Boom, done. Thanks for the advice

what do you use for a user to run as root? sudo doas or something else?

I've juggled both, which isn't something you're supposed to do as it becomes more difficult to ensure you keep policy between the two synced

well, plain su?

ubuntu complicates it by making sudo the only way to get root, it's technically more secure but just makes your password the root password

It's just a default in Ubuntu. If you set a password for root then "su -" works to jump to root okay no problem.

yeah it's discouraged though

i like the simplicity of simple su

And sshd by default is configured not to allow root logins with "PermitRootLogin no" or "PermitRootLogin prohibit-password" but again can be changed if desired.

ok, i'll just use su.

Plain "su" or plain as in not "su -"? I say always use "su -" in order to load the root environment. Otherwise it all depends upon your user environment, and the OS, as to what environment is set.

yes, `su -l` always.

Whew! Good! :-)

i like the tips i see when i ssh into a user or su -l user

Noting that "su -" and "su -l" are the same. They added -l for those who felt uneasy about the old style "su -".

I don't trust users with root access

they could be engaged in hacking!

You can run "fortune freebsd-tips" at any time and see those tips. Good stuff in there!

same, i don't have sudo installed on my linux box.

But rdr... Aren't *you* also a user? At least on your own box? :-)

yeah but i only have me on my box so i can't abuse anyone's privacy

absolute power corrupts absolutely

Power corrupts! And ABSOLUTE POWER... Is kind'a nice. :-)

just installed freebsd on QEMU on macbook M1 yesterday, it's up 10h. first try. i already liked it. a little bit unconfortable about csh though. but it has default settings of history-search which is nice.

csh is in base and so a default root shell to keep everything booting and able to log in with only components from base.

i am just tired of inconsistence of Linux distro after being a Linux user for 15+ years.

But you are free to install other shells. I always install bash and I use bash as my own personal command line shell. But never for scripting! bash is a slow pig at scripts. #!/bin/sh FTW! Standard is better than better.

A more traditional shell for the Unix user might be ksh. But zsh is also popular.

i'm pretty sure zsh can be shipped in base only by removing one piece of code

which is not crucia

l

But none of those are in the base. So it is not recommended to set any of those as the root shell. Or after a major OS upgrade those ports might not run and you might not be able to log in as root. Ask me how I know! :-)

but it never will be (which is probably a good thing)

rwp, core memory unlocked!

i am gonna keep csh and learn it instead. i am already very familar with bash though. i like what's default on a new system. less third party more lightweight

i use zsh i like it

it's fancy

I locked myself out by setting the root shell to bash and upgrading from 12 to 13. But was able to save it easily with Boot Environments. Booted back to the previous, logged in, fixed the passwd entry for the system, rebooted. All good! Woot!

nice

Then of course upgraded all of the ports and then all of those were okay.

so apparently windows was using my efi partition on my linux drive

so when i wiped that drive for freebsd it wiped all my boot stuff

I know I could have booted rescue media like the install ISO but with Boot Environments there those worked much nicer.

rwp: isn't pkg install good enough?

is multiple efi partitions really discouraged or something?

That's one of the problems with UEFI IMNHO that all of the OSes share one ESP.

rwp: is the boot environment auto added to the boot menu? or do you have to configure something

uuidNuniq, I was using the binary pkgs which are precompiled ports but the shared libraries in base had been changed out from under all of those /usr/local/bin/* executables causing them to fail to run.

Boot Environments are automatically added to the boot menu. You can boot any of them by manually selecting one.

You can see your current boot environments with "bectl list" (and noting that bectl is in base and beadm is the original version in ports, basically the same)

nice. how often does libc breaks /usr/local/bin/* ?

what if you also upgrade third party pkgs along with system upgrade?

It was not unexpected breakage. It was expected changes. But I tripped over my own shoelaces! Major upgrades change out shared libraries. Meaning that all pkgs need to be re-installed at major upgrade points.

During a point release cycle such as 13.0 to 13.1 to 13.2 all of the pkgs are compiled using the oldest supported point release. So simple upgrades are usually okay.

But at major OS releases everything is reset to the new .0 version in base. And all pkgs need to be re-installed.

i am not clear, you mean from like 13 to 14, i have to reinstall pkgs listed by `pkg info`?

If 3rd party compiled executables, or your own locally compiled executables, depend upon a shared library in base that has been upgraded then those will also need to be recompiled.

uuidNuniq, Yes. Use "pkg upgrade -f" which will force re-install all automatically. Very simple.

ah, thanks for the tip

If, and I say if, pkg itself becomes broken by the shared libraries being swapped out from under it then there is a special pkg-static shipped which is statically linked which can be used to bootstrap pkg itself.

"pkg-static upgrade -f pkg" would force bootstrap itself using static libraries. That's if it is needed. It's not usually.

i'll install that in case of things went mad.

You will install pkg? Doesn't everyone? :-)

ohh it is already installed

i mean pkg-static. i thought it was a separate package

There is a /usr/sbin/pkg which is nothing but a bootstrap tool to bootstrap install /usr/local/sbin/pkg in order to get things going.

And pkg-static is part of pkg.

rdr, Regarding UEFI one of the things I miss about Legacy BIOS booting is that everything about booting was on disk. One could just pick up a disk and move it and boot.

i have /usr/local/sbin/pkg-static.pkgsave as well

But with UEFI half is on disk in the ESP partition and half is in CMOS RAM with the efi vars. Those efi vars need to be set with efibootmgr in order to boot.

rwp: with UEFI you can do --removable?

So picking up a disk and moving it and trying to boot now with UEFI no longer Just Works but now on my systems at least I have to boot rescue media and then run efibootmgr to set the vars to enable the disk to boot.

there is a standard place that UEFI firmware searches for . which is ESP/EFI/BOOT/BOOTX64.efi

I have heard about people who set their disks to be removable media. I have not done that myself yet. Sounds attractive.

i always do --removable e.g. with grub-install

but it will replace any BOOTX64.efi other bootloader created.

Is that the "fallback" boot? I have one system that refuses to boot anything but the fallback and have installed rEFIng there in order to allow it to boot sanely.

but why do we need multi bootloader? use a good one that boots all OSes is fine.

But mostly I never dual boot anymore. It's always booted to the wrong system. So now I just boot one main OS and then run the other in a virtual machine if needed.

uuidNuniq, The reason is BUGGY UEFI firmware. I have an HP workstation that clears the efi vars at boot time. So I can set things once and boot but it fails on the second boot.

rwp: ESP/EFI/BOOT/BOOTX64.efi works for buggy and sane firmware.

any firmware will search for that...

I have exactly this problem described here at problem #2. wiki.debian.org/GrubEFIReinstall#Pr…boot_entries_disappear_after_reboot

Title: GrubEFIReinstall - Debian Wiki

i think you can just cp your whatever .efi to that location, it will boot.

/dev/da1p2 on /a (ufs, local, soft-updates, journaled soft-updates) is there any danger of soft updates journalled?

my current machien has more ram 8g

should I move to zfs?

rwp: for example. on my vm freebsd: md5sum /boot/efi/efi/freebsd/loader.efi /boot/efi/efi/boot/bootaa64.efi

they are identical.

concrete_houses, I am not knowledgeable about UFS, sorry, so cannot advise. But of course I do advise *everyone* to use zfs because it is quite nice.

it's in a macbook m1 QEMU

anyway the standard search path for UEFI is efi/boot/bootaa64.efi or efi/boot/bootx64.efi

uuidNuniq, I think you are using the fallback system, which is perfectly fine when booting a single OS. But I think systems that have those problems cannot multi-boot different operating systems. Not easily. That's the reason for refing.

this way, you don't need to add boot entries to the UEFI firmware. just select to boot from the disk ... it will find bootx64.efi or bootaa64.efi

rwp: for multiboot, try to boot other OSes from THAT bootloader (which efi/boot/bootx64.efi finds)...

if you're doing grub, then use grub to boot everything. if you are doing refind... same

I wish there was a way to specify a list of "fallback" shells for any user.

kenrap: you mean /etc/shells?

That's a

make a little script that tries to load bash and falls back on a default shell if it exits incorrectly

list of acceptable shells for any user

kenrap: just put a `if command -v bash ; then exec bash ; elif .... `

but how could it trigger? i mean, if the shell specified in /etc/password for the user does not exists... the user basically can't login

that's what I'm trying to ponder about

so, what do you mean fallback

the code is easy to write, just a bunch of if elif ... , the trigger is the hard part

but sh is always there, right?

you could set your default shell to the script

so, i would set the user to use /bin/sh in /etc/password. and in user's $HOME/.cshrc, write that if elif

using exec

Let's say since I use zsh for root and for whatever reason I uninstalled it from local and tried to login as root again. So in this list of shells assoociated to root, it will then fallback to prior shell before zsh and load that for root without not being able to log in again. If that shell doesn't exist, then fallback on the next.

kenrap: set csh for root, and in ~/.cshrc : if ( $?zsh ) then exec zsh endif

if ( $?zsh ) then exec zsh ; else if ( $?bash ) then exec bash ; endif

I'll play around this idea and thanks

ohh don't use $?, it test a varable set or not set

i am new to csh, but i think there is like a command -v or something

maybe if ( which zsh )

if ( { which zsh } ) exec zsh -l

lol, all my bash/posix sh experience are obsolete in csh...

Title: linux - How to write If-else statement in one line in csh? - Stack Overflow

looks like you can't write oneliner with csh...

uuidNuniq: You don't need to use csh for scripting ever.,

kenrap: For fallback, for root anyway which should be the only account for which this matters, it's tradition to set up toor.

also >/dev/null does not work.

mason: i think i may need to install bash and use bash instead

uuidNuniq: Or just write scripts against /bin/sh.

mason: oh, very interesting, good to know

But yeah, I use bash.

I set root to bash, and toor to /bin/sh

what is toor?

Title: toor (Unix) - Wikipedia

uuidNuniq: reverse of root - typically a back-up account in case root has issues

It feels good when I discover more "hidden gems" like this in FreeBSD

Enjoy it while it lasts. Folks want to get rid of it.

...ah, I see

You can always add it in again, but that's not entirely the same.

i am on bash now. all my bash config are ported there.

is there a replacement of the command `tac` on freebsd?

I can imagine the OpenBSD devs ripping toor out of their OS ages ago.

i use tac and awk to make history file uniq

You might also like the ksh variants. I only use bash over them because of how easily it integrates emacs mode keystrokes inside vi insert mode.

I think mksh is pretty close to bash in terms of features.

or oksh, but neither does emacs keys inside vi insert mode

Gotcha

at least last time I looked - if this has changed I'll hop right over

toor is already with /bin/sh

kenrap: thanks, I got that from earlier chat. Do you have a Reddit account? If not, can you post the hint somewhere as a point of reference?

Apparently mksh has an "Emacs mode" and the devs have been steadily implementing more features for it over many releases

so, if csh is not suitable for scripting, why does freebsd have csh around

uuidNuniq: yewtu.be/watch?v=gRdfX7ut8gw

Title: Fiddler on the roof - Tradition ( with subtitles ) - Invidious

meena: I guess that online manual pages for CURRENT will be refreshed when we reach pre-release (pre-alpha) or alpha

I guess I will to create a reddit account, I had a few in the past but deleted them eventually.

mason: the explains

*that

pretty much :)

i got to set a password for toor

that and the shell and you should be good

What's wrong with my use of fwget(8)? <pastebin.com/raw/tzyRUwjL>

Maybe I don't have the subsystem. It's a circa 2012 notebook.

grahamperrin: Also, try -nv in case it's doing funny argument parsing.

tail -r can be tac replacement. good

if [ $(uname) = FreeBSD ]; then _rev="tail -r" ; else _rev=tac ; fi

is this ungoogled chrome slower?

concrete_houses: chromium?

yes

yeah update removed chrome and I had to install ungoogled-chromium

<reviews.freebsd.org/D40565>

Title: ⚙ D40565 Remove toor backup account

how can i run man against /dev/stdin

curl some_manual | man /dev/stdin

does not work. however, curl -O .. and man ./some_manul works

grahamperrin: I left my comment in the amdgpu post

mason: DAI DAI DAI DAI DUM

Title: k3nrap comments on AMDGPU fails to boot

grahamperrin: I don't know why it's removed either. When I logged out I noticed that but when I log in my comment remains.

I assumed it was some automated subreddit thing you configured as admin to prevent spammers

Woah, I was somehow given a free Reddit Premium access and an award for my comment. I feel all warm and fuzzy all of sudden

Not configured by me, but maybe it's automated when a newcomer's first post includes a link.

Gotcha

This automation occurs silently, unfortunately (for me as a mod)

grahamperrin: thank you for the reward btw, I just found out by looking through my email.

is the big reddit fight still going on?

The whole blackout thing?

looks like it is

Honestly, I find that type of protest both a waste of time and hurtful towards users of those blacked out subreddits that need to reference helpful info for solving their problems

It's pretty much a "cut off one's nose to spite one's face"

nah. it's a valid response to corporate bullshittery

I wholeheartly disagree. Reddit can do whatever they want. If people don't like what they are doing, they should explore alternatives.

whatever, this channel probably isn't the place for that

Understood. Sorry for the offtopic-ness.

Speaking of off-topic, this month's FreeBSD Journal is pretty amazing, with a ton of history from a bunch of FreeBSD old-timers (freebsdfoundation.org/wp-content/up…325213a554aba948ae00a2e4bba06.2.pdf)

hi, I am struggling to mount an ext4 usb flash partiton

What I am supposed to do?

Depends on what errors you get; how you are trying to mount the partition; etc? Probably need "ext4" support from some package.

best bet is the fuse extfs package

fusefs-ext2 is the name, but it supports ext2,3,4

you can try mounting it readonly using the base system ext2fs, but that's not guaranteed to work

base system ext2fs doesn't support journalling, I think, which is likely to be a problem

what am i missing in this bectl call:

meena@vault13 ~> sudo -H bectl jail -o host=inherit -o ip4=inherit -o ip6=inherit -o allow.chflags=1 -o allow.mount=1 -o allow.mount.devfs=1 -o devfs_ruleset=4 PkgBase

# pkg update -r FreeBSD-base

pkg: Cannot open dev/null

forgot to mount /dev in the jail?

RhodiumToad: i thought devfs_ruleset=4 would do that?

nope, mount.devfs=1 does that…

Starting jails: cannot start jail "builder":

jail: builder: unknown parameter: vnet

wat

this is 13-STABLE, btw

this is coming directly from the jail command.

for anyone curious, here's my /etc/jail.builder.conf: gist.github.com/igalic/4c1d7417a83dfa796b7e105b006c4347

Title: gist:4c1d7417a83dfa796b7e105b006c4347 · GitHub

oh weird…

i don't see options VIMAGE in kern.conftxt …

good grief, how have i not noticed that GENERIC-MMCCAM doesn't include options VIMAGE yet?

good day

I yesterday upgrade a box with opnsense, now running 13.1-RELEASE-p7

Since then some pf rules seem not to match anymore that used to match

I have a rule "pass in quick inet from any to 10.200.32.0/24 flags S/SA keep state" that does not match a packet going towards 10.200.32.239, the packet is blocked by the default "block drop in log inet all" rule

has anyone else seeing something similar happening (recently)?

x39

telmich: do you have logging enabled?

meena: yes

it says it's caught by the default block rule

telmich: so something has been restructured that's changed priorities.

meena: exactly, but it doesn't make sense as there is no quick in the default block

telmich: what does the config look like when you ask pfctl to print it?

meena: it's quite a long list as it is generated by opnsense, but essentially:

- default blocks at the beginning

- lots of exceptions/passes

- pass quick inet from any to 10.200.32.0/24 flags S/SA keep state at the bottom

what I can see in the log is that the "block drop in log inet all" is matched for packets that are destined within 10.200.32.0/24

even if I widen the rule to "pass in log quick inet from any to 10.0.0.0/8 flags S/SA keep state", in case I am stupid and mismatch the network, it still does not match

found it

I removed the state S/SA requirement, even though state was SA

And now it passes through

"pass in log quick inet from any to 10.0.0.0/8 no state"

it's strange, as even with state tracking this should work

what's S/SA vs SA?

telmich: ^

meena: it's referring to syn/syn ack bit set

uuidNuniq: Sorry I dropped off last night. My ISP went away. I could get packets several hops in but then some router other other in their infrastructure was dead.

mason: that was very sad

i watched the whole movie... because of that "tradition"

uuidNuniq: Oh, good. Not time wasted. It's one of my favourites.

mount /dev/vtbd1p1 /mnt/ : mount: /dev/vtbd1p1: Read-only file system

mount -o autoro /dev/vtbd1p1 /mnt/ : mount: /dev/vtbd1p1: No such file or directory

i am not so confortable with these commands. my first try was always error

if this is in Linux, it will just mount, no complains

i miss commands like : lsblk, sfdisk , ss ...

uuidNuniq: Yeah, there are some conveniences missing. The one that I have to remember is to specify the -t type

uuidNuniq: Try with an explicit -t and see if maybe it's happier.

mount_cd9660 /dev/iso9660/13_2_RELEASE_AARCH64_BO /mnt/

this works

i don't know what this is : 0x0.st/H13s.txt

it looks like the cdrom

from gpart show: 0x0.st/H13z.txt

this guy created a lsblk.sh : github.com/vermaden/scripts/blob/master/lsblk.sh

Title: scripts/lsblk.sh at master · vermaden/scripts · GitHub

FreeBSD is not Linux.

du -a -c -h -d0 ./

The partition you're showing is what's known as an EFI System Partition; it's a FAT12/FAT16/FAT32 used by UEFI implementations to boot.

-a and -d can't be used together?

Seems like it.

uuidNuniq: Interesting. I was thinking of something a bit simpler. But look at "geom part list".

uuidNuniq: Interesting thing, I noticed that even for things with GPT labels, not everything shows up in 'glabel list' or 'geom label list'

debdrup: but which efi? the ISO?

glabel is for GEOM labels

GEOM labels aren't partitions

mason: geom part list outputs the same thing as gpart list

uuidNuniq: I mostly want to do a quick look-up tool where you say 'glookup foo' and it'll spit back just the entities that have foo as provider, consumer, or label.

uuidNuniq: yes

debdrup: The interesting thing is I never set explicit geom labels, ever. I set GPT labels. And some of my GPT labels show up in 'geom label list' and some do not.

debdrup: They all show up in 'geom part list'.

uuidNuniq: the ISO uses a hybrid CD-9660/USB boot schema that makes it act as either a CD/DVD or a disk (typically, BIOS implementations expect you to define whether a USB drive is either a CD/DVD or a harddisk, but with the hybriad approach, that isn't necessary).

mason: which 'all'?

debdrup: Example: bpa.st/RIU3G So, this was done the same way I do all my GPT labels, but in this case, da5/da5p1 is missing from 'glabel list'

Title: View paste RIU3G

debdrup: i see 3 "disks", so vtbd1 is basically the same as iso9660/13_2_RELEASE_AARCH64_BO in this case

mason: I can't keep track of things when you elid everything

uuidNuniq: yes, the iso images will show up as both the actual partition and the hybrid image; what exactly is the problem you're facing?

debdrup: That's the relevant bit from 'geom part list' - and that label doesn't appear anywhere in 'geom label list'.

mason: only things with a GEOM label can appear in `geom label list`, whereas things with a GPT label can appear in `geom part list`

debdrup: Right! So why do most of my things evidently have a GEOM label? That's puzzling since I never explicitly assign them.

If they're inferred, why aren't they ALL inferred?

mason: most of your things don't appear to have GEOM labels if they're not appearing in `geom label list`

finally : mount_msdosfs -o ro /dev/vtbd1p1 /mnt

debdrup: bpa.st/AHCUW

Title: View paste AHCUW

I've never once set an explicit geom label.

so the issue is , the mount command on freebsd lack of auto detect filesystem type

uuidNuniq: Yes. You can also say mount -t foo if you prefer that form.,

mason: but i don't know what the filesystem is called on freebsd. i tried fat fat32 vfat efi ...

but it is actually msdosfs...

uuidNuniq: msdosfs

yes

gparted list shows it's type is efi

uuidNuniq: mount<tabtab> will show some common ones

ok. is there a command to know the filesystem type before mount? or else i don't know what to feed to -t

uuidNuniq: I'm not sure where there's an authorative (or queryable) list of mount types currently supported.

uuidNuniq: file(1) perhaps?

uuidNuniq: In practise I don't see enough different types for it to be an issue.

Mm, not file(1). Unsure.

file says : /dev/vtbd1p1: character special (0/72)

Yeah, sorry, that was my just being hopeful.

uuidNuniq: My advice, don't let this kind of thing throw you off. It's minor in the grand scheme of things.

I'm too prone to let minor annoyances redirect me, and it's not as useful as it might be.

mason: thanks, i agree.

mason: wrong property

label != name

file -sL

lesta: cool, thanks

welcome

mason: I dunno what you've done to your install, I only have a single label and that's from the swap zvol.

file -sL is a very good command.

debdrup: Beats me. This is just my experience of what the system is giving me.

Anyway, the upshot is that everything shows up in geom part list, and I'm going to whip up a little tool to ease mapping.

i am goona read through the handbook. looks like lots of commands and things are totally different in here

uuidNuniq: see what I said earlier. ;)

uuidNuniq: Random note, if you've grown accustomed to 'watch', realize that it's something else in FreeBSD, and what you'll want is the 'cmdwatch' port.

there is a util-linux package, i did pkg install. and then it has wipefs blkid findfs... but they all print empty stuff...

debdrup: yeah, i thought based on my Linux (cli only) and macOS (mostly GUI) it's gonna be easy for me to play with Freebsd but i was wrong

uuidNuniq: It will be. Just takes some time to learn the dialect.

mason: watch is another command here : snoop on another tty line

uuidNuniq: I had the same jump when I moved from NetBSD to Debian years ago, and a similar jump picking up what's different about FreeBSD some years after that.

uuidNuniq: it's pretty much agreed-upon that the more flavours of Unix-likes you're intimately familiar with, the better the sysadmin you are.

mason: i was hopping distro a lot, and settled on Debian. all my servers are debian right now. but i found that debian is really complicated inside. so many wrappers with complex code, stuff like debconf, dpkg-reconfigure, they are perl + sh combined... and it's too complex when i read the code... so i decided to give FreeBSD a try

There's also something called the "UNIX Rosetta Stone" which you might look for, to help you move from one Unix-like to another.

uuidNuniq: this is exactly what led me to FreeBSD

It hasn't been updated since The Event, so there are things that might require you to be familiar with how things used to be - but it's still quite handy for everything else.

Debian is great for stability, but don't ever look under the hood, lol

uuidNuniq: Yeah. There's a ton of infrastructure in Debian. FreeBSD is certainly more straightforward. That said, I run both as they're both worth using and strong in different ways. Shifting more towards FreeBSD of late.

bbiab

Debian is ~70 million lines of code (of which ~24 million is the Linux kernel itself); FreeBSD comes in at a svelte 14 million lines of code.

I find FreeBSD strikes a nice balance of up-to-date ports and a stable base. Debian gets a bit stale with its long release cycle.

For comparison, Windows and macOS are estimated to be north of 100 million.

I've attempted several times, but nothing ever beats macOS on the desktop for me. FreeBSD for servers.

debdrup: i liked the stability of Debian too and mutil arch support, but it is now mixed with the Debian way and the systemd way... network configuration for example ... and cron vs timer, and they have init scripts along with systemd services ... however, i wrote a sh script to bootstrap debian (400 lines of posix sh code ), it takes 2mins to install, and everything is setup inside my script.

uuidNuniq: have you ever tried Void? It's like BSD and Linux had a child. :)

markmcb: i use only macOS for desktop.

I like the fact that FreeBSD even has a seperation of base and local. Linux doesn't have. If I want to "clean out" my package installations without breaking the system, then not a problem.

markmcb: yes, i also wrote a void bootstrap script using xbps-static. it is simpler but what i worried about is that they don't have enough devs, some packages are not up to date due to that.

uuidNuniq: i agree. i was a long time user, but lxc (fairly critical IMHO) went like 4 months with a blocking bug on btrfs due to lack of maintainer. i had to ditch it.

and Void devs don't (have time to) deal with bugs and PR/Mrs

MRs

and others i liked once are Alpine (lack of devs), Arch (pretty good, but only x86_64) Gentoo (good, but not for lowend hardware)

in the realm of *nix desktops, i always found Arch to be the best experience. up-to-date and no junk. plus wayland+gnome is so good on a hidpi laptop. i know X gets a lot of love here, but i feel it's daylight and dark from an experience standpoin.

does most Software even work on wayland without Xorg compat?

I've never had issues

i intentionally don't install xwayland and everything i use works great

wayland is very good and adoption is high

markmcb: yes, arch is very good for daily use except critital servers. i was a long time Arch user.

uuidNuniq: funny you say that. after void failed me, i attempted arch as my server OS because I like it on laptops ... bad idea. lol. so i found myself choosing between Debian and FreeBSD. I'm still in the honeymoon period, but I think I'm in love with FreeBSD :)

uuidNuniq: FWIW, nothing forces you to use systemd with Debian: github.com/ChibaPet/install-debian-zfs-sysvinit

Title: GitHub - ChibaPet/install-debian-zfs-sysvinit: Install Debian on ZFS, with sysvinit, optionally encrypted.

Arch is ok for not important servers, i used arch as a VPS OS for some years, never breaks, it is a pity they don't include aarch64 (it is another project , only one maintainer.). i fall in love with FreeBSD when the installation menu shows up, and the whole installation is fantastic simple and clean.

FWIW, as we drift afield, #freebsd-social exists. I'm there and happy to carry on there.

mason: i use btrfs on Linux because it is supported natively. i will never go back to sysvinit.

uuidNuniq: I used Debian for all of a week back in 2000, then got frustrated with the lack of documentation. Since then I've just used FreeBSD.

debdrup: speaking of documentations... yes, that's a sad story, debian's apt dpkg docs are confusing. Arch wiki is the best.

i think this is a bug, the du command , -a can't be used with -d

I can't spot why it shouldn't work, so it probably is a bug, yeah. bugs.freebsd.org is the place where reports go :)

Title: FreeBSD Bugzilla Main Page

Are there any PCIe-SSD devices that are well-supported in FreeBSD? I need ZIL devices for a 4-6-way NVMe pool, but it's silly if they're slower than the pool devices. :)

Ltning: PCIe SSDs can either work via da(4) (as a SATA device, which I assume you don't want), as an NVMe disk via nvd(4) or nda(4), with the latter being preferred because it supports the CAM I/O scheduler

Ltning: also, you might wanna make note of openzfs/zfs #14520

Title: Implement shared log pool for ZFS by pcd1193182 · Pull Request #14520 · openzfs/zfs · GitHub

Hasn't nvd(4) been dropped in favor of nda(4) only in 14-CURRENT though?

I'm not sure I'm comfortable recommending any specific device, but a log device generally doesn't need to be big, it just needs to have MASSIVE write tolerances - so datacenter NVMe SSDs are generally preferred, since they can always be overprovisioned.

kenrap: well, that switch is pretty recent, but yes

Yea we're using Intel Optane for ZIL on our current database servers

I suspected as much.

Those have insane IOPS and high write tolerances. Also they fail in a predictable fashion.

ublock origin not available for ungoogled-chromium :(

Given Intel just killed Optane, it seemed like a bit too much of a coincidence.

But those are not faster (by any meaningful degree) than the NVMe SSDs we're looking at, and yes Intel just killed off their most interesting SSD product.

Ltning: for what it's worth, the team at Intel went to work with SK Hynix; this is what caused me to buy a Solidigm P44 for my laptop.

concrete_houses: you need to install "chromium web store" extension first

err, new workstation*

Our ZIL is like 4GB mirrored per pool (and yes I'm watching that PR closely).

kenra: from the extension or using pkg

the former

But my understanding is it's not much point having a ZIL that is slower than the data disks - unless you have very high read load

concrete_houses: go to its github repo and follow the instructions in the README.md

Ltning: it's not a ZIL, though. It's a Separate Intent Log

SLOG devices. Sorry.

the ZIL is still on the actual pool devices, the log device just exists to speed up synchronous writes.

If you have an NVMe pool, having a log device won't help you.

That depends on the size (in device count) and usage of the pool.

Do you have actual numbers showing that it makes a difference?

We're maxing out 14 SATA SSDs now, so even when switching to nvme we'll have pretty high per-device load

SATA SSDs are still painfully slow.

Sure, but we go from 14 to 4 devices in the pool.

And no, I don't have numbers because this isn't tested yet. I just want to know what kind of devices I might want to consider IFF it turns out to be useful/needed.

Not only is the interface limited to ~550MBps due to having to send 10 bits for every 8 bits transferred over the SATA connection, but SATA NCQ is limited to 1 queue of only 32 items. NVMe has 64k queues of 64k items.

If nothing else they reduce the write counts on the main SSDs

The point is, you don't have an NVME pool.

So yes, you still get the benefit from a couple of NVMe log devices.

You're forgetting the batched writes. We have constant writes to the SLOG devices, and flushing every 10s. Since we run 128k block sizes (for compression), but mysql I/O is in 16kb blocks, the SLOG devices save us from a lot of the write amplification.

I don't know if that is going to be an issue with new NVMe SSDs.

Whenever we need to turn off the SLOG devices now, latency goes up, busy rate on the main SSDs goes way up, and bytes written (in the smart data) accelerates a lot.

(Happened a couple years ago when our first-gen optanes suddenly stopped working - they both failed a couple hours apart, after ~2 years :D )

firefox here I come

memory leaks and all

no firefox is so sluggish agggkgkgk

github crap ok

🤷

I beleive sql db are sscam

ever read henry baker on files being faster?

werc.cat-v.org + varnish+ha-proxy

brb, going to do the CURRENT upgrade-rebooting dance.

concrete_houses: use Safari

the best browser

LOL

I wish chuck moore would do a web browser in FORTH :) might work in 1/10,000th the code

I am using ungoogled-chromeium

ignition source located !!! now slapping with a large amount of trout juice

still feels way more zippy that firefox

CmdLnKid, "trout juice" lol

is that a zoomer diss?

IM too old to understand

maybe a military phrase?

What is this ungoogled thing?

ltning, plenty of things im sure that don't involve the goog

concrete_houses mentioned "ungoogled-chromium"

Ltning, I take that to mean removal of various tracking, ad showing, & privacy busting code removal

Whoa, it's actually in ports. *jexec poudriere.child*

Ltning, If that is actually so, do not know the extent

latest pkg upgrade removed charome

parv: which part?

and you have to use ungoogled-chromium

its probably for my own good but now no ublock origin and screenshot plugin until i goto github on a safari to install something

Ltning, What do you mean?

parv: "..do not know the extent"

Ltning, Do not know the extent of (as wrote earlier) of "removal[/proper replacement] of various tracking, ad showing, & privacy busting code removal"

Ah I see. Sorry, I'm dense.

From horse's mouth: github.com/ungoogled-software/ungoogled-chromium (see the "Objectives")

Title: GitHub - ungoogled-software/ungoogled-chromium: Google Chromium, sans integration with Google

Yep, I am not only dense, but also lazy. I did locate it. Now figuring out why I broke my poudriere.

Wonder if LTO is safe .. :D

Once again to the rescue: `top -m io`

kenrap: are you thinking of these commits? <freebsd/freebsd-src bdc81ee>,

Title: nvme: Switch to nda by default · freebsd/freebsd-src@bdc81ee · GitHub

<freebsd/freebsd-src 6a8d407>

Title: nvd/nda: Add release notes info. · freebsd/freebsd-src@6a8d407 · GitHub

grahamperrin: yep

Ltning: re LTO: depends on the project. sometimes, the speedup isn't worth it. MongDB took 9 hours to link, so LTO was disabled

Ltning: reference bugs.freebsd.org/bugzilla/show_bug.cgi?id=271919

Title: 271919 – databases/mongodb50: ld.lld takes over 10h to complete on arm64 on 32-core machine

so i have a single device pool with a failed drive that I don't care about the data. i can't remove the device from the pool because there are no other devices in the pool. is my only option is to remove the pool?

kenrap: I'm curious, do you know about src/tools/build/beinstall.sh ?

CCFL_Man, zpool destory?

You'll also want zpool-labelclear(8)

zpool destroy

debdrup: I don't unfortunately.

kenrap: if you use zfs, it can create boot environments and install the new base system into it using chroot, which you can then boot to temporarily using bectl (it does this by default)

Oh, awesome!

if it fails to boot, you can use the loader to switch back to the old boot environment

it makes running -CURRENT exceptionally easy :3

Sweet! \o/

also this: wiki.freebsd.org/MetaMode

Title: MetaMode - FreeBSD Wiki

is this channel for all arch's or is there a dedicated one for arm64?

im trying to find out if there are any semi-recent arm64 laptops that are known to mostly work with freebsd but haven't had much luck

TIL FreeBSD has the new procmail: freshports.org/mail/procmail-bgrb

Title: FreshPorts -- mail/procmail-bgrb: BuGlessRB fork of the procmail local mail delivery agent

debdrup: I'm all configured with metamode. Thanks again for sharing about those two methods.

I was planning on learning to use bectl with CURRENT upgrades. So coincidentally, beinstall.sh conveniently fills in that idea for me.

derzahl: #freebsd-mips on EFnet is probably a better place to ask

#bsdmips

...though that's developers' channel and the question is rather a user level thing

#freebsd-xorg on EFnet might have opinions, and that's end-user friendly.

ok thanks

VVD: thanks

debdrup: why labelclear?

prevents data from being restored?

and also, why would I want "IT mode" firmware on an HBA?

CCFL_Man: IT mode is the "let the system manage it" mode, as compared with hardware RAID.

CCFL_Man: You generally want to let the system manage the attached devices directly.

mason: understood. otherwise, the HBA would?

CCFL_Man: That's generally the idea, yeah - if not in IT mode, the computer can't actually see the underlying drives as they are.

I could be confused, but this is how I understand it.

mason: understood. that makes sense. i'll have to flash my HBA with IT mode firmware

Anyone recognize this number?: 1688768316000000000

CrtxReavr: DDG thinks it's a FedEx tracking number.

DDG?

echo $(( $(date +%s) * 1000000000 ))

It's nanoseconds past the epoch.

DuckDuckGo

I want to like DuckDuckGo, I really do. . .

I just don't find their search results very good.

re: DDG - when I was searching for some stuff while reading about Unix/FreeBSD history, I realized Google is more awful than I realized. I hit some of the same stuff on Bing (which apparently is where DDG pulls from) and had way more info.

DDG's main annoyance for me is that dumb infinite scroll crap - I really, really prefer pagination.

DDG had an option for disabling infinite scroll, or had one last I checked.