5175 implement dmu_read_uio_dbuf() to improve cached read performance

Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Alex Reece <alex.reece@delphix.com>
Reviewed by: George Wilson <george@delphix.com>
Reviewed by: Richard Elling <richard.elling@gmail.com>
Approved by: Robert Mustacchi <rm@joyent.com>

4 files changed, 81 insertions, 17 deletions

diff --git a/usr/src/uts/common/fs/zfs/dmu.c b/usr/src/uts/common/fs/zfs/dmu.c
index 4fdcb14750..13c189831f 100644
--- a/usr/src/uts/common/fs/zfs/dmu.c
+++ b/usr/src/uts/common/fs/zfs/dmu.c
@@ -1015,8 +1015,8 @@ xuio_stat_wbuf_nocopy()
 }
 
 #ifdef _KERNEL
-int
-dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
+static int
+dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
 {
 	dmu_buf_t **dbp;
 	int numbufs, i, err;
@@ -1026,8 +1026,8 @@ dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
 	 * NB: we could do this block-at-a-time, but it's nice
 	 * to be reading in parallel.
 	 */
-	err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
-	    &numbufs, &dbp);
+	err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
+	    TRUE, FTAG, &numbufs, &dbp, 0);
 	if (err)
 		return (err);
 
@@ -1072,6 +1072,58 @@ dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
 	return (err);
 }
 
+/*
+ * Read 'size' bytes into the uio buffer.
+ * From object zdb->db_object.
+ * Starting at offset uio->uio_loffset.
+ *
+ * If the caller already has a dbuf in the target object
+ * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
+ * because we don't have to find the dnode_t for the object.
+ */
+int
+dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size)
+{
+	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
+	dnode_t *dn;
+	int err;
+
+	if (size == 0)
+		return (0);
+
+	DB_DNODE_ENTER(db);
+	dn = DB_DNODE(db);
+	err = dmu_read_uio_dnode(dn, uio, size);
+	DB_DNODE_EXIT(db);
+
+	return (err);
+}
+
+/*
+ * Read 'size' bytes into the uio buffer.
+ * From the specified object
+ * Starting at offset uio->uio_loffset.
+ */
+int
+dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
+{
+	dnode_t *dn;
+	int err;
+
+	if (size == 0)
+		return (0);
+
+	err = dnode_hold(os, object, FTAG, &dn);
+	if (err)
+		return (err);
+
+	err = dmu_read_uio_dnode(dn, uio, size);
+
+	dnode_rele(dn, FTAG);
+
+	return (err);
+}
+
 static int
 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
 {
@@ -1124,6 +1176,15 @@ dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
 	return (err);
 }
 
+/*
+ * Write 'size' bytes from the uio buffer.
+ * To object zdb->db_object.
+ * Starting at offset uio->uio_loffset.
+ *
+ * If the caller already has a dbuf in the target object
+ * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
+ * because we don't have to find the dnode_t for the object.
+ */
 int
 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
     dmu_tx_t *tx)
@@ -1143,6 +1204,11 @@ dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
 	return (err);
 }
 
+/*
+ * Write 'size' bytes from the uio buffer.
+ * To the specified object.
+ * Starting at offset uio->uio_loffset.
+ */
 int
 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
     dmu_tx_t *tx)
diff --git a/usr/src/uts/common/fs/zfs/sys/dmu.h b/usr/src/uts/common/fs/zfs/sys/dmu.h
index b73de06493..ed374d5f80 100644
--- a/usr/src/uts/common/fs/zfs/sys/dmu.h
+++ b/usr/src/uts/common/fs/zfs/sys/dmu.h
@@ -614,6 +614,7 @@ void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 	dmu_tx_t *tx);
 int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
+int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
 int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
     dmu_tx_t *tx);
 int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
diff --git a/usr/src/uts/common/fs/zfs/zfs_vnops.c b/usr/src/uts/common/fs/zfs/zfs_vnops.c
index e87c002737..c608a79fd2 100644
--- a/usr/src/uts/common/fs/zfs/zfs_vnops.c
+++ b/usr/src/uts/common/fs/zfs/zfs_vnops.c
@@ -406,7 +406,6 @@ static int
 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
 {
 	znode_t *zp = VTOZ(vp);
-	objset_t *os = zp->z_zfsvfs->z_os;
 	int64_t	start, off;
 	int len = nbytes;
 	int error = 0;
@@ -425,7 +424,8 @@ mappedread(vnode_t *vp, int nbytes, uio_t *uio)
 			zfs_unmap_page(pp, va);
 			page_unlock(pp);
 		} else {
-			error = dmu_read_uio(os, zp->z_id, uio, bytes);
+			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
+			    uio, bytes);
 		}
 		len -= bytes;
 		off = 0;
@@ -460,7 +460,6 @@ zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 {
 	znode_t		*zp = VTOZ(vp);
 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
-	objset_t	*os;
 	ssize_t		n, nbytes;
 	int		error = 0;
 	rl_t		*rl;
@@ -468,7 +467,6 @@ zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 
 	ZFS_ENTER(zfsvfs);
 	ZFS_VERIFY_ZP(zp);
-	os = zfsvfs->z_os;
 
 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
 		ZFS_EXIT(zfsvfs);
@@ -558,10 +556,12 @@ zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 		nbytes = MIN(n, zfs_read_chunk_size -
 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
 
-		if (vn_has_cached_data(vp))
+		if (vn_has_cached_data(vp)) {
 			error = mappedread(vp, nbytes, uio);
-		else
-			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
+		} else {
+			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
+			    uio, nbytes);
+		}
 		if (error) {
 			/* convert checksum errors into IO errors */
 			if (error == ECKSUM)
diff --git a/usr/src/uts/common/io/comstar/lu/stmf_sbd/sbd_zvol.c b/usr/src/uts/common/io/comstar/lu/stmf_sbd/sbd_zvol.c
index 40b1aeeac3..0e96e2ec96 100644
--- a/usr/src/uts/common/io/comstar/lu/stmf_sbd/sbd_zvol.c
+++ b/usr/src/uts/common/io/comstar/lu/stmf_sbd/sbd_zvol.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014 by Delphix. All rights reserved.
  */
 
 #include <sys/conf.h>
@@ -402,7 +403,7 @@ sbd_zvol_copy_read(sbd_lu_t *sl, uio_t *uio)
 
 	rl = zfs_range_lock(sl->sl_zvol_rl_hdl, offset, len, RL_READER);
 
-	error =  dmu_read_uio(sl->sl_zvol_objset_hdl, ZVOL_OBJ, uio, len);
+	error = dmu_read_uio_dbuf(sl->sl_zvol_bonus_hdl, uio, len);
 
 	zfs_range_unlock(rl);
 	if (error == ECKSUM)
@@ -441,11 +442,7 @@ sbd_zvol_copy_write(sbd_lu_t *sl, uio_t *uio, int flags)
 	if (error) {
 		dmu_tx_abort(tx);
 	} else {
-		/*
-		 * XXX use the new bonus handle entry.
-		 */
-		error = dmu_write_uio(sl->sl_zvol_objset_hdl, ZVOL_OBJ,
-		    uio, len, tx);
+		error = dmu_write_uio_dbuf(sl->sl_zvol_bonus_hdl, uio, len, tx);
 		if (error == 0) {
 			zvol_log_write_minor(sl->sl_zvol_minor_hdl, tx, offset,
 			    (ssize_t)len, sync);