linux-kernel - Re: CVE-2021-3640 and the unlimited block of lock

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] [day] [month] [year] [list]

Date:   Sat, 28 Aug 2021 18:06:43 +0200
From:   Takashi Iwai <tiwai@...e.de>
To:     Luiz Augusto von Dentz <luiz.dentz@...il.com>
Cc:     "linux-bluetooth@...r.kernel.org" <linux-bluetooth@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: CVE-2021-3640 and the unlimited block of lock_sock()

On Fri, 27 Aug 2021 03:28:09 +0200,
Luiz Augusto von Dentz wrote:
> 
> Hi Takashi,
> 
> On Thu, Aug 26, 2021 at 3:29 AM Takashi Iwai <tiwai@...e.de> wrote:
> >
> > On Thu, 19 Aug 2021 17:46:39 +0200,
> > Takashi Iwai wrote:
> > >
> > > Hi,
> > >
> > > it seems that the recent fixes in bluetooth tree address most of
> > > issues in CVE-2021-3640 ("Use-After-Free vulnerability in function
> > > sco_sock_sendmsg()").  But there is still a problem left: although we
> > > cover the race with lock_sock() now, the lock may be blocked endlessly
> > > (as the task takes over with userfaultd), which result in the trigger
> > > of watchdog like:
> > >
> > > -- 8< --
> > > [   23.226767][    T7] Bluetooth: hci0: command 0x0419 tx timeout
> > > [  284.985881][ T1529] INFO: task poc:7603 blocked for more than 143 seconds.
> > > [  284.989134][ T1529]       Not tainted 5.13.0-rc4+ #48
> > > [  284.990098][ T1529] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> > > [  284.991705][ T1529] task:poc             state:D stack:13784 pid: 7603 ppid:  7593 flags:0x00000000
> > > [  284.993414][ T1529] Call Trace:
> > > [  284.994025][ T1529]  __schedule+0x32e/0xb90
> > > [  284.994842][ T1529]  ? __local_bh_enable_ip+0x72/0xe0
> > > [  284.995987][ T1529]  schedule+0x38/0xe0
> > > [  284.996723][ T1529]  __lock_sock+0xa1/0x130
> > > [  284.997434][ T1529]  ? finish_wait+0x80/0x80
> > > [  284.998150][ T1529]  lock_sock_nested+0x9f/0xb0
> > > [  284.998914][ T1529]  sco_conn_del+0xb1/0x1a0
> > > [  284.999619][ T1529]  ? sco_conn_del+0x1a0/0x1a0
> > > [  285.000361][ T1529]  sco_disconn_cfm+0x3a/0x60
> > > [  285.001116][ T1529]  hci_conn_hash_flush+0x95/0x130
> > > [  285.001921][ T1529]  hci_dev_do_close+0x298/0x680
> > > [  285.002687][ T1529]  ? up_write+0x12/0x130
> > > [  285.003367][ T1529]  ? vhci_close_dev+0x20/0x20
> > > [  285.004107][ T1529]  hci_unregister_dev+0x9f/0x240
> > > [  285.004886][ T1529]  vhci_release+0x35/0x70
> > > [  285.005602][ T1529]  __fput+0xdf/0x360
> > > [  285.006225][ T1529]  task_work_run+0x86/0xd0
> > > [  285.006927][ T1529]  exit_to_user_mode_prepare+0x267/0x270
> > > [  285.007824][ T1529]  syscall_exit_to_user_mode+0x19/0x60
> > > [  285.008694][ T1529]  do_syscall_64+0x42/0xa0
> > > [  285.009393][ T1529]  entry_SYSCALL_64_after_hwframe+0x44/0xae
> > > [  285.010321][ T1529] RIP: 0033:0x4065c7
> > > -- 8< --
> > >
> > > Is there any plan to address this?
> > >
> > > As a quick hack, I confirmed a workaround like below:
> > >
> > > -- 8< --
> > > --- a/net/core/sock.c
> > > +++ b/net/core/sock.c
> > > @@ -2628,7 +2628,7 @@ void __lock_sock(struct sock *sk)
> > >               prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
> > >                                       TASK_UNINTERRUPTIBLE);
> > >               spin_unlock_bh(&sk->sk_lock.slock);
> > > -             schedule();
> > > +             schedule_timeout(msecs_to_jiffies(10 * 1000));
> > >               spin_lock_bh(&sk->sk_lock.slock);
> > >               if (!sock_owned_by_user(sk))
> > >                       break;
> > > -- 8< --
> > >
> > > .... but I'm not sure whether it's the right way to go.
> >
> > Does anyone has an idea?
> 
> It seems that we need to rework some code so the functions affected by
> userfaultfd are not used with sock_lock held.

OK, now I tried a similar way like the commit 92c685dc5de0 to move the
memcpy_from_msg() call out of lock_sock(), and it seems working.
I'm going to submit the fix.


thanks,

Takashi