[fuse-devel] fuse lseek bug

[Mohit A. <ext...@gm...>]

Hello,

I tried running a test by having an open source database (postgres) write to
a fuse mounted filesystem. I used the fusexmp_fh to do the fuse mount.
Postgres hit into what seems to be a fuse bug related to lseek - basically,
it seems that if there's a race between an lseek and a write, fuse can
sometimes return wrong results with an lseek.

A patch for the bug seems to be available at:
http://linux.derkeiler.com/Mailing-Lists/Kernel/2008-04/msg10586.html

First, I'd like someone to confirm that the above bug is indeed what
postgres is running into. If this is a separate bug, is this a known one ?
Is a patch available ?

It seems the above mentioned patch is only available for 2.6.26 which isn't
a kernel that's been released. I'm using 2.6.20 - will the patch work for
this kernel ? I'm using fuse version 2.7.0 - is this bug otherwise fixed in
later fuse versions ? I'd appreciate some help knowing how can I upgrade
either my current kernel or the fuse version I'm using to get around this
bug. Actually using an 'rc' version of the 2.6.26 kernel is probably not an
option for me.


- Mohit