So...I'm trying to run some Linux applications and I have a Ubuntu system setup in Linuxulator but I'm still having issues where I get this error:

ELF interpreter /lib64/ld-linux-x86-64.so.2 not found, error 2

I ran "brandelf -t Linux *linux-bin*" but I still get the same error. I have this in my Linux path on sysctl: compat.linux.emul_path: /compat/ubuntu so I have no clue why I have to get into my chroot on my Linuxulator to even execute Linux binaries.

Stupid PoE switch I bought a couple of months ago looks like it just went to the big LAN in the sky

BillyJoeBob: and presumably /compat/ubuntu/lib64/ld-linux-x86-64.so.2 exists?

It does. ls -lah shows root:wheel 42B /compat/ubuntu/lib64/ld-linux-x86-64.so.2 -> /lib/x86_64-linux-gnu/ld-linux-x86-64.so.2

And I can run everything just fine if I'm chrooted in the Linuxulator.

/usr/local/etc/rc.d/ncidd: Permission denied.

i'm runnin g this as root. how can one investigate this?

you're not supposed to run rc.d scripts directly

oh, that's right

how can i add it as a service?

that said, are the permissions on it correct?

just the presence of the file makes it a service

i'm not sure. this is part of the ncid package

ls -l /usr/local/etc/rc.d/ncidd

-rw-r--r-- 1 root wheel

RhodiumToad: maybe, but they (nearly always) work when run directly

service ncidd start does not start it, but it looks like it's not executable

CCFL_Man: did you set ncidd_enable=yes in rc.conf or 'service ncidd enable' ?

oh

now it starts

well, /usr/local/etc/rc.d/ncidd: WARNING: failed to start ncidd

how can i see why it did not start?

does ncidd have a log?

it seems to exit without writing any logs

have you properly set a config file?

i think that might be it

i'll have to refer to ncid support. the program just exist and prints nothing

rc.d scripts should be executable. Look at the other scripts for example: ls -l /etc/rc.d/

does it have any debug option?

rwp: it might be the package

RhodiumToad: good thinking. it's the config file

Just for the sake of discussion... What is ncidd?

rwp: it's a network caller id server that passed caller id information from a modem over a network socket to clients

yeah, i see it here in comms/ncid... but what uses it?

will ... xbmc/kodi use it and show who's calling?

How was it installed? pkg binary package or source port compile install?

It bothers me that /usr/local/etc/rc.d/ncidd after install was mode -rw-r--r-- which implies to me that something is broken from the start.

rwp: the port doesn't have a maintainer, maybe you should step up and fix that :P

rtprio: i primarily wanted to use to to log caller id data to a file

yes, pkg install ncid

For one I had never heard of this before a few minutes ago. For another I had previously written my own caller id thingie using perl-expect and voiced with festival. And finally, who has a house phone that would make use of this anymore? I mean besides me. But even I have the ring bell taped off. It's just a honeypot now.

ok, it required /dev/ttyU1 as the device, but it won't print caller id information

i have a house phone. i just had one installed, actually

My house phone is now VOIP through my ISP but I did keep the number. Version ncid-1.11_1 from packages?

yep

I am playing with jails tonight. I'm going to eat something first but then I will set up a jail and install this and poke at it with a stick.

do you have a pstn modem you can use with your provider voip?

the modem i have does not return caller id

i don't know if it's the wrong init string or not

RhodiumToad: thank you again for the help!

i should look at the windows driver ini files for the init strings

the product page for this modem has someone asking why the default init string does not activate caller id

i haven't had a pstn modem in a long time

HKR, EnableCallerID,1,,"at+vcid=1<cr>"

that's in the winders driver

yep, that's in all the winders drivers

I installed it. Poked at it with a stick. And I think it definitely needs some work. For one it comes with 16 startup scripts!

Obviously not all of them are needed. None of them are installed executable.

The /usr/local/share/doc/ncid/INSTALL-FreeBSD.md and /usr/local/share/doc/ncid/README-FreeBSD files have some information about installing it manually.

Seems like the port does not quite finish all of the needed bits.

rwp: there is no maintainer

I think you are doing more work on it now than has been done in a while. Just say'in...

so for this modem, i might have to set the country code first

another user used a different baud rate to get caller id working

Can you talk to the modem yourself? Does it respond normally?

For my modem I would get its attention with sleep 1; +++; sleep 1; then send an ATZ and wait for the OK, then send it AT#CID=1\r to enable caller id.

That's all historical. I don't have a caller id modem on my phone line since it is now VOIP.

I don't know if the modem would work or not. Probably the modem has died in the years since it was last powered up.

i can talk to the modem's tty and send it commands manually

it does respond and returns ok when i send it the caller id string

Call your modem while talking to it. It should send you the caller id string.

it doesn't return anything

well, no caller ID, but it sees the ring

so it sees the ring, but no caller id information

but right, it creates lots of startup scripts that are not executable

broken package

So two problems. The modem is not (as of yet) emitting the caller id information. The package is not working.

But nothing can work unless the modem will produce the data so that's the place to start.

ncidd -Dv3 -C /usr/local/etc/ncid/ncidd.conf allows you to start in debug mode. you can see it exchanging with the modem and also print status output

the application runs fine, but this specific model is not emitting caller id data

i have ncid running on my tivo using the tivo

i have ncid running on my tivo using the tivo's dial up modem and it works quite well

Modem Identifier: CX93001-EIS_V0.2013-V92

You have a TiVo with a dial-up modem? That's pretty cool. I don't think either of my old beloved TiVo models had a phone modem on them.

the Tivo HD does. i hacked it and installed ncid on it and the built in dial up modem works great with caller id

CCFL_Man, Respect!

why doesn't tzsetup work?

Select local or UTC (Greenwhich Mean Time) clock

I tried both Yes and No

Does the abbreviation `EEST` look reasonable? I press Yes here!

# date

Mon Oct 16 05:40:07 EEST 2023

but why on planet it is not 08:40:07? how it would be in EEST here

m

bapt: ?

TIL about "m" urbandictionary.com/define.php?term=m

Title: Urban Dictionary: m

V-T60, The timezone names are ambiguous in some cases. It is better to always use an unambiguous format such as "date -R" provides.

V-T60, Also my previous question. Is "date -uR" correct for UTC time?

release candidates are based off the releng branch, correct? i.e. i can update to -RC1 via freebsd-update?

rwp: thanks for the quip about dpms tho. added 'xset -dpms' to my .xinitrc, that should fix things for now. gonna investigate and see about where to report this. far as i can tell asus probably is the culprit here, but there still is the memory leak in the amdgpu driver, so I'm not yet ruling out that having something to do with it.

going on with a completely different topic: sector sizes / alignment – can somebody give me a summary? i vaguely remember that 4k was recommended under… some conditions?

but i don't recall what the specifics where, i think either ssds or zfs, but it's been a long time.

background is i currently have my system on mechanical drives, but recently added 2 nvme drives to my desktop. i've noticed that my performance on these mechanical drives is quite awful (using defaults, so i assume an alignment of 512) and am wondering what to look out for to get the most performance out of both platter disks and ssds.

other pertinent info: running on ufs, planning to reinstall on the nvme drives with gmirror for full system redundancy.

yup, beta & rc are based on releng branches, and you can download them using freebsd-update

babz: great. then i'm gonna update to RC1 as soon as i moved to the nvme drives.

mhh, testing with unencrypted ufs, it seems default, 512 and 4k alignments don't make much if any difference on the nvme drives…

tho the performance is only about 20% of what's advertised (~530MB/s vs 2.6GB/s) – is that expected?

is the disk trimmed ?

do you have enough pcie lanes to support 2 nvme drives?

babz: i don't even know what that means. bradd i think so, i mean it has to m.2 slots. second one is slower from what i've read, but haven't tested that one yet.

pre-erase the blocks

come to think of it, maybe /dev/urandom might've been the bottleneck in my test… i don't think it's often used to pull 10 gigabytes from?^^

babz: what does that mean? dd if=/dev/zero of=/dev/nda0?

erasing a block is much lower than writing a page (what manufacturer advertises as "write thoughtput")

s/lower/slower/

ah, so i do that dd of /dev/zero and see if the speed there aligns more with what's advertised?

also, i read that nvme has some weird hybrid sata mode. when I'm using nda, does that already imply i'm *not* using the sata mode?

mhh, yeah, seems like urandom was actually the bottleneck. not used to that.^^

bradd: thanks for pointing out trim, not sure if it matters for performance but sounds like it's important for ssd longevity. :)

tho i'm not sure if/how this will translate to things once geli is in play…

last time I checked, it didn't pass throughout the geli layer

so using geli will by default result in more wear for the ssd?

I don't think it will impact wear that much

(Unless the ssd's firmware is plain stupid)

good^^

But you'll have faster writes with trim

anyhow, creating my 10G file in /tmp first and then cp'ing that to the test partitions has it finish in ~2.4s, independent of alignment used. that would be way above the advertised speed…

mhh yeah, with sync after it, it's closer to 5s, tho i had to measure that externally because i can't figure out time(1) ^^

will geli authentication make use of hardware acceleration if it is available? is hw accel for sha even a thing freebsd supports?

phryk > aesni module enable

if you support cpu hw support aesni is working ...

IRRC

vxwarlock: but… isn't that specific to… actually AESNI? not the SSE/AVX acceleration for SHA?

Title: aesni

ah, nice.

and this time, i'm putting geli on the mirrors instead of creating mirrors of multiple .eli devices. hoping this will be more performant. :)

I am a bit confused about the -j, -J and -U flags for newfs. They all kind of sound like they do the same thing, especially -j and -U…

soft updates and <whatever> journaling are different things

I personally have had only bad experience with journaling

the man page literally calls it "soft updates journaling" tho…

Yeah, but the important part is the "journaling" part

so I should only use -U and that already gives me the faster fsck?

And since it's built on soft updates mechanism it's called "soft updates journaning"

mhh, no that seems wrong…

Not exacly, it just enables fsck to run in background, it still takes it's time though

i *think* i understand now. fsck only works faster with journaling, and -j and -J are two different implementations for journaling, with -j building on top of (and possibly implying) -U?

ah, okay background fsck is another nice thing.

so -U alone will already have me boot faster because of background fsck in case of a previous crash, but with -j *xor* -J the fsck (whether in foreground or background) will finish faster, too.

so I *definitely* want -U, possibly -j but should probably ignore -J, does that sound about right?

is it expected that using HMAC/SHA-256 authentication with geli literally halves the capacity?

nvm, it's a sector size thing.

(i.e. the fix is supplying "-s 4096" to geli init)

OpenZFS Developer Summit 2023 is starting now, broadcasting live at: youtu.be/bvTVPwofvLg

Title: OpenZFS Developer Summit 2023 - YouTube

fairly obscure question, but does FreeBSD have support for NVMe Reservations? (I know that SCSI-3 Persistent Reservations support exists for iSCSI as an example here, but wanted to know about the 'updated' version that goes over NVMe)

use case being SAN-style storage failover for certain high availability workloads like databases

oh gosh, how silly, It's been implemented there with no issue

(since 12.1)

I'm trying out gelis -g ("geliboot") feature, but am having trouble getting it to boot.

I have a gmirror that's encrypted passphrase-only (i.e. no keyfile), has the geliboot flag set and contains /boot. I have an efi partition, copied loader.efi to EFI/BOOT/BOOTX64.EFI on that and I registered that loader with the uefi or whatever it is `efibootmgr -a -c -l <path/to/BOOTX64.EFI>` does. the efi loader itself starts fine, but gives an error that it can't find any bootable partitions.

phyber: need -b as well, iirc

kevans: i'm somewhat sure i already tried that 15 minutes ago, but no harm in trying that again to be extra sure, i guess.^^

is the encryption inside or outside the mirror?

i.e. is it a mirror of encrypted providers, or did you create the mirror first and then use that as provider for geli?

(because that second one won't work)

RhodiumToad: inside. right, that's also a difference to how my homeserver does this.

damn

why not?

(it will work when the kernel's up, but not for booting)

i thought i was halving computation for the encryption here :<

it won't work because loader doesn't do mirrors and so it won't find the geli metadata in the expected place

it looks at the last sector of the partition for geli metadata, but the mirror metadata will be there, with the geli metadata before it

(yes, encrypting the unmirrored partitions doubles the crypto work needed for writes (but not reads), but if the mirror is just /boot then you do not care about that)

yeah, just wanted to say, if it works as soon as the kernel is loaded, then my root fs can remain geli'd inside the mirror. :)

btw, does efibootmgr do some black magic to store settings from loader.conf somewhere? because it seems to honor the beastie_disabled i only have on the encrypted mirror…

don't think so... but if it can't find /boot it can't load any of the fancy scripts

ah, that makes sense.

all of the menu and fancy stuff is not in the loader binary itself but in lua scripts

ah right, i did see an error that it couldn't read some lua file from /boot

RhodiumToad: do i interpret it correctly, that what you said implies that single mirror providers are still normally mountable filesystems without gmirror running?

err, "consumers" rather, i think.

providers

they kind of are, but there are risks

so this even working at all on my server is already lucky? good to know. :'D

in the normal case if you need to handle the mirrors independently you should make gmirror instances with one provider

but for purposes of booting, the individual provider can be read as if it were a plain filesystem

are the dangers only providers going async or something else?

(writing to it is more of an issue, since that makes it inconsistent, which is why mirroring the ESP itself is tricky since the firmware or other OSes can write to that)

there's no real danger as long as you're just reading it

(putting the metadata at the end was explicitly done to allow this sort of thing)

mhh, sounds like fixing this wouldn't be too involved. might open an issue for that.

fixing what exactly?

geli-inside-mirror boot

eh

how?

i'm not sure i follow. how what?^^

you need actual mirror support in loader since we do actually try to write in some parts that aren't critical until you need them, not as simple as just understanding there's mirror metadata to blow past

ah, alright. i thought you could just bastardize whatever code gmirror has to recognize its metadata and condense it for loader.

you probably could if you kill nextboot in that configuration, but we only recently were able to declare support for nextboot pretty much anywhere and it'd be nice to not break that already

s/any/every/

ahh the lovely freebsd folks :)

hm, nextboot is hoping it can write to /boot/nextboot.conf ?

ah, the uefi setting for which entry is booted the next time the system starts. already thought uefi was already being deprecated.

(or some other configured file)

that one for ufs, the padded area of the pool for zfs

padding

i don't recall the technical name for it; iirc it was called pad1 in stand/libsa/zfs

okay, got my boot mirror redone, retrying boot now. :)

mhhh, now it asks for the passphrase, but still says it can't find a boot partition.

"cp -a /boot /my/mirror/mount" should be sufficient to prepare /boot on the new disk, shouldn't it?

grmpf. i'm not finding anything wrong with the way i set things up, but loader vehemently refuses to recognize the boot partition…

i even tried mounting one of the bare .eli providers for the boot mirror as plain ufs and that works without a problem.

is anything besides /boot required on a geliboot partition?

paste.xinu.at/igMX """screenshot""" of the error

ls / or ls /boot show anything?

didn't try. i just did a plain "ls" and that showed the contents of the efi partition.

ls disk0p2: perhaps?

giving up for today, just gonna do this unencrypted and look into it next weekend or whatever.

henlo from ssd. i guess i at least know that the problem is specific to gelis -g feature.

also a nice thing i noticed: if you geli a mirror instead of creating a mirrof of multiple .eli devices, it only asks for the passphrase for the mirror/foo device, not all the constituent .elis. :)

now to finally get that rc1 going…

rwp: always be sure that your phone line even has caller id enabled, as well

my line does not have caller id. i tried my ATA and caller id comes up with no issues

the phone company charges $13.95 for caller ID service is you have a low use line

but nccid works pretty well

and so does the modem

i should reallly set up up a test vm or two with root encryption

really, it's amazing. i've done open-heart surgery of degrading mirrors and just setting up the whole OS again, switch over, kill the old mirror, add the old device to the new mirror and it never failed me.

gmirror is a superpower.

these days, though, zfs is encroaching on that space

heh, i first did that with zfs and only recently migrated away from it.

case in point, my online server still is zfs. only migrated my homeserver yet.

somehow, i always had problems with zfs… performance that was mysteriously degraded to complete shit, impossible to triage (even after days with help from people here). i never quite got rid of weird problems with automatic mounting either and it was just an almost universally miserable experience for me. tho i have to say i never lost data with it. tho i did come close once – but hey, what do

you expect when you just rip out running drives? ¯\_(ツ)_/¯

people always going "fragmentation is not an issue" while my pools had like 30% fragmentation and performance was degraded to like 800KB/s. :F

and no, it wasn't dedup. i mean, it was once, but i learned from that and gave it a huge berth afterwards. :P