dm bufio: fix buffer alignment

Commit 6b5e718cc138 ("dm bufio: relax alignment constraint on slab
cache") relaxed alignment on dm-bufio cache, however it may break
dm-crypt or dm-integrity.

dm-crypt and dm-integrity require that the size of bio vector entries
(bv_len) is aligned on its sector size. bv_offset doesn't have to be
aligned, but bv_len must be. XFS sends unaligned bios, but they do not
cross page boundary, so the requirement for aligned bv_len is met.

Commit 6b5e718cc138 made dm-bufio send unaligned bios that cross page
boundary, this could break dm-crypt and dm-integrity.

Reinstates the alignment. Note that misaligned entries only happen when
we use slab/slub debugging. Without debugging, the entries are always
aligned.

Fixes: 6b5e718cc138 ("dm bufio: relax alignment constraint on slab cache")
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c
index 12aa9ca21d8c656166d819011adb00e4e7a5e4e0..dc385b70e4c336fc2322c9538f86538b000a8556 100644 (file)
--- a/drivers/md/dm-bufio.c
+++ b/drivers/md/dm-bufio.c
@@ -1681,8 +1681,9 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign
 
        if (block_size <= KMALLOC_MAX_SIZE &&
            (block_size < PAGE_SIZE || !is_power_of_2(block_size))) {
-               snprintf(slab_name, sizeof slab_name, "dm_bufio_cache-%u", c->block_size);
-               c->slab_cache = kmem_cache_create(slab_name, c->block_size, ARCH_KMALLOC_MINALIGN,
+               unsigned align = min(1U << __ffs(block_size), (unsigned)PAGE_SIZE);
+               snprintf(slab_name, sizeof slab_name, "dm_bufio_cache-%u", block_size);
+               c->slab_cache = kmem_cache_create(slab_name, block_size, align,
                                                  SLAB_RECLAIM_ACCOUNT, NULL);
                if (!c->slab_cache) {
                        r = -ENOMEM;