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, 2014 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 TUNABLE_INT("vfs.zfs.vol.mode", &volmode);
137 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
138 "Expose as GEOM providers (1), device files (2) or neither");
141 typedef struct zvol_extent {
143 dva_t ze_dva; /* dva associated with this extent */
144 uint64_t ze_nblks; /* number of blocks in extent */
148 * The in-core state of each volume.
150 typedef struct zvol_state {
152 LIST_ENTRY(zvol_state) zv_links;
154 char zv_name[MAXPATHLEN]; /* pool/dd name */
155 uint64_t zv_volsize; /* amount of space we advertise */
156 uint64_t zv_volblocksize; /* volume block size */
158 minor_t zv_minor; /* minor number */
160 struct cdev *zv_dev; /* non-GEOM device */
161 struct g_provider *zv_provider; /* GEOM provider */
163 uint8_t zv_min_bs; /* minimum addressable block shift */
164 uint8_t zv_flags; /* readonly, dumpified, etc. */
165 objset_t *zv_objset; /* objset handle */
167 uint32_t zv_open_count[OTYPCNT]; /* open counts */
169 uint32_t zv_total_opens; /* total open count */
170 zilog_t *zv_zilog; /* ZIL handle */
171 list_t zv_extents; /* List of extents for dump */
172 znode_t zv_znode; /* for range locking */
173 dmu_buf_t *zv_dbuf; /* bonus handle */
176 int zv_volmode; /* Provide GEOM or cdev */
177 struct bio_queue_head zv_queue;
178 struct mtx zv_queue_mtx; /* zv_queue mutex */
183 static LIST_HEAD(, zvol_state) all_zvols;
186 * zvol specific flags
188 #define ZVOL_RDONLY 0x1
189 #define ZVOL_DUMPIFIED 0x2
190 #define ZVOL_EXCL 0x4
194 * zvol maximum transfer in one DMU tx.
196 int zvol_maxphys = DMU_MAX_ACCESS/2;
199 * Toggle unmap functionality.
201 boolean_t zvol_unmap_enabled = B_TRUE;
203 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
204 &zvol_unmap_enabled, 0,
205 "Enable UNMAP functionality");
207 static d_open_t zvol_d_open;
208 static d_close_t zvol_d_close;
209 static d_read_t zvol_read;
210 static d_write_t zvol_write;
211 static d_ioctl_t zvol_d_ioctl;
212 static d_strategy_t zvol_strategy;
214 static struct cdevsw zvol_cdevsw = {
215 .d_version = D_VERSION,
216 .d_open = zvol_d_open,
217 .d_close = zvol_d_close,
219 .d_write = zvol_write,
220 .d_ioctl = zvol_d_ioctl,
221 .d_strategy = zvol_strategy,
223 .d_flags = D_DISK | D_TRACKCLOSE,
226 static void zvol_geom_run(zvol_state_t *zv);
227 static void zvol_geom_destroy(zvol_state_t *zv);
228 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
229 static void zvol_geom_start(struct bio *bp);
230 static void zvol_geom_worker(void *arg);
231 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
232 uint64_t len, boolean_t sync);
233 #endif /* !illumos */
235 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
236 nvlist_t *, nvlist_t *);
237 static int zvol_remove_zv(zvol_state_t *);
238 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
239 static int zvol_dumpify(zvol_state_t *zv);
240 static int zvol_dump_fini(zvol_state_t *zv);
241 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
244 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
247 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
249 zv->zv_volsize = volsize;
250 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
251 "Size", volsize) == DDI_SUCCESS);
252 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
253 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
255 /* Notify specfs to invalidate the cached size */
256 spec_size_invalidate(dev, VBLK);
257 spec_size_invalidate(dev, VCHR);
259 zv->zv_volsize = volsize;
260 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
261 struct g_provider *pp;
263 pp = zv->zv_provider;
267 g_resize_provider(pp, zv->zv_volsize);
274 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
277 return (SET_ERROR(EINVAL));
279 if (volsize % blocksize != 0)
280 return (SET_ERROR(EINVAL));
283 if (volsize - 1 > SPEC_MAXOFFSET_T)
284 return (SET_ERROR(EOVERFLOW));
290 zvol_check_volblocksize(uint64_t volblocksize)
292 if (volblocksize < SPA_MINBLOCKSIZE ||
293 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
295 return (SET_ERROR(EDOM));
301 zvol_get_stats(objset_t *os, nvlist_t *nv)
304 dmu_object_info_t doi;
307 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
311 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
313 error = dmu_object_info(os, ZVOL_OBJ, &doi);
316 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
317 doi.doi_data_block_size);
323 static zvol_state_t *
324 zvol_minor_lookup(const char *name)
331 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
334 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
335 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
339 LIST_FOREACH(zv, &all_zvols, zv_links) {
341 if (strcmp(zv->zv_name, name) == 0)
348 /* extent mapping arg */
356 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
357 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
359 struct maparg *ma = arg;
361 int bs = ma->ma_zv->zv_volblocksize;
363 if (bp == NULL || BP_IS_HOLE(bp) ||
364 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
367 VERIFY(!BP_IS_EMBEDDED(bp));
369 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
372 /* Abort immediately if we have encountered gang blocks */
374 return (SET_ERROR(EFRAGS));
377 * See if the block is at the end of the previous extent.
379 ze = list_tail(&ma->ma_zv->zv_extents);
381 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
382 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
383 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
388 dprintf_bp(bp, "%s", "next blkptr:");
390 /* start a new extent */
391 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
392 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
394 list_insert_tail(&ma->ma_zv->zv_extents, ze);
399 zvol_free_extents(zvol_state_t *zv)
403 while (ze = list_head(&zv->zv_extents)) {
404 list_remove(&zv->zv_extents, ze);
405 kmem_free(ze, sizeof (zvol_extent_t));
410 zvol_get_lbas(zvol_state_t *zv)
412 objset_t *os = zv->zv_objset;
418 zvol_free_extents(zv);
420 /* commit any in-flight changes before traversing the dataset */
421 txg_wait_synced(dmu_objset_pool(os), 0);
422 err = traverse_dataset(dmu_objset_ds(os), 0,
423 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
424 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
425 zvol_free_extents(zv);
426 return (err ? err : EIO);
434 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
436 zfs_creat_t *zct = arg;
437 nvlist_t *nvprops = zct->zct_props;
439 uint64_t volblocksize, volsize;
441 VERIFY(nvlist_lookup_uint64(nvprops,
442 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
443 if (nvlist_lookup_uint64(nvprops,
444 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
445 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
448 * These properties must be removed from the list so the generic
449 * property setting step won't apply to them.
451 VERIFY(nvlist_remove_all(nvprops,
452 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
453 (void) nvlist_remove_all(nvprops,
454 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
456 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
460 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
464 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
469 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
470 * implement DKIOCFREE/free-long-range.
473 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
475 uint64_t offset, length;
478 byteswap_uint64_array(lr, sizeof (*lr));
480 offset = lr->lr_offset;
481 length = lr->lr_length;
483 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
487 * Replay a TX_WRITE ZIL transaction that didn't get committed
488 * after a system failure
491 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
493 objset_t *os = zv->zv_objset;
494 char *data = (char *)(lr + 1); /* data follows lr_write_t */
495 uint64_t offset, length;
500 byteswap_uint64_array(lr, sizeof (*lr));
502 offset = lr->lr_offset;
503 length = lr->lr_length;
505 /* If it's a dmu_sync() block, write the whole block */
506 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
507 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
508 if (length < blocksize) {
509 offset -= offset % blocksize;
514 tx = dmu_tx_create(os);
515 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
516 error = dmu_tx_assign(tx, TXG_WAIT);
520 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
529 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
531 return (SET_ERROR(ENOTSUP));
535 * Callback vectors for replaying records.
536 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
538 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
539 zvol_replay_err, /* 0 no such transaction type */
540 zvol_replay_err, /* TX_CREATE */
541 zvol_replay_err, /* TX_MKDIR */
542 zvol_replay_err, /* TX_MKXATTR */
543 zvol_replay_err, /* TX_SYMLINK */
544 zvol_replay_err, /* TX_REMOVE */
545 zvol_replay_err, /* TX_RMDIR */
546 zvol_replay_err, /* TX_LINK */
547 zvol_replay_err, /* TX_RENAME */
548 zvol_replay_write, /* TX_WRITE */
549 zvol_replay_truncate, /* TX_TRUNCATE */
550 zvol_replay_err, /* TX_SETATTR */
551 zvol_replay_err, /* TX_ACL */
552 zvol_replay_err, /* TX_CREATE_ACL */
553 zvol_replay_err, /* TX_CREATE_ATTR */
554 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
555 zvol_replay_err, /* TX_MKDIR_ACL */
556 zvol_replay_err, /* TX_MKDIR_ATTR */
557 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
558 zvol_replay_err, /* TX_WRITE2 */
563 zvol_name2minor(const char *name, minor_t *minor)
567 mutex_enter(&zfsdev_state_lock);
568 zv = zvol_minor_lookup(name);
570 *minor = zv->zv_minor;
571 mutex_exit(&zfsdev_state_lock);
572 return (zv ? 0 : -1);
577 * Create a minor node (plus a whole lot more) for the specified volume.
580 zvol_create_minor(const char *name)
582 zfs_soft_state_t *zs;
585 dmu_object_info_t doi;
588 char chrbuf[30], blkbuf[30];
590 struct g_provider *pp;
592 uint64_t volsize, mode;
597 ZFS_LOG(1, "Creating ZVOL %s...", name);
600 mutex_enter(&zfsdev_state_lock);
602 if (zvol_minor_lookup(name) != NULL) {
603 mutex_exit(&zfsdev_state_lock);
604 return (SET_ERROR(EEXIST));
607 /* lie and say we're read-only */
608 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
611 mutex_exit(&zfsdev_state_lock);
616 if ((minor = zfsdev_minor_alloc()) == 0) {
617 dmu_objset_disown(os, FTAG);
618 mutex_exit(&zfsdev_state_lock);
619 return (SET_ERROR(ENXIO));
622 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
623 dmu_objset_disown(os, FTAG);
624 mutex_exit(&zfsdev_state_lock);
625 return (SET_ERROR(EAGAIN));
627 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
630 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
632 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
633 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
634 ddi_soft_state_free(zfsdev_state, minor);
635 dmu_objset_disown(os, FTAG);
636 mutex_exit(&zfsdev_state_lock);
637 return (SET_ERROR(EAGAIN));
640 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
642 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
643 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
644 ddi_remove_minor_node(zfs_dip, chrbuf);
645 ddi_soft_state_free(zfsdev_state, minor);
646 dmu_objset_disown(os, FTAG);
647 mutex_exit(&zfsdev_state_lock);
648 return (SET_ERROR(EAGAIN));
651 zs = ddi_get_soft_state(zfsdev_state, minor);
652 zs->zss_type = ZSST_ZVOL;
653 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
656 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
658 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
660 kmem_free(zv, sizeof(*zv));
661 dmu_objset_disown(os, zvol_tag);
662 mutex_exit(&zfsdev_state_lock);
665 error = dsl_prop_get_integer(name,
666 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
667 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
671 zv->zv_volsize = volsize;
672 zv->zv_volmode = mode;
673 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
675 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
676 gp->start = zvol_geom_start;
677 gp->access = zvol_geom_access;
678 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
679 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
680 pp->sectorsize = DEV_BSIZE;
681 pp->mediasize = zv->zv_volsize;
684 zv->zv_provider = pp;
685 bioq_init(&zv->zv_queue);
686 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
687 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
688 struct make_dev_args args;
690 make_dev_args_init(&args);
691 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
692 args.mda_devsw = &zvol_cdevsw;
694 args.mda_uid = UID_ROOT;
695 args.mda_gid = GID_OPERATOR;
696 args.mda_mode = 0640;
697 args.mda_si_drv2 = zv;
698 error = make_dev_s(&args, &zv->zv_dev,
699 "%s/%s", ZVOL_DRIVER, name);
701 kmem_free(zv, sizeof(*zv));
702 dmu_objset_disown(os, FTAG);
703 mutex_exit(&zfsdev_state_lock);
706 zv->zv_dev->si_iosize_max = MAXPHYS;
708 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
711 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
712 zv->zv_min_bs = DEV_BSHIFT;
714 zv->zv_minor = minor;
717 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
718 zv->zv_flags |= ZVOL_RDONLY;
719 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
720 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
721 sizeof (rl_t), offsetof(rl_t, r_node));
722 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
723 offsetof(zvol_extent_t, ze_node));
724 /* get and cache the blocksize */
725 error = dmu_object_info(os, ZVOL_OBJ, &doi);
727 zv->zv_volblocksize = doi.doi_data_block_size;
729 if (spa_writeable(dmu_objset_spa(os))) {
730 if (zil_replay_disable)
731 zil_destroy(dmu_objset_zil(os), B_FALSE);
733 zil_replay(os, zv, zvol_replay_vector);
735 dmu_objset_disown(os, FTAG);
736 zv->zv_objset = NULL;
740 mutex_exit(&zfsdev_state_lock);
742 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
748 ZFS_LOG(1, "ZVOL %s created.", name);
755 * Remove minor node for the specified volume.
758 zvol_remove_zv(zvol_state_t *zv)
762 minor_t minor = zv->zv_minor;
765 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
766 if (zv->zv_total_opens != 0)
767 return (SET_ERROR(EBUSY));
770 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
771 ddi_remove_minor_node(zfs_dip, nmbuf);
773 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
774 ddi_remove_minor_node(zfs_dip, nmbuf);
776 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
778 LIST_REMOVE(zv, zv_links);
779 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
781 zvol_geom_destroy(zv);
783 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
784 if (zv->zv_dev != NULL)
785 destroy_dev(zv->zv_dev);
789 avl_destroy(&zv->zv_znode.z_range_avl);
790 mutex_destroy(&zv->zv_znode.z_range_lock);
792 kmem_free(zv, sizeof (zvol_state_t));
794 ddi_soft_state_free(zfsdev_state, minor);
801 zvol_remove_minor(const char *name)
806 mutex_enter(&zfsdev_state_lock);
807 if ((zv = zvol_minor_lookup(name)) == NULL) {
808 mutex_exit(&zfsdev_state_lock);
809 return (SET_ERROR(ENXIO));
811 rc = zvol_remove_zv(zv);
812 mutex_exit(&zfsdev_state_lock);
817 zvol_first_open(zvol_state_t *zv)
824 /* lie and say we're read-only */
825 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
831 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
834 dmu_objset_disown(os, zvol_tag);
838 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
840 dmu_objset_disown(os, zvol_tag);
844 zvol_size_changed(zv, volsize);
845 zv->zv_zilog = zil_open(os, zvol_get_data);
847 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
849 if (readonly || dmu_objset_is_snapshot(os) ||
850 !spa_writeable(dmu_objset_spa(os)))
851 zv->zv_flags |= ZVOL_RDONLY;
853 zv->zv_flags &= ~ZVOL_RDONLY;
858 zvol_last_close(zvol_state_t *zv)
860 zil_close(zv->zv_zilog);
863 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
869 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
870 !(zv->zv_flags & ZVOL_RDONLY))
871 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
872 dmu_objset_evict_dbufs(zv->zv_objset);
874 dmu_objset_disown(zv->zv_objset, zvol_tag);
875 zv->zv_objset = NULL;
880 zvol_prealloc(zvol_state_t *zv)
882 objset_t *os = zv->zv_objset;
884 uint64_t refd, avail, usedobjs, availobjs;
885 uint64_t resid = zv->zv_volsize;
888 /* Check the space usage before attempting to allocate the space */
889 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
890 if (avail < zv->zv_volsize)
891 return (SET_ERROR(ENOSPC));
893 /* Free old extents if they exist */
894 zvol_free_extents(zv);
898 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
900 tx = dmu_tx_create(os);
901 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
902 error = dmu_tx_assign(tx, TXG_WAIT);
905 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
908 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
913 txg_wait_synced(dmu_objset_pool(os), 0);
920 zvol_update_volsize(objset_t *os, uint64_t volsize)
925 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
927 tx = dmu_tx_create(os);
928 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
929 dmu_tx_mark_netfree(tx);
930 error = dmu_tx_assign(tx, TXG_WAIT);
936 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
941 error = dmu_free_long_range(os,
942 ZVOL_OBJ, volsize, DMU_OBJECT_END);
947 zvol_remove_minors(const char *name)
954 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
955 (void) strncpy(namebuf, name, strlen(name));
956 (void) strcat(namebuf, "/");
957 mutex_enter(&zfsdev_state_lock);
958 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
960 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
963 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
964 (void) zvol_remove_zv(zv);
966 kmem_free(namebuf, strlen(name) + 2);
968 mutex_exit(&zfsdev_state_lock);
970 zvol_state_t *zv, *tzv;
973 namelen = strlen(name);
976 mutex_enter(&zfsdev_state_lock);
978 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
979 if (strcmp(zv->zv_name, name) == 0 ||
980 (strncmp(zv->zv_name, name, namelen) == 0 &&
981 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
982 zv->zv_name[namelen] == '@'))) {
983 (void) zvol_remove_zv(zv);
987 mutex_exit(&zfsdev_state_lock);
993 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
995 uint64_t old_volsize = 0ULL;
998 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1001 * Reinitialize the dump area to the new size. If we
1002 * failed to resize the dump area then restore it back to
1003 * its original size. We must set the new volsize prior
1004 * to calling dumpvp_resize() to ensure that the devices'
1005 * size(9P) is not visible by the dump subsystem.
1007 old_volsize = zv->zv_volsize;
1008 zvol_size_changed(zv, volsize);
1011 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1012 if ((error = zvol_dumpify(zv)) != 0 ||
1013 (error = dumpvp_resize()) != 0) {
1016 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1017 zvol_size_changed(zv, old_volsize);
1018 dumpify_error = zvol_dumpify(zv);
1019 error = dumpify_error ? dumpify_error : error;
1022 #endif /* ZVOL_DUMP */
1026 * Generate a LUN expansion event.
1031 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1033 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1036 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1037 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1039 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1040 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1043 kmem_free(physpath, MAXPATHLEN);
1045 #endif /* illumos */
1050 zvol_set_volsize(const char *name, uint64_t volsize)
1052 zvol_state_t *zv = NULL;
1055 dmu_object_info_t doi;
1057 boolean_t owned = B_FALSE;
1059 error = dsl_prop_get_integer(name,
1060 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1064 return (SET_ERROR(EROFS));
1066 mutex_enter(&zfsdev_state_lock);
1067 zv = zvol_minor_lookup(name);
1069 if (zv == NULL || zv->zv_objset == NULL) {
1070 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1072 mutex_exit(&zfsdev_state_lock);
1082 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1083 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1086 error = zvol_update_volsize(os, volsize);
1088 if (error == 0 && zv != NULL)
1089 error = zvol_update_live_volsize(zv, volsize);
1092 dmu_objset_disown(os, FTAG);
1094 zv->zv_objset = NULL;
1096 mutex_exit(&zfsdev_state_lock);
1103 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1106 zvol_open(struct g_provider *pp, int flag, int count)
1113 mutex_enter(&zfsdev_state_lock);
1115 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1117 mutex_exit(&zfsdev_state_lock);
1118 return (SET_ERROR(ENXIO));
1121 if (zv->zv_total_opens == 0)
1122 err = zvol_first_open(zv);
1124 mutex_exit(&zfsdev_state_lock);
1127 #else /* !illumos */
1128 if (tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1130 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1131 * attempting to probe geom providers while looking for a
1132 * replacement for a missing VDEV. In this case, the
1133 * spa_namespace_lock will not be held, but it is still illegal
1134 * to use a zvol as a vdev. Deadlocks can result if another
1135 * thread has spa_namespace_lock
1137 return (EOPNOTSUPP);
1140 mutex_enter(&zfsdev_state_lock);
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 pp->mediasize = zv->zv_volsize;
1155 pp->stripeoffset = 0;
1156 pp->stripesize = zv->zv_volblocksize;
1158 #endif /* illumos */
1159 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1160 err = SET_ERROR(EROFS);
1163 if (zv->zv_flags & ZVOL_EXCL) {
1164 err = SET_ERROR(EBUSY);
1169 if (zv->zv_total_opens != 0) {
1170 err = SET_ERROR(EBUSY);
1173 zv->zv_flags |= ZVOL_EXCL;
1178 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1179 zv->zv_open_count[otyp]++;
1180 zv->zv_total_opens++;
1182 mutex_exit(&zfsdev_state_lock);
1184 zv->zv_total_opens += count;
1185 mutex_exit(&zfsdev_state_lock);
1190 if (zv->zv_total_opens == 0)
1191 zvol_last_close(zv);
1193 mutex_exit(&zfsdev_state_lock);
1195 mutex_exit(&zfsdev_state_lock);
1203 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1205 minor_t minor = getminor(dev);
1209 mutex_enter(&zfsdev_state_lock);
1211 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1213 mutex_exit(&zfsdev_state_lock);
1214 #else /* !illumos */
1216 zvol_close(struct g_provider *pp, int flag, int count)
1220 boolean_t locked = B_FALSE;
1222 /* See comment in zvol_open(). */
1223 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1224 mutex_enter(&zfsdev_state_lock);
1231 mutex_exit(&zfsdev_state_lock);
1232 #endif /* illumos */
1233 return (SET_ERROR(ENXIO));
1236 if (zv->zv_flags & ZVOL_EXCL) {
1237 ASSERT(zv->zv_total_opens == 1);
1238 zv->zv_flags &= ~ZVOL_EXCL;
1242 * If the open count is zero, this is a spurious close.
1243 * That indicates a bug in the kernel / DDI framework.
1246 ASSERT(zv->zv_open_count[otyp] != 0);
1248 ASSERT(zv->zv_total_opens != 0);
1251 * You may get multiple opens, but only one close.
1254 zv->zv_open_count[otyp]--;
1255 zv->zv_total_opens--;
1257 zv->zv_total_opens -= count;
1260 if (zv->zv_total_opens == 0)
1261 zvol_last_close(zv);
1264 mutex_exit(&zfsdev_state_lock);
1267 mutex_exit(&zfsdev_state_lock);
1273 zvol_get_done(zgd_t *zgd, int error)
1276 dmu_buf_rele(zgd->zgd_db, zgd);
1278 zfs_range_unlock(zgd->zgd_rl);
1280 if (error == 0 && zgd->zgd_bp)
1281 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1283 kmem_free(zgd, sizeof (zgd_t));
1287 * Get data to generate a TX_WRITE intent log record.
1290 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1292 zvol_state_t *zv = arg;
1293 objset_t *os = zv->zv_objset;
1294 uint64_t object = ZVOL_OBJ;
1295 uint64_t offset = lr->lr_offset;
1296 uint64_t size = lr->lr_length; /* length of user data */
1297 blkptr_t *bp = &lr->lr_blkptr;
1302 ASSERT(zio != NULL);
1305 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1306 zgd->zgd_zilog = zv->zv_zilog;
1307 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1310 * Write records come in two flavors: immediate and indirect.
1311 * For small writes it's cheaper to store the data with the
1312 * log record (immediate); for large writes it's cheaper to
1313 * sync the data and get a pointer to it (indirect) so that
1314 * we don't have to write the data twice.
1316 if (buf != NULL) { /* immediate write */
1317 error = dmu_read(os, object, offset, size, buf,
1318 DMU_READ_NO_PREFETCH);
1320 size = zv->zv_volblocksize;
1321 offset = P2ALIGN(offset, size);
1322 error = dmu_buf_hold(os, object, offset, zgd, &db,
1323 DMU_READ_NO_PREFETCH);
1325 blkptr_t *obp = dmu_buf_get_blkptr(db);
1327 ASSERT(BP_IS_HOLE(bp));
1334 ASSERT(db->db_offset == offset);
1335 ASSERT(db->db_size == size);
1337 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1338 zvol_get_done, zgd);
1345 zvol_get_done(zgd, error);
1351 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1353 * We store data in the log buffers if it's small enough.
1354 * Otherwise we will later flush the data out via dmu_sync().
1356 ssize_t zvol_immediate_write_sz = 32768;
1359 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1362 uint32_t blocksize = zv->zv_volblocksize;
1363 zilog_t *zilog = zv->zv_zilog;
1365 ssize_t immediate_write_sz;
1367 if (zil_replaying(zilog, tx))
1370 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1371 ? 0 : zvol_immediate_write_sz;
1373 slogging = spa_has_slogs(zilog->zl_spa) &&
1374 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1380 itx_wr_state_t write_state;
1383 * Unlike zfs_log_write() we can be called with
1384 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1386 if (blocksize > immediate_write_sz && !slogging &&
1387 resid >= blocksize && off % blocksize == 0) {
1388 write_state = WR_INDIRECT; /* uses dmu_sync */
1391 write_state = WR_COPIED;
1392 len = MIN(ZIL_MAX_LOG_DATA, resid);
1394 write_state = WR_NEED_COPY;
1395 len = MIN(ZIL_MAX_LOG_DATA, resid);
1398 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1399 (write_state == WR_COPIED ? len : 0));
1400 lr = (lr_write_t *)&itx->itx_lr;
1401 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1402 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1403 zil_itx_destroy(itx);
1404 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1405 lr = (lr_write_t *)&itx->itx_lr;
1406 write_state = WR_NEED_COPY;
1409 itx->itx_wr_state = write_state;
1410 if (write_state == WR_NEED_COPY)
1411 itx->itx_sod += len;
1412 lr->lr_foid = ZVOL_OBJ;
1413 lr->lr_offset = off;
1414 lr->lr_length = len;
1416 BP_ZERO(&lr->lr_blkptr);
1418 itx->itx_private = zv;
1419 itx->itx_sync = sync;
1421 zil_itx_assign(zilog, itx, tx);
1430 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1431 uint64_t size, boolean_t doread, boolean_t isdump)
1437 if (vd->vdev_ops == &vdev_mirror_ops ||
1438 vd->vdev_ops == &vdev_replacing_ops ||
1439 vd->vdev_ops == &vdev_spare_ops) {
1440 for (c = 0; c < vd->vdev_children; c++) {
1441 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1442 addr, offset, origoffset, size, doread, isdump);
1445 } else if (doread) {
1451 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1452 return (numerrors < vd->vdev_children ? 0 : EIO);
1454 if (doread && !vdev_readable(vd))
1455 return (SET_ERROR(EIO));
1456 else if (!doread && !vdev_writeable(vd))
1457 return (SET_ERROR(EIO));
1459 if (vd->vdev_ops == &vdev_raidz_ops) {
1460 return (vdev_raidz_physio(vd,
1461 addr, size, offset, origoffset, doread, isdump));
1464 offset += VDEV_LABEL_START_SIZE;
1466 if (ddi_in_panic() || isdump) {
1469 return (SET_ERROR(EIO));
1471 ASSERT3P(dvd, !=, NULL);
1472 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1476 ASSERT3P(dvd, !=, NULL);
1477 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1478 offset, doread ? B_READ : B_WRITE));
1483 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1484 boolean_t doread, boolean_t isdump)
1489 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1491 /* Must be sector aligned, and not stradle a block boundary. */
1492 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1493 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1494 return (SET_ERROR(EINVAL));
1496 ASSERT(size <= zv->zv_volblocksize);
1498 /* Locate the extent this belongs to */
1499 ze = list_head(&zv->zv_extents);
1500 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1501 offset -= ze->ze_nblks * zv->zv_volblocksize;
1502 ze = list_next(&zv->zv_extents, ze);
1506 return (SET_ERROR(EINVAL));
1508 if (!ddi_in_panic())
1509 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1511 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1512 offset += DVA_GET_OFFSET(&ze->ze_dva);
1513 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1514 size, doread, isdump);
1516 if (!ddi_in_panic())
1517 spa_config_exit(spa, SCL_STATE, FTAG);
1523 zvol_strategy(buf_t *bp)
1525 zfs_soft_state_t *zs = NULL;
1526 #else /* !illumos */
1528 zvol_strategy(struct bio *bp)
1530 #endif /* illumos */
1532 uint64_t off, volsize;
1539 boolean_t doread = bp->b_flags & B_READ;
1541 boolean_t doread = 0;
1543 boolean_t is_dumpified;
1547 if (getminor(bp->b_edev) == 0) {
1548 error = SET_ERROR(EINVAL);
1550 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1552 error = SET_ERROR(ENXIO);
1553 else if (zs->zss_type != ZSST_ZVOL)
1554 error = SET_ERROR(EINVAL);
1558 bioerror(bp, error);
1565 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1566 bioerror(bp, EROFS);
1571 off = ldbtob(bp->b_blkno);
1572 #else /* !illumos */
1574 zv = bp->bio_to->private;
1576 zv = bp->bio_dev->si_drv2;
1579 error = SET_ERROR(ENXIO);
1583 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1584 error = SET_ERROR(EROFS);
1588 switch (bp->bio_cmd) {
1601 off = bp->bio_offset;
1602 #endif /* illumos */
1603 volsize = zv->zv_volsize;
1610 addr = bp->b_un.b_addr;
1611 resid = bp->b_bcount;
1613 if (resid > 0 && (off < 0 || off >= volsize)) {
1619 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1620 sync = ((!(bp->b_flags & B_ASYNC) &&
1621 !(zv->zv_flags & ZVOL_WCE)) ||
1622 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1623 !doread && !is_dumpified;
1624 #else /* !illumos */
1625 addr = bp->bio_data;
1626 resid = bp->bio_length;
1628 if (resid > 0 && (off < 0 || off >= volsize)) {
1629 error = SET_ERROR(EIO);
1633 is_dumpified = B_FALSE;
1634 sync = !doread && !is_dumpified &&
1635 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1636 #endif /* illumos */
1639 * There must be no buffer changes when doing a dmu_sync() because
1640 * we can't change the data whilst calculating the checksum.
1642 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1643 doread ? RL_READER : RL_WRITER);
1646 if (bp->bio_cmd == BIO_DELETE) {
1647 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1648 error = dmu_tx_assign(tx, TXG_WAIT);
1652 zvol_log_truncate(zv, tx, off, resid, B_TRUE);
1654 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1661 while (resid != 0 && off < volsize) {
1662 size_t size = MIN(resid, zvol_maxphys);
1665 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1666 error = zvol_dumpio(zv, addr, off, size,
1668 } else if (doread) {
1672 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1675 dmu_tx_t *tx = dmu_tx_create(os);
1676 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1677 error = dmu_tx_assign(tx, TXG_WAIT);
1681 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1682 zvol_log_write(zv, tx, off, size, sync);
1687 /* convert checksum errors into IO errors */
1688 if (error == ECKSUM)
1689 error = SET_ERROR(EIO);
1699 zfs_range_unlock(rl);
1702 if ((bp->b_resid = resid) == bp->b_bcount)
1703 bioerror(bp, off > volsize ? EINVAL : error);
1706 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1710 #else /* !illumos */
1711 bp->bio_completed = bp->bio_length - resid;
1712 if (bp->bio_completed < bp->bio_length && off > volsize)
1717 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1721 g_io_deliver(bp, error);
1723 biofinish(bp, NULL, error);
1724 #endif /* illumos */
1729 * Set the buffer count to the zvol maximum transfer.
1730 * Using our own routine instead of the default minphys()
1731 * means that for larger writes we write bigger buffers on X86
1732 * (128K instead of 56K) and flush the disk write cache less often
1733 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1734 * 56K on X86 and 128K on sparc).
1737 zvol_minphys(struct buf *bp)
1739 if (bp->b_bcount > zvol_maxphys)
1740 bp->b_bcount = zvol_maxphys;
1744 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1746 minor_t minor = getminor(dev);
1753 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1755 return (SET_ERROR(ENXIO));
1757 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1758 return (SET_ERROR(EINVAL));
1760 boff = ldbtob(blkno);
1761 resid = ldbtob(nblocks);
1763 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1766 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1767 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1780 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1782 minor_t minor = getminor(dev);
1783 #else /* !illumos */
1785 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1787 #endif /* illumos */
1794 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1796 return (SET_ERROR(ENXIO));
1801 volsize = zv->zv_volsize;
1802 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1803 if (uio->uio_resid > 0 &&
1804 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1805 return (SET_ERROR(EIO));
1808 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1809 error = physio(zvol_strategy, NULL, dev, B_READ,
1815 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1817 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1818 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1820 /* don't read past the end */
1821 if (bytes > volsize - uio->uio_loffset)
1822 bytes = volsize - uio->uio_loffset;
1824 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
1826 /* convert checksum errors into IO errors */
1827 if (error == ECKSUM)
1828 error = SET_ERROR(EIO);
1832 zfs_range_unlock(rl);
1839 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1841 minor_t minor = getminor(dev);
1842 #else /* !illumos */
1844 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1846 #endif /* illumos */
1854 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1856 return (SET_ERROR(ENXIO));
1861 volsize = zv->zv_volsize;
1862 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1863 if (uio->uio_resid > 0 &&
1864 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1865 return (SET_ERROR(EIO));
1868 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1869 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1874 sync = !(zv->zv_flags & ZVOL_WCE) ||
1876 sync = (ioflag & IO_SYNC) ||
1878 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1880 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1882 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1883 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1884 uint64_t off = uio->uio_loffset;
1885 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1887 if (bytes > volsize - off) /* don't write past the end */
1888 bytes = volsize - off;
1890 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1891 error = dmu_tx_assign(tx, TXG_WAIT);
1896 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1898 zvol_log_write(zv, tx, off, bytes, sync);
1904 zfs_range_unlock(rl);
1906 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1912 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1914 struct uuid uuid = EFI_RESERVED;
1915 efi_gpe_t gpe = { 0 };
1921 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1922 return (SET_ERROR(EFAULT));
1923 ptr = (char *)(uintptr_t)efi.dki_data_64;
1924 length = efi.dki_length;
1926 * Some clients may attempt to request a PMBR for the
1927 * zvol. Currently this interface will return EINVAL to
1928 * such requests. These requests could be supported by
1929 * adding a check for lba == 0 and consing up an appropriate
1932 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1933 return (SET_ERROR(EINVAL));
1935 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1936 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1937 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1939 if (efi.dki_lba == 1) {
1940 efi_gpt_t gpt = { 0 };
1942 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1943 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1944 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1945 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1946 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1947 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1948 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1949 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1950 gpt.efi_gpt_SizeOfPartitionEntry =
1951 LE_32(sizeof (efi_gpe_t));
1952 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1953 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1954 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1955 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1956 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1958 return (SET_ERROR(EFAULT));
1959 ptr += sizeof (gpt);
1960 length -= sizeof (gpt);
1962 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1964 return (SET_ERROR(EFAULT));
1969 * BEGIN entry points to allow external callers access to the volume.
1972 * Return the volume parameters needed for access from an external caller.
1973 * These values are invariant as long as the volume is held open.
1976 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1977 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1978 void **rl_hdl, void **bonus_hdl)
1982 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1984 return (SET_ERROR(ENXIO));
1985 if (zv->zv_flags & ZVOL_DUMPIFIED)
1986 return (SET_ERROR(ENXIO));
1988 ASSERT(blksize && max_xfer_len && minor_hdl &&
1989 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1991 *blksize = zv->zv_volblocksize;
1992 *max_xfer_len = (uint64_t)zvol_maxphys;
1994 *objset_hdl = zv->zv_objset;
1995 *zil_hdl = zv->zv_zilog;
1996 *rl_hdl = &zv->zv_znode;
1997 *bonus_hdl = zv->zv_dbuf;
2002 * Return the current volume size to an external caller.
2003 * The size can change while the volume is open.
2006 zvol_get_volume_size(void *minor_hdl)
2008 zvol_state_t *zv = minor_hdl;
2010 return (zv->zv_volsize);
2014 * Return the current WCE setting to an external caller.
2015 * The WCE setting can change while the volume is open.
2018 zvol_get_volume_wce(void *minor_hdl)
2020 zvol_state_t *zv = minor_hdl;
2022 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2026 * Entry point for external callers to zvol_log_write
2029 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2032 zvol_state_t *zv = minor_hdl;
2034 zvol_log_write(zv, tx, off, resid, sync);
2037 * END entry points to allow external callers access to the volume.
2039 #endif /* illumos */
2042 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2045 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2050 zilog_t *zilog = zv->zv_zilog;
2052 if (zil_replaying(zilog, tx))
2055 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2056 lr = (lr_truncate_t *)&itx->itx_lr;
2057 lr->lr_foid = ZVOL_OBJ;
2058 lr->lr_offset = off;
2059 lr->lr_length = len;
2061 itx->itx_sync = sync;
2062 zil_itx_assign(zilog, itx, tx);
2067 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2068 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2072 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2075 struct dk_callback *dkc;
2079 mutex_enter(&zfsdev_state_lock);
2081 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2084 mutex_exit(&zfsdev_state_lock);
2085 return (SET_ERROR(ENXIO));
2087 ASSERT(zv->zv_total_opens > 0);
2093 struct dk_cinfo dki;
2095 bzero(&dki, sizeof (dki));
2096 (void) strcpy(dki.dki_cname, "zvol");
2097 (void) strcpy(dki.dki_dname, "zvol");
2098 dki.dki_ctype = DKC_UNKNOWN;
2099 dki.dki_unit = getminor(dev);
2100 dki.dki_maxtransfer =
2101 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2102 mutex_exit(&zfsdev_state_lock);
2103 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2104 error = SET_ERROR(EFAULT);
2108 case DKIOCGMEDIAINFO:
2110 struct dk_minfo dkm;
2112 bzero(&dkm, sizeof (dkm));
2113 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2114 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2115 dkm.dki_media_type = DK_UNKNOWN;
2116 mutex_exit(&zfsdev_state_lock);
2117 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2118 error = SET_ERROR(EFAULT);
2122 case DKIOCGMEDIAINFOEXT:
2124 struct dk_minfo_ext dkmext;
2126 bzero(&dkmext, sizeof (dkmext));
2127 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2128 dkmext.dki_pbsize = zv->zv_volblocksize;
2129 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2130 dkmext.dki_media_type = DK_UNKNOWN;
2131 mutex_exit(&zfsdev_state_lock);
2132 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2133 error = SET_ERROR(EFAULT);
2139 uint64_t vs = zv->zv_volsize;
2140 uint8_t bs = zv->zv_min_bs;
2142 mutex_exit(&zfsdev_state_lock);
2143 error = zvol_getefi((void *)arg, flag, vs, bs);
2147 case DKIOCFLUSHWRITECACHE:
2148 dkc = (struct dk_callback *)arg;
2149 mutex_exit(&zfsdev_state_lock);
2150 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2151 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2152 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2159 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2160 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2162 error = SET_ERROR(EFAULT);
2168 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2170 error = SET_ERROR(EFAULT);
2174 zv->zv_flags |= ZVOL_WCE;
2175 mutex_exit(&zfsdev_state_lock);
2177 zv->zv_flags &= ~ZVOL_WCE;
2178 mutex_exit(&zfsdev_state_lock);
2179 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2187 * commands using these (like prtvtoc) expect ENOTSUP
2188 * since we're emulating an EFI label
2190 error = SET_ERROR(ENOTSUP);
2194 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2196 error = zvol_dumpify(zv);
2197 zfs_range_unlock(rl);
2201 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2203 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2205 error = zvol_dump_fini(zv);
2206 zfs_range_unlock(rl);
2214 if (!zvol_unmap_enabled)
2217 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2218 error = SET_ERROR(EFAULT);
2223 * Apply Postel's Law to length-checking. If they overshoot,
2224 * just blank out until the end, if there's a need to blank
2227 if (df.df_start >= zv->zv_volsize)
2228 break; /* No need to do anything... */
2230 mutex_exit(&zfsdev_state_lock);
2232 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2234 tx = dmu_tx_create(zv->zv_objset);
2235 dmu_tx_mark_netfree(tx);
2236 error = dmu_tx_assign(tx, TXG_WAIT);
2240 zvol_log_truncate(zv, tx, df.df_start,
2241 df.df_length, B_TRUE);
2243 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2244 df.df_start, df.df_length);
2247 zfs_range_unlock(rl);
2251 * If the write-cache is disabled or 'sync' property
2252 * is set to 'always' then treat this as a synchronous
2253 * operation (i.e. commit to zil).
2255 if (!(zv->zv_flags & ZVOL_WCE) ||
2256 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2257 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2260 * If the caller really wants synchronous writes, and
2261 * can't wait for them, don't return until the write
2264 if (df.df_flags & DF_WAIT_SYNC) {
2266 dmu_objset_pool(zv->zv_objset), 0);
2273 error = SET_ERROR(ENOTTY);
2277 mutex_exit(&zfsdev_state_lock);
2280 #endif /* illumos */
2285 return (zvol_minors != 0);
2291 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2294 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2296 ZFS_LOG(1, "ZVOL Initialized.");
2304 mutex_destroy(&zfsdev_state_lock);
2306 ddi_soft_state_fini(&zfsdev_state);
2307 ZFS_LOG(1, "ZVOL Deinitialized.");
2313 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2315 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2317 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2324 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2326 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2328 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2332 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2336 objset_t *os = zv->zv_objset;
2337 spa_t *spa = dmu_objset_spa(os);
2338 vdev_t *vd = spa->spa_root_vdev;
2339 nvlist_t *nv = NULL;
2340 uint64_t version = spa_version(spa);
2341 uint64_t checksum, compress, refresrv, vbs, dedup;
2343 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2344 ASSERT(vd->vdev_ops == &vdev_root_ops);
2346 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2350 /* wait for dmu_free_long_range to actually free the blocks */
2351 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2354 * If the pool on which the dump device is being initialized has more
2355 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2356 * enabled. If so, bump that feature's counter to indicate that the
2357 * feature is active. We also check the vdev type to handle the
2359 * # zpool create test raidz disk1 disk2 disk3
2360 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2361 * the raidz vdev itself has 3 children.
2363 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2364 if (!spa_feature_is_enabled(spa,
2365 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2366 return (SET_ERROR(ENOTSUP));
2367 (void) dsl_sync_task(spa_name(spa),
2368 zfs_mvdev_dump_feature_check,
2369 zfs_mvdev_dump_activate_feature_sync, NULL,
2370 2, ZFS_SPACE_CHECK_RESERVED);
2374 error = dsl_prop_get_integer(zv->zv_name,
2375 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2377 error = dsl_prop_get_integer(zv->zv_name,
2378 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2382 error = dsl_prop_get_integer(zv->zv_name,
2383 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2387 error = dsl_prop_get_integer(zv->zv_name,
2388 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2391 if (version >= SPA_VERSION_DEDUP && error == 0) {
2392 error = dsl_prop_get_integer(zv->zv_name,
2393 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2399 tx = dmu_tx_create(os);
2400 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2401 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2402 error = dmu_tx_assign(tx, TXG_WAIT);
2409 * If we are resizing the dump device then we only need to
2410 * update the refreservation to match the newly updated
2411 * zvolsize. Otherwise, we save off the original state of the
2412 * zvol so that we can restore them if the zvol is ever undumpified.
2415 error = zap_update(os, ZVOL_ZAP_OBJ,
2416 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2417 &zv->zv_volsize, tx);
2419 error = zap_update(os, ZVOL_ZAP_OBJ,
2420 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2423 error = zap_update(os, ZVOL_ZAP_OBJ,
2424 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2428 error = zap_update(os, ZVOL_ZAP_OBJ,
2429 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2433 error = zap_update(os, ZVOL_ZAP_OBJ,
2434 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2438 error = dmu_object_set_blocksize(
2439 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2441 if (version >= SPA_VERSION_DEDUP && error == 0) {
2442 error = zap_update(os, ZVOL_ZAP_OBJ,
2443 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2447 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2452 * We only need update the zvol's property if we are initializing
2453 * the dump area for the first time.
2455 if (error == 0 && !resize) {
2457 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2458 * function. Otherwise, use the old default -- OFF.
2460 checksum = spa_feature_is_active(spa,
2461 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2464 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2465 VERIFY(nvlist_add_uint64(nv,
2466 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2467 VERIFY(nvlist_add_uint64(nv,
2468 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2469 ZIO_COMPRESS_OFF) == 0);
2470 VERIFY(nvlist_add_uint64(nv,
2471 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2473 if (version >= SPA_VERSION_DEDUP) {
2474 VERIFY(nvlist_add_uint64(nv,
2475 zfs_prop_to_name(ZFS_PROP_DEDUP),
2476 ZIO_CHECKSUM_OFF) == 0);
2479 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2484 /* Allocate the space for the dump */
2486 error = zvol_prealloc(zv);
2491 zvol_dumpify(zvol_state_t *zv)
2494 uint64_t dumpsize = 0;
2496 objset_t *os = zv->zv_objset;
2498 if (zv->zv_flags & ZVOL_RDONLY)
2499 return (SET_ERROR(EROFS));
2501 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2502 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2503 boolean_t resize = (dumpsize > 0);
2505 if ((error = zvol_dump_init(zv, resize)) != 0) {
2506 (void) zvol_dump_fini(zv);
2512 * Build up our lba mapping.
2514 error = zvol_get_lbas(zv);
2516 (void) zvol_dump_fini(zv);
2520 tx = dmu_tx_create(os);
2521 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2522 error = dmu_tx_assign(tx, TXG_WAIT);
2525 (void) zvol_dump_fini(zv);
2529 zv->zv_flags |= ZVOL_DUMPIFIED;
2530 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2531 &zv->zv_volsize, tx);
2535 (void) zvol_dump_fini(zv);
2539 txg_wait_synced(dmu_objset_pool(os), 0);
2544 zvol_dump_fini(zvol_state_t *zv)
2547 objset_t *os = zv->zv_objset;
2550 uint64_t checksum, compress, refresrv, vbs, dedup;
2551 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2554 * Attempt to restore the zvol back to its pre-dumpified state.
2555 * This is a best-effort attempt as it's possible that not all
2556 * of these properties were initialized during the dumpify process
2557 * (i.e. error during zvol_dump_init).
2560 tx = dmu_tx_create(os);
2561 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2562 error = dmu_tx_assign(tx, TXG_WAIT);
2567 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2570 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2571 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2572 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2573 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2574 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2575 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2576 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2577 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2579 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2580 (void) nvlist_add_uint64(nv,
2581 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2582 (void) nvlist_add_uint64(nv,
2583 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2584 (void) nvlist_add_uint64(nv,
2585 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2586 if (version >= SPA_VERSION_DEDUP &&
2587 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2588 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2589 (void) nvlist_add_uint64(nv,
2590 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2592 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2596 zvol_free_extents(zv);
2597 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2598 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2599 /* wait for dmu_free_long_range to actually free the blocks */
2600 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2601 tx = dmu_tx_create(os);
2602 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2603 error = dmu_tx_assign(tx, TXG_WAIT);
2608 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2609 zv->zv_volblocksize = vbs;
2614 #else /* !illumos */
2617 zvol_geom_run(zvol_state_t *zv)
2619 struct g_provider *pp;
2621 pp = zv->zv_provider;
2622 g_error_provider(pp, 0);
2624 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2625 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2629 zvol_geom_destroy(zvol_state_t *zv)
2631 struct g_provider *pp;
2633 g_topology_assert();
2635 mtx_lock(&zv->zv_queue_mtx);
2637 wakeup_one(&zv->zv_queue);
2638 while (zv->zv_state != 2)
2639 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2640 mtx_destroy(&zv->zv_queue_mtx);
2642 pp = zv->zv_provider;
2643 zv->zv_provider = NULL;
2645 g_wither_geom(pp->geom, ENXIO);
2649 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2651 int count, error, flags;
2653 g_topology_assert();
2656 * To make it easier we expect either open or close, but not both
2659 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2660 (acr <= 0 && acw <= 0 && ace <= 0),
2661 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2662 pp->name, acr, acw, ace));
2664 if (pp->private == NULL) {
2665 if (acr <= 0 && acw <= 0 && ace <= 0)
2671 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2672 * because GEOM already handles that and handles it a bit differently.
2673 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2674 * only one exclusive consumer, no matter if it is reader or writer.
2675 * I like better the way GEOM works so I'll leave it for GEOM to
2676 * decide what to do.
2679 count = acr + acw + ace;
2684 if (acr != 0 || ace != 0)
2689 g_topology_unlock();
2691 error = zvol_open(pp, flags, count);
2693 error = zvol_close(pp, flags, -count);
2699 zvol_geom_start(struct bio *bp)
2704 zv = bp->bio_to->private;
2706 switch (bp->bio_cmd) {
2708 if (!THREAD_CAN_SLEEP())
2710 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2711 g_io_deliver(bp, 0);
2716 if (!THREAD_CAN_SLEEP())
2721 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2722 uint64_t refd, avail, usedobjs, availobjs, val;
2724 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2726 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2727 dmu_objset_space(zv->zv_objset, &refd, &avail,
2728 &usedobjs, &availobjs);
2729 if (g_handleattr_off_t(bp, "blocksavail",
2732 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2733 dmu_objset_space(zv->zv_objset, &refd, &avail,
2734 &usedobjs, &availobjs);
2735 if (g_handleattr_off_t(bp, "blocksused",
2738 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2739 avail = metaslab_class_get_space(spa_normal_class(spa));
2740 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2741 if (g_handleattr_off_t(bp, "poolblocksavail",
2744 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2745 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2746 if (g_handleattr_off_t(bp, "poolblocksused",
2753 g_io_deliver(bp, EOPNOTSUPP);
2759 mtx_lock(&zv->zv_queue_mtx);
2760 first = (bioq_first(&zv->zv_queue) == NULL);
2761 bioq_insert_tail(&zv->zv_queue, bp);
2762 mtx_unlock(&zv->zv_queue_mtx);
2764 wakeup_one(&zv->zv_queue);
2768 zvol_geom_worker(void *arg)
2773 thread_lock(curthread);
2774 sched_prio(curthread, PRIBIO);
2775 thread_unlock(curthread);
2779 mtx_lock(&zv->zv_queue_mtx);
2780 bp = bioq_takefirst(&zv->zv_queue);
2782 if (zv->zv_state == 1) {
2784 wakeup(&zv->zv_state);
2785 mtx_unlock(&zv->zv_queue_mtx);
2788 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2792 mtx_unlock(&zv->zv_queue_mtx);
2793 switch (bp->bio_cmd) {
2795 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2796 g_io_deliver(bp, 0);
2804 g_io_deliver(bp, EOPNOTSUPP);
2810 extern boolean_t dataset_name_hidden(const char *name);
2813 zvol_create_snapshots(objset_t *os, const char *name)
2815 uint64_t cookie, obj;
2820 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2823 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2828 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2829 if (len >= MAXPATHLEN) {
2830 dmu_objset_rele(os, FTAG);
2831 error = ENAMETOOLONG;
2835 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2836 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2837 sname + len, &obj, &cookie, NULL);
2838 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2840 if (error == ENOENT)
2845 error = zvol_create_minor(sname);
2846 if (error != 0 && error != EEXIST) {
2847 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2853 kmem_free(sname, MAXPATHLEN);
2858 zvol_create_minors(const char *name)
2865 if (dataset_name_hidden(name))
2868 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2869 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2873 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2874 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2875 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2876 error = zvol_create_minor(name);
2877 if (error == 0 || error == EEXIST) {
2878 error = zvol_create_snapshots(os, name);
2880 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2883 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2884 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2887 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2888 dmu_objset_rele(os, FTAG);
2892 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2893 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2894 dmu_objset_rele(os, FTAG);
2895 kmem_free(osname, MAXPATHLEN);
2898 p = osname + strlen(osname);
2899 len = MAXPATHLEN - (p - osname);
2902 /* Prefetch the datasets. */
2904 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2905 if (!dataset_name_hidden(osname))
2906 (void) dmu_objset_prefetch(osname, NULL);
2911 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2913 dmu_objset_rele(os, FTAG);
2914 (void)zvol_create_minors(osname);
2915 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2916 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2922 dmu_objset_rele(os, FTAG);
2923 kmem_free(osname, MAXPATHLEN);
2928 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2931 struct g_provider *pp;
2934 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2936 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2938 pp = zv->zv_provider;
2943 zv->zv_provider = NULL;
2944 g_wither_provider(pp, ENXIO);
2946 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2947 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2948 pp->sectorsize = DEV_BSIZE;
2949 pp->mediasize = zv->zv_volsize;
2951 zv->zv_provider = pp;
2952 g_error_provider(pp, 0);
2953 g_topology_unlock();
2954 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2955 struct make_dev_args args;
2957 if ((dev = zv->zv_dev) != NULL) {
2960 if (zv->zv_total_opens > 0) {
2961 zv->zv_flags &= ~ZVOL_EXCL;
2962 zv->zv_total_opens = 0;
2963 zvol_last_close(zv);
2967 make_dev_args_init(&args);
2968 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
2969 args.mda_devsw = &zvol_cdevsw;
2971 args.mda_uid = UID_ROOT;
2972 args.mda_gid = GID_OPERATOR;
2973 args.mda_mode = 0640;
2974 args.mda_si_drv2 = zv;
2975 if (make_dev_s(&args, &zv->zv_dev,
2976 "%s/%s", ZVOL_DRIVER, newname) == 0)
2977 zv->zv_dev->si_iosize_max = MAXPHYS;
2979 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
2983 zvol_rename_minors(const char *oldname, const char *newname)
2985 char name[MAXPATHLEN];
2986 struct g_provider *pp;
2988 size_t oldnamelen, newnamelen;
2991 boolean_t locked = B_FALSE;
2993 oldnamelen = strlen(oldname);
2994 newnamelen = strlen(newname);
2997 /* See comment in zvol_open(). */
2998 if (!MUTEX_HELD(&zfsdev_state_lock)) {
2999 mutex_enter(&zfsdev_state_lock);
3003 LIST_FOREACH(zv, &all_zvols, zv_links) {
3004 if (strcmp(zv->zv_name, oldname) == 0) {
3005 zvol_rename_minor(zv, newname);
3006 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3007 (zv->zv_name[oldnamelen] == '/' ||
3008 zv->zv_name[oldnamelen] == '@')) {
3009 snprintf(name, sizeof(name), "%s%c%s", newname,
3010 zv->zv_name[oldnamelen],
3011 zv->zv_name + oldnamelen + 1);
3012 zvol_rename_minor(zv, name);
3017 mutex_exit(&zfsdev_state_lock);
3022 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3024 zvol_state_t *zv = dev->si_drv2;
3027 mutex_enter(&zfsdev_state_lock);
3028 if (zv->zv_total_opens == 0)
3029 err = zvol_first_open(zv);
3031 mutex_exit(&zfsdev_state_lock);
3034 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3035 err = SET_ERROR(EROFS);
3038 if (zv->zv_flags & ZVOL_EXCL) {
3039 err = SET_ERROR(EBUSY);
3043 if (flags & FEXCL) {
3044 if (zv->zv_total_opens != 0) {
3045 err = SET_ERROR(EBUSY);
3048 zv->zv_flags |= ZVOL_EXCL;
3052 zv->zv_total_opens++;
3053 mutex_exit(&zfsdev_state_lock);
3056 if (zv->zv_total_opens == 0)
3057 zvol_last_close(zv);
3058 mutex_exit(&zfsdev_state_lock);
3063 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3065 zvol_state_t *zv = dev->si_drv2;
3067 mutex_enter(&zfsdev_state_lock);
3068 if (zv->zv_flags & ZVOL_EXCL) {
3069 ASSERT(zv->zv_total_opens == 1);
3070 zv->zv_flags &= ~ZVOL_EXCL;
3074 * If the open count is zero, this is a spurious close.
3075 * That indicates a bug in the kernel / DDI framework.
3077 ASSERT(zv->zv_total_opens != 0);
3080 * You may get multiple opens, but only one close.
3082 zv->zv_total_opens--;
3084 if (zv->zv_total_opens == 0)
3085 zvol_last_close(zv);
3087 mutex_exit(&zfsdev_state_lock);
3092 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3096 off_t offset, length, chunk;
3103 KASSERT(zv->zv_total_opens > 0,
3104 ("Device with zero access count in zvol_d_ioctl"));
3106 i = IOCPARM_LEN(cmd);
3108 case DIOCGSECTORSIZE:
3109 *(u_int *)data = DEV_BSIZE;
3111 case DIOCGMEDIASIZE:
3112 *(off_t *)data = zv->zv_volsize;
3115 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3118 if (!zvol_unmap_enabled)
3121 offset = ((off_t *)data)[0];
3122 length = ((off_t *)data)[1];
3123 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3124 offset < 0 || offset >= zv->zv_volsize ||
3126 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3132 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3133 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3134 error = dmu_tx_assign(tx, TXG_WAIT);
3138 zvol_log_truncate(zv, tx, offset, length, B_TRUE);
3140 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3143 zfs_range_unlock(rl);
3144 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
3145 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3147 case DIOCGSTRIPESIZE:
3148 *(off_t *)data = zv->zv_volblocksize;
3150 case DIOCGSTRIPEOFFSET:
3154 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3155 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3156 uint64_t refd, avail, usedobjs, availobjs;
3158 if (strcmp(arg->name, "GEOM::candelete") == 0)
3160 else if (strcmp(arg->name, "blocksavail") == 0) {
3161 dmu_objset_space(zv->zv_objset, &refd, &avail,
3162 &usedobjs, &availobjs);
3163 arg->value.off = avail / DEV_BSIZE;
3164 } else if (strcmp(arg->name, "blocksused") == 0) {
3165 dmu_objset_space(zv->zv_objset, &refd, &avail,
3166 &usedobjs, &availobjs);
3167 arg->value.off = refd / DEV_BSIZE;
3168 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3169 avail = metaslab_class_get_space(spa_normal_class(spa));
3170 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3171 arg->value.off = avail / DEV_BSIZE;
3172 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3173 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3174 arg->value.off = refd / DEV_BSIZE;
3181 off_t *off = (off_t *)data;
3185 hole = (cmd == FIOSEEKHOLE);
3187 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3197 #endif /* illumos */