libceph, rbd: ceph_osd_linger_request, watch/notify v2

This adds support and switches rbd to a new, more reliable version of
watch/notify protocol.  As with the OSD client update, this is mostly
about getting the right structures linked into the right places so that
reconnects are properly sent when needed.  watch/notify v2 also
requires sending regular pings to the OSDs - send_linger_ping().

A major change from the old watch/notify implementation is the
introduction of ceph_osd_linger_request - linger requests no longer
piggy back on ceph_osd_request.  ceph_osd_event has been merged into
ceph_osd_linger_request.

All the details are now hidden within libceph, the interface consists
of a simple pair of watch/unwatch functions and ceph_osdc_notify_ack().
ceph_osdc_watch() does return ceph_osd_linger_request, but only to keep
the lifetime management simple.

ceph_osdc_notify_ack() accepts an optional data payload, which is
relayed back to the notifier.

Portions of this patch are loosely based on work by Douglas Fuller
<dfuller@redhat.com> and Mike Christie <michaelc@cs.wisc.edu>.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

drivers/block/rbd.c

@@ -351,11 +351,11 @@ struct rbd_device {
 	struct rbd_options	*opts;
 
 	struct ceph_object_id	header_oid;
+	struct ceph_object_locator header_oloc;
 
 	struct ceph_file_layout	layout;
 
-	struct ceph_osd_event   *watch_event;
-	struct rbd_obj_request	*watch_request;
+	struct ceph_osd_linger_request *watch_handle;
 
 	struct rbd_spec		*parent_spec;
 	u64			parent_overlap;
@@ -1596,12 +1596,6 @@ static int rbd_obj_request_wait(struct rbd_obj_request *obj_request)
 	return __rbd_obj_request_wait(obj_request, 0);
 }
 
-static int rbd_obj_request_wait_timeout(struct rbd_obj_request *obj_request,
-					unsigned long timeout)
-{
-	return __rbd_obj_request_wait(obj_request, timeout);
-}
-
 static void rbd_img_request_complete(struct rbd_img_request *img_request)
 {
 
@@ -1751,12 +1745,6 @@ static void rbd_obj_request_complete(struct rbd_obj_request *obj_request)
 		complete_all(&obj_request->completion);
 }
 
-static void rbd_osd_trivial_callback(struct rbd_obj_request *obj_request)
-{
-	dout("%s: obj %p\n", __func__, obj_request);
-	obj_request_done_set(obj_request);
-}
-
 static void rbd_osd_read_callback(struct rbd_obj_request *obj_request)
 {
 	struct rbd_img_request *img_request = NULL;
@@ -1877,10 +1865,6 @@ static void rbd_osd_req_callback(struct ceph_osd_request *osd_req)
 	case CEPH_OSD_OP_CALL:
 		rbd_osd_call_callback(obj_request);
 		break;
-	case CEPH_OSD_OP_NOTIFY_ACK:
-	case CEPH_OSD_OP_WATCH:
-		rbd_osd_trivial_callback(obj_request);
-		break;
 	default:
 		rbd_warn(NULL, "%s: unsupported op %hu",
 			obj_request->object_name, (unsigned short) opcode);
@@ -3100,45 +3084,18 @@ static void rbd_img_parent_read(struct rbd_obj_request *obj_request)
 	obj_request_done_set(obj_request);
 }
 
-static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id)
-{
-	struct rbd_obj_request *obj_request;
-	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
-	int ret;
-
-	obj_request = rbd_obj_request_create(rbd_dev->header_oid.name, 0, 0,
-							OBJ_REQUEST_NODATA);
-	if (!obj_request)
-		return -ENOMEM;
-
-	ret = -ENOMEM;
-	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_READ, 1,
-						  obj_request);
-	if (!obj_request->osd_req)
-		goto out;
-
-	osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
-					notify_id, 0, 0);
-	rbd_osd_req_format_read(obj_request);
-
-	ret = rbd_obj_request_submit(osdc, obj_request);
-	if (ret)
-		goto out;
-	ret = rbd_obj_request_wait(obj_request);
-out:
-	rbd_obj_request_put(obj_request);
-
-	return ret;
-}
+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 void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
+static void rbd_watch_cb(void *arg, u64 notify_id, u64 cookie,
+			 u64 notifier_id, void *data, size_t data_len)
 {
-	struct rbd_device *rbd_dev = (struct rbd_device *)data;
+	struct rbd_device *rbd_dev = arg;
+	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
 	int ret;
 
-	dout("%s: \"%s\" notify_id %llu opcode %u\n", __func__,
-		rbd_dev->header_oid.name, (unsigned long long)notify_id,
-		(unsigned int)opcode);
+	dout("%s rbd_dev %p cookie %llu notify_id %llu\n", __func__, rbd_dev,
+	     cookie, notify_id);
 
 	/*
 	 * Until adequate refresh error handling is in place, there is
@@ -3150,63 +3107,31 @@ static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
 	if (ret)
 		rbd_warn(rbd_dev, "refresh failed: %d", ret);
 
-	ret = rbd_obj_notify_ack_sync(rbd_dev, notify_id);
+	ret = ceph_osdc_notify_ack(osdc, &rbd_dev->header_oid,
+				   &rbd_dev->header_oloc, notify_id, cookie,
+				   NULL, 0);
 	if (ret)
 		rbd_warn(rbd_dev, "notify_ack ret %d", ret);
 }
 
-/*
- * Send a (un)watch request and wait for the ack.  Return a request
- * with a ref held on success or error.
- */
-static struct rbd_obj_request *rbd_obj_watch_request_helper(
-						struct rbd_device *rbd_dev,
-						bool watch)
+static void rbd_watch_errcb(void *arg, u64 cookie, int err)
 {
-	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
-	struct ceph_options *opts = osdc->client->options;
-	struct rbd_obj_request *obj_request;
+	struct rbd_device *rbd_dev = arg;
 	int ret;
 
-	obj_request = rbd_obj_request_create(rbd_dev->header_oid.name, 0, 0,
-					     OBJ_REQUEST_NODATA);
-	if (!obj_request)
-		return ERR_PTR(-ENOMEM);
+	rbd_warn(rbd_dev, "encountered watch error: %d", err);
 
-	obj_request->osd_req = rbd_osd_req_create(rbd_dev, OBJ_OP_WRITE, 1,
-						  obj_request);
-	if (!obj_request->osd_req) {
-		ret = -ENOMEM;
-		goto out;
-	}
-
-	osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_WATCH,
-			      rbd_dev->watch_event->cookie, 0, watch);
-	rbd_osd_req_format_write(obj_request);
-
-	if (watch)
-		ceph_osdc_set_request_linger(osdc, obj_request->osd_req);
-
-	ret = rbd_obj_request_submit(osdc, obj_request);
-	if (ret)
-		goto out;
-
-	ret = rbd_obj_request_wait_timeout(obj_request, opts->mount_timeout);
-	if (ret)
-		goto out;
+	__rbd_dev_header_unwatch_sync(rbd_dev);
 
-	ret = obj_request->result;
+	ret = rbd_dev_header_watch_sync(rbd_dev);
 	if (ret) {
-		if (watch)
-			rbd_obj_request_end(obj_request);
-		goto out;
+		rbd_warn(rbd_dev, "failed to reregister watch: %d", ret);
+		return;
 	}
 
-	return obj_request;
-
-out:
-	rbd_obj_request_put(obj_request);
-	return ERR_PTR(ret);
+	ret = rbd_dev_refresh(rbd_dev);
+	if (ret)
+		rbd_warn(rbd_dev, "reregisteration refresh failed: %d", ret);
 }
 
 /*
@@ -3215,57 +3140,33 @@ static struct rbd_obj_request *rbd_obj_watch_request_helper(
 static int rbd_dev_header_watch_sync(struct rbd_device *rbd_dev)
 {
 	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
-	struct rbd_obj_request *obj_request;
-	int ret;
+	struct ceph_osd_linger_request *handle;
 
-	rbd_assert(!rbd_dev->watch_event);
-	rbd_assert(!rbd_dev->watch_request);
+	rbd_assert(!rbd_dev->watch_handle);
 
-	ret = ceph_osdc_create_event(osdc, rbd_watch_cb, rbd_dev,
-				     &rbd_dev->watch_event);
-	if (ret < 0)
-		return ret;
-
-	obj_request = rbd_obj_watch_request_helper(rbd_dev, true);
-	if (IS_ERR(obj_request)) {
-		ceph_osdc_cancel_event(rbd_dev->watch_event);
-		rbd_dev->watch_event = NULL;
-		return PTR_ERR(obj_request);
-	}
-
-	/*
-	 * A watch request is set to linger, so the underlying osd
-	 * request won't go away until we unregister it.  We retain
-	 * a pointer to the object request during that time (in
-	 * rbd_dev->watch_request), so we'll keep a reference to it.
-	 * We'll drop that reference after we've unregistered it in
-	 * rbd_dev_header_unwatch_sync().
-	 */
-	rbd_dev->watch_request = obj_request;
+	handle = ceph_osdc_watch(osdc, &rbd_dev->header_oid,
+				 &rbd_dev->header_oloc, rbd_watch_cb,
+				 rbd_watch_errcb, rbd_dev);
+	if (IS_ERR(handle))
+		return PTR_ERR(handle);
 
+	rbd_dev->watch_handle = handle;
 	return 0;
 }
 
 static void __rbd_dev_header_unwatch_sync(struct rbd_device *rbd_dev)
 {
-	struct rbd_obj_request *obj_request;
-
-	rbd_assert(rbd_dev->watch_event);
-	rbd_assert(rbd_dev->watch_request);
+	struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
+	int ret;
 
-	rbd_obj_request_end(rbd_dev->watch_request);
-	rbd_obj_request_put(rbd_dev->watch_request);
-	rbd_dev->watch_request = NULL;
+	if (!rbd_dev->watch_handle)
+		return;
 
-	obj_request = rbd_obj_watch_request_helper(rbd_dev, false);
-	if (!IS_ERR(obj_request))
-		rbd_obj_request_put(obj_request);
-	else
-		rbd_warn(rbd_dev, "unable to tear down watch request (%ld)",
-			 PTR_ERR(obj_request));
+	ret = ceph_osdc_unwatch(osdc, rbd_dev->watch_handle);
+	if (ret)
+		rbd_warn(rbd_dev, "failed to unwatch: %d", ret);
 
-	ceph_osdc_cancel_event(rbd_dev->watch_event);
-	rbd_dev->watch_event = NULL;
+	rbd_dev->watch_handle = NULL;
 }
 
 /*
@@ -4081,6 +3982,7 @@ static struct rbd_device *rbd_dev_create(struct rbd_client *rbdc,
 	init_rwsem(&rbd_dev->header_rwsem);
 
 	ceph_oid_init(&rbd_dev->header_oid);
+	ceph_oloc_init(&rbd_dev->header_oloc);
 
 	rbd_dev->dev.bus = &rbd_bus_type;
 	rbd_dev->dev.type = &rbd_device_type;
@@ -5285,6 +5187,7 @@ static int rbd_dev_header_name(struct rbd_device *rbd_dev)
 
 	rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
 
+	rbd_dev->header_oloc.pool = ceph_file_layout_pg_pool(rbd_dev->layout);
 	if (rbd_dev->image_format == 1)
 		ret = ceph_oid_aprintf(&rbd_dev->header_oid, GFP_KERNEL, "%s%s",
 				       spec->image_name, RBD_SUFFIX);

include/linux/ceph/ceph_fs.h

@@ -153,8 +153,9 @@ struct ceph_dir_layout {
 
 /* watch-notify operations */
 enum {
-  WATCH_NOTIFY				= 1, /* notifying watcher */
-  WATCH_NOTIFY_COMPLETE			= 2, /* notifier notified when done */
+	CEPH_WATCH_EVENT_NOTIFY		  = 1, /* notifying watcher */
+	CEPH_WATCH_EVENT_NOTIFY_COMPLETE  = 2, /* notifier notified when done */
+	CEPH_WATCH_EVENT_DISCONNECT       = 3, /* we were disconnected */
 };
 
 

include/linux/ceph/osd_client.h

@@ -34,7 +34,7 @@ struct ceph_osd {
 	struct rb_node o_node;
 	struct ceph_connection o_con;
 	struct rb_root o_requests;
-	struct list_head o_linger_requests;
+	struct rb_root o_linger_requests;
 	struct list_head o_osd_lru;
 	struct ceph_auth_handshake o_auth;
 	unsigned long lru_ttl;
@@ -108,11 +108,12 @@ struct ceph_osd_req_op {
 		} cls;
 		struct {
 			u64 cookie;
-			u64 ver;
-			u32 prot_ver;
-			u32 timeout;
-			__u8 flag;
+			__u8 op;           /* CEPH_OSD_WATCH_OP_ */
+			u32 gen;
 		} watch;
+		struct {
+			struct ceph_osd_data request_data;
+		} notify_ack;
 		struct {
 			u64 expected_object_size;
 			u64 expected_write_size;
@@ -145,8 +146,6 @@ struct ceph_osd_request_target {
 struct ceph_osd_request {
 	u64             r_tid;              /* unique for this client */
 	struct rb_node  r_node;
-	struct list_head r_linger_item;
-	struct list_head r_linger_osd_item;
 	struct ceph_osd *r_osd;
 
 	struct ceph_osd_request_target r_t;
@@ -162,7 +161,6 @@ struct ceph_osd_request {
 
 	int               r_result;
 	bool              r_got_reply;
-	int		  r_linger;
 
 	struct ceph_osd_client *r_osdc;
 	struct kref       r_kref;
@@ -181,6 +179,7 @@ struct ceph_osd_request {
 	struct ceph_snap_context *r_snapc;    /* for writes */
 	struct timespec r_mtime;              /* ditto */
 	u64 r_data_offset;                    /* ditto */
+	bool r_linger;                        /* don't resend on failure */
 
 	/* internal */
 	unsigned long r_stamp;                /* jiffies, send or check time */
@@ -195,23 +194,40 @@ struct ceph_request_redirect {
 	struct ceph_object_locator oloc;
 };
 
-struct ceph_osd_event {
-	u64 cookie;
-	int one_shot;
+typedef void (*rados_watchcb2_t)(void *arg, u64 notify_id, u64 cookie,
+				 u64 notifier_id, void *data, size_t data_len);
+typedef void (*rados_watcherrcb_t)(void *arg, u64 cookie, int err);
+
+struct ceph_osd_linger_request {
 	struct ceph_osd_client *osdc;
-	void (*cb)(u64, u64, u8, void *);
-	void *data;
-	struct rb_node node;
-	struct list_head osd_node;
+	u64 linger_id;
+	bool committed;
+
+	struct ceph_osd *osd;
+	struct ceph_osd_request *reg_req;
+	struct ceph_osd_request *ping_req;
+	unsigned long ping_sent;
+
+	struct ceph_osd_request_target t;
+	u32 last_force_resend;
+
+	struct timespec mtime;
+
 	struct kref kref;
-};
+	struct mutex lock;
+	struct rb_node node;            /* osd */
+	struct rb_node osdc_node;       /* osdc */
+	struct list_head scan_item;
+
+	struct completion reg_commit_wait;
+	int reg_commit_error;
+	int last_error;
 
-struct ceph_osd_event_work {
-	struct work_struct work;
-	struct ceph_osd_event *event;
-        u64 ver;
-        u64 notify_id;
-        u8 opcode;
+	u32 register_gen;
+
+	rados_watchcb2_t wcb;
+	rados_watcherrcb_t errcb;
+	void *data;
 };
 
 struct ceph_osd_client {
@@ -223,9 +239,10 @@ struct ceph_osd_client {
 	struct rb_root         osds;          /* osds */
 	struct list_head       osd_lru;       /* idle osds */
 	spinlock_t             osd_lru_lock;
-	struct list_head       req_linger;    /* lingering requests */
 	struct ceph_osd        homeless_osd;
 	atomic64_t             last_tid;      /* tid of last request */
+	u64                    last_linger_id;
+	struct rb_root         linger_requests; /* lingering requests */
 	atomic_t               num_requests;
 	atomic_t               num_homeless;
 	struct delayed_work    timeout_work;
@@ -239,10 +256,6 @@ struct ceph_osd_client {
 	struct ceph_msgpool	msgpool_op;
 	struct ceph_msgpool	msgpool_op_reply;
 
-	spinlock_t		event_lock;
-	struct rb_root		event_tree;
-	u64			event_count;
-
 	struct workqueue_struct	*notify_wq;
 };
 
@@ -314,9 +327,6 @@ extern void osd_req_op_cls_init(struct ceph_osd_request *osd_req,
 extern int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
 				 u16 opcode, const char *name, const void *value,
 				 size_t size, u8 cmp_op, u8 cmp_mode);
-extern void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
-					unsigned int which, u16 opcode,
-					u64 cookie, u64 version, int flag);
 extern void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
 				       unsigned int which,
 				       u64 expected_object_size,
@@ -339,9 +349,6 @@ extern struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *,
 				      u32 truncate_seq, u64 truncate_size,
 				      bool use_mempool);
 
-extern void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
-					 struct ceph_osd_request *req);
-
 extern void ceph_osdc_get_request(struct ceph_osd_request *req);
 extern void ceph_osdc_put_request(struct ceph_osd_request *req);
 
@@ -372,11 +379,23 @@ extern int ceph_osdc_writepages(struct ceph_osd_client *osdc,
 				struct timespec *mtime,
 				struct page **pages, int nr_pages);
 
-/* watch/notify events */
-extern int ceph_osdc_create_event(struct ceph_osd_client *osdc,
-				  void (*event_cb)(u64, u64, u8, void *),
-				  void *data, struct ceph_osd_event **pevent);
-extern void ceph_osdc_cancel_event(struct ceph_osd_event *event);
-extern void ceph_osdc_put_event(struct ceph_osd_event *event);
+/* watch/notify */
+struct ceph_osd_linger_request *
+ceph_osdc_watch(struct ceph_osd_client *osdc,
+		struct ceph_object_id *oid,
+		struct ceph_object_locator *oloc,
+		rados_watchcb2_t wcb,
+		rados_watcherrcb_t errcb,
+		void *data);
+int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
+		      struct ceph_osd_linger_request *lreq);
+
+int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
+			 struct ceph_object_id *oid,
+			 struct ceph_object_locator *oloc,
+			 u64 notify_id,
+			 u64 cookie,
+			 void *payload,
+			 size_t payload_len);
 #endif
 

include/linux/ceph/rados.h

@@ -427,7 +427,17 @@ enum {
 	CEPH_OSD_CMPXATTR_MODE_U64    = 2
 };
 
-#define RADOS_NOTIFY_VER	1
+enum {
+	CEPH_OSD_WATCH_OP_UNWATCH = 0,
+	CEPH_OSD_WATCH_OP_LEGACY_WATCH = 1,
+	/* note: use only ODD ids to prevent pre-giant code from
+	   interpreting the op as UNWATCH */
+	CEPH_OSD_WATCH_OP_WATCH = 3,
+	CEPH_OSD_WATCH_OP_RECONNECT = 5,
+	CEPH_OSD_WATCH_OP_PING = 7,
+};
+
+const char *ceph_osd_watch_op_name(int o);
 
 /*
  * an individual object operation.  each may be accompanied by some data
@@ -462,8 +472,9 @@ struct ceph_osd_op {
 	        } __attribute__ ((packed)) snap;
 		struct {
 			__le64 cookie;
-			__le64 ver;
-			__u8 flag;	/* 0 = unwatch, 1 = watch */
+			__le64 ver;     /* no longer used */
+			__u8 op;	/* CEPH_OSD_WATCH_OP_* */
+			__le32 gen;     /* registration generation */
 		} __attribute__ ((packed)) watch;
 		struct {
 			__le64 offset, length;

net/ceph/ceph_strings.c

@@ -27,6 +27,22 @@ __CEPH_FORALL_OSD_OPS(GENERATE_CASE)
 	}
 }
 
+const char *ceph_osd_watch_op_name(int o)
+{
+	switch (o) {
+	case CEPH_OSD_WATCH_OP_UNWATCH:
+		return "unwatch";
+	case CEPH_OSD_WATCH_OP_WATCH:
+		return "watch";
+	case CEPH_OSD_WATCH_OP_RECONNECT:
+		return "reconnect";
+	case CEPH_OSD_WATCH_OP_PING:
+		return "ping";
+	default:
+		return "???";
+	}
+}
+
 const char *ceph_osd_state_name(int s)
 {
 	switch (s) {

net/ceph/debugfs.c

@@ -177,6 +177,9 @@ static void dump_request(struct seq_file *s, struct ceph_osd_request *req)
 
 		seq_printf(s, "%s%s", (i == 0 ? "\t" : ","),
 			   ceph_osd_op_name(op->op));
+		if (op->op == CEPH_OSD_OP_WATCH)
+			seq_printf(s, "-%s",
+				   ceph_osd_watch_op_name(op->watch.op));
 	}
 
 	seq_putc(s, '\n');
@@ -197,6 +200,31 @@ static void dump_requests(struct seq_file *s, struct ceph_osd *osd)
 	mutex_unlock(&osd->lock);
 }
 
+static void dump_linger_request(struct seq_file *s,
+				struct ceph_osd_linger_request *lreq)
+{
+	seq_printf(s, "%llu\t", lreq->linger_id);
+	dump_target(s, &lreq->t);
+
+	seq_printf(s, "\t%u\t%s/%d\n", lreq->register_gen,
+		   lreq->committed ? "C" : "", lreq->last_error);
+}
+
+static void dump_linger_requests(struct seq_file *s, struct ceph_osd *osd)
+{
+	struct rb_node *n;
+
+	mutex_lock(&osd->lock);
+	for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
+		struct ceph_osd_linger_request *lreq =
+		    rb_entry(n, struct ceph_osd_linger_request, node);
+
+		dump_linger_request(s, lreq);
+	}
+
+	mutex_unlock(&osd->lock);
+}
+
 static int osdc_show(struct seq_file *s, void *pp)
 {
 	struct ceph_client *client = s->private;
@@ -214,6 +242,14 @@ static int osdc_show(struct seq_file *s, void *pp)
 	}
 	dump_requests(s, &osdc->homeless_osd);
 
+	seq_puts(s, "LINGER REQUESTS\n");
+	for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
+		struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
+
+		dump_linger_requests(s, osd);
+	}
+	dump_linger_requests(s, &osdc->homeless_osd);
+
 	up_read(&osdc->lock);
 	return 0;
 }