Matthew Dillon has made significant progress on DragonFly’s journaling code, which can now mirror partitions. Essentially all that remains now is for the “reverse journaling” work to be completed (allowing for filesystem rollback), and for stability to be increased. More information here and here
This seems to be more far-reaching than traditional metadata journalling support found on filesystems like ext3 or reiserfs.
Sounds like it will work on any supported filesystem, and is (or is intended to be) capable of mirroring volumes on different machines, not just on the local box.
It will definately be interesting to see in action once the bits are fully in place and tested…
I don’t use any unix, but this is pretty darn neat!
This kind of stuff theoretically allows live proces migration in an easy way.
i’m impressed by the sheer will and ability of the dragonfly team. personally i think all the BSDs are missing a security/audit/event log mechanism such as OpenBSM – analogous to solarism BSM.
the dragonfly journalling is different from the filesystem-fast-recovery journalling filesystems implement.
DragonFlys journalling mirrors each VFS operation to a filedescriptor (e. g. a pipe). A program listening to that descriptor can do anything you imagine. At the moment mirroring to disk is implemented. But you could as well analyse the output to, say, find a virus or such.
btw: a pipe can surely point to a ssh tunnel. so you get near real-time remote mirroring. Without any specific networking code in the kernel. For Free! Great, heh.
~ibotty
FreeBSD-CURRENT is getting an import of OpenBSM. From what I have been reading its going to be easily portable to other OSes.
I agree with you about dragonfly, they seem to be making a great deal of progress, it would be nice if BSDs shared a unified driver interface. Last time I tried dragonfly I was turned back by lack of hardware support.
It’s good to see DragonFly living up to its promise to add new and useful features instead of just re-implementing existing features (a trap that many other OSes have fallen into).
Dragonfly has some great concepts for the multi-core concept.
I tell you what, DragonFly might have a big life in the near future where other don’t.
I am not a DFBSD user but I can say one thing.
1) I am damn impressed with the work of the DF crew, especially Matthew Dillon.
Cheers to the team.
Matt has been impressing me greatly since the Amiga days. May he ever do so.
Matt has another response on the kernal mailing list with regards to journaling.
http://leaf.dragonflybsd.org/mailarchive/kernel/2005-07/msg00020.ht…
anyone know if linux has support for anything similar?
“anyone know if linux has support for anything similar?”
Of course. Linux has ext3,reiserfs, xfs ans jfs all of which are journalling filesystems. none of the mainstream bsd’s have anything like that
“Of course. Linux has ext3,reiserfs, xfs ans jfs all of which are journalling filesystems. none of the mainstream bsd’s have anything like that”
Linux has nothing like what is going into DragonFly now. Linux has several meta-data jounaling filesystems, and due to the thin VFS layer, that adds up to a lot of duplicated functionality (read code bloat).
DragonFly’s journaling code supports all the filesystems that DragonFly itself supports as it’s implemented within the VFS layer, and not in individual filesystems. In fairly short order, DragonFly’s also will be far more capable, as it does not merely journal filesystem meta-data, but can mirror *entire* filesystems, and not just localy, but over a network as well.
Of course. Linux has ext3,reiserfs, xfs ans jfs all of which are journalling filesystems. none of the mainstream bsd’s have anything like that
Hehe, I suggest you read the articles before embarassing yourself like that 😉
”
Linux has nothing like what is going into DragonFly now. Linux has several meta-data jounaling filesystems, and due to the thin VFS layer, that adds up to a lot of duplicated functionality (read code bloat). ”
Linux VFS and file system code is entirely clean. You can have both meta data and data journalling with mount options.
“Hehe, I suggest you read the articles before embarassing yourself like that ;-)”
name a single production level journalling code in any bsd system. then we will talk
“Of course. Linux has ext3,reiserfs, … none of the mainstream bsd’s have anything like that”
FreeBSD 6.0-CURRENT is getting reiserfs support, I believe it is currently read-only, not full support I know, but it is in the works.
There is also work going into a journaling layer for the native UFS. It will be interesting to see if freebsd references any of Matt’s work.
“Linux VFS and file system code is entirely clean.
Clean? Where did that come from? I said it was “thin” as in lacking just enough functionality, that many filesystem drivers in Linux need to contain essentially duplicate code when they share similar features with other drivers.
It’s nice to see progress from the DFBSD folks, and I’m still keeping a close eye on them for when FreeBSD 4.x support goes away.
As for the journalling, I’m not sure where BSD is going to end up with this. Adding a journal to FFS seems like a bit of a waste to me, considering the other drawbacks inherent to FFS. FWIW, there already is a journaled FFS available from Wasabi Systems.
http://storagebuilder.com/products/storagebuilder/iscsi/articles/9….
I really wonder if the future of BSD filesystems, however, is HFS+. NetBSD has an HFS+ project for Google’s Summer of Code. HFS+ has become a very nice filesystem to use with unix-like systems with the addition of journalling and case-sensitivity. It’s also pretty damn fast. 🙂
> name a single production level journalling code
> in any bsd system. then we will talk
Wasabi Systems had OTFS (now WasabiJFS), which was sheer brilliance; Crescent Anchor sells FireflyBSD systems with a functional journalling filesystem…