I'm gonna have to test it further, but it seems like setting compat.linuxkpi.semaphores=1 in /boot/loader.conf has maybe made an impact on the fault errors

I feel like this requires a pretty firm knocking on wood though, since several things so far have given the impression that the fault errors are fixed

oh good

anybody know the answer to my earlier question of gstat vs camcontrol?

basically i added a sata, ide, and scsi drive to this vm but i'm not sure how to identify that in freebsd

i gave them different sizes so i can cheat though

sata comes up as vtbd0

I prefer camcontrol because it identifies the device by model. If I want the serial number though I query the device with "smartctl -i /dev/...". There is also "geom -t" which is very useful after it has been partitioned.

Let's not forget diskinfo -cit /dev/... either.

do virtual sata disks have a serial ?

i see

rtprio: that'

s part of the problem, vm's aren't exactly making it easy to identify the drive

it just says VIRTUAL DRIVE or QEMU DISK

bit weird how many dropped frames I get on youtube under firefox on FreeBSD, even with hardware accelerated decoding, while on my PC, using firefox on Linux, with a CPU from the same exact generation, I basically get a small fraction of this number despite using software decoding

where are you trying to identify th drive? what's the vm guest?

freebsd?

yes

like here I've got about 120 dropped out of about 6300 total frames

i realized i couldn't tell sata from ide or whatnot unless maybe it gave it a different device type

and what vm host?

i use virtualbox on windows and libvirt on linux. linux in this case

I am looking at my vm and I cannot run smartctl on a virtual drive. diskinfo -v /dev/vtbd0 does not report any id information.

i couldn't either

Maybe it would be good to add the devices one at a time and then get them configured with a UUID so that the confusion is avoided. :-)

johnjaye: do they not iterate in order?

sure. but i wanted to know the way to do it for real in freebsd

But you said they were different sizes. So you can tell them apart that way, right?

they do. in this case

well yes. but that's because i did that on purpose

i was trying to see if i could tell them apart based on the type of device

On bare metal systems which I assume is what is meant by "for real" then the devices do have models and serial numbers.

ah ok

yep, they do.

as in the device id and vendor id too right

$ sudo camcontrol devlist

<WDC WD4005FZBX-00K5WB0 01.01A01> at scbus0 target 0 lun 0 (pass0,ada0)

seems picom is responsible for a lot of these frame drops. guess I shouldn't be surprised

what is ada vs da again?

johnjaye, Two examples: bsd.to/06in/raw

Title: 06in

da is scsi, ada is sata

SATA controllers are adaX and SCSI controllers are daX.

is 'attachment' the type of controller?

Generally random desktop hardware will have SATA controllers and use ada names. Fancy systems with high performance SAS/SATA/SCSI and use da names.

ok. i think libvirt got that right because it made the scsi one da0

But... USB drives also use da names too. So... I wish USB controllers were different.

eh

ok

On my desktop when I plug in a USB SD card then I get /dev/da0 for it and my SATA disks are /dev/ada0 and /dev/ada1 so there is no confusing. But on my NAS with the SAS controller all of the disks have /dev/da0 through da5 names and plugging in a USB there results in a /dev/da6 name. And I am VERY careful about which is which.

rwp: same: ada0 and ada1, ada4-7 ar the onboard controller here, and ada2,ada3 are the expansion card? for some reason?

rtprio, No idea. But they don't have names unless something is attached there, right? What are those?

yes, i have 8 drives.

You have 8 drives so it makes sense you would have /dev/ada0 through ada7 then, right? What am I missing?

that the expansion card is assigned ada2 and 3 rather than ada0,1 or ada 6,7

What does "sysctl kern.disks" say?

kern.disks: ada6 ada5 ada4 ada0 ada7 ada1 ada3 ada2

That looks perfectly regular to me. Out of order, yes, but all of the devices are there. Those are the dynamically assigned kernel names anyway. Which is why UUIDs or gpt labels are recommended.

i labeled the OS drives, but one label was lost

Very likely your motherboard is using two different SATA controllers, at the least. It would be good to check that all are 6Gbs as many motherboards might only have that for two of the ports.

yes, it is, the onboard controller only has 6 ports

One of my motherboards has huge silk screen lettering 6Gbps SATA! but only two of the ports are 6 and the rest on that motherboard are 3.

i don't recall seeing that

i have " block size: 512B configured, 4096B native " also, which i expect is not doing me any favors either

For the SATA port speed the best I know is smartctl for example: bsd.to/hVUe/raw

Title: hVUe

i'll be damned bsd.to/7Bep

Title: dpaste/7Bep (Plain Text)

And here is a 3Gbs here too: bsd.to/JynG/raw

Title: JynG

my motherboard is like that. half the sata ports are 6Gb the other 3Gb

why? no idea.

Honestly that is pretty typical vendor marketing. It's a buyer beware market.

And also honestly it must be fast enough because it wasn't stopping any of us from using it. Sorry to disillusion.

'current' does that mean the drive supports 6 but the controller does not?

Yes.

you can see which drives i've updated over time with that too

Yes. By the different SATA versions. However that is going to be fuzzy because it all depends upon vendor and such.

they're all wd. the oldest drives sadly have 94500 hours

I have a couple of Hitachis that are past 100,000 hours!

i'm impressed

I assume these are in a RAID configuration running zfs? If so then there is no concern even if one turns up failed. Just replace it when it does. Until then use it until it fails.

2 os drives, mirrored. 6 raidz2. yep.

at some point i need to buy a whole set of drives and migrate because i'm out of space and this sector size thing is annoying and can't be repaired

On 2022-10-18 I recorded this for one of the drives: "9 Power_On_Hours 0x0012 085 085 000 Old_age Always - 108935" that's 109k hours! One of the drives in that pair did start to throw an error last year.

12.4 years. oof

A 1TB Deskstar. Those Hitachis were awesomely good.

The other one in the pair on that day was recorded at 102878 hours.

Oh, and it was the 102k hour drive that started to throw errors. The 109k drive is still going. I just don't have it plugged into anything at the moment. 1TB is so small now afterall. :-)

yeah it is

To avoid the "slow" 3Gbps port problem I am using three different LSI SAS 2008 HBA cards running with IT Initiator Target firmware. Mine are Dell surplus. You can get them inexpensively surplus now. Those are SAS/SATA controllers and will provide 6Gbps to 8 ports using multiple aggregated PCI channel bandwidth. If you have a spare PCI-e slot then that's a good option for increasing disk I/O performance.

i actually had a dell hba card in this, which freebsd casually dropped support for between 13 and 14

twa i believe it was. that's what started my whole mess

That's so annoying when things are working and then support is dropped for it. I have been there many times and find it terribly frustrating.

yeah, just a harmless buildworld/installworld and ... where's the pool

Oh that would be annoying. But you can boot to the previous boot environment and get back going again. But not able to upgrade. Drat!

I think I was able to buy each of these Dell LSI HBA adaptor cards for less than US$50 each at different times. It's a way forward.

And there are fancier cards for more dollars too.

my other system has a mfi one and sas drives, using hardware raid since passthrough on that card seemed annoying

Note that these do get hot without airflow. I have ensured a fan for them in order to keep the temperature down. With a fan they are fine. Without they became so hot that I could not keep a finger on the heat sink.

When it comes to CVEs in packages, who is responsible to updating the package. The security team or the package manager?

port maintainer generally

remiliascarlet: tykling: yep, same zfs code used for linux and freebsd. but linux and freebsd are different kernels, so freebsd-only and linux-only zfs bugs are possible if they involve some kernel-specific interaction

someone (can't remember who offhand) was asserting that no one on freebsd had encountered the zfs encryption + snapshotting + zfs send issue, and there was speculation that it was all due to a linux kernel specific interaction of some kind

but i think that's been disproven

i mean, some freebsd users have seen the same problem right?

andjjj23: yes, i'm having that exact problem right now on 14.0

I'm sending unencrypted data snapshots to a remote encrypted dataset (trough zrepl) and I don't have any problem (13.2)

(offsite backup)

5

mage: the problem seems to be if the sending side is encrypted

andjjj23: I wonder what about the ZFS in OpenIndiana?

Or the one in the mother of ZFS, Sun Solaris?

remiliascarlet: Pretty sure that's still Illumos-based and hasn't moved to OpenZFS.

Or rather, the new definition of OpenZFS.

is openindiana reengineered all the code or is it just Sun's code?

Sun's largely as I understand it.

ok. so kinda different to linux and the *bsds which reengineered all the original AT&T stuff over time

That's part of the unix history I didn't understand initially. That the CSRG rewrote the entire system

Might be more accurate to say they zealously wrote so much that there wasn't a lot of AT&T left that mattered.

CSRG didn't rewrite the entire system, they just released a non-functional OS called 4.4BSD-Lite which removed the remaining AT&T code but as a result didn't compile or boot

386BSD and others rewrote the removed code to make it functional again

That's the last chapter. CSRG did a ton before then.

yeah, i mean at that point the vast majority of the code was probably CSRG, just not all of it

Yeah.

i wonder what BSD/OS was based off... did they have a UNIX source license to use the full 4.4BSD code? or did they start from 386BSD

that was mostly ex-CSRG people, right

They had tapes from AT&T.

The legality of the arrangement was all very relaxed in the very early days.

lw: If you haven't read it, you'd love Salus.

There's a digital copy available on ... um. Forgetting the name. TUHS?

mason: hmm, that's an interesting way to look at it. so if i join your open source project and write 10x as much code as you have... now it's my project?

mason: this one? wiki.tuhs.org/doku.php?id=publications:quarter_century_of_unix

Title: publications:quarter_century_of_unix [Unix Heritage Wiki]

johnjaye: Licensing isn't done on the project level.

mason: Forked from OpenSolaris's last (and half broken) release, and continued work more or less independently from there on.

lw: That's the one. I'd recommend a paper copy if you can get one, but that works.

mason: i'll have a look, ty

i started using BSD around 1996 so i just missed a lot of this earlier stuff (and wasn't really paying attention at the time)

mason: i'm not saying you're wrong. it's an interesting idea

I started using Linux in the early 1990s, and only recently discovered BSD.

johnjaye: Not even an idea so much as a description of what happened.

lw: Kirk McKusick and Michael Karels have given talks at BSD confs that were recorded about a lot of the history

There's a good textual example at the beginning of D&I.

i lost my copy of D&I :-( probably time to replace it with the new freebsd edition anyway

Well, I know FreeBSD existed much earlier on, but I thought "FreeBSD runs Gnome, and so does Debian, so what's even the difference?" at that time. But I recently found out that in order to appreciate FreeBSD and OpenBSD, you need to use it for at least a while, and now I love both of them more than Linux.

lost my two volumes of UNP and TAOCP at the same time, that was annoying, they're all a bit dated but still useful references

the TLAs are getting a bit heavy. what's UNP?

johnjaye: "UNIX Network Programming", a popular book on writing network applications for Unix

Richard Stevens' books, yes.

What happened to to "TCP/IP Illustrated?"

Stevens passed I think.,

Title: W. Richard Stevens - Wikipedia

So he's not writing more.

mason: yes, i believe someone else updated the books for a new edition a few years back though

Ah, I'd missed that.

yes, RIP Richard Stevens. I have I think all his editions.

one of my shelves: lowsig.com/shelf1.jpg

all those books helped with my early career. the magic garden books were useful when working on SGI/MIPS/IRIX.

There were 2 new editions of APUE with Seven Rago (?) writing the updates - Pearson has the copyright.

Steven Rago

Another great early Linux book is "Operating Systems - Design and Impl" by Andrew S. Tannenbaum

OK it's Stepehen

I got my copy when I bought Minix for ATART ST. Took all weekend to build Minux from source on an ATARI but was super fun.

Tannenbaum is Minix oops, not Linux but Linus used that book too IIRC

am i missing something incredibly obvious here? /dev/md0s2 exists, but mount says no such file or directory... le-fay.org/tmp/30d/M026Wo.txt

oh, it's md0s2a, stupid BSD labels

weird error though as md0s2 clearly does exist

filed bug as i think this is wrong bugs.freebsd.org/bugzilla/show_bug.cgi?id=277264

Title: 277264 – mount(8)ing incorrect device gives confusing ENOENT

am i missing something or is there no way to tell wireguard what address to listen / bind on?

it can be any address you want, as long as the one you want is *

yeah, might go back to IPsec if this isn't supported...

or, considerig all my traffic should be encrypted anyway, maybe i'll just stop using a vpn and run over ipv6

right, you get just the port at best

why would you not use ipv6 over that vpn?

nimaje: i am using ipv6 over the vpn?

nimaje: what i'm considering is to just not use a vpn at all and use ipv6 directly between the two hosts

because all the traffic is over tls or ssh anyway, so... what's the vpn really giving me in terms of security

ok, that 'maybe i'll just stop using a vpn and run over ipv6' read like it was a choise between vpn and ipv6

oh, no

the choice is vpn-over-ipv6 + ipv6-over-vpn vs just ipv6-over-internet

i think i set up these vpns initially because a) ipv4 brain (everything goes over vpn) and b) it means everything can have an address in my local /48, but neither of those are really valid reasons

Hi, I have a jail inside a zfs dataset. think of it as tank/jails/jail01 and I want my jail to be able to manage its own dataset, for example create dataset under its dataset like tank/jails/jail01/test. the zfs device has been unhided with devfs.rules, and all of the allow.mount, allow.mount.zfs parameters are have been set. However my jail can't create a dataset under its own. I read the jailed property

of the zfs. But when I want to set it, it need to remount the dataset and obviously it is not possible when the jail is alive. even before starting the jail, you can not use the jailed property because the it will be hide from the host point of view and you can not start the jail because the root directory of it is not mounted.

s/it need/it needs/g s/because the it/because it/g

lw, I need a C++ library that does utf-8 which comes with algorithms such as starts_with(), ends_with(), substring(), replace(), count_occurrences() and that sort of thing. Ideally with conversion to/from other encodings but that is optional. Any recommendations?

lw, I was looking at github.com/tzlaine/text but...

Title: GitHub - tzlaine/text: What a c++ standard Unicode library might look like.

jbo: i don't know, i was in the middle of writing my own but got distracted. there was a reference implementation of std::text i came across once but i could never find it again

:<

spmzt: pretty sure jailed=on should be set before starting the jail

or before attaching the dataset to it

you will probably need to add `zfs jail ....` to exec.poststart

i've never had any luck delegating zfs to a jail - even if you get the jail to start, it hangs on shutdown because it doesn't unmount the filesystems properly (zfs unmount rc script is disabled for jails)

would be very interested if anyone got that working, even though i've mostly migrated to bhyve now

use exec.*stop

I had no issues running poudriere in a jail with ZFS

and poudriere loves managing dataset

Remilia: what do you put in exec.stop?

you can use zfs.unjail

er

does that unmount the filesystems?

`zfs unjail` obviously

lw: I forgot since poudriere unmounts everything after a run

so maybe I never considered that myself

It'd be pretty annoying if it didn't clean up after itself.

but if you are *attaching* a dataset you do not need to unmount it

ah yeah. i wanted to run an app that just had a filesystem and some children

you need to detach it

'detach'?

zfs unjail

so unjail unmounts the filesystems?

it detaches the dataset from the jail

but does it unmount the filesystems? the problem i had is the jail was stuck in dying state because the zfs filesystems were still mounted

as soon as i manually unmounted them, the jail would stop

if you are mounting them manually, unmount them in jail.stop

i was not mounting them manually, the jail did that from rc.d

I think we had very different scenarios

and right now it is way too late for me to start experimenting with this, maybe I will try tomorrow

loooking at /etc/rc.d/zfs, it seems like this is *meant* to work, there is specific support for unmounting filesystems in jail

well, i don't use jails anymore, but maybe i'll have another poke at it one day

I am guessing in my case it just called rc.d/zfs

Remilia: you can't set it before starting the jail. check this: bsd.to/1V2c

Title: dpaste/1V2c (Bash)

Yea, zfs jail and unjail needs to be done exec.poststart and exec.prestop respectively.