MFV r254749:

Don't hold dd_lock for long by breaking it when not doing dsl_dir
accounting.  It is not necessary to hold the lock while manipulating
the parent's accounting, because there is no interface for userland
to see a consistent picture of both parent and child at the same
time anyway.

Illumos ZFS issues:
  4046 dsl_dataset_t ds_dir->dd_lock is highly contended

2 files changed, 14 insertions, 38 deletions

diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c
index ff6e682..1b69ed52 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c
@@ -101,9 +101,8 @@ dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 		    used, compressed, uncompressed);
 		return;
 	}
-	dmu_buf_will_dirty(ds->ds_dbuf, tx);
 
-	mutex_enter(&ds->ds_dir->dd_lock);
+	dmu_buf_will_dirty(ds->ds_dbuf, tx);
 	mutex_enter(&ds->ds_lock);
 	delta = parent_delta(ds, used);
 	ds->ds_phys->ds_referenced_bytes += used;
@@ -115,7 +114,6 @@ dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 	    compressed, uncompressed, tx);
 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
-	mutex_exit(&ds->ds_dir->dd_lock);
 }
 
 int
@@ -150,7 +148,6 @@ dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
 
-		mutex_enter(&ds->ds_dir->dd_lock);
 		mutex_enter(&ds->ds_lock);
 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
 		    !DS_UNIQUE_IS_ACCURATE(ds));
@@ -161,7 +158,6 @@ dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 		    delta, -compressed, -uncompressed, tx);
 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
-		mutex_exit(&ds->ds_dir->dd_lock);
 	} else {
 		dprintf_bp(bp, "putting on dead list: %s", "");
 		if (async) {
@@ -596,31 +592,6 @@ dsl_dataset_name(dsl_dataset_t *ds, char *name)
 	}
 }
 
-static int
-dsl_dataset_namelen(dsl_dataset_t *ds)
-{
-	int result;
-
-	if (ds == NULL) {
-		result = 3;	/* "mos" */
-	} else {
-		result = dsl_dir_namelen(ds->ds_dir);
-		VERIFY0(dsl_dataset_get_snapname(ds));
-		if (ds->ds_snapname[0]) {
-			++result;	/* adding one for the @-sign */
-			if (!MUTEX_HELD(&ds->ds_lock)) {
-				mutex_enter(&ds->ds_lock);
-				result += strlen(ds->ds_snapname);
-				mutex_exit(&ds->ds_lock);
-			} else {
-				result += strlen(ds->ds_snapname);
-			}
-		}
-	}
-
-	return (result);
-}
-
 void
 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 {
diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dir.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dir.c
index 7d0b104..4c86f9c 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dir.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dir.c
@@ -845,11 +845,21 @@ dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
     int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx)
 {
 	int64_t accounted_delta;
+
+	/*
+	 * dsl_dataset_set_refreservation_sync_impl() calls this with
+	 * dd_lock held, so that it can atomically update
+	 * ds->ds_reserved and the dsl_dir accounting, so that
+	 * dsl_dataset_check_quota() can see dataset and dir accounting
+	 * consistently.
+	 */
 	boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
 
 	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT(type < DD_USED_NUM);
 
+	dmu_buf_will_dirty(dd->dd_dbuf, tx);
+
 	if (needlock)
 		mutex_enter(&dd->dd_lock);
 	accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
@@ -858,7 +868,6 @@ dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 	    dd->dd_phys->dd_compressed_bytes >= -compressed);
 	ASSERT(uncompressed >= 0 ||
 	    dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
-	dmu_buf_will_dirty(dd->dd_dbuf, tx);
 	dd->dd_phys->dd_used_bytes += used;
 	dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 	dd->dd_phys->dd_compressed_bytes += compressed;
@@ -891,8 +900,6 @@ void
 dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
     dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx)
 {
-	boolean_t needlock = !MUTEX_HELD(&dd->dd_lock);
-
 	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT(oldtype < DD_USED_NUM);
 	ASSERT(newtype < DD_USED_NUM);
@@ -900,17 +907,15 @@ dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 	if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 		return;
 
-	if (needlock)
-		mutex_enter(&dd->dd_lock);
+	dmu_buf_will_dirty(dd->dd_dbuf, tx);
+	mutex_enter(&dd->dd_lock);
 	ASSERT(delta > 0 ?
 	    dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 	    dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 	ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
-	dmu_buf_will_dirty(dd->dd_dbuf, tx);
 	dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 	dd->dd_phys->dd_used_breakdown[newtype] += delta;
-	if (needlock)
-		mutex_exit(&dd->dd_lock);
+	mutex_exit(&dd->dd_lock);
 }
 
 typedef struct dsl_dir_set_qr_arg {