[OpenAFS] OpenAFS on Linux 2.5.x

[chas williams]

In message <Pine.LNX.4.53.0304161028171.26304@scully.trafford.dementia.org>,Der
rick J Brashear writes:
>this still has an air of "flaky" about it. still, if we can do it without
>relying on any kernel resources being provided it will help insulate
>against changes adversely affecting us.

i dont know.  it seems about as flaky as tacking on groups to a process
by patching the system calls (ngroups being a rather limited and abused
resource at times).   the previously mentioned scheme would be a bit cleaner
imho since the pgid is really the thing to track.  its too bad there isnt
a way to get a kernel event/callback for the process death.  (this is
possible but only when profiling is enabled).

>i wish i kept notes on this, since i punted doing something like this
>because of (something) late last year, where (something) was a concern
>that emerged after i progressed beyond sketching something on a napkin.

well i might try to give this a little more thought and possibly make a
test implementation.  i dont think it would take much effort really.

i do see a litte member in the 2.5 kernel task_struct though that seems
interesting -- from linux/sched.h:

...
        int (*notifier)(void *priv);
        void *notifier_data;
        sigset_t *notifier_mask;

        void *security;

/* Thread group tracking */
        u32 parent_exec_id;
        u32 self_exec_id;
...

fork seems to set this to NULL.  its probably meant to be part of the
security plugin nonsense.  however, i suppose openafs could borrow it.