ext2 - ext3 - ext4 [LWN.net]

ext2 - ext3 - ext4

Posted Jun 15, 2006 11:26 UTC (Thu) by nix (subscriber, #2304) [Link]

This is an argument for calling it 'ext3'.

If the extents code were added in a copy of the filesystem, called ext4, then you could go to ext4+extents easily, but going back would be impossible unless extents were optional even in ext4 (which introduces to ext4 all the complexity problems which ruled out its being called ext3!)

If extents are optional, then you can remove extent blocks from files by mounting without extents (but *with an extent-capable fs*, so new ext3 or ext4 or whatever), then copying and renaming all files which have extents on them. It's fiddly, but doable.

If you introduce an ext4 with non-optional extents, then there's no filesystem which can read extents but prefers not to write them. Oops, *now* you can't go back.

ext2 - ext3 - ext4

Posted Jun 15, 2006 12:31 UTC (Thu) by NAR (subscriber, #1313) [Link]

which introduces to ext4 all the complexity problems which ruled out its being called ext3!

No, I don't think this is the reason. The problem with the "new ext3" is that someone upgrades the kernel, accidentally or intentionally mounts the filesystem with the same fstype, but with different options, then goes back to the previous kernel and finds that some of the files can't be read. I believe that a filesystem should be compatible with itself regardless of mount options.

Bye,NAR

ext2 - ext3 - ext4

Posted Jun 15, 2006 12:54 UTC (Thu) by ekj (guest, #1524) [Link]

Not being able to go back is perfectly acceptable for a new filesystem. Arguably less so for the same filesystem.

I would except a kernel with support for ext3 to be able to mount a ext3 filesystems. Everything else is, imho, lunacy.

There's no reason to expect though, that an ext3-driver should be able to mount a ext4-filesystem. If the oposite works as a migration-path it's a big benefit, but that's it.

ext2 - ext3 - ext4

Posted Jun 16, 2006 2:15 UTC (Fri) by jzbiciak (✭ supporter ✭, #5246) [Link]

And, since the change is isolated to meta-data, what really stops there from being an e4fsck flag that finds ext4-specific files and rewrites them as ext2 format (space permitting)?

As for a "start over again" filesystem, I believe Andrew Morton's position is non sequitur. We should keep evolving ext with whatever steps we think are appropriate, and let someone dedicated to filesystem design come up w/ the next big thing. It'll have to be obviously superior to ext3 (or ext4 or ext99 or whatever) to win, but if there was true benefit to the new filesystem, then it's worth it.

The corollary to my argument is that while Reiser4 seems to be just that sort of "rethink and do-over" type of filesystem, it's not obviously superior to ext3, and so it hasn't won a place in the Linux kernel yet. At least, that's how I see it.

reiser4, and general administration of Linux

Posted Jun 16, 2006 5:16 UTC (Fri) by zmi (guest, #4829) [Link]

They way they wrote it on lwn.net, I understand that reiser4 is too
intrusive for inclusion yet, as it makes many traditional things change. I
believe time has come for some "revolution" in FS, because the actual
handling is quite limited. I'm still missing the ease-of-administration I
had with Novell's Netware 4 back in 1996, where groups could be assigned
rights in the FS very simple. That's (a bit) possible with ACL's now, but
not as easy and smooth.

And, a bit OT, I'm missing something like "make that user having the right
to administrate his domain - manage users, ftp, and such". This is
currently a PITA, and not really funny. Linux is stuck with the
traditional "root is king" view, where it should adopt. SELinux seems to
be a step in the right direction, but all that SELinux/ACLs security tools
forget about the admin, who should be able to quickly manage things.
Commandline is nice and a must, but some GFX tools for the
not-so-much-keyboard-king type of admins should be there also. Just look
at Novells NDS, it's a fun to administer it.

mfg zmi

ext2 - ext3 - ext4

Posted Jun 18, 2006 11:31 UTC (Sun) by nix (subscriber, #2304) [Link]

I would except a kernel with support for ext3 to be able to mount a ext3 filesystems. Everything else is, imho, lunacy.

In that case ext2 and ext3 have always been lunatic. There have always been flags in the superblock indicating features that the fs driver must possess before the FS can be mounted read-only, and features that it must possess before the FS can be mounted writable.