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 2011 Nexenta Systems, Inc. All rights reserved.
30 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
31 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32 * Copyright (c) 2014 Integros [integros.com]
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/filio.h>
97 #include <geom/geom.h>
99 #include "zfs_namecheck.h"
102 struct g_class zfs_zvol_class = {
104 .version = G_VERSION,
107 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
111 static char *zvol_tag = "zvol_tag";
113 #define ZVOL_DUMPSIZE "dumpsize"
116 * This lock protects the zfsdev_state structure from being modified
117 * while it's being used, e.g. an open that comes in before a create
118 * finishes. It also protects temporary opens of the dataset so that,
119 * e.g., an open doesn't get a spurious EBUSY.
122 kmutex_t zfsdev_state_lock;
125 * In FreeBSD we've replaced the upstream zfsdev_state_lock with the
126 * spa_namespace_lock in the ZVOL code.
128 #define zfsdev_state_lock spa_namespace_lock
130 static uint32_t zvol_minors;
133 SYSCTL_DECL(_vfs_zfs);
134 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
135 static int volmode = ZFS_VOLMODE_GEOM;
136 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
137 "Expose as GEOM providers (1), device files (2) or neither");
138 static boolean_t zpool_on_zvol = B_FALSE;
139 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, recursive, CTLFLAG_RWTUN, &zpool_on_zvol, 0,
140 "Allow zpools to use zvols as vdevs (DANGEROUS)");
143 typedef struct zvol_extent {
145 dva_t ze_dva; /* dva associated with this extent */
146 uint64_t ze_nblks; /* number of blocks in extent */
150 * The in-core state of each volume.
152 typedef struct zvol_state {
154 LIST_ENTRY(zvol_state) zv_links;
156 char zv_name[MAXPATHLEN]; /* pool/dd name */
157 uint64_t zv_volsize; /* amount of space we advertise */
158 uint64_t zv_volblocksize; /* volume block size */
160 minor_t zv_minor; /* minor number */
162 struct cdev *zv_dev; /* non-GEOM device */
163 struct g_provider *zv_provider; /* GEOM provider */
165 uint8_t zv_min_bs; /* minimum addressable block shift */
166 uint8_t zv_flags; /* readonly, dumpified, etc. */
167 objset_t *zv_objset; /* objset handle */
169 uint32_t zv_open_count[OTYPCNT]; /* open counts */
171 uint32_t zv_total_opens; /* total open count */
172 uint32_t zv_sync_cnt; /* synchronous open count */
173 zilog_t *zv_zilog; /* ZIL handle */
174 list_t zv_extents; /* List of extents for dump */
175 znode_t zv_znode; /* for range locking */
176 dmu_buf_t *zv_dbuf; /* bonus handle */
179 int zv_volmode; /* Provide GEOM or cdev */
180 struct bio_queue_head zv_queue;
181 struct mtx zv_queue_mtx; /* zv_queue mutex */
186 static LIST_HEAD(, zvol_state) all_zvols;
189 * zvol specific flags
191 #define ZVOL_RDONLY 0x1
192 #define ZVOL_DUMPIFIED 0x2
193 #define ZVOL_EXCL 0x4
197 * zvol maximum transfer in one DMU tx.
199 int zvol_maxphys = DMU_MAX_ACCESS/2;
202 * Toggle unmap functionality.
204 boolean_t zvol_unmap_enabled = B_TRUE;
207 * If true, unmaps requested as synchronous are executed synchronously,
208 * otherwise all unmaps are asynchronous.
210 boolean_t zvol_unmap_sync_enabled = B_FALSE;
213 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
214 &zvol_unmap_enabled, 0,
215 "Enable UNMAP functionality");
217 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_sync_enabled, CTLFLAG_RWTUN,
218 &zvol_unmap_sync_enabled, 0,
219 "UNMAPs requested as sync are executed synchronously");
221 static d_open_t zvol_d_open;
222 static d_close_t zvol_d_close;
223 static d_read_t zvol_read;
224 static d_write_t zvol_write;
225 static d_ioctl_t zvol_d_ioctl;
226 static d_strategy_t zvol_strategy;
228 static struct cdevsw zvol_cdevsw = {
229 .d_version = D_VERSION,
230 .d_open = zvol_d_open,
231 .d_close = zvol_d_close,
233 .d_write = zvol_write,
234 .d_ioctl = zvol_d_ioctl,
235 .d_strategy = zvol_strategy,
237 .d_flags = D_DISK | D_TRACKCLOSE,
240 static void zvol_geom_run(zvol_state_t *zv);
241 static void zvol_geom_destroy(zvol_state_t *zv);
242 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
243 static void zvol_geom_start(struct bio *bp);
244 static void zvol_geom_worker(void *arg);
245 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
246 uint64_t len, boolean_t sync);
247 #endif /* !illumos */
249 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
250 nvlist_t *, nvlist_t *);
251 static int zvol_remove_zv(zvol_state_t *);
252 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
253 static int zvol_dumpify(zvol_state_t *zv);
254 static int zvol_dump_fini(zvol_state_t *zv);
255 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
258 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
261 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
263 zv->zv_volsize = volsize;
264 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
265 "Size", volsize) == DDI_SUCCESS);
266 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
267 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
269 /* Notify specfs to invalidate the cached size */
270 spec_size_invalidate(dev, VBLK);
271 spec_size_invalidate(dev, VCHR);
273 zv->zv_volsize = volsize;
274 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
275 struct g_provider *pp;
277 pp = zv->zv_provider;
283 * Do not invoke resize event when initial size was zero.
284 * ZVOL initializes the size on first open, this is not
287 if (pp->mediasize == 0)
288 pp->mediasize = zv->zv_volsize;
290 g_resize_provider(pp, zv->zv_volsize);
297 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
300 return (SET_ERROR(EINVAL));
302 if (volsize % blocksize != 0)
303 return (SET_ERROR(EINVAL));
306 if (volsize - 1 > SPEC_MAXOFFSET_T)
307 return (SET_ERROR(EOVERFLOW));
313 zvol_check_volblocksize(uint64_t volblocksize)
315 if (volblocksize < SPA_MINBLOCKSIZE ||
316 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
318 return (SET_ERROR(EDOM));
324 zvol_get_stats(objset_t *os, nvlist_t *nv)
327 dmu_object_info_t doi;
330 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
334 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
336 error = dmu_object_info(os, ZVOL_OBJ, &doi);
339 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
340 doi.doi_data_block_size);
346 static zvol_state_t *
347 zvol_minor_lookup(const char *name)
354 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
357 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
358 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
362 LIST_FOREACH(zv, &all_zvols, zv_links) {
364 if (strcmp(zv->zv_name, name) == 0)
371 /* extent mapping arg */
379 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
380 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
382 struct maparg *ma = arg;
384 int bs = ma->ma_zv->zv_volblocksize;
386 if (bp == NULL || BP_IS_HOLE(bp) ||
387 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
390 VERIFY(!BP_IS_EMBEDDED(bp));
392 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
395 /* Abort immediately if we have encountered gang blocks */
397 return (SET_ERROR(EFRAGS));
400 * See if the block is at the end of the previous extent.
402 ze = list_tail(&ma->ma_zv->zv_extents);
404 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
405 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
406 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
411 dprintf_bp(bp, "%s", "next blkptr:");
413 /* start a new extent */
414 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
415 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
417 list_insert_tail(&ma->ma_zv->zv_extents, ze);
422 zvol_free_extents(zvol_state_t *zv)
426 while (ze = list_head(&zv->zv_extents)) {
427 list_remove(&zv->zv_extents, ze);
428 kmem_free(ze, sizeof (zvol_extent_t));
433 zvol_get_lbas(zvol_state_t *zv)
435 objset_t *os = zv->zv_objset;
441 zvol_free_extents(zv);
443 /* commit any in-flight changes before traversing the dataset */
444 txg_wait_synced(dmu_objset_pool(os), 0);
445 err = traverse_dataset(dmu_objset_ds(os), 0,
446 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
447 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
448 zvol_free_extents(zv);
449 return (err ? err : EIO);
457 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
459 zfs_creat_t *zct = arg;
460 nvlist_t *nvprops = zct->zct_props;
462 uint64_t volblocksize, volsize;
464 VERIFY(nvlist_lookup_uint64(nvprops,
465 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
466 if (nvlist_lookup_uint64(nvprops,
467 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
468 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
471 * These properties must be removed from the list so the generic
472 * property setting step won't apply to them.
474 VERIFY(nvlist_remove_all(nvprops,
475 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
476 (void) nvlist_remove_all(nvprops,
477 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
479 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
483 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
487 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
492 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
493 * implement DKIOCFREE/free-long-range.
496 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
498 uint64_t offset, length;
501 byteswap_uint64_array(lr, sizeof (*lr));
503 offset = lr->lr_offset;
504 length = lr->lr_length;
506 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
510 * Replay a TX_WRITE ZIL transaction that didn't get committed
511 * after a system failure
514 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
516 objset_t *os = zv->zv_objset;
517 char *data = (char *)(lr + 1); /* data follows lr_write_t */
518 uint64_t offset, length;
523 byteswap_uint64_array(lr, sizeof (*lr));
525 offset = lr->lr_offset;
526 length = lr->lr_length;
528 /* If it's a dmu_sync() block, write the whole block */
529 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
530 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
531 if (length < blocksize) {
532 offset -= offset % blocksize;
537 tx = dmu_tx_create(os);
538 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
539 error = dmu_tx_assign(tx, TXG_WAIT);
543 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
552 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
554 return (SET_ERROR(ENOTSUP));
558 * Callback vectors for replaying records.
559 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
561 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
562 zvol_replay_err, /* 0 no such transaction type */
563 zvol_replay_err, /* TX_CREATE */
564 zvol_replay_err, /* TX_MKDIR */
565 zvol_replay_err, /* TX_MKXATTR */
566 zvol_replay_err, /* TX_SYMLINK */
567 zvol_replay_err, /* TX_REMOVE */
568 zvol_replay_err, /* TX_RMDIR */
569 zvol_replay_err, /* TX_LINK */
570 zvol_replay_err, /* TX_RENAME */
571 zvol_replay_write, /* TX_WRITE */
572 zvol_replay_truncate, /* TX_TRUNCATE */
573 zvol_replay_err, /* TX_SETATTR */
574 zvol_replay_err, /* TX_ACL */
575 zvol_replay_err, /* TX_CREATE_ACL */
576 zvol_replay_err, /* TX_CREATE_ATTR */
577 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
578 zvol_replay_err, /* TX_MKDIR_ACL */
579 zvol_replay_err, /* TX_MKDIR_ATTR */
580 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
581 zvol_replay_err, /* TX_WRITE2 */
586 zvol_name2minor(const char *name, minor_t *minor)
590 mutex_enter(&zfsdev_state_lock);
591 zv = zvol_minor_lookup(name);
593 *minor = zv->zv_minor;
594 mutex_exit(&zfsdev_state_lock);
595 return (zv ? 0 : -1);
600 * Create a minor node (plus a whole lot more) for the specified volume.
603 zvol_create_minor(const char *name)
605 zfs_soft_state_t *zs;
609 dmu_object_info_t doi;
611 char chrbuf[30], blkbuf[30];
613 struct g_provider *pp;
620 ZFS_LOG(1, "Creating ZVOL %s...", name);
623 mutex_enter(&zfsdev_state_lock);
625 if (zvol_minor_lookup(name) != NULL) {
626 mutex_exit(&zfsdev_state_lock);
627 return (SET_ERROR(EEXIST));
630 /* lie and say we're read-only */
631 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
634 mutex_exit(&zfsdev_state_lock);
639 if ((minor = zfsdev_minor_alloc()) == 0) {
640 dmu_objset_disown(os, FTAG);
641 mutex_exit(&zfsdev_state_lock);
642 return (SET_ERROR(ENXIO));
645 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
646 dmu_objset_disown(os, FTAG);
647 mutex_exit(&zfsdev_state_lock);
648 return (SET_ERROR(EAGAIN));
650 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
653 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
655 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
656 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
657 ddi_soft_state_free(zfsdev_state, minor);
658 dmu_objset_disown(os, FTAG);
659 mutex_exit(&zfsdev_state_lock);
660 return (SET_ERROR(EAGAIN));
663 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
665 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
666 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
667 ddi_remove_minor_node(zfs_dip, chrbuf);
668 ddi_soft_state_free(zfsdev_state, minor);
669 dmu_objset_disown(os, FTAG);
670 mutex_exit(&zfsdev_state_lock);
671 return (SET_ERROR(EAGAIN));
674 zs = ddi_get_soft_state(zfsdev_state, minor);
675 zs->zss_type = ZSST_ZVOL;
676 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
679 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
681 error = dsl_prop_get_integer(name,
682 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
683 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
687 zv->zv_volmode = mode;
688 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
690 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
691 gp->start = zvol_geom_start;
692 gp->access = zvol_geom_access;
693 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
694 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
695 pp->sectorsize = DEV_BSIZE;
699 zv->zv_provider = pp;
700 bioq_init(&zv->zv_queue);
701 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
702 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
703 struct make_dev_args args;
705 make_dev_args_init(&args);
706 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
707 args.mda_devsw = &zvol_cdevsw;
709 args.mda_uid = UID_ROOT;
710 args.mda_gid = GID_OPERATOR;
711 args.mda_mode = 0640;
712 args.mda_si_drv2 = zv;
713 error = make_dev_s(&args, &zv->zv_dev,
714 "%s/%s", ZVOL_DRIVER, name);
716 kmem_free(zv, sizeof(*zv));
717 dmu_objset_disown(os, FTAG);
718 mutex_exit(&zfsdev_state_lock);
721 zv->zv_dev->si_iosize_max = MAXPHYS;
723 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
726 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
727 zv->zv_min_bs = DEV_BSHIFT;
729 zv->zv_minor = minor;
732 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
733 zv->zv_flags |= ZVOL_RDONLY;
734 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
735 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
736 sizeof (rl_t), offsetof(rl_t, r_node));
737 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
738 offsetof(zvol_extent_t, ze_node));
740 /* get and cache the blocksize */
741 error = dmu_object_info(os, ZVOL_OBJ, &doi);
743 zv->zv_volblocksize = doi.doi_data_block_size;
746 if (spa_writeable(dmu_objset_spa(os))) {
747 if (zil_replay_disable)
748 zil_destroy(dmu_objset_zil(os), B_FALSE);
750 zil_replay(os, zv, zvol_replay_vector);
752 dmu_objset_disown(os, FTAG);
753 zv->zv_objset = NULL;
757 mutex_exit(&zfsdev_state_lock);
759 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
765 ZFS_LOG(1, "ZVOL %s created.", name);
772 * Remove minor node for the specified volume.
775 zvol_remove_zv(zvol_state_t *zv)
779 minor_t minor = zv->zv_minor;
782 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
783 if (zv->zv_total_opens != 0)
784 return (SET_ERROR(EBUSY));
787 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
788 ddi_remove_minor_node(zfs_dip, nmbuf);
790 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
791 ddi_remove_minor_node(zfs_dip, nmbuf);
793 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
795 LIST_REMOVE(zv, zv_links);
796 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
798 zvol_geom_destroy(zv);
800 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
801 if (zv->zv_dev != NULL)
802 destroy_dev(zv->zv_dev);
806 avl_destroy(&zv->zv_znode.z_range_avl);
807 mutex_destroy(&zv->zv_znode.z_range_lock);
809 kmem_free(zv, sizeof (zvol_state_t));
811 ddi_soft_state_free(zfsdev_state, minor);
818 zvol_remove_minor(const char *name)
823 mutex_enter(&zfsdev_state_lock);
824 if ((zv = zvol_minor_lookup(name)) == NULL) {
825 mutex_exit(&zfsdev_state_lock);
826 return (SET_ERROR(ENXIO));
828 rc = zvol_remove_zv(zv);
829 mutex_exit(&zfsdev_state_lock);
834 zvol_first_open(zvol_state_t *zv)
836 dmu_object_info_t doi;
842 /* lie and say we're read-only */
843 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
849 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
852 dmu_objset_disown(os, zvol_tag);
856 /* get and cache the blocksize */
857 error = dmu_object_info(os, ZVOL_OBJ, &doi);
860 dmu_objset_disown(os, zvol_tag);
863 zv->zv_volblocksize = doi.doi_data_block_size;
865 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
867 dmu_objset_disown(os, zvol_tag);
871 zvol_size_changed(zv, volsize);
872 zv->zv_zilog = zil_open(os, zvol_get_data);
874 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
876 if (readonly || dmu_objset_is_snapshot(os) ||
877 !spa_writeable(dmu_objset_spa(os)))
878 zv->zv_flags |= ZVOL_RDONLY;
880 zv->zv_flags &= ~ZVOL_RDONLY;
885 zvol_last_close(zvol_state_t *zv)
887 zil_close(zv->zv_zilog);
890 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
896 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
897 !(zv->zv_flags & ZVOL_RDONLY))
898 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
899 dmu_objset_evict_dbufs(zv->zv_objset);
901 dmu_objset_disown(zv->zv_objset, zvol_tag);
902 zv->zv_objset = NULL;
907 zvol_prealloc(zvol_state_t *zv)
909 objset_t *os = zv->zv_objset;
911 uint64_t refd, avail, usedobjs, availobjs;
912 uint64_t resid = zv->zv_volsize;
915 /* Check the space usage before attempting to allocate the space */
916 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
917 if (avail < zv->zv_volsize)
918 return (SET_ERROR(ENOSPC));
920 /* Free old extents if they exist */
921 zvol_free_extents(zv);
925 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
927 tx = dmu_tx_create(os);
928 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
929 error = dmu_tx_assign(tx, TXG_WAIT);
932 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
935 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
940 txg_wait_synced(dmu_objset_pool(os), 0);
947 zvol_update_volsize(objset_t *os, uint64_t volsize)
952 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
954 tx = dmu_tx_create(os);
955 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
956 dmu_tx_mark_netfree(tx);
957 error = dmu_tx_assign(tx, TXG_WAIT);
963 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
968 error = dmu_free_long_range(os,
969 ZVOL_OBJ, volsize, DMU_OBJECT_END);
974 zvol_remove_minors(const char *name)
981 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
982 (void) strncpy(namebuf, name, strlen(name));
983 (void) strcat(namebuf, "/");
984 mutex_enter(&zfsdev_state_lock);
985 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
987 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
990 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
991 (void) zvol_remove_zv(zv);
993 kmem_free(namebuf, strlen(name) + 2);
995 mutex_exit(&zfsdev_state_lock);
997 zvol_state_t *zv, *tzv;
1000 namelen = strlen(name);
1003 mutex_enter(&zfsdev_state_lock);
1005 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
1006 if (strcmp(zv->zv_name, name) == 0 ||
1007 (strncmp(zv->zv_name, name, namelen) == 0 &&
1008 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
1009 zv->zv_name[namelen] == '@'))) {
1010 (void) zvol_remove_zv(zv);
1014 mutex_exit(&zfsdev_state_lock);
1016 #endif /* illumos */
1020 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
1022 uint64_t old_volsize = 0ULL;
1025 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1028 * Reinitialize the dump area to the new size. If we
1029 * failed to resize the dump area then restore it back to
1030 * its original size. We must set the new volsize prior
1031 * to calling dumpvp_resize() to ensure that the devices'
1032 * size(9P) is not visible by the dump subsystem.
1034 old_volsize = zv->zv_volsize;
1035 zvol_size_changed(zv, volsize);
1038 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1039 if ((error = zvol_dumpify(zv)) != 0 ||
1040 (error = dumpvp_resize()) != 0) {
1043 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1044 zvol_size_changed(zv, old_volsize);
1045 dumpify_error = zvol_dumpify(zv);
1046 error = dumpify_error ? dumpify_error : error;
1049 #endif /* ZVOL_DUMP */
1053 * Generate a LUN expansion event.
1058 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1060 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1063 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1064 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1066 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1067 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1070 kmem_free(physpath, MAXPATHLEN);
1072 #endif /* illumos */
1077 zvol_set_volsize(const char *name, uint64_t volsize)
1079 zvol_state_t *zv = NULL;
1082 dmu_object_info_t doi;
1084 boolean_t owned = B_FALSE;
1086 error = dsl_prop_get_integer(name,
1087 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1091 return (SET_ERROR(EROFS));
1093 mutex_enter(&zfsdev_state_lock);
1094 zv = zvol_minor_lookup(name);
1096 if (zv == NULL || zv->zv_objset == NULL) {
1097 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1099 mutex_exit(&zfsdev_state_lock);
1109 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1110 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1113 error = zvol_update_volsize(os, volsize);
1115 if (error == 0 && zv != NULL)
1116 error = zvol_update_live_volsize(zv, volsize);
1119 dmu_objset_disown(os, FTAG);
1121 zv->zv_objset = NULL;
1123 mutex_exit(&zfsdev_state_lock);
1130 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1133 zvol_open(struct g_provider *pp, int flag, int count)
1140 mutex_enter(&zfsdev_state_lock);
1142 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1144 mutex_exit(&zfsdev_state_lock);
1145 return (SET_ERROR(ENXIO));
1148 if (zv->zv_total_opens == 0)
1149 err = zvol_first_open(zv);
1151 mutex_exit(&zfsdev_state_lock);
1154 #else /* !illumos */
1155 boolean_t locked = B_FALSE;
1157 if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1159 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1160 * attempting to probe geom providers while looking for a
1161 * replacement for a missing VDEV. In this case, the
1162 * spa_namespace_lock will not be held, but it is still illegal
1163 * to use a zvol as a vdev. Deadlocks can result if another
1164 * thread has spa_namespace_lock
1166 return (EOPNOTSUPP);
1169 * Protect against recursively entering spa_namespace_lock
1170 * when spa_open() is used for a pool on a (local) ZVOL(s).
1171 * This is needed since we replaced upstream zfsdev_state_lock
1172 * with spa_namespace_lock in the ZVOL code.
1173 * We are using the same trick as spa_open().
1174 * Note that calls in zvol_first_open which need to resolve
1175 * pool name to a spa object will enter spa_open()
1176 * recursively, but that function already has all the
1177 * necessary protection.
1179 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1180 mutex_enter(&zfsdev_state_lock);
1187 mutex_exit(&zfsdev_state_lock);
1188 return (SET_ERROR(ENXIO));
1191 if (zv->zv_total_opens == 0) {
1192 err = zvol_first_open(zv);
1195 mutex_exit(&zfsdev_state_lock);
1198 pp->mediasize = zv->zv_volsize;
1199 pp->stripeoffset = 0;
1200 pp->stripesize = zv->zv_volblocksize;
1202 #endif /* illumos */
1203 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1204 err = SET_ERROR(EROFS);
1207 if (zv->zv_flags & ZVOL_EXCL) {
1208 err = SET_ERROR(EBUSY);
1213 if (zv->zv_total_opens != 0) {
1214 err = SET_ERROR(EBUSY);
1217 zv->zv_flags |= ZVOL_EXCL;
1222 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1223 zv->zv_open_count[otyp]++;
1224 zv->zv_total_opens++;
1226 mutex_exit(&zfsdev_state_lock);
1228 zv->zv_total_opens += count;
1230 mutex_exit(&zfsdev_state_lock);
1235 if (zv->zv_total_opens == 0)
1236 zvol_last_close(zv);
1238 mutex_exit(&zfsdev_state_lock);
1241 mutex_exit(&zfsdev_state_lock);
1249 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1251 minor_t minor = getminor(dev);
1255 mutex_enter(&zfsdev_state_lock);
1257 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1259 mutex_exit(&zfsdev_state_lock);
1260 #else /* !illumos */
1262 zvol_close(struct g_provider *pp, int flag, int count)
1266 boolean_t locked = B_FALSE;
1268 /* See comment in zvol_open(). */
1269 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1270 mutex_enter(&zfsdev_state_lock);
1277 mutex_exit(&zfsdev_state_lock);
1278 #endif /* illumos */
1279 return (SET_ERROR(ENXIO));
1282 if (zv->zv_flags & ZVOL_EXCL) {
1283 ASSERT(zv->zv_total_opens == 1);
1284 zv->zv_flags &= ~ZVOL_EXCL;
1288 * If the open count is zero, this is a spurious close.
1289 * That indicates a bug in the kernel / DDI framework.
1292 ASSERT(zv->zv_open_count[otyp] != 0);
1294 ASSERT(zv->zv_total_opens != 0);
1297 * You may get multiple opens, but only one close.
1300 zv->zv_open_count[otyp]--;
1301 zv->zv_total_opens--;
1303 zv->zv_total_opens -= count;
1306 if (zv->zv_total_opens == 0)
1307 zvol_last_close(zv);
1310 mutex_exit(&zfsdev_state_lock);
1313 mutex_exit(&zfsdev_state_lock);
1319 zvol_get_done(zgd_t *zgd, int error)
1322 dmu_buf_rele(zgd->zgd_db, zgd);
1324 zfs_range_unlock(zgd->zgd_rl);
1326 if (error == 0 && zgd->zgd_bp)
1327 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1329 kmem_free(zgd, sizeof (zgd_t));
1333 * Get data to generate a TX_WRITE intent log record.
1336 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1338 zvol_state_t *zv = arg;
1339 objset_t *os = zv->zv_objset;
1340 uint64_t object = ZVOL_OBJ;
1341 uint64_t offset = lr->lr_offset;
1342 uint64_t size = lr->lr_length; /* length of user data */
1347 ASSERT(zio != NULL);
1350 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1351 zgd->zgd_zilog = zv->zv_zilog;
1352 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1355 * Write records come in two flavors: immediate and indirect.
1356 * For small writes it's cheaper to store the data with the
1357 * log record (immediate); for large writes it's cheaper to
1358 * sync the data and get a pointer to it (indirect) so that
1359 * we don't have to write the data twice.
1361 if (buf != NULL) { /* immediate write */
1362 error = dmu_read(os, object, offset, size, buf,
1363 DMU_READ_NO_PREFETCH);
1365 size = zv->zv_volblocksize;
1366 offset = P2ALIGN(offset, size);
1367 error = dmu_buf_hold(os, object, offset, zgd, &db,
1368 DMU_READ_NO_PREFETCH);
1370 blkptr_t *bp = &lr->lr_blkptr;
1375 ASSERT(db->db_offset == offset);
1376 ASSERT(db->db_size == size);
1378 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1379 zvol_get_done, zgd);
1386 zvol_get_done(zgd, error);
1392 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1394 * We store data in the log buffers if it's small enough.
1395 * Otherwise we will later flush the data out via dmu_sync().
1397 ssize_t zvol_immediate_write_sz = 32768;
1399 SYSCTL_LONG(_vfs_zfs_vol, OID_AUTO, immediate_write_sz, CTLFLAG_RWTUN,
1400 &zvol_immediate_write_sz, 0, "Minimal size for indirect log write");
1404 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1407 uint32_t blocksize = zv->zv_volblocksize;
1408 zilog_t *zilog = zv->zv_zilog;
1409 itx_wr_state_t write_state;
1411 if (zil_replaying(zilog, tx))
1414 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1415 write_state = WR_INDIRECT;
1416 else if (!spa_has_slogs(zilog->zl_spa) &&
1417 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1418 write_state = WR_INDIRECT;
1420 write_state = WR_COPIED;
1422 write_state = WR_NEED_COPY;
1427 itx_wr_state_t wr_state = write_state;
1428 ssize_t len = resid;
1430 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1431 wr_state = WR_NEED_COPY;
1432 else if (wr_state == WR_INDIRECT)
1433 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1435 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1436 (wr_state == WR_COPIED ? len : 0));
1437 lr = (lr_write_t *)&itx->itx_lr;
1438 if (wr_state == WR_COPIED && dmu_read(zv->zv_objset,
1439 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1440 zil_itx_destroy(itx);
1441 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1442 lr = (lr_write_t *)&itx->itx_lr;
1443 wr_state = WR_NEED_COPY;
1446 itx->itx_wr_state = wr_state;
1447 lr->lr_foid = ZVOL_OBJ;
1448 lr->lr_offset = off;
1449 lr->lr_length = len;
1451 BP_ZERO(&lr->lr_blkptr);
1453 itx->itx_private = zv;
1455 if (!sync && (zv->zv_sync_cnt == 0))
1456 itx->itx_sync = B_FALSE;
1458 zil_itx_assign(zilog, itx, tx);
1467 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1468 uint64_t size, boolean_t doread, boolean_t isdump)
1474 if (vd->vdev_ops == &vdev_mirror_ops ||
1475 vd->vdev_ops == &vdev_replacing_ops ||
1476 vd->vdev_ops == &vdev_spare_ops) {
1477 for (c = 0; c < vd->vdev_children; c++) {
1478 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1479 addr, offset, origoffset, size, doread, isdump);
1482 } else if (doread) {
1488 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1489 return (numerrors < vd->vdev_children ? 0 : EIO);
1491 if (doread && !vdev_readable(vd))
1492 return (SET_ERROR(EIO));
1493 else if (!doread && !vdev_writeable(vd))
1494 return (SET_ERROR(EIO));
1496 if (vd->vdev_ops == &vdev_raidz_ops) {
1497 return (vdev_raidz_physio(vd,
1498 addr, size, offset, origoffset, doread, isdump));
1501 offset += VDEV_LABEL_START_SIZE;
1503 if (ddi_in_panic() || isdump) {
1506 return (SET_ERROR(EIO));
1508 ASSERT3P(dvd, !=, NULL);
1509 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1513 ASSERT3P(dvd, !=, NULL);
1514 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1515 offset, doread ? B_READ : B_WRITE));
1520 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1521 boolean_t doread, boolean_t isdump)
1526 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1528 /* Must be sector aligned, and not stradle a block boundary. */
1529 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1530 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1531 return (SET_ERROR(EINVAL));
1533 ASSERT(size <= zv->zv_volblocksize);
1535 /* Locate the extent this belongs to */
1536 ze = list_head(&zv->zv_extents);
1537 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1538 offset -= ze->ze_nblks * zv->zv_volblocksize;
1539 ze = list_next(&zv->zv_extents, ze);
1543 return (SET_ERROR(EINVAL));
1545 if (!ddi_in_panic())
1546 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1548 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1549 offset += DVA_GET_OFFSET(&ze->ze_dva);
1550 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1551 size, doread, isdump);
1553 if (!ddi_in_panic())
1554 spa_config_exit(spa, SCL_STATE, FTAG);
1560 zvol_strategy(buf_t *bp)
1562 zfs_soft_state_t *zs = NULL;
1563 #else /* !illumos */
1565 zvol_strategy(struct bio *bp)
1567 #endif /* illumos */
1569 uint64_t off, volsize;
1576 boolean_t doread = bp->b_flags & B_READ;
1578 boolean_t doread = 0;
1580 boolean_t is_dumpified;
1584 if (getminor(bp->b_edev) == 0) {
1585 error = SET_ERROR(EINVAL);
1587 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1589 error = SET_ERROR(ENXIO);
1590 else if (zs->zss_type != ZSST_ZVOL)
1591 error = SET_ERROR(EINVAL);
1595 bioerror(bp, error);
1602 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1603 bioerror(bp, EROFS);
1608 off = ldbtob(bp->b_blkno);
1609 #else /* !illumos */
1611 zv = bp->bio_to->private;
1613 zv = bp->bio_dev->si_drv2;
1616 error = SET_ERROR(ENXIO);
1620 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1621 error = SET_ERROR(EROFS);
1625 switch (bp->bio_cmd) {
1638 off = bp->bio_offset;
1639 #endif /* illumos */
1640 volsize = zv->zv_volsize;
1647 addr = bp->b_un.b_addr;
1648 resid = bp->b_bcount;
1650 if (resid > 0 && (off < 0 || off >= volsize)) {
1656 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1657 sync = ((!(bp->b_flags & B_ASYNC) &&
1658 !(zv->zv_flags & ZVOL_WCE)) ||
1659 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1660 !doread && !is_dumpified;
1661 #else /* !illumos */
1662 addr = bp->bio_data;
1663 resid = bp->bio_length;
1665 if (resid > 0 && (off < 0 || off >= volsize)) {
1666 error = SET_ERROR(EIO);
1670 is_dumpified = B_FALSE;
1671 sync = !doread && !is_dumpified &&
1672 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1673 #endif /* illumos */
1676 * There must be no buffer changes when doing a dmu_sync() because
1677 * we can't change the data whilst calculating the checksum.
1679 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1680 doread ? RL_READER : RL_WRITER);
1683 if (bp->bio_cmd == BIO_DELETE) {
1684 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1685 error = dmu_tx_assign(tx, TXG_WAIT);
1689 zvol_log_truncate(zv, tx, off, resid, sync);
1691 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1698 while (resid != 0 && off < volsize) {
1699 size_t size = MIN(resid, zvol_maxphys);
1702 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1703 error = zvol_dumpio(zv, addr, off, size,
1705 } else if (doread) {
1709 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1712 dmu_tx_t *tx = dmu_tx_create(os);
1713 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1714 error = dmu_tx_assign(tx, TXG_WAIT);
1718 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1719 zvol_log_write(zv, tx, off, size, sync);
1724 /* convert checksum errors into IO errors */
1725 if (error == ECKSUM)
1726 error = SET_ERROR(EIO);
1736 zfs_range_unlock(rl);
1739 if ((bp->b_resid = resid) == bp->b_bcount)
1740 bioerror(bp, off > volsize ? EINVAL : error);
1743 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1747 #else /* !illumos */
1748 bp->bio_completed = bp->bio_length - resid;
1749 if (bp->bio_completed < bp->bio_length && off > volsize)
1754 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1758 g_io_deliver(bp, error);
1760 biofinish(bp, NULL, error);
1761 #endif /* illumos */
1766 * Set the buffer count to the zvol maximum transfer.
1767 * Using our own routine instead of the default minphys()
1768 * means that for larger writes we write bigger buffers on X86
1769 * (128K instead of 56K) and flush the disk write cache less often
1770 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1771 * 56K on X86 and 128K on sparc).
1774 zvol_minphys(struct buf *bp)
1776 if (bp->b_bcount > zvol_maxphys)
1777 bp->b_bcount = zvol_maxphys;
1781 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1783 minor_t minor = getminor(dev);
1790 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1792 return (SET_ERROR(ENXIO));
1794 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1795 return (SET_ERROR(EINVAL));
1797 boff = ldbtob(blkno);
1798 resid = ldbtob(nblocks);
1800 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1803 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1804 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1817 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1819 minor_t minor = getminor(dev);
1820 #else /* !illumos */
1822 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1824 #endif /* illumos */
1831 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1833 return (SET_ERROR(ENXIO));
1838 volsize = zv->zv_volsize;
1839 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1840 if (uio->uio_resid > 0 &&
1841 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1842 return (SET_ERROR(EIO));
1845 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1846 error = physio(zvol_strategy, NULL, dev, B_READ,
1852 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1854 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1855 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1857 /* don't read past the end */
1858 if (bytes > volsize - uio->uio_loffset)
1859 bytes = volsize - uio->uio_loffset;
1861 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
1863 /* convert checksum errors into IO errors */
1864 if (error == ECKSUM)
1865 error = SET_ERROR(EIO);
1869 zfs_range_unlock(rl);
1876 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1878 minor_t minor = getminor(dev);
1879 #else /* !illumos */
1881 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1883 #endif /* illumos */
1891 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1893 return (SET_ERROR(ENXIO));
1898 volsize = zv->zv_volsize;
1899 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1900 if (uio->uio_resid > 0 &&
1901 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1902 return (SET_ERROR(EIO));
1905 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1906 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1911 sync = !(zv->zv_flags & ZVOL_WCE) ||
1913 sync = (ioflag & IO_SYNC) ||
1915 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1917 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1919 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1920 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1921 uint64_t off = uio->uio_loffset;
1922 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1924 if (bytes > volsize - off) /* don't write past the end */
1925 bytes = volsize - off;
1927 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1928 error = dmu_tx_assign(tx, TXG_WAIT);
1933 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1935 zvol_log_write(zv, tx, off, bytes, sync);
1941 zfs_range_unlock(rl);
1943 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1949 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1951 struct uuid uuid = EFI_RESERVED;
1952 efi_gpe_t gpe = { 0 };
1958 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1959 return (SET_ERROR(EFAULT));
1960 ptr = (char *)(uintptr_t)efi.dki_data_64;
1961 length = efi.dki_length;
1963 * Some clients may attempt to request a PMBR for the
1964 * zvol. Currently this interface will return EINVAL to
1965 * such requests. These requests could be supported by
1966 * adding a check for lba == 0 and consing up an appropriate
1969 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1970 return (SET_ERROR(EINVAL));
1972 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1973 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1974 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1976 if (efi.dki_lba == 1) {
1977 efi_gpt_t gpt = { 0 };
1979 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1980 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1981 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1982 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1983 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1984 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1985 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1986 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1987 gpt.efi_gpt_SizeOfPartitionEntry =
1988 LE_32(sizeof (efi_gpe_t));
1989 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1990 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1991 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1992 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1993 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1995 return (SET_ERROR(EFAULT));
1996 ptr += sizeof (gpt);
1997 length -= sizeof (gpt);
1999 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
2001 return (SET_ERROR(EFAULT));
2006 * BEGIN entry points to allow external callers access to the volume.
2009 * Return the volume parameters needed for access from an external caller.
2010 * These values are invariant as long as the volume is held open.
2013 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
2014 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
2015 void **rl_hdl, void **bonus_hdl)
2019 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2021 return (SET_ERROR(ENXIO));
2022 if (zv->zv_flags & ZVOL_DUMPIFIED)
2023 return (SET_ERROR(ENXIO));
2025 ASSERT(blksize && max_xfer_len && minor_hdl &&
2026 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
2028 *blksize = zv->zv_volblocksize;
2029 *max_xfer_len = (uint64_t)zvol_maxphys;
2031 *objset_hdl = zv->zv_objset;
2032 *zil_hdl = zv->zv_zilog;
2033 *rl_hdl = &zv->zv_znode;
2034 *bonus_hdl = zv->zv_dbuf;
2039 * Return the current volume size to an external caller.
2040 * The size can change while the volume is open.
2043 zvol_get_volume_size(void *minor_hdl)
2045 zvol_state_t *zv = minor_hdl;
2047 return (zv->zv_volsize);
2051 * Return the current WCE setting to an external caller.
2052 * The WCE setting can change while the volume is open.
2055 zvol_get_volume_wce(void *minor_hdl)
2057 zvol_state_t *zv = minor_hdl;
2059 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2063 * Entry point for external callers to zvol_log_write
2066 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2069 zvol_state_t *zv = minor_hdl;
2071 zvol_log_write(zv, tx, off, resid, sync);
2074 * END entry points to allow external callers access to the volume.
2076 #endif /* illumos */
2079 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2082 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2087 zilog_t *zilog = zv->zv_zilog;
2089 if (zil_replaying(zilog, tx))
2092 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2093 lr = (lr_truncate_t *)&itx->itx_lr;
2094 lr->lr_foid = ZVOL_OBJ;
2095 lr->lr_offset = off;
2096 lr->lr_length = len;
2098 itx->itx_sync = (sync || zv->zv_sync_cnt != 0);
2099 zil_itx_assign(zilog, itx, tx);
2104 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2105 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2109 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2112 struct dk_callback *dkc;
2116 mutex_enter(&zfsdev_state_lock);
2118 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2121 mutex_exit(&zfsdev_state_lock);
2122 return (SET_ERROR(ENXIO));
2124 ASSERT(zv->zv_total_opens > 0);
2130 struct dk_cinfo dki;
2132 bzero(&dki, sizeof (dki));
2133 (void) strcpy(dki.dki_cname, "zvol");
2134 (void) strcpy(dki.dki_dname, "zvol");
2135 dki.dki_ctype = DKC_UNKNOWN;
2136 dki.dki_unit = getminor(dev);
2137 dki.dki_maxtransfer =
2138 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2139 mutex_exit(&zfsdev_state_lock);
2140 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2141 error = SET_ERROR(EFAULT);
2145 case DKIOCGMEDIAINFO:
2147 struct dk_minfo dkm;
2149 bzero(&dkm, sizeof (dkm));
2150 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2151 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2152 dkm.dki_media_type = DK_UNKNOWN;
2153 mutex_exit(&zfsdev_state_lock);
2154 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2155 error = SET_ERROR(EFAULT);
2159 case DKIOCGMEDIAINFOEXT:
2161 struct dk_minfo_ext dkmext;
2163 bzero(&dkmext, sizeof (dkmext));
2164 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2165 dkmext.dki_pbsize = zv->zv_volblocksize;
2166 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2167 dkmext.dki_media_type = DK_UNKNOWN;
2168 mutex_exit(&zfsdev_state_lock);
2169 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2170 error = SET_ERROR(EFAULT);
2176 uint64_t vs = zv->zv_volsize;
2177 uint8_t bs = zv->zv_min_bs;
2179 mutex_exit(&zfsdev_state_lock);
2180 error = zvol_getefi((void *)arg, flag, vs, bs);
2184 case DKIOCFLUSHWRITECACHE:
2185 dkc = (struct dk_callback *)arg;
2186 mutex_exit(&zfsdev_state_lock);
2187 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2188 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2189 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2196 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2197 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2199 error = SET_ERROR(EFAULT);
2205 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2207 error = SET_ERROR(EFAULT);
2211 zv->zv_flags |= ZVOL_WCE;
2212 mutex_exit(&zfsdev_state_lock);
2214 zv->zv_flags &= ~ZVOL_WCE;
2215 mutex_exit(&zfsdev_state_lock);
2216 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2224 * commands using these (like prtvtoc) expect ENOTSUP
2225 * since we're emulating an EFI label
2227 error = SET_ERROR(ENOTSUP);
2231 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2233 error = zvol_dumpify(zv);
2234 zfs_range_unlock(rl);
2238 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2240 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2242 error = zvol_dump_fini(zv);
2243 zfs_range_unlock(rl);
2251 if (!zvol_unmap_enabled)
2254 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2255 error = SET_ERROR(EFAULT);
2260 * Apply Postel's Law to length-checking. If they overshoot,
2261 * just blank out until the end, if there's a need to blank
2264 if (df.df_start >= zv->zv_volsize)
2265 break; /* No need to do anything... */
2267 mutex_exit(&zfsdev_state_lock);
2269 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2271 tx = dmu_tx_create(zv->zv_objset);
2272 dmu_tx_mark_netfree(tx);
2273 error = dmu_tx_assign(tx, TXG_WAIT);
2277 zvol_log_truncate(zv, tx, df.df_start,
2278 df.df_length, B_TRUE);
2280 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2281 df.df_start, df.df_length);
2284 zfs_range_unlock(rl);
2287 * If the write-cache is disabled, 'sync' property
2288 * is set to 'always', or if the caller is asking for
2289 * a synchronous free, commit this operation to the zil.
2290 * This will sync any previous uncommitted writes to the
2292 * Can be overridden by the zvol_unmap_sync_enabled tunable.
2294 if ((error == 0) && zvol_unmap_sync_enabled &&
2295 (!(zv->zv_flags & ZVOL_WCE) ||
2296 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
2297 (df.df_flags & DF_WAIT_SYNC))) {
2298 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2305 error = SET_ERROR(ENOTTY);
2309 mutex_exit(&zfsdev_state_lock);
2312 #endif /* illumos */
2317 return (zvol_minors != 0);
2323 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2326 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2328 ZFS_LOG(1, "ZVOL Initialized.");
2336 mutex_destroy(&zfsdev_state_lock);
2338 ddi_soft_state_fini(&zfsdev_state);
2339 ZFS_LOG(1, "ZVOL Deinitialized.");
2345 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2347 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2349 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2356 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2358 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2360 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2364 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2368 objset_t *os = zv->zv_objset;
2369 spa_t *spa = dmu_objset_spa(os);
2370 vdev_t *vd = spa->spa_root_vdev;
2371 nvlist_t *nv = NULL;
2372 uint64_t version = spa_version(spa);
2373 uint64_t checksum, compress, refresrv, vbs, dedup;
2375 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2376 ASSERT(vd->vdev_ops == &vdev_root_ops);
2378 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2382 /* wait for dmu_free_long_range to actually free the blocks */
2383 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2386 * If the pool on which the dump device is being initialized has more
2387 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2388 * enabled. If so, bump that feature's counter to indicate that the
2389 * feature is active. We also check the vdev type to handle the
2391 * # zpool create test raidz disk1 disk2 disk3
2392 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2393 * the raidz vdev itself has 3 children.
2395 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2396 if (!spa_feature_is_enabled(spa,
2397 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2398 return (SET_ERROR(ENOTSUP));
2399 (void) dsl_sync_task(spa_name(spa),
2400 zfs_mvdev_dump_feature_check,
2401 zfs_mvdev_dump_activate_feature_sync, NULL,
2402 2, ZFS_SPACE_CHECK_RESERVED);
2406 error = dsl_prop_get_integer(zv->zv_name,
2407 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2409 error = dsl_prop_get_integer(zv->zv_name,
2410 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2414 error = dsl_prop_get_integer(zv->zv_name,
2415 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2419 error = dsl_prop_get_integer(zv->zv_name,
2420 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2423 if (version >= SPA_VERSION_DEDUP && error == 0) {
2424 error = dsl_prop_get_integer(zv->zv_name,
2425 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2431 tx = dmu_tx_create(os);
2432 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2433 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2434 error = dmu_tx_assign(tx, TXG_WAIT);
2441 * If we are resizing the dump device then we only need to
2442 * update the refreservation to match the newly updated
2443 * zvolsize. Otherwise, we save off the original state of the
2444 * zvol so that we can restore them if the zvol is ever undumpified.
2447 error = zap_update(os, ZVOL_ZAP_OBJ,
2448 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2449 &zv->zv_volsize, tx);
2451 error = zap_update(os, ZVOL_ZAP_OBJ,
2452 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2455 error = zap_update(os, ZVOL_ZAP_OBJ,
2456 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2460 error = zap_update(os, ZVOL_ZAP_OBJ,
2461 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2465 error = zap_update(os, ZVOL_ZAP_OBJ,
2466 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2470 error = dmu_object_set_blocksize(
2471 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2473 if (version >= SPA_VERSION_DEDUP && error == 0) {
2474 error = zap_update(os, ZVOL_ZAP_OBJ,
2475 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2479 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2484 * We only need update the zvol's property if we are initializing
2485 * the dump area for the first time.
2487 if (error == 0 && !resize) {
2489 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2490 * function. Otherwise, use the old default -- OFF.
2492 checksum = spa_feature_is_active(spa,
2493 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2496 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2497 VERIFY(nvlist_add_uint64(nv,
2498 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2499 VERIFY(nvlist_add_uint64(nv,
2500 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2501 ZIO_COMPRESS_OFF) == 0);
2502 VERIFY(nvlist_add_uint64(nv,
2503 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2505 if (version >= SPA_VERSION_DEDUP) {
2506 VERIFY(nvlist_add_uint64(nv,
2507 zfs_prop_to_name(ZFS_PROP_DEDUP),
2508 ZIO_CHECKSUM_OFF) == 0);
2511 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2516 /* Allocate the space for the dump */
2518 error = zvol_prealloc(zv);
2523 zvol_dumpify(zvol_state_t *zv)
2526 uint64_t dumpsize = 0;
2528 objset_t *os = zv->zv_objset;
2530 if (zv->zv_flags & ZVOL_RDONLY)
2531 return (SET_ERROR(EROFS));
2533 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2534 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2535 boolean_t resize = (dumpsize > 0);
2537 if ((error = zvol_dump_init(zv, resize)) != 0) {
2538 (void) zvol_dump_fini(zv);
2544 * Build up our lba mapping.
2546 error = zvol_get_lbas(zv);
2548 (void) zvol_dump_fini(zv);
2552 tx = dmu_tx_create(os);
2553 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2554 error = dmu_tx_assign(tx, TXG_WAIT);
2557 (void) zvol_dump_fini(zv);
2561 zv->zv_flags |= ZVOL_DUMPIFIED;
2562 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2563 &zv->zv_volsize, tx);
2567 (void) zvol_dump_fini(zv);
2571 txg_wait_synced(dmu_objset_pool(os), 0);
2576 zvol_dump_fini(zvol_state_t *zv)
2579 objset_t *os = zv->zv_objset;
2582 uint64_t checksum, compress, refresrv, vbs, dedup;
2583 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2586 * Attempt to restore the zvol back to its pre-dumpified state.
2587 * This is a best-effort attempt as it's possible that not all
2588 * of these properties were initialized during the dumpify process
2589 * (i.e. error during zvol_dump_init).
2592 tx = dmu_tx_create(os);
2593 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2594 error = dmu_tx_assign(tx, TXG_WAIT);
2599 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2602 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2603 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2604 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2605 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2606 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2607 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2608 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2609 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2611 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2612 (void) nvlist_add_uint64(nv,
2613 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2614 (void) nvlist_add_uint64(nv,
2615 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2616 (void) nvlist_add_uint64(nv,
2617 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2618 if (version >= SPA_VERSION_DEDUP &&
2619 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2620 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2621 (void) nvlist_add_uint64(nv,
2622 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2624 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2628 zvol_free_extents(zv);
2629 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2630 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2631 /* wait for dmu_free_long_range to actually free the blocks */
2632 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2633 tx = dmu_tx_create(os);
2634 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2635 error = dmu_tx_assign(tx, TXG_WAIT);
2640 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2641 zv->zv_volblocksize = vbs;
2646 #else /* !illumos */
2649 zvol_geom_run(zvol_state_t *zv)
2651 struct g_provider *pp;
2653 pp = zv->zv_provider;
2654 g_error_provider(pp, 0);
2656 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2657 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2661 zvol_geom_destroy(zvol_state_t *zv)
2663 struct g_provider *pp;
2665 g_topology_assert();
2667 mtx_lock(&zv->zv_queue_mtx);
2669 wakeup_one(&zv->zv_queue);
2670 while (zv->zv_state != 2)
2671 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2672 mtx_destroy(&zv->zv_queue_mtx);
2674 pp = zv->zv_provider;
2675 zv->zv_provider = NULL;
2677 g_wither_geom(pp->geom, ENXIO);
2681 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2683 int count, error, flags;
2685 g_topology_assert();
2688 * To make it easier we expect either open or close, but not both
2691 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2692 (acr <= 0 && acw <= 0 && ace <= 0),
2693 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2694 pp->name, acr, acw, ace));
2696 if (pp->private == NULL) {
2697 if (acr <= 0 && acw <= 0 && ace <= 0)
2703 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2704 * because GEOM already handles that and handles it a bit differently.
2705 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2706 * only one exclusive consumer, no matter if it is reader or writer.
2707 * I like better the way GEOM works so I'll leave it for GEOM to
2708 * decide what to do.
2711 count = acr + acw + ace;
2716 if (acr != 0 || ace != 0)
2721 g_topology_unlock();
2723 error = zvol_open(pp, flags, count);
2725 error = zvol_close(pp, flags, -count);
2731 zvol_geom_start(struct bio *bp)
2736 zv = bp->bio_to->private;
2738 switch (bp->bio_cmd) {
2740 if (!THREAD_CAN_SLEEP())
2742 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2743 g_io_deliver(bp, 0);
2748 if (!THREAD_CAN_SLEEP())
2753 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2754 uint64_t refd, avail, usedobjs, availobjs, val;
2756 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2758 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2759 dmu_objset_space(zv->zv_objset, &refd, &avail,
2760 &usedobjs, &availobjs);
2761 if (g_handleattr_off_t(bp, "blocksavail",
2764 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2765 dmu_objset_space(zv->zv_objset, &refd, &avail,
2766 &usedobjs, &availobjs);
2767 if (g_handleattr_off_t(bp, "blocksused",
2770 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2771 avail = metaslab_class_get_space(spa_normal_class(spa));
2772 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2773 if (g_handleattr_off_t(bp, "poolblocksavail",
2776 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2777 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2778 if (g_handleattr_off_t(bp, "poolblocksused",
2785 g_io_deliver(bp, EOPNOTSUPP);
2791 mtx_lock(&zv->zv_queue_mtx);
2792 first = (bioq_first(&zv->zv_queue) == NULL);
2793 bioq_insert_tail(&zv->zv_queue, bp);
2794 mtx_unlock(&zv->zv_queue_mtx);
2796 wakeup_one(&zv->zv_queue);
2800 zvol_geom_worker(void *arg)
2805 thread_lock(curthread);
2806 sched_prio(curthread, PRIBIO);
2807 thread_unlock(curthread);
2811 mtx_lock(&zv->zv_queue_mtx);
2812 bp = bioq_takefirst(&zv->zv_queue);
2814 if (zv->zv_state == 1) {
2816 wakeup(&zv->zv_state);
2817 mtx_unlock(&zv->zv_queue_mtx);
2820 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2824 mtx_unlock(&zv->zv_queue_mtx);
2825 switch (bp->bio_cmd) {
2827 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2828 g_io_deliver(bp, 0);
2836 g_io_deliver(bp, EOPNOTSUPP);
2842 extern boolean_t dataset_name_hidden(const char *name);
2845 zvol_create_snapshots(objset_t *os, const char *name)
2847 uint64_t cookie, obj;
2852 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2855 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2860 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2861 if (len >= MAXPATHLEN) {
2862 dmu_objset_rele(os, FTAG);
2863 error = ENAMETOOLONG;
2867 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2868 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2869 sname + len, &obj, &cookie, NULL);
2870 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2872 if (error == ENOENT)
2877 error = zvol_create_minor(sname);
2878 if (error != 0 && error != EEXIST) {
2879 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2885 kmem_free(sname, MAXPATHLEN);
2890 zvol_create_minors(const char *name)
2897 if (dataset_name_hidden(name))
2900 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2901 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2905 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2906 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2907 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2908 error = zvol_create_minor(name);
2909 if (error == 0 || error == EEXIST) {
2910 error = zvol_create_snapshots(os, name);
2912 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2915 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2916 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2919 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2920 dmu_objset_rele(os, FTAG);
2924 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2925 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2926 dmu_objset_rele(os, FTAG);
2927 kmem_free(osname, MAXPATHLEN);
2930 p = osname + strlen(osname);
2931 len = MAXPATHLEN - (p - osname);
2934 /* Prefetch the datasets. */
2936 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2937 if (!dataset_name_hidden(osname))
2938 (void) dmu_objset_prefetch(osname, NULL);
2943 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2945 dmu_objset_rele(os, FTAG);
2946 (void)zvol_create_minors(osname);
2947 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2948 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2954 dmu_objset_rele(os, FTAG);
2955 kmem_free(osname, MAXPATHLEN);
2960 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2963 struct g_provider *pp;
2966 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2968 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2970 pp = zv->zv_provider;
2975 zv->zv_provider = NULL;
2976 g_wither_provider(pp, ENXIO);
2978 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2979 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2980 pp->sectorsize = DEV_BSIZE;
2981 pp->mediasize = zv->zv_volsize;
2983 zv->zv_provider = pp;
2984 g_error_provider(pp, 0);
2985 g_topology_unlock();
2986 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2987 struct make_dev_args args;
2989 if ((dev = zv->zv_dev) != NULL) {
2992 if (zv->zv_total_opens > 0) {
2993 zv->zv_flags &= ~ZVOL_EXCL;
2994 zv->zv_total_opens = 0;
2995 zvol_last_close(zv);
2999 make_dev_args_init(&args);
3000 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
3001 args.mda_devsw = &zvol_cdevsw;
3003 args.mda_uid = UID_ROOT;
3004 args.mda_gid = GID_OPERATOR;
3005 args.mda_mode = 0640;
3006 args.mda_si_drv2 = zv;
3007 if (make_dev_s(&args, &zv->zv_dev,
3008 "%s/%s", ZVOL_DRIVER, newname) == 0)
3009 zv->zv_dev->si_iosize_max = MAXPHYS;
3011 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3015 zvol_rename_minors(const char *oldname, const char *newname)
3017 char name[MAXPATHLEN];
3018 struct g_provider *pp;
3020 size_t oldnamelen, newnamelen;
3023 boolean_t locked = B_FALSE;
3025 oldnamelen = strlen(oldname);
3026 newnamelen = strlen(newname);
3029 /* See comment in zvol_open(). */
3030 if (!MUTEX_HELD(&zfsdev_state_lock)) {
3031 mutex_enter(&zfsdev_state_lock);
3035 LIST_FOREACH(zv, &all_zvols, zv_links) {
3036 if (strcmp(zv->zv_name, oldname) == 0) {
3037 zvol_rename_minor(zv, newname);
3038 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3039 (zv->zv_name[oldnamelen] == '/' ||
3040 zv->zv_name[oldnamelen] == '@')) {
3041 snprintf(name, sizeof(name), "%s%c%s", newname,
3042 zv->zv_name[oldnamelen],
3043 zv->zv_name + oldnamelen + 1);
3044 zvol_rename_minor(zv, name);
3049 mutex_exit(&zfsdev_state_lock);
3054 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3056 zvol_state_t *zv = dev->si_drv2;
3059 mutex_enter(&zfsdev_state_lock);
3060 if (zv->zv_total_opens == 0)
3061 err = zvol_first_open(zv);
3063 mutex_exit(&zfsdev_state_lock);
3066 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3067 err = SET_ERROR(EROFS);
3070 if (zv->zv_flags & ZVOL_EXCL) {
3071 err = SET_ERROR(EBUSY);
3075 if (flags & FEXCL) {
3076 if (zv->zv_total_opens != 0) {
3077 err = SET_ERROR(EBUSY);
3080 zv->zv_flags |= ZVOL_EXCL;
3084 zv->zv_total_opens++;
3085 if (flags & (FSYNC | FDSYNC)) {
3087 if (zv->zv_sync_cnt == 1)
3088 zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
3090 mutex_exit(&zfsdev_state_lock);
3093 if (zv->zv_total_opens == 0)
3094 zvol_last_close(zv);
3095 mutex_exit(&zfsdev_state_lock);
3100 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3102 zvol_state_t *zv = dev->si_drv2;
3104 mutex_enter(&zfsdev_state_lock);
3105 if (zv->zv_flags & ZVOL_EXCL) {
3106 ASSERT(zv->zv_total_opens == 1);
3107 zv->zv_flags &= ~ZVOL_EXCL;
3111 * If the open count is zero, this is a spurious close.
3112 * That indicates a bug in the kernel / DDI framework.
3114 ASSERT(zv->zv_total_opens != 0);
3117 * You may get multiple opens, but only one close.
3119 zv->zv_total_opens--;
3120 if (flags & (FSYNC | FDSYNC))
3123 if (zv->zv_total_opens == 0)
3124 zvol_last_close(zv);
3126 mutex_exit(&zfsdev_state_lock);
3131 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3135 off_t offset, length;
3142 KASSERT(zv->zv_total_opens > 0,
3143 ("Device with zero access count in zvol_d_ioctl"));
3145 i = IOCPARM_LEN(cmd);
3147 case DIOCGSECTORSIZE:
3148 *(u_int *)data = DEV_BSIZE;
3150 case DIOCGMEDIASIZE:
3151 *(off_t *)data = zv->zv_volsize;
3154 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3157 if (!zvol_unmap_enabled)
3160 offset = ((off_t *)data)[0];
3161 length = ((off_t *)data)[1];
3162 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3163 offset < 0 || offset >= zv->zv_volsize ||
3165 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3171 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3172 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3173 error = dmu_tx_assign(tx, TXG_WAIT);
3178 sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
3179 zvol_log_truncate(zv, tx, offset, length, sync);
3181 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3184 zfs_range_unlock(rl);
3186 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3188 case DIOCGSTRIPESIZE:
3189 *(off_t *)data = zv->zv_volblocksize;
3191 case DIOCGSTRIPEOFFSET:
3195 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3196 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3197 uint64_t refd, avail, usedobjs, availobjs;
3199 if (strcmp(arg->name, "GEOM::candelete") == 0)
3201 else if (strcmp(arg->name, "blocksavail") == 0) {
3202 dmu_objset_space(zv->zv_objset, &refd, &avail,
3203 &usedobjs, &availobjs);
3204 arg->value.off = avail / DEV_BSIZE;
3205 } else if (strcmp(arg->name, "blocksused") == 0) {
3206 dmu_objset_space(zv->zv_objset, &refd, &avail,
3207 &usedobjs, &availobjs);
3208 arg->value.off = refd / DEV_BSIZE;
3209 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3210 avail = metaslab_class_get_space(spa_normal_class(spa));
3211 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3212 arg->value.off = avail / DEV_BSIZE;
3213 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3214 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3215 arg->value.off = refd / DEV_BSIZE;
3222 off_t *off = (off_t *)data;
3226 hole = (cmd == FIOSEEKHOLE);
3228 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3238 #endif /* illumos */