xref: /freebsd-src/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c (revision c14e17a49cdd72c8c05d12fbec631c409b14b9cb)
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  * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
25  * All rights reserved.
26  *
27  * Portions Copyright 2010 Robert Milkowski
28  *
29  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
30  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
31  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32  * Copyright (c) 2014 Integros [integros.com]
33  */
34 
35 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
36 
37 /*
38  * ZFS volume emulation driver.
39  *
40  * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
41  * Volumes are accessed through the symbolic links named:
42  *
43  * /dev/zvol/<pool_name>/<dataset_name>
44  *
45  * Volumes are persistent through reboot.  No user command needs to be
46  * run before opening and using a device.
47  *
48  * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
49  * in the system. Except when they're simply character devices (volmode=dev).
50  */
51 
52 #include <sys/types.h>
53 #include <sys/param.h>
54 #include <sys/kernel.h>
55 #include <sys/errno.h>
56 #include <sys/uio.h>
57 #include <sys/bio.h>
58 #include <sys/buf.h>
59 #include <sys/kmem.h>
60 #include <sys/conf.h>
61 #include <sys/cmn_err.h>
62 #include <sys/stat.h>
63 #include <sys/proc.h>
64 #include <sys/zap.h>
65 #include <sys/spa.h>
66 #include <sys/spa_impl.h>
67 #include <sys/zio.h>
68 #include <sys/disk.h>
69 #include <sys/dmu_traverse.h>
70 #include <sys/dnode.h>
71 #include <sys/dsl_dataset.h>
72 #include <sys/dsl_prop.h>
73 #include <sys/dsl_dir.h>
74 #include <sys/byteorder.h>
75 #include <sys/sunddi.h>
76 #include <sys/dirent.h>
77 #include <sys/policy.h>
78 #include <sys/queue.h>
79 #include <sys/fs/zfs.h>
80 #include <sys/zfs_ioctl.h>
81 #include <sys/zil.h>
82 #include <sys/zfs_znode.h>
83 #include <sys/zfs_rlock.h>
84 #include <sys/vdev_impl.h>
85 #include <sys/vdev_raidz.h>
86 #include <sys/zvol.h>
87 #include <sys/zil_impl.h>
88 #include <sys/dataset_kstats.h>
89 #include <sys/dbuf.h>
90 #include <sys/dmu_tx.h>
91 #include <sys/zfeature.h>
92 #include <sys/zio_checksum.h>
93 #include <sys/zil_impl.h>
94 #include <sys/filio.h>
95 
96 #include <geom/geom.h>
97 #include <sys/zvol.h>
98 #include <sys/zvol_impl.h>
99 
100 #include "zfs_namecheck.h"
101 
102 #define	ZVOL_DUMPSIZE		"dumpsize"
103 
104 #ifdef ZVOL_LOCK_DEBUG
105 #define	ZVOL_RW_READER		RW_WRITER
106 #define	ZVOL_RW_READ_HELD	RW_WRITE_HELD
107 #else
108 #define	ZVOL_RW_READER		RW_READER
109 #define	ZVOL_RW_READ_HELD	RW_READ_HELD
110 #endif
111 
112 enum zvol_geom_state {
113 	ZVOL_GEOM_UNINIT,
114 	ZVOL_GEOM_STOPPED,
115 	ZVOL_GEOM_RUNNING,
116 };
117 
118 struct zvol_state_os {
119 #define	zso_dev		_zso_state._zso_dev
120 #define	zso_geom	_zso_state._zso_geom
121 	union {
122 		/* volmode=dev */
123 		struct zvol_state_dev {
124 			struct cdev *zsd_cdev;
125 			uint64_t zsd_sync_cnt;
126 		} _zso_dev;
127 
128 		/* volmode=geom */
129 		struct zvol_state_geom {
130 			struct g_provider *zsg_provider;
131 			struct bio_queue_head zsg_queue;
132 			struct mtx zsg_queue_mtx;
133 			enum zvol_geom_state zsg_state;
134 		} _zso_geom;
135 	} _zso_state;
136 	int zso_dying;
137 };
138 
139 static uint32_t zvol_minors;
140 
141 SYSCTL_DECL(_vfs_zfs);
142 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
143 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &zvol_volmode, 0,
144 	"Expose as GEOM providers (1), device files (2) or neither");
145 static boolean_t zpool_on_zvol = B_FALSE;
146 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, recursive, CTLFLAG_RWTUN, &zpool_on_zvol, 0,
147 	"Allow zpools to use zvols as vdevs (DANGEROUS)");
148 
149 /*
150  * Toggle unmap functionality.
151  */
152 boolean_t zvol_unmap_enabled = B_TRUE;
153 
154 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
155 	&zvol_unmap_enabled, 0, "Enable UNMAP functionality");
156 
157 /*
158  * zvol maximum transfer in one DMU tx.
159  */
160 int zvol_maxphys = DMU_MAX_ACCESS / 2;
161 
162 static void zvol_ensure_zilog(zvol_state_t *zv);
163 
164 static d_open_t		zvol_cdev_open;
165 static d_close_t	zvol_cdev_close;
166 static d_ioctl_t	zvol_cdev_ioctl;
167 static d_read_t		zvol_cdev_read;
168 static d_write_t	zvol_cdev_write;
169 static d_strategy_t	zvol_geom_bio_strategy;
170 
171 static struct cdevsw zvol_cdevsw = {
172 	.d_name =	"zvol",
173 	.d_version =	D_VERSION,
174 	.d_flags =	D_DISK | D_TRACKCLOSE,
175 	.d_open =	zvol_cdev_open,
176 	.d_close =	zvol_cdev_close,
177 	.d_ioctl =	zvol_cdev_ioctl,
178 	.d_read =	zvol_cdev_read,
179 	.d_write =	zvol_cdev_write,
180 	.d_strategy =	zvol_geom_bio_strategy,
181 };
182 
183 extern uint_t zfs_geom_probe_vdev_key;
184 
185 struct g_class zfs_zvol_class = {
186 	.name = "ZFS::ZVOL",
187 	.version = G_VERSION,
188 };
189 
190 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
191 
192 static int zvol_geom_open(struct g_provider *pp, int flag, int count);
193 static int zvol_geom_close(struct g_provider *pp, int flag, int count);
194 static void zvol_geom_run(zvol_state_t *zv);
195 static void zvol_geom_destroy(zvol_state_t *zv);
196 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
197 static void zvol_geom_worker(void *arg);
198 static void zvol_geom_bio_start(struct bio *bp);
199 static int zvol_geom_bio_getattr(struct bio *bp);
200 /* static d_strategy_t	zvol_geom_bio_strategy; (declared elsewhere) */
201 
202 /*
203  * GEOM mode implementation
204  */
205 
206 /*ARGSUSED*/
207 static int
208 zvol_geom_open(struct g_provider *pp, int flag, int count)
209 {
210 	zvol_state_t *zv;
211 	int err = 0;
212 	boolean_t drop_suspend = B_FALSE;
213 	boolean_t drop_namespace = B_FALSE;
214 
215 	if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
216 		/*
217 		 * if zfs_geom_probe_vdev_key is set, that means that zfs is
218 		 * attempting to probe geom providers while looking for a
219 		 * replacement for a missing VDEV.  In this case, the
220 		 * spa_namespace_lock will not be held, but it is still illegal
221 		 * to use a zvol as a vdev.  Deadlocks can result if another
222 		 * thread has spa_namespace_lock
223 		 */
224 		return (SET_ERROR(EOPNOTSUPP));
225 	}
226 
227 retry:
228 	rw_enter(&zvol_state_lock, ZVOL_RW_READER);
229 	zv = pp->private;
230 	if (zv == NULL) {
231 		rw_exit(&zvol_state_lock);
232 		err = SET_ERROR(ENXIO);
233 		goto out_locked;
234 	}
235 
236 	if (zv->zv_open_count == 0 && !mutex_owned(&spa_namespace_lock)) {
237 		/*
238 		 * We need to guarantee that the namespace lock is held
239 		 * to avoid spurious failures in zvol_first_open.
240 		 */
241 		drop_namespace = B_TRUE;
242 		if (!mutex_tryenter(&spa_namespace_lock)) {
243 			rw_exit(&zvol_state_lock);
244 			mutex_enter(&spa_namespace_lock);
245 			goto retry;
246 		}
247 	}
248 	mutex_enter(&zv->zv_state_lock);
249 	if (zv->zv_zso->zso_dying) {
250 		rw_exit(&zvol_state_lock);
251 		err = SET_ERROR(ENXIO);
252 		goto out_zv_locked;
253 	}
254 	ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM);
255 
256 	/*
257 	 * make sure zvol is not suspended during first open
258 	 * (hold zv_suspend_lock) and respect proper lock acquisition
259 	 * ordering - zv_suspend_lock before zv_state_lock
260 	 */
261 	if (zv->zv_open_count == 0) {
262 		drop_suspend = B_TRUE;
263 		if (!rw_tryenter(&zv->zv_suspend_lock, ZVOL_RW_READER)) {
264 			mutex_exit(&zv->zv_state_lock);
265 			rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
266 			mutex_enter(&zv->zv_state_lock);
267 			/* check to see if zv_suspend_lock is needed */
268 			if (zv->zv_open_count != 0) {
269 				rw_exit(&zv->zv_suspend_lock);
270 				drop_suspend = B_FALSE;
271 			}
272 		}
273 	}
274 	rw_exit(&zvol_state_lock);
275 
276 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
277 
278 	if (zv->zv_open_count == 0) {
279 		ASSERT(ZVOL_RW_READ_HELD(&zv->zv_suspend_lock));
280 		err = zvol_first_open(zv, !(flag & FWRITE));
281 		if (err)
282 			goto out_zv_locked;
283 		pp->mediasize = zv->zv_volsize;
284 		pp->stripeoffset = 0;
285 		pp->stripesize = zv->zv_volblocksize;
286 	}
287 
288 	/*
289 	 * Check for a bad on-disk format version now since we
290 	 * lied about owning the dataset readonly before.
291 	 */
292 	if ((flag & FWRITE) && ((zv->zv_flags & ZVOL_RDONLY) ||
293 	    dmu_objset_incompatible_encryption_version(zv->zv_objset))) {
294 		err = SET_ERROR(EROFS);
295 		goto out_opened;
296 	}
297 	if (zv->zv_flags & ZVOL_EXCL) {
298 		err = SET_ERROR(EBUSY);
299 		goto out_opened;
300 	}
301 #ifdef FEXCL
302 	if (flag & FEXCL) {
303 		if (zv->zv_open_count != 0) {
304 			err = SET_ERROR(EBUSY);
305 			goto out_opened;
306 		}
307 		zv->zv_flags |= ZVOL_EXCL;
308 	}
309 #endif
310 
311 	zv->zv_open_count += count;
312 out_opened:
313 	if (zv->zv_open_count == 0) {
314 		zvol_last_close(zv);
315 		wakeup(zv);
316 	}
317 out_zv_locked:
318 	mutex_exit(&zv->zv_state_lock);
319 out_locked:
320 	if (drop_namespace)
321 		mutex_exit(&spa_namespace_lock);
322 	if (drop_suspend)
323 		rw_exit(&zv->zv_suspend_lock);
324 	return (err);
325 }
326 
327 /*ARGSUSED*/
328 static int
329 zvol_geom_close(struct g_provider *pp, int flag, int count)
330 {
331 	zvol_state_t *zv;
332 	boolean_t drop_suspend = B_TRUE;
333 	int new_open_count;
334 
335 	rw_enter(&zvol_state_lock, ZVOL_RW_READER);
336 	zv = pp->private;
337 	if (zv == NULL) {
338 		rw_exit(&zvol_state_lock);
339 		return (SET_ERROR(ENXIO));
340 	}
341 
342 	mutex_enter(&zv->zv_state_lock);
343 	if (zv->zv_flags & ZVOL_EXCL) {
344 		ASSERT3U(zv->zv_open_count, ==, 1);
345 		zv->zv_flags &= ~ZVOL_EXCL;
346 	}
347 
348 	ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM);
349 
350 	/*
351 	 * If the open count is zero, this is a spurious close.
352 	 * That indicates a bug in the kernel / DDI framework.
353 	 */
354 	ASSERT3U(zv->zv_open_count, >, 0);
355 
356 	/*
357 	 * make sure zvol is not suspended during last close
358 	 * (hold zv_suspend_lock) and respect proper lock acquisition
359 	 * ordering - zv_suspend_lock before zv_state_lock
360 	 */
361 	new_open_count = zv->zv_open_count - count;
362 	if (new_open_count == 0) {
363 		if (!rw_tryenter(&zv->zv_suspend_lock, ZVOL_RW_READER)) {
364 			mutex_exit(&zv->zv_state_lock);
365 			rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
366 			mutex_enter(&zv->zv_state_lock);
367 			/* check to see if zv_suspend_lock is needed */
368 			new_open_count = zv->zv_open_count - count;
369 			if (new_open_count != 0) {
370 				rw_exit(&zv->zv_suspend_lock);
371 				drop_suspend = B_FALSE;
372 			}
373 		}
374 	} else {
375 		drop_suspend = B_FALSE;
376 	}
377 	rw_exit(&zvol_state_lock);
378 
379 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
380 
381 	/*
382 	 * You may get multiple opens, but only one close.
383 	 */
384 	zv->zv_open_count = new_open_count;
385 	if (zv->zv_open_count == 0) {
386 		ASSERT(ZVOL_RW_READ_HELD(&zv->zv_suspend_lock));
387 		zvol_last_close(zv);
388 		wakeup(zv);
389 	}
390 
391 	mutex_exit(&zv->zv_state_lock);
392 
393 	if (drop_suspend)
394 		rw_exit(&zv->zv_suspend_lock);
395 	return (0);
396 }
397 
398 static void
399 zvol_geom_run(zvol_state_t *zv)
400 {
401 	struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
402 	struct g_provider *pp = zsg->zsg_provider;
403 
404 	ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM);
405 
406 	g_error_provider(pp, 0);
407 
408 	kproc_kthread_add(zvol_geom_worker, zv, &system_proc, NULL, 0, 0,
409 	    "zfskern", "zvol %s", pp->name + sizeof (ZVOL_DRIVER));
410 }
411 
412 static void
413 zvol_geom_destroy(zvol_state_t *zv)
414 {
415 	struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
416 	struct g_provider *pp = zsg->zsg_provider;
417 
418 	ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM);
419 
420 	g_topology_assert();
421 
422 	mutex_enter(&zv->zv_state_lock);
423 	VERIFY(zsg->zsg_state == ZVOL_GEOM_RUNNING);
424 	mutex_exit(&zv->zv_state_lock);
425 	zsg->zsg_provider = NULL;
426 	g_wither_geom(pp->geom, ENXIO);
427 }
428 
429 void
430 zvol_wait_close(zvol_state_t *zv)
431 {
432 
433 	if (zv->zv_volmode != ZFS_VOLMODE_GEOM)
434 		return;
435 	mutex_enter(&zv->zv_state_lock);
436 	zv->zv_zso->zso_dying = B_TRUE;
437 
438 	if (zv->zv_open_count)
439 		msleep(zv, &zv->zv_state_lock,
440 		    PRIBIO, "zvol:dying", 10*hz);
441 	mutex_exit(&zv->zv_state_lock);
442 }
443 
444 
445 static int
446 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
447 {
448 	int count, error, flags;
449 
450 	g_topology_assert();
451 
452 	/*
453 	 * To make it easier we expect either open or close, but not both
454 	 * at the same time.
455 	 */
456 	KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
457 	    (acr <= 0 && acw <= 0 && ace <= 0),
458 	    ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
459 	    pp->name, acr, acw, ace));
460 
461 	if (pp->private == NULL) {
462 		if (acr <= 0 && acw <= 0 && ace <= 0)
463 			return (0);
464 		return (pp->error);
465 	}
466 
467 	/*
468 	 * We don't pass FEXCL flag to zvol_geom_open()/zvol_geom_close() if
469 	 * ace != 0, because GEOM already handles that and handles it a bit
470 	 * differently. GEOM allows for multiple read/exclusive consumers and
471 	 * ZFS allows only one exclusive consumer, no matter if it is reader or
472 	 * writer. I like better the way GEOM works so I'll leave it for GEOM
473 	 * to decide what to do.
474 	 */
475 
476 	count = acr + acw + ace;
477 	if (count == 0)
478 		return (0);
479 
480 	flags = 0;
481 	if (acr != 0 || ace != 0)
482 		flags |= FREAD;
483 	if (acw != 0)
484 		flags |= FWRITE;
485 
486 	g_topology_unlock();
487 	if (count > 0)
488 		error = zvol_geom_open(pp, flags, count);
489 	else
490 		error = zvol_geom_close(pp, flags, -count);
491 	g_topology_lock();
492 	return (error);
493 }
494 
495 static void
496 zvol_geom_worker(void *arg)
497 {
498 	zvol_state_t *zv = arg;
499 	struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
500 	struct bio *bp;
501 
502 	ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM);
503 
504 	thread_lock(curthread);
505 	sched_prio(curthread, PRIBIO);
506 	thread_unlock(curthread);
507 
508 	for (;;) {
509 		mtx_lock(&zsg->zsg_queue_mtx);
510 		bp = bioq_takefirst(&zsg->zsg_queue);
511 		if (bp == NULL) {
512 			if (zsg->zsg_state == ZVOL_GEOM_STOPPED) {
513 				zsg->zsg_state = ZVOL_GEOM_RUNNING;
514 				wakeup(&zsg->zsg_state);
515 				mtx_unlock(&zsg->zsg_queue_mtx);
516 				kthread_exit();
517 			}
518 			msleep(&zsg->zsg_queue, &zsg->zsg_queue_mtx,
519 			    PRIBIO | PDROP, "zvol:io", 0);
520 			continue;
521 		}
522 		mtx_unlock(&zsg->zsg_queue_mtx);
523 		zvol_geom_bio_strategy(bp);
524 	}
525 }
526 
527 static void
528 zvol_geom_bio_start(struct bio *bp)
529 {
530 	zvol_state_t *zv = bp->bio_to->private;
531 	struct zvol_state_geom *zsg;
532 	boolean_t first;
533 
534 	if (zv == NULL) {
535 		g_io_deliver(bp, ENXIO);
536 		return;
537 	}
538 	if (bp->bio_cmd == BIO_GETATTR) {
539 		if (zvol_geom_bio_getattr(bp))
540 			g_io_deliver(bp, EOPNOTSUPP);
541 		return;
542 	}
543 
544 	if (!THREAD_CAN_SLEEP()) {
545 		zsg = &zv->zv_zso->zso_geom;
546 		mtx_lock(&zsg->zsg_queue_mtx);
547 		first = (bioq_first(&zsg->zsg_queue) == NULL);
548 		bioq_insert_tail(&zsg->zsg_queue, bp);
549 		mtx_unlock(&zsg->zsg_queue_mtx);
550 		if (first)
551 			wakeup_one(&zsg->zsg_queue);
552 		return;
553 	}
554 
555 	zvol_geom_bio_strategy(bp);
556 }
557 
558 static int
559 zvol_geom_bio_getattr(struct bio *bp)
560 {
561 	zvol_state_t *zv;
562 
563 	zv = bp->bio_to->private;
564 	ASSERT3P(zv, !=, NULL);
565 
566 	spa_t *spa = dmu_objset_spa(zv->zv_objset);
567 	uint64_t refd, avail, usedobjs, availobjs;
568 
569 	if (g_handleattr_int(bp, "GEOM::candelete", 1))
570 		return (0);
571 	if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
572 		dmu_objset_space(zv->zv_objset, &refd, &avail,
573 		    &usedobjs, &availobjs);
574 		if (g_handleattr_off_t(bp, "blocksavail", avail / DEV_BSIZE))
575 			return (0);
576 	} else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
577 		dmu_objset_space(zv->zv_objset, &refd, &avail,
578 		    &usedobjs, &availobjs);
579 		if (g_handleattr_off_t(bp, "blocksused", refd / DEV_BSIZE))
580 			return (0);
581 	} else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
582 		avail = metaslab_class_get_space(spa_normal_class(spa));
583 		avail -= metaslab_class_get_alloc(spa_normal_class(spa));
584 		if (g_handleattr_off_t(bp, "poolblocksavail",
585 		    avail / DEV_BSIZE))
586 			return (0);
587 	} else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
588 		refd = metaslab_class_get_alloc(spa_normal_class(spa));
589 		if (g_handleattr_off_t(bp, "poolblocksused", refd / DEV_BSIZE))
590 			return (0);
591 	}
592 	return (1);
593 }
594 
595 static void
596 zvol_geom_bio_strategy(struct bio *bp)
597 {
598 	zvol_state_t *zv;
599 	uint64_t off, volsize;
600 	size_t resid;
601 	char *addr;
602 	objset_t *os;
603 	zfs_locked_range_t *lr;
604 	int error = 0;
605 	boolean_t doread = B_FALSE;
606 	boolean_t is_dumpified;
607 	boolean_t sync;
608 
609 	if (bp->bio_to)
610 		zv = bp->bio_to->private;
611 	else
612 		zv = bp->bio_dev->si_drv2;
613 
614 	if (zv == NULL) {
615 		error = SET_ERROR(ENXIO);
616 		goto out;
617 	}
618 
619 	rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
620 
621 	switch (bp->bio_cmd) {
622 	case BIO_READ:
623 		doread = B_TRUE;
624 		break;
625 	case BIO_WRITE:
626 	case BIO_FLUSH:
627 	case BIO_DELETE:
628 		if (zv->zv_flags & ZVOL_RDONLY) {
629 			error = SET_ERROR(EROFS);
630 			goto resume;
631 		}
632 		zvol_ensure_zilog(zv);
633 		if (bp->bio_cmd == BIO_FLUSH)
634 			goto sync;
635 		break;
636 	default:
637 		error = SET_ERROR(EOPNOTSUPP);
638 		goto resume;
639 	}
640 
641 	off = bp->bio_offset;
642 	volsize = zv->zv_volsize;
643 
644 	os = zv->zv_objset;
645 	ASSERT3P(os, !=, NULL);
646 
647 	addr = bp->bio_data;
648 	resid = bp->bio_length;
649 
650 	if (resid > 0 && off >= volsize) {
651 		error = SET_ERROR(EIO);
652 		goto resume;
653 	}
654 
655 	is_dumpified = B_FALSE;
656 	sync = !doread && !is_dumpified &&
657 	    zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
658 
659 	/*
660 	 * There must be no buffer changes when doing a dmu_sync() because
661 	 * we can't change the data whilst calculating the checksum.
662 	 */
663 	lr = zfs_rangelock_enter(&zv->zv_rangelock, off, resid,
664 	    doread ? RL_READER : RL_WRITER);
665 
666 	if (bp->bio_cmd == BIO_DELETE) {
667 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
668 		error = dmu_tx_assign(tx, TXG_WAIT);
669 		if (error != 0) {
670 			dmu_tx_abort(tx);
671 		} else {
672 			zvol_log_truncate(zv, tx, off, resid, sync);
673 			dmu_tx_commit(tx);
674 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
675 			    off, resid);
676 			resid = 0;
677 		}
678 		goto unlock;
679 	}
680 	while (resid != 0 && off < volsize) {
681 		size_t size = MIN(resid, zvol_maxphys);
682 		if (doread) {
683 			error = dmu_read(os, ZVOL_OBJ, off, size, addr,
684 			    DMU_READ_PREFETCH);
685 		} else {
686 			dmu_tx_t *tx = dmu_tx_create(os);
687 			dmu_tx_hold_write_by_dnode(tx, zv->zv_dn, off, size);
688 			error = dmu_tx_assign(tx, TXG_WAIT);
689 			if (error) {
690 				dmu_tx_abort(tx);
691 			} else {
692 				dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
693 				zvol_log_write(zv, tx, off, size, sync);
694 				dmu_tx_commit(tx);
695 			}
696 		}
697 		if (error) {
698 			/* convert checksum errors into IO errors */
699 			if (error == ECKSUM)
700 				error = SET_ERROR(EIO);
701 			break;
702 		}
703 		off += size;
704 		addr += size;
705 		resid -= size;
706 	}
707 unlock:
708 	zfs_rangelock_exit(lr);
709 
710 	bp->bio_completed = bp->bio_length - resid;
711 	if (bp->bio_completed < bp->bio_length && off > volsize)
712 		error = SET_ERROR(EINVAL);
713 
714 	switch (bp->bio_cmd) {
715 	case BIO_FLUSH:
716 		break;
717 	case BIO_READ:
718 		dataset_kstats_update_read_kstats(&zv->zv_kstat,
719 		    bp->bio_completed);
720 		break;
721 	case BIO_WRITE:
722 		dataset_kstats_update_write_kstats(&zv->zv_kstat,
723 		    bp->bio_completed);
724 		break;
725 	case BIO_DELETE:
726 		break;
727 	default:
728 		break;
729 	}
730 
731 	if (sync) {
732 sync:
733 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
734 	}
735 resume:
736 	rw_exit(&zv->zv_suspend_lock);
737 out:
738 	if (bp->bio_to)
739 		g_io_deliver(bp, error);
740 	else
741 		biofinish(bp, NULL, error);
742 }
743 
744 /*
745  * Character device mode implementation
746  */
747 
748 static int
749 zvol_cdev_read(struct cdev *dev, struct uio *uio, int ioflag)
750 {
751 	zvol_state_t *zv;
752 	uint64_t volsize;
753 	zfs_locked_range_t *lr;
754 	int error = 0;
755 
756 	zv = dev->si_drv2;
757 
758 	volsize = zv->zv_volsize;
759 	/*
760 	 * uio_loffset == volsize isn't an error as
761 	 * its required for EOF processing.
762 	 */
763 	if (uio->uio_resid > 0 &&
764 	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
765 		return (SET_ERROR(EIO));
766 
767 	lr = zfs_rangelock_enter(&zv->zv_rangelock, uio->uio_loffset,
768 	    uio->uio_resid, RL_READER);
769 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
770 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
771 
772 		/* don't read past the end */
773 		if (bytes > volsize - uio->uio_loffset)
774 			bytes = volsize - uio->uio_loffset;
775 
776 		error =  dmu_read_uio_dnode(zv->zv_dn, uio, bytes);
777 		if (error) {
778 			/* convert checksum errors into IO errors */
779 			if (error == ECKSUM)
780 				error = SET_ERROR(EIO);
781 			break;
782 		}
783 	}
784 	zfs_rangelock_exit(lr);
785 
786 	return (error);
787 }
788 
789 static int
790 zvol_cdev_write(struct cdev *dev, struct uio *uio, int ioflag)
791 {
792 	zvol_state_t *zv;
793 	uint64_t volsize;
794 	zfs_locked_range_t *lr;
795 	int error = 0;
796 	boolean_t sync;
797 
798 	zv = dev->si_drv2;
799 
800 	volsize = zv->zv_volsize;
801 
802 	if (uio->uio_resid > 0 &&
803 	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
804 		return (SET_ERROR(EIO));
805 
806 	sync = (ioflag & IO_SYNC) ||
807 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
808 
809 	rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
810 	zvol_ensure_zilog(zv);
811 
812 	lr = zfs_rangelock_enter(&zv->zv_rangelock, uio->uio_loffset,
813 	    uio->uio_resid, RL_WRITER);
814 	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
815 		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
816 		uint64_t off = uio->uio_loffset;
817 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
818 
819 		if (bytes > volsize - off)	/* don't write past the end */
820 			bytes = volsize - off;
821 
822 		dmu_tx_hold_write_by_dnode(tx, zv->zv_dn, off, bytes);
823 		error = dmu_tx_assign(tx, TXG_WAIT);
824 		if (error) {
825 			dmu_tx_abort(tx);
826 			break;
827 		}
828 		error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
829 		if (error == 0)
830 			zvol_log_write(zv, tx, off, bytes, sync);
831 		dmu_tx_commit(tx);
832 
833 		if (error)
834 			break;
835 	}
836 	zfs_rangelock_exit(lr);
837 	if (sync)
838 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
839 	rw_exit(&zv->zv_suspend_lock);
840 	return (error);
841 }
842 
843 static int
844 zvol_cdev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
845 {
846 	zvol_state_t *zv;
847 	struct zvol_state_dev *zsd;
848 	int err = 0;
849 	boolean_t drop_suspend = B_FALSE;
850 	boolean_t drop_namespace = B_FALSE;
851 
852 retry:
853 	rw_enter(&zvol_state_lock, ZVOL_RW_READER);
854 	zv = dev->si_drv2;
855 	if (zv == NULL) {
856 		rw_exit(&zvol_state_lock);
857 		err = SET_ERROR(ENXIO);
858 		goto out_locked;
859 	}
860 
861 	if (zv->zv_open_count == 0 && !mutex_owned(&spa_namespace_lock)) {
862 		/*
863 		 * We need to guarantee that the namespace lock is held
864 		 * to avoid spurious failures in zvol_first_open.
865 		 */
866 		drop_namespace = B_TRUE;
867 		if (!mutex_tryenter(&spa_namespace_lock)) {
868 			rw_exit(&zvol_state_lock);
869 			mutex_enter(&spa_namespace_lock);
870 			goto retry;
871 		}
872 	}
873 	mutex_enter(&zv->zv_state_lock);
874 
875 	ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_DEV);
876 
877 	/*
878 	 * make sure zvol is not suspended during first open
879 	 * (hold zv_suspend_lock) and respect proper lock acquisition
880 	 * ordering - zv_suspend_lock before zv_state_lock
881 	 */
882 	if (zv->zv_open_count == 0) {
883 		drop_suspend = B_TRUE;
884 		if (!rw_tryenter(&zv->zv_suspend_lock, ZVOL_RW_READER)) {
885 			mutex_exit(&zv->zv_state_lock);
886 			rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
887 			mutex_enter(&zv->zv_state_lock);
888 			/* check to see if zv_suspend_lock is needed */
889 			if (zv->zv_open_count != 0) {
890 				rw_exit(&zv->zv_suspend_lock);
891 				drop_suspend = B_FALSE;
892 			}
893 		}
894 	}
895 	rw_exit(&zvol_state_lock);
896 
897 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
898 
899 	if (zv->zv_open_count == 0) {
900 		ASSERT(ZVOL_RW_READ_HELD(&zv->zv_suspend_lock));
901 		err = zvol_first_open(zv, !(flags & FWRITE));
902 		if (err)
903 			goto out_zv_locked;
904 	}
905 
906 	if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
907 		err = SET_ERROR(EROFS);
908 		goto out_opened;
909 	}
910 	if (zv->zv_flags & ZVOL_EXCL) {
911 		err = SET_ERROR(EBUSY);
912 		goto out_opened;
913 	}
914 #ifdef FEXCL
915 	if (flags & FEXCL) {
916 		if (zv->zv_open_count != 0) {
917 			err = SET_ERROR(EBUSY);
918 			goto out_opened;
919 		}
920 		zv->zv_flags |= ZVOL_EXCL;
921 	}
922 #endif
923 
924 	zv->zv_open_count++;
925 	if (flags & (FSYNC | FDSYNC)) {
926 		zsd = &zv->zv_zso->zso_dev;
927 		zsd->zsd_sync_cnt++;
928 		if (zsd->zsd_sync_cnt == 1 &&
929 		    (zv->zv_flags & ZVOL_WRITTEN_TO) != 0)
930 			zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
931 	}
932 out_opened:
933 	if (zv->zv_open_count == 0) {
934 		zvol_last_close(zv);
935 		wakeup(zv);
936 	}
937 out_zv_locked:
938 	mutex_exit(&zv->zv_state_lock);
939 out_locked:
940 	if (drop_namespace)
941 		mutex_exit(&spa_namespace_lock);
942 	if (drop_suspend)
943 		rw_exit(&zv->zv_suspend_lock);
944 	return (err);
945 }
946 
947 static int
948 zvol_cdev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
949 {
950 	zvol_state_t *zv;
951 	struct zvol_state_dev *zsd;
952 	boolean_t drop_suspend = B_TRUE;
953 
954 	rw_enter(&zvol_state_lock, ZVOL_RW_READER);
955 	zv = dev->si_drv2;
956 	if (zv == NULL) {
957 		rw_exit(&zvol_state_lock);
958 		return (SET_ERROR(ENXIO));
959 	}
960 
961 	mutex_enter(&zv->zv_state_lock);
962 	if (zv->zv_flags & ZVOL_EXCL) {
963 		ASSERT3U(zv->zv_open_count, ==, 1);
964 		zv->zv_flags &= ~ZVOL_EXCL;
965 	}
966 
967 	ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_DEV);
968 
969 	/*
970 	 * If the open count is zero, this is a spurious close.
971 	 * That indicates a bug in the kernel / DDI framework.
972 	 */
973 	ASSERT3U(zv->zv_open_count, >, 0);
974 	/*
975 	 * make sure zvol is not suspended during last close
976 	 * (hold zv_suspend_lock) and respect proper lock acquisition
977 	 * ordering - zv_suspend_lock before zv_state_lock
978 	 */
979 	if (zv->zv_open_count == 1) {
980 		if (!rw_tryenter(&zv->zv_suspend_lock, ZVOL_RW_READER)) {
981 			mutex_exit(&zv->zv_state_lock);
982 			rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
983 			mutex_enter(&zv->zv_state_lock);
984 			/* check to see if zv_suspend_lock is needed */
985 			if (zv->zv_open_count != 1) {
986 				rw_exit(&zv->zv_suspend_lock);
987 				drop_suspend = B_FALSE;
988 			}
989 		}
990 	} else {
991 		drop_suspend = B_FALSE;
992 	}
993 	rw_exit(&zvol_state_lock);
994 
995 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
996 
997 	/*
998 	 * You may get multiple opens, but only one close.
999 	 */
1000 	zv->zv_open_count--;
1001 	if (flags & (FSYNC | FDSYNC)) {
1002 		zsd = &zv->zv_zso->zso_dev;
1003 		zsd->zsd_sync_cnt--;
1004 	}
1005 
1006 	if (zv->zv_open_count == 0) {
1007 		ASSERT(ZVOL_RW_READ_HELD(&zv->zv_suspend_lock));
1008 		zvol_last_close(zv);
1009 		wakeup(zv);
1010 	}
1011 
1012 	mutex_exit(&zv->zv_state_lock);
1013 
1014 	if (drop_suspend)
1015 		rw_exit(&zv->zv_suspend_lock);
1016 	return (0);
1017 }
1018 
1019 static int
1020 zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data,
1021     int fflag, struct thread *td)
1022 {
1023 	zvol_state_t *zv;
1024 	zfs_locked_range_t *lr;
1025 	off_t offset, length;
1026 	int i, error;
1027 	boolean_t sync;
1028 
1029 	zv = dev->si_drv2;
1030 
1031 	error = 0;
1032 	KASSERT(zv->zv_open_count > 0,
1033 	    ("Device with zero access count in %s", __func__));
1034 
1035 	i = IOCPARM_LEN(cmd);
1036 	switch (cmd) {
1037 	case DIOCGSECTORSIZE:
1038 		*(uint32_t *)data = DEV_BSIZE;
1039 		break;
1040 	case DIOCGMEDIASIZE:
1041 		*(off_t *)data = zv->zv_volsize;
1042 		break;
1043 	case DIOCGFLUSH:
1044 		rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
1045 		if (zv->zv_zilog != NULL)
1046 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1047 		rw_exit(&zv->zv_suspend_lock);
1048 		break;
1049 	case DIOCGDELETE:
1050 		if (!zvol_unmap_enabled)
1051 			break;
1052 
1053 		offset = ((off_t *)data)[0];
1054 		length = ((off_t *)data)[1];
1055 		if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
1056 		    offset < 0 || offset >= zv->zv_volsize ||
1057 		    length <= 0) {
1058 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
1059 			    length);
1060 			error = SET_ERROR(EINVAL);
1061 			break;
1062 		}
1063 		rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER);
1064 		zvol_ensure_zilog(zv);
1065 		lr = zfs_rangelock_enter(&zv->zv_rangelock, offset, length,
1066 		    RL_WRITER);
1067 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1068 		error = dmu_tx_assign(tx, TXG_WAIT);
1069 		if (error != 0) {
1070 			sync = FALSE;
1071 			dmu_tx_abort(tx);
1072 		} else {
1073 			sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1074 			zvol_log_truncate(zv, tx, offset, length, sync);
1075 			dmu_tx_commit(tx);
1076 			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1077 			    offset, length);
1078 		}
1079 		zfs_rangelock_exit(lr);
1080 		if (sync)
1081 			zil_commit(zv->zv_zilog, ZVOL_OBJ);
1082 		rw_exit(&zv->zv_suspend_lock);
1083 		break;
1084 	case DIOCGSTRIPESIZE:
1085 		*(off_t *)data = zv->zv_volblocksize;
1086 		break;
1087 	case DIOCGSTRIPEOFFSET:
1088 		*(off_t *)data = 0;
1089 		break;
1090 	case DIOCGATTR: {
1091 		spa_t *spa = dmu_objset_spa(zv->zv_objset);
1092 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
1093 		uint64_t refd, avail, usedobjs, availobjs;
1094 
1095 		if (strcmp(arg->name, "GEOM::candelete") == 0)
1096 			arg->value.i = 1;
1097 		else if (strcmp(arg->name, "blocksavail") == 0) {
1098 			dmu_objset_space(zv->zv_objset, &refd, &avail,
1099 			    &usedobjs, &availobjs);
1100 			arg->value.off = avail / DEV_BSIZE;
1101 		} else if (strcmp(arg->name, "blocksused") == 0) {
1102 			dmu_objset_space(zv->zv_objset, &refd, &avail,
1103 			    &usedobjs, &availobjs);
1104 			arg->value.off = refd / DEV_BSIZE;
1105 		} else if (strcmp(arg->name, "poolblocksavail") == 0) {
1106 			avail = metaslab_class_get_space(spa_normal_class(spa));
1107 			avail -= metaslab_class_get_alloc(
1108 			    spa_normal_class(spa));
1109 			arg->value.off = avail / DEV_BSIZE;
1110 		} else if (strcmp(arg->name, "poolblocksused") == 0) {
1111 			refd = metaslab_class_get_alloc(spa_normal_class(spa));
1112 			arg->value.off = refd / DEV_BSIZE;
1113 		} else
1114 			error = SET_ERROR(ENOIOCTL);
1115 		break;
1116 	}
1117 	case FIOSEEKHOLE:
1118 	case FIOSEEKDATA: {
1119 		off_t *off = (off_t *)data;
1120 		uint64_t noff;
1121 		boolean_t hole;
1122 
1123 		hole = (cmd == FIOSEEKHOLE);
1124 		noff = *off;
1125 		error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
1126 		*off = noff;
1127 		break;
1128 	}
1129 	default:
1130 		error = SET_ERROR(ENOIOCTL);
1131 	}
1132 
1133 	return (error);
1134 }
1135 
1136 /*
1137  * Misc. helpers
1138  */
1139 
1140 static void
1141 zvol_ensure_zilog(zvol_state_t *zv)
1142 {
1143 	ASSERT(ZVOL_RW_READ_HELD(&zv->zv_suspend_lock));
1144 
1145 	/*
1146 	 * Open a ZIL if this is the first time we have written to this
1147 	 * zvol. We protect zv->zv_zilog with zv_suspend_lock rather
1148 	 * than zv_state_lock so that we don't need to acquire an
1149 	 * additional lock in this path.
1150 	 */
1151 	if (zv->zv_zilog == NULL) {
1152 		if (!rw_tryupgrade(&zv->zv_suspend_lock)) {
1153 			rw_exit(&zv->zv_suspend_lock);
1154 			rw_enter(&zv->zv_suspend_lock, RW_WRITER);
1155 		}
1156 		if (zv->zv_zilog == NULL) {
1157 			zv->zv_zilog = zil_open(zv->zv_objset,
1158 			    zvol_get_data);
1159 			zv->zv_flags |= ZVOL_WRITTEN_TO;
1160 		}
1161 		rw_downgrade(&zv->zv_suspend_lock);
1162 	}
1163 }
1164 
1165 static boolean_t
1166 zvol_is_zvol_impl(const char *device)
1167 {
1168 	return (device && strncmp(device, ZVOL_DIR, strlen(ZVOL_DIR)) == 0);
1169 }
1170 
1171 static void
1172 zvol_rename_minor(zvol_state_t *zv, const char *newname)
1173 {
1174 	ASSERT(RW_LOCK_HELD(&zvol_state_lock));
1175 	ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1176 
1177 	/* move to new hashtable entry  */
1178 	zv->zv_hash = zvol_name_hash(zv->zv_name);
1179 	hlist_del(&zv->zv_hlink);
1180 	hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash));
1181 
1182 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
1183 		struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
1184 		struct g_provider *pp = zsg->zsg_provider;
1185 		struct g_geom *gp;
1186 
1187 		g_topology_lock();
1188 		gp = pp->geom;
1189 		ASSERT3P(gp, !=, NULL);
1190 
1191 		zsg->zsg_provider = NULL;
1192 		g_wither_provider(pp, ENXIO);
1193 
1194 		pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
1195 		pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
1196 		pp->sectorsize = DEV_BSIZE;
1197 		pp->mediasize = zv->zv_volsize;
1198 		pp->private = zv;
1199 		zsg->zsg_provider = pp;
1200 		g_error_provider(pp, 0);
1201 		g_topology_unlock();
1202 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
1203 		struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev;
1204 		struct cdev *dev;
1205 		struct make_dev_args args;
1206 
1207 		dev = zsd->zsd_cdev;
1208 		if (dev != NULL) {
1209 			destroy_dev(dev);
1210 			dev = zsd->zsd_cdev = NULL;
1211 			if (zv->zv_open_count > 0) {
1212 				zv->zv_flags &= ~ZVOL_EXCL;
1213 				zv->zv_open_count = 0;
1214 				/* XXX  need suspend lock but lock order */
1215 				zvol_last_close(zv);
1216 			}
1217 		}
1218 
1219 		make_dev_args_init(&args);
1220 		args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
1221 		args.mda_devsw = &zvol_cdevsw;
1222 		args.mda_cr = NULL;
1223 		args.mda_uid = UID_ROOT;
1224 		args.mda_gid = GID_OPERATOR;
1225 		args.mda_mode = 0640;
1226 		args.mda_si_drv2 = zv;
1227 		if (make_dev_s(&args, &dev, "%s/%s", ZVOL_DRIVER, newname)
1228 		    == 0) {
1229 			dev->si_iosize_max = maxphys;
1230 			zsd->zsd_cdev = dev;
1231 		}
1232 	}
1233 	strlcpy(zv->zv_name, newname, sizeof (zv->zv_name));
1234 }
1235 
1236 /*
1237  * Remove minor node for the specified volume.
1238  */
1239 static void
1240 zvol_free(zvol_state_t *zv)
1241 {
1242 	ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock));
1243 	ASSERT(!MUTEX_HELD(&zv->zv_state_lock));
1244 	ASSERT0(zv->zv_open_count);
1245 
1246 	ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
1247 
1248 	rw_destroy(&zv->zv_suspend_lock);
1249 	zfs_rangelock_fini(&zv->zv_rangelock);
1250 
1251 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
1252 		struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
1253 		struct g_provider *pp __maybe_unused = zsg->zsg_provider;
1254 
1255 		ASSERT3P(pp->private, ==, NULL);
1256 
1257 		g_topology_lock();
1258 		zvol_geom_destroy(zv);
1259 		g_topology_unlock();
1260 		mtx_destroy(&zsg->zsg_queue_mtx);
1261 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
1262 		struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev;
1263 		struct cdev *dev = zsd->zsd_cdev;
1264 
1265 		ASSERT3P(dev->si_drv2, ==, NULL);
1266 
1267 		destroy_dev(dev);
1268 	}
1269 
1270 	mutex_destroy(&zv->zv_state_lock);
1271 	dataset_kstats_destroy(&zv->zv_kstat);
1272 	kmem_free(zv->zv_zso, sizeof (struct zvol_state_os));
1273 	kmem_free(zv, sizeof (zvol_state_t));
1274 	zvol_minors--;
1275 }
1276 
1277 /*
1278  * Create a minor node (plus a whole lot more) for the specified volume.
1279  */
1280 static int
1281 zvol_create_minor_impl(const char *name)
1282 {
1283 	zvol_state_t *zv;
1284 	objset_t *os;
1285 	dmu_object_info_t *doi;
1286 	uint64_t volsize;
1287 	uint64_t volmode, hash;
1288 	int error;
1289 
1290 	ZFS_LOG(1, "Creating ZVOL %s...", name);
1291 	hash = zvol_name_hash(name);
1292 	if ((zv = zvol_find_by_name_hash(name, hash, RW_NONE)) != NULL) {
1293 		ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1294 		mutex_exit(&zv->zv_state_lock);
1295 		return (SET_ERROR(EEXIST));
1296 	}
1297 
1298 	DROP_GIANT();
1299 
1300 	doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);
1301 
1302 	/* lie and say we're read-only */
1303 	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, B_TRUE, FTAG, &os);
1304 	if (error)
1305 		goto out_doi;
1306 
1307 	error = dmu_object_info(os, ZVOL_OBJ, doi);
1308 	if (error)
1309 		goto out_dmu_objset_disown;
1310 
1311 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
1312 	if (error)
1313 		goto out_dmu_objset_disown;
1314 
1315 	error = dsl_prop_get_integer(name,
1316 	    zfs_prop_to_name(ZFS_PROP_VOLMODE), &volmode, NULL);
1317 	if (error || volmode == ZFS_VOLMODE_DEFAULT)
1318 		volmode = zvol_volmode;
1319 	error = 0;
1320 
1321 	/*
1322 	 * zvol_alloc equivalent ...
1323 	 */
1324 	zv = kmem_zalloc(sizeof (*zv), KM_SLEEP);
1325 	zv->zv_hash = hash;
1326 	mutex_init(&zv->zv_state_lock, NULL, MUTEX_DEFAULT, NULL);
1327 	zv->zv_zso = kmem_zalloc(sizeof (struct zvol_state_os), KM_SLEEP);
1328 	zv->zv_volmode = volmode;
1329 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
1330 		struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
1331 		struct g_provider *pp;
1332 		struct g_geom *gp;
1333 
1334 		zsg->zsg_state = ZVOL_GEOM_UNINIT;
1335 		mtx_init(&zsg->zsg_queue_mtx, "zvol", NULL, MTX_DEF);
1336 
1337 		g_topology_lock();
1338 		gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
1339 		gp->start = zvol_geom_bio_start;
1340 		gp->access = zvol_geom_access;
1341 		pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
1342 		pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
1343 		pp->sectorsize = DEV_BSIZE;
1344 		pp->mediasize = 0;
1345 		pp->private = zv;
1346 
1347 		zsg->zsg_provider = pp;
1348 		bioq_init(&zsg->zsg_queue);
1349 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
1350 		struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev;
1351 		struct cdev *dev;
1352 		struct make_dev_args args;
1353 
1354 		make_dev_args_init(&args);
1355 		args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
1356 		args.mda_devsw = &zvol_cdevsw;
1357 		args.mda_cr = NULL;
1358 		args.mda_uid = UID_ROOT;
1359 		args.mda_gid = GID_OPERATOR;
1360 		args.mda_mode = 0640;
1361 		args.mda_si_drv2 = zv;
1362 		error = make_dev_s(&args, &dev, "%s/%s", ZVOL_DRIVER, name);
1363 		if (error) {
1364 			kmem_free(zv->zv_zso, sizeof (struct zvol_state_os));
1365 			mutex_destroy(&zv->zv_state_lock);
1366 			kmem_free(zv, sizeof (*zv));
1367 			dmu_objset_disown(os, B_TRUE, FTAG);
1368 			goto out_doi;
1369 		}
1370 		dev->si_iosize_max = maxphys;
1371 		zsd->zsd_cdev = dev;
1372 	}
1373 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
1374 	rw_init(&zv->zv_suspend_lock, NULL, RW_DEFAULT, NULL);
1375 	zfs_rangelock_init(&zv->zv_rangelock, NULL, NULL);
1376 
1377 	if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
1378 		zv->zv_flags |= ZVOL_RDONLY;
1379 
1380 	zv->zv_volblocksize = doi->doi_data_block_size;
1381 	zv->zv_volsize = volsize;
1382 	zv->zv_objset = os;
1383 
1384 	if (spa_writeable(dmu_objset_spa(os))) {
1385 		if (zil_replay_disable)
1386 			zil_destroy(dmu_objset_zil(os), B_FALSE);
1387 		else
1388 			zil_replay(os, zv, zvol_replay_vector);
1389 	}
1390 	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
1391 	dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
1392 
1393 	/* TODO: prefetch for geom tasting */
1394 
1395 	zv->zv_objset = NULL;
1396 out_dmu_objset_disown:
1397 	dmu_objset_disown(os, B_TRUE, FTAG);
1398 
1399 	if (error == 0 && volmode == ZFS_VOLMODE_GEOM) {
1400 		zvol_geom_run(zv);
1401 		g_topology_unlock();
1402 	}
1403 out_doi:
1404 	kmem_free(doi, sizeof (dmu_object_info_t));
1405 	if (error == 0) {
1406 		rw_enter(&zvol_state_lock, RW_WRITER);
1407 		zvol_insert(zv);
1408 		zvol_minors++;
1409 		rw_exit(&zvol_state_lock);
1410 		ZFS_LOG(1, "ZVOL %s created.", name);
1411 	}
1412 	PICKUP_GIANT();
1413 	return (error);
1414 }
1415 
1416 static void
1417 zvol_clear_private(zvol_state_t *zv)
1418 {
1419 	ASSERT(RW_LOCK_HELD(&zvol_state_lock));
1420 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
1421 		struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
1422 		struct g_provider *pp = zsg->zsg_provider;
1423 
1424 		if (pp->private == NULL) /* already cleared */
1425 			return;
1426 
1427 		mtx_lock(&zsg->zsg_queue_mtx);
1428 		zsg->zsg_state = ZVOL_GEOM_STOPPED;
1429 		pp->private = NULL;
1430 		wakeup_one(&zsg->zsg_queue);
1431 		while (zsg->zsg_state != ZVOL_GEOM_RUNNING)
1432 			msleep(&zsg->zsg_state, &zsg->zsg_queue_mtx,
1433 			    0, "zvol:w", 0);
1434 		mtx_unlock(&zsg->zsg_queue_mtx);
1435 		ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock));
1436 	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
1437 		struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev;
1438 		struct cdev *dev = zsd->zsd_cdev;
1439 
1440 		dev->si_drv2 = NULL;
1441 	}
1442 }
1443 
1444 static int
1445 zvol_update_volsize(zvol_state_t *zv, uint64_t volsize)
1446 {
1447 	zv->zv_volsize = volsize;
1448 	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
1449 		struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom;
1450 		struct g_provider *pp = zsg->zsg_provider;
1451 
1452 		g_topology_lock();
1453 
1454 		if (pp->private == NULL) {
1455 			g_topology_unlock();
1456 			return (SET_ERROR(ENXIO));
1457 		}
1458 
1459 		/*
1460 		 * Do not invoke resize event when initial size was zero.
1461 		 * ZVOL initializes the size on first open, this is not
1462 		 * real resizing.
1463 		 */
1464 		if (pp->mediasize == 0)
1465 			pp->mediasize = zv->zv_volsize;
1466 		else
1467 			g_resize_provider(pp, zv->zv_volsize);
1468 
1469 		g_topology_unlock();
1470 	}
1471 	return (0);
1472 }
1473 
1474 static void
1475 zvol_set_disk_ro_impl(zvol_state_t *zv, int flags)
1476 {
1477 	// XXX? set_disk_ro(zv->zv_zso->zvo_disk, flags);
1478 }
1479 
1480 static void
1481 zvol_set_capacity_impl(zvol_state_t *zv, uint64_t capacity)
1482 {
1483 	// XXX? set_capacity(zv->zv_zso->zvo_disk, capacity);
1484 }
1485 
1486 const static zvol_platform_ops_t zvol_freebsd_ops = {
1487 	.zv_free = zvol_free,
1488 	.zv_rename_minor = zvol_rename_minor,
1489 	.zv_create_minor = zvol_create_minor_impl,
1490 	.zv_update_volsize = zvol_update_volsize,
1491 	.zv_clear_private = zvol_clear_private,
1492 	.zv_is_zvol = zvol_is_zvol_impl,
1493 	.zv_set_disk_ro = zvol_set_disk_ro_impl,
1494 	.zv_set_capacity = zvol_set_capacity_impl,
1495 };
1496 
1497 /*
1498  * Public interfaces
1499  */
1500 
1501 int
1502 zvol_busy(void)
1503 {
1504 	return (zvol_minors != 0);
1505 }
1506 
1507 int
1508 zvol_init(void)
1509 {
1510 	zvol_init_impl();
1511 	zvol_register_ops(&zvol_freebsd_ops);
1512 	return (0);
1513 }
1514 
1515 void
1516 zvol_fini(void)
1517 {
1518 	zvol_fini_impl();
1519 }
1520