I haven't been able to find a clear answer on this, but is there a way to zfs send (or something else) a whole pool to make a full backup, which can be zfs received later? I have a broken configuration on my vdev (ashift=9 instead of 12) so I want to dump it somewhere, recreate the vdev and then send it back, maintaining all the datasets and snapshots

Axman6: You can capture the stream to a file, although this is more fragile than having it temporarily live in another pool.

hmm, may be I missed this, but sending the pool using zfs send tank would actually work, and retain all the snapshots etc?

Axman6: You'd want to send a replication stream, -R, but it's just a stream of data at that point, while it's in-flight. But you'd be far, far better having it live inside another pool for the purpose, rather than just sitting as a bare file somewhere.

yeah that's probably the way I'll need to go. Sadly to do that I'll need to buy a lot of new hardware

Axman6: Look on the bright side. That extra hardware can then turn into a dedicated back-up pool.

Hello when I build curl 7.88.1 from port I have still a security report about vulnerabilities but when I look on the official website I don't see any threat on the last 2 versions .

Maybe it takes time for the site to update but I would like to know the exact threat. How can we get this information?

Ok I have found the reference thanks to vuxml

and what was it?

Indeed these threat are resolved in the last versions and are not referenced in vuxml anymore.

meena, they are here curl.se/docs/security.html

Title: curl - Security

Title: FreeBSD VuXML - curl

maybe there is something which is not up to date on my system

pkg audit doesn't show any threat too

So I'm not sure whether to rely on the security report provided by the port system

are you on quarterly?

yes

no sorry I am on main

main on both kernel and port. Is it bad ?

Midjak: mind showing the exact security report itself?

they are 3 fixed by 7.88.0 cve.mitre.org/cgi-bin/cvename.cgi?name=2023-23914 cve.mitre.org/cgi-bin/cvename.cgi?name=2023-23915 and cve.mitre.org/cgi-bin/cvename.cgi?name=2023-23916

Title: CVE - CVE-2023-23914

yuripv

Well I think it's not a big deal as pkg audit doesn't show any threat on 7.88

maybe port tree will be updated later

i just had curl update ( in ports )

oh I think I understood sorry

this report ony informs that curl can run as server and subject to threat

but doesn't indicate there is a threat actually

only*

Is thunderbolt working on freebsd

Midjak: that's not "exact security report"

i'm just wondering *why* it is shown to you, so wanted to see the exact text the ports show

oh yes sorry

servers and may therefore pose a remote security risk to the system.

This port has installed the following files which may act as network servers and may therefore pose a remote security risk to the system.

it's just a warning not a big deal

I was cheated by the extreme capitalization :-))

sorry for that

:)

Hi

which path of snapshot in zroot refer to OS settings and apps installed? I installed a wacom pen driver that I think made my pen not work. I only want to revert the software settings part. Not the database etc part

Can someone share /dev/MAKEDEV with me ?

it seems netflix and amazon prime won't run on freebsd?

ox1eef_: isn't that a linux thing?

Indeed, looks like 'mknod' is the fix in my case (deleted large parts of /dev by mistake). OpenBSD has /dev/MAKEDEV - not sure about Linux.

Letiute: Yeah, I think "widevine" is the issue there.

mason and there is no clean way out?

Letiute: Conceivably you could run Chrome in a Linuxulator, but that has reports of success that vary over time. A VM might be another option.

ok

my wacom pen/ table stopped working since followed this: freshports.org/x11-drivers/xf86-input-wacom

Title: FreshPorts -- x11-drivers/xf86-input-wacom: X.Org legacy Wacom tablet driver

xinput does not shows wacom pen

uhid2: <Wacom Co.,Ltd. CTL-672, class 0/0, rev 2.00/1.00, addr 1> on usbus0

 in dmesg

i never ha wacom devices but i think modern way to deal with them is with libinput, not legacy xorg drivers

pkg remove xf86-input-wacom

willr eboot

or how can i rever to a snapshot for this issue and not revert / change database files (postgres)?

Letiute: you can also copy the file change out of the snapshot

rtprio "copy teh file change"?

yeah, the snapshot lives in .zfs/snapshots/<snapshot-name> and you can treat it like a normal file

or read about zfs-rollback

rtprio ok will read

 which of the following I need to revert to: zroot, zroot/ROOT, zroot/tmp, zroot/usr, zroot/usr/home, zroot/usr/ports, zroot/usr/src, zroot/var, there are many

rtprio I thought those were meta data and pointers :)

and real files can't be extracted from snapshots unless restored

I didn't send/rec yet

they are just on my system

which one depends on what file you're tring to recover

no idea

if the file is in /var it will probably reset your postgrs

if it's the package you're trying to revert, why not just uninstall the package

rtprio I think, thigns messed when Iinstalled xf86-input-wacom

then uninstall that package, or did you already?

and did kldload cuse

and did  pw groupmod webcamd -m user1

also did  webcamd -d ugenx.y

I uninstalled few moments ago. Do I need to reboot rtprio?

wacom was not smooth before. Not its not working.

anyway rebooting

if the module doesn't unload, perhaps so. it would be the easiest to reboot

how to unload?

via terminal

kldunload

ok

rebooting

futile

as in the module still loaded?

and can it not be unloaded?

no idea

how can I check

kldstat

ok

don't see it though

that was the point of rebooting, yeah

still best to reboot

let see

I am trying to unlock my account when I lock screen with xscreensaver when I type my password it doesn't unlock my screen anybody know whats wrong

capslock?

its off

rtprio works like a charm after reboot. no idea what changed though

Letiute: if the module was not loaded it sounded like perhaps that was it

sn00p: there should be a log for what's wrong

rtprio speaking of snapshots, I guess the postgres data is in /var/db/postgres/...  so while making snapshots, I see zroot/var also. so if I restore it, pg will be affected.? so anything OS related also there in that dir?

rtprio can you also paste the path of file view/recover from snapshots (that was new forme)

when things like happen, how can I treat OS files snapshot and postgres files snapshot differently and separatly?

yes, you're corect about the /var snapshot

if you want it to be ... detached from /var you'd want to put /var/db/postgres on it's on partition

( stop postgres; mv /var/db/postgres to a different name; zfs create -o mountpoint=/var/db/postgres zroot/postgres; chown, copy data back in, restart postgres )

Letiute: os snapshots would primarly be ROOT and perhaps usr, not much else,

or set postgres_data to difrerent path in different dataset

and again, if i say, botched a file in /etc, i'd probably cp /.zfs/snapshots/mysnapshot/etc/rc.conf /etc/rc.conf and not restore the whole snapshot

rtprio otis I see

whats better, postgres_data var change or mount -o?

by the way, rare question, anyone you happen to change keymas of keyboard in freebsd. I used to have capslock and escape swapped and left alt & ctrl swapped

i don't understand

mount -o? you don't use /sbin/mount with zfs

rtprio I meant, whats better way of achieving the same thing: postgres_data var change in pg settings or mount -o?

my `moutn -0 ` i meant zfs create -o mountpoint=/var/db/postgres zroot/postgres;

you can create another file system and configure postgres to it, if you want. the way i posted preserves the default config which can be easier

last question: I am doing nvme raid. I have postgres and os, what compression level with zstd is ok if capacity is a concern but not the level where it gives huge performance issues

rtprio understood!

you might have to test some things, but i'd probably not use compression;

rtprio another filesystem? like it would be treated as a partition/drive by zfs?

rtprio ok

any idea who to contact to remove a bugreport from freebsd bugzilla?

remember, only new files are compressed when written if you change the compression setting

without -o mountpoint=/var/db/postgres it would appar as /postgres which i suppose you could configure in postgres, but it doesn't follow the common pattern

yes, I was wondering if there would be a way to recompress them

that might be worth some of your own investigation

ok

untitled, I think the most you can do is have it marked as spam, if it is spam. Otherwise, the bug reporter can close it.

freshbsd.org/freebsd/src/commit/1dc1f6bd313876 probably about time

Title: FreeBSD / src / 1dc1f6bd313876 - FreshBSD

No commits found in 4 milliseconds

Title: src - FreeBSD source tree

not sure why freshbsd failed

debdrup: freshbsd.org/freebsd/src/commit/1dc…6bd3138760a9e96e13017cc3c05e5e1b1e9

Title: FreeBSD / src / 1dc1f6bd3138760a9e96e13017cc3c05e5e1b1e9 - FreshBSD

it wants the full commit hash

i'm pretty sure it didn't used to want the full commit hash

git rev-parse --short produces short hashes that're shorter than what I stripped that hash down to (and it can get even shorter than that, but at that point there's a risk of collision)

debdrup: I think git gives you the shortest possible that won't result in that.

Oh, that's possible, yeah.

Currently it looks like it can make do with only 13 bytes.

I can assume an init. repo with one commit isn't the shortest possible but some something like src or ports its at the min. limt and you're getting the smallest without issues.

debdrup: oh look, Freaky just joined

meena: I asked him to :3

.o/

But if you're saving commits outside of git then you might have issues in the future. So make you want something longer.

Freaky: meena noticed that freshbsd needs full hashes now, is that correct?

I linked to freshbsd.org/freebsd/src/commit/1dc1f6bd313876 which doesn't work, but meena pointed out that freshbsd.org/freebsd/src/commit/1dc…6bd3138760a9e96e13017cc3c05e5e1b1e9 does.

Title: FreeBSD / src / 1dc1f6bd313876 - FreshBSD

Also, thank you for taking time for this silly nonsense. :P

I don't think partial hashes have ever worked

Hm, I thought I remembered having used them, but maybe I'm wrong?

though you can query them: freshbsd.org/freebsd/src?q=commit%3A1dc1f6bd313876*

Title: FreeBSD / src - FreshBSD

That makes sense, though, since it's using wildcards.

Well, a wildcard.

I wouldn't dream of asking you to implement gits rev-parse --short functionality in whatever language it's using, so it's probably fine to leave it as is.

Well, I guess it would be the reverse of that functionality, but it's still a way too big ask, I think.

were there some ... accidents in the source tree

git history eventually leads to a commit that looks somewhat out of place

rtprio: I'm not sure what you mean

There are probably unavoidable artfacts from the source having gone from cvs over svn to git. :P

i don't know if i could find it again

debdrup: I don't think it's an unreasonable request, shouldn't be too hard

didn't realize this was your site, Freaky

I think most of the community sites are made by people that're fairly active in the community, which is quite nice.

debdrup: links with short hashes are now converted into prefix commit hash searches

Freaky: that was quick!

only a couple of lines in the router

this power consumption thing I've been seeing whenever I reboot into FreeBSD has had me very confused

although I am also using hardendebsd which has some dev-friendly (as opposed to power-friendly) options which may impact matters

I pull like 0.7W more than I do under Linux. Is kern.hz still a thing in 2023, and if so, should I be setting it low?

Try it.

Doesn't seem like 0.7W is worth the extra time it makes everything take, but aside from kern.hz, FreeBSD is basically a soft-realtime OS already, so changing it doesn't hurt anything.

0.7W is probably also within the margin of error for any kind of testing you're doing, given how many lines of code are involved.

does kern.hz do anything by default now?

or is it just if kern.eventtimer.periodic=1

Those are two different things.

I'll set kern.hz to 100

(I wonder howit'd go if I set it to 10)

Try it.

You should probably investigate what it actually does.

Freaky: the periodic mode forces eventtimers to only use the hardware clock, which as you probably know can drift.

I thought it switched between ticking at hz and "tickless" mode with dynamic oneshot timers

Right, if it's set to 0, as is the default, FreeBSD uses a dynamic t ickless mode.

tickless*

I think there's varying efficiency for various timer coalescing algorithms in any event.

Freaky: using the periodic mode has it relying on the actual hardware though.

I think even with tickless mode, kern.hz=100 in hypervisor guests still has some effect, due to hypervisors not handling tickless kernels very well.

hz(9) has more information on it all, and I don't remember most of it I think. :(

There's also some link to timecounter(4), but I can't remember what that's all about.

rtprio: re: commits out of place; wasn't it you bisecting into some openzfs merge commit which contents looked completely out of place?

i might have asked about it before

but i'm having trouble finding it again

blah no xorg update.

haven't you heard, the cool kids are all wayland these days

doh none there either.

If I want to treat a `/dir` differently in terms of backups, compression, even mirroring etc, do I need to create a new pool for it?

What do you mean by "mirroring", as in live copy?

parv   `mirror` as in raid .

stripe, mirror, raidz1.2 are some features of zfs

hi folks :) in FreeBSD-13.1 do we have a kernel option to contol the buffer size for TTYs? I've already changed consmsgbuf_size but it seems not related. My problem is when in the terminal I do (cu -l /dev/ttyu1) and another terminal (printf '<\n\n>') I can only get like 382 characters exactly. There must be something which is blocking the next output. I already had the same issue in NetBSD and in there I changed (kern.tty.qsize = 4096) in the kernel

options and it wored properly. So I'm wondering if we have such an option here in FreeBSD 13.1

parv do I make sense?

Letiute: yes, for 'mirroring' no for the rest

Letiute, Backups can be made either via ZFS snapshots or via rsync on a directory of a dataset. Separate ZFS compression does require at least a separate dataset. ZFS mirror would mean starting with new pool layout. So all the things you mentioned could relate to each other.

you can zfs create zroot/blah and set different compression on /blah

but you can

can

can't make /blah a raidz2 when the rest is a mirror

i SEE

I see*

if you have a specific example / configuration it could help us advise you

what is required for /bla to have raidz2?

rtprio I will share  a tiny topology diagrame soon for the big picture but for now I was thinking to make /postgresdb1  and /postgresdb2 (2 databases living on different dirs via TABLESPACE), both having different compression levels

Letiute: at least four drives

rtprio the defintion of a drive , for zfs, is just a partition of physical drive?

putting postgres data onto it's on zpool is not a bad idea, but i think you might be expecting too much of compression

rtprio y ou mean its 'own' zpool?

Letiute: yes, own zpool

I am setting ztsd-3 for db1 and ztsd-7 or 9 for db2

it could be a partition but if you're going to make multiple pools from partitions you're going to take a performace hit and lose out on some feature of zfs

so I "can" make a new zpool for few directories?

Using 4 partitions of a disk for RAIDZ2 is pointless

Letiute: have you tested that compression with postgres is enough compresion to be worth the effort

what features of zfs will be lost in that case?

performance hit due to simultanious data acces on same physcial drive?

Letiute: you have some reading to do, as it seem you might misunderstand the topology of a pool and a filesystem

pool is a set of disks with a specific configuration (mirror, raidz1, etc) and can have many partitiosn (from zfs create)

rtprio no I have not tested. This would be the test

so it sounds like you'd need a box with 5 drives, at least one for the os and 4 for the raidz2 for postgres

but database purists would say to put postgres on mirrors instead

for now I have only one dsk

rtprio: simple example: pastebin.com/aSJmwYfk

Title: titan:yuri:/usr/src$ git branch --contains f04cb31e7c17* maintitan:yuri:/usr - Pastebin.com

(just look for any openzfs commits :)

rtprio as you said "you can zfs create zroot/blah and set different compression on /blah" so I have to create a new pool / dataset for /db1 /db2?

rtprio which reading doc would you redommend for zfs?

Do check the block size used by Postgresql in order to set the same for "recordsize" of the pool for performance, etc

I see

parv default value is 8192 bytes in pg. what is iin zfs?

128 kiB

oh. so ..

8kb vs 128k

yuripv: yep

Letiute: you're free to do what you want, but if you're testing compression on zfs you only need a single disk. to test performance of single disk vs raidz probably should be a different test

rtprio `zpool create pg1pool  /pg1` `zpool set compression=... pg1pool`

rtprio ok I will talk about raidz in a moment

no, it's zpool create zroot/pgpool1

no, it's zfs create zroot/pgpool1

damnit

rtprio parv jamboard.google.com/d/18hni1wEORDsO…f8y7Mjur4in1ecJMuM/viewer?pli=1&f=0

Title: Untitled Jam - Google Jamboard

maybe you should reread `man zfs` and `man zpool`

rtprio so I have to create pgpool1 "inside" zroot? I cant create an indepednat pool of pg1/2?

rtprio ok

what do you think of the drawing?

i don't have a web browser

so you have 2 2tb disks?

this manual* mirroring thing... i don't think you want that

rtprio nvme is 1tb each

and you definately don't want to mirror it from a striped 1tb

rtprio well i can keep it there plugged all the time too

mirror the 1tb for what you're doing

why is that?

1tb don't have the full data. 1tb raid0 canlt live without the other 1tb making it 2tb total

because raid0 is fragile for starters

its striped raid0

yes

fragie

and depending on how you're planning mirroring ... i just. it's just gross

could you backup 2tb of snapshots onto that drive? that would be great

you could find a file you deleted six months ago.

rtprio yes but mirroring will happening online/live

i also see you're planning on a raidz2 of ssd and hard drives... are those all the same size? i don't think that's goin to work, or it's going to work poorly

and I don't have to restore the snapshot if its already mirrored. I just plug and boot from hdd if nvme fails

Letiute: a zpool mirror will happen live

rtprio all radz2 are 1tb each

the speed of ssd will be just limited to the speed of hdd

thats all

right?

rtprio yes

you're paddling upstream of how zfs is generally accepted and while that might work for you it might not work well

zpool mirror will happen live.

and anyone you ask for help is going to have a wtf moment as you describe your topology

I see

also i don't think you can have a nested mirror like you're describing

which one?

you can't have a raid0 raid a member of a zpool mirror

again, you can do it how you want, but if i had this pile of drives on this list, i told you what i'd do

I see.

is #2 a usb or estata or what?

so #1,2 is not doable?

*esata

#2,4 are in hdd drive docker box (external)

navme is m.2

nvme*

other drives are sata

2.5"

you could manually maintain the mirror from 1 to 2 but that seems like a lot of work and #1 is still raid0 so i wouldn't trust it with any data

manually? rsync?

yes, something like that

so I can"not" make #1 to #2 mirror?

Letiute, that "mirror" looks like a backup than ZFS mirror

parv ya but technically a miror

can it not happen?

i'll say again, you can copy data to it, but it's not automaitic

so parv is right, that's a backup, not a mirror

but technically i cannot make a mirror like that?

or can I?

you can't

mirror nvme1,2 and then you are fast and redundant

why because the soruce is raid0?

you could put postgres data on ssd1,2 mirrored, that would also be fast and redundant

Letiute: yes, mirror members must be identical

so the cause of not being able to be mirrored by #1,2 is that #2 is not two drives?

its one

does it have the exact same disk geomerty? no it doesn't not, can't be mirroed

i suggest we move on so we don't debate this al day

ok..

i donno parv, how would you set this up

rtprio, Just as you have mentioned

For the need of 2TiB data on ZFS stripe, buy 2x 2 TiB SSD for ZFS mirror

2x 1TB drives you mean?

and at least 1 spare for all the disks

or even 2x 1tb hdd

s/all the disks/all the topologies/

2x 1TB drives you mean?   not 2x 2tb

so just for understanding, if I had 1tb + 1tb hdd, I could mirror #1?

the reason I can't is the disk geometry? and that the 2tb is a drive with 2 partitions

correct?

if you had 2 more nvme, made another raid0, maybe

you can only mirror the same thing

Letiute, If your data will fit in less than 1 TiB of space, then 2x 1TB for ZFS mirror will be enough

git it out of your head you can just make up blocks of data like linux fakeraid and everything will be o

k, real data on real filesystems need to be the same

s/1TB/1 TiB/ # for shame!

ok

rtprio why nvm raid has to mirror to another nvme? can't they be ssds on one mirror and nvme on other?

are they the exact same size? sectors?

no but I thought zfs would just take the least one

rtprio if this is the case, then I can't mix ssd and hdd in #3 as well?

i don't understand why you feel the need to do things the hard way

it might work, but no one is goign to want to help you with it

I see

you mean #3 might work?

you know you could also try these things and read the messages zpool tell you

ok

it might warn you that they're different sizes and might perform lie dogshit

if that's what you want to do, then perhaps i have better things to do now

got it

you mean #3 might work? or might not work?

might work

ok

why not try it and see how it goes

I think then, #2 might not even boot...

ok.

the hard way :)

but I should rethink

about your advice. my topology is off

#4 is ok?

a usb drive that you rsync to; sure can't see a problem with that

not even with -c in rsync?

checkum

checksum*

oh . I thought you mean "rsync cant see a problem with data corruption"

ok. so #4 is ok.

thank you rtprio parv

yes, rsync will happily sync your corrupted data to it's backup destination

which is why i suggested saving snapshots onto that 2tb drive

am.. ya but the source (zfs) should and would correct it

am ok..

what if you bungle a file? something crashes and eats the file.

rsync would just wisk that along and there's no corruption as far as zfs is concerned

this one, still?

understood. the only problem is restor time of snapshots. e.g if #1 fails, I have to formate, reinstall freebsd, the restore snapshort from 2tb

AmyMalik evolving proces :)

not the way i'm describing you should do it

rtprio snaps make sense

just boot into the other nvme and update bios order. order a replacement drive. no downtime

rtprio how are you proposing ?

zroot mirror on nvme1,2

postgres mirror on ssd1,2

zfs-send snapshots to the 2tb drive

rtprio wait. if #1 crashes, I can't make mirrors/plug play

the only option I would have is to plug a new nvme, install freebsd, restore snapshot

why do you thin that

correct?

why do you think that?

no you'd just boot off the other nvme, that's the whole point of a mirror

and then swap it out, and rebuild the mirror, re-redundant again

which mirror? we are putting snapshots. didn't made a mirror

always this one, apparently.

two distinct strategies

This is going in circles; meow out

yeah man, have you done any reading about zfs before coming in here with the worst desgined zfs layout?

rtprio  yes since 2 days. Are  we on same page that #1 is raid 0 and i just save snapshots to another hdd. If #1 crashes, what will I swap? which mirror?

let it be the last point for today :)