[patch] [rfc] udhcpc eating too much cpu

[Stefan Rompf]

Am Freitag, 18. Januar 2008 21:54 schrieb Cristian Ionescu-Idbohrn:

> Makes me happy too ;)  But the credit should go to a collegue of mine
> (Bcc:ed) who checked the code and suggested that.

I can credit to "unknown colleague of Cristian Ionescu-Idbohrn" ;-)

> By the way, you wodn't happen to be excited over cotributing your netlink
> code (adjusted) to busybox needs?

Feel free, but from looking at udhcpc, our codebases are quite different so 
that it will be mostly recoding the idea. Important points that come to my 
mind:

-When doing RFC4436 like address validation by arping the default gateway, 
there should be a time limit, e.g. don't do this when the interface has been 
down for longer than 60 seconds. Main reason is that the MAC address isn't 
that unique when rundancy protocols like VRRP or HSRP are used. Using a time 
limit, you can guess whether the down/up sequence is from roaming between 
access points of one ESSID, or a real network disconnection. Existing 
problem: On suspend to disk/RAM, many drivers report the down/up after 
resume.
-Kernel 2.4 compatibility: 2.4 already has the IFF_RUNNING, and some drivers 
set it, but there is no notification. So for 2.4, you either have to poll the 
kernel or completely ignore the flag.
-You should at least migrate read_interface() to use the same netlink socket 
as the notification to avoid race conditions.

Anything left... surely, I'll mail you when I remember it.

Stefan