Skip to contentSkip to conversation

IPVS: Add handling of incoming ICMPV6_PKT_TOOBIG messages

PATCH In review Julius Volz 4 participants 9 messages Jun 24, 2009 – Jul 27, 2009

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 IPVS: Add handling of incoming ICMPV6_PKT_TOOBIG messages

Add handling of incoming ICMPv6 Packet Too Big messages. This message is received when a realserver sends a packet >PMTU to the client. The hop on this path with insufficient MTU will generate an ICMPv6 Packet Too Big message back to the VIP. The LVS server receives this message, but the call to the function handling this has been missing. Thus, IPVS fails to forward the message to the real server, which then does not adjust the path MTU. This patch adds the missing call to ip_vs_in_icmp_v6() in ip_vs_in() to handle this situation.

Thanks to Rob Gallagher from HEAnet for reporting this issue and for testing this patch in production (with direct routing mode).

---
 net/netfilter/ipvs/ip_vs_core.c |   23 +++++++++++++++++------
 1 files changed, 17 insertions(+), 6 deletions(-)
net/netfilter/ipvs/ip_vs_core.c +17 -6
index 8dddb17..5750800 100644
@@ -1274,13 +1274,24 @@ ip_vs_in(unsigned int hooknum, struct sk_buff *skb,
 		return NF_ACCEPT;
 	}
 
-	if (unlikely(iph.protocol == IPPROTO_ICMP)) {
-		int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+#ifdef CONFIG_IP_VS_IPV6
+	if (af == AF_INET6) {
+		if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
+			int related, verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
 
-		if (related)
-			return verdict;
-		ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
-	}
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
+	} else
+#endif
+		if (unlikely(iph.protocol == IPPROTO_ICMP)) {
+			int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
 
 	/* Protocol supported? */
 	pp = ip_vs_proto_get(iph.protocol);
Signature · 1 line
1.6.0.4
maintainer
#1 lore raw
Quoted 9 lines · Julius Volz

Add handling of incoming ICMPv6 Packet Too Big messages. This message is received when a realserver sends a packet >PMTU to the client. The hop on this path with insufficient MTU will generate an ICMPv6 Packet Too Big message back to the VIP. The LVS server receives this message, but the call to the function handling this has been missing. Thus, IPVS fails to forward the message to the real server, which then does not adjust the path MTU. This patch adds the missing call to ip_vs_in_icmp_v6() in ip_vs_in() to handle this situation.

Quoted 5 lines

Thanks to Rob Gallagher from HEAnet for reporting this issue and for testing this patch in production (with direct routing mode).

Hi Julius, Hi Rob,

this seems reasonable to me, although it seems that the following code is common. I wonder if its repetition could be removed.

			if (related)
				return verdict;
			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);

On a not very related note, I'm currently on holidays and my net access is very sporadic. I'll be back at my desk on the 8th.

Quoted 41 lines
---
 net/netfilter/ipvs/ip_vs_core.c |   23 +++++++++++++++++------
 1 files changed, 17 insertions(+), 6 deletions(-)
net/netfilter/ipvs/ip_vs_core.c +17 -6
index 8dddb17..5750800 100644
@@ -1274,13 +1274,24 @@ ip_vs_in(unsigned int hooknum, struct sk_buff *skb,
 		return NF_ACCEPT;
 	}
 
-	if (unlikely(iph.protocol == IPPROTO_ICMP)) {
-		int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+#ifdef CONFIG_IP_VS_IPV6
+	if (af == AF_INET6) {
+		if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
+			int related, verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
 
-		if (related)
-			return verdict;
-		ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
-	}
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
+	} else
+#endif
+		if (unlikely(iph.protocol == IPPROTO_ICMP)) {
+			int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
 
 	/* Protocol supported? */
 	pp = ip_vs_proto_get(iph.protocol);
Signature · 1 line
1.6.0.4
#2 lore raw

Hi Simon,

Quoted 16 lines · Simon Horman
Quoted 9 lines · Julius Volz

Add handling of incoming ICMPv6 Packet Too Big messages. This message is received when a realserver sends a packet >PMTU to the client. The hop on this path with insufficient MTU will generate an ICMPv6 Packet Too Big message back to the VIP. The LVS server receives this message, but the call to the function handling this has been missing. Thus, IPVS fails to forward the message to the real server, which then does not adjust the path MTU. This patch adds the missing call to ip_vs_in_icmp_v6() in ip_vs_in() to handle this situation.

Quoted 5 lines

Thanks to Rob Gallagher from HEAnet for reporting this issue and for testing this patch in production (with direct routing mode).

Quoted 8 lines

Hi Julius, Hi Rob,

this seems reasonable to me, although it seems that the following code is common. I wonder if its repetition could be removed.

                       if (related)
                               return verdict; ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);

I agree, though I see no "nice" way to remove this duplication considering the ifs and #ifdefs around this. You could move the related and verdict variables to the top of the function and then recheck afterwards whether one of these ICMP-handling branches was entered and put the common code in there. But this seems more cumbersome to me than repeating the code. Maybe you see a nicer way? Btw., exactly this structure already exists in ip_vs_out(), which is why I adopted it like this for ip_vs_in().

Cheers,
Julius

maintainer Acked-by
#3 lore raw
Quoted 28 lines · Julius Volz

Hi Simon,

Quoted 16 lines · Simon Horman
Quoted 9 lines · Julius Volz

Add handling of incoming ICMPv6 Packet Too Big messages. This message is received when a realserver sends a packet >PMTU to the client. The hop on this path with insufficient MTU will generate an ICMPv6 Packet Too Big message back to the VIP. The LVS server receives this message, but the call to the function handling this has been missing. Thus, IPVS fails to forward the message to the real server, which then does not adjust the path MTU. This patch adds the missing call to ip_vs_in_icmp_v6() in ip_vs_in() to handle this situation.

Quoted 5 lines

Thanks to Rob Gallagher from HEAnet for reporting this issue and for testing this patch in production (with direct routing mode).

Quoted 8 lines

Hi Julius, Hi Rob,

this seems reasonable to me, although it seems that the following code is common. I wonder if its repetition could be removed.

                       if (related)
                               return verdict; ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);

Quoted 8 lines

I agree, though I see no "nice" way to remove this duplication considering the ifs and #ifdefs around this. You could move the related and verdict variables to the top of the function and then recheck afterwards whether one of these ICMP-handling branches was entered and put the common code in there. But this seems more cumbersome to me than repeating the code. Maybe you see a nicer way? Btw., exactly this structure already exists in ip_vs_out(), which is why I adopted it like this for ip_vs_in().

Hi Julius,

sorry for the delay in responding, I've been off-line / recovering from being off-line.

I couldn't see an obvious way either, though I was hoping that you might :-) As you can't I'm happy to go with what you originally posted.

maintainer Tested-by Rob Gallagher on v1 1/1
#4 lore raw
In reply to Julius Volz

From: Julius Volz <[email protected]>

IPVS: Add handling of incoming ICMPV6_PKT_TOOBIG messages

Add handling of incoming ICMPv6 Packet Too Big messages. This message is received when a realserver sends a packet >PMTU to the client. The hop on this path with insufficient MTU will generate an ICMPv6 Packet Too Big message back to the VIP. The LVS server receives this message, but the call to the function handling this has been missing. Thus, IPVS fails to forward the message to the real server, which then does not adjust the path MTU. This patch adds the missing call to ip_vs_in_icmp_v6() in ip_vs_in() to handle this situation.

Thanks to Rob Gallagher from HEAnet for reporting this issue and for testing this patch in production (with direct routing mode).

---
 net/netfilter/ipvs/ip_vs_core.c |   23 +++++++++++++++++------
 1 files changed, 17 insertions(+), 6 deletions(-)

Dave, please consider applying this change.

I'm ok with it not going into 2.6.31 as I don't think that many people are affected by this problem.

net/netfilter/ipvs/ip_vs_core.c +17 -6
index 8dddb17..5750800 100644
@@ -1274,13 +1274,24 @@ ip_vs_in(unsigned int hooknum, struct sk_buff *skb,
 		return NF_ACCEPT;
 	}
 
-	if (unlikely(iph.protocol == IPPROTO_ICMP)) {
-		int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+#ifdef CONFIG_IP_VS_IPV6
+	if (af == AF_INET6) {
+		if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
+			int related, verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
 
-		if (related)
-			return verdict;
-		ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
-	}
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
+	} else
+#endif
+		if (unlikely(iph.protocol == IPPROTO_ICMP)) {
+			int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
 
 	/* Protocol supported? */
 	pp = ip_vs_proto_get(iph.protocol);
#5 lore raw
Quoted 54 lines · Simon Horman

From: Julius Volz <[email protected]>

IPVS: Add handling of incoming ICMPV6_PKT_TOOBIG messages

Add handling of incoming ICMPv6 Packet Too Big messages. This message is received when a realserver sends a packet >PMTU to the client. The hop on this path with insufficient MTU will generate an ICMPv6 Packet Too Big message back to the VIP. The LVS server receives this message, but the call to the function handling this has been missing. Thus, IPVS fails to forward the message to the real server, which then does not adjust the path MTU. This patch adds the missing call to ip_vs_in_icmp_v6() in ip_vs_in() to handle this situation.

Thanks to Rob Gallagher from HEAnet for reporting this issue and for testing this patch in production (with direct routing mode).

---
 net/netfilter/ipvs/ip_vs_core.c |   23 +++++++++++++++++------
 1 files changed, 17 insertions(+), 6 deletions(-)

Dave, please consider applying this change.

I'm ok with it not going into 2.6.31 as I don't think that many people are affected by this problem.

net/netfilter/ipvs/ip_vs_core.c +10 -6
index 8dddb17..5750800 100644
@@ -1274,13 +1274,24 @@ ip_vs_in(unsigned int hooknum, struct sk_buff *skb,
 		return NF_ACCEPT;
 	}
 
-	if (unlikely(iph.protocol == IPPROTO_ICMP)) {
-		int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+#ifdef CONFIG_IP_VS_IPV6
+	if (af == AF_INET6) {
+		if (unlikely(iph.protocol == IPPROTO_ICMPV6)) {
+			int related, verdict = ip_vs_in_icmp_v6(skb, &related, hooknum);
 
-		if (related)
-			return verdict;
-		ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
-	}
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
+	} else
+#endif

net/netfilter/ipvs/ip_vs_core.c ip_vs_in(unsigned int hooknum, struct sk_buff *skb, · line 1297

Quoted 10 lines
net/netfilter/ipvs/ip_vs_core.c +7 -0
+		if (unlikely(iph.protocol == IPPROTO_ICMP)) {
+			int related, verdict = ip_vs_in_icmp(skb, &related, hooknum);
+
+			if (related)
+				return verdict;
+			ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
+		}
 
 	/* Protocol supported? */
 	pp = ip_vs_proto_get(iph.protocol);

I see no reference to ICMPV6_PKT_TOOBIG in this patch, so ChangeLog might be misleading or uncomplete, since other ICMPV6 message types (ICMPV6_DEST_UNREACH/ICMPV6_TIME_EXCEED) will also be forwarded/handled ?

maintainer
#6 lore raw

From: Eric Dumazet <[email protected]>
Date: Fri, 24 Jul 2009 06:25:25 +0200

Quoted 3 lines

I see no reference to ICMPV6_PKT_TOOBIG in this patch, so ChangeLog might be misleading or uncomplete, since other ICMPV6 message types (ICMPV6_DEST_UNREACH/ICMPV6_TIME_EXCEED) will also be forwarded/handled ?

Agreed, this commit message could use a tidy. What the patch actually is doing is adding the handling of ipv6 icmp messages at all.

Simon could you clarify the commit message a bit and resubmit?

Thanks!

maintainer
#7 lore raw
Quoted 8 lines · David Miller

From: Eric Dumazet <[email protected]>
Date: Fri, 24 Jul 2009 06:25:25 +0200

Quoted 4 lines

I see no reference to ICMPV6_PKT_TOOBIG in this patch, so ChangeLog might be misleading or uncomplete, since other ICMPV6 message types (ICMPV6_DEST_UNREACH/ICMPV6_TIME_EXCEED) will also be forwarded/handled ?

Quoted 4 lines

Agreed, this commit message could use a tidy. What the patch actually is doing is adding the handling of ipv6 icmp messages at all.

Simon could you clarify the commit message a bit and resubmit?

Will do.

#8 lore raw
Quoted 9 lines · Simon Horman
Quoted 8 lines · David Miller

From: Eric Dumazet <[email protected]>
Date: Fri, 24 Jul 2009 06:25:25 +0200

Quoted 4 lines

I see no reference to ICMPV6_PKT_TOOBIG in this patch, so ChangeLog might be misleading or uncomplete, since other ICMPV6 message types (ICMPV6_DEST_UNREACH/ICMPV6_TIME_EXCEED) will also be forwarded/handled ?

Quoted 6 lines
Quoted 4 lines

Agreed, this commit message could use a tidy.  What the patch actually is doing is adding the handling of ipv6 icmp messages at all.

Simon could you clarify the commit message a bit and resubmit?

Will do.

Oops, yes. For some weird reason, only the specific problem that caused the patch was in my head when writing that commit message. Feel free to edit it to include ICMPv6 handling for in-out packets in general.

Julius