More on RLIMIT_DATA vs. RLIMIT_AS

Following up to my previous post, here's a thread on the linux-kernel mailing list (from 2007) about the meanings of RLIMIT_DATA and RLIMIT_AS and their usefulness, with notes on the history of how these limits came about, and more importantly, why RLIMIT_DATA is ignored. The POSIX standard is referred to, as well as the implementation of malloc() in GNU libc.

A comment from Hugh Dickins:

I remember thinking that the idea of data ulimit had become obsolete (just preserved for compatibility) back when mmap() got invented and used for dynamic libraries. I think that's when they brought in RLIMIT_AS, something which could make sense in the mmap() world.

Regardless, this patch does not seem to have made it in: I looked at linux-2.6.36-rc8. A bug was filed about this in 2008, updated most recently in March 2010. And this issue is now also being tracked on Fedora/RedHat's bugzilla.

EDIT: This fortune is relevant.

Note that nobody reads every post in linux-kernel. In fact, nobody who expects to have time left over to actually do any real kernel work will read even half. Except Alan Cox, but he's actually not human, but about a thousand gnomes working in under-ground caves in Swansea. None of the individual gnomes read all the postings either, they just work together really well.  -- Linus Torvalds