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  * Copyright (c) 2006-2010 Pawel Jakub Dawidek <[email protected]>
25  * All rights reserved.
26  *
27  * Portions Copyright 2010 Robert Milkowski
28  *
29  * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
30  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
31  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32  * Copyright (c) 2014 Integros [integros.com]
33  */
34 
35 /* Portions Copyright 2011 Martin Matuska <[email protected]> */
36 
37 /*
38  * ZFS volume emulation driver.
39  *
40  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
41  * Volumes are accessed through the symbolic links named:
42  *
43  * /dev/zvol/dsk/<pool_name>/<dataset_name>
44  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
45  *
46  * These links are created by the /dev filesystem (sdev_zvolops.c).
47  * Volumes are persistent through reboot.  No user command needs to be
48  * run before opening and using a device.
49  *
50  * FreeBSD notes.
51  * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
52  * in the system.
53  */
54 
55 #include <sys/types.h>
56 #include <sys/param.h>
57 #include <sys/kernel.h>
58 #include <sys/errno.h>
59 #include <sys/uio.h>
60 #include <sys/bio.h>
61 #include <sys/buf.h>
62 #include <sys/kmem.h>
63 #include <sys/conf.h>
64 #include <sys/cmn_err.h>
65 #include <sys/stat.h>
66 #include <sys/zap.h>
67 #include <sys/spa.h>
68 #include <sys/spa_impl.h>
69 #include <sys/zio.h>
70 #include <sys/disk.h>
71 #include <sys/dmu_traverse.h>
72 #include <sys/dnode.h>
73 #include <sys/dsl_dataset.h>
74 #include <sys/dsl_prop.h>
75 #include <sys/dkio.h>
76 #include <sys/byteorder.h>
77 #include <sys/sunddi.h>
78 #include <sys/dirent.h>
79 #include <sys/policy.h>
80 #include <sys/queue.h>
81 #include <sys/fs/zfs.h>
82 #include <sys/zfs_ioctl.h>
83 #include <sys/zil.h>
84 #include <sys/refcount.h>
85 #include <sys/zfs_znode.h>
86 #include <sys/zfs_rlock.h>
87 #include <sys/vdev_impl.h>
88 #include <sys/vdev_raidz.h>
89 #include <sys/zvol.h>
90 #include <sys/zil_impl.h>
91 #include <sys/dbuf.h>
92 #include <sys/dmu_tx.h>
93 #include <sys/zfeature.h>
94 #include <sys/zio_checksum.h>
95 #include <sys/zil_impl.h>
96 #include <sys/filio.h>
97 
98 #include <geom/geom.h>
99 
100 #include "zfs_namecheck.h"
101 
102 #ifndef illumos
103 struct g_class zfs_zvol_class = {
104 	.name = "ZFS::ZVOL",
105 	.version = G_VERSION,
106 };
107 
108 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
109 
110 #endif
111 void *zfsdev_state;
112 static char *zvol_tag = "zvol_tag";
113 
114 #define	ZVOL_DUMPSIZE		"dumpsize"
115 
116 /*
117  * This lock protects the zfsdev_state structure from being modified
118  * while it's being used, e.g. an open that comes in before a create
119  * finishes.  It also protects temporary opens of the dataset so that,
120  * e.g., an open doesn't get a spurious EBUSY.
121  */
122 #ifdef illumos
123 kmutex_t zfsdev_state_lock;
124 #else
125 /*
126  * In FreeBSD we've replaced the upstream zfsdev_state_lock with the
127  * spa_namespace_lock in the ZVOL code.
128  */
129 #define zfsdev_state_lock spa_namespace_lock
130 #endif
131 static uint32_t zvol_minors;
132 
133 #ifndef illumos
134 SYSCTL_DECL(_vfs_zfs);
135 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
136 static int	volmode = ZFS_VOLMODE_GEOM;
137 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
138     "Expose as GEOM providers (1), device files (2) or neither");
139 static boolean_t zpool_on_zvol = B_FALSE;
140 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, recursive, CTLFLAG_RWTUN, &zpool_on_zvol, 0,
141     "Allow zpools to use zvols as vdevs (DANGEROUS)");
142 
143 #endif
144 typedef struct zvol_extent {
145 	list_node_t	ze_node;
146 	dva_t		ze_dva;		/* dva associated with this extent */
147 	uint64_t	ze_nblks;	/* number of blocks in extent */
148 } zvol_extent_t;
149 
150 /*
151  * The in-core state of each volume.
152  */
153 typedef struct zvol_state {
154 #ifndef illumos
155 	LIST_ENTRY(zvol_state)	zv_links;
156 #endif
157 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
158 	uint64_t	zv_volsize;	/* amount of space we advertise */
159 	uint64_t	zv_volblocksize; /* volume block size */
160 #ifdef illumos
161 	minor_t		zv_minor;	/* minor number */
162 #else
163 	struct cdev	*zv_dev;	/* non-GEOM device */
164 	struct g_provider *zv_provider;	/* GEOM provider */
165 #endif
166 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
167 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
168 	objset_t	*zv_objset;	/* objset handle */
169 #ifdef illumos
170 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
171 #endif
172 	uint32_t	zv_total_opens;	/* total open count */
173 	uint32_t	zv_sync_cnt;	/* synchronous open count */
174 	zilog_t		*zv_zilog;	/* ZIL handle */
175 	list_t		zv_extents;	/* List of extents for dump */
176 	znode_t		zv_znode;	/* for range locking */
177 	dnode_t		*zv_dn;		/* dnode hold */
178 #ifndef illumos
179 	int		zv_state;
180 	int		zv_volmode;	/* Provide GEOM or cdev */
181 	struct bio_queue_head zv_queue;
182 	struct mtx	zv_queue_mtx;	/* zv_queue mutex */
183 #endif
184 } zvol_state_t;
185 
186 #ifndef illumos
187 static LIST_HEAD(, zvol_state) all_zvols;
188 #endif
189 /*
190  * zvol specific flags
191  */
192 #define	ZVOL_RDONLY	0x1
193 #define	ZVOL_DUMPIFIED	0x2
194 #define	ZVOL_EXCL	0x4
195 #define	ZVOL_WCE	0x8
196 
197 /*
198  * zvol maximum transfer in one DMU tx.
199  */
200 int zvol_maxphys = DMU_MAX_ACCESS/2;
201 
202 /*
203  * Toggle unmap functionality.
204  */
205 boolean_t zvol_unmap_enabled = B_TRUE;
206 
207 /*
208  * If true, unmaps requested as synchronous are executed synchronously,
209  * otherwise all unmaps are asynchronous.
210  */
211 boolean_t zvol_unmap_sync_enabled = B_FALSE;
212 
213 #ifndef illumos
214 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
215     &zvol_unmap_enabled, 0,
216     "Enable UNMAP functionality");
217 
218 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_sync_enabled, CTLFLAG_RWTUN,
219     &zvol_unmap_sync_enabled, 0,
220     "UNMAPs requested as sync are executed synchronously");
221 
222 static d_open_t		zvol_d_open;
223 static d_close_t	zvol_d_close;
224 static d_read_t		zvol_read;
225 static d_write_t	zvol_write;
226 static d_ioctl_t	zvol_d_ioctl;
227 static d_strategy_t	zvol_strategy;
228 
229 static struct cdevsw zvol_cdevsw = {
230 	.d_version =	D_VERSION,
231 	.d_open =	zvol_d_open,
232 	.d_close =	zvol_d_close,
233 	.d_read =	zvol_read,
234 	.d_write =	zvol_write,
235 	.d_ioctl =	zvol_d_ioctl,
236 	.d_strategy =	zvol_strategy,
237 	.d_name =	"zvol",
238 	.d_flags =	D_DISK | D_TRACKCLOSE,
239 };
240 
241 static void zvol_geom_run(zvol_state_t *zv);
242 static void zvol_geom_destroy(zvol_state_t *zv);
243 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
244 static void zvol_geom_start(struct bio *bp);
245 static void zvol_geom_worker(void *arg);
246 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
247     uint64_t len, boolean_t sync);
248 #endif	/* !illumos */
249 
250 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
251     nvlist_t *, nvlist_t *);
252 static int zvol_remove_zv(zvol_state_t *);
253 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
254     struct lwb *lwb, zio_t *zio);
255 static int zvol_dumpify(zvol_state_t *zv);
256 static int zvol_dump_fini(zvol_state_t *zv);
257 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
258 
259 static void
zvol_size_changed(zvol_state_t * zv,uint64_t volsize)260 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
261 {
262 #ifdef illumos
263 	dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
264 
265 	zv->zv_volsize = volsize;
266 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
267 	    "Size", volsize) == DDI_SUCCESS);
268 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
269 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
270 
271 	/* Notify specfs to invalidate the cached size */
272 	spec_size_invalidate(dev, VBLK);
273 	spec_size_invalidate(dev, VCHR);
274 #else	/* !illumos */
275 	zv->zv_volsize = volsize;
276 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
277 		struct g_provider *pp;
278 
279 		pp = zv->zv_provider;
280 		if (pp == NULL)
281 			return;
282 		g_topology_lock();
283 
284 		/*
285 		 * Do not invoke resize event when initial size was zero.
286 		 * ZVOL initializes the size on first open, this is not
287 		 * real resizing.
288 		 */
289 		if (pp->mediasize == 0)
290 			pp->mediasize = zv->zv_volsize;
291 		else
292 			g_resize_provider(pp, zv->zv_volsize);
293 		g_topology_unlock();
294 	}
295 #endif	/* illumos */
296 }
297 
298 int
zvol_check_volsize(uint64_t volsize,uint64_t blocksize)299 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
300 {
301 	if (volsize == 0)
302 		return (SET_ERROR(EINVAL));
303 
304 	if (volsize % blocksize != 0)
305 		return (SET_ERROR(EINVAL));
306 
307 #ifdef _ILP32
308 	if (volsize - 1 > SPEC_MAXOFFSET_T)
309 		return (SET_ERROR(EOVERFLOW));
310 #endif
311 	return (0);
312 }
313 
314 int
zvol_check_volblocksize(uint64_t volblocksize)315 zvol_check_volblocksize(uint64_t volblocksize)
316 {
317 	if (volblocksize < SPA_MINBLOCKSIZE ||
318 	    volblocksize > SPA_OLD_MAXBLOCKSIZE ||
319 	    !ISP2(volblocksize))
320 		return (SET_ERROR(EDOM));
321 
322 	return (0);
323 }
324 
325 int
zvol_get_stats(objset_t * os,nvlist_t * nv)326 zvol_get_stats(objset_t *os, nvlist_t *nv)
327 {
328 	int error;
329 	dmu_object_info_t doi;
330 	uint64_t val;
331 
332 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
333 	if (error)
334 		return (error);
335 
336 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
337 
338 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
339 
340 	if (error == 0) {
341 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
342 		    doi.doi_data_block_size);
343 	}
344 
345 	return (error);
346 }
347 
348 static zvol_state_t *
zvol_minor_lookup(const char * name)349 zvol_minor_lookup(const char *name)
350 {
351 #ifdef illumos
352 	minor_t minor;
353 #endif
354 	zvol_state_t *zv;
355 
356 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
357 
358 #ifdef illumos
359 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
360 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
361 		if (zv == NULL)
362 			continue;
363 #else
364 	LIST_FOREACH(zv, &all_zvols, zv_links) {
365 #endif
366 		if (strcmp(zv->zv_name, name) == 0)
367 			return (zv);
368 	}
369 
370 	return (NULL);
371 }
372 
373 /* extent mapping arg */
374 struct maparg {
375 	zvol_state_t	*ma_zv;
376 	uint64_t	ma_blks;
377 };
378 
379 /*ARGSUSED*/
380 static int
381 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
382     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
383 {
384 	struct maparg *ma = arg;
385 	zvol_extent_t *ze;
386 	int bs = ma->ma_zv->zv_volblocksize;
387 
388 	if (bp == NULL || BP_IS_HOLE(bp) ||
389 	    zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
390 		return (0);
391 
392 	VERIFY(!BP_IS_EMBEDDED(bp));
393 
394 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
395 	ma->ma_blks++;
396 
397 	/* Abort immediately if we have encountered gang blocks */
398 	if (BP_IS_GANG(bp))
399 		return (SET_ERROR(EFRAGS));
400 
401 	/*
402 	 * See if the block is at the end of the previous extent.
403 	 */
404 	ze = list_tail(&ma->ma_zv->zv_extents);
405 	if (ze &&
406 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
407 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
408 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
409 		ze->ze_nblks++;
410 		return (0);
411 	}
412 
413 	dprintf_bp(bp, "%s", "next blkptr:");
414 
415 	/* start a new extent */
416 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
417 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
418 	ze->ze_nblks = 1;
419 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
420 	return (0);
421 }
422 
423 static void
424 zvol_free_extents(zvol_state_t *zv)
425 {
426 	zvol_extent_t *ze;
427 
428 	while (ze = list_head(&zv->zv_extents)) {
429 		list_remove(&zv->zv_extents, ze);
430 		kmem_free(ze, sizeof (zvol_extent_t));
431 	}
432 }
433 
434 static int
435 zvol_get_lbas(zvol_state_t *zv)
436 {
437 	objset_t *os = zv->zv_objset;
438 	struct maparg	ma;
439 	int		err;
440 
441 	ma.ma_zv = zv;
442 	ma.ma_blks = 0;
443 	zvol_free_extents(zv);
444 
445 	/* commit any in-flight changes before traversing the dataset */
446 	txg_wait_synced(dmu_objset_pool(os), 0);
447 	err = traverse_dataset(dmu_objset_ds(os), 0,
448 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
449 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
450 		zvol_free_extents(zv);
451 		return (err ? err : EIO);
452 	}
453 
454 	return (0);
455 }
456 
457 /* ARGSUSED */
458 void
459 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
460 {
461 	zfs_creat_t *zct = arg;
462 	nvlist_t *nvprops = zct->zct_props;
463 	int error;
464 	uint64_t volblocksize, volsize;
465 
466 	VERIFY(nvlist_lookup_uint64(nvprops,
467 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
468 	if (nvlist_lookup_uint64(nvprops,
469 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
470 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
471 
472 	/*
473 	 * These properties must be removed from the list so the generic
474 	 * property setting step won't apply to them.
475 	 */
476 	VERIFY(nvlist_remove_all(nvprops,
477 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
478 	(void) nvlist_remove_all(nvprops,
479 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
480 
481 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
482 	    DMU_OT_NONE, 0, tx);
483 	ASSERT(error == 0);
484 
485 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
486 	    DMU_OT_NONE, 0, tx);
487 	ASSERT(error == 0);
488 
489 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
490 	ASSERT(error == 0);
491 }
492 
493 /*
494  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
495  * implement DKIOCFREE/free-long-range.
496  */
497 static int
498 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
499 {
500 	zvol_state_t *zv = arg1;
501 	lr_truncate_t *lr = arg2;
502 	uint64_t offset, length;
503 
504 	if (byteswap)
505 		byteswap_uint64_array(lr, sizeof (*lr));
506 
507 	offset = lr->lr_offset;
508 	length = lr->lr_length;
509 
510 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
511 }
512 
513 /*
514  * Replay a TX_WRITE ZIL transaction that didn't get committed
515  * after a system failure
516  */
517 static int
518 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
519 {
520 	zvol_state_t *zv = arg1;
521 	lr_write_t *lr = arg2;
522 	objset_t *os = zv->zv_objset;
523 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
524 	uint64_t offset, length;
525 	dmu_tx_t *tx;
526 	int error;
527 
528 	if (byteswap)
529 		byteswap_uint64_array(lr, sizeof (*lr));
530 
531 	offset = lr->lr_offset;
532 	length = lr->lr_length;
533 
534 	/* If it's a dmu_sync() block, write the whole block */
535 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
536 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
537 		if (length < blocksize) {
538 			offset -= offset % blocksize;
539 			length = blocksize;
540 		}
541 	}
542 
543 	tx = dmu_tx_create(os);
544 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
545 	error = dmu_tx_assign(tx, TXG_WAIT);
546 	if (error) {
547 		dmu_tx_abort(tx);
548 	} else {
549 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
550 		dmu_tx_commit(tx);
551 	}
552 
553 	return (error);
554 }
555 
556 /* ARGSUSED */
557 static int
558 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
559 {
560 	return (SET_ERROR(ENOTSUP));
561 }
562 
563 /*
564  * Callback vectors for replaying records.
565  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
566  */
567 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
568 	zvol_replay_err,	/* 0 no such transaction type */
569 	zvol_replay_err,	/* TX_CREATE */
570 	zvol_replay_err,	/* TX_MKDIR */
571 	zvol_replay_err,	/* TX_MKXATTR */
572 	zvol_replay_err,	/* TX_SYMLINK */
573 	zvol_replay_err,	/* TX_REMOVE */
574 	zvol_replay_err,	/* TX_RMDIR */
575 	zvol_replay_err,	/* TX_LINK */
576 	zvol_replay_err,	/* TX_RENAME */
577 	zvol_replay_write,	/* TX_WRITE */
578 	zvol_replay_truncate,	/* TX_TRUNCATE */
579 	zvol_replay_err,	/* TX_SETATTR */
580 	zvol_replay_err,	/* TX_ACL */
581 	zvol_replay_err,	/* TX_CREATE_ACL */
582 	zvol_replay_err,	/* TX_CREATE_ATTR */
583 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
584 	zvol_replay_err,	/* TX_MKDIR_ACL */
585 	zvol_replay_err,	/* TX_MKDIR_ATTR */
586 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
587 	zvol_replay_err,	/* TX_WRITE2 */
588 };
589 
590 #ifdef illumos
591 int
592 zvol_name2minor(const char *name, minor_t *minor)
593 {
594 	zvol_state_t *zv;
595 
596 	mutex_enter(&zfsdev_state_lock);
597 	zv = zvol_minor_lookup(name);
598 	if (minor && zv)
599 		*minor = zv->zv_minor;
600 	mutex_exit(&zfsdev_state_lock);
601 	return (zv ? 0 : -1);
602 }
603 #endif	/* illumos */
604 
605 /*
606  * Create a minor node (plus a whole lot more) for the specified volume.
607  */
608 int
609 zvol_create_minor(const char *name)
610 {
611 	zfs_soft_state_t *zs;
612 	zvol_state_t *zv;
613 	objset_t *os;
614 #ifdef illumos
615 	dmu_object_info_t doi;
616 	minor_t minor = 0;
617 	char chrbuf[30], blkbuf[30];
618 #else
619 	struct g_provider *pp;
620 	struct g_geom *gp;
621 	uint64_t mode;
622 #endif
623 	int error;
624 
625 #ifndef illumos
626 	ZFS_LOG(1, "Creating ZVOL %s...", name);
627 #endif
628 
629 	mutex_enter(&zfsdev_state_lock);
630 
631 	if (zvol_minor_lookup(name) != NULL) {
632 		mutex_exit(&zfsdev_state_lock);
633 		return (SET_ERROR(EEXIST));
634 	}
635 
636 	/* lie and say we're read-only */
637 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
638 
639 	if (error) {
640 		mutex_exit(&zfsdev_state_lock);
641 		return (error);
642 	}
643 
644 #ifdef illumos
645 	if ((minor = zfsdev_minor_alloc()) == 0) {
646 		dmu_objset_disown(os, FTAG);
647 		mutex_exit(&zfsdev_state_lock);
648 		return (SET_ERROR(ENXIO));
649 	}
650 
651 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
652 		dmu_objset_disown(os, FTAG);
653 		mutex_exit(&zfsdev_state_lock);
654 		return (SET_ERROR(EAGAIN));
655 	}
656 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
657 	    (char *)name);
658 
659 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
660 
661 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
662 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
663 		ddi_soft_state_free(zfsdev_state, minor);
664 		dmu_objset_disown(os, FTAG);
665 		mutex_exit(&zfsdev_state_lock);
666 		return (SET_ERROR(EAGAIN));
667 	}
668 
669 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
670 
671 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
672 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
673 		ddi_remove_minor_node(zfs_dip, chrbuf);
674 		ddi_soft_state_free(zfsdev_state, minor);
675 		dmu_objset_disown(os, FTAG);
676 		mutex_exit(&zfsdev_state_lock);
677 		return (SET_ERROR(EAGAIN));
678 	}
679 
680 	zs = ddi_get_soft_state(zfsdev_state, minor);
681 	zs->zss_type = ZSST_ZVOL;
682 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
683 #else	/* !illumos */
684 
685 	zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
686 	zv->zv_state = 0;
687 	error = dsl_prop_get_integer(name,
688 	    zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
689 	if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
690 		mode = volmode;
691 
692 	DROP_GIANT();
693 	zv->zv_volmode = mode;
694 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
695 		g_topology_lock();
696 		gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
697 		gp->start = zvol_geom_start;
698 		gp->access = zvol_geom_access;
699 		pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
700 		pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
701 		pp->sectorsize = DEV_BSIZE;
702 		pp->mediasize = 0;
703 		pp->private = zv;
704 
705 		zv->zv_provider = pp;
706 		bioq_init(&zv->zv_queue);
707 		mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
708 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
709 		struct make_dev_args args;
710 
711 		make_dev_args_init(&args);
712 		args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
713 		args.mda_devsw = &zvol_cdevsw;
714 		args.mda_cr = NULL;
715 		args.mda_uid = UID_ROOT;
716 		args.mda_gid = GID_OPERATOR;
717 		args.mda_mode = 0640;
718 		args.mda_si_drv2 = zv;
719 		error = make_dev_s(&args, &zv->zv_dev,
720 		    "%s/%s", ZVOL_DRIVER, name);
721 		if (error != 0) {
722 			kmem_free(zv, sizeof(*zv));
723 			dmu_objset_disown(os, FTAG);
724 			mutex_exit(&zfsdev_state_lock);
725 			return (error);
726 		}
727 		zv->zv_dev->si_iosize_max = MAXPHYS;
728 	}
729 	LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
730 #endif	/* illumos */
731 
732 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
733 	zv->zv_min_bs = DEV_BSHIFT;
734 #ifdef illumos
735 	zv->zv_minor = minor;
736 #endif
737 	zv->zv_objset = os;
738 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
739 		zv->zv_flags |= ZVOL_RDONLY;
740 	mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
741 	avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
742 	    sizeof (rl_t), offsetof(rl_t, r_node));
743 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
744 	    offsetof(zvol_extent_t, ze_node));
745 #ifdef illumos
746 	/* get and cache the blocksize */
747 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
748 	ASSERT(error == 0);
749 	zv->zv_volblocksize = doi.doi_data_block_size;
750 #endif
751 
752 	if (spa_writeable(dmu_objset_spa(os))) {
753 		if (zil_replay_disable)
754 			zil_destroy(dmu_objset_zil(os), B_FALSE);
755 		else
756 			zil_replay(os, zv, zvol_replay_vector);
757 	}
758 	dmu_objset_disown(os, FTAG);
759 	zv->zv_objset = NULL;
760 
761 	zvol_minors++;
762 
763 	mutex_exit(&zfsdev_state_lock);
764 #ifndef illumos
765 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
766 		zvol_geom_run(zv);
767 		g_topology_unlock();
768 	}
769 	PICKUP_GIANT();
770 
771 	ZFS_LOG(1, "ZVOL %s created.", name);
772 #endif
773 
774 	return (0);
775 }
776 
777 /*
778  * Remove minor node for the specified volume.
779  */
780 static int
781 zvol_remove_zv(zvol_state_t *zv)
782 {
783 #ifdef illumos
784 	char nmbuf[20];
785 	minor_t minor = zv->zv_minor;
786 #endif
787 
788 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
789 	if (zv->zv_total_opens != 0)
790 		return (SET_ERROR(EBUSY));
791 
792 #ifdef illumos
793 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
794 	ddi_remove_minor_node(zfs_dip, nmbuf);
795 
796 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
797 	ddi_remove_minor_node(zfs_dip, nmbuf);
798 #else
799 	ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
800 
801 	LIST_REMOVE(zv, zv_links);
802 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
803 		g_topology_lock();
804 		zvol_geom_destroy(zv);
805 		g_topology_unlock();
806 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
807 		if (zv->zv_dev != NULL)
808 			destroy_dev(zv->zv_dev);
809 	}
810 #endif
811 
812 	avl_destroy(&zv->zv_znode.z_range_avl);
813 	mutex_destroy(&zv->zv_znode.z_range_lock);
814 
815 	kmem_free(zv, sizeof (zvol_state_t));
816 #ifdef illumos
817 	ddi_soft_state_free(zfsdev_state, minor);
818 #endif
819 	zvol_minors--;
820 	return (0);
821 }
822 
823 int
824 zvol_remove_minor(const char *name)
825 {
826 	zvol_state_t *zv;
827 	int rc;
828 
829 	mutex_enter(&zfsdev_state_lock);
830 	if ((zv = zvol_minor_lookup(name)) == NULL) {
831 		mutex_exit(&zfsdev_state_lock);
832 		return (SET_ERROR(ENXIO));
833 	}
834 	rc = zvol_remove_zv(zv);
835 	mutex_exit(&zfsdev_state_lock);
836 	return (rc);
837 }
838 
839 int
840 zvol_first_open(zvol_state_t *zv)
841 {
842 	dmu_object_info_t doi;
843 	objset_t *os;
844 	uint64_t volsize;
845 	int error;
846 	uint64_t readonly;
847 
848 	/* lie and say we're read-only */
849 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
850 	    zvol_tag, &os);
851 	if (error)
852 		return (error);
853 
854 	zv->zv_objset = os;
855 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
856 	if (error) {
857 		ASSERT(error == 0);
858 		dmu_objset_disown(os, zvol_tag);
859 		return (error);
860 	}
861 
862 	/* get and cache the blocksize */
863 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
864 	if (error) {
865 		ASSERT(error == 0);
866 		dmu_objset_disown(os, zvol_tag);
867 		return (error);
868 	}
869 	zv->zv_volblocksize = doi.doi_data_block_size;
870 
871 	error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
872 	if (error) {
873 		dmu_objset_disown(os, zvol_tag);
874 		return (error);
875 	}
876 
877 	zvol_size_changed(zv, volsize);
878 	zv->zv_zilog = zil_open(os, zvol_get_data);
879 
880 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
881 	    NULL) == 0);
882 	if (readonly || dmu_objset_is_snapshot(os) ||
883 	    !spa_writeable(dmu_objset_spa(os)))
884 		zv->zv_flags |= ZVOL_RDONLY;
885 	else
886 		zv->zv_flags &= ~ZVOL_RDONLY;
887 	return (error);
888 }
889 
890 void
891 zvol_last_close(zvol_state_t *zv)
892 {
893 	zil_close(zv->zv_zilog);
894 	zv->zv_zilog = NULL;
895 
896 	dnode_rele(zv->zv_dn, zvol_tag);
897 	zv->zv_dn = NULL;
898 
899 	/*
900 	 * Evict cached data
901 	 */
902 	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
903 	    !(zv->zv_flags & ZVOL_RDONLY))
904 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
905 	dmu_objset_evict_dbufs(zv->zv_objset);
906 
907 	dmu_objset_disown(zv->zv_objset, zvol_tag);
908 	zv->zv_objset = NULL;
909 }
910 
911 #ifdef illumos
912 int
913 zvol_prealloc(zvol_state_t *zv)
914 {
915 	objset_t *os = zv->zv_objset;
916 	dmu_tx_t *tx;
917 	uint64_t refd, avail, usedobjs, availobjs;
918 	uint64_t resid = zv->zv_volsize;
919 	uint64_t off = 0;
920 
921 	/* Check the space usage before attempting to allocate the space */
922 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
923 	if (avail < zv->zv_volsize)
924 		return (SET_ERROR(ENOSPC));
925 
926 	/* Free old extents if they exist */
927 	zvol_free_extents(zv);
928 
929 	while (resid != 0) {
930 		int error;
931 		uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
932 
933 		tx = dmu_tx_create(os);
934 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
935 		error = dmu_tx_assign(tx, TXG_WAIT);
936 		if (error) {
937 			dmu_tx_abort(tx);
938 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
939 			return (error);
940 		}
941 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
942 		dmu_tx_commit(tx);
943 		off += bytes;
944 		resid -= bytes;
945 	}
946 	txg_wait_synced(dmu_objset_pool(os), 0);
947 
948 	return (0);
949 }
950 #endif	/* illumos */
951 
952 static int
953 zvol_update_volsize(objset_t *os, uint64_t volsize)
954 {
955 	dmu_tx_t *tx;
956 	int error;
957 
958 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
959 
960 	tx = dmu_tx_create(os);
961 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
962 	dmu_tx_mark_netfree(tx);
963 	error = dmu_tx_assign(tx, TXG_WAIT);
964 	if (error) {
965 		dmu_tx_abort(tx);
966 		return (error);
967 	}
968 
969 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
970 	    &volsize, tx);
971 	dmu_tx_commit(tx);
972 
973 	if (error == 0)
974 		error = dmu_free_long_range(os,
975 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
976 	return (error);
977 }
978 
979 void
980 zvol_remove_minors(const char *name)
981 {
982 #ifdef illumos
983 	zvol_state_t *zv;
984 	char *namebuf;
985 	minor_t minor;
986 
987 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
988 	(void) strncpy(namebuf, name, strlen(name));
989 	(void) strcat(namebuf, "/");
990 	mutex_enter(&zfsdev_state_lock);
991 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
992 
993 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
994 		if (zv == NULL)
995 			continue;
996 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
997 			(void) zvol_remove_zv(zv);
998 	}
999 	kmem_free(namebuf, strlen(name) + 2);
1000 
1001 	mutex_exit(&zfsdev_state_lock);
1002 #else	/* !illumos */
1003 	zvol_state_t *zv, *tzv;
1004 	size_t namelen;
1005 
1006 	namelen = strlen(name);
1007 
1008 	DROP_GIANT();
1009 	mutex_enter(&zfsdev_state_lock);
1010 
1011 	LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
1012 		if (strcmp(zv->zv_name, name) == 0 ||
1013 		    (strncmp(zv->zv_name, name, namelen) == 0 &&
1014 		    strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
1015 		    zv->zv_name[namelen] == '@'))) {
1016 			(void) zvol_remove_zv(zv);
1017 		}
1018 	}
1019 
1020 	mutex_exit(&zfsdev_state_lock);
1021 	PICKUP_GIANT();
1022 #endif	/* illumos */
1023 }
1024 
1025 static int
1026 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
1027 {
1028 	uint64_t old_volsize = 0ULL;
1029 	int error = 0;
1030 
1031 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1032 
1033 	/*
1034 	 * Reinitialize the dump area to the new size. If we
1035 	 * failed to resize the dump area then restore it back to
1036 	 * its original size.  We must set the new volsize prior
1037 	 * to calling dumpvp_resize() to ensure that the devices'
1038 	 * size(9P) is not visible by the dump subsystem.
1039 	 */
1040 	old_volsize = zv->zv_volsize;
1041 	zvol_size_changed(zv, volsize);
1042 
1043 #ifdef ZVOL_DUMP
1044 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1045 		if ((error = zvol_dumpify(zv)) != 0 ||
1046 		    (error = dumpvp_resize()) != 0) {
1047 			int dumpify_error;
1048 
1049 			(void) zvol_update_volsize(zv->zv_objset, old_volsize);
1050 			zvol_size_changed(zv, old_volsize);
1051 			dumpify_error = zvol_dumpify(zv);
1052 			error = dumpify_error ? dumpify_error : error;
1053 		}
1054 	}
1055 #endif	/* ZVOL_DUMP */
1056 
1057 #ifdef illumos
1058 	/*
1059 	 * Generate a LUN expansion event.
1060 	 */
1061 	if (error == 0) {
1062 		sysevent_id_t eid;
1063 		nvlist_t *attr;
1064 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1065 
1066 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1067 		    zv->zv_minor);
1068 
1069 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1070 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1071 
1072 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1073 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1074 
1075 		nvlist_free(attr);
1076 		kmem_free(physpath, MAXPATHLEN);
1077 	}
1078 #endif	/* illumos */
1079 	return (error);
1080 }
1081 
1082 int
1083 zvol_set_volsize(const char *name, uint64_t volsize)
1084 {
1085 	zvol_state_t *zv = NULL;
1086 	objset_t *os;
1087 	int error;
1088 	dmu_object_info_t doi;
1089 	uint64_t readonly;
1090 	boolean_t owned = B_FALSE;
1091 
1092 	error = dsl_prop_get_integer(name,
1093 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1094 	if (error != 0)
1095 		return (error);
1096 	if (readonly)
1097 		return (SET_ERROR(EROFS));
1098 
1099 	mutex_enter(&zfsdev_state_lock);
1100 	zv = zvol_minor_lookup(name);
1101 
1102 	if (zv == NULL || zv->zv_objset == NULL) {
1103 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1104 		    FTAG, &os)) != 0) {
1105 			mutex_exit(&zfsdev_state_lock);
1106 			return (error);
1107 		}
1108 		owned = B_TRUE;
1109 		if (zv != NULL)
1110 			zv->zv_objset = os;
1111 	} else {
1112 		os = zv->zv_objset;
1113 	}
1114 
1115 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1116 	    (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1117 		goto out;
1118 
1119 	error = zvol_update_volsize(os, volsize);
1120 
1121 	if (error == 0 && zv != NULL)
1122 		error = zvol_update_live_volsize(zv, volsize);
1123 out:
1124 	if (owned) {
1125 		dmu_objset_disown(os, FTAG);
1126 		if (zv != NULL)
1127 			zv->zv_objset = NULL;
1128 	}
1129 	mutex_exit(&zfsdev_state_lock);
1130 	return (error);
1131 }
1132 
1133 /*ARGSUSED*/
1134 #ifdef illumos
1135 int
1136 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1137 #else
1138 static int
1139 zvol_open(struct g_provider *pp, int flag, int count)
1140 #endif
1141 {
1142 	zvol_state_t *zv;
1143 	int err = 0;
1144 #ifdef illumos
1145 
1146 	mutex_enter(&zfsdev_state_lock);
1147 
1148 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1149 	if (zv == NULL) {
1150 		mutex_exit(&zfsdev_state_lock);
1151 		return (SET_ERROR(ENXIO));
1152 	}
1153 
1154 	if (zv->zv_total_opens == 0)
1155 		err = zvol_first_open(zv);
1156 	if (err) {
1157 		mutex_exit(&zfsdev_state_lock);
1158 		return (err);
1159 	}
1160 #else	/* !illumos */
1161 	boolean_t locked = B_FALSE;
1162 
1163 	if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1164 		/*
1165 		 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1166 		 * attempting to probe geom providers while looking for a
1167 		 * replacement for a missing VDEV.  In this case, the
1168 		 * spa_namespace_lock will not be held, but it is still illegal
1169 		 * to use a zvol as a vdev.  Deadlocks can result if another
1170 		 * thread has spa_namespace_lock
1171 		 */
1172 		return (EOPNOTSUPP);
1173 	}
1174 	/*
1175 	 * Protect against recursively entering spa_namespace_lock
1176 	 * when spa_open() is used for a pool on a (local) ZVOL(s).
1177 	 * This is needed since we replaced upstream zfsdev_state_lock
1178 	 * with spa_namespace_lock in the ZVOL code.
1179 	 * We are using the same trick as spa_open().
1180 	 * Note that calls in zvol_first_open which need to resolve
1181 	 * pool name to a spa object will enter spa_open()
1182 	 * recursively, but that function already has all the
1183 	 * necessary protection.
1184 	 */
1185 	if (!MUTEX_HELD(&zfsdev_state_lock)) {
1186 		mutex_enter(&zfsdev_state_lock);
1187 		locked = B_TRUE;
1188 	}
1189 
1190 	zv = pp->private;
1191 	if (zv == NULL) {
1192 		if (locked)
1193 			mutex_exit(&zfsdev_state_lock);
1194 		return (SET_ERROR(ENXIO));
1195 	}
1196 
1197 	if (zv->zv_total_opens == 0) {
1198 		err = zvol_first_open(zv);
1199 		if (err) {
1200 			if (locked)
1201 				mutex_exit(&zfsdev_state_lock);
1202 			return (err);
1203 		}
1204 		pp->mediasize = zv->zv_volsize;
1205 		pp->stripeoffset = 0;
1206 		pp->stripesize = zv->zv_volblocksize;
1207 	}
1208 #endif	/* illumos */
1209 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1210 		err = SET_ERROR(EROFS);
1211 		goto out;
1212 	}
1213 	if (zv->zv_flags & ZVOL_EXCL) {
1214 		err = SET_ERROR(EBUSY);
1215 		goto out;
1216 	}
1217 #ifdef FEXCL
1218 	if (flag & FEXCL) {
1219 		if (zv->zv_total_opens != 0) {
1220 			err = SET_ERROR(EBUSY);
1221 			goto out;
1222 		}
1223 		zv->zv_flags |= ZVOL_EXCL;
1224 	}
1225 #endif
1226 
1227 #ifdef illumos
1228 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1229 		zv->zv_open_count[otyp]++;
1230 		zv->zv_total_opens++;
1231 	}
1232 	mutex_exit(&zfsdev_state_lock);
1233 #else
1234 	zv->zv_total_opens += count;
1235 	if (locked)
1236 		mutex_exit(&zfsdev_state_lock);
1237 #endif
1238 
1239 	return (err);
1240 out:
1241 	if (zv->zv_total_opens == 0)
1242 		zvol_last_close(zv);
1243 #ifdef illumos
1244 	mutex_exit(&zfsdev_state_lock);
1245 #else
1246 	if (locked)
1247 		mutex_exit(&zfsdev_state_lock);
1248 #endif
1249 	return (err);
1250 }
1251 
1252 /*ARGSUSED*/
1253 #ifdef illumos
1254 int
1255 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1256 {
1257 	minor_t minor = getminor(dev);
1258 	zvol_state_t *zv;
1259 	int error = 0;
1260 
1261 	mutex_enter(&zfsdev_state_lock);
1262 
1263 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1264 	if (zv == NULL) {
1265 		mutex_exit(&zfsdev_state_lock);
1266 #else	/* !illumos */
1267 static int
1268 zvol_close(struct g_provider *pp, int flag, int count)
1269 {
1270 	zvol_state_t *zv;
1271 	int error = 0;
1272 	boolean_t locked = B_FALSE;
1273 
1274 	/* See comment in zvol_open(). */
1275 	if (!MUTEX_HELD(&zfsdev_state_lock)) {
1276 		mutex_enter(&zfsdev_state_lock);
1277 		locked = B_TRUE;
1278 	}
1279 
1280 	zv = pp->private;
1281 	if (zv == NULL) {
1282 		if (locked)
1283 			mutex_exit(&zfsdev_state_lock);
1284 #endif	/* illumos */
1285 		return (SET_ERROR(ENXIO));
1286 	}
1287 
1288 	if (zv->zv_flags & ZVOL_EXCL) {
1289 		ASSERT(zv->zv_total_opens == 1);
1290 		zv->zv_flags &= ~ZVOL_EXCL;
1291 	}
1292 
1293 	/*
1294 	 * If the open count is zero, this is a spurious close.
1295 	 * That indicates a bug in the kernel / DDI framework.
1296 	 */
1297 #ifdef illumos
1298 	ASSERT(zv->zv_open_count[otyp] != 0);
1299 #endif
1300 	ASSERT(zv->zv_total_opens != 0);
1301 
1302 	/*
1303 	 * You may get multiple opens, but only one close.
1304 	 */
1305 #ifdef illumos
1306 	zv->zv_open_count[otyp]--;
1307 	zv->zv_total_opens--;
1308 #else
1309 	zv->zv_total_opens -= count;
1310 #endif
1311 
1312 	if (zv->zv_total_opens == 0)
1313 		zvol_last_close(zv);
1314 
1315 #ifdef illumos
1316 	mutex_exit(&zfsdev_state_lock);
1317 #else
1318 	if (locked)
1319 		mutex_exit(&zfsdev_state_lock);
1320 #endif
1321 	return (error);
1322 }
1323 
1324 /* ARGSUSED */
1325 static void
1326 zvol_get_done(zgd_t *zgd, int error)
1327 {
1328 	if (zgd->zgd_db)
1329 		dmu_buf_rele(zgd->zgd_db, zgd);
1330 
1331 	zfs_range_unlock(zgd->zgd_rl);
1332 
1333 	kmem_free(zgd, sizeof (zgd_t));
1334 }
1335 
1336 /*
1337  * Get data to generate a TX_WRITE intent log record.
1338  */
1339 static int
1340 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1341 {
1342 	zvol_state_t *zv = arg;
1343 	uint64_t offset = lr->lr_offset;
1344 	uint64_t size = lr->lr_length;	/* length of user data */
1345 	dmu_buf_t *db;
1346 	zgd_t *zgd;
1347 	int error;
1348 
1349 	ASSERT3P(lwb, !=, NULL);
1350 	ASSERT3P(zio, !=, NULL);
1351 	ASSERT3U(size, !=, 0);
1352 
1353 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1354 	zgd->zgd_lwb = lwb;
1355 
1356 	/*
1357 	 * Write records come in two flavors: immediate and indirect.
1358 	 * For small writes it's cheaper to store the data with the
1359 	 * log record (immediate); for large writes it's cheaper to
1360 	 * sync the data and get a pointer to it (indirect) so that
1361 	 * we don't have to write the data twice.
1362 	 */
1363 	if (buf != NULL) { /* immediate write */
1364 		zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1365 		    RL_READER);
1366 		error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1367 		    DMU_READ_NO_PREFETCH);
1368 	} else { /* indirect write */
1369 		/*
1370 		 * Have to lock the whole block to ensure when it's written out
1371 		 * and its checksum is being calculated that no one can change
1372 		 * the data. Contrarily to zfs_get_data we need not re-check
1373 		 * blocksize after we get the lock because it cannot be changed.
1374 		 */
1375 		size = zv->zv_volblocksize;
1376 		offset = P2ALIGN(offset, size);
1377 		zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1378 		    RL_READER);
1379 		error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1380 		    DMU_READ_NO_PREFETCH);
1381 		if (error == 0) {
1382 			blkptr_t *bp = &lr->lr_blkptr;
1383 
1384 			zgd->zgd_db = db;
1385 			zgd->zgd_bp = bp;
1386 
1387 			ASSERT(db->db_offset == offset);
1388 			ASSERT(db->db_size == size);
1389 
1390 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1391 			    zvol_get_done, zgd);
1392 
1393 			if (error == 0)
1394 				return (0);
1395 		}
1396 	}
1397 
1398 	zvol_get_done(zgd, error);
1399 
1400 	return (error);
1401 }
1402 
1403 /*
1404  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1405  *
1406  * We store data in the log buffers if it's small enough.
1407  * Otherwise we will later flush the data out via dmu_sync().
1408  */
1409 ssize_t zvol_immediate_write_sz = 32768;
1410 #ifdef _KERNEL
1411 SYSCTL_LONG(_vfs_zfs_vol, OID_AUTO, immediate_write_sz, CTLFLAG_RWTUN,
1412     &zvol_immediate_write_sz, 0, "Minimal size for indirect log write");
1413 #endif
1414 
1415 static void
1416 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1417     boolean_t sync)
1418 {
1419 	uint32_t blocksize = zv->zv_volblocksize;
1420 	zilog_t *zilog = zv->zv_zilog;
1421 	itx_wr_state_t write_state;
1422 
1423 	if (zil_replaying(zilog, tx))
1424 		return;
1425 
1426 	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1427 		write_state = WR_INDIRECT;
1428 	else if (!spa_has_slogs(zilog->zl_spa) &&
1429 	    resid >= blocksize && blocksize > zvol_immediate_write_sz)
1430 		write_state = WR_INDIRECT;
1431 	else if (sync)
1432 		write_state = WR_COPIED;
1433 	else
1434 		write_state = WR_NEED_COPY;
1435 
1436 	while (resid) {
1437 		itx_t *itx;
1438 		lr_write_t *lr;
1439 		itx_wr_state_t wr_state = write_state;
1440 		ssize_t len = resid;
1441 
1442 		if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1443 			wr_state = WR_NEED_COPY;
1444 		else if (wr_state == WR_INDIRECT)
1445 			len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1446 
1447 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1448 		    (wr_state == WR_COPIED ? len : 0));
1449 		lr = (lr_write_t *)&itx->itx_lr;
1450 		if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1451 		    off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1452 			zil_itx_destroy(itx);
1453 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1454 			lr = (lr_write_t *)&itx->itx_lr;
1455 			wr_state = WR_NEED_COPY;
1456 		}
1457 
1458 		itx->itx_wr_state = wr_state;
1459 		lr->lr_foid = ZVOL_OBJ;
1460 		lr->lr_offset = off;
1461 		lr->lr_length = len;
1462 		lr->lr_blkoff = 0;
1463 		BP_ZERO(&lr->lr_blkptr);
1464 
1465 		itx->itx_private = zv;
1466 
1467 		if (!sync && (zv->zv_sync_cnt == 0))
1468 			itx->itx_sync = B_FALSE;
1469 
1470 		zil_itx_assign(zilog, itx, tx);
1471 
1472 		off += len;
1473 		resid -= len;
1474 	}
1475 }
1476 
1477 #ifdef illumos
1478 static int
1479 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1480     uint64_t size, boolean_t doread, boolean_t isdump)
1481 {
1482 	vdev_disk_t *dvd;
1483 	int c;
1484 	int numerrors = 0;
1485 
1486 	if (vd->vdev_ops == &vdev_mirror_ops ||
1487 	    vd->vdev_ops == &vdev_replacing_ops ||
1488 	    vd->vdev_ops == &vdev_spare_ops) {
1489 		for (c = 0; c < vd->vdev_children; c++) {
1490 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1491 			    addr, offset, origoffset, size, doread, isdump);
1492 			if (err != 0) {
1493 				numerrors++;
1494 			} else if (doread) {
1495 				break;
1496 			}
1497 		}
1498 	}
1499 
1500 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1501 		return (numerrors < vd->vdev_children ? 0 : EIO);
1502 
1503 	if (doread && !vdev_readable(vd))
1504 		return (SET_ERROR(EIO));
1505 	else if (!doread && !vdev_writeable(vd))
1506 		return (SET_ERROR(EIO));
1507 
1508 	if (vd->vdev_ops == &vdev_raidz_ops) {
1509 		return (vdev_raidz_physio(vd,
1510 		    addr, size, offset, origoffset, doread, isdump));
1511 	}
1512 
1513 	offset += VDEV_LABEL_START_SIZE;
1514 
1515 	if (ddi_in_panic() || isdump) {
1516 		ASSERT(!doread);
1517 		if (doread)
1518 			return (SET_ERROR(EIO));
1519 		dvd = vd->vdev_tsd;
1520 		ASSERT3P(dvd, !=, NULL);
1521 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1522 		    lbtodb(size)));
1523 	} else {
1524 		dvd = vd->vdev_tsd;
1525 		ASSERT3P(dvd, !=, NULL);
1526 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1527 		    offset, doread ? B_READ : B_WRITE));
1528 	}
1529 }
1530 
1531 static int
1532 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1533     boolean_t doread, boolean_t isdump)
1534 {
1535 	vdev_t *vd;
1536 	int error;
1537 	zvol_extent_t *ze;
1538 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1539 
1540 	/* Must be sector aligned, and not stradle a block boundary. */
1541 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1542 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1543 		return (SET_ERROR(EINVAL));
1544 	}
1545 	ASSERT(size <= zv->zv_volblocksize);
1546 
1547 	/* Locate the extent this belongs to */
1548 	ze = list_head(&zv->zv_extents);
1549 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1550 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1551 		ze = list_next(&zv->zv_extents, ze);
1552 	}
1553 
1554 	if (ze == NULL)
1555 		return (SET_ERROR(EINVAL));
1556 
1557 	if (!ddi_in_panic())
1558 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1559 
1560 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1561 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1562 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1563 	    size, doread, isdump);
1564 
1565 	if (!ddi_in_panic())
1566 		spa_config_exit(spa, SCL_STATE, FTAG);
1567 
1568 	return (error);
1569 }
1570 
1571 int
1572 zvol_strategy(buf_t *bp)
1573 {
1574 	zfs_soft_state_t *zs = NULL;
1575 #else	/* !illumos */
1576 void
1577 zvol_strategy(struct bio *bp)
1578 {
1579 #endif	/* illumos */
1580 	zvol_state_t *zv;
1581 	uint64_t off, volsize;
1582 	size_t resid;
1583 	char *addr;
1584 	objset_t *os;
1585 	rl_t *rl;
1586 	int error = 0;
1587 #ifdef illumos
1588 	boolean_t doread = bp->b_flags & B_READ;
1589 #else
1590 	boolean_t doread = 0;
1591 #endif
1592 	boolean_t is_dumpified;
1593 	boolean_t sync;
1594 
1595 #ifdef illumos
1596 	if (getminor(bp->b_edev) == 0) {
1597 		error = SET_ERROR(EINVAL);
1598 	} else {
1599 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1600 		if (zs == NULL)
1601 			error = SET_ERROR(ENXIO);
1602 		else if (zs->zss_type != ZSST_ZVOL)
1603 			error = SET_ERROR(EINVAL);
1604 	}
1605 
1606 	if (error) {
1607 		bioerror(bp, error);
1608 		biodone(bp);
1609 		return (0);
1610 	}
1611 
1612 	zv = zs->zss_data;
1613 
1614 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1615 		bioerror(bp, EROFS);
1616 		biodone(bp);
1617 		return (0);
1618 	}
1619 
1620 	off = ldbtob(bp->b_blkno);
1621 #else	/* !illumos */
1622 	if (bp->bio_to)
1623 		zv = bp->bio_to->private;
1624 	else
1625 		zv = bp->bio_dev->si_drv2;
1626 
1627 	if (zv == NULL) {
1628 		error = SET_ERROR(ENXIO);
1629 		goto out;
1630 	}
1631 
1632 	if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1633 		error = SET_ERROR(EROFS);
1634 		goto out;
1635 	}
1636 
1637 	switch (bp->bio_cmd) {
1638 	case BIO_FLUSH:
1639 		goto sync;
1640 	case BIO_READ:
1641 		doread = 1;
1642 	case BIO_WRITE:
1643 	case BIO_DELETE:
1644 		break;
1645 	default:
1646 		error = EOPNOTSUPP;
1647 		goto out;
1648 	}
1649 
1650 	off = bp->bio_offset;
1651 #endif	/* illumos */
1652 	volsize = zv->zv_volsize;
1653 
1654 	os = zv->zv_objset;
1655 	ASSERT(os != NULL);
1656 
1657 #ifdef illumos
1658 	bp_mapin(bp);
1659 	addr = bp->b_un.b_addr;
1660 	resid = bp->b_bcount;
1661 
1662 	if (resid > 0 && (off < 0 || off >= volsize)) {
1663 		bioerror(bp, EIO);
1664 		biodone(bp);
1665 		return (0);
1666 	}
1667 
1668 	is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1669 	sync = ((!(bp->b_flags & B_ASYNC) &&
1670 	    !(zv->zv_flags & ZVOL_WCE)) ||
1671 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1672 	    !doread && !is_dumpified;
1673 #else	/* !illumos */
1674 	addr = bp->bio_data;
1675 	resid = bp->bio_length;
1676 
1677 	if (resid > 0 && (off < 0 || off >= volsize)) {
1678 		error = SET_ERROR(EIO);
1679 		goto out;
1680 	}
1681 
1682 	is_dumpified = B_FALSE;
1683 	sync = !doread && !is_dumpified &&
1684 	    zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1685 #endif	/* illumos */
1686 
1687 	/*
1688 	 * There must be no buffer changes when doing a dmu_sync() because
1689 	 * we can't change the data whilst calculating the checksum.
1690 	 */
1691 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1692 	    doread ? RL_READER : RL_WRITER);
1693 
1694 #ifndef illumos
1695 	if (bp->bio_cmd == BIO_DELETE) {
1696 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1697 		error = dmu_tx_assign(tx, TXG_WAIT);
1698 		if (error != 0) {
1699 			dmu_tx_abort(tx);
1700 		} else {
1701 			zvol_log_truncate(zv, tx, off, resid, sync);
1702 			dmu_tx_commit(tx);
1703 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1704 			    off, resid);
1705 			resid = 0;
1706 		}
1707 		goto unlock;
1708 	}
1709 #endif
1710 	while (resid != 0 && off < volsize) {
1711 		size_t size = MIN(resid, zvol_maxphys);
1712 #ifdef illumos
1713 		if (is_dumpified) {
1714 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1715 			error = zvol_dumpio(zv, addr, off, size,
1716 			    doread, B_FALSE);
1717 		} else if (doread) {
1718 #else
1719 		if (doread) {
1720 #endif
1721 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1722 			    DMU_READ_PREFETCH);
1723 		} else {
1724 			dmu_tx_t *tx = dmu_tx_create(os);
1725 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1726 			error = dmu_tx_assign(tx, TXG_WAIT);
1727 			if (error) {
1728 				dmu_tx_abort(tx);
1729 			} else {
1730 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1731 				zvol_log_write(zv, tx, off, size, sync);
1732 				dmu_tx_commit(tx);
1733 			}
1734 		}
1735 		if (error) {
1736 			/* convert checksum errors into IO errors */
1737 			if (error == ECKSUM)
1738 				error = SET_ERROR(EIO);
1739 			break;
1740 		}
1741 		off += size;
1742 		addr += size;
1743 		resid -= size;
1744 	}
1745 #ifndef illumos
1746 unlock:
1747 #endif
1748 	zfs_range_unlock(rl);
1749 
1750 #ifdef illumos
1751 	if ((bp->b_resid = resid) == bp->b_bcount)
1752 		bioerror(bp, off > volsize ? EINVAL : error);
1753 
1754 	if (sync)
1755 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1756 	biodone(bp);
1757 
1758 	return (0);
1759 #else	/* !illumos */
1760 	bp->bio_completed = bp->bio_length - resid;
1761 	if (bp->bio_completed < bp->bio_length && off > volsize)
1762 		error = EINVAL;
1763 
1764 	if (sync) {
1765 sync:
1766 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1767 	}
1768 out:
1769 	if (bp->bio_to)
1770 		g_io_deliver(bp, error);
1771 	else
1772 		biofinish(bp, NULL, error);
1773 #endif	/* illumos */
1774 }
1775 
1776 #ifdef illumos
1777 /*
1778  * Set the buffer count to the zvol maximum transfer.
1779  * Using our own routine instead of the default minphys()
1780  * means that for larger writes we write bigger buffers on X86
1781  * (128K instead of 56K) and flush the disk write cache less often
1782  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1783  * 56K on X86 and 128K on sparc).
1784  */
1785 void
1786 zvol_minphys(struct buf *bp)
1787 {
1788 	if (bp->b_bcount > zvol_maxphys)
1789 		bp->b_bcount = zvol_maxphys;
1790 }
1791 
1792 int
1793 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1794 {
1795 	minor_t minor = getminor(dev);
1796 	zvol_state_t *zv;
1797 	int error = 0;
1798 	uint64_t size;
1799 	uint64_t boff;
1800 	uint64_t resid;
1801 
1802 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1803 	if (zv == NULL)
1804 		return (SET_ERROR(ENXIO));
1805 
1806 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1807 		return (SET_ERROR(EINVAL));
1808 
1809 	boff = ldbtob(blkno);
1810 	resid = ldbtob(nblocks);
1811 
1812 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1813 
1814 	while (resid) {
1815 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1816 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1817 		if (error)
1818 			break;
1819 		boff += size;
1820 		addr += size;
1821 		resid -= size;
1822 	}
1823 
1824 	return (error);
1825 }
1826 
1827 /*ARGSUSED*/
1828 int
1829 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1830 {
1831 	minor_t minor = getminor(dev);
1832 #else	/* !illumos */
1833 int
1834 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1835 {
1836 #endif	/* illumos */
1837 	zvol_state_t *zv;
1838 	uint64_t volsize;
1839 	rl_t *rl;
1840 	int error = 0;
1841 
1842 #ifdef illumos
1843 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1844 	if (zv == NULL)
1845 		return (SET_ERROR(ENXIO));
1846 #else
1847 	zv = dev->si_drv2;
1848 #endif
1849 
1850 	volsize = zv->zv_volsize;
1851 	/* uio_loffset == volsize isn't an error as its required for EOF processing. */
1852 	if (uio->uio_resid > 0 &&
1853 	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1854 		return (SET_ERROR(EIO));
1855 
1856 #ifdef illumos
1857 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1858 		error = physio(zvol_strategy, NULL, dev, B_READ,
1859 		    zvol_minphys, uio);
1860 		return (error);
1861 	}
1862 #endif
1863 
1864 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1865 	    RL_READER);
1866 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1867 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1868 
1869 		/* don't read past the end */
1870 		if (bytes > volsize - uio->uio_loffset)
1871 			bytes = volsize - uio->uio_loffset;
1872 
1873 		error =  dmu_read_uio_dnode(zv->zv_dn, uio, bytes);
1874 		if (error) {
1875 			/* convert checksum errors into IO errors */
1876 			if (error == ECKSUM)
1877 				error = SET_ERROR(EIO);
1878 			break;
1879 		}
1880 	}
1881 	zfs_range_unlock(rl);
1882 	return (error);
1883 }
1884 
1885 #ifdef illumos
1886 /*ARGSUSED*/
1887 int
1888 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1889 {
1890 	minor_t minor = getminor(dev);
1891 #else	/* !illumos */
1892 int
1893 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1894 {
1895 #endif	/* illumos */
1896 	zvol_state_t *zv;
1897 	uint64_t volsize;
1898 	rl_t *rl;
1899 	int error = 0;
1900 	boolean_t sync;
1901 
1902 #ifdef illumos
1903 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1904 	if (zv == NULL)
1905 		return (SET_ERROR(ENXIO));
1906 #else
1907 	zv = dev->si_drv2;
1908 #endif
1909 
1910 	volsize = zv->zv_volsize;
1911 	/* uio_loffset == volsize isn't an error as its required for EOF processing. */
1912 	if (uio->uio_resid > 0 &&
1913 	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1914 		return (SET_ERROR(EIO));
1915 
1916 #ifdef illumos
1917 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1918 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1919 		    zvol_minphys, uio);
1920 		return (error);
1921 	}
1922 
1923 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1924 #else
1925 	sync = (ioflag & IO_SYNC) ||
1926 #endif
1927 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1928 
1929 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1930 	    RL_WRITER);
1931 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1932 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1933 		uint64_t off = uio->uio_loffset;
1934 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1935 
1936 		if (bytes > volsize - off)	/* don't write past the end */
1937 			bytes = volsize - off;
1938 
1939 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1940 		error = dmu_tx_assign(tx, TXG_WAIT);
1941 		if (error) {
1942 			dmu_tx_abort(tx);
1943 			break;
1944 		}
1945 		error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1946 		if (error == 0)
1947 			zvol_log_write(zv, tx, off, bytes, sync);
1948 		dmu_tx_commit(tx);
1949 
1950 		if (error)
1951 			break;
1952 	}
1953 	zfs_range_unlock(rl);
1954 	if (sync)
1955 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1956 	return (error);
1957 }
1958 
1959 #ifdef illumos
1960 int
1961 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1962 {
1963 	struct uuid uuid = EFI_RESERVED;
1964 	efi_gpe_t gpe = { 0 };
1965 	uint32_t crc;
1966 	dk_efi_t efi;
1967 	int length;
1968 	char *ptr;
1969 
1970 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1971 		return (SET_ERROR(EFAULT));
1972 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1973 	length = efi.dki_length;
1974 	/*
1975 	 * Some clients may attempt to request a PMBR for the
1976 	 * zvol.  Currently this interface will return EINVAL to
1977 	 * such requests.  These requests could be supported by
1978 	 * adding a check for lba == 0 and consing up an appropriate
1979 	 * PMBR.
1980 	 */
1981 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1982 		return (SET_ERROR(EINVAL));
1983 
1984 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1985 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1986 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1987 
1988 	if (efi.dki_lba == 1) {
1989 		efi_gpt_t gpt = { 0 };
1990 
1991 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1992 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1993 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1994 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1995 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1996 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1997 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1998 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1999 		gpt.efi_gpt_SizeOfPartitionEntry =
2000 		    LE_32(sizeof (efi_gpe_t));
2001 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
2002 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
2003 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
2004 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
2005 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
2006 		    flag))
2007 			return (SET_ERROR(EFAULT));
2008 		ptr += sizeof (gpt);
2009 		length -= sizeof (gpt);
2010 	}
2011 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
2012 	    length), flag))
2013 		return (SET_ERROR(EFAULT));
2014 	return (0);
2015 }
2016 
2017 /*
2018  * BEGIN entry points to allow external callers access to the volume.
2019  */
2020 /*
2021  * Return the volume parameters needed for access from an external caller.
2022  * These values are invariant as long as the volume is held open.
2023  */
2024 int
2025 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
2026     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
2027     void **rl_hdl, void **dnode_hdl)
2028 {
2029 	zvol_state_t *zv;
2030 
2031 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2032 	if (zv == NULL)
2033 		return (SET_ERROR(ENXIO));
2034 	if (zv->zv_flags & ZVOL_DUMPIFIED)
2035 		return (SET_ERROR(ENXIO));
2036 
2037 	ASSERT(blksize && max_xfer_len && minor_hdl &&
2038 	    objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
2039 
2040 	*blksize = zv->zv_volblocksize;
2041 	*max_xfer_len = (uint64_t)zvol_maxphys;
2042 	*minor_hdl = zv;
2043 	*objset_hdl = zv->zv_objset;
2044 	*zil_hdl = zv->zv_zilog;
2045 	*rl_hdl = &zv->zv_znode;
2046 	*dnode_hdl = zv->zv_dn;
2047 	return (0);
2048 }
2049 
2050 /*
2051  * Return the current volume size to an external caller.
2052  * The size can change while the volume is open.
2053  */
2054 uint64_t
2055 zvol_get_volume_size(void *minor_hdl)
2056 {
2057 	zvol_state_t *zv = minor_hdl;
2058 
2059 	return (zv->zv_volsize);
2060 }
2061 
2062 /*
2063  * Return the current WCE setting to an external caller.
2064  * The WCE setting can change while the volume is open.
2065  */
2066 int
2067 zvol_get_volume_wce(void *minor_hdl)
2068 {
2069 	zvol_state_t *zv = minor_hdl;
2070 
2071 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2072 }
2073 
2074 /*
2075  * Entry point for external callers to zvol_log_write
2076  */
2077 void
2078 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2079     boolean_t sync)
2080 {
2081 	zvol_state_t *zv = minor_hdl;
2082 
2083 	zvol_log_write(zv, tx, off, resid, sync);
2084 }
2085 /*
2086  * END entry points to allow external callers access to the volume.
2087  */
2088 #endif	/* illumos */
2089 
2090 /*
2091  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2092  */
2093 static void
2094 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2095     boolean_t sync)
2096 {
2097 	itx_t *itx;
2098 	lr_truncate_t *lr;
2099 	zilog_t *zilog = zv->zv_zilog;
2100 
2101 	if (zil_replaying(zilog, tx))
2102 		return;
2103 
2104 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2105 	lr = (lr_truncate_t *)&itx->itx_lr;
2106 	lr->lr_foid = ZVOL_OBJ;
2107 	lr->lr_offset = off;
2108 	lr->lr_length = len;
2109 
2110 	itx->itx_sync = (sync || zv->zv_sync_cnt != 0);
2111 	zil_itx_assign(zilog, itx, tx);
2112 }
2113 
2114 #ifdef illumos
2115 /*
2116  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
2117  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2118  */
2119 /*ARGSUSED*/
2120 int
2121 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2122 {
2123 	zvol_state_t *zv;
2124 	struct dk_callback *dkc;
2125 	int error = 0;
2126 	rl_t *rl;
2127 
2128 	mutex_enter(&zfsdev_state_lock);
2129 
2130 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2131 
2132 	if (zv == NULL) {
2133 		mutex_exit(&zfsdev_state_lock);
2134 		return (SET_ERROR(ENXIO));
2135 	}
2136 	ASSERT(zv->zv_total_opens > 0);
2137 
2138 	switch (cmd) {
2139 
2140 	case DKIOCINFO:
2141 	{
2142 		struct dk_cinfo dki;
2143 
2144 		bzero(&dki, sizeof (dki));
2145 		(void) strcpy(dki.dki_cname, "zvol");
2146 		(void) strcpy(dki.dki_dname, "zvol");
2147 		dki.dki_ctype = DKC_UNKNOWN;
2148 		dki.dki_unit = getminor(dev);
2149 		dki.dki_maxtransfer =
2150 		    1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2151 		mutex_exit(&zfsdev_state_lock);
2152 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2153 			error = SET_ERROR(EFAULT);
2154 		return (error);
2155 	}
2156 
2157 	case DKIOCGMEDIAINFO:
2158 	{
2159 		struct dk_minfo dkm;
2160 
2161 		bzero(&dkm, sizeof (dkm));
2162 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
2163 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2164 		dkm.dki_media_type = DK_UNKNOWN;
2165 		mutex_exit(&zfsdev_state_lock);
2166 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2167 			error = SET_ERROR(EFAULT);
2168 		return (error);
2169 	}
2170 
2171 	case DKIOCGMEDIAINFOEXT:
2172 	{
2173 		struct dk_minfo_ext dkmext;
2174 
2175 		bzero(&dkmext, sizeof (dkmext));
2176 		dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2177 		dkmext.dki_pbsize = zv->zv_volblocksize;
2178 		dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2179 		dkmext.dki_media_type = DK_UNKNOWN;
2180 		mutex_exit(&zfsdev_state_lock);
2181 		if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2182 			error = SET_ERROR(EFAULT);
2183 		return (error);
2184 	}
2185 
2186 	case DKIOCGETEFI:
2187 	{
2188 		uint64_t vs = zv->zv_volsize;
2189 		uint8_t bs = zv->zv_min_bs;
2190 
2191 		mutex_exit(&zfsdev_state_lock);
2192 		error = zvol_getefi((void *)arg, flag, vs, bs);
2193 		return (error);
2194 	}
2195 
2196 	case DKIOCFLUSHWRITECACHE:
2197 		dkc = (struct dk_callback *)arg;
2198 		mutex_exit(&zfsdev_state_lock);
2199 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
2200 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2201 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
2202 			error = 0;
2203 		}
2204 		return (error);
2205 
2206 	case DKIOCGETWCE:
2207 	{
2208 		int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2209 		if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2210 		    flag))
2211 			error = SET_ERROR(EFAULT);
2212 		break;
2213 	}
2214 	case DKIOCSETWCE:
2215 	{
2216 		int wce;
2217 		if (ddi_copyin((void *)arg, &wce, sizeof (int),
2218 		    flag)) {
2219 			error = SET_ERROR(EFAULT);
2220 			break;
2221 		}
2222 		if (wce) {
2223 			zv->zv_flags |= ZVOL_WCE;
2224 			mutex_exit(&zfsdev_state_lock);
2225 		} else {
2226 			zv->zv_flags &= ~ZVOL_WCE;
2227 			mutex_exit(&zfsdev_state_lock);
2228 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
2229 		}
2230 		return (0);
2231 	}
2232 
2233 	case DKIOCGGEOM:
2234 	case DKIOCGVTOC:
2235 		/*
2236 		 * commands using these (like prtvtoc) expect ENOTSUP
2237 		 * since we're emulating an EFI label
2238 		 */
2239 		error = SET_ERROR(ENOTSUP);
2240 		break;
2241 
2242 	case DKIOCDUMPINIT:
2243 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2244 		    RL_WRITER);
2245 		error = zvol_dumpify(zv);
2246 		zfs_range_unlock(rl);
2247 		break;
2248 
2249 	case DKIOCDUMPFINI:
2250 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2251 			break;
2252 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2253 		    RL_WRITER);
2254 		error = zvol_dump_fini(zv);
2255 		zfs_range_unlock(rl);
2256 		break;
2257 
2258 	case DKIOCFREE:
2259 	{
2260 		dkioc_free_list_t *dfl;
2261 		dmu_tx_t *tx;
2262 
2263 		if (!zvol_unmap_enabled)
2264 			break;
2265 
2266 		if (!(flag & FKIOCTL)) {
2267 			error = dfl_copyin((void *)arg, &dfl, flag, KM_SLEEP);
2268 			if (error != 0)
2269 				break;
2270 		} else {
2271 			dfl = (dkioc_free_list_t *)arg;
2272 			ASSERT3U(dfl->dfl_num_exts, <=, DFL_COPYIN_MAX_EXTS);
2273 			if (dfl->dfl_num_exts > DFL_COPYIN_MAX_EXTS) {
2274 				error = SET_ERROR(EINVAL);
2275 				break;
2276 			}
2277 		}
2278 
2279 		mutex_exit(&zfsdev_state_lock);
2280 
2281 		for (int i = 0; i < dfl->dfl_num_exts; i++) {
2282 			uint64_t start = dfl->dfl_exts[i].dfle_start,
2283 			    length = dfl->dfl_exts[i].dfle_length,
2284 			    end = start + length;
2285 
2286 			/*
2287 			 * Apply Postel's Law to length-checking.  If they
2288 			 * overshoot, just blank out until the end, if there's
2289 			 * a need to blank out anything.
2290 			 */
2291 			if (start >= zv->zv_volsize)
2292 				continue;	/* No need to do anything... */
2293 			if (end > zv->zv_volsize) {
2294 				end = DMU_OBJECT_END;
2295 				length = end - start;
2296 			}
2297 
2298 			rl = zfs_range_lock(&zv->zv_znode, start, length,
2299 			    RL_WRITER);
2300 			tx = dmu_tx_create(zv->zv_objset);
2301 			error = dmu_tx_assign(tx, TXG_WAIT);
2302 			if (error != 0) {
2303 				dmu_tx_abort(tx);
2304 			} else {
2305 				zvol_log_truncate(zv, tx, start, length,
2306 				    B_TRUE);
2307 				dmu_tx_commit(tx);
2308 				error = dmu_free_long_range(zv->zv_objset,
2309 				    ZVOL_OBJ, start, length);
2310 			}
2311 
2312 			zfs_range_unlock(rl);
2313 
2314 			if (error != 0)
2315 				break;
2316 		}
2317 
2318 		/*
2319 		 * If the write-cache is disabled, 'sync' property
2320 		 * is set to 'always', or if the caller is asking for
2321 		 * a synchronous free, commit this operation to the zil.
2322 		 * This will sync any previous uncommitted writes to the
2323 		 * zvol object.
2324 		 * Can be overridden by the zvol_unmap_sync_enabled tunable.
2325 		 */
2326 		if ((error == 0) && zvol_unmap_sync_enabled &&
2327 		    (!(zv->zv_flags & ZVOL_WCE) ||
2328 		    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
2329 		    (dfl->dfl_flags & DF_WAIT_SYNC))) {
2330 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
2331 		}
2332 
2333 		if (!(flag & FKIOCTL))
2334 			dfl_free(dfl);
2335 
2336 		return (error);
2337 	}
2338 
2339 	default:
2340 		error = SET_ERROR(ENOTTY);
2341 		break;
2342 
2343 	}
2344 	mutex_exit(&zfsdev_state_lock);
2345 	return (error);
2346 }
2347 #endif	/* illumos */
2348 
2349 int
2350 zvol_busy(void)
2351 {
2352 	return (zvol_minors != 0);
2353 }
2354 
2355 void
2356 zvol_init(void)
2357 {
2358 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2359 	    1) == 0);
2360 #ifdef illumos
2361 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2362 #else
2363 	ZFS_LOG(1, "ZVOL Initialized.");
2364 #endif
2365 }
2366 
2367 void
2368 zvol_fini(void)
2369 {
2370 #ifdef illumos
2371 	mutex_destroy(&zfsdev_state_lock);
2372 #endif
2373 	ddi_soft_state_fini(&zfsdev_state);
2374 	ZFS_LOG(1, "ZVOL Deinitialized.");
2375 }
2376 
2377 #ifdef illumos
2378 /*ARGSUSED*/
2379 static int
2380 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2381 {
2382 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2383 
2384 	if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2385 		return (1);
2386 	return (0);
2387 }
2388 
2389 /*ARGSUSED*/
2390 static void
2391 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2392 {
2393 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2394 
2395 	spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2396 }
2397 
2398 static int
2399 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2400 {
2401 	dmu_tx_t *tx;
2402 	int error;
2403 	objset_t *os = zv->zv_objset;
2404 	spa_t *spa = dmu_objset_spa(os);
2405 	vdev_t *vd = spa->spa_root_vdev;
2406 	nvlist_t *nv = NULL;
2407 	uint64_t version = spa_version(spa);
2408 	uint64_t checksum, compress, refresrv, vbs, dedup;
2409 
2410 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2411 	ASSERT(vd->vdev_ops == &vdev_root_ops);
2412 
2413 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2414 	    DMU_OBJECT_END);
2415 	if (error != 0)
2416 		return (error);
2417 	/* wait for dmu_free_long_range to actually free the blocks */
2418 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2419 
2420 	/*
2421 	 * If the pool on which the dump device is being initialized has more
2422 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2423 	 * enabled.  If so, bump that feature's counter to indicate that the
2424 	 * feature is active. We also check the vdev type to handle the
2425 	 * following case:
2426 	 *   # zpool create test raidz disk1 disk2 disk3
2427 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2428 	 *   the raidz vdev itself has 3 children.
2429 	 */
2430 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2431 		if (!spa_feature_is_enabled(spa,
2432 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2433 			return (SET_ERROR(ENOTSUP));
2434 		(void) dsl_sync_task(spa_name(spa),
2435 		    zfs_mvdev_dump_feature_check,
2436 		    zfs_mvdev_dump_activate_feature_sync, NULL,
2437 		    2, ZFS_SPACE_CHECK_RESERVED);
2438 	}
2439 
2440 	if (!resize) {
2441 		error = dsl_prop_get_integer(zv->zv_name,
2442 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2443 		if (error == 0) {
2444 			error = dsl_prop_get_integer(zv->zv_name,
2445 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2446 			    NULL);
2447 		}
2448 		if (error == 0) {
2449 			error = dsl_prop_get_integer(zv->zv_name,
2450 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2451 			    &refresrv, NULL);
2452 		}
2453 		if (error == 0) {
2454 			error = dsl_prop_get_integer(zv->zv_name,
2455 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2456 			    NULL);
2457 		}
2458 		if (version >= SPA_VERSION_DEDUP && error == 0) {
2459 			error = dsl_prop_get_integer(zv->zv_name,
2460 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2461 		}
2462 	}
2463 	if (error != 0)
2464 		return (error);
2465 
2466 	tx = dmu_tx_create(os);
2467 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2468 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2469 	error = dmu_tx_assign(tx, TXG_WAIT);
2470 	if (error != 0) {
2471 		dmu_tx_abort(tx);
2472 		return (error);
2473 	}
2474 
2475 	/*
2476 	 * If we are resizing the dump device then we only need to
2477 	 * update the refreservation to match the newly updated
2478 	 * zvolsize. Otherwise, we save off the original state of the
2479 	 * zvol so that we can restore them if the zvol is ever undumpified.
2480 	 */
2481 	if (resize) {
2482 		error = zap_update(os, ZVOL_ZAP_OBJ,
2483 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2484 		    &zv->zv_volsize, tx);
2485 	} else {
2486 		error = zap_update(os, ZVOL_ZAP_OBJ,
2487 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2488 		    &compress, tx);
2489 		if (error == 0) {
2490 			error = zap_update(os, ZVOL_ZAP_OBJ,
2491 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2492 			    &checksum, tx);
2493 		}
2494 		if (error == 0) {
2495 			error = zap_update(os, ZVOL_ZAP_OBJ,
2496 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2497 			    &refresrv, tx);
2498 		}
2499 		if (error == 0) {
2500 			error = zap_update(os, ZVOL_ZAP_OBJ,
2501 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2502 			    &vbs, tx);
2503 		}
2504 		if (error == 0) {
2505 			error = dmu_object_set_blocksize(
2506 			    os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2507 		}
2508 		if (version >= SPA_VERSION_DEDUP && error == 0) {
2509 			error = zap_update(os, ZVOL_ZAP_OBJ,
2510 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2511 			    &dedup, tx);
2512 		}
2513 		if (error == 0)
2514 			zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2515 	}
2516 	dmu_tx_commit(tx);
2517 
2518 	/*
2519 	 * We only need update the zvol's property if we are initializing
2520 	 * the dump area for the first time.
2521 	 */
2522 	if (error == 0 && !resize) {
2523 		/*
2524 		 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2525 		 * function.  Otherwise, use the old default -- OFF.
2526 		 */
2527 		checksum = spa_feature_is_active(spa,
2528 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2529 		    ZIO_CHECKSUM_OFF;
2530 
2531 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2532 		VERIFY(nvlist_add_uint64(nv,
2533 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2534 		VERIFY(nvlist_add_uint64(nv,
2535 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2536 		    ZIO_COMPRESS_OFF) == 0);
2537 		VERIFY(nvlist_add_uint64(nv,
2538 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2539 		    checksum) == 0);
2540 		if (version >= SPA_VERSION_DEDUP) {
2541 			VERIFY(nvlist_add_uint64(nv,
2542 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
2543 			    ZIO_CHECKSUM_OFF) == 0);
2544 		}
2545 
2546 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2547 		    nv, NULL);
2548 		nvlist_free(nv);
2549 	}
2550 
2551 	/* Allocate the space for the dump */
2552 	if (error == 0)
2553 		error = zvol_prealloc(zv);
2554 	return (error);
2555 }
2556 
2557 static int
2558 zvol_dumpify(zvol_state_t *zv)
2559 {
2560 	int error = 0;
2561 	uint64_t dumpsize = 0;
2562 	dmu_tx_t *tx;
2563 	objset_t *os = zv->zv_objset;
2564 
2565 	if (zv->zv_flags & ZVOL_RDONLY)
2566 		return (SET_ERROR(EROFS));
2567 
2568 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2569 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2570 		boolean_t resize = (dumpsize > 0);
2571 
2572 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2573 			(void) zvol_dump_fini(zv);
2574 			return (error);
2575 		}
2576 	}
2577 
2578 	/*
2579 	 * Build up our lba mapping.
2580 	 */
2581 	error = zvol_get_lbas(zv);
2582 	if (error) {
2583 		(void) zvol_dump_fini(zv);
2584 		return (error);
2585 	}
2586 
2587 	tx = dmu_tx_create(os);
2588 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2589 	error = dmu_tx_assign(tx, TXG_WAIT);
2590 	if (error) {
2591 		dmu_tx_abort(tx);
2592 		(void) zvol_dump_fini(zv);
2593 		return (error);
2594 	}
2595 
2596 	zv->zv_flags |= ZVOL_DUMPIFIED;
2597 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2598 	    &zv->zv_volsize, tx);
2599 	dmu_tx_commit(tx);
2600 
2601 	if (error) {
2602 		(void) zvol_dump_fini(zv);
2603 		return (error);
2604 	}
2605 
2606 	txg_wait_synced(dmu_objset_pool(os), 0);
2607 	return (0);
2608 }
2609 
2610 static int
2611 zvol_dump_fini(zvol_state_t *zv)
2612 {
2613 	dmu_tx_t *tx;
2614 	objset_t *os = zv->zv_objset;
2615 	nvlist_t *nv;
2616 	int error = 0;
2617 	uint64_t checksum, compress, refresrv, vbs, dedup;
2618 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2619 
2620 	/*
2621 	 * Attempt to restore the zvol back to its pre-dumpified state.
2622 	 * This is a best-effort attempt as it's possible that not all
2623 	 * of these properties were initialized during the dumpify process
2624 	 * (i.e. error during zvol_dump_init).
2625 	 */
2626 
2627 	tx = dmu_tx_create(os);
2628 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2629 	error = dmu_tx_assign(tx, TXG_WAIT);
2630 	if (error) {
2631 		dmu_tx_abort(tx);
2632 		return (error);
2633 	}
2634 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2635 	dmu_tx_commit(tx);
2636 
2637 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2638 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2639 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2640 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2641 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2642 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2643 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2644 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2645 
2646 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2647 	(void) nvlist_add_uint64(nv,
2648 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2649 	(void) nvlist_add_uint64(nv,
2650 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2651 	(void) nvlist_add_uint64(nv,
2652 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2653 	if (version >= SPA_VERSION_DEDUP &&
2654 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2655 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2656 		(void) nvlist_add_uint64(nv,
2657 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2658 	}
2659 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2660 	    nv, NULL);
2661 	nvlist_free(nv);
2662 
2663 	zvol_free_extents(zv);
2664 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2665 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2666 	/* wait for dmu_free_long_range to actually free the blocks */
2667 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2668 	tx = dmu_tx_create(os);
2669 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2670 	error = dmu_tx_assign(tx, TXG_WAIT);
2671 	if (error) {
2672 		dmu_tx_abort(tx);
2673 		return (error);
2674 	}
2675 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2676 		zv->zv_volblocksize = vbs;
2677 	dmu_tx_commit(tx);
2678 
2679 	return (0);
2680 }
2681 #else	/* !illumos */
2682 
2683 static void
2684 zvol_geom_run(zvol_state_t *zv)
2685 {
2686 	struct g_provider *pp;
2687 
2688 	pp = zv->zv_provider;
2689 	g_error_provider(pp, 0);
2690 
2691 	kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2692 	    "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2693 }
2694 
2695 static void
2696 zvol_geom_destroy(zvol_state_t *zv)
2697 {
2698 	struct g_provider *pp;
2699 
2700 	g_topology_assert();
2701 
2702 	mtx_lock(&zv->zv_queue_mtx);
2703 	zv->zv_state = 1;
2704 	wakeup_one(&zv->zv_queue);
2705 	while (zv->zv_state != 2)
2706 		msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2707 	mtx_destroy(&zv->zv_queue_mtx);
2708 
2709 	pp = zv->zv_provider;
2710 	zv->zv_provider = NULL;
2711 	pp->private = NULL;
2712 	g_wither_geom(pp->geom, ENXIO);
2713 }
2714 
2715 static int
2716 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2717 {
2718 	int count, error, flags;
2719 
2720 	g_topology_assert();
2721 
2722 	/*
2723 	 * To make it easier we expect either open or close, but not both
2724 	 * at the same time.
2725 	 */
2726 	KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2727 	    (acr <= 0 && acw <= 0 && ace <= 0),
2728 	    ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2729 	    pp->name, acr, acw, ace));
2730 
2731 	if (pp->private == NULL) {
2732 		if (acr <= 0 && acw <= 0 && ace <= 0)
2733 			return (0);
2734 		return (pp->error);
2735 	}
2736 
2737 	/*
2738 	 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2739 	 * because GEOM already handles that and handles it a bit differently.
2740 	 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2741 	 * only one exclusive consumer, no matter if it is reader or writer.
2742 	 * I like better the way GEOM works so I'll leave it for GEOM to
2743 	 * decide what to do.
2744 	 */
2745 
2746 	count = acr + acw + ace;
2747 	if (count == 0)
2748 		return (0);
2749 
2750 	flags = 0;
2751 	if (acr != 0 || ace != 0)
2752 		flags |= FREAD;
2753 	if (acw != 0)
2754 		flags |= FWRITE;
2755 
2756 	g_topology_unlock();
2757 	if (count > 0)
2758 		error = zvol_open(pp, flags, count);
2759 	else
2760 		error = zvol_close(pp, flags, -count);
2761 	g_topology_lock();
2762 	return (error);
2763 }
2764 
2765 static void
2766 zvol_geom_start(struct bio *bp)
2767 {
2768 	zvol_state_t *zv;
2769 	boolean_t first;
2770 
2771 	zv = bp->bio_to->private;
2772 	ASSERT(zv != NULL);
2773 	switch (bp->bio_cmd) {
2774 	case BIO_FLUSH:
2775 		if (!THREAD_CAN_SLEEP())
2776 			goto enqueue;
2777 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
2778 		g_io_deliver(bp, 0);
2779 		break;
2780 	case BIO_READ:
2781 	case BIO_WRITE:
2782 	case BIO_DELETE:
2783 		if (!THREAD_CAN_SLEEP())
2784 			goto enqueue;
2785 		zvol_strategy(bp);
2786 		break;
2787 	case BIO_GETATTR: {
2788 		spa_t *spa = dmu_objset_spa(zv->zv_objset);
2789 		uint64_t refd, avail, usedobjs, availobjs, val;
2790 
2791 		if (g_handleattr_int(bp, "GEOM::candelete", 1))
2792 			return;
2793 		if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2794 			dmu_objset_space(zv->zv_objset, &refd, &avail,
2795 			    &usedobjs, &availobjs);
2796 			if (g_handleattr_off_t(bp, "blocksavail",
2797 			    avail / DEV_BSIZE))
2798 				return;
2799 		} else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2800 			dmu_objset_space(zv->zv_objset, &refd, &avail,
2801 			    &usedobjs, &availobjs);
2802 			if (g_handleattr_off_t(bp, "blocksused",
2803 			    refd / DEV_BSIZE))
2804 				return;
2805 		} else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2806 			avail = metaslab_class_get_space(spa_normal_class(spa));
2807 			avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2808 			if (g_handleattr_off_t(bp, "poolblocksavail",
2809 			    avail / DEV_BSIZE))
2810 				return;
2811 		} else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2812 			refd = metaslab_class_get_alloc(spa_normal_class(spa));
2813 			if (g_handleattr_off_t(bp, "poolblocksused",
2814 			    refd / DEV_BSIZE))
2815 				return;
2816 		}
2817 		/* FALLTHROUGH */
2818 	}
2819 	default:
2820 		g_io_deliver(bp, EOPNOTSUPP);
2821 		break;
2822 	}
2823 	return;
2824 
2825 enqueue:
2826 	mtx_lock(&zv->zv_queue_mtx);
2827 	first = (bioq_first(&zv->zv_queue) == NULL);
2828 	bioq_insert_tail(&zv->zv_queue, bp);
2829 	mtx_unlock(&zv->zv_queue_mtx);
2830 	if (first)
2831 		wakeup_one(&zv->zv_queue);
2832 }
2833 
2834 static void
2835 zvol_geom_worker(void *arg)
2836 {
2837 	zvol_state_t *zv;
2838 	struct bio *bp;
2839 
2840 	thread_lock(curthread);
2841 	sched_prio(curthread, PRIBIO);
2842 	thread_unlock(curthread);
2843 
2844 	zv = arg;
2845 	for (;;) {
2846 		mtx_lock(&zv->zv_queue_mtx);
2847 		bp = bioq_takefirst(&zv->zv_queue);
2848 		if (bp == NULL) {
2849 			if (zv->zv_state == 1) {
2850 				zv->zv_state = 2;
2851 				wakeup(&zv->zv_state);
2852 				mtx_unlock(&zv->zv_queue_mtx);
2853 				kthread_exit();
2854 			}
2855 			msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2856 			    "zvol:io", 0);
2857 			continue;
2858 		}
2859 		mtx_unlock(&zv->zv_queue_mtx);
2860 		switch (bp->bio_cmd) {
2861 		case BIO_FLUSH:
2862 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
2863 			g_io_deliver(bp, 0);
2864 			break;
2865 		case BIO_READ:
2866 		case BIO_WRITE:
2867 		case BIO_DELETE:
2868 			zvol_strategy(bp);
2869 			break;
2870 		default:
2871 			g_io_deliver(bp, EOPNOTSUPP);
2872 			break;
2873 		}
2874 	}
2875 }
2876 
2877 extern boolean_t dataset_name_hidden(const char *name);
2878 
2879 static int
2880 zvol_create_snapshots(objset_t *os, const char *name)
2881 {
2882 	uint64_t cookie, obj;
2883 	char *sname;
2884 	int error, len;
2885 
2886 	cookie = obj = 0;
2887 	sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2888 
2889 #if 0
2890 	(void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2891 	    DS_FIND_SNAPSHOTS);
2892 #endif
2893 
2894 	for (;;) {
2895 		len = snprintf(sname, MAXPATHLEN, "%s@", name);
2896 		if (len >= MAXPATHLEN) {
2897 			dmu_objset_rele(os, FTAG);
2898 			error = ENAMETOOLONG;
2899 			break;
2900 		}
2901 
2902 		dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2903 		error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2904 		    sname + len, &obj, &cookie, NULL);
2905 		dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2906 		if (error != 0) {
2907 			if (error == ENOENT)
2908 				error = 0;
2909 			break;
2910 		}
2911 
2912 		error = zvol_create_minor(sname);
2913 		if (error != 0 && error != EEXIST) {
2914 			printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2915 			    sname, error);
2916 			break;
2917 		}
2918 	}
2919 
2920 	kmem_free(sname, MAXPATHLEN);
2921 	return (error);
2922 }
2923 
2924 int
2925 zvol_create_minors(const char *name)
2926 {
2927 	uint64_t cookie;
2928 	objset_t *os;
2929 	char *osname, *p;
2930 	int error, len;
2931 
2932 	if (dataset_name_hidden(name))
2933 		return (0);
2934 
2935 	if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2936 		printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2937 		    name, error);
2938 		return (error);
2939 	}
2940 	if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2941 		dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2942 		dsl_pool_rele(dmu_objset_pool(os), FTAG);
2943 		error = zvol_create_minor(name);
2944 		if (error == 0 || error == EEXIST) {
2945 			error = zvol_create_snapshots(os, name);
2946 		} else {
2947 			printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2948 			    name, error);
2949 		}
2950 		dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2951 		dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2952 		return (error);
2953 	}
2954 	if (dmu_objset_type(os) != DMU_OST_ZFS) {
2955 		dmu_objset_rele(os, FTAG);
2956 		return (0);
2957 	}
2958 
2959 	osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2960 	if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2961 		dmu_objset_rele(os, FTAG);
2962 		kmem_free(osname, MAXPATHLEN);
2963 		return (ENOENT);
2964 	}
2965 	p = osname + strlen(osname);
2966 	len = MAXPATHLEN - (p - osname);
2967 
2968 #if 0
2969 	/* Prefetch the datasets. */
2970 	cookie = 0;
2971 	while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2972 		if (!dataset_name_hidden(osname))
2973 			(void) dmu_objset_prefetch(osname, NULL);
2974 	}
2975 #endif
2976 
2977 	cookie = 0;
2978 	while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2979 	    &cookie) == 0) {
2980 		dmu_objset_rele(os, FTAG);
2981 		(void)zvol_create_minors(osname);
2982 		if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2983 			printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2984 			    name, error);
2985 			return (error);
2986 		}
2987 	}
2988 
2989 	dmu_objset_rele(os, FTAG);
2990 	kmem_free(osname, MAXPATHLEN);
2991 	return (0);
2992 }
2993 
2994 static void
2995 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2996 {
2997 	struct g_geom *gp;
2998 	struct g_provider *pp;
2999 	struct cdev *dev;
3000 
3001 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
3002 
3003 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
3004 		g_topology_lock();
3005 		pp = zv->zv_provider;
3006 		ASSERT(pp != NULL);
3007 		gp = pp->geom;
3008 		ASSERT(gp != NULL);
3009 
3010 		zv->zv_provider = NULL;
3011 		g_wither_provider(pp, ENXIO);
3012 
3013 		pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
3014 		pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
3015 		pp->sectorsize = DEV_BSIZE;
3016 		pp->mediasize = zv->zv_volsize;
3017 		pp->private = zv;
3018 		zv->zv_provider = pp;
3019 		g_error_provider(pp, 0);
3020 		g_topology_unlock();
3021 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
3022 		struct make_dev_args args;
3023 
3024 		if ((dev = zv->zv_dev) != NULL) {
3025 			zv->zv_dev = NULL;
3026 			destroy_dev(dev);
3027 			if (zv->zv_total_opens > 0) {
3028 				zv->zv_flags &= ~ZVOL_EXCL;
3029 				zv->zv_total_opens = 0;
3030 				zvol_last_close(zv);
3031 			}
3032 		}
3033 
3034 		make_dev_args_init(&args);
3035 		args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
3036 		args.mda_devsw = &zvol_cdevsw;
3037 		args.mda_cr = NULL;
3038 		args.mda_uid = UID_ROOT;
3039 		args.mda_gid = GID_OPERATOR;
3040 		args.mda_mode = 0640;
3041 		args.mda_si_drv2 = zv;
3042 		if (make_dev_s(&args, &zv->zv_dev,
3043 		    "%s/%s", ZVOL_DRIVER, newname) == 0)
3044 			zv->zv_dev->si_iosize_max = MAXPHYS;
3045 	}
3046 	strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3047 }
3048 
3049 void
3050 zvol_rename_minors(const char *oldname, const char *newname)
3051 {
3052 	char name[MAXPATHLEN];
3053 	struct g_provider *pp;
3054 	struct g_geom *gp;
3055 	size_t oldnamelen, newnamelen;
3056 	zvol_state_t *zv;
3057 	char *namebuf;
3058 	boolean_t locked = B_FALSE;
3059 
3060 	oldnamelen = strlen(oldname);
3061 	newnamelen = strlen(newname);
3062 
3063 	DROP_GIANT();
3064 	/* See comment in zvol_open(). */
3065 	if (!MUTEX_HELD(&zfsdev_state_lock)) {
3066 		mutex_enter(&zfsdev_state_lock);
3067 		locked = B_TRUE;
3068 	}
3069 
3070 	LIST_FOREACH(zv, &all_zvols, zv_links) {
3071 		if (strcmp(zv->zv_name, oldname) == 0) {
3072 			zvol_rename_minor(zv, newname);
3073 		} else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3074 		    (zv->zv_name[oldnamelen] == '/' ||
3075 		     zv->zv_name[oldnamelen] == '@')) {
3076 			snprintf(name, sizeof(name), "%s%c%s", newname,
3077 			    zv->zv_name[oldnamelen],
3078 			    zv->zv_name + oldnamelen + 1);
3079 			zvol_rename_minor(zv, name);
3080 		}
3081 	}
3082 
3083 	if (locked)
3084 		mutex_exit(&zfsdev_state_lock);
3085 	PICKUP_GIANT();
3086 }
3087 
3088 static int
3089 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3090 {
3091 	zvol_state_t *zv = dev->si_drv2;
3092 	int err = 0;
3093 
3094 	mutex_enter(&zfsdev_state_lock);
3095 	if (zv->zv_total_opens == 0)
3096 		err = zvol_first_open(zv);
3097 	if (err) {
3098 		mutex_exit(&zfsdev_state_lock);
3099 		return (err);
3100 	}
3101 	if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3102 		err = SET_ERROR(EROFS);
3103 		goto out;
3104 	}
3105 	if (zv->zv_flags & ZVOL_EXCL) {
3106 		err = SET_ERROR(EBUSY);
3107 		goto out;
3108 	}
3109 #ifdef FEXCL
3110 	if (flags & FEXCL) {
3111 		if (zv->zv_total_opens != 0) {
3112 			err = SET_ERROR(EBUSY);
3113 			goto out;
3114 		}
3115 		zv->zv_flags |= ZVOL_EXCL;
3116 	}
3117 #endif
3118 
3119 	zv->zv_total_opens++;
3120 	if (flags & (FSYNC | FDSYNC)) {
3121 		zv->zv_sync_cnt++;
3122 		if (zv->zv_sync_cnt == 1)
3123 			zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
3124 	}
3125 	mutex_exit(&zfsdev_state_lock);
3126 	return (err);
3127 out:
3128 	if (zv->zv_total_opens == 0)
3129 		zvol_last_close(zv);
3130 	mutex_exit(&zfsdev_state_lock);
3131 	return (err);
3132 }
3133 
3134 static int
3135 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3136 {
3137 	zvol_state_t *zv = dev->si_drv2;
3138 
3139 	mutex_enter(&zfsdev_state_lock);
3140 	if (zv->zv_flags & ZVOL_EXCL) {
3141 		ASSERT(zv->zv_total_opens == 1);
3142 		zv->zv_flags &= ~ZVOL_EXCL;
3143 	}
3144 
3145 	/*
3146 	 * If the open count is zero, this is a spurious close.
3147 	 * That indicates a bug in the kernel / DDI framework.
3148 	 */
3149 	ASSERT(zv->zv_total_opens != 0);
3150 
3151 	/*
3152 	 * You may get multiple opens, but only one close.
3153 	 */
3154 	zv->zv_total_opens--;
3155 	if (flags & (FSYNC | FDSYNC))
3156 		zv->zv_sync_cnt--;
3157 
3158 	if (zv->zv_total_opens == 0)
3159 		zvol_last_close(zv);
3160 
3161 	mutex_exit(&zfsdev_state_lock);
3162 	return (0);
3163 }
3164 
3165 static int
3166 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3167 {
3168 	zvol_state_t *zv;
3169 	rl_t *rl;
3170 	off_t offset, length;
3171 	int i, error;
3172 	boolean_t sync;
3173 
3174 	zv = dev->si_drv2;
3175 
3176 	error = 0;
3177 	KASSERT(zv->zv_total_opens > 0,
3178 	    ("Device with zero access count in zvol_d_ioctl"));
3179 
3180 	i = IOCPARM_LEN(cmd);
3181 	switch (cmd) {
3182 	case DIOCGSECTORSIZE:
3183 		*(u_int *)data = DEV_BSIZE;
3184 		break;
3185 	case DIOCGMEDIASIZE:
3186 		*(off_t *)data = zv->zv_volsize;
3187 		break;
3188 	case DIOCGFLUSH:
3189 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
3190 		break;
3191 	case DIOCGDELETE:
3192 		if (!zvol_unmap_enabled)
3193 			break;
3194 
3195 		offset = ((off_t *)data)[0];
3196 		length = ((off_t *)data)[1];
3197 		if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3198 		    offset < 0 || offset >= zv->zv_volsize ||
3199 		    length <= 0) {
3200 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
3201 			    length);
3202 			error = EINVAL;
3203 			break;
3204 		}
3205 
3206 		rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3207 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3208 		error = dmu_tx_assign(tx, TXG_WAIT);
3209 		if (error != 0) {
3210 			sync = FALSE;
3211 			dmu_tx_abort(tx);
3212 		} else {
3213 			sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
3214 			zvol_log_truncate(zv, tx, offset, length, sync);
3215 			dmu_tx_commit(tx);
3216 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3217 			    offset, length);
3218 		}
3219 		zfs_range_unlock(rl);
3220 		if (sync)
3221 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
3222 		break;
3223 	case DIOCGSTRIPESIZE:
3224 		*(off_t *)data = zv->zv_volblocksize;
3225 		break;
3226 	case DIOCGSTRIPEOFFSET:
3227 		*(off_t *)data = 0;
3228 		break;
3229 	case DIOCGATTR: {
3230 		spa_t *spa = dmu_objset_spa(zv->zv_objset);
3231 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3232 		uint64_t refd, avail, usedobjs, availobjs;
3233 
3234 		if (strcmp(arg->name, "GEOM::candelete") == 0)
3235 			arg->value.i = 1;
3236 		else if (strcmp(arg->name, "blocksavail") == 0) {
3237 			dmu_objset_space(zv->zv_objset, &refd, &avail,
3238 			    &usedobjs, &availobjs);
3239 			arg->value.off = avail / DEV_BSIZE;
3240 		} else if (strcmp(arg->name, "blocksused") == 0) {
3241 			dmu_objset_space(zv->zv_objset, &refd, &avail,
3242 			    &usedobjs, &availobjs);
3243 			arg->value.off = refd / DEV_BSIZE;
3244 		} else if (strcmp(arg->name, "poolblocksavail") == 0) {
3245 			avail = metaslab_class_get_space(spa_normal_class(spa));
3246 			avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3247 			arg->value.off = avail / DEV_BSIZE;
3248 		} else if (strcmp(arg->name, "poolblocksused") == 0) {
3249 			refd = metaslab_class_get_alloc(spa_normal_class(spa));
3250 			arg->value.off = refd / DEV_BSIZE;
3251 		} else
3252 			error = ENOIOCTL;
3253 		break;
3254 	}
3255 	case FIOSEEKHOLE:
3256 	case FIOSEEKDATA: {
3257 		off_t *off = (off_t *)data;
3258 		uint64_t noff;
3259 		boolean_t hole;
3260 
3261 		hole = (cmd == FIOSEEKHOLE);
3262 		noff = *off;
3263 		error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3264 		*off = noff;
3265 		break;
3266 	}
3267 	default:
3268 		error = ENOIOCTL;
3269 	}
3270 
3271 	return (error);
3272 }
3273 #endif	/* illumos */
3274