|| ||Linus Torvalds <torvalds-AT-linux-foundation.org>|
|| ||Matt Mackall <mpm-AT-selenic.com>|
|| ||Re: [patch 7/8] fdmap v2 - implement sys_socket2|
|| ||Thu, 7 Jun 2007 19:25:03 -0700 (PDT)|
|| ||Alan Cox <alan-AT-lxorguk.ukuu.org.uk>,
Davide Libenzi <davidel-AT-xmailserver.org>,
Linux Kernel Mailing List <linux-kernel-AT-vger.kernel.org>,
Andrew Morton <akpm-AT-linux-foundation.org>,
Ulrich Drepper <drepper-AT-redhat.com>,
Ingo Molnar <mingo-AT-elte.hu>, Eric Dumazet <dada1-AT-cosmosbay.com>|
On Thu, 7 Jun 2007, Matt Mackall wrote:
> First, how does this work in-kernel? Does it set a flag in the thread
> struct that magically gets used in the actual syscall? Or do we pass
> flags down to the sys_foo() function in some manner?
Set a flag in the thread-struct.
In fact, that's how "access()" already works.
And yes, syslets would need to have their own thread-structs and/or
save/restore the thing.
> Second, I think we're likely to run out of flag bits really quickly as
> this is a good dumping spot for patching up our many slightly
> brain-damaged APIs (be they POSIX or Linux-specific).
Well, I do suspect that we'd need to basically make the flags be
per-system call. With "common features" (ie a system call that doesn't
return a file descriptor would re-use the bit for "nonlinear-fd" for
something else, while a system call that doesn't do path lookup would use
all the LOOKUP_xyzzy bits for something else).
I agree that if we kept flags _totally_ separate, we'd run out of them
really quickly. But I don't think we want to ever be in the situation
where _one_ set of system calls would need that many flags. If we get
there, we'd really be much better off with a new system call!
> Third, can I do sys_indirect(sys_indirect(foo, args), flags1), flags2)?
I'd say no.
> Fourth, can we do sys_indirect(foo, args, flags | ASYNC) and get most
> of the way to merging this with the syslet proposal?
I think that may well be a really good idea.
to post comments)