xref: /titanic_41/usr/src/uts/common/fs/zfs/zvol.c (revision a31148363f598def767ac48c5d82e1572e44b935)
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 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * ZFS volume emulation driver.
28  *
29  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
30  * Volumes are accessed through the symbolic links named:
31  *
32  * /dev/zvol/dsk/<pool_name>/<dataset_name>
33  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
34  *
35  * These links are created by the /dev filesystem (sdev_zvolops.c).
36  * Volumes are persistent through reboot.  No user command needs to be
37  * run before opening and using a device.
38  */
39 
40 #include <sys/types.h>
41 #include <sys/param.h>
42 #include <sys/errno.h>
43 #include <sys/uio.h>
44 #include <sys/buf.h>
45 #include <sys/modctl.h>
46 #include <sys/open.h>
47 #include <sys/kmem.h>
48 #include <sys/conf.h>
49 #include <sys/cmn_err.h>
50 #include <sys/stat.h>
51 #include <sys/zap.h>
52 #include <sys/spa.h>
53 #include <sys/zio.h>
54 #include <sys/dmu_traverse.h>
55 #include <sys/dnode.h>
56 #include <sys/dsl_dataset.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dkio.h>
59 #include <sys/efi_partition.h>
60 #include <sys/byteorder.h>
61 #include <sys/pathname.h>
62 #include <sys/ddi.h>
63 #include <sys/sunddi.h>
64 #include <sys/crc32.h>
65 #include <sys/dirent.h>
66 #include <sys/policy.h>
67 #include <sys/fs/zfs.h>
68 #include <sys/zfs_ioctl.h>
69 #include <sys/mkdev.h>
70 #include <sys/zil.h>
71 #include <sys/refcount.h>
72 #include <sys/zfs_znode.h>
73 #include <sys/zfs_rlock.h>
74 #include <sys/vdev_disk.h>
75 #include <sys/vdev_impl.h>
76 #include <sys/zvol.h>
77 #include <sys/dumphdr.h>
78 #include <sys/zil_impl.h>
79 
80 #include "zfs_namecheck.h"
81 
82 static void *zvol_state;
83 static char *zvol_tag = "zvol_tag";
84 
85 #define	ZVOL_DUMPSIZE		"dumpsize"
86 
87 /*
88  * This lock protects the zvol_state structure from being modified
89  * while it's being used, e.g. an open that comes in before a create
90  * finishes.  It also protects temporary opens of the dataset so that,
91  * e.g., an open doesn't get a spurious EBUSY.
92  */
93 static kmutex_t zvol_state_lock;
94 static uint32_t zvol_minors;
95 
96 typedef struct zvol_extent {
97 	list_node_t	ze_node;
98 	dva_t		ze_dva;		/* dva associated with this extent */
99 	uint64_t	ze_nblks;	/* number of blocks in extent */
100 } zvol_extent_t;
101 
102 /*
103  * The in-core state of each volume.
104  */
105 typedef struct zvol_state {
106 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
107 	uint64_t	zv_volsize;	/* amount of space we advertise */
108 	uint64_t	zv_volblocksize; /* volume block size */
109 	minor_t		zv_minor;	/* minor number */
110 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
111 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
112 	objset_t	*zv_objset;	/* objset handle */
113 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
114 	uint32_t	zv_total_opens;	/* total open count */
115 	zilog_t		*zv_zilog;	/* ZIL handle */
116 	list_t		zv_extents;	/* List of extents for dump */
117 	znode_t		zv_znode;	/* for range locking */
118 } zvol_state_t;
119 
120 /*
121  * zvol specific flags
122  */
123 #define	ZVOL_RDONLY	0x1
124 #define	ZVOL_DUMPIFIED	0x2
125 #define	ZVOL_EXCL	0x4
126 #define	ZVOL_WCE	0x8
127 
128 /*
129  * zvol maximum transfer in one DMU tx.
130  */
131 int zvol_maxphys = DMU_MAX_ACCESS/2;
132 
133 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
134     nvlist_t *, nvlist_t **);
135 static int zvol_remove_zv(zvol_state_t *);
136 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
137 static int zvol_dumpify(zvol_state_t *zv);
138 static int zvol_dump_fini(zvol_state_t *zv);
139 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
140 
141 static void
142 zvol_size_changed(uint64_t volsize, major_t maj, minor_t min)
143 {
144 	dev_t dev = makedevice(maj, min);
145 
146 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
147 	    "Size", volsize) == DDI_SUCCESS);
148 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
149 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
150 
151 	/* Notify specfs to invalidate the cached size */
152 	spec_size_invalidate(dev, VBLK);
153 	spec_size_invalidate(dev, VCHR);
154 }
155 
156 int
157 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
158 {
159 	if (volsize == 0)
160 		return (EINVAL);
161 
162 	if (volsize % blocksize != 0)
163 		return (EINVAL);
164 
165 #ifdef _ILP32
166 	if (volsize - 1 > SPEC_MAXOFFSET_T)
167 		return (EOVERFLOW);
168 #endif
169 	return (0);
170 }
171 
172 int
173 zvol_check_volblocksize(uint64_t volblocksize)
174 {
175 	if (volblocksize < SPA_MINBLOCKSIZE ||
176 	    volblocksize > SPA_MAXBLOCKSIZE ||
177 	    !ISP2(volblocksize))
178 		return (EDOM);
179 
180 	return (0);
181 }
182 
183 int
184 zvol_get_stats(objset_t *os, nvlist_t *nv)
185 {
186 	int error;
187 	dmu_object_info_t doi;
188 	uint64_t val;
189 
190 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
191 	if (error)
192 		return (error);
193 
194 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
195 
196 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
197 
198 	if (error == 0) {
199 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
200 		    doi.doi_data_block_size);
201 	}
202 
203 	return (error);
204 }
205 
206 /*
207  * Find a free minor number.
208  */
209 static minor_t
210 zvol_minor_alloc(void)
211 {
212 	minor_t minor;
213 
214 	ASSERT(MUTEX_HELD(&zvol_state_lock));
215 
216 	for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++)
217 		if (ddi_get_soft_state(zvol_state, minor) == NULL)
218 			return (minor);
219 
220 	return (0);
221 }
222 
223 static zvol_state_t *
224 zvol_minor_lookup(const char *name)
225 {
226 	minor_t minor;
227 	zvol_state_t *zv;
228 
229 	ASSERT(MUTEX_HELD(&zvol_state_lock));
230 
231 	for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) {
232 		zv = ddi_get_soft_state(zvol_state, minor);
233 		if (zv == NULL)
234 			continue;
235 		if (strcmp(zv->zv_name, name) == 0)
236 			break;
237 	}
238 
239 	return (zv);
240 }
241 
242 /* extent mapping arg */
243 struct maparg {
244 	zvol_state_t	*ma_zv;
245 	uint64_t	ma_blks;
246 };
247 
248 /*ARGSUSED*/
249 static int
250 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
251     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
252 {
253 	struct maparg *ma = arg;
254 	zvol_extent_t *ze;
255 	int bs = ma->ma_zv->zv_volblocksize;
256 
257 	if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
258 		return (0);
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 (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_WRITE ZIL transaction that didn't get committed
361  * after a system failure
362  */
363 static int
364 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
365 {
366 	objset_t *os = zv->zv_objset;
367 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
368 	uint64_t offset, length;
369 	dmu_tx_t *tx;
370 	int error;
371 
372 	if (byteswap)
373 		byteswap_uint64_array(lr, sizeof (*lr));
374 
375 	offset = lr->lr_offset;
376 	length = lr->lr_length;
377 
378 	/* If it's a dmu_sync() block, write the whole block */
379 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
380 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
381 		if (length < blocksize) {
382 			offset -= offset % blocksize;
383 			length = blocksize;
384 		}
385 	}
386 
387 	tx = dmu_tx_create(os);
388 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
389 	error = dmu_tx_assign(tx, TXG_WAIT);
390 	if (error) {
391 		dmu_tx_abort(tx);
392 	} else {
393 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
394 		dmu_tx_commit(tx);
395 	}
396 
397 	return (error);
398 }
399 
400 /* ARGSUSED */
401 static int
402 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
403 {
404 	return (ENOTSUP);
405 }
406 
407 /*
408  * Callback vectors for replaying records.
409  * Only TX_WRITE is needed for zvol.
410  */
411 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
412 	zvol_replay_err,	/* 0 no such transaction type */
413 	zvol_replay_err,	/* TX_CREATE */
414 	zvol_replay_err,	/* TX_MKDIR */
415 	zvol_replay_err,	/* TX_MKXATTR */
416 	zvol_replay_err,	/* TX_SYMLINK */
417 	zvol_replay_err,	/* TX_REMOVE */
418 	zvol_replay_err,	/* TX_RMDIR */
419 	zvol_replay_err,	/* TX_LINK */
420 	zvol_replay_err,	/* TX_RENAME */
421 	zvol_replay_write,	/* TX_WRITE */
422 	zvol_replay_err,	/* TX_TRUNCATE */
423 	zvol_replay_err,	/* TX_SETATTR */
424 	zvol_replay_err,	/* TX_ACL */
425 	zvol_replay_err,	/* TX_CREATE_ACL */
426 	zvol_replay_err,	/* TX_CREATE_ATTR */
427 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
428 	zvol_replay_err,	/* TX_MKDIR_ACL */
429 	zvol_replay_err,	/* TX_MKDIR_ATTR */
430 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
431 	zvol_replay_err,	/* TX_WRITE2 */
432 };
433 
434 int
435 zvol_name2minor(const char *name, minor_t *minor)
436 {
437 	zvol_state_t *zv;
438 
439 	mutex_enter(&zvol_state_lock);
440 	zv = zvol_minor_lookup(name);
441 	if (minor && zv)
442 		*minor = zv->zv_minor;
443 	mutex_exit(&zvol_state_lock);
444 	return (zv ? 0 : -1);
445 }
446 
447 /*
448  * Create a minor node (plus a whole lot more) for the specified volume.
449  */
450 int
451 zvol_create_minor(const char *name)
452 {
453 	zvol_state_t *zv;
454 	objset_t *os;
455 	dmu_object_info_t doi;
456 	minor_t minor = 0;
457 	char chrbuf[30], blkbuf[30];
458 	int error;
459 
460 	mutex_enter(&zvol_state_lock);
461 
462 	if (zvol_minor_lookup(name) != NULL) {
463 		mutex_exit(&zvol_state_lock);
464 		return (EEXIST);
465 	}
466 
467 	/* lie and say we're read-only */
468 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os);
469 
470 	if (error) {
471 		mutex_exit(&zvol_state_lock);
472 		return (error);
473 	}
474 
475 	if ((minor = zvol_minor_alloc()) == 0) {
476 		dmu_objset_disown(os, zvol_tag);
477 		mutex_exit(&zvol_state_lock);
478 		return (ENXIO);
479 	}
480 
481 	if (ddi_soft_state_zalloc(zvol_state, minor) != DDI_SUCCESS) {
482 		dmu_objset_disown(os, zvol_tag);
483 		mutex_exit(&zvol_state_lock);
484 		return (EAGAIN);
485 	}
486 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
487 	    (char *)name);
488 
489 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
490 
491 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
492 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
493 		ddi_soft_state_free(zvol_state, minor);
494 		dmu_objset_disown(os, zvol_tag);
495 		mutex_exit(&zvol_state_lock);
496 		return (EAGAIN);
497 	}
498 
499 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
500 
501 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
502 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
503 		ddi_remove_minor_node(zfs_dip, chrbuf);
504 		ddi_soft_state_free(zvol_state, minor);
505 		dmu_objset_disown(os, zvol_tag);
506 		mutex_exit(&zvol_state_lock);
507 		return (EAGAIN);
508 	}
509 
510 	zv = ddi_get_soft_state(zvol_state, minor);
511 
512 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
513 	zv->zv_min_bs = DEV_BSHIFT;
514 	zv->zv_minor = minor;
515 	zv->zv_objset = os;
516 	if (dmu_objset_is_snapshot(os))
517 		zv->zv_flags |= ZVOL_RDONLY;
518 	mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
519 	avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
520 	    sizeof (rl_t), offsetof(rl_t, r_node));
521 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
522 	    offsetof(zvol_extent_t, ze_node));
523 	/* get and cache the blocksize */
524 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
525 	ASSERT(error == 0);
526 	zv->zv_volblocksize = doi.doi_data_block_size;
527 
528 	zil_replay(os, zv, zvol_replay_vector);
529 	dmu_objset_disown(os, zvol_tag);
530 	zv->zv_objset = NULL;
531 
532 	zvol_minors++;
533 
534 	mutex_exit(&zvol_state_lock);
535 
536 	return (0);
537 }
538 
539 /*
540  * Remove minor node for the specified volume.
541  */
542 static int
543 zvol_remove_zv(zvol_state_t *zv)
544 {
545 	char nmbuf[20];
546 
547 	ASSERT(MUTEX_HELD(&zvol_state_lock));
548 	if (zv->zv_total_opens != 0)
549 		return (EBUSY);
550 
551 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", zv->zv_minor);
552 	ddi_remove_minor_node(zfs_dip, nmbuf);
553 
554 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", zv->zv_minor);
555 	ddi_remove_minor_node(zfs_dip, nmbuf);
556 
557 	avl_destroy(&zv->zv_znode.z_range_avl);
558 	mutex_destroy(&zv->zv_znode.z_range_lock);
559 
560 	ddi_soft_state_free(zvol_state, zv->zv_minor);
561 
562 	zvol_minors--;
563 	return (0);
564 }
565 
566 int
567 zvol_remove_minor(const char *name)
568 {
569 	zvol_state_t *zv;
570 	int rc;
571 
572 	mutex_enter(&zvol_state_lock);
573 	if ((zv = zvol_minor_lookup(name)) == NULL) {
574 		mutex_exit(&zvol_state_lock);
575 		return (ENXIO);
576 	}
577 	rc = zvol_remove_zv(zv);
578 	mutex_exit(&zvol_state_lock);
579 	return (rc);
580 }
581 
582 int
583 zvol_first_open(zvol_state_t *zv)
584 {
585 	objset_t *os;
586 	uint64_t volsize;
587 	int error;
588 	uint64_t readonly;
589 
590 	/* lie and say we're read-only */
591 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
592 	    zvol_tag, &os);
593 	if (error)
594 		return (error);
595 
596 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
597 	if (error) {
598 		ASSERT(error == 0);
599 		dmu_objset_disown(os, zvol_tag);
600 		return (error);
601 	}
602 	zv->zv_objset = os;
603 	zv->zv_volsize = volsize;
604 	zv->zv_zilog = zil_open(os, zvol_get_data);
605 	zvol_size_changed(zv->zv_volsize, ddi_driver_major(zfs_dip),
606 	    zv->zv_minor);
607 
608 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
609 	    NULL) == 0);
610 	if (readonly || dmu_objset_is_snapshot(os))
611 		zv->zv_flags |= ZVOL_RDONLY;
612 	else
613 		zv->zv_flags &= ~ZVOL_RDONLY;
614 	return (error);
615 }
616 
617 void
618 zvol_last_close(zvol_state_t *zv)
619 {
620 	zil_close(zv->zv_zilog);
621 	zv->zv_zilog = NULL;
622 	dmu_objset_disown(zv->zv_objset, zvol_tag);
623 	zv->zv_objset = NULL;
624 }
625 
626 int
627 zvol_prealloc(zvol_state_t *zv)
628 {
629 	objset_t *os = zv->zv_objset;
630 	dmu_tx_t *tx;
631 	uint64_t refd, avail, usedobjs, availobjs;
632 	uint64_t resid = zv->zv_volsize;
633 	uint64_t off = 0;
634 
635 	/* Check the space usage before attempting to allocate the space */
636 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
637 	if (avail < zv->zv_volsize)
638 		return (ENOSPC);
639 
640 	/* Free old extents if they exist */
641 	zvol_free_extents(zv);
642 
643 	while (resid != 0) {
644 		int error;
645 		uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
646 
647 		tx = dmu_tx_create(os);
648 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
649 		error = dmu_tx_assign(tx, TXG_WAIT);
650 		if (error) {
651 			dmu_tx_abort(tx);
652 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
653 			return (error);
654 		}
655 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
656 		dmu_tx_commit(tx);
657 		off += bytes;
658 		resid -= bytes;
659 	}
660 	txg_wait_synced(dmu_objset_pool(os), 0);
661 
662 	return (0);
663 }
664 
665 int
666 zvol_update_volsize(objset_t *os, uint64_t volsize)
667 {
668 	dmu_tx_t *tx;
669 	int error;
670 
671 	ASSERT(MUTEX_HELD(&zvol_state_lock));
672 
673 	tx = dmu_tx_create(os);
674 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
675 	error = dmu_tx_assign(tx, TXG_WAIT);
676 	if (error) {
677 		dmu_tx_abort(tx);
678 		return (error);
679 	}
680 
681 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
682 	    &volsize, tx);
683 	dmu_tx_commit(tx);
684 
685 	if (error == 0)
686 		error = dmu_free_long_range(os,
687 		    ZVOL_OBJ, volsize, DMU_OBJECT_END);
688 	return (error);
689 }
690 
691 void
692 zvol_remove_minors(const char *name)
693 {
694 	zvol_state_t *zv;
695 	char *namebuf;
696 	minor_t minor;
697 
698 	namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
699 	(void) strncpy(namebuf, name, strlen(name));
700 	(void) strcat(namebuf, "/");
701 	mutex_enter(&zvol_state_lock);
702 	for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) {
703 
704 		zv = ddi_get_soft_state(zvol_state, minor);
705 		if (zv == NULL)
706 			continue;
707 		if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
708 			(void) zvol_remove_zv(zv);
709 	}
710 	kmem_free(namebuf, strlen(name) + 2);
711 
712 	mutex_exit(&zvol_state_lock);
713 }
714 
715 int
716 zvol_set_volsize(const char *name, major_t maj, uint64_t volsize)
717 {
718 	zvol_state_t *zv = NULL;
719 	objset_t *os;
720 	int error;
721 	dmu_object_info_t doi;
722 	uint64_t old_volsize = 0ULL;
723 	uint64_t readonly;
724 
725 	mutex_enter(&zvol_state_lock);
726 	zv = zvol_minor_lookup(name);
727 	if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
728 		mutex_exit(&zvol_state_lock);
729 		return (error);
730 	}
731 
732 	if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
733 	    (error = zvol_check_volsize(volsize,
734 	    doi.doi_data_block_size)) != 0)
735 		goto out;
736 
737 	VERIFY(dsl_prop_get_integer(name, "readonly", &readonly,
738 	    NULL) == 0);
739 	if (readonly) {
740 		error = EROFS;
741 		goto out;
742 	}
743 
744 	error = zvol_update_volsize(os, volsize);
745 	/*
746 	 * Reinitialize the dump area to the new size. If we
747 	 * failed to resize the dump area then restore it back to
748 	 * its original size.
749 	 */
750 	if (zv && error == 0) {
751 		if (zv->zv_flags & ZVOL_DUMPIFIED) {
752 			old_volsize = zv->zv_volsize;
753 			zv->zv_volsize = volsize;
754 			if ((error = zvol_dumpify(zv)) != 0 ||
755 			    (error = dumpvp_resize()) != 0) {
756 				(void) zvol_update_volsize(os, old_volsize);
757 				zv->zv_volsize = old_volsize;
758 				error = zvol_dumpify(zv);
759 			}
760 		}
761 		if (error == 0) {
762 			zv->zv_volsize = volsize;
763 			zvol_size_changed(volsize, maj, zv->zv_minor);
764 		}
765 	}
766 
767 	/*
768 	 * Generate a LUN expansion event.
769 	 */
770 	if (zv && error == 0) {
771 		sysevent_id_t eid;
772 		nvlist_t *attr;
773 		char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
774 
775 		(void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
776 		    zv->zv_minor);
777 
778 		VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
779 		VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
780 
781 		(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
782 		    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
783 
784 		nvlist_free(attr);
785 		kmem_free(physpath, MAXPATHLEN);
786 	}
787 
788 out:
789 	dmu_objset_rele(os, FTAG);
790 
791 	mutex_exit(&zvol_state_lock);
792 
793 	return (error);
794 }
795 
796 /*ARGSUSED*/
797 int
798 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
799 {
800 	minor_t minor = getminor(*devp);
801 	zvol_state_t *zv;
802 	int err = 0;
803 
804 	if (minor == 0)			/* This is the control device */
805 		return (0);
806 
807 	mutex_enter(&zvol_state_lock);
808 
809 	zv = ddi_get_soft_state(zvol_state, minor);
810 	if (zv == NULL) {
811 		mutex_exit(&zvol_state_lock);
812 		return (ENXIO);
813 	}
814 
815 	if (zv->zv_total_opens == 0)
816 		err = zvol_first_open(zv);
817 	if (err) {
818 		mutex_exit(&zvol_state_lock);
819 		return (err);
820 	}
821 	if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
822 		err = EROFS;
823 		goto out;
824 	}
825 	if (zv->zv_flags & ZVOL_EXCL) {
826 		err = EBUSY;
827 		goto out;
828 	}
829 	if (flag & FEXCL) {
830 		if (zv->zv_total_opens != 0) {
831 			err = EBUSY;
832 			goto out;
833 		}
834 		zv->zv_flags |= ZVOL_EXCL;
835 	}
836 
837 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
838 		zv->zv_open_count[otyp]++;
839 		zv->zv_total_opens++;
840 	}
841 	mutex_exit(&zvol_state_lock);
842 
843 	return (err);
844 out:
845 	if (zv->zv_total_opens == 0)
846 		zvol_last_close(zv);
847 	mutex_exit(&zvol_state_lock);
848 	return (err);
849 }
850 
851 /*ARGSUSED*/
852 int
853 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
854 {
855 	minor_t minor = getminor(dev);
856 	zvol_state_t *zv;
857 	int error = 0;
858 
859 	if (minor == 0)		/* This is the control device */
860 		return (0);
861 
862 	mutex_enter(&zvol_state_lock);
863 
864 	zv = ddi_get_soft_state(zvol_state, minor);
865 	if (zv == NULL) {
866 		mutex_exit(&zvol_state_lock);
867 		return (ENXIO);
868 	}
869 
870 	if (zv->zv_flags & ZVOL_EXCL) {
871 		ASSERT(zv->zv_total_opens == 1);
872 		zv->zv_flags &= ~ZVOL_EXCL;
873 	}
874 
875 	/*
876 	 * If the open count is zero, this is a spurious close.
877 	 * That indicates a bug in the kernel / DDI framework.
878 	 */
879 	ASSERT(zv->zv_open_count[otyp] != 0);
880 	ASSERT(zv->zv_total_opens != 0);
881 
882 	/*
883 	 * You may get multiple opens, but only one close.
884 	 */
885 	zv->zv_open_count[otyp]--;
886 	zv->zv_total_opens--;
887 
888 	if (zv->zv_total_opens == 0)
889 		zvol_last_close(zv);
890 
891 	mutex_exit(&zvol_state_lock);
892 	return (error);
893 }
894 
895 static void
896 zvol_get_done(zgd_t *zgd, int error)
897 {
898 	if (zgd->zgd_db)
899 		dmu_buf_rele(zgd->zgd_db, zgd);
900 
901 	zfs_range_unlock(zgd->zgd_rl);
902 
903 	if (error == 0 && zgd->zgd_bp)
904 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
905 
906 	kmem_free(zgd, sizeof (zgd_t));
907 }
908 
909 /*
910  * Get data to generate a TX_WRITE intent log record.
911  */
912 static int
913 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
914 {
915 	zvol_state_t *zv = arg;
916 	objset_t *os = zv->zv_objset;
917 	uint64_t object = ZVOL_OBJ;
918 	uint64_t offset = lr->lr_offset;
919 	uint64_t size = lr->lr_length;	/* length of user data */
920 	blkptr_t *bp = &lr->lr_blkptr;
921 	dmu_buf_t *db;
922 	zgd_t *zgd;
923 	int error;
924 
925 	ASSERT(zio != NULL);
926 	ASSERT(size != 0);
927 
928 	zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
929 	zgd->zgd_zilog = zv->zv_zilog;
930 	zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
931 
932 	/*
933 	 * Write records come in two flavors: immediate and indirect.
934 	 * For small writes it's cheaper to store the data with the
935 	 * log record (immediate); for large writes it's cheaper to
936 	 * sync the data and get a pointer to it (indirect) so that
937 	 * we don't have to write the data twice.
938 	 */
939 	if (buf != NULL) {	/* immediate write */
940 		error = dmu_read(os, object, offset, size, buf,
941 		    DMU_READ_NO_PREFETCH);
942 	} else {
943 		size = zv->zv_volblocksize;
944 		offset = P2ALIGN(offset, size);
945 		error = dmu_buf_hold(os, object, offset, zgd, &db);
946 		if (error == 0) {
947 			zgd->zgd_db = db;
948 			zgd->zgd_bp = bp;
949 
950 			ASSERT(db->db_offset == offset);
951 			ASSERT(db->db_size == size);
952 
953 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
954 			    zvol_get_done, zgd);
955 
956 			if (error == 0)
957 				return (0);
958 		}
959 	}
960 
961 	zvol_get_done(zgd, error);
962 
963 	return (error);
964 }
965 
966 /*
967  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
968  *
969  * We store data in the log buffers if it's small enough.
970  * Otherwise we will later flush the data out via dmu_sync().
971  */
972 ssize_t zvol_immediate_write_sz = 32768;
973 
974 static void
975 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
976     boolean_t sync)
977 {
978 	uint32_t blocksize = zv->zv_volblocksize;
979 	zilog_t *zilog = zv->zv_zilog;
980 	boolean_t slogging;
981 	ssize_t immediate_write_sz;
982 
983 	if (zil_disable)
984 		return;
985 
986 	if (zil_replaying(zilog, tx))
987 		return;
988 
989 	immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
990 	    ? 0 : zvol_immediate_write_sz;
991 
992 	slogging = spa_has_slogs(zilog->zl_spa) &&
993 	    (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
994 
995 	while (resid) {
996 		itx_t *itx;
997 		lr_write_t *lr;
998 		ssize_t len;
999 		itx_wr_state_t write_state;
1000 
1001 		/*
1002 		 * Unlike zfs_log_write() we can be called with
1003 		 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1004 		 */
1005 		if (blocksize > immediate_write_sz && !slogging &&
1006 		    resid >= blocksize && off % blocksize == 0) {
1007 			write_state = WR_INDIRECT; /* uses dmu_sync */
1008 			len = blocksize;
1009 		} else if (sync) {
1010 			write_state = WR_COPIED;
1011 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1012 		} else {
1013 			write_state = WR_NEED_COPY;
1014 			len = MIN(ZIL_MAX_LOG_DATA, resid);
1015 		}
1016 
1017 		itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1018 		    (write_state == WR_COPIED ? len : 0));
1019 		lr = (lr_write_t *)&itx->itx_lr;
1020 		if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1021 		    ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1022 			zil_itx_destroy(itx);
1023 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1024 			lr = (lr_write_t *)&itx->itx_lr;
1025 			write_state = WR_NEED_COPY;
1026 		}
1027 
1028 		itx->itx_wr_state = write_state;
1029 		if (write_state == WR_NEED_COPY)
1030 			itx->itx_sod += len;
1031 		lr->lr_foid = ZVOL_OBJ;
1032 		lr->lr_offset = off;
1033 		lr->lr_length = len;
1034 		lr->lr_blkoff = 0;
1035 		BP_ZERO(&lr->lr_blkptr);
1036 
1037 		itx->itx_private = zv;
1038 		itx->itx_sync = sync;
1039 
1040 		(void) zil_itx_assign(zilog, itx, tx);
1041 
1042 		off += len;
1043 		resid -= len;
1044 	}
1045 }
1046 
1047 static int
1048 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size,
1049     boolean_t doread, boolean_t isdump)
1050 {
1051 	vdev_disk_t *dvd;
1052 	int c;
1053 	int numerrors = 0;
1054 
1055 	for (c = 0; c < vd->vdev_children; c++) {
1056 		ASSERT(vd->vdev_ops == &vdev_mirror_ops ||
1057 		    vd->vdev_ops == &vdev_replacing_ops ||
1058 		    vd->vdev_ops == &vdev_spare_ops);
1059 		int err = zvol_dumpio_vdev(vd->vdev_child[c],
1060 		    addr, offset, size, doread, isdump);
1061 		if (err != 0) {
1062 			numerrors++;
1063 		} else if (doread) {
1064 			break;
1065 		}
1066 	}
1067 
1068 	if (!vd->vdev_ops->vdev_op_leaf)
1069 		return (numerrors < vd->vdev_children ? 0 : EIO);
1070 
1071 	if (doread && !vdev_readable(vd))
1072 		return (EIO);
1073 	else if (!doread && !vdev_writeable(vd))
1074 		return (EIO);
1075 
1076 	dvd = vd->vdev_tsd;
1077 	ASSERT3P(dvd, !=, NULL);
1078 	offset += VDEV_LABEL_START_SIZE;
1079 
1080 	if (ddi_in_panic() || isdump) {
1081 		ASSERT(!doread);
1082 		if (doread)
1083 			return (EIO);
1084 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1085 		    lbtodb(size)));
1086 	} else {
1087 		return (vdev_disk_physio(dvd->vd_lh, addr, size, offset,
1088 		    doread ? B_READ : B_WRITE));
1089 	}
1090 }
1091 
1092 static int
1093 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1094     boolean_t doread, boolean_t isdump)
1095 {
1096 	vdev_t *vd;
1097 	int error;
1098 	zvol_extent_t *ze;
1099 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1100 
1101 	/* Must be sector aligned, and not stradle a block boundary. */
1102 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1103 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1104 		return (EINVAL);
1105 	}
1106 	ASSERT(size <= zv->zv_volblocksize);
1107 
1108 	/* Locate the extent this belongs to */
1109 	ze = list_head(&zv->zv_extents);
1110 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1111 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1112 		ze = list_next(&zv->zv_extents, ze);
1113 	}
1114 
1115 	if (!ddi_in_panic())
1116 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1117 
1118 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1119 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1120 	error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump);
1121 
1122 	if (!ddi_in_panic())
1123 		spa_config_exit(spa, SCL_STATE, FTAG);
1124 
1125 	return (error);
1126 }
1127 
1128 int
1129 zvol_strategy(buf_t *bp)
1130 {
1131 	zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
1132 	uint64_t off, volsize;
1133 	size_t resid;
1134 	char *addr;
1135 	objset_t *os;
1136 	rl_t *rl;
1137 	int error = 0;
1138 	boolean_t doread = bp->b_flags & B_READ;
1139 	boolean_t is_dump = zv->zv_flags & ZVOL_DUMPIFIED;
1140 	boolean_t sync;
1141 
1142 	if (zv == NULL) {
1143 		bioerror(bp, ENXIO);
1144 		biodone(bp);
1145 		return (0);
1146 	}
1147 
1148 	if (getminor(bp->b_edev) == 0) {
1149 		bioerror(bp, EINVAL);
1150 		biodone(bp);
1151 		return (0);
1152 	}
1153 
1154 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1155 		bioerror(bp, EROFS);
1156 		biodone(bp);
1157 		return (0);
1158 	}
1159 
1160 	off = ldbtob(bp->b_blkno);
1161 	volsize = zv->zv_volsize;
1162 
1163 	os = zv->zv_objset;
1164 	ASSERT(os != NULL);
1165 
1166 	bp_mapin(bp);
1167 	addr = bp->b_un.b_addr;
1168 	resid = bp->b_bcount;
1169 
1170 	if (resid > 0 && (off < 0 || off >= volsize)) {
1171 		bioerror(bp, EIO);
1172 		biodone(bp);
1173 		return (0);
1174 	}
1175 
1176 	sync = !(bp->b_flags & B_ASYNC) && !doread && !is_dump &&
1177 	    !(zv->zv_flags & ZVOL_WCE) && !zil_disable;
1178 
1179 	/*
1180 	 * There must be no buffer changes when doing a dmu_sync() because
1181 	 * we can't change the data whilst calculating the checksum.
1182 	 */
1183 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1184 	    doread ? RL_READER : RL_WRITER);
1185 
1186 	while (resid != 0 && off < volsize) {
1187 		size_t size = MIN(resid, zvol_maxphys);
1188 		if (is_dump) {
1189 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1190 			error = zvol_dumpio(zv, addr, off, size,
1191 			    doread, B_FALSE);
1192 		} else if (doread) {
1193 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1194 			    DMU_READ_PREFETCH);
1195 		} else {
1196 			dmu_tx_t *tx = dmu_tx_create(os);
1197 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1198 			error = dmu_tx_assign(tx, TXG_WAIT);
1199 			if (error) {
1200 				dmu_tx_abort(tx);
1201 			} else {
1202 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1203 				zvol_log_write(zv, tx, off, size, sync);
1204 				dmu_tx_commit(tx);
1205 			}
1206 		}
1207 		if (error) {
1208 			/* convert checksum errors into IO errors */
1209 			if (error == ECKSUM)
1210 				error = EIO;
1211 			break;
1212 		}
1213 		off += size;
1214 		addr += size;
1215 		resid -= size;
1216 	}
1217 	zfs_range_unlock(rl);
1218 
1219 	if ((bp->b_resid = resid) == bp->b_bcount)
1220 		bioerror(bp, off > volsize ? EINVAL : error);
1221 
1222 	if (sync)
1223 		zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
1224 	biodone(bp);
1225 
1226 	return (0);
1227 }
1228 
1229 /*
1230  * Set the buffer count to the zvol maximum transfer.
1231  * Using our own routine instead of the default minphys()
1232  * means that for larger writes we write bigger buffers on X86
1233  * (128K instead of 56K) and flush the disk write cache less often
1234  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1235  * 56K on X86 and 128K on sparc).
1236  */
1237 void
1238 zvol_minphys(struct buf *bp)
1239 {
1240 	if (bp->b_bcount > zvol_maxphys)
1241 		bp->b_bcount = zvol_maxphys;
1242 }
1243 
1244 int
1245 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1246 {
1247 	minor_t minor = getminor(dev);
1248 	zvol_state_t *zv;
1249 	int error = 0;
1250 	uint64_t size;
1251 	uint64_t boff;
1252 	uint64_t resid;
1253 
1254 	if (minor == 0)			/* This is the control device */
1255 		return (ENXIO);
1256 
1257 	zv = ddi_get_soft_state(zvol_state, minor);
1258 	if (zv == NULL)
1259 		return (ENXIO);
1260 
1261 	boff = ldbtob(blkno);
1262 	resid = ldbtob(nblocks);
1263 
1264 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1265 
1266 	while (resid) {
1267 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1268 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1269 		if (error)
1270 			break;
1271 		boff += size;
1272 		addr += size;
1273 		resid -= size;
1274 	}
1275 
1276 	return (error);
1277 }
1278 
1279 /*ARGSUSED*/
1280 int
1281 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1282 {
1283 	minor_t minor = getminor(dev);
1284 	zvol_state_t *zv;
1285 	uint64_t volsize;
1286 	rl_t *rl;
1287 	int error = 0;
1288 
1289 	if (minor == 0)			/* This is the control device */
1290 		return (ENXIO);
1291 
1292 	zv = ddi_get_soft_state(zvol_state, minor);
1293 	if (zv == NULL)
1294 		return (ENXIO);
1295 
1296 	volsize = zv->zv_volsize;
1297 	if (uio->uio_resid > 0 &&
1298 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1299 		return (EIO);
1300 
1301 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1302 		error = physio(zvol_strategy, NULL, dev, B_READ,
1303 		    zvol_minphys, uio);
1304 		return (error);
1305 	}
1306 
1307 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1308 	    RL_READER);
1309 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1310 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1311 
1312 		/* don't read past the end */
1313 		if (bytes > volsize - uio->uio_loffset)
1314 			bytes = volsize - uio->uio_loffset;
1315 
1316 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1317 		if (error) {
1318 			/* convert checksum errors into IO errors */
1319 			if (error == ECKSUM)
1320 				error = EIO;
1321 			break;
1322 		}
1323 	}
1324 	zfs_range_unlock(rl);
1325 	return (error);
1326 }
1327 
1328 /*ARGSUSED*/
1329 int
1330 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1331 {
1332 	minor_t minor = getminor(dev);
1333 	zvol_state_t *zv;
1334 	uint64_t volsize;
1335 	rl_t *rl;
1336 	int error = 0;
1337 	boolean_t sync;
1338 
1339 	if (minor == 0)			/* This is the control device */
1340 		return (ENXIO);
1341 
1342 	zv = ddi_get_soft_state(zvol_state, minor);
1343 	if (zv == NULL)
1344 		return (ENXIO);
1345 
1346 	volsize = zv->zv_volsize;
1347 	if (uio->uio_resid > 0 &&
1348 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1349 		return (EIO);
1350 
1351 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1352 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1353 		    zvol_minphys, uio);
1354 		return (error);
1355 	}
1356 
1357 	sync = !(zv->zv_flags & ZVOL_WCE) && !zil_disable;
1358 
1359 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1360 	    RL_WRITER);
1361 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1362 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1363 		uint64_t off = uio->uio_loffset;
1364 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1365 
1366 		if (bytes > volsize - off)	/* don't write past the end */
1367 			bytes = volsize - off;
1368 
1369 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1370 		error = dmu_tx_assign(tx, TXG_WAIT);
1371 		if (error) {
1372 			dmu_tx_abort(tx);
1373 			break;
1374 		}
1375 		error = dmu_write_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes, tx);
1376 		if (error == 0)
1377 			zvol_log_write(zv, tx, off, bytes, sync);
1378 		dmu_tx_commit(tx);
1379 
1380 		if (error)
1381 			break;
1382 	}
1383 	zfs_range_unlock(rl);
1384 	if (sync)
1385 		zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
1386 	return (error);
1387 }
1388 
1389 int
1390 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1391 {
1392 	struct uuid uuid = EFI_RESERVED;
1393 	efi_gpe_t gpe = { 0 };
1394 	uint32_t crc;
1395 	dk_efi_t efi;
1396 	int length;
1397 	char *ptr;
1398 
1399 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1400 		return (EFAULT);
1401 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1402 	length = efi.dki_length;
1403 	/*
1404 	 * Some clients may attempt to request a PMBR for the
1405 	 * zvol.  Currently this interface will return EINVAL to
1406 	 * such requests.  These requests could be supported by
1407 	 * adding a check for lba == 0 and consing up an appropriate
1408 	 * PMBR.
1409 	 */
1410 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1411 		return (EINVAL);
1412 
1413 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1414 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1415 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1416 
1417 	if (efi.dki_lba == 1) {
1418 		efi_gpt_t gpt = { 0 };
1419 
1420 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1421 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1422 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1423 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1424 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1425 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1426 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1427 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1428 		gpt.efi_gpt_SizeOfPartitionEntry =
1429 		    LE_32(sizeof (efi_gpe_t));
1430 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1431 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1432 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1433 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1434 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1435 		    flag))
1436 			return (EFAULT);
1437 		ptr += sizeof (gpt);
1438 		length -= sizeof (gpt);
1439 	}
1440 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1441 	    length), flag))
1442 		return (EFAULT);
1443 	return (0);
1444 }
1445 
1446 /*
1447  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1448  */
1449 /*ARGSUSED*/
1450 int
1451 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1452 {
1453 	zvol_state_t *zv;
1454 	struct dk_cinfo dki;
1455 	struct dk_minfo dkm;
1456 	struct dk_callback *dkc;
1457 	int error = 0;
1458 	rl_t *rl;
1459 
1460 	mutex_enter(&zvol_state_lock);
1461 
1462 	zv = ddi_get_soft_state(zvol_state, getminor(dev));
1463 
1464 	if (zv == NULL) {
1465 		mutex_exit(&zvol_state_lock);
1466 		return (ENXIO);
1467 	}
1468 	ASSERT(zv->zv_total_opens > 0);
1469 
1470 	switch (cmd) {
1471 
1472 	case DKIOCINFO:
1473 		bzero(&dki, sizeof (dki));
1474 		(void) strcpy(dki.dki_cname, "zvol");
1475 		(void) strcpy(dki.dki_dname, "zvol");
1476 		dki.dki_ctype = DKC_UNKNOWN;
1477 		dki.dki_unit = getminor(dev);
1478 		dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1479 		mutex_exit(&zvol_state_lock);
1480 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1481 			error = EFAULT;
1482 		return (error);
1483 
1484 	case DKIOCGMEDIAINFO:
1485 		bzero(&dkm, sizeof (dkm));
1486 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1487 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1488 		dkm.dki_media_type = DK_UNKNOWN;
1489 		mutex_exit(&zvol_state_lock);
1490 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1491 			error = EFAULT;
1492 		return (error);
1493 
1494 	case DKIOCGETEFI:
1495 		{
1496 			uint64_t vs = zv->zv_volsize;
1497 			uint8_t bs = zv->zv_min_bs;
1498 
1499 			mutex_exit(&zvol_state_lock);
1500 			error = zvol_getefi((void *)arg, flag, vs, bs);
1501 			return (error);
1502 		}
1503 
1504 	case DKIOCFLUSHWRITECACHE:
1505 		dkc = (struct dk_callback *)arg;
1506 		mutex_exit(&zvol_state_lock);
1507 		zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
1508 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1509 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1510 			error = 0;
1511 		}
1512 		return (error);
1513 
1514 	case DKIOCGETWCE:
1515 		{
1516 			int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1517 			if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1518 			    flag))
1519 				error = EFAULT;
1520 			break;
1521 		}
1522 	case DKIOCSETWCE:
1523 		{
1524 			int wce;
1525 			if (ddi_copyin((void *)arg, &wce, sizeof (int),
1526 			    flag)) {
1527 				error = EFAULT;
1528 				break;
1529 			}
1530 			if (wce) {
1531 				zv->zv_flags |= ZVOL_WCE;
1532 				mutex_exit(&zvol_state_lock);
1533 			} else {
1534 				zv->zv_flags &= ~ZVOL_WCE;
1535 				mutex_exit(&zvol_state_lock);
1536 				zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
1537 			}
1538 			return (0);
1539 		}
1540 
1541 	case DKIOCGGEOM:
1542 	case DKIOCGVTOC:
1543 		/*
1544 		 * commands using these (like prtvtoc) expect ENOTSUP
1545 		 * since we're emulating an EFI label
1546 		 */
1547 		error = ENOTSUP;
1548 		break;
1549 
1550 	case DKIOCDUMPINIT:
1551 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1552 		    RL_WRITER);
1553 		error = zvol_dumpify(zv);
1554 		zfs_range_unlock(rl);
1555 		break;
1556 
1557 	case DKIOCDUMPFINI:
1558 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1559 			break;
1560 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1561 		    RL_WRITER);
1562 		error = zvol_dump_fini(zv);
1563 		zfs_range_unlock(rl);
1564 		break;
1565 
1566 	default:
1567 		error = ENOTTY;
1568 		break;
1569 
1570 	}
1571 	mutex_exit(&zvol_state_lock);
1572 	return (error);
1573 }
1574 
1575 int
1576 zvol_busy(void)
1577 {
1578 	return (zvol_minors != 0);
1579 }
1580 
1581 void
1582 zvol_init(void)
1583 {
1584 	VERIFY(ddi_soft_state_init(&zvol_state, sizeof (zvol_state_t), 1) == 0);
1585 	mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
1586 }
1587 
1588 void
1589 zvol_fini(void)
1590 {
1591 	mutex_destroy(&zvol_state_lock);
1592 	ddi_soft_state_fini(&zvol_state);
1593 }
1594 
1595 static int
1596 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1597 {
1598 	dmu_tx_t *tx;
1599 	int error = 0;
1600 	objset_t *os = zv->zv_objset;
1601 	nvlist_t *nv = NULL;
1602 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1603 
1604 	ASSERT(MUTEX_HELD(&zvol_state_lock));
1605 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1606 	    DMU_OBJECT_END);
1607 	/* wait for dmu_free_long_range to actually free the blocks */
1608 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1609 
1610 	tx = dmu_tx_create(os);
1611 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1612 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1613 	error = dmu_tx_assign(tx, TXG_WAIT);
1614 	if (error) {
1615 		dmu_tx_abort(tx);
1616 		return (error);
1617 	}
1618 
1619 	/*
1620 	 * If we are resizing the dump device then we only need to
1621 	 * update the refreservation to match the newly updated
1622 	 * zvolsize. Otherwise, we save off the original state of the
1623 	 * zvol so that we can restore them if the zvol is ever undumpified.
1624 	 */
1625 	if (resize) {
1626 		error = zap_update(os, ZVOL_ZAP_OBJ,
1627 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1628 		    &zv->zv_volsize, tx);
1629 	} else {
1630 		uint64_t checksum, compress, refresrv, vbs, dedup;
1631 
1632 		error = dsl_prop_get_integer(zv->zv_name,
1633 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1634 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1635 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1636 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1637 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1638 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1639 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1640 		if (version >= SPA_VERSION_DEDUP) {
1641 			error = error ? error :
1642 			    dsl_prop_get_integer(zv->zv_name,
1643 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1644 		}
1645 
1646 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1647 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1648 		    &compress, tx);
1649 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1650 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1651 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1652 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1653 		    &refresrv, tx);
1654 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1655 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1656 		    &vbs, tx);
1657 		error = error ? error : dmu_object_set_blocksize(
1658 		    os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1659 		if (version >= SPA_VERSION_DEDUP) {
1660 			error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1661 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1662 			    &dedup, tx);
1663 		}
1664 		if (error == 0)
1665 			zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1666 	}
1667 	dmu_tx_commit(tx);
1668 
1669 	/*
1670 	 * We only need update the zvol's property if we are initializing
1671 	 * the dump area for the first time.
1672 	 */
1673 	if (!resize) {
1674 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1675 		VERIFY(nvlist_add_uint64(nv,
1676 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1677 		VERIFY(nvlist_add_uint64(nv,
1678 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1679 		    ZIO_COMPRESS_OFF) == 0);
1680 		VERIFY(nvlist_add_uint64(nv,
1681 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1682 		    ZIO_CHECKSUM_OFF) == 0);
1683 		if (version >= SPA_VERSION_DEDUP) {
1684 			VERIFY(nvlist_add_uint64(nv,
1685 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
1686 			    ZIO_CHECKSUM_OFF) == 0);
1687 		}
1688 
1689 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1690 		    nv, NULL);
1691 		nvlist_free(nv);
1692 
1693 		if (error)
1694 			return (error);
1695 	}
1696 
1697 	/* Allocate the space for the dump */
1698 	error = zvol_prealloc(zv);
1699 	return (error);
1700 }
1701 
1702 static int
1703 zvol_dumpify(zvol_state_t *zv)
1704 {
1705 	int error = 0;
1706 	uint64_t dumpsize = 0;
1707 	dmu_tx_t *tx;
1708 	objset_t *os = zv->zv_objset;
1709 
1710 	if (zv->zv_flags & ZVOL_RDONLY)
1711 		return (EROFS);
1712 
1713 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
1714 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
1715 		boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE;
1716 
1717 		if ((error = zvol_dump_init(zv, resize)) != 0) {
1718 			(void) zvol_dump_fini(zv);
1719 			return (error);
1720 		}
1721 	}
1722 
1723 	/*
1724 	 * Build up our lba mapping.
1725 	 */
1726 	error = zvol_get_lbas(zv);
1727 	if (error) {
1728 		(void) zvol_dump_fini(zv);
1729 		return (error);
1730 	}
1731 
1732 	tx = dmu_tx_create(os);
1733 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1734 	error = dmu_tx_assign(tx, TXG_WAIT);
1735 	if (error) {
1736 		dmu_tx_abort(tx);
1737 		(void) zvol_dump_fini(zv);
1738 		return (error);
1739 	}
1740 
1741 	zv->zv_flags |= ZVOL_DUMPIFIED;
1742 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
1743 	    &zv->zv_volsize, tx);
1744 	dmu_tx_commit(tx);
1745 
1746 	if (error) {
1747 		(void) zvol_dump_fini(zv);
1748 		return (error);
1749 	}
1750 
1751 	txg_wait_synced(dmu_objset_pool(os), 0);
1752 	return (0);
1753 }
1754 
1755 static int
1756 zvol_dump_fini(zvol_state_t *zv)
1757 {
1758 	dmu_tx_t *tx;
1759 	objset_t *os = zv->zv_objset;
1760 	nvlist_t *nv;
1761 	int error = 0;
1762 	uint64_t checksum, compress, refresrv, vbs, dedup;
1763 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1764 
1765 	/*
1766 	 * Attempt to restore the zvol back to its pre-dumpified state.
1767 	 * This is a best-effort attempt as it's possible that not all
1768 	 * of these properties were initialized during the dumpify process
1769 	 * (i.e. error during zvol_dump_init).
1770 	 */
1771 
1772 	tx = dmu_tx_create(os);
1773 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1774 	error = dmu_tx_assign(tx, TXG_WAIT);
1775 	if (error) {
1776 		dmu_tx_abort(tx);
1777 		return (error);
1778 	}
1779 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
1780 	dmu_tx_commit(tx);
1781 
1782 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1783 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
1784 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1785 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
1786 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1787 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
1788 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1789 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
1790 
1791 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1792 	(void) nvlist_add_uint64(nv,
1793 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
1794 	(void) nvlist_add_uint64(nv,
1795 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
1796 	(void) nvlist_add_uint64(nv,
1797 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
1798 	if (version >= SPA_VERSION_DEDUP &&
1799 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1800 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
1801 		(void) nvlist_add_uint64(nv,
1802 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
1803 	}
1804 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1805 	    nv, NULL);
1806 	nvlist_free(nv);
1807 
1808 	zvol_free_extents(zv);
1809 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
1810 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
1811 	/* wait for dmu_free_long_range to actually free the blocks */
1812 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1813 	tx = dmu_tx_create(os);
1814 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1815 	error = dmu_tx_assign(tx, TXG_WAIT);
1816 	if (error) {
1817 		dmu_tx_abort(tx);
1818 		return (error);
1819 	}
1820 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
1821 		zv->zv_volblocksize = vbs;
1822 	dmu_tx_commit(tx);
1823 
1824 	return (0);
1825 }
1826