[OpenAFS-devel] PAGs/tokens/threads

[Chas Williams (CONTRACTOR)]

In message <Pine.LNX.4.64.0908071446570.25892@ossigeno.math.unifi.it>,Alberto M
ancini writes:
>> The current implementation only looks at the keyring if it can't determine 
>> the PAG from the process's groups. If it has to use the keyring, then it 
>> attempts to set the process's group list, so that the group is used for all 
>> future lookups.
>
>Is this for performance reasons (or for something i do not see) or
>is just an implementation side-effect ?

i am surpised the behavior is this way.  probably came about due to the
way i wrote the code.  i added pag support to get check if the group check
failed.  then i added the feature to put the pag groups back into the
list (so that legacy applications dependent on this worked).  i suppose
we should really check the keyring first and fall back to the groups.
see PagInCred().  just make the LINUX_KEYRING_SUPPORT stuff happen first.

>This really forbid having a pag-per-thread (at least with POSIX threads
>where groups are shared).

right.

>First of all, to go ahead, I should change this (looking at the code this 
>seems not terribly tricky) but, is this change something that can be
>reasonably applied to ``some`` openafs upstream ?
>
>Responding to Derrick's question ``For yourself, or to us?''
>
>I am writing this code and if there is some convergence on the
>interface the new setpag should have I can volunteer to write (an rewrite) 
>the code for inclusion.
>
>The facts seems to be:
>
>Adding an argument to AFSCALL_SETPAG is ``no way`` (Jeff)
>A new ioctl to pioctl can be a choice (Chas)

yeah, i didnt think anyone would like changing setpag().  anything we
probably just need a variation of setpag() (in LINUX/osi_groups.c) 
that has an extra flag to indicate the session keyring or the thread
keyring.  this would also need to flag install_session_keyring, or
just simply the keyring of interest to install_session_keyring() and
rename to install_keyring().

calling it a pag is a misnomer though at this point.  oh well.