Signed-off-by: John Howard <jhoward@microsoft.com>
Fixes#38719
Fixes some subtle bugs on Windows
- Fixes https://github.com/moby/moby/issues/38719. This one is the most important
as failure to start the init process in a Windows container will cause leaked
handles. (ie where the `ctr.hcsContainer.CreateProcess(...)` call fails).
The solution to the leak is to split out the `reapContainer` part of `reapProcess`
into a separate function. This ensures HCS resources are cleaned up correctly and
not leaked.
- Ensuring the reapProcess goroutine is started immediately the process
is actually started, so we don't leak in the case of failures such as
from `newIOFromProcess` or `attachStdio`
- libcontainerd on Windows (local, not containerd) was not sending the EventCreate
back to the monitor on Windows. Just LCOW. This was just an oversight from
refactoring a couple of years ago by Mikael as far as I can tell. Technically
not needed for functionality except for the logging being missing, but is correct.
Signed-off-by: John Howard <jhoward@microsoft.com>
Also fixes https://github.com/moby/moby/issues/22874
This commit is a pre-requisite to moving moby/moby on Windows to using
Containerd for its runtime.
The reason for this is that the interface between moby and containerd
for the runtime is an OCI spec which must be unambigious.
It is the responsibility of the runtime (runhcs in the case of
containerd on Windows) to ensure that arguments are escaped prior
to calling into HCS and onwards to the Win32 CreateProcess call.
Previously, the builder was always escaping arguments which has
led to several bugs in moby. Because the local runtime in
libcontainerd had context of whether or not arguments were escaped,
it was possible to hack around in daemon/oci_windows.go with
knowledge of the context of the call (from builder or not).
With a remote runtime, this is not possible as there's rightly
no context of the caller passed across in the OCI spec. Put another
way, as I put above, the OCI spec must be unambigious.
The other previous limitation (which leads to various subtle bugs)
is that moby is coded entirely from a Linux-centric point of view.
Unfortunately, Windows != Linux. Windows CreateProcess uses a
command line, not an array of arguments. And it has very specific
rules about how to escape a command line. Some interesting reading
links about this are:
https://blogs.msdn.microsoft.com/twistylittlepassagesallalike/2011/04/23/everyone-quotes-command-line-arguments-the-wrong-way/https://stackoverflow.com/questions/31838469/how-do-i-convert-argv-to-lpcommandline-parameter-of-createprocesshttps://docs.microsoft.com/en-us/cpp/cpp/parsing-cpp-command-line-arguments?view=vs-2017
For this reason, the OCI spec has recently been updated to cater
for more natural syntax by including a CommandLine option in
Process.
What does this commit do?
Primary objective is to ensure that the built OCI spec is unambigious.
It changes the builder so that `ArgsEscaped` as commited in a
layer is only controlled by the use of CMD or ENTRYPOINT.
Subsequently, when calling in to create a container from the builder,
if follows a different path to both `docker run` and `docker create`
using the added `ContainerCreateIgnoreImagesArgsEscaped`. This allows
a RUN from the builder to control how to escape in the OCI spec.
It changes the builder so that when shell form is used for RUN,
CMD or ENTRYPOINT, it builds (for WCOW) a more natural command line
using the original as put by the user in the dockerfile, not
the parsed version as a set of args which loses fidelity.
This command line is put into args[0] and `ArgsEscaped` is set
to true for CMD or ENTRYPOINT. A RUN statement does not commit
`ArgsEscaped` to the commited layer regardless or whether shell
or exec form were used.
Signed-off-by: John Howard <jhoward@microsoft.com>
This is the first step in refactoring moby (dockerd) to use containerd on Windows.
Similar to the current model in Linux, this adds the option to enable it for runtime.
It does not switch the graphdriver to containerd snapshotters.
- Refactors libcontainerd to a series of subpackages so that either a
"local" containerd (1) or a "remote" (2) containerd can be loaded as opposed
to conditional compile as "local" for Windows and "remote" for Linux.
- Updates libcontainerd such that Windows has an option to allow the use of a
"remote" containerd. Here, it communicates over a named pipe using GRPC.
This is currently guarded behind the experimental flag, an environment variable,
and the providing of a pipename to connect to containerd.
- Infrastructure pieces such as under pkg/system to have helper functions for
determining whether containerd is being used.
(1) "local" containerd is what the daemon on Windows has used since inception.
It's not really containerd at all - it's simply local invocation of HCS APIs
directly in-process from the daemon through the Microsoft/hcsshim library.
(2) "remote" containerd is what docker on Linux uses for it's runtime. It means
that there is a separate containerd service running, and docker communicates over
GRPC to it.
To try this out, you will need to start with something like the following:
Window 1:
containerd --log-level debug
Window 2:
$env:DOCKER_WINDOWS_CONTAINERD=1
dockerd --experimental -D --containerd \\.\pipe\containerd-containerd
You will need the following binary from github.com/containerd/containerd in your path:
- containerd.exe
You will need the following binaries from github.com/Microsoft/hcsshim in your path:
- runhcs.exe
- containerd-shim-runhcs-v1.exe
For LCOW, it will require and initrd.img and kernel in `C:\Program Files\Linux Containers`.
This is no different to the current requirements. However, you may need updated binaries,
particularly initrd.img built from Microsoft/opengcs as (at the time of writing), Linuxkit
binaries are somewhat out of date.
Note that containerd and hcsshim for HCS v2 APIs do not yet support all the required
functionality needed for docker. This will come in time - this is a baby (although large)
step to migrating Docker on Windows to containerd.
Note that the HCS v2 APIs are only called on RS5+ builds. RS1..RS4 will still use
HCS v1 APIs as the v2 APIs were not fully developed enough on these builds to be usable.
This abstraction is done in HCSShim. (Referring specifically to runtime)
Note the LCOW graphdriver still uses HCS v1 APIs regardless.
Note also that this does not migrate docker to use containerd snapshotters
rather than graphdrivers. This needs to be done in conjunction with Linux also
doing the same switch.
