Wednesday, September 19, 2007

Improving Samba4 winbind, a look back

What the project is about Samba4 contains a basic winbind implementation, but it is still lacking many features. The goal of my project was to improve Samba4 winbind so that the nsswitch and pam functionality provided by Samba3's winbindd would be present in Samba4, too. Simply copying code over from Samba3 would not do, of course, as the underlying architecture in Samba4 is different. Also, the goal was to improve readability of the code, as opposed to the more organically "grown" look of the Samba3 winbindd code. In Samba4, there is a library taking care of user/group management called libnet. Samba4 winbind uses this library extensively. What was done so far
  • Porting nsstest to Samba4 nsstest is a binary that test basic functionality of a nsswitch library. Once all the tests in nsstest work for libnss_winbind from Samba4, the winbind implementation is useable.
  • Getting information about users / SIDs This is the basic nsswitch getpwnam/getpwuid functionality returning a pwent structure. It is also possible to query for AD / NT domain information about the user/sid.
  • User listing / enumeration nsswitch provides a set of functions (setpwent/getpwent/endpwent) to iterate over a password database. The classic database is the /etc/passwd file, but of course using nsswitch, it's possible to use a directory like LDAP or AD.
  • Mapping SIDs to user ids and back This one is only stubbed out, as Samba4 doesn't handle that mapping yet. However, functions for this were needed to make winbind work, so I had to stub these out. The advantage is that the other code will automagically start to work correctly once these functions are implemented for real.
  • Mapping SIDs to group ids and back Much the same applies here, once idmapping is supported in Samba4, this will be replaced by real code.
What is left to do
  • Group enumeration The libnet functions for group enumeration were not implemented by the time GSoC was up. Now these functions are in, so support for groups identical to the user functions will follow soon.
  • NTLM caching Due to time constraints, caching of NTLM blobs was discarded. The nsswitch/pam functionality was regarded as core importance.
  • PAC/info3 caching As with NTLM caching, PAC/info3 caching was discarded. Caching is only interesting once the other features are working and will be implemented eventually.
  • Automated tests Currently the only way to test all of the functionality is to wrap the wbinfo binary and let that take care of constructing the necessary winbind queries. This is a bit clumsy. Jerry Carter is currently working on a winbind client library that will allow to access the functionality of wbinfo without a wrapper. The tests will be implemented using that API once it is in the tree.
Future (related) work First of all, the features still left on the TODO list will be implemented. Group functions first, testing next if possible. There is more to winbind than this GSoC project was about, so the more missing features will be implemented. The caching will follow once the other features are working and tested. An improved winbind will help Samba4 to not only act as an AD controller but also as a domain member. A look back Complying with long-standing computer science tradition, I underestimated the amount of work that had to be done before I could start working on the actual features I was planning to implement. In the end I had to prioritize features and drop the least important ones to get finished in time. I did not expect to spend so much time figuring out my way around the libnet code. However, the foundation for implementing the dropped features is laid, so I do not feel too bad about it. Samba4 winbind already works better than before. Pending group support and id mapping, it will be usable for simple scenarios. Conclusions I still need to be more careful about the scheduling of projects and estimating the amount of work required to get features to work. Still, the only way to improve is to try and adjust the estimations accordingly. I feel more confident around the Samba4 code now, thanks to Metze, Jelmer and Andrew's help. Of course thanks to all the other team members for the help and advice offered, on IRC and the mailing lists. Last but not least I would like to thank Google in general and Leslie Hawthorn in particular for running the third Summer of Code program in an efficient manner, making this a really enjoyable experience.