Add manual scsi probe of a Lun

[alessandrocarminati]

This commit introduces manual probing of a specified SCSI LUN. The LUN can be specified either via the kernel command line scsi.addr=<host>:<channel>:<target>:<lun> or by adding a line to the initoverlayfs.config file scsi.addr <host>:<channel>:<target>:<lun>. In case both are present, the configuration takes precedence. Any operation depends on the presence of the scsi_mod.scan=manual kernel argument.
If it is not present in the kernel command line, no action will occur. The directory scsi_probe contains the main functions needed to complete the task and some related tests.
To run the tests, simply run make in the scsi_probe subdirectory.

[ericcurtin]

I can't wait to give this a run!

I'm probably taking a few day off from Wednesday (I'll be out around 7 days), so I dunno if I will get the opportunity to look at this in detail.

But it's pretty ifdef'd out, so I don't mind merging this if it helps things progress.

[alessandrocarminati]

Take your time; I took mine, so I can't complain if things go long.
In any case, to be more confident about the code quality, I bothered to add tests to my functions.
I hope it helps to maintain the quality of this contribution and not degrade the rest.

[alessandrocarminati]

This statement here is not ifdef'd.
For some reason, I thought it should be enabled, no matter what.
On a second thought, I wonder if you prefer this to be ifdef dependent.

[ericcurtin]

It was false deliberately because in the use case this code was intended for initoverlayfs is the init system, but exec's itself as a systemd process really quickly (and then systemd mounts /proc).

But tbh, it was a micro-optimization, so I don't mind it being true, it was just one less mount call.

[ericcurtin]

Actually, some things are broken here in general (not in this PR, in initoverlayfs main branch) that I have to fix when I come back.

I think if you run the upstream version of this now on a Fedora Workstation machine, it doesn't boot anymore and they are really trivial silly bugs.

I just got distracted by other things the last week or two.

[alessandrocarminati]

I updated my PR because of a typo I discovered. It doesn't change much in practice, but it bothered me. In scsi_probe/tests/parse_kernel_cmdline.c, the Italian version of Donald Duck, paperino, was misspelled.
Now, it's fixed! :-P

[ericcurtin]

Merging before I go on PTO, I didn't give the most thorough review, but it's all ifdef'd out anyway, better to merge to not block progress...

[dougsland]

agreed, sorry i couldn't review before. If we find anything later in the road we improve it.

[ericcurtin]

It will be interesting to try the boot direct to erofs idea you had @alessandrocarminati via "root=" karg and then pivot_rooting to the real rootfs using "systemd.root=" karg @rrendec is working on.

Of course this is only suitable for embedded because of the dependency on the storage drivers being directly into the kernel, but it's one of the cases we care about.