I am not sure if we should make our services read load more configuration files or perhaps solve this issue with symlinks.

There is a second problem I've discovered... ifconfig_DEFAULT= only works in /etc/rc.conf. It does not work when put into /etc/rc.conf.local or any of these files in /etc/rc.conf.d/

I kinda wonder if load_rc_config $name should be the last one loaded so /etc/rc.conf.d/dhclient overrides the other files...

Not in -current, tagging for srcmgr review.

Hmm, actually, our documentation is slightly contradicting.

• Rebase
• Add tests

Thanks!

In D45855#1160605, @hrs wrote:

https://people.allbsd.org/~hrs/FreeBSD/f1692c.diff

In D45855#1160514, @kenrap_kennethraplee.com wrote:

In D45855#1160497, @hrs wrote:

Could you explain your configuation and behavior change in more detail?

Sure thing. I use the following in my rc.conf which enlarges the virtual terminal fonts from being small after loading a drm-kmod driver during the booting process:

allscreens_flags="-f spleen-16x32"

After doing a src tree build upgrade today this functionality had stopped. By manually removing both set -o noglob and set +o noglob lines, I restored the original functionality.

The problem is that allscreens_flags relies on the pathname expansion in two places:
https://github.com/freebsd/freebsd-src/blob/4deb9760a9d84d5861ee45162ffebe83f13503b8/libexec/rc/rc.d/syscons#L237-L243
https://github.com/freebsd/freebsd-src/blob/4deb9760a9d84d5861ee45162ffebe83f13503b8/libexec/rc/rc.d/syscons#L381-L387

In D45855#1160497, @hrs wrote:

Could you explain your configuation and behavior change in more detail?

Sure thing. I use the following in my rc.conf which enlarges the virtual terminal fonts from being small after loading a drm-kmod driver during the booting process:

allscreens_flags="-f spleen-16x32"

After doing a src tree build upgrade today this functionality had stopped. By manually removing both set -o noglob and set +o noglob lines, I restored the original functionality.

In D45855#1160402, @kenrap_kennethraplee.com wrote:

This change introduced a regression where my allscreens_flags settings gets ignored at boot.

This change introduced a regression where my allscreens_flags settings gets ignored at boot.

Hi Mark, thanks for your review.

Could you please explain what the diff is actually changing? It is quite hard to understand, for someone not very familiar with this code.

Removed local _ifconfig_descr leftover from previously corrected autoif() hack

Replaced PoC diff with real tested diff - sorry for the mess.
This version corresponds to what the summary and the bug report (https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=285947) describe.

Better 'ifconfig description' handling - previous hack wasn't limited to ifn_start condition.

Looks good (not run tested).

Approved.

Seems sane enough. It's not something we use today, and I've verified that run_rc_command should do the right thing w/o side effects (although I'd think that the -z rc_arg test should be higher in that function to avoid pid file name eval, etc).

Approved.

Thank you all for feedback! :)

• Add UPDATING entry
• Fix typos in rc.subr.8 and do some extra wordsmithing there

I'm sure it adds appreciable speedup in some cases, but I can't see us fixing support and keeping it working. Best to stop pretending and focus on lower hanging fruit.

In D47264#1078101, @0mp wrote:

I've got some reports from the NetBSD community that setting rc_fast_and_loose on NetBSD booting on vax simh makes rc finish in 109 seconds instead of 183 seconds (https://mastodon.sdf.org/@mrgtwentythree/113364676304360235).

At the same time, brooks@ mentioned in Bugzilla (https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=282255#c4) that the benefits of rc_fast_and_loose are probably negligible.

I've got some reports from the NetBSD community that setting rc_fast_and_loose on NetBSD booting on vax simh makes rc finish in 109 seconds instead of 183 seconds (https://mastodon.sdf.org/@mrgtwentythree/113364676304360235).

Approved, plus a minor nit.

In D47264#1077729, @markj wrote:

Approved.

This probably deserves an entry in UPDATING as well as the relnotes tag, since it breaks backward compatibility and applies to those who track main as well as release users.

This probably deserves an entry in UPDATING as well as the relnotes tag, since it breaks backward compatibility and applies to those who track main as well as release users.

Update docs and remove an unused boottrace function

Thanks!

Approved.

I'll set MFC to 2 weeks so that it does not land in 14.2.

Set exit status by calling exit in a subshell.

I opened a PR for investigating the rc_fast_and_loose support: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=282255

Check all invocations.

In D46862#1076197, @gshapiro wrote:

This is an improvement over the current functionality, but I think we would want all invocations to be checked. I wonder if it would be better to separate these into separate rc scripts instead of using one for four possible different daemons. That shouldn't prevent this change from going forward while alternatives are formulated.

This is an improvement over the current functionality, but I think we would want all invocations to be checked. I wonder if it would be better to separate these into separate rc scripts instead of using one for four possible different daemons. That shouldn't prevent this change from going forward while alternatives are formulated.

LGTM. Only a minor nit: in the commit message there is a typo: "sendmai daemon ...".

Committed in https://cgit.freebsd.org/src/commit/?id=6fa42b91ca3f481912af98c4d49c44507eb1b8e1

Thanks!

Approved.