Since OpenBSD Network Stack Internals by Claudio Jeker at AsiaBSDCon 2008, there's a noticeable changes in OpenBSD network stack. obviously the removal of using softnet task queue for forwarding IP packets which is now done by almost a complete run (passing packets from Incoming Interface to outgoing Interface). For me to understand the processing of incoming packets, I did a quick glance of OpenBSD's network stack and did a little graph below:
Stern Warning: This is my understanding of OpenBSD's network stack internals (by reading the source code) and this is by no mean a case study. So please feel free to fact check and let me know where I miss-understood.
The above graph, shows the input of the packet to network stack by the ethernet
device driver to the network stack. This is done at the lowest level and is the lowest
layer which is Layer 1. At the next Layer (Layer 2),
the ethernet header and by the value of it's type, passes the packet to the Layer 3.
This used to be done by enqueuing the packet (in our case, IP Packet) into a
queue and schedule a software interrupt to process the packet later, but apparently
with the new changes in OpenBSD, the queuing process is removed, letting
forwarding the packets without scheduling any interrupts.
ip_input_if, the IP packet's header is ispected to decide if it's destined to us
or should be forwared. if it's not destined locally,
ip_forward will do the forwarding
the packet. if the destincation is not reachable, it produces an ICMP Unreachable
message and sends it to the sender, otherwise the
ip_output will send the packet
to the right interface. Now the packet enter's the Layer 2 again and will be
enqueued into outgoing interface to be send out later.
But, if the IP packet is destined for us, it will be enqueued by
This process is done at the Layer 3 and later
by scheduling an interrupt, the packet will be dequeued by the
if_netisr and be
passed to the propper internet protocol by
ip_deliver. This is where the packet
leaves the Layer 3 and enters the Layer 4 of our network stack.
Now, deppending on the payload and opened sockets on the user land and a few
other conditions (like if the socket is spliced), the packet is processed differently.
But typically it is done by copying the mbuf packet into the right socket's received
buffer (if it's not FULL) and will be notified by
If you find these topics interesting or you're a newbie like me, you should read the followings as well. Hope to be useful: