On shutdown we move physical network interfaces back to the
host namespace and rename them afterwards as well as in the
later lxc_network_delete() step. However, if the device had
a name which already exists in the host namespace then the
moving fails and so do the subsequent rename attempts. When
the namespace ceases to exist the devices finally end up
in the host namespace named 'dev<ID>' by the kernel.
In order to avoid this, we do the moving and renaming in a
single step (lxc_netdev_move_by_*()'s move & rename happen
in a single netlink transaction).
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
We switched to --ext-mount-map auto because of "system" (liblxc) added
mounts like the cgmanager socket that weren't in the config file. This had
the added advantage that we could drop all the mount processing code,
because we no longer needed an --ext-mount-map argument.
The problem here is that mounts can move between hosts. While
--ext-mount-map auto does its best to detect this situation, it explicitly
disallows moves that change the path name. In LXD, we bind mount
/var/lib/lxd/shmounts/$container to /dev/.lxd-mounts for each container,
and so when a container is renamed in a migration, the name changes.
--ext-mount-map auto won't detect this, and so the migration fails.
We *could* implement mount rewriting in CRIU, but my experience with cgroup
and apparmor rewriting is that this is painful and error prone. Instead, it
is much easier to go back to explicitly listing --ext-mount-map arguments
from the config file, and allow the source of the bind to change. We leave
--ext-mount-map auto to catch any stragling (or future) system added
mounts.
I believe this should fix Launchpad Bug 1580765
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
When the container is already running our manpage promises to exit with 2.
Let's make it so.
Signed-off-by: Christian Brauner <christian.brauner@canonical.com>
Somehow this implementation of a cgroupfs backend decided to use the hierarchy
numbers it detects in /proc/cgroups and /proc/self/cgroups as indices for
the hierarchy struct. Controller numbering usually starts at 1 but may start at
0 if:
a) the controller is not mounted on a cgroups v1 hierarchy;
b) the controller is bound to the cgroups v2 single unified hierarchy; or
c) the controller is disabled
To avoid having to rework our fallback backend significantly, we should
explicitly check for each controller if hierarchy[i] != NULL.
Signed-off-by: Christian Brauner <christian.brauner@canonical.com>
If cgroupv2 is enabled either alone or together with legacy hierarchies
/proc/self/cgroup can contain entries of the form:
0::/
These entries need to be skipped.
Signed-off-by: Christian Brauner <christian.brauner@canonical.com>
This will never actually overflow, because %d is 32 bits and eth is 128
bytes long, but safety first :)
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
We need this for calculating the name of unnamed interfaces in the config.
But we also need to remember to increment it :)
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
This should allow proper filtering of build flags for libraries and make
it easier to use PIE/PIC.
Signed-off-by: Stéphane Graber <stgraber@ubuntu.com>
So far, users could only create overlay snapshots by specifying -B overlayfs
and not with -B overlay. This adds support for -B overlay.
Signed-off-by: Christian Brauner <christian.brauner@canonical.com>
