Re: [f2fs-dev] [RFC PATCH v4 0/3] f2fs: introduce inline extent mapping for inode data blocks

On 8/6/26 21:19, Yongpeng Yang wrote:
> From: Yongpeng Yang <[email protected]>
> 
> Changes since v3:
> - Do not cache NEW_ADDR in the inline extent area. A NEW_ADDR write is
>   now handled exactly like NULL_ADDR: it only unmaps the block.
> - Simplify the insert path. A single-block overwrite is now expressed as
>   four small steps:
>   1) reserve room for a possible split.
>   2) unmap the old mapping of the block.
>   3) insert the new mapping.
>   4) try to merge the inserted mapping with its left and right.
> - Remove the inline-extent code-coverage statistics.
> 
> v3: https://lore.kernel.org/all/[email protected]/
> 
> Changes since v2:
> - Inline extent no longer covers direct block mappings. The inline
>   extent area is now placed after struct f2fs_inode->i_extra_end,
>   inside the i_extra_isize region. This removes the need for format
>   conversion and for f2fs-tools to understand/modify the inline extent
>   on-disk format.
> - Split struct f2fs_inode->i_compr_blocks (was __le64) into __le32
>   i_compr_blocks + __le32 i_inline_ext_capacity, where
>   i_inline_ext_capacity records the per-inode inline extent capacity.
> - Dropped [RFC PATCH v2 1/5] "replace raw dnode pointer arithmetic with
>   f2fs_data_blkaddr()": f2fs_truncate_data_blocks_range no longer needs
>   changes to accommodate the redesigned inline extent.
> - Dropped [RFC PATCH v2 3/5] "support setting inline extent flag via
>   ioctl": format conversion between inline extent and the direct block
>   address array is no longer needed.
> - Re-ran the performance tests (see updated numbers below).
> 
> v2: https://lore.kernel.org/all/[email protected]/
> 
> Changes since v1:
> - Introduce tracepoints for f2fs_iext_update_data_blkaddr and
>   f2fs_iext_lookup_blkaddr to aid debugging (new patch 5/5).
> - Bypass inline extent lookup for F2FS_GET_BLOCK_PRECACHE to ensure all
>   mappings are loaded into the read extent cache.
> - Unify the check for fofs exceeding direct_blocks range to use
>   "fofs >= direct_blocks" consistently.
> - Remove support for caching NULL_ADDR in inline extent area. If a fofs
>   within [0, direct_blocks) is not found in inline extent, it implies
>   NULL_ADDR. This simplifies merge and split logic.
> - Fix f2fs_iext_enable_inline_extent to use PTR_ERR instead of -ENOMEM.
> - Change f2fs_iext_convert_to_inline_extent return type to bool.
> - Add benchmark data covering 4K/8K/32K/64K random read.
> - Rename __is_extent_mergeable to __is_iextent_mergeable to avoid
>   naming collision with extent cache code.
> - Remove inode parameter from f2fs_iext_sanity_check (always NULL).
> - Reduce #ifdef CONFIG_F2FS_INLINE_EXTENT nesting in node.c.
> - Code style fixes to comply with kernel coding style.
> 
> v1: https://lore.kernel.org/all/[email protected]/
> 
> This patchset introduces an inline extent mapping mechanism for f2fs.
> Instead of storing individual block addresses for indirect-node blocks,
> this feature packs contiguous block ranges into compact extent entries
> stored directly in the inode, reducing indirect/double-indirect node
> page reads and enabling O(log n) block address lookup via binary search.
> 
> Design overview:
> - The inline extent area is placed immediately after i_extra_end in the
>   on-disk inode, within the i_extra_isize region. Its size (number of
>   extent entries) is determined by the new field i_inline_ext_capacity.
> - On-disk layout:
>     [i_extra_isize .. i_extra_end] [f2fs_iext_header | f2fs_extent[cap]]
>     |<------------------ i_extra_isize (in bytes) ------------------->|
> - Inline extent only caches mappings for indirect blocks
>   (fofs >= ADDRS_PER_INODE). Direct block mappings continue to use
>   i_addr[], so no format conversion is required.
> - Per-inode capacity is decided at file creation time based on the
>   configured file-extension matching rules (see sysfs below).
> - Mutually exclusive with compression.
> 
> Patch 1: Core implementation -- data structures, extent operations
>          (lookup, insert, merge, split, truncate), and integration
>          with the f2fs data/node/inode/recovery paths.
> Patch 2: sysfs interface -- runtime enable/disable toggle and the file
>          extension list (with optional per-extension capacity) for
>          automatic inline extent activation.
> Patch 3: Tracepoints for inline extent lookup and update operations.
> 
> Test setup (Xiaomi smartphone, UFS 4.0 storage):
> 
>   echo 1 > /sys/fs/f2fs/<dev>/inline_extent_enable
>   echo 'mp4:256' > /sys/fs/f2fs/<dev>/inline_extent_extension_list
>   fio --name=test --filename=data.mp4 --rw=write:4k --bs=64M \
>       --size=8G --ioengine=libaio --direct=1
>   sync
>   fio --name=test --filename=data.mp4 --rw=write --bs=64M \
>       --size=8G --ioengine=libaio --direct=1
>   sync
>   echo 3 > /proc/sys/vm/drop_caches
>   fio --name=buffer-read --ioengine=libaio --rw=randread --bs=$BS \
>       --size=8G --io_size=1G --numjobs=1 --filename=data.mp4
> 
> Results (random read bandwidth, MiB/s):
> +---------------------------------------------------+
> | BS     | baseline | inline ext | improvement      |
> |--------+----------+------------+------------------|
> | 4K     | 31.6     | 32.4       | +2.5%            |
> | 8K     | 55.4     | 58.5       | +5.6%            |
> | 32K    | 155.3    | 166.8      | +7.4%            |
> | 64K    | 229.8    | 255.3      | +11.1%           |
> | 128K   | 337.8    | 388        | +14.9%           |
> +---------------------------------------------------+
> 
> The improvement comes from eliminating indirect/double-indirect node
> page reads during block address lookup -- indirect-block mappings are
> stored directly in the inode page and found via binary search.
> 
> Yongpeng Yang (3):
>   f2fs: introduce inline extent mapping for inode data blocks
>   f2fs: add sysfs interface for inline extent management
>   f2fs: introduce tracepoints for inline extent lookup and update
> 
>  fs/f2fs/Kconfig             |  18 ++
>  fs/f2fs/Makefile            |   1 +
>  fs/f2fs/data.c              | 146 +++++++++-
>  fs/f2fs/dir.c               |   1 +
>  fs/f2fs/f2fs.h              |  44 ++-
>  fs/f2fs/file.c              |   1 +
>  fs/f2fs/iextent.c           | 535 ++++++++++++++++++++++++++++++++++++
>  fs/f2fs/iextent.h           | 136 +++++++++
>  fs/f2fs/inline.c            |   1 +
>  fs/f2fs/inode.c             |  41 ++-
>  fs/f2fs/namei.c             |  67 +++++
>  fs/f2fs/node.c              |  38 ++-
>  fs/f2fs/node.h              |   4 +
>  fs/f2fs/recovery.c          |  15 +
>  fs/f2fs/super.c             |  29 ++
>  fs/f2fs/sysfs.c             |  91 ++++++
>  include/linux/f2fs_fs.h     |   3 +-
>  include/trace/events/f2fs.h |  79 ++++++
>  18 files changed, 1229 insertions(+), 21 deletions(-)
>  create mode 100644 fs/f2fs/iextent.c
>  create mode 100644 fs/f2fs/iextent.h
> 

ping

Thanks,
Yongpeng



_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel