I'm looking at an issue with upgrading a system from 14.3 -> 15.1.

I think /var overflowed.

`freebsd-update install` => "ld-elf.so.1: Shared object "libsys.so.7" not found, required by "libc.so.7""

Ohkay, `setenv PATH /rescue:"$PATH"` so still have the most crucial utilities.

Every time I see where BSD features such as /rescue are available I am reminded that FreeBSD is a Real Operating System there to help you not punish you like some other operating systems.

Rule 1: DON"T PANIC!

Rule 2: Keep calm.

in a jail config, can i use append instead of assign for every instance of a var? like exec.start += "foo";

i reorder them sometimes and get bit by the odd exec.start = "..."; (=, no +=)

if it's valid, i'd rather just use += everywhere

kerneldove: without actually knowing I removed the = once on exec.prestart (and only left with the +='s) and apparantly it never broke

i'll try that tyvm

i'm only using it on the lifecycle hooks, so exec.start/stop/poststop/prestart

#metoo

karolyi: I think your problem report would have been better if you mentioned your use case of running fail2ban in a jail at the start, so the reader doesn't have to wonder why you would want to manage pf from a jail

if in jail config exec.prestart i create an epair, then rename it (exec.prestart += "ifconfig epair create" \n exec.prestart += "ifconfig epair0a foobar0a"), the jail temporarily uses epair0. so if a bunch of jails are configured this same way, then are configured to start in parallel, won't that mean using "epair0" could could get mixed up? like 1

jail while starting is epair0, another jail starting at the same time might get epair1, but then the rename part would fail from referencing the wrong epair

I determine which epair to create with "ifconfig bridge${vlan} addm ${epair}a" and have every jail definition have a ' $epair = "epair5"; # must be unique in every jail'

ya but that breaks down if you want to support jails having multiple network interfaces

ermm, yes, indeed (I do not have a use case for that)

so there's no way to atomically create and rename an epair nif?

arg. so i have a local samba server (in a jail), and i don't generally care about the performance, but at the moment it keeps having issues with my music player, where it's halting reading a file frequently. if i restart the track (reread the file), it works

as usual, nothing in the logs indicating the error

@joemie still there?

the longest epair i could make is epair11110(a/b), how's that make sense when ifconfig spec says name can be 15 chars long?

just, was driving home from work

I am having really weird behaviour with IPv6 and jails

neighbour solicit doesnt seem to be working despite icmpv6 being passed quickly as the first pf rule

I decided to ping6 the LLA of the host

and then global packets could pass

but before that they couldn't

wait no its working just fine

ugh

I asked this many times but I will ask this again, how do I specify multiple ifconfig command in rc.conf

I want to set a LLA of fe80::2/64 and also set the IPv6 address statically

but if I try to append them to the same ifconfig command in rc.conf

it errors

ifconfig_blah0_aliasN=

N being?

anything?

i think you'll want to start at 0

search for _alias in `man rc.conf`

rtprio: I only see this being used for IPv4

#ifconfig_em0_alias0="inet6 2001:db8:2::1 prefixlen 64" # Sample IPv6 alias

honestly it should work anyways, because ifconfig is run either way, I can just put inet6 in the alias even without ipv6_

yeah I know

is in /etc/defaults/rc.conf

lemme test this

rtprio: thx

life saver

rtprio: will relay to doc@ because this is not in the handbook and would be EXTREMELY useful

kerneldove: oh I bet this is going to be a really hinky bug

i figured something out just by brainstorming. didn't see it in any docs so dunno if it's intended to work. ifconfig create epair myifname0

yeah, that's fine. officially that's just interface renaming

ah shit it only renames the a side of it, not the b side, due to the epair bug in freebsd

ffs that bug needs fixing lol

my b-side all are known by the name jail0

how can they all have the same name?

and anyway the point is there's no way to rename the b side because i need to know the epair unit number to then do that

and the epair unit numbering is really restrictive/weird

from the jail pov it doesn't really matter how they are known as

hmm

but it does because in the jail's exec.poststop you gotta delete it

well you can just destroy the a side, but that still requires knowing the exact nif

ifconfig ${epair}b name jail0

I throught you could assign your own unit numbers when creating via ifconfig epair<number> create

you can't just ifconfig epair create in exec.prestart and somehow use a reference to the epair instance that was created

you can, but only kinda

joemie that depends on you being able to choose the $epair you want to use

the destroy only happens on the a-part: ifconfig ${epair}a destroy

yes, that's true

ya but your $epair still needs to be something you choose

correct, as stated before

so you can have gaps due to jails being temporarily disabled, no longer in service, etc

liek epair0, epair1, epair5...

the number of the epair is identical to the last octet of the ip address

kerneldove: just tested ifconfig epair3 create works fine and creates epair3{a,b} without any need for having epair1 or epair2

k but what if you have have more than a /24 going to the box?

nimaje ya but try ifconfig epair000000010 and it fails but it should be allowed because nif names are limited to 15 chars

then, try ifconfig epair0010 and it fails because leading 0 are disallowed ig? because ifconfig epair1110 create works

that's what i mean by the epair naming restrictions are quirky and limiting

i'm planning to look into this when i have a minute

dude if we could just do ifconfig epair create name foobar0 and get foobar0a/b it would be perfect

kevans ok very cool! the bug causing trouble here is that ifconfig <epairif> name ... has a bug that only the a side is renamed, not the b side too

well, how many epairs do you plan to use per jail? more than 10? if not you could use epair<id><digit> to name your epairs until that create name bug is fixed and you would be able to create at least 1000 jails, as ifconfig epair9999 create works too (epair99999 doesn't for some reason, but no idea how unit numbers work for network interfaces)

ya that was my idea, but that's how i found out ifconfig epair0...010 didn't work. the idea being nif names are 15 chars max, epair is 5 leaving 10, -1 for a/b, so 9 for the jail, then -1 so each jail can have up to 10 nifs, leaving 8, so epair000000010 is first jail nif on box, and first nif of jail

I used to admin hundreds of Solaris x86 9 & 10 boxes - most of them with Intel Gbit NICs.

well, the number there is not just a name, so maybe that is why additional restrictions apply

gonna have to use 1 instead of 0 as base jail id

The driver had to be manually in stalled (well, it was part of the image we deployed).

then just deal with the limited length

But most NICs in Solaris were called <some letters><digit>

Bit the Intel gbit driver called the NIC e1000gX

That caused so many problems.

ifconfig epair11110 create is longest possible

ah no because epair99999 was rejected

epair1110-epair9999 is practical limit, + a/b

kevans ^ fwiw

kerneldove: ./if.h:56:#define IF_MAXUNIT 0x7fff /* historical value */

ahh

manage to find where the bug is that ifconfig epair create name foo only renames the a side of an epair?

kerneldove: that's not strictly a bug, renaming is a separate operation that happens after creation

but every other interface gets a complete, coherent rename, except epair

i think the a/b side split was an oversight

because every other interface doesn't have the janky paired-interface notion

:-)

but shouldn't our goal be to make it not janky? it doesn't seem like a flawed concept, just a lil buggy implementation

i think it's going to end up getting rewritten in some way in 16.0

wow, what's in store?

glebius is trying to re-work how operating between different vnets works

the current technique of creating it in the home vnet and 'moving' it out ends up creating a lot of hassle

so i think epair ends up getting hit hard here in some way

ya would be cool to refactor it so its init procedure is refactored to match how it works for host networking subsystem, just independent

hmmm yeah I cant get IPv6 networking to work in my jail unless I ping its LLA on the host os first

maybe there is a solicitation problem...

i have config snippet files for different daemons that i include in the config files for the daemons. where's the right place to copy that big dir tree of config snippet files?

i was thinking /usr/local/share/config-snippets

If you are storing for your own use, it's your perogative. Backup's not withstanding, I currently keep a copy of my base configs in my ~

we do the thing like $id then use it as the last octet for the ip and other uses of vars for per-jail config. but what do you do to extend the convention for jails that have >1 network interface? create 10 versions of the vars? (jails can have up to 10 nifs in my convention) but then even if you do that, you gotta generate up to 10 blocks of the

network related exec.prestart/poststart/start/stop?

looking for ideas to make that nicer

anyone else miss the lmdb snafu, recently? forums.freebsd.org/threads/how-do-i…-stuff-mentioned-in-updating.103179

drop.17es.dev/-GJ7bF7Ekoj any idea what's going on with all these in my zfs?

is the output for sysctl dev.hdaa is safe to post online for troubleshooting?

kevans: well, I would assume the renaming to apply to the pair when done as ifconfig epair create name something as I don't choose a part of the pair there

kevans 100% agree with nimaje there

yes, i don't think I'm arguing that you're wrong for expecting this

it's just not surprising because naming is a logically separate operation that happens on an interface

cloner drivers don't really get a vote in how renaming semantics work, you have to listen for renames and decide if you should do something about it

by the time that event comes in, you're unlocked and could already be in the process of being renamed again, iirc, so it's a little tricky

(then you also need to worry about racing against another thread trying to rename the other side of the pair, etc etc)

that also means you probably can't give them different names depending on where they're going, which maybe breaks something people do today

('them' being different sides of the epair; this one going in the host bridge you may want to name after a jail, while the jail side you might want to name 'wan0' or something)

yes, that's something i do today

i can see why one might choose to punt on figuring out how to handle renaming epairs specifically

how would it not be safe?

rtprio: I think I only saw something similar when docker decided to do some bullshit on some linux system that used zfs, because hashes are oviously good names for stuff without any further context (took a while to figuring out it was docker, as the dataset names didn't say anything about it and it just assumed it should make that mess, because it was running on zfs)

Given that ifconfig has worked like this for some years what's so hard about renaming both halfs of the epair? That's what I have always done. It works.

Example of creating an epair, renaming both sides of the pair, attaching it to the bridge, cleaning it up at jail shutdown. bpa.st/ASPQ

i think the problem is that we have syntax for renaming-at-creation time, and it doesn't clearly indicate that it's actually a rename rather than atomic creation+rename

So... Improving the documentation would solve the problem?

It's really not a serious problem. Just perform the steps individually and it all works as expected.

I am not even sure how one would modify the ifconfig syntax for naming the associated device. Call it "name" for the first and "name2" for the associated pair?

just add nameall for multicomponent interfaces

i'm already sobbing

ifconfig epair create nameall foobar

nameeach maybe

I am with kevans already. You can't name both interfaces the same name.

this is also a problem for exactly one cloned interface type, no?

I think epair is the only thing that creates two devices in one step. So it is the only one that is susceptible to this.

i keep also paging in dirty stuff with nmdm, which isn't a netif at all but obviously has similar semantics

i made a post on the FreeBSD forums about my sound card not outputting sound through the speakers. my post was awaiting approval. i don't think it got approved and i haven't received a reason as to why. what should i do?

glenny2: is your card supported ? what is your problem ?

it's the Realtek ALC257. there is no output from the speakers. i've tried searching for a solution. i've added (i think correctly) the device to device.hints. i've tried making sure the sound card drivers are loaded, etc. every post i've searched i've tried.

the mic input works but there is no output from the speakers. sysctl hdaa lists the speakers as [DISABLED]

what about 'cat /dev/sndstat' ? can you paste it somewhere ?

mzar: some users here tried troubleshooting the problem here in the #channel but were largely unsuccessful

and show us

ok

so perhaps this device requires some platform-specific settings

i tried adding the device to device.hints using the pin output (i think those are the correct terms) as this was successful for others with a similar issue

glenny2: have you tried switch units with command sysctl hw.snd.default_unit ?

mzar: yes

mzar: microphone input works correctly too

is it laptop ?

so input works

mzar: yep

if your laptop yes not supported yet, you need something similiar to this review reviews.freebsd.org/D49002

search reviews.freebsd.org, perhaps fix for your laptop is in the review

mzar: i saw that but i don't know what to do with it or the patch

I made my Dell working this way

you have to apply it on the sources, rebuild and reinstall the kernel

i don't think i have any sources?

if you really want to play with sources, you have to fetch them

i'm just using whatever was installed from the USB install

mzar: i don't want to play with sources really :( i haven't configured a kernel for linux in years. i don't know if i could do it on an OS i just started using a week or two ago

if you are not ready for the experimentation, plese wait until the official fix will be applied - but it will take some time - months

that patch was made over a year ago

yes, perhaps years

...really?

sure, but if you will help testing it, will give some feedback it will expedite the update

and if i wait, and it isn't solved, i'll have to wait even longer...if ever

this way you can contribute

i don't think this is a viable option. if i configure the kernel wrong, and the system breaks, it will take forever for my potato of a laptop to recompile.

OS: Linux ('Gentoo' '2.18') (6.18.35-gentoo-dist) CPU: Intel(R) Pentium(R) Silver N6000 @ 1.10GHz MEM: 4 GB GPU: Intel(R) UHD Graphics (JSL) (Vulkan) UP: 6 days 18 hours 36 minutes 19 seconds ago

this OS is not supported here

that's the hardware for the laptop

i know. the devices are the same

i don't have the /sysinfo command available on my FreeBSD machine as it's using a cloud based IRC client

yep, bulding kernel on latpop could be challenging, but with regard to configuration you are fine - just build GENERIC

mzar: now to get my wifi card to work >:|

rtprio: you need CURRENT, fwget(8) and it will work

for iwlwifi(4) ?

i suppose current wouldl't hurt; it's a bit awkward being installed on a usb stick but i don't want to blast the os that's there until it know it works

rtprio: Can you use another USB stick for -CURRENT?

Recompiling the kernel on freebsd is easy and imo a right of passage.

Rite, not right.

i don't have any wifi usb sticks, i don't think

Thanks

bummer !

i donno maybe, it's probably only 'b' if i do have one

rtprio: Oh! I thought you meant you were running FBSD from a stick.

Anyone remember launching a FreeBSD install from a single floppy disk, via ppp & ftp?

sure, but you needed a bunch of floppies

So, if you *DO* have another USB stick, maybe just run -CURRENT from that with your wifi USB and see if it works?

Then, you don't need to upgrade your current system. Just run a live one.

yep, please follow ek's guidance and your wifi will work

i'm just updating this fresh stick to 16

it was just a test to see if 'having the firmware made it work' since the installer wasn't going to do that

Fair enough.

I'm curious to hear the results.

well, they're in

could not load binary firmware /boot/firmware/iwl-debug-yoyo.bin either

RFIm is deactivated, reason = 4

bummer

Bah!

reason =4 is like 'fn key to make the wifi work' but that doesn't seem to do anything

I haven't played with wireless in a while.

on my lenovo x1 it just worked

i'm considering swapping the cards

I think I do have a laptop that I can use to test stuff. But, it's not a REALTEK.

That reminds me of real talk by tupac shakur

what is my best support venue for digging into a crazy drive controller problem? IRC is probably not the best...

basically I saw some drive issues (SMART giving some poor numbers on SSDs), so i just assumed replacing drives would fix me right up, but no...

new drives also just timing out then being tossed by the controller (mps). no load really.

same hardware (Dell R720) ran vmware for quite a few years and putting FreeBSD on here (15.1) to test out Sylve has been quite a little ordeal (and I probably spent about $500 on drives I didn't need to, oops!).

I am far from an expert on cross-flashing Dell/LSI RAID cards from IR to IT, but that is my biggest change.

spork_css: are you using mr. sas

I need to double-check cabling again and all, but it's a server with a drive backplane and all that so I doubt I'll find an issue there.

mr. or mrs. SAS? :)

i mean, MRSAS(4)

mps

i didn't have great luck without it

wait is mrsas a driver or utility?

pool: zroot

state: SUSPENDED

(can't look at the moment, oops)

it's a driver

that you put in your loader.conf

i haven't flashed my lsi card either, it's on my list

I think that driver is for something much newer than what I have. mps appears to be the right thing.

way back in 8.4 we had some onboard LSI controllers and I just hated them. They seemed to work well elsewhere, but on FreeBSD kept running into problems not unlike what I'm seeing now, but there it seemed to correlate to load (these were PGSQL servers with a bunch of SSDs).

It would dump a drive, I'd pull it, analyze it elsewhere, bring it back to the colo and then in a few months, different drive, different slot, same thing and it just continued until we brought in new hardware years later.

I'm kind of thinking that since this is just a home server (used for work at times tho) and it's old VMs are not currently essential to anything I'm working on I should really dig in.

But also in the past I've spent days on troubleshooting things, did a big bug writeup, but then couldn't get anyone interested in it.

I found a few extra PCIe SATA cards last night while unpacking, considering putting those in and bypassing the onboard controller to see what happens and kind of write off drive/power/cabling issues.

Then I could move to looking at why mps and these drives aren't happy. Problem moves to all slots (currently I have six SSDs in).

I'm also a little concerned that the cross-flashing tool I used is quite old and maybe my mps firmware is just crap (I did update firmware on two samsung drives that are older).

CrtxReavr: Not that far back for me, but I did build a freebsd router that would ppp dial on demand to connect ot my ISP when something on the network requested internet resources.

I don't want the majority of the world to live dial on demand, but I do want to live in a world where the majority could easily live dial on demand if they wanted or had to.

That ship has probably sailed long ago.

public-owned FTTH and you "dial" on demand to the ISP of your choice (dial = pick a VLAN). :)

i dont miss 3kb/s

*sound of many modems dialing at once*

spork_css: that sounds like a fun utility billing puzzle. how do you decide "link activity" and then pass on fees from the ISP?

how do you make the network detect "oh, this ONT was active for this VLAN between now and now"

Well, I think the owner of the infrastructure would probably mandate a particular ONT and such, so the customer couldn't just "fake" being another user.

I'm going to assume that if things have progressed since the DSL days when I played with those access concentrators, this is probably a total no-brainer on billing/access control.

The politics of it and finding ISPs wanting to compete in that scenario are the more difficult part. :)

I mean, once you have that, it's a puzzle in the sense of, "how do you consider the line active, should data have been sent within the last X minutes?", and "should ISPs be able to charge clock-on and clock-off fees?"

how do you ensure equitable accounting for: the fixed costs of the installed plant, the (minuscule, possibly left unaccounted for convenience's sake) incremental costs of delivering each data unit to/from the ISP, the ISP's opportunity costs in terms of address resource usage during that time, and the ISP's costs to transmit the packet onto the Internet?

having a static IP also throws the whole calculation because the ISP might on occasion need to dial *you*