lw: botsin.space/@lexknowssystems/111791073584612720

Title: Lex knows systems: "" - botsin.space

_xor: what kind of doorbell camera is it?

sorry was scrolled far up for some reason

if a daemon starts up and creates a socket file, should it also delete the file when terminating? or is there another way to cleanly shut down sockets?

mason: yo, were you the one with the jail(8) segfault?

kevans: Yeah, parsing a bad config.

bad

mason: full name or just "mason" for attribution?

Mason Loring Bliss, although I only whined about it.

This is much different from fixing.

kevans: Abuse and scorn for not doing the legwork and submitting a patch might be more appropriate.

Sir, this is a Wendy's.

heh

mason: finding bugs is easy, actually reporting them deserves recognition imo

good repro is gold

Next thing to figure out (or implement if it doesn't exist) is to figure out jail numbers that's fixed on jail start but dynamic at rest. Which is to say, rather than having $ep = <somenumber> in my /etc/jail.conf.d/foo.conf I have $ep = $magicnumber for purposes of epair allocation.

like numbered taps for bhyves?

Possible? I haven't used Bhyve for more than "oh yeah, that seems to work" yet.

alepzi: Example /etc/jail.conf config: bpa.st/G7BA and then in /etc/jail.conf.d/foo.conf: bpa.st/4YMA

Title: View paste G7BA

So I'd like to have a magic number rather than hardcoding $ep in the config, with the notion that I could then ship the jail dataset (and reconstitute the config from properties) and not have to figure out a system-specific number.

Hm, I bet I can just use the jail ID.

Got to figure out when the jail ID is first known.

mason: cgit.freebsd.org/src/commit/?id=768…9ff156e7d34a035853030bf946ca94cb461

Title: src - FreeBSD source tree

parv: missed you on the lagg thing from yesterday. I have been running lagg with NIC and wifi since 11 or 12 on my laptop. always worked great. same subnet of course.

my router has been crashing, always during the night. To debug it I left a couple of tail -Fs running from my laptop. The problem is, the FreeBSD running laptop crashes too...

pstef_: no dumps?

surprisingly, no. Probably I don't know how to turn them on

_xor: relevant to your interests lists.freebsd.org/archives/freebsd-ports/2024-January/005181.html

Title: Call for help: moving manpages to share/man

how can I see where a certain port is defining it's dependenciens ?

last1: dependencies should always be in Makefile

that I knew, but this one I Have no clue where it's coming from

somehow my nginx port is depending on libcrypto.so.12 which is openssl 3

which I don't have and don't want to install

nginx-1.24.0_14,3 depends on file: /usr/local/lib/libcrypto.so.12 - not found

is that defined somewhere else ?

that comes from security/openssl, if you don't want the nginx package to use that, then select base openssl when configuring it

hmm, or does nginx not have an option for this?

last1: try setting DEFAULT_VERSIONS+=ssl=base in make.conf and rebuilding it

(if you use poudriere, make sure to set that in poudriere's make.conf)

Seems I fixed my lagg setup on my laptop. I had some weird setup mistakes in /etc/rc.conf which somehow worked fine on 13.2 and before. But actually should have given issues back then too...

that did it

weust: what was the issue?

meena: for some reason in the past I had SYNCDHCP included on my ifconfig_wlan0 line, while having DHCP on the ifconfig_lagg0 line.

That meant that the default router was using wlan0 instead of lagg0

oooohhhh

and I had the MAC address to em0 set to the MAC address of em0 instead of iwm0. but in my dhcpd config I had a commented out line for the iwm0 MAC, and one for the m0 MAC.

em0*

No idea when or why I did that. the laptop is from mid 2018 so somewhere in the past 5 years I must have messed with it for some reason

Since I copied that setup over to the RPi4 I am now changeing that setup to reflect the laptop

Hello. I have installed FreeBSD 14. During boot it seems to be trying to mount da0 but "Medium not present". Could it be trying to mount the disk I installed from?

or ... maybe not trying to mount but looking for it

msamsing: like this? "cd0: Attempt to query device size failed: NOT READY, Medium not present - tray closed"? that's just reporting that it can't identify the disc in the drive because there isn't one, it's not trying to mount it

lw: Yes. Like that. Is it possible to fix it so it is not looking for da0?

you may be able to forcibly hide it with hints somehow, but... why?

lw: I was just wondering why it is looking for something that isn't there - and is not supposed to be there

how would it know it's not present until it looks? ... this is just a normal part of querying available devices on startup, every OS does this

it's not trying to do anything with it, it just wants some basic information about the device

lw: Ok. Good to know it's not a problem

am i reading the -CURRENT makefile right? that i can not just build man pages, in a /usr/src of current?

i see that i can do a mandoc -Tlint <source file>

kevans: Neat. Thanks for helping fix it!

nginx-devel from ports, with lua enabled: [alert] failed to load the 'resty.core' module (github.com/openresty/lua-resty-core); ensure you are using an OpenResty release from openresty.org/en/download.html (reason: /usr/local/share/lua/5.1/resty/core/base.lua:24: ngx_http_lua_module 0.10.25 required)

Title: OpenResty - Download

any ideas?

13-stable, nginx 1.25.3, openresty 1.21.4, lua-resty-core 0.1.27, luajit-openresty 2.1.20230911_2

Onepamopa, have you tried the nginx port to see if the same issue occurs? Do you need 1.25.x?

I'm installing it from port www/nginx-devel

And yes, I do.

Have you tried using the latest openresty module? Might be a compatibility thing...

installed all I mentioned from ports ... same thing

maybe the ports in mention haven't been updated recently by the respective maintainers, idk

Well the error message seems to say it is looking for a later version of the ngx_http_lua_module than what you have...

isn't ngx_http_lua_module built when "lua" is selected in port's make config ?

cause I've installed everything I could think of - no change

I think it probably comes from the openresty port... but I'm not 100% sure.

does phabricator email you automatically if you're the author on something that's committed? i got some mails from it about commits with my name on and i'm not certain why

solved by: /usr/local/share/lua/5.1/resty/core/base.lua ==> or ngx.config.ngx_lua_version ~= 10025  --> or ngx.config.ngx_lua_version ~= 10026

kqueue(2): "EV_CLEAR: After the event is retrieved by the user, its state is reset." - what does this mean? what state?

it also says, for EV_TIMER, "On return, fflags contains the events which triggered the filter." - does that mean fflags will contain NOTE_ABSTIME if it's an absolute timer?

lw: it's probably intentionally vague because the state it clears is different based on the hint, but see conditionals that check for EV_CLEAR, here is one example: github.com/freebsd/freebsd-src/blob…n/sys/kern/kern_event.c#L1026-L1029.

Title: freebsd-src/sys/kern/kern_event.c at main · freebsd/freebsd-src · GitHub

dstolfa: hmm. i'm curious specifically about the behaviour for EVFILT_SIGNAL because it says EV_CLEAR is set, but i don't see anything related to EVFILT_SIGNAL there...

lw: github.com/freebsd/freebsd-src/blob…461/sys/kern/kern_sig.c#L4308-L4319 this is where the data gets populated, and this is what it means by EV_CLEAR being set: github.com/freebsd/freebsd-src/blob…461/sys/kern/kern_sig.c#L4283-L4294

Title: freebsd-src/sys/kern/kern_sig.c at 768149ff156e7d34a035853030bf946ca94cb461 · freebsd/freebsd-src · GitHub

dstolfa: but what does EV_CLEAR actually do there? what changes about the event?

lw: in every knote that has it set, it will clear out the data and fflags. this usually means that it won't signal the event over and over again. for EVFILT_SIGNAL, this means that you can build your signal handling event loop in the "obvious" way

however, again, depending on the event type and various other things, it could mean something else which is why it's probably documented as vaguely as it is. i'm looking through the man pages to see if there's a good example of what it does, but i can't really find one that explicitly shows it

there's a lot of implicit stuff that makes sense if you know how it works, but could be confusing otherwise

hmm.

dstolfa: so the way i'd expect EVFILT_SIGNAL to work is i get one event per signal delivered to my process... but i don't want it to clear out fflags because what if another signal arrives before i can rearm the event?

has anyone used these instructions: wiki.freebsd.org/PkgBase to upgrade a jail/bastille to current? for testing purposes?

Title: PkgBase - FreeBSD Wiki

dstolfa: it's not clear to me if i reset fflags and re-arm the event, will it deliver the 'missing' signals then?

(this is at least how Solaris event ports worked, from what i remember... you set up your signal with sigaction() to tell it to deliver to the event port, then you just get one event per signal)

ofc freebsd is not solaris but this seems like the obvious way to implement it

lw: it's been a while since i looked at kqueue internals but i'm pretty sure you get all the signals, you just don't get the same signal multiple times when EV_CLEAR is set (but you would if it wasn't). that's the idea that the man page talks about when it comes to wanting to get notified about state transitions. there are a few examples on how you might implement a signal handling event loop in base,

see camdd or pmcstat

i'll have a look, thanks. this is basically described "edge-triggered" behaviour, right? whereas kevent is normally level-triggered

(might be easier if the manpage just said that...)

actually wait, is EV_CLEAR really just the "make this event edge-triggered" flag?

if so i am going to just add EV_CLEAR to all my kevents because i hate level-triggered behaviour

lw: depends on how you think of edge-triggered i guess. in this case, it would only trigger when the signal is delivered, but not when it's handled (for obvious reasons)

right

so like, a signal is delivered, then i call kevent() another 20 times, i don't get any more notifications, until i turn the event back on

wait, no. until another signal is delivered.

(otherwise that would be EV_ONESHOT which is different)

... i might write a test program to understand how this works then submit a diff to the manpage

EV_CLEAR sounds like what we want for threaded programs though where you get an event, dispatch it to a thread pool, and might end up calling kevent() again before the handler has had a chance to read() or whatever

lw: better docs are always welcome :). might be helpful to even include an example that clears up stuff that you've run into so even if it's confusing in the text, there's at least code that demonstrates the interesting bits of behaviour someone might care about

am i going nuts or is there NO limit to how many "clones" of a nic you can make?

voy4g3r2: what do you mean by clones

i have the physical re0 and i want to have one for jails (VNET) and one for bhyve

errr, you probably don't want that

it looks right now, when i try to "connect" the bhyve to re0 it is "busy" and it keeps mentioning clone the interface

haha

could explain why it does not work

create a bridge, put re0 in the bridge, create an epair for the jail, put that in the bridge, create a tap for the bhyve, put that in the bridge

a tap is NOT a bridge? or am i just overthinking the terms

correct, a tap is not a bridge

a tap has exactly two endpoints, one is the kernel device, the other end is whichever program has the tap device opened

and if i want to have -CURRENT builld i CAN NOT put it in a jail.. think jail

an epair also has two endpoints, but both of them are kernel devices, so you can put one end in a VNET jail

yes, do not use tap for jail, use epair

okay.. and i need to use bhyve for -CURRENT?

voy4g3r2: it's a bit long (sorry) but here's a real-life example: le-fay.org/tmp/30d/RPXZZo.txt - tap0 is a bhyve vm, all those epairs are jails, ix1 is the external interface, and bridge0 is the one bridge to rule them all

if you want to run -current on 14.0 (or whatever) then yes you need bhyve

okay.. i am at point hwere bhyve has a freebsd-current image just the network is NOT working

and if i use that txt (thank you and nice on tolkein reference) i should get network

this is the rc.conf for that setup which is actually much simpler: le-fay.org/tmp/30d/6Vle6n.txt

and now i get why some of these tutorials i am reading, say have a multiple nic card in your machine :)

notice the two instances of 'auto_linklocal' which are important, ipv6 will break without that

wow, very clean and easy to follow.. i think you have been doing this quite a while

i am not cool enough to have ipv6

i tried to setup on home network and it was just acting up..

voy4g3r2: this machine has an onboard re(4) (unused) and an Intel X540-AT2 with two ports (ix0 and ix1)

that was going to be next question is ix your nic card

s/card//g

yes, ix1 is plugged into the switch

lets see what i can break for being the bull in china shop

thank you for the guidance

i waant to get this man page pull request over to current because seems like "wasted" effort to do on 14.0

and i probably annoyed some people with all the udpates to the pull request.. i got some awesome help to break up the files into smaller commits

voy4g3r2: fwiw, rather than trying to do this with ifconfig, i suggest editing rc.conf based on what i pasted above, reboot, and see if it works. (take a back of rc.conf first in case it doesn't.) the main thing to be aware of is for a bridge, you have to put your IP addresses on the bridge instead of on the existing member interface

s/a back/a backup/

so for a single bhyve vm you want something like: cloned_interfaces="bridge0 tap0"; ifconfig_tap0="link0 up"; ifconfig_bridge0="inet 10.1.1.1/24 addm tap0 addm ix1 up" except replace ix1 with your actual nic

and replace 10.1.1.1/24 with your actual IP address

the reason you set link0 is on tap is so it stays up even when bhyve exits, otherwise by default it will go down

ah

i was noticing an entry when bhyve instance was up

then when i turned it off to do "work" it went away

but i also was not going down the path you were

so everytime you add a new bhyve.. you update your rc.conf to then add more taps/clones depending on what youa re using?

yes

vm-bhyve can actually do this automatically but i found that annoying, i prefer to do it by hand

don't like random scripts messing with my interfaces

yeah, that would make sense

i was watching garyhtech youtube and he was using BVCP..

i don't know that

so trying that.. but adding a nichttps://bhyve.npulse.net/

Title: BVCP: FreeBSD Bhyve Project

i really need to NOT have ifconfig_re0="DHCP"

it makes setting up bhyve a lot easier but it falls when it comes to this NIC stuff

i don't know how to run DHCP on a bridge and also add members to it, maybe someone else does...

i have been delaying the transition.. DHCP should not be on a server.. clients yes, servers no

it goes on the list of fixing today

thank you for the example lw

lw, does sublime-music really require python >=3.10 ?

jbo: yes, the setup.py is full of lies

sorry, i know this makes it really annoying to build. i went through the same thing.

yeah, that's why I haven't committed this yet

lw, how did you poudriere-testport this?

jbo: i use python 3.11

it appears to ask for dependencies with flavor 3.10 that do not exist or something

I hate this.

so DEFAULT_VERSIONS

hmm really?

Title: # poudriere testport -p main -j 132Ramd64 -o audio/py-sublime-music[00:00:00] - Pastebin.com

uh that's weird

jbo: does it work if you set DEFAULT_VERSIONS+=python=3.10 ?

that's AFAIK not how ports are supposed to behave so I didn't test that :p

i would have thought it would automatically build some 3.10 packages if you had 3.9 as default but i guess not

are you sure that RUN_DEPENDS is sound?

this also applies to something else i want to port soon that requires 3.11 so it would be nice to solve it...

===> py310-sublime-music-0.12.0 depends on package: py310-gobject3>0 - not found

===> py310-sublime-music-0.12.0 depends on package: /packages/All/py310-gobject3-3.42.2.pkg - not found

hang on, let me git am and test again

I attempted this three times last week and luckily always got some customer crap to take care of instead :p

but also want to get the PR done...

RUN_DEPENDS= ${PYTHON_PKGNAMEPREFIX}gobject3>0:devel/py-gobject3@${PY_FLAVOR} \

i mean, this seems right

let's ask #freebsd-ports

I'm gonna let you handle this one, ok? once it passed poudriere-testport I'll take over again

please do make sure that it works without "hacks" such as DEFAULT_VERSIONS

ok, it's not a "hack", that's just how i build my ports because i want a newer python. i will testport it a blank make.conf though

so poudriere-testport on a 13.2-RELEASE jail worked for you out-of-the-box with the latest patch of the PR?

you're 100% sure that you don't have a /usr/local/etc/poudriere.d/make.conf (or make.conf per-jail) that does funky things?

jbo: on -current. but with my local make.conf (which includes python 3.11) which i didn't realise it would be an issue, so i will fix that. i did not deliberately "hack" this to make it build...

err - I didn't mean to insinuate something or blame you - trying to find solutions so we can get this moving :p

apologies if my wording was... suboptimal

god i wish testporting was less of a fucking pain in the ass (not a rant at you, just in general)

trying to maintain a development branch of ports and get poudriere to actually use it is far more annoying than it should be

huh? poudriere-testport is like the least pain-in-the-ass I have ever experienced working on any system that had something to do with "integrating" upstream software

just register multiple ports

that would be easier if poudriere ports -B actually worked to switch branches (unless that got fixed...)

nah, just use METHOD=null

right now i have to commit to my dev branch, then push it somewhere, then pull that in the poudriere checkout...

that way you can just handle the ports repo manually

nah

it's like 5 commits just to testport a one-line change

s/commits/commands

hehe.. without fail. typos are evil!

fun fact, poudriere supports supplying it a patch, but not for git

Title: poudriere jail -P doesn't work with -u · Issue #1057 · freebsd/poudriere · GitHub

1057 – this is a junk PR bugs.freebsd.org/bugzilla/show_bug.cgi?id=1057

jbo: there are zero (0) ports which require python 3.10 or later and have RUN_DEPENDS other than audio/sublime-music. so... i don't know. maybe this isn't meant to work?

there are 3 ports which require 3.10+ and have BUILD_DEPENDS, but it's the RUN_DEPENDS which are the problem here

lw, is python 3.10 supported on 13.2 (and 14.0)?

meena, might now. I'm not a python guy.

s/now/know

jbo: i don't know, i can test that but it's definitely supported on 15.0 and it still doesn't work for me

:D

so for example i did # poudriere testport -p sublimemusic -j main textproc/py-TatSu

it would be nice to bump the default python…

and it's off building lang/python311 like you'd expect

but py-TatSu only has BUILD_DEPENDS

jbo: i will ask python@

lw, alright! wanna assign the PR to yourself for the time being? :D

jbo: i don't think i have the magic bugzilla flag to let me assign bugs to people

this is restricted to committers iirc

are you mad at me if I assign it back to you?

that's fine, i will comment when i solve this

sounds good :)

lw, mediaelch still waiting for approval from mentors btw

jbo: makes no darn sense that a Qt app is easier to port than a Python app

well, that is one of the reasons I try to stay away from python where reasonable :p

I'm sure python@ will be able to shed some light on your particular issue - I just don't have enough python knowledge to help more directly

I just see that it's my duty to make sure that it builds under reasonable configurations and it doesn't seem to :D

i didn't realise sublime was python when i first looked at it but it's literally the only navidrome client that exists that isn't written in electron

(well maybe not literally the only one but the only useful one)

electron... probably the only thing worse than python :p </SCNR>

lw, I can mostly recommend emby

(although you are probably more interested in jellyfin)

i already use jellyfin but it's really bad at reading music tags for some reason, half my music ends up under 'unknown artist'

i could probably fix that with mediaelch but tagging 1TB of music in mediaelch would take years

jbo: interestingly now that you assigned it to me i can edit the assignee, so i will give it you back once i solve this

works for me

lw, I have to admit that my "listen to music" experience is not as great with emby

lw, I usually make sure that the audio files I import are already pre-tagged

jbo: all my music is 100% correctly tagged (i am really anal about this) but jellyfin still doesn't like it :-(

i'm using musickube on an nfs share for now which actually works quite well, i'm not sure if i'll end up actually using sublime

but even if not i'll keep maintaining it, nice to have more music apps

sorry, musikcube, not musickube

as a maintainer of a port you're somewhat expected to actually maintain the port - just saying ;)

> but even if not i'll keep maintaining it, nice to have more music apps

i've already submitted upstream patches to it, i think that counts as maintaining it :-)

it counts as maintaining if you _keep_ maintaining it ;)

i mean i just said i would do that. i already submitted an update to py-pychromecast and tested it with sublime even though it's not even in ports yet

good - that's nice/appreciated

jbo: can you please re-paste your paste with the sublime error? pastebin.com says it's expired

lw, not really. that ship has sailed :p

lw, would have to get back to that machine tomorrow

ah nm then

lw: sorry to bother again, i am looking throguh your output and i am curious, why are is an 'a' added to the ifconfig but when i looked at cloned_interfaces there is no 'a'?

i see the bridge0 uses the a version.. just not sure i understand the cloned_interfaces line

random question, is there a specific reason the emacs port doesnt have nativecomp enabled by default? i usually compile with that option but dont feel like starting to build ports on this machine

cedb, default options questions are usually best directed to the package maintainer (i.e. by e-mailing them)

cedb, also: forums.freebsd.org/threads/ports-hu…he-disadvantages.91969/#post-639550

Title: Ports huh? What are the advantages and do they outweigh the disadvantages? | The FreeBSD Forums

voy4g3r2: epair is magical, when you clone one it creates both 'a' and 'b' automagically

voy4g3r2: i put the 'a' interface in the bridge and the 'b' version in /etc/jail.conf, but it doesn't matter, you could do it the other way around, both ends of the interface are identical

jbo: right sorry

ah that explains my next question on the epair2 it was the only one with both :)

you have a running jail most likely

voy4g3r2: the jail that normally has epair2b is not currently running

cedb, no need to apologize for anything

lw: and the ip address you are using as bridge, that is your "main" one

the .16

voy4g3r2: yes

got it, my freaking cheap router was offriding my ip address, got that fixed.. i guess the next thing is to just reboot

wait, no

and see what happens

this system's addresses are 10.1.6.13 and 2001:8b0:aab5:106::12

that was my typo.. i mean 13 not 16

i have way too many lines between these two files

i am old and had to print out to trace it all

well lets see what happens, thanks for confirmation

wow i got them, sweet!

err bvcp just use my bridge!! stop making your own@