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 uint32_t zv_sync_cnt; /* synchronous open count */
173 zilog_t *zv_zilog; /* ZIL handle */
174 list_t zv_extents; /* List of extents for dump */
175 znode_t zv_znode; /* for range locking */
176 dmu_buf_t *zv_dbuf; /* bonus handle */
179 int zv_volmode; /* Provide GEOM or cdev */
180 struct bio_queue_head zv_queue;
181 struct mtx zv_queue_mtx; /* zv_queue mutex */
186 static LIST_HEAD(, zvol_state) all_zvols;
189 * zvol specific flags
191 #define ZVOL_RDONLY 0x1
192 #define ZVOL_DUMPIFIED 0x2
193 #define ZVOL_EXCL 0x4
197 * zvol maximum transfer in one DMU tx.
199 int zvol_maxphys = DMU_MAX_ACCESS/2;
202 * Toggle unmap functionality.
204 boolean_t zvol_unmap_enabled = B_TRUE;
206 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
207 &zvol_unmap_enabled, 0,
208 "Enable UNMAP functionality");
210 static d_open_t zvol_d_open;
211 static d_close_t zvol_d_close;
212 static d_read_t zvol_read;
213 static d_write_t zvol_write;
214 static d_ioctl_t zvol_d_ioctl;
215 static d_strategy_t zvol_strategy;
217 static struct cdevsw zvol_cdevsw = {
218 .d_version = D_VERSION,
219 .d_open = zvol_d_open,
220 .d_close = zvol_d_close,
222 .d_write = zvol_write,
223 .d_ioctl = zvol_d_ioctl,
224 .d_strategy = zvol_strategy,
226 .d_flags = D_DISK | D_TRACKCLOSE,
229 static void zvol_geom_run(zvol_state_t *zv);
230 static void zvol_geom_destroy(zvol_state_t *zv);
231 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
232 static void zvol_geom_start(struct bio *bp);
233 static void zvol_geom_worker(void *arg);
234 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
235 uint64_t len, boolean_t sync);
236 #endif /* !illumos */
238 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
239 nvlist_t *, nvlist_t *);
240 static int zvol_remove_zv(zvol_state_t *);
241 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
242 static int zvol_dumpify(zvol_state_t *zv);
243 static int zvol_dump_fini(zvol_state_t *zv);
244 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
247 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
250 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
252 zv->zv_volsize = volsize;
253 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
254 "Size", volsize) == DDI_SUCCESS);
255 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
256 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
258 /* Notify specfs to invalidate the cached size */
259 spec_size_invalidate(dev, VBLK);
260 spec_size_invalidate(dev, VCHR);
262 zv->zv_volsize = volsize;
263 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
264 struct g_provider *pp;
266 pp = zv->zv_provider;
270 g_resize_provider(pp, zv->zv_volsize);
277 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
280 return (SET_ERROR(EINVAL));
282 if (volsize % blocksize != 0)
283 return (SET_ERROR(EINVAL));
286 if (volsize - 1 > SPEC_MAXOFFSET_T)
287 return (SET_ERROR(EOVERFLOW));
293 zvol_check_volblocksize(uint64_t volblocksize)
295 if (volblocksize < SPA_MINBLOCKSIZE ||
296 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
298 return (SET_ERROR(EDOM));
304 zvol_get_stats(objset_t *os, nvlist_t *nv)
307 dmu_object_info_t doi;
310 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
314 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
316 error = dmu_object_info(os, ZVOL_OBJ, &doi);
319 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
320 doi.doi_data_block_size);
326 static zvol_state_t *
327 zvol_minor_lookup(const char *name)
334 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
337 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
338 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
342 LIST_FOREACH(zv, &all_zvols, zv_links) {
344 if (strcmp(zv->zv_name, name) == 0)
351 /* extent mapping arg */
359 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
360 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
362 struct maparg *ma = arg;
364 int bs = ma->ma_zv->zv_volblocksize;
366 if (bp == NULL || BP_IS_HOLE(bp) ||
367 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
370 VERIFY(!BP_IS_EMBEDDED(bp));
372 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
375 /* Abort immediately if we have encountered gang blocks */
377 return (SET_ERROR(EFRAGS));
380 * See if the block is at the end of the previous extent.
382 ze = list_tail(&ma->ma_zv->zv_extents);
384 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
385 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
386 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
391 dprintf_bp(bp, "%s", "next blkptr:");
393 /* start a new extent */
394 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
395 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
397 list_insert_tail(&ma->ma_zv->zv_extents, ze);
402 zvol_free_extents(zvol_state_t *zv)
406 while (ze = list_head(&zv->zv_extents)) {
407 list_remove(&zv->zv_extents, ze);
408 kmem_free(ze, sizeof (zvol_extent_t));
413 zvol_get_lbas(zvol_state_t *zv)
415 objset_t *os = zv->zv_objset;
421 zvol_free_extents(zv);
423 /* commit any in-flight changes before traversing the dataset */
424 txg_wait_synced(dmu_objset_pool(os), 0);
425 err = traverse_dataset(dmu_objset_ds(os), 0,
426 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
427 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
428 zvol_free_extents(zv);
429 return (err ? err : EIO);
437 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
439 zfs_creat_t *zct = arg;
440 nvlist_t *nvprops = zct->zct_props;
442 uint64_t volblocksize, volsize;
444 VERIFY(nvlist_lookup_uint64(nvprops,
445 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
446 if (nvlist_lookup_uint64(nvprops,
447 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
448 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
451 * These properties must be removed from the list so the generic
452 * property setting step won't apply to them.
454 VERIFY(nvlist_remove_all(nvprops,
455 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
456 (void) nvlist_remove_all(nvprops,
457 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
459 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
463 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
467 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
472 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
473 * implement DKIOCFREE/free-long-range.
476 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
478 uint64_t offset, length;
481 byteswap_uint64_array(lr, sizeof (*lr));
483 offset = lr->lr_offset;
484 length = lr->lr_length;
486 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
490 * Replay a TX_WRITE ZIL transaction that didn't get committed
491 * after a system failure
494 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
496 objset_t *os = zv->zv_objset;
497 char *data = (char *)(lr + 1); /* data follows lr_write_t */
498 uint64_t offset, length;
503 byteswap_uint64_array(lr, sizeof (*lr));
505 offset = lr->lr_offset;
506 length = lr->lr_length;
508 /* If it's a dmu_sync() block, write the whole block */
509 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
510 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
511 if (length < blocksize) {
512 offset -= offset % blocksize;
517 tx = dmu_tx_create(os);
518 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
519 error = dmu_tx_assign(tx, TXG_WAIT);
523 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
532 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
534 return (SET_ERROR(ENOTSUP));
538 * Callback vectors for replaying records.
539 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
541 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
542 zvol_replay_err, /* 0 no such transaction type */
543 zvol_replay_err, /* TX_CREATE */
544 zvol_replay_err, /* TX_MKDIR */
545 zvol_replay_err, /* TX_MKXATTR */
546 zvol_replay_err, /* TX_SYMLINK */
547 zvol_replay_err, /* TX_REMOVE */
548 zvol_replay_err, /* TX_RMDIR */
549 zvol_replay_err, /* TX_LINK */
550 zvol_replay_err, /* TX_RENAME */
551 zvol_replay_write, /* TX_WRITE */
552 zvol_replay_truncate, /* TX_TRUNCATE */
553 zvol_replay_err, /* TX_SETATTR */
554 zvol_replay_err, /* TX_ACL */
555 zvol_replay_err, /* TX_CREATE_ACL */
556 zvol_replay_err, /* TX_CREATE_ATTR */
557 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
558 zvol_replay_err, /* TX_MKDIR_ACL */
559 zvol_replay_err, /* TX_MKDIR_ATTR */
560 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
561 zvol_replay_err, /* TX_WRITE2 */
566 zvol_name2minor(const char *name, minor_t *minor)
570 mutex_enter(&zfsdev_state_lock);
571 zv = zvol_minor_lookup(name);
573 *minor = zv->zv_minor;
574 mutex_exit(&zfsdev_state_lock);
575 return (zv ? 0 : -1);
580 * Create a minor node (plus a whole lot more) for the specified volume.
583 zvol_create_minor(const char *name)
585 zfs_soft_state_t *zs;
589 dmu_object_info_t doi;
591 char chrbuf[30], blkbuf[30];
593 struct g_provider *pp;
600 ZFS_LOG(1, "Creating ZVOL %s...", name);
603 mutex_enter(&zfsdev_state_lock);
605 if (zvol_minor_lookup(name) != NULL) {
606 mutex_exit(&zfsdev_state_lock);
607 return (SET_ERROR(EEXIST));
610 /* lie and say we're read-only */
611 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
614 mutex_exit(&zfsdev_state_lock);
619 if ((minor = zfsdev_minor_alloc()) == 0) {
620 dmu_objset_disown(os, FTAG);
621 mutex_exit(&zfsdev_state_lock);
622 return (SET_ERROR(ENXIO));
625 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
626 dmu_objset_disown(os, FTAG);
627 mutex_exit(&zfsdev_state_lock);
628 return (SET_ERROR(EAGAIN));
630 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
633 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
635 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
636 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
637 ddi_soft_state_free(zfsdev_state, minor);
638 dmu_objset_disown(os, FTAG);
639 mutex_exit(&zfsdev_state_lock);
640 return (SET_ERROR(EAGAIN));
643 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
645 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
646 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
647 ddi_remove_minor_node(zfs_dip, chrbuf);
648 ddi_soft_state_free(zfsdev_state, minor);
649 dmu_objset_disown(os, FTAG);
650 mutex_exit(&zfsdev_state_lock);
651 return (SET_ERROR(EAGAIN));
654 zs = ddi_get_soft_state(zfsdev_state, minor);
655 zs->zss_type = ZSST_ZVOL;
656 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
659 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
661 error = dsl_prop_get_integer(name,
662 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
663 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
667 zv->zv_volmode = mode;
668 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
670 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
671 gp->start = zvol_geom_start;
672 gp->access = zvol_geom_access;
673 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
674 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
675 pp->sectorsize = DEV_BSIZE;
679 zv->zv_provider = pp;
680 bioq_init(&zv->zv_queue);
681 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
682 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
683 struct make_dev_args args;
685 make_dev_args_init(&args);
686 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
687 args.mda_devsw = &zvol_cdevsw;
689 args.mda_uid = UID_ROOT;
690 args.mda_gid = GID_OPERATOR;
691 args.mda_mode = 0640;
692 args.mda_si_drv2 = zv;
693 error = make_dev_s(&args, &zv->zv_dev,
694 "%s/%s", ZVOL_DRIVER, name);
696 kmem_free(zv, sizeof(*zv));
697 dmu_objset_disown(os, FTAG);
698 mutex_exit(&zfsdev_state_lock);
701 zv->zv_dev->si_iosize_max = MAXPHYS;
703 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
706 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
707 zv->zv_min_bs = DEV_BSHIFT;
709 zv->zv_minor = minor;
712 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
713 zv->zv_flags |= ZVOL_RDONLY;
714 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
715 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
716 sizeof (rl_t), offsetof(rl_t, r_node));
717 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
718 offsetof(zvol_extent_t, ze_node));
720 /* get and cache the blocksize */
721 error = dmu_object_info(os, ZVOL_OBJ, &doi);
723 zv->zv_volblocksize = doi.doi_data_block_size;
726 if (spa_writeable(dmu_objset_spa(os))) {
727 if (zil_replay_disable)
728 zil_destroy(dmu_objset_zil(os), B_FALSE);
730 zil_replay(os, zv, zvol_replay_vector);
732 dmu_objset_disown(os, FTAG);
733 zv->zv_objset = NULL;
737 mutex_exit(&zfsdev_state_lock);
739 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
745 ZFS_LOG(1, "ZVOL %s created.", name);
752 * Remove minor node for the specified volume.
755 zvol_remove_zv(zvol_state_t *zv)
759 minor_t minor = zv->zv_minor;
762 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
763 if (zv->zv_total_opens != 0)
764 return (SET_ERROR(EBUSY));
767 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
768 ddi_remove_minor_node(zfs_dip, nmbuf);
770 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
771 ddi_remove_minor_node(zfs_dip, nmbuf);
773 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
775 LIST_REMOVE(zv, zv_links);
776 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
778 zvol_geom_destroy(zv);
780 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
781 if (zv->zv_dev != NULL)
782 destroy_dev(zv->zv_dev);
786 avl_destroy(&zv->zv_znode.z_range_avl);
787 mutex_destroy(&zv->zv_znode.z_range_lock);
789 kmem_free(zv, sizeof (zvol_state_t));
791 ddi_soft_state_free(zfsdev_state, minor);
798 zvol_remove_minor(const char *name)
803 mutex_enter(&zfsdev_state_lock);
804 if ((zv = zvol_minor_lookup(name)) == NULL) {
805 mutex_exit(&zfsdev_state_lock);
806 return (SET_ERROR(ENXIO));
808 rc = zvol_remove_zv(zv);
809 mutex_exit(&zfsdev_state_lock);
814 zvol_first_open(zvol_state_t *zv)
816 dmu_object_info_t doi;
822 /* lie and say we're read-only */
823 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
829 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
832 dmu_objset_disown(os, zvol_tag);
836 /* get and cache the blocksize */
837 error = dmu_object_info(os, ZVOL_OBJ, &doi);
840 dmu_objset_disown(os, zvol_tag);
843 zv->zv_volblocksize = doi.doi_data_block_size;
845 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
847 dmu_objset_disown(os, zvol_tag);
851 zvol_size_changed(zv, volsize);
852 zv->zv_zilog = zil_open(os, zvol_get_data);
854 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
856 if (readonly || dmu_objset_is_snapshot(os) ||
857 !spa_writeable(dmu_objset_spa(os)))
858 zv->zv_flags |= ZVOL_RDONLY;
860 zv->zv_flags &= ~ZVOL_RDONLY;
865 zvol_last_close(zvol_state_t *zv)
867 zil_close(zv->zv_zilog);
870 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
876 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
877 !(zv->zv_flags & ZVOL_RDONLY))
878 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
879 dmu_objset_evict_dbufs(zv->zv_objset);
881 dmu_objset_disown(zv->zv_objset, zvol_tag);
882 zv->zv_objset = NULL;
887 zvol_prealloc(zvol_state_t *zv)
889 objset_t *os = zv->zv_objset;
891 uint64_t refd, avail, usedobjs, availobjs;
892 uint64_t resid = zv->zv_volsize;
895 /* Check the space usage before attempting to allocate the space */
896 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
897 if (avail < zv->zv_volsize)
898 return (SET_ERROR(ENOSPC));
900 /* Free old extents if they exist */
901 zvol_free_extents(zv);
905 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
907 tx = dmu_tx_create(os);
908 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
909 error = dmu_tx_assign(tx, TXG_WAIT);
912 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
915 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
920 txg_wait_synced(dmu_objset_pool(os), 0);
927 zvol_update_volsize(objset_t *os, uint64_t volsize)
932 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
934 tx = dmu_tx_create(os);
935 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
936 dmu_tx_mark_netfree(tx);
937 error = dmu_tx_assign(tx, TXG_WAIT);
943 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
948 error = dmu_free_long_range(os,
949 ZVOL_OBJ, volsize, DMU_OBJECT_END);
954 zvol_remove_minors(const char *name)
961 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
962 (void) strncpy(namebuf, name, strlen(name));
963 (void) strcat(namebuf, "/");
964 mutex_enter(&zfsdev_state_lock);
965 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
967 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
970 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
971 (void) zvol_remove_zv(zv);
973 kmem_free(namebuf, strlen(name) + 2);
975 mutex_exit(&zfsdev_state_lock);
977 zvol_state_t *zv, *tzv;
980 namelen = strlen(name);
983 mutex_enter(&zfsdev_state_lock);
985 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
986 if (strcmp(zv->zv_name, name) == 0 ||
987 (strncmp(zv->zv_name, name, namelen) == 0 &&
988 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
989 zv->zv_name[namelen] == '@'))) {
990 (void) zvol_remove_zv(zv);
994 mutex_exit(&zfsdev_state_lock);
1000 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
1002 uint64_t old_volsize = 0ULL;
1005 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1008 * Reinitialize the dump area to the new size. If we
1009 * failed to resize the dump area then restore it back to
1010 * its original size. We must set the new volsize prior
1011 * to calling dumpvp_resize() to ensure that the devices'
1012 * size(9P) is not visible by the dump subsystem.
1014 old_volsize = zv->zv_volsize;
1015 zvol_size_changed(zv, volsize);
1018 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1019 if ((error = zvol_dumpify(zv)) != 0 ||
1020 (error = dumpvp_resize()) != 0) {
1023 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1024 zvol_size_changed(zv, old_volsize);
1025 dumpify_error = zvol_dumpify(zv);
1026 error = dumpify_error ? dumpify_error : error;
1029 #endif /* ZVOL_DUMP */
1033 * Generate a LUN expansion event.
1038 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1040 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1043 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1044 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1046 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1047 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1050 kmem_free(physpath, MAXPATHLEN);
1052 #endif /* illumos */
1057 zvol_set_volsize(const char *name, uint64_t volsize)
1059 zvol_state_t *zv = NULL;
1062 dmu_object_info_t doi;
1064 boolean_t owned = B_FALSE;
1066 error = dsl_prop_get_integer(name,
1067 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1071 return (SET_ERROR(EROFS));
1073 mutex_enter(&zfsdev_state_lock);
1074 zv = zvol_minor_lookup(name);
1076 if (zv == NULL || zv->zv_objset == NULL) {
1077 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1079 mutex_exit(&zfsdev_state_lock);
1089 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1090 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1093 error = zvol_update_volsize(os, volsize);
1095 if (error == 0 && zv != NULL)
1096 error = zvol_update_live_volsize(zv, volsize);
1099 dmu_objset_disown(os, FTAG);
1101 zv->zv_objset = NULL;
1103 mutex_exit(&zfsdev_state_lock);
1110 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1113 zvol_open(struct g_provider *pp, int flag, int count)
1120 mutex_enter(&zfsdev_state_lock);
1122 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1124 mutex_exit(&zfsdev_state_lock);
1125 return (SET_ERROR(ENXIO));
1128 if (zv->zv_total_opens == 0)
1129 err = zvol_first_open(zv);
1131 mutex_exit(&zfsdev_state_lock);
1134 #else /* !illumos */
1135 boolean_t locked = B_FALSE;
1137 if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1139 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1140 * attempting to probe geom providers while looking for a
1141 * replacement for a missing VDEV. In this case, the
1142 * spa_namespace_lock will not be held, but it is still illegal
1143 * to use a zvol as a vdev. Deadlocks can result if another
1144 * thread has spa_namespace_lock
1146 return (EOPNOTSUPP);
1149 * Protect against recursively entering spa_namespace_lock
1150 * when spa_open() is used for a pool on a (local) ZVOL(s).
1151 * This is needed since we replaced upstream zfsdev_state_lock
1152 * with spa_namespace_lock in the ZVOL code.
1153 * We are using the same trick as spa_open().
1154 * Note that calls in zvol_first_open which need to resolve
1155 * pool name to a spa object will enter spa_open()
1156 * recursively, but that function already has all the
1157 * necessary protection.
1159 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1160 mutex_enter(&zfsdev_state_lock);
1167 mutex_exit(&zfsdev_state_lock);
1168 return (SET_ERROR(ENXIO));
1171 if (zv->zv_total_opens == 0) {
1172 err = zvol_first_open(zv);
1175 mutex_exit(&zfsdev_state_lock);
1178 pp->mediasize = zv->zv_volsize;
1179 pp->stripeoffset = 0;
1180 pp->stripesize = zv->zv_volblocksize;
1182 #endif /* illumos */
1183 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1184 err = SET_ERROR(EROFS);
1187 if (zv->zv_flags & ZVOL_EXCL) {
1188 err = SET_ERROR(EBUSY);
1193 if (zv->zv_total_opens != 0) {
1194 err = SET_ERROR(EBUSY);
1197 zv->zv_flags |= ZVOL_EXCL;
1202 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1203 zv->zv_open_count[otyp]++;
1204 zv->zv_total_opens++;
1206 mutex_exit(&zfsdev_state_lock);
1208 zv->zv_total_opens += count;
1210 mutex_exit(&zfsdev_state_lock);
1215 if (zv->zv_total_opens == 0)
1216 zvol_last_close(zv);
1218 mutex_exit(&zfsdev_state_lock);
1221 mutex_exit(&zfsdev_state_lock);
1229 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1231 minor_t minor = getminor(dev);
1235 mutex_enter(&zfsdev_state_lock);
1237 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1239 mutex_exit(&zfsdev_state_lock);
1240 #else /* !illumos */
1242 zvol_close(struct g_provider *pp, int flag, int count)
1246 boolean_t locked = B_FALSE;
1248 /* See comment in zvol_open(). */
1249 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1250 mutex_enter(&zfsdev_state_lock);
1257 mutex_exit(&zfsdev_state_lock);
1258 #endif /* illumos */
1259 return (SET_ERROR(ENXIO));
1262 if (zv->zv_flags & ZVOL_EXCL) {
1263 ASSERT(zv->zv_total_opens == 1);
1264 zv->zv_flags &= ~ZVOL_EXCL;
1268 * If the open count is zero, this is a spurious close.
1269 * That indicates a bug in the kernel / DDI framework.
1272 ASSERT(zv->zv_open_count[otyp] != 0);
1274 ASSERT(zv->zv_total_opens != 0);
1277 * You may get multiple opens, but only one close.
1280 zv->zv_open_count[otyp]--;
1281 zv->zv_total_opens--;
1283 zv->zv_total_opens -= count;
1286 if (zv->zv_total_opens == 0)
1287 zvol_last_close(zv);
1290 mutex_exit(&zfsdev_state_lock);
1293 mutex_exit(&zfsdev_state_lock);
1299 zvol_get_done(zgd_t *zgd, int error)
1302 dmu_buf_rele(zgd->zgd_db, zgd);
1304 zfs_range_unlock(zgd->zgd_rl);
1306 if (error == 0 && zgd->zgd_bp)
1307 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1309 kmem_free(zgd, sizeof (zgd_t));
1313 * Get data to generate a TX_WRITE intent log record.
1316 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1318 zvol_state_t *zv = arg;
1319 objset_t *os = zv->zv_objset;
1320 uint64_t object = ZVOL_OBJ;
1321 uint64_t offset = lr->lr_offset;
1322 uint64_t size = lr->lr_length; /* length of user data */
1323 blkptr_t *bp = &lr->lr_blkptr;
1328 ASSERT(zio != NULL);
1331 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1332 zgd->zgd_zilog = zv->zv_zilog;
1333 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1336 * Write records come in two flavors: immediate and indirect.
1337 * For small writes it's cheaper to store the data with the
1338 * log record (immediate); for large writes it's cheaper to
1339 * sync the data and get a pointer to it (indirect) so that
1340 * we don't have to write the data twice.
1342 if (buf != NULL) { /* immediate write */
1343 error = dmu_read(os, object, offset, size, buf,
1344 DMU_READ_NO_PREFETCH);
1346 size = zv->zv_volblocksize;
1347 offset = P2ALIGN(offset, size);
1348 error = dmu_buf_hold(os, object, offset, zgd, &db,
1349 DMU_READ_NO_PREFETCH);
1351 blkptr_t *obp = dmu_buf_get_blkptr(db);
1353 ASSERT(BP_IS_HOLE(bp));
1360 ASSERT(db->db_offset == offset);
1361 ASSERT(db->db_size == size);
1363 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1364 zvol_get_done, zgd);
1371 zvol_get_done(zgd, error);
1377 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1379 * We store data in the log buffers if it's small enough.
1380 * Otherwise we will later flush the data out via dmu_sync().
1382 ssize_t zvol_immediate_write_sz = 32768;
1384 SYSCTL_LONG(_vfs_zfs_vol, OID_AUTO, immediate_write_sz, CTLFLAG_RWTUN,
1385 &zvol_immediate_write_sz, 0, "Minimal size for indirect log write");
1389 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1392 uint32_t blocksize = zv->zv_volblocksize;
1393 zilog_t *zilog = zv->zv_zilog;
1394 itx_wr_state_t write_state;
1396 if (zil_replaying(zilog, tx))
1399 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1400 write_state = WR_INDIRECT;
1401 else if (!spa_has_slogs(zilog->zl_spa) &&
1402 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1403 write_state = WR_INDIRECT;
1405 write_state = WR_COPIED;
1407 write_state = WR_NEED_COPY;
1412 itx_wr_state_t wr_state = write_state;
1413 ssize_t len = resid;
1415 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1416 wr_state = WR_NEED_COPY;
1417 else if (wr_state == WR_INDIRECT)
1418 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1420 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1421 (wr_state == WR_COPIED ? len : 0));
1422 lr = (lr_write_t *)&itx->itx_lr;
1423 if (wr_state == WR_COPIED && dmu_read(zv->zv_objset,
1424 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1425 zil_itx_destroy(itx);
1426 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1427 lr = (lr_write_t *)&itx->itx_lr;
1428 wr_state = WR_NEED_COPY;
1431 itx->itx_wr_state = wr_state;
1432 lr->lr_foid = ZVOL_OBJ;
1433 lr->lr_offset = off;
1434 lr->lr_length = len;
1436 BP_ZERO(&lr->lr_blkptr);
1438 itx->itx_private = zv;
1440 if (!sync && (zv->zv_sync_cnt == 0))
1441 itx->itx_sync = B_FALSE;
1443 zil_itx_assign(zilog, itx, tx);
1452 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1453 uint64_t size, boolean_t doread, boolean_t isdump)
1459 if (vd->vdev_ops == &vdev_mirror_ops ||
1460 vd->vdev_ops == &vdev_replacing_ops ||
1461 vd->vdev_ops == &vdev_spare_ops) {
1462 for (c = 0; c < vd->vdev_children; c++) {
1463 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1464 addr, offset, origoffset, size, doread, isdump);
1467 } else if (doread) {
1473 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1474 return (numerrors < vd->vdev_children ? 0 : EIO);
1476 if (doread && !vdev_readable(vd))
1477 return (SET_ERROR(EIO));
1478 else if (!doread && !vdev_writeable(vd))
1479 return (SET_ERROR(EIO));
1481 if (vd->vdev_ops == &vdev_raidz_ops) {
1482 return (vdev_raidz_physio(vd,
1483 addr, size, offset, origoffset, doread, isdump));
1486 offset += VDEV_LABEL_START_SIZE;
1488 if (ddi_in_panic() || isdump) {
1491 return (SET_ERROR(EIO));
1493 ASSERT3P(dvd, !=, NULL);
1494 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1498 ASSERT3P(dvd, !=, NULL);
1499 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1500 offset, doread ? B_READ : B_WRITE));
1505 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1506 boolean_t doread, boolean_t isdump)
1511 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1513 /* Must be sector aligned, and not stradle a block boundary. */
1514 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1515 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1516 return (SET_ERROR(EINVAL));
1518 ASSERT(size <= zv->zv_volblocksize);
1520 /* Locate the extent this belongs to */
1521 ze = list_head(&zv->zv_extents);
1522 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1523 offset -= ze->ze_nblks * zv->zv_volblocksize;
1524 ze = list_next(&zv->zv_extents, ze);
1528 return (SET_ERROR(EINVAL));
1530 if (!ddi_in_panic())
1531 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1533 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1534 offset += DVA_GET_OFFSET(&ze->ze_dva);
1535 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1536 size, doread, isdump);
1538 if (!ddi_in_panic())
1539 spa_config_exit(spa, SCL_STATE, FTAG);
1545 zvol_strategy(buf_t *bp)
1547 zfs_soft_state_t *zs = NULL;
1548 #else /* !illumos */
1550 zvol_strategy(struct bio *bp)
1552 #endif /* illumos */
1554 uint64_t off, volsize;
1561 boolean_t doread = bp->b_flags & B_READ;
1563 boolean_t doread = 0;
1565 boolean_t is_dumpified;
1569 if (getminor(bp->b_edev) == 0) {
1570 error = SET_ERROR(EINVAL);
1572 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1574 error = SET_ERROR(ENXIO);
1575 else if (zs->zss_type != ZSST_ZVOL)
1576 error = SET_ERROR(EINVAL);
1580 bioerror(bp, error);
1587 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1588 bioerror(bp, EROFS);
1593 off = ldbtob(bp->b_blkno);
1594 #else /* !illumos */
1596 zv = bp->bio_to->private;
1598 zv = bp->bio_dev->si_drv2;
1601 error = SET_ERROR(ENXIO);
1605 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1606 error = SET_ERROR(EROFS);
1610 switch (bp->bio_cmd) {
1623 off = bp->bio_offset;
1624 #endif /* illumos */
1625 volsize = zv->zv_volsize;
1632 addr = bp->b_un.b_addr;
1633 resid = bp->b_bcount;
1635 if (resid > 0 && (off < 0 || off >= volsize)) {
1641 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1642 sync = ((!(bp->b_flags & B_ASYNC) &&
1643 !(zv->zv_flags & ZVOL_WCE)) ||
1644 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1645 !doread && !is_dumpified;
1646 #else /* !illumos */
1647 addr = bp->bio_data;
1648 resid = bp->bio_length;
1650 if (resid > 0 && (off < 0 || off >= volsize)) {
1651 error = SET_ERROR(EIO);
1655 is_dumpified = B_FALSE;
1656 sync = !doread && !is_dumpified &&
1657 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1658 #endif /* illumos */
1661 * There must be no buffer changes when doing a dmu_sync() because
1662 * we can't change the data whilst calculating the checksum.
1664 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1665 doread ? RL_READER : RL_WRITER);
1668 if (bp->bio_cmd == BIO_DELETE) {
1669 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1670 error = dmu_tx_assign(tx, TXG_WAIT);
1674 zvol_log_truncate(zv, tx, off, resid, sync);
1676 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1683 while (resid != 0 && off < volsize) {
1684 size_t size = MIN(resid, zvol_maxphys);
1687 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1688 error = zvol_dumpio(zv, addr, off, size,
1690 } else if (doread) {
1694 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1697 dmu_tx_t *tx = dmu_tx_create(os);
1698 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1699 error = dmu_tx_assign(tx, TXG_WAIT);
1703 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1704 zvol_log_write(zv, tx, off, size, sync);
1709 /* convert checksum errors into IO errors */
1710 if (error == ECKSUM)
1711 error = SET_ERROR(EIO);
1721 zfs_range_unlock(rl);
1724 if ((bp->b_resid = resid) == bp->b_bcount)
1725 bioerror(bp, off > volsize ? EINVAL : error);
1728 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1732 #else /* !illumos */
1733 bp->bio_completed = bp->bio_length - resid;
1734 if (bp->bio_completed < bp->bio_length && off > volsize)
1739 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1743 g_io_deliver(bp, error);
1745 biofinish(bp, NULL, error);
1746 #endif /* illumos */
1751 * Set the buffer count to the zvol maximum transfer.
1752 * Using our own routine instead of the default minphys()
1753 * means that for larger writes we write bigger buffers on X86
1754 * (128K instead of 56K) and flush the disk write cache less often
1755 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1756 * 56K on X86 and 128K on sparc).
1759 zvol_minphys(struct buf *bp)
1761 if (bp->b_bcount > zvol_maxphys)
1762 bp->b_bcount = zvol_maxphys;
1766 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1768 minor_t minor = getminor(dev);
1775 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1777 return (SET_ERROR(ENXIO));
1779 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1780 return (SET_ERROR(EINVAL));
1782 boff = ldbtob(blkno);
1783 resid = ldbtob(nblocks);
1785 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1788 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1789 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1802 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1804 minor_t minor = getminor(dev);
1805 #else /* !illumos */
1807 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1809 #endif /* illumos */
1816 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1818 return (SET_ERROR(ENXIO));
1823 volsize = zv->zv_volsize;
1824 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1825 if (uio->uio_resid > 0 &&
1826 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1827 return (SET_ERROR(EIO));
1830 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1831 error = physio(zvol_strategy, NULL, dev, B_READ,
1837 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1839 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1840 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1842 /* don't read past the end */
1843 if (bytes > volsize - uio->uio_loffset)
1844 bytes = volsize - uio->uio_loffset;
1846 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
1848 /* convert checksum errors into IO errors */
1849 if (error == ECKSUM)
1850 error = SET_ERROR(EIO);
1854 zfs_range_unlock(rl);
1861 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1863 minor_t minor = getminor(dev);
1864 #else /* !illumos */
1866 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1868 #endif /* illumos */
1876 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1878 return (SET_ERROR(ENXIO));
1883 volsize = zv->zv_volsize;
1884 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1885 if (uio->uio_resid > 0 &&
1886 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1887 return (SET_ERROR(EIO));
1890 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1891 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1896 sync = !(zv->zv_flags & ZVOL_WCE) ||
1898 sync = (ioflag & IO_SYNC) ||
1900 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1902 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1904 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1905 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1906 uint64_t off = uio->uio_loffset;
1907 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1909 if (bytes > volsize - off) /* don't write past the end */
1910 bytes = volsize - off;
1912 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1913 error = dmu_tx_assign(tx, TXG_WAIT);
1918 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1920 zvol_log_write(zv, tx, off, bytes, sync);
1926 zfs_range_unlock(rl);
1928 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1934 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1936 struct uuid uuid = EFI_RESERVED;
1937 efi_gpe_t gpe = { 0 };
1943 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1944 return (SET_ERROR(EFAULT));
1945 ptr = (char *)(uintptr_t)efi.dki_data_64;
1946 length = efi.dki_length;
1948 * Some clients may attempt to request a PMBR for the
1949 * zvol. Currently this interface will return EINVAL to
1950 * such requests. These requests could be supported by
1951 * adding a check for lba == 0 and consing up an appropriate
1954 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1955 return (SET_ERROR(EINVAL));
1957 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1958 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1959 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1961 if (efi.dki_lba == 1) {
1962 efi_gpt_t gpt = { 0 };
1964 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1965 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1966 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1967 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1968 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1969 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1970 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1971 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1972 gpt.efi_gpt_SizeOfPartitionEntry =
1973 LE_32(sizeof (efi_gpe_t));
1974 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1975 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1976 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1977 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1978 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1980 return (SET_ERROR(EFAULT));
1981 ptr += sizeof (gpt);
1982 length -= sizeof (gpt);
1984 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1986 return (SET_ERROR(EFAULT));
1991 * BEGIN entry points to allow external callers access to the volume.
1994 * Return the volume parameters needed for access from an external caller.
1995 * These values are invariant as long as the volume is held open.
1998 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1999 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
2000 void **rl_hdl, void **bonus_hdl)
2004 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2006 return (SET_ERROR(ENXIO));
2007 if (zv->zv_flags & ZVOL_DUMPIFIED)
2008 return (SET_ERROR(ENXIO));
2010 ASSERT(blksize && max_xfer_len && minor_hdl &&
2011 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
2013 *blksize = zv->zv_volblocksize;
2014 *max_xfer_len = (uint64_t)zvol_maxphys;
2016 *objset_hdl = zv->zv_objset;
2017 *zil_hdl = zv->zv_zilog;
2018 *rl_hdl = &zv->zv_znode;
2019 *bonus_hdl = zv->zv_dbuf;
2024 * Return the current volume size to an external caller.
2025 * The size can change while the volume is open.
2028 zvol_get_volume_size(void *minor_hdl)
2030 zvol_state_t *zv = minor_hdl;
2032 return (zv->zv_volsize);
2036 * Return the current WCE setting to an external caller.
2037 * The WCE setting can change while the volume is open.
2040 zvol_get_volume_wce(void *minor_hdl)
2042 zvol_state_t *zv = minor_hdl;
2044 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2048 * Entry point for external callers to zvol_log_write
2051 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2054 zvol_state_t *zv = minor_hdl;
2056 zvol_log_write(zv, tx, off, resid, sync);
2059 * END entry points to allow external callers access to the volume.
2061 #endif /* illumos */
2064 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2067 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2072 zilog_t *zilog = zv->zv_zilog;
2074 if (zil_replaying(zilog, tx))
2077 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2078 lr = (lr_truncate_t *)&itx->itx_lr;
2079 lr->lr_foid = ZVOL_OBJ;
2080 lr->lr_offset = off;
2081 lr->lr_length = len;
2083 itx->itx_sync = (sync || zv->zv_sync_cnt != 0);
2084 zil_itx_assign(zilog, itx, tx);
2089 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2090 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2094 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2097 struct dk_callback *dkc;
2101 mutex_enter(&zfsdev_state_lock);
2103 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2106 mutex_exit(&zfsdev_state_lock);
2107 return (SET_ERROR(ENXIO));
2109 ASSERT(zv->zv_total_opens > 0);
2115 struct dk_cinfo dki;
2117 bzero(&dki, sizeof (dki));
2118 (void) strcpy(dki.dki_cname, "zvol");
2119 (void) strcpy(dki.dki_dname, "zvol");
2120 dki.dki_ctype = DKC_UNKNOWN;
2121 dki.dki_unit = getminor(dev);
2122 dki.dki_maxtransfer =
2123 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2124 mutex_exit(&zfsdev_state_lock);
2125 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2126 error = SET_ERROR(EFAULT);
2130 case DKIOCGMEDIAINFO:
2132 struct dk_minfo dkm;
2134 bzero(&dkm, sizeof (dkm));
2135 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2136 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2137 dkm.dki_media_type = DK_UNKNOWN;
2138 mutex_exit(&zfsdev_state_lock);
2139 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2140 error = SET_ERROR(EFAULT);
2144 case DKIOCGMEDIAINFOEXT:
2146 struct dk_minfo_ext dkmext;
2148 bzero(&dkmext, sizeof (dkmext));
2149 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2150 dkmext.dki_pbsize = zv->zv_volblocksize;
2151 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2152 dkmext.dki_media_type = DK_UNKNOWN;
2153 mutex_exit(&zfsdev_state_lock);
2154 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2155 error = SET_ERROR(EFAULT);
2161 uint64_t vs = zv->zv_volsize;
2162 uint8_t bs = zv->zv_min_bs;
2164 mutex_exit(&zfsdev_state_lock);
2165 error = zvol_getefi((void *)arg, flag, vs, bs);
2169 case DKIOCFLUSHWRITECACHE:
2170 dkc = (struct dk_callback *)arg;
2171 mutex_exit(&zfsdev_state_lock);
2172 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2173 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2174 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2181 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2182 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2184 error = SET_ERROR(EFAULT);
2190 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2192 error = SET_ERROR(EFAULT);
2196 zv->zv_flags |= ZVOL_WCE;
2197 mutex_exit(&zfsdev_state_lock);
2199 zv->zv_flags &= ~ZVOL_WCE;
2200 mutex_exit(&zfsdev_state_lock);
2201 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2209 * commands using these (like prtvtoc) expect ENOTSUP
2210 * since we're emulating an EFI label
2212 error = SET_ERROR(ENOTSUP);
2216 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2218 error = zvol_dumpify(zv);
2219 zfs_range_unlock(rl);
2223 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2225 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2227 error = zvol_dump_fini(zv);
2228 zfs_range_unlock(rl);
2236 if (!zvol_unmap_enabled)
2239 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2240 error = SET_ERROR(EFAULT);
2245 * Apply Postel's Law to length-checking. If they overshoot,
2246 * just blank out until the end, if there's a need to blank
2249 if (df.df_start >= zv->zv_volsize)
2250 break; /* No need to do anything... */
2252 mutex_exit(&zfsdev_state_lock);
2254 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2256 tx = dmu_tx_create(zv->zv_objset);
2257 dmu_tx_mark_netfree(tx);
2258 error = dmu_tx_assign(tx, TXG_WAIT);
2262 zvol_log_truncate(zv, tx, df.df_start,
2263 df.df_length, B_TRUE);
2265 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2266 df.df_start, df.df_length);
2269 zfs_range_unlock(rl);
2273 * If the write-cache is disabled or 'sync' property
2274 * is set to 'always' then treat this as a synchronous
2275 * operation (i.e. commit to zil).
2277 if (!(zv->zv_flags & ZVOL_WCE) ||
2278 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2279 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2282 * If the caller really wants synchronous writes, and
2283 * can't wait for them, don't return until the write
2286 if (df.df_flags & DF_WAIT_SYNC) {
2288 dmu_objset_pool(zv->zv_objset), 0);
2295 error = SET_ERROR(ENOTTY);
2299 mutex_exit(&zfsdev_state_lock);
2302 #endif /* illumos */
2307 return (zvol_minors != 0);
2313 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2316 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2318 ZFS_LOG(1, "ZVOL Initialized.");
2326 mutex_destroy(&zfsdev_state_lock);
2328 ddi_soft_state_fini(&zfsdev_state);
2329 ZFS_LOG(1, "ZVOL Deinitialized.");
2335 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2337 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2339 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2346 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2348 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2350 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2354 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2358 objset_t *os = zv->zv_objset;
2359 spa_t *spa = dmu_objset_spa(os);
2360 vdev_t *vd = spa->spa_root_vdev;
2361 nvlist_t *nv = NULL;
2362 uint64_t version = spa_version(spa);
2363 uint64_t checksum, compress, refresrv, vbs, dedup;
2365 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2366 ASSERT(vd->vdev_ops == &vdev_root_ops);
2368 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2372 /* wait for dmu_free_long_range to actually free the blocks */
2373 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2376 * If the pool on which the dump device is being initialized has more
2377 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2378 * enabled. If so, bump that feature's counter to indicate that the
2379 * feature is active. We also check the vdev type to handle the
2381 * # zpool create test raidz disk1 disk2 disk3
2382 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2383 * the raidz vdev itself has 3 children.
2385 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2386 if (!spa_feature_is_enabled(spa,
2387 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2388 return (SET_ERROR(ENOTSUP));
2389 (void) dsl_sync_task(spa_name(spa),
2390 zfs_mvdev_dump_feature_check,
2391 zfs_mvdev_dump_activate_feature_sync, NULL,
2392 2, ZFS_SPACE_CHECK_RESERVED);
2396 error = dsl_prop_get_integer(zv->zv_name,
2397 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2399 error = dsl_prop_get_integer(zv->zv_name,
2400 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2404 error = dsl_prop_get_integer(zv->zv_name,
2405 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2409 error = dsl_prop_get_integer(zv->zv_name,
2410 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2413 if (version >= SPA_VERSION_DEDUP && error == 0) {
2414 error = dsl_prop_get_integer(zv->zv_name,
2415 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2421 tx = dmu_tx_create(os);
2422 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2423 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2424 error = dmu_tx_assign(tx, TXG_WAIT);
2431 * If we are resizing the dump device then we only need to
2432 * update the refreservation to match the newly updated
2433 * zvolsize. Otherwise, we save off the original state of the
2434 * zvol so that we can restore them if the zvol is ever undumpified.
2437 error = zap_update(os, ZVOL_ZAP_OBJ,
2438 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2439 &zv->zv_volsize, tx);
2441 error = zap_update(os, ZVOL_ZAP_OBJ,
2442 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2445 error = zap_update(os, ZVOL_ZAP_OBJ,
2446 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2450 error = zap_update(os, ZVOL_ZAP_OBJ,
2451 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2455 error = zap_update(os, ZVOL_ZAP_OBJ,
2456 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2460 error = dmu_object_set_blocksize(
2461 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2463 if (version >= SPA_VERSION_DEDUP && error == 0) {
2464 error = zap_update(os, ZVOL_ZAP_OBJ,
2465 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2469 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2474 * We only need update the zvol's property if we are initializing
2475 * the dump area for the first time.
2477 if (error == 0 && !resize) {
2479 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2480 * function. Otherwise, use the old default -- OFF.
2482 checksum = spa_feature_is_active(spa,
2483 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2486 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2487 VERIFY(nvlist_add_uint64(nv,
2488 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2489 VERIFY(nvlist_add_uint64(nv,
2490 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2491 ZIO_COMPRESS_OFF) == 0);
2492 VERIFY(nvlist_add_uint64(nv,
2493 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2495 if (version >= SPA_VERSION_DEDUP) {
2496 VERIFY(nvlist_add_uint64(nv,
2497 zfs_prop_to_name(ZFS_PROP_DEDUP),
2498 ZIO_CHECKSUM_OFF) == 0);
2501 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2506 /* Allocate the space for the dump */
2508 error = zvol_prealloc(zv);
2513 zvol_dumpify(zvol_state_t *zv)
2516 uint64_t dumpsize = 0;
2518 objset_t *os = zv->zv_objset;
2520 if (zv->zv_flags & ZVOL_RDONLY)
2521 return (SET_ERROR(EROFS));
2523 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2524 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2525 boolean_t resize = (dumpsize > 0);
2527 if ((error = zvol_dump_init(zv, resize)) != 0) {
2528 (void) zvol_dump_fini(zv);
2534 * Build up our lba mapping.
2536 error = zvol_get_lbas(zv);
2538 (void) zvol_dump_fini(zv);
2542 tx = dmu_tx_create(os);
2543 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2544 error = dmu_tx_assign(tx, TXG_WAIT);
2547 (void) zvol_dump_fini(zv);
2551 zv->zv_flags |= ZVOL_DUMPIFIED;
2552 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2553 &zv->zv_volsize, tx);
2557 (void) zvol_dump_fini(zv);
2561 txg_wait_synced(dmu_objset_pool(os), 0);
2566 zvol_dump_fini(zvol_state_t *zv)
2569 objset_t *os = zv->zv_objset;
2572 uint64_t checksum, compress, refresrv, vbs, dedup;
2573 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2576 * Attempt to restore the zvol back to its pre-dumpified state.
2577 * This is a best-effort attempt as it's possible that not all
2578 * of these properties were initialized during the dumpify process
2579 * (i.e. error during zvol_dump_init).
2582 tx = dmu_tx_create(os);
2583 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2584 error = dmu_tx_assign(tx, TXG_WAIT);
2589 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2592 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2593 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2594 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2595 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2596 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2597 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2598 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2599 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2601 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2602 (void) nvlist_add_uint64(nv,
2603 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2604 (void) nvlist_add_uint64(nv,
2605 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2606 (void) nvlist_add_uint64(nv,
2607 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2608 if (version >= SPA_VERSION_DEDUP &&
2609 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2610 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2611 (void) nvlist_add_uint64(nv,
2612 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2614 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2618 zvol_free_extents(zv);
2619 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2620 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2621 /* wait for dmu_free_long_range to actually free the blocks */
2622 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2623 tx = dmu_tx_create(os);
2624 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2625 error = dmu_tx_assign(tx, TXG_WAIT);
2630 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2631 zv->zv_volblocksize = vbs;
2636 #else /* !illumos */
2639 zvol_geom_run(zvol_state_t *zv)
2641 struct g_provider *pp;
2643 pp = zv->zv_provider;
2644 g_error_provider(pp, 0);
2646 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2647 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2651 zvol_geom_destroy(zvol_state_t *zv)
2653 struct g_provider *pp;
2655 g_topology_assert();
2657 mtx_lock(&zv->zv_queue_mtx);
2659 wakeup_one(&zv->zv_queue);
2660 while (zv->zv_state != 2)
2661 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2662 mtx_destroy(&zv->zv_queue_mtx);
2664 pp = zv->zv_provider;
2665 zv->zv_provider = NULL;
2667 g_wither_geom(pp->geom, ENXIO);
2671 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2673 int count, error, flags;
2675 g_topology_assert();
2678 * To make it easier we expect either open or close, but not both
2681 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2682 (acr <= 0 && acw <= 0 && ace <= 0),
2683 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2684 pp->name, acr, acw, ace));
2686 if (pp->private == NULL) {
2687 if (acr <= 0 && acw <= 0 && ace <= 0)
2693 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2694 * because GEOM already handles that and handles it a bit differently.
2695 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2696 * only one exclusive consumer, no matter if it is reader or writer.
2697 * I like better the way GEOM works so I'll leave it for GEOM to
2698 * decide what to do.
2701 count = acr + acw + ace;
2706 if (acr != 0 || ace != 0)
2711 g_topology_unlock();
2713 error = zvol_open(pp, flags, count);
2715 error = zvol_close(pp, flags, -count);
2721 zvol_geom_start(struct bio *bp)
2726 zv = bp->bio_to->private;
2728 switch (bp->bio_cmd) {
2730 if (!THREAD_CAN_SLEEP())
2732 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2733 g_io_deliver(bp, 0);
2738 if (!THREAD_CAN_SLEEP())
2743 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2744 uint64_t refd, avail, usedobjs, availobjs, val;
2746 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2748 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2749 dmu_objset_space(zv->zv_objset, &refd, &avail,
2750 &usedobjs, &availobjs);
2751 if (g_handleattr_off_t(bp, "blocksavail",
2754 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2755 dmu_objset_space(zv->zv_objset, &refd, &avail,
2756 &usedobjs, &availobjs);
2757 if (g_handleattr_off_t(bp, "blocksused",
2760 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2761 avail = metaslab_class_get_space(spa_normal_class(spa));
2762 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2763 if (g_handleattr_off_t(bp, "poolblocksavail",
2766 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2767 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2768 if (g_handleattr_off_t(bp, "poolblocksused",
2775 g_io_deliver(bp, EOPNOTSUPP);
2781 mtx_lock(&zv->zv_queue_mtx);
2782 first = (bioq_first(&zv->zv_queue) == NULL);
2783 bioq_insert_tail(&zv->zv_queue, bp);
2784 mtx_unlock(&zv->zv_queue_mtx);
2786 wakeup_one(&zv->zv_queue);
2790 zvol_geom_worker(void *arg)
2795 thread_lock(curthread);
2796 sched_prio(curthread, PRIBIO);
2797 thread_unlock(curthread);
2801 mtx_lock(&zv->zv_queue_mtx);
2802 bp = bioq_takefirst(&zv->zv_queue);
2804 if (zv->zv_state == 1) {
2806 wakeup(&zv->zv_state);
2807 mtx_unlock(&zv->zv_queue_mtx);
2810 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2814 mtx_unlock(&zv->zv_queue_mtx);
2815 switch (bp->bio_cmd) {
2817 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2818 g_io_deliver(bp, 0);
2826 g_io_deliver(bp, EOPNOTSUPP);
2832 extern boolean_t dataset_name_hidden(const char *name);
2835 zvol_create_snapshots(objset_t *os, const char *name)
2837 uint64_t cookie, obj;
2842 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2845 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2850 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2851 if (len >= MAXPATHLEN) {
2852 dmu_objset_rele(os, FTAG);
2853 error = ENAMETOOLONG;
2857 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2858 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2859 sname + len, &obj, &cookie, NULL);
2860 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2862 if (error == ENOENT)
2867 error = zvol_create_minor(sname);
2868 if (error != 0 && error != EEXIST) {
2869 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2875 kmem_free(sname, MAXPATHLEN);
2880 zvol_create_minors(const char *name)
2887 if (dataset_name_hidden(name))
2890 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2891 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2895 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2896 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2897 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2898 error = zvol_create_minor(name);
2899 if (error == 0 || error == EEXIST) {
2900 error = zvol_create_snapshots(os, name);
2902 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2905 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2906 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2909 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2910 dmu_objset_rele(os, FTAG);
2914 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2915 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2916 dmu_objset_rele(os, FTAG);
2917 kmem_free(osname, MAXPATHLEN);
2920 p = osname + strlen(osname);
2921 len = MAXPATHLEN - (p - osname);
2924 /* Prefetch the datasets. */
2926 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2927 if (!dataset_name_hidden(osname))
2928 (void) dmu_objset_prefetch(osname, NULL);
2933 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2935 dmu_objset_rele(os, FTAG);
2936 (void)zvol_create_minors(osname);
2937 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2938 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2944 dmu_objset_rele(os, FTAG);
2945 kmem_free(osname, MAXPATHLEN);
2950 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2953 struct g_provider *pp;
2956 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2958 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2960 pp = zv->zv_provider;
2965 zv->zv_provider = NULL;
2966 g_wither_provider(pp, ENXIO);
2968 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2969 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2970 pp->sectorsize = DEV_BSIZE;
2971 pp->mediasize = zv->zv_volsize;
2973 zv->zv_provider = pp;
2974 g_error_provider(pp, 0);
2975 g_topology_unlock();
2976 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2977 struct make_dev_args args;
2979 if ((dev = zv->zv_dev) != NULL) {
2982 if (zv->zv_total_opens > 0) {
2983 zv->zv_flags &= ~ZVOL_EXCL;
2984 zv->zv_total_opens = 0;
2985 zvol_last_close(zv);
2989 make_dev_args_init(&args);
2990 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
2991 args.mda_devsw = &zvol_cdevsw;
2993 args.mda_uid = UID_ROOT;
2994 args.mda_gid = GID_OPERATOR;
2995 args.mda_mode = 0640;
2996 args.mda_si_drv2 = zv;
2997 if (make_dev_s(&args, &zv->zv_dev,
2998 "%s/%s", ZVOL_DRIVER, newname) == 0)
2999 zv->zv_dev->si_iosize_max = MAXPHYS;
3001 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3005 zvol_rename_minors(const char *oldname, const char *newname)
3007 char name[MAXPATHLEN];
3008 struct g_provider *pp;
3010 size_t oldnamelen, newnamelen;
3013 boolean_t locked = B_FALSE;
3015 oldnamelen = strlen(oldname);
3016 newnamelen = strlen(newname);
3019 /* See comment in zvol_open(). */
3020 if (!MUTEX_HELD(&zfsdev_state_lock)) {
3021 mutex_enter(&zfsdev_state_lock);
3025 LIST_FOREACH(zv, &all_zvols, zv_links) {
3026 if (strcmp(zv->zv_name, oldname) == 0) {
3027 zvol_rename_minor(zv, newname);
3028 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3029 (zv->zv_name[oldnamelen] == '/' ||
3030 zv->zv_name[oldnamelen] == '@')) {
3031 snprintf(name, sizeof(name), "%s%c%s", newname,
3032 zv->zv_name[oldnamelen],
3033 zv->zv_name + oldnamelen + 1);
3034 zvol_rename_minor(zv, name);
3039 mutex_exit(&zfsdev_state_lock);
3044 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3046 zvol_state_t *zv = dev->si_drv2;
3049 mutex_enter(&zfsdev_state_lock);
3050 if (zv->zv_total_opens == 0)
3051 err = zvol_first_open(zv);
3053 mutex_exit(&zfsdev_state_lock);
3056 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3057 err = SET_ERROR(EROFS);
3060 if (zv->zv_flags & ZVOL_EXCL) {
3061 err = SET_ERROR(EBUSY);
3065 if (flags & FEXCL) {
3066 if (zv->zv_total_opens != 0) {
3067 err = SET_ERROR(EBUSY);
3070 zv->zv_flags |= ZVOL_EXCL;
3074 zv->zv_total_opens++;
3075 if (flags & (FSYNC | FDSYNC)) {
3077 if (zv->zv_sync_cnt == 1)
3078 zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
3080 mutex_exit(&zfsdev_state_lock);
3083 if (zv->zv_total_opens == 0)
3084 zvol_last_close(zv);
3085 mutex_exit(&zfsdev_state_lock);
3090 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3092 zvol_state_t *zv = dev->si_drv2;
3094 mutex_enter(&zfsdev_state_lock);
3095 if (zv->zv_flags & ZVOL_EXCL) {
3096 ASSERT(zv->zv_total_opens == 1);
3097 zv->zv_flags &= ~ZVOL_EXCL;
3101 * If the open count is zero, this is a spurious close.
3102 * That indicates a bug in the kernel / DDI framework.
3104 ASSERT(zv->zv_total_opens != 0);
3107 * You may get multiple opens, but only one close.
3109 zv->zv_total_opens--;
3110 if (flags & (FSYNC | FDSYNC))
3113 if (zv->zv_total_opens == 0)
3114 zvol_last_close(zv);
3116 mutex_exit(&zfsdev_state_lock);
3121 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3125 off_t offset, length;
3132 KASSERT(zv->zv_total_opens > 0,
3133 ("Device with zero access count in zvol_d_ioctl"));
3135 i = IOCPARM_LEN(cmd);
3137 case DIOCGSECTORSIZE:
3138 *(u_int *)data = DEV_BSIZE;
3140 case DIOCGMEDIASIZE:
3141 *(off_t *)data = zv->zv_volsize;
3144 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3147 if (!zvol_unmap_enabled)
3150 offset = ((off_t *)data)[0];
3151 length = ((off_t *)data)[1];
3152 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3153 offset < 0 || offset >= zv->zv_volsize ||
3155 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3161 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3162 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3163 error = dmu_tx_assign(tx, TXG_WAIT);
3168 sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
3169 zvol_log_truncate(zv, tx, offset, length, sync);
3171 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3174 zfs_range_unlock(rl);
3176 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3178 case DIOCGSTRIPESIZE:
3179 *(off_t *)data = zv->zv_volblocksize;
3181 case DIOCGSTRIPEOFFSET:
3185 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3186 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3187 uint64_t refd, avail, usedobjs, availobjs;
3189 if (strcmp(arg->name, "GEOM::candelete") == 0)
3191 else if (strcmp(arg->name, "blocksavail") == 0) {
3192 dmu_objset_space(zv->zv_objset, &refd, &avail,
3193 &usedobjs, &availobjs);
3194 arg->value.off = avail / DEV_BSIZE;
3195 } else if (strcmp(arg->name, "blocksused") == 0) {
3196 dmu_objset_space(zv->zv_objset, &refd, &avail,
3197 &usedobjs, &availobjs);
3198 arg->value.off = refd / DEV_BSIZE;
3199 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3200 avail = metaslab_class_get_space(spa_normal_class(spa));
3201 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3202 arg->value.off = avail / DEV_BSIZE;
3203 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3204 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3205 arg->value.off = refd / DEV_BSIZE;
3212 off_t *off = (off_t *)data;
3216 hole = (cmd == FIOSEEKHOLE);
3218 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3228 #endif /* illumos */