dm log userspace: allow mark requests to piggyback on flush requests

In the cluster evironment, cluster write has poor performance because
userspace_flush() has to contact a userspace program (cmirrord) for
clear/mark/flush requests. But both mark and flush requests require
cmirrord to communicate the message to all the cluster nodes for each
flush call. This behaviour is really slow.

To address this we now merge mark and flush requests together to reduce
the kernel-userspace-kernel time. We allow a new directive,
"integrated_flush" that can be used to instruct the kernel log code to
combine flush and mark requests when directed by userspace. If not
directed by userspace (due to an older version of the userspace code
perhaps), the kernel will function as it did previously - preserving
backwards compatibility. Additionally, flush requests are performed
lazily when only clear requests exist.

Signed-off-by: Dongmao Zhang <[e-mail]>
Signed-off-by: Jonathan Brassow <[e-mail]>
Signed-off-by: Alasdair G Kergon <[e-mail]>
Signed-off-by: Mike Snitzer <[e-mail]>

drivers/md/dm-log-userspace-base.c

···
 #include <linux/device-mapper.h>
 #include <linux/dm-log-userspace.h>
 #include <linux/module.h>
+#include <linux/workqueue.h>
 
 #include "dm-log-userspace-transfer.h"
 
-#define DM_LOG_USERSPACE_VSN "1.1.0"
+#define DM_LOG_USERSPACE_VSN "1.3.0"
 
 struct flush_entry {
 	int type;
···
 	spinlock_t flush_lock;
 	struct list_head mark_list;
 	struct list_head clear_list;
+
+	/*
+	 * Workqueue for flush of clear region requests.
+	 */
+	struct workqueue_struct *dmlog_wq;
+	struct delayed_work flush_log_work;
+	atomic_t sched_flush;
+
+	/*
+	 * Combine userspace flush and mark requests for efficiency.
+	 */
+	uint32_t integrated_flush;
 };
 
 static mempool_t *flush_entry_pool;
···
 
 	*ctr_str = NULL;
 
+	/*
+	 * Determine overall size of the string.
+	 */
 	for (i = 0, str_size = 0; i < argc; i++)
 		str_size += strlen(argv[i]) + 1; /* +1 for space between args */
 
···
 	return str_size;
 }
 
+static void do_flush(struct work_struct *work)
+{
+	int r;
+	struct log_c *lc = container_of(work, struct log_c, flush_log_work.work);
+
+	atomic_set(&lc->sched_flush, 0);
+
+	r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, NULL, 0, NULL, NULL);
+
+	if (r)
+		dm_table_event(lc->ti->table);
+}
+
 /*
  * userspace_ctr
  *
  * argv contains:
- *	<UUID> <other args>
- * Where 'other args' is the userspace implementation specific log
- * arguments.  An example might be:
- *	<UUID> clustered-disk <arg count> <log dev> <region_size> [[no]sync]
+ *	<UUID> [integrated_flush] <other args>
+ * Where 'other args' are the userspace implementation-specific log
+ * arguments.
  *
- * So, this module will strip off the <UUID> for identification purposes
- * when communicating with userspace about a log; but will pass on everything
- * else.
+ * Example:
+ *	<UUID> [integrated_flush] clustered-disk <arg count> <log dev>
+ *	<region_size> [[no]sync]
+ *
+ * This module strips off the <UUID> and uses it for identification
+ * purposes when communicating with userspace about a log.
+ *
+ * If integrated_flush is defined, the kernel combines flush
+ * and mark requests.
+ *
+ * The rest of the line, beginning with 'clustered-disk', is passed
+ * to the userspace ctr function.
  */
 static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
 			 unsigned argc, char **argv)
···
 		return -EINVAL;
 	}
 
+	lc->usr_argc = argc;
+
 	strncpy(lc->uuid, argv[0], DM_UUID_LEN);
+	argc--;
+	argv++;
 	spin_lock_init(&lc->flush_lock);
 	INIT_LIST_HEAD(&lc->mark_list);
 	INIT_LIST_HEAD(&lc->clear_list);
 
-	str_size = build_constructor_string(ti, argc - 1, argv + 1, &ctr_str);
+	if (!strcasecmp(argv[0], "integrated_flush")) {
+		lc->integrated_flush = 1;
+		argc--;
+		argv++;
+	}
+
+	str_size = build_constructor_string(ti, argc, argv, &ctr_str);
 	if (str_size < 0) {
 		kfree(lc);
 		return str_size;
···
 			DMERR("Failed to register %s with device-mapper",
 			      devices_rdata);
 	}
+
+	if (lc->integrated_flush) {
+		lc->dmlog_wq = alloc_workqueue("dmlogd", WQ_MEM_RECLAIM, 0);
+		if (!lc->dmlog_wq) {
+			DMERR("couldn't start dmlogd");
+			r = -ENOMEM;
+			goto out;
+		}
+
+		INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
+		atomic_set(&lc->sched_flush, 0);
+	}
+
 out:
 	kfree(devices_rdata);
 	if (r) {
···
 		kfree(ctr_str);
 	} else {
 		lc->usr_argv_str = ctr_str;
-		lc->usr_argc = argc;
 		log->context = lc;
 	}
 
···
 {
 	struct log_c *lc = log->context;
 
+	if (lc->integrated_flush) {
+		/* flush workqueue */
+		if (atomic_read(&lc->sched_flush))
+			flush_delayed_work(&lc->flush_log_work);
+
+		destroy_workqueue(lc->dmlog_wq);
+	}
+
 	(void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
-				 NULL, 0,
-				 NULL, NULL);
+				    NULL, 0, NULL, NULL);
 
 	if (lc->log_dev)
 		dm_put_device(lc->ti, lc->log_dev);
···
 	struct log_c *lc = log->context;
 
 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
-				 NULL, 0,
-				 NULL, NULL);
+				 NULL, 0, NULL, NULL);
 
 	return r;
 }
···
 	int r;
 	struct log_c *lc = log->context;
 
+	/*
+	 * Run planned flush earlier.
+	 */
+	if (lc->integrated_flush && atomic_read(&lc->sched_flush))
+		flush_delayed_work(&lc->flush_log_work);
+
 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
-				 NULL, 0,
-				 NULL, NULL);
+				 NULL, 0, NULL, NULL);
 
 	return r;
 }
···
 
 	lc->in_sync_hint = 0;
 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
-				 NULL, 0,
-				 NULL, NULL);
+				 NULL, 0, NULL, NULL);
 
 	return r;
 }
···
 	return r;
 }
 
-static int flush_by_group(struct log_c *lc, struct list_head *flush_list)
+static int flush_by_group(struct log_c *lc, struct list_head *flush_list,
+			  int flush_with_payload)
 {
 	int r = 0;
 	int count;
···
 				break;
 		}
 
-		r = userspace_do_request(lc, lc->uuid, type,
-					 (char *)(group),
-					 count * sizeof(uint64_t),
-					 NULL, NULL);
-		if (r) {
-			/* Group send failed.  Attempt one-by-one. */
-			list_splice_init(&tmp_list, flush_list);
-			r = flush_one_by_one(lc, flush_list);
-			break;
+		if (flush_with_payload) {
+			r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
+						 (char *)(group),
+						 count * sizeof(uint64_t),
+						 NULL, NULL);
+			/*
+			 * Integrated flush failed.
+			 */
+			if (r)
+				break;
+		} else {
+			r = userspace_do_request(lc, lc->uuid, type,
+						 (char *)(group),
+						 count * sizeof(uint64_t),
+						 NULL, NULL);
+			if (r) {
+				/*
+				 * Group send failed.  Attempt one-by-one.
+				 */
+				list_splice_init(&tmp_list, flush_list);
+				r = flush_one_by_one(lc, flush_list);
+				break;
+			}
 		}
 	}
 
···
 	struct log_c *lc = log->context;
 	LIST_HEAD(mark_list);
 	LIST_HEAD(clear_list);
+	int mark_list_is_empty;
+	int clear_list_is_empty;
 	struct flush_entry *fe, *tmp_fe;
 
 	spin_lock_irqsave(&lc->flush_lock, flags);
···
 	list_splice_init(&lc->clear_list, &clear_list);
 	spin_unlock_irqrestore(&lc->flush_lock, flags);
 
-	if (list_empty(&mark_list) && list_empty(&clear_list))
+	mark_list_is_empty = list_empty(&mark_list);
+	clear_list_is_empty = list_empty(&clear_list);
+
+	if (mark_list_is_empty && clear_list_is_empty)
 		return 0;
 
-	r = flush_by_group(lc, &mark_list);
+	r = flush_by_group(lc, &clear_list, 0);
 	if (r)
-		goto fail;
+		goto out;
 
-	r = flush_by_group(lc, &clear_list);
-	if (r)
-		goto fail;
+	if (!lc->integrated_flush) {
+		r = flush_by_group(lc, &mark_list, 0);
+		if (r)
+			goto out;
+		r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
+					 NULL, 0, NULL, NULL);
+		goto out;
+	}
 
-	r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
-				 NULL, 0, NULL, NULL);
-
-fail:
 	/*
-	 * We can safely remove these entries, even if failure.
+	 * Send integrated flush request with mark_list as payload.
+	 */
+	r = flush_by_group(lc, &mark_list, 1);
+	if (r)
+		goto out;
+
+	if (mark_list_is_empty && !atomic_read(&lc->sched_flush)) {
+		/*
+		 * When there are only clear region requests,
+		 * we schedule a flush in the future.
+		 */
+		queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ);
+		atomic_set(&lc->sched_flush, 1);
+	} else {
+		/*
+		 * Cancel pending flush because we
+		 * have already flushed in mark_region.
+		 */
+		cancel_delayed_work(&lc->flush_log_work);
+		atomic_set(&lc->sched_flush, 0);
+	}
+
+out:
+	/*
+	 * We can safely remove these entries, even after failure.
 	 * Calling code will receive an error and will know that
 	 * the log facility has failed.
 	 */
···
 
 	rdata_size = sizeof(pkg);
 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
-				 NULL, 0,
-				 (char *)&pkg, &rdata_size);
+				 NULL, 0, (char *)&pkg, &rdata_size);
 
 	*region = pkg.r;
 	return (r) ? r : (int)pkg.i;
···
 	pkg.i = (int64_t)in_sync;
 
 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
-				 (char *)&pkg, sizeof(pkg),
-				 NULL, NULL);
+				 (char *)&pkg, sizeof(pkg), NULL, NULL);
 
 	/*
 	 * It would be nice to be able to report failures.
···
 
 	rdata_size = sizeof(sync_count);
 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
-				 NULL, 0,
-				 (char *)&sync_count, &rdata_size);
+				 NULL, 0, (char *)&sync_count, &rdata_size);
 
 	if (r)
 		return 0;
···
 	switch (status_type) {
 	case STATUSTYPE_INFO:
 		r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
-					 NULL, 0,
-					 result, &sz);
+					 NULL, 0, result, &sz);
 
 		if (r) {
 			sz = 0;
···
 		BUG_ON(!table_args); /* There will always be a ' ' */
 		table_args++;
 
-		DMEMIT("%s %u %s %s ", log->type->name, lc->usr_argc,
-		       lc->uuid, table_args);
+		DMEMIT("%s %u %s ", log->type->name, lc->usr_argc, lc->uuid);
+		if (lc->integrated_flush)
+			DMEMIT("integrated_flush ");
+		DMEMIT("%s ", table_args);
 		break;
 	}
 	return (r) ? 0 : (int)sz;

include/uapi/linux/dm-log-userspace.h

···
  * int (*flush)(struct dm_dirty_log *log);
  *
  * Payload-to-userspace:
- *	None.
+ *	If the 'integrated_flush' directive is present in the constructor
+ *	table, the payload is as same as DM_ULOG_MARK_REGION:
+ *		uint64_t [] - region(s) to mark
+ *	else
+ *		None
  * Payload-to-kernel:
  *	None.
  *
- * No incoming or outgoing payload.  Simply flush log state to disk.
+ * If the 'integrated_flush' option was used during the creation of the
+ * log, mark region requests are carried as payload in the flush request.
+ * Piggybacking the mark requests in this way allows for fewer communications
+ * between kernel and userspace.
  *
  * When the request has been processed, user-space must return the
  * dm_ulog_request to the kernel - setting the 'error' field and clearing
···
  *	version 2:  DM_ULOG_CTR allowed to return a string containing a
  *	            device name that is to be registered with DM via
  *	            'dm_get_device'.
+ *	version 3:  DM_ULOG_FLUSH is capable of carrying payload for marking
+ *		    regions.  This "integrated flush" reduces the number of
+ *		    requests between the kernel and userspace by effectively
+ *		    merging 'mark' and 'flush' requests.  A constructor table
+ *		    argument ('integrated_flush') is required to turn this
+ *		    feature on, so it is backwards compatible with older
+ *		    userspace versions.
  */
-#define DM_ULOG_REQUEST_VERSION 2
+#define DM_ULOG_REQUEST_VERSION 3
 
 struct dm_ulog_request {
 	/*