From: Sukadev B. <su...@li...> - 2009-07-02 17:28:51
|
Serge E. Hallyn [se...@us...] wrote: | Quoting M. Mohan Kumar (mo...@in...): | > [PATCH] pidns14 | > | > Container-init may be immune to unhandled fatal signals (like SIGUSR1) | > even if they are from ancestor namespace. SIGKILL/SIGSTOP are the only | > reliable signals to a container-init from ancestor namespace. Make sure | > that container-init will not respond to signals other than | > SIGKILL/SIGSTOP | | Hmm? This may or may not be right... but you start out by saying 'may be | immune to', then provide a patch making the testcase TFAIL if is not immune | to. So at the very least anyone on a slightly older kernel will get TFAILs. | | I don't think that immunity to SIGUSR1 from ancestor pidns is something we | want to guarantee, it's just what is happening. The proper thing is to | not depend on either getting or not getting SIGUSR1, in my opinion. Suka? Yes we did confirm that there is a test for SIGKILL from parent ns. We discussed in the bug report on whether to drop or modify the test, but leaned towards modifying the test bc if SIGUSR1 does kill a container init, then something has changed in the sig_ignored() checks in the kernel. Hmm, not a very strong reason to keep the test. Lets just drop the test :-) Sukadev |