xref: /titanic_41/usr/src/uts/common/fs/zfs/zvol.c (revision e521049d0fdbd76727f036cb77c99c98d7ee1125)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  *
24  * Portions Copyright 2010 Robert Milkowski
25  *
26  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
27  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
28  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29  */
30 
31 /*
32  * ZFS volume emulation driver.
33  *
34  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
35  * Volumes are accessed through the symbolic links named:
36  *
37  * /dev/zvol/dsk/<pool_name>/<dataset_name>
38  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
39  *
40  * These links are created by the /dev filesystem (sdev_zvolops.c).
41  * Volumes are persistent through reboot.  No user command needs to be
42  * run before opening and using a device.
43  */
44 
45 #include <sys/types.h>
46 #include <sys/param.h>
47 #include <sys/errno.h>
48 #include <sys/uio.h>
49 #include <sys/buf.h>
50 #include <sys/modctl.h>
51 #include <sys/open.h>
52 #include <sys/kmem.h>
53 #include <sys/conf.h>
54 #include <sys/cmn_err.h>
55 #include <sys/stat.h>
56 #include <sys/zap.h>
57 #include <sys/spa.h>
58 #include <sys/spa_impl.h>
59 #include <sys/zio.h>
60 #include <sys/dmu_traverse.h>
61 #include <sys/dnode.h>
62 #include <sys/dsl_dataset.h>
63 #include <sys/dsl_prop.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/byteorder.h>
67 #include <sys/pathname.h>
68 #include <sys/ddi.h>
69 #include <sys/sunddi.h>
70 #include <sys/crc32.h>
71 #include <sys/dirent.h>
72 #include <sys/policy.h>
73 #include <sys/fs/zfs.h>
74 #include <sys/zfs_ioctl.h>
75 #include <sys/mkdev.h>
76 #include <sys/zil.h>
77 #include <sys/refcount.h>
78 #include <sys/zfs_znode.h>
79 #include <sys/zfs_rlock.h>
80 #include <sys/vdev_disk.h>
81 #include <sys/vdev_impl.h>
82 #include <sys/vdev_raidz.h>
83 #include <sys/zvol.h>
84 #include <sys/dumphdr.h>
85 #include <sys/zil_impl.h>
86 #include <sys/dbuf.h>
87 #include <sys/zfs_events.h>
88 #include <sys/dmu_tx.h>
89 #include <sys/zfeature.h>
90 #include <sys/zio_checksum.h>
91 
92 #include "zfs_namecheck.h"
93 
94 void *zfsdev_state;
95 static char *zvol_tag = "zvol_tag";
96 
97 #define	ZVOL_DUMPSIZE		"dumpsize"
98 
99 /*
100  * This lock protects the zfsdev_state structure from being modified
101  * while it's being used, e.g. an open that comes in before a create
102  * finishes.  It also protects temporary opens of the dataset so that,
103  * e.g., an open doesn't get a spurious EBUSY.
104  */
105 kmutex_t zfsdev_state_lock;
106 static uint32_t zvol_minors;
107 
108 typedef struct zvol_extent {
109 	list_node_t	ze_node;
110 	dva_t		ze_dva;		/* dva associated with this extent */
111 	uint64_t	ze_nblks;	/* number of blocks in extent */
112 } zvol_extent_t;
113 
114 /*
115  * The in-core state of each volume.
116  */
117 typedef struct zvol_state {
118 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
119 	uint64_t	zv_volsize;	/* amount of space we advertise */
120 	uint64_t	zv_volblocksize; /* volume block size */
121 	minor_t		zv_minor;	/* minor number */
122 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
123 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
124 	objset_t	*zv_objset;	/* objset handle */
125 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
126 	uint32_t	zv_total_opens;	/* total open count */
127 	zilog_t		*zv_zilog;	/* ZIL handle */
128 	list_t		zv_extents;	/* List of extents for dump */
129 	znode_t		zv_znode;	/* for range locking */
130 	dmu_buf_t	*zv_dbuf;	/* bonus handle */
131 } zvol_state_t;
132 
133 /*
134  * zvol specific flags
135  */
136 #define	ZVOL_RDONLY	0x1
137 #define	ZVOL_DUMPIFIED	0x2
138 #define	ZVOL_EXCL	0x4
139 #define	ZVOL_WCE	0x8
140 
141 /*
142  * zvol maximum transfer in one DMU tx.
143  */
144 int zvol_maxphys = DMU_MAX_ACCESS/2;
145 
146 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
147     nvlist_t *, nvlist_t *);
148 static int zvol_remove_zv(zvol_state_t *);
149 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
150 static int zvol_dumpify(zvol_state_t *zv);
151 static int zvol_dump_fini(zvol_state_t *zv);
152 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
153 
154 static void
155 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
156 {
157 	dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
158 
159 	zv->zv_volsize = volsize;
160 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
161 	    "Size", volsize) == DDI_SUCCESS);
162 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
163 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
164 
165 	/* Notify specfs to invalidate the cached size */
166 	spec_size_invalidate(dev, VBLK);
167 	spec_size_invalidate(dev, VCHR);
168 }
169 
170 int
171 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
172 {
173 	if (volsize == 0)
174 		return (SET_ERROR(EINVAL));
175 
176 	if (volsize % blocksize != 0)
177 		return (SET_ERROR(EINVAL));
178 
179 #ifdef _ILP32
180 	if (volsize - 1 > SPEC_MAXOFFSET_T)
181 		return (SET_ERROR(EOVERFLOW));
182 #endif
183 	return (0);
184 }
185 
186 int
187 zvol_check_volblocksize(uint64_t volblocksize)
188 {
189 	if (volblocksize < SPA_MINBLOCKSIZE ||
190 	    volblocksize > SPA_MAXBLOCKSIZE ||
191 	    !ISP2(volblocksize))
192 		return (SET_ERROR(EDOM));
193 
194 	return (0);
195 }
196 
197 int
198 zvol_get_stats(objset_t *os, nvlist_t *nv)
199 {
200 	int error;
201 	dmu_object_info_t doi;
202 	uint64_t val;
203 
204 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
205 	if (error)
206 		return (error);
207 
208 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
209 
210 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
211 
212 	if (error == 0) {
213 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
214 		    doi.doi_data_block_size);
215 	}
216 
217 	return (error);
218 }
219 
220 static zvol_state_t *
221 zvol_minor_lookup(const char *name)
222 {
223 	minor_t minor;
224 	zvol_state_t *zv;
225 
226 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
227 
228 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
229 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
230 		if (zv == NULL)
231 			continue;
232 		if (strcmp(zv->zv_name, name) == 0)
233 			return (zv);
234 	}
235 
236 	return (NULL);
237 }
238 
239 /* extent mapping arg */
240 struct maparg {
241 	zvol_state_t	*ma_zv;
242 	uint64_t	ma_blks;
243 };
244 
245 /*ARGSUSED*/
246 static int
247 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
248     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
249 {
250 	struct maparg *ma = arg;
251 	zvol_extent_t *ze;
252 	int bs = ma->ma_zv->zv_volblocksize;
253 
254 	if (BP_IS_HOLE(bp) ||
255 	    zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
256 		return (0);
257 
258 	VERIFY(!BP_IS_EMBEDDED(bp));
259 
260 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
261 	ma->ma_blks++;
262 
263 	/* Abort immediately if we have encountered gang blocks */
264 	if (BP_IS_GANG(bp))
265 		return (SET_ERROR(EFRAGS));
266 
267 	/*
268 	 * See if the block is at the end of the previous extent.
269 	 */
270 	ze = list_tail(&ma->ma_zv->zv_extents);
271 	if (ze &&
272 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
273 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
274 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
275 		ze->ze_nblks++;
276 		return (0);
277 	}
278 
279 	dprintf_bp(bp, "%s", "next blkptr:");
280 
281 	/* start a new extent */
282 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
283 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
284 	ze->ze_nblks = 1;
285 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
286 	return (0);
287 }
288 
289 static void
290 zvol_free_extents(zvol_state_t *zv)
291 {
292 	zvol_extent_t *ze;
293 
294 	while (ze = list_head(&zv->zv_extents)) {
295 		list_remove(&zv->zv_extents, ze);
296 		kmem_free(ze, sizeof (zvol_extent_t));
297 	}
298 }
299 
300 static int
301 zvol_get_lbas(zvol_state_t *zv)
302 {
303 	objset_t *os = zv->zv_objset;
304 	struct maparg	ma;
305 	int		err;
306 
307 	ma.ma_zv = zv;
308 	ma.ma_blks = 0;
309 	zvol_free_extents(zv);
310 
311 	/* commit any in-flight changes before traversing the dataset */
312 	txg_wait_synced(dmu_objset_pool(os), 0);
313 	err = traverse_dataset(dmu_objset_ds(os), 0,
314 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
315 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
316 		zvol_free_extents(zv);
317 		return (err ? err : EIO);
318 	}
319 
320 	return (0);
321 }
322 
323 /* ARGSUSED */
324 void
325 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
326 {
327 	zfs_creat_t *zct = arg;
328 	nvlist_t *nvprops = zct->zct_props;
329 	int error;
330 	uint64_t volblocksize, volsize;
331 
332 	VERIFY(nvlist_lookup_uint64(nvprops,
333 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
334 	if (nvlist_lookup_uint64(nvprops,
335 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
336 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
337 
338 	/*
339 	 * These properties must be removed from the list so the generic
340 	 * property setting step won't apply to them.
341 	 */
342 	VERIFY(nvlist_remove_all(nvprops,
343 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
344 	(void) nvlist_remove_all(nvprops,
345 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
346 
347 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
348 	    DMU_OT_NONE, 0, tx);
349 	ASSERT(error == 0);
350 
351 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
352 	    DMU_OT_NONE, 0, tx);
353 	ASSERT(error == 0);
354 
355 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
356 	ASSERT(error == 0);
357 }
358 
359 /*
360  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
361  * implement DKIOCFREE/free-long-range.
362  */
363 static int
364 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
365 {
366 	uint64_t offset, length;
367 
368 	if (byteswap)
369 		byteswap_uint64_array(lr, sizeof (*lr));
370 
371 	offset = lr->lr_offset;
372 	length = lr->lr_length;
373 
374 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
375 }
376 
377 /*
378  * Replay a TX_WRITE ZIL transaction that didn't get committed
379  * after a system failure
380  */
381 static int
382 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
383 {
384 	objset_t *os = zv->zv_objset;
385 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
386 	uint64_t offset, length;
387 	dmu_tx_t *tx;
388 	int error;
389 
390 	if (byteswap)
391 		byteswap_uint64_array(lr, sizeof (*lr));
392 
393 	offset = lr->lr_offset;
394 	length = lr->lr_length;
395 
396 	/* If it's a dmu_sync() block, write the whole block */
397 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
398 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
399 		if (length < blocksize) {
400 			offset -= offset % blocksize;
401 			length = blocksize;
402 		}
403 	}
404 
405 	tx = dmu_tx_create(os);
406 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
407 	error = dmu_tx_assign(tx, TXG_WAIT);
408 	if (error) {
409 		dmu_tx_abort(tx);
410 	} else {
411 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
412 		dmu_tx_commit(tx);
413 	}
414 
415 	return (error);
416 }
417 
418 /* ARGSUSED */
419 static int
420 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
421 {
422 	return (SET_ERROR(ENOTSUP));
423 }
424 
425 /*
426  * Callback vectors for replaying records.
427  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
428  */
429 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
430 	zvol_replay_err,	/* 0 no such transaction type */
431 	zvol_replay_err,	/* TX_CREATE */
432 	zvol_replay_err,	/* TX_MKDIR */
433 	zvol_replay_err,	/* TX_MKXATTR */
434 	zvol_replay_err,	/* TX_SYMLINK */
435 	zvol_replay_err,	/* TX_REMOVE */
436 	zvol_replay_err,	/* TX_RMDIR */
437 	zvol_replay_err,	/* TX_LINK */
438 	zvol_replay_err,	/* TX_RENAME */
439 	zvol_replay_write,	/* TX_WRITE */
440 	zvol_replay_truncate,	/* TX_TRUNCATE */
441 	zvol_replay_err,	/* TX_SETATTR */
442 	zvol_replay_err,	/* TX_ACL */
443 	zvol_replay_err,	/* TX_CREATE_ACL */
444 	zvol_replay_err,	/* TX_CREATE_ATTR */
445 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
446 	zvol_replay_err,	/* TX_MKDIR_ACL */
447 	zvol_replay_err,	/* TX_MKDIR_ATTR */
448 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
449 	zvol_replay_err,	/* TX_WRITE2 */
450 };
451 
452 int
453 zvol_name2minor(const char *name, minor_t *minor)
454 {
455 	zvol_state_t *zv;
456 
457 	mutex_enter(&zfsdev_state_lock);
458 	zv = zvol_minor_lookup(name);
459 	if (minor && zv)
460 		*minor = zv->zv_minor;
461 	mutex_exit(&zfsdev_state_lock);
462 	return (zv ? 0 : -1);
463 }
464 
465 /*
466  * Create a minor node (plus a whole lot more) for the specified volume.
467  */
468 int
469 zvol_create_minor(const char *name)
470 {
471 	zfs_soft_state_t *zs;
472 	zvol_state_t *zv;
473 	objset_t *os;
474 	dmu_object_info_t doi;
475 	minor_t minor = 0;
476 	char chrbuf[30], blkbuf[30];
477 	int error;
478 
479 	mutex_enter(&zfsdev_state_lock);
480 
481 	if (zvol_minor_lookup(name) != NULL) {
482 		mutex_exit(&zfsdev_state_lock);
483 		return (SET_ERROR(EEXIST));
484 	}
485 
486 	/* lie and say we're read-only */
487 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
488 
489 	if (error) {
490 		mutex_exit(&zfsdev_state_lock);
491 		return (error);
492 	}
493 
494 	if ((minor = zfsdev_minor_alloc()) == 0) {
495 		dmu_objset_disown(os, FTAG);
496 		mutex_exit(&zfsdev_state_lock);
497 		return (SET_ERROR(ENXIO));
498 	}
499 
500 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
501 		dmu_objset_disown(os, FTAG);
502 		mutex_exit(&zfsdev_state_lock);
503 		return (SET_ERROR(EAGAIN));
504 	}
505 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
506 	    (char *)name);
507 
508 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
509 
510 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
511 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
512 		ddi_soft_state_free(zfsdev_state, minor);
513 		dmu_objset_disown(os, FTAG);
514 		mutex_exit(&zfsdev_state_lock);
515 		return (SET_ERROR(EAGAIN));
516 	}
517 
518 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
519 
520 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
521 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
522 		ddi_remove_minor_node(zfs_dip, chrbuf);
523 		ddi_soft_state_free(zfsdev_state, minor);
524 		dmu_objset_disown(os, FTAG);
525 		mutex_exit(&zfsdev_state_lock);
526 		return (SET_ERROR(EAGAIN));
527 	}
528 
529 	zs = ddi_get_soft_state(zfsdev_state, minor);
530 	zs->zss_type = ZSST_ZVOL;
531 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
532 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
533 	zv->zv_min_bs = DEV_BSHIFT;
534 	zv->zv_minor = minor;
535 	zv->zv_objset = os;
536 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
537 		zv->zv_flags |= ZVOL_RDONLY;
538 	mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
539 	avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
540 	    sizeof (rl_t), offsetof(rl_t, r_node));
541 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
542 	    offsetof(zvol_extent_t, ze_node));
543 	/* get and cache the blocksize */
544 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
545 	ASSERT(error == 0);
546 	zv->zv_volblocksize = doi.doi_data_block_size;
547 
548 	if (spa_writeable(dmu_objset_spa(os))) {
549 		if (zil_replay_disable)
550 			zil_destroy(dmu_objset_zil(os), B_FALSE);
551 		else
552 			zil_replay(os, zv, zvol_replay_vector);
553 	}
554 	dmu_objset_disown(os, FTAG);
555 	zv->zv_objset = NULL;
556 
557 	zvol_minors++;
558 
559 	mutex_exit(&zfsdev_state_lock);
560 
561 	return (0);
562 }
563 
564 /*
565  * Remove minor node for the specified volume.
566  */
567 static int
568 zvol_remove_zv(zvol_state_t *zv)
569 {
570 	char nmbuf[20];
571 	minor_t minor = zv->zv_minor;
572 
573 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
574 	if (zv->zv_total_opens != 0)
575 		return (SET_ERROR(EBUSY));
576 
577 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
578 	ddi_remove_minor_node(zfs_dip, nmbuf);
579 
580 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
581 	ddi_remove_minor_node(zfs_dip, nmbuf);
582 
583 	avl_destroy(&zv->zv_znode.z_range_avl);
584 	mutex_destroy(&zv->zv_znode.z_range_lock);
585 
586 	kmem_free(zv, sizeof (zvol_state_t));
587 
588 	ddi_soft_state_free(zfsdev_state, minor);
589 
590 	zvol_minors--;
591 	return (0);
592 }
593 
594 int
595 zvol_remove_minor(const char *name)
596 {
597 	zvol_state_t *zv;
598 	int rc;
599 
600 	mutex_enter(&zfsdev_state_lock);
601 	if ((zv = zvol_minor_lookup(name)) == NULL) {
602 		mutex_exit(&zfsdev_state_lock);
603 		return (SET_ERROR(ENXIO));
604 	}
605 	rc = zvol_remove_zv(zv);
606 	mutex_exit(&zfsdev_state_lock);
607 	return (rc);
608 }
609 
610 int
611 zvol_first_open(zvol_state_t *zv)
612 {
613 	objset_t *os;
614 	uint64_t volsize;
615 	int error;
616 	uint64_t readonly;
617 
618 	/* lie and say we're read-only */
619 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
620 	    zvol_tag, &os);
621 	if (error)
622 		return (error);
623 
624 	zv->zv_objset = os;
625 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
626 	if (error) {
627 		ASSERT(error == 0);
628 		dmu_objset_disown(os, zvol_tag);
629 		return (error);
630 	}
631 
632 	error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
633 	if (error) {
634 		dmu_objset_disown(os, zvol_tag);
635 		return (error);
636 	}
637 
638 	zvol_size_changed(zv, volsize);
639 	zv->zv_zilog = zil_open(os, zvol_get_data);
640 
641 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
642 	    NULL) == 0);
643 	if (readonly || dmu_objset_is_snapshot(os) ||
644 	    !spa_writeable(dmu_objset_spa(os)))
645 		zv->zv_flags |= ZVOL_RDONLY;
646 	else
647 		zv->zv_flags &= ~ZVOL_RDONLY;
648 	return (error);
649 }
650 
651 void
652 zvol_last_close(zvol_state_t *zv)
653 {
654 	zil_close(zv->zv_zilog);
655 	zv->zv_zilog = NULL;
656 
657 	dmu_buf_rele(zv->zv_dbuf, zvol_tag);
658 	zv->zv_dbuf = NULL;
659 
660 	/*
661 	 * Evict cached data
662 	 */
663 	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
664 	    !(zv->zv_flags & ZVOL_RDONLY))
665 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
666 	dmu_objset_evict_dbufs(zv->zv_objset);
667 
668 	dmu_objset_disown(zv->zv_objset, zvol_tag);
669 	zv->zv_objset = NULL;
670 }
671 
672 int
673 zvol_prealloc(zvol_state_t *zv)
674 {
675 	objset_t *os = zv->zv_objset;
676 	dmu_tx_t *tx;
677 	uint64_t refd, avail, usedobjs, availobjs;
678 	uint64_t resid = zv->zv_volsize;
679 	uint64_t off = 0;
680 
681 	/* Check the space usage before attempting to allocate the space */
682 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
683 	if (avail < zv->zv_volsize)
684 		return (SET_ERROR(ENOSPC));
685 
686 	/* Free old extents if they exist */
687 	zvol_free_extents(zv);
688 
689 	while (resid != 0) {
690 		int error;
691 		uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
692 
693 		tx = dmu_tx_create(os);
694 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
695 		error = dmu_tx_assign(tx, TXG_WAIT);
696 		if (error) {
697 			dmu_tx_abort(tx);
698 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
699 			return (error);
700 		}
701 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
702 		dmu_tx_commit(tx);
703 		off += bytes;
704 		resid -= bytes;
705 	}
706 	txg_wait_synced(dmu_objset_pool(os), 0);
707 
708 	return (0);
709 }
710 
711 static int
712 zvol_update_volsize(objset_t *os, uint64_t volsize)
713 {
714 	dmu_tx_t *tx;
715 	int error;
716 
717 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
718 
719 	tx = dmu_tx_create(os);
720 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
721 	dmu_tx_mark_netfree(tx);
722 	error = dmu_tx_assign(tx, TXG_WAIT);
723 	if (error) {
724 		dmu_tx_abort(tx);
725 		return (error);
726 	}
727 
728 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
729 	    &volsize, tx);
730 	dmu_tx_commit(tx);
731 
732 	if (error == 0)
733 		error = dmu_free_long_range(os,
734 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
735 	return (error);
736 }
737 
738 void
739 zvol_remove_minors(const char *name)
740 {
741 	zvol_state_t *zv;
742 	char *namebuf;
743 	minor_t minor;
744 
745 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
746 	(void) strncpy(namebuf, name, strlen(name));
747 	(void) strcat(namebuf, "/");
748 	mutex_enter(&zfsdev_state_lock);
749 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
750 
751 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
752 		if (zv == NULL)
753 			continue;
754 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
755 			(void) zvol_remove_zv(zv);
756 	}
757 	kmem_free(namebuf, strlen(name) + 2);
758 
759 	mutex_exit(&zfsdev_state_lock);
760 }
761 
762 static int
763 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
764 {
765 	uint64_t old_volsize = 0ULL;
766 	int error = 0;
767 
768 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
769 
770 	/*
771 	 * Reinitialize the dump area to the new size. If we
772 	 * failed to resize the dump area then restore it back to
773 	 * its original size.  We must set the new volsize prior
774 	 * to calling dumpvp_resize() to ensure that the devices'
775 	 * size(9P) is not visible by the dump subsystem.
776 	 */
777 	old_volsize = zv->zv_volsize;
778 	zvol_size_changed(zv, volsize);
779 
780 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
781 		if ((error = zvol_dumpify(zv)) != 0 ||
782 		    (error = dumpvp_resize()) != 0) {
783 			int dumpify_error;
784 
785 			(void) zvol_update_volsize(zv->zv_objset, old_volsize);
786 			zvol_size_changed(zv, old_volsize);
787 			dumpify_error = zvol_dumpify(zv);
788 			error = dumpify_error ? dumpify_error : error;
789 		}
790 	}
791 
792 	/*
793 	 * Generate a LUN expansion event.
794 	 */
795 	if (error == 0) {
796 		sysevent_id_t eid;
797 		nvlist_t *attr;
798 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
799 
800 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
801 		    zv->zv_minor);
802 
803 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
804 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
805 
806 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
807 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
808 
809 		nvlist_free(attr);
810 		kmem_free(physpath, MAXPATHLEN);
811 	}
812 	return (error);
813 }
814 
815 int
816 zvol_set_volsize(const char *name, uint64_t volsize)
817 {
818 	zvol_state_t *zv = NULL;
819 	objset_t *os;
820 	int error;
821 	dmu_object_info_t doi;
822 	uint64_t readonly;
823 	boolean_t owned = B_FALSE;
824 
825 	error = dsl_prop_get_integer(name,
826 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
827 	if (error != 0)
828 		return (error);
829 	if (readonly)
830 		return (SET_ERROR(EROFS));
831 
832 	mutex_enter(&zfsdev_state_lock);
833 	zv = zvol_minor_lookup(name);
834 
835 	if (zv == NULL || zv->zv_objset == NULL) {
836 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
837 		    FTAG, &os)) != 0) {
838 			mutex_exit(&zfsdev_state_lock);
839 			return (error);
840 		}
841 		owned = B_TRUE;
842 		if (zv != NULL)
843 			zv->zv_objset = os;
844 	} else {
845 		os = zv->zv_objset;
846 	}
847 
848 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
849 	    (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
850 		goto out;
851 
852 	error = zvol_update_volsize(os, volsize);
853 
854 	if (error == 0 && zv != NULL)
855 		error = zvol_update_live_volsize(zv, volsize);
856 out:
857 	if (owned) {
858 		dmu_objset_disown(os, FTAG);
859 		if (zv != NULL)
860 			zv->zv_objset = NULL;
861 	}
862 	mutex_exit(&zfsdev_state_lock);
863 	return (error);
864 }
865 
866 /*ARGSUSED*/
867 int
868 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
869 {
870 	zvol_state_t *zv;
871 	int err = 0;
872 
873 	mutex_enter(&zfsdev_state_lock);
874 
875 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
876 	if (zv == NULL) {
877 		mutex_exit(&zfsdev_state_lock);
878 		return (SET_ERROR(ENXIO));
879 	}
880 
881 	if (zv->zv_total_opens == 0)
882 		err = zvol_first_open(zv);
883 	if (err) {
884 		mutex_exit(&zfsdev_state_lock);
885 		return (err);
886 	}
887 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
888 		err = SET_ERROR(EROFS);
889 		goto out;
890 	}
891 	if (zv->zv_flags & ZVOL_EXCL) {
892 		err = SET_ERROR(EBUSY);
893 		goto out;
894 	}
895 	if (flag & FEXCL) {
896 		if (zv->zv_total_opens != 0) {
897 			err = SET_ERROR(EBUSY);
898 			goto out;
899 		}
900 		zv->zv_flags |= ZVOL_EXCL;
901 	}
902 
903 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
904 		zv->zv_open_count[otyp]++;
905 		zv->zv_total_opens++;
906 	}
907 	mutex_exit(&zfsdev_state_lock);
908 
909 	return (err);
910 out:
911 	if (zv->zv_total_opens == 0)
912 		zvol_last_close(zv);
913 	mutex_exit(&zfsdev_state_lock);
914 	return (err);
915 }
916 
917 /*ARGSUSED*/
918 int
919 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
920 {
921 	minor_t minor = getminor(dev);
922 	zvol_state_t *zv;
923 	int error = 0;
924 
925 	mutex_enter(&zfsdev_state_lock);
926 
927 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
928 	if (zv == NULL) {
929 		mutex_exit(&zfsdev_state_lock);
930 		return (SET_ERROR(ENXIO));
931 	}
932 
933 	if (zv->zv_flags & ZVOL_EXCL) {
934 		ASSERT(zv->zv_total_opens == 1);
935 		zv->zv_flags &= ~ZVOL_EXCL;
936 	}
937 
938 	/*
939 	 * If the open count is zero, this is a spurious close.
940 	 * That indicates a bug in the kernel / DDI framework.
941 	 */
942 	ASSERT(zv->zv_open_count[otyp] != 0);
943 	ASSERT(zv->zv_total_opens != 0);
944 
945 	/*
946 	 * You may get multiple opens, but only one close.
947 	 */
948 	zv->zv_open_count[otyp]--;
949 	zv->zv_total_opens--;
950 
951 	if (zv->zv_total_opens == 0)
952 		zvol_last_close(zv);
953 
954 	mutex_exit(&zfsdev_state_lock);
955 	return (error);
956 }
957 
958 static void
959 zvol_get_done(zgd_t *zgd, int error)
960 {
961 	if (zgd->zgd_db)
962 		dmu_buf_rele(zgd->zgd_db, zgd);
963 
964 	zfs_range_unlock(zgd->zgd_rl);
965 
966 	if (error == 0 && zgd->zgd_bp)
967 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
968 
969 	kmem_free(zgd, sizeof (zgd_t));
970 }
971 
972 /*
973  * Get data to generate a TX_WRITE intent log record.
974  */
975 static int
976 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
977 {
978 	zvol_state_t *zv = arg;
979 	objset_t *os = zv->zv_objset;
980 	uint64_t object = ZVOL_OBJ;
981 	uint64_t offset = lr->lr_offset;
982 	uint64_t size = lr->lr_length;	/* length of user data */
983 	blkptr_t *bp = &lr->lr_blkptr;
984 	dmu_buf_t *db;
985 	zgd_t *zgd;
986 	int error;
987 
988 	ASSERT(zio != NULL);
989 	ASSERT(size != 0);
990 
991 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
992 	zgd->zgd_zilog = zv->zv_zilog;
993 	zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
994 
995 	/*
996 	 * Write records come in two flavors: immediate and indirect.
997 	 * For small writes it's cheaper to store the data with the
998 	 * log record (immediate); for large writes it's cheaper to
999 	 * sync the data and get a pointer to it (indirect) so that
1000 	 * we don't have to write the data twice.
1001 	 */
1002 	if (buf != NULL) {	/* immediate write */
1003 		error = dmu_read(os, object, offset, size, buf,
1004 		    DMU_READ_NO_PREFETCH);
1005 	} else {
1006 		size = zv->zv_volblocksize;
1007 		offset = P2ALIGN(offset, size);
1008 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1009 		    DMU_READ_NO_PREFETCH);
1010 		if (error == 0) {
1011 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1012 			if (obp) {
1013 				ASSERT(BP_IS_HOLE(bp));
1014 				*bp = *obp;
1015 			}
1016 
1017 			zgd->zgd_db = db;
1018 			zgd->zgd_bp = bp;
1019 
1020 			ASSERT(db->db_offset == offset);
1021 			ASSERT(db->db_size == size);
1022 
1023 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1024 			    zvol_get_done, zgd);
1025 
1026 			if (error == 0)
1027 				return (0);
1028 		}
1029 	}
1030 
1031 	zvol_get_done(zgd, error);
1032 
1033 	return (error);
1034 }
1035 
1036 /*
1037  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1038  *
1039  * We store data in the log buffers if it's small enough.
1040  * Otherwise we will later flush the data out via dmu_sync().
1041  */
1042 ssize_t zvol_immediate_write_sz = 32768;
1043 
1044 static void
1045 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1046     boolean_t sync)
1047 {
1048 	uint32_t blocksize = zv->zv_volblocksize;
1049 	zilog_t *zilog = zv->zv_zilog;
1050 	boolean_t slogging;
1051 	ssize_t immediate_write_sz;
1052 
1053 	if (zil_replaying(zilog, tx))
1054 		return;
1055 
1056 	immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1057 	    ? 0 : zvol_immediate_write_sz;
1058 
1059 	slogging = spa_has_slogs(zilog->zl_spa) &&
1060 	    (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1061 
1062 	while (resid) {
1063 		itx_t *itx;
1064 		lr_write_t *lr;
1065 		ssize_t len;
1066 		itx_wr_state_t write_state;
1067 
1068 		/*
1069 		 * Unlike zfs_log_write() we can be called with
1070 		 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1071 		 */
1072 		if (blocksize > immediate_write_sz && !slogging &&
1073 		    resid >= blocksize && off % blocksize == 0) {
1074 			write_state = WR_INDIRECT; /* uses dmu_sync */
1075 			len = blocksize;
1076 		} else if (sync) {
1077 			write_state = WR_COPIED;
1078 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1079 		} else {
1080 			write_state = WR_NEED_COPY;
1081 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1082 		}
1083 
1084 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1085 		    (write_state == WR_COPIED ? len : 0));
1086 		lr = (lr_write_t *)&itx->itx_lr;
1087 		if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1088 		    ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1089 			zil_itx_destroy(itx);
1090 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1091 			lr = (lr_write_t *)&itx->itx_lr;
1092 			write_state = WR_NEED_COPY;
1093 		}
1094 
1095 		itx->itx_wr_state = write_state;
1096 		if (write_state == WR_NEED_COPY)
1097 			itx->itx_sod += len;
1098 		lr->lr_foid = ZVOL_OBJ;
1099 		lr->lr_offset = off;
1100 		lr->lr_length = len;
1101 		lr->lr_blkoff = 0;
1102 		BP_ZERO(&lr->lr_blkptr);
1103 
1104 		itx->itx_private = zv;
1105 		itx->itx_sync = sync;
1106 
1107 		zil_itx_assign(zilog, itx, tx);
1108 
1109 		rw_enter(&rz_zev_rwlock, RW_READER);
1110 		if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_zvol_write)
1111 			rz_zev_callbacks->rz_zev_zvol_write(zv->zv_name,
1112 			    zv->zv_objset, tx, off, len);
1113 		rw_exit(&rz_zev_rwlock);
1114 
1115 		off += len;
1116 		resid -= len;
1117 	}
1118 }
1119 
1120 static int
1121 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1122     uint64_t size, boolean_t doread, boolean_t isdump)
1123 {
1124 	vdev_disk_t *dvd;
1125 	int c;
1126 	int numerrors = 0;
1127 
1128 	if (vd->vdev_ops == &vdev_mirror_ops ||
1129 	    vd->vdev_ops == &vdev_replacing_ops ||
1130 	    vd->vdev_ops == &vdev_spare_ops) {
1131 		for (c = 0; c < vd->vdev_children; c++) {
1132 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1133 			    addr, offset, origoffset, size, doread, isdump);
1134 			if (err != 0) {
1135 				numerrors++;
1136 			} else if (doread) {
1137 				break;
1138 			}
1139 		}
1140 	}
1141 
1142 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1143 		return (numerrors < vd->vdev_children ? 0 : EIO);
1144 
1145 	if (doread && !vdev_readable(vd))
1146 		return (SET_ERROR(EIO));
1147 	else if (!doread && !vdev_writeable(vd))
1148 		return (SET_ERROR(EIO));
1149 
1150 	if (vd->vdev_ops == &vdev_raidz_ops) {
1151 		return (vdev_raidz_physio(vd,
1152 		    addr, size, offset, origoffset, doread, isdump));
1153 	}
1154 
1155 	offset += VDEV_LABEL_START_SIZE;
1156 
1157 	if (ddi_in_panic() || isdump) {
1158 		ASSERT(!doread);
1159 		if (doread)
1160 			return (SET_ERROR(EIO));
1161 		dvd = vd->vdev_tsd;
1162 		ASSERT3P(dvd, !=, NULL);
1163 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1164 		    lbtodb(size)));
1165 	} else {
1166 		dvd = vd->vdev_tsd;
1167 		ASSERT3P(dvd, !=, NULL);
1168 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1169 		    offset, doread ? B_READ : B_WRITE));
1170 	}
1171 }
1172 
1173 static int
1174 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1175     boolean_t doread, boolean_t isdump)
1176 {
1177 	vdev_t *vd;
1178 	int error;
1179 	zvol_extent_t *ze;
1180 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1181 
1182 	/* Must be sector aligned, and not stradle a block boundary. */
1183 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1184 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1185 		return (SET_ERROR(EINVAL));
1186 	}
1187 	ASSERT(size <= zv->zv_volblocksize);
1188 
1189 	/* Locate the extent this belongs to */
1190 	ze = list_head(&zv->zv_extents);
1191 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1192 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1193 		ze = list_next(&zv->zv_extents, ze);
1194 	}
1195 
1196 	if (ze == NULL)
1197 		return (SET_ERROR(EINVAL));
1198 
1199 	if (!ddi_in_panic())
1200 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1201 
1202 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1203 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1204 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1205 	    size, doread, isdump);
1206 
1207 	if (!ddi_in_panic())
1208 		spa_config_exit(spa, SCL_STATE, FTAG);
1209 
1210 	return (error);
1211 }
1212 
1213 int
1214 zvol_strategy(buf_t *bp)
1215 {
1216 	zfs_soft_state_t *zs = NULL;
1217 	zvol_state_t *zv;
1218 	uint64_t off, volsize;
1219 	size_t resid;
1220 	char *addr;
1221 	objset_t *os;
1222 	rl_t *rl;
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 	rl = zfs_range_lock(&zv->zv_znode, 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 	zfs_range_unlock(rl);
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 	rl_t *rl;
1382 	int error = 0;
1383 
1384 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1385 	if (zv == NULL)
1386 		return (SET_ERROR(ENXIO));
1387 
1388 	volsize = zv->zv_volsize;
1389 	if (uio->uio_resid > 0 &&
1390 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1391 		return (SET_ERROR(EIO));
1392 
1393 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1394 		error = physio(zvol_strategy, NULL, dev, B_READ,
1395 		    zvol_minphys, uio);
1396 		return (error);
1397 	}
1398 
1399 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1400 	    RL_READER);
1401 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1402 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1403 
1404 		/* don't read past the end */
1405 		if (bytes > volsize - uio->uio_loffset)
1406 			bytes = volsize - uio->uio_loffset;
1407 
1408 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1409 		if (error) {
1410 			/* convert checksum errors into IO errors */
1411 			if (error == ECKSUM)
1412 				error = SET_ERROR(EIO);
1413 			break;
1414 		}
1415 	}
1416 	zfs_range_unlock(rl);
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 	rl_t *rl;
1428 	int error = 0;
1429 	boolean_t sync;
1430 
1431 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1432 	if (zv == NULL)
1433 		return (SET_ERROR(ENXIO));
1434 
1435 	volsize = zv->zv_volsize;
1436 	if (uio->uio_resid > 0 &&
1437 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1438 		return (SET_ERROR(EIO));
1439 
1440 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1441 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1442 		    zvol_minphys, uio);
1443 		return (error);
1444 	}
1445 
1446 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1447 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1448 
1449 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1450 	    RL_WRITER);
1451 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1452 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1453 		uint64_t off = uio->uio_loffset;
1454 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1455 
1456 		if (bytes > volsize - off)	/* don't write past the end */
1457 			bytes = volsize - off;
1458 
1459 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1460 		error = dmu_tx_assign(tx, TXG_WAIT);
1461 		if (error) {
1462 			dmu_tx_abort(tx);
1463 			break;
1464 		}
1465 		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1466 		if (error == 0)
1467 			zvol_log_write(zv, tx, off, bytes, sync);
1468 		dmu_tx_commit(tx);
1469 
1470 		if (error)
1471 			break;
1472 	}
1473 	zfs_range_unlock(rl);
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 **bonus_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 && bonus_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_znode;
1565 	*bonus_hdl = zv->zv_dbuf;
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 	rw_enter(&rz_zev_rwlock, RW_READER);
1632 	if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_zvol_truncate)
1633 		rz_zev_callbacks->rz_zev_zvol_truncate(zv->zv_name,
1634 		    zv->zv_objset, tx, off, len);
1635 	rw_exit(&rz_zev_rwlock);
1636 }
1637 
1638 /*
1639  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1640  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1641  */
1642 /*ARGSUSED*/
1643 int
1644 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1645 {
1646 	zvol_state_t *zv;
1647 	struct dk_callback *dkc;
1648 	int error = 0;
1649 	rl_t *rl;
1650 
1651 	mutex_enter(&zfsdev_state_lock);
1652 
1653 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1654 
1655 	if (zv == NULL) {
1656 		mutex_exit(&zfsdev_state_lock);
1657 		return (SET_ERROR(ENXIO));
1658 	}
1659 	ASSERT(zv->zv_total_opens > 0);
1660 
1661 	switch (cmd) {
1662 
1663 	case DKIOCINFO:
1664 	{
1665 		struct dk_cinfo dki;
1666 
1667 		bzero(&dki, sizeof (dki));
1668 		(void) strcpy(dki.dki_cname, "zvol");
1669 		(void) strcpy(dki.dki_dname, "zvol");
1670 		dki.dki_ctype = DKC_UNKNOWN;
1671 		dki.dki_unit = getminor(dev);
1672 		dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1673 		mutex_exit(&zfsdev_state_lock);
1674 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1675 			error = SET_ERROR(EFAULT);
1676 		return (error);
1677 	}
1678 
1679 	case DKIOCGMEDIAINFO:
1680 	{
1681 		struct dk_minfo dkm;
1682 
1683 		bzero(&dkm, sizeof (dkm));
1684 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1685 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1686 		dkm.dki_media_type = DK_UNKNOWN;
1687 		mutex_exit(&zfsdev_state_lock);
1688 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1689 			error = SET_ERROR(EFAULT);
1690 		return (error);
1691 	}
1692 
1693 	case DKIOCGMEDIAINFOEXT:
1694 	{
1695 		struct dk_minfo_ext dkmext;
1696 
1697 		bzero(&dkmext, sizeof (dkmext));
1698 		dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1699 		dkmext.dki_pbsize = zv->zv_volblocksize;
1700 		dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1701 		dkmext.dki_media_type = DK_UNKNOWN;
1702 		mutex_exit(&zfsdev_state_lock);
1703 		if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1704 			error = SET_ERROR(EFAULT);
1705 		return (error);
1706 	}
1707 
1708 	case DKIOCGETEFI:
1709 	{
1710 		uint64_t vs = zv->zv_volsize;
1711 		uint8_t bs = zv->zv_min_bs;
1712 
1713 		mutex_exit(&zfsdev_state_lock);
1714 		error = zvol_getefi((void *)arg, flag, vs, bs);
1715 		return (error);
1716 	}
1717 
1718 	case DKIOCFLUSHWRITECACHE:
1719 		dkc = (struct dk_callback *)arg;
1720 		mutex_exit(&zfsdev_state_lock);
1721 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1722 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1723 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1724 			error = 0;
1725 		}
1726 		return (error);
1727 
1728 	case DKIOCGETWCE:
1729 	{
1730 		int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1731 		if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1732 		    flag))
1733 			error = SET_ERROR(EFAULT);
1734 		break;
1735 	}
1736 	case DKIOCSETWCE:
1737 	{
1738 		int wce;
1739 		if (ddi_copyin((void *)arg, &wce, sizeof (int),
1740 		    flag)) {
1741 			error = SET_ERROR(EFAULT);
1742 			break;
1743 		}
1744 		if (wce) {
1745 			zv->zv_flags |= ZVOL_WCE;
1746 			mutex_exit(&zfsdev_state_lock);
1747 		} else {
1748 			zv->zv_flags &= ~ZVOL_WCE;
1749 			mutex_exit(&zfsdev_state_lock);
1750 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1751 		}
1752 		return (0);
1753 	}
1754 
1755 	case DKIOCGGEOM:
1756 	case DKIOCGVTOC:
1757 		/*
1758 		 * commands using these (like prtvtoc) expect ENOTSUP
1759 		 * since we're emulating an EFI label
1760 		 */
1761 		error = SET_ERROR(ENOTSUP);
1762 		break;
1763 
1764 	case DKIOCDUMPINIT:
1765 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1766 		    RL_WRITER);
1767 		error = zvol_dumpify(zv);
1768 		zfs_range_unlock(rl);
1769 		break;
1770 
1771 	case DKIOCDUMPFINI:
1772 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1773 			break;
1774 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1775 		    RL_WRITER);
1776 		error = zvol_dump_fini(zv);
1777 		zfs_range_unlock(rl);
1778 		break;
1779 
1780 	case DKIOCFREE:
1781 	{
1782 		dkioc_free_t df;
1783 		dmu_tx_t *tx;
1784 
1785 		if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1786 			error = SET_ERROR(EFAULT);
1787 			break;
1788 		}
1789 
1790 		/*
1791 		 * Apply Postel's Law to length-checking.  If they overshoot,
1792 		 * just blank out until the end, if there's a need to blank
1793 		 * out anything.
1794 		 */
1795 		if (df.df_start >= zv->zv_volsize)
1796 			break;	/* No need to do anything... */
1797 		if (df.df_start + df.df_length > zv->zv_volsize)
1798 			df.df_length = DMU_OBJECT_END;
1799 
1800 		rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1801 		    RL_WRITER);
1802 		tx = dmu_tx_create(zv->zv_objset);
1803 		dmu_tx_mark_netfree(tx);
1804 		error = dmu_tx_assign(tx, TXG_WAIT);
1805 		if (error != 0) {
1806 			dmu_tx_abort(tx);
1807 		} else {
1808 			zvol_log_truncate(zv, tx, df.df_start,
1809 			    df.df_length, B_TRUE);
1810 			dmu_tx_commit(tx);
1811 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1812 			    df.df_start, df.df_length);
1813 		}
1814 
1815 		zfs_range_unlock(rl);
1816 
1817 		if (error == 0) {
1818 			/*
1819 			 * If the write-cache is disabled or 'sync' property
1820 			 * is set to 'always' then treat this as a synchronous
1821 			 * operation (i.e. commit to zil).
1822 			 */
1823 			if (!(zv->zv_flags & ZVOL_WCE) ||
1824 			    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
1825 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1826 
1827 			/*
1828 			 * If the caller really wants synchronous writes, and
1829 			 * can't wait for them, don't return until the write
1830 			 * is done.
1831 			 */
1832 			if (df.df_flags & DF_WAIT_SYNC) {
1833 				txg_wait_synced(
1834 				    dmu_objset_pool(zv->zv_objset), 0);
1835 			}
1836 		}
1837 		break;
1838 	}
1839 
1840 	default:
1841 		error = SET_ERROR(ENOTTY);
1842 		break;
1843 
1844 	}
1845 	mutex_exit(&zfsdev_state_lock);
1846 	return (error);
1847 }
1848 
1849 int
1850 zvol_busy(void)
1851 {
1852 	return (zvol_minors != 0);
1853 }
1854 
1855 void
1856 zvol_init(void)
1857 {
1858 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1859 	    1) == 0);
1860 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1861 }
1862 
1863 void
1864 zvol_fini(void)
1865 {
1866 	mutex_destroy(&zfsdev_state_lock);
1867 	ddi_soft_state_fini(&zfsdev_state);
1868 }
1869 
1870 /*ARGSUSED*/
1871 static int
1872 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1873 {
1874 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1875 
1876 	if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1877 		return (1);
1878 	return (0);
1879 }
1880 
1881 /*ARGSUSED*/
1882 static void
1883 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1884 {
1885 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1886 
1887 	spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1888 }
1889 
1890 static int
1891 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1892 {
1893 	dmu_tx_t *tx;
1894 	int error;
1895 	objset_t *os = zv->zv_objset;
1896 	spa_t *spa = dmu_objset_spa(os);
1897 	vdev_t *vd = spa->spa_root_vdev;
1898 	nvlist_t *nv = NULL;
1899 	uint64_t version = spa_version(spa);
1900 	enum zio_checksum checksum;
1901 
1902 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1903 	ASSERT(vd->vdev_ops == &vdev_root_ops);
1904 
1905 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1906 	    DMU_OBJECT_END);
1907 	/* wait for dmu_free_long_range to actually free the blocks */
1908 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1909 
1910 	/*
1911 	 * If the pool on which the dump device is being initialized has more
1912 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1913 	 * enabled.  If so, bump that feature's counter to indicate that the
1914 	 * feature is active. We also check the vdev type to handle the
1915 	 * following case:
1916 	 *   # zpool create test raidz disk1 disk2 disk3
1917 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1918 	 *   the raidz vdev itself has 3 children.
1919 	 */
1920 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1921 		if (!spa_feature_is_enabled(spa,
1922 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1923 			return (SET_ERROR(ENOTSUP));
1924 		(void) dsl_sync_task(spa_name(spa),
1925 		    zfs_mvdev_dump_feature_check,
1926 		    zfs_mvdev_dump_activate_feature_sync, NULL,
1927 		    2, ZFS_SPACE_CHECK_RESERVED);
1928 	}
1929 
1930 	tx = dmu_tx_create(os);
1931 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1932 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1933 	error = dmu_tx_assign(tx, TXG_WAIT);
1934 	if (error) {
1935 		dmu_tx_abort(tx);
1936 		return (error);
1937 	}
1938 
1939 	/*
1940 	 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
1941 	 * function.  Otherwise, use the old default -- OFF.
1942 	 */
1943 	checksum = spa_feature_is_active(spa,
1944 	    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
1945 	    ZIO_CHECKSUM_OFF;
1946 
1947 	/*
1948 	 * If we are resizing the dump device then we only need to
1949 	 * update the refreservation to match the newly updated
1950 	 * zvolsize. Otherwise, we save off the original state of the
1951 	 * zvol so that we can restore them if the zvol is ever undumpified.
1952 	 */
1953 	if (resize) {
1954 		error = zap_update(os, ZVOL_ZAP_OBJ,
1955 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1956 		    &zv->zv_volsize, tx);
1957 	} else {
1958 		uint64_t checksum, compress, refresrv, vbs, dedup;
1959 
1960 		error = dsl_prop_get_integer(zv->zv_name,
1961 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1962 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1963 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1964 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1965 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1966 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1967 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1968 		if (version >= SPA_VERSION_DEDUP) {
1969 			error = error ? error :
1970 			    dsl_prop_get_integer(zv->zv_name,
1971 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1972 		}
1973 
1974 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1975 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1976 		    &compress, tx);
1977 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1978 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1979 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1980 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1981 		    &refresrv, tx);
1982 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1983 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1984 		    &vbs, tx);
1985 		error = error ? error : dmu_object_set_blocksize(
1986 		    os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1987 		if (version >= SPA_VERSION_DEDUP) {
1988 			error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1989 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1990 			    &dedup, tx);
1991 		}
1992 		if (error == 0)
1993 			zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1994 	}
1995 	dmu_tx_commit(tx);
1996 
1997 	/*
1998 	 * We only need update the zvol's property if we are initializing
1999 	 * the dump area for the first time.
2000 	 */
2001 	if (!resize) {
2002 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2003 		VERIFY(nvlist_add_uint64(nv,
2004 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2005 		VERIFY(nvlist_add_uint64(nv,
2006 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2007 		    ZIO_COMPRESS_OFF) == 0);
2008 		VERIFY(nvlist_add_uint64(nv,
2009 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2010 		    checksum) == 0);
2011 		if (version >= SPA_VERSION_DEDUP) {
2012 			VERIFY(nvlist_add_uint64(nv,
2013 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
2014 			    ZIO_CHECKSUM_OFF) == 0);
2015 		}
2016 
2017 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2018 		    nv, NULL);
2019 		nvlist_free(nv);
2020 
2021 		if (error)
2022 			return (error);
2023 	}
2024 
2025 	/* Allocate the space for the dump */
2026 	error = zvol_prealloc(zv);
2027 	return (error);
2028 }
2029 
2030 static int
2031 zvol_dumpify(zvol_state_t *zv)
2032 {
2033 	int error = 0;
2034 	uint64_t dumpsize = 0;
2035 	dmu_tx_t *tx;
2036 	objset_t *os = zv->zv_objset;
2037 
2038 	if (zv->zv_flags & ZVOL_RDONLY)
2039 		return (SET_ERROR(EROFS));
2040 
2041 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2042 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2043 		boolean_t resize = (dumpsize > 0);
2044 
2045 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2046 			(void) zvol_dump_fini(zv);
2047 			return (error);
2048 		}
2049 	}
2050 
2051 	/*
2052 	 * Build up our lba mapping.
2053 	 */
2054 	error = zvol_get_lbas(zv);
2055 	if (error) {
2056 		(void) zvol_dump_fini(zv);
2057 		return (error);
2058 	}
2059 
2060 	tx = dmu_tx_create(os);
2061 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2062 	error = dmu_tx_assign(tx, TXG_WAIT);
2063 	if (error) {
2064 		dmu_tx_abort(tx);
2065 		(void) zvol_dump_fini(zv);
2066 		return (error);
2067 	}
2068 
2069 	zv->zv_flags |= ZVOL_DUMPIFIED;
2070 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2071 	    &zv->zv_volsize, tx);
2072 	dmu_tx_commit(tx);
2073 
2074 	if (error) {
2075 		(void) zvol_dump_fini(zv);
2076 		return (error);
2077 	}
2078 
2079 	txg_wait_synced(dmu_objset_pool(os), 0);
2080 	return (0);
2081 }
2082 
2083 static int
2084 zvol_dump_fini(zvol_state_t *zv)
2085 {
2086 	dmu_tx_t *tx;
2087 	objset_t *os = zv->zv_objset;
2088 	nvlist_t *nv;
2089 	int error = 0;
2090 	uint64_t checksum, compress, refresrv, vbs, dedup;
2091 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2092 
2093 	/*
2094 	 * Attempt to restore the zvol back to its pre-dumpified state.
2095 	 * This is a best-effort attempt as it's possible that not all
2096 	 * of these properties were initialized during the dumpify process
2097 	 * (i.e. error during zvol_dump_init).
2098 	 */
2099 
2100 	tx = dmu_tx_create(os);
2101 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2102 	error = dmu_tx_assign(tx, TXG_WAIT);
2103 	if (error) {
2104 		dmu_tx_abort(tx);
2105 		return (error);
2106 	}
2107 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2108 	dmu_tx_commit(tx);
2109 
2110 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2111 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2112 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2113 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2114 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2115 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2116 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2117 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2118 
2119 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2120 	(void) nvlist_add_uint64(nv,
2121 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2122 	(void) nvlist_add_uint64(nv,
2123 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2124 	(void) nvlist_add_uint64(nv,
2125 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2126 	if (version >= SPA_VERSION_DEDUP &&
2127 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2128 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2129 		(void) nvlist_add_uint64(nv,
2130 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2131 	}
2132 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2133 	    nv, NULL);
2134 	nvlist_free(nv);
2135 
2136 	zvol_free_extents(zv);
2137 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2138 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2139 	/* wait for dmu_free_long_range to actually free the blocks */
2140 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2141 	tx = dmu_tx_create(os);
2142 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2143 	error = dmu_tx_assign(tx, TXG_WAIT);
2144 	if (error) {
2145 		dmu_tx_abort(tx);
2146 		return (error);
2147 	}
2148 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2149 		zv->zv_volblocksize = vbs;
2150 	dmu_tx_commit(tx);
2151 
2152 	return (0);
2153 }
2154