PDP contexts can be deleted through GTP_CMD_DELPDP or while the GTP
network device is being unregistered. The latter is serialized by RTNL,
but the generic-netlink delete path only holds RCU.

Running both paths concurrently can therefore make both paths delete the
same PDP context. On a KASAN-enabled kernel, a reproducer racing DELPDP
against RTM_DELLINK triggered:

  Oops: general protection fault, probably for non-canonical address
  KASAN: maybe wild-memory-access in range 
         [0xdead000000000120-0xdead000000000127]
  RIP: gtp_genl_del_pdp+0x1c1/0x420 [gtp]
  RBP: dead000000000122

The second deletion dereferenced the poisoned hlist pprev pointer.

Take RTNL around the DELPDP lookup and deletion so that PDP creation,
generic-netlink deletion and link teardown use the same serialization
domain.

Fixes: 459aa660eb1d ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)")
Signed-off-by: Qing Ming <a0yami@mailbox.org>
---
 drivers/net/gtp.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c
index 427b91aca50d..7c4a99e2fd52 100644
--- a/drivers/net/gtp.c
+++ b/drivers/net/gtp.c
@@ -2127,6 +2127,8 @@ static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
 	if (!info->attrs[GTPA_VERSION])
 		return -EINVAL;
 
+	rtnl_lock();
+
 	rcu_read_lock();
 
 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
@@ -2147,6 +2149,7 @@ static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
 
 out_unlock:
 	rcu_read_unlock();
+	rtnl_unlock();
 	return err;
 }
 
-- 
2.53.0

Hi,

On Thu, Aug 06, 2026 at 10:32:26AM +0800, Qing Ming wrote:
> PDP contexts can be deleted through GTP_CMD_DELPDP or while the GTP
> network device is being unregistered. The latter is serialized by RTNL,
> but the generic-netlink delete path only holds RCU.
> 
> Running both paths concurrently can therefore make both paths delete the
> same PDP context. On a KASAN-enabled kernel, a reproducer racing DELPDP
> against RTM_DELLINK triggered:
> 
>   Oops: general protection fault, probably for non-canonical address
>   KASAN: maybe wild-memory-access in range 
>          [0xdead000000000120-0xdead000000000127]
>   RIP: gtp_genl_del_pdp+0x1c1/0x420 [gtp]
>   RBP: dead000000000122
> 
> The second deletion dereferenced the poisoned hlist pprev pointer.
> 
> Take RTNL around the DELPDP lookup and deletion so that PDP creation,
> generic-netlink deletion and link teardown use the same serialization
> domain.

Could please you instead add a mutex to a use it to protect PDP ctx
updates? genetlink mutex is not enough, and I'd prefer not to fix this
with the rtnl lock.

Please, use this new mutex from gtp_dellink() path to protect the
iteration over the hashtable.

gtp_genl_del_pdp() must use it too as well as gtp_pdp_add().

Thanks.

On Thu, Aug 06, 2026 at 10:32:26AM +0800, Qing Ming wrote:
> PDP contexts can be deleted through GTP_CMD_DELPDP or while the GTP
> network device is being unregistered. The latter is serialized by RTNL,
> but the generic-netlink delete path only holds RCU.
> 
> Running both paths concurrently can therefore make both paths delete the
> same PDP context. On a KASAN-enabled kernel, a reproducer racing DELPDP
> against RTM_DELLINK triggered:
> 
>   Oops: general protection fault, probably for non-canonical address
>   KASAN: maybe wild-memory-access in range 
>          [0xdead000000000120-0xdead000000000127]
>   RIP: gtp_genl_del_pdp+0x1c1/0x420 [gtp]
>   RBP: dead000000000122
> 
> The second deletion dereferenced the poisoned hlist pprev pointer.
> 
> Take RTNL around the DELPDP lookup and deletion so that PDP creation,
> generic-netlink deletion and link teardown use the same serialization
> domain.
> 
> Fixes: 459aa660eb1d ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)")
> Signed-off-by: Qing Ming <a0yami@mailbox.org>

Hi,

I think it would be good to mention how this problem was found,
and to what extent it has been tested.

You may also want to consider adding an Assisted-by tag as appropriate.

Hi Simon,

On Fri, Aug 07, 2026 at 02:13:35PM +0100, Simon Horman wrote:
> On Thu, Aug 06, 2026 at 10:32:26AM +0800, Qing Ming wrote:
> > PDP contexts can be deleted through GTP_CMD_DELPDP or while the GTP
> > network device is being unregistered. The latter is serialized by RTNL,
> > but the generic-netlink delete path only holds RCU.
> > 
> > Running both paths concurrently can therefore make both paths delete the
> > same PDP context. On a KASAN-enabled kernel, a reproducer racing DELPDP
> > against RTM_DELLINK triggered:
> > 
> >   Oops: general protection fault, probably for non-canonical address
> >   KASAN: maybe wild-memory-access in range 
> >          [0xdead000000000120-0xdead000000000127]
> >   RIP: gtp_genl_del_pdp+0x1c1/0x420 [gtp]
> >   RBP: dead000000000122
> > 
> > The second deletion dereferenced the poisoned hlist pprev pointer.
> > 
> > Take RTNL around the DELPDP lookup and deletion so that PDP creation,
> > generic-netlink deletion and link teardown use the same serialization
> > domain.
> > 
> > Fixes: 459aa660eb1d ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)")
> > Signed-off-by: Qing Ming <a0yami@mailbox.org>
> 
> Hi,
> 
> I think it would be good to mention how this problem was found,
> and to what extent it has been tested.

I know what it is going on, I can take care of this, thanks!

> You may also want to consider adding an Assisted-by tag as appropriate.

PDP contexts can be deleted through GTP_CMD_DELPDP or while the GTP
network device is being unregistered. The latter is serialized by RTNL,
but the generic-netlink delete path only holds RCU.

Running both paths concurrently can therefore make both paths delete the
same PDP context. The issue was found through static analysis and
reproduced on a KASAN-enabled kernel by a two-thread program
racing GTP_CMD_DELPDP against RTM_DELLINK:

  Oops: general protection fault, probably for non-canonical address
  KASAN: maybe wild-memory-access in range
         [0xdead000000000120-0xdead000000000127]
  RIP: gtp_genl_del_pdp+0x1c1/0x420 [gtp]
  RBP: dead000000000122

The second deletion dereferenced the poisoned hlist pprev pointer.

Serialize PDP context updates with a mutex shared by gtp_pdp_add(),
gtp_genl_del_pdp(), and gtp_dellink().

Fixes: 459aa660eb1d ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)")
Assisted-by: Codex:gpt-5
Signed-off-by: Qing Ming <a0yami@mailbox.org>
---
v2:
- Use a dedicated mutex instead of RTNL.
- Protect PDP add, delete, and link teardown with the same mutex.
 drivers/net/gtp.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c
index 427b91aca50d..161d004e34b6 100644
--- a/drivers/net/gtp.c
+++ b/drivers/net/gtp.c
@@ -12,6 +12,7 @@
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 
 #include <linux/module.h>
+#include <linux/mutex.h>
 #include <linux/skbuff.h>
 #include <linux/udp.h>
 #include <linux/rculist.h>
@@ -105,6 +106,7 @@ struct gtp_net {
 };
 
 static u32 gtp_h_initval;
+static DEFINE_MUTEX(gtp_pdp_lock);
 
 static struct genl_family gtp_genl_family;
 
@@ -1546,9 +1548,11 @@ static void gtp_dellink(struct net_device *dev, struct list_head *head)
 	struct pdp_ctx *pctx;
 	int i;
 
+	mutex_lock(&gtp_pdp_lock);
 	for (i = 0; i < gtp->hash_size; i++)
 		hlist_for_each_entry_safe(pctx, next, &gtp->tid_hash[i], hlist_tid)
 			pdp_context_delete(pctx);
+	mutex_unlock(&gtp_pdp_lock);
 
 	list_del_rcu(&gtp->list);
 	unregister_netdevice_queue(dev, head);
@@ -1820,6 +1824,8 @@ static struct pdp_ctx *gtp_pdp_add(struct gtp_dev *gtp, struct sock *sk,
 	__be32 ms_addr;
 	int family;
 
+	guard(mutex)(&gtp_pdp_lock);
+
 	version = nla_get_u32(info->attrs[GTPA_VERSION]);
 
 	if (info->attrs[GTPA_FAMILY])
@@ -2127,6 +2133,8 @@ static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
 	if (!info->attrs[GTPA_VERSION])
 		return -EINVAL;
 
+	mutex_lock(&gtp_pdp_lock);
+
 	rcu_read_lock();
 
 	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);
@@ -2147,6 +2155,7 @@ static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
 
 out_unlock:
 	rcu_read_unlock();
+	mutex_unlock(&gtp_pdp_lock);
 	return err;
 }
 
-- 
2.53.0

On Sat,  8 Aug 2026 20:43:22 +0800 Qing Ming wrote:
> Subject: [PATCH net v2] gtp: serialize PDP context updates

This patch does not apply to net/main, please rebase and repost.