[HACKERS] semaphore usage "port based"?

[Daniel Eischen]

On Tue, 4 Apr 2006, Peter Jeremy wrote:

> On Mon, 2006-Apr-03 08:19:00 -0400, Daniel Eischen wrote:
> >I don't really see what the problem is.  ESRCH seems perfectly
> >reasonable for trying to kill (even sig 0) a process from a
> >different jail.  If you're in a jail, then you shouldn't have
> >knowledge of processes from other jails.
>
> I agree in general.  The problem here is that SysV IPC isn't
> jail-aware - there's a single SysV IPC address space across the
> physical system.  This confuses (eg) postgres because it can
> see the SHM for a postgres instance in another jail but kill(2)
> claims that the process associated with that SHM doesn't exist.
>
> There appear to be two solutions:
> 1) Add a sysctl to change cr_cansignal() and/or prison_check() to
>    make processes visible between jails.
> 2) Change SysV IPC to be jail-aware.
>
> The former is trivial - but has a number of security implications.
> The latter is much harder, there is apparently a RELENG_4 patch in
> kern/48471 but it's not clear how much work would be necessary to
> being it up to scratch.

Or:

  3) Run postgres in such a way that it doesn't look for
     remnant IPC information from other instances (use a
     per-jail-specific port #?).

Postgres has no business cleaning up after different jailed
instances of itself, which it wouldn't do if IPC's were
per-jail.  So since IPC's don't currently work that way,
account for it by the way you run postgres.

-- 
DE