[Top] [All Lists]

Re: 2.4.11 and large files

To: Robert Sander <robert.sander@xxxxxxxxxxxxxxx>
Subject: Re: 2.4.11 and large files
From: Nathan Scott <nathans@xxxxxxx>
Date: Fri, 12 Oct 2001 11:33:07 +1100
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <20011011095412.A9418@xxxxxxxxxxxxxx>; from robert.sander@xxxxxxxxxxxxxxx on Thu, Oct 11, 2001 at 09:54:12AM +0200
References: <news2mail-slrn9s8feg.ti2.gurubert-news@xxxxxxxxxxxxxxxxxxxxx> <3BC46867.3895800C@xxxxxxxxxxxxxxx> <news2mail-slrn9s8tn8.v5f.gurubert-news@xxxxxxxxxxxxxxxxxxxxx> <3BC48047.5959D636@xxxxxxxxxxxxxxx> <news2mail-slrn9s91kr.a8p.gurubert-news@xxxxxxxxxxxxxxxxxxxxx> <20011011122630.C486887@xxxxxxxxxxxxxxxxxxxxxxxx> <20011011095412.A9418@xxxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
User-agent: Mutt/1.2.5i

On Thu, Oct 11, 2001 at 09:54:12AM +0200, Robert Sander wrote:
> On Thu, Oct 11, 2001 at 12:26:31PM +1100, Nathan Scott wrote:
> > Could you try switching OFF XFS quota support
> > (but leaving the normal Quota support ON), and see if the problem
> > persists?
> When switching off only XFS quotas it works! Seems to be nailed down:

Uhrm.  Well, thats very odd.

Let me see if we're on the same page - you have a system with no XFS
filesystems, and no filesystems of any type with any form of quota
at all.  And then you dd one >2Gb partition to another, same size.
This fails with CONFIG_XFS_QUOTA set, but works without it.  It works
also before 2.4.10, independent of any quota config options.

Your failure is seen in dd(1), copying at the 2Gb mark, it prints out
something like "File size limit exceeded" (ie. it receives a SIGXFSZ
signal), and exits.

Is that all correct? 

Some points of interest, may or may not be relevent in helping find
the problem:
- So far, I haven't been able to reproduce the problem, eg. when
dd'ing one 3Gb partition to another on my local machine, all quota
switched on, with an XFS CVS kernel;
- There is no possibility that any XFS quota code is being exectued
in your setup (no XFS code at all), other than at bootup - where we
don't do much, mainly just some global initialization (locks, memory
allocation, etc) - so, I'm a bit amazed that switching off XFS quota
can change the outcome at all here;
- I'm not sure how even the VFS quota can be involved here, since
that quota code should only come into play for a filesystem where we
have quota enabled;
- SIGXFSZ has nothing to do with filesystem quota, it is sent to a
process from truncate/write when the file offset reaches certain
limits (in contrast, quota being exceeded would fail the write with
errno set to EDQUOT, not by sending a signal).

Its all very, very strange... I'm a little inclined to suspect the
recent block device changes in 2.4.10/11 - the only XFS quota change
recently was some minor changes to the quotactl interface, and this
code wouldn't be executed at all in your setup.

Do you see any system console messages when you get the error?



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