f2fs: fix data loss caused by incorrect use of nat_entry flag

Data loss can occur when fsync is performed on a newly created file
(before any checkpoint has been written) concurrently with a checkpoint
operation. The scenario is as follows:

create & write & fsync 'file A'                 write checkpoint
- f2fs_do_sync_file // inline inode
 - f2fs_write_inode // inode folio is dirty
                                                - f2fs_write_checkpoint
                                                 - f2fs_flush_merged_writes
                                                 - f2fs_sync_node_pages
                                                 - f2fs_flush_nat_entries
 - f2fs_fsync_node_pages // no dirty node
 - f2fs_need_inode_block_update // return false
 SPO and lost 'file A'

f2fs_flush_nat_entries() sets the IS_CHECKPOINTED and HAS_LAST_FSYNC
flags for the nat_entry, but this does not mean that the checkpoint has
actually completed successfully. However, f2fs_need_inode_block_update()
checks these flags and incorrectly assumes that the checkpoint has
finished.

The root cause is that the semantics of IS_CHECKPOINTED and
HAS_LAST_FSYNC are only guaranteed after the checkpoint write fully
completes.

This patch modifies f2fs_need_inode_block_update() to acquire the
sbi->node_write lock before reading the nat_entry flags, ensuring that
once IS_CHECKPOINTED and HAS_LAST_FSYNC are observed to be set, the
checkpoint operation has already completed.

Fixes: e05df3b115 ("f2fs: add node operations")
Signed-off-by: Yongpeng Yang <yangyongpeng@xiaomi.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This commit is contained in:
Yongpeng Yang 2026-03-10 17:36:14 +08:00 • committed by Jaegeuk Kim
commit 238e14eb72