What steps will reproduce the problem?
libdnet-1.12 uses a deprecated function prototype of PyObject_AsReadBuffer,
causing an overflow on 64-bit builds. Calling dnet.ip_cksum_add(buf, x) (with
non-zero x) will cause ip_cksum_add in ip-util.c to get cksum = 0.
An example of this bug breaking something can be found in
dpkt/ip.py::IP:__str__ :
s = dpkt.in_cksum_add(0, s)
s = dpkt.in_cksum_add(s, p)
(where dpkt.in_cksum_add is aliased to dnet.ip_cksum_add in dpkt.py)
This bug causes dpkt to insert incorrect checksums for UDP/TCP packets, as the
first result from ip_cksum_add is essentially ignored.
What is the expected output? What do you see instead?
I expect cksum to be maintained between the python-c call, instead of getting
overwritten as 0.
What version of the product are you using? On what operating system?
dnet-1.12, ubuntu 10.10.
Please provide any additional information below.
I've provided a patch to libdnet-1.12/python/dnet.pyx that fixes this problem.
The solution is to use Py_ssize_t instead of int. (this patch also includes
modifications to allow dnet.pyx to build under the version of pyrexc I pulled
out of the Ubuntu repository (pyrexc 0.9.8.5))
Original issue reported on code.google.com by nexter...@gmail.com on 31 Dec 2011 at 7:38
Original issue reported on code.google.com by
nexter...@gmail.comon 31 Dec 2011 at 7:38Attachments: