ibmcloud: add support for SSH keys

[erlarese]

This PR adds support to the ibmcloud provider for parsing SSH keys. This applies to the IBM Cloud VPC Gen 2 environment. It does not apply to other aspects of IBM Cloud including the "classic" environment.

This code looks for the vendor-data file in the cidata drive and finds the cloud-config section within that file. A YAML parser is used to pull the SSH keys out from this section.

This is my first time writing any Rust code, so I apologize for anything that's obviously wrong. I have attempted to update the unit tests and verify they still pass. Please let me know if anything else is needed !

[lucab]

@erlarese Thanks for the PR! I currently have a release in progress at #473, so I'll queue the review here after that. Also, the CI is on fire today, so don't worry about that right now.

Before I start with a first pass on this, can you please squash everything in a single commit?

[erlarese]

I attempted to squash all 13 commits into one, but I appear to have done it poorly and somehow pulled in a bunch of other commits. I think I'm just going to create a new fork and start over with a new PR. Oops.

[lucab]

@erlarese as you prefer, but from the current branch status I think it is only missing a rebase on master (which should automatically detect and drop the other commits), see https://git-scm.com/book/en/v2/Git-Branching-Rebasing for guidance.

[erlarese]

Thanks Luca for the pointer. I think I hopefully got everything squashed into a single commit and rebased now.

[cgwalters]

This is my first time writing any Rust code, so I apologize for anything that's obviously wrong.

No worries, and thanks for submitting this PR! Personally it took me quite a while to learn Rust but it was also a rewarding process. Luca already went over some things, but in general don't hesitate to reach out with questions and the like on e.g. #forum-coreos on freenode.

[erlarese]

Hi Luca, thanks for the feedback! I have attempted to implement everything. Please let me know what you think now. Thanks !

[lucab]

@erlarese thanks, great work! I've left a few minor nits and a bunch of suggestions in order to avoid a couple of unwrapping points.
In the failed CI jobs there are a few more suggestions from clippy and rustfmt (you can run those locally too, check Travis config to see how they run).

[erlarese]

I'm not sure if this is candidate for merging yet, but please hold off on merging for now. I just built an FCOS image with the latest patches and afterburn is failing to run, I'm still debugging what exactly is wrong.

afterburn-sshkeys@core.service: Main process exited, code=exited, status=1/FAILURE
afterburn-sshkeys@core.service: Failed with result 'exit-code'.

[lucab]

@erlarese no hurry. A journalctl -u afterburn-sshkeys@core.service should you all the details.
However, when trying to investigate issues, I usually do a debug build (which produces a larger binary, with all symbols and full debugging on) and run it manually on the host.

[erlarese]

Hi @lucab , thanks for the tip. The problem was that the IBM Cloud VPC control plane was generating a MIME vendor-data file that actually contained two sub sections with the cloud-config header. The switch to mailparse and serde caused it to match on the second cloud-config section which doesn't contain the SSH keys. My unit test passed anyway because I had used an abbreviated version of the vendor-data file to keep the UT smaller.

I fixed this by adding an additional check for a cloud-config section with ssh_authorized_keys inside, and I also expanded the UT to a full actual vendor-data file generated by the IBM Cloud (with SSH keys snipped).

With these updates, it appears the unit tests pass, the linters pass, and an FCOS image built with this boots successfully on the IBM Cloud and I can connect via SSH.

Please re-review and let me know if anything else is needed, thanks !

[erlarese]

Hi @lucab, I believe I've made the requested changes.

One confusing thing is the linter is failing saying the import of use std::fs is unused, but it's pretty clearly used in that file to read the new fixture, and if I remove the import the code doesn't compile because fs can't be found. I tried a few variations of prefixing the read call with std:: and such, but I can't quite figure out why the linter thinks this import isn't used when the compiler seems to require it. Any idea on how to fix this?

[erlarese]

Ah, ignore the earlier comment about the import error, I realized the unit tests actually have a separate dependency declaration and moving the import down there seems to have fixed the issue.

[lucab]

@erlarese thanks, code now looks overall fine to me. Can you please squash everything into a single commit, so that I'll do the final pass?

[erlarese]

Hi @lucab -- the commits should be squashed back into one again. Please let me know if anything else needs to be changed!

[lucab]

LGTM, thanks for the patch!