Switch synchronous IO on zvols to use _by_dbuf routines

This saves a small amount of CPU time and slightly improves latencies on synchronous operations on zvols. Signed-off-by: Richard Yao <[email protected]>
openzfs · Apr 26, 2017 · 7687327 · 7687327
1 parent 0091d66
commit 7687327
Show file tree

Hide file tree

Showing 3 changed files with 52 additions and 4 deletions.
diff --git a/include/sys/dmu.h b/include/sys/dmu.h
@@ -476,6 +476,8 @@ int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp);
  */
 int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
     void *tag, dmu_buf_t **, int flags);
+int dmu_buf_hold_by_dbuf(dmu_buf_t *zdb, uint64_t offset,
+    void *tag, dmu_buf_t **, int flags);
 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
     void *tag, dmu_buf_t **dbp, int flags);
 
@@ -741,6 +743,8 @@ int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 	void *buf, uint32_t flags);
 int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
     uint32_t flags);
+int dmu_read_by_dbuf(dmu_buf_t *zdb, uint64_t offset, uint64_t size, void *buf,
+    uint32_t flags);
 void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 	const void *buf, dmu_tx_t *tx);
 void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,

diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c
@@ -212,6 +212,23 @@ dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
 	return (err);
 }
 
+int
+dmu_buf_hold_by_dbuf(dmu_buf_t *zdb, uint64_t offset,
+    void *tag, dmu_buf_t **dbp, int flags)
+{
+	dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
+	dnode_t *dn;
+	int err;
+
+
+	DB_DNODE_ENTER(db);
+	dn = DB_DNODE(db);
+	err = dmu_buf_hold_by_dnode(dn, offset, tag, dbp, flags);
+	DB_DNODE_EXIT(db);
+
+	return (err);
+}
+
 int
 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
     void *tag, dmu_buf_t **dbp, int flags)
@@ -968,6 +985,33 @@ dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
 	return (dmu_read_impl(dn, offset, size, buf, flags));
 }
 
+/*
+ * Read 'size' bytes into the buffer.
+ * From object zdb->db_object.
+ *
+ * If the caller already has a dbuf in the target object
+ * (e.g. its bonus buffer), this routine is faster than dmu_read(),
+ * because we don't have to find the dnode_t for the object.
+ */
+int
+dmu_read_by_dbuf(dmu_buf_t *zdb, uint64_t offset, uint64_t size, void *buf,
+    uint32_t flags)
+{
+	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_impl(dn, offset, size, buf, flags);
+	DB_DNODE_EXIT(db);
+
+	return (err);
+}
+
 static void
 dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size,
     const void *buf, dmu_tx_t *tx)

diff --git a/module/zfs/zvol.c b/module/zfs/zvol.c
@@ -609,8 +609,8 @@ zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
 		    (write_state == WR_COPIED ? len : 0));
 		lr = (lr_write_t *)&itx->itx_lr;
-		if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
-		    ZVOL_OBJ, offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
+		if (write_state == WR_COPIED && dmu_read_by_dbuf(zv->zv_dbuf,
+		    offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
 			zil_itx_destroy(itx);
 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
 			lr = (lr_write_t *)&itx->itx_lr;
@@ -901,12 +901,12 @@ zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
 	 * we don't have to write the data twice.
 	 */
 	if (buf != NULL) { /* immediate write */
-		error = dmu_read(os, object, offset, size, buf,
+		error = dmu_read_by_dbuf(zv->zv_dbuf, offset, size, buf,
 		    DMU_READ_NO_PREFETCH);
 	} else {
 		size = zv->zv_volblocksize;
 		offset = P2ALIGN_TYPED(offset, size, uint64_t);
-		error = dmu_buf_hold(os, object, offset, zgd, &db,
+		error = dmu_buf_hold_by_dbuf(zv->zv_dbuf, offset, zgd, &db,
 		    DMU_READ_NO_PREFETCH);
 		if (error == 0) {
 			blkptr_t *obp = dmu_buf_get_blkptr(db);