xref: /titanic_41/usr/src/uts/common/fs/zfs/zvol.c (revision dd7bc1797d9dc4e3b17c9a517508e879e16ef1b2)
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/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_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 	error = dmu_tx_assign(tx, TXG_WAIT);
722 	if (error) {
723 		dmu_tx_abort(tx);
724 		return (error);
725 	}
726 
727 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
728 	    &volsize, tx);
729 	dmu_tx_commit(tx);
730 
731 	if (error == 0)
732 		error = dmu_free_long_range(os,
733 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
734 	return (error);
735 }
736 
737 void
738 zvol_remove_minors(const char *name)
739 {
740 	zvol_state_t *zv;
741 	char *namebuf;
742 	minor_t minor;
743 
744 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
745 	(void) strncpy(namebuf, name, strlen(name));
746 	(void) strcat(namebuf, "/");
747 	mutex_enter(&zfsdev_state_lock);
748 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
749 
750 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
751 		if (zv == NULL)
752 			continue;
753 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
754 			(void) zvol_remove_zv(zv);
755 	}
756 	kmem_free(namebuf, strlen(name) + 2);
757 
758 	mutex_exit(&zfsdev_state_lock);
759 }
760 
761 static int
762 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
763 {
764 	uint64_t old_volsize = 0ULL;
765 	int error = 0;
766 
767 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
768 
769 	/*
770 	 * Reinitialize the dump area to the new size. If we
771 	 * failed to resize the dump area then restore it back to
772 	 * its original size.  We must set the new volsize prior
773 	 * to calling dumpvp_resize() to ensure that the devices'
774 	 * size(9P) is not visible by the dump subsystem.
775 	 */
776 	old_volsize = zv->zv_volsize;
777 	zvol_size_changed(zv, volsize);
778 
779 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
780 		if ((error = zvol_dumpify(zv)) != 0 ||
781 		    (error = dumpvp_resize()) != 0) {
782 			int dumpify_error;
783 
784 			(void) zvol_update_volsize(zv->zv_objset, old_volsize);
785 			zvol_size_changed(zv, old_volsize);
786 			dumpify_error = zvol_dumpify(zv);
787 			error = dumpify_error ? dumpify_error : error;
788 		}
789 	}
790 
791 	/*
792 	 * Generate a LUN expansion event.
793 	 */
794 	if (error == 0) {
795 		sysevent_id_t eid;
796 		nvlist_t *attr;
797 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
798 
799 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
800 		    zv->zv_minor);
801 
802 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
803 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
804 
805 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
806 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
807 
808 		nvlist_free(attr);
809 		kmem_free(physpath, MAXPATHLEN);
810 	}
811 	return (error);
812 }
813 
814 int
815 zvol_set_volsize(const char *name, uint64_t volsize)
816 {
817 	zvol_state_t *zv = NULL;
818 	objset_t *os;
819 	int error;
820 	dmu_object_info_t doi;
821 	uint64_t readonly;
822 	boolean_t owned = B_FALSE;
823 
824 	error = dsl_prop_get_integer(name,
825 	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
826 	if (error != 0)
827 		return (error);
828 	if (readonly)
829 		return (SET_ERROR(EROFS));
830 
831 	mutex_enter(&zfsdev_state_lock);
832 	zv = zvol_minor_lookup(name);
833 
834 	if (zv == NULL || zv->zv_objset == NULL) {
835 		if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
836 		    FTAG, &os)) != 0) {
837 			mutex_exit(&zfsdev_state_lock);
838 			return (error);
839 		}
840 		owned = B_TRUE;
841 		if (zv != NULL)
842 			zv->zv_objset = os;
843 	} else {
844 		os = zv->zv_objset;
845 	}
846 
847 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
848 	    (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
849 		goto out;
850 
851 	error = zvol_update_volsize(os, volsize);
852 
853 	if (error == 0 && zv != NULL)
854 		error = zvol_update_live_volsize(zv, volsize);
855 out:
856 	if (owned) {
857 		dmu_objset_disown(os, FTAG);
858 		if (zv != NULL)
859 			zv->zv_objset = NULL;
860 	}
861 	mutex_exit(&zfsdev_state_lock);
862 	return (error);
863 }
864 
865 /*ARGSUSED*/
866 int
867 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
868 {
869 	zvol_state_t *zv;
870 	int err = 0;
871 
872 	mutex_enter(&zfsdev_state_lock);
873 
874 	zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
875 	if (zv == NULL) {
876 		mutex_exit(&zfsdev_state_lock);
877 		return (SET_ERROR(ENXIO));
878 	}
879 
880 	if (zv->zv_total_opens == 0)
881 		err = zvol_first_open(zv);
882 	if (err) {
883 		mutex_exit(&zfsdev_state_lock);
884 		return (err);
885 	}
886 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
887 		err = SET_ERROR(EROFS);
888 		goto out;
889 	}
890 	if (zv->zv_flags & ZVOL_EXCL) {
891 		err = SET_ERROR(EBUSY);
892 		goto out;
893 	}
894 	if (flag & FEXCL) {
895 		if (zv->zv_total_opens != 0) {
896 			err = SET_ERROR(EBUSY);
897 			goto out;
898 		}
899 		zv->zv_flags |= ZVOL_EXCL;
900 	}
901 
902 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
903 		zv->zv_open_count[otyp]++;
904 		zv->zv_total_opens++;
905 	}
906 	mutex_exit(&zfsdev_state_lock);
907 
908 	return (err);
909 out:
910 	if (zv->zv_total_opens == 0)
911 		zvol_last_close(zv);
912 	mutex_exit(&zfsdev_state_lock);
913 	return (err);
914 }
915 
916 /*ARGSUSED*/
917 int
918 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
919 {
920 	minor_t minor = getminor(dev);
921 	zvol_state_t *zv;
922 	int error = 0;
923 
924 	mutex_enter(&zfsdev_state_lock);
925 
926 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
927 	if (zv == NULL) {
928 		mutex_exit(&zfsdev_state_lock);
929 		return (SET_ERROR(ENXIO));
930 	}
931 
932 	if (zv->zv_flags & ZVOL_EXCL) {
933 		ASSERT(zv->zv_total_opens == 1);
934 		zv->zv_flags &= ~ZVOL_EXCL;
935 	}
936 
937 	/*
938 	 * If the open count is zero, this is a spurious close.
939 	 * That indicates a bug in the kernel / DDI framework.
940 	 */
941 	ASSERT(zv->zv_open_count[otyp] != 0);
942 	ASSERT(zv->zv_total_opens != 0);
943 
944 	/*
945 	 * You may get multiple opens, but only one close.
946 	 */
947 	zv->zv_open_count[otyp]--;
948 	zv->zv_total_opens--;
949 
950 	if (zv->zv_total_opens == 0)
951 		zvol_last_close(zv);
952 
953 	mutex_exit(&zfsdev_state_lock);
954 	return (error);
955 }
956 
957 static void
958 zvol_get_done(zgd_t *zgd, int error)
959 {
960 	if (zgd->zgd_db)
961 		dmu_buf_rele(zgd->zgd_db, zgd);
962 
963 	zfs_range_unlock(zgd->zgd_rl);
964 
965 	if (error == 0 && zgd->zgd_bp)
966 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
967 
968 	kmem_free(zgd, sizeof (zgd_t));
969 }
970 
971 /*
972  * Get data to generate a TX_WRITE intent log record.
973  */
974 static int
975 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
976 {
977 	zvol_state_t *zv = arg;
978 	objset_t *os = zv->zv_objset;
979 	uint64_t object = ZVOL_OBJ;
980 	uint64_t offset = lr->lr_offset;
981 	uint64_t size = lr->lr_length;	/* length of user data */
982 	blkptr_t *bp = &lr->lr_blkptr;
983 	dmu_buf_t *db;
984 	zgd_t *zgd;
985 	int error;
986 
987 	ASSERT(zio != NULL);
988 	ASSERT(size != 0);
989 
990 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
991 	zgd->zgd_zilog = zv->zv_zilog;
992 	zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
993 
994 	/*
995 	 * Write records come in two flavors: immediate and indirect.
996 	 * For small writes it's cheaper to store the data with the
997 	 * log record (immediate); for large writes it's cheaper to
998 	 * sync the data and get a pointer to it (indirect) so that
999 	 * we don't have to write the data twice.
1000 	 */
1001 	if (buf != NULL) {	/* immediate write */
1002 		error = dmu_read(os, object, offset, size, buf,
1003 		    DMU_READ_NO_PREFETCH);
1004 	} else {
1005 		size = zv->zv_volblocksize;
1006 		offset = P2ALIGN(offset, size);
1007 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1008 		    DMU_READ_NO_PREFETCH);
1009 		if (error == 0) {
1010 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1011 			if (obp) {
1012 				ASSERT(BP_IS_HOLE(bp));
1013 				*bp = *obp;
1014 			}
1015 
1016 			zgd->zgd_db = db;
1017 			zgd->zgd_bp = bp;
1018 
1019 			ASSERT(db->db_offset == offset);
1020 			ASSERT(db->db_size == size);
1021 
1022 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1023 			    zvol_get_done, zgd);
1024 
1025 			if (error == 0)
1026 				return (0);
1027 		}
1028 	}
1029 
1030 	zvol_get_done(zgd, error);
1031 
1032 	return (error);
1033 }
1034 
1035 /*
1036  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1037  *
1038  * We store data in the log buffers if it's small enough.
1039  * Otherwise we will later flush the data out via dmu_sync().
1040  */
1041 ssize_t zvol_immediate_write_sz = 32768;
1042 
1043 static void
1044 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1045     boolean_t sync)
1046 {
1047 	uint32_t blocksize = zv->zv_volblocksize;
1048 	zilog_t *zilog = zv->zv_zilog;
1049 	boolean_t slogging;
1050 	ssize_t immediate_write_sz;
1051 
1052 	if (zil_replaying(zilog, tx))
1053 		return;
1054 
1055 	immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1056 	    ? 0 : zvol_immediate_write_sz;
1057 
1058 	slogging = spa_has_slogs(zilog->zl_spa) &&
1059 	    (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1060 
1061 	while (resid) {
1062 		itx_t *itx;
1063 		lr_write_t *lr;
1064 		ssize_t len;
1065 		itx_wr_state_t write_state;
1066 
1067 		/*
1068 		 * Unlike zfs_log_write() we can be called with
1069 		 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1070 		 */
1071 		if (blocksize > immediate_write_sz && !slogging &&
1072 		    resid >= blocksize && off % blocksize == 0) {
1073 			write_state = WR_INDIRECT; /* uses dmu_sync */
1074 			len = blocksize;
1075 		} else if (sync) {
1076 			write_state = WR_COPIED;
1077 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1078 		} else {
1079 			write_state = WR_NEED_COPY;
1080 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1081 		}
1082 
1083 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1084 		    (write_state == WR_COPIED ? len : 0));
1085 		lr = (lr_write_t *)&itx->itx_lr;
1086 		if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1087 		    ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1088 			zil_itx_destroy(itx);
1089 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1090 			lr = (lr_write_t *)&itx->itx_lr;
1091 			write_state = WR_NEED_COPY;
1092 		}
1093 
1094 		itx->itx_wr_state = write_state;
1095 		if (write_state == WR_NEED_COPY)
1096 			itx->itx_sod += len;
1097 		lr->lr_foid = ZVOL_OBJ;
1098 		lr->lr_offset = off;
1099 		lr->lr_length = len;
1100 		lr->lr_blkoff = 0;
1101 		BP_ZERO(&lr->lr_blkptr);
1102 
1103 		itx->itx_private = zv;
1104 		itx->itx_sync = sync;
1105 
1106 		zil_itx_assign(zilog, itx, tx);
1107 
1108 		rw_enter(&rz_zev_rwlock, RW_READER);
1109 		if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_zvol_write)
1110 			rz_zev_callbacks->rz_zev_zvol_write(zv->zv_name,
1111 			    zv->zv_objset, tx, off, len);
1112 		rw_exit(&rz_zev_rwlock);
1113 
1114 		off += len;
1115 		resid -= len;
1116 	}
1117 }
1118 
1119 static int
1120 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1121     uint64_t size, boolean_t doread, boolean_t isdump)
1122 {
1123 	vdev_disk_t *dvd;
1124 	int c;
1125 	int numerrors = 0;
1126 
1127 	if (vd->vdev_ops == &vdev_mirror_ops ||
1128 	    vd->vdev_ops == &vdev_replacing_ops ||
1129 	    vd->vdev_ops == &vdev_spare_ops) {
1130 		for (c = 0; c < vd->vdev_children; c++) {
1131 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1132 			    addr, offset, origoffset, size, doread, isdump);
1133 			if (err != 0) {
1134 				numerrors++;
1135 			} else if (doread) {
1136 				break;
1137 			}
1138 		}
1139 	}
1140 
1141 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1142 		return (numerrors < vd->vdev_children ? 0 : EIO);
1143 
1144 	if (doread && !vdev_readable(vd))
1145 		return (SET_ERROR(EIO));
1146 	else if (!doread && !vdev_writeable(vd))
1147 		return (SET_ERROR(EIO));
1148 
1149 	if (vd->vdev_ops == &vdev_raidz_ops) {
1150 		return (vdev_raidz_physio(vd,
1151 		    addr, size, offset, origoffset, doread, isdump));
1152 	}
1153 
1154 	offset += VDEV_LABEL_START_SIZE;
1155 
1156 	if (ddi_in_panic() || isdump) {
1157 		ASSERT(!doread);
1158 		if (doread)
1159 			return (SET_ERROR(EIO));
1160 		dvd = vd->vdev_tsd;
1161 		ASSERT3P(dvd, !=, NULL);
1162 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1163 		    lbtodb(size)));
1164 	} else {
1165 		dvd = vd->vdev_tsd;
1166 		ASSERT3P(dvd, !=, NULL);
1167 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1168 		    offset, doread ? B_READ : B_WRITE));
1169 	}
1170 }
1171 
1172 static int
1173 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1174     boolean_t doread, boolean_t isdump)
1175 {
1176 	vdev_t *vd;
1177 	int error;
1178 	zvol_extent_t *ze;
1179 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1180 
1181 	/* Must be sector aligned, and not stradle a block boundary. */
1182 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1183 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1184 		return (SET_ERROR(EINVAL));
1185 	}
1186 	ASSERT(size <= zv->zv_volblocksize);
1187 
1188 	/* Locate the extent this belongs to */
1189 	ze = list_head(&zv->zv_extents);
1190 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1191 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1192 		ze = list_next(&zv->zv_extents, ze);
1193 	}
1194 
1195 	if (ze == NULL)
1196 		return (SET_ERROR(EINVAL));
1197 
1198 	if (!ddi_in_panic())
1199 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1200 
1201 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1202 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1203 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1204 	    size, doread, isdump);
1205 
1206 	if (!ddi_in_panic())
1207 		spa_config_exit(spa, SCL_STATE, FTAG);
1208 
1209 	return (error);
1210 }
1211 
1212 int
1213 zvol_strategy(buf_t *bp)
1214 {
1215 	zfs_soft_state_t *zs = NULL;
1216 	zvol_state_t *zv;
1217 	uint64_t off, volsize;
1218 	size_t resid;
1219 	char *addr;
1220 	objset_t *os;
1221 	rl_t *rl;
1222 	int error = 0;
1223 	boolean_t doread = bp->b_flags & B_READ;
1224 	boolean_t is_dumpified;
1225 	boolean_t sync;
1226 
1227 	if (getminor(bp->b_edev) == 0) {
1228 		error = SET_ERROR(EINVAL);
1229 	} else {
1230 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1231 		if (zs == NULL)
1232 			error = SET_ERROR(ENXIO);
1233 		else if (zs->zss_type != ZSST_ZVOL)
1234 			error = SET_ERROR(EINVAL);
1235 	}
1236 
1237 	if (error) {
1238 		bioerror(bp, error);
1239 		biodone(bp);
1240 		return (0);
1241 	}
1242 
1243 	zv = zs->zss_data;
1244 
1245 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1246 		bioerror(bp, EROFS);
1247 		biodone(bp);
1248 		return (0);
1249 	}
1250 
1251 	off = ldbtob(bp->b_blkno);
1252 	volsize = zv->zv_volsize;
1253 
1254 	os = zv->zv_objset;
1255 	ASSERT(os != NULL);
1256 
1257 	bp_mapin(bp);
1258 	addr = bp->b_un.b_addr;
1259 	resid = bp->b_bcount;
1260 
1261 	if (resid > 0 && (off < 0 || off >= volsize)) {
1262 		bioerror(bp, EIO);
1263 		biodone(bp);
1264 		return (0);
1265 	}
1266 
1267 	is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1268 	sync = ((!(bp->b_flags & B_ASYNC) &&
1269 	    !(zv->zv_flags & ZVOL_WCE)) ||
1270 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1271 	    !doread && !is_dumpified;
1272 
1273 	/*
1274 	 * There must be no buffer changes when doing a dmu_sync() because
1275 	 * we can't change the data whilst calculating the checksum.
1276 	 */
1277 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1278 	    doread ? RL_READER : RL_WRITER);
1279 
1280 	while (resid != 0 && off < volsize) {
1281 		size_t size = MIN(resid, zvol_maxphys);
1282 		if (is_dumpified) {
1283 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1284 			error = zvol_dumpio(zv, addr, off, size,
1285 			    doread, B_FALSE);
1286 		} else if (doread) {
1287 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1288 			    DMU_READ_PREFETCH);
1289 		} else {
1290 			dmu_tx_t *tx = dmu_tx_create(os);
1291 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1292 			error = dmu_tx_assign(tx, TXG_WAIT);
1293 			if (error) {
1294 				dmu_tx_abort(tx);
1295 			} else {
1296 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1297 				zvol_log_write(zv, tx, off, size, sync);
1298 				dmu_tx_commit(tx);
1299 			}
1300 		}
1301 		if (error) {
1302 			/* convert checksum errors into IO errors */
1303 			if (error == ECKSUM)
1304 				error = SET_ERROR(EIO);
1305 			break;
1306 		}
1307 		off += size;
1308 		addr += size;
1309 		resid -= size;
1310 	}
1311 	zfs_range_unlock(rl);
1312 
1313 	if ((bp->b_resid = resid) == bp->b_bcount)
1314 		bioerror(bp, off > volsize ? EINVAL : error);
1315 
1316 	if (sync)
1317 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1318 	biodone(bp);
1319 
1320 	return (0);
1321 }
1322 
1323 /*
1324  * Set the buffer count to the zvol maximum transfer.
1325  * Using our own routine instead of the default minphys()
1326  * means that for larger writes we write bigger buffers on X86
1327  * (128K instead of 56K) and flush the disk write cache less often
1328  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1329  * 56K on X86 and 128K on sparc).
1330  */
1331 void
1332 zvol_minphys(struct buf *bp)
1333 {
1334 	if (bp->b_bcount > zvol_maxphys)
1335 		bp->b_bcount = zvol_maxphys;
1336 }
1337 
1338 int
1339 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1340 {
1341 	minor_t minor = getminor(dev);
1342 	zvol_state_t *zv;
1343 	int error = 0;
1344 	uint64_t size;
1345 	uint64_t boff;
1346 	uint64_t resid;
1347 
1348 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1349 	if (zv == NULL)
1350 		return (SET_ERROR(ENXIO));
1351 
1352 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1353 		return (SET_ERROR(EINVAL));
1354 
1355 	boff = ldbtob(blkno);
1356 	resid = ldbtob(nblocks);
1357 
1358 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1359 
1360 	while (resid) {
1361 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1362 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1363 		if (error)
1364 			break;
1365 		boff += size;
1366 		addr += size;
1367 		resid -= size;
1368 	}
1369 
1370 	return (error);
1371 }
1372 
1373 /*ARGSUSED*/
1374 int
1375 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1376 {
1377 	minor_t minor = getminor(dev);
1378 	zvol_state_t *zv;
1379 	uint64_t volsize;
1380 	rl_t *rl;
1381 	int error = 0;
1382 
1383 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1384 	if (zv == NULL)
1385 		return (SET_ERROR(ENXIO));
1386 
1387 	volsize = zv->zv_volsize;
1388 	if (uio->uio_resid > 0 &&
1389 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1390 		return (SET_ERROR(EIO));
1391 
1392 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1393 		error = physio(zvol_strategy, NULL, dev, B_READ,
1394 		    zvol_minphys, uio);
1395 		return (error);
1396 	}
1397 
1398 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1399 	    RL_READER);
1400 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1401 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1402 
1403 		/* don't read past the end */
1404 		if (bytes > volsize - uio->uio_loffset)
1405 			bytes = volsize - uio->uio_loffset;
1406 
1407 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1408 		if (error) {
1409 			/* convert checksum errors into IO errors */
1410 			if (error == ECKSUM)
1411 				error = SET_ERROR(EIO);
1412 			break;
1413 		}
1414 	}
1415 	zfs_range_unlock(rl);
1416 	return (error);
1417 }
1418 
1419 /*ARGSUSED*/
1420 int
1421 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1422 {
1423 	minor_t minor = getminor(dev);
1424 	zvol_state_t *zv;
1425 	uint64_t volsize;
1426 	rl_t *rl;
1427 	int error = 0;
1428 	boolean_t sync;
1429 
1430 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1431 	if (zv == NULL)
1432 		return (SET_ERROR(ENXIO));
1433 
1434 	volsize = zv->zv_volsize;
1435 	if (uio->uio_resid > 0 &&
1436 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1437 		return (SET_ERROR(EIO));
1438 
1439 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1440 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1441 		    zvol_minphys, uio);
1442 		return (error);
1443 	}
1444 
1445 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1446 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1447 
1448 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1449 	    RL_WRITER);
1450 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1451 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1452 		uint64_t off = uio->uio_loffset;
1453 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1454 
1455 		if (bytes > volsize - off)	/* don't write past the end */
1456 			bytes = volsize - off;
1457 
1458 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1459 		error = dmu_tx_assign(tx, TXG_WAIT);
1460 		if (error) {
1461 			dmu_tx_abort(tx);
1462 			break;
1463 		}
1464 		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1465 		if (error == 0)
1466 			zvol_log_write(zv, tx, off, bytes, sync);
1467 		dmu_tx_commit(tx);
1468 
1469 		if (error)
1470 			break;
1471 	}
1472 	zfs_range_unlock(rl);
1473 	if (sync)
1474 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1475 	return (error);
1476 }
1477 
1478 int
1479 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1480 {
1481 	struct uuid uuid = EFI_RESERVED;
1482 	efi_gpe_t gpe = { 0 };
1483 	uint32_t crc;
1484 	dk_efi_t efi;
1485 	int length;
1486 	char *ptr;
1487 
1488 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1489 		return (SET_ERROR(EFAULT));
1490 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1491 	length = efi.dki_length;
1492 	/*
1493 	 * Some clients may attempt to request a PMBR for the
1494 	 * zvol.  Currently this interface will return EINVAL to
1495 	 * such requests.  These requests could be supported by
1496 	 * adding a check for lba == 0 and consing up an appropriate
1497 	 * PMBR.
1498 	 */
1499 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1500 		return (SET_ERROR(EINVAL));
1501 
1502 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1503 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1504 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1505 
1506 	if (efi.dki_lba == 1) {
1507 		efi_gpt_t gpt = { 0 };
1508 
1509 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1510 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1511 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1512 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1513 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1514 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1515 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1516 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1517 		gpt.efi_gpt_SizeOfPartitionEntry =
1518 		    LE_32(sizeof (efi_gpe_t));
1519 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1520 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1521 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1522 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1523 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1524 		    flag))
1525 			return (SET_ERROR(EFAULT));
1526 		ptr += sizeof (gpt);
1527 		length -= sizeof (gpt);
1528 	}
1529 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1530 	    length), flag))
1531 		return (SET_ERROR(EFAULT));
1532 	return (0);
1533 }
1534 
1535 /*
1536  * BEGIN entry points to allow external callers access to the volume.
1537  */
1538 /*
1539  * Return the volume parameters needed for access from an external caller.
1540  * These values are invariant as long as the volume is held open.
1541  */
1542 int
1543 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1544     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1545     void **rl_hdl, void **bonus_hdl)
1546 {
1547 	zvol_state_t *zv;
1548 
1549 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1550 	if (zv == NULL)
1551 		return (SET_ERROR(ENXIO));
1552 	if (zv->zv_flags & ZVOL_DUMPIFIED)
1553 		return (SET_ERROR(ENXIO));
1554 
1555 	ASSERT(blksize && max_xfer_len && minor_hdl &&
1556 	    objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1557 
1558 	*blksize = zv->zv_volblocksize;
1559 	*max_xfer_len = (uint64_t)zvol_maxphys;
1560 	*minor_hdl = zv;
1561 	*objset_hdl = zv->zv_objset;
1562 	*zil_hdl = zv->zv_zilog;
1563 	*rl_hdl = &zv->zv_znode;
1564 	*bonus_hdl = zv->zv_dbuf;
1565 	return (0);
1566 }
1567 
1568 /*
1569  * Return the current volume size to an external caller.
1570  * The size can change while the volume is open.
1571  */
1572 uint64_t
1573 zvol_get_volume_size(void *minor_hdl)
1574 {
1575 	zvol_state_t *zv = minor_hdl;
1576 
1577 	return (zv->zv_volsize);
1578 }
1579 
1580 /*
1581  * Return the current WCE setting to an external caller.
1582  * The WCE setting can change while the volume is open.
1583  */
1584 int
1585 zvol_get_volume_wce(void *minor_hdl)
1586 {
1587 	zvol_state_t *zv = minor_hdl;
1588 
1589 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1590 }
1591 
1592 /*
1593  * Entry point for external callers to zvol_log_write
1594  */
1595 void
1596 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1597     boolean_t sync)
1598 {
1599 	zvol_state_t *zv = minor_hdl;
1600 
1601 	zvol_log_write(zv, tx, off, resid, sync);
1602 }
1603 /*
1604  * END entry points to allow external callers access to the volume.
1605  */
1606 
1607 /*
1608  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1609  */
1610 static void
1611 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1612     boolean_t sync)
1613 {
1614 	itx_t *itx;
1615 	lr_truncate_t *lr;
1616 	zilog_t *zilog = zv->zv_zilog;
1617 
1618 	if (zil_replaying(zilog, tx))
1619 		return;
1620 
1621 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1622 	lr = (lr_truncate_t *)&itx->itx_lr;
1623 	lr->lr_foid = ZVOL_OBJ;
1624 	lr->lr_offset = off;
1625 	lr->lr_length = len;
1626 
1627 	itx->itx_sync = sync;
1628 	zil_itx_assign(zilog, itx, tx);
1629 
1630 	rw_enter(&rz_zev_rwlock, RW_READER);
1631 	if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_zvol_truncate)
1632 		rz_zev_callbacks->rz_zev_zvol_truncate(zv->zv_name,
1633 		    zv->zv_objset, tx, off, len);
1634 	rw_exit(&rz_zev_rwlock);
1635 }
1636 
1637 /*
1638  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1639  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1640  */
1641 /*ARGSUSED*/
1642 int
1643 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1644 {
1645 	zvol_state_t *zv;
1646 	struct dk_callback *dkc;
1647 	int error = 0;
1648 	rl_t *rl;
1649 
1650 	mutex_enter(&zfsdev_state_lock);
1651 
1652 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1653 
1654 	if (zv == NULL) {
1655 		mutex_exit(&zfsdev_state_lock);
1656 		return (SET_ERROR(ENXIO));
1657 	}
1658 	ASSERT(zv->zv_total_opens > 0);
1659 
1660 	switch (cmd) {
1661 
1662 	case DKIOCINFO:
1663 	{
1664 		struct dk_cinfo dki;
1665 
1666 		bzero(&dki, sizeof (dki));
1667 		(void) strcpy(dki.dki_cname, "zvol");
1668 		(void) strcpy(dki.dki_dname, "zvol");
1669 		dki.dki_ctype = DKC_UNKNOWN;
1670 		dki.dki_unit = getminor(dev);
1671 		dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1672 		mutex_exit(&zfsdev_state_lock);
1673 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1674 			error = SET_ERROR(EFAULT);
1675 		return (error);
1676 	}
1677 
1678 	case DKIOCGMEDIAINFO:
1679 	{
1680 		struct dk_minfo dkm;
1681 
1682 		bzero(&dkm, sizeof (dkm));
1683 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1684 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1685 		dkm.dki_media_type = DK_UNKNOWN;
1686 		mutex_exit(&zfsdev_state_lock);
1687 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1688 			error = SET_ERROR(EFAULT);
1689 		return (error);
1690 	}
1691 
1692 	case DKIOCGMEDIAINFOEXT:
1693 	{
1694 		struct dk_minfo_ext dkmext;
1695 
1696 		bzero(&dkmext, sizeof (dkmext));
1697 		dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1698 		dkmext.dki_pbsize = zv->zv_volblocksize;
1699 		dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1700 		dkmext.dki_media_type = DK_UNKNOWN;
1701 		mutex_exit(&zfsdev_state_lock);
1702 		if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1703 			error = SET_ERROR(EFAULT);
1704 		return (error);
1705 	}
1706 
1707 	case DKIOCGETEFI:
1708 	{
1709 		uint64_t vs = zv->zv_volsize;
1710 		uint8_t bs = zv->zv_min_bs;
1711 
1712 		mutex_exit(&zfsdev_state_lock);
1713 		error = zvol_getefi((void *)arg, flag, vs, bs);
1714 		return (error);
1715 	}
1716 
1717 	case DKIOCFLUSHWRITECACHE:
1718 		dkc = (struct dk_callback *)arg;
1719 		mutex_exit(&zfsdev_state_lock);
1720 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1721 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1722 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1723 			error = 0;
1724 		}
1725 		return (error);
1726 
1727 	case DKIOCGETWCE:
1728 	{
1729 		int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1730 		if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1731 		    flag))
1732 			error = SET_ERROR(EFAULT);
1733 		break;
1734 	}
1735 	case DKIOCSETWCE:
1736 	{
1737 		int wce;
1738 		if (ddi_copyin((void *)arg, &wce, sizeof (int),
1739 		    flag)) {
1740 			error = SET_ERROR(EFAULT);
1741 			break;
1742 		}
1743 		if (wce) {
1744 			zv->zv_flags |= ZVOL_WCE;
1745 			mutex_exit(&zfsdev_state_lock);
1746 		} else {
1747 			zv->zv_flags &= ~ZVOL_WCE;
1748 			mutex_exit(&zfsdev_state_lock);
1749 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1750 		}
1751 		return (0);
1752 	}
1753 
1754 	case DKIOCGGEOM:
1755 	case DKIOCGVTOC:
1756 		/*
1757 		 * commands using these (like prtvtoc) expect ENOTSUP
1758 		 * since we're emulating an EFI label
1759 		 */
1760 		error = SET_ERROR(ENOTSUP);
1761 		break;
1762 
1763 	case DKIOCDUMPINIT:
1764 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1765 		    RL_WRITER);
1766 		error = zvol_dumpify(zv);
1767 		zfs_range_unlock(rl);
1768 		break;
1769 
1770 	case DKIOCDUMPFINI:
1771 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1772 			break;
1773 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1774 		    RL_WRITER);
1775 		error = zvol_dump_fini(zv);
1776 		zfs_range_unlock(rl);
1777 		break;
1778 
1779 	case DKIOCFREE:
1780 	{
1781 		dkioc_free_t df;
1782 		dmu_tx_t *tx;
1783 
1784 		if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1785 			error = SET_ERROR(EFAULT);
1786 			break;
1787 		}
1788 
1789 		/*
1790 		 * Apply Postel's Law to length-checking.  If they overshoot,
1791 		 * just blank out until the end, if there's a need to blank
1792 		 * out anything.
1793 		 */
1794 		if (df.df_start >= zv->zv_volsize)
1795 			break;	/* No need to do anything... */
1796 		if (df.df_start + df.df_length > zv->zv_volsize)
1797 			df.df_length = DMU_OBJECT_END;
1798 
1799 		rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1800 		    RL_WRITER);
1801 		tx = dmu_tx_create(zv->zv_objset);
1802 		error = dmu_tx_assign(tx, TXG_WAIT);
1803 		if (error != 0) {
1804 			dmu_tx_abort(tx);
1805 		} else {
1806 			zvol_log_truncate(zv, tx, df.df_start,
1807 			    df.df_length, B_TRUE);
1808 			dmu_tx_commit(tx);
1809 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1810 			    df.df_start, df.df_length);
1811 		}
1812 
1813 		zfs_range_unlock(rl);
1814 
1815 		if (error == 0) {
1816 			/*
1817 			 * If the write-cache is disabled or 'sync' property
1818 			 * is set to 'always' then treat this as a synchronous
1819 			 * operation (i.e. commit to zil).
1820 			 */
1821 			if (!(zv->zv_flags & ZVOL_WCE) ||
1822 			    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
1823 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1824 
1825 			/*
1826 			 * If the caller really wants synchronous writes, and
1827 			 * can't wait for them, don't return until the write
1828 			 * is done.
1829 			 */
1830 			if (df.df_flags & DF_WAIT_SYNC) {
1831 				txg_wait_synced(
1832 				    dmu_objset_pool(zv->zv_objset), 0);
1833 			}
1834 		}
1835 		break;
1836 	}
1837 
1838 	default:
1839 		error = SET_ERROR(ENOTTY);
1840 		break;
1841 
1842 	}
1843 	mutex_exit(&zfsdev_state_lock);
1844 	return (error);
1845 }
1846 
1847 int
1848 zvol_busy(void)
1849 {
1850 	return (zvol_minors != 0);
1851 }
1852 
1853 void
1854 zvol_init(void)
1855 {
1856 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1857 	    1) == 0);
1858 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1859 }
1860 
1861 void
1862 zvol_fini(void)
1863 {
1864 	mutex_destroy(&zfsdev_state_lock);
1865 	ddi_soft_state_fini(&zfsdev_state);
1866 }
1867 
1868 /*ARGSUSED*/
1869 static int
1870 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1871 {
1872 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1873 
1874 	if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1875 		return (1);
1876 	return (0);
1877 }
1878 
1879 /*ARGSUSED*/
1880 static void
1881 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1882 {
1883 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1884 
1885 	spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1886 }
1887 
1888 static int
1889 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1890 {
1891 	dmu_tx_t *tx;
1892 	int error;
1893 	objset_t *os = zv->zv_objset;
1894 	spa_t *spa = dmu_objset_spa(os);
1895 	vdev_t *vd = spa->spa_root_vdev;
1896 	nvlist_t *nv = NULL;
1897 	uint64_t version = spa_version(spa);
1898 	enum zio_checksum checksum;
1899 
1900 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1901 	ASSERT(vd->vdev_ops == &vdev_root_ops);
1902 
1903 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1904 	    DMU_OBJECT_END);
1905 	/* wait for dmu_free_long_range to actually free the blocks */
1906 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1907 
1908 	/*
1909 	 * If the pool on which the dump device is being initialized has more
1910 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1911 	 * enabled.  If so, bump that feature's counter to indicate that the
1912 	 * feature is active. We also check the vdev type to handle the
1913 	 * following case:
1914 	 *   # zpool create test raidz disk1 disk2 disk3
1915 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1916 	 *   the raidz vdev itself has 3 children.
1917 	 */
1918 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1919 		if (!spa_feature_is_enabled(spa,
1920 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1921 			return (SET_ERROR(ENOTSUP));
1922 		(void) dsl_sync_task(spa_name(spa),
1923 		    zfs_mvdev_dump_feature_check,
1924 		    zfs_mvdev_dump_activate_feature_sync, NULL, 2);
1925 	}
1926 
1927 	tx = dmu_tx_create(os);
1928 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1929 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1930 	error = dmu_tx_assign(tx, TXG_WAIT);
1931 	if (error) {
1932 		dmu_tx_abort(tx);
1933 		return (error);
1934 	}
1935 
1936 	/*
1937 	 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
1938 	 * function.  Otherwise, use the old default -- OFF.
1939 	 */
1940 	checksum = spa_feature_is_active(spa,
1941 	    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
1942 	    ZIO_CHECKSUM_OFF;
1943 
1944 	/*
1945 	 * If we are resizing the dump device then we only need to
1946 	 * update the refreservation to match the newly updated
1947 	 * zvolsize. Otherwise, we save off the original state of the
1948 	 * zvol so that we can restore them if the zvol is ever undumpified.
1949 	 */
1950 	if (resize) {
1951 		error = zap_update(os, ZVOL_ZAP_OBJ,
1952 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1953 		    &zv->zv_volsize, tx);
1954 	} else {
1955 		uint64_t checksum, compress, refresrv, vbs, dedup;
1956 
1957 		error = dsl_prop_get_integer(zv->zv_name,
1958 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1959 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1960 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1961 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1962 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1963 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1964 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1965 		if (version >= SPA_VERSION_DEDUP) {
1966 			error = error ? error :
1967 			    dsl_prop_get_integer(zv->zv_name,
1968 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1969 		}
1970 
1971 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1972 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1973 		    &compress, tx);
1974 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1975 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1976 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1977 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1978 		    &refresrv, tx);
1979 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1980 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1981 		    &vbs, tx);
1982 		error = error ? error : dmu_object_set_blocksize(
1983 		    os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1984 		if (version >= SPA_VERSION_DEDUP) {
1985 			error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1986 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1987 			    &dedup, tx);
1988 		}
1989 		if (error == 0)
1990 			zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1991 	}
1992 	dmu_tx_commit(tx);
1993 
1994 	/*
1995 	 * We only need update the zvol's property if we are initializing
1996 	 * the dump area for the first time.
1997 	 */
1998 	if (!resize) {
1999 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2000 		VERIFY(nvlist_add_uint64(nv,
2001 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2002 		VERIFY(nvlist_add_uint64(nv,
2003 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2004 		    ZIO_COMPRESS_OFF) == 0);
2005 		VERIFY(nvlist_add_uint64(nv,
2006 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2007 		    checksum) == 0);
2008 		if (version >= SPA_VERSION_DEDUP) {
2009 			VERIFY(nvlist_add_uint64(nv,
2010 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
2011 			    ZIO_CHECKSUM_OFF) == 0);
2012 		}
2013 
2014 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2015 		    nv, NULL);
2016 		nvlist_free(nv);
2017 
2018 		if (error)
2019 			return (error);
2020 	}
2021 
2022 	/* Allocate the space for the dump */
2023 	error = zvol_prealloc(zv);
2024 	return (error);
2025 }
2026 
2027 static int
2028 zvol_dumpify(zvol_state_t *zv)
2029 {
2030 	int error = 0;
2031 	uint64_t dumpsize = 0;
2032 	dmu_tx_t *tx;
2033 	objset_t *os = zv->zv_objset;
2034 
2035 	if (zv->zv_flags & ZVOL_RDONLY)
2036 		return (SET_ERROR(EROFS));
2037 
2038 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2039 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2040 		boolean_t resize = (dumpsize > 0);
2041 
2042 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2043 			(void) zvol_dump_fini(zv);
2044 			return (error);
2045 		}
2046 	}
2047 
2048 	/*
2049 	 * Build up our lba mapping.
2050 	 */
2051 	error = zvol_get_lbas(zv);
2052 	if (error) {
2053 		(void) zvol_dump_fini(zv);
2054 		return (error);
2055 	}
2056 
2057 	tx = dmu_tx_create(os);
2058 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2059 	error = dmu_tx_assign(tx, TXG_WAIT);
2060 	if (error) {
2061 		dmu_tx_abort(tx);
2062 		(void) zvol_dump_fini(zv);
2063 		return (error);
2064 	}
2065 
2066 	zv->zv_flags |= ZVOL_DUMPIFIED;
2067 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2068 	    &zv->zv_volsize, tx);
2069 	dmu_tx_commit(tx);
2070 
2071 	if (error) {
2072 		(void) zvol_dump_fini(zv);
2073 		return (error);
2074 	}
2075 
2076 	txg_wait_synced(dmu_objset_pool(os), 0);
2077 	return (0);
2078 }
2079 
2080 static int
2081 zvol_dump_fini(zvol_state_t *zv)
2082 {
2083 	dmu_tx_t *tx;
2084 	objset_t *os = zv->zv_objset;
2085 	nvlist_t *nv;
2086 	int error = 0;
2087 	uint64_t checksum, compress, refresrv, vbs, dedup;
2088 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2089 
2090 	/*
2091 	 * Attempt to restore the zvol back to its pre-dumpified state.
2092 	 * This is a best-effort attempt as it's possible that not all
2093 	 * of these properties were initialized during the dumpify process
2094 	 * (i.e. error during zvol_dump_init).
2095 	 */
2096 
2097 	tx = dmu_tx_create(os);
2098 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2099 	error = dmu_tx_assign(tx, TXG_WAIT);
2100 	if (error) {
2101 		dmu_tx_abort(tx);
2102 		return (error);
2103 	}
2104 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2105 	dmu_tx_commit(tx);
2106 
2107 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2108 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2109 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2110 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2111 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2112 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2113 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2114 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2115 
2116 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2117 	(void) nvlist_add_uint64(nv,
2118 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2119 	(void) nvlist_add_uint64(nv,
2120 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2121 	(void) nvlist_add_uint64(nv,
2122 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2123 	if (version >= SPA_VERSION_DEDUP &&
2124 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2125 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2126 		(void) nvlist_add_uint64(nv,
2127 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2128 	}
2129 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2130 	    nv, NULL);
2131 	nvlist_free(nv);
2132 
2133 	zvol_free_extents(zv);
2134 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2135 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2136 	/* wait for dmu_free_long_range to actually free the blocks */
2137 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2138 	tx = dmu_tx_create(os);
2139 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2140 	error = dmu_tx_assign(tx, TXG_WAIT);
2141 	if (error) {
2142 		dmu_tx_abort(tx);
2143 		return (error);
2144 	}
2145 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2146 		zv->zv_volblocksize = vbs;
2147 	dmu_tx_commit(tx);
2148 
2149 	return (0);
2150 }
2151