Skip to contentSkip to conversation

netfilter: nft_reject: prevent unbounded recursion in output hooks

PATCH In review Fernando Fernandez Mancera 2 participants 5 messages Sep 24 – Nov 3

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 netfilter: nft_reject: prevent unbounded recursion in output hooks

nftables payload-mangling rules can rewrite packet contents (such as TCP flags) before they are evaluated by reject targets. This defeats the content-based loop guards, allowing unbounded hook recursion that leads to kernel stack exhaustion and panics.

Fix this by extracting the per-CPU xmit recursion limit previously used by the netdev duplication/forwarding paths into a shared header, and apply it to the inet and inet6 reject transmit functions. It disables BH while transmitting the skb to address a possible migration to different CPU leading to imbalanced decrementation of the recursion counters.

Note that the dev_queue_xmit() path used when CONFIG_BRIDGE_NETFILTER is enabled does not require this recursion guard. In that scenario, the generated reset packet is transmitted directly at Layer 2. Because the packet egresses from a different hook, it cannot synchronously loop. In addition nf_send_unreach() and nf_send_unreach6() are safe as the ICMP transmit functions are protected against recursion by the icmp_xmit_lock() and icmpv6_xmit_lock() guards.

Fixes bee11dc78fc8 ("netfilter: nft_reject: support for IPv6 and TCP reset")

---
v3: fixed build robot errors due to the struct member being guarded by CONFIG_NF_DUP_NETDEV instead of CONFIG_NETFILTER and renamed the struct member too.

---
 include/linux/netdevice_xmit.h        |  4 +--
 include/net/netfilter/nf_dup_netdev.h | 35 ----------------------
 include/net/netfilter/nf_recursion.h  | 42 +++++++++++++++++++++++++++
 net/ipv4/netfilter/nf_reject_ipv4.c   | 12 ++++++++
 net/ipv6/netfilter/nf_reject_ipv6.c   | 13 +++++++++
 net/netfilter/nf_dup_netdev.c         |  1 +
 net/netfilter/nft_fwd_netdev.c        |  1 +
 7 files changed, 71 insertions(+), 37 deletions(-)
 create mode 100644 include/net/netfilter/nf_recursion.h
include/linux/netdevice_xmit.h +2 -2
index cc232508e695..286ae5a6101e 100644
@@ -18,8 +18,8 @@ struct netdev_xmit {
 	u8			sched_mirred_nest;
 	struct net_device	*sched_mirred_dev[MIRRED_NEST_LIMIT];
 #endif
-#if IS_ENABLED(CONFIG_NF_DUP_NETDEV)
-	u8 nf_dup_skb_recursion;
+#if IS_ENABLED(CONFIG_NETFILTER)
+	u8 nf_xmit_skb_recursion;
 #endif
 };
 
include/net/netfilter/nf_dup_netdev.h +0 -35
index f6b05bd80c3f..a39411264614 100644
@@ -9,41 +9,6 @@
 void nf_dup_netdev_egress(const struct nft_pktinfo *pkt, int oif);
 void nf_fwd_netdev_egress(const struct nft_pktinfo *pkt, int oif);
 
-#define NF_RECURSION_LIMIT	2
-
-#ifndef CONFIG_PREEMPT_RT
-static inline bool nf_dev_xmit_recursion(void)
-{
-	return unlikely(__this_cpu_read(softnet_data.xmit.nf_dup_skb_recursion) >
-			NF_RECURSION_LIMIT);
-}
-
-static inline void nf_dev_xmit_recursion_inc(void)
-{
-	__this_cpu_inc(softnet_data.xmit.nf_dup_skb_recursion);
-}
-
-static inline void nf_dev_xmit_recursion_dec(void)
-{
-	__this_cpu_dec(softnet_data.xmit.nf_dup_skb_recursion);
-}
-#else
-static inline bool nf_dev_xmit_recursion(void)
-{
-	return unlikely(current->net_xmit.nf_dup_skb_recursion > NF_RECURSION_LIMIT);
-}
-
-static inline void nf_dev_xmit_recursion_inc(void)
-{
-	current->net_xmit.nf_dup_skb_recursion++;
-}
-
-static inline void nf_dev_xmit_recursion_dec(void)
-{
-	current->net_xmit.nf_dup_skb_recursion--;
-}
-#endif
-
 struct nft_offload_ctx;
 struct nft_flow_rule;
 
include/net/netfilter/nf_recursion.h new file +42 -0

42 lines not on this page · raw message

net/ipv4/netfilter/nf_reject_ipv4.c +12 -0
index 59ec465a9df9..3f227a094312 100644
@@ -10,6 +10,7 @@
 #include <net/dst.h>
 #include <net/dst_metadata.h>
 #include <net/netfilter/ipv4/nf_reject.h>
+#include <net/netfilter/nf_recursion.h>
 #include <linux/netfilter_ipv4.h>
 #include <linux/netfilter_bridge.h>
 
@@ -336,7 +337,18 @@ void nf_send_reset(struct net *net, struct sock *sk, struct sk_buff *oldskb,
 		dev_queue_xmit(nskb);
 	} else
 #endif
+	{
+		local_bh_disable();
+		if (nf_dev_xmit_recursion()) {
+			local_bh_enable();
+			goto free_nskb;
+		}
+
+		nf_dev_xmit_recursion_inc();
 		ip_local_out(net, nskb->sk, nskb);
+		nf_dev_xmit_recursion_dec();
+		local_bh_enable();
+	}
 
 	return;
 
net/ipv6/netfilter/nf_reject_ipv6.c +13 -0
index 07cdaa10da0d..f69ddfe72030 100644
@@ -10,6 +10,7 @@
 #include <net/ip6_checksum.h>
 #include <net/dst_metadata.h>
 #include <net/netfilter/ipv6/nf_reject.h>
+#include <net/netfilter/nf_recursion.h>
 #include <linux/netfilter_ipv6.h>
 #include <linux/netfilter_bridge.h>
 
@@ -409,7 +410,19 @@ void nf_send_reset6(struct net *net, struct sock *sk, struct sk_buff *oldskb,
 		dev_queue_xmit(nskb);
 	} else
 #endif
+	{
+		local_bh_disable();
+		if (nf_dev_xmit_recursion()) {
+			local_bh_enable();
+			kfree_skb(nskb);
+			return;
+		}
+
+		nf_dev_xmit_recursion_inc();
 		ip6_local_out(net, sk, nskb);
+		nf_dev_xmit_recursion_dec();
+		local_bh_enable();
+	}
 }
 EXPORT_SYMBOL_GPL(nf_send_reset6);
 
net/netfilter/nf_dup_netdev.c +1 -0
index c6bd5c29bed6..918716629c15 100644
@@ -12,6 +12,7 @@
 #include <net/netfilter/nf_tables.h>
 #include <net/netfilter/nf_tables_offload.h>
 #include <net/netfilter/nf_dup_netdev.h>
+#include <net/netfilter/nf_recursion.h>
 
 static void nf_do_netdev_egress(struct sk_buff *skb, struct net_device *dev,
 				enum nf_dev_hooks hook)
net/netfilter/nft_fwd_netdev.c +1 -0
index a48c2f765bba..7447da931a95 100644
@@ -14,6 +14,7 @@
 #include <net/netfilter/nf_tables.h>
 #include <net/netfilter/nf_tables_offload.h>
 #include <net/netfilter/nf_dup_netdev.h>
+#include <net/netfilter/nf_recursion.h>
 #include <net/neighbour.h>
 #include <net/ip.h>
 
Signature · 1 line
2.55.0
#1 lore raw
Quoted 19 lines · Fernando Fernandez Mancera

nftables payload-mangling rules can rewrite packet contents (such as TCP flags) before they are evaluated by reject targets. This defeats the content-based loop guards, allowing unbounded hook recursion that leads to kernel stack exhaustion and panics.

Fix this by extracting the per-CPU xmit recursion limit previously used by the netdev duplication/forwarding paths into a shared header, and apply it to the inet and inet6 reject transmit functions. It disables BH while transmitting the skb to address a possible migration to different CPU leading to imbalanced decrementation of the recursion counters.

Note that the dev_queue_xmit() path used when CONFIG_BRIDGE_NETFILTER is enabled does not require this recursion guard. In that scenario, the generated reset packet is transmitted directly at Layer 2. Because the packet egresses from a different hook, it cannot synchronously loop. In addition nf_send_unreach() and nf_send_unreach6() are safe as the ICMP transmit functions are protected against recursion by the icmp_xmit_lock() and icmpv6_xmit_lock() guards.

Quoted 8 lines
Fixes bee11dc78fc8 ("netfilter: nft_reject: support for IPv6 and TCP reset")

---
v3: fixed build robot errors due to the struct member being guarded by CONFIG_NF_DUP_NETDEV instead of CONFIG_NETFILTER and renamed the struct member too.

Is there something missing for this patch? It would be nice if we can get this in the next nf.git since it was discover by a real user (or it looks like so).

Quoted 207 lines
---
  include/linux/netdevice_xmit.h        |  4 +--
  include/net/netfilter/nf_dup_netdev.h | 35 ----------------------
  include/net/netfilter/nf_recursion.h  | 42 +++++++++++++++++++++++++++
  net/ipv4/netfilter/nf_reject_ipv4.c   | 12 ++++++++
  net/ipv6/netfilter/nf_reject_ipv6.c   | 13 +++++++++
  net/netfilter/nf_dup_netdev.c         |  1 +
  net/netfilter/nft_fwd_netdev.c        |  1 +
  7 files changed, 71 insertions(+), 37 deletions(-)

create mode 100644 include/net/netfilter/nf_recursion.h

include/linux/netdevice_xmit.h +2 -2
index cc232508e695..286ae5a6101e 100644
@@ -18,8 +18,8 @@ struct netdev_xmit {
  	u8			sched_mirred_nest;
  	struct net_device	*sched_mirred_dev[MIRRED_NEST_LIMIT];
  #endif
-#if IS_ENABLED(CONFIG_NF_DUP_NETDEV)
-	u8 nf_dup_skb_recursion;
+#if IS_ENABLED(CONFIG_NETFILTER)
+	u8 nf_xmit_skb_recursion;
  #endif
  };
 
include/net/netfilter/nf_dup_netdev.h +0 -35
index f6b05bd80c3f..a39411264614 100644
@@ -9,41 +9,6 @@
  void nf_dup_netdev_egress(const struct nft_pktinfo *pkt, int oif);
  void nf_fwd_netdev_egress(const struct nft_pktinfo *pkt, int oif);
 
-#define NF_RECURSION_LIMIT	2
-
-#ifndef CONFIG_PREEMPT_RT
-static inline bool nf_dev_xmit_recursion(void)
-{
-	return unlikely(__this_cpu_read(softnet_data.xmit.nf_dup_skb_recursion) >
-			NF_RECURSION_LIMIT);
-}
-
-static inline void nf_dev_xmit_recursion_inc(void)
-{
-	__this_cpu_inc(softnet_data.xmit.nf_dup_skb_recursion);
-}
-
-static inline void nf_dev_xmit_recursion_dec(void)
-{
-	__this_cpu_dec(softnet_data.xmit.nf_dup_skb_recursion);
-}
-#else
-static inline bool nf_dev_xmit_recursion(void)
-{
-	return unlikely(current->net_xmit.nf_dup_skb_recursion > NF_RECURSION_LIMIT);
-}
-
-static inline void nf_dev_xmit_recursion_inc(void)
-{
-	current->net_xmit.nf_dup_skb_recursion++;
-}
-
-static inline void nf_dev_xmit_recursion_dec(void)
-{
-	current->net_xmit.nf_dup_skb_recursion--;
-}
-#endif
-
  struct nft_offload_ctx;
  struct nft_flow_rule;
 
include/net/netfilter/nf_recursion.h new file +42 -0
new file mode 100644
index 000000000000..1ac65f7793f1
@@ -0,0 +1,42 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _NF_RECURSION_H_
+#define _NF_RECURSION_H_
+
+#include <linux/netdevice.h>
+
+#define NF_RECURSION_LIMIT	2
+
+#ifndef CONFIG_PREEMPT_RT
+static inline bool nf_dev_xmit_recursion(void)
+{
+	return unlikely(__this_cpu_read(softnet_data.xmit.nf_xmit_skb_recursion) >
+			NF_RECURSION_LIMIT);
+}
+
+static inline void nf_dev_xmit_recursion_inc(void)
+{
+	__this_cpu_inc(softnet_data.xmit.nf_xmit_skb_recursion);
+}
+
+static inline void nf_dev_xmit_recursion_dec(void)
+{
+	__this_cpu_dec(softnet_data.xmit.nf_xmit_skb_recursion);
+}
+#else
+static inline bool nf_dev_xmit_recursion(void)
+{
+	return unlikely(current->net_xmit.nf_xmit_skb_recursion > NF_RECURSION_LIMIT);
+}
+
+static inline void nf_dev_xmit_recursion_inc(void)
+{
+	current->net_xmit.nf_xmit_skb_recursion++;
+}
+
+static inline void nf_dev_xmit_recursion_dec(void)
+{
+	current->net_xmit.nf_xmit_skb_recursion--;
+}
+#endif
+
+#endif /* _NF_RECURSION_H_ */
net/ipv4/netfilter/nf_reject_ipv4.c +12 -0
index 59ec465a9df9..3f227a094312 100644
@@ -10,6 +10,7 @@
  #include <net/dst.h>
  #include <net/dst_metadata.h>
  #include <net/netfilter/ipv4/nf_reject.h>
+#include <net/netfilter/nf_recursion.h>
  #include <linux/netfilter_ipv4.h>
  #include <linux/netfilter_bridge.h>
 
@@ -336,7 +337,18 @@ void nf_send_reset(struct net *net, struct sock *sk, struct sk_buff *oldskb,
  		dev_queue_xmit(nskb);
  	} else
  #endif
+	{
+		local_bh_disable();
+		if (nf_dev_xmit_recursion()) {
+			local_bh_enable();
+			goto free_nskb;
+		}
+
+		nf_dev_xmit_recursion_inc();
  		ip_local_out(net, nskb->sk, nskb);
+		nf_dev_xmit_recursion_dec();
+		local_bh_enable();
+	}
 
  	return;
 
net/ipv6/netfilter/nf_reject_ipv6.c +13 -0
index 07cdaa10da0d..f69ddfe72030 100644
@@ -10,6 +10,7 @@
  #include <net/ip6_checksum.h>
  #include <net/dst_metadata.h>
  #include <net/netfilter/ipv6/nf_reject.h>
+#include <net/netfilter/nf_recursion.h>
  #include <linux/netfilter_ipv6.h>
  #include <linux/netfilter_bridge.h>
 
@@ -409,7 +410,19 @@ void nf_send_reset6(struct net *net, struct sock *sk, struct sk_buff *oldskb,
  		dev_queue_xmit(nskb);
  	} else
  #endif
+	{
+		local_bh_disable();
+		if (nf_dev_xmit_recursion()) {
+			local_bh_enable();
+			kfree_skb(nskb);
+			return;
+		}
+
+		nf_dev_xmit_recursion_inc();
  		ip6_local_out(net, sk, nskb);
+		nf_dev_xmit_recursion_dec();
+		local_bh_enable();
+	}
  }
  EXPORT_SYMBOL_GPL(nf_send_reset6);
 
net/netfilter/nf_dup_netdev.c +1 -0
index c6bd5c29bed6..918716629c15 100644
@@ -12,6 +12,7 @@
  #include <net/netfilter/nf_tables.h>
  #include <net/netfilter/nf_tables_offload.h>
  #include <net/netfilter/nf_dup_netdev.h>
+#include <net/netfilter/nf_recursion.h>
 
  static void nf_do_netdev_egress(struct sk_buff *skb, struct net_device *dev,
  				enum nf_dev_hooks hook)
net/netfilter/nft_fwd_netdev.c +1 -0
index a48c2f765bba..7447da931a95 100644
@@ -14,6 +14,7 @@
  #include <net/netfilter/nf_tables.h>
  #include <net/netfilter/nf_tables_offload.h>
  #include <net/netfilter/nf_dup_netdev.h>
+#include <net/netfilter/nf_recursion.h>
  #include <net/neighbour.h>
  #include <net/ip.h>
 
#2 lore raw
Quoted 29 lines · Fernando Fernandez Mancera
Quoted 19 lines · Fernando Fernandez Mancera

nftables payload-mangling rules can rewrite packet contents (such as TCP flags) before they are evaluated by reject targets. This defeats the content-based loop guards, allowing unbounded hook recursion that leads to kernel stack exhaustion and panics.

Fix this by extracting the per-CPU xmit recursion limit previously used by the netdev duplication/forwarding paths into a shared header, and apply it to the inet and inet6 reject transmit functions. It disables BH while transmitting the skb to address a possible migration to different CPU leading to imbalanced decrementation of the recursion counters.

Note that the dev_queue_xmit() path used when CONFIG_BRIDGE_NETFILTER is enabled does not require this recursion guard. In that scenario, the generated reset packet is transmitted directly at Layer 2. Because the packet egresses from a different hook, it cannot synchronously loop. In addition nf_send_unreach() and nf_send_unreach6() are safe as the ICMP transmit functions are protected against recursion by the icmp_xmit_lock() and icmpv6_xmit_lock() guards.

Quoted 8 lines
Fixes bee11dc78fc8 ("netfilter: nft_reject: support for IPv6 and TCP reset")

---
v3: fixed build robot errors due to the struct member being guarded by CONFIG_NF_DUP_NETDEV instead of CONFIG_NETFILTER and renamed the struct member too.

Quoted 3 lines

Is there something missing for this patch? It would be nice if we can get this in the next nf.git since it was discover by a real user (or it looks like so).

I will include this in the next nf.git batch.

#3 lore raw
In reply to Fernando Fernandez Mancera
Quoted 11 lines · Fernando Fernandez Mancera

nftables payload-mangling rules can rewrite packet contents (such as TCP flags) before they are evaluated by reject targets. This defeats the content-based loop guards, allowing unbounded hook recursion that leads to kernel stack exhaustion and panics.

Fix this by extracting the per-CPU xmit recursion limit previously used by the netdev duplication/forwarding paths into a shared header, and apply it to the inet and inet6 reject transmit functions. It disables BH while transmitting the skb to address a possible migration to different CPU leading to imbalanced decrementation of the recursion counters.

Quoted 7 lines

Note that the dev_queue_xmit() path used when CONFIG_BRIDGE_NETFILTER is enabled does not require this recursion guard. In that scenario, the generated reset packet is transmitted directly at Layer 2. Because the packet egresses from a different hook, it cannot synchronously loop. In addition nf_send_unreach() and nf_send_unreach6() are safe as the ICMP transmit functions are protected against recursion by the icmp_xmit_lock() and icmpv6_xmit_lock() guards.

Unrelated to this patch, but should ip_local_out() call in synproxy_send_tcp() be tracked by the recursion counter too?

#4 lore raw
Quoted 20 lines · Pablo Neira Ayuso
Quoted 11 lines · Fernando Fernandez Mancera

nftables payload-mangling rules can rewrite packet contents (such as TCP flags) before they are evaluated by reject targets. This defeats the content-based loop guards, allowing unbounded hook recursion that leads to kernel stack exhaustion and panics.

Fix this by extracting the per-CPU xmit recursion limit previously used by the netdev duplication/forwarding paths into a shared header, and apply it to the inet and inet6 reject transmit functions. It disables BH while transmitting the skb to address a possible migration to different CPU leading to imbalanced decrementation of the recursion counters.

Quoted 7 lines

Note that the dev_queue_xmit() path used when CONFIG_BRIDGE_NETFILTER is enabled does not require this recursion guard. In that scenario, the generated reset packet is transmitted directly at Layer 2. Because the packet egresses from a different hook, it cannot synchronously loop. In addition nf_send_unreach() and nf_send_unreach6() are safe as the ICMP transmit functions are protected against recursion by the icmp_xmit_lock() and icmpv6_xmit_lock() guards.

Quoted 2 lines

Unrelated to this patch, but should ip_local_out() call in synproxy_send_tcp() be tracked by the recursion counter too?

Hi Pablo, you are right. I have reproduced the same on synproxy. I am sending a patch to nf tree.

Thanks for pointing it out Pablo!