Subject: Re: problem with tmpfs and linux emulation?
To: None <>
From: Michael van Elst <>
List: current-users
Date: 11/14/2005 01:05:19 (Pavel Cahyna) writes:

>On Mon, Nov 14, 2005 at 09:34:29AM +1100, Daniel Carosone wrote:
>> On Sun, Nov 13, 2005 at 11:31:55PM +0100, Pavel Cahyna wrote:
>> > How can d_off be invalid? It was returned by the kernel before, so it
>> > should be valid, no?
>> Which takes us right back to tmpfs directory cookies.

>So is it a bug in tmpfs or glibc? If I understand it correctly, the
>problem is in tmpfs. (Does it return directory cookies that it does not
>accept as offsets for lseek() ?)

tmpfs (and NFS) do return cookies in d_off that are not usuable
for lseek(). If that is a bug then it is not in tmpfs but in the
Linux emulation layer.

But in the openoffice case this is not the problem.

The problem is that openoffice is compiled for a 32bit d_off and gets
a 64bit d_off from the kernel. The same problem can hit Linux natively
when it reads from a NFSv3 server.

                                Michael van Elst
                                "A potential Snark may lurk in every tree."