Cannot use special characters/Umlauts when datasets are mounted

[derritter88]

System information

Type	Version/Name
Distribution Name	Win 11
OpenZFS Version	zfs-2.2.99-5-ga6951e43bf

When having a pool mounted with multiple datasets like

NAME                USED  AVAIL  REFER  MOUNTPOINT
Backup-Data3       3.38T  3.76T   132K  /Backup-Data3
Backup-Data3/Data  2.08T  3.76T  2.08T  /Backup-Data3/Data
Backup-Data3/VM    1.29T  3.76T  1.29T  /Backup-Data3/VM

in a layout like:

F:\
F:\Data
F:\VM

then I cannot create files with special characters and I cannot access any existing files.

When using driveletter=on for data sets then there are no problems for special characters.

[lundman]

OK, so playing with: E:\lower\Segeln_FB2-Prüfungstörn_2019-05

where E:\lower is a sub-dataset.

What happens is we get a request to open \lower\Segeln_FB2-Prüfungstörn_2019-05 and we detect \lower is a REPARSE point, and correctly return STATUS_REPARSE.

However, we are also supposed to work out how much is remaining of the filename:

            rpb->Reserved = strlen(finalname) * sizeof (WCHAR);
            if (rpb->Reserved != 0)
                rpb->Reserved += sizeof (WCHAR); // the slash

So finalname here is utf8 of Segeln_FB2-Prüfungstörn_2019-05 and comes to 0x25 bytes and at 2-byte chars, 0x4a. Plus a slash, making a total of 0x4c. But, we only used \lower, some 0x6 bytes (0xa). The full 2byte char string given is 0x52 so we should actually return 0x48, and not 0x4c. If I hardcode 0x4c->0x48 I can successfully use the name Segeln_FB2-Prüfungstörn_2019-05

The reason here is strlen() returns bytes, and not characters. Windows always uses 2byte chars, so doing the math without utf8 would be easier, but not fit with ZFS. Doing the math based on "given minus used" would work, but you can also have umlauts in the directory name \löwer - so it would just shift the issue. Having a u8_strlen() to count chars might work.

Pondering.

[lundman]

f886377

[derritter88]

Thanks for that - will you release a new version for me to test it?

[lundman]

New build is up

[derritter88]

This works now - thanks!

[derritter88]

Found a different thing (let me know if you need a new issue ticket): When copying stuff via driveletter=off FreeFileSync has issues with creating new folders ERROR_PATH_NOT_FOUND - [FindFirstFileEx] and translated "cannot read file attributes of folder F:\Data\Mathias\Apple Backup"

I then unmounted the subset and remounted it with driveletter=on - backup/sync worked without any issues.

[lundman]

Yeah make a new tickets, with details on how I can reproduce it