Re: [PATCH net v2] net: vlan: fix a UAF in vlan_dev_real_dev()

On Thu, 18 Nov 2021 09:46:24 +0800 Ziyang Xuan (William) wrote:

quoted

quoted

quoted

I think we should move the dev_hold() to ndo_init(), otherwise 
it's hard to reason if destructor was invoked or not if
register_netdevice() errors out.  

Ziyang Xuan, do you intend to take care of this?
.  

I am reading the related processes according to the problem scenario.
And I will give a more clear sequence and root cause as soon as possible
by some necessary tests.

Okay, I still don't see the fix. 

Petr would you mind submitting since you have a repro handy?
.

The sequence for the problem maybe as following:

=============================================================
# create vlan device
vlan_newlink [assume real_dev refcnt == 2 just referenced by itself]
	register_vlan_dev
		register_netdevice(vlan_dev) [success]
		netdev_upper_dev_link [failed]
		unregister_netdevice(vlan_dev) [failure process]
		...
		netdev_run_todo [vlan_dev]
			vlan_dev_free(vlan_dev) [priv_destructor cb]
				dev_put(real_dev) [real_dev refcnt == 1]

# delete real device
unregister_netdevice(real_dev) [real_dev refcnt == 1]
	unregister_netdevice_many
		dev_put(real_dev) [real_dev refcnt == 0]
		net_set_todo(real_dev)
...
netdev_run_todo [real_dev]
	netdev_wait_allrefs(real_dev) [real_dev refcnt == 0]
		pr_emerg("unregister_netdevice: ...", ...)

I am thinking about how to fix the problem. priv_destructor() of the
net_device referenced not only by net_set_todo() but also failure process
in register_netdevice().

I need some time to test my some ideas. And anyone has good ideas, please
do not be stingy.

Thank you!