Subject: Re: placement of PFIL_HOOKS filtering points
To: Darren Reed <>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: tech-net
Date: 11/07/2000 13:24:21
In message <>Darren Reed writes

>> SIGCOMM '96.  a more complete description is at:
>That was something I tried to mention to you on icb, before you started
>on this, Jason, and which you obviously ignored as you and Chris seemed
>intent on "reinventing the wheel".

>I'm not sure that DPF is what you want to do with BPF - typically a BPF
>filter is "compiled" and put in place as a static filter for a program
>such as dhcpd, rarpd, etc.  DPF is more than just "covert opcodes into
>machine code".

Yeah, but the part of DPF which does "safe" dynamic generaton of
native machine code for a give filter seems pretty applicable.
That part is done, for mips and alpha.

I think it'd be kewld to have a systemw hich generates native
machine-codeto evaluate ipfil rules, too :).

DPF's perforamnce compares favoruably with handwritten classifiers,
but that relies on having annotations which wouldnt be present in BPF
binary rules.  I don't know how well the codegen does without the annotations.