On Wed, 24 Oct 2001 10:19:46 EDT, Kyle Lussier said:
and the many other contributors to Linux have offered. What really got my
attention was the shipping of a journaling file system (I believe with
Red Hat 7.2?) so that you can back out hard drive changes. Is that even
Actually, ReiserFS and ext3 have been available for quite some time, as have
both IBM's JFS and SGI's XFS, all of which do journalling, but *NOT* in
the "back out file changes" sense (at least not currently).
What all 4 of these file systems do is maintain a journal of disk block
allocation and structure, so that if the directory/file/free structure is
known good at time X, and the system then crashes at time X+Y, to restore
the disk to a "known good" state only requires replaying the journal.
So for instance, if you checkpoint the known good state to disk, and then
remove a file, and then the system crashes, the 'fsck' (the moral equivalent
of the Windows 'chkdsk') doesn't have to walk across the *entire* free list,
and *all* the files, and figure out that 23 blocks are not on the free list
even though they should be. It can just replay the journal, which only has
"unlink this file" "put the 23 blocks on the free list" - and the disk is
once again in a consistent state.
Having said that, I'll note that the ReiserFS guys *have* specifically designed
their code to allow the plug-in of commit/rollback features, and I'm sure
that the IBM and SGI crew are both thinking about the issues for their
high-end boxes... ;)
Operating Systems Analyst
Description: PGP signature