Re: [syzbot] [net?] [bpf?] possible deadlock in sock_hash_delete_elem (2)

[Hillf Danton <hdanton@sina.com>]

On Fri, 26 Apr 2024 23:08:19 -0700
> syzbot has found a reproducer for the following issue on:
> 
> HEAD commit:    b2ff42c6d3ab Merge tag 'for-netdev' of https://git.kernel...
> git tree:       bpf
> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=13c06aa0980000

#syz test https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf.git  b2ff42c6d3ab

--- x/net/core/sock_map.c
+++ y/net/core/sock_map.c
@@ -928,6 +928,7 @@ static void sock_hash_delete_from_link(s
 	spin_unlock_bh(&bucket->lock);
 }
 
+static DEFINE_PER_CPU(int, subclass);
 static long sock_hash_delete_elem(struct bpf_map *map, void *key)
 {
 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
@@ -935,6 +936,7 @@ static long sock_hash_delete_elem(struct
 	struct bpf_shtab_bucket *bucket;
 	struct bpf_shtab_elem *elem;
 	int ret = -ENOENT;
+	int *class;
 
 	if (irqs_disabled())
 		return -EOPNOTSUPP; /* locks here are hardirq-unsafe */
@@ -942,7 +944,10 @@ static long sock_hash_delete_elem(struct
 	hash = sock_hash_bucket_hash(key, key_size);
 	bucket = sock_hash_select_bucket(htab, hash);
 
-	spin_lock_bh(&bucket->lock);
+	local_bh_disable();
+	class = this_cpu_ptr(&subclass);
+	*class += 1;
+	spin_lock_nested(&bucket->lock, *class);
 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 	if (elem) {
 		hlist_del_rcu(&elem->node);
@@ -950,6 +955,7 @@ static long sock_hash_delete_elem(struct
 		sock_hash_free_elem(htab, elem);
 		ret = 0;
 	}
+	*class -= 1;
 	spin_unlock_bh(&bucket->lock);
 	return ret;
 }
--