Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sat, 06 Feb 2021 09:28:51 +0100
Message-ID: <s5hlfc11tx8.wl-tiwai@suse.de>
From:   Takashi Iwai <tiwai@suse.de>
To:     Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Cc:     Hillf Danton <hdanton@sina.com>, zonque@gmail.com,
        LKML <linux-kernel@vger.kernel.org>, alsa-devel@alsa-project.org,
        linux-usb@vger.kernel.org
Subject: Re: BUG: KASAN: use-after-free in snd_complete_urb+0x109e/0x1740 [snd_usb_audio] (5.11-rc6)
In-Reply-To: <s5hpn1d1vt6.wl-tiwai@suse.de>
References: <CABXGCsOposU1A_HavA_jmtkJMKhDZgh5m1b_YJK1Es5wyE-hZw@mail.gmail.com>
        <20210206054533.120-1-hdanton@sina.com>
        <s5hpn1d1vt6.wl-tiwai@suse.de>
User-Agent: Wanderlust/2.15.9 (Almost Unreal) SEMI/1.14.6 (Maruoka)
 FLIM/1.14.9 (=?UTF-8?B?R29qxY0=?=) APEL/10.8 Emacs/25.3
 (x86_64-suse-linux-gnu) MULE/6.0 (HANACHIRUSATO)
MIME-Version: 1.0 (generated by SEMI 1.14.6 - "Maruoka")
Content-Type: text/plain; charset=US-ASCII
Precedence: bulk

On Sat, 06 Feb 2021 08:48:05 +0100,
Takashi Iwai wrote:
> 
> On Sat, 06 Feb 2021 06:45:32 +0100,
> Hillf Danton wrote:
> > 
> > Due to the reconnecting key word mentioned, no fix to
> > d0f09d1e4a88 ("ALSA: usb-audio: Refactoring endpoint URB deactivation")
> > will be added.
> > 
> > What is added is to capture EP_FLAG_STOPPING and remove the one
> > second wait limit if the reconnecting acts may make it easier to
> > repro the uaf. The diff is only for idea show.
> 
> If my understanding is right, this won't change.  The problem is
> rather the lack of this function call itself, i.e. the missing
> synchronization for the stream stop.
> 
> It worked casually in the past because the endpoint resource is
> released at a later point that is after all streams are really closed.
> Now it's released earlier and hitting the UAF.

... and reading the code in a closer look, my guess was also wrong.
The sync should have happened in snd_usb_endpoint_release(), and this
didn't change for quite some time.  So my previous fix won't be
effective, too, I'm afraid.  (And Hilif's patch won't help, either; if
it's effective, there must have been a timeout error in the original
case.)

That said, I don't think this is a newly introduced regression, and
race the condition could be in a hairy detail.

Mikhail, can you reproduce this bug reliably?


Takashi