Skip to contentSkip to conversation

macvlan: fix gso_max_size setting

PATCH Patrick McHardy 3 participants 7 messages Nov 23, 2009 – Nov 24, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Fix macvlan gso_max_size setting. Based on net-next-2.6.git.

commit 198a1fd488e7ebec080d1d2da7947cb9e1aacebf

Author: Patrick McHardy <[email protected]>
Date:   Mon Nov 23 11:28:22 2009 +0100

macvlan: fix gso_max_size setting

gso_max_size must be set based on the value of the underlying device to support devices not using the full 64k.

drivers/net/macvlan.c +2 -0
index ae2b5c7..7b0ef0c 100644
@@ -376,6 +376,7 @@ static int macvlan_init(struct net_device *dev)
 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
 				  (lowerdev->state & MACVLAN_STATE_MASK);
 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
+	dev->gso_max_size	= lowerdev->gso_max_size;
 	dev->iflink		= lowerdev->ifindex;
 	dev->hard_header_len	= lowerdev->hard_header_len;
 
@@ -652,6 +653,7 @@ static int macvlan_device_event(struct notifier_block *unused,
 	case NETDEV_FEAT_CHANGE:
 		list_for_each_entry(vlan, &port->vlans, list) {
 			vlan->dev->features = dev->features & MACVLAN_FEATURES;
+			vlan->dev->gso_max_size = dev->gso_max_size;
 			netdev_features_change(vlan->dev);
 		}
 		break;
maintainer
#1 lore raw

From: Patrick McHardy <[email protected]>
Date: Mon, 23 Nov 2009 11:33:06 +0100

Quoted 1 line

Fix macvlan gso_max_size setting. Based on net-next-2.6.git.

Pretty serious defect, maybe net-2.6 material?

If someone hits this, it's going to trigger something awful.

#2 lore raw
Quoted 7 lines · David Miller

From: Patrick McHardy <[email protected]>
Date: Mon, 23 Nov 2009 11:33:06 +0100

Quoted 1 line

Fix macvlan gso_max_size setting. Based on net-next-2.6.git.

Pretty serious defect, maybe net-2.6 material?

Sure, please take it in net-2.6.

Quoted 2 lines

If someone hits this, it's going to trigger something awful.

Don't drivers simply drop the packet? Otherwise I think you should be able to trigger the same problem using the TC mirred action.

maintainer
#3 lore raw

From: Patrick McHardy <[email protected]>
Date: Mon, 23 Nov 2009 23:11:07 +0100

Quoted 10 lines
Quoted 7 lines · David Miller

From: Patrick McHardy <[email protected]>
Date: Mon, 23 Nov 2009 11:33:06 +0100

Quoted 1 line

Fix macvlan gso_max_size setting. Based on net-next-2.6.git.

Pretty serious defect, maybe net-2.6 material?

Sure, please take it in net-2.6.

Done, thanks.

Quoted 6 lines
Quoted 2 lines

If someone hits this, it's going to trigger something awful.

Don't drivers simply drop the packet? Otherwise I think you should be able to trigger the same problem using the TC mirred action.

I see nothing in the IXGBE driver's transmit code paths which guard against this.

#4 lore raw
Quoted 23 lines · David Miller

From: Patrick McHardy <[email protected]>
Date: Mon, 23 Nov 2009 23:11:07 +0100

Quoted 7 lines
Quoted 5 lines · David Miller

From: Patrick McHardy <[email protected]>
Date: Mon, 23 Nov 2009 11:33:06 +0100

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

Fix macvlan gso_max_size setting. Based on net-next-2.6.git.

Pretty serious defect, maybe net-2.6 material?

Sure, please take it in net-2.6.

Done, thanks.

Quoted 2 lines
Quoted 2 lines

If someone hits this, it's going to trigger something awful.

Quoted 8 lines
Quoted 4 lines

Don't drivers simply drop the packet? Otherwise I think you should be able to trigger the same problem using the TC mirred action.

I see nothing in the IXGBE driver's transmit code paths which guard against this.

Perhaps we should simply fall back to software in that case. Compile tested only.

net: fall back to software segmentation if skb length exceeds dev->gso_max_size

include/linux/netdevice.h +2 -1
index ffc3106..edd1e2b 100644
@@ -1943,7 +1943,8 @@ static inline int netif_needs_gso(struct net_device *dev, struct sk_buff *skb)
 {
 	return skb_is_gso(skb) &&
 	       (!skb_gso_ok(skb, dev->features) ||
-		unlikely(skb->ip_summed != CHECKSUM_PARTIAL));
+		unlikely(skb->ip_summed != CHECKSUM_PARTIAL) ||
+		unlikely(skb->len > dev->gso_max_size));
 }
 
 static inline void netif_set_gso_max_size(struct net_device *dev,
#5 lore raw
Quoted 3 lines · Patrick McHardy

Perhaps we should simply fall back to software in that case. Compile tested only.

Sure.

Since ixgbe appears to be the only driver in our entire tree that has this limitation, I think it might make more sense to move this special-case check into it and get it to do software GSO just like tg3 does for a few other cases that the hardware can't handle.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#6 lore raw
Quoted 11 lines · Herbert Xu
Quoted 2 lines · Patrick McHardy

Perhaps we should simply fall back to software in that case. Compile tested only.

Sure.

Since ixgbe appears to be the only driver in our entire tree that has this limitation, I think it might make more sense to move this special-case check into it and get it to do software GSO just like tg3 does for a few other cases that the hardware can't handle.

Sure, that would also be fine. I'll leave that for the Intel guys however :)