is the compression in zfs similar to gzipping a file ?

I believe each record is compressed separately

(so the recordsize matters)

win 14

I think my compressratio is very poor because my files are smaller than 128k which is the default recordsize

any downsides to having a recordsize of say 8k ?

small recordsizes reduce the compression opportunities (for files larger than the record size)

I have an efi system running freebsd with a zfs mirror for /. One of the drives failed and I have resilvered it onto a new one.

How do I restore the efi partition on the new drive

Create the same size partition of FAT16 or -32 (whatever that is); then copy the files

the efi partition appears to be there but it's not mountable like it is on the other drive

Title: => 40 234441568 ada0 GPT (112G) 40 409600 1 efi (2 - Pastebin.com

ada0 is the replacement

ikkuranus-: what did you do to create the partition?

I don't remember. It was a while ago

ikkuranus-: what do you get from file -s /dev/ada0p1 and file -s /dev/ada1p1

. /dev/ada0p1: data

./dev/ada1p1: DOS/MBR boot sector, code offset 0x3c+2, OEM-ID "BSD4.4 ", root entries 512, sectors 1600 (volumes <=32 MB), sectors/FAT 5, sectors/track 63, heads 1, serial number 0xbd4111ee, label: "EFISYS ", FAT (12 bit), followed by FAT

ok, so you never did a newfs_msdos on the new partition

since it seems to be the exact same size, you could just dd the old one onto the new one

that said, the old one looks like it might have been copied from an install image rather than properly created

if you mount the old one, how large is it?

780K 387Kused

that's rather small given that it's on a 200MB partition. definitely copied from an image

the EFI spec requires a larger size (minimum 100MB I think), whereas install images are created as small as possible

so the best thing would be to newfs_msdos the new one, and copy the files

I assume I need to delete it first before running that

no

you already have the partition, you just don't have a filesystem in it

just to check, though, what is your zpool configuration

mirror

and it shows two devices, ada0p3 and ada1p3 ? exactly those names?

yup

TIL android does now support ikev2 for ipsec

that makes it much more attractive

and swapinfo shows ada0p2 and ada1p2 ?

wait, that was in -social, sorry

yup

ok, so newfs_msdos /dev/ada0p1 should be safe

ok it mounted now I can copy it over

copy it as files, yes

though probably you only have one or two files

that's right

thank you

bsdnow.tv/504 FreeBSD 13.2 release episode

Title: BSD Now 504: Release the BSD

test test

_0pr_: syn ack

_0pr_: test successfully failed

with status code 0

cgit.freebsd.org/ports/log/?qt=grep&q=capsicum this seems worth keeping an eye on

So... I'm trying to pass through a pci device to a bhyve vm. Followed wiki.freebsd.org/bhyve/pci_passthru and the churchers/vm-bhyve man page, and from the host side I guess things look good, but my guest doesn't see a new pci device.

Title: bhyve/pci_passthru - FreeBSD Wiki

I see that `vm passthru` will tell be what devices are *ready* to be passed through, but is there a way to see what is currently being passed through in-the-moment?

what exactly did you do?

The guest is a HAOS, a linux appliance.

I used `pciconf` to identify the peripheral's PCI location, then added that to 'pptdevs' in /etc/loader.conf, then added `passthru0="..."' to /vm/haos/haos.conf

The card is at 3/0/0, perhaps I should have used that instead of "...". :)

Oh, and there was a host reboot in there, so `pciconf` identified the card as "ppt0" instead of "xhci0".

are you loading vmm in loader.conf or in rc.conf?

loader.conf.

ok, and did you also specify to the relevant bhyve instance that it should claim the device?

I think when I did not, I didn't get a ppt0.

Er, that would be the vm configuration, would it not?

yes

I'm using churchers/vm-bhyve

ppt0 is visible in the host when you have pptdevs= and vmm loaded in loader.conf

but you have to explicitly specify that bhyve should pass it through to a specific guest

> grep passthru /vm/haos/haos.conf

passthru0="3/0/0"

Is there some other place as well?

I don't know vm-bhyve

can you see what command-line options it used when starting the guest?

Ya, that would help. I'll try. It changes its text for ps.

I can't find a way to see the options that vm-bhyve generates, but all the instructions I've seen make sense and I've followed them. :(

vm-bhyve also has a "bhyve_options" config option, so perhaps I'll try the freebsd-wiki-way with that.

have you specified that the guest memory is wired down?

(passthrough requires that)

No luck with bhyve_options, the `lspci` output is the same. I have not specified that; if it's required, I would have expected vm-bhyve to "take care of it" for me if I specified a passthru device in the config. Is that mentions in the wiki?

It doesn't seem to be in the Requirements section of wiki.freebsd.org/bhyve/pci_passthru ..

Title: bhyve/pci_passthru - FreeBSD Wiki

"The "-S" option must be passed to both bhyveload(8) and bhyve(8) to wire guest memory."

Does that mean that -S is required for passthru, even though it hasn't been mentioned earlier, or just that it's required for wiring guest memory, which also hasn't been mentioned earlier?

you can't use passthrough without wired guest memory

"Guest memory must be wired using the -S option when a pass-through device is configured." from bhyve(8)

if -S is not given, passthrough configuration will be ignored with a warning

as you said, I just haven't found documentation that says that. The "additional note" you quoted doesn't quite say that, though the wording is unclear.

I would expect vm-bhyve to have some specific option for specifying wired memory

if -S is required, why would the wiki page not even mention its syntax?

... it does?

not that I saw.

Step 5 is all about "-s" which is a different option..

it's just -S, there is no more to it than that

vm-bhyve has a "wired_memory" config option, but doesn't mention any connection to passthru.

well, you need to enable that

Okie, I'll try it with -S added as a bhyve option. We'll see if it changes. :)

no, use vm-bhyve's wired_memory option

Tried both, neither changed anything. Perhaps my get OS is at fault, I dunno.

guest

I'll try again using a freebsd guest for testing, where I'll know the diagnostic tools better.

ghoti: is "/etc/loader.conf" a typo?

yuripv: yes, that was a typo.

Hmm. I'm getting ld-elf.so.1: /usr/local/bin/gimp: Undefined symbol "g_string_free_and_steal" when I try to launch a freshly-installed gimp. I see that `g_string_free_and_steal()' was added to GLib 2.75.3 in February, but I'm running version 2.74. Upgrading to the latest, 2.76, lets gimp run.

Can packages not have version dependencies, or only ports? Or would it have to be done with a bespoke pre-install script?

Were no dependencies updated with update of gimp?

OTOH if something else was updated under gimp, that would also explain the error

hi

i came here to ask if qemu has kernel acceleration under freebsd?

i wonder what is better qemu or virtual box or something else

wikan, for amd64 use bhyve.

so if I get this right, in zfs a small recordsize is ideal for applications that need to modify small blocks within large files

aka vmdk files, or other large binaries

and 128k is sort of the default middle-ground

last1, In all file systems the choice of block size is a compromise. Smaller is more packed but more overhead and larger is more efficient with less overhead but more wasted unusable fragments.

Historically block size used to be smaller but that caused impractically large amount of overhead and counter overflows.

So block size has been steadily increased over the years as storage sizes has increased to what feels like huge sizes now compared to a decade ago.

This results in more inefficient use for storing small bits of data. But we now have support for this much larger total amount of storage to compensate.

I don't know what would be the best block size for vmdk virtual machine back end storage files. In the end there is no substitute for benchmarking.

well here's the thing about zfs recordsize, from what I read

if I set it to 1M and my file is 4k it will still occupy 4k

At least use compression=lz4. :-o

so large recordsizes in zfs doesn't command wasted fragments

"Therefore, for simple file sharing you should typically set recordsize=1M—tiny files take care of themselves by being stored in tiny blocks, requiring no additional tuning. But your large files get the benefit of higher compression ratios, fewer I/O operations required to read or write the file, fewer indirect blocks, and less impact from on-disk fragmentation."

from: klarasystems.com/articles/tuning-recordsize-in-openzfs

Title: Tuning recordsize in OpenZFS | Klara Inc

wikan: I use bhyve. It is excellent. I used to use VirtualBox. bhyve performs WAAY better than vbox.

ghoti thanks

Would it be silly to think that I could use a GPU in a FreeBSD bare metal machine from within a linux bhyve VM running on that machine? (Knowing that I need the GPU full power for CUDA processing...)

it is demonstrated, but experimentally

I see.

is freebsd performance near archlinux? how does netbsd compare? dragonfly?

performance at what ?

concrete_houses: for all the BSDs the WiFi is slower than Linux but inroads are being made