Wlroots: Add ConsoleKit2 support

[PureTryOut]

I've been trying to package wlroots for Alpine Linux/postmarketOS, with the intention to eventually both run Sway and Phosh. wlroots compiled without any issues on Musl (for which I applaude you guys) and rootston runs fine when running as root.

From what I understand though, to run it without root rights, it needs a logind implementation. Since we do not have systemd (and thus no logind) or elogind, I'd like to request support for ConsoleKit2 as well.

[emersion]

On non-logind systems, you can set the suid bit on the compositor binary. The compositor should become DRM master and then drop privileges. Is that enough?

[PureTryOut]

It would do for now I guess, but the suid bit is discouraged and frowned upon on this distro due to security issues, so I'd still like ConsoleKit2 support.

[ddevault]

For Alpine in particular I've been slowly working on porting elogind. For now I use suid.

[PureTryOut]

Even then I'd prefer ConsoleKit2 support rather than using an alternative session manager than all current packaged DE's use :wink:

It does mean it's not as high priority any more though.

[unrelentingtech]

I have ConsoleKit2 working on FreeBSD, so I'll try implementing support for it here soon because of that :)

[ascent12]

With FreeBSD's lack of major/minor numbers for device nodes, I don't see how https://consolekit2.github.io/ConsoleKit2/#Session.TakeDevice is going to work.
That's the way to open devices without requiring root, so if that doesn't work, this whole thing would be pretty pointless for FreeBSD.

Being a DBus API, that means we'd need to pull in another dependency implementing that, as the current code uses libsystemd/libelogind. I don't like the idea of having multiple DBus libraries, but I dislike the idea of using something crappy like libdbus/gdbus even more.

Also, considering the lack of any major distro which uses this, I'm not convinced this is worth the effort.

[unrelentingtech]

When I said I have it working, I mean really actually fully working — I'm currently writing this from modified Weston that was launched with CK2 :) I can also start KWin, but that doesn't pick up input devices for some reason (needs more work).

lack of major/minor numbers for device nodes

That's not true, they are there, they just don't mean the same thing as on Linux:

These numbers are not divided in device classes and are not guaranteed to be stable upon reboot or driver reload.

Which works well enough for identifying evdev nodes:

% la /dev/input
total 1
dr-xr-xr-x   2 root  wheel    512 Jan 15 12:41 .
dr-xr-xr-x  16 root  wheel    512 Jan 15 15:41 ..
crw-------   1 root  wheel   0x17 Jan 15 12:41 event0
crw-------   1 root  wheel   0x1a Jan 15 12:41 event1
crw-------   1 root  wheel   0x53 Jan 15 12:41 event2
crw-------   1 root  wheel  0x2c0 Jan 15 15:41 event3
crw-------   1 root  wheel  0x2c9 Jan 15 15:41 event4
crw-------   1 root  wheel  0x2d0 Jan 15 12:42 event5

crappy like libdbus

What's so crappy about it? And why does that even matter? It's the standard, systemd-independent client library.

worth the effort

I'll do it either way, and if you don't end up accepting it, it'll have to live in FreeBSD's and Alpine's patches :(

[emersion]

Related: https://github.com/emersion/mako/pull/92

[emersion]

(Personal note: a logind alternative not relying on D-Bus would be nice to have)

[emersion]

We have no plans to support ConsoleKit2, as we now entirely rely on libseat for our session code. Maybe have a look at seatd for a maintained replacement.