1 files changed, 299 insertions, 238 deletions

diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index 55745d6953f0..fd0cdced8024 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -626,9 +626,8 @@ void rbd_warn(struct rbd_device *rbd_dev, const char *fmt, ...)
 static void rbd_dev_remove_parent(struct rbd_device *rbd_dev);
 
 static int rbd_dev_refresh(struct rbd_device *rbd_dev);
-static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev);
-static int rbd_dev_header_info(struct rbd_device *rbd_dev);
-static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev);
+static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev,
+				     struct rbd_image_header *header);
 static const char *rbd_dev_v2_snap_name(struct rbd_device *rbd_dev,
 					u64 snap_id);
 static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
@@ -1098,15 +1097,24 @@ static void rbd_init_layout(struct rbd_device *rbd_dev)
 	RCU_INIT_POINTER(rbd_dev->layout.pool_ns, NULL);
 }
 
+static void rbd_image_header_cleanup(struct rbd_image_header *header)
+{
+	kfree(header->object_prefix);
+	ceph_put_snap_context(header->snapc);
+	kfree(header->snap_sizes);
+	kfree(header->snap_names);
+
+	memset(header, 0, sizeof(*header));
+}
+
 /*
  * Fill an rbd image header with information from the given format 1
  * on-disk header.
  */
-static int rbd_header_from_disk(struct rbd_device *rbd_dev,
-				 struct rbd_image_header_ondisk *ondisk)
+static int rbd_header_from_disk(struct rbd_image_header *header,
+				struct rbd_image_header_ondisk *ondisk,
+				bool first_time)
 {
-	struct rbd_image_header *header = &rbd_dev->header;
-	bool first_time = header->object_prefix == NULL;
 	struct ceph_snap_context *snapc;
 	char *object_prefix = NULL;
 	char *snap_names = NULL;
@@ -1173,11 +1181,6 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,
 	if (first_time) {
 		header->object_prefix = object_prefix;
 		header->obj_order = ondisk->options.order;
-		rbd_init_layout(rbd_dev);
-	} else {
-		ceph_put_snap_context(header->snapc);
-		kfree(header->snap_names);
-		kfree(header->snap_sizes);
 	}
 
 	/* The remaining fields always get updated (when we refresh) */
@@ -1493,14 +1496,30 @@ static bool rbd_obj_is_tail(struct rbd_obj_request *obj_req)
 /*
  * Must be called after rbd_obj_calc_img_extents().
  */
-static bool rbd_obj_copyup_enabled(struct rbd_obj_request *obj_req)
+static void rbd_obj_set_copyup_enabled(struct rbd_obj_request *obj_req)
 {
-	if (!obj_req->num_img_extents ||
-	    (rbd_obj_is_entire(obj_req) &&
-	     !obj_req->img_request->snapc->num_snaps))
-		return false;
+	rbd_assert(obj_req->img_request->snapc);
 
-	return true;
+	if (obj_req->img_request->op_type == OBJ_OP_DISCARD) {
+		dout("%s %p objno %llu discard\n", __func__, obj_req,
+		     obj_req->ex.oe_objno);
+		return;
+	}
+
+	if (!obj_req->num_img_extents) {
+		dout("%s %p objno %llu not overlapping\n", __func__, obj_req,
+		     obj_req->ex.oe_objno);
+		return;
+	}
+
+	if (rbd_obj_is_entire(obj_req) &&
+	    !obj_req->img_request->snapc->num_snaps) {
+		dout("%s %p objno %llu entire\n", __func__, obj_req,
+		     obj_req->ex.oe_objno);
+		return;
+	}
+
+	obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
 }
 
 static u64 rbd_obj_img_extents_bytes(struct rbd_obj_request *obj_req)
@@ -1599,6 +1618,7 @@ __rbd_obj_add_osd_request(struct rbd_obj_request *obj_req,
 static struct ceph_osd_request *
 rbd_obj_add_osd_request(struct rbd_obj_request *obj_req, int num_ops)
 {
+	rbd_assert(obj_req->img_request->snapc);
 	return __rbd_obj_add_osd_request(obj_req, obj_req->img_request->snapc,
 					 num_ops);
 }
@@ -1727,11 +1747,14 @@ static bool rbd_dev_parent_get(struct rbd_device *rbd_dev)
  * Caller is responsible for filling in the list of object requests
  * that comprises the image request, and the Linux request pointer
  * (if there is one).
+ *
+ * Only snap_id is captured here, for reads.  For writes, snapshot
+ * context is captured in rbd_img_object_requests() after exclusive
+ * lock is ensured to be held.
  */
 static struct rbd_img_request *rbd_img_request_create(
 					struct rbd_device *rbd_dev,
-					enum obj_operation_type op_type,
-					struct ceph_snap_context *snapc)
+					enum obj_operation_type op_type)
 {
 	struct rbd_img_request *img_request;
 
@@ -1743,8 +1766,6 @@ static struct rbd_img_request *rbd_img_request_create(
 	img_request->op_type = op_type;
 	if (!rbd_img_is_write(img_request))
 		img_request->snap_id = rbd_dev->spec->snap_id;
-	else
-		img_request->snapc = snapc;
 
 	if (rbd_dev_parent_get(rbd_dev))
 		img_request_layered_set(img_request);
@@ -2087,7 +2108,7 @@ static int rbd_object_map_update_finish(struct rbd_obj_request *obj_req,
 	struct rbd_device *rbd_dev = obj_req->img_request->rbd_dev;
 	struct ceph_osd_data *osd_data;
 	u64 objno;
-	u8 state, new_state, uninitialized_var(current_state);
+	u8 state, new_state, current_state;
 	bool has_current_state;
 	void *p;
 
@@ -2389,9 +2410,6 @@ static int rbd_obj_init_write(struct rbd_obj_request *obj_req)
 	if (ret)
 		return ret;
 
-	if (rbd_obj_copyup_enabled(obj_req))
-		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
-
 	obj_req->write_state = RBD_OBJ_WRITE_START;
 	return 0;
 }
@@ -2497,8 +2515,6 @@ static int rbd_obj_init_zeroout(struct rbd_obj_request *obj_req)
 	if (ret)
 		return ret;
 
-	if (rbd_obj_copyup_enabled(obj_req))
-		obj_req->flags |= RBD_OBJ_FLAG_COPYUP_ENABLED;
 	if (!obj_req->num_img_extents) {
 		obj_req->flags |= RBD_OBJ_FLAG_NOOP_FOR_NONEXISTENT;
 		if (rbd_obj_is_entire(obj_req))
@@ -2935,7 +2951,7 @@ static int rbd_obj_read_from_parent(struct rbd_obj_request *obj_req)
 	int ret;
 
 	child_img_req = rbd_img_request_create(img_req->rbd_dev->parent,
-					       OBJ_OP_READ, NULL);
+					       OBJ_OP_READ);
 	if (!child_img_req)
 		return -ENOMEM;
 
@@ -3439,6 +3455,7 @@ again:
 	case RBD_OBJ_WRITE_START:
 		rbd_assert(!*result);
 
+		rbd_obj_set_copyup_enabled(obj_req);
 		if (rbd_obj_write_is_noop(obj_req))
 			return true;
 
@@ -3586,14 +3603,15 @@ static bool rbd_lock_add_request(struct rbd_img_request *img_req)
 static void rbd_lock_del_request(struct rbd_img_request *img_req)
 {
 	struct rbd_device *rbd_dev = img_req->rbd_dev;
-	bool need_wakeup;
+	bool need_wakeup = false;
 
 	lockdep_assert_held(&rbd_dev->lock_rwsem);
 	spin_lock(&rbd_dev->lock_lists_lock);
-	rbd_assert(!list_empty(&img_req->lock_item));
-	list_del_init(&img_req->lock_item);
-	need_wakeup = (rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING &&
-		       list_empty(&rbd_dev->running_list));
+	if (!list_empty(&img_req->lock_item)) {
+		list_del_init(&img_req->lock_item);
+		need_wakeup = (rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING &&
+			       list_empty(&rbd_dev->running_list));
+	}
 	spin_unlock(&rbd_dev->lock_lists_lock);
 	if (need_wakeup)
 		complete(&rbd_dev->releasing_wait);
@@ -3625,9 +3643,19 @@ static int rbd_img_exclusive_lock(struct rbd_img_request *img_req)
 
 static void rbd_img_object_requests(struct rbd_img_request *img_req)
 {
+	struct rbd_device *rbd_dev = img_req->rbd_dev;
 	struct rbd_obj_request *obj_req;
 
 	rbd_assert(!img_req->pending.result && !img_req->pending.num_pending);
+	rbd_assert(!need_exclusive_lock(img_req) ||
+		   __rbd_is_lock_owner(rbd_dev));
+
+	if (rbd_img_is_write(img_req)) {
+		rbd_assert(!img_req->snapc);
+		down_read(&rbd_dev->header_rwsem);
+		img_req->snapc = ceph_get_snap_context(rbd_dev->header.snapc);
+		up_read(&rbd_dev->header_rwsem);
+	}
 
 	for_each_obj_request(img_req, obj_req) {
 		int result = 0;
@@ -3645,7 +3673,6 @@ static void rbd_img_object_requests(struct rbd_img_request *img_req)
 
 static bool rbd_img_advance(struct rbd_img_request *img_req, int *result)
 {
-	struct rbd_device *rbd_dev = img_req->rbd_dev;
 	int ret;
 
 again:
@@ -3666,9 +3693,6 @@ again:
 		if (*result)
 			return true;
 
-		rbd_assert(!need_exclusive_lock(img_req) ||
-			   __rbd_is_lock_owner(rbd_dev));
-
 		rbd_img_object_requests(img_req);
 		if (!img_req->pending.num_pending) {
 			*result = img_req->pending.result;
@@ -3974,14 +3998,19 @@ static void wake_lock_waiters(struct rbd_device *rbd_dev, int result)
 		return;
 	}
 
-	list_for_each_entry(img_req, &rbd_dev->acquiring_list, lock_item) {
+	while (!list_empty(&rbd_dev->acquiring_list)) {
+		img_req = list_first_entry(&rbd_dev->acquiring_list,
+					   struct rbd_img_request, lock_item);
 		mutex_lock(&img_req->state_mutex);
 		rbd_assert(img_req->state == RBD_IMG_EXCLUSIVE_LOCK);
+		if (!result)
+			list_move_tail(&img_req->lock_item,
+				       &rbd_dev->running_list);
+		else
+			list_del_init(&img_req->lock_item);
 		rbd_img_schedule(img_req, result);
 		mutex_unlock(&img_req->state_mutex);
 	}
-
-	list_splice_tail_init(&rbd_dev->acquiring_list, &rbd_dev->running_list);
 }
 
 static int get_lock_owner_info(struct rbd_device *rbd_dev,
@@ -4130,6 +4159,10 @@ static int rbd_post_acquire_action(struct rbd_device *rbd_dev)
 {
 	int ret;
 
+	ret = rbd_dev_refresh(rbd_dev);
+	if (ret)
+		return ret;
+
 	if (rbd_dev->header.features & RBD_FEATURE_OBJECT_MAP) {
 		ret = rbd_object_map_open(rbd_dev);
 		if (ret)
@@ -4788,7 +4821,6 @@ static void rbd_queue_workfn(struct work_struct *work)
 	struct request *rq = blk_mq_rq_from_pdu(work);
 	struct rbd_device *rbd_dev = rq->q->queuedata;
 	struct rbd_img_request *img_request;
-	struct ceph_snap_context *snapc = NULL;
 	u64 offset = (u64)blk_rq_pos(rq) << SECTOR_SHIFT;
 	u64 length = blk_rq_bytes(rq);
 	enum obj_operation_type op_type;
@@ -4853,10 +4885,6 @@ static void rbd_queue_workfn(struct work_struct *work)
 
 	down_read(&rbd_dev->header_rwsem);
 	mapping_size = rbd_dev->mapping.size;
-	if (op_type != OBJ_OP_READ) {
-		snapc = rbd_dev->header.snapc;
-		ceph_get_snap_context(snapc);
-	}
 	up_read(&rbd_dev->header_rwsem);
 
 	if (offset + length > mapping_size) {
@@ -4866,13 +4894,12 @@ static void rbd_queue_workfn(struct work_struct *work)
 		goto err_rq;
 	}
 
-	img_request = rbd_img_request_create(rbd_dev, op_type, snapc);
+	img_request = rbd_img_request_create(rbd_dev, op_type);
 	if (!img_request) {
 		result = -ENOMEM;
 		goto err_rq;
 	}
 	img_request->rq = rq;
-	snapc = NULL; /* img_request consumes a ref */
 
 	dout("%s rbd_dev %p img_req %p %s %llu~%llu\n", __func__, rbd_dev,
 	     img_request, obj_op_name(op_type), offset, length);
@@ -4894,7 +4921,6 @@ err_rq:
 	if (result)
 		rbd_warn(rbd_dev, "%s %llx at %llx result %d",
 			 obj_op_name(op_type), length, offset, result);
-	ceph_put_snap_context(snapc);
 err:
 	blk_mq_end_request(rq, errno_to_blk_status(result));
 }
@@ -4966,7 +4992,9 @@ out_req:
  * return, the rbd_dev->header field will contain up-to-date
  * information about the image.
  */
-static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
+static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev,
+				  struct rbd_image_header *header,
+				  bool first_time)
 {
 	struct rbd_image_header_ondisk *ondisk = NULL;
 	u32 snap_count = 0;
@@ -5014,7 +5042,7 @@ static int rbd_dev_v1_header_info(struct rbd_device *rbd_dev)
 		snap_count = le32_to_cpu(ondisk->snap_count);
 	} while (snap_count != want_count);
 
-	ret = rbd_header_from_disk(rbd_dev, ondisk);
+	ret = rbd_header_from_disk(header, ondisk, first_time);
 out:
 	kfree(ondisk);
 
@@ -5058,43 +5086,6 @@ static void rbd_dev_update_size(struct rbd_device *rbd_dev)
 	}
 }
 
-static int rbd_dev_refresh(struct rbd_device *rbd_dev)
-{
-	u64 mapping_size;
-	int ret;
-
-	down_write(&rbd_dev->header_rwsem);
-	mapping_size = rbd_dev->mapping.size;
-
-	ret = rbd_dev_header_info(rbd_dev);
-	if (ret)
-		goto out;
-
-	/*
-	 * If there is a parent, see if it has disappeared due to the
-	 * mapped image getting flattened.
-	 */
-	if (rbd_dev->parent) {
-		ret = rbd_dev_v2_parent_info(rbd_dev);
-		if (ret)
-			goto out;
-	}
-
-	if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
-		rbd_dev->mapping.size = rbd_dev->header.image_size;
-	} else {
-		/* validate mapped snapshot's EXISTS flag */
-		rbd_exists_validate(rbd_dev);
-	}
-
-out:
-	up_write(&rbd_dev->header_rwsem);
-	if (!ret && mapping_size != rbd_dev->mapping.size)
-		rbd_dev_update_size(rbd_dev);
-
-	return ret;
-}
-
 static int rbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
 		unsigned int hctx_idx, unsigned int numa_node)
 {
@@ -5529,8 +5520,7 @@ static void rbd_dev_release(struct device *dev)
 		module_put(THIS_MODULE);
 }
 
-static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
-					   struct rbd_spec *spec)
+static struct rbd_device *__rbd_dev_create(struct rbd_spec *spec)
 {
 	struct rbd_device *rbd_dev;
 
@@ -5575,9 +5565,6 @@ static struct rbd_device *__rbd_dev_create(struct rbd_client *rbdc,
 	rbd_dev->dev.parent = &rbd_root_dev;
 	device_initialize(&rbd_dev->dev);
 
-	rbd_dev->rbd_client = rbdc;
-	rbd_dev->spec = spec;
-
 	return rbd_dev;
 }
 
@@ -5590,12 +5577,10 @@ static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
 {
 	struct rbd_device *rbd_dev;
 
-	rbd_dev = __rbd_dev_create(rbdc, spec);
+	rbd_dev = __rbd_dev_create(spec);
 	if (!rbd_dev)
 		return NULL;
 
-	rbd_dev->opts = opts;
-
 	/* get an id and fill in device name */
 	rbd_dev->dev_id = ida_simple_get(&rbd_dev_id_ida, 0,
 					 minor_to_rbd_dev_id(1 << MINORBITS),
@@ -5612,6 +5597,10 @@ static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
 	/* we have a ref from do_rbd_add() */
 	__module_get(THIS_MODULE);
 
+	rbd_dev->rbd_client = rbdc;
+	rbd_dev->spec = spec;
+	rbd_dev->opts = opts;
+
 	dout("%s rbd_dev %p dev_id %d\n", __func__, rbd_dev, rbd_dev->dev_id);
 	return rbd_dev;
 
@@ -5666,17 +5655,12 @@ static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
 	return 0;
 }
 
-static int rbd_dev_v2_image_size(struct rbd_device *rbd_dev)
-{
-	return _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
-					&rbd_dev->header.obj_order,
-					&rbd_dev->header.image_size);
-}
-
-static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
+static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev,
+				    char **pobject_prefix)
 {
 	size_t size;
 	void *reply_buf;
+	char *object_prefix;
 	int ret;
 	void *p;
 
@@ -5694,16 +5678,16 @@ static int rbd_dev_v2_object_prefix(struct rbd_device *rbd_dev)
 		goto out;
 
 	p = reply_buf;
-	rbd_dev->header.object_prefix = ceph_extract_encoded_string(&p,
-						p + ret, NULL, GFP_NOIO);
+	object_prefix = ceph_extract_encoded_string(&p, p + ret, NULL,
+						    GFP_NOIO);
+	if (IS_ERR(object_prefix)) {
+		ret = PTR_ERR(object_prefix);
+		goto out;
+	}
 	ret = 0;
 
-	if (IS_ERR(rbd_dev->header.object_prefix)) {
-		ret = PTR_ERR(rbd_dev->header.object_prefix);
-		rbd_dev->header.object_prefix = NULL;
-	} else {
-		dout("  object_prefix = %s\n", rbd_dev->header.object_prefix);
-	}
+	*pobject_prefix = object_prefix;
+	dout("  object_prefix = %s\n", object_prefix);
 out:
 	kfree(reply_buf);
 
@@ -5748,12 +5732,6 @@ static int _rbd_dev_v2_snap_features(struct rbd_device *rbd_dev, u64 snap_id,
 	return 0;
 }
 
-static int rbd_dev_v2_features(struct rbd_device *rbd_dev)
-{
-	return _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
-						&rbd_dev->header.features);
-}
-
 /*
  * These are generic image flags, but since they are used only for
  * object map, store them in rbd_dev->object_map_flags.
@@ -5790,6 +5768,14 @@ struct parent_image_info {
 	u64		overlap;
 };
 
+static void rbd_parent_info_cleanup(struct parent_image_info *pii)
+{
+	kfree(pii->pool_ns);
+	kfree(pii->image_id);
+
+	memset(pii, 0, sizeof(*pii));
+}
+
 /*
  * The caller is responsible for @pii.
  */
@@ -5859,6 +5845,9 @@ static int __get_parent_info(struct rbd_device *rbd_dev,
 	if (pii->has_overlap)
 		ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
 
+	dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n",
+	     __func__, pii->pool_id, pii->pool_ns, pii->image_id, pii->snap_id,
+	     pii->has_overlap, pii->overlap);
 	return 0;
 
 e_inval:
@@ -5897,14 +5886,17 @@ static int __get_parent_info_legacy(struct rbd_device *rbd_dev,
 	pii->has_overlap = true;
 	ceph_decode_64_safe(&p, end, pii->overlap, e_inval);
 
+	dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n",
+	     __func__, pii->pool_id, pii->pool_ns, pii->image_id, pii->snap_id,
+	     pii->has_overlap, pii->overlap);
 	return 0;
 
 e_inval:
 	return -EINVAL;
 }
 
-static int get_parent_info(struct rbd_device *rbd_dev,
-			   struct parent_image_info *pii)
+static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev,
+				  struct parent_image_info *pii)
 {
 	struct page *req_page, *reply_page;
 	void *p;
@@ -5932,7 +5924,7 @@ static int get_parent_info(struct rbd_device *rbd_dev,
 	return ret;
 }
 
-static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
+static int rbd_dev_setup_parent(struct rbd_device *rbd_dev)
 {
 	struct rbd_spec *parent_spec;
 	struct parent_image_info pii = { 0 };
@@ -5942,37 +5934,12 @@ static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
 	if (!parent_spec)
 		return -ENOMEM;
 
-	ret = get_parent_info(rbd_dev, &pii);
+	ret = rbd_dev_v2_parent_info(rbd_dev, &pii);
 	if (ret)
 		goto out_err;
 
-	dout("%s pool_id %llu pool_ns %s image_id %s snap_id %llu has_overlap %d overlap %llu\n",
-	     __func__, pii.pool_id, pii.pool_ns, pii.image_id, pii.snap_id,
-	     pii.has_overlap, pii.overlap);
-
-	if (pii.pool_id == CEPH_NOPOOL || !pii.has_overlap) {
-		/*
-		 * Either the parent never existed, or we have
-		 * record of it but the image got flattened so it no
-		 * longer has a parent.  When the parent of a
-		 * layered image disappears we immediately set the
-		 * overlap to 0.  The effect of this is that all new
-		 * requests will be treated as if the image had no
-		 * parent.
-		 *
-		 * If !pii.has_overlap, the parent image spec is not
-		 * applicable.  It's there to avoid duplication in each
-		 * snapshot record.
-		 */
-		if (rbd_dev->parent_overlap) {
-			rbd_dev->parent_overlap = 0;
-			rbd_dev_parent_put(rbd_dev);
-			pr_info("%s: clone image has been flattened\n",
-				rbd_dev->disk->disk_name);
-		}
-
+	if (pii.pool_id == CEPH_NOPOOL || !pii.has_overlap)
 		goto out;	/* No parent?  No problem. */
-	}
 
 	/* The ceph file layout needs to fit pool id in 32 bits */
 
@@ -5984,58 +5951,46 @@ static int rbd_dev_v2_parent_info(struct rbd_device *rbd_dev)
 	}
 
 	/*
-	 * The parent won't change (except when the clone is
-	 * flattened, already handled that).  So we only need to
-	 * record the parent spec we have not already done so.
+	 * The parent won't change except when the clone is flattened,
+	 * so we only need to record the parent image spec once.
 	 */
-	if (!rbd_dev->parent_spec) {
-		parent_spec->pool_id = pii.pool_id;
-		if (pii.pool_ns && *pii.pool_ns) {
-			parent_spec->pool_ns = pii.pool_ns;
-			pii.pool_ns = NULL;
-		}
-		parent_spec->image_id = pii.image_id;
-		pii.image_id = NULL;
-		parent_spec->snap_id = pii.snap_id;
-
-		rbd_dev->parent_spec = parent_spec;
-		parent_spec = NULL;	/* rbd_dev now owns this */
+	parent_spec->pool_id = pii.pool_id;
+	if (pii.pool_ns && *pii.pool_ns) {
+		parent_spec->pool_ns = pii.pool_ns;
+		pii.pool_ns = NULL;
 	}
+	parent_spec->image_id = pii.image_id;
+	pii.image_id = NULL;
+	parent_spec->snap_id = pii.snap_id;
+
+	rbd_assert(!rbd_dev->parent_spec);
+	rbd_dev->parent_spec = parent_spec;
+	parent_spec = NULL;	/* rbd_dev now owns this */
 
 	/*
-	 * We always update the parent overlap.  If it's zero we issue
-	 * a warning, as we will proceed as if there was no parent.
+	 * Record the parent overlap.  If it's zero, issue a warning as
+	 * we will proceed as if there is no parent.
 	 */
-	if (!pii.overlap) {
-		if (parent_spec) {
-			/* refresh, careful to warn just once */
-			if (rbd_dev->parent_overlap)
-				rbd_warn(rbd_dev,
-				    "clone now standalone (overlap became 0)");
-		} else {
-			/* initial probe */
-			rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
-		}
-	}
+	if (!pii.overlap)
+		rbd_warn(rbd_dev, "clone is standalone (overlap 0)");
 	rbd_dev->parent_overlap = pii.overlap;
 
 out:
 	ret = 0;
 out_err:
-	kfree(pii.pool_ns);
-	kfree(pii.image_id);
+	rbd_parent_info_cleanup(&pii);
 	rbd_spec_put(parent_spec);
 	return ret;
 }
 
-static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
+static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev,
+				    u64 *stripe_unit, u64 *stripe_count)
 {
 	struct {
 		__le64 stripe_unit;
 		__le64 stripe_count;
 	} __attribute__ ((packed)) striping_info_buf = { 0 };
 	size_t size = sizeof (striping_info_buf);
-	void *p;
 	int ret;
 
 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
@@ -6047,27 +6002,33 @@ static int rbd_dev_v2_striping_info(struct rbd_device *rbd_dev)
 	if (ret < size)
 		return -ERANGE;
 
-	p = &striping_info_buf;
-	rbd_dev->header.stripe_unit = ceph_decode_64(&p);
-	rbd_dev->header.stripe_count = ceph_decode_64(&p);
+	*stripe_unit = le64_to_cpu(striping_info_buf.stripe_unit);
+	*stripe_count = le64_to_cpu(striping_info_buf.stripe_count);
+	dout("  stripe_unit = %llu stripe_count = %llu\n", *stripe_unit,
+	     *stripe_count);
+
 	return 0;
 }
 
-static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev)
+static int rbd_dev_v2_data_pool(struct rbd_device *rbd_dev, s64 *data_pool_id)
 {
-	__le64 data_pool_id;
+	__le64 data_pool_buf;
 	int ret;
 
 	ret = rbd_obj_method_sync(rbd_dev, &rbd_dev->header_oid,
 				  &rbd_dev->header_oloc, "get_data_pool",
-				  NULL, 0, &data_pool_id, sizeof(data_pool_id));
+				  NULL, 0, &data_pool_buf,
+				  sizeof(data_pool_buf));
+	dout("%s: rbd_obj_method_sync returned %d\n", __func__, ret);
 	if (ret < 0)
 		return ret;
-	if (ret < sizeof(data_pool_id))
+	if (ret < sizeof(data_pool_buf))
 		return -EBADMSG;
 
-	rbd_dev->header.data_pool_id = le64_to_cpu(data_pool_id);
-	WARN_ON(rbd_dev->header.data_pool_id == CEPH_NOPOOL);
+	*data_pool_id = le64_to_cpu(data_pool_buf);
+	dout("  data_pool_id = %lld\n", *data_pool_id);
+	WARN_ON(*data_pool_id == CEPH_NOPOOL);
+
 	return 0;
 }
 
@@ -6259,7 +6220,8 @@ out_err:
 	return ret;
 }
 
-static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
+static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev,
+				   struct ceph_snap_context **psnapc)
 {
 	size_t size;
 	int ret;
@@ -6320,9 +6282,7 @@ static int rbd_dev_v2_snap_context(struct rbd_device *rbd_dev)
 	for (i = 0; i < snap_count; i++)
 		snapc->snaps[i] = ceph_decode_64(&p);
 
-	ceph_put_snap_context(rbd_dev->header.snapc);
-	rbd_dev->header.snapc = snapc;
-
+	*psnapc = snapc;
 	dout("  snap context seq = %llu, snap_count = %u\n",
 		(unsigned long long)seq, (unsigned int)snap_count);
 out:
@@ -6371,38 +6331,42 @@ out:
 	return snap_name;
 }
 
-static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev)
+static int rbd_dev_v2_header_info(struct rbd_device *rbd_dev,
+				  struct rbd_image_header *header,
+				  bool first_time)
 {
-	bool first_time = rbd_dev->header.object_prefix == NULL;
 	int ret;
 
-	ret = rbd_dev_v2_image_size(rbd_dev);
+	ret = _rbd_dev_v2_snap_size(rbd_dev, CEPH_NOSNAP,
+				    first_time ? &header->obj_order : NULL,
+				    &header->image_size);
 	if (ret)
 		return ret;
 
 	if (first_time) {
-		ret = rbd_dev_v2_header_onetime(rbd_dev);
+		ret = rbd_dev_v2_header_onetime(rbd_dev, header);
 		if (ret)
 			return ret;
 	}
 
-	ret = rbd_dev_v2_snap_context(rbd_dev);
-	if (ret && first_time) {
-		kfree(rbd_dev->header.object_prefix);
-		rbd_dev->header.object_prefix = NULL;
-	}
+	ret = rbd_dev_v2_snap_context(rbd_dev, &header->snapc);
+	if (ret)
+		return ret;
 
-	return ret;
+	return 0;
 }
 
-static int rbd_dev_header_info(struct rbd_device *rbd_dev)
+static int rbd_dev_header_info(struct rbd_device *rbd_dev,
+			       struct rbd_image_header *header,
+			       bool first_time)
 {
 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
+	rbd_assert(!header->object_prefix && !header->snapc);
 
 	if (rbd_dev->image_format == 1)
-		return rbd_dev_v1_header_info(rbd_dev);
+		return rbd_dev_v1_header_info(rbd_dev, header, first_time);
 
-	return rbd_dev_v2_header_info(rbd_dev);
+	return rbd_dev_v2_header_info(rbd_dev, header, first_time);
 }
 
 /*
@@ -6752,60 +6716,49 @@ out:
  */
 static void rbd_dev_unprobe(struct rbd_device *rbd_dev)
 {
-	struct rbd_image_header	*header;
-
 	rbd_dev_parent_put(rbd_dev);
 	rbd_object_map_free(rbd_dev);
 	rbd_dev_mapping_clear(rbd_dev);
 
 	/* Free dynamic fields from the header, then zero it out */
 
-	header = &rbd_dev->header;
-	ceph_put_snap_context(header->snapc);
-	kfree(header->snap_sizes);
-	kfree(header->snap_names);
-	kfree(header->object_prefix);
-	memset(header, 0, sizeof (*header));
+	rbd_image_header_cleanup(&rbd_dev->header);
 }
 
-static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev)
+static int rbd_dev_v2_header_onetime(struct rbd_device *rbd_dev,
+				     struct rbd_image_header *header)
 {
 	int ret;
 
-	ret = rbd_dev_v2_object_prefix(rbd_dev);
+	ret = rbd_dev_v2_object_prefix(rbd_dev, &header->object_prefix);
 	if (ret)
-		goto out_err;
+		return ret;
 
 	/*
 	 * Get the and check features for the image.  Currently the
 	 * features are assumed to never change.
 	 */
-	ret = rbd_dev_v2_features(rbd_dev);
+	ret = _rbd_dev_v2_snap_features(rbd_dev, CEPH_NOSNAP,
+					&header->features);
 	if (ret)
-		goto out_err;
+		return ret;
 
 	/* If the image supports fancy striping, get its parameters */
 
-	if (rbd_dev->header.features & RBD_FEATURE_STRIPINGV2) {
-		ret = rbd_dev_v2_striping_info(rbd_dev);
-		if (ret < 0)
-			goto out_err;
+	if (header->features & RBD_FEATURE_STRIPINGV2) {
+		ret = rbd_dev_v2_striping_info(rbd_dev, &header->stripe_unit,
+					       &header->stripe_count);
+		if (ret)
+			return ret;
 	}
 
-	if (rbd_dev->header.features & RBD_FEATURE_DATA_POOL) {
-		ret = rbd_dev_v2_data_pool(rbd_dev);
+	if (header->features & RBD_FEATURE_DATA_POOL) {
+		ret = rbd_dev_v2_data_pool(rbd_dev, &header->data_pool_id);
 		if (ret)
-			goto out_err;
+			return ret;
 	}
 
-	rbd_init_layout(rbd_dev);
 	return 0;
-
-out_err:
-	rbd_dev->header.features = 0;
-	kfree(rbd_dev->header.object_prefix);
-	rbd_dev->header.object_prefix = NULL;
-	return ret;
 }
 
 /*
@@ -6827,7 +6780,7 @@ static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth)
 		goto out_err;
 	}
 
-	parent = __rbd_dev_create(rbd_dev->rbd_client, rbd_dev->parent_spec);
+	parent = __rbd_dev_create(rbd_dev->parent_spec);
 	if (!parent) {
 		ret = -ENOMEM;
 		goto out_err;
@@ -6837,8 +6790,8 @@ static int rbd_dev_probe_parent(struct rbd_device *rbd_dev, int depth)
 	 * Images related by parent/child relationships always share
 	 * rbd_client and spec/parent_spec, so bump their refcounts.
 	 */
-	__rbd_get_client(rbd_dev->rbd_client);
-	rbd_spec_get(rbd_dev->parent_spec);
+	parent->rbd_client = __rbd_get_client(rbd_dev->rbd_client);
+	parent->spec = rbd_spec_get(rbd_dev->parent_spec);
 
 	ret = rbd_dev_image_probe(parent, depth);
 	if (ret < 0)
@@ -6983,10 +6936,12 @@ static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
 	if (!depth)
 		down_write(&rbd_dev->header_rwsem);
 
-	ret = rbd_dev_header_info(rbd_dev);
+	ret = rbd_dev_header_info(rbd_dev, &rbd_dev->header, true);
 	if (ret)
 		goto err_out_probe;
 
+	rbd_init_layout(rbd_dev);
+
 	/*
 	 * If this image is the one being mapped, we have pool name and
 	 * id, image name and id, and snap name - need to fill snap id.
@@ -7020,7 +6975,7 @@ static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
 	}
 
 	if (rbd_dev->header.features & RBD_FEATURE_LAYERING) {
-		ret = rbd_dev_v2_parent_info(rbd_dev);
+		ret = rbd_dev_setup_parent(rbd_dev);
 		if (ret)
 			goto err_out_probe;
 	}
@@ -7046,6 +7001,112 @@ err_out_format:
 	return ret;
 }
 
+static void rbd_dev_update_header(struct rbd_device *rbd_dev,
+				  struct rbd_image_header *header)
+{
+	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
+	rbd_assert(rbd_dev->header.object_prefix); /* !first_time */
+
+	if (rbd_dev->header.image_size != header->image_size) {
+		rbd_dev->header.image_size = header->image_size;
+
+		if (rbd_dev->spec->snap_id == CEPH_NOSNAP) {
+			rbd_dev->mapping.size = header->image_size;
+			rbd_dev_update_size(rbd_dev);
+		}
+	}
+
+	if (rbd_dev->spec->snap_id != CEPH_NOSNAP) {
+		/* validate mapped snapshot's EXISTS flag */
+		rbd_exists_validate(rbd_dev);
+	}
+
+	ceph_put_snap_context(rbd_dev->header.snapc);
+	rbd_dev->header.snapc = header->snapc;
+	header->snapc = NULL;
+
+	if (rbd_dev->image_format == 1) {
+		kfree(rbd_dev->header.snap_names);
+		rbd_dev->header.snap_names = header->snap_names;
+		header->snap_names = NULL;
+
+		kfree(rbd_dev->header.snap_sizes);
+		rbd_dev->header.snap_sizes = header->snap_sizes;
+		header->snap_sizes = NULL;
+	}
+}
+
+static void rbd_dev_update_parent(struct rbd_device *rbd_dev,
+				  struct parent_image_info *pii)
+{
+	if (pii->pool_id == CEPH_NOPOOL || !pii->has_overlap) {
+		/*
+		 * Either the parent never existed, or we have
+		 * record of it but the image got flattened so it no
+		 * longer has a parent.  When the parent of a
+		 * layered image disappears we immediately set the
+		 * overlap to 0.  The effect of this is that all new
+		 * requests will be treated as if the image had no
+		 * parent.
+		 *
+		 * If !pii.has_overlap, the parent image spec is not
+		 * applicable.  It's there to avoid duplication in each
+		 * snapshot record.
+		 */
+		if (rbd_dev->parent_overlap) {
+			rbd_dev->parent_overlap = 0;
+			rbd_dev_parent_put(rbd_dev);
+			pr_info("%s: clone has been flattened\n",
+				rbd_dev->disk->disk_name);
+		}
+	} else {
+		rbd_assert(rbd_dev->parent_spec);
+
+		/*
+		 * Update the parent overlap.  If it became zero, issue
+		 * a warning as we will proceed as if there is no parent.
+		 */
+		if (!pii->overlap && rbd_dev->parent_overlap)
+			rbd_warn(rbd_dev,
+				 "clone has become standalone (overlap 0)");
+		rbd_dev->parent_overlap = pii->overlap;
+	}
+}
+
+static int rbd_dev_refresh(struct rbd_device *rbd_dev)
+{
+	struct rbd_image_header	header = { 0 };
+	struct parent_image_info pii = { 0 };
+	int ret;
+
+	dout("%s rbd_dev %p\n", __func__, rbd_dev);
+
+	ret = rbd_dev_header_info(rbd_dev, &header, false);
+	if (ret)
+		goto out;
+
+	/*
+	 * If there is a parent, see if it has disappeared due to the
+	 * mapped image getting flattened.
+	 */
+	if (rbd_dev->parent) {
+		ret = rbd_dev_v2_parent_info(rbd_dev, &pii);
+		if (ret)
+			goto out;
+	}
+
+	down_write(&rbd_dev->header_rwsem);
+	rbd_dev_update_header(rbd_dev, &header);
+	if (rbd_dev->parent)
+		rbd_dev_update_parent(rbd_dev, &pii);
+	up_write(&rbd_dev->header_rwsem);
+
+out:
+	rbd_parent_info_cleanup(&pii);
+	rbd_image_header_cleanup(&header);
+	return ret;
+}
+
 static ssize_t do_rbd_add(struct bus_type *bus,
 			  const char *buf,
 			  size_t count)