Linux Developers Discuss Deprecating & Removing ReiserFS

[brent]

reiserfs is hardly maintained. The first page (200 commits) of git commits in the commit log on kernel.org goes back to 2013. So we're talking about less than 200 commits in the last 10 years! For ext4, 200 commits back is mid-2021. Sure that doesn't tell everything, but it is a usable indicator of the maintenance state.

[Berniyh]

Reiser5 looks pretty good to me. If it was in-kernel, I'd give it a chance. For real.

Sure, feel free to do so. Although I kind of doubt it'll ever happen.

Why is maintaining in-kernel and FUSE versions of each file system difficult enough that it is not common practice?

Because there is no point? Even if the additional effort was mild it would be just a waste of resources.

I wish there were FUSE versions for all file systems supported by Linux.

Why? It does not give you any advantage whatsoever.

And if one gets deprecated in the kernel, it gets maintained in userspace, with the associated performance degradation, but the data on existing drives is still accessible.

If a file system is removed from the kernel, then it might be time to create a fuse version of it.
And if ReiserFS is that fs, then it's likely that some will do exactly that.
At least if there is somebody actually needing it.

Also, Michael, in the article you say "Plus these days EXT4, XFS, and Btrfs are all better choices and even OpenZFS." -- do you have the benchmarks to defend your point? Would be nice to see those, especially with a lot of small files on the partition, like in the real world as opposed to the synthetic lab tests.

How important is the performance of a filesystem?
Even if ReiserFS would be 20% faster, does it matter?
Nowadays pretty much everybody is using SSDs that can shovel up to 5 GB/s around and have very low access times.
I doubt that you'll notice the improvement ReiserFS may give you in any real world scenario.
Does ReiserFS even understand the semantics of SSDs? Doubt that.
And don't get me started on HDDs. If you using those, then performance is obviously not important to you.
Don't get me wrong. Using HDDs is fine, but you don't use them where performance is important.

Where fs performance may still be important is server applications and workstations. But noone with a fully working brain would rely on a mediocre maintained fs like ReiserFS in a production system. In such cases you use the likes of ZFS, btrfs, XFS etc.

tbh, the only reason fs benchmarks are still a thing is to find performance regression. The rest is just another form of masturbation...

[Vistaus]

For the moment it would run just fine (unless you hit a bug due to code rot), but the kernel will spit out a warning that the filesystem is deprecated.
Just like it already does due to the 2038 incompatibility (didn't try, but the problem is mentioned in the text, so there will be a warning).
At some point it will be removed and in that case the filesystem won't mount anymore, because the kernel does not know it.
But of course you can still use some live system or a custom patched kernel to get your data.

Honestly though... noone with a functioning brain would use this filesystem these days. There just is exactly zero reasons to do that.
Personally, I would also claim that everybody using Reiser4/5 is out of their mind, too. But that's a different story.

Linux has enough great filesystems, there is no point in having Reiser*.

And if you happen to find some old disk formatted with ReiserFS that you need to get data from, then it should be possible to build the module out of kernel.
Or maybe there will be a fuse driver.

Reiser5 looks pretty good to me. If it was in-kernel, I'd give it a chance. For real.

[rmfx]

Yes, remove the old stuff nobody uses. If they want to use old stuff, they can use old Linux versions.

[sinepgib]

Just to add to my original comment, not only you don't need the in-kernel performance when you keep a filesystem alive for archival reasons, but you only need a read only implementation (tho you may need some write abilities to fix errors, but that's a reduced set of functionality anyway).

I wonder if it takes such people to push forward big things, society pushes people to be average, like using bad programming languages and concepts like Java/C# and OOP or now Python, why so that you can replace people easier when each retard and herd follower does the exactly same it's easier to replace them, even if the resulting products are 100 times worse, it does not matter you can put it throw 100 times as much people onte the problem, developers are very cheap for what they do (replace the work of thousends or more permanently often). People even like the guy behind the Python language also cut down his social live totally when creating the language.
So you have to be either asocial, slightly autistic or something and know that every person on the planet fights against you if you try to do something great.

And it's similar in business there you also have this psychopaths, it's just that if they murder 100 people with some business transaction about weapons or food, this murder acts are somehow legalized. Steve Ballmer and Steve Jobs are also not really considered neurologically totally normal, Ballmer is at least a colerical person and Jobs was a narcicist and probably a sociopath.

I don't think perfectly mentally healthy people can do something great in computer science. Ohh we can even go further, Elon Mask has clearly some personality disorder(s), genious or just the will to not be average and crazyness is very close.

You read too much José Ingenieros. You can be social all the same, your circles may be of different interests than the common folk's.

Why is maintaining in-kernel and FUSE versions of each file system difficult enough that it is not common practice? Just how different are the interfaces exposed by the kernel and by FUSE?

I wish there were FUSE versions for all file systems supported by Linux. And if one gets deprecated in the kernel, it gets maintained in userspace, with the associated performance degradation, but the data on existing drives is still accessible.

Maintaining a filesystem is hard by itself because you need real robustness. Most bugs you can recover from, but filesystem bugs destroy data and, while everyone should have a good backup policy, none will ever be faultless (at the very least, if you're obsessive enough, you'll lose a day worth of data), so that's something you simply can't recover from. This means extensive testing, among other things.
That said, FUSE filesystems are much easier to maintain because there are stable interfaces with userspace, while the kernel doesn't keep a stable API or ABI for modules, which means extra work even for in-tree drivers, and a lot more for out-of-tree ones. It's not static, so you need to keep track of changes.

At lest 5.10 (in debian stable) seems to have it still.
I remember trying all kinds of filesystems when I started with linux, and JFS was interesting for being the most quiet of them all (and IIRC, Reiser3 was the loudest).

What do you mean by quiet and loudest? Like how much they made the disk spin?

[bnolsen]

I used to use reiserfs because it had good performance for small files. Hopefully I don't have any of those partitions around anymore. With how long linux servers can just run and run and the number of hard drives I've transferred from machine to machine you never know.

[pmorph]

JFS was an interesting filesystem.
It had very slow meta data operations (slow to find files etc)

But is had synchronous write performance that was out of this world.
Unfortunately it had a tendency to lock up under heavy load (I won't be surprised if that bug is still present)

I am not following filesystem news anymore maybe JFS has also been dropped from the kernel.

At lest 5.10 (in debian stable) seems to have it still.
I remember trying all kinds of filesystems when I started with linux, and JFS was interesting for being the most quiet of them all (and IIRC, Reiser3 was the loudest).

[Tobu]

What happens if the system still runs ReiserFS and then someone upgrades the kernel?

libguestfs, a VM, or a live distro could help you get the data out.

[sinepgib]

While I don't really see the appeal of ReiserFS (not that I did any research anyway) and I wouldn't use it in new storage, block device interfaces (and only the standard interfaces) and filesystems are pretty much the only feature that I'd advice to never drop from an OS, mostly because of preservation. That said, if a FUSE implementation comes up, you don't really crave the performance for that edge case, so you can drop it from the kernel in particular.

[cb88]

I remember doing file system comparisons when I was first getting into Linux, and reiserfs was one of the standouts at the time. Jeez I feel old.



Yeah, Human Resources are the worst

I used to run reiser3 .... ext2/3 sucked compared to it usability wise. Especially when you had to sit through fsck at boot for the others.

It also handles folders like the lots of many small files portage tree on Gentoo quite well.