[PATCH net 0/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[PATCH net 0/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[Tobias Waldekranz]

This bug has been latent in the source for quite some time, I suspect
due to the homogeneity of both typical configurations and hardware.

On singlechip systems, this would never be triggered. The only reason
I saw it on my multichip system was because not all chips had the same
number of ports, which means that the misdemeanor alien call turned
into a felony array-out-of-bounds access.

Tobias Waldekranz (2):
  net: dsa: Move VLAN filtering syncing out of dsa_switch_bridge_leave
  net: dsa: Avoid cross-chip syncing of VLAN filtering

 net/dsa/switch.c | 40 +++++++++++++++++++++++++++-------------
 1 file changed, 27 insertions(+), 13 deletions(-)

-- 
2.25.1

[PATCH net 1/2] net: dsa: Move VLAN filtering syncing out of dsa_switch_bridge_leave

[Tobias Waldekranz]

Most of dsa_switch_bridge_leave was, in fact, dealing with the syncing
of VLAN filtering for switches on which that is a global
setting. Separate the two phases to prepare for the cross-chip related
bugfix in the following commit.

Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
---
 net/dsa/switch.c | 38 +++++++++++++++++++++++++-------------
 1 file changed, 25 insertions(+), 13 deletions(-)

diff --git a/net/dsa/switch.c b/net/dsa/switch.c
index e3c7d2627a61..9f9b70d6070a 100644
--- a/net/dsa/switch.c
+++ b/net/dsa/switch.c
@@ -113,26 +113,15 @@ static int dsa_switch_bridge_join(struct dsa_switch *ds,
 	return dsa_tag_8021q_bridge_join(ds, info);
 }
 
-static int dsa_switch_bridge_leave(struct dsa_switch *ds,
-				   struct dsa_notifier_bridge_info *info)
+static int dsa_switch_sync_vlan_filtering(struct dsa_switch *ds,
+					  struct dsa_notifier_bridge_info *info)
 {
-	struct dsa_switch_tree *dst = ds->dst;
 	struct netlink_ext_ack extack = {0};
 	bool change_vlan_filtering = false;
 	bool vlan_filtering;
 	struct dsa_port *dp;
 	int err;
 
-	if (dst->index == info->tree_index && ds->index == info->sw_index &&
-	    ds->ops->port_bridge_leave)
-		ds->ops->port_bridge_leave(ds, info->port, info->bridge);
-
-	if ((dst->index != info->tree_index || ds->index != info->sw_index) &&
-	    ds->ops->crosschip_bridge_leave)
-		ds->ops->crosschip_bridge_leave(ds, info->tree_index,
-						info->sw_index, info->port,
-						info->bridge);
-
 	if (ds->needs_standalone_vlan_filtering &&
 	    !br_vlan_enabled(info->bridge.dev)) {
 		change_vlan_filtering = true;
@@ -172,6 +161,29 @@ static int dsa_switch_bridge_leave(struct dsa_switch *ds,
 			return err;
 	}
 
+	return 0;
+}
+
+static int dsa_switch_bridge_leave(struct dsa_switch *ds,
+				   struct dsa_notifier_bridge_info *info)
+{
+	struct dsa_switch_tree *dst = ds->dst;
+	int err;
+
+	if (dst->index == info->tree_index && ds->index == info->sw_index &&
+	    ds->ops->port_bridge_leave)
+		ds->ops->port_bridge_leave(ds, info->port, info->bridge);
+
+	if ((dst->index != info->tree_index || ds->index != info->sw_index) &&
+	    ds->ops->crosschip_bridge_leave)
+		ds->ops->crosschip_bridge_leave(ds, info->tree_index,
+						info->sw_index, info->port,
+						info->bridge);
+
+	err = dsa_switch_sync_vlan_filtering(ds, info);
+	if (err)
+		return err;
+
 	return dsa_tag_8021q_bridge_leave(ds, info);
 }
 
-- 
2.25.1

Re: [PATCH net 1/2] net: dsa: Move VLAN filtering syncing out of dsa_switch_bridge_leave

[Vladimir Oltean]

On Mon, Jan 24, 2022 at 10:09:43PM +0100, Tobias Waldekranz wrote:
> Most of dsa_switch_bridge_leave was, in fact, dealing with the syncing
> of VLAN filtering for switches on which that is a global
> setting. Separate the two phases to prepare for the cross-chip related
> bugfix in the following commit.
> 
> Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
> ---

Reviewed-by: Vladimir Oltean <olteanv@gmail.com>

[PATCH net 2/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[Tobias Waldekranz]

Changes to VLAN filtering are not applicable to cross-chip
notifications.

On a system like this:

.-----.   .-----.   .-----.
| sw1 +---+ sw2 +---+ sw3 |
'-1-2-'   '-1-2-'   '-1-2-'

Before this change, upon sw1p1 leaving a bridge, a call to
dsa_port_vlan_filtering would also be made to sw2p1 and sw3p1.

In this scenario:

.---------.   .-----.   .-----.
|   sw1   +---+ sw2 +---+ sw3 |
'-1-2-3-4-'   '-1-2-'   '-1-2-'

When sw1p4 would leave a bridge, dsa_port_vlan_filtering would be
called for sw2 and sw3 with a non-existing port - leading to array
out-of-bounds accesses and crashes on mv88e6xxx.

Fixes: d371b7c92d19 ("net: dsa: Unset vlan_filtering when ports leave the bridge")
Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
---
 net/dsa/switch.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/net/dsa/switch.c b/net/dsa/switch.c
index 9f9b70d6070a..517cc83d13cc 100644
--- a/net/dsa/switch.c
+++ b/net/dsa/switch.c
@@ -180,9 +180,11 @@ static int dsa_switch_bridge_leave(struct dsa_switch *ds,
 						info->sw_index, info->port,
 						info->bridge);
 
-	err = dsa_switch_sync_vlan_filtering(ds, info);
-	if (err)
-		return err;
+	if (ds->dst->index == info->tree_index && ds->index == info->sw_index) {
+		err = dsa_switch_sync_vlan_filtering(ds, info);
+		if (err)
+			return err;
+	}
 
 	return dsa_tag_8021q_bridge_leave(ds, info);
 }
-- 
2.25.1

Re: [PATCH net 2/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[Vladimir Oltean]

On Mon, Jan 24, 2022 at 10:09:44PM +0100, Tobias Waldekranz wrote:
> Changes to VLAN filtering are not applicable to cross-chip
> notifications.

Yes, it seems so. In a cross-chip setup, ports will individually leave
the bridge, leaving every switch a chance to unset VLAN filtering.
We have this check in dsa_port_vlan_filtering(), so it's easy to forget
that the function is called more times than actually needed:

	if (dsa_port_is_vlan_filtering(dp) == vlan_filtering)
		return 0;

Sorry.

> On a system like this:
> 
> .-----.   .-----.   .-----.
> | sw1 +---+ sw2 +---+ sw3 |
> '-1-2-'   '-1-2-'   '-1-2-'
> 
> Before this change, upon sw1p1 leaving a bridge, a call to
> dsa_port_vlan_filtering would also be made to sw2p1 and sw3p1.
> 
> In this scenario:
> 
> .---------.   .-----.   .-----.
> |   sw1   +---+ sw2 +---+ sw3 |
> '-1-2-3-4-'   '-1-2-'   '-1-2-'
> 
> When sw1p4 would leave a bridge, dsa_port_vlan_filtering would be
> called for sw2 and sw3 with a non-existing port - leading to array
> out-of-bounds accesses and crashes on mv88e6xxx.
> 
> Fixes: d371b7c92d19 ("net: dsa: Unset vlan_filtering when ports leave the bridge")
> Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
> ---
>  net/dsa/switch.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/net/dsa/switch.c b/net/dsa/switch.c
> index 9f9b70d6070a..517cc83d13cc 100644
> --- a/net/dsa/switch.c
> +++ b/net/dsa/switch.c
> @@ -180,9 +180,11 @@ static int dsa_switch_bridge_leave(struct dsa_switch *ds,
>  						info->sw_index, info->port,
>  						info->bridge);
>  
> -	err = dsa_switch_sync_vlan_filtering(ds, info);
> -	if (err)
> -		return err;
> +	if (ds->dst->index == info->tree_index && ds->index == info->sw_index) {
> +		err = dsa_switch_sync_vlan_filtering(ds, info);
> +		if (err)
> +			return err;
> +	}

As net-next material, we could probably move this call to
dsa_port_switchdev_unsync_attrs() where there's even a comment that
references it, and do away with the targeted switch check.

>  
>  	return dsa_tag_8021q_bridge_leave(ds, info);
>  }
> -- 
> 2.25.1
> 

Reviewed-by: Vladimir Oltean <olteanv@gmail.com>

Re: [PATCH net 0/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[Jakub Kicinski]

On Mon, 24 Jan 2022 22:09:42 +0100 Tobias Waldekranz wrote:
> This bug has been latent in the source for quite some time, I suspect
> due to the homogeneity of both typical configurations and hardware.
> 
> On singlechip systems, this would never be triggered. The only reason
> I saw it on my multichip system was because not all chips had the same
> number of ports, which means that the misdemeanor alien call turned
> into a felony array-out-of-bounds access.

Applied, thanks, 934d0f039959 ("Merge branch
'dsa-avoid-cross-chip-vlan-sync'") in net-next.

Re: [PATCH net 0/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[Tobias Waldekranz]

On Tue, Jan 25, 2022 at 10:01, Jakub Kicinski <kuba@kernel.org> wrote:
> On Mon, 24 Jan 2022 22:09:42 +0100 Tobias Waldekranz wrote:
>> This bug has been latent in the source for quite some time, I suspect
>> due to the homogeneity of both typical configurations and hardware.
>> 
>> On singlechip systems, this would never be triggered. The only reason
>> I saw it on my multichip system was because not all chips had the same
>> number of ports, which means that the misdemeanor alien call turned
>> into a felony array-out-of-bounds access.
>
> Applied, thanks, 934d0f039959 ("Merge branch
> 'dsa-avoid-cross-chip-vlan-sync'") in net-next.

Thank you! Is there a particular reason that this was applied to
net-next? I guess my question is really: will it still be considered for
upcoming stable kernel releases?

Re: [PATCH net 0/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[Jakub Kicinski]

On Tue, 25 Jan 2022 20:05:45 +0100 Tobias Waldekranz wrote:
> On Tue, Jan 25, 2022 at 10:01, Jakub Kicinski <kuba@kernel.org> wrote:
> > On Mon, 24 Jan 2022 22:09:42 +0100 Tobias Waldekranz wrote:  
> >> This bug has been latent in the source for quite some time, I suspect
> >> due to the homogeneity of both typical configurations and hardware.
> >> 
> >> On singlechip systems, this would never be triggered. The only reason
> >> I saw it on my multichip system was because not all chips had the same
> >> number of ports, which means that the misdemeanor alien call turned
> >> into a felony array-out-of-bounds access.  
> >
> > Applied, thanks, 934d0f039959 ("Merge branch
> > 'dsa-avoid-cross-chip-vlan-sync'") in net-next.  
> 
> Is there a particular reason that this was applied to net-next?

Not sure, there were issues with kernel.org infra during the night,
could be unintentional.

> I guess my question is really: will it still be considered for
> upcoming stable kernel releases?

Only after the next merge window, but yes.

Re: [PATCH net 0/2] net: dsa: Avoid cross-chip syncing of VLAN filtering

[Tobias Waldekranz]

On Tue, Jan 25, 2022 at 12:41, Jakub Kicinski <kuba@kernel.org> wrote:
> On Tue, 25 Jan 2022 20:05:45 +0100 Tobias Waldekranz wrote:
>> On Tue, Jan 25, 2022 at 10:01, Jakub Kicinski <kuba@kernel.org> wrote:
>> > On Mon, 24 Jan 2022 22:09:42 +0100 Tobias Waldekranz wrote:  
>> >> This bug has been latent in the source for quite some time, I suspect
>> >> due to the homogeneity of both typical configurations and hardware.
>> >> 
>> >> On singlechip systems, this would never be triggered. The only reason
>> >> I saw it on my multichip system was because not all chips had the same
>> >> number of ports, which means that the misdemeanor alien call turned
>> >> into a felony array-out-of-bounds access.  
>> >
>> > Applied, thanks, 934d0f039959 ("Merge branch
>> > 'dsa-avoid-cross-chip-vlan-sync'") in net-next.  
>> 
>> Is there a particular reason that this was applied to net-next?
>
> Not sure, there were issues with kernel.org infra during the night,
> could be unintentional.

Ahh ok, hope it gets sorted quickly!

>> I guess my question is really: will it still be considered for
>> upcoming stable kernel releases?
>
> Only after the next merge window, but yes.

I had a feeling that would be the case. Aright, not optimal, but not a
big deal either. Thanks for taking the time to respond.