Skip to contentSkip to conversation

netfilter: arptables: add mcmangle target

PATCH In review Pablo Neira Ayuso 4 participants 23 messages Jan 28, 2009 – Feb 11, 2009

Subsystems (2)
lore
Share & tools
lore

Pablo Neira Ayuso

author maintainer
#0 lore raw

PATCH 1/3 netfilter: arptables: add mcmangle target

This patch adds the mcmangle target for arptables which allows altering the source hardware address in ARP with a multicast hardware address. This target can be used to make a switch flood packets to the ports that use the same MAC multicast address. This is useful to deploy load-sharing clusters in environments in which the switch does not provide a way to flood packets to several ports.

Since all the nodes receives the same packets, each decides if it handles the packet based on hashing approach (See the `cluster' iptables match that comes with this patchset).

Theoretically, the use of the reserved VRRP hardware address should be fine for this, however, switches generally treat this hardware address space as normal unicast hardware address. Thus, in practise, it is not possible to have two nodes with the same VRRP hardware address.

Please, note that this target violates RFC 1812 (section 3.3.2) since an ethernet device must not use a multicast link address.

An example of the use of this target:

arptables -I OUTPUT -o eth0 -j mcmangle --h-length 6 \
	--mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth0
arptables -I INPUT -i eth0 --h-length 6 --destination-mac \
	01:00:5e:00:01:01 -j mangle --mangle-mac-d 00:zz:yy:xx:5a:27

Where 00:zz:yy:xx:5a:27 is the original hardware address of this node. Note that the mcmangle target registers an entry in the multicast list that is required to get this working:

$ cat /proc/net/dev_mcast | grep eth0 | head -1
2    eth0            1     0     01005e000101

You need the PKTTYPE iptables target (included in this patchset) to set skb->type to PACKET_HOST. Otherwise, you would be only able to ICMP ping nodes in the network ;).

---

 include/linux/netfilter_arp/arpt_mcmangle.h |   14 ++++
 net/ipv4/netfilter/Kconfig                  |   12 +++
 net/ipv4/netfilter/Makefile                 |    1 
 net/ipv4/netfilter/arpt_mcmangle.c          |   98 +++++++++++++++++++++++++++
 4 files changed, 125 insertions(+), 0 deletions(-)
 create mode 100644 include/linux/netfilter_arp/arpt_mcmangle.h
 create mode 100644 net/ipv4/netfilter/arpt_mcmangle.c
include/linux/netfilter_arp/arpt_mcmangle.h new file +14 -0
new file mode 100644
index 0000000..d14a1ab
@@ -0,0 +1,14 @@
+#ifndef _ARPT_MCMANGLE_H
+#define _ARPT_MCMANGLE_H
+#include <linux/netfilter_arp/arp_tables.h>
+
+struct net_device;
+
+struct arpt_mcmangle
+{
+	char ifname[IFNAMSIZ];
+	char mc_devaddr[ETH_ALEN];
+	struct net_device  __attribute__((aligned(8))) *dev;
+};
+
+#endif /* _ARPT_MANGLE_H */
net/ipv4/netfilter/Kconfig +12 -0
index 3816e1d..50f38b2 100644
@@ -392,6 +392,18 @@ config IP_NF_ARP_MANGLE
 	  Allows altering the ARP packet payload: source and destination
 	  hardware and network addresses.
 
+config IP_NF_ARP_MCMANGLE
+	tristate "ARP multicast address mangling"
+	help
+	  Allows altering the source unicast hardware address in ARP messages
+	  with a multicast hardware address. This target is useful to make a
+	  switch flood to all ports whose ethernet device have a multicast
+	  hardware address. Please, see that this target violates RFC 1812
+	  (section 3.3.2) since an ethernet device must not use a multicast
+	  link address.
+
+	  To compile it as a module, choose M here.  If unsure, say N.
+
 endif # IP_NF_ARPTABLES
 
 endmenu
net/ipv4/netfilter/Makefile +1 -0
index 5f9b650..5c4cc3e 100644
@@ -67,6 +67,7 @@ obj-$(CONFIG_IP_NF_TARGET_ULOG) += ipt_ULOG.o
 # generic ARP tables
 obj-$(CONFIG_IP_NF_ARPTABLES) += arp_tables.o
 obj-$(CONFIG_IP_NF_ARP_MANGLE) += arpt_mangle.o
+obj-$(CONFIG_IP_NF_ARP_MCMANGLE) += arpt_mcmangle.o
 
 # just filtering instance of ARP tables for now
 obj-$(CONFIG_IP_NF_ARPFILTER) += arptable_filter.o
net/ipv4/netfilter/arpt_mcmangle.c new file +98 -0

98 lines not on this page · raw message

author maintainer
#1 lore raw

PATCH 2/3 netfilter: xtables: add PKTTYPE target

This patch adds the PKTTYPE target which can be used to mangle the skbuff packet type field. This target is useful in conjunction with the arptables mcmangle target to TCP working again when a multicast hardware address is used. An example of its use:

iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
	-j PKTTYPE --to-pkt-type unicast

Given the following arptables rule-set:

arptables -I OUTPUT -o eth0 -j mcmangle --h-length 6

	\ --mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth0
arptables -I INPUT --h-length 6 --destination-mac 01:00:5e:00:01:01
	\ -j mangle --mangle-mac-d 00:zz:yy:xx:5a:27

See arptables mcmangle target for further information.

---

 include/linux/netfilter/xt_PKTTYPE.h |    8 ++++
 net/netfilter/Kconfig                |   18 ++++++++++
 net/netfilter/Makefile               |    1 +
 net/netfilter/xt_PKTTYPE.c           |   61 ++++++++++++++++++++++++++++++++++
 4 files changed, 88 insertions(+), 0 deletions(-)
 create mode 100644 include/linux/netfilter/xt_PKTTYPE.h
 create mode 100644 net/netfilter/xt_PKTTYPE.c
include/linux/netfilter/xt_PKTTYPE.h new file +8 -0
new file mode 100644
index 0000000..cc67cbf
@@ -0,0 +1,8 @@
+#ifndef _XT_PKTTYPE_TARGET_H
+#define _XT_PKTTYPE_TARGET_H
+
+struct xt_pkttype_target_info {
+	u_int8_t pkt_type;
+};
+
+#endif /* _XT_PKTTYPE_TARGET_H */
net/netfilter/Kconfig +18 -0
index 25dcef9..9ed1ccf 100644
@@ -489,6 +489,24 @@ config NETFILTER_XT_TARGET_TCPOPTSTRIP
 	  This option adds a "TCPOPTSTRIP" target, which allows you to strip
 	  TCP options from TCP packets.
 
+config NETFILTER_XT_TARGET_PKTTYPE
+	tristate  '"PKTTYPE" target support'
+	depends on IP_NF_RAW || IP6_NF_RAW
+	depends on NETFILTER_ADVANCED
+	help
+	  The PKTTYPE target allows you to change the link layer packet type.
+	  This target is useful if you have set up a multicast MAC address (via
+	  arptables) for a given interface and you want the packets to reach
+	  the layer 4 stack (which would drop packet tagged as multicast
+	  from the link layer).
+
+	  This target can be used in conjunction with arptables and the cluster
+	  match to setup cluster of stateful firewalls which are connected
+	  through a switch.
+
+	  If you want to compile it as a module, say M here and read
+	  <file:Documentation/kbuild/modules.txt>.  If unsure, say `N'.
+
 config NETFILTER_XT_MATCH_COMMENT
 	tristate  '"comment" match support'
 	depends on NETFILTER_ADVANCED
net/netfilter/Makefile +1 -0
index da3d909..dd43ba9 100644
@@ -55,6 +55,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_TPROXY) += xt_TPROXY.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_TCPMSS) += xt_TCPMSS.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_TCPOPTSTRIP) += xt_TCPOPTSTRIP.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_TRACE) += xt_TRACE.o
+obj-$(CONFIG_NETFILTER_XT_TARGET_PKTTYPE) += xt_PKTTYPE.o
 
 # matches
 obj-$(CONFIG_NETFILTER_XT_MATCH_COMMENT) += xt_comment.o
net/netfilter/xt_PKTTYPE.c new file +61 -0

61 lines not on this page · raw message

author maintainer
#15 lore raw

PATCH 3/3 netfilter: xtables: add cluster match

In reply to Pablo Neira Ayuso

This patch adds the iptables cluster match. This match can be used to deploy gateway and back-end load-sharing clusters. Assuming that all the nodes see all packets (see arptables mcmangle target and PKTTYPE iptables targets on how to do that), the cluster match decides if this node has to handle a packet given:

	jhash(source IP) % nodeID == 0

For related connections, the master conntrack is used. An example of its use for a gateway cluster, in one of the cluster nodes:

iptables -I PREROUTING -t mangle -i eth1 -m cluster \
	--cluster-total-nodes 2 --cluster-local-node 1 \
	--cluster-proc-name eth1 -j MARK --set-mark 0xffff
iptables -A PREROUTING -t mangle -i eth1 \
	-m mark ! --mark 0xffff -j DROP
iptables -A PREROUTING -t mangle -i eth2 -m cluster \
	--cluster-total-nodes 2 --cluster-local-node 1 \
	--cluster-proc-name eth2 -j MARK --set-mark 0xffff
iptables -A PREROUTING -t mangle -i eth2 \
	-m mark ! --mark 0xffff -j DROP

And the following rule-set to make all nodes see all the packets:

arptables -I OUTPUT -o eth1 -j mcmangle --h-length 6 \
	--mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth1
arptables -I INPUT -i eth1 --h-length 6 \
	--destination-mac 01:00:5e:00:01:01 \
	-j mangle --mangle-mac-d 00:zz:yy:xx:5a:27
arptables -I OUTPUT -o eth2 -j mcmangle --h-length 6 \
	--mc-mangle-mac 01:00:5e:00:01:02 --mc-mangle-dev eth2
arptables -I INPUT -i eth2 --h-length 6 \
	--destination-mac 01:00:5e:00:01:02 \
	-j mangle --mangle-mac-d 00:zz:yy:xx:5a:27
iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
	-j PKTTYPE --to-pkt-type unicast

In the case of TCP connections, pickup facility has to be disabled to avoid marking TCP ACK packets coming in the reply direction as valid.

echo 0 > /proc/sys/net/netfilter/nf_conntrack_tcp_loose

The match also provides a /proc entry under:

/proc/sys/net/netfilter/cluster/$PROC_NAME

where PROC_NAME is set via --cluster-proc-name. This is useful to include possible cluster reconfigurations via fail-over scripts.

---

 include/linux/netfilter/xt_cluster.h |   21 ++
 net/netfilter/Kconfig                |   17 ++
 net/netfilter/Makefile               |    1 
 net/netfilter/xt_cluster.c           |  333 ++++++++++++++++++++++++++++++++++
 4 files changed, 372 insertions(+), 0 deletions(-)
 create mode 100644 include/linux/netfilter/xt_cluster.h
 create mode 100644 net/netfilter/xt_cluster.c
include/linux/netfilter/xt_cluster.h new file +21 -0
new file mode 100644
index 0000000..85ab07f
@@ -0,0 +1,21 @@
+#ifndef _XT_CLUSTER_MATCH_H
+#define _XT_CLUSTER_MATCH_H
+
+struct proc_dir_entry;
+
+enum xt_cluster_flags {
+	XT_CLUSTER_F_INV = 0,
+};
+
+struct xt_cluster_match_info {
+	u_int16_t		total_nodes;
+	u_int16_t		node_id;
+	u_int32_t		hash_seed;
+	char			proc_name[16];
+	unsigned int		flags;
+
+	/* Used internally by the kernel */
+	void			*data __attribute__((aligned(8)));
+};
+
+#endif /* _XT_CLUSTER_MATCH_H */
net/netfilter/Kconfig +17 -0
index 9ed1ccf..33d37d9 100644
@@ -872,6 +872,23 @@ config NETFILTER_XT_MATCH_U32
 
 	  Details and examples are in the kernel module source.
 
+config NETFILTER_XT_MATCH_CLUSTER
+	tristate '"cluster" match support'
+	depends on NETFILTER_ADVANCED
+	---help---
+	  This option allows you to build work-load-sharing clusters of
+	  network servers/stateful firewalls without having a dedicated
+	  load-balancing router/server/switch. Basically, this match returns
+	  true when the packet must be handled by this cluster node. Thus,
+	  all nodes see all packets and this match decides which node handles
+	  what packets. The work-load sharing algorithm is based on source
+	  address hashing.
+
+	  If you say Y here, try `iptables -m cluster --help` for
+	  more information. See the PKTTYPE target and the mcmangle arptables
+	  target on how to make your nodes see all packets. You can also have
+	  a look at man iptables(8) for some examples on the usage.
+
 endif # NETFILTER_XTABLES
 
 endmenu
net/netfilter/Makefile +1 -0
index dd43ba9..29049e4 100644
@@ -90,6 +90,7 @@ obj-$(CONFIG_NETFILTER_XT_MATCH_STRING) += xt_string.o
 obj-$(CONFIG_NETFILTER_XT_MATCH_TCPMSS) += xt_tcpmss.o
 obj-$(CONFIG_NETFILTER_XT_MATCH_TIME) += xt_time.o
 obj-$(CONFIG_NETFILTER_XT_MATCH_U32) += xt_u32.o
+obj-$(CONFIG_NETFILTER_XT_MATCH_CLUSTER) += xt_cluster.o
 
 # IPVS
 obj-$(CONFIG_IP_VS) += ipvs/
net/netfilter/xt_cluster.c new file +333 -0

333 lines not on this page · raw message

#16 lore raw

Re: [PATCH 3/3] netfilter: xtables: add cluster match

Quoted 6 lines · Pablo Neira Ayuso

This patch adds the iptables cluster match. This match can be used to deploy gateway and back-end load-sharing clusters. Assuming that all the nodes see all packets
(see arptables mcmangle target and PKTTYPE iptables targets on how to do that), the cluster match decides if this node has to handle a packet given:

Is not this what CLUSTERIP is essentially supposed to do?

include/linux/netfilter/xt_cluster.h line 15

Quoted 9 lines
	jhash(source IP) % nodeID == 0
include/linux/netfilter/xt_cluster.h new file +6 -0
+struct xt_cluster_match_info {
+	u_int16_t		total_nodes;
+	u_int16_t		node_id;
+	u_int32_t		hash_seed;
+	char			proc_name[16];
+	unsigned int		flags;

Ouch, don't use unfixated types like unsigned int.

Quoted 17 lines
net/netfilter/Kconfig +17 -0
+config NETFILTER_XT_MATCH_CLUSTER
+	tristate '"cluster" match support'
+	depends on NETFILTER_ADVANCED
+	---help---
+	  This option allows you to build work-load-sharing clusters of
+	  network servers/stateful firewalls without having a dedicated
+	  load-balancing router/server/switch. Basically, this match returns
+	  true when the packet must be handled by this cluster node. Thus,
+	  all nodes see all packets and this match decides which node handles
+	  what packets. The work-load sharing algorithm is based on source
+	  address hashing.
+
+	  If you say Y here, try `iptables -m cluster --help` for
+	  more information. See the PKTTYPE target and the mcmangle arptables
+	  target on how to make your nodes see all packets. You can also have
+	  a look at man iptables(8) for some examples on the usage.
+
Quoted 1 line

endif # NETFILTER_XTABLES

Please keep the list alphabetical.

Quoted 10 lines

endmenu

net/netfilter/Makefile +0 -0
index dd43ba9..29049e4 100644
@@ -90,6 +90,7 @@ obj-$(CONFIG_NETFILTER_XT_MATCH_STRING) += xt_string.o

obj-$(CONFIG_NETFILTER_XT_MATCH_TCPMSS) += xt_tcpmss.o obj-$(CONFIG_NETFILTER_XT_MATCH_TIME) += xt_time.o obj-$(CONFIG_NETFILTER_XT_MATCH_U32) += xt_u32.o +obj-$(CONFIG_NETFILTER_XT_MATCH_CLUSTER) += xt_cluster.o

Similarly.

net/netfilter/Makefile

Quoted 2 lines
net/netfilter/Makefile +2 -0
+struct xt_cluster_internal {
+	unsigned long		node_mask;

In the back of my head I seem to remember a cpumask_t.

net/netfilter/Makefile

Quoted 10 lines
net/netfilter/Makefile +10 -0
+	struct proc_dir_entry	*proc;
+	atomic_t		use;
+};
+
+#ifdef CONFIG_PROC_FS
+static void *xt_cluster_seq_start(struct seq_file *s, loff_t *pos)
+{
+	if (*pos == 0) {
+		struct xt_cluster_internal *data =
+			(struct xt_cluster_internal *)s->private;

Redundant cast ;)

net/netfilter/Makefile

Quoted 3 lines
net/netfilter/Makefile +3 -0
+static int xt_cluster_seq_show(struct seq_file *s, void *v)
+{
+	unsigned long *mask = (unsigned long *)v;

Same.

Quoted 1 line

+static struct seq_operations xt_cluster_seq_ops = {

static const struct ...

net/netfilter/Makefile

Quoted 8 lines
net/netfilter/Makefile +8 -0
+#define XT_CLUSTER_PROC_WRITELEN	10
+
+static ssize_t 
+xt_cluster_write_proc(struct file *file, const char __user *input,
+		      size_t size, loff_t *ofs)
+{
+	const struct proc_dir_entry *pde = PDE(file->f_path.dentry->d_inode);
+        struct xt_cluster_internal *info = pde->data;

^^

net/netfilter/Makefile

Quoted 9 lines
net/netfilter/Makefile +9 -0
+	char buffer[XT_CLUSTER_PROC_WRITELEN+1];
+	unsigned int new_node_id;
+
+	if (copy_from_user(buffer, input, XT_CLUSTER_PROC_WRITELEN))
+		return -EFAULT;
+
+	switch(*buffer) {
+	case '+':
+		new_node_id = simple_strtoul(buffer+1, NULL, 10);

Perhaps strict_strtoul would be preferred?

Quoted 13 lines
net/netfilter/Makefile +12 -0
+	case '-':
+		new_node_id = simple_strtoul(buffer+1, NULL, 10);
 
+static struct xt_match xt_cluster_match[] __read_mostly = {
+	{
+		.name		= "cluster",
+		.family		= AF_INET,
+		.match		= xt_cluster_mt,
+		.checkentry	= xt_cluster_mt_checkentry,
+		.destroy	= xt_cluster_mt_destroy,
+		.matchsize	= sizeof(struct xt_cluster_match_info),
+		.me		= THIS_MODULE,
+	},

net/netfilter/Makefile

Quoted 10 lines
net/netfilter/Makefile +10 -0
+	{
+		.name		= "cluster",
+		.family		= AF_INET6,
+		.match		= xt_cluster_mt,
+		.checkentry	= xt_cluster_mt_checkentry,
+		.destroy	= xt_cluster_mt_destroy,
+		.matchsize	= sizeof(struct xt_cluster_match_info),
+		.me		= THIS_MODULE,
+	},
+};

First, NFPROTO_IPV4 instead of AF_INET, and NFPROTO_IPV6 instead of AF_INET6.
Since these are the same, too, they can be combined with .family = NFPROTO_UNSPEC.

net/netfilter/Makefile

Quoted 13 lines
net/netfilter/Makefile +13 -0
+static int __init xt_cluster_mt_init(void)
+{
+	int ret;
+
+#ifdef CONFIG_PROC_FS
+	proc_cluster = proc_mkdir("cluster", proc_net_netfilter);
+	if (!proc_cluster)
+		return -ENOMEM;
+#endif
+	ret = xt_register_matches(xt_cluster_match,
+				  ARRAY_SIZE(xt_cluster_match));
+	if (ret < 0)
+		return ret;

When register_match fails, you still have to remove the proc directory.

net/netfilter/Makefile

Quoted 3 lines
net/netfilter/Makefile +3 -0
+MODULE_AUTHOR("Pablo Neira Ayuso <[email protected]>");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Xtables: xt_cluster match");

A better short-description would be nice (the user could have probably guessed already that xt_cluster.ko is a cluster match..), maybe something like "cluster multicast filtering"?

#2 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

In reply to Pablo Neira Ayuso
Quoted 7 lines · Pablo Neira Ayuso

This patch adds the PKTTYPE target which can be used to mangle the skbuff packet type field. This target is useful in conjunction with the arptables mcmangle target to TCP working again when a multicast hardware address is used. An example of its use:

iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
	-j PKTTYPE --to-pkt-type unicast

Are not packets not destined for 224/4 - assuming standard route setups - unicast anyway?

net/netfilter/Kconfig config NETFILTER_XT_TARGET_TCPOPTSTRIP · line 492

Quoted 5 lines
net/netfilter/Kconfig +1 -0
@@ -489,6 +489,24 @@ config NETFILTER_XT_TARGET_TCPOPTSTRIP
 	  This option adds a "TCPOPTSTRIP" target, which allows you to strip
 	  TCP options from TCP packets.
 
+config NETFILTER_XT_TARGET_PKTTYPE

Please adhere to alphabetical ordering. (Andrew Morton mentioned it time and again that it causes the least merge conflicts where conflicts would happen if multiple people always add to the last item.) Also, searching for "PKTTYPE"/("cluster") in make menuconfig in the otherwise-sorted Kconfig/Makefile list could provide confusing if it was out-of-order.

net/netfilter/Kconfig config NETFILTER_XT_TARGET_TCPOPTSTRIP · line 508

Quoted 2 lines
net/netfilter/Kconfig +2 -0
+	  If you want to compile it as a module, say M here and read
+	  <file:Documentation/kbuild/modules.txt>.  If unsure, say `N'.

A lovely paragraph, but I think it has lost its purpose given that NETFILTER_ADVANCED can now be general-tagged "if unsure, pick N" and the rest will be automagic.

Quoted 15 lines
(quoted diff) +1 -0
+++ b/net/netfilter/Makefile
@@ -55,6 +55,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_TPROXY) += xt_TPROXY.o

obj-$(CONFIG_NETFILTER_XT_TARGET_TCPMSS) += xt_TCPMSS.o obj-$(CONFIG_NETFILTER_XT_TARGET_TCPOPTSTRIP) += xt_TCPOPTSTRIP.o obj-$(CONFIG_NETFILTER_XT_TARGET_TRACE) += xt_TRACE.o

net/netfilter/xt_PKTTYPE.c new file +9 -0
+obj-$(CONFIG_NETFILTER_XT_TARGET_PKTTYPE) += xt_PKTTYPE.o
 
+static struct xt_target xt_pkttype_target[] __read_mostly = {
+	{
+		.family		= AF_INET,
+		.name		= "PKTTYPE",
+		.table		= "mangle",
+		.target		= xt_pkttype_tg,
+		.targetsize	= sizeof(struct xt_pkttype_target_info),
+		.me		= THIS_MODULE,

net/netfilter/xt_PKTTYPE.c line 42

Quoted 10 lines
net/netfilter/xt_PKTTYPE.c new file +10 -0
+	},
+	{
+		.family		= AF_INET6,
+		.name		= "PKTTYPE",
+		.table		= "mangle",
+		.target		= xt_pkttype_tg,
+		.targetsize	= sizeof(struct xt_pkttype_target_info),
+		.me		= THIS_MODULE,
+	},
+};

Combine to NFPROTO_UNSPEC.

net/netfilter/xt_PKTTYPE.c line 57

Quoted 3 lines
net/netfilter/xt_PKTTYPE.c new file +3 -0
+MODULE_AUTHOR("Pablo Neira Ayuso <[email protected]>");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Xtables: xt_PKTTYPE target");

I propose
MODULE_DESCRIPTION("Xtables: set skbuff packet type");
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

author maintainer
#17 lore raw

Re: [PATCH 3/3] netfilter: xtables: add cluster match

In reply to Jan Engelhardt
Quoted 10 lines · Jan Engelhardt
Quoted 6 lines · Pablo Neira Ayuso

This patch adds the iptables cluster match. This match can be used to deploy gateway and back-end load-sharing clusters. Assuming that all the nodes see all packets
(see arptables mcmangle target and PKTTYPE iptables targets on how to do that), the cluster match decides if this node has to handle a packet given:

Is not this what CLUSTERIP is essentially supposed to do?

No, you can't deploy load-sharing setups for gateways as the -d parameter is mandatory, CLUSTERIP only works for back-end nodes and have a couple of significant problems. Indeed, these stuff supersedes CLUSTERIP.

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
author maintainer
#3 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

In reply to Jan Engelhardt
Quoted 7 lines · Jan Engelhardt
Quoted 5 lines · Pablo Neira Ayuso

This patch adds the PKTTYPE target which can be used to mangle the skbuff packet type field. This target is useful in conjunction with the arptables mcmangle target to TCP working again when a multicast hardware address is used. An example of its use:

Quoted 5 lines
Quoted 2 lines
iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
	-j PKTTYPE --to-pkt-type unicast

Are not packets not destined for 224/4 - assuming standard route setups - unicast anyway?

Not if you use the mcmangle arptables target to use a multicast hardware address instead of the normal unicast, in that case, the link layer set them to PACKET_MULTICAST. Indeed, the only use that I can find for this target at the moment is to combine it with mcmangle.

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
#20 lore raw
In reply to Pablo Neira Ayuso
Quoted 7 lines · Pablo Neira Ayuso

This patch adds the mcmangle target for arptables which allows altering the source hardware address in ARP with a multicast hardware address. This target can be used to make a switch flood packets to the ports that use the same MAC multicast address. This is useful to deploy load-sharing clusters in environments in which the switch does not provide a way to flood packets to several ports.

Great name. Took me a bit to realize the connection to "multicast" :)

Quoted 13 lines

Since all the nodes receives the same packets, each decides if it handles the packet based on hashing approach (See the `cluster' iptables match that comes with this patchset).

Theoretically, the use of the reserved VRRP hardware address should be fine for this, however, switches generally treat this hardware address space as normal unicast hardware address. Thus, in practise, it is not possible to have two nodes with the same VRRP hardware address.

Please, note that this target violates RFC 1812 (section 3.3.2) since an ethernet device must not use a multicast link address.

Quoted 6 lines

An example of the use of this target:

arptables -I OUTPUT -o eth0 -j mcmangle --h-length 6 \
	--mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth0
arptables -I INPUT -i eth0 --h-length 6 --destination-mac \
	01:00:5e:00:01:01 -j mangle --mangle-mac-d 00:zz:yy:xx:5a:27

Wouldn't it be more generically usable if it was a simple "set either source or destination mac address to any value" target? In fact thats what it seems to be (or is there a multicast check?), so perhaps we should call it S/DNAT for consistency with iptables and ebtables?

I'm wondering though why the device is needed as a parameter .. ah I see:

+ if (dev_mc_add(dev, mangle->mc_devaddr, ETH_ALEN, 0) < 0) {
+ printk(KERN_ERR "arpt_mcmangle: cannot set multicast "
+ "address\n");
+ return false;
+ }

Continuing the idea of a generic ARP address mangling target, this would have to be done in userspace using SIOCADDMULTI.

I would also prefer that approach because the multicast mangling seems a bit like a hack which only works when the requesting host accepts a multicast MAC address in the ARP reply.

#4 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

In reply to Pablo Neira Ayuso
Quoted 8 lines · Pablo Neira Ayuso

This patch adds the PKTTYPE target which can be used to mangle the skbuff packet type field. This target is useful in conjunction with the arptables mcmangle target to TCP working again when a multicast hardware address is used. An example of its use:

iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
	-j PKTTYPE --to-pkt-type unicast
Quoted 8 lines

Given the following arptables rule-set:

arptables -I OUTPUT -o eth0 -j mcmangle --h-length 6

	\ --mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth0
arptables -I INPUT --h-length 6 --destination-mac 01:00:5e:00:01:01
	\ -j mangle --mangle-mac-d 00:zz:yy:xx:5a:27

See arptables mcmangle target for further information.

That one refers to this patch :) So you're actually communicating using TCP and multicast? Why don't you use UDP, which works fine using multicast without pkttype mangling?

#18 lore raw

Re: [PATCH 3/3] netfilter: xtables: add cluster match

In reply to Pablo Neira Ayuso
Quoted 19 lines · Pablo Neira Ayuso

This patch adds the iptables cluster match. This match can be used to deploy gateway and back-end load-sharing clusters. Assuming that all the nodes see all packets (see arptables mcmangle target and PKTTYPE iptables targets on how to do that), the cluster match decides if this node has to handle a packet given:

	jhash(source IP) % nodeID == 0

For related connections, the master conntrack is used. An example of its use for a gateway cluster, in one of the cluster nodes:

iptables -I PREROUTING -t mangle -i eth1 -m cluster \
	--cluster-total-nodes 2 --cluster-local-node 1 \
	--cluster-proc-name eth1 -j MARK --set-mark 0xffff
iptables -A PREROUTING -t mangle -i eth1 \
	-m mark ! --mark 0xffff -j DROP
iptables -A PREROUTING -t mangle -i eth2 -m cluster \
	--cluster-total-nodes 2 --cluster-local-node 1 \
	--cluster-proc-name eth2 -j MARK --set-mark 0xffff
Quoted 1 line

iptables -A PREROUTING -t mangle -i eth2 \

While its simple and probably gives good distribution, using the source IP like this has the major disadvantage that when one node fails, all nodes need to change rulesets and the existing mapping of connections becomes at least partially invalid.

Also using conntrack data implies that this can't be used in the raw table, which seems like the best place to drop foreign packets.

author maintainer
#21 lore raw
In reply to Patrick McHardy
Quoted 1 line · Patrick McHardy

Great name. Took me a bit to realize the connection to "multicast" :)

I thought about mcnugget target, but it seems to be already copyrighted ;)

Quoted 6 lines

Continuing the idea of a generic ARP address mangling target, this would have to be done in userspace using SIOCADDMULTI.

I would also prefer that approach because the multicast mangling seems a bit like a hack which only works when the requesting host accepts a multicast MAC address in the ARP reply.

Indeed, I didn't know about SIOCADDMULTI, great. I'll do it like you have proposed. Thanks.

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
author maintainer
#5 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

In reply to Patrick McHardy
Quoted 16 lines · Patrick McHardy
Quoted 15 lines · Pablo Neira Ayuso

This patch adds the PKTTYPE target which can be used to mangle the skbuff packet type field. This target is useful in conjunction with the arptables mcmangle target to TCP working again when a multicast hardware address is used. An example of its use:

iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
    -j PKTTYPE --to-pkt-type unicast

Given the following arptables rule-set:

arptables -I OUTPUT -o eth0 -j mcmangle --h-length 6

    \ --mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth0
arptables -I INPUT --h-length 6 --destination-mac 01:00:5e:00:01:01
    \ -j mangle --mangle-mac-d 00:zz:yy:xx:5a:27
Quoted 5 lines
Quoted 1 line

See arptables mcmangle target for further information.

That one refers to this patch :) So you're actually communicating using TCP and multicast? Why don't you use UDP, which works fine using multicast without pkttype mangling?

If the netdevice uses multicast MAC address, the link layer sets skbuff pkttype to PACKET_MULTICAST and TCP (among others) doesn't like this. This target is required to make TCP work again if a multicast MAC address is used.

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
author maintainer
#19 lore raw

Re: [PATCH 3/3] netfilter: xtables: add cluster match

In reply to Patrick McHardy
Quoted 22 lines · Patrick McHardy
Quoted 19 lines · Pablo Neira Ayuso

This patch adds the iptables cluster match. This match can be used to deploy gateway and back-end load-sharing clusters. Assuming that all the nodes see all packets (see arptables mcmangle target and PKTTYPE iptables targets on how to do that), the cluster match decides if this node has to handle a packet given:

    jhash(source IP) % nodeID == 0

For related connections, the master conntrack is used. An example of its use for a gateway cluster, in one of the cluster nodes:

iptables -I PREROUTING -t mangle -i eth1 -m cluster \
    --cluster-total-nodes 2 --cluster-local-node 1 \
    --cluster-proc-name eth1 -j MARK --set-mark 0xffff
iptables -A PREROUTING -t mangle -i eth1 \
    -m mark ! --mark 0xffff -j DROP
iptables -A PREROUTING -t mangle -i eth2 -m cluster \
    --cluster-total-nodes 2 --cluster-local-node 1 \
    --cluster-proc-name eth2 -j MARK --set-mark 0xffff
Quoted 1 line

iptables -A PREROUTING -t mangle -i eth2 \

Quoted 4 lines

While its simple and probably gives good distribution, using the source IP like this has the major disadvantage that when one node fails, all nodes need to change rulesets and the existing mapping of connections becomes at least partially invalid.

But this is the same problem that people have in fault-tolerant primary-backup stateful firewall setups. That's why the conntrack-tools are there, just in case that you want fault-tolerant stateful firewalls ;).

Quoted 2 lines

Also using conntrack data implies that this can't be used in the raw table, which seems like the best place to drop foreign packets.

Indeed, but this relies on the conntrack data to hash the same tuple for all the packets.

Look at the numbers, they are not that bad, I'm getting ~45% more performance while having my two firewalls running (before one was idle in my primary-backup setup).

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
author maintainer
#22 lore raw
In reply to Pablo Neira Ayuso
Quoted 8 lines · Pablo Neira Ayuso
Quoted 1 line · Patrick McHardy

Great name. Took me a bit to realize the connection to "multicast" :)

I thought about mcnugget target, but it seems to be already copyrighted ;)

Quoted 3 lines

Continuing the idea of a generic ARP address mangling target, this would have to be done in userspace using SIOCADDMULTI.

Quoted 6 lines
Quoted 3 lines

I would also prefer that approach because the multicast mangling seems a bit like a hack which only works when the requesting host accepts a multicast MAC address in the ARP reply.

Indeed, I didn't know about SIOCADDMULTI, great. I'll do it like you have proposed. Thanks.

Hey, just to let you know ;). The 'ip' tool already allows adding static multicast MAC addresses via SIOCADDMULTI. The following commands do the trick:

ip maddr add 01:00:5e:00:01:01 dev eth1
arptables -I OUTPUT -o eth1 --h-length 6 \
-j mangle --mangle-mac-s 01:00:5e:00:01:01 arptables -I INPUT -i eth1 --h-length 6 --destination-mac \
01:00:5e:00:01:01 -j mangle --mangle-mac-d $REAL_HWADDR1

Still the PKTTYPE iptables target that I posted is needed to make TCP and friends work in a devide that uses a multicast MAC.

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
#6 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

In reply to Pablo Neira Ayuso
Quoted 18 lines · Pablo Neira Ayuso
Quoted 17 lines · Patrick McHardy
Quoted 16 lines · Pablo Neira Ayuso

This patch adds the PKTTYPE target which can be used to mangle the skbuff packet type field. This target is useful in conjunction with the arptables mcmangle target to TCP working again when a multicast hardware address is used. An example of its use:

iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
    -j PKTTYPE --to-pkt-type unicast

Given the following arptables rule-set:

arptables -I OUTPUT -o eth0 -j mcmangle --h-length 6

    \ --mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth0
arptables -I INPUT --h-length 6 --destination-mac 01:00:5e:00:01:01
    \ -j mangle --mangle-mac-d 00:zz:yy:xx:5a:27

See arptables mcmangle target for further information.

Quoted 8 lines
Quoted 3 lines

That one refers to this patch :) So you're actually communicating using TCP and multicast? Why don't you use UDP, which works fine using multicast without pkttype mangling?

If the netdevice uses multicast MAC address, the link layer sets skbuff pkttype to PACKET_MULTICAST and TCP (among others) doesn't like this. This target is required to make TCP work again if a multicast MAC address is used.

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

It really seems pretty hackish to add netfilter modules to work around valid checks in the stack. I'd prefer if we can come up with a nicer way that offers you the same functionality.

#7 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 23 lines · Patrick McHardy
Quoted 22 lines · Pablo Neira Ayuso
Quoted 20 lines · Patrick McHardy
Quoted 8 lines · Pablo Neira Ayuso

This patch adds the PKTTYPE target which can be used to mangle the skbuff packet type field. This target is useful in conjunction with the arptables mcmangle target to TCP working again when a multicast hardware address is used. An example of its use:

iptables -I PREROUTING ! -s 224.0.0.0/4 -t mangle \
    -j PKTTYPE --to-pkt-type unicast
Quoted 8 lines

Given the following arptables rule-set:

arptables -I OUTPUT -o eth0 -j mcmangle --h-length 6

    \ --mc-mangle-mac 01:00:5e:00:01:01 --mc-mangle-dev eth0
arptables -I INPUT --h-length 6 --destination-mac 01:00:5e:00:01:01
    \ -j mangle --mangle-mac-d 00:zz:yy:xx:5a:27

See arptables mcmangle target for further information.

That one refers to this patch :) So you're actually communicating using TCP and multicast? Why don't you use UDP, which works fine using multicast without pkttype mangling?

Quoted 8 lines
Quoted 4 lines

If the netdevice uses multicast MAC address, the link layer sets skbuff pkttype to PACKET_MULTICAST and TCP (among others) doesn't like this. This target is required to make TCP work again if a multicast MAC address is used.

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

No, that's not the syncronizing traffic, but the "normal" TCP traffic to be filtered by the firewalls, which have got multicast MAC addresses on their interfaces.

Best regards,
Jzosef
-
E-mail : [email protected], [email protected] PGP key : http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address : KFKI Research Institute for Particle and Nuclear Physics

          H-1525 Budapest 114, POB. 49, Hungary
#8 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 7 lines · Jozsef Kadlecsik
Quoted 3 lines · Patrick McHardy

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

No, that's not the syncronizing traffic, but the "normal" TCP traffic to be filtered by the firewalls, which have got multicast MAC addresses on their interfaces.

Multicast traffic is accepted for forwarding just fine, its just local TCP delivery thats refusing it. So it can't be forwarded traffic.

author maintainer
#9 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 7 lines · Patrick McHardy
Quoted 6 lines · Jozsef Kadlecsik
Quoted 3 lines · Patrick McHardy

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

Quoted 7 lines
Quoted 3 lines

No, that's not the syncronizing traffic, but the "normal" TCP traffic to be filtered by the firewalls, which have got multicast MAC addresses on their interfaces.

Multicast traffic is accepted for forwarding just fine, its just local TCP delivery thats refusing it. So it can't be forwarded traffic.

You usually have some administration facility (like ssh) that would break. Please, think that this can be also used to replace CLUSTERIP (to be used in back-end servers, not only stateful firewalls).

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
#10 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 10 lines · Pablo Neira Ayuso
Quoted 9 lines · Patrick McHardy
Quoted 8 lines · Jozsef Kadlecsik
Quoted 3 lines · Patrick McHardy

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

No, that's not the syncronizing traffic, but the "normal" TCP traffic to be filtered by the firewalls, which have got multicast MAC addresses on their interfaces.

Quoted 7 lines
Quoted 3 lines

Multicast traffic is accepted for forwarding just fine, its just local TCP delivery thats refusing it. So it can't be forwarded traffic.

You usually have some administration facility (like ssh) that would break. Please, think that this can be also used to replace CLUSTERIP (to be used in back-end servers, not only stateful firewalls).

I see. Still, this module has only one purpose, which is to circumvent valid checks to make a different hack (although a nicer one) work. Perhaps simply move it into the cluster match (yes yes, I know). That way we can also avoid a bit of the new-module overhead.

author maintainer
#11 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 15 lines · Patrick McHardy
Quoted 14 lines · Pablo Neira Ayuso
Quoted 12 lines · Patrick McHardy
Quoted 8 lines · Jozsef Kadlecsik
Quoted 3 lines · Patrick McHardy

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

No, that's not the syncronizing traffic, but the "normal" TCP traffic to be filtered by the firewalls, which have got multicast MAC addresses on their interfaces.

Multicast traffic is accepted for forwarding just fine, its just local TCP delivery thats refusing it. So it can't be forwarded traffic.

Quoted 8 lines
Quoted 3 lines

You usually have some administration facility (like ssh) that would break. Please, think that this can be also used to replace CLUSTERIP (to be used in back-end servers, not only stateful firewalls).

I see. Still, this module has only one purpose, which is to circumvent valid checks to make a different hack (although a nicer one) work. Perhaps simply move it into the cluster match (yes yes, I know). That way we can also avoid a bit of the new-module overhead.

Then, the cluster match would become the target match, since skbuff cannot be modified from matches (they are passed as const). Becoming a target means less flexibility :(

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
#12 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 18 lines · Pablo Neira Ayuso
Quoted 17 lines · Patrick McHardy
Quoted 16 lines · Pablo Neira Ayuso
Quoted 6 lines · Patrick McHardy
Quoted 5 lines · Jozsef Kadlecsik
Quoted 3 lines · Patrick McHardy

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

Quoted 6 lines
Quoted 3 lines

No, that's not the syncronizing traffic, but the "normal" TCP traffic to be filtered by the firewalls, which have got multicast MAC addresses on their interfaces.

Multicast traffic is accepted for forwarding just fine, its just local TCP delivery thats refusing it. So it can't be forwarded traffic.

You usually have some administration facility (like ssh) that would break. Please, think that this can be also used to replace CLUSTERIP (to be used in back-end servers, not only stateful firewalls).

Quoted 8 lines
Quoted 4 lines

I see. Still, this module has only one purpose, which is to circumvent valid checks to make a different hack (although a nicer one) work. Perhaps simply move it into the cluster match (yes yes, I know). That way we can also avoid a bit of the new-module overhead.

Then, the cluster match would become the target match, since skbuff cannot be modified from matches (they are passed as const). Becoming a target means less flexibility :(

Well, a cast should "fix" that :) But feel free to suggest a better method that doesn't need to expose this as a standalone feature.

author maintainer
#13 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 25 lines · Patrick McHardy
Quoted 24 lines · Pablo Neira Ayuso
Quoted 22 lines · Patrick McHardy
Quoted 10 lines · Pablo Neira Ayuso
Quoted 9 lines · Patrick McHardy
Quoted 8 lines · Jozsef Kadlecsik
Quoted 3 lines · Patrick McHardy

Yes, I know, I'm just wondering why you're using TCP at all for synchronizing. Its not for traffic from the Internet I assume since the node it ends up on is unknown to the outside anyways.

No, that's not the syncronizing traffic, but the "normal" TCP traffic to be filtered by the firewalls, which have got multicast MAC addresses on their interfaces.

Quoted 7 lines
Quoted 3 lines

Multicast traffic is accepted for forwarding just fine, its just local TCP delivery thats refusing it. So it can't be forwarded traffic.

You usually have some administration facility (like ssh) that would break. Please, think that this can be also used to replace CLUSTERIP (to
be used in back-end servers, not only stateful firewalls).

I see. Still, this module has only one purpose, which is to circumvent valid checks to make a different hack (although a nicer one) work. Perhaps simply move it into the cluster match (yes yes, I know). That way we can also avoid a bit of the new-module overhead.

Quoted 7 lines
Quoted 3 lines

Then, the cluster match would become the target match, since skbuff cannot be modified from matches (they are passed as const). Becoming a target means less flexibility :(

Well, a cast should "fix" that :) But feel free to suggest a better method that doesn't need to expose this as a standalone feature.

Hm, I forgot another point that is the fact that the PKTTYPE target is not always required. Actually, if the switch can flood the same packets to a set of nodes that are part of the cluster via "port mirroring" or if the switch interprets the VRRP reserved MAC address range correctly (I did not find any yet), the nodes in the cluster would not need to use with a multicast MAC address, in that case, the PKTTYPE target would not be required.

Well, I'll do the hackish solution if that relieves the possible extra bloat of one target that indeed only has this purpose ;). I'll send you a new version of the patch asap.

Signature · 1 line
"Los honestos son inadaptados sociales" -- Les Luthiers
#14 lore raw

Re: [PATCH 2/3] netfilter: xtables: add PKTTYPE target

Quoted 13 lines · Pablo Neira Ayuso
Quoted 3 lines · Patrick McHardy

Well, a cast should "fix" that :) But feel free to suggest a better method that doesn't need to expose this as a standalone feature.

Hm, I forgot another point that is the fact that the PKTTYPE target is not always required. Actually, if the switch can flood the same packets to a set of nodes that are part of the cluster via "port mirroring" or if the switch interprets the VRRP reserved MAC address range correctly (I did not find any yet), the nodes in the cluster would not need to use with a multicast MAC address, in that case, the PKTTYPE target would not be required.

Quoted 3 lines

Well, I'll do the hackish solution if that relieves the possible extra bloat of one target that indeed only has this purpose ;). I'll send you a new version of the patch asap.

Yeah, its a bit hackish too. Ideally we find a cleaner way, but the best I could come up with (use MACVLAN to make the multicast address handled as unicast) isn't any cleaner itself.