I can send you the exact confiuration

Anyway, the first thing would be the capsicumize the base system ... that would be a step

I would love to help

if a pledged application tries a system call, it is not allowed to, it will be aborted

but the idea behind unveil was that an attacker can not see the difference between an unveil and an unexisting file, therefore ENOEnT

poke through src/usr.bin and see which ones (a) don't already link to libcasper (see Makefile) and (b) look like possible candidates for it

@ RhodiumToad I will do

Though I think the daemons are appropriate candidates ... but maybe that's just what I have been doing for too long

usr.bin is fine

what is the general philosophy ? trying to keep with Linux ?

philosophy of what exactly?

well, one thing I like about OpenBSD is the idea of radical simplicity

have a look at the echo.c at FreeBSD ... is this a show off of C11 ? NetBSD and OpenBSD keep it with a simple iteration over argv

this is what I sometimes don't understand

in the source code

what in there is C11?

looking at the history, it seems that once upon a time someone decided that echo could be made smaller by not using stdio (back when /bin could not be dynamic-linked this was more important, and still matters a bit for /rescue),

and then it was changed to use writev() to restore the previous performance

(we're talking 2002-2003 here)

once upon time ... I would like to meet Ken Thompaon ... but it will never happen ,,,

oh, writev() was 99 ... sorry, I get old

writev has been in BSD since... about forever?

since 4.2BSD anyway

and therefore since 1983

I remember a Schneider CPC128 ... you didn't need floppies to boot

1983 was UNIX very far away from me, I was 8

I first used unix in about 1988

and freebsd in 1997

writev is that old ... ok ... forgive the young guy

My first Linux was in 1996, the X configuration shredded my gf's monitor

looks like writev was added to the spec in XPG4.2

hahaha

ah, the days of writing your own modelines for X

I am not good in history ... I still think it's overkill for echo

using writev?

oh yes, the modelines

looping over write() uses a lot of system cpu for echoing many arguments

echo can be written in still plain 89 C in 20 lines ...

hey, it was just an example .. the simplest I could think of

looping over puts() doesn't use much system cpu but, as I said, back in the days when /bin had to be static linked it bloats the binary quite a bit

arguably it could be changed back to using stdio now that it's dynamic linked

FreeBSD tends to have bloated coat ... but hey it's Friday

I don't care much about the 1970s

UNIX survived becaused it was simple and therefore adoptable

Greetings VMS ... rest in peace ... I love you

I will go through usr.bin tomorrow

Good night everyone

anyone know any reason why a usb thumb drive, unused but plugged in for some weeks/months, would decide to become extremely slow to read?

as in ~3MB/sec for a usb3 device on a usb3 port

I'm pretty sure it was not this slow originally

RhodiumToad: electrons caught in NAND cells become tired and need rest ?

that is _almost_ a thing :-)

flash cells do need to be written again after a certain number of reads

but (a) nothing was reading the drive and (b) I just completed a full read of all blocks and it's _still_ slow

is writing slow as well ?

haven't tried

but writing is usually pretty slow anyway

3MB/s is way too slow

indeed

esp. for a usb3 device

good old SD/MMC class 10 does 15MB/s on regular 4 bit socket

did you try reading it in different USB port ?

yes, no change

also if I stick a usb3 sdcard reader in the same port I can get 80MB/sec reads

sec, I should have another thumbdrive of the same type, unused since purchase - lemme find it

as it was long time powered, perhaps a voltage regulator got broken which leads to higher error rate

hence more retransmits/re-reads

ok, unused device of the same type is giving ~95MB/s

so it's definitely not the port :-)

whew!

no errors are being shown at the USB bus level, so if there are errors/retries then it's inside the device

interesting. write speed when writing zeros to the dodgy device is very high

oh wait, scratch that, it just dropped

95MB/s? I don't get but 55MB/s or so

it'll depend a lot on the device I suspect

fastest usb3 drive I have does >220MB/s

i've hit 200MB/s or so but it doesn't sustain

that's good

220 *whistles*

Anyone know how to set what IP a jail uses as a source address when using iocage?

I've got two external IPs on a jail - one "fixed", the other a CARP IP, which may or may not be up depending on the status of the other CARP member on another host.

vnet jail?

The jail insists on sending traffic out the CARP interface even when it's in "BACKUP" mode.

Nope, old-fashioned non-VNET.

VNET is my next step if I can't figure this out, but I've never used VNET in production, so wasn't crazy about testing it out.

what do ip4.addr and ip4.saddrsel end up set to on the jail?

I've tried changing the order in the "ip4_addr" parameter, and tried "ip4_saddrsel" set to "0" and "1" with the same result.

Testing at home this was fine, but I also had a simpler setup with only a single interface - the actual hosts have an internal and external NIC which seems to play some role in this not quite working.

is the CARP address uses only for a service running in this jail, and not for anything else on the host?

Actually, I take that back about "ip4_saddrsel" - with it set to "0", I can reach the internal network, the external only if the CARP IP is MASTER, and if I set "ip4_saddrsel" to "1" I lose connectivity to the internal network (which is where my DNS resides).

Correct only on the jail.

And I may have a need to move it from host to host in the future.

vnet's probably the way to go then

Yeah, that would make it even more "portable" in that I wouldn't have to have any CARP junk in rc.conf.

ALthough I'm not finding confirmation CARP on VNET is actually supported.

I'm really puzzled on what setting "ip4_saddrsel" does to break the internal network.

The jail (and iocage) manpages are not super clear on what exactly "ip4_saddrsel" - they both use language that references some other part of the manpage that I'm not seeing.

But I feel like what it might be doing is forcing all the internal net traffic out the primary (external).

Yep, pinging something in the internal net sources the ping from the external.

remember that source address selection doesn't necessarily control what interface is used, just what the source address in the packet is

Yeah, I'm seeing that with tcpdump - source from an external IP, but going out the internal net interface.

if source address selection isn't forced, then it should normally match the interface

Just wish there was a way to specify (with "ip4_saddrsel" set to "0"), which IP, by default, traffic is sourced from inside the jail.

since the default is to try and route the packet to its destination, and pick the outgoing interface's primary address as the source address

Config order doesn't do it.

the order of ip4.addrs matters only when normal source address selection fails to match one of the IPs that the jail is allowed to use

And I have it backwards up there, I meant "(with "ip4_saddrsel" set to "1")" not "0".

vnet is at least potentially simpler to understand since the jail has its own network stack to itself

This kind of puzzles me:

[root@clwebX /home/spork]# iocage get ip4_addr haproxy02

ext0|1.2.3.247,1.2.3.248,int0|10.99.88.241/32

I would think 1.2.3.247 would be the default source address, but in any order, it always sources from 1.2.3.248 (which is the CARP IP). Just odd, seems like it should be configurable in some way.

I do not use the "interface|IP" form for the CARP IP because then I end up with the an extra alias for the CARP IP.

(manpage says specifying an interface forces creating an alias)

the one thing I'm unsure of is whether epair supports carp, it's possible it does not

Anyhow, just wanted to check if I was missing anything obvious before I move to a completely new and different setup. :)

I'm about to find out I think!

This just does not seem to be true:

"If a jail requests a specific IP address, the host attempts to add that address as a new alias. All traffic sourced from the jail uses the first IP assigned as the source IP."

I'd _guess_ it should because the common ethernet code handles it, and I think epair uses that

But that's MWL and not the manpage...

Maybe at some point I'm just going to end up using bhyve. :)

I think that "first assigned IP is the source" really translates to "first IP *configured by the jail/iocage* is the source *if* no existing IPs are used by the jail" - meaning the CARP IP is already there, which in some sense makes it "first".

I could also not assign the CARP IP to the jail and instead let the host's devd add/remove it with CARP state changes...

Anyhow, thanks so much for talking me through this mess!

fascinating, the dodgy thumbdrive is sustaining a 20MB/s write speed (which is what it's rated for)

RhodiumToad: 👍

RhodiumToad: s/👍/*Thumbs up*/

:)

and having zeroed the whole drive, I now get ~130MB/s read speed

dang, that's very good

would be for me anyways

anyone want to help me fix the scipy port?

i'm new to porting but i'd like to help. have room for a noob?

how is it broken?

it was the last time I installed it, trying again, to see if it still is, hold please?

WARNING: Ignoring invalid distribution ~equests (/mnt/extra/ec2-user/.virtualenvs/units/env/lib/python3.11/site-packages)

WARNING: Ignoring invalid distribution ~equests (/mnt/extra/ec2-user/.virtualenvs/units/env/lib/python3.11/site-packages)

those are the first lines when I try doing a pip install -e from  the extracted make patch directory

pythonpool.com/solving-python-ignoring-invalid-distribution basically says to delete the "~package name>" dirs

Title: Solving Python Ignoring Invalid Distribution - Python Pool

ok, pythran reinstalled without issue

^5

Whats the best way to turn off drives power to avoid them from spinning?

depends on the drive

RhodiumToad pastebin.mozilla.org/EFA3i7VM

Title: Mozilla Community Pastebin/EFA3i7VM (Plain Code)

you can probably set the standby timer using smartctl

Can I unmount the drives and use camcontrol standby adaX to turn off the spindle on HDDs and camcontrol idle adaX to spin it back up.

RhodiumToad ^

or I need `powerd`?

do you particularly need to unmount them?

(rather than just having them spin down when idle)

those are backup drives. I don't want to keep them running except a bi weekly backup time

RhodiumToad I would prefer full power off. But as those are on sata cable, I can't just plug them off

updated pastebin.mozilla.org/AH4Ktr2i#L1

Title: Mozilla Community Pastebin/AH4Ktr2i (Plain Code)

but I may want them to reuse them. so what are my options

if they're not mounted, then by all means try camcontrol standby

RhodiumToad so umount, standy and mount again to use? thats all

should do

however, rebooting the machine will probably spin them up again even if not mounted

RhodiumToad my/all hdd support this? I don't need EPC or whatever? also i don't need powed?

you don't need powerd. not all hdd support it, try it and see if it works

ok. I usually do this : gelli attach /dev/adaX   <- this i the point where it is mounted?

Current power state: PM0:Active or PM1:Idle(0xff)  <---- its onn, as you said, even not mounted

that doesn't mount it, that just attaches it for decryption, but that will access the drive so it should spin up

I am doing standby to check

then maybe `zpool import ..` will

then maybe `zpool import ..` will mount it

How to check the device state after standby?

that it was really on standby now

what does it report now for current power state?

camcontrol epc ada3 -c status  -> Current power state: PM0:Active or PM1:Idle(0xff)

same as before. nothing changed

well it might not support it then

despite standby

EPC: Supported, NOT Enabled

Low Power Standby NOT Supported

Set EPC Power Source NOT Supported

APM: NOT Supported, NOT Enabled

RhodiumToad I am doing camcontrol epc ada3 -c enable

Now, EPC: Supported, Enabled

its still. Current power state: PM0:Active or PM1:Idle(0xff)

when you query the status, use camcontrol epc ada3 -c status -P

same

Current power state: PM0:Active or PM1:Idle(0xff)

camcontrol epc ada3 -c goto -D -p Standby_z

then camcontrol epc ada3 -c status -P

that worked

Current power state: Standby_z(0x00)

I wonder

Why it says active if I don't use -P

camcontrol idle ada3 , camcontrol epc ada3 -c status -P

Current power state: Idle_a(0x81)

some commands will bring it out of standby

without -P, that status option issues several commands

so I need EPC: Supported for any of this to work?

yes

no other option?

so without -P , the other commadns make it active again?

Title: In file included from scipy/stats/_stats_pythran.cpp:34: In file inc - Pastebin.com

that's the error pip install encounters from the port

incidentally, an install into the linuxulator works without problem

(using ubuntu)

hd1: what are you doing exactly?

still messing with the scipy port

no, what exact commands are you doing

pip install -e . from the work-py311/scipy-1.10.1 directory

why are you doing that?

because make install installs the package globally and I want it only in this vitrualenv

RhodiumToad confusion. as per serverfault.com/questions/1047331/h…pin-down-or-head-parking-in-freebsd , `camcontrol epc ada3 -c state -d -p standby_z` is used to disable spin-down (Standby_z) via EPC?

Title: How do I disable hard disk spin down or head parking in FreeBSD? - Server Fault

but you don't want to disable it, do you

no

but the command says that it will disable it

hd1: that's not the port's problem. the port exists to build a package which you can install on your system

but actually it PUT it on standby <-- desired

hence the confusion

hd1: if you want to do something else, you're on your own

Beladona: that command is not the one I told you to use

goto != state, -d != -D

oh I missed that

thanks, i think that is all I need

(I guess)

No need for head parking either

(I guess)

probably ok if you don't plan on dropping your computer

rtprio so power outage will hurt?

How to know in depth, where is the disk consumed? which files take most space

maybe ncdu

du -k | sort -rn

Beladona: the heads won't hit the discs in that case

RhodiumToad du -h -depth=2 /* | sort -rhb > /tmp/disk.txt ?

no

I see

rtprio on reboot, they will. I mean no danger in doing this standby/idle thing in power managment?

it's in all respects safer than having the drives running

RhodiumToad you mean, having it on standby / idle is better and safer?

yes

ok

Error loading module '/usr/local/lib/samba4/modules/vfs/fruit.so': /usr/local/lib/samba4/private/libsmbd-base-samba4.so: version SAMBA_4.16.11_SAMBA4 required by /usr/local/lib/samba4/modules/vfs/fruit.so not found

both of those files exist

huh restarting samba and it's working now

let me guess, you upgraded it while it was running

it wasn't complaining that the file didn't exist, it was complaining that it didn't define the symbol version that it was expecting

very possible

... fascinating, having overwritten this thumbdrive with random data, it's now performing at full speed for reads, >30x faster than before

RhodiumToad: would this be something that's usb related or filesystem related?

can't be either

mostly I was testing with dd, bypassing the filesystem

no UAS(P)

?

I'm "complaining" about the lack of UAS(P) support in FreeBSD

Title: USB Attached SCSI (UAS/UASP) support and fallback to BOT

for this example, that's not really relevant

for this example, perhaps not, I don't know what this example is

ludicrously slow reads from a thumbdrive, but it's now really fast again after overwriting it all

ah, well, without much thought I'd be quick to blame the firmware

Saved by randomness

Any ideas how my zfs on root resized before my eyes? it was 4.9 of 4.9 , then when I'm struggling to free some space 4.9 out of 16

it's very easy to forget about space used by snapshots

yes I removed many. but I'm confused why it would change the available space? the snapshots were also only hundreds of megs at most

then it suddenly became 4.9 out of 16. some housekeeping that happened many minutes later?

hello,I met trap about the freebsd install mountroot error 19 problem. Even in usb mode or isoemu mode to load freebsd13.2-amd64-bootonly.iso.

Unable to understand why I got rebooted when I tried to resume screen off by mouse ( I did this few hours ago `sleep 1; xset -display :0.0 dpms force off`)  /var/log/messages: pastebin.mozilla.org/1vTsbXLn

Title: Mozilla Community Pastebin/1vTsbXLn (Plain Code)

I'm having a hell of a time installing alacritty.info on FreeBSD

I can't get it to set terminal capabilities correctly

the weirdest part is that installing alacritty on the freebsd host doesn't even seem to allow TERM to be set to `alacritty`

damn,I create a new usb stick with ventroy, debian could install. but freebsd13.2 still enter mountroot error 19 problem.`

Disconnected

Unable to understand why I got rebooted when I tried to resume screen off by mouse ( I did this few hours ago `sleep 1; xset -display :0.0 dpms force off`) /var/log/messages: pastebin.mozilla.org/1vTsbXLn

Title: Mozilla Community Pastebin/1vTsbXLn (Plain Code)

OK,I think it's a bug of freebsd13.2.iso. I just try ventroy+freebsd12.4-bootonly.iso, it cloud install normally.

redlegion: how are you installing it?

kyonsalt: use the .img for usb sticks; what are you installing on?

qmr: yes, i believe snapshot deletes happen in the background

How can I know who triggered this boot? pastebin.mozilla.org/1vTsbXLn#L90,91,92,93

Title: Mozilla Community Pastebin/1vTsbXLn (Plain Code)

a crash would be my guess

anyone know if much change is coming to bhyve sr-iov with 14? i can consistently trigger a crash when i start the install process for freebsd in bhyve using a chelsio T520 passed through pci device. the whole card will freeze about 2-3 mins. any attempt to reset the device triggers a panic.

it's not critical. just wondering if i should report a bug, or wait and see in a few weeks.

markmcb: Several of us have been tracking this on the bhyve call, notably John D and Santiago M. What is the key PR that is impacting you?

michaeldexter: PR=pull request? i'm running on 13.2 RELEASE. Haven't tested at all with changes in STABLE or CURRENT.

Problem Report.

ah, sorry. i haven't reported it yet. i can browse and look for a match.

i assume bugs.freebsd.org is the place to look?

Do look at 234073 271456 270966 and 266325

thanks, will do

And possibly 269133 269133 263046. Thank you! Do report back and share any new information you can in the tickets.

234073 for sure is still an issue. i'm the last comment on that one. that was on an X710-DA2, which seemed quirky in general. i've replaced it with a Chelsio T520.

the rest don't quite match. i can passthrough the device and load the driver for it. shortly after though i'll experience the freeze, which kills the whole card, i.e., event the host loses networking.

*even

markmcb: Is the Chelsio giving you issues too?

Yes

the freeze i was describing is with the Chelsio

Do mention that in the ticket if it’s not there already.

I use Chelsio cards passed through into bhyve VMs for almost all cxgbe(4) development.

Are you using SR-IOV?

Are you passing through SR-IOV devices or the main PF (physical function)

no I'm not using SR-IOV

I know of a problem in 13.2 that seems to be fixed in 14-Stable

T520-SO-CR, have 4 SR-IOV VFs set to passthough

markmcb: Have you tested on 14?

what is the exact failure? a bhyve hang or something else?

SR-IOV works without issues in jails and the VFs not set to passthrough in iovctl.conf

michaeldexter: not yet, was waiting for the first version i can get to with freebsd-update

file a bug if you see the problem with 14

and I'll fix it before release

(if it's in the driver)

npn: i'd have to trigger it again for specifics. i'll do that today and capture it all in a PR

markmcb: Do you have hands-on access to the hardware or is it in a distant colo?

hands on, it's in a rack in my garage

markmcb: For quick tests, I use this script, which can dd a VM-IMAGE on a physical device, and boot it with a USB to SATA adapter: github.com/michaeldexter/occambsd/blob/main/imagine.sh

Title: occambsd/imagine.sh at main · michaeldexter/occambsd · GitHub

Ping me if you have any questions though I am not always on IRC.

ok, thanks

NP

I had a report of the 13.2 SR-IOV cxgbev driver failing to attach on a hyper-v host (no bhyve there). I could see PCIe problems (misconfigured BARs) even before an attempt to load the driver. I assumed there was a bug in the pcie code in 13.2 and tried 14.0-alpha1, which worked., and I dropped the debug.

take a look at pciconf -lvbc before you load the driver (in case you see miconfigured BARs too)

Noted

npn: you check on the host or guest? i'm not sure what to look for to identify a misconfigured BAR. i see bar line items on the host, but not sure what they should/shouldn't be

look in the guest. In my case one of the BARs had an address with all ffs in the upper bits and that's not a valid phys addr, and the BARs were not enabled either.

boot the guest in verbose mode and the pcie drivers will print this info too when they enumerate the devices

ok, i'll give that a look. thanks.

What are my options for a command included in base, that supports libxo ideally, to see which PIDs have open handle(s) to a given file? (specifically in my case, I'm looking to add a rule to devd that sends SIGHUP to any PIDs holding open /dev/dsp6 when my headphones become unplugged)

I know there's fstat, but from what I can tell, it doesn't support libxo, which means awk/cut/whatever to hack the output text.

My USB headphones come unplugged every once in a while and Firefox doesn't release /dev/dsp6 when there are tabs open with an audio player (even if paused). So I figure I'll add a short script that's triggered by devd for the headphones to get a list of PIDs holding it open and send SIGHUP so that it gets released. When the headphones are plugged

back in, they don't show up as a dev device until dsp6 is released (and I don't want to just tell it to create a new dspX device and set it as the default).

Anyone know of a tool that would let me spoof a response to a broadcast packet to then reply with an address on a different subnet? I have a device that only uses broadcast packets to find the server to connect to, and doesn't permit direct ip entry in the client. I'd like to have the server on a different subnet, but the client can't find it because it will only find it via broadcast.

Erhard > broadcast L2 feature | cannot leave the subnet it is on

Yeah, that's the issue. I think I may need to listen for the broadcast and sniff the protocol and reply with the correct ip

The client is closed source and they refuse to allow direct ip entry because they think it might confuse users (idiots)

But my wifi is on a separate subnet and I would like to connect from it...

If you are in a different subnet, there should be a routing once in a while.

Not sure I know what that means. There is routing, and that is via a FreeBSD box.

BUt the clinet owuld be on wifi, which is a different subnet from the LAN where the server is.

192.168.1.0/24 wifi ; 192.168.2.0/24 cllient = L3 routing

Yeah. I have that setup... but the broadcast bit is layer 2

Maybe I can do an IP alias.. I guess I will do some more research

Sorry, I don't fully understand what you want to do. I can't help.

Erhard: do you need a fully fledged tool, or will a framework that's done some heavy lifting be enough?

(albeit you have to know how to use the framework, but still)

It's OK. I will do some more research. What I want is client on 192.168.0.0/24 to connect to server on 192.168.1.0/24, but the client will ONLY locate the server via broadcast, and does not permit entry of a server IP.

It's a one off for this one client/server.

Can be hacky.

Erhard: If you're familiar with python, I'm 99% sure the python scapy module can do what you want.

If I can sniff the ptotocol the client uses to sommunicate with the server via broadcast then perhaps I can mimic that with python

scapy? OK. I will look at that, thanks.

This looks good. Thanks. bbl

michaeldexter: following up from earlier, I added 273372

npn: see above message

Hey, new to freebsd I have issues with pkg's network speed I tried using mirrors but fastest I get is 400kb/s

I tried to post the issue on freebsd forums before but coudn't find a solution

Who do you have to poke to get zfs-get to list props in alpha sort? All things zfs are now an upstream project?

sys/contrib/openzfs makes me thing it is

I see "ALPHA" referenced in the freebsd-update man page. is that a normally available option? looking at the 14 schedule, i only see BETA referenced.

markmcb: I think it's an artefact of the manual pages being built on 14-CURRENT, which is technically speaking 14.0-ALPHA2 right now (if it hasn't moved onto ALPHA3)

skered: github.com/openzfs/zfs