[RFC][PATCH 0/6][v3] Container-init signal semantics

[Sukadev Bhattiprolu]

Eric W. Biederman [ebiederm at xmission.com] wrote:
| Sukadev Bhattiprolu <sukadev at linux.vnet.ibm.com> writes:
| 
| > This patchset implements the design/simplified semantics suggested by
| > Oleg Nesterov.  The simplified semantics for container-init are:
| >
| > 	- container-init must never be terminated by a signal from a
| > 	  descendant process.
| >
| > 	- container-init must never be immune to SIGKILL from an ancestor
| > 	  namespace (so a process in parent namespace must always be able
| > 	  to terminate a descendant container).
| >
| > 	- container-init may be immune to unhandled fatal signals (like
| > 	  SIGUSR1) even if they are from ancestor namespace (SIGKILL is
| > 	  the only reliable signal from ancestor namespace).
| 
| It sounds you are still struggling to get something that works and gets
| done what needs to be done.  So let me suggest a simplified semantic that
| should be easier to implement and test, and solves the biggest problem
| that we must solve in the kernel.
| 
| - container-init ignores SIGKILL and SIGSTOP.

Yes.

| 
| - container-init is responsible for setting the rest of the signals
|   to SIG_IGN.

Oleg pointed out that we could drop SIG_DFL signals to global init early
to ensure wait_for_completion_killable/lock_page_killable don't incorrectly
believe that a fatal signal is pending. (patch 2/6).

If that patch is valid regardless of containers, it would be a minor
extension to get container-inits to drop SIG_DFL signals too, right ?

So the bigger problem/unknown for me is the sig_from_user() in patch 4/6
(i.e determining if it safe to deref the pid-ns of sender). We went from
!in_interrupt() to the SIG_FROM_USER flag to this.

If that is correct, I am hoping it would come down to opitmizing the code
if possible (eg: can/should we avoid passing same_ns into sig_ignored()

There is probably some ugliness :-) but do you see any other correctness 
issues ?