Re: For the almost 4-year anniversary: O_CLOEXEC again

[Ulrich Drepper <drepper@redhat.com>]

Jamie Lokier wrote:

> Since O_CLOEXEC is non-portable, why not implement the per-thread
> switch?
> 
> Signal handlers that call open() will have to be aware of it, but
> that's ok: an application will only set the switch if it knows what
> that implies.

It's not OK since library functions can install signal handlers and they
need not be aware of everything the application does.

Any globally visible change in the behavior can have negative effects.

Yes, there are other calls but open() is far more critical since it is
now promoted not only to an interface to actually do work.  We are
supported to read kernel and system informatino from /proc which
requires open().

And no, oen() is not the only signal-safe function.  If you don't have
the POSIX specs handy, look at a recent signal(2) man page which lists
the signal-safe functions.  accept(), socket() are on the list among others.

-- 
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖