|
Nikolaus Schlemm
nikl at nikl.net
Tue Jun 27 06:54:44 EDT 2006
Am Dienstag, 27. Juni 2006 09:04 schrieb Deron Meranda: > On 6/27/06, Mike Looijmans <nlv11281 at natlab.research.philips.com> wrote: > > I'm using a 64-bit AMD linux machine here - and I'm doing lots of MD5 > > checksumming without problems while processing terabytes of data. in modpython? I only have the problem there not in "regular" python.. > > I can probably use some 64 bit intel as well, but in the hundreds of > > mails I got during my vacation, I lost track of the original post for > > this problem. I found the problem while moving a django powered site from django's builtin dev-server to an modpy setup - it was complaining that I had tampered with the session cookies. md5 is used in django for generating a) session ids and b) as a checksum for the session data. when I saw that the session ids looked odd, I first tried to generate some md5-hashes in django and when they were flawed, I wrote the little test-handler Deron quoted. In django I got "solved" the problem by generating the hashes with sha instead of md5 - but that's only a little hack, not a solution ;) but at least this is not a blocker right now. also not that other django-users on irc have the same problem. unfortunately, I forgot to ask them for details like arch, distro, versions. > Good to hear it's working for you on AMD 64-bit. > > Just a quick summary of of the emails so far about what Nikolaus is > seeing, on two machines I believe. (Nikl, correct me if I'm wrong) yep > One is a debian Linux (celeron 2ghz), the other ubuntu Linux (opteron > 144). The Ubuntu Linux is 64-bit with python 2.4.3-1ubuntu1 and > libapache2-mod-python-3.1.4-0ubuntu1 (both stock from the distro). > > When he does MD5s, he always gets the same bogus hexdigest value > regardless of input, namely 0000000000000000d09215a3e97561b3 in his > report. Supposedly the value will be different on each mod_python > request (but constant for every MD5 within the same request). Also, > although random the initial half is always zeros. The confusing part > is that when he runs Python outside of mod_python (such as > interactively) MD5s always work correctly. Supposedly SHAs work > regardless. > > He ran a simple test handler: > > def handler(req): > import md5 > req.content_type = 'text/plain' > req.write('md5: %s\n' % md5.__file__) > for k, v in MD5_TESTS.items(): > m = md5.new(k).hexdigest() > req.write('"%s" -> "%s" ? %s\n' % (k, m, (m == v))) > return apache.OK > > Using RFC 1321 test vectors. MD5s were all wrong. > > I had him dump sys.modules, as well as looking at /proc/xxxx/map for > the httpd process, and saw nothing unusual. The md5.so module is > loaded into the process from the expected Python lib-dynload standard > path. > > I did see that his Apache has a ton of modules loaded though (many > PHP, MySQL, X11, etc), so I've recommended he run a stripped-down > apache. Haven't heard results yet. unfortunately I can't strip down the apache as those dependencies are used by my clients and I don't have any spare machines right now... :( might be possible in a couple days though :) -- cheers Nikl
|