Received-SPF: softfail (google.com: domain of transitioning linux-wireless-owner@vger.kernel.org does not designate 23.128.96.19 as permitted sender) client-ip=23.128.96.19;
Message-ID: <1e4ef09dea2c22a03fa2303a0e909b99e8ed9028.camel@sipsolutions.net>
Subject: Re: [BUG] deadlock in nl80211_vendor_cmd
From:   Johannes Berg <johannes@sipsolutions.net>
To:     William McVicker <willmcvicker@google.com>
Cc:     Jakub Kicinski <kuba@kernel.org>, linux-wireless@vger.kernel.org,
        Marek Szyprowski <m.szyprowski@samsung.com>,
        Kalle Valo <kvalo@codeaurora.org>,
        "David S. Miller" <davem@davemloft.net>, netdev@vger.kernel.org,
        Amitkumar Karwar <amitkarwar@gmail.com>,
        Ganapathi Bhat <ganapathi.bhat@nxp.com>,
        Xinming Hu <huxinming820@gmail.com>, kernel-team@android.com,
        Paolo Abeni <pabeni@redhat.com>
Date:   Fri, 25 Mar 2022 22:54:38 +0100
In-Reply-To: <976e8cf697c7e5bc3a752e758a484b69a058710a.camel@sipsolutions.net>
References: <0000000000009e9b7105da6d1779@google.com>
         <99eda6d1dad3ff49435b74e539488091642b10a8.camel@sipsolutions.net>
         <5d5cf050-7de0-7bad-2407-276970222635@quicinc.com>
         <YjpGlRvcg72zNo8s@google.com>
         <dc556455-51a2-06e8-8ec5-b807c2901b7e@quicinc.com>
         <Yjzpo3TfZxtKPMAG@google.com>
         <19e12e6b5f04ba9e5b192001fbe31a3fc47d380a.camel@sipsolutions.net>
         <20220325094952.10c46350@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
         <f4f8a27dc07c1adaab470fde302ed841113e6b7f.camel@sipsolutions.net>
         <Yj4FFIXi//ivQC3X@google.com> <Yj4ntUejxaPhrM5b@google.com>
         <976e8cf697c7e5bc3a752e758a484b69a058710a.camel@sipsolutions.net>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.42.4 (3.42.4-1.fc35) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Fri, 2022-03-25 at 22:16 +0100, Johannes Berg wrote:
> 
> > Thread 1                         Thread 2
> >  nl80211_pre_doit():
> >    rtnl_lock()
> >    wiphy_lock()                   nl80211_pre_doit():
> >                                     rtnl_lock() // blocked by Thread 1
> >  nl80211_vendor_cmd():
> >    doit()
> >      cfg80211_unregister_netdevice()
> >    rtnl_unlock():
> >      netdev_run_todo():
> >        __rtnl_unlock()
> >                                     <got RTNL lock>
> >                                     wiphy_lock() // blocked by Thread 1
> >        rtnl_lock(); // DEADLOCK
> >  nl80211_post_doit():
> >    wiphy_unlock();
> 
> 
> Right, this is what I had discussed in my other mails.
> 
> Basically, you're actually doing (some form of) unregister_netdevice()
> before rtnl_unlock().
> 
> Clearly this isn't possible in cfg80211 itself.
> 
> However, I couldn't entirely discount the possibility that this is
> possible:
> 
> Thread 1                   Thread 2
>                             rtnl_lock()
>                             unregister_netdevice()
>                             __rtnl_unlock()
> rtnl_lock()
> wiphy_lock()
> netdev_run_todo()
>  __rtnl_unlock()
>  // list not empty now    
>  // because of thread 2     rtnl_lock()
>  rtnl_lock()
>                             wiphy_lock()
> 
> ** DEADLOCK **
> 
> 
> Given my other discussion with Jakub though, it seems that we can indeed
> make sure that this cannot happen, and then this scenario is impossible
> without the unregistration you're doing.
> 

I just sent a patch for this then, forgot to CC everyone:

https://lore.kernel.org/r/20220325225055.37e89a72f814.Ic73d206e217db20fd22dcec14fe5442ca732804b@changeid

But basically it changes nothing, just adds a WARN_ON with documentation
ensuring that the invariant never breaks, i.e. that Thread 2 can't
happen.

And maybe I should've written that with 3 Threads, but the setup of
unregister_netdevice()/__rtnl_unlock() could happen anywhere in the
system anyway.


johannes