xref: /titanic_51/usr/src/uts/common/fs/zfs/zvol.c (revision 11cb6ef00e4913c161e88403c11c037fb2afab44)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  *
24  * Portions Copyright 2010 Robert Milkowski
25  *
26  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
27  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
28  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29  */
30 
31 /*
32  * ZFS volume emulation driver.
33  *
34  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
35  * Volumes are accessed through the symbolic links named:
36  *
37  * /dev/zvol/dsk/<pool_name>/<dataset_name>
38  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
39  *
40  * These links are created by the /dev filesystem (sdev_zvolops.c).
41  * Volumes are persistent through reboot.  No user command needs to be
42  * run before opening and using a device.
43  */
44 
45 #include <sys/types.h>
46 #include <sys/param.h>
47 #include <sys/errno.h>
48 #include <sys/uio.h>
49 #include <sys/buf.h>
50 #include <sys/modctl.h>
51 #include <sys/open.h>
52 #include <sys/kmem.h>
53 #include <sys/conf.h>
54 #include <sys/cmn_err.h>
55 #include <sys/stat.h>
56 #include <sys/zap.h>
57 #include <sys/spa.h>
58 #include <sys/spa_impl.h>
59 #include <sys/zio.h>
60 #include <sys/dmu_traverse.h>
61 #include <sys/dnode.h>
62 #include <sys/dsl_dataset.h>
63 #include <sys/dsl_prop.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/byteorder.h>
67 #include <sys/pathname.h>
68 #include <sys/ddi.h>
69 #include <sys/sunddi.h>
70 #include <sys/crc32.h>
71 #include <sys/dirent.h>
72 #include <sys/policy.h>
73 #include <sys/fs/zfs.h>
74 #include <sys/zfs_ioctl.h>
75 #include <sys/mkdev.h>
76 #include <sys/zil.h>
77 #include <sys/refcount.h>
78 #include <sys/zfs_znode.h>
79 #include <sys/zfs_rlock.h>
80 #include <sys/vdev_disk.h>
81 #include <sys/vdev_impl.h>
82 #include <sys/vdev_raidz.h>
83 #include <sys/zvol.h>
84 #include <sys/dumphdr.h>
85 #include <sys/zil_impl.h>
86 #include <sys/dbuf.h>
87 #include <sys/dmu_tx.h>
88 #include <sys/zfeature.h>
89 #include <sys/zio_checksum.h>
90 
91 #include "zfs_namecheck.h"
92 
93 void *zfsdev_state;
94 static char *zvol_tag = "zvol_tag";
95 
96 #define	ZVOL_DUMPSIZE		"dumpsize"
97 
98 /*
99  * This lock protects the zfsdev_state structure from being modified
100  * while it's being used, e.g. an open that comes in before a create
101  * finishes.  It also protects temporary opens of the dataset so that,
102  * e.g., an open doesn't get a spurious EBUSY.
103  */
104 kmutex_t zfsdev_state_lock;
105 static uint32_t zvol_minors;
106 
107 typedef struct zvol_extent {
108 	list_node_t	ze_node;
109 	dva_t		ze_dva;		/* dva associated with this extent */
110 	uint64_t	ze_nblks;	/* number of blocks in extent */
111 } zvol_extent_t;
112 
113 /*
114  * The in-core state of each volume.
115  */
116 typedef struct zvol_state {
117 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
118 	uint64_t	zv_volsize;	/* amount of space we advertise */
119 	uint64_t	zv_volblocksize; /* volume block size */
120 	minor_t		zv_minor;	/* minor number */
121 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
122 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
123 	objset_t	*zv_objset;	/* objset handle */
124 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
125 	uint32_t	zv_total_opens;	/* total open count */
126 	zilog_t		*zv_zilog;	/* ZIL handle */
127 	list_t		zv_extents;	/* List of extents for dump */
128 	znode_t		zv_znode;	/* for range locking */
129 	dmu_buf_t	*zv_dbuf;	/* bonus handle */
130 } zvol_state_t;
131 
132 /*
133  * zvol specific flags
134  */
135 #define	ZVOL_RDONLY	0x1
136 #define	ZVOL_DUMPIFIED	0x2
137 #define	ZVOL_EXCL	0x4
138 #define	ZVOL_WCE	0x8
139 
140 /*
141  * zvol maximum transfer in one DMU tx.
142  */
143 int zvol_maxphys = DMU_MAX_ACCESS/2;
144 
145 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
146     nvlist_t *, nvlist_t *);
147 static int zvol_remove_zv(zvol_state_t *);
148 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
149 static int zvol_dumpify(zvol_state_t *zv);
150 static int zvol_dump_fini(zvol_state_t *zv);
151 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
152 
153 static void
154 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
155 {
156 	dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
157 
158 	zv->zv_volsize = volsize;
159 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
160 	    "Size", volsize) == DDI_SUCCESS);
161 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
162 	    "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
163 
164 	/* Notify specfs to invalidate the cached size */
165 	spec_size_invalidate(dev, VBLK);
166 	spec_size_invalidate(dev, VCHR);
167 }
168 
169 int
170 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
171 {
172 	if (volsize == 0)
173 		return (SET_ERROR(EINVAL));
174 
175 	if (volsize % blocksize != 0)
176 		return (SET_ERROR(EINVAL));
177 
178 #ifdef _ILP32
179 	if (volsize - 1 > SPEC_MAXOFFSET_T)
180 		return (SET_ERROR(EOVERFLOW));
181 #endif
182 	return (0);
183 }
184 
185 int
186 zvol_check_volblocksize(uint64_t volblocksize)
187 {
188 	if (volblocksize < SPA_MINBLOCKSIZE ||
189 	    volblocksize > SPA_MAXBLOCKSIZE ||
190 	    !ISP2(volblocksize))
191 		return (SET_ERROR(EDOM));
192 
193 	return (0);
194 }
195 
196 int
197 zvol_get_stats(objset_t *os, nvlist_t *nv)
198 {
199 	int error;
200 	dmu_object_info_t doi;
201 	uint64_t val;
202 
203 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
204 	if (error)
205 		return (error);
206 
207 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
208 
209 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
210 
211 	if (error == 0) {
212 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
213 		    doi.doi_data_block_size);
214 	}
215 
216 	return (error);
217 }
218 
219 static zvol_state_t *
220 zvol_minor_lookup(const char *name)
221 {
222 	minor_t minor;
223 	zvol_state_t *zv;
224 
225 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
226 
227 	for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
228 		zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
229 		if (zv == NULL)
230 			continue;
231 		if (strcmp(zv->zv_name, name) == 0)
232 			return (zv);
233 	}
234 
235 	return (NULL);
236 }
237 
238 /* extent mapping arg */
239 struct maparg {
240 	zvol_state_t	*ma_zv;
241 	uint64_t	ma_blks;
242 };
243 
244 /*ARGSUSED*/
245 static int
246 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
247     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
248 {
249 	struct maparg *ma = arg;
250 	zvol_extent_t *ze;
251 	int bs = ma->ma_zv->zv_volblocksize;
252 
253 	if (BP_IS_HOLE(bp) ||
254 	    zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
255 		return (0);
256 
257 	VERIFY(!BP_IS_EMBEDDED(bp));
258 
259 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
260 	ma->ma_blks++;
261 
262 	/* Abort immediately if we have encountered gang blocks */
263 	if (BP_IS_GANG(bp))
264 		return (SET_ERROR(EFRAGS));
265 
266 	/*
267 	 * See if the block is at the end of the previous extent.
268 	 */
269 	ze = list_tail(&ma->ma_zv->zv_extents);
270 	if (ze &&
271 	    DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
272 	    DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
273 	    DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
274 		ze->ze_nblks++;
275 		return (0);
276 	}
277 
278 	dprintf_bp(bp, "%s", "next blkptr:");
279 
280 	/* start a new extent */
281 	ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
282 	ze->ze_dva = bp->blk_dva[0];	/* structure assignment */
283 	ze->ze_nblks = 1;
284 	list_insert_tail(&ma->ma_zv->zv_extents, ze);
285 	return (0);
286 }
287 
288 static void
289 zvol_free_extents(zvol_state_t *zv)
290 {
291 	zvol_extent_t *ze;
292 
293 	while (ze = list_head(&zv->zv_extents)) {
294 		list_remove(&zv->zv_extents, ze);
295 		kmem_free(ze, sizeof (zvol_extent_t));
296 	}
297 }
298 
299 static int
300 zvol_get_lbas(zvol_state_t *zv)
301 {
302 	objset_t *os = zv->zv_objset;
303 	struct maparg	ma;
304 	int		err;
305 
306 	ma.ma_zv = zv;
307 	ma.ma_blks = 0;
308 	zvol_free_extents(zv);
309 
310 	/* commit any in-flight changes before traversing the dataset */
311 	txg_wait_synced(dmu_objset_pool(os), 0);
312 	err = traverse_dataset(dmu_objset_ds(os), 0,
313 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
314 	if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
315 		zvol_free_extents(zv);
316 		return (err ? err : EIO);
317 	}
318 
319 	return (0);
320 }
321 
322 /* ARGSUSED */
323 void
324 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
325 {
326 	zfs_creat_t *zct = arg;
327 	nvlist_t *nvprops = zct->zct_props;
328 	int error;
329 	uint64_t volblocksize, volsize;
330 
331 	VERIFY(nvlist_lookup_uint64(nvprops,
332 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
333 	if (nvlist_lookup_uint64(nvprops,
334 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
335 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
336 
337 	/*
338 	 * These properties must be removed from the list so the generic
339 	 * property setting step won't apply to them.
340 	 */
341 	VERIFY(nvlist_remove_all(nvprops,
342 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
343 	(void) nvlist_remove_all(nvprops,
344 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
345 
346 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
347 	    DMU_OT_NONE, 0, tx);
348 	ASSERT(error == 0);
349 
350 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
351 	    DMU_OT_NONE, 0, tx);
352 	ASSERT(error == 0);
353 
354 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
355 	ASSERT(error == 0);
356 }
357 
358 /*
359  * Replay a TX_TRUNCATE ZIL transaction if asked.  TX_TRUNCATE is how we
360  * implement DKIOCFREE/free-long-range.
361  */
362 static int
363 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
364 {
365 	uint64_t offset, length;
366 
367 	if (byteswap)
368 		byteswap_uint64_array(lr, sizeof (*lr));
369 
370 	offset = lr->lr_offset;
371 	length = lr->lr_length;
372 
373 	return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
374 }
375 
376 /*
377  * Replay a TX_WRITE ZIL transaction that didn't get committed
378  * after a system failure
379  */
380 static int
381 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
382 {
383 	objset_t *os = zv->zv_objset;
384 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
385 	uint64_t offset, length;
386 	dmu_tx_t *tx;
387 	int error;
388 
389 	if (byteswap)
390 		byteswap_uint64_array(lr, sizeof (*lr));
391 
392 	offset = lr->lr_offset;
393 	length = lr->lr_length;
394 
395 	/* If it's a dmu_sync() block, write the whole block */
396 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
397 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
398 		if (length < blocksize) {
399 			offset -= offset % blocksize;
400 			length = blocksize;
401 		}
402 	}
403 
404 	tx = dmu_tx_create(os);
405 	dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
406 	error = dmu_tx_assign(tx, TXG_WAIT);
407 	if (error) {
408 		dmu_tx_abort(tx);
409 	} else {
410 		dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
411 		dmu_tx_commit(tx);
412 	}
413 
414 	return (error);
415 }
416 
417 /* ARGSUSED */
418 static int
419 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
420 {
421 	return (SET_ERROR(ENOTSUP));
422 }
423 
424 /*
425  * Callback vectors for replaying records.
426  * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
427  */
428 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
429 	zvol_replay_err,	/* 0 no such transaction type */
430 	zvol_replay_err,	/* TX_CREATE */
431 	zvol_replay_err,	/* TX_MKDIR */
432 	zvol_replay_err,	/* TX_MKXATTR */
433 	zvol_replay_err,	/* TX_SYMLINK */
434 	zvol_replay_err,	/* TX_REMOVE */
435 	zvol_replay_err,	/* TX_RMDIR */
436 	zvol_replay_err,	/* TX_LINK */
437 	zvol_replay_err,	/* TX_RENAME */
438 	zvol_replay_write,	/* TX_WRITE */
439 	zvol_replay_truncate,	/* TX_TRUNCATE */
440 	zvol_replay_err,	/* TX_SETATTR */
441 	zvol_replay_err,	/* TX_ACL */
442 	zvol_replay_err,	/* TX_CREATE_ACL */
443 	zvol_replay_err,	/* TX_CREATE_ATTR */
444 	zvol_replay_err,	/* TX_CREATE_ACL_ATTR */
445 	zvol_replay_err,	/* TX_MKDIR_ACL */
446 	zvol_replay_err,	/* TX_MKDIR_ATTR */
447 	zvol_replay_err,	/* TX_MKDIR_ACL_ATTR */
448 	zvol_replay_err,	/* TX_WRITE2 */
449 };
450 
451 int
452 zvol_name2minor(const char *name, minor_t *minor)
453 {
454 	zvol_state_t *zv;
455 
456 	mutex_enter(&zfsdev_state_lock);
457 	zv = zvol_minor_lookup(name);
458 	if (minor && zv)
459 		*minor = zv->zv_minor;
460 	mutex_exit(&zfsdev_state_lock);
461 	return (zv ? 0 : -1);
462 }
463 
464 /*
465  * Create a minor node (plus a whole lot more) for the specified volume.
466  */
467 int
468 zvol_create_minor(const char *name)
469 {
470 	zfs_soft_state_t *zs;
471 	zvol_state_t *zv;
472 	objset_t *os;
473 	dmu_object_info_t doi;
474 	minor_t minor = 0;
475 	char chrbuf[30], blkbuf[30];
476 	int error;
477 
478 	mutex_enter(&zfsdev_state_lock);
479 
480 	if (zvol_minor_lookup(name) != NULL) {
481 		mutex_exit(&zfsdev_state_lock);
482 		return (SET_ERROR(EEXIST));
483 	}
484 
485 	/* lie and say we're read-only */
486 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
487 
488 	if (error) {
489 		mutex_exit(&zfsdev_state_lock);
490 		return (error);
491 	}
492 
493 	if ((minor = zfsdev_minor_alloc()) == 0) {
494 		dmu_objset_disown(os, FTAG);
495 		mutex_exit(&zfsdev_state_lock);
496 		return (SET_ERROR(ENXIO));
497 	}
498 
499 	if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
500 		dmu_objset_disown(os, FTAG);
501 		mutex_exit(&zfsdev_state_lock);
502 		return (SET_ERROR(EAGAIN));
503 	}
504 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
505 	    (char *)name);
506 
507 	(void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
508 
509 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
510 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
511 		ddi_soft_state_free(zfsdev_state, minor);
512 		dmu_objset_disown(os, FTAG);
513 		mutex_exit(&zfsdev_state_lock);
514 		return (SET_ERROR(EAGAIN));
515 	}
516 
517 	(void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
518 
519 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
520 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
521 		ddi_remove_minor_node(zfs_dip, chrbuf);
522 		ddi_soft_state_free(zfsdev_state, minor);
523 		dmu_objset_disown(os, FTAG);
524 		mutex_exit(&zfsdev_state_lock);
525 		return (SET_ERROR(EAGAIN));
526 	}
527 
528 	zs = ddi_get_soft_state(zfsdev_state, minor);
529 	zs->zss_type = ZSST_ZVOL;
530 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
531 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
532 	zv->zv_min_bs = DEV_BSHIFT;
533 	zv->zv_minor = minor;
534 	zv->zv_objset = os;
535 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
536 		zv->zv_flags |= ZVOL_RDONLY;
537 	mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
538 	avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
539 	    sizeof (rl_t), offsetof(rl_t, r_node));
540 	list_create(&zv->zv_extents, sizeof (zvol_extent_t),
541 	    offsetof(zvol_extent_t, ze_node));
542 	/* get and cache the blocksize */
543 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
544 	ASSERT(error == 0);
545 	zv->zv_volblocksize = doi.doi_data_block_size;
546 
547 	if (spa_writeable(dmu_objset_spa(os))) {
548 		if (zil_replay_disable)
549 			zil_destroy(dmu_objset_zil(os), B_FALSE);
550 		else
551 			zil_replay(os, zv, zvol_replay_vector);
552 	}
553 	dmu_objset_disown(os, FTAG);
554 	zv->zv_objset = NULL;
555 
556 	zvol_minors++;
557 
558 	mutex_exit(&zfsdev_state_lock);
559 
560 	return (0);
561 }
562 
563 /*
564  * Remove minor node for the specified volume.
565  */
566 static int
567 zvol_remove_zv(zvol_state_t *zv)
568 {
569 	char nmbuf[20];
570 	minor_t minor = zv->zv_minor;
571 
572 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
573 	if (zv->zv_total_opens != 0)
574 		return (SET_ERROR(EBUSY));
575 
576 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
577 	ddi_remove_minor_node(zfs_dip, nmbuf);
578 
579 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
580 	ddi_remove_minor_node(zfs_dip, nmbuf);
581 
582 	avl_destroy(&zv->zv_znode.z_range_avl);
583 	mutex_destroy(&zv->zv_znode.z_range_lock);
584 
585 	kmem_free(zv, sizeof (zvol_state_t));
586 
587 	ddi_soft_state_free(zfsdev_state, minor);
588 
589 	zvol_minors--;
590 	return (0);
591 }
592 
593 int
594 zvol_remove_minor(const char *name)
595 {
596 	zvol_state_t *zv;
597 	int rc;
598 
599 	mutex_enter(&zfsdev_state_lock);
600 	if ((zv = zvol_minor_lookup(name)) == NULL) {
601 		mutex_exit(&zfsdev_state_lock);
602 		return (SET_ERROR(ENXIO));
603 	}
604 	rc = zvol_remove_zv(zv);
605 	mutex_exit(&zfsdev_state_lock);
606 	return (rc);
607 }
608 
609 int
610 zvol_first_open(zvol_state_t *zv)
611 {
612 	objset_t *os;
613 	uint64_t volsize;
614 	int error;
615 	uint64_t readonly;
616 
617 	/* lie and say we're read-only */
618 	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
619 	    zvol_tag, &os);
620 	if (error)
621 		return (error);
622 
623 	zv->zv_objset = os;
624 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
625 	if (error) {
626 		ASSERT(error == 0);
627 		dmu_objset_disown(os, zvol_tag);
628 		return (error);
629 	}
630 
631 	error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
632 	if (error) {
633 		dmu_objset_disown(os, zvol_tag);
634 		return (error);
635 	}
636 
637 	zvol_size_changed(zv, volsize);
638 	zv->zv_zilog = zil_open(os, zvol_get_data);
639 
640 	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
641 	    NULL) == 0);
642 	if (readonly || dmu_objset_is_snapshot(os) ||
643 	    !spa_writeable(dmu_objset_spa(os)))
644 		zv->zv_flags |= ZVOL_RDONLY;
645 	else
646 		zv->zv_flags &= ~ZVOL_RDONLY;
647 	return (error);
648 }
649 
650 void
651 zvol_last_close(zvol_state_t *zv)
652 {
653 	zil_close(zv->zv_zilog);
654 	zv->zv_zilog = NULL;
655 
656 	dmu_buf_rele(zv->zv_dbuf, zvol_tag);
657 	zv->zv_dbuf = NULL;
658 
659 	/*
660 	 * Evict cached data
661 	 */
662 	if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
663 	    !(zv->zv_flags & ZVOL_RDONLY))
664 		txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
665 	dmu_objset_evict_dbufs(zv->zv_objset);
666 
667 	dmu_objset_disown(zv->zv_objset, zvol_tag);
668 	zv->zv_objset = NULL;
669 }
670 
671 int
672 zvol_prealloc(zvol_state_t *zv)
673 {
674 	objset_t *os = zv->zv_objset;
675 	dmu_tx_t *tx;
676 	uint64_t refd, avail, usedobjs, availobjs;
677 	uint64_t resid = zv->zv_volsize;
678 	uint64_t off = 0;
679 
680 	/* Check the space usage before attempting to allocate the space */
681 	dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
682 	if (avail < zv->zv_volsize)
683 		return (SET_ERROR(ENOSPC));
684 
685 	/* Free old extents if they exist */
686 	zvol_free_extents(zv);
687 
688 	while (resid != 0) {
689 		int error;
690 		uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
691 
692 		tx = dmu_tx_create(os);
693 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
694 		error = dmu_tx_assign(tx, TXG_WAIT);
695 		if (error) {
696 			dmu_tx_abort(tx);
697 			(void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
698 			return (error);
699 		}
700 		dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
701 		dmu_tx_commit(tx);
702 		off += bytes;
703 		resid -= bytes;
704 	}
705 	txg_wait_synced(dmu_objset_pool(os), 0);
706 
707 	return (0);
708 }
709 
710 static int
711 zvol_update_volsize(objset_t *os, uint64_t volsize)
712 {
713 	dmu_tx_t *tx;
714 	int error;
715 
716 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
717 
718 	tx = dmu_tx_create(os);
719 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
720 	dmu_tx_mark_netfree(tx);
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 		off += len;
1109 		resid -= len;
1110 	}
1111 }
1112 
1113 static int
1114 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1115     uint64_t size, boolean_t doread, boolean_t isdump)
1116 {
1117 	vdev_disk_t *dvd;
1118 	int c;
1119 	int numerrors = 0;
1120 
1121 	if (vd->vdev_ops == &vdev_mirror_ops ||
1122 	    vd->vdev_ops == &vdev_replacing_ops ||
1123 	    vd->vdev_ops == &vdev_spare_ops) {
1124 		for (c = 0; c < vd->vdev_children; c++) {
1125 			int err = zvol_dumpio_vdev(vd->vdev_child[c],
1126 			    addr, offset, origoffset, size, doread, isdump);
1127 			if (err != 0) {
1128 				numerrors++;
1129 			} else if (doread) {
1130 				break;
1131 			}
1132 		}
1133 	}
1134 
1135 	if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1136 		return (numerrors < vd->vdev_children ? 0 : EIO);
1137 
1138 	if (doread && !vdev_readable(vd))
1139 		return (SET_ERROR(EIO));
1140 	else if (!doread && !vdev_writeable(vd))
1141 		return (SET_ERROR(EIO));
1142 
1143 	if (vd->vdev_ops == &vdev_raidz_ops) {
1144 		return (vdev_raidz_physio(vd,
1145 		    addr, size, offset, origoffset, doread, isdump));
1146 	}
1147 
1148 	offset += VDEV_LABEL_START_SIZE;
1149 
1150 	if (ddi_in_panic() || isdump) {
1151 		ASSERT(!doread);
1152 		if (doread)
1153 			return (SET_ERROR(EIO));
1154 		dvd = vd->vdev_tsd;
1155 		ASSERT3P(dvd, !=, NULL);
1156 		return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1157 		    lbtodb(size)));
1158 	} else {
1159 		dvd = vd->vdev_tsd;
1160 		ASSERT3P(dvd, !=, NULL);
1161 		return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1162 		    offset, doread ? B_READ : B_WRITE));
1163 	}
1164 }
1165 
1166 static int
1167 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1168     boolean_t doread, boolean_t isdump)
1169 {
1170 	vdev_t *vd;
1171 	int error;
1172 	zvol_extent_t *ze;
1173 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
1174 
1175 	/* Must be sector aligned, and not stradle a block boundary. */
1176 	if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1177 	    P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1178 		return (SET_ERROR(EINVAL));
1179 	}
1180 	ASSERT(size <= zv->zv_volblocksize);
1181 
1182 	/* Locate the extent this belongs to */
1183 	ze = list_head(&zv->zv_extents);
1184 	while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1185 		offset -= ze->ze_nblks * zv->zv_volblocksize;
1186 		ze = list_next(&zv->zv_extents, ze);
1187 	}
1188 
1189 	if (ze == NULL)
1190 		return (SET_ERROR(EINVAL));
1191 
1192 	if (!ddi_in_panic())
1193 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1194 
1195 	vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1196 	offset += DVA_GET_OFFSET(&ze->ze_dva);
1197 	error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1198 	    size, doread, isdump);
1199 
1200 	if (!ddi_in_panic())
1201 		spa_config_exit(spa, SCL_STATE, FTAG);
1202 
1203 	return (error);
1204 }
1205 
1206 int
1207 zvol_strategy(buf_t *bp)
1208 {
1209 	zfs_soft_state_t *zs = NULL;
1210 	zvol_state_t *zv;
1211 	uint64_t off, volsize;
1212 	size_t resid;
1213 	char *addr;
1214 	objset_t *os;
1215 	rl_t *rl;
1216 	int error = 0;
1217 	boolean_t doread = bp->b_flags & B_READ;
1218 	boolean_t is_dumpified;
1219 	boolean_t sync;
1220 
1221 	if (getminor(bp->b_edev) == 0) {
1222 		error = SET_ERROR(EINVAL);
1223 	} else {
1224 		zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1225 		if (zs == NULL)
1226 			error = SET_ERROR(ENXIO);
1227 		else if (zs->zss_type != ZSST_ZVOL)
1228 			error = SET_ERROR(EINVAL);
1229 	}
1230 
1231 	if (error) {
1232 		bioerror(bp, error);
1233 		biodone(bp);
1234 		return (0);
1235 	}
1236 
1237 	zv = zs->zss_data;
1238 
1239 	if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1240 		bioerror(bp, EROFS);
1241 		biodone(bp);
1242 		return (0);
1243 	}
1244 
1245 	off = ldbtob(bp->b_blkno);
1246 	volsize = zv->zv_volsize;
1247 
1248 	os = zv->zv_objset;
1249 	ASSERT(os != NULL);
1250 
1251 	bp_mapin(bp);
1252 	addr = bp->b_un.b_addr;
1253 	resid = bp->b_bcount;
1254 
1255 	if (resid > 0 && (off < 0 || off >= volsize)) {
1256 		bioerror(bp, EIO);
1257 		biodone(bp);
1258 		return (0);
1259 	}
1260 
1261 	is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1262 	sync = ((!(bp->b_flags & B_ASYNC) &&
1263 	    !(zv->zv_flags & ZVOL_WCE)) ||
1264 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1265 	    !doread && !is_dumpified;
1266 
1267 	/*
1268 	 * There must be no buffer changes when doing a dmu_sync() because
1269 	 * we can't change the data whilst calculating the checksum.
1270 	 */
1271 	rl = zfs_range_lock(&zv->zv_znode, off, resid,
1272 	    doread ? RL_READER : RL_WRITER);
1273 
1274 	while (resid != 0 && off < volsize) {
1275 		size_t size = MIN(resid, zvol_maxphys);
1276 		if (is_dumpified) {
1277 			size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1278 			error = zvol_dumpio(zv, addr, off, size,
1279 			    doread, B_FALSE);
1280 		} else if (doread) {
1281 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1282 			    DMU_READ_PREFETCH);
1283 		} else {
1284 			dmu_tx_t *tx = dmu_tx_create(os);
1285 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1286 			error = dmu_tx_assign(tx, TXG_WAIT);
1287 			if (error) {
1288 				dmu_tx_abort(tx);
1289 			} else {
1290 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1291 				zvol_log_write(zv, tx, off, size, sync);
1292 				dmu_tx_commit(tx);
1293 			}
1294 		}
1295 		if (error) {
1296 			/* convert checksum errors into IO errors */
1297 			if (error == ECKSUM)
1298 				error = SET_ERROR(EIO);
1299 			break;
1300 		}
1301 		off += size;
1302 		addr += size;
1303 		resid -= size;
1304 	}
1305 	zfs_range_unlock(rl);
1306 
1307 	if ((bp->b_resid = resid) == bp->b_bcount)
1308 		bioerror(bp, off > volsize ? EINVAL : error);
1309 
1310 	if (sync)
1311 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1312 	biodone(bp);
1313 
1314 	return (0);
1315 }
1316 
1317 /*
1318  * Set the buffer count to the zvol maximum transfer.
1319  * Using our own routine instead of the default minphys()
1320  * means that for larger writes we write bigger buffers on X86
1321  * (128K instead of 56K) and flush the disk write cache less often
1322  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1323  * 56K on X86 and 128K on sparc).
1324  */
1325 void
1326 zvol_minphys(struct buf *bp)
1327 {
1328 	if (bp->b_bcount > zvol_maxphys)
1329 		bp->b_bcount = zvol_maxphys;
1330 }
1331 
1332 int
1333 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1334 {
1335 	minor_t minor = getminor(dev);
1336 	zvol_state_t *zv;
1337 	int error = 0;
1338 	uint64_t size;
1339 	uint64_t boff;
1340 	uint64_t resid;
1341 
1342 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1343 	if (zv == NULL)
1344 		return (SET_ERROR(ENXIO));
1345 
1346 	if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1347 		return (SET_ERROR(EINVAL));
1348 
1349 	boff = ldbtob(blkno);
1350 	resid = ldbtob(nblocks);
1351 
1352 	VERIFY3U(boff + resid, <=, zv->zv_volsize);
1353 
1354 	while (resid) {
1355 		size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1356 		error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1357 		if (error)
1358 			break;
1359 		boff += size;
1360 		addr += size;
1361 		resid -= size;
1362 	}
1363 
1364 	return (error);
1365 }
1366 
1367 /*ARGSUSED*/
1368 int
1369 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1370 {
1371 	minor_t minor = getminor(dev);
1372 	zvol_state_t *zv;
1373 	uint64_t volsize;
1374 	rl_t *rl;
1375 	int error = 0;
1376 
1377 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1378 	if (zv == NULL)
1379 		return (SET_ERROR(ENXIO));
1380 
1381 	volsize = zv->zv_volsize;
1382 	if (uio->uio_resid > 0 &&
1383 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1384 		return (SET_ERROR(EIO));
1385 
1386 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1387 		error = physio(zvol_strategy, NULL, dev, B_READ,
1388 		    zvol_minphys, uio);
1389 		return (error);
1390 	}
1391 
1392 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1393 	    RL_READER);
1394 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1395 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1396 
1397 		/* don't read past the end */
1398 		if (bytes > volsize - uio->uio_loffset)
1399 			bytes = volsize - uio->uio_loffset;
1400 
1401 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1402 		if (error) {
1403 			/* convert checksum errors into IO errors */
1404 			if (error == ECKSUM)
1405 				error = SET_ERROR(EIO);
1406 			break;
1407 		}
1408 	}
1409 	zfs_range_unlock(rl);
1410 	return (error);
1411 }
1412 
1413 /*ARGSUSED*/
1414 int
1415 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1416 {
1417 	minor_t minor = getminor(dev);
1418 	zvol_state_t *zv;
1419 	uint64_t volsize;
1420 	rl_t *rl;
1421 	int error = 0;
1422 	boolean_t sync;
1423 
1424 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1425 	if (zv == NULL)
1426 		return (SET_ERROR(ENXIO));
1427 
1428 	volsize = zv->zv_volsize;
1429 	if (uio->uio_resid > 0 &&
1430 	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1431 		return (SET_ERROR(EIO));
1432 
1433 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
1434 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
1435 		    zvol_minphys, uio);
1436 		return (error);
1437 	}
1438 
1439 	sync = !(zv->zv_flags & ZVOL_WCE) ||
1440 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1441 
1442 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1443 	    RL_WRITER);
1444 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1445 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1446 		uint64_t off = uio->uio_loffset;
1447 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1448 
1449 		if (bytes > volsize - off)	/* don't write past the end */
1450 			bytes = volsize - off;
1451 
1452 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1453 		error = dmu_tx_assign(tx, TXG_WAIT);
1454 		if (error) {
1455 			dmu_tx_abort(tx);
1456 			break;
1457 		}
1458 		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1459 		if (error == 0)
1460 			zvol_log_write(zv, tx, off, bytes, sync);
1461 		dmu_tx_commit(tx);
1462 
1463 		if (error)
1464 			break;
1465 	}
1466 	zfs_range_unlock(rl);
1467 	if (sync)
1468 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1469 	return (error);
1470 }
1471 
1472 int
1473 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1474 {
1475 	struct uuid uuid = EFI_RESERVED;
1476 	efi_gpe_t gpe = { 0 };
1477 	uint32_t crc;
1478 	dk_efi_t efi;
1479 	int length;
1480 	char *ptr;
1481 
1482 	if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1483 		return (SET_ERROR(EFAULT));
1484 	ptr = (char *)(uintptr_t)efi.dki_data_64;
1485 	length = efi.dki_length;
1486 	/*
1487 	 * Some clients may attempt to request a PMBR for the
1488 	 * zvol.  Currently this interface will return EINVAL to
1489 	 * such requests.  These requests could be supported by
1490 	 * adding a check for lba == 0 and consing up an appropriate
1491 	 * PMBR.
1492 	 */
1493 	if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1494 		return (SET_ERROR(EINVAL));
1495 
1496 	gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1497 	gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1498 	UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1499 
1500 	if (efi.dki_lba == 1) {
1501 		efi_gpt_t gpt = { 0 };
1502 
1503 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1504 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1505 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1506 		gpt.efi_gpt_MyLBA = LE_64(1ULL);
1507 		gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1508 		gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1509 		gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1510 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1511 		gpt.efi_gpt_SizeOfPartitionEntry =
1512 		    LE_32(sizeof (efi_gpe_t));
1513 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1514 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1515 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1516 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1517 		if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1518 		    flag))
1519 			return (SET_ERROR(EFAULT));
1520 		ptr += sizeof (gpt);
1521 		length -= sizeof (gpt);
1522 	}
1523 	if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1524 	    length), flag))
1525 		return (SET_ERROR(EFAULT));
1526 	return (0);
1527 }
1528 
1529 /*
1530  * BEGIN entry points to allow external callers access to the volume.
1531  */
1532 /*
1533  * Return the volume parameters needed for access from an external caller.
1534  * These values are invariant as long as the volume is held open.
1535  */
1536 int
1537 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1538     uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1539     void **rl_hdl, void **bonus_hdl)
1540 {
1541 	zvol_state_t *zv;
1542 
1543 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1544 	if (zv == NULL)
1545 		return (SET_ERROR(ENXIO));
1546 	if (zv->zv_flags & ZVOL_DUMPIFIED)
1547 		return (SET_ERROR(ENXIO));
1548 
1549 	ASSERT(blksize && max_xfer_len && minor_hdl &&
1550 	    objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1551 
1552 	*blksize = zv->zv_volblocksize;
1553 	*max_xfer_len = (uint64_t)zvol_maxphys;
1554 	*minor_hdl = zv;
1555 	*objset_hdl = zv->zv_objset;
1556 	*zil_hdl = zv->zv_zilog;
1557 	*rl_hdl = &zv->zv_znode;
1558 	*bonus_hdl = zv->zv_dbuf;
1559 	return (0);
1560 }
1561 
1562 /*
1563  * Return the current volume size to an external caller.
1564  * The size can change while the volume is open.
1565  */
1566 uint64_t
1567 zvol_get_volume_size(void *minor_hdl)
1568 {
1569 	zvol_state_t *zv = minor_hdl;
1570 
1571 	return (zv->zv_volsize);
1572 }
1573 
1574 /*
1575  * Return the current WCE setting to an external caller.
1576  * The WCE setting can change while the volume is open.
1577  */
1578 int
1579 zvol_get_volume_wce(void *minor_hdl)
1580 {
1581 	zvol_state_t *zv = minor_hdl;
1582 
1583 	return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1584 }
1585 
1586 /*
1587  * Entry point for external callers to zvol_log_write
1588  */
1589 void
1590 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1591     boolean_t sync)
1592 {
1593 	zvol_state_t *zv = minor_hdl;
1594 
1595 	zvol_log_write(zv, tx, off, resid, sync);
1596 }
1597 /*
1598  * END entry points to allow external callers access to the volume.
1599  */
1600 
1601 /*
1602  * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1603  */
1604 static void
1605 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1606     boolean_t sync)
1607 {
1608 	itx_t *itx;
1609 	lr_truncate_t *lr;
1610 	zilog_t *zilog = zv->zv_zilog;
1611 
1612 	if (zil_replaying(zilog, tx))
1613 		return;
1614 
1615 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1616 	lr = (lr_truncate_t *)&itx->itx_lr;
1617 	lr->lr_foid = ZVOL_OBJ;
1618 	lr->lr_offset = off;
1619 	lr->lr_length = len;
1620 
1621 	itx->itx_sync = sync;
1622 	zil_itx_assign(zilog, itx, tx);
1623 }
1624 
1625 /*
1626  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
1627  * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1628  */
1629 /*ARGSUSED*/
1630 int
1631 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1632 {
1633 	zvol_state_t *zv;
1634 	struct dk_callback *dkc;
1635 	int error = 0;
1636 	rl_t *rl;
1637 
1638 	mutex_enter(&zfsdev_state_lock);
1639 
1640 	zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1641 
1642 	if (zv == NULL) {
1643 		mutex_exit(&zfsdev_state_lock);
1644 		return (SET_ERROR(ENXIO));
1645 	}
1646 	ASSERT(zv->zv_total_opens > 0);
1647 
1648 	switch (cmd) {
1649 
1650 	case DKIOCINFO:
1651 	{
1652 		struct dk_cinfo dki;
1653 
1654 		bzero(&dki, sizeof (dki));
1655 		(void) strcpy(dki.dki_cname, "zvol");
1656 		(void) strcpy(dki.dki_dname, "zvol");
1657 		dki.dki_ctype = DKC_UNKNOWN;
1658 		dki.dki_unit = getminor(dev);
1659 		dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1660 		mutex_exit(&zfsdev_state_lock);
1661 		if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1662 			error = SET_ERROR(EFAULT);
1663 		return (error);
1664 	}
1665 
1666 	case DKIOCGMEDIAINFO:
1667 	{
1668 		struct dk_minfo dkm;
1669 
1670 		bzero(&dkm, sizeof (dkm));
1671 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
1672 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1673 		dkm.dki_media_type = DK_UNKNOWN;
1674 		mutex_exit(&zfsdev_state_lock);
1675 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1676 			error = SET_ERROR(EFAULT);
1677 		return (error);
1678 	}
1679 
1680 	case DKIOCGMEDIAINFOEXT:
1681 	{
1682 		struct dk_minfo_ext dkmext;
1683 
1684 		bzero(&dkmext, sizeof (dkmext));
1685 		dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1686 		dkmext.dki_pbsize = zv->zv_volblocksize;
1687 		dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1688 		dkmext.dki_media_type = DK_UNKNOWN;
1689 		mutex_exit(&zfsdev_state_lock);
1690 		if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1691 			error = SET_ERROR(EFAULT);
1692 		return (error);
1693 	}
1694 
1695 	case DKIOCGETEFI:
1696 	{
1697 		uint64_t vs = zv->zv_volsize;
1698 		uint8_t bs = zv->zv_min_bs;
1699 
1700 		mutex_exit(&zfsdev_state_lock);
1701 		error = zvol_getefi((void *)arg, flag, vs, bs);
1702 		return (error);
1703 	}
1704 
1705 	case DKIOCFLUSHWRITECACHE:
1706 		dkc = (struct dk_callback *)arg;
1707 		mutex_exit(&zfsdev_state_lock);
1708 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
1709 		if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1710 			(*dkc->dkc_callback)(dkc->dkc_cookie, error);
1711 			error = 0;
1712 		}
1713 		return (error);
1714 
1715 	case DKIOCGETWCE:
1716 	{
1717 		int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1718 		if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1719 		    flag))
1720 			error = SET_ERROR(EFAULT);
1721 		break;
1722 	}
1723 	case DKIOCSETWCE:
1724 	{
1725 		int wce;
1726 		if (ddi_copyin((void *)arg, &wce, sizeof (int),
1727 		    flag)) {
1728 			error = SET_ERROR(EFAULT);
1729 			break;
1730 		}
1731 		if (wce) {
1732 			zv->zv_flags |= ZVOL_WCE;
1733 			mutex_exit(&zfsdev_state_lock);
1734 		} else {
1735 			zv->zv_flags &= ~ZVOL_WCE;
1736 			mutex_exit(&zfsdev_state_lock);
1737 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1738 		}
1739 		return (0);
1740 	}
1741 
1742 	case DKIOCGGEOM:
1743 	case DKIOCGVTOC:
1744 		/*
1745 		 * commands using these (like prtvtoc) expect ENOTSUP
1746 		 * since we're emulating an EFI label
1747 		 */
1748 		error = SET_ERROR(ENOTSUP);
1749 		break;
1750 
1751 	case DKIOCDUMPINIT:
1752 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1753 		    RL_WRITER);
1754 		error = zvol_dumpify(zv);
1755 		zfs_range_unlock(rl);
1756 		break;
1757 
1758 	case DKIOCDUMPFINI:
1759 		if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1760 			break;
1761 		rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1762 		    RL_WRITER);
1763 		error = zvol_dump_fini(zv);
1764 		zfs_range_unlock(rl);
1765 		break;
1766 
1767 	case DKIOCFREE:
1768 	{
1769 		dkioc_free_t df;
1770 		dmu_tx_t *tx;
1771 
1772 		if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1773 			error = SET_ERROR(EFAULT);
1774 			break;
1775 		}
1776 
1777 		/*
1778 		 * Apply Postel's Law to length-checking.  If they overshoot,
1779 		 * just blank out until the end, if there's a need to blank
1780 		 * out anything.
1781 		 */
1782 		if (df.df_start >= zv->zv_volsize)
1783 			break;	/* No need to do anything... */
1784 		if (df.df_start + df.df_length > zv->zv_volsize)
1785 			df.df_length = DMU_OBJECT_END;
1786 
1787 		rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1788 		    RL_WRITER);
1789 		tx = dmu_tx_create(zv->zv_objset);
1790 		dmu_tx_mark_netfree(tx);
1791 		error = dmu_tx_assign(tx, TXG_WAIT);
1792 		if (error != 0) {
1793 			dmu_tx_abort(tx);
1794 		} else {
1795 			zvol_log_truncate(zv, tx, df.df_start,
1796 			    df.df_length, B_TRUE);
1797 			dmu_tx_commit(tx);
1798 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1799 			    df.df_start, df.df_length);
1800 		}
1801 
1802 		zfs_range_unlock(rl);
1803 
1804 		if (error == 0) {
1805 			/*
1806 			 * If the write-cache is disabled or 'sync' property
1807 			 * is set to 'always' then treat this as a synchronous
1808 			 * operation (i.e. commit to zil).
1809 			 */
1810 			if (!(zv->zv_flags & ZVOL_WCE) ||
1811 			    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
1812 				zil_commit(zv->zv_zilog, ZVOL_OBJ);
1813 
1814 			/*
1815 			 * If the caller really wants synchronous writes, and
1816 			 * can't wait for them, don't return until the write
1817 			 * is done.
1818 			 */
1819 			if (df.df_flags & DF_WAIT_SYNC) {
1820 				txg_wait_synced(
1821 				    dmu_objset_pool(zv->zv_objset), 0);
1822 			}
1823 		}
1824 		break;
1825 	}
1826 
1827 	default:
1828 		error = SET_ERROR(ENOTTY);
1829 		break;
1830 
1831 	}
1832 	mutex_exit(&zfsdev_state_lock);
1833 	return (error);
1834 }
1835 
1836 int
1837 zvol_busy(void)
1838 {
1839 	return (zvol_minors != 0);
1840 }
1841 
1842 void
1843 zvol_init(void)
1844 {
1845 	VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1846 	    1) == 0);
1847 	mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1848 }
1849 
1850 void
1851 zvol_fini(void)
1852 {
1853 	mutex_destroy(&zfsdev_state_lock);
1854 	ddi_soft_state_fini(&zfsdev_state);
1855 }
1856 
1857 /*ARGSUSED*/
1858 static int
1859 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1860 {
1861 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1862 
1863 	if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1864 		return (1);
1865 	return (0);
1866 }
1867 
1868 /*ARGSUSED*/
1869 static void
1870 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1871 {
1872 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1873 
1874 	spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1875 }
1876 
1877 static int
1878 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1879 {
1880 	dmu_tx_t *tx;
1881 	int error;
1882 	objset_t *os = zv->zv_objset;
1883 	spa_t *spa = dmu_objset_spa(os);
1884 	vdev_t *vd = spa->spa_root_vdev;
1885 	nvlist_t *nv = NULL;
1886 	uint64_t version = spa_version(spa);
1887 	enum zio_checksum checksum;
1888 
1889 	ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1890 	ASSERT(vd->vdev_ops == &vdev_root_ops);
1891 
1892 	error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1893 	    DMU_OBJECT_END);
1894 	/* wait for dmu_free_long_range to actually free the blocks */
1895 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1896 
1897 	/*
1898 	 * If the pool on which the dump device is being initialized has more
1899 	 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1900 	 * enabled.  If so, bump that feature's counter to indicate that the
1901 	 * feature is active. We also check the vdev type to handle the
1902 	 * following case:
1903 	 *   # zpool create test raidz disk1 disk2 disk3
1904 	 *   Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1905 	 *   the raidz vdev itself has 3 children.
1906 	 */
1907 	if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1908 		if (!spa_feature_is_enabled(spa,
1909 		    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1910 			return (SET_ERROR(ENOTSUP));
1911 		(void) dsl_sync_task(spa_name(spa),
1912 		    zfs_mvdev_dump_feature_check,
1913 		    zfs_mvdev_dump_activate_feature_sync, NULL,
1914 		    2, ZFS_SPACE_CHECK_RESERVED);
1915 	}
1916 
1917 	tx = dmu_tx_create(os);
1918 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1919 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1920 	error = dmu_tx_assign(tx, TXG_WAIT);
1921 	if (error) {
1922 		dmu_tx_abort(tx);
1923 		return (error);
1924 	}
1925 
1926 	/*
1927 	 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
1928 	 * function.  Otherwise, use the old default -- OFF.
1929 	 */
1930 	checksum = spa_feature_is_active(spa,
1931 	    SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
1932 	    ZIO_CHECKSUM_OFF;
1933 
1934 	/*
1935 	 * If we are resizing the dump device then we only need to
1936 	 * update the refreservation to match the newly updated
1937 	 * zvolsize. Otherwise, we save off the original state of the
1938 	 * zvol so that we can restore them if the zvol is ever undumpified.
1939 	 */
1940 	if (resize) {
1941 		error = zap_update(os, ZVOL_ZAP_OBJ,
1942 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1943 		    &zv->zv_volsize, tx);
1944 	} else {
1945 		uint64_t checksum, compress, refresrv, vbs, dedup;
1946 
1947 		error = dsl_prop_get_integer(zv->zv_name,
1948 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1949 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1950 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1951 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1952 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1953 		error = error ? error : dsl_prop_get_integer(zv->zv_name,
1954 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1955 		if (version >= SPA_VERSION_DEDUP) {
1956 			error = error ? error :
1957 			    dsl_prop_get_integer(zv->zv_name,
1958 			    zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1959 		}
1960 
1961 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1962 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1963 		    &compress, tx);
1964 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1965 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1966 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1967 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1968 		    &refresrv, tx);
1969 		error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1970 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1971 		    &vbs, tx);
1972 		error = error ? error : dmu_object_set_blocksize(
1973 		    os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1974 		if (version >= SPA_VERSION_DEDUP) {
1975 			error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1976 			    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1977 			    &dedup, tx);
1978 		}
1979 		if (error == 0)
1980 			zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1981 	}
1982 	dmu_tx_commit(tx);
1983 
1984 	/*
1985 	 * We only need update the zvol's property if we are initializing
1986 	 * the dump area for the first time.
1987 	 */
1988 	if (!resize) {
1989 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1990 		VERIFY(nvlist_add_uint64(nv,
1991 		    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1992 		VERIFY(nvlist_add_uint64(nv,
1993 		    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1994 		    ZIO_COMPRESS_OFF) == 0);
1995 		VERIFY(nvlist_add_uint64(nv,
1996 		    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1997 		    checksum) == 0);
1998 		if (version >= SPA_VERSION_DEDUP) {
1999 			VERIFY(nvlist_add_uint64(nv,
2000 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
2001 			    ZIO_CHECKSUM_OFF) == 0);
2002 		}
2003 
2004 		error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2005 		    nv, NULL);
2006 		nvlist_free(nv);
2007 
2008 		if (error)
2009 			return (error);
2010 	}
2011 
2012 	/* Allocate the space for the dump */
2013 	error = zvol_prealloc(zv);
2014 	return (error);
2015 }
2016 
2017 static int
2018 zvol_dumpify(zvol_state_t *zv)
2019 {
2020 	int error = 0;
2021 	uint64_t dumpsize = 0;
2022 	dmu_tx_t *tx;
2023 	objset_t *os = zv->zv_objset;
2024 
2025 	if (zv->zv_flags & ZVOL_RDONLY)
2026 		return (SET_ERROR(EROFS));
2027 
2028 	if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2029 	    8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2030 		boolean_t resize = (dumpsize > 0);
2031 
2032 		if ((error = zvol_dump_init(zv, resize)) != 0) {
2033 			(void) zvol_dump_fini(zv);
2034 			return (error);
2035 		}
2036 	}
2037 
2038 	/*
2039 	 * Build up our lba mapping.
2040 	 */
2041 	error = zvol_get_lbas(zv);
2042 	if (error) {
2043 		(void) zvol_dump_fini(zv);
2044 		return (error);
2045 	}
2046 
2047 	tx = dmu_tx_create(os);
2048 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2049 	error = dmu_tx_assign(tx, TXG_WAIT);
2050 	if (error) {
2051 		dmu_tx_abort(tx);
2052 		(void) zvol_dump_fini(zv);
2053 		return (error);
2054 	}
2055 
2056 	zv->zv_flags |= ZVOL_DUMPIFIED;
2057 	error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2058 	    &zv->zv_volsize, tx);
2059 	dmu_tx_commit(tx);
2060 
2061 	if (error) {
2062 		(void) zvol_dump_fini(zv);
2063 		return (error);
2064 	}
2065 
2066 	txg_wait_synced(dmu_objset_pool(os), 0);
2067 	return (0);
2068 }
2069 
2070 static int
2071 zvol_dump_fini(zvol_state_t *zv)
2072 {
2073 	dmu_tx_t *tx;
2074 	objset_t *os = zv->zv_objset;
2075 	nvlist_t *nv;
2076 	int error = 0;
2077 	uint64_t checksum, compress, refresrv, vbs, dedup;
2078 	uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2079 
2080 	/*
2081 	 * Attempt to restore the zvol back to its pre-dumpified state.
2082 	 * This is a best-effort attempt as it's possible that not all
2083 	 * of these properties were initialized during the dumpify process
2084 	 * (i.e. error during zvol_dump_init).
2085 	 */
2086 
2087 	tx = dmu_tx_create(os);
2088 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2089 	error = dmu_tx_assign(tx, TXG_WAIT);
2090 	if (error) {
2091 		dmu_tx_abort(tx);
2092 		return (error);
2093 	}
2094 	(void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2095 	dmu_tx_commit(tx);
2096 
2097 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2098 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2099 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2100 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2101 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2102 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2103 	(void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2104 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2105 
2106 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2107 	(void) nvlist_add_uint64(nv,
2108 	    zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2109 	(void) nvlist_add_uint64(nv,
2110 	    zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2111 	(void) nvlist_add_uint64(nv,
2112 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2113 	if (version >= SPA_VERSION_DEDUP &&
2114 	    zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2115 	    zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2116 		(void) nvlist_add_uint64(nv,
2117 		    zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2118 	}
2119 	(void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2120 	    nv, NULL);
2121 	nvlist_free(nv);
2122 
2123 	zvol_free_extents(zv);
2124 	zv->zv_flags &= ~ZVOL_DUMPIFIED;
2125 	(void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2126 	/* wait for dmu_free_long_range to actually free the blocks */
2127 	txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2128 	tx = dmu_tx_create(os);
2129 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2130 	error = dmu_tx_assign(tx, TXG_WAIT);
2131 	if (error) {
2132 		dmu_tx_abort(tx);
2133 		return (error);
2134 	}
2135 	if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2136 		zv->zv_volblocksize = vbs;
2137 	dmu_tx_commit(tx);
2138 
2139 	return (0);
2140 }
2141