Re: raw socket, IP_HDRINCL byteorder

>	we have bpf write/read for pure on-wire access.

...which requires the application to construct the l2 header and do
the l2 routing.  i've done it a few times, but i never really enjoyed
it.

>	based on udp socket behavior, i thought it makes more sense to let
>	kernel do fragment processing.

sure, but i still feel it's reasonable to allow the application to
construct and transmit fragments.  i might even go so far as to
suggest that the kernel only calculate the ip_sum if the application
feeds it a datagram where it's set to zero, that perhaps that's too
much.

-- 
|-----< "CODE WARRIOR" >-----|
[email protected]             * "ah!  i see you have the internet
twofsonet-X5LO1RRdYLVWk0Htik3J/[email protected] (Andrew Brown)                that goes *ping*!"
werdna-aTR67+QfzahWk0Htik3J/[email protected]       * "information is power -- share the wealth."

---------------------------------------------------------------------
The BSD APIs Discussion Mailing List
To unsubscribe:
send "unsubscribe bsd-api-discuss" to majordomo-DZEk9q7Sfju/3pe1ocb+swC/[email protected]