[ECOS] Problem with IP Stack ...

[Thomas BINDER]

I have a serious problem with the IP Stack of ecos. Whenever UDP packets are
sent to an open socket and nobody cares to read the packets, ecos crashes after
a while. The problem seems to be related to the long known `out of MBUFs'
problem. At least one occasionally reads this message before the crash happens.
As I understand, however, running out of MBUFs should not actually crash the
system.

The attached code will reproduce the problem.

target code:
	mbuf.cpp

host code (linux):
	client.cpp


To reproduce the problem first load and start mbuf. It will open an udp socket
and then print mbuf statistics.

Then start the client tool:
	./client 10.16.2.238 9001 100000

The last argument determines an interval (realized as a for loop) between two
consecutive transmissions of data. A smaller value results in faster
transmissions.

With an interval value of 0 (up to 100000 on my machine) ecos crashes
immediately.

The problem is not restricted to UDP sockets. It also happens after return from
accept of a TCP socket. Note, however, that the problem does not occur *before*
accept returns.

I use ecos version 2.0 (from June or Juli).

Any ideas on what is going on here or where I could start searching?


Thx,
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mbuf-udp.cpp
Type: text/x-c++src
Size: 4892 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/ecos-discuss/attachments/20020906/1c969593/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: client.cpp
Type: text/x-c++src
Size: 2500 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/ecos-discuss/attachments/20020906/1c969593/attachment-0001.bin>
-------------- next part --------------
-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss