Tux3 File System Could Finally Make It Into the Mainline Linux Kernel

An anonymous reader writes "The Tux3 file-system that's been in development since 2008 as the public replacement to the patent-blocked Tux2 file-system is now under review for inclusion into the Linux kernel. Tux3 tries to act as a 'light, tight, modern file-system. We offer a fresh approach to some ancient problems,' according to its lead developer, Daniel Phillips. Tux3 strives for minimal resource consumption but lacks enterprise-grade reliability at this point. Tux3, at the end of the day, tries to be 'robust, fast, and simple' with the Linux FS reportedly being as fast as other well known file-systems. Details on the project are at Tux3.org."

Comments

Ambitious but not much has happened in 6 yrs

[haruchai]

and they expect to be competitive with ZFS?? They have a LOT of work to do.

Re:Ambitious but not much has happened in 6 yrs

[Anonymous Coward]

and they expect to be competitive with ZFS?? They have a LOT of work to do.

It's just another run of the mill linux filesystem which is to say completely useless.
The only real viable filesystems on linux are XFS, followed by EXT 4 and BTRFS (only in experimental form).

Re:Ambitious but not much has happened in 6 yrs

[Bengie]

It's a worthy goal to have. We need more competition in the FS sector. Many times competition is the inspiration for new features, even if some of these FS don't even make it off the ground. ZFS is great, but it's not perfect, and they only have so many resources to throw at new ideas to test. Monoculture is never a good thing.

Re:Ambitious but not much has happened in 6 yrs

[bheading]

On the contrary. What we have in filesystems at the moment is fragmentation.

We need people pitching in with stabilizing and fixing one major FS in Linux. It looks to me as if that should be btrfs.

Re:Ambitious but not much has happened in 6 yrs

[Bengie]

btrfs is interesting, but it's taking a long time to get anywhere and it has some big backers. I've also read some really well written blogs from sysadmins who have been Unix admins since the beginning of time, and they had some really good examples of some "Features" of btrfs that a sysadmin should never-ever use under any circumstance, and some features that are half-asses that are nearly a requirement for any good sysadmin, but cannot be done because of those other "bad" features.

One such example is btrfs allows a volume to be mounted under multiple parents. In order to handle this "awesome" feature, they had to give up the ability to atomically snapshot across volumes. In ZFS, if you mount a volume under another volume and snapshot the parent, the children will automatically be atomically included, not so with btrfs, that's an impossibility to add a feature that should never be used.

Re:Ambitious but not much has happened in 6 yrs

[dbIII]

Different aims.

Re:Ambitious but not much has happened in 6 yrs

[dbIII]

I disagree. The different approach of ZFS means it should be far better than btrfs when you have many disks, yet it makes almost no sense at all with single disks which is where btrfs makes sense.
Different tools for different jobs.

Re:Ambitious but not much has happened in 6 yrs

[philip.paradis]

I only use XFS and ZFS on Linux these days, with exception of ext2 on some /boot volumes. Nobody has shown me a compelling reason to deviate from what works.

Re:Ambitious but not much has happened in 6 yrs

[UnknownSoldier]

> ZFS is great, but it's not perfect

Serious question: I'm curious as to what those would be?

The license?

Re:Ambitious but not much has happened in 6 yrs

[cmurf]

What blogs discuss btrfs features that should never be used under any circumstances?

Could you elaborate on your example? I not understanding what commands make it possible to mount a volume "under multiple parents" how this differs from shared or bind mounts. I can mount an XFS volume on two separate mountpoints, not a big deal. Btrfs volumes can't be snapshot, just their subvolumes, maybe that's what you're referring to? The lack of recursively snapshotable nested subvolumes?

Btrfs and ZFS are different when it comes to snapshots. In Btrfs snapshots are subvolumes that are "prefilled". They aren't otherwise unique like they are on ZFS. And that means there isn't really a parent child relationship on Btrfs, you can delete "parent" subvolumes that have "child" snapshots, unlike on ZFS.

Re:Ambitious but not much has happened in 6 yrs

[Daniel Phillips]

Indeed, different aims. Tux3 has the modest goal of being a light, tight and fast filesystem without ambition of also being a volume manager.

Re:Ambitious but not much has happened in 6 yrs

[nctritech]

This is probably irrelevant for you, but I ran into issues with software running on i386 with XFS and newer kernels. Programs not compiled with -D_FILE_OFFSET_BITS=64 used the 32-bit versions of certain file-related system calls, and the default mount options for XFS changed at some point to allow 64-bit inode numbers to be created. What would happen is the program would readdir and choke the instant it hit a file or directory on an inode number greater than 2^32; the fstat calls returned EOVERFLOW and processing aborted. You'd go into a directory with GQView, for example, and mysteriously see i.e. three images and one directory where you knew there were tens of directories and hundreds of images.

Obviously, x86_64 platforms don't have this issue, but I was operating an i386 server since 2008 until just a few months ago and I found it to be extremely annoying and (at first) difficult to figure out what was happening. There is surprisingly little information about XFS and 64-bit file syscall issues when all you have is strace spouting EOVERFLOW at you and don't immediately pin the issue to the filesystem in use.

Re:Ambitious but not much has happened in 6 yrs

[sjames]

There are valid reasons to want to mount a volume in more than one place. For example, strong namespace based isolation for sensitive processes/users.

Over all, btrfs is much more flezible than ZFS. In the end, it looks like it will be the superior filesystem. For example, it has a much greater flexibility in changing the underlying storage over time. Why should gradually upgrading the underlying pool be a disruptive process. It seems natural that as drives age out to replace them with bigger drives. Ideally, the filesystem will move things around as needed to maintain the specified level of redundancy and to utilize as much of the new storage as possible.

I am currently using ZFS because btrfs isn't yet sufficiently mature, especially wrt redundancy

Re:Ambitious but not much has happened in 6 yrs

[sjames]

The license is an annoyance, but can be lived with, more or less.

I would like to see more flexibility in re-structuring the zpool. I see no intrinsic reason why a pool can't start out without redundancy and then have it added after the fact (the equivilent of bringing a soft raid up in degraded mode and adding in the other devices later) or have the geometry changed later. It should be perfectly feasible to start small and over time add more disks and replace small disks with larger ones. I would really like to be able to evacuate a disk before pulling and replacing it (given enough free capacity elsewhere in the pool.

Re:Ambitious but not much has happened in 6 yrs

[philip.paradis]

That's a darn good find, and definitely a really annoying issue. You're correct that 99% of the stuff I manage is x86-64; I've only got a couple of legacy x86 systems floating around, and fortunately in this case they don't use XFS. I can only imagine the amount of $head_desk you went through before realizing what the root cause was in the case you described.

Re:Ambitious but not much has happened in 6 yrs

[palion]

"Monoculture is never a good thing." Sentences with "sth. is never/always X" are generally not trustworthy... Always doubt. Maybe monoculture sometimes IS a good thing, who knows?

Re:Ambitious but not much has happened in 6 yrs

[Bill, Shooter of Bul]

There are more i/o workloads and problem domains than the enterprise grade storage system that zfs is designed for. Its like declaring that the Honda fit can never compete with a dump truck.

Re:Ambitious but not much has happened in 6 yrs

[Bill, Shooter of Bul]

Memory requirements, for one. ZFS is probably a terrible choice for tizen, or firefox os, or raspbian.

Re:Ambitious but not much has happened in 6 yrs

[haruchai]

This particular Honda Fit is not even properly assembled so it couldn't even bring a new set of tires for the dump truck.
Did you miss the part where it was stated "and a tertiary goal is to be better than ZFS"
The implication is that it will eventually be a ZFS replacement. My point is that they are so far behind and have accomplished so little to date, that will NEVER happen.

If they prove me wrong, good for them. I'm always on the lookout for a better tool or way to do things but I stand by my remark that this project is going nowhere fast.

Re:Ambitious but not much has happened in 6 yrs

[ruir]

Quite interesting your insight and finding. One more argument for me to complete my migration to 64-bits. Many thanks!

Re:Ambitious but not much has happened in 6 yrs

[Bill, Shooter of Bul]

Yes, I did miss that part where they pointed to zfs as an aspiration. Strange, weird, and just wrong.

TFS misses one point

[NotInHere]

From TFA: "Tux3 is yet another interesting open-source file-system designed for specialized cases."

NIHFS?

[BaronM]

First off, I think that 'better than ZFS' is a good and legitimate goal, seeing as how ZFS is very, very good, but not perfect.

That said, there's also BTFS and HAMMER aiming to be 'better than ZFS'.

I know: everyone wants to scratch their own itch, and there is no reason that multiple projects in the same area should necessarily been see as competing, and if I'm unhappy about it, I should just go write my own instead of complaining. Did I cover everything? :)

I just wonder sometimes if Linux wouldn't have moved beyond EXT4, X11, and the desktop environment wars if the 'not invented here' syndrome were just a little less prevalent.

Re:NIHFS?

[Anonymous Coward]

this is not the only weasel word infested post by you to this same article.
who claims that zfs is $GOLD_STANDARD? and which version of zfs
are you talking about? oracle keeps putting out incompatible versions, and
then there are the various opensolaris and bsd verions. ... saying $GOLD_STANDARD
implies something that is clearly not true --- that there is "a" zfs.

Re:NIHFS?

[Anonymous Coward]

I can see Tux3 used in some special cases, but there are so many filesystems that should be tossed in the Linux kernel first, IMHO. A few thoughts/ideas:

1: A successor for SquashFS except with the option of some ECC to detect corruption and optionally to fix it. That way, a compressed image can be somewhat self-healing.

2: Hooks for ZFS on Linux. Because of the license differences, it can't be included into the kernel proper, but might as well make it as easy as possible to have it used on bootup, perhaps as a root filesystem.

3: Not really a filesystem, but extensions to LVM3 to give features similar to products like UnRAID. That way, one can have a drive dedicated for parity, a drive as a write "landing zone", autotiering, hot spares, encryption on that level, etc.

4: A filesystem designed for WORM, with each set of writes cryptographically signed, perhaps with signed hashes from an external timestamping service. Similar to UDF, except with cryptography and maybe even error correction so if there is damage, it can be repaired. That way, it will be hard to tamper with the FS unless it is zeroed and rewritten or rolled back from snapshots. This way, removable hard drives can be used for archival storage and there is protection from tampering.

Re:NIHFS?

[Bengie]

OpenZFS. They're getting rid of "versions" and just having "Feature flags". This will allow you to create ZFS pools on one system, and just make sure what ever features that are only supported on another target system, are enable when you create the pool.

Re:NIHFS?

[BaronM]

Apparently, I grew up speaking 'typo'.

I suspect it's an editing error where I was writing in present tense (be seen as...), started a switch to past tense (been seen as...), and instead ended up with that ridiculous construction.

Oops.

Re:NIHFS?

[unixisc]

I just wonder sometimes if Linux wouldn't have moved beyond EXT4, X11, and the desktop environment wars if the 'not invented here' syndrome were just a little less prevalent.

The EXT* have been the only native filesystems of Linux, and BTRFS is something recent. With ZFS, the issue is an incompatible license, since CDDL code can't legally be combined w/ GPL code. Wayland again is new and likely to be the standard, aside from Ubuntu Unity, which will use Mir, and Android, which uses SurfaceFlinger

I do agree that there is a plethora of DEs. In the beginning, there was just KDE. Between that, and FOSS ports of CDE or Motif/OpenLook, they'd have been fine. Then came GNOME (due to the controversy over Qt licensing) and then a number of others. I do think that too much development effort is put into developing something that's already plentiful in the market.

File systems - not too many. You have ext2,3,4, and then you have file systems that have come from elsewhere - XFS from SGI, ReiserFS (the butt of all the dead wife jokes), VjFS from Veritas and that's about it. In the BSDs, you have UFS and the BSD FFS, and in addition to that, HAMMER/HAMMER2 and ZFS, that's come from Solaris.

As an old OpenVMS fan, this isn't...

[Nutria]

what I think of when someone writes "versioning filesystem".

Re:As an old OpenVMS fan, this isn't...

[MichaelSmith]

As versioning goes, FILES-11 was barely good enough to be called built in un-delete. Nowhere near as good as just checking all your files into git.

Re:As an old OpenVMS fan, this isn't...

[Nutria]

As versioning goes, FILES-11 was barely good enough to be called built in un-delete.

Mainly because -- in my 25 years using FILES-11 -- I've never heard of versioning called "undelete".

"Persistent undo" is the best term to describe going back to a prior file version.

parent delays

[KiloByte]

So tux2 was ready in 2000, and it took 14 years to rewrite it to avoid parents? Oh how much patents help innovation!

Re:parent delays

[Anonymous Coward]

So tux2 was ready in 2000, and it took 14 years to rewrite it to avoid parents? Oh how much patents help innovation!

Few more years and those patents will expire and we can use both!

Re:parent delays

[MichaelSmith]

took 14 years to rewrite it to avoid parents!

A lot of these linux developers are pretty young.

Re:parent delays

[NotInHere]

Politicians could at least recognize the faster pace in the IT world compared to other technology industries, and lessen the patent terms for software patents.

I also don't know why there should be a difference between a patent troll and a large company with lots of 'defensive' patents suing other companies because of "swipe to unlock" features.

Re:parent delays

[phantomfive]

That's lucky, it took me 18 years to get free of my parents. Although it might be more accurately described as, "they kicked me out of the house." But let's not go there.

Re:parent delays

[Daniel Phillips]

So tux2 was ready in 2000, and it took 14 years to rewrite it to avoid parents? Oh how much patents help innovation!

Few more years and those patents will expire and we can use both!

Tux3 is a better design. Tux2 was more along the lines of ZFS and Btrfs, that is, multiply-rooted trees sharing subtrees. Tux3 is a single tree with exactly one pointer to each extent. Considerably easier to check and repair. Of course we need to see if it turns out that way so please stay tuned.

ZFS or fail

[grub]

Come to the 21st century, use ZFS.

Re:ZFS or fail

[Anonymous Coward]

yes, if zfs worked correctly, it would be the ultimate evolution of 70s technology --- a non-distributed file system that may take an indeterminate time to export and recover. but it doesn't. i've seen customers lose lots of data with zfs. i have not had this experience with other file systems.

Re:ZFS or fail

[Bengie]

I've never heard of anyone ever losing data with ZFS, even people with petabytes of R&D data.

Re: ZFS or fail

[Bengie]

And it uses that detection to fix the bad data from good data from your other RAID drives. Not just using one drive, right? Can't blame the FS for mistakes of the user, you also can't blame the FS for really bad luck. The Earth blew up when the Sun went super-nova. ZFS is a bad FS because it didn't protect my data.

Re:ZFS or fail

[dbIII]

It doesn't entirely save you from failing disk controllers but does warn you that it is going on. I've lost four files that way, all new multi-GB temporary files that were easily recreated or no longer required, but it does happen when you keep failing systems limping along until the replacement arrives.

Re: ZFS or fail

[cbreak]

Any type of redundancy is enough to fix errors, as long as you have at least one correct copy. raidz has one copy redundancy, raidz2 has two copies redundancy, and mirrored drives have as much redundancy as you want.

Clearly

[smittyoneeach]

The way ZFS saved Sun marks it as an über-technology.

Since Trolling is an art, and stuff.

Re:Clearly

[dbIII]

Being seen to be worth being eaten by a larger fish is not a sign of failure in technology companies.

It's old and mature -- so let's come up

[fisted]

..with something new already.

Is the general problem of the GNU- and Linux world.

Backstory

[eclectro]

This is the story of the patents involved. It's not so much that there was any litigation, but rather the ongoing threat that there would be (for arguably stuff that was already being done.)

Untested and unused always tests as faster

[Anonymous Coward]

Systems that no one uses always test out faster than ones that actually work, deal with edge cases, do reliable recovery from hard crashes, etc. That's why ReiserFS was always faster, but kept hiding the corpses in the woods and pretending complete ignorance of having destroyed your data.

            http://en.wikipedia.org/wiki/Hans_Reiser

Come on, how could you not trust yet another "fan boi" burdened filesystem that's never been been shown as stable?

Choice

[warrax_666]

Oh, please. A modern Linux distro actually needs to provide hotplug that actually works, a tear-free desktop experience, reliable service termination/startup/restarts, etc.

Stop living in the past.

Re:Choice

[K. S. Kyosuke]

If you haven't noticed, he *was* talking about simplicity. Just the MIT type, not the New Jersey one.

Re:Choice

[Bengie]

If you don't like shiney, then there's always the console. For everyone else who likes desktops for stuff like videos or web browsing, annoyances like tearing is a distraction, like a 3 year old throwing a temper tantrum. Tearing was an issue fixed back in the 90s.

Re:Choice

[causality]

If you don't like shiney, then there's always the console. For everyone else who likes desktops for stuff like videos or web browsing, annoyances like tearing is a distraction, like a 3 year old throwing a temper tantrum. Tearing was an issue fixed back in the 90s.

I've used X11 and later Xorg for about the last 15 years. I've never had issues with tearing or any other visual artifacts. Back in the day I certainly had a real good time with modelines (fun!) but I've never had problems with tearing.

Am I the only one who hears complaints like this and scratches his head wondering what the fuss is all about? Lots of these mysterious complaints have never happened to me. I don't believe you are being dishonest, I just would like a more in-depth explanation about what problem you had, what solutions you tried that didn't work, and what you think the working solution would or should be.

Re:Choice

[shutdown -p now]

So what's you're saying is that a modern Linux distro should be more like FreeBSD?

YASF (c) forver by me

[FlyingGuy]

YASF or Yet Another File System.

Well someone has yet another personal project they want us all to take seriously. Really? I mean Really?

Of the numerous file systems out there, and I have tried a whole boat load of them, the one that is the most mature, most reliable, arguably the fastest is... Wait for it... From the company that everyone loves to hate...

Is NSS from Novell. It has more posix features then all of the rest of them, it is insanely fast, it provides undelete and complete repeatability and Novell has open sourced it. Nuff said.

Re:YASF (c) forver by me

[flux]

Not sure if trolling.. but I looked it up, and on the paper it seems interesting, but for use today it has limitations: 2 TB maximum device size, 8 TB maximum volume size. So that's a non-contender. Seems quite advanced for its day, though (introduced 1998).

Re:YASF (c) forver by me

[FlyingGuy]

Nope, not trolling..

Since it is open source and has all those goodies it would seem to me that increasing the volume size capabilities would not be nearly as tall an order as starting over from scratch.

Re:YASF (c) forver by me

[Anonymous Coward]

> YASF or Yet Another File System.

Woldn't that be "Yet Another Sile Fystem"? Or something?

Look: I don't know who this Flying Guy is nor what he has done for me recently. But Daniel Philips... this is an awesomely smart guy, and friendly and all. And he has contributed to free software which I use every day.

Now imagine how much weight your rant has in my eyes. This might be a view shared by many.

Re:At least...

[dbIII]

But then people with no understanding

That's a bit rich after some of your earlier posts about X and Wayland. I suggest you join the wayland mailing list and get some understanding before making such comments.

Re:Waste of time and effort

[Daniel Phillips]

Those shiny distributed file systems run on top of boring local filesystems.

Ready for mainline?

[rew]

To be taken serously, the home page needs to mention something more recent than 2008 in the "on the web" section. And the "we're active, see the git log" link needs to point somewhere other than a 404....

Re:Ready for mainline?

[Daniel Phillips]

Thanks for the bug report :)

Re:Ready for mainline?

[rew]

Daniel phillips, where have I heard that name before? It was in the last few days.... :-) Ah! :-)

versioning

[allo]

what about the cleanup?
Nilfs2 is quite cool, but the cleaner-daemon causes a lot of disk io, which is not only annoying, but makes me think about disk lifetime as well.

Re:versioning

[Tough Love]

Tux3 does not need a cleaner. At the moment, Tux3 also does not have snapshots. When it does, it still won't need a cleaner, although snapshot delete will run in the background because many inodes may need to be updated. This only touches metadata, which is normally less than 1/100th of the volume so it should be OK: some activity after the snapshot delete, but otherwise, changes commit immediately to disk then the disk goes quiet.