Le lundi 12 avril 2010 à 09:08 +0530, Krishna Kumar2 a écrit :
quoted text
> Hi Eric,
> 
> Eric Dumazet <eric.dumazet@gmail.com> wrote on 04/12/2010 04:05:53 AM:
> 
> > I believe the following lines from dev_pick_tx() are not the problem :
> >
> >    if (sk && sk->sk_dst_cache)
> >       sk_tx_queue_set(sk, queue_index);
> >
> > It is IMHO not safe, because route for this socket might have just
> > changed and we are transmitting an old packet (queued some milli seconds
> > before, when route was different).
> >
> > We then memorize a queue_index that might be too big for the new device
> > of new selected route.
> >
> > Next packet we want to transmit will take the cached value of
> > queue_index, correct for old device, maybe not correct for new device.
> 
> When route changes, I think my patch had reset sk->sk_tx_queue_mapping
> by calling sk_tx_queue_clear. I don't know if I missed any path where
> the route changes and sk_dst_reset() was not called.
> 

Problem is when you reset sk->sk_tx_queue_mapping at the very moment
route (or destination) changes, we might have old packets queued in tx
queues, of the old ethernet device (eth0 : multi queue compatable)

2) Application does a sendmsg() or connect() call and sk->sk_dst_cache
is rebuild, it points to a dst_entry referring a new device (eth1 : non
multiqueue)

3) When one old packet finally is transmitted, we do :

	queue_index = 1; // any value > 0

	if (sk && sk->sk_dst_cache)
		sk_tx_queue_set(sk, queue_index); // remember a >0 value

4) application does a sendmsg(), enqueues a new skb on eth1

5) We re-enter dev_pick_tx(), and consider cached value in 3) is valid.
   we pick a non existent txq for eth1 device.

6) We crash.


quoted text
> The following might be better to prove the panic is due to this, since
> your suggestion will hide a panic that happens somewhat rare (according
> to Denys):
> 
>       if (sk_tx_queue_recorded(sk)) {
>             queue_index = sk_tx_queue_get(sk);
> +           queue_index = dev_cap_txqueue(dev, queue_index);
>       } else {
> 

Sure, but I thought I was clear enough to prove this commit was wrong,
and we have to find a fix.

quoted text
> Thanks,
> 
> - KK
> 
> > You could try to revert commit a4ee3ce3293dc931fab19beb472a8bde1295aebe
> >
> > commit a4ee3ce3293dc931fab19beb472a8bde1295aebe
> > Author: Krishna Kumar <krkumar2@in.ibm.com>
> > Date:   Mon Oct 19 23:50:07 2009 +0000
> >
> >     net: Use sk_tx_queue_mapping for connected sockets
> >
> >     For connected sockets, the first run of dev_pick_tx saves the
> >     calculated txq in sk_tx_queue_mapping. This is not saved if
> >     either the device has a queue select or the socket is not
> >     connected. Next iterations of dev_pick_tx uses the cached value
> >     of sk_tx_queue_mapping.
> >
> >     Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
> >     Signed-off-by: David S. Miller <davem@davemloft.net>
> >
> > diff --git a/net/core/dev.c b/net/core/dev.c
> > index 28b0b9e..fa88dcd 100644
> > --- a/net/core/dev.c
> > +++ b/net/core/dev.c
> > @@ -1791,13 +1791,25 @@ EXPORT_SYMBOL(skb_tx_hash);
> >  static struct netdev_queue *dev_pick_tx(struct net_device *dev,
> >                 struct sk_buff *skb)
> >  {
> > -   const struct net_device_ops *ops = dev->netdev_ops;
> > -   u16 queue_index = 0;
> > +   u16 queue_index;
> > +   struct sock *sk = skb->sk;
> > +
> > +   if (sk_tx_queue_recorded(sk)) {
> > +      queue_index = sk_tx_queue_get(sk);
> > +   } else {
> > +      const struct net_device_ops *ops = dev->netdev_ops;
> >
> > -   if (ops->ndo_select_queue)
> > -      queue_index = ops->ndo_select_queue(dev, skb);
> > -   else if (dev->real_num_tx_queues > 1)
> > -      queue_index = skb_tx_hash(dev, skb);
> > +      if (ops->ndo_select_queue) {
> > +         queue_index = ops->ndo_select_queue(dev, skb);
> > +      } else {
> > +         queue_index = 0;
> > +         if (dev->real_num_tx_queues > 1)
> > +            queue_index = skb_tx_hash(dev, skb);
> > +
> > +         if (sk && sk->sk_dst_cache)
> > +            sk_tx_queue_set(sk, queue_index);
> > +      }
> > +   }
> >
> >     skb_set_queue_mapping(skb, queue_index);
> >     return netdev_get_tx_queue(dev, queue_index);
> >
> >
> 
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 


--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html