}
 
        if (behind) {
-               /* we release behind master bio when all write are done */
-               if (r1_bio->behind_master_bio == bio)
-                       to_put = NULL;
-
                if (test_bit(WriteMostly, &rdev->flags))
                        atomic_dec(&r1_bio->behind_remaining);
 
        wake_up(&conf->wait_barrier);
 }
 
-static struct bio *alloc_behind_master_bio(struct r1bio *r1_bio,
+static void alloc_behind_master_bio(struct r1bio *r1_bio,
                                           struct bio *bio)
 {
        int size = bio->bi_iter.bi_size;
 
        behind_bio = bio_alloc_mddev(GFP_NOIO, vcnt, r1_bio->mddev);
        if (!behind_bio)
-               goto fail;
+               return;
 
        /* discard op, we don't support writezero/writesame yet */
-       if (!bio_has_data(bio))
+       if (!bio_has_data(bio)) {
+               behind_bio->bi_iter.bi_size = size;
                goto skip_copy;
+       }
 
        while (i < vcnt && size) {
                struct page *page;
        r1_bio->behind_master_bio = behind_bio;;
        set_bit(R1BIO_BehindIO, &r1_bio->state);
 
-       return behind_bio;
+       return;
 
 free_pages:
        pr_debug("%dB behind alloc failed, doing sync I/O\n",
                 bio->bi_iter.bi_size);
        bio_free_pages(behind_bio);
-fail:
-       return behind_bio;
+       bio_put(behind_bio);
 }
 
 struct raid1_plug_cb {
                            (atomic_read(&bitmap->behind_writes)
                             < mddev->bitmap_info.max_write_behind) &&
                            !waitqueue_active(&bitmap->behind_wait)) {
-                               mbio = alloc_behind_master_bio(r1_bio, bio);
+                               alloc_behind_master_bio(r1_bio, bio);
                        }
 
                        bitmap_startwrite(bitmap, r1_bio->sector,
                        first_clone = 0;
                }
 
-               if (!mbio) {
-                       if (r1_bio->behind_master_bio)
-                               mbio = bio_clone_fast(r1_bio->behind_master_bio,
-                                                     GFP_NOIO,
-                                                     mddev->bio_set);
-                       else
-                               mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set);
-               }
+               if (r1_bio->behind_master_bio)
+                       mbio = bio_clone_fast(r1_bio->behind_master_bio,
+                                             GFP_NOIO, mddev->bio_set);
+               else
+                       mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set);
 
                if (r1_bio->behind_master_bio) {
                        if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
                        wbio = bio_clone_fast(r1_bio->behind_master_bio,
                                              GFP_NOIO,
                                              mddev->bio_set);
-                       /* We really need a _all clone */
-                       wbio->bi_iter = (struct bvec_iter){ 0 };
                } else {
                        wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO,
                                              mddev->bio_set);