinion on what impact that'd have on potential damages.

I certainly very strongly suspect any the damages would be *much* less than you'd get from losing a patent suit. I would be surprised if the CDDL/GPL incompatibility risks were anything but trivial compared to the risks of choosing to use Oracle ZFS patents *without* availing oneself of the licence to those patents offered by the CDDLed ZFS code. ;)

That said, most people will choose to avoid both those risks, of course. However, again, if you simply can not avoid those patents, then your smartest move may well be to avail of the CDDL licence to them, rather than avoid a perfectly good, free patent licence.

The OpenZFS project launches

BenHutchings — Wed, 18 Sep 2013 02:03:09 +0000

Since Oracle has some substantial contributions in the Linux kernel (Btrfs, among others), could they not sue people who commercialize OpenZFS/Linux on the ground of GPL - CDDL license incompatibilities?

That would be very interesting to see, as Oracle itself ships a Linux kernel including DTrace (and I don't believe they have explicitly relicensed it).

The OpenZFS project launches

salimma — Wed, 18 Sep 2013 01:29:32 +0000

Since Oracle has some substantial contributions in the Linux kernel (Btrfs, among others), could they not sue people who commercialize OpenZFS/Linux on the ground of GPL - CDDL license incompatibilities?

This wouldn't affect OpenZFS on other platforms, of course, so I agree with the previous commenter - this would be of benefit mostly to FreeBSD, Solaris forks and maybe OS X (I wonder what happened to ZFS there, after some positive noises before OS X 10.5)

The OpenZFS project launches

Cyberax — Wed, 18 Sep 2013 00:16:31 +0000

This project is probably irrelevant for Linux, since there's no way to get ZFS accepted into the mainline kernel. The main benefactors of this project are *BSDs and various OpenSolaris forks.

Anyway, btrfs is clearly a replacement for ZFS on Linux. It doesn't have all the ZFS features right now, but it'll probably get them in the near future.

The OpenZFS project launches

proski — Tue, 17 Sep 2013 21:05:05 +0000

I'd rather see another situation. If ZFS on FUSE becomes popular, major distros would provide an option to install on ZFS, which eventually could become the default. If FUSE is widely used, many other filesystems and filesystem encryption could be ported to the userspace. That would make Linux one step closer to a microkernel.

The OpenZFS project launches

gus3 — Tue, 17 Sep 2013 20:54:15 +0000

Technically, you are correct. However, bear in mind two points:

1. Oracle's reputation has tanked, thanks in large part to their abusive treatment of former Sun customers and collaborators. Of love, respect, or fear, Ellison chose the least productive one to instill in people. The exodus from Java, OpenOffice, and MySQL falls under "what did he expect?". Any new collateral bad PR for Oracle would diminish their standing even more. It's no way to grow the customer base, and it will make more corporate enemies than corporate friends.

2. I can't think of anything that would more accelerate the demise of software-only patents, than a lawsuit from Oracle over ZFS. Oracle is vulnerable right now, and many parties are more willing to take them on in court than they would have been even three years ago. A court case would be a gamble for all parties involved, but I doubt it's a gamble Oracle can afford right now.

The OpenZFS project launches

paulj — Tue, 17 Sep 2013 20:45:38 +0000

How exactly? As long as they're using the Sun^WOracle implementation of ZFS, then they got the required patent licence along with the licence to allow them to use/copy the code via the CDDL.

Indeed, if you want to avail of any features found in ZFS that happened to be patented by Oracle, then *NOT* using ZFS would be the *best* way to open yourself to the risk of being sued by Oracle. The best way then to *avoid* that risk would be ensure you use the CDDLed Oracle code implementing those patents.

E.g., similar to how Google, had they used the GPL OpenJDK, rather than another implementation, could then not have been sued by Oracle for infringement of JVM patents.

The OpenZFS project launches

lyda — Tue, 17 Sep 2013 17:25:02 +0000

Besides license issues, my biggest concern for ZFS is Oracle and whatever patents of theirs ZFS uses. If someone with money starts using ZFS in a visible way, Oracle's legal department will lock their sights on them and the legal billable hours clock will start ticking rapido.

Regardless of the technical merit of ZFS, it's a giant lawsuit just awaiting a victim.

The OpenZFS project launches

rahulsundaram — Tue, 17 Sep 2013 17:06:02 +0000

Linux kernel? No. a community of fork cannot unilaterally change the license of a weak copyleft licensed codebase.

The OpenZFS project launches

Xiol — Tue, 17 Sep 2013 16:58:48 +0000

I'm guessing this won't change the situation with regards to getting patches in the kernel?