[coyotos-dev] Thoughts on (non)persistence

Marcus Sundman sundman at iki.fi
Fri Jan 21 17:33:35 EST 2005


On Friday 21 January 2005 23:45, Jonathan S. Shapiro wrote:
> On Fri, 2005-01-21 at 23:15 +0200, Marcus Sundman wrote:
> > On Friday 21 January 2005 23:04, Jonathan Adams wrote:
> > > On Fri, 21 Jan 2005 22:55:50 +0200, Marcus Sundman wrote:
> > > > On Friday 21 January 2005 22:02, Jonathan S. Shapiro wrote:
> > > > >   1. If processes are not persistent, how can we contrive to give
> > > > >      them some form of unique identity that survives system
> > > > > restart?
> > > >
> > > > Umm.. If the only relevant difference between two instances
> > > > (processes) of a program is the sets of capabilities they have then
> > > > I see no problem. If a process dies (e.g. because of a crash) the
> > > > system will just create a new process and give it the capabilities
> > > > of the one that died, right? Why would some kind of "unique
> > > > identity" be needed? What am I missing here?
> > >
> > > What if two programs have capabitilities to *each other*?
> >
> > Why not associate the capability with the set of capabilities that a
> > process holds instead of with that process itself?
>
> Because you might have two different processes with the same capability
> set that run different binary images.

The collection of process-less capability sets isn't a set. No two processes 
would have the same capability set. And obviously you won't give a 
process-less capability set to an instance of another program than the one 
that used to have that set.

Or are you saying that that the process is no longer an instance of the 
program it used to be but has been modified in memory? If so and if that 
binary image is lost then I don't see what could be done.

> Also, *some* of these capabilities are capabilities to things that have
> ceased to exist, such as memory maps and pages...

That's another problem and one that is not related to remapping capabilities 
to processes. I suppose the entity that manages memory maps and pages have 
to support remapping such stale capabilities, or something.


- Marcus Sundman


More information about the coyotos-dev mailing list