md/raid1: fix data lost for writemostly rdev

[ Upstream commit 93dec51e716db88f32d770dc9ab268964fff320b ]

If writemostly is enabled, alloc_behind_master_bio() will allocate a new
bio for rdev, with bi_opf set to 0. Later, raid1_write_request() will
clone from this bio, hence bi_opf is still 0 for the cloned bio. Submit
this cloned bio will end up to be read, causing write data lost.

Fix this problem by inheriting bi_opf from original bio for
behind_mast_bio.

Fixes: e879a0d9cb08 ("md/raid1,raid10: don't ignore IO flags")
Reported-and-tested-by: Ian Dall <ian@beware.dropbear.id.au>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220507
Link: https://lore.kernel.org/linux-raid/20250903014140.3690499-1-yukuai1@huaweicloud.com
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Li Nan <linan122@huawei.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
index 6cee738a645ff22c5542f77765c77e294cc3ac70..6d52336057946e1fab59a0ab812ea1979b5b6463 100644 (file)
--- a/drivers/md/raid1.c
+++ b/drivers/md/raid1.c
@@ -1226,7 +1226,7 @@ static void alloc_behind_master_bio(struct r1bio *r1_bio,
        int i = 0;
        struct bio *behind_bio = NULL;
 
-       behind_bio = bio_alloc_bioset(NULL, vcnt, 0, GFP_NOIO,
+       behind_bio = bio_alloc_bioset(NULL, vcnt, bio->bi_opf, GFP_NOIO,
                                      &r1_bio->mddev->bio_set);
 
        /* discard op, we don't support writezero/writesame yet */