Re: [RESEND 1/1] Bluetooth: Use chan_list_lock to protect the whole put/destroy invokation

[Luiz Augusto von Dentz]

Hi Eric, Lee,

On Mon, Jun 27, 2022 at 7:41 AM Eric Dumazet <edumazet@xxxxxxxxxx> wrote:
>
> On Wed, Jun 22, 2022 at 10:27 AM Lee Jones <lee.jones@xxxxxxxxxx> wrote:
> >
> > This change prevents a use-after-free caused by one of the worker
> > threads starting up (see below) *after* the final channel reference
> > has been put() during sock_close() but *before* the references to the
> > channel have been destroyed.
> >
> >   refcount_t: increment on 0; use-after-free.
> >   BUG: KASAN: use-after-free in refcount_dec_and_test+0x20/0xd0
> >   Read of size 4 at addr ffffffc114f5bf18 by task kworker/u17:14/705
> >
> >   CPU: 4 PID: 705 Comm: kworker/u17:14 Tainted: G S      W       4.14.234-00003-g1fb6d0bd49a4-dirty #28
> >   Hardware name: Qualcomm Technologies, Inc. SM8150 V2 PM8150 Google Inc. MSM sm8150 Flame DVT (DT)
> >   Workqueue: hci0 hci_rx_work
> >   Call trace:
> >    dump_backtrace+0x0/0x378
> >    show_stack+0x20/0x2c
> >    dump_stack+0x124/0x148
> >    print_address_description+0x80/0x2e8
> >    __kasan_report+0x168/0x188
> >    kasan_report+0x10/0x18
> >    __asan_load4+0x84/0x8c
> >    refcount_dec_and_test+0x20/0xd0
> >    l2cap_chan_put+0x48/0x12c
> >    l2cap_recv_frame+0x4770/0x6550
> >    l2cap_recv_acldata+0x44c/0x7a4
> >    hci_acldata_packet+0x100/0x188
> >    hci_rx_work+0x178/0x23c
> >    process_one_work+0x35c/0x95c
> >    worker_thread+0x4cc/0x960
> >    kthread+0x1a8/0x1c4
> >    ret_from_fork+0x10/0x18
> >
> > Cc: stable@xxxxxxxxxx
>
> When was the bug added ? (Fixes: tag please)
>
> > Cc: Marcel Holtmann <marcel@xxxxxxxxxxxx>
> > Cc: Johan Hedberg <johan.hedberg@xxxxxxxxx>
> > Cc: Luiz Augusto von Dentz <luiz.dentz@xxxxxxxxx>
> > Cc: "David S. Miller" <davem@xxxxxxxxxxxxx>
> > Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> > Cc: Jakub Kicinski <kuba@xxxxxxxxxx>
> > Cc: Paolo Abeni <pabeni@xxxxxxxxxx>
> > Cc: linux-bluetooth@xxxxxxxxxxxxxxx
> > Cc: netdev@xxxxxxxxxxxxxxx
> > Signed-off-by: Lee Jones <lee.jones@xxxxxxxxxx>
> > ---
> >  net/bluetooth/l2cap_core.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> > index ae78490ecd3d4..82279c5919fd8 100644
> > --- a/net/bluetooth/l2cap_core.c
> > +++ b/net/bluetooth/l2cap_core.c
> > @@ -483,9 +483,7 @@ static void l2cap_chan_destroy(struct kref *kref)
> >
> >         BT_DBG("chan %p", chan);
> >
> > -       write_lock(&chan_list_lock);
> >         list_del(&chan->global_l);
> > -       write_unlock(&chan_list_lock);
> >
> >         kfree(chan);
> >  }
> > @@ -501,7 +499,9 @@ void l2cap_chan_put(struct l2cap_chan *c)
> >  {
> >         BT_DBG("chan %p orig refcnt %u", c, kref_read(&c->kref));
> >
> > +       write_lock(&chan_list_lock);
> >         kref_put(&c->kref, l2cap_chan_destroy);
> > +       write_unlock(&chan_list_lock);
> >  }
> >  EXPORT_SYMBOL_GPL(l2cap_chan_put);
> >
> > --
> > 2.36.1.255.ge46751e96f-goog
> >
>
> I do not think this patch is correct.
>
> a kref does not need to be protected by a write lock.
>
> This might shuffle things enough to work around a particular repro you have.
>
> If the patch was correct why not protect kref_get() sides ?
>
> Before the &hdev->rx_work is scheduled (queue_work(hdev->workqueue,
> &hdev->rx_work),
> a reference must be taken.
>
> Then this reference must be released at the end of hci_rx_work() or
> when hdev->workqueue
> is canceled.
>
> This refcount is not needed _if_ the workqueue is properly canceled at
> device dismantle,
> in a synchronous way.
>
> I do not see this hdev->rx_work being canceled, maybe this is the real issue.
>
> There is a call to drain_workqueue() but this is not enough I think,
> because hci_recv_frame()
> can re-arm
>    queue_work(hdev->workqueue, &hdev->rx_work);

I suspect this likely a refcount problem, we do l2cap_get_chan_by_scid:

/* Find channel with given SCID.
 * Returns locked channel. */
static struct l2cap_chan *l2cap_get_chan_by_scid(struct l2cap_conn
*conn, u16 cid)

So we return a locked channel but that doesn't prevent another thread
to call l2cap_chan_put which doesn't care about l2cap_chan_lock so
perhaps we actually need to host a reference while we have the lock,
at least we do something like that on l2cap_sock.c:

l2cap_chan_hold(chan);
l2cap_chan_lock(chan);

__clear_chan_timer(chan);
l2cap_chan_close(chan, ECONNRESET);
l2cap_sock_kill(sk);

l2cap_chan_unlock(chan);
l2cap_chan_put(chan);


-- 
Luiz Augusto von Dentz