xref: /titanic_44/usr/src/uts/common/fs/zfs/zvol.c (revision a970c705050f9e90c4b6d7982a9b3211719353fb)
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) 2012 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 
84 #include "zfs_namecheck.h"
85 
86 void *zfsdev_state;
87 static char *zvol_tag = "zvol_tag";
88 
89 #define	ZVOL_DUMPSIZE		"dumpsize"
90 
91 /*
92  * This lock protects the zfsdev_state structure from being modified
93  * while it's being used, e.g. an open that comes in before a create
94  * finishes.  It also protects temporary opens of the dataset so that,
95  * e.g., an open doesn't get a spurious EBUSY.
96  */
97 kmutex_t zfsdev_state_lock;
98 static uint32_t zvol_minors;
99 
100 typedef struct zvol_extent {
101 	list_node_t	ze_node;
102 	dva_t		ze_dva;		/* dva associated with this extent */
103 	uint64_t	ze_nblks;	/* number of blocks in extent */
104 } zvol_extent_t;
105 
106 /*
107  * The in-core state of each volume.
108  */
109 typedef struct zvol_state {
110 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
111 	uint64_t	zv_volsize;	/* amount of space we advertise */
112 	uint64_t	zv_volblocksize; /* volume block size */
113 	minor_t		zv_minor;	/* minor number */
114 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
115 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
116 	objset_t	*zv_objset;	/* objset handle */
117 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
118 	uint32_t	zv_total_opens;	/* total open count */
119 	zilog_t		*zv_zilog;	/* ZIL handle */
120 	list_t		zv_extents;	/* List of extents for dump */
121 	znode_t		zv_znode;	/* for range locking */
122 	dmu_buf_t	*zv_dbuf;	/* bonus handle */
123 } zvol_state_t;
124 
125 /*
126  * zvol specific flags
127  */
128 #define	ZVOL_RDONLY	0x1
129 #define	ZVOL_DUMPIFIED	0x2
130 #define	ZVOL_EXCL	0x4
131 #define	ZVOL_WCE	0x8
132 
133 /*
134  * zvol maximum transfer in one DMU tx.
135  */
136 int zvol_maxphys = DMU_MAX_ACCESS/2;
137 
138 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
139     nvlist_t *, nvlist_t *);
140 static int zvol_remove_zv(zvol_state_t *);
141 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
142 static int zvol_dumpify(zvol_state_t *zv);
143 static int zvol_dump_fini(zvol_state_t *zv);
144 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
145 
146 static void
147 zvol_size_changed(uint64_t volsize, major_t maj, minor_t min)
148 {
149 	dev_t dev = makedevice(maj, min);
150 
151 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
152 	    "Size", volsize) == DDI_SUCCESS);
153 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
154 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
155 
156 	/* Notify specfs to invalidate the cached size */
157 	spec_size_invalidate(dev, VBLK);
158 	spec_size_invalidate(dev, VCHR);
159 }
160 
161 int
162 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
163 {
164 	if (volsize == 0)
165 		return (EINVAL);
166 
167 	if (volsize % blocksize != 0)
168 		return (EINVAL);
169 
170 #ifdef _ILP32
171 	if (volsize - 1 > SPEC_MAXOFFSET_T)
172 		return (EOVERFLOW);
173 #endif
174 	return (0);
175 }
176 
177 int
178 zvol_check_volblocksize(uint64_t volblocksize)
179 {
180 	if (volblocksize < SPA_MINBLOCKSIZE ||
181 	    volblocksize > SPA_MAXBLOCKSIZE ||
182 	    !ISP2(volblocksize))
183 		return (EDOM);
184 
185 	return (0);
186 }
187 
188 int
189 zvol_get_stats(objset_t *os, nvlist_t *nv)
190 {
191 	int error;
192 	dmu_object_info_t doi;
193 	uint64_t val;
194 
195 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
196 	if (error)
197 		return (error);
198 
199 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
200 
201 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
202 
203 	if (error == 0) {
204 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
205 		    doi.doi_data_block_size);
206 	}
207 
208 	return (error);
209 }
210 
211 static zvol_state_t *
212 zvol_minor_lookup(const char *name)
213 {
214 	minor_t minor;
215 	zvol_state_t *zv;
216 
217 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
218 
219 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
220 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
221 		if (zv == NULL)
222 			continue;
223 		if (strcmp(zv->zv_name, name) == 0)
224 			return (zv);
225 	}
226 
227 	return (NULL);
228 }
229 
230 /* extent mapping arg */
231 struct maparg {
232 	zvol_state_t	*ma_zv;
233 	uint64_t	ma_blks;
234 };
235 
236 /*ARGSUSED*/
237 static int
238 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
239     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
240 {
241 	struct maparg *ma = arg;
242 	zvol_extent_t *ze;
243 	int bs = ma->ma_zv->zv_volblocksize;
244 
245 	if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
246 		return (0);
247 
248 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
249 	ma->ma_blks++;
250 
251 	/* Abort immediately if we have encountered gang blocks */
252 	if (BP_IS_GANG(bp))
253 		return (EFRAGS);
254 
255 	/*
256 	 * See if the block is at the end of the previous extent.
257 	 */
258 	ze = list_tail(&ma->ma_zv->zv_extents);
259 	if (ze &&
260 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
261 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
262 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
263 		ze->ze_nblks++;
264 		return (0);
265 	}
266 
267 	dprintf_bp(bp, "%s", "next blkptr:");
268 
269 	/* start a new extent */
270 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
271 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
272 	ze->ze_nblks = 1;
273 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
274 	return (0);
275 }
276 
277 static void
278 zvol_free_extents(zvol_state_t *zv)
279 {
280 	zvol_extent_t *ze;
281 
282 	while (ze = list_head(&zv->zv_extents)) {
283 		list_remove(&zv->zv_extents, ze);
284 		kmem_free(ze, sizeof (zvol_extent_t));
285 	}
286 }
287 
288 static int
289 zvol_get_lbas(zvol_state_t *zv)
290 {
291 	objset_t *os = zv->zv_objset;
292 	struct maparg	ma;
293 	int		err;
294 
295 	ma.ma_zv = zv;
296 	ma.ma_blks = 0;
297 	zvol_free_extents(zv);
298 
299 	/* commit any in-flight changes before traversing the dataset */
300 	txg_wait_synced(dmu_objset_pool(os), 0);
301 	err = traverse_dataset(dmu_objset_ds(os), 0,
302 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
303 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
304 		zvol_free_extents(zv);
305 		return (err ? err : EIO);
306 	}
307 
308 	return (0);
309 }
310 
311 /* ARGSUSED */
312 void
313 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
314 {
315 	zfs_creat_t *zct = arg;
316 	nvlist_t *nvprops = zct->zct_props;
317 	int error;
318 	uint64_t volblocksize, volsize;
319 
320 	VERIFY(nvlist_lookup_uint64(nvprops,
321 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
322 	if (nvlist_lookup_uint64(nvprops,
323 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
324 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
325 
326 	/*
327 	 * These properties must be removed from the list so the generic
328 	 * property setting step won't apply to them.
329 	 */
330 	VERIFY(nvlist_remove_all(nvprops,
331 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
332 	(void) nvlist_remove_all(nvprops,
333 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
334 
335 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
336 	    DMU_OT_NONE, 0, tx);
337 	ASSERT(error == 0);
338 
339 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
340 	    DMU_OT_NONE, 0, tx);
341 	ASSERT(error == 0);
342 
343 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
344 	ASSERT(error == 0);
345 }
346 
347 /*
348  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
349  * implement DKIOCFREE/free-long-range.
350  */
351 static int
352 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
353 {
354 	uint64_t offset, length;
355 
356 	if (byteswap)
357 		byteswap_uint64_array(lr, sizeof (*lr));
358 
359 	offset = lr->lr_offset;
360 	length = lr->lr_length;
361 
362 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
363 }
364 
365 /*
366  * Replay a TX_WRITE ZIL transaction that didn't get committed
367  * after a system failure
368  */
369 static int
370 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
371 {
372 	objset_t *os = zv->zv_objset;
373 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
374 	uint64_t offset, length;
375 	dmu_tx_t *tx;
376 	int error;
377 
378 	if (byteswap)
379 		byteswap_uint64_array(lr, sizeof (*lr));
380 
381 	offset = lr->lr_offset;
382 	length = lr->lr_length;
383 
384 	/* If it's a dmu_sync() block, write the whole block */
385 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
386 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
387 		if (length < blocksize) {
388 			offset -= offset % blocksize;
389 			length = blocksize;
390 		}
391 	}
392 
393 	tx = dmu_tx_create(os);
394 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
395 	error = dmu_tx_assign(tx, TXG_WAIT);
396 	if (error) {
397 		dmu_tx_abort(tx);
398 	} else {
399 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
400 		dmu_tx_commit(tx);
401 	}
402 
403 	return (error);
404 }
405 
406 /* ARGSUSED */
407 static int
408 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
409 {
410 	return (ENOTSUP);
411 }
412 
413 /*
414  * Callback vectors for replaying records.
415  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
416  */
417 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
418 	zvol_replay_err,	/* 0 no such transaction type */
419 	zvol_replay_err,	/* TX_CREATE */
420 	zvol_replay_err,	/* TX_MKDIR */
421 	zvol_replay_err,	/* TX_MKXATTR */
422 	zvol_replay_err,	/* TX_SYMLINK */
423 	zvol_replay_err,	/* TX_REMOVE */
424 	zvol_replay_err,	/* TX_RMDIR */
425 	zvol_replay_err,	/* TX_LINK */
426 	zvol_replay_err,	/* TX_RENAME */
427 	zvol_replay_write,	/* TX_WRITE */
428 	zvol_replay_truncate,	/* TX_TRUNCATE */
429 	zvol_replay_err,	/* TX_SETATTR */
430 	zvol_replay_err,	/* TX_ACL */
431 	zvol_replay_err,	/* TX_CREATE_ACL */
432 	zvol_replay_err,	/* TX_CREATE_ATTR */
433 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
434 	zvol_replay_err,	/* TX_MKDIR_ACL */
435 	zvol_replay_err,	/* TX_MKDIR_ATTR */
436 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
437 	zvol_replay_err,	/* TX_WRITE2 */
438 };
439 
440 int
441 zvol_name2minor(const char *name, minor_t *minor)
442 {
443 	zvol_state_t *zv;
444 
445 	mutex_enter(&zfsdev_state_lock);
446 	zv = zvol_minor_lookup(name);
447 	if (minor && zv)
448 		*minor = zv->zv_minor;
449 	mutex_exit(&zfsdev_state_lock);
450 	return (zv ? 0 : -1);
451 }
452 
453 /*
454  * Create a minor node (plus a whole lot more) for the specified volume.
455  */
456 int
457 zvol_create_minor(const char *name)
458 {
459 	zfs_soft_state_t *zs;
460 	zvol_state_t *zv;
461 	objset_t *os;
462 	dmu_object_info_t doi;
463 	minor_t minor = 0;
464 	char chrbuf[30], blkbuf[30];
465 	int error;
466 
467 	mutex_enter(&zfsdev_state_lock);
468 
469 	if (zvol_minor_lookup(name) != NULL) {
470 		mutex_exit(&zfsdev_state_lock);
471 		return (EEXIST);
472 	}
473 
474 	/* lie and say we're read-only */
475 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
476 
477 	if (error) {
478 		mutex_exit(&zfsdev_state_lock);
479 		return (error);
480 	}
481 
482 	if ((minor = zfsdev_minor_alloc()) == 0) {
483 		dmu_objset_disown(os, FTAG);
484 		mutex_exit(&zfsdev_state_lock);
485 		return (ENXIO);
486 	}
487 
488 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
489 		dmu_objset_disown(os, FTAG);
490 		mutex_exit(&zfsdev_state_lock);
491 		return (EAGAIN);
492 	}
493 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
494 	    (char *)name);
495 
496 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
497 
498 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
499 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
500 		ddi_soft_state_free(zfsdev_state, minor);
501 		dmu_objset_disown(os, FTAG);
502 		mutex_exit(&zfsdev_state_lock);
503 		return (EAGAIN);
504 	}
505 
506 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
507 
508 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
509 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
510 		ddi_remove_minor_node(zfs_dip, chrbuf);
511 		ddi_soft_state_free(zfsdev_state, minor);
512 		dmu_objset_disown(os, FTAG);
513 		mutex_exit(&zfsdev_state_lock);
514 		return (EAGAIN);
515 	}
516 
517 	zs = ddi_get_soft_state(zfsdev_state, minor);
518 	zs->zss_type = ZSST_ZVOL;
519 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
520 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
521 	zv->zv_min_bs = DEV_BSHIFT;
522 	zv->zv_minor = minor;
523 	zv->zv_objset = os;
524 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
525 		zv->zv_flags |= ZVOL_RDONLY;
526 	mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
527 	avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
528 	    sizeof (rl_t), offsetof(rl_t, r_node));
529 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
530 	    offsetof(zvol_extent_t, ze_node));
531 	/* get and cache the blocksize */
532 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
533 	ASSERT(error == 0);
534 	zv->zv_volblocksize = doi.doi_data_block_size;
535 
536 	if (spa_writeable(dmu_objset_spa(os))) {
537 		if (zil_replay_disable)
538 			zil_destroy(dmu_objset_zil(os), B_FALSE);
539 		else
540 			zil_replay(os, zv, zvol_replay_vector);
541 	}
542 	dmu_objset_disown(os, FTAG);
543 	zv->zv_objset = NULL;
544 
545 	zvol_minors++;
546 
547 	mutex_exit(&zfsdev_state_lock);
548 
549 	return (0);
550 }
551 
552 /*
553  * Remove minor node for the specified volume.
554  */
555 static int
556 zvol_remove_zv(zvol_state_t *zv)
557 {
558 	char nmbuf[20];
559 	minor_t minor = zv->zv_minor;
560 
561 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
562 	if (zv->zv_total_opens != 0)
563 		return (EBUSY);
564 
565 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
566 	ddi_remove_minor_node(zfs_dip, nmbuf);
567 
568 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
569 	ddi_remove_minor_node(zfs_dip, nmbuf);
570 
571 	avl_destroy(&zv->zv_znode.z_range_avl);
572 	mutex_destroy(&zv->zv_znode.z_range_lock);
573 
574 	kmem_free(zv, sizeof (zvol_state_t));
575 
576 	ddi_soft_state_free(zfsdev_state, minor);
577 
578 	zvol_minors--;
579 	return (0);
580 }
581 
582 int
583 zvol_remove_minor(const char *name)
584 {
585 	zvol_state_t *zv;
586 	int rc;
587 
588 	mutex_enter(&zfsdev_state_lock);
589 	if ((zv = zvol_minor_lookup(name)) == NULL) {
590 		mutex_exit(&zfsdev_state_lock);
591 		return (ENXIO);
592 	}
593 	rc = zvol_remove_zv(zv);
594 	mutex_exit(&zfsdev_state_lock);
595 	return (rc);
596 }
597 
598 int
599 zvol_first_open(zvol_state_t *zv)
600 {
601 	objset_t *os;
602 	uint64_t volsize;
603 	int error;
604 	uint64_t readonly;
605 
606 	/* lie and say we're read-only */
607 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
608 	    zvol_tag, &os);
609 	if (error)
610 		return (error);
611 
612 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
613 	if (error) {
614 		ASSERT(error == 0);
615 		dmu_objset_disown(os, zvol_tag);
616 		return (error);
617 	}
618 	zv->zv_objset = os;
619 	error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
620 	if (error) {
621 		dmu_objset_disown(os, zvol_tag);
622 		return (error);
623 	}
624 	zv->zv_volsize = volsize;
625 	zv->zv_zilog = zil_open(os, zvol_get_data);
626 	zvol_size_changed(zv->zv_volsize, ddi_driver_major(zfs_dip),
627 	    zv->zv_minor);
628 
629 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
630 	    NULL) == 0);
631 	if (readonly || dmu_objset_is_snapshot(os) ||
632 	    !spa_writeable(dmu_objset_spa(os)))
633 		zv->zv_flags |= ZVOL_RDONLY;
634 	else
635 		zv->zv_flags &= ~ZVOL_RDONLY;
636 	return (error);
637 }
638 
639 void
640 zvol_last_close(zvol_state_t *zv)
641 {
642 	zil_close(zv->zv_zilog);
643 	zv->zv_zilog = NULL;
644 	dmu_buf_rele(zv->zv_dbuf, zvol_tag);
645 	zv->zv_dbuf = NULL;
646 	dmu_objset_disown(zv->zv_objset, zvol_tag);
647 	zv->zv_objset = NULL;
648 }
649 
650 int
651 zvol_prealloc(zvol_state_t *zv)
652 {
653 	objset_t *os = zv->zv_objset;
654 	dmu_tx_t *tx;
655 	uint64_t refd, avail, usedobjs, availobjs;
656 	uint64_t resid = zv->zv_volsize;
657 	uint64_t off = 0;
658 
659 	/* Check the space usage before attempting to allocate the space */
660 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
661 	if (avail < zv->zv_volsize)
662 		return (ENOSPC);
663 
664 	/* Free old extents if they exist */
665 	zvol_free_extents(zv);
666 
667 	while (resid != 0) {
668 		int error;
669 		uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
670 
671 		tx = dmu_tx_create(os);
672 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
673 		error = dmu_tx_assign(tx, TXG_WAIT);
674 		if (error) {
675 			dmu_tx_abort(tx);
676 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
677 			return (error);
678 		}
679 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
680 		dmu_tx_commit(tx);
681 		off += bytes;
682 		resid -= bytes;
683 	}
684 	txg_wait_synced(dmu_objset_pool(os), 0);
685 
686 	return (0);
687 }
688 
689 int
690 zvol_update_volsize(objset_t *os, uint64_t volsize)
691 {
692 	dmu_tx_t *tx;
693 	int error;
694 
695 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
696 
697 	tx = dmu_tx_create(os);
698 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
699 	error = dmu_tx_assign(tx, TXG_WAIT);
700 	if (error) {
701 		dmu_tx_abort(tx);
702 		return (error);
703 	}
704 
705 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
706 	    &volsize, tx);
707 	dmu_tx_commit(tx);
708 
709 	if (error == 0)
710 		error = dmu_free_long_range(os,
711 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
712 	return (error);
713 }
714 
715 void
716 zvol_remove_minors(const char *name)
717 {
718 	zvol_state_t *zv;
719 	char *namebuf;
720 	minor_t minor;
721 
722 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
723 	(void) strncpy(namebuf, name, strlen(name));
724 	(void) strcat(namebuf, "/");
725 	mutex_enter(&zfsdev_state_lock);
726 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
727 
728 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
729 		if (zv == NULL)
730 			continue;
731 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
732 			(void) zvol_remove_zv(zv);
733 	}
734 	kmem_free(namebuf, strlen(name) + 2);
735 
736 	mutex_exit(&zfsdev_state_lock);
737 }
738 
739 int
740 zvol_set_volsize(const char *name, major_t maj, uint64_t volsize)
741 {
742 	zvol_state_t *zv = NULL;
743 	objset_t *os;
744 	int error;
745 	dmu_object_info_t doi;
746 	uint64_t old_volsize = 0ULL;
747 	uint64_t readonly;
748 
749 	mutex_enter(&zfsdev_state_lock);
750 	zv = zvol_minor_lookup(name);
751 	if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
752 		mutex_exit(&zfsdev_state_lock);
753 		return (error);
754 	}
755 
756 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
757 	    (error = zvol_check_volsize(volsize,
758 	    doi.doi_data_block_size)) != 0)
759 		goto out;
760 
761 	VERIFY(dsl_prop_get_integer(name, "readonly", &readonly,
762 	    NULL) == 0);
763 	if (readonly) {
764 		error = EROFS;
765 		goto out;
766 	}
767 
768 	error = zvol_update_volsize(os, volsize);
769 	/*
770 	 * Reinitialize the dump area to the new size. If we
771 	 * failed to resize the dump area then restore it back to
772 	 * its original size.
773 	 */
774 	if (zv && error == 0) {
775 		if (zv->zv_flags & ZVOL_DUMPIFIED) {
776 			old_volsize = zv->zv_volsize;
777 			zv->zv_volsize = volsize;
778 			if ((error = zvol_dumpify(zv)) != 0 ||
779 			    (error = dumpvp_resize()) != 0) {
780 				(void) zvol_update_volsize(os, old_volsize);
781 				zv->zv_volsize = old_volsize;
782 				error = zvol_dumpify(zv);
783 			}
784 		}
785 		if (error == 0) {
786 			zv->zv_volsize = volsize;
787 			zvol_size_changed(volsize, maj, zv->zv_minor);
788 		}
789 	}
790 
791 	/*
792 	 * Generate a LUN expansion event.
793 	 */
794 	if (zv && error == 0) {
795 		sysevent_id_t eid;
796 		nvlist_t *attr;
797 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
798 
799 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
800 		    zv->zv_minor);
801 
802 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
803 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
804 
805 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
806 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
807 
808 		nvlist_free(attr);
809 		kmem_free(physpath, MAXPATHLEN);
810 	}
811 
812 out:
813 	dmu_objset_rele(os, FTAG);
814 
815 	mutex_exit(&zfsdev_state_lock);
816 
817 	return (error);
818 }
819 
820 /*ARGSUSED*/
821 int
822 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
823 {
824 	zvol_state_t *zv;
825 	int err = 0;
826 
827 	mutex_enter(&zfsdev_state_lock);
828 
829 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
830 	if (zv == NULL) {
831 		mutex_exit(&zfsdev_state_lock);
832 		return (ENXIO);
833 	}
834 
835 	if (zv->zv_total_opens == 0)
836 		err = zvol_first_open(zv);
837 	if (err) {
838 		mutex_exit(&zfsdev_state_lock);
839 		return (err);
840 	}
841 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
842 		err = EROFS;
843 		goto out;
844 	}
845 	if (zv->zv_flags & ZVOL_EXCL) {
846 		err = EBUSY;
847 		goto out;
848 	}
849 	if (flag & FEXCL) {
850 		if (zv->zv_total_opens != 0) {
851 			err = EBUSY;
852 			goto out;
853 		}
854 		zv->zv_flags |= ZVOL_EXCL;
855 	}
856 
857 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
858 		zv->zv_open_count[otyp]++;
859 		zv->zv_total_opens++;
860 	}
861 	mutex_exit(&zfsdev_state_lock);
862 
863 	return (err);
864 out:
865 	if (zv->zv_total_opens == 0)
866 		zvol_last_close(zv);
867 	mutex_exit(&zfsdev_state_lock);
868 	return (err);
869 }
870 
871 /*ARGSUSED*/
872 int
873 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
874 {
875 	minor_t minor = getminor(dev);
876 	zvol_state_t *zv;
877 	int error = 0;
878 
879 	mutex_enter(&zfsdev_state_lock);
880 
881 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
882 	if (zv == NULL) {
883 		mutex_exit(&zfsdev_state_lock);
884 		return (ENXIO);
885 	}
886 
887 	if (zv->zv_flags & ZVOL_EXCL) {
888 		ASSERT(zv->zv_total_opens == 1);
889 		zv->zv_flags &= ~ZVOL_EXCL;
890 	}
891 
892 	/*
893 	 * If the open count is zero, this is a spurious close.
894 	 * That indicates a bug in the kernel / DDI framework.
895 	 */
896 	ASSERT(zv->zv_open_count[otyp] != 0);
897 	ASSERT(zv->zv_total_opens != 0);
898 
899 	/*
900 	 * You may get multiple opens, but only one close.
901 	 */
902 	zv->zv_open_count[otyp]--;
903 	zv->zv_total_opens--;
904 
905 	if (zv->zv_total_opens == 0)
906 		zvol_last_close(zv);
907 
908 	mutex_exit(&zfsdev_state_lock);
909 	return (error);
910 }
911 
912 static void
913 zvol_get_done(zgd_t *zgd, int error)
914 {
915 	if (zgd->zgd_db)
916 		dmu_buf_rele(zgd->zgd_db, zgd);
917 
918 	zfs_range_unlock(zgd->zgd_rl);
919 
920 	if (error == 0 && zgd->zgd_bp)
921 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
922 
923 	kmem_free(zgd, sizeof (zgd_t));
924 }
925 
926 /*
927  * Get data to generate a TX_WRITE intent log record.
928  */
929 static int
930 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
931 {
932 	zvol_state_t *zv = arg;
933 	objset_t *os = zv->zv_objset;
934 	uint64_t object = ZVOL_OBJ;
935 	uint64_t offset = lr->lr_offset;
936 	uint64_t size = lr->lr_length;	/* length of user data */
937 	blkptr_t *bp = &lr->lr_blkptr;
938 	dmu_buf_t *db;
939 	zgd_t *zgd;
940 	int error;
941 
942 	ASSERT(zio != NULL);
943 	ASSERT(size != 0);
944 
945 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
946 	zgd->zgd_zilog = zv->zv_zilog;
947 	zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
948 
949 	/*
950 	 * Write records come in two flavors: immediate and indirect.
951 	 * For small writes it's cheaper to store the data with the
952 	 * log record (immediate); for large writes it's cheaper to
953 	 * sync the data and get a pointer to it (indirect) so that
954 	 * we don't have to write the data twice.
955 	 */
956 	if (buf != NULL) {	/* immediate write */
957 		error = dmu_read(os, object, offset, size, buf,
958 		    DMU_READ_NO_PREFETCH);
959 	} else {
960 		size = zv->zv_volblocksize;
961 		offset = P2ALIGN(offset, size);
962 		error = dmu_buf_hold(os, object, offset, zgd, &db,
963 		    DMU_READ_NO_PREFETCH);
964 		if (error == 0) {
965 			zgd->zgd_db = db;
966 			zgd->zgd_bp = bp;
967 
968 			ASSERT(db->db_offset == offset);
969 			ASSERT(db->db_size == size);
970 
971 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
972 			    zvol_get_done, zgd);
973 
974 			if (error == 0)
975 				return (0);
976 		}
977 	}
978 
979 	zvol_get_done(zgd, error);
980 
981 	return (error);
982 }
983 
984 /*
985  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
986  *
987  * We store data in the log buffers if it's small enough.
988  * Otherwise we will later flush the data out via dmu_sync().
989  */
990 ssize_t zvol_immediate_write_sz = 32768;
991 
992 static void
993 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
994     boolean_t sync)
995 {
996 	uint32_t blocksize = zv->zv_volblocksize;
997 	zilog_t *zilog = zv->zv_zilog;
998 	boolean_t slogging;
999 	ssize_t immediate_write_sz;
1000 
1001 	if (zil_replaying(zilog, tx))
1002 		return;
1003 
1004 	immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1005 	    ? 0 : zvol_immediate_write_sz;
1006 
1007 	slogging = spa_has_slogs(zilog->zl_spa) &&
1008 	    (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1009 
1010 	while (resid) {
1011 		itx_t *itx;
1012 		lr_write_t *lr;
1013 		ssize_t len;
1014 		itx_wr_state_t write_state;
1015 
1016 		/*
1017 		 * Unlike zfs_log_write() we can be called with
1018 		 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1019 		 */
1020 		if (blocksize > immediate_write_sz && !slogging &&
1021 		    resid >= blocksize && off % blocksize == 0) {
1022 			write_state = WR_INDIRECT; /* uses dmu_sync */
1023 			len = blocksize;
1024 		} else if (sync) {
1025 			write_state = WR_COPIED;
1026 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1027 		} else {
1028 			write_state = WR_NEED_COPY;
1029 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1030 		}
1031 
1032 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1033 		    (write_state == WR_COPIED ? len : 0));
1034 		lr = (lr_write_t *)&itx->itx_lr;
1035 		if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1036 		    ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1037 			zil_itx_destroy(itx);
1038 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1039 			lr = (lr_write_t *)&itx->itx_lr;
1040 			write_state = WR_NEED_COPY;
1041 		}
1042 
1043 		itx->itx_wr_state = write_state;
1044 		if (write_state == WR_NEED_COPY)
1045 			itx->itx_sod += len;
1046 		lr->lr_foid = ZVOL_OBJ;
1047 		lr->lr_offset = off;
1048 		lr->lr_length = len;
1049 		lr->lr_blkoff = 0;
1050 		BP_ZERO(&lr->lr_blkptr);
1051 
1052 		itx->itx_private = zv;
1053 		itx->itx_sync = sync;
1054 
1055 		zil_itx_assign(zilog, itx, tx);
1056 
1057 		off += len;
1058 		resid -= len;
1059 	}
1060 }
1061 
1062 static int
1063 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size,
1064     boolean_t doread, boolean_t isdump)
1065 {
1066 	vdev_disk_t *dvd;
1067 	int c;
1068 	int numerrors = 0;
1069 
1070 	for (c = 0; c < vd->vdev_children; c++) {
1071 		ASSERT(vd->vdev_ops == &vdev_mirror_ops ||
1072 		    vd->vdev_ops == &vdev_replacing_ops ||
1073 		    vd->vdev_ops == &vdev_spare_ops);
1074 		int err = zvol_dumpio_vdev(vd->vdev_child[c],
1075 		    addr, offset, size, doread, isdump);
1076 		if (err != 0) {
1077 			numerrors++;
1078 		} else if (doread) {
1079 			break;
1080 		}
1081 	}
1082 
1083 	if (!vd->vdev_ops->vdev_op_leaf)
1084 		return (numerrors < vd->vdev_children ? 0 : EIO);
1085 
1086 	if (doread && !vdev_readable(vd))
1087 		return (EIO);
1088 	else if (!doread && !vdev_writeable(vd))
1089 		return (EIO);
1090 
1091 	dvd = vd->vdev_tsd;
1092 	ASSERT3P(dvd, !=, NULL);
1093 	offset += VDEV_LABEL_START_SIZE;
1094 
1095 	if (ddi_in_panic() || isdump) {
1096 		ASSERT(!doread);
1097 		if (doread)
1098 			return (EIO);
1099 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1100 		    lbtodb(size)));
1101 	} else {
1102 		return (vdev_disk_physio(dvd->vd_lh, addr, size, offset,
1103 		    doread ? B_READ : B_WRITE));
1104 	}
1105 }
1106 
1107 static int
1108 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1109     boolean_t doread, boolean_t isdump)
1110 {
1111 	vdev_t *vd;
1112 	int error;
1113 	zvol_extent_t *ze;
1114 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1115 
1116 	/* Must be sector aligned, and not stradle a block boundary. */
1117 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1118 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1119 		return (EINVAL);
1120 	}
1121 	ASSERT(size <= zv->zv_volblocksize);
1122 
1123 	/* Locate the extent this belongs to */
1124 	ze = list_head(&zv->zv_extents);
1125 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1126 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1127 		ze = list_next(&zv->zv_extents, ze);
1128 	}
1129 
1130 	if (!ddi_in_panic())
1131 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1132 
1133 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1134 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1135 	error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump);
1136 
1137 	if (!ddi_in_panic())
1138 		spa_config_exit(spa, SCL_STATE, FTAG);
1139 
1140 	return (error);
1141 }
1142 
1143 int
1144 zvol_strategy(buf_t *bp)
1145 {
1146 	zfs_soft_state_t *zs = NULL;
1147 	zvol_state_t *zv;
1148 	uint64_t off, volsize;
1149 	size_t resid;
1150 	char *addr;
1151 	objset_t *os;
1152 	rl_t *rl;
1153 	int error = 0;
1154 	boolean_t doread = bp->b_flags & B_READ;
1155 	boolean_t is_dump;
1156 	boolean_t sync;
1157 
1158 	if (getminor(bp->b_edev) == 0) {
1159 		error = EINVAL;
1160 	} else {
1161 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1162 		if (zs == NULL)
1163 			error = ENXIO;
1164 		else if (zs->zss_type != ZSST_ZVOL)
1165 			error = EINVAL;
1166 	}
1167 
1168 	if (error) {
1169 		bioerror(bp, error);
1170 		biodone(bp);
1171 		return (0);
1172 	}
1173 
1174 	zv = zs->zss_data;
1175 
1176 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1177 		bioerror(bp, EROFS);
1178 		biodone(bp);
1179 		return (0);
1180 	}
1181 
1182 	off = ldbtob(bp->b_blkno);
1183 	volsize = zv->zv_volsize;
1184 
1185 	os = zv->zv_objset;
1186 	ASSERT(os != NULL);
1187 
1188 	bp_mapin(bp);
1189 	addr = bp->b_un.b_addr;
1190 	resid = bp->b_bcount;
1191 
1192 	if (resid > 0 && (off < 0 || off >= volsize)) {
1193 		bioerror(bp, EIO);
1194 		biodone(bp);
1195 		return (0);
1196 	}
1197 
1198 	is_dump = zv->zv_flags & ZVOL_DUMPIFIED;
1199 	sync = ((!(bp->b_flags & B_ASYNC) &&
1200 	    !(zv->zv_flags & ZVOL_WCE)) ||
1201 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1202 	    !doread && !is_dump;
1203 
1204 	/*
1205 	 * There must be no buffer changes when doing a dmu_sync() because
1206 	 * we can't change the data whilst calculating the checksum.
1207 	 */
1208 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1209 	    doread ? RL_READER : RL_WRITER);
1210 
1211 	while (resid != 0 && off < volsize) {
1212 		size_t size = MIN(resid, zvol_maxphys);
1213 		if (is_dump) {
1214 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1215 			error = zvol_dumpio(zv, addr, off, size,
1216 			    doread, B_FALSE);
1217 		} else if (doread) {
1218 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1219 			    DMU_READ_PREFETCH);
1220 		} else {
1221 			dmu_tx_t *tx = dmu_tx_create(os);
1222 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1223 			error = dmu_tx_assign(tx, TXG_WAIT);
1224 			if (error) {
1225 				dmu_tx_abort(tx);
1226 			} else {
1227 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1228 				zvol_log_write(zv, tx, off, size, sync);
1229 				dmu_tx_commit(tx);
1230 			}
1231 		}
1232 		if (error) {
1233 			/* convert checksum errors into IO errors */
1234 			if (error == ECKSUM)
1235 				error = EIO;
1236 			break;
1237 		}
1238 		off += size;
1239 		addr += size;
1240 		resid -= size;
1241 	}
1242 	zfs_range_unlock(rl);
1243 
1244 	if ((bp->b_resid = resid) == bp->b_bcount)
1245 		bioerror(bp, off > volsize ? EINVAL : error);
1246 
1247 	if (sync)
1248 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1249 	biodone(bp);
1250 
1251 	return (0);
1252 }
1253 
1254 /*
1255  * Set the buffer count to the zvol maximum transfer.
1256  * Using our own routine instead of the default minphys()
1257  * means that for larger writes we write bigger buffers on X86
1258  * (128K instead of 56K) and flush the disk write cache less often
1259  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1260  * 56K on X86 and 128K on sparc).
1261  */
1262 void
1263 zvol_minphys(struct buf *bp)
1264 {
1265 	if (bp->b_bcount > zvol_maxphys)
1266 		bp->b_bcount = zvol_maxphys;
1267 }
1268 
1269 int
1270 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1271 {
1272 	minor_t minor = getminor(dev);
1273 	zvol_state_t *zv;
1274 	int error = 0;
1275 	uint64_t size;
1276 	uint64_t boff;
1277 	uint64_t resid;
1278 
1279 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1280 	if (zv == NULL)
1281 		return (ENXIO);
1282 
1283 	boff = ldbtob(blkno);
1284 	resid = ldbtob(nblocks);
1285 
1286 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1287 
1288 	while (resid) {
1289 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1290 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1291 		if (error)
1292 			break;
1293 		boff += size;
1294 		addr += size;
1295 		resid -= size;
1296 	}
1297 
1298 	return (error);
1299 }
1300 
1301 /*ARGSUSED*/
1302 int
1303 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1304 {
1305 	minor_t minor = getminor(dev);
1306 	zvol_state_t *zv;
1307 	uint64_t volsize;
1308 	rl_t *rl;
1309 	int error = 0;
1310 
1311 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1312 	if (zv == NULL)
1313 		return (ENXIO);
1314 
1315 	volsize = zv->zv_volsize;
1316 	if (uio->uio_resid > 0 &&
1317 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1318 		return (EIO);
1319 
1320 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1321 		error = physio(zvol_strategy, NULL, dev, B_READ,
1322 		    zvol_minphys, uio);
1323 		return (error);
1324 	}
1325 
1326 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1327 	    RL_READER);
1328 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1329 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1330 
1331 		/* don't read past the end */
1332 		if (bytes > volsize - uio->uio_loffset)
1333 			bytes = volsize - uio->uio_loffset;
1334 
1335 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1336 		if (error) {
1337 			/* convert checksum errors into IO errors */
1338 			if (error == ECKSUM)
1339 				error = EIO;
1340 			break;
1341 		}
1342 	}
1343 	zfs_range_unlock(rl);
1344 	return (error);
1345 }
1346 
1347 /*ARGSUSED*/
1348 int
1349 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1350 {
1351 	minor_t minor = getminor(dev);
1352 	zvol_state_t *zv;
1353 	uint64_t volsize;
1354 	rl_t *rl;
1355 	int error = 0;
1356 	boolean_t sync;
1357 
1358 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1359 	if (zv == NULL)
1360 		return (ENXIO);
1361 
1362 	volsize = zv->zv_volsize;
1363 	if (uio->uio_resid > 0 &&
1364 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1365 		return (EIO);
1366 
1367 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1368 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1369 		    zvol_minphys, uio);
1370 		return (error);
1371 	}
1372 
1373 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1374 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1375 
1376 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1377 	    RL_WRITER);
1378 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1379 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1380 		uint64_t off = uio->uio_loffset;
1381 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1382 
1383 		if (bytes > volsize - off)	/* don't write past the end */
1384 			bytes = volsize - off;
1385 
1386 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1387 		error = dmu_tx_assign(tx, TXG_WAIT);
1388 		if (error) {
1389 			dmu_tx_abort(tx);
1390 			break;
1391 		}
1392 		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1393 		if (error == 0)
1394 			zvol_log_write(zv, tx, off, bytes, sync);
1395 		dmu_tx_commit(tx);
1396 
1397 		if (error)
1398 			break;
1399 	}
1400 	zfs_range_unlock(rl);
1401 	if (sync)
1402 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1403 	return (error);
1404 }
1405 
1406 int
1407 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1408 {
1409 	struct uuid uuid = EFI_RESERVED;
1410 	efi_gpe_t gpe = { 0 };
1411 	uint32_t crc;
1412 	dk_efi_t efi;
1413 	int length;
1414 	char *ptr;
1415 
1416 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1417 		return (EFAULT);
1418 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1419 	length = efi.dki_length;
1420 	/*
1421 	 * Some clients may attempt to request a PMBR for the
1422 	 * zvol.  Currently this interface will return EINVAL to
1423 	 * such requests.  These requests could be supported by
1424 	 * adding a check for lba == 0 and consing up an appropriate
1425 	 * PMBR.
1426 	 */
1427 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1428 		return (EINVAL);
1429 
1430 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1431 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1432 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1433 
1434 	if (efi.dki_lba == 1) {
1435 		efi_gpt_t gpt = { 0 };
1436 
1437 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1438 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1439 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1440 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1441 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1442 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1443 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1444 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1445 		gpt.efi_gpt_SizeOfPartitionEntry =
1446 		    LE_32(sizeof (efi_gpe_t));
1447 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1448 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1449 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1450 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1451 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1452 		    flag))
1453 			return (EFAULT);
1454 		ptr += sizeof (gpt);
1455 		length -= sizeof (gpt);
1456 	}
1457 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1458 	    length), flag))
1459 		return (EFAULT);
1460 	return (0);
1461 }
1462 
1463 /*
1464  * BEGIN entry points to allow external callers access to the volume.
1465  */
1466 /*
1467  * Return the volume parameters needed for access from an external caller.
1468  * These values are invariant as long as the volume is held open.
1469  */
1470 int
1471 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1472     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1473     void **rl_hdl, void **bonus_hdl)
1474 {
1475 	zvol_state_t *zv;
1476 
1477 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1478 	if (zv == NULL)
1479 		return (ENXIO);
1480 	if (zv->zv_flags & ZVOL_DUMPIFIED)
1481 		return (ENXIO);
1482 
1483 	ASSERT(blksize && max_xfer_len && minor_hdl &&
1484 	    objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1485 
1486 	*blksize = zv->zv_volblocksize;
1487 	*max_xfer_len = (uint64_t)zvol_maxphys;
1488 	*minor_hdl = zv;
1489 	*objset_hdl = zv->zv_objset;
1490 	*zil_hdl = zv->zv_zilog;
1491 	*rl_hdl = &zv->zv_znode;
1492 	*bonus_hdl = zv->zv_dbuf;
1493 	return (0);
1494 }
1495 
1496 /*
1497  * Return the current volume size to an external caller.
1498  * The size can change while the volume is open.
1499  */
1500 uint64_t
1501 zvol_get_volume_size(void *minor_hdl)
1502 {
1503 	zvol_state_t *zv = minor_hdl;
1504 
1505 	return (zv->zv_volsize);
1506 }
1507 
1508 /*
1509  * Return the current WCE setting to an external caller.
1510  * The WCE setting can change while the volume is open.
1511  */
1512 int
1513 zvol_get_volume_wce(void *minor_hdl)
1514 {
1515 	zvol_state_t *zv = minor_hdl;
1516 
1517 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1518 }
1519 
1520 /*
1521  * Entry point for external callers to zvol_log_write
1522  */
1523 void
1524 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1525     boolean_t sync)
1526 {
1527 	zvol_state_t *zv = minor_hdl;
1528 
1529 	zvol_log_write(zv, tx, off, resid, sync);
1530 }
1531 /*
1532  * END entry points to allow external callers access to the volume.
1533  */
1534 
1535 /*
1536  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1537  */
1538 static void
1539 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1540     boolean_t sync)
1541 {
1542 	itx_t *itx;
1543 	lr_truncate_t *lr;
1544 	zilog_t *zilog = zv->zv_zilog;
1545 
1546 	if (zil_replaying(zilog, tx))
1547 		return;
1548 
1549 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1550 	lr = (lr_truncate_t *)&itx->itx_lr;
1551 	lr->lr_foid = ZVOL_OBJ;
1552 	lr->lr_offset = off;
1553 	lr->lr_length = len;
1554 
1555 	itx->itx_sync = sync;
1556 	zil_itx_assign(zilog, itx, tx);
1557 }
1558 
1559 /*
1560  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1561  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1562  */
1563 /*ARGSUSED*/
1564 int
1565 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1566 {
1567 	zvol_state_t *zv;
1568 	struct dk_cinfo dki;
1569 	struct dk_minfo dkm;
1570 	struct dk_callback *dkc;
1571 	int error = 0;
1572 	rl_t *rl;
1573 
1574 	mutex_enter(&zfsdev_state_lock);
1575 
1576 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1577 
1578 	if (zv == NULL) {
1579 		mutex_exit(&zfsdev_state_lock);
1580 		return (ENXIO);
1581 	}
1582 	ASSERT(zv->zv_total_opens > 0);
1583 
1584 	switch (cmd) {
1585 
1586 	case DKIOCINFO:
1587 		bzero(&dki, sizeof (dki));
1588 		(void) strcpy(dki.dki_cname, "zvol");
1589 		(void) strcpy(dki.dki_dname, "zvol");
1590 		dki.dki_ctype = DKC_UNKNOWN;
1591 		dki.dki_unit = getminor(dev);
1592 		dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1593 		mutex_exit(&zfsdev_state_lock);
1594 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1595 			error = EFAULT;
1596 		return (error);
1597 
1598 	case DKIOCGMEDIAINFO:
1599 		bzero(&dkm, sizeof (dkm));
1600 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1601 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1602 		dkm.dki_media_type = DK_UNKNOWN;
1603 		mutex_exit(&zfsdev_state_lock);
1604 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1605 			error = EFAULT;
1606 		return (error);
1607 
1608 	case DKIOCGETEFI:
1609 		{
1610 			uint64_t vs = zv->zv_volsize;
1611 			uint8_t bs = zv->zv_min_bs;
1612 
1613 			mutex_exit(&zfsdev_state_lock);
1614 			error = zvol_getefi((void *)arg, flag, vs, bs);
1615 			return (error);
1616 		}
1617 
1618 	case DKIOCFLUSHWRITECACHE:
1619 		dkc = (struct dk_callback *)arg;
1620 		mutex_exit(&zfsdev_state_lock);
1621 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1622 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1623 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1624 			error = 0;
1625 		}
1626 		return (error);
1627 
1628 	case DKIOCGETWCE:
1629 		{
1630 			int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1631 			if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1632 			    flag))
1633 				error = EFAULT;
1634 			break;
1635 		}
1636 	case DKIOCSETWCE:
1637 		{
1638 			int wce;
1639 			if (ddi_copyin((void *)arg, &wce, sizeof (int),
1640 			    flag)) {
1641 				error = EFAULT;
1642 				break;
1643 			}
1644 			if (wce) {
1645 				zv->zv_flags |= ZVOL_WCE;
1646 				mutex_exit(&zfsdev_state_lock);
1647 			} else {
1648 				zv->zv_flags &= ~ZVOL_WCE;
1649 				mutex_exit(&zfsdev_state_lock);
1650 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1651 			}
1652 			return (0);
1653 		}
1654 
1655 	case DKIOCGGEOM:
1656 	case DKIOCGVTOC:
1657 		/*
1658 		 * commands using these (like prtvtoc) expect ENOTSUP
1659 		 * since we're emulating an EFI label
1660 		 */
1661 		error = ENOTSUP;
1662 		break;
1663 
1664 	case DKIOCDUMPINIT:
1665 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1666 		    RL_WRITER);
1667 		error = zvol_dumpify(zv);
1668 		zfs_range_unlock(rl);
1669 		break;
1670 
1671 	case DKIOCDUMPFINI:
1672 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1673 			break;
1674 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1675 		    RL_WRITER);
1676 		error = zvol_dump_fini(zv);
1677 		zfs_range_unlock(rl);
1678 		break;
1679 
1680 	case DKIOCFREE:
1681 	{
1682 		dkioc_free_t df;
1683 		dmu_tx_t *tx;
1684 
1685 		if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1686 			error = EFAULT;
1687 			break;
1688 		}
1689 
1690 		/*
1691 		 * Apply Postel's Law to length-checking.  If they overshoot,
1692 		 * just blank out until the end, if there's a need to blank
1693 		 * out anything.
1694 		 */
1695 		if (df.df_start >= zv->zv_volsize)
1696 			break;	/* No need to do anything... */
1697 		if (df.df_start + df.df_length > zv->zv_volsize)
1698 			df.df_length = DMU_OBJECT_END;
1699 
1700 		rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1701 		    RL_WRITER);
1702 		tx = dmu_tx_create(zv->zv_objset);
1703 		error = dmu_tx_assign(tx, TXG_WAIT);
1704 		if (error != 0) {
1705 			dmu_tx_abort(tx);
1706 		} else {
1707 			zvol_log_truncate(zv, tx, df.df_start,
1708 			    df.df_length, B_TRUE);
1709 			dmu_tx_commit(tx);
1710 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1711 			    df.df_start, df.df_length);
1712 		}
1713 
1714 		zfs_range_unlock(rl);
1715 
1716 		if (error == 0) {
1717 			/*
1718 			 * If the write-cache is disabled or 'sync' property
1719 			 * is set to 'always' then treat this as a synchronous
1720 			 * operation (i.e. commit to zil).
1721 			 */
1722 			if (!(zv->zv_flags & ZVOL_WCE) ||
1723 			    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
1724 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1725 
1726 			/*
1727 			 * If the caller really wants synchronous writes, and
1728 			 * can't wait for them, don't return until the write
1729 			 * is done.
1730 			 */
1731 			if (df.df_flags & DF_WAIT_SYNC) {
1732 				txg_wait_synced(
1733 				    dmu_objset_pool(zv->zv_objset), 0);
1734 			}
1735 		}
1736 		break;
1737 	}
1738 
1739 	default:
1740 		error = ENOTTY;
1741 		break;
1742 
1743 	}
1744 	mutex_exit(&zfsdev_state_lock);
1745 	return (error);
1746 }
1747 
1748 int
1749 zvol_busy(void)
1750 {
1751 	return (zvol_minors != 0);
1752 }
1753 
1754 void
1755 zvol_init(void)
1756 {
1757 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1758 	    1) == 0);
1759 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1760 }
1761 
1762 void
1763 zvol_fini(void)
1764 {
1765 	mutex_destroy(&zfsdev_state_lock);
1766 	ddi_soft_state_fini(&zfsdev_state);
1767 }
1768 
1769 static int
1770 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1771 {
1772 	dmu_tx_t *tx;
1773 	int error = 0;
1774 	objset_t *os = zv->zv_objset;
1775 	nvlist_t *nv = NULL;
1776 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1777 
1778 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1779 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1780 	    DMU_OBJECT_END);
1781 	/* wait for dmu_free_long_range to actually free the blocks */
1782 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1783 
1784 	tx = dmu_tx_create(os);
1785 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1786 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1787 	error = dmu_tx_assign(tx, TXG_WAIT);
1788 	if (error) {
1789 		dmu_tx_abort(tx);
1790 		return (error);
1791 	}
1792 
1793 	/*
1794 	 * If we are resizing the dump device then we only need to
1795 	 * update the refreservation to match the newly updated
1796 	 * zvolsize. Otherwise, we save off the original state of the
1797 	 * zvol so that we can restore them if the zvol is ever undumpified.
1798 	 */
1799 	if (resize) {
1800 		error = zap_update(os, ZVOL_ZAP_OBJ,
1801 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1802 		    &zv->zv_volsize, tx);
1803 	} else {
1804 		uint64_t checksum, compress, refresrv, vbs, dedup;
1805 
1806 		error = dsl_prop_get_integer(zv->zv_name,
1807 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1808 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1809 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1810 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1811 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1812 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1813 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1814 		if (version >= SPA_VERSION_DEDUP) {
1815 			error = error ? error :
1816 			    dsl_prop_get_integer(zv->zv_name,
1817 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1818 		}
1819 
1820 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1821 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1822 		    &compress, tx);
1823 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1824 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1825 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1826 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1827 		    &refresrv, tx);
1828 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1829 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1830 		    &vbs, tx);
1831 		error = error ? error : dmu_object_set_blocksize(
1832 		    os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1833 		if (version >= SPA_VERSION_DEDUP) {
1834 			error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1835 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1836 			    &dedup, tx);
1837 		}
1838 		if (error == 0)
1839 			zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1840 	}
1841 	dmu_tx_commit(tx);
1842 
1843 	/*
1844 	 * We only need update the zvol's property if we are initializing
1845 	 * the dump area for the first time.
1846 	 */
1847 	if (!resize) {
1848 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1849 		VERIFY(nvlist_add_uint64(nv,
1850 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1851 		VERIFY(nvlist_add_uint64(nv,
1852 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1853 		    ZIO_COMPRESS_OFF) == 0);
1854 		VERIFY(nvlist_add_uint64(nv,
1855 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1856 		    ZIO_CHECKSUM_OFF) == 0);
1857 		if (version >= SPA_VERSION_DEDUP) {
1858 			VERIFY(nvlist_add_uint64(nv,
1859 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
1860 			    ZIO_CHECKSUM_OFF) == 0);
1861 		}
1862 
1863 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1864 		    nv, NULL);
1865 		nvlist_free(nv);
1866 
1867 		if (error)
1868 			return (error);
1869 	}
1870 
1871 	/* Allocate the space for the dump */
1872 	error = zvol_prealloc(zv);
1873 	return (error);
1874 }
1875 
1876 static int
1877 zvol_dumpify(zvol_state_t *zv)
1878 {
1879 	int error = 0;
1880 	uint64_t dumpsize = 0;
1881 	dmu_tx_t *tx;
1882 	objset_t *os = zv->zv_objset;
1883 
1884 	if (zv->zv_flags & ZVOL_RDONLY)
1885 		return (EROFS);
1886 
1887 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
1888 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
1889 		boolean_t resize = (dumpsize > 0);
1890 
1891 		if ((error = zvol_dump_init(zv, resize)) != 0) {
1892 			(void) zvol_dump_fini(zv);
1893 			return (error);
1894 		}
1895 	}
1896 
1897 	/*
1898 	 * Build up our lba mapping.
1899 	 */
1900 	error = zvol_get_lbas(zv);
1901 	if (error) {
1902 		(void) zvol_dump_fini(zv);
1903 		return (error);
1904 	}
1905 
1906 	tx = dmu_tx_create(os);
1907 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1908 	error = dmu_tx_assign(tx, TXG_WAIT);
1909 	if (error) {
1910 		dmu_tx_abort(tx);
1911 		(void) zvol_dump_fini(zv);
1912 		return (error);
1913 	}
1914 
1915 	zv->zv_flags |= ZVOL_DUMPIFIED;
1916 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
1917 	    &zv->zv_volsize, tx);
1918 	dmu_tx_commit(tx);
1919 
1920 	if (error) {
1921 		(void) zvol_dump_fini(zv);
1922 		return (error);
1923 	}
1924 
1925 	txg_wait_synced(dmu_objset_pool(os), 0);
1926 	return (0);
1927 }
1928 
1929 static int
1930 zvol_dump_fini(zvol_state_t *zv)
1931 {
1932 	dmu_tx_t *tx;
1933 	objset_t *os = zv->zv_objset;
1934 	nvlist_t *nv;
1935 	int error = 0;
1936 	uint64_t checksum, compress, refresrv, vbs, dedup;
1937 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1938 
1939 	/*
1940 	 * Attempt to restore the zvol back to its pre-dumpified state.
1941 	 * This is a best-effort attempt as it's possible that not all
1942 	 * of these properties were initialized during the dumpify process
1943 	 * (i.e. error during zvol_dump_init).
1944 	 */
1945 
1946 	tx = dmu_tx_create(os);
1947 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1948 	error = dmu_tx_assign(tx, TXG_WAIT);
1949 	if (error) {
1950 		dmu_tx_abort(tx);
1951 		return (error);
1952 	}
1953 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
1954 	dmu_tx_commit(tx);
1955 
1956 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1957 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
1958 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1959 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
1960 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1961 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
1962 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1963 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
1964 
1965 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1966 	(void) nvlist_add_uint64(nv,
1967 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
1968 	(void) nvlist_add_uint64(nv,
1969 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
1970 	(void) nvlist_add_uint64(nv,
1971 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
1972 	if (version >= SPA_VERSION_DEDUP &&
1973 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1974 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
1975 		(void) nvlist_add_uint64(nv,
1976 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
1977 	}
1978 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1979 	    nv, NULL);
1980 	nvlist_free(nv);
1981 
1982 	zvol_free_extents(zv);
1983 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
1984 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
1985 	/* wait for dmu_free_long_range to actually free the blocks */
1986 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1987 	tx = dmu_tx_create(os);
1988 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1989 	error = dmu_tx_assign(tx, TXG_WAIT);
1990 	if (error) {
1991 		dmu_tx_abort(tx);
1992 		return (error);
1993 	}
1994 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
1995 		zv->zv_volblocksize = vbs;
1996 	dmu_tx_commit(tx);
1997 
1998 	return (0);
1999 }
2000