Skip to contentSkip to conversation

ipv6: Queue fragments per interface for multicast/link-local addresses.

RFC PATCH In review YOSHIFUJI Hideaki 5 participants 9 messages Feb 16, 2013 – Mar 17, 2013

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

RFC PATCH 1/1 ipv6: Queue fragments per interface for multicast/link-local addresses.

We should queue fragments for the same link-local address on different interfaces (e.g. fe80::1%eth0 and fe80::1%eth1) to the different queue, because of nature of addressing architecture.

Similarly, we should queue fragments for multicast on different interface to the different queue. This is okay because application joins group on speicific interface, and multicast traffic is expected only on that interface.

CC Ben Greear <[email protected]>
CC Vlad Yasevich <[email protected]>
CC Eric Dumazet <[email protected]>
---
 include/net/ipv6.h                      |    1 +
 net/ipv6/netfilter/nf_conntrack_reasm.c |    9 +++++++--
 net/ipv6/reassembly.c                   |   10 ++++++++--
 3 files changed, 16 insertions(+), 4 deletions(-)
include/net/ipv6.h +1 -0
index 851d541..1af98aa 100644
@@ -463,6 +463,7 @@ enum ip6_defrag_users {
 struct ip6_create_arg {
 	__be32 id;
 	u32 user;
+	int ifindex;
 	const struct in6_addr *src;
 	const struct in6_addr *dst;
 };
net/ipv6/netfilter/nf_conntrack_reasm.c +7 -2
index c674f15..acd1820 100644
@@ -163,7 +163,8 @@ static void nf_ct_frag6_expire(unsigned long data)
 
 /* Creation primitives. */
 static inline struct frag_queue *fq_find(struct net *net, __be32 id,
-					 u32 user, struct in6_addr *src,
+					 u32 user, int ifindex,
+					 struct in6_addr *src,
 					 struct in6_addr *dst)
 {
 	struct inet_frag_queue *q;
@@ -171,6 +172,10 @@ static inline struct frag_queue *fq_find(struct net *net, __be32 id,
 	unsigned int hash;
 
 	arg.id = id;
+	if (ipv6_addr_type(dst) & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL))
+		arg.ifindex = ifindex;
+	else
+		arg.ifindex = 0;
 	arg.user = user;
 	arg.src = src;
 	arg.dst = dst;
@@ -572,7 +577,7 @@ struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
 	inet_frag_evictor(&net->nf_frag.frags, &nf_frags, false);
 	local_bh_enable();
 
-	fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr);
+	fq = fq_find(net, fhdr->identification, user, dev->ifindex, &hdr->saddr, &hdr->daddr);
 	if (fq == NULL) {
 		pr_debug("Can't find and can't create new queue\n");
 		goto ret_orig;
net/ipv6/reassembly.c +8 -2
index bab2c27..18c5e3c 100644
@@ -113,6 +113,7 @@ bool ip6_frag_match(struct inet_frag_queue *q, void *a)
 
 	fq = container_of(q, struct frag_queue, q);
 	return	fq->id == arg->id &&
+		(!arg->ifindex || fq->iif == arg->ifindex) &&
 		fq->user == arg->user &&
 		ipv6_addr_equal(&fq->saddr, arg->src) &&
 		ipv6_addr_equal(&fq->daddr, arg->dst);
@@ -182,13 +183,17 @@ static void ip6_frag_expire(unsigned long data)
 }
 
 static __inline__ struct frag_queue *
-fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
+fq_find(struct net *net, __be32 id, int ifindex, const struct in6_addr *src, const struct in6_addr *dst)
 {
 	struct inet_frag_queue *q;
 	struct ip6_create_arg arg;
 	unsigned int hash;
 
 	arg.id = id;
+	if (ipv6_addr_type(dst) & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL))
+		arg.ifindex = ifindex;
+	else
+		arg.ifindex = 0;
 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 	arg.src = src;
 	arg.dst = dst;
@@ -534,7 +539,8 @@ static int ipv6_frag_rcv(struct sk_buff *skb)
 		IP6_ADD_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 				 IPSTATS_MIB_REASMFAILS, evicted);
 
-	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
+	fq = fq_find(net, fhdr->identification, skb->dev->ifindex,
+		     &hdr->saddr, &hdr->daddr);
 	if (fq != NULL) {
 		int ret;
 
Signature · 1 line
1.7.9.5
#1 lore raw
Quoted 9 lines · YOSHIFUJI Hideaki

We should queue fragments for the same link-local address on different interfaces (e.g. fe80::1%eth0 and fe80::1%eth1) to the different queue, because of nature of addressing architecture.

Similarly, we should queue fragments for multicast on different interface to the different queue. This is okay because application joins group on speicific interface, and multicast traffic is expected only on that interface.

Quoted 5 lines
CC Ben Greear <[email protected]>
CC Vlad Yasevich <[email protected]>
CC Eric Dumazet <[email protected]>

---

I would definitely ask advice from Patrick & Pablo on this patch.

If a router uses several links in aggregation (but no bonding dev), we might break fragmentation/reassembly.

#2 lore raw
Quoted 16 lines · Eric Dumazet
Quoted 9 lines · YOSHIFUJI Hideaki

We should queue fragments for the same link-local address on different interfaces (e.g. fe80::1%eth0 and fe80::1%eth1) to the different queue, because of nature of addressing architecture.

Similarly, we should queue fragments for multicast on different interface to the different queue. This is okay because application joins group on speicific interface, and multicast traffic is expected only on that interface.

Quoted 5 lines
CC Ben Greear <[email protected]>
CC Vlad Yasevich <[email protected]>
CC Eric Dumazet <[email protected]>

---

Quoted 4 lines

I would definitely ask advice from Patrick & Pablo on this patch.

If a router uses several links in aggregation (but no bonding dev), we might break fragmentation/reassembly.

Could you elaborate, please?

The patch does not compare incoming interface if address is non-link-local unicast address.

--yoshfuji

#3 lore raw
Quoted 4 lines · YOSHIFUJI Hideaki

Could you elaborate, please?

The patch does not compare incoming interface if address is non-link-local unicast address.

There must be a reason ipv6 reasm is duplicated in net/ipv6/netfilter/nf_conntrack_reasm.c

netfilter uses the notion of ct zone, and several nics can belong to same zone.

Anyway your patch touches netfilter land, so must be CC to netfilter guys.

M:      Pablo Neira Ayuso <[email protected]>
M:      Patrick McHardy <[email protected]>
L:      [email protected]
#4 lore raw
Quoted 13 lines · Eric Dumazet
Quoted 4 lines · YOSHIFUJI Hideaki

Could you elaborate, please?

The patch does not compare incoming interface if address is non-link-local unicast address.

There must be a reason ipv6 reasm is duplicated in net/ipv6/netfilter/nf_conntrack_reasm.c

netfilter uses the notion of ct zone, and several nics can belong to same zone.

Quoted 7 lines

Anyway your patch touches netfilter land, so must be CC to netfilter guys.

M:      Pablo Neira Ayuso <[email protected]>
M:      Patrick McHardy <[email protected]>
L:      [email protected]

Looks like netfilter implementation will benefit from a similar patch as well. I like the idea of tagging the reassembly queue with the interface and I think it would have application in netfilter as well. Link-local traffic is limited to the interface already, so that shouldn't break netfilter assumptions. Multicast traffic is also bound
to an interface since group membership is per interface. If multiple interfaces are receiving the same fragmented multicast traffic, we want
multiple reassembly queues, or we'd end up discarding.

-vlad

#5 lore raw

RFC PATCH 1/1 ipv6: Queue fragments per interface for multicast/link-local addresses.

In reply to YOSHIFUJI Hideaki

We should queue fragments for the same link-local address on different interfaces (e.g. fe80::1%eth0 and fe80::1%eth1) to the different queue, because of nature of addressing architecture.

Similarly, we should queue fragments for multicast on different interface to the different queue. This is okay because application joins group on speicific interface, and multicast traffic is expected only on that interface.

CC Ben Greear <[email protected]>
CC Vlad Yasevich <[email protected]>
CC Eric Dumazet <[email protected]>
---
 include/net/ipv6.h                      |    1 +
 net/ipv6/netfilter/nf_conntrack_reasm.c |    9 +++++++--
 net/ipv6/reassembly.c                   |   10 ++++++++--
 3 files changed, 16 insertions(+), 4 deletions(-)
include/net/ipv6.h +1 -0
index 851d541..1af98aa 100644
@@ -463,6 +463,7 @@ enum ip6_defrag_users {
 struct ip6_create_arg {
 	__be32 id;
 	u32 user;
+	int ifindex;
 	const struct in6_addr *src;
 	const struct in6_addr *dst;
 };
net/ipv6/netfilter/nf_conntrack_reasm.c +7 -2
index c674f15..acd1820 100644
@@ -163,7 +163,8 @@ static void nf_ct_frag6_expire(unsigned long data)
 
 /* Creation primitives. */
 static inline struct frag_queue *fq_find(struct net *net, __be32 id,
-					 u32 user, struct in6_addr *src,
+					 u32 user, int ifindex,
+					 struct in6_addr *src,
 					 struct in6_addr *dst)
 {
 	struct inet_frag_queue *q;
@@ -171,6 +172,10 @@ static inline struct frag_queue *fq_find(struct net *net, __be32 id,
 	unsigned int hash;
 
 	arg.id = id;
+	if (ipv6_addr_type(dst) & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL))
+		arg.ifindex = ifindex;
+	else
+		arg.ifindex = 0;
 	arg.user = user;
 	arg.src = src;
 	arg.dst = dst;
@@ -572,7 +577,7 @@ struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb, u32 user)
 	inet_frag_evictor(&net->nf_frag.frags, &nf_frags, false);
 	local_bh_enable();
 
-	fq = fq_find(net, fhdr->identification, user, &hdr->saddr, &hdr->daddr);
+	fq = fq_find(net, fhdr->identification, user, dev->ifindex, &hdr->saddr, &hdr->daddr);
 	if (fq == NULL) {
 		pr_debug("Can't find and can't create new queue\n");
 		goto ret_orig;
net/ipv6/reassembly.c +8 -2
index bab2c27..18c5e3c 100644
@@ -113,6 +113,7 @@ bool ip6_frag_match(struct inet_frag_queue *q, void *a)
 
 	fq = container_of(q, struct frag_queue, q);
 	return	fq->id == arg->id &&
+		(!arg->ifindex || fq->iif == arg->ifindex) &&
 		fq->user == arg->user &&
 		ipv6_addr_equal(&fq->saddr, arg->src) &&
 		ipv6_addr_equal(&fq->daddr, arg->dst);
@@ -182,13 +183,17 @@ static void ip6_frag_expire(unsigned long data)
 }
 
 static __inline__ struct frag_queue *
-fq_find(struct net *net, __be32 id, const struct in6_addr *src, const struct in6_addr *dst)
+fq_find(struct net *net, __be32 id, int ifindex, const struct in6_addr *src, const struct in6_addr *dst)
 {
 	struct inet_frag_queue *q;
 	struct ip6_create_arg arg;
 	unsigned int hash;
 
 	arg.id = id;
+	if (ipv6_addr_type(dst) & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL))
+		arg.ifindex = ifindex;
+	else
+		arg.ifindex = 0;
 	arg.user = IP6_DEFRAG_LOCAL_DELIVER;
 	arg.src = src;
 	arg.dst = dst;
@@ -534,7 +539,8 @@ static int ipv6_frag_rcv(struct sk_buff *skb)
 		IP6_ADD_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
 				 IPSTATS_MIB_REASMFAILS, evicted);
 
-	fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr);
+	fq = fq_find(net, fhdr->identification, skb->dev->ifindex,
+		     &hdr->saddr, &hdr->daddr);
 	if (fq != NULL) {
 		int ret;
 
Signature · 6 lines
1.7.9.5

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#6 lore raw
Quoted 9 lines · YOSHIFUJI Hideaki

We should queue fragments for the same link-local address on different interfaces (e.g. fe80::1%eth0 and fe80::1%eth1) to the different queue, because of nature of addressing architecture.

Similarly, we should queue fragments for multicast on different interface to the different queue. This is okay because application joins group on speicific interface, and multicast traffic is expected only on that interface.

Your patch does solve the reassembly problem when macvlans are defined, thanks!

A tad unrelated, but i think there's still some ipv6 multicast filtering work to be done in the macvlan driver. If you for example join an all scope mcast address, ff09::1 on macvlan 0 it will implicitly be 'joined' on all other macvlans aswell..

I can buy that linklocal multicast packets are cloned out to all macvlan devices, but if a specific prefix is joined on one of them, i dont think it's correct that all sibling devices should receive that traffic, unless they asked for it.

//E

#7 lore raw
In reply to YOSHIFUJI Hideaki
Quoted 9 lines · YOSHIFUJI Hideaki

We should queue fragments for the same link-local address on different interfaces (e.g. fe80::1%eth0 and fe80::1%eth1) to the different queue, because of nature of addressing architecture.

Similarly, we should queue fragments for multicast on different interface to the different queue. This is okay because application joins group on speicific interface, and multicast traffic is expected only on that interface.

Quoted 4 lines
CC Ben Greear <[email protected]>
CC Vlad Yasevich <[email protected]>
CC Eric Dumazet <[email protected]>

I just found this patch while cleaning up my tree. I don't know its state (netdev patchworks says RFC and netfilter patchworks still lists it as new). However, I also do think that the per interface matching would be the right thing to do for multicast|linklocal fragments. Perhaps this patch should be resend?

Yoshifuji, do you think we should also implement RFC 3168 5.3 ECN fragmentation protection in reassembly.c? I think it should be straightforward because it is already implemented for ipv4 and the relevant bits just need to be moved to inet_fragment.c and become a bit more generalized.

Thanks,

Hannes

#8 lore raw
Quoted 15 lines · Hannes Frederic Sowa
Quoted 9 lines · YOSHIFUJI Hideaki

We should queue fragments for the same link-local address on different interfaces (e.g. fe80::1%eth0 and fe80::1%eth1) to the different queue, because of nature of addressing architecture.

Similarly, we should queue fragments for multicast on different interface to the different queue. This is okay because application joins group on speicific interface, and multicast traffic is expected only on that interface.

Quoted 4 lines
CC Ben Greear <[email protected]>
CC Vlad Yasevich <[email protected]>
CC Eric Dumazet <[email protected]>
Quoted 5 lines

I just found this patch while cleaning up my tree. I don't know its state (netdev patchworks says RFC and netfilter patchworks still lists it as new). However, I also do think that the per interface matching would be the right thing to do for multicast|linklocal fragments. Perhaps this patch should be resend?

Will do.

Quoted 5 lines

Yoshifuji, do you think we should also implement RFC 3168 5.3 ECN fragmentation protection in reassembly.c? I think it should be straightforward because it is already implemented for ipv4 and the relevant bits just need to be moved to inet_fragment.c and become a bit more generalized.

OK.

--yoshfuji