Skip to contentSkip to conversation

net: multipath routing: configurable seed

PATCH Superseded Balaev Pavel 4 participants 7 messages Apr 13, 2021 – Apr 15, 2021

Subsystems (4)

v3 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 net: multipath routing: configurable seed

Ability for a user to assign seed value to multipath route hashes. Now kernel uses random seed value to prevent hash-flooding DoS attacks; however, it disables some use cases, f.e:

+-------+        +------+        +--------+
|       |-eth0---| FW0  |---eth0-|        |
|       |        +------+        |        |
|  GW0  |ECMP                ECMP|  GW1   |
|       |        +------+        |        |
|       |-eth1---| FW1  |---eth1-|        |
+-------+        +------+        +--------+

In this use case, two ECMP routers balance traffic between
two firewalls. If some flow transmits a response over a different channel than request, such flow will be dropped, because keep-state rules are created on
the other firewall.

This patch adds sysctl variable: net.ipv4.fib_multipath_hash_seed. User can set the same seed value on GW0 and GW1 for traffic to be mirror-balanced. By default, random value is used.

---
 Documentation/networking/ip-sysctl.rst |  14 ++++
 include/net/flow_dissector.h           |   4 +
 include/net/netns/ipv4.h               |  20 +++++
 net/core/flow_dissector.c              |   9 +++
 net/ipv4/af_inet.c                     |   5 ++
 net/ipv4/route.c                       |  10 ++-
 net/ipv4/sysctl_net_ipv4.c             | 104 +++++++++++++++++++++++++
 7 files changed, 165 insertions(+), 1 deletion(-)
Documentation/networking/ip-sysctl.rst +14 -0
index 9701906f6..d1a67e6fe 100644
@@ -100,6 +100,20 @@ fib_multipath_hash_policy - INTEGER
 	- 1 - Layer 4
 	- 2 - Layer 3 or inner Layer 3 if present
 
+fib_multipath_hash_seed - STRING
+	Controls seed value for multipath route hashes. By default
+	random value is used. Only valid for kernels built with
+	CONFIG_IP_ROUTE_MULTIPATH enabled.
+
+	Valid format: two hex values set off with comma or "random"
+	keyword.
+
+	Example to generate the seed value::
+
+		RAND=$(openssl rand -hex 16) && echo "${RAND:0:16},${RAND:16:16}"
+
+	Default: "random"
+
 fib_sync_mem - UNSIGNED INTEGER
 	Amount of dirty memory from fib entries that can be backlogged before
 	synchronize_rcu is forced.
include/net/flow_dissector.h +4 -0
index ffd386ea0..2bd4e28de 100644
@@ -348,6 +348,10 @@ static inline bool flow_keys_have_l4(const struct flow_keys *keys)
 }
 
 u32 flow_hash_from_keys(struct flow_keys *keys);
+#ifdef CONFIG_IP_ROUTE_MULTIPATH
+u32 flow_multipath_hash_from_keys(struct flow_keys *keys,
+			   const siphash_key_t *seed);
+#endif /* CONFIG_IP_ROUTE_MULTIPATH */
 void skb_flow_get_icmp_tci(const struct sk_buff *skb,
 			   struct flow_dissector_key_icmp *key_icmp,
 			   const void *data, int thoff, int hlen);
include/net/netns/ipv4.h +20 -0
index 87e161249..70a01817e 100644
@@ -11,6 +11,14 @@
 #include <linux/rcupdate.h>
 #include <linux/siphash.h>
 
+#ifdef CONFIG_IP_ROUTE_MULTIPATH
+/* Multipath seed key context */
+struct multipath_seed_ctx {
+	siphash_key_t	seed;
+	struct rcu_head	rcu;
+};
+#endif
+
 struct ctl_table_header;
 struct ipv4_devconf;
 struct fib_rules_ops;
@@ -222,6 +230,9 @@ struct netns_ipv4 {
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 	u8 sysctl_fib_multipath_use_neigh;
 	u8 sysctl_fib_multipath_hash_policy;
+	int sysctl_fib_multipath_hash_seed;
+	struct multipath_seed_ctx __rcu *fib_multipath_hash_seed_ctx;
+	spinlock_t fib_multipath_hash_seed_ctx_lock;
 #endif
 
 	struct fib_notifier_ops	*notifier_ops;
@@ -233,4 +244,13 @@ struct netns_ipv4 {
 	atomic_t	rt_genid;
 	siphash_key_t	ip_id_key;
 };
+
+#ifdef CONFIG_IP_ROUTE_MULTIPATH
+/* Caller needs to wrap with rcu_read_(un)lock() */
+static inline
+struct multipath_seed_ctx *fib_multipath_seed_get_ctx(const struct netns_ipv4 *ipv4)
+{
+	return rcu_dereference(ipv4->fib_multipath_hash_seed_ctx);
+}
+#endif
 #endif
net/core/flow_dissector.c +9 -0
index 5985029e4..457dfdab8 100644
@@ -1560,6 +1560,15 @@ u32 flow_hash_from_keys(struct flow_keys *keys)
 }
 EXPORT_SYMBOL(flow_hash_from_keys);
 
+#ifdef CONFIG_IP_ROUTE_MULTIPATH
+u32 flow_multipath_hash_from_keys(struct flow_keys *keys,
+					const siphash_key_t *seed)
+{
+	return __flow_hash_from_keys(keys, seed);
+}
+EXPORT_SYMBOL(flow_multipath_hash_from_keys);
+#endif /* CONFIG_IP_ROUTE_MULTIPATH */
+
 static inline u32 ___skb_get_hash(const struct sk_buff *skb,
 				  struct flow_keys *keys,
 				  const siphash_key_t *keyval)
net/ipv4/af_inet.c +5 -0
index f17870ee5..fd4c68966 100644
@@ -1874,6 +1874,11 @@ static __net_init int inet_init_net(struct net *net)
 
 	net->ipv4.sysctl_fib_notify_on_flag_change = 0;
 
+#ifdef CONFIG_IP_ROUTE_MULTIPATH
+	net->ipv4.fib_multipath_hash_seed_ctx = NULL;
+	spin_lock_init(&net->ipv4.fib_multipath_hash_seed_ctx_lock);
+#endif
+
 	return 0;
 }
 
net/ipv4/route.c +9 -1
index f6787c55f..36f72738d 100644
@@ -1911,6 +1911,7 @@ int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
 		       const struct sk_buff *skb, struct flow_keys *flkeys)
 {
 	u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0;
+	struct multipath_seed_ctx *seed_ctx;
 	struct flow_keys hash_keys;
 	u32 mhash;
 
@@ -1989,7 +1990,14 @@ int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
 		}
 		break;
 	}
-	mhash = flow_hash_from_keys(&hash_keys);
+
+	rcu_read_lock();
+	seed_ctx = fib_multipath_seed_get_ctx(&net->ipv4);
+	if (seed_ctx)
+		mhash = flow_multipath_hash_from_keys(&hash_keys, &seed_ctx->seed);
+	else
+		mhash = flow_hash_from_keys(&hash_keys);
+	rcu_read_unlock();
 
 	if (multipath_hash)
 		mhash = jhash_2words(mhash, multipath_hash, 0);
net/ipv4/sysctl_net_ipv4.c +104 -0

122 lines not on this page · raw message

Signature · 1 line
2.31.1
maintainer
#1 lore raw

From: Balaev Pavel <[email protected]>
Date: Tue, 13 Apr 2021 14:55:04 +0300

include/net/netns/ipv4.h struct netns_ipv4 { · line 235

Quoted 7 lines
include/net/netns/ipv4.h +3 -0
@@ -222,6 +230,9 @@ struct netns_ipv4 {
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 	u8 sysctl_fib_multipath_use_neigh;
 	u8 sysctl_fib_multipath_hash_policy;
+	int sysctl_fib_multipath_hash_seed;
+	struct multipath_seed_ctx __rcu *fib_multipath_hash_seed_ctx;
+	spinlock_t fib_multipath_hash_seed_ctx_lock;

Maybe use the rtnl mutex instead of this custom spinlock?

Thanks.

#2 lore raw
Quoted 13 lines · David Miller

From: Balaev Pavel <[email protected]>
Date: Tue, 13 Apr 2021 14:55:04 +0300

include/net/netns/ipv4.h struct netns_ipv4 { · line 235

Quoted 7 lines
include/net/netns/ipv4.h +3 -0
@@ -222,6 +230,9 @@ struct netns_ipv4 {
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 	u8 sysctl_fib_multipath_use_neigh;
 	u8 sysctl_fib_multipath_hash_policy;
+	int sysctl_fib_multipath_hash_seed;
+	struct multipath_seed_ctx __rcu *fib_multipath_hash_seed_ctx;
+	spinlock_t fib_multipath_hash_seed_ctx_lock;

Maybe use the rtnl mutex instead of this custom spinlock?

Quoted 1 line

Thanks.

Thanks for advise, I will use it in next patch version.

#3 lore raw
In reply to Balaev Pavel
Quoted 22 lines · Balaev Pavel

Ability for a user to assign seed value to multipath route hashes. Now kernel uses random seed value to prevent hash-flooding DoS attacks; however, it disables some use cases, f.e:

+-------+        +------+        +--------+
|       |-eth0---| FW0  |---eth0-|        |
|       |        +------+        |        |
|  GW0  |ECMP                ECMP|  GW1   |
|       |        +------+        |        |
|       |-eth1---| FW1  |---eth1-|        |
+-------+        +------+        +--------+

In this use case, two ECMP routers balance traffic between
two firewalls. If some flow transmits a response over a different channel than request, such flow will be dropped, because keep-state rules are created on
the other firewall.

This patch adds sysctl variable: net.ipv4.fib_multipath_hash_seed. User can set the same seed value on GW0 and GW1 for traffic to be mirror-balanced. By default, random value is used.

Quoted 10 lines
---
 Documentation/networking/ip-sysctl.rst |  14 ++++
 include/net/flow_dissector.h           |   4 +
 include/net/netns/ipv4.h               |  20 +++++
 net/core/flow_dissector.c              |   9 +++
 net/ipv4/af_inet.c                     |   5 ++
 net/ipv4/route.c                       |  10 ++-
 net/ipv4/sysctl_net_ipv4.c             | 104 +++++++++++++++++++++++++
 7 files changed, 165 insertions(+), 1 deletion(-)

This should work the same for IPv6.

And please add test cases under tools/testing/selftests/net.

#4 lore raw
Quoted 34 lines · David Ahern
Quoted 22 lines · Balaev Pavel

Ability for a user to assign seed value to multipath route hashes. Now kernel uses random seed value to prevent hash-flooding DoS attacks; however, it disables some use cases, f.e:

+-------+        +------+        +--------+
|       |-eth0---| FW0  |---eth0-|        |
|       |        +------+        |        |
|  GW0  |ECMP                ECMP|  GW1   |
|       |        +------+        |        |
|       |-eth1---| FW1  |---eth1-|        |
+-------+        +------+        +--------+

In this use case, two ECMP routers balance traffic between
two firewalls. If some flow transmits a response over a different channel than request, such flow will be dropped, because keep-state rules are created on
the other firewall.

This patch adds sysctl variable: net.ipv4.fib_multipath_hash_seed. User can set the same seed value on GW0 and GW1 for traffic to be mirror-balanced. By default, random value is used.

Quoted 10 lines
---
 Documentation/networking/ip-sysctl.rst |  14 ++++
 include/net/flow_dissector.h           |   4 +
 include/net/netns/ipv4.h               |  20 +++++
 net/core/flow_dissector.c              |   9 +++
 net/ipv4/af_inet.c                     |   5 ++
 net/ipv4/route.c                       |  10 ++-
 net/ipv4/sysctl_net_ipv4.c             | 104 +++++++++++++++++++++++++
 7 files changed, 165 insertions(+), 1 deletion(-)
Quoted 1 line

This should work the same for IPv6.

I wanted to add IPv6 support after IPv4 will be approved, anyway no problem, will add IPv6 in next version

Quoted 1 line

And please add test cases under tools/testing/selftests/net.

This feature cannot be tested whithin one host instance, becasue the same seed will be used by default for all netns, so results will be the same anyway, should I use QEMU for this tests?

#5 lore raw
Quoted 10 lines · Pavel Balaev
Quoted 2 lines

This should work the same for IPv6.

I wanted to add IPv6 support after IPv4 will be approved, anyway no problem, will add IPv6 in next version

Quoted 1 line

And please add test cases under tools/testing/selftests/net.

This feature cannot be tested whithin one host instance, becasue the same seed will be used by default for all netns, so results will be the same anyway, should I use QEMU for this tests?

why not make the seed per namespace?

#6 lore raw
Quoted 5 lines · David Ahern
Quoted 4 lines · Pavel Balaev
Quoted 2 lines

This should work the same for IPv6.

I wanted to add IPv6 support after IPv4 will be approved, anyway no problem, will add IPv6 in next version

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

And please add test cases under tools/testing/selftests/net.

This feature cannot be tested whithin one host instance, becasue the same seed will be used by default for all netns, so results will be the same anyway, should I use QEMU for this tests?

why not make the seed per namespace?

In patch seed is maked per namespace. I mean that I cannot check default behaviour whitin one host: sysctl net.ipv4.fib_multipath_hash_seed=random. In this case system random seed will be used (same for all netns, as it was before my patch).

We can only test two cases:

netns0: net.ipv4.fib_multipath_hash_seed=${SEED_0}
netns1: net.ipv4.fib_multipath_hash_seed=${SEED_1}

flows direction will not be the same.

netns0: net.ipv4.fib_multipath_hash_seed=${SEED_0}
netns1: net.ipv4.fib_multipath_hash_seed=${SEED_0}

flows direction will be the same.

Is this enough?