[OpenAFS-devel] Honoring g+s on directories

Todd M. Lewis Todd_Lewis@unc.edu
Tue, 13 Jun 2006 15:45:27 -0400


How big a deal would it be to make OpenAFS clients honor the g+s bits on 
directories?  In normal UNIX directories, new files created in a g+s 
directory are owned by the same group that owns the directory, but 
OpenAFS (like AFS before it) creates files owned by the users current 
effective gid. I can see why, given that AFS doesn't really care so much 
about UNIX groups. But then why pick the process egid instead of the 
directory's group for new files in g+s directories?

I started looking for the responsible code, but quickly realizing the 
haystack was a lot larger than the needle I was looking for (Is this a 
totally client-side issue, or does the server get involved?), I though 
I'd ask you guys.

So, this boils down to two questions:

* Would this be a big change to implement? and

* If a suitable patch were created, would such a change in behavior even 
be considered?

Rationale: 1. AFS already throws enough quirks at potential adopters. 
This is one difference between traditional UNIX semantics and AFS that 
seems, to me at least, unnecessary. 2. I have at least one project that 
tends to migrate among file systems, sometimes including AFS. The 
"from-AFS" transition always causes a hiccup because new files have 
unexpected group owners.
-- 
    +--------------------------------------------------------------+
   / Todd_Lewis@unc.edu  919-445-9302  http://www.unc.edu/~utoddl /
  /    The only substitute for good manners is fast reflexes.    /
+--------------------------------------------------------------+