imm_: pretty sure not possible

i set "gdm_enable=yes" in rc.conf file, after reboot or startup , it directly go into default user, no chance to select root user, how to select root user in system startup

built first custom kernel.. feels good

owo

uwu

after installing gnome desktop , the system haven't sound , paste.centos.org/view/6eeed335

did you test sound before? does cat /dev/random > /dev/dsp give you sound (random noise)?

nimaje: it's ok, that pc didn't connect soundbar

I've got ipfw redirecting inbound traffic to diffferent jails using redirect_port. Things were working fine until a few days ago when the redirect stopped happening on aliased IPs. I don't knnow what I might have changed.:[ Can someone help? Does anyone still remember ipfw? termbin.com/arks

ivy et al: A couple of days ago we discussed installing FreeBSD on ZFS, dual-booting, with existing ESP. I tried your suggestions - copied loader.efi to EFS, set it up with efibootmgr, set the "decrypt me" boot-time flag for my GELI partition holding the ZFS pool, then got stuck. loader.efi wants to load /boot/lua/something and /boot/kernel/kernel, both of which are in ZFS, not EFI (I was able to copy some

stuff to EFI as a temp hack, and got kernel to boot, but not mount /). How do I tell loader.efi to mount ZFS at /? loader.efi(8) says rootdev=disk0s1a should be inlodader.eli. How do I specify ZFS there? I tried zfs:zroot/ROOT/default

oops, scratch that. wrong button. I'm happy to report that I got it finally booted.

Plenty of footguns around, with misleading or confusing documentation. What's the procedure to suggest improvements to docs? For example, the Handbook speaks about keyfiles in GELI, tho it's actually entirely easy to just _omit_ the keyfile, and it'll prompt for password.

cizra: You could submit a PR for a change. Or you could detail what you found here or on a mailing list (we still use those) and hope an interested party does it for you.

But if you want someone else to do it, follow up, don't trust a "yes, I'll do that" until you see the PR. There are a lot of square tuits around here. :]

Haven't found the source code for the handbook, yet, but I suppose I'm not in a hurry anymore. First I need to figure out how to set up the rest of my system :) Reading the handbook now.

Hm, permissions of /tmp reset to 0644 on every reboot. How do I make them 1777, permanently? I'm mounting tmpfs over ZFS, using fstab (no fstab for ZFS, tho)

the handbooks should be in cgit.freebsd.org/doc

cizra: add mode=1777 in your line in fstab

sweet, the long cables for my terminal arrived. 20 meters at 19200 haha

i think this is Cat3 cable even

is there a way to make USB-serial device names survive reboots?

i tried some udev rules but i'm not sure about the syntax. i wasn't able to match by vendor or by serial number

hjf, did you see this post? unix.stackexchange.com/questions/19…bsd-permanent-nodes-for-usb-devices

CrtxReavr: thanks. my google fu is weak today

3rd hit.

this is very strange. this FTDI USB adapter that was giving me trouble, kept disconnecting with USB_ERR_TIMEOUT is working just fine now that i'm using that very long serial cable

i wonder if it had anything to do with voltage drop or something

maybe the +/-12V from the terminal was too hot for the USB adapter? and it's dropping enough over 20M of wire that it's no longer glitching

Help needed again! I realized that following wiki.freebsd.org/RootOnZFS/GPTZFSBoot I neglected to set sector size for GELI and ashift for ZFS, so I recreated my install. It went a lot easier, but now ZFS datasets (/usr/, etc) aren't mounting automatically on boot, except / itself! I'm absolutely certain I didn't have them in /etc/fstab previously, how'd they get mounted then? And what am I missing

now?

Perhaps missing the -o mountpoint=... options when the datasets were created? Or the canmount=off is mistakenly set? Otherwise they would be mounted.

I could triple-check these with `zfs set`...

Here is an example of an install I did recently using the freebsd installer and this is how it created the zfs datasets from "zpool history": paste.debian.net/plain/1331797

iirc there is also an automount property you could have deactivated by accident

Thanks, I'll cross-check against that paste, and double-check the properties.

For exploring this type of problem I like this set of zfs list options: zfs list -o name,compression,exec,setuid,atime,canmount,mounted,mountpoint

Also note that the /usr dataset itself is not mounted. It's there only as a template for datasets created under it such as /usr/src to inherit from.

Same with the /var dataset. The /var dataset is the template to inherit for /var/log, /var/tmp, /var/mail, and the others. But /var itself is not mounted.

This is one of those recurring topics and I have saved off this reference to previous discussion of it: forums.freebsd.org/threads/why-are-…var-not-in-a-boot-environment.59844

Here is the corresponding zfs list output from a default freebsd installer run: paste.debian.net/plain/1331799

I guess I should complete the set with a df listing too. paste.debian.net/plain/1331800

With zfs things are mapped to df but they don't really correspond directly. So zfs list is the better tool. And I had been awake I would have put all of those into one paste for easy comparison but I wasn't thinking straight yet.

rwp: I guess that just means that /usr/whatever-else is mounted under the / dataset, just inheriting its options for /usr. While a bit surprising, perhaps, it's not confusing.

rwp, do you ever listen to 2.5 admins? i want to say they are of the opinion that if you're going with four disks, you're better off with a mirror rather than raidz2.. but i forget the reasons why .. /me hides

2.5 admins?

scoobybejesus, I am not really aware of 2.5 admins but I assume it is a podcast. I do listen to a lot of podcasts. But that means all of my listening time is pretty much already consumed.

That particular case has 4x hot swap 3.5 inch drive bays. It needs to be reliable but isn't requiring ultimate performance. Several options available.

podcast with Jim Salter and Allan Jude. and boy do they love them some zfs, so it's a very satisfying show

With a RAID10 type organization with 2x 2-disk mirrors is simple and hard to mess up. But then IF there are two drives that fail then they have to be opposite drives in the RAID10 organization. If unlucky and two drives in a mirror fail then all is lost. With raidz2 any two drives can fail and everything survives. I have had very good luck with raidz2 pulling me through problems and so that's what I chose.

I'll have to look for the Salter and Jude podcast as yes both of them are the experts in the field. I definitely enjoy reading their written articles.

If I "had my druthers" I would always have an 8x or 12x hot swap drive bay case and then populate with raidz2 as that has been a great combination for me. But alas this system has only 4x drive bays and I wanted to fully populate the storage. You can see it is empty still so you could convince me I should have done a 2x disk mirror now and had spare bays for use later. I agonized over that decision!

rwp: i have an 8 bay case, (2 for os) 6 for storage and it doesn't feel like enough bays either

it's probably from this: jrs-s.net/2015/02/06/zfs-you-should-use-mirror-vdevs-not-raidz and finding "Saving the best for last," those next two paragraphs would make his case, i guess, if you're curious

scoobybejesus, That section is basically the description of the "problem with RAID5" and why I don't like raid5 either and will only use RAID6. raidz2 is close enough to RAID6 for me to describe that way. And I totally agree with it. The resilver time can be a big problem. Especially on highly active production systems.

Ah, there you go, coolio

In RAID5 it's fine with all drives but when a drive drops out then the system loaded up more and resilvering is a high load and if the system is already highly loaded it might be TOO highly loaded and be unable to keep up. It's been a long running problem with very active storage systems.

However this particular system is for me and it's in my queue to replace an Intel Atom D525 system from 2010 or something that currently is maxed out at 4GB RAM and has a 2-disk mirror now. If that system can handle the load (just barely, always at a load of 2-5, but it keeps up) then this replacement system will have no trouble.

For me and this particular system the fractional extra robustness of raidz2/raid6 is more important than the more robust resilvering load. Because if it needed it then I could simply take this particular system down to single user mode and let it run the resilver to completion and live with it being offline during that time.

The Salter & Jude article I totally agree with. If it was a production datacenter and every day an IT tech walks the aisle looking for red lights on drives, pulls then, replaces them, always very quickly with a new replacement drive then it is very unlikely to have a 2nd drive fail in the array before the 1st one was replaced and resilvered. In that situation using mirrors is always better.

I am not always so organized with my own stuff! :-)

i'd much rather have the burden of choosing the style of zfs redundancy that is right for me, over not having that burden due to zfs not existing!

scoobybejesus, Yes! You have it EXACTLY! :-)

anybody been able to pull from git.FreeBSD.org/ports.git? I'm getting 504 errors... presumably because something is overwhelmed? Maybe due to the quarterly branch switch over?

ok so i'm using the text terminal but i can't sudo

what i've noticed is that whats actually happening is that i can't run sh

jmpp: looks ok now

with my user i'm using bash. if i try to sh, that's when it locks up

what could be causing sh to lock up with this term/termcap/whatever

hjf: what does control-t show?

ober: makes a beep

^t?

if i do ctrl-v ctrl-t it prints ^T

just ctrl-t

ctrl-t makes a beep

odd

how are you accessing console? directly? thru a browser/vnc?

crt serial terminal to a serial port

but bash works?

flawlessly. i'm even in screen right now

ober: imgur.com/a/NNjDZRP

nice

oprs: thanks, trying again

hjf: what is your $TERM set to?

oprs: still failing just the same :(

is git.FreeBSD.org some kind of proxy with geobalancing, and perhaps some close by mirror is down?

ah, yeap, git.freebsd.org says so

jmpp: tbf it did take a while, so maybe I was just lucky

that would require confirmation from someone who actualy know what they're talikng about, but I'm pretty sure you can add the github mirror as an additional remote, just in case

ober: myvp3a+ which is a customized version (cursor keys to ^[OA etc) of vp3a+

git remote add github github.com/freebsd/freebsd-ports.git

yeah, I might as well do that

(as long as it is just for fetching)

yeah, poudriere ports tree

works for me, the webserver says geodns gave me gitmir.fra.FreeBSD.org

assuming git.FreeBSD.org is directing me to my local gitmir.nyi.FreeBSD.org mirror, then why would curl -k -v gitmir.nyi.FreeBSD.org/ports.git respond with a redirect back to cgit.freebsd.org/ports ?

probably it want to see cgit.freebsd.org as host

was trying to probe the mirror directly, to see if it was suffering from any issue when trying to hit the ports repo

in any case:

git remote -v set-url origin github.com/freebsd/freebsd-ports.git && git fetch -v --unshallow origin

working! :D

objects being fetched, rather than the error with the FreeBSD mirror of some small amount of bytes being received, followed by a long wait, and then the 502 error

so it definitely seems to me like the infrastructure is struggling

a lot of people hammering it with the quarterly branch swich over, surely

stop it! :P

you know what else doesnt work besides su, sh, sudo? more and less

but all works if i go inside a screen session

hjf, I know that problem. Look at "ls -ld /dev/tty" and you don't own it.

If you "su - user37" you will end up with the problem. Instead use "sudo -i -u user37". Compare the tty, pty, /dev/tty devices in both cases. sudo sets up a new pty for the user but su does not.

why does that happen though

do i need to be operator

Why is su insufficiently programed to do the right thing there? No idea.

i'll try chmod 777 /dev/cuaU3

hahaha

It's not a question of being the operator or not. Since you ask it I say that you don't yet understand the problem.

I am not trying to be mean saying it! But... It's not a question of operator or chmod and those are hacks which will probably create more problems later.

well i suppose the serial ports don't have the right permissions for regular users

This is not related to your using an actual serial port.

oh

A non-root user using "su -" changes from me rwp to root and root has no permission problems with ioctl() calls on /dev/tty owned by rwp.

but i'm assuming it tried to do something with /dev/tty first where the user doesn't have full permissions or something like that

But then the opposite creates the problem. A root user using "su - rwp" becomes rwp and the /dev/tty is owned by root. All ioctl() calls upon it fail. That's what kills more, less, and now sh which is setting up a line editor now. sh doing that is new in the grand scheme of things and doesn't handle the case well. bash has been doing that longer and handles it better.

but the problem is that i'm hjf and want to become root

via sudo or whatever

How did you become hjf initially? I am assuming that if you are seeing this problem that you had to have gotten there with "su - hjf" to have this problem. No?

getty got me there

getty + login got you there? There is a bug in there somewhere then.

the problem is that nothing works, not even truss sudo

all i can do is ctrl-c and then sudo will print out its usage info

After a fresh getty+login do a "ls -ld /dev/tty" and a "tty" and please tell me what it says.

The reason screen&tmux cause things to work is that those set up fresh ptys for each window. And those ptys are owned by hjf and so everything works.

should bios fan speed curves continue to work under freebsd? there is no way to control fan speeds right?

crb: on Dell servers, IPMI works

Dereckson: I'm asking more about consumer grade Risen motherboards from th 5900 series

On workstation I would perhaps try to control them indirectly through powerd

That's what I did on a laptop

I am pretty sure the ACPI bios stuff takes care of that. My mone FBSD machine has fan controls like that ob it, seems to work ok. It's around 3-4 years old with Ryzen 6 core

s/mone/one/

I feel temps ramping up and don't hear the fans ramp so I'm concerned, on my EPYC server however there is IPMI control for the fans but boy is it confusing as to how to set it up

powerd > performance mode

they should kick more quickly

release.sh just hates smp when it hits perl.

I do not believe powerd does fan control. It is setting CPU frequency scaling and ACPI misc

if someone can clarify this, go ahead

servers are different and have crazy static pressure and usually IPMI does this, the OS has limited maneuverability

you have to install fand don't you?