Re: raw socket, IP_HDRINCL byteorder

> ip_id is simply a token, not something that has an endianness per se.
> i understand that you may end up stepping on your own feet if you
> reuse an ip_id the kernel had previously assigned, but aren't there
> *tons* of ways to step on your feet via the use of raw ip?  why focus
> on this one?

I guess the distinction here is whether you want to limit this to
"well formed IP packets", or whether you also want to use it to inject
random broken crap for testing purposes.  

There's a need to reduce the chance of ip_id collisions -- even if we
were to provide a mechanism to let the application get allocated an
ip_id for each packet it wants to send, there may still be an
arbitrary delay (due to scheduling/preemption/ process
suspention/etc.) between when the ip_id value is obtained and the
packet is actually sent.

Letting the caller supply the ip_id leans more towards allowing you to
send random broken crap.

						- Bill

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