Re: [syzbot] [net?] KASAN: slab-use-after-free Write in __xfrm_state_delete

[Hillf Danton <hdanton@sina.com>]

> Date: Sun, 24 Aug 2025 12:21:23 -0700	[thread overview]
> syzbot has found a reproducer for the following issue on:
> 
> HEAD commit:    b1c92cdf5af3 Merge branch 'net-wangxun-complete-ethtool-co..
> git tree:       net-next
> console output: https://syzkaller.appspot.com/x/log.txt?x=1411b062580000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=67b99ceb67d33475
> dashboard link: https://syzkaller.appspot.com/bug?extid=a25ee9d20d31e483ba7b
> compiler:       Debian clang version 20.1.7 (++20250616065708+6146a88f6049-1~exp1~20250616065826.132), Debian LLD 20.1.7
> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=14221862580000
> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=159fba34580000

#syz test

--- x/include/net/xfrm.h
+++ y/include/net/xfrm.h
@@ -202,6 +202,7 @@ struct xfrm_state {
 
 	refcount_t		refcnt;
 	spinlock_t		lock;
+	int deleted;
 
 	u32			pcpu_num;
 	struct xfrm_id		id;
--- x/net/xfrm/xfrm_state.c
+++ y/net/xfrm/xfrm_state.c
@@ -615,6 +615,7 @@ static void xfrm_state_gc_destroy(struct
 		put_page(x->xfrag.page);
 	xfrm_dev_state_free(x);
 	security_xfrm_state_free(x);
+	xfrm_state_delete(x);
 	xfrm_state_free(x);
 }
 
@@ -812,10 +813,16 @@ int __xfrm_state_delete(struct xfrm_stat
 	struct net *net = xs_net(x);
 	int err = -ESRCH;
 
-	if (x->km.state != XFRM_STATE_DEAD) {
-		x->km.state = XFRM_STATE_DEAD;
+	for (;;) {
+		if (x->km.state != XFRM_STATE_DEAD)
+			x->km.state = XFRM_STATE_DEAD;
 
 		spin_lock(&net->xfrm.xfrm_state_lock);
+		if (x->deleted) {
+			spin_unlock(&net->xfrm.xfrm_state_lock);
+			return 0;
+		}
+		x->deleted++;
 		list_del(&x->km.all);
 		hlist_del_rcu(&x->bydst);
 		hlist_del_rcu(&x->bysrc);
@@ -833,14 +840,7 @@ int __xfrm_state_delete(struct xfrm_stat
 		spin_unlock(&net->xfrm.xfrm_state_lock);
 
 		xfrm_dev_state_delete(x);
-
 		xfrm_state_delete_tunnel(x);
-
-		/* All xfrm_state objects are created by xfrm_state_alloc.
-		 * The xfrm_state_alloc call gives a reference, and that
-		 * is what we are dropping here.
-		 */
-		xfrm_state_put(x);
 		err = 0;
 	}
 
@@ -929,22 +929,29 @@ int xfrm_state_flush(struct net *net, u8
 	err = -ESRCH;
 	for (i = 0; i <= net->xfrm.state_hmask; i++) {
 		struct xfrm_state *x;
+		bool dead;
 restart:
 		hlist_for_each_entry(x, net->xfrm.state_bydst+i, bydst) {
 			if (!xfrm_state_kern(x) &&
 			    xfrm_id_proto_match(x->id.proto, proto)) {
-				xfrm_state_hold(x);
+				dead = x->km.state == XFRM_STATE_DEAD;
+				x->km.state = XFRM_STATE_DEAD;
 				spin_unlock_bh(&net->xfrm.xfrm_state_lock);
 
-				err = xfrm_state_delete(x);
+				if (dead) {
+			flush:
+					schedule_work(&xfrm_state_gc_work);
+					flush_work(&xfrm_state_gc_work);
+					spin_lock_bh(&net->xfrm.xfrm_state_lock);
+					goto restart;
+				}
+				err = 0;
 				xfrm_audit_state_delete(x, err ? 0 : 1,
 							task_valid);
 				xfrm_state_put(x);
 				if (!err)
 					cnt++;
-
-				spin_lock_bh(&net->xfrm.xfrm_state_lock);
-				goto restart;
+				goto flush;
 			}
 		}
 	}
@@ -1863,6 +1870,7 @@ static struct xfrm_state *__find_acq_cor
 		x->mark.m = m->m;
 		x->lft.hard_add_expires_seconds = net->xfrm.sysctl_acq_expires;
 		xfrm_state_hold(x);
+		xfrm_state_hold(x);
 		hrtimer_start(&x->mtimer,
 			      ktime_set(net->xfrm.sysctl_acq_expires, 0),
 			      HRTIMER_MODE_REL_SOFT);
--