Illumos 5056 - ZFS deadlock on db_mtx and dn_holds

5056 ZFS deadlock on db_mtx and dn_holds
Author: Justin Gibbs <justing@spectralogic.com>
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>

References:
  https://www.illumos.org/issues/5056
  https://github.com/illumos/illumos-gate/commit/bc9014e

Porting Notes:

sa_handle_get_from_db():
  - the original patch includes an otherwise unmentioned fix for a
    possible usage of an uninitialised variable

dmu_objset_open_impl():
  - Under Illumos list_link_init() is the same as filling a list_node_t
    with NULLs, so they don't notice if they miss doing list_link_init()
    on a zero'd containing structure (e.g. allocated with kmem_zalloc as
    here). Under Linux, not so much: an uninitialised list_node_t goes
    "Boom!" some time later when it's used or destroyed.

dmu_objset_evict_dbufs():
  - reduce stack usage using kmem_alloc()

Ported-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>

include/sys/dbuf.h

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef	_SYS_DBUF_H
@@ -226,9 +227,8 @@ typedef struct dmu_buf_impl {
 
 	/* Data which is unique to data (leaf) blocks: */
 
-	/* stuff we store for the user (see dmu_buf_set_user) */
+	/* User callback information. */
-	void *db_user_ptr;
+	dmu_buf_user_t *db_user;
-	dmu_buf_evict_func_t *db_evict_func;
 
 	uint8_t db_immediate_evict;
 	uint8_t db_freed_in_flight;

include/sys/dmu.h

@@ -24,6 +24,7 @@
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright 2014 HybridCluster. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -39,11 +40,9 @@
  * dmu_spa.h.
  */
 
+#include <sys/zfs_context.h>
 #include <sys/inttypes.h>
-#include <sys/types.h>
-#include <sys/param.h>
 #include <sys/cred.h>
-#include <sys/time.h>
 #include <sys/fs/zfs.h>
 #include <sys/uio.h>
 
@@ -288,8 +287,6 @@ typedef struct dmu_buf {
 	void *db_data;			/* data in buffer */
 } dmu_buf_t;
 
-typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
-
 /*
  * The names of zap entries in the DIRECTORY_OBJECT of the MOS.
  */
@@ -475,36 +472,126 @@ int dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
     uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
 void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
 
-/*
+typedef void dmu_buf_evict_func_t(void *user_ptr);
- * Returns NULL on success, or the existing user ptr if it's already
- * been set.
- *
- * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
- *
- * If non-NULL, pageout func will be called when this buffer is being
- * excised from the cache, so that you can clean up the data structure
- * pointed to by user_ptr.
- *
- * dmu_evict_user() will call the pageout func for all buffers in a
- * objset with a given pageout func.
- */
-void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr,
-    dmu_buf_evict_func_t *pageout_func);
-/*
- * set_user_ie is the same as set_user, but request immediate eviction
- * when hold count goes to zero.
- */
-void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
-    dmu_buf_evict_func_t *pageout_func);
-void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
-    void *user_ptr, dmu_buf_evict_func_t *pageout_func);
-void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
 
 /*
- * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
+ * A DMU buffer user object may be associated with a dbuf for the
+ * duration of its lifetime.  This allows the user of a dbuf (client)
+ * to attach private data to a dbuf (e.g. in-core only data such as a
+ * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
+ * when that dbuf has been evicted.  Clients typically respond to the
+ * eviction notification by freeing their private data, thus ensuring
+ * the same lifetime for both dbuf and private data.
+ *
+ * The mapping from a dmu_buf_user_t to any client private data is the
+ * client's responsibility.  All current consumers of the API with private
+ * data embed a dmu_buf_user_t as the first member of the structure for
+ * their private data.  This allows conversions between the two types
+ * with a simple cast.  Since the DMU buf user API never needs access
+ * to the private data, other strategies can be employed if necessary
+ * or convenient for the client (e.g. using container_of() to do the
+ * conversion for private data that cannot have the dmu_buf_user_t as
+ * its first member).
+ *
+ * Eviction callbacks are executed without the dbuf mutex held or any
+ * other type of mechanism to guarantee that the dbuf is still available.
+ * For this reason, users must assume the dbuf has already been freed
+ * and not reference the dbuf from the callback context.
+ *
+ * Users requesting "immediate eviction" are notified as soon as the dbuf
+ * is only referenced by dirty records (dirties == holds).  Otherwise the
+ * notification occurs after eviction processing for the dbuf begins.
+ */
+typedef struct dmu_buf_user {
+	/*
+	 * Asynchronous user eviction callback state.
+	 */
+	taskq_ent_t	dbu_tqent;
+
+	/* This instance's eviction function pointer. */
+	dmu_buf_evict_func_t *dbu_evict_func;
+#ifdef ZFS_DEBUG
+	/*
+	 * Pointer to user's dbuf pointer.  NULL for clients that do
+	 * not associate a dbuf with their user data.
+	 *
+	 * The dbuf pointer is cleared upon eviction so as to catch
+	 * use-after-evict bugs in clients.
+	 */
+	dmu_buf_t **dbu_clear_on_evict_dbufp;
+#endif
+} dmu_buf_user_t;
+
+/*
+ * Initialize the given dmu_buf_user_t instance with the eviction function
+ * evict_func, to be called when the user is evicted.
+ *
+ * NOTE: This function should only be called once on a given dmu_buf_user_t.
+ *       To allow enforcement of this, dbu must already be zeroed on entry.
+ */
+#ifdef __lint
+/* Very ugly, but it beats issuing suppression directives in many Makefiles. */
+extern void
+dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
+    dmu_buf_t **clear_on_evict_dbufp);
+#else /* __lint */
+static inline void
+dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
+    dmu_buf_t **clear_on_evict_dbufp)
+{
+	ASSERT(dbu->dbu_evict_func == NULL);
+	ASSERT(evict_func != NULL);
+	dbu->dbu_evict_func = evict_func;
+#ifdef ZFS_DEBUG
+	dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
+#endif
+}
+#endif /* __lint */
+
+/*
+ * Attach user data to a dbuf and mark it for normal (when the dbuf's
+ * data is cleared or its reference count goes to zero) eviction processing.
+ *
+ * Returns NULL on success, or the existing user if another user currently
+ * owns the buffer.
+ */
+void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
+
+/*
+ * Attach user data to a dbuf and mark it for immediate (its dirty and
+ * reference counts are equal) eviction processing.
+ *
+ * Returns NULL on success, or the existing user if another user currently
+ * owns the buffer.
+ */
+void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
+
+/*
+ * Replace the current user of a dbuf.
+ *
+ * If given the current user of a dbuf, replaces the dbuf's user with
+ * "new_user" and returns the user data pointer that was replaced.
+ * Otherwise returns the current, and unmodified, dbuf user pointer.
+ */
+void *dmu_buf_replace_user(dmu_buf_t *db,
+    dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
+
+/*
+ * Remove the specified user data for a DMU buffer.
+ *
+ * Returns the user that was removed on success, or the current user if
+ * another user currently owns the buffer.
+ */
+void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
+
+/*
+ * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
  */
 void *dmu_buf_get_user(dmu_buf_t *db);
 
+/* Block until any in-progress dmu buf user evictions complete. */
+void dmu_buf_user_evict_wait(void);
+
 /*
  * Returns the blkptr associated with this dbuf, or NULL if not set.
  */

include/sys/dmu_objset.h

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -74,22 +75,25 @@ struct objset {
 	arc_buf_t *os_phys_buf;
 	objset_phys_t *os_phys;
 	/*
-	 * The following "special" dnodes have no parent and are exempt from
+	 * The following "special" dnodes have no parent, are exempt
-	 * dnode_move(), but they root their descendents in this objset using
+	 * from dnode_move(), and are not recorded in os_dnodes, but they
-	 * handles anyway, so that all access to dnodes from dbufs consistently
+	 * root their descendents in this objset using handles anyway, so
-	 * uses handles.
+	 * that all access to dnodes from dbufs consistently uses handles.
 	 */
 	dnode_handle_t os_meta_dnode;
 	dnode_handle_t os_userused_dnode;
 	dnode_handle_t os_groupused_dnode;
 	zilog_t *os_zil;
 
+	list_node_t os_evicting_node;
+
 	/* can change, under dsl_dir's locks: */
 	enum zio_checksum os_checksum;
 	enum zio_compress os_compress;
 	uint8_t os_copies;
 	enum zio_checksum os_dedup_checksum;
 	boolean_t os_dedup_verify;
+	boolean_t os_evicting;
 	zfs_logbias_op_t os_logbias;
 	zfs_cache_type_t os_primary_cache;
 	zfs_cache_type_t os_secondary_cache;
@@ -168,6 +172,8 @@ int dmu_objset_userspace_upgrade(objset_t *os);
 boolean_t dmu_objset_userspace_present(objset_t *os);
 int dmu_fsname(const char *snapname, char *buf);
 
+void dmu_objset_evict_done(objset_t *os);
+
 void dmu_objset_init(void);
 void dmu_objset_fini(void);
 

include/sys/dnode.h

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef	_SYS_DNODE_H
@@ -277,6 +278,7 @@ typedef struct dnode_handle {
 } dnode_handle_t;
 
 typedef struct dnode_children {
+	dmu_buf_user_t dnc_dbu;		/* User evict data */
 	size_t dnc_count;		/* number of children */
 	dnode_handle_t dnc_children[];	/* sized dynamically */
 } dnode_children_t;
@@ -287,7 +289,7 @@ typedef struct free_range {
 	uint64_t fr_nblks;
 } free_range_t;
 
-dnode_t *dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
+void dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
     uint64_t object, dnode_handle_t *dnh);
 void dnode_special_close(dnode_handle_t *dnh);
 

include/sys/dsl_dataset.h

@@ -23,6 +23,7 @@
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  * Copyright (c) 2013 Steven Hartland. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef	_SYS_DSL_DATASET_H
@@ -125,11 +126,14 @@ typedef struct dsl_dataset_phys {
 } dsl_dataset_phys_t;
 
 typedef struct dsl_dataset {
+	dmu_buf_user_t ds_dbu;
+
 	/* Immutable: */
 	struct dsl_dir *ds_dir;
 	dmu_buf_t *ds_dbuf;
 	uint64_t ds_object;
 	uint64_t ds_fsid_guid;
+	boolean_t ds_is_snapshot;
 
 	/* only used in syncing context, only valid for non-snapshots: */
 	struct dsl_dataset *ds_prev;
@@ -188,12 +192,6 @@ dsl_dataset_phys(dsl_dataset_t *ds)
  */
 #define	MAX_TAG_PREFIX_LEN	17
 
-static inline boolean_t
-dsl_dataset_is_snapshot(dsl_dataset_t *ds)
-{
-	return (dsl_dataset_phys(ds)->ds_num_children != 0);
-}
-
 #define	DS_UNIQUE_IS_ACCURATE(ds)	\
 	((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_UNIQUE_ACCURATE) != 0)
 

include/sys/dsl_dir.h

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef	_SYS_DSL_DIR_H
@@ -84,6 +85,8 @@ typedef struct dsl_dir_phys {
 } dsl_dir_phys_t;
 
 struct dsl_dir {
+	dmu_buf_user_t dd_dbu;
+
 	/* These are immutable; no lock needed: */
 	uint64_t dd_object;
 	dsl_pool_t *dd_pool;
@@ -119,6 +122,7 @@ dsl_dir_phys(dsl_dir_t *dd)
 }
 
 void dsl_dir_rele(dsl_dir_t *dd, void *tag);
+void dsl_dir_async_rele(dsl_dir_t *dd, void *tag);
 int dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
     dsl_dir_t **, const char **tail);
 int dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,

include/sys/sa.h

@@ -133,7 +133,6 @@ int sa_update_from_cb(sa_handle_t *, sa_attr_type_t,
     uint32_t buflen, sa_data_locator_t *, void *userdata, dmu_tx_t *);
 void sa_object_info(sa_handle_t *, dmu_object_info_t *);
 void sa_object_size(sa_handle_t *, uint32_t *, u_longlong_t *);
-void sa_update_user(sa_handle_t *, sa_handle_t *);
 void *sa_get_userdata(sa_handle_t *);
 void sa_set_userp(sa_handle_t *, void *);
 dmu_buf_t *sa_get_db(sa_handle_t *);

include/sys/sa_impl.h

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef	_SYS_SA_IMPL_H
@@ -208,11 +209,12 @@ typedef enum sa_data_op {
  */
 
 struct sa_handle {
+	dmu_buf_user_t	sa_dbu;
 	kmutex_t	sa_lock;
 	dmu_buf_t	*sa_bonus;
 	dmu_buf_t	*sa_spill;
 	objset_t	*sa_os;
-	void 		*sa_userp;
+	void		*sa_userp;
 	sa_idx_tab_t	*sa_bonus_tab;	 /* idx of bonus */
 	sa_idx_tab_t	*sa_spill_tab; /* only present if spill activated */
 };

include/sys/spa.h

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef _SYS_SPA_H
@@ -680,6 +681,7 @@ extern spa_t *spa_next(spa_t *prev);
 /* Refcount functions */
 extern void spa_open_ref(spa_t *spa, void *tag);
 extern void spa_close(spa_t *spa, void *tag);
+extern void spa_async_close(spa_t *spa, void *tag);
 extern boolean_t spa_refcount_zero(spa_t *spa);
 
 #define	SCL_NONE	0x00
@@ -789,6 +791,9 @@ extern uint64_t spa_version(spa_t *spa);
 extern boolean_t spa_deflate(spa_t *spa);
 extern metaslab_class_t *spa_normal_class(spa_t *spa);
 extern metaslab_class_t *spa_log_class(spa_t *spa);
+extern void spa_evicting_os_register(spa_t *, objset_t *os);
+extern void spa_evicting_os_deregister(spa_t *, objset_t *os);
+extern void spa_evicting_os_wait(spa_t *spa);
 extern int spa_max_replication(spa_t *spa);
 extern int spa_prev_software_version(spa_t *spa);
 extern uint8_t spa_get_failmode(spa_t *spa);

include/sys/spa_impl.h

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef _SYS_SPA_IMPL_H
@@ -144,6 +145,9 @@ struct spa {
 	uint64_t	spa_claim_max_txg;	/* highest claimed birth txg */
 	timespec_t	spa_loaded_ts;		/* 1st successful open time */
 	objset_t	*spa_meta_objset;	/* copy of dp->dp_meta_objset */
+	kmutex_t	spa_evicting_os_lock;	/* Evicting objset list lock */
+	list_t		spa_evicting_os_list;	/* Objsets being evicted. */
+	kcondvar_t	spa_evicting_os_cv;	/* Objset Eviction Completion */
 	txg_list_t	spa_vdev_txg_list;	/* per-txg dirty vdev list */
 	vdev_t		*spa_root_vdev;		/* top-level vdev container */
 	uint64_t	spa_config_guid;	/* config pool guid */

include/sys/zap_impl.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef	_SYS_ZAP_IMPL_H
@@ -140,6 +141,7 @@ typedef struct zap_phys {
 typedef struct zap_table_phys zap_table_phys_t;
 
 typedef struct zap {
+	dmu_buf_user_t zap_dbu;
 	objset_t *zap_objset;
 	uint64_t zap_object;
 	struct dmu_buf *zap_dbuf;
@@ -196,7 +198,7 @@ boolean_t zap_match(zap_name_t *zn, const char *matchname);
 int zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
     krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp);
 void zap_unlockdir(zap_t *zap);
-void zap_evict(dmu_buf_t *db, void *vmzap);
+void zap_evict(void *dbu);
 zap_name_t *zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt);
 void zap_name_free(zap_name_t *zn);
 int zap_hashbits(zap_t *zap);

include/sys/zap_leaf.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #ifndef	_SYS_ZAP_LEAF_H
@@ -152,6 +153,7 @@ typedef union zap_leaf_chunk {
 } zap_leaf_chunk_t;
 
 typedef struct zap_leaf {
+	dmu_buf_user_t l_dbu;
 	krwlock_t l_rwlock;
 	uint64_t l_blkid;		/* 1<<ZAP_BLOCK_SHIFT byte block off */
 	int l_bs;			/* block size shift */

module/zfs/dbuf.c

@@ -23,6 +23,7 @@
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -78,10 +79,16 @@ static void dbuf_destroy(dmu_buf_impl_t *db);
 static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx);
 
+#ifndef __lint
+extern inline void dmu_buf_init_user(dmu_buf_user_t *dbu,
+    dmu_buf_evict_func_t *evict_func, dmu_buf_t **clear_on_evict_dbufp);
+#endif /* ! __lint */
+
 /*
  * Global data structures and functions for the dbuf cache.
  */
 static kmem_cache_t *dbuf_cache;
+static taskq_t *dbu_evict_taskq;
 
 /* ARGSUSED */
 static int
@@ -247,17 +254,72 @@ dbuf_hash_remove(dmu_buf_impl_t *db)
 
 static arc_evict_func_t dbuf_do_evict;
 
+typedef enum {
+	DBVU_EVICTING,
+	DBVU_NOT_EVICTING
+} dbvu_verify_type_t;
+
+static void
+dbuf_verify_user(dmu_buf_impl_t *db, dbvu_verify_type_t verify_type)
+{
+#ifdef ZFS_DEBUG
+	int64_t holds;
+
+	if (db->db_user == NULL)
+		return;
+
+	/* Only data blocks support the attachment of user data. */
+	ASSERT(db->db_level == 0);
+
+	/* Clients must resolve a dbuf before attaching user data. */
+	ASSERT(db->db.db_data != NULL);
+	ASSERT3U(db->db_state, ==, DB_CACHED);
+
+	holds = refcount_count(&db->db_holds);
+	if (verify_type == DBVU_EVICTING) {
+		/*
+		 * Immediate eviction occurs when holds == dirtycnt.
+		 * For normal eviction buffers, holds is zero on
+		 * eviction, except when dbuf_fix_old_data() calls
+		 * dbuf_clear_data().  However, the hold count can grow
+		 * during eviction even though db_mtx is held (see
+		 * dmu_bonus_hold() for an example), so we can only
+		 * test the generic invariant that holds >= dirtycnt.
+		 */
+		ASSERT3U(holds, >=, db->db_dirtycnt);
+	} else {
+		if (db->db_immediate_evict == TRUE)
+			ASSERT3U(holds, >=, db->db_dirtycnt);
+		else
+			ASSERT3U(holds, >, 0);
+	}
+#endif
+}
+
 static void
 dbuf_evict_user(dmu_buf_impl_t *db)
 {
+	dmu_buf_user_t *dbu = db->db_user;
+
 	ASSERT(MUTEX_HELD(&db->db_mtx));
 
-	if (db->db_level != 0 || db->db_evict_func == NULL)
+	if (dbu == NULL)
 		return;
 
-	db->db_evict_func(&db->db, db->db_user_ptr);
+	dbuf_verify_user(db, DBVU_EVICTING);
-	db->db_user_ptr = NULL;
+	db->db_user = NULL;
-	db->db_evict_func = NULL;
+
+#ifdef ZFS_DEBUG
+	if (dbu->dbu_clear_on_evict_dbufp != NULL)
+		*dbu->dbu_clear_on_evict_dbufp = NULL;
+#endif
+
+	/*
+	 * Invoke the callback from a taskq to avoid lock order reversals
+	 * and limit stack depth.
+	 */
+	taskq_dispatch_ent(dbu_evict_taskq, dbu->dbu_evict_func, dbu, 0,
+	    &dbu->dbu_tqent);
 }
 
 boolean_t
@@ -331,6 +393,12 @@ retry:
 		mutex_init(&h->hash_mutexes[i], NULL, MUTEX_DEFAULT, NULL);
 
 	dbuf_stats_init(h);
+
+	/*
+	 * All entries are queued via taskq_dispatch_ent(), so min/maxalloc
+	 * configuration is not required.
+	 */
+	dbu_evict_taskq = taskq_create("dbu_evict", 1, minclsyspri, 0, 0, 0);
 }
 
 void
@@ -353,6 +421,7 @@ dbuf_fini(void)
 	kmem_free(h->hash_table, (h->hash_table_mask + 1) * sizeof (void *));
 #endif
 	kmem_cache_destroy(dbuf_cache);
+	taskq_destroy(dbu_evict_taskq);
 }
 
 /*
@@ -470,22 +539,28 @@ dbuf_verify(dmu_buf_impl_t *db)
 }
 #endif
 
+static void
+dbuf_clear_data(dmu_buf_impl_t *db)
+{
+	ASSERT(MUTEX_HELD(&db->db_mtx));
+	dbuf_evict_user(db);
+	db->db_buf = NULL;
+	db->db.db_data = NULL;
+	if (db->db_state != DB_NOFILL)
+		db->db_state = DB_UNCACHED;
+}
+
 static void
 dbuf_set_data(dmu_buf_impl_t *db, arc_buf_t *buf)
 {
 	ASSERT(MUTEX_HELD(&db->db_mtx));
+	ASSERT(buf != NULL);
+
 	db->db_buf = buf;
-	if (buf != NULL) {
+	ASSERT(buf->b_data != NULL);
-		ASSERT(buf->b_data != NULL);
+	db->db.db_data = buf->b_data;
-		db->db.db_data = buf->b_data;
+	if (!arc_released(buf))
-		if (!arc_released(buf))
+		arc_set_callback(buf, dbuf_do_evict, db);
-			arc_set_callback(buf, dbuf_do_evict, db);
-	} else {
-		dbuf_evict_user(db);
-		db->db.db_data = NULL;
-		if (db->db_state != DB_NOFILL)
-			db->db_state = DB_UNCACHED;
-	}
 }
 
 /*
@@ -507,7 +582,7 @@ dbuf_loan_arcbuf(dmu_buf_impl_t *db)
 	} else {
 		abuf = db->db_buf;
 		arc_loan_inuse_buf(abuf, db);
-		dbuf_set_data(db, NULL);
+		dbuf_clear_data(db);
 		mutex_exit(&db->db_mtx);
 	}
 	return (abuf);
@@ -747,7 +822,7 @@ dbuf_noread(dmu_buf_impl_t *db)
 		dbuf_set_data(db, arc_buf_alloc(spa, db->db.db_size, db, type));
 		db->db_state = DB_FILL;
 	} else if (db->db_state == DB_NOFILL) {
-		dbuf_set_data(db, NULL);
+		dbuf_clear_data(db);
 	} else {
 		ASSERT3U(db->db_state, ==, DB_CACHED);
 	}
@@ -803,7 +878,7 @@ dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg)
 		dr->dt.dl.dr_data = arc_buf_alloc(spa, size, db, type);
 		bcopy(db->db.db_data, dr->dt.dl.dr_data->b_data, size);
 	} else {
-		dbuf_set_data(db, NULL);
+		dbuf_clear_data(db);
 	}
 }
 
@@ -854,7 +929,8 @@ void
 dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid,
     dmu_tx_t *tx)
 {
-	dmu_buf_impl_t *db, *db_next, *db_search;
+	dmu_buf_impl_t *db_search;
+	dmu_buf_impl_t *db, *db_next;
 	uint64_t txg = tx->tx_txg;
 	avl_index_t where;
 	boolean_t freespill =
@@ -864,7 +940,7 @@ dbuf_free_range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid,
 		end_blkid = dn->dn_maxblkid;
 	dprintf_dnode(dn, "start=%llu end=%llu\n", start_blkid, end_blkid);
 
-	db_seach = kmem_alloc(sizeof (dmu_buf_impl_t), KM_SLEEP);
+	db_search = kmem_alloc(sizeof (dmu_buf_impl_t), KM_SLEEP);
 	db_search->db_level = 0;
 	db_search->db_blkid = start_blkid;
 	db_search->db_state = DB_SEARCH;
@@ -1436,7 +1512,7 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
 		arc_buf_t *buf = db->db_buf;
 
 		ASSERT(db->db_state == DB_NOFILL || arc_released(buf));
-		dbuf_set_data(db, NULL);
+		dbuf_clear_data(db);
 		VERIFY(arc_buf_remove_ref(buf, db));
 		dbuf_evict(db);
 		return (B_TRUE);
@@ -1785,8 +1861,7 @@ dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid,
 	db->db_parent = parent;
 	db->db_blkptr = blkptr;
 
-	db->db_user_ptr = NULL;
+	db->db_user = NULL;
-	db->db_evict_func = NULL;
 	db->db_immediate_evict = 0;
 	db->db_freed_in_flight = 0;
 
@@ -2273,7 +2348,7 @@ dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag)
 			/*
 			 * This dbuf has anonymous data associated with it.
 			 */
-			dbuf_set_data(db, NULL);
+			dbuf_clear_data(db);
 			VERIFY(arc_buf_remove_ref(buf, db));
 			dbuf_evict(db);
 		} else {
@@ -2306,7 +2381,8 @@ dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag)
 				} else {
 					dbuf_clear(db);
 				}
-			} else if (arc_buf_eviction_needed(db->db_buf)) {
+			} else if (db->db_objset->os_evicting ||
+			    arc_buf_eviction_needed(db->db_buf)) {
 				dbuf_clear(db);
 			} else {
 				mutex_exit(&db->db_mtx);
@@ -2325,51 +2401,57 @@ dbuf_refcount(dmu_buf_impl_t *db)
 }
 
 void *
-dmu_buf_set_user(dmu_buf_t *db_fake, void *user_ptr,
+dmu_buf_replace_user(dmu_buf_t *db_fake, dmu_buf_user_t *old_user,
-    dmu_buf_evict_func_t *evict_func)
+    dmu_buf_user_t *new_user)
 {
-	return (dmu_buf_update_user(db_fake, NULL, user_ptr, evict_func));
+	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
+
+	mutex_enter(&db->db_mtx);
+	dbuf_verify_user(db, DBVU_NOT_EVICTING);
+	if (db->db_user == old_user)
+		db->db_user = new_user;
+	else
+		old_user = db->db_user;
+	dbuf_verify_user(db, DBVU_NOT_EVICTING);
+	mutex_exit(&db->db_mtx);
+
+	return (old_user);
 }
 
 void *
-dmu_buf_set_user_ie(dmu_buf_t *db_fake, void *user_ptr,
+dmu_buf_set_user(dmu_buf_t *db_fake, dmu_buf_user_t *user)
-    dmu_buf_evict_func_t *evict_func)
+{
+	return (dmu_buf_replace_user(db_fake, NULL, user));
+}
+
+void *
+dmu_buf_set_user_ie(dmu_buf_t *db_fake, dmu_buf_user_t *user)
 {
 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
 
 	db->db_immediate_evict = TRUE;
-	return (dmu_buf_update_user(db_fake, NULL, user_ptr, evict_func));
+	return (dmu_buf_set_user(db_fake, user));
 }
 
 void *
-dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr, void *user_ptr,
+dmu_buf_remove_user(dmu_buf_t *db_fake, dmu_buf_user_t *user)
-    dmu_buf_evict_func_t *evict_func)
 {
-	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
+	return (dmu_buf_replace_user(db_fake, user, NULL));
-	ASSERT(db->db_level == 0);
-
-	ASSERT((user_ptr == NULL) == (evict_func == NULL));
-
-	mutex_enter(&db->db_mtx);
-
-	if (db->db_user_ptr == old_user_ptr) {
-		db->db_user_ptr = user_ptr;
-		db->db_evict_func = evict_func;
-	} else {
-		old_user_ptr = db->db_user_ptr;
-	}
-
-	mutex_exit(&db->db_mtx);
-	return (old_user_ptr);
 }
 
 void *
 dmu_buf_get_user(dmu_buf_t *db_fake)
 {
 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
-	ASSERT(!refcount_is_zero(&db->db_holds));
 
-	return (db->db_user_ptr);
+	dbuf_verify_user(db, DBVU_NOT_EVICTING);
+	return (db->db_user);
+}
+
+void
+dmu_buf_user_evict_wait()
+{
+	taskq_wait(dbu_evict_taskq);
 }
 
 boolean_t
@@ -3037,7 +3119,6 @@ EXPORT_SYMBOL(dbuf_refcount);
 EXPORT_SYMBOL(dbuf_sync_list);
 EXPORT_SYMBOL(dmu_buf_set_user);
 EXPORT_SYMBOL(dmu_buf_set_user_ie);
-EXPORT_SYMBOL(dmu_buf_update_user);
 EXPORT_SYMBOL(dmu_buf_get_user);
 EXPORT_SYMBOL(dmu_buf_freeable);
 EXPORT_SYMBOL(dmu_buf_get_blkptr);

module/zfs/dmu_objset.c

@@ -23,6 +23,7 @@
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -347,7 +348,7 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 			    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
 			    secondary_cache_changed_cb, os);
 		}
-		if (!dsl_dataset_is_snapshot(ds)) {
+		if (!ds->ds_is_snapshot) {
 			if (err == 0) {
 				err = dsl_prop_register(ds,
 				    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
@@ -404,7 +405,7 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 		os->os_secondary_cache = ZFS_CACHE_ALL;
 	}
 
-	if (ds == NULL || !dsl_dataset_is_snapshot(ds))
+	if (ds == NULL || !ds->ds_is_snapshot)
 		os->os_zil_header = os->os_phys->os_zil_header;
 	os->os_zil = zil_alloc(os, &os->os_zil_header);
 
@@ -419,20 +420,19 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
 	    offsetof(dmu_buf_impl_t, db_link));
 
+	list_link_init(&os->os_evicting_node);
+
 	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
 
-	DMU_META_DNODE(os) = dnode_special_open(os,
+	dnode_special_open(os, &os->os_phys->os_meta_dnode,
-	    &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
+	    DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
-	    &os->os_meta_dnode);
 	if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
-		DMU_USERUSED_DNODE(os) = dnode_special_open(os,
+		dnode_special_open(os, &os->os_phys->os_userused_dnode,
-		    &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
+		    DMU_USERUSED_OBJECT, &os->os_userused_dnode);
-		    &os->os_userused_dnode);
+		dnode_special_open(os, &os->os_phys->os_groupused_dnode,
-		DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
+		    DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
-		    &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
-		    &os->os_groupused_dnode);
 	}
 
 	*osp = os;
@@ -520,7 +520,7 @@ dmu_objset_own(const char *name, dmu_objset_type_t type,
 	} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
 		dsl_dataset_disown(ds, tag);
 		return (SET_ERROR(EINVAL));
-	} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
+	} else if (!readonly && ds->ds_is_snapshot) {
 		dsl_dataset_disown(ds, tag);
 		return (SET_ERROR(EROFS));
 	}
@@ -576,41 +576,57 @@ dmu_objset_disown(objset_t *os, void *tag)
 void
 dmu_objset_evict_dbufs(objset_t *os)
 {
+	dnode_t *dn_marker;
 	dnode_t *dn;
 
+	dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP);
+
 	mutex_enter(&os->os_lock);
+	dn = list_head(&os->os_dnodes);
+	while (dn != NULL) {
+		/*
+		 * Skip dnodes without holds.  We have to do this dance
+		 * because dnode_add_ref() only works if there is already a
+		 * hold.  If the dnode has no holds, then it has no dbufs.
+		 */
+		if (dnode_add_ref(dn, FTAG)) {
+			list_insert_after(&os->os_dnodes, dn, dn_marker);
+			mutex_exit(&os->os_lock);
 
-	/* process the mdn last, since the other dnodes have holds on it */
+			dnode_evict_dbufs(dn);
-	list_remove(&os->os_dnodes, DMU_META_DNODE(os));
+			dnode_rele(dn, FTAG);
-	list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
 
-	/*
+			mutex_enter(&os->os_lock);
-	 * Find the first dnode with holds.  We have to do this dance
+			dn = list_next(&os->os_dnodes, dn_marker);
-	 * because dnode_add_ref() only works if you already have a
+			list_remove(&os->os_dnodes, dn_marker);
-	 * hold.  If there are no holds then it has no dbufs so OK to
+		} else {
-	 * skip.
+			dn = list_next(&os->os_dnodes, dn);
-	 */
+		}
-	for (dn = list_head(&os->os_dnodes);
-	    dn && !dnode_add_ref(dn, FTAG);
-	    dn = list_next(&os->os_dnodes, dn))
-		continue;
-
-	while (dn) {
-		dnode_t *next_dn = dn;
-
-		do {
-			next_dn = list_next(&os->os_dnodes, next_dn);
-		} while (next_dn && !dnode_add_ref(next_dn, FTAG));
-
-		mutex_exit(&os->os_lock);
-		dnode_evict_dbufs(dn);
-		dnode_rele(dn, FTAG);
-		mutex_enter(&os->os_lock);
-		dn = next_dn;
 	}
 	mutex_exit(&os->os_lock);
+
+	kmem_free(dn_marker, sizeof (dnode_t));
+
+	if (DMU_USERUSED_DNODE(os) != NULL) {
+		dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
+		dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
+	}
+	dnode_evict_dbufs(DMU_META_DNODE(os));
 }
 
+/*
+ * Objset eviction processing is split into into two pieces.
+ * The first marks the objset as evicting, evicts any dbufs that
+ * have a refcount of zero, and then queues up the objset for the
+ * second phase of eviction.  Once os->os_dnodes has been cleared by
+ * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
+ * The second phase closes the special dnodes, dequeues the objset from
+ * the list of those undergoing eviction, and finally frees the objset.
+ *
+ * NOTE: Due to asynchronous eviction processing (invocation of
+ *       dnode_buf_pageout()), it is possible for the meta dnode for the
+ *       objset to have no holds even though os->os_dnodes is not empty.
+ */
 void
 dmu_objset_evict(objset_t *os)
 {
@@ -622,7 +638,7 @@ dmu_objset_evict(objset_t *os)
 		ASSERT(!dmu_objset_is_dirty(os, t));
 
 	if (ds) {
-		if (!dsl_dataset_is_snapshot(ds)) {
+		if (!ds->ds_is_snapshot) {
 			VERIFY0(dsl_prop_unregister(ds,
 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
 			    checksum_changed_cb, os));
@@ -656,8 +672,24 @@ dmu_objset_evict(objset_t *os)
 	if (os->os_sa)
 		sa_tear_down(os);
 
+	os->os_evicting = B_TRUE;
 	dmu_objset_evict_dbufs(os);
 
+	mutex_enter(&os->os_lock);
+	spa_evicting_os_register(os->os_spa, os);
+	if (list_is_empty(&os->os_dnodes)) {
+		mutex_exit(&os->os_lock);
+		dmu_objset_evict_done(os);
+	} else {
+		mutex_exit(&os->os_lock);
+	}
+}
+
+void
+dmu_objset_evict_done(objset_t *os)
+{
+	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
+
 	dnode_special_close(&os->os_meta_dnode);
 	if (DMU_USERUSED_DNODE(os)) {
 		dnode_special_close(&os->os_userused_dnode);
@@ -665,8 +697,6 @@ dmu_objset_evict(objset_t *os)
 	}
 	zil_free(os->os_zil);
 
-	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
-
 	VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
 
 	/*
@@ -681,6 +711,7 @@ dmu_objset_evict(objset_t *os)
 	mutex_destroy(&os->os_lock);
 	mutex_destroy(&os->os_obj_lock);
 	mutex_destroy(&os->os_user_ptr_lock);
+	spa_evicting_os_deregister(os->os_spa, os);
 	kmem_free(os, sizeof (objset_t));
 }
 
@@ -888,7 +919,7 @@ dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
 	}
 
 	/* You can only clone snapshots, not the head datasets. */
-	if (!dsl_dataset_is_snapshot(origin)) {
+	if (!origin->ds_is_snapshot) {
 		dsl_dataset_rele(origin, FTAG);
 		return (SET_ERROR(EINVAL));
 	}
@@ -1453,7 +1484,7 @@ int
 dmu_objset_is_snapshot(objset_t *os)
 {
 	if (os->os_dsl_dataset != NULL)
-		return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
+		return (os->os_dsl_dataset->ds_is_snapshot);
 	else
 		return (B_FALSE);
 }

module/zfs/dmu_send.c

@@ -615,7 +615,7 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 		fromtxg = fromzb->zbm_creation_txg;
 	}
 	dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
-	if (!dsl_dataset_is_snapshot(ds)) {
+	if (!ds->ds_is_snapshot) {
 		(void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
 		    sizeof (drr->drr_u.drr_begin.drr_toname));
 	}
@@ -820,7 +820,7 @@ dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
 	ASSERT(dsl_pool_config_held(dp));
 
 	/* tosnap must be a snapshot */
-	if (!dsl_dataset_is_snapshot(ds))
+	if (!ds->ds_is_snapshot)
 		return (SET_ERROR(EINVAL));
 
 	/*
@@ -1057,7 +1057,7 @@ dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 				dsl_dataset_rele(ds, FTAG);
 				return (error);
 			}
-			if (!dsl_dataset_is_snapshot(origin)) {
+			if (!origin->ds_is_snapshot) {
 				dsl_dataset_rele(origin, FTAG);
 				dsl_dataset_rele(ds, FTAG);
 				return (SET_ERROR(EINVAL));

module/zfs/dmu_traverse.c

@@ -541,7 +541,7 @@ traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
 
 	/* See comment on ZIL traversal in dsl_scan_visitds. */
-	if (ds != NULL && !dsl_dataset_is_snapshot(ds) && !BP_IS_HOLE(rootbp)) {
+	if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) {
 		uint32_t flags = ARC_WAIT;
 		objset_phys_t *osp;
 		arc_buf_t *buf;

module/zfs/dnode.c

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -405,8 +406,9 @@ static dnode_t *
 dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db,
     uint64_t object, dnode_handle_t *dnh)
 {
-	dnode_t *dn = kmem_cache_alloc(dnode_cache, KM_SLEEP);
+	dnode_t *dn;
 
+	dn = kmem_cache_alloc(dnode_cache, KM_SLEEP);
 	ASSERT(!POINTER_IS_VALID(dn->dn_objset));
 	dn->dn_moved = 0;
 
@@ -443,13 +445,31 @@ dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db,
 	ASSERT(DMU_OT_IS_VALID(dn->dn_phys->dn_type));
 
 	mutex_enter(&os->os_lock);
-	list_insert_head(&os->os_dnodes, dn);
+	if (dnh->dnh_dnode != NULL) {
-	membar_producer();
+		/* Lost the allocation race. */
+		mutex_exit(&os->os_lock);
+		kmem_cache_free(dnode_cache, dn);
+		return (dnh->dnh_dnode);
+	}
+
 	/*
-	 * Everything else must be valid before assigning dn_objset makes the
+	 * Exclude special dnodes from os_dnodes so an empty os_dnodes
-	 * dnode eligible for dnode_move().
+	 * signifies that the special dnodes have no references from
+	 * their children (the entries in os_dnodes).  This allows
+	 * dnode_destroy() to easily determine if the last child has
+	 * been removed and then complete eviction of the objset.
+	 */
+	if (!DMU_OBJECT_IS_SPECIAL(object))
+		list_insert_head(&os->os_dnodes, dn);
+	membar_producer();
+
+	/*
+	 * Everything else must be valid before assigning dn_objset
+	 * makes the dnode eligible for dnode_move().
 	 */
 	dn->dn_objset = os;
+
+	dnh->dnh_dnode = dn;
 	mutex_exit(&os->os_lock);
 
 	arc_space_consume(sizeof (dnode_t), ARC_SPACE_OTHER);
@@ -463,12 +483,18 @@ static void
 dnode_destroy(dnode_t *dn)
 {
 	objset_t *os = dn->dn_objset;
+	boolean_t complete_os_eviction = B_FALSE;
 
 	ASSERT((dn->dn_id_flags & DN_ID_NEW_EXIST) == 0);
 
 	mutex_enter(&os->os_lock);
 	POINTER_INVALIDATE(&dn->dn_objset);
-	list_remove(&os->os_dnodes, dn);
+	if (!DMU_OBJECT_IS_SPECIAL(dn->dn_object)) {
+		list_remove(&os->os_dnodes, dn);
+		complete_os_eviction =
+		    list_is_empty(&os->os_dnodes) &&
+		    list_link_active(&os->os_evicting_node);
+	}
 	mutex_exit(&os->os_lock);
 
 	/* the dnode can no longer move, so we can release the handle */
@@ -503,6 +529,9 @@ dnode_destroy(dnode_t *dn)
 	dmu_zfetch_rele(&dn->dn_zfetch);
 	kmem_cache_free(dnode_cache, dn);
 	arc_space_return(sizeof (dnode_t), ARC_SPACE_OTHER);
+
+	if (complete_os_eviction)
+		dmu_objset_evict_done(os);
 }
 
 void
@@ -968,33 +997,32 @@ dnode_special_close(dnode_handle_t *dnh)
 	 */
 	while (refcount_count(&dn->dn_holds) > 0)
 		delay(1);
+	ASSERT(dn->dn_dbuf == NULL ||
+	    dmu_buf_get_user(&dn->dn_dbuf->db) == NULL);
 	zrl_add(&dnh->dnh_zrlock);
 	dnode_destroy(dn); /* implicit zrl_remove() */
 	zrl_destroy(&dnh->dnh_zrlock);
 	dnh->dnh_dnode = NULL;
 }
 
-dnode_t *
+void
 dnode_special_open(objset_t *os, dnode_phys_t *dnp, uint64_t object,
     dnode_handle_t *dnh)
 {
-	dnode_t *dn = dnode_create(os, dnp, NULL, object, dnh);
+	dnode_t *dn;
-	dnh->dnh_dnode = dn;
+
+	dn = dnode_create(os, dnp, NULL, object, dnh);
 	zrl_init(&dnh->dnh_zrlock);
 	DNODE_VERIFY(dn);
-	return (dn);
 }
 
 static void
-dnode_buf_pageout(dmu_buf_t *db, void *arg)
+dnode_buf_pageout(void *dbu)
 {
-	dnode_children_t *children_dnodes = arg;
+	dnode_children_t *children_dnodes = dbu;
 	int i;
-	int epb = db->db_size >> DNODE_SHIFT;
 
-	ASSERT(epb == children_dnodes->dnc_count);
+	for (i = 0; i < children_dnodes->dnc_count; i++) {
-
-	for (i = 0; i < epb; i++) {
 		dnode_handle_t *dnh = &children_dnodes->dnc_children[i];
 		dnode_t *dn;
 
@@ -1024,7 +1052,7 @@ dnode_buf_pageout(dmu_buf_t *db, void *arg)
 		dnh->dnh_dnode = NULL;
 	}
 	kmem_free(children_dnodes, sizeof (dnode_children_t) +
-	    epb * sizeof (dnode_handle_t));
+	    children_dnodes->dnc_count * sizeof (dnode_handle_t));
 }
 
 /*
@@ -1108,16 +1136,17 @@ dnode_hold_impl(objset_t *os, uint64_t object, int flag,
 	if (children_dnodes == NULL) {
 		int i;
 		dnode_children_t *winner;
-		children_dnodes = kmem_alloc(sizeof (dnode_children_t) +
+		children_dnodes = kmem_zalloc(sizeof (dnode_children_t) +
 		    epb * sizeof (dnode_handle_t), KM_SLEEP);
 		children_dnodes->dnc_count = epb;
 		dnh = &children_dnodes->dnc_children[0];
 		for (i = 0; i < epb; i++) {
 			zrl_init(&dnh[i].dnh_zrlock);
-			dnh[i].dnh_dnode = NULL;
 		}
-		if ((winner = dmu_buf_set_user(&db->db, children_dnodes,
+		dmu_buf_init_user(&children_dnodes->dnc_dbu,
-		    dnode_buf_pageout))) {
+		    dnode_buf_pageout, NULL);
+		winner = dmu_buf_set_user(&db->db, &children_dnodes->dnc_dbu);
+		if (winner != NULL) {
 
 			for (i = 0; i < epb; i++) {
 				zrl_destroy(&dnh[i].dnh_zrlock);
@@ -1132,17 +1161,11 @@ dnode_hold_impl(objset_t *os, uint64_t object, int flag,
 
 	dnh = &children_dnodes->dnc_children[idx];
 	zrl_add(&dnh->dnh_zrlock);
-	if ((dn = dnh->dnh_dnode) == NULL) {
+	dn = dnh->dnh_dnode;
+	if (dn == NULL) {
 		dnode_phys_t *phys = (dnode_phys_t *)db->db.db_data+idx;
-		dnode_t *winner;
 
 		dn = dnode_create(os, phys, db, object, dnh);
-		winner = atomic_cas_ptr(&dnh->dnh_dnode, NULL, dn);
-		if (winner != NULL) {
-			zrl_add(&dnh->dnh_zrlock);
-			dnode_destroy(dn); /* implicit zrl_remove() */
-			dn = winner;
-		}
 	}
 
 	mutex_enter(&dn->dn_mtx);
@@ -1156,10 +1179,10 @@ dnode_hold_impl(objset_t *os, uint64_t object, int flag,
 		dbuf_rele(db, FTAG);
 		return (type == DMU_OT_NONE ? ENOENT : EEXIST);
 	}
-	mutex_exit(&dn->dn_mtx);
-
 	if (refcount_add(&dn->dn_holds, tag) == 1)
 		dbuf_add_ref(db, dnh);
+	mutex_exit(&dn->dn_mtx);
+
 	/* Now we can rely on the hold to prevent the dnode from moving. */
 	zrl_remove(&dnh->dnh_zrlock);
 

module/zfs/dnode_sync.c

@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -402,53 +403,41 @@ dnode_sync_free_range(void *arg, uint64_t blkid, uint64_t nblks)
 void
 dnode_evict_dbufs(dnode_t *dn)
 {
-	int progress;
+	dmu_buf_impl_t *db_marker;
-	int pass = 0;
+	dmu_buf_impl_t *db, *db_next;
 
-	do {
+	db_marker = kmem_alloc(sizeof (dmu_buf_impl_t), KM_SLEEP);
-		dmu_buf_impl_t *db, *db_next;
+
-		int evicting = FALSE;
+	mutex_enter(&dn->dn_dbufs_mtx);
+	for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) {
 
-		progress = FALSE;
-		mutex_enter(&dn->dn_dbufs_mtx);
-		for (db = avl_first(&dn->dn_dbufs); db != NULL; db = db_next) {
-			db_next = AVL_NEXT(&dn->dn_dbufs, db);
 #ifdef	DEBUG
-			DB_DNODE_ENTER(db);
+		DB_DNODE_ENTER(db);
-			ASSERT3P(DB_DNODE(db), ==, dn);
+		ASSERT3P(DB_DNODE(db), ==, dn);
-			DB_DNODE_EXIT(db);
+		DB_DNODE_EXIT(db);
 #endif	/* DEBUG */
 
-			mutex_enter(&db->db_mtx);
+		mutex_enter(&db->db_mtx);
-			if (db->db_state == DB_EVICTING) {
+		if (db->db_state != DB_EVICTING &&
-				progress = TRUE;
+		    refcount_is_zero(&db->db_holds)) {
-				evicting = TRUE;
+			db_marker->db_level = db->db_level;
-				mutex_exit(&db->db_mtx);
+			db_marker->db_blkid = db->db_blkid;
-			} else if (refcount_is_zero(&db->db_holds)) {
+			db_marker->db_state = DB_SEARCH;
-				progress = TRUE;
+			avl_insert_here(&dn->dn_dbufs, db_marker, db,
-				dbuf_clear(db); /* exits db_mtx for us */
+			    AVL_BEFORE);
-			} else {
-				mutex_exit(&db->db_mtx);
-			}
 
+			dbuf_clear(db);
+
+			db_next = AVL_NEXT(&dn->dn_dbufs, db_marker);
+			avl_remove(&dn->dn_dbufs, db_marker);
+		} else {
+			mutex_exit(&db->db_mtx);
+			db_next = AVL_NEXT(&dn->dn_dbufs, db);
 		}
-		/*
+	}
-		 * NB: we need to drop dn_dbufs_mtx between passes so
+	mutex_exit(&dn->dn_dbufs_mtx);
-		 * that any DB_EVICTING dbufs can make progress.
-		 * Ideally, we would have some cv we could wait on, but
-		 * since we don't, just wait a bit to give the other
-		 * thread a chance to run.
-		 */
-		mutex_exit(&dn->dn_dbufs_mtx);
-		if (evicting)
-			delay(1);
-		pass++;
-		if ((pass % 100) == 0)
-			dprintf("Exceeded %d passes evicting dbufs\n", pass);
-	} while (progress);
 
-	if (pass >= 100)
+	kmem_free(db_marker, sizeof (dmu_buf_impl_t));
-		dprintf("Required %d passes to evict dbufs\n", pass);
 
 	dnode_evict_bonus(dn);
 }
@@ -513,7 +502,6 @@ dnode_sync_free(dnode_t *dn, dmu_tx_t *tx)
 	dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]);
 	dnode_evict_dbufs(dn);
 	ASSERT(avl_is_empty(&dn->dn_dbufs));
-	ASSERT3P(dn->dn_bonus, ==, NULL);
 
 	/*
 	 * XXX - It would be nice to assert this, but we may still

module/zfs/dsl_bookmark.c

@@ -120,7 +120,7 @@ dsl_bookmark_create_check_impl(dsl_dataset_t *snapds, const char *bookmark_name,
 	int error;
 	zfs_bookmark_phys_t bmark_phys;
 
-	if (!dsl_dataset_is_snapshot(snapds))
+	if (!snapds->ds_is_snapshot)
 		return (SET_ERROR(EINVAL));
 
 	error = dsl_bookmark_hold_ds(dp, bookmark_name,

module/zfs/dsl_dataset.c

@@ -23,6 +23,7 @@
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  * Copyright (c) 2014 RackTop Systems.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/dmu_objset.h>
@@ -62,7 +63,6 @@
 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
 
 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
-extern inline boolean_t dsl_dataset_is_snapshot(dsl_dataset_t *ds);
 
 /*
  * Figure out how much of this delta should be propogated to the dsl_dir
@@ -146,7 +146,7 @@ dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 	}
 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 
-	ASSERT(!dsl_dataset_is_snapshot(ds));
+	ASSERT(!ds->ds_is_snapshot);
 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
 
 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
@@ -244,14 +244,15 @@ dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 	return (B_TRUE);
 }
 
-/* ARGSUSED */
 static void
-dsl_dataset_evict(dmu_buf_t *db, void *dsv)
+dsl_dataset_evict(void *dbu)
 {
-	dsl_dataset_t *ds = dsv;
+	dsl_dataset_t *ds = dbu;
 
 	ASSERT(ds->ds_owner == NULL);
 
+	ds->ds_dbuf = NULL;
+
 	unique_remove(ds->ds_fsid_guid);
 
 	if (ds->ds_objset != NULL)
@@ -263,10 +264,10 @@ dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 	}
 
 	bplist_destroy(&ds->ds_pending_deadlist);
-	if (dsl_dataset_phys(ds)->ds_deadlist_obj != 0)
+	if (ds->ds_deadlist.dl_os != NULL)
 		dsl_deadlist_close(&ds->ds_deadlist);
 	if (ds->ds_dir)
-		dsl_dir_rele(ds->ds_dir, ds);
+		dsl_dir_async_rele(ds->ds_dir, ds);
 
 	ASSERT(!list_link_active(&ds->ds_synced_link));
 
@@ -380,6 +381,7 @@ dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 		ds->ds_dbuf = dbuf;
 		ds->ds_object = dsobj;
+		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
 		list_link_init(&ds->ds_synced_link);
 
 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
@@ -411,7 +413,7 @@ dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 			return (err);
 		}
 
-		if (!dsl_dataset_is_snapshot(ds)) {
+		if (!ds->ds_is_snapshot) {
 			ds->ds_snapname[0] = '\0';
 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 				err = dsl_dataset_hold_obj(dp,
@@ -438,7 +440,7 @@ dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 			}
 		}
 
-		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
+		if (err == 0 && !ds->ds_is_snapshot) {
 			err = dsl_prop_get_int_ds(ds,
 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 			    &ds->ds_reserved);
@@ -451,8 +453,11 @@ dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 			ds->ds_reserved = ds->ds_quota = 0;
 		}
 
-		if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
+		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
-		    dsl_dataset_evict)) != NULL) {
+		if (err == 0)
+			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
+
+		if (err != 0 || winner != NULL) {
 			bplist_destroy(&ds->ds_pending_deadlist);
 			dsl_deadlist_close(&ds->ds_deadlist);
 			if (ds->ds_prev)
@@ -829,7 +834,7 @@ dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
 	uint64_t mrs_used;
 	uint64_t dlused, dlcomp, dluncomp;
 
-	ASSERT(!dsl_dataset_is_snapshot(ds));
+	ASSERT(!ds->ds_is_snapshot);
 
 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
@@ -1579,7 +1584,7 @@ dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
 
-	if (dsl_dataset_is_snapshot(ds)) {
+	if (ds->ds_is_snapshot) {
 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
 		    dsl_dataset_phys(ds)->ds_unique_bytes);
@@ -1643,7 +1648,7 @@ dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
 	stat->dds_origin[0] = '\0';
-	if (dsl_dataset_is_snapshot(ds)) {
+	if (ds->ds_is_snapshot) {
 		stat->dds_is_snapshot = B_TRUE;
 		stat->dds_num_clones =
 		    dsl_dataset_phys(ds)->ds_num_children - 1;
@@ -1920,7 +1925,7 @@ dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
 		return (error);
 
 	/* must not be a snapshot */
-	if (dsl_dataset_is_snapshot(ds)) {
+	if (ds->ds_is_snapshot) {
 		dsl_dataset_rele(ds, FTAG);
 		return (SET_ERROR(EINVAL));
 	}
@@ -2491,7 +2496,7 @@ promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
 		return (error);
 	dd = ddpa->ddpa_clone->ds_dir;
 
-	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
+	if (ddpa->ddpa_clone->ds_is_snapshot ||
 	    !dsl_dir_is_clone(dd)) {
 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
 		return (SET_ERROR(EINVAL));
@@ -2582,8 +2587,8 @@ dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
 	int64_t unused_refres_delta;
 
 	/* they should both be heads */
-	if (dsl_dataset_is_snapshot(clone) ||
+	if (clone->ds_is_snapshot ||
-	    dsl_dataset_is_snapshot(origin_head))
+	    origin_head->ds_is_snapshot)
 		return (SET_ERROR(EINVAL));
 
 	/* if we are not forcing, the branch point should be just before them */
@@ -2862,7 +2867,7 @@ dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
 	if (error != 0)
 		return (error);
 
-	if (dsl_dataset_is_snapshot(ds)) {
+	if (ds->ds_is_snapshot) {
 		dsl_dataset_rele(ds, FTAG);
 		return (SET_ERROR(EINVAL));
 	}
@@ -2945,7 +2950,7 @@ dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
 	if (error != 0)
 		return (error);
 
-	if (dsl_dataset_is_snapshot(ds)) {
+	if (ds->ds_is_snapshot) {
 		dsl_dataset_rele(ds, FTAG);
 		return (SET_ERROR(EINVAL));
 	}
@@ -3158,8 +3163,8 @@ dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
 	uint64_t snapobj;
 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
 
-	ASSERT(dsl_dataset_is_snapshot(firstsnap));
+	ASSERT(firstsnap->ds_is_snapshot);
-	ASSERT(dsl_dataset_is_snapshot(lastsnap));
+	ASSERT(lastsnap->ds_is_snapshot);
 
 	/*
 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
@@ -3214,12 +3219,12 @@ dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
 	dsl_dataset_t *origin;
 
 	ASSERT(dsl_pool_config_held(dp));
-	ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
+	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
 
 	if (earlier_txg == 0)
 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
 
-	if (dsl_dataset_is_snapshot(later) &&
+	if (later->ds_is_snapshot &&
 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
 		return (B_FALSE);
 

module/zfs/dsl_deadlist.c

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/dsl_dataset.h>
@@ -121,6 +122,8 @@ dsl_deadlist_close(dsl_deadlist_t *dl)
 	void *cookie = NULL;
 	dsl_deadlist_entry_t *dle;
 
+	dl->dl_os = NULL;
+
 	if (dl->dl_oldfmt) {
 		dl->dl_oldfmt = B_FALSE;
 		bpobj_close(&dl->dl_bpobj);

module/zfs/dsl_deleg.c

@@ -570,7 +570,7 @@ dsl_deleg_access_impl(dsl_dataset_t *ds, const char *perm, cred_t *cr)
 	    SPA_VERSION_DELEGATED_PERMS)
 		return (SET_ERROR(EPERM));
 
-	if (dsl_dataset_is_snapshot(ds)) {
+	if (ds->ds_is_snapshot) {
 		/*
 		 * Snapshots are treated as descendents only,
 		 * local permissions do not apply.

module/zfs/dsl_destroy.c

@@ -51,7 +51,7 @@ typedef struct dmu_snapshots_destroy_arg {
 int
 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
 {
-	if (!dsl_dataset_is_snapshot(ds))
+	if (!ds->ds_is_snapshot)
 		return (SET_ERROR(EINVAL));
 
 	if (dsl_dataset_long_held(ds))
@@ -356,7 +356,7 @@ dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
 	dsl_dataset_remove_clones_key(ds,
 	    dsl_dataset_phys(ds)->ds_creation_txg, tx);
 
-	if (dsl_dataset_is_snapshot(ds_next)) {
+	if (ds_next->ds_is_snapshot) {
 		dsl_dataset_t *ds_nextnext;
 
 		/*
@@ -605,8 +605,8 @@ dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
 	uint64_t count;
 	objset_t *mos;
 
-	ASSERT(!dsl_dataset_is_snapshot(ds));
+	ASSERT(!ds->ds_is_snapshot);
-	if (dsl_dataset_is_snapshot(ds))
+	if (ds->ds_is_snapshot)
 		return (SET_ERROR(EINVAL));
 
 	if (refcount_count(&ds->ds_longholds) != expected_holds)

module/zfs/dsl_dir.c

@@ -23,6 +23,7 @@
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2013 Martin Matuska. All rights reserved.
  * Copyright (c) 2014 Joyent, Inc. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/dmu.h>
@@ -126,14 +127,15 @@ extern inline dsl_dir_phys_t *dsl_dir_phys(dsl_dir_t *dd);
 
 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
 
-/* ARGSUSED */
 static void
-dsl_dir_evict(dmu_buf_t *db, void *arg)
+dsl_dir_evict(void *dbu)
 {
-	dsl_dir_t *dd = arg;
+	dsl_dir_t *dd = dbu;
 	int t;
 	ASSERTV(dsl_pool_t *dp = dd->dd_pool);
 
+	dd->dd_dbuf = NULL;
+
 	for (t = 0; t < TXG_SIZE; t++) {
 		ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t));
 		ASSERT(dd->dd_tempreserved[t] == 0);
@@ -141,9 +143,9 @@ dsl_dir_evict(dmu_buf_t *db, void *arg)
 	}
 
 	if (dd->dd_parent)
-		dsl_dir_rele(dd->dd_parent, dd);
+		dsl_dir_async_rele(dd->dd_parent, dd);
 
-	spa_close(dd->dd_pool->dp_spa, dd);
+	spa_async_close(dd->dd_pool->dp_spa, dd);
 
 	/*
 	 * The props callback list should have been cleaned up by
@@ -239,8 +241,9 @@ dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
 			dmu_buf_rele(origin_bonus, FTAG);
 		}
 
-		winner = dmu_buf_set_user_ie(dbuf, dd, dsl_dir_evict);
+		dmu_buf_init_user(&dd->dd_dbu, dsl_dir_evict, &dd->dd_dbuf);
-		if (winner) {
+		winner = dmu_buf_set_user_ie(dbuf, &dd->dd_dbu);
+		if (winner != NULL) {
 			if (dd->dd_parent)
 				dsl_dir_rele(dd->dd_parent, dd);
 			mutex_destroy(&dd->dd_lock);
@@ -284,6 +287,21 @@ dsl_dir_rele(dsl_dir_t *dd, void *tag)
 	dmu_buf_rele(dd->dd_dbuf, tag);
 }
 
+/*
+ * Remove a reference to the given dsl dir that is being asynchronously
+ * released.  Async releases occur from a taskq performing eviction of
+ * dsl datasets and dirs.  This process is identical to a normal release
+ * with the exception of using the async API for releasing the reference on
+ * the spa.
+ */
+void
+dsl_dir_async_rele(dsl_dir_t *dd, void *tag)
+{
+	dprintf_dd(dd, "%s\n", "");
+	spa_async_close(dd->dd_pool->dp_spa, tag);
+	dmu_buf_rele(dd->dd_dbuf, tag);
+}
+
 /* buf must be long enough (MAXNAMELEN + strlen(MOS_DIR_NAME) + 1 should do) */
 void
 dsl_dir_name(dsl_dir_t *dd, char *buf)
@@ -417,7 +435,7 @@ dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
 	}
 
 	while (next != NULL) {
-		dsl_dir_t *child_ds;
+		dsl_dir_t *child_dd;
 		err = getcomponent(next, buf, &nextnext);
 		if (err != 0)
 			break;
@@ -436,11 +454,11 @@ dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
 			break;
 		}
 
-		err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
+		err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_dd);
 		if (err != 0)
 			break;
 		dsl_dir_rele(dd, tag);
-		dd = child_ds;
+		dd = child_dd;
 		next = nextnext;
 	}
 

module/zfs/dsl_pool.c

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright (c) 2013 Steven Hartland. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/dsl_pool.h>
@@ -319,6 +320,8 @@ dsl_pool_close(dsl_pool_t *dp)
 	arc_flush(dp->dp_spa);
 	txg_fini(dp);
 	dsl_scan_fini(dp);
+	dmu_buf_user_evict_wait();
+
 	rrw_destroy(&dp->dp_config_rwlock);
 	mutex_destroy(&dp->dp_lock);
 	taskq_destroy(dp->dp_iput_taskq);

module/zfs/dsl_prop.c

@@ -163,19 +163,17 @@ dsl_prop_get_ds(dsl_dataset_t *ds, const char *propname,
 {
 	zfs_prop_t prop = zfs_name_to_prop(propname);
 	boolean_t inheritable;
-	boolean_t snapshot;
 	uint64_t zapobj;
 
 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 	inheritable = (prop == ZPROP_INVAL || zfs_prop_inheritable(prop));
-	snapshot = dsl_dataset_is_snapshot(ds);
 	zapobj = dsl_dataset_phys(ds)->ds_props_obj;
 
 	if (zapobj != 0) {
 		objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 		int err;
 
-		ASSERT(snapshot);
+		ASSERT(ds->ds_is_snapshot);
 
 		/* Check for a local value. */
 		err = zap_lookup(mos, zapobj, propname, intsz, numints, buf);
@@ -215,7 +213,7 @@ dsl_prop_get_ds(dsl_dataset_t *ds, const char *propname,
 	}
 
 	return (dsl_prop_get_dd(ds->ds_dir, propname,
-	    intsz, numints, buf, setpoint, snapshot));
+	    intsz, numints, buf, setpoint, ds->ds_is_snapshot));
 }
 
 /*
@@ -545,7 +543,7 @@ dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
 
 	isint = (dodefault(propname, 8, 1, &intval) == 0);
 
-	if (dsl_dataset_is_snapshot(ds)) {
+	if (ds->ds_is_snapshot) {
 		ASSERT(version >= SPA_VERSION_SNAP_PROPS);
 		if (dsl_dataset_phys(ds)->ds_props_obj == 0) {
 			dmu_buf_will_dirty(ds->ds_dbuf, tx);
@@ -642,7 +640,7 @@ dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
 	if (isint) {
 		VERIFY0(dsl_prop_get_int_ds(ds, propname, &intval));
 
-		if (dsl_dataset_is_snapshot(ds)) {
+		if (ds->ds_is_snapshot) {
 			dsl_prop_cb_record_t *cbr;
 			/*
 			 * It's a snapshot; nothing can inherit this
@@ -760,7 +758,7 @@ dsl_props_set_check(void *arg, dmu_tx_t *tx)
 		}
 	}
 
-	if (dsl_dataset_is_snapshot(ds) && version < SPA_VERSION_SNAP_PROPS) {
+	if (ds->ds_is_snapshot && version < SPA_VERSION_SNAP_PROPS) {
 		dsl_dataset_rele(ds, FTAG);
 		return (SET_ERROR(ENOTSUP));
 	}
@@ -983,7 +981,7 @@ dsl_prop_get_all_ds(dsl_dataset_t *ds, nvlist_t **nvp,
 
 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 
-	if (dsl_dataset_is_snapshot(ds))
+	if (ds->ds_is_snapshot)
 		flags |= DSL_PROP_GET_SNAPSHOT;
 
 	ASSERT(dsl_pool_config_held(dp));

module/zfs/dsl_scan.c

@@ -415,7 +415,7 @@ static uint64_t
 dsl_scan_ds_maxtxg(dsl_dataset_t *ds)
 {
 	uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg;
-	if (dsl_dataset_is_snapshot(ds))
+	if (ds->ds_is_snapshot)
 		return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg));
 	return (smt);
 }
@@ -846,7 +846,7 @@ dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx)
 		return;
 
 	if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
-		if (dsl_dataset_is_snapshot(ds)) {
+		if (ds->ds_is_snapshot) {
 			/* Note, scn_cur_{min,max}_txg stays the same. */
 			scn->scn_phys.scn_bookmark.zb_objset =
 			    dsl_dataset_phys(ds)->ds_next_snap_obj;
@@ -868,7 +868,7 @@ dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx)
 		ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
 		    scn->scn_phys.scn_queue_obj, ds->ds_object, tx));
-		if (dsl_dataset_is_snapshot(ds)) {
+		if (ds->ds_is_snapshot) {
 			/*
 			 * We keep the same mintxg; it could be >
 			 * ds_creation_txg if the previous snapshot was
@@ -1053,7 +1053,7 @@ dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx)
 	 * ZIL here, rather than in scan_recurse(), because the regular
 	 * snapshot block-sharing rules don't apply to it.
 	 */
-	if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !dsl_dataset_is_snapshot(ds))
+	if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !ds->ds_is_snapshot)
 		dsl_scan_zil(dp, &os->os_zil_header);
 
 	/*

module/zfs/dsl_userhold.c

@@ -354,7 +354,7 @@ dsl_dataset_user_release_check_one(dsl_dataset_user_release_arg_t *ddura,
 	objset_t *mos;
 	int numholds;
 
-	if (!dsl_dataset_is_snapshot(ds))
+	if (!ds->ds_is_snapshot)
 		return (SET_ERROR(EINVAL));
 
 	if (nvlist_empty(holds))

module/zfs/sa.c

@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -209,6 +210,7 @@ sa_cache_constructor(void *buf, void *unused, int kmflag)
 {
 	sa_handle_t *hdl = buf;
 
+	hdl->sa_dbu.dbu_evict_func = NULL;
 	hdl->sa_bonus_tab = NULL;
 	hdl->sa_spill_tab = NULL;
 	hdl->sa_os = NULL;
@@ -224,6 +226,7 @@ static void
 sa_cache_destructor(void *buf, void *unused)
 {
 	sa_handle_t *hdl = buf;
+	hdl->sa_dbu.dbu_evict_func = NULL;
 	mutex_destroy(&hdl->sa_lock);
 }
 
@@ -1306,10 +1309,10 @@ sa_build_index(sa_handle_t *hdl, sa_buf_type_t buftype)
 }
 
 /*ARGSUSED*/
-void
+static void
-sa_evict(dmu_buf_t *db, void *sap)
+sa_evict(void *dbu)
 {
-	panic("evicting sa dbuf %p\n", (void *)db);
+	panic("evicting sa dbuf\n");
 }
 
 static void
@@ -1361,9 +1364,10 @@ sa_spill_rele(sa_handle_t *hdl)
 void
 sa_handle_destroy(sa_handle_t *hdl)
 {
+	dmu_buf_t *db = hdl->sa_bonus;
+
 	mutex_enter(&hdl->sa_lock);
-	(void) dmu_buf_update_user((dmu_buf_t *)hdl->sa_bonus, hdl,
+	(void) dmu_buf_remove_user(db, &hdl->sa_dbu);
-	    NULL, NULL);
 
 	if (hdl->sa_bonus_tab) {
 		sa_idx_tab_rele(hdl->sa_os, hdl->sa_bonus_tab);
@@ -1388,7 +1392,7 @@ sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
     sa_handle_type_t hdl_type, sa_handle_t **handlepp)
 {
 	int error = 0;
-	sa_handle_t *handle;
+	sa_handle_t *handle = NULL;
 #ifdef ZFS_DEBUG
 	dmu_object_info_t doi;
 
@@ -1399,9 +1403,12 @@ sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
 	/* find handle, if it exists */
 	/* if one doesn't exist then create a new one, and initialize it */
 
-	handle = (hdl_type == SA_HDL_SHARED) ? dmu_buf_get_user(db) : NULL;
+	if (hdl_type == SA_HDL_SHARED)
+		handle = dmu_buf_get_user(db);
+
 	if (handle == NULL) {
-		sa_handle_t *newhandle;
+		sa_handle_t *winner = NULL;
+
 		handle = kmem_cache_alloc(sa_cache, KM_SLEEP);
 		handle->sa_userp = userp;
 		handle->sa_bonus = db;
@@ -1409,12 +1416,15 @@ sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
 		handle->sa_spill = NULL;
 
 		error = sa_build_index(handle, SA_BONUS);
-		newhandle = (hdl_type == SA_HDL_SHARED) ?
-		    dmu_buf_set_user_ie(db, handle, sa_evict) : NULL;
 
-		if (newhandle != NULL) {
+		if (hdl_type == SA_HDL_SHARED) {
+			dmu_buf_init_user(&handle->sa_dbu, sa_evict, NULL);
+			winner = dmu_buf_set_user_ie(db, &handle->sa_dbu);
+		}
+
+		if (winner != NULL) {
 			kmem_cache_free(sa_cache, handle);
-			handle = newhandle;
+			handle = winner;
 		}
 	}
 	*handlepp = handle;
@@ -1946,14 +1956,6 @@ sa_object_size(sa_handle_t *hdl, uint32_t *blksize, u_longlong_t *nblocks)
 	    blksize, nblocks);
 }
 
-void
-sa_update_user(sa_handle_t *newhdl, sa_handle_t *oldhdl)
-{
-	(void) dmu_buf_update_user((dmu_buf_t *)newhdl->sa_bonus,
-	    oldhdl, newhdl, sa_evict);
-	oldhdl->sa_bonus = NULL;
-}
-
 void
 sa_set_userp(sa_handle_t *hdl, void *ptr)
 {
@@ -2052,7 +2054,6 @@ EXPORT_SYMBOL(sa_size);
 EXPORT_SYMBOL(sa_update_from_cb);
 EXPORT_SYMBOL(sa_object_info);
 EXPORT_SYMBOL(sa_object_size);
-EXPORT_SYMBOL(sa_update_user);
 EXPORT_SYMBOL(sa_get_userdata);
 EXPORT_SYMBOL(sa_set_userp);
 EXPORT_SYMBOL(sa_get_db);

module/zfs/spa.c

@@ -23,6 +23,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2013, 2014, Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 /*
@@ -1095,6 +1096,8 @@ spa_activate(spa_t *spa, int mode)
 
 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
 	    offsetof(vdev_t, vdev_config_dirty_node));
+	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
+	    offsetof(objset_t, os_evicting_node));
 	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
 	    offsetof(vdev_t, vdev_state_dirty_node));
 
@@ -1123,9 +1126,12 @@ spa_deactivate(spa_t *spa)
 	ASSERT(spa->spa_async_zio_root == NULL);
 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
 
+	spa_evicting_os_wait(spa);
+
 	txg_list_destroy(&spa->spa_vdev_txg_list);
 
 	list_destroy(&spa->spa_config_dirty_list);
+	list_destroy(&spa->spa_evicting_os_list);
 	list_destroy(&spa->spa_state_dirty_list);
 
 	taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
@@ -2136,6 +2142,11 @@ spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
 		    mosconfig, &ereport);
 	}
 
+	/*
+	 * Don't count references from objsets that are already closed
+	 * and are making their way through the eviction process.
+	 */
+	spa_evicting_os_wait(spa);
 	spa->spa_minref = refcount_count(&spa->spa_refcount);
 	if (error) {
 		if (error != EEXIST) {
@@ -3775,6 +3786,11 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
 	spa_history_log_version(spa, "create");
 
+	/*
+	 * Don't count references from objsets that are already closed
+	 * and are making their way through the eviction process.
+	 */
+	spa_evicting_os_wait(spa);
 	spa->spa_minref = refcount_count(&spa->spa_refcount);
 
 	mutex_exit(&spa_namespace_lock);
@@ -4314,8 +4330,10 @@ spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
 	 * modify its state.  Objsets may be open only because they're dirty,
 	 * so we have to force it to sync before checking spa_refcnt.
 	 */
-	if (spa->spa_sync_on)
+	if (spa->spa_sync_on) {
 		txg_wait_synced(spa->spa_dsl_pool, 0);
+		spa_evicting_os_wait(spa);
+	}
 
 	/*
 	 * A pool cannot be exported or destroyed if there are active

module/zfs/spa_misc.c

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -524,6 +525,7 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
+	mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
@@ -533,6 +535,7 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 	mutex_init(&spa->spa_feat_stats_lock, NULL, MUTEX_DEFAULT, NULL);
 
 	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
+	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
 	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
 	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
 	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
@@ -619,6 +622,7 @@ spa_remove(spa_t *spa)
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
+	ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
 
 	nvlist_free(spa->spa_config_splitting);
 
@@ -651,6 +655,7 @@ spa_remove(spa_t *spa)
 		bplist_destroy(&spa->spa_free_bplist[t]);
 
 	cv_destroy(&spa->spa_async_cv);
+	cv_destroy(&spa->spa_evicting_os_cv);
 	cv_destroy(&spa->spa_proc_cv);
 	cv_destroy(&spa->spa_scrub_io_cv);
 	cv_destroy(&spa->spa_suspend_cv);
@@ -658,6 +663,7 @@ spa_remove(spa_t *spa)
 	mutex_destroy(&spa->spa_async_lock);
 	mutex_destroy(&spa->spa_errlist_lock);
 	mutex_destroy(&spa->spa_errlog_lock);
+	mutex_destroy(&spa->spa_evicting_os_lock);
 	mutex_destroy(&spa->spa_history_lock);
 	mutex_destroy(&spa->spa_proc_lock);
 	mutex_destroy(&spa->spa_props_lock);
@@ -714,6 +720,20 @@ spa_close(spa_t *spa, void *tag)
 	(void) refcount_remove(&spa->spa_refcount, tag);
 }
 
+/*
+ * Remove a reference to the given spa_t held by a dsl dir that is
+ * being asynchronously released.  Async releases occur from a taskq
+ * performing eviction of dsl datasets and dirs.  The namespace lock
+ * isn't held and the hold by the object being evicted may contribute to
+ * spa_minref (e.g. dataset or directory released during pool export),
+ * so the asserts in spa_close() do not apply.
+ */
+void
+spa_async_close(spa_t *spa, void *tag)
+{
+	(void) refcount_remove(&spa->spa_refcount, tag);
+}
+
 /*
  * Check to see if the spa refcount is zero.  Must be called with
  * spa_namespace_lock held.  We really compare against spa_minref, which is the
@@ -1597,6 +1617,34 @@ spa_log_class(spa_t *spa)
 	return (spa->spa_log_class);
 }
 
+void
+spa_evicting_os_register(spa_t *spa, objset_t *os)
+{
+	mutex_enter(&spa->spa_evicting_os_lock);
+	list_insert_head(&spa->spa_evicting_os_list, os);
+	mutex_exit(&spa->spa_evicting_os_lock);
+}
+
+void
+spa_evicting_os_deregister(spa_t *spa, objset_t *os)
+{
+	mutex_enter(&spa->spa_evicting_os_lock);
+	list_remove(&spa->spa_evicting_os_list, os);
+	cv_broadcast(&spa->spa_evicting_os_cv);
+	mutex_exit(&spa->spa_evicting_os_lock);
+}
+
+void
+spa_evicting_os_wait(spa_t *spa)
+{
+	mutex_enter(&spa->spa_evicting_os_lock);
+	while (!list_is_empty(&spa->spa_evicting_os_list))
+		cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
+	mutex_exit(&spa->spa_evicting_os_lock);
+
+	dmu_buf_user_evict_wait();
+}
+
 int
 spa_max_replication(spa_t *spa)
 {

module/zfs/zap.c

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 /*
@@ -52,7 +53,6 @@ int fzap_default_block_shift = 14; /* 16k blocksize */
 
 extern inline zap_phys_t *zap_f_phys(zap_t *zap);
 
-static void zap_leaf_pageout(dmu_buf_t *db, void *vl);
 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
 
 void
@@ -81,7 +81,7 @@ fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 	zap->zap_ismicro = FALSE;
 
-	(void) dmu_buf_update_user(zap->zap_dbuf, zap, zap, zap_evict);
+	zap->zap_dbu.dbu_evict_func = zap_evict;
 
 	mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
 	zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
@@ -387,11 +387,20 @@ zap_allocate_blocks(zap_t *zap, int nblocks)
 	return (newblk);
 }
 
+static void
+zap_leaf_pageout(void *dbu)
+{
+	zap_leaf_t *l = dbu;
+
+	rw_destroy(&l->l_rwlock);
+	kmem_free(l, sizeof (zap_leaf_t));
+}
+
 static zap_leaf_t *
 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
 {
 	void *winner;
-	zap_leaf_t *l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
+	zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
 
 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
 
@@ -403,7 +412,8 @@ zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
 	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
 	    l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
 	    DMU_READ_NO_PREFETCH));
-	winner = dmu_buf_set_user(l->l_dbuf, l, zap_leaf_pageout);
+	dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
+	winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu);
 	ASSERT(winner == NULL);
 	dmu_buf_will_dirty(l->l_dbuf, tx);
 
@@ -435,16 +445,6 @@ zap_put_leaf(zap_leaf_t *l)
 	dmu_buf_rele(l->l_dbuf, NULL);
 }
 
-_NOTE(ARGSUSED(0))
-static void
-zap_leaf_pageout(dmu_buf_t *db, void *vl)
-{
-	zap_leaf_t *l = vl;
-
-	rw_destroy(&l->l_rwlock);
-	kmem_free(l, sizeof (zap_leaf_t));
-}
-
 static zap_leaf_t *
 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
 {
@@ -452,19 +452,20 @@ zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
 
 	ASSERT(blkid != 0);
 
-	l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
+	l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
 	rw_init(&l->l_rwlock, NULL, RW_DEFAULT, NULL);
 	rw_enter(&l->l_rwlock, RW_WRITER);
 	l->l_blkid = blkid;
 	l->l_bs = highbit64(db->db_size) - 1;
 	l->l_dbuf = db;
 
-	winner = dmu_buf_set_user(db, l, zap_leaf_pageout);
+	dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
+	winner = dmu_buf_set_user(db, &l->l_dbu);
 
 	rw_exit(&l->l_rwlock);
 	if (winner != NULL) {
 		/* someone else set it first */
-		zap_leaf_pageout(NULL, l);
+		zap_leaf_pageout(&l->l_dbu);
 		l = winner;
 	}
 

module/zfs/zap_micro.c

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
 #include <sys/zio.h>
@@ -386,7 +387,8 @@ mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
 	 * it, because zap_lockdir() checks zap_ismicro without the lock
 	 * held.
 	 */
-	winner = dmu_buf_set_user(db, zap, zap_evict);
+	dmu_buf_init_user(&zap->zap_dbu, zap_evict, &zap->zap_dbuf);
+	winner = dmu_buf_set_user(db, &zap->zap_dbu);
 
 	if (winner != NULL) {
 		rw_exit(&zap->zap_rwlock);
@@ -675,11 +677,10 @@ zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
 	return (dmu_object_free(os, zapobj, tx));
 }
 
-_NOTE(ARGSUSED(0))
 void
-zap_evict(dmu_buf_t *db, void *vzap)
+zap_evict(void *dbu)
 {
-	zap_t *zap = vzap;
+	zap_t *zap = dbu;
 
 	rw_destroy(&zap->zap_rwlock);
 

module/zfs/zfs_sa.c

@@ -22,8 +22,7 @@
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  */
 
-#include <sys/types.h>
+#include <sys/zfs_context.h>
-#include <sys/param.h>
 #include <sys/vnode.h>
 #include <sys/sa.h>
 #include <sys/zfs_acl.h>

module/zfs/zil.c

@@ -497,7 +497,7 @@ zilog_dirty(zilog_t *zilog, uint64_t txg)
 	dsl_pool_t *dp = zilog->zl_dmu_pool;
 	dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os);
 
-	if (dsl_dataset_is_snapshot(ds))
+	if (ds->ds_is_snapshot)
 		panic("dirtying snapshot!");
 
 	if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) {