xfs
[Top] [All Lists]

Re: Patch to make XFS to work on ramdisk

To: Bo Yang <kealia@xxxxxxxxx>
Subject: Re: Patch to make XFS to work on ramdisk
From: Nathan Scott <nathans@xxxxxxx>
Date: Fri, 6 Aug 2004 18:55:07 +1000
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <3b94088e04080517166fd22dd3@xxxxxxxxxxxxxx>
References: <3b94088e04080517166fd22dd3@xxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/1.5.3i
On Thu, Aug 05, 2004 at 05:16:45PM -0700, Bo Yang wrote:
> Hi,
> 
> Appended is a patch to make XFS to work on ramdisk.

Thanks.

> (rd_ioctl() in kernel 2.4.25).  I assume the expected semantics of
> BLKFLSBUG on a normal block device is to: 1) flush buffers to device;

Heh, "BLKFLSBUG" - Freudian slip?

> 2)  release buffers from main memory. Apparently a ramdisk cannot do
> both at the same time. It just invalidates all its pages in page cache
> and returns. So the XFS meta data is lost.

Yep.  Actually saw discussion of this on linux-kernel not too
long ago, just didn't register that it'd affect us in mkfs.

> it's the ramdisk implementation that should get fixed. I changed XFS
> tools a little bit so it does not call BLKFLSBUF on a ramdisk.

Its mkfs.xfs that should be fixed, yes - I see the kernel code
does a very similar thing when initialising the initrd.

I'll push a slight variant on your fix into the oss tree on Monday
(after a bit more testing here).

thanks!

-- 
Nathan


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