Skip to contentSkip to conversation

netfilter: nft_reject: prevent unbounded recursion in output hooks

PATCH Superseded Fernando Fernandez Mancera 2 participants 3 messages Sep 21 – Sep 22

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 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.

Fixes bee11dc78fc8 ("netfilter: nft_reject: support for IPv6 and TCP reset")
---
 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 +
 6 files changed, 69 insertions(+), 35 deletions(-)
 create mode 100644 include/net/netfilter/nf_recursion.h
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
maintainer
#1 lore raw
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 4 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.

Won't nf_send_unreach() / unrach6 need the same guard?

If not, please explain why those are safe, its not clear to me.

Patch looks good otherwise, thanks Fernando.

#2 lore raw
Quoted 12 lines · Florian Westphal
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 8 lines
Quoted 4 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.

Won't nf_send_unreach() / unrach6 need the same guard?

If not, please explain why those are safe, its not clear to me.

Hi Florian,

If I am not wrong, those functions are safe. I initially thought it was needed too but after failing to reproduce the issue in that path I noticed that both icmp_send() and icmp6_send() are protected for recursions by disabling BH and using a xmit lock icmp_xmit_lock()/icmpv6_xmit_lock().

I will send a V2 mentioning this in the commit message so it is documented.

Quoted 3 lines

Patch looks good otherwise, thanks Fernando.

Thanks Florian!