xfs
[Top] [All Lists]

Re: Minor bug with xfs_iget

To: William L Jones <jones@xxxxxxxxxxxxxxxxxx>
Subject: Re: Minor bug with xfs_iget
From: "Andi Kleen" <ak@xxxxxxx>
Date: Sun, 20 Aug 2000 00:40:40 +0200
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <200008192149.QAA28106@xxxxxxxxxxxxxxxxxxx>; from jones@xxxxxxxxxxxxxxxxxx on Sat, Aug 19, 2000 at 04:49:00PM -0500
References: <200008192149.QAA28106@xxxxxxxxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
On Sat, Aug 19, 2000 at 04:49:00PM -0500, William L Jones wrote:
> I think xfs_iget should always make sure that cached linux inode entries 
> should have the
> correct file/inode/... operatoins initialized in the linux inode. 
> 
> I only stumbled on this do to my work with open_inode_by_handle.   It is not 
> something
> that needs to be fixed quickly.  It is very unlikely to cause any problem 
> except with nfsd
> and only after the server crashes and some one does a 
> XFS_IOC_FSBULKSTAT_SINGLE
> with out doing a open_by_inode_handle immediately after. open_by_inode_handle 
> will fix 
> the file operation fields.

Actually knfsd does the iget too when its clients overrun the fhcache and it
has to reconstruct file handles, so you would likely see it over high load. 
An easy way to overrun the fhcache is just to do a recursive grep in a big 
source tree.

-Andi

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