[PATCH] ipv6: mcast: extend RCU protection in igmp6_send()

[PATCH] ipv6: mcast: extend RCU protection in igmp6_send()

[Chanho Min]

From: Eric Dumazet <edumazet@google.com>

[ Upstream commit 087c1faa594fa07a66933d750c0b2610aa1a2946 ]

igmp6_send() can be called without RTNL or RCU being held.

Extend RCU protection so that we can safely fetch the net pointer
and avoid a potential UAF.

Note that we no longer can use sock_alloc_send_skb() because
ipv6.igmp_sk uses GFP_KERNEL allocations which can sleep.

Instead use alloc_skb() and charge the net->ipv6.igmp_sk
socket under RCU protection.

Cc: stable@vger.kernel.org # 5.4
Fixes: b8ad0cbc58f7 ("[NETNS][IPV6] mcast - handle several network namespace")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://patch.msgid.link/20250207135841.1948589-9-edumazet@google.com
[ chanho: Backports to v5.4.y. v5.4.y does not include
commit b4a11b2033b7(net: fix IPSTATS_MIB_OUTPKGS increment in OutForwDatagrams),
so IPSTATS_MIB_OUTREQUESTS was changed to IPSTATS_MIB_OUTPKGS defined as
'OutRequests'. ]
Signed-off-by: Chanho Min <chanho.min@lge.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 net/ipv6/mcast.c | 32 +++++++++++++++-----------------
 1 file changed, 15 insertions(+), 17 deletions(-)

diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c
index 7d0a6a7c9d283..c2dc70a69be94 100644
--- a/net/ipv6/mcast.c
+++ b/net/ipv6/mcast.c
@@ -1977,21 +1977,21 @@ static void mld_send_cr(struct inet6_dev *idev)
 
 static void igmp6_send(struct in6_addr *addr, struct net_device *dev, int type)
 {
-	struct net *net = dev_net(dev);
-	struct sock *sk = net->ipv6.igmp_sk;
+	const struct in6_addr *snd_addr, *saddr;
+	int err, len, payload_len, full_len;
+	struct in6_addr addr_buf;
 	struct inet6_dev *idev;
 	struct sk_buff *skb;
 	struct mld_msg *hdr;
-	const struct in6_addr *snd_addr, *saddr;
-	struct in6_addr addr_buf;
 	int hlen = LL_RESERVED_SPACE(dev);
 	int tlen = dev->needed_tailroom;
-	int err, len, payload_len, full_len;
 	u8 ra[8] = { IPPROTO_ICMPV6, 0,
 		     IPV6_TLV_ROUTERALERT, 2, 0, 0,
 		     IPV6_TLV_PADN, 0 };
-	struct flowi6 fl6;
 	struct dst_entry *dst;
+	struct flowi6 fl6;
+	struct net *net;
+	struct sock *sk;
 
 	if (type == ICMPV6_MGM_REDUCTION)
 		snd_addr = &in6addr_linklocal_allrouters;
@@ -2002,20 +2002,21 @@ static void igmp6_send(struct in6_addr *addr, struct net_device *dev, int type)
 	payload_len = len + sizeof(ra);
 	full_len = sizeof(struct ipv6hdr) + payload_len;
 
-	rcu_read_lock();
-	IP6_UPD_PO_STATS(net, __in6_dev_get(dev),
-		      IPSTATS_MIB_OUT, full_len);
-	rcu_read_unlock();
+	skb = alloc_skb(hlen + tlen + full_len, GFP_KERNEL);
 
-	skb = sock_alloc_send_skb(sk, hlen + tlen + full_len, 1, &err);
+	rcu_read_lock();
 
+	net = dev_net_rcu(dev);
+	idev = __in6_dev_get(dev);
+	IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTPKTS);
 	if (!skb) {
-		rcu_read_lock();
-		IP6_INC_STATS(net, __in6_dev_get(dev),
-			      IPSTATS_MIB_OUTDISCARDS);
+		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
 		rcu_read_unlock();
 		return;
 	}
+	sk = net->ipv6.igmp_sk;
+	skb_set_owner_w(skb, sk);
+
 	skb->priority = TC_PRIO_CONTROL;
 	skb_reserve(skb, hlen);
 
@@ -2040,9 +2041,6 @@ static void igmp6_send(struct in6_addr *addr, struct net_device *dev, int type)
 					 IPPROTO_ICMPV6,
 					 csum_partial(hdr, len, 0));
 
-	rcu_read_lock();
-	idev = __in6_dev_get(skb->dev);
-
 	icmpv6_flow_init(sk, &fl6, type,
 			 &ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr,
 			 skb->dev->ifindex);

Re: [PATCH] ipv6: mcast: extend RCU protection in igmp6_send()

[Paolo Abeni]

On 8/18/25 11:24 AM, Chanho Min wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> [ Upstream commit 087c1faa594fa07a66933d750c0b2610aa1a2946 ]
> 
> igmp6_send() can be called without RTNL or RCU being held.
> 
> Extend RCU protection so that we can safely fetch the net pointer
> and avoid a potential UAF.
> 
> Note that we no longer can use sock_alloc_send_skb() because
> ipv6.igmp_sk uses GFP_KERNEL allocations which can sleep.
> 
> Instead use alloc_skb() and charge the net->ipv6.igmp_sk
> socket under RCU protection.
> 
> Cc: stable@vger.kernel.org # 5.4
> Fixes: b8ad0cbc58f7 ("[NETNS][IPV6] mcast - handle several network namespace")
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Reviewed-by: David Ahern <dsahern@kernel.org>
> Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
> Link: https://patch.msgid.link/20250207135841.1948589-9-edumazet@google.com
> [ chanho: Backports to v5.4.y. v5.4.y does not include
> commit b4a11b2033b7(net: fix IPSTATS_MIB_OUTPKGS increment in OutForwDatagrams),
> so IPSTATS_MIB_OUTREQUESTS was changed to IPSTATS_MIB_OUTPKGS defined as
> 'OutRequests'. ]
> Signed-off-by: Chanho Min <chanho.min@lge.com>
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
> Signed-off-by: Sasha Levin <sashal@kernel.org>

FWIW, the SoB chain above looks incorrect, as I think that neither Jakub
nor Sasha have touched yet this patch.

/P

Re: [PATCH] ipv6: mcast: extend RCU protection in igmp6_send()

[Sasha Levin]

On Thu, Aug 21, 2025 at 11:23:23AM +0200, Paolo Abeni wrote:
>On 8/18/25 11:24 AM, Chanho Min wrote:
>> From: Eric Dumazet <edumazet@google.com>
>>
>> [ Upstream commit 087c1faa594fa07a66933d750c0b2610aa1a2946 ]
>>
>> igmp6_send() can be called without RTNL or RCU being held.
>>
>> Extend RCU protection so that we can safely fetch the net pointer
>> and avoid a potential UAF.
>>
>> Note that we no longer can use sock_alloc_send_skb() because
>> ipv6.igmp_sk uses GFP_KERNEL allocations which can sleep.
>>
>> Instead use alloc_skb() and charge the net->ipv6.igmp_sk
>> socket under RCU protection.
>>
>> Cc: stable@vger.kernel.org # 5.4
>> Fixes: b8ad0cbc58f7 ("[NETNS][IPV6] mcast - handle several network namespace")
>> Signed-off-by: Eric Dumazet <edumazet@google.com>
>> Reviewed-by: David Ahern <dsahern@kernel.org>
>> Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
>> Link: https://patch.msgid.link/20250207135841.1948589-9-edumazet@google.com
>> [ chanho: Backports to v5.4.y. v5.4.y does not include
>> commit b4a11b2033b7(net: fix IPSTATS_MIB_OUTPKGS increment in OutForwDatagrams),
>> so IPSTATS_MIB_OUTREQUESTS was changed to IPSTATS_MIB_OUTPKGS defined as
>> 'OutRequests'. ]
>> Signed-off-by: Chanho Min <chanho.min@lge.com>
>> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
>> Signed-off-by: Sasha Levin <sashal@kernel.org>
>
>FWIW, the SoB chain above looks incorrect, as I think that neither Jakub
>nor Sasha have touched yet this patch.

It's a backport of an older backport where there already exists a commit with
both Jakub's any my signoff:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?id=81b25a07ebf53f9ef4ca8f3d96a8ddb94561dd5a

But yes, the SoB chain is wrong because Chanho Min's SoB should come after
ours, not before.

-- 
Thanks,
Sasha

Re: [PATCH] ipv6: mcast: extend RCU protection in igmp6_send()

[Greg Kroah-Hartman]

On Mon, Aug 18, 2025 at 06:24:53PM +0900, Chanho Min wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> [ Upstream commit 087c1faa594fa07a66933d750c0b2610aa1a2946 ]
> 
> igmp6_send() can be called without RTNL or RCU being held.
> 
> Extend RCU protection so that we can safely fetch the net pointer
> and avoid a potential UAF.
> 
> Note that we no longer can use sock_alloc_send_skb() because
> ipv6.igmp_sk uses GFP_KERNEL allocations which can sleep.
> 
> Instead use alloc_skb() and charge the net->ipv6.igmp_sk
> socket under RCU protection.
> 
> Cc: stable@vger.kernel.org # 5.4
> Fixes: b8ad0cbc58f7 ("[NETNS][IPV6] mcast - handle several network namespace")
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Reviewed-by: David Ahern <dsahern@kernel.org>
> Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
> Link: https://patch.msgid.link/20250207135841.1948589-9-edumazet@google.com
> [ chanho: Backports to v5.4.y. v5.4.y does not include
> commit b4a11b2033b7(net: fix IPSTATS_MIB_OUTPKGS increment in OutForwDatagrams),
> so IPSTATS_MIB_OUTREQUESTS was changed to IPSTATS_MIB_OUTPKGS defined as
> 'OutRequests'. ]

We can not take patches for only older kernel versions, as when you
upgrade you would have a regression.  Please send backports for all of
the missing stable versions and we will be glad to queue them up.

thanks,

greg k-h