xref: /illumos-gate/usr/src/uts/common/fs/zfs/zvol.c (revision dcdeca0a948b9d3139743db085d193c9db8ff2a3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  *
24  * Portions Copyright 2010 Robert Milkowski
25  *
26  * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
27  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
28  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29  * Copyright (c) 2014 Integros [integros.com]
30  */
31 
32 /*
33  * ZFS volume emulation driver.
34  *
35  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
36  * Volumes are accessed through the symbolic links named:
37  *
38  * /dev/zvol/dsk/<pool_name>/<dataset_name>
39  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
40  *
41  * These links are created by the /dev filesystem (sdev_zvolops.c).
42  * Volumes are persistent through reboot.  No user command needs to be
43  * run before opening and using a device.
44  */
45 
46 #include <sys/types.h>
47 #include <sys/param.h>
48 #include <sys/errno.h>
49 #include <sys/uio.h>
50 #include <sys/buf.h>
51 #include <sys/modctl.h>
52 #include <sys/open.h>
53 #include <sys/kmem.h>
54 #include <sys/conf.h>
55 #include <sys/cmn_err.h>
56 #include <sys/stat.h>
57 #include <sys/zap.h>
58 #include <sys/spa.h>
59 #include <sys/spa_impl.h>
60 #include <sys/zio.h>
61 #include <sys/dmu_traverse.h>
62 #include <sys/dnode.h>
63 #include <sys/dsl_dataset.h>
64 #include <sys/dsl_prop.h>
65 #include <sys/dkio.h>
66 #include <sys/efi_partition.h>
67 #include <sys/byteorder.h>
68 #include <sys/pathname.h>
69 #include <sys/ddi.h>
70 #include <sys/sunddi.h>
71 #include <sys/crc32.h>
72 #include <sys/dirent.h>
73 #include <sys/policy.h>
74 #include <sys/fs/zfs.h>
75 #include <sys/zfs_ioctl.h>
76 #include <sys/mkdev.h>
77 #include <sys/zil.h>
78 #include <sys/refcount.h>
79 #include <sys/zfs_znode.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/vdev_disk.h>
82 #include <sys/vdev_impl.h>
83 #include <sys/vdev_raidz.h>
84 #include <sys/zvol.h>
85 #include <sys/dumphdr.h>
86 #include <sys/zil_impl.h>
87 #include <sys/dbuf.h>
88 #include <sys/dmu_tx.h>
89 #include <sys/zfeature.h>
90 #include <sys/zio_checksum.h>
91 #include <sys/zil_impl.h>
92 #include <sys/dkioc_free_util.h>
93 #include <sys/zfs_rlock.h>
94 
95 #include "zfs_namecheck.h"
96 
97 void *zfsdev_state;
98 static char *zvol_tag = "zvol_tag";
99 
100 #define	ZVOL_DUMPSIZE		"dumpsize"
101 
102 /*
103  * This lock protects the zfsdev_state structure from being modified
104  * while it's being used, e.g. an open that comes in before a create
105  * finishes.  It also protects temporary opens of the dataset so that,
106  * e.g., an open doesn't get a spurious EBUSY.
107  */
108 kmutex_t zfsdev_state_lock;
109 static uint32_t zvol_minors;
110 
111 typedef struct zvol_extent {
112 	list_node_t	ze_node;
113 	dva_t		ze_dva;		/* dva associated with this extent */
114 	uint64_t	ze_nblks;	/* number of blocks in extent */
115 } zvol_extent_t;
116 
117 /*
118  * The in-core state of each volume.
119  */
120 typedef struct zvol_state {
121 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
122 	uint64_t	zv_volsize;	/* amount of space we advertise */
123 	uint64_t	zv_volblocksize; /* volume block size */
124 	minor_t		zv_minor;	/* minor number */
125 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
126 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
127 	objset_t	*zv_objset;	/* objset handle */
128 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
129 	uint32_t	zv_total_opens;	/* total open count */
130 	zilog_t		*zv_zilog;	/* ZIL handle */
131 	list_t		zv_extents;	/* List of extents for dump */
132 	rangelock_t	zv_rangelock;
133 	dnode_t		*zv_dn;		/* dnode hold */
134 } zvol_state_t;
135 
136 /*
137  * zvol specific flags
138  */
139 #define	ZVOL_RDONLY	0x1
140 #define	ZVOL_DUMPIFIED	0x2
141 #define	ZVOL_EXCL	0x4
142 #define	ZVOL_WCE	0x8
143 
144 /*
145  * zvol maximum transfer in one DMU tx.
146  */
147 int zvol_maxphys = DMU_MAX_ACCESS/2;
148 
149 /*
150  * Toggle unmap functionality.
151  */
152 boolean_t zvol_unmap_enabled = B_TRUE;
153 
154 /*
155  * If true, unmaps requested as synchronous are executed synchronously,
156  * otherwise all unmaps are asynchronous.
157  */
158 boolean_t zvol_unmap_sync_enabled = B_FALSE;
159 
160 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
161     nvlist_t *, nvlist_t *);
162 static int zvol_remove_zv(zvol_state_t *);
163 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
164     struct lwb *lwb, zio_t *zio);
165 static int zvol_dumpify(zvol_state_t *zv);
166 static int zvol_dump_fini(zvol_state_t *zv);
167 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
168 
169 static void
170 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
171 {
172 	dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
173 
174 	zv->zv_volsize = volsize;
175 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
176 	    "Size", volsize) == DDI_SUCCESS);
177 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
178 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
179 
180 	/* Notify specfs to invalidate the cached size */
181 	spec_size_invalidate(dev, VBLK);
182 	spec_size_invalidate(dev, VCHR);
183 }
184 
185 int
186 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
187 {
188 	if (volsize == 0)
189 		return (SET_ERROR(EINVAL));
190 
191 	if (volsize % blocksize != 0)
192 		return (SET_ERROR(EINVAL));
193 
194 #ifdef _ILP32
195 	if (volsize - 1 > SPEC_MAXOFFSET_T)
196 		return (SET_ERROR(EOVERFLOW));
197 #endif
198 	return (0);
199 }
200 
201 int
202 zvol_check_volblocksize(uint64_t volblocksize)
203 {
204 	if (volblocksize < SPA_MINBLOCKSIZE ||
205 	    volblocksize > SPA_OLD_MAXBLOCKSIZE ||
206 	    !ISP2(volblocksize))
207 		return (SET_ERROR(EDOM));
208 
209 	return (0);
210 }
211 
212 int
213 zvol_get_stats(objset_t *os, nvlist_t *nv)
214 {
215 	int error;
216 	dmu_object_info_t doi;
217 	uint64_t val;
218 
219 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
220 	if (error)
221 		return (error);
222 
223 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
224 
225 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
226 
227 	if (error == 0) {
228 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
229 		    doi.doi_data_block_size);
230 	}
231 
232 	return (error);
233 }
234 
235 static zvol_state_t *
236 zvol_minor_lookup(const char *name)
237 {
238 	minor_t minor;
239 	zvol_state_t *zv;
240 
241 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
242 
243 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
244 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
245 		if (zv == NULL)
246 			continue;
247 		if (strcmp(zv->zv_name, name) == 0)
248 			return (zv);
249 	}
250 
251 	return (NULL);
252 }
253 
254 /* extent mapping arg */
255 struct maparg {
256 	zvol_state_t	*ma_zv;
257 	uint64_t	ma_blks;
258 };
259 
260 /*ARGSUSED*/
261 static int
262 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
263     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
264 {
265 	struct maparg *ma = arg;
266 	zvol_extent_t *ze;
267 	int bs = ma->ma_zv->zv_volblocksize;
268 
269 	if (bp == NULL || BP_IS_HOLE(bp) ||
270 	    zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
271 		return (0);
272 
273 	VERIFY(!BP_IS_EMBEDDED(bp));
274 
275 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
276 	ma->ma_blks++;
277 
278 	/* Abort immediately if we have encountered gang blocks */
279 	if (BP_IS_GANG(bp))
280 		return (SET_ERROR(EFRAGS));
281 
282 	/*
283 	 * See if the block is at the end of the previous extent.
284 	 */
285 	ze = list_tail(&ma->ma_zv->zv_extents);
286 	if (ze &&
287 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
288 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
289 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
290 		ze->ze_nblks++;
291 		return (0);
292 	}
293 
294 	dprintf_bp(bp, "%s", "next blkptr:");
295 
296 	/* start a new extent */
297 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
298 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
299 	ze->ze_nblks = 1;
300 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
301 	return (0);
302 }
303 
304 static void
305 zvol_free_extents(zvol_state_t *zv)
306 {
307 	zvol_extent_t *ze;
308 
309 	while (ze = list_head(&zv->zv_extents)) {
310 		list_remove(&zv->zv_extents, ze);
311 		kmem_free(ze, sizeof (zvol_extent_t));
312 	}
313 }
314 
315 static int
316 zvol_get_lbas(zvol_state_t *zv)
317 {
318 	objset_t *os = zv->zv_objset;
319 	struct maparg	ma;
320 	int		err;
321 
322 	ma.ma_zv = zv;
323 	ma.ma_blks = 0;
324 	zvol_free_extents(zv);
325 
326 	/* commit any in-flight changes before traversing the dataset */
327 	txg_wait_synced(dmu_objset_pool(os), 0);
328 	err = traverse_dataset(dmu_objset_ds(os), 0,
329 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
330 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
331 		zvol_free_extents(zv);
332 		return (err ? err : EIO);
333 	}
334 
335 	return (0);
336 }
337 
338 /* ARGSUSED */
339 void
340 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
341 {
342 	zfs_creat_t *zct = arg;
343 	nvlist_t *nvprops = zct->zct_props;
344 	int error;
345 	uint64_t volblocksize, volsize;
346 
347 	VERIFY(nvlist_lookup_uint64(nvprops,
348 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
349 	if (nvlist_lookup_uint64(nvprops,
350 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
351 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
352 
353 	/*
354 	 * These properties must be removed from the list so the generic
355 	 * property setting step won't apply to them.
356 	 */
357 	VERIFY(nvlist_remove_all(nvprops,
358 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
359 	(void) nvlist_remove_all(nvprops,
360 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
361 
362 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
363 	    DMU_OT_NONE, 0, tx);
364 	ASSERT(error == 0);
365 
366 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
367 	    DMU_OT_NONE, 0, tx);
368 	ASSERT(error == 0);
369 
370 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
371 	ASSERT(error == 0);
372 }
373 
374 /*
375  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
376  * implement DKIOCFREE/free-long-range.
377  */
378 static int
379 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
380 {
381 	zvol_state_t *zv = arg1;
382 	lr_truncate_t *lr = arg2;
383 	uint64_t offset, length;
384 
385 	if (byteswap)
386 		byteswap_uint64_array(lr, sizeof (*lr));
387 
388 	offset = lr->lr_offset;
389 	length = lr->lr_length;
390 
391 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
392 }
393 
394 /*
395  * Replay a TX_WRITE ZIL transaction that didn't get committed
396  * after a system failure
397  */
398 static int
399 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
400 {
401 	zvol_state_t *zv = arg1;
402 	lr_write_t *lr = arg2;
403 	objset_t *os = zv->zv_objset;
404 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
405 	uint64_t offset, length;
406 	dmu_tx_t *tx;
407 	int error;
408 
409 	if (byteswap)
410 		byteswap_uint64_array(lr, sizeof (*lr));
411 
412 	offset = lr->lr_offset;
413 	length = lr->lr_length;
414 
415 	/* If it's a dmu_sync() block, write the whole block */
416 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
417 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
418 		if (length < blocksize) {
419 			offset -= offset % blocksize;
420 			length = blocksize;
421 		}
422 	}
423 
424 	tx = dmu_tx_create(os);
425 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
426 	error = dmu_tx_assign(tx, TXG_WAIT);
427 	if (error) {
428 		dmu_tx_abort(tx);
429 	} else {
430 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
431 		dmu_tx_commit(tx);
432 	}
433 
434 	return (error);
435 }
436 
437 /* ARGSUSED */
438 static int
439 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
440 {
441 	return (SET_ERROR(ENOTSUP));
442 }
443 
444 /*
445  * Callback vectors for replaying records.
446  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
447  */
448 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
449 	zvol_replay_err,	/* 0 no such transaction type */
450 	zvol_replay_err,	/* TX_CREATE */
451 	zvol_replay_err,	/* TX_MKDIR */
452 	zvol_replay_err,	/* TX_MKXATTR */
453 	zvol_replay_err,	/* TX_SYMLINK */
454 	zvol_replay_err,	/* TX_REMOVE */
455 	zvol_replay_err,	/* TX_RMDIR */
456 	zvol_replay_err,	/* TX_LINK */
457 	zvol_replay_err,	/* TX_RENAME */
458 	zvol_replay_write,	/* TX_WRITE */
459 	zvol_replay_truncate,	/* TX_TRUNCATE */
460 	zvol_replay_err,	/* TX_SETATTR */
461 	zvol_replay_err,	/* TX_ACL */
462 	zvol_replay_err,	/* TX_CREATE_ACL */
463 	zvol_replay_err,	/* TX_CREATE_ATTR */
464 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
465 	zvol_replay_err,	/* TX_MKDIR_ACL */
466 	zvol_replay_err,	/* TX_MKDIR_ATTR */
467 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
468 	zvol_replay_err,	/* TX_WRITE2 */
469 };
470 
471 int
472 zvol_name2minor(const char *name, minor_t *minor)
473 {
474 	zvol_state_t *zv;
475 
476 	mutex_enter(&zfsdev_state_lock);
477 	zv = zvol_minor_lookup(name);
478 	if (minor && zv)
479 		*minor = zv->zv_minor;
480 	mutex_exit(&zfsdev_state_lock);
481 	return (zv ? 0 : -1);
482 }
483 
484 /*
485  * Create a minor node (plus a whole lot more) for the specified volume.
486  */
487 int
488 zvol_create_minor(const char *name)
489 {
490 	zfs_soft_state_t *zs;
491 	zvol_state_t *zv;
492 	objset_t *os;
493 	dmu_object_info_t doi;
494 	minor_t minor = 0;
495 	char chrbuf[30], blkbuf[30];
496 	int error;
497 
498 	mutex_enter(&zfsdev_state_lock);
499 
500 	if (zvol_minor_lookup(name) != NULL) {
501 		mutex_exit(&zfsdev_state_lock);
502 		return (SET_ERROR(EEXIST));
503 	}
504 
505 	/* lie and say we're read-only */
506 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
507 
508 	if (error) {
509 		mutex_exit(&zfsdev_state_lock);
510 		return (error);
511 	}
512 
513 	if ((minor = zfsdev_minor_alloc()) == 0) {
514 		dmu_objset_disown(os, FTAG);
515 		mutex_exit(&zfsdev_state_lock);
516 		return (SET_ERROR(ENXIO));
517 	}
518 
519 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
520 		dmu_objset_disown(os, FTAG);
521 		mutex_exit(&zfsdev_state_lock);
522 		return (SET_ERROR(EAGAIN));
523 	}
524 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
525 	    (char *)name);
526 
527 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
528 
529 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
530 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
531 		ddi_soft_state_free(zfsdev_state, minor);
532 		dmu_objset_disown(os, FTAG);
533 		mutex_exit(&zfsdev_state_lock);
534 		return (SET_ERROR(EAGAIN));
535 	}
536 
537 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
538 
539 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
540 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
541 		ddi_remove_minor_node(zfs_dip, chrbuf);
542 		ddi_soft_state_free(zfsdev_state, minor);
543 		dmu_objset_disown(os, FTAG);
544 		mutex_exit(&zfsdev_state_lock);
545 		return (SET_ERROR(EAGAIN));
546 	}
547 
548 	zs = ddi_get_soft_state(zfsdev_state, minor);
549 	zs->zss_type = ZSST_ZVOL;
550 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
551 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
552 	zv->zv_min_bs = DEV_BSHIFT;
553 	zv->zv_minor = minor;
554 	zv->zv_objset = os;
555 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
556 		zv->zv_flags |= ZVOL_RDONLY;
557 	rangelock_init(&zv->zv_rangelock, NULL, NULL);
558 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
559 	    offsetof(zvol_extent_t, ze_node));
560 	/* get and cache the blocksize */
561 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
562 	ASSERT(error == 0);
563 	zv->zv_volblocksize = doi.doi_data_block_size;
564 
565 	if (spa_writeable(dmu_objset_spa(os))) {
566 		if (zil_replay_disable)
567 			zil_destroy(dmu_objset_zil(os), B_FALSE);
568 		else
569 			zil_replay(os, zv, zvol_replay_vector);
570 	}
571 	dmu_objset_disown(os, FTAG);
572 	zv->zv_objset = NULL;
573 
574 	zvol_minors++;
575 
576 	mutex_exit(&zfsdev_state_lock);
577 
578 	return (0);
579 }
580 
581 /*
582  * Remove minor node for the specified volume.
583  */
584 static int
585 zvol_remove_zv(zvol_state_t *zv)
586 {
587 	char nmbuf[20];
588 	minor_t minor = zv->zv_minor;
589 
590 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
591 	if (zv->zv_total_opens != 0)
592 		return (SET_ERROR(EBUSY));
593 
594 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
595 	ddi_remove_minor_node(zfs_dip, nmbuf);
596 
597 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
598 	ddi_remove_minor_node(zfs_dip, nmbuf);
599 
600 	rangelock_fini(&zv->zv_rangelock);
601 
602 	kmem_free(zv, sizeof (zvol_state_t));
603 
604 	ddi_soft_state_free(zfsdev_state, minor);
605 
606 	zvol_minors--;
607 	return (0);
608 }
609 
610 int
611 zvol_remove_minor(const char *name)
612 {
613 	zvol_state_t *zv;
614 	int rc;
615 
616 	mutex_enter(&zfsdev_state_lock);
617 	if ((zv = zvol_minor_lookup(name)) == NULL) {
618 		mutex_exit(&zfsdev_state_lock);
619 		return (SET_ERROR(ENXIO));
620 	}
621 	rc = zvol_remove_zv(zv);
622 	mutex_exit(&zfsdev_state_lock);
623 	return (rc);
624 }
625 
626 int
627 zvol_first_open(zvol_state_t *zv)
628 {
629 	objset_t *os;
630 	uint64_t volsize;
631 	int error;
632 	uint64_t readonly;
633 
634 	/* lie and say we're read-only */
635 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
636 	    zvol_tag, &os);
637 	if (error)
638 		return (error);
639 
640 	zv->zv_objset = os;
641 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
642 	if (error) {
643 		ASSERT(error == 0);
644 		dmu_objset_disown(os, zvol_tag);
645 		return (error);
646 	}
647 
648 	error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
649 	if (error) {
650 		dmu_objset_disown(os, zvol_tag);
651 		return (error);
652 	}
653 
654 	zvol_size_changed(zv, volsize);
655 	zv->zv_zilog = zil_open(os, zvol_get_data);
656 
657 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
658 	    NULL) == 0);
659 	if (readonly || dmu_objset_is_snapshot(os) ||
660 	    !spa_writeable(dmu_objset_spa(os)))
661 		zv->zv_flags |= ZVOL_RDONLY;
662 	else
663 		zv->zv_flags &= ~ZVOL_RDONLY;
664 	return (error);
665 }
666 
667 void
668 zvol_last_close(zvol_state_t *zv)
669 {
670 	zil_close(zv->zv_zilog);
671 	zv->zv_zilog = NULL;
672 
673 	dnode_rele(zv->zv_dn, zvol_tag);
674 	zv->zv_dn = NULL;
675 
676 	/*
677 	 * Evict cached data
678 	 */
679 	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
680 	    !(zv->zv_flags & ZVOL_RDONLY))
681 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
682 	dmu_objset_evict_dbufs(zv->zv_objset);
683 
684 	dmu_objset_disown(zv->zv_objset, zvol_tag);
685 	zv->zv_objset = NULL;
686 }
687 
688 int
689 zvol_prealloc(zvol_state_t *zv)
690 {
691 	objset_t *os = zv->zv_objset;
692 	dmu_tx_t *tx;
693 	uint64_t refd, avail, usedobjs, availobjs;
694 	uint64_t resid = zv->zv_volsize;
695 	uint64_t off = 0;
696 
697 	/* Check the space usage before attempting to allocate the space */
698 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
699 	if (avail < zv->zv_volsize)
700 		return (SET_ERROR(ENOSPC));
701 
702 	/* Free old extents if they exist */
703 	zvol_free_extents(zv);
704 
705 	while (resid != 0) {
706 		int error;
707 		uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
708 
709 		tx = dmu_tx_create(os);
710 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
711 		error = dmu_tx_assign(tx, TXG_WAIT);
712 		if (error) {
713 			dmu_tx_abort(tx);
714 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
715 			return (error);
716 		}
717 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
718 		dmu_tx_commit(tx);
719 		off += bytes;
720 		resid -= bytes;
721 	}
722 	txg_wait_synced(dmu_objset_pool(os), 0);
723 
724 	return (0);
725 }
726 
727 static int
728 zvol_update_volsize(objset_t *os, uint64_t volsize)
729 {
730 	dmu_tx_t *tx;
731 	int error;
732 
733 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
734 
735 	tx = dmu_tx_create(os);
736 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
737 	dmu_tx_mark_netfree(tx);
738 	error = dmu_tx_assign(tx, TXG_WAIT);
739 	if (error) {
740 		dmu_tx_abort(tx);
741 		return (error);
742 	}
743 
744 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
745 	    &volsize, tx);
746 	dmu_tx_commit(tx);
747 
748 	if (error == 0)
749 		error = dmu_free_long_range(os,
750 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
751 	return (error);
752 }
753 
754 void
755 zvol_remove_minors(const char *name)
756 {
757 	zvol_state_t *zv;
758 	char *namebuf;
759 	minor_t minor;
760 
761 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
762 	(void) strncpy(namebuf, name, strlen(name));
763 	(void) strcat(namebuf, "/");
764 	mutex_enter(&zfsdev_state_lock);
765 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
766 
767 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
768 		if (zv == NULL)
769 			continue;
770 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
771 			(void) zvol_remove_zv(zv);
772 	}
773 	kmem_free(namebuf, strlen(name) + 2);
774 
775 	mutex_exit(&zfsdev_state_lock);
776 }
777 
778 static int
779 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
780 {
781 	uint64_t old_volsize = 0ULL;
782 	int error = 0;
783 
784 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
785 
786 	/*
787 	 * Reinitialize the dump area to the new size. If we
788 	 * failed to resize the dump area then restore it back to
789 	 * its original size.  We must set the new volsize prior
790 	 * to calling dumpvp_resize() to ensure that the devices'
791 	 * size(9P) is not visible by the dump subsystem.
792 	 */
793 	old_volsize = zv->zv_volsize;
794 	zvol_size_changed(zv, volsize);
795 
796 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
797 		if ((error = zvol_dumpify(zv)) != 0 ||
798 		    (error = dumpvp_resize()) != 0) {
799 			int dumpify_error;
800 
801 			(void) zvol_update_volsize(zv->zv_objset, old_volsize);
802 			zvol_size_changed(zv, old_volsize);
803 			dumpify_error = zvol_dumpify(zv);
804 			error = dumpify_error ? dumpify_error : error;
805 		}
806 	}
807 
808 	/*
809 	 * Generate a LUN expansion event.
810 	 */
811 	if (error == 0) {
812 		sysevent_id_t eid;
813 		nvlist_t *attr;
814 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
815 
816 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
817 		    zv->zv_minor);
818 
819 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
820 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
821 
822 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
823 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
824 
825 		nvlist_free(attr);
826 		kmem_free(physpath, MAXPATHLEN);
827 	}
828 	return (error);
829 }
830 
831 int
832 zvol_set_volsize(const char *name, uint64_t volsize)
833 {
834 	zvol_state_t *zv = NULL;
835 	objset_t *os;
836 	int error;
837 	dmu_object_info_t doi;
838 	uint64_t readonly;
839 	boolean_t owned = B_FALSE;
840 
841 	error = dsl_prop_get_integer(name,
842 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
843 	if (error != 0)
844 		return (error);
845 	if (readonly)
846 		return (SET_ERROR(EROFS));
847 
848 	mutex_enter(&zfsdev_state_lock);
849 	zv = zvol_minor_lookup(name);
850 
851 	if (zv == NULL || zv->zv_objset == NULL) {
852 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
853 		    FTAG, &os)) != 0) {
854 			mutex_exit(&zfsdev_state_lock);
855 			return (error);
856 		}
857 		owned = B_TRUE;
858 		if (zv != NULL)
859 			zv->zv_objset = os;
860 	} else {
861 		os = zv->zv_objset;
862 	}
863 
864 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
865 	    (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
866 		goto out;
867 
868 	error = zvol_update_volsize(os, volsize);
869 
870 	if (error == 0 && zv != NULL)
871 		error = zvol_update_live_volsize(zv, volsize);
872 out:
873 	if (owned) {
874 		dmu_objset_disown(os, FTAG);
875 		if (zv != NULL)
876 			zv->zv_objset = NULL;
877 	}
878 	mutex_exit(&zfsdev_state_lock);
879 	return (error);
880 }
881 
882 /*ARGSUSED*/
883 int
884 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
885 {
886 	zvol_state_t *zv;
887 	int err = 0;
888 
889 	mutex_enter(&zfsdev_state_lock);
890 
891 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
892 	if (zv == NULL) {
893 		mutex_exit(&zfsdev_state_lock);
894 		return (SET_ERROR(ENXIO));
895 	}
896 
897 	if (zv->zv_total_opens == 0)
898 		err = zvol_first_open(zv);
899 	if (err) {
900 		mutex_exit(&zfsdev_state_lock);
901 		return (err);
902 	}
903 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
904 		err = SET_ERROR(EROFS);
905 		goto out;
906 	}
907 	if (zv->zv_flags & ZVOL_EXCL) {
908 		err = SET_ERROR(EBUSY);
909 		goto out;
910 	}
911 	if (flag & FEXCL) {
912 		if (zv->zv_total_opens != 0) {
913 			err = SET_ERROR(EBUSY);
914 			goto out;
915 		}
916 		zv->zv_flags |= ZVOL_EXCL;
917 	}
918 
919 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
920 		zv->zv_open_count[otyp]++;
921 		zv->zv_total_opens++;
922 	}
923 	mutex_exit(&zfsdev_state_lock);
924 
925 	return (err);
926 out:
927 	if (zv->zv_total_opens == 0)
928 		zvol_last_close(zv);
929 	mutex_exit(&zfsdev_state_lock);
930 	return (err);
931 }
932 
933 /*ARGSUSED*/
934 int
935 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
936 {
937 	minor_t minor = getminor(dev);
938 	zvol_state_t *zv;
939 	int error = 0;
940 
941 	mutex_enter(&zfsdev_state_lock);
942 
943 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
944 	if (zv == NULL) {
945 		mutex_exit(&zfsdev_state_lock);
946 		return (SET_ERROR(ENXIO));
947 	}
948 
949 	if (zv->zv_flags & ZVOL_EXCL) {
950 		ASSERT(zv->zv_total_opens == 1);
951 		zv->zv_flags &= ~ZVOL_EXCL;
952 	}
953 
954 	/*
955 	 * If the open count is zero, this is a spurious close.
956 	 * That indicates a bug in the kernel / DDI framework.
957 	 */
958 	ASSERT(zv->zv_open_count[otyp] != 0);
959 	ASSERT(zv->zv_total_opens != 0);
960 
961 	/*
962 	 * You may get multiple opens, but only one close.
963 	 */
964 	zv->zv_open_count[otyp]--;
965 	zv->zv_total_opens--;
966 
967 	if (zv->zv_total_opens == 0)
968 		zvol_last_close(zv);
969 
970 	mutex_exit(&zfsdev_state_lock);
971 	return (error);
972 }
973 
974 /* ARGSUSED */
975 static void
976 zvol_get_done(zgd_t *zgd, int error)
977 {
978 	if (zgd->zgd_db)
979 		dmu_buf_rele(zgd->zgd_db, zgd);
980 
981 	rangelock_exit(zgd->zgd_lr);
982 
983 	kmem_free(zgd, sizeof (zgd_t));
984 }
985 
986 /*
987  * Get data to generate a TX_WRITE intent log record.
988  */
989 static int
990 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
991 {
992 	zvol_state_t *zv = arg;
993 	uint64_t offset = lr->lr_offset;
994 	uint64_t size = lr->lr_length;	/* length of user data */
995 	dmu_buf_t *db;
996 	zgd_t *zgd;
997 	int error;
998 
999 	ASSERT3P(lwb, !=, NULL);
1000 	ASSERT3P(zio, !=, NULL);
1001 	ASSERT3U(size, !=, 0);
1002 
1003 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1004 	zgd->zgd_lwb = lwb;
1005 
1006 	/*
1007 	 * Write records come in two flavors: immediate and indirect.
1008 	 * For small writes it's cheaper to store the data with the
1009 	 * log record (immediate); for large writes it's cheaper to
1010 	 * sync the data and get a pointer to it (indirect) so that
1011 	 * we don't have to write the data twice.
1012 	 */
1013 	if (buf != NULL) { /* immediate write */
1014 		zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1015 		    RL_READER);
1016 		error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1017 		    DMU_READ_NO_PREFETCH);
1018 	} else { /* indirect write */
1019 		/*
1020 		 * Have to lock the whole block to ensure when it's written out
1021 		 * and its checksum is being calculated that no one can change
1022 		 * the data. Contrarily to zfs_get_data we need not re-check
1023 		 * blocksize after we get the lock because it cannot be changed.
1024 		 */
1025 		size = zv->zv_volblocksize;
1026 		offset = P2ALIGN(offset, size);
1027 		zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1028 		    RL_READER);
1029 		error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1030 		    DMU_READ_NO_PREFETCH);
1031 		if (error == 0) {
1032 			blkptr_t *bp = &lr->lr_blkptr;
1033 
1034 			zgd->zgd_db = db;
1035 			zgd->zgd_bp = bp;
1036 
1037 			ASSERT(db->db_offset == offset);
1038 			ASSERT(db->db_size == size);
1039 
1040 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1041 			    zvol_get_done, zgd);
1042 
1043 			if (error == 0)
1044 				return (0);
1045 		}
1046 	}
1047 
1048 	zvol_get_done(zgd, error);
1049 
1050 	return (error);
1051 }
1052 
1053 /*
1054  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1055  *
1056  * We store data in the log buffers if it's small enough.
1057  * Otherwise we will later flush the data out via dmu_sync().
1058  */
1059 ssize_t zvol_immediate_write_sz = 32768;
1060 
1061 static void
1062 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1063     boolean_t sync)
1064 {
1065 	uint32_t blocksize = zv->zv_volblocksize;
1066 	zilog_t *zilog = zv->zv_zilog;
1067 	itx_wr_state_t write_state;
1068 
1069 	if (zil_replaying(zilog, tx))
1070 		return;
1071 
1072 	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1073 		write_state = WR_INDIRECT;
1074 	else if (!spa_has_slogs(zilog->zl_spa) &&
1075 	    resid >= blocksize && blocksize > zvol_immediate_write_sz)
1076 		write_state = WR_INDIRECT;
1077 	else if (sync)
1078 		write_state = WR_COPIED;
1079 	else
1080 		write_state = WR_NEED_COPY;
1081 
1082 	while (resid) {
1083 		itx_t *itx;
1084 		lr_write_t *lr;
1085 		itx_wr_state_t wr_state = write_state;
1086 		ssize_t len = resid;
1087 
1088 		if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1089 			wr_state = WR_NEED_COPY;
1090 		else if (wr_state == WR_INDIRECT)
1091 			len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1092 
1093 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1094 		    (wr_state == WR_COPIED ? len : 0));
1095 		lr = (lr_write_t *)&itx->itx_lr;
1096 		if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1097 		    off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1098 			zil_itx_destroy(itx);
1099 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1100 			lr = (lr_write_t *)&itx->itx_lr;
1101 			wr_state = WR_NEED_COPY;
1102 		}
1103 
1104 		itx->itx_wr_state = wr_state;
1105 		lr->lr_foid = ZVOL_OBJ;
1106 		lr->lr_offset = off;
1107 		lr->lr_length = len;
1108 		lr->lr_blkoff = 0;
1109 		BP_ZERO(&lr->lr_blkptr);
1110 
1111 		itx->itx_private = zv;
1112 		itx->itx_sync = sync;
1113 
1114 		zil_itx_assign(zilog, itx, tx);
1115 
1116 		off += len;
1117 		resid -= len;
1118 	}
1119 }
1120 
1121 static int
1122 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1123     uint64_t size, boolean_t doread, boolean_t isdump)
1124 {
1125 	vdev_disk_t *dvd;
1126 	int c;
1127 	int numerrors = 0;
1128 
1129 	if (vd->vdev_ops == &vdev_mirror_ops ||
1130 	    vd->vdev_ops == &vdev_replacing_ops ||
1131 	    vd->vdev_ops == &vdev_spare_ops) {
1132 		for (c = 0; c < vd->vdev_children; c++) {
1133 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1134 			    addr, offset, origoffset, size, doread, isdump);
1135 			if (err != 0) {
1136 				numerrors++;
1137 			} else if (doread) {
1138 				break;
1139 			}
1140 		}
1141 	}
1142 
1143 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1144 		return (numerrors < vd->vdev_children ? 0 : EIO);
1145 
1146 	if (doread && !vdev_readable(vd))
1147 		return (SET_ERROR(EIO));
1148 	else if (!doread && !vdev_writeable(vd))
1149 		return (SET_ERROR(EIO));
1150 
1151 	if (vd->vdev_ops == &vdev_raidz_ops) {
1152 		return (vdev_raidz_physio(vd,
1153 		    addr, size, offset, origoffset, doread, isdump));
1154 	}
1155 
1156 	offset += VDEV_LABEL_START_SIZE;
1157 
1158 	if (ddi_in_panic() || isdump) {
1159 		ASSERT(!doread);
1160 		if (doread)
1161 			return (SET_ERROR(EIO));
1162 		dvd = vd->vdev_tsd;
1163 		ASSERT3P(dvd, !=, NULL);
1164 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1165 		    lbtodb(size)));
1166 	} else {
1167 		dvd = vd->vdev_tsd;
1168 		ASSERT3P(dvd, !=, NULL);
1169 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1170 		    offset, doread ? B_READ : B_WRITE));
1171 	}
1172 }
1173 
1174 static int
1175 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1176     boolean_t doread, boolean_t isdump)
1177 {
1178 	vdev_t *vd;
1179 	int error;
1180 	zvol_extent_t *ze;
1181 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1182 
1183 	/* Must be sector aligned, and not stradle a block boundary. */
1184 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1185 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1186 		return (SET_ERROR(EINVAL));
1187 	}
1188 	ASSERT(size <= zv->zv_volblocksize);
1189 
1190 	/* Locate the extent this belongs to */
1191 	ze = list_head(&zv->zv_extents);
1192 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1193 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1194 		ze = list_next(&zv->zv_extents, ze);
1195 	}
1196 
1197 	if (ze == NULL)
1198 		return (SET_ERROR(EINVAL));
1199 
1200 	if (!ddi_in_panic())
1201 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1202 
1203 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1204 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1205 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1206 	    size, doread, isdump);
1207 
1208 	if (!ddi_in_panic())
1209 		spa_config_exit(spa, SCL_STATE, FTAG);
1210 
1211 	return (error);
1212 }
1213 
1214 int
1215 zvol_strategy(buf_t *bp)
1216 {
1217 	zfs_soft_state_t *zs = NULL;
1218 	zvol_state_t *zv;
1219 	uint64_t off, volsize;
1220 	size_t resid;
1221 	char *addr;
1222 	objset_t *os;
1223 	int error = 0;
1224 	boolean_t doread = bp->b_flags & B_READ;
1225 	boolean_t is_dumpified;
1226 	boolean_t sync;
1227 
1228 	if (getminor(bp->b_edev) == 0) {
1229 		error = SET_ERROR(EINVAL);
1230 	} else {
1231 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1232 		if (zs == NULL)
1233 			error = SET_ERROR(ENXIO);
1234 		else if (zs->zss_type != ZSST_ZVOL)
1235 			error = SET_ERROR(EINVAL);
1236 	}
1237 
1238 	if (error) {
1239 		bioerror(bp, error);
1240 		biodone(bp);
1241 		return (0);
1242 	}
1243 
1244 	zv = zs->zss_data;
1245 
1246 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1247 		bioerror(bp, EROFS);
1248 		biodone(bp);
1249 		return (0);
1250 	}
1251 
1252 	off = ldbtob(bp->b_blkno);
1253 	volsize = zv->zv_volsize;
1254 
1255 	os = zv->zv_objset;
1256 	ASSERT(os != NULL);
1257 
1258 	bp_mapin(bp);
1259 	addr = bp->b_un.b_addr;
1260 	resid = bp->b_bcount;
1261 
1262 	if (resid > 0 && (off < 0 || off >= volsize)) {
1263 		bioerror(bp, EIO);
1264 		biodone(bp);
1265 		return (0);
1266 	}
1267 
1268 	is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1269 	sync = ((!(bp->b_flags & B_ASYNC) &&
1270 	    !(zv->zv_flags & ZVOL_WCE)) ||
1271 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1272 	    !doread && !is_dumpified;
1273 
1274 	/*
1275 	 * There must be no buffer changes when doing a dmu_sync() because
1276 	 * we can't change the data whilst calculating the checksum.
1277 	 */
1278 	locked_range_t *lr = rangelock_enter(&zv->zv_rangelock, off, resid,
1279 	    doread ? RL_READER : RL_WRITER);
1280 
1281 	while (resid != 0 && off < volsize) {
1282 		size_t size = MIN(resid, zvol_maxphys);
1283 		if (is_dumpified) {
1284 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1285 			error = zvol_dumpio(zv, addr, off, size,
1286 			    doread, B_FALSE);
1287 		} else if (doread) {
1288 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1289 			    DMU_READ_PREFETCH);
1290 		} else {
1291 			dmu_tx_t *tx = dmu_tx_create(os);
1292 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1293 			error = dmu_tx_assign(tx, TXG_WAIT);
1294 			if (error) {
1295 				dmu_tx_abort(tx);
1296 			} else {
1297 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1298 				zvol_log_write(zv, tx, off, size, sync);
1299 				dmu_tx_commit(tx);
1300 			}
1301 		}
1302 		if (error) {
1303 			/* convert checksum errors into IO errors */
1304 			if (error == ECKSUM)
1305 				error = SET_ERROR(EIO);
1306 			break;
1307 		}
1308 		off += size;
1309 		addr += size;
1310 		resid -= size;
1311 	}
1312 	rangelock_exit(lr);
1313 
1314 	if ((bp->b_resid = resid) == bp->b_bcount)
1315 		bioerror(bp, off > volsize ? EINVAL : error);
1316 
1317 	if (sync)
1318 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1319 	biodone(bp);
1320 
1321 	return (0);
1322 }
1323 
1324 /*
1325  * Set the buffer count to the zvol maximum transfer.
1326  * Using our own routine instead of the default minphys()
1327  * means that for larger writes we write bigger buffers on X86
1328  * (128K instead of 56K) and flush the disk write cache less often
1329  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1330  * 56K on X86 and 128K on sparc).
1331  */
1332 void
1333 zvol_minphys(struct buf *bp)
1334 {
1335 	if (bp->b_bcount > zvol_maxphys)
1336 		bp->b_bcount = zvol_maxphys;
1337 }
1338 
1339 int
1340 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1341 {
1342 	minor_t minor = getminor(dev);
1343 	zvol_state_t *zv;
1344 	int error = 0;
1345 	uint64_t size;
1346 	uint64_t boff;
1347 	uint64_t resid;
1348 
1349 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1350 	if (zv == NULL)
1351 		return (SET_ERROR(ENXIO));
1352 
1353 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1354 		return (SET_ERROR(EINVAL));
1355 
1356 	boff = ldbtob(blkno);
1357 	resid = ldbtob(nblocks);
1358 
1359 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1360 
1361 	while (resid) {
1362 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1363 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1364 		if (error)
1365 			break;
1366 		boff += size;
1367 		addr += size;
1368 		resid -= size;
1369 	}
1370 
1371 	return (error);
1372 }
1373 
1374 /*ARGSUSED*/
1375 int
1376 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1377 {
1378 	minor_t minor = getminor(dev);
1379 	zvol_state_t *zv;
1380 	uint64_t volsize;
1381 	int error = 0;
1382 
1383 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1384 	if (zv == NULL)
1385 		return (SET_ERROR(ENXIO));
1386 
1387 	volsize = zv->zv_volsize;
1388 	if (uio->uio_resid > 0 &&
1389 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1390 		return (SET_ERROR(EIO));
1391 
1392 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1393 		error = physio(zvol_strategy, NULL, dev, B_READ,
1394 		    zvol_minphys, uio);
1395 		return (error);
1396 	}
1397 
1398 	locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1399 	    uio->uio_loffset, uio->uio_resid, RL_READER);
1400 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1401 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1402 
1403 		/* don't read past the end */
1404 		if (bytes > volsize - uio->uio_loffset)
1405 			bytes = volsize - uio->uio_loffset;
1406 
1407 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1408 		if (error) {
1409 			/* convert checksum errors into IO errors */
1410 			if (error == ECKSUM)
1411 				error = SET_ERROR(EIO);
1412 			break;
1413 		}
1414 	}
1415 	rangelock_exit(lr);
1416 
1417 	return (error);
1418 }
1419 
1420 /*ARGSUSED*/
1421 int
1422 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1423 {
1424 	minor_t minor = getminor(dev);
1425 	zvol_state_t *zv;
1426 	uint64_t volsize;
1427 	int error = 0;
1428 	boolean_t sync;
1429 
1430 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1431 	if (zv == NULL)
1432 		return (SET_ERROR(ENXIO));
1433 
1434 	volsize = zv->zv_volsize;
1435 	if (uio->uio_resid > 0 &&
1436 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1437 		return (SET_ERROR(EIO));
1438 
1439 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1440 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1441 		    zvol_minphys, uio);
1442 		return (error);
1443 	}
1444 
1445 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1446 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1447 
1448 	locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1449 	    uio->uio_loffset, uio->uio_resid, RL_WRITER);
1450 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1451 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1452 		uint64_t off = uio->uio_loffset;
1453 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1454 
1455 		if (bytes > volsize - off)	/* don't write past the end */
1456 			bytes = volsize - off;
1457 
1458 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1459 		error = dmu_tx_assign(tx, TXG_WAIT);
1460 		if (error) {
1461 			dmu_tx_abort(tx);
1462 			break;
1463 		}
1464 		error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1465 		if (error == 0)
1466 			zvol_log_write(zv, tx, off, bytes, sync);
1467 		dmu_tx_commit(tx);
1468 
1469 		if (error)
1470 			break;
1471 	}
1472 	rangelock_exit(lr);
1473 
1474 	if (sync)
1475 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1476 	return (error);
1477 }
1478 
1479 int
1480 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1481 {
1482 	struct uuid uuid = EFI_RESERVED;
1483 	efi_gpe_t gpe = { 0 };
1484 	uint32_t crc;
1485 	dk_efi_t efi;
1486 	int length;
1487 	char *ptr;
1488 
1489 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1490 		return (SET_ERROR(EFAULT));
1491 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1492 	length = efi.dki_length;
1493 	/*
1494 	 * Some clients may attempt to request a PMBR for the
1495 	 * zvol.  Currently this interface will return EINVAL to
1496 	 * such requests.  These requests could be supported by
1497 	 * adding a check for lba == 0 and consing up an appropriate
1498 	 * PMBR.
1499 	 */
1500 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1501 		return (SET_ERROR(EINVAL));
1502 
1503 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1504 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1505 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1506 
1507 	if (efi.dki_lba == 1) {
1508 		efi_gpt_t gpt = { 0 };
1509 
1510 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1511 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1512 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1513 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1514 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1515 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1516 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1517 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1518 		gpt.efi_gpt_SizeOfPartitionEntry =
1519 		    LE_32(sizeof (efi_gpe_t));
1520 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1521 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1522 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1523 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1524 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1525 		    flag))
1526 			return (SET_ERROR(EFAULT));
1527 		ptr += sizeof (gpt);
1528 		length -= sizeof (gpt);
1529 	}
1530 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1531 	    length), flag))
1532 		return (SET_ERROR(EFAULT));
1533 	return (0);
1534 }
1535 
1536 /*
1537  * BEGIN entry points to allow external callers access to the volume.
1538  */
1539 /*
1540  * Return the volume parameters needed for access from an external caller.
1541  * These values are invariant as long as the volume is held open.
1542  */
1543 int
1544 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1545     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1546     void **rl_hdl, void **dnode_hdl)
1547 {
1548 	zvol_state_t *zv;
1549 
1550 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1551 	if (zv == NULL)
1552 		return (SET_ERROR(ENXIO));
1553 	if (zv->zv_flags & ZVOL_DUMPIFIED)
1554 		return (SET_ERROR(ENXIO));
1555 
1556 	ASSERT(blksize && max_xfer_len && minor_hdl &&
1557 	    objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
1558 
1559 	*blksize = zv->zv_volblocksize;
1560 	*max_xfer_len = (uint64_t)zvol_maxphys;
1561 	*minor_hdl = zv;
1562 	*objset_hdl = zv->zv_objset;
1563 	*zil_hdl = zv->zv_zilog;
1564 	*rl_hdl = &zv->zv_rangelock;
1565 	*dnode_hdl = zv->zv_dn;
1566 	return (0);
1567 }
1568 
1569 /*
1570  * Return the current volume size to an external caller.
1571  * The size can change while the volume is open.
1572  */
1573 uint64_t
1574 zvol_get_volume_size(void *minor_hdl)
1575 {
1576 	zvol_state_t *zv = minor_hdl;
1577 
1578 	return (zv->zv_volsize);
1579 }
1580 
1581 /*
1582  * Return the current WCE setting to an external caller.
1583  * The WCE setting can change while the volume is open.
1584  */
1585 int
1586 zvol_get_volume_wce(void *minor_hdl)
1587 {
1588 	zvol_state_t *zv = minor_hdl;
1589 
1590 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1591 }
1592 
1593 /*
1594  * Entry point for external callers to zvol_log_write
1595  */
1596 void
1597 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1598     boolean_t sync)
1599 {
1600 	zvol_state_t *zv = minor_hdl;
1601 
1602 	zvol_log_write(zv, tx, off, resid, sync);
1603 }
1604 /*
1605  * END entry points to allow external callers access to the volume.
1606  */
1607 
1608 /*
1609  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1610  */
1611 static void
1612 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1613     boolean_t sync)
1614 {
1615 	itx_t *itx;
1616 	lr_truncate_t *lr;
1617 	zilog_t *zilog = zv->zv_zilog;
1618 
1619 	if (zil_replaying(zilog, tx))
1620 		return;
1621 
1622 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1623 	lr = (lr_truncate_t *)&itx->itx_lr;
1624 	lr->lr_foid = ZVOL_OBJ;
1625 	lr->lr_offset = off;
1626 	lr->lr_length = len;
1627 
1628 	itx->itx_sync = sync;
1629 	zil_itx_assign(zilog, itx, tx);
1630 }
1631 
1632 /*
1633  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1634  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1635  */
1636 /*ARGSUSED*/
1637 int
1638 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1639 {
1640 	zvol_state_t *zv;
1641 	struct dk_callback *dkc;
1642 	int error = 0;
1643 	locked_range_t *lr;
1644 
1645 	mutex_enter(&zfsdev_state_lock);
1646 
1647 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1648 
1649 	if (zv == NULL) {
1650 		mutex_exit(&zfsdev_state_lock);
1651 		return (SET_ERROR(ENXIO));
1652 	}
1653 	ASSERT(zv->zv_total_opens > 0);
1654 
1655 	switch (cmd) {
1656 
1657 	case DKIOCINFO:
1658 	{
1659 		struct dk_cinfo dki;
1660 
1661 		bzero(&dki, sizeof (dki));
1662 		(void) strcpy(dki.dki_cname, "zvol");
1663 		(void) strcpy(dki.dki_dname, "zvol");
1664 		dki.dki_ctype = DKC_UNKNOWN;
1665 		dki.dki_unit = getminor(dev);
1666 		dki.dki_maxtransfer =
1667 		    1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
1668 		mutex_exit(&zfsdev_state_lock);
1669 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1670 			error = SET_ERROR(EFAULT);
1671 		return (error);
1672 	}
1673 
1674 	case DKIOCGMEDIAINFO:
1675 	{
1676 		struct dk_minfo dkm;
1677 
1678 		bzero(&dkm, sizeof (dkm));
1679 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1680 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1681 		dkm.dki_media_type = DK_UNKNOWN;
1682 		mutex_exit(&zfsdev_state_lock);
1683 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1684 			error = SET_ERROR(EFAULT);
1685 		return (error);
1686 	}
1687 
1688 	case DKIOCGMEDIAINFOEXT:
1689 	{
1690 		struct dk_minfo_ext dkmext;
1691 
1692 		bzero(&dkmext, sizeof (dkmext));
1693 		dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1694 		dkmext.dki_pbsize = zv->zv_volblocksize;
1695 		dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1696 		dkmext.dki_media_type = DK_UNKNOWN;
1697 		mutex_exit(&zfsdev_state_lock);
1698 		if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1699 			error = SET_ERROR(EFAULT);
1700 		return (error);
1701 	}
1702 
1703 	case DKIOCGETEFI:
1704 	{
1705 		uint64_t vs = zv->zv_volsize;
1706 		uint8_t bs = zv->zv_min_bs;
1707 
1708 		mutex_exit(&zfsdev_state_lock);
1709 		error = zvol_getefi((void *)arg, flag, vs, bs);
1710 		return (error);
1711 	}
1712 
1713 	case DKIOCFLUSHWRITECACHE:
1714 		dkc = (struct dk_callback *)arg;
1715 		mutex_exit(&zfsdev_state_lock);
1716 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1717 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1718 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1719 			error = 0;
1720 		}
1721 		return (error);
1722 
1723 	case DKIOCGETWCE:
1724 	{
1725 		int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1726 		if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1727 		    flag))
1728 			error = SET_ERROR(EFAULT);
1729 		break;
1730 	}
1731 	case DKIOCSETWCE:
1732 	{
1733 		int wce;
1734 		if (ddi_copyin((void *)arg, &wce, sizeof (int),
1735 		    flag)) {
1736 			error = SET_ERROR(EFAULT);
1737 			break;
1738 		}
1739 		if (wce) {
1740 			zv->zv_flags |= ZVOL_WCE;
1741 			mutex_exit(&zfsdev_state_lock);
1742 		} else {
1743 			zv->zv_flags &= ~ZVOL_WCE;
1744 			mutex_exit(&zfsdev_state_lock);
1745 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1746 		}
1747 		return (0);
1748 	}
1749 
1750 	case DKIOCGGEOM:
1751 	case DKIOCGVTOC:
1752 		/*
1753 		 * commands using these (like prtvtoc) expect ENOTSUP
1754 		 * since we're emulating an EFI label
1755 		 */
1756 		error = SET_ERROR(ENOTSUP);
1757 		break;
1758 
1759 	case DKIOCDUMPINIT:
1760 		lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
1761 		    RL_WRITER);
1762 		error = zvol_dumpify(zv);
1763 		rangelock_exit(lr);
1764 		break;
1765 
1766 	case DKIOCDUMPFINI:
1767 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1768 			break;
1769 		lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
1770 		    RL_WRITER);
1771 		error = zvol_dump_fini(zv);
1772 		rangelock_exit(lr);
1773 		break;
1774 
1775 	case DKIOCFREE:
1776 	{
1777 		dkioc_free_list_t *dfl;
1778 		dmu_tx_t *tx;
1779 
1780 		if (!zvol_unmap_enabled)
1781 			break;
1782 
1783 		if (!(flag & FKIOCTL)) {
1784 			error = dfl_copyin((void *)arg, &dfl, flag, KM_SLEEP);
1785 			if (error != 0)
1786 				break;
1787 		} else {
1788 			dfl = (dkioc_free_list_t *)arg;
1789 			ASSERT3U(dfl->dfl_num_exts, <=, DFL_COPYIN_MAX_EXTS);
1790 			if (dfl->dfl_num_exts > DFL_COPYIN_MAX_EXTS) {
1791 				error = SET_ERROR(EINVAL);
1792 				break;
1793 			}
1794 		}
1795 
1796 		mutex_exit(&zfsdev_state_lock);
1797 
1798 		for (int i = 0; i < dfl->dfl_num_exts; i++) {
1799 			uint64_t start = dfl->dfl_exts[i].dfle_start,
1800 			    length = dfl->dfl_exts[i].dfle_length,
1801 			    end = start + length;
1802 
1803 			/*
1804 			 * Apply Postel's Law to length-checking.  If they
1805 			 * overshoot, just blank out until the end, if there's
1806 			 * a need to blank out anything.
1807 			 */
1808 			if (start >= zv->zv_volsize)
1809 				continue;	/* No need to do anything... */
1810 			if (end > zv->zv_volsize) {
1811 				end = DMU_OBJECT_END;
1812 				length = end - start;
1813 			}
1814 
1815 			lr = rangelock_enter(&zv->zv_rangelock, start, length,
1816 			    RL_WRITER);
1817 			tx = dmu_tx_create(zv->zv_objset);
1818 			error = dmu_tx_assign(tx, TXG_WAIT);
1819 			if (error != 0) {
1820 				dmu_tx_abort(tx);
1821 			} else {
1822 				zvol_log_truncate(zv, tx, start, length,
1823 				    B_TRUE);
1824 				dmu_tx_commit(tx);
1825 				error = dmu_free_long_range(zv->zv_objset,
1826 				    ZVOL_OBJ, start, length);
1827 			}
1828 
1829 			rangelock_exit(lr);
1830 
1831 			if (error != 0)
1832 				break;
1833 		}
1834 
1835 		/*
1836 		 * If the write-cache is disabled, 'sync' property
1837 		 * is set to 'always', or if the caller is asking for
1838 		 * a synchronous free, commit this operation to the zil.
1839 		 * This will sync any previous uncommitted writes to the
1840 		 * zvol object.
1841 		 * Can be overridden by the zvol_unmap_sync_enabled tunable.
1842 		 */
1843 		if ((error == 0) && zvol_unmap_sync_enabled &&
1844 		    (!(zv->zv_flags & ZVOL_WCE) ||
1845 		    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
1846 		    (dfl->dfl_flags & DF_WAIT_SYNC))) {
1847 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1848 		}
1849 
1850 		if (!(flag & FKIOCTL))
1851 			dfl_free(dfl);
1852 
1853 		return (error);
1854 	}
1855 
1856 	default:
1857 		error = SET_ERROR(ENOTTY);
1858 		break;
1859 
1860 	}
1861 	mutex_exit(&zfsdev_state_lock);
1862 	return (error);
1863 }
1864 
1865 int
1866 zvol_busy(void)
1867 {
1868 	return (zvol_minors != 0);
1869 }
1870 
1871 void
1872 zvol_init(void)
1873 {
1874 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1875 	    1) == 0);
1876 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1877 }
1878 
1879 void
1880 zvol_fini(void)
1881 {
1882 	mutex_destroy(&zfsdev_state_lock);
1883 	ddi_soft_state_fini(&zfsdev_state);
1884 }
1885 
1886 /*ARGSUSED*/
1887 static int
1888 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1889 {
1890 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1891 
1892 	if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1893 		return (1);
1894 	return (0);
1895 }
1896 
1897 /*ARGSUSED*/
1898 static void
1899 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1900 {
1901 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1902 
1903 	spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1904 }
1905 
1906 static int
1907 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1908 {
1909 	dmu_tx_t *tx;
1910 	int error;
1911 	objset_t *os = zv->zv_objset;
1912 	spa_t *spa = dmu_objset_spa(os);
1913 	vdev_t *vd = spa->spa_root_vdev;
1914 	nvlist_t *nv = NULL;
1915 	uint64_t version = spa_version(spa);
1916 	uint64_t checksum, compress, refresrv, vbs, dedup;
1917 
1918 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1919 	ASSERT(vd->vdev_ops == &vdev_root_ops);
1920 
1921 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1922 	    DMU_OBJECT_END);
1923 	if (error != 0)
1924 		return (error);
1925 	/* wait for dmu_free_long_range to actually free the blocks */
1926 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1927 
1928 	/*
1929 	 * If the pool on which the dump device is being initialized has more
1930 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1931 	 * enabled.  If so, bump that feature's counter to indicate that the
1932 	 * feature is active. We also check the vdev type to handle the
1933 	 * following case:
1934 	 *   # zpool create test raidz disk1 disk2 disk3
1935 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1936 	 *   the raidz vdev itself has 3 children.
1937 	 */
1938 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1939 		if (!spa_feature_is_enabled(spa,
1940 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1941 			return (SET_ERROR(ENOTSUP));
1942 		(void) dsl_sync_task(spa_name(spa),
1943 		    zfs_mvdev_dump_feature_check,
1944 		    zfs_mvdev_dump_activate_feature_sync, NULL,
1945 		    2, ZFS_SPACE_CHECK_RESERVED);
1946 	}
1947 
1948 	if (!resize) {
1949 		error = dsl_prop_get_integer(zv->zv_name,
1950 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1951 		if (error == 0) {
1952 			error = dsl_prop_get_integer(zv->zv_name,
1953 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
1954 			    NULL);
1955 		}
1956 		if (error == 0) {
1957 			error = dsl_prop_get_integer(zv->zv_name,
1958 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
1959 			    &refresrv, NULL);
1960 		}
1961 		if (error == 0) {
1962 			error = dsl_prop_get_integer(zv->zv_name,
1963 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
1964 			    NULL);
1965 		}
1966 		if (version >= SPA_VERSION_DEDUP && error == 0) {
1967 			error = dsl_prop_get_integer(zv->zv_name,
1968 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1969 		}
1970 	}
1971 	if (error != 0)
1972 		return (error);
1973 
1974 	tx = dmu_tx_create(os);
1975 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1976 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1977 	error = dmu_tx_assign(tx, TXG_WAIT);
1978 	if (error != 0) {
1979 		dmu_tx_abort(tx);
1980 		return (error);
1981 	}
1982 
1983 	/*
1984 	 * If we are resizing the dump device then we only need to
1985 	 * update the refreservation to match the newly updated
1986 	 * zvolsize. Otherwise, we save off the original state of the
1987 	 * zvol so that we can restore them if the zvol is ever undumpified.
1988 	 */
1989 	if (resize) {
1990 		error = zap_update(os, ZVOL_ZAP_OBJ,
1991 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1992 		    &zv->zv_volsize, tx);
1993 	} else {
1994 		error = zap_update(os, ZVOL_ZAP_OBJ,
1995 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1996 		    &compress, tx);
1997 		if (error == 0) {
1998 			error = zap_update(os, ZVOL_ZAP_OBJ,
1999 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2000 			    &checksum, tx);
2001 		}
2002 		if (error == 0) {
2003 			error = zap_update(os, ZVOL_ZAP_OBJ,
2004 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2005 			    &refresrv, tx);
2006 		}
2007 		if (error == 0) {
2008 			error = zap_update(os, ZVOL_ZAP_OBJ,
2009 			    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2010 			    &vbs, tx);
2011 		}
2012 		if (error == 0) {
2013 			error = dmu_object_set_blocksize(
2014 			    os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2015 		}
2016 		if (version >= SPA_VERSION_DEDUP && error == 0) {
2017 			error = zap_update(os, ZVOL_ZAP_OBJ,
2018 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2019 			    &dedup, tx);
2020 		}
2021 		if (error == 0)
2022 			zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2023 	}
2024 	dmu_tx_commit(tx);
2025 
2026 	/*
2027 	 * We only need update the zvol's property if we are initializing
2028 	 * the dump area for the first time.
2029 	 */
2030 	if (error == 0 && !resize) {
2031 		/*
2032 		 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2033 		 * function.  Otherwise, use the old default -- OFF.
2034 		 */
2035 		checksum = spa_feature_is_active(spa,
2036 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2037 		    ZIO_CHECKSUM_OFF;
2038 
2039 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2040 		VERIFY(nvlist_add_uint64(nv,
2041 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2042 		VERIFY(nvlist_add_uint64(nv,
2043 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2044 		    ZIO_COMPRESS_OFF) == 0);
2045 		VERIFY(nvlist_add_uint64(nv,
2046 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2047 		    checksum) == 0);
2048 		if (version >= SPA_VERSION_DEDUP) {
2049 			VERIFY(nvlist_add_uint64(nv,
2050 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
2051 			    ZIO_CHECKSUM_OFF) == 0);
2052 		}
2053 
2054 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2055 		    nv, NULL);
2056 		nvlist_free(nv);
2057 	}
2058 
2059 	/* Allocate the space for the dump */
2060 	if (error == 0)
2061 		error = zvol_prealloc(zv);
2062 	return (error);
2063 }
2064 
2065 static int
2066 zvol_dumpify(zvol_state_t *zv)
2067 {
2068 	int error = 0;
2069 	uint64_t dumpsize = 0;
2070 	dmu_tx_t *tx;
2071 	objset_t *os = zv->zv_objset;
2072 
2073 	if (zv->zv_flags & ZVOL_RDONLY)
2074 		return (SET_ERROR(EROFS));
2075 
2076 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2077 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2078 		boolean_t resize = (dumpsize > 0);
2079 
2080 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2081 			(void) zvol_dump_fini(zv);
2082 			return (error);
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * Build up our lba mapping.
2088 	 */
2089 	error = zvol_get_lbas(zv);
2090 	if (error) {
2091 		(void) zvol_dump_fini(zv);
2092 		return (error);
2093 	}
2094 
2095 	tx = dmu_tx_create(os);
2096 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2097 	error = dmu_tx_assign(tx, TXG_WAIT);
2098 	if (error) {
2099 		dmu_tx_abort(tx);
2100 		(void) zvol_dump_fini(zv);
2101 		return (error);
2102 	}
2103 
2104 	zv->zv_flags |= ZVOL_DUMPIFIED;
2105 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2106 	    &zv->zv_volsize, tx);
2107 	dmu_tx_commit(tx);
2108 
2109 	if (error) {
2110 		(void) zvol_dump_fini(zv);
2111 		return (error);
2112 	}
2113 
2114 	txg_wait_synced(dmu_objset_pool(os), 0);
2115 	return (0);
2116 }
2117 
2118 static int
2119 zvol_dump_fini(zvol_state_t *zv)
2120 {
2121 	dmu_tx_t *tx;
2122 	objset_t *os = zv->zv_objset;
2123 	nvlist_t *nv;
2124 	int error = 0;
2125 	uint64_t checksum, compress, refresrv, vbs, dedup;
2126 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2127 
2128 	/*
2129 	 * Attempt to restore the zvol back to its pre-dumpified state.
2130 	 * This is a best-effort attempt as it's possible that not all
2131 	 * of these properties were initialized during the dumpify process
2132 	 * (i.e. error during zvol_dump_init).
2133 	 */
2134 
2135 	tx = dmu_tx_create(os);
2136 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2137 	error = dmu_tx_assign(tx, TXG_WAIT);
2138 	if (error) {
2139 		dmu_tx_abort(tx);
2140 		return (error);
2141 	}
2142 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2143 	dmu_tx_commit(tx);
2144 
2145 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2146 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2147 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2148 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2149 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2150 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2151 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2152 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2153 
2154 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2155 	(void) nvlist_add_uint64(nv,
2156 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2157 	(void) nvlist_add_uint64(nv,
2158 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2159 	(void) nvlist_add_uint64(nv,
2160 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2161 	if (version >= SPA_VERSION_DEDUP &&
2162 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2163 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2164 		(void) nvlist_add_uint64(nv,
2165 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2166 	}
2167 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2168 	    nv, NULL);
2169 	nvlist_free(nv);
2170 
2171 	zvol_free_extents(zv);
2172 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2173 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2174 	/* wait for dmu_free_long_range to actually free the blocks */
2175 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2176 	tx = dmu_tx_create(os);
2177 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2178 	error = dmu_tx_assign(tx, TXG_WAIT);
2179 	if (error) {
2180 		dmu_tx_abort(tx);
2181 		return (error);
2182 	}
2183 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2184 		zv->zv_volblocksize = vbs;
2185 	dmu_tx_commit(tx);
2186 
2187 	return (0);
2188 }
2189