netlink: fix potential deadlock in netlink_set_err()

syzbot reported a possible deadlock in netlink_set_err() [1]

A similar issue was fixed in commit 1d482e66 ("netlink: disable IRQs
for netlink_lock_table()") in netlink_lock_table()

This patch adds IRQ safety to netlink_set_err() and __netlink_diag_dump()
which were not covered by cited commit.

[1]

WARNING: possible irq lock inversion dependency detected
6.4.0-rc6-syzkaller-00240-g4e9f0ec3 #0 Not tainted

syz-executor.2/23011 just changed the state of lock:
ffffffff8e1a7a58 (nl_table_lock){.+.?}-{2:2}, at: netlink_set_err+0x2e/0x3a0 net/netlink/af_netlink.c:1612
but this lock was taken by another, SOFTIRQ-safe lock in the past:
 (&local->queue_stop_reason_lock){..-.}-{2:2}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
 Possible interrupt unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(nl_table_lock);
                               local_irq_disable();
                               lock(&local->queue_stop_reason_lock);
                               lock(nl_table_lock);
  <Interrupt>
    lock(&local->queue_stop_reason_lock);

 *** DEADLOCK ***

Fixes: 1d482e66 ("netlink: disable IRQs for netlink_lock_table()")
Reported-by: syzbot+a7d200a347f912723e5c@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?extid=a7d200a347f912723e5c
Link: https://lore.kernel.org/netdev/000000000000e38d1605fea5747e@google.com/T/#uSigned-off-by: Eric Dumazet <edumazet@google.com>
Cc: Johannes Berg <johannes.berg@intel.com>
Link: https://lore.kernel.org/r/20230621154337.1668594-1-edumazet@google.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

net/netlink/af_netlink.c

@@ -1600,6 +1600,7 @@ static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
 {
 	struct netlink_set_err_data info;
+	unsigned long flags;
 	struct sock *sk;
 	int ret = 0;
 
@@ -1609,12 +1610,12 @@ int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
 	/* sk->sk_err wants a positive error value */
 	info.code = -code;
 
-	read_lock(&nl_table_lock);
+	read_lock_irqsave(&nl_table_lock, flags);
 
 	sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
 		ret += do_one_set_err(sk, &info);
 
-	read_unlock(&nl_table_lock);
+	read_unlock_irqrestore(&nl_table_lock, flags);
 	return ret;
 }
 EXPORT_SYMBOL(netlink_set_err);

net/netlink/diag.c

@@ -94,6 +94,7 @@ static int __netlink_diag_dump(struct sk_buff *skb, struct netlink_callback *cb,
 	struct net *net = sock_net(skb->sk);
 	struct netlink_diag_req *req;
 	struct netlink_sock *nlsk;
+	unsigned long flags;
 	struct sock *sk;
 	int num = 2;
 	int ret = 0;
@@ -152,7 +153,7 @@ static int __netlink_diag_dump(struct sk_buff *skb, struct netlink_callback *cb,
 	num++;
 
 mc_list:
-	read_lock(&nl_table_lock);
+	read_lock_irqsave(&nl_table_lock, flags);
 	sk_for_each_bound(sk, &tbl->mc_list) {
 		if (sk_hashed(sk))
 			continue;
@@ -173,7 +174,7 @@ static int __netlink_diag_dump(struct sk_buff *skb, struct netlink_callback *cb,
 		}
 		num++;
 	}
-	read_unlock(&nl_table_lock);
+	read_unlock_irqrestore(&nl_table_lock, flags);
 
 done:
 	cb->args[0] = num;