xref: /titanic_52/usr/src/uts/common/fs/zfs/zvol.c (revision e1c679fa4b0ab8c4bcaa6263974ca0c46e5b027f)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * ZFS volume emulation driver.
30  *
31  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
32  * Volumes are accessed through the symbolic links named:
33  *
34  * /dev/zvol/dsk/<pool_name>/<dataset_name>
35  * /dev/zvol/rdsk/<pool_name>/<dataset_name>
36  *
37  * These links are created by the ZFS-specific devfsadm link generator.
38  * Volumes are persistent through reboot.  No user command needs to be
39  * run before opening and using a device.
40  */
41 
42 #include <sys/types.h>
43 #include <sys/param.h>
44 #include <sys/errno.h>
45 #include <sys/uio.h>
46 #include <sys/buf.h>
47 #include <sys/modctl.h>
48 #include <sys/open.h>
49 #include <sys/kmem.h>
50 #include <sys/conf.h>
51 #include <sys/cmn_err.h>
52 #include <sys/stat.h>
53 #include <sys/zap.h>
54 #include <sys/spa.h>
55 #include <sys/zio.h>
56 #include <sys/dsl_prop.h>
57 #include <sys/dkio.h>
58 #include <sys/efi_partition.h>
59 #include <sys/byteorder.h>
60 #include <sys/pathname.h>
61 #include <sys/ddi.h>
62 #include <sys/sunddi.h>
63 #include <sys/crc32.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/fs/zfs.h>
67 #include <sys/zfs_ioctl.h>
68 #include <sys/mkdev.h>
69 #include <sys/zil.h>
70 #include <sys/refcount.h>
71 
72 #include "zfs_namecheck.h"
73 
74 #define	ZVOL_OBJ		1ULL
75 #define	ZVOL_ZAP_OBJ		2ULL
76 
77 static void *zvol_state;
78 
79 /*
80  * This lock protects the zvol_state structure from being modified
81  * while it's being used, e.g. an open that comes in before a create
82  * finishes.  It also protects temporary opens of the dataset so that,
83  * e.g., an open doesn't get a spurious EBUSY.
84  */
85 static kmutex_t zvol_state_lock;
86 static uint32_t zvol_minors;
87 
88 /*
89  * The in-core state of each volume.
90  */
91 typedef struct zvol_state {
92 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
93 	uint64_t	zv_volsize;	/* amount of space we advertise */
94 	uint64_t	zv_volblocksize; /* volume block size */
95 	minor_t		zv_minor;	/* minor number */
96 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
97 	uint8_t		zv_readonly;	/* hard readonly; like write-protect */
98 	objset_t	*zv_objset;	/* objset handle */
99 	uint32_t	zv_mode;	/* DS_MODE_* flags at open time */
100 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
101 	uint32_t	zv_total_opens;	/* total open count */
102 	zilog_t		*zv_zilog;	/* ZIL handle */
103 	uint64_t	zv_txg_assign;	/* txg to assign during ZIL replay */
104 	krwlock_t	zv_dslock;	/* dmu_sync() rwlock */
105 } zvol_state_t;
106 
107 /*
108  * zvol maximum transfer in one DMU tx.
109  */
110 int zvol_maxphys = DMU_MAX_ACCESS/2;
111 
112 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
113 
114 static void
115 zvol_size_changed(zvol_state_t *zv, dev_t dev)
116 {
117 	dev = makedevice(getmajor(dev), zv->zv_minor);
118 
119 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
120 	    "Size", zv->zv_volsize) == DDI_SUCCESS);
121 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
122 	    "Nblocks", lbtodb(zv->zv_volsize)) == DDI_SUCCESS);
123 }
124 
125 int
126 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
127 {
128 	if (volsize == 0)
129 		return (EINVAL);
130 
131 	if (volsize % blocksize != 0)
132 		return (EINVAL);
133 
134 #ifdef _ILP32
135 	if (volsize - 1 > SPEC_MAXOFFSET_T)
136 		return (EOVERFLOW);
137 #endif
138 	return (0);
139 }
140 
141 int
142 zvol_check_volblocksize(uint64_t volblocksize)
143 {
144 	if (volblocksize < SPA_MINBLOCKSIZE ||
145 	    volblocksize > SPA_MAXBLOCKSIZE ||
146 	    !ISP2(volblocksize))
147 		return (EDOM);
148 
149 	return (0);
150 }
151 
152 static void
153 zvol_readonly_changed_cb(void *arg, uint64_t newval)
154 {
155 	zvol_state_t *zv = arg;
156 
157 	zv->zv_readonly = (uint8_t)newval;
158 }
159 
160 int
161 zvol_get_stats(objset_t *os, nvlist_t *nv)
162 {
163 	int error;
164 	dmu_object_info_t doi;
165 	uint64_t val;
166 
167 
168 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
169 	if (error)
170 		return (error);
171 
172 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
173 
174 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
175 
176 	if (error == 0) {
177 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
178 		    doi.doi_data_block_size);
179 	}
180 
181 	return (error);
182 }
183 
184 /*
185  * Find a free minor number.
186  */
187 static minor_t
188 zvol_minor_alloc(void)
189 {
190 	minor_t minor;
191 
192 	ASSERT(MUTEX_HELD(&zvol_state_lock));
193 
194 	for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++)
195 		if (ddi_get_soft_state(zvol_state, minor) == NULL)
196 			return (minor);
197 
198 	return (0);
199 }
200 
201 static zvol_state_t *
202 zvol_minor_lookup(const char *name)
203 {
204 	minor_t minor;
205 	zvol_state_t *zv;
206 
207 	ASSERT(MUTEX_HELD(&zvol_state_lock));
208 
209 	for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) {
210 		zv = ddi_get_soft_state(zvol_state, minor);
211 		if (zv == NULL)
212 			continue;
213 		if (strcmp(zv->zv_name, name) == 0)
214 			break;
215 	}
216 
217 	return (zv);
218 }
219 
220 void
221 zvol_create_cb(objset_t *os, void *arg, dmu_tx_t *tx)
222 {
223 	zfs_create_data_t *zc = arg;
224 	int error;
225 	uint64_t volblocksize, volsize;
226 
227 	VERIFY(nvlist_lookup_uint64(zc->zc_props,
228 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
229 	if (nvlist_lookup_uint64(zc->zc_props,
230 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
231 		volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
232 
233 	/*
234 	 * These properites must be removed from the list so the generic
235 	 * property setting step won't apply to them.
236 	 */
237 	VERIFY(nvlist_remove_all(zc->zc_props,
238 	    zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
239 	(void) nvlist_remove_all(zc->zc_props,
240 	    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
241 
242 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
243 	    DMU_OT_NONE, 0, tx);
244 	ASSERT(error == 0);
245 
246 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
247 	    DMU_OT_NONE, 0, tx);
248 	ASSERT(error == 0);
249 
250 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
251 	ASSERT(error == 0);
252 }
253 
254 /*
255  * Replay a TX_WRITE ZIL transaction that didn't get committed
256  * after a system failure
257  */
258 static int
259 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
260 {
261 	objset_t *os = zv->zv_objset;
262 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
263 	uint64_t off = lr->lr_offset;
264 	uint64_t len = lr->lr_length;
265 	dmu_tx_t *tx;
266 	int error;
267 
268 	if (byteswap)
269 		byteswap_uint64_array(lr, sizeof (*lr));
270 
271 	tx = dmu_tx_create(os);
272 	dmu_tx_hold_write(tx, ZVOL_OBJ, off, len);
273 	error = dmu_tx_assign(tx, zv->zv_txg_assign);
274 	if (error) {
275 		dmu_tx_abort(tx);
276 	} else {
277 		dmu_write(os, ZVOL_OBJ, off, len, data, tx);
278 		dmu_tx_commit(tx);
279 	}
280 
281 	return (error);
282 }
283 
284 /* ARGSUSED */
285 static int
286 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
287 {
288 	return (ENOTSUP);
289 }
290 
291 /*
292  * Callback vectors for replaying records.
293  * Only TX_WRITE is needed for zvol.
294  */
295 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
296 	zvol_replay_err,	/* 0 no such transaction type */
297 	zvol_replay_err,	/* TX_CREATE */
298 	zvol_replay_err,	/* TX_MKDIR */
299 	zvol_replay_err,	/* TX_MKXATTR */
300 	zvol_replay_err,	/* TX_SYMLINK */
301 	zvol_replay_err,	/* TX_REMOVE */
302 	zvol_replay_err,	/* TX_RMDIR */
303 	zvol_replay_err,	/* TX_LINK */
304 	zvol_replay_err,	/* TX_RENAME */
305 	zvol_replay_write,	/* TX_WRITE */
306 	zvol_replay_err,	/* TX_TRUNCATE */
307 	zvol_replay_err,	/* TX_SETATTR */
308 	zvol_replay_err,	/* TX_ACL */
309 };
310 
311 /*
312  * Create a minor node for the specified volume.
313  */
314 int
315 zvol_create_minor(const char *name, dev_t dev)
316 {
317 	zvol_state_t *zv;
318 	objset_t *os;
319 	dmu_object_info_t doi;
320 	uint64_t volsize;
321 	minor_t minor = 0;
322 	struct pathname linkpath;
323 	int ds_mode = DS_MODE_PRIMARY;
324 	vnode_t *vp = NULL;
325 	char *devpath;
326 	size_t devpathlen = strlen(ZVOL_FULL_DEV_DIR) + 1 + strlen(name) + 1;
327 	char chrbuf[30], blkbuf[30];
328 	int error;
329 
330 	mutex_enter(&zvol_state_lock);
331 
332 	if ((zv = zvol_minor_lookup(name)) != NULL) {
333 		mutex_exit(&zvol_state_lock);
334 		return (EEXIST);
335 	}
336 
337 	if (strchr(name, '@') != 0)
338 		ds_mode |= DS_MODE_READONLY;
339 
340 	error = dmu_objset_open(name, DMU_OST_ZVOL, ds_mode, &os);
341 
342 	if (error) {
343 		mutex_exit(&zvol_state_lock);
344 		return (error);
345 	}
346 
347 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
348 
349 	if (error) {
350 		dmu_objset_close(os);
351 		mutex_exit(&zvol_state_lock);
352 		return (error);
353 	}
354 
355 	/*
356 	 * If there's an existing /dev/zvol symlink, try to use the
357 	 * same minor number we used last time.
358 	 */
359 	devpath = kmem_alloc(devpathlen, KM_SLEEP);
360 
361 	(void) sprintf(devpath, "%s/%s", ZVOL_FULL_DEV_DIR, name);
362 
363 	error = lookupname(devpath, UIO_SYSSPACE, NO_FOLLOW, NULL, &vp);
364 
365 	kmem_free(devpath, devpathlen);
366 
367 	if (error == 0 && vp->v_type != VLNK)
368 		error = EINVAL;
369 
370 	if (error == 0) {
371 		pn_alloc(&linkpath);
372 		error = pn_getsymlink(vp, &linkpath, kcred);
373 		if (error == 0) {
374 			char *ms = strstr(linkpath.pn_path, ZVOL_PSEUDO_DEV);
375 			if (ms != NULL) {
376 				ms += strlen(ZVOL_PSEUDO_DEV);
377 				minor = stoi(&ms);
378 			}
379 		}
380 		pn_free(&linkpath);
381 	}
382 
383 	if (vp != NULL)
384 		VN_RELE(vp);
385 
386 	/*
387 	 * If we found a minor but it's already in use, we must pick a new one.
388 	 */
389 	if (minor != 0 && ddi_get_soft_state(zvol_state, minor) != NULL)
390 		minor = 0;
391 
392 	if (minor == 0)
393 		minor = zvol_minor_alloc();
394 
395 	if (minor == 0) {
396 		dmu_objset_close(os);
397 		mutex_exit(&zvol_state_lock);
398 		return (ENXIO);
399 	}
400 
401 	if (ddi_soft_state_zalloc(zvol_state, minor) != DDI_SUCCESS) {
402 		dmu_objset_close(os);
403 		mutex_exit(&zvol_state_lock);
404 		return (EAGAIN);
405 	}
406 
407 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
408 	    (char *)name);
409 
410 	(void) sprintf(chrbuf, "%uc,raw", minor);
411 
412 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
413 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
414 		ddi_soft_state_free(zvol_state, minor);
415 		dmu_objset_close(os);
416 		mutex_exit(&zvol_state_lock);
417 		return (EAGAIN);
418 	}
419 
420 	(void) sprintf(blkbuf, "%uc", minor);
421 
422 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
423 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
424 		ddi_remove_minor_node(zfs_dip, chrbuf);
425 		ddi_soft_state_free(zvol_state, minor);
426 		dmu_objset_close(os);
427 		mutex_exit(&zvol_state_lock);
428 		return (EAGAIN);
429 	}
430 
431 	zv = ddi_get_soft_state(zvol_state, minor);
432 
433 	(void) strcpy(zv->zv_name, name);
434 	zv->zv_min_bs = DEV_BSHIFT;
435 	zv->zv_minor = minor;
436 	zv->zv_volsize = volsize;
437 	zv->zv_objset = os;
438 	zv->zv_mode = ds_mode;
439 	zv->zv_zilog = zil_open(os, zvol_get_data);
440 
441 	/* get and cache the blocksize */
442 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
443 	ASSERT(error == 0);
444 	zv->zv_volblocksize = doi.doi_data_block_size;
445 
446 	rw_init(&zv->zv_dslock, NULL, RW_DEFAULT, NULL);
447 
448 	zil_replay(os, zv, &zv->zv_txg_assign, zvol_replay_vector);
449 
450 	zvol_size_changed(zv, dev);
451 
452 	/* XXX this should handle the possible i/o error */
453 	VERIFY(dsl_prop_register(dmu_objset_ds(zv->zv_objset),
454 	    "readonly", zvol_readonly_changed_cb, zv) == 0);
455 
456 	zvol_minors++;
457 
458 	mutex_exit(&zvol_state_lock);
459 
460 	return (0);
461 }
462 
463 /*
464  * Remove minor node for the specified volume.
465  */
466 int
467 zvol_remove_minor(const char *name)
468 {
469 	zvol_state_t *zv;
470 	char namebuf[30];
471 
472 	mutex_enter(&zvol_state_lock);
473 
474 	if ((zv = zvol_minor_lookup(name)) == NULL) {
475 		mutex_exit(&zvol_state_lock);
476 		return (ENXIO);
477 	}
478 
479 	if (zv->zv_total_opens != 0) {
480 		mutex_exit(&zvol_state_lock);
481 		return (EBUSY);
482 	}
483 
484 	(void) sprintf(namebuf, "%uc,raw", zv->zv_minor);
485 	ddi_remove_minor_node(zfs_dip, namebuf);
486 
487 	(void) sprintf(namebuf, "%uc", zv->zv_minor);
488 	ddi_remove_minor_node(zfs_dip, namebuf);
489 
490 	VERIFY(dsl_prop_unregister(dmu_objset_ds(zv->zv_objset),
491 	    "readonly", zvol_readonly_changed_cb, zv) == 0);
492 
493 	zil_close(zv->zv_zilog);
494 	zv->zv_zilog = NULL;
495 	dmu_objset_close(zv->zv_objset);
496 	zv->zv_objset = NULL;
497 
498 	ddi_soft_state_free(zvol_state, zv->zv_minor);
499 
500 	zvol_minors--;
501 
502 	mutex_exit(&zvol_state_lock);
503 
504 	return (0);
505 }
506 
507 int
508 zvol_set_volsize(const char *name, dev_t dev, uint64_t volsize)
509 {
510 	zvol_state_t *zv;
511 	dmu_tx_t *tx;
512 	int error;
513 	dmu_object_info_t doi;
514 
515 	mutex_enter(&zvol_state_lock);
516 
517 	if ((zv = zvol_minor_lookup(name)) == NULL) {
518 		mutex_exit(&zvol_state_lock);
519 		return (ENXIO);
520 	}
521 
522 	if ((error = dmu_object_info(zv->zv_objset, ZVOL_OBJ, &doi)) != 0 ||
523 	    (error = zvol_check_volsize(volsize,
524 	    doi.doi_data_block_size)) != 0) {
525 		mutex_exit(&zvol_state_lock);
526 		return (error);
527 	}
528 
529 	if (zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY)) {
530 		mutex_exit(&zvol_state_lock);
531 		return (EROFS);
532 	}
533 
534 	tx = dmu_tx_create(zv->zv_objset);
535 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
536 	dmu_tx_hold_free(tx, ZVOL_OBJ, volsize, DMU_OBJECT_END);
537 	error = dmu_tx_assign(tx, TXG_WAIT);
538 	if (error) {
539 		dmu_tx_abort(tx);
540 		mutex_exit(&zvol_state_lock);
541 		return (error);
542 	}
543 
544 	error = zap_update(zv->zv_objset, ZVOL_ZAP_OBJ, "size", 8, 1,
545 	    &volsize, tx);
546 	if (error == 0) {
547 		error = dmu_free_range(zv->zv_objset, ZVOL_OBJ, volsize,
548 		    DMU_OBJECT_END, tx);
549 	}
550 
551 	dmu_tx_commit(tx);
552 
553 	if (error == 0) {
554 		zv->zv_volsize = volsize;
555 		zvol_size_changed(zv, dev);
556 	}
557 
558 	mutex_exit(&zvol_state_lock);
559 
560 	return (error);
561 }
562 
563 int
564 zvol_set_volblocksize(const char *name, uint64_t volblocksize)
565 {
566 	zvol_state_t *zv;
567 	dmu_tx_t *tx;
568 	int error;
569 
570 	mutex_enter(&zvol_state_lock);
571 
572 	if ((zv = zvol_minor_lookup(name)) == NULL) {
573 		mutex_exit(&zvol_state_lock);
574 		return (ENXIO);
575 	}
576 
577 	if (zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY)) {
578 		mutex_exit(&zvol_state_lock);
579 		return (EROFS);
580 	}
581 
582 	tx = dmu_tx_create(zv->zv_objset);
583 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
584 	error = dmu_tx_assign(tx, TXG_WAIT);
585 	if (error) {
586 		dmu_tx_abort(tx);
587 	} else {
588 		error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
589 		    volblocksize, 0, tx);
590 		if (error == ENOTSUP)
591 			error = EBUSY;
592 		dmu_tx_commit(tx);
593 	}
594 
595 	mutex_exit(&zvol_state_lock);
596 
597 	return (error);
598 }
599 
600 /*ARGSUSED*/
601 int
602 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
603 {
604 	minor_t minor = getminor(*devp);
605 	zvol_state_t *zv;
606 
607 	if (minor == 0)			/* This is the control device */
608 		return (0);
609 
610 	mutex_enter(&zvol_state_lock);
611 
612 	zv = ddi_get_soft_state(zvol_state, minor);
613 	if (zv == NULL) {
614 		mutex_exit(&zvol_state_lock);
615 		return (ENXIO);
616 	}
617 
618 	ASSERT(zv->zv_objset != NULL);
619 
620 	if ((flag & FWRITE) &&
621 	    (zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY))) {
622 		mutex_exit(&zvol_state_lock);
623 		return (EROFS);
624 	}
625 
626 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
627 		zv->zv_open_count[otyp]++;
628 		zv->zv_total_opens++;
629 	}
630 
631 	mutex_exit(&zvol_state_lock);
632 
633 	return (0);
634 }
635 
636 /*ARGSUSED*/
637 int
638 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
639 {
640 	minor_t minor = getminor(dev);
641 	zvol_state_t *zv;
642 
643 	if (minor == 0)		/* This is the control device */
644 		return (0);
645 
646 	mutex_enter(&zvol_state_lock);
647 
648 	zv = ddi_get_soft_state(zvol_state, minor);
649 	if (zv == NULL) {
650 		mutex_exit(&zvol_state_lock);
651 		return (ENXIO);
652 	}
653 
654 	/*
655 	 * The next statement is a workaround for the following DDI bug:
656 	 * 6343604 specfs race: multiple "last-close" of the same device
657 	 */
658 	if (zv->zv_total_opens == 0) {
659 		mutex_exit(&zvol_state_lock);
660 		return (0);
661 	}
662 
663 	/*
664 	 * If the open count is zero, this is a spurious close.
665 	 * That indicates a bug in the kernel / DDI framework.
666 	 */
667 	ASSERT(zv->zv_open_count[otyp] != 0);
668 	ASSERT(zv->zv_total_opens != 0);
669 
670 	/*
671 	 * You may get multiple opens, but only one close.
672 	 */
673 	zv->zv_open_count[otyp]--;
674 	zv->zv_total_opens--;
675 
676 	mutex_exit(&zvol_state_lock);
677 
678 	return (0);
679 }
680 
681 static void
682 zvol_get_done(dmu_buf_t *db, void *vzgd)
683 {
684 	zgd_t *zgd = (zgd_t *)vzgd;
685 
686 	dmu_buf_rele(db, vzgd);
687 	zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp)));
688 	kmem_free(zgd, sizeof (zgd_t));
689 }
690 
691 /*
692  * Get data to generate a TX_WRITE intent log record.
693  */
694 static int
695 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
696 {
697 	zvol_state_t *zv = arg;
698 	objset_t *os = zv->zv_objset;
699 	dmu_buf_t *db;
700 	zgd_t *zgd;
701 	int error;
702 
703 	ASSERT(zio);
704 	ASSERT(lr->lr_length != 0);
705 
706 	if (buf != NULL)
707 		return (dmu_read(os, ZVOL_OBJ,
708 		    lr->lr_offset, lr->lr_length, buf));
709 
710 	zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
711 	zgd->zgd_zilog = zv->zv_zilog;
712 	zgd->zgd_bp = &lr->lr_blkptr;
713 
714 	VERIFY(0 == dmu_buf_hold(os, ZVOL_OBJ, lr->lr_offset, zgd, &db));
715 	/*
716 	 * Have to lock to ensure when when the data is
717 	 * written out and it's checksum is being calculated
718 	 * that no one can change the data.
719 	 */
720 	rw_enter(&zv->zv_dslock, RW_READER);
721 	error = dmu_sync(zio, db, &lr->lr_blkptr,
722 	    lr->lr_common.lrc_txg, zvol_get_done, zgd);
723 	rw_exit(&zv->zv_dslock);
724 	if (error == 0)
725 		zil_add_vdev(zv->zv_zilog,
726 		    DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr)));
727 	/*
728 	 * If we get EINPROGRESS, then we need to wait for a
729 	 * write IO initiated by dmu_sync() to complete before
730 	 * we can release this dbuf.  We will finish everything
731 	 * up in the zvol_get_done() callback.
732 	 */
733 	if (error == EINPROGRESS)
734 		return (0);
735 	dmu_buf_rele(db, zgd);
736 	kmem_free(zgd, sizeof (zgd_t));
737 	return (error);
738 }
739 
740 /*
741  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
742  *
743  * We store data in the log buffers if it's small enough.
744  * Otherwise we will later flush the data out via dmu_sync().
745  */
746 ssize_t zvol_immediate_write_sz = 32768;
747 
748 static void
749 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t len)
750 {
751 	uint32_t blocksize = zv->zv_volblocksize;
752 	lr_write_t *lr;
753 
754 	while (len) {
755 		ssize_t nbytes = MIN(len, blocksize - P2PHASE(off, blocksize));
756 		itx_t *itx = zil_itx_create(TX_WRITE, sizeof (*lr));
757 
758 		itx->itx_wr_state =
759 		    len > zvol_immediate_write_sz ?  WR_INDIRECT : WR_NEED_COPY;
760 		itx->itx_private = zv;
761 		lr = (lr_write_t *)&itx->itx_lr;
762 		lr->lr_foid = ZVOL_OBJ;
763 		lr->lr_offset = off;
764 		lr->lr_length = nbytes;
765 		lr->lr_blkoff = off - P2ALIGN_TYPED(off, blocksize, uint64_t);
766 		BP_ZERO(&lr->lr_blkptr);
767 
768 		(void) zil_itx_assign(zv->zv_zilog, itx, tx);
769 		len -= nbytes;
770 		off += nbytes;
771 	}
772 }
773 
774 int
775 zvol_strategy(buf_t *bp)
776 {
777 	zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
778 	uint64_t off, volsize;
779 	size_t size, resid;
780 	char *addr;
781 	objset_t *os;
782 	int error = 0;
783 	int sync;
784 	int reading;
785 
786 	if (zv == NULL) {
787 		bioerror(bp, ENXIO);
788 		biodone(bp);
789 		return (0);
790 	}
791 
792 	if (getminor(bp->b_edev) == 0) {
793 		bioerror(bp, EINVAL);
794 		biodone(bp);
795 		return (0);
796 	}
797 
798 	if ((zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY)) &&
799 	    !(bp->b_flags & B_READ)) {
800 		bioerror(bp, EROFS);
801 		biodone(bp);
802 		return (0);
803 	}
804 
805 	off = ldbtob(bp->b_blkno);
806 	volsize = zv->zv_volsize;
807 
808 	os = zv->zv_objset;
809 	ASSERT(os != NULL);
810 	sync = !(bp->b_flags & B_ASYNC) && !(zil_disable);
811 
812 	bp_mapin(bp);
813 	addr = bp->b_un.b_addr;
814 	resid = bp->b_bcount;
815 
816 	/*
817 	 * There must be no buffer changes when doing a dmu_sync() because
818 	 * we can't change the data whilst calculating the checksum.
819 	 * A better approach than a per zvol rwlock would be to lock ranges.
820 	 */
821 	reading = bp->b_flags & B_READ;
822 	if (reading || resid <= zvol_immediate_write_sz)
823 		rw_enter(&zv->zv_dslock, RW_READER);
824 	else
825 		rw_enter(&zv->zv_dslock, RW_WRITER);
826 
827 	while (resid != 0 && off < volsize) {
828 
829 		size = MIN(resid, zvol_maxphys); /* zvol_maxphys per tx */
830 
831 		if (size > volsize - off)	/* don't write past the end */
832 			size = volsize - off;
833 
834 		if (reading) {
835 			error = dmu_read(os, ZVOL_OBJ, off, size, addr);
836 		} else {
837 			dmu_tx_t *tx = dmu_tx_create(os);
838 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
839 			error = dmu_tx_assign(tx, TXG_WAIT);
840 			if (error) {
841 				dmu_tx_abort(tx);
842 			} else {
843 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
844 				zvol_log_write(zv, tx, off, size);
845 				dmu_tx_commit(tx);
846 			}
847 		}
848 		if (error)
849 			break;
850 		off += size;
851 		addr += size;
852 		resid -= size;
853 	}
854 	rw_exit(&zv->zv_dslock);
855 
856 	if ((bp->b_resid = resid) == bp->b_bcount)
857 		bioerror(bp, off > volsize ? EINVAL : error);
858 
859 	if (sync)
860 		zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
861 
862 	biodone(bp);
863 
864 	return (0);
865 }
866 
867 /*
868  * Set the buffer count to the zvol maximum transfer.
869  * Using our own routine instead of the default minphys()
870  * means that for larger writes we write bigger buffers on X86
871  * (128K instead of 56K) and flush the disk write cache less often
872  * (every zvol_maxphys - currently 1MB) instead of minphys (currently
873  * 56K on X86 and 128K on sparc).
874  */
875 void
876 zvol_minphys(struct buf *bp)
877 {
878 	if (bp->b_bcount > zvol_maxphys)
879 		bp->b_bcount = zvol_maxphys;
880 }
881 
882 /*ARGSUSED*/
883 int
884 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
885 {
886 	zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(dev));
887 	int error = 0;
888 
889 	while (uio->uio_resid > 0) {
890 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
891 
892 		error =  dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
893 		if (error)
894 			break;
895 	}
896 	return (error);
897 }
898 
899 /*ARGSUSED*/
900 int
901 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
902 {
903 	zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(dev));
904 	int error = 0;
905 
906 	while (uio->uio_resid > 0) {
907 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
908 		uint64_t off = uio->uio_loffset;
909 
910 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
911 		dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
912 		error = dmu_tx_assign(tx, TXG_WAIT);
913 		if (error) {
914 			dmu_tx_abort(tx);
915 			break;
916 		}
917 		error = dmu_write_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes, tx);
918 		if (error == 0)
919 			zvol_log_write(zv, tx, off, bytes);
920 		dmu_tx_commit(tx);
921 
922 		if (error)
923 			break;
924 	}
925 	return (error);
926 }
927 
928 /*
929  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
930  */
931 /*ARGSUSED*/
932 int
933 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
934 {
935 	zvol_state_t *zv;
936 	struct dk_cinfo dkc;
937 	struct dk_minfo dkm;
938 	dk_efi_t efi;
939 	struct uuid uuid = EFI_RESERVED;
940 	uint32_t crc;
941 	int error = 0;
942 
943 	mutex_enter(&zvol_state_lock);
944 
945 	zv = ddi_get_soft_state(zvol_state, getminor(dev));
946 
947 	if (zv == NULL) {
948 		mutex_exit(&zvol_state_lock);
949 		return (ENXIO);
950 	}
951 
952 	switch (cmd) {
953 
954 	case DKIOCINFO:
955 		bzero(&dkc, sizeof (dkc));
956 		(void) strcpy(dkc.dki_cname, "zvol");
957 		(void) strcpy(dkc.dki_dname, "zvol");
958 		dkc.dki_ctype = DKC_UNKNOWN;
959 		dkc.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
960 		mutex_exit(&zvol_state_lock);
961 		if (ddi_copyout(&dkc, (void *)arg, sizeof (dkc), flag))
962 			error = EFAULT;
963 		return (error);
964 
965 	case DKIOCGMEDIAINFO:
966 		bzero(&dkm, sizeof (dkm));
967 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
968 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
969 		dkm.dki_media_type = DK_UNKNOWN;
970 		mutex_exit(&zvol_state_lock);
971 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
972 			error = EFAULT;
973 		return (error);
974 
975 	case DKIOCGETEFI:
976 		if (ddi_copyin((void *)arg, &efi, sizeof (dk_efi_t), flag)) {
977 			mutex_exit(&zvol_state_lock);
978 			return (EFAULT);
979 		}
980 		efi.dki_data = (void *)(uintptr_t)efi.dki_data_64;
981 
982 		/*
983 		 * Some clients may attempt to request a PMBR for the
984 		 * zvol.  Currently this interface will return ENOTTY to
985 		 * such requests.  These requests could be supported by
986 		 * adding a check for lba == 0 and consing up an appropriate
987 		 * RMBR.
988 		 */
989 		if (efi.dki_lba == 1) {
990 			efi_gpt_t gpt;
991 			efi_gpe_t gpe;
992 
993 			bzero(&gpt, sizeof (gpt));
994 			bzero(&gpe, sizeof (gpe));
995 
996 			if (efi.dki_length < sizeof (gpt)) {
997 				mutex_exit(&zvol_state_lock);
998 				return (EINVAL);
999 			}
1000 
1001 			gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1002 			gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1003 			gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1004 			gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1005 			gpt.efi_gpt_LastUsableLBA =
1006 			    LE_64((zv->zv_volsize >> zv->zv_min_bs) - 1);
1007 			gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1008 			gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1009 			gpt.efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (gpe));
1010 
1011 			UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1012 			gpe.efi_gpe_StartingLBA = gpt.efi_gpt_FirstUsableLBA;
1013 			gpe.efi_gpe_EndingLBA = gpt.efi_gpt_LastUsableLBA;
1014 
1015 			CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1016 			gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1017 
1018 			CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1019 			gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1020 
1021 			mutex_exit(&zvol_state_lock);
1022 			if (ddi_copyout(&gpt, efi.dki_data, sizeof (gpt), flag))
1023 				error = EFAULT;
1024 		} else if (efi.dki_lba == 2) {
1025 			efi_gpe_t gpe;
1026 
1027 			bzero(&gpe, sizeof (gpe));
1028 
1029 			if (efi.dki_length < sizeof (gpe)) {
1030 				mutex_exit(&zvol_state_lock);
1031 				return (EINVAL);
1032 			}
1033 
1034 			UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1035 			gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1036 			gpe.efi_gpe_EndingLBA =
1037 			    LE_64((zv->zv_volsize >> zv->zv_min_bs) - 1);
1038 
1039 			mutex_exit(&zvol_state_lock);
1040 			if (ddi_copyout(&gpe, efi.dki_data, sizeof (gpe), flag))
1041 				error = EFAULT;
1042 		} else {
1043 			mutex_exit(&zvol_state_lock);
1044 			error = EINVAL;
1045 		}
1046 		return (error);
1047 
1048 	case DKIOCFLUSHWRITECACHE:
1049 		zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
1050 		break;
1051 
1052 	case DKIOCGGEOM:
1053 	case DKIOCGVTOC:
1054 		/* commands using these (like prtvtoc) expect ENOTSUP */
1055 		error = ENOTSUP;
1056 		break;
1057 
1058 	default:
1059 		error = ENOTTY;
1060 		break;
1061 
1062 	}
1063 	mutex_exit(&zvol_state_lock);
1064 	return (error);
1065 }
1066 
1067 int
1068 zvol_busy(void)
1069 {
1070 	return (zvol_minors != 0);
1071 }
1072 
1073 void
1074 zvol_init(void)
1075 {
1076 	VERIFY(ddi_soft_state_init(&zvol_state, sizeof (zvol_state_t), 1) == 0);
1077 	mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
1078 }
1079 
1080 void
1081 zvol_fini(void)
1082 {
1083 	mutex_destroy(&zvol_state_lock);
1084 	ddi_soft_state_fini(&zvol_state);
1085 }
1086