Re: [PATCH] ns: Wire up the setns system call for 2.6.40-rc1 or whatever

[Ralf Baechle <ralf@linux-mips.org>]

On Sat, May 28, 2011 at 10:55:55AM -0700, Linus Torvalds wrote:

> On Fri, May 27, 2011 at 7:28 PM, Eric W. Biederman
> <ebiederm@xmission.com> wrote:
> >
> > 32bit and 64bit on x86 are tested and working.  The rest I have looked
> > at closely and I can't find any problems.
> 
> So I really don't think this was even worth it. I applied the patch,
> but I think that you should just have done the architecture you
> tested, and left it to arch maintainers to add it as they will.
> 
> That's how we tend to do this, and it works. It also avoids surprises
> when people then invariably end up having clashes due to system calls
> being added. Even in just the 15 hours since you sent the email, I had
> merged more code from ARM, and the patch no longer applied to my tree.
> It's trivial to fix up, so that's not the problem, but the problem is
> with different people adding system calls resulting in re-numbering.

Which just happened on MIPS; I had a conflict between sendmmsg and sysns.

People other than the maintainer adding new syscalls routinely goes wrong
for this or other reasons.

> In other words, it's simply better to strive to have *one* entity in
> charge of picking the system call number, rather than do it like this.
> Ergo: leave it to architecture maintainers to minimize the issue of
> system call renumbering.

Amen.

  Ralf