xfs
[Top] [All Lists]

Re: Detected potential for stack overflows, stack left: 796 bytes

To: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Subject: Re: Detected potential for stack overflows, stack left: 796 bytes
From: Andi Kleen <ak@xxxxxxx>
Date: Thu, 22 Aug 2002 21:11:53 +0200
Cc: Andi Kleen <ak@xxxxxxx>, Steve Lord <lord@xxxxxxx>, Eric Sandeen <sandeen@xxxxxxx>, linux-xfs@xxxxxxxxxxx
In-reply-to: <20020822200644.A29301@infradead.org>
References: <3D652F1A.3080005@Lehigh.EDU> <1030041962.8450.18.camel@stout.americas.sgi.com> <20020822205014.A9535@wotan.suse.de> <1030042219.10044.107.camel@jen.americas.sgi.com> <20020822210134.A11739@wotan.suse.de> <20020822200644.A29301@infradead.org>
Sender: owner-linux-xfs@xxxxxxxxxxx
User-agent: Mutt/1.3.22.1i
On Thu, Aug 22, 2002 at 08:06:44PM +0100, Christoph Hellwig wrote:
> On Thu, Aug 22, 2002 at 09:01:34PM +0200, Andi Kleen wrote:
> > The only thing I have pending is a partial ioctl32 translation for x86-64
> > (and possible ia64 too) for some XFS ioctls.
> 
> Any reason why it doesn't work for sparc64/parisc/mips64?

Those don't need it (at least not for those ioctls I converted - i didn't
attempt the more complex xfsdump ioctls) 

IA64/x86-64 have the unique problem that the 32bit port has a different
alignment (non natural 4 bytes) for long long than the 64bit port.
Some of the XFS ioctls hit this because they do not explicitely pad
u64 to natural boundaries. sparc64 etc. are fine because their 32bit
ports have naturally aligned long long.

I implemented it only for x86-64 now. IA64 currently has no dynamic
ioctl32 registration interface for modules.

If you wanted to make xfsdump which has complex data structures with pointers
in ioctls you would also need to cover the other architectures. Would be 
lots of work though and probably not worth the effort.

-Andi


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