Resolve code review.

Updated man/man4/zfs.4

Added dmu_tx_wrlog_over_max to dmu_tx kstat

Added ASSERT0 for dp_wrlog_total and dp_wrlog_pertxg[]
on dsl_pool_close()

Fixed some grammar in comments.

Signed-off-by: jxdking <lostking2008@hotmail.com>

include/sys/dmu_tx.h

@@ -124,6 +124,7 @@ typedef struct dmu_tx_stats {
 	kstat_named_t dmu_tx_dirty_throttle;
 	kstat_named_t dmu_tx_dirty_delay;
 	kstat_named_t dmu_tx_dirty_over_max;
+	kstat_named_t dmu_tx_wrlog_over_max;
 	kstat_named_t dmu_tx_dirty_frees_delay;
 	kstat_named_t dmu_tx_quota;
 } dmu_tx_stats_t;

man/man4/zfs.4

@@ -1066,6 +1066,15 @@ Start syncing out a transaction group if there's at least this much dirty data
 This should be less than
 .Sy zfs_vdev_async_write_active_min_dirty_percent .
 .
+.It Sy zfs_wrlog_data_max Ns = Pq int
+The upper limit of write-transaction zil log data size in bytes. Once it is reached, write operation is blocked, until log data is cleared out after transaction group sync. Because of some overhead, it should be set at least 2 times the size of
+.Sy zfs_dirty_data_max
+.No to prevent harming normal write throughput.
+It also should be smaller than the size of the slog device if slog is present.
+.Pp
+Defaults to
+.Sy zfs_dirty_data_max*2
+.
 .It Sy zfs_fallocate_reserve_percent Ns = Ns Sy 110 Ns % Pq uint
 Since ZFS is a copy-on-write filesystem with snapshots, blocks cannot be
 preallocated for a file in order to guarantee that later writes will not

module/zfs/arc.c

@@ -7975,7 +7975,7 @@ arc_init(void)
 		/*
 		 * dp_wrlog_total is reduced for each txg at the end of
 		 * spa_sync(). However, dp_dirty_total is reduced every time
-		 * a block being written out. Thus under normal operation,
+		 * a block is written out. Thus under normal operation,
 		 * dp_wrlog_total could grow 2 times as big as
 		 * zfs_dirty_data_max.
 		 */

module/zfs/dmu_tx.c

@@ -53,6 +53,7 @@ dmu_tx_stats_t dmu_tx_stats = {
 	{ "dmu_tx_dirty_throttle",	KSTAT_DATA_UINT64 },
 	{ "dmu_tx_dirty_delay",		KSTAT_DATA_UINT64 },
 	{ "dmu_tx_dirty_over_max",	KSTAT_DATA_UINT64 },
+	{ "dmu_tx_wrlog_over_max",	KSTAT_DATA_UINT64 },
 	{ "dmu_tx_dirty_frees_delay",	KSTAT_DATA_UINT64 },
 	{ "dmu_tx_quota",		KSTAT_DATA_UINT64 },
 };
@@ -886,6 +887,7 @@ dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how)
 
 	if (!tx->tx_dirty_delayed &&
 	    dsl_pool_wrlog_over_max(tx->tx_pool)) {
+		DMU_TX_STAT_BUMP(dmu_tx_wrlog_over_max);
 		return (SET_ERROR(ERESTART));
 	}
 

module/zfs/dsl_pool.c

@@ -430,8 +430,10 @@ dsl_pool_close(dsl_pool_t *dp)
 	mutex_destroy(&dp->dp_lock);
 	cv_destroy(&dp->dp_spaceavail_cv);
 
+	ASSERT0(aggsum_value(&dp->dp_wrlog_total));
 	aggsum_fini(&dp->dp_wrlog_total);
 	for (int i = 0; i < TXG_SIZE; i++) {
+		ASSERT0(aggsum_value(&dp->dp_wrlog_pertxg[i]));
 		aggsum_fini(&dp->dp_wrlog_pertxg[i]);
 	}