Merge commit 'refs/top-bases/linux-zfs-branch' into linux-zfs-branch

cmd/zdb/zdb.c

@@ -52,7 +52,6 @@
 #include <sys/zfs_fuid.h>
 #include <sys/arc.h>
 #undef ZFS_MAXNAMELEN
-#undef verify
 #include <libzfs.h>
 
 const char cmdname[] = "zdb";

cmd/zinject/translate.c

@@ -25,8 +25,6 @@
 
 #include <libzfs.h>
 
-#undef verify	/* both libzfs.h and zfs_context.h want to define this */
-
 #include <sys/zfs_context.h>
 
 #include <errno.h>

cmd/ztest/ztest.c

@@ -138,7 +138,7 @@ typedef struct ztest_args {
 	spa_t		*za_spa;
 	objset_t	*za_os;
 	zilog_t		*za_zilog;
-	thread_t	za_thread;
+	pthread_t	za_thread;
 	uint64_t	za_instance;
 	uint64_t	za_random;
 	uint64_t	za_diroff;
@@ -164,6 +164,7 @@ typedef void ztest_func_t(ztest_args_t *);
 ztest_func_t ztest_dmu_read_write;
 ztest_func_t ztest_dmu_write_parallel;
 ztest_func_t ztest_dmu_object_alloc_free;
+ztest_func_t ztest_dmu_commit_callbacks;
 ztest_func_t ztest_zap;
 ztest_func_t ztest_zap_parallel;
 ztest_func_t ztest_traverse;
@@ -198,6 +199,7 @@ ztest_info_t ztest_info[] = {
 	{ ztest_dmu_read_write,			1,	&zopt_always	},
 	{ ztest_dmu_write_parallel,		30,	&zopt_always	},
 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
+	{ ztest_dmu_commit_callbacks,		10,	&zopt_always	},
 	{ ztest_zap,				30,	&zopt_always	},
 	{ ztest_zap_parallel,			100,	&zopt_always	},
 	{ ztest_dsl_prop_get_set,		1,	&zopt_sometimes	},
@@ -217,12 +219,22 @@ ztest_info_t ztest_info[] = {
 
 #define	ZTEST_SYNC_LOCKS	16
 
+/*
+ * The following struct is used to hold a list of uncalled commit callbacks.
+ *
+ * The callbacks are ordered by txg number.
+ */
+typedef struct ztest_cb_list {
+	pthread_mutex_t		zcl_callbacks_lock;
+	list_t			zcl_callbacks;
+} ztest_cb_list_t;
+
 /*
  * Stuff we need to share writably between parent and child.
  */
 typedef struct ztest_shared {
-	mutex_t		zs_vdev_lock;
+	pthread_mutex_t		zs_vdev_lock;
-	rwlock_t	zs_name_lock;
+	pthread_rwlock_t	zs_name_lock;
 	uint64_t		zs_vdev_primaries;
 	uint64_t		zs_vdev_aux;
 	uint64_t		zs_enospc_count;
@@ -231,8 +243,9 @@ typedef struct ztest_shared {
 	uint64_t		zs_alloc;
 	uint64_t		zs_space;
 	ztest_info_t		zs_info[ZTEST_FUNCS];
-	mutex_t		zs_sync_lock[ZTEST_SYNC_LOCKS];
+	pthread_mutex_t		zs_sync_lock[ZTEST_SYNC_LOCKS];
 	uint64_t		zs_seq[ZTEST_SYNC_LOCKS];
+	ztest_cb_list_t		zs_cb_list;
 } ztest_shared_t;
 
 static char ztest_dev_template[] = "%s/%s.%llua";
@@ -804,7 +817,7 @@ ztest_spa_create_destroy(ztest_args_t *za)
 	 * Attempt to create an existing pool.  It shouldn't matter
 	 * what's in the nvroot; we should fail with EEXIST.
 	 */
-	(void) rw_rdlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_rdlock(&ztest_shared->zs_name_lock);
 	nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
 	error = spa_create(za->za_pool, nvroot, NULL, NULL, NULL);
 	nvlist_free(nvroot);
@@ -820,7 +833,7 @@ ztest_spa_create_destroy(ztest_args_t *za)
 		fatal(0, "spa_destroy() = %d", error);
 
 	spa_close(spa, FTAG);
-	(void) rw_unlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 }
 
 static vdev_t *
@@ -851,7 +864,7 @@ ztest_vdev_add_remove(ztest_args_t *za)
 	nvlist_t *nvroot;
 	int error;
 
-	(void) mutex_lock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_lock(&ztest_shared->zs_vdev_lock);
 
 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 
@@ -869,7 +882,7 @@ ztest_vdev_add_remove(ztest_args_t *za)
 	error = spa_vdev_add(spa, nvroot);
 	nvlist_free(nvroot);
 
-	(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_unlock(&ztest_shared->zs_vdev_lock);
 
 	if (error == ENOSPC)
 		ztest_record_enospc("spa_vdev_add");
@@ -898,7 +911,7 @@ ztest_vdev_aux_add_remove(ztest_args_t *za)
 		aux = ZPOOL_CONFIG_L2CACHE;
 	}
 
-	(void) mutex_lock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_lock(&ztest_shared->zs_vdev_lock);
 
 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 
@@ -954,7 +967,7 @@ ztest_vdev_aux_add_remove(ztest_args_t *za)
 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
 	}
 
-	(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_unlock(&ztest_shared->zs_vdev_lock);
 }
 
 /*
@@ -980,7 +993,7 @@ ztest_vdev_attach_detach(ztest_args_t *za)
 	int oldvd_is_log;
 	int error, expected_error;
 
-	(void) mutex_lock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_lock(&ztest_shared->zs_vdev_lock);
 
 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 
@@ -1040,7 +1053,7 @@ ztest_vdev_attach_detach(ztest_args_t *za)
 		if (error != 0 && error != ENODEV && error != EBUSY &&
 		    error != ENOTSUP)
 			fatal(0, "detach (%s) returned %d", oldpath, error);
-		(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
+		(void) pthread_mutex_unlock(&ztest_shared->zs_vdev_lock);
 		return;
 	}
 
@@ -1133,7 +1146,7 @@ ztest_vdev_attach_detach(ztest_args_t *za)
 		    (longlong_t)newsize, replacing, error, expected_error);
 	}
 
-	(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_unlock(&ztest_shared->zs_vdev_lock);
 }
 
 /*
@@ -1149,7 +1162,7 @@ ztest_vdev_LUN_growth(ztest_args_t *za)
 	size_t fsize;
 	int fd;
 
-	(void) mutex_lock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_lock(&ztest_shared->zs_vdev_lock);
 
 	/*
 	 * Pick a random leaf vdev.
@@ -1180,7 +1193,7 @@ ztest_vdev_LUN_growth(ztest_args_t *za)
 		(void) close(fd);
 	}
 
-	(void) mutex_unlock(&ztest_shared->zs_vdev_lock);
+	(void) pthread_mutex_unlock(&ztest_shared->zs_vdev_lock);
 }
 
 /* ARGSUSED */
@@ -1279,7 +1292,7 @@ ztest_dmu_objset_create_destroy(ztest_args_t *za)
 	uint64_t seq;
 	uint64_t objects;
 
-	(void) rw_rdlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_rdlock(&ztest_shared->zs_name_lock);
 	(void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool,
 	    (u_longlong_t)za->za_instance);
 
@@ -1322,7 +1335,7 @@ ztest_dmu_objset_create_destroy(ztest_args_t *za)
 	if (error) {
 		if (error == ENOSPC) {
 			ztest_record_enospc("dmu_objset_create");
-			(void) rw_unlock(&ztest_shared->zs_name_lock);
+			(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 			return;
 		}
 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
@@ -1404,7 +1417,7 @@ ztest_dmu_objset_create_destroy(ztest_args_t *za)
 	if (error)
 		fatal(0, "dmu_objset_destroy(%s) = %d", name, error);
 
-	(void) rw_unlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 }
 
 /*
@@ -1418,7 +1431,7 @@ ztest_dmu_snapshot_create_destroy(ztest_args_t *za)
 	char snapname[100];
 	char osname[MAXNAMELEN];
 
-	(void) rw_rdlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_rdlock(&ztest_shared->zs_name_lock);
 	dmu_objset_name(os, osname);
 	(void) snprintf(snapname, 100, "%s@%llu", osname,
 	    (u_longlong_t)za->za_instance);
@@ -1431,7 +1444,7 @@ ztest_dmu_snapshot_create_destroy(ztest_args_t *za)
 		ztest_record_enospc("dmu_take_snapshot");
 	else if (error != 0 && error != EEXIST)
 		fatal(0, "dmu_take_snapshot() = %d", error);
-	(void) rw_unlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 }
 
 /*
@@ -1929,7 +1942,7 @@ ztest_dmu_write_parallel(ztest_args_t *za)
 	int bs = ZTEST_DIROBJ_BLOCKSIZE;
 	int do_free = 0;
 	uint64_t off, txg, txg_how;
-	mutex_t *lp;
+	pthread_mutex_t *lp;
 	char osname[MAXNAMELEN];
 	char iobuf[SPA_MAXBLOCKSIZE];
 	blkptr_t blk;
@@ -1980,7 +1993,7 @@ ztest_dmu_write_parallel(ztest_args_t *za)
 	txg = dmu_tx_get_txg(tx);
 
 	lp = &ztest_shared->zs_sync_lock[b];
-	(void) mutex_lock(lp);
+	(void) pthread_mutex_lock(lp);
 
 	wbt->bt_objset = dmu_objset_id(os);
 	wbt->bt_object = ZTEST_DIROBJ;
@@ -2033,7 +2046,7 @@ ztest_dmu_write_parallel(ztest_args_t *za)
 		dmu_write(os, ZTEST_DIROBJ, off, btsize, wbt, tx);
 	}
 
-	(void) mutex_unlock(lp);
+	(void) pthread_mutex_unlock(lp);
 
 	if (ztest_random(1000) == 0)
 		(void) poll(NULL, 0, 1); /* open dn_notxholds window */
@@ -2052,13 +2065,13 @@ ztest_dmu_write_parallel(ztest_args_t *za)
 	/*
 	 * dmu_sync() the block we just wrote.
 	 */
-	(void) mutex_lock(lp);
+	(void) pthread_mutex_lock(lp);
 
 	blkoff = P2ALIGN_TYPED(off, bs, uint64_t);
 	error = dmu_buf_hold(os, ZTEST_DIROBJ, blkoff, FTAG, &db);
 	za->za_dbuf = db;
 	if (error) {
-		(void) mutex_unlock(lp);
+		(void) pthread_mutex_unlock(lp);
 		return;
 	}
 	blkoff = off - blkoff;
@@ -2066,7 +2079,7 @@ ztest_dmu_write_parallel(ztest_args_t *za)
 	dmu_buf_rele(db, FTAG);
 	za->za_dbuf = NULL;
 
-	(void) mutex_unlock(lp);
+	(void) pthread_mutex_unlock(lp);
 
 	if (error)
 		return;
@@ -2435,6 +2448,205 @@ ztest_zap_parallel(ztest_args_t *za)
 		dmu_tx_commit(tx);
 }
 
+/*
+ * Commit callback data.
+ */
+typedef struct ztest_cb_data {
+	list_node_t		zcd_node;
+	ztest_cb_list_t		*zcd_zcl;
+	uint64_t		zcd_txg;
+	int			zcd_expected_err;
+	boolean_t		zcd_added;
+	boolean_t		zcd_called;
+	spa_t			*zcd_spa;
+} ztest_cb_data_t;
+
+static void
+ztest_commit_callback(void *arg, int error)
+{
+	ztest_cb_data_t *data = arg;
+	ztest_cb_list_t *zcl;
+	uint64_t synced_txg;
+
+	VERIFY(data != NULL);
+	VERIFY3S(data->zcd_expected_err, ==, error);
+	VERIFY(!data->zcd_called);
+
+	synced_txg = spa_last_synced_txg(data->zcd_spa);
+	if (data->zcd_txg > synced_txg)
+		fatal(0, "commit callback of txg %llu prematurely called, last"
+		    " synced txg = %llu\n", data->zcd_txg, synced_txg);
+
+	zcl = data->zcd_zcl;
+	data->zcd_called = B_TRUE;
+
+	if (error == ECANCELED) {
+		ASSERT3U(data->zcd_txg, ==, 0);
+		ASSERT(!data->zcd_added);
+
+		/*
+		 * The private callback data should be destroyed here, but
+		 * since we are going to check the zcd_called field after
+		 * dmu_tx_abort(), we will destroy it there.
+		 */
+		return;
+	}
+
+	if (!data->zcd_added)
+		goto out;
+
+	ASSERT3U(data->zcd_txg, !=, 0);
+
+	/* Remove our callback from the list */
+	(void) mutex_lock(&zcl->zcl_callbacks_lock);
+	list_remove(&zcl->zcl_callbacks, data);
+	(void) mutex_unlock(&zcl->zcl_callbacks_lock);
+
+out:
+	umem_free(data, sizeof (ztest_cb_data_t));
+}
+
+static ztest_cb_data_t *
+ztest_create_cb_data(objset_t *os, ztest_cb_list_t *zcl, uint64_t txg)
+{
+	ztest_cb_data_t *cb_data;
+
+	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
+
+	cb_data->zcd_zcl = zcl;
+	cb_data->zcd_txg = txg;
+	cb_data->zcd_spa = os->os->os_spa;
+
+	return (cb_data);
+}
+
+/*
+ * If a number of txgs equal to this threshold have been created after a commit
+ * callback has been registered but not called, then we assume there is an
+ * implementation bug.
+ */
+#define	ZTEST_COMMIT_CALLBACK_THRESH	3
+
+/*
+ * Commit callback test.
+ */
+void
+ztest_dmu_commit_callbacks(ztest_args_t *za)
+{
+	objset_t *os = za->za_os;
+	dmu_tx_t *tx;
+	ztest_cb_list_t *zcl = &ztest_shared->zs_cb_list;
+	ztest_cb_data_t *cb_data[3], *tmp_cb;
+	uint64_t old_txg, txg;
+	int i, error;
+
+	tx = dmu_tx_create(os);
+
+	cb_data[0] = ztest_create_cb_data(os, zcl, 0);
+	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
+
+	dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t));
+
+	/* Every once in a while, abort the transaction on purpose */
+	if (ztest_random(100) == 0)
+		error = -1;
+
+	if (!error)
+		error = dmu_tx_assign(tx, TXG_NOWAIT);
+
+	txg = error ? 0 : dmu_tx_get_txg(tx);
+
+	cb_data[1] = ztest_create_cb_data(os, zcl, txg);
+	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
+
+	if (error) {
+		/*
+		 * It's not a strict requirement to call the registered
+		 * callbacks from inside dmu_tx_abort(), but that's what
+		 * happens in the current implementation so we will check for
+		 * that.
+		 */
+		for (i = 0; i < 2; i++) {
+			cb_data[i]->zcd_expected_err = ECANCELED;
+			VERIFY(!cb_data[i]->zcd_called);
+		}
+
+		dmu_tx_abort(tx);
+
+		for (i = 0; i < 2; i++) {
+			VERIFY(cb_data[i]->zcd_called);
+			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
+		}
+
+		return;
+	}
+
+	cb_data[0]->zcd_txg = txg;
+	cb_data[2] = ztest_create_cb_data(os, zcl, txg);
+	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
+
+	/*
+	 * Read existing data to make sure there isn't a future leak.
+	 */
+	VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t),
+	    &old_txg));
+
+	if (old_txg > txg)
+		fatal(0, "future leak: got %llx, open txg is %llx", old_txg,
+		    txg);
+
+	dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t), &txg, tx);
+
+	(void) mutex_lock(&zcl->zcl_callbacks_lock);
+
+	/*
+	 * Since commit callbacks don't have any ordering requirement and since
+	 * it is theoretically possible for a commit callback to be called
+	 * after an arbitrary amount of time has elapsed since its txg has been
+	 * synced, it is difficult to reliably determine whether a commit
+	 * callback hasn't been called due to high load or due to a flawed
+	 * implementation.
+	 *
+	 * In practice, we will assume that if after a certain number of txgs a
+	 * commit callback hasn't been called, then most likely there's an
+	 * implementation bug..
+	 */
+	tmp_cb = list_head(&zcl->zcl_callbacks);
+	if (tmp_cb != NULL)
+		VERIFY3U(tmp_cb->zcd_txg, >,
+		    txg - ZTEST_COMMIT_CALLBACK_THRESH);
+
+	/*
+	 * Let's find the place to insert our callbacks.
+	 *
+	 * Even though the list is ordered by txg, it is possible for the
+	 * insertion point to not be the end because our txg may already be
+	 * quiescing at this point and other callbacks in the open txg may
+	 * have sneaked in.
+	 */
+	tmp_cb = list_tail(&zcl->zcl_callbacks);
+	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
+		tmp_cb = list_prev(&zcl->zcl_callbacks, tmp_cb);
+
+	/* Add the 3 callbacks to the list */
+	for (i = 0; i < 3; i++) {
+		if (tmp_cb == NULL)
+			list_insert_head(&zcl->zcl_callbacks, cb_data[i]);
+		else
+			list_insert_after(&zcl->zcl_callbacks, tmp_cb,
+			    cb_data[i]);
+
+		cb_data[i]->zcd_added = B_TRUE;
+		VERIFY(!cb_data[i]->zcd_called);
+
+		tmp_cb = cb_data[i];
+	}
+
+	(void) mutex_unlock(&zcl->zcl_callbacks_lock);
+
+	dmu_tx_commit(tx);
+}
+
 void
 ztest_dsl_prop_get_set(ztest_args_t *za)
 {
@@ -2446,7 +2658,7 @@ ztest_dsl_prop_get_set(ztest_args_t *za)
 	char osname[MAXNAMELEN];
 	int error;
 
-	(void) rw_rdlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_rdlock(&ztest_shared->zs_name_lock);
 
 	dmu_objset_name(os, osname);
 
@@ -2485,7 +2697,7 @@ ztest_dsl_prop_get_set(ztest_args_t *za)
 		}
 	}
 
-	(void) rw_unlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 }
 
 /*
@@ -2649,7 +2861,7 @@ ztest_spa_rename(ztest_args_t *za)
 	int error;
 	spa_t *spa;
 
-	(void) rw_wrlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_wrlock(&ztest_shared->zs_name_lock);
 
 	oldname = za->za_pool;
 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
@@ -2701,7 +2913,7 @@ ztest_spa_rename(ztest_args_t *za)
 
 	umem_free(newname, strlen(newname) + 1);
 
-	(void) rw_unlock(&ztest_shared->zs_name_lock);
+	(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 }
 
 
@@ -2929,15 +3141,18 @@ ztest_spa_import_export(char *oldname, char *newname)
 	nvlist_free(config);
 }
 
-static void
+static void *
-ztest_resume(spa_t *spa)
+ztest_resume(void *arg)
 {
+	spa_t *spa = arg;
+
 	if (spa_suspended(spa)) {
 		spa_vdev_state_enter(spa);
 		vdev_clear(spa, NULL);
 		(void) spa_vdev_state_exit(spa, NULL, 0);
 		zio_resume(spa);
 	}
+	return (NULL);
 }
 
 static void *
@@ -3027,15 +3242,19 @@ ztest_run(char *pool)
 	ztest_args_t *za;
 	spa_t *spa;
 	char name[100];
-	thread_t resume_tid;
+	pthread_t resume_tid;
 
 	ztest_exiting = B_FALSE;
 
-	(void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL);
+	(void) pthread_mutex_init(&zs->zs_vdev_lock, NULL);
-	(void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL);
+	(void) pthread_rwlock_init(&zs->zs_name_lock, NULL);
+	(void) pthread_mutex_init(&zs->zs_cb_list.zcl_callbacks_lock, NULL);
+
+	list_create(&zs->zs_cb_list.zcl_callbacks, sizeof (ztest_cb_data_t),
+	    offsetof(ztest_cb_data_t, zcd_node));
 
 	for (t = 0; t < ZTEST_SYNC_LOCKS; t++)
-		(void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL);
+		(void) pthread_mutex_init(&zs->zs_sync_lock[t], NULL);
 
 	/*
 	 * Destroy one disk before we even start.
@@ -3102,8 +3321,7 @@ ztest_run(char *pool)
 	/*
 	 * Create a thread to periodically resume suspended I/O.
 	 */
-	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
+	VERIFY(pthread_create(&resume_tid, NULL, ztest_resume_thread, spa)==0);
-	    &resume_tid) == 0);
 
 	/*
 	 * Verify that we can safely inquire about about any object,
@@ -3152,7 +3370,7 @@ ztest_run(char *pool)
 
 		if (t < zopt_datasets) {
 			int test_future = FALSE;
-			(void) rw_rdlock(&ztest_shared->zs_name_lock);
+			(void) pthread_rwlock_rdlock(&ztest_shared->zs_name_lock);
 			(void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
 			error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0,
 			    ztest_create_cb, NULL);
@@ -3160,7 +3378,7 @@ ztest_run(char *pool)
 				test_future = TRUE;
 			} else if (error == ENOSPC) {
 				zs->zs_enospc_count++;
-				(void) rw_unlock(&ztest_shared->zs_name_lock);
+				(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 				break;
 			} else if (error != 0) {
 				fatal(0, "dmu_objset_create(%s) = %d",
@@ -3171,7 +3389,7 @@ ztest_run(char *pool)
 			if (error)
 				fatal(0, "dmu_objset_open('%s') = %d",
 				    name, error);
-			(void) rw_unlock(&ztest_shared->zs_name_lock);
+			(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 			if (test_future)
 				ztest_dmu_check_future_leak(&za[t]);
 			zil_replay(za[d].za_os, za[d].za_os,
@@ -3179,12 +3397,12 @@ ztest_run(char *pool)
 			za[d].za_zilog = zil_open(za[d].za_os, NULL);
 		}
 
-		VERIFY(thr_create(0, 0, ztest_thread, &za[t], THR_BOUND,
+		VERIFY(pthread_create(&za[t].za_thread, NULL, ztest_thread,
-		    &za[t].za_thread) == 0);
+		    &za[t]) == 0);
 	}
 
 	while (--t >= 0) {
-		VERIFY(thr_join(za[t].za_thread, NULL, NULL) == 0);
+		VERIFY(pthread_join(za[t].za_thread, NULL) == 0);
 		if (t < zopt_datasets) {
 			zil_close(za[t].za_zilog);
 			dmu_objset_close(za[t].za_os);
@@ -3203,14 +3421,14 @@ ztest_run(char *pool)
 	 * If we had out-of-space errors, destroy a random objset.
 	 */
 	if (zs->zs_enospc_count != 0) {
-		(void) rw_rdlock(&ztest_shared->zs_name_lock);
+		(void) pthread_rwlock_rdlock(&ztest_shared->zs_name_lock);
 		d = (int)ztest_random(zopt_datasets);
 		(void) snprintf(name, 100, "%s/%s_%d", pool, pool, d);
 		if (zopt_verbose >= 3)
 			(void) printf("Destroying %s to free up space\n", name);
 		(void) dmu_objset_find(name, ztest_destroy_cb, &za[d],
 		    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
-		(void) rw_unlock(&ztest_shared->zs_name_lock);
+		(void) pthread_rwlock_unlock(&ztest_shared->zs_name_lock);
 	}
 
 	txg_wait_synced(spa_get_dsl(spa), 0);
@@ -3219,7 +3437,7 @@ ztest_run(char *pool)
 
 	/* Kill the resume thread */
 	ztest_exiting = B_TRUE;
-	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
+	VERIFY(pthread_join(resume_tid, NULL) == 0);
 	ztest_resume(spa);
 
 	/*
@@ -3232,6 +3450,12 @@ ztest_run(char *pool)
 	spa_close(spa, FTAG);
 
 	kernel_fini();
+
+	list_destroy(&zs->zs_cb_list.zcl_callbacks);
+
+	(void) pthread_mutex_destroy(&zs->zs_cb_list.zcl_callbacks_lock);
+	(void) pthread_rwlock_destroy(&zs->zs_name_lock);
+	(void) pthread_mutex_destroy(&zs->zs_vdev_lock);
 }
 
 void

lib/libuutil/uu_misc.c

@@ -37,7 +37,6 @@
 #include <stdlib.h>
 #include <string.h>
 #include <sys/debug.h>
-#include <thread.h>
 #include <unistd.h>
 
 #if !defined(TEXT_DOMAIN)
@@ -70,11 +69,12 @@ static va_list		uu_panic_args;
 static pthread_t	uu_panic_thread;
 
 static uint32_t		_uu_main_error;
+static __thread int	_uu_main_thread = 0;
 
 void
 uu_set_error(uint_t code)
 {
-	if (thr_main() != 0) {
+	if (_uu_main_thread) {
 		_uu_main_error = code;
 		return;
 	}
@@ -103,7 +103,7 @@ uu_set_error(uint_t code)
 uint32_t
 uu_error(void)
 {
-	if (thr_main() != 0)
+	if (_uu_main_thread)
 		return (_uu_main_error);
 
 	if (uu_error_key_setup < 0)	/* can't happen? */
@@ -249,5 +249,6 @@ uu_init(void) __attribute__((constructor));
 static void
 uu_init(void)
 {
+	_uu_main_thread = 1;
 	(void) pthread_atfork(uu_lockup, uu_release, uu_release_child);
 }

lib/libzpool/include/sys/zfs_context.h

@@ -50,8 +50,7 @@ extern "C" {
 #include <errno.h>
 #include <string.h>
 #include <strings.h>
-#include <synch.h>
+#include <pthread.h>
-#include <thread.h>
 #include <assert.h>
 #include <alloca.h>
 #include <umem.h>
@@ -150,15 +149,18 @@ extern void vpanic(const char *, __va_list);
 /*
  * Threads
  */
-#define	curthread	((void *)(uintptr_t)thr_self())
+#define	curthread		((void *)(uintptr_t)pthread_self())
+#define	tsd_get(key)		pthread_getspecific(key)
+#define	tsd_set(key, val)	pthread_setspecific(key, val)
 
 typedef struct kthread kthread_t;
+typedef void (*thread_func_t)(void *);
 
 #define	thread_create(stk, stksize, func, arg, len, pp, state, pri)	\
-	zk_thread_create(func, arg)
+	zk_thread_create((thread_func_t)func, arg)
-#define	thread_exit() thr_exit(NULL)
+#define	thread_exit() pthread_exit(NULL)
 
-extern kthread_t *zk_thread_create(void (*func)(), void *arg);
+extern kthread_t *zk_thread_create(thread_func_t func, void *arg);
 
 #define	issig(why)	(FALSE)
 #define	ISSIG(thr, why)	(FALSE)
@@ -166,28 +168,18 @@ extern kthread_t *zk_thread_create(void (*func)(), void *arg);
 /*
  * Mutexes
  */
+#define MTX_MAGIC 0x9522f51362a6e326ull
 typedef struct kmutex {
 	void		*m_owner;
-	boolean_t	initialized;
+	uint64_t	m_magic;
-	mutex_t		m_lock;
+	pthread_mutex_t	m_lock;
 } kmutex_t;
 
-#define	MUTEX_DEFAULT	USYNC_THREAD
+#define	MUTEX_DEFAULT	0
-#undef MUTEX_HELD
+#define	MUTEX_HELD(m)	((m)->m_owner == curthread)
-#define	MUTEX_HELD(m) _mutex_held(&(m)->m_lock)
 
-/*
+extern void mutex_init(kmutex_t *mp, char *name, int type, void *cookie);
- * Argh -- we have to get cheesy here because the kernel and userland
+extern void mutex_destroy(kmutex_t *mp);
- * have different signatures for the same routine.
- */
-extern int _mutex_init(mutex_t *mp, int type, void *arg);
-extern int _mutex_destroy(mutex_t *mp);
-
-#define	mutex_init(mp, b, c, d)		zmutex_init((kmutex_t *)(mp))
-#define	mutex_destroy(mp)		zmutex_destroy((kmutex_t *)(mp))
-
-extern void zmutex_init(kmutex_t *mp);
-extern void zmutex_destroy(kmutex_t *mp);
 extern void mutex_enter(kmutex_t *mp);
 extern void mutex_exit(kmutex_t *mp);
 extern int mutex_tryenter(kmutex_t *mp);
@@ -196,23 +188,24 @@ extern void *mutex_owner(kmutex_t *mp);
 /*
  * RW locks
  */
+#define RW_MAGIC 0x4d31fb123648e78aull
 typedef struct krwlock {
 	void			*rw_owner;
-	boolean_t	initialized;
+	void			*rw_wr_owner;
-	rwlock_t	rw_lock;
+	uint64_t		rw_magic;
+	pthread_rwlock_t	rw_lock;
+	uint_t			rw_readers;
 } krwlock_t;
 
 typedef int krw_t;
 
 #define	RW_READER	0
 #define	RW_WRITER	1
-#define	RW_DEFAULT	USYNC_THREAD
+#define	RW_DEFAULT	0
 
-#undef RW_READ_HELD
+#define	RW_READ_HELD(x)		((x)->rw_readers > 0)
-#define	RW_READ_HELD(x)		_rw_read_held(&(x)->rw_lock)
+#define	RW_WRITE_HELD(x)	((x)->rw_wr_owner == curthread)
-
+#define	RW_LOCK_HELD(x)		(RW_READ_HELD(x) || RW_WRITE_HELD(x))
-#undef RW_WRITE_HELD
-#define	RW_WRITE_HELD(x)	_rw_write_held(&(x)->rw_lock)
 
 extern void rw_init(krwlock_t *rwlp, char *name, int type, void *arg);
 extern void rw_destroy(krwlock_t *rwlp);
@@ -230,9 +223,13 @@ extern gid_t *crgetgroups(cred_t *cr);
 /*
  * Condition variables
  */
-typedef cond_t kcondvar_t;
+#define CV_MAGIC 0xd31ea9a83b1b30c4ull
+typedef struct kcondvar {
+	uint64_t cv_magic;
+	pthread_cond_t cv;
+} kcondvar_t;
 
-#define	CV_DEFAULT	USYNC_THREAD
+#define	CV_DEFAULT	0
 
 extern void cv_init(kcondvar_t *cv, char *name, int type, void *arg);
 extern void cv_destroy(kcondvar_t *cv);
@@ -409,7 +406,8 @@ extern void delay(clock_t ticks);
 #define	minclsyspri	60
 #define	maxclsyspri	99
 
-#define	CPU_SEQID	(thr_self() & (max_ncpus - 1))
+/* XXX: not portable */
+#define	CPU_SEQID	(pthread_self() & (max_ncpus - 1))
 
 #define	kcred		NULL
 #define	CRED()		NULL

lib/libzpool/kernel.c

@@ -35,6 +35,7 @@
 #include <sys/processor.h>
 #include <sys/zfs_context.h>
 #include <sys/utsname.h>
+#include <sys/time.h>
 #include <sys/systeminfo.h>
 
 /*
@@ -56,13 +57,17 @@ struct utsname utsname = {
  */
 /*ARGSUSED*/
 kthread_t *
-zk_thread_create(void (*func)(), void *arg)
+zk_thread_create(thread_func_t func, void *arg)
 {
-	thread_t tid;
+	pthread_t tid;
 
-	VERIFY(thr_create(0, 0, (void *(*)(void *))func, arg, THR_DETACHED,
+	pthread_attr_t attr;
-	    &tid) == 0);
+	VERIFY(pthread_attr_init(&attr) == 0);
+	VERIFY(pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED) == 0);
 
+	VERIFY(pthread_create(&tid, &attr, (void *(*)(void *))func, arg) == 0);
+
+	/* XXX: not portable */
 	return ((void *)(uintptr_t)tid);
 }
 
@@ -95,30 +100,37 @@ kstat_delete(kstat_t *ksp)
  * =========================================================================
  */
 void
-zmutex_init(kmutex_t *mp)
+mutex_init(kmutex_t *mp, char *name, int type, void *cookie)
 {
+	ASSERT(type == MUTEX_DEFAULT);
+	ASSERT(cookie == NULL);
+
+#ifdef IM_FEELING_LUCKY
+	ASSERT(mp->m_magic != MTX_MAGIC);
+#endif
+
 	mp->m_owner = NULL;
-	mp->initialized = B_TRUE;
+	mp->m_magic = MTX_MAGIC;
-	(void) _mutex_init(&mp->m_lock, USYNC_THREAD, NULL);
+	VERIFY3S(pthread_mutex_init(&mp->m_lock, NULL), ==, 0);
 }
 
 void
-zmutex_destroy(kmutex_t *mp)
+mutex_destroy(kmutex_t *mp)
 {
-	ASSERT(mp->initialized == B_TRUE);
+	ASSERT(mp->m_magic == MTX_MAGIC);
 	ASSERT(mp->m_owner == NULL);
-	(void) _mutex_destroy(&(mp)->m_lock);
+	VERIFY3S(pthread_mutex_destroy(&(mp)->m_lock), ==, 0);
 	mp->m_owner = (void *)-1UL;
-	mp->initialized = B_FALSE;
+	mp->m_magic = 0;
 }
 
 void
 mutex_enter(kmutex_t *mp)
 {
-	ASSERT(mp->initialized == B_TRUE);
+	ASSERT(mp->m_magic == MTX_MAGIC);
 	ASSERT(mp->m_owner != (void *)-1UL);
 	ASSERT(mp->m_owner != curthread);
-	VERIFY(mutex_lock(&mp->m_lock) == 0);
+	VERIFY3S(pthread_mutex_lock(&mp->m_lock), ==, 0);
 	ASSERT(mp->m_owner == NULL);
 	mp->m_owner = curthread;
 }
@@ -126,9 +138,9 @@ mutex_enter(kmutex_t *mp)
 int
 mutex_tryenter(kmutex_t *mp)
 {
-	ASSERT(mp->initialized == B_TRUE);
+	ASSERT(mp->m_magic == MTX_MAGIC);
 	ASSERT(mp->m_owner != (void *)-1UL);
-	if (0 == mutex_trylock(&mp->m_lock)) {
+	if (0 == pthread_mutex_trylock(&mp->m_lock)) {
 		ASSERT(mp->m_owner == NULL);
 		mp->m_owner = curthread;
 		return (1);
@@ -140,16 +152,16 @@ mutex_tryenter(kmutex_t *mp)
 void
 mutex_exit(kmutex_t *mp)
 {
-	ASSERT(mp->initialized == B_TRUE);
+	ASSERT(mp->m_magic == MTX_MAGIC);
 	ASSERT(mutex_owner(mp) == curthread);
 	mp->m_owner = NULL;
-	VERIFY(mutex_unlock(&mp->m_lock) == 0);
+	VERIFY3S(pthread_mutex_unlock(&mp->m_lock), ==, 0);
 }
 
 void *
 mutex_owner(kmutex_t *mp)
 {
-	ASSERT(mp->initialized == B_TRUE);
+	ASSERT(mp->m_magic == MTX_MAGIC);
 	return (mp->m_owner);
 }
 
@@ -162,31 +174,48 @@ mutex_owner(kmutex_t *mp)
 void
 rw_init(krwlock_t *rwlp, char *name, int type, void *arg)
 {
-	rwlock_init(&rwlp->rw_lock, USYNC_THREAD, NULL);
+	ASSERT(type == RW_DEFAULT);
+	ASSERT(arg == NULL);
+
+#ifdef IM_FEELING_LUCKY
+	ASSERT(rwlp->rw_magic != RW_MAGIC);
+#endif
+
+	VERIFY3S(pthread_rwlock_init(&rwlp->rw_lock, NULL), ==, 0);
 	rwlp->rw_owner = NULL;
-	rwlp->initialized = B_TRUE;
+	rwlp->rw_wr_owner = NULL;
+	rwlp->rw_readers = 0;
+	rwlp->rw_magic = RW_MAGIC;
 }
 
 void
 rw_destroy(krwlock_t *rwlp)
 {
-	rwlock_destroy(&rwlp->rw_lock);
+	ASSERT(rwlp->rw_magic == RW_MAGIC);
-	rwlp->rw_owner = (void *)-1UL;
+
-	rwlp->initialized = B_FALSE;
+	VERIFY3S(pthread_rwlock_destroy(&rwlp->rw_lock), ==, 0);
+	rwlp->rw_magic = 0;
 }
 
 void
 rw_enter(krwlock_t *rwlp, krw_t rw)
 {
-	ASSERT(!RW_LOCK_HELD(rwlp));
+	ASSERT(rwlp->rw_magic == RW_MAGIC);
-	ASSERT(rwlp->initialized == B_TRUE);
-	ASSERT(rwlp->rw_owner != (void *)-1UL);
 	ASSERT(rwlp->rw_owner != curthread);
+	ASSERT(rwlp->rw_wr_owner != curthread);
 
-	if (rw == RW_READER)
+	if (rw == RW_READER) {
-		VERIFY(rw_rdlock(&rwlp->rw_lock) == 0);
+		VERIFY3S(pthread_rwlock_rdlock(&rwlp->rw_lock), ==, 0);
-	else
+		ASSERT(rwlp->rw_wr_owner == NULL);
-		VERIFY(rw_wrlock(&rwlp->rw_lock) == 0);
+
+		atomic_inc_uint(&rwlp->rw_readers);
+	} else {
+		VERIFY3S(pthread_rwlock_wrlock(&rwlp->rw_lock), ==, 0);
+		ASSERT(rwlp->rw_wr_owner == NULL);
+		ASSERT3U(rwlp->rw_readers, ==, 0);
+
+		rwlp->rw_wr_owner = curthread;
+	}
 
 	rwlp->rw_owner = curthread;
 }
@@ -194,11 +223,16 @@ rw_enter(krwlock_t *rwlp, krw_t rw)
 void
 rw_exit(krwlock_t *rwlp)
 {
-	ASSERT(rwlp->initialized == B_TRUE);
+	ASSERT(rwlp->rw_magic == RW_MAGIC);
-	ASSERT(rwlp->rw_owner != (void *)-1UL);
+	ASSERT(RW_LOCK_HELD(rwlp));
+
+	if (RW_READ_HELD(rwlp))
+		atomic_dec_uint(&rwlp->rw_readers);
+	else
+		rwlp->rw_wr_owner = NULL;
 
 	rwlp->rw_owner = NULL;
-	VERIFY(rw_unlock(&rwlp->rw_lock) == 0);
+	VERIFY3S(pthread_rwlock_unlock(&rwlp->rw_lock), ==, 0);
 }
 
 int
@@ -206,19 +240,29 @@ rw_tryenter(krwlock_t *rwlp, krw_t rw)
 {
 	int rv;
 
-	ASSERT(rwlp->initialized == B_TRUE);
+	ASSERT(rwlp->rw_magic == RW_MAGIC);
-	ASSERT(rwlp->rw_owner != (void *)-1UL);
 
 	if (rw == RW_READER)
-		rv = rw_tryrdlock(&rwlp->rw_lock);
+		rv = pthread_rwlock_tryrdlock(&rwlp->rw_lock);
 	else
-		rv = rw_trywrlock(&rwlp->rw_lock);
+		rv = pthread_rwlock_trywrlock(&rwlp->rw_lock);
 
 	if (rv == 0) {
+		ASSERT(rwlp->rw_wr_owner == NULL);
+
+		if (rw == RW_READER)
+			atomic_inc_uint(&rwlp->rw_readers);
+		else {
+			ASSERT3U(rwlp->rw_readers, ==, 0);
+			rwlp->rw_wr_owner = curthread;
+		}
+
 		rwlp->rw_owner = curthread;
 		return (1);
 	}
 
+	VERIFY3S(rv, ==, EBUSY);
+
 	return (0);
 }
 
@@ -226,8 +270,7 @@ rw_tryenter(krwlock_t *rwlp, krw_t rw)
 int
 rw_tryupgrade(krwlock_t *rwlp)
 {
-	ASSERT(rwlp->initialized == B_TRUE);
+	ASSERT(rwlp->rw_magic == RW_MAGIC);
-	ASSERT(rwlp->rw_owner != (void *)-1UL);
 
 	return (0);
 }
@@ -241,22 +284,34 @@ rw_tryupgrade(krwlock_t *rwlp)
 void
 cv_init(kcondvar_t *cv, char *name, int type, void *arg)
 {
-	VERIFY(cond_init(cv, type, NULL) == 0);
+	ASSERT(type == CV_DEFAULT);
+
+#ifdef IM_FEELING_LUCKY
+	ASSERT(cv->cv_magic != CV_MAGIC);
+#endif
+
+	cv->cv_magic = CV_MAGIC;
+
+	VERIFY3S(pthread_cond_init(&cv->cv, NULL), ==, 0);
 }
 
 void
 cv_destroy(kcondvar_t *cv)
 {
-	VERIFY(cond_destroy(cv) == 0);
+	ASSERT(cv->cv_magic == CV_MAGIC);
+	VERIFY3S(pthread_cond_destroy(&cv->cv), ==, 0);
+	cv->cv_magic = 0;
 }
 
 void
 cv_wait(kcondvar_t *cv, kmutex_t *mp)
 {
+	ASSERT(cv->cv_magic == CV_MAGIC);
 	ASSERT(mutex_owner(mp) == curthread);
 	mp->m_owner = NULL;
-	int ret = cond_wait(cv, &mp->m_lock);
+	int ret = pthread_cond_wait(&cv->cv, &mp->m_lock);
-	VERIFY(ret == 0 || ret == EINTR);
+	if (ret != 0)
+		VERIFY3S(ret, ==, EINTR);
 	mp->m_owner = curthread;
 }
 
@@ -264,29 +319,38 @@ clock_t
 cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime)
 {
 	int error;
+	struct timeval tv;
 	timestruc_t ts;
 	clock_t delta;
 
+	ASSERT(cv->cv_magic == CV_MAGIC);
+
 top:
 	delta = abstime - lbolt;
 	if (delta <= 0)
 		return (-1);
 
-	ts.tv_sec = delta / hz;
+	VERIFY(gettimeofday(&tv, NULL) == 0);
-	ts.tv_nsec = (delta % hz) * (NANOSEC / hz);
+
+	ts.tv_sec = tv.tv_sec + delta / hz;
+	ts.tv_nsec = tv.tv_usec * 1000 + (delta % hz) * (NANOSEC / hz);
+	if (ts.tv_nsec >= NANOSEC) {
+		ts.tv_sec++;
+		ts.tv_nsec -= NANOSEC;
+	}
 
 	ASSERT(mutex_owner(mp) == curthread);
 	mp->m_owner = NULL;
-	error = cond_reltimedwait(cv, &mp->m_lock, &ts);
+	error = pthread_cond_timedwait(&cv->cv, &mp->m_lock, &ts);
 	mp->m_owner = curthread;
 
-	if (error == ETIME)
+	if (error == ETIMEDOUT)
 		return (-1);
 
 	if (error == EINTR)
 		goto top;
 
-	ASSERT(error == 0);
+	VERIFY3S(error, ==, 0);
 
 	return (1);
 }
@@ -294,13 +358,15 @@ top:
 void
 cv_signal(kcondvar_t *cv)
 {
-	VERIFY(cond_signal(cv) == 0);
+	ASSERT(cv->cv_magic == CV_MAGIC);
+	VERIFY3S(pthread_cond_signal(&cv->cv), ==, 0);
 }
 
 void
 cv_broadcast(kcondvar_t *cv)
 {
-	VERIFY(cond_broadcast(cv) == 0);
+	ASSERT(cv->cv_magic == CV_MAGIC);
+	VERIFY3S(pthread_cond_broadcast(&cv->cv), ==, 0);
 }
 
 /*
@@ -554,7 +620,7 @@ __dprintf(const char *file, const char *func, int line, const char *fmt, ...)
 		if (dprintf_find_string("pid"))
 			(void) printf("%d ", getpid());
 		if (dprintf_find_string("tid"))
-			(void) printf("%u ", thr_self());
+			(void) printf("%u ", (uint_t) pthread_self());
 		if (dprintf_find_string("cpu"))
 			(void) printf("%u ", getcpuid());
 		if (dprintf_find_string("time"))

lib/libzpool/taskq.c

@@ -42,7 +42,7 @@ struct taskq {
 	krwlock_t	tq_threadlock;
 	kcondvar_t	tq_dispatch_cv;
 	kcondvar_t	tq_wait_cv;
-	thread_t	*tq_threadlist;
+	pthread_t	*tq_threadlist;
 	int		tq_flags;
 	int		tq_active;
 	int		tq_nthreads;
@@ -185,7 +185,7 @@ taskq_create(const char *name, int nthreads, pri_t pri,
 	tq->tq_maxalloc = maxalloc;
 	tq->tq_task.task_next = &tq->tq_task;
 	tq->tq_task.task_prev = &tq->tq_task;
-	tq->tq_threadlist = kmem_alloc(nthreads * sizeof (thread_t), KM_SLEEP);
+	tq->tq_threadlist = kmem_alloc(nthreads * sizeof (pthread_t), KM_SLEEP);
 
 	if (flags & TASKQ_PREPOPULATE) {
 		mutex_enter(&tq->tq_lock);
@@ -195,8 +195,8 @@ taskq_create(const char *name, int nthreads, pri_t pri,
 	}
 
 	for (t = 0; t < nthreads; t++)
-		(void) thr_create(0, 0, taskq_thread,
+		VERIFY(pthread_create(&tq->tq_threadlist[t],
-		    tq, THR_BOUND, &tq->tq_threadlist[t]);
+		    NULL, taskq_thread, tq) == 0);
 
 	return (tq);
 }
@@ -226,9 +226,9 @@ taskq_destroy(taskq_t *tq)
 	mutex_exit(&tq->tq_lock);
 
 	for (t = 0; t < nthreads; t++)
-		(void) thr_join(tq->tq_threadlist[t], NULL, NULL);
+		VERIFY(pthread_join(tq->tq_threadlist[t], NULL) == 0);
 
-	kmem_free(tq->tq_threadlist, nthreads * sizeof (thread_t));
+	kmem_free(tq->tq_threadlist, nthreads * sizeof (pthread_t));
 
 	rw_destroy(&tq->tq_threadlock);
 	mutex_destroy(&tq->tq_lock);
@@ -247,7 +247,7 @@ taskq_member(taskq_t *tq, void *t)
 		return (1);
 
 	for (i = 0; i < tq->tq_nthreads; i++)
-		if (tq->tq_threadlist[i] == (thread_t)(uintptr_t)t)
+		if (tq->tq_threadlist[i] == (pthread_t)(uintptr_t)t)
 			return (1);
 
 	return (0);

module/zfs/dmu_tx.c

@@ -48,6 +48,8 @@ dmu_tx_create_dd(dsl_dir_t *dd)
 		tx->tx_pool = dd->dd_pool;
 	list_create(&tx->tx_holds, sizeof (dmu_tx_hold_t),
 	    offsetof(dmu_tx_hold_t, txh_node));
+	list_create(&tx->tx_callbacks, sizeof (dmu_tx_callback_t),
+	    offsetof(dmu_tx_callback_t, dcb_node));
 #ifdef ZFS_DEBUG
 	refcount_create(&tx->tx_space_written);
 	refcount_create(&tx->tx_space_freed);
@@ -1020,8 +1022,13 @@ dmu_tx_commit(dmu_tx_t *tx)
 	if (tx->tx_tempreserve_cookie)
 		dsl_dir_tempreserve_clear(tx->tx_tempreserve_cookie, tx);
 
+	if (!list_is_empty(&tx->tx_callbacks))
+		txg_register_callbacks(&tx->tx_txgh, &tx->tx_callbacks);
+
 	if (tx->tx_anyobj == FALSE)
 		txg_rele_to_sync(&tx->tx_txgh);
+
+	list_destroy(&tx->tx_callbacks);
 	list_destroy(&tx->tx_holds);
 #ifdef ZFS_DEBUG
 	dprintf("towrite=%llu written=%llu tofree=%llu freed=%llu\n",
@@ -1050,6 +1057,14 @@ dmu_tx_abort(dmu_tx_t *tx)
 		if (dn != NULL)
 			dnode_rele(dn, tx);
 	}
+
+	/*
+	 * Call any registered callbacks with an error code.
+	 */
+	if (!list_is_empty(&tx->tx_callbacks))
+		dmu_tx_callback(&tx->tx_callbacks, ECANCELED);
+
+	list_destroy(&tx->tx_callbacks);
 	list_destroy(&tx->tx_holds);
 #ifdef ZFS_DEBUG
 	refcount_destroy_many(&tx->tx_space_written,
@@ -1067,6 +1082,34 @@ dmu_tx_get_txg(dmu_tx_t *tx)
 	return (tx->tx_txg);
 }
 
+void
+dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *func, void *data)
+{
+	dmu_tx_callback_t *dcb;
+
+	dcb = kmem_alloc(sizeof (dmu_tx_callback_t), KM_SLEEP);
+
+	dcb->dcb_func = func;
+	dcb->dcb_data = data;
+
+	list_insert_tail(&tx->tx_callbacks, dcb);
+}
+
+/*
+ * Call all the commit callbacks on a list, with a given error code.
+ */
+void
+dmu_tx_callback(list_t *cb_list, int error)
+{
+	dmu_tx_callback_t *dcb;
+
+	while (dcb = list_head(cb_list)) {
+		list_remove(cb_list, dcb);
+		dcb->dcb_func(dcb->dcb_data, error);
+		kmem_free(dcb, sizeof (dmu_tx_callback_t));
+	}
+}
+
 #if defined(_KERNEL) && defined(HAVE_SPL)
 EXPORT_SYMBOL(dmu_tx_create);
 EXPORT_SYMBOL(dmu_tx_hold_write);

module/zfs/include/sys/dmu.h

@@ -431,6 +431,26 @@ int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
 void dmu_tx_wait(dmu_tx_t *tx);
 void dmu_tx_commit(dmu_tx_t *tx);
 
+/*
+ * To register a commit callback, dmu_tx_callback_register() must be called.
+ *
+ * dcb_data is a pointer to caller private data that is passed on as a
+ * callback parameter. The caller is responsible for properly allocating and
+ * freeing it.
+ *
+ * When registering a callback, the transaction must be already created, but
+ * it cannot be committed or aborted. It can be assigned to a txg or not.
+ *
+ * The callback will be called after the transaction has been safely written
+ * to stable storage and will also be called if the dmu_tx is aborted.
+ * If there is any error which prevents the transaction from being committed to
+ * disk, the callback will be called with a value of error != 0.
+ */
+typedef void dmu_tx_callback_func_t(void *dcb_data, int error);
+
+void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
+    void *dcb_data);
+
 /*
  * Free up the data blocks for a defined range of a file.  If size is
  * zero, the range from offset to end-of-file is freed.

module/zfs/include/sys/dmu_tx.h

@@ -26,8 +26,6 @@
 #ifndef	_SYS_DMU_TX_H
 #define	_SYS_DMU_TX_H
 
-
-
 #include <sys/inttypes.h>
 #include <sys/dmu.h>
 #include <sys/txg.h>
@@ -59,6 +57,7 @@ struct dmu_tx {
 	txg_handle_t tx_txgh;
 	void *tx_tempreserve_cookie;
 	struct dmu_tx_hold *tx_needassign_txh;
+	list_t tx_callbacks; /* list of dmu_tx_callback_t on this dmu_tx */
 	uint8_t tx_anyobj;
 	int tx_err;
 #ifdef ZFS_DEBUG
@@ -98,6 +97,11 @@ typedef struct dmu_tx_hold {
 #endif
 } dmu_tx_hold_t;
 
+typedef struct dmu_tx_callback {
+	list_node_t		dcb_node;    /* linked to tx_callbacks list */
+	dmu_tx_callback_func_t	*dcb_func;   /* caller function pointer */
+	void			*dcb_data;   /* caller private data */
+} dmu_tx_callback_t;
 
 /*
  * These routines are defined in dmu.h, and are called by the user.
@@ -109,6 +113,10 @@ void dmu_tx_abort(dmu_tx_t *tx);
 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
 void dmu_tx_wait(dmu_tx_t *tx);
 
+void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,
+    void *dcb_data);
+void dmu_tx_callback(list_t *cb_list, int error);
+
 /*
  * These routines are defined in dmu_spa.h, and are called by the SPA.
  */

module/zfs/include/sys/txg.h

@@ -26,8 +26,6 @@
 #ifndef _SYS_TXG_H
 #define	_SYS_TXG_H
 
-
-
 #include <sys/spa.h>
 #include <sys/zfs_context.h>
 
@@ -71,6 +69,7 @@ extern void txg_sync_stop(struct dsl_pool *dp);
 extern uint64_t txg_hold_open(struct dsl_pool *dp, txg_handle_t *txghp);
 extern void txg_rele_to_quiesce(txg_handle_t *txghp);
 extern void txg_rele_to_sync(txg_handle_t *txghp);
+extern void txg_register_callbacks(txg_handle_t *txghp, list_t *tx_callbacks);
 extern void txg_suspend(struct dsl_pool *dp);
 extern void txg_resume(struct dsl_pool *dp);
 

module/zfs/include/sys/txg_impl.h

@@ -33,10 +33,16 @@
 extern "C" {
 #endif
 
+typedef struct tx_cb {
+	tx_cpu_t	*tcb_tc;
+	uint64_t	tcb_txg;
+} tx_cb_t;
+
 struct tx_cpu {
 	kmutex_t	tc_lock;
 	kcondvar_t	tc_cv[TXG_SIZE];
 	uint64_t	tc_count[TXG_SIZE];
+	list_t		tc_callbacks[TXG_SIZE]; /* commit cb list */
 	char		tc_pad[16];
 };
 
@@ -64,6 +70,8 @@ typedef struct tx_state {
 
 	kthread_t	*tx_sync_thread;
 	kthread_t	*tx_quiesce_thread;
+
+	taskq_t		*tx_commit_cb_taskq; /* commit callback taskq */
 } tx_state_t;
 
 #ifdef	__cplusplus

module/zfs/txg.c

@@ -26,6 +26,7 @@
 #include <sys/zfs_context.h>
 #include <sys/txg_impl.h>
 #include <sys/dmu_impl.h>
+#include <sys/dmu_tx.h>
 #include <sys/dsl_pool.h>
 #include <sys/callb.h>
 
@@ -57,6 +58,9 @@ txg_init(dsl_pool_t *dp, uint64_t txg)
 		for (i = 0; i < TXG_SIZE; i++) {
 			cv_init(&tx->tx_cpu[c].tc_cv[i], NULL, CV_DEFAULT,
 			    NULL);
+			list_create(&tx->tx_cpu[c].tc_callbacks[i],
+			    sizeof (dmu_tx_callback_t),
+			    offsetof(dmu_tx_callback_t, dcb_node));
 		}
 	}
 
@@ -96,9 +100,14 @@ txg_fini(dsl_pool_t *dp)
 		int i;
 
 		mutex_destroy(&tx->tx_cpu[c].tc_lock);
-		for (i = 0; i < TXG_SIZE; i++)
+		for (i = 0; i < TXG_SIZE; i++) {
 			cv_destroy(&tx->tx_cpu[c].tc_cv[i]);
+			list_destroy(&tx->tx_cpu[c].tc_callbacks[i]);
 		}
+	}
+
+	if (tx->tx_commit_cb_taskq != NULL)
+		taskq_destroy(tx->tx_commit_cb_taskq);
 
 	vmem_free(tx->tx_cpu, max_ncpus * sizeof (tx_cpu_t));
 
@@ -229,25 +238,55 @@ txg_rele_to_quiesce(txg_handle_t *th)
 }
 
 void
-txg_rele_to_sync(txg_handle_t *th)
+txg_register_callbacks(txg_handle_t *th, list_t *tx_callbacks)
 {
 	tx_cpu_t *tc = th->th_cpu;
 	int g = th->th_txg & TXG_MASK;
 
+	mutex_enter(&tc->tc_lock);
+	list_move_tail(&tc->tc_callbacks[g], tx_callbacks);
+	mutex_exit(&tc->tc_lock);
+}
+
+static void
+txg_exit(tx_cpu_t *tc, uint64_t txg)
+{
+	int g = txg & TXG_MASK;
+
 	mutex_enter(&tc->tc_lock);
 	ASSERT(tc->tc_count[g] != 0);
 	if (--tc->tc_count[g] == 0)
 		cv_broadcast(&tc->tc_cv[g]);
 	mutex_exit(&tc->tc_lock);
+}
+
+void
+txg_rele_to_sync(txg_handle_t *th)
+{
+	txg_exit(th->th_cpu, th->th_txg);
 
 	th->th_cpu = NULL;	/* defensive */
 }
 
+static void
+txg_wait_exit(tx_state_t *tx, uint64_t txg)
+{
+	int g = txg & TXG_MASK;
+	int c;
+
+	for (c = 0; c < max_ncpus; c++) {
+		tx_cpu_t *tc = &tx->tx_cpu[c];
+		mutex_enter(&tc->tc_lock);
+		while (tc->tc_count[g] != 0)
+			cv_wait(&tc->tc_cv[g], &tc->tc_lock);
+		mutex_exit(&tc->tc_lock);
+	}
+}
+
 static void
 txg_quiesce(dsl_pool_t *dp, uint64_t txg)
 {
 	tx_state_t *tx = &dp->dp_tx;
-	int g = txg & TXG_MASK;
 	int c;
 
 	/*
@@ -269,12 +308,60 @@ txg_quiesce(dsl_pool_t *dp, uint64_t txg)
 	/*
 	 * Quiesce the transaction group by waiting for everyone to txg_exit().
 	 */
+	txg_wait_exit(tx, txg);
+}
+
+static void
+txg_callback(tx_cb_t *tcb)
+{
+	tx_cpu_t *tc = tcb->tcb_tc;
+	int g = tcb->tcb_txg & TXG_MASK;
+
+	dmu_tx_callback(&tc->tc_callbacks[g], 0);
+
+	txg_exit(tc, tcb->tcb_txg);
+
+	kmem_free(tcb, sizeof (tx_cb_t));
+}
+
+/*
+ * Dispatch the commit callbacks registered on this txg to worker threads.
+ */
+static void
+txg_dispatch_callbacks(dsl_pool_t *dp, uint64_t txg)
+{
+	int c;
+	tx_state_t *tx = &dp->dp_tx;
+	tx_cb_t *tcb;
+
 	for (c = 0; c < max_ncpus; c++) {
 		tx_cpu_t *tc = &tx->tx_cpu[c];
-		mutex_enter(&tc->tc_lock);
+		/* No need to lock tx_cpu_t at this point */
-		while (tc->tc_count[g] != 0)
+
-			cv_wait(&tc->tc_cv[g], &tc->tc_lock);
+		int g = txg & TXG_MASK;
-		mutex_exit(&tc->tc_lock);
+
+		if (list_is_empty(&tc->tc_callbacks[g]))
+			continue;
+
+		if (tx->tx_commit_cb_taskq == NULL) {
+			/*
+			 * Commit callback taskq hasn't been created yet.
+			 */
+			tx->tx_commit_cb_taskq = taskq_create("tx_commit_cb",
+			    max_ncpus, minclsyspri, max_ncpus, max_ncpus * 4,
+			    TASKQ_PREPOPULATE);
+		}
+
+		tcb = kmem_alloc(sizeof (tx_cb_t), KM_SLEEP);
+		tcb->tcb_txg = txg;
+		tcb->tcb_tc = tc;
+
+		/* There shouldn't be any holders on this txg at this point */
+		ASSERT3U(tc->tc_count[g], ==, 0);
+		tc->tc_count[g]++;
+
+		(void) taskq_dispatch(tx->tx_commit_cb_taskq, (task_func_t *)
+		    txg_callback, tcb, TQ_SLEEP);
 	}
 }
 
@@ -345,6 +432,13 @@ txg_sync_thread(dsl_pool_t *dp)
 		spa_sync(dp->dp_spa, txg);
 		delta = lbolt - start;
 
+		/*
+		 * Dispatch commit callbacks to worker threads and wait for
+		 * them to finish.
+		 */
+		txg_dispatch_callbacks(dp, txg);
+		txg_wait_exit(tx, txg);
+
 		mutex_enter(&tx->tx_sync_lock);
 		rw_enter(&tx->tx_suspend, RW_WRITER);
 		tx->tx_synced_txg = txg;