drbd: wait for meta data IO completion even with failed disk, unless force-detached

The intention of force-detach is to be able to deal with a completely
unresponsive lower level IO stack, which does not even deliver error
completions anymore, but no completion at all.

In all other cases, we must still wait for the meta data IO completion.
Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>

drivers/block/drbd/drbd_actlog.c

@@ -123,13 +123,7 @@ void drbd_md_put_buffer(struct drbd_conf *mdev)
 		wake_up(&mdev->misc_wait);
 }
 
-static bool md_io_allowed(struct drbd_conf *mdev)
-{
-	enum drbd_disk_state ds = mdev->state.disk;
-	return ds >= D_NEGOTIATING || ds == D_ATTACHING;
-}
-
-void wait_until_done_or_disk_failure(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
+void wait_until_done_or_force_detached(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 				     unsigned int *done)
 {
 	long dt;
@@ -141,9 +135,12 @@ void wait_until_done_or_disk_failure(struct drbd_conf *mdev, struct drbd_backing
 	if (dt == 0)
 		dt = MAX_SCHEDULE_TIMEOUT;
 
-	dt = wait_event_timeout(mdev->misc_wait, *done || !md_io_allowed(mdev), dt);
-	if (dt == 0)
+	dt = wait_event_timeout(mdev->misc_wait,
+			*done || test_bit(FORCE_DETACH, &mdev->flags), dt);
+	if (dt == 0) {
 		dev_err(DEV, "meta-data IO operation timed out\n");
+		drbd_chk_io_error(mdev, 1, DRBD_FORCE_DETACH);
+	}
 }
 
 static int _drbd_md_sync_page_io(struct drbd_conf *mdev,
@@ -183,7 +180,7 @@ static int _drbd_md_sync_page_io(struct drbd_conf *mdev,
 		bio_endio(bio, -EIO);
 	else
 		submit_bio(rw, bio);
-	wait_until_done_or_disk_failure(mdev, bdev, &mdev->md_io.done);
+	wait_until_done_or_force_detached(mdev, bdev, &mdev->md_io.done);
 	if (bio_flagged(bio, BIO_UPTODATE))
 		err = mdev->md_io.error;
 

drivers/block/drbd/drbd_bitmap.c

@@ -1128,7 +1128,7 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_w
 	 * "in_flight reached zero, all done" event.
 	 */
 	if (!atomic_dec_and_test(&ctx->in_flight))
-		wait_until_done_or_disk_failure(mdev, mdev->ldev, &ctx->done);
+		wait_until_done_or_force_detached(mdev, mdev->ldev, &ctx->done);
 	else
 		kref_put(&ctx->kref, &bm_aio_ctx_destroy);
 
@@ -1145,7 +1145,7 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_w
 	}
 
 	if (atomic_read(&ctx->in_flight))
-		err = -EIO; /* Disk failed during IO... */
+		err = -EIO; /* Disk timeout/force-detach during IO... */
 
 	now = jiffies;
 	if (rw == WRITE) {
@@ -1273,11 +1273,11 @@ int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(loc
 	}
 
 	bm_page_io_async(ctx, idx, WRITE_SYNC);
-	wait_until_done_or_disk_failure(mdev, mdev->ldev, &ctx->done);
+	wait_until_done_or_force_detached(mdev, mdev->ldev, &ctx->done);
 
 	if (ctx->error)
 		drbd_chk_io_error(mdev, 1, DRBD_META_IO_ERROR);
-		/* that should force detach, so the in memory bitmap will be
+		/* that causes us to detach, so the in memory bitmap will be
 		 * gone in a moment as well. */
 
 	mdev->bm_writ_cnt++;

drivers/block/drbd/drbd_int.h

@@ -1452,8 +1452,8 @@ extern void drbd_md_put_buffer(struct drbd_conf *mdev);
 extern int drbd_md_sync_page_io(struct drbd_conf *mdev,
 		struct drbd_backing_dev *bdev, sector_t sector, int rw);
 extern void drbd_ov_out_of_sync_found(struct drbd_conf *, sector_t, int);
-extern void wait_until_done_or_disk_failure(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
-					    unsigned int *done);
+extern void wait_until_done_or_force_detached(struct drbd_conf *mdev,
+		struct drbd_backing_dev *bdev, unsigned int *done);
 extern void drbd_rs_controller_reset(struct drbd_conf *mdev);
 
 static inline void ov_out_of_sync_print(struct drbd_conf *mdev)