Re: [PATCH 1/3] setup_arg_pages: diagnose excessive argument size

[pageexec () freemail hu]

On 14 Sep 2010 at 14:16, Roland McGrath wrote:

> > obviously an AT_ARGMAX computed at execve time would be based on the rlimits
> > as well and if later userland changed the rlimits, it'd be userland's problem,
> > not that of the kernel (or the kernel could refuse a change that would violate
> > its earlier promise).
>
> This would thoroughly defeat the purpose of adding the thing.  The
> only reason to have a new thing is so that userland does not have to
> mirror the kernel's policy (as it attempts to do now, with the 1/4
> calculation).  If the new thing is not something that userland can
> use consistently so as not to have to know what the kernel's actual
> policy is, then I don't see the point of it at all.

userland could never rely on the kernel's policy at all since get_arg_page
could have failed for more reasons than overstepping the currently hardcoded
ARG_MAX check in there. so what AT_ARGMAX would buy us is to allow the kernel
policy to change over time, but it's never been about guarantees, whether
POSIX wants such a thing or not.

> > >  auxv is only appropriate for things that
> > > are known at the time of the exec and won't change thereafter.
> >
> > you mean stuff like AT_EUID et al.? ;)
>
> The information that these give is about the conditions at startup.
> That's what they mean to userland, and userland only uses them to know
> the situation before it has made any calls.  The definition of AT_EUID
> is "effective user ID at program startup", and that fact does not
> change.

just for my own curiosity, where does this definition come from?