fuse: zero the correct range on a short read reply

When a read reply is shorter than the requested range, fuse_copy_folio()
zeroes everything in the folio outside the copied bytes. However, for
folios with block sizes smaller than the folio size, this can clear
blocks outside the requested range which are uptodate, or dirty and have
not yet been written back.

Move the zeroing logic into fuse_copy_folios(), since descs[i].length
needs to be used, and clear only the tail of the requested range that
the server did not send.

The range is also no longer cleared upfront in the non
cs->skip_folio_copy case. This is fine since a failed copy leaves those
blocks non-uptodate, so the uncopied bytes are never visible. This also
matches the pre-existing behavior in the non-short-read case (a failed
copy doesn't zero out the range in the folio). This lets the zeroing
logic stay the same regardless of whether the folio copy is skipped or
not.

Fixes: a4c9ab1d4975 ("fuse: use iomap for buffered writes")
Reported-by: Sashiko <sashiko-bot@kernel.org>
Reported-by: Kanishka De Silva <kpskanna1915@gmail.com>
Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

0 files changed, 0 insertions, 0 deletions