On 11/08/2018 10:16 PM, Nikolay Borisov wrote:
Before btrfs_map_bio submits all stripe bio it does a number of checks to ensure the device for every stripe is present. However, it doesn't do a DEV_STATE_MISSING check, instead this is relegated to the lower level btrfs_schedule_bio (in the async submission case, sync submission doesn't check DEV_STATE_MISSING at all). Additionally btrfs_schedule_bios does the duplicate device->bdev check which has already been performed in btrfs_map_bio. This patch moves the DEV_STATE_MISSING check in btrfs_map_bio and removes the duplicate device->bdev check. Doing so ensures that no bio cloning/submission happens for both async/sync requests in the face of missing device. This makes the async io submission path slightly shorter in terms of instruction count. No functional changes. Signed-off-by: Nikolay Borisov <nbori...@suse.com>
Reviewed-by: Anand Jain <anand.j...@oracle.com> Thanks, Anand
--- fs/btrfs/volumes.c | 9 ++------- 1 file changed, 2 insertions(+), 7 deletions(-) diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 44c5e8ccb644..3312cad65209 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -6106,12 +6106,6 @@ static noinline void btrfs_schedule_bio(struct btrfs_device *device, int should_queue = 1; struct btrfs_pending_bios *pending_bios;- if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||- !device->bdev) { - bio_io_error(bio); - return; - } - /* don't bother with additional async steps for reads, right now */ if (bio_op(bio) == REQ_OP_READ) { btrfsic_submit_bio(bio); @@ -6240,7 +6234,8 @@ blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {dev = bbio->stripes[dev_nr].dev; - if (!dev || !dev->bdev || + if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, + &dev->dev_state) || (bio_op(first_bio) == REQ_OP_WRITE && !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { bbio_error(bbio, first_bio, logical);
