xref: /freebsd/sys/contrib/openzfs/module/zfs/zvol.c (revision 6c05f3a74f30934ee60919cc97e16ec69b542b06)
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 https://opensource.org/licenses/CDDL-1.0.
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) 2008-2010 Lawrence Livermore National Security, LLC.
23  * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
24  * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
25  * LLNL-CODE-403049.
26  *
27  * ZFS volume emulation driver.
28  *
29  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
30  * Volumes are accessed through the symbolic links named:
31  *
32  * /dev/<pool_name>/<dataset_name>
33  *
34  * Volumes are persistent through reboot and module load.  No user command
35  * needs to be run before opening and using a device.
36  *
37  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
38  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
39  * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
40  * Copyright (c) 2024, Klara, Inc.
41  */
42 
43 /*
44  * Note on locking of zvol state structures.
45  *
46  * These structures are used to maintain internal state used to emulate block
47  * devices on top of zvols. In particular, management of device minor number
48  * operations - create, remove, rename, and set_snapdev - involves access to
49  * these structures. The zvol_state_lock is primarily used to protect the
50  * zvol_state_list. The zv->zv_state_lock is used to protect the contents
51  * of the zvol_state_t structures, as well as to make sure that when the
52  * time comes to remove the structure from the list, it is not in use, and
53  * therefore, it can be taken off zvol_state_list and freed.
54  *
55  * The zv_suspend_lock was introduced to allow for suspending I/O to a zvol,
56  * e.g. for the duration of receive and rollback operations. This lock can be
57  * held for significant periods of time. Given that it is undesirable to hold
58  * mutexes for long periods of time, the following lock ordering applies:
59  * - take zvol_state_lock if necessary, to protect zvol_state_list
60  * - take zv_suspend_lock if necessary, by the code path in question
61  * - take zv_state_lock to protect zvol_state_t
62  *
63  * The minor operations are issued to spa->spa_zvol_taskq queues, that are
64  * single-threaded (to preserve order of minor operations), and are executed
65  * through the zvol_task_cb that dispatches the specific operations. Therefore,
66  * these operations are serialized per pool. Consequently, we can be certain
67  * that for a given zvol, there is only one operation at a time in progress.
68  * That is why one can be sure that first, zvol_state_t for a given zvol is
69  * allocated and placed on zvol_state_list, and then other minor operations
70  * for this zvol are going to proceed in the order of issue.
71  *
72  */
73 
74 #include <sys/dataset_kstats.h>
75 #include <sys/dbuf.h>
76 #include <sys/dmu_traverse.h>
77 #include <sys/dsl_dataset.h>
78 #include <sys/dsl_prop.h>
79 #include <sys/dsl_dir.h>
80 #include <sys/zap.h>
81 #include <sys/zfeature.h>
82 #include <sys/zil_impl.h>
83 #include <sys/dmu_tx.h>
84 #include <sys/zio.h>
85 #include <sys/zfs_rlock.h>
86 #include <sys/spa_impl.h>
87 #include <sys/zvol.h>
88 #include <sys/zvol_impl.h>
89 
90 unsigned int zvol_inhibit_dev = 0;
91 unsigned int zvol_volmode = ZFS_VOLMODE_GEOM;
92 
93 struct hlist_head *zvol_htable;
94 static list_t zvol_state_list;
95 krwlock_t zvol_state_lock;
96 extern int zfs_bclone_wait_dirty;
97 
98 typedef enum {
99 	ZVOL_ASYNC_REMOVE_MINORS,
100 	ZVOL_ASYNC_RENAME_MINORS,
101 	ZVOL_ASYNC_SET_SNAPDEV,
102 	ZVOL_ASYNC_SET_VOLMODE,
103 	ZVOL_ASYNC_MAX
104 } zvol_async_op_t;
105 
106 typedef struct {
107 	zvol_async_op_t op;
108 	char name1[MAXNAMELEN];
109 	char name2[MAXNAMELEN];
110 	uint64_t value;
111 } zvol_task_t;
112 
113 uint64_t
114 zvol_name_hash(const char *name)
115 {
116 	uint64_t crc = -1ULL;
117 	ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
118 	for (const uint8_t *p = (const uint8_t *)name; *p != 0; p++)
119 		crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (*p)) & 0xFF];
120 	return (crc);
121 }
122 
123 /*
124  * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
125  * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
126  * return (NULL) without the taking locks. The zv_suspend_lock is always taken
127  * before zv_state_lock. The mode argument indicates the mode (including none)
128  * for zv_suspend_lock to be taken.
129  */
130 zvol_state_t *
131 zvol_find_by_name_hash(const char *name, uint64_t hash, int mode)
132 {
133 	zvol_state_t *zv;
134 	struct hlist_node *p = NULL;
135 
136 	rw_enter(&zvol_state_lock, RW_READER);
137 	hlist_for_each(p, ZVOL_HT_HEAD(hash)) {
138 		zv = hlist_entry(p, zvol_state_t, zv_hlink);
139 		mutex_enter(&zv->zv_state_lock);
140 		if (zv->zv_hash == hash && strcmp(zv->zv_name, name) == 0) {
141 			/*
142 			 * this is the right zvol, take the locks in the
143 			 * right order
144 			 */
145 			if (mode != RW_NONE &&
146 			    !rw_tryenter(&zv->zv_suspend_lock, mode)) {
147 				mutex_exit(&zv->zv_state_lock);
148 				rw_enter(&zv->zv_suspend_lock, mode);
149 				mutex_enter(&zv->zv_state_lock);
150 				/*
151 				 * zvol cannot be renamed as we continue
152 				 * to hold zvol_state_lock
153 				 */
154 				ASSERT(zv->zv_hash == hash &&
155 				    strcmp(zv->zv_name, name) == 0);
156 			}
157 			rw_exit(&zvol_state_lock);
158 			return (zv);
159 		}
160 		mutex_exit(&zv->zv_state_lock);
161 	}
162 	rw_exit(&zvol_state_lock);
163 
164 	return (NULL);
165 }
166 
167 /*
168  * Find a zvol_state_t given the name.
169  * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
170  * return (NULL) without the taking locks. The zv_suspend_lock is always taken
171  * before zv_state_lock. The mode argument indicates the mode (including none)
172  * for zv_suspend_lock to be taken.
173  */
174 static zvol_state_t *
175 zvol_find_by_name(const char *name, int mode)
176 {
177 	return (zvol_find_by_name_hash(name, zvol_name_hash(name), mode));
178 }
179 
180 /*
181  * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
182  */
183 void
184 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
185 {
186 	zfs_creat_t *zct = arg;
187 	nvlist_t *nvprops = zct->zct_props;
188 	int error;
189 	uint64_t volblocksize, volsize;
190 
191 	VERIFY(nvlist_lookup_uint64(nvprops,
192 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
193 	if (nvlist_lookup_uint64(nvprops,
194 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
195 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
196 
197 	/*
198 	 * These properties must be removed from the list so the generic
199 	 * property setting step won't apply to them.
200 	 */
201 	VERIFY(nvlist_remove_all(nvprops,
202 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
203 	(void) nvlist_remove_all(nvprops,
204 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
205 
206 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
207 	    DMU_OT_NONE, 0, tx);
208 	ASSERT(error == 0);
209 
210 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
211 	    DMU_OT_NONE, 0, tx);
212 	ASSERT(error == 0);
213 
214 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
215 	ASSERT(error == 0);
216 }
217 
218 /*
219  * ZFS_IOC_OBJSET_STATS entry point.
220  */
221 int
222 zvol_get_stats(objset_t *os, nvlist_t *nv)
223 {
224 	int error;
225 	dmu_object_info_t *doi;
226 	uint64_t val;
227 
228 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
229 	if (error)
230 		return (SET_ERROR(error));
231 
232 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
233 	doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);
234 	error = dmu_object_info(os, ZVOL_OBJ, doi);
235 
236 	if (error == 0) {
237 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
238 		    doi->doi_data_block_size);
239 	}
240 
241 	kmem_free(doi, sizeof (dmu_object_info_t));
242 
243 	return (SET_ERROR(error));
244 }
245 
246 /*
247  * Sanity check volume size.
248  */
249 int
250 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
251 {
252 	if (volsize == 0)
253 		return (SET_ERROR(EINVAL));
254 
255 	if (volsize % blocksize != 0)
256 		return (SET_ERROR(EINVAL));
257 
258 #ifdef _ILP32
259 	if (volsize - 1 > SPEC_MAXOFFSET_T)
260 		return (SET_ERROR(EOVERFLOW));
261 #endif
262 	return (0);
263 }
264 
265 /*
266  * Ensure the zap is flushed then inform the VFS of the capacity change.
267  */
268 static int
269 zvol_update_volsize(uint64_t volsize, objset_t *os)
270 {
271 	dmu_tx_t *tx;
272 	int error;
273 	uint64_t txg;
274 
275 	tx = dmu_tx_create(os);
276 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
277 	dmu_tx_mark_netfree(tx);
278 	error = dmu_tx_assign(tx, TXG_WAIT);
279 	if (error) {
280 		dmu_tx_abort(tx);
281 		return (SET_ERROR(error));
282 	}
283 	txg = dmu_tx_get_txg(tx);
284 
285 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
286 	    &volsize, tx);
287 	dmu_tx_commit(tx);
288 
289 	txg_wait_synced(dmu_objset_pool(os), txg);
290 
291 	if (error == 0)
292 		error = dmu_free_long_range(os,
293 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
294 
295 	return (error);
296 }
297 
298 /*
299  * Set ZFS_PROP_VOLSIZE set entry point.  Note that modifying the volume
300  * size will result in a udev "change" event being generated.
301  */
302 int
303 zvol_set_volsize(const char *name, uint64_t volsize)
304 {
305 	objset_t *os = NULL;
306 	uint64_t readonly;
307 	int error;
308 	boolean_t owned = B_FALSE;
309 
310 	error = dsl_prop_get_integer(name,
311 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
312 	if (error != 0)
313 		return (SET_ERROR(error));
314 	if (readonly)
315 		return (SET_ERROR(EROFS));
316 
317 	zvol_state_t *zv = zvol_find_by_name(name, RW_READER);
318 
319 	ASSERT(zv == NULL || (MUTEX_HELD(&zv->zv_state_lock) &&
320 	    RW_READ_HELD(&zv->zv_suspend_lock)));
321 
322 	if (zv == NULL || zv->zv_objset == NULL) {
323 		if (zv != NULL)
324 			rw_exit(&zv->zv_suspend_lock);
325 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE,
326 		    FTAG, &os)) != 0) {
327 			if (zv != NULL)
328 				mutex_exit(&zv->zv_state_lock);
329 			return (SET_ERROR(error));
330 		}
331 		owned = B_TRUE;
332 		if (zv != NULL)
333 			zv->zv_objset = os;
334 	} else {
335 		os = zv->zv_objset;
336 	}
337 
338 	dmu_object_info_t *doi = kmem_alloc(sizeof (*doi), KM_SLEEP);
339 
340 	if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) ||
341 	    (error = zvol_check_volsize(volsize, doi->doi_data_block_size)))
342 		goto out;
343 
344 	error = zvol_update_volsize(volsize, os);
345 	if (error == 0 && zv != NULL) {
346 		zv->zv_volsize = volsize;
347 		zv->zv_changed = 1;
348 	}
349 out:
350 	kmem_free(doi, sizeof (dmu_object_info_t));
351 
352 	if (owned) {
353 		dmu_objset_disown(os, B_TRUE, FTAG);
354 		if (zv != NULL)
355 			zv->zv_objset = NULL;
356 	} else {
357 		rw_exit(&zv->zv_suspend_lock);
358 	}
359 
360 	if (zv != NULL)
361 		mutex_exit(&zv->zv_state_lock);
362 
363 	if (error == 0 && zv != NULL)
364 		zvol_os_update_volsize(zv, volsize);
365 
366 	return (SET_ERROR(error));
367 }
368 
369 /*
370  * Update volthreading.
371  */
372 int
373 zvol_set_volthreading(const char *name, boolean_t value)
374 {
375 	zvol_state_t *zv = zvol_find_by_name(name, RW_NONE);
376 	if (zv == NULL)
377 		return (ENOENT);
378 	zv->zv_threading = value;
379 	mutex_exit(&zv->zv_state_lock);
380 	return (0);
381 }
382 
383 /*
384  * Update zvol ro property.
385  */
386 int
387 zvol_set_ro(const char *name, boolean_t value)
388 {
389 	zvol_state_t *zv = zvol_find_by_name(name, RW_NONE);
390 	if (zv == NULL)
391 		return (-1);
392 	if (value) {
393 		zvol_os_set_disk_ro(zv, 1);
394 		zv->zv_flags |= ZVOL_RDONLY;
395 	} else {
396 		zvol_os_set_disk_ro(zv, 0);
397 		zv->zv_flags &= ~ZVOL_RDONLY;
398 	}
399 	mutex_exit(&zv->zv_state_lock);
400 	return (0);
401 }
402 
403 /*
404  * Sanity check volume block size.
405  */
406 int
407 zvol_check_volblocksize(const char *name, uint64_t volblocksize)
408 {
409 	/* Record sizes above 128k need the feature to be enabled */
410 	if (volblocksize > SPA_OLD_MAXBLOCKSIZE) {
411 		spa_t *spa;
412 		int error;
413 
414 		if ((error = spa_open(name, &spa, FTAG)) != 0)
415 			return (error);
416 
417 		if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
418 			spa_close(spa, FTAG);
419 			return (SET_ERROR(ENOTSUP));
420 		}
421 
422 		/*
423 		 * We don't allow setting the property above 1MB,
424 		 * unless the tunable has been changed.
425 		 */
426 		if (volblocksize > zfs_max_recordsize)
427 			return (SET_ERROR(EDOM));
428 
429 		spa_close(spa, FTAG);
430 	}
431 
432 	if (volblocksize < SPA_MINBLOCKSIZE ||
433 	    volblocksize > SPA_MAXBLOCKSIZE ||
434 	    !ISP2(volblocksize))
435 		return (SET_ERROR(EDOM));
436 
437 	return (0);
438 }
439 
440 /*
441  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
442  * implement DKIOCFREE/free-long-range.
443  */
444 static int
445 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
446 {
447 	zvol_state_t *zv = arg1;
448 	lr_truncate_t *lr = arg2;
449 	uint64_t offset, length;
450 
451 	ASSERT3U(lr->lr_common.lrc_reclen, >=, sizeof (*lr));
452 
453 	if (byteswap)
454 		byteswap_uint64_array(lr, sizeof (*lr));
455 
456 	offset = lr->lr_offset;
457 	length = lr->lr_length;
458 
459 	dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
460 	dmu_tx_mark_netfree(tx);
461 	int error = dmu_tx_assign(tx, TXG_WAIT);
462 	if (error != 0) {
463 		dmu_tx_abort(tx);
464 	} else {
465 		(void) zil_replaying(zv->zv_zilog, tx);
466 		dmu_tx_commit(tx);
467 		error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset,
468 		    length);
469 	}
470 
471 	return (error);
472 }
473 
474 /*
475  * Replay a TX_WRITE ZIL transaction that didn't get committed
476  * after a system failure
477  */
478 static int
479 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
480 {
481 	zvol_state_t *zv = arg1;
482 	lr_write_t *lr = arg2;
483 	objset_t *os = zv->zv_objset;
484 	char *data = (char *)(lr + 1);  /* data follows lr_write_t */
485 	uint64_t offset, length;
486 	dmu_tx_t *tx;
487 	int error;
488 
489 	ASSERT3U(lr->lr_common.lrc_reclen, >=, sizeof (*lr));
490 
491 	if (byteswap)
492 		byteswap_uint64_array(lr, sizeof (*lr));
493 
494 	offset = lr->lr_offset;
495 	length = lr->lr_length;
496 
497 	/* If it's a dmu_sync() block, write the whole block */
498 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
499 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
500 		if (length < blocksize) {
501 			offset -= offset % blocksize;
502 			length = blocksize;
503 		}
504 	}
505 
506 	tx = dmu_tx_create(os);
507 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
508 	error = dmu_tx_assign(tx, TXG_WAIT);
509 	if (error) {
510 		dmu_tx_abort(tx);
511 	} else {
512 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
513 		(void) zil_replaying(zv->zv_zilog, tx);
514 		dmu_tx_commit(tx);
515 	}
516 
517 	return (error);
518 }
519 
520 /*
521  * Replay a TX_CLONE_RANGE ZIL transaction that didn't get committed
522  * after a system failure
523  */
524 static int
525 zvol_replay_clone_range(void *arg1, void *arg2, boolean_t byteswap)
526 {
527 	zvol_state_t *zv = arg1;
528 	lr_clone_range_t *lr = arg2;
529 	objset_t *os = zv->zv_objset;
530 	dmu_tx_t *tx;
531 	int error;
532 	uint64_t blksz;
533 	uint64_t off;
534 	uint64_t len;
535 
536 	ASSERT3U(lr->lr_common.lrc_reclen, >=, sizeof (*lr));
537 	ASSERT3U(lr->lr_common.lrc_reclen, >=, offsetof(lr_clone_range_t,
538 	    lr_bps[lr->lr_nbps]));
539 
540 	if (byteswap)
541 		byteswap_uint64_array(lr, sizeof (*lr));
542 
543 	ASSERT(spa_feature_is_enabled(dmu_objset_spa(os),
544 	    SPA_FEATURE_BLOCK_CLONING));
545 
546 	off = lr->lr_offset;
547 	len = lr->lr_length;
548 	blksz = lr->lr_blksz;
549 
550 	if ((off % blksz) != 0) {
551 		return (SET_ERROR(EINVAL));
552 	}
553 
554 	error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn);
555 	if (error != 0 || !zv->zv_dn)
556 		return (error);
557 	tx = dmu_tx_create(os);
558 	dmu_tx_hold_clone_by_dnode(tx, zv->zv_dn, off, len);
559 	error = dmu_tx_assign(tx, TXG_WAIT);
560 	if (error != 0) {
561 		dmu_tx_abort(tx);
562 		goto out;
563 	}
564 	error = dmu_brt_clone(zv->zv_objset, ZVOL_OBJ, off, len,
565 	    tx, lr->lr_bps, lr->lr_nbps);
566 	if (error != 0) {
567 		dmu_tx_commit(tx);
568 		goto out;
569 	}
570 
571 	/*
572 	 * zil_replaying() not only check if we are replaying ZIL, but also
573 	 * updates the ZIL header to record replay progress.
574 	 */
575 	VERIFY(zil_replaying(zv->zv_zilog, tx));
576 	dmu_tx_commit(tx);
577 
578 out:
579 	dnode_rele(zv->zv_dn, zv);
580 	zv->zv_dn = NULL;
581 	return (error);
582 }
583 
584 int
585 zvol_clone_range(zvol_state_t *zv_src, uint64_t inoff, zvol_state_t *zv_dst,
586     uint64_t outoff, uint64_t len)
587 {
588 	zilog_t	*zilog_dst;
589 	zfs_locked_range_t *inlr, *outlr;
590 	objset_t *inos, *outos;
591 	dmu_tx_t *tx;
592 	blkptr_t *bps;
593 	size_t maxblocks;
594 	int error = EINVAL;
595 
596 	rw_enter(&zv_dst->zv_suspend_lock, RW_READER);
597 	if (zv_dst->zv_zilog == NULL) {
598 		rw_exit(&zv_dst->zv_suspend_lock);
599 		rw_enter(&zv_dst->zv_suspend_lock, RW_WRITER);
600 		if (zv_dst->zv_zilog == NULL) {
601 			zv_dst->zv_zilog = zil_open(zv_dst->zv_objset,
602 			    zvol_get_data, &zv_dst->zv_kstat.dk_zil_sums);
603 			zv_dst->zv_flags |= ZVOL_WRITTEN_TO;
604 			VERIFY0((zv_dst->zv_zilog->zl_header->zh_flags &
605 			    ZIL_REPLAY_NEEDED));
606 		}
607 		rw_downgrade(&zv_dst->zv_suspend_lock);
608 	}
609 	if (zv_src != zv_dst)
610 		rw_enter(&zv_src->zv_suspend_lock, RW_READER);
611 
612 	inos = zv_src->zv_objset;
613 	outos = zv_dst->zv_objset;
614 
615 	/*
616 	 * Sanity checks
617 	 */
618 	if (!spa_feature_is_enabled(dmu_objset_spa(outos),
619 	    SPA_FEATURE_BLOCK_CLONING)) {
620 		error = EOPNOTSUPP;
621 		goto out;
622 	}
623 	if (dmu_objset_spa(inos) != dmu_objset_spa(outos)) {
624 		error = EXDEV;
625 		goto out;
626 	}
627 	if (inos->os_encrypted != outos->os_encrypted) {
628 		error = EXDEV;
629 		goto out;
630 	}
631 	if (zv_src->zv_volblocksize != zv_dst->zv_volblocksize) {
632 		error = EINVAL;
633 		goto out;
634 	}
635 	if (inoff >= zv_src->zv_volsize || outoff >= zv_dst->zv_volsize) {
636 		error = 0;
637 		goto out;
638 	}
639 
640 	/*
641 	 * Do not read beyond boundary
642 	 */
643 	if (len > zv_src->zv_volsize - inoff)
644 		len = zv_src->zv_volsize - inoff;
645 	if (len > zv_dst->zv_volsize - outoff)
646 		len = zv_dst->zv_volsize - outoff;
647 	if (len == 0) {
648 		error = 0;
649 		goto out;
650 	}
651 
652 	/*
653 	 * No overlapping if we are cloning within the same file
654 	 */
655 	if (zv_src == zv_dst) {
656 		if (inoff < outoff + len && outoff < inoff + len) {
657 			error = EINVAL;
658 			goto out;
659 		}
660 	}
661 
662 	/*
663 	 * Offsets and length must be at block boundaries
664 	 */
665 	if ((inoff % zv_src->zv_volblocksize) != 0 ||
666 	    (outoff % zv_dst->zv_volblocksize) != 0) {
667 		error = EINVAL;
668 		goto out;
669 	}
670 
671 	/*
672 	 * Length must be multiple of block size
673 	 */
674 	if ((len % zv_src->zv_volblocksize) != 0) {
675 		error = EINVAL;
676 		goto out;
677 	}
678 
679 	zilog_dst = zv_dst->zv_zilog;
680 	maxblocks = zil_max_log_data(zilog_dst, sizeof (lr_clone_range_t)) /
681 	    sizeof (bps[0]);
682 	bps = vmem_alloc(sizeof (bps[0]) * maxblocks, KM_SLEEP);
683 	/*
684 	 * Maintain predictable lock order.
685 	 */
686 	if (zv_src < zv_dst || (zv_src == zv_dst && inoff < outoff)) {
687 		inlr = zfs_rangelock_enter(&zv_src->zv_rangelock, inoff, len,
688 		    RL_READER);
689 		outlr = zfs_rangelock_enter(&zv_dst->zv_rangelock, outoff, len,
690 		    RL_WRITER);
691 	} else {
692 		outlr = zfs_rangelock_enter(&zv_dst->zv_rangelock, outoff, len,
693 		    RL_WRITER);
694 		inlr = zfs_rangelock_enter(&zv_src->zv_rangelock, inoff, len,
695 		    RL_READER);
696 	}
697 
698 	while (len > 0) {
699 		uint64_t size, last_synced_txg;
700 		size_t nbps = maxblocks;
701 		size = MIN(zv_src->zv_volblocksize * maxblocks, len);
702 		last_synced_txg = spa_last_synced_txg(
703 		    dmu_objset_spa(zv_src->zv_objset));
704 		error = dmu_read_l0_bps(zv_src->zv_objset, ZVOL_OBJ, inoff,
705 		    size, bps, &nbps);
706 		if (error != 0) {
707 			/*
708 			 * If we are trying to clone a block that was created
709 			 * in the current transaction group, the error will be
710 			 * EAGAIN here.  Based on zfs_bclone_wait_dirty either
711 			 * return a shortened range to the caller so it can
712 			 * fallback, or wait for the next TXG and check again.
713 			 */
714 			if (error == EAGAIN && zfs_bclone_wait_dirty) {
715 				txg_wait_synced(dmu_objset_pool
716 				    (zv_src->zv_objset), last_synced_txg + 1);
717 					continue;
718 			}
719 			break;
720 		}
721 
722 		tx = dmu_tx_create(zv_dst->zv_objset);
723 		dmu_tx_hold_clone_by_dnode(tx, zv_dst->zv_dn, outoff, size);
724 		error = dmu_tx_assign(tx, TXG_WAIT);
725 		if (error != 0) {
726 			dmu_tx_abort(tx);
727 			break;
728 		}
729 		error = dmu_brt_clone(zv_dst->zv_objset, ZVOL_OBJ, outoff, size,
730 		    tx, bps, nbps);
731 		if (error != 0) {
732 			dmu_tx_commit(tx);
733 			break;
734 		}
735 		zvol_log_clone_range(zilog_dst, tx, TX_CLONE_RANGE, outoff,
736 		    size, zv_src->zv_volblocksize, bps, nbps);
737 		dmu_tx_commit(tx);
738 		inoff += size;
739 		outoff += size;
740 		len -= size;
741 	}
742 	vmem_free(bps, sizeof (bps[0]) * maxblocks);
743 	zfs_rangelock_exit(outlr);
744 	zfs_rangelock_exit(inlr);
745 	if (error == 0 && zv_dst->zv_objset->os_sync == ZFS_SYNC_ALWAYS) {
746 		zil_commit(zilog_dst, ZVOL_OBJ);
747 	}
748 out:
749 	if (zv_src != zv_dst)
750 		rw_exit(&zv_src->zv_suspend_lock);
751 	rw_exit(&zv_dst->zv_suspend_lock);
752 	return (SET_ERROR(error));
753 }
754 
755 /*
756  * Handles TX_CLONE_RANGE transactions.
757  */
758 void
759 zvol_log_clone_range(zilog_t *zilog, dmu_tx_t *tx, int txtype, uint64_t off,
760     uint64_t len, uint64_t blksz, const blkptr_t *bps, size_t nbps)
761 {
762 	itx_t *itx;
763 	lr_clone_range_t *lr;
764 	uint64_t partlen, max_log_data;
765 	size_t partnbps;
766 
767 	if (zil_replaying(zilog, tx))
768 		return;
769 
770 	max_log_data = zil_max_log_data(zilog, sizeof (lr_clone_range_t));
771 
772 	while (nbps > 0) {
773 		partnbps = MIN(nbps, max_log_data / sizeof (bps[0]));
774 		partlen = partnbps * blksz;
775 		ASSERT3U(partlen, <, len + blksz);
776 		partlen = MIN(partlen, len);
777 
778 		itx = zil_itx_create(txtype,
779 		    sizeof (*lr) + sizeof (bps[0]) * partnbps);
780 		lr = (lr_clone_range_t *)&itx->itx_lr;
781 		lr->lr_foid = ZVOL_OBJ;
782 		lr->lr_offset = off;
783 		lr->lr_length = partlen;
784 		lr->lr_blksz = blksz;
785 		lr->lr_nbps = partnbps;
786 		memcpy(lr->lr_bps, bps, sizeof (bps[0]) * partnbps);
787 
788 		zil_itx_assign(zilog, itx, tx);
789 
790 		bps += partnbps;
791 		ASSERT3U(nbps, >=, partnbps);
792 		nbps -= partnbps;
793 		off += partlen;
794 		ASSERT3U(len, >=, partlen);
795 		len -= partlen;
796 	}
797 }
798 
799 static int
800 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
801 {
802 	(void) arg1, (void) arg2, (void) byteswap;
803 	return (SET_ERROR(ENOTSUP));
804 }
805 
806 /*
807  * Callback vectors for replaying records.
808  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
809  */
810 zil_replay_func_t *const zvol_replay_vector[TX_MAX_TYPE] = {
811 	zvol_replay_err,	/* no such transaction type */
812 	zvol_replay_err,	/* TX_CREATE */
813 	zvol_replay_err,	/* TX_MKDIR */
814 	zvol_replay_err,	/* TX_MKXATTR */
815 	zvol_replay_err,	/* TX_SYMLINK */
816 	zvol_replay_err,	/* TX_REMOVE */
817 	zvol_replay_err,	/* TX_RMDIR */
818 	zvol_replay_err,	/* TX_LINK */
819 	zvol_replay_err,	/* TX_RENAME */
820 	zvol_replay_write,	/* TX_WRITE */
821 	zvol_replay_truncate,	/* TX_TRUNCATE */
822 	zvol_replay_err,	/* TX_SETATTR */
823 	zvol_replay_err,	/* TX_ACL_V0 */
824 	zvol_replay_err,	/* TX_ACL */
825 	zvol_replay_err,	/* TX_CREATE_ACL */
826 	zvol_replay_err,	/* TX_CREATE_ATTR */
827 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
828 	zvol_replay_err,	/* TX_MKDIR_ACL */
829 	zvol_replay_err,	/* TX_MKDIR_ATTR */
830 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
831 	zvol_replay_err,	/* TX_WRITE2 */
832 	zvol_replay_err,	/* TX_SETSAXATTR */
833 	zvol_replay_err,	/* TX_RENAME_EXCHANGE */
834 	zvol_replay_err,	/* TX_RENAME_WHITEOUT */
835 	zvol_replay_clone_range,	/* TX_CLONE_RANGE */
836 };
837 
838 /*
839  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
840  *
841  * We store data in the log buffers if it's small enough.
842  * Otherwise we will later flush the data out via dmu_sync().
843  */
844 static const ssize_t zvol_immediate_write_sz = 32768;
845 
846 void
847 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
848     uint64_t size, boolean_t commit)
849 {
850 	uint32_t blocksize = zv->zv_volblocksize;
851 	zilog_t *zilog = zv->zv_zilog;
852 	itx_wr_state_t write_state;
853 	uint64_t sz = size;
854 
855 	if (zil_replaying(zilog, tx))
856 		return;
857 
858 	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
859 		write_state = WR_INDIRECT;
860 	else if (!spa_has_slogs(zilog->zl_spa) &&
861 	    size >= blocksize && blocksize > zvol_immediate_write_sz)
862 		write_state = WR_INDIRECT;
863 	else if (commit)
864 		write_state = WR_COPIED;
865 	else
866 		write_state = WR_NEED_COPY;
867 
868 	while (size) {
869 		itx_t *itx;
870 		lr_write_t *lr;
871 		itx_wr_state_t wr_state = write_state;
872 		ssize_t len = size;
873 
874 		if (wr_state == WR_COPIED && size > zil_max_copied_data(zilog))
875 			wr_state = WR_NEED_COPY;
876 		else if (wr_state == WR_INDIRECT)
877 			len = MIN(blocksize - P2PHASE(offset, blocksize), size);
878 
879 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
880 		    (wr_state == WR_COPIED ? len : 0));
881 		lr = (lr_write_t *)&itx->itx_lr;
882 		if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
883 		    offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
884 			zil_itx_destroy(itx);
885 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
886 			lr = (lr_write_t *)&itx->itx_lr;
887 			wr_state = WR_NEED_COPY;
888 		}
889 
890 		itx->itx_wr_state = wr_state;
891 		lr->lr_foid = ZVOL_OBJ;
892 		lr->lr_offset = offset;
893 		lr->lr_length = len;
894 		lr->lr_blkoff = 0;
895 		BP_ZERO(&lr->lr_blkptr);
896 
897 		itx->itx_private = zv;
898 
899 		(void) zil_itx_assign(zilog, itx, tx);
900 
901 		offset += len;
902 		size -= len;
903 	}
904 
905 	if (write_state == WR_COPIED || write_state == WR_NEED_COPY) {
906 		dsl_pool_wrlog_count(zilog->zl_dmu_pool, sz, tx->tx_txg);
907 	}
908 }
909 
910 /*
911  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
912  */
913 void
914 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len)
915 {
916 	itx_t *itx;
917 	lr_truncate_t *lr;
918 	zilog_t *zilog = zv->zv_zilog;
919 
920 	if (zil_replaying(zilog, tx))
921 		return;
922 
923 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
924 	lr = (lr_truncate_t *)&itx->itx_lr;
925 	lr->lr_foid = ZVOL_OBJ;
926 	lr->lr_offset = off;
927 	lr->lr_length = len;
928 
929 	zil_itx_assign(zilog, itx, tx);
930 }
931 
932 
933 static void
934 zvol_get_done(zgd_t *zgd, int error)
935 {
936 	(void) error;
937 	if (zgd->zgd_db)
938 		dmu_buf_rele(zgd->zgd_db, zgd);
939 
940 	zfs_rangelock_exit(zgd->zgd_lr);
941 
942 	kmem_free(zgd, sizeof (zgd_t));
943 }
944 
945 /*
946  * Get data to generate a TX_WRITE intent log record.
947  */
948 int
949 zvol_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
950     struct lwb *lwb, zio_t *zio)
951 {
952 	zvol_state_t *zv = arg;
953 	uint64_t offset = lr->lr_offset;
954 	uint64_t size = lr->lr_length;
955 	dmu_buf_t *db;
956 	zgd_t *zgd;
957 	int error;
958 
959 	ASSERT3P(lwb, !=, NULL);
960 	ASSERT3U(size, !=, 0);
961 
962 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
963 	zgd->zgd_lwb = lwb;
964 
965 	/*
966 	 * Write records come in two flavors: immediate and indirect.
967 	 * For small writes it's cheaper to store the data with the
968 	 * log record (immediate); for large writes it's cheaper to
969 	 * sync the data and get a pointer to it (indirect) so that
970 	 * we don't have to write the data twice.
971 	 */
972 	if (buf != NULL) { /* immediate write */
973 		zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
974 		    size, RL_READER);
975 		error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
976 		    DMU_READ_NO_PREFETCH);
977 	} else { /* indirect write */
978 		ASSERT3P(zio, !=, NULL);
979 		/*
980 		 * Have to lock the whole block to ensure when it's written out
981 		 * and its checksum is being calculated that no one can change
982 		 * the data. Contrarily to zfs_get_data we need not re-check
983 		 * blocksize after we get the lock because it cannot be changed.
984 		 */
985 		size = zv->zv_volblocksize;
986 		offset = P2ALIGN_TYPED(offset, size, uint64_t);
987 		zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
988 		    size, RL_READER);
989 		error = dmu_buf_hold_noread_by_dnode(zv->zv_dn, offset, zgd,
990 		    &db);
991 		if (error == 0) {
992 			blkptr_t *bp = &lr->lr_blkptr;
993 
994 			zgd->zgd_db = db;
995 			zgd->zgd_bp = bp;
996 
997 			ASSERT(db != NULL);
998 			ASSERT(db->db_offset == offset);
999 			ASSERT(db->db_size == size);
1000 
1001 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1002 			    zvol_get_done, zgd);
1003 
1004 			if (error == 0)
1005 				return (0);
1006 		}
1007 	}
1008 
1009 	zvol_get_done(zgd, error);
1010 
1011 	return (SET_ERROR(error));
1012 }
1013 
1014 /*
1015  * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
1016  */
1017 
1018 void
1019 zvol_insert(zvol_state_t *zv)
1020 {
1021 	ASSERT(RW_WRITE_HELD(&zvol_state_lock));
1022 	list_insert_head(&zvol_state_list, zv);
1023 	hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash));
1024 }
1025 
1026 /*
1027  * Simply remove the zvol from to list of zvols.
1028  */
1029 static void
1030 zvol_remove(zvol_state_t *zv)
1031 {
1032 	ASSERT(RW_WRITE_HELD(&zvol_state_lock));
1033 	list_remove(&zvol_state_list, zv);
1034 	hlist_del(&zv->zv_hlink);
1035 }
1036 
1037 /*
1038  * Setup zv after we just own the zv->objset
1039  */
1040 static int
1041 zvol_setup_zv(zvol_state_t *zv)
1042 {
1043 	uint64_t volsize;
1044 	int error;
1045 	uint64_t ro;
1046 	objset_t *os = zv->zv_objset;
1047 
1048 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1049 	ASSERT(RW_LOCK_HELD(&zv->zv_suspend_lock));
1050 
1051 	zv->zv_zilog = NULL;
1052 	zv->zv_flags &= ~ZVOL_WRITTEN_TO;
1053 
1054 	error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL);
1055 	if (error)
1056 		return (SET_ERROR(error));
1057 
1058 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
1059 	if (error)
1060 		return (SET_ERROR(error));
1061 
1062 	error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn);
1063 	if (error)
1064 		return (SET_ERROR(error));
1065 
1066 	zvol_os_set_capacity(zv, volsize >> 9);
1067 	zv->zv_volsize = volsize;
1068 
1069 	if (ro || dmu_objset_is_snapshot(os) ||
1070 	    !spa_writeable(dmu_objset_spa(os))) {
1071 		zvol_os_set_disk_ro(zv, 1);
1072 		zv->zv_flags |= ZVOL_RDONLY;
1073 	} else {
1074 		zvol_os_set_disk_ro(zv, 0);
1075 		zv->zv_flags &= ~ZVOL_RDONLY;
1076 	}
1077 	return (0);
1078 }
1079 
1080 /*
1081  * Shutdown every zv_objset related stuff except zv_objset itself.
1082  * The is the reverse of zvol_setup_zv.
1083  */
1084 static void
1085 zvol_shutdown_zv(zvol_state_t *zv)
1086 {
1087 	ASSERT(MUTEX_HELD(&zv->zv_state_lock) &&
1088 	    RW_LOCK_HELD(&zv->zv_suspend_lock));
1089 
1090 	if (zv->zv_flags & ZVOL_WRITTEN_TO) {
1091 		ASSERT(zv->zv_zilog != NULL);
1092 		zil_close(zv->zv_zilog);
1093 	}
1094 
1095 	zv->zv_zilog = NULL;
1096 
1097 	dnode_rele(zv->zv_dn, zv);
1098 	zv->zv_dn = NULL;
1099 
1100 	/*
1101 	 * Evict cached data. We must write out any dirty data before
1102 	 * disowning the dataset.
1103 	 */
1104 	if (zv->zv_flags & ZVOL_WRITTEN_TO)
1105 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1106 	(void) dmu_objset_evict_dbufs(zv->zv_objset);
1107 }
1108 
1109 /*
1110  * return the proper tag for rollback and recv
1111  */
1112 void *
1113 zvol_tag(zvol_state_t *zv)
1114 {
1115 	ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1116 	return (zv->zv_open_count > 0 ? zv : NULL);
1117 }
1118 
1119 /*
1120  * Suspend the zvol for recv and rollback.
1121  */
1122 zvol_state_t *
1123 zvol_suspend(const char *name)
1124 {
1125 	zvol_state_t *zv;
1126 
1127 	zv = zvol_find_by_name(name, RW_WRITER);
1128 
1129 	if (zv == NULL)
1130 		return (NULL);
1131 
1132 	/* block all I/O, release in zvol_resume. */
1133 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1134 	ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1135 
1136 	atomic_inc(&zv->zv_suspend_ref);
1137 
1138 	if (zv->zv_open_count > 0)
1139 		zvol_shutdown_zv(zv);
1140 
1141 	/*
1142 	 * do not hold zv_state_lock across suspend/resume to
1143 	 * avoid locking up zvol lookups
1144 	 */
1145 	mutex_exit(&zv->zv_state_lock);
1146 
1147 	/* zv_suspend_lock is released in zvol_resume() */
1148 	return (zv);
1149 }
1150 
1151 int
1152 zvol_resume(zvol_state_t *zv)
1153 {
1154 	int error = 0;
1155 
1156 	ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1157 
1158 	mutex_enter(&zv->zv_state_lock);
1159 
1160 	if (zv->zv_open_count > 0) {
1161 		VERIFY0(dmu_objset_hold(zv->zv_name, zv, &zv->zv_objset));
1162 		VERIFY3P(zv->zv_objset->os_dsl_dataset->ds_owner, ==, zv);
1163 		VERIFY(dsl_dataset_long_held(zv->zv_objset->os_dsl_dataset));
1164 		dmu_objset_rele(zv->zv_objset, zv);
1165 
1166 		error = zvol_setup_zv(zv);
1167 	}
1168 
1169 	mutex_exit(&zv->zv_state_lock);
1170 
1171 	rw_exit(&zv->zv_suspend_lock);
1172 	/*
1173 	 * We need this because we don't hold zvol_state_lock while releasing
1174 	 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
1175 	 * zv_suspend_lock to determine it is safe to free because rwlock is
1176 	 * not inherent atomic.
1177 	 */
1178 	atomic_dec(&zv->zv_suspend_ref);
1179 
1180 	if (zv->zv_flags & ZVOL_REMOVING)
1181 		cv_broadcast(&zv->zv_removing_cv);
1182 
1183 	return (SET_ERROR(error));
1184 }
1185 
1186 int
1187 zvol_first_open(zvol_state_t *zv, boolean_t readonly)
1188 {
1189 	objset_t *os;
1190 	int error;
1191 
1192 	ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
1193 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1194 	ASSERT(mutex_owned(&spa_namespace_lock));
1195 
1196 	boolean_t ro = (readonly || (strchr(zv->zv_name, '@') != NULL));
1197 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os);
1198 	if (error)
1199 		return (SET_ERROR(error));
1200 
1201 	zv->zv_objset = os;
1202 
1203 	error = zvol_setup_zv(zv);
1204 	if (error) {
1205 		dmu_objset_disown(os, 1, zv);
1206 		zv->zv_objset = NULL;
1207 	}
1208 
1209 	return (error);
1210 }
1211 
1212 void
1213 zvol_last_close(zvol_state_t *zv)
1214 {
1215 	ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
1216 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1217 
1218 	if (zv->zv_flags & ZVOL_REMOVING)
1219 		cv_broadcast(&zv->zv_removing_cv);
1220 
1221 	zvol_shutdown_zv(zv);
1222 
1223 	dmu_objset_disown(zv->zv_objset, 1, zv);
1224 	zv->zv_objset = NULL;
1225 }
1226 
1227 typedef struct minors_job {
1228 	list_t *list;
1229 	list_node_t link;
1230 	/* input */
1231 	char *name;
1232 	/* output */
1233 	int error;
1234 } minors_job_t;
1235 
1236 /*
1237  * Prefetch zvol dnodes for the minors_job
1238  */
1239 static void
1240 zvol_prefetch_minors_impl(void *arg)
1241 {
1242 	minors_job_t *job = arg;
1243 	char *dsname = job->name;
1244 	objset_t *os = NULL;
1245 
1246 	job->error = dmu_objset_own(dsname, DMU_OST_ZVOL, B_TRUE, B_TRUE,
1247 	    FTAG, &os);
1248 	if (job->error == 0) {
1249 		dmu_prefetch_dnode(os, ZVOL_OBJ, ZIO_PRIORITY_SYNC_READ);
1250 		dmu_objset_disown(os, B_TRUE, FTAG);
1251 	}
1252 }
1253 
1254 /*
1255  * Mask errors to continue dmu_objset_find() traversal
1256  */
1257 static int
1258 zvol_create_snap_minor_cb(const char *dsname, void *arg)
1259 {
1260 	minors_job_t *j = arg;
1261 	list_t *minors_list = j->list;
1262 	const char *name = j->name;
1263 
1264 	ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1265 
1266 	/* skip the designated dataset */
1267 	if (name && strcmp(dsname, name) == 0)
1268 		return (0);
1269 
1270 	/* at this point, the dsname should name a snapshot */
1271 	if (strchr(dsname, '@') == 0) {
1272 		dprintf("zvol_create_snap_minor_cb(): "
1273 		    "%s is not a snapshot name\n", dsname);
1274 	} else {
1275 		minors_job_t *job;
1276 		char *n = kmem_strdup(dsname);
1277 		if (n == NULL)
1278 			return (0);
1279 
1280 		job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1281 		job->name = n;
1282 		job->list = minors_list;
1283 		job->error = 0;
1284 		list_insert_tail(minors_list, job);
1285 		/* don't care if dispatch fails, because job->error is 0 */
1286 		taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1287 		    TQ_SLEEP);
1288 	}
1289 
1290 	return (0);
1291 }
1292 
1293 /*
1294  * If spa_keystore_load_wkey() is called for an encrypted zvol,
1295  * we need to look for any clones also using the key. This function
1296  * is "best effort" - so we just skip over it if there are failures.
1297  */
1298 static void
1299 zvol_add_clones(const char *dsname, list_t *minors_list)
1300 {
1301 	/* Also check if it has clones */
1302 	dsl_dir_t *dd = NULL;
1303 	dsl_pool_t *dp = NULL;
1304 
1305 	if (dsl_pool_hold(dsname, FTAG, &dp) != 0)
1306 		return;
1307 
1308 	if (!spa_feature_is_enabled(dp->dp_spa,
1309 	    SPA_FEATURE_ENCRYPTION))
1310 		goto out;
1311 
1312 	if (dsl_dir_hold(dp, dsname, FTAG, &dd, NULL) != 0)
1313 		goto out;
1314 
1315 	if (dsl_dir_phys(dd)->dd_clones == 0)
1316 		goto out;
1317 
1318 	zap_cursor_t *zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
1319 	zap_attribute_t *za = zap_attribute_alloc();
1320 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1321 
1322 	for (zap_cursor_init(zc, mos, dsl_dir_phys(dd)->dd_clones);
1323 	    zap_cursor_retrieve(zc, za) == 0;
1324 	    zap_cursor_advance(zc)) {
1325 		dsl_dataset_t *clone;
1326 		minors_job_t *job;
1327 
1328 		if (dsl_dataset_hold_obj(dd->dd_pool,
1329 		    za->za_first_integer, FTAG, &clone) == 0) {
1330 
1331 			char name[ZFS_MAX_DATASET_NAME_LEN];
1332 			dsl_dataset_name(clone, name);
1333 
1334 			char *n = kmem_strdup(name);
1335 			job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1336 			job->name = n;
1337 			job->list = minors_list;
1338 			job->error = 0;
1339 			list_insert_tail(minors_list, job);
1340 
1341 			dsl_dataset_rele(clone, FTAG);
1342 		}
1343 	}
1344 	zap_cursor_fini(zc);
1345 	zap_attribute_free(za);
1346 	kmem_free(zc, sizeof (zap_cursor_t));
1347 
1348 out:
1349 	if (dd != NULL)
1350 		dsl_dir_rele(dd, FTAG);
1351 	dsl_pool_rele(dp, FTAG);
1352 }
1353 
1354 /*
1355  * Mask errors to continue dmu_objset_find() traversal
1356  */
1357 static int
1358 zvol_create_minors_cb(const char *dsname, void *arg)
1359 {
1360 	uint64_t snapdev;
1361 	int error;
1362 	list_t *minors_list = arg;
1363 
1364 	ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1365 
1366 	error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL);
1367 	if (error)
1368 		return (0);
1369 
1370 	/*
1371 	 * Given the name and the 'snapdev' property, create device minor nodes
1372 	 * with the linkages to zvols/snapshots as needed.
1373 	 * If the name represents a zvol, create a minor node for the zvol, then
1374 	 * check if its snapshots are 'visible', and if so, iterate over the
1375 	 * snapshots and create device minor nodes for those.
1376 	 */
1377 	if (strchr(dsname, '@') == 0) {
1378 		minors_job_t *job;
1379 		char *n = kmem_strdup(dsname);
1380 		if (n == NULL)
1381 			return (0);
1382 
1383 		job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1384 		job->name = n;
1385 		job->list = minors_list;
1386 		job->error = 0;
1387 		list_insert_tail(minors_list, job);
1388 		/* don't care if dispatch fails, because job->error is 0 */
1389 		taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1390 		    TQ_SLEEP);
1391 
1392 		zvol_add_clones(dsname, minors_list);
1393 
1394 		if (snapdev == ZFS_SNAPDEV_VISIBLE) {
1395 			/*
1396 			 * traverse snapshots only, do not traverse children,
1397 			 * and skip the 'dsname'
1398 			 */
1399 			(void) dmu_objset_find(dsname,
1400 			    zvol_create_snap_minor_cb, (void *)job,
1401 			    DS_FIND_SNAPSHOTS);
1402 		}
1403 	} else {
1404 		dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
1405 		    dsname);
1406 	}
1407 
1408 	return (0);
1409 }
1410 
1411 /*
1412  * Create minors for the specified dataset, including children and snapshots.
1413  * Pay attention to the 'snapdev' property and iterate over the snapshots
1414  * only if they are 'visible'. This approach allows one to assure that the
1415  * snapshot metadata is read from disk only if it is needed.
1416  *
1417  * The name can represent a dataset to be recursively scanned for zvols and
1418  * their snapshots, or a single zvol snapshot. If the name represents a
1419  * dataset, the scan is performed in two nested stages:
1420  * - scan the dataset for zvols, and
1421  * - for each zvol, create a minor node, then check if the zvol's snapshots
1422  *   are 'visible', and only then iterate over the snapshots if needed
1423  *
1424  * If the name represents a snapshot, a check is performed if the snapshot is
1425  * 'visible' (which also verifies that the parent is a zvol), and if so,
1426  * a minor node for that snapshot is created.
1427  */
1428 void
1429 zvol_create_minors_recursive(const char *name)
1430 {
1431 	list_t minors_list;
1432 	minors_job_t *job;
1433 
1434 	if (zvol_inhibit_dev)
1435 		return;
1436 
1437 	/*
1438 	 * This is the list for prefetch jobs. Whenever we found a match
1439 	 * during dmu_objset_find, we insert a minors_job to the list and do
1440 	 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
1441 	 * any lock because all list operation is done on the current thread.
1442 	 *
1443 	 * We will use this list to do zvol_os_create_minor after prefetch
1444 	 * so we don't have to traverse using dmu_objset_find again.
1445 	 */
1446 	list_create(&minors_list, sizeof (minors_job_t),
1447 	    offsetof(minors_job_t, link));
1448 
1449 
1450 	if (strchr(name, '@') != NULL) {
1451 		uint64_t snapdev;
1452 
1453 		int error = dsl_prop_get_integer(name, "snapdev",
1454 		    &snapdev, NULL);
1455 
1456 		if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1457 			(void) zvol_os_create_minor(name);
1458 	} else {
1459 		fstrans_cookie_t cookie = spl_fstrans_mark();
1460 		(void) dmu_objset_find(name, zvol_create_minors_cb,
1461 		    &minors_list, DS_FIND_CHILDREN);
1462 		spl_fstrans_unmark(cookie);
1463 	}
1464 
1465 	taskq_wait_outstanding(system_taskq, 0);
1466 
1467 	/*
1468 	 * Prefetch is completed, we can do zvol_os_create_minor
1469 	 * sequentially.
1470 	 */
1471 	while ((job = list_remove_head(&minors_list)) != NULL) {
1472 		if (!job->error)
1473 			(void) zvol_os_create_minor(job->name);
1474 		kmem_strfree(job->name);
1475 		kmem_free(job, sizeof (minors_job_t));
1476 	}
1477 
1478 	list_destroy(&minors_list);
1479 }
1480 
1481 void
1482 zvol_create_minor(const char *name)
1483 {
1484 	/*
1485 	 * Note: the dsl_pool_config_lock must not be held.
1486 	 * Minor node creation needs to obtain the zvol_state_lock.
1487 	 * zvol_open() obtains the zvol_state_lock and then the dsl pool
1488 	 * config lock.  Therefore, we can't have the config lock now if
1489 	 * we are going to wait for the zvol_state_lock, because it
1490 	 * would be a lock order inversion which could lead to deadlock.
1491 	 */
1492 
1493 	if (zvol_inhibit_dev)
1494 		return;
1495 
1496 	if (strchr(name, '@') != NULL) {
1497 		uint64_t snapdev;
1498 
1499 		int error = dsl_prop_get_integer(name,
1500 		    "snapdev", &snapdev, NULL);
1501 
1502 		if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1503 			(void) zvol_os_create_minor(name);
1504 	} else {
1505 		(void) zvol_os_create_minor(name);
1506 	}
1507 }
1508 
1509 /*
1510  * Remove minors for specified dataset including children and snapshots.
1511  */
1512 
1513 /*
1514  * Remove the minor for a given zvol. This will do it all:
1515  *  - flag the zvol for removal, so new requests are rejected
1516  *  - wait until outstanding requests are completed
1517  *  - remove it from lists
1518  *  - free it
1519  * It's also usable as a taskq task, and smells nice too.
1520  */
1521 static void
1522 zvol_remove_minor_task(void *arg)
1523 {
1524 	zvol_state_t *zv = (zvol_state_t *)arg;
1525 
1526 	ASSERT(!RW_LOCK_HELD(&zvol_state_lock));
1527 	ASSERT(!MUTEX_HELD(&zv->zv_state_lock));
1528 
1529 	mutex_enter(&zv->zv_state_lock);
1530 	while (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) {
1531 		zv->zv_flags |= ZVOL_REMOVING;
1532 		cv_wait(&zv->zv_removing_cv, &zv->zv_state_lock);
1533 	}
1534 	mutex_exit(&zv->zv_state_lock);
1535 
1536 	rw_enter(&zvol_state_lock, RW_WRITER);
1537 	mutex_enter(&zv->zv_state_lock);
1538 
1539 	zvol_remove(zv);
1540 	zvol_os_clear_private(zv);
1541 
1542 	mutex_exit(&zv->zv_state_lock);
1543 	rw_exit(&zvol_state_lock);
1544 
1545 	zvol_os_free(zv);
1546 }
1547 
1548 static void
1549 zvol_free_task(void *arg)
1550 {
1551 	zvol_os_free(arg);
1552 }
1553 
1554 void
1555 zvol_remove_minors_impl(const char *name)
1556 {
1557 	zvol_state_t *zv, *zv_next;
1558 	int namelen = ((name) ? strlen(name) : 0);
1559 	taskqid_t t;
1560 	list_t delay_list, free_list;
1561 
1562 	if (zvol_inhibit_dev)
1563 		return;
1564 
1565 	list_create(&delay_list, sizeof (zvol_state_t),
1566 	    offsetof(zvol_state_t, zv_next));
1567 	list_create(&free_list, sizeof (zvol_state_t),
1568 	    offsetof(zvol_state_t, zv_next));
1569 
1570 	rw_enter(&zvol_state_lock, RW_WRITER);
1571 
1572 	for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1573 		zv_next = list_next(&zvol_state_list, zv);
1574 
1575 		mutex_enter(&zv->zv_state_lock);
1576 		if (name == NULL || strcmp(zv->zv_name, name) == 0 ||
1577 		    (strncmp(zv->zv_name, name, namelen) == 0 &&
1578 		    (zv->zv_name[namelen] == '/' ||
1579 		    zv->zv_name[namelen] == '@'))) {
1580 			/*
1581 			 * By holding zv_state_lock here, we guarantee that no
1582 			 * one is currently using this zv
1583 			 */
1584 
1585 			/*
1586 			 * If in use, try to throw everyone off and try again
1587 			 * later.
1588 			 */
1589 			if (zv->zv_open_count > 0 ||
1590 			    atomic_read(&zv->zv_suspend_ref)) {
1591 				zv->zv_flags |= ZVOL_REMOVING;
1592 				t = taskq_dispatch(
1593 				    zv->zv_objset->os_spa->spa_zvol_taskq,
1594 				    zvol_remove_minor_task, zv, TQ_SLEEP);
1595 				if (t == TASKQID_INVALID) {
1596 					/*
1597 					 * Couldn't create the task, so we'll
1598 					 * do it in place once the loop is
1599 					 * finished.
1600 					 */
1601 					list_insert_head(&delay_list, zv);
1602 				}
1603 				mutex_exit(&zv->zv_state_lock);
1604 				continue;
1605 			}
1606 
1607 			zvol_remove(zv);
1608 
1609 			/*
1610 			 * Cleared while holding zvol_state_lock as a writer
1611 			 * which will prevent zvol_open() from opening it.
1612 			 */
1613 			zvol_os_clear_private(zv);
1614 
1615 			/* Drop zv_state_lock before zvol_free() */
1616 			mutex_exit(&zv->zv_state_lock);
1617 
1618 			/* Try parallel zv_free, if failed do it in place */
1619 			t = taskq_dispatch(system_taskq, zvol_free_task, zv,
1620 			    TQ_SLEEP);
1621 			if (t == TASKQID_INVALID)
1622 				list_insert_head(&free_list, zv);
1623 		} else {
1624 			mutex_exit(&zv->zv_state_lock);
1625 		}
1626 	}
1627 	rw_exit(&zvol_state_lock);
1628 
1629 	/* Wait for zvols that we couldn't create a remove task for */
1630 	while ((zv = list_remove_head(&delay_list)) != NULL)
1631 		zvol_remove_minor_task(zv);
1632 
1633 	/* Free any that we couldn't free in parallel earlier */
1634 	while ((zv = list_remove_head(&free_list)) != NULL)
1635 		zvol_os_free(zv);
1636 }
1637 
1638 /* Remove minor for this specific volume only */
1639 static void
1640 zvol_remove_minor_impl(const char *name)
1641 {
1642 	zvol_state_t *zv = NULL, *zv_next;
1643 
1644 	if (zvol_inhibit_dev)
1645 		return;
1646 
1647 	rw_enter(&zvol_state_lock, RW_WRITER);
1648 
1649 	for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1650 		zv_next = list_next(&zvol_state_list, zv);
1651 
1652 		mutex_enter(&zv->zv_state_lock);
1653 		if (strcmp(zv->zv_name, name) == 0)
1654 			/* Found, leave the the loop with zv_lock held */
1655 			break;
1656 		mutex_exit(&zv->zv_state_lock);
1657 	}
1658 
1659 	if (zv == NULL) {
1660 		rw_exit(&zvol_state_lock);
1661 		return;
1662 	}
1663 
1664 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1665 
1666 	if (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) {
1667 		/*
1668 		 * In use, so try to throw everyone off, then wait
1669 		 * until finished.
1670 		 */
1671 		zv->zv_flags |= ZVOL_REMOVING;
1672 		mutex_exit(&zv->zv_state_lock);
1673 		rw_exit(&zvol_state_lock);
1674 		zvol_remove_minor_task(zv);
1675 		return;
1676 	}
1677 
1678 	zvol_remove(zv);
1679 	zvol_os_clear_private(zv);
1680 
1681 	mutex_exit(&zv->zv_state_lock);
1682 	rw_exit(&zvol_state_lock);
1683 
1684 	zvol_os_free(zv);
1685 }
1686 
1687 /*
1688  * Rename minors for specified dataset including children and snapshots.
1689  */
1690 static void
1691 zvol_rename_minors_impl(const char *oldname, const char *newname)
1692 {
1693 	zvol_state_t *zv, *zv_next;
1694 	int oldnamelen;
1695 
1696 	if (zvol_inhibit_dev)
1697 		return;
1698 
1699 	oldnamelen = strlen(oldname);
1700 
1701 	rw_enter(&zvol_state_lock, RW_READER);
1702 
1703 	for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1704 		zv_next = list_next(&zvol_state_list, zv);
1705 
1706 		mutex_enter(&zv->zv_state_lock);
1707 
1708 		if (strcmp(zv->zv_name, oldname) == 0) {
1709 			zvol_os_rename_minor(zv, newname);
1710 		} else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
1711 		    (zv->zv_name[oldnamelen] == '/' ||
1712 		    zv->zv_name[oldnamelen] == '@')) {
1713 			char *name = kmem_asprintf("%s%c%s", newname,
1714 			    zv->zv_name[oldnamelen],
1715 			    zv->zv_name + oldnamelen + 1);
1716 			zvol_os_rename_minor(zv, name);
1717 			kmem_strfree(name);
1718 		}
1719 
1720 		mutex_exit(&zv->zv_state_lock);
1721 	}
1722 
1723 	rw_exit(&zvol_state_lock);
1724 }
1725 
1726 typedef struct zvol_snapdev_cb_arg {
1727 	uint64_t snapdev;
1728 } zvol_snapdev_cb_arg_t;
1729 
1730 static int
1731 zvol_set_snapdev_cb(const char *dsname, void *param)
1732 {
1733 	zvol_snapdev_cb_arg_t *arg = param;
1734 
1735 	if (strchr(dsname, '@') == NULL)
1736 		return (0);
1737 
1738 	switch (arg->snapdev) {
1739 		case ZFS_SNAPDEV_VISIBLE:
1740 			(void) zvol_os_create_minor(dsname);
1741 			break;
1742 		case ZFS_SNAPDEV_HIDDEN:
1743 			(void) zvol_remove_minor_impl(dsname);
1744 			break;
1745 	}
1746 
1747 	return (0);
1748 }
1749 
1750 static void
1751 zvol_set_snapdev_impl(char *name, uint64_t snapdev)
1752 {
1753 	zvol_snapdev_cb_arg_t arg = {snapdev};
1754 	fstrans_cookie_t cookie = spl_fstrans_mark();
1755 	/*
1756 	 * The zvol_set_snapdev_sync() sets snapdev appropriately
1757 	 * in the dataset hierarchy. Here, we only scan snapshots.
1758 	 */
1759 	dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS);
1760 	spl_fstrans_unmark(cookie);
1761 }
1762 
1763 static void
1764 zvol_set_volmode_impl(char *name, uint64_t volmode)
1765 {
1766 	fstrans_cookie_t cookie;
1767 	uint64_t old_volmode;
1768 	zvol_state_t *zv;
1769 
1770 	if (strchr(name, '@') != NULL)
1771 		return;
1772 
1773 	/*
1774 	 * It's unfortunate we need to remove minors before we create new ones:
1775 	 * this is necessary because our backing gendisk (zvol_state->zv_disk)
1776 	 * could be different when we set, for instance, volmode from "geom"
1777 	 * to "dev" (or vice versa).
1778 	 */
1779 	zv = zvol_find_by_name(name, RW_NONE);
1780 	if (zv == NULL && volmode == ZFS_VOLMODE_NONE)
1781 			return;
1782 	if (zv != NULL) {
1783 		old_volmode = zv->zv_volmode;
1784 		mutex_exit(&zv->zv_state_lock);
1785 		if (old_volmode == volmode)
1786 			return;
1787 		zvol_wait_close(zv);
1788 	}
1789 	cookie = spl_fstrans_mark();
1790 	switch (volmode) {
1791 		case ZFS_VOLMODE_NONE:
1792 			(void) zvol_remove_minor_impl(name);
1793 			break;
1794 		case ZFS_VOLMODE_GEOM:
1795 		case ZFS_VOLMODE_DEV:
1796 			(void) zvol_remove_minor_impl(name);
1797 			(void) zvol_os_create_minor(name);
1798 			break;
1799 		case ZFS_VOLMODE_DEFAULT:
1800 			(void) zvol_remove_minor_impl(name);
1801 			if (zvol_volmode == ZFS_VOLMODE_NONE)
1802 				break;
1803 			else /* if zvol_volmode is invalid defaults to "geom" */
1804 				(void) zvol_os_create_minor(name);
1805 			break;
1806 	}
1807 	spl_fstrans_unmark(cookie);
1808 }
1809 
1810 static zvol_task_t *
1811 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2,
1812     uint64_t value)
1813 {
1814 	zvol_task_t *task;
1815 
1816 	/* Never allow tasks on hidden names. */
1817 	if (name1[0] == '$')
1818 		return (NULL);
1819 
1820 	task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP);
1821 	task->op = op;
1822 	task->value = value;
1823 
1824 	strlcpy(task->name1, name1, sizeof (task->name1));
1825 	if (name2 != NULL)
1826 		strlcpy(task->name2, name2, sizeof (task->name2));
1827 
1828 	return (task);
1829 }
1830 
1831 static void
1832 zvol_task_free(zvol_task_t *task)
1833 {
1834 	kmem_free(task, sizeof (zvol_task_t));
1835 }
1836 
1837 /*
1838  * The worker thread function performed asynchronously.
1839  */
1840 static void
1841 zvol_task_cb(void *arg)
1842 {
1843 	zvol_task_t *task = arg;
1844 
1845 	switch (task->op) {
1846 	case ZVOL_ASYNC_REMOVE_MINORS:
1847 		zvol_remove_minors_impl(task->name1);
1848 		break;
1849 	case ZVOL_ASYNC_RENAME_MINORS:
1850 		zvol_rename_minors_impl(task->name1, task->name2);
1851 		break;
1852 	case ZVOL_ASYNC_SET_SNAPDEV:
1853 		zvol_set_snapdev_impl(task->name1, task->value);
1854 		break;
1855 	case ZVOL_ASYNC_SET_VOLMODE:
1856 		zvol_set_volmode_impl(task->name1, task->value);
1857 		break;
1858 	default:
1859 		VERIFY(0);
1860 		break;
1861 	}
1862 
1863 	zvol_task_free(task);
1864 }
1865 
1866 typedef struct zvol_set_prop_int_arg {
1867 	const char *zsda_name;
1868 	uint64_t zsda_value;
1869 	zprop_source_t zsda_source;
1870 	zfs_prop_t zsda_prop;
1871 } zvol_set_prop_int_arg_t;
1872 
1873 /*
1874  * Sanity check the dataset for safe use by the sync task.  No additional
1875  * conditions are imposed.
1876  */
1877 static int
1878 zvol_set_common_check(void *arg, dmu_tx_t *tx)
1879 {
1880 	zvol_set_prop_int_arg_t *zsda = arg;
1881 	dsl_pool_t *dp = dmu_tx_pool(tx);
1882 	dsl_dir_t *dd;
1883 	int error;
1884 
1885 	error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL);
1886 	if (error != 0)
1887 		return (error);
1888 
1889 	dsl_dir_rele(dd, FTAG);
1890 
1891 	return (error);
1892 }
1893 
1894 static int
1895 zvol_set_common_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1896 {
1897 	zvol_set_prop_int_arg_t *zsda = arg;
1898 	char dsname[ZFS_MAX_DATASET_NAME_LEN];
1899 	zvol_task_t *task;
1900 	uint64_t prop;
1901 
1902 	const char *prop_name = zfs_prop_to_name(zsda->zsda_prop);
1903 	dsl_dataset_name(ds, dsname);
1904 
1905 	if (dsl_prop_get_int_ds(ds, prop_name, &prop) != 0)
1906 		return (0);
1907 
1908 	switch (zsda->zsda_prop) {
1909 		case ZFS_PROP_VOLMODE:
1910 			task = zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE, dsname,
1911 			    NULL, prop);
1912 			break;
1913 		case ZFS_PROP_SNAPDEV:
1914 			task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname,
1915 			    NULL, prop);
1916 			break;
1917 		default:
1918 			task = NULL;
1919 			break;
1920 	}
1921 
1922 	if (task == NULL)
1923 		return (0);
1924 
1925 	(void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb,
1926 	    task, TQ_SLEEP);
1927 	return (0);
1928 }
1929 
1930 /*
1931  * Traverse all child datasets and apply the property appropriately.
1932  * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
1933  * dataset and read the effective "property" on every child in the callback
1934  * function: this is because the value is not guaranteed to be the same in the
1935  * whole dataset hierarchy.
1936  */
1937 static void
1938 zvol_set_common_sync(void *arg, dmu_tx_t *tx)
1939 {
1940 	zvol_set_prop_int_arg_t *zsda = arg;
1941 	dsl_pool_t *dp = dmu_tx_pool(tx);
1942 	dsl_dir_t *dd;
1943 	dsl_dataset_t *ds;
1944 	int error;
1945 
1946 	VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL));
1947 
1948 	error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds);
1949 	if (error == 0) {
1950 		dsl_prop_set_sync_impl(ds, zfs_prop_to_name(zsda->zsda_prop),
1951 		    zsda->zsda_source, sizeof (zsda->zsda_value), 1,
1952 		    &zsda->zsda_value, tx);
1953 		dsl_dataset_rele(ds, FTAG);
1954 	}
1955 
1956 	dmu_objset_find_dp(dp, dd->dd_object, zvol_set_common_sync_cb,
1957 	    zsda, DS_FIND_CHILDREN);
1958 
1959 	dsl_dir_rele(dd, FTAG);
1960 }
1961 
1962 int
1963 zvol_set_common(const char *ddname, zfs_prop_t prop, zprop_source_t source,
1964     uint64_t val)
1965 {
1966 	zvol_set_prop_int_arg_t zsda;
1967 
1968 	zsda.zsda_name = ddname;
1969 	zsda.zsda_source = source;
1970 	zsda.zsda_value = val;
1971 	zsda.zsda_prop = prop;
1972 
1973 	return (dsl_sync_task(ddname, zvol_set_common_check,
1974 	    zvol_set_common_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
1975 }
1976 
1977 void
1978 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async)
1979 {
1980 	zvol_task_t *task;
1981 	taskqid_t id;
1982 
1983 	task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL);
1984 	if (task == NULL)
1985 		return;
1986 
1987 	id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
1988 	if ((async == B_FALSE) && (id != TASKQID_INVALID))
1989 		taskq_wait_id(spa->spa_zvol_taskq, id);
1990 }
1991 
1992 void
1993 zvol_rename_minors(spa_t *spa, const char *name1, const char *name2,
1994     boolean_t async)
1995 {
1996 	zvol_task_t *task;
1997 	taskqid_t id;
1998 
1999 	task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL);
2000 	if (task == NULL)
2001 		return;
2002 
2003 	id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
2004 	if ((async == B_FALSE) && (id != TASKQID_INVALID))
2005 		taskq_wait_id(spa->spa_zvol_taskq, id);
2006 }
2007 
2008 boolean_t
2009 zvol_is_zvol(const char *name)
2010 {
2011 
2012 	return (zvol_os_is_zvol(name));
2013 }
2014 
2015 int
2016 zvol_init_impl(void)
2017 {
2018 	int i;
2019 
2020 	list_create(&zvol_state_list, sizeof (zvol_state_t),
2021 	    offsetof(zvol_state_t, zv_next));
2022 	rw_init(&zvol_state_lock, NULL, RW_DEFAULT, NULL);
2023 
2024 	zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head),
2025 	    KM_SLEEP);
2026 	for (i = 0; i < ZVOL_HT_SIZE; i++)
2027 		INIT_HLIST_HEAD(&zvol_htable[i]);
2028 
2029 	return (0);
2030 }
2031 
2032 void
2033 zvol_fini_impl(void)
2034 {
2035 	zvol_remove_minors_impl(NULL);
2036 
2037 	/*
2038 	 * The call to "zvol_remove_minors_impl" may dispatch entries to
2039 	 * the system_taskq, but it doesn't wait for those entries to
2040 	 * complete before it returns. Thus, we must wait for all of the
2041 	 * removals to finish, before we can continue.
2042 	 */
2043 	taskq_wait_outstanding(system_taskq, 0);
2044 
2045 	kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
2046 	list_destroy(&zvol_state_list);
2047 	rw_destroy(&zvol_state_lock);
2048 }
2049