Re: [PATCH] nbd: Fix NULL pointer in flush_workqueue

[Sun Ke]

hi，Markus

在 2021/1/29 3:42, Markus Elfring 写道:

…

+++ b/drivers/block/nbd.c
@@ -2011,12 +2011,20 @@ static int nbd_genl_disconnect(struct sk_buff *skb, 
struct genl_info *info)
                       index);
                return -EINVAL;
        }
+       mutex_lock(&nbd->config_lock);
        if (!refcount_inc_not_zero(&nbd->refs)) {
                mutex_unlock(&nbd_index_mutex);
+               mutex_unlock(&nbd->config_lock);

Can an other function call order become relevant for the unlocking of these 
mutexes?

Do you think the nbd->config_lock  mutex here is useless?

                printk(KERN_ERR "nbd: device at index %d is going down\n",
                       index);

May such an error message be moved into the lock scope?

Sure.

                return -EINVAL;
        }
+       if (!nbd->recv_workq) {
+               mutex_unlock(&nbd->config_lock);
+               mutex_unlock(&nbd_index_mutex);
+               return -EINVAL;
+       }

How do you think about to connect the code from this if branch
with a jump target like “unlock” so that such statements would be shareable
for the desired exception handling?

OK, I will improve it in V2 patch.

+       mutex_unlock(&nbd->config_lock);
        mutex_unlock(&nbd_index_mutex);
        if (!refcount_inc_not_zero(&nbd->config_refs)) {
                nbd_put(nbd);

Regards,
Markus
.