xref: /titanic_52/usr/src/uts/common/fs/zfs/zvol.c (revision b49b27dcb66b2c7f4a23f7bc158e2dde5cd79030)
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 2011 Nexenta Systems, Inc.  All rights reserved.
27  * Copyright (c) 2013 by Delphix. All rights reserved.
28  */
29 
30 /*
31  * ZFS volume emulation driver.
32  *
33  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
34  * Volumes are accessed through the symbolic links named:
35  *
36  * /dev/zvol/dsk/<pool_name>/<dataset_name>
37  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
38  *
39  * These links are created by the /dev filesystem (sdev_zvolops.c).
40  * Volumes are persistent through reboot.  No user command needs to be
41  * run before opening and using a device.
42  */
43 
44 #include <sys/types.h>
45 #include <sys/param.h>
46 #include <sys/errno.h>
47 #include <sys/uio.h>
48 #include <sys/buf.h>
49 #include <sys/modctl.h>
50 #include <sys/open.h>
51 #include <sys/kmem.h>
52 #include <sys/conf.h>
53 #include <sys/cmn_err.h>
54 #include <sys/stat.h>
55 #include <sys/zap.h>
56 #include <sys/spa.h>
57 #include <sys/zio.h>
58 #include <sys/dmu_traverse.h>
59 #include <sys/dnode.h>
60 #include <sys/dsl_dataset.h>
61 #include <sys/dsl_prop.h>
62 #include <sys/dkio.h>
63 #include <sys/efi_partition.h>
64 #include <sys/byteorder.h>
65 #include <sys/pathname.h>
66 #include <sys/ddi.h>
67 #include <sys/sunddi.h>
68 #include <sys/crc32.h>
69 #include <sys/dirent.h>
70 #include <sys/policy.h>
71 #include <sys/fs/zfs.h>
72 #include <sys/zfs_ioctl.h>
73 #include <sys/mkdev.h>
74 #include <sys/zil.h>
75 #include <sys/refcount.h>
76 #include <sys/zfs_znode.h>
77 #include <sys/zfs_rlock.h>
78 #include <sys/vdev_disk.h>
79 #include <sys/vdev_impl.h>
80 #include <sys/zvol.h>
81 #include <sys/dumphdr.h>
82 #include <sys/zil_impl.h>
83 #include <sys/dbuf.h>
84 
85 #include "zfs_namecheck.h"
86 
87 void *zfsdev_state;
88 static char *zvol_tag = "zvol_tag";
89 
90 #define	ZVOL_DUMPSIZE		"dumpsize"
91 
92 /*
93  * This lock protects the zfsdev_state structure from being modified
94  * while it's being used, e.g. an open that comes in before a create
95  * finishes.  It also protects temporary opens of the dataset so that,
96  * e.g., an open doesn't get a spurious EBUSY.
97  */
98 kmutex_t zfsdev_state_lock;
99 static uint32_t zvol_minors;
100 
101 typedef struct zvol_extent {
102 	list_node_t	ze_node;
103 	dva_t		ze_dva;		/* dva associated with this extent */
104 	uint64_t	ze_nblks;	/* number of blocks in extent */
105 } zvol_extent_t;
106 
107 /*
108  * The in-core state of each volume.
109  */
110 typedef struct zvol_state {
111 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
112 	uint64_t	zv_volsize;	/* amount of space we advertise */
113 	uint64_t	zv_volblocksize; /* volume block size */
114 	minor_t		zv_minor;	/* minor number */
115 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
116 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
117 	objset_t	*zv_objset;	/* objset handle */
118 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
119 	uint32_t	zv_total_opens;	/* total open count */
120 	zilog_t		*zv_zilog;	/* ZIL handle */
121 	list_t		zv_extents;	/* List of extents for dump */
122 	znode_t		zv_znode;	/* for range locking */
123 	dmu_buf_t	*zv_dbuf;	/* bonus handle */
124 } zvol_state_t;
125 
126 /*
127  * zvol specific flags
128  */
129 #define	ZVOL_RDONLY	0x1
130 #define	ZVOL_DUMPIFIED	0x2
131 #define	ZVOL_EXCL	0x4
132 #define	ZVOL_WCE	0x8
133 
134 /*
135  * zvol maximum transfer in one DMU tx.
136  */
137 int zvol_maxphys = DMU_MAX_ACCESS/2;
138 
139 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
140     nvlist_t *, nvlist_t *);
141 static int zvol_remove_zv(zvol_state_t *);
142 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
143 static int zvol_dumpify(zvol_state_t *zv);
144 static int zvol_dump_fini(zvol_state_t *zv);
145 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
146 
147 static void
148 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
149 {
150 	dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
151 
152 	zv->zv_volsize = volsize;
153 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
154 	    "Size", volsize) == DDI_SUCCESS);
155 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
156 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
157 
158 	/* Notify specfs to invalidate the cached size */
159 	spec_size_invalidate(dev, VBLK);
160 	spec_size_invalidate(dev, VCHR);
161 }
162 
163 int
164 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
165 {
166 	if (volsize == 0)
167 		return (SET_ERROR(EINVAL));
168 
169 	if (volsize % blocksize != 0)
170 		return (SET_ERROR(EINVAL));
171 
172 #ifdef _ILP32
173 	if (volsize - 1 > SPEC_MAXOFFSET_T)
174 		return (SET_ERROR(EOVERFLOW));
175 #endif
176 	return (0);
177 }
178 
179 int
180 zvol_check_volblocksize(uint64_t volblocksize)
181 {
182 	if (volblocksize < SPA_MINBLOCKSIZE ||
183 	    volblocksize > SPA_MAXBLOCKSIZE ||
184 	    !ISP2(volblocksize))
185 		return (SET_ERROR(EDOM));
186 
187 	return (0);
188 }
189 
190 int
191 zvol_get_stats(objset_t *os, nvlist_t *nv)
192 {
193 	int error;
194 	dmu_object_info_t doi;
195 	uint64_t val;
196 
197 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
198 	if (error)
199 		return (error);
200 
201 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
202 
203 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
204 
205 	if (error == 0) {
206 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
207 		    doi.doi_data_block_size);
208 	}
209 
210 	return (error);
211 }
212 
213 static zvol_state_t *
214 zvol_minor_lookup(const char *name)
215 {
216 	minor_t minor;
217 	zvol_state_t *zv;
218 
219 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
220 
221 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
222 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
223 		if (zv == NULL)
224 			continue;
225 		if (strcmp(zv->zv_name, name) == 0)
226 			return (zv);
227 	}
228 
229 	return (NULL);
230 }
231 
232 /* extent mapping arg */
233 struct maparg {
234 	zvol_state_t	*ma_zv;
235 	uint64_t	ma_blks;
236 };
237 
238 /*ARGSUSED*/
239 static int
240 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
241     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
242 {
243 	struct maparg *ma = arg;
244 	zvol_extent_t *ze;
245 	int bs = ma->ma_zv->zv_volblocksize;
246 
247 	if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
248 		return (0);
249 
250 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
251 	ma->ma_blks++;
252 
253 	/* Abort immediately if we have encountered gang blocks */
254 	if (BP_IS_GANG(bp))
255 		return (SET_ERROR(EFRAGS));
256 
257 	/*
258 	 * See if the block is at the end of the previous extent.
259 	 */
260 	ze = list_tail(&ma->ma_zv->zv_extents);
261 	if (ze &&
262 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
263 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
264 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
265 		ze->ze_nblks++;
266 		return (0);
267 	}
268 
269 	dprintf_bp(bp, "%s", "next blkptr:");
270 
271 	/* start a new extent */
272 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
273 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
274 	ze->ze_nblks = 1;
275 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
276 	return (0);
277 }
278 
279 static void
280 zvol_free_extents(zvol_state_t *zv)
281 {
282 	zvol_extent_t *ze;
283 
284 	while (ze = list_head(&zv->zv_extents)) {
285 		list_remove(&zv->zv_extents, ze);
286 		kmem_free(ze, sizeof (zvol_extent_t));
287 	}
288 }
289 
290 static int
291 zvol_get_lbas(zvol_state_t *zv)
292 {
293 	objset_t *os = zv->zv_objset;
294 	struct maparg	ma;
295 	int		err;
296 
297 	ma.ma_zv = zv;
298 	ma.ma_blks = 0;
299 	zvol_free_extents(zv);
300 
301 	/* commit any in-flight changes before traversing the dataset */
302 	txg_wait_synced(dmu_objset_pool(os), 0);
303 	err = traverse_dataset(dmu_objset_ds(os), 0,
304 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
305 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
306 		zvol_free_extents(zv);
307 		return (err ? err : EIO);
308 	}
309 
310 	return (0);
311 }
312 
313 /* ARGSUSED */
314 void
315 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
316 {
317 	zfs_creat_t *zct = arg;
318 	nvlist_t *nvprops = zct->zct_props;
319 	int error;
320 	uint64_t volblocksize, volsize;
321 
322 	VERIFY(nvlist_lookup_uint64(nvprops,
323 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
324 	if (nvlist_lookup_uint64(nvprops,
325 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
326 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
327 
328 	/*
329 	 * These properties must be removed from the list so the generic
330 	 * property setting step won't apply to them.
331 	 */
332 	VERIFY(nvlist_remove_all(nvprops,
333 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
334 	(void) nvlist_remove_all(nvprops,
335 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
336 
337 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
338 	    DMU_OT_NONE, 0, tx);
339 	ASSERT(error == 0);
340 
341 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
342 	    DMU_OT_NONE, 0, tx);
343 	ASSERT(error == 0);
344 
345 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
346 	ASSERT(error == 0);
347 }
348 
349 /*
350  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
351  * implement DKIOCFREE/free-long-range.
352  */
353 static int
354 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
355 {
356 	uint64_t offset, length;
357 
358 	if (byteswap)
359 		byteswap_uint64_array(lr, sizeof (*lr));
360 
361 	offset = lr->lr_offset;
362 	length = lr->lr_length;
363 
364 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
365 }
366 
367 /*
368  * Replay a TX_WRITE ZIL transaction that didn't get committed
369  * after a system failure
370  */
371 static int
372 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
373 {
374 	objset_t *os = zv->zv_objset;
375 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
376 	uint64_t offset, length;
377 	dmu_tx_t *tx;
378 	int error;
379 
380 	if (byteswap)
381 		byteswap_uint64_array(lr, sizeof (*lr));
382 
383 	offset = lr->lr_offset;
384 	length = lr->lr_length;
385 
386 	/* If it's a dmu_sync() block, write the whole block */
387 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
388 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
389 		if (length < blocksize) {
390 			offset -= offset % blocksize;
391 			length = blocksize;
392 		}
393 	}
394 
395 	tx = dmu_tx_create(os);
396 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
397 	error = dmu_tx_assign(tx, TXG_WAIT);
398 	if (error) {
399 		dmu_tx_abort(tx);
400 	} else {
401 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
402 		dmu_tx_commit(tx);
403 	}
404 
405 	return (error);
406 }
407 
408 /* ARGSUSED */
409 static int
410 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
411 {
412 	return (SET_ERROR(ENOTSUP));
413 }
414 
415 /*
416  * Callback vectors for replaying records.
417  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
418  */
419 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
420 	zvol_replay_err,	/* 0 no such transaction type */
421 	zvol_replay_err,	/* TX_CREATE */
422 	zvol_replay_err,	/* TX_MKDIR */
423 	zvol_replay_err,	/* TX_MKXATTR */
424 	zvol_replay_err,	/* TX_SYMLINK */
425 	zvol_replay_err,	/* TX_REMOVE */
426 	zvol_replay_err,	/* TX_RMDIR */
427 	zvol_replay_err,	/* TX_LINK */
428 	zvol_replay_err,	/* TX_RENAME */
429 	zvol_replay_write,	/* TX_WRITE */
430 	zvol_replay_truncate,	/* TX_TRUNCATE */
431 	zvol_replay_err,	/* TX_SETATTR */
432 	zvol_replay_err,	/* TX_ACL */
433 	zvol_replay_err,	/* TX_CREATE_ACL */
434 	zvol_replay_err,	/* TX_CREATE_ATTR */
435 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
436 	zvol_replay_err,	/* TX_MKDIR_ACL */
437 	zvol_replay_err,	/* TX_MKDIR_ATTR */
438 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
439 	zvol_replay_err,	/* TX_WRITE2 */
440 };
441 
442 int
443 zvol_name2minor(const char *name, minor_t *minor)
444 {
445 	zvol_state_t *zv;
446 
447 	mutex_enter(&zfsdev_state_lock);
448 	zv = zvol_minor_lookup(name);
449 	if (minor && zv)
450 		*minor = zv->zv_minor;
451 	mutex_exit(&zfsdev_state_lock);
452 	return (zv ? 0 : -1);
453 }
454 
455 /*
456  * Create a minor node (plus a whole lot more) for the specified volume.
457  */
458 int
459 zvol_create_minor(const char *name)
460 {
461 	zfs_soft_state_t *zs;
462 	zvol_state_t *zv;
463 	objset_t *os;
464 	dmu_object_info_t doi;
465 	minor_t minor = 0;
466 	char chrbuf[30], blkbuf[30];
467 	int error;
468 
469 	mutex_enter(&zfsdev_state_lock);
470 
471 	if (zvol_minor_lookup(name) != NULL) {
472 		mutex_exit(&zfsdev_state_lock);
473 		return (SET_ERROR(EEXIST));
474 	}
475 
476 	/* lie and say we're read-only */
477 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
478 
479 	if (error) {
480 		mutex_exit(&zfsdev_state_lock);
481 		return (error);
482 	}
483 
484 	if ((minor = zfsdev_minor_alloc()) == 0) {
485 		dmu_objset_disown(os, FTAG);
486 		mutex_exit(&zfsdev_state_lock);
487 		return (SET_ERROR(ENXIO));
488 	}
489 
490 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
491 		dmu_objset_disown(os, FTAG);
492 		mutex_exit(&zfsdev_state_lock);
493 		return (SET_ERROR(EAGAIN));
494 	}
495 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
496 	    (char *)name);
497 
498 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
499 
500 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
501 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
502 		ddi_soft_state_free(zfsdev_state, minor);
503 		dmu_objset_disown(os, FTAG);
504 		mutex_exit(&zfsdev_state_lock);
505 		return (SET_ERROR(EAGAIN));
506 	}
507 
508 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
509 
510 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
511 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
512 		ddi_remove_minor_node(zfs_dip, chrbuf);
513 		ddi_soft_state_free(zfsdev_state, minor);
514 		dmu_objset_disown(os, FTAG);
515 		mutex_exit(&zfsdev_state_lock);
516 		return (SET_ERROR(EAGAIN));
517 	}
518 
519 	zs = ddi_get_soft_state(zfsdev_state, minor);
520 	zs->zss_type = ZSST_ZVOL;
521 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
522 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
523 	zv->zv_min_bs = DEV_BSHIFT;
524 	zv->zv_minor = minor;
525 	zv->zv_objset = os;
526 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
527 		zv->zv_flags |= ZVOL_RDONLY;
528 	mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
529 	avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
530 	    sizeof (rl_t), offsetof(rl_t, r_node));
531 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
532 	    offsetof(zvol_extent_t, ze_node));
533 	/* get and cache the blocksize */
534 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
535 	ASSERT(error == 0);
536 	zv->zv_volblocksize = doi.doi_data_block_size;
537 
538 	if (spa_writeable(dmu_objset_spa(os))) {
539 		if (zil_replay_disable)
540 			zil_destroy(dmu_objset_zil(os), B_FALSE);
541 		else
542 			zil_replay(os, zv, zvol_replay_vector);
543 	}
544 	dmu_objset_disown(os, FTAG);
545 	zv->zv_objset = NULL;
546 
547 	zvol_minors++;
548 
549 	mutex_exit(&zfsdev_state_lock);
550 
551 	return (0);
552 }
553 
554 /*
555  * Remove minor node for the specified volume.
556  */
557 static int
558 zvol_remove_zv(zvol_state_t *zv)
559 {
560 	char nmbuf[20];
561 	minor_t minor = zv->zv_minor;
562 
563 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
564 	if (zv->zv_total_opens != 0)
565 		return (SET_ERROR(EBUSY));
566 
567 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
568 	ddi_remove_minor_node(zfs_dip, nmbuf);
569 
570 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
571 	ddi_remove_minor_node(zfs_dip, nmbuf);
572 
573 	avl_destroy(&zv->zv_znode.z_range_avl);
574 	mutex_destroy(&zv->zv_znode.z_range_lock);
575 
576 	kmem_free(zv, sizeof (zvol_state_t));
577 
578 	ddi_soft_state_free(zfsdev_state, minor);
579 
580 	zvol_minors--;
581 	return (0);
582 }
583 
584 int
585 zvol_remove_minor(const char *name)
586 {
587 	zvol_state_t *zv;
588 	int rc;
589 
590 	mutex_enter(&zfsdev_state_lock);
591 	if ((zv = zvol_minor_lookup(name)) == NULL) {
592 		mutex_exit(&zfsdev_state_lock);
593 		return (SET_ERROR(ENXIO));
594 	}
595 	rc = zvol_remove_zv(zv);
596 	mutex_exit(&zfsdev_state_lock);
597 	return (rc);
598 }
599 
600 int
601 zvol_first_open(zvol_state_t *zv)
602 {
603 	objset_t *os;
604 	uint64_t volsize;
605 	int error;
606 	uint64_t readonly;
607 
608 	/* lie and say we're read-only */
609 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
610 	    zvol_tag, &os);
611 	if (error)
612 		return (error);
613 
614 	zv->zv_objset = os;
615 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
616 	if (error) {
617 		ASSERT(error == 0);
618 		dmu_objset_disown(os, zvol_tag);
619 		return (error);
620 	}
621 
622 	error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
623 	if (error) {
624 		dmu_objset_disown(os, zvol_tag);
625 		return (error);
626 	}
627 
628 	zvol_size_changed(zv, volsize);
629 	zv->zv_zilog = zil_open(os, zvol_get_data);
630 
631 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
632 	    NULL) == 0);
633 	if (readonly || dmu_objset_is_snapshot(os) ||
634 	    !spa_writeable(dmu_objset_spa(os)))
635 		zv->zv_flags |= ZVOL_RDONLY;
636 	else
637 		zv->zv_flags &= ~ZVOL_RDONLY;
638 	return (error);
639 }
640 
641 void
642 zvol_last_close(zvol_state_t *zv)
643 {
644 	zil_close(zv->zv_zilog);
645 	zv->zv_zilog = NULL;
646 
647 	dmu_buf_rele(zv->zv_dbuf, zvol_tag);
648 	zv->zv_dbuf = NULL;
649 
650 	/*
651 	 * Evict cached data
652 	 */
653 	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
654 	    !(zv->zv_flags & ZVOL_RDONLY))
655 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
656 	dmu_objset_evict_dbufs(zv->zv_objset);
657 
658 	dmu_objset_disown(zv->zv_objset, zvol_tag);
659 	zv->zv_objset = NULL;
660 }
661 
662 int
663 zvol_prealloc(zvol_state_t *zv)
664 {
665 	objset_t *os = zv->zv_objset;
666 	dmu_tx_t *tx;
667 	uint64_t refd, avail, usedobjs, availobjs;
668 	uint64_t resid = zv->zv_volsize;
669 	uint64_t off = 0;
670 
671 	/* Check the space usage before attempting to allocate the space */
672 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
673 	if (avail < zv->zv_volsize)
674 		return (SET_ERROR(ENOSPC));
675 
676 	/* Free old extents if they exist */
677 	zvol_free_extents(zv);
678 
679 	while (resid != 0) {
680 		int error;
681 		uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
682 
683 		tx = dmu_tx_create(os);
684 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
685 		error = dmu_tx_assign(tx, TXG_WAIT);
686 		if (error) {
687 			dmu_tx_abort(tx);
688 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
689 			return (error);
690 		}
691 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
692 		dmu_tx_commit(tx);
693 		off += bytes;
694 		resid -= bytes;
695 	}
696 	txg_wait_synced(dmu_objset_pool(os), 0);
697 
698 	return (0);
699 }
700 
701 static int
702 zvol_update_volsize(objset_t *os, uint64_t volsize)
703 {
704 	dmu_tx_t *tx;
705 	int error;
706 
707 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
708 
709 	tx = dmu_tx_create(os);
710 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
711 	error = dmu_tx_assign(tx, TXG_WAIT);
712 	if (error) {
713 		dmu_tx_abort(tx);
714 		return (error);
715 	}
716 
717 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
718 	    &volsize, tx);
719 	dmu_tx_commit(tx);
720 
721 	if (error == 0)
722 		error = dmu_free_long_range(os,
723 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
724 	return (error);
725 }
726 
727 void
728 zvol_remove_minors(const char *name)
729 {
730 	zvol_state_t *zv;
731 	char *namebuf;
732 	minor_t minor;
733 
734 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
735 	(void) strncpy(namebuf, name, strlen(name));
736 	(void) strcat(namebuf, "/");
737 	mutex_enter(&zfsdev_state_lock);
738 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
739 
740 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
741 		if (zv == NULL)
742 			continue;
743 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
744 			(void) zvol_remove_zv(zv);
745 	}
746 	kmem_free(namebuf, strlen(name) + 2);
747 
748 	mutex_exit(&zfsdev_state_lock);
749 }
750 
751 static int
752 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
753 {
754 	uint64_t old_volsize = 0ULL;
755 	int error = 0;
756 
757 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
758 
759 	/*
760 	 * Reinitialize the dump area to the new size. If we
761 	 * failed to resize the dump area then restore it back to
762 	 * its original size.  We must set the new volsize prior
763 	 * to calling dumpvp_resize() to ensure that the devices'
764 	 * size(9P) is not visible by the dump subsystem.
765 	 */
766 	old_volsize = zv->zv_volsize;
767 	zvol_size_changed(zv, volsize);
768 
769 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
770 		if ((error = zvol_dumpify(zv)) != 0 ||
771 		    (error = dumpvp_resize()) != 0) {
772 			int dumpify_error;
773 
774 			(void) zvol_update_volsize(zv->zv_objset, old_volsize);
775 			zvol_size_changed(zv, old_volsize);
776 			dumpify_error = zvol_dumpify(zv);
777 			error = dumpify_error ? dumpify_error : error;
778 		}
779 	}
780 
781 	/*
782 	 * Generate a LUN expansion event.
783 	 */
784 	if (error == 0) {
785 		sysevent_id_t eid;
786 		nvlist_t *attr;
787 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
788 
789 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
790 		    zv->zv_minor);
791 
792 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
793 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
794 
795 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
796 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
797 
798 		nvlist_free(attr);
799 		kmem_free(physpath, MAXPATHLEN);
800 	}
801 	return (error);
802 }
803 
804 int
805 zvol_set_volsize(const char *name, uint64_t volsize)
806 {
807 	zvol_state_t *zv = NULL;
808 	objset_t *os;
809 	int error;
810 	dmu_object_info_t doi;
811 	uint64_t readonly;
812 	boolean_t owned = B_FALSE;
813 
814 	error = dsl_prop_get_integer(name,
815 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
816 	if (error != 0)
817 		return (error);
818 	if (readonly)
819 		return (SET_ERROR(EROFS));
820 
821 	mutex_enter(&zfsdev_state_lock);
822 	zv = zvol_minor_lookup(name);
823 
824 	if (zv == NULL || zv->zv_objset == NULL) {
825 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
826 		    FTAG, &os)) != 0) {
827 			mutex_exit(&zfsdev_state_lock);
828 			return (error);
829 		}
830 		owned = B_TRUE;
831 		if (zv != NULL)
832 			zv->zv_objset = os;
833 	} else {
834 		os = zv->zv_objset;
835 	}
836 
837 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
838 	    (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
839 		goto out;
840 
841 	error = zvol_update_volsize(os, volsize);
842 
843 	if (error == 0 && zv != NULL)
844 		error = zvol_update_live_volsize(zv, volsize);
845 out:
846 	if (owned) {
847 		dmu_objset_disown(os, FTAG);
848 		if (zv != NULL)
849 			zv->zv_objset = NULL;
850 	}
851 	mutex_exit(&zfsdev_state_lock);
852 	return (error);
853 }
854 
855 /*ARGSUSED*/
856 int
857 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
858 {
859 	zvol_state_t *zv;
860 	int err = 0;
861 
862 	mutex_enter(&zfsdev_state_lock);
863 
864 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
865 	if (zv == NULL) {
866 		mutex_exit(&zfsdev_state_lock);
867 		return (SET_ERROR(ENXIO));
868 	}
869 
870 	if (zv->zv_total_opens == 0)
871 		err = zvol_first_open(zv);
872 	if (err) {
873 		mutex_exit(&zfsdev_state_lock);
874 		return (err);
875 	}
876 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
877 		err = SET_ERROR(EROFS);
878 		goto out;
879 	}
880 	if (zv->zv_flags & ZVOL_EXCL) {
881 		err = SET_ERROR(EBUSY);
882 		goto out;
883 	}
884 	if (flag & FEXCL) {
885 		if (zv->zv_total_opens != 0) {
886 			err = SET_ERROR(EBUSY);
887 			goto out;
888 		}
889 		zv->zv_flags |= ZVOL_EXCL;
890 	}
891 
892 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
893 		zv->zv_open_count[otyp]++;
894 		zv->zv_total_opens++;
895 	}
896 	mutex_exit(&zfsdev_state_lock);
897 
898 	return (err);
899 out:
900 	if (zv->zv_total_opens == 0)
901 		zvol_last_close(zv);
902 	mutex_exit(&zfsdev_state_lock);
903 	return (err);
904 }
905 
906 /*ARGSUSED*/
907 int
908 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
909 {
910 	minor_t minor = getminor(dev);
911 	zvol_state_t *zv;
912 	int error = 0;
913 
914 	mutex_enter(&zfsdev_state_lock);
915 
916 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
917 	if (zv == NULL) {
918 		mutex_exit(&zfsdev_state_lock);
919 		return (SET_ERROR(ENXIO));
920 	}
921 
922 	if (zv->zv_flags & ZVOL_EXCL) {
923 		ASSERT(zv->zv_total_opens == 1);
924 		zv->zv_flags &= ~ZVOL_EXCL;
925 	}
926 
927 	/*
928 	 * If the open count is zero, this is a spurious close.
929 	 * That indicates a bug in the kernel / DDI framework.
930 	 */
931 	ASSERT(zv->zv_open_count[otyp] != 0);
932 	ASSERT(zv->zv_total_opens != 0);
933 
934 	/*
935 	 * You may get multiple opens, but only one close.
936 	 */
937 	zv->zv_open_count[otyp]--;
938 	zv->zv_total_opens--;
939 
940 	if (zv->zv_total_opens == 0)
941 		zvol_last_close(zv);
942 
943 	mutex_exit(&zfsdev_state_lock);
944 	return (error);
945 }
946 
947 static void
948 zvol_get_done(zgd_t *zgd, int error)
949 {
950 	if (zgd->zgd_db)
951 		dmu_buf_rele(zgd->zgd_db, zgd);
952 
953 	zfs_range_unlock(zgd->zgd_rl);
954 
955 	if (error == 0 && zgd->zgd_bp)
956 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
957 
958 	kmem_free(zgd, sizeof (zgd_t));
959 }
960 
961 /*
962  * Get data to generate a TX_WRITE intent log record.
963  */
964 static int
965 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
966 {
967 	zvol_state_t *zv = arg;
968 	objset_t *os = zv->zv_objset;
969 	uint64_t object = ZVOL_OBJ;
970 	uint64_t offset = lr->lr_offset;
971 	uint64_t size = lr->lr_length;	/* length of user data */
972 	blkptr_t *bp = &lr->lr_blkptr;
973 	dmu_buf_t *db;
974 	zgd_t *zgd;
975 	int error;
976 
977 	ASSERT(zio != NULL);
978 	ASSERT(size != 0);
979 
980 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
981 	zgd->zgd_zilog = zv->zv_zilog;
982 	zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
983 
984 	/*
985 	 * Write records come in two flavors: immediate and indirect.
986 	 * For small writes it's cheaper to store the data with the
987 	 * log record (immediate); for large writes it's cheaper to
988 	 * sync the data and get a pointer to it (indirect) so that
989 	 * we don't have to write the data twice.
990 	 */
991 	if (buf != NULL) {	/* immediate write */
992 		error = dmu_read(os, object, offset, size, buf,
993 		    DMU_READ_NO_PREFETCH);
994 	} else {
995 		size = zv->zv_volblocksize;
996 		offset = P2ALIGN(offset, size);
997 		error = dmu_buf_hold(os, object, offset, zgd, &db,
998 		    DMU_READ_NO_PREFETCH);
999 		if (error == 0) {
1000 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1001 			if (obp) {
1002 				ASSERT(BP_IS_HOLE(bp));
1003 				*bp = *obp;
1004 			}
1005 
1006 			zgd->zgd_db = db;
1007 			zgd->zgd_bp = bp;
1008 
1009 			ASSERT(db->db_offset == offset);
1010 			ASSERT(db->db_size == size);
1011 
1012 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1013 			    zvol_get_done, zgd);
1014 
1015 			if (error == 0)
1016 				return (0);
1017 		}
1018 	}
1019 
1020 	zvol_get_done(zgd, error);
1021 
1022 	return (error);
1023 }
1024 
1025 /*
1026  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1027  *
1028  * We store data in the log buffers if it's small enough.
1029  * Otherwise we will later flush the data out via dmu_sync().
1030  */
1031 ssize_t zvol_immediate_write_sz = 32768;
1032 
1033 static void
1034 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1035     boolean_t sync)
1036 {
1037 	uint32_t blocksize = zv->zv_volblocksize;
1038 	zilog_t *zilog = zv->zv_zilog;
1039 	boolean_t slogging;
1040 	ssize_t immediate_write_sz;
1041 
1042 	if (zil_replaying(zilog, tx))
1043 		return;
1044 
1045 	immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1046 	    ? 0 : zvol_immediate_write_sz;
1047 
1048 	slogging = spa_has_slogs(zilog->zl_spa) &&
1049 	    (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1050 
1051 	while (resid) {
1052 		itx_t *itx;
1053 		lr_write_t *lr;
1054 		ssize_t len;
1055 		itx_wr_state_t write_state;
1056 
1057 		/*
1058 		 * Unlike zfs_log_write() we can be called with
1059 		 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1060 		 */
1061 		if (blocksize > immediate_write_sz && !slogging &&
1062 		    resid >= blocksize && off % blocksize == 0) {
1063 			write_state = WR_INDIRECT; /* uses dmu_sync */
1064 			len = blocksize;
1065 		} else if (sync) {
1066 			write_state = WR_COPIED;
1067 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1068 		} else {
1069 			write_state = WR_NEED_COPY;
1070 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1071 		}
1072 
1073 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1074 		    (write_state == WR_COPIED ? len : 0));
1075 		lr = (lr_write_t *)&itx->itx_lr;
1076 		if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1077 		    ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1078 			zil_itx_destroy(itx);
1079 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1080 			lr = (lr_write_t *)&itx->itx_lr;
1081 			write_state = WR_NEED_COPY;
1082 		}
1083 
1084 		itx->itx_wr_state = write_state;
1085 		if (write_state == WR_NEED_COPY)
1086 			itx->itx_sod += len;
1087 		lr->lr_foid = ZVOL_OBJ;
1088 		lr->lr_offset = off;
1089 		lr->lr_length = len;
1090 		lr->lr_blkoff = 0;
1091 		BP_ZERO(&lr->lr_blkptr);
1092 
1093 		itx->itx_private = zv;
1094 		itx->itx_sync = sync;
1095 
1096 		zil_itx_assign(zilog, itx, tx);
1097 
1098 		off += len;
1099 		resid -= len;
1100 	}
1101 }
1102 
1103 static int
1104 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size,
1105     boolean_t doread, boolean_t isdump)
1106 {
1107 	vdev_disk_t *dvd;
1108 	int c;
1109 	int numerrors = 0;
1110 
1111 	for (c = 0; c < vd->vdev_children; c++) {
1112 		ASSERT(vd->vdev_ops == &vdev_mirror_ops ||
1113 		    vd->vdev_ops == &vdev_replacing_ops ||
1114 		    vd->vdev_ops == &vdev_spare_ops);
1115 		int err = zvol_dumpio_vdev(vd->vdev_child[c],
1116 		    addr, offset, size, doread, isdump);
1117 		if (err != 0) {
1118 			numerrors++;
1119 		} else if (doread) {
1120 			break;
1121 		}
1122 	}
1123 
1124 	if (!vd->vdev_ops->vdev_op_leaf)
1125 		return (numerrors < vd->vdev_children ? 0 : EIO);
1126 
1127 	if (doread && !vdev_readable(vd))
1128 		return (SET_ERROR(EIO));
1129 	else if (!doread && !vdev_writeable(vd))
1130 		return (SET_ERROR(EIO));
1131 
1132 	dvd = vd->vdev_tsd;
1133 	ASSERT3P(dvd, !=, NULL);
1134 	offset += VDEV_LABEL_START_SIZE;
1135 
1136 	if (ddi_in_panic() || isdump) {
1137 		ASSERT(!doread);
1138 		if (doread)
1139 			return (SET_ERROR(EIO));
1140 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1141 		    lbtodb(size)));
1142 	} else {
1143 		return (vdev_disk_physio(dvd->vd_lh, addr, size, offset,
1144 		    doread ? B_READ : B_WRITE));
1145 	}
1146 }
1147 
1148 static int
1149 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1150     boolean_t doread, boolean_t isdump)
1151 {
1152 	vdev_t *vd;
1153 	int error;
1154 	zvol_extent_t *ze;
1155 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1156 
1157 	/* Must be sector aligned, and not stradle a block boundary. */
1158 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1159 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1160 		return (SET_ERROR(EINVAL));
1161 	}
1162 	ASSERT(size <= zv->zv_volblocksize);
1163 
1164 	/* Locate the extent this belongs to */
1165 	ze = list_head(&zv->zv_extents);
1166 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1167 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1168 		ze = list_next(&zv->zv_extents, ze);
1169 	}
1170 
1171 	if (ze == NULL)
1172 		return (SET_ERROR(EINVAL));
1173 
1174 	if (!ddi_in_panic())
1175 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1176 
1177 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1178 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1179 	error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump);
1180 
1181 	if (!ddi_in_panic())
1182 		spa_config_exit(spa, SCL_STATE, FTAG);
1183 
1184 	return (error);
1185 }
1186 
1187 int
1188 zvol_strategy(buf_t *bp)
1189 {
1190 	zfs_soft_state_t *zs = NULL;
1191 	zvol_state_t *zv;
1192 	uint64_t off, volsize;
1193 	size_t resid;
1194 	char *addr;
1195 	objset_t *os;
1196 	rl_t *rl;
1197 	int error = 0;
1198 	boolean_t doread = bp->b_flags & B_READ;
1199 	boolean_t is_dump;
1200 	boolean_t sync;
1201 
1202 	if (getminor(bp->b_edev) == 0) {
1203 		error = SET_ERROR(EINVAL);
1204 	} else {
1205 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1206 		if (zs == NULL)
1207 			error = SET_ERROR(ENXIO);
1208 		else if (zs->zss_type != ZSST_ZVOL)
1209 			error = SET_ERROR(EINVAL);
1210 	}
1211 
1212 	if (error) {
1213 		bioerror(bp, error);
1214 		biodone(bp);
1215 		return (0);
1216 	}
1217 
1218 	zv = zs->zss_data;
1219 
1220 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1221 		bioerror(bp, EROFS);
1222 		biodone(bp);
1223 		return (0);
1224 	}
1225 
1226 	off = ldbtob(bp->b_blkno);
1227 	volsize = zv->zv_volsize;
1228 
1229 	os = zv->zv_objset;
1230 	ASSERT(os != NULL);
1231 
1232 	bp_mapin(bp);
1233 	addr = bp->b_un.b_addr;
1234 	resid = bp->b_bcount;
1235 
1236 	if (resid > 0 && (off < 0 || off >= volsize)) {
1237 		bioerror(bp, EIO);
1238 		biodone(bp);
1239 		return (0);
1240 	}
1241 
1242 	is_dump = zv->zv_flags & ZVOL_DUMPIFIED;
1243 	sync = ((!(bp->b_flags & B_ASYNC) &&
1244 	    !(zv->zv_flags & ZVOL_WCE)) ||
1245 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1246 	    !doread && !is_dump;
1247 
1248 	/*
1249 	 * There must be no buffer changes when doing a dmu_sync() because
1250 	 * we can't change the data whilst calculating the checksum.
1251 	 */
1252 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1253 	    doread ? RL_READER : RL_WRITER);
1254 
1255 	while (resid != 0 && off < volsize) {
1256 		size_t size = MIN(resid, zvol_maxphys);
1257 		if (is_dump) {
1258 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1259 			error = zvol_dumpio(zv, addr, off, size,
1260 			    doread, B_FALSE);
1261 		} else if (doread) {
1262 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1263 			    DMU_READ_PREFETCH);
1264 		} else {
1265 			dmu_tx_t *tx = dmu_tx_create(os);
1266 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1267 			error = dmu_tx_assign(tx, TXG_WAIT);
1268 			if (error) {
1269 				dmu_tx_abort(tx);
1270 			} else {
1271 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1272 				zvol_log_write(zv, tx, off, size, sync);
1273 				dmu_tx_commit(tx);
1274 			}
1275 		}
1276 		if (error) {
1277 			/* convert checksum errors into IO errors */
1278 			if (error == ECKSUM)
1279 				error = SET_ERROR(EIO);
1280 			break;
1281 		}
1282 		off += size;
1283 		addr += size;
1284 		resid -= size;
1285 	}
1286 	zfs_range_unlock(rl);
1287 
1288 	if ((bp->b_resid = resid) == bp->b_bcount)
1289 		bioerror(bp, off > volsize ? EINVAL : error);
1290 
1291 	if (sync)
1292 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1293 	biodone(bp);
1294 
1295 	return (0);
1296 }
1297 
1298 /*
1299  * Set the buffer count to the zvol maximum transfer.
1300  * Using our own routine instead of the default minphys()
1301  * means that for larger writes we write bigger buffers on X86
1302  * (128K instead of 56K) and flush the disk write cache less often
1303  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1304  * 56K on X86 and 128K on sparc).
1305  */
1306 void
1307 zvol_minphys(struct buf *bp)
1308 {
1309 	if (bp->b_bcount > zvol_maxphys)
1310 		bp->b_bcount = zvol_maxphys;
1311 }
1312 
1313 int
1314 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1315 {
1316 	minor_t minor = getminor(dev);
1317 	zvol_state_t *zv;
1318 	int error = 0;
1319 	uint64_t size;
1320 	uint64_t boff;
1321 	uint64_t resid;
1322 
1323 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1324 	if (zv == NULL)
1325 		return (SET_ERROR(ENXIO));
1326 
1327 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1328 		return (SET_ERROR(EINVAL));
1329 
1330 	boff = ldbtob(blkno);
1331 	resid = ldbtob(nblocks);
1332 
1333 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1334 
1335 	while (resid) {
1336 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1337 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1338 		if (error)
1339 			break;
1340 		boff += size;
1341 		addr += size;
1342 		resid -= size;
1343 	}
1344 
1345 	return (error);
1346 }
1347 
1348 /*ARGSUSED*/
1349 int
1350 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1351 {
1352 	minor_t minor = getminor(dev);
1353 	zvol_state_t *zv;
1354 	uint64_t volsize;
1355 	rl_t *rl;
1356 	int error = 0;
1357 
1358 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1359 	if (zv == NULL)
1360 		return (SET_ERROR(ENXIO));
1361 
1362 	volsize = zv->zv_volsize;
1363 	if (uio->uio_resid > 0 &&
1364 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1365 		return (SET_ERROR(EIO));
1366 
1367 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1368 		error = physio(zvol_strategy, NULL, dev, B_READ,
1369 		    zvol_minphys, uio);
1370 		return (error);
1371 	}
1372 
1373 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1374 	    RL_READER);
1375 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1376 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1377 
1378 		/* don't read past the end */
1379 		if (bytes > volsize - uio->uio_loffset)
1380 			bytes = volsize - uio->uio_loffset;
1381 
1382 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1383 		if (error) {
1384 			/* convert checksum errors into IO errors */
1385 			if (error == ECKSUM)
1386 				error = SET_ERROR(EIO);
1387 			break;
1388 		}
1389 	}
1390 	zfs_range_unlock(rl);
1391 	return (error);
1392 }
1393 
1394 /*ARGSUSED*/
1395 int
1396 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1397 {
1398 	minor_t minor = getminor(dev);
1399 	zvol_state_t *zv;
1400 	uint64_t volsize;
1401 	rl_t *rl;
1402 	int error = 0;
1403 	boolean_t sync;
1404 
1405 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1406 	if (zv == NULL)
1407 		return (SET_ERROR(ENXIO));
1408 
1409 	volsize = zv->zv_volsize;
1410 	if (uio->uio_resid > 0 &&
1411 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1412 		return (SET_ERROR(EIO));
1413 
1414 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1415 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1416 		    zvol_minphys, uio);
1417 		return (error);
1418 	}
1419 
1420 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1421 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1422 
1423 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1424 	    RL_WRITER);
1425 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1426 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1427 		uint64_t off = uio->uio_loffset;
1428 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1429 
1430 		if (bytes > volsize - off)	/* don't write past the end */
1431 			bytes = volsize - off;
1432 
1433 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1434 		error = dmu_tx_assign(tx, TXG_WAIT);
1435 		if (error) {
1436 			dmu_tx_abort(tx);
1437 			break;
1438 		}
1439 		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1440 		if (error == 0)
1441 			zvol_log_write(zv, tx, off, bytes, sync);
1442 		dmu_tx_commit(tx);
1443 
1444 		if (error)
1445 			break;
1446 	}
1447 	zfs_range_unlock(rl);
1448 	if (sync)
1449 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1450 	return (error);
1451 }
1452 
1453 int
1454 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1455 {
1456 	struct uuid uuid = EFI_RESERVED;
1457 	efi_gpe_t gpe = { 0 };
1458 	uint32_t crc;
1459 	dk_efi_t efi;
1460 	int length;
1461 	char *ptr;
1462 
1463 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1464 		return (SET_ERROR(EFAULT));
1465 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1466 	length = efi.dki_length;
1467 	/*
1468 	 * Some clients may attempt to request a PMBR for the
1469 	 * zvol.  Currently this interface will return EINVAL to
1470 	 * such requests.  These requests could be supported by
1471 	 * adding a check for lba == 0 and consing up an appropriate
1472 	 * PMBR.
1473 	 */
1474 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1475 		return (SET_ERROR(EINVAL));
1476 
1477 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1478 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1479 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1480 
1481 	if (efi.dki_lba == 1) {
1482 		efi_gpt_t gpt = { 0 };
1483 
1484 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1485 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1486 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1487 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1488 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1489 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1490 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1491 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1492 		gpt.efi_gpt_SizeOfPartitionEntry =
1493 		    LE_32(sizeof (efi_gpe_t));
1494 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1495 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1496 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1497 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1498 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1499 		    flag))
1500 			return (SET_ERROR(EFAULT));
1501 		ptr += sizeof (gpt);
1502 		length -= sizeof (gpt);
1503 	}
1504 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1505 	    length), flag))
1506 		return (SET_ERROR(EFAULT));
1507 	return (0);
1508 }
1509 
1510 /*
1511  * BEGIN entry points to allow external callers access to the volume.
1512  */
1513 /*
1514  * Return the volume parameters needed for access from an external caller.
1515  * These values are invariant as long as the volume is held open.
1516  */
1517 int
1518 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1519     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1520     void **rl_hdl, void **bonus_hdl)
1521 {
1522 	zvol_state_t *zv;
1523 
1524 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1525 	if (zv == NULL)
1526 		return (SET_ERROR(ENXIO));
1527 	if (zv->zv_flags & ZVOL_DUMPIFIED)
1528 		return (SET_ERROR(ENXIO));
1529 
1530 	ASSERT(blksize && max_xfer_len && minor_hdl &&
1531 	    objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1532 
1533 	*blksize = zv->zv_volblocksize;
1534 	*max_xfer_len = (uint64_t)zvol_maxphys;
1535 	*minor_hdl = zv;
1536 	*objset_hdl = zv->zv_objset;
1537 	*zil_hdl = zv->zv_zilog;
1538 	*rl_hdl = &zv->zv_znode;
1539 	*bonus_hdl = zv->zv_dbuf;
1540 	return (0);
1541 }
1542 
1543 /*
1544  * Return the current volume size to an external caller.
1545  * The size can change while the volume is open.
1546  */
1547 uint64_t
1548 zvol_get_volume_size(void *minor_hdl)
1549 {
1550 	zvol_state_t *zv = minor_hdl;
1551 
1552 	return (zv->zv_volsize);
1553 }
1554 
1555 /*
1556  * Return the current WCE setting to an external caller.
1557  * The WCE setting can change while the volume is open.
1558  */
1559 int
1560 zvol_get_volume_wce(void *minor_hdl)
1561 {
1562 	zvol_state_t *zv = minor_hdl;
1563 
1564 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1565 }
1566 
1567 /*
1568  * Entry point for external callers to zvol_log_write
1569  */
1570 void
1571 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1572     boolean_t sync)
1573 {
1574 	zvol_state_t *zv = minor_hdl;
1575 
1576 	zvol_log_write(zv, tx, off, resid, sync);
1577 }
1578 /*
1579  * END entry points to allow external callers access to the volume.
1580  */
1581 
1582 /*
1583  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1584  */
1585 static void
1586 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1587     boolean_t sync)
1588 {
1589 	itx_t *itx;
1590 	lr_truncate_t *lr;
1591 	zilog_t *zilog = zv->zv_zilog;
1592 
1593 	if (zil_replaying(zilog, tx))
1594 		return;
1595 
1596 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1597 	lr = (lr_truncate_t *)&itx->itx_lr;
1598 	lr->lr_foid = ZVOL_OBJ;
1599 	lr->lr_offset = off;
1600 	lr->lr_length = len;
1601 
1602 	itx->itx_sync = sync;
1603 	zil_itx_assign(zilog, itx, tx);
1604 }
1605 
1606 /*
1607  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1608  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1609  */
1610 /*ARGSUSED*/
1611 int
1612 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1613 {
1614 	zvol_state_t *zv;
1615 	struct dk_cinfo dki;
1616 	struct dk_minfo dkm;
1617 	struct dk_callback *dkc;
1618 	int error = 0;
1619 	rl_t *rl;
1620 
1621 	mutex_enter(&zfsdev_state_lock);
1622 
1623 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1624 
1625 	if (zv == NULL) {
1626 		mutex_exit(&zfsdev_state_lock);
1627 		return (SET_ERROR(ENXIO));
1628 	}
1629 	ASSERT(zv->zv_total_opens > 0);
1630 
1631 	switch (cmd) {
1632 
1633 	case DKIOCINFO:
1634 		bzero(&dki, sizeof (dki));
1635 		(void) strcpy(dki.dki_cname, "zvol");
1636 		(void) strcpy(dki.dki_dname, "zvol");
1637 		dki.dki_ctype = DKC_UNKNOWN;
1638 		dki.dki_unit = getminor(dev);
1639 		dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1640 		mutex_exit(&zfsdev_state_lock);
1641 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1642 			error = SET_ERROR(EFAULT);
1643 		return (error);
1644 
1645 	case DKIOCGMEDIAINFO:
1646 		bzero(&dkm, sizeof (dkm));
1647 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1648 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1649 		dkm.dki_media_type = DK_UNKNOWN;
1650 		mutex_exit(&zfsdev_state_lock);
1651 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1652 			error = SET_ERROR(EFAULT);
1653 		return (error);
1654 
1655 	case DKIOCGETEFI:
1656 		{
1657 			uint64_t vs = zv->zv_volsize;
1658 			uint8_t bs = zv->zv_min_bs;
1659 
1660 			mutex_exit(&zfsdev_state_lock);
1661 			error = zvol_getefi((void *)arg, flag, vs, bs);
1662 			return (error);
1663 		}
1664 
1665 	case DKIOCFLUSHWRITECACHE:
1666 		dkc = (struct dk_callback *)arg;
1667 		mutex_exit(&zfsdev_state_lock);
1668 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1669 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1670 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1671 			error = 0;
1672 		}
1673 		return (error);
1674 
1675 	case DKIOCGETWCE:
1676 		{
1677 			int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1678 			if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1679 			    flag))
1680 				error = SET_ERROR(EFAULT);
1681 			break;
1682 		}
1683 	case DKIOCSETWCE:
1684 		{
1685 			int wce;
1686 			if (ddi_copyin((void *)arg, &wce, sizeof (int),
1687 			    flag)) {
1688 				error = SET_ERROR(EFAULT);
1689 				break;
1690 			}
1691 			if (wce) {
1692 				zv->zv_flags |= ZVOL_WCE;
1693 				mutex_exit(&zfsdev_state_lock);
1694 			} else {
1695 				zv->zv_flags &= ~ZVOL_WCE;
1696 				mutex_exit(&zfsdev_state_lock);
1697 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1698 			}
1699 			return (0);
1700 		}
1701 
1702 	case DKIOCGGEOM:
1703 	case DKIOCGVTOC:
1704 		/*
1705 		 * commands using these (like prtvtoc) expect ENOTSUP
1706 		 * since we're emulating an EFI label
1707 		 */
1708 		error = SET_ERROR(ENOTSUP);
1709 		break;
1710 
1711 	case DKIOCDUMPINIT:
1712 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1713 		    RL_WRITER);
1714 		error = zvol_dumpify(zv);
1715 		zfs_range_unlock(rl);
1716 		break;
1717 
1718 	case DKIOCDUMPFINI:
1719 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1720 			break;
1721 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1722 		    RL_WRITER);
1723 		error = zvol_dump_fini(zv);
1724 		zfs_range_unlock(rl);
1725 		break;
1726 
1727 	case DKIOCFREE:
1728 	{
1729 		dkioc_free_t df;
1730 		dmu_tx_t *tx;
1731 
1732 		if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1733 			error = SET_ERROR(EFAULT);
1734 			break;
1735 		}
1736 
1737 		/*
1738 		 * Apply Postel's Law to length-checking.  If they overshoot,
1739 		 * just blank out until the end, if there's a need to blank
1740 		 * out anything.
1741 		 */
1742 		if (df.df_start >= zv->zv_volsize)
1743 			break;	/* No need to do anything... */
1744 		if (df.df_start + df.df_length > zv->zv_volsize)
1745 			df.df_length = DMU_OBJECT_END;
1746 
1747 		rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1748 		    RL_WRITER);
1749 		tx = dmu_tx_create(zv->zv_objset);
1750 		error = dmu_tx_assign(tx, TXG_WAIT);
1751 		if (error != 0) {
1752 			dmu_tx_abort(tx);
1753 		} else {
1754 			zvol_log_truncate(zv, tx, df.df_start,
1755 			    df.df_length, B_TRUE);
1756 			dmu_tx_commit(tx);
1757 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1758 			    df.df_start, df.df_length);
1759 		}
1760 
1761 		zfs_range_unlock(rl);
1762 
1763 		if (error == 0) {
1764 			/*
1765 			 * If the write-cache is disabled or 'sync' property
1766 			 * is set to 'always' then treat this as a synchronous
1767 			 * operation (i.e. commit to zil).
1768 			 */
1769 			if (!(zv->zv_flags & ZVOL_WCE) ||
1770 			    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
1771 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1772 
1773 			/*
1774 			 * If the caller really wants synchronous writes, and
1775 			 * can't wait for them, don't return until the write
1776 			 * is done.
1777 			 */
1778 			if (df.df_flags & DF_WAIT_SYNC) {
1779 				txg_wait_synced(
1780 				    dmu_objset_pool(zv->zv_objset), 0);
1781 			}
1782 		}
1783 		break;
1784 	}
1785 
1786 	default:
1787 		error = SET_ERROR(ENOTTY);
1788 		break;
1789 
1790 	}
1791 	mutex_exit(&zfsdev_state_lock);
1792 	return (error);
1793 }
1794 
1795 int
1796 zvol_busy(void)
1797 {
1798 	return (zvol_minors != 0);
1799 }
1800 
1801 void
1802 zvol_init(void)
1803 {
1804 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1805 	    1) == 0);
1806 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1807 }
1808 
1809 void
1810 zvol_fini(void)
1811 {
1812 	mutex_destroy(&zfsdev_state_lock);
1813 	ddi_soft_state_fini(&zfsdev_state);
1814 }
1815 
1816 static int
1817 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1818 {
1819 	dmu_tx_t *tx;
1820 	int error = 0;
1821 	objset_t *os = zv->zv_objset;
1822 	nvlist_t *nv = NULL;
1823 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1824 
1825 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1826 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1827 	    DMU_OBJECT_END);
1828 	/* wait for dmu_free_long_range to actually free the blocks */
1829 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1830 
1831 	tx = dmu_tx_create(os);
1832 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1833 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1834 	error = dmu_tx_assign(tx, TXG_WAIT);
1835 	if (error) {
1836 		dmu_tx_abort(tx);
1837 		return (error);
1838 	}
1839 
1840 	/*
1841 	 * If we are resizing the dump device then we only need to
1842 	 * update the refreservation to match the newly updated
1843 	 * zvolsize. Otherwise, we save off the original state of the
1844 	 * zvol so that we can restore them if the zvol is ever undumpified.
1845 	 */
1846 	if (resize) {
1847 		error = zap_update(os, ZVOL_ZAP_OBJ,
1848 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1849 		    &zv->zv_volsize, tx);
1850 	} else {
1851 		uint64_t checksum, compress, refresrv, vbs, dedup;
1852 
1853 		error = dsl_prop_get_integer(zv->zv_name,
1854 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1855 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1856 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1857 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1858 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1859 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1860 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1861 		if (version >= SPA_VERSION_DEDUP) {
1862 			error = error ? error :
1863 			    dsl_prop_get_integer(zv->zv_name,
1864 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1865 		}
1866 
1867 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1868 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1869 		    &compress, tx);
1870 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1871 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1872 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1873 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1874 		    &refresrv, tx);
1875 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1876 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1877 		    &vbs, tx);
1878 		error = error ? error : dmu_object_set_blocksize(
1879 		    os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1880 		if (version >= SPA_VERSION_DEDUP) {
1881 			error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1882 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1883 			    &dedup, tx);
1884 		}
1885 		if (error == 0)
1886 			zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1887 	}
1888 	dmu_tx_commit(tx);
1889 
1890 	/*
1891 	 * We only need update the zvol's property if we are initializing
1892 	 * the dump area for the first time.
1893 	 */
1894 	if (!resize) {
1895 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1896 		VERIFY(nvlist_add_uint64(nv,
1897 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1898 		VERIFY(nvlist_add_uint64(nv,
1899 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1900 		    ZIO_COMPRESS_OFF) == 0);
1901 		VERIFY(nvlist_add_uint64(nv,
1902 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1903 		    ZIO_CHECKSUM_OFF) == 0);
1904 		if (version >= SPA_VERSION_DEDUP) {
1905 			VERIFY(nvlist_add_uint64(nv,
1906 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
1907 			    ZIO_CHECKSUM_OFF) == 0);
1908 		}
1909 
1910 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1911 		    nv, NULL);
1912 		nvlist_free(nv);
1913 
1914 		if (error)
1915 			return (error);
1916 	}
1917 
1918 	/* Allocate the space for the dump */
1919 	error = zvol_prealloc(zv);
1920 	return (error);
1921 }
1922 
1923 static int
1924 zvol_dumpify(zvol_state_t *zv)
1925 {
1926 	int error = 0;
1927 	uint64_t dumpsize = 0;
1928 	dmu_tx_t *tx;
1929 	objset_t *os = zv->zv_objset;
1930 
1931 	if (zv->zv_flags & ZVOL_RDONLY)
1932 		return (SET_ERROR(EROFS));
1933 
1934 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
1935 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
1936 		boolean_t resize = (dumpsize > 0);
1937 
1938 		if ((error = zvol_dump_init(zv, resize)) != 0) {
1939 			(void) zvol_dump_fini(zv);
1940 			return (error);
1941 		}
1942 	}
1943 
1944 	/*
1945 	 * Build up our lba mapping.
1946 	 */
1947 	error = zvol_get_lbas(zv);
1948 	if (error) {
1949 		(void) zvol_dump_fini(zv);
1950 		return (error);
1951 	}
1952 
1953 	tx = dmu_tx_create(os);
1954 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1955 	error = dmu_tx_assign(tx, TXG_WAIT);
1956 	if (error) {
1957 		dmu_tx_abort(tx);
1958 		(void) zvol_dump_fini(zv);
1959 		return (error);
1960 	}
1961 
1962 	zv->zv_flags |= ZVOL_DUMPIFIED;
1963 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
1964 	    &zv->zv_volsize, tx);
1965 	dmu_tx_commit(tx);
1966 
1967 	if (error) {
1968 		(void) zvol_dump_fini(zv);
1969 		return (error);
1970 	}
1971 
1972 	txg_wait_synced(dmu_objset_pool(os), 0);
1973 	return (0);
1974 }
1975 
1976 static int
1977 zvol_dump_fini(zvol_state_t *zv)
1978 {
1979 	dmu_tx_t *tx;
1980 	objset_t *os = zv->zv_objset;
1981 	nvlist_t *nv;
1982 	int error = 0;
1983 	uint64_t checksum, compress, refresrv, vbs, dedup;
1984 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1985 
1986 	/*
1987 	 * Attempt to restore the zvol back to its pre-dumpified state.
1988 	 * This is a best-effort attempt as it's possible that not all
1989 	 * of these properties were initialized during the dumpify process
1990 	 * (i.e. error during zvol_dump_init).
1991 	 */
1992 
1993 	tx = dmu_tx_create(os);
1994 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1995 	error = dmu_tx_assign(tx, TXG_WAIT);
1996 	if (error) {
1997 		dmu_tx_abort(tx);
1998 		return (error);
1999 	}
2000 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2001 	dmu_tx_commit(tx);
2002 
2003 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2004 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2005 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2006 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2007 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2008 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2009 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2010 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2011 
2012 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2013 	(void) nvlist_add_uint64(nv,
2014 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2015 	(void) nvlist_add_uint64(nv,
2016 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2017 	(void) nvlist_add_uint64(nv,
2018 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2019 	if (version >= SPA_VERSION_DEDUP &&
2020 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2021 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2022 		(void) nvlist_add_uint64(nv,
2023 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2024 	}
2025 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2026 	    nv, NULL);
2027 	nvlist_free(nv);
2028 
2029 	zvol_free_extents(zv);
2030 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2031 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2032 	/* wait for dmu_free_long_range to actually free the blocks */
2033 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2034 	tx = dmu_tx_create(os);
2035 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2036 	error = dmu_tx_assign(tx, TXG_WAIT);
2037 	if (error) {
2038 		dmu_tx_abort(tx);
2039 		return (error);
2040 	}
2041 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2042 		zv->zv_volblocksize = vbs;
2043 	dmu_tx_commit(tx);
2044 
2045 	return (0);
2046 }
2047