[PATCH v4 0/5] exfat: fix valid_size update races

From: Chi Zhiling

Date: Mon Nov 16 2026 - 03:59:41 EST


From: Chi Zhiling <chizhiling@xxxxxxxxxx>

exFAT uses valid_size (ValidDataLength) as the on-disk boundary of
initialised data, but the current code advances it too late and without
proper serialisation. Buffered writes can silently lose data reclaimed
before valid_size covers it, file expansion and page faults can
initialise the same range concurrently, and a failed or short direct
write can still grow the file on disk.

This series advances valid_size under the folio lock, keeps zeroed_size
in sync with the block tails that were actually initialised, and
revalidates cached iomaps for buffered writes. File expansion holds
invalidate_lock across allocation, EOF cleanup and the i_size update.
Direct writes are prepared and written back through their destination
boundary folio before submission, publish i_size and valid_size only
from ->end_io(), and hold invalidate_lock across completion, post-write
invalidation, gap initialisation and rollback so that mmap faults
cannot interfere.

V4:
- Fold the series into five patches and rebase onto current exfat/dev.
- Add a patch that serialises file expansion with mmap write faults.
- Rework direct writes: prepare and write back the invalid range through
the destination boundary folio, initialise the gap after the post-write
invalidation, and keep pure overwrites asynchronous.
- Track initialised block tails in zeroed_size instead of
IOMAP_F_ZERO_TAIL, record the on-disk length in i_size, and finish
short direct reads with filemap_read().
- Pick up the Reviewed-by on the iomap patch.

V3: https://lore.kernel.org/exfat/20261016033349.3104509-1-chizhiling@xxxxxxx/
V2: https://lore.kernel.org/exfat/20261013033837.2748258-1-chizhiling@xxxxxxx/
V1: https://lore.kernel.org/exfat/59056131-e3bd-4003-8c30-0a356cf0bd82@xxxxxxx/
Link: https://lore.kernel.org/exfat/10029d6f-13cc-4d17-a2f9-b093a8339dde/T/#t

Chi Zhiling (3):
iomap: export iomap_bio_read_folio_range_sync()
exfat: increase VDL before unlocking initialized folios
exfat: zero invalid VDL tails in buffered and direct reads

Namjae Jeon (2):
exfat: serialize file expansion with mmap write faults
exfat: coordinate direct writes with mmap faults

fs/exfat/exfat_fs.h | 2 +-
fs/exfat/fatent.c | 6 +-
fs/exfat/file.c | 427 ++++++++++++++++++++++++++++++++----------
fs/exfat/inode.c | 19 +-
fs/exfat/iomap.c | 368 ++++++++++++++++++++++++------------
fs/exfat/iomap.h | 10 +-
fs/iomap/bio.c | 1 +
fs/iomap/internal.h | 12 --
include/linux/iomap.h | 9 +
9 files changed, 599 insertions(+), 255 deletions(-)

--
2.53.0