netdev - Re: net: Fix inconsistent teardown and release of private netdev state.

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: <CAHmME9qjhOOGvOuXi0LEcEPE6Z5dK8MO-01dxb++RkAkpr91dw@mail.gmail.com>
Date:   Mon, 10 Jul 2017 04:07:02 +0200
From:   "Jason A. Donenfeld" <Jason@...c4.com>
To:     Cong Wang <xiyou.wangcong@...il.com>
Cc:     Netdev <netdev@...r.kernel.org>
Subject: Re: net: Fix inconsistent teardown and release of private netdev state.

On Sat, Jul 8, 2017 at 12:39 AM, Cong Wang <xiyou.wangcong@...il.com> wrote:
> On Thu, Jul 6, 2017 at 7:24 AM, Jason A. Donenfeld <Jason@...c4.com> wrote:
>>         list_add(&priv->list, &list_of_things);
>>
>>         ret = register_netdevice(); // if ret is < 0, then destruct above is automatically called
>>
>>         // RACE WITH LIST_ADD/LIST_DEL!! It's impossible to call list_add only after
>>         // things are brought up successfully. This is problematic.
>>
>>         if (!ret)
>>                 pr_info("Yay it worked!\n");
>
> I fail to understand what you mean by RACE here.
>
> Here you should already have RTNL lock, so it can't race with any other
> newlink() calls. In fact you can't acquire RTNL lock in your destructor
> since register_netdevice() already gets it. Perhaps you mean
> netdev_run_todo() calls it without RTNL lock?
>
> I don't know why you reorder the above list_add(), you can order it
> as it was before, aka, call it after register_netdevice(), but you have to
> init the priv->list now for the list_del() on error path.

The race is that there's a state in which priv->list is part of
list_of_things before the interface is actually successfully setup and
ready to go.

And no, it's not possible to order it _after_ register_netdevice,
since register_netdevice might call priv_destructor, and
priv_destructor calls list_del, so if it's not already on the list,
we'll OOPS. In otherwords, API problem.

To work around this shortcoming, I'm actually just assigning
dev->priv_device *after* a successful call to register_netdevice. This
seems to be working well and allows me to retain ± the old behavior.
Hopefully this is an okay way to go about things?

Jason