wouldn't a better abrouch be try to invert the O_CLOEXEC logic in kernel?

no

I'm not touching the kernel

I also can't assume that fdescfs is mounted

there is a way to get the process's whole fd table via sysctl, but that's way more overhead than I want

the problem I see is that you might close an fd which expected to not be closed on exec but ovesee by the user because it's created inside a lib

well that's kind of the point - the child program in this case must not inherit fds that are not explicitly specified by the parent

no library has any business passing down fds that aren't exposed to the main program anyway

the answer in general has been "use the CLOEXEC flag everywhere" but that's messy and error-prone

thats why I proposed to invert the CLOEXEC logic in libc/kernel

that's a non-starter because of contravening all the existing standards

but using your wrapper will do the same

no it won't, because my code will only affect programs that actually use it

(my lower-level improvements to posix_spawn could be implemented as posix_*_np functions in libc, but the higher-level interface is wholly new)

(and also is not in C)

yeah, looking at the call to get the whole fd table, it's pretty costly, though only O(numfds) rather than O(maxfd)

it returns a huge struct for each open fd

it's annoying because the kernel has a perfectly good fdlastfile(fdp) function that's not exposed via any reasonable interface

my alternative plan is just to say "don't do that" and enforce the constraint that the parent can not specify that any files remain open above the child's soft fd limit

.vim

I finally compiled xfce, installed it by first making pkg install from the stuff built by poudriere as opposed to the online binary repos and I ended up installing… all 395 packages thwt were built alongside xfce

I tried to use pkg autoremove but none of them go away

How do I get rid of the packages i dont need?

what command did you use to install them?

it wouldn't surprise me if xfce had that many dependencies, either

I first modified the freebsd,conf to disable freebsd and install from my poudriere folder instead, then pkg install

I remember xfce from pkg pulling less than 395

packages including dependencies though

pkg install with what options?

Just pkg install

not specifying a package name?

Pkg install xfce

ah ok

I should have used like, xfce4-desktop instead

how many packages are listed by pkg query -e '%a==1' '%n-%v' ?

That didn’t trigger anything other than usage help

Like nothing

absolutely nothing useful other than muh help

Oh wait

I did a typo

And now I typed it properly

As for how many packages are listed, I cannot even count

pkg query -e '%a==1' '%n-%v' | wc -l

you have a computer to count for you

Ikr

(that's a lowercase L after the -, not a 1)

Kek I thought it was an |

And I got a 0 from that command

Got disconnected for whatever reason, sorry

you must have had some other typo

What was it again

I can’t see the comma dnd because kiwi irc refreshes afain

pkg query -e '%a==1' '%n-%v' | wc -l

Nope, nothing

In fact I get told that - and l are “no such file or directory”

you mistyped something

G'day all

Nope, nada

I did not mistype anythig

there's no space between - and l

Oh

If so then 401 packages total

I suspect a lot of these packages were already in my system

ok. so most likely installing xfce did install 395 dependencies

Many of these being actually just reinstalls of packages that were already on my system

how many packages have %a==0 ?

I added a new mirror drive to my zfs on root setup, how can I install the bootcode on this device as well ?

nineteen

What does this nineteen to 401 ratio mean

last1: is it GPT partitioned, and does the partitioning match the existing drive? and is this UEFI or legacy boot?

Got disconnected again, my bad

puddinghead: the ratio means nothing, the 19 packages are just the ones you explicitly installed (rather than were installed as dependencies)

I see

it's GPT partitioned, I copied the partitions 100% the same from the original drive

autoremove will remove packages which are both %a==1 AND not needed as a (direct or indirect) dependency of any package with %a==0

last1: UEFI or legacy?

Title: gpart show=> 40 15002931808 diskid/DISK-PHYI2423002X7P6FGN GPT (7. - Pastebin.com

legacy

then you need to run gpart bootcode -b /boot/pmbr -p /boot/gptzfsboot -i 1 diskname on whichever is the new disk

that will populate the freebsd-boot partition (index 1) with /boot/gptzfsboot which is the legacy first-stage ZFS-capable boot program

(gptzfsboot then goes on to load /boot/loader from the appropriate zfs dataset)

alright, let's see how this goes

just of curiosity, can't I do something like dd if=/dev/ada0p1 of=/dev/ada1p1 bs=1M ?

Thanks a lot!

I have a last question though, how easy it should be to dual boot freebsd with Linux? Im considering to also keep them in the same drive since I only have one ssd

well, yes, you could, but that wouldn't cover the /boot/pmbr part

I mean, im planning to use grub on the Linux partition

puddinghead: legacy bios or UEFI?

Uefi

any UEFI boot manager will handle it just fine

Njce

Partitioning while on the live Linux environment should be fine rifgt

freebsd should have installed a copy of loader.efi on the ESP, and running that from within UEFI, whether as the default or via a boot manager will boot freebsd

as long as you don't try and move or overwrite any freebsd partition

Yeah all im planning to do is to resize a partition

currently, freebsd installs loader.efi both as /efi/boot/bootx64.efi and /efi/freebsd/loader.efi

resize a partition not used by freebsd?

A partition used by freebsd

uh

used for what?

Storage

what filesystem?

Zfs

To ext4 on linux

this is a zfs filesystem you're not using and are going to destroy?

no, I am using it rn

because reducing the size of a zfs partition WILL destroy it

I see

So im better off just making a persistent usb then

you may be able to resize it from within zfs, I'd have to check

Although my largest usb drive is 32gb and I plan to use this usb for ai stuff

anyone happens to know the order in which FreeBSD starts the various nfs daemon services ?

aka, rpcbind -> lockd -> nfsd ?

it always works when I reboot the box but if I do it manually I must get the order wrong or something

look at the provide/require lines in the rc.d files

Although tbf

or do rcorder /etc/rc.d/* and pick out the nfs ones

I feel like im tired of configuring freebsd

I guess I’ll prob go back to Linux and testing freebsd on vms again until I feel like I have enough experience and resources (read: a better Ethernet cable since the one I have doesnt exactly work the way I want it to on my intel either net port, as well as a second ssd and gpu for pass through with bhyve) to painlessly use freebsd

what do you mean by pick the nfs ones ?

in any case, I think I got it

it's: rpcbind, statd, lockd, mountd, nfsd

nfsv4 requires at least two more, I think

though they may be optional

never really made the switch to nfsv4

Inb4 I get a spare laptop that can run freebsd

ah, you're back

so what makes you want to give up on Linux ?

Well, I think its just how fragmented it feels like

It feels like patchwork and how the distros just dont fit what my ideal os should be

Im actually just going to start running freebsd on a vm as opposed to bare metal though

I run FreeBSD on a 10 year old Alienware, it's my Python programming pc, works well

I wonder

Althouh tbh last time I was using FreeBSD as a vm I didn’t do much stuff

Heya FreeBSD people!

Hi

hello

I feel like changing the desktop enviroment From xfce to something else on Linux and keeping xfce for the freebsd vm in order to motivate myself to use thwt vm more

But at the same time its also quite intimidating since I like using my xfce config a lot whenever I can

G'day kenrap

o/

I feel victorious after porting a linux application that I was struggling with for a few days

Nice

What did you port over?

BraveheartBSD: github.com/easymodo/qimgv

Title: GitHub - easymodo/qimgv: Image viewer. Fast, easy to use. Optional video support.

RhodiumToad: just updating you because I know this is a riveting story - I can find no reference to the zpool that doesn't exist, not in zdb or anything, but just mounted the ufs boot drive and I did find references in /boot/zfs/zpool.cache

So I could see when booting off those drives that gets looked at, but odd that booting off a cd the loader would want to look at every damn drive on the system for a zpool.cache file - seems a bit aggressive

aha

I wondered if there was a cache, but I looked around on my zfs vm and didn't find one

obviously didn't look hard enough

and I can't find whatever code reads or writes it, either

ah, here it is

Yet after removing it... still the same issue, lol. I am so lost as to where it's finding this. The problem is I currently have time to tinker, and I'm having a hard time not trying the "If I get rid of all references to hdpool, does that solve the problem?" option.

BraveheartBSD: I'm also interested learning about writing or modifying FreeBSD drivers. I'm reading an ebook about it. The C code after the first chapter is intimidating though, however this would be the best way for me to get the most out of FreeBSD.

spork_css: do you still get messages about hdpool, or is it just slow now?

still get the message, which is what's driving me totally nuts

I moved the zpool.cache out of the way, and after booting into a livecd (12.2), still had the hdpool message from the loader, so I mounted my ufs root again and zpool.cache is still gone.

ran "zdb -l" on all the other drives, no mention of "hdpool"

There is another zfs zpool cache in /etc, but that probably references my data pool

on what fbsd version?

13.2

I don't see a cache in /etc

(also 13.2)

I have /etc/zfs - compatabilty.d exports and zpool.cache

I only have compatibility.d and exports

this is in a pretty freshly-installed vm

It could be my 20tb data pool. zfs is on my boot ssd, it has the same cache in /boot.

G'night

Night

A 20TB pool is pretty impressive

It's old ...... an old Dell T610, Perc H700 with 8 x 3gb sas drives, i saved it from going to landfill

Truly a learning experience, lol

Awesome!

20Tb pool impressive ? that's like 1 HDD these days

The age of the thing maybe working against me, ubunutu-server/zfs failed horribly, the pool survived, the data on it didn't

All I've got is /etc/zfs/exports

I think /etc/zfs/zpool.cache is created when the first zpool import <pool_name> is run

I see reference to that location in the zdb manpage (-C flag is the first mention).

doing two zfs send/receive in parallel, this box rocks: >>> Transfer 61% 503MB/s (total 690813MB, 8 minutes left) >>> Transfer 16% 583MB/s (total 2076016MB, 49 minutes left)

almost 1.1GB/sec

i'm writing from a headless weechat install in a jail connected with wireguard on geli+zfs ... not too exciting, but it's my first "on metal" freebsd install, so i'm pretty proud of myself :)

markmcb: nice!

^C

--- google.com ping statistics ---

2 packets transmitted, 2 packets received, 0.0% packet loss

hm, someone was asking here recently about blocking ips for sending SYN with no follow-up?

Hello, do you know how can install 'pkg install misc/compat4 / pkg: No packages available to install matching 'misc/compat4' have been found in the repositories ' on '13.1-RELEASE-p6 FreeBSD 13.1-RELEASE-p6 GENERIC arm64' ?

compat4 ?

there never was a freebsd 4 on arm64, so there is no such package

(and the port origin names for the compat libs are misc/compat4x etc., but that won't help you)

other than a couple of versions that supported sparc64, the compat libs are only available for non-intel architectures from compat12x onward

Drat! broken apachectl now is preventing certbot from renewing my cert. I previously worked around bugs.freebsd.org/bugzilla/show_bug.cgi?id=268318 by using service stop/start, but py39-certbot-apache insists on using "apachectl graceful"

Title: 268318 – www/apache24 with www/mod_php82: opcache + ASLR turned on crashes Apache

Time to get rid of my last apache install and transition that machine to nginx as well.

can I drop PORTREVISION when DISTVERSION is increased?

yes, and you should

thank you!

fireglow: as RhodiumToad mentioned. the only exception is when the "newew" version actually goes backwards. then you need to increase PORTEPOCH and never decrease it.

right. but even then you should drop PORTREVISION.

anyone know offhand if there exists a SIMD-optimized way to convert decimal ipv4 addresses to binary?

greetings everyone

i'm interested in helping with package maintenance

is this a good place to start?

Bah. xrs left. I was going to point them to docs.freebsd.org/en/articles/contributing and specifically docs.freebsd.org/en/articles/contributing/#ports-contributing.

Title: Contributing to FreeBSD | FreeBSD Documentation Portal

zfs list says 824G used by postresdb but `select pg_size_pretty(pg_database_size('mydb') );`  in postgres says 149G. My drive got full and only 1 Gb left. What can i do here?

du -h -depth=1 /mydb/* | sort -rh also says 137G

why zfs says 842.

and my OS gets unusable?

pg_database_size doesn't include WAL

RhodiumToad I wonder why WAL would go that big

How to remove all WAL files for now?

did it? how big is the pg_wal dir?

do not ever touch the WAL files

How can I check?

I see

du -hs pg_wal

PG_15_202209061/pgsql_tmp

if that's not very big, the next question is, do you have any snapshots on this dataset?

what's that supposed to mean?

137G /mydb/ PG_15_202209061/12054723

512B /mydb/PG_15_202209061/pgsql_tmp

no snaps

so this dataset has a tablespace on it, and the main data dir is somewhere else?

yes

but the size zfs shows is on that dataset

du -ks mydb says what?

despite the files on that location are 140GG

checking

143782304

what exactly was the zfs list output? use a paste site

ok

Title: Mozilla Community Pastebin/y861J1kS (Plain Text)

pastebin.mozilla.org/y861J1kS/raw  better format

RhodiumToad confusing. isn't it

zfs compression is enabled?

compression zstd-7

what's the total file size of the files in /mydb/PG_15_202209061/12054723 ?

(not as reported by du, but the sum of the sizes shown by e.g. ls -l)

also, how big is your zroot pool actually?

RhodiumToad pastebin.mozilla.org/KxOLjWYA/slim

Title: Mozilla Community Pastebin/KxOLjWYA (Plain Text)

zroot drive is 2tb in total having other things, os, etc as well

use ls -l | awk '{ t += $5 }; END { print t; }' (note, no -h or similar crap)

doing, its taking time

it shouldn't do

you probably did it wrong

still taking high cpu and time for that command

one sec. pasting

I did this: `ls -l | awk '{ t += $5 }; END { print t; }' /mydb/PG_15_202209061/12054723/`

no

kill that

ok

ls -l /mydb/PG_15_202209061/12054723/ | awk '{ t += $5 }; END { print t; }'

oops

doing

159679648559

zpool list says what?

same. the size avaialble got smaller now. from 1.3T to 850M

checking zpool again

also what's the output of zfs list -o space

RhodiumToad pastebin.mozilla.org/XPhLUQPd/raw

USEDSNAP is 686G for mydb

that's the space consumed by snapshots of mydb

RhodiumToad do I nees crub?

need*

scrub*

zfs list -t all

but I never made a snape since months

just look at zfs list -t all

right. got it

maybe you were using some tool that created snapshots?

will this be accurate commadn to destroy the snap? `zfs destroy zroot/mydb⊙z2`  ?

no idea.. really

didn't used any tools lately

I believe that is correct. but my zfs experience is minimal

RhodiumToad your experience solved the issue. I am getting the space back with time in `zfs list` . thanks

can i use gpart with ufs to extend the size of a virtual machine disk?

i think i tried it before and the vm wasn't bootable

has the underlying physical storage been extended?

in theory you can do it, but it depends on the current layout, where any swap partition is, etc

there's even a builtin script for it that usually works

i did some command with qemu-img to extend the qcow2 file and then i couldnt' boot multi-user

i guess there is a way to do it i just don't know

in what way did it fail?

well you got me there. if you like i can try to recreate it

but it was a boot failure

so presumably it didn't like the fact the disk drive was different now

that in itself shouldn't cause failure, after all that's the normal case when booting of e.g. memstick images

(where the image is usually much smaller than the drive)

hmm. maybe i should use a faster drive for this

is there a good way to benchmark the different encryption options zfs provides? I want this external drive used by an rpi to be encrypted, but I worry about performance, as freebsd doesn't have the aes-adiantum cipher

what model of rpi and what cpu?

rpi4

4gb ram

hrm. dmesg read the new size correctly. that's odd. maybe the problem was i just didn't know how to extend the disk right

cortex z72, running at 1.5G

*A72

on linux I used aes-adiandum with it, as its optimized on run on cpu-only, without aes hardware accelleration

*t

which the pi sadly lacks

does the cpu in fact have the armv8-a encryption instructions?

hm, are you sure it lacks them?

i'm looking into this rn

freebsd-13.2 at least seems to have a module for armv8crypto, which mentions AES instructions

if that module loads, then you should be good

RhodiumToad: i'm not sure but i think it's frozen at devd right after bringing up the network

although cpu % is still high

RhodiumToad: `armv8crypto0: CPU lacks AES instructions`

in dmesg

oh. well that settles that then

i'm on 13.2 release

ye, that's why linux people recommended that google adiantum cipher, they developed that for fs encryption on phone SoCs, and its fast

if you want to test crypto overhead, I'd try creating an md device, putting a geli partition on it, and using dd to read and write it

I'd rather just test it on the actual drive

Does GELI provides all the encryptions schemes that ZFS does?

I have no idea what zfs does

RhodiumToad: actually it does boot surprisingly

so you were definitely right. now i guess would be the question how to extend ufs so it takes the entire new disk size.

gpart list reports the new size. i added +1G to it

service growfs enable; touch /firstboot; shutdown -r now

but the partition or slice or whatever you call it is the same

the growfs script will try and adjust the partitioning to use the new space, and extend the filesystem

it only runs once, if /firstboot is present (that's removed after one boot)

amazing

did it work?

df says it did.

of course i'm not going to actually try writing a huge file because i think i probably shouldn't be using an sd card for this at all.

but yes... i think it worked. you're a genius!

well reality sides with you on this one. chapter 19 of the handbook does indeed mention growfs

anyway thanks again

yw

o/ I cannot seem to catch a break with these network cards of mine. This 13.1 machine with a re0 device doesn't want to route any traffic despite having an active link, and if I switch over to em0 that seems to work fine until some NIC bug causes it to stop sending traffic (says things like sendto: no buffer space)

so, you have the bug on both cards?

costly: heh, hard to tell for sure. The machine has always felt haunted

on reboot it kernel panicked with some unrecoverable EHCI error

rtyler: hmm, sounds a bit more like an issue on the motherboard than on each card?

I cannot remember for sure, but I could have sworn I swapped this motherboard out earlier this year. Waiting for some more errors before I shut it down and take a look

why does it seem chromium is missing from pkg repos to me? I am on latest

i see chromium no longer accepts BSD patches

chromium build often times out

latest for what fbsd version?

13.2

13.x release latest build seems to be running right now, chromium build is 11.5 hours in

so it should be available in the next few days?

the build is about 20% done out of 14399 ports

ungoogled-chromium is availavle but brings in Pulse Audio and sndio as a dependacy... firefox does not support 'sandboxing'

would I be ok to mix my own build of chromium with the rest of my system using precompiled pkg's

sure, if you have the patience

that said, I looked at the failed build log, and it doesn't look like a timeout

unfortunately it's uninformative, it just says "FAILED" for some of the compiles, and stops

the containerisation built around Chromium on openbsd is a great feature

could I run my webbrowser is a Linux bhyve or something?

still learning freebsd

without GPU passthrough you won't get any accelerated rendering

I have firefox open in a bhyve vm at the moment for various reasons, it works ok other than the rendering performance

(I wouldn't try and play video with it for example)

any fix for that?

do you have an extra GPU available? :-)

I don't know the status of GPU passthrough, whether it works properly yet or not

power intel integrated

on wayland too, very unchanted territory

hi

I've started porting components from FreeBSD's source tree to Windows (Cygwin, MSVC, MinGW).

that does not sound like a pleasant project

True dat

I'm almost 2 down so far - github.com/SamuelMarks/freebsd-getopt-portable - github.com/SamuelMarks/freebsd-m4-portable

Title: GitHub - SamuelMarks/freebsd-getopt-portable: FreeBSD's `getopt` and `getopt_long` ported to MSVC

SamuelMarks: what's the motivation / goal?

meena - I'm porting random interesting C projects to Windows and some depend on m4 or getopt or <insert name here>; and the FreeBSD ones are nicely licensed and pretty portable

I may possibly be mad

I just wrote an optimized IP address reader

RhodiumToad: SIMD?

yes

nice

now to find out if it's actually any faster than the obvious method

If it is send link I'll send PR ;)

well, it's not slower.

(I tried the binary to decimal example from the intel optimization manual once. it's slower than a decent non-SIMD approach.)

on the other hand it's less than 10% faster.

mmmm that's within the margin of error I think

oh also why is fetch so weird in FreeBSD, I tried porting it years ago but it failed without error

This attempt: github.com/offscale/libacquire/tree/master/acquire/freebsd_libfetch

Title: libacquire/acquire/freebsd_libfetch at master · offscale/libacquire · GitHub

hm so I've put a new NIC in this machine and configured that statically. It's not able to ping the router at all however which has me a little confused

ifconfig shows that it has a link

checked with and without pf, still am unable to ping the gateway

netmask and broadcast domain all look correct too >_<

definitely ghosts. Was doing a `service netif restart re1` and the machine panicked and rebooted, so that's no good

another reboot and somehow re1 is now properly connected. GHOSTS

rtyler: did the ghosts leave a crashdump?

meena: the crash file in /var/crash just says to install a kernel debugger xD but I did get a picture of the trace

I'm going to bump this host up to 13.2 while it's offline

well upgrade went fine. but I get this panic on reboot xD gist.github.com/rtyler/d64ccb5ecfcd832b08bec23f4723f838

Title: watermelon.reboot.txt · GitHub

when the machine comes back up, it doesn't have network until I `service netif restart re1` and `service routing restart`, which leads me to the ghosts conclusion again

I am pretty sure I should just go buy a different less haunted box and move my disks over