xref: /titanic_44/usr/src/uts/common/fs/zfs/zvol.c (revision 5c46c6a053defc1bf25f2e08df2653bc7717ed0b)
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 2006 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/aio_req.h>
46 #include <sys/uio.h>
47 #include <sys/buf.h>
48 #include <sys/modctl.h>
49 #include <sys/open.h>
50 #include <sys/kmem.h>
51 #include <sys/conf.h>
52 #include <sys/cmn_err.h>
53 #include <sys/stat.h>
54 #include <sys/zap.h>
55 #include <sys/spa.h>
56 #include <sys/zio.h>
57 #include <sys/dsl_prop.h>
58 #include <sys/dkio.h>
59 #include <sys/efi_partition.h>
60 #include <sys/byteorder.h>
61 #include <sys/pathname.h>
62 #include <sys/ddi.h>
63 #include <sys/sunddi.h>
64 #include <sys/crc32.h>
65 #include <sys/dirent.h>
66 #include <sys/policy.h>
67 #include <sys/fs/zfs.h>
68 #include <sys/zfs_ioctl.h>
69 #include <sys/mkdev.h>
70 #include <sys/zil.h>
71 
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 	minor_t		zv_minor;	/* minor number */
95 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
96 	uint8_t		zv_readonly;	/* hard readonly; like write-protect */
97 	objset_t	*zv_objset;	/* objset handle */
98 	uint32_t	zv_mode;	/* DS_MODE_* flags at open time */
99 	uint32_t	zv_open_count[OTYPCNT];	/* open counts */
100 	uint32_t	zv_total_opens;	/* total open count */
101 	zilog_t		*zv_zilog;	/* ZIL handle */
102 	uint64_t	zv_txg_assign;	/* txg to assign during ZIL replay */
103 	krwlock_t	zv_dslock;	/* dmu_sync() rwlock */
104 } zvol_state_t;
105 
106 static void
107 zvol_size_changed(zvol_state_t *zv, dev_t dev)
108 {
109 	dev = makedevice(getmajor(dev), zv->zv_minor);
110 
111 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
112 	    "Size", zv->zv_volsize) == DDI_SUCCESS);
113 	VERIFY(ddi_prop_update_int64(dev, zfs_dip,
114 	    "Nblocks", lbtodb(zv->zv_volsize)) == DDI_SUCCESS);
115 }
116 
117 int
118 zvol_check_volsize(zfs_cmd_t *zc, uint64_t blocksize)
119 {
120 	if (zc->zc_volsize == 0)
121 		return (EINVAL);
122 
123 	if (zc->zc_volsize % blocksize != 0)
124 		return (EINVAL);
125 
126 #ifdef _ILP32
127 	if (zc->zc_volsize - 1 > SPEC_MAXOFFSET_T)
128 		return (EOVERFLOW);
129 #endif
130 	return (0);
131 }
132 
133 int
134 zvol_check_volblocksize(zfs_cmd_t *zc)
135 {
136 	if (zc->zc_volblocksize < SPA_MINBLOCKSIZE ||
137 	    zc->zc_volblocksize > SPA_MAXBLOCKSIZE ||
138 	    !ISP2(zc->zc_volblocksize))
139 		return (EDOM);
140 
141 	return (0);
142 }
143 
144 static void
145 zvol_readonly_changed_cb(void *arg, uint64_t newval)
146 {
147 	zvol_state_t *zv = arg;
148 
149 	zv->zv_readonly = (uint8_t)newval;
150 }
151 
152 int
153 zvol_get_stats(zfs_cmd_t *zc, objset_t *os)
154 {
155 	int error;
156 	dmu_object_info_t doi;
157 
158 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &zc->zc_volsize);
159 
160 	if (error)
161 		return (error);
162 
163 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
164 
165 	if (error == 0)
166 		zc->zc_volblocksize = doi.doi_data_block_size;
167 
168 	return (error);
169 }
170 
171 /*
172  * Find a free minor number.
173  */
174 static minor_t
175 zvol_minor_alloc(void)
176 {
177 	minor_t minor;
178 
179 	ASSERT(MUTEX_HELD(&zvol_state_lock));
180 
181 	for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++)
182 		if (ddi_get_soft_state(zvol_state, minor) == NULL)
183 			return (minor);
184 
185 	return (0);
186 }
187 
188 static zvol_state_t *
189 zvol_minor_lookup(char *name)
190 {
191 	minor_t minor;
192 	zvol_state_t *zv;
193 
194 	ASSERT(MUTEX_HELD(&zvol_state_lock));
195 
196 	for (minor = 1; minor <= ZVOL_MAX_MINOR; minor++) {
197 		zv = ddi_get_soft_state(zvol_state, minor);
198 		if (zv == NULL)
199 			continue;
200 		if (strcmp(zv->zv_name, name) == 0)
201 			break;
202 	}
203 
204 	return (zv);
205 }
206 
207 void
208 zvol_create_cb(objset_t *os, void *arg, dmu_tx_t *tx)
209 {
210 	zfs_cmd_t *zc = arg;
211 	int error;
212 
213 	error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, zc->zc_volblocksize,
214 	    DMU_OT_NONE, 0, tx);
215 	ASSERT(error == 0);
216 
217 	error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
218 	    DMU_OT_NONE, 0, tx);
219 	ASSERT(error == 0);
220 
221 	error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &zc->zc_volsize, tx);
222 	ASSERT(error == 0);
223 }
224 
225 /*
226  * Replay a TX_WRITE ZIL transaction that didn't get committed
227  * after a system failure
228  */
229 static int
230 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
231 {
232 	objset_t *os = zv->zv_objset;
233 	char *data = (char *)(lr + 1);	/* data follows lr_write_t */
234 	uint64_t off = lr->lr_offset;
235 	uint64_t len = lr->lr_length;
236 	dmu_tx_t *tx;
237 	int error;
238 
239 	if (byteswap)
240 		byteswap_uint64_array(lr, sizeof (*lr));
241 
242 	tx = dmu_tx_create(os);
243 	dmu_tx_hold_write(tx, ZVOL_OBJ, off, len);
244 	error = dmu_tx_assign(tx, zv->zv_txg_assign);
245 	if (error) {
246 		dmu_tx_abort(tx);
247 	} else {
248 		dmu_write(os, ZVOL_OBJ, off, len, data, tx);
249 		dmu_tx_commit(tx);
250 	}
251 
252 	return (error);
253 }
254 
255 /* ARGSUSED */
256 static int
257 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
258 {
259 	return (ENOTSUP);
260 }
261 
262 /*
263  * Callback vectors for replaying records.
264  * Only TX_WRITE is needed for zvol.
265  */
266 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
267 	zvol_replay_err,	/* 0 no such transaction type */
268 	zvol_replay_err,	/* TX_CREATE */
269 	zvol_replay_err,	/* TX_MKDIR */
270 	zvol_replay_err,	/* TX_MKXATTR */
271 	zvol_replay_err,	/* TX_SYMLINK */
272 	zvol_replay_err,	/* TX_REMOVE */
273 	zvol_replay_err,	/* TX_RMDIR */
274 	zvol_replay_err,	/* TX_LINK */
275 	zvol_replay_err,	/* TX_RENAME */
276 	zvol_replay_write,	/* TX_WRITE */
277 	zvol_replay_err,	/* TX_TRUNCATE */
278 	zvol_replay_err,	/* TX_SETATTR */
279 	zvol_replay_err,	/* TX_ACL */
280 };
281 
282 /*
283  * Create a minor node for the specified volume.
284  */
285 int
286 zvol_create_minor(zfs_cmd_t *zc)
287 {
288 	char *name = zc->zc_name;
289 	dev_t dev = zc->zc_dev;
290 	zvol_state_t *zv;
291 	objset_t *os;
292 	uint64_t volsize;
293 	minor_t minor = 0;
294 	struct pathname linkpath;
295 	int ds_mode = DS_MODE_PRIMARY;
296 	vnode_t *vp = NULL;
297 	char *devpath;
298 	size_t devpathlen = strlen(ZVOL_FULL_DEV_DIR) + 1 + strlen(name) + 1;
299 	char chrbuf[30], blkbuf[30];
300 	int error;
301 
302 	mutex_enter(&zvol_state_lock);
303 
304 	if ((zv = zvol_minor_lookup(name)) != NULL) {
305 		mutex_exit(&zvol_state_lock);
306 		return (EEXIST);
307 	}
308 
309 	if (strchr(name, '@') != 0)
310 		ds_mode |= DS_MODE_READONLY;
311 
312 	error = dmu_objset_open(name, DMU_OST_ZVOL, ds_mode, &os);
313 
314 	if (error) {
315 		mutex_exit(&zvol_state_lock);
316 		return (error);
317 	}
318 
319 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
320 
321 	if (error) {
322 		dmu_objset_close(os);
323 		mutex_exit(&zvol_state_lock);
324 		return (error);
325 	}
326 
327 	/*
328 	 * If there's an existing /dev/zvol symlink, try to use the
329 	 * same minor number we used last time.
330 	 */
331 	devpath = kmem_alloc(devpathlen, KM_SLEEP);
332 
333 	(void) sprintf(devpath, "%s/%s", ZVOL_FULL_DEV_DIR, name);
334 
335 	error = lookupname(devpath, UIO_SYSSPACE, NO_FOLLOW, NULL, &vp);
336 
337 	kmem_free(devpath, devpathlen);
338 
339 	if (error == 0 && vp->v_type != VLNK)
340 		error = EINVAL;
341 
342 	if (error == 0) {
343 		pn_alloc(&linkpath);
344 		error = pn_getsymlink(vp, &linkpath, kcred);
345 		if (error == 0) {
346 			char *ms = strstr(linkpath.pn_path, ZVOL_PSEUDO_DEV);
347 			if (ms != NULL) {
348 				ms += strlen(ZVOL_PSEUDO_DEV);
349 				minor = stoi(&ms);
350 			}
351 		}
352 		pn_free(&linkpath);
353 	}
354 
355 	if (vp != NULL)
356 		VN_RELE(vp);
357 
358 	/*
359 	 * If we found a minor but it's already in use, we must pick a new one.
360 	 */
361 	if (minor != 0 && ddi_get_soft_state(zvol_state, minor) != NULL)
362 		minor = 0;
363 
364 	if (minor == 0)
365 		minor = zvol_minor_alloc();
366 
367 	if (minor == 0) {
368 		dmu_objset_close(os);
369 		mutex_exit(&zvol_state_lock);
370 		return (ENXIO);
371 	}
372 
373 	if (ddi_soft_state_zalloc(zvol_state, minor) != DDI_SUCCESS) {
374 		dmu_objset_close(os);
375 		mutex_exit(&zvol_state_lock);
376 		return (EAGAIN);
377 	}
378 
379 	(void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, name);
380 
381 	(void) sprintf(chrbuf, "%uc,raw", minor);
382 
383 	if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
384 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
385 		ddi_soft_state_free(zvol_state, minor);
386 		dmu_objset_close(os);
387 		mutex_exit(&zvol_state_lock);
388 		return (EAGAIN);
389 	}
390 
391 	(void) sprintf(blkbuf, "%uc", minor);
392 
393 	if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
394 	    minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
395 		ddi_remove_minor_node(zfs_dip, chrbuf);
396 		ddi_soft_state_free(zvol_state, minor);
397 		dmu_objset_close(os);
398 		mutex_exit(&zvol_state_lock);
399 		return (EAGAIN);
400 	}
401 
402 	zv = ddi_get_soft_state(zvol_state, minor);
403 
404 	(void) strcpy(zv->zv_name, name);
405 	zv->zv_min_bs = DEV_BSHIFT;
406 	zv->zv_minor = minor;
407 	zv->zv_volsize = volsize;
408 	zv->zv_objset = os;
409 	zv->zv_mode = ds_mode;
410 	zv->zv_zilog = zil_open(os, NULL);
411 
412 	rw_init(&zv->zv_dslock, NULL, RW_DEFAULT, NULL);
413 
414 	zil_replay(os, zv, &zv->zv_txg_assign, zvol_replay_vector, NULL);
415 
416 	zvol_size_changed(zv, dev);
417 
418 	/* XXX this should handle the possible i/o error */
419 	VERIFY(dsl_prop_register(dmu_objset_ds(zv->zv_objset),
420 	    "readonly", zvol_readonly_changed_cb, zv) == 0);
421 
422 	zvol_minors++;
423 
424 	mutex_exit(&zvol_state_lock);
425 
426 	return (0);
427 }
428 
429 /*
430  * Remove minor node for the specified volume.
431  */
432 int
433 zvol_remove_minor(zfs_cmd_t *zc)
434 {
435 	zvol_state_t *zv;
436 	char namebuf[30];
437 
438 	mutex_enter(&zvol_state_lock);
439 
440 	if ((zv = zvol_minor_lookup(zc->zc_name)) == NULL) {
441 		mutex_exit(&zvol_state_lock);
442 		return (ENXIO);
443 	}
444 
445 	if (zv->zv_total_opens != 0) {
446 		mutex_exit(&zvol_state_lock);
447 		return (EBUSY);
448 	}
449 
450 	(void) sprintf(namebuf, "%uc,raw", zv->zv_minor);
451 	ddi_remove_minor_node(zfs_dip, namebuf);
452 
453 	(void) sprintf(namebuf, "%uc", zv->zv_minor);
454 	ddi_remove_minor_node(zfs_dip, namebuf);
455 
456 	VERIFY(dsl_prop_unregister(dmu_objset_ds(zv->zv_objset),
457 	    "readonly", zvol_readonly_changed_cb, zv) == 0);
458 
459 	zil_close(zv->zv_zilog);
460 	zv->zv_zilog = NULL;
461 	dmu_objset_close(zv->zv_objset);
462 	zv->zv_objset = NULL;
463 
464 	ddi_soft_state_free(zvol_state, zv->zv_minor);
465 
466 	zvol_minors--;
467 
468 	mutex_exit(&zvol_state_lock);
469 
470 	return (0);
471 }
472 
473 int
474 zvol_set_volsize(zfs_cmd_t *zc)
475 {
476 	zvol_state_t *zv;
477 	dev_t dev = zc->zc_dev;
478 	dmu_tx_t *tx;
479 	int error;
480 	dmu_object_info_t doi;
481 
482 	mutex_enter(&zvol_state_lock);
483 
484 	if ((zv = zvol_minor_lookup(zc->zc_name)) == NULL) {
485 		mutex_exit(&zvol_state_lock);
486 		return (ENXIO);
487 	}
488 
489 	if ((error = dmu_object_info(zv->zv_objset, ZVOL_OBJ, &doi)) != 0 ||
490 	    (error = zvol_check_volsize(zc, doi.doi_data_block_size)) != 0) {
491 		mutex_exit(&zvol_state_lock);
492 		return (error);
493 	}
494 
495 	if (zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY)) {
496 		mutex_exit(&zvol_state_lock);
497 		return (EROFS);
498 	}
499 
500 	tx = dmu_tx_create(zv->zv_objset);
501 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
502 	dmu_tx_hold_free(tx, ZVOL_OBJ, zc->zc_volsize, DMU_OBJECT_END);
503 	error = dmu_tx_assign(tx, TXG_WAIT);
504 	if (error) {
505 		dmu_tx_abort(tx);
506 		mutex_exit(&zvol_state_lock);
507 		return (error);
508 	}
509 
510 	error = zap_update(zv->zv_objset, ZVOL_ZAP_OBJ, "size", 8, 1,
511 	    &zc->zc_volsize, tx);
512 	if (error == 0) {
513 		error = dmu_free_range(zv->zv_objset, ZVOL_OBJ, zc->zc_volsize,
514 		    DMU_OBJECT_END, tx);
515 	}
516 
517 	dmu_tx_commit(tx);
518 
519 	if (error == 0) {
520 		zv->zv_volsize = zc->zc_volsize;
521 		zvol_size_changed(zv, dev);
522 	}
523 
524 	mutex_exit(&zvol_state_lock);
525 
526 	return (error);
527 }
528 
529 int
530 zvol_set_volblocksize(zfs_cmd_t *zc)
531 {
532 	zvol_state_t *zv;
533 	dmu_tx_t *tx;
534 	int error;
535 
536 	mutex_enter(&zvol_state_lock);
537 
538 	if ((zv = zvol_minor_lookup(zc->zc_name)) == NULL) {
539 		mutex_exit(&zvol_state_lock);
540 		return (ENXIO);
541 	}
542 
543 	if (zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY)) {
544 		mutex_exit(&zvol_state_lock);
545 		return (EROFS);
546 	}
547 
548 	tx = dmu_tx_create(zv->zv_objset);
549 	dmu_tx_hold_bonus(tx, ZVOL_OBJ);
550 	error = dmu_tx_assign(tx, TXG_WAIT);
551 	if (error) {
552 		dmu_tx_abort(tx);
553 	} else {
554 		error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
555 		    zc->zc_volblocksize, 0, tx);
556 		if (error == ENOTSUP)
557 			error = EBUSY;
558 		dmu_tx_commit(tx);
559 	}
560 
561 	mutex_exit(&zvol_state_lock);
562 
563 	return (error);
564 }
565 
566 /*ARGSUSED*/
567 int
568 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
569 {
570 	minor_t minor = getminor(*devp);
571 	zvol_state_t *zv;
572 
573 	if (minor == 0)			/* This is the control device */
574 		return (0);
575 
576 	mutex_enter(&zvol_state_lock);
577 
578 	zv = ddi_get_soft_state(zvol_state, minor);
579 	if (zv == NULL) {
580 		mutex_exit(&zvol_state_lock);
581 		return (ENXIO);
582 	}
583 
584 	ASSERT(zv->zv_objset != NULL);
585 
586 	if ((flag & FWRITE) &&
587 	    (zv->zv_readonly || (zv->zv_mode & DS_MODE_READONLY))) {
588 		mutex_exit(&zvol_state_lock);
589 		return (EROFS);
590 	}
591 
592 	if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
593 		zv->zv_open_count[otyp]++;
594 		zv->zv_total_opens++;
595 	}
596 
597 	mutex_exit(&zvol_state_lock);
598 
599 	return (0);
600 }
601 
602 /*ARGSUSED*/
603 int
604 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
605 {
606 	minor_t minor = getminor(dev);
607 	zvol_state_t *zv;
608 
609 	if (minor == 0)		/* This is the control device */
610 		return (0);
611 
612 	mutex_enter(&zvol_state_lock);
613 
614 	zv = ddi_get_soft_state(zvol_state, minor);
615 	if (zv == NULL) {
616 		mutex_exit(&zvol_state_lock);
617 		return (ENXIO);
618 	}
619 
620 	/*
621 	 * The next statement is a workaround for the following DDI bug:
622 	 * 6343604 specfs race: multiple "last-close" of the same device
623 	 */
624 	if (zv->zv_total_opens == 0) {
625 		mutex_exit(&zvol_state_lock);
626 		return (0);
627 	}
628 
629 	/*
630 	 * If the open count is zero, this is a spurious close.
631 	 * That indicates a bug in the kernel / DDI framework.
632 	 */
633 	ASSERT(zv->zv_open_count[otyp] != 0);
634 	ASSERT(zv->zv_total_opens != 0);
635 
636 	/*
637 	 * You may get multiple opens, but only one close.
638 	 */
639 	zv->zv_open_count[otyp]--;
640 	zv->zv_total_opens--;
641 
642 	mutex_exit(&zvol_state_lock);
643 
644 	return (0);
645 }
646 
647 /*
648  * Create and return an immediate write ZIL transaction.
649  */
650 itx_t *
651 zvol_immediate_itx(offset_t off, ssize_t len, char *addr)
652 {
653 	itx_t *itx;
654 	lr_write_t *lr;
655 
656 	itx = zil_itx_create(TX_WRITE, sizeof (*lr) + len);
657 	lr = (lr_write_t *)&itx->itx_lr;
658 	lr->lr_foid = ZVOL_OBJ;
659 	lr->lr_offset = off;
660 	lr->lr_length = len;
661 	lr->lr_blkoff = 0;
662 	BP_ZERO(&lr->lr_blkptr);
663 	bcopy(addr, (char *)itx + offsetof(itx_t, itx_lr) +
664 	    sizeof (*lr), len);
665 	itx->itx_wr_state = WR_COPIED;
666 	return (itx);
667 }
668 
669 /*
670  * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
671  *
672  * We store data in the log buffers if it's small enough.
673  * Otherwise we flush the data out via dmu_sync().
674  */
675 ssize_t zvol_immediate_write_sz = 65536;
676 
677 int
678 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t len,
679     char *addr)
680 {
681 	dmu_object_info_t doi;
682 	ssize_t nbytes;
683 	itx_t *itx;
684 	lr_write_t *lr;
685 	objset_t *os;
686 	uint64_t txg;
687 	int error;
688 	uint32_t blocksize;
689 
690 	/* handle common case */
691 	if (len <= zvol_immediate_write_sz) {
692 		itx = zvol_immediate_itx(off, len, addr);
693 		(void) zil_itx_assign(zv->zv_zilog, itx, tx);
694 		return (0);
695 	}
696 
697 	txg = dmu_tx_get_txg(tx);
698 	os = zv->zv_objset;
699 
700 	/*
701 	 * We need to dmu_sync() each block in the range.
702 	 * For this we need the blocksize.
703 	 */
704 	error = dmu_object_info(os, ZVOL_OBJ, &doi);
705 	if (error)
706 		return (error);
707 	blocksize = doi.doi_data_block_size;
708 
709 	/*
710 	 * We need to immediate write or dmu_sync() each block in the range.
711 	 */
712 	while (len) {
713 		nbytes = MIN(len, blocksize - P2PHASE(off, blocksize));
714 		if (nbytes <= zvol_immediate_write_sz) {
715 			itx = zvol_immediate_itx(off, nbytes, addr);
716 		} else {
717 			itx = zil_itx_create(TX_WRITE, sizeof (*lr));
718 			lr = (lr_write_t *)&itx->itx_lr;
719 			lr->lr_foid = ZVOL_OBJ;
720 			lr->lr_offset = off;
721 			lr->lr_length = nbytes;
722 			lr->lr_blkoff = 0;
723 			BP_ZERO(&lr->lr_blkptr);
724 
725 			txg_suspend(dmu_objset_pool(os));
726 			error = dmu_sync(os, ZVOL_OBJ, off, &lr->lr_blkoff,
727 			    &lr->lr_blkptr, txg);
728 			txg_resume(dmu_objset_pool(os));
729 			if (error) {
730 				kmem_free(itx, offsetof(itx_t, itx_lr));
731 				return (error);
732 			}
733 			itx->itx_wr_state = WR_COPIED;
734 		}
735 		(void) zil_itx_assign(zv->zv_zilog, itx, tx);
736 		len -= nbytes;
737 		off += nbytes;
738 	}
739 	return (0);
740 }
741 
742 int
743 zvol_strategy(buf_t *bp)
744 {
745 	zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
746 	uint64_t off, volsize;
747 	size_t size, resid;
748 	char *addr;
749 	objset_t *os;
750 	int error = 0;
751 	int sync;
752 	int reading;
753 	int txg_sync_needed = B_FALSE;
754 
755 	if (zv == NULL) {
756 		bioerror(bp, ENXIO);
757 		biodone(bp);
758 		return (0);
759 	}
760 
761 	if (getminor(bp->b_edev) == 0) {
762 		bioerror(bp, EINVAL);
763 		biodone(bp);
764 		return (0);
765 	}
766 
767 	if (zv->zv_readonly && !(bp->b_flags & B_READ)) {
768 		bioerror(bp, EROFS);
769 		biodone(bp);
770 		return (0);
771 	}
772 
773 	off = ldbtob(bp->b_blkno);
774 	volsize = zv->zv_volsize;
775 
776 	os = zv->zv_objset;
777 	ASSERT(os != NULL);
778 	sync = !(bp->b_flags & B_ASYNC) && !(zil_disable);
779 
780 	bp_mapin(bp);
781 	addr = bp->b_un.b_addr;
782 	resid = bp->b_bcount;
783 
784 	/*
785 	 * There must be no buffer changes when doing a dmu_sync() because
786 	 * we can't change the data whilst calculating the checksum.
787 	 * A better approach than a per zvol rwlock would be to lock ranges.
788 	 */
789 	reading = bp->b_flags & B_READ;
790 	if (reading || resid <= zvol_immediate_write_sz)
791 		rw_enter(&zv->zv_dslock, RW_READER);
792 	else
793 		rw_enter(&zv->zv_dslock, RW_WRITER);
794 
795 	while (resid != 0 && off < volsize) {
796 
797 		size = MIN(resid, 1UL << 20);	/* cap at 1MB per tx */
798 
799 		if (size > volsize - off)	/* don't write past the end */
800 			size = volsize - off;
801 
802 		if (reading) {
803 			error = dmu_read(os, ZVOL_OBJ, off, size, addr);
804 		} else {
805 			dmu_tx_t *tx = dmu_tx_create(os);
806 			dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
807 			error = dmu_tx_assign(tx, TXG_WAIT);
808 			if (error) {
809 				dmu_tx_abort(tx);
810 			} else {
811 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
812 				if (sync) {
813 					/* use the ZIL to commit this write */
814 					if (zvol_log_write(zv, tx, off, size,
815 					    addr) != 0) {
816 						txg_sync_needed = B_TRUE;
817 					}
818 				}
819 				dmu_tx_commit(tx);
820 			}
821 		}
822 		if (error)
823 			break;
824 		off += size;
825 		addr += size;
826 		resid -= size;
827 	}
828 	rw_exit(&zv->zv_dslock);
829 
830 	if ((bp->b_resid = resid) == bp->b_bcount)
831 		bioerror(bp, off > volsize ? EINVAL : error);
832 
833 	biodone(bp);
834 
835 	if (sync) {
836 		if (txg_sync_needed)
837 			txg_wait_synced(dmu_objset_pool(os), 0);
838 		else
839 			zil_commit(zv->zv_zilog, UINT64_MAX, FDSYNC);
840 	}
841 
842 	return (0);
843 }
844 
845 /*ARGSUSED*/
846 int
847 zvol_read(dev_t dev, uio_t *uiop, cred_t *cr)
848 {
849 	return (physio(zvol_strategy, NULL, dev, B_READ, minphys, uiop));
850 }
851 
852 /*ARGSUSED*/
853 int
854 zvol_write(dev_t dev, uio_t *uiop, cred_t *cr)
855 {
856 	return (physio(zvol_strategy, NULL, dev, B_WRITE, minphys, uiop));
857 }
858 
859 /*ARGSUSED*/
860 int
861 zvol_aread(dev_t dev, struct aio_req *aio, cred_t *cr)
862 {
863 	return (aphysio(zvol_strategy, anocancel, dev, B_READ, minphys, aio));
864 }
865 
866 /*ARGSUSED*/
867 int
868 zvol_awrite(dev_t dev, struct aio_req *aio, cred_t *cr)
869 {
870 	return (aphysio(zvol_strategy, anocancel, dev, B_WRITE, minphys, aio));
871 }
872 
873 /*
874  * Dirtbag ioctls to support mkfs(1M) for UFS filesystems.  See dkio(7I).
875  */
876 /*ARGSUSED*/
877 int
878 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
879 {
880 	zvol_state_t *zv;
881 	struct dk_cinfo dkc;
882 	struct dk_minfo dkm;
883 	dk_efi_t efi;
884 	efi_gpt_t gpt;
885 	efi_gpe_t gpe;
886 	struct uuid uuid = EFI_RESERVED;
887 	uint32_t crc;
888 	int error = 0;
889 
890 	mutex_enter(&zvol_state_lock);
891 
892 	zv = ddi_get_soft_state(zvol_state, getminor(dev));
893 
894 	if (zv == NULL) {
895 		mutex_exit(&zvol_state_lock);
896 		return (ENXIO);
897 	}
898 
899 	switch (cmd) {
900 
901 	case DKIOCINFO:
902 		bzero(&dkc, sizeof (dkc));
903 		(void) strcpy(dkc.dki_cname, "zvol");
904 		(void) strcpy(dkc.dki_dname, "zvol");
905 		dkc.dki_ctype = DKC_UNKNOWN;
906 		dkc.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
907 		mutex_exit(&zvol_state_lock);
908 		if (ddi_copyout(&dkc, (void *)arg, sizeof (dkc), flag))
909 			error = EFAULT;
910 		return (error);
911 
912 	case DKIOCGMEDIAINFO:
913 		bzero(&dkm, sizeof (dkm));
914 		dkm.dki_lbsize = 1U << zv->zv_min_bs;
915 		dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
916 		dkm.dki_media_type = DK_UNKNOWN;
917 		mutex_exit(&zvol_state_lock);
918 		if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
919 			error = EFAULT;
920 		return (error);
921 
922 	case DKIOCGETEFI:
923 		if (ddi_copyin((void *)arg, &efi, sizeof (dk_efi_t), flag)) {
924 			mutex_exit(&zvol_state_lock);
925 			return (EFAULT);
926 		}
927 
928 		bzero(&gpt, sizeof (gpt));
929 		bzero(&gpe, sizeof (gpe));
930 
931 		efi.dki_data = (void *)(uintptr_t)efi.dki_data_64;
932 
933 		if (efi.dki_length < sizeof (gpt) + sizeof (gpe)) {
934 			mutex_exit(&zvol_state_lock);
935 			return (EINVAL);
936 		}
937 
938 		efi.dki_length = sizeof (gpt) + sizeof (gpe);
939 
940 		gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
941 		gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
942 		gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
943 		gpt.efi_gpt_FirstUsableLBA = LE_64(0ULL);
944 		gpt.efi_gpt_LastUsableLBA =
945 		    LE_64((zv->zv_volsize >> zv->zv_min_bs) - 1);
946 		gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
947 		gpt.efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (gpe));
948 
949 		UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
950 		gpe.efi_gpe_StartingLBA = gpt.efi_gpt_FirstUsableLBA;
951 		gpe.efi_gpe_EndingLBA = gpt.efi_gpt_LastUsableLBA;
952 
953 		CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
954 		gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
955 
956 		CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
957 		gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
958 
959 		mutex_exit(&zvol_state_lock);
960 		if (ddi_copyout(&gpt, efi.dki_data, sizeof (gpt), flag) ||
961 		    ddi_copyout(&gpe, efi.dki_data + 1, sizeof (gpe), flag))
962 			error = EFAULT;
963 		return (error);
964 
965 	default:
966 		error = ENOTSUP;
967 		break;
968 
969 	}
970 	mutex_exit(&zvol_state_lock);
971 	return (error);
972 }
973 
974 int
975 zvol_busy(void)
976 {
977 	return (zvol_minors != 0);
978 }
979 
980 void
981 zvol_init(void)
982 {
983 	VERIFY(ddi_soft_state_init(&zvol_state, sizeof (zvol_state_t), 1) == 0);
984 	mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
985 }
986 
987 void
988 zvol_fini(void)
989 {
990 	mutex_destroy(&zvol_state_lock);
991 	ddi_soft_state_fini(&zvol_state);
992 }
993