thr3ads.net - Linux Ethernet Bridging - [Bridge] [PATCH v2 0/1] UDP traceroute packets with no checksum [May 2022]

If this information is useful, please help other people find it:
Share via:

Kevin Mitchell

2022-Apr-05 23:51 UTC

[Bridge] [PATCH v2 0/1] UDP traceroute packets with no checksum

This is v2 of https://lkml.org/lkml/2022/1/14/1060

That patch was discovered to cause problems with UDP tunnels as
described here:

https://lore.kernel.org/netdev/7eed8111-42d7-63e1-d289-346a596fc933 at
nvidia.com/

This version addresses the issue by instead explicitly handling zero UDP
checksum in the nf_reject_verify_csum() helper function.

Unlike the previous patch, this one only allows zero UDP checksum in
IPv4. I discovered that the non-netfilter IPv6 path would indeed drop
zero UDP checksum packets, so it's probably best to remain consistent.

-- 
2.35.1

Kevin Mitchell

2022-Apr-05 23:51 UTC

head link

[Bridge] [PATCH 1/1] netfilter: conntrack: skip verification of zero UDP checksum

The checksum is optional for UDP packets in IPv4. However nf_reject
would previously require a valid checksum to illicit a response such as
ICMP_DEST_UNREACH.

Add some logic to nf_reject_verify_csum to determine if a UDP packet has
a zero checksum and should therefore not be verified. Explicitly require
a valid checksum for IPv6 consistent RFC 2460 and with the non-netfilter
stack (see udp6_csum_zero_error).

Signed-off-by: Kevin Mitchell <kevmitch at arista.com>
---
 include/net/netfilter/nf_reject.h        | 27 ++++++++++++++++++++----
 net/bridge/netfilter/nft_reject_bridge.c |  6 ++++--
 net/ipv4/netfilter/nf_reject_ipv4.c      |  6 ++++--
 net/ipv6/netfilter/nf_reject_ipv6.c      |  2 +-
 4 files changed, 32 insertions(+), 9 deletions(-)

diff --git a/include/net/netfilter/nf_reject.h
b/include/net/netfilter/nf_reject.h
index 9051c3a0c8e7..f248c1ff8b22 100644
--- a/include/net/netfilter/nf_reject.h
+++ b/include/net/netfilter/nf_reject.h
@@ -5,12 +5,34 @@
 #include <linux/types.h>
 #include <uapi/linux/in.h>
 
-static inline bool nf_reject_verify_csum(__u8 proto)
+static inline bool nf_reject_verify_csum(struct sk_buff *skb, int dataoff,
+					  __u8 proto)
 {
 	/* Skip protocols that don't use 16-bit one's complement checksum
 	 * of the entire payload.
 	 */
 	switch (proto) {
+		/* Protocols with optional checksums. */
+		case IPPROTO_UDP: {
+			const struct udphdr *udp_hdr;
+			struct udphdr _udp_hdr;
+
+			/* Checksum is required in IPv6
+			 * see RFC 2460 section 8.1
+			 */
+			if (skb->protocol == htons(ETH_P_IPV6))
+				return true;
+
+			udp_hdr = skb_header_pointer(skb, dataoff,
+						     sizeof(_udp_hdr),
+						     &_udp_hdr);
+			if (!udp_hdr || udp_hdr->check)
+				return true;
+
+			return false;
+		}
+		case IPPROTO_GRE:
+
 		/* Protocols with other integrity checks. */
 		case IPPROTO_AH:
 		case IPPROTO_ESP:
@@ -19,9 +41,6 @@ static inline bool nf_reject_verify_csum(__u8 proto)
 		/* Protocols with partial checksums. */
 		case IPPROTO_UDPLITE:
 		case IPPROTO_DCCP:
-
-		/* Protocols with optional checksums. */
-		case IPPROTO_GRE:
 			return false;
 	}
 	return true;
diff --git a/net/bridge/netfilter/nft_reject_bridge.c
b/net/bridge/netfilter/nft_reject_bridge.c
index deae2c9a0f69..e6d62de95378 100644
--- a/net/bridge/netfilter/nft_reject_bridge.c
+++ b/net/bridge/netfilter/nft_reject_bridge.c
@@ -111,6 +111,7 @@ static void nft_reject_br_send_v4_unreach(struct net *net,
 	unsigned int len;
 	__wsum csum;
 	u8 proto;
+	int dataoff;
 
 	if (!nft_bridge_iphdr_validate(oldskb))
 		return;
@@ -129,9 +130,10 @@ static void nft_reject_br_send_v4_unreach(struct net *net,
 		return;
 
 	proto = ip_hdr(oldskb)->protocol;
+	dataoff = ip_hdrlen(oldskb);
 
 	if (!skb_csum_unnecessary(oldskb) &&
-	    nf_reject_verify_csum(proto) &&
+	    nf_reject_verify_csum(oldskb, dataoff, proto) &&
 	    nf_ip_checksum(oldskb, hook, ip_hdrlen(oldskb), proto))
 		return;
 
@@ -234,7 +236,7 @@ static bool reject6_br_csum_ok(struct sk_buff *skb, int
hook)
 	if (thoff < 0 || thoff >= skb->len || (fo & htons(~0x7)) != 0)
 		return false;
 
-	if (!nf_reject_verify_csum(proto))
+	if (!nf_reject_verify_csum(skb, thoff, proto))
 		return true;
 
 	return nf_ip6_checksum(skb, hook, thoff, proto) == 0;
diff --git a/net/ipv4/netfilter/nf_reject_ipv4.c
b/net/ipv4/netfilter/nf_reject_ipv4.c
index 93b07739807b..7a83fe1c2675 100644
--- a/net/ipv4/netfilter/nf_reject_ipv4.c
+++ b/net/ipv4/netfilter/nf_reject_ipv4.c
@@ -189,6 +189,7 @@ void nf_send_unreach(struct sk_buff *skb_in, int code, int
hook)
 {
 	struct iphdr *iph = ip_hdr(skb_in);
 	u8 proto = iph->protocol;
+	int dataoff = ip_hdrlen(skb_in);
 
 	if (iph->frag_off & htons(IP_OFFSET))
 		return;
@@ -196,12 +197,13 @@ void nf_send_unreach(struct sk_buff *skb_in, int code, int
hook)
 	if (hook == NF_INET_PRE_ROUTING && nf_reject_fill_skb_dst(skb_in))
 		return;
 
-	if (skb_csum_unnecessary(skb_in) || !nf_reject_verify_csum(proto)) {
+	if (skb_csum_unnecessary(skb_in) ||
+	    !nf_reject_verify_csum(skb_in, dataoff, proto)) {
 		icmp_send(skb_in, ICMP_DEST_UNREACH, code, 0);
 		return;
 	}
 
-	if (nf_ip_checksum(skb_in, hook, ip_hdrlen(skb_in), proto) == 0)
+	if (nf_ip_checksum(skb_in, hook, dataoff, proto) == 0)
 		icmp_send(skb_in, ICMP_DEST_UNREACH, code, 0);
 }
 EXPORT_SYMBOL_GPL(nf_send_unreach);
diff --git a/net/ipv6/netfilter/nf_reject_ipv6.c
b/net/ipv6/netfilter/nf_reject_ipv6.c
index 4aef6baaa55e..ac9574c6068f 100644
--- a/net/ipv6/netfilter/nf_reject_ipv6.c
+++ b/net/ipv6/netfilter/nf_reject_ipv6.c
@@ -253,7 +253,7 @@ static bool reject6_csum_ok(struct sk_buff *skb, int hook)
 	if (thoff < 0 || thoff >= skb->len || (fo & htons(~0x7)) != 0)
 		return false;
 
-	if (!nf_reject_verify_csum(proto))
+	if (!nf_reject_verify_csum(skb, thoff, proto))
 		return true;
 
 	return nf_ip6_checksum(skb, hook, thoff, proto) == 0;
-- 
2.35.1

Linus Lüssing

2022-May-04 10:06 UTC

head link

[Bridge] [PATCH v2 0/1] UDP traceroute packets with no checksum

On Tue, Apr 05, 2022 at 04:51:15PM -0700, Kevin Mitchell via Bridge
wrote:> This is v2 of https://lkml.org/lkml/2022/1/14/1060
> 
> That patch was discovered to cause problems with UDP tunnels as
> described here:
> 
> https://lore.kernel.org/netdev/7eed8111-42d7-63e1-d289-346a596fc933 at
nvidia.com/
> 
> This version addresses the issue by instead explicitly handling zero UDP
> checksum in the nf_reject_verify_csum() helper function.
> 
> Unlike the previous patch, this one only allows zero UDP checksum in
> IPv4. I discovered that the non-netfilter IPv6 path would indeed drop
> zero UDP checksum packets, so it's probably best to remain consistent.
Are you sure that a UDP zero checksum is not working for IPv6
packets? We are using it here without any issues with VXLAN
tunnels.

Yes, the original RFC did not allow UDP zero checksums in IPv6
packets, but I believe this has changed:

https://www.rfc-editor.org/rfc/rfc6936
(https://www.ietf.org/archive/id/draft-ietf-6man-udpzero-01.html)

Regards, Linus

Linux Ethernet Bridging - May 2022 - [Bridge] [PATCH v2 0/1] UDP traceroute packets with no checksum

[Bridge] [PATCH v2 0/1] UDP traceroute packets with no checksum

[Bridge] [PATCH 1/1] netfilter: conntrack: skip verification of zero UDP checksum

[Bridge] [PATCH v2 0/1] UDP traceroute packets with no checksum