Re: [RACE] net: in process_backlog

[Stephen Hemminger <shemminger@vyatta.com>]

On Fri, 13 Nov 2009 07:54:14 +0800
Changli Gao <xiaosuo@gmail.com> wrote:

> On Fri, Nov 13, 2009 at 12:57 AM, Stephen Hemminger
> <shemminger@vyatta.com> wrote:
> > On Thu, 12 Nov 2009 16:50:53 +0800
> > Changli Gao <xiaosuo@gmail.com> wrote:
> >
> >
> > There is are a couple of issues here, but it is not what you thought
> > you saw.
> >
> > The receive process is always done in soft IRQ context. The backlog queue's
> > are per-cpu. When a device is deleted an IPI is sent to all cpu's to
> > scan there backlog queue.  What should protect the skb is the fact that
> > the network device destruction process waits for an RCU grace period.
> > So skb->dev points to valid data.
> 
> Yea, if the process waits for a RCU grace period, there will be no
> race. But think about another case:
> 1. flush_backlog().
After flush backlog there should be no more skb's with that device
in the queue, and if more are added, the device is buggy.

> 2. dev_hold(skb->dev); netif_rx(). dev_put(skb->dev);
There is no dev_hold in netif_rx path.

> 3. wait_for_refs();
> 4. free(dev);
> 5. netif_receive_skb(); //skb->dev doesn't present.
> flush_backlog() can't prevent new skbs are added to backlog. If we
> swap the flush_backlog() and wait_for_refs(), this case will be OK
> too.

It is still up to device driver not to add skb's to queue when stopped.

--