Re: [PATCH RESEND 0/2] j1939: make sure that sent DAT/CTL frames are marked as TX

[Oleksij Rempel <o.rempel@pengutronix.de>]

Hi Devid,

On Tue, May 10, 2022 at 08:12:32PM +0200, Devid Antonio Filoni wrote:
> Hi Oleksij,
> 
> On Tue, 2022-05-10 at 06:34 +0200, Oleksij Rempel wrote:
> > Hi Devid,
> > 
> > On Mon, May 09, 2022 at 07:07:44PM +0200, Devid Antonio Filoni wrote:
> > > Hello,
> > > 
> > > If candump -x is used to dump CAN bus traffic on an interface while a J1939
> > > socket is sending multi-packet messages, then the DAT and CTL frames
> > > show up as RX instead of TX.
> > > 
> > > This patch series sets to generated struct sk_buff the owning struct sock
> > > pointer so that the MSG_DONTROUTE flag can be set by recv functions.
> > > 
> > > I'm not sure that j1939_session_skb_get is needed, I think that session->sk
> > > could be directly passed as can_skb_set_owner parameter. This patch
> > > is based on j1939_simple_txnext function which uses j1939_session_skb_get.
> > > I can provide an additional patch to remove the calls to
> > > j1939_session_skb_get function if you think they are not needed.
> > 
> > Thank you for your patches. By testing it I noticed that there is a memory
> > leak in current kernel and it seems to be even worse after this patches.
> > Found by this test:
> > https://github.com/linux-can/can-tests/blob/master/j1939/run_all.sh#L13
> > 
> > 
> > Can you please investigate it (or wait until I get time to do it).
> > 
> > Regards,
> > Oleksij
> > 
> 
> I checked the test you linked and I can see that the number of the
> instances of the can_j1939 module increases on each
> j1939_ac_100k_dual_can.sh test execution (then the script exits),
> however this doesn't seem to be worse with my patches, I have the same
> results with the original kernel. Did you execute a particular test to
> verify that the memory leak is worse with my patches?
> I tried to take a look at all code that I changed in my patches but the
> used ref counters seem to be handled correctly in called functions. I
> suspected that the issue may be caused by the ref counter increased
> in can_skb_set_owner() function but, even if I remove that call from the
> j1939_simple_txnext() function in original kernel, I can still reproduce
> the memory leak.
> I think the issue is somewhere else, I'll try to give another look but I
> can't assure nothing.

Suddenly detecting local frames by skb->sk will not work for all control
packets. I'll send different patch solving it for all j1939 and raw
variants.

Regards,
Oleksij
-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |