[OpenAFS-devel] Re: RFC: move rx epoch/cid generation into the rx core

[Andrew Deason]

On Tue, 11 Feb 2014 16:46:31 -0500 (EST)
Benjamin Kaduk <kaduk@MIT.EDU> wrote:

> [HEIMDAL_MUTEX_* initialization]
> Is there a better place for something at this level?
[...]
> [fortuna rand seeding]
> being the AFSOP_SET_THISCELL easily visible in afsd.c.  Since that
> doesn't fire up rx, we are safe for now, but how future-proof are we?

osi_Init() seems like a good place for initializing a mutex. That is
called when the openafs/libafs module is loaded. That's where GLOCK is
initialized on platforms that need procedural GLOCK initialization.

Or an alternative is afs_InitSetup, which will be called when afsd is
run, but it supposed to be called before anything hits the net (this is
where we call rx_InitHost). This isn't really _enforced_, but -rxbind
will not work if something else hits the net before that runs. So if the
code is rearchitected or whatever, it's likely that whatever's in there
will be moved to the appropriate place.

So, it seems like you could collect entropy in that syscall (calling the
syscall 32+ times doesn't seem so bad really, but obviously a buffer
would be better). And when we hit afs_InitSetup, you can track yourself
if you've got the right amount of entropy, and fail to continue if
something is wrong.

-- 
Andrew Deason
adeason@sinenomine.net