James Dykman
2006-Apr-10 19:59 UTC
[Xen-devel] [PATCH] [BUG 143] [2nd try] Fix checksum errors over IPSec ESP tunnels
This patch moves the prep for checksum calculations from dev_queue_xmit()
into netback. The pointers in the skb headerare
set up for all IPv4 TCP/UDP traffic passing through netback, but the
checksums are not actually calculated until they absolutely
have to be (currently in dev_queue_xmit() and xfrm4_output()).
This removes some xen-specific code from the net/core/dev.c, at the
expense of some overhead for domU-domU traffic. Also
note that xfrm4_output() has to recalculate skb->h.raw, which gets
stomped. I have not bothered to track down where this happens;
I think it''s a sign that setting up these pointers so far from where
they
will be used is a bit fragile. But, it does work/fix the problem.
Tested on:
changeset: 9533:806d04252761
tag: tip
user: kaf24@firebug.cl.cam.ac.uk
date: Wed Apr 5 05:37:37 2006 +0100
summary: Fix checksum-offload problems introduced in c/s 9514, due to
With xm-test w/network-route/vif-route, and also with the failing config
from bug 143.
For reference, the first try:
http://lists.xensource.com/archives/html/xen-devel/2006-02/msg00162.html
Comments were:
http://lists.xensource.com/archives/html/xen-devel/2006-03/msg00521.html
Signed-off-by: Jim Dykman <dykman@us.ibm.com>
Jim
diff -r a8b1d4fad72d linux-2.6-xen-sparse/drivers/xen/netback/netback.c
--- a/linux-2.6-xen-sparse/drivers/xen/netback/netback.c Mon Mar 20
11:01:32 2006
+++ b/linux-2.6-xen-sparse/drivers/xen/netback/netback.c Fri Mar 24
16:45:17 2006
@@ -13,6 +13,9 @@
#include "common.h"
#include <xen/balloon.h>
#include <xen/interface/memory.h>
+#include <net/ip.h>
+#include <linux/tcp.h>
+#include <linux/udp.h>
/*#define NETBE_DEBUG_INTERRUPT*/
@@ -451,6 +454,33 @@
}
}
+inline static int checksum_setup(struct sk_buff *skb)
+{
+ if (skb->protocol != htons(ETH_P_IP))
+ return -1;
+ skb->nh.raw = skb->data;
+ skb->h.raw = (unsigned char *)skb->nh.iph +
4*skb->nh.iph->ihl;
+ if (skb->h.raw >= skb->tail)
+ return -1;
+ switch (skb->nh.iph->protocol) {
+ case IPPROTO_TCP:
+ skb->csum = offsetof(struct tcphdr, check);
+ break;
+ case IPPROTO_UDP:
+ skb->csum = offsetof(struct udphdr, check);
+ break;
+ default:
+ if (net_ratelimit())
+ printk(KERN_ERR "Attempting to checksum a
non-"
+ "TCP/UDP packet, dropping a protocol"
diff -r a8b1d4fad72d linux-2.6-xen-sparse/drivers/xen/netback/netback.c
--- a/linux-2.6-xen-sparse/drivers/xen/netback/netback.c Mon Mar 20
11:01:32 2006
+++ b/linux-2.6-xen-sparse/drivers/xen/netback/netback.c Fri Mar 24
16:45:17 2006
@@ -13,6 +13,9 @@
#include "common.h"
#include <xen/balloon.h>
#include <xen/interface/memory.h>
+#include <net/ip.h>
+#include <linux/tcp.h>
+#include <linux/udp.h>
/*#define NETBE_DEBUG_INTERRUPT*/
@@ -451,6 +454,33 @@
}
}
+inline static int checksum_setup(struct sk_buff *skb)
+{
+ if (skb->protocol != htons(ETH_P_IP))
+ return -1;
+ skb->nh.raw = skb->data;
+ skb->h.raw = (unsigned char *)skb->nh.iph +
4*skb->nh.iph->ihl;
+ if (skb->h.raw >= skb->tail)
+ return -1;
+ switch (skb->nh.iph->protocol) {
+ case IPPROTO_TCP:
+ skb->csum = offsetof(struct tcphdr, check);
+ break;
+ case IPPROTO_UDP:
+ skb->csum = offsetof(struct udphdr, check);
+ break;
+ default:
+ if (net_ratelimit())
+ printk(KERN_ERR "Attempting to checksum a
non-"
+ "TCP/UDP packet, dropping a protocol"
+ " %d packet",
skb->nh.iph->protocol);
+ return -1;
+ }
+ if ((skb->h.raw + skb->csum + 2) > skb->tail)
+ return -1;
+ return 0;
+}
+
/* Called after netfront has transmitted */
static void net_tx_action(unsigned long unused)
{
@@ -637,6 +667,14 @@
skb->ip_summed = CHECKSUM_UNNECESSARY;
skb->proto_csum_valid = 1;
skb->proto_csum_blank = !!(txreq.flags &
NETTXF_csum_blank);
+ if (skb->proto_csum_blank) {
+ ret = checksum_setup(skb);
+ if (ret < 0) {
+ kfree_skb(skb);
+ mop++;
+ continue;
+ }
+ }
netif->stats.rx_bytes += txreq.size;
netif->stats.rx_packets++;
diff -r a8b1d4fad72d linux-2.6-xen-sparse/net/core/dev.c
--- a/linux-2.6-xen-sparse/net/core/dev.c Mon Mar 20 11:01:32 2006
+++ b/linux-2.6-xen-sparse/net/core/dev.c Fri Mar 24 16:45:17 2006
@@ -116,12 +116,6 @@
#endif /* CONFIG_NET_RADIO */
#include <asm/current.h>
-#ifdef CONFIG_XEN
-#include <net/ip.h>
-#include <linux/tcp.h>
-#include <linux/udp.h>
-#endif
-
/*
* The list of packet types we will receive (as opposed to discard)
* and the routines to invoke.
@@ -1270,31 +1264,8 @@
/* If a checksum-deferred packet is forwarded to a device that
needs a
* checksum, correct the pointers and force checksumming.
*/
- if (skb->proto_csum_blank) {
- if (skb->protocol != htons(ETH_P_IP))
- goto out_kfree_skb;
- skb->h.raw = (unsigned char *)skb->nh.iph +
4*skb->nh.iph->ihl;
- if (skb->h.raw >= skb->tail)
- goto out_kfree_skb;
- switch (skb->nh.iph->protocol) {
- case IPPROTO_TCP:
- skb->csum = offsetof(struct tcphdr, check);
- break;
- case IPPROTO_UDP:
- skb->csum = offsetof(struct udphdr, check);
- break;
- default:
- if (net_ratelimit())
- printk(KERN_ERR "Attempting to checksum a
non-"
- "TCP/UDP packet, dropping a
protocol"
- " %d packet",
skb->nh.iph->protocol);
- rc = -EPROTO;
- goto out_kfree_skb;
- }
- if ((skb->h.raw + skb->csum + 2) > skb->tail)
- goto out_kfree_skb;
+ if (skb->proto_csum_blank)
skb->ip_summed = CHECKSUM_HW;
- }
#endif
/* If packet is not checksummed and device does not support
diff -r a8b1d4fad72d patches/linux-2.6.16-rc6/net-csum.patch
--- a/patches/linux-2.6.16/net-csum.patch Mon Mar 20 11:01:32 2006
+++ b/patches/linux-2.6.16/net-csum.patch Fri Mar 24 16:45:17 2006
@@ -39,3 +39,18 @@
*portptr = newport;
return 1;
}
+--- ../pristine-linux-2.6.16/net/ipv4/xfrm4_output.c 2006-03-24
16:17:38.000000000 -0500
++++ ./net/ipv4/xfrm4_output.c 2006-03-24 16:13:24.000000000 -0500
+@@ -103,6 +103,12 @@ static int xfrm4_output_one(struct sk_bu
+ struct xfrm_state *x = dst->xfrm;
+ int err;
+
++ if (skb->proto_csum_blank) {
++ skb->ip_summed = CHECKSUM_HW;
++ skb->proto_csum_blank = 0;
++ skb->h.raw = (unsigned char *)skb->nh.iph +
4*skb->nh.iph->ihl;
++ }
++
+ if (skb->ip_summed == CHECKSUM_HW) {
+ err = skb_checksum_help(skb, 0);
+ if (err)
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel