Microsoft Engineer Ports EXT2 File-System Driver To Rust

[oiaohm]

Something to remember lets say your goal is to make a rust ext4 driver where would you start that would be ext2 driver as there is really common code between ext2, ext3 and ext4.

[Amarildo]

EXT2 originates from the 90's and doesn't see much real-world Linux use on modern systems

I guess most of it is Debian, then? I just installed Debian 12 and it defaulted to ext2 for the /boot partition.

[Artim]

When my teams create a POC, they generally aren't jumping immediately to the most complex use cases. EXT2 seems like a very logical starting point. You don't need journaling to do some good real world testing let alone snapshotting or other complex features.

Of course, as a proof that this works on a general level, that's enough. But implementing a much more complex file system would show of the strengths and weaknesses of the implementation. You could show that this implementation can on one hand handle these cases well enough and e.g. be faster as the already existing implementation.

[Artim]

It's possible Microsoft wants to have native ext2/ext4 drivers on windows to read Android MicroSD cards and the likes so they're rewriting the linux driver in rust as a win-win solution of a sort.

Bad example. Android never formated microSD cards with ext file systems, only fat32 or (much later) exFAT. So the only real world usage would be to access Linux file systems from within Windows, as you can access NTFS file systems from Linux.

[Artim]

I guess most of it is Debian, then? I just installed Debian 12 and it defaulted to ext2 for the /boot partition.

Very interesting. Because for me it defaulted to fat32. If you do manual partitioning, that's what will be recommended to you too.

[c117152]

Bad example. Android never formated microSD cards with ext file systems, only fat32 or (much later) exFAT. So the only real world usage would be to access Linux file systems from within Windows, as you can access NTFS file systems from Linux.

Strange, I might be PTSDing about some specific variant like Samsung's or Chromebooks but I definitely remember being asked by some devices to encrypt the microsd card which resulted in a wipe and an ext2 partition... Regardless, they might want to let users mount the WSL file system or whatever weird hyperv setups they have going more directly.

[mdedetrich]

That's why full NTFS support - or even ReFS - would that much of a better proof of concept.

Thats a terrible method to undergo software rewrites, you always start with the smallest demonstrable proof of concept which in this case would either be FAT32 or ext2 especially if the various rust specific interfaces still need to be worked out.

[Artim]

Strange, I might be PTSDing about some specific variant like Samsung's or Chromebooks but I definitely remember being asked by some devices to encrypt the microsd card which resulted in a wipe and an ext2 partition... Regardless, they might want to let users mount the WSL file system or whatever weird hyperv setups they have going more directly.

I think to remember that some devices were able to format the microSD card in a way that it can expand the internal storage. But that were some rare cases.

[Artim]

Thats a terrible method to undergo software rewrites, you always start with the smallest demonstrable proof of concept which in this case would either be FAT32 or ext2 especially if the various rust specific interfaces still need to be worked out.

I didn't say it wasn't a proof of concept.

[spicfoo]

I didn't say it wasn't a proof of concept.

You said implementing NTFS or Refs would be a better proof of concept. What others are pointing out to you is that implementing an advanced filesystem is no longer just a proof of concept. It is a full fledged project. That's not a good starting point.