1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Martin Matuska. All rights reserved.
25 * Copyright 2015, Joyent, Inc.
26 */
27
28#include <sys/zfs_context.h>
29#include <sys/dmu.h>
30#include <sys/dmu_objset.h>
31#include <sys/dmu_tx.h>
32#include <sys/dsl_dataset.h>
33#include <sys/dsl_dir.h>
34#include <sys/dsl_prop.h>
35#include <sys/dsl_synctask.h>
36#include <sys/spa.h>
37#include <sys/zap.h>
38#include <sys/fs/zfs.h>
39
40#include "zfs_prop.h"
41
42#define	ZPROP_INHERIT_SUFFIX "$inherit"
43#define	ZPROP_RECVD_SUFFIX "$recvd"
44
45static int
46dodefault(zfs_prop_t prop, int intsz, int numints, void *buf)
47{
48	/*
49	 * The setonce properties are read-only, BUT they still
50	 * have a default value that can be used as the initial
51	 * value.
52	 */
53	if (prop == ZPROP_INVAL ||
54	    (zfs_prop_readonly(prop) && !zfs_prop_setonce(prop)))
55		return (SET_ERROR(ENOENT));
56
57	if (zfs_prop_get_type(prop) == PROP_TYPE_STRING) {
58		if (intsz != 1)
59			return (SET_ERROR(EOVERFLOW));
60		(void) strncpy(buf, zfs_prop_default_string(prop),
61		    numints);
62	} else {
63		if (intsz != 8 || numints < 1)
64			return (SET_ERROR(EOVERFLOW));
65
66		*(uint64_t *)buf = zfs_prop_default_numeric(prop);
67	}
68
69	return (0);
70}
71
72int
73dsl_prop_get_dd(dsl_dir_t *dd, const char *propname,
74    int intsz, int numints, void *buf, char *setpoint, boolean_t snapshot)
75{
76	int err = ENOENT;
77	dsl_dir_t *target = dd;
78	objset_t *mos = dd->dd_pool->dp_meta_objset;
79	zfs_prop_t prop;
80	boolean_t inheritable;
81	boolean_t inheriting = B_FALSE;
82	char *inheritstr;
83	char *recvdstr;
84
85	ASSERT(dsl_pool_config_held(dd->dd_pool));
86
87	if (setpoint)
88		setpoint[0] = '\0';
89
90	prop = zfs_name_to_prop(propname);
91	inheritable = (prop == ZPROP_INVAL || zfs_prop_inheritable(prop));
92	inheritstr = kmem_asprintf("%s%s", propname, ZPROP_INHERIT_SUFFIX);
93	recvdstr = kmem_asprintf("%s%s", propname, ZPROP_RECVD_SUFFIX);
94
95	/*
96	 * Note: dd may become NULL, therefore we shouldn't dereference it
97	 * after this loop.
98	 */
99	for (; dd != NULL; dd = dd->dd_parent) {
100		if (dd != target || snapshot) {
101			if (!inheritable)
102				break;
103			inheriting = B_TRUE;
104		}
105
106		/* Check for a local value. */
107		err = zap_lookup(mos, dsl_dir_phys(dd)->dd_props_zapobj,
108		    propname, intsz, numints, buf);
109		if (err != ENOENT) {
110			if (setpoint != NULL && err == 0)
111				dsl_dir_name(dd, setpoint);
112			break;
113		}
114
115		/*
116		 * Skip the check for a received value if there is an explicit
117		 * inheritance entry.
118		 */
119		err = zap_contains(mos, dsl_dir_phys(dd)->dd_props_zapobj,
120		    inheritstr);
121		if (err != 0 && err != ENOENT)
122			break;
123
124		if (err == ENOENT) {
125			/* Check for a received value. */
126			err = zap_lookup(mos, dsl_dir_phys(dd)->dd_props_zapobj,
127			    recvdstr, intsz, numints, buf);
128			if (err != ENOENT) {
129				if (setpoint != NULL && err == 0) {
130					if (inheriting) {
131						dsl_dir_name(dd, setpoint);
132					} else {
133						(void) strcpy(setpoint,
134						    ZPROP_SOURCE_VAL_RECVD);
135					}
136				}
137				break;
138			}
139		}
140
141		/*
142		 * If we found an explicit inheritance entry, err is zero even
143		 * though we haven't yet found the value, so reinitializing err
144		 * at the end of the loop (instead of at the beginning) ensures
145		 * that err has a valid post-loop value.
146		 */
147		err = SET_ERROR(ENOENT);
148	}
149
150	if (err == ENOENT)
151		err = dodefault(prop, intsz, numints, buf);
152
153	strfree(inheritstr);
154	strfree(recvdstr);
155
156	return (err);
157}
158
159int
160dsl_prop_get_ds(dsl_dataset_t *ds, const char *propname,
161    int intsz, int numints, void *buf, char *setpoint)
162{
163	zfs_prop_t prop = zfs_name_to_prop(propname);
164	boolean_t inheritable;
165	uint64_t zapobj;
166
167	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
168	inheritable = (prop == ZPROP_INVAL || zfs_prop_inheritable(prop));
169	zapobj = dsl_dataset_phys(ds)->ds_props_obj;
170
171	if (zapobj != 0) {
172		objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
173		int err;
174
175		ASSERT(ds->ds_is_snapshot);
176
177		/* Check for a local value. */
178		err = zap_lookup(mos, zapobj, propname, intsz, numints, buf);
179		if (err != ENOENT) {
180			if (setpoint != NULL && err == 0)
181				dsl_dataset_name(ds, setpoint);
182			return (err);
183		}
184
185		/*
186		 * Skip the check for a received value if there is an explicit
187		 * inheritance entry.
188		 */
189		if (inheritable) {
190			char *inheritstr = kmem_asprintf("%s%s", propname,
191			    ZPROP_INHERIT_SUFFIX);
192			err = zap_contains(mos, zapobj, inheritstr);
193			strfree(inheritstr);
194			if (err != 0 && err != ENOENT)
195				return (err);
196		}
197
198		if (err == ENOENT) {
199			/* Check for a received value. */
200			char *recvdstr = kmem_asprintf("%s%s", propname,
201			    ZPROP_RECVD_SUFFIX);
202			err = zap_lookup(mos, zapobj, recvdstr,
203			    intsz, numints, buf);
204			strfree(recvdstr);
205			if (err != ENOENT) {
206				if (setpoint != NULL && err == 0)
207					(void) strcpy(setpoint,
208					    ZPROP_SOURCE_VAL_RECVD);
209				return (err);
210			}
211		}
212	}
213
214	return (dsl_prop_get_dd(ds->ds_dir, propname,
215	    intsz, numints, buf, setpoint, ds->ds_is_snapshot));
216}
217
218static dsl_prop_record_t *
219dsl_prop_record_find(dsl_dir_t *dd, const char *propname)
220{
221	dsl_prop_record_t *pr = NULL;
222
223	ASSERT(MUTEX_HELD(&dd->dd_lock));
224
225	for (pr = list_head(&dd->dd_props);
226	    pr != NULL; pr = list_next(&dd->dd_props, pr)) {
227		if (strcmp(pr->pr_propname, propname) == 0)
228			break;
229	}
230
231	return (pr);
232}
233
234static dsl_prop_record_t *
235dsl_prop_record_create(dsl_dir_t *dd, const char *propname)
236{
237	dsl_prop_record_t *pr;
238
239	ASSERT(MUTEX_HELD(&dd->dd_lock));
240
241	pr = kmem_alloc(sizeof (dsl_prop_record_t), KM_SLEEP);
242	pr->pr_propname = spa_strdup(propname);
243	list_create(&pr->pr_cbs, sizeof (dsl_prop_cb_record_t),
244	    offsetof(dsl_prop_cb_record_t, cbr_pr_node));
245	list_insert_head(&dd->dd_props, pr);
246
247	return (pr);
248}
249
250void
251dsl_prop_init(dsl_dir_t *dd)
252{
253	list_create(&dd->dd_props, sizeof (dsl_prop_record_t),
254	    offsetof(dsl_prop_record_t, pr_node));
255}
256
257void
258dsl_prop_fini(dsl_dir_t *dd)
259{
260	dsl_prop_record_t *pr;
261
262	while ((pr = list_remove_head(&dd->dd_props)) != NULL) {
263		list_destroy(&pr->pr_cbs);
264		strfree((char *)pr->pr_propname);
265		kmem_free(pr, sizeof (dsl_prop_record_t));
266	}
267	list_destroy(&dd->dd_props);
268}
269
270/*
271 * Register interest in the named property.  We'll call the callback
272 * once to notify it of the current property value, and again each time
273 * the property changes, until this callback is unregistered.
274 *
275 * Return 0 on success, errno if the prop is not an integer value.
276 */
277int
278dsl_prop_register(dsl_dataset_t *ds, const char *propname,
279    dsl_prop_changed_cb_t *callback, void *cbarg)
280{
281	dsl_dir_t *dd = ds->ds_dir;
282	dsl_pool_t *dp = dd->dd_pool;
283	uint64_t value;
284	dsl_prop_record_t *pr;
285	dsl_prop_cb_record_t *cbr;
286	int err;
287
288	ASSERT(dsl_pool_config_held(dp));
289
290	err = dsl_prop_get_int_ds(ds, propname, &value);
291	if (err != 0)
292		return (err);
293
294	cbr = kmem_alloc(sizeof (dsl_prop_cb_record_t), KM_SLEEP);
295	cbr->cbr_ds = ds;
296	cbr->cbr_func = callback;
297	cbr->cbr_arg = cbarg;
298
299	mutex_enter(&dd->dd_lock);
300	pr = dsl_prop_record_find(dd, propname);
301	if (pr == NULL)
302		pr = dsl_prop_record_create(dd, propname);
303	cbr->cbr_pr = pr;
304	list_insert_head(&pr->pr_cbs, cbr);
305	list_insert_head(&ds->ds_prop_cbs, cbr);
306	mutex_exit(&dd->dd_lock);
307
308	cbr->cbr_func(cbr->cbr_arg, value);
309	return (0);
310}
311
312int
313dsl_prop_get(const char *dsname, const char *propname,
314    int intsz, int numints, void *buf, char *setpoint)
315{
316	objset_t *os;
317	int error;
318
319	error = dmu_objset_hold(dsname, FTAG, &os);
320	if (error != 0)
321		return (error);
322
323	error = dsl_prop_get_ds(dmu_objset_ds(os), propname,
324	    intsz, numints, buf, setpoint);
325
326	dmu_objset_rele(os, FTAG);
327	return (error);
328}
329
330/*
331 * Get the current property value.  It may have changed by the time this
332 * function returns, so it is NOT safe to follow up with
333 * dsl_prop_register() and assume that the value has not changed in
334 * between.
335 *
336 * Return 0 on success, ENOENT if ddname is invalid.
337 */
338int
339dsl_prop_get_integer(const char *ddname, const char *propname,
340    uint64_t *valuep, char *setpoint)
341{
342	return (dsl_prop_get(ddname, propname, 8, 1, valuep, setpoint));
343}
344
345int
346dsl_prop_get_int_ds(dsl_dataset_t *ds, const char *propname,
347    uint64_t *valuep)
348{
349	return (dsl_prop_get_ds(ds, propname, 8, 1, valuep, NULL));
350}
351
352/*
353 * Predict the effective value of the given special property if it were set with
354 * the given value and source. This is not a general purpose function. It exists
355 * only to handle the special requirements of the quota and reservation
356 * properties. The fact that these properties are non-inheritable greatly
357 * simplifies the prediction logic.
358 *
359 * Returns 0 on success, a positive error code on failure, or -1 if called with
360 * a property not handled by this function.
361 */
362int
363dsl_prop_predict(dsl_dir_t *dd, const char *propname,
364    zprop_source_t source, uint64_t value, uint64_t *newvalp)
365{
366	zfs_prop_t prop = zfs_name_to_prop(propname);
367	objset_t *mos;
368	uint64_t zapobj;
369	uint64_t version;
370	char *recvdstr;
371	int err = 0;
372
373	switch (prop) {
374	case ZFS_PROP_QUOTA:
375	case ZFS_PROP_RESERVATION:
376	case ZFS_PROP_REFQUOTA:
377	case ZFS_PROP_REFRESERVATION:
378		break;
379	default:
380		return (-1);
381	}
382
383	mos = dd->dd_pool->dp_meta_objset;
384	zapobj = dsl_dir_phys(dd)->dd_props_zapobj;
385	recvdstr = kmem_asprintf("%s%s", propname, ZPROP_RECVD_SUFFIX);
386
387	version = spa_version(dd->dd_pool->dp_spa);
388	if (version < SPA_VERSION_RECVD_PROPS) {
389		if (source & ZPROP_SRC_NONE)
390			source = ZPROP_SRC_NONE;
391		else if (source & ZPROP_SRC_RECEIVED)
392			source = ZPROP_SRC_LOCAL;
393	}
394
395	switch (source) {
396	case ZPROP_SRC_NONE:
397		/* Revert to the received value, if any. */
398		err = zap_lookup(mos, zapobj, recvdstr, 8, 1, newvalp);
399		if (err == ENOENT)
400			*newvalp = 0;
401		break;
402	case ZPROP_SRC_LOCAL:
403		*newvalp = value;
404		break;
405	case ZPROP_SRC_RECEIVED:
406		/*
407		 * If there's no local setting, then the new received value will
408		 * be the effective value.
409		 */
410		err = zap_lookup(mos, zapobj, propname, 8, 1, newvalp);
411		if (err == ENOENT)
412			*newvalp = value;
413		break;
414	case (ZPROP_SRC_NONE | ZPROP_SRC_RECEIVED):
415		/*
416		 * We're clearing the received value, so the local setting (if
417		 * it exists) remains the effective value.
418		 */
419		err = zap_lookup(mos, zapobj, propname, 8, 1, newvalp);
420		if (err == ENOENT)
421			*newvalp = 0;
422		break;
423	default:
424		panic("unexpected property source: %d", source);
425	}
426
427	strfree(recvdstr);
428
429	if (err == ENOENT)
430		return (0);
431
432	return (err);
433}
434
435/*
436 * Unregister all callbacks that are registered with the
437 * given callback argument.
438 */
439void
440dsl_prop_unregister_all(dsl_dataset_t *ds, void *cbarg)
441{
442	dsl_prop_cb_record_t *cbr, *next_cbr;
443
444	dsl_dir_t *dd = ds->ds_dir;
445
446	mutex_enter(&dd->dd_lock);
447	next_cbr = list_head(&ds->ds_prop_cbs);
448	while (next_cbr != NULL) {
449		cbr = next_cbr;
450		next_cbr = list_next(&ds->ds_prop_cbs, cbr);
451		if (cbr->cbr_arg == cbarg) {
452			list_remove(&ds->ds_prop_cbs, cbr);
453			list_remove(&cbr->cbr_pr->pr_cbs, cbr);
454			kmem_free(cbr, sizeof (dsl_prop_cb_record_t));
455		}
456	}
457	mutex_exit(&dd->dd_lock);
458}
459
460boolean_t
461dsl_prop_hascb(dsl_dataset_t *ds)
462{
463	return (!list_is_empty(&ds->ds_prop_cbs));
464}
465
466/* ARGSUSED */
467static int
468dsl_prop_notify_all_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
469{
470	dsl_dir_t *dd = ds->ds_dir;
471	dsl_prop_record_t *pr;
472	dsl_prop_cb_record_t *cbr;
473
474	mutex_enter(&dd->dd_lock);
475	for (pr = list_head(&dd->dd_props);
476	    pr; pr = list_next(&dd->dd_props, pr)) {
477		for (cbr = list_head(&pr->pr_cbs); cbr;
478		    cbr = list_next(&pr->pr_cbs, cbr)) {
479			uint64_t value;
480
481			/*
482			 * Callback entries do not have holds on their
483			 * datasets so that datasets with registered
484			 * callbacks are still eligible for eviction.
485			 * Unlike operations to update properties on a
486			 * single dataset, we are performing a recursive
487			 * descent of related head datasets.  The caller
488			 * of this function only has a dataset hold on
489			 * the passed in head dataset, not the snapshots
490			 * associated with this dataset.  Without a hold,
491			 * the dataset pointer within callback records
492			 * for snapshots can be invalidated by eviction
493			 * at any time.
494			 *
495			 * Use dsl_dataset_try_add_ref() to verify
496			 * that the dataset for a snapshot has not
497			 * begun eviction processing and to prevent
498			 * eviction from occurring for the duration of
499			 * the callback.  If the hold attempt fails,
500			 * this object is already being evicted and the
501			 * callback can be safely ignored.
502			 */
503			if (ds != cbr->cbr_ds &&
504			    !dsl_dataset_try_add_ref(dp, cbr->cbr_ds, FTAG))
505				continue;
506
507			if (dsl_prop_get_ds(cbr->cbr_ds,
508			    cbr->cbr_pr->pr_propname, sizeof (value), 1,
509			    &value, NULL) == 0)
510				cbr->cbr_func(cbr->cbr_arg, value);
511
512			if (ds != cbr->cbr_ds)
513				dsl_dataset_rele(cbr->cbr_ds, FTAG);
514		}
515	}
516	mutex_exit(&dd->dd_lock);
517
518	return (0);
519}
520
521/*
522 * Update all property values for ddobj & its descendants.  This is used
523 * when renaming the dir.
524 */
525void
526dsl_prop_notify_all(dsl_dir_t *dd)
527{
528	dsl_pool_t *dp = dd->dd_pool;
529	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
530	(void) dmu_objset_find_dp(dp, dd->dd_object, dsl_prop_notify_all_cb,
531	    NULL, DS_FIND_CHILDREN);
532}
533
534static void
535dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj,
536    const char *propname, uint64_t value, int first)
537{
538	dsl_dir_t *dd;
539	dsl_prop_record_t *pr;
540	dsl_prop_cb_record_t *cbr;
541	objset_t *mos = dp->dp_meta_objset;
542	zap_cursor_t zc;
543	zap_attribute_t *za;
544	int err;
545
546	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
547	err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
548	if (err)
549		return;
550
551	if (!first) {
552		/*
553		 * If the prop is set here, then this change is not
554		 * being inherited here or below; stop the recursion.
555		 */
556		err = zap_contains(mos, dsl_dir_phys(dd)->dd_props_zapobj,
557		    propname);
558		if (err == 0) {
559			dsl_dir_rele(dd, FTAG);
560			return;
561		}
562		ASSERT3U(err, ==, ENOENT);
563	}
564
565	mutex_enter(&dd->dd_lock);
566	pr = dsl_prop_record_find(dd, propname);
567	if (pr != NULL) {
568		for (cbr = list_head(&pr->pr_cbs); cbr;
569		    cbr = list_next(&pr->pr_cbs, cbr)) {
570			uint64_t propobj;
571
572			/*
573			 * cbr->cbr_ds may be invalidated due to eviction,
574			 * requiring the use of dsl_dataset_try_add_ref().
575			 * See comment block in dsl_prop_notify_all_cb()
576			 * for details.
577			 */
578			if (!dsl_dataset_try_add_ref(dp, cbr->cbr_ds, FTAG))
579				continue;
580
581			propobj = dsl_dataset_phys(cbr->cbr_ds)->ds_props_obj;
582
583			/*
584			 * If the property is not set on this ds, then it is
585			 * inherited here; call the callback.
586			 */
587			if (propobj == 0 ||
588			    zap_contains(mos, propobj, propname) != 0)
589				cbr->cbr_func(cbr->cbr_arg, value);
590
591			dsl_dataset_rele(cbr->cbr_ds, FTAG);
592		}
593	}
594	mutex_exit(&dd->dd_lock);
595
596	za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
597	for (zap_cursor_init(&zc, mos,
598	    dsl_dir_phys(dd)->dd_child_dir_zapobj);
599	    zap_cursor_retrieve(&zc, za) == 0;
600	    zap_cursor_advance(&zc)) {
601		dsl_prop_changed_notify(dp, za->za_first_integer,
602		    propname, value, FALSE);
603	}
604	kmem_free(za, sizeof (zap_attribute_t));
605	zap_cursor_fini(&zc);
606	dsl_dir_rele(dd, FTAG);
607}
608
609void
610dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
611    zprop_source_t source, int intsz, int numints, const void *value,
612    dmu_tx_t *tx)
613{
614	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
615	uint64_t zapobj, intval, dummy;
616	int isint;
617	char valbuf[32];
618	const char *valstr = NULL;
619	char *inheritstr;
620	char *recvdstr;
621	char *tbuf = NULL;
622	int err;
623	uint64_t version = spa_version(ds->ds_dir->dd_pool->dp_spa);
624
625	isint = (dodefault(zfs_name_to_prop(propname), 8, 1, &intval) == 0);
626
627	if (ds->ds_is_snapshot) {
628		ASSERT(version >= SPA_VERSION_SNAP_PROPS);
629		if (dsl_dataset_phys(ds)->ds_props_obj == 0) {
630			dmu_buf_will_dirty(ds->ds_dbuf, tx);
631			dsl_dataset_phys(ds)->ds_props_obj =
632			    zap_create(mos,
633			    DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx);
634		}
635		zapobj = dsl_dataset_phys(ds)->ds_props_obj;
636	} else {
637		zapobj = dsl_dir_phys(ds->ds_dir)->dd_props_zapobj;
638	}
639
640	if (version < SPA_VERSION_RECVD_PROPS) {
641		if (source & ZPROP_SRC_NONE)
642			source = ZPROP_SRC_NONE;
643		else if (source & ZPROP_SRC_RECEIVED)
644			source = ZPROP_SRC_LOCAL;
645	}
646
647	inheritstr = kmem_asprintf("%s%s", propname, ZPROP_INHERIT_SUFFIX);
648	recvdstr = kmem_asprintf("%s%s", propname, ZPROP_RECVD_SUFFIX);
649
650	switch (source) {
651	case ZPROP_SRC_NONE:
652		/*
653		 * revert to received value, if any (inherit -S)
654		 * - remove propname
655		 * - remove propname$inherit
656		 */
657		err = zap_remove(mos, zapobj, propname, tx);
658		ASSERT(err == 0 || err == ENOENT);
659		err = zap_remove(mos, zapobj, inheritstr, tx);
660		ASSERT(err == 0 || err == ENOENT);
661		break;
662	case ZPROP_SRC_LOCAL:
663		/*
664		 * remove propname$inherit
665		 * set propname -> value
666		 */
667		err = zap_remove(mos, zapobj, inheritstr, tx);
668		ASSERT(err == 0 || err == ENOENT);
669		VERIFY0(zap_update(mos, zapobj, propname,
670		    intsz, numints, value, tx));
671		break;
672	case ZPROP_SRC_INHERITED:
673		/*
674		 * explicitly inherit
675		 * - remove propname
676		 * - set propname$inherit
677		 */
678		err = zap_remove(mos, zapobj, propname, tx);
679		ASSERT(err == 0 || err == ENOENT);
680		if (version >= SPA_VERSION_RECVD_PROPS &&
681		    dsl_prop_get_int_ds(ds, ZPROP_HAS_RECVD, &dummy) == 0) {
682			dummy = 0;
683			VERIFY0(zap_update(mos, zapobj, inheritstr,
684			    8, 1, &dummy, tx));
685		}
686		break;
687	case ZPROP_SRC_RECEIVED:
688		/*
689		 * set propname$recvd -> value
690		 */
691		err = zap_update(mos, zapobj, recvdstr,
692		    intsz, numints, value, tx);
693		ASSERT(err == 0);
694		break;
695	case (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED):
696		/*
697		 * clear local and received settings
698		 * - remove propname
699		 * - remove propname$inherit
700		 * - remove propname$recvd
701		 */
702		err = zap_remove(mos, zapobj, propname, tx);
703		ASSERT(err == 0 || err == ENOENT);
704		err = zap_remove(mos, zapobj, inheritstr, tx);
705		ASSERT(err == 0 || err == ENOENT);
706		/* FALLTHRU */
707	case (ZPROP_SRC_NONE | ZPROP_SRC_RECEIVED):
708		/*
709		 * remove propname$recvd
710		 */
711		err = zap_remove(mos, zapobj, recvdstr, tx);
712		ASSERT(err == 0 || err == ENOENT);
713		break;
714	default:
715		cmn_err(CE_PANIC, "unexpected property source: %d", source);
716	}
717
718	strfree(inheritstr);
719	strfree(recvdstr);
720
721	if (isint) {
722		VERIFY0(dsl_prop_get_int_ds(ds, propname, &intval));
723
724		if (ds->ds_is_snapshot) {
725			dsl_prop_cb_record_t *cbr;
726			/*
727			 * It's a snapshot; nothing can inherit this
728			 * property, so just look for callbacks on this
729			 * ds here.
730			 */
731			mutex_enter(&ds->ds_dir->dd_lock);
732			for (cbr = list_head(&ds->ds_prop_cbs); cbr;
733			    cbr = list_next(&ds->ds_prop_cbs, cbr)) {
734				if (strcmp(cbr->cbr_pr->pr_propname,
735				    propname) == 0)
736					cbr->cbr_func(cbr->cbr_arg, intval);
737			}
738			mutex_exit(&ds->ds_dir->dd_lock);
739		} else {
740			dsl_prop_changed_notify(ds->ds_dir->dd_pool,
741			    ds->ds_dir->dd_object, propname, intval, TRUE);
742		}
743
744		(void) snprintf(valbuf, sizeof (valbuf),
745		    "%lld", (longlong_t)intval);
746		valstr = valbuf;
747	} else {
748		if (source == ZPROP_SRC_LOCAL) {
749			valstr = value;
750		} else {
751			tbuf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
752			if (dsl_prop_get_ds(ds, propname, 1,
753			    ZAP_MAXVALUELEN, tbuf, NULL) == 0)
754				valstr = tbuf;
755		}
756	}
757
758	spa_history_log_internal_ds(ds, (source == ZPROP_SRC_NONE ||
759	    source == ZPROP_SRC_INHERITED) ? "inherit" : "set", tx,
760	    "%s=%s", propname, (valstr == NULL ? "" : valstr));
761
762	if (tbuf != NULL)
763		kmem_free(tbuf, ZAP_MAXVALUELEN);
764}
765
766int
767dsl_prop_set_int(const char *dsname, const char *propname,
768    zprop_source_t source, uint64_t value)
769{
770	nvlist_t *nvl = fnvlist_alloc();
771	int error;
772
773	fnvlist_add_uint64(nvl, propname, value);
774	error = dsl_props_set(dsname, source, nvl);
775	fnvlist_free(nvl);
776	return (error);
777}
778
779int
780dsl_prop_set_string(const char *dsname, const char *propname,
781    zprop_source_t source, const char *value)
782{
783	nvlist_t *nvl = fnvlist_alloc();
784	int error;
785
786	fnvlist_add_string(nvl, propname, value);
787	error = dsl_props_set(dsname, source, nvl);
788	fnvlist_free(nvl);
789	return (error);
790}
791
792int
793dsl_prop_inherit(const char *dsname, const char *propname,
794    zprop_source_t source)
795{
796	nvlist_t *nvl = fnvlist_alloc();
797	int error;
798
799	fnvlist_add_boolean(nvl, propname);
800	error = dsl_props_set(dsname, source, nvl);
801	fnvlist_free(nvl);
802	return (error);
803}
804
805typedef struct dsl_props_set_arg {
806	const char *dpsa_dsname;
807	zprop_source_t dpsa_source;
808	nvlist_t *dpsa_props;
809} dsl_props_set_arg_t;
810
811static int
812dsl_props_set_check(void *arg, dmu_tx_t *tx)
813{
814	dsl_props_set_arg_t *dpsa = arg;
815	dsl_pool_t *dp = dmu_tx_pool(tx);
816	dsl_dataset_t *ds;
817	uint64_t version;
818	nvpair_t *elem = NULL;
819	int err;
820
821	err = dsl_dataset_hold(dp, dpsa->dpsa_dsname, FTAG, &ds);
822	if (err != 0)
823		return (err);
824
825	version = spa_version(ds->ds_dir->dd_pool->dp_spa);
826	while ((elem = nvlist_next_nvpair(dpsa->dpsa_props, elem)) != NULL) {
827		if (strlen(nvpair_name(elem)) >= ZAP_MAXNAMELEN) {
828			dsl_dataset_rele(ds, FTAG);
829			return (SET_ERROR(ENAMETOOLONG));
830		}
831		if (nvpair_type(elem) == DATA_TYPE_STRING) {
832			char *valstr = fnvpair_value_string(elem);
833			if (strlen(valstr) >= (version <
834			    SPA_VERSION_STMF_PROP ?
835			    ZAP_OLDMAXVALUELEN : ZAP_MAXVALUELEN)) {
836				dsl_dataset_rele(ds, FTAG);
837				return (E2BIG);
838			}
839		}
840	}
841
842	if (ds->ds_is_snapshot && version < SPA_VERSION_SNAP_PROPS) {
843		dsl_dataset_rele(ds, FTAG);
844		return (SET_ERROR(ENOTSUP));
845	}
846	dsl_dataset_rele(ds, FTAG);
847	return (0);
848}
849
850void
851dsl_props_set_sync_impl(dsl_dataset_t *ds, zprop_source_t source,
852    nvlist_t *props, dmu_tx_t *tx)
853{
854	nvpair_t *elem = NULL;
855
856	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
857		nvpair_t *pair = elem;
858		const char *name = nvpair_name(pair);
859
860		if (nvpair_type(pair) == DATA_TYPE_NVLIST) {
861			/*
862			 * This usually happens when we reuse the nvlist_t data
863			 * returned by the counterpart dsl_prop_get_all_impl().
864			 * For instance we do this to restore the original
865			 * received properties when an error occurs in the
866			 * zfs_ioc_recv() codepath.
867			 */
868			nvlist_t *attrs = fnvpair_value_nvlist(pair);
869			pair = fnvlist_lookup_nvpair(attrs, ZPROP_VALUE);
870		}
871
872		if (nvpair_type(pair) == DATA_TYPE_STRING) {
873			const char *value = fnvpair_value_string(pair);
874			dsl_prop_set_sync_impl(ds, name,
875			    source, 1, strlen(value) + 1, value, tx);
876		} else if (nvpair_type(pair) == DATA_TYPE_UINT64) {
877			uint64_t intval = fnvpair_value_uint64(pair);
878			dsl_prop_set_sync_impl(ds, name,
879			    source, sizeof (intval), 1, &intval, tx);
880		} else if (nvpair_type(pair) == DATA_TYPE_BOOLEAN) {
881			dsl_prop_set_sync_impl(ds, name,
882			    source, 0, 0, NULL, tx);
883		} else {
884			panic("invalid nvpair type");
885		}
886	}
887}
888
889static void
890dsl_props_set_sync(void *arg, dmu_tx_t *tx)
891{
892	dsl_props_set_arg_t *dpsa = arg;
893	dsl_pool_t *dp = dmu_tx_pool(tx);
894	dsl_dataset_t *ds;
895
896	VERIFY0(dsl_dataset_hold(dp, dpsa->dpsa_dsname, FTAG, &ds));
897	dsl_props_set_sync_impl(ds, dpsa->dpsa_source, dpsa->dpsa_props, tx);
898	dsl_dataset_rele(ds, FTAG);
899}
900
901/*
902 * All-or-nothing; if any prop can't be set, nothing will be modified.
903 */
904int
905dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *props)
906{
907	dsl_props_set_arg_t dpsa;
908	int nblks = 0;
909
910	dpsa.dpsa_dsname = dsname;
911	dpsa.dpsa_source = source;
912	dpsa.dpsa_props = props;
913
914	/*
915	 * If the source includes NONE, then we will only be removing entries
916	 * from the ZAP object.  In that case don't check for ENOSPC.
917	 */
918	if ((source & ZPROP_SRC_NONE) == 0)
919		nblks = 2 * fnvlist_num_pairs(props);
920
921	return (dsl_sync_task(dsname, dsl_props_set_check, dsl_props_set_sync,
922	    &dpsa, nblks, ZFS_SPACE_CHECK_RESERVED));
923}
924
925typedef enum dsl_prop_getflags {
926	DSL_PROP_GET_INHERITING = 0x1,	/* searching parent of target ds */
927	DSL_PROP_GET_SNAPSHOT = 0x2,	/* snapshot dataset */
928	DSL_PROP_GET_LOCAL = 0x4,	/* local properties */
929	DSL_PROP_GET_RECEIVED = 0x8,	/* received properties */
930} dsl_prop_getflags_t;
931
932static int
933dsl_prop_get_all_impl(objset_t *mos, uint64_t propobj,
934    const char *setpoint, dsl_prop_getflags_t flags, nvlist_t *nv)
935{
936	zap_cursor_t zc;
937	zap_attribute_t za;
938	int err = 0;
939
940	for (zap_cursor_init(&zc, mos, propobj);
941	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
942	    zap_cursor_advance(&zc)) {
943		nvlist_t *propval;
944		zfs_prop_t prop;
945		char buf[ZAP_MAXNAMELEN];
946		char *valstr;
947		const char *suffix;
948		const char *propname;
949		const char *source;
950
951		suffix = strchr(za.za_name, '$');
952
953		if (suffix == NULL) {
954			/*
955			 * Skip local properties if we only want received
956			 * properties.
957			 */
958			if (flags & DSL_PROP_GET_RECEIVED)
959				continue;
960
961			propname = za.za_name;
962			source = setpoint;
963		} else if (strcmp(suffix, ZPROP_INHERIT_SUFFIX) == 0) {
964			/* Skip explicitly inherited entries. */
965			continue;
966		} else if (strcmp(suffix, ZPROP_RECVD_SUFFIX) == 0) {
967			if (flags & DSL_PROP_GET_LOCAL)
968				continue;
969
970			(void) strncpy(buf, za.za_name, (suffix - za.za_name));
971			buf[suffix - za.za_name] = '\0';
972			propname = buf;
973
974			if (!(flags & DSL_PROP_GET_RECEIVED)) {
975				/* Skip if locally overridden. */
976				err = zap_contains(mos, propobj, propname);
977				if (err == 0)
978					continue;
979				if (err != ENOENT)
980					break;
981
982				/* Skip if explicitly inherited. */
983				valstr = kmem_asprintf("%s%s", propname,
984				    ZPROP_INHERIT_SUFFIX);
985				err = zap_contains(mos, propobj, valstr);
986				strfree(valstr);
987				if (err == 0)
988					continue;
989				if (err != ENOENT)
990					break;
991			}
992
993			source = ((flags & DSL_PROP_GET_INHERITING) ?
994			    setpoint : ZPROP_SOURCE_VAL_RECVD);
995		} else {
996			/*
997			 * For backward compatibility, skip suffixes we don't
998			 * recognize.
999			 */
1000			continue;
1001		}
1002
1003		prop = zfs_name_to_prop(propname);
1004
1005		/* Skip non-inheritable properties. */
1006		if ((flags & DSL_PROP_GET_INHERITING) && prop != ZPROP_INVAL &&
1007		    !zfs_prop_inheritable(prop))
1008			continue;
1009
1010		/* Skip properties not valid for this type. */
1011		if ((flags & DSL_PROP_GET_SNAPSHOT) && prop != ZPROP_INVAL &&
1012		    !zfs_prop_valid_for_type(prop, ZFS_TYPE_SNAPSHOT))
1013			continue;
1014
1015		/* Skip properties already defined. */
1016		if (nvlist_exists(nv, propname))
1017			continue;
1018
1019		VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1020		if (za.za_integer_length == 1) {
1021			/*
1022			 * String property
1023			 */
1024			char *tmp = kmem_alloc(za.za_num_integers,
1025			    KM_SLEEP);
1026			err = zap_lookup(mos, propobj,
1027			    za.za_name, 1, za.za_num_integers, tmp);
1028			if (err != 0) {
1029				kmem_free(tmp, za.za_num_integers);
1030				break;
1031			}
1032			VERIFY(nvlist_add_string(propval, ZPROP_VALUE,
1033			    tmp) == 0);
1034			kmem_free(tmp, za.za_num_integers);
1035		} else {
1036			/*
1037			 * Integer property
1038			 */
1039			ASSERT(za.za_integer_length == 8);
1040			(void) nvlist_add_uint64(propval, ZPROP_VALUE,
1041			    za.za_first_integer);
1042		}
1043
1044		VERIFY(nvlist_add_string(propval, ZPROP_SOURCE, source) == 0);
1045		VERIFY(nvlist_add_nvlist(nv, propname, propval) == 0);
1046		nvlist_free(propval);
1047	}
1048	zap_cursor_fini(&zc);
1049	if (err == ENOENT)
1050		err = 0;
1051	return (err);
1052}
1053
1054/*
1055 * Iterate over all properties for this dataset and return them in an nvlist.
1056 */
1057static int
1058dsl_prop_get_all_ds(dsl_dataset_t *ds, nvlist_t **nvp,
1059    dsl_prop_getflags_t flags)
1060{
1061	dsl_dir_t *dd = ds->ds_dir;
1062	dsl_pool_t *dp = dd->dd_pool;
1063	objset_t *mos = dp->dp_meta_objset;
1064	int err = 0;
1065	char setpoint[ZFS_MAX_DATASET_NAME_LEN];
1066
1067	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1068
1069	if (ds->ds_is_snapshot)
1070		flags |= DSL_PROP_GET_SNAPSHOT;
1071
1072	ASSERT(dsl_pool_config_held(dp));
1073
1074	if (dsl_dataset_phys(ds)->ds_props_obj != 0) {
1075		ASSERT(flags & DSL_PROP_GET_SNAPSHOT);
1076		dsl_dataset_name(ds, setpoint);
1077		err = dsl_prop_get_all_impl(mos,
1078		    dsl_dataset_phys(ds)->ds_props_obj, setpoint, flags, *nvp);
1079		if (err)
1080			goto out;
1081	}
1082
1083	for (; dd != NULL; dd = dd->dd_parent) {
1084		if (dd != ds->ds_dir || (flags & DSL_PROP_GET_SNAPSHOT)) {
1085			if (flags & (DSL_PROP_GET_LOCAL |
1086			    DSL_PROP_GET_RECEIVED))
1087				break;
1088			flags |= DSL_PROP_GET_INHERITING;
1089		}
1090		dsl_dir_name(dd, setpoint);
1091		err = dsl_prop_get_all_impl(mos,
1092		    dsl_dir_phys(dd)->dd_props_zapobj, setpoint, flags, *nvp);
1093		if (err)
1094			break;
1095	}
1096
1097out:
1098	return (err);
1099}
1100
1101boolean_t
1102dsl_prop_get_hasrecvd(const char *dsname)
1103{
1104	uint64_t dummy;
1105
1106	return (0 ==
1107	    dsl_prop_get_integer(dsname, ZPROP_HAS_RECVD, &dummy, NULL));
1108}
1109
1110static int
1111dsl_prop_set_hasrecvd_impl(const char *dsname, zprop_source_t source)
1112{
1113	uint64_t version;
1114	spa_t *spa;
1115	int error = 0;
1116
1117	VERIFY0(spa_open(dsname, &spa, FTAG));
1118	version = spa_version(spa);
1119	spa_close(spa, FTAG);
1120
1121	if (version >= SPA_VERSION_RECVD_PROPS)
1122		error = dsl_prop_set_int(dsname, ZPROP_HAS_RECVD, source, 0);
1123	return (error);
1124}
1125
1126/*
1127 * Call after successfully receiving properties to ensure that only the first
1128 * receive on or after SPA_VERSION_RECVD_PROPS blows away local properties.
1129 */
1130int
1131dsl_prop_set_hasrecvd(const char *dsname)
1132{
1133	int error = 0;
1134	if (!dsl_prop_get_hasrecvd(dsname))
1135		error = dsl_prop_set_hasrecvd_impl(dsname, ZPROP_SRC_LOCAL);
1136	return (error);
1137}
1138
1139void
1140dsl_prop_unset_hasrecvd(const char *dsname)
1141{
1142	VERIFY0(dsl_prop_set_hasrecvd_impl(dsname, ZPROP_SRC_NONE));
1143}
1144
1145int
1146dsl_prop_get_all(objset_t *os, nvlist_t **nvp)
1147{
1148	return (dsl_prop_get_all_ds(os->os_dsl_dataset, nvp, 0));
1149}
1150
1151int
1152dsl_prop_get_received(const char *dsname, nvlist_t **nvp)
1153{
1154	objset_t *os;
1155	int error;
1156
1157	/*
1158	 * Received properties are not distinguishable from local properties
1159	 * until the dataset has received properties on or after
1160	 * SPA_VERSION_RECVD_PROPS.
1161	 */
1162	dsl_prop_getflags_t flags = (dsl_prop_get_hasrecvd(dsname) ?
1163	    DSL_PROP_GET_RECEIVED : DSL_PROP_GET_LOCAL);
1164
1165	error = dmu_objset_hold(dsname, FTAG, &os);
1166	if (error != 0)
1167		return (error);
1168	error = dsl_prop_get_all_ds(os->os_dsl_dataset, nvp, flags);
1169	dmu_objset_rele(os, FTAG);
1170	return (error);
1171}
1172
1173void
1174dsl_prop_nvlist_add_uint64(nvlist_t *nv, zfs_prop_t prop, uint64_t value)
1175{
1176	nvlist_t *propval;
1177	const char *propname = zfs_prop_to_name(prop);
1178	uint64_t default_value;
1179
1180	if (nvlist_lookup_nvlist(nv, propname, &propval) == 0) {
1181		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, value) == 0);
1182		return;
1183	}
1184
1185	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1186	VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, value) == 0);
1187	/* Indicate the default source if we can. */
1188	if (dodefault(prop, 8, 1, &default_value) == 0 &&
1189	    value == default_value) {
1190		VERIFY(nvlist_add_string(propval, ZPROP_SOURCE, "") == 0);
1191	}
1192	VERIFY(nvlist_add_nvlist(nv, propname, propval) == 0);
1193	nvlist_free(propval);
1194}
1195
1196void
1197dsl_prop_nvlist_add_string(nvlist_t *nv, zfs_prop_t prop, const char *value)
1198{
1199	nvlist_t *propval;
1200	const char *propname = zfs_prop_to_name(prop);
1201
1202	if (nvlist_lookup_nvlist(nv, propname, &propval) == 0) {
1203		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, value) == 0);
1204		return;
1205	}
1206
1207	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1208	VERIFY(nvlist_add_string(propval, ZPROP_VALUE, value) == 0);
1209	VERIFY(nvlist_add_nvlist(nv, propname, propval) == 0);
1210	nvlist_free(propval);
1211}
1212