btrfs: initial bs > ps support

[CHANGELOG]
v2:
- Add a new patch to fix the incorrect @max_bytes of find_lock_delalloc_range()
This in fact also fixes a very rare corner case where bs < ps support is also affected.

This allows us to re-enable extra large folios (folio > bs) for bs > ps cases.

RFC->v1
- Disable extra large folios for bs > ps mounts Such extra large folios are larger than a block.

Still debugging, but disabling it makes 8K block size runs survive the full fs tests, with some minor failures due to the limitations.

This may be something affecting regular large folios (folio > bs, but bs <= ps).

This series enables the initial bs > ps support, with several limitations:

- No direct IO support
All direct IOs fall back to buffered ones.

- No RAID56 support
Any fs with RAID56 feature will be rejected at mount time.

- No encoded read/write/send
Encoded send will fallback to the regular send (reading from page cache).
Encoded read/write utilized by send/receive will fallback to regular ones.

Above limits are introduced by the fact that, we require large folios to cover at least one fs block, so that no block can cross large folio boundaries.

This simplifies our checksum and RAID56 handling.

The problem is, user space programs can only ensure their memory is properly aligned in virtual addresses, but have no control on the backing folios. Thus they can got a contiguous memory but is backed by incontiguous pages.

In that case, it will break the "no block can cross large folio boundaries" assumption, and will need a very complex mechanism to handle checksum, especially for RAID56.

The same applies to encoded send, which uses vmallocated memory.

In the long run, we will need to support all those complex mechanism.

[FUTURE ROADMAP]
Currently bs > ps support is only to allow extra compatibility, e.g. allowing x86_64 to mount a btrfs which is originally created on ppc64 (64K page size, 64K block size).

But this should also open a new door for btrfs RAID56 write hole problems in the future, by enforcing a larger block size and fixed power-of-2 data stripes, so that every write can fill a full stripe, just like RAIDZ.

E.g. with 8K block size, all data writes are now in 8K sizes, and will always be a full stripe write for a 3 disks RAID5 with a stripe length of 4K.

This RAIDZ like solution will allow a much simpler RAID56 (no more RMW any more), at the cost of a larger block size (more write-amplification, higher memory usage etc).

Qu Wenruo (9):
btrfs: fix the incorrect @max_bytes value for
find_lock_delalloc_range()
btrfs: prepare compression folio alloc/free for bs > ps cases
btrfs: prepare zstd to support bs > ps cases
btrfs: prepare lzo to support bs > ps cases
btrfs: prepare zlib to support bs > ps cases
btrfs: prepare scrub to support bs > ps cases
btrfs: fix symbolic link reading when bs > ps
btrfs: add extra ASSERT()s to catch unaligned bios
btrfs: enable experimental bs > ps support

 fs/btrfs/bio.c         | 27 +++++++++++++++++++
 fs/btrfs/compression.c | 42 ++++++++++++++++++++---------
 fs/btrfs/compression.h |  2 +-
 fs/btrfs/direct-io.c   | 12 +++++++++
 fs/btrfs/disk-io.c     | 14 ++++++++--
 fs/btrfs/extent_io.c   | 21 +++++++++++----
 fs/btrfs/extent_io.h   |  3 ++-
 fs/btrfs/fs.c          | 20 ++++++++++++--
 fs/btrfs/fs.h          |  6 +++++
 fs/btrfs/inode.c       | 18 +++++++------
 fs/btrfs/ioctl.c       | 35 +++++++++++++++++-------
 fs/btrfs/lzo.c         | 59 ++++++++++++++++++++++-------------------
 fs/btrfs/raid56.c      | 42 +++++++++++++++++++----------
 fs/btrfs/raid56.h      |  4 +--
 fs/btrfs/scrub.c       | 51 +++++++++++++++++++----------------
 fs/btrfs/send.c        |  9 ++++++-
 fs/btrfs/zlib.c        | 60 +++++++++++++++++++++++++++---------------
 fs/btrfs/zstd.c        | 44 +++++++++++++++++--------------
 18 files changed, 321 insertions(+), 148 deletions(-)

2.50.1