Commit d279c80e0bac ("iomap: inline iomap_dio_bio_opflags()") has broken the logic in iomap_dio_bio_iter() in a way that when the device does support FUA (or has no writeback cache) and the direct IO happens to freshly allocated or unwritten extents, we will *not* issue fsync after completing direct IO O_SYNC / O_DSYNC write because the IOMAP_DIO_WRITE_THROUGH flag stays mistakenly set. Fix the problem by clearing IOMAP_DIO_WRITE_THROUGH whenever we do not perform FUA write as it was originally intended.
CC: John Garry john.g.garry@oracle.com CC: "Ritesh Harjani (IBM)" ritesh.list@gmail.com Fixes: d279c80e0bac ("iomap: inline iomap_dio_bio_opflags()") CC: stable@vger.kernel.org Signed-off-by: Jan Kara jack@suse.cz --- fs/iomap/direct-io.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)
BTW, I've spotted this because some performance tests got suspiciously fast on recent kernels :) Sadly no easy improvement to cherry-pick for me to fix customer issue I'm chasing...
diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c index 6f25d4cfea9f..b84f6af2eb4c 100644 --- a/fs/iomap/direct-io.c +++ b/fs/iomap/direct-io.c @@ -363,14 +363,14 @@ static int iomap_dio_bio_iter(struct iomap_iter *iter, struct iomap_dio *dio) if (iomap->flags & IOMAP_F_SHARED) dio->flags |= IOMAP_DIO_COW;
- if (iomap->flags & IOMAP_F_NEW) { + if (iomap->flags & IOMAP_F_NEW) need_zeroout = true; - } else if (iomap->type == IOMAP_MAPPED) { - if (iomap_dio_can_use_fua(iomap, dio)) - bio_opf |= REQ_FUA; - else - dio->flags &= ~IOMAP_DIO_WRITE_THROUGH; - } + else if (iomap->type == IOMAP_MAPPED && + iomap_dio_can_use_fua(iomap, dio)) + bio_opf |= REQ_FUA; + + if (!(bio_opf & REQ_FUA)) + dio->flags &= ~IOMAP_DIO_WRITE_THROUGH;
/* * We can only do deferred completion for pure overwrites that