It's great to finally have something systems-related to post about amidst the endless bureaucracy that fills my days lately. Of course that means that — yup, you guessed it — something broke. But hey, that's what it's all about. Well, that and the fixing of said brokeness, of course.
So we recently discovered that our Windows clients were suddenly, and without explanation, able to greatly exceed their roaming profile quotas. In fact, looking at the roaming profile drive showed users with upwards of 25 GBs in their roaming profiles, which have quota limits of 50 MB. Not only that, but further testing revealed that Windows client machines wouldn't even complain if they went over quota. Any SMB connection to the roaming profile drive could exceed the quota limit without so much as a complaint from server or client. AFP worked. UNIX worked. But quotas were ignored over SMB. What the fuck?
For three days I've been trying to track this problem down, testing all sorts of quota scenarios and SMB configurations in between meetings and meetings and more meetings. Eventually, when I can't make headway on a problem, I start thinking it might just be a bug. So I started poking around in the Apple Discussions, and I found one and only one complaint of a similar nature: 10.4.8 Server with broken quotas on Windows. Had I recently done a system update that perhaps broke quotas?
So I started thinking about what in a system update could break such a thing. How do quotas work? There is no daemon. A colleague suggested that they were part of the kernel. Had I done anything that would have replaced the kernel in the last month or two?
The answer was yes. Over the winter break I had decided to update the server to version 10.4.8. Upon realizing this I began to strongly suspect that Mac OS X Server 10.4.8 contained a bug that broke quotas over SMB. Fortunately, as is often my practice, I'd made a clone of my 10.4.7 server to a portable firewire drive before upgrading. Testing my theory would be a simple matter of booting off the clone.
Sure enough, after booting from the clone, quotas began behaving properly on Windows clients again. Because I had the clone, reverting the 10.4.8 server back to 10.4.7 was a simple matter of cloning the contents of the firewire to the server's internal drive and rebooting. Voilà! Problem solved!
From now on I think I'll hold off on server updates unless I really, really need them. When it comes to servers, I think the old adage is best: If it ain't broke, don't fix it.