Re: [PATCH v2] Implement close-on-fork

[Christian Brauner <brauner@kernel.org>]

On Tue, Jun 28, 2022 at 01:38:07PM +0000, David Laight wrote:
> From: Christian Brauner
> > Sent: 28 June 2022 14:13
> > 
> > On Sun, Jun 19, 2022 at 11:42:28AM +0100, Ralph Corderoy wrote:
> > > Hi Matthew, thanks for replying.
> > >
> > > > > The need for O_CLOFORK might be made more clear by looking at a
> > > > > long-standing Go issue, i.e. unrelated to system(3), which was started
> > > > > in 2017 by Russ Cox when he summed up the current race-condition
> > > > > behaviour of trying to execve(2) a newly created file:
> > > > > https://github.com/golang/go/issues/22315.
> > > >
> > > > The problem is that people advocating for O_CLOFORK understand its
> > > > value, but not its cost.  Other google employees have a system which
> > > > has literally millions of file descriptors in a single process.
> > > > Having to maintain this extra state per-fd is a cost they don't want
> > > > to pay (and have been quite vocal about earlier in this thread).
> > >
> > > So do you agree the userspace issue is best solved by *_CLOFORK and the
> > > problem is how to implement *_CLOFORK at an acceptable cost?
> > >
> > > OTOH David Laight was making suggestions on moving the load to the
> > > fork/exec path earlier in the thread, but OTOH Al Viro mentioned a
> > > ‘portable solution’, though that could have been to a specific issue
> > > rather than the more general case.
> > >
> > > How would you recommend approaching an acceptable cost is progressed?
> > > Iterate on patch versions?  Open a bugzilla.kernel.org for central
> > > tracking and linking from the other projects?  ..?
> > 
> > Quoting from that go thread
> > 
> > "If the OS had a "close all fds above x", we could use that. (I don't know of any that do, but it sure
> > would help.)"
> > 
> > So why can't this be solved with:
> > close_range(fd_first, fd_last, CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_UNSHARE)?
> > e.g.
> > close_range(100, ~0U, CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_UNSHARE)?
> 
> That is a relatively recent linux system call.
> Although it can be (mostly) emulated by reading /proc/fd
> - but that may not be mounted.
> 
> In any case another thread can open an fd between the close_range()
> and fork() calls.

The CLOSE_RANGE_UNSHARE gives the calling thread a private file
descriptor table before marking fs close-on-exec.

close_range(100, ~0U, CLOSE_RANGE_CLOEXEC | CLOSE_RANGE_UNSHARE)?