xfs
[Top] [All Lists]

Re: Dumping a mounted filesystem

To: Chris Pascoe <c.pascoe@xxxxxxxxxxxxxx>, <linux-xfs@xxxxxxxxxxx>
Subject: Re: Dumping a mounted filesystem
From: "Nathan Scott" <nathans@xxxxxxxxxxxxxxxxxxxxxxxx>
Date: Tue, 1 May 2001 11:01:28 -0500
In-reply-to: Chris Pascoe <c.pascoe@csee.uq.edu.au> "Dumping a mounted filesystem" (May 1, 12:43am)
References: <Pine.GSO.4.33.0105010033270.21451-100000@mango.csee.uq.edu.au>
Sender: owner-linux-xfs@xxxxxxxxxxx
hi Chris,

On May 1, 12:43am, Chris Pascoe wrote:
> Subject: Dumping a mounted filesystem
> Hi,
> 
> I've been reading the linux-kernel mailing list and looking at the
> conversation regarding "SMP race in ext2 - metadata corruption.", where
> reading from a mounted filesystem's block device on 2.4 could cause
> corruption.  As such, I'm curious if any of the things mentioned there
> also affect XFS.
> 

Yes, I also read this thread - it was quite interesting.

Though it wasn't explicitly spelt out, I think the problem
that was being refered to with dump (not xfsdump) is that
it looks at the on-disk metadata structures (via libext2fs)
to figure out what needs dumping, etc.  tar, et al. use the
standard system call interfaces for traversing metadata and
obtaining file data, so aren't exposed to the problems
associated with dump-on-a-mounted-filesystem.

xfsdump uses an xfs-specific ioctl (bulkstat) to walk the
filesystem inodes, and doesn't refer directly to the on-disk
data structures either, so it should be as safe as tar.

cheers.

-- 
Nathan

<Prev in Thread] Current Thread [Next in Thread>