Re:?

Re:?

[Ian Latter]

Hello Eugen,

  I'm sorry but I'm not the H323 author ... and I haven't used netfilter H323
support since ipchains.  I have CCd the netfilter list in case someone 
there can help.





----- Original Message -----
>From: "Soporte Meranetwork" <mera@fibertel.com.ar>
>To: <Ian.Latter@mq.edu.au>
>Subject:  ?
>Date: Mon, 31 Mar 2003 18:56:17 -0300
>
> Good day.I have used IPTabels 1.2.5-3,and dont know that it have or not H323 modul .But 
for recive Call from outside(internet) to inside(throught NAT) - Netmeeting(ATA 186) i use 
thet rules: 
> $IPTABLES -t nat -A POSTROUTING -o $EXTIF -s 192.168.0.0/24 -j MASQUERADE 
>   (intern LAN)
> PORTFWIP="192.168.0.201" ( PC with Netmeting or ATA 186)
> $IPTABLES -A FORWARD -i $EXTIF -o $INTIF -p tcp --dport 1720 -m state --state
> NEW,ESTABLISHED,RELATED -j ACCEPT
> $IPTABLES -A PREROUTING -t nat -p tcp -d $EXTIP --dport 1720 -j DNAT --to 
$PORTFWIP:1720
> But i want to use NAT without  FORWARD ,PREROUTING with H323 ability for any IP in 
LAN.Where i can find IPtables with ability of H323(version) or where is the patch(modul) for 
that ability? Thank your. 
> Buy,
> Eugen

--
Ian Latter
Internet and Networking Security Officer
Macquarie University

Re:

[David Stes]

> Date: Mon, 12 Sep 2005 16:29:38 +0000
> From: Cory Visi <merlin@gentoo.org>
> Subject: patchlets patch submission
> To: netfilter-devel@lists.netfilter.org
> Message-ID: <20050912162938.GB15843@toucan.gentoo.org>
> Content-Type: text/plain; charset="us-ascii"
> 
> I did some work on fixing the conntrack pom addons for the lockhelp.h 
> patch (2.6.13 support essentially). I think I got everything setup except 
> for 2 files.
> 
> Attached is all my work in patch form against a full kernel source. I have 
> a feeling you'd prefer I sent this in some other format or in some other 
> structure of a diff. This is my first time contributing so just let me 
> know how you'd prefer this in the future.
> 
> What's left is:
> - rtsp (part of rsh)

Did you also do the RPC modules ? (RPC for UDP/TCP)

I wonder whether the author(s) of RSH/RPC modules are still reading this list.

If they are, then maybe your 2.6.13 patch (and my older 2.6.12 patches)
could be incorporated.  If they are not, then maybe it's time to see how
future changes to these modules can be coordinated.

Re:

[Patrick McHardy]

Jan Engelhardt wrote:
> Hi,
> 
> 
> here are a number of patches I am suggesting for the -master branch
> (-stable is separate in my view and my management, but see other
> discussion thread).
> Pullable from
> 	git://dev.medozas.de/iptables master
> 
> 
> Jan Engelhardt (9):
>       libiptc: split v4 and v6
>       extensions: collapse registration structures
>       iptables: allow for parse-less extensions
>       iptables: allow for help-less extensions
>       extensions: remove empty help and parse functions
>       xtables: add multi-registration functions
>       extensions: collapse data variables to use multi-reg calls
>       xtables: warn of missing version identifier in extensions
>       COMMIT_NOTES: notice to check for soversion bumps
> 
> Michael Granzow (1):
>       iptables: accept multiple IP address specifications for -s, -d

Looks good, pulled and pushed out again, thanks.

> (Shall I post the entire mergestat, or just the "X files changed" line?)

This is fine, I mainly want something comparable to the git pull output.

Re:

[Patrick McHardy]

Jan Engelhardt wrote:
> Hi Patrick,
> 
> Please pull from
> 	git://dev.medozas.de/iptables master
> 
> which contains a pack of patches to build iptables without libdl,
> obsoleting iptables-static (leaving -multi) and using the -multi
> program exclusively.

Pulled and pushed out again, thanks.

Re:

[Patrick McHardy]

Jan Engelhardt wrote:
> Please pull from
> 	git://dev.medozas.de/iptables master
> 
> to receive
> 
> Jan Engelhardt (2+1):
>       xt_conntrack: revision 2 for enlarged state_mask member
>       libxt_helper: fix invalid passed option to check_inverse

Pulled and pushed back out again, thanks.

Re:

[Patrick McHardy]

Jan Engelhardt wrote:
> here are three commits that fix bugzilla entries and/or other
> problems encountered. There are also two extra commits prepended
> without any changes, which only provide missing log entries for
> already-merged commits.

Just to clarify before I apply this - how does adding changelog
entries afterwards work? Am I correct to assume that this won't
affect this history of the tree and existing clones?

Re:

[Jan Engelhardt]

On Thursday 2009-10-29 23:26, Patrick McHardy wrote:

>Jan Engelhardt wrote:
>> here are three commits that fix bugzilla entries and/or other
>> problems encountered. There are also two extra commits prepended
>> without any changes, which only provide missing log entries for
>> already-merged commits.
>
>Just to clarify before I apply this - how does adding changelog
>entries afterwards work? Am I correct to assume that this won't
>affect this history of the tree and existing clones?

I just used `git commit --allow-empty -e` to record a plain commit on 
top, just without any change in the tree object. Take a look in 
git-forest/gitk if in doubt ;-)

Re:

[Patrick McHardy]

Jan Engelhardt wrote:
> On Thursday 2009-10-29 23:26, Patrick McHardy wrote:
> 
>> Jan Engelhardt wrote:
>>> here are three commits that fix bugzilla entries and/or other
>>> problems encountered. There are also two extra commits prepended
>>> without any changes, which only provide missing log entries for
>>> already-merged commits.
>> Just to clarify before I apply this - how does adding changelog
>> entries afterwards work? Am I correct to assume that this won't
>> affect this history of the tree and existing clones?
> 
> I just used `git commit --allow-empty -e` to record a plain commit on 
> top, just without any change in the tree object. Take a look in 
> git-forest/gitk if in doubt ;-)

Nice. Pulled and pushed out again, thanks.

Re:

[William Wilcox]

Good day!
My name is Sir William Wilcox,I work with the Euro Lottery. I can help you
win 4,528,000 GBP.But I charge 40% of the winning.Can we do this deal
together? Email me; william.wilcox98@gmail.com

Re:

[Mistick Levi]

What's up with all those mail's about business proposal's that are
sent to the mailing list?!

On Sat, Oct 9, 2010 at 7:26 PM, Mr.Young Chang <info@yahoo.com> wrote:
> My name is Mr.Young Chang,Credit officer MEVAS BANK,HK.I have a Business
> Proposal of $19.7 million usd for you to handle with me.Are you interested?
>
>
>
>
>
>
> ----------------------------------------------------------------
> This message was sent using IMP, the Internet Messaging Program.
>
> --
> To unsubscribe from this list: send the line "unsubscribe netfilter" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re:

[James Brown]

https://docs.google.com/document/d/1yAkUys2osN7co_KbzphWLLsoe-TPq7ELZhoySYvzjF0/edit

RE:

[W. Cheung]

 I have a very lucrative business transaction which requires the utmost discretion. If you are interested, kindly contact me ASAP for full details.

Warm Regards,
William Cheung

Re:

[christain147]

Good day,hoping you read this email and respond to me in good time.I do not intend to solicit for funds but  your time and energy in using my own resources to assist the less privileged.I am medically confined at the moment hence I request your indulgence.
I will give you a comprehensive brief once I hear from you.

Please forward your response to my private email address:
gudworks104@yahoo.com

Thanks and reply.

Robert Grondahl

RE:

[JO Bower]

Your email address has brought you an unexpected luck, which was selected in The Euro Millions Lottery and subsequently won you the sum of €1,000,000.00 Euros. Contact Monica Torres Email: monicatorresesp@gmail.com to claim your prize.
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE:

[Qin's Yanjun]

----
How are you today and your family? I require your attention and honest
co-operation about some issues which i will really want to discuss with you
which.  Looking forward to read from you soon.  

Qin's


______________________________

Sky Silk, http://aknet.kz

Re:

[Amos Kalonzo]

Attn:

I am wondering why You haven't respond to my email for some days now.
reference to my client's contract balance payment of (11.7M,USD)
Kindly get back to me for more details.

Best Regards

Amos Kalonzo

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

[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.

Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>
---
 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(-)

diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 29a36709e7f3..58b2ad7906e8 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -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)
diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
index 965f3c8e5089..5c97f85929b4 100644
--- a/net/netfilter/ipvs/ip_vs_conn.c
+++ b/net/netfilter/ipvs/ip_vs_conn.c
@@ -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();
diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
index c7a8a08b7308..84eed2e45927 100644
--- a/net/netfilter/ipvs/ip_vs_core.c
+++ b/net/netfilter/ipvs/ip_vs_core.c
@@ -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);
 	}
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index 6a6fc4478533..ad7e1c387c1f 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -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);
diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
index 15049b826732..c5aa2660de92 100644
--- a/net/netfilter/ipvs/ip_vs_est.c
+++ b/net/netfilter/ipvs/ip_vs_est.c
@@ -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)
-- 
2.47.3

Re: [PATCH] net/netfilter/ipvs: Fix data-race in ip_vs_add_service / ip_vs_out_hook

[Florian Westphal]

Zhang Tengfei <zhtfdev@gmail.com> wrote:
> 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?

> 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.

> -	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).

Re: [PATCH] net/netfilter/ipvs: Fix data-race in ip_vs_add_service / ip_vs_out_hook

[Eric Dumazet]

On Tue, Aug 26, 2025 at 7:18 AM Florian Westphal <fw@strlen.de> wrote:
>
> Zhang Tengfei <zhtfdev@gmail.com> wrote:
> > 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

>
> > 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.
>
> > -     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

Re: [PATCH] net/netfilter/ipvs: Fix data-race in ip_vs_add_service / ip_vs_out_hook

[Julian Anastasov]

	Hello,

On Tue, 26 Aug 2025, Zhang Tengfei wrote:

> 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).

> 
> 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.

> 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.

> Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
> Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>
> ---
>  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(-)
> 
> diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
> index 29a36709e7f3..58b2ad7906e8 100644
> --- a/include/net/ip_vs.h
> +++ b/include/net/ip_vs.h
> @@ -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)
> diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
> index 965f3c8e5089..5c97f85929b4 100644
> --- a/net/netfilter/ipvs/ip_vs_conn.c
> +++ b/net/netfilter/ipvs/ip_vs_conn.c
> @@ -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.

>  				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();
> diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
> index c7a8a08b7308..84eed2e45927 100644
> --- a/net/netfilter/ipvs/ip_vs_core.c
> +++ b/net/netfilter/ipvs/ip_vs_core.c
> @@ -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.

> +	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);

	No new/flying packets after this point...

> -		ipvs->enable = 0;	/* Disable packet reception */

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

> +		atomic_set(&ipvs->enable, 0);	/* Disable packet reception */
>  		smp_wmb();
>  		ip_vs_sync_net_cleanup(ipvs);
>  	}
> diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
> index 6a6fc4478533..ad7e1c387c1f 100644
> --- a/net/netfilter/ipvs/ip_vs_ctl.c
> +++ b/net/netfilter/ipvs/ip_vs_ctl.c
> @@ -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.

> +		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);
> diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
> index 15049b826732..c5aa2660de92 100644
> --- a/net/netfilter/ipvs/ip_vs_est.c
> +++ b/net/netfilter/ipvs/ip_vs_est.c
> @@ -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)
> -- 

	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 <ja@ssi.bg>

Re: [PATCH] net/netfilter/ipvs: Fix data-race in ip_vs_add_service / ip_vs_out_hook

[Julian Anastasov]

	Hello,

On Tue, 26 Aug 2025, Zhang Tengfei wrote:

> 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.
> 
> Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
> Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>
> ---
>  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(-)
> 

> diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
> index 15049b826732..c5aa2660de92 100644
> --- a/net/netfilter/ipvs/ip_vs_est.c
> +++ b/net/netfilter/ipvs/ip_vs_est.c
...
> @@ -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 <ja@ssi.bg>

(no subject)

[Zhang Tengfei]

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").

Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
Suggested-by: Julian Anastasov <ja@ssi.bg>
Link: https://lore.kernel.org/lvs-devel/2189fc62-e51e-78c9-d1de-d35b8e3657e3@ssi.bg/
Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>

---
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(-)

diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
index 965f3c8e5..37ebb0cb6 100644
--- a/net/netfilter/ipvs/ip_vs_conn.c
+++ b/net/netfilter/ipvs/ip_vs_conn.c
@@ -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();
diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
index c7a8a08b7..5ea7ab8bf 100644
--- a/net/netfilter/ipvs/ip_vs_core.c
+++ b/net/netfilter/ipvs/ip_vs_core.c
@@ -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);
 	}
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index 6a6fc4478..4c8fa22be 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -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);
diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
index 15049b826..93a925f1e 100644
--- a/net/netfilter/ipvs/ip_vs_est.c
+++ b/net/netfilter/ipvs/ip_vs_est.c
@@ -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)
-- 
2.34.1

Re:

[Pablo Neira Ayuso]

On Wed, Aug 27, 2025 at 10:43:42PM +0800, Zhang Tengfei wrote:
> 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.

> 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").
> 
> Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
> Suggested-by: Julian Anastasov <ja@ssi.bg>
> Link: https://lore.kernel.org/lvs-devel/2189fc62-e51e-78c9-d1de-d35b8e3657e3@ssi.bg/
> Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>
> 
> ---
> 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(-)
[...]
> diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
> index c7a8a08b7..5ea7ab8bf 100644
> --- a/net/netfilter/ipvs/ip_vs_core.c
> +++ b/net/netfilter/ipvs/ip_vs_core.c
> @@ -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

>  	ip_vs_fill_iph_skb(af, skb, false, &iph);
>  #ifdef CONFIG_IP_VS_IPV6
>  	if (af == AF_INET6) {

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

[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. 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.

Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
Suggested-by: Julian Anastasov <ja@ssi.bg>
Link: https://lore.kernel.org/lvs-devel/2189fc62-e51e-78c9-d1de-d35b8e3657e3@ssi.bg/
Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>

---
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(-)

diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
index 965f3c8e5..37ebb0cb6 100644
--- a/net/netfilter/ipvs/ip_vs_conn.c
+++ b/net/netfilter/ipvs/ip_vs_conn.c
@@ -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();
diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
index c7a8a08b7..5ea7ab8bf 100644
--- a/net/netfilter/ipvs/ip_vs_core.c
+++ b/net/netfilter/ipvs/ip_vs_core.c
@@ -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);
 	}
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index 6a6fc4478..4c8fa22be 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -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);
diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
index 15049b826..93a925f1e 100644
--- a/net/netfilter/ipvs/ip_vs_est.c
+++ b/net/netfilter/ipvs/ip_vs_est.c
@@ -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)
-- 
2.34.1

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

[Zhang Tengfei]

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.

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

[Julian Anastasov]

	Hello,

On Thu, 28 Aug 2025, Zhang Tengfei wrote:

> 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`

	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.

> 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.
> 
> Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
> Suggested-by: Julian Anastasov <ja@ssi.bg>
> Link: https://lore.kernel.org/lvs-devel/2189fc62-e51e-78c9-d1de-d35b8e3657e3@ssi.bg/
> Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>
> 
> ---
> 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 <ja@ssi.bg>

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

[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.

Fixes: 857ca89711de ("ipvs: register hooks only with services")
Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
Suggested-by: Julian Anastasov <ja@ssi.bg>
Link: https://lore.kernel.org/lvs-devel/2189fc62-e51e-78c9-d1de-d35b8e3657e3@ssi.bg/
Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>
---
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(-)

diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
index 965f3c8e508..37ebb0cb62b 100644
--- a/net/netfilter/ipvs/ip_vs_conn.c
+++ b/net/netfilter/ipvs/ip_vs_conn.c
@@ -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();
diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
index c7a8a08b730..5ea7ab8bf4d 100644
--- a/net/netfilter/ipvs/ip_vs_core.c
+++ b/net/netfilter/ipvs/ip_vs_core.c
@@ -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);
 	}
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index 6a6fc447853..4c8fa22be88 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -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);
diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
index 15049b82673..93a925f1ed9 100644
--- a/net/netfilter/ipvs/ip_vs_est.c
+++ b/net/netfilter/ipvs/ip_vs_est.c
@@ -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)
-- 
2.34.1

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

[Julian Anastasov]

	Hello,

On Mon, 1 Sep 2025, Zhang Tengfei wrote:

> 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.
> 
> Fixes: 857ca89711de ("ipvs: register hooks only with services")
> Reported-by: syzbot+1651b5234028c294c339@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=1651b5234028c294c339
> Suggested-by: Julian Anastasov <ja@ssi.bg>
> Link: https://lore.kernel.org/lvs-devel/2189fc62-e51e-78c9-d1de-d35b8e3657e3@ssi.bg/
> Signed-off-by: Zhang Tengfei <zhtfdev@gmail.com>

	Looks good to me, thanks!

Acked-by: Julian Anastasov <ja@ssi.bg>

> ---
> 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(-)
> 
> diff --git a/net/netfilter/ipvs/ip_vs_conn.c b/net/netfilter/ipvs/ip_vs_conn.c
> index 965f3c8e508..37ebb0cb62b 100644
> --- a/net/netfilter/ipvs/ip_vs_conn.c
> +++ b/net/netfilter/ipvs/ip_vs_conn.c
> @@ -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();
> diff --git a/net/netfilter/ipvs/ip_vs_core.c b/net/netfilter/ipvs/ip_vs_core.c
> index c7a8a08b730..5ea7ab8bf4d 100644
> --- a/net/netfilter/ipvs/ip_vs_core.c
> +++ b/net/netfilter/ipvs/ip_vs_core.c
> @@ -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);
>  	}
> diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
> index 6a6fc447853..4c8fa22be88 100644
> --- a/net/netfilter/ipvs/ip_vs_ctl.c
> +++ b/net/netfilter/ipvs/ip_vs_ctl.c
> @@ -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);
> diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
> index 15049b82673..93a925f1ed9 100644
> --- a/net/netfilter/ipvs/ip_vs_est.c
> +++ b/net/netfilter/ipvs/ip_vs_est.c
> @@ -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)
> -- 
> 2.34.1

Regards

--
Julian Anastasov <ja@ssi.bg>