On Wed, Apr 15, 2009 at 01:55:18PM -0500, Serge E. Hallyn wrote:
...
Index: src/lxc_container.c =================================================================== ...
This all looks good, though I haven't tested it yet.
+ /* + * If doing a chroot style setup, we need to prepare + * a private /dev/pts for the child now, which they + * will later move into position. + * + * This is complex because 'virsh console' needs to + * use /dev/pts from the host OS, and the guest OS + * needs to use /dev/pts from the guest. + * + * This means that we (libvirt_lxc) need to see and + * use both /dev/pts instances. We're running in the + * host OS context though and don't want to expose + * the guest OS /dev/pts there. + * + * Thus we call unshare(CLONE_NS) so that we can see + * the guest's new /dev/pts, without it becoming + * visible to the host OS. + */
Calling unshare(CLONE_NEWNS) will not prevent the host OS from seeing the new /dev/pts if / was MS_SHARED. That isn't taken care of anywhere else for this process's namespace, is it?
Yeah, so this is the place where I think we must still have a difference in our host setups. I'm testing this patch on a Fedora 11 host, and with my current code, the new /dev/pts is not visible in the host. So I can only assume this means my host / is *not* MS_SHARED, while yours is. I'm struggling to find out why this is different because I'm testing on an Fedora 11 up2date system. Anyway, would it be sufficiently to add in a call if (mount("", "/", NULL, MS_PRIVATE|MS_REC, NULL) < 0) { virReportSystemError(NULL, errno, "%s", _("failed to make root private")); goto cleanup; } Just after the 'unshare' call, to make sure our / is private before we setup the new /dev/pts
I assume the reason you want the new devpts not visible in the host OS is so that it will be auto-umounted when the container is released?
Yes, that's the primary reason, although I also just don't want anything accessing the container's PTYs directly from the host. I wanted them to remain a hidden impl detail, used only by the libvirt controller process. So its preferrable that this new dev/pts instance is not visible in the host (at least not easily). Daniel -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|