[PATCH] x25: defer socket cleanup on link termination to process context

From: Shuai Liu

Date: Wed Nov 18 2026 - 20:24:28 EST


x25_kill_by_neigh() walks all sockets bound to a neighbour and calls
lock_sock() and x25_disconnect() on each one. It is invoked from
x25_link_terminated() and x25_link_control(), which both run in softirq
context when the LAPB link goes down or a restart frame arrives, e.g.
from the lapbether NAPI poll path.

lock_sock() takes a mutex in its slow path, so calling it from softirq
sleeps in atomic context:

BUG: sleeping function called from invalid context at net/core/sock.c:3839
in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 14, name: ksoftirqd/0
...
lock_sock_nested+0x66/0x70
x25_kill_by_neigh+0x90/0x110
x25_link_terminated+0x18/0x50
x25_lapb_receive_frame+0x1c2/0x1f0

and lockdep flags the sk_lock-AF_X25 usage violation because the same
lock is registered in process context by x25_bind():

WARNING: inconsistent lock state
inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
ksoftirqd/0/14 [HC0[0]:SC1[1]:HE1:SE0] takes:
...
x25_kill_by_neigh+0x90/0x110
x25_link_terminated+0x18/0x50
x25_lapb_receive_frame+0x1c2/0x1f0

Defer the neighbour socket cleanup to a per-neighbour work item so it
always runs in process context. The work item holds a reference on the
neighbour and on its net_device, so the deferred cleanup cannot race
with neighbour or device teardown; both references are dropped again if
queueing fails because the work is already pending. A dedicated
workqueue is drained in x25_exit() so that the module stays loaded until
pending work has run, and NETDEV_DOWN flushes the work to keep device
teardown synchronous as before.

Fixes: 7781607938c8 ("net/x25: Fix null-ptr-deref caused by x25_disconnect")
Reported-by: syzbot+300375dfd44665f063a6@xxxxxxxxxxxxxxxxxxxxxxxxx
Closes: https://syzkaller.appspot.com/bug?extid=300375dfd44665f063a6
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Shuai Liu <847807867@xxxxxx>
---

Changes in v2:
- Use a dedicated workqueue and drain it in x25_exit() so the module
stays loaded until pending work has run (Sashiko)
- Hold a reference on the neighbour's net_device for the deferred work,
and drop it again when queueing fails because the work is already
pending (Sashiko)
- Flush the work in NETDEV_DOWN to keep device teardown synchronous
(Sashiko)
- Explicitly request WQ_PERCPU to satisfy the workqueue API
- Add Fixes tag and Cc stable

include/net/x25.h | 3 +++
net/x25/af_x25.c | 11 ++++++++++-
net/x25/x25_link.c | 28 +++++++++++++++++++++++++---
3 files changed, 38 insertions(+), 4 deletions(-)

diff --git a/include/net/x25.h b/include/net/x25.h
index 414f3fd993454..125803d4f0ee7 100644
--- a/include/net/x25.h
+++ b/include/net/x25.h
@@ -13,6 +13,7 @@
#include <linux/x25.h>
#include <linux/slab.h>
#include <linux/refcount.h>
+#include <linux/workqueue.h>
#include <net/sock.h>

#define X25_ADDR_LEN 16
@@ -143,6 +144,7 @@ struct x25_neigh {
struct sk_buff_head queue;
unsigned long t20;
struct timer_list t20timer;
+ struct work_struct kill_work;
unsigned long global_facil_mask;
refcount_t refcnt;
};
@@ -317,6 +319,7 @@ extern struct list_head x25_forward_list;
extern rwlock_t x25_forward_list_lock;
extern struct list_head x25_neigh_list;
extern rwlock_t x25_neigh_list_lock;
+extern struct workqueue_struct *x25_wq;

int x25_proc_init(void);
void x25_proc_exit(void);
diff --git a/net/x25/af_x25.c b/net/x25/af_x25.c
index f45c9ec7d2c68..a1a8bc203a24f 100644
--- a/net/x25/af_x25.c
+++ b/net/x25/af_x25.c
@@ -222,6 +222,7 @@ static int x25_device_event(struct notifier_block *this, unsigned long event,
nb = x25_get_neigh(dev);
if (nb) {
x25_link_terminated(nb);
+ flush_work(&nb->kill_work);
x25_neigh_put(nb);
}
x25_route_device_down(dev);
@@ -1790,9 +1791,13 @@ static int __init x25_init(void)
{
int rc;

+ x25_wq = alloc_workqueue("x25", WQ_PERCPU, 0);
+ if (!x25_wq)
+ return -ENOMEM;
+
rc = proto_register(&x25_proto, 0);
if (rc)
- goto out;
+ goto out_wq;

rc = sock_register(&x25_family_ops);
if (rc)
@@ -1825,6 +1830,8 @@ static int __init x25_init(void)
sock_unregister(AF_X25);
out_proto:
proto_unregister(&x25_proto);
+out_wq:
+ destroy_workqueue(x25_wq);
goto out;
}
module_init(x25_init);
@@ -1843,6 +1850,8 @@ static void __exit x25_exit(void)

sock_unregister(AF_X25);
proto_unregister(&x25_proto);
+
+ destroy_workqueue(x25_wq);
}
module_exit(x25_exit);

diff --git a/net/x25/x25_link.c b/net/x25/x25_link.c
index d1e8cd81c2143..424044c1cf974 100644
--- a/net/x25/x25_link.c
+++ b/net/x25/x25_link.c
@@ -31,6 +31,8 @@
LIST_HEAD(x25_neigh_list);
DEFINE_RWLOCK(x25_neigh_list_lock);

+struct workqueue_struct *x25_wq;
+
static void x25_t20timer_expiry(struct timer_list *);

static void x25_transmit_restart_confirmation(struct x25_neigh *nb);
@@ -58,6 +60,25 @@ static inline void x25_stop_t20timer(struct x25_neigh *nb)
timer_delete(&nb->t20timer);
}

+static void x25_kill_by_neigh_work(struct work_struct *work)
+{
+ struct x25_neigh *nb = container_of(work, struct x25_neigh, kill_work);
+
+ x25_kill_by_neigh(nb);
+ dev_put(nb->dev);
+ x25_neigh_put(nb);
+}
+
+static void x25_defer_kill_by_neigh(struct x25_neigh *nb)
+{
+ x25_neigh_hold(nb);
+ dev_hold(nb->dev);
+ if (!queue_work(x25_wq, &nb->kill_work)) {
+ dev_put(nb->dev);
+ x25_neigh_put(nb);
+ }
+}
+
/*
* This handles all restart and diagnostic frames.
*/
@@ -82,7 +103,7 @@ void x25_link_control(struct sk_buff *skb, struct x25_neigh *nb,
break;
case X25_LINK_STATE_3:
/* clear existing virtual calls */
- x25_kill_by_neigh(nb);
+ x25_defer_kill_by_neigh(nb);

x25_transmit_restart_confirmation(nb);
break;
@@ -97,7 +118,7 @@ void x25_link_control(struct sk_buff *skb, struct x25_neigh *nb,
break;
case X25_LINK_STATE_3:
/* clear existing virtual calls */
- x25_kill_by_neigh(nb);
+ x25_defer_kill_by_neigh(nb);

x25_transmit_restart_request(nb);
nb->state = X25_LINK_STATE_2;
@@ -254,7 +275,7 @@ void x25_link_terminated(struct x25_neigh *nb)
x25_stop_t20timer(nb);

/* Out of order: clear existing virtual calls (X.25 03/93 4.6.3) */
- x25_kill_by_neigh(nb);
+ x25_defer_kill_by_neigh(nb);
}

/*
@@ -269,6 +290,7 @@ void x25_link_device_up(struct net_device *dev)

skb_queue_head_init(&nb->queue);
timer_setup(&nb->t20timer, x25_t20timer_expiry, 0);
+ INIT_WORK(&nb->kill_work, x25_kill_by_neigh_work);

dev_hold(dev);
nb->dev = dev;
--
2.43.0