Skip to contentSkip to conversation

FAILED: patch "[PATCH] netfilter: cttimeout: prevent UAF during module unload" failed to apply to 6.18-stable tree

PATCH New [email protected] 2 participants 3 messages Sep 16

Subsystems (3)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

The patch below does not apply to the 6.18-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

To reproduce the conflict and resubmit, you may use the following commands:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.18.y
git checkout FETCH_HEAD
git cherry-pick -x fec9b1de0d02de8dafa3cc344bcb91cf28660643
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2026091639-unmade-pouch-5365@gregkh' --subject-prefix 'PATCH 6.18.y' 'HEAD^..'

Possible dependencies:

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From fec9b1de0d02de8dafa3cc344bcb91cf28660643 Mon Sep 17 00:00:00 2001
From: Chengfeng Ye <[email protected]>
Date: Mon, 24 Aug 2026 20:12:38 +0800
Subject: [PATCH] netfilter: cttimeout: prevent UAF during module unload

nf_ct_set_timeout() protects the timeout hook dereference and policy lookup with rcu_read_lock(). cttimeout_exit(), however, unregisters the per-net operations before it clears the hook.

This allows the following interleaving:

CPU 0 CPU 1
cttimeout_exit() nf_ct_set_timeout()
unregister_pernet_subsys() rcu_read_lock()
kfree(pernet) h = nf_ct_timeout_hook
h->timeout_find_get()
nfct_timeout_pernet()

The hook still points to ctnl_timeout_find_get() when CPU 1 looks up the already freed per-net timeout list. KASAN reported:

  BUG: KASAN: slab-use-after-free in ctnl_timeout_find_get
  Read of size 8 by task poc/90
  Call Trace:
   ctnl_timeout_find_get+0x271/0x2a0 [nfnetlink_cttimeout]
   nf_ct_set_timeout+0x7b/0x3c0
   xt_ct_tg_check+0x724/0xb20
   xt_check_target+0x234/0xa90
   do_ipt_set_ctl+0x570/0x1270
  Allocated by task 89:
   __kmalloc_noprof+0x16e/0x460
   ops_init+0x6d/0x420
   register_pernet_operations+0x2f6/0x670
  Freed by task 91:
   kfree+0x131/0x390
   ops_undo_list+0x3d4/0x730
   unregister_pernet_operations+0x232/0x490
   unregister_pernet_subsys+0x1c/0x30
   cttimeout_exit+0x52/0x970 [nfnetlink_cttimeout]

Clear the hook and wait for existing readers before unregistering the per-net operations. This blocks new policy lookups and ensures readers that observed the hook finish before the per-net storage is freed.

Fixes ebfbe67568a7 ("netfilter: cttimeout: use net_generic infra")
net/netfilter/nfnetlink_cttimeout.c +1 -1
index 66c2016f6049..132c02ac7c4e 100644
@@ -652,9 +652,9 @@ static void __exit cttimeout_exit(void)
 {
 	nfnetlink_subsys_unregister(&cttimeout_subsys);
 
-	unregister_pernet_subsys(&cttimeout_ops);
 	RCU_INIT_POINTER(nf_ct_timeout_hook, NULL);
 	synchronize_net();
+	unregister_pernet_subsys(&cttimeout_ops);
 }
 
 module_init(cttimeout_init);
maintainer
#1 lore raw

PATCH 1/2 netfilter: cttimeout: detach dataplane timeout policy and repurpose refcount

From: Pablo Neira Ayuso <[email protected]>

[ Upstream commit 7d6a9cdb8d3a51d9cfe546a09a518ab3d2671549 ]

Add a refcount for struct nf_ct_timeout which is used by ct extension to set the custom ct timeout policy, this tells us that the ct timeout is being used by a conntrack entry. When the last conntrack entry drops the refcount on the ct timeout, the ct timeout is released.

Remove the refcount for control plane which controls if the ruleset refers to the timeout policy. After this update, it is possible to remove the ct timeout policy from nfnetlink_cttimeout immediately. This is for simplicity not to handle two refcounts on a single object.

Remove nf_queue_nf_hook_drop(): a packet sitting in nfqueue will just hold a reference to the nf_ct_timeout object until packet is reinjected, since this is part of the ct extension, this will be released by the time the conntrack is freed.

nf_ct_untimeout() is still called to clean up in a best effort basis:
the ct timeout on existing entries gets removed when the ct timeout goes away, but as long as the iptables ruleset still refers to the ct timeout through a template, new conntracks may keep attaching it and extend its lifetime until the rule is removed.

nf_ct_untimeout() is not called anymore from module removal path, this is unlikely to find timeouts give module refcount is bumped, and the new refcount already tracks the ct timeout policy use so it is released when unused.

Fixes 50978462300f ("netfilter: add cttimeout infrastructure for fine timeout tuning")
Fixes 7e0b2b57f01d ("netfilter: nft_ct: add ct timeout support")

Stable-dep-of: fec9b1de0d02 ("netfilter: cttimeout: prevent UAF during module unload")

---
 include/net/netfilter/nf_conntrack_timeout.h |  27 ++++-
 net/netfilter/nf_conntrack_core.c            |   6 +-
 net/netfilter/nf_conntrack_timeout.c         |  27 ++++-
 net/netfilter/nfnetlink_cttimeout.c          | 112 +++++++++----------
 net/netfilter/nft_ct.c                       |   7 +-
 net/netfilter/xt_CT.c                        |   2 +-
 6 files changed, 107 insertions(+), 74 deletions(-)
include/net/netfilter/nf_conntrack_timeout.h +25 -2
index 3a66d4abb6d68..d60aa86be0194 100644
@@ -12,6 +12,7 @@
 #define CTNL_TIMEOUT_NAME_MAX	32
 
 struct nf_ct_timeout {
+	refcount_t		refcnt;
 	__u16			l3num;
 	const struct nf_conntrack_l4proto *l4proto;
 	struct rcu_head		rcu;
@@ -22,6 +23,22 @@ struct nf_conn_timeout {
 	struct nf_ct_timeout __rcu *timeout;
 };
 
+static inline void nf_ct_timeout_put(const struct nf_conn *ct)
+{
+#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
+	struct nf_conn_timeout *timeout_ext;
+	struct nf_ct_timeout *timeout;
+
+	timeout_ext = nf_ct_ext_find(ct, NF_CT_EXT_TIMEOUT);
+	if (!timeout_ext)
+		return;
+
+	timeout = rcu_dereference(timeout_ext->timeout);
+	if (timeout && refcount_dec_and_test(&timeout->refcnt))
+		kfree_rcu(timeout, rcu);
+#endif
+}
+
 static inline unsigned int *
 nf_ct_timeout_data(const struct nf_conn_timeout *t)
 {
@@ -56,8 +73,14 @@ struct nf_conn_timeout *nf_ct_timeout_ext_add(struct nf_conn *ct,
 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
 	struct nf_conn_timeout *timeout_ext;
 
+	if (!timeout)
+		return NULL;
+
 	timeout_ext = nf_ct_ext_add(ct, NF_CT_EXT_TIMEOUT, gfp);
-	if (timeout_ext == NULL)
+	if (!timeout_ext || timeout_ext->timeout)
+		return NULL;
+
+	if (!refcount_inc_not_zero(&timeout->refcnt))
 		return NULL;
 
 	rcu_assign_pointer(timeout_ext->timeout, timeout);
@@ -75,7 +98,7 @@ static inline unsigned int *nf_ct_timeout_lookup(const struct nf_conn *ct)
 	struct nf_conn_timeout *timeout_ext;
 
 	timeout_ext = nf_ct_timeout_find(ct);
-	if (timeout_ext)
+	if (timeout_ext && rcu_access_pointer(timeout_ext->timeout))
 		timeouts = nf_ct_timeout_data(timeout_ext);
 #endif
 	return timeouts;
net/netfilter/nf_conntrack_core.c +4 -2
index 44fa3224cc870..201d2ef8a6f02 100644
@@ -1690,16 +1690,18 @@ void nf_conntrack_free(struct nf_conn *ct)
 	 */
 	WARN_ON(refcount_read(&ct->ct_general.use) != 0);
 
+	rcu_read_lock();
 	if (ct->status & IPS_SRC_NAT_DONE) {
 		const struct nf_nat_hook *nat_hook;
 
-		rcu_read_lock();
 		nat_hook = rcu_dereference(nf_nat_hook);
 		if (nat_hook)
 			nat_hook->remove_nat_bysrc(ct);
-		rcu_read_unlock();
 	}
 
+	nf_ct_timeout_put(ct);
+	rcu_read_unlock();
+
 	kfree(ct->ext);
 	kmem_cache_free(nf_conntrack_cachep, ct);
 	cnet = nf_ct_pernet(net);
net/netfilter/nf_conntrack_timeout.c +24 -3
index 0cc584d3dbb1d..c81becde2afa9 100644
@@ -25,17 +25,32 @@
 const struct nf_ct_timeout_hooks __rcu *nf_ct_timeout_hook __read_mostly;
 EXPORT_SYMBOL_GPL(nf_ct_timeout_hook);
 
+/* nf_ct_iterate_cleanup() holds the conntrack lock. */
 static int untimeout(struct nf_conn *ct, void *timeout)
 {
 	struct nf_conn_timeout *timeout_ext = nf_ct_timeout_find(ct);
 
 	if (timeout_ext) {
-		const struct nf_ct_timeout *t;
+		struct nf_ct_timeout *t;
 
-		t = rcu_access_pointer(timeout_ext->timeout);
+		rcu_read_lock();
+		t = rcu_dereference(timeout_ext->timeout);
+		if (!t) {
+			rcu_read_unlock();
+			return 0;
+		}
 
-		if (!timeout || t == timeout)
+		if (!timeout || t == timeout) {
 			RCU_INIT_POINTER(timeout_ext->timeout, NULL);
+
+			/* No race with nf_conntrack_free() which is called
+			 * only after the conntrack has been removed from
+			 * the hashes.
+			 */
+			if (refcount_dec_and_test(&t->refcnt))
+				kfree_rcu(t, rcu);
+		}
+		rcu_read_unlock();
 	}
 
 	/* We are not intended to delete this conntrack. */
@@ -70,6 +85,8 @@ int nf_ct_set_timeout(struct net *net, struct nf_conn *ct,
 	const char *errmsg = NULL;
 	int ret = 0;
 
+	WARN_ON_ONCE(!nf_ct_is_template(ct));
+
 	rcu_read_lock();
 	h = rcu_dereference(nf_ct_timeout_hook);
 	if (!h) {
@@ -127,6 +144,8 @@ void nf_ct_destroy_timeout(struct nf_conn *ct)
 	struct nf_conn_timeout *timeout_ext;
 	const struct nf_ct_timeout_hooks *h;
 
+	WARN_ON_ONCE(!nf_ct_is_template(ct));
+
 	rcu_read_lock();
 	h = rcu_dereference(nf_ct_timeout_hook);
 
@@ -139,6 +158,8 @@ void nf_ct_destroy_timeout(struct nf_conn *ct)
 			if (t)
 				h->timeout_put(t);
 			RCU_INIT_POINTER(timeout_ext->timeout, NULL);
+			if (t && refcount_dec_and_test(&t->refcnt))
+				kfree_rcu(t, rcu);
 		}
 	}
 	rcu_read_unlock();
net/netfilter/nfnetlink_cttimeout.c +50 -62

219 lines not on this page · raw message

net/netfilter/nft_ct.c +3 -4
index b29ff555979b2..fa32bbb313065 100644
@@ -898,8 +898,6 @@ static void nft_ct_timeout_obj_eval(struct nft_object *obj,
 		}
 	}
 
-	rcu_assign_pointer(timeout->timeout, priv->timeout);
-
 	/* adjust the timeout as per 'new' state. ct is unconfirmed,
 	 * so the current timestamp must not be added.
 	 */
@@ -950,6 +948,7 @@ static int nft_ct_timeout_obj_init(const struct nft_ctx *ctx,
 
 	timeout->l3num = l3num;
 	timeout->l4proto = l4proto;
+	refcount_set(&timeout->refcnt, 1);
 
 	ret = nf_ct_netns_get(ctx->net, ctx->family);
 	if (ret < 0)
@@ -970,10 +969,10 @@ static void nft_ct_timeout_obj_destroy(const struct nft_ctx *ctx,
 	struct nft_ct_timeout_obj *priv = nft_obj_data(obj);
 	struct nf_ct_timeout *timeout = priv->timeout;
 
-	nf_queue_nf_hook_drop(ctx->net);
 	nf_ct_untimeout(ctx->net, timeout);
 	nf_ct_netns_put(ctx->net, ctx->family);
-	kfree_rcu(priv->timeout, rcu);
+	if (refcount_dec_and_test(&timeout->refcnt))
+		kfree_rcu(priv->timeout, rcu);
 }
 
 static int nft_ct_timeout_obj_dump(struct sk_buff *skb,
net/netfilter/xt_CT.c +1 -1
index 498f5871c84a0..90a295cf5ec6b 100644
@@ -284,7 +284,7 @@ static void xt_ct_tg_destroy(const struct xt_tgdtor_param *par,
 	struct nf_conn_help *help;
 
 	if (ct) {
-		if (info->helper[0] || info->timeout[0])
+		if (info->helper[0])
 			nf_queue_nf_hook_drop(par->net);
 
 		help = nfct_help(ct);
Signature · 1 line
2.53.0
maintainer
#2 lore raw

PATCH 2/2 netfilter: cttimeout: prevent UAF during module unload

From: Chengfeng Ye <[email protected]>

[ Upstream commit fec9b1de0d02de8dafa3cc344bcb91cf28660643 ]

nf_ct_set_timeout() protects the timeout hook dereference and policy lookup with rcu_read_lock(). cttimeout_exit(), however, unregisters the per-net operations before it clears the hook.

This allows the following interleaving:

CPU 0 CPU 1
cttimeout_exit() nf_ct_set_timeout()
unregister_pernet_subsys() rcu_read_lock()
kfree(pernet) h = nf_ct_timeout_hook
h->timeout_find_get()
nfct_timeout_pernet()

The hook still points to ctnl_timeout_find_get() when CPU 1 looks up the already freed per-net timeout list. KASAN reported:

  BUG: KASAN: slab-use-after-free in ctnl_timeout_find_get
  Read of size 8 by task poc/90
  Call Trace:
   ctnl_timeout_find_get+0x271/0x2a0 [nfnetlink_cttimeout]
   nf_ct_set_timeout+0x7b/0x3c0
   xt_ct_tg_check+0x724/0xb20
   xt_check_target+0x234/0xa90
   do_ipt_set_ctl+0x570/0x1270
  Allocated by task 89:
   __kmalloc_noprof+0x16e/0x460
   ops_init+0x6d/0x420
   register_pernet_operations+0x2f6/0x670
  Freed by task 91:
   kfree+0x131/0x390
   ops_undo_list+0x3d4/0x730
   unregister_pernet_operations+0x232/0x490
   unregister_pernet_subsys+0x1c/0x30
   cttimeout_exit+0x52/0x970 [nfnetlink_cttimeout]

Clear the hook and wait for existing readers before unregistering the per-net operations. This blocks new policy lookups and ensures readers that observed the hook finish before the per-net storage is freed.

Fixes ebfbe67568a7 ("netfilter: cttimeout: use net_generic infra")
---
 net/netfilter/nfnetlink_cttimeout.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/netfilter/nfnetlink_cttimeout.c +1 -1
index c87335d78dfa0..564e4642467dc 100644
@@ -654,9 +654,9 @@ static void __exit cttimeout_exit(void)
 {
 	nfnetlink_subsys_unregister(&cttimeout_subsys);
 
-	unregister_pernet_subsys(&cttimeout_ops);
 	RCU_INIT_POINTER(nf_ct_timeout_hook, NULL);
 	synchronize_net();
+	unregister_pernet_subsys(&cttimeout_ops);
 }
 
 module_init(cttimeout_init);
Signature · 1 line
2.53.0