Skip to contentSkip to conversation

ipv4: add link_filter sysctl

RFC In review Stephen Hemminger 2 participants 4 messages Mar 13, 2009 – Mar 19, 2009

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

RFC 1/1 ipv4: add link_filter sysctl

Add a new parameter that controls how kernel responds to packets when interface is down. This is done to solve the problem of:

Assume topology of:

      A <----------->       Router             X--- down link

10.1.1.2/24 10.1.1.1/24 10.2.1.1/24
eth0 eth1

If A pings 10.2.1.1 then with normal Linux semantics Router would respond even if eth1 link on 10.2.1.1 was down. This causes some network management tools (that work with other router OS's) to falsely report that link is okay.

The problem is that a Linux router does not respond the way other systems do. This is the router equivalent of "Strong ES" model, it is not the same as "Strong ES" as defined in Host Requirements.

The new parameter adds an additional check on slow input packet path, and causes route cache flush if enabled and carrier is lost.

---
Patch against net-next-2.6

 Documentation/networking/ip-sysctl.txt |   13 +++++++++++++
 include/linux/inetdevice.h             |    2 ++
 include/linux/sysctl.h                 |    1 +
 kernel/sysctl_check.c                  |    1 +
 net/ipv4/devinet.c                     |    1 +
 net/ipv4/fib_frontend.c                |    7 +++++++
 net/ipv4/route.c                       |    9 +++++++++
 7 files changed, 34 insertions(+)
Documentation/networking/ip-sysctl.txt +13 -0
@@ -720,6 +720,19 @@ rp_filter - INTEGER
 	Default value is 0. Note that some distributions enable it
 	in startup scripts.
 
+link_filter - INTEGER
+        0 - Allow packets to be received for the address on this interface
+	even if interface is disabled or no carrier.
+
+	1 - Ignore packets received if interface associated with the incoming
+	address is down.
+
+	2 - Ignore packets received if interface associated with the incoming
+	address is down or has no carrier.
+
+	Default value is 0. Note that some distributions enable it
+	in startup scripts.
+
 arp_filter - BOOLEAN
 	1 - Allows you to have multiple network interfaces on the same
 	subnet, and have the ARPs for each interface be answered
include/linux/inetdevice.h +2 -0
@@ -83,6 +83,7 @@ static inline void ipv4_devconf_setall(s
 #define IN_DEV_FORWARD(in_dev)		IN_DEV_CONF_GET((in_dev), FORWARDING)
 #define IN_DEV_MFORWARD(in_dev)		IN_DEV_ANDCONF((in_dev), MC_FORWARDING)
 #define IN_DEV_RPFILTER(in_dev)		IN_DEV_ANDCONF((in_dev), RP_FILTER)
+#define IN_DEV_LINKFILTER(in_dev)	IN_DEV_ORCONF((in_dev), LINKFILTER)
 #define IN_DEV_SOURCE_ROUTE(in_dev)	IN_DEV_ANDCONF((in_dev), \
 						       ACCEPT_SOURCE_ROUTE)
 #define IN_DEV_BOOTP_RELAY(in_dev)	IN_DEV_ANDCONF((in_dev), BOOTP_RELAY)
@@ -110,6 +111,7 @@ static inline void ipv4_devconf_setall(s
 #define IN_DEV_ARP_IGNORE(in_dev)	IN_DEV_MAXCONF((in_dev), ARP_IGNORE)
 #define IN_DEV_ARP_NOTIFY(in_dev)	IN_DEV_MAXCONF((in_dev), ARP_NOTIFY)
 
+
 struct in_ifaddr
 {
 	struct in_ifaddr	*ifa_next;
include/linux/sysctl.h +1 -0
@@ -491,6 +491,7 @@ enum
 	NET_IPV4_CONF_PROMOTE_SECONDARIES=20,
 	NET_IPV4_CONF_ARP_ACCEPT=21,
 	NET_IPV4_CONF_ARP_NOTIFY=22,
+ 	NET_IPV4_CONF_LINKFILTER=23,
 	__NET_IPV4_CONF_MAX
 };
 
kernel/sysctl_check.c +1 -0
@@ -220,6 +220,7 @@ static const struct trans_ctl_table tran
 	{ NET_IPV4_CONF_PROMOTE_SECONDARIES,	"promote_secondaries" },
 	{ NET_IPV4_CONF_ARP_ACCEPT,		"arp_accept" },
 	{ NET_IPV4_CONF_ARP_NOTIFY,		"arp_notify" },
+	{ NET_IPV4_CONF_LINKFILTER,		"link_filter" },
 	{}
 };
 
net/ipv4/devinet.c +1 -0
@@ -1456,6 +1456,7 @@ static struct devinet_sysctl_table {
 					      "force_igmp_version"),
 		DEVINET_SYSCTL_FLUSHING_ENTRY(PROMOTE_SECONDARIES,
 					      "promote_secondaries"),
+		DEVINET_SYSCTL_RW_ENTRY(LINKFILTER, "link_filter"),
 	},
 };
 
net/ipv4/fib_frontend.c +7 -0
@@ -914,6 +914,13 @@ static int fib_inetaddr_event(struct not
 #endif
 		rt_cache_flush(dev_net(dev), -1);
 		break;
+	case NETDEV_CHANGE:
+		if (!netif_carrier_ok(dev)) {
+			struct in_device *in_dev = __in_dev_get_rtnl(dev);
+			if (in_dev && IN_DEV_LINKFILTER(in_dev) > 1)
+				rt_cache_flush(dev_net(dev), -1);
+		}
+		break;
 	case NETDEV_DOWN:
 		fib_del_ifaddr(ifa);
 		if (ifa->ifa_dev->ifa_list == NULL) {
net/ipv4/route.c +9 -0
@@ -2117,6 +2117,15 @@ static int ip_route_input_slow(struct sk
 
 	if (res.type == RTN_LOCAL) {
 		int result;
+		int linkf = IN_DEV_LINKFILTER(in_dev);
+
+		if (linkf) {
+			if (!netif_running(res.fi->fib_dev))
+				goto e_inval;
+			if (linkf > 1 && !netif_carrier_ok(res.fi->fib_dev))
+				goto e_inval;
+		}
+
 		result = fib_validate_source(saddr, daddr, tos,
 					     net->loopback_dev->ifindex,
 					     dev, &spec_dst, &itag);
maintainer
#1 lore raw

From: Stephen Hemminger <[email protected]>
Date: Fri, 13 Mar 2009 16:12:53 -0700

Quoted 18 lines

Add a new parameter that controls how kernel responds to packets when interface is down. This is done to solve the problem of:

Assume topology of:

      A <----------->       Router             X--- down link

10.1.1.2/24 10.1.1.1/24 10.2.1.1/24
eth0 eth1

If A pings 10.2.1.1 then with normal Linux semantics Router would respond even if eth1 link on 10.2.1.1 was down. This causes some network management tools (that work with other router OS's) to falsely report that link is okay.

The problem is that a Linux router does not respond the way other systems do. This is the router equivalent of "Strong ES" model, it is not the same as "Strong ES" as defined in Host Requirements.

Quoted 5 lines

The new parameter adds an additional check on slow input packet path, and causes route cache flush if enabled and carrier is lost.

There is nothing "router" about this situation.

When 10.2.1.1 is being pinged, it is in the role of an end-system in that transaction.

The "router" is reachable by "A", and as a consequence so is that IP address 10.2.1.1, and therefore the ping should succeed.

#2 lore raw
Quoted 27 lines · On Wed, 18 Mar 2009 18:21:15 -0700 (PDT) David Miller <[email protected]>…

From: Stephen Hemminger <[email protected]>
Date: Fri, 13 Mar 2009 16:12:53 -0700

Quoted 18 lines

Add a new parameter that controls how kernel responds to packets when interface is down. This is done to solve the problem of:

Assume topology of:

      A <----------->       Router             X--- down link

10.1.1.2/24 10.1.1.1/24 10.2.1.1/24
eth0 eth1

If A pings 10.2.1.1 then with normal Linux semantics Router would respond even if eth1 link on 10.2.1.1 was down. This causes some network management tools (that work with other router OS's) to falsely report that link is okay.

The problem is that a Linux router does not respond the way other systems do. This is the router equivalent of "Strong ES" model, it is not the same as "Strong ES" as defined in Host Requirements.

Quoted 5 lines

The new parameter adds an additional check on slow input packet path, and causes route cache flush if enabled and carrier is lost.

Quoted 7 lines

There is nothing "router" about this situation.

When 10.2.1.1 is being pinged, it is in the role of an end-system in that transaction.

The "router" is reachable by "A", and as a consequence so is that IP address 10.2.1.1, and therefore the ping should succeed.

Unfortunately, network management tools expect routers to behave this
way: WWCD

maintainer
#3 lore raw

From: Stephen Hemminger <[email protected]>
Date: Wed, 18 Mar 2009 19:42:39 -0700

Quoted 2 lines

Unfortunately, network management tools expect routers to behave this
way: WWCD

I understand your concern but I would never approach the specific problem you stated that way.

If I want to know that the link to the next hop of the router is down, I'd ping the next hop not that router's interface IP address. That is pretty much the stupidest thing I've ever heard of.

Or, if I specifically wanted to diagnose connectivity to "B" I'd use traceroute and/or work my way back with pings, hop by hop.

People expect Linux to do a lot of things the way other systems do, so what? If we have a reasonable way to approach solving a particular problem, and in this case we certainly do, we gain nothing by adding the knob besides placating a robot unwilling to learn new things.

Sorry, I'm not applying this.