This is part of The Pile, a partial archive of some open source mailing lists and newsgroups.
Date: Fri, 14 Dec 2001 11:00:04 -0500 From: Bryan-TheBS-Smith <b.j.smith@ieee.org> To: Alexander Udalov <aludal@softhome.net> Subject: Re: [svlug] Re: OFF TOPIC fsck for ext3 Alexander Udalov wrote: > Thank you Drew, fsck did the fscking job (or was it actually > fsck.ext3? hadn't the clue at the time, to check). Ext3 uses the same, reliable fsck as Ext2, as their structures are basically the same -- as long as your e2fsprogs package is up-to-date for your Ext3 version, it knows how to handle the journal components. Just for further note, on my RedHat 7.2 system, I did: $ ls -la /sbin/fsck.ext* -rwxr-xr-x 3 root root 561928 Aug 29 18:36 fsck.ext2 -rwxr-xr-x 3 root root 561928 Aug 29 18:36 fsck.ext3 A hard link??? Let's check: $ ls -li fsck.ext* 32927 -rwxr-xr-x 3 root root 561928 Aug 29 18:36 fsck.ext2 32927 -rwxr-xr-x 3 root root 561928 Aug 29 18:36 fsck.ext3 Yep, inodes match. > Well, fsck-ing was accomplished successfuly only in manual operation, > so to speak, and with my unprintable curses about some losses of data > discovered after the dust settled. At least Ext3 is _paranoid_enough_ to say "hey, something happened, I don't trust my journal." Trust me, I'd rather have that _any_day_! I've been running Ext3 since early 2000 on dozens of workstations and servers and it's nice to know that it uses Ext2's proven fsck, and its not afraid to go to it if necessary. After years of using HPFS and NTFS, it is obvious that IBM/Microsoft are a little "too aggressive" in always going to the journal. I've had 2 NT Servers toast themselves in the past. As such, I _force_ a full CHKDSK everytime NT crashes -- which defeats the purpose of journaling in the first place! Several people have noted that IBM's JFS continues this "aggressive" tradition as well. I don't know about ReiserFS. It seems that it actually knows when not to trust its journal. Unfortunately, in both cases where people I know have had issues and dropped down to a full reiserfsck, they've _completely_toasted_ their volume! I guess good journaling means little if the recovery tools are not mature enough -- I'm not saying ReiserFS isn't, it just doesn't look that way. Yeah, I'm sure that will net a little flammage, but realize that ReiserFS is about features and "pushing the envelope," whereas Ext3 benefits from Ext2's proven fsck with little additional effort on the part of the Ext3 team. And lastly, there is SGI's XFS. I have been using it on over a dozen Linux systems since February, and dozens of Irix systems since the mid-90s. XFS seems to be a filesystem that actively fixes and defragments itself constantly, consistently and correctly. On Linux, provided I stick with the "releases" or be careful to use a good CVS tag and configure/compile it correctly (and *NOT* with GCC 2.95/2.96!), I haven't had any data loss yet. The feature-set used to be unique, with ACLs and official quota support, but even Ext3 has that now. But there is one small bonus I feel there is to using XFS, it seems (or at least seemed to, at the time) _fix_ all the VM crap in the kernel -- because XFS is integrated in the VM, and SGI tested the hell out of their releases before putting the version stamp on them. But those are just my opinions. ===