xref: /freebsd-11-stable/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c (revision 45b7761575f02b13dbdabab87a8bb9bdbac0d0a0)
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 <pjd@FreeBSD.org>
25  * All rights reserved.
26  *
27  * Portions Copyright 2010 Robert Milkowski
28  *
29  * Copyright 2011 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 <mm@FreeBSD.org> */
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 static void
1325 zvol_get_done(zgd_t *zgd, int error)
1326 {
1327 	if (zgd->zgd_db)
1328 		dmu_buf_rele(zgd->zgd_db, zgd);
1329 
1330 	zfs_range_unlock(zgd->zgd_rl);
1331 
1332 	if (error == 0 && zgd->zgd_bp)
1333 		zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1334 
1335 	kmem_free(zgd, sizeof (zgd_t));
1336 }
1337 
1338 /*
1339  * Get data to generate a TX_WRITE intent log record.
1340  */
1341 static int
1342 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1343 {
1344 	zvol_state_t *zv = arg;
1345 	uint64_t offset = lr->lr_offset;
1346 	uint64_t size = lr->lr_length;	/* length of user data */
1347 	dmu_buf_t *db;
1348 	zgd_t *zgd;
1349 	int error;
1350 
1351 	ASSERT3P(lwb, !=, NULL);
1352 	ASSERT3P(zio, !=, NULL);
1353 	ASSERT3U(size, !=, 0);
1354 
1355 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1356 	zgd->zgd_lwb = lwb;
1357 
1358 	/*
1359 	 * Write records come in two flavors: immediate and indirect.
1360 	 * For small writes it's cheaper to store the data with the
1361 	 * log record (immediate); for large writes it's cheaper to
1362 	 * sync the data and get a pointer to it (indirect) so that
1363 	 * we don't have to write the data twice.
1364 	 */
1365 	if (buf != NULL) { /* immediate write */
1366 		zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1367 		    RL_READER);
1368 		error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1369 		    DMU_READ_NO_PREFETCH);
1370 	} else { /* indirect write */
1371 		/*
1372 		 * Have to lock the whole block to ensure when it's written out
1373 		 * and its checksum is being calculated that no one can change
1374 		 * the data. Contrarily to zfs_get_data we need not re-check
1375 		 * blocksize after we get the lock because it cannot be changed.
1376 		 */
1377 		size = zv->zv_volblocksize;
1378 		offset = P2ALIGN(offset, size);
1379 		zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1380 		    RL_READER);
1381 		error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1382 		    DMU_READ_NO_PREFETCH);
1383 		if (error == 0) {
1384 			blkptr_t *bp = &lr->lr_blkptr;
1385 
1386 			zgd->zgd_db = db;
1387 			zgd->zgd_bp = bp;
1388 
1389 			ASSERT(db->db_offset == offset);
1390 			ASSERT(db->db_size == size);
1391 
1392 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1393 			    zvol_get_done, zgd);
1394 
1395 			if (error == 0)
1396 				return (0);
1397 		}
1398 	}
1399 
1400 	zvol_get_done(zgd, error);
1401 
1402 	return (error);
1403 }
1404 
1405 /*
1406  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1407  *
1408  * We store data in the log buffers if it's small enough.
1409  * Otherwise we will later flush the data out via dmu_sync().
1410  */
1411 ssize_t zvol_immediate_write_sz = 32768;
1412 #ifdef _KERNEL
1413 SYSCTL_LONG(_vfs_zfs_vol, OID_AUTO, immediate_write_sz, CTLFLAG_RWTUN,
1414     &zvol_immediate_write_sz, 0, "Minimal size for indirect log write");
1415 #endif
1416 
1417 static void
1418 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1419     boolean_t sync)
1420 {
1421 	uint32_t blocksize = zv->zv_volblocksize;
1422 	zilog_t *zilog = zv->zv_zilog;
1423 	itx_wr_state_t write_state;
1424 
1425 	if (zil_replaying(zilog, tx))
1426 		return;
1427 
1428 	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1429 		write_state = WR_INDIRECT;
1430 	else if (!spa_has_slogs(zilog->zl_spa) &&
1431 	    resid >= blocksize && blocksize > zvol_immediate_write_sz)
1432 		write_state = WR_INDIRECT;
1433 	else if (sync)
1434 		write_state = WR_COPIED;
1435 	else
1436 		write_state = WR_NEED_COPY;
1437 
1438 	while (resid) {
1439 		itx_t *itx;
1440 		lr_write_t *lr;
1441 		itx_wr_state_t wr_state = write_state;
1442 		ssize_t len = resid;
1443 
1444 		if (wr_state == WR_COPIED && resid > zil_max_copied_data(zilog))
1445 			wr_state = WR_NEED_COPY;
1446 		else if (wr_state == WR_INDIRECT)
1447 			len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1448 
1449 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1450 		    (wr_state == WR_COPIED ? len : 0));
1451 		lr = (lr_write_t *)&itx->itx_lr;
1452 		if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1453 		    off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1454 			zil_itx_destroy(itx);
1455 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1456 			lr = (lr_write_t *)&itx->itx_lr;
1457 			wr_state = WR_NEED_COPY;
1458 		}
1459 
1460 		itx->itx_wr_state = wr_state;
1461 		lr->lr_foid = ZVOL_OBJ;
1462 		lr->lr_offset = off;
1463 		lr->lr_length = len;
1464 		lr->lr_blkoff = 0;
1465 		BP_ZERO(&lr->lr_blkptr);
1466 
1467 		itx->itx_private = zv;
1468 
1469 		if (!sync && (zv->zv_sync_cnt == 0))
1470 			itx->itx_sync = B_FALSE;
1471 
1472 		zil_itx_assign(zilog, itx, tx);
1473 
1474 		off += len;
1475 		resid -= len;
1476 	}
1477 }
1478 
1479 #ifdef illumos
1480 static int
1481 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1482     uint64_t size, boolean_t doread, boolean_t isdump)
1483 {
1484 	vdev_disk_t *dvd;
1485 	int c;
1486 	int numerrors = 0;
1487 
1488 	if (vd->vdev_ops == &vdev_mirror_ops ||
1489 	    vd->vdev_ops == &vdev_replacing_ops ||
1490 	    vd->vdev_ops == &vdev_spare_ops) {
1491 		for (c = 0; c < vd->vdev_children; c++) {
1492 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1493 			    addr, offset, origoffset, size, doread, isdump);
1494 			if (err != 0) {
1495 				numerrors++;
1496 			} else if (doread) {
1497 				break;
1498 			}
1499 		}
1500 	}
1501 
1502 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1503 		return (numerrors < vd->vdev_children ? 0 : EIO);
1504 
1505 	if (doread && !vdev_readable(vd))
1506 		return (SET_ERROR(EIO));
1507 	else if (!doread && !vdev_writeable(vd))
1508 		return (SET_ERROR(EIO));
1509 
1510 	if (vd->vdev_ops == &vdev_raidz_ops) {
1511 		return (vdev_raidz_physio(vd,
1512 		    addr, size, offset, origoffset, doread, isdump));
1513 	}
1514 
1515 	offset += VDEV_LABEL_START_SIZE;
1516 
1517 	if (ddi_in_panic() || isdump) {
1518 		ASSERT(!doread);
1519 		if (doread)
1520 			return (SET_ERROR(EIO));
1521 		dvd = vd->vdev_tsd;
1522 		ASSERT3P(dvd, !=, NULL);
1523 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1524 		    lbtodb(size)));
1525 	} else {
1526 		dvd = vd->vdev_tsd;
1527 		ASSERT3P(dvd, !=, NULL);
1528 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1529 		    offset, doread ? B_READ : B_WRITE));
1530 	}
1531 }
1532 
1533 static int
1534 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1535     boolean_t doread, boolean_t isdump)
1536 {
1537 	vdev_t *vd;
1538 	int error;
1539 	zvol_extent_t *ze;
1540 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1541 
1542 	/* Must be sector aligned, and not stradle a block boundary. */
1543 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1544 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1545 		return (SET_ERROR(EINVAL));
1546 	}
1547 	ASSERT(size <= zv->zv_volblocksize);
1548 
1549 	/* Locate the extent this belongs to */
1550 	ze = list_head(&zv->zv_extents);
1551 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1552 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1553 		ze = list_next(&zv->zv_extents, ze);
1554 	}
1555 
1556 	if (ze == NULL)
1557 		return (SET_ERROR(EINVAL));
1558 
1559 	if (!ddi_in_panic())
1560 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1561 
1562 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1563 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1564 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1565 	    size, doread, isdump);
1566 
1567 	if (!ddi_in_panic())
1568 		spa_config_exit(spa, SCL_STATE, FTAG);
1569 
1570 	return (error);
1571 }
1572 
1573 int
1574 zvol_strategy(buf_t *bp)
1575 {
1576 	zfs_soft_state_t *zs = NULL;
1577 #else	/* !illumos */
1578 void
1579 zvol_strategy(struct bio *bp)
1580 {
1581 #endif	/* illumos */
1582 	zvol_state_t *zv;
1583 	uint64_t off, volsize;
1584 	size_t resid;
1585 	char *addr;
1586 	objset_t *os;
1587 	rl_t *rl;
1588 	int error = 0;
1589 #ifdef illumos
1590 	boolean_t doread = bp->b_flags & B_READ;
1591 #else
1592 	boolean_t doread = 0;
1593 #endif
1594 	boolean_t is_dumpified;
1595 	boolean_t sync;
1596 
1597 #ifdef illumos
1598 	if (getminor(bp->b_edev) == 0) {
1599 		error = SET_ERROR(EINVAL);
1600 	} else {
1601 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1602 		if (zs == NULL)
1603 			error = SET_ERROR(ENXIO);
1604 		else if (zs->zss_type != ZSST_ZVOL)
1605 			error = SET_ERROR(EINVAL);
1606 	}
1607 
1608 	if (error) {
1609 		bioerror(bp, error);
1610 		biodone(bp);
1611 		return (0);
1612 	}
1613 
1614 	zv = zs->zss_data;
1615 
1616 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1617 		bioerror(bp, EROFS);
1618 		biodone(bp);
1619 		return (0);
1620 	}
1621 
1622 	off = ldbtob(bp->b_blkno);
1623 #else	/* !illumos */
1624 	if (bp->bio_to)
1625 		zv = bp->bio_to->private;
1626 	else
1627 		zv = bp->bio_dev->si_drv2;
1628 
1629 	if (zv == NULL) {
1630 		error = SET_ERROR(ENXIO);
1631 		goto out;
1632 	}
1633 
1634 	if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1635 		error = SET_ERROR(EROFS);
1636 		goto out;
1637 	}
1638 
1639 	switch (bp->bio_cmd) {
1640 	case BIO_FLUSH:
1641 		goto sync;
1642 	case BIO_READ:
1643 		doread = 1;
1644 	case BIO_WRITE:
1645 	case BIO_DELETE:
1646 		break;
1647 	default:
1648 		error = EOPNOTSUPP;
1649 		goto out;
1650 	}
1651 
1652 	off = bp->bio_offset;
1653 #endif	/* illumos */
1654 	volsize = zv->zv_volsize;
1655 
1656 	os = zv->zv_objset;
1657 	ASSERT(os != NULL);
1658 
1659 #ifdef illumos
1660 	bp_mapin(bp);
1661 	addr = bp->b_un.b_addr;
1662 	resid = bp->b_bcount;
1663 
1664 	if (resid > 0 && (off < 0 || off >= volsize)) {
1665 		bioerror(bp, EIO);
1666 		biodone(bp);
1667 		return (0);
1668 	}
1669 
1670 	is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1671 	sync = ((!(bp->b_flags & B_ASYNC) &&
1672 	    !(zv->zv_flags & ZVOL_WCE)) ||
1673 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1674 	    !doread && !is_dumpified;
1675 #else	/* !illumos */
1676 	addr = bp->bio_data;
1677 	resid = bp->bio_length;
1678 
1679 	if (resid > 0 && (off < 0 || off >= volsize)) {
1680 		error = SET_ERROR(EIO);
1681 		goto out;
1682 	}
1683 
1684 	is_dumpified = B_FALSE;
1685 	sync = !doread && !is_dumpified &&
1686 	    zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1687 #endif	/* illumos */
1688 
1689 	/*
1690 	 * There must be no buffer changes when doing a dmu_sync() because
1691 	 * we can't change the data whilst calculating the checksum.
1692 	 */
1693 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1694 	    doread ? RL_READER : RL_WRITER);
1695 
1696 #ifndef illumos
1697 	if (bp->bio_cmd == BIO_DELETE) {
1698 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1699 		error = dmu_tx_assign(tx, TXG_WAIT);
1700 		if (error != 0) {
1701 			dmu_tx_abort(tx);
1702 		} else {
1703 			zvol_log_truncate(zv, tx, off, resid, sync);
1704 			dmu_tx_commit(tx);
1705 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1706 			    off, resid);
1707 			resid = 0;
1708 		}
1709 		goto unlock;
1710 	}
1711 #endif
1712 	while (resid != 0 && off < volsize) {
1713 		size_t size = MIN(resid, zvol_maxphys);
1714 #ifdef illumos
1715 		if (is_dumpified) {
1716 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1717 			error = zvol_dumpio(zv, addr, off, size,
1718 			    doread, B_FALSE);
1719 		} else if (doread) {
1720 #else
1721 		if (doread) {
1722 #endif
1723 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1724 			    DMU_READ_PREFETCH);
1725 		} else {
1726 			dmu_tx_t *tx = dmu_tx_create(os);
1727 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1728 			error = dmu_tx_assign(tx, TXG_WAIT);
1729 			if (error) {
1730 				dmu_tx_abort(tx);
1731 			} else {
1732 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1733 				zvol_log_write(zv, tx, off, size, sync);
1734 				dmu_tx_commit(tx);
1735 			}
1736 		}
1737 		if (error) {
1738 			/* convert checksum errors into IO errors */
1739 			if (error == ECKSUM)
1740 				error = SET_ERROR(EIO);
1741 			break;
1742 		}
1743 		off += size;
1744 		addr += size;
1745 		resid -= size;
1746 	}
1747 #ifndef illumos
1748 unlock:
1749 #endif
1750 	zfs_range_unlock(rl);
1751 
1752 #ifdef illumos
1753 	if ((bp->b_resid = resid) == bp->b_bcount)
1754 		bioerror(bp, off > volsize ? EINVAL : error);
1755 
1756 	if (sync)
1757 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1758 	biodone(bp);
1759 
1760 	return (0);
1761 #else	/* !illumos */
1762 	bp->bio_completed = bp->bio_length - resid;
1763 	if (bp->bio_completed < bp->bio_length && off > volsize)
1764 		error = EINVAL;
1765 
1766 	if (sync) {
1767 sync:
1768 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1769 	}
1770 out:
1771 	if (bp->bio_to)
1772 		g_io_deliver(bp, error);
1773 	else
1774 		biofinish(bp, NULL, error);
1775 #endif	/* illumos */
1776 }
1777 
1778 #ifdef illumos
1779 /*
1780  * Set the buffer count to the zvol maximum transfer.
1781  * Using our own routine instead of the default minphys()
1782  * means that for larger writes we write bigger buffers on X86
1783  * (128K instead of 56K) and flush the disk write cache less often
1784  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1785  * 56K on X86 and 128K on sparc).
1786  */
1787 void
1788 zvol_minphys(struct buf *bp)
1789 {
1790 	if (bp->b_bcount > zvol_maxphys)
1791 		bp->b_bcount = zvol_maxphys;
1792 }
1793 
1794 int
1795 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1796 {
1797 	minor_t minor = getminor(dev);
1798 	zvol_state_t *zv;
1799 	int error = 0;
1800 	uint64_t size;
1801 	uint64_t boff;
1802 	uint64_t resid;
1803 
1804 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1805 	if (zv == NULL)
1806 		return (SET_ERROR(ENXIO));
1807 
1808 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1809 		return (SET_ERROR(EINVAL));
1810 
1811 	boff = ldbtob(blkno);
1812 	resid = ldbtob(nblocks);
1813 
1814 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1815 
1816 	while (resid) {
1817 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1818 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1819 		if (error)
1820 			break;
1821 		boff += size;
1822 		addr += size;
1823 		resid -= size;
1824 	}
1825 
1826 	return (error);
1827 }
1828 
1829 /*ARGSUSED*/
1830 int
1831 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1832 {
1833 	minor_t minor = getminor(dev);
1834 #else	/* !illumos */
1835 int
1836 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1837 {
1838 #endif	/* illumos */
1839 	zvol_state_t *zv;
1840 	uint64_t volsize;
1841 	rl_t *rl;
1842 	int error = 0;
1843 
1844 #ifdef illumos
1845 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1846 	if (zv == NULL)
1847 		return (SET_ERROR(ENXIO));
1848 #else
1849 	zv = dev->si_drv2;
1850 #endif
1851 
1852 	volsize = zv->zv_volsize;
1853 	/* uio_loffset == volsize isn't an error as its required for EOF processing. */
1854 	if (uio->uio_resid > 0 &&
1855 	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1856 		return (SET_ERROR(EIO));
1857 
1858 #ifdef illumos
1859 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1860 		error = physio(zvol_strategy, NULL, dev, B_READ,
1861 		    zvol_minphys, uio);
1862 		return (error);
1863 	}
1864 #endif
1865 
1866 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1867 	    RL_READER);
1868 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1869 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1870 
1871 		/* don't read past the end */
1872 		if (bytes > volsize - uio->uio_loffset)
1873 			bytes = volsize - uio->uio_loffset;
1874 
1875 		error =  dmu_read_uio_dnode(zv->zv_dn, uio, bytes);
1876 		if (error) {
1877 			/* convert checksum errors into IO errors */
1878 			if (error == ECKSUM)
1879 				error = SET_ERROR(EIO);
1880 			break;
1881 		}
1882 	}
1883 	zfs_range_unlock(rl);
1884 	return (error);
1885 }
1886 
1887 #ifdef illumos
1888 /*ARGSUSED*/
1889 int
1890 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1891 {
1892 	minor_t minor = getminor(dev);
1893 #else	/* !illumos */
1894 int
1895 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1896 {
1897 #endif	/* illumos */
1898 	zvol_state_t *zv;
1899 	uint64_t volsize;
1900 	rl_t *rl;
1901 	int error = 0;
1902 	boolean_t sync;
1903 
1904 #ifdef illumos
1905 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1906 	if (zv == NULL)
1907 		return (SET_ERROR(ENXIO));
1908 #else
1909 	zv = dev->si_drv2;
1910 #endif
1911 
1912 	volsize = zv->zv_volsize;
1913 	/* uio_loffset == volsize isn't an error as its required for EOF processing. */
1914 	if (uio->uio_resid > 0 &&
1915 	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1916 		return (SET_ERROR(EIO));
1917 
1918 #ifdef illumos
1919 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1920 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1921 		    zvol_minphys, uio);
1922 		return (error);
1923 	}
1924 
1925 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1926 #else
1927 	sync = (ioflag & IO_SYNC) ||
1928 #endif
1929 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1930 
1931 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1932 	    RL_WRITER);
1933 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1934 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1935 		uint64_t off = uio->uio_loffset;
1936 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1937 
1938 		if (bytes > volsize - off)	/* don't write past the end */
1939 			bytes = volsize - off;
1940 
1941 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1942 		error = dmu_tx_assign(tx, TXG_WAIT);
1943 		if (error) {
1944 			dmu_tx_abort(tx);
1945 			break;
1946 		}
1947 		error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1948 		if (error == 0)
1949 			zvol_log_write(zv, tx, off, bytes, sync);
1950 		dmu_tx_commit(tx);
1951 
1952 		if (error)
1953 			break;
1954 	}
1955 	zfs_range_unlock(rl);
1956 	if (sync)
1957 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1958 	return (error);
1959 }
1960 
1961 #ifdef illumos
1962 int
1963 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1964 {
1965 	struct uuid uuid = EFI_RESERVED;
1966 	efi_gpe_t gpe = { 0 };
1967 	uint32_t crc;
1968 	dk_efi_t efi;
1969 	int length;
1970 	char *ptr;
1971 
1972 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1973 		return (SET_ERROR(EFAULT));
1974 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1975 	length = efi.dki_length;
1976 	/*
1977 	 * Some clients may attempt to request a PMBR for the
1978 	 * zvol.  Currently this interface will return EINVAL to
1979 	 * such requests.  These requests could be supported by
1980 	 * adding a check for lba == 0 and consing up an appropriate
1981 	 * PMBR.
1982 	 */
1983 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1984 		return (SET_ERROR(EINVAL));
1985 
1986 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1987 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1988 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1989 
1990 	if (efi.dki_lba == 1) {
1991 		efi_gpt_t gpt = { 0 };
1992 
1993 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1994 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1995 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1996 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1997 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1998 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1999 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
2000 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
2001 		gpt.efi_gpt_SizeOfPartitionEntry =
2002 		    LE_32(sizeof (efi_gpe_t));
2003 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
2004 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
2005 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
2006 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
2007 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
2008 		    flag))
2009 			return (SET_ERROR(EFAULT));
2010 		ptr += sizeof (gpt);
2011 		length -= sizeof (gpt);
2012 	}
2013 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
2014 	    length), flag))
2015 		return (SET_ERROR(EFAULT));
2016 	return (0);
2017 }
2018 
2019 /*
2020  * BEGIN entry points to allow external callers access to the volume.
2021  */
2022 /*
2023  * Return the volume parameters needed for access from an external caller.
2024  * These values are invariant as long as the volume is held open.
2025  */
2026 int
2027 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
2028     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
2029     void **rl_hdl, void **dnode_hdl)
2030 {
2031 	zvol_state_t *zv;
2032 
2033 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2034 	if (zv == NULL)
2035 		return (SET_ERROR(ENXIO));
2036 	if (zv->zv_flags & ZVOL_DUMPIFIED)
2037 		return (SET_ERROR(ENXIO));
2038 
2039 	ASSERT(blksize && max_xfer_len && minor_hdl &&
2040 	    objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
2041 
2042 	*blksize = zv->zv_volblocksize;
2043 	*max_xfer_len = (uint64_t)zvol_maxphys;
2044 	*minor_hdl = zv;
2045 	*objset_hdl = zv->zv_objset;
2046 	*zil_hdl = zv->zv_zilog;
2047 	*rl_hdl = &zv->zv_znode;
2048 	*dnode_hdl = zv->zv_dn;
2049 	return (0);
2050 }
2051 
2052 /*
2053  * Return the current volume size to an external caller.
2054  * The size can change while the volume is open.
2055  */
2056 uint64_t
2057 zvol_get_volume_size(void *minor_hdl)
2058 {
2059 	zvol_state_t *zv = minor_hdl;
2060 
2061 	return (zv->zv_volsize);
2062 }
2063 
2064 /*
2065  * Return the current WCE setting to an external caller.
2066  * The WCE setting can change while the volume is open.
2067  */
2068 int
2069 zvol_get_volume_wce(void *minor_hdl)
2070 {
2071 	zvol_state_t *zv = minor_hdl;
2072 
2073 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2074 }
2075 
2076 /*
2077  * Entry point for external callers to zvol_log_write
2078  */
2079 void
2080 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2081     boolean_t sync)
2082 {
2083 	zvol_state_t *zv = minor_hdl;
2084 
2085 	zvol_log_write(zv, tx, off, resid, sync);
2086 }
2087 /*
2088  * END entry points to allow external callers access to the volume.
2089  */
2090 #endif	/* illumos */
2091 
2092 /*
2093  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2094  */
2095 static void
2096 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2097     boolean_t sync)
2098 {
2099 	itx_t *itx;
2100 	lr_truncate_t *lr;
2101 	zilog_t *zilog = zv->zv_zilog;
2102 
2103 	if (zil_replaying(zilog, tx))
2104 		return;
2105 
2106 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2107 	lr = (lr_truncate_t *)&itx->itx_lr;
2108 	lr->lr_foid = ZVOL_OBJ;
2109 	lr->lr_offset = off;
2110 	lr->lr_length = len;
2111 
2112 	itx->itx_sync = (sync || zv->zv_sync_cnt != 0);
2113 	zil_itx_assign(zilog, itx, tx);
2114 }
2115 
2116 #ifdef illumos
2117 /*
2118  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
2119  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2120  */
2121 /*ARGSUSED*/
2122 int
2123 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2124 {
2125 	zvol_state_t *zv;
2126 	struct dk_callback *dkc;
2127 	int error = 0;
2128 	rl_t *rl;
2129 
2130 	mutex_enter(&zfsdev_state_lock);
2131 
2132 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2133 
2134 	if (zv == NULL) {
2135 		mutex_exit(&zfsdev_state_lock);
2136 		return (SET_ERROR(ENXIO));
2137 	}
2138 	ASSERT(zv->zv_total_opens > 0);
2139 
2140 	switch (cmd) {
2141 
2142 	case DKIOCINFO:
2143 	{
2144 		struct dk_cinfo dki;
2145 
2146 		bzero(&dki, sizeof (dki));
2147 		(void) strcpy(dki.dki_cname, "zvol");
2148 		(void) strcpy(dki.dki_dname, "zvol");
2149 		dki.dki_ctype = DKC_UNKNOWN;
2150 		dki.dki_unit = getminor(dev);
2151 		dki.dki_maxtransfer =
2152 		    1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2153 		mutex_exit(&zfsdev_state_lock);
2154 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2155 			error = SET_ERROR(EFAULT);
2156 		return (error);
2157 	}
2158 
2159 	case DKIOCGMEDIAINFO:
2160 	{
2161 		struct dk_minfo dkm;
2162 
2163 		bzero(&dkm, sizeof (dkm));
2164 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
2165 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2166 		dkm.dki_media_type = DK_UNKNOWN;
2167 		mutex_exit(&zfsdev_state_lock);
2168 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2169 			error = SET_ERROR(EFAULT);
2170 		return (error);
2171 	}
2172 
2173 	case DKIOCGMEDIAINFOEXT:
2174 	{
2175 		struct dk_minfo_ext dkmext;
2176 
2177 		bzero(&dkmext, sizeof (dkmext));
2178 		dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2179 		dkmext.dki_pbsize = zv->zv_volblocksize;
2180 		dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2181 		dkmext.dki_media_type = DK_UNKNOWN;
2182 		mutex_exit(&zfsdev_state_lock);
2183 		if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2184 			error = SET_ERROR(EFAULT);
2185 		return (error);
2186 	}
2187 
2188 	case DKIOCGETEFI:
2189 	{
2190 		uint64_t vs = zv->zv_volsize;
2191 		uint8_t bs = zv->zv_min_bs;
2192 
2193 		mutex_exit(&zfsdev_state_lock);
2194 		error = zvol_getefi((void *)arg, flag, vs, bs);
2195 		return (error);
2196 	}
2197 
2198 	case DKIOCFLUSHWRITECACHE:
2199 		dkc = (struct dk_callback *)arg;
2200 		mutex_exit(&zfsdev_state_lock);
2201 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
2202 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2203 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
2204 			error = 0;
2205 		}
2206 		return (error);
2207 
2208 	case DKIOCGETWCE:
2209 	{
2210 		int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2211 		if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2212 		    flag))
2213 			error = SET_ERROR(EFAULT);
2214 		break;
2215 	}
2216 	case DKIOCSETWCE:
2217 	{
2218 		int wce;
2219 		if (ddi_copyin((void *)arg, &wce, sizeof (int),
2220 		    flag)) {
2221 			error = SET_ERROR(EFAULT);
2222 			break;
2223 		}
2224 		if (wce) {
2225 			zv->zv_flags |= ZVOL_WCE;
2226 			mutex_exit(&zfsdev_state_lock);
2227 		} else {
2228 			zv->zv_flags &= ~ZVOL_WCE;
2229 			mutex_exit(&zfsdev_state_lock);
2230 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
2231 		}
2232 		return (0);
2233 	}
2234 
2235 	case DKIOCGGEOM:
2236 	case DKIOCGVTOC:
2237 		/*
2238 		 * commands using these (like prtvtoc) expect ENOTSUP
2239 		 * since we're emulating an EFI label
2240 		 */
2241 		error = SET_ERROR(ENOTSUP);
2242 		break;
2243 
2244 	case DKIOCDUMPINIT:
2245 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2246 		    RL_WRITER);
2247 		error = zvol_dumpify(zv);
2248 		zfs_range_unlock(rl);
2249 		break;
2250 
2251 	case DKIOCDUMPFINI:
2252 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2253 			break;
2254 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2255 		    RL_WRITER);
2256 		error = zvol_dump_fini(zv);
2257 		zfs_range_unlock(rl);
2258 		break;
2259 
2260 	case DKIOCFREE:
2261 	{
2262 		dkioc_free_t df;
2263 		dmu_tx_t *tx;
2264 
2265 		if (!zvol_unmap_enabled)
2266 			break;
2267 
2268 		if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2269 			error = SET_ERROR(EFAULT);
2270 			break;
2271 		}
2272 
2273 		/*
2274 		 * Apply Postel's Law to length-checking.  If they overshoot,
2275 		 * just blank out until the end, if there's a need to blank
2276 		 * out anything.
2277 		 */
2278 		if (df.df_start >= zv->zv_volsize)
2279 			break;	/* No need to do anything... */
2280 
2281 		mutex_exit(&zfsdev_state_lock);
2282 
2283 		rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2284 		    RL_WRITER);
2285 		tx = dmu_tx_create(zv->zv_objset);
2286 		dmu_tx_mark_netfree(tx);
2287 		error = dmu_tx_assign(tx, TXG_WAIT);
2288 		if (error != 0) {
2289 			dmu_tx_abort(tx);
2290 		} else {
2291 			zvol_log_truncate(zv, tx, df.df_start,
2292 			    df.df_length, B_TRUE);
2293 			dmu_tx_commit(tx);
2294 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2295 			    df.df_start, df.df_length);
2296 		}
2297 
2298 		zfs_range_unlock(rl);
2299 
2300 		/*
2301 		 * If the write-cache is disabled, 'sync' property
2302 		 * is set to 'always', or if the caller is asking for
2303 		 * a synchronous free, commit this operation to the zil.
2304 		 * This will sync any previous uncommitted writes to the
2305 		 * zvol object.
2306 		 * Can be overridden by the zvol_unmap_sync_enabled tunable.
2307 		 */
2308 		if ((error == 0) && zvol_unmap_sync_enabled &&
2309 		    (!(zv->zv_flags & ZVOL_WCE) ||
2310 		    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
2311 		    (df.df_flags & DF_WAIT_SYNC))) {
2312 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
2313 		}
2314 
2315 		return (error);
2316 	}
2317 
2318 	default:
2319 		error = SET_ERROR(ENOTTY);
2320 		break;
2321 
2322 	}
2323 	mutex_exit(&zfsdev_state_lock);
2324 	return (error);
2325 }
2326 #endif	/* illumos */
2327 
2328 int
2329 zvol_busy(void)
2330 {
2331 	return (zvol_minors != 0);
2332 }
2333 
2334 void
2335 zvol_init(void)
2336 {
2337 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2338 	    1) == 0);
2339 #ifdef illumos
2340 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2341 #else
2342 	ZFS_LOG(1, "ZVOL Initialized.");
2343 #endif
2344 }
2345 
2346 void
2347 zvol_fini(void)
2348 {
2349 #ifdef illumos
2350 	mutex_destroy(&zfsdev_state_lock);
2351 #endif
2352 	ddi_soft_state_fini(&zfsdev_state);
2353 	ZFS_LOG(1, "ZVOL Deinitialized.");
2354 }
2355 
2356 #ifdef illumos
2357 /*ARGSUSED*/
2358 static int
2359 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2360 {
2361 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2362 
2363 	if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2364 		return (1);
2365 	return (0);
2366 }
2367 
2368 /*ARGSUSED*/
2369 static void
2370 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2371 {
2372 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2373 
2374 	spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2375 }
2376 
2377 static int
2378 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2379 {
2380 	dmu_tx_t *tx;
2381 	int error;
2382 	objset_t *os = zv->zv_objset;
2383 	spa_t *spa = dmu_objset_spa(os);
2384 	vdev_t *vd = spa->spa_root_vdev;
2385 	nvlist_t *nv = NULL;
2386 	uint64_t version = spa_version(spa);
2387 	uint64_t checksum, compress, refresrv, vbs, dedup;
2388 
2389 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2390 	ASSERT(vd->vdev_ops == &vdev_root_ops);
2391 
2392 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2393 	    DMU_OBJECT_END);
2394 	if (error != 0)
2395 		return (error);
2396 	/* wait for dmu_free_long_range to actually free the blocks */
2397 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2398 
2399 	/*
2400 	 * If the pool on which the dump device is being initialized has more
2401 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2402 	 * enabled.  If so, bump that feature's counter to indicate that the
2403 	 * feature is active. We also check the vdev type to handle the
2404 	 * following case:
2405 	 *   # zpool create test raidz disk1 disk2 disk3
2406 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2407 	 *   the raidz vdev itself has 3 children.
2408 	 */
2409 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2410 		if (!spa_feature_is_enabled(spa,
2411 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2412 			return (SET_ERROR(ENOTSUP));
2413 		(void) dsl_sync_task(spa_name(spa),
2414 		    zfs_mvdev_dump_feature_check,
2415 		    zfs_mvdev_dump_activate_feature_sync, NULL,
2416 		    2, ZFS_SPACE_CHECK_RESERVED);
2417 	}
2418 
2419 	if (!resize) {
2420 		error = dsl_prop_get_integer(zv->zv_name,
2421 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2422 		if (error == 0) {
2423 			error = dsl_prop_get_integer(zv->zv_name,
2424 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2425 			    NULL);
2426 		}
2427 		if (error == 0) {
2428 			error = dsl_prop_get_integer(zv->zv_name,
2429 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2430 			    &refresrv, NULL);
2431 		}
2432 		if (error == 0) {
2433 			error = dsl_prop_get_integer(zv->zv_name,
2434 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2435 			    NULL);
2436 		}
2437 		if (version >= SPA_VERSION_DEDUP && error == 0) {
2438 			error = dsl_prop_get_integer(zv->zv_name,
2439 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2440 		}
2441 	}
2442 	if (error != 0)
2443 		return (error);
2444 
2445 	tx = dmu_tx_create(os);
2446 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2447 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2448 	error = dmu_tx_assign(tx, TXG_WAIT);
2449 	if (error != 0) {
2450 		dmu_tx_abort(tx);
2451 		return (error);
2452 	}
2453 
2454 	/*
2455 	 * If we are resizing the dump device then we only need to
2456 	 * update the refreservation to match the newly updated
2457 	 * zvolsize. Otherwise, we save off the original state of the
2458 	 * zvol so that we can restore them if the zvol is ever undumpified.
2459 	 */
2460 	if (resize) {
2461 		error = zap_update(os, ZVOL_ZAP_OBJ,
2462 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2463 		    &zv->zv_volsize, tx);
2464 	} else {
2465 		error = zap_update(os, ZVOL_ZAP_OBJ,
2466 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2467 		    &compress, tx);
2468 		if (error == 0) {
2469 			error = zap_update(os, ZVOL_ZAP_OBJ,
2470 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2471 			    &checksum, tx);
2472 		}
2473 		if (error == 0) {
2474 			error = zap_update(os, ZVOL_ZAP_OBJ,
2475 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2476 			    &refresrv, tx);
2477 		}
2478 		if (error == 0) {
2479 			error = zap_update(os, ZVOL_ZAP_OBJ,
2480 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2481 			    &vbs, tx);
2482 		}
2483 		if (error == 0) {
2484 			error = dmu_object_set_blocksize(
2485 			    os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2486 		}
2487 		if (version >= SPA_VERSION_DEDUP && error == 0) {
2488 			error = zap_update(os, ZVOL_ZAP_OBJ,
2489 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2490 			    &dedup, tx);
2491 		}
2492 		if (error == 0)
2493 			zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2494 	}
2495 	dmu_tx_commit(tx);
2496 
2497 	/*
2498 	 * We only need update the zvol's property if we are initializing
2499 	 * the dump area for the first time.
2500 	 */
2501 	if (error == 0 && !resize) {
2502 		/*
2503 		 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2504 		 * function.  Otherwise, use the old default -- OFF.
2505 		 */
2506 		checksum = spa_feature_is_active(spa,
2507 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2508 		    ZIO_CHECKSUM_OFF;
2509 
2510 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2511 		VERIFY(nvlist_add_uint64(nv,
2512 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2513 		VERIFY(nvlist_add_uint64(nv,
2514 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2515 		    ZIO_COMPRESS_OFF) == 0);
2516 		VERIFY(nvlist_add_uint64(nv,
2517 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2518 		    checksum) == 0);
2519 		if (version >= SPA_VERSION_DEDUP) {
2520 			VERIFY(nvlist_add_uint64(nv,
2521 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
2522 			    ZIO_CHECKSUM_OFF) == 0);
2523 		}
2524 
2525 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2526 		    nv, NULL);
2527 		nvlist_free(nv);
2528 	}
2529 
2530 	/* Allocate the space for the dump */
2531 	if (error == 0)
2532 		error = zvol_prealloc(zv);
2533 	return (error);
2534 }
2535 
2536 static int
2537 zvol_dumpify(zvol_state_t *zv)
2538 {
2539 	int error = 0;
2540 	uint64_t dumpsize = 0;
2541 	dmu_tx_t *tx;
2542 	objset_t *os = zv->zv_objset;
2543 
2544 	if (zv->zv_flags & ZVOL_RDONLY)
2545 		return (SET_ERROR(EROFS));
2546 
2547 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2548 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2549 		boolean_t resize = (dumpsize > 0);
2550 
2551 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2552 			(void) zvol_dump_fini(zv);
2553 			return (error);
2554 		}
2555 	}
2556 
2557 	/*
2558 	 * Build up our lba mapping.
2559 	 */
2560 	error = zvol_get_lbas(zv);
2561 	if (error) {
2562 		(void) zvol_dump_fini(zv);
2563 		return (error);
2564 	}
2565 
2566 	tx = dmu_tx_create(os);
2567 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2568 	error = dmu_tx_assign(tx, TXG_WAIT);
2569 	if (error) {
2570 		dmu_tx_abort(tx);
2571 		(void) zvol_dump_fini(zv);
2572 		return (error);
2573 	}
2574 
2575 	zv->zv_flags |= ZVOL_DUMPIFIED;
2576 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2577 	    &zv->zv_volsize, tx);
2578 	dmu_tx_commit(tx);
2579 
2580 	if (error) {
2581 		(void) zvol_dump_fini(zv);
2582 		return (error);
2583 	}
2584 
2585 	txg_wait_synced(dmu_objset_pool(os), 0);
2586 	return (0);
2587 }
2588 
2589 static int
2590 zvol_dump_fini(zvol_state_t *zv)
2591 {
2592 	dmu_tx_t *tx;
2593 	objset_t *os = zv->zv_objset;
2594 	nvlist_t *nv;
2595 	int error = 0;
2596 	uint64_t checksum, compress, refresrv, vbs, dedup;
2597 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2598 
2599 	/*
2600 	 * Attempt to restore the zvol back to its pre-dumpified state.
2601 	 * This is a best-effort attempt as it's possible that not all
2602 	 * of these properties were initialized during the dumpify process
2603 	 * (i.e. error during zvol_dump_init).
2604 	 */
2605 
2606 	tx = dmu_tx_create(os);
2607 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2608 	error = dmu_tx_assign(tx, TXG_WAIT);
2609 	if (error) {
2610 		dmu_tx_abort(tx);
2611 		return (error);
2612 	}
2613 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2614 	dmu_tx_commit(tx);
2615 
2616 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2617 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2618 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2619 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2620 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2621 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2622 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2623 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2624 
2625 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2626 	(void) nvlist_add_uint64(nv,
2627 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2628 	(void) nvlist_add_uint64(nv,
2629 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2630 	(void) nvlist_add_uint64(nv,
2631 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2632 	if (version >= SPA_VERSION_DEDUP &&
2633 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2634 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2635 		(void) nvlist_add_uint64(nv,
2636 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2637 	}
2638 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2639 	    nv, NULL);
2640 	nvlist_free(nv);
2641 
2642 	zvol_free_extents(zv);
2643 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2644 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2645 	/* wait for dmu_free_long_range to actually free the blocks */
2646 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2647 	tx = dmu_tx_create(os);
2648 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2649 	error = dmu_tx_assign(tx, TXG_WAIT);
2650 	if (error) {
2651 		dmu_tx_abort(tx);
2652 		return (error);
2653 	}
2654 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2655 		zv->zv_volblocksize = vbs;
2656 	dmu_tx_commit(tx);
2657 
2658 	return (0);
2659 }
2660 #else	/* !illumos */
2661 
2662 static void
2663 zvol_geom_run(zvol_state_t *zv)
2664 {
2665 	struct g_provider *pp;
2666 
2667 	pp = zv->zv_provider;
2668 	g_error_provider(pp, 0);
2669 
2670 	kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2671 	    "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2672 }
2673 
2674 static void
2675 zvol_geom_destroy(zvol_state_t *zv)
2676 {
2677 	struct g_provider *pp;
2678 
2679 	g_topology_assert();
2680 
2681 	mtx_lock(&zv->zv_queue_mtx);
2682 	zv->zv_state = 1;
2683 	wakeup_one(&zv->zv_queue);
2684 	while (zv->zv_state != 2)
2685 		msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2686 	mtx_destroy(&zv->zv_queue_mtx);
2687 
2688 	pp = zv->zv_provider;
2689 	zv->zv_provider = NULL;
2690 	pp->private = NULL;
2691 	g_wither_geom(pp->geom, ENXIO);
2692 }
2693 
2694 static int
2695 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2696 {
2697 	int count, error, flags;
2698 
2699 	g_topology_assert();
2700 
2701 	/*
2702 	 * To make it easier we expect either open or close, but not both
2703 	 * at the same time.
2704 	 */
2705 	KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2706 	    (acr <= 0 && acw <= 0 && ace <= 0),
2707 	    ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2708 	    pp->name, acr, acw, ace));
2709 
2710 	if (pp->private == NULL) {
2711 		if (acr <= 0 && acw <= 0 && ace <= 0)
2712 			return (0);
2713 		return (pp->error);
2714 	}
2715 
2716 	/*
2717 	 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2718 	 * because GEOM already handles that and handles it a bit differently.
2719 	 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2720 	 * only one exclusive consumer, no matter if it is reader or writer.
2721 	 * I like better the way GEOM works so I'll leave it for GEOM to
2722 	 * decide what to do.
2723 	 */
2724 
2725 	count = acr + acw + ace;
2726 	if (count == 0)
2727 		return (0);
2728 
2729 	flags = 0;
2730 	if (acr != 0 || ace != 0)
2731 		flags |= FREAD;
2732 	if (acw != 0)
2733 		flags |= FWRITE;
2734 
2735 	g_topology_unlock();
2736 	if (count > 0)
2737 		error = zvol_open(pp, flags, count);
2738 	else
2739 		error = zvol_close(pp, flags, -count);
2740 	g_topology_lock();
2741 	return (error);
2742 }
2743 
2744 static void
2745 zvol_geom_start(struct bio *bp)
2746 {
2747 	zvol_state_t *zv;
2748 	boolean_t first;
2749 
2750 	zv = bp->bio_to->private;
2751 	ASSERT(zv != NULL);
2752 	switch (bp->bio_cmd) {
2753 	case BIO_FLUSH:
2754 		if (!THREAD_CAN_SLEEP())
2755 			goto enqueue;
2756 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
2757 		g_io_deliver(bp, 0);
2758 		break;
2759 	case BIO_READ:
2760 	case BIO_WRITE:
2761 	case BIO_DELETE:
2762 		if (!THREAD_CAN_SLEEP())
2763 			goto enqueue;
2764 		zvol_strategy(bp);
2765 		break;
2766 	case BIO_GETATTR: {
2767 		spa_t *spa = dmu_objset_spa(zv->zv_objset);
2768 		uint64_t refd, avail, usedobjs, availobjs, val;
2769 
2770 		if (g_handleattr_int(bp, "GEOM::candelete", 1))
2771 			return;
2772 		if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2773 			dmu_objset_space(zv->zv_objset, &refd, &avail,
2774 			    &usedobjs, &availobjs);
2775 			if (g_handleattr_off_t(bp, "blocksavail",
2776 			    avail / DEV_BSIZE))
2777 				return;
2778 		} else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2779 			dmu_objset_space(zv->zv_objset, &refd, &avail,
2780 			    &usedobjs, &availobjs);
2781 			if (g_handleattr_off_t(bp, "blocksused",
2782 			    refd / DEV_BSIZE))
2783 				return;
2784 		} else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2785 			avail = metaslab_class_get_space(spa_normal_class(spa));
2786 			avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2787 			if (g_handleattr_off_t(bp, "poolblocksavail",
2788 			    avail / DEV_BSIZE))
2789 				return;
2790 		} else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2791 			refd = metaslab_class_get_alloc(spa_normal_class(spa));
2792 			if (g_handleattr_off_t(bp, "poolblocksused",
2793 			    refd / DEV_BSIZE))
2794 				return;
2795 		}
2796 		/* FALLTHROUGH */
2797 	}
2798 	default:
2799 		g_io_deliver(bp, EOPNOTSUPP);
2800 		break;
2801 	}
2802 	return;
2803 
2804 enqueue:
2805 	mtx_lock(&zv->zv_queue_mtx);
2806 	first = (bioq_first(&zv->zv_queue) == NULL);
2807 	bioq_insert_tail(&zv->zv_queue, bp);
2808 	mtx_unlock(&zv->zv_queue_mtx);
2809 	if (first)
2810 		wakeup_one(&zv->zv_queue);
2811 }
2812 
2813 static void
2814 zvol_geom_worker(void *arg)
2815 {
2816 	zvol_state_t *zv;
2817 	struct bio *bp;
2818 
2819 	thread_lock(curthread);
2820 	sched_prio(curthread, PRIBIO);
2821 	thread_unlock(curthread);
2822 
2823 	zv = arg;
2824 	for (;;) {
2825 		mtx_lock(&zv->zv_queue_mtx);
2826 		bp = bioq_takefirst(&zv->zv_queue);
2827 		if (bp == NULL) {
2828 			if (zv->zv_state == 1) {
2829 				zv->zv_state = 2;
2830 				wakeup(&zv->zv_state);
2831 				mtx_unlock(&zv->zv_queue_mtx);
2832 				kthread_exit();
2833 			}
2834 			msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2835 			    "zvol:io", 0);
2836 			continue;
2837 		}
2838 		mtx_unlock(&zv->zv_queue_mtx);
2839 		switch (bp->bio_cmd) {
2840 		case BIO_FLUSH:
2841 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
2842 			g_io_deliver(bp, 0);
2843 			break;
2844 		case BIO_READ:
2845 		case BIO_WRITE:
2846 		case BIO_DELETE:
2847 			zvol_strategy(bp);
2848 			break;
2849 		default:
2850 			g_io_deliver(bp, EOPNOTSUPP);
2851 			break;
2852 		}
2853 	}
2854 }
2855 
2856 extern boolean_t dataset_name_hidden(const char *name);
2857 
2858 static int
2859 zvol_create_snapshots(objset_t *os, const char *name)
2860 {
2861 	uint64_t cookie, obj;
2862 	char *sname;
2863 	int error, len;
2864 
2865 	cookie = obj = 0;
2866 	sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2867 
2868 #if 0
2869 	(void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2870 	    DS_FIND_SNAPSHOTS);
2871 #endif
2872 
2873 	for (;;) {
2874 		len = snprintf(sname, MAXPATHLEN, "%s@", name);
2875 		if (len >= MAXPATHLEN) {
2876 			dmu_objset_rele(os, FTAG);
2877 			error = ENAMETOOLONG;
2878 			break;
2879 		}
2880 
2881 		dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2882 		error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2883 		    sname + len, &obj, &cookie, NULL);
2884 		dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2885 		if (error != 0) {
2886 			if (error == ENOENT)
2887 				error = 0;
2888 			break;
2889 		}
2890 
2891 		error = zvol_create_minor(sname);
2892 		if (error != 0 && error != EEXIST) {
2893 			printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2894 			    sname, error);
2895 			break;
2896 		}
2897 	}
2898 
2899 	kmem_free(sname, MAXPATHLEN);
2900 	return (error);
2901 }
2902 
2903 int
2904 zvol_create_minors(const char *name)
2905 {
2906 	uint64_t cookie;
2907 	objset_t *os;
2908 	char *osname, *p;
2909 	int error, len;
2910 
2911 	if (dataset_name_hidden(name))
2912 		return (0);
2913 
2914 	if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2915 		printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2916 		    name, error);
2917 		return (error);
2918 	}
2919 	if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2920 		dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2921 		dsl_pool_rele(dmu_objset_pool(os), FTAG);
2922 		error = zvol_create_minor(name);
2923 		if (error == 0 || error == EEXIST) {
2924 			error = zvol_create_snapshots(os, name);
2925 		} else {
2926 			printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2927 			    name, error);
2928 		}
2929 		dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2930 		dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2931 		return (error);
2932 	}
2933 	if (dmu_objset_type(os) != DMU_OST_ZFS) {
2934 		dmu_objset_rele(os, FTAG);
2935 		return (0);
2936 	}
2937 
2938 	osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2939 	if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2940 		dmu_objset_rele(os, FTAG);
2941 		kmem_free(osname, MAXPATHLEN);
2942 		return (ENOENT);
2943 	}
2944 	p = osname + strlen(osname);
2945 	len = MAXPATHLEN - (p - osname);
2946 
2947 #if 0
2948 	/* Prefetch the datasets. */
2949 	cookie = 0;
2950 	while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2951 		if (!dataset_name_hidden(osname))
2952 			(void) dmu_objset_prefetch(osname, NULL);
2953 	}
2954 #endif
2955 
2956 	cookie = 0;
2957 	while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2958 	    &cookie) == 0) {
2959 		dmu_objset_rele(os, FTAG);
2960 		(void)zvol_create_minors(osname);
2961 		if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2962 			printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2963 			    name, error);
2964 			return (error);
2965 		}
2966 	}
2967 
2968 	dmu_objset_rele(os, FTAG);
2969 	kmem_free(osname, MAXPATHLEN);
2970 	return (0);
2971 }
2972 
2973 static void
2974 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2975 {
2976 	struct g_geom *gp;
2977 	struct g_provider *pp;
2978 	struct cdev *dev;
2979 
2980 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2981 
2982 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2983 		g_topology_lock();
2984 		pp = zv->zv_provider;
2985 		ASSERT(pp != NULL);
2986 		gp = pp->geom;
2987 		ASSERT(gp != NULL);
2988 
2989 		zv->zv_provider = NULL;
2990 		g_wither_provider(pp, ENXIO);
2991 
2992 		pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2993 		pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2994 		pp->sectorsize = DEV_BSIZE;
2995 		pp->mediasize = zv->zv_volsize;
2996 		pp->private = zv;
2997 		zv->zv_provider = pp;
2998 		g_error_provider(pp, 0);
2999 		g_topology_unlock();
3000 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
3001 		struct make_dev_args args;
3002 
3003 		if ((dev = zv->zv_dev) != NULL) {
3004 			zv->zv_dev = NULL;
3005 			destroy_dev(dev);
3006 			if (zv->zv_total_opens > 0) {
3007 				zv->zv_flags &= ~ZVOL_EXCL;
3008 				zv->zv_total_opens = 0;
3009 				zvol_last_close(zv);
3010 			}
3011 		}
3012 
3013 		make_dev_args_init(&args);
3014 		args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
3015 		args.mda_devsw = &zvol_cdevsw;
3016 		args.mda_cr = NULL;
3017 		args.mda_uid = UID_ROOT;
3018 		args.mda_gid = GID_OPERATOR;
3019 		args.mda_mode = 0640;
3020 		args.mda_si_drv2 = zv;
3021 		if (make_dev_s(&args, &zv->zv_dev,
3022 		    "%s/%s", ZVOL_DRIVER, newname) == 0)
3023 			zv->zv_dev->si_iosize_max = MAXPHYS;
3024 	}
3025 	strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3026 }
3027 
3028 void
3029 zvol_rename_minors(const char *oldname, const char *newname)
3030 {
3031 	char name[MAXPATHLEN];
3032 	struct g_provider *pp;
3033 	struct g_geom *gp;
3034 	size_t oldnamelen, newnamelen;
3035 	zvol_state_t *zv;
3036 	char *namebuf;
3037 	boolean_t locked = B_FALSE;
3038 
3039 	oldnamelen = strlen(oldname);
3040 	newnamelen = strlen(newname);
3041 
3042 	DROP_GIANT();
3043 	/* See comment in zvol_open(). */
3044 	if (!MUTEX_HELD(&zfsdev_state_lock)) {
3045 		mutex_enter(&zfsdev_state_lock);
3046 		locked = B_TRUE;
3047 	}
3048 
3049 	LIST_FOREACH(zv, &all_zvols, zv_links) {
3050 		if (strcmp(zv->zv_name, oldname) == 0) {
3051 			zvol_rename_minor(zv, newname);
3052 		} else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3053 		    (zv->zv_name[oldnamelen] == '/' ||
3054 		     zv->zv_name[oldnamelen] == '@')) {
3055 			snprintf(name, sizeof(name), "%s%c%s", newname,
3056 			    zv->zv_name[oldnamelen],
3057 			    zv->zv_name + oldnamelen + 1);
3058 			zvol_rename_minor(zv, name);
3059 		}
3060 	}
3061 
3062 	if (locked)
3063 		mutex_exit(&zfsdev_state_lock);
3064 	PICKUP_GIANT();
3065 }
3066 
3067 static int
3068 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3069 {
3070 	zvol_state_t *zv = dev->si_drv2;
3071 	int err = 0;
3072 
3073 	mutex_enter(&zfsdev_state_lock);
3074 	if (zv->zv_total_opens == 0)
3075 		err = zvol_first_open(zv);
3076 	if (err) {
3077 		mutex_exit(&zfsdev_state_lock);
3078 		return (err);
3079 	}
3080 	if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3081 		err = SET_ERROR(EROFS);
3082 		goto out;
3083 	}
3084 	if (zv->zv_flags & ZVOL_EXCL) {
3085 		err = SET_ERROR(EBUSY);
3086 		goto out;
3087 	}
3088 #ifdef FEXCL
3089 	if (flags & FEXCL) {
3090 		if (zv->zv_total_opens != 0) {
3091 			err = SET_ERROR(EBUSY);
3092 			goto out;
3093 		}
3094 		zv->zv_flags |= ZVOL_EXCL;
3095 	}
3096 #endif
3097 
3098 	zv->zv_total_opens++;
3099 	if (flags & (FSYNC | FDSYNC)) {
3100 		zv->zv_sync_cnt++;
3101 		if (zv->zv_sync_cnt == 1)
3102 			zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
3103 	}
3104 	mutex_exit(&zfsdev_state_lock);
3105 	return (err);
3106 out:
3107 	if (zv->zv_total_opens == 0)
3108 		zvol_last_close(zv);
3109 	mutex_exit(&zfsdev_state_lock);
3110 	return (err);
3111 }
3112 
3113 static int
3114 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3115 {
3116 	zvol_state_t *zv = dev->si_drv2;
3117 
3118 	mutex_enter(&zfsdev_state_lock);
3119 	if (zv->zv_flags & ZVOL_EXCL) {
3120 		ASSERT(zv->zv_total_opens == 1);
3121 		zv->zv_flags &= ~ZVOL_EXCL;
3122 	}
3123 
3124 	/*
3125 	 * If the open count is zero, this is a spurious close.
3126 	 * That indicates a bug in the kernel / DDI framework.
3127 	 */
3128 	ASSERT(zv->zv_total_opens != 0);
3129 
3130 	/*
3131 	 * You may get multiple opens, but only one close.
3132 	 */
3133 	zv->zv_total_opens--;
3134 	if (flags & (FSYNC | FDSYNC))
3135 		zv->zv_sync_cnt--;
3136 
3137 	if (zv->zv_total_opens == 0)
3138 		zvol_last_close(zv);
3139 
3140 	mutex_exit(&zfsdev_state_lock);
3141 	return (0);
3142 }
3143 
3144 static int
3145 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3146 {
3147 	zvol_state_t *zv;
3148 	rl_t *rl;
3149 	off_t offset, length;
3150 	int i, error;
3151 	boolean_t sync;
3152 
3153 	zv = dev->si_drv2;
3154 
3155 	error = 0;
3156 	KASSERT(zv->zv_total_opens > 0,
3157 	    ("Device with zero access count in zvol_d_ioctl"));
3158 
3159 	i = IOCPARM_LEN(cmd);
3160 	switch (cmd) {
3161 	case DIOCGSECTORSIZE:
3162 		*(u_int *)data = DEV_BSIZE;
3163 		break;
3164 	case DIOCGMEDIASIZE:
3165 		*(off_t *)data = zv->zv_volsize;
3166 		break;
3167 	case DIOCGFLUSH:
3168 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
3169 		break;
3170 	case DIOCGDELETE:
3171 		if (!zvol_unmap_enabled)
3172 			break;
3173 
3174 		offset = ((off_t *)data)[0];
3175 		length = ((off_t *)data)[1];
3176 		if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3177 		    offset < 0 || offset >= zv->zv_volsize ||
3178 		    length <= 0) {
3179 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
3180 			    length);
3181 			error = EINVAL;
3182 			break;
3183 		}
3184 
3185 		rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3186 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3187 		error = dmu_tx_assign(tx, TXG_WAIT);
3188 		if (error != 0) {
3189 			sync = FALSE;
3190 			dmu_tx_abort(tx);
3191 		} else {
3192 			sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
3193 			zvol_log_truncate(zv, tx, offset, length, sync);
3194 			dmu_tx_commit(tx);
3195 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3196 			    offset, length);
3197 		}
3198 		zfs_range_unlock(rl);
3199 		if (sync)
3200 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
3201 		break;
3202 	case DIOCGSTRIPESIZE:
3203 		*(off_t *)data = zv->zv_volblocksize;
3204 		break;
3205 	case DIOCGSTRIPEOFFSET:
3206 		*(off_t *)data = 0;
3207 		break;
3208 	case DIOCGATTR: {
3209 		spa_t *spa = dmu_objset_spa(zv->zv_objset);
3210 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3211 		uint64_t refd, avail, usedobjs, availobjs;
3212 
3213 		if (strcmp(arg->name, "GEOM::candelete") == 0)
3214 			arg->value.i = 1;
3215 		else if (strcmp(arg->name, "blocksavail") == 0) {
3216 			dmu_objset_space(zv->zv_objset, &refd, &avail,
3217 			    &usedobjs, &availobjs);
3218 			arg->value.off = avail / DEV_BSIZE;
3219 		} else if (strcmp(arg->name, "blocksused") == 0) {
3220 			dmu_objset_space(zv->zv_objset, &refd, &avail,
3221 			    &usedobjs, &availobjs);
3222 			arg->value.off = refd / DEV_BSIZE;
3223 		} else if (strcmp(arg->name, "poolblocksavail") == 0) {
3224 			avail = metaslab_class_get_space(spa_normal_class(spa));
3225 			avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3226 			arg->value.off = avail / DEV_BSIZE;
3227 		} else if (strcmp(arg->name, "poolblocksused") == 0) {
3228 			refd = metaslab_class_get_alloc(spa_normal_class(spa));
3229 			arg->value.off = refd / DEV_BSIZE;
3230 		} else
3231 			error = ENOIOCTL;
3232 		break;
3233 	}
3234 	case FIOSEEKHOLE:
3235 	case FIOSEEKDATA: {
3236 		off_t *off = (off_t *)data;
3237 		uint64_t noff;
3238 		boolean_t hole;
3239 
3240 		hole = (cmd == FIOSEEKHOLE);
3241 		noff = *off;
3242 		error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3243 		*off = noff;
3244 		break;
3245 	}
3246 	default:
3247 		error = ENOIOCTL;
3248 	}
3249 
3250 	return (error);
3251 }
3252 #endif	/* illumos */
3253