[OpenAFS-devel] 1.3.86 LINUX largefile-support broken

Niklas Edmundsson Niklas.Edmundsson@hpc2n.umu.se
Mon, 1 Aug 2005 14:08:51 +0200 (MEST)


On Mon, 1 Aug 2005 chas@cmf.nrl.navy.mil wrote:

> In message <42EE0716.4000102@rzg.mpg.de>,Hartmut Reuter writes:
>> 3 gb is under the 32bit line! if you try to write a 5 gb file you will
>
> i believe the 32-bit line is traditionally considered 2G.  from the
> comments in afs, i would have guessed you were concerned about a
> 4G boundary:

2GB is the "traditional" non-largefile barrier (signed 32bit).

Breaking this, you move up to the 4GB (unsigned 32bit) barrier. Lots 
of code breaks here (mostly userland apps that tries to do clever 
things).

You usually want to test that crossing both barriers work. Remember 
that the writing usually just wraps around instead of fail when you're 
messing with the 4GB-barrier, so you'll want to check the output data 
and not just relying on that dd if=/dev/zero of=largefile succeeds...

Oh, and for those seeking good test data, StarWars Revelations DVD#2 
available at http://ftp.acc.umu.se/mirror/media/StarWars-Revelations/ 
has proven to be an excellent file to break applications with ;-)


/Nikke
-- 
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
  Niklas Edmundsson, Admin @ {acc,hpc2n}.umu.se     |    nikke@hpc2n.umu.se
---------------------------------------------------------------------------
  Take pot to the beach; leave no tern unstoned.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=