ppp_cp_event is called directly or indirectly by ppp_rx with "ppp->lock"
held. It may call mod_timer to add a new timer. However, at the same time
ppp_timer may be already running and waiting for "ppp->lock". In this
case, there's no need for ppp_timer to continue running and it can just
exit.
If we let ppp_timer continue running, it may call add_timer. This causes
kernel panic because add_timer can't be called with a timer pending.
This patch fixes this problem.
Fixes: e022c2f07ae5 ("WAN: new synchronous PPP implementation for generic HDLC.")
Cc: Krzysztof Halasa <[email protected]>
Signed-off-by: Xie He <[email protected]>
---
drivers/net/wan/hdlc_ppp.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/drivers/net/wan/hdlc_ppp.c b/drivers/net/wan/hdlc_ppp.c
index 64f855651336..261b53fc8e04 100644
--- a/drivers/net/wan/hdlc_ppp.c
+++ b/drivers/net/wan/hdlc_ppp.c
@@ -569,6 +569,13 @@ static void ppp_timer(struct timer_list *t)
unsigned long flags;
spin_lock_irqsave(&ppp->lock, flags);
+ /* mod_timer could be called after we entered this function but
+ * before we got the lock.
+ */
+ if (timer_pending(&proto->timer)) {
+ spin_unlock_irqrestore(&ppp->lock, flags);
+ return;
+ }
switch (proto->state) {
case STOPPING:
case REQ_SENT:
--
2.27.0
On Mon, Dec 28, 2020 at 1:17 AM Hillf Danton <[email protected]> wrote:
>
> On Sun, 27 Dec 2020 18:53:39 -0800 Xie He wrote:
> > ppp_cp_event is called directly or indirectly by ppp_rx with "ppp->lock"
> > held. It may call mod_timer to add a new timer. However, at the same time
> > ppp_timer may be already running and waiting for "ppp->lock". In this
> > case, there's no need for ppp_timer to continue running and it can just
> > exit.
>
> Because the timer callback loses the race in acquiring the ppp->lock
> does not mean it should abort.
I think aborting ppp_timer is the correct solution. When mod_timer is
called by ppp_cp_event, which is (directly or indirectly) called by
ppp_rx, this means we received something on the line that makes the
original timer no longer necessary. If the timer is pending, mod_timer
will delete it. If the timer is running and waiting for the lock, I
think it should abort. This way it would appear that the timer hasn't
fired and is deleted before it fires.
> > If we let ppp_timer continue running, it may call add_timer. This causes
> > kernel panic because add_timer can't be called with a timer pending.
>
> Meanwhie we can defuse the peril following add_timer() added in
> e022c2f07ae5, say by replacing add_timer() with mod_timer().
The code path that calls add_timer is for sending keep-alive packets
when operating in the OPENED state. If we have just changed to the
OPENED state in ppp_cp_event, we should wait for the amount of time
set by the (2nd) mod_timer call in ppp_cp_event, before firing the
timer. We shouldn't fire the timer immediately after we change to the
OPENED state. This is not the intention of the (2nd) mod_timer call in
ppp_cp_event. Therefore aborting ppp_timer is the correct solution.
From: Xie He <[email protected]>
Date: Sun, 27 Dec 2020 18:53:39 -0800
> ppp_cp_event is called directly or indirectly by ppp_rx with "ppp->lock"
> held. It may call mod_timer to add a new timer. However, at the same time
> ppp_timer may be already running and waiting for "ppp->lock". In this
> case, there's no need for ppp_timer to continue running and it can just
> exit.
>
> If we let ppp_timer continue running, it may call add_timer. This causes
> kernel panic because add_timer can't be called with a timer pending.
> This patch fixes this problem.
>
> Fixes: e022c2f07ae5 ("WAN: new synchronous PPP implementation for generic HDLC.")
> Cc: Krzysztof Halasa <[email protected]>
> Signed-off-by: Xie He <[email protected]>
Applied and queued up for -stable, thanks!
On Tue, Dec 29, 2020 at 12:10 AM Hillf Danton <[email protected]> wrote:
>
> >The code path that calls add_timer is for sending keep-alive packets
> >when operating in the OPENED state. If we have just changed to the
> >OPENED state in ppp_cp_event, we should wait for the amount of time
> >set by the (2nd) mod_timer call in ppp_cp_event, before firing the
>
> What if your change also covers the first case of mod_timer() in
> ppp_cp_event()?
Yes, for the 1st mod_timer call in ppp_cp_event, the situation is the
same. If it is called, it means we are sending out a Configure Request
or Terminate Request. In this case, we should wait for the amount of
time set by this mod_timer call, before firing the timer. We shouldn't
fire the timer immediately because this is not the intention of this
mod_timer call.
> >timer. We shouldn't fire the timer immediately after we change to the
> >OPENED state. This is not the intention of the (2nd) mod_timer call in
> >ppp_cp_event. Therefore aborting ppp_timer is the correct solution.
> >
> Given an expiring timer, is it the right time to call ppp_tx_flush() in
> addition to add/mod_timer()?
Do you mean when we are aborting ppp_timer, whether we need to call
ppp_tx_flush in the aborted ppp_timer? I don't think so. Because when
ppp_rx (which directly or indirectly calls ppp_cp_event) releases the
lock, it will call ppp_tx_flush. So we don't need to call it again.