Add support for per dataset zil stats and use wmsum counters

ZIL kstats are reported in an inclusive way, i.e., same counters are
shared to capture all the activities happening in zil. Added support
to report zil stats for every datset individually by combining them
with already exposed dataset kstats.

Wmsum uses per cpu counters and provide less overhead as compared
to atomic operations. Updated zil kstats to replace wmsum counters
to avoid atomic operations.

Reviewed-by: Christian Schwarz <christian.schwarz@nutanix.com>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Ameer Hamza <ahamza@ixsystems.com>
Closes #13636

cmd/ztest.c

@@ -2908,7 +2908,7 @@ ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
 	zil_close(zd->zd_zilog);
 
 	/* zfsvfs_setup() */
-	VERIFY3P(zil_open(os, ztest_get_data), ==, zd->zd_zilog);
+	VERIFY3P(zil_open(os, ztest_get_data, NULL), ==, zd->zd_zilog);
 	zil_replay(os, zd, ztest_replay_vector);
 
 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
@@ -4378,7 +4378,7 @@ ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
 	/*
 	 * Open the intent log for it.
 	 */
-	zilog = zil_open(os, ztest_get_data);
+	zilog = zil_open(os, ztest_get_data, NULL);
 
 	/*
 	 * Put some objects in there, do a little I/O to them,
@@ -7304,7 +7304,7 @@ ztest_dataset_open(int d)
 		    zilog->zl_parse_lr_count,
 		    zilog->zl_replaying_seq);
 
-	zilog = zil_open(os, ztest_get_data);
+	zilog = zil_open(os, ztest_get_data, NULL);
 
 	if (zilog->zl_replaying_seq != 0 &&
 	    zilog->zl_replaying_seq < committed_seq)

include/sys/dataset_kstats.h

@@ -30,6 +30,7 @@
 #include <sys/wmsum.h>
 #include <sys/dmu.h>
 #include <sys/kstat.h>
+#include <sys/zil.h>
 
 typedef struct dataset_sum_stats_t {
 	wmsum_t dss_writes;
@@ -56,14 +57,19 @@ typedef struct dataset_kstat_values {
 	 * entry is removed from the unlinked set
 	 */
 	kstat_named_t dkv_nunlinked;
+	/*
+	 * Per dataset zil kstats
+	 */
+	zil_kstat_values_t dkv_zil_stats;
 } dataset_kstat_values_t;
 
 typedef struct dataset_kstats {
 	dataset_sum_stats_t dk_sums;
+	zil_sums_t dk_zil_sums;
 	kstat_t *dk_kstats;
 } dataset_kstats_t;
 
-void dataset_kstats_create(dataset_kstats_t *, objset_t *);
+int dataset_kstats_create(dataset_kstats_t *, objset_t *);
 void dataset_kstats_destroy(dataset_kstats_t *);
 
 void dataset_kstats_update_write_kstats(dataset_kstats_t *, int64_t);

include/sys/zil.h

@@ -33,6 +33,7 @@
 #include <sys/zio.h>
 #include <sys/dmu.h>
 #include <sys/zio_crypt.h>
+#include <sys/wmsum.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -472,12 +473,34 @@ typedef struct zil_stats {
 	 */
 	kstat_named_t zil_itx_metaslab_slog_count;
 	kstat_named_t zil_itx_metaslab_slog_bytes;
-} zil_stats_t;
+} zil_kstat_values_t;
 
-#define	ZIL_STAT_INCR(stat, val) \
+typedef struct zil_sums {
-    atomic_add_64(&zil_stats.stat.value.ui64, (val));
+	wmsum_t zil_commit_count;
-#define	ZIL_STAT_BUMP(stat) \
+	wmsum_t zil_commit_writer_count;
-    ZIL_STAT_INCR(stat, 1);
+	wmsum_t zil_itx_count;
+	wmsum_t zil_itx_indirect_count;
+	wmsum_t zil_itx_indirect_bytes;
+	wmsum_t zil_itx_copied_count;
+	wmsum_t zil_itx_copied_bytes;
+	wmsum_t zil_itx_needcopy_count;
+	wmsum_t zil_itx_needcopy_bytes;
+	wmsum_t zil_itx_metaslab_normal_count;
+	wmsum_t zil_itx_metaslab_normal_bytes;
+	wmsum_t zil_itx_metaslab_slog_count;
+	wmsum_t zil_itx_metaslab_slog_bytes;
+} zil_sums_t;
+
+#define	ZIL_STAT_INCR(zil, stat, val) \
+	do { \
+		int64_t tmpval = (val); \
+		wmsum_add(&(zil_sums_global.stat), tmpval); \
+		if ((zil)->zl_sums) \
+			wmsum_add(&((zil)->zl_sums->stat), tmpval); \
+	} while (0)
+
+#define	ZIL_STAT_BUMP(zil, stat) \
+    ZIL_STAT_INCR(zil, stat, 1);
 
 typedef int zil_parse_blk_func_t(zilog_t *zilog, const blkptr_t *bp, void *arg,
     uint64_t txg);
@@ -497,7 +520,8 @@ extern void	zil_fini(void);
 extern zilog_t	*zil_alloc(objset_t *os, zil_header_t *zh_phys);
 extern void	zil_free(zilog_t *zilog);
 
-extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data);
+extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data,
+    zil_sums_t *zil_sums);
 extern void	zil_close(zilog_t *zilog);
 
 extern void	zil_replay(objset_t *os, void *arg,
@@ -537,6 +561,11 @@ extern void	zil_set_logbias(zilog_t *zilog, uint64_t slogval);
 extern uint64_t	zil_max_copied_data(zilog_t *zilog);
 extern uint64_t	zil_max_log_data(zilog_t *zilog);
 
+extern void zil_sums_init(zil_sums_t *zs);
+extern void zil_sums_fini(zil_sums_t *zs);
+extern void zil_kstat_values_update(zil_kstat_values_t *zs,
+    zil_sums_t *zil_sums);
+
 extern int zil_replay_disable;
 
 #ifdef	__cplusplus

include/sys/zil_impl.h

@@ -222,6 +222,9 @@ struct zilog {
 	 * (see zil_max_copied_data()).
 	 */
 	uint64_t	zl_max_block_size;
+
+	/* Pointer for per dataset zil sums */
+	zil_sums_t *zl_sums;
 };
 
 typedef struct zil_bp_node {

module/os/freebsd/zfs/zfs_vfsops.c

@@ -1027,8 +1027,6 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 	if (error)
 		return (error);
 
-	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
-
 	/*
 	 * If we are not mounting (ie: online recv), then we don't
 	 * have to worry about replaying the log as we blocked all
@@ -1038,7 +1036,11 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 		boolean_t readonly;
 
 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
-		dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		if (error)
+			return (error);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 
 		/*
 		 * During replay we remove the read only flag to
@@ -1109,6 +1111,10 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 		/* restore readonly bit */
 		if (readonly != 0)
 			zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
+	} else {
+		ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 	}
 
 	/*

module/os/freebsd/zfs/zvol_os.c

@@ -1189,7 +1189,7 @@ zvol_ensure_zilog(zvol_state_t *zv)
 		}
 		if (zv->zv_zilog == NULL) {
 			zv->zv_zilog = zil_open(zv->zv_objset,
-			    zvol_get_data);
+			    zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 			zv->zv_flags |= ZVOL_WRITTEN_TO;
 			/* replay / destroy done in zvol_os_create_minor() */
 			VERIFY0(zv->zv_zilog->zl_header->zh_flags &
@@ -1422,8 +1422,12 @@ zvol_os_create_minor(const char *name)
 	zv->zv_volsize = volsize;
 	zv->zv_objset = os;
 
+	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
+	error = dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
+	if (error)
+		goto out_dmu_objset_disown;
 	ASSERT3P(zv->zv_zilog, ==, NULL);
-	zv->zv_zilog = zil_open(os, zvol_get_data);
+	zv->zv_zilog = zil_open(os, zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 	if (spa_writeable(dmu_objset_spa(os))) {
 		if (zil_replay_disable)
 			zil_destroy(zv->zv_zilog, B_FALSE);
@@ -1432,8 +1436,6 @@ zvol_os_create_minor(const char *name)
 	}
 	zil_close(zv->zv_zilog);
 	zv->zv_zilog = NULL;
-	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
-	dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
 
 	/* TODO: prefetch for geom tasting */
 

module/os/linux/zfs/zfs_vfsops.c

@@ -848,8 +848,6 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 	if (error)
 		return (error);
 
-	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
-
 	/*
 	 * If we are not mounting (ie: online recv), then we don't
 	 * have to worry about replaying the log as we blocked all
@@ -857,7 +855,11 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 	 */
 	if (mounting) {
 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
-		dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		if (error)
+			return (error);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 
 		/*
 		 * During replay we remove the read only flag to
@@ -921,6 +923,10 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 		/* restore readonly bit */
 		if (readonly != 0)
 			readonly_changed_cb(zfsvfs, B_TRUE);
+	} else {
+		ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 	}
 
 	/*

module/os/linux/zfs/zvol_os.c

@@ -558,7 +558,7 @@ zvol_request_impl(zvol_state_t *zv, struct bio *bio, struct request *rq,
 			rw_enter(&zv->zv_suspend_lock, RW_WRITER);
 			if (zv->zv_zilog == NULL) {
 				zv->zv_zilog = zil_open(zv->zv_objset,
-				    zvol_get_data);
+				    zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 				zv->zv_flags |= ZVOL_WRITTEN_TO;
 				/* replay / destroy done in zvol_create_minor */
 				VERIFY0((zv->zv_zilog->zl_header->zh_flags &
@@ -1408,8 +1408,12 @@ zvol_os_create_minor(const char *name)
 	blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, zv->zv_zso->zvo_queue);
 #endif
 
+	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
+	error = dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
+	if (error)
+		goto out_dmu_objset_disown;
 	ASSERT3P(zv->zv_zilog, ==, NULL);
-	zv->zv_zilog = zil_open(os, zvol_get_data);
+	zv->zv_zilog = zil_open(os, zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 	if (spa_writeable(dmu_objset_spa(os))) {
 		if (zil_replay_disable)
 			zil_destroy(zv->zv_zilog, B_FALSE);
@@ -1418,8 +1422,6 @@ zvol_os_create_minor(const char *name)
 	}
 	zil_close(zv->zv_zilog);
 	zv->zv_zilog = NULL;
-	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
-	dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
 
 	/*
 	 * When udev detects the addition of the device it will immediately

module/zfs/dataset_kstats.c

@@ -37,18 +37,33 @@ static dataset_kstat_values_t empty_dataset_kstats = {
 	{ "nread",	KSTAT_DATA_UINT64 },
 	{ "nunlinks",	KSTAT_DATA_UINT64 },
 	{ "nunlinked",	KSTAT_DATA_UINT64 },
+	{
+	{ "zil_commit_count",			KSTAT_DATA_UINT64 },
+	{ "zil_commit_writer_count",		KSTAT_DATA_UINT64 },
+	{ "zil_itx_count",			KSTAT_DATA_UINT64 },
+	{ "zil_itx_indirect_count",		KSTAT_DATA_UINT64 },
+	{ "zil_itx_indirect_bytes",		KSTAT_DATA_UINT64 },
+	{ "zil_itx_copied_count",		KSTAT_DATA_UINT64 },
+	{ "zil_itx_copied_bytes",		KSTAT_DATA_UINT64 },
+	{ "zil_itx_needcopy_count",		KSTAT_DATA_UINT64 },
+	{ "zil_itx_needcopy_bytes",		KSTAT_DATA_UINT64 },
+	{ "zil_itx_metaslab_normal_count",	KSTAT_DATA_UINT64 },
+	{ "zil_itx_metaslab_normal_bytes",	KSTAT_DATA_UINT64 },
+	{ "zil_itx_metaslab_slog_count",	KSTAT_DATA_UINT64 },
+	{ "zil_itx_metaslab_slog_bytes",	KSTAT_DATA_UINT64 }
+	}
 };
 
 static int
 dataset_kstats_update(kstat_t *ksp, int rw)
 {
 	dataset_kstats_t *dk = ksp->ks_private;
-	ASSERT3P(dk->dk_kstats->ks_data, ==, ksp->ks_data);
+	dataset_kstat_values_t *dkv = ksp->ks_data;
+	ASSERT3P(dk->dk_kstats->ks_data, ==, dkv);
 
 	if (rw == KSTAT_WRITE)
 		return (EACCES);
 
-	dataset_kstat_values_t *dkv = dk->dk_kstats->ks_data;
 	dkv->dkv_writes.value.ui64 =
 	    wmsum_value(&dk->dk_sums.dss_writes);
 	dkv->dkv_nwritten.value.ui64 =
@@ -62,10 +77,12 @@ dataset_kstats_update(kstat_t *ksp, int rw)
 	dkv->dkv_nunlinked.value.ui64 =
 	    wmsum_value(&dk->dk_sums.dss_nunlinked);
 
+	zil_kstat_values_update(&dkv->dkv_zil_stats, &dk->dk_zil_sums);
+
 	return (0);
 }
 
-void
+int
 dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
 {
 	/*
@@ -75,7 +92,7 @@ dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
 	 * a filesystem with many snapshots, we skip them for now.
 	 */
 	if (dmu_objset_is_snapshot(objset))
-		return;
+		return (0);
 
 	/*
 	 * At the time of this writing, KSTAT_STRLEN is 255 in Linux,
@@ -94,13 +111,13 @@ dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
 		zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
 		    " snprintf() for kstat module name returned %d",
 		    (unsigned long long)dmu_objset_id(objset), n);
-		return;
+		return (SET_ERROR(EINVAL));
 	} else if (n >= KSTAT_STRLEN) {
 		zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
 		    "kstat module name length (%d) exceeds limit (%d)",
 		    (unsigned long long)dmu_objset_id(objset),
 		    n, KSTAT_STRLEN);
-		return;
+		return (SET_ERROR(ENAMETOOLONG));
 	}
 
 	char kstat_name[KSTAT_STRLEN];
@@ -110,7 +127,7 @@ dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
 		zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
 		    " snprintf() for kstat name returned %d",
 		    (unsigned long long)dmu_objset_id(objset), n);
-		return;
+		return (SET_ERROR(EINVAL));
 	}
 	ASSERT3U(n, <, KSTAT_STRLEN);
 
@@ -119,7 +136,7 @@ dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
 	    sizeof (empty_dataset_kstats) / sizeof (kstat_named_t),
 	    KSTAT_FLAG_VIRTUAL);
 	if (kstat == NULL)
-		return;
+		return (SET_ERROR(ENOMEM));
 
 	dataset_kstat_values_t *dk_kstats =
 	    kmem_alloc(sizeof (empty_dataset_kstats), KM_SLEEP);
@@ -137,15 +154,17 @@ dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
 	kstat->ks_private = dk;
 	kstat->ks_data_size += ZFS_MAX_DATASET_NAME_LEN;
 
-	kstat_install(kstat);
-	dk->dk_kstats = kstat;
-
 	wmsum_init(&dk->dk_sums.dss_writes, 0);
 	wmsum_init(&dk->dk_sums.dss_nwritten, 0);
 	wmsum_init(&dk->dk_sums.dss_reads, 0);
 	wmsum_init(&dk->dk_sums.dss_nread, 0);
 	wmsum_init(&dk->dk_sums.dss_nunlinks, 0);
 	wmsum_init(&dk->dk_sums.dss_nunlinked, 0);
+	zil_sums_init(&dk->dk_zil_sums);
+
+	dk->dk_kstats = kstat;
+	kstat_install(kstat);
+	return (0);
 }
 
 void
@@ -155,19 +174,19 @@ dataset_kstats_destroy(dataset_kstats_t *dk)
 		return;
 
 	dataset_kstat_values_t *dkv = dk->dk_kstats->ks_data;
+	kstat_delete(dk->dk_kstats);
+	dk->dk_kstats = NULL;
 	kmem_free(KSTAT_NAMED_STR_PTR(&dkv->dkv_ds_name),
 	    KSTAT_NAMED_STR_BUFLEN(&dkv->dkv_ds_name));
 	kmem_free(dkv, sizeof (empty_dataset_kstats));
 
-	kstat_delete(dk->dk_kstats);
-	dk->dk_kstats = NULL;
-
 	wmsum_fini(&dk->dk_sums.dss_writes);
 	wmsum_fini(&dk->dk_sums.dss_nwritten);
 	wmsum_fini(&dk->dk_sums.dss_reads);
 	wmsum_fini(&dk->dk_sums.dss_nread);
 	wmsum_fini(&dk->dk_sums.dss_nunlinks);
 	wmsum_fini(&dk->dk_sums.dss_nunlinked);
+	zil_sums_fini(&dk->dk_zil_sums);
 }
 
 void

module/zfs/zil.c

@@ -43,6 +43,7 @@
 #include <sys/metaslab.h>
 #include <sys/trace_zfs.h>
 #include <sys/abd.h>
+#include <sys/wmsum.h>
 
 /*
  * The ZFS Intent Log (ZIL) saves "transaction records" (itxs) of system
@@ -94,7 +95,7 @@ static int zfs_commit_timeout_pct = 5;
 /*
  * See zil.h for more information about these fields.
  */
-static zil_stats_t zil_stats = {
+static zil_kstat_values_t zil_stats = {
 	{ "zil_commit_count",			KSTAT_DATA_UINT64 },
 	{ "zil_commit_writer_count",		KSTAT_DATA_UINT64 },
 	{ "zil_itx_count",			KSTAT_DATA_UINT64 },
@@ -110,7 +111,8 @@ static zil_stats_t zil_stats = {
 	{ "zil_itx_metaslab_slog_bytes",	KSTAT_DATA_UINT64 },
 };
 
-static kstat_t *zil_ksp;
+static zil_sums_t zil_sums_global;
+static kstat_t *zil_kstats_global;
 
 /*
  * Disable intent logging replay.  This global ZIL switch affects all pools.
@@ -213,6 +215,21 @@ zil_init_log_chain(zilog_t *zilog, blkptr_t *bp)
 	zc->zc_word[ZIL_ZC_SEQ] = 1ULL;
 }
 
+static int
+zil_kstats_global_update(kstat_t *ksp, int rw)
+{
+	zil_kstat_values_t *zs = ksp->ks_data;
+	ASSERT3P(&zil_stats, ==, zs);
+
+	if (rw == KSTAT_WRITE) {
+		return (SET_ERROR(EACCES));
+	}
+
+	zil_kstat_values_update(zs, &zil_sums_global);
+
+	return (0);
+}
+
 /*
  * Read a log block and make sure it's valid.
  */
@@ -337,6 +354,73 @@ zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf)
 	return (error);
 }
 
+void
+zil_sums_init(zil_sums_t *zs)
+{
+	wmsum_init(&zs->zil_commit_count, 0);
+	wmsum_init(&zs->zil_commit_writer_count, 0);
+	wmsum_init(&zs->zil_itx_count, 0);
+	wmsum_init(&zs->zil_itx_indirect_count, 0);
+	wmsum_init(&zs->zil_itx_indirect_bytes, 0);
+	wmsum_init(&zs->zil_itx_copied_count, 0);
+	wmsum_init(&zs->zil_itx_copied_bytes, 0);
+	wmsum_init(&zs->zil_itx_needcopy_count, 0);
+	wmsum_init(&zs->zil_itx_needcopy_bytes, 0);
+	wmsum_init(&zs->zil_itx_metaslab_normal_count, 0);
+	wmsum_init(&zs->zil_itx_metaslab_normal_bytes, 0);
+	wmsum_init(&zs->zil_itx_metaslab_slog_count, 0);
+	wmsum_init(&zs->zil_itx_metaslab_slog_bytes, 0);
+}
+
+void
+zil_sums_fini(zil_sums_t *zs)
+{
+	wmsum_fini(&zs->zil_commit_count);
+	wmsum_fini(&zs->zil_commit_writer_count);
+	wmsum_fini(&zs->zil_itx_count);
+	wmsum_fini(&zs->zil_itx_indirect_count);
+	wmsum_fini(&zs->zil_itx_indirect_bytes);
+	wmsum_fini(&zs->zil_itx_copied_count);
+	wmsum_fini(&zs->zil_itx_copied_bytes);
+	wmsum_fini(&zs->zil_itx_needcopy_count);
+	wmsum_fini(&zs->zil_itx_needcopy_bytes);
+	wmsum_fini(&zs->zil_itx_metaslab_normal_count);
+	wmsum_fini(&zs->zil_itx_metaslab_normal_bytes);
+	wmsum_fini(&zs->zil_itx_metaslab_slog_count);
+	wmsum_fini(&zs->zil_itx_metaslab_slog_bytes);
+}
+
+void
+zil_kstat_values_update(zil_kstat_values_t *zs, zil_sums_t *zil_sums)
+{
+	zs->zil_commit_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_commit_count);
+	zs->zil_commit_writer_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_commit_writer_count);
+	zs->zil_itx_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_count);
+	zs->zil_itx_indirect_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_indirect_count);
+	zs->zil_itx_indirect_bytes.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_indirect_bytes);
+	zs->zil_itx_copied_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_copied_count);
+	zs->zil_itx_copied_bytes.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_copied_bytes);
+	zs->zil_itx_needcopy_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_needcopy_count);
+	zs->zil_itx_needcopy_bytes.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_needcopy_bytes);
+	zs->zil_itx_metaslab_normal_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_metaslab_normal_count);
+	zs->zil_itx_metaslab_normal_bytes.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_metaslab_normal_bytes);
+	zs->zil_itx_metaslab_slog_count.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_metaslab_slog_count);
+	zs->zil_itx_metaslab_slog_bytes.value.ui64 =
+	    wmsum_value(&zil_sums->zil_itx_metaslab_slog_bytes);
+}
+
 /*
  * Parse the intent log, and call parse_func for each valid record within.
  */
@@ -1644,11 +1728,13 @@ zil_lwb_write_issue(zilog_t *zilog, lwb_t *lwb)
 	BP_ZERO(bp);
 	error = zio_alloc_zil(spa, zilog->zl_os, txg, bp, zil_blksz, &slog);
 	if (slog) {
-		ZIL_STAT_BUMP(zil_itx_metaslab_slog_count);
+		ZIL_STAT_BUMP(zilog, zil_itx_metaslab_slog_count);
-		ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused);
+		ZIL_STAT_INCR(zilog, zil_itx_metaslab_slog_bytes,
+		    lwb->lwb_nused);
 	} else {
-		ZIL_STAT_BUMP(zil_itx_metaslab_normal_count);
+		ZIL_STAT_BUMP(zilog, zil_itx_metaslab_normal_count);
-		ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused);
+		ZIL_STAT_INCR(zilog, zil_itx_metaslab_normal_bytes,
+		    lwb->lwb_nused);
 	}
 	if (error == 0) {
 		ASSERT3U(bp->blk_birth, ==, txg);
@@ -1818,7 +1904,7 @@ cont:
 	lrcb = (lr_t *)lr_buf;		/* Like lrc, but inside lwb. */
 	lrwb = (lr_write_t *)lrcb;	/* Like lrw, but inside lwb. */
 
-	ZIL_STAT_BUMP(zil_itx_count);
+	ZIL_STAT_BUMP(zilog, zil_itx_count);
 
 	/*
 	 * If it's a write, fetch the data or get its blkptr as appropriate.
@@ -1827,8 +1913,9 @@ cont:
 		if (txg > spa_freeze_txg(zilog->zl_spa))
 			txg_wait_synced(zilog->zl_dmu_pool, txg);
 		if (itx->itx_wr_state == WR_COPIED) {
-			ZIL_STAT_BUMP(zil_itx_copied_count);
+			ZIL_STAT_BUMP(zilog, zil_itx_copied_count);
-			ZIL_STAT_INCR(zil_itx_copied_bytes, lrw->lr_length);
+			ZIL_STAT_INCR(zilog, zil_itx_copied_bytes,
+			    lrw->lr_length);
 		} else {
 			char *dbuf;
 			int error;
@@ -1840,13 +1927,14 @@ cont:
 					lrwb->lr_length = dnow;
 				lrw->lr_offset += dnow;
 				lrw->lr_length -= dnow;
-				ZIL_STAT_BUMP(zil_itx_needcopy_count);
+				ZIL_STAT_BUMP(zilog, zil_itx_needcopy_count);
-				ZIL_STAT_INCR(zil_itx_needcopy_bytes, dnow);
+				ZIL_STAT_INCR(zilog, zil_itx_needcopy_bytes,
+				    dnow);
 			} else {
 				ASSERT3S(itx->itx_wr_state, ==, WR_INDIRECT);
 				dbuf = NULL;
-				ZIL_STAT_BUMP(zil_itx_indirect_count);
+				ZIL_STAT_BUMP(zilog, zil_itx_indirect_count);
-				ZIL_STAT_INCR(zil_itx_indirect_bytes,
+				ZIL_STAT_INCR(zilog, zil_itx_indirect_bytes,
 				    lrw->lr_length);
 			}
 
@@ -2611,7 +2699,7 @@ zil_commit_writer(zilog_t *zilog, zil_commit_waiter_t *zcw)
 		goto out;
 	}
 
-	ZIL_STAT_BUMP(zil_commit_writer_count);
+	ZIL_STAT_BUMP(zilog, zil_commit_writer_count);
 
 	zil_get_commit_list(zilog);
 	zil_prune_commit_list(zilog);
@@ -3088,7 +3176,7 @@ zil_commit(zilog_t *zilog, uint64_t foid)
 void
 zil_commit_impl(zilog_t *zilog, uint64_t foid)
 {
-	ZIL_STAT_BUMP(zil_commit_count);
+	ZIL_STAT_BUMP(zilog, zil_commit_count);
 
 	/*
 	 * Move the "async" itxs for the specified foid to the "sync"
@@ -3271,13 +3359,16 @@ zil_init(void)
 	zil_zcw_cache = kmem_cache_create("zil_zcw_cache",
 	    sizeof (zil_commit_waiter_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
 
-	zil_ksp = kstat_create("zfs", 0, "zil", "misc",
+	zil_sums_init(&zil_sums_global);
+	zil_kstats_global = kstat_create("zfs", 0, "zil", "misc",
 	    KSTAT_TYPE_NAMED, sizeof (zil_stats) / sizeof (kstat_named_t),
 	    KSTAT_FLAG_VIRTUAL);
 
-	if (zil_ksp != NULL) {
+	if (zil_kstats_global != NULL) {
-		zil_ksp->ks_data = &zil_stats;
+		zil_kstats_global->ks_data = &zil_stats;
-		kstat_install(zil_ksp);
+		zil_kstats_global->ks_update = zil_kstats_global_update;
+		zil_kstats_global->ks_private = NULL;
+		kstat_install(zil_kstats_global);
 	}
 }
 
@@ -3287,10 +3378,12 @@ zil_fini(void)
 	kmem_cache_destroy(zil_zcw_cache);
 	kmem_cache_destroy(zil_lwb_cache);
 
-	if (zil_ksp != NULL) {
+	if (zil_kstats_global != NULL) {
-		kstat_delete(zil_ksp);
+		kstat_delete(zil_kstats_global);
-		zil_ksp = NULL;
+		zil_kstats_global = NULL;
 	}
+
+	zil_sums_fini(&zil_sums_global);
 }
 
 void
@@ -3388,7 +3481,7 @@ zil_free(zilog_t *zilog)
  * Open an intent log.
  */
 zilog_t *
-zil_open(objset_t *os, zil_get_data_t *get_data)
+zil_open(objset_t *os, zil_get_data_t *get_data, zil_sums_t *zil_sums)
 {
 	zilog_t *zilog = dmu_objset_zil(os);
 
@@ -3397,6 +3490,7 @@ zil_open(objset_t *os, zil_get_data_t *get_data)
 	ASSERT(list_is_empty(&zilog->zl_lwb_list));
 
 	zilog->zl_get_data = get_data;
+	zilog->zl_sums = zil_sums;
 
 	return (zilog);
 }
@@ -3838,6 +3932,9 @@ EXPORT_SYMBOL(zil_lwb_add_block);
 EXPORT_SYMBOL(zil_bp_tree_add);
 EXPORT_SYMBOL(zil_set_sync);
 EXPORT_SYMBOL(zil_set_logbias);
+EXPORT_SYMBOL(zil_sums_init);
+EXPORT_SYMBOL(zil_sums_fini);
+EXPORT_SYMBOL(zil_kstat_values_update);
 
 ZFS_MODULE_PARAM(zfs, zfs_, commit_timeout_pct, INT, ZMOD_RW,
 	"ZIL block open timeout percentage");