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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
25 * All rights reserved.
27 * Portions Copyright 2010 Robert Milkowski
29 * Copyright 2017 Nexenta Systems, Inc. All rights reserved.
30 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
31 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32 * Copyright (c) 2014 Integros [integros.com]
35 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
38 * ZFS volume emulation driver.
40 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
41 * Volumes are accessed through the symbolic links named:
43 * /dev/zvol/dsk/<pool_name>/<dataset_name>
44 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
46 * These links are created by the /dev filesystem (sdev_zvolops.c).
47 * Volumes are persistent through reboot. No user command needs to be
48 * run before opening and using a device.
51 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
55 #include <sys/types.h>
56 #include <sys/param.h>
57 #include <sys/kernel.h>
58 #include <sys/errno.h>
64 #include <sys/cmn_err.h>
68 #include <sys/spa_impl.h>
71 #include <sys/dmu_traverse.h>
72 #include <sys/dnode.h>
73 #include <sys/dsl_dataset.h>
74 #include <sys/dsl_prop.h>
76 #include <sys/byteorder.h>
77 #include <sys/sunddi.h>
78 #include <sys/dirent.h>
79 #include <sys/policy.h>
80 #include <sys/queue.h>
81 #include <sys/fs/zfs.h>
82 #include <sys/zfs_ioctl.h>
84 #include <sys/refcount.h>
85 #include <sys/zfs_znode.h>
86 #include <sys/zfs_rlock.h>
87 #include <sys/vdev_impl.h>
88 #include <sys/vdev_raidz.h>
90 #include <sys/zil_impl.h>
92 #include <sys/dmu_tx.h>
93 #include <sys/zfeature.h>
94 #include <sys/zio_checksum.h>
95 #include <sys/zil_impl.h>
96 #include <sys/filio.h>
98 #include <geom/geom.h>
100 #include "zfs_namecheck.h"
103 struct g_class zfs_zvol_class = {
105 .version = G_VERSION,
108 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
112 static char *zvol_tag = "zvol_tag";
114 #define ZVOL_DUMPSIZE "dumpsize"
117 * This lock protects the zfsdev_state structure from being modified
118 * while it's being used, e.g. an open that comes in before a create
119 * finishes. It also protects temporary opens of the dataset so that,
120 * e.g., an open doesn't get a spurious EBUSY.
123 kmutex_t zfsdev_state_lock;
126 * In FreeBSD we've replaced the upstream zfsdev_state_lock with the
127 * spa_namespace_lock in the ZVOL code.
129 #define zfsdev_state_lock spa_namespace_lock
131 static uint32_t zvol_minors;
134 SYSCTL_DECL(_vfs_zfs);
135 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
136 static int volmode = ZFS_VOLMODE_GEOM;
137 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
138 "Expose as GEOM providers (1), device files (2) or neither");
139 static boolean_t zpool_on_zvol = B_FALSE;
140 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, recursive, CTLFLAG_RWTUN, &zpool_on_zvol, 0,
141 "Allow zpools to use zvols as vdevs (DANGEROUS)");
144 typedef struct zvol_extent {
146 dva_t ze_dva; /* dva associated with this extent */
147 uint64_t ze_nblks; /* number of blocks in extent */
151 * The in-core state of each volume.
153 typedef struct zvol_state {
155 LIST_ENTRY(zvol_state) zv_links;
157 char zv_name[MAXPATHLEN]; /* pool/dd name */
158 uint64_t zv_volsize; /* amount of space we advertise */
159 uint64_t zv_volblocksize; /* volume block size */
161 minor_t zv_minor; /* minor number */
163 struct cdev *zv_dev; /* non-GEOM device */
164 struct g_provider *zv_provider; /* GEOM provider */
166 uint8_t zv_min_bs; /* minimum addressable block shift */
167 uint8_t zv_flags; /* readonly, dumpified, etc. */
168 objset_t *zv_objset; /* objset handle */
170 uint32_t zv_open_count[OTYPCNT]; /* open counts */
172 uint32_t zv_total_opens; /* total open count */
173 uint32_t zv_sync_cnt; /* synchronous open count */
174 zilog_t *zv_zilog; /* ZIL handle */
175 list_t zv_extents; /* List of extents for dump */
176 znode_t zv_znode; /* for range locking */
177 dnode_t *zv_dn; /* dnode hold */
180 int zv_volmode; /* Provide GEOM or cdev */
181 struct bio_queue_head zv_queue;
182 struct mtx zv_queue_mtx; /* zv_queue mutex */
187 static LIST_HEAD(, zvol_state) all_zvols;
190 * zvol specific flags
192 #define ZVOL_RDONLY 0x1
193 #define ZVOL_DUMPIFIED 0x2
194 #define ZVOL_EXCL 0x4
198 * zvol maximum transfer in one DMU tx.
200 int zvol_maxphys = DMU_MAX_ACCESS/2;
203 * Toggle unmap functionality.
205 boolean_t zvol_unmap_enabled = B_TRUE;
208 * If true, unmaps requested as synchronous are executed synchronously,
209 * otherwise all unmaps are asynchronous.
211 boolean_t zvol_unmap_sync_enabled = B_FALSE;
214 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
215 &zvol_unmap_enabled, 0,
216 "Enable UNMAP functionality");
218 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_sync_enabled, CTLFLAG_RWTUN,
219 &zvol_unmap_sync_enabled, 0,
220 "UNMAPs requested as sync are executed synchronously");
222 static d_open_t zvol_d_open;
223 static d_close_t zvol_d_close;
224 static d_read_t zvol_read;
225 static d_write_t zvol_write;
226 static d_ioctl_t zvol_d_ioctl;
227 static d_strategy_t zvol_strategy;
229 static struct cdevsw zvol_cdevsw = {
230 .d_version = D_VERSION,
231 .d_open = zvol_d_open,
232 .d_close = zvol_d_close,
234 .d_write = zvol_write,
235 .d_ioctl = zvol_d_ioctl,
236 .d_strategy = zvol_strategy,
238 .d_flags = D_DISK | D_TRACKCLOSE,
241 static void zvol_geom_run(zvol_state_t *zv);
242 static void zvol_geom_destroy(zvol_state_t *zv);
243 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
244 static void zvol_geom_start(struct bio *bp);
245 static void zvol_geom_worker(void *arg);
246 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
247 uint64_t len, boolean_t sync);
248 #endif /* !illumos */
250 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
251 nvlist_t *, nvlist_t *);
252 static int zvol_remove_zv(zvol_state_t *);
253 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
254 struct lwb *lwb, zio_t *zio);
255 static int zvol_dumpify(zvol_state_t *zv);
256 static int zvol_dump_fini(zvol_state_t *zv);
257 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
260 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
263 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
265 zv->zv_volsize = volsize;
266 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
267 "Size", volsize) == DDI_SUCCESS);
268 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
269 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
271 /* Notify specfs to invalidate the cached size */
272 spec_size_invalidate(dev, VBLK);
273 spec_size_invalidate(dev, VCHR);
275 zv->zv_volsize = volsize;
276 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
277 struct g_provider *pp;
279 pp = zv->zv_provider;
285 * Do not invoke resize event when initial size was zero.
286 * ZVOL initializes the size on first open, this is not
289 if (pp->mediasize == 0)
290 pp->mediasize = zv->zv_volsize;
292 g_resize_provider(pp, zv->zv_volsize);
299 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
302 return (SET_ERROR(EINVAL));
304 if (volsize % blocksize != 0)
305 return (SET_ERROR(EINVAL));
308 if (volsize - 1 > SPEC_MAXOFFSET_T)
309 return (SET_ERROR(EOVERFLOW));
315 zvol_check_volblocksize(uint64_t volblocksize)
317 if (volblocksize < SPA_MINBLOCKSIZE ||
318 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
320 return (SET_ERROR(EDOM));
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);
336 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
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);
348 static zvol_state_t *
349 zvol_minor_lookup(const char *name)
356 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
359 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
360 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
364 LIST_FOREACH(zv, &all_zvols, zv_links) {
366 if (strcmp(zv->zv_name, name) == 0)
373 /* extent mapping arg */
381 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
382 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
384 struct maparg *ma = arg;
386 int bs = ma->ma_zv->zv_volblocksize;
388 if (bp == NULL || BP_IS_HOLE(bp) ||
389 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
392 VERIFY(!BP_IS_EMBEDDED(bp));
394 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
397 /* Abort immediately if we have encountered gang blocks */
399 return (SET_ERROR(EFRAGS));
402 * See if the block is at the end of the previous extent.
404 ze = list_tail(&ma->ma_zv->zv_extents);
406 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
407 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
408 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
413 dprintf_bp(bp, "%s", "next blkptr:");
415 /* start a new extent */
416 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
417 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
419 list_insert_tail(&ma->ma_zv->zv_extents, ze);
424 zvol_free_extents(zvol_state_t *zv)
428 while (ze = list_head(&zv->zv_extents)) {
429 list_remove(&zv->zv_extents, ze);
430 kmem_free(ze, sizeof (zvol_extent_t));
435 zvol_get_lbas(zvol_state_t *zv)
437 objset_t *os = zv->zv_objset;
443 zvol_free_extents(zv);
445 /* commit any in-flight changes before traversing the dataset */
446 txg_wait_synced(dmu_objset_pool(os), 0);
447 err = traverse_dataset(dmu_objset_ds(os), 0,
448 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
449 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
450 zvol_free_extents(zv);
451 return (err ? err : EIO);
459 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
461 zfs_creat_t *zct = arg;
462 nvlist_t *nvprops = zct->zct_props;
464 uint64_t volblocksize, volsize;
466 VERIFY(nvlist_lookup_uint64(nvprops,
467 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
468 if (nvlist_lookup_uint64(nvprops,
469 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
470 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
473 * These properties must be removed from the list so the generic
474 * property setting step won't apply to them.
476 VERIFY(nvlist_remove_all(nvprops,
477 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
478 (void) nvlist_remove_all(nvprops,
479 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
481 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
485 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
489 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
494 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
495 * implement DKIOCFREE/free-long-range.
498 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
500 zvol_state_t *zv = arg1;
501 lr_truncate_t *lr = arg2;
502 uint64_t offset, length;
505 byteswap_uint64_array(lr, sizeof (*lr));
507 offset = lr->lr_offset;
508 length = lr->lr_length;
510 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
514 * Replay a TX_WRITE ZIL transaction that didn't get committed
515 * after a system failure
518 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
520 zvol_state_t *zv = arg1;
521 lr_write_t *lr = arg2;
522 objset_t *os = zv->zv_objset;
523 char *data = (char *)(lr + 1); /* data follows lr_write_t */
524 uint64_t offset, length;
529 byteswap_uint64_array(lr, sizeof (*lr));
531 offset = lr->lr_offset;
532 length = lr->lr_length;
534 /* If it's a dmu_sync() block, write the whole block */
535 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
536 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
537 if (length < blocksize) {
538 offset -= offset % blocksize;
543 tx = dmu_tx_create(os);
544 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
545 error = dmu_tx_assign(tx, TXG_WAIT);
549 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
558 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
560 return (SET_ERROR(ENOTSUP));
564 * Callback vectors for replaying records.
565 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
567 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
568 zvol_replay_err, /* 0 no such transaction type */
569 zvol_replay_err, /* TX_CREATE */
570 zvol_replay_err, /* TX_MKDIR */
571 zvol_replay_err, /* TX_MKXATTR */
572 zvol_replay_err, /* TX_SYMLINK */
573 zvol_replay_err, /* TX_REMOVE */
574 zvol_replay_err, /* TX_RMDIR */
575 zvol_replay_err, /* TX_LINK */
576 zvol_replay_err, /* TX_RENAME */
577 zvol_replay_write, /* TX_WRITE */
578 zvol_replay_truncate, /* TX_TRUNCATE */
579 zvol_replay_err, /* TX_SETATTR */
580 zvol_replay_err, /* TX_ACL */
581 zvol_replay_err, /* TX_CREATE_ACL */
582 zvol_replay_err, /* TX_CREATE_ATTR */
583 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
584 zvol_replay_err, /* TX_MKDIR_ACL */
585 zvol_replay_err, /* TX_MKDIR_ATTR */
586 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
587 zvol_replay_err, /* TX_WRITE2 */
592 zvol_name2minor(const char *name, minor_t *minor)
596 mutex_enter(&zfsdev_state_lock);
597 zv = zvol_minor_lookup(name);
599 *minor = zv->zv_minor;
600 mutex_exit(&zfsdev_state_lock);
601 return (zv ? 0 : -1);
606 * Create a minor node (plus a whole lot more) for the specified volume.
609 zvol_create_minor(const char *name)
611 zfs_soft_state_t *zs;
615 dmu_object_info_t doi;
617 char chrbuf[30], blkbuf[30];
619 struct g_provider *pp;
626 ZFS_LOG(1, "Creating ZVOL %s...", name);
629 mutex_enter(&zfsdev_state_lock);
631 if (zvol_minor_lookup(name) != NULL) {
632 mutex_exit(&zfsdev_state_lock);
633 return (SET_ERROR(EEXIST));
636 /* lie and say we're read-only */
637 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
640 mutex_exit(&zfsdev_state_lock);
645 if ((minor = zfsdev_minor_alloc()) == 0) {
646 dmu_objset_disown(os, FTAG);
647 mutex_exit(&zfsdev_state_lock);
648 return (SET_ERROR(ENXIO));
651 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
652 dmu_objset_disown(os, FTAG);
653 mutex_exit(&zfsdev_state_lock);
654 return (SET_ERROR(EAGAIN));
656 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
659 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
661 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
662 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
663 ddi_soft_state_free(zfsdev_state, minor);
664 dmu_objset_disown(os, FTAG);
665 mutex_exit(&zfsdev_state_lock);
666 return (SET_ERROR(EAGAIN));
669 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
671 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
672 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
673 ddi_remove_minor_node(zfs_dip, chrbuf);
674 ddi_soft_state_free(zfsdev_state, minor);
675 dmu_objset_disown(os, FTAG);
676 mutex_exit(&zfsdev_state_lock);
677 return (SET_ERROR(EAGAIN));
680 zs = ddi_get_soft_state(zfsdev_state, minor);
681 zs->zss_type = ZSST_ZVOL;
682 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
685 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
687 error = dsl_prop_get_integer(name,
688 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
689 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
693 zv->zv_volmode = mode;
694 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
696 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
697 gp->start = zvol_geom_start;
698 gp->access = zvol_geom_access;
699 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
700 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
701 pp->sectorsize = DEV_BSIZE;
705 zv->zv_provider = pp;
706 bioq_init(&zv->zv_queue);
707 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
708 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
709 struct make_dev_args args;
711 make_dev_args_init(&args);
712 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
713 args.mda_devsw = &zvol_cdevsw;
715 args.mda_uid = UID_ROOT;
716 args.mda_gid = GID_OPERATOR;
717 args.mda_mode = 0640;
718 args.mda_si_drv2 = zv;
719 error = make_dev_s(&args, &zv->zv_dev,
720 "%s/%s", ZVOL_DRIVER, name);
722 kmem_free(zv, sizeof(*zv));
723 dmu_objset_disown(os, FTAG);
724 mutex_exit(&zfsdev_state_lock);
727 zv->zv_dev->si_iosize_max = MAXPHYS;
729 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
732 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
733 zv->zv_min_bs = DEV_BSHIFT;
735 zv->zv_minor = minor;
738 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
739 zv->zv_flags |= ZVOL_RDONLY;
740 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
741 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
742 sizeof (rl_t), offsetof(rl_t, r_node));
743 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
744 offsetof(zvol_extent_t, ze_node));
746 /* get and cache the blocksize */
747 error = dmu_object_info(os, ZVOL_OBJ, &doi);
749 zv->zv_volblocksize = doi.doi_data_block_size;
752 if (spa_writeable(dmu_objset_spa(os))) {
753 if (zil_replay_disable)
754 zil_destroy(dmu_objset_zil(os), B_FALSE);
756 zil_replay(os, zv, zvol_replay_vector);
758 dmu_objset_disown(os, FTAG);
759 zv->zv_objset = NULL;
763 mutex_exit(&zfsdev_state_lock);
765 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
771 ZFS_LOG(1, "ZVOL %s created.", name);
778 * Remove minor node for the specified volume.
781 zvol_remove_zv(zvol_state_t *zv)
785 minor_t minor = zv->zv_minor;
788 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
789 if (zv->zv_total_opens != 0)
790 return (SET_ERROR(EBUSY));
793 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
794 ddi_remove_minor_node(zfs_dip, nmbuf);
796 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
797 ddi_remove_minor_node(zfs_dip, nmbuf);
799 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
801 LIST_REMOVE(zv, zv_links);
802 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
804 zvol_geom_destroy(zv);
806 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
807 if (zv->zv_dev != NULL)
808 destroy_dev(zv->zv_dev);
812 avl_destroy(&zv->zv_znode.z_range_avl);
813 mutex_destroy(&zv->zv_znode.z_range_lock);
815 kmem_free(zv, sizeof (zvol_state_t));
817 ddi_soft_state_free(zfsdev_state, minor);
824 zvol_remove_minor(const char *name)
829 mutex_enter(&zfsdev_state_lock);
830 if ((zv = zvol_minor_lookup(name)) == NULL) {
831 mutex_exit(&zfsdev_state_lock);
832 return (SET_ERROR(ENXIO));
834 rc = zvol_remove_zv(zv);
835 mutex_exit(&zfsdev_state_lock);
840 zvol_first_open(zvol_state_t *zv)
842 dmu_object_info_t doi;
848 /* lie and say we're read-only */
849 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
855 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
858 dmu_objset_disown(os, zvol_tag);
862 /* get and cache the blocksize */
863 error = dmu_object_info(os, ZVOL_OBJ, &doi);
866 dmu_objset_disown(os, zvol_tag);
869 zv->zv_volblocksize = doi.doi_data_block_size;
871 error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
873 dmu_objset_disown(os, zvol_tag);
877 zvol_size_changed(zv, volsize);
878 zv->zv_zilog = zil_open(os, zvol_get_data);
880 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
882 if (readonly || dmu_objset_is_snapshot(os) ||
883 !spa_writeable(dmu_objset_spa(os)))
884 zv->zv_flags |= ZVOL_RDONLY;
886 zv->zv_flags &= ~ZVOL_RDONLY;
891 zvol_last_close(zvol_state_t *zv)
893 zil_close(zv->zv_zilog);
896 dnode_rele(zv->zv_dn, zvol_tag);
902 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
903 !(zv->zv_flags & ZVOL_RDONLY))
904 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
905 dmu_objset_evict_dbufs(zv->zv_objset);
907 dmu_objset_disown(zv->zv_objset, zvol_tag);
908 zv->zv_objset = NULL;
913 zvol_prealloc(zvol_state_t *zv)
915 objset_t *os = zv->zv_objset;
917 uint64_t refd, avail, usedobjs, availobjs;
918 uint64_t resid = zv->zv_volsize;
921 /* Check the space usage before attempting to allocate the space */
922 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
923 if (avail < zv->zv_volsize)
924 return (SET_ERROR(ENOSPC));
926 /* Free old extents if they exist */
927 zvol_free_extents(zv);
931 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
933 tx = dmu_tx_create(os);
934 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
935 error = dmu_tx_assign(tx, TXG_WAIT);
938 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
941 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
946 txg_wait_synced(dmu_objset_pool(os), 0);
953 zvol_update_volsize(objset_t *os, uint64_t volsize)
958 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
960 tx = dmu_tx_create(os);
961 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
962 dmu_tx_mark_netfree(tx);
963 error = dmu_tx_assign(tx, TXG_WAIT);
969 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
974 error = dmu_free_long_range(os,
975 ZVOL_OBJ, volsize, DMU_OBJECT_END);
980 zvol_remove_minors(const char *name)
987 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
988 (void) strncpy(namebuf, name, strlen(name));
989 (void) strcat(namebuf, "/");
990 mutex_enter(&zfsdev_state_lock);
991 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
993 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
996 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
997 (void) zvol_remove_zv(zv);
999 kmem_free(namebuf, strlen(name) + 2);
1001 mutex_exit(&zfsdev_state_lock);
1002 #else /* !illumos */
1003 zvol_state_t *zv, *tzv;
1006 namelen = strlen(name);
1009 mutex_enter(&zfsdev_state_lock);
1011 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
1012 if (strcmp(zv->zv_name, name) == 0 ||
1013 (strncmp(zv->zv_name, name, namelen) == 0 &&
1014 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
1015 zv->zv_name[namelen] == '@'))) {
1016 (void) zvol_remove_zv(zv);
1020 mutex_exit(&zfsdev_state_lock);
1022 #endif /* illumos */
1026 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
1028 uint64_t old_volsize = 0ULL;
1031 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1034 * Reinitialize the dump area to the new size. If we
1035 * failed to resize the dump area then restore it back to
1036 * its original size. We must set the new volsize prior
1037 * to calling dumpvp_resize() to ensure that the devices'
1038 * size(9P) is not visible by the dump subsystem.
1040 old_volsize = zv->zv_volsize;
1041 zvol_size_changed(zv, volsize);
1044 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1045 if ((error = zvol_dumpify(zv)) != 0 ||
1046 (error = dumpvp_resize()) != 0) {
1049 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1050 zvol_size_changed(zv, old_volsize);
1051 dumpify_error = zvol_dumpify(zv);
1052 error = dumpify_error ? dumpify_error : error;
1055 #endif /* ZVOL_DUMP */
1059 * Generate a LUN expansion event.
1064 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1066 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1069 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1070 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1072 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1073 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1076 kmem_free(physpath, MAXPATHLEN);
1078 #endif /* illumos */
1083 zvol_set_volsize(const char *name, uint64_t volsize)
1085 zvol_state_t *zv = NULL;
1088 dmu_object_info_t doi;
1090 boolean_t owned = B_FALSE;
1092 error = dsl_prop_get_integer(name,
1093 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1097 return (SET_ERROR(EROFS));
1099 mutex_enter(&zfsdev_state_lock);
1100 zv = zvol_minor_lookup(name);
1102 if (zv == NULL || zv->zv_objset == NULL) {
1103 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1105 mutex_exit(&zfsdev_state_lock);
1115 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1116 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1119 error = zvol_update_volsize(os, volsize);
1121 if (error == 0 && zv != NULL)
1122 error = zvol_update_live_volsize(zv, volsize);
1125 dmu_objset_disown(os, FTAG);
1127 zv->zv_objset = NULL;
1129 mutex_exit(&zfsdev_state_lock);
1136 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1139 zvol_open(struct g_provider *pp, int flag, int count)
1146 mutex_enter(&zfsdev_state_lock);
1148 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1150 mutex_exit(&zfsdev_state_lock);
1151 return (SET_ERROR(ENXIO));
1154 if (zv->zv_total_opens == 0)
1155 err = zvol_first_open(zv);
1157 mutex_exit(&zfsdev_state_lock);
1160 #else /* !illumos */
1161 boolean_t locked = B_FALSE;
1163 if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1165 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1166 * attempting to probe geom providers while looking for a
1167 * replacement for a missing VDEV. In this case, the
1168 * spa_namespace_lock will not be held, but it is still illegal
1169 * to use a zvol as a vdev. Deadlocks can result if another
1170 * thread has spa_namespace_lock
1172 return (EOPNOTSUPP);
1175 * Protect against recursively entering spa_namespace_lock
1176 * when spa_open() is used for a pool on a (local) ZVOL(s).
1177 * This is needed since we replaced upstream zfsdev_state_lock
1178 * with spa_namespace_lock in the ZVOL code.
1179 * We are using the same trick as spa_open().
1180 * Note that calls in zvol_first_open which need to resolve
1181 * pool name to a spa object will enter spa_open()
1182 * recursively, but that function already has all the
1183 * necessary protection.
1185 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1186 mutex_enter(&zfsdev_state_lock);
1193 mutex_exit(&zfsdev_state_lock);
1194 return (SET_ERROR(ENXIO));
1197 if (zv->zv_total_opens == 0) {
1198 err = zvol_first_open(zv);
1201 mutex_exit(&zfsdev_state_lock);
1204 pp->mediasize = zv->zv_volsize;
1205 pp->stripeoffset = 0;
1206 pp->stripesize = zv->zv_volblocksize;
1208 #endif /* illumos */
1209 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1210 err = SET_ERROR(EROFS);
1213 if (zv->zv_flags & ZVOL_EXCL) {
1214 err = SET_ERROR(EBUSY);
1219 if (zv->zv_total_opens != 0) {
1220 err = SET_ERROR(EBUSY);
1223 zv->zv_flags |= ZVOL_EXCL;
1228 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1229 zv->zv_open_count[otyp]++;
1230 zv->zv_total_opens++;
1232 mutex_exit(&zfsdev_state_lock);
1234 zv->zv_total_opens += count;
1236 mutex_exit(&zfsdev_state_lock);
1241 if (zv->zv_total_opens == 0)
1242 zvol_last_close(zv);
1244 mutex_exit(&zfsdev_state_lock);
1247 mutex_exit(&zfsdev_state_lock);
1255 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1257 minor_t minor = getminor(dev);
1261 mutex_enter(&zfsdev_state_lock);
1263 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1265 mutex_exit(&zfsdev_state_lock);
1266 #else /* !illumos */
1268 zvol_close(struct g_provider *pp, int flag, int count)
1272 boolean_t locked = B_FALSE;
1274 /* See comment in zvol_open(). */
1275 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1276 mutex_enter(&zfsdev_state_lock);
1283 mutex_exit(&zfsdev_state_lock);
1284 #endif /* illumos */
1285 return (SET_ERROR(ENXIO));
1288 if (zv->zv_flags & ZVOL_EXCL) {
1289 ASSERT(zv->zv_total_opens == 1);
1290 zv->zv_flags &= ~ZVOL_EXCL;
1294 * If the open count is zero, this is a spurious close.
1295 * That indicates a bug in the kernel / DDI framework.
1298 ASSERT(zv->zv_open_count[otyp] != 0);
1300 ASSERT(zv->zv_total_opens != 0);
1303 * You may get multiple opens, but only one close.
1306 zv->zv_open_count[otyp]--;
1307 zv->zv_total_opens--;
1309 zv->zv_total_opens -= count;
1312 if (zv->zv_total_opens == 0)
1313 zvol_last_close(zv);
1316 mutex_exit(&zfsdev_state_lock);
1319 mutex_exit(&zfsdev_state_lock);
1326 zvol_get_done(zgd_t *zgd, int error)
1329 dmu_buf_rele(zgd->zgd_db, zgd);
1331 zfs_range_unlock(zgd->zgd_rl);
1333 kmem_free(zgd, sizeof (zgd_t));
1337 * Get data to generate a TX_WRITE intent log record.
1340 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1342 zvol_state_t *zv = arg;
1343 uint64_t offset = lr->lr_offset;
1344 uint64_t size = lr->lr_length; /* length of user data */
1349 ASSERT3P(lwb, !=, NULL);
1350 ASSERT3P(zio, !=, NULL);
1351 ASSERT3U(size, !=, 0);
1353 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1357 * Write records come in two flavors: immediate and indirect.
1358 * For small writes it's cheaper to store the data with the
1359 * log record (immediate); for large writes it's cheaper to
1360 * sync the data and get a pointer to it (indirect) so that
1361 * we don't have to write the data twice.
1363 if (buf != NULL) { /* immediate write */
1364 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1366 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1367 DMU_READ_NO_PREFETCH);
1368 } else { /* indirect write */
1370 * Have to lock the whole block to ensure when it's written out
1371 * and its checksum is being calculated that no one can change
1372 * the data. Contrarily to zfs_get_data we need not re-check
1373 * blocksize after we get the lock because it cannot be changed.
1375 size = zv->zv_volblocksize;
1376 offset = P2ALIGN(offset, size);
1377 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1379 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1380 DMU_READ_NO_PREFETCH);
1382 blkptr_t *bp = &lr->lr_blkptr;
1387 ASSERT(db->db_offset == offset);
1388 ASSERT(db->db_size == size);
1390 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1391 zvol_get_done, zgd);
1398 zvol_get_done(zgd, error);
1404 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1406 * We store data in the log buffers if it's small enough.
1407 * Otherwise we will later flush the data out via dmu_sync().
1409 ssize_t zvol_immediate_write_sz = 32768;
1411 SYSCTL_LONG(_vfs_zfs_vol, OID_AUTO, immediate_write_sz, CTLFLAG_RWTUN,
1412 &zvol_immediate_write_sz, 0, "Minimal size for indirect log write");
1416 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1419 uint32_t blocksize = zv->zv_volblocksize;
1420 zilog_t *zilog = zv->zv_zilog;
1421 itx_wr_state_t write_state;
1423 if (zil_replaying(zilog, tx))
1426 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1427 write_state = WR_INDIRECT;
1428 else if (!spa_has_slogs(zilog->zl_spa) &&
1429 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1430 write_state = WR_INDIRECT;
1432 write_state = WR_COPIED;
1434 write_state = WR_NEED_COPY;
1439 itx_wr_state_t wr_state = write_state;
1440 ssize_t len = resid;
1442 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1443 wr_state = WR_NEED_COPY;
1444 else if (wr_state == WR_INDIRECT)
1445 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1447 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1448 (wr_state == WR_COPIED ? len : 0));
1449 lr = (lr_write_t *)&itx->itx_lr;
1450 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1451 off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1452 zil_itx_destroy(itx);
1453 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1454 lr = (lr_write_t *)&itx->itx_lr;
1455 wr_state = WR_NEED_COPY;
1458 itx->itx_wr_state = wr_state;
1459 lr->lr_foid = ZVOL_OBJ;
1460 lr->lr_offset = off;
1461 lr->lr_length = len;
1463 BP_ZERO(&lr->lr_blkptr);
1465 itx->itx_private = zv;
1467 if (!sync && (zv->zv_sync_cnt == 0))
1468 itx->itx_sync = B_FALSE;
1470 zil_itx_assign(zilog, itx, tx);
1479 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1480 uint64_t size, boolean_t doread, boolean_t isdump)
1486 if (vd->vdev_ops == &vdev_mirror_ops ||
1487 vd->vdev_ops == &vdev_replacing_ops ||
1488 vd->vdev_ops == &vdev_spare_ops) {
1489 for (c = 0; c < vd->vdev_children; c++) {
1490 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1491 addr, offset, origoffset, size, doread, isdump);
1494 } else if (doread) {
1500 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1501 return (numerrors < vd->vdev_children ? 0 : EIO);
1503 if (doread && !vdev_readable(vd))
1504 return (SET_ERROR(EIO));
1505 else if (!doread && !vdev_writeable(vd))
1506 return (SET_ERROR(EIO));
1508 if (vd->vdev_ops == &vdev_raidz_ops) {
1509 return (vdev_raidz_physio(vd,
1510 addr, size, offset, origoffset, doread, isdump));
1513 offset += VDEV_LABEL_START_SIZE;
1515 if (ddi_in_panic() || isdump) {
1518 return (SET_ERROR(EIO));
1520 ASSERT3P(dvd, !=, NULL);
1521 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1525 ASSERT3P(dvd, !=, NULL);
1526 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1527 offset, doread ? B_READ : B_WRITE));
1532 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1533 boolean_t doread, boolean_t isdump)
1538 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1540 /* Must be sector aligned, and not stradle a block boundary. */
1541 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1542 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1543 return (SET_ERROR(EINVAL));
1545 ASSERT(size <= zv->zv_volblocksize);
1547 /* Locate the extent this belongs to */
1548 ze = list_head(&zv->zv_extents);
1549 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1550 offset -= ze->ze_nblks * zv->zv_volblocksize;
1551 ze = list_next(&zv->zv_extents, ze);
1555 return (SET_ERROR(EINVAL));
1557 if (!ddi_in_panic())
1558 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1560 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1561 offset += DVA_GET_OFFSET(&ze->ze_dva);
1562 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1563 size, doread, isdump);
1565 if (!ddi_in_panic())
1566 spa_config_exit(spa, SCL_STATE, FTAG);
1572 zvol_strategy(buf_t *bp)
1574 zfs_soft_state_t *zs = NULL;
1575 #else /* !illumos */
1577 zvol_strategy(struct bio *bp)
1579 #endif /* illumos */
1581 uint64_t off, volsize;
1588 boolean_t doread = bp->b_flags & B_READ;
1590 boolean_t doread = 0;
1592 boolean_t is_dumpified;
1596 if (getminor(bp->b_edev) == 0) {
1597 error = SET_ERROR(EINVAL);
1599 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1601 error = SET_ERROR(ENXIO);
1602 else if (zs->zss_type != ZSST_ZVOL)
1603 error = SET_ERROR(EINVAL);
1607 bioerror(bp, error);
1614 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1615 bioerror(bp, EROFS);
1620 off = ldbtob(bp->b_blkno);
1621 #else /* !illumos */
1623 zv = bp->bio_to->private;
1625 zv = bp->bio_dev->si_drv2;
1628 error = SET_ERROR(ENXIO);
1632 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1633 error = SET_ERROR(EROFS);
1637 switch (bp->bio_cmd) {
1650 off = bp->bio_offset;
1651 #endif /* illumos */
1652 volsize = zv->zv_volsize;
1659 addr = bp->b_un.b_addr;
1660 resid = bp->b_bcount;
1662 if (resid > 0 && (off < 0 || off >= volsize)) {
1668 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1669 sync = ((!(bp->b_flags & B_ASYNC) &&
1670 !(zv->zv_flags & ZVOL_WCE)) ||
1671 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1672 !doread && !is_dumpified;
1673 #else /* !illumos */
1674 addr = bp->bio_data;
1675 resid = bp->bio_length;
1677 if (resid > 0 && (off < 0 || off >= volsize)) {
1678 error = SET_ERROR(EIO);
1682 is_dumpified = B_FALSE;
1683 sync = !doread && !is_dumpified &&
1684 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1685 #endif /* illumos */
1688 * There must be no buffer changes when doing a dmu_sync() because
1689 * we can't change the data whilst calculating the checksum.
1691 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1692 doread ? RL_READER : RL_WRITER);
1695 if (bp->bio_cmd == BIO_DELETE) {
1696 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1697 error = dmu_tx_assign(tx, TXG_WAIT);
1701 zvol_log_truncate(zv, tx, off, resid, sync);
1703 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1710 while (resid != 0 && off < volsize) {
1711 size_t size = MIN(resid, zvol_maxphys);
1714 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1715 error = zvol_dumpio(zv, addr, off, size,
1717 } else if (doread) {
1721 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1724 dmu_tx_t *tx = dmu_tx_create(os);
1725 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1726 error = dmu_tx_assign(tx, TXG_WAIT);
1730 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1731 zvol_log_write(zv, tx, off, size, sync);
1736 /* convert checksum errors into IO errors */
1737 if (error == ECKSUM)
1738 error = SET_ERROR(EIO);
1748 zfs_range_unlock(rl);
1751 if ((bp->b_resid = resid) == bp->b_bcount)
1752 bioerror(bp, off > volsize ? EINVAL : error);
1755 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1759 #else /* !illumos */
1760 bp->bio_completed = bp->bio_length - resid;
1761 if (bp->bio_completed < bp->bio_length && off > volsize)
1766 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1770 g_io_deliver(bp, error);
1772 biofinish(bp, NULL, error);
1773 #endif /* illumos */
1778 * Set the buffer count to the zvol maximum transfer.
1779 * Using our own routine instead of the default minphys()
1780 * means that for larger writes we write bigger buffers on X86
1781 * (128K instead of 56K) and flush the disk write cache less often
1782 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1783 * 56K on X86 and 128K on sparc).
1786 zvol_minphys(struct buf *bp)
1788 if (bp->b_bcount > zvol_maxphys)
1789 bp->b_bcount = zvol_maxphys;
1793 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1795 minor_t minor = getminor(dev);
1802 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1804 return (SET_ERROR(ENXIO));
1806 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1807 return (SET_ERROR(EINVAL));
1809 boff = ldbtob(blkno);
1810 resid = ldbtob(nblocks);
1812 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1815 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1816 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1829 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1831 minor_t minor = getminor(dev);
1832 #else /* !illumos */
1834 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1836 #endif /* illumos */
1843 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1845 return (SET_ERROR(ENXIO));
1850 volsize = zv->zv_volsize;
1851 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1852 if (uio->uio_resid > 0 &&
1853 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1854 return (SET_ERROR(EIO));
1857 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1858 error = physio(zvol_strategy, NULL, dev, B_READ,
1864 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1866 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1867 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1869 /* don't read past the end */
1870 if (bytes > volsize - uio->uio_loffset)
1871 bytes = volsize - uio->uio_loffset;
1873 error = dmu_read_uio_dnode(zv->zv_dn, uio, bytes);
1875 /* convert checksum errors into IO errors */
1876 if (error == ECKSUM)
1877 error = SET_ERROR(EIO);
1881 zfs_range_unlock(rl);
1888 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1890 minor_t minor = getminor(dev);
1891 #else /* !illumos */
1893 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1895 #endif /* illumos */
1903 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1905 return (SET_ERROR(ENXIO));
1910 volsize = zv->zv_volsize;
1911 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1912 if (uio->uio_resid > 0 &&
1913 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1914 return (SET_ERROR(EIO));
1917 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1918 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1923 sync = !(zv->zv_flags & ZVOL_WCE) ||
1925 sync = (ioflag & IO_SYNC) ||
1927 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1929 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1931 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1932 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1933 uint64_t off = uio->uio_loffset;
1934 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1936 if (bytes > volsize - off) /* don't write past the end */
1937 bytes = volsize - off;
1939 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1940 error = dmu_tx_assign(tx, TXG_WAIT);
1945 error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1947 zvol_log_write(zv, tx, off, bytes, sync);
1953 zfs_range_unlock(rl);
1955 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1961 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1963 struct uuid uuid = EFI_RESERVED;
1964 efi_gpe_t gpe = { 0 };
1970 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1971 return (SET_ERROR(EFAULT));
1972 ptr = (char *)(uintptr_t)efi.dki_data_64;
1973 length = efi.dki_length;
1975 * Some clients may attempt to request a PMBR for the
1976 * zvol. Currently this interface will return EINVAL to
1977 * such requests. These requests could be supported by
1978 * adding a check for lba == 0 and consing up an appropriate
1981 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1982 return (SET_ERROR(EINVAL));
1984 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1985 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1986 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1988 if (efi.dki_lba == 1) {
1989 efi_gpt_t gpt = { 0 };
1991 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1992 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1993 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1994 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1995 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1996 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1997 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1998 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1999 gpt.efi_gpt_SizeOfPartitionEntry =
2000 LE_32(sizeof (efi_gpe_t));
2001 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
2002 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
2003 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
2004 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
2005 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
2007 return (SET_ERROR(EFAULT));
2008 ptr += sizeof (gpt);
2009 length -= sizeof (gpt);
2011 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
2013 return (SET_ERROR(EFAULT));
2018 * BEGIN entry points to allow external callers access to the volume.
2021 * Return the volume parameters needed for access from an external caller.
2022 * These values are invariant as long as the volume is held open.
2025 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
2026 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
2027 void **rl_hdl, void **dnode_hdl)
2031 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2033 return (SET_ERROR(ENXIO));
2034 if (zv->zv_flags & ZVOL_DUMPIFIED)
2035 return (SET_ERROR(ENXIO));
2037 ASSERT(blksize && max_xfer_len && minor_hdl &&
2038 objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
2040 *blksize = zv->zv_volblocksize;
2041 *max_xfer_len = (uint64_t)zvol_maxphys;
2043 *objset_hdl = zv->zv_objset;
2044 *zil_hdl = zv->zv_zilog;
2045 *rl_hdl = &zv->zv_znode;
2046 *dnode_hdl = zv->zv_dn;
2051 * Return the current volume size to an external caller.
2052 * The size can change while the volume is open.
2055 zvol_get_volume_size(void *minor_hdl)
2057 zvol_state_t *zv = minor_hdl;
2059 return (zv->zv_volsize);
2063 * Return the current WCE setting to an external caller.
2064 * The WCE setting can change while the volume is open.
2067 zvol_get_volume_wce(void *minor_hdl)
2069 zvol_state_t *zv = minor_hdl;
2071 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2075 * Entry point for external callers to zvol_log_write
2078 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2081 zvol_state_t *zv = minor_hdl;
2083 zvol_log_write(zv, tx, off, resid, sync);
2086 * END entry points to allow external callers access to the volume.
2088 #endif /* illumos */
2091 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2094 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2099 zilog_t *zilog = zv->zv_zilog;
2101 if (zil_replaying(zilog, tx))
2104 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2105 lr = (lr_truncate_t *)&itx->itx_lr;
2106 lr->lr_foid = ZVOL_OBJ;
2107 lr->lr_offset = off;
2108 lr->lr_length = len;
2110 itx->itx_sync = (sync || zv->zv_sync_cnt != 0);
2111 zil_itx_assign(zilog, itx, tx);
2116 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2117 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2121 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2124 struct dk_callback *dkc;
2128 mutex_enter(&zfsdev_state_lock);
2130 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2133 mutex_exit(&zfsdev_state_lock);
2134 return (SET_ERROR(ENXIO));
2136 ASSERT(zv->zv_total_opens > 0);
2142 struct dk_cinfo dki;
2144 bzero(&dki, sizeof (dki));
2145 (void) strcpy(dki.dki_cname, "zvol");
2146 (void) strcpy(dki.dki_dname, "zvol");
2147 dki.dki_ctype = DKC_UNKNOWN;
2148 dki.dki_unit = getminor(dev);
2149 dki.dki_maxtransfer =
2150 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2151 mutex_exit(&zfsdev_state_lock);
2152 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2153 error = SET_ERROR(EFAULT);
2157 case DKIOCGMEDIAINFO:
2159 struct dk_minfo dkm;
2161 bzero(&dkm, sizeof (dkm));
2162 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2163 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2164 dkm.dki_media_type = DK_UNKNOWN;
2165 mutex_exit(&zfsdev_state_lock);
2166 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2167 error = SET_ERROR(EFAULT);
2171 case DKIOCGMEDIAINFOEXT:
2173 struct dk_minfo_ext dkmext;
2175 bzero(&dkmext, sizeof (dkmext));
2176 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2177 dkmext.dki_pbsize = zv->zv_volblocksize;
2178 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2179 dkmext.dki_media_type = DK_UNKNOWN;
2180 mutex_exit(&zfsdev_state_lock);
2181 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2182 error = SET_ERROR(EFAULT);
2188 uint64_t vs = zv->zv_volsize;
2189 uint8_t bs = zv->zv_min_bs;
2191 mutex_exit(&zfsdev_state_lock);
2192 error = zvol_getefi((void *)arg, flag, vs, bs);
2196 case DKIOCFLUSHWRITECACHE:
2197 dkc = (struct dk_callback *)arg;
2198 mutex_exit(&zfsdev_state_lock);
2199 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2200 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2201 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2208 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2209 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2211 error = SET_ERROR(EFAULT);
2217 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2219 error = SET_ERROR(EFAULT);
2223 zv->zv_flags |= ZVOL_WCE;
2224 mutex_exit(&zfsdev_state_lock);
2226 zv->zv_flags &= ~ZVOL_WCE;
2227 mutex_exit(&zfsdev_state_lock);
2228 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2236 * commands using these (like prtvtoc) expect ENOTSUP
2237 * since we're emulating an EFI label
2239 error = SET_ERROR(ENOTSUP);
2243 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2245 error = zvol_dumpify(zv);
2246 zfs_range_unlock(rl);
2250 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2252 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2254 error = zvol_dump_fini(zv);
2255 zfs_range_unlock(rl);
2260 dkioc_free_list_t *dfl;
2263 if (!zvol_unmap_enabled)
2266 if (!(flag & FKIOCTL)) {
2267 error = dfl_copyin((void *)arg, &dfl, flag, KM_SLEEP);
2271 dfl = (dkioc_free_list_t *)arg;
2272 ASSERT3U(dfl->dfl_num_exts, <=, DFL_COPYIN_MAX_EXTS);
2273 if (dfl->dfl_num_exts > DFL_COPYIN_MAX_EXTS) {
2274 error = SET_ERROR(EINVAL);
2279 mutex_exit(&zfsdev_state_lock);
2281 for (int i = 0; i < dfl->dfl_num_exts; i++) {
2282 uint64_t start = dfl->dfl_exts[i].dfle_start,
2283 length = dfl->dfl_exts[i].dfle_length,
2284 end = start + length;
2287 * Apply Postel's Law to length-checking. If they
2288 * overshoot, just blank out until the end, if there's
2289 * a need to blank out anything.
2291 if (start >= zv->zv_volsize)
2292 continue; /* No need to do anything... */
2293 if (end > zv->zv_volsize) {
2294 end = DMU_OBJECT_END;
2295 length = end - start;
2298 rl = zfs_range_lock(&zv->zv_znode, start, length,
2300 tx = dmu_tx_create(zv->zv_objset);
2301 error = dmu_tx_assign(tx, TXG_WAIT);
2305 zvol_log_truncate(zv, tx, start, length,
2308 error = dmu_free_long_range(zv->zv_objset,
2309 ZVOL_OBJ, start, length);
2312 zfs_range_unlock(rl);
2319 * If the write-cache is disabled, 'sync' property
2320 * is set to 'always', or if the caller is asking for
2321 * a synchronous free, commit this operation to the zil.
2322 * This will sync any previous uncommitted writes to the
2324 * Can be overridden by the zvol_unmap_sync_enabled tunable.
2326 if ((error == 0) && zvol_unmap_sync_enabled &&
2327 (!(zv->zv_flags & ZVOL_WCE) ||
2328 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
2329 (dfl->dfl_flags & DF_WAIT_SYNC))) {
2330 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2333 if (!(flag & FKIOCTL))
2340 error = SET_ERROR(ENOTTY);
2344 mutex_exit(&zfsdev_state_lock);
2347 #endif /* illumos */
2352 return (zvol_minors != 0);
2358 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2361 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2363 ZFS_LOG(1, "ZVOL Initialized.");
2371 mutex_destroy(&zfsdev_state_lock);
2373 ddi_soft_state_fini(&zfsdev_state);
2374 ZFS_LOG(1, "ZVOL Deinitialized.");
2380 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2382 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2384 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2391 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2393 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2395 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2399 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2403 objset_t *os = zv->zv_objset;
2404 spa_t *spa = dmu_objset_spa(os);
2405 vdev_t *vd = spa->spa_root_vdev;
2406 nvlist_t *nv = NULL;
2407 uint64_t version = spa_version(spa);
2408 uint64_t checksum, compress, refresrv, vbs, dedup;
2410 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2411 ASSERT(vd->vdev_ops == &vdev_root_ops);
2413 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2417 /* wait for dmu_free_long_range to actually free the blocks */
2418 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2421 * If the pool on which the dump device is being initialized has more
2422 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2423 * enabled. If so, bump that feature's counter to indicate that the
2424 * feature is active. We also check the vdev type to handle the
2426 * # zpool create test raidz disk1 disk2 disk3
2427 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2428 * the raidz vdev itself has 3 children.
2430 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2431 if (!spa_feature_is_enabled(spa,
2432 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2433 return (SET_ERROR(ENOTSUP));
2434 (void) dsl_sync_task(spa_name(spa),
2435 zfs_mvdev_dump_feature_check,
2436 zfs_mvdev_dump_activate_feature_sync, NULL,
2437 2, ZFS_SPACE_CHECK_RESERVED);
2441 error = dsl_prop_get_integer(zv->zv_name,
2442 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2444 error = dsl_prop_get_integer(zv->zv_name,
2445 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2449 error = dsl_prop_get_integer(zv->zv_name,
2450 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2454 error = dsl_prop_get_integer(zv->zv_name,
2455 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2458 if (version >= SPA_VERSION_DEDUP && error == 0) {
2459 error = dsl_prop_get_integer(zv->zv_name,
2460 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2466 tx = dmu_tx_create(os);
2467 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2468 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2469 error = dmu_tx_assign(tx, TXG_WAIT);
2476 * If we are resizing the dump device then we only need to
2477 * update the refreservation to match the newly updated
2478 * zvolsize. Otherwise, we save off the original state of the
2479 * zvol so that we can restore them if the zvol is ever undumpified.
2482 error = zap_update(os, ZVOL_ZAP_OBJ,
2483 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2484 &zv->zv_volsize, tx);
2486 error = zap_update(os, ZVOL_ZAP_OBJ,
2487 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2490 error = zap_update(os, ZVOL_ZAP_OBJ,
2491 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2495 error = zap_update(os, ZVOL_ZAP_OBJ,
2496 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2500 error = zap_update(os, ZVOL_ZAP_OBJ,
2501 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2505 error = dmu_object_set_blocksize(
2506 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2508 if (version >= SPA_VERSION_DEDUP && error == 0) {
2509 error = zap_update(os, ZVOL_ZAP_OBJ,
2510 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2514 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2519 * We only need update the zvol's property if we are initializing
2520 * the dump area for the first time.
2522 if (error == 0 && !resize) {
2524 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2525 * function. Otherwise, use the old default -- OFF.
2527 checksum = spa_feature_is_active(spa,
2528 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2531 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2532 VERIFY(nvlist_add_uint64(nv,
2533 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2534 VERIFY(nvlist_add_uint64(nv,
2535 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2536 ZIO_COMPRESS_OFF) == 0);
2537 VERIFY(nvlist_add_uint64(nv,
2538 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2540 if (version >= SPA_VERSION_DEDUP) {
2541 VERIFY(nvlist_add_uint64(nv,
2542 zfs_prop_to_name(ZFS_PROP_DEDUP),
2543 ZIO_CHECKSUM_OFF) == 0);
2546 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2551 /* Allocate the space for the dump */
2553 error = zvol_prealloc(zv);
2558 zvol_dumpify(zvol_state_t *zv)
2561 uint64_t dumpsize = 0;
2563 objset_t *os = zv->zv_objset;
2565 if (zv->zv_flags & ZVOL_RDONLY)
2566 return (SET_ERROR(EROFS));
2568 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2569 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2570 boolean_t resize = (dumpsize > 0);
2572 if ((error = zvol_dump_init(zv, resize)) != 0) {
2573 (void) zvol_dump_fini(zv);
2579 * Build up our lba mapping.
2581 error = zvol_get_lbas(zv);
2583 (void) zvol_dump_fini(zv);
2587 tx = dmu_tx_create(os);
2588 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2589 error = dmu_tx_assign(tx, TXG_WAIT);
2592 (void) zvol_dump_fini(zv);
2596 zv->zv_flags |= ZVOL_DUMPIFIED;
2597 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2598 &zv->zv_volsize, tx);
2602 (void) zvol_dump_fini(zv);
2606 txg_wait_synced(dmu_objset_pool(os), 0);
2611 zvol_dump_fini(zvol_state_t *zv)
2614 objset_t *os = zv->zv_objset;
2617 uint64_t checksum, compress, refresrv, vbs, dedup;
2618 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2621 * Attempt to restore the zvol back to its pre-dumpified state.
2622 * This is a best-effort attempt as it's possible that not all
2623 * of these properties were initialized during the dumpify process
2624 * (i.e. error during zvol_dump_init).
2627 tx = dmu_tx_create(os);
2628 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2629 error = dmu_tx_assign(tx, TXG_WAIT);
2634 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2637 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2638 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2639 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2640 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2641 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2642 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2643 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2644 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2646 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2647 (void) nvlist_add_uint64(nv,
2648 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2649 (void) nvlist_add_uint64(nv,
2650 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2651 (void) nvlist_add_uint64(nv,
2652 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2653 if (version >= SPA_VERSION_DEDUP &&
2654 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2655 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2656 (void) nvlist_add_uint64(nv,
2657 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2659 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2663 zvol_free_extents(zv);
2664 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2665 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2666 /* wait for dmu_free_long_range to actually free the blocks */
2667 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2668 tx = dmu_tx_create(os);
2669 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2670 error = dmu_tx_assign(tx, TXG_WAIT);
2675 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2676 zv->zv_volblocksize = vbs;
2681 #else /* !illumos */
2684 zvol_geom_run(zvol_state_t *zv)
2686 struct g_provider *pp;
2688 pp = zv->zv_provider;
2689 g_error_provider(pp, 0);
2691 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2692 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2696 zvol_geom_destroy(zvol_state_t *zv)
2698 struct g_provider *pp;
2700 g_topology_assert();
2702 mtx_lock(&zv->zv_queue_mtx);
2704 wakeup_one(&zv->zv_queue);
2705 while (zv->zv_state != 2)
2706 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2707 mtx_destroy(&zv->zv_queue_mtx);
2709 pp = zv->zv_provider;
2710 zv->zv_provider = NULL;
2712 g_wither_geom(pp->geom, ENXIO);
2716 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2718 int count, error, flags;
2720 g_topology_assert();
2723 * To make it easier we expect either open or close, but not both
2726 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2727 (acr <= 0 && acw <= 0 && ace <= 0),
2728 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2729 pp->name, acr, acw, ace));
2731 if (pp->private == NULL) {
2732 if (acr <= 0 && acw <= 0 && ace <= 0)
2738 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2739 * because GEOM already handles that and handles it a bit differently.
2740 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2741 * only one exclusive consumer, no matter if it is reader or writer.
2742 * I like better the way GEOM works so I'll leave it for GEOM to
2743 * decide what to do.
2746 count = acr + acw + ace;
2751 if (acr != 0 || ace != 0)
2756 g_topology_unlock();
2758 error = zvol_open(pp, flags, count);
2760 error = zvol_close(pp, flags, -count);
2766 zvol_geom_start(struct bio *bp)
2771 zv = bp->bio_to->private;
2773 switch (bp->bio_cmd) {
2775 if (!THREAD_CAN_SLEEP())
2777 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2778 g_io_deliver(bp, 0);
2783 if (!THREAD_CAN_SLEEP())
2788 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2789 uint64_t refd, avail, usedobjs, availobjs, val;
2791 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2793 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2794 dmu_objset_space(zv->zv_objset, &refd, &avail,
2795 &usedobjs, &availobjs);
2796 if (g_handleattr_off_t(bp, "blocksavail",
2799 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2800 dmu_objset_space(zv->zv_objset, &refd, &avail,
2801 &usedobjs, &availobjs);
2802 if (g_handleattr_off_t(bp, "blocksused",
2805 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2806 avail = metaslab_class_get_space(spa_normal_class(spa));
2807 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2808 if (g_handleattr_off_t(bp, "poolblocksavail",
2811 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2812 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2813 if (g_handleattr_off_t(bp, "poolblocksused",
2820 g_io_deliver(bp, EOPNOTSUPP);
2826 mtx_lock(&zv->zv_queue_mtx);
2827 first = (bioq_first(&zv->zv_queue) == NULL);
2828 bioq_insert_tail(&zv->zv_queue, bp);
2829 mtx_unlock(&zv->zv_queue_mtx);
2831 wakeup_one(&zv->zv_queue);
2835 zvol_geom_worker(void *arg)
2840 thread_lock(curthread);
2841 sched_prio(curthread, PRIBIO);
2842 thread_unlock(curthread);
2846 mtx_lock(&zv->zv_queue_mtx);
2847 bp = bioq_takefirst(&zv->zv_queue);
2849 if (zv->zv_state == 1) {
2851 wakeup(&zv->zv_state);
2852 mtx_unlock(&zv->zv_queue_mtx);
2855 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2859 mtx_unlock(&zv->zv_queue_mtx);
2860 switch (bp->bio_cmd) {
2862 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2863 g_io_deliver(bp, 0);
2871 g_io_deliver(bp, EOPNOTSUPP);
2877 extern boolean_t dataset_name_hidden(const char *name);
2880 zvol_create_snapshots(objset_t *os, const char *name)
2882 uint64_t cookie, obj;
2887 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2890 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2895 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2896 if (len >= MAXPATHLEN) {
2897 dmu_objset_rele(os, FTAG);
2898 error = ENAMETOOLONG;
2902 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2903 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2904 sname + len, &obj, &cookie, NULL);
2905 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2907 if (error == ENOENT)
2912 error = zvol_create_minor(sname);
2913 if (error != 0 && error != EEXIST) {
2914 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2920 kmem_free(sname, MAXPATHLEN);
2925 zvol_create_minors(const char *name)
2932 if (dataset_name_hidden(name))
2935 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2936 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2940 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2941 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2942 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2943 error = zvol_create_minor(name);
2944 if (error == 0 || error == EEXIST) {
2945 error = zvol_create_snapshots(os, name);
2947 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2950 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2951 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2954 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2955 dmu_objset_rele(os, FTAG);
2959 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2960 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2961 dmu_objset_rele(os, FTAG);
2962 kmem_free(osname, MAXPATHLEN);
2965 p = osname + strlen(osname);
2966 len = MAXPATHLEN - (p - osname);
2969 /* Prefetch the datasets. */
2971 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2972 if (!dataset_name_hidden(osname))
2973 (void) dmu_objset_prefetch(osname, NULL);
2978 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2980 dmu_objset_rele(os, FTAG);
2981 (void)zvol_create_minors(osname);
2982 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2983 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2989 dmu_objset_rele(os, FTAG);
2990 kmem_free(osname, MAXPATHLEN);
2995 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2998 struct g_provider *pp;
3001 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
3003 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
3005 pp = zv->zv_provider;
3010 zv->zv_provider = NULL;
3011 g_wither_provider(pp, ENXIO);
3013 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
3014 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
3015 pp->sectorsize = DEV_BSIZE;
3016 pp->mediasize = zv->zv_volsize;
3018 zv->zv_provider = pp;
3019 g_error_provider(pp, 0);
3020 g_topology_unlock();
3021 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
3022 struct make_dev_args args;
3024 if ((dev = zv->zv_dev) != NULL) {
3027 if (zv->zv_total_opens > 0) {
3028 zv->zv_flags &= ~ZVOL_EXCL;
3029 zv->zv_total_opens = 0;
3030 zvol_last_close(zv);
3034 make_dev_args_init(&args);
3035 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
3036 args.mda_devsw = &zvol_cdevsw;
3038 args.mda_uid = UID_ROOT;
3039 args.mda_gid = GID_OPERATOR;
3040 args.mda_mode = 0640;
3041 args.mda_si_drv2 = zv;
3042 if (make_dev_s(&args, &zv->zv_dev,
3043 "%s/%s", ZVOL_DRIVER, newname) == 0)
3044 zv->zv_dev->si_iosize_max = MAXPHYS;
3046 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3050 zvol_rename_minors(const char *oldname, const char *newname)
3052 char name[MAXPATHLEN];
3053 struct g_provider *pp;
3055 size_t oldnamelen, newnamelen;
3058 boolean_t locked = B_FALSE;
3060 oldnamelen = strlen(oldname);
3061 newnamelen = strlen(newname);
3064 /* See comment in zvol_open(). */
3065 if (!MUTEX_HELD(&zfsdev_state_lock)) {
3066 mutex_enter(&zfsdev_state_lock);
3070 LIST_FOREACH(zv, &all_zvols, zv_links) {
3071 if (strcmp(zv->zv_name, oldname) == 0) {
3072 zvol_rename_minor(zv, newname);
3073 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3074 (zv->zv_name[oldnamelen] == '/' ||
3075 zv->zv_name[oldnamelen] == '@')) {
3076 snprintf(name, sizeof(name), "%s%c%s", newname,
3077 zv->zv_name[oldnamelen],
3078 zv->zv_name + oldnamelen + 1);
3079 zvol_rename_minor(zv, name);
3084 mutex_exit(&zfsdev_state_lock);
3089 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3091 zvol_state_t *zv = dev->si_drv2;
3094 mutex_enter(&zfsdev_state_lock);
3095 if (zv->zv_total_opens == 0)
3096 err = zvol_first_open(zv);
3098 mutex_exit(&zfsdev_state_lock);
3101 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3102 err = SET_ERROR(EROFS);
3105 if (zv->zv_flags & ZVOL_EXCL) {
3106 err = SET_ERROR(EBUSY);
3110 if (flags & FEXCL) {
3111 if (zv->zv_total_opens != 0) {
3112 err = SET_ERROR(EBUSY);
3115 zv->zv_flags |= ZVOL_EXCL;
3119 zv->zv_total_opens++;
3120 if (flags & (FSYNC | FDSYNC)) {
3122 if (zv->zv_sync_cnt == 1)
3123 zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
3125 mutex_exit(&zfsdev_state_lock);
3128 if (zv->zv_total_opens == 0)
3129 zvol_last_close(zv);
3130 mutex_exit(&zfsdev_state_lock);
3135 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3137 zvol_state_t *zv = dev->si_drv2;
3139 mutex_enter(&zfsdev_state_lock);
3140 if (zv->zv_flags & ZVOL_EXCL) {
3141 ASSERT(zv->zv_total_opens == 1);
3142 zv->zv_flags &= ~ZVOL_EXCL;
3146 * If the open count is zero, this is a spurious close.
3147 * That indicates a bug in the kernel / DDI framework.
3149 ASSERT(zv->zv_total_opens != 0);
3152 * You may get multiple opens, but only one close.
3154 zv->zv_total_opens--;
3155 if (flags & (FSYNC | FDSYNC))
3158 if (zv->zv_total_opens == 0)
3159 zvol_last_close(zv);
3161 mutex_exit(&zfsdev_state_lock);
3166 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3170 off_t offset, length;
3177 KASSERT(zv->zv_total_opens > 0,
3178 ("Device with zero access count in zvol_d_ioctl"));
3180 i = IOCPARM_LEN(cmd);
3182 case DIOCGSECTORSIZE:
3183 *(u_int *)data = DEV_BSIZE;
3185 case DIOCGMEDIASIZE:
3186 *(off_t *)data = zv->zv_volsize;
3189 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3192 if (!zvol_unmap_enabled)
3195 offset = ((off_t *)data)[0];
3196 length = ((off_t *)data)[1];
3197 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3198 offset < 0 || offset >= zv->zv_volsize ||
3200 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3206 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3207 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3208 error = dmu_tx_assign(tx, TXG_WAIT);
3213 sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
3214 zvol_log_truncate(zv, tx, offset, length, sync);
3216 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3219 zfs_range_unlock(rl);
3221 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3223 case DIOCGSTRIPESIZE:
3224 *(off_t *)data = zv->zv_volblocksize;
3226 case DIOCGSTRIPEOFFSET:
3230 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3231 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3232 uint64_t refd, avail, usedobjs, availobjs;
3234 if (strcmp(arg->name, "GEOM::candelete") == 0)
3236 else if (strcmp(arg->name, "blocksavail") == 0) {
3237 dmu_objset_space(zv->zv_objset, &refd, &avail,
3238 &usedobjs, &availobjs);
3239 arg->value.off = avail / DEV_BSIZE;
3240 } else if (strcmp(arg->name, "blocksused") == 0) {
3241 dmu_objset_space(zv->zv_objset, &refd, &avail,
3242 &usedobjs, &availobjs);
3243 arg->value.off = refd / DEV_BSIZE;
3244 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3245 avail = metaslab_class_get_space(spa_normal_class(spa));
3246 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3247 arg->value.off = avail / DEV_BSIZE;
3248 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3249 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3250 arg->value.off = refd / DEV_BSIZE;
3257 off_t *off = (off_t *)data;
3261 hole = (cmd == FIOSEEKHOLE);
3263 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3273 #endif /* illumos */