Lutz, do you have any plans for the upcoming changes?
I also thought about getting rid of ng_bridge from NG_NODE_FORCE_WRITER. Since rcv_data is always called in the NET_EPOCH context, I think we can do it like @kp did it for if_bridge(4) (see D24250).

In D28123#628673, @afedorov wrote:

Lutz, do you have any plans for the upcoming changes?

Yes, I have to solve the problem ASAP.
Steps include (as written in the comment):

• modify revdata path so work read only
• changes to the host or node private data need to be done via control messages (send to self)
• stats and alive counter will become counters

I also thought about getting rid of ng_bridge from NG_NODE_FORCE_WRITER. Since rcv_data is always called in the NET_EPOCH context, I think we can do it like @kp did it for if_bridge(4) (see D24250).

Thanks, I'll have a look.

In D28123#628673, @afedorov wrote:

I also thought about getting rid of ng_bridge from NG_NODE_FORCE_WRITER. Since rcv_data is always called in the NET_EPOCH context, I think we can do it like @kp did it for if_bridge(4) (see D24250).

The idea in of this patch is to enter and leave the epoch for critical paths, and delay destroying actions by callbacks to the epoch framework.
if_bridge can do this, because the epoch framework is voluntary to it's work.

On contrast ng_bridge is called in epoch itself, it's a functional addon to data paths (i.e. if_bridge) which are already covered by epoch. Despite it's unwise to interfere with this out of scope usage (we may break things on physical interface level), there is a the synchronization from the netgraph framework itself. An out-of-epoch callback does not know anything about the current state of the netgraph control message processing (which is not necessarily covered by epoch).

So I fear, that the approach of using epoch will cause subtle race conditions.

For out-of-current callbacks the netgraph framework offers control messages with direct function calls and can end in externally usable table modification. Personally I like the idea of manipulating the host table of ng_bridge from user space. I would allow me to backup and restore the table for maintenance, pre-populate the table for a quicker response (especially if D23963 does not learn all addresses automatically)

So let me try to nail down this path of using the netgraph framework directly.