[Top] [All Lists]

Re: Filesystem Consistency Issues

To: Steve Lord <lord@xxxxxxx>
Subject: Re: Filesystem Consistency Issues
From: Eric Sandeen <sandeen@xxxxxxx>
Date: Wed, 07 Dec 2005 23:39:58 -0600
Cc: Linux-XFS Mailing List <linux-xfs@xxxxxxxxxxx>
In-reply-to: <439745EB.3050903@xxxxxxx>
References: <20051204142506.GE2605@xxxxxxxxxxx> <17299.32438.100855.316477@xxxxxxxxxxxxxxxxxx> <20051207200333.GA5279@xxxxxxxxxxxxxx> <439745EB.3050903@xxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mozilla Thunderbird 1.0.6 (Macintosh/20050716)
Steve Lord wrote:
XFS always does a scan of the log on mount, it always replays the log if it
is not terminated with an unmount record. The fact that the filesystem is
mounted readonly during startup should have no affect on this. XFS will run
recovery on a read only mount.

The only time xfs will not replay the log on mount is if the -block device- itself is readonly; for example a read-only snapshot or... um... maybe a cdrom? :) But Steve is right (of course!) - the "ro" option does not prevent log replay.

On inode unlink, XFS adds inodes to an unlinked inode list until the space in the inode is freed when the last reference count is released. Removing these inodes is part of the recovery process. It sounds like there may be an issue
with this part of recovery.



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