Netdev List
 help / color / mirror / Atom feed
* [PATCH net 0/1] ipv6: mcast: fix default-interface source filter selection
@ 2026-11-12 15:19 Zihan Xi
  2026-11-12 15:19 ` [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters Zihan Xi
  0 siblings, 1 reply; 5+ messages in thread
From: Zihan Xi @ 2026-11-12 15:19 UTC (permalink / raw)
  To: netdev; +Cc: dsahern, idosch, davem, edumazet, pabeni, horms, vega, zihanx

Hi Linux kernel maintainers,

We found an issue in net/ipv6/mcast.c. The bug is reachable by an
unprivileged process using the IPv6 multicast source-filter socket API.
The relevant details are provided below.

---- details below ----

An LLM was used to help create this patch.

This bug is tracked at: https://bugtracker.nebusec.ai/f/Z8p (Z8p)

Bug details:

When gsr_interface is zero, ip6_mc_source() resolves the multicast route
to a device but previously searched socket memberships with a wildcard
interface index. With the same group joined on multiple interfaces, the
socket filter could be updated on one interface while the aggregate
filter was updated on the route-selected interface. The patch uses the
resolved interface index for both membership lookup and last-source
membership removal.

Reproducer:

Build and run as an unprivileged user in a Linux guest where ff0e::1234
routes through eth0 and lo is available:

    gcc -O2 -Wall -Wextra -o mini_poc mini_poc.c
    ip -6 route get ff0e::1234
    ./mini_poc

The test used a 2 vCPU, 2 GiB x86 QEMU/KVM environment. On the baseline,
the source appeared in socket[lo] while socket[eth0] stayed empty. With
the patch, it appeared in socket[eth0] while socket[lo] stayed empty.
No kernel crash was observed.

------BEGIN PoC------

#define _GNU_SOURCE

#include <arpa/inet.h>
#include <net/if.h>
#include <netinet/in.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/socket.h>
#include <unistd.h>

#ifndef GROUP_FILTER_SIZE
#define GROUP_FILTER_SIZE(n) \
	(sizeof(struct group_filter) - sizeof(struct sockaddr_storage) + \
	 (n) * sizeof(struct sockaddr_storage))
#endif

static void set_group(struct sockaddr_storage *ss, const char *addr)
{
	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)ss;

	sin6->sin6_family = AF_INET6;
	if (inet_pton(AF_INET6, addr, &sin6->sin6_addr) != 1) {
		perror("inet_pton");
		exit(EXIT_FAILURE);
	}
}

static void show_filter(int fd, unsigned int ifindex, const char *ifname)
{
	struct group_filter *gf = calloc(1, GROUP_FILTER_SIZE(1));
	struct sockaddr_in6 *sin6;
	socklen_t len = GROUP_FILTER_SIZE(1);

	if (!gf) {
		perror("calloc");
		exit(EXIT_FAILURE);
	}
	gf->gf_interface = ifindex;
	gf->gf_numsrc = 1;
	set_group(&gf->gf_group, "ff0e::1234");
	if (getsockopt(fd, SOL_IPV6, MCAST_MSFILTER, gf, &len)) {
		perror("getsockopt(MCAST_MSFILTER)");
		exit(EXIT_FAILURE);
	}
	printf("%s: mode=%s sources=%u", ifname,
	       gf->gf_fmode == MCAST_INCLUDE ? "INCLUDE" : "EXCLUDE",
	       gf->gf_numsrc);
	if (gf->gf_numsrc) {
		char source[INET6_ADDRSTRLEN];

		sin6 = (struct sockaddr_in6 *)&gf->gf_slist[0];
		inet_ntop(AF_INET6, &sin6->sin6_addr, source, sizeof(source));
		printf(" first-source=%s", source);
	}
	putchar('\n');
	free(gf);
}

int main(void)
{
	const char *group = "ff0e::1234";
	const char *source = "2001:db8::1";
	unsigned int route_if = if_nametoindex("eth0");
	unsigned int shadow_if = if_nametoindex("lo");
	struct ipv6_mreq mreq = { 0 };
	struct group_source_req gsr = { 0 };
	int fd;

	if (!route_if || !shadow_if) {
		fprintf(stderr, "eth0 or lo is unavailable\n");
		return EXIT_FAILURE;
	}
	fd = socket(AF_INET6, SOCK_DGRAM, 0);
	if (fd < 0) {
		perror("socket");
		return EXIT_FAILURE;
	}
	if (inet_pton(AF_INET6, group, &mreq.ipv6mr_multiaddr) != 1) {
		perror("inet_pton(group)");
		return EXIT_FAILURE;
	}
	mreq.ipv6mr_interface = route_if;
	if (setsockopt(fd, SOL_IPV6, IPV6_ADD_MEMBERSHIP, &mreq, sizeof(mreq))) {
		perror("IPV6_ADD_MEMBERSHIP(eth0)");
		return EXIT_FAILURE;
	}
	mreq.ipv6mr_interface = shadow_if;
	if (setsockopt(fd, SOL_IPV6, IPV6_ADD_MEMBERSHIP, &mreq, sizeof(mreq))) {
		perror("IPV6_ADD_MEMBERSHIP(lo)");
		return EXIT_FAILURE;
	}

	gsr.gsr_interface = 0;
	set_group(&gsr.gsr_group, group);
	set_group(&gsr.gsr_source, source);
	show_filter(fd, route_if, "eth0 before");
	show_filter(fd, shadow_if, "lo before");
	if (setsockopt(fd, SOL_IPV6, MCAST_BLOCK_SOURCE, &gsr, sizeof(gsr))) {
		perror("setsockopt(MCAST_BLOCK_SOURCE)");
		return EXIT_FAILURE;
	}
	show_filter(fd, route_if, "eth0 after");
	show_filter(fd, shadow_if, "lo after");
	close(fd);
	return 0;
}

------END PoC--------

Best regards,
Zihan Xi


Zihan Xi (1):
  ipv6: mcast: use resolved interface for source filters

 net/ipv6/mcast.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

-- 
2.43.0


^ permalink raw reply	[flat|nested] 5+ messages in thread

* [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters
  2026-11-12 15:19 [PATCH net 0/1] ipv6: mcast: fix default-interface source filter selection Zihan Xi
@ 2026-11-12 15:19 ` Zihan Xi
  2026-11-19  8:39   ` Paolo Abeni
                     ` (2 more replies)
  0 siblings, 3 replies; 5+ messages in thread
From: Zihan Xi @ 2026-11-12 15:19 UTC (permalink / raw)
  To: netdev; +Cc: dsahern, idosch, davem, edumazet, pabeni, horms, vega, zihanx

When no interface is specified, ip6_mc_source() resolves the multicast
route to an interface but searches socket memberships with a wildcard
index. A socket joined to the same group on multiple interfaces can
therefore update its source filter on a different interface from the
interface filter.

Use the resolved interface index for the membership lookup and for
dropping the membership after leaving the last source.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Reported-by: Vega <vega@nebusec.ai>
Signed-off-by: Zihan Xi <zihanx@nebusec.ai>
---
 net/ipv6/mcast.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c
index ecef55f26189..63770b3a7bb2 100644
--- a/net/ipv6/mcast.c
+++ b/net/ipv6/mcast.c
@@ -362,6 +362,7 @@ int ip6_mc_source(int add, int omode, struct sock *sk,
 	struct net *net = sock_net(sk);
 	struct ipv6_mc_socklist *pmc;
 	struct inet6_dev *idev;
+	int ifindex;
 	int leavegroup = 0;
 	int i, j;
 	int err;
@@ -375,6 +376,7 @@ int ip6_mc_source(int add, int omode, struct sock *sk,
 	idev = ip6_mc_find_idev(net, group, pgsr->gsr_interface);
 	if (!idev)
 		return -ENODEV;
+	ifindex = idev->dev->ifindex;
 
 	mutex_lock(&idev->mc_lock);
 
@@ -386,7 +388,7 @@ int ip6_mc_source(int add, int omode, struct sock *sk,
 	err = -EADDRNOTAVAIL;
 
 	for_each_pmc_socklock(inet6, sk, pmc) {
-		if (pgsr->gsr_interface && pmc->ifindex != pgsr->gsr_interface)
+		if (pmc->ifindex != ifindex)
 			continue;
 		if (ipv6_addr_equal(&pmc->addr, group))
 			break;
@@ -497,7 +499,7 @@ int ip6_mc_source(int add, int omode, struct sock *sk,
 	mutex_unlock(&idev->mc_lock);
 	in6_dev_put(idev);
 	if (leavegroup)
-		err = ipv6_sock_mc_drop(sk, pgsr->gsr_interface, group);
+		err = ipv6_sock_mc_drop(sk, ifindex, group);
 	return err;
 }
 
-- 
2.43.0


^ permalink raw reply related	[flat|nested] 5+ messages in thread

* Re: [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters
  2026-11-12 15:19 ` [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters Zihan Xi
@ 2026-11-19  8:39   ` Paolo Abeni
  2026-11-19  8:43   ` Paolo Abeni
  2026-11-19 10:42   ` Simon Horman
  2 siblings, 0 replies; 5+ messages in thread
From: Paolo Abeni @ 2026-11-19  8:39 UTC (permalink / raw)
  To: zihanx; +Cc: netdev, dsahern, idosch, davem, edumazet, pabeni, horms, vega

This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.
Full review at: https://netdev-ai.bots.linux.dev/sashiko/
---
--- Patch [1]: [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters ---
commit 65be18fe1d53e3d386a91fc23d1c7ba9cc2e7b28
Author: Zihan Xi <zihanx@nebusec.ai>

ipv6: mcast: use resolved interface for source filters

This patch makes ip6_mc_source() match socket memberships against the
ifindex of the device resolved by ip6_mc_find_idev(). Before, it treated
gsr_interface == 0 as a wildcard. The resolved ifindex is also passed to
ipv6_sock_mc_drop() when the last source is removed.

> diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c
> index ecef55f261890..63770b3a7bb25 100644
> --- a/net/ipv6/mcast.c
> +++ b/net/ipv6/mcast.c

[ ... ]

> @@ -375,6 +376,7 @@ int ip6_mc_source(int add, int omode, struct sock *sk,
>  	idev = ip6_mc_find_idev(net, group, pgsr->gsr_interface);
>  	if (!idev)
>  		return -ENODEV;
> +	ifindex = idev->dev->ifindex;
>  
>  	mutex_lock(&idev->mc_lock);
>  
> @@ -386,7 +388,7 @@ int ip6_mc_source(int add, int omode, struct sock *sk,
>  	err = -EADDRNOTAVAIL;
>  
>  	for_each_pmc_socklock(inet6, sk, pmc) {
> -		if (pgsr->gsr_interface && pmc->ifindex != pgsr->gsr_interface)
> +		if (pmc->ifindex != ifindex)
>  			continue;
>  		if (ipv6_addr_equal(&pmc->addr, group))
>  			break;

[Severity: Medium]
[Finding: main-1-0]
[Sources: opus-5-5]
Does this break MCAST_JOIN_SOURCE_GROUP with gsr_interface == 0 when the
socket already has a membership for the group on another interface?

__ipv6_sock_mc_join() still treats ifindex 0 as "any interface". Its
duplicate check also runs before ip6_mc_find_dev() resolves the device:

net/ipv6/mcast.c:__ipv6_sock_mc_join() {
    ...
	for_each_pmc_socklock(np, sk, mc_lst) {
		if ((ifindex == 0 || mc_lst->ifindex == ifindex) &&
		    ipv6_addr_equal(&mc_lst->addr, addr))
			return -EADDRINUSE;
	}
    ...
	dev = ip6_mc_find_dev(net, addr, ifindex);
    ...
}

The caller accepts -EADDRINUSE and goes on to call ip6_mc_source():

net/ipv6/ipv6_sockglue.c:do_ipv6_mcast_group_source() {
    ...
		retv = ipv6_sock_mc_join_ssm(sk, greqs.gsr_interface,
					     &psin6->sin6_addr,
					     MCAST_INCLUDE);
		/* prior join w/ different source is ok */
		if (retv && retv != -EADDRINUSE)
			return retv;
    ...
}

Suppose the socket joined group G on interface Y, and the route for G
resolves to interface X:

do_ipv6_mcast_group_source()
  ipv6_sock_mc_join_ssm(sk, 0, G, MCAST_INCLUDE)
    __ipv6_sock_mc_join()
      returns -EADDRINUSE (matches Y, nothing created on X)
  ip6_mc_source()
    ip6_mc_find_idev() resolves to X
    no pmc with ifindex == X
    returns -EINVAL ("must have a prior join")

Before this patch the call returned 0, but it changed the socket filter
on Y while updating the interface filter on X. After this patch it fails.
In this state, a source-specific join with interface 0 can never join on
the resolved interface.

IPv4 does not have this gap. __ip_mc_join_group() calls ip_mc_find_dev()
first, which writes the resolved ifindex into imr. The duplicate check
then uses that ifindex.

MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE and MCAST_LEAVE_SOURCE_GROUP
with interface 0 also behave differently when the only membership is on
an interface the route does not resolve to, for example after a route
change.

One case that used to work now fails with -EINVAL: removing the last
INCLUDE source with MCAST_LEAVE_SOURCE_GROUP. Before this patch, that
path reached ipv6_sock_mc_drop(), which cleaned up using the pmc's own
ifindex.

Should __ipv6_sock_mc_join() resolve the device before its -EADDRINUSE
check, as IPv4 does? This is tagged for stable, so could the commit
message also say that these setsockopt() calls can now return -EINVAL?
-- 
This is an AI-generated review.


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters
  2026-11-12 15:19 ` [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters Zihan Xi
  2026-11-19  8:39   ` Paolo Abeni
@ 2026-11-19  8:43   ` Paolo Abeni
  2026-11-19 10:42   ` Simon Horman
  2 siblings, 0 replies; 5+ messages in thread
From: Paolo Abeni @ 2026-11-19  8:43 UTC (permalink / raw)
  To: Zihan Xi, netdev; +Cc: dsahern, idosch, davem, edumazet, horms, vega

On 11/12/26 16:19, Zihan Xi wrote:
> When no interface is specified, ip6_mc_source() resolves the multicast
> route to an interface but searches socket memberships with a wildcard
> index. A socket joined to the same group on multiple interfaces can
> therefore update its source filter on a different interface from the
> interface filter.
> 
> Use the resolved interface index for the membership lookup and for
> dropping the membership after leaving the last source.
> 
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Cc: stable@vger.kernel.org
> Reported-by: Vega <vega@nebusec.ai>
> Signed-off-by: Zihan Xi <zihanx@nebusec.ai>

This is changing an established behavior. While the old one is likely
"uncorrect", existing application may depend on it.
This is at best net-next material (with no fixes nor stable tag), but
lacking an end-user report, but I tent to think we are better off
without this change entirely.

/P


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters
  2026-11-12 15:19 ` [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters Zihan Xi
  2026-11-19  8:39   ` Paolo Abeni
  2026-11-19  8:43   ` Paolo Abeni
@ 2026-11-19 10:42   ` Simon Horman
  2 siblings, 0 replies; 5+ messages in thread
From: Simon Horman @ 2026-11-19 10:42 UTC (permalink / raw)
  To: Zihan Xi; +Cc: netdev, dsahern, idosch, davem, edumazet, pabeni, vega

On Thu, Nov 12, 2026 at 03:19:04PM +0000, Zihan Xi wrote:
> When no interface is specified, ip6_mc_source() resolves the multicast
> route to an interface but searches socket memberships with a wildcard
> index. A socket joined to the same group on multiple interfaces can
> therefore update its source filter on a different interface from the
> interface filter.
> 
> Use the resolved interface index for the membership lookup and for
> dropping the membership after leaving the last source.
> 
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")

Hi Zihan,

I see that Paolo has responded, and his comments should stand.
But I had been looking into this patch and I think it's worth
sharing two points, in case they turn out to be useful in future.

There is an AI-generated review of this patch available at
https://sashiko.dev/#/patchset/cover.1794738094.git.zihanx%40nebusec.ai

It states:

  Furthermore, the duplicate membership check with ifindex == 0 in
  ipv6_sock_mc_join(), the -EADDRINUSE handling in MCAST_JOIN_SOURCE_GROUP,
  and the leavegroup handling in ip6_mc_source() were introduced later by
  c9e3e8b6958e ("[IPV6]: multicast join and misc"), while
  ipv6_sock_mc_join_ssm() was introduced by c7ea20c9da5b ("ipv6/mcast: init
  as INCLUDE when join SSM INCLUDE group").  Should the Fixes: tag point to
  those commits instead?

I would lean towards using c9e3e8b6958e as the Fixes tag as it seems to me
that is the commit that that first introduced this bug.

> Cc: stable@vger.kernel.org
> Reported-by: Vega <vega@nebusec.ai>
> Signed-off-by: Zihan Xi <zihanx@nebusec.ai>
> ---
>  net/ipv6/mcast.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c
> index ecef55f26189..63770b3a7bb2 100644
> --- a/net/ipv6/mcast.c
> +++ b/net/ipv6/mcast.c
> @@ -362,6 +362,7 @@ int ip6_mc_source(int add, int omode, struct sock *sk,
>  	struct net *net = sock_net(sk);
>  	struct ipv6_mc_socklist *pmc;
>  	struct inet6_dev *idev;
> +	int ifindex;
>  	int leavegroup = 0;
>  	int i, j;
>  	int err;

For reference, please preserve reverse xmas tree - longest line to shortest -
ordering for local variable declarations in Networking code.

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2026-11-19 10:42 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-12 15:19 [PATCH net 0/1] ipv6: mcast: fix default-interface source filter selection Zihan Xi
2026-11-12 15:19 ` [PATCH net 1/1] ipv6: mcast: use resolved interface for source filters Zihan Xi
2026-11-19  8:39   ` Paolo Abeni
2026-11-19  8:43   ` Paolo Abeni
2026-11-19 10:42   ` Simon Horman

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox