Re: 2.6.28, rlimits, performance and debian etch

[Florian Weimer <fw@deneb.enyo.de>]

* David Daney:

> One problem is that for values of RLIMIT_NOFILE less than something
> like 4096, it is much faster to call sys_close() on all possible
> values than iterate through a handful of open files from /proc/self/fd
> using opendir(3)/readdir(3).

Really?  Yuck.

> The real solution is to convert your user space programs to use the
> new syscalls that allow for race-free setting of close-on-exec.
> Then you no longer need to mess around with iterating over these
> things.

These system calls are too recent to use in the next two Debian
releases.  In addition, we can't really be sure that all libraries use
the new calls.

I find the design of the CLOEXEC business somewhat revulsive, by the
way.  It reminds me of those DoSomethingEx APIs in another platform.
The *_at system calls are a similar wart.  Even with this stuff, I
still can't safely open a file with a different effective user ID in a
multithreaded application, or create a AF_UNIX socket with specific
file system permissions.  Some thread-specific context with what have
been traditionally considered per-process attributes might have been
better. 8-(