xref: /illumos-gate/usr/src/uts/common/fs/zfs/zvol.c (revision 455e370ca67aeea268bdfbcf581b6a05547f6636)
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  *
24  * Portions Copyright 2010 Robert Milkowski
25  *
26  * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
27  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
28  * Copyright (c) 2014 Integros [integros.com]
29  * Copyright 2019 Joyent, Inc.
30  */
31 
32 /*
33  * ZFS volume emulation driver.
34  *
35  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
36  * Volumes are accessed through the symbolic links named:
37  *
38  * /dev/zvol/dsk/<pool_name>/<dataset_name>
39  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
40  *
41  * These links are created by the /dev filesystem (sdev_zvolops.c).
42  * Volumes are persistent through reboot.  No user command needs to be
43  * run before opening and using a device.
44  */
45 
46 #include <sys/types.h>
47 #include <sys/param.h>
48 #include <sys/errno.h>
49 #include <sys/uio.h>
50 #include <sys/buf.h>
51 #include <sys/modctl.h>
52 #include <sys/open.h>
53 #include <sys/kmem.h>
54 #include <sys/conf.h>
55 #include <sys/cmn_err.h>
56 #include <sys/stat.h>
57 #include <sys/zap.h>
58 #include <sys/spa.h>
59 #include <sys/spa_impl.h>
60 #include <sys/zio.h>
61 #include <sys/dmu_traverse.h>
62 #include <sys/dnode.h>
63 #include <sys/dsl_dataset.h>
64 #include <sys/dsl_prop.h>
65 #include <sys/dkio.h>
66 #include <sys/efi_partition.h>
67 #include <sys/byteorder.h>
68 #include <sys/pathname.h>
69 #include <sys/ddi.h>
70 #include <sys/sunddi.h>
71 #include <sys/crc32.h>
72 #include <sys/dirent.h>
73 #include <sys/policy.h>
74 #include <sys/fs/zfs.h>
75 #include <sys/zfs_ioctl.h>
76 #include <sys/mkdev.h>
77 #include <sys/zil.h>
78 #include <sys/refcount.h>
79 #include <sys/zfs_znode.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/vdev_disk.h>
82 #include <sys/vdev_impl.h>
83 #include <sys/vdev_raidz.h>
84 #include <sys/zvol.h>
85 #include <sys/dumphdr.h>
86 #include <sys/zil_impl.h>
87 #include <sys/dbuf.h>
88 #include <sys/dmu_tx.h>
89 #include <sys/zfeature.h>
90 #include <sys/zio_checksum.h>
91 #include <sys/zil_impl.h>
92 #include <sys/ht.h>
93 #include <sys/dkioc_free_util.h>
94 #include <sys/zfs_rlock.h>
95 
96 #include "zfs_namecheck.h"
97 
98 void *zfsdev_state;
99 static char *zvol_tag = "zvol_tag";
100 
101 #define	ZVOL_DUMPSIZE		"dumpsize"
102 
103 /*
104  * This lock protects the zfsdev_state structure from being modified
105  * while it's being used, e.g. an open that comes in before a create
106  * finishes.  It also protects temporary opens of the dataset so that,
107  * e.g., an open doesn't get a spurious EBUSY.
108  */
109 kmutex_t zfsdev_state_lock;
110 static uint32_t zvol_minors;
111 
112 typedef struct zvol_extent {
113 	list_node_t	ze_node;
114 	dva_t		ze_dva;		/* dva associated with this extent */
115 	uint64_t	ze_nblks;	/* number of blocks in extent */
116 } zvol_extent_t;
117 
118 /*
119  * The in-core state of each volume.
120  */
121 typedef struct zvol_state {
122 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
123 	uint64_t	zv_volsize;	/* amount of space we advertise */
124 	uint64_t	zv_volblocksize; /* volume block size */
125 	minor_t		zv_minor;	/* minor number */
126 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
127 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
128 	objset_t	*zv_objset;	/* objset handle */
129 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
130 	uint32_t	zv_total_opens;	/* total open count */
131 	zilog_t		*zv_zilog;	/* ZIL handle */
132 	list_t		zv_extents;	/* List of extents for dump */
133 	rangelock_t	zv_rangelock;
134 	dnode_t		*zv_dn;		/* dnode hold */
135 } zvol_state_t;
136 
137 /*
138  * zvol specific flags
139  */
140 #define	ZVOL_RDONLY	0x1
141 #define	ZVOL_DUMPIFIED	0x2
142 #define	ZVOL_EXCL	0x4
143 #define	ZVOL_WCE	0x8
144 
145 /*
146  * zvol maximum transfer in one DMU tx.
147  */
148 int zvol_maxphys = DMU_MAX_ACCESS/2;
149 
150 /*
151  * Toggle unmap functionality.
152  */
153 boolean_t zvol_unmap_enabled = B_TRUE;
154 
155 /*
156  * If true, unmaps requested as synchronous are executed synchronously,
157  * otherwise all unmaps are asynchronous.
158  */
159 boolean_t zvol_unmap_sync_enabled = B_FALSE;
160 
161 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
162     nvlist_t *, nvlist_t *);
163 static int zvol_remove_zv(zvol_state_t *);
164 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
165     struct lwb *lwb, zio_t *zio);
166 static int zvol_dumpify(zvol_state_t *zv);
167 static int zvol_dump_fini(zvol_state_t *zv);
168 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
169 
170 static void
171 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
172 {
173 	dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
174 
175 	zv->zv_volsize = volsize;
176 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
177 	    "Size", volsize) == DDI_SUCCESS);
178 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
179 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
180 
181 	/* Notify specfs to invalidate the cached size */
182 	spec_size_invalidate(dev, VBLK);
183 	spec_size_invalidate(dev, VCHR);
184 }
185 
186 int
187 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
188 {
189 	if (volsize == 0)
190 		return (SET_ERROR(EINVAL));
191 
192 	if (volsize % blocksize != 0)
193 		return (SET_ERROR(EINVAL));
194 
195 #ifdef _ILP32
196 	if (volsize - 1 > SPEC_MAXOFFSET_T)
197 		return (SET_ERROR(EOVERFLOW));
198 #endif
199 	return (0);
200 }
201 
202 int
203 zvol_check_volblocksize(uint64_t volblocksize)
204 {
205 	if (volblocksize < SPA_MINBLOCKSIZE ||
206 	    volblocksize > SPA_OLD_MAXBLOCKSIZE ||
207 	    !ISP2(volblocksize))
208 		return (SET_ERROR(EDOM));
209 
210 	return (0);
211 }
212 
213 int
214 zvol_get_stats(objset_t *os, nvlist_t *nv)
215 {
216 	int error;
217 	dmu_object_info_t doi;
218 	uint64_t val;
219 
220 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
221 	if (error)
222 		return (error);
223 
224 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
225 
226 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
227 
228 	if (error == 0) {
229 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
230 		    doi.doi_data_block_size);
231 	}
232 
233 	return (error);
234 }
235 
236 static zvol_state_t *
237 zvol_minor_lookup(const char *name)
238 {
239 	minor_t minor;
240 	zvol_state_t *zv;
241 
242 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
243 
244 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
245 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
246 		if (zv == NULL)
247 			continue;
248 		if (strcmp(zv->zv_name, name) == 0)
249 			return (zv);
250 	}
251 
252 	return (NULL);
253 }
254 
255 /* extent mapping arg */
256 struct maparg {
257 	zvol_state_t	*ma_zv;
258 	uint64_t	ma_blks;
259 };
260 
261 /*ARGSUSED*/
262 static int
263 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
264     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
265 {
266 	struct maparg *ma = arg;
267 	zvol_extent_t *ze;
268 	int bs = ma->ma_zv->zv_volblocksize;
269 
270 	if (bp == NULL || BP_IS_HOLE(bp) ||
271 	    zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
272 		return (0);
273 
274 	VERIFY(!BP_IS_EMBEDDED(bp));
275 
276 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
277 	ma->ma_blks++;
278 
279 	/* Abort immediately if we have encountered gang blocks */
280 	if (BP_IS_GANG(bp))
281 		return (SET_ERROR(EFRAGS));
282 
283 	/*
284 	 * See if the block is at the end of the previous extent.
285 	 */
286 	ze = list_tail(&ma->ma_zv->zv_extents);
287 	if (ze &&
288 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
289 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
290 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
291 		ze->ze_nblks++;
292 		return (0);
293 	}
294 
295 	dprintf_bp(bp, "%s", "next blkptr:");
296 
297 	/* start a new extent */
298 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
299 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
300 	ze->ze_nblks = 1;
301 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
302 	return (0);
303 }
304 
305 static void
306 zvol_free_extents(zvol_state_t *zv)
307 {
308 	zvol_extent_t *ze;
309 
310 	while (ze = list_head(&zv->zv_extents)) {
311 		list_remove(&zv->zv_extents, ze);
312 		kmem_free(ze, sizeof (zvol_extent_t));
313 	}
314 }
315 
316 static int
317 zvol_get_lbas(zvol_state_t *zv)
318 {
319 	objset_t *os = zv->zv_objset;
320 	struct maparg	ma;
321 	int		err;
322 
323 	ma.ma_zv = zv;
324 	ma.ma_blks = 0;
325 	zvol_free_extents(zv);
326 
327 	/* commit any in-flight changes before traversing the dataset */
328 	txg_wait_synced(dmu_objset_pool(os), 0);
329 	err = traverse_dataset(dmu_objset_ds(os), 0,
330 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
331 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
332 		zvol_free_extents(zv);
333 		return (err ? err : EIO);
334 	}
335 
336 	return (0);
337 }
338 
339 /* ARGSUSED */
340 void
341 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
342 {
343 	zfs_creat_t *zct = arg;
344 	nvlist_t *nvprops = zct->zct_props;
345 	int error;
346 	uint64_t volblocksize, volsize;
347 
348 	VERIFY(nvlist_lookup_uint64(nvprops,
349 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
350 	if (nvlist_lookup_uint64(nvprops,
351 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
352 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
353 
354 	/*
355 	 * These properties must be removed from the list so the generic
356 	 * property setting step won't apply to them.
357 	 */
358 	VERIFY(nvlist_remove_all(nvprops,
359 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
360 	(void) nvlist_remove_all(nvprops,
361 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
362 
363 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
364 	    DMU_OT_NONE, 0, tx);
365 	ASSERT(error == 0);
366 
367 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
368 	    DMU_OT_NONE, 0, tx);
369 	ASSERT(error == 0);
370 
371 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
372 	ASSERT(error == 0);
373 }
374 
375 /*
376  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
377  * implement DKIOCFREE/free-long-range.
378  */
379 static int
380 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
381 {
382 	zvol_state_t *zv = arg1;
383 	lr_truncate_t *lr = arg2;
384 	uint64_t offset, length;
385 
386 	if (byteswap)
387 		byteswap_uint64_array(lr, sizeof (*lr));
388 
389 	offset = lr->lr_offset;
390 	length = lr->lr_length;
391 
392 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
393 }
394 
395 /*
396  * Replay a TX_WRITE ZIL transaction that didn't get committed
397  * after a system failure
398  */
399 static int
400 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
401 {
402 	zvol_state_t *zv = arg1;
403 	lr_write_t *lr = arg2;
404 	objset_t *os = zv->zv_objset;
405 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
406 	uint64_t offset, length;
407 	dmu_tx_t *tx;
408 	int error;
409 
410 	if (byteswap)
411 		byteswap_uint64_array(lr, sizeof (*lr));
412 
413 	offset = lr->lr_offset;
414 	length = lr->lr_length;
415 
416 	/* If it's a dmu_sync() block, write the whole block */
417 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
418 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
419 		if (length < blocksize) {
420 			offset -= offset % blocksize;
421 			length = blocksize;
422 		}
423 	}
424 
425 	tx = dmu_tx_create(os);
426 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
427 	error = dmu_tx_assign(tx, TXG_WAIT);
428 	if (error) {
429 		dmu_tx_abort(tx);
430 	} else {
431 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
432 		dmu_tx_commit(tx);
433 	}
434 
435 	return (error);
436 }
437 
438 /* ARGSUSED */
439 static int
440 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
441 {
442 	return (SET_ERROR(ENOTSUP));
443 }
444 
445 /*
446  * Callback vectors for replaying records.
447  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
448  */
449 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
450 	zvol_replay_err,	/* 0 no such transaction type */
451 	zvol_replay_err,	/* TX_CREATE */
452 	zvol_replay_err,	/* TX_MKDIR */
453 	zvol_replay_err,	/* TX_MKXATTR */
454 	zvol_replay_err,	/* TX_SYMLINK */
455 	zvol_replay_err,	/* TX_REMOVE */
456 	zvol_replay_err,	/* TX_RMDIR */
457 	zvol_replay_err,	/* TX_LINK */
458 	zvol_replay_err,	/* TX_RENAME */
459 	zvol_replay_write,	/* TX_WRITE */
460 	zvol_replay_truncate,	/* TX_TRUNCATE */
461 	zvol_replay_err,	/* TX_SETATTR */
462 	zvol_replay_err,	/* TX_ACL */
463 	zvol_replay_err,	/* TX_CREATE_ACL */
464 	zvol_replay_err,	/* TX_CREATE_ATTR */
465 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
466 	zvol_replay_err,	/* TX_MKDIR_ACL */
467 	zvol_replay_err,	/* TX_MKDIR_ATTR */
468 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
469 	zvol_replay_err,	/* TX_WRITE2 */
470 };
471 
472 int
473 zvol_name2minor(const char *name, minor_t *minor)
474 {
475 	zvol_state_t *zv;
476 
477 	mutex_enter(&zfsdev_state_lock);
478 	zv = zvol_minor_lookup(name);
479 	if (minor && zv)
480 		*minor = zv->zv_minor;
481 	mutex_exit(&zfsdev_state_lock);
482 	return (zv ? 0 : -1);
483 }
484 
485 /*
486  * Create a minor node (plus a whole lot more) for the specified volume.
487  */
488 int
489 zvol_create_minor(const char *name)
490 {
491 	zfs_soft_state_t *zs;
492 	zvol_state_t *zv;
493 	objset_t *os;
494 	dmu_object_info_t doi;
495 	minor_t minor = 0;
496 	char chrbuf[30], blkbuf[30];
497 	int error;
498 
499 	mutex_enter(&zfsdev_state_lock);
500 
501 	if (zvol_minor_lookup(name) != NULL) {
502 		mutex_exit(&zfsdev_state_lock);
503 		return (SET_ERROR(EEXIST));
504 	}
505 
506 	/* lie and say we're read-only */
507 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
508 
509 	if (error) {
510 		mutex_exit(&zfsdev_state_lock);
511 		return (error);
512 	}
513 
514 	if ((minor = zfsdev_minor_alloc()) == 0) {
515 		dmu_objset_disown(os, FTAG);
516 		mutex_exit(&zfsdev_state_lock);
517 		return (SET_ERROR(ENXIO));
518 	}
519 
520 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
521 		dmu_objset_disown(os, FTAG);
522 		mutex_exit(&zfsdev_state_lock);
523 		return (SET_ERROR(EAGAIN));
524 	}
525 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
526 	    (char *)name);
527 
528 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
529 
530 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
531 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
532 		ddi_soft_state_free(zfsdev_state, minor);
533 		dmu_objset_disown(os, FTAG);
534 		mutex_exit(&zfsdev_state_lock);
535 		return (SET_ERROR(EAGAIN));
536 	}
537 
538 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
539 
540 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
541 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
542 		ddi_remove_minor_node(zfs_dip, chrbuf);
543 		ddi_soft_state_free(zfsdev_state, minor);
544 		dmu_objset_disown(os, FTAG);
545 		mutex_exit(&zfsdev_state_lock);
546 		return (SET_ERROR(EAGAIN));
547 	}
548 
549 	zs = ddi_get_soft_state(zfsdev_state, minor);
550 	zs->zss_type = ZSST_ZVOL;
551 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
552 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
553 	zv->zv_min_bs = DEV_BSHIFT;
554 	zv->zv_minor = minor;
555 	zv->zv_objset = os;
556 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
557 		zv->zv_flags |= ZVOL_RDONLY;
558 	rangelock_init(&zv->zv_rangelock, NULL, NULL);
559 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
560 	    offsetof(zvol_extent_t, ze_node));
561 	/* get and cache the blocksize */
562 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
563 	ASSERT(error == 0);
564 	zv->zv_volblocksize = doi.doi_data_block_size;
565 
566 	if (spa_writeable(dmu_objset_spa(os))) {
567 		if (zil_replay_disable)
568 			zil_destroy(dmu_objset_zil(os), B_FALSE);
569 		else
570 			zil_replay(os, zv, zvol_replay_vector);
571 	}
572 	dmu_objset_disown(os, FTAG);
573 	zv->zv_objset = NULL;
574 
575 	zvol_minors++;
576 
577 	mutex_exit(&zfsdev_state_lock);
578 
579 	return (0);
580 }
581 
582 /*
583  * Remove minor node for the specified volume.
584  */
585 static int
586 zvol_remove_zv(zvol_state_t *zv)
587 {
588 	char nmbuf[20];
589 	minor_t minor = zv->zv_minor;
590 
591 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
592 	if (zv->zv_total_opens != 0)
593 		return (SET_ERROR(EBUSY));
594 
595 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
596 	ddi_remove_minor_node(zfs_dip, nmbuf);
597 
598 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
599 	ddi_remove_minor_node(zfs_dip, nmbuf);
600 
601 	rangelock_fini(&zv->zv_rangelock);
602 
603 	kmem_free(zv, sizeof (zvol_state_t));
604 
605 	ddi_soft_state_free(zfsdev_state, minor);
606 
607 	zvol_minors--;
608 	return (0);
609 }
610 
611 int
612 zvol_remove_minor(const char *name)
613 {
614 	zvol_state_t *zv;
615 	int rc;
616 
617 	mutex_enter(&zfsdev_state_lock);
618 	if ((zv = zvol_minor_lookup(name)) == NULL) {
619 		mutex_exit(&zfsdev_state_lock);
620 		return (SET_ERROR(ENXIO));
621 	}
622 	rc = zvol_remove_zv(zv);
623 	mutex_exit(&zfsdev_state_lock);
624 	return (rc);
625 }
626 
627 int
628 zvol_first_open(zvol_state_t *zv)
629 {
630 	objset_t *os;
631 	uint64_t volsize;
632 	int error;
633 	uint64_t readonly;
634 
635 	/* lie and say we're read-only */
636 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
637 	    zvol_tag, &os);
638 	if (error)
639 		return (error);
640 
641 	zv->zv_objset = os;
642 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
643 	if (error) {
644 		ASSERT(error == 0);
645 		dmu_objset_disown(os, zvol_tag);
646 		return (error);
647 	}
648 
649 	error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
650 	if (error) {
651 		dmu_objset_disown(os, zvol_tag);
652 		return (error);
653 	}
654 
655 	zvol_size_changed(zv, volsize);
656 	zv->zv_zilog = zil_open(os, zvol_get_data);
657 
658 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
659 	    NULL) == 0);
660 	if (readonly || dmu_objset_is_snapshot(os) ||
661 	    !spa_writeable(dmu_objset_spa(os)))
662 		zv->zv_flags |= ZVOL_RDONLY;
663 	else
664 		zv->zv_flags &= ~ZVOL_RDONLY;
665 	return (error);
666 }
667 
668 void
669 zvol_last_close(zvol_state_t *zv)
670 {
671 	zil_close(zv->zv_zilog);
672 	zv->zv_zilog = NULL;
673 
674 	dnode_rele(zv->zv_dn, zvol_tag);
675 	zv->zv_dn = NULL;
676 
677 	/*
678 	 * Evict cached data
679 	 */
680 	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
681 	    !(zv->zv_flags & ZVOL_RDONLY))
682 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
683 	dmu_objset_evict_dbufs(zv->zv_objset);
684 
685 	dmu_objset_disown(zv->zv_objset, zvol_tag);
686 	zv->zv_objset = NULL;
687 }
688 
689 int
690 zvol_prealloc(zvol_state_t *zv)
691 {
692 	objset_t *os = zv->zv_objset;
693 	dmu_tx_t *tx;
694 	uint64_t refd, avail, usedobjs, availobjs;
695 	uint64_t resid = zv->zv_volsize;
696 	uint64_t off = 0;
697 
698 	/* Check the space usage before attempting to allocate the space */
699 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
700 	if (avail < zv->zv_volsize)
701 		return (SET_ERROR(ENOSPC));
702 
703 	/* Free old extents if they exist */
704 	zvol_free_extents(zv);
705 
706 	while (resid != 0) {
707 		int error;
708 		uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
709 
710 		tx = dmu_tx_create(os);
711 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
712 		error = dmu_tx_assign(tx, TXG_WAIT);
713 		if (error) {
714 			dmu_tx_abort(tx);
715 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
716 			return (error);
717 		}
718 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
719 		dmu_tx_commit(tx);
720 		off += bytes;
721 		resid -= bytes;
722 	}
723 	txg_wait_synced(dmu_objset_pool(os), 0);
724 
725 	return (0);
726 }
727 
728 static int
729 zvol_update_volsize(objset_t *os, uint64_t volsize)
730 {
731 	dmu_tx_t *tx;
732 	int error;
733 
734 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
735 
736 	tx = dmu_tx_create(os);
737 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
738 	dmu_tx_mark_netfree(tx);
739 	error = dmu_tx_assign(tx, TXG_WAIT);
740 	if (error) {
741 		dmu_tx_abort(tx);
742 		return (error);
743 	}
744 
745 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
746 	    &volsize, tx);
747 	dmu_tx_commit(tx);
748 
749 	if (error == 0)
750 		error = dmu_free_long_range(os,
751 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
752 	return (error);
753 }
754 
755 void
756 zvol_remove_minors(const char *name)
757 {
758 	zvol_state_t *zv;
759 	char *namebuf;
760 	minor_t minor;
761 
762 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
763 	(void) strncpy(namebuf, name, strlen(name));
764 	(void) strcat(namebuf, "/");
765 	mutex_enter(&zfsdev_state_lock);
766 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
767 
768 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
769 		if (zv == NULL)
770 			continue;
771 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
772 			(void) zvol_remove_zv(zv);
773 	}
774 	kmem_free(namebuf, strlen(name) + 2);
775 
776 	mutex_exit(&zfsdev_state_lock);
777 }
778 
779 static int
780 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
781 {
782 	uint64_t old_volsize = 0ULL;
783 	int error = 0;
784 
785 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
786 
787 	/*
788 	 * Reinitialize the dump area to the new size. If we
789 	 * failed to resize the dump area then restore it back to
790 	 * its original size.  We must set the new volsize prior
791 	 * to calling dumpvp_resize() to ensure that the devices'
792 	 * size(9P) is not visible by the dump subsystem.
793 	 */
794 	old_volsize = zv->zv_volsize;
795 	zvol_size_changed(zv, volsize);
796 
797 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
798 		if ((error = zvol_dumpify(zv)) != 0 ||
799 		    (error = dumpvp_resize()) != 0) {
800 			int dumpify_error;
801 
802 			(void) zvol_update_volsize(zv->zv_objset, old_volsize);
803 			zvol_size_changed(zv, old_volsize);
804 			dumpify_error = zvol_dumpify(zv);
805 			error = dumpify_error ? dumpify_error : error;
806 		}
807 	}
808 
809 	/*
810 	 * Generate a LUN expansion event.
811 	 */
812 	if (error == 0) {
813 		sysevent_id_t eid;
814 		nvlist_t *attr;
815 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
816 
817 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
818 		    zv->zv_minor);
819 
820 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
821 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
822 
823 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
824 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
825 
826 		nvlist_free(attr);
827 		kmem_free(physpath, MAXPATHLEN);
828 	}
829 	return (error);
830 }
831 
832 int
833 zvol_set_volsize(const char *name, uint64_t volsize)
834 {
835 	zvol_state_t *zv = NULL;
836 	objset_t *os;
837 	int error;
838 	dmu_object_info_t doi;
839 	uint64_t readonly;
840 	boolean_t owned = B_FALSE;
841 
842 	error = dsl_prop_get_integer(name,
843 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
844 	if (error != 0)
845 		return (error);
846 	if (readonly)
847 		return (SET_ERROR(EROFS));
848 
849 	mutex_enter(&zfsdev_state_lock);
850 	zv = zvol_minor_lookup(name);
851 
852 	if (zv == NULL || zv->zv_objset == NULL) {
853 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
854 		    FTAG, &os)) != 0) {
855 			mutex_exit(&zfsdev_state_lock);
856 			return (error);
857 		}
858 		owned = B_TRUE;
859 		if (zv != NULL)
860 			zv->zv_objset = os;
861 	} else {
862 		os = zv->zv_objset;
863 	}
864 
865 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
866 	    (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
867 		goto out;
868 
869 	error = zvol_update_volsize(os, volsize);
870 
871 	if (error == 0 && zv != NULL)
872 		error = zvol_update_live_volsize(zv, volsize);
873 out:
874 	if (owned) {
875 		dmu_objset_disown(os, FTAG);
876 		if (zv != NULL)
877 			zv->zv_objset = NULL;
878 	}
879 	mutex_exit(&zfsdev_state_lock);
880 	return (error);
881 }
882 
883 /*ARGSUSED*/
884 int
885 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
886 {
887 	zvol_state_t *zv;
888 	int err = 0;
889 
890 	mutex_enter(&zfsdev_state_lock);
891 
892 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
893 	if (zv == NULL) {
894 		mutex_exit(&zfsdev_state_lock);
895 		return (SET_ERROR(ENXIO));
896 	}
897 
898 	if (zv->zv_total_opens == 0)
899 		err = zvol_first_open(zv);
900 	if (err) {
901 		mutex_exit(&zfsdev_state_lock);
902 		return (err);
903 	}
904 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
905 		err = SET_ERROR(EROFS);
906 		goto out;
907 	}
908 	if (zv->zv_flags & ZVOL_EXCL) {
909 		err = SET_ERROR(EBUSY);
910 		goto out;
911 	}
912 	if (flag & FEXCL) {
913 		if (zv->zv_total_opens != 0) {
914 			err = SET_ERROR(EBUSY);
915 			goto out;
916 		}
917 		zv->zv_flags |= ZVOL_EXCL;
918 	}
919 
920 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
921 		zv->zv_open_count[otyp]++;
922 		zv->zv_total_opens++;
923 	}
924 	mutex_exit(&zfsdev_state_lock);
925 
926 	return (err);
927 out:
928 	if (zv->zv_total_opens == 0)
929 		zvol_last_close(zv);
930 	mutex_exit(&zfsdev_state_lock);
931 	return (err);
932 }
933 
934 /*ARGSUSED*/
935 int
936 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
937 {
938 	minor_t minor = getminor(dev);
939 	zvol_state_t *zv;
940 	int error = 0;
941 
942 	mutex_enter(&zfsdev_state_lock);
943 
944 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
945 	if (zv == NULL) {
946 		mutex_exit(&zfsdev_state_lock);
947 		return (SET_ERROR(ENXIO));
948 	}
949 
950 	if (zv->zv_flags & ZVOL_EXCL) {
951 		ASSERT(zv->zv_total_opens == 1);
952 		zv->zv_flags &= ~ZVOL_EXCL;
953 	}
954 
955 	/*
956 	 * If the open count is zero, this is a spurious close.
957 	 * That indicates a bug in the kernel / DDI framework.
958 	 */
959 	ASSERT(zv->zv_open_count[otyp] != 0);
960 	ASSERT(zv->zv_total_opens != 0);
961 
962 	/*
963 	 * You may get multiple opens, but only one close.
964 	 */
965 	zv->zv_open_count[otyp]--;
966 	zv->zv_total_opens--;
967 
968 	if (zv->zv_total_opens == 0)
969 		zvol_last_close(zv);
970 
971 	mutex_exit(&zfsdev_state_lock);
972 	return (error);
973 }
974 
975 /* ARGSUSED */
976 static void
977 zvol_get_done(zgd_t *zgd, int error)
978 {
979 	if (zgd->zgd_db)
980 		dmu_buf_rele(zgd->zgd_db, zgd);
981 
982 	rangelock_exit(zgd->zgd_lr);
983 
984 	kmem_free(zgd, sizeof (zgd_t));
985 }
986 
987 /*
988  * Get data to generate a TX_WRITE intent log record.
989  */
990 static int
991 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
992 {
993 	zvol_state_t *zv = arg;
994 	uint64_t offset = lr->lr_offset;
995 	uint64_t size = lr->lr_length;	/* length of user data */
996 	dmu_buf_t *db;
997 	zgd_t *zgd;
998 	int error;
999 
1000 	ASSERT3P(lwb, !=, NULL);
1001 	ASSERT3P(zio, !=, NULL);
1002 	ASSERT3U(size, !=, 0);
1003 
1004 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1005 	zgd->zgd_lwb = lwb;
1006 
1007 	/*
1008 	 * Write records come in two flavors: immediate and indirect.
1009 	 * For small writes it's cheaper to store the data with the
1010 	 * log record (immediate); for large writes it's cheaper to
1011 	 * sync the data and get a pointer to it (indirect) so that
1012 	 * we don't have to write the data twice.
1013 	 */
1014 	if (buf != NULL) { /* immediate write */
1015 		zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1016 		    RL_READER);
1017 		error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1018 		    DMU_READ_NO_PREFETCH);
1019 	} else { /* indirect write */
1020 		/*
1021 		 * Have to lock the whole block to ensure when it's written out
1022 		 * and its checksum is being calculated that no one can change
1023 		 * the data. Contrarily to zfs_get_data we need not re-check
1024 		 * blocksize after we get the lock because it cannot be changed.
1025 		 */
1026 		size = zv->zv_volblocksize;
1027 		offset = P2ALIGN(offset, size);
1028 		zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1029 		    RL_READER);
1030 		error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1031 		    DMU_READ_NO_PREFETCH);
1032 		if (error == 0) {
1033 			blkptr_t *bp = &lr->lr_blkptr;
1034 
1035 			zgd->zgd_db = db;
1036 			zgd->zgd_bp = bp;
1037 
1038 			ASSERT(db->db_offset == offset);
1039 			ASSERT(db->db_size == size);
1040 
1041 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1042 			    zvol_get_done, zgd);
1043 
1044 			if (error == 0)
1045 				return (0);
1046 		}
1047 	}
1048 
1049 	zvol_get_done(zgd, error);
1050 
1051 	return (error);
1052 }
1053 
1054 /*
1055  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1056  *
1057  * We store data in the log buffers if it's small enough.
1058  * Otherwise we will later flush the data out via dmu_sync().
1059  */
1060 ssize_t zvol_immediate_write_sz = 32768;
1061 
1062 static void
1063 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1064     boolean_t sync)
1065 {
1066 	uint32_t blocksize = zv->zv_volblocksize;
1067 	zilog_t *zilog = zv->zv_zilog;
1068 	itx_wr_state_t write_state;
1069 
1070 	if (zil_replaying(zilog, tx))
1071 		return;
1072 
1073 	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1074 		write_state = WR_INDIRECT;
1075 	else if (!spa_has_slogs(zilog->zl_spa) &&
1076 	    resid >= blocksize && blocksize > zvol_immediate_write_sz)
1077 		write_state = WR_INDIRECT;
1078 	else if (sync)
1079 		write_state = WR_COPIED;
1080 	else
1081 		write_state = WR_NEED_COPY;
1082 
1083 	while (resid) {
1084 		itx_t *itx;
1085 		lr_write_t *lr;
1086 		itx_wr_state_t wr_state = write_state;
1087 		ssize_t len = resid;
1088 
1089 		if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1090 			wr_state = WR_NEED_COPY;
1091 		else if (wr_state == WR_INDIRECT)
1092 			len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1093 
1094 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1095 		    (wr_state == WR_COPIED ? len : 0));
1096 		lr = (lr_write_t *)&itx->itx_lr;
1097 		if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1098 		    off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1099 			zil_itx_destroy(itx);
1100 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1101 			lr = (lr_write_t *)&itx->itx_lr;
1102 			wr_state = WR_NEED_COPY;
1103 		}
1104 
1105 		itx->itx_wr_state = wr_state;
1106 		lr->lr_foid = ZVOL_OBJ;
1107 		lr->lr_offset = off;
1108 		lr->lr_length = len;
1109 		lr->lr_blkoff = 0;
1110 		BP_ZERO(&lr->lr_blkptr);
1111 
1112 		itx->itx_private = zv;
1113 		itx->itx_sync = sync;
1114 
1115 		zil_itx_assign(zilog, itx, tx);
1116 
1117 		off += len;
1118 		resid -= len;
1119 	}
1120 }
1121 
1122 static int
1123 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1124     uint64_t size, boolean_t doread, boolean_t isdump)
1125 {
1126 	vdev_disk_t *dvd;
1127 	int c;
1128 	int numerrors = 0;
1129 
1130 	if (vd->vdev_ops == &vdev_mirror_ops ||
1131 	    vd->vdev_ops == &vdev_replacing_ops ||
1132 	    vd->vdev_ops == &vdev_spare_ops) {
1133 		for (c = 0; c < vd->vdev_children; c++) {
1134 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1135 			    addr, offset, origoffset, size, doread, isdump);
1136 			if (err != 0) {
1137 				numerrors++;
1138 			} else if (doread) {
1139 				break;
1140 			}
1141 		}
1142 	}
1143 
1144 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1145 		return (numerrors < vd->vdev_children ? 0 : EIO);
1146 
1147 	if (doread && !vdev_readable(vd))
1148 		return (SET_ERROR(EIO));
1149 	else if (!doread && !vdev_writeable(vd))
1150 		return (SET_ERROR(EIO));
1151 
1152 	if (vd->vdev_ops == &vdev_raidz_ops) {
1153 		return (vdev_raidz_physio(vd,
1154 		    addr, size, offset, origoffset, doread, isdump));
1155 	}
1156 
1157 	offset += VDEV_LABEL_START_SIZE;
1158 
1159 	if (ddi_in_panic() || isdump) {
1160 		ASSERT(!doread);
1161 		if (doread)
1162 			return (SET_ERROR(EIO));
1163 		dvd = vd->vdev_tsd;
1164 		ASSERT3P(dvd, !=, NULL);
1165 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1166 		    lbtodb(size)));
1167 	} else {
1168 		dvd = vd->vdev_tsd;
1169 		ASSERT3P(dvd, !=, NULL);
1170 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1171 		    offset, doread ? B_READ : B_WRITE));
1172 	}
1173 }
1174 
1175 static int
1176 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1177     boolean_t doread, boolean_t isdump)
1178 {
1179 	vdev_t *vd;
1180 	int error;
1181 	zvol_extent_t *ze;
1182 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1183 
1184 	/* Must be sector aligned, and not stradle a block boundary. */
1185 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1186 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1187 		return (SET_ERROR(EINVAL));
1188 	}
1189 	ASSERT(size <= zv->zv_volblocksize);
1190 
1191 	/* Locate the extent this belongs to */
1192 	ze = list_head(&zv->zv_extents);
1193 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1194 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1195 		ze = list_next(&zv->zv_extents, ze);
1196 	}
1197 
1198 	if (ze == NULL)
1199 		return (SET_ERROR(EINVAL));
1200 
1201 	if (!ddi_in_panic())
1202 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1203 
1204 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1205 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1206 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1207 	    size, doread, isdump);
1208 
1209 	if (!ddi_in_panic())
1210 		spa_config_exit(spa, SCL_STATE, FTAG);
1211 
1212 	return (error);
1213 }
1214 
1215 int
1216 zvol_strategy(buf_t *bp)
1217 {
1218 	zfs_soft_state_t *zs = NULL;
1219 	zvol_state_t *zv;
1220 	uint64_t off, volsize;
1221 	size_t resid;
1222 	char *addr;
1223 	objset_t *os;
1224 	int error = 0;
1225 	boolean_t doread = bp->b_flags & B_READ;
1226 	boolean_t is_dumpified;
1227 	boolean_t sync;
1228 
1229 	if (getminor(bp->b_edev) == 0) {
1230 		error = SET_ERROR(EINVAL);
1231 	} else {
1232 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1233 		if (zs == NULL)
1234 			error = SET_ERROR(ENXIO);
1235 		else if (zs->zss_type != ZSST_ZVOL)
1236 			error = SET_ERROR(EINVAL);
1237 	}
1238 
1239 	if (error) {
1240 		bioerror(bp, error);
1241 		biodone(bp);
1242 		return (0);
1243 	}
1244 
1245 	zv = zs->zss_data;
1246 
1247 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1248 		bioerror(bp, EROFS);
1249 		biodone(bp);
1250 		return (0);
1251 	}
1252 
1253 	off = ldbtob(bp->b_blkno);
1254 	volsize = zv->zv_volsize;
1255 
1256 	os = zv->zv_objset;
1257 	ASSERT(os != NULL);
1258 
1259 	bp_mapin(bp);
1260 	addr = bp->b_un.b_addr;
1261 	resid = bp->b_bcount;
1262 
1263 	if (resid > 0 && (off < 0 || off >= volsize)) {
1264 		bioerror(bp, EIO);
1265 		biodone(bp);
1266 		return (0);
1267 	}
1268 
1269 	is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1270 	sync = ((!(bp->b_flags & B_ASYNC) &&
1271 	    !(zv->zv_flags & ZVOL_WCE)) ||
1272 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1273 	    !doread && !is_dumpified;
1274 
1275 	ht_begin_unsafe();
1276 
1277 	/*
1278 	 * There must be no buffer changes when doing a dmu_sync() because
1279 	 * we can't change the data whilst calculating the checksum.
1280 	 */
1281 	locked_range_t *lr = rangelock_enter(&zv->zv_rangelock, off, resid,
1282 	    doread ? RL_READER : RL_WRITER);
1283 
1284 	while (resid != 0 && off < volsize) {
1285 		size_t size = MIN(resid, zvol_maxphys);
1286 		if (is_dumpified) {
1287 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1288 			error = zvol_dumpio(zv, addr, off, size,
1289 			    doread, B_FALSE);
1290 		} else if (doread) {
1291 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1292 			    DMU_READ_PREFETCH);
1293 		} else {
1294 			dmu_tx_t *tx = dmu_tx_create(os);
1295 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1296 			error = dmu_tx_assign(tx, TXG_WAIT);
1297 			if (error) {
1298 				dmu_tx_abort(tx);
1299 			} else {
1300 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1301 				zvol_log_write(zv, tx, off, size, sync);
1302 				dmu_tx_commit(tx);
1303 			}
1304 		}
1305 		if (error) {
1306 			/* convert checksum errors into IO errors */
1307 			if (error == ECKSUM)
1308 				error = SET_ERROR(EIO);
1309 			break;
1310 		}
1311 		off += size;
1312 		addr += size;
1313 		resid -= size;
1314 	}
1315 	rangelock_exit(lr);
1316 
1317 	if ((bp->b_resid = resid) == bp->b_bcount)
1318 		bioerror(bp, off > volsize ? EINVAL : error);
1319 
1320 	if (sync)
1321 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1322 	biodone(bp);
1323 
1324 	ht_end_unsafe();
1325 
1326 	return (0);
1327 }
1328 
1329 /*
1330  * Set the buffer count to the zvol maximum transfer.
1331  * Using our own routine instead of the default minphys()
1332  * means that for larger writes we write bigger buffers on X86
1333  * (128K instead of 56K) and flush the disk write cache less often
1334  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1335  * 56K on X86 and 128K on sparc).
1336  */
1337 void
1338 zvol_minphys(struct buf *bp)
1339 {
1340 	if (bp->b_bcount > zvol_maxphys)
1341 		bp->b_bcount = zvol_maxphys;
1342 }
1343 
1344 int
1345 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1346 {
1347 	minor_t minor = getminor(dev);
1348 	zvol_state_t *zv;
1349 	int error = 0;
1350 	uint64_t size;
1351 	uint64_t boff;
1352 	uint64_t resid;
1353 
1354 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1355 	if (zv == NULL)
1356 		return (SET_ERROR(ENXIO));
1357 
1358 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1359 		return (SET_ERROR(EINVAL));
1360 
1361 	boff = ldbtob(blkno);
1362 	resid = ldbtob(nblocks);
1363 
1364 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1365 
1366 	while (resid) {
1367 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1368 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1369 		if (error)
1370 			break;
1371 		boff += size;
1372 		addr += size;
1373 		resid -= size;
1374 	}
1375 
1376 	return (error);
1377 }
1378 
1379 /*ARGSUSED*/
1380 int
1381 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1382 {
1383 	minor_t minor = getminor(dev);
1384 	zvol_state_t *zv;
1385 	uint64_t volsize;
1386 	int error = 0;
1387 
1388 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1389 	if (zv == NULL)
1390 		return (SET_ERROR(ENXIO));
1391 
1392 	volsize = zv->zv_volsize;
1393 	if (uio->uio_resid > 0 &&
1394 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1395 		return (SET_ERROR(EIO));
1396 
1397 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1398 		error = physio(zvol_strategy, NULL, dev, B_READ,
1399 		    zvol_minphys, uio);
1400 		return (error);
1401 	}
1402 
1403 	ht_begin_unsafe();
1404 
1405 	locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1406 	    uio->uio_loffset, uio->uio_resid, RL_READER);
1407 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1408 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1409 
1410 		/* don't read past the end */
1411 		if (bytes > volsize - uio->uio_loffset)
1412 			bytes = volsize - uio->uio_loffset;
1413 
1414 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1415 		if (error) {
1416 			/* convert checksum errors into IO errors */
1417 			if (error == ECKSUM)
1418 				error = SET_ERROR(EIO);
1419 			break;
1420 		}
1421 	}
1422 	rangelock_exit(lr);
1423 
1424 	ht_end_unsafe();
1425 
1426 	return (error);
1427 }
1428 
1429 /*ARGSUSED*/
1430 int
1431 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1432 {
1433 	minor_t minor = getminor(dev);
1434 	zvol_state_t *zv;
1435 	uint64_t volsize;
1436 	int error = 0;
1437 	boolean_t sync;
1438 
1439 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1440 	if (zv == NULL)
1441 		return (SET_ERROR(ENXIO));
1442 
1443 	volsize = zv->zv_volsize;
1444 	if (uio->uio_resid > 0 &&
1445 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1446 		return (SET_ERROR(EIO));
1447 
1448 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1449 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1450 		    zvol_minphys, uio);
1451 		return (error);
1452 	}
1453 
1454 	ht_begin_unsafe();
1455 
1456 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1457 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1458 
1459 	locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1460 	    uio->uio_loffset, uio->uio_resid, RL_WRITER);
1461 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1462 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1463 		uint64_t off = uio->uio_loffset;
1464 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1465 
1466 		if (bytes > volsize - off)	/* don't write past the end */
1467 			bytes = volsize - off;
1468 
1469 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1470 		error = dmu_tx_assign(tx, TXG_WAIT);
1471 		if (error) {
1472 			dmu_tx_abort(tx);
1473 			break;
1474 		}
1475 		error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1476 		if (error == 0)
1477 			zvol_log_write(zv, tx, off, bytes, sync);
1478 		dmu_tx_commit(tx);
1479 
1480 		if (error)
1481 			break;
1482 	}
1483 	rangelock_exit(lr);
1484 
1485 	if (sync)
1486 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1487 
1488 	ht_end_unsafe();
1489 
1490 	return (error);
1491 }
1492 
1493 int
1494 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1495 {
1496 	struct uuid uuid = EFI_RESERVED;
1497 	efi_gpe_t gpe = { 0 };
1498 	uint32_t crc;
1499 	dk_efi_t efi;
1500 	int length;
1501 	char *ptr;
1502 
1503 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1504 		return (SET_ERROR(EFAULT));
1505 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1506 	length = efi.dki_length;
1507 	/*
1508 	 * Some clients may attempt to request a PMBR for the
1509 	 * zvol.  Currently this interface will return EINVAL to
1510 	 * such requests.  These requests could be supported by
1511 	 * adding a check for lba == 0 and consing up an appropriate
1512 	 * PMBR.
1513 	 */
1514 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1515 		return (SET_ERROR(EINVAL));
1516 
1517 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1518 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1519 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1520 
1521 	if (efi.dki_lba == 1) {
1522 		efi_gpt_t gpt = { 0 };
1523 
1524 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1525 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1526 		gpt.efi_gpt_HeaderSize = LE_32(EFI_HEADER_SIZE);
1527 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1528 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1529 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1530 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1531 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1532 		gpt.efi_gpt_SizeOfPartitionEntry =
1533 		    LE_32(sizeof (efi_gpe_t));
1534 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1535 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1536 		CRC32(crc, &gpt, EFI_HEADER_SIZE, -1U, crc32_table);
1537 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1538 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1539 		    flag))
1540 			return (SET_ERROR(EFAULT));
1541 		ptr += sizeof (gpt);
1542 		length -= sizeof (gpt);
1543 	}
1544 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1545 	    length), flag))
1546 		return (SET_ERROR(EFAULT));
1547 	return (0);
1548 }
1549 
1550 /*
1551  * BEGIN entry points to allow external callers access to the volume.
1552  */
1553 /*
1554  * Return the volume parameters needed for access from an external caller.
1555  * These values are invariant as long as the volume is held open.
1556  */
1557 int
1558 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1559     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1560     void **rl_hdl, void **dnode_hdl)
1561 {
1562 	zvol_state_t *zv;
1563 
1564 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1565 	if (zv == NULL)
1566 		return (SET_ERROR(ENXIO));
1567 	if (zv->zv_flags & ZVOL_DUMPIFIED)
1568 		return (SET_ERROR(ENXIO));
1569 
1570 	ASSERT(blksize && max_xfer_len && minor_hdl &&
1571 	    objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
1572 
1573 	*blksize = zv->zv_volblocksize;
1574 	*max_xfer_len = (uint64_t)zvol_maxphys;
1575 	*minor_hdl = zv;
1576 	*objset_hdl = zv->zv_objset;
1577 	*zil_hdl = zv->zv_zilog;
1578 	*rl_hdl = &zv->zv_rangelock;
1579 	*dnode_hdl = zv->zv_dn;
1580 	return (0);
1581 }
1582 
1583 /*
1584  * Return the current volume size to an external caller.
1585  * The size can change while the volume is open.
1586  */
1587 uint64_t
1588 zvol_get_volume_size(void *minor_hdl)
1589 {
1590 	zvol_state_t *zv = minor_hdl;
1591 
1592 	return (zv->zv_volsize);
1593 }
1594 
1595 /*
1596  * Return the current WCE setting to an external caller.
1597  * The WCE setting can change while the volume is open.
1598  */
1599 int
1600 zvol_get_volume_wce(void *minor_hdl)
1601 {
1602 	zvol_state_t *zv = minor_hdl;
1603 
1604 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1605 }
1606 
1607 /*
1608  * Entry point for external callers to zvol_log_write
1609  */
1610 void
1611 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1612     boolean_t sync)
1613 {
1614 	zvol_state_t *zv = minor_hdl;
1615 
1616 	zvol_log_write(zv, tx, off, resid, sync);
1617 }
1618 /*
1619  * END entry points to allow external callers access to the volume.
1620  */
1621 
1622 /*
1623  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1624  */
1625 static void
1626 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1627     boolean_t sync)
1628 {
1629 	itx_t *itx;
1630 	lr_truncate_t *lr;
1631 	zilog_t *zilog = zv->zv_zilog;
1632 
1633 	if (zil_replaying(zilog, tx))
1634 		return;
1635 
1636 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1637 	lr = (lr_truncate_t *)&itx->itx_lr;
1638 	lr->lr_foid = ZVOL_OBJ;
1639 	lr->lr_offset = off;
1640 	lr->lr_length = len;
1641 
1642 	itx->itx_sync = sync;
1643 	zil_itx_assign(zilog, itx, tx);
1644 }
1645 
1646 /*
1647  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1648  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1649  */
1650 /*ARGSUSED*/
1651 int
1652 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1653 {
1654 	zvol_state_t *zv;
1655 	struct dk_callback *dkc;
1656 	int error = 0;
1657 	locked_range_t *lr;
1658 
1659 	mutex_enter(&zfsdev_state_lock);
1660 
1661 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1662 
1663 	if (zv == NULL) {
1664 		mutex_exit(&zfsdev_state_lock);
1665 		return (SET_ERROR(ENXIO));
1666 	}
1667 	ASSERT(zv->zv_total_opens > 0);
1668 
1669 	switch (cmd) {
1670 
1671 	case DKIOCINFO:
1672 	{
1673 		struct dk_cinfo dki;
1674 
1675 		bzero(&dki, sizeof (dki));
1676 		(void) strcpy(dki.dki_cname, "zvol");
1677 		(void) strcpy(dki.dki_dname, "zvol");
1678 		dki.dki_ctype = DKC_UNKNOWN;
1679 		dki.dki_unit = getminor(dev);
1680 		dki.dki_maxtransfer =
1681 		    1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
1682 		mutex_exit(&zfsdev_state_lock);
1683 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1684 			error = SET_ERROR(EFAULT);
1685 		return (error);
1686 	}
1687 
1688 	case DKIOCGMEDIAINFO:
1689 	{
1690 		struct dk_minfo dkm;
1691 
1692 		bzero(&dkm, sizeof (dkm));
1693 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1694 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1695 		dkm.dki_media_type = DK_UNKNOWN;
1696 		mutex_exit(&zfsdev_state_lock);
1697 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1698 			error = SET_ERROR(EFAULT);
1699 		return (error);
1700 	}
1701 
1702 	case DKIOCGMEDIAINFOEXT:
1703 	{
1704 		struct dk_minfo_ext dkmext;
1705 
1706 		bzero(&dkmext, sizeof (dkmext));
1707 		dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1708 		dkmext.dki_pbsize = zv->zv_volblocksize;
1709 		dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1710 		dkmext.dki_media_type = DK_UNKNOWN;
1711 		mutex_exit(&zfsdev_state_lock);
1712 		if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1713 			error = SET_ERROR(EFAULT);
1714 		return (error);
1715 	}
1716 
1717 	case DKIOCGETEFI:
1718 	{
1719 		uint64_t vs = zv->zv_volsize;
1720 		uint8_t bs = zv->zv_min_bs;
1721 
1722 		mutex_exit(&zfsdev_state_lock);
1723 		error = zvol_getefi((void *)arg, flag, vs, bs);
1724 		return (error);
1725 	}
1726 
1727 	case DKIOCFLUSHWRITECACHE:
1728 		dkc = (struct dk_callback *)arg;
1729 		mutex_exit(&zfsdev_state_lock);
1730 
1731 		ht_begin_unsafe();
1732 
1733 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1734 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1735 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1736 			error = 0;
1737 		}
1738 
1739 		ht_end_unsafe();
1740 
1741 		return (error);
1742 
1743 	case DKIOCGETWCE:
1744 	{
1745 		int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1746 		if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1747 		    flag))
1748 			error = SET_ERROR(EFAULT);
1749 		break;
1750 	}
1751 	case DKIOCSETWCE:
1752 	{
1753 		int wce;
1754 		if (ddi_copyin((void *)arg, &wce, sizeof (int),
1755 		    flag)) {
1756 			error = SET_ERROR(EFAULT);
1757 			break;
1758 		}
1759 		if (wce) {
1760 			zv->zv_flags |= ZVOL_WCE;
1761 			mutex_exit(&zfsdev_state_lock);
1762 		} else {
1763 			zv->zv_flags &= ~ZVOL_WCE;
1764 			mutex_exit(&zfsdev_state_lock);
1765 			ht_begin_unsafe();
1766 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1767 			ht_end_unsafe();
1768 		}
1769 		return (0);
1770 	}
1771 
1772 	case DKIOCGGEOM:
1773 	case DKIOCGVTOC:
1774 		/*
1775 		 * commands using these (like prtvtoc) expect ENOTSUP
1776 		 * since we're emulating an EFI label
1777 		 */
1778 		error = SET_ERROR(ENOTSUP);
1779 		break;
1780 
1781 	case DKIOCDUMPINIT:
1782 		lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
1783 		    RL_WRITER);
1784 		error = zvol_dumpify(zv);
1785 		rangelock_exit(lr);
1786 		break;
1787 
1788 	case DKIOCDUMPFINI:
1789 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1790 			break;
1791 		lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
1792 		    RL_WRITER);
1793 		error = zvol_dump_fini(zv);
1794 		rangelock_exit(lr);
1795 		break;
1796 
1797 	case DKIOCFREE:
1798 	{
1799 		dkioc_free_list_t *dfl;
1800 		dmu_tx_t *tx;
1801 
1802 		if (!zvol_unmap_enabled)
1803 			break;
1804 
1805 		if (!(flag & FKIOCTL)) {
1806 			error = dfl_copyin((void *)arg, &dfl, flag, KM_SLEEP);
1807 			if (error != 0)
1808 				break;
1809 		} else {
1810 			dfl = (dkioc_free_list_t *)arg;
1811 			ASSERT3U(dfl->dfl_num_exts, <=, DFL_COPYIN_MAX_EXTS);
1812 			if (dfl->dfl_num_exts > DFL_COPYIN_MAX_EXTS) {
1813 				error = SET_ERROR(EINVAL);
1814 				break;
1815 			}
1816 		}
1817 
1818 		mutex_exit(&zfsdev_state_lock);
1819 
1820 		ht_begin_unsafe();
1821 
1822 		for (int i = 0; i < dfl->dfl_num_exts; i++) {
1823 			uint64_t start = dfl->dfl_exts[i].dfle_start,
1824 			    length = dfl->dfl_exts[i].dfle_length,
1825 			    end = start + length;
1826 
1827 			/*
1828 			 * Apply Postel's Law to length-checking.  If they
1829 			 * overshoot, just blank out until the end, if there's
1830 			 * a need to blank out anything.
1831 			 */
1832 			if (start >= zv->zv_volsize)
1833 				continue;	/* No need to do anything... */
1834 			if (end > zv->zv_volsize) {
1835 				end = DMU_OBJECT_END;
1836 				length = end - start;
1837 			}
1838 
1839 			lr = rangelock_enter(&zv->zv_rangelock, start, length,
1840 			    RL_WRITER);
1841 			tx = dmu_tx_create(zv->zv_objset);
1842 			error = dmu_tx_assign(tx, TXG_WAIT);
1843 			if (error != 0) {
1844 				dmu_tx_abort(tx);
1845 			} else {
1846 				zvol_log_truncate(zv, tx, start, length,
1847 				    B_TRUE);
1848 				dmu_tx_commit(tx);
1849 				error = dmu_free_long_range(zv->zv_objset,
1850 				    ZVOL_OBJ, start, length);
1851 			}
1852 
1853 			rangelock_exit(lr);
1854 
1855 			if (error != 0)
1856 				break;
1857 		}
1858 
1859 		/*
1860 		 * If the write-cache is disabled, 'sync' property
1861 		 * is set to 'always', or if the caller is asking for
1862 		 * a synchronous free, commit this operation to the zil.
1863 		 * This will sync any previous uncommitted writes to the
1864 		 * zvol object.
1865 		 * Can be overridden by the zvol_unmap_sync_enabled tunable.
1866 		 */
1867 		if ((error == 0) && zvol_unmap_sync_enabled &&
1868 		    (!(zv->zv_flags & ZVOL_WCE) ||
1869 		    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
1870 		    (dfl->dfl_flags & DF_WAIT_SYNC))) {
1871 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1872 		}
1873 
1874 		if (!(flag & FKIOCTL))
1875 			dfl_free(dfl);
1876 
1877 		ht_end_unsafe();
1878 
1879 		return (error);
1880 	}
1881 
1882 	default:
1883 		error = SET_ERROR(ENOTTY);
1884 		break;
1885 
1886 	}
1887 	mutex_exit(&zfsdev_state_lock);
1888 	return (error);
1889 }
1890 
1891 int
1892 zvol_busy(void)
1893 {
1894 	return (zvol_minors != 0);
1895 }
1896 
1897 void
1898 zvol_init(void)
1899 {
1900 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1901 	    1) == 0);
1902 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1903 }
1904 
1905 void
1906 zvol_fini(void)
1907 {
1908 	mutex_destroy(&zfsdev_state_lock);
1909 	ddi_soft_state_fini(&zfsdev_state);
1910 }
1911 
1912 /*ARGSUSED*/
1913 static int
1914 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1915 {
1916 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1917 
1918 	if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1919 		return (1);
1920 	return (0);
1921 }
1922 
1923 /*ARGSUSED*/
1924 static void
1925 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1926 {
1927 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1928 
1929 	spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1930 }
1931 
1932 static int
1933 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1934 {
1935 	dmu_tx_t *tx;
1936 	int error;
1937 	objset_t *os = zv->zv_objset;
1938 	spa_t *spa = dmu_objset_spa(os);
1939 	vdev_t *vd = spa->spa_root_vdev;
1940 	nvlist_t *nv = NULL;
1941 	uint64_t version = spa_version(spa);
1942 	uint64_t checksum, compress, refresrv, vbs, dedup;
1943 
1944 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1945 	ASSERT(vd->vdev_ops == &vdev_root_ops);
1946 
1947 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1948 	    DMU_OBJECT_END);
1949 	if (error != 0)
1950 		return (error);
1951 	/* wait for dmu_free_long_range to actually free the blocks */
1952 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1953 
1954 	/*
1955 	 * If the pool on which the dump device is being initialized has more
1956 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1957 	 * enabled.  If so, bump that feature's counter to indicate that the
1958 	 * feature is active. We also check the vdev type to handle the
1959 	 * following case:
1960 	 *   # zpool create test raidz disk1 disk2 disk3
1961 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1962 	 *   the raidz vdev itself has 3 children.
1963 	 */
1964 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1965 		if (!spa_feature_is_enabled(spa,
1966 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1967 			return (SET_ERROR(ENOTSUP));
1968 		(void) dsl_sync_task(spa_name(spa),
1969 		    zfs_mvdev_dump_feature_check,
1970 		    zfs_mvdev_dump_activate_feature_sync, NULL,
1971 		    2, ZFS_SPACE_CHECK_RESERVED);
1972 	}
1973 
1974 	if (!resize) {
1975 		error = dsl_prop_get_integer(zv->zv_name,
1976 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1977 		if (error == 0) {
1978 			error = dsl_prop_get_integer(zv->zv_name,
1979 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
1980 			    NULL);
1981 		}
1982 		if (error == 0) {
1983 			error = dsl_prop_get_integer(zv->zv_name,
1984 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
1985 			    &refresrv, NULL);
1986 		}
1987 		if (error == 0) {
1988 			error = dsl_prop_get_integer(zv->zv_name,
1989 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
1990 			    NULL);
1991 		}
1992 		if (version >= SPA_VERSION_DEDUP && error == 0) {
1993 			error = dsl_prop_get_integer(zv->zv_name,
1994 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1995 		}
1996 	}
1997 	if (error != 0)
1998 		return (error);
1999 
2000 	tx = dmu_tx_create(os);
2001 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2002 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2003 	error = dmu_tx_assign(tx, TXG_WAIT);
2004 	if (error != 0) {
2005 		dmu_tx_abort(tx);
2006 		return (error);
2007 	}
2008 
2009 	/*
2010 	 * If we are resizing the dump device then we only need to
2011 	 * update the refreservation to match the newly updated
2012 	 * zvolsize. Otherwise, we save off the original state of the
2013 	 * zvol so that we can restore them if the zvol is ever undumpified.
2014 	 */
2015 	if (resize) {
2016 		error = zap_update(os, ZVOL_ZAP_OBJ,
2017 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2018 		    &zv->zv_volsize, tx);
2019 	} else {
2020 		error = zap_update(os, ZVOL_ZAP_OBJ,
2021 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2022 		    &compress, tx);
2023 		if (error == 0) {
2024 			error = zap_update(os, ZVOL_ZAP_OBJ,
2025 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2026 			    &checksum, tx);
2027 		}
2028 		if (error == 0) {
2029 			error = zap_update(os, ZVOL_ZAP_OBJ,
2030 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2031 			    &refresrv, tx);
2032 		}
2033 		if (error == 0) {
2034 			error = zap_update(os, ZVOL_ZAP_OBJ,
2035 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2036 			    &vbs, tx);
2037 		}
2038 		if (error == 0) {
2039 			error = dmu_object_set_blocksize(
2040 			    os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2041 		}
2042 		if (version >= SPA_VERSION_DEDUP && error == 0) {
2043 			error = zap_update(os, ZVOL_ZAP_OBJ,
2044 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2045 			    &dedup, tx);
2046 		}
2047 		if (error == 0)
2048 			zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2049 	}
2050 	dmu_tx_commit(tx);
2051 
2052 	/*
2053 	 * We only need update the zvol's property if we are initializing
2054 	 * the dump area for the first time.
2055 	 */
2056 	if (error == 0 && !resize) {
2057 		/*
2058 		 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2059 		 * function.  Otherwise, use the old default -- OFF.
2060 		 */
2061 		checksum = spa_feature_is_active(spa,
2062 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2063 		    ZIO_CHECKSUM_OFF;
2064 
2065 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2066 		VERIFY(nvlist_add_uint64(nv,
2067 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2068 		VERIFY(nvlist_add_uint64(nv,
2069 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2070 		    ZIO_COMPRESS_OFF) == 0);
2071 		VERIFY(nvlist_add_uint64(nv,
2072 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2073 		    checksum) == 0);
2074 		if (version >= SPA_VERSION_DEDUP) {
2075 			VERIFY(nvlist_add_uint64(nv,
2076 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
2077 			    ZIO_CHECKSUM_OFF) == 0);
2078 		}
2079 
2080 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2081 		    nv, NULL);
2082 		nvlist_free(nv);
2083 	}
2084 
2085 	/* Allocate the space for the dump */
2086 	if (error == 0)
2087 		error = zvol_prealloc(zv);
2088 	return (error);
2089 }
2090 
2091 static int
2092 zvol_dumpify(zvol_state_t *zv)
2093 {
2094 	int error = 0;
2095 	uint64_t dumpsize = 0;
2096 	dmu_tx_t *tx;
2097 	objset_t *os = zv->zv_objset;
2098 
2099 	if (zv->zv_flags & ZVOL_RDONLY)
2100 		return (SET_ERROR(EROFS));
2101 
2102 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2103 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2104 		boolean_t resize = (dumpsize > 0);
2105 
2106 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2107 			(void) zvol_dump_fini(zv);
2108 			return (error);
2109 		}
2110 	}
2111 
2112 	/*
2113 	 * Build up our lba mapping.
2114 	 */
2115 	error = zvol_get_lbas(zv);
2116 	if (error) {
2117 		(void) zvol_dump_fini(zv);
2118 		return (error);
2119 	}
2120 
2121 	tx = dmu_tx_create(os);
2122 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2123 	error = dmu_tx_assign(tx, TXG_WAIT);
2124 	if (error) {
2125 		dmu_tx_abort(tx);
2126 		(void) zvol_dump_fini(zv);
2127 		return (error);
2128 	}
2129 
2130 	zv->zv_flags |= ZVOL_DUMPIFIED;
2131 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2132 	    &zv->zv_volsize, tx);
2133 	dmu_tx_commit(tx);
2134 
2135 	if (error) {
2136 		(void) zvol_dump_fini(zv);
2137 		return (error);
2138 	}
2139 
2140 	txg_wait_synced(dmu_objset_pool(os), 0);
2141 	return (0);
2142 }
2143 
2144 static int
2145 zvol_dump_fini(zvol_state_t *zv)
2146 {
2147 	dmu_tx_t *tx;
2148 	objset_t *os = zv->zv_objset;
2149 	nvlist_t *nv;
2150 	int error = 0;
2151 	uint64_t checksum, compress, refresrv, vbs, dedup;
2152 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2153 
2154 	/*
2155 	 * Attempt to restore the zvol back to its pre-dumpified state.
2156 	 * This is a best-effort attempt as it's possible that not all
2157 	 * of these properties were initialized during the dumpify process
2158 	 * (i.e. error during zvol_dump_init).
2159 	 */
2160 
2161 	tx = dmu_tx_create(os);
2162 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2163 	error = dmu_tx_assign(tx, TXG_WAIT);
2164 	if (error) {
2165 		dmu_tx_abort(tx);
2166 		return (error);
2167 	}
2168 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2169 	dmu_tx_commit(tx);
2170 
2171 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2172 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2173 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2174 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2175 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2176 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2177 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2178 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2179 
2180 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2181 	(void) nvlist_add_uint64(nv,
2182 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2183 	(void) nvlist_add_uint64(nv,
2184 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2185 	(void) nvlist_add_uint64(nv,
2186 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2187 	if (version >= SPA_VERSION_DEDUP &&
2188 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2189 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2190 		(void) nvlist_add_uint64(nv,
2191 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2192 	}
2193 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2194 	    nv, NULL);
2195 	nvlist_free(nv);
2196 
2197 	zvol_free_extents(zv);
2198 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2199 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2200 	/* wait for dmu_free_long_range to actually free the blocks */
2201 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2202 	tx = dmu_tx_create(os);
2203 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2204 	error = dmu_tx_assign(tx, TXG_WAIT);
2205 	if (error) {
2206 		dmu_tx_abort(tx);
2207 		return (error);
2208 	}
2209 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2210 		zv->zv_volblocksize = vbs;
2211 	dmu_tx_commit(tx);
2212 
2213 	return (0);
2214 }
2215