Re: Ping with a 65510 bytes pack

Philippe Strauss (philou@sicel-home-1-4.urbanet.ch)
Sun, 20 Oct 1996 20:56:14 +0200 (MET DST)

Alan Cox wrote:
>
> > Yes it fix the problem. Applied against 2.1.5 rather than 2.0.22.
> > I will provide some dump of this strange icmp packet later.
>
> This is indeed a combination of a windows 95 bug - apparently fixed in later
> releases and a Linux bug when we construct a packet over 64K long (not valid)
> Unfortunately the ip_send test program I had didnt build a packet that got
> high enough up the stack after reassembly to cause an explosion.
>
> Fun bug ;). Now perhaps you can get the tcp fixes into the 2.0.24 release
> and the route cache leak fix hey Linus 8)
>

I've played further with big sized ICMP_ECHO, and found some interesting
behavior of the ping prog. (Mine come from debian 1.1 netbase package).

pinging myself (127.0.0.1) give that:

packetsize (-s)

56-24372 work ok
24373-65332 nothing _seems_ to get through
(though i can't put tcpdump on the loopback :)
65333-65464 work ok
65465-65468 really weird, seem to loose some bits, print a warning message
and dump the packet on the screen.
On 2.1.5, i rather get a 'Couldn't get a free page....'
65469 packet too large

On the wire, bheavior seem different, between to linux box
(a 2.1.5 and a 2.0.10, no router, just a hub)

56-48948 work well
48949-65469 ping doesn't give a line of result, though tcpdump see all the
packet coming and going (request *and* reply). Also the LED's on
the hub are riding high every second :) but this is OK.

Any ideas? well, ok this is more a ping soft. problem (except maybe for the
Couldn't get a free page).

-- 
Philippe Strauss, CH-1092 Belmont


Email:		<philippe.strauss@urbanet.ch>
Homepage:	http://sicel-home-1-4.urbanet.ch