xfs
[Top] [All Lists]

Re: XFS as Root filesystem

To: stb@xxxxxxxxx (Klaus Strebel)
Subject: Re: XFS as Root filesystem
From: Jim Mostek <mostek@xxxxxxx>
Date: Wed, 5 Apr 2000 07:48:21 -0500 (CDT)
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <38EB2FEA.E22A0E8B@xxxxxxxxx> from "Klaus Strebel" at Apr 05, 2000 02:22:02 PM
Sender: owner-linux-xfs@xxxxxxxxxxx
        .
        .
        .
>the other OS's root, ex. 6.5, mounting ro of the 6.4 filesystems (esp.
>the root) on boot failes, due to dirty filesystem (on XFS impossible
>;-). So the filesystems must 1. be mounted rw, umounted, and then
>mounted ro (what i wanted). Behaps, it helps, if you ensure, that lilo
        .
        .
        .

It would be pretty simple to have a new flag that let recovery run for
a read-only mount. Recovery checks to see if the head/tail are == and then
allows the read-only mount.

If you look at the code, the XFS_MOUNT_NORECOVERY flag is set
to skip recovery. But, we must already have a consistent file system when
this flag is used.  This is not what we should use for a read-only mount
of root. We need a separate flag. xlog_recovery would then do recovery
even on read-only mounts. Anyone have time to do this? I'll add this to
the mount as root work item.

Jim

>
>Tan Pong Heng wrote:
>> 
>> Already done that, that is why I am certain that the hang occur after the
>> successful
>> mount. The last place I checked was in main.c which was trying to do
>> "exec(init...)"
>> it never reach the printk after all the possible init path. (Or, it may be
>> taking too long
>> for me to notice...)
>Hi all,
>
>the issue might be, that xfs has big problems, being mounted ro on boot.
>This is the same behaviour as on IRIX (i have a Origin 200 with IRIX 6.4
>on dksc(0,1,0) and IRIX 6.5 on dksc(0,2,0)). When i halt and reboot from
>the other OS's root, ex. 6.5, mounting ro of the 6.4 filesystems (esp.
>the root) on boot failes, due to dirty filesystem (on XFS impossible
>;-). So the filesystems must 1. be mounted rw, umounted, and then
>mounted ro (what i wanted). Behaps, it helps, if you ensure, that lilo
>isn't booting with to ro-flag. Btw. ext3fs has the same problem. I think
>Steve Tweedie solved it in ext2-0.0.2d (didn't he, had no time to test).
>
>Bye
>Klaus
>-- 
>Klaus Strebel
>stb@xxxxxxxxx
>EIGNER + PARTNER AG   - The Engineering Warehouse Company -
><http://www.ep-ag.com>
>-----------------------------------------------------------------------
>


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