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 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 zilog_t *zv_zilog; /* ZIL handle */
173 list_t zv_extents; /* List of extents for dump */
174 znode_t zv_znode; /* for range locking */
175 dmu_buf_t *zv_dbuf; /* bonus handle */
178 int zv_volmode; /* Provide GEOM or cdev */
179 struct bio_queue_head zv_queue;
180 struct mtx zv_queue_mtx; /* zv_queue mutex */
185 static LIST_HEAD(, zvol_state) all_zvols;
188 * zvol specific flags
190 #define ZVOL_RDONLY 0x1
191 #define ZVOL_DUMPIFIED 0x2
192 #define ZVOL_EXCL 0x4
196 * zvol maximum transfer in one DMU tx.
198 int zvol_maxphys = DMU_MAX_ACCESS/2;
201 * Toggle unmap functionality.
203 boolean_t zvol_unmap_enabled = B_TRUE;
205 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
206 &zvol_unmap_enabled, 0,
207 "Enable UNMAP functionality");
209 static d_open_t zvol_d_open;
210 static d_close_t zvol_d_close;
211 static d_read_t zvol_read;
212 static d_write_t zvol_write;
213 static d_ioctl_t zvol_d_ioctl;
214 static d_strategy_t zvol_strategy;
216 static struct cdevsw zvol_cdevsw = {
217 .d_version = D_VERSION,
218 .d_open = zvol_d_open,
219 .d_close = zvol_d_close,
221 .d_write = zvol_write,
222 .d_ioctl = zvol_d_ioctl,
223 .d_strategy = zvol_strategy,
225 .d_flags = D_DISK | D_TRACKCLOSE,
228 static void zvol_geom_run(zvol_state_t *zv);
229 static void zvol_geom_destroy(zvol_state_t *zv);
230 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
231 static void zvol_geom_start(struct bio *bp);
232 static void zvol_geom_worker(void *arg);
233 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
234 uint64_t len, boolean_t sync);
235 #endif /* !illumos */
237 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
238 nvlist_t *, nvlist_t *);
239 static int zvol_remove_zv(zvol_state_t *);
240 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
241 static int zvol_dumpify(zvol_state_t *zv);
242 static int zvol_dump_fini(zvol_state_t *zv);
243 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
246 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
249 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
251 zv->zv_volsize = volsize;
252 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
253 "Size", volsize) == DDI_SUCCESS);
254 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
255 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
257 /* Notify specfs to invalidate the cached size */
258 spec_size_invalidate(dev, VBLK);
259 spec_size_invalidate(dev, VCHR);
261 zv->zv_volsize = volsize;
262 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
263 struct g_provider *pp;
265 pp = zv->zv_provider;
269 g_resize_provider(pp, zv->zv_volsize);
276 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
279 return (SET_ERROR(EINVAL));
281 if (volsize % blocksize != 0)
282 return (SET_ERROR(EINVAL));
285 if (volsize - 1 > SPEC_MAXOFFSET_T)
286 return (SET_ERROR(EOVERFLOW));
292 zvol_check_volblocksize(uint64_t volblocksize)
294 if (volblocksize < SPA_MINBLOCKSIZE ||
295 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
297 return (SET_ERROR(EDOM));
303 zvol_get_stats(objset_t *os, nvlist_t *nv)
306 dmu_object_info_t doi;
309 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
313 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
315 error = dmu_object_info(os, ZVOL_OBJ, &doi);
318 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
319 doi.doi_data_block_size);
325 static zvol_state_t *
326 zvol_minor_lookup(const char *name)
333 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
336 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
337 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
341 LIST_FOREACH(zv, &all_zvols, zv_links) {
343 if (strcmp(zv->zv_name, name) == 0)
350 /* extent mapping arg */
358 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
359 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
361 struct maparg *ma = arg;
363 int bs = ma->ma_zv->zv_volblocksize;
365 if (bp == NULL || BP_IS_HOLE(bp) ||
366 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
369 VERIFY(!BP_IS_EMBEDDED(bp));
371 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
374 /* Abort immediately if we have encountered gang blocks */
376 return (SET_ERROR(EFRAGS));
379 * See if the block is at the end of the previous extent.
381 ze = list_tail(&ma->ma_zv->zv_extents);
383 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
384 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
385 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
390 dprintf_bp(bp, "%s", "next blkptr:");
392 /* start a new extent */
393 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
394 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
396 list_insert_tail(&ma->ma_zv->zv_extents, ze);
401 zvol_free_extents(zvol_state_t *zv)
405 while (ze = list_head(&zv->zv_extents)) {
406 list_remove(&zv->zv_extents, ze);
407 kmem_free(ze, sizeof (zvol_extent_t));
412 zvol_get_lbas(zvol_state_t *zv)
414 objset_t *os = zv->zv_objset;
420 zvol_free_extents(zv);
422 /* commit any in-flight changes before traversing the dataset */
423 txg_wait_synced(dmu_objset_pool(os), 0);
424 err = traverse_dataset(dmu_objset_ds(os), 0,
425 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
426 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
427 zvol_free_extents(zv);
428 return (err ? err : EIO);
436 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
438 zfs_creat_t *zct = arg;
439 nvlist_t *nvprops = zct->zct_props;
441 uint64_t volblocksize, volsize;
443 VERIFY(nvlist_lookup_uint64(nvprops,
444 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
445 if (nvlist_lookup_uint64(nvprops,
446 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
447 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
450 * These properties must be removed from the list so the generic
451 * property setting step won't apply to them.
453 VERIFY(nvlist_remove_all(nvprops,
454 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
455 (void) nvlist_remove_all(nvprops,
456 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
458 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
462 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
466 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
471 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
472 * implement DKIOCFREE/free-long-range.
475 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
477 uint64_t offset, length;
480 byteswap_uint64_array(lr, sizeof (*lr));
482 offset = lr->lr_offset;
483 length = lr->lr_length;
485 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
489 * Replay a TX_WRITE ZIL transaction that didn't get committed
490 * after a system failure
493 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
495 objset_t *os = zv->zv_objset;
496 char *data = (char *)(lr + 1); /* data follows lr_write_t */
497 uint64_t offset, length;
502 byteswap_uint64_array(lr, sizeof (*lr));
504 offset = lr->lr_offset;
505 length = lr->lr_length;
507 /* If it's a dmu_sync() block, write the whole block */
508 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
509 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
510 if (length < blocksize) {
511 offset -= offset % blocksize;
516 tx = dmu_tx_create(os);
517 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
518 error = dmu_tx_assign(tx, TXG_WAIT);
522 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
531 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
533 return (SET_ERROR(ENOTSUP));
537 * Callback vectors for replaying records.
538 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
540 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
541 zvol_replay_err, /* 0 no such transaction type */
542 zvol_replay_err, /* TX_CREATE */
543 zvol_replay_err, /* TX_MKDIR */
544 zvol_replay_err, /* TX_MKXATTR */
545 zvol_replay_err, /* TX_SYMLINK */
546 zvol_replay_err, /* TX_REMOVE */
547 zvol_replay_err, /* TX_RMDIR */
548 zvol_replay_err, /* TX_LINK */
549 zvol_replay_err, /* TX_RENAME */
550 zvol_replay_write, /* TX_WRITE */
551 zvol_replay_truncate, /* TX_TRUNCATE */
552 zvol_replay_err, /* TX_SETATTR */
553 zvol_replay_err, /* TX_ACL */
554 zvol_replay_err, /* TX_CREATE_ACL */
555 zvol_replay_err, /* TX_CREATE_ATTR */
556 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
557 zvol_replay_err, /* TX_MKDIR_ACL */
558 zvol_replay_err, /* TX_MKDIR_ATTR */
559 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
560 zvol_replay_err, /* TX_WRITE2 */
565 zvol_name2minor(const char *name, minor_t *minor)
569 mutex_enter(&zfsdev_state_lock);
570 zv = zvol_minor_lookup(name);
572 *minor = zv->zv_minor;
573 mutex_exit(&zfsdev_state_lock);
574 return (zv ? 0 : -1);
579 * Create a minor node (plus a whole lot more) for the specified volume.
582 zvol_create_minor(const char *name)
584 zfs_soft_state_t *zs;
587 dmu_object_info_t doi;
590 char chrbuf[30], blkbuf[30];
592 struct g_provider *pp;
594 uint64_t volsize, mode;
599 ZFS_LOG(1, "Creating ZVOL %s...", name);
602 mutex_enter(&zfsdev_state_lock);
604 if (zvol_minor_lookup(name) != NULL) {
605 mutex_exit(&zfsdev_state_lock);
606 return (SET_ERROR(EEXIST));
609 /* lie and say we're read-only */
610 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
613 mutex_exit(&zfsdev_state_lock);
618 if ((minor = zfsdev_minor_alloc()) == 0) {
619 dmu_objset_disown(os, FTAG);
620 mutex_exit(&zfsdev_state_lock);
621 return (SET_ERROR(ENXIO));
624 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
625 dmu_objset_disown(os, FTAG);
626 mutex_exit(&zfsdev_state_lock);
627 return (SET_ERROR(EAGAIN));
629 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
632 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
634 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
635 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
636 ddi_soft_state_free(zfsdev_state, minor);
637 dmu_objset_disown(os, FTAG);
638 mutex_exit(&zfsdev_state_lock);
639 return (SET_ERROR(EAGAIN));
642 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
644 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
645 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
646 ddi_remove_minor_node(zfs_dip, chrbuf);
647 ddi_soft_state_free(zfsdev_state, minor);
648 dmu_objset_disown(os, FTAG);
649 mutex_exit(&zfsdev_state_lock);
650 return (SET_ERROR(EAGAIN));
653 zs = ddi_get_soft_state(zfsdev_state, minor);
654 zs->zss_type = ZSST_ZVOL;
655 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
658 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
660 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
662 kmem_free(zv, sizeof(*zv));
663 dmu_objset_disown(os, zvol_tag);
664 mutex_exit(&zfsdev_state_lock);
667 error = dsl_prop_get_integer(name,
668 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
669 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
673 zv->zv_volsize = volsize;
674 zv->zv_volmode = mode;
675 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
677 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
678 gp->start = zvol_geom_start;
679 gp->access = zvol_geom_access;
680 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
681 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
682 pp->sectorsize = DEV_BSIZE;
683 pp->mediasize = zv->zv_volsize;
686 zv->zv_provider = pp;
687 bioq_init(&zv->zv_queue);
688 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
689 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
690 struct make_dev_args args;
692 make_dev_args_init(&args);
693 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
694 args.mda_devsw = &zvol_cdevsw;
696 args.mda_uid = UID_ROOT;
697 args.mda_gid = GID_OPERATOR;
698 args.mda_mode = 0640;
699 args.mda_si_drv2 = zv;
700 error = make_dev_s(&args, &zv->zv_dev,
701 "%s/%s", ZVOL_DRIVER, name);
703 kmem_free(zv, sizeof(*zv));
704 dmu_objset_disown(os, FTAG);
705 mutex_exit(&zfsdev_state_lock);
708 zv->zv_dev->si_iosize_max = MAXPHYS;
710 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
713 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
714 zv->zv_min_bs = DEV_BSHIFT;
716 zv->zv_minor = minor;
719 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
720 zv->zv_flags |= ZVOL_RDONLY;
721 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
722 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
723 sizeof (rl_t), offsetof(rl_t, r_node));
724 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
725 offsetof(zvol_extent_t, ze_node));
726 /* get and cache the blocksize */
727 error = dmu_object_info(os, ZVOL_OBJ, &doi);
729 zv->zv_volblocksize = doi.doi_data_block_size;
731 if (spa_writeable(dmu_objset_spa(os))) {
732 if (zil_replay_disable)
733 zil_destroy(dmu_objset_zil(os), B_FALSE);
735 zil_replay(os, zv, zvol_replay_vector);
737 dmu_objset_disown(os, FTAG);
738 zv->zv_objset = NULL;
742 mutex_exit(&zfsdev_state_lock);
744 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
750 ZFS_LOG(1, "ZVOL %s created.", name);
757 * Remove minor node for the specified volume.
760 zvol_remove_zv(zvol_state_t *zv)
764 minor_t minor = zv->zv_minor;
767 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
768 if (zv->zv_total_opens != 0)
769 return (SET_ERROR(EBUSY));
772 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
773 ddi_remove_minor_node(zfs_dip, nmbuf);
775 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
776 ddi_remove_minor_node(zfs_dip, nmbuf);
778 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
780 LIST_REMOVE(zv, zv_links);
781 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
783 zvol_geom_destroy(zv);
785 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV)
786 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 boolean_t locked = B_FALSE;
1130 if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1132 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1133 * attempting to probe geom providers while looking for a
1134 * replacement for a missing VDEV. In this case, the
1135 * spa_namespace_lock will not be held, but it is still illegal
1136 * to use a zvol as a vdev. Deadlocks can result if another
1137 * thread has spa_namespace_lock
1139 return (EOPNOTSUPP);
1142 * Protect against recursively entering spa_namespace_lock
1143 * when spa_open() is used for a pool on a (local) ZVOL(s).
1144 * This is needed since we replaced upstream zfsdev_state_lock
1145 * with spa_namespace_lock in the ZVOL code.
1146 * We are using the same trick as spa_open().
1147 * Note that calls in zvol_first_open which need to resolve
1148 * pool name to a spa object will enter spa_open()
1149 * recursively, but that function already has all the
1150 * necessary protection.
1152 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1153 mutex_enter(&zfsdev_state_lock);
1160 mutex_exit(&zfsdev_state_lock);
1161 return (SET_ERROR(ENXIO));
1164 if (zv->zv_total_opens == 0) {
1165 err = zvol_first_open(zv);
1168 mutex_exit(&zfsdev_state_lock);
1171 pp->mediasize = zv->zv_volsize;
1172 pp->stripeoffset = 0;
1173 pp->stripesize = zv->zv_volblocksize;
1175 #endif /* illumos */
1176 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1177 err = SET_ERROR(EROFS);
1180 if (zv->zv_flags & ZVOL_EXCL) {
1181 err = SET_ERROR(EBUSY);
1186 if (zv->zv_total_opens != 0) {
1187 err = SET_ERROR(EBUSY);
1190 zv->zv_flags |= ZVOL_EXCL;
1195 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1196 zv->zv_open_count[otyp]++;
1197 zv->zv_total_opens++;
1199 mutex_exit(&zfsdev_state_lock);
1201 zv->zv_total_opens += count;
1203 mutex_exit(&zfsdev_state_lock);
1208 if (zv->zv_total_opens == 0)
1209 zvol_last_close(zv);
1211 mutex_exit(&zfsdev_state_lock);
1214 mutex_exit(&zfsdev_state_lock);
1222 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1224 minor_t minor = getminor(dev);
1228 mutex_enter(&zfsdev_state_lock);
1230 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1232 mutex_exit(&zfsdev_state_lock);
1233 #else /* !illumos */
1235 zvol_close(struct g_provider *pp, int flag, int count)
1239 boolean_t locked = B_FALSE;
1241 /* See comment in zvol_open(). */
1242 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1243 mutex_enter(&zfsdev_state_lock);
1250 mutex_exit(&zfsdev_state_lock);
1251 #endif /* illumos */
1252 return (SET_ERROR(ENXIO));
1255 if (zv->zv_flags & ZVOL_EXCL) {
1256 ASSERT(zv->zv_total_opens == 1);
1257 zv->zv_flags &= ~ZVOL_EXCL;
1261 * If the open count is zero, this is a spurious close.
1262 * That indicates a bug in the kernel / DDI framework.
1265 ASSERT(zv->zv_open_count[otyp] != 0);
1267 ASSERT(zv->zv_total_opens != 0);
1270 * You may get multiple opens, but only one close.
1273 zv->zv_open_count[otyp]--;
1274 zv->zv_total_opens--;
1276 zv->zv_total_opens -= count;
1279 if (zv->zv_total_opens == 0)
1280 zvol_last_close(zv);
1283 mutex_exit(&zfsdev_state_lock);
1286 mutex_exit(&zfsdev_state_lock);
1292 zvol_get_done(zgd_t *zgd, int error)
1295 dmu_buf_rele(zgd->zgd_db, zgd);
1297 zfs_range_unlock(zgd->zgd_rl);
1299 if (error == 0 && zgd->zgd_bp)
1300 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1302 kmem_free(zgd, sizeof (zgd_t));
1306 * Get data to generate a TX_WRITE intent log record.
1309 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1311 zvol_state_t *zv = arg;
1312 objset_t *os = zv->zv_objset;
1313 uint64_t object = ZVOL_OBJ;
1314 uint64_t offset = lr->lr_offset;
1315 uint64_t size = lr->lr_length; /* length of user data */
1316 blkptr_t *bp = &lr->lr_blkptr;
1321 ASSERT(zio != NULL);
1324 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1325 zgd->zgd_zilog = zv->zv_zilog;
1326 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1329 * Write records come in two flavors: immediate and indirect.
1330 * For small writes it's cheaper to store the data with the
1331 * log record (immediate); for large writes it's cheaper to
1332 * sync the data and get a pointer to it (indirect) so that
1333 * we don't have to write the data twice.
1335 if (buf != NULL) { /* immediate write */
1336 error = dmu_read(os, object, offset, size, buf,
1337 DMU_READ_NO_PREFETCH);
1339 size = zv->zv_volblocksize;
1340 offset = P2ALIGN(offset, size);
1341 error = dmu_buf_hold(os, object, offset, zgd, &db,
1342 DMU_READ_NO_PREFETCH);
1344 blkptr_t *obp = dmu_buf_get_blkptr(db);
1346 ASSERT(BP_IS_HOLE(bp));
1353 ASSERT(db->db_offset == offset);
1354 ASSERT(db->db_size == size);
1356 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1357 zvol_get_done, zgd);
1364 zvol_get_done(zgd, error);
1370 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1372 * We store data in the log buffers if it's small enough.
1373 * Otherwise we will later flush the data out via dmu_sync().
1375 ssize_t zvol_immediate_write_sz = 32768;
1378 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1381 uint32_t blocksize = zv->zv_volblocksize;
1382 zilog_t *zilog = zv->zv_zilog;
1384 ssize_t immediate_write_sz;
1386 if (zil_replaying(zilog, tx))
1389 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1390 ? 0 : zvol_immediate_write_sz;
1392 slogging = spa_has_slogs(zilog->zl_spa) &&
1393 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1399 itx_wr_state_t write_state;
1402 * Unlike zfs_log_write() we can be called with
1403 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1405 if (blocksize > immediate_write_sz && !slogging &&
1406 resid >= blocksize && off % blocksize == 0) {
1407 write_state = WR_INDIRECT; /* uses dmu_sync */
1410 write_state = WR_COPIED;
1411 len = MIN(ZIL_MAX_LOG_DATA, resid);
1413 write_state = WR_NEED_COPY;
1414 len = MIN(ZIL_MAX_LOG_DATA, resid);
1417 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1418 (write_state == WR_COPIED ? len : 0));
1419 lr = (lr_write_t *)&itx->itx_lr;
1420 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1421 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1422 zil_itx_destroy(itx);
1423 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1424 lr = (lr_write_t *)&itx->itx_lr;
1425 write_state = WR_NEED_COPY;
1428 itx->itx_wr_state = write_state;
1429 if (write_state == WR_NEED_COPY)
1430 itx->itx_sod += len;
1431 lr->lr_foid = ZVOL_OBJ;
1432 lr->lr_offset = off;
1433 lr->lr_length = len;
1435 BP_ZERO(&lr->lr_blkptr);
1437 itx->itx_private = zv;
1438 itx->itx_sync = sync;
1440 zil_itx_assign(zilog, itx, tx);
1449 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1450 uint64_t size, boolean_t doread, boolean_t isdump)
1456 if (vd->vdev_ops == &vdev_mirror_ops ||
1457 vd->vdev_ops == &vdev_replacing_ops ||
1458 vd->vdev_ops == &vdev_spare_ops) {
1459 for (c = 0; c < vd->vdev_children; c++) {
1460 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1461 addr, offset, origoffset, size, doread, isdump);
1464 } else if (doread) {
1470 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1471 return (numerrors < vd->vdev_children ? 0 : EIO);
1473 if (doread && !vdev_readable(vd))
1474 return (SET_ERROR(EIO));
1475 else if (!doread && !vdev_writeable(vd))
1476 return (SET_ERROR(EIO));
1478 if (vd->vdev_ops == &vdev_raidz_ops) {
1479 return (vdev_raidz_physio(vd,
1480 addr, size, offset, origoffset, doread, isdump));
1483 offset += VDEV_LABEL_START_SIZE;
1485 if (ddi_in_panic() || isdump) {
1488 return (SET_ERROR(EIO));
1490 ASSERT3P(dvd, !=, NULL);
1491 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1495 ASSERT3P(dvd, !=, NULL);
1496 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1497 offset, doread ? B_READ : B_WRITE));
1502 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1503 boolean_t doread, boolean_t isdump)
1508 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1510 /* Must be sector aligned, and not stradle a block boundary. */
1511 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1512 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1513 return (SET_ERROR(EINVAL));
1515 ASSERT(size <= zv->zv_volblocksize);
1517 /* Locate the extent this belongs to */
1518 ze = list_head(&zv->zv_extents);
1519 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1520 offset -= ze->ze_nblks * zv->zv_volblocksize;
1521 ze = list_next(&zv->zv_extents, ze);
1525 return (SET_ERROR(EINVAL));
1527 if (!ddi_in_panic())
1528 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1530 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1531 offset += DVA_GET_OFFSET(&ze->ze_dva);
1532 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1533 size, doread, isdump);
1535 if (!ddi_in_panic())
1536 spa_config_exit(spa, SCL_STATE, FTAG);
1542 zvol_strategy(buf_t *bp)
1544 zfs_soft_state_t *zs = NULL;
1545 #else /* !illumos */
1547 zvol_strategy(struct bio *bp)
1549 #endif /* illumos */
1551 uint64_t off, volsize;
1558 boolean_t doread = bp->b_flags & B_READ;
1560 boolean_t doread = 0;
1562 boolean_t is_dumpified;
1566 if (getminor(bp->b_edev) == 0) {
1567 error = SET_ERROR(EINVAL);
1569 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1571 error = SET_ERROR(ENXIO);
1572 else if (zs->zss_type != ZSST_ZVOL)
1573 error = SET_ERROR(EINVAL);
1577 bioerror(bp, error);
1584 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1585 bioerror(bp, EROFS);
1590 off = ldbtob(bp->b_blkno);
1591 #else /* !illumos */
1593 zv = bp->bio_to->private;
1595 zv = bp->bio_dev->si_drv2;
1598 error = SET_ERROR(ENXIO);
1602 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1603 error = SET_ERROR(EROFS);
1607 switch (bp->bio_cmd) {
1620 off = bp->bio_offset;
1621 #endif /* illumos */
1622 volsize = zv->zv_volsize;
1629 addr = bp->b_un.b_addr;
1630 resid = bp->b_bcount;
1632 if (resid > 0 && (off < 0 || off >= volsize)) {
1638 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1639 sync = ((!(bp->b_flags & B_ASYNC) &&
1640 !(zv->zv_flags & ZVOL_WCE)) ||
1641 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1642 !doread && !is_dumpified;
1643 #else /* !illumos */
1644 addr = bp->bio_data;
1645 resid = bp->bio_length;
1647 if (resid > 0 && (off < 0 || off >= volsize)) {
1648 error = SET_ERROR(EIO);
1652 is_dumpified = B_FALSE;
1653 sync = !doread && !is_dumpified &&
1654 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1655 #endif /* illumos */
1658 * There must be no buffer changes when doing a dmu_sync() because
1659 * we can't change the data whilst calculating the checksum.
1661 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1662 doread ? RL_READER : RL_WRITER);
1665 if (bp->bio_cmd == BIO_DELETE) {
1666 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1667 error = dmu_tx_assign(tx, TXG_WAIT);
1671 zvol_log_truncate(zv, tx, off, resid, B_TRUE);
1673 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1680 while (resid != 0 && off < volsize) {
1681 size_t size = MIN(resid, zvol_maxphys);
1684 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1685 error = zvol_dumpio(zv, addr, off, size,
1687 } else if (doread) {
1691 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1694 dmu_tx_t *tx = dmu_tx_create(os);
1695 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1696 error = dmu_tx_assign(tx, TXG_WAIT);
1700 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1701 zvol_log_write(zv, tx, off, size, sync);
1706 /* convert checksum errors into IO errors */
1707 if (error == ECKSUM)
1708 error = SET_ERROR(EIO);
1718 zfs_range_unlock(rl);
1721 if ((bp->b_resid = resid) == bp->b_bcount)
1722 bioerror(bp, off > volsize ? EINVAL : error);
1725 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1729 #else /* !illumos */
1730 bp->bio_completed = bp->bio_length - resid;
1731 if (bp->bio_completed < bp->bio_length && off > volsize)
1736 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1740 g_io_deliver(bp, error);
1742 biofinish(bp, NULL, error);
1743 #endif /* illumos */
1748 * Set the buffer count to the zvol maximum transfer.
1749 * Using our own routine instead of the default minphys()
1750 * means that for larger writes we write bigger buffers on X86
1751 * (128K instead of 56K) and flush the disk write cache less often
1752 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1753 * 56K on X86 and 128K on sparc).
1756 zvol_minphys(struct buf *bp)
1758 if (bp->b_bcount > zvol_maxphys)
1759 bp->b_bcount = zvol_maxphys;
1763 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1765 minor_t minor = getminor(dev);
1772 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1774 return (SET_ERROR(ENXIO));
1776 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1777 return (SET_ERROR(EINVAL));
1779 boff = ldbtob(blkno);
1780 resid = ldbtob(nblocks);
1782 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1785 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1786 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1799 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1801 minor_t minor = getminor(dev);
1802 #else /* !illumos */
1804 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1806 #endif /* illumos */
1813 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1815 return (SET_ERROR(ENXIO));
1820 volsize = zv->zv_volsize;
1821 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1822 if (uio->uio_resid > 0 &&
1823 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1824 return (SET_ERROR(EIO));
1827 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1828 error = physio(zvol_strategy, NULL, dev, B_READ,
1834 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1836 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1837 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1839 /* don't read past the end */
1840 if (bytes > volsize - uio->uio_loffset)
1841 bytes = volsize - uio->uio_loffset;
1843 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
1845 /* convert checksum errors into IO errors */
1846 if (error == ECKSUM)
1847 error = SET_ERROR(EIO);
1851 zfs_range_unlock(rl);
1858 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1860 minor_t minor = getminor(dev);
1861 #else /* !illumos */
1863 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1865 #endif /* illumos */
1873 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1875 return (SET_ERROR(ENXIO));
1880 volsize = zv->zv_volsize;
1881 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1882 if (uio->uio_resid > 0 &&
1883 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1884 return (SET_ERROR(EIO));
1887 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1888 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1893 sync = !(zv->zv_flags & ZVOL_WCE) ||
1895 sync = (ioflag & IO_SYNC) ||
1897 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1899 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1901 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1902 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1903 uint64_t off = uio->uio_loffset;
1904 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1906 if (bytes > volsize - off) /* don't write past the end */
1907 bytes = volsize - off;
1909 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1910 error = dmu_tx_assign(tx, TXG_WAIT);
1915 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1917 zvol_log_write(zv, tx, off, bytes, sync);
1923 zfs_range_unlock(rl);
1925 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1931 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1933 struct uuid uuid = EFI_RESERVED;
1934 efi_gpe_t gpe = { 0 };
1940 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1941 return (SET_ERROR(EFAULT));
1942 ptr = (char *)(uintptr_t)efi.dki_data_64;
1943 length = efi.dki_length;
1945 * Some clients may attempt to request a PMBR for the
1946 * zvol. Currently this interface will return EINVAL to
1947 * such requests. These requests could be supported by
1948 * adding a check for lba == 0 and consing up an appropriate
1951 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1952 return (SET_ERROR(EINVAL));
1954 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1955 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1956 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1958 if (efi.dki_lba == 1) {
1959 efi_gpt_t gpt = { 0 };
1961 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1962 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1963 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1964 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1965 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1966 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1967 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1968 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1969 gpt.efi_gpt_SizeOfPartitionEntry =
1970 LE_32(sizeof (efi_gpe_t));
1971 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1972 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1973 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1974 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1975 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1977 return (SET_ERROR(EFAULT));
1978 ptr += sizeof (gpt);
1979 length -= sizeof (gpt);
1981 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1983 return (SET_ERROR(EFAULT));
1988 * BEGIN entry points to allow external callers access to the volume.
1991 * Return the volume parameters needed for access from an external caller.
1992 * These values are invariant as long as the volume is held open.
1995 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1996 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1997 void **rl_hdl, void **bonus_hdl)
2001 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2003 return (SET_ERROR(ENXIO));
2004 if (zv->zv_flags & ZVOL_DUMPIFIED)
2005 return (SET_ERROR(ENXIO));
2007 ASSERT(blksize && max_xfer_len && minor_hdl &&
2008 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
2010 *blksize = zv->zv_volblocksize;
2011 *max_xfer_len = (uint64_t)zvol_maxphys;
2013 *objset_hdl = zv->zv_objset;
2014 *zil_hdl = zv->zv_zilog;
2015 *rl_hdl = &zv->zv_znode;
2016 *bonus_hdl = zv->zv_dbuf;
2021 * Return the current volume size to an external caller.
2022 * The size can change while the volume is open.
2025 zvol_get_volume_size(void *minor_hdl)
2027 zvol_state_t *zv = minor_hdl;
2029 return (zv->zv_volsize);
2033 * Return the current WCE setting to an external caller.
2034 * The WCE setting can change while the volume is open.
2037 zvol_get_volume_wce(void *minor_hdl)
2039 zvol_state_t *zv = minor_hdl;
2041 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2045 * Entry point for external callers to zvol_log_write
2048 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2051 zvol_state_t *zv = minor_hdl;
2053 zvol_log_write(zv, tx, off, resid, sync);
2056 * END entry points to allow external callers access to the volume.
2058 #endif /* illumos */
2061 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2064 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2069 zilog_t *zilog = zv->zv_zilog;
2071 if (zil_replaying(zilog, tx))
2074 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2075 lr = (lr_truncate_t *)&itx->itx_lr;
2076 lr->lr_foid = ZVOL_OBJ;
2077 lr->lr_offset = off;
2078 lr->lr_length = len;
2080 itx->itx_sync = sync;
2081 zil_itx_assign(zilog, itx, tx);
2086 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2087 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2091 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2094 struct dk_callback *dkc;
2098 mutex_enter(&zfsdev_state_lock);
2100 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2103 mutex_exit(&zfsdev_state_lock);
2104 return (SET_ERROR(ENXIO));
2106 ASSERT(zv->zv_total_opens > 0);
2112 struct dk_cinfo dki;
2114 bzero(&dki, sizeof (dki));
2115 (void) strcpy(dki.dki_cname, "zvol");
2116 (void) strcpy(dki.dki_dname, "zvol");
2117 dki.dki_ctype = DKC_UNKNOWN;
2118 dki.dki_unit = getminor(dev);
2119 dki.dki_maxtransfer =
2120 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2121 mutex_exit(&zfsdev_state_lock);
2122 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2123 error = SET_ERROR(EFAULT);
2127 case DKIOCGMEDIAINFO:
2129 struct dk_minfo dkm;
2131 bzero(&dkm, sizeof (dkm));
2132 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2133 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2134 dkm.dki_media_type = DK_UNKNOWN;
2135 mutex_exit(&zfsdev_state_lock);
2136 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2137 error = SET_ERROR(EFAULT);
2141 case DKIOCGMEDIAINFOEXT:
2143 struct dk_minfo_ext dkmext;
2145 bzero(&dkmext, sizeof (dkmext));
2146 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2147 dkmext.dki_pbsize = zv->zv_volblocksize;
2148 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2149 dkmext.dki_media_type = DK_UNKNOWN;
2150 mutex_exit(&zfsdev_state_lock);
2151 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2152 error = SET_ERROR(EFAULT);
2158 uint64_t vs = zv->zv_volsize;
2159 uint8_t bs = zv->zv_min_bs;
2161 mutex_exit(&zfsdev_state_lock);
2162 error = zvol_getefi((void *)arg, flag, vs, bs);
2166 case DKIOCFLUSHWRITECACHE:
2167 dkc = (struct dk_callback *)arg;
2168 mutex_exit(&zfsdev_state_lock);
2169 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2170 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2171 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2178 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2179 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2181 error = SET_ERROR(EFAULT);
2187 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2189 error = SET_ERROR(EFAULT);
2193 zv->zv_flags |= ZVOL_WCE;
2194 mutex_exit(&zfsdev_state_lock);
2196 zv->zv_flags &= ~ZVOL_WCE;
2197 mutex_exit(&zfsdev_state_lock);
2198 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2206 * commands using these (like prtvtoc) expect ENOTSUP
2207 * since we're emulating an EFI label
2209 error = SET_ERROR(ENOTSUP);
2213 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2215 error = zvol_dumpify(zv);
2216 zfs_range_unlock(rl);
2220 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2222 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2224 error = zvol_dump_fini(zv);
2225 zfs_range_unlock(rl);
2233 if (!zvol_unmap_enabled)
2236 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2237 error = SET_ERROR(EFAULT);
2242 * Apply Postel's Law to length-checking. If they overshoot,
2243 * just blank out until the end, if there's a need to blank
2246 if (df.df_start >= zv->zv_volsize)
2247 break; /* No need to do anything... */
2249 mutex_exit(&zfsdev_state_lock);
2251 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2253 tx = dmu_tx_create(zv->zv_objset);
2254 dmu_tx_mark_netfree(tx);
2255 error = dmu_tx_assign(tx, TXG_WAIT);
2259 zvol_log_truncate(zv, tx, df.df_start,
2260 df.df_length, B_TRUE);
2262 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2263 df.df_start, df.df_length);
2266 zfs_range_unlock(rl);
2270 * If the write-cache is disabled or 'sync' property
2271 * is set to 'always' then treat this as a synchronous
2272 * operation (i.e. commit to zil).
2274 if (!(zv->zv_flags & ZVOL_WCE) ||
2275 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2276 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2279 * If the caller really wants synchronous writes, and
2280 * can't wait for them, don't return until the write
2283 if (df.df_flags & DF_WAIT_SYNC) {
2285 dmu_objset_pool(zv->zv_objset), 0);
2292 error = SET_ERROR(ENOTTY);
2296 mutex_exit(&zfsdev_state_lock);
2299 #endif /* illumos */
2304 return (zvol_minors != 0);
2310 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2313 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2315 ZFS_LOG(1, "ZVOL Initialized.");
2323 mutex_destroy(&zfsdev_state_lock);
2325 ddi_soft_state_fini(&zfsdev_state);
2326 ZFS_LOG(1, "ZVOL Deinitialized.");
2332 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2334 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2336 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2343 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2345 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2347 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2351 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2355 objset_t *os = zv->zv_objset;
2356 spa_t *spa = dmu_objset_spa(os);
2357 vdev_t *vd = spa->spa_root_vdev;
2358 nvlist_t *nv = NULL;
2359 uint64_t version = spa_version(spa);
2360 uint64_t checksum, compress, refresrv, vbs, dedup;
2362 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2363 ASSERT(vd->vdev_ops == &vdev_root_ops);
2365 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2369 /* wait for dmu_free_long_range to actually free the blocks */
2370 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2373 * If the pool on which the dump device is being initialized has more
2374 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2375 * enabled. If so, bump that feature's counter to indicate that the
2376 * feature is active. We also check the vdev type to handle the
2378 * # zpool create test raidz disk1 disk2 disk3
2379 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2380 * the raidz vdev itself has 3 children.
2382 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2383 if (!spa_feature_is_enabled(spa,
2384 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2385 return (SET_ERROR(ENOTSUP));
2386 (void) dsl_sync_task(spa_name(spa),
2387 zfs_mvdev_dump_feature_check,
2388 zfs_mvdev_dump_activate_feature_sync, NULL,
2389 2, ZFS_SPACE_CHECK_RESERVED);
2393 error = dsl_prop_get_integer(zv->zv_name,
2394 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2396 error = dsl_prop_get_integer(zv->zv_name,
2397 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2401 error = dsl_prop_get_integer(zv->zv_name,
2402 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2406 error = dsl_prop_get_integer(zv->zv_name,
2407 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2410 if (version >= SPA_VERSION_DEDUP && error == 0) {
2411 error = dsl_prop_get_integer(zv->zv_name,
2412 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2418 tx = dmu_tx_create(os);
2419 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2420 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2421 error = dmu_tx_assign(tx, TXG_WAIT);
2428 * If we are resizing the dump device then we only need to
2429 * update the refreservation to match the newly updated
2430 * zvolsize. Otherwise, we save off the original state of the
2431 * zvol so that we can restore them if the zvol is ever undumpified.
2434 error = zap_update(os, ZVOL_ZAP_OBJ,
2435 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2436 &zv->zv_volsize, tx);
2438 error = zap_update(os, ZVOL_ZAP_OBJ,
2439 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2442 error = zap_update(os, ZVOL_ZAP_OBJ,
2443 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2447 error = zap_update(os, ZVOL_ZAP_OBJ,
2448 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2452 error = zap_update(os, ZVOL_ZAP_OBJ,
2453 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2457 error = dmu_object_set_blocksize(
2458 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2460 if (version >= SPA_VERSION_DEDUP && error == 0) {
2461 error = zap_update(os, ZVOL_ZAP_OBJ,
2462 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2466 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2471 * We only need update the zvol's property if we are initializing
2472 * the dump area for the first time.
2474 if (error == 0 && !resize) {
2476 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2477 * function. Otherwise, use the old default -- OFF.
2479 checksum = spa_feature_is_active(spa,
2480 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2483 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2484 VERIFY(nvlist_add_uint64(nv,
2485 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2486 VERIFY(nvlist_add_uint64(nv,
2487 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2488 ZIO_COMPRESS_OFF) == 0);
2489 VERIFY(nvlist_add_uint64(nv,
2490 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2492 if (version >= SPA_VERSION_DEDUP) {
2493 VERIFY(nvlist_add_uint64(nv,
2494 zfs_prop_to_name(ZFS_PROP_DEDUP),
2495 ZIO_CHECKSUM_OFF) == 0);
2498 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2503 /* Allocate the space for the dump */
2505 error = zvol_prealloc(zv);
2510 zvol_dumpify(zvol_state_t *zv)
2513 uint64_t dumpsize = 0;
2515 objset_t *os = zv->zv_objset;
2517 if (zv->zv_flags & ZVOL_RDONLY)
2518 return (SET_ERROR(EROFS));
2520 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2521 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2522 boolean_t resize = (dumpsize > 0);
2524 if ((error = zvol_dump_init(zv, resize)) != 0) {
2525 (void) zvol_dump_fini(zv);
2531 * Build up our lba mapping.
2533 error = zvol_get_lbas(zv);
2535 (void) zvol_dump_fini(zv);
2539 tx = dmu_tx_create(os);
2540 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2541 error = dmu_tx_assign(tx, TXG_WAIT);
2544 (void) zvol_dump_fini(zv);
2548 zv->zv_flags |= ZVOL_DUMPIFIED;
2549 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2550 &zv->zv_volsize, tx);
2554 (void) zvol_dump_fini(zv);
2558 txg_wait_synced(dmu_objset_pool(os), 0);
2563 zvol_dump_fini(zvol_state_t *zv)
2566 objset_t *os = zv->zv_objset;
2569 uint64_t checksum, compress, refresrv, vbs, dedup;
2570 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2573 * Attempt to restore the zvol back to its pre-dumpified state.
2574 * This is a best-effort attempt as it's possible that not all
2575 * of these properties were initialized during the dumpify process
2576 * (i.e. error during zvol_dump_init).
2579 tx = dmu_tx_create(os);
2580 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2581 error = dmu_tx_assign(tx, TXG_WAIT);
2586 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2589 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2590 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2591 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2592 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2593 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2594 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2595 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2596 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2598 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2599 (void) nvlist_add_uint64(nv,
2600 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2601 (void) nvlist_add_uint64(nv,
2602 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2603 (void) nvlist_add_uint64(nv,
2604 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2605 if (version >= SPA_VERSION_DEDUP &&
2606 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2607 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2608 (void) nvlist_add_uint64(nv,
2609 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2611 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2615 zvol_free_extents(zv);
2616 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2617 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2618 /* wait for dmu_free_long_range to actually free the blocks */
2619 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2620 tx = dmu_tx_create(os);
2621 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2622 error = dmu_tx_assign(tx, TXG_WAIT);
2627 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2628 zv->zv_volblocksize = vbs;
2633 #else /* !illumos */
2636 zvol_geom_run(zvol_state_t *zv)
2638 struct g_provider *pp;
2640 pp = zv->zv_provider;
2641 g_error_provider(pp, 0);
2643 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2644 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2648 zvol_geom_destroy(zvol_state_t *zv)
2650 struct g_provider *pp;
2652 g_topology_assert();
2654 mtx_lock(&zv->zv_queue_mtx);
2656 wakeup_one(&zv->zv_queue);
2657 while (zv->zv_state != 2)
2658 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2659 mtx_destroy(&zv->zv_queue_mtx);
2661 pp = zv->zv_provider;
2662 zv->zv_provider = NULL;
2664 g_wither_geom(pp->geom, ENXIO);
2668 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2670 int count, error, flags;
2672 g_topology_assert();
2675 * To make it easier we expect either open or close, but not both
2678 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2679 (acr <= 0 && acw <= 0 && ace <= 0),
2680 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2681 pp->name, acr, acw, ace));
2683 if (pp->private == NULL) {
2684 if (acr <= 0 && acw <= 0 && ace <= 0)
2690 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2691 * because GEOM already handles that and handles it a bit differently.
2692 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2693 * only one exclusive consumer, no matter if it is reader or writer.
2694 * I like better the way GEOM works so I'll leave it for GEOM to
2695 * decide what to do.
2698 count = acr + acw + ace;
2703 if (acr != 0 || ace != 0)
2708 g_topology_unlock();
2710 error = zvol_open(pp, flags, count);
2712 error = zvol_close(pp, flags, -count);
2718 zvol_geom_start(struct bio *bp)
2723 zv = bp->bio_to->private;
2725 switch (bp->bio_cmd) {
2727 if (!THREAD_CAN_SLEEP())
2729 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2730 g_io_deliver(bp, 0);
2735 if (!THREAD_CAN_SLEEP())
2740 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2741 uint64_t refd, avail, usedobjs, availobjs, val;
2743 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2745 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2746 dmu_objset_space(zv->zv_objset, &refd, &avail,
2747 &usedobjs, &availobjs);
2748 if (g_handleattr_off_t(bp, "blocksavail",
2751 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2752 dmu_objset_space(zv->zv_objset, &refd, &avail,
2753 &usedobjs, &availobjs);
2754 if (g_handleattr_off_t(bp, "blocksused",
2757 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2758 avail = metaslab_class_get_space(spa_normal_class(spa));
2759 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2760 if (g_handleattr_off_t(bp, "poolblocksavail",
2763 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2764 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2765 if (g_handleattr_off_t(bp, "poolblocksused",
2772 g_io_deliver(bp, EOPNOTSUPP);
2778 mtx_lock(&zv->zv_queue_mtx);
2779 first = (bioq_first(&zv->zv_queue) == NULL);
2780 bioq_insert_tail(&zv->zv_queue, bp);
2781 mtx_unlock(&zv->zv_queue_mtx);
2783 wakeup_one(&zv->zv_queue);
2787 zvol_geom_worker(void *arg)
2792 thread_lock(curthread);
2793 sched_prio(curthread, PRIBIO);
2794 thread_unlock(curthread);
2798 mtx_lock(&zv->zv_queue_mtx);
2799 bp = bioq_takefirst(&zv->zv_queue);
2801 if (zv->zv_state == 1) {
2803 wakeup(&zv->zv_state);
2804 mtx_unlock(&zv->zv_queue_mtx);
2807 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2811 mtx_unlock(&zv->zv_queue_mtx);
2812 switch (bp->bio_cmd) {
2814 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2815 g_io_deliver(bp, 0);
2823 g_io_deliver(bp, EOPNOTSUPP);
2829 extern boolean_t dataset_name_hidden(const char *name);
2832 zvol_create_snapshots(objset_t *os, const char *name)
2834 uint64_t cookie, obj;
2839 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2842 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2847 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2848 if (len >= MAXPATHLEN) {
2849 dmu_objset_rele(os, FTAG);
2850 error = ENAMETOOLONG;
2854 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2855 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2856 sname + len, &obj, &cookie, NULL);
2857 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2859 if (error == ENOENT)
2864 error = zvol_create_minor(sname);
2865 if (error != 0 && error != EEXIST) {
2866 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2872 kmem_free(sname, MAXPATHLEN);
2877 zvol_create_minors(const char *name)
2884 if (dataset_name_hidden(name))
2887 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2888 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2892 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2893 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2894 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2895 error = zvol_create_minor(name);
2896 if (error == 0 || error == EEXIST) {
2897 error = zvol_create_snapshots(os, name);
2899 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2902 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2903 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2906 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2907 dmu_objset_rele(os, FTAG);
2911 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2912 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2913 dmu_objset_rele(os, FTAG);
2914 kmem_free(osname, MAXPATHLEN);
2917 p = osname + strlen(osname);
2918 len = MAXPATHLEN - (p - osname);
2921 /* Prefetch the datasets. */
2923 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2924 if (!dataset_name_hidden(osname))
2925 (void) dmu_objset_prefetch(osname, NULL);
2930 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2932 dmu_objset_rele(os, FTAG);
2933 (void)zvol_create_minors(osname);
2934 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2935 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2941 dmu_objset_rele(os, FTAG);
2942 kmem_free(osname, MAXPATHLEN);
2947 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2950 struct g_provider *pp;
2953 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2955 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2957 pp = zv->zv_provider;
2962 zv->zv_provider = NULL;
2963 g_wither_provider(pp, ENXIO);
2965 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2966 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2967 pp->sectorsize = DEV_BSIZE;
2968 pp->mediasize = zv->zv_volsize;
2970 zv->zv_provider = pp;
2971 g_error_provider(pp, 0);
2972 g_topology_unlock();
2973 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2974 struct make_dev_args args;
2977 ASSERT(dev != NULL);
2980 if (zv->zv_total_opens > 0) {
2981 zv->zv_flags &= ~ZVOL_EXCL;
2982 zv->zv_total_opens = 0;
2983 zvol_last_close(zv);
2986 make_dev_args_init(&args);
2987 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
2988 args.mda_devsw = &zvol_cdevsw;
2990 args.mda_uid = UID_ROOT;
2991 args.mda_gid = GID_OPERATOR;
2992 args.mda_mode = 0640;
2993 args.mda_si_drv2 = zv;
2994 if (make_dev_s(&args, &zv->zv_dev,
2995 "%s/%s", ZVOL_DRIVER, newname) == 0)
2996 zv->zv_dev->si_iosize_max = MAXPHYS;
2998 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3002 zvol_rename_minors(const char *oldname, const char *newname)
3004 char name[MAXPATHLEN];
3005 struct g_provider *pp;
3007 size_t oldnamelen, newnamelen;
3010 boolean_t locked = B_FALSE;
3012 oldnamelen = strlen(oldname);
3013 newnamelen = strlen(newname);
3016 /* See comment in zvol_open(). */
3017 if (!MUTEX_HELD(&zfsdev_state_lock)) {
3018 mutex_enter(&zfsdev_state_lock);
3022 LIST_FOREACH(zv, &all_zvols, zv_links) {
3023 if (strcmp(zv->zv_name, oldname) == 0) {
3024 zvol_rename_minor(zv, newname);
3025 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3026 (zv->zv_name[oldnamelen] == '/' ||
3027 zv->zv_name[oldnamelen] == '@')) {
3028 snprintf(name, sizeof(name), "%s%c%s", newname,
3029 zv->zv_name[oldnamelen],
3030 zv->zv_name + oldnamelen + 1);
3031 zvol_rename_minor(zv, name);
3036 mutex_exit(&zfsdev_state_lock);
3041 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3043 zvol_state_t *zv = dev->si_drv2;
3046 mutex_enter(&zfsdev_state_lock);
3047 if (zv->zv_total_opens == 0)
3048 err = zvol_first_open(zv);
3050 mutex_exit(&zfsdev_state_lock);
3053 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3054 err = SET_ERROR(EROFS);
3057 if (zv->zv_flags & ZVOL_EXCL) {
3058 err = SET_ERROR(EBUSY);
3062 if (flags & FEXCL) {
3063 if (zv->zv_total_opens != 0) {
3064 err = SET_ERROR(EBUSY);
3067 zv->zv_flags |= ZVOL_EXCL;
3071 zv->zv_total_opens++;
3072 mutex_exit(&zfsdev_state_lock);
3075 if (zv->zv_total_opens == 0)
3076 zvol_last_close(zv);
3077 mutex_exit(&zfsdev_state_lock);
3082 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3084 zvol_state_t *zv = dev->si_drv2;
3086 mutex_enter(&zfsdev_state_lock);
3087 if (zv->zv_flags & ZVOL_EXCL) {
3088 ASSERT(zv->zv_total_opens == 1);
3089 zv->zv_flags &= ~ZVOL_EXCL;
3093 * If the open count is zero, this is a spurious close.
3094 * That indicates a bug in the kernel / DDI framework.
3096 ASSERT(zv->zv_total_opens != 0);
3099 * You may get multiple opens, but only one close.
3101 zv->zv_total_opens--;
3103 if (zv->zv_total_opens == 0)
3104 zvol_last_close(zv);
3106 mutex_exit(&zfsdev_state_lock);
3111 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3115 off_t offset, length, chunk;
3122 KASSERT(zv->zv_total_opens > 0,
3123 ("Device with zero access count in zvol_d_ioctl"));
3125 i = IOCPARM_LEN(cmd);
3127 case DIOCGSECTORSIZE:
3128 *(u_int *)data = DEV_BSIZE;
3130 case DIOCGMEDIASIZE:
3131 *(off_t *)data = zv->zv_volsize;
3134 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3137 if (!zvol_unmap_enabled)
3140 offset = ((off_t *)data)[0];
3141 length = ((off_t *)data)[1];
3142 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3143 offset < 0 || offset >= zv->zv_volsize ||
3145 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3151 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3152 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3153 error = dmu_tx_assign(tx, TXG_WAIT);
3157 zvol_log_truncate(zv, tx, offset, length, B_TRUE);
3159 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3162 zfs_range_unlock(rl);
3163 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
3164 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3166 case DIOCGSTRIPESIZE:
3167 *(off_t *)data = zv->zv_volblocksize;
3169 case DIOCGSTRIPEOFFSET:
3173 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3174 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3175 uint64_t refd, avail, usedobjs, availobjs;
3177 if (strcmp(arg->name, "GEOM::candelete") == 0)
3179 else if (strcmp(arg->name, "blocksavail") == 0) {
3180 dmu_objset_space(zv->zv_objset, &refd, &avail,
3181 &usedobjs, &availobjs);
3182 arg->value.off = avail / DEV_BSIZE;
3183 } else if (strcmp(arg->name, "blocksused") == 0) {
3184 dmu_objset_space(zv->zv_objset, &refd, &avail,
3185 &usedobjs, &availobjs);
3186 arg->value.off = refd / DEV_BSIZE;
3187 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3188 avail = metaslab_class_get_space(spa_normal_class(spa));
3189 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3190 arg->value.off = avail / DEV_BSIZE;
3191 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3192 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3193 arg->value.off = refd / DEV_BSIZE;
3200 off_t *off = (off_t *)data;
3204 hole = (cmd == FIOSEEKHOLE);
3206 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3216 #endif /* illumos */