BSOD while install driver, work rsync in cygwin, export pool and unmount pool, can't change atime

[leonik3]

System information

Type	Version/Name
Distribution Name	Windows
Distribution Version	10
Kernel Version
Architecture	x64
OpenZFS Version	zfswin2.3.0rc2 and zfswin2.3.0rc3 and 2.2.6rc11

1. BSOD When installing zfswin2.3.0rc2 on fresh new windows.
   I solved by:
   a. installing in Windows safe mode (the zpool command was not available after this step),
   b. reinstalling in normal Windows mode.

After that I did an upgrade to zfswin2.3.0rc3 and on that version:

2. BSOD when I do the command “zpool export zw”

3. BSOD when I do the command “zpool umount zw”

4. BSOD when synchronizing a pool as a target. I give the rsync command on another computer. For about two hours rsync almost does not copy new files, but only compares directories, until finally BSOD. I repeated this several times.

rsync -P -h -r -v -i --append-verify --suffix=-ver-$(date "+%Y_%m_%d-%H_%M_%S") --exclude={'hiberfil.sys','pagefile.sys','swapfile.sys'} -s '/cygdrive/x/' [email protected]:'/cygdrive/w/x'

Unfortunately, during this sync time, in addition to reading on this target pool, zpool is also constantly writing (I see in zpool iostat). My guess is that the issue is with atime. On this pool I had atime set to off. I don't know what switched this parameter and turned it on - still somehow strangely as “temporary”. I wanted to change it, but can't. I typed another point on it - below.

5. can't change atime
   zfs set atime=off zw
   cannot mount 'zw': Unknown error
   property may be set but unable to remount filesystem

[lundman]

You can run zpool version to confirm userland and kernel are the same version, but it looks correct from here.

See if you can paste the stack from the crash using memory.dmp, or failing that, share the memory.dmp - it is the only way we will know what went wrong.

[leonik3]

I have now changed the version to zfswin-2.2.6rc11 and got a BSOD on the command “zpool export zw”.

I had no such problem on another computer with zfswin-2.2.6rc11.

Maybe the reason is my new computer I am testing. Maybe it's because it has too many cores, or too much memory or the new disk controller I used to connect the drives?

On this older version of zfs I have the same problem with atime - it's on and I can't turn it off.

Only reading from the pool, but there are writings;

[leonik3]

Memory dump on zfswin-2.2.6rc11
https://www.uschovna.cz/zasilka/RH5VTF87JDXF36FR-NW6/

[lundman]

Don't worry about atime - it doesn't exist on Windows. Defaults to on true, but that is Linux default.

[leonik3]

OK, but I'm worried about these writes when it only goes read.

[lundman]

Ok mostly reading, but there are some writes, and you are curious why? Yeah I see why you look at atime, I suppose even though Windows doesn't know what it is, but ZFS does internally, so if it is on by default maybe that does affect us.

You can try rc4, it will let you set atime to off, I confirmed it in debugger - but it will incorrectly say "on" if you check the value.

[leonik3]

Thank you.

When there is dedup and millions of files, one worries about these writes, because they slow down the reading and tire traditional hard drives completely unnecessarily.

[leonik3]

On the version zfswin-2.3.0rc4 I can execute the "zfs set atime=off zw" command but it doesn't work for me. The child dataset now has atime=off, but unfortunately the main one still has atime=on. When I run rscync, nothing has changed. I still have disk writes whenever I do only reading. I do the reading from the main dataset.

[guenther-alka]

are those problems still there in newest 2.3.0rc5?

[lundman]

I am not convinced it is from atime now, perhaps FileSpy.exe will show were the write requests come from, if they are external.