Illumos #3086: unnecessarily setting DS_FLAG_INCONSISTENT on async

3086 unnecessarily setting DS_FLAG_INCONSISTENT on async
destroyed datasets
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Approved by: Eric Schrock <Eric.Schrock@delphix.com>

References:
  illumos/illumos-gate@ce636f8
  illumos changeset: 13776:cd512c80fd75
  https://www.illumos.org/issues/3086

Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>

Author: ahrens

cmd/ztest/ztest.c

@@ -2277,6 +2277,7 @@ ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
 {
 	objset_t *os = zd->zd_os;
 
+	mutex_enter(&zd->zd_dirobj_lock);
 	(void) rw_enter(&zd->zd_zilog_lock, RW_WRITER);
 
 	/* zfs_sb_teardown() */
@@ -2287,6 +2288,7 @@ ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
 	zil_replay(os, zd, ztest_replay_vector);
 
 	(void) rw_exit(&zd->zd_zilog_lock);
+	mutex_exit(&zd->zd_dirobj_lock);
 }
 
 /*
@@ -5743,6 +5745,7 @@ ztest_freeze(void)
 	 */
 	kernel_init(FREAD | FWRITE);
 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
+	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
 	VERIFY3U(0, ==, ztest_dataset_open(0));
 	ztest_dataset_close(0);
 

include/sys/dsl_pool.h

@@ -90,7 +90,6 @@ typedef struct dsl_pool {
 
 	/* No lock needed - sync context only */
 	blkptr_t dp_meta_rootbp;
-	list_t dp_synced_datasets;
 	hrtime_t dp_read_overhead;
 	uint64_t dp_throughput; /* bytes per millisec */
 	uint64_t dp_write_limit;
@@ -104,13 +103,17 @@ typedef struct dsl_pool {
 	kmutex_t dp_lock;
 	uint64_t dp_space_towrite[TXG_SIZE];
 	uint64_t dp_tempreserved[TXG_SIZE];
+	uint64_t dp_mos_used_delta;
+	uint64_t dp_mos_compressed_delta;
+	uint64_t dp_mos_uncompressed_delta;
 	uint64_t dp_txg_history_size;
 	list_t dp_txg_history;
 
 
 	/* Has its own locking */
 	tx_state_t dp_tx;
 	txg_list_t dp_dirty_datasets;
+	txg_list_t dp_dirty_zilogs;
 	txg_list_t dp_dirty_dirs;
 	txg_list_t dp_sync_tasks;
 
@@ -150,6 +153,8 @@ int dsl_read_nolock(zio_t *pio, spa_t *spa, const blkptr_t *bpp,
 void dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx);
 void dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx);
 void dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx);
+void dsl_pool_mos_diduse_space(dsl_pool_t *dp,
+    int64_t used, int64_t comp, int64_t uncomp);
 
 taskq_t *dsl_pool_iput_taskq(dsl_pool_t *dp);
 

include/sys/txg.h

@@ -22,6 +22,9 @@
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
+/*
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ */
 
 #ifndef _SYS_TXG_H
 #define	_SYS_TXG_H
@@ -121,7 +124,7 @@ extern void txg_wait_callbacks(struct dsl_pool *dp);
 
 extern void txg_list_create(txg_list_t *tl, size_t offset);
 extern void txg_list_destroy(txg_list_t *tl);
-extern int txg_list_empty(txg_list_t *tl, uint64_t txg);
+extern boolean_t txg_list_empty(txg_list_t *tl, uint64_t txg);
 extern int txg_list_add(txg_list_t *tl, void *p, uint64_t txg);
 extern int txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg);
 extern void *txg_list_remove(txg_list_t *tl, uint64_t txg);

include/sys/zil.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -454,6 +455,7 @@ extern void	zil_replay(objset_t *os, void *arg,
     zil_replay_func_t *replay_func[TX_MAX_TYPE]);
 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
 extern void	zil_destroy(zilog_t *zilog, boolean_t keep_first);
+extern void	zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
 extern void	zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
 
 extern itx_t	*zil_itx_create(uint64_t txtype, size_t lrsize);

include/sys/zil_impl.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -131,6 +132,7 @@ struct zilog {
 	zil_header_t	zl_old_header;	/* debugging aid */
 	uint_t		zl_prev_blks[ZIL_PREV_BLKS]; /* size - sector rounded */
 	uint_t		zl_prev_rotor;	/* rotor for zl_prev[] */
+	txg_node_t	zl_dirty_link;	/* protected by dp_dirty_zilogs list */
 };
 
 typedef struct zil_bp_node {

module/zfs/dmu.c

@@ -1953,15 +1953,15 @@ dmu_init(void)
 	dbuf_init();
 	zfetch_init();
 	dmu_tx_init();
-	arc_init();
 	l2arc_init();
+	arc_init();
 }
 
 void
 dmu_fini(void)
 {
-	l2arc_fini();
 	arc_fini();
+	l2arc_fini();
 	dmu_tx_fini();
 	zfetch_fini();
 	dbuf_fini();

module/zfs/dmu_send.c

@@ -1609,13 +1609,6 @@ dmu_recv_existing_end(dmu_recv_cookie_t *drc)
 	dsl_dataset_t *ds = drc->drc_logical_ds;
 	int err, myerr;
 
-	/*
-	 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
-	 * expects it to have a ds_user_ptr (and zil), but clone_swap()
-	 * can close it.
-	 */
-	txg_wait_synced(ds->ds_dir->dd_pool, 0);
-
 	if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
 		err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
 		    drc->drc_force);

module/zfs/dsl_dataset.c

@@ -105,14 +105,8 @@ dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 	if (ds == NULL) {
-		/*
-		 * Account for the meta-objset space in its placeholder
-		 * dsl_dir.
-		 */
-		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
-		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
-		    used, compressed, uncompressed, tx);
-		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
+		dsl_pool_mos_diduse_space(tx->tx_pool,
+		    used, compressed, uncompressed);
 		return;
 	}
 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
@@ -150,15 +144,9 @@ dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 
 	ASSERT(used > 0);
 	if (ds == NULL) {
-		/*
-		 * Account for the meta-objset space in its placeholder
-		 * dataset.
-		 */
 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
-
-		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
-		    -used, -compressed, -uncompressed, tx);
-		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
+		dsl_pool_mos_diduse_space(tx->tx_pool,
+		    -used, -compressed, -uncompressed);
 		return (used);
 	}
 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
@@ -1074,26 +1062,26 @@ dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
 	dummy_ds->ds_dir = dd;
 	dummy_ds->ds_object = ds->ds_object;
 
-	/*
-	 * Check for errors and mark this ds as inconsistent, in
-	 * case we crash while freeing the objects.
-	 */
-	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
-	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
-	if (err)
-		goto out_free;
-
-	err = dmu_objset_from_ds(ds, &os);
-	if (err)
-		goto out_free;
-
-	/*
-	 * If async destruction is not enabled try to remove all objects
-	 * while in the open context so that there is less work to do in
-	 * the syncing context.
-	 */
 	if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
 	    &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
+		/*
+		 * Check for errors and mark this ds as inconsistent, in
+		 * case we crash while freeing the objects.
+		 */
+		err = dsl_sync_task_do(dd->dd_pool,
+		    dsl_dataset_destroy_begin_check,
+		    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
+		if (err)
+			goto out_free;
+
+		err = dmu_objset_from_ds(ds, &os);
+		if (err)
+			goto out_free;
+
+		/*
+		 * Remove all objects while in the open context so that
+		 * there is less work to do in the syncing context.
+		 */
 		for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
 		    ds->ds_phys->ds_prev_snap_txg)) {
 			/*
@@ -1104,29 +1092,25 @@ dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
 		}
 		if (err != ESRCH)
 			goto out_free;
-	}
 
-	/*
-	 * Only the ZIL knows how to free log blocks.
-	 */
-	zil_destroy(dmu_objset_zil(os), B_FALSE);
-
-	/*
-	 * Sync out all in-flight IO.
-	 */
-	txg_wait_synced(dd->dd_pool, 0);
+		/*
+		 * Sync out all in-flight IO.
+		 */
+		txg_wait_synced(dd->dd_pool, 0);
 
-	/*
-	 * If we managed to free all the objects in open
-	 * context, the user space accounting should be zero.
-	 */
-	if (ds->ds_phys->ds_bp.blk_fill == 0 &&
-	    dmu_objset_userused_enabled(os)) {
-		ASSERTV(uint64_t count);
-		ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
-		    count == 0);
-		ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
-		    count == 0);
+		/*
+		 * If we managed to free all the objects in open
+		 * context, the user space accounting should be zero.
+		 */
+		if (ds->ds_phys->ds_bp.blk_fill == 0 &&
+		    dmu_objset_userused_enabled(os)) {
+			ASSERTV(uint64_t count);
+
+			ASSERT(zap_count(os, DMU_USERUSED_OBJECT,
+			    &count) != 0 || count == 0);
+			ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT,
+			    &count) != 0 || count == 0);
+		}
 	}
 
 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
@@ -1878,6 +1862,7 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
 	} else {
 		zfeature_info_t *async_destroy =
 		    &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
+		objset_t *os;
 
 		/*
 		 * There's no next snapshot, so this is a head dataset.
@@ -1889,6 +1874,8 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
 		dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
 		ds->ds_phys->ds_deadlist_obj = 0;
 
+		VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
+
 		if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
 			err = old_synchronous_dataset_destroy(ds, tx);
 		} else {
@@ -1898,12 +1885,12 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
 			 */
 			uint64_t used, comp, uncomp;
 
-			ASSERT(err == 0 || err == EBUSY);
+			zil_destroy_sync(dmu_objset_zil(os), tx);
+
 			if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
 				spa_feature_incr(dp->dp_spa, async_destroy, tx);
-				dp->dp_bptree_obj = bptree_alloc(
-				    dp->dp_meta_objset, tx);
-				VERIFY(zap_add(dp->dp_meta_objset,
+				dp->dp_bptree_obj = bptree_alloc(mos, tx);
+				VERIFY(zap_add(mos,
 				    DMU_POOL_DIRECTORY_OBJECT,
 				    DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
 				    &dp->dp_bptree_obj, tx) == 0);
@@ -1916,7 +1903,7 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
 			ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
 			    ds->ds_phys->ds_unique_bytes == used);
 
-			bptree_add(dp->dp_meta_objset, dp->dp_bptree_obj,
+			bptree_add(mos, dp->dp_bptree_obj,
 			    &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
 			    used, comp, uncomp, tx);
 			dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
@@ -2203,7 +2190,6 @@ dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
 
-	dsl_dir_dirty(ds->ds_dir, tx);
 	dmu_objset_sync(ds->ds_objset, zio, tx);
 }
 

module/zfs/dsl_dir.c

@@ -189,7 +189,6 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 	kmem_free(dd, sizeof (dsl_dir_t));
 	dmu_buf_rele(dbuf, tag);
 	return (err);
-
 }
 
 void
@@ -223,7 +222,7 @@ dsl_dir_name(dsl_dir_t *dd, char *buf)
 	}
 }
 
-/* Calculate name legnth, avoiding all the strcat calls of dsl_dir_name */
+/* Calculate name length, avoiding all the strcat calls of dsl_dir_name */
 int
 dsl_dir_namelen(dsl_dir_t *dd)
 {
@@ -592,8 +591,6 @@ dsl_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
 {
 	ASSERT(dmu_tx_is_syncing(tx));
 
-	dmu_buf_will_dirty(dd->dd_dbuf, tx);
-
 	mutex_enter(&dd->dd_lock);
 	ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0);
 	dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
@@ -950,8 +947,6 @@ dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT(type < DD_USED_NUM);
 
-	dsl_dir_dirty(dd, tx);
-
 	if (needlock)
 		mutex_enter(&dd->dd_lock);
 	accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used);
@@ -960,6 +955,7 @@ dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type,
 	    dd->dd_phys->dd_compressed_bytes >= -compressed);
 	ASSERT(uncompressed >= 0 ||
 	    dd->dd_phys->dd_uncompressed_bytes >= -uncompressed);
+	dmu_buf_will_dirty(dd->dd_dbuf, tx);
 	dd->dd_phys->dd_used_bytes += used;
 	dd->dd_phys->dd_uncompressed_bytes += uncompressed;
 	dd->dd_phys->dd_compressed_bytes += compressed;
@@ -1003,13 +999,13 @@ dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 	if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN))
 		return;
 
-	dsl_dir_dirty(dd, tx);
 	if (needlock)
 		mutex_enter(&dd->dd_lock);
 	ASSERT(delta > 0 ?
 	    dd->dd_phys->dd_used_breakdown[oldtype] >= delta :
 	    dd->dd_phys->dd_used_breakdown[newtype] >= -delta);
 	ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta));
+	dmu_buf_will_dirty(dd->dd_dbuf, tx);
 	dd->dd_phys->dd_used_breakdown[oldtype] -= delta;
 	dd->dd_phys->dd_used_breakdown[newtype] += delta;
 	if (needlock)