Re: [patch net-next v2 01/15] net: introduce upper device lists

Wed, Aug 15, 2012 at 12:33:44AM CEST, bhutchings@solarflare.com wrote:

On Tue, 2012-08-14 at 17:05 +0200, Jiri Pirko wrote:

quoted

This lists are supposed to serve for storing pointers to all upper devices.
Eventually it will replace dev->master pointer which is used for
bonding, bridge, team but it cannot be used for vlan, macvlan where
there might be multiple upper present. In case the upper link is
replacement for dev->master, it is marked with "master" flag.

Something I found interesting is that the dev->master pointer and now
netdev_master_upper_dev_get{,_rcu}() are hardly used by the stackled
drivers that set the master.  They also have to set an rx_handler on the
lower device (which is itself mutually exclusive) which gets its own
context pointer (rx_handler_data).

Instead, the master pointer is mostly used by device drivers to find out
about a bridge or bonding device above *their* devices.  And that seems
to work only for those specific device drivers, not e.g. openvswitch or
team.  I wonder if we could find a better way to encapsulate the things
they want do do, in a later step (not holding up this change!).

Yes. I was thinking about this as well. I believe that we should follow up
with this.

[...]

quoted

+static int __netdev_upper_dev_link(struct net_device *dev,
+                                  struct net_device *upper_dev, bool master)
+{
+       struct netdev_upper *upper;
+
+       ASSERT_RTNL();
+
+       if (dev == upper_dev)
+               return -EBUSY;
+       /*
+        * To prevent loops, check if dev is not upper device to upper_dev.
+        */
+       if (__netdev_has_upper_dev(upper_dev, dev, true))
+               return -EBUSY;

[...]

I think we will also need to limit the depth of the device stack so we
don't run out of stack space here.  __netif_receive() implements a kind
of tail recursion whenever a packet is passed up, but
__netdev_has_upper_dev() can't avoid doing real recursion (without the
addition of a flag to net_device so it can mark its progress).

You are probably right. I'm not sure how to handle this correctly
though. Adding some hard limit number might not be correct.

The problem could be also resolved by adding another struct list_head
into struct upper and use this inside __netdev_has_upper_dev(). But that
does not seem right to me as well (Considering the fact that walking
through the tree could be in future done under _rcu).

Ben.

-- 
Ben Hutchings, Staff Engineer, Solarflare
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.