Minimum runc shim implementation

[Burning1020]

Implement shim cli commands and some basic task APIs to run a runc container

Co-authored-by: Feng Shaobao fshb1988@gmail.com
Co-authored-by: Zhang Tianyang burning9699@gmail.com
Co-authored-by: Yu Qitao yuqitao1024@qq.com

Signed-off-by: Zhang Tianyang burning9699@gmail.com

[Burning1020]

Hi, all! I'm from Huawei Cloud Company. My team has implemented a whole runc shim that can truly run a runc container and it works well with containerd and even kubernetes for mounths in production .

Now that having noticed this "rust-extensions" project, we'd like to open our work and provide LTS as well. Actually, this commit is only a part just including some basic tasks(eg. create/start/exec/wait/kill/delete/shutdown...), the other tasks will be brought up if this commit can be accepted. Hope you can take our implementation into consinderations.

You can build it and replace to where the containerd-shim-runc-v2 (go verison) is for your own, use ctr or crictl to run pod/containers then I'm sure you'll get a wonderful experience!

[Burning1020]

This commit has many features as you see, no limited to this:

1. The completed shim commands(start/delete/default) imps based on your current work
2. Impls for basic tasks, with runc cli underlay.
3. A signal handler wording with a static subscribe/nortify monitor and a subreaper
4. The abstract of ContainerFactory, Container, Process, CommandRuntime and related RuncFactory, RuncContainer, InitProcess/ExitProcess, Runc and so on. (we'll continuous optimize this part to make it more clear)
5. A checker for process exits and so on

Honestly, this commit may still have defects or something like that:

1. We've defined a crate level error in error.rs, which is different from your mod level error. The reason is that, it is much complicated to return each mod error layer by layer and there would left many duplicated errors. BTW, our solation is not perfect either, we can discuss this point lately.
2. Exec process with tty works well in rust 1.48 but in 1.52, the terminal display may be a little unfriendly but that doesn't affect function. I believe there are something wrong in std::io::copy(), I will figure out and push again at the end.
3. Not enouth UT and comments in our code.

There is still some work needed to be done, perhaps we can develop together!

[Burning1020]

I'll fix the CI checks and continue to add changes lately, don't worry~

[AkihiroSuda]

Seems a duplicate of #21 and #22

[fuweid]

@AkihiroSuda I don't think it is duplicate 😄 I Checkout this patch and it provides more helpers like go-runc-shim, for instance，fifo scheme support and pure runC commandline. I would like to say it can work with #21 or #27 together. WDYT?

Cc @mxpv

[Burning1020]

I just pushed again and it passed CI finally.
cc plz. @AkihiroSuda @mxpv
@jiangliu plz give review it to avoid same impls within we two

[Burning1020]

When handling container IO, we take a method that pass the stdio from request task to runc directly, which is quite different from pipeIO. Cause we think our solution has higher efficiency than CopyIO in shim layer.

[Burning1020]

@mxpv async feature defined in Cargo.toml used here

[Burning1020]

update done. ping @mxpv

[mxpv]

LGTM

[mxpv]

@jiangliu @kzys PTAL?

[mxpv]

Should we move runc dependent bits from shim crate to runc-shim ? (container, task).
Not a blocker for this PR, I can follow up on this.

[fuweid]

LGTM

[Burning1020]

ping @mxpv

[Burning1020]

thanks for reviewing! @mxpv @fuweid @kzys