linux-kernel - Re: [PATCH net] macvlan: Fix use-after-free in macvlan_common

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20260108191930.145655-1-boudewijn@delta-utec.com>
Date: Thu,  8 Jan 2026 20:19:30 +0100
From: Boudewijn van der Heide <boudewijn@...ta-utec.com>
To: edumazet@...gle.com
Cc: netdev@...r.kernel.org,
	andrew+netdev@...n.ch,
	davem@...emloft.net,
	kuba@...nel.org,
	pabeni@...hat.com,
	linux-kernel@...r.kernel.org,
	syzbot+7182fbe91e58602ec1fe@...kaller.appspotmail.com
Subject: Re: [PATCH net] macvlan: Fix use-after-free in macvlan_common_newlink

Hi Eric,

Thanks for the patch, I agree it improves safety for source-entry deletion.

However, I believe it does not fix the specific KASAN report from syzbot, 
which indicates a UAF on the struct macvlan_port itself, not a source entry.

The report shows the freed object is in kmalloc-cg-4k (size 4096):

The buggy address belongs to the object at ffff888030eda000
which belongs to the cache kmalloc-cg-4k of size 4096

struct macvlan_port fits this size (due to the large hash tables), 
whereas struct macvlan_source_entry is much smaller. 
The crash happens at offset 3580, which corresponds to the vlan_source_hash array inside the port:

The race occurs in the register_netdevice() error path in macvlan_common_newlink():

1. netdev_rx_handler_register() succeeds 
2. register_netdevice() fails.
3. macvlan_port_destroy() is called, performing a synchronous kfree(port).

If a packet arrives during step 3, 
macvlan_handle_frame() accesses port->vlan_source_hash (via macvlan_forward_source),
after it is freed.

My patch restores the kfree_rcu behavior for the port (removed in 2016). 
I believe both fixes are needed: yours for the source entries, and mine for the port itself.

Thanks,
Boudewijn