Re: [patch 7/8] fdmap v2 - implement sys_socket2

[Matt Mackall <mpm@selenic.com>]

On Sat, Jun 09, 2007 at 09:49:07PM +0100, Al Viro wrote:
> On Sat, Jun 09, 2007 at 01:21:24PM -0700, Linus Torvalds wrote:
> > Which is why you'd like to do the *initial* operation with a flag that 
> > says "please set the FD_CLOEXEC flag on the file descriptor", so that you 
> > *atomically* install the file file descriptor and set the FD_CLOEXEC bit.
> > 
> > It's trivial to do for open(), but there are about a million ways to get a 
> > file descriptor, and open() is just about the *only* one of those that 
> > actually takes a "flags" field that can be used to tell the kernel.
> 
> Eww...  Idea of pipe(2) taking flags as argument...
> 
> BTW, you also need that for recvmsg() (SCM_RIGHTS) and fsckloads of
> syscalls we've got duplicating open() for no good reason (and no, "BSD
> folks did it for sockets, so we'll do it for tons of new subsystems" doesn't
> really qualify ;-/).
> 
> I don't know if your indirect is a good idea in that respect, actually.
> AFAICS, you are suggesting per-syscall meanings of the flags, so it really
> smells like YAMultiplexor, free for abuse.

Well, of course it sucks. The question is does it suck less than
adding dozens of new syscalls to patch up problems with POSIX?

I don't think we can get this right in one iteration, so I actually
prefer Linus's approach as less likely to leave numerous vestiges of
incremental improvements around.

On the other hand, we don't want to end up in a future where the bulk
of our calls are through sys_indirect! Given the number of important
APIs that have gotten slight semantic bumps, that's a real
possibility.

-- 
Mathematics is the supreme nostalgia of our time.