Subpage support

Subpage block size support, or just subpage for short, is a feature to allow using a filesystem that has different size of data block size (sectorsize) and the host CPU page size. For easier implementation the support was limited to the exactly same size of the block and page. On x86_64 this is typically 4KiB, but there are other architectures commonly used that make use of larger pages, like 64KiB on 64bit ARM or PowerPC. This means filesystems created with 64KiB sector size cannot be mounted on a system with 4KiB page size.

While with subpage support, systems with 64KiB page size can create (still needs “-s 4k” option for mkfs.btrfs) and mount filesystems with 4KiB sectorsize, allowing us to push 4KiB sectorsize as default sectorsize for all platforms in the near future.

Requirements, limitations

The initial subpage support has been added in v5.15, although it’s still considered as experimental at the time of writing (v5.18), most features are already working without problems.

End users can mount filesystems with 4KiB sectorsize and do their usual workload, while should not notice any obvious change, as long as the initial mount succeeded (there are cases a mount will be rejected though).

The following features has some limitations for subpage:

  • RAID56 support This support is already queued for v5.19 cycle. Any fs with RAID56 chunks will be rejected at mount time for now.

  • Support for page size other than 64KiB The support for other page sizes (16KiB, 32KiB and more) are already queued for v5.19 cycle. Initially the subpage support is only for 64KiB support, but the design makes it pretty easy to enable support for other page sizes.

  • No inline extent creation This is an artificial limit, to prevent mixed inline and regular extents.

    It’s possible to create mixed inline and regular extents even with non-subpage mount for certain corner cases, it’s way easier to create such mixed extents for subpage.

    Thus max_inline mount option will be silently ignored for subpage mounts, and it always acts as “max_inline=0”.

  • Compression write is limited to page aligned ranges Compression write for subpage is introduced in v5.16, with the limitation that only page aligned range can be compressed. This limitation is due to how btrfs handles delayed allocation.

  • No support for v1 space cache V1 space cache is considered deprecated, and we’re defaulting to v2 cache in btrfs-progs already. The old v1 cache has quite some hard coded page size usage, and consider it is already deprecated, we force v2 cache for subpage.

  • Slightly higher memory usage for scrub This is due to how we allocate pages for scrub, and will be fixed in the coming releases soon.