OpenZFS 7004 - dmu_tx_hold_zap() does dnode_hold() 7x on same object

Using a benchmark which has 32 threads creating 2 million files in the
same directory, on a machine with 16 CPU cores, I observed poor
performance. I noticed that dmu_tx_hold_zap() was using about 30% of
all CPU, and doing dnode_hold() 7 times on the same object (the ZAP
object that is being held).

dmu_tx_hold_zap() keeps a hold on the dnode_t the entire time it is
running, in dmu_tx_hold_t:txh_dnode, so it would be nice to use the
dnode_t that we already have in hand, rather than repeatedly calling
dnode_hold(). To do this, we need to pass the dnode_t down through
all the intermediate calls that dmu_tx_hold_zap() makes, making these
routines take the dnode_t* rather than an objset_t* and a uint64_t
object number. In particular, the following routines will need to have
analogous *_by_dnode() variants created:

dmu_buf_hold_noread()
dmu_buf_hold()
zap_lookup()
zap_lookup_norm()
zap_count_write()
zap_lockdir()
zap_count_write()

This can improve performance on the benchmark described above by 100%,
from 30,000 file creations per second to 60,000. (This improvement is on
top of that provided by working around the object allocation issue. Peak
performance of ~90,000 creations per second was observed with 8 CPUs;
adding CPUs past that decreased performance due to lock contention.) The
CPU used by dmu_tx_hold_zap() was reduced by 88%, from 340 CPU-seconds
to 40 CPU-seconds.

Sponsored by: Intel Corp.

Signed-off-by: Matthew Ahrens <[email protected]>
Signed-off-by: Ned Bass <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/7004
OpenZFS-commit: https://github.com/openzfs/openzfs/pull/109
Closes #4641
Closes #4972

5 files changed, 126 insertions, 11 deletions

diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c
index d89d030d3..d297506e9 100644
--- a/module/zfs/dbuf.c
+++ b/module/zfs/dbuf.c
@@ -21,7 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
@@ -2843,6 +2843,21 @@ dmu_buf_get_objset(dmu_buf_t *db)
 	return (dbi->db_objset);
 }
 
+dnode_t *
+dmu_buf_dnode_enter(dmu_buf_t *db)
+{
+	dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
+	DB_DNODE_ENTER(dbi);
+	return (DB_DNODE(dbi));
+}
+
+void
+dmu_buf_dnode_exit(dmu_buf_t *db)
+{
+	dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
+	DB_DNODE_EXIT(dbi);
+}
+
 static void
 dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db)
 {
diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c
index e1dfb41ff..6ed4743ee 100644
--- a/module/zfs/dmu.c
+++ b/module/zfs/dmu.c
@@ -128,6 +128,26 @@ const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
 };
 
 int
+dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset,
+    void *tag, dmu_buf_t **dbp)
+{
+	uint64_t blkid;
+	dmu_buf_impl_t *db;
+
+	blkid = dbuf_whichblock(dn, 0, offset);
+	rw_enter(&dn->dn_struct_rwlock, RW_READER);
+	db = dbuf_hold(dn, blkid, tag);
+	rw_exit(&dn->dn_struct_rwlock);
+
+	if (db == NULL) {
+		*dbp = NULL;
+		return (SET_ERROR(EIO));
+	}
+
+	*dbp = &db->db;
+	return (0);
+}
+int
 dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
     void *tag, dmu_buf_t **dbp)
 {
@@ -155,6 +175,29 @@ dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
 }
 
 int
+dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
+    void *tag, dmu_buf_t **dbp, int flags)
+{
+	int err;
+	int db_flags = DB_RF_CANFAIL;
+
+	if (flags & DMU_READ_NO_PREFETCH)
+		db_flags |= DB_RF_NOPREFETCH;
+
+	err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp);
+	if (err == 0) {
+		dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
+		err = dbuf_read(db, NULL, db_flags);
+		if (err != 0) {
+			dbuf_rele(db, tag);
+			*dbp = NULL;
+		}
+	}
+
+	return (err);
+}
+
+int
 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
     void *tag, dmu_buf_t **dbp, int flags)
 {
diff --git a/module/zfs/dmu_tx.c b/module/zfs/dmu_tx.c
index ed29bfbc6..c50f732a7 100644
--- a/module/zfs/dmu_tx.c
+++ b/module/zfs/dmu_tx.c
@@ -21,7 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
  */
 
 #include <sys/dmu.h>
@@ -794,15 +794,14 @@ dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name)
 		 * access the name in this fat-zap so that we'll check
 		 * for i/o errors to the leaf blocks, etc.
 		 */
-		err = zap_lookup(dn->dn_objset, dn->dn_object, name,
-		    8, 0, NULL);
+		err = zap_lookup_by_dnode(dn, name, 8, 0, NULL);
 		if (err == EIO) {
 			tx->tx_err = err;
 			return;
 		}
 	}
 
-	err = zap_count_write(dn->dn_objset, dn->dn_object, name, add,
+	err = zap_count_write_by_dnode(dn, name, add,
 	    &txh->txh_space_towrite, &txh->txh_space_tooverwrite);
 
 	/*
diff --git a/module/zfs/zap.c b/module/zfs/zap.c
index d0d438fe3..0ed6d7f49 100644
--- a/module/zfs/zap.c
+++ b/module/zfs/zap.c
@@ -270,6 +270,7 @@ zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
 	uint64_t blk, off;
 	int err;
 	dmu_buf_t *db;
+	dnode_t *dn;
 	int bs = FZAP_BLOCK_SHIFT(zap);
 
 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
@@ -277,8 +278,15 @@ zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
 	blk = idx >> (bs-3);
 	off = idx & ((1<<(bs-3))-1);
 
-	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
+	/*
+	 * Note: this is equivalent to dmu_buf_hold(), but we use
+	 * _dnode_enter / _by_dnode because it's faster because we don't
+	 * have to hold the dnode.
+	 */
+	dn = dmu_buf_dnode_enter(zap->zap_dbuf);
+	err = dmu_buf_hold_by_dnode(dn,
 	    (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
+	dmu_buf_dnode_exit(zap->zap_dbuf);
 	if (err)
 		return (err);
 	*valp = ((uint64_t *)db->db_data)[off];
@@ -292,9 +300,11 @@ zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
 		 */
 		blk = (idx*2) >> (bs-3);
 
-		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
+		dn = dmu_buf_dnode_enter(zap->zap_dbuf);
+		err = dmu_buf_hold_by_dnode(dn,
 		    (tbl->zt_nextblk + blk) << bs, FTAG, &db,
 		    DMU_READ_NO_PREFETCH);
+		dmu_buf_dnode_exit(zap->zap_dbuf);
 		if (err == 0)
 			dmu_buf_rele(db, FTAG);
 	}
@@ -502,6 +512,7 @@ zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
 	zap_leaf_t *l;
 	int bs = FZAP_BLOCK_SHIFT(zap);
 	int err;
+	dnode_t *dn;
 
 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
 
@@ -515,8 +526,10 @@ zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
 	if (blkid == 0)
 		return (ENOENT);
 
-	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
+	dn = dmu_buf_dnode_enter(zap->zap_dbuf);
+	err = dmu_buf_hold_by_dnode(dn,
 	    blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
+	dmu_buf_dnode_exit(zap->zap_dbuf);
 	if (err)
 		return (err);
 
diff --git a/module/zfs/zap_micro.c b/module/zfs/zap_micro.c
index de646287e..bc0a5e71d 100644
--- a/module/zfs/zap_micro.c
+++ b/module/zfs/zap_micro.c
@@ -536,6 +536,24 @@ zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
 	return (0);
 }
 
+static int
+zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
+    krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
+{
+	dmu_buf_t *db;
+	int err;
+
+	err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
+	if (err != 0) {
+		return (err);
+	}
+	err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
+	if (err != 0) {
+		dmu_buf_rele(db, tag);
+	}
+	return (err);
+}
+
 int
 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
     krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
@@ -928,6 +946,33 @@ zap_prefetch(objset_t *os, uint64_t zapobj, const char *name)
 }
 
 int
+zap_lookup_by_dnode(dnode_t *dn, const char *name,
+    uint64_t integer_size, uint64_t num_integers, void *buf)
+{
+	return (zap_lookup_norm_by_dnode(dn, name, integer_size,
+	    num_integers, buf, MT_EXACT, NULL, 0, NULL));
+}
+
+int
+zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
+    uint64_t integer_size, uint64_t num_integers, void *buf,
+    matchtype_t mt, char *realname, int rn_len,
+    boolean_t *ncp)
+{
+	zap_t *zap;
+	int err;
+
+	err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
+	    FTAG, &zap);
+	if (err != 0)
+		return (err);
+	err = zap_lookup_impl(zap, name, integer_size,
+	    num_integers, buf, mt, realname, rn_len, ncp);
+	zap_unlockdir(zap, FTAG);
+	return (err);
+}
+
+int
 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
     int key_numints)
 {
@@ -1460,7 +1505,7 @@ zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
 }
 
 int
-zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
+zap_count_write_by_dnode(dnode_t *dn, const char *name, int add,
     uint64_t *towrite, uint64_t *tooverwrite)
 {
 	zap_t *zap;
@@ -1488,7 +1533,7 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
 	 * At present we are just evaluating the possibility of this operation
 	 * and hence we do not want to trigger an upgrade.
 	 */
-	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE,
+	err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
 	    FTAG, &zap);
 	if (err != 0)
 		return (err);
@@ -1552,7 +1597,7 @@ EXPORT_SYMBOL(zap_lookup_uint64);
 EXPORT_SYMBOL(zap_contains);
 EXPORT_SYMBOL(zap_prefetch);
 EXPORT_SYMBOL(zap_prefetch_uint64);
-EXPORT_SYMBOL(zap_count_write);
+EXPORT_SYMBOL(zap_count_write_by_dnode);
 EXPORT_SYMBOL(zap_add);
 EXPORT_SYMBOL(zap_add_uint64);
 EXPORT_SYMBOL(zap_update);