From: Harry Coin <redacted>
Date: Tue, 11 Jul 2023 16:40:03 -0500

On 7/11/23 15:44, Andrew Lunn wrote:

quoted

quoted

quoted

quoted

quoted

quoted

The current llc_rcv.c around line 166 in net/llc/llc_input.c  has

           if (!net_eq(dev_net(dev), &init_net))
                   goto drop;

Thank you!  When you offer your patches, and you hear worries about being
'invasive', it's worth asking 'compared to what' -- since the 'status quo'
is every bridge with STP in a non default namespace with a loop in it
somewhere will freeze every connected system more solid than ice in
Antarctica.

https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html

say:

o It must be obviously correct and tested.
o It cannot be bigger than 100 lines, with context.
o It must fix only one thing.
o It must fix a real bug that bothers people (not a, "This could be a problem..." type thing).

git blame shows:

commit 721499e8931c5732202481ae24f2dfbf9910f129
Author: YOSHIFUJI Hideaki [off-list ref]
Date:   Sat Jul 19 22:34:43 2008 -0700

     netns: Use net_eq() to compare net-namespaces for optimization.

diff --git a/net/llc/llc_input.c b/net/llc/llc_input.c
index 1c45f172991e..57ad974e4d94 100644
--- a/net/llc/llc_input.c
+++ b/net/llc/llc_input.c

@@ -150,7 +150,7 @@ int llc_rcv(struct sk_buff *skb, struct net_device *dev,
         int (*rcv)(struct sk_buff *, struct net_device *,
                    struct packet_type *, struct net_device *);
  
-       if (dev_net(dev) != &init_net)
+       if (!net_eq(dev_net(dev), &init_net))
                 goto drop;
  
         /*

So this is just an optimization.

The test itself was added in

ommit e730c15519d09ea528b4d2f1103681fa5937c0e6
Author: Eric W. Biederman [off-list ref]
Date:   Mon Sep 17 11:53:39 2007 -0700

     [NET]: Make packet reception network namespace safe
     
     This patch modifies every packet receive function
     registered with dev_add_pack() to drop packets if they
     are not from the initial network namespace.
     
     This should ensure that the various network stacks do
     not receive packets in a anything but the initial network
     namespace until the code has been converted and is ready
     for them.
     
     Signed-off-by: Eric W. Biederman [off-list ref]
     Signed-off-by: David S. Miller [off-list ref]

So that was over 15 years ago.

It appears it has not bothered people for over 15 years.

Lets wait until we get to see the actual fix. We can then decide how
simple/hard it is to back port to stable, if it fulfils the stable
rules or not.

       Andrew

Andrew, fair enough.  In the time until it's fixed, the kernel folks 
should publish an advisory and block any attempt to set bridge stp state 
to other than 0 if in a non-default namespace. The alternative is a 
packet flood at whatever the top line speed is should there be a loop 
somewhere in even one connected link.

Like this ?  Someone who didn't notice the issue might complain about
it as regression.

---8<---

diff --git a/net/bridge/br_stp_if.c b/net/bridge/br_stp_if.c
index 75204d36d7f9..a807996ac56b 100644
--- a/net/bridge/br_stp_if.c
+++ b/net/bridge/br_stp_if.c

@@ -201,6 +201,11 @@ int br_stp_set_enabled(struct net_bridge *br, unsigned long val,
 {
 	ASSERT_RTNL();
 
+	if (!net_eq(dev_net(br->dev), &init_net)) {
+		NL_SET_ERR_MSG_MOD(extack, "STP can't be enabled in non-root netns");
+		return -EINVAL;
+	}
+
 	if (br_mrp_enabled(br)) {
 		NL_SET_ERR_MSG_MOD(extack,
 				   "STP can't be enabled if MRP is already enabled");
---8<---