Re: [PATCH net v4] gtp: serialize PDP context updates

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Fri, Aug 14, 2026 at 11:15:26AM +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. The issue was found through static analysis and
> reproduced on a KASAN-enabled kernel by a simple 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 gtp_pdp_add(), gtp_genl_del_pdp(), and gtp_dellink() with a
> shared mutex. With this change, the same reproducer completed 50 races
> on a KASAN- and lockdep-enabled kernel without a KASAN report, oops, or
> lockdep warning.
> 
> Fixes: 459aa660eb1d ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)")
> Cc: stable@vger.kernel.org
> Assisted-by: Codex:gpt-5
> Signed-off-by: Qing Ming <a0yami@mailbox.org>
> ---
> v4:
> - Replace guard() with plain mutex_lock()/mutex_unlock() in the
>   GTP_CMD_NEWPDP path.
> - Narrow the RCU read-side section in gtp_genl_del_pdp() to
>   gtp_find_pdp(); gtp_find_dev() uses dev_get_by_index_rcu().
> - Add mutex-aware lockdep conditions to the shared PDP lookup helpers.
> 
> v3: https://lore.kernel.org/netdev/20260811022012.5416-1-a0yami@mailbox.org/
>  drivers/net/gtp.c | 25 +++++++++++++++++++------
>  1 file changed, 19 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/gtp.c b/drivers/net/gtp.c
> index 9a12cc53da00..ea83a6b6aa1a 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>
> @@ -108,6 +109,7 @@ struct gtp_net {
>  };
>  
>  static u32 gtp_h_initval;
> +static DEFINE_MUTEX(gtp_pdp_lock);
>  
>  static struct genl_family gtp_genl_family;
>  
> @@ -151,7 +153,8 @@ static struct pdp_ctx *gtp0_pdp_find(struct gtp_dev *gtp, u64 tid, u16 family)
>  
>  	head = &gtp->tid_hash[gtp0_hashfn(tid) % gtp->hash_size];
>  
> -	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
> +	hlist_for_each_entry_rcu(pdp, head, hlist_tid,
> +				 lockdep_is_held(&gtp_pdp_lock)) {
>  		if (pdp->af == family &&
>  		    pdp->gtp_version == GTP_V0 &&
>  		    pdp->u.v0.tid == tid)
> @@ -168,7 +171,8 @@ static struct pdp_ctx *gtp1_pdp_find(struct gtp_dev *gtp, u32 tid, u16 family)
>  
>  	head = &gtp->tid_hash[gtp1u_hashfn(tid) % gtp->hash_size];
>  
> -	hlist_for_each_entry_rcu(pdp, head, hlist_tid) {
> +	hlist_for_each_entry_rcu(pdp, head, hlist_tid,
> +				 lockdep_is_held(&gtp_pdp_lock)) {
>  		if (pdp->af == family &&
>  		    pdp->gtp_version == GTP_V1 &&
>  		    pdp->u.v1.i_tei == tid)
> @@ -185,7 +189,8 @@ static struct pdp_ctx *ipv4_pdp_find(struct gtp_dev *gtp, __be32 ms_addr)
>  
>  	head = &gtp->addr_hash[ipv4_hashfn(ms_addr) % gtp->hash_size];
>  
> -	hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
> +	hlist_for_each_entry_rcu(pdp, head, hlist_addr,
> +				 lockdep_is_held(&gtp_pdp_lock)) {
>  		if (pdp->af == AF_INET &&
>  		    pdp->ms.addr.s_addr == ms_addr)
>  			return pdp;
> @@ -220,7 +225,8 @@ static struct pdp_ctx *ipv6_pdp_find(struct gtp_dev *gtp,
>  
>  	head = &gtp->addr_hash[ipv6_hashfn(ms_addr) % gtp->hash_size];
>  
> -	hlist_for_each_entry_rcu(pdp, head, hlist_addr) {
> +	hlist_for_each_entry_rcu(pdp, head, hlist_addr,
> +				 lockdep_is_held(&gtp_pdp_lock)) {
>  		if (pdp->af == AF_INET6 &&
>  		    ipv6_pdp_addr_equal(&pdp->ms.addr6, ms_addr))
>  			return pdp;
> @@ -1555,9 +1561,11 @@ 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(&gtp->list);
>  	unregister_netdevice_queue(dev, head);
> @@ -2053,7 +2061,9 @@ static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
>  		goto out_unlock;
>  	}
>  
> +	mutex_lock(&gtp_pdp_lock);
>  	pctx = gtp_pdp_add(gtp, sk, info)

This does not look fine, it should protect the entire section where
this pctx object is used.

> +	mutex_unlock(&gtp_pdp_lock);
>  	if (IS_ERR(pctx)) {
>  		err = PTR_ERR(pctx);
>  	} else {
> @@ -2134,9 +2144,12 @@ static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
>  	if (!info->attrs[GTPA_VERSION])
>  		return -EINVAL;
>  
> -	rcu_read_lock();
> +	mutex_lock(&gtp_pdp_lock);
>  
> +	/* gtp_find_dev() uses dev_get_by_index_rcu(). */
> +	rcu_read_lock();
>  	pctx = gtp_find_pdp(sock_net(skb->sk), info->attrs);

Same here.

> +	rcu_read_unlock();
>  	if (IS_ERR(pctx)) {
>  		err = PTR_ERR(pctx);
>  		goto out_unlock;
> @@ -2153,7 +2166,7 @@ static int gtp_genl_del_pdp(struct sk_buff *skb, struct genl_info *info)
>  	pdp_context_delete(pctx);
>  
>  out_unlock:
> -	rcu_read_unlock();
> +	mutex_unlock(&gtp_pdp_lock);
>  	return err;
>  }
>  
> 
> base-commit: 24ef02f934eeb48830cff6b739abc3c62b1d107b
> -- 
> 2.53.0
>