23 нояб. 2015 г. 17:15 пользователь Andrey Ryabinin <aryabi...@odin.com>
написал:
>
>
>
> On 11/23/2015 07:04 PM, Stanislav Kinsburskiy wrote:
> >
> >
> > 23.11.2015 16:53, Andrey Ryabinin пишет:
> >>
> >> On 11/23/2015 06:46 PM, Stanislav Kinsburskiу wrote:
> >>
> >>> +1
> >>> I'm curious, why network namespace is not reported as leaked...
> >>>
> >>>>> 3) This object holds network namespace. And it also have to be leaked
> >>>>> leaked after CT stop. Is it?
> >> Why do you think it leaked?
> >> How the rule could hold the namespace?
> > int fib_default_rule_add(struct fib_rules_ops *ops,
> > u32 pref, u32 table, u32 flags)
> > {
> > <snip>
> > r->fr_net = hold_net(ops->fro_net);
>
> That is fib_rules_ops. ops not leaked, so net ns is also not leaking.
> We leaking fib_rule.
>
I was not clear enough, sorry.
I meant debug message in dmesg, which should be printed in net_free() if
NETNS_REFCNT_DEBUG is set.
I didn't see it.
>
> > <snip>
> > return 0;
> > }
> >
> > But, probably, we have NETNS_REFCNT_DEBUG unset. And that's bad...
> >
_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel
