Re: select() regression in libpcap-devel?

[Guy Harris <guy () alum mit edu>]

On Jun 24, 2009, at 1:40 PM, Mike Kershaw wrote:

> Yes, it's linux, sorry for not including that.

As you might guess, I suspected it was a problem with the memory- 
mapped capture stuff in Linux; that and the FreeBSD memory-mapped  
capture stuff were the most likely 1.0 changes to cause this,  
although...

> dragorn@drd1813 ~/src/libpcap $ tcpdump -V
> tcpdump version 4.1-PRE-CVS_2009_06_22
> libpcap version 1.1-PRE-CVS

...does this happen with standard 1.0?  If not, this might be a side- 
effect of a bug fix in top-of-tree - it turns out that you are *not*  
finished with a packet when the callback returns; consider how  
pcap_next() and pcap_next_ex() are implemented - they call the  
callback and, when it returns, hand the packet back.  That's not an  
issue when the packet buffer is owned solely by libpcap, but if it's  
shared between the kernel and libpcap, the kernel might reuse a buffer  
slot in the memory-mapped region after libpcap releases it, so it  
can't release it until the *next* packet is read.
-
This is the tcpdump-workers list.
Visit https://cod.sandelman.ca/ to unsubscribe.