Demosthenex: well, they're not block devices.

Are you running securelevel though?

JKHII: that's a question only the security officer can answer

is there a plan to rewrite freebsd in rust so it's blazing-fast and secure?

alip: Better if it were Common Lisp.

alip: unlikely in its enirety, but there are attempts of writing drivers in Rust out there

alip: I hope no. You already created your own

alip: why are you doing this? now i'm expecting nightmares

Drivers in rust are interesting, because it offers safer way of writing kernel modules. I can see a future where FreeBSD maintains rust KPI bindings, but rewriting entire OS in Rust has negative ROI, it is easier to write a new OS from scratch. Same deal with Linux. It will never be rewritten in Rust completely

ngortheone: another one :(

?

i will see nightmares after that. looks like you didn't read last news

well, thankfully the quality of your sleep has nothing to do with the question at hand :P

didn't read or just a mantra fan, in this does not matter, fans just does not see. it always just different

just read

and try to forget about mantra

i saw how thy created their own system, i saw how they tried to rewrite ThreadX

believe me, it is nightmare

ok, ok, take it easy, deeaitch, nobody is rewriting FreeBSD in Rust

...yet :P

alip: in case your question was not a flame starter - take a look at wiki.freebsd.org/Rust, there you will see links to attempts writing Rust kernel modules

Why just do not finish own system? Do you know why they cant?

because i know, in our company we tried big project for a 2 years

we tired embedded system, irmc

Hi all, I tried the re_kmod package for realtek (my NIC has realtek 8111C and generic kernel does not support it). This card is new, besides the fact it is former PCI format. The NIC is not detected and I am still looking how to configure/fix.

pciconfig gives me this as output: "none3@pci0:1:6:0: class=0x000000 rev=0x10 hdr=0x00 vendor=0x10e8 device=0x82b0 subvendor=0x0004 subdevice=0x0000"

Kalten: no, i was deliberately trying to sabotage that disk by overwriting it online.

Kalten: on linux there is no protection, and every article including the dd command has huge warnings that you need to triplecheck what you dd over

debdrup: well, block devices... erm. i'm making a general statement i suppose ;]

debdrup: as to secure level, just a default install with no packages yet. i did check most of the security boxes in the installer

When I say securelevel, I mean the kern.securitylevel OID in sysctl(8)

Except it's kern.securelevel. :D

Good day

I'm having an issues with freebsd stock tftpd during PXE booting devices

any advise which one to use?

i prefer atftpd (different ones worked with PXE very unreliably)

otis, thanks for reply!

cant find it in ports/ftp

hm, it's expired from ports, indeed. it's been very long time since i last used pxe+tftp :-9

:-(

in my case old machines booting ok, but new ones shows weird behavior, like instant brake

anyway thanks for suggestion

pxe linux manual contains a good review about tftp-hpa serer

oh yes, that one is also a good alternative. i couldn't recall it (but i've also used it)

otis thank you!

What issues are you having with tftpd?

hi. let's say you're running poudriere and it usually builds against say 2022Q4. if a port gets updated in HEAD but not in the 2022Q4 ports branch, how can one apply the update from HEAD to that (local) branch?

debdrup, DATA block 1, attempt 0 failed (Error 40: Message too long)

something like this one

dell optiplex desktop workstation, not able to transfer file at all while tftpd say - transfer successful

in general mu guess is a kinda of packet size ussue

*issue

didn't go deeper at the moment, ( not followed tcpdump traffic )

f451: You can just copy the directory from HEAD. But it's not pretty though, and I would not recommend it.

and if it depends on more current dependencies, it won't work...

souji: ty

i thought there might have been a git command to take care of the deps

f451: you can use overlay

so, make an overlay for that one port

i use overlay in the context of sccache

yes, I do that for phpfpm exporter

er

what is that?

like ccache but for rust

really speeds up rust

oh

anyway, you create a new ports tree in poudriere which will just be the overrides of the ports you need

and specify that with the -O option

theres sccache and there's the poudriere thing for it

yes

it works fine for me

ports-mgmt/sccache-overlay

uhh

I am not sure if we are talking about the same thing

sorry, I thought you were talking about poudriere

ignore my comments above

its my fault - i brought sccache into the discussion ;)

i am

so I was talking about poudriere bulk -O

yes got sidetracked soz

yes i use -O in the context of sccache

-O overlay \Specify an extra poudriere-ports(8) tree to use as an overlay. \Multiple -O overlay arguments may be specified to stack them.

afaik sccache-overlay is an overlay to use sccache with all ports and poudriere has an mechanism to use overlays

sorry for the confusion

so your overlay for yr php program - it's just that one port?

yes

awesome. i have my answer :D

means that if a port gets an important update, i can overlay that update

poudriere% ls …/localpatches \net-mgmt

poudriere% ls …/localpatches/net-mgmt/phpfpm_exporter \distinfo distinfo.orig files Makefile Makefile.orig pkg-descr

if i understand coorrectly

yes

just create a new null-sourced 'tree' that points at a directory with the usual structure

excellent, thats just what i need

thank you

note it might be a poudriere-devel feature

i use that anyway

mainly for it fetching deps from the pkg cluster

i have two encrypted ZFS drives in a mirror zpool. i only have two hard drive slots in my computer, so i want to replace the disks one at a time (with bigger disks) and re-silver after each replacement. however, when i use zpool-attach or zpool-replace, i get "one or more devices is currently unavailable"

boot into a rescue disk, import the pool and unmount all filesystems then try again.

with replace...

The update to 13.1-p6 did not go well. The /boot/efi slice is marked as dirty. fsck won't fix it.

You tried from single user, without it mounted and used 'fsck -y'

?

No problem when in single user mode. However, the moment I try to do a normal boot I get the same error and get dropped into the singleu user mode.

Not really what I asked.

Yes. I had done that. The fs did not come back as dirty.

I now see it is complaining about a different drive.

CmdLnKid: what's the fundemental problem that requires a rescue disk? i've tried more things here (other than rescue disk) including creating a brand new mirrored zpool and a few other things and the result is always the same "one or more devices is currently unavailable"

The old boot drive. A couple of reboots later fsck is running on it!

14.0 rather

ober, disklabel and bsdlabel have been hard linked for ages. .

I guess they decided to drop the disklabel name finally.

I see. /me shows his age

Back up! Thanks

ok so i fixed my issue. it was caused by not initializing/zeroing the new device

i zero'd the first 500mb and then it was accepted to replace an offline device

type: netbsd-ffs

morning

i'm still pumped over my storage testing. freebsd's storage layer is so.... SANE

and mirroring my OS, a dream come true

(gpt/bios mode only, no uefi, some restrictions may apply, batteries included)

If it's a system that's been manufactured in the last decade, it's probably UEFI with CSM.

well, i tried uefi, but that bug in the freebsd uefi installer didn't mirror

which i confirmed.

so i reverted to bios

because... though i could build anything i wanted, i desired the out of box experience

There's no way to "revert to BIOS" with UEFI.

Title: 258987 – 13.0-RELEASE installer broken redundancy with UEFI and ZFS

oh i just told the bios to use legacy mode and disabled uefi

and reinstalled

i think i reinstalled like 4 times tweaking

and then i beat the snot out of the disk redundancy

i'm very pleased

UEFI isn't disabled, "legacy-mode" is UEFI-CSM.

ah.

It's UEFI that's emulating the INT 0x17 call, which is the way that most BIOS' boot.

well freebsd could tell. when i had the bios in "bios" mode, it's text only in teh boot screen. UEFI in bios had the graphical loader

yep

so rctl needs kern.racct.enable=1 to work, but login.conf has no mention of this sysctl. does this mean that login.conf resource limits work in some different way?

A real BIOS only works if your CPU is starting in realmode, whereas UEFI starts in longmode and doesn't ever drop below that.

debdrup: greek to me. i haven't done intel assembly since highschool. but i can talk about making storage reliable all day

ngortheone: rctl uses an entirely different mechanism than the ones you can define in login.conf and display/set via limits(1).

If memory serves, rctl used to be called "resource containers" to distinguish it a bit more.

i was even more thrilled with the gmirror man page. so common sense

ngortheone: wiki.freebsd.org/Hierarchical_Resource_Limits#Implementation_details

Title: Hierarchical_Resource_Limits - FreeBSD Wiki

Oh, resource containers were an intermediary name. Hierarchical resource limits was the original name.

having trouble with my "das keyboard" and swedish layout. trying to type gt & lt symbols. But I get ‹ and › instead... =) any good ideas on how to solve this?

in linux and previously in openbsd (as far as I can remember). I was able to produce the symbols with alt-gr+shift+z and x

drobban: are you sure the das keyboard implements the HID classes properly? Lots of newer keyboards that do n-key rollover have a rather nasty habit of making a mess of the HID specification.

debdrup: oooh, not sure.

N-key rollover on USB isn't really doable within the HID spec, if memory serves.

okey.

any ideas on how to error

search it

No clue.

I suspect you'll need to talk with the engineers who built the keyboard.

=D

how can I make x run with the same keymap as console. The mapping seems to work fine in console

Oh, it's xorg that's causing problems? I thought it was FreeBSD.

Xorg has nothing to do with FreeBSD other than that it's provided in FreeBSD Ports

you are right

I think you might be able to use the xf86-input-evdev driver.

Although I'm not entirely sure if that applies to the character map.

Oh wait, it's for v4l2 devices.

Nothing to do but try and mess with xorg keyboard mapping then, I suspect.

I do run with evdev. get different result between freebsd and debian with the same setxkbmap

freebsd seems to give the "correct" - unfortunately for me =D

I'm not really sure I'm qualified to help with xorg-related stuff, as I use wayland+sway (launched using the ly display manager)

check =)

with zfs mirror devices, is data read from both devices mirrored ? or just the 'main' one ?

can you rephrase the question?

Demosthenex: I think what the question is... does zfs do parallel reads

like if you're asking for an 8k block, does it read 4k from A and 4k from B

Some mirroring systems will do that.. some mirroring systems will read from the 'primary' only

that's a great way to put it

correct

last1: the only way to ensure every record on every device is read is through scrub, but to answer the specifics of your question, whichever disk first returns an answer that matches the checksum is the one that it gets read from as)

is that true of zfs only or are you aware of hardware implementations that can do different reads on all mirror devices ?

so if I have a 10 drive raid 10, it would read different parts of the file from all 10 drives at once

I avoid hardware RAID like the plague, because it's just software RAID written by someone who used to work for the company, but they're still builting it out of his home directory because it's so brittle that if they move it it breaks.

last1: that's how most striped mirrors work, yes - otherwise, you wouldn't get the increased IOPS from using striped mirrors.

debdrup: well... unless you're using a giant flash array that does 750k iops. then you might stick to hw raid ;]

Most hardware RAID is a RTOS running on a ~500MHz MIPS CPU (or maybe ARM nowadays).

ah ok, and would you happen to know why zfs doesn't read like that ?

Demosthenex: absolutely not.

ZFS or bust.

last1: it does.

Remember, the data is striped across the mirrors.

yes, but if I have to read a 1 GB file striped across 8 mirrors. I want to read block 1 from mirror1-A, block 2 from mirror2-B, etc

you said right now zfs just requests a read block and one of the mirorred drives answers fastest, it will go with that

it's not really using both mirrored devices for independent reads (?)

or have I misunderstood

last1: for a striped pair of mirrors, it'll read in the pattern you described because that's how the data was written.

err, sorry

Your first question was about mirrors only, not striped mirrors.

I meant block1 from mirror1-A, block2 from mirror1-B

When a series of zfs records is written to a striped pair of mirrors, it writes the first record to mirror0 then the the second record to mirror1, and so on and so forth. When you then read that data, the reads will be striped as a function of how they were written.

If it didn't work like that, there wouldn't be an IOPS speedup in using striped mirrors.

yes, that's at the mirror level

but each mirror is made up of two ( or more ) devices

with identical copies of the data

can't reads be leveraged against those copies in parallel ?

That doesn't make sense.

hmm, maybe I'm imagining things, let me rephrase

A striped mirror is two sets of mirrors where one pair receives half of the writes, the other pair receives the other half of the writes.

That's how it's always worked.

Are you suggesting you want mirrored stripes?

no no, so with striped mirrors, I have copies of my stripes

can't those copies be read in parallel ?

What would that achieve?

I'm guessing more speed ?

right now I believe only one member of the mirror gets hit with all the requests and the other sits (almost) idle, doing just copies of the first one

You'd have to have some way of interleaving the two datastreams, and I'm not sure that could be done reliably.

Reads don't exhaust disks, unless you buy disks so cheap that they're not worth the money you paid for them.

ok, so to recap: I am correct that only one member of the mirror gets hit with most requests, right ?

No.

right, it's the one that answers quickest

Yes, and statistically that's a 50/50 if they're the exact same make and model with the same firmware et cetera ad nauseum.

debdrup, that's going to cost you a demerit.

CrtxReavr: what.

Using "et cetera" and "ad nauseum" in the same sentence.

"Et cetera ad nauseum" means "and so forth until the point of nausea".

last1: but if you have multi-threaded reads, they are going to be spread out on both mirrors. but single-threaded reads, probably not in practice. I think it would depend on read-aheads acting like concurrent reads of sorts. but it doesn't seem to increase the read bandwidth in practice. probably depends on a ton of stuff though. possibly depends on fragmentation even.

sphex: it's all pointless, because there's much easier ways to depessimise performance on NVMe disks (which ZFS isn't especially good at utilizing, because it was designed about two decades ago, see youtube.com/watch?v=v8sl8gj9UnA).

Title: Scaling ZFS for NVMe - Allan Jude - EuroBSDcon 2022 - YouTube

CrtxReavr: the two words are not synonymous.

Didn't say they were.

sphex: and rotational media would be depessimized by the fact that reading data off the platter against the speed of rotation is different from reading it with the direction of rotation.

Though I think they're phrases, not words.

CrtxReavr: then I'm not sure what your point is.

CrtxReavr: pedantry will get you nowhere.

it was a joke. . . lighten up.

debdrup: I think it was your pedantry that started it :)

Kobaz: I suggest you find anything better to do than proking people.

proking? is that a word?

debdrup: my understanding is that the read-ahead would have to be very large to consistently speed up single-threaded mirrored reads. because it works on raid-z, it does speed up single-threaded reads. if you interleave the reads enough, it could have a similar effect on mirrors. but on NVME yeah I can assume it's a whole different story. kind of crazy how fast these things can be.

looks like proking makes vehicle transmissions

s/proking/provoking/

Multiplayer vi still doesn't implement regexs though.

oh dear

sphex: take some time to watch Allans talk, it's quite informative (as one might expect).

are there other FS's which are faster on nvme than zfs ?

FAT32.

UFS is faster than ZFS on NVMe.

That doesn't mean it's better.

debdrup: yeah I will later.

Having "speed" be the only factor of importance in a conversation about filesystems, that ought to include data availability, administrative ease, and many other subjects feels.. lacking, to say the least.

ah always nice to see that ZFS is still getting improved.

debdrup: I was just following up on your comment from above: there's much easier ways to depessimise performance on NVMe disks (which ZFS isn't especially good at utilizing, because it was designed about two decades ago,

Preventing data corruption ought to be the most important thing that a filesystem should do, if you ask me.

and I equaled performance with speed

last1: depessiming performance means making things go faster, but the important part of that sentence that was unsaid (because I thought it needn't be said) is that ZFS has a proven record with first priority as mentioned above.

ZFS has a pretty good track record when it comes to preventing data corruption.

it does, I only survived an unexpected power outage due to our bkp server running zfs

our ufs file servers took 3 days to complete fsck

( years ago, using hdd )

good ol' fsck

Another example of a piece of software getting things right is Postgres. Unlike other databases that shall remain nameless, it was written to be ACID compliant first, and then focus got shifted onto making it performant.

aaahahaa

mysql has a nice long and fun history of being silently failing or broken in some way

It's gotten a lot better...

That wasn't really my point, though.

mysql... no thanks

mysql... well, being stable wasn't the primary goal

or enforce any kind of compliance... so yeah

Being acid is a good way to date a base.

Kobaz are you alleging that Oracle's acquisition of MySQL was a good thing?

CrtxReavr: I haven't followed it much since we switched to postgres. I think making it more of a proper database would have happened regardless who owned it.... me personally I would rather have more things more community-based and less controlled by corporate

But those good-natured community projects crumble under the weight of selling out for millions of dollars.. and it's hard to blame them.. that's a lot of friggin money

But now we have maria which is basically mysql-community continued... so in the end it probably didn't matter a whole lot

anyone self-hosting a matrix homeserver?

true, but not everything needs to be perfect from the beginning

and 95% of small business cases do just fine with mysql

yourfate: I'm not sure that question belongs in #FreeBSD

I wanted to ask if one of those is easier to set up / better supported on freeBSD

as my server runs that :D

like, one of the different server implementations

last1: we started with mysql, and then moved to firebird when interbase got open sourced, and then ran into tons of limitations and then switched the postgres and never looked back. Had I known what I know now, I would have done postgres from the getgo no questions asked

and i would never recommend mysql for anyone after what I went through... heh

sure it's fast and easy to use, but at what cost when you outgrow it and need an actual properly acid compliant database with things like transaction safe DDL and whatnot... you'll be surprised how fast you need that stuff

ah, dendride is in pkg even

Kobaz, from my experience, the vast majority of what's done with SQL can be accomplished with SQLite, faster and more reliably.

depends what you're doing... sqlite can only have one writer at a time

There's certainly implementations that need more, but the majority of SQL deployments are needless overkill.

quickbooks could use some sqlite

that would be a million percent upgrade

Hello! It's me again with my questions from a newbie learning the FreeBSD operating system. Yesterday I installed the FreeBSD operating system on a VirtualBox virtual machine. I have a question about the kernel of the system. Please tell me why the /kernel.old/ directory appeared in the /boot/ directory, despite the fact that I did not build a new kernel? When you turn on a virtual machine with the FreeBSD operating system, in the

operating system bootloader in paragraph 6 there is an option to select a kernel between "defaul/kernel" and "kernel.old".

Kit_Leopold: there's absolutely nothing bad about having a kernel.old directory (though I can't tell you exactly where it comes from).

did you run freebsd-update?

gotta update those updates

I ran the "freebsd-update fetch" and "freebsd-update install" commands right after installing the system.

That'll probably be it, then.

Kit_Leopold: that's the reason

Oh, I beg your pardon, as I myself could not understand this reason.

Kit_Leopold, is this kernel.old directory populated?

CrtxReavr: Yes, there are many ".ko" files.

There's probably also a kernel file.

I didn't read the scroll well. .

But I concurr it was probably freebsd-update that did that.

debdrup: Yes, in this directory there is a kernel file called "kernel" without extension.

That'd be the kernel binary, yes.

I ran the command "diff -r /boot/kernel/ /boot/kernel.old/" in the console and saw the changes that were made during the system update that I called with the commands "freebsd-update fetch" and "freebsd-update-install". Thank you all for your help, you helped me a lot with solving this riddle.

cgit.freebsd.org/src/commit/?id=48bfd3597654 Neat!

Title: src - FreeBSD source tree

"Print the current program version and exit. Don't use this option."

now i really need to try it