Messages in this thread | | | Date | Tue, 23 Jun 1998 23:41:25 -0400 | From | "Theodore Y. Ts'o" <> | Subject | Re: (reiserfs) Re: LVM / Filesystems / High availability |
| |
Date: Wed, 24 Jun 1998 03:59:35 +0200 From: Michael Marxmeier <mike@msede.com>
For example locking the fs without making it effectively unusable (this is about on-line resizing): - negotiating with LVM about PEs which should be freed - the fs may no longer use that PEs for allocation - in case an inode must be moved - how to find all directories an inode is in on a life fs w/o all the possible races (on a big fs) - how to deal with open files when inode needs to be moved - how to do it safe so fsck is able to cleanup - some ext2 metadata must be aligned (block/cylinder groups)?
Just wondering if this is achievable with ext2?
Oh sure, it's doable. It's just a mere matter of programming. :-)
The hardest part is finding all directorys which an inode is in, but that can be done by searching through the filesystem, and if things change out from under the user mode program, it can just retry the search. Most of the time the directory for which an inode belongs to will be in the same block group, and this can be used to optimize the search. (Most inodes also only have one link.)
Doing it safely so that fsck can cleanup is easy; you have the kernel copy the the data block first, and only then does it update the references.
As far as metadata being aligned, see my previous comments about why the LVM should be done at the filesystem. It makes this problem easy to solve (where as doing it at the fake pseudo-device layer makes this problem very hard indeed, especially when you're removing a physical device in the middle of the filesystem).
- Ted
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |