1 files changed, 104 insertions, 55 deletions

diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c
index aaeec418d..04b02c7d2 100644
--- a/module/zfs/dmu.c
+++ b/module/zfs/dmu.c
@@ -381,7 +381,7 @@ dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
 		}
 		nblks = 1;
 	}
-	dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
+	dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP | KM_NODEBUG);
 
 	if (dn->dn_objset->os_dsl_dataset)
 		dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
@@ -1122,75 +1122,82 @@ dmu_write_req(objset_t *os, uint64_t object, struct request *req, dmu_tx_t *tx)
 	dmu_buf_rele_array(dbp, numbufs, FTAG);
 	return (err);
 }
-#endif
 
-#ifdef HAVE_ZPL
 int
-dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
-    dmu_tx_t *tx)
+dmu_read_uio(objset_t *os, uint64_t object, 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);
+	dmu_buf_t **dbp;
+	int numbufs, i, err;
+	xuio_t *xuio = NULL;
 
-	DB_DNODE_ENTER(db);
-	dn = DB_DNODE(db);
-	err = dmu_write_uio_dnode(dn, uio, size, tx);
-	DB_DNODE_EXIT(db);
+	/*
+	 * 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);
+	if (err)
+		return (err);
 
-	return (err);
-}
+	for (i = 0; i < numbufs; i++) {
+		int tocpy;
+		int bufoff;
+		dmu_buf_t *db = dbp[i];
 
-int
-dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
-    dmu_tx_t *tx)
-{
-	dnode_t *dn;
-	int err;
+		ASSERT(size > 0);
 
-	if (size == 0)
-		return (0);
+		bufoff = uio->uio_loffset - db->db_offset;
+		tocpy = (int)MIN(db->db_size - bufoff, size);
 
-	err = dnode_hold(os, object, FTAG, &dn);
-	if (err)
-		return (err);
+		if (xuio) {
+			dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
+			arc_buf_t *dbuf_abuf = dbi->db_buf;
+			arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
+			err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
+			if (!err) {
+				uio->uio_resid -= tocpy;
+				uio->uio_loffset += tocpy;
+			}
 
-	err = dmu_write_uio_dnode(dn, uio, size, tx);
+			if (abuf == dbuf_abuf)
+				XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
+			else
+				XUIOSTAT_BUMP(xuiostat_rbuf_copied);
+		} else {
+			err = uiomove((char *)db->db_data + bufoff, tocpy,
+			    UIO_READ, uio);
+		}
+		if (err)
+			break;
 
-	dnode_rele(dn, FTAG);
+		size -= tocpy;
+	}
+	dmu_buf_rele_array(dbp, numbufs, FTAG);
 
 	return (err);
 }
 
-int
-dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
-    page_t *pp, dmu_tx_t *tx)
+static int
+dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
 {
 	dmu_buf_t **dbp;
-	int numbufs, i;
-	int err;
-
-	if (size == 0)
-		return (0);
+	int numbufs;
+	int err = 0;
+	int i;
 
-	err = dmu_buf_hold_array(os, object, offset, size,
-	    FALSE, FTAG, &numbufs, &dbp);
+	err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
+	    FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
 	if (err)
 		return (err);
 
 	for (i = 0; i < numbufs; i++) {
-		int tocpy, copied, thiscpy;
+		int tocpy;
 		int bufoff;
 		dmu_buf_t *db = dbp[i];
-		caddr_t va;
 
 		ASSERT(size > 0);
-		ASSERT3U(db->db_size, >=, PAGESIZE);
 
-		bufoff = offset - db->db_offset;
+		bufoff = uio->uio_loffset - db->db_offset;
 		tocpy = (int)MIN(db->db_size - bufoff, size);
 
 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
@@ -1200,26 +1207,68 @@ dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 		else
 			dmu_buf_will_dirty(db, tx);
 
-		for (copied = 0; copied < tocpy; copied += PAGESIZE) {
-			ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
-			thiscpy = MIN(PAGESIZE, tocpy - copied);
-			va = zfs_map_page(pp, S_READ);
-			bcopy(va, (char *)db->db_data + bufoff, thiscpy);
-			zfs_unmap_page(pp, va);
-			pp = pp->p_next;
-			bufoff += PAGESIZE;
-		}
+		/*
+		 * XXX uiomove could block forever (eg.nfs-backed
+		 * pages).  There needs to be a uiolockdown() function
+		 * to lock the pages in memory, so that uiomove won't
+		 * block.
+		 */
+		err = uiomove((char *)db->db_data + bufoff, tocpy,
+		    UIO_WRITE, uio);
 
 		if (tocpy == db->db_size)
 			dmu_buf_fill_done(db, tx);
 
-		offset += tocpy;
+		if (err)
+			break;
+
 		size -= tocpy;
 	}
+
 	dmu_buf_rele_array(dbp, numbufs, FTAG);
 	return (err);
 }
-#endif
+
+int
+dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
+    dmu_tx_t *tx)
+{
+	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_write_uio_dnode(dn, uio, size, tx);
+	DB_DNODE_EXIT(db);
+
+	return (err);
+}
+
+int
+dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
+    dmu_tx_t *tx)
+{
+	dnode_t *dn;
+	int err;
+
+	if (size == 0)
+		return (0);
+
+	err = dnode_hold(os, object, FTAG, &dn);
+	if (err)
+		return (err);
+
+	err = dmu_write_uio_dnode(dn, uio, size, tx);
+
+	dnode_rele(dn, FTAG);
+
+	return (err);
+}
+#endif /* _KERNEL */
 
 /*
  * Allocate a loaned anonymous arc buffer.