Re: Resource limits interface proposal [was: pull request for writable limits]

[Linus Torvalds <torvalds@linux-foundation.org>]

On Thu, 6 May 2010, Alexey Dobriyan wrote:

> On Wed, May 5, 2010 at 6:08 PM, Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
> >
> > Ok, I'm not entirely sure we need to care specially about INFINITY,
> > _especially_ since INF is really rather big in 64 bits. So to some degree,
> > making things 64-bit is _less_ likely to make INFINITIES a problem.
> 
> I'm _sure_, someone will mention bloat and performance degradation
> on 32-bit, so there will be config option for 32-bit in-kernel limits.
> And if there will be config option, infinities are better be separated.

Umm. Why?

Adding an INF field will make it essentially impossible to use that thing 
as a compatibility function, and/or together with the legacy rlimit() 
system call. And that is going to be 99.999% of all uses.

I don't think you understand how system calls work. People don't use 
Linux-only features. It's been shown over and over and over again. People 
use the standard interfaces, and they don't _have_ that INF field. We can 
extend ranges by just updating structures (see "stat()" and friends) and a 
recompile will make it use higher limits, but we can't make programs use 
new features.

So there is _no_ upside. There would be _no_ programs ever using it. 

Whether the internal limits for the kernel are 32-bit or not makes 
absolutely zero difference. And my argument is that _if_ they are 64-bit, 
that makes the INF field even _less_ interesting. So they go from "zero 
practical difference" in 32-bit to "f*ck it, nobody can _possibly_ care 
even in theory" in 64-bit.

If the internal limits are 32-bit, then you just turn any 64-bit value 
larger than the (32-bit) RLIM_INFINITY into a RLIM_INFINITY. And that's as 
good as you can get with mixing with the legacy interface anyway.

		Linus