xref: /illumos-gate/usr/src/uts/common/fs/zfs/zvol.c (revision 522c768b8cd12cb4e80ed84c1eb5df0ed2fd2baa)
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) 2013 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/dmu_tx.h>
88 #include <sys/zfeature.h>
89 #include <sys/zio_checksum.h>
90 
91 #include "zfs_namecheck.h"
92 
93 void *zfsdev_state;
94 static char *zvol_tag = "zvol_tag";
95 
96 #define	ZVOL_DUMPSIZE		"dumpsize"
97 
98 /*
99  * This lock protects the zfsdev_state structure from being modified
100  * while it's being used, e.g. an open that comes in before a create
101  * finishes.  It also protects temporary opens of the dataset so that,
102  * e.g., an open doesn't get a spurious EBUSY.
103  */
104 kmutex_t zfsdev_state_lock;
105 static uint32_t zvol_minors;
106 
107 typedef struct zvol_extent {
108 	list_node_t	ze_node;
109 	dva_t		ze_dva;		/* dva associated with this extent */
110 	uint64_t	ze_nblks;	/* number of blocks in extent */
111 } zvol_extent_t;
112 
113 /*
114  * The in-core state of each volume.
115  */
116 typedef struct zvol_state {
117 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
118 	uint64_t	zv_volsize;	/* amount of space we advertise */
119 	uint64_t	zv_volblocksize; /* volume block size */
120 	minor_t		zv_minor;	/* minor number */
121 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
122 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
123 	objset_t	*zv_objset;	/* objset handle */
124 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
125 	uint32_t	zv_total_opens;	/* total open count */
126 	zilog_t		*zv_zilog;	/* ZIL handle */
127 	list_t		zv_extents;	/* List of extents for dump */
128 	znode_t		zv_znode;	/* for range locking */
129 	dmu_buf_t	*zv_dbuf;	/* bonus handle */
130 } zvol_state_t;
131 
132 /*
133  * zvol specific flags
134  */
135 #define	ZVOL_RDONLY	0x1
136 #define	ZVOL_DUMPIFIED	0x2
137 #define	ZVOL_EXCL	0x4
138 #define	ZVOL_WCE	0x8
139 
140 /*
141  * zvol maximum transfer in one DMU tx.
142  */
143 int zvol_maxphys = DMU_MAX_ACCESS/2;
144 
145 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
146     nvlist_t *, nvlist_t *);
147 static int zvol_remove_zv(zvol_state_t *);
148 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
149 static int zvol_dumpify(zvol_state_t *zv);
150 static int zvol_dump_fini(zvol_state_t *zv);
151 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
152 
153 static void
154 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
155 {
156 	dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
157 
158 	zv->zv_volsize = volsize;
159 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
160 	    "Size", volsize) == DDI_SUCCESS);
161 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
162 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
163 
164 	/* Notify specfs to invalidate the cached size */
165 	spec_size_invalidate(dev, VBLK);
166 	spec_size_invalidate(dev, VCHR);
167 }
168 
169 int
170 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
171 {
172 	if (volsize == 0)
173 		return (SET_ERROR(EINVAL));
174 
175 	if (volsize % blocksize != 0)
176 		return (SET_ERROR(EINVAL));
177 
178 #ifdef _ILP32
179 	if (volsize - 1 > SPEC_MAXOFFSET_T)
180 		return (SET_ERROR(EOVERFLOW));
181 #endif
182 	return (0);
183 }
184 
185 int
186 zvol_check_volblocksize(uint64_t volblocksize)
187 {
188 	if (volblocksize < SPA_MINBLOCKSIZE ||
189 	    volblocksize > SPA_MAXBLOCKSIZE ||
190 	    !ISP2(volblocksize))
191 		return (SET_ERROR(EDOM));
192 
193 	return (0);
194 }
195 
196 int
197 zvol_get_stats(objset_t *os, nvlist_t *nv)
198 {
199 	int error;
200 	dmu_object_info_t doi;
201 	uint64_t val;
202 
203 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
204 	if (error)
205 		return (error);
206 
207 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
208 
209 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
210 
211 	if (error == 0) {
212 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
213 		    doi.doi_data_block_size);
214 	}
215 
216 	return (error);
217 }
218 
219 static zvol_state_t *
220 zvol_minor_lookup(const char *name)
221 {
222 	minor_t minor;
223 	zvol_state_t *zv;
224 
225 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
226 
227 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
228 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
229 		if (zv == NULL)
230 			continue;
231 		if (strcmp(zv->zv_name, name) == 0)
232 			return (zv);
233 	}
234 
235 	return (NULL);
236 }
237 
238 /* extent mapping arg */
239 struct maparg {
240 	zvol_state_t	*ma_zv;
241 	uint64_t	ma_blks;
242 };
243 
244 /*ARGSUSED*/
245 static int
246 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
247     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
248 {
249 	struct maparg *ma = arg;
250 	zvol_extent_t *ze;
251 	int bs = ma->ma_zv->zv_volblocksize;
252 
253 	if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
254 		return (0);
255 
256 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
257 	ma->ma_blks++;
258 
259 	/* Abort immediately if we have encountered gang blocks */
260 	if (BP_IS_GANG(bp))
261 		return (SET_ERROR(EFRAGS));
262 
263 	/*
264 	 * See if the block is at the end of the previous extent.
265 	 */
266 	ze = list_tail(&ma->ma_zv->zv_extents);
267 	if (ze &&
268 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
269 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
270 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
271 		ze->ze_nblks++;
272 		return (0);
273 	}
274 
275 	dprintf_bp(bp, "%s", "next blkptr:");
276 
277 	/* start a new extent */
278 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
279 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
280 	ze->ze_nblks = 1;
281 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
282 	return (0);
283 }
284 
285 static void
286 zvol_free_extents(zvol_state_t *zv)
287 {
288 	zvol_extent_t *ze;
289 
290 	while (ze = list_head(&zv->zv_extents)) {
291 		list_remove(&zv->zv_extents, ze);
292 		kmem_free(ze, sizeof (zvol_extent_t));
293 	}
294 }
295 
296 static int
297 zvol_get_lbas(zvol_state_t *zv)
298 {
299 	objset_t *os = zv->zv_objset;
300 	struct maparg	ma;
301 	int		err;
302 
303 	ma.ma_zv = zv;
304 	ma.ma_blks = 0;
305 	zvol_free_extents(zv);
306 
307 	/* commit any in-flight changes before traversing the dataset */
308 	txg_wait_synced(dmu_objset_pool(os), 0);
309 	err = traverse_dataset(dmu_objset_ds(os), 0,
310 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
311 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
312 		zvol_free_extents(zv);
313 		return (err ? err : EIO);
314 	}
315 
316 	return (0);
317 }
318 
319 /* ARGSUSED */
320 void
321 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
322 {
323 	zfs_creat_t *zct = arg;
324 	nvlist_t *nvprops = zct->zct_props;
325 	int error;
326 	uint64_t volblocksize, volsize;
327 
328 	VERIFY(nvlist_lookup_uint64(nvprops,
329 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
330 	if (nvlist_lookup_uint64(nvprops,
331 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
332 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
333 
334 	/*
335 	 * These properties must be removed from the list so the generic
336 	 * property setting step won't apply to them.
337 	 */
338 	VERIFY(nvlist_remove_all(nvprops,
339 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
340 	(void) nvlist_remove_all(nvprops,
341 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
342 
343 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
344 	    DMU_OT_NONE, 0, tx);
345 	ASSERT(error == 0);
346 
347 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
348 	    DMU_OT_NONE, 0, tx);
349 	ASSERT(error == 0);
350 
351 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
352 	ASSERT(error == 0);
353 }
354 
355 /*
356  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
357  * implement DKIOCFREE/free-long-range.
358  */
359 static int
360 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
361 {
362 	uint64_t offset, length;
363 
364 	if (byteswap)
365 		byteswap_uint64_array(lr, sizeof (*lr));
366 
367 	offset = lr->lr_offset;
368 	length = lr->lr_length;
369 
370 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
371 }
372 
373 /*
374  * Replay a TX_WRITE ZIL transaction that didn't get committed
375  * after a system failure
376  */
377 static int
378 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
379 {
380 	objset_t *os = zv->zv_objset;
381 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
382 	uint64_t offset, length;
383 	dmu_tx_t *tx;
384 	int error;
385 
386 	if (byteswap)
387 		byteswap_uint64_array(lr, sizeof (*lr));
388 
389 	offset = lr->lr_offset;
390 	length = lr->lr_length;
391 
392 	/* If it's a dmu_sync() block, write the whole block */
393 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
394 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
395 		if (length < blocksize) {
396 			offset -= offset % blocksize;
397 			length = blocksize;
398 		}
399 	}
400 
401 	tx = dmu_tx_create(os);
402 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
403 	error = dmu_tx_assign(tx, TXG_WAIT);
404 	if (error) {
405 		dmu_tx_abort(tx);
406 	} else {
407 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
408 		dmu_tx_commit(tx);
409 	}
410 
411 	return (error);
412 }
413 
414 /* ARGSUSED */
415 static int
416 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
417 {
418 	return (SET_ERROR(ENOTSUP));
419 }
420 
421 /*
422  * Callback vectors for replaying records.
423  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
424  */
425 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
426 	zvol_replay_err,	/* 0 no such transaction type */
427 	zvol_replay_err,	/* TX_CREATE */
428 	zvol_replay_err,	/* TX_MKDIR */
429 	zvol_replay_err,	/* TX_MKXATTR */
430 	zvol_replay_err,	/* TX_SYMLINK */
431 	zvol_replay_err,	/* TX_REMOVE */
432 	zvol_replay_err,	/* TX_RMDIR */
433 	zvol_replay_err,	/* TX_LINK */
434 	zvol_replay_err,	/* TX_RENAME */
435 	zvol_replay_write,	/* TX_WRITE */
436 	zvol_replay_truncate,	/* TX_TRUNCATE */
437 	zvol_replay_err,	/* TX_SETATTR */
438 	zvol_replay_err,	/* TX_ACL */
439 	zvol_replay_err,	/* TX_CREATE_ACL */
440 	zvol_replay_err,	/* TX_CREATE_ATTR */
441 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
442 	zvol_replay_err,	/* TX_MKDIR_ACL */
443 	zvol_replay_err,	/* TX_MKDIR_ATTR */
444 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
445 	zvol_replay_err,	/* TX_WRITE2 */
446 };
447 
448 int
449 zvol_name2minor(const char *name, minor_t *minor)
450 {
451 	zvol_state_t *zv;
452 
453 	mutex_enter(&zfsdev_state_lock);
454 	zv = zvol_minor_lookup(name);
455 	if (minor && zv)
456 		*minor = zv->zv_minor;
457 	mutex_exit(&zfsdev_state_lock);
458 	return (zv ? 0 : -1);
459 }
460 
461 /*
462  * Create a minor node (plus a whole lot more) for the specified volume.
463  */
464 int
465 zvol_create_minor(const char *name)
466 {
467 	zfs_soft_state_t *zs;
468 	zvol_state_t *zv;
469 	objset_t *os;
470 	dmu_object_info_t doi;
471 	minor_t minor = 0;
472 	char chrbuf[30], blkbuf[30];
473 	int error;
474 
475 	mutex_enter(&zfsdev_state_lock);
476 
477 	if (zvol_minor_lookup(name) != NULL) {
478 		mutex_exit(&zfsdev_state_lock);
479 		return (SET_ERROR(EEXIST));
480 	}
481 
482 	/* lie and say we're read-only */
483 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
484 
485 	if (error) {
486 		mutex_exit(&zfsdev_state_lock);
487 		return (error);
488 	}
489 
490 	if ((minor = zfsdev_minor_alloc()) == 0) {
491 		dmu_objset_disown(os, FTAG);
492 		mutex_exit(&zfsdev_state_lock);
493 		return (SET_ERROR(ENXIO));
494 	}
495 
496 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
497 		dmu_objset_disown(os, FTAG);
498 		mutex_exit(&zfsdev_state_lock);
499 		return (SET_ERROR(EAGAIN));
500 	}
501 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
502 	    (char *)name);
503 
504 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
505 
506 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
507 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
508 		ddi_soft_state_free(zfsdev_state, minor);
509 		dmu_objset_disown(os, FTAG);
510 		mutex_exit(&zfsdev_state_lock);
511 		return (SET_ERROR(EAGAIN));
512 	}
513 
514 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
515 
516 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
517 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
518 		ddi_remove_minor_node(zfs_dip, chrbuf);
519 		ddi_soft_state_free(zfsdev_state, minor);
520 		dmu_objset_disown(os, FTAG);
521 		mutex_exit(&zfsdev_state_lock);
522 		return (SET_ERROR(EAGAIN));
523 	}
524 
525 	zs = ddi_get_soft_state(zfsdev_state, minor);
526 	zs->zss_type = ZSST_ZVOL;
527 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
528 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
529 	zv->zv_min_bs = DEV_BSHIFT;
530 	zv->zv_minor = minor;
531 	zv->zv_objset = os;
532 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
533 		zv->zv_flags |= ZVOL_RDONLY;
534 	mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
535 	avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
536 	    sizeof (rl_t), offsetof(rl_t, r_node));
537 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
538 	    offsetof(zvol_extent_t, ze_node));
539 	/* get and cache the blocksize */
540 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
541 	ASSERT(error == 0);
542 	zv->zv_volblocksize = doi.doi_data_block_size;
543 
544 	if (spa_writeable(dmu_objset_spa(os))) {
545 		if (zil_replay_disable)
546 			zil_destroy(dmu_objset_zil(os), B_FALSE);
547 		else
548 			zil_replay(os, zv, zvol_replay_vector);
549 	}
550 	dmu_objset_disown(os, FTAG);
551 	zv->zv_objset = NULL;
552 
553 	zvol_minors++;
554 
555 	mutex_exit(&zfsdev_state_lock);
556 
557 	return (0);
558 }
559 
560 /*
561  * Remove minor node for the specified volume.
562  */
563 static int
564 zvol_remove_zv(zvol_state_t *zv)
565 {
566 	char nmbuf[20];
567 	minor_t minor = zv->zv_minor;
568 
569 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
570 	if (zv->zv_total_opens != 0)
571 		return (SET_ERROR(EBUSY));
572 
573 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
574 	ddi_remove_minor_node(zfs_dip, nmbuf);
575 
576 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
577 	ddi_remove_minor_node(zfs_dip, nmbuf);
578 
579 	avl_destroy(&zv->zv_znode.z_range_avl);
580 	mutex_destroy(&zv->zv_znode.z_range_lock);
581 
582 	kmem_free(zv, sizeof (zvol_state_t));
583 
584 	ddi_soft_state_free(zfsdev_state, minor);
585 
586 	zvol_minors--;
587 	return (0);
588 }
589 
590 int
591 zvol_remove_minor(const char *name)
592 {
593 	zvol_state_t *zv;
594 	int rc;
595 
596 	mutex_enter(&zfsdev_state_lock);
597 	if ((zv = zvol_minor_lookup(name)) == NULL) {
598 		mutex_exit(&zfsdev_state_lock);
599 		return (SET_ERROR(ENXIO));
600 	}
601 	rc = zvol_remove_zv(zv);
602 	mutex_exit(&zfsdev_state_lock);
603 	return (rc);
604 }
605 
606 int
607 zvol_first_open(zvol_state_t *zv)
608 {
609 	objset_t *os;
610 	uint64_t volsize;
611 	int error;
612 	uint64_t readonly;
613 
614 	/* lie and say we're read-only */
615 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
616 	    zvol_tag, &os);
617 	if (error)
618 		return (error);
619 
620 	zv->zv_objset = os;
621 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
622 	if (error) {
623 		ASSERT(error == 0);
624 		dmu_objset_disown(os, zvol_tag);
625 		return (error);
626 	}
627 
628 	error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
629 	if (error) {
630 		dmu_objset_disown(os, zvol_tag);
631 		return (error);
632 	}
633 
634 	zvol_size_changed(zv, volsize);
635 	zv->zv_zilog = zil_open(os, zvol_get_data);
636 
637 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
638 	    NULL) == 0);
639 	if (readonly || dmu_objset_is_snapshot(os) ||
640 	    !spa_writeable(dmu_objset_spa(os)))
641 		zv->zv_flags |= ZVOL_RDONLY;
642 	else
643 		zv->zv_flags &= ~ZVOL_RDONLY;
644 	return (error);
645 }
646 
647 void
648 zvol_last_close(zvol_state_t *zv)
649 {
650 	zil_close(zv->zv_zilog);
651 	zv->zv_zilog = NULL;
652 
653 	dmu_buf_rele(zv->zv_dbuf, zvol_tag);
654 	zv->zv_dbuf = NULL;
655 
656 	/*
657 	 * Evict cached data
658 	 */
659 	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
660 	    !(zv->zv_flags & ZVOL_RDONLY))
661 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
662 	dmu_objset_evict_dbufs(zv->zv_objset);
663 
664 	dmu_objset_disown(zv->zv_objset, zvol_tag);
665 	zv->zv_objset = NULL;
666 }
667 
668 int
669 zvol_prealloc(zvol_state_t *zv)
670 {
671 	objset_t *os = zv->zv_objset;
672 	dmu_tx_t *tx;
673 	uint64_t refd, avail, usedobjs, availobjs;
674 	uint64_t resid = zv->zv_volsize;
675 	uint64_t off = 0;
676 
677 	/* Check the space usage before attempting to allocate the space */
678 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
679 	if (avail < zv->zv_volsize)
680 		return (SET_ERROR(ENOSPC));
681 
682 	/* Free old extents if they exist */
683 	zvol_free_extents(zv);
684 
685 	while (resid != 0) {
686 		int error;
687 		uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
688 
689 		tx = dmu_tx_create(os);
690 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
691 		error = dmu_tx_assign(tx, TXG_WAIT);
692 		if (error) {
693 			dmu_tx_abort(tx);
694 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
695 			return (error);
696 		}
697 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
698 		dmu_tx_commit(tx);
699 		off += bytes;
700 		resid -= bytes;
701 	}
702 	txg_wait_synced(dmu_objset_pool(os), 0);
703 
704 	return (0);
705 }
706 
707 static int
708 zvol_update_volsize(objset_t *os, uint64_t volsize)
709 {
710 	dmu_tx_t *tx;
711 	int error;
712 
713 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
714 
715 	tx = dmu_tx_create(os);
716 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
717 	error = dmu_tx_assign(tx, TXG_WAIT);
718 	if (error) {
719 		dmu_tx_abort(tx);
720 		return (error);
721 	}
722 
723 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
724 	    &volsize, tx);
725 	dmu_tx_commit(tx);
726 
727 	if (error == 0)
728 		error = dmu_free_long_range(os,
729 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
730 	return (error);
731 }
732 
733 void
734 zvol_remove_minors(const char *name)
735 {
736 	zvol_state_t *zv;
737 	char *namebuf;
738 	minor_t minor;
739 
740 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
741 	(void) strncpy(namebuf, name, strlen(name));
742 	(void) strcat(namebuf, "/");
743 	mutex_enter(&zfsdev_state_lock);
744 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
745 
746 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
747 		if (zv == NULL)
748 			continue;
749 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
750 			(void) zvol_remove_zv(zv);
751 	}
752 	kmem_free(namebuf, strlen(name) + 2);
753 
754 	mutex_exit(&zfsdev_state_lock);
755 }
756 
757 static int
758 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
759 {
760 	uint64_t old_volsize = 0ULL;
761 	int error = 0;
762 
763 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
764 
765 	/*
766 	 * Reinitialize the dump area to the new size. If we
767 	 * failed to resize the dump area then restore it back to
768 	 * its original size.  We must set the new volsize prior
769 	 * to calling dumpvp_resize() to ensure that the devices'
770 	 * size(9P) is not visible by the dump subsystem.
771 	 */
772 	old_volsize = zv->zv_volsize;
773 	zvol_size_changed(zv, volsize);
774 
775 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
776 		if ((error = zvol_dumpify(zv)) != 0 ||
777 		    (error = dumpvp_resize()) != 0) {
778 			int dumpify_error;
779 
780 			(void) zvol_update_volsize(zv->zv_objset, old_volsize);
781 			zvol_size_changed(zv, old_volsize);
782 			dumpify_error = zvol_dumpify(zv);
783 			error = dumpify_error ? dumpify_error : error;
784 		}
785 	}
786 
787 	/*
788 	 * Generate a LUN expansion event.
789 	 */
790 	if (error == 0) {
791 		sysevent_id_t eid;
792 		nvlist_t *attr;
793 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
794 
795 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
796 		    zv->zv_minor);
797 
798 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
799 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
800 
801 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
802 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
803 
804 		nvlist_free(attr);
805 		kmem_free(physpath, MAXPATHLEN);
806 	}
807 	return (error);
808 }
809 
810 int
811 zvol_set_volsize(const char *name, uint64_t volsize)
812 {
813 	zvol_state_t *zv = NULL;
814 	objset_t *os;
815 	int error;
816 	dmu_object_info_t doi;
817 	uint64_t readonly;
818 	boolean_t owned = B_FALSE;
819 
820 	error = dsl_prop_get_integer(name,
821 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
822 	if (error != 0)
823 		return (error);
824 	if (readonly)
825 		return (SET_ERROR(EROFS));
826 
827 	mutex_enter(&zfsdev_state_lock);
828 	zv = zvol_minor_lookup(name);
829 
830 	if (zv == NULL || zv->zv_objset == NULL) {
831 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
832 		    FTAG, &os)) != 0) {
833 			mutex_exit(&zfsdev_state_lock);
834 			return (error);
835 		}
836 		owned = B_TRUE;
837 		if (zv != NULL)
838 			zv->zv_objset = os;
839 	} else {
840 		os = zv->zv_objset;
841 	}
842 
843 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
844 	    (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
845 		goto out;
846 
847 	error = zvol_update_volsize(os, volsize);
848 
849 	if (error == 0 && zv != NULL)
850 		error = zvol_update_live_volsize(zv, volsize);
851 out:
852 	if (owned) {
853 		dmu_objset_disown(os, FTAG);
854 		if (zv != NULL)
855 			zv->zv_objset = NULL;
856 	}
857 	mutex_exit(&zfsdev_state_lock);
858 	return (error);
859 }
860 
861 /*ARGSUSED*/
862 int
863 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
864 {
865 	zvol_state_t *zv;
866 	int err = 0;
867 
868 	mutex_enter(&zfsdev_state_lock);
869 
870 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
871 	if (zv == NULL) {
872 		mutex_exit(&zfsdev_state_lock);
873 		return (SET_ERROR(ENXIO));
874 	}
875 
876 	if (zv->zv_total_opens == 0)
877 		err = zvol_first_open(zv);
878 	if (err) {
879 		mutex_exit(&zfsdev_state_lock);
880 		return (err);
881 	}
882 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
883 		err = SET_ERROR(EROFS);
884 		goto out;
885 	}
886 	if (zv->zv_flags & ZVOL_EXCL) {
887 		err = SET_ERROR(EBUSY);
888 		goto out;
889 	}
890 	if (flag & FEXCL) {
891 		if (zv->zv_total_opens != 0) {
892 			err = SET_ERROR(EBUSY);
893 			goto out;
894 		}
895 		zv->zv_flags |= ZVOL_EXCL;
896 	}
897 
898 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
899 		zv->zv_open_count[otyp]++;
900 		zv->zv_total_opens++;
901 	}
902 	mutex_exit(&zfsdev_state_lock);
903 
904 	return (err);
905 out:
906 	if (zv->zv_total_opens == 0)
907 		zvol_last_close(zv);
908 	mutex_exit(&zfsdev_state_lock);
909 	return (err);
910 }
911 
912 /*ARGSUSED*/
913 int
914 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
915 {
916 	minor_t minor = getminor(dev);
917 	zvol_state_t *zv;
918 	int error = 0;
919 
920 	mutex_enter(&zfsdev_state_lock);
921 
922 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
923 	if (zv == NULL) {
924 		mutex_exit(&zfsdev_state_lock);
925 		return (SET_ERROR(ENXIO));
926 	}
927 
928 	if (zv->zv_flags & ZVOL_EXCL) {
929 		ASSERT(zv->zv_total_opens == 1);
930 		zv->zv_flags &= ~ZVOL_EXCL;
931 	}
932 
933 	/*
934 	 * If the open count is zero, this is a spurious close.
935 	 * That indicates a bug in the kernel / DDI framework.
936 	 */
937 	ASSERT(zv->zv_open_count[otyp] != 0);
938 	ASSERT(zv->zv_total_opens != 0);
939 
940 	/*
941 	 * You may get multiple opens, but only one close.
942 	 */
943 	zv->zv_open_count[otyp]--;
944 	zv->zv_total_opens--;
945 
946 	if (zv->zv_total_opens == 0)
947 		zvol_last_close(zv);
948 
949 	mutex_exit(&zfsdev_state_lock);
950 	return (error);
951 }
952 
953 static void
954 zvol_get_done(zgd_t *zgd, int error)
955 {
956 	if (zgd->zgd_db)
957 		dmu_buf_rele(zgd->zgd_db, zgd);
958 
959 	zfs_range_unlock(zgd->zgd_rl);
960 
961 	if (error == 0 && zgd->zgd_bp)
962 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
963 
964 	kmem_free(zgd, sizeof (zgd_t));
965 }
966 
967 /*
968  * Get data to generate a TX_WRITE intent log record.
969  */
970 static int
971 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
972 {
973 	zvol_state_t *zv = arg;
974 	objset_t *os = zv->zv_objset;
975 	uint64_t object = ZVOL_OBJ;
976 	uint64_t offset = lr->lr_offset;
977 	uint64_t size = lr->lr_length;	/* length of user data */
978 	blkptr_t *bp = &lr->lr_blkptr;
979 	dmu_buf_t *db;
980 	zgd_t *zgd;
981 	int error;
982 
983 	ASSERT(zio != NULL);
984 	ASSERT(size != 0);
985 
986 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
987 	zgd->zgd_zilog = zv->zv_zilog;
988 	zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
989 
990 	/*
991 	 * Write records come in two flavors: immediate and indirect.
992 	 * For small writes it's cheaper to store the data with the
993 	 * log record (immediate); for large writes it's cheaper to
994 	 * sync the data and get a pointer to it (indirect) so that
995 	 * we don't have to write the data twice.
996 	 */
997 	if (buf != NULL) {	/* immediate write */
998 		error = dmu_read(os, object, offset, size, buf,
999 		    DMU_READ_NO_PREFETCH);
1000 	} else {
1001 		size = zv->zv_volblocksize;
1002 		offset = P2ALIGN(offset, size);
1003 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1004 		    DMU_READ_NO_PREFETCH);
1005 		if (error == 0) {
1006 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1007 			if (obp) {
1008 				ASSERT(BP_IS_HOLE(bp));
1009 				*bp = *obp;
1010 			}
1011 
1012 			zgd->zgd_db = db;
1013 			zgd->zgd_bp = bp;
1014 
1015 			ASSERT(db->db_offset == offset);
1016 			ASSERT(db->db_size == size);
1017 
1018 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1019 			    zvol_get_done, zgd);
1020 
1021 			if (error == 0)
1022 				return (0);
1023 		}
1024 	}
1025 
1026 	zvol_get_done(zgd, error);
1027 
1028 	return (error);
1029 }
1030 
1031 /*
1032  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1033  *
1034  * We store data in the log buffers if it's small enough.
1035  * Otherwise we will later flush the data out via dmu_sync().
1036  */
1037 ssize_t zvol_immediate_write_sz = 32768;
1038 
1039 static void
1040 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1041     boolean_t sync)
1042 {
1043 	uint32_t blocksize = zv->zv_volblocksize;
1044 	zilog_t *zilog = zv->zv_zilog;
1045 	boolean_t slogging;
1046 	ssize_t immediate_write_sz;
1047 
1048 	if (zil_replaying(zilog, tx))
1049 		return;
1050 
1051 	immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1052 	    ? 0 : zvol_immediate_write_sz;
1053 
1054 	slogging = spa_has_slogs(zilog->zl_spa) &&
1055 	    (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1056 
1057 	while (resid) {
1058 		itx_t *itx;
1059 		lr_write_t *lr;
1060 		ssize_t len;
1061 		itx_wr_state_t write_state;
1062 
1063 		/*
1064 		 * Unlike zfs_log_write() we can be called with
1065 		 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1066 		 */
1067 		if (blocksize > immediate_write_sz && !slogging &&
1068 		    resid >= blocksize && off % blocksize == 0) {
1069 			write_state = WR_INDIRECT; /* uses dmu_sync */
1070 			len = blocksize;
1071 		} else if (sync) {
1072 			write_state = WR_COPIED;
1073 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1074 		} else {
1075 			write_state = WR_NEED_COPY;
1076 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1077 		}
1078 
1079 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1080 		    (write_state == WR_COPIED ? len : 0));
1081 		lr = (lr_write_t *)&itx->itx_lr;
1082 		if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1083 		    ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1084 			zil_itx_destroy(itx);
1085 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1086 			lr = (lr_write_t *)&itx->itx_lr;
1087 			write_state = WR_NEED_COPY;
1088 		}
1089 
1090 		itx->itx_wr_state = write_state;
1091 		if (write_state == WR_NEED_COPY)
1092 			itx->itx_sod += len;
1093 		lr->lr_foid = ZVOL_OBJ;
1094 		lr->lr_offset = off;
1095 		lr->lr_length = len;
1096 		lr->lr_blkoff = 0;
1097 		BP_ZERO(&lr->lr_blkptr);
1098 
1099 		itx->itx_private = zv;
1100 		itx->itx_sync = sync;
1101 
1102 		zil_itx_assign(zilog, itx, tx);
1103 
1104 		off += len;
1105 		resid -= len;
1106 	}
1107 }
1108 
1109 static int
1110 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1111     uint64_t size, boolean_t doread, boolean_t isdump)
1112 {
1113 	vdev_disk_t *dvd;
1114 	int c;
1115 	int numerrors = 0;
1116 
1117 	if (vd->vdev_ops == &vdev_mirror_ops ||
1118 	    vd->vdev_ops == &vdev_replacing_ops ||
1119 	    vd->vdev_ops == &vdev_spare_ops) {
1120 		for (c = 0; c < vd->vdev_children; c++) {
1121 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1122 			    addr, offset, origoffset, size, doread, isdump);
1123 			if (err != 0) {
1124 				numerrors++;
1125 			} else if (doread) {
1126 				break;
1127 			}
1128 		}
1129 	}
1130 
1131 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1132 		return (numerrors < vd->vdev_children ? 0 : EIO);
1133 
1134 	if (doread && !vdev_readable(vd))
1135 		return (SET_ERROR(EIO));
1136 	else if (!doread && !vdev_writeable(vd))
1137 		return (SET_ERROR(EIO));
1138 
1139 	if (vd->vdev_ops == &vdev_raidz_ops) {
1140 		return (vdev_raidz_physio(vd,
1141 		    addr, size, offset, origoffset, doread, isdump));
1142 	}
1143 
1144 	offset += VDEV_LABEL_START_SIZE;
1145 
1146 	if (ddi_in_panic() || isdump) {
1147 		ASSERT(!doread);
1148 		if (doread)
1149 			return (SET_ERROR(EIO));
1150 		dvd = vd->vdev_tsd;
1151 		ASSERT3P(dvd, !=, NULL);
1152 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1153 		    lbtodb(size)));
1154 	} else {
1155 		dvd = vd->vdev_tsd;
1156 		ASSERT3P(dvd, !=, NULL);
1157 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1158 		    offset, doread ? B_READ : B_WRITE));
1159 	}
1160 }
1161 
1162 static int
1163 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1164     boolean_t doread, boolean_t isdump)
1165 {
1166 	vdev_t *vd;
1167 	int error;
1168 	zvol_extent_t *ze;
1169 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1170 
1171 	/* Must be sector aligned, and not stradle a block boundary. */
1172 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1173 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1174 		return (SET_ERROR(EINVAL));
1175 	}
1176 	ASSERT(size <= zv->zv_volblocksize);
1177 
1178 	/* Locate the extent this belongs to */
1179 	ze = list_head(&zv->zv_extents);
1180 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1181 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1182 		ze = list_next(&zv->zv_extents, ze);
1183 	}
1184 
1185 	if (ze == NULL)
1186 		return (SET_ERROR(EINVAL));
1187 
1188 	if (!ddi_in_panic())
1189 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1190 
1191 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1192 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1193 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1194 	    size, doread, isdump);
1195 
1196 	if (!ddi_in_panic())
1197 		spa_config_exit(spa, SCL_STATE, FTAG);
1198 
1199 	return (error);
1200 }
1201 
1202 int
1203 zvol_strategy(buf_t *bp)
1204 {
1205 	zfs_soft_state_t *zs = NULL;
1206 	zvol_state_t *zv;
1207 	uint64_t off, volsize;
1208 	size_t resid;
1209 	char *addr;
1210 	objset_t *os;
1211 	rl_t *rl;
1212 	int error = 0;
1213 	boolean_t doread = bp->b_flags & B_READ;
1214 	boolean_t is_dumpified;
1215 	boolean_t sync;
1216 
1217 	if (getminor(bp->b_edev) == 0) {
1218 		error = SET_ERROR(EINVAL);
1219 	} else {
1220 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1221 		if (zs == NULL)
1222 			error = SET_ERROR(ENXIO);
1223 		else if (zs->zss_type != ZSST_ZVOL)
1224 			error = SET_ERROR(EINVAL);
1225 	}
1226 
1227 	if (error) {
1228 		bioerror(bp, error);
1229 		biodone(bp);
1230 		return (0);
1231 	}
1232 
1233 	zv = zs->zss_data;
1234 
1235 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1236 		bioerror(bp, EROFS);
1237 		biodone(bp);
1238 		return (0);
1239 	}
1240 
1241 	off = ldbtob(bp->b_blkno);
1242 	volsize = zv->zv_volsize;
1243 
1244 	os = zv->zv_objset;
1245 	ASSERT(os != NULL);
1246 
1247 	bp_mapin(bp);
1248 	addr = bp->b_un.b_addr;
1249 	resid = bp->b_bcount;
1250 
1251 	if (resid > 0 && (off < 0 || off >= volsize)) {
1252 		bioerror(bp, EIO);
1253 		biodone(bp);
1254 		return (0);
1255 	}
1256 
1257 	is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1258 	sync = ((!(bp->b_flags & B_ASYNC) &&
1259 	    !(zv->zv_flags & ZVOL_WCE)) ||
1260 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1261 	    !doread && !is_dumpified;
1262 
1263 	/*
1264 	 * There must be no buffer changes when doing a dmu_sync() because
1265 	 * we can't change the data whilst calculating the checksum.
1266 	 */
1267 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1268 	    doread ? RL_READER : RL_WRITER);
1269 
1270 	while (resid != 0 && off < volsize) {
1271 		size_t size = MIN(resid, zvol_maxphys);
1272 		if (is_dumpified) {
1273 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1274 			error = zvol_dumpio(zv, addr, off, size,
1275 			    doread, B_FALSE);
1276 		} else if (doread) {
1277 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1278 			    DMU_READ_PREFETCH);
1279 		} else {
1280 			dmu_tx_t *tx = dmu_tx_create(os);
1281 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1282 			error = dmu_tx_assign(tx, TXG_WAIT);
1283 			if (error) {
1284 				dmu_tx_abort(tx);
1285 			} else {
1286 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1287 				zvol_log_write(zv, tx, off, size, sync);
1288 				dmu_tx_commit(tx);
1289 			}
1290 		}
1291 		if (error) {
1292 			/* convert checksum errors into IO errors */
1293 			if (error == ECKSUM)
1294 				error = SET_ERROR(EIO);
1295 			break;
1296 		}
1297 		off += size;
1298 		addr += size;
1299 		resid -= size;
1300 	}
1301 	zfs_range_unlock(rl);
1302 
1303 	if ((bp->b_resid = resid) == bp->b_bcount)
1304 		bioerror(bp, off > volsize ? EINVAL : error);
1305 
1306 	if (sync)
1307 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1308 	biodone(bp);
1309 
1310 	return (0);
1311 }
1312 
1313 /*
1314  * Set the buffer count to the zvol maximum transfer.
1315  * Using our own routine instead of the default minphys()
1316  * means that for larger writes we write bigger buffers on X86
1317  * (128K instead of 56K) and flush the disk write cache less often
1318  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1319  * 56K on X86 and 128K on sparc).
1320  */
1321 void
1322 zvol_minphys(struct buf *bp)
1323 {
1324 	if (bp->b_bcount > zvol_maxphys)
1325 		bp->b_bcount = zvol_maxphys;
1326 }
1327 
1328 int
1329 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1330 {
1331 	minor_t minor = getminor(dev);
1332 	zvol_state_t *zv;
1333 	int error = 0;
1334 	uint64_t size;
1335 	uint64_t boff;
1336 	uint64_t resid;
1337 
1338 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1339 	if (zv == NULL)
1340 		return (SET_ERROR(ENXIO));
1341 
1342 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1343 		return (SET_ERROR(EINVAL));
1344 
1345 	boff = ldbtob(blkno);
1346 	resid = ldbtob(nblocks);
1347 
1348 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1349 
1350 	while (resid) {
1351 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1352 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1353 		if (error)
1354 			break;
1355 		boff += size;
1356 		addr += size;
1357 		resid -= size;
1358 	}
1359 
1360 	return (error);
1361 }
1362 
1363 /*ARGSUSED*/
1364 int
1365 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1366 {
1367 	minor_t minor = getminor(dev);
1368 	zvol_state_t *zv;
1369 	uint64_t volsize;
1370 	rl_t *rl;
1371 	int error = 0;
1372 
1373 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1374 	if (zv == NULL)
1375 		return (SET_ERROR(ENXIO));
1376 
1377 	volsize = zv->zv_volsize;
1378 	if (uio->uio_resid > 0 &&
1379 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1380 		return (SET_ERROR(EIO));
1381 
1382 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1383 		error = physio(zvol_strategy, NULL, dev, B_READ,
1384 		    zvol_minphys, uio);
1385 		return (error);
1386 	}
1387 
1388 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1389 	    RL_READER);
1390 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1391 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1392 
1393 		/* don't read past the end */
1394 		if (bytes > volsize - uio->uio_loffset)
1395 			bytes = volsize - uio->uio_loffset;
1396 
1397 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1398 		if (error) {
1399 			/* convert checksum errors into IO errors */
1400 			if (error == ECKSUM)
1401 				error = SET_ERROR(EIO);
1402 			break;
1403 		}
1404 	}
1405 	zfs_range_unlock(rl);
1406 	return (error);
1407 }
1408 
1409 /*ARGSUSED*/
1410 int
1411 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1412 {
1413 	minor_t minor = getminor(dev);
1414 	zvol_state_t *zv;
1415 	uint64_t volsize;
1416 	rl_t *rl;
1417 	int error = 0;
1418 	boolean_t sync;
1419 
1420 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1421 	if (zv == NULL)
1422 		return (SET_ERROR(ENXIO));
1423 
1424 	volsize = zv->zv_volsize;
1425 	if (uio->uio_resid > 0 &&
1426 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1427 		return (SET_ERROR(EIO));
1428 
1429 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1430 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1431 		    zvol_minphys, uio);
1432 		return (error);
1433 	}
1434 
1435 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1436 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1437 
1438 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1439 	    RL_WRITER);
1440 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1441 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1442 		uint64_t off = uio->uio_loffset;
1443 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1444 
1445 		if (bytes > volsize - off)	/* don't write past the end */
1446 			bytes = volsize - off;
1447 
1448 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1449 		error = dmu_tx_assign(tx, TXG_WAIT);
1450 		if (error) {
1451 			dmu_tx_abort(tx);
1452 			break;
1453 		}
1454 		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1455 		if (error == 0)
1456 			zvol_log_write(zv, tx, off, bytes, sync);
1457 		dmu_tx_commit(tx);
1458 
1459 		if (error)
1460 			break;
1461 	}
1462 	zfs_range_unlock(rl);
1463 	if (sync)
1464 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1465 	return (error);
1466 }
1467 
1468 int
1469 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1470 {
1471 	struct uuid uuid = EFI_RESERVED;
1472 	efi_gpe_t gpe = { 0 };
1473 	uint32_t crc;
1474 	dk_efi_t efi;
1475 	int length;
1476 	char *ptr;
1477 
1478 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1479 		return (SET_ERROR(EFAULT));
1480 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1481 	length = efi.dki_length;
1482 	/*
1483 	 * Some clients may attempt to request a PMBR for the
1484 	 * zvol.  Currently this interface will return EINVAL to
1485 	 * such requests.  These requests could be supported by
1486 	 * adding a check for lba == 0 and consing up an appropriate
1487 	 * PMBR.
1488 	 */
1489 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1490 		return (SET_ERROR(EINVAL));
1491 
1492 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1493 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1494 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1495 
1496 	if (efi.dki_lba == 1) {
1497 		efi_gpt_t gpt = { 0 };
1498 
1499 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1500 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1501 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1502 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1503 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1504 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1505 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1506 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1507 		gpt.efi_gpt_SizeOfPartitionEntry =
1508 		    LE_32(sizeof (efi_gpe_t));
1509 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1510 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1511 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1512 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1513 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1514 		    flag))
1515 			return (SET_ERROR(EFAULT));
1516 		ptr += sizeof (gpt);
1517 		length -= sizeof (gpt);
1518 	}
1519 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1520 	    length), flag))
1521 		return (SET_ERROR(EFAULT));
1522 	return (0);
1523 }
1524 
1525 /*
1526  * BEGIN entry points to allow external callers access to the volume.
1527  */
1528 /*
1529  * Return the volume parameters needed for access from an external caller.
1530  * These values are invariant as long as the volume is held open.
1531  */
1532 int
1533 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1534     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1535     void **rl_hdl, void **bonus_hdl)
1536 {
1537 	zvol_state_t *zv;
1538 
1539 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1540 	if (zv == NULL)
1541 		return (SET_ERROR(ENXIO));
1542 	if (zv->zv_flags & ZVOL_DUMPIFIED)
1543 		return (SET_ERROR(ENXIO));
1544 
1545 	ASSERT(blksize && max_xfer_len && minor_hdl &&
1546 	    objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1547 
1548 	*blksize = zv->zv_volblocksize;
1549 	*max_xfer_len = (uint64_t)zvol_maxphys;
1550 	*minor_hdl = zv;
1551 	*objset_hdl = zv->zv_objset;
1552 	*zil_hdl = zv->zv_zilog;
1553 	*rl_hdl = &zv->zv_znode;
1554 	*bonus_hdl = zv->zv_dbuf;
1555 	return (0);
1556 }
1557 
1558 /*
1559  * Return the current volume size to an external caller.
1560  * The size can change while the volume is open.
1561  */
1562 uint64_t
1563 zvol_get_volume_size(void *minor_hdl)
1564 {
1565 	zvol_state_t *zv = minor_hdl;
1566 
1567 	return (zv->zv_volsize);
1568 }
1569 
1570 /*
1571  * Return the current WCE setting to an external caller.
1572  * The WCE setting can change while the volume is open.
1573  */
1574 int
1575 zvol_get_volume_wce(void *minor_hdl)
1576 {
1577 	zvol_state_t *zv = minor_hdl;
1578 
1579 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1580 }
1581 
1582 /*
1583  * Entry point for external callers to zvol_log_write
1584  */
1585 void
1586 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1587     boolean_t sync)
1588 {
1589 	zvol_state_t *zv = minor_hdl;
1590 
1591 	zvol_log_write(zv, tx, off, resid, sync);
1592 }
1593 /*
1594  * END entry points to allow external callers access to the volume.
1595  */
1596 
1597 /*
1598  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1599  */
1600 static void
1601 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1602     boolean_t sync)
1603 {
1604 	itx_t *itx;
1605 	lr_truncate_t *lr;
1606 	zilog_t *zilog = zv->zv_zilog;
1607 
1608 	if (zil_replaying(zilog, tx))
1609 		return;
1610 
1611 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1612 	lr = (lr_truncate_t *)&itx->itx_lr;
1613 	lr->lr_foid = ZVOL_OBJ;
1614 	lr->lr_offset = off;
1615 	lr->lr_length = len;
1616 
1617 	itx->itx_sync = sync;
1618 	zil_itx_assign(zilog, itx, tx);
1619 }
1620 
1621 /*
1622  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1623  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1624  */
1625 /*ARGSUSED*/
1626 int
1627 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1628 {
1629 	zvol_state_t *zv;
1630 	struct dk_cinfo dki;
1631 	struct dk_minfo dkm;
1632 	struct dk_callback *dkc;
1633 	int error = 0;
1634 	rl_t *rl;
1635 
1636 	mutex_enter(&zfsdev_state_lock);
1637 
1638 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1639 
1640 	if (zv == NULL) {
1641 		mutex_exit(&zfsdev_state_lock);
1642 		return (SET_ERROR(ENXIO));
1643 	}
1644 	ASSERT(zv->zv_total_opens > 0);
1645 
1646 	switch (cmd) {
1647 
1648 	case DKIOCINFO:
1649 		bzero(&dki, sizeof (dki));
1650 		(void) strcpy(dki.dki_cname, "zvol");
1651 		(void) strcpy(dki.dki_dname, "zvol");
1652 		dki.dki_ctype = DKC_UNKNOWN;
1653 		dki.dki_unit = getminor(dev);
1654 		dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1655 		mutex_exit(&zfsdev_state_lock);
1656 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1657 			error = SET_ERROR(EFAULT);
1658 		return (error);
1659 
1660 	case DKIOCGMEDIAINFO:
1661 		bzero(&dkm, sizeof (dkm));
1662 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1663 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1664 		dkm.dki_media_type = DK_UNKNOWN;
1665 		mutex_exit(&zfsdev_state_lock);
1666 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1667 			error = SET_ERROR(EFAULT);
1668 		return (error);
1669 
1670 	case DKIOCGETEFI:
1671 		{
1672 			uint64_t vs = zv->zv_volsize;
1673 			uint8_t bs = zv->zv_min_bs;
1674 
1675 			mutex_exit(&zfsdev_state_lock);
1676 			error = zvol_getefi((void *)arg, flag, vs, bs);
1677 			return (error);
1678 		}
1679 
1680 	case DKIOCFLUSHWRITECACHE:
1681 		dkc = (struct dk_callback *)arg;
1682 		mutex_exit(&zfsdev_state_lock);
1683 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1684 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1685 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1686 			error = 0;
1687 		}
1688 		return (error);
1689 
1690 	case DKIOCGETWCE:
1691 		{
1692 			int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1693 			if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1694 			    flag))
1695 				error = SET_ERROR(EFAULT);
1696 			break;
1697 		}
1698 	case DKIOCSETWCE:
1699 		{
1700 			int wce;
1701 			if (ddi_copyin((void *)arg, &wce, sizeof (int),
1702 			    flag)) {
1703 				error = SET_ERROR(EFAULT);
1704 				break;
1705 			}
1706 			if (wce) {
1707 				zv->zv_flags |= ZVOL_WCE;
1708 				mutex_exit(&zfsdev_state_lock);
1709 			} else {
1710 				zv->zv_flags &= ~ZVOL_WCE;
1711 				mutex_exit(&zfsdev_state_lock);
1712 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1713 			}
1714 			return (0);
1715 		}
1716 
1717 	case DKIOCGGEOM:
1718 	case DKIOCGVTOC:
1719 		/*
1720 		 * commands using these (like prtvtoc) expect ENOTSUP
1721 		 * since we're emulating an EFI label
1722 		 */
1723 		error = SET_ERROR(ENOTSUP);
1724 		break;
1725 
1726 	case DKIOCDUMPINIT:
1727 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1728 		    RL_WRITER);
1729 		error = zvol_dumpify(zv);
1730 		zfs_range_unlock(rl);
1731 		break;
1732 
1733 	case DKIOCDUMPFINI:
1734 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1735 			break;
1736 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1737 		    RL_WRITER);
1738 		error = zvol_dump_fini(zv);
1739 		zfs_range_unlock(rl);
1740 		break;
1741 
1742 	case DKIOCFREE:
1743 	{
1744 		dkioc_free_t df;
1745 		dmu_tx_t *tx;
1746 
1747 		if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1748 			error = SET_ERROR(EFAULT);
1749 			break;
1750 		}
1751 
1752 		/*
1753 		 * Apply Postel's Law to length-checking.  If they overshoot,
1754 		 * just blank out until the end, if there's a need to blank
1755 		 * out anything.
1756 		 */
1757 		if (df.df_start >= zv->zv_volsize)
1758 			break;	/* No need to do anything... */
1759 		if (df.df_start + df.df_length > zv->zv_volsize)
1760 			df.df_length = DMU_OBJECT_END;
1761 
1762 		rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1763 		    RL_WRITER);
1764 		tx = dmu_tx_create(zv->zv_objset);
1765 		error = dmu_tx_assign(tx, TXG_WAIT);
1766 		if (error != 0) {
1767 			dmu_tx_abort(tx);
1768 		} else {
1769 			zvol_log_truncate(zv, tx, df.df_start,
1770 			    df.df_length, B_TRUE);
1771 			dmu_tx_commit(tx);
1772 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1773 			    df.df_start, df.df_length);
1774 		}
1775 
1776 		zfs_range_unlock(rl);
1777 
1778 		if (error == 0) {
1779 			/*
1780 			 * If the write-cache is disabled or 'sync' property
1781 			 * is set to 'always' then treat this as a synchronous
1782 			 * operation (i.e. commit to zil).
1783 			 */
1784 			if (!(zv->zv_flags & ZVOL_WCE) ||
1785 			    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
1786 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1787 
1788 			/*
1789 			 * If the caller really wants synchronous writes, and
1790 			 * can't wait for them, don't return until the write
1791 			 * is done.
1792 			 */
1793 			if (df.df_flags & DF_WAIT_SYNC) {
1794 				txg_wait_synced(
1795 				    dmu_objset_pool(zv->zv_objset), 0);
1796 			}
1797 		}
1798 		break;
1799 	}
1800 
1801 	default:
1802 		error = SET_ERROR(ENOTTY);
1803 		break;
1804 
1805 	}
1806 	mutex_exit(&zfsdev_state_lock);
1807 	return (error);
1808 }
1809 
1810 int
1811 zvol_busy(void)
1812 {
1813 	return (zvol_minors != 0);
1814 }
1815 
1816 void
1817 zvol_init(void)
1818 {
1819 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1820 	    1) == 0);
1821 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1822 }
1823 
1824 void
1825 zvol_fini(void)
1826 {
1827 	mutex_destroy(&zfsdev_state_lock);
1828 	ddi_soft_state_fini(&zfsdev_state);
1829 }
1830 
1831 /*ARGSUSED*/
1832 static int
1833 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1834 {
1835 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1836 
1837 	if (spa_feature_is_active(spa,
1838 	    &spa_feature_table[SPA_FEATURE_MULTI_VDEV_CRASH_DUMP]))
1839 		return (1);
1840 	return (0);
1841 }
1842 
1843 /*ARGSUSED*/
1844 static void
1845 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1846 {
1847 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1848 
1849 	spa_feature_incr(spa,
1850 	    &spa_feature_table[SPA_FEATURE_MULTI_VDEV_CRASH_DUMP], tx);
1851 }
1852 
1853 static int
1854 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1855 {
1856 	dmu_tx_t *tx;
1857 	int error;
1858 	objset_t *os = zv->zv_objset;
1859 	spa_t *spa = dmu_objset_spa(os);
1860 	vdev_t *vd = spa->spa_root_vdev;
1861 	nvlist_t *nv = NULL;
1862 	uint64_t version = spa_version(spa);
1863 	enum zio_checksum checksum;
1864 
1865 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1866 	ASSERT(vd->vdev_ops == &vdev_root_ops);
1867 
1868 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1869 	    DMU_OBJECT_END);
1870 	/* wait for dmu_free_long_range to actually free the blocks */
1871 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1872 
1873 	/*
1874 	 * If the pool on which the dump device is being initialized has more
1875 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1876 	 * enabled.  If so, bump that feature's counter to indicate that the
1877 	 * feature is active. We also check the vdev type to handle the
1878 	 * following case:
1879 	 *   # zpool create test raidz disk1 disk2 disk3
1880 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1881 	 *   the raidz vdev itself has 3 children.
1882 	 */
1883 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1884 		if (!spa_feature_is_enabled(spa,
1885 		    &spa_feature_table[SPA_FEATURE_MULTI_VDEV_CRASH_DUMP]))
1886 			return (SET_ERROR(ENOTSUP));
1887 		(void) dsl_sync_task(spa_name(spa),
1888 		    zfs_mvdev_dump_feature_check,
1889 		    zfs_mvdev_dump_activate_feature_sync, NULL, 2);
1890 	}
1891 
1892 	tx = dmu_tx_create(os);
1893 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1894 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1895 	error = dmu_tx_assign(tx, TXG_WAIT);
1896 	if (error) {
1897 		dmu_tx_abort(tx);
1898 		return (error);
1899 	}
1900 
1901 	/*
1902 	 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
1903 	 * function.  Otherwise, use the old default -- OFF.
1904 	 */
1905 	checksum = spa_feature_is_active(spa,
1906 	    &spa_feature_table[SPA_FEATURE_MULTI_VDEV_CRASH_DUMP]) ?
1907 	    ZIO_CHECKSUM_NOPARITY : ZIO_CHECKSUM_OFF;
1908 
1909 	/*
1910 	 * If we are resizing the dump device then we only need to
1911 	 * update the refreservation to match the newly updated
1912 	 * zvolsize. Otherwise, we save off the original state of the
1913 	 * zvol so that we can restore them if the zvol is ever undumpified.
1914 	 */
1915 	if (resize) {
1916 		error = zap_update(os, ZVOL_ZAP_OBJ,
1917 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1918 		    &zv->zv_volsize, tx);
1919 	} else {
1920 		uint64_t checksum, compress, refresrv, vbs, dedup;
1921 
1922 		error = dsl_prop_get_integer(zv->zv_name,
1923 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1924 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1925 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1926 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1927 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1928 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1929 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1930 		if (version >= SPA_VERSION_DEDUP) {
1931 			error = error ? error :
1932 			    dsl_prop_get_integer(zv->zv_name,
1933 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1934 		}
1935 
1936 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1937 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1938 		    &compress, tx);
1939 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1940 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1941 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1942 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1943 		    &refresrv, tx);
1944 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1945 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1946 		    &vbs, tx);
1947 		error = error ? error : dmu_object_set_blocksize(
1948 		    os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1949 		if (version >= SPA_VERSION_DEDUP) {
1950 			error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1951 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1952 			    &dedup, tx);
1953 		}
1954 		if (error == 0)
1955 			zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1956 	}
1957 	dmu_tx_commit(tx);
1958 
1959 	/*
1960 	 * We only need update the zvol's property if we are initializing
1961 	 * the dump area for the first time.
1962 	 */
1963 	if (!resize) {
1964 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1965 		VERIFY(nvlist_add_uint64(nv,
1966 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1967 		VERIFY(nvlist_add_uint64(nv,
1968 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1969 		    ZIO_COMPRESS_OFF) == 0);
1970 		VERIFY(nvlist_add_uint64(nv,
1971 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1972 		    checksum) == 0);
1973 		if (version >= SPA_VERSION_DEDUP) {
1974 			VERIFY(nvlist_add_uint64(nv,
1975 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
1976 			    ZIO_CHECKSUM_OFF) == 0);
1977 		}
1978 
1979 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1980 		    nv, NULL);
1981 		nvlist_free(nv);
1982 
1983 		if (error)
1984 			return (error);
1985 	}
1986 
1987 	/* Allocate the space for the dump */
1988 	error = zvol_prealloc(zv);
1989 	return (error);
1990 }
1991 
1992 static int
1993 zvol_dumpify(zvol_state_t *zv)
1994 {
1995 	int error = 0;
1996 	uint64_t dumpsize = 0;
1997 	dmu_tx_t *tx;
1998 	objset_t *os = zv->zv_objset;
1999 
2000 	if (zv->zv_flags & ZVOL_RDONLY)
2001 		return (SET_ERROR(EROFS));
2002 
2003 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2004 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2005 		boolean_t resize = (dumpsize > 0);
2006 
2007 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2008 			(void) zvol_dump_fini(zv);
2009 			return (error);
2010 		}
2011 	}
2012 
2013 	/*
2014 	 * Build up our lba mapping.
2015 	 */
2016 	error = zvol_get_lbas(zv);
2017 	if (error) {
2018 		(void) zvol_dump_fini(zv);
2019 		return (error);
2020 	}
2021 
2022 	tx = dmu_tx_create(os);
2023 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2024 	error = dmu_tx_assign(tx, TXG_WAIT);
2025 	if (error) {
2026 		dmu_tx_abort(tx);
2027 		(void) zvol_dump_fini(zv);
2028 		return (error);
2029 	}
2030 
2031 	zv->zv_flags |= ZVOL_DUMPIFIED;
2032 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2033 	    &zv->zv_volsize, tx);
2034 	dmu_tx_commit(tx);
2035 
2036 	if (error) {
2037 		(void) zvol_dump_fini(zv);
2038 		return (error);
2039 	}
2040 
2041 	txg_wait_synced(dmu_objset_pool(os), 0);
2042 	return (0);
2043 }
2044 
2045 static int
2046 zvol_dump_fini(zvol_state_t *zv)
2047 {
2048 	dmu_tx_t *tx;
2049 	objset_t *os = zv->zv_objset;
2050 	nvlist_t *nv;
2051 	int error = 0;
2052 	uint64_t checksum, compress, refresrv, vbs, dedup;
2053 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2054 
2055 	/*
2056 	 * Attempt to restore the zvol back to its pre-dumpified state.
2057 	 * This is a best-effort attempt as it's possible that not all
2058 	 * of these properties were initialized during the dumpify process
2059 	 * (i.e. error during zvol_dump_init).
2060 	 */
2061 
2062 	tx = dmu_tx_create(os);
2063 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2064 	error = dmu_tx_assign(tx, TXG_WAIT);
2065 	if (error) {
2066 		dmu_tx_abort(tx);
2067 		return (error);
2068 	}
2069 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2070 	dmu_tx_commit(tx);
2071 
2072 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2073 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2074 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2075 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2076 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2077 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2078 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2079 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2080 
2081 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2082 	(void) nvlist_add_uint64(nv,
2083 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2084 	(void) nvlist_add_uint64(nv,
2085 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2086 	(void) nvlist_add_uint64(nv,
2087 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2088 	if (version >= SPA_VERSION_DEDUP &&
2089 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2090 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2091 		(void) nvlist_add_uint64(nv,
2092 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2093 	}
2094 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2095 	    nv, NULL);
2096 	nvlist_free(nv);
2097 
2098 	zvol_free_extents(zv);
2099 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2100 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2101 	/* wait for dmu_free_long_range to actually free the blocks */
2102 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2103 	tx = dmu_tx_create(os);
2104 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2105 	error = dmu_tx_assign(tx, TXG_WAIT);
2106 	if (error) {
2107 		dmu_tx_abort(tx);
2108 		return (error);
2109 	}
2110 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2111 		zv->zv_volblocksize = vbs;
2112 	dmu_tx_commit(tx);
2113 
2114 	return (0);
2115 }
2116