zfs-windows-2.3.0rc5

[lundman]

Testing refresh with upstream at 2.3.0.

---

This discussion was created from the release zfs-windows-2.3.0rc1.

[guenther-alka]

updated from 2.6 rc11:
still complains about unable to overwrite zed, ignore and ok

zfs list
NAME          USED  AVAIL  REFER  MOUNTPOINT
usb          1.27M  13.6G   168K  /usb
usb/data      332K  13.6G   184K  /usb/data
usb/data/ds   148K  13.6G   148K  /usb/data/ds

data/ds visible under SMB share data
still able to delete data/ds in Explorer

I tried a export/import to make data/ds visible again: BSOD

After a reboot pool is automatically mounted by zed and data/ds visible in the SMB share data
(the switch to legacy and back is no longer needed)

Tried again to delete usb/data/ds via Explorer and SMB
data/ds disappeared but pool export/import worked and folder ds visible so the BSOD could be due a problem from 2.2.6

[lundman]

It isn't clear to me if you mean there is an issue with data/ds here, and 2.3.0rc1 ?

Checking out the zed issue in installer for rc2

[lundman]

I believe zed issue has been addressed. It was surprisingly unkillable - I'd be proud except I wanted it to go away when installing.

[guenther-alka]

yes 2.3.0 rc1

A nested ZFS filesystem like usb/data/ds is visible in Explorer under data as a regular subfolder that you can delete
(should give access denied)

I then tried a zpool export what results in a BSOD

After reboot the pool is imported but the nested ZFS filesystem not visible in Explorer

Workaround to fix it after a reboot
zfs set mountpoint=legacy usb/data/ds
zfs set mountpoint=/usb/data/ds usb/data/ds

pool export and import ok then

[lundman]

OK so usb/data/ds is a ZFS dataset, as in:

zfs list
usb
usb/data
usb/data/ds

and gets mounted as expected (right?)
but you can go and delete it - even though it is a mountpoint, and therefor should have an iocount (lock) to make sure it stays busy while "mounted upon" - then if you export, since the mountpoint is in limbo, there is fireworks.

yes?

[lundman]

Hmm yes, certainly let me delete the mountpoint, so I should increase usecount (not iocount) when mounted. But also add check to delete that if file has usecount, to not allow delete.

[guenther-alka]

Update from rc1-> rc2: Ok
but had to close the black driver install cmd Window manually, not sure if this is due my Avast Antivir

Reinstall rc2: no complain about zed

Delete of folder data or ds (filesystem usb/data or usb/data/ds in Explorer: still possible
export/import usb afterwards works and data or data/ds folder in Explorer visible again

[lundman]

Yeo, only just tried this bug - I will lock mountpoints in future

[lundman]

OK so uh, I looked into this, created 2 NTFS Volumes, D: and D:\lower and sure enough, I can go and delete lower. So it seems Windows does not protect mountpoints. So works as expected? I must admit I find that unpleasant, but we would have to decide if we want to separate from Windows practises?

[inoperable]

@lundman this was acting up on and off again, again, if i do mount each dataset too fast for windows its get confused and results in this, any ideas? happens 2 times out of 10 during logon script (which i use to mount pools, if i do zpool import -a i can forget it, i need it done gradually)

[inoperable]

must use timeout.exe /T proven tech :>

[lundman]

It's weird, it has a condvar, that it waits for the mount to finish, before moving on to next. I mean, I can stick in a sleep(1) and it would be fine, but I'd rather it worked better

[inoperable]

otherwise it perfect, im running those pools since 2020 on windows, figures :]

[inoperable]

@lundman let there be colors 👍

stats work as well, and colors too - however (surprise!) not in windows terminal since it does redmond ansi

[lundman]

I am quite surprised that works, not had time to confirm that it would ;)

[guenther-alka]

off topic
How did you created a partition on say physicaldrive6 to access it under ZFS as Harddisk6Partition0 ?

[inoperable]

off topic How did you created a partition on say physicaldrive6 to access it under ZFS as Harddisk6Partition0 ?

What do you mean exactly? You can explicitly import pool direct in with the \HardDisk... wierd path, I dont recall now if it was \.\ or \ (qemu on windows btw. can access raw devices by this path as well) - if thats what you mean by it

[guenther-alka]

The question is If you want to use partitions of a disk ex for Slog, L2Arc or special vdev:, can you actively partition a disk in diskmanager and assign a partition as vdev to a pool or have you used whole disks and for whatever reason ZFS reports them as HarddiskNPartition0

[inoperable]

@lundman Great job making Windows great again :> since the 2.2.x-rc6~ it's been a very smooth ride for me I haven't got a single BSOD with any past that so.. its getting kind of suspicious, both w10 ltsc and w11 insiders 27xxx branch.

[lundman]

I've also not had time to look at what happens with long filenames, going over 256 to 1024. It is supposed to enable the feature automatically and so on.

[lundman]

OK with a few more fixes to longname, I can use FileTest.exe to create a name like:

\??\D:\0___________________________________________________________________________________________________1___________________________________________________________________________________________________2aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa12345678901234567890123456789hello

and also using FileTest.exe I can confirm directory listing returning the full name ending in "hello".

But from UI, like Explorer, Bash for Windows, CMDline, and Powershell, they all truncate at 250 chars. (After first 123456789).

It is working, now it is more of a question regarding its usefulness. Maybe SMB will show longer? For rc3.

[sskras]

I hope everyone who test this, has following key set:

C:\Users\saukrs> reg query HKLM\SYSTEM\CurrentControlSet\Control\FileSystem /v LongPathsEnabled

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem
    LongPathsEnabled    REG_DWORD    0x1

I mean w10 versions 1607 and up:
https://www.mindgems.com/article/find-long-paths-long-file-names/#p5

[lundman]

What the - what is that? I wish to subscribe

[sskras]

@lundman, I doubt I get your message correctly (was it ironic? :) but anyways:

File Explorer isn't going to support this, at least on w10, I think.
https://superuser.com/questions/1783364/still-no-file-paths-longer-than-260-characters-in-win-explorer-in-2023/1783451#1783451

[lundman]

Interesting, well, it made no difference that I could see - in that it works in FileTest (the app) but nothing outside it in regular Windows. But the superuser URL lists apps that do support it, so perhaps it would be useful after all. Cheers for sharing

[guenther-alka]

Update rc2-> rc3: no problem

Bug with deleting mountpoint in Explorer remains
delete a subfolder /data/ds (=filesystem usb/data/ds): possible
To make it visible again: zpool export +zpool import or switch mountpoint to legacy and back to normal
but zpool export usb -> BSOD

memory dump
https://drive.google.com/file/d/1gZdPfuTjP4iP_-BUgilBakdNab4pZHTv/view?usp=sharing

[guenther-alka]

Update rc4-> rc5: no problem
I had to wait over 5 min until the reboot menu.
As the black installer window covers the setup window, first impression is thar os hangs.
It would be fine if the setuo window is ontop the installer window with a message please wait up to a few minutes

a short test
Explorer: delete a folder that is a mountpoint: fixed, del not possible
start an .exe from ZFS: ok
duplicate a folder with ctrl-c followed by ctrl-v: does not duplicate the folder but the files within: todo

[lundman]

That paste directory is neat, how can that go wrong

[Orenjdos]

I am seeing huge unpaged pool memory usage with I/O over (not necessarily a lot of) time, up to and including with the latest 2.3.0rc6d. For example right now I have

Computer\HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Services\OpenZFS\zfs_arc\zfs_arc_max = 0x280000000 (10737418240) (10 GB)

but task manager's telling me I have 36 GB of nonpaged pool in use, and poolmon.exe -iZFS! tells me ZFS is responsible for 36493844480 bytes, i.e. basically all of that. This memory usage does not go away when I unmount everything or even export my pool (is there a Windows equivalent of rmmod I can use to yoink the driver entirely without rebooting?)

This smells like a memory leak to me, but maybe I'm missing something... attaching a kstat dump; I have no idea what I should be looking at in there.
kstat-1.txt

[inoperable]

@sskras I deleted my own comments -personal rant does not add anything here so not really to add t here, sorry about that

[inoperable]

@sskras I frankly tried to convert my .cmd/.bat batch files to powershell, but the more i dived into it the more M$ tooling stands in its way- so i got some a wierd combo of .cmd/.bat msys2 mix-ins, ansible and lua, works fine and is close enough to unixish shell so i forgot a lot of time its running ntoskrnl :)

[sskras]

As for me, the comments could have stayed. Your unixish experience is definitely of interest for me, so we may communicate in future about other subjects and by other means. May I write you an email message? Would extract your address from some of your PR. Cheers! :)

PS. Have you heard about the Midipix project ? It's like Cygwin/MSYS2, but tends to be faster, more POSIXish, and even allows mixing of POSIX code with WinAPI code (except for some specifics coming from GDI and DirectX).

It's pretty sad it has very low manpower at the moment, hence extremely slow development.

Would be a nice base for porting userland of OpenZFS to Windows.

[inoperable]

@sskras I wrote you an mail btw. didnt got it?

[inoperable]

World would be so much more bearable (if someone really needs MS Windows) with MS Windows without the MS parts :) (allthough, impossible to compile a empty file but i bet coopilot knows how to do it anyways, but not on the free tier, thats pro level stuff)

[lundman]

Hmm when did hostid break, seems to be 0 now

[mio-19]

I was expecting to encounter data loss when sharing ZFS between Linux, macOS, and Windows using experimental OpenZFS forks, but your forks have been remarkably stable. Thank you for your work!