Re: Full support for more than 1024fds.

Andi Kleen (andi@mlm.extern.lrz-muenchen.de)
19 Jan 1997 18:30:52 +0100


jum@anubis.han.de (Jens-Uwe Mager) writes:

> In article <97Jan19.015504-0500_est.212997-11748+2516@vger.rutgers.edu>,
> Andi Kleen <andi@mlm.extern.lrz-muenchen.de> wrote:
>
> > Here is a new version of the patch to add support for more than
> > 1024 files per process. I added a /proc/sys/kernel/nr-open now,
> > so you can tune NR_OPEN dynamically now. Applications have still
> > to be recompiled (with an appropiate -D__USER_FDSETSIZE value).
>
> Shouldn't that be -DFD_SETSIZE=value to be compatible with the
> implementation that a lot of other Unix variants use, notably the BSD Unix
> family?

I choosed __USER_FDSETSIZE to not conflict with the ANSI C namespace.
It's defined in linux/posix_types.h. This file has to conform to the ANSI-C
namespace so I can't add a name without __ in front of it. So someone
has to add a:

#ifdef FD_SETSIZE
#define __USER_SETSIZE FD_SETSIZE
#endif

to the libc sys/types.h (that's where *BSD defines this). Any better ideas?

-Andi