Skip to contentSkip to conversation

xen-netback: fix fragment detection in checksum setup

PATCH Superseded Paul Durrant 3 participants 5 messages Nov 28, 2013

Subsystems (2)

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xen-netback: fix fragment detection in checksum setup

The code to detect fragments in checksum_setup() was missing for IPv4 and too eager for IPv6. (It transpires that Windows seems to send IPv6 packets with a fragment header even if they are not a fragment - i.e. offset is zero, and M bit is not set).

Cc Wei Liu <[email protected]>
Cc Ian Campbell <[email protected]>
Cc David Vrabel <[email protected]>
---
 drivers/net/xen-netback/netback.c |   31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)
drivers/net/xen-netback/netback.c +28 -3
index 919b650..eea7ff2 100644
@@ -1165,15 +1165,27 @@ static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb,
 	struct iphdr *iph = (void *)skb->data;
 	unsigned int header_size;
 	unsigned int off;
+	bool fragment;
 	int err = -EPROTO;
 
+	fragment = false;
+
 	off = sizeof(struct iphdr);
 
 	header_size = skb->network_header + off + MAX_IPOPTLEN;
 	maybe_pull_tail(skb, header_size);
 
+	if (ntohs(iph->frag_off) & 0x3fff)
+		fragment = true;
+
 	off = iph->ihl * 4;
 
+	if (fragment) {
+		if (net_ratelimit())
+			netdev_err(vif->dev, "Packet is a fragment!\n");
+		goto out;
+	}
+
 	switch (iph->protocol) {
 	case IPPROTO_TCP:
 		if (!skb_partial_csum_set(skb, off,
@@ -1237,6 +1249,7 @@ static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb,
 	bool fragment;
 	bool done;
 
+	fragment = false;
 	done = false;
 
 	off = sizeof(struct ipv6hdr);
@@ -1275,9 +1288,21 @@ static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb,
 			off += (hp->hdrlen+2)<<2;
 			break;
 		}
-		case IPPROTO_FRAGMENT:
-			fragment = true;
-			/* fall through */
+		case IPPROTO_FRAGMENT: {
+			struct frag_hdr *hp = (void *)(skb->data + off);
+
+			header_size = skb->network_header +
+				off +
+				sizeof(struct frag_hdr);
+			maybe_pull_tail(skb, header_size);
+
+			if (ntohs(hp->frag_off) & 0xFFF9)
+				fragment = true;
+
+			nexthdr = hp->nexthdr;
+			off += sizeof(struct frag_hdr);
+			break;
+		}
 		default:
 			done = true;
 			break;
Signature · 1 line
1.7.10.4
#1 lore raw

You seem to have forgotten to Cc netdev@.

Quoted 22 lines · Paul Durrant

The code to detect fragments in checksum_setup() was missing for IPv4 and too eager for IPv6. (It transpires that Windows seems to send IPv6 packets with a fragment header even if they are not a fragment - i.e. offset is zero, and M bit is not set).

Cc Wei Liu <[email protected]>
Cc Ian Campbell <[email protected]>
Cc David Vrabel <[email protected]>
---
 drivers/net/xen-netback/netback.c |   31 ++++++++++++++++++++++++++++---
 1 file changed, 28 insertions(+), 3 deletions(-)
drivers/net/xen-netback/netback.c +1 -0
index 919b650..eea7ff2 100644
@@ -1165,15 +1165,27 @@ static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb,
 	struct iphdr *iph = (void *)skb->data;
 	unsigned int header_size;
 	unsigned int off;
+	bool fragment;

drivers/net/xen-netback/netback.c static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb, · line 1178

Quoted 10 lines
drivers/net/xen-netback/netback.c +3 -0
 	int err = -EPROTO;
 
+	fragment = false;
+
 	off = sizeof(struct iphdr);
 
 	header_size = skb->network_header + off + MAX_IPOPTLEN;
 	maybe_pull_tail(skb, header_size);
 
+	if (ntohs(iph->frag_off) & 0x3fff)

I'm actually quite suprised that these magic numbers don't have a #define in network core code.

0x3fff, if I'm not mistaken, is to check if a) M bit is set b) frag offset is not zero. I think this warrants a line of comment.

Quoted 29 lines
drivers/net/xen-netback/netback.c +10 -3
+		fragment = true;
+
 	off = iph->ihl * 4;
 
+	if (fragment) {
+		if (net_ratelimit())
+			netdev_err(vif->dev, "Packet is a fragment!\n");
+		goto out;
+	}
+
 	switch (iph->protocol) {
 	case IPPROTO_TCP:
 		if (!skb_partial_csum_set(skb, off,
@@ -1237,6 +1249,7 @@ static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb,
 	bool fragment;
 	bool done;
 
+	fragment = false;
 	done = false;
 
 	off = sizeof(struct ipv6hdr);
@@ -1275,9 +1288,21 @@ static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb,
 			off += (hp->hdrlen+2)<<2;
 			break;
 		}
-		case IPPROTO_FRAGMENT:
-			fragment = true;
-			/* fall through */
+		case IPPROTO_FRAGMENT: {

drivers/net/xen-netback/netback.c static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb, · line 1301

Quoted 10 lines
drivers/net/xen-netback/netback.c +10 -0
+			struct frag_hdr *hp = (void *)(skb->data + off);
+
+			header_size = skb->network_header +
+				off +
+				sizeof(struct frag_hdr);
+			maybe_pull_tail(skb, header_size);
+
+			if (ntohs(hp->frag_off) & 0xFFF9)
+				fragment = true;
+

Same here, would be helpful if you add comment to clarify what 0xFFF9 is.

Wei.

Quoted 9 lines
drivers/net/xen-netback/netback.c +4 -0
+			nexthdr = hp->nexthdr;
+			off += sizeof(struct frag_hdr);
+			break;
+		}
 		default:
 			done = true;
 			break;
Signature · 1 line
1.7.10.4
#2 lore raw
Quoted 10 lines

-----Original Message-----
From: Wei Liu [mailto:[email protected]]
Sent: 28 November 2013 13:13
To: Paul Durrant
Cc: [email protected]; Wei Liu; Ian Campbell; David Vrabel
Subject: Re: [PATCH net] xen-netback: fix fragment detection in checksum setup

You seem to have forgotten to Cc netdev@.

No, that was deliberate: I thought first review was supposed to be limited to xen-devel. Is that not the case any more?

Paul

Quoted 100 lines · Paul Durrant
Quoted 1 line

The code to detect fragments in checksum_setup() was missing for IPv4

and

Quoted 10 lines

too eager for IPv6. (It transpires that Windows seems to send IPv6 packets with a fragment header even if they are not a fragment - i.e. offset is zero, and M bit is not set).

Cc Wei Liu <[email protected]>
Cc Ian Campbell <[email protected]>
Cc David Vrabel <[email protected]>
---
 drivers/net/xen-netback/netback.c |   31

++++++++++++++++++++++++++++---

drivers/net/xen-

Quoted 3 lines
 1 file changed, 28 insertions(+), 3 deletions(-)
drivers/net/xen- renamed from drivers/net/xen-netback/netback.c +0 -0

netback/netback.c

drivers/net/xen-netback/netback.c static int checksum_setup_ip(struct xenvif *vif,

Quoted 4 lines

index 919b650..eea7ff2 100644

drivers/net/xen-netback/netback.c +0 -0
@@ -1165,15 +1165,27 @@ static int checksum_setup_ip(struct xenvif *vif,

struct sk_buff *skb,

Quoted 4 lines
drivers/net/xen-netback/netback.c +1 -0
 	struct iphdr *iph = (void *)skb->data;
 	unsigned int header_size;
 	unsigned int off;
+	bool fragment;

drivers/net/xen-netback/netback.c static int checksum_setup_ip(struct xenvif *vif, struct sk_buff *skb, · line 1178

Quoted 10 lines
drivers/net/xen-netback/netback.c +3 -0
 	int err = -EPROTO;
 
+	fragment = false;
+
 	off = sizeof(struct iphdr);
 
 	header_size = skb->network_header + off + MAX_IPOPTLEN;
 	maybe_pull_tail(skb, header_size);
 
+	if (ntohs(iph->frag_off) & 0x3fff)

I'm actually quite suprised that these magic numbers don't have a #define in network core code.

0x3fff, if I'm not mistaken, is to check if a) M bit is set b) frag offset is not zero. I think this warrants a line of comment.

drivers/net/xen-netback/netback.c static int checksum_setup_ipv6(struct xenvif *vif,

Quoted 14 lines
drivers/net/xen-netback/netback.c +8 -0
+		fragment = true;
+
 	off = iph->ihl * 4;
 
+	if (fragment) {
+		if (net_ratelimit())
+			netdev_err(vif->dev, "Packet is a fragment!\n");
+		goto out;
+	}
+
 	switch (iph->protocol) {
 	case IPPROTO_TCP:
 		if (!skb_partial_csum_set(skb, off,
@@ -1237,6 +1249,7 @@ static int checksum_setup_ipv6(struct xenvif *vif,

struct sk_buff *skb,

drivers/net/xen-netback/netback.c static int checksum_setup_ipv6(struct xenvif

Quoted 8 lines
drivers/net/xen-netback/netback.c +1 -0
 	bool fragment;
 	bool done;
 
+	fragment = false;
 	done = false;
 
 	off = sizeof(struct ipv6hdr);
@@ -1275,9 +1288,21 @@ static int checksum_setup_ipv6(struct xenvif

*vif, struct sk_buff *skb,

Quoted 7 lines
drivers/net/xen-netback/netback.c +1 -3
 			off += (hp->hdrlen+2)<<2;
 			break;
 		}
-		case IPPROTO_FRAGMENT:
-			fragment = true;
-			/* fall through */
+		case IPPROTO_FRAGMENT: {

drivers/net/xen-netback/netback.c static int checksum_setup_ipv6(struct xenvif *vif, struct sk_buff *skb, · line 1301

Quoted 10 lines
drivers/net/xen-netback/netback.c +10 -0
+			struct frag_hdr *hp = (void *)(skb->data + off);
+
+			header_size = skb->network_header +
+				off +
+				sizeof(struct frag_hdr);
+			maybe_pull_tail(skb, header_size);
+
+			if (ntohs(hp->frag_off) & 0xFFF9)
+				fragment = true;
+

Same here, would be helpful if you add comment to clarify what 0xFFF9 is.

Wei.

Quoted 9 lines
drivers/net/xen-netback/netback.c +4 -1
+			nexthdr = hp->nexthdr;
+			off += sizeof(struct frag_hdr);
+			break;
+		}
 		default:
 			done = true;
 			break;
--

1.7.10.4

#3 lore raw
Quoted 8 lines · Paul Durrant
Quoted 8 lines

-----Original Message-----
From: Wei Liu [mailto:[email protected]]
Sent: 28 November 2013 13:13
To: Paul Durrant
Cc: [email protected]; Wei Liu; Ian Campbell; David Vrabel
Subject: Re: [PATCH net] xen-netback: fix fragment detection in checksum setup

Quoted 5 lines
Quoted 2 lines

You seem to have forgotten to Cc netdev@.

No, that was deliberate: I thought first review was supposed to be limited to xen-devel. Is that not the case any more?

I think this works too. There's no guideline for this. The downside is that if you post something like 'V2' to netdev@ people there might wonder where V1 has gone. ;-)

Wei.

#4 lore raw
In reply to Paul Durrant
Quoted 8 lines · Paul Durrant
Quoted 8 lines

-----Original Message-----
From: Wei Liu [mailto:[email protected]]
Sent: 28 November 2013 13:13
To: Paul Durrant
Cc: [email protected]; Wei Liu; Ian Campbell; David Vrabel
Subject: Re: [PATCH net] xen-netback: fix fragment detection in checksum setup

Quoted 5 lines
Quoted 2 lines

You seem to have forgotten to Cc netdev@.

No, that was deliberate: I thought first review was supposed to be limited to xen-devel. Is that not the case any more?

For stuff which is big an complex or adds new protocol features I think discussing it on xen-devel and agreeing on the interface etc before taking it to netdev makes sense, simply because 99% of the folks there don't care one jot and don't want to read some enormous thread where we nitpick the name of the xenbus feature node ;-).

Obviously trivial stuff can go straight to netdev.

This one is somewhere in the middle and is a bit of a judgement call, but I think erring on the side of xen-devel only first is fine.

Ian.