On Fri, Jul 31, 2015 at 02:57:36PM +0300, Cyrill Gorcunov wrote: > On Fri, Jul 31, 2015 at 01:53:19PM +0300, Vladimir Davydov wrote: > > > > > > The distinguishing between VEs is happening on "opener" context, > > > ie join to VE cgroup, and open /dev/vztm1 the kernel calls > > > for get_exec_env and allocates appropriate vzt pair. > > > > i.e. if I want to open a container's console as a slave I have to put > > myself into its ve cgroup first? > > Exactly. On open the kernel will find that the global character > device associated with > tty_fops > tty_open > tty_lookup_driver -> vtty_slave > tty_driver_lookup_tty -> get_exec_env -> map for tty -> lookup for tty in > array >
I really don't like it. You know why? Because if I were a VNC client, I'd do my best to avoid dealing with the ve cgroup :-) Seriously, you won't even be able to get out of it once you are done - you're doomed to die and be buried there :-) Let's instead introduce an ioctl for it. To be more exact, let's port VE_CONFIGURE_OPEN_TTY from PCS6. What do you think about it? My point is that it's a legacy thing which will hopefully pass away in Vz8, because the whole vtty thing can be perfectly done from userspace using ptys - you only need a daemon flushing dummy slaves. So using ioctls for it should be OK. _______________________________________________ Devel mailing list [email protected] https://lists.openvz.org/mailman/listinfo/devel
