Add new rc: machine_id to generate /etc/machine-id

[tcberner]

Summary

This new default-enabled rc will generate a /etc/machine-id file if it
does not exist, and prefill it with a newly generated UUID of version 4
[2].

This file is amongst other things used by libraries like GLib.

Bump FreeBSD version to be able to easily add support for older version
of FreeBSD via a package.

[1] Linux machine-id(5): https://www.man7.org/linux/man-pages/man5/machine-id.5.html
[2] f176fe8e7f638e585afcd2f4dd52a522c4648f63

[tcberner]

• fix names

[arichardson]

libexec/rc/rc.d/machine_id

25

[tcberner]

• Fix message

[tijl]

libexec/rc/rc.conf

698

I think osrelease above uses /var to support read-only /etc. So maybe you should use something like /var/db/machine-id here and make /etc/machine-id a symlink like /etc/os-release.

[jlduran]

libexec/rc/rc.d/machine_id

32

Do we want this message displayed every time after the machine-id file is created?

[jlduran]

libexec/rc/rc.d/machine_id

32

Please, discard the comment.
It was supposed to be just the removal of -n (should be echoed with a trailing newline character) and the removal of the space at the beginning of the sentence.

[bcr]

libexec/rc/rc.conf

698

Does the comment have a superfluous "if" in it (the first one)?
I think it would be better if the comment would state:
"Create /etc/machine-id on boot if missing (or NO)."

[arichardson]

libexec/rc/rc.conf

698

I think osrelease above uses /var to support read-only /etc. So maybe you should use something like /var/db/machine-id here and make /etc/machine-id a symlink like /etc/os-release.

If you decide to do this (which probably makes sense), you'll also need to add the symlink to etc/Makefile (see rS354922)

[tcberner]

libexec/rc/rc.conf

698

does it make sense to have an "ideally" static file in /var/db if so, sure, I'll move it there, and create a symlink

libexec/rc/rc.d/machine_id

32

the machine-id file should only be generated once, ideally.

[tcberner]

• Don't spam message log if machine-id exists
• Generate machine-id in /var/db
• Fix comment
• Create symlink from /var/db/machine-id to /etc/machine-id

[tcberner]

libexec/rc/rc.conf

698

^ does it make sense that machine_id_file this is configurable given the etc/Makefile change?

[tcberner]

• Add machine_id to libexec/rc/rc.d/Makefile

[imp]

All generated files should wind up in /var/db, imho. I have lots of read-only root machines that would need yet another special case for this. They have writeable /var/db partitions though.

[bapt]

LGTM,
I would personnally suppress all configurability in rc.conf, but I don't think it hurts to be able to do extra config via rc.conf.

please MFC After: 1 week

[freebsd_igalic.co]

this says MFC 1 week. that was two weeks ago.
This revision links to machine-id(5) on Linux, but perhaps we should have that man page ourselves

[dan.mcgregor_usask.ca]

This is good, but it doesn't match what Linux does. Both systemd and dbus generate 32 character UUID strings, while this creates 36 character ones.

I'd suggest we strip the '-' characters from the machine-id file, and possibly think about ensuring that dbus's idea of the machine-id is the same as the one generated here, or since dbus will use /etc/machine-id if /var/lib/dbus/machine-id doesn't exist, maybe not generating the latter at all.

[jlduran]

In D37722#863538, @dan.mcgregor_usask.ca wrote:

This is good, but it doesn't match what Linux does. Both systemd and dbus generate 32 character UUID strings, while this creates 36 character ones.

I'd suggest we strip the '-' characters from the machine-id file, and possibly think about ensuring that dbus's idea of the machine-id is the same as the one generated here, or since dbus will use /etc/machine-id if /var/lib/dbus/machine-id doesn't exist, maybe not generating the latter at all.

@tcberner regarding this comment, should we change it to /bin/uuidgen -r | tr -d - > $t, or something similar? or is not really necessary to strip the -s?

[tcberner]

In D37722#879899, @jlduran_gmail.com wrote:

In D37722#863538, @dan.mcgregor_usask.ca wrote:

This is good, but it doesn't match what Linux does. Both systemd and dbus generate 32 character UUID strings, while this creates 36 character ones.

I'd suggest we strip the '-' characters from the machine-id file, and possibly think about ensuring that dbus's idea of the machine-id is the same as the one generated here, or since dbus will use /etc/machine-id if /var/lib/dbus/machine-id doesn't exist, maybe not generating the latter at all.

@tcberner regarding this comment, should we change it to /bin/uuidgen -r | tr -d - > $t, or something similar? or is not really necessary to strip the -s?

Moin moin

I probably misread the documentation that said that modern systemd creates valid UUIDs. But that probably did not mean also it uses the 'dashed' format.

mfg Tobias

[jlduran]

In D37722#879900, @tcberner wrote:

In D37722#879899, @jlduran_gmail.com wrote:

In D37722#863538, @dan.mcgregor_usask.ca wrote:

This is good, but it doesn't match what Linux does. Both systemd and dbus generate 32 character UUID strings, while this creates 36 character ones.

I'd suggest we strip the '-' characters from the machine-id file, and possibly think about ensuring that dbus's idea of the machine-id is the same as the one generated here, or since dbus will use /etc/machine-id if /var/lib/dbus/machine-id doesn't exist, maybe not generating the latter at all.

@tcberner regarding this comment, should we change it to /bin/uuidgen -r | tr -d - > $t, or something similar? or is not really necessary to strip the -s?

Moin moin

I probably misread the documentation that said that modern systemd creates valid UUIDs. But that probably did not mean also it uses the 'dashed' format.

mfg Tobias

Ahh, I see:
https://www.man7.org/linux/man-pages/man5/machine-id.5.html#RELATION_TO_OSF_UUIDS