On Wed, Jan 08, 2014 at 12:26:41PM -0700, Eric Blake wrote:
On 01/08/2014 12:10 PM, Guido Günther wrote:
> On Mon, Dec 23, 2013 at 10:55:50PM -0700, Eric Blake wrote:
> [..snip..]
>> + if (virAsprintf(&path, "/proc/%llu/ns/mnt", (unsigned long
long)pid) < 0)
>> + goto cleanup;
>> +
>> + if ((fd = open(path, O_RDONLY)) < 0) {
>> + virReportSystemError(errno, "%s",
>> + _("Kernel does not provide mount
namespace"));
>> + goto cleanup;
>> + }
>
> So in case mount namespaces are unavailable we'll fail these operations
> entirely? I think this is the right thing to do but it will break
> distros that have a too old kernel. So shutting down of containers will
> no longer work (as it did before).
We'll fail the attempt to use initctl as the shutdown mechanism, but
should still gracefully fall back to the attempt to use signals (once
this patch is in [1]). Or, if the user explicitly requested intictl
only, then they WANT to know that initctl didn't work.
Ahh...the signals. Forgat that we have this too. Thanks!
-- Guido