Merge tag 'ceph-for-4.9-rc1' of git://github.com/ceph/ceph-client

Pull Ceph updates from Ilya Dryomov:
 "The big ticket item here is support for rbd exclusive-lock feature,
  with maintenance operations offloaded to userspace (Douglas Fuller,
  Mike Christie and myself). Another block device bullet is a series
  fixing up layering error paths (myself).

  On the filesystem side, we've got patches that improve our handling of
  buffered vs dio write races (Neil Brown) and a few assorted fixes from
  Zheng. Also included a couple of random cleanups and a minor CRUSH
  update"

* tag 'ceph-for-4.9-rc1' of git://github.com/ceph/ceph-client: (39 commits)
  crush: remove redundant local variable
  crush: don't normalize input of crush_ln iteratively
  libceph: ceph_build_auth() doesn't need ceph_auth_build_hello()
  libceph: use CEPH_AUTH_UNKNOWN in ceph_auth_build_hello()
  ceph: fix description for rsize and rasize mount options
  rbd: use kmalloc_array() in rbd_header_from_disk()
  ceph: use list_move instead of list_del/list_add
  ceph: handle CEPH_SESSION_REJECT message
  ceph: avoid accessing / when mounting a subpath
  ceph: fix mandatory flock check
  ceph: remove warning when ceph_releasepage() is called on dirty page
  ceph: ignore error from invalidate_inode_pages2_range() in direct write
  ceph: fix error handling of start_read()
  rbd: add rbd_obj_request_error() helper
  rbd: img_data requests don't own their page array
  rbd: don't call rbd_osd_req_format_read() for !img_data requests
  rbd: rework rbd_img_obj_exists_submit() error paths
  rbd: don't crash or leak on errors in rbd_img_obj_parent_read_full_callback()
  rbd: move bumping img_request refcount into rbd_obj_request_submit()
  rbd: mark the original request as done if stat request fails
  ...

2 files changed, 1179 insertions, 264 deletions

diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index c1f84df7838b..abb71628ab61 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -31,6 +31,7 @@
 #include <linux/ceph/libceph.h>
 #include <linux/ceph/osd_client.h>
 #include <linux/ceph/mon_client.h>
+#include <linux/ceph/cls_lock_client.h>
 #include <linux/ceph/decode.h>
 #include <linux/parser.h>
 #include <linux/bsearch.h>
@@ -114,12 +115,17 @@ static int atomic_dec_return_safe(atomic_t *v)
 
 #define RBD_OBJ_PREFIX_LEN_MAX	64
 
+#define RBD_NOTIFY_TIMEOUT	5	/* seconds */
+#define RBD_RETRY_DELAY		msecs_to_jiffies(1000)
+
 /* Feature bits */
 
 #define RBD_FEATURE_LAYERING	(1<<0)
 #define RBD_FEATURE_STRIPINGV2	(1<<1)
-#define RBD_FEATURES_ALL \
-	    (RBD_FEATURE_LAYERING | RBD_FEATURE_STRIPINGV2)
+#define RBD_FEATURE_EXCLUSIVE_LOCK (1<<2)
+#define RBD_FEATURES_ALL	(RBD_FEATURE_LAYERING |		\
+				 RBD_FEATURE_STRIPINGV2 |	\
+				 RBD_FEATURE_EXCLUSIVE_LOCK)
 
 /* Features supported by this (client software) implementation. */
 
@@ -128,11 +134,8 @@ static int atomic_dec_return_safe(atomic_t *v)
 /*
  * An RBD device name will be "rbd#", where the "rbd" comes from
  * RBD_DRV_NAME above, and # is a unique integer identifier.
- * MAX_INT_FORMAT_WIDTH is used in ensuring DEV_NAME_LEN is big
- * enough to hold all possible device names.
  */
 #define DEV_NAME_LEN		32
-#define MAX_INT_FORMAT_WIDTH	((5 * sizeof (int)) / 2 + 1)
 
 /*
  * block device image metadata (in-memory version)
@@ -322,6 +325,24 @@ struct rbd_img_request {
 #define for_each_obj_request_safe(ireq, oreq, n) \
 	list_for_each_entry_safe_reverse(oreq, n, &(ireq)->obj_requests, links)
 
+enum rbd_watch_state {
+	RBD_WATCH_STATE_UNREGISTERED,
+	RBD_WATCH_STATE_REGISTERED,
+	RBD_WATCH_STATE_ERROR,
+};
+
+enum rbd_lock_state {
+	RBD_LOCK_STATE_UNLOCKED,
+	RBD_LOCK_STATE_LOCKED,
+	RBD_LOCK_STATE_RELEASING,
+};
+
+/* WatchNotify::ClientId */
+struct rbd_client_id {
+	u64 gid;
+	u64 handle;
+};
+
 struct rbd_mapping {
 	u64                     size;
 	u64                     features;
@@ -349,13 +370,29 @@ struct rbd_device {
 	unsigned long		flags;		/* possibly lock protected */
 	struct rbd_spec		*spec;
 	struct rbd_options	*opts;
+	char			*config_info;	/* add{,_single_major} string */
 
 	struct ceph_object_id	header_oid;
 	struct ceph_object_locator header_oloc;
 
-	struct ceph_file_layout	layout;
+	struct ceph_file_layout	layout;		/* used for all rbd requests */
 
+	struct mutex		watch_mutex;
+	enum rbd_watch_state	watch_state;
 	struct ceph_osd_linger_request *watch_handle;
+	u64			watch_cookie;
+	struct delayed_work	watch_dwork;
+
+	struct rw_semaphore	lock_rwsem;
+	enum rbd_lock_state	lock_state;
+	struct rbd_client_id	owner_cid;
+	struct work_struct	acquired_lock_work;
+	struct work_struct	released_lock_work;
+	struct delayed_work	lock_dwork;
+	struct work_struct	unlock_work;
+	wait_queue_head_t	lock_waitq;
+
+	struct workqueue_struct	*task_wq;
 
 	struct rbd_spec		*parent_spec;
 	u64			parent_overlap;
@@ -439,6 +476,29 @@ static int minor_to_rbd_dev_id(int minor)
 	return minor >> RBD_SINGLE_MAJOR_PART_SHIFT;
 }
 
+static bool rbd_is_lock_supported(struct rbd_device *rbd_dev)
+{
+	return (rbd_dev->header.features & RBD_FEATURE_EXCLUSIVE_LOCK) &&
+	       rbd_dev->spec->snap_id == CEPH_NOSNAP &&
+	       !rbd_dev->mapping.read_only;
+}
+
+static bool __rbd_is_lock_owner(struct rbd_device *rbd_dev)
+{
+	return rbd_dev->lock_state == RBD_LOCK_STATE_LOCKED ||
+	       rbd_dev->lock_state == RBD_LOCK_STATE_RELEASING;
+}
+
+static bool rbd_is_lock_owner(struct rbd_device *rbd_dev)
+{
+	bool is_lock_owner;
+
+	down_read(&rbd_dev->lock_rwsem);
+	is_lock_owner = __rbd_is_lock_owner(rbd_dev);
+	up_read(&rbd_dev->lock_rwsem);
+	return is_lock_owner;
+}
+
 static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
 static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
 static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major);
@@ -735,6 +795,7 @@ enum {
 	/* string args above */
 	Opt_read_only,
 	Opt_read_write,
+	Opt_lock_on_read,
 	Opt_err
 };
 
@@ -746,16 +807,19 @@ static match_table_t rbd_opts_tokens = {
 	{Opt_read_only, "ro"},		/* Alternate spelling */
 	{Opt_read_write, "read_write"},
 	{Opt_read_write, "rw"},		/* Alternate spelling */
+	{Opt_lock_on_read, "lock_on_read"},
 	{Opt_err, NULL}
 };
 
 struct rbd_options {
 	int	queue_depth;
 	bool	read_only;
+	bool	lock_on_read;
 };
 
 #define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
 #define RBD_READ_ONLY_DEFAULT	false
+#define RBD_LOCK_ON_READ_DEFAULT false
 
 static int parse_rbd_opts_token(char *c, void *private)
 {
@@ -791,6 +855,9 @@ static int parse_rbd_opts_token(char *c, void *private)
 	case Opt_read_write:
 		rbd_opts->read_only = false;
 		break;
+	case Opt_lock_on_read:
+		rbd_opts->lock_on_read = true;
+		break;
 	default:
 		/* libceph prints "bad option" msg */
 		return -EINVAL;
@@ -919,7 +986,6 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,
 	char *snap_names = NULL;
 	u64 *snap_sizes = NULL;
 	u32 snap_count;
-	size_t size;
 	int ret = -ENOMEM;
 	u32 i;
 
@@ -957,9 +1023,9 @@ static int rbd_header_from_disk(struct rbd_device *rbd_dev,
 			goto out_err;
 
 		/* ...as well as the array of their sizes. */
-
-		size = snap_count * sizeof (*header->snap_sizes);
-		snap_sizes = kmalloc(size, GFP_KERNEL);
+		snap_sizes = kmalloc_array(snap_count,
+					   sizeof(*header->snap_sizes),
+					   GFP_KERNEL);
 		if (!snap_sizes)
 			goto out_err;
 
@@ -1551,11 +1617,18 @@ static bool obj_request_type_valid(enum obj_request_type type)
 	}
 }
 
-static int rbd_obj_request_submit(struct ceph_osd_client *osdc,
-				struct rbd_obj_request *obj_request)
+static void rbd_img_obj_callback(struct rbd_obj_request *obj_request);
+
+static void rbd_obj_request_submit(struct rbd_obj_request *obj_request)
 {
-	dout("%s %p\n", __func__, obj_request);
-	return ceph_osdc_start_request(osdc, obj_request->osd_req, false);
+	struct ceph_osd_request *osd_req = obj_request->osd_req;
+
+	dout("%s %p osd_req %p\n", __func__, obj_request, osd_req);
+	if (obj_request_img_data_test(obj_request)) {
+		WARN_ON(obj_request->callback != rbd_img_obj_callback);
+		rbd_img_request_get(obj_request->img_request);
+	}
+	ceph_osdc_start_request(osd_req->r_osdc, osd_req, false);
 }
 
 static void rbd_obj_request_end(struct rbd_obj_request *obj_request)
@@ -1745,6 +1818,22 @@ static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
 		complete_all(&obj_request->completion);
 }
 
+static void rbd_obj_request_error(struct rbd_obj_request *obj_request, int err)
+{
+	obj_request->result = err;
+	obj_request->xferred = 0;
+	/*
+	 * kludge - mirror rbd_obj_request_submit() to match a put in
+	 * rbd_img_obj_callback()
+	 */
+	if (obj_request_img_data_test(obj_request)) {
+		WARN_ON(obj_request->callback != rbd_img_obj_callback);
+		rbd_img_request_get(obj_request->img_request);
+	}
+	obj_request_done_set(obj_request);
+	rbd_obj_request_complete(obj_request);
+}
+
 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
 {
 	struct rbd_img_request *img_request = NULL;
@@ -1877,11 +1966,10 @@ static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
 
 static void rbd_osd_req_format_read(struct rbd_obj_request *obj_request)
 {
-	struct rbd_img_request *img_request = obj_request->img_request;
 	struct ceph_osd_request *osd_req = obj_request->osd_req;
 
-	if (img_request)
-		osd_req->r_snapid = img_request->snap_id;
+	rbd_assert(obj_request_img_data_test(obj_request));
+	osd_req->r_snapid = obj_request->img_request->snap_id;
 }
 
 static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
@@ -2074,7 +2162,9 @@ static void rbd_obj_request_destroy(struct kref *kref)
 			bio_chain_put(obj_request->bio_list);
 		break;
 	case OBJ_REQUEST_PAGES:
-		if (obj_request->pages)
+		/* img_data requests don't own their page array */
+		if (obj_request->pages &&
+		    !obj_request_img_data_test(obj_request))
 			ceph_release_page_vector(obj_request->pages,
 						obj_request->page_count);
 		break;
@@ -2295,13 +2385,6 @@ static bool rbd_img_obj_end_request(struct rbd_obj_request *obj_request)
 		xferred = obj_request->length;
 	}
 
-	/* Image object requests don't own their page array */
-
-	if (obj_request->type == OBJ_REQUEST_PAGES) {
-		obj_request->pages = NULL;
-		obj_request->page_count = 0;
-	}
-
 	if (img_request_child_test(img_request)) {
 		rbd_assert(img_request->obj_request != NULL);
 		more = obj_request->which < img_request->obj_request_count - 1;
@@ -2520,8 +2603,6 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,
 
 		rbd_img_obj_request_fill(obj_request, osd_req, op_type, 0);
 
-		rbd_img_request_get(img_request);
-
 		img_offset += length;
 		resid -= length;
 	}
@@ -2579,7 +2660,6 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
 {
 	struct rbd_obj_request *orig_request;
 	struct ceph_osd_request *osd_req;
-	struct ceph_osd_client *osdc;
 	struct rbd_device *rbd_dev;
 	struct page **pages;
 	enum obj_operation_type op_type;
@@ -2603,7 +2683,7 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
 	rbd_assert(obj_request_type_valid(orig_request->type));
 	img_result = img_request->result;
 	parent_length = img_request->length;
-	rbd_assert(parent_length == img_request->xferred);
+	rbd_assert(img_result || parent_length == img_request->xferred);
 	rbd_img_request_put(img_request);
 
 	rbd_assert(orig_request->img_request);
@@ -2616,13 +2696,9 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
 	 * and re-submit the original write request.
 	 */
 	if (!rbd_dev->parent_overlap) {
-		struct ceph_osd_client *osdc;
-
 		ceph_release_page_vector(pages, page_count);
-		osdc = &rbd_dev->rbd_client->client->osdc;
-		img_result = rbd_obj_request_submit(osdc, orig_request);
-		if (!img_result)
-			return;
+		rbd_obj_request_submit(orig_request);
+		return;
 	}
 
 	if (img_result)
@@ -2656,17 +2732,12 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
 
 	/* All set, send it off. */
 
-	osdc = &rbd_dev->rbd_client->client->osdc;
-	img_result = rbd_obj_request_submit(osdc, orig_request);
-	if (!img_result)
-		return;
-out_err:
-	/* Record the error code and complete the request */
+	rbd_obj_request_submit(orig_request);
+	return;
 
-	orig_request->result = img_result;
-	orig_request->xferred = 0;
-	obj_request_done_set(orig_request);
-	rbd_obj_request_complete(orig_request);
+out_err:
+	ceph_release_page_vector(pages, page_count);
+	rbd_obj_request_error(orig_request, img_result);
 }
 
 /*
@@ -2680,26 +2751,19 @@ out_err:
  * When the read completes, this page array will be transferred to
  * the original object request for the copyup operation.
  *
- * If an error occurs, record it as the result of the original
- * object request and mark it done so it gets completed.
+ * If an error occurs, it is recorded as the result of the original
+ * object request in rbd_img_obj_exists_callback().
  */
 static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
 {
-	struct rbd_img_request *img_request = NULL;
+	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
 	struct rbd_img_request *parent_request = NULL;
-	struct rbd_device *rbd_dev;
 	u64 img_offset;
 	u64 length;
 	struct page **pages = NULL;
 	u32 page_count;
 	int result;
 
-	rbd_assert(obj_request_img_data_test(obj_request));
-	rbd_assert(obj_request_type_valid(obj_request->type));
-
-	img_request = obj_request->img_request;
-	rbd_assert(img_request != NULL);
-	rbd_dev = img_request->rbd_dev;
 	rbd_assert(rbd_dev->parent != NULL);
 
 	/*
@@ -2740,10 +2804,11 @@ static int rbd_img_obj_parent_read_full(struct rbd_obj_request *obj_request)
 	result = rbd_img_request_fill(parent_request, OBJ_REQUEST_PAGES, pages);
 	if (result)
 		goto out_err;
+
 	parent_request->copyup_pages = pages;
 	parent_request->copyup_page_count = page_count;
-
 	parent_request->callback = rbd_img_obj_parent_read_full_callback;
+
 	result = rbd_img_request_submit(parent_request);
 	if (!result)
 		return 0;
@@ -2757,10 +2822,6 @@ out_err:
 		ceph_release_page_vector(pages, page_count);
 	if (parent_request)
 		rbd_img_request_put(parent_request);
-	obj_request->result = result;
-	obj_request->xferred = 0;
-	obj_request_done_set(obj_request);
-
 	return result;
 }
 
@@ -2793,17 +2854,13 @@ static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
 
 	/*
 	 * If the overlap has become 0 (most likely because the
-	 * image has been flattened) we need to free the pages
-	 * and re-submit the original write request.
+	 * image has been flattened) we need to re-submit the
+	 * original request.
 	 */
 	rbd_dev = orig_request->img_request->rbd_dev;
 	if (!rbd_dev->parent_overlap) {
-		struct ceph_osd_client *osdc;
-
-		osdc = &rbd_dev->rbd_client->client->osdc;
-		result = rbd_obj_request_submit(osdc, orig_request);
-		if (!result)
-			return;
+		rbd_obj_request_submit(orig_request);
+		return;
 	}
 
 	/*
@@ -2816,31 +2873,45 @@ static void rbd_img_obj_exists_callback(struct rbd_obj_request *obj_request)
 		obj_request_existence_set(orig_request, true);
 	} else if (result == -ENOENT) {
 		obj_request_existence_set(orig_request, false);
-	} else if (result) {
-		orig_request->result = result;
-		goto out;
+	} else {
+		goto fail_orig_request;
 	}
 
 	/*
 	 * Resubmit the original request now that we have recorded
 	 * whether the target object exists.
 	 */
-	orig_request->result = rbd_img_obj_request_submit(orig_request);
-out:
-	if (orig_request->result)
-		rbd_obj_request_complete(orig_request);
+	result = rbd_img_obj_request_submit(orig_request);
+	if (result)
+		goto fail_orig_request;
+
+	return;
+
+fail_orig_request:
+	rbd_obj_request_error(orig_request, result);
 }
 
 static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
 {
+	struct rbd_device *rbd_dev = obj_request->img_request->rbd_dev;
 	struct rbd_obj_request *stat_request;
-	struct rbd_device *rbd_dev;
-	struct ceph_osd_client *osdc;
-	struct page **pages = NULL;
+	struct page **pages;
 	u32 page_count;
 	size_t size;
 	int ret;
 
+	stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
+					      OBJ_REQUEST_PAGES);
+	if (!stat_request)
+		return -ENOMEM;
+
+	stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
+						   stat_request);
+	if (!stat_request->osd_req) {
+		ret = -ENOMEM;
+		goto fail_stat_request;
+	}
+
 	/*
 	 * The response data for a STAT call consists of:
 	 *     le64 length;
@@ -2852,52 +2923,33 @@ static int rbd_img_obj_exists_submit(struct rbd_obj_request *obj_request)
 	size = sizeof (__le64) + sizeof (__le32) + sizeof (__le32);
 	page_count = (u32)calc_pages_for(0, size);
 	pages = ceph_alloc_page_vector(page_count, GFP_KERNEL);
-	if (IS_ERR(pages))
-		return PTR_ERR(pages);
+	if (IS_ERR(pages)) {
+		ret = PTR_ERR(pages);
+		goto fail_stat_request;
+	}
 
-	ret = -ENOMEM;
-	stat_request = rbd_obj_request_create(obj_request->object_name, 0, 0,
-							OBJ_REQUEST_PAGES);
-	if (!stat_request)
-		goto out;
+	osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0);
+	osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
+				     false, false);
 
 	rbd_obj_request_get(obj_request);
 	stat_request->obj_request = obj_request;
 	stat_request->pages = pages;
 	stat_request->page_count = page_count;
-
-	rbd_assert(obj_request->img_request);
-	rbd_dev = obj_request->img_request->rbd_dev;
-	stat_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
-						   stat_request);
-	if (!stat_request->osd_req)
-		goto out;
 	stat_request->callback = rbd_img_obj_exists_callback;
 
-	osd_req_op_init(stat_request->osd_req, 0, CEPH_OSD_OP_STAT, 0);
-	osd_req_op_raw_data_in_pages(stat_request->osd_req, 0, pages, size, 0,
-					false, false);
-	rbd_osd_req_format_read(stat_request);
-
-	osdc = &rbd_dev->rbd_client->client->osdc;
-	ret = rbd_obj_request_submit(osdc, stat_request);
-out:
-	if (ret)
-		rbd_obj_request_put(obj_request);
+	rbd_obj_request_submit(stat_request);
+	return 0;
 
+fail_stat_request:
+	rbd_obj_request_put(stat_request);
 	return ret;
 }
 
 static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
 {
-	struct rbd_img_request *img_request;
-	struct rbd_device *rbd_dev;
-
-	rbd_assert(obj_request_img_data_test(obj_request));
-
-	img_request = obj_request->img_request;
-	rbd_assert(img_request);
-	rbd_dev = img_request->rbd_dev;
+	struct rbd_img_request *img_request = obj_request->img_request;
+	struct rbd_device *rbd_dev = img_request->rbd_dev;
 
 	/* Reads */
 	if (!img_request_write_test(img_request) &&
@@ -2936,14 +2988,13 @@ static bool img_obj_request_simple(struct rbd_obj_request *obj_request)
 
 static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request)
 {
-	if (img_obj_request_simple(obj_request)) {
-		struct rbd_device *rbd_dev;
-		struct ceph_osd_client *osdc;
-
-		rbd_dev = obj_request->img_request->rbd_dev;
-		osdc = &rbd_dev->rbd_client->client->osdc;
+	rbd_assert(obj_request_img_data_test(obj_request));
+	rbd_assert(obj_request_type_valid(obj_request->type));
+	rbd_assert(obj_request->img_request);
 
-		return rbd_obj_request_submit(osdc, obj_request);
+	if (img_obj_request_simple(obj_request)) {
+		rbd_obj_request_submit(obj_request);
+		return 0;
 	}
 
 	/*
@@ -3006,12 +3057,8 @@ static void rbd_img_parent_read_callback(struct rbd_img_request *img_request)
 	rbd_assert(obj_request->img_request);
 	rbd_dev = obj_request->img_request->rbd_dev;
 	if (!rbd_dev->parent_overlap) {
-		struct ceph_osd_client *osdc;
-
-		osdc = &rbd_dev->rbd_client->client->osdc;
-		img_result = rbd_obj_request_submit(osdc, obj_request);
-		if (!img_result)
-			return;
+		rbd_obj_request_submit(obj_request);
+		return;
 	}
 
 	obj_request->result = img_result;
@@ -3084,65 +3131,724 @@ out_err:
 	obj_request_done_set(obj_request);
 }
 
-static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev);
-static void __rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev);
+static const struct rbd_client_id rbd_empty_cid;
 
-static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie,
-			 u64 notifier_id, void *data, size_t data_len)
+static bool rbd_cid_equal(const struct rbd_client_id *lhs,
+			  const struct rbd_client_id *rhs)
+{
+	return lhs->gid == rhs->gid && lhs->handle == rhs->handle;
+}
+
+static struct rbd_client_id rbd_get_cid(struct rbd_device *rbd_dev)
+{
+	struct rbd_client_id cid;
+
+	mutex_lock(&rbd_dev->watch_mutex);
+	cid.gid = ceph_client_gid(rbd_dev->rbd_client->client);
+	cid.handle = rbd_dev->watch_cookie;
+	mutex_unlock(&rbd_dev->watch_mutex);
+	return cid;
+}
+
+/*
+ * lock_rwsem must be held for write
+ */
+static void rbd_set_owner_cid(struct rbd_device *rbd_dev,
+			      const struct rbd_client_id *cid)
+{
+	dout("%s rbd_dev %p %llu-%llu -> %llu-%llu\n", __func__, rbd_dev,
+	     rbd_dev->owner_cid.gid, rbd_dev->owner_cid.handle,
+	     cid->gid, cid->handle);
+	rbd_dev->owner_cid = *cid; /* struct */
+}
+
+static void format_lock_cookie(struct rbd_device *rbd_dev, char *buf)
+{
+	mutex_lock(&rbd_dev->watch_mutex);
+	sprintf(buf, "%s %llu", RBD_LOCK_COOKIE_PREFIX, rbd_dev->watch_cookie);
+	mutex_unlock(&rbd_dev->watch_mutex);
+}
+
+/*
+ * lock_rwsem must be held for write
+ */
+static int rbd_lock(struct rbd_device *rbd_dev)
 {
-	struct rbd_device *rbd_dev = arg;
 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
+	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
+	char cookie[32];
+	int ret;
+
+	WARN_ON(__rbd_is_lock_owner(rbd_dev));
+
+	format_lock_cookie(rbd_dev, cookie);
+	ret = ceph_cls_lock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
+			    RBD_LOCK_NAME, CEPH_CLS_LOCK_EXCLUSIVE, cookie,
+			    RBD_LOCK_TAG, "", 0);
+	if (ret)
+		return ret;
+
+	rbd_dev->lock_state = RBD_LOCK_STATE_LOCKED;
+	rbd_set_owner_cid(rbd_dev, &cid);
+	queue_work(rbd_dev->task_wq, &rbd_dev->acquired_lock_work);
+	return 0;
+}
+
+/*
+ * lock_rwsem must be held for write
+ */
+static int rbd_unlock(struct rbd_device *rbd_dev)
+{
+	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
+	char cookie[32];
+	int ret;
+
+	WARN_ON(!__rbd_is_lock_owner(rbd_dev));
+
+	rbd_dev->lock_state = RBD_LOCK_STATE_UNLOCKED;
+
+	format_lock_cookie(rbd_dev, cookie);
+	ret = ceph_cls_unlock(osdc, &rbd_dev->header_oid, &rbd_dev->header_oloc,
+			      RBD_LOCK_NAME, cookie);
+	if (ret && ret != -ENOENT) {
+		rbd_warn(rbd_dev, "cls_unlock failed: %d", ret);
+		return ret;
+	}
+
+	rbd_set_owner_cid(rbd_dev, &rbd_empty_cid);
+	queue_work(rbd_dev->task_wq, &rbd_dev->released_lock_work);
+	return 0;
+}
+
+static int __rbd_notify_op_lock(struct rbd_device *rbd_dev,
+				enum rbd_notify_op notify_op,
+				struct page ***preply_pages,
+				size_t *preply_len)
+{
+	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
+	struct rbd_client_id cid = rbd_get_cid(rbd_dev);
+	int buf_size = 4 + 8 + 8 + CEPH_ENCODING_START_BLK_LEN;
+	char buf[buf_size];
+	void *p = buf;
+
+	dout("%s rbd_dev %p notify_op %d\n", __func__, rbd_dev, notify_op);
+
+	/* encode *LockPayload NotifyMessage (op + ClientId) */
+	ceph_start_encoding(&p, 2, 1, buf_size - CEPH_ENCODING_START_BLK_LEN);
+	ceph_encode_32(&p, notify_op);
+	ceph_encode_64(&p, cid.gid);
+	ceph_encode_64(&p, cid.handle);
+
+	return ceph_osdc_notify(osdc, &rbd_dev->header_oid,
+				&rbd_dev->header_oloc, buf, buf_size,
+				RBD_NOTIFY_TIMEOUT, preply_pages, preply_len);
+}
+
+static void rbd_notify_op_lock(struct rbd_device *rbd_dev,
+			       enum rbd_notify_op notify_op)
+{
+	struct page **reply_pages;
+	size_t reply_len;
+
+	__rbd_notify_op_lock(rbd_dev, notify_op, &reply_pages, &reply_len);
+	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
+}
+
+static void rbd_notify_acquired_lock(struct work_struct *work)
+{
+	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
+						  acquired_lock_work);
+
+	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_ACQUIRED_LOCK);
+}
+
+static void rbd_notify_released_lock(struct work_struct *work)
+{
+	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
+						  released_lock_work);
+
+	rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_RELEASED_LOCK);
+}
+
+static int rbd_request_lock(struct rbd_device *rbd_dev)
+{
+	struct page **reply_pages;
+	size_t reply_len;
+	bool lock_owner_responded = false;
 	int ret;
 
-	dout("%s rbd_dev %p cookie %llu notify_id %llu\n", __func__, rbd_dev,
-	     cookie, notify_id);
+	dout("%s rbd_dev %p\n", __func__, rbd_dev);
+
+	ret = __rbd_notify_op_lock(rbd_dev, RBD_NOTIFY_OP_REQUEST_LOCK,
+				   &reply_pages, &reply_len);
+	if (ret && ret != -ETIMEDOUT) {
+		rbd_warn(rbd_dev, "failed to request lock: %d", ret);
+		goto out;
+	}
+
+	if (reply_len > 0 && reply_len <= PAGE_SIZE) {
+		void *p = page_address(reply_pages[0]);
+		void *const end = p + reply_len;
+		u32 n;
+
+		ceph_decode_32_safe(&p, end, n, e_inval); /* num_acks */
+		while (n--) {
+			u8 struct_v;
+			u32 len;
 
+			ceph_decode_need(&p, end, 8 + 8, e_inval);
+			p += 8 + 8; /* skip gid and cookie */
+
+			ceph_decode_32_safe(&p, end, len, e_inval);
+			if (!len)
+				continue;
+
+			if (lock_owner_responded) {
+				rbd_warn(rbd_dev,
+					 "duplicate lock owners detected");
+				ret = -EIO;
+				goto out;
+			}
+
+			lock_owner_responded = true;
+			ret = ceph_start_decoding(&p, end, 1, "ResponseMessage",
+						  &struct_v, &len);
+			if (ret) {
+				rbd_warn(rbd_dev,
+					 "failed to decode ResponseMessage: %d",
+					 ret);
+				goto e_inval;
+			}
+
+			ret = ceph_decode_32(&p);
+		}
+	}
+
+	if (!lock_owner_responded) {
+		rbd_warn(rbd_dev, "no lock owners detected");
+		ret = -ETIMEDOUT;
+	}
+
+out:
+	ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len));
+	return ret;
+
+e_inval:
+	ret = -EINVAL;
+	goto out;
+}
+
+static void wake_requests(struct rbd_device *rbd_dev, bool wake_all)
+{
+	dout("%s rbd_dev %p wake_all %d\n", __func__, rbd_dev, wake_all);
+
+	cancel_delayed_work(&rbd_dev->lock_dwork);
+	if (wake_all)
+		wake_up_all(&rbd_dev->lock_waitq);
+	else
+		wake_up(&rbd_dev->lock_waitq);
+}
+
+static int get_lock_owner_info(struct rbd_device *rbd_dev,
+			       struct ceph_locker **lockers, u32 *num_lockers)
+{
+	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
+	u8 lock_type;
+	char *lock_tag;
+	int ret;
+
+	dout("%s rbd_dev %p\n", __func__, rbd_dev);
+
+	ret = ceph_cls_lock_info(osdc, &rbd_dev->header_oid,
+				 &rbd_dev->header_oloc, RBD_LOCK_NAME,
+				 &lock_type, &lock_tag, lockers, num_lockers);
+	if (ret)
+		return ret;
+
+	if (*num_lockers == 0) {
+		dout("%s rbd_dev %p no lockers detected\n", __func__, rbd_dev);
+		goto out;
+	}
+
+	if (strcmp(lock_tag, RBD_LOCK_TAG)) {
+		rbd_warn(rbd_dev, "locked by external mechanism, tag %s",
+			 lock_tag);
+		ret = -EBUSY;
+		goto out;
+	}
+
+	if (lock_type == CEPH_CLS_LOCK_SHARED) {
+		rbd_warn(rbd_dev, "shared lock type detected");
+		ret = -EBUSY;
+		goto out;
+	}
+
+	if (strncmp((*lockers)[0].id.cookie, RBD_LOCK_COOKIE_PREFIX,
+		    strlen(RBD_LOCK_COOKIE_PREFIX))) {
+		rbd_warn(rbd_dev, "locked by external mechanism, cookie %s",
+			 (*lockers)[0].id.cookie);
+		ret = -EBUSY;
+		goto out;
+	}
+
+out:
+	kfree(lock_tag);
+	return ret;
+}
+
+static int find_watcher(struct rbd_device *rbd_dev,
+			const struct ceph_locker *locker)
+{
+	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
+	struct ceph_watch_item *watchers;
+	u32 num_watchers;
+	u64 cookie;
+	int i;
+	int ret;
+
+	ret = ceph_osdc_list_watchers(osdc, &rbd_dev->header_oid,
+				      &rbd_dev->header_oloc, &watchers,
+				      &num_watchers);
+	if (ret)
+		return ret;
+
+	sscanf(locker->id.cookie, RBD_LOCK_COOKIE_PREFIX " %llu", &cookie);
+	for (i = 0; i < num_watchers; i++) {
+		if (!memcmp(&watchers[i].addr, &locker->info.addr,
+			    sizeof(locker->info.addr)) &&
+		    watchers[i].cookie == cookie) {
+			struct rbd_client_id cid = {
+				.gid = le64_to_cpu(watchers[i].name.num),
+				.handle = cookie,
+			};
+
+			dout("%s rbd_dev %p found cid %llu-%llu\n", __func__,
+			     rbd_dev, cid.gid, cid.handle);
+			rbd_set_owner_cid(rbd_dev, &cid);
+			ret = 1;
+			goto out;
+		}
+	}
+
+	dout("%s rbd_dev %p no watchers\n", __func__, rbd_dev);
+	ret = 0;
+out:
+	kfree(watchers);
+	return ret;
+}
+
+/*
+ * lock_rwsem must be held for write
+ */
+static int rbd_try_lock(struct rbd_device *rbd_dev)
+{
+	struct ceph_client *client = rbd_dev->rbd_client->client;
+	struct ceph_locker *lockers;
+	u32 num_lockers;
+	int ret;
+
+	for (;;) {
+		ret = rbd_lock(rbd_dev);
+		if (ret != -EBUSY)
+			return ret;
+
+		/* determine if the current lock holder is still alive */
+		ret = get_lock_owner_info(rbd_dev, &lockers, &num_lockers);
+		if (ret)
+			return ret;
+
+		if (num_lockers == 0)
+			goto again;
+
+		ret = find_watcher(rbd_dev, lockers);
+		if (ret) {
+			if (ret > 0)
+				ret = 0; /* have to request lock */
+			goto out;
+		}
+
+		rbd_warn(rbd_dev, "%s%llu seems dead, breaking lock",
+			 ENTITY_NAME(lockers[0].id.name));
+
+		ret = ceph_monc_blacklist_add(&client->monc,
+					      &lockers[0].info.addr);
+		if (ret) {
+			rbd_warn(rbd_dev, "blacklist of %s%llu failed: %d",
+				 ENTITY_NAME(lockers[0].id.name), ret);
+			goto out;
+		}
+
+		ret = ceph_cls_break_lock(&client->osdc, &rbd_dev->header_oid,
+					  &rbd_dev->header_oloc, RBD_LOCK_NAME,
+					  lockers[0].id.cookie,
+					  &lockers[0].id.name);
+		if (ret && ret != -ENOENT)
+			goto out;
+
+again:
+		ceph_free_lockers(lockers, num_lockers);
+	}
+
+out:
+	ceph_free_lockers(lockers, num_lockers);
+	return ret;
+}
+
+/*
+ * ret is set only if lock_state is RBD_LOCK_STATE_UNLOCKED
+ */
+static enum rbd_lock_state rbd_try_acquire_lock(struct rbd_device *rbd_dev,
+						int *pret)
+{
+	enum rbd_lock_state lock_state;
+
+	down_read(&rbd_dev->lock_rwsem);
+	dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
+	     rbd_dev->lock_state);
+	if (__rbd_is_lock_owner(rbd_dev)) {
+		lock_state = rbd_dev->lock_state;
+		up_read(&rbd_dev->lock_rwsem);
+		return lock_state;
+	}
+
+	up_read(&rbd_dev->lock_rwsem);
+	down_write(&rbd_dev->lock_rwsem);
+	dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
+	     rbd_dev->lock_state);
+	if (!__rbd_is_lock_owner(rbd_dev)) {
+		*pret = rbd_try_lock(rbd_dev);
+		if (*pret)
+			rbd_warn(rbd_dev, "failed to acquire lock: %d", *pret);
+	}
+
+	lock_state = rbd_dev->lock_state;
+	up_write(&rbd_dev->lock_rwsem);
+	return lock_state;
+}
+
+static void rbd_acquire_lock(struct work_struct *work)
+{
+	struct rbd_device *rbd_dev = container_of(to_delayed_work(work),
+					    struct rbd_device, lock_dwork);
+	enum rbd_lock_state lock_state;
+	int ret;
+
+	dout("%s rbd_dev %p\n", __func__, rbd_dev);
+again:
+	lock_state = rbd_try_acquire_lock(rbd_dev, &ret);
+	if (lock_state != RBD_LOCK_STATE_UNLOCKED || ret == -EBLACKLISTED) {
+		if (lock_state == RBD_LOCK_STATE_LOCKED)
+			wake_requests(rbd_dev, true);
+		dout("%s rbd_dev %p lock_state %d ret %d - done\n", __func__,
+		     rbd_dev, lock_state, ret);
+		return;
+	}
+
+	ret = rbd_request_lock(rbd_dev);
+	if (ret == -ETIMEDOUT) {
+		goto again; /* treat this as a dead client */
+	} else if (ret < 0) {
+		rbd_warn(rbd_dev, "error requesting lock: %d", ret);
+		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
+				 RBD_RETRY_DELAY);
+	} else {
+		/*
+		 * lock owner acked, but resend if we don't see them
+		 * release the lock
+		 */
+		dout("%s rbd_dev %p requeueing lock_dwork\n", __func__,
+		     rbd_dev);
+		mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork,
+		    msecs_to_jiffies(2 * RBD_NOTIFY_TIMEOUT * MSEC_PER_SEC));
+	}
+}
+
+/*
+ * lock_rwsem must be held for write
+ */
+static bool rbd_release_lock(struct rbd_device *rbd_dev)
+{
+	dout("%s rbd_dev %p read lock_state %d\n", __func__, rbd_dev,
+	     rbd_dev->lock_state);
+	if (rbd_dev->lock_state != RBD_LOCK_STATE_LOCKED)
+		return false;
+
+	rbd_dev->lock_state = RBD_LOCK_STATE_RELEASING;
+	downgrade_write(&rbd_dev->lock_rwsem);
 	/*
-	 * Until adequate refresh error handling is in place, there is
-	 * not much we can do here, except warn.
+	 * Ensure that all in-flight IO is flushed.
 	 *
-	 * See http://tracker.ceph.com/issues/5040
+	 * FIXME: ceph_osdc_sync() flushes the entire OSD client, which
+	 * may be shared with other devices.
 	 */
-	ret = rbd_dev_refresh(rbd_dev);
-	if (ret)
-		rbd_warn(rbd_dev, "refresh failed: %d", ret);
+	ceph_osdc_sync(&rbd_dev->rbd_client->client->osdc);
+	up_read(&rbd_dev->lock_rwsem);
+
+	down_write(&rbd_dev->lock_rwsem);
+	dout("%s rbd_dev %p write lock_state %d\n", __func__, rbd_dev,
+	     rbd_dev->lock_state);
+	if (rbd_dev->lock_state != RBD_LOCK_STATE_RELEASING)
+		return false;
+
+	if (!rbd_unlock(rbd_dev))
+		/*
+		 * Give others a chance to grab the lock - we would re-acquire
+		 * almost immediately if we got new IO during ceph_osdc_sync()
+		 * otherwise.  We need to ack our own notifications, so this
+		 * lock_dwork will be requeued from rbd_wait_state_locked()
+		 * after wake_requests() in rbd_handle_released_lock().
+		 */
+		cancel_delayed_work(&rbd_dev->lock_dwork);
+
+	return true;
+}
+
+static void rbd_release_lock_work(struct work_struct *work)
+{
+	struct rbd_device *rbd_dev = container_of(work, struct rbd_device,
+						  unlock_work);
+
+	down_write(&rbd_dev->lock_rwsem);
+	rbd_release_lock(rbd_dev);
+	up_write(&rbd_dev->lock_rwsem);
+}
+
+static void rbd_handle_acquired_lock(struct rbd_device *rbd_dev, u8 struct_v,
+				     void **p)
+{
+	struct rbd_client_id cid = { 0 };
+
+	if (struct_v >= 2) {
+		cid.gid = ceph_decode_64(p);
+		cid