[PATCH] RDS: Add AF and PF defines for RDS sockets

[PATCH] RDS: Add AF and PF defines for RDS sockets

[Andy Grover]

RDS is a reliable datagram protocol used for IPC on Oracle
database clusters. This adds address and protocol family numbers
for it.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
---
 include/linux/socket.h |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)

diff --git a/include/linux/socket.h b/include/linux/socket.h
index dc5086f..e1f7764 100644
--- a/include/linux/socket.h
+++ b/include/linux/socket.h
@@ -190,7 +190,8 @@ struct ucred {
 #define AF_IUCV		32	/* IUCV sockets			*/
 #define AF_RXRPC	33	/* RxRPC sockets 		*/
 #define AF_ISDN		34	/* mISDN sockets 		*/
-#define AF_MAX		35	/* For now.. */
+#define AF_RDS		35	/* RDS sockets 			*/
+#define AF_MAX		36	/* For now.. */
 
 /* Protocol families, same as address families. */
 #define PF_UNSPEC	AF_UNSPEC
@@ -227,6 +228,7 @@ struct ucred {
 #define PF_IUCV		AF_IUCV
 #define PF_RXRPC	AF_RXRPC
 #define PF_ISDN		AF_ISDN
+#define PF_RDS		AF_RDS
 #define PF_MAX		AF_MAX
 
 /* Maximum queue length specifiable by listen.  */
-- 
1.5.4.rc3

Re: [PATCH] RDS: Add AF and PF defines for RDS sockets

[Marcel Holtmann]

Hi Andy,

> RDS is a reliable datagram protocol used for IPC on Oracle
> database clusters. This adds address and protocol family numbers
> for it.

and where is the protocol? Adding new constants without posting the  
implementation normally doesn't fly anymore.

Regards

Marcel

Re: [PATCH] RDS: Add AF and PF defines for RDS sockets

[David Miller]

From: Marcel Holtmann <marcel@holtmann.org>
Date: Tue, 26 Aug 2008 06:16:28 +0200

> and where is the protocol? Adding new constants without posting the
> implementation normally doesn't fly anymore.

That's correct :)

Re: [PATCH] RDS: Add AF and PF defines for RDS sockets

[David Woodhouse]

On Mon, 2008-08-25 at 21:29 -0700, David Miller wrote:
> From: Marcel Holtmann <marcel@holtmann.org>
> Date: Tue, 26 Aug 2008 06:16:28 +0200
> 
> > and where is the protocol? Adding new constants without posting the
> > implementation normally doesn't fly anymore.
> 
> That's correct :)

That's kind of unfortunate, and leads to the problems we have with
compatibility for AF_ISDN, where userspace built for the _older_
versions has a different value.

Is there really any harm in reserving it up-front?

-- 
David Woodhouse                            Open Source Technology Centre
David.Woodhouse@intel.com                              Intel Corporation

Re: [PATCH] RDS: Add AF and PF defines for RDS sockets

[David Miller]

From: David Woodhouse <dwmw2@infradead.org>
Date: Thu, 28 Aug 2008 08:06:40 +0100

> On Mon, 2008-08-25 at 21:29 -0700, David Miller wrote:
> > From: Marcel Holtmann <marcel@holtmann.org>
> > Date: Tue, 26 Aug 2008 06:16:28 +0200
> > 
> > > and where is the protocol? Adding new constants without posting the
> > > implementation normally doesn't fly anymore.
> > 
> > That's correct :)
> 
> That's kind of unfortunate, and leads to the problems we have with
> compatibility for AF_ISDN, where userspace built for the _older_
> versions has a different value.
> 
> Is there really any harm in reserving it up-front?

If you're commiting a userspace to a kernel component
that hasn't been blessed upstream yet, you're already
up the creek without a paddle.

Re: [PATCH] RDS: Add AF and PF defines for RDS sockets

[Andrew Grover]

(reviving an old thread, new comments at bottom)

On Thu, Aug 28, 2008 at 12:24 AM, David Miller <davem@davemloft.net> wrote:
> From: David Woodhouse <dwmw2@infradead.org>
> Date: Thu, 28 Aug 2008 08:06:40 +0100
>
>> On Mon, 2008-08-25 at 21:29 -0700, David Miller wrote:
>> > From: Marcel Holtmann <marcel@holtmann.org>
>> > Date: Tue, 26 Aug 2008 06:16:28 +0200
>> >
>> > > and where is the protocol? Adding new constants without posting the
>> > > implementation normally doesn't fly anymore.
>> >
>> > That's correct :)
>>
>> That's kind of unfortunate, and leads to the problems we have with
>> compatibility for AF_ISDN, where userspace built for the _older_
>> versions has a different value.
>>
>> Is there really any harm in reserving it up-front?
>
> If you're commiting a userspace to a kernel component
> that hasn't been blessed upstream yet, you're already
> up the creek without a paddle.

The RDS protocol is already in production use by Oracle, and has been
in the OFED distribution for a long time. It's under active
development, and we'll be submitting it soon for mainline.

So userspace is already committed to a kernel component that isn't in
mainline, and we *are* up a creek a little bit. User code has to do
some hacky things to determine RDS's protocol number (it's variable).
The sooner we get an actual number allocated, the sooner userspace can
start using it, and start the clock ticking on when it can remove the
old variable protocol num hack.

Basically: we're gonna submit the code. Who knows how many rounds of
reviews it will take before it goes in, but we're committed to getting
it in. We just want a number now so userspace can start doing things
correctly right away.

Let me know and I can repost the patch.

Regards -- Andy

Re: [PATCH] RDS: Add AF and PF defines for RDS sockets

[David Miller]

From: "Andrew Grover" <andy.grover@gmail.com>
Date: Thu, 11 Sep 2008 11:44:39 -0700

> (reviving an old thread, new comments at bottom)
> 
> On Thu, Aug 28, 2008 at 12:24 AM, David Miller <davem@davemloft.net> wrote:
> > From: David Woodhouse <dwmw2@infradead.org>
> > Date: Thu, 28 Aug 2008 08:06:40 +0100
> >
> >> On Mon, 2008-08-25 at 21:29 -0700, David Miller wrote:
> >> > From: Marcel Holtmann <marcel@holtmann.org>
> >> > Date: Tue, 26 Aug 2008 06:16:28 +0200
> >> >
> >> > > and where is the protocol? Adding new constants without posting the
> >> > > implementation normally doesn't fly anymore.
> >> >
> >> > That's correct :)
> >>
> >> That's kind of unfortunate, and leads to the problems we have with
> >> compatibility for AF_ISDN, where userspace built for the _older_
> >> versions has a different value.
> >>
> >> Is there really any harm in reserving it up-front?
> >
> > If you're commiting a userspace to a kernel component
> > that hasn't been blessed upstream yet, you're already
> > up the creek without a paddle.
> 
> The RDS protocol is already in production use by Oracle, and has been
> in the OFED distribution for a long time. It's under active
> development, and we'll be submitting it soon for mainline.

So?!??!  Upstream doesn't give a crap about what someone is happening
internally.  It doesn't matter, it's irrelevant.

> We just want a number now so userspace can start doing things
> correctly right away.

N-O spells "no".