[PATCH 0/2] ksmbd: fix SetAllocation EOF and rounding

[DaeMyung Kang]

In response to Namjae's review of my earlier SetAllocation patch, this
series separates two fixes. Patch 1/2 prevents the existing cleanup
branch from extending EOF on disk and then restoring only the in-memory
size. It carries Fixes and Cc: stable. Patch 2/2 rounds the request to
f_bsize, the allocation unit ksmbd advertises in FS_SIZE_INFORMATION,
instead of 512 bytes.

The other review point, reserving the requested range when shrinking,
remains open. I tried fallocate before truncate, but the decision to
shrink uses an EOF value read before vfs_truncate() takes inode_lock.
The existing block-count branch already has this race; extending
truncate to additional sparse-file cases would expose it more widely.
If another client truncates and writes in that interval, our truncate
can extend the file again with zeros. I reproduced this with an
injected delay in QEMU. Rechecking EOF after fallocate narrows the
window but does not close it.

A VFS operation that checks EOF and truncates under the same inode_lock
without extending the file may be needed. I plan to discuss the
reservation fix separately with linux-fsdevel and would welcome your
thoughts on that approach.

DaeMyung Kang (2):
  ksmbd: avoid restoring only in-memory EOF after allocation change
  ksmbd: round SetAllocation to the advertised allocation unit

 fs/smb/server/smb2pdu.c | 48 ++++++++++++++++++++---------------------
 1 file changed, 23 insertions(+), 25 deletions(-)


base-commit: bf3465919b0337dfef8798da01469145887c4ef5
-- 
2.43.0