It says: The prefix path **must be absolute, have all symlinks resolved, and cleaned**. But those requirements are violated in lots of places.
What happens when it is given a non-canonicalized path is that `mountinfo.GetMounts` will not find mounts.
The trivial case is:
```
$ mkdir a && ln -s a b && mkdir b/c b/d && mount --bind b/c b/d && cat /proc/mounts | grep -- '[ab]/d'
/dev/sdd3 /home/user/a/d ext4 rw,noatime,discard 0 0
```
We asked to bind-mount b/c to b/d, but ended up with mount in a/d.
So, mount table always contains canonicalized mount points, and it is an error to look for non-canonicalized paths in it.
Signed-off-by: Marat Radchenko <marat@slonopotamus.org>
runc considers libcontainer to be "unstable" (not for external use),
so we try not to use it. Commit ed47d6ba76
brought back the dependency on other parts of libcontainer, but looks to
be only depending on a single utility, which in itself was borrowed from
github.com/coreos/go-systemd to not introduce CGO code in the same package.
This patch copies the version from github.com/coreos/go-systemd (adding
proper attribution, although the function is pretty trivial).
runc is in process of moving the libcontainer/user package to an external
module, which means we can remove the dependency on libcontainer entirely
in the near future. There is one more use of `libcontainer` in our vendor
tree; it looks like CDI is depending on one utility (devices.DeviceFromPath);
a943033a8b/vendor/github.com/container-orchestrated-devices/container-device-interface/pkg/cdi/container-edits_unix.go (L38)
We should remove the dependency on that utility, and add a CI check to
prevent bringing it back.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
The failed to recover state message didn't include the ID making this
not as useful as it could be..
This additionally moves some of the other logs to include the id for
the sandbox/container as a field instead of part of a format string.
Signed-off-by: Danny Canter <danny@dcantah.dev>
The reference/docker package was a fork of github.com/distribution/distribution,
which could not easily be used as a direct dependency, as it brought many other
dependencies with it.
The "reference' package has now moved to a separate repository, which means
we can replace the local fork, and use the upstream implementation again.
The new module was extracted from the distribution repository at commit:
b9b19409cf
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
This is a partial revert of "cri/sbserver: Use platform instead of GOOS
for userns detection".
While what that commit did is 100% the right thing to do, when the
sandbox_mode is "shim" all controller.XXX() calls are RPCs and the
controller.Create() call initializes the controller. Therefore, things
like "getSandboxController()" don't work in the case of "shim"
sandbox_mode until after the controller.Create().
Due to this asymmetry and the lack of tests for shim mode, we didn't
catch it before.
This patch just reverts that commit so that the Create() and
getSandboxController() calls remain where they were, and just relies on
the config Linux section as a hack to detect if the pod sandbox will use
user namespaces or not.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
In the sbserver we should not use the GOOS, as windows hosts can run
linux containers. On the sbserver we should use the platform param.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
Runc 1.1 throws a warning when using rel destination paths, and runc 1.2
is planning to thow an error (i.e. won't start the container).
Let's just make this an abs path in the only place it might not be: the
mounts created due to `VOLUME` directives in the Dockerfile.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
The rpc only reports one field, i.e. the cgroup driver, to kubelet.
Containerd determines the effective cgroup driver by looking at all
runtime handlers, starting from the default runtime handler (the rest in
alphabetical order), and returning the cgroup driver setting of the
first runtime handler that supports one. If no runtime handler supports
cgroup driver (i.e. has a config option for it) containerd falls back to
auto-detection, returning systemd if systemd is running and cgroupfs
otherwise.
This patch implements the CRI server side of Kubernetes KEP-4033:
https://github.com/kubernetes/enhancements/tree/master/keps/sig-node/4033-group-driver-detection-over-cri
Signed-off-by: Markus Lehtonen <markus.lehtonen@intel.com>
When the pods are transitioning there are several
cases where containers might not be in valid state.
There were several cases where the stats where
failing hard but we should just continue on as
they are transient and will be picked up again
when kubelet queries for the stats again.
Signed-off-by: James Sturtevant <jstur@microsoft.com>
Signed-off-by: Mark Rossetti <marosset@microsoft.com>
This commit just updates the sbserver with the same fix we did on main:
9bf5aeca77 ("cri: Fix net.ipv4.ping_group_range with userns ")
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
This is a port of 31a6449734 ("Add capability for snapshotters to
declare support for UID remapping") to sbserver.
This patch remaps the rootfs in the platform-specific if user namespaces
are in use, so the pod can read/write to the rootfs.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
This patch requests the OCI runtime to create a userns when the CRI
message includes such request.
This is an adaptation of a7adeb6976 ("cri: Support pods with user
namespaces") to sbserver, although the container_create.go parts were
already ported as part of 40be96efa9 ("Have separate spec builder for
each platform"),
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
This commit just ports 36f520dc04 ("Let OCI runtime create netns when
userns is used") to sbserver.
The CNI network setup is done after OCI start, as it didn't seem simple
to get the sandbox PID we need for the netns otherwise.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
Currently there is a big c&p of the helpers between these two folders
and a TODO in the platform agnostic file to organize them in the future,
when some other things settle.
So, let's just copy them for now.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
Commit c085fac1e5 ("Move sandbox start behind controller") moved the
runtimeStart to only account for time _after_ the netns has been
created.
To match what we currently do in cri/server, let's move it to just after
the get the sandbox runtime.
This come up when porting userns to sbserver, as the CNI network setup
needs to be done at a later stage and runtimeStart was accounting for
the CNI network setup time only when userns is enabled.
To avoid that discrepancy, let's just move it earlier, that also matches
what we do in cri/server.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
Beside the "in future the when" typo, we take the chance to reflect that
user namespaces are already merged.
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
userns.RunningInUserNS() checks if the code calling that function is
running inside a user namespace. But we need to check if the container
we will create will use a user namespace, in that case we need to
disable the sysctl too (or we would need to take the userns mapping into
account to set the IDs).
This was added in PR:
https://github.com/containerd/containerd/pull/6170/
And the param documentation says it is not enabled when user namespaces
are in use:
https://github.com/containerd/containerd/pull/6170/files#diff-91d0a4c61f6d3523b5a19717d1b40b5fffd7e392d8fe22aed7c905fe195b8902R118
I'm not sure if the intention was to disable this if containerd is
running inside a userns (rootless, if that is even supported) or just
when the pod has user namespaces.
Out of an abundance of caution, I'm keeping the userns.RunningInUserNS()
so it is still not used if containerd runs inside a user namespace.
With this patch and "enable_unprivileged_icmp = true" in the config,
running containerd as root on the host, pods with user namespaces start
just fine. Without this patch they fail with:
... failed to create containerd task: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: error during container init: w
/proc/sys/net/ipv4/ping_group_range: invalid argument: unknown
Thanks a lot to Andy on the k8s slack for reporting the issue. He also
mentions he hits this with k3s on a default installation (the param
is off by default on containerd, but k3s turns that on by default it
seems). He also debugged which part of the stack was setting that
sysctl, found the PR that added this code in containerd and a workaround
(to turn the bool off).
Signed-off-by: Rodrigo Campos <rodrigoca@microsoft.com>
As a follow up change to adding a SandboxMetrics rpc to the core
sandbox service, the controller needed a corresponding rpc for CRI
and others to eventually implement.
This leaves the CRI (non-shim mode) controller unimplemented just to
have a change with the API addition to start.
Signed-off-by: Danny Canter <danny@dcantah.dev>
When a container is just created, exited state the container will not have stats. A common case for this in k8s is the init containers for a pod. The will be present in the listed containers but will not have a running task and there for no stats.
Signed-off-by: James Sturtevant <jstur@microsoft.com>
