Subject: Re: struct ifnet's if_softc field?
To: None <>
From: Dave McConnell <>
List: tech-kern
Date: 05/16/1999 11:55:43
>  > Why does "struct ifnet" have a "void *if_softc" as its first field. It 
>  > seems unnecessary...
> It's quite necessary, actually...
I saw that the "struct ifnet" documented in Wright/Stevens' Book 
(TCP/IP illustrated Vol2) which deals with the 4.4BSD net-3 code, 
doesn't have a "void* if_softc" anywhere in it which is why I asked.

I'm pretty new to this stuff and I'm really looking for insight.

>  > Isn't ifnet always the first field in a network drivers softc structure?
> No, in fact, it's never the first field in a network driver's softc.  A
> "struct device" is the first structure _always_ in a real device.
OK, once again, I was going by Wright/Steven's book (TCP/IP 
Illustrated, Vol2, section 3.6, page 76) where he essentially states 
that the network interface softc's always have a "struct ifnet" as the 
first field (which he illustrates with the Lance ETHERNET (le_softc), 
SLIP (sl_softc) and the (trivial) loopback case. 

Subsequent to that I had looked quite carefully at the SLIP and 
PPP code for NetBSD where indeed the sl_softc and ppp_softc are 
headed by a "struct ifnet". 

However, as you point out and UNLIKE in Wright/Steven's Lance 
ethernet example, the "struct le_softc" in NetBSD begins with a 
"struct device", followed by a struct ifnet(embedded in the struct 
ethercom). I assume it is like this for all "real" network devices?

I expect that this way of doing things is one of the reasons for the 
first field of "struct ifnet" being a "void *if_softc" ?

BTW is this a NetBSD specific way of doing things or do all the 
BSD flavours do it like this?

Of course I now have a new problem. I can't fully rely on the 
Wright/Stevens book anymore :-( :-(

Thanks for the help!