F2FS currently describes several on-disk metadata layouts and filesystem
geometry values with constants derived from PAGE_SIZE or a fixed 4KB block
size. This series derives those values from the mounted filesystem's block
size instead, as preparation for supporting filesystem blocks smaller than
PAGE_SIZE.

The patches initialize sb_info early, describe the SIT, NAT, orphan, dentry,
node, and xattr layouts dynamically, parameterize sector/byte/block
conversion helpers, and derive node-tree geometry at runtime.

This series does not relax the block-size mount check or enable subpage I/O.
It is intended to be a behavior-preserving refactoring for currently
supported 4KB-block filesystems.

Changes since v5:
- no code changes
- send each diff as a base64-encoded text/x-patch attachment because Gmail's
  outgoing transport reflowed long lines in inline text/plain bodies
- verified that an actual received raw message applies with git am and
  produces the expected tree; the list-side copy of patch 01/11 will be
  checked before completing the series

v5:
https://lore.kernel.org/linux-f2fs-devel/CAH=xXfH7TPGt-zKXvJRNTgGDZKa1Xoq_sF3-WGs5kdsQLuYn4g@mail.gmail.com/

Changes since v4:
- no code changes
- resend with CRLF-canonicalized RFC822 messages; the delivered inline patch
  was still reflowed by Gmail's outgoing transport

v4:
https://lore.kernel.org/linux-f2fs-devel/CAH=xXfHWPg6qRwW+N2ziH1cvG3OD3wJc1z_Tm04MtFCpqYmQFg@mail.gmail.com/

Changes since v3:
- no code changes
- resend with git send-email because the previous mail-sending path produced
  hard-wrapped hunk headers in the delivered messages

v3:
https://lore.kernel.org/linux-f2fs-devel/CAH=xXfEFL7w7nsU3xjNubKJ1soYMqT-DbJBuyzfEX4Rv3rdFEA@mail.gmail.com/

Changes since v2:
- use DECLARE_FLEX_ARRAY() for flexible arrays that are alone in a struct or
  used in a union, fixing the reported m68k GCC build failures
- use F2FS_BLK_ALIGN() in f2fs_sec_trim_file() to avoid a runtime 64-bit
  division and an unresolved __divdi3 reference on 32-bit SH
- no other code changes

Changes since v1:
- rebase onto f2fs/dev at c966d29e01bb
- drop the unrelated free-NID underflow fix from this series
- split the preparatory changes into 11 focused patches
- initialize sb_info earlier and describe the xattr layout explicitly
- separate sector, byte/block, and block-size macro parameterization
- update commit messages and address checkpatch warnings

v2:
https://lore.kernel.org/linux-f2fs-devel/CAH=xXfHWW_J92z3YgvoLSx8+z6rekq=p2kiP15KE9=2+Ju2RjQ@mail.gmail.com/
v1:
https://lore.kernel.org/linux-f2fs-devel/CAH=xXfFVAwQ3r_NdUDtmXhtg6Lw8kDWoNTzGL5oy4mwHMsJLJw@mail.gmail.com/

Tested with:
- x86_64 W=1 F2FS build at every patch
- arm64 16KB-page W=1 F2FS build at the series tip
- m68k W=1 F2FS builds with the two reported randconfigs at the affected
  patch points
- SH W=1 F2FS build with the reported randconfig at the affected patch;
  fs/f2fs/file.o has no __divdi3 reference
- scripts/checkpatch.pl --strict (no errors or warnings)

Kelvin Zhang (11):
  f2fs: initialize sb_info early in f2fs_fill_super
  f2fs: describe SIT block layout dynamically
  f2fs: describe NAT block layout dynamically
  f2fs: describe orphan block layout dynamically
  f2fs: describe dentry block layout dynamically
  f2fs: describe {i,d,id}node block layout dynamically
  f2fs: describe xattr block layout dynamically
  f2fs: parameterize sector conversion macros
  f2fs: parameterize byte and block conversion macros
  f2fs: parameterize block size and mask macros
  f2fs: describe node tree geometry dynamically

 fs/f2fs/checkpoint.c    |  75 ++++++++++--------
 fs/f2fs/data.c          |  85 +++++++++++---------
 fs/f2fs/debug.c         |   4 +-
 fs/f2fs/dir.c           |  51 ++++++------
 fs/f2fs/extent_cache.c  |   6 +-
 fs/f2fs/f2fs.h          | 122 +++++++++++++++++++++-------
 fs/f2fs/file.c          |  84 +++++++++++---------
 fs/f2fs/gc.c            |  27 ++++---
 fs/f2fs/inline.c        |  11 ++-
 fs/f2fs/inode.c         |  22 +++---
 fs/f2fs/node.c          | 171 +++++++++++++++++++++-------------------
 fs/f2fs/node.h          |  89 +++++++++++----------
 fs/f2fs/recovery.c      |   4 +-
 fs/f2fs/segment.c       |  60 +++++++-------
 fs/f2fs/segment.h       |  32 ++++----
 fs/f2fs/shrinker.c      |   2 +-
 fs/f2fs/super.c         |  83 +++++++++++++------
 fs/f2fs/sysfs.c         |  10 +--
 fs/f2fs/verity.c        |   6 +-
 fs/f2fs/xattr.c         |  10 ++-
 fs/f2fs/xattr.h         |  15 ++--
 include/linux/f2fs_fs.h | 131 ++++++++++++++++--------------
 22 files changed, 636 insertions(+), 464 deletions(-)


base-commit: c966d29e01bbf829f8bb4a39a49811c56cdb49c3
-- 
2.53.0

Hi Kelvin,

I see only the first patch in this series. Am I missing something?

On 08/31, Kelvin Zhang wrote:
> F2FS currently describes several on-disk metadata layouts and filesystem
> geometry values with constants derived from PAGE_SIZE or a fixed 4KB block
> size. This series derives those values from the mounted filesystem's block
> size instead, as preparation for supporting filesystem blocks smaller than
> PAGE_SIZE.
> 
> The patches initialize sb_info early, describe the SIT, NAT, orphan, dentry,
> node, and xattr layouts dynamically, parameterize sector/byte/block
> conversion helpers, and derive node-tree geometry at runtime.
> 
> This series does not relax the block-size mount check or enable subpage I/O.
> It is intended to be a behavior-preserving refactoring for currently
> supported 4KB-block filesystems.
> 
> Changes since v5:
> - no code changes
> - send each diff as a base64-encoded text/x-patch attachment because Gmail's
>   outgoing transport reflowed long lines in inline text/plain bodies
> - verified that an actual received raw message applies with git am and
>   produces the expected tree; the list-side copy of patch 01/11 will be
>   checked before completing the series
> 
> v5:
> https://lore.kernel.org/linux-f2fs-devel/CAH=xXfH7TPGt-zKXvJRNTgGDZKa1Xoq_sF3-WGs5kdsQLuYn4g@mail.gmail.com/
> 
> Changes since v4:
> - no code changes
> - resend with CRLF-canonicalized RFC822 messages; the delivered inline patch
>   was still reflowed by Gmail's outgoing transport
> 
> v4:
> https://lore.kernel.org/linux-f2fs-devel/CAH=xXfHWPg6qRwW+N2ziH1cvG3OD3wJc1z_Tm04MtFCpqYmQFg@mail.gmail.com/
> 
> Changes since v3:
> - no code changes
> - resend with git send-email because the previous mail-sending path produced
>   hard-wrapped hunk headers in the delivered messages
> 
> v3:
> https://lore.kernel.org/linux-f2fs-devel/CAH=xXfEFL7w7nsU3xjNubKJ1soYMqT-DbJBuyzfEX4Rv3rdFEA@mail.gmail.com/
> 
> Changes since v2:
> - use DECLARE_FLEX_ARRAY() for flexible arrays that are alone in a struct or
>   used in a union, fixing the reported m68k GCC build failures
> - use F2FS_BLK_ALIGN() in f2fs_sec_trim_file() to avoid a runtime 64-bit
>   division and an unresolved __divdi3 reference on 32-bit SH
> - no other code changes
> 
> Changes since v1:
> - rebase onto f2fs/dev at c966d29e01bb
> - drop the unrelated free-NID underflow fix from this series
> - split the preparatory changes into 11 focused patches
> - initialize sb_info earlier and describe the xattr layout explicitly
> - separate sector, byte/block, and block-size macro parameterization
> - update commit messages and address checkpatch warnings
> 
> v2:
> https://lore.kernel.org/linux-f2fs-devel/CAH=xXfHWW_J92z3YgvoLSx8+z6rekq=p2kiP15KE9=2+Ju2RjQ@mail.gmail.com/
> v1:
> https://lore.kernel.org/linux-f2fs-devel/CAH=xXfFVAwQ3r_NdUDtmXhtg6Lw8kDWoNTzGL5oy4mwHMsJLJw@mail.gmail.com/
> 
> Tested with:
> - x86_64 W=1 F2FS build at every patch
> - arm64 16KB-page W=1 F2FS build at the series tip
> - m68k W=1 F2FS builds with the two reported randconfigs at the affected
>   patch points
> - SH W=1 F2FS build with the reported randconfig at the affected patch;
>   fs/f2fs/file.o has no __divdi3 reference
> - scripts/checkpatch.pl --strict (no errors or warnings)
> 
> Kelvin Zhang (11):
>   f2fs: initialize sb_info early in f2fs_fill_super
>   f2fs: describe SIT block layout dynamically
>   f2fs: describe NAT block layout dynamically
>   f2fs: describe orphan block layout dynamically
>   f2fs: describe dentry block layout dynamically
>   f2fs: describe {i,d,id}node block layout dynamically
>   f2fs: describe xattr block layout dynamically
>   f2fs: parameterize sector conversion macros
>   f2fs: parameterize byte and block conversion macros
>   f2fs: parameterize block size and mask macros
>   f2fs: describe node tree geometry dynamically
> 
>  fs/f2fs/checkpoint.c    |  75 ++++++++++--------
>  fs/f2fs/data.c          |  85 +++++++++++---------
>  fs/f2fs/debug.c         |   4 +-
>  fs/f2fs/dir.c           |  51 ++++++------
>  fs/f2fs/extent_cache.c  |   6 +-
>  fs/f2fs/f2fs.h          | 122 +++++++++++++++++++++-------
>  fs/f2fs/file.c          |  84 +++++++++++---------
>  fs/f2fs/gc.c            |  27 ++++---
>  fs/f2fs/inline.c        |  11 ++-
>  fs/f2fs/inode.c         |  22 +++---
>  fs/f2fs/node.c          | 171 +++++++++++++++++++++-------------------
>  fs/f2fs/node.h          |  89 +++++++++++----------
>  fs/f2fs/recovery.c      |   4 +-
>  fs/f2fs/segment.c       |  60 +++++++-------
>  fs/f2fs/segment.h       |  32 ++++----
>  fs/f2fs/shrinker.c      |   2 +-
>  fs/f2fs/super.c         |  83 +++++++++++++------
>  fs/f2fs/sysfs.c         |  10 +--
>  fs/f2fs/verity.c        |   6 +-
>  fs/f2fs/xattr.c         |  10 ++-
>  fs/f2fs/xattr.h         |  15 ++--
>  include/linux/f2fs_fs.h | 131 ++++++++++++++++--------------
>  22 files changed, 636 insertions(+), 464 deletions(-)
> 
> 
> base-commit: c966d29e01bbf829f8bb4a39a49811c56cdb49c3
> -- 
> 2.53.0
> 
> 
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Previously, init_sb_info() was invoked late in f2fs_fill_super(), after
processing superblock error tracking, mount options, and sb->s_maxbytes.

However, read_raw_super_block() reads the superblock folio directly
from the block device mapping using folio page offsets. This makes the
initial static sb_set_blocksize(sb, F2FS_BLKSIZE) call redundant.

Move init_sb_info() to execute immediately after sbi->raw_super is
attached, allowing the block device block size to be set dynamically via
sb_set_blocksize(sb, sbi->blocksize).

This early initialization also prepares for upcoming patches that add
dynamically computed filesystem geometry fields, such as
sbi->addrs_per_block and sbi->nids_per_block, to init_sb_info(). These
fields are required by earlier initialization steps, including mount option
consistency checks and sb->s_maxbytes computation.

Signed-off-by: Kelvin Zhang <zhangxp1998@gmail.com>
---
 fs/f2fs/super.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)