Skip to contentSkip to conversation

via-rhine: fix VLAN receive handling error in 4.2.x

PATCH In review Andrej 2 participants 4 messages Oct 3, 2015 – Oct 6, 2015

Subsystems (2)
lore
Share & tools
lore

Andrej

author
#0 lore raw

Hi,

via-rhine driver in 4.2.x kernels doesn’t correctly parse VLAN ID on receive. A bug was introduced in the commit 810f19bcb862f8889b27e0c9d9eceac9593925dd. All 4.2.x kernels are affected. 4.1.x and older kernels are not affected.

During code refactoring, the sequence of calls changed which introduced a regression. Original sequence was:
1) Read TCI from skb->data
2) Determine eth protocol using eth_type_trans (which calls skb_pull_inline)
3) Write TCI to skb->vlan_tci

After the change, the sequence is:
1) Determine protocol using eth_type_trans (which calls skb_pull_inline)
2) Read TCI from skb->data
3) Write TCI to skb->vlan_tci

Because eth_type_trans consumes ethernet header worth of bytes, a call to read TCI from packet no longer works as expected as it’s reading from invalid offset.

Choosing between changing rhine_get_vlan_tci(), which retrieves TCI from skb->data, or moving eth_type_trans() invocation after rhine_rx_vlan_tag(), I chose the latter.

Andrej.

drivers/net/ethernet/via/via-rhine.c 2015-10-03 18:53:51.799000000 +0200 renamed from drivers/net/ethernet/via/via-rhine.c 2015-10-03 15:46:59.817000000 +0200 +2 -1
@@ -2134,10 +2134,11 @@
 			}
 
 			skb_put(skb, pkt_len);
-			skb->protocol = eth_type_trans(skb, dev);
 
 			rhine_rx_vlan_tag(skb, desc, data_size);
 
+			skb->protocol = eth_type_trans(skb, dev);
+
 			netif_receive_skb(skb);
 
 			u64_stats_update_begin(&rp->rx_stats.syncp);
#1 lore raw

Andrej <[email protected]> :
[...]

Quoted 3 lines

Choosing between changing rhine_get_vlan_tci(), which retrieves TCI from skb->data, or moving eth_type_trans() invocation after rhine_rx_vlan_tag(), I chose the latter.

Can you send a patch with a proper Signed-off-by and a single line 'Fixes: 810f19bcb862 ("via-rhine: add consistent memory barrier in vlan receive code.")' or do you want me to do that ?

The patch ought to be based against davem's net but it wouldn't make any noticeable difference.

Signature · 1 line
Ueimor
author
#2 lore raw

PATCH 1/1 via-rhine: fix VLAN receive handling error in 4.2.x

Fixes commit 810f19bcb862 ("via-rhine: add consistent memory barrier in vlan receive code.") which broke VLAN tag parsing on receive.

Because eth_type_trans() consumes ethernet header worth of bytes, a call to read TCI from packet using rhine_rx_vlan_tag() no longer works as it's reading from an invalid offset.

Tested to be working on PCEngines Alix board.

Fixes 810f19bcb862 ("via-rhine: add consistent memory barrier in vlan receive code.")
---
 drivers/net/ethernet/via/via-rhine.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/via/via-rhine.c +2 -1
index a832637..2b7550c 100644
@@ -2134,10 +2134,11 @@ static int rhine_rx(struct net_device *dev, int limit)
 			}
 
 			skb_put(skb, pkt_len);
-			skb->protocol = eth_type_trans(skb, dev);
 
 			rhine_rx_vlan_tag(skb, desc, data_size);
 
+			skb->protocol = eth_type_trans(skb, dev);
+
 			netif_receive_skb(skb);
 
 			u64_stats_update_begin(&rp->rx_stats.syncp);
Signature · 1 line
2.4.3
#3 lore raw

Andrej Ota <[email protected]> :

Quoted 11 lines

Fixes commit 810f19bcb862 ("via-rhine: add consistent memory barrier in vlan receive code.") which broke VLAN tag parsing on receive.

Because eth_type_trans() consumes ethernet header worth of bytes, a call to read TCI from packet using rhine_rx_vlan_tag() no longer works as it's reading from an invalid offset.

Tested to be working on PCEngines Alix board.

Fixes 810f19bcb862 ("via-rhine: add consistent memory barrier in vlan receive code.")

Minor nits:
- Ethernet driver patches should usually be sent to [email protected]. Their state is thus tracked at
http://patchwork.ozlabs.org/project/netdev/list/

- The more specific about the target branch, the better:
[PATCH net 1/1] via-rhine: blah
-> patch against davem's bugfixes branch [PATCH net-next 19/73] foo: more balh
-> patch against davem's dev branch

- davem will take care of -stable propagation

Signature · 1 line
Ueimor