Add tool to display emergency log message full-sceen on boot failure.

[1awesomeJ]

No description provided.

[1awesomeJ]

I want to start with a function that fetches the oldest message in the journal and displays it to the console.
Currently I set PRIORITY to 5(notice) to see if I'll get any messages. I'm currently not getting any.

[1awesomeJ]

@bluca
Following Poettering's guidance, I created this bsod tool.
Currently it just tries to fetch the oldest message from the current boot with a log level of "emergency", and prints it to a console.

Subsequently, I have to modify it to take over the entire screen, turn it blue, and display the QR code.

Howevever, I am currently not getting an output at all, even when i change the PRIORITY to numbers other than 0.
What am I doing wrong sir?

Thank you.

[poettering]

so for this I think we should not go to /dev/console (which is a magic device that basically points to wherever boot time logs and stuff should go to which can be serial or similar) but the VT subsystem (which is the virtual terminal, i.e. a textual interface to a local physical display device, i.e. no serial), and I think we should not interfere with the usual log/progress output that /dev/console gets. Hence I'd suggest we allocate a full new VT for this, and then switch to that.

(in case you wonder what a VT is, it's this archaic textual display logic that the linux kernel uses to do early boot logging before wayland/x11 take over, and that you can log into via Alt-F2, Alt-F3, …)

i.e. open /dev/tty1 temporarily, then issue the VT_GETSTATE ioctl call on it which tells you which VTs are currently allocated via a bitmask. Look for the first free VT (i.e. determine lowest unset bit, then format /dev/tty%i with it plus one. Then open that, and use that. switch to it via the VT_ACTIVATE ioctl, then clear it by output ANSI_HOME_CLEAR on it, then display the message there.

If this sounds like a bit much, grep our sources for VT_GETSTAT, we already call that ioctl elsewhere for other reasons.

[halfline]

VT_OPENQRY might be easier than VT_GETSTATE and __builtin_ctz(~vt_stat.v_state) or whatever.

[1awesomeJ]

I read that the APIs return errno-style errors on failure. Kindly review my choice of error types sir.

I made some of the requested changes.

I still have the SD_ID128_FORMAT_STR() and SD_ID128_FORMAT_VAL() outstanding. Still need to study them more.
And of course VT.

But I don't want to miss the chance of getting a review in case you view the PR, so I decided to push some first.

[1awesomeJ]

I looked up the format string and format value macros sir, and I've now used them in the patch.

So it's VT I need to look up now.

perhaps I have too many log messages?

[1awesomeJ]

Thank you so much @yuwata for the very detailed review.
Let me fix.

[1awesomeJ]

@yuwata, I made all the requested changes.
Sorry I took long.

[bluca]

needs a rebase on main due to conflicts

[1awesomeJ]

needs a rebase on main due to conflicts

Okay sir.
I'll fix in about an hour.

[yuwata]

LGTM. Let's go in this form at least now.
Future tasks are

• merge two qrcode logic more,
• add tests for bsod.service.

[bluca]

When you rebase and push, please also amend the commit message, as this is not about PID1, it's a different tool

[1awesomeJ]

When you rebase and push, please also amend the commit message, as this is not about PID1, it's a different tool

Certainly sir.

[1awesomeJ]

LGTM. Let's go in this form at least now. Future tasks are

• merge two qrcode logic more,
• add tests for bsod.service.

Alright sir.
I hope "future" turns out to be next week or next month.