[PATCH v2 net 0/3] ipv4/ipv6: Fix UAF and memory leak in IGMP/MLD

[PATCH v2 net 0/3] ipv4/ipv6: Fix UAF and memory leak in IGMP/MLD

[Eric Dumazet]

This series addresses two potential UAF vulnerabilities
and memory leaks in the IPv4 IGMP and IPv6 MLD subsystems.

The first two patches fix a UAF where the packet receive path races with
device teardown. If the device refcount has already hit 0 (but the memory
is still held by RCU), incoming IGMP/MLD packets trying to schedule delayed
work or timers would call refcount_inc() on the 0 refcount, triggering a
warning and eventually leading to a UAF when the work runs after the device
has been freed. This is fixed by introducing safe hold helpers using
refcount_inc_not_zero(). In MLD, we also ensure we only enqueue the skb
if we successfully acquired the device reference, to avoid leaking skbs
when the device is being destroyed.

The third patch fixes memory leaks in IPv4 IGMP when timers are deleted or
stopped. When a timer is deleted (in igmp_mod_timer) or stopped (in
igmp_stop_timer) and not re-armed, the code dropped the group refcount using
refcount_dec(). However, if the group was concurrently removed from the list,
this decrement could drop the refcount to 0 without triggering the
cleanup/free path, leaking the group structure. This is fixed by using
ip_ma_put() instead, and deferring the put until after the lock is released.

Eric Dumazet (3):
  ipv4: igmp: Fix potential UAF in igmp_gq_start_timer()
  ipv6: mcast: Fix potential UAF in MLD delayed work
  ipv4: igmp: Fix potential memory leaks in igmp_mod_timer() and
    igmp_stop_timer()

 include/linux/inetdevice.h |  5 +++++
 include/net/addrconf.h     |  5 +++++
 net/ipv4/igmp.c            | 28 +++++++++++++++++++-------
 net/ipv6/mcast.c           | 40 ++++++++++++++++++++++++++------------
 4 files changed, 59 insertions(+), 19 deletions(-)

-- 
2.55.0.rc0.799.gd6f94ed593-goog

[PATCH v2 net 1/3] ipv4: igmp: Fix potential UAF in igmp_gq_start_timer()

[Eric Dumazet]

A race condition exists between device teardown (inetdev_destroy) and
incoming IGMP query processing (igmp_rcv), leading to a Use-After-Free
in the IGMP timer callback.

During device destruction, inetdev_destroy() drops the primary reference
to in_device, which can drop its refcount to 0. The actual freeing of
in_device memory is deferred via RCU (using call_rcu()).

Concurrently, igmp_rcv() runs under RCU read lock and obtains the
in_device pointer. Because the memory is RCU-protected, CPU-0 can safely
dereference in_device even if its refcount has hit 0.

However, if CPU-0 calls igmp_gq_start_timer() and re-arms the timer, it
attempts to acquire a reference using in_dev_hold(). This increments the
refcount from 0 to 1, triggering a "refcount_t: addition on 0" warning.
Since the in_device memory is still scheduled to be freed after the RCU
grace period (as the free callback does not check the refcount again),
the device is freed while the timer is still armed. When the timer
expires, it accesses the freed memory, causing a kernel panic.

Fix this by using refcount_inc_not_zero() (via a new helper
in_dev_hold_safe()) to prevent acquiring a reference if the device is
already being destroyed. If the refcount is 0, we do not arm the timer.

A similar issue in IPv6 MLD is fixed in a subsequent patch.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: Zero Day Initiative <zdi-disclosures@trendmicro.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/inetdevice.h |  5 +++++
 net/ipv4/igmp.c            | 14 +++++++++-----
 2 files changed, 14 insertions(+), 5 deletions(-)

diff --git a/include/linux/inetdevice.h b/include/linux/inetdevice.h
index dccbeb25f70141982160c776f3cc727296def2b7..6032eea2539a60d0476d85667bda3bfcad8f1425 100644
--- a/include/linux/inetdevice.h
+++ b/include/linux/inetdevice.h
@@ -293,6 +293,11 @@ static inline void in_dev_put(struct in_device *idev)
 #define __in_dev_put(idev)  refcount_dec(&(idev)->refcnt)
 #define in_dev_hold(idev)   refcount_inc(&(idev)->refcnt)
 
+static inline bool in_dev_hold_safe(struct in_device *idev)
+{
+	return refcount_inc_not_zero(&idev->refcnt);
+}
+
 #endif /* __KERNEL__ */
 
 static __inline__ __be32 inet_make_mask(int logmask)
diff --git a/net/ipv4/igmp.c b/net/ipv4/igmp.c
index b6337a47c1418589bf8ef31e00fd98b6187f5271..f5f9763895641bf86bfcf9fd7fd7b06012fa4ece 100644
--- a/net/ipv4/igmp.c
+++ b/net/ipv4/igmp.c
@@ -248,16 +248,20 @@ static void igmp_gq_start_timer(struct in_device *in_dev)
 		return;
 
 	in_dev->mr_gq_running = 1;
-	if (!mod_timer(&in_dev->mr_gq_timer, exp))
-		in_dev_hold(in_dev);
+	if (in_dev_hold_safe(in_dev)) {
+		if (mod_timer(&in_dev->mr_gq_timer, exp))
+			in_dev_put(in_dev);
+	}
 }
 
 static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 {
-	int tv = get_random_u32_below(delay);
+	if (in_dev_hold_safe(in_dev)) {
+		int tv = get_random_u32_below(delay);
 
-	if (!mod_timer(&in_dev->mr_ifc_timer, jiffies+tv+2))
-		in_dev_hold(in_dev);
+		if (mod_timer(&in_dev->mr_ifc_timer, jiffies + tv + 2))
+			in_dev_put(in_dev);
+	}
 }
 
 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
-- 
2.55.0.rc0.799.gd6f94ed593-goog

Re: [PATCH v2 net 1/3] ipv4: igmp: Fix potential UAF in igmp_gq_start_timer()

[Ido Schimmel]

On Sun, Jul 05, 2026 at 06:17:54PM +0000, Eric Dumazet wrote:
> A race condition exists between device teardown (inetdev_destroy) and
> incoming IGMP query processing (igmp_rcv), leading to a Use-After-Free
> in the IGMP timer callback.
> 
> During device destruction, inetdev_destroy() drops the primary reference
> to in_device, which can drop its refcount to 0. The actual freeing of
> in_device memory is deferred via RCU (using call_rcu()).
> 
> Concurrently, igmp_rcv() runs under RCU read lock and obtains the
> in_device pointer. Because the memory is RCU-protected, CPU-0 can safely
> dereference in_device even if its refcount has hit 0.
> 
> However, if CPU-0 calls igmp_gq_start_timer() and re-arms the timer, it
> attempts to acquire a reference using in_dev_hold(). This increments the
> refcount from 0 to 1, triggering a "refcount_t: addition on 0" warning.
> Since the in_device memory is still scheduled to be freed after the RCU
> grace period (as the free callback does not check the refcount again),
> the device is freed while the timer is still armed. When the timer
> expires, it accesses the freed memory, causing a kernel panic.
> 
> Fix this by using refcount_inc_not_zero() (via a new helper
> in_dev_hold_safe()) to prevent acquiring a reference if the device is
> already being destroyed. If the refcount is 0, we do not arm the timer.
> 
> A similar issue in IPv6 MLD is fixed in a subsequent patch.
> 
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Reported-by: Zero Day Initiative <zdi-disclosures@trendmicro.com>
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Reviewed-by: Ido Schimmel <idosch@nvidia.com>

[PATCH v2 net 2/3] ipv6: mcast: Fix potential UAF in MLD delayed work

[Eric Dumazet]

A race condition exists between device teardown and incoming MLD query
processing, leading to a Use-After-Free in the MLD delayed work.

During device destruction, the primary reference to inet6_dev is dropped,
which can drop its refcount to 0. The actual freeing of inet6_dev memory
is deferred via RCU.

Concurrently, the packet receive path runs under RCU read lock and obtains
the inet6_dev pointer. Because the memory is RCU-protected, CPU-0 can
safely dereference inet6_dev even if its refcount has hit 0.

However, if CPU-0 calls igmp6_event_query() and schedules delayed work, it
attempts to acquire a reference using in6_dev_hold(). This increments the
refcount from 0 to 1, triggering a "refcount_t: addition on 0" warning.
Since the inet6_dev memory is still scheduled to be freed after the RCU
grace period, the device is freed while the work is still scheduled.
When the work runs, it accesses the freed memory, causing a kernel panic.

Fix this by using refcount_inc_not_zero() (via a new helper
in6_dev_hold_safe()) to prevent acquiring a reference if the device is
already being destroyed. If the refcount is 0, we do not schedule the work.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 include/net/addrconf.h |  5 +++++
 net/ipv6/mcast.c       | 40 ++++++++++++++++++++++++++++------------
 2 files changed, 33 insertions(+), 12 deletions(-)

diff --git a/include/net/addrconf.h b/include/net/addrconf.h
index 539bbbe54b14e8108ff7304d7a08bc605655cc31..8ced27a8229b6e0580f934be2223676cc123307b 100644
--- a/include/net/addrconf.h
+++ b/include/net/addrconf.h
@@ -446,6 +446,11 @@ static inline void in6_dev_hold(struct inet6_dev *idev)
 	refcount_inc(&idev->refcnt);
 }
 
+static inline bool in6_dev_hold_safe(struct inet6_dev *idev)
+{
+	return refcount_inc_not_zero(&idev->refcnt);
+}
+
 /* called with rcu_read_lock held */
 static inline bool ip6_ignore_linkdown(const struct net_device *dev)
 {
diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c
index 04b811b3be978e36b0fd4ecd8312313d73ed2ed9..4d2b9377ba2de5ad2b6d503bb4b285bfd1307a55 100644
--- a/net/ipv6/mcast.c
+++ b/net/ipv6/mcast.c
@@ -1083,8 +1083,10 @@ static void mld_gq_start_work(struct inet6_dev *idev)
 	mc_assert_locked(idev);
 
 	idev->mc_gq_running = 1;
-	if (!mod_delayed_work(mld_wq, &idev->mc_gq_work, tv + 2))
-		in6_dev_hold(idev);
+	if (in6_dev_hold_safe(idev)) {
+		if (mod_delayed_work(mld_wq, &idev->mc_gq_work, tv + 2))
+			in6_dev_put(idev);
+	}
 }
 
 static void mld_gq_stop_work(struct inet6_dev *idev)
@@ -1102,8 +1104,10 @@ static void mld_ifc_start_work(struct inet6_dev *idev, unsigned long delay)
 
 	mc_assert_locked(idev);
 
-	if (!mod_delayed_work(mld_wq, &idev->mc_ifc_work, tv + 2))
-		in6_dev_hold(idev);
+	if (in6_dev_hold_safe(idev)) {
+		if (mod_delayed_work(mld_wq, &idev->mc_ifc_work, tv + 2))
+			in6_dev_put(idev);
+	}
 }
 
 static void mld_ifc_stop_work(struct inet6_dev *idev)
@@ -1121,8 +1125,10 @@ static void mld_dad_start_work(struct inet6_dev *idev, unsigned long delay)
 
 	mc_assert_locked(idev);
 
-	if (!mod_delayed_work(mld_wq, &idev->mc_dad_work, tv + 2))
-		in6_dev_hold(idev);
+	if (in6_dev_hold_safe(idev)) {
+		if (mod_delayed_work(mld_wq, &idev->mc_dad_work, tv + 2))
+			in6_dev_put(idev);
+	}
 }
 
 static void mld_dad_stop_work(struct inet6_dev *idev)
@@ -1395,18 +1401,23 @@ static void mld_process_v2(struct inet6_dev *idev, struct mld2_query *mld,
 void igmp6_event_query(struct sk_buff *skb)
 {
 	struct inet6_dev *idev = __in6_dev_get(skb->dev);
+	bool put = false;
 
 	if (!idev || idev->dead)
 		goto out;
 
 	spin_lock_bh(&idev->mc_query_lock);
-	if (skb_queue_len(&idev->mc_query_queue) < MLD_MAX_SKBS) {
+	if (skb_queue_len(&idev->mc_query_queue) < MLD_MAX_SKBS &&
+	    in6_dev_hold_safe(idev)) {
 		__skb_queue_tail(&idev->mc_query_queue, skb);
-		if (!mod_delayed_work(mld_wq, &idev->mc_query_work, 0))
-			in6_dev_hold(idev);
+		if (mod_delayed_work(mld_wq, &idev->mc_query_work, 0))
+			put = true;
 		skb = NULL;
 	}
 	spin_unlock_bh(&idev->mc_query_lock);
+
+	if (put)
+		in6_dev_put(idev);
 out:
 	kfree_skb(skb);
 }
@@ -1570,18 +1581,23 @@ static void mld_query_work(struct work_struct *work)
 void igmp6_event_report(struct sk_buff *skb)
 {
 	struct inet6_dev *idev = __in6_dev_get(skb->dev);
+	bool put = false;
 
 	if (!idev || idev->dead)
 		goto out;
 
 	spin_lock_bh(&idev->mc_report_lock);
-	if (skb_queue_len(&idev->mc_report_queue) < MLD_MAX_SKBS) {
+	if (skb_queue_len(&idev->mc_report_queue) < MLD_MAX_SKBS &&
+	    in6_dev_hold_safe(idev)) {
 		__skb_queue_tail(&idev->mc_report_queue, skb);
-		if (!mod_delayed_work(mld_wq, &idev->mc_report_work, 0))
-			in6_dev_hold(idev);
+		if (mod_delayed_work(mld_wq, &idev->mc_report_work, 0))
+			put = true;
 		skb = NULL;
 	}
 	spin_unlock_bh(&idev->mc_report_lock);
+
+	if (put)
+		in6_dev_put(idev);
 out:
 	kfree_skb(skb);
 }
-- 
2.55.0.rc0.799.gd6f94ed593-goog

Re: [PATCH v2 net 2/3] ipv6: mcast: Fix potential UAF in MLD delayed work

[Ido Schimmel]

On Sun, Jul 05, 2026 at 06:17:55PM +0000, Eric Dumazet wrote:
> A race condition exists between device teardown and incoming MLD query
> processing, leading to a Use-After-Free in the MLD delayed work.
> 
> During device destruction, the primary reference to inet6_dev is dropped,
> which can drop its refcount to 0. The actual freeing of inet6_dev memory
> is deferred via RCU.
> 
> Concurrently, the packet receive path runs under RCU read lock and obtains
> the inet6_dev pointer. Because the memory is RCU-protected, CPU-0 can
> safely dereference inet6_dev even if its refcount has hit 0.
> 
> However, if CPU-0 calls igmp6_event_query() and schedules delayed work, it
> attempts to acquire a reference using in6_dev_hold(). This increments the
> refcount from 0 to 1, triggering a "refcount_t: addition on 0" warning.
> Since the inet6_dev memory is still scheduled to be freed after the RCU
> grace period, the device is freed while the work is still scheduled.
> When the work runs, it accesses the freed memory, causing a kernel panic.
> 
> Fix this by using refcount_inc_not_zero() (via a new helper
> in6_dev_hold_safe()) to prevent acquiring a reference if the device is
> already being destroyed. If the refcount is 0, we do not schedule the work.
> 
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Reviewed-by: Ido Schimmel <idosch@nvidia.com>

[PATCH v2 net 3/3] ipv4: igmp: Fix potential memory leaks in igmp_mod_timer() and igmp_stop_timer()

[Eric Dumazet]

When a timer is deleted and not re-armed in igmp_mod_timer(), or stopped
in igmp_stop_timer(), the code currently decrements the reference counter
of the multicast list entry @im using refcount_dec(&im->refcnt).

However, both functions can be called from the RCU reader path:
- igmp_mod_timer() via igmp_heard_query() -> for_each_pmc_rcu()
- igmp_stop_timer() via igmp_rcv() -> igmp_heard_report()

If the group im was concurrently removed from the list by ip_mc_dec_group(),
its reference count might have already been decremented to 1.

In this case, timer_delete() succeeds, and refcount_dec() decrements
the refcount from 1 to 0. Since refcount_dec() does not free the object
when it hits 0 (unlike ip_ma_put()), the im structure is leaked.

Fix this by using ip_ma_put(im) instead of refcount_dec(&im->refcnt),
and deferring the put until after the spinlock is released.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/igmp.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/igmp.c b/net/ipv4/igmp.c
index f5f9763895641bf86bfcf9fd7fd7b06012fa4ece..957013e7f7a25f32d848822caab322147db3dc6e 100644
--- a/net/ipv4/igmp.c
+++ b/net/ipv4/igmp.c
@@ -217,13 +217,18 @@ static void ip_sf_list_clear_all(struct ip_sf_list *psf)
 
 static void igmp_stop_timer(struct ip_mc_list *im)
 {
+	bool put = false;
+
 	spin_lock_bh(&im->lock);
 	if (timer_delete(&im->timer))
-		refcount_dec(&im->refcnt);
+		put = true;
 	WRITE_ONCE(im->tm_running, 0);
 	WRITE_ONCE(im->reporter, 0);
 	im->unsolicit_count = 0;
 	spin_unlock_bh(&im->lock);
+
+	if (put)
+		ip_ma_put(im);
 }
 
 /* It must be called with locked im->lock */
@@ -266,6 +271,8 @@ static void igmp_ifc_start_timer(struct in_device *in_dev, int delay)
 
 static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 {
+	bool put = false;
+
 	spin_lock_bh(&im->lock);
 	im->unsolicit_count = 0;
 	if (timer_delete(&im->timer)) {
@@ -275,10 +282,13 @@ static void igmp_mod_timer(struct ip_mc_list *im, int max_delay)
 			spin_unlock_bh(&im->lock);
 			return;
 		}
-		refcount_dec(&im->refcnt);
+		put = true;
 	}
 	igmp_start_timer(im, max_delay);
 	spin_unlock_bh(&im->lock);
+
+	if (put)
+		ip_ma_put(im);
 }
 
 
-- 
2.55.0.rc0.799.gd6f94ed593-goog

Re: [PATCH v2 net 3/3] ipv4: igmp: Fix potential memory leaks in igmp_mod_timer() and igmp_stop_timer()

[Ido Schimmel]

On Sun, Jul 05, 2026 at 06:17:56PM +0000, Eric Dumazet wrote:
> When a timer is deleted and not re-armed in igmp_mod_timer(), or stopped
> in igmp_stop_timer(), the code currently decrements the reference counter
> of the multicast list entry @im using refcount_dec(&im->refcnt).
> 
> However, both functions can be called from the RCU reader path:
> - igmp_mod_timer() via igmp_heard_query() -> for_each_pmc_rcu()
> - igmp_stop_timer() via igmp_rcv() -> igmp_heard_report()
> 
> If the group im was concurrently removed from the list by ip_mc_dec_group(),
> its reference count might have already been decremented to 1.
> 
> In this case, timer_delete() succeeds, and refcount_dec() decrements
> the refcount from 1 to 0. Since refcount_dec() does not free the object
> when it hits 0 (unlike ip_ma_put()), the im structure is leaked.
> 
> Fix this by using ip_ma_put(im) instead of refcount_dec(&im->refcnt),
> and deferring the put until after the spinlock is released.
> 
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Reviewed-by: Ido Schimmel <idosch@nvidia.com>