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 http://www.opensolaris.org/os/licensing.
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.
71 * It is also worth keeping in mind that once add_disk() is called, the zvol is
72 * announced to the world, and zvol_open()/zvol_release() can be called at any
73 * time. Incidentally, add_disk() itself calls zvol_open()->zvol_first_open()
74 * and zvol_release()->zvol_last_close() directly as well.
77 #include <sys/dataset_kstats.h>
79 #include <sys/dmu_traverse.h>
80 #include <sys/dsl_dataset.h>
81 #include <sys/dsl_prop.h>
82 #include <sys/dsl_dir.h>
84 #include <sys/zfeature.h>
85 #include <sys/zil_impl.h>
86 #include <sys/dmu_tx.h>
88 #include <sys/zfs_rlock.h>
89 #include <sys/spa_impl.h>
92 #include <linux/blkdev_compat.h>
93 #include <linux/task_io_accounting_ops.h>
95 unsigned int zvol_inhibit_dev = 0;
96 unsigned int zvol_major = ZVOL_MAJOR;
97 unsigned int zvol_threads = 32;
98 unsigned int zvol_request_sync = 0;
99 unsigned int zvol_prefetch_bytes = (128 * 1024);
100 unsigned long zvol_max_discard_blocks = 16384;
101 unsigned int zvol_volmode = ZFS_VOLMODE_GEOM;
103 static taskq_t *zvol_taskq;
104 static krwlock_t zvol_state_lock;
105 static list_t zvol_state_list;
107 #define ZVOL_HT_SIZE 1024
108 static struct hlist_head *zvol_htable;
109 #define ZVOL_HT_HEAD(hash) (&zvol_htable[(hash) & (ZVOL_HT_SIZE-1)])
111 static struct ida zvol_ida;
114 * The in-core state of each volume.
117 char zv_name[MAXNAMELEN]; /* name */
118 uint64_t zv_volsize; /* advertised space */
119 uint64_t zv_volblocksize; /* volume block size */
120 objset_t *zv_objset; /* objset handle */
121 uint32_t zv_flags; /* ZVOL_* flags */
122 uint32_t zv_open_count; /* open counts */
123 uint32_t zv_changed; /* disk changed */
124 zilog_t *zv_zilog; /* ZIL handle */
125 rangelock_t zv_rangelock; /* for range locking */
126 dnode_t *zv_dn; /* dnode hold */
127 dev_t zv_dev; /* device id */
128 struct gendisk *zv_disk; /* generic disk */
129 struct request_queue *zv_queue; /* request queue */
130 dataset_kstats_t zv_kstat; /* zvol kstats */
131 list_node_t zv_next; /* next zvol_state_t linkage */
132 uint64_t zv_hash; /* name hash */
133 struct hlist_node zv_hlink; /* hash link */
134 kmutex_t zv_state_lock; /* protects zvol_state_t */
135 atomic_t zv_suspend_ref; /* refcount for suspend */
136 krwlock_t zv_suspend_lock; /* suspend lock */
140 ZVOL_ASYNC_CREATE_MINORS,
141 ZVOL_ASYNC_REMOVE_MINORS,
142 ZVOL_ASYNC_RENAME_MINORS,
143 ZVOL_ASYNC_SET_SNAPDEV,
144 ZVOL_ASYNC_SET_VOLMODE,
150 char pool[MAXNAMELEN];
151 char name1[MAXNAMELEN];
152 char name2[MAXNAMELEN];
153 zprop_source_t source;
157 #define ZVOL_RDONLY 0x1
159 * Whether the zvol has been written to (as opposed to ZVOL_RDONLY, which
160 * specifies whether or not the zvol _can_ be written to)
162 #define ZVOL_WRITTEN_TO 0x2
165 zvol_name_hash(const char *name)
168 uint64_t crc = -1ULL;
169 uint8_t *p = (uint8_t *)name;
170 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
171 for (i = 0; i < MAXNAMELEN - 1 && *p; i++, p++) {
172 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (*p)) & 0xFF];
178 * Find a zvol_state_t given the full major+minor dev_t. If found,
179 * return with zv_state_lock taken, otherwise, return (NULL) without
180 * taking zv_state_lock.
182 static zvol_state_t *
183 zvol_find_by_dev(dev_t dev)
187 rw_enter(&zvol_state_lock, RW_READER);
188 for (zv = list_head(&zvol_state_list); zv != NULL;
189 zv = list_next(&zvol_state_list, zv)) {
190 mutex_enter(&zv->zv_state_lock);
191 if (zv->zv_dev == dev) {
192 rw_exit(&zvol_state_lock);
195 mutex_exit(&zv->zv_state_lock);
197 rw_exit(&zvol_state_lock);
203 * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
204 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
205 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
206 * before zv_state_lock. The mode argument indicates the mode (including none)
207 * for zv_suspend_lock to be taken.
209 static zvol_state_t *
210 zvol_find_by_name_hash(const char *name, uint64_t hash, int mode)
213 struct hlist_node *p = NULL;
215 rw_enter(&zvol_state_lock, RW_READER);
216 hlist_for_each(p, ZVOL_HT_HEAD(hash)) {
217 zv = hlist_entry(p, zvol_state_t, zv_hlink);
218 mutex_enter(&zv->zv_state_lock);
219 if (zv->zv_hash == hash &&
220 strncmp(zv->zv_name, name, MAXNAMELEN) == 0) {
222 * this is the right zvol, take the locks in the
225 if (mode != RW_NONE &&
226 !rw_tryenter(&zv->zv_suspend_lock, mode)) {
227 mutex_exit(&zv->zv_state_lock);
228 rw_enter(&zv->zv_suspend_lock, mode);
229 mutex_enter(&zv->zv_state_lock);
231 * zvol cannot be renamed as we continue
232 * to hold zvol_state_lock
234 ASSERT(zv->zv_hash == hash &&
235 strncmp(zv->zv_name, name, MAXNAMELEN)
238 rw_exit(&zvol_state_lock);
241 mutex_exit(&zv->zv_state_lock);
243 rw_exit(&zvol_state_lock);
249 * Find a zvol_state_t given the name.
250 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
251 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
252 * before zv_state_lock. The mode argument indicates the mode (including none)
253 * for zv_suspend_lock to be taken.
255 static zvol_state_t *
256 zvol_find_by_name(const char *name, int mode)
258 return (zvol_find_by_name_hash(name, zvol_name_hash(name), mode));
263 * Given a path, return TRUE if path is a ZVOL.
266 zvol_is_zvol(const char *device)
268 struct block_device *bdev;
271 bdev = vdev_lookup_bdev(device);
275 major = MAJOR(bdev->bd_dev);
278 if (major == zvol_major)
285 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
288 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
290 zfs_creat_t *zct = arg;
291 nvlist_t *nvprops = zct->zct_props;
293 uint64_t volblocksize, volsize;
295 VERIFY(nvlist_lookup_uint64(nvprops,
296 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
297 if (nvlist_lookup_uint64(nvprops,
298 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
299 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
302 * These properties must be removed from the list so the generic
303 * property setting step won't apply to them.
305 VERIFY(nvlist_remove_all(nvprops,
306 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
307 (void) nvlist_remove_all(nvprops,
308 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
310 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
314 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
318 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
323 * ZFS_IOC_OBJSET_STATS entry point.
326 zvol_get_stats(objset_t *os, nvlist_t *nv)
329 dmu_object_info_t *doi;
332 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
334 return (SET_ERROR(error));
336 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
337 doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);
338 error = dmu_object_info(os, ZVOL_OBJ, doi);
341 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
342 doi->doi_data_block_size);
345 kmem_free(doi, sizeof (dmu_object_info_t));
347 return (SET_ERROR(error));
351 * Sanity check volume size.
354 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
357 return (SET_ERROR(EINVAL));
359 if (volsize % blocksize != 0)
360 return (SET_ERROR(EINVAL));
363 if (volsize - 1 > SPEC_MAXOFFSET_T)
364 return (SET_ERROR(EOVERFLOW));
370 * Ensure the zap is flushed then inform the VFS of the capacity change.
373 zvol_update_volsize(uint64_t volsize, objset_t *os)
379 tx = dmu_tx_create(os);
380 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
381 dmu_tx_mark_netfree(tx);
382 error = dmu_tx_assign(tx, TXG_WAIT);
385 return (SET_ERROR(error));
387 txg = dmu_tx_get_txg(tx);
389 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
393 txg_wait_synced(dmu_objset_pool(os), txg);
396 error = dmu_free_long_range(os,
397 ZVOL_OBJ, volsize, DMU_OBJECT_END);
403 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume
404 * size will result in a udev "change" event being generated.
407 zvol_set_volsize(const char *name, uint64_t volsize)
410 struct gendisk *disk = NULL;
413 boolean_t owned = B_FALSE;
415 error = dsl_prop_get_integer(name,
416 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
418 return (SET_ERROR(error));
420 return (SET_ERROR(EROFS));
422 zvol_state_t *zv = zvol_find_by_name(name, RW_READER);
424 ASSERT(zv == NULL || (MUTEX_HELD(&zv->zv_state_lock) &&
425 RW_READ_HELD(&zv->zv_suspend_lock)));
427 if (zv == NULL || zv->zv_objset == NULL) {
429 rw_exit(&zv->zv_suspend_lock);
430 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE,
433 mutex_exit(&zv->zv_state_lock);
434 return (SET_ERROR(error));
443 dmu_object_info_t *doi = kmem_alloc(sizeof (*doi), KM_SLEEP);
445 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) ||
446 (error = zvol_check_volsize(volsize, doi->doi_data_block_size)))
449 error = zvol_update_volsize(volsize, os);
450 if (error == 0 && zv != NULL) {
451 zv->zv_volsize = volsize;
456 kmem_free(doi, sizeof (dmu_object_info_t));
459 dmu_objset_disown(os, B_TRUE, FTAG);
461 zv->zv_objset = NULL;
463 rw_exit(&zv->zv_suspend_lock);
467 mutex_exit(&zv->zv_state_lock);
470 revalidate_disk(disk);
472 return (SET_ERROR(error));
476 * Sanity check volume block size.
479 zvol_check_volblocksize(const char *name, uint64_t volblocksize)
481 /* Record sizes above 128k need the feature to be enabled */
482 if (volblocksize > SPA_OLD_MAXBLOCKSIZE) {
486 if ((error = spa_open(name, &spa, FTAG)) != 0)
489 if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
490 spa_close(spa, FTAG);
491 return (SET_ERROR(ENOTSUP));
495 * We don't allow setting the property above 1MB,
496 * unless the tunable has been changed.
498 if (volblocksize > zfs_max_recordsize)
499 return (SET_ERROR(EDOM));
501 spa_close(spa, FTAG);
504 if (volblocksize < SPA_MINBLOCKSIZE ||
505 volblocksize > SPA_MAXBLOCKSIZE ||
507 return (SET_ERROR(EDOM));
513 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
516 zvol_set_volblocksize(const char *name, uint64_t volblocksize)
522 zv = zvol_find_by_name(name, RW_READER);
525 return (SET_ERROR(ENXIO));
527 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
528 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
530 if (zv->zv_flags & ZVOL_RDONLY) {
531 mutex_exit(&zv->zv_state_lock);
532 rw_exit(&zv->zv_suspend_lock);
533 return (SET_ERROR(EROFS));
536 tx = dmu_tx_create(zv->zv_objset);
537 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
538 error = dmu_tx_assign(tx, TXG_WAIT);
542 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
543 volblocksize, 0, tx);
544 if (error == ENOTSUP)
545 error = SET_ERROR(EBUSY);
548 zv->zv_volblocksize = volblocksize;
551 mutex_exit(&zv->zv_state_lock);
552 rw_exit(&zv->zv_suspend_lock);
554 return (SET_ERROR(error));
558 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
559 * implement DKIOCFREE/free-long-range.
562 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
564 zvol_state_t *zv = arg1;
565 lr_truncate_t *lr = arg2;
566 uint64_t offset, length;
569 byteswap_uint64_array(lr, sizeof (*lr));
571 offset = lr->lr_offset;
572 length = lr->lr_length;
574 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
578 * Replay a TX_WRITE ZIL transaction that didn't get committed
579 * after a system failure
582 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
584 zvol_state_t *zv = arg1;
585 lr_write_t *lr = arg2;
586 objset_t *os = zv->zv_objset;
587 char *data = (char *)(lr + 1); /* data follows lr_write_t */
588 uint64_t offset, length;
593 byteswap_uint64_array(lr, sizeof (*lr));
595 offset = lr->lr_offset;
596 length = lr->lr_length;
598 /* If it's a dmu_sync() block, write the whole block */
599 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
600 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
601 if (length < blocksize) {
602 offset -= offset % blocksize;
607 tx = dmu_tx_create(os);
608 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
609 error = dmu_tx_assign(tx, TXG_WAIT);
613 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
621 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
623 return (SET_ERROR(ENOTSUP));
627 * Callback vectors for replaying records.
628 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
630 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
631 zvol_replay_err, /* no such transaction type */
632 zvol_replay_err, /* TX_CREATE */
633 zvol_replay_err, /* TX_MKDIR */
634 zvol_replay_err, /* TX_MKXATTR */
635 zvol_replay_err, /* TX_SYMLINK */
636 zvol_replay_err, /* TX_REMOVE */
637 zvol_replay_err, /* TX_RMDIR */
638 zvol_replay_err, /* TX_LINK */
639 zvol_replay_err, /* TX_RENAME */
640 zvol_replay_write, /* TX_WRITE */
641 zvol_replay_truncate, /* TX_TRUNCATE */
642 zvol_replay_err, /* TX_SETATTR */
643 zvol_replay_err, /* TX_ACL */
644 zvol_replay_err, /* TX_CREATE_ATTR */
645 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
646 zvol_replay_err, /* TX_MKDIR_ACL */
647 zvol_replay_err, /* TX_MKDIR_ATTR */
648 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
649 zvol_replay_err, /* TX_WRITE2 */
653 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
655 * We store data in the log buffers if it's small enough.
656 * Otherwise we will later flush the data out via dmu_sync().
658 ssize_t zvol_immediate_write_sz = 32768;
661 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
662 uint64_t size, int sync)
664 uint32_t blocksize = zv->zv_volblocksize;
665 zilog_t *zilog = zv->zv_zilog;
666 itx_wr_state_t write_state;
668 if (zil_replaying(zilog, tx))
671 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
672 write_state = WR_INDIRECT;
673 else if (!spa_has_slogs(zilog->zl_spa) &&
674 size >= blocksize && blocksize > zvol_immediate_write_sz)
675 write_state = WR_INDIRECT;
677 write_state = WR_COPIED;
679 write_state = WR_NEED_COPY;
684 itx_wr_state_t wr_state = write_state;
687 if (wr_state == WR_COPIED && size > zil_max_copied_data(zilog))
688 wr_state = WR_NEED_COPY;
689 else if (wr_state == WR_INDIRECT)
690 len = MIN(blocksize - P2PHASE(offset, blocksize), size);
692 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
693 (wr_state == WR_COPIED ? len : 0));
694 lr = (lr_write_t *)&itx->itx_lr;
695 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
696 offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
697 zil_itx_destroy(itx);
698 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
699 lr = (lr_write_t *)&itx->itx_lr;
700 wr_state = WR_NEED_COPY;
703 itx->itx_wr_state = wr_state;
704 lr->lr_foid = ZVOL_OBJ;
705 lr->lr_offset = offset;
708 BP_ZERO(&lr->lr_blkptr);
710 itx->itx_private = zv;
711 itx->itx_sync = sync;
713 (void) zil_itx_assign(zilog, itx, tx);
720 typedef struct zv_request {
727 uio_from_bio(uio_t *uio, struct bio *bio)
729 uio->uio_bvec = &bio->bi_io_vec[BIO_BI_IDX(bio)];
730 uio->uio_iovcnt = bio->bi_vcnt - BIO_BI_IDX(bio);
731 uio->uio_loffset = BIO_BI_SECTOR(bio) << 9;
732 uio->uio_segflg = UIO_BVEC;
733 uio->uio_limit = MAXOFFSET_T;
734 uio->uio_resid = BIO_BI_SIZE(bio);
735 uio->uio_skip = BIO_BI_SKIP(bio);
739 zvol_write(void *arg)
743 zv_request_t *zvr = arg;
744 struct bio *bio = zvr->bio;
745 uio_t uio = { { 0 }, 0 };
746 uio_from_bio(&uio, bio);
748 zvol_state_t *zv = zvr->zv;
749 ASSERT(zv && zv->zv_open_count > 0);
750 ASSERT(zv->zv_zilog != NULL);
752 ssize_t start_resid = uio.uio_resid;
753 unsigned long start_jif = jiffies;
754 blk_generic_start_io_acct(zv->zv_queue, WRITE, bio_sectors(bio),
755 &zv->zv_disk->part0);
758 bio_is_fua(bio) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
760 uint64_t volsize = zv->zv_volsize;
761 while (uio.uio_resid > 0 && uio.uio_loffset < volsize) {
762 uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1);
763 uint64_t off = uio.uio_loffset;
764 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
766 if (bytes > volsize - off) /* don't write past the end */
767 bytes = volsize - off;
769 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
771 /* This will only fail for ENOSPC */
772 error = dmu_tx_assign(tx, TXG_WAIT);
777 error = dmu_write_uio_dnode(zv->zv_dn, &uio, bytes, tx);
779 zvol_log_write(zv, tx, off, bytes, sync);
786 rangelock_exit(zvr->lr);
788 int64_t nwritten = start_resid - uio.uio_resid;
789 dataset_kstats_update_write_kstats(&zv->zv_kstat, nwritten);
790 task_io_account_write(nwritten);
793 zil_commit(zv->zv_zilog, ZVOL_OBJ);
795 rw_exit(&zv->zv_suspend_lock);
796 blk_generic_end_io_acct(zv->zv_queue, WRITE, &zv->zv_disk->part0,
798 BIO_END_IO(bio, -error);
799 kmem_free(zvr, sizeof (zv_request_t));
803 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
806 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
811 zilog_t *zilog = zv->zv_zilog;
813 if (zil_replaying(zilog, tx))
816 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
817 lr = (lr_truncate_t *)&itx->itx_lr;
818 lr->lr_foid = ZVOL_OBJ;
822 itx->itx_sync = sync;
823 zil_itx_assign(zilog, itx, tx);
827 zvol_discard(void *arg)
829 zv_request_t *zvr = arg;
830 struct bio *bio = zvr->bio;
831 zvol_state_t *zv = zvr->zv;
832 uint64_t start = BIO_BI_SECTOR(bio) << 9;
833 uint64_t size = BIO_BI_SIZE(bio);
834 uint64_t end = start + size;
838 unsigned long start_jif;
840 ASSERT(zv && zv->zv_open_count > 0);
841 ASSERT(zv->zv_zilog != NULL);
844 blk_generic_start_io_acct(zv->zv_queue, WRITE, bio_sectors(bio),
845 &zv->zv_disk->part0);
847 sync = bio_is_fua(bio) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
849 if (end > zv->zv_volsize) {
850 error = SET_ERROR(EIO);
855 * Align the request to volume block boundaries when a secure erase is
856 * not required. This will prevent dnode_free_range() from zeroing out
857 * the unaligned parts which is slow (read-modify-write) and useless
858 * since we are not freeing any space by doing so.
860 if (!bio_is_secure_erase(bio)) {
861 start = P2ROUNDUP(start, zv->zv_volblocksize);
862 end = P2ALIGN(end, zv->zv_volblocksize);
869 tx = dmu_tx_create(zv->zv_objset);
870 dmu_tx_mark_netfree(tx);
871 error = dmu_tx_assign(tx, TXG_WAIT);
875 zvol_log_truncate(zv, tx, start, size, B_TRUE);
877 error = dmu_free_long_range(zv->zv_objset,
878 ZVOL_OBJ, start, size);
881 rangelock_exit(zvr->lr);
883 if (error == 0 && sync)
884 zil_commit(zv->zv_zilog, ZVOL_OBJ);
886 rw_exit(&zv->zv_suspend_lock);
887 blk_generic_end_io_acct(zv->zv_queue, WRITE, &zv->zv_disk->part0,
889 BIO_END_IO(bio, -error);
890 kmem_free(zvr, sizeof (zv_request_t));
898 zv_request_t *zvr = arg;
899 struct bio *bio = zvr->bio;
900 uio_t uio = { { 0 }, 0 };
901 uio_from_bio(&uio, bio);
903 zvol_state_t *zv = zvr->zv;
904 ASSERT(zv && zv->zv_open_count > 0);
906 ssize_t start_resid = uio.uio_resid;
907 unsigned long start_jif = jiffies;
908 blk_generic_start_io_acct(zv->zv_queue, READ, bio_sectors(bio),
909 &zv->zv_disk->part0);
911 uint64_t volsize = zv->zv_volsize;
912 while (uio.uio_resid > 0 && uio.uio_loffset < volsize) {
913 uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1);
915 /* don't read past the end */
916 if (bytes > volsize - uio.uio_loffset)
917 bytes = volsize - uio.uio_loffset;
919 error = dmu_read_uio_dnode(zv->zv_dn, &uio, bytes);
921 /* convert checksum errors into IO errors */
923 error = SET_ERROR(EIO);
927 rangelock_exit(zvr->lr);
929 int64_t nread = start_resid - uio.uio_resid;
930 dataset_kstats_update_read_kstats(&zv->zv_kstat, nread);
931 task_io_account_read(nread);
933 rw_exit(&zv->zv_suspend_lock);
934 blk_generic_end_io_acct(zv->zv_queue, READ, &zv->zv_disk->part0,
936 BIO_END_IO(bio, -error);
937 kmem_free(zvr, sizeof (zv_request_t));
942 zvol_get_done(zgd_t *zgd, int error)
945 dmu_buf_rele(zgd->zgd_db, zgd);
947 rangelock_exit(zgd->zgd_lr);
949 kmem_free(zgd, sizeof (zgd_t));
953 * Get data to generate a TX_WRITE intent log record.
956 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
958 zvol_state_t *zv = arg;
959 uint64_t offset = lr->lr_offset;
960 uint64_t size = lr->lr_length;
965 ASSERT3P(lwb, !=, NULL);
966 ASSERT3P(zio, !=, NULL);
967 ASSERT3U(size, !=, 0);
969 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
973 * Write records come in two flavors: immediate and indirect.
974 * For small writes it's cheaper to store the data with the
975 * log record (immediate); for large writes it's cheaper to
976 * sync the data and get a pointer to it (indirect) so that
977 * we don't have to write the data twice.
979 if (buf != NULL) { /* immediate write */
980 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
982 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
983 DMU_READ_NO_PREFETCH);
984 } else { /* indirect write */
986 * Have to lock the whole block to ensure when it's written out
987 * and its checksum is being calculated that no one can change
988 * the data. Contrarily to zfs_get_data we need not re-check
989 * blocksize after we get the lock because it cannot be changed.
991 size = zv->zv_volblocksize;
992 offset = P2ALIGN_TYPED(offset, size, uint64_t);
993 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
995 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
996 DMU_READ_NO_PREFETCH);
998 blkptr_t *bp = &lr->lr_blkptr;
1004 ASSERT(db->db_offset == offset);
1005 ASSERT(db->db_size == size);
1007 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1008 zvol_get_done, zgd);
1015 zvol_get_done(zgd, error);
1017 return (SET_ERROR(error));
1020 static MAKE_REQUEST_FN_RET
1021 zvol_request(struct request_queue *q, struct bio *bio)
1023 zvol_state_t *zv = q->queuedata;
1024 fstrans_cookie_t cookie = spl_fstrans_mark();
1025 uint64_t offset = BIO_BI_SECTOR(bio) << 9;
1026 uint64_t size = BIO_BI_SIZE(bio);
1027 int rw = bio_data_dir(bio);
1030 if (bio_has_data(bio) && offset + size > zv->zv_volsize) {
1032 "%s: bad access: offset=%llu, size=%lu\n",
1033 zv->zv_disk->disk_name,
1034 (long long unsigned)offset,
1035 (long unsigned)size);
1037 BIO_END_IO(bio, -SET_ERROR(EIO));
1042 boolean_t need_sync = B_FALSE;
1044 if (unlikely(zv->zv_flags & ZVOL_RDONLY)) {
1045 BIO_END_IO(bio, -SET_ERROR(EROFS));
1050 * To be released in the I/O function. See the comment on
1051 * rangelock_enter() below.
1053 rw_enter(&zv->zv_suspend_lock, RW_READER);
1056 * Open a ZIL if this is the first time we have written to this
1057 * zvol. We protect zv->zv_zilog with zv_suspend_lock rather
1058 * than zv_state_lock so that we don't need to acquire an
1059 * additional lock in this path.
1061 if (zv->zv_zilog == NULL) {
1062 rw_exit(&zv->zv_suspend_lock);
1063 rw_enter(&zv->zv_suspend_lock, RW_WRITER);
1064 if (zv->zv_zilog == NULL) {
1065 zv->zv_zilog = zil_open(zv->zv_objset,
1067 zv->zv_flags |= ZVOL_WRITTEN_TO;
1069 rw_downgrade(&zv->zv_suspend_lock);
1072 /* bio marked as FLUSH need to flush before write */
1073 if (bio_is_flush(bio))
1074 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1076 /* Some requests are just for flush and nothing else. */
1078 rw_exit(&zv->zv_suspend_lock);
1083 zvr = kmem_alloc(sizeof (zv_request_t), KM_SLEEP);
1088 * To be released in the I/O function. Since the I/O functions
1089 * are asynchronous, we take it here synchronously to make
1090 * sure overlapped I/Os are properly ordered.
1092 zvr->lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1095 * Sync writes and discards execute zil_commit() which may need
1096 * to take a RL_READER lock on the whole block being modified
1097 * via its zillog->zl_get_data(): to avoid circular dependency
1098 * issues with taskq threads execute these requests
1099 * synchronously here in zvol_request().
1101 need_sync = bio_is_fua(bio) ||
1102 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1103 if (bio_is_discard(bio) || bio_is_secure_erase(bio)) {
1104 if (zvol_request_sync || need_sync ||
1105 taskq_dispatch(zvol_taskq, zvol_discard, zvr,
1106 TQ_SLEEP) == TASKQID_INVALID)
1109 if (zvol_request_sync || need_sync ||
1110 taskq_dispatch(zvol_taskq, zvol_write, zvr,
1111 TQ_SLEEP) == TASKQID_INVALID)
1116 * The SCST driver, and possibly others, may issue READ I/Os
1117 * with a length of zero bytes. These empty I/Os contain no
1118 * data and require no additional handling.
1125 zvr = kmem_alloc(sizeof (zv_request_t), KM_SLEEP);
1129 rw_enter(&zv->zv_suspend_lock, RW_READER);
1131 zvr->lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1133 if (zvol_request_sync || taskq_dispatch(zvol_taskq,
1134 zvol_read, zvr, TQ_SLEEP) == TASKQID_INVALID)
1139 spl_fstrans_unmark(cookie);
1140 #ifdef HAVE_MAKE_REQUEST_FN_RET_INT
1142 #elif defined(HAVE_MAKE_REQUEST_FN_RET_QC)
1143 return (BLK_QC_T_NONE);
1148 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
1151 zvol_insert(zvol_state_t *zv)
1153 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
1154 ASSERT3U(MINOR(zv->zv_dev) & ZVOL_MINOR_MASK, ==, 0);
1155 list_insert_head(&zvol_state_list, zv);
1156 hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash));
1160 * Simply remove the zvol from to list of zvols.
1163 zvol_remove(zvol_state_t *zv)
1165 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
1166 list_remove(&zvol_state_list, zv);
1167 hlist_del(&zv->zv_hlink);
1171 * Setup zv after we just own the zv->objset
1174 zvol_setup_zv(zvol_state_t *zv)
1179 objset_t *os = zv->zv_objset;
1181 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1182 ASSERT(RW_LOCK_HELD(&zv->zv_suspend_lock));
1184 zv->zv_zilog = NULL;
1185 zv->zv_flags &= ~ZVOL_WRITTEN_TO;
1187 error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL);
1189 return (SET_ERROR(error));
1191 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
1193 return (SET_ERROR(error));
1195 error = dnode_hold(os, ZVOL_OBJ, FTAG, &zv->zv_dn);
1197 return (SET_ERROR(error));
1199 set_capacity(zv->zv_disk, volsize >> 9);
1200 zv->zv_volsize = volsize;
1202 if (ro || dmu_objset_is_snapshot(os) ||
1203 !spa_writeable(dmu_objset_spa(os))) {
1204 set_disk_ro(zv->zv_disk, 1);
1205 zv->zv_flags |= ZVOL_RDONLY;
1207 set_disk_ro(zv->zv_disk, 0);
1208 zv->zv_flags &= ~ZVOL_RDONLY;
1214 * Shutdown every zv_objset related stuff except zv_objset itself.
1215 * The is the reverse of zvol_setup_zv.
1218 zvol_shutdown_zv(zvol_state_t *zv)
1220 ASSERT(MUTEX_HELD(&zv->zv_state_lock) &&
1221 RW_LOCK_HELD(&zv->zv_suspend_lock));
1223 if (zv->zv_flags & ZVOL_WRITTEN_TO) {
1224 ASSERT(zv->zv_zilog != NULL);
1225 zil_close(zv->zv_zilog);
1228 zv->zv_zilog = NULL;
1230 dnode_rele(zv->zv_dn, FTAG);
1234 * Evict cached data. We must write out any dirty data before
1235 * disowning the dataset.
1237 if (zv->zv_flags & ZVOL_WRITTEN_TO)
1238 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1239 (void) dmu_objset_evict_dbufs(zv->zv_objset);
1243 * return the proper tag for rollback and recv
1246 zvol_tag(zvol_state_t *zv)
1248 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1249 return (zv->zv_open_count > 0 ? zv : NULL);
1253 * Suspend the zvol for recv and rollback.
1256 zvol_suspend(const char *name)
1260 zv = zvol_find_by_name(name, RW_WRITER);
1265 /* block all I/O, release in zvol_resume. */
1266 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1267 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1269 atomic_inc(&zv->zv_suspend_ref);
1271 if (zv->zv_open_count > 0)
1272 zvol_shutdown_zv(zv);
1275 * do not hold zv_state_lock across suspend/resume to
1276 * avoid locking up zvol lookups
1278 mutex_exit(&zv->zv_state_lock);
1280 /* zv_suspend_lock is released in zvol_resume() */
1285 zvol_resume(zvol_state_t *zv)
1289 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1291 mutex_enter(&zv->zv_state_lock);
1293 if (zv->zv_open_count > 0) {
1294 VERIFY0(dmu_objset_hold(zv->zv_name, zv, &zv->zv_objset));
1295 VERIFY3P(zv->zv_objset->os_dsl_dataset->ds_owner, ==, zv);
1296 VERIFY(dsl_dataset_long_held(zv->zv_objset->os_dsl_dataset));
1297 dmu_objset_rele(zv->zv_objset, zv);
1299 error = zvol_setup_zv(zv);
1302 mutex_exit(&zv->zv_state_lock);
1304 rw_exit(&zv->zv_suspend_lock);
1306 * We need this because we don't hold zvol_state_lock while releasing
1307 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
1308 * zv_suspend_lock to determine it is safe to free because rwlock is
1309 * not inherent atomic.
1311 atomic_dec(&zv->zv_suspend_ref);
1313 return (SET_ERROR(error));
1317 zvol_first_open(zvol_state_t *zv, boolean_t readonly)
1320 int error, locked = 0;
1323 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
1324 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1327 * In all other cases the spa_namespace_lock is taken before the
1328 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
1329 * function calls fops->open() with the bdev->bd_mutex lock held.
1330 * This deadlock can be easily observed with zvols used as vdevs.
1332 * To avoid a potential lock inversion deadlock we preemptively
1333 * try to take the spa_namespace_lock(). Normally it will not
1334 * be contended and this is safe because spa_open_common() handles
1335 * the case where the caller already holds the spa_namespace_lock.
1337 * When it is contended we risk a lock inversion if we were to
1338 * block waiting for the lock. Luckily, the __blkdev_get()
1339 * function allows us to return -ERESTARTSYS which will result in
1340 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
1341 * called again. This process can be repeated safely until both
1342 * locks are acquired.
1344 if (!mutex_owned(&spa_namespace_lock)) {
1345 locked = mutex_tryenter(&spa_namespace_lock);
1347 return (-SET_ERROR(ERESTARTSYS));
1350 ro = (readonly || (strchr(zv->zv_name, '@') != NULL));
1351 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os);
1357 error = zvol_setup_zv(zv);
1360 dmu_objset_disown(os, 1, zv);
1361 zv->zv_objset = NULL;
1366 mutex_exit(&spa_namespace_lock);
1367 return (SET_ERROR(-error));
1371 zvol_last_close(zvol_state_t *zv)
1373 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
1374 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1376 zvol_shutdown_zv(zv);
1378 dmu_objset_disown(zv->zv_objset, 1, zv);
1379 zv->zv_objset = NULL;
1383 zvol_open(struct block_device *bdev, fmode_t flag)
1387 boolean_t drop_suspend = B_TRUE;
1389 rw_enter(&zvol_state_lock, RW_READER);
1391 * Obtain a copy of private_data under the zvol_state_lock to make
1392 * sure that either the result of zvol free code path setting
1393 * bdev->bd_disk->private_data to NULL is observed, or zvol_free()
1394 * is not called on this zv because of the positive zv_open_count.
1396 zv = bdev->bd_disk->private_data;
1398 rw_exit(&zvol_state_lock);
1399 return (SET_ERROR(-ENXIO));
1402 mutex_enter(&zv->zv_state_lock);
1404 * make sure zvol is not suspended during first open
1405 * (hold zv_suspend_lock) and respect proper lock acquisition
1406 * ordering - zv_suspend_lock before zv_state_lock
1408 if (zv->zv_open_count == 0) {
1409 if (!rw_tryenter(&zv->zv_suspend_lock, RW_READER)) {
1410 mutex_exit(&zv->zv_state_lock);
1411 rw_enter(&zv->zv_suspend_lock, RW_READER);
1412 mutex_enter(&zv->zv_state_lock);
1413 /* check to see if zv_suspend_lock is needed */
1414 if (zv->zv_open_count != 0) {
1415 rw_exit(&zv->zv_suspend_lock);
1416 drop_suspend = B_FALSE;
1420 drop_suspend = B_FALSE;
1422 rw_exit(&zvol_state_lock);
1424 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1425 ASSERT(zv->zv_open_count != 0 || RW_READ_HELD(&zv->zv_suspend_lock));
1427 if (zv->zv_open_count == 0) {
1428 error = zvol_first_open(zv, !(flag & FMODE_WRITE));
1433 if ((flag & FMODE_WRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1435 goto out_open_count;
1438 zv->zv_open_count++;
1440 mutex_exit(&zv->zv_state_lock);
1442 rw_exit(&zv->zv_suspend_lock);
1444 check_disk_change(bdev);
1449 if (zv->zv_open_count == 0)
1450 zvol_last_close(zv);
1453 mutex_exit(&zv->zv_state_lock);
1455 rw_exit(&zv->zv_suspend_lock);
1456 if (error == -ERESTARTSYS)
1459 return (SET_ERROR(error));
1462 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1467 zvol_release(struct gendisk *disk, fmode_t mode)
1470 boolean_t drop_suspend = B_TRUE;
1472 rw_enter(&zvol_state_lock, RW_READER);
1473 zv = disk->private_data;
1475 mutex_enter(&zv->zv_state_lock);
1476 ASSERT(zv->zv_open_count > 0);
1478 * make sure zvol is not suspended during last close
1479 * (hold zv_suspend_lock) and respect proper lock acquisition
1480 * ordering - zv_suspend_lock before zv_state_lock
1482 if (zv->zv_open_count == 1) {
1483 if (!rw_tryenter(&zv->zv_suspend_lock, RW_READER)) {
1484 mutex_exit(&zv->zv_state_lock);
1485 rw_enter(&zv->zv_suspend_lock, RW_READER);
1486 mutex_enter(&zv->zv_state_lock);
1487 /* check to see if zv_suspend_lock is needed */
1488 if (zv->zv_open_count != 1) {
1489 rw_exit(&zv->zv_suspend_lock);
1490 drop_suspend = B_FALSE;
1494 drop_suspend = B_FALSE;
1496 rw_exit(&zvol_state_lock);
1498 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1499 ASSERT(zv->zv_open_count != 1 || RW_READ_HELD(&zv->zv_suspend_lock));
1501 zv->zv_open_count--;
1502 if (zv->zv_open_count == 0)
1503 zvol_last_close(zv);
1505 mutex_exit(&zv->zv_state_lock);
1508 rw_exit(&zv->zv_suspend_lock);
1510 #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1516 zvol_ioctl(struct block_device *bdev, fmode_t mode,
1517 unsigned int cmd, unsigned long arg)
1519 zvol_state_t *zv = bdev->bd_disk->private_data;
1522 ASSERT3U(zv->zv_open_count, >, 0);
1527 invalidate_bdev(bdev);
1528 rw_enter(&zv->zv_suspend_lock, RW_READER);
1530 if (!(zv->zv_flags & ZVOL_RDONLY))
1531 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1533 rw_exit(&zv->zv_suspend_lock);
1537 mutex_enter(&zv->zv_state_lock);
1538 error = copy_to_user((void *)arg, zv->zv_name, MAXNAMELEN);
1539 mutex_exit(&zv->zv_state_lock);
1547 return (SET_ERROR(error));
1550 #ifdef CONFIG_COMPAT
1552 zvol_compat_ioctl(struct block_device *bdev, fmode_t mode,
1553 unsigned cmd, unsigned long arg)
1555 return (zvol_ioctl(bdev, mode, cmd, arg));
1558 #define zvol_compat_ioctl NULL
1562 * Linux 2.6.38 preferred interface.
1564 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1566 zvol_check_events(struct gendisk *disk, unsigned int clearing)
1568 unsigned int mask = 0;
1570 rw_enter(&zvol_state_lock, RW_READER);
1572 zvol_state_t *zv = disk->private_data;
1574 mutex_enter(&zv->zv_state_lock);
1575 mask = zv->zv_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1577 mutex_exit(&zv->zv_state_lock);
1580 rw_exit(&zvol_state_lock);
1585 static int zvol_media_changed(struct gendisk *disk)
1589 rw_enter(&zvol_state_lock, RW_READER);
1591 zvol_state_t *zv = disk->private_data;
1593 mutex_enter(&zv->zv_state_lock);
1594 changed = zv->zv_changed;
1596 mutex_exit(&zv->zv_state_lock);
1599 rw_exit(&zvol_state_lock);
1605 static int zvol_revalidate_disk(struct gendisk *disk)
1607 rw_enter(&zvol_state_lock, RW_READER);
1609 zvol_state_t *zv = disk->private_data;
1611 mutex_enter(&zv->zv_state_lock);
1612 set_capacity(zv->zv_disk, zv->zv_volsize >> SECTOR_BITS);
1613 mutex_exit(&zv->zv_state_lock);
1616 rw_exit(&zvol_state_lock);
1622 * Provide a simple virtual geometry for legacy compatibility. For devices
1623 * smaller than 1 MiB a small head and sector count is used to allow very
1624 * tiny devices. For devices over 1 Mib a standard head and sector count
1625 * is used to keep the cylinders count reasonable.
1628 zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1630 zvol_state_t *zv = bdev->bd_disk->private_data;
1633 ASSERT3U(zv->zv_open_count, >, 0);
1635 sectors = get_capacity(zv->zv_disk);
1637 if (sectors > 2048) {
1646 geo->cylinders = sectors / (geo->heads * geo->sectors);
1651 static struct kobject *
1652 zvol_probe(dev_t dev, int *part, void *arg)
1655 struct kobject *kobj;
1657 zv = zvol_find_by_dev(dev);
1658 kobj = zv ? get_disk_and_module(zv->zv_disk) : NULL;
1659 ASSERT(zv == NULL || MUTEX_HELD(&zv->zv_state_lock));
1661 mutex_exit(&zv->zv_state_lock);
1666 static struct block_device_operations zvol_ops = {
1668 .release = zvol_release,
1669 .ioctl = zvol_ioctl,
1670 .compat_ioctl = zvol_compat_ioctl,
1671 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1672 .check_events = zvol_check_events,
1674 .media_changed = zvol_media_changed,
1676 .revalidate_disk = zvol_revalidate_disk,
1677 .getgeo = zvol_getgeo,
1678 .owner = THIS_MODULE,
1682 * Allocate memory for a new zvol_state_t and setup the required
1683 * request queue and generic disk structures for the block device.
1685 static zvol_state_t *
1686 zvol_alloc(dev_t dev, const char *name)
1691 if (dsl_prop_get_integer(name, "volmode", &volmode, NULL) != 0)
1694 if (volmode == ZFS_VOLMODE_DEFAULT)
1695 volmode = zvol_volmode;
1697 if (volmode == ZFS_VOLMODE_NONE)
1700 zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
1702 list_link_init(&zv->zv_next);
1704 mutex_init(&zv->zv_state_lock, NULL, MUTEX_DEFAULT, NULL);
1706 zv->zv_queue = blk_alloc_queue(GFP_ATOMIC);
1707 if (zv->zv_queue == NULL)
1710 blk_queue_make_request(zv->zv_queue, zvol_request);
1711 blk_queue_set_write_cache(zv->zv_queue, B_TRUE, B_TRUE);
1713 /* Limit read-ahead to a single page to prevent over-prefetching. */
1714 blk_queue_set_read_ahead(zv->zv_queue, 1);
1716 /* Disable write merging in favor of the ZIO pipeline. */
1717 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, zv->zv_queue);
1719 zv->zv_disk = alloc_disk(ZVOL_MINORS);
1720 if (zv->zv_disk == NULL)
1723 zv->zv_queue->queuedata = zv;
1725 zv->zv_open_count = 0;
1726 strlcpy(zv->zv_name, name, MAXNAMELEN);
1728 rangelock_init(&zv->zv_rangelock, NULL, NULL);
1729 rw_init(&zv->zv_suspend_lock, NULL, RW_DEFAULT, NULL);
1731 zv->zv_disk->major = zvol_major;
1732 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1733 zv->zv_disk->events = DISK_EVENT_MEDIA_CHANGE;
1736 if (volmode == ZFS_VOLMODE_DEV) {
1738 * ZFS_VOLMODE_DEV disable partitioning on ZVOL devices: set
1739 * gendisk->minors = 1 as noted in include/linux/genhd.h.
1740 * Also disable extended partition numbers (GENHD_FL_EXT_DEVT)
1741 * and suppresses partition scanning (GENHD_FL_NO_PART_SCAN)
1742 * setting gendisk->flags accordingly.
1744 zv->zv_disk->minors = 1;
1745 #if defined(GENHD_FL_EXT_DEVT)
1746 zv->zv_disk->flags &= ~GENHD_FL_EXT_DEVT;
1748 #if defined(GENHD_FL_NO_PART_SCAN)
1749 zv->zv_disk->flags |= GENHD_FL_NO_PART_SCAN;
1752 zv->zv_disk->first_minor = (dev & MINORMASK);
1753 zv->zv_disk->fops = &zvol_ops;
1754 zv->zv_disk->private_data = zv;
1755 zv->zv_disk->queue = zv->zv_queue;
1756 snprintf(zv->zv_disk->disk_name, DISK_NAME_LEN, "%s%d",
1757 ZVOL_DEV_NAME, (dev & MINORMASK));
1762 blk_cleanup_queue(zv->zv_queue);
1764 kmem_free(zv, sizeof (zvol_state_t));
1770 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1771 * At this time, the structure is not opened by anyone, is taken off
1772 * the zvol_state_list, and has its private data set to NULL.
1773 * The zvol_state_lock is dropped.
1776 zvol_free(void *arg)
1778 zvol_state_t *zv = arg;
1780 ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock));
1781 ASSERT(!MUTEX_HELD(&zv->zv_state_lock));
1782 ASSERT(zv->zv_open_count == 0);
1783 ASSERT(zv->zv_disk->private_data == NULL);
1785 rw_destroy(&zv->zv_suspend_lock);
1786 rangelock_fini(&zv->zv_rangelock);
1788 del_gendisk(zv->zv_disk);
1789 blk_cleanup_queue(zv->zv_queue);
1790 put_disk(zv->zv_disk);
1792 ida_simple_remove(&zvol_ida, MINOR(zv->zv_dev) >> ZVOL_MINOR_BITS);
1794 mutex_destroy(&zv->zv_state_lock);
1795 dataset_kstats_destroy(&zv->zv_kstat);
1797 kmem_free(zv, sizeof (zvol_state_t));
1801 * Create a block device minor node and setup the linkage between it
1802 * and the specified volume. Once this function returns the block
1803 * device is live and ready for use.
1806 zvol_create_minor_impl(const char *name)
1810 dmu_object_info_t *doi;
1816 uint64_t hash = zvol_name_hash(name);
1818 if (zvol_inhibit_dev)
1821 idx = ida_simple_get(&zvol_ida, 0, 0, kmem_flags_convert(KM_SLEEP));
1823 return (SET_ERROR(-idx));
1824 minor = idx << ZVOL_MINOR_BITS;
1826 zv = zvol_find_by_name_hash(name, hash, RW_NONE);
1828 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1829 mutex_exit(&zv->zv_state_lock);
1830 ida_simple_remove(&zvol_ida, idx);
1831 return (SET_ERROR(EEXIST));
1834 doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);
1836 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, B_TRUE, FTAG, &os);
1840 error = dmu_object_info(os, ZVOL_OBJ, doi);
1842 goto out_dmu_objset_disown;
1844 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
1846 goto out_dmu_objset_disown;
1848 zv = zvol_alloc(MKDEV(zvol_major, minor), name);
1850 error = SET_ERROR(EAGAIN);
1851 goto out_dmu_objset_disown;
1855 if (dmu_objset_is_snapshot(os))
1856 zv->zv_flags |= ZVOL_RDONLY;
1858 zv->zv_volblocksize = doi->doi_data_block_size;
1859 zv->zv_volsize = volsize;
1862 set_capacity(zv->zv_disk, zv->zv_volsize >> 9);
1864 blk_queue_max_hw_sectors(zv->zv_queue, (DMU_MAX_ACCESS / 4) >> 9);
1865 blk_queue_max_segments(zv->zv_queue, UINT16_MAX);
1866 blk_queue_max_segment_size(zv->zv_queue, UINT_MAX);
1867 blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize);
1868 blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize);
1869 blk_queue_max_discard_sectors(zv->zv_queue,
1870 (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9);
1871 blk_queue_discard_granularity(zv->zv_queue, zv->zv_volblocksize);
1872 blk_queue_flag_set(QUEUE_FLAG_DISCARD, zv->zv_queue);
1873 #ifdef QUEUE_FLAG_NONROT
1874 blk_queue_flag_set(QUEUE_FLAG_NONROT, zv->zv_queue);
1876 #ifdef QUEUE_FLAG_ADD_RANDOM
1877 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zv->zv_queue);
1880 if (spa_writeable(dmu_objset_spa(os))) {
1881 if (zil_replay_disable)
1882 zil_destroy(dmu_objset_zil(os), B_FALSE);
1884 zil_replay(os, zv, zvol_replay_vector);
1886 ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
1887 dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
1890 * When udev detects the addition of the device it will immediately
1891 * invoke blkid(8) to determine the type of content on the device.
1892 * Prefetching the blocks commonly scanned by blkid(8) will speed
1895 len = MIN(MAX(zvol_prefetch_bytes, 0), SPA_MAXBLOCKSIZE);
1897 dmu_prefetch(os, ZVOL_OBJ, 0, 0, len, ZIO_PRIORITY_SYNC_READ);
1898 dmu_prefetch(os, ZVOL_OBJ, 0, volsize - len, len,
1899 ZIO_PRIORITY_SYNC_READ);
1902 zv->zv_objset = NULL;
1903 out_dmu_objset_disown:
1904 dmu_objset_disown(os, B_TRUE, FTAG);
1906 kmem_free(doi, sizeof (dmu_object_info_t));
1909 rw_enter(&zvol_state_lock, RW_WRITER);
1911 rw_exit(&zvol_state_lock);
1912 add_disk(zv->zv_disk);
1914 ida_simple_remove(&zvol_ida, idx);
1917 return (SET_ERROR(error));
1921 * Rename a block device minor mode for the specified volume.
1924 zvol_rename_minor(zvol_state_t *zv, const char *newname)
1926 int readonly = get_disk_ro(zv->zv_disk);
1928 ASSERT(RW_LOCK_HELD(&zvol_state_lock));
1929 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1931 strlcpy(zv->zv_name, newname, sizeof (zv->zv_name));
1933 /* move to new hashtable entry */
1934 zv->zv_hash = zvol_name_hash(zv->zv_name);
1935 hlist_del(&zv->zv_hlink);
1936 hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash));
1939 * The block device's read-only state is briefly changed causing
1940 * a KOBJ_CHANGE uevent to be issued. This ensures udev detects
1941 * the name change and fixes the symlinks. This does not change
1942 * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never
1943 * changes. This would normally be done using kobject_uevent() but
1944 * that is a GPL-only symbol which is why we need this workaround.
1946 set_disk_ro(zv->zv_disk, !readonly);
1947 set_disk_ro(zv->zv_disk, readonly);
1950 typedef struct minors_job {
1960 * Prefetch zvol dnodes for the minors_job
1963 zvol_prefetch_minors_impl(void *arg)
1965 minors_job_t *job = arg;
1966 char *dsname = job->name;
1967 objset_t *os = NULL;
1969 job->error = dmu_objset_own(dsname, DMU_OST_ZVOL, B_TRUE, B_TRUE,
1971 if (job->error == 0) {
1972 dmu_prefetch(os, ZVOL_OBJ, 0, 0, 0, ZIO_PRIORITY_SYNC_READ);
1973 dmu_objset_disown(os, B_TRUE, FTAG);
1978 * Mask errors to continue dmu_objset_find() traversal
1981 zvol_create_snap_minor_cb(const char *dsname, void *arg)
1983 minors_job_t *j = arg;
1984 list_t *minors_list = j->list;
1985 const char *name = j->name;
1987 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1989 /* skip the designated dataset */
1990 if (name && strcmp(dsname, name) == 0)
1993 /* at this point, the dsname should name a snapshot */
1994 if (strchr(dsname, '@') == 0) {
1995 dprintf("zvol_create_snap_minor_cb(): "
1996 "%s is not a shapshot name\n", dsname);
1999 char *n = strdup(dsname);
2003 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
2005 job->list = minors_list;
2007 list_insert_tail(minors_list, job);
2008 /* don't care if dispatch fails, because job->error is 0 */
2009 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
2017 * Mask errors to continue dmu_objset_find() traversal
2020 zvol_create_minors_cb(const char *dsname, void *arg)
2024 list_t *minors_list = arg;
2026 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
2028 error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL);
2033 * Given the name and the 'snapdev' property, create device minor nodes
2034 * with the linkages to zvols/snapshots as needed.
2035 * If the name represents a zvol, create a minor node for the zvol, then
2036 * check if its snapshots are 'visible', and if so, iterate over the
2037 * snapshots and create device minor nodes for those.
2039 if (strchr(dsname, '@') == 0) {
2041 char *n = strdup(dsname);
2045 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
2047 job->list = minors_list;
2049 list_insert_tail(minors_list, job);
2050 /* don't care if dispatch fails, because job->error is 0 */
2051 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
2054 if (snapdev == ZFS_SNAPDEV_VISIBLE) {
2056 * traverse snapshots only, do not traverse children,
2057 * and skip the 'dsname'
2059 error = dmu_objset_find((char *)dsname,
2060 zvol_create_snap_minor_cb, (void *)job,
2064 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
2072 * Create minors for the specified dataset, including children and snapshots.
2073 * Pay attention to the 'snapdev' property and iterate over the snapshots
2074 * only if they are 'visible'. This approach allows one to assure that the
2075 * snapshot metadata is read from disk only if it is needed.
2077 * The name can represent a dataset to be recursively scanned for zvols and
2078 * their snapshots, or a single zvol snapshot. If the name represents a
2079 * dataset, the scan is performed in two nested stages:
2080 * - scan the dataset for zvols, and
2081 * - for each zvol, create a minor node, then check if the zvol's snapshots
2082 * are 'visible', and only then iterate over the snapshots if needed
2084 * If the name represents a snapshot, a check is performed if the snapshot is
2085 * 'visible' (which also verifies that the parent is a zvol), and if so,
2086 * a minor node for that snapshot is created.
2089 zvol_create_minors_impl(const char *name)
2092 fstrans_cookie_t cookie;
2097 if (zvol_inhibit_dev)
2101 * This is the list for prefetch jobs. Whenever we found a match
2102 * during dmu_objset_find, we insert a minors_job to the list and do
2103 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
2104 * any lock because all list operation is done on the current thread.
2106 * We will use this list to do zvol_create_minor_impl after prefetch
2107 * so we don't have to traverse using dmu_objset_find again.
2109 list_create(&minors_list, sizeof (minors_job_t),
2110 offsetof(minors_job_t, link));
2112 parent = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2113 (void) strlcpy(parent, name, MAXPATHLEN);
2115 if ((atp = strrchr(parent, '@')) != NULL) {
2119 error = dsl_prop_get_integer(parent, "snapdev",
2122 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
2123 error = zvol_create_minor_impl(name);
2125 cookie = spl_fstrans_mark();
2126 error = dmu_objset_find(parent, zvol_create_minors_cb,
2127 &minors_list, DS_FIND_CHILDREN);
2128 spl_fstrans_unmark(cookie);
2131 kmem_free(parent, MAXPATHLEN);
2132 taskq_wait_outstanding(system_taskq, 0);
2135 * Prefetch is completed, we can do zvol_create_minor_impl
2138 while ((job = list_head(&minors_list)) != NULL) {
2139 list_remove(&minors_list, job);
2141 zvol_create_minor_impl(job->name);
2143 kmem_free(job, sizeof (minors_job_t));
2146 list_destroy(&minors_list);
2148 return (SET_ERROR(error));
2152 * Remove minors for specified dataset including children and snapshots.
2155 zvol_remove_minors_impl(const char *name)
2157 zvol_state_t *zv, *zv_next;
2158 int namelen = ((name) ? strlen(name) : 0);
2159 taskqid_t t, tid = TASKQID_INVALID;
2162 if (zvol_inhibit_dev)
2165 list_create(&free_list, sizeof (zvol_state_t),
2166 offsetof(zvol_state_t, zv_next));
2168 rw_enter(&zvol_state_lock, RW_WRITER);
2170 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
2171 zv_next = list_next(&zvol_state_list, zv);
2173 mutex_enter(&zv->zv_state_lock);
2174 if (name == NULL || strcmp(zv->zv_name, name) == 0 ||
2175 (strncmp(zv->zv_name, name, namelen) == 0 &&
2176 (zv->zv_name[namelen] == '/' ||
2177 zv->zv_name[namelen] == '@'))) {
2179 * By holding zv_state_lock here, we guarantee that no
2180 * one is currently using this zv
2183 /* If in use, leave alone */
2184 if (zv->zv_open_count > 0 ||
2185 atomic_read(&zv->zv_suspend_ref)) {
2186 mutex_exit(&zv->zv_state_lock);
2193 * Cleared while holding zvol_state_lock as a writer
2194 * which will prevent zvol_open() from opening it.
2196 zv->zv_disk->private_data = NULL;
2198 /* Drop zv_state_lock before zvol_free() */
2199 mutex_exit(&zv->zv_state_lock);
2201 /* Try parallel zv_free, if failed do it in place */
2202 t = taskq_dispatch(system_taskq, zvol_free, zv,
2204 if (t == TASKQID_INVALID)
2205 list_insert_head(&free_list, zv);
2209 mutex_exit(&zv->zv_state_lock);
2212 rw_exit(&zvol_state_lock);
2214 /* Drop zvol_state_lock before calling zvol_free() */
2215 while ((zv = list_head(&free_list)) != NULL) {
2216 list_remove(&free_list, zv);
2220 if (tid != TASKQID_INVALID)
2221 taskq_wait_outstanding(system_taskq, tid);
2224 /* Remove minor for this specific volume only */
2226 zvol_remove_minor_impl(const char *name)
2228 zvol_state_t *zv = NULL, *zv_next;
2230 if (zvol_inhibit_dev)
2233 rw_enter(&zvol_state_lock, RW_WRITER);
2235 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
2236 zv_next = list_next(&zvol_state_list, zv);
2238 mutex_enter(&zv->zv_state_lock);
2239 if (strcmp(zv->zv_name, name) == 0) {
2241 * By holding zv_state_lock here, we guarantee that no
2242 * one is currently using this zv
2245 /* If in use, leave alone */
2246 if (zv->zv_open_count > 0 ||
2247 atomic_read(&zv->zv_suspend_ref)) {
2248 mutex_exit(&zv->zv_state_lock);
2254 * Cleared while holding zvol_state_lock as a writer
2255 * which will prevent zvol_open() from opening it.
2257 zv->zv_disk->private_data = NULL;
2259 mutex_exit(&zv->zv_state_lock);
2262 mutex_exit(&zv->zv_state_lock);
2266 /* Drop zvol_state_lock before calling zvol_free() */
2267 rw_exit(&zvol_state_lock);
2274 * Rename minors for specified dataset including children and snapshots.
2277 zvol_rename_minors_impl(const char *oldname, const char *newname)
2279 zvol_state_t *zv, *zv_next;
2280 int oldnamelen, newnamelen;
2282 if (zvol_inhibit_dev)
2285 oldnamelen = strlen(oldname);
2286 newnamelen = strlen(newname);
2288 rw_enter(&zvol_state_lock, RW_READER);
2290 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
2291 zv_next = list_next(&zvol_state_list, zv);
2293 mutex_enter(&zv->zv_state_lock);
2295 if (strcmp(zv->zv_name, oldname) == 0) {
2296 zvol_rename_minor(zv, newname);
2297 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
2298 (zv->zv_name[oldnamelen] == '/' ||
2299 zv->zv_name[oldnamelen] == '@')) {
2300 char *name = kmem_asprintf("%s%c%s", newname,
2301 zv->zv_name[oldnamelen],
2302 zv->zv_name + oldnamelen + 1);
2303 zvol_rename_minor(zv, name);
2307 mutex_exit(&zv->zv_state_lock);
2310 rw_exit(&zvol_state_lock);
2313 typedef struct zvol_snapdev_cb_arg {
2315 } zvol_snapdev_cb_arg_t;
2318 zvol_set_snapdev_cb(const char *dsname, void *param)
2320 zvol_snapdev_cb_arg_t *arg = param;
2322 if (strchr(dsname, '@') == NULL)
2325 switch (arg->snapdev) {
2326 case ZFS_SNAPDEV_VISIBLE:
2327 (void) zvol_create_minor_impl(dsname);
2329 case ZFS_SNAPDEV_HIDDEN:
2330 (void) zvol_remove_minor_impl(dsname);
2338 zvol_set_snapdev_impl(char *name, uint64_t snapdev)
2340 zvol_snapdev_cb_arg_t arg = {snapdev};
2341 fstrans_cookie_t cookie = spl_fstrans_mark();
2343 * The zvol_set_snapdev_sync() sets snapdev appropriately
2344 * in the dataset hierarchy. Here, we only scan snapshots.
2346 dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS);
2347 spl_fstrans_unmark(cookie);
2350 typedef struct zvol_volmode_cb_arg {
2352 } zvol_volmode_cb_arg_t;
2355 zvol_set_volmode_impl(char *name, uint64_t volmode)
2357 fstrans_cookie_t cookie = spl_fstrans_mark();
2359 if (strchr(name, '@') != NULL)
2363 * It's unfortunate we need to remove minors before we create new ones:
2364 * this is necessary because our backing gendisk (zvol_state->zv_disk)
2365 * coule be different when we set, for instance, volmode from "geom"
2366 * to "dev" (or vice versa).
2367 * A possible optimization is to modify our consumers so we don't get
2368 * called when "volmode" does not change.
2371 case ZFS_VOLMODE_NONE:
2372 (void) zvol_remove_minor_impl(name);
2374 case ZFS_VOLMODE_GEOM:
2375 case ZFS_VOLMODE_DEV:
2376 (void) zvol_remove_minor_impl(name);
2377 (void) zvol_create_minor_impl(name);
2379 case ZFS_VOLMODE_DEFAULT:
2380 (void) zvol_remove_minor_impl(name);
2381 if (zvol_volmode == ZFS_VOLMODE_NONE)
2383 else /* if zvol_volmode is invalid defaults to "geom" */
2384 (void) zvol_create_minor_impl(name);
2388 spl_fstrans_unmark(cookie);
2391 static zvol_task_t *
2392 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2,
2398 /* Never allow tasks on hidden names. */
2399 if (name1[0] == '$')
2402 task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP);
2404 task->value = value;
2405 delim = strchr(name1, '/');
2406 strlcpy(task->pool, name1, delim ? (delim - name1 + 1) : MAXNAMELEN);
2408 strlcpy(task->name1, name1, MAXNAMELEN);
2410 strlcpy(task->name2, name2, MAXNAMELEN);
2416 zvol_task_free(zvol_task_t *task)
2418 kmem_free(task, sizeof (zvol_task_t));
2422 * The worker thread function performed asynchronously.
2425 zvol_task_cb(void *param)
2427 zvol_task_t *task = (zvol_task_t *)param;
2430 case ZVOL_ASYNC_CREATE_MINORS:
2431 (void) zvol_create_minors_impl(task->name1);
2433 case ZVOL_ASYNC_REMOVE_MINORS:
2434 zvol_remove_minors_impl(task->name1);
2436 case ZVOL_ASYNC_RENAME_MINORS:
2437 zvol_rename_minors_impl(task->name1, task->name2);
2439 case ZVOL_ASYNC_SET_SNAPDEV:
2440 zvol_set_snapdev_impl(task->name1, task->value);
2442 case ZVOL_ASYNC_SET_VOLMODE:
2443 zvol_set_volmode_impl(task->name1, task->value);
2450 zvol_task_free(task);
2453 typedef struct zvol_set_prop_int_arg {
2454 const char *zsda_name;
2455 uint64_t zsda_value;
2456 zprop_source_t zsda_source;
2458 } zvol_set_prop_int_arg_t;
2461 * Sanity check the dataset for safe use by the sync task. No additional
2462 * conditions are imposed.
2465 zvol_set_snapdev_check(void *arg, dmu_tx_t *tx)
2467 zvol_set_prop_int_arg_t *zsda = arg;
2468 dsl_pool_t *dp = dmu_tx_pool(tx);
2472 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL);
2476 dsl_dir_rele(dd, FTAG);
2483 zvol_set_snapdev_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2485 char dsname[MAXNAMELEN];
2489 dsl_dataset_name(ds, dsname);
2490 if (dsl_prop_get_int_ds(ds, "snapdev", &snapdev) != 0)
2492 task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname, NULL, snapdev);
2496 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb,
2502 * Traverse all child datasets and apply snapdev appropriately.
2503 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2504 * dataset and read the effective "snapdev" on every child in the callback
2505 * function: this is because the value is not guaranteed to be the same in the
2506 * whole dataset hierarchy.
2509 zvol_set_snapdev_sync(void *arg, dmu_tx_t *tx)
2511 zvol_set_prop_int_arg_t *zsda = arg;
2512 dsl_pool_t *dp = dmu_tx_pool(tx);
2517 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL));
2520 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds);
2522 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_SNAPDEV),
2523 zsda->zsda_source, sizeof (zsda->zsda_value), 1,
2524 &zsda->zsda_value, zsda->zsda_tx);
2525 dsl_dataset_rele(ds, FTAG);
2527 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_snapdev_sync_cb,
2528 zsda, DS_FIND_CHILDREN);
2530 dsl_dir_rele(dd, FTAG);
2534 zvol_set_snapdev(const char *ddname, zprop_source_t source, uint64_t snapdev)
2536 zvol_set_prop_int_arg_t zsda;
2538 zsda.zsda_name = ddname;
2539 zsda.zsda_source = source;
2540 zsda.zsda_value = snapdev;
2542 return (dsl_sync_task(ddname, zvol_set_snapdev_check,
2543 zvol_set_snapdev_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
2547 * Sanity check the dataset for safe use by the sync task. No additional
2548 * conditions are imposed.
2551 zvol_set_volmode_check(void *arg, dmu_tx_t *tx)
2553 zvol_set_prop_int_arg_t *zsda = arg;
2554 dsl_pool_t *dp = dmu_tx_pool(tx);
2558 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL);
2562 dsl_dir_rele(dd, FTAG);
2569 zvol_set_volmode_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2571 char dsname[MAXNAMELEN];
2575 dsl_dataset_name(ds, dsname);
2576 if (dsl_prop_get_int_ds(ds, "volmode", &volmode) != 0)
2578 task = zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE, dsname, NULL, volmode);
2582 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb,
2588 * Traverse all child datasets and apply volmode appropriately.
2589 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2590 * dataset and read the effective "volmode" on every child in the callback
2591 * function: this is because the value is not guaranteed to be the same in the
2592 * whole dataset hierarchy.
2595 zvol_set_volmode_sync(void *arg, dmu_tx_t *tx)
2597 zvol_set_prop_int_arg_t *zsda = arg;
2598 dsl_pool_t *dp = dmu_tx_pool(tx);
2603 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL));
2606 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds);
2608 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_VOLMODE),
2609 zsda->zsda_source, sizeof (zsda->zsda_value), 1,
2610 &zsda->zsda_value, zsda->zsda_tx);
2611 dsl_dataset_rele(ds, FTAG);
2614 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_volmode_sync_cb,
2615 zsda, DS_FIND_CHILDREN);
2617 dsl_dir_rele(dd, FTAG);
2621 zvol_set_volmode(const char *ddname, zprop_source_t source, uint64_t volmode)
2623 zvol_set_prop_int_arg_t zsda;
2625 zsda.zsda_name = ddname;
2626 zsda.zsda_source = source;
2627 zsda.zsda_value = volmode;
2629 return (dsl_sync_task(ddname, zvol_set_volmode_check,
2630 zvol_set_volmode_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
2634 zvol_create_minors(spa_t *spa, const char *name, boolean_t async)
2639 task = zvol_task_alloc(ZVOL_ASYNC_CREATE_MINORS, name, NULL, ~0ULL);
2643 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
2644 if ((async == B_FALSE) && (id != TASKQID_INVALID))
2645 taskq_wait_id(spa->spa_zvol_taskq, id);
2649 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async)
2654 task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL);
2658 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
2659 if ((async == B_FALSE) && (id != TASKQID_INVALID))
2660 taskq_wait_id(spa->spa_zvol_taskq, id);
2664 zvol_rename_minors(spa_t *spa, const char *name1, const char *name2,
2670 task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL);
2674 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
2675 if ((async == B_FALSE) && (id != TASKQID_INVALID))
2676 taskq_wait_id(spa->spa_zvol_taskq, id);
2682 int threads = MIN(MAX(zvol_threads, 1), 1024);
2685 list_create(&zvol_state_list, sizeof (zvol_state_t),
2686 offsetof(zvol_state_t, zv_next));
2687 rw_init(&zvol_state_lock, NULL, RW_DEFAULT, NULL);
2688 ida_init(&zvol_ida);
2690 zvol_taskq = taskq_create(ZVOL_DRIVER, threads, maxclsyspri,
2691 threads * 2, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC);
2692 if (zvol_taskq == NULL) {
2693 printk(KERN_INFO "ZFS: taskq_create() failed\n");
2698 zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head),
2704 for (i = 0; i < ZVOL_HT_SIZE; i++)
2705 INIT_HLIST_HEAD(&zvol_htable[i]);
2707 error = register_blkdev(zvol_major, ZVOL_DRIVER);
2709 printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error);
2713 blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS,
2714 THIS_MODULE, zvol_probe, NULL, NULL);
2719 kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
2721 taskq_destroy(zvol_taskq);
2723 ida_destroy(&zvol_ida);
2724 rw_destroy(&zvol_state_lock);
2725 list_destroy(&zvol_state_list);
2727 return (SET_ERROR(error));
2733 zvol_remove_minors_impl(NULL);
2735 blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS);
2736 unregister_blkdev(zvol_major, ZVOL_DRIVER);
2737 kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
2739 taskq_destroy(zvol_taskq);
2740 list_destroy(&zvol_state_list);
2741 rw_destroy(&zvol_state_lock);
2743 ida_destroy(&zvol_ida);
2747 module_param(zvol_inhibit_dev, uint, 0644);
2748 MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes");
2750 module_param(zvol_major, uint, 0444);
2751 MODULE_PARM_DESC(zvol_major, "Major number for zvol device");
2753 module_param(zvol_threads, uint, 0444);
2754 MODULE_PARM_DESC(zvol_threads, "Max number of threads to handle I/O requests");
2756 module_param(zvol_request_sync, uint, 0644);
2757 MODULE_PARM_DESC(zvol_request_sync, "Synchronously handle bio requests");
2759 module_param(zvol_max_discard_blocks, ulong, 0444);
2760 MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard");
2762 module_param(zvol_prefetch_bytes, uint, 0644);
2763 MODULE_PARM_DESC(zvol_prefetch_bytes, "Prefetch N bytes at zvol start+end");
2765 module_param(zvol_volmode, uint, 0644);
2766 MODULE_PARM_DESC(zvol_volmode, "Default volmode property value");