Subpagesize-blocksize: Allow I/O on blocks whose size is less than page size

Subpagesize-blocksize: Allow I/O on blocks whose size is less than page size

[Niccolò Belli]

Hi,
What happened to the subpagesize-blocksize patchset? It seems to be 
untouched since 3 years.
I'm currently stuck on Fedora ppc64le because it ships with a 64k page 
size, but I need it to be 4k in order to be able to chroot into foreign 
architectures (like x86_64) with qemu-user.

If I try to mount my btrfs partition with a kernel with page size set to 
4k I get the following error:

sectorsize 65536 not supported yet, only support 4094

I didn't try the patchset, because it doesn't support compression and my 
whole fs is compressed.

It looked like an awesome work, why didn't it get merged?

Bests,
Niccolo'

Re: Subpagesize-blocksize: Allow I/O on blocks whose size is less than page size

[Nikolay Borisov]

On 17.09.19 г. 11:31 ч., Niccolò Belli  wrote:
> Hi,
> What happened to the subpagesize-blocksize patchset? It seems to be
> untouched since 3 years.
> I'm currently stuck on Fedora ppc64le because it ships with a 64k page
> size, but I need it to be 4k in order to be able to chroot into foreign
> architectures (like x86_64) with qemu-user.
> 
> If I try to mount my btrfs partition with a kernel with page size set to
> 4k I get the following error:
> 
> sectorsize 65536 not supported yet, only support 4094
> 
> I didn't try the patchset, because it doesn't support compression and my
> whole fs is compressed.
> 
> It looked like an awesome work, why didn't it get merged?

TLDR: It's not ready and likely won't be in the upcoming months.

There are a bunch of stuff that need to happen before subpage blocksize
support can land in. Namely, the btree inode needs to be removed. Josef
has a WIP branch doing that but it in turn is dependent on a new
mechanism to support proper writeback throttling of metadata from the
superblock. OTOH I have an idea of how to remove the btree indoe by
exploiting the existing shrinkers mechanism but that's only a concept
and there are some reservations whether it's doable (It is but there are
concerns shrinker infrastructure cannot cope with the rate of dirty
metadata that btrfs produces.)

> 
> Bests,
> Niccolo'
>