X-Git-Url: https://git.camperquake.de/gitweb.cgi?a=blobdiff_plain;f=module%2Fzfs%2Fzio.c;h=66f228bc7807fbab4fae6b1dc9f0306e9df400f7;hb=9ae529ec5dbdc828ff8326beae58062971d74b2e;hp=6b03be6f3bbf62e8d7c442e145e8bd5fd8ca90cb;hpb=ae6ba3dbe618bb7dbc46f2a3fb54c58243835d6b;p=zfs.git

diff --git a/module/zfs/zio.c b/module/zfs/zio.c
index 6b03be6..66f228b 100644
--- a/module/zfs/zio.c
+++ b/module/zfs/zio.c
@@ -20,7 +20,8 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011 by Delphix. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -72,6 +73,7 @@ char *zio_type_name[ZIO_TYPES] = {
  */
 kmem_cache_t *zio_cache;
 kmem_cache_t *zio_link_cache;
+kmem_cache_t *zio_vdev_cache;
 kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
 kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
 int zio_bulk_flags = 0;
@@ -98,6 +100,35 @@ int zio_buf_debug_limit = 0;
 
 static inline void __zio_execute(zio_t *zio);
 
+static int
+zio_cons(void *arg, void *unused, int kmflag)
+{
+	zio_t *zio = arg;
+
+	bzero(zio, sizeof (zio_t));
+
+	mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL);
+	cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL);
+
+	list_create(&zio->io_parent_list, sizeof (zio_link_t),
+	    offsetof(zio_link_t, zl_parent_node));
+	list_create(&zio->io_child_list, sizeof (zio_link_t),
+	    offsetof(zio_link_t, zl_child_node));
+
+	return (0);
+}
+
+static void
+zio_dest(void *arg, void *unused)
+{
+	zio_t *zio = arg;
+
+	mutex_destroy(&zio->io_lock);
+	cv_destroy(&zio->io_cv);
+	list_destroy(&zio->io_parent_list);
+	list_destroy(&zio->io_child_list);
+}
+
 void
 zio_init(void)
 {
@@ -107,10 +138,12 @@ zio_init(void)
 #ifdef _KERNEL
 	data_alloc_arena = zio_alloc_arena;
 #endif
-	zio_cache = kmem_cache_create("zio_cache",
-	    sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM);
+	zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0,
+	    zio_cons, zio_dest, NULL, NULL, NULL, KMC_KMEM);
 	zio_link_cache = kmem_cache_create("zio_link_cache",
 	    sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM);
+	zio_vdev_cache = kmem_cache_create("zio_vdev_cache", sizeof(vdev_io_t),
+	    PAGESIZE, NULL, NULL, NULL, NULL, NULL, KMC_VMEM);
 
 	/*
 	 * For small buffers, we want a cache for each multiple of
@@ -200,6 +233,7 @@ zio_fini(void)
 		zio_data_buf_cache[c] = NULL;
 	}
 
+	kmem_cache_destroy(zio_vdev_cache);
 	kmem_cache_destroy(zio_link_cache);
 	kmem_cache_destroy(zio_cache);
 
@@ -225,7 +259,7 @@ zio_buf_alloc(size_t size)
 
 	ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
 
-	return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE));
+	return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE | KM_NODEBUG));
 }
 
 /*
@@ -241,7 +275,8 @@ zio_data_buf_alloc(size_t size)
 
 	ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
 
-	return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE));
+	return (kmem_cache_alloc(zio_data_buf_cache[c],
+	    KM_PUSHPAGE | KM_NODEBUG));
 }
 
 void
@@ -265,6 +300,24 @@ zio_data_buf_free(void *buf, size_t size)
 }
 
 /*
+ * Dedicated I/O buffers to ensure that memory fragmentation never prevents
+ * or significantly delays the issuing of a zio.   These buffers are used
+ * to aggregate I/O and could be used for raidz stripes.
+ */
+void *
+zio_vdev_alloc(void)
+{
+	return (kmem_cache_alloc(zio_vdev_cache, KM_PUSHPAGE));
+}
+
+void
+zio_vdev_free(void *buf)
+{
+	kmem_cache_free(zio_vdev_cache, buf);
+
+}
+
+/*
  * ==========================================================================
  * Push and pop I/O transform buffers
  * ==========================================================================
@@ -510,15 +563,6 @@ zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 	ASSERT(vd || stage == ZIO_STAGE_OPEN);
 
 	zio = kmem_cache_alloc(zio_cache, KM_PUSHPAGE);
-	bzero(zio, sizeof (zio_t));
-
-	mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL);
-	cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL);
-
-	list_create(&zio->io_parent_list, sizeof (zio_link_t),
-	    offsetof(zio_link_t, zl_parent_node));
-	list_create(&zio->io_child_list, sizeof (zio_link_t),
-	    offsetof(zio_link_t, zl_child_node));
 
 	if (vd != NULL)
 		zio->io_child_type = ZIO_CHILD_VDEV;
@@ -530,6 +574,7 @@ zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 		zio->io_child_type = ZIO_CHILD_LOGICAL;
 
 	if (bp != NULL) {
+		zio->io_logical = NULL;
 		zio->io_bp = (blkptr_t *)bp;
 		zio->io_bp_copy = *bp;
 		zio->io_bp_orig = *bp;
@@ -540,21 +585,52 @@ zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 			zio->io_logical = zio;
 		if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp))
 			pipeline |= ZIO_GANG_STAGES;
+	} else {
+		zio->io_logical = NULL;
+		zio->io_bp = NULL;
+		bzero(&zio->io_bp_copy, sizeof (blkptr_t));
+		bzero(&zio->io_bp_orig, sizeof (blkptr_t));
 	}
 
 	zio->io_spa = spa;
 	zio->io_txg = txg;
+	zio->io_ready = NULL;
 	zio->io_done = done;
 	zio->io_private = private;
+	zio->io_prev_space_delta = 0;
 	zio->io_type = type;
 	zio->io_priority = priority;
 	zio->io_vd = vd;
+	zio->io_vsd = NULL;
+	zio->io_vsd_ops = NULL;
 	zio->io_offset = offset;
+	zio->io_deadline = 0;
 	zio->io_orig_data = zio->io_data = data;
 	zio->io_orig_size = zio->io_size = size;
 	zio->io_orig_flags = zio->io_flags = flags;
 	zio->io_orig_stage = zio->io_stage = stage;
 	zio->io_orig_pipeline = zio->io_pipeline = pipeline;
+	bzero(&zio->io_prop, sizeof (zio_prop_t));
+	zio->io_cmd = 0;
+	zio->io_reexecute = 0;
+	zio->io_bp_override = NULL;
+	zio->io_walk_link = NULL;
+	zio->io_transform_stack = NULL;
+	zio->io_delay = 0;
+	zio->io_error = 0;
+	zio->io_child_count = 0;
+	zio->io_parent_count = 0;
+	zio->io_stall = NULL;
+	zio->io_gang_leader = NULL;
+	zio->io_gang_tree = NULL;
+	zio->io_executor = NULL;
+	zio->io_waiter = NULL;
+	zio->io_cksum_report = NULL;
+	zio->io_ena = 0;
+	bzero(zio->io_child_error, sizeof (int) * ZIO_CHILD_TYPES);
+	bzero(zio->io_children,
+	    sizeof (uint64_t) * ZIO_CHILD_TYPES * ZIO_WAIT_TYPES);
+	bzero(&zio->io_bookmark, sizeof (zbookmark_t));
 
 	zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY);
 	zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE);
@@ -570,16 +646,14 @@ zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 		zio_add_child(pio, zio);
 	}
 
+	taskq_init_ent(&zio->io_tqent);
+
 	return (zio);
 }
 
 static void
 zio_destroy(zio_t *zio)
 {
-	list_destroy(&zio->io_parent_list);
-	list_destroy(&zio->io_child_list);
-	mutex_destroy(&zio->io_lock);
-	cv_destroy(&zio->io_cv);
 	kmem_cache_free(zio_cache, zio);
 }
 
@@ -630,7 +704,7 @@ zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 	    zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS &&
 	    zp->zp_compress >= ZIO_COMPRESS_OFF &&
 	    zp->zp_compress < ZIO_COMPRESS_FUNCTIONS &&
-	    zp->zp_type < DMU_OT_NUMTYPES &&
+	    DMU_OT_IS_VALID(zp->zp_type) &&
 	    zp->zp_level < 32 &&
 	    zp->zp_copies > 0 &&
 	    zp->zp_copies <= spa_max_replication(spa) &&
@@ -914,7 +988,7 @@ zio_read_bp_init(zio_t *zio)
 		zio_push_transform(zio, cbuf, psize, psize, zio_decompress);
 	}
 
-	if (!dmu_ot[BP_GET_TYPE(bp)].ot_metadata && BP_GET_LEVEL(bp) == 0)
+	if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0)
 		zio->io_flags |= ZIO_FLAG_DONT_CACHE;
 
 	if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP)
@@ -1073,7 +1147,7 @@ zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline)
 {
 	spa_t *spa = zio->io_spa;
 	zio_type_t t = zio->io_type;
-	int flags = TQ_NOSLEEP | (cutinline ? TQ_FRONT : 0);
+	int flags = (cutinline ? TQ_FRONT : 0);
 
 	/*
 	 * If we're a config writer or a probe, the normal issue and
@@ -1098,8 +1172,14 @@ zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline)
 
 	ASSERT3U(q, <, ZIO_TASKQ_TYPES);
 
-	while (taskq_dispatch(spa->spa_zio_taskq[t][q],
-	    (task_func_t *)zio_execute, zio, flags) == 0); /* do nothing */
+	/*
+	 * NB: We are assuming that the zio can only be dispatched
+	 * to a single taskq at a time.  It would be a grievous error
+	 * to dispatch the zio to another taskq at the same time.
+	 */
+	ASSERT(taskq_empty_ent(&zio->io_tqent));
+	taskq_dispatch_ent(spa->spa_zio_taskq[t][q],
+	    (task_func_t *)zio_execute, zio, flags, &zio->io_tqent);
 }
 
 static boolean_t
@@ -1236,7 +1316,7 @@ zio_wait(zio_t *zio)
 
 	mutex_enter(&zio->io_lock);
 	while (zio->io_executor != NULL)
-		cv_wait(&zio->io_cv, &zio->io_lock);
+		cv_wait_io(&zio->io_cv, &zio->io_lock);
 	mutex_exit(&zio->io_lock);
 
 	error = zio->io_error;
@@ -1781,6 +1861,11 @@ zio_write_gang_block(zio_t *pio)
 	 */
 	pio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
 
+	/*
+	 * We didn't allocate this bp, so make sure it doesn't get unmarked.
+	 */
+	pio->io_flags &= ~ZIO_FLAG_FASTWRITE;
+
 	zio_nowait(zio);
 
 	return (ZIO_PIPELINE_CONTINUE);
@@ -2190,6 +2275,7 @@ zio_dva_allocate(zio_t *zio)
 	flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0;
 	flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ?
 	    METASLAB_GANG_CHILD : 0;
+	flags |= (zio->io_flags & ZIO_FLAG_FASTWRITE) ? METASLAB_FASTWRITE : 0;
 	error = metaslab_alloc(spa, mc, zio->io_size, bp,
 	    zio->io_prop.zp_copies, zio->io_txg, NULL, flags);
 
@@ -2253,20 +2339,29 @@ zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp)
  * Try to allocate an intent log block.  Return 0 on success, errno on failure.
  */
 int
-zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp,
-    uint64_t size, boolean_t use_slog)
+zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, uint64_t size,
+    boolean_t use_slog)
 {
 	int error = 1;
 
 	ASSERT(txg > spa_syncing_txg(spa));
 
-	if (use_slog)
+	/*
+	 * ZIL blocks are always contiguous (i.e. not gang blocks) so we
+	 * set the METASLAB_GANG_AVOID flag so that they don't "fast gang"
+	 * when allocating them.
+	 */
+	if (use_slog) {
 		error = metaslab_alloc(spa, spa_log_class(spa), size,
-		    new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID);
+		    new_bp, 1, txg, NULL,
+		    METASLAB_FASTWRITE | METASLAB_GANG_AVOID);
+	}
 
-	if (error)
+	if (error) {
 		error = metaslab_alloc(spa, spa_normal_class(spa), size,
-		    new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID);
+		    new_bp, 1, txg, NULL,
+		    METASLAB_FASTWRITE | METASLAB_GANG_AVOID);
+	}
 
 	if (error == 0) {
 		BP_SET_LSIZE(new_bp, size);
@@ -2947,9 +3042,11 @@ zio_done(zio_t *zio)
 			 * Reexecution is potentially a huge amount of work.
 			 * Hand it off to the otherwise-unused claim taskq.
 			 */
-			(void) taskq_dispatch(
+			ASSERT(taskq_empty_ent(&zio->io_tqent));
+			(void) taskq_dispatch_ent(
 			    zio->io_spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE],
-			    (task_func_t *)zio_reexecute, zio, TQ_SLEEP);
+			    (task_func_t *)zio_reexecute, zio, 0,
+			    &zio->io_tqent);
 		}
 		return (ZIO_PIPELINE_STOP);
 	}
@@ -2969,6 +3066,11 @@ zio_done(zio_t *zio)
 		zfs_ereport_free_checksum(zcr);
 	}
 
+	if (zio->io_flags & ZIO_FLAG_FASTWRITE && zio->io_bp &&
+	    !BP_IS_HOLE(zio->io_bp)) {
+		metaslab_fastwrite_unmark(zio->io_spa, zio->io_bp);
+	}
+
 	/*
 	 * It is the responsibility of the done callback to ensure that this
 	 * particular zio is no longer discoverable for adoption, and as
@@ -3029,6 +3131,48 @@ static zio_pipe_stage_t *zio_pipeline[] = {
 	zio_done
 };
 
+/* dnp is the dnode for zb1->zb_object */
+boolean_t
+zbookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1,
+    const zbookmark_t *zb2)
+{
+	uint64_t zb1nextL0, zb2thisobj;
+
+	ASSERT(zb1->zb_objset == zb2->zb_objset);
+	ASSERT(zb2->zb_level == 0);
+
+	/*
+	 * A bookmark in the deadlist is considered to be after
+	 * everything else.
+	 */
+	if (zb2->zb_object == DMU_DEADLIST_OBJECT)
+		return (B_TRUE);
+
+	/* The objset_phys_t isn't before anything. */
+	if (dnp == NULL)
+		return (B_FALSE);
+
+	zb1nextL0 = (zb1->zb_blkid + 1) <<
+	    ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT));
+
+	zb2thisobj = zb2->zb_object ? zb2->zb_object :
+	    zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT);
+
+	if (zb1->zb_object == DMU_META_DNODE_OBJECT) {
+		uint64_t nextobj = zb1nextL0 *
+		    (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT;
+		return (nextobj <= zb2thisobj);
+	}
+
+	if (zb1->zb_object < zb2thisobj)
+		return (B_TRUE);
+	if (zb1->zb_object > zb2thisobj)
+		return (B_FALSE);
+	if (zb2->zb_object == DMU_META_DNODE_OBJECT)
+		return (B_FALSE);
+	return (zb1nextL0 <= zb2->zb_blkid);
+}
+
 #if defined(_KERNEL) && defined(HAVE_SPL)
 /* Fault injection */
 EXPORT_SYMBOL(zio_injection_enabled);