On Mon, 29 Oct 2018 06:18:11 +0000 chouryzhou(周威) <chouryzhou@xxxxxxxxxxx> wrote: > We are working for running android in container, but we found that binder is > not isolated by ipc namespace. Since binder is a form of IPC and therefore should > be tied to ipc namespace. With this patch, we can run more than one android > container on one host. > This patch move "binder_procs" and "binder_context" into ipc_namespace, > driver will find the context from it when opening. Althought statistics in debugfs > remain global. > > ... > > --- a/ipc/namespace.c > +++ b/ipc/namespace.c > @@ -56,6 +56,9 @@ static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns, > ns->ucounts = ucounts; > > err = mq_init_ns(ns); > + if (err) > + goto fail_put; > + err = binder_init_ns(ns); > if (err) > goto fail_put; > Don't we need an mq_put_mnt() if binder_init_ns() fails? free_ipc_ns() seems to have forgotten about that too. In which case it must be madly leaking mounts so probably I'm wrong. Confused. _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
