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.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
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]
22 * Copyright (C) 2008-2010 Lawrence Livermore National Security, LLC.
23 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
24 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
27 * ZFS volume emulation driver.
29 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
30 * Volumes are accessed through the symbolic links named:
32 * /dev/<pool_name>/<dataset_name>
34 * Volumes are persistent through reboot and module load. No user command
35 * needs to be run before opening and using a device.
37 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
38 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
39 * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
43 * Note on locking of zvol state structures.
45 * These structures are used to maintain internal state used to emulate block
46 * devices on top of zvols. In particular, management of device minor number
47 * operations - create, remove, rename, and set_snapdev - involves access to
48 * these structures. The zvol_state_lock is primarily used to protect the
49 * zvol_state_list. The zv->zv_state_lock is used to protect the contents
50 * of the zvol_state_t structures, as well as to make sure that when the
51 * time comes to remove the structure from the list, it is not in use, and
52 * therefore, it can be taken off zvol_state_list and freed.
54 * The zv_suspend_lock was introduced to allow for suspending I/O to a zvol,
55 * e.g. for the duration of receive and rollback operations. This lock can be
56 * held for significant periods of time. Given that it is undesirable to hold
57 * mutexes for long periods of time, the following lock ordering applies:
58 * - take zvol_state_lock if necessary, to protect zvol_state_list
59 * - take zv_suspend_lock if necessary, by the code path in question
60 * - take zv_state_lock to protect zvol_state_t
62 * The minor operations are issued to spa->spa_zvol_taskq queues, that are
63 * single-threaded (to preserve order of minor operations), and are executed
64 * through the zvol_task_cb that dispatches the specific operations. Therefore,
65 * these operations are serialized per pool. Consequently, we can be certain
66 * that for a given zvol, there is only one operation at a time in progress.
67 * That is why one can be sure that first, zvol_state_t for a given zvol is
68 * allocated and placed on zvol_state_list, and then other minor operations
69 * for this zvol are going to proceed in the order of issue.
73 #include <sys/dataset_kstats.h>
75 #include <sys/dmu_traverse.h>
76 #include <sys/dsl_dataset.h>
77 #include <sys/dsl_prop.h>
78 #include <sys/dsl_dir.h>
80 #include <sys/zfeature.h>
81 #include <sys/zil_impl.h>
82 #include <sys/dmu_tx.h>
84 #include <sys/zfs_rlock.h>
85 #include <sys/spa_impl.h>
87 #include <sys/zvol_impl.h>
89 unsigned int zvol_inhibit_dev = 0;
90 unsigned int zvol_volmode = ZFS_VOLMODE_GEOM;
92 struct hlist_head *zvol_htable;
93 static list_t zvol_state_list;
94 krwlock_t zvol_state_lock;
97 ZVOL_ASYNC_REMOVE_MINORS,
98 ZVOL_ASYNC_RENAME_MINORS,
99 ZVOL_ASYNC_SET_SNAPDEV,
100 ZVOL_ASYNC_SET_VOLMODE,
106 char name1[MAXNAMELEN];
107 char name2[MAXNAMELEN];
112 zvol_name_hash(const char *name)
115 uint64_t crc = -1ULL;
116 const uint8_t *p = (const uint8_t *)name;
117 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
118 for (i = 0; i < MAXNAMELEN - 1 && *p; i++, p++) {
119 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (*p)) & 0xFF];
125 * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
126 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
127 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
128 * before zv_state_lock. The mode argument indicates the mode (including none)
129 * for zv_suspend_lock to be taken.
132 zvol_find_by_name_hash(const char *name, uint64_t hash, int mode)
135 struct hlist_node *p = NULL;
137 rw_enter(&zvol_state_lock, RW_READER);
138 hlist_for_each(p, ZVOL_HT_HEAD(hash)) {
139 zv = hlist_entry(p, zvol_state_t, zv_hlink);
140 mutex_enter(&zv->zv_state_lock);
141 if (zv->zv_hash == hash &&
142 strncmp(zv->zv_name, name, MAXNAMELEN) == 0) {
144 * this is the right zvol, take the locks in the
147 if (mode != RW_NONE &&
148 !rw_tryenter(&zv->zv_suspend_lock, mode)) {
149 mutex_exit(&zv->zv_state_lock);
150 rw_enter(&zv->zv_suspend_lock, mode);
151 mutex_enter(&zv->zv_state_lock);
153 * zvol cannot be renamed as we continue
154 * to hold zvol_state_lock
156 ASSERT(zv->zv_hash == hash &&
157 strncmp(zv->zv_name, name, MAXNAMELEN)
160 rw_exit(&zvol_state_lock);
163 mutex_exit(&zv->zv_state_lock);
165 rw_exit(&zvol_state_lock);
171 * Find a zvol_state_t given the name.
172 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
173 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
174 * before zv_state_lock. The mode argument indicates the mode (including none)
175 * for zv_suspend_lock to be taken.
177 static zvol_state_t *
178 zvol_find_by_name(const char *name, int mode)
180 return (zvol_find_by_name_hash(name, zvol_name_hash(name), mode));
184 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
187 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
189 zfs_creat_t *zct = arg;
190 nvlist_t *nvprops = zct->zct_props;
192 uint64_t volblocksize, volsize;
194 VERIFY(nvlist_lookup_uint64(nvprops,
195 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
196 if (nvlist_lookup_uint64(nvprops,
197 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
198 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
201 * These properties must be removed from the list so the generic
202 * property setting step won't apply to them.
204 VERIFY(nvlist_remove_all(nvprops,
205 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
206 (void) nvlist_remove_all(nvprops,
207 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
209 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
213 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
217 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
222 * ZFS_IOC_OBJSET_STATS entry point.
225 zvol_get_stats(objset_t *os, nvlist_t *nv)
228 dmu_object_info_t *doi;
231 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
233 return (SET_ERROR(error));
235 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
236 doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);
237 error = dmu_object_info(os, ZVOL_OBJ, doi);
240 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
241 doi->doi_data_block_size);
244 kmem_free(doi, sizeof (dmu_object_info_t));
246 return (SET_ERROR(error));
250 * Sanity check volume size.
253 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
256 return (SET_ERROR(EINVAL));
258 if (volsize % blocksize != 0)
259 return (SET_ERROR(EINVAL));
262 if (volsize - 1 > SPEC_MAXOFFSET_T)
263 return (SET_ERROR(EOVERFLOW));
269 * Ensure the zap is flushed then inform the VFS of the capacity change.
272 zvol_update_volsize(uint64_t volsize, objset_t *os)
278 tx = dmu_tx_create(os);
279 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
280 dmu_tx_mark_netfree(tx);
281 error = dmu_tx_assign(tx, TXG_WAIT);
284 return (SET_ERROR(error));
286 txg = dmu_tx_get_txg(tx);
288 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
292 txg_wait_synced(dmu_objset_pool(os), txg);
295 error = dmu_free_long_range(os,
296 ZVOL_OBJ, volsize, DMU_OBJECT_END);
302 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume
303 * size will result in a udev "change" event being generated.
306 zvol_set_volsize(const char *name, uint64_t volsize)
311 boolean_t owned = B_FALSE;
313 error = dsl_prop_get_integer(name,
314 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
316 return (SET_ERROR(error));
318 return (SET_ERROR(EROFS));
320 zvol_state_t *zv = zvol_find_by_name(name, RW_READER);
322 ASSERT(zv == NULL || (MUTEX_HELD(&zv->zv_state_lock) &&
323 RW_READ_HELD(&zv->zv_suspend_lock)));
325 if (zv == NULL || zv->zv_objset == NULL) {
327 rw_exit(&zv->zv_suspend_lock);
328 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE,
331 mutex_exit(&zv->zv_state_lock);
332 return (SET_ERROR(error));
341 dmu_object_info_t *doi = kmem_alloc(sizeof (*doi), KM_SLEEP);
343 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) ||
344 (error = zvol_check_volsize(volsize, doi->doi_data_block_size)))
347 error = zvol_update_volsize(volsize, os);
348 if (error == 0 && zv != NULL) {
349 zv->zv_volsize = volsize;
353 kmem_free(doi, sizeof (dmu_object_info_t));
356 dmu_objset_disown(os, B_TRUE, FTAG);
358 zv->zv_objset = NULL;
360 rw_exit(&zv->zv_suspend_lock);
364 mutex_exit(&zv->zv_state_lock);
366 if (error == 0 && zv != NULL)
367 zvol_os_update_volsize(zv, volsize);
369 return (SET_ERROR(error));
373 * Update volthreading.
376 zvol_set_volthreading(const char *name, boolean_t value)
378 zvol_state_t *zv = zvol_find_by_name(name, RW_NONE);
381 zv->zv_threading = value;
382 mutex_exit(&zv->zv_state_lock);
387 * Update zvol ro property.
390 zvol_set_ro(const char *name, boolean_t value)
392 zvol_state_t *zv = zvol_find_by_name(name, RW_NONE);
396 zvol_os_set_disk_ro(zv, 1);
397 zv->zv_flags |= ZVOL_RDONLY;
399 zvol_os_set_disk_ro(zv, 0);
400 zv->zv_flags &= ~ZVOL_RDONLY;
402 mutex_exit(&zv->zv_state_lock);
407 * Sanity check volume block size.
410 zvol_check_volblocksize(const char *name, uint64_t volblocksize)
412 /* Record sizes above 128k need the feature to be enabled */
413 if (volblocksize > SPA_OLD_MAXBLOCKSIZE) {
417 if ((error = spa_open(name, &spa, FTAG)) != 0)
420 if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
421 spa_close(spa, FTAG);
422 return (SET_ERROR(ENOTSUP));
426 * We don't allow setting the property above 1MB,
427 * unless the tunable has been changed.
429 if (volblocksize > zfs_max_recordsize)
430 return (SET_ERROR(EDOM));
432 spa_close(spa, FTAG);
435 if (volblocksize < SPA_MINBLOCKSIZE ||
436 volblocksize > SPA_MAXBLOCKSIZE ||
438 return (SET_ERROR(EDOM));
444 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
445 * implement DKIOCFREE/free-long-range.
448 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
450 zvol_state_t *zv = arg1;
451 lr_truncate_t *lr = arg2;
452 uint64_t offset, length;
455 byteswap_uint64_array(lr, sizeof (*lr));
457 offset = lr->lr_offset;
458 length = lr->lr_length;
460 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
461 dmu_tx_mark_netfree(tx);
462 int error = dmu_tx_assign(tx, TXG_WAIT);
466 (void) zil_replaying(zv->zv_zilog, tx);
468 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset,
476 * Replay a TX_WRITE ZIL transaction that didn't get committed
477 * after a system failure
480 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
482 zvol_state_t *zv = arg1;
483 lr_write_t *lr = arg2;
484 objset_t *os = zv->zv_objset;
485 char *data = (char *)(lr + 1); /* data follows lr_write_t */
486 uint64_t offset, length;
491 byteswap_uint64_array(lr, sizeof (*lr));
493 offset = lr->lr_offset;
494 length = lr->lr_length;
496 /* If it's a dmu_sync() block, write the whole block */
497 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
498 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
499 if (length < blocksize) {
500 offset -= offset % blocksize;
505 tx = dmu_tx_create(os);
506 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
507 error = dmu_tx_assign(tx, TXG_WAIT);
511 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
512 (void) zil_replaying(zv->zv_zilog, tx);
520 * Replay a TX_CLONE_RANGE ZIL transaction that didn't get committed
521 * after a system failure.
523 * TODO: For now we drop block cloning transations for ZVOLs as they are
524 * unsupported, but we still need to inform BRT about that as we
525 * claimed them during pool import.
526 * This situation can occur when we try to import a pool from a ZFS
527 * version supporting block cloning for ZVOLs into a system that
528 * has this ZFS version, that doesn't support block cloning for ZVOLs.
531 zvol_replay_clone_range(void *arg1, void *arg2, boolean_t byteswap)
533 char name[ZFS_MAX_DATASET_NAME_LEN];
534 zvol_state_t *zv = arg1;
535 objset_t *os = zv->zv_objset;
536 lr_clone_range_t *lr = arg2;
543 dmu_objset_name(os, name);
544 cmn_err(CE_WARN, "ZFS dropping block cloning transaction for %s.",
548 byteswap_uint64_array(lr, sizeof (*lr));
550 tx = dmu_tx_create(os);
551 error = dmu_tx_assign(tx, TXG_WAIT);
559 for (ii = 0; ii < lr->lr_nbps; ii++) {
560 bp = &lr->lr_bps[ii];
562 if (!BP_IS_HOLE(bp)) {
563 zio_free(spa, dmu_tx_get_txg(tx), bp);
567 (void) zil_replaying(zv->zv_zilog, tx);
574 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
576 (void) arg1, (void) arg2, (void) byteswap;
577 return (SET_ERROR(ENOTSUP));
581 * Callback vectors for replaying records.
582 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
584 zil_replay_func_t *const zvol_replay_vector[TX_MAX_TYPE] = {
585 zvol_replay_err, /* no such transaction type */
586 zvol_replay_err, /* TX_CREATE */
587 zvol_replay_err, /* TX_MKDIR */
588 zvol_replay_err, /* TX_MKXATTR */
589 zvol_replay_err, /* TX_SYMLINK */
590 zvol_replay_err, /* TX_REMOVE */
591 zvol_replay_err, /* TX_RMDIR */
592 zvol_replay_err, /* TX_LINK */
593 zvol_replay_err, /* TX_RENAME */
594 zvol_replay_write, /* TX_WRITE */
595 zvol_replay_truncate, /* TX_TRUNCATE */
596 zvol_replay_err, /* TX_SETATTR */
597 zvol_replay_err, /* TX_ACL */
598 zvol_replay_err, /* TX_CREATE_ATTR */
599 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
600 zvol_replay_err, /* TX_MKDIR_ACL */
601 zvol_replay_err, /* TX_MKDIR_ATTR */
602 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
603 zvol_replay_err, /* TX_WRITE2 */
604 zvol_replay_err, /* TX_SETSAXATTR */
605 zvol_replay_err, /* TX_RENAME_EXCHANGE */
606 zvol_replay_err, /* TX_RENAME_WHITEOUT */
607 zvol_replay_clone_range /* TX_CLONE_RANGE */
611 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
613 * We store data in the log buffers if it's small enough.
614 * Otherwise we will later flush the data out via dmu_sync().
616 static const ssize_t zvol_immediate_write_sz = 32768;
619 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
620 uint64_t size, boolean_t commit)
622 uint32_t blocksize = zv->zv_volblocksize;
623 zilog_t *zilog = zv->zv_zilog;
624 itx_wr_state_t write_state;
627 if (zil_replaying(zilog, tx))
630 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
631 write_state = WR_INDIRECT;
632 else if (!spa_has_slogs(zilog->zl_spa) &&
633 size >= blocksize && blocksize > zvol_immediate_write_sz)
634 write_state = WR_INDIRECT;
636 write_state = WR_COPIED;
638 write_state = WR_NEED_COPY;
643 itx_wr_state_t wr_state = write_state;
646 if (wr_state == WR_COPIED && size > zil_max_copied_data(zilog))
647 wr_state = WR_NEED_COPY;
648 else if (wr_state == WR_INDIRECT)
649 len = MIN(blocksize - P2PHASE(offset, blocksize), size);
651 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
652 (wr_state == WR_COPIED ? len : 0));
653 lr = (lr_write_t *)&itx->itx_lr;
654 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
655 offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
656 zil_itx_destroy(itx);
657 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
658 lr = (lr_write_t *)&itx->itx_lr;
659 wr_state = WR_NEED_COPY;
662 itx->itx_wr_state = wr_state;
663 lr->lr_foid = ZVOL_OBJ;
664 lr->lr_offset = offset;
667 BP_ZERO(&lr->lr_blkptr);
669 itx->itx_private = zv;
671 (void) zil_itx_assign(zilog, itx, tx);
677 if (write_state == WR_COPIED || write_state == WR_NEED_COPY) {
678 dsl_pool_wrlog_count(zilog->zl_dmu_pool, sz, tx->tx_txg);
683 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
686 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len)
690 zilog_t *zilog = zv->zv_zilog;
692 if (zil_replaying(zilog, tx))
695 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
696 lr = (lr_truncate_t *)&itx->itx_lr;
697 lr->lr_foid = ZVOL_OBJ;
701 zil_itx_assign(zilog, itx, tx);
706 zvol_get_done(zgd_t *zgd, int error)
710 dmu_buf_rele(zgd->zgd_db, zgd);
712 zfs_rangelock_exit(zgd->zgd_lr);
714 kmem_free(zgd, sizeof (zgd_t));
718 * Get data to generate a TX_WRITE intent log record.
721 zvol_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
722 struct lwb *lwb, zio_t *zio)
724 zvol_state_t *zv = arg;
725 uint64_t offset = lr->lr_offset;
726 uint64_t size = lr->lr_length;
731 ASSERT3P(lwb, !=, NULL);
732 ASSERT3U(size, !=, 0);
734 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
738 * Write records come in two flavors: immediate and indirect.
739 * For small writes it's cheaper to store the data with the
740 * log record (immediate); for large writes it's cheaper to
741 * sync the data and get a pointer to it (indirect) so that
742 * we don't have to write the data twice.
744 if (buf != NULL) { /* immediate write */
745 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
747 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
748 DMU_READ_NO_PREFETCH);
749 } else { /* indirect write */
750 ASSERT3P(zio, !=, NULL);
752 * Have to lock the whole block to ensure when it's written out
753 * and its checksum is being calculated that no one can change
754 * the data. Contrarily to zfs_get_data we need not re-check
755 * blocksize after we get the lock because it cannot be changed.
757 size = zv->zv_volblocksize;
758 offset = P2ALIGN_TYPED(offset, size, uint64_t);
759 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
761 error = dmu_buf_hold_noread_by_dnode(zv->zv_dn, offset, zgd,
764 blkptr_t *bp = &lr->lr_blkptr;
770 ASSERT(db->db_offset == offset);
771 ASSERT(db->db_size == size);
773 error = dmu_sync(zio, lr->lr_common.lrc_txg,
781 zvol_get_done(zgd, error);
783 return (SET_ERROR(error));
787 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
791 zvol_insert(zvol_state_t *zv)
793 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
794 list_insert_head(&zvol_state_list, zv);
795 hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash));
799 * Simply remove the zvol from to list of zvols.
802 zvol_remove(zvol_state_t *zv)
804 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
805 list_remove(&zvol_state_list, zv);
806 hlist_del(&zv->zv_hlink);
810 * Setup zv after we just own the zv->objset
813 zvol_setup_zv(zvol_state_t *zv)
818 objset_t *os = zv->zv_objset;
820 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
821 ASSERT(RW_LOCK_HELD(&zv->zv_suspend_lock));
824 zv->zv_flags &= ~ZVOL_WRITTEN_TO;
826 error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL);
828 return (SET_ERROR(error));
830 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
832 return (SET_ERROR(error));
834 error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn);
836 return (SET_ERROR(error));
838 zvol_os_set_capacity(zv, volsize >> 9);
839 zv->zv_volsize = volsize;
841 if (ro || dmu_objset_is_snapshot(os) ||
842 !spa_writeable(dmu_objset_spa(os))) {
843 zvol_os_set_disk_ro(zv, 1);
844 zv->zv_flags |= ZVOL_RDONLY;
846 zvol_os_set_disk_ro(zv, 0);
847 zv->zv_flags &= ~ZVOL_RDONLY;
853 * Shutdown every zv_objset related stuff except zv_objset itself.
854 * The is the reverse of zvol_setup_zv.
857 zvol_shutdown_zv(zvol_state_t *zv)
859 ASSERT(MUTEX_HELD(&zv->zv_state_lock) &&
860 RW_LOCK_HELD(&zv->zv_suspend_lock));
862 if (zv->zv_flags & ZVOL_WRITTEN_TO) {
863 ASSERT(zv->zv_zilog != NULL);
864 zil_close(zv->zv_zilog);
869 dnode_rele(zv->zv_dn, zv);
873 * Evict cached data. We must write out any dirty data before
874 * disowning the dataset.
876 if (zv->zv_flags & ZVOL_WRITTEN_TO)
877 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
878 (void) dmu_objset_evict_dbufs(zv->zv_objset);
882 * return the proper tag for rollback and recv
885 zvol_tag(zvol_state_t *zv)
887 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
888 return (zv->zv_open_count > 0 ? zv : NULL);
892 * Suspend the zvol for recv and rollback.
895 zvol_suspend(const char *name)
899 zv = zvol_find_by_name(name, RW_WRITER);
904 /* block all I/O, release in zvol_resume. */
905 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
906 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
908 atomic_inc(&zv->zv_suspend_ref);
910 if (zv->zv_open_count > 0)
911 zvol_shutdown_zv(zv);
914 * do not hold zv_state_lock across suspend/resume to
915 * avoid locking up zvol lookups
917 mutex_exit(&zv->zv_state_lock);
919 /* zv_suspend_lock is released in zvol_resume() */
924 zvol_resume(zvol_state_t *zv)
928 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
930 mutex_enter(&zv->zv_state_lock);
932 if (zv->zv_open_count > 0) {
933 VERIFY0(dmu_objset_hold(zv->zv_name, zv, &zv->zv_objset));
934 VERIFY3P(zv->zv_objset->os_dsl_dataset->ds_owner, ==, zv);
935 VERIFY(dsl_dataset_long_held(zv->zv_objset->os_dsl_dataset));
936 dmu_objset_rele(zv->zv_objset, zv);
938 error = zvol_setup_zv(zv);
941 mutex_exit(&zv->zv_state_lock);
943 rw_exit(&zv->zv_suspend_lock);
945 * We need this because we don't hold zvol_state_lock while releasing
946 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
947 * zv_suspend_lock to determine it is safe to free because rwlock is
948 * not inherent atomic.
950 atomic_dec(&zv->zv_suspend_ref);
952 return (SET_ERROR(error));
956 zvol_first_open(zvol_state_t *zv, boolean_t readonly)
961 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
962 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
963 ASSERT(mutex_owned(&spa_namespace_lock));
965 boolean_t ro = (readonly || (strchr(zv->zv_name, '@') != NULL));
966 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os);
968 return (SET_ERROR(error));
972 error = zvol_setup_zv(zv);
974 dmu_objset_disown(os, 1, zv);
975 zv->zv_objset = NULL;
982 zvol_last_close(zvol_state_t *zv)
984 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
985 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
987 zvol_shutdown_zv(zv);
989 dmu_objset_disown(zv->zv_objset, 1, zv);
990 zv->zv_objset = NULL;
993 typedef struct minors_job {
1003 * Prefetch zvol dnodes for the minors_job
1006 zvol_prefetch_minors_impl(void *arg)
1008 minors_job_t *job = arg;
1009 char *dsname = job->name;
1010 objset_t *os = NULL;
1012 job->error = dmu_objset_own(dsname, DMU_OST_ZVOL, B_TRUE, B_TRUE,
1014 if (job->error == 0) {
1015 dmu_prefetch_dnode(os, ZVOL_OBJ, ZIO_PRIORITY_SYNC_READ);
1016 dmu_objset_disown(os, B_TRUE, FTAG);
1021 * Mask errors to continue dmu_objset_find() traversal
1024 zvol_create_snap_minor_cb(const char *dsname, void *arg)
1026 minors_job_t *j = arg;
1027 list_t *minors_list = j->list;
1028 const char *name = j->name;
1030 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1032 /* skip the designated dataset */
1033 if (name && strcmp(dsname, name) == 0)
1036 /* at this point, the dsname should name a snapshot */
1037 if (strchr(dsname, '@') == 0) {
1038 dprintf("zvol_create_snap_minor_cb(): "
1039 "%s is not a snapshot name\n", dsname);
1042 char *n = kmem_strdup(dsname);
1046 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1048 job->list = minors_list;
1050 list_insert_tail(minors_list, job);
1051 /* don't care if dispatch fails, because job->error is 0 */
1052 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1060 * If spa_keystore_load_wkey() is called for an encrypted zvol,
1061 * we need to look for any clones also using the key. This function
1062 * is "best effort" - so we just skip over it if there are failures.
1065 zvol_add_clones(const char *dsname, list_t *minors_list)
1067 /* Also check if it has clones */
1068 dsl_dir_t *dd = NULL;
1069 dsl_pool_t *dp = NULL;
1071 if (dsl_pool_hold(dsname, FTAG, &dp) != 0)
1074 if (!spa_feature_is_enabled(dp->dp_spa,
1075 SPA_FEATURE_ENCRYPTION))
1078 if (dsl_dir_hold(dp, dsname, FTAG, &dd, NULL) != 0)
1081 if (dsl_dir_phys(dd)->dd_clones == 0)
1084 zap_cursor_t *zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
1085 zap_attribute_t *za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1086 objset_t *mos = dd->dd_pool->dp_meta_objset;
1088 for (zap_cursor_init(zc, mos, dsl_dir_phys(dd)->dd_clones);
1089 zap_cursor_retrieve(zc, za) == 0;
1090 zap_cursor_advance(zc)) {
1091 dsl_dataset_t *clone;
1094 if (dsl_dataset_hold_obj(dd->dd_pool,
1095 za->za_first_integer, FTAG, &clone) == 0) {
1097 char name[ZFS_MAX_DATASET_NAME_LEN];
1098 dsl_dataset_name(clone, name);
1100 char *n = kmem_strdup(name);
1101 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1103 job->list = minors_list;
1105 list_insert_tail(minors_list, job);
1107 dsl_dataset_rele(clone, FTAG);
1110 zap_cursor_fini(zc);
1111 kmem_free(za, sizeof (zap_attribute_t));
1112 kmem_free(zc, sizeof (zap_cursor_t));
1116 dsl_dir_rele(dd, FTAG);
1117 dsl_pool_rele(dp, FTAG);
1121 * Mask errors to continue dmu_objset_find() traversal
1124 zvol_create_minors_cb(const char *dsname, void *arg)
1128 list_t *minors_list = arg;
1130 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1132 error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL);
1137 * Given the name and the 'snapdev' property, create device minor nodes
1138 * with the linkages to zvols/snapshots as needed.
1139 * If the name represents a zvol, create a minor node for the zvol, then
1140 * check if its snapshots are 'visible', and if so, iterate over the
1141 * snapshots and create device minor nodes for those.
1143 if (strchr(dsname, '@') == 0) {
1145 char *n = kmem_strdup(dsname);
1149 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1151 job->list = minors_list;
1153 list_insert_tail(minors_list, job);
1154 /* don't care if dispatch fails, because job->error is 0 */
1155 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1158 zvol_add_clones(dsname, minors_list);
1160 if (snapdev == ZFS_SNAPDEV_VISIBLE) {
1162 * traverse snapshots only, do not traverse children,
1163 * and skip the 'dsname'
1165 (void) dmu_objset_find(dsname,
1166 zvol_create_snap_minor_cb, (void *)job,
1170 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
1178 * Create minors for the specified dataset, including children and snapshots.
1179 * Pay attention to the 'snapdev' property and iterate over the snapshots
1180 * only if they are 'visible'. This approach allows one to assure that the
1181 * snapshot metadata is read from disk only if it is needed.
1183 * The name can represent a dataset to be recursively scanned for zvols and
1184 * their snapshots, or a single zvol snapshot. If the name represents a
1185 * dataset, the scan is performed in two nested stages:
1186 * - scan the dataset for zvols, and
1187 * - for each zvol, create a minor node, then check if the zvol's snapshots
1188 * are 'visible', and only then iterate over the snapshots if needed
1190 * If the name represents a snapshot, a check is performed if the snapshot is
1191 * 'visible' (which also verifies that the parent is a zvol), and if so,
1192 * a minor node for that snapshot is created.
1195 zvol_create_minors_recursive(const char *name)
1200 if (zvol_inhibit_dev)
1204 * This is the list for prefetch jobs. Whenever we found a match
1205 * during dmu_objset_find, we insert a minors_job to the list and do
1206 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
1207 * any lock because all list operation is done on the current thread.
1209 * We will use this list to do zvol_os_create_minor after prefetch
1210 * so we don't have to traverse using dmu_objset_find again.
1212 list_create(&minors_list, sizeof (minors_job_t),
1213 offsetof(minors_job_t, link));
1216 if (strchr(name, '@') != NULL) {
1219 int error = dsl_prop_get_integer(name, "snapdev",
1222 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1223 (void) zvol_os_create_minor(name);
1225 fstrans_cookie_t cookie = spl_fstrans_mark();
1226 (void) dmu_objset_find(name, zvol_create_minors_cb,
1227 &minors_list, DS_FIND_CHILDREN);
1228 spl_fstrans_unmark(cookie);
1231 taskq_wait_outstanding(system_taskq, 0);
1234 * Prefetch is completed, we can do zvol_os_create_minor
1237 while ((job = list_remove_head(&minors_list)) != NULL) {
1239 (void) zvol_os_create_minor(job->name);
1240 kmem_strfree(job->name);
1241 kmem_free(job, sizeof (minors_job_t));
1244 list_destroy(&minors_list);
1248 zvol_create_minor(const char *name)
1251 * Note: the dsl_pool_config_lock must not be held.
1252 * Minor node creation needs to obtain the zvol_state_lock.
1253 * zvol_open() obtains the zvol_state_lock and then the dsl pool
1254 * config lock. Therefore, we can't have the config lock now if
1255 * we are going to wait for the zvol_state_lock, because it
1256 * would be a lock order inversion which could lead to deadlock.
1259 if (zvol_inhibit_dev)
1262 if (strchr(name, '@') != NULL) {
1265 int error = dsl_prop_get_integer(name,
1266 "snapdev", &snapdev, NULL);
1268 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1269 (void) zvol_os_create_minor(name);
1271 (void) zvol_os_create_minor(name);
1276 * Remove minors for specified dataset including children and snapshots.
1280 zvol_free_task(void *arg)
1286 zvol_remove_minors_impl(const char *name)
1288 zvol_state_t *zv, *zv_next;
1289 int namelen = ((name) ? strlen(name) : 0);
1293 if (zvol_inhibit_dev)
1296 list_create(&free_list, sizeof (zvol_state_t),
1297 offsetof(zvol_state_t, zv_next));
1299 rw_enter(&zvol_state_lock, RW_WRITER);
1301 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1302 zv_next = list_next(&zvol_state_list, zv);
1304 mutex_enter(&zv->zv_state_lock);
1305 if (name == NULL || strcmp(zv->zv_name, name) == 0 ||
1306 (strncmp(zv->zv_name, name, namelen) == 0 &&
1307 (zv->zv_name[namelen] == '/' ||
1308 zv->zv_name[namelen] == '@'))) {
1310 * By holding zv_state_lock here, we guarantee that no
1311 * one is currently using this zv
1314 /* If in use, leave alone */
1315 if (zv->zv_open_count > 0 ||
1316 atomic_read(&zv->zv_suspend_ref)) {
1317 mutex_exit(&zv->zv_state_lock);
1324 * Cleared while holding zvol_state_lock as a writer
1325 * which will prevent zvol_open() from opening it.
1327 zvol_os_clear_private(zv);
1329 /* Drop zv_state_lock before zvol_free() */
1330 mutex_exit(&zv->zv_state_lock);
1332 /* Try parallel zv_free, if failed do it in place */
1333 t = taskq_dispatch(system_taskq, zvol_free_task, zv,
1335 if (t == TASKQID_INVALID)
1336 list_insert_head(&free_list, zv);
1338 mutex_exit(&zv->zv_state_lock);
1341 rw_exit(&zvol_state_lock);
1343 /* Drop zvol_state_lock before calling zvol_free() */
1344 while ((zv = list_remove_head(&free_list)) != NULL)
1348 /* Remove minor for this specific volume only */
1350 zvol_remove_minor_impl(const char *name)
1352 zvol_state_t *zv = NULL, *zv_next;
1354 if (zvol_inhibit_dev)
1357 rw_enter(&zvol_state_lock, RW_WRITER);
1359 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1360 zv_next = list_next(&zvol_state_list, zv);
1362 mutex_enter(&zv->zv_state_lock);
1363 if (strcmp(zv->zv_name, name) == 0) {
1365 * By holding zv_state_lock here, we guarantee that no
1366 * one is currently using this zv
1369 /* If in use, leave alone */
1370 if (zv->zv_open_count > 0 ||
1371 atomic_read(&zv->zv_suspend_ref)) {
1372 mutex_exit(&zv->zv_state_lock);
1377 zvol_os_clear_private(zv);
1378 mutex_exit(&zv->zv_state_lock);
1381 mutex_exit(&zv->zv_state_lock);
1385 /* Drop zvol_state_lock before calling zvol_free() */
1386 rw_exit(&zvol_state_lock);
1393 * Rename minors for specified dataset including children and snapshots.
1396 zvol_rename_minors_impl(const char *oldname, const char *newname)
1398 zvol_state_t *zv, *zv_next;
1401 if (zvol_inhibit_dev)
1404 oldnamelen = strlen(oldname);
1406 rw_enter(&zvol_state_lock, RW_READER);
1408 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1409 zv_next = list_next(&zvol_state_list, zv);
1411 mutex_enter(&zv->zv_state_lock);
1413 if (strcmp(zv->zv_name, oldname) == 0) {
1414 zvol_os_rename_minor(zv, newname);
1415 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
1416 (zv->zv_name[oldnamelen] == '/' ||
1417 zv->zv_name[oldnamelen] == '@')) {
1418 char *name = kmem_asprintf("%s%c%s", newname,
1419 zv->zv_name[oldnamelen],
1420 zv->zv_name + oldnamelen + 1);
1421 zvol_os_rename_minor(zv, name);
1425 mutex_exit(&zv->zv_state_lock);
1428 rw_exit(&zvol_state_lock);
1431 typedef struct zvol_snapdev_cb_arg {
1433 } zvol_snapdev_cb_arg_t;
1436 zvol_set_snapdev_cb(const char *dsname, void *param)
1438 zvol_snapdev_cb_arg_t *arg = param;
1440 if (strchr(dsname, '@') == NULL)
1443 switch (arg->snapdev) {
1444 case ZFS_SNAPDEV_VISIBLE:
1445 (void) zvol_os_create_minor(dsname);
1447 case ZFS_SNAPDEV_HIDDEN:
1448 (void) zvol_remove_minor_impl(dsname);
1456 zvol_set_snapdev_impl(char *name, uint64_t snapdev)
1458 zvol_snapdev_cb_arg_t arg = {snapdev};
1459 fstrans_cookie_t cookie = spl_fstrans_mark();
1461 * The zvol_set_snapdev_sync() sets snapdev appropriately
1462 * in the dataset hierarchy. Here, we only scan snapshots.
1464 dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS);
1465 spl_fstrans_unmark(cookie);
1469 zvol_set_volmode_impl(char *name, uint64_t volmode)
1471 fstrans_cookie_t cookie;
1472 uint64_t old_volmode;
1475 if (strchr(name, '@') != NULL)
1479 * It's unfortunate we need to remove minors before we create new ones:
1480 * this is necessary because our backing gendisk (zvol_state->zv_disk)
1481 * could be different when we set, for instance, volmode from "geom"
1482 * to "dev" (or vice versa).
1484 zv = zvol_find_by_name(name, RW_NONE);
1485 if (zv == NULL && volmode == ZFS_VOLMODE_NONE)
1488 old_volmode = zv->zv_volmode;
1489 mutex_exit(&zv->zv_state_lock);
1490 if (old_volmode == volmode)
1492 zvol_wait_close(zv);
1494 cookie = spl_fstrans_mark();
1496 case ZFS_VOLMODE_NONE:
1497 (void) zvol_remove_minor_impl(name);
1499 case ZFS_VOLMODE_GEOM:
1500 case ZFS_VOLMODE_DEV:
1501 (void) zvol_remove_minor_impl(name);
1502 (void) zvol_os_create_minor(name);
1504 case ZFS_VOLMODE_DEFAULT:
1505 (void) zvol_remove_minor_impl(name);
1506 if (zvol_volmode == ZFS_VOLMODE_NONE)
1508 else /* if zvol_volmode is invalid defaults to "geom" */
1509 (void) zvol_os_create_minor(name);
1512 spl_fstrans_unmark(cookie);
1515 static zvol_task_t *
1516 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2,
1521 /* Never allow tasks on hidden names. */
1522 if (name1[0] == '$')
1525 task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP);
1527 task->value = value;
1529 strlcpy(task->name1, name1, MAXNAMELEN);
1531 strlcpy(task->name2, name2, MAXNAMELEN);
1537 zvol_task_free(zvol_task_t *task)
1539 kmem_free(task, sizeof (zvol_task_t));
1543 * The worker thread function performed asynchronously.
1546 zvol_task_cb(void *arg)
1548 zvol_task_t *task = arg;
1551 case ZVOL_ASYNC_REMOVE_MINORS:
1552 zvol_remove_minors_impl(task->name1);
1554 case ZVOL_ASYNC_RENAME_MINORS:
1555 zvol_rename_minors_impl(task->name1, task->name2);
1557 case ZVOL_ASYNC_SET_SNAPDEV:
1558 zvol_set_snapdev_impl(task->name1, task->value);
1560 case ZVOL_ASYNC_SET_VOLMODE:
1561 zvol_set_volmode_impl(task->name1, task->value);
1568 zvol_task_free(task);
1571 typedef struct zvol_set_prop_int_arg {
1572 const char *zsda_name;
1573 uint64_t zsda_value;
1574 zprop_source_t zsda_source;
1575 zfs_prop_t zsda_prop;
1577 } zvol_set_prop_int_arg_t;
1580 * Sanity check the dataset for safe use by the sync task. No additional
1581 * conditions are imposed.
1584 zvol_set_common_check(void *arg, dmu_tx_t *tx)
1586 zvol_set_prop_int_arg_t *zsda = arg;
1587 dsl_pool_t *dp = dmu_tx_pool(tx);
1591 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL);
1595 dsl_dir_rele(dd, FTAG);
1601 zvol_set_common_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1603 zvol_set_prop_int_arg_t *zsda = arg;
1604 char dsname[MAXNAMELEN];
1608 const char *prop_name = zfs_prop_to_name(zsda->zsda_prop);
1609 dsl_dataset_name(ds, dsname);
1611 if (dsl_prop_get_int_ds(ds, prop_name, &prop) != 0)
1614 switch (zsda->zsda_prop) {
1615 case ZFS_PROP_VOLMODE:
1616 task = zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE, dsname,
1619 case ZFS_PROP_SNAPDEV:
1620 task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname,
1631 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb,
1637 * Traverse all child datasets and apply the property appropriately.
1638 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
1639 * dataset and read the effective "property" on every child in the callback
1640 * function: this is because the value is not guaranteed to be the same in the
1641 * whole dataset hierarchy.
1644 zvol_set_common_sync(void *arg, dmu_tx_t *tx)
1646 zvol_set_prop_int_arg_t *zsda = arg;
1647 dsl_pool_t *dp = dmu_tx_pool(tx);
1652 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL));
1655 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds);
1657 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(zsda->zsda_prop),
1658 zsda->zsda_source, sizeof (zsda->zsda_value), 1,
1659 &zsda->zsda_value, zsda->zsda_tx);
1660 dsl_dataset_rele(ds, FTAG);
1663 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_common_sync_cb,
1664 zsda, DS_FIND_CHILDREN);
1666 dsl_dir_rele(dd, FTAG);
1670 zvol_set_common(const char *ddname, zfs_prop_t prop, zprop_source_t source,
1673 zvol_set_prop_int_arg_t zsda;
1675 zsda.zsda_name = ddname;
1676 zsda.zsda_source = source;
1677 zsda.zsda_value = val;
1678 zsda.zsda_prop = prop;
1680 return (dsl_sync_task(ddname, zvol_set_common_check,
1681 zvol_set_common_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
1685 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async)
1690 task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL);
1694 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
1695 if ((async == B_FALSE) && (id != TASKQID_INVALID))
1696 taskq_wait_id(spa->spa_zvol_taskq, id);
1700 zvol_rename_minors(spa_t *spa, const char *name1, const char *name2,
1706 task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL);
1710 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
1711 if ((async == B_FALSE) && (id != TASKQID_INVALID))
1712 taskq_wait_id(spa->spa_zvol_taskq, id);
1716 zvol_is_zvol(const char *name)
1719 return (zvol_os_is_zvol(name));
1723 zvol_init_impl(void)
1727 list_create(&zvol_state_list, sizeof (zvol_state_t),
1728 offsetof(zvol_state_t, zv_next));
1729 rw_init(&zvol_state_lock, NULL, RW_DEFAULT, NULL);
1731 zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head),
1733 for (i = 0; i < ZVOL_HT_SIZE; i++)
1734 INIT_HLIST_HEAD(&zvol_htable[i]);
1740 zvol_fini_impl(void)
1742 zvol_remove_minors_impl(NULL);
1745 * The call to "zvol_remove_minors_impl" may dispatch entries to
1746 * the system_taskq, but it doesn't wait for those entries to
1747 * complete before it returns. Thus, we must wait for all of the
1748 * removals to finish, before we can continue.
1750 taskq_wait_outstanding(system_taskq, 0);
1752 kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
1753 list_destroy(&zvol_state_list);
1754 rw_destroy(&zvol_state_lock);