you should be able to zpool replace zroot ada2p4 /dev/diskid/DISK-1234p4

aye aye, that's more or less what I was thinking

(Okay, so I did it the hard way....)

maybe a stupid question, but how do you get `/dev/diskid` populated ?

is it not?

doth not exist on this fresh 13.1-RELEASE install

If the drives are not mounted via some other identity, that's automatic?

what's `sysctl kern.geom.label.disk_ident.enable`

oh, that's zero, set in `/boot/loader.conf` I was looking at that recently too

probably reading the same page i was

think that's safe to enable and reboot?

yeah

yolo *reboots*

there `/dev/diskid` is, let's party

I vaguely recall that if you're using gpt partition labels, those work too. /dev/gpt/ IIRC.

hm, rtprio I don't think this replace approach will work properly. The Hard Way that tmp_ tried might be the path. There's no entry in `/dev/diskid` for any of the ZFS owned disks

aw nuts

That's probably why I did it the hard way. I'm pretty sure they only show up if not mounted via some other id.

yeah, I think that's probably right based on the tea leaves from forums.freebsd.org I have seen :)

Same with /dev/gpt/ labels.

what if you offline one of them first?

trying that now :P

offlining doesn't result in any entries showing up

zpool-labelclear(8) ?

i'm spitballing here

 /dev/ada5p4 is a member (ACTIVE) of pool "zroot"

womp womp

strawberry# zpool remove zroot ada5p4

cannot remove ada5p4: operation not supported on this type of pool

drats

how fond of this install are you?

it's intentionally fresh, so not at all xD

I had to remove each one.

Which, probably succeed due to being raidz1?

this is raidz1 too fwiw

(Or maybe it was that I had a spare on standby....)

I don't know ZFS well enough to understand why I cannot remove this device since there are four drives here

Did this during the FreeBSD 9 days.

rtprio do you think from bsdinstall I'd be able to get the diskids set up correctly?

boot the installer and check those two loader variables, to see if you can add them via diskid

Alternate: The gpt labels?

or gpt labels

so that setting is certainly enabled, but I cannot select by disk id at least through `bsdinstall`.

Dredging up more memory on this. I think there's something about zpool offline before zpool remove ?

I wonder if I could mount this zpool from the live shell here and do the replacearoo

to blast the old pool?

to IMPROVE the old pool!

I just might work.

s/I/It/

I'm well out of my area of expertise right now so I'm reading the zpool-import manpage

you'll want to set altroot

that's the big one, and not write to the zfs cache

on the liver system everything is definitely there in `/dev/diskid`

rtprio: altroot being a property on the import?

oh `-R`, I sees it

yes, otherwise it will trounce the current running system and things get very confusing from there.

okay, so I imported the pool with all the disk IDs

_now what_ xD

zpool offline one drive?

now you need to save the cache onto that pool or it will just import with ada next time

and i'm not sure how to do that

Let me check my notes.

the import was done with `cachefile` set to `none` because of the `-R`

zpool set cachefile=/altroot/etc/zfs/zpool.cache zroot

If you're chrooted to the /altroot, something like: zpool set cachefile=/boot/zfs/zpool.cache zroot

good news/bad news. The altroot contains `usr` `var` `tmp` and an empty `zroot` directory . Very confused now

pastebin `zfs list` plz

and you did specify an altroot, right?

I did! `/tmp/altroot`  which is where these are

this is running from a console, so it's not like I can gist the output from this live boot :P

I didn't look at the mounted volumes before I booted into bsdinstall

zfs get mountpoint zroot/root # (you'll need to figure out which)

is it none?

oh, it's probably "legacy"

for lols I rebooted, and those disk IDs persisted on the pool, which means because some things are missing that I've booted into single user mode. How interesting

I have a hunch on what I did wrong, back to the live shell

the pool came back as gpt ?

or ada

when I rebooted back to the installed system the zpool was listing disk IDs, but failed to into multi-user because it was complaining about missing stuff.  I've just rebooted and imported the pool again but this time rather than importing p4 I imported the whole device, which resulted in different behavior but the altroot still is missing

mountpoints

i haven't had a boot pool in a long time; and to think i was missing out on this sort of fun

you definitely are :)

I'm starting a re-install and going to see if I can swap to diskids before I reboot

i think if you had a mirror as your zroot you could have done the `zpool replacing`

Did you try zpool offline?

at which stage? I had tried that on the live system already

Any. Was just wondering how that went.

offline works fine, I cannot replace with the pool

Oh. Okay. I'm pretty sure when I did mine, I used zpool offline before zpool remove.

... And then add and resilver...

I'm wondering if I'll be able to do that from the live system, because modifying online is causing problems

check NomadBSD

hrm, I cannot remove. It says the operation is not available on this type of pool

Found one of my notes. I did something like offline <id>, add <spare>, replace <id> <spare>, remove <id>. Looks like you'd need a fifth drive to pull this off.

You're probably right, so this is a four disk raidz1 pool which makes me think I need to find another way

tmp_ would you mind dropping the `zfs list` output (sans any non-standard mountpoints if you'd like). I'm curious if I can do this manual at install time and get it right then

All 22? I mostly got them from wiki.freebsd.org/RootOnZFS/UFSBoot

Title: RootOnZFS/UFSBoot - FreeBSD Wiki

thanks

Critical points: The zpool itself "has" a mount point of /zpool, but is set to noautomount. zpool/ROOT mounts on / and then zpool/usr on /usr, zpool/var on /var, etc.

I'm actually really surprised at how difficult this is

I've worked with zfs in an enterprise setting. I'm not.

haha

That was many zpools on multiple versions of Solaris, for the most part.

that was a rocky time

Oracle put their oar in part way through, it got even rockier.

you mean like how oracle does with anything they touch

"We're not restricting you from using that; Oracle didn't provide it. It's not there."

Mind you, that was a set of things that Sun deprecated in 1994. As in, "This will go away someday. Don't make anything new rely on it." Oracle decided it was time for all those things to go away.

Notably, it was the developer teams that had the most problems with that....

huh, I cannot seem to remove this damn device

I thought maybe I would try raidz2

hrm, I cannot even detach a device from the pool after offlining it

I think the zpool is preserving the place for the device, so replace is the only option when it's offline. But replace requires a new device.

that sounds about right

is there a file floating around somewhere or something like that which I can just edit to change ada5 to the diskid xD

I have a different idea, I'll create the pool with 3 disks in a new install, and use the fourth like you did tmp_

Have the fourth as a spare?

yeah, there are four disks that are supposed to be there. Maybe I make a three node pool, do the swapping around there, and then grow it with the fourth disk after the shuffling to diskids is complete

I'm drawing a blank on whether expanding the number of active disks in raidz is possible. I'm suspecting not.

The notes I've dug up suggest I used a USB disk as the spare when migrating.

It was kind of unstable. A little. Just a tad.

laffo

Is anyone aware if there are any PRs for "bsdinstall" to create a (GPT one if using GPT for example) label for ZFS root pool?

parv, welcome to the party  ;)

tmp_: no, can grow from 4x 2tb to 4x 4tb but not 8x2tb

rtyler, Hi there

Yeah, I've grown 4x 1tb to 4x 2tb a time or two.

we've already been fighting a lot against ZFS tonight, I'm not sure there's anything planned for bsdinstall

honestly this would be a much less frustrating problem for me if the onboard sata controllers devices were starting at 0 rather than the pcie card's

put the boot pool on the pcie cards?

wait, rtyler are you going to install to 3 +spare and then convert to gpt?

if you're going to install fresh i have a different idea

I've installed fresh a few times tonight, my brain's fried, whatcha got

trying to find the wiki article

basically you shell out and set up the pool, and then exit back to the installer

and since you shell out rather than letting bsdinstall do it, you could add them by the gpt or diskid names

mason's notes on "ZFS and GELI by Hand" : wiki.freebsd.org/MasonLoringBliss/ZFSandGELIbyHAND

Title: MasonLoringBliss/ZFSandGELIbyHAND - FreeBSD Wiki

rtprio, Is the one you are looking for listed at wiki.freebsd.org/CategoryZfs ?

Title: CategoryZfs - FreeBSD Wiki

the one i followed was installing for dualbooting but the procedure would work

let me spin up a vm quick and see

hmm,

manual -> gpt -> label it -> create pool

rtyler, Since you are willing to reinstall, I propose to use manual ZFS install: (GPT) partition the disk as you like; add GPT labels; install ZFS on root manually on one of the partitions. Note that everything will be dumped directly on the "pool". Then create separate datasets; move stuff out from root pool to separate datasets

but 'shell' was the document i followed

it did it all by hand: When finished, mount the system at /mnt and place an fstab file for the new system at /tmp/bsdinstall_etc/fstab

parv: manual won't create the pools it seems you have to do it by shell

# ls /dev/diskid/

DISK-BHYVE-DAA9-3899-15E2 DISK-BHYVE-E466-C286-81C7

but there are the disks; create the pool and off you go

rtprio, bsdinstall would create a single root pool & no separate datasets (which otherwise are created in auto mode)

yah, to do what you want, you've got to pick the 'shell' option

Title: 262461 – bsdinstall: After manual partitioning the disk, installer installs "ZFS on root" directly on the pool not under separate ZFS datasets

wiki.freebsd.org/RootOnZFS/GPTZFSBoot is what i follow, it guides you through the shell method

Title: RootOnZFS/GPTZFSBoot - FreeBSD Wiki

rtprio, Thanks for the link (GPTZFSBoot)

parv: i don't recall having that problem

rtprio, Problem, as stated in PR 262461, is that manual partitioning & installing ZFS-root -- all via bsdinstall UI -- does not create different ZFS datasets unlike when auto ZFS is used (which consumes whole disk & no label is created for the root pool)

yeah, so create them yourself

i don't like a few of the ones bsdinstall picks out anyway

*sigh* Yeah that was what I did eventually as clean up

oof

Re default ZFS datasets, I also have the same opinion but that is still far more preferable, workable than dumping everything on the pool dataset directly.

but hey, once you get it working, you won't have to touch it for a while

History for 'wd':

2013-02-16.21:19:45 zpool create wd raidz2 /dev/ada0 /dev/ada1 /dev/ada2 /dev/ada3 /dev/ada4 /dev/ada5

some of those drives had 2 hours on them when i created the pool

<antranigv> "What's the 2022 style of..." <- antranigv: I don't doubt that arcanist can be useful, but I haven't got that far. (I commit without using it.)

grahamperrin really? how?

grahamperrin I'm having a hard time contributing to FreeBSD. All I wanna do is just fork it on GitHub and make a pull request, or maybe use git's mail pull request functionality.

antranigv: I pinged you in FreeBSD in Matrix, continue there? Thanks

grahamperrin I'm in FreeBSD's matrix? hah I didn't know that :D

played with pf today and couldnt get a set of rules working that didnt block me from ssh! im so lame

antranigv: irc.libera.chat private message failed for some reason, do you have a Matrix client handy? (I use <app.element.io>, YMMV.)

Title: Element

Ah we're there

hi, I just click on restart Firefox in my BSD and I get this: XPCOMGlueLoad error for file /usr/local/lib/firefox/libxul.so /usr/local/lib/libnss3.so: version NSS_3.79 required by /usr/local/lib/firefox/libxul.so not found Couldn't load XPCOM.

any tip is welcomed

pkg install nss <- fixed the issue

what is the difference between pkg update vs pkg upgrade

?

pretty sure POWER is not a mainframe architecture

I was scrolled up

uskerine, "update" only "update[s] the local catalogues of the enabled package repositories" (from "pkg-update(8)" manual page) ; installed packages are not changed

uskerine: if you were _without_ an installation of security/nss, you should wonder how the absence occurred.

<freshports.org/www/firefox/#requiredlib> security/nss is a required library for www/firefox.

Title: FreshPorts -- www/firefox: Web browser based on the browser portion of Mozilla

I am using firefox-esr

uskerine: thanks, there's the same requirement <freshports.org/www/firefox-esr/#requiredlib>.

Title: FreshPorts -- www/firefox-esr: Web browser based on the browser portion of Mozilla

Reviewing, committing. <docs.freebsd.org/en/articles/commit…de_appropriate_metadata_in_a_footer> gives 're' as an example of the name of a team, however (unless I'm missing something) there's no such team in Phabricator.

Title: Committer's Guide | FreeBSD Documentation Portal

the nss was upgraded when I did pkg install nss

maybe I did not run pkg upgrade in a long time (as I see I have 209 packages to upgrade)

"DHCPACK from ... Starting sendmail_submit. sm-mta[814]: My unqualified host name (freebsd) unknown; sleeping for retry". The VM-IMAGES download just hangs at boot for no good reason :(

two minutes of my life - gone

Molnija: POWER is, POWERNV is what you find on anything else.

debdrup: I didn't think either of them was MF.

Well, it's possible I'm misremembering.

publibfp.boulder.ibm.com/epubs/pdf/dz9zr010.pdf sure makes it seem separate, so it's probably just me misremembering

yeah, the machine check handling is completely different, so it's gotta be a different thing all together

i mean, it still uses ECC, because pretty much everything does, but it's described entirely differently from how the OpenPOWER spec describes it

anyone: i think you can ^c during that

When I do a "service unbound restart" it works fine. If I do "service unbound reload" I get: " fatal error: Could not read config file: /unbound.conf. Maybe try unbound -dd, it stays on the commandline to see more errors, or unbound-checkconf"

the checkconf retuns no errors. Running unbound -dd shows no erros either.

path to file looks wrong, is it really in /?

No, that is why I am confused.

I am looking at the rc script

so with that path $local_unbound_workdir is not defined, which is why it expands to /unbound.conf

Where is that supposed to be defined?

and there's no local_unbound_reload function; it might be a bug

it's defined one line up,

Which script, sorry. I am not finding in /usr/local/etc/rc.d/unbound

oh, i was looking at /etc/rc.d/local_unbound

Oh, I am not running local_unbound

the real unbound

But I guess that applies workdir

I wonder if it is being borked by load_rc_config "${name}" (like it wasn't save dproperly)

eh?

It calls that twice.

When I do a "service unbound restart" it works fine. If I do "service unbound reload" I get: " fatal error: Could not read config file: /unbound.conf. Maybe try unbound -dd, it stays on the commandline to see more errors, or unbound-checkconf"

That, cpet

is it a ports rc.d?

Yes, pkg install unbound

why do you install unbound if its already in base ?

and why arent you emailing the maintainer rather here ?

Only local_unbound is in base, no?

Because I am trying to see if there actually is a problem

one listen on localhost other listend on all

basically what that means

Right, I need to listen on external ips

And I have it doing void-zones as well.

Is it preferred to use "local_unbound" and change it's conf?

depend son use I personally hate it and wish it uberly death but thats me

Hate what? unbound in general?

I hate bind, and djb is getting very old

it works

but the error is self explainitory

I use djb for lots of stuff, still.

it wants /unbound.conf and cant find it

Of course, but trying to figure out WHY it is looking for that

(where is the bug)

djbdns?

human error ?

I am troubleshooting unbound, but yes, I mean djbdns

I pronounce that "jebdns" fwiw

Likely, but which human, as I don't think it is this one.

lol

well you fail to point configs and errors

so good lukc getting help :)

so start pasting some pastbins

It has something to do with the script installed by the package

(semi-serious)

my car makes a weird noise but without install a ODB scanner on it I can guess it's the belt ?

I kinda figured other people used unbound as well. So they would have the same installed files.

But I guess nobody uses the pkg version ??

I use bind

After bind 4 bugs I never went back

all software has bugs

Of course.

all sotftware has bugs freebsd is no different

Some developers don't code well. They make more bugs

stoping the use of a program casue it has 4 CVE's wihc arent even remote is well ...

Title: ports - FreeBSD ports tree

try changing that back

There was also the speed issue. Back then it was old HW. djb was significantly faster

But the issue now is with unbound scripts ;-)

seeing how the root DNS still use NSD,bind

i fint that BS

Erhard: are you trying to turn off Unbound?

hes using the port version

No, I want service unbound reload to work correctly

what?

right

any reaosn why yopu dont email the maintainer ?

Becaue I am trying to find the bug first.

Erhard: how's /usr/local/etc/rc.d/ look with unbound installed from ports?

I just gave you saomething to try

That is what I am looking at Molnija

so...

maybe do `sh -x /usr/local/etc/unbound reload` and pastebin the output

: ${unbound_config:=%%PREFIX%%/etc/unbound/unbound.conf}

default from port

which tells me human error

sysrc unbound_enable=YES

er, nvm

this is too advanced for me

any reason why you dont paste your rc.conf or any other relative items instead of just saying "it doesnt work"

sh -x /usr/local/etc/unbound reload works fine. It is service unbound reload that fails.

Because it's not in rc.conf

bleh

you seem to know more yet your asking here for help

so enjoy

cpet: chill out man

LOL, ok

`sh -x /usr/sbin/service unbound reload`

rtprio: I've had a few tussles with cpet i think

rtprio: Good call

you think ? its either a truse or false statement

+ exec env -i -L -/daemon 'HOME=/' 'PATH=/sbin:/bin:/usr/sbin:/usr/bin' /usr/local/etc/rc.d/unbound reload

oh good lord now you're words lawyering

is that it?

I've never been able to work unbound anyway.

it requires a little more than cp .sample to .conf and running it

Erhard: ` env -i -L -/daemon 'HOME=/' 'PATH=/sbin:/bin:/usr/sbin:/usr/bin' /bin/sh -x /usr/local/etc/rc.d/unbound reload`

A friendly question: If you like djbdns, why not use djbdns?

root@vm:/home/chris # service unbound reload

rtprio: That has the same effect. Still debugging.

Title: Nov 19 12:48:26 vm unbound[82204]: [82204:0] info: server stats for thread 0: re - Pastebin.com

Just simply HUPping unbound causes the error

So perhaps the unbound config itself... Checking that

Nothing untoward there, must have to so with unbound being a jail.

Forgot to run "etcupdate extract && etcupdate diff" before source update & world building (docs.freebsd.org/en/books/handbook/cutting-edge/#makeworld ). Could|Should I do something else?

Title: Chapter 25. Updating and Upgrading FreeBSD | FreeBSD Documentation Portal

Here root is on a ZFS dataset, along with boot environment, if that would matter

question. say i want to build pkgs from source, can i build them on a beefier machine running a VM and then upload it to a repo and use them on the wimpier laptop?

Answer is "yes". "How?" That I do not know (involved setting up the other 'puter as the package repository)

s/involved/involves/

i know the how. i just needed the reassurance that i could

Ok then, get building!

not until the holidays.

and i have to set up my diagram for what i want done and what will need to be done

richardbanger: you might want to check out poudriere (if you aren't familiar with it)

i was reading about poudriere which prompted the question

i just wanted reassurance.

parv, and?

cpet, "and?" for what?

etcupdate*

I don’t get zus snapshot and rollback. I create a snapshot of zroot@test with -r, create a file with content, do a rollback of zroot@test and the file still exists. What am I’m missing here?

I had asked: I forgot to run etcupdate extract" before source update & world building per the handbook. Is there anything I could do to have the effect of "etcupdate extract" which would have the effect of running it before (source update & world building)?

s/zus/zfs (damn autocorrect)

parv: probably fine to leave until the next make world process

rtprio, Certainly can do

So, I can rollback a dataset like zroot/usr/home@test, but not the entire zroot@test?

weust: yes

OK. And there is no recursive option like there is with snapshot. So, I need to rollback every single dataset separately?

i believe so

From "zfs-rollback(8)" manual page: The -rR options do not recursively destroy the child snapshots of a recursive snapshot. Only direct snapshots of the specified filesystem are destroyed by either of these options. To completely roll back a recursive snapshot, you must roll back the individual child snapshots.

I overlooked that. I did read the man.

it happens, no problem

But it sucks. Imagine wanting to rollback zroot after a failed system update. Why is there a recursive for snapshot, but not rollback. That last sentence doesn’t say why. At least, to me.

just script it

Yeah, will have too.

zfs list -H -o name -t snapshot |grep 2022-09 |xargs -n1 zfs rollback

Thanks. Will save that. Will come in handy again soon, no doubt. Trying to setup my laptop partially with ansible.

Please make liberal use of the -n and -v flags on all zfs commands unless you use zpool checkpoints.

Liberal as in use whenever possible, you mean?

Sadly for both snapshot and rollback such flags do not exist

What is a "stable" or "current" installation of the OS called: branch, version, other? Not "release", right?

release

Hmm. Ok.

cpet, Thanks

called where?

So for the 802.11s mesh support - is there anyway to make freebsd use ethernet for the mesh backhaul?

If I had installed a "stable/13" from source, what do I tell someone else what "it" is besides being installed from source or snapshot?

stable

you say "13-stable"

(instead of 13.something release)

Right. Thank you

that was interdasting.

What is "stable" in the context, a "version", "type"?

version

Looking for a noun here

dev

non-release-engineered

after running freebsd for 25+ years still puzzles me why release stable and current is so hard

parv, 14.0-CURRENT what version am I running ?

i gave up running stable partly for that reason and have stuck with -current since 5

cpet, "14.0-CURRENT"

14-CURRENT

STABLE denotes the KBI/ABI won't change, CURRENT denotes that it stays current with the head of the tree.

s/won't/isn't meant to/

That's all there is to it, really.

RELEASE is the product that the FreeBSD project publishes, along with its documentation in the form of the handbook and articles.

If you're not running RELEASE, you're expected to be willing to learn how it works and/or already know.

KungFuJesus: docs.freebsd.org/en/books/handbook/…#networking-lagg-wired-and-wireless

Title: Chapter 33. Advanced Networking | FreeBSD Documentation Portal

weust: yeah, liberal as in use it as often as possible. I didn't know zfs-rollback didn't support it though, that seems like an oversight - but it's all the more reason to use zpool checkpoints then.

I understand the purposes of "stable" & "current" branches. It is just that when I was trying to mention elsewhere that one should subscribe to appropriate -(stable|current)@ mailing list, I could not come up with general word to refer to each (regular "RELEASE", STABLE, & CURRENT)

They're branches in the VCS, if that's what you're asking for.

STABLE is branched from main (which is the CURRENT branch), and RELEASE is branched from STABLE.

It's described in more detail here: docs.freebsd.org/en/articles/freebsd-releng

Title: FreeBSD Release Engineering | FreeBSD Documentation Portal

e

It's not entirely accurate anymore since nowadays even a dot-zero RELEASE involves creating a new STABLE branch.

..it also still references svn because 12 and 13 are still made using that.

I figure out why cpet got unbound to reload, and I did not. It seems void-zones-tools does not place the zone in the correct location under the config directory. This works fine with s start or restart, but not with a reload.

Not sure if that owuld be considered an unbound bug or a void-zones-tools bug.

Or a bit of both. Seems wrong that unbound would have no problem with starting, but would have a problem reloading. Seemes like a separate issue as to where void-zones-tools put the files.

And I think I found a security flaw in void-zones-tools I will have to address anyway.

probably a good idea to take it up with the author

it's not inconceivable that the problem with file location is an issue that can be solved in the port, either

worth looking into, at least

Yeah, I see why void-zones-tools is puttin ghte files in theother directory. It's because it is designed to work with local_unbound rather than the ports unbound.

So there would need to be a config option for using it with either.

However, I still think the behavior of unbound is a bug in terms of working on startup, but not on reload.

And passing the checkconf

I'm curious why you're using unbound from ports

Not that I think it's wrong by any means, just curious

is there an error in your config that maybe slips by, that causes it to crash in reload

I perhaps incorrectly thought that local_unbound was only for listening on localhost.

Erhard: heh

rtprio: Kinda. There is an include: that doesn't work with the reload

I mean, I can understand making that mistake if you just didn't think about it any more (which is something I'm sure I'd be capable of, too)

Erhard: and if you remove the include does it reload?

debdrup: Can you just as easily modify the local_unbound to listen everywhere? Is it really just unbound?

man local-unbound.conf looks like yes, it will listen on 0.0.0.0

rtprio: Yes, or after I fixed the location to be under /usr/local/etc/unbound

🤷

so its fixed now?

Erhard: src/contrib/unbound is the location of the files

I guess the question is.. what is different between local_unbound and unbound?

it really is just a vendor import

Title: src - FreeBSD source tree

Hmm. Then I guess I don't need the port version.

Annoying.

cgit.freebsd.org/src/tree/usr.sbin/unbound freebsd adds/changes a few things apparently, but i don't think it's major

Title: unbound « usr.sbin - src - FreeBSD source tree

When I saw local_unbound I stupidly assumed it was FreeBSD's answer to the systemd-resolver.. A neurtered local-only resolver that I didn't want.

I guess I was wrong.

Why the heck do they callit local_unbound and set it up like that?

It could just be configured to listen on localhost by default

isn't it?

It is.

But I mean rather than the rename and having a port

I just switched from djsdns a few years back.

djbdns even

How do I set options for ports in batch mode? Like `make -DWITH_MYSQL install BATCH=1` or something like that?

cgit.freebsd.org/src/commit/usr.sbi…e74eecf425ae72d87eb6e8a4f77eec87a5b has a bit more information it seems

Title: src - FreeBSD source tree

I _think_ it might also be related to being able still install unbound from a package repo and have it not conflict in any way, if for whatever reason?

Well, I think I will stick with the port for now, but good to know it works fine the other way.

my brain is inconceivable

it might also be because local_unbound can be controlled by a socket: cgit.freebsd.org/src/commit/usr.sbi…40c9a9ccc46480c36781e2205ec9565be45'

And there definitely is a security flaw in void-zones-tools. Rudimentary, local only, DoD mainly. But stupid practice nonetheless

* DoS mainly

INteresting, debdrup

I prefer as few vendor patches as possible.

hernan: BATCH is an environment variable, you can either set it for that instance of make or you can set it via setenv (or whichever method your shell uses)

well I'm trying to inject it into a jail with a jexec one-liner

`make BATCH=1` actually works. I'm just trying to set port options non-interactively

like we used to do before dialog allowed us to forget :)

i think you can set them in the environment also

right, but what's the right way to do it? and can I one-liner it as a make argument?

you should be able to; i check the makefile for which options to use

hmmm, `make showconfig` is pretty cool and instructive

ya

anyone here running FreeBSD aarch64 on M1 macOS with QEMU/UTM? I am able to run it, but it uses like 100% CPU on the host.

haroldp: you can also use make.conf to set things

There are multiple ways of doing it, no "right" way as that depends on what in particular you want to accomplish

yeah, or i could inject a /var/db/ports/${NAME}/options, but I don't think I love either option

build the port on the host and `pkg install` it on the guest

antranigv: there's also vmware fusion tech preview for m1, try that

I just remember we used to do this back in the good old bad old days before ports were tricked out with dialog-ed options

got it

make -C /usr/ports/mail/maildrop OPTIONS_SET="AUTHLIB AUTH_MYSQL" BATCH=1 install

OPTIONS_SET and OPTIONS_UNSET

use poudriere!

(I'm _almost_ maybe not entirely possibly perhaps semi kidding)

... no; poudriere is the hammer used to solve this (thumbtack)

I'd use it but I am too dumb to spell it correctly.

aaa finally

kern.hz=100

..but why?

according to the handbook: The most important step is to reduce the kern.hz tunable to reduce the CPU utilization of FreeBSD

it also reduces the task switching frequency

for what purpose, specifically?

so, don't do that if you care about response

no, freebsd is closer to soft-realtime in quite a lot of respects, so kern.hz doesn't control nearly as much as it does on other unix-likes

kernel scheduling is controlled by a quanta, a unit in the scheduler itself

Yes somehow, lowering that got my CPU from 350% to 5%

:D

its value can be found in the kern.sched.quantum OID

I assume there's SOMETHING in the system that's using that.

ohh lemme check that

what was using 350% of your CPU, and how did you meassuure it?

debdrup the VM was using 350% CPU, but the guest itself was idle (according to top)

what VM?

"according to top"?

which flags were you using?

debdrup I mean, top(1). the VM is the FreeBSD VM running on QEMU on M1 macOS

I get kern.sched.quantum: 94488; debdrup what do you get on bare-metal?

if memory serves, that's the default value

I wonder what uses kern.hz that affected this change

VM can stand for virtual memory and virtual machine, so it's not exactly easy to know which one people are talking about

debdrup oh, a virtual machine.

the real question is, what happens wne you set kern.hz to 1000 again?

oh I tried that! the CPU goes back to 350% :( and my laptop gets hot! :D

if you run top with the right flags, you should see in particular which kernel thread/or light-weight process is using what

I think it's polling something.

debdrup oh true. I should check

i thought fbsd called lwps threads, not lwps as illumos does

no, freebsd has both threads and lwps

ahhh, I can't believe the pkt server is dead

wHAT

i need to go code diving

Molnija: you don't even need to go code diving, both ps and top will let you see IDs to confirm it

I don't disbelieve you

I mean that I need to go code diving to understand

antranigv: -IqSuZ as root should get you closer

Molnija: if memory serves, it dates back to kernel M:N threading that was eventually backed out, whereas this stayed.

Or maybe it came in with libthr? I don't remember :/

huh, it's rand_harvestq

so it's something specific to the hypervisor that the qemu frontend is using

I have considered implementing user-level M:N M>N threading before

am I likely to go insane?

seems likely.

antranigv: what's your kern.random.random_sources OID set to?

debdrup VirtIO Entropy Adapter

Yep, I thought so.

mew

antranigv: what version is this?

debdrup FreeBSD 14.0-CURRENT #0 main-n259276-5d42ef55dec1

what should I do? remove the RNG device?

that's probably the easiest

but that isn't really going to solve the issue

i'd suggest you file a bug