[PATCH] Multicast socket option

[PATCH] Multicast socket option

[Nivedita Singhvi]

After some discussion offline with Christoph Lameter and David Stevens 
regarding multicast behaviour in Linux, I'm submitting a slightly
modified patch from the one Christoph submitted earlier.

This patch provides a new socket option IP_MULTICAST_ALL.

In this case, default behaviour is _unchanged_ from the current
Linux standard. The socket option is set by default to provide 
original behaviour. Sockets wishing to receive data only from 
multicast groups they join explicitly will need to clear this 
socket option. 


Signed-off-by: Nivedita Singhvi <niv@us.ibm.com>
Signed-off-by: Christoph Lameter<cl@linux.com>
Acked-by: David Stevens <dlstevens@us.ibm.com>


diff -urN linux-2.6.29.2/include/linux/in.h linux-2.6.29.2.new/include/linux/in.h
--- linux-2.6.29.2/include/linux/in.h	2009-04-27 13:37:11.000000000 -0400
+++ linux-2.6.29.2.new/include/linux/in.h	2009-05-18 16:22:06.000000000 -0400
@@ -107,6 +107,7 @@
 #define MCAST_JOIN_SOURCE_GROUP		46
 #define MCAST_LEAVE_SOURCE_GROUP	47
 #define MCAST_MSFILTER			48
+#define IP_MULTICAST_ALL		49
 
 #define MCAST_EXCLUDE	0
 #define MCAST_INCLUDE	1
diff -urN linux-2.6.29.2/include/net/inet_sock.h linux-2.6.29.2.new/include/net/inet_sock.h
--- linux-2.6.29.2/include/net/inet_sock.h	2009-04-27 13:37:11.000000000 -0400
+++ linux-2.6.29.2.new/include/net/inet_sock.h	2009-05-18 16:22:06.000000000 -0400
@@ -130,7 +130,8 @@
 				freebind:1,
 				hdrincl:1,
 				mc_loop:1,
-				transparent:1;
+				transparent:1,
+				mc_all:1;
 	int			mc_index;
 	__be32			mc_addr;
 	struct ip_mc_socklist	*mc_list;
diff -urN linux-2.6.29.2/net/ipv4/af_inet.c linux-2.6.29.2.new/net/ipv4/af_inet.c
--- linux-2.6.29.2/net/ipv4/af_inet.c	2009-04-27 13:37:11.000000000 -0400
+++ linux-2.6.29.2.new/net/ipv4/af_inet.c	2009-05-18 16:22:06.000000000 -0400
@@ -376,6 +376,7 @@
 	inet->uc_ttl	= -1;
 	inet->mc_loop	= 1;
 	inet->mc_ttl	= 1;
+	inet->mc_all	= 1;
 	inet->mc_index	= 0;
 	inet->mc_list	= NULL;
 
diff -urN linux-2.6.29.2/net/ipv4/igmp.c linux-2.6.29.2.new/net/ipv4/igmp.c
--- linux-2.6.29.2/net/ipv4/igmp.c	2009-04-27 13:37:11.000000000 -0400
+++ linux-2.6.29.2.new/net/ipv4/igmp.c	2009-05-18 16:22:33.000000000 -0400
@@ -2196,7 +2196,7 @@
 			break;
 	}
 	if (!pmc)
-		return 1;
+		return inet->mc_all;
 	psl = pmc->sflist;
 	if (!psl)
 		return pmc->sfmode == MCAST_EXCLUDE;
diff -urN linux-2.6.29.2/net/ipv4/ip_sockglue.c linux-2.6.29.2.new/net/ipv4/ip_sockglue.c
--- linux-2.6.29.2/net/ipv4/ip_sockglue.c	2009-04-27 13:37:11.000000000 -0400
+++ linux-2.6.29.2.new/net/ipv4/ip_sockglue.c	2009-05-18 16:22:06.000000000 -0400
@@ -449,6 +449,7 @@
 			     (1<<IP_ROUTER_ALERT) | (1<<IP_FREEBIND) |
 			     (1<<IP_PASSSEC) | (1<<IP_TRANSPARENT))) ||
 	    optname == IP_MULTICAST_TTL ||
+	    optname == IP_MULTICAST_ALL ||
 	    optname == IP_MULTICAST_LOOP ||
 	    optname == IP_RECVORIGDSTADDR) {
 		if (optlen >= sizeof(int)) {
@@ -895,6 +896,13 @@
 		kfree(gsf);
 		break;
 	}
+	case IP_MULTICAST_ALL:
+		if (optlen<1)
+			goto e_inval;
+		if (val != 0 && val != 1)
+			goto e_inval;
+		inet->mc_all = val;
+		break;
 	case IP_ROUTER_ALERT:
 		err = ip_ra_control(sk, val ? 1 : 0, NULL);
 		break;
@@ -1147,6 +1155,9 @@
 		release_sock(sk);
 		return err;
 	}
+	case IP_MULTICAST_ALL:
+		val = inet->mc_all;
+		break;
 	case IP_PKTOPTIONS:
 	{
 		struct msghdr msg;

Re: [PATCH] Multicast socket option

[Rémi Denis-Courmont]

On Thursday 28 May 2009 20:00:46 ext Nivedita Singhvi wrote:
> After some discussion offline with Christoph Lameter and David Stevens
> regarding multicast behaviour in Linux, I'm submitting a slightly
> modified patch from the one Christoph submitted earlier.
>
> This patch provides a new socket option IP_MULTICAST_ALL.
>
> In this case, default behaviour is _unchanged_ from the current
> Linux standard. The socket option is set by default to provide
> original behaviour. Sockets wishing to receive data only from
> multicast groups they join explicitly will need to clear this
> socket option.

You can already achieve this by checking the destination address in the 
SOL_PKTINFO ancilliary data. Sure, it will cause extra context switches to 
process unwanted packets but it will work with any kernel version.

-- 
Rémi Denis-Courmont
Nokia Devices R&D, Maemo Software, Helsinki

Re: [PATCH] Multicast socket option

[Nivedita Singhvi]

Rémi Denis-Courmont wrote:

>> In this case, default behaviour is _unchanged_ from the current
>> Linux standard. The socket option is set by default to provide
>> original behaviour. Sockets wishing to receive data only from
>> multicast groups they join explicitly will need to clear this
>> socket option.
> 
> You can already achieve this by checking the destination address in the 
> SOL_PKTINFO ancilliary data. Sure, it will cause extra context switches to 
> process unwanted packets but it will work with any kernel version.
> 

True, and depending on your environment and workload, this
is a non-event or a big deal. For low latency, real time
and other performance-sensitive applications, this can be
an issue. Think lots (hundreds of threads, perhaps) listening
on different channels(groups). They will likely all get
woken, scheduled, possibly preempt running lower-priority
tasks to process this delivery that they don't need. Seen
across the system, the delivery by the kernel of packets
to user space processes which don't want them can be very
significant overhead (context switches, scheduling out/in,...).

thanks,
Nivedita

Re: [PATCH] Multicast socket option

[Neil Horman]

On Thu, May 28, 2009 at 10:00:46AM -0700, Nivedita Singhvi wrote:
> After some discussion offline with Christoph Lameter and David Stevens  
> regarding multicast behaviour in Linux, I'm submitting a slightly
> modified patch from the one Christoph submitted earlier.
>
> This patch provides a new socket option IP_MULTICAST_ALL.
>
> In this case, default behaviour is _unchanged_ from the current
> Linux standard. The socket option is set by default to provide original 
> behaviour. Sockets wishing to receive data only from multicast groups 
> they join explicitly will need to clear this socket option. 
>
>
> Signed-off-by: Nivedita Singhvi <niv@us.ibm.com>
> Signed-off-by: Christoph Lameter<cl@linux.com>
> Acked-by: David Stevens <dlstevens@us.ibm.com>
>
>
> diff -urN linux-2.6.29.2/include/linux/in.h linux-2.6.29.2.new/include/linux/in.h
> --- linux-2.6.29.2/include/linux/in.h	2009-04-27 13:37:11.000000000 -0400
> +++ linux-2.6.29.2.new/include/linux/in.h	2009-05-18 16:22:06.000000000 -0400
> @@ -107,6 +107,7 @@
> #define MCAST_JOIN_SOURCE_GROUP		46
> #define MCAST_LEAVE_SOURCE_GROUP	47
> #define MCAST_MSFILTER			48
> +#define IP_MULTICAST_ALL		49
>
> #define MCAST_EXCLUDE	0
> #define MCAST_INCLUDE	1
> diff -urN linux-2.6.29.2/include/net/inet_sock.h linux-2.6.29.2.new/include/net/inet_sock.h
> --- linux-2.6.29.2/include/net/inet_sock.h	2009-04-27 13:37:11.000000000 -0400
> +++ linux-2.6.29.2.new/include/net/inet_sock.h	2009-05-18 16:22:06.000000000 -0400
> @@ -130,7 +130,8 @@
> 				freebind:1,
> 				hdrincl:1,
> 				mc_loop:1,
> -				transparent:1;
> +				transparent:1,
> +				mc_all:1;
> 	int			mc_index;
> 	__be32			mc_addr;
> 	struct ip_mc_socklist	*mc_list;
> diff -urN linux-2.6.29.2/net/ipv4/af_inet.c linux-2.6.29.2.new/net/ipv4/af_inet.c
> --- linux-2.6.29.2/net/ipv4/af_inet.c	2009-04-27 13:37:11.000000000 -0400
> +++ linux-2.6.29.2.new/net/ipv4/af_inet.c	2009-05-18 16:22:06.000000000 -0400
> @@ -376,6 +376,7 @@
> 	inet->uc_ttl	= -1;
> 	inet->mc_loop	= 1;
> 	inet->mc_ttl	= 1;
> +	inet->mc_all	= 1;
> 	inet->mc_index	= 0;
> 	inet->mc_list	= NULL;
>
> diff -urN linux-2.6.29.2/net/ipv4/igmp.c linux-2.6.29.2.new/net/ipv4/igmp.c
> --- linux-2.6.29.2/net/ipv4/igmp.c	2009-04-27 13:37:11.000000000 -0400
> +++ linux-2.6.29.2.new/net/ipv4/igmp.c	2009-05-18 16:22:33.000000000 -0400
> @@ -2196,7 +2196,7 @@
> 			break;
> 	}
> 	if (!pmc)
> -		return 1;
> +		return inet->mc_all;

This change also filters out broadcasts sent to sockets listening on a given
port.  Is that what you want?

Neil

>

Re: [PATCH] Multicast socket option

[Christoph Lameter]

On Fri, 29 May 2009, Neil Horman wrote:

> > diff -urN linux-2.6.29.2/net/ipv4/igmp.c linux-2.6.29.2.new/net/ipv4/igmp.c
> > --- linux-2.6.29.2/net/ipv4/igmp.c	2009-04-27 13:37:11.000000000 -0400
> > +++ linux-2.6.29.2.new/net/ipv4/igmp.c	2009-05-18 16:22:33.000000000 -0400
> > @@ -2196,7 +2196,7 @@
> > 			break;
> > 	}
> > 	if (!pmc)
> > -		return 1;
> > +		return inet->mc_all;
>
> This change also filters out broadcasts sent to sockets listening on a given
> port.  Is that what you want?

Look at the beginning of the function:

   if (!ipv4_is_multicast(loc_addr))
                return 1;


will return 1 for broadcast.

Re: [PATCH] Multicast socket option

[Nivedita Singhvi]

Neil Horman wrote:

>> 	if (!pmc)
>> -		return 1;
>> +		return inet->mc_all;
> 
> This change also filters out broadcasts sent to sockets listening on a given
> port.  Is that what you want?

Yes, as Christoph pointed out, this does not impact
broadcast delivery, which will remain consistent with
original behaviour.

thanks,
Nivedita

Re: [PATCH] Multicast socket option

[David Miller]

From: Nivedita Singhvi <niv@us.ibm.com>
Date: Thu, 28 May 2009 10:00:46 -0700

> After some discussion offline with Christoph Lameter and David Stevens
> regarding multicast behaviour in Linux, I'm submitting a slightly
> modified patch from the one Christoph submitted earlier.
> 
> This patch provides a new socket option IP_MULTICAST_ALL.
> 
> In this case, default behaviour is _unchanged_ from the current
> Linux standard. The socket option is set by default to provide
> original behaviour. Sockets wishing to receive data only from
> multicast groups they join explicitly will need to clear this socket
> option.
> 
> 
> Signed-off-by: Nivedita Singhvi <niv@us.ibm.com>
> Signed-off-by: Christoph Lameter<cl@linux.com>
> Acked-by: David Stevens <dlstevens@us.ibm.com>

Applied, thanks.