Skip to contentSkip to conversation

net/netfilter/ipvs: Fix data-race in ip_vs_add_service / ip_vs_out_hook

PATCH In review Zhang Tengfei 5 participants 12 messages Aug 26, 2025 – Sep 3, 2025

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/netfilter/ipvs: Fix data-race in ip_vs_add_service / ip_vs_out_hook

A data-race was detected by KCSAN between ip_vs_add_service() which acts as a writer, and ip_vs_out_hook() which acts as a reader. This can lead to unpredictable behavior and crashes. One observed symptom is the "no destination available" error when processing packets.

The race occurs on the enable flag within the netns_ipvs struct. This flag was being written in the configuration path without any protection, while concurrently being read in the packet processing path. This lack of synchronization means a reader on one CPU could see a partially initialized service, leading to incorrect behavior.

To fix this, convert the enable flag from a plain integer to an atomic_t. This ensures that all reads and writes to the flag are atomic. More importantly, using atomic_set() and atomic_read() provides the necessary memory barriers to guarantee that changes to other fields of the service are visible to the reader CPU before the service is marked as enabled.

---
 include/net/ip_vs.h             |  2 +-
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 10 +++++-----
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 5 files changed, 19 insertions(+), 19 deletions(-)
include/net/ip_vs.h +1 -1
index 29a36709e7f3..58b2ad7906e8 100644
@@ -895,7 +895,7 @@ struct ipvs_sync_daemon_cfg {
 /* IPVS in network namespace */
 struct netns_ipvs {
 	int			gen;		/* Generation */
-	int			enable;		/* enable like nf_hooks do */
+	atomic_t	enable;		/* enable like nf_hooks do */
 	/* Hash table: for real service lookups */
 	#define IP_VS_RTAB_BITS 4
 	#define IP_VS_RTAB_SIZE (1 << IP_VS_RTAB_BITS)
net/netfilter/ipvs/ip_vs_conn.c +2 -2
index 965f3c8e5089..5c97f85929b4 100644
@@ -885,7 +885,7 @@ static void ip_vs_conn_expire(struct timer_list *t)
 			 * conntrack cleanup for the net.
 			 */
 			smp_rmb();
-			if (ipvs->enable)
+			if (atomic_read(&ipvs->enable))
 				ip_vs_conn_drop_conntrack(cp);
 		}
 
@@ -1439,7 +1439,7 @@ void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
 		cond_resched_rcu();
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!atomic_read(&ipvs->enable))
 			break;
 	}
 	rcu_read_unlock();
net/netfilter/ipvs/ip_vs_core.c +5 -5
index c7a8a08b7308..84eed2e45927 100644
@@ -1353,7 +1353,7 @@ ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat
 	if (unlikely(!skb_dst(skb)))
 		return NF_ACCEPT;
 
-	if (!ipvs->enable)
+	if (!atomic_read(&ipvs->enable))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -1940,7 +1940,7 @@ ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state
 		return NF_ACCEPT;
 	}
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs) || !atomic_read(&ipvs->enable)))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -2108,7 +2108,7 @@ ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
 	int r;
 
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs) || !atomic_read(&ipvs->enable)))
 		return NF_ACCEPT;
 
 	if (state->pf == NFPROTO_IPV4) {
@@ -2295,7 +2295,7 @@ static int __net_init __ip_vs_init(struct net *net)
 		return -ENOMEM;
 
 	/* Hold the beast until a service is registered */
-	ipvs->enable = 0;
+	atomic_set(&ipvs->enable, 0);
 	ipvs->net = net;
 	/* Counters used for creating unique names */
 	ipvs->gen = atomic_read(&ipvs_netns_cnt);
@@ -2367,7 +2367,7 @@ static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
 		ipvs = net_ipvs(net);
 		ip_vs_unregister_hooks(ipvs, AF_INET);
 		ip_vs_unregister_hooks(ipvs, AF_INET6);
-		ipvs->enable = 0;	/* Disable packet reception */
+		atomic_set(&ipvs->enable, 0);	/* Disable packet reception */
 		smp_wmb();
 		ip_vs_sync_net_cleanup(ipvs);
 	}
net/netfilter/ipvs/ip_vs_ctl.c +3 -3
index 6a6fc4478533..ad7e1c387c1f 100644
@@ -256,7 +256,7 @@ static void est_reload_work_handler(struct work_struct *work)
 		struct ip_vs_est_kt_data *kd = ipvs->est_kt_arr[id];
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!atomic_read(&ipvs->enable))
 			goto unlock;
 		if (!kd)
 			continue;
@@ -1483,9 +1483,9 @@ ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 
 	*svc_p = svc;
 
-	if (!ipvs->enable) {
+	if (!atomic_read(&ipvs->enable)) {
 		/* Now there is a service - full throttle */
-		ipvs->enable = 1;
+		atomic_set(&ipvs->enable, 1);
 
 		/* Start estimation for first time */
 		ip_vs_est_reload_start(ipvs);
net/netfilter/ipvs/ip_vs_est.c +8 -8
index 15049b826732..c5aa2660de92 100644
@@ -231,7 +231,7 @@ static int ip_vs_estimation_kthread(void *data)
 void ip_vs_est_reload_start(struct netns_ipvs *ipvs)
 {
 	/* Ignore reloads before first service is added */
-	if (!ipvs->enable)
+	if (!atomic_read(&ipvs->enable))
 		return;
 	ip_vs_est_stopped_recalc(ipvs);
 	/* Bump the kthread configuration genid */
@@ -306,7 +306,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	int i;
 
 	if ((unsigned long)ipvs->est_kt_count >= ipvs->est_max_threads &&
-	    ipvs->enable && ipvs->est_max_threads)
+	    atomic_read(&ipvs->enable) && ipvs->est_max_threads)
 		return -EINVAL;
 
 	mutex_lock(&ipvs->est_mutex);
@@ -343,7 +343,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	}
 
 	/* Start kthread tasks only when services are present */
-	if (ipvs->enable && !ip_vs_est_stopped(ipvs)) {
+	if (atomic_read(&ipvs->enable) && !ip_vs_est_stopped(ipvs)) {
 		ret = ip_vs_est_kthread_start(ipvs, kd);
 		if (ret < 0)
 			goto out;
@@ -486,7 +486,7 @@ int ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 	struct ip_vs_estimator *est = &stats->est;
 	int ret;
 
-	if (!ipvs->est_max_threads && ipvs->enable)
+	if (!ipvs->est_max_threads && atomic_read(&ipvs->enable))
 		ipvs->est_max_threads = ip_vs_est_max_threads(ipvs);
 
 	est->ktid = -1;
@@ -663,7 +663,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 			/* Wait for cpufreq frequency transition */
 			wait_event_idle_timeout(wq, kthread_should_stop(),
 						HZ / 50);
-			if (!ipvs->enable || kthread_should_stop())
+			if (!atomic_read(&ipvs->enable) || kthread_should_stop())
 				goto stop;
 		}
 
@@ -681,7 +681,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 		rcu_read_unlock();
 		local_bh_enable();
 
-		if (!ipvs->enable || kthread_should_stop())
+		if (!atomic_read(&ipvs->enable) || kthread_should_stop())
 			goto stop;
 		cond_resched();
 
@@ -757,7 +757,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	mutex_lock(&ipvs->est_mutex);
 	for (id = 1; id < ipvs->est_kt_count; id++) {
 		/* netns clean up started, abort */
-		if (!ipvs->enable)
+		if (!atomic_read(&ipvs->enable))
 			goto unlock2;
 		kd = ipvs->est_kt_arr[id];
 		if (!kd)
@@ -787,7 +787,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	id = ipvs->est_kt_count;
 
 next_kt:
-	if (!ipvs->enable || kthread_should_stop())
+	if (!atomic_read(&ipvs->enable) || kthread_should_stop())
 		goto unlock;
 	id--;
 	if (id < 0)
Signature · 1 line
2.47.3
maintainer
#1 lore raw
Quoted 3 lines · Zhang Tengfei

A data-race was detected by KCSAN between ip_vs_add_service() which acts as a writer, and ip_vs_out_hook() which acts as a reader. This can lead to unpredictable behavior and crashes.

Really? How can this cause a crash?

Quoted 13 lines

The race occurs on the enable flag within the netns_ipvs struct. This flag was being written in the configuration path without any protection, while concurrently being read in the packet processing path. This lack of synchronization means a reader on one CPU could see a partially initialized service, leading to incorrect behavior.

To fix this, convert the enable flag from a plain integer to an atomic_t. This ensures that all reads and writes to the flag are atomic. More importantly, using atomic_set() and atomic_read() provides the necessary memory barriers to guarantee that changes to other fields of the service are visible to the reader CPU before the service is marked as enabled.

include/net/ip_vs.h struct ipvs_sync_daemon_cfg { · line 898

Quoted 2 lines
include/net/ip_vs.h +1 -1
-	int			enable;		/* enable like nf_hooks do */
+	atomic_t	enable;		/* enable like nf_hooks do */

Julian, Simon, I will defer to your judgment but I dislike this, because I see no reason for atomic_t. To me is seems better to use READ/WRITE_ONCE as ->enable is only ever set but not modified (increment for instance).

#2 lore raw
Quoted 7 lines · Florian Westphal
Quoted 3 lines · Zhang Tengfei

A data-race was detected by KCSAN between ip_vs_add_service() which acts as a writer, and ip_vs_out_hook() which acts as a reader. This can lead to unpredictable behavior and crashes.

Really? How can this cause a crash?

KCSAN + panic_on_warn=1 : Only in debug environment

Quoted 14 lines
Quoted 13 lines

The race occurs on the enable flag within the netns_ipvs struct. This flag was being written in the configuration path without any protection, while concurrently being read in the packet processing path. This lack of synchronization means a reader on one CPU could see a partially initialized service, leading to incorrect behavior.

To fix this, convert the enable flag from a plain integer to an atomic_t. This ensures that all reads and writes to the flag are atomic. More importantly, using atomic_set() and atomic_read() provides the necessary memory barriers to guarantee that changes to other fields of the service are visible to the reader CPU before the service is marked as enabled.

Quoted 7 lines

include/net/ip_vs.h struct ipvs_sync_daemon_cfg { · line 898

Quoted 2 lines
include/net/ip_vs.h +1 -1
-     int                     enable;         /* enable like nf_hooks do */
+     atomic_t        enable;         /* enable like nf_hooks do */

Julian, Simon, I will defer to your judgment but I dislike this, because I see no reason for atomic_t. To me is seems better to use READ/WRITE_ONCE as ->enable is only ever set but not modified (increment for instance).

+2

maintainer
#3 lore raw
In reply to Zhang Tengfei

Hello,

Quoted 4 lines · Zhang Tengfei

A data-race was detected by KCSAN between ip_vs_add_service() which acts as a writer, and ip_vs_out_hook() which acts as a reader. This can lead to unpredictable behavior and crashes. One observed symptom is the "no destination available" error when processing packets.

You can not solve the "no destination available" race
in IPVS itself. When service is added in ip_vs_add_service() things happen in some order:

- hooks are registered (if first service)
- service is registered
- enable flag is set (if first service)

All this is part of single IP_VS_SO_SET_ADD call.
You can reorder the above actions as you wish but without any gain because the dests (real servers) are added with a following IP_VS_SO_SET_ADDDEST call. There will be always a gap between the both calls where packets can hit service without any dests, with enable=0 or 1.

One can stop the traffic with firewall rules until
all IPVS rules are added. It is decided by user space tools when to route the traffic via IPVS (after or during configuration).

Quoted 6 lines

The race occurs on the enable flag within the netns_ipvs struct. This flag was being written in the configuration path without any protection, while concurrently being read in the packet processing path. This lack of synchronization means a reader on one CPU could see a partially initialized service, leading to incorrect behavior.

No, service is added with hlist_add_head_rcu() in
ip_vs_svc_hash() which commits all writes before adding svc to list. This is also an answer for the concerns in your paragraph below.

If "partially initialized" refers to the missing
dests, then see above, they are added later with IP_VS_SO_SET_ADDDEST.

Quoted 6 lines

To fix this, convert the enable flag from a plain integer to an atomic_t. This ensures that all reads and writes to the flag are atomic. More importantly, using atomic_set() and atomic_read() provides the necessary memory barriers to guarantee that changes to other fields of the service are visible to the reader CPU before the service is marked as enabled.

We use RCU for proper svc registration.

Quoted 29 lines
---
 include/net/ip_vs.h             |  2 +-
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 10 +++++-----
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 5 files changed, 19 insertions(+), 19 deletions(-)
include/net/ip_vs.h +1 -1
index 29a36709e7f3..58b2ad7906e8 100644
@@ -895,7 +895,7 @@ struct ipvs_sync_daemon_cfg {
 /* IPVS in network namespace */
 struct netns_ipvs {
 	int			gen;		/* Generation */
-	int			enable;		/* enable like nf_hooks do */
+	atomic_t	enable;		/* enable like nf_hooks do */
 	/* Hash table: for real service lookups */
 	#define IP_VS_RTAB_BITS 4
 	#define IP_VS_RTAB_SIZE (1 << IP_VS_RTAB_BITS)
net/netfilter/ipvs/ip_vs_conn.c +0 -0
index 965f3c8e5089..5c97f85929b4 100644

net/netfilter/ipvs/ip_vs_conn.c static void ip_vs_conn_expire(struct timer_list *t) · line 888

Quoted 5 lines
net/netfilter/ipvs/ip_vs_conn.c +1 -1
@@ -885,7 +885,7 @@ static void ip_vs_conn_expire(struct timer_list *t)
 			 * conntrack cleanup for the net.
 			 */
 			smp_rmb();
-			if (ipvs->enable)
+			if (atomic_read(&ipvs->enable))

This place is a valid user of this flag.
This is one of the reasons we should keep such flag.

Quoted 17 lines
net/netfilter/ipvs/ip_vs_conn.c +1 -1
 				ip_vs_conn_drop_conntrack(cp);
 		}
 
@@ -1439,7 +1439,7 @@ void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
 		cond_resched_rcu();
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!atomic_read(&ipvs->enable))
 			break;
 	}
 	rcu_read_unlock();
net/netfilter/ipvs/ip_vs_core.c +0 -0
index c7a8a08b7308..84eed2e45927 100644

net/netfilter/ipvs/ip_vs_core.c ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat · line 1356

Quoted 4 lines
net/netfilter/ipvs/ip_vs_core.c +0 -1
@@ -1353,7 +1353,7 @@ ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat
 	if (unlikely(!skb_dst(skb)))
 		return NF_ACCEPT;
 
-	if (!ipvs->enable)

The checks in the hooks are not needed anymore after
commit 857ca89711de ("ipvs: register hooks only with services") because the hooks are registered when the first service is added.

So, you can not see enable=0 in hooks anymore which
was possible before first service was added. Or it is possible for small time between adding the hooks and setting the flag but it is irrelevant because there are no dests yet.

Quoted 32 lines
net/netfilter/ipvs/ip_vs_core.c +4 -3
+	if (!atomic_read(&ipvs->enable))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -1940,7 +1940,7 @@ ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state
 		return NF_ACCEPT;
 	}
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs) || !atomic_read(&ipvs->enable)))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -2108,7 +2108,7 @@ ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
 	int r;
 
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs) || !atomic_read(&ipvs->enable)))
 		return NF_ACCEPT;
 
 	if (state->pf == NFPROTO_IPV4) {
@@ -2295,7 +2295,7 @@ static int __net_init __ip_vs_init(struct net *net)
 		return -ENOMEM;
 
 	/* Hold the beast until a service is registered */
-	ipvs->enable = 0;
+	atomic_set(&ipvs->enable, 0);
 	ipvs->net = net;
 	/* Counters used for creating unique names */
 	ipvs->gen = atomic_read(&ipvs_netns_cnt);

net/netfilter/ipvs/ip_vs_core.c static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list) · line 2369

Quoted 3 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
@@ -2367,7 +2367,7 @@ static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
 		ipvs = net_ipvs(net);
 		ip_vs_unregister_hooks(ipvs, AF_INET);
 		ip_vs_unregister_hooks(ipvs, AF_INET6);

No new/flying packets after this point...

net/netfilter/ipvs/ip_vs_core.c static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list) · line 2370

Quoted 1 line
net/netfilter/ipvs/ip_vs_core.c +0 -1
-		ipvs->enable = 0;	/* Disable packet reception */

It was an old method to disable reception but now
the flag helps for other purposes.

net/netfilter/ipvs/ip_vs_ctl.c static void est_reload_work_handler(struct work_struct *work) · line 259

Quoted 13 lines
net/netfilter/ipvs/ip_vs_core.c +1 -0
+		atomic_set(&ipvs->enable, 0);	/* Disable packet reception */
 		smp_wmb();
 		ip_vs_sync_net_cleanup(ipvs);
 	}
net/netfilter/ipvs/ip_vs_ctl.c +0 -1
index 6a6fc4478533..ad7e1c387c1f 100644
@@ -256,7 +256,7 @@ static void est_reload_work_handler(struct work_struct *work)
 		struct ip_vs_est_kt_data *kd = ipvs->est_kt_arr[id];
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)

Such checks in slow path just speedup the cleanup.

Quoted 85 lines
net/netfilter/ipvs/ip_vs_ctl.c +3 -2
+		if (!atomic_read(&ipvs->enable))
 			goto unlock;
 		if (!kd)
 			continue;
@@ -1483,9 +1483,9 @@ ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 
 	*svc_p = svc;
 
-	if (!ipvs->enable) {
+	if (!atomic_read(&ipvs->enable)) {
 		/* Now there is a service - full throttle */
-		ipvs->enable = 1;
+		atomic_set(&ipvs->enable, 1);
 
 		/* Start estimation for first time */
 		ip_vs_est_reload_start(ipvs);
net/netfilter/ipvs/ip_vs_est.c +7 -7
index 15049b826732..c5aa2660de92 100644
@@ -231,7 +231,7 @@ static int ip_vs_estimation_kthread(void *data)
 void ip_vs_est_reload_start(struct netns_ipvs *ipvs)
 {
 	/* Ignore reloads before first service is added */
-	if (!ipvs->enable)
+	if (!atomic_read(&ipvs->enable))
 		return;
 	ip_vs_est_stopped_recalc(ipvs);
 	/* Bump the kthread configuration genid */
@@ -306,7 +306,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	int i;
 
 	if ((unsigned long)ipvs->est_kt_count >= ipvs->est_max_threads &&
-	    ipvs->enable && ipvs->est_max_threads)
+	    atomic_read(&ipvs->enable) && ipvs->est_max_threads)
 		return -EINVAL;
 
 	mutex_lock(&ipvs->est_mutex);
@@ -343,7 +343,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	}
 
 	/* Start kthread tasks only when services are present */
-	if (ipvs->enable && !ip_vs_est_stopped(ipvs)) {
+	if (atomic_read(&ipvs->enable) && !ip_vs_est_stopped(ipvs)) {
 		ret = ip_vs_est_kthread_start(ipvs, kd);
 		if (ret < 0)
 			goto out;
@@ -486,7 +486,7 @@ int ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 	struct ip_vs_estimator *est = &stats->est;
 	int ret;
 
-	if (!ipvs->est_max_threads && ipvs->enable)
+	if (!ipvs->est_max_threads && atomic_read(&ipvs->enable))
 		ipvs->est_max_threads = ip_vs_est_max_threads(ipvs);
 
 	est->ktid = -1;
@@ -663,7 +663,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 			/* Wait for cpufreq frequency transition */
 			wait_event_idle_timeout(wq, kthread_should_stop(),
 						HZ / 50);
-			if (!ipvs->enable || kthread_should_stop())
+			if (!atomic_read(&ipvs->enable) || kthread_should_stop())
 				goto stop;
 		}
 
@@ -681,7 +681,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 		rcu_read_unlock();
 		local_bh_enable();
 
-		if (!ipvs->enable || kthread_should_stop())
+		if (!atomic_read(&ipvs->enable) || kthread_should_stop())
 			goto stop;
 		cond_resched();
 
@@ -757,7 +757,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	mutex_lock(&ipvs->est_mutex);
 	for (id = 1; id < ipvs->est_kt_count; id++) {
 		/* netns clean up started, abort */
-		if (!ipvs->enable)
+		if (!atomic_read(&ipvs->enable))
 			goto unlock2;
 		kd = ipvs->est_kt_arr[id];
 		if (!kd)

net/netfilter/ipvs/ip_vs_est.c static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs) · line 793

Quoted 8 lines
net/netfilter/ipvs/ip_vs_est.c +1 -1
@@ -787,7 +787,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	id = ipvs->est_kt_count;
 
 next_kt:
-	if (!ipvs->enable || kthread_should_stop())
+	if (!atomic_read(&ipvs->enable) || kthread_should_stop())
 		goto unlock;
 	id--;
 	if (id < 0)

In summary, the checks in fast path (in hooks) are
useless/obsolete while the other checks in slow path do not need atomic operations. Such races are normal to happen because service and its dests are not added in same atomic transaction.

Regards

--
Julian Anastasov <[email protected]>

maintainer
#4 lore raw
In reply to Zhang Tengfei

Hello,

Quoted 34 lines · Zhang Tengfei

A data-race was detected by KCSAN between ip_vs_add_service() which acts as a writer, and ip_vs_out_hook() which acts as a reader. This can lead to unpredictable behavior and crashes. One observed symptom is the "no destination available" error when processing packets.

The race occurs on the enable flag within the netns_ipvs struct. This flag was being written in the configuration path without any protection, while concurrently being read in the packet processing path. This lack of synchronization means a reader on one CPU could see a partially initialized service, leading to incorrect behavior.

To fix this, convert the enable flag from a plain integer to an atomic_t. This ensures that all reads and writes to the flag are atomic. More importantly, using atomic_set() and atomic_read() provides the necessary memory barriers to guarantee that changes to other fields of the service are visible to the reader CPU before the service is marked as enabled.

---
 include/net/ip_vs.h             |  2 +-
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 10 +++++-----
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 5 files changed, 19 insertions(+), 19 deletions(-)
net/netfilter/ipvs/ip_vs_est.c +0 -0
index 15049b826732..c5aa2660de92 100644

...

net/netfilter/ipvs/ip_vs_est.c static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs) · line 761

Quoted 7 lines
net/netfilter/ipvs/ip_vs_est.c +1 -1
@@ -757,7 +757,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	mutex_lock(&ipvs->est_mutex);
 	for (id = 1; id < ipvs->est_kt_count; id++) {
 		/* netns clean up started, abort */
-		if (!ipvs->enable)
+		if (!atomic_read(&ipvs->enable))
 			goto unlock2;

It is a simple flag but as it is checked in loops
in a few places in ip_vs_est.c, lets use READ_ONCE/WRITE_ONCE as suggested by Florian and Eric. The 3 checks in hooks in ip_vs_core.c can be simply removed: in ip_vs_out_hook, ip_vs_in_hook and ip_vs_forward_icmp. We can see enable=0 in rare cases which is not fatal. It is a flying packet in two possible cases:

1. after hooks are registered but before the flag is set
2. after the hooks are unregistered on cleanup_net

Regards

--
Julian Anastasov <[email protected]>

#5 lore raw

Hi everyone,

Here is the v2 patch that incorporates the feedback.

Many thanks to Julian for his thorough review and for providing the detailed plan for this new version, and thanks to Florian and Eric for suggestions.

Subject: [PATCH v2] net/netfilter/ipvs: Use READ_ONCE/WRITE_ONCE for
ipvs->enable

KCSAN reported a data-race on the ipvs->enable flag, which is written in the control path and read concurrently from many other contexts.

Following a suggestion by Julian, this patch fixes the race by converting all accesses to use WRITE_ONCE()/READ_ONCE(). This lightweight approach ensures atomic access and acts as a compiler barrier, preventing unsafe optimizations where the flag is checked in loops (e.g., in ip_vs_est.c).

Additionally, the now-obsolete enable checks in the fast path hooks (ip_vs_in_hook, ip_vs_out_hook, ip_vs_forward_icmp) are removed. These are unnecessary since commit 857ca89711de ("ipvs: register hooks only with services").

---
v2:
- Switched from atomic_t to the suggested READ_ONCE()/WRITE_ONCE().
- Removed obsolete checks from the packet processing hooks.
- Polished commit message based on feedback.

---
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 11 ++++-------
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 4 files changed, 17 insertions(+), 20 deletions(-)
net/netfilter/ipvs/ip_vs_conn.c +2 -2
index 965f3c8e5..37ebb0cb6 100644
@@ -885,7 +885,7 @@ static void ip_vs_conn_expire(struct timer_list *t)
 			 * conntrack cleanup for the net.
 			 */
 			smp_rmb();
-			if (ipvs->enable)
+			if (READ_ONCE(ipvs->enable))
 				ip_vs_conn_drop_conntrack(cp);
 		}
 
@@ -1439,7 +1439,7 @@ void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
 		cond_resched_rcu();
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			break;
 	}
 	rcu_read_unlock();
net/netfilter/ipvs/ip_vs_core.c +4 -7
index c7a8a08b7..5ea7ab8bf 100644
@@ -1353,9 +1353,6 @@ ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat
 	if (unlikely(!skb_dst(skb)))
 		return NF_ACCEPT;
 
-	if (!ipvs->enable)
-		return NF_ACCEPT;
-
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
 #ifdef CONFIG_IP_VS_IPV6
 	if (af == AF_INET6) {
@@ -1940,7 +1937,7 @@ ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state
 		return NF_ACCEPT;
 	}
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -2108,7 +2105,7 @@ ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
 	int r;
 
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	if (state->pf == NFPROTO_IPV4) {
@@ -2295,7 +2292,7 @@ static int __net_init __ip_vs_init(struct net *net)
 		return -ENOMEM;
 
 	/* Hold the beast until a service is registered */
-	ipvs->enable = 0;
+	WRITE_ONCE(ipvs->enable, 0);
 	ipvs->net = net;
 	/* Counters used for creating unique names */
 	ipvs->gen = atomic_read(&ipvs_netns_cnt);
@@ -2367,7 +2364,7 @@ static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
 		ipvs = net_ipvs(net);
 		ip_vs_unregister_hooks(ipvs, AF_INET);
 		ip_vs_unregister_hooks(ipvs, AF_INET6);
-		ipvs->enable = 0;	/* Disable packet reception */
+		WRITE_ONCE(ipvs->enable, 0);	/* Disable packet reception */
 		smp_wmb();
 		ip_vs_sync_net_cleanup(ipvs);
 	}
net/netfilter/ipvs/ip_vs_ctl.c +3 -3
index 6a6fc4478..4c8fa22be 100644
@@ -256,7 +256,7 @@ static void est_reload_work_handler(struct work_struct *work)
 		struct ip_vs_est_kt_data *kd = ipvs->est_kt_arr[id];
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock;
 		if (!kd)
 			continue;
@@ -1483,9 +1483,9 @@ ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 
 	*svc_p = svc;
 
-	if (!ipvs->enable) {
+	if (!READ_ONCE(ipvs->enable)) {
 		/* Now there is a service - full throttle */
-		ipvs->enable = 1;
+		WRITE_ONCE(ipvs->enable, 1);
 
 		/* Start estimation for first time */
 		ip_vs_est_reload_start(ipvs);
net/netfilter/ipvs/ip_vs_est.c +8 -8
index 15049b826..93a925f1e 100644
@@ -231,7 +231,7 @@ static int ip_vs_estimation_kthread(void *data)
 void ip_vs_est_reload_start(struct netns_ipvs *ipvs)
 {
 	/* Ignore reloads before first service is added */
-	if (!ipvs->enable)
+	if (!READ_ONCE(ipvs->enable))
 		return;
 	ip_vs_est_stopped_recalc(ipvs);
 	/* Bump the kthread configuration genid */
@@ -306,7 +306,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	int i;
 
 	if ((unsigned long)ipvs->est_kt_count >= ipvs->est_max_threads &&
-	    ipvs->enable && ipvs->est_max_threads)
+	    READ_ONCE(ipvs->enable) && ipvs->est_max_threads)
 		return -EINVAL;
 
 	mutex_lock(&ipvs->est_mutex);
@@ -343,7 +343,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	}
 
 	/* Start kthread tasks only when services are present */
-	if (ipvs->enable && !ip_vs_est_stopped(ipvs)) {
+	if (READ_ONCE(ipvs->enable) && !ip_vs_est_stopped(ipvs)) {
 		ret = ip_vs_est_kthread_start(ipvs, kd);
 		if (ret < 0)
 			goto out;
@@ -486,7 +486,7 @@ int ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 	struct ip_vs_estimator *est = &stats->est;
 	int ret;
 
-	if (!ipvs->est_max_threads && ipvs->enable)
+	if (!ipvs->est_max_threads && READ_ONCE(ipvs->enable))
 		ipvs->est_max_threads = ip_vs_est_max_threads(ipvs);
 
 	est->ktid = -1;
@@ -663,7 +663,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 			/* Wait for cpufreq frequency transition */
 			wait_event_idle_timeout(wq, kthread_should_stop(),
 						HZ / 50);
-			if (!ipvs->enable || kthread_should_stop())
+			if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 				goto stop;
 		}
 
@@ -681,7 +681,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 		rcu_read_unlock();
 		local_bh_enable();
 
-		if (!ipvs->enable || kthread_should_stop())
+		if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 			goto stop;
 		cond_resched();
 
@@ -757,7 +757,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	mutex_lock(&ipvs->est_mutex);
 	for (id = 1; id < ipvs->est_kt_count; id++) {
 		/* netns clean up started, abort */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock2;
 		kd = ipvs->est_kt_arr[id];
 		if (!kd)
@@ -787,7 +787,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	id = ipvs->est_kt_count;
 
 next_kt:
-	if (!ipvs->enable || kthread_should_stop())
+	if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 		goto unlock;
 	id--;
 	if (id < 0)
Signature · 1 line
2.34.1
#6 lore raw

Re:

Quoted 3 lines · Zhang Tengfei

Hi everyone,

Here is the v2 patch that incorporates the feedback.

Patch without subject will not fly too far, I'm afraid you will have to resubmit. One more comment below.

Quoted 39 lines

Many thanks to Julian for his thorough review and for providing the detailed plan for this new version, and thanks to Florian and Eric for suggestions.

Subject: [PATCH v2] net/netfilter/ipvs: Use READ_ONCE/WRITE_ONCE for
ipvs->enable

KCSAN reported a data-race on the ipvs->enable flag, which is written in the control path and read concurrently from many other contexts.

Following a suggestion by Julian, this patch fixes the race by converting all accesses to use WRITE_ONCE()/READ_ONCE(). This lightweight approach ensures atomic access and acts as a compiler barrier, preventing unsafe optimizations where the flag is checked in loops (e.g., in ip_vs_est.c).

Additionally, the now-obsolete enable checks in the fast path hooks (ip_vs_in_hook, ip_vs_out_hook, ip_vs_forward_icmp) are removed. These are unnecessary since commit 857ca89711de ("ipvs: register hooks only with services").

---
v2:
- Switched from atomic_t to the suggested READ_ONCE()/WRITE_ONCE().
- Removed obsolete checks from the packet processing hooks.
- Polished commit message based on feedback.

---
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 11 ++++-------
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 4 files changed, 17 insertions(+), 20 deletions(-)

[...]

net/netfilter/ipvs/ip_vs_core.c ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat · line 1357

Quoted 10 lines
net/netfilter/ipvs/ip_vs_core.c +0 -2
index c7a8a08b7..5ea7ab8bf 100644
@@ -1353,9 +1353,6 @@ ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat
 	if (unlikely(!skb_dst(skb)))
 		return NF_ACCEPT;
 
-	if (!ipvs->enable)
-		return NF_ACCEPT;

Patch does say why is this going away? If you think this is not necessary, then make a separated patch and example why this is needed?

Thanks

Quoted 3 lines
net/netfilter/ipvs/ip_vs_core.c +0 -0
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
 #ifdef CONFIG_IP_VS_IPV6
 	if (af == AF_INET6) {
#7 lore raw

PATCH v2 1/1 net/netfilter/ipvs: Use READ_ONCE/WRITE_ONCE for ipvs->enable

In reply to Julian Anastasov

KCSAN reported a data-race on the ipvs->enable flag, which is written in the control path and read concurrently from many other contexts.

Following a suggestion by Julian, this patch fixes the race by converting all accesses to use WRITE_ONCE()/READ_ONCE(). This lightweight approach ensures atomic access and acts as a compiler barrier, preventing unsafe optimizations where the flag is checked in loops (e.g., in ip_vs_est.c).

Additionally, the enable checks in the fast-path hooks (ip_vs_in_hook, ip_vs_out_hook, ip_vs_forward_icmp) are removed. They are considered unnecessary because the enable=0 condition they check for can only occur in two rare and non-fatal
scenarios: 1) after hooks are registered but before the flag is set, and 2) after hooks are unregistered on cleanup_net. In the worst case, a single packet might be mishandled (e.g., dropped), which does not lead to a system crash or data corruption. Adding a check in the performance-critical fast-path to handle this harmless condition is not a worthwhile trade-off.

---
v2:
- Switched from atomic_t to the suggested READ_ONCE()/WRITE_ONCE().
- Removed obsolete checks from the packet processing hooks.
- Polished commit message based on feedback from maintainers.

---
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 11 ++++-------
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 4 files changed, 17 insertions(+), 20 deletions(-)
net/netfilter/ipvs/ip_vs_conn.c +2 -2
index 965f3c8e5..37ebb0cb6 100644
@@ -885,7 +885,7 @@ static void ip_vs_conn_expire(struct timer_list *t)
 			 * conntrack cleanup for the net.
 			 */
 			smp_rmb();
-			if (ipvs->enable)
+			if (READ_ONCE(ipvs->enable))
 				ip_vs_conn_drop_conntrack(cp);
 		}
 
@@ -1439,7 +1439,7 @@ void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
 		cond_resched_rcu();
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			break;
 	}
 	rcu_read_unlock();
net/netfilter/ipvs/ip_vs_core.c +4 -7
index c7a8a08b7..5ea7ab8bf 100644
@@ -1353,9 +1353,6 @@ ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat
 	if (unlikely(!skb_dst(skb)))
 		return NF_ACCEPT;
 
-	if (!ipvs->enable)
-		return NF_ACCEPT;
-
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
 #ifdef CONFIG_IP_VS_IPV6
 	if (af == AF_INET6) {
@@ -1940,7 +1937,7 @@ ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state
 		return NF_ACCEPT;
 	}
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -2108,7 +2105,7 @@ ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
 	int r;
 
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	if (state->pf == NFPROTO_IPV4) {
@@ -2295,7 +2292,7 @@ static int __net_init __ip_vs_init(struct net *net)
 		return -ENOMEM;
 
 	/* Hold the beast until a service is registered */
-	ipvs->enable = 0;
+	WRITE_ONCE(ipvs->enable, 0);
 	ipvs->net = net;
 	/* Counters used for creating unique names */
 	ipvs->gen = atomic_read(&ipvs_netns_cnt);
@@ -2367,7 +2364,7 @@ static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
 		ipvs = net_ipvs(net);
 		ip_vs_unregister_hooks(ipvs, AF_INET);
 		ip_vs_unregister_hooks(ipvs, AF_INET6);
-		ipvs->enable = 0;	/* Disable packet reception */
+		WRITE_ONCE(ipvs->enable, 0);	/* Disable packet reception */
 		smp_wmb();
 		ip_vs_sync_net_cleanup(ipvs);
 	}
net/netfilter/ipvs/ip_vs_ctl.c +3 -3
index 6a6fc4478..4c8fa22be 100644
@@ -256,7 +256,7 @@ static void est_reload_work_handler(struct work_struct *work)
 		struct ip_vs_est_kt_data *kd = ipvs->est_kt_arr[id];
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock;
 		if (!kd)
 			continue;
@@ -1483,9 +1483,9 @@ ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 
 	*svc_p = svc;
 
-	if (!ipvs->enable) {
+	if (!READ_ONCE(ipvs->enable)) {
 		/* Now there is a service - full throttle */
-		ipvs->enable = 1;
+		WRITE_ONCE(ipvs->enable, 1);
 
 		/* Start estimation for first time */
 		ip_vs_est_reload_start(ipvs);
net/netfilter/ipvs/ip_vs_est.c +8 -8
index 15049b826..93a925f1e 100644
@@ -231,7 +231,7 @@ static int ip_vs_estimation_kthread(void *data)
 void ip_vs_est_reload_start(struct netns_ipvs *ipvs)
 {
 	/* Ignore reloads before first service is added */
-	if (!ipvs->enable)
+	if (!READ_ONCE(ipvs->enable))
 		return;
 	ip_vs_est_stopped_recalc(ipvs);
 	/* Bump the kthread configuration genid */
@@ -306,7 +306,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	int i;
 
 	if ((unsigned long)ipvs->est_kt_count >= ipvs->est_max_threads &&
-	    ipvs->enable && ipvs->est_max_threads)
+	    READ_ONCE(ipvs->enable) && ipvs->est_max_threads)
 		return -EINVAL;
 
 	mutex_lock(&ipvs->est_mutex);
@@ -343,7 +343,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	}
 
 	/* Start kthread tasks only when services are present */
-	if (ipvs->enable && !ip_vs_est_stopped(ipvs)) {
+	if (READ_ONCE(ipvs->enable) && !ip_vs_est_stopped(ipvs)) {
 		ret = ip_vs_est_kthread_start(ipvs, kd);
 		if (ret < 0)
 			goto out;
@@ -486,7 +486,7 @@ int ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 	struct ip_vs_estimator *est = &stats->est;
 	int ret;
 
-	if (!ipvs->est_max_threads && ipvs->enable)
+	if (!ipvs->est_max_threads && READ_ONCE(ipvs->enable))
 		ipvs->est_max_threads = ip_vs_est_max_threads(ipvs);
 
 	est->ktid = -1;
@@ -663,7 +663,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 			/* Wait for cpufreq frequency transition */
 			wait_event_idle_timeout(wq, kthread_should_stop(),
 						HZ / 50);
-			if (!ipvs->enable || kthread_should_stop())
+			if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 				goto stop;
 		}
 
@@ -681,7 +681,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 		rcu_read_unlock();
 		local_bh_enable();
 
-		if (!ipvs->enable || kthread_should_stop())
+		if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 			goto stop;
 		cond_resched();
 
@@ -757,7 +757,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	mutex_lock(&ipvs->est_mutex);
 	for (id = 1; id < ipvs->est_kt_count; id++) {
 		/* netns clean up started, abort */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock2;
 		kd = ipvs->est_kt_arr[id];
 		if (!kd)
@@ -787,7 +787,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	id = ipvs->est_kt_count;
 
 next_kt:
-	if (!ipvs->enable || kthread_should_stop())
+	if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 		goto unlock;
 	id--;
 	if (id < 0)
Signature · 1 line
2.34.1
#8 lore raw

Re: [PATCH v2] net/netfilter/ipvs: Use READ_ONCE/WRITE_ONCE for ipvs->enable

Hi everyone,

My apologies for the noise and confusion with the multiple v2 submissions. I sent a malformed reply earlier with no subject.

This patch in *this* thread is the correct and final version for review. Please disregard the previous malformed email.

Many thanks to Pablo Neira Ayuso for pointing out the formatting error.

maintainer
#9 lore raw

Re: [PATCH v2] net/netfilter/ipvs: Use READ_ONCE/WRITE_ONCE for ipvs->enable

In reply to Zhang Tengfei

Hello,

Quoted 10 lines · Zhang Tengfei

KCSAN reported a data-race on the ipvs->enable flag, which is written in the control path and read concurrently from many other contexts.

Following a suggestion by Julian, this patch fixes the race by converting all accesses to use WRITE_ONCE()/READ_ONCE(). This lightweight approach ensures atomic access and acts as a compiler barrier, preventing unsafe optimizations where the flag is checked in loops (e.g., in ip_vs_est.c).

Quoted 3 lines

Additionally, the enable checks in the fast-path hooks (ip_vs_in_hook, ip_vs_out_hook, ip_vs_forward_icmp) are removed. They are considered unnecessary because the enable=0

It was good idea to mention about the 857ca89711de
commit here as in the previous v2 version. You can even add it as Fixes tag as suggested here:

scripts/checkpatch.pl --strict /tmp/file.patch

As for the Subject line, you probably can use
[PATCH v3 nf-next] ipvs: Use READ_ONCE/WRITE_ONCE for ipvs->enable to specify the desired target tree ('nf-next' or 'nf' if such data-race needs it).

As for the patch code, it looks ok.

Quoted 14 lines

condition they check for can only occur in two rare and non-fatal
scenarios: 1) after hooks are registered but before the flag is set, and 2) after hooks are unregistered on cleanup_net. In the worst case, a single packet might be mishandled (e.g., dropped), which does not lead to a system crash or data corruption. Adding a check in the performance-critical fast-path to handle this harmless condition is not a worthwhile trade-off.

Quoted 5 lines

---
v2:
- Switched from atomic_t to the suggested READ_ONCE()/WRITE_ONCE().
- Removed obsolete checks from the packet processing hooks.
- Polished commit message based on feedback from maintainers.

...

Regards

--
Julian Anastasov <[email protected]>

#10 lore raw

PATCH v3 1/1 ipvs: Use READ_ONCE/WRITE_ONCE for ipvs->enable

KCSAN reported a data-race on the ipvs->enable flag, which is written in the control path and read concurrently from many other contexts.

Following a suggestion by Julian, this patch fixes the race by converting all accesses to use WRITE_ONCE()/READ_ONCE(). This lightweight approach ensures atomic access and acts as a compiler barrier, preventing unsafe optimizations where the flag is checked in loops (e.g., in ip_vs_est.c).

Additionally, the enable checks in the fast-path hooks (ip_vs_in_hook, ip_vs_out_hook, ip_vs_forward_icmp) are removed. These are unnecessary since commit 857ca89711de ("ipvs: register hooks only with services"). The enable=0 condition they check for can only occur in two rare and non-fatal
scenarios: 1) after hooks are registered but before the flag is set, and 2) after hooks are unregistered on cleanup_net. In the worst case, a single packet might be mishandled (e.g., dropped), which does not lead to a system crash or data corruption. Adding a check in the performance-critical fast-path to handle this harmless condition is not a worthwhile trade-off.

---
v3:
- Restore reference to commit 857ca89711de in commit message.
- Add corresponding Fixes tag.
v2:
- Switched from atomic_t to the suggested READ_ONCE()/WRITE_ONCE().
- Removed obsolete checks from the packet processing hooks.
- Polished commit message based on feedback from maintainers.

---
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 11 ++++-------
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 4 files changed, 17 insertions(+), 20 deletions(-)
net/netfilter/ipvs/ip_vs_conn.c +2 -2
index 965f3c8e508..37ebb0cb62b 100644
@@ -885,7 +885,7 @@ static void ip_vs_conn_expire(struct timer_list *t)
 			 * conntrack cleanup for the net.
 			 */
 			smp_rmb();
-			if (ipvs->enable)
+			if (READ_ONCE(ipvs->enable))
 				ip_vs_conn_drop_conntrack(cp);
 		}
 
@@ -1439,7 +1439,7 @@ void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
 		cond_resched_rcu();
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			break;
 	}
 	rcu_read_unlock();
net/netfilter/ipvs/ip_vs_core.c +4 -7
index c7a8a08b730..5ea7ab8bf4d 100644
@@ -1353,9 +1353,6 @@ ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat
 	if (unlikely(!skb_dst(skb)))
 		return NF_ACCEPT;
 
-	if (!ipvs->enable)
-		return NF_ACCEPT;
-
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
 #ifdef CONFIG_IP_VS_IPV6
 	if (af == AF_INET6) {
@@ -1940,7 +1937,7 @@ ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state
 		return NF_ACCEPT;
 	}
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -2108,7 +2105,7 @@ ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
 	int r;
 
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	if (state->pf == NFPROTO_IPV4) {
@@ -2295,7 +2292,7 @@ static int __net_init __ip_vs_init(struct net *net)
 		return -ENOMEM;
 
 	/* Hold the beast until a service is registered */
-	ipvs->enable = 0;
+	WRITE_ONCE(ipvs->enable, 0);
 	ipvs->net = net;
 	/* Counters used for creating unique names */
 	ipvs->gen = atomic_read(&ipvs_netns_cnt);
@@ -2367,7 +2364,7 @@ static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
 		ipvs = net_ipvs(net);
 		ip_vs_unregister_hooks(ipvs, AF_INET);
 		ip_vs_unregister_hooks(ipvs, AF_INET6);
-		ipvs->enable = 0;	/* Disable packet reception */
+		WRITE_ONCE(ipvs->enable, 0);	/* Disable packet reception */
 		smp_wmb();
 		ip_vs_sync_net_cleanup(ipvs);
 	}
net/netfilter/ipvs/ip_vs_ctl.c +3 -3
index 6a6fc447853..4c8fa22be88 100644
@@ -256,7 +256,7 @@ static void est_reload_work_handler(struct work_struct *work)
 		struct ip_vs_est_kt_data *kd = ipvs->est_kt_arr[id];
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock;
 		if (!kd)
 			continue;
@@ -1483,9 +1483,9 @@ ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 
 	*svc_p = svc;
 
-	if (!ipvs->enable) {
+	if (!READ_ONCE(ipvs->enable)) {
 		/* Now there is a service - full throttle */
-		ipvs->enable = 1;
+		WRITE_ONCE(ipvs->enable, 1);
 
 		/* Start estimation for first time */
 		ip_vs_est_reload_start(ipvs);
net/netfilter/ipvs/ip_vs_est.c +8 -8
index 15049b82673..93a925f1ed9 100644
@@ -231,7 +231,7 @@ static int ip_vs_estimation_kthread(void *data)
 void ip_vs_est_reload_start(struct netns_ipvs *ipvs)
 {
 	/* Ignore reloads before first service is added */
-	if (!ipvs->enable)
+	if (!READ_ONCE(ipvs->enable))
 		return;
 	ip_vs_est_stopped_recalc(ipvs);
 	/* Bump the kthread configuration genid */
@@ -306,7 +306,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	int i;
 
 	if ((unsigned long)ipvs->est_kt_count >= ipvs->est_max_threads &&
-	    ipvs->enable && ipvs->est_max_threads)
+	    READ_ONCE(ipvs->enable) && ipvs->est_max_threads)
 		return -EINVAL;
 
 	mutex_lock(&ipvs->est_mutex);
@@ -343,7 +343,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	}
 
 	/* Start kthread tasks only when services are present */
-	if (ipvs->enable && !ip_vs_est_stopped(ipvs)) {
+	if (READ_ONCE(ipvs->enable) && !ip_vs_est_stopped(ipvs)) {
 		ret = ip_vs_est_kthread_start(ipvs, kd);
 		if (ret < 0)
 			goto out;
@@ -486,7 +486,7 @@ int ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 	struct ip_vs_estimator *est = &stats->est;
 	int ret;
 
-	if (!ipvs->est_max_threads && ipvs->enable)
+	if (!ipvs->est_max_threads && READ_ONCE(ipvs->enable))
 		ipvs->est_max_threads = ip_vs_est_max_threads(ipvs);
 
 	est->ktid = -1;
@@ -663,7 +663,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 			/* Wait for cpufreq frequency transition */
 			wait_event_idle_timeout(wq, kthread_should_stop(),
 						HZ / 50);
-			if (!ipvs->enable || kthread_should_stop())
+			if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 				goto stop;
 		}
 
@@ -681,7 +681,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 		rcu_read_unlock();
 		local_bh_enable();
 
-		if (!ipvs->enable || kthread_should_stop())
+		if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 			goto stop;
 		cond_resched();
 
@@ -757,7 +757,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	mutex_lock(&ipvs->est_mutex);
 	for (id = 1; id < ipvs->est_kt_count; id++) {
 		/* netns clean up started, abort */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock2;
 		kd = ipvs->est_kt_arr[id];
 		if (!kd)
@@ -787,7 +787,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	id = ipvs->est_kt_count;
 
 next_kt:
-	if (!ipvs->enable || kthread_should_stop())
+	if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 		goto unlock;
 	id--;
 	if (id < 0)
Signature · 1 line
2.34.1
maintainer Acked-by on v3 1/1
#11 lore raw

Re: [PATCH v3 nf-next] ipvs: Use READ_ONCE/WRITE_ONCE for ipvs->enable

Hello,

Quoted 22 lines · Zhang Tengfei

KCSAN reported a data-race on the ipvs->enable flag, which is written in the control path and read concurrently from many other contexts.

Following a suggestion by Julian, this patch fixes the race by converting all accesses to use WRITE_ONCE()/READ_ONCE(). This lightweight approach ensures atomic access and acts as a compiler barrier, preventing unsafe optimizations where the flag is checked in loops (e.g., in ip_vs_est.c).

Additionally, the enable checks in the fast-path hooks (ip_vs_in_hook, ip_vs_out_hook, ip_vs_forward_icmp) are removed. These are unnecessary since commit 857ca89711de ("ipvs: register hooks only with services"). The enable=0 condition they check for can only occur in two rare and non-fatal
scenarios: 1) after hooks are registered but before the flag is set, and 2) after hooks are unregistered on cleanup_net. In the worst case, a single packet might be mishandled (e.g., dropped), which does not lead to a system crash or data corruption. Adding a check in the performance-critical fast-path to handle this harmless condition is not a worthwhile trade-off.

Quoted 6 lines

Looks good to me, thanks!

Quoted 190 lines

---
v3:
- Restore reference to commit 857ca89711de in commit message.
- Add corresponding Fixes tag.
v2:
- Switched from atomic_t to the suggested READ_ONCE()/WRITE_ONCE().
- Removed obsolete checks from the packet processing hooks.
- Polished commit message based on feedback from maintainers.

---
 net/netfilter/ipvs/ip_vs_conn.c |  4 ++--
 net/netfilter/ipvs/ip_vs_core.c | 11 ++++-------
 net/netfilter/ipvs/ip_vs_ctl.c  |  6 +++---
 net/netfilter/ipvs/ip_vs_est.c  | 16 ++++++++--------
 4 files changed, 17 insertions(+), 20 deletions(-)
net/netfilter/ipvs/ip_vs_conn.c +2 -2
index 965f3c8e508..37ebb0cb62b 100644
@@ -885,7 +885,7 @@ static void ip_vs_conn_expire(struct timer_list *t)
 			 * conntrack cleanup for the net.
 			 */
 			smp_rmb();
-			if (ipvs->enable)
+			if (READ_ONCE(ipvs->enable))
 				ip_vs_conn_drop_conntrack(cp);
 		}
 
@@ -1439,7 +1439,7 @@ void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
 		cond_resched_rcu();
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			break;
 	}
 	rcu_read_unlock();
net/netfilter/ipvs/ip_vs_core.c +4 -7
index c7a8a08b730..5ea7ab8bf4d 100644
@@ -1353,9 +1353,6 @@ ip_vs_out_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *stat
 	if (unlikely(!skb_dst(skb)))
 		return NF_ACCEPT;
 
-	if (!ipvs->enable)
-		return NF_ACCEPT;
-
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
 #ifdef CONFIG_IP_VS_IPV6
 	if (af == AF_INET6) {
@@ -1940,7 +1937,7 @@ ip_vs_in_hook(void *priv, struct sk_buff *skb, const struct nf_hook_state *state
 		return NF_ACCEPT;
 	}
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	ip_vs_fill_iph_skb(af, skb, false, &iph);
@@ -2108,7 +2105,7 @@ ip_vs_forward_icmp(void *priv, struct sk_buff *skb,
 	int r;
 
 	/* ipvs enabled in this netns ? */
-	if (unlikely(sysctl_backup_only(ipvs) || !ipvs->enable))
+	if (unlikely(sysctl_backup_only(ipvs)))
 		return NF_ACCEPT;
 
 	if (state->pf == NFPROTO_IPV4) {
@@ -2295,7 +2292,7 @@ static int __net_init __ip_vs_init(struct net *net)
 		return -ENOMEM;
 
 	/* Hold the beast until a service is registered */
-	ipvs->enable = 0;
+	WRITE_ONCE(ipvs->enable, 0);
 	ipvs->net = net;
 	/* Counters used for creating unique names */
 	ipvs->gen = atomic_read(&ipvs_netns_cnt);
@@ -2367,7 +2364,7 @@ static void __net_exit __ip_vs_dev_cleanup_batch(struct list_head *net_list)
 		ipvs = net_ipvs(net);
 		ip_vs_unregister_hooks(ipvs, AF_INET);
 		ip_vs_unregister_hooks(ipvs, AF_INET6);
-		ipvs->enable = 0;	/* Disable packet reception */
+		WRITE_ONCE(ipvs->enable, 0);	/* Disable packet reception */
 		smp_wmb();
 		ip_vs_sync_net_cleanup(ipvs);
 	}
net/netfilter/ipvs/ip_vs_ctl.c +3 -3
index 6a6fc447853..4c8fa22be88 100644
@@ -256,7 +256,7 @@ static void est_reload_work_handler(struct work_struct *work)
 		struct ip_vs_est_kt_data *kd = ipvs->est_kt_arr[id];
 
 		/* netns clean up started, abort delayed work */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock;
 		if (!kd)
 			continue;
@@ -1483,9 +1483,9 @@ ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 
 	*svc_p = svc;
 
-	if (!ipvs->enable) {
+	if (!READ_ONCE(ipvs->enable)) {
 		/* Now there is a service - full throttle */
-		ipvs->enable = 1;
+		WRITE_ONCE(ipvs->enable, 1);
 
 		/* Start estimation for first time */
 		ip_vs_est_reload_start(ipvs);
net/netfilter/ipvs/ip_vs_est.c +8 -8
index 15049b82673..93a925f1ed9 100644
@@ -231,7 +231,7 @@ static int ip_vs_estimation_kthread(void *data)
 void ip_vs_est_reload_start(struct netns_ipvs *ipvs)
 {
 	/* Ignore reloads before first service is added */
-	if (!ipvs->enable)
+	if (!READ_ONCE(ipvs->enable))
 		return;
 	ip_vs_est_stopped_recalc(ipvs);
 	/* Bump the kthread configuration genid */
@@ -306,7 +306,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	int i;
 
 	if ((unsigned long)ipvs->est_kt_count >= ipvs->est_max_threads &&
-	    ipvs->enable && ipvs->est_max_threads)
+	    READ_ONCE(ipvs->enable) && ipvs->est_max_threads)
 		return -EINVAL;
 
 	mutex_lock(&ipvs->est_mutex);
@@ -343,7 +343,7 @@ static int ip_vs_est_add_kthread(struct netns_ipvs *ipvs)
 	}
 
 	/* Start kthread tasks only when services are present */
-	if (ipvs->enable && !ip_vs_est_stopped(ipvs)) {
+	if (READ_ONCE(ipvs->enable) && !ip_vs_est_stopped(ipvs)) {
 		ret = ip_vs_est_kthread_start(ipvs, kd);
 		if (ret < 0)
 			goto out;
@@ -486,7 +486,7 @@ int ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 	struct ip_vs_estimator *est = &stats->est;
 	int ret;
 
-	if (!ipvs->est_max_threads && ipvs->enable)
+	if (!ipvs->est_max_threads && READ_ONCE(ipvs->enable))
 		ipvs->est_max_threads = ip_vs_est_max_threads(ipvs);
 
 	est->ktid = -1;
@@ -663,7 +663,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 			/* Wait for cpufreq frequency transition */
 			wait_event_idle_timeout(wq, kthread_should_stop(),
 						HZ / 50);
-			if (!ipvs->enable || kthread_should_stop())
+			if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 				goto stop;
 		}
 
@@ -681,7 +681,7 @@ static int ip_vs_est_calc_limits(struct netns_ipvs *ipvs, int *chain_max)
 		rcu_read_unlock();
 		local_bh_enable();
 
-		if (!ipvs->enable || kthread_should_stop())
+		if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 			goto stop;
 		cond_resched();
 
@@ -757,7 +757,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	mutex_lock(&ipvs->est_mutex);
 	for (id = 1; id < ipvs->est_kt_count; id++) {
 		/* netns clean up started, abort */
-		if (!ipvs->enable)
+		if (!READ_ONCE(ipvs->enable))
 			goto unlock2;
 		kd = ipvs->est_kt_arr[id];
 		if (!kd)
@@ -787,7 +787,7 @@ static void ip_vs_est_calc_phase(struct netns_ipvs *ipvs)
 	id = ipvs->est_kt_count;
 
 next_kt:
-	if (!ipvs->enable || kthread_should_stop())
+	if (!READ_ONCE(ipvs->enable) || kthread_should_stop())
 		goto unlock;
 	id--;
 	if (id < 0)
Signature · 1 line
2.34.1

Regards

--
Julian Anastasov <[email protected]>