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;
588 dmu_object_info_t doi;
591 char chrbuf[30], blkbuf[30];
593 struct g_provider *pp;
595 uint64_t volsize, mode;
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 = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
663 kmem_free(zv, sizeof(*zv));
664 dmu_objset_disown(os, zvol_tag);
665 mutex_exit(&zfsdev_state_lock);
668 error = dsl_prop_get_integer(name,
669 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
670 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
674 zv->zv_volsize = volsize;
675 zv->zv_volmode = mode;
676 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
678 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
679 gp->start = zvol_geom_start;
680 gp->access = zvol_geom_access;
681 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
682 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
683 pp->sectorsize = DEV_BSIZE;
684 pp->mediasize = zv->zv_volsize;
687 zv->zv_provider = pp;
688 bioq_init(&zv->zv_queue);
689 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
690 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
691 struct make_dev_args args;
693 make_dev_args_init(&args);
694 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
695 args.mda_devsw = &zvol_cdevsw;
697 args.mda_uid = UID_ROOT;
698 args.mda_gid = GID_OPERATOR;
699 args.mda_mode = 0640;
700 args.mda_si_drv2 = zv;
701 error = make_dev_s(&args, &zv->zv_dev,
702 "%s/%s", ZVOL_DRIVER, name);
704 kmem_free(zv, sizeof(*zv));
705 dmu_objset_disown(os, FTAG);
706 mutex_exit(&zfsdev_state_lock);
709 zv->zv_dev->si_iosize_max = MAXPHYS;
711 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
714 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
715 zv->zv_min_bs = DEV_BSHIFT;
717 zv->zv_minor = minor;
720 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
721 zv->zv_flags |= ZVOL_RDONLY;
722 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
723 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
724 sizeof (rl_t), offsetof(rl_t, r_node));
725 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
726 offsetof(zvol_extent_t, ze_node));
727 /* get and cache the blocksize */
728 error = dmu_object_info(os, ZVOL_OBJ, &doi);
730 zv->zv_volblocksize = doi.doi_data_block_size;
732 if (spa_writeable(dmu_objset_spa(os))) {
733 if (zil_replay_disable)
734 zil_destroy(dmu_objset_zil(os), B_FALSE);
736 zil_replay(os, zv, zvol_replay_vector);
738 dmu_objset_disown(os, FTAG);
739 zv->zv_objset = NULL;
743 mutex_exit(&zfsdev_state_lock);
745 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
751 ZFS_LOG(1, "ZVOL %s created.", name);
758 * Remove minor node for the specified volume.
761 zvol_remove_zv(zvol_state_t *zv)
765 minor_t minor = zv->zv_minor;
768 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
769 if (zv->zv_total_opens != 0)
770 return (SET_ERROR(EBUSY));
773 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
774 ddi_remove_minor_node(zfs_dip, nmbuf);
776 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
777 ddi_remove_minor_node(zfs_dip, nmbuf);
779 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
781 LIST_REMOVE(zv, zv_links);
782 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
784 zvol_geom_destroy(zv);
786 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
787 if (zv->zv_dev != NULL)
788 destroy_dev(zv->zv_dev);
792 avl_destroy(&zv->zv_znode.z_range_avl);
793 mutex_destroy(&zv->zv_znode.z_range_lock);
795 kmem_free(zv, sizeof (zvol_state_t));
797 ddi_soft_state_free(zfsdev_state, minor);
804 zvol_remove_minor(const char *name)
809 mutex_enter(&zfsdev_state_lock);
810 if ((zv = zvol_minor_lookup(name)) == NULL) {
811 mutex_exit(&zfsdev_state_lock);
812 return (SET_ERROR(ENXIO));
814 rc = zvol_remove_zv(zv);
815 mutex_exit(&zfsdev_state_lock);
820 zvol_first_open(zvol_state_t *zv)
827 /* lie and say we're read-only */
828 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
834 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
837 dmu_objset_disown(os, zvol_tag);
841 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
843 dmu_objset_disown(os, zvol_tag);
847 zvol_size_changed(zv, volsize);
848 zv->zv_zilog = zil_open(os, zvol_get_data);
850 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
852 if (readonly || dmu_objset_is_snapshot(os) ||
853 !spa_writeable(dmu_objset_spa(os)))
854 zv->zv_flags |= ZVOL_RDONLY;
856 zv->zv_flags &= ~ZVOL_RDONLY;
861 zvol_last_close(zvol_state_t *zv)
863 zil_close(zv->zv_zilog);
866 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
872 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
873 !(zv->zv_flags & ZVOL_RDONLY))
874 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
875 dmu_objset_evict_dbufs(zv->zv_objset);
877 dmu_objset_disown(zv->zv_objset, zvol_tag);
878 zv->zv_objset = NULL;
883 zvol_prealloc(zvol_state_t *zv)
885 objset_t *os = zv->zv_objset;
887 uint64_t refd, avail, usedobjs, availobjs;
888 uint64_t resid = zv->zv_volsize;
891 /* Check the space usage before attempting to allocate the space */
892 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
893 if (avail < zv->zv_volsize)
894 return (SET_ERROR(ENOSPC));
896 /* Free old extents if they exist */
897 zvol_free_extents(zv);
901 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
903 tx = dmu_tx_create(os);
904 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
905 error = dmu_tx_assign(tx, TXG_WAIT);
908 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
911 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
916 txg_wait_synced(dmu_objset_pool(os), 0);
923 zvol_update_volsize(objset_t *os, uint64_t volsize)
928 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
930 tx = dmu_tx_create(os);
931 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
932 dmu_tx_mark_netfree(tx);
933 error = dmu_tx_assign(tx, TXG_WAIT);
939 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
944 error = dmu_free_long_range(os,
945 ZVOL_OBJ, volsize, DMU_OBJECT_END);
950 zvol_remove_minors(const char *name)
957 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
958 (void) strncpy(namebuf, name, strlen(name));
959 (void) strcat(namebuf, "/");
960 mutex_enter(&zfsdev_state_lock);
961 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
963 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
966 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
967 (void) zvol_remove_zv(zv);
969 kmem_free(namebuf, strlen(name) + 2);
971 mutex_exit(&zfsdev_state_lock);
973 zvol_state_t *zv, *tzv;
976 namelen = strlen(name);
979 mutex_enter(&zfsdev_state_lock);
981 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
982 if (strcmp(zv->zv_name, name) == 0 ||
983 (strncmp(zv->zv_name, name, namelen) == 0 &&
984 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
985 zv->zv_name[namelen] == '@'))) {
986 (void) zvol_remove_zv(zv);
990 mutex_exit(&zfsdev_state_lock);
996 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
998 uint64_t old_volsize = 0ULL;
1001 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1004 * Reinitialize the dump area to the new size. If we
1005 * failed to resize the dump area then restore it back to
1006 * its original size. We must set the new volsize prior
1007 * to calling dumpvp_resize() to ensure that the devices'
1008 * size(9P) is not visible by the dump subsystem.
1010 old_volsize = zv->zv_volsize;
1011 zvol_size_changed(zv, volsize);
1014 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1015 if ((error = zvol_dumpify(zv)) != 0 ||
1016 (error = dumpvp_resize()) != 0) {
1019 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1020 zvol_size_changed(zv, old_volsize);
1021 dumpify_error = zvol_dumpify(zv);
1022 error = dumpify_error ? dumpify_error : error;
1025 #endif /* ZVOL_DUMP */
1029 * Generate a LUN expansion event.
1034 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1036 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1039 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1040 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1042 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1043 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1046 kmem_free(physpath, MAXPATHLEN);
1048 #endif /* illumos */
1053 zvol_set_volsize(const char *name, uint64_t volsize)
1055 zvol_state_t *zv = NULL;
1058 dmu_object_info_t doi;
1060 boolean_t owned = B_FALSE;
1062 error = dsl_prop_get_integer(name,
1063 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1067 return (SET_ERROR(EROFS));
1069 mutex_enter(&zfsdev_state_lock);
1070 zv = zvol_minor_lookup(name);
1072 if (zv == NULL || zv->zv_objset == NULL) {
1073 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1075 mutex_exit(&zfsdev_state_lock);
1085 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1086 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1089 error = zvol_update_volsize(os, volsize);
1091 if (error == 0 && zv != NULL)
1092 error = zvol_update_live_volsize(zv, volsize);
1095 dmu_objset_disown(os, FTAG);
1097 zv->zv_objset = NULL;
1099 mutex_exit(&zfsdev_state_lock);
1106 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1109 zvol_open(struct g_provider *pp, int flag, int count)
1116 mutex_enter(&zfsdev_state_lock);
1118 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1120 mutex_exit(&zfsdev_state_lock);
1121 return (SET_ERROR(ENXIO));
1124 if (zv->zv_total_opens == 0)
1125 err = zvol_first_open(zv);
1127 mutex_exit(&zfsdev_state_lock);
1130 #else /* !illumos */
1131 boolean_t locked = B_FALSE;
1133 if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1135 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1136 * attempting to probe geom providers while looking for a
1137 * replacement for a missing VDEV. In this case, the
1138 * spa_namespace_lock will not be held, but it is still illegal
1139 * to use a zvol as a vdev. Deadlocks can result if another
1140 * thread has spa_namespace_lock
1142 return (EOPNOTSUPP);
1145 * Protect against recursively entering spa_namespace_lock
1146 * when spa_open() is used for a pool on a (local) ZVOL(s).
1147 * This is needed since we replaced upstream zfsdev_state_lock
1148 * with spa_namespace_lock in the ZVOL code.
1149 * We are using the same trick as spa_open().
1150 * Note that calls in zvol_first_open which need to resolve
1151 * pool name to a spa object will enter spa_open()
1152 * recursively, but that function already has all the
1153 * necessary protection.
1155 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1156 mutex_enter(&zfsdev_state_lock);
1163 mutex_exit(&zfsdev_state_lock);
1164 return (SET_ERROR(ENXIO));
1167 if (zv->zv_total_opens == 0) {
1168 err = zvol_first_open(zv);
1171 mutex_exit(&zfsdev_state_lock);
1174 pp->mediasize = zv->zv_volsize;
1175 pp->stripeoffset = 0;
1176 pp->stripesize = zv->zv_volblocksize;
1178 #endif /* illumos */
1179 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1180 err = SET_ERROR(EROFS);
1183 if (zv->zv_flags & ZVOL_EXCL) {
1184 err = SET_ERROR(EBUSY);
1189 if (zv->zv_total_opens != 0) {
1190 err = SET_ERROR(EBUSY);
1193 zv->zv_flags |= ZVOL_EXCL;
1198 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1199 zv->zv_open_count[otyp]++;
1200 zv->zv_total_opens++;
1202 mutex_exit(&zfsdev_state_lock);
1204 zv->zv_total_opens += count;
1206 mutex_exit(&zfsdev_state_lock);
1211 if (zv->zv_total_opens == 0)
1212 zvol_last_close(zv);
1214 mutex_exit(&zfsdev_state_lock);
1217 mutex_exit(&zfsdev_state_lock);
1225 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1227 minor_t minor = getminor(dev);
1231 mutex_enter(&zfsdev_state_lock);
1233 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1235 mutex_exit(&zfsdev_state_lock);
1236 #else /* !illumos */
1238 zvol_close(struct g_provider *pp, int flag, int count)
1242 boolean_t locked = B_FALSE;
1244 /* See comment in zvol_open(). */
1245 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1246 mutex_enter(&zfsdev_state_lock);
1253 mutex_exit(&zfsdev_state_lock);
1254 #endif /* illumos */
1255 return (SET_ERROR(ENXIO));
1258 if (zv->zv_flags & ZVOL_EXCL) {
1259 ASSERT(zv->zv_total_opens == 1);
1260 zv->zv_flags &= ~ZVOL_EXCL;
1264 * If the open count is zero, this is a spurious close.
1265 * That indicates a bug in the kernel / DDI framework.
1268 ASSERT(zv->zv_open_count[otyp] != 0);
1270 ASSERT(zv->zv_total_opens != 0);
1273 * You may get multiple opens, but only one close.
1276 zv->zv_open_count[otyp]--;
1277 zv->zv_total_opens--;
1279 zv->zv_total_opens -= count;
1282 if (zv->zv_total_opens == 0)
1283 zvol_last_close(zv);
1286 mutex_exit(&zfsdev_state_lock);
1289 mutex_exit(&zfsdev_state_lock);
1295 zvol_get_done(zgd_t *zgd, int error)
1298 dmu_buf_rele(zgd->zgd_db, zgd);
1300 zfs_range_unlock(zgd->zgd_rl);
1302 if (error == 0 && zgd->zgd_bp)
1303 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1305 kmem_free(zgd, sizeof (zgd_t));
1309 * Get data to generate a TX_WRITE intent log record.
1312 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1314 zvol_state_t *zv = arg;
1315 objset_t *os = zv->zv_objset;
1316 uint64_t object = ZVOL_OBJ;
1317 uint64_t offset = lr->lr_offset;
1318 uint64_t size = lr->lr_length; /* length of user data */
1319 blkptr_t *bp = &lr->lr_blkptr;
1324 ASSERT(zio != NULL);
1327 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1328 zgd->zgd_zilog = zv->zv_zilog;
1329 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1332 * Write records come in two flavors: immediate and indirect.
1333 * For small writes it's cheaper to store the data with the
1334 * log record (immediate); for large writes it's cheaper to
1335 * sync the data and get a pointer to it (indirect) so that
1336 * we don't have to write the data twice.
1338 if (buf != NULL) { /* immediate write */
1339 error = dmu_read(os, object, offset, size, buf,
1340 DMU_READ_NO_PREFETCH);
1342 size = zv->zv_volblocksize;
1343 offset = P2ALIGN(offset, size);
1344 error = dmu_buf_hold(os, object, offset, zgd, &db,
1345 DMU_READ_NO_PREFETCH);
1347 blkptr_t *obp = dmu_buf_get_blkptr(db);
1349 ASSERT(BP_IS_HOLE(bp));
1356 ASSERT(db->db_offset == offset);
1357 ASSERT(db->db_size == size);
1359 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1360 zvol_get_done, zgd);
1367 zvol_get_done(zgd, error);
1373 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1375 * We store data in the log buffers if it's small enough.
1376 * Otherwise we will later flush the data out via dmu_sync().
1378 ssize_t zvol_immediate_write_sz = 32768;
1380 SYSCTL_LONG(_vfs_zfs_vol, OID_AUTO, immediate_write_sz, CTLFLAG_RWTUN,
1381 &zvol_immediate_write_sz, 0, "Minimal size for indirect log write");
1385 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1388 uint32_t blocksize = zv->zv_volblocksize;
1389 zilog_t *zilog = zv->zv_zilog;
1390 itx_wr_state_t write_state;
1392 if (zil_replaying(zilog, tx))
1395 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1396 write_state = WR_INDIRECT;
1397 else if (!spa_has_slogs(zilog->zl_spa) &&
1398 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1399 write_state = WR_INDIRECT;
1401 write_state = WR_COPIED;
1403 write_state = WR_NEED_COPY;
1408 itx_wr_state_t wr_state = write_state;
1409 ssize_t len = resid;
1411 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1412 wr_state = WR_NEED_COPY;
1413 else if (wr_state == WR_INDIRECT)
1414 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1416 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1417 (wr_state == WR_COPIED ? len : 0));
1418 lr = (lr_write_t *)&itx->itx_lr;
1419 if (wr_state == WR_COPIED && dmu_read(zv->zv_objset,
1420 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1421 zil_itx_destroy(itx);
1422 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1423 lr = (lr_write_t *)&itx->itx_lr;
1424 wr_state = WR_NEED_COPY;
1427 itx->itx_wr_state = wr_state;
1428 lr->lr_foid = ZVOL_OBJ;
1429 lr->lr_offset = off;
1430 lr->lr_length = len;
1432 BP_ZERO(&lr->lr_blkptr);
1434 itx->itx_private = zv;
1436 if (!sync && (zv->zv_sync_cnt == 0))
1437 itx->itx_sync = B_FALSE;
1439 zil_itx_assign(zilog, itx, tx);
1448 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1449 uint64_t size, boolean_t doread, boolean_t isdump)
1455 if (vd->vdev_ops == &vdev_mirror_ops ||
1456 vd->vdev_ops == &vdev_replacing_ops ||
1457 vd->vdev_ops == &vdev_spare_ops) {
1458 for (c = 0; c < vd->vdev_children; c++) {
1459 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1460 addr, offset, origoffset, size, doread, isdump);
1463 } else if (doread) {
1469 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1470 return (numerrors < vd->vdev_children ? 0 : EIO);
1472 if (doread && !vdev_readable(vd))
1473 return (SET_ERROR(EIO));
1474 else if (!doread && !vdev_writeable(vd))
1475 return (SET_ERROR(EIO));
1477 if (vd->vdev_ops == &vdev_raidz_ops) {
1478 return (vdev_raidz_physio(vd,
1479 addr, size, offset, origoffset, doread, isdump));
1482 offset += VDEV_LABEL_START_SIZE;
1484 if (ddi_in_panic() || isdump) {
1487 return (SET_ERROR(EIO));
1489 ASSERT3P(dvd, !=, NULL);
1490 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1494 ASSERT3P(dvd, !=, NULL);
1495 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1496 offset, doread ? B_READ : B_WRITE));
1501 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1502 boolean_t doread, boolean_t isdump)
1507 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1509 /* Must be sector aligned, and not stradle a block boundary. */
1510 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1511 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1512 return (SET_ERROR(EINVAL));
1514 ASSERT(size <= zv->zv_volblocksize);
1516 /* Locate the extent this belongs to */
1517 ze = list_head(&zv->zv_extents);
1518 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1519 offset -= ze->ze_nblks * zv->zv_volblocksize;
1520 ze = list_next(&zv->zv_extents, ze);
1524 return (SET_ERROR(EINVAL));
1526 if (!ddi_in_panic())
1527 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1529 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1530 offset += DVA_GET_OFFSET(&ze->ze_dva);
1531 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1532 size, doread, isdump);
1534 if (!ddi_in_panic())
1535 spa_config_exit(spa, SCL_STATE, FTAG);
1541 zvol_strategy(buf_t *bp)
1543 zfs_soft_state_t *zs = NULL;
1544 #else /* !illumos */
1546 zvol_strategy(struct bio *bp)
1548 #endif /* illumos */
1550 uint64_t off, volsize;
1557 boolean_t doread = bp->b_flags & B_READ;
1559 boolean_t doread = 0;
1561 boolean_t is_dumpified;
1565 if (getminor(bp->b_edev) == 0) {
1566 error = SET_ERROR(EINVAL);
1568 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1570 error = SET_ERROR(ENXIO);
1571 else if (zs->zss_type != ZSST_ZVOL)
1572 error = SET_ERROR(EINVAL);
1576 bioerror(bp, error);
1583 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1584 bioerror(bp, EROFS);
1589 off = ldbtob(bp->b_blkno);
1590 #else /* !illumos */
1592 zv = bp->bio_to->private;
1594 zv = bp->bio_dev->si_drv2;
1597 error = SET_ERROR(ENXIO);
1601 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1602 error = SET_ERROR(EROFS);
1606 switch (bp->bio_cmd) {
1619 off = bp->bio_offset;
1620 #endif /* illumos */
1621 volsize = zv->zv_volsize;
1628 addr = bp->b_un.b_addr;
1629 resid = bp->b_bcount;
1631 if (resid > 0 && (off < 0 || off >= volsize)) {
1637 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1638 sync = ((!(bp->b_flags & B_ASYNC) &&
1639 !(zv->zv_flags & ZVOL_WCE)) ||
1640 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1641 !doread && !is_dumpified;
1642 #else /* !illumos */
1643 addr = bp->bio_data;
1644 resid = bp->bio_length;
1646 if (resid > 0 && (off < 0 || off >= volsize)) {
1647 error = SET_ERROR(EIO);
1651 is_dumpified = B_FALSE;
1652 sync = !doread && !is_dumpified &&
1653 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1654 #endif /* illumos */
1657 * There must be no buffer changes when doing a dmu_sync() because
1658 * we can't change the data whilst calculating the checksum.
1660 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1661 doread ? RL_READER : RL_WRITER);
1664 if (bp->bio_cmd == BIO_DELETE) {
1665 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1666 error = dmu_tx_assign(tx, TXG_WAIT);
1670 zvol_log_truncate(zv, tx, off, resid, sync);
1672 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1679 while (resid != 0 && off < volsize) {
1680 size_t size = MIN(resid, zvol_maxphys);
1683 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1684 error = zvol_dumpio(zv, addr, off, size,
1686 } else if (doread) {
1690 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1693 dmu_tx_t *tx = dmu_tx_create(os);
1694 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1695 error = dmu_tx_assign(tx, TXG_WAIT);
1699 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1700 zvol_log_write(zv, tx, off, size, sync);
1705 /* convert checksum errors into IO errors */
1706 if (error == ECKSUM)
1707 error = SET_ERROR(EIO);
1717 zfs_range_unlock(rl);
1720 if ((bp->b_resid = resid) == bp->b_bcount)
1721 bioerror(bp, off > volsize ? EINVAL : error);
1724 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1728 #else /* !illumos */
1729 bp->bio_completed = bp->bio_length - resid;
1730 if (bp->bio_completed < bp->bio_length && off > volsize)
1735 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1739 g_io_deliver(bp, error);
1741 biofinish(bp, NULL, error);
1742 #endif /* illumos */
1747 * Set the buffer count to the zvol maximum transfer.
1748 * Using our own routine instead of the default minphys()
1749 * means that for larger writes we write bigger buffers on X86
1750 * (128K instead of 56K) and flush the disk write cache less often
1751 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1752 * 56K on X86 and 128K on sparc).
1755 zvol_minphys(struct buf *bp)
1757 if (bp->b_bcount > zvol_maxphys)
1758 bp->b_bcount = zvol_maxphys;
1762 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1764 minor_t minor = getminor(dev);
1771 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1773 return (SET_ERROR(ENXIO));
1775 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1776 return (SET_ERROR(EINVAL));
1778 boff = ldbtob(blkno);
1779 resid = ldbtob(nblocks);
1781 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1784 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1785 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1798 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1800 minor_t minor = getminor(dev);
1801 #else /* !illumos */
1803 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1805 #endif /* illumos */
1812 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1814 return (SET_ERROR(ENXIO));
1819 volsize = zv->zv_volsize;
1820 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1821 if (uio->uio_resid > 0 &&
1822 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1823 return (SET_ERROR(EIO));
1826 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1827 error = physio(zvol_strategy, NULL, dev, B_READ,
1833 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1835 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1836 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1838 /* don't read past the end */
1839 if (bytes > volsize - uio->uio_loffset)
1840 bytes = volsize - uio->uio_loffset;
1842 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
1844 /* convert checksum errors into IO errors */
1845 if (error == ECKSUM)
1846 error = SET_ERROR(EIO);
1850 zfs_range_unlock(rl);
1857 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1859 minor_t minor = getminor(dev);
1860 #else /* !illumos */
1862 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1864 #endif /* illumos */
1872 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1874 return (SET_ERROR(ENXIO));
1879 volsize = zv->zv_volsize;
1880 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1881 if (uio->uio_resid > 0 &&
1882 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1883 return (SET_ERROR(EIO));
1886 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1887 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1892 sync = !(zv->zv_flags & ZVOL_WCE) ||
1894 sync = (ioflag & IO_SYNC) ||
1896 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1898 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1900 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1901 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1902 uint64_t off = uio->uio_loffset;
1903 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1905 if (bytes > volsize - off) /* don't write past the end */
1906 bytes = volsize - off;
1908 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1909 error = dmu_tx_assign(tx, TXG_WAIT);
1914 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1916 zvol_log_write(zv, tx, off, bytes, sync);
1922 zfs_range_unlock(rl);
1924 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1930 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1932 struct uuid uuid = EFI_RESERVED;
1933 efi_gpe_t gpe = { 0 };
1939 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1940 return (SET_ERROR(EFAULT));
1941 ptr = (char *)(uintptr_t)efi.dki_data_64;
1942 length = efi.dki_length;
1944 * Some clients may attempt to request a PMBR for the
1945 * zvol. Currently this interface will return EINVAL to
1946 * such requests. These requests could be supported by
1947 * adding a check for lba == 0 and consing up an appropriate
1950 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1951 return (SET_ERROR(EINVAL));
1953 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1954 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1955 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1957 if (efi.dki_lba == 1) {
1958 efi_gpt_t gpt = { 0 };
1960 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1961 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1962 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1963 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1964 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1965 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1966 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1967 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1968 gpt.efi_gpt_SizeOfPartitionEntry =
1969 LE_32(sizeof (efi_gpe_t));
1970 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1971 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1972 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1973 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1974 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1976 return (SET_ERROR(EFAULT));
1977 ptr += sizeof (gpt);
1978 length -= sizeof (gpt);
1980 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1982 return (SET_ERROR(EFAULT));
1987 * BEGIN entry points to allow external callers access to the volume.
1990 * Return the volume parameters needed for access from an external caller.
1991 * These values are invariant as long as the volume is held open.
1994 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1995 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1996 void **rl_hdl, void **bonus_hdl)
2000 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2002 return (SET_ERROR(ENXIO));
2003 if (zv->zv_flags & ZVOL_DUMPIFIED)
2004 return (SET_ERROR(ENXIO));
2006 ASSERT(blksize && max_xfer_len && minor_hdl &&
2007 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
2009 *blksize = zv->zv_volblocksize;
2010 *max_xfer_len = (uint64_t)zvol_maxphys;
2012 *objset_hdl = zv->zv_objset;
2013 *zil_hdl = zv->zv_zilog;
2014 *rl_hdl = &zv->zv_znode;
2015 *bonus_hdl = zv->zv_dbuf;
2020 * Return the current volume size to an external caller.
2021 * The size can change while the volume is open.
2024 zvol_get_volume_size(void *minor_hdl)
2026 zvol_state_t *zv = minor_hdl;
2028 return (zv->zv_volsize);
2032 * Return the current WCE setting to an external caller.
2033 * The WCE setting can change while the volume is open.
2036 zvol_get_volume_wce(void *minor_hdl)
2038 zvol_state_t *zv = minor_hdl;
2040 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2044 * Entry point for external callers to zvol_log_write
2047 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2050 zvol_state_t *zv = minor_hdl;
2052 zvol_log_write(zv, tx, off, resid, sync);
2055 * END entry points to allow external callers access to the volume.
2057 #endif /* illumos */
2060 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2063 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2068 zilog_t *zilog = zv->zv_zilog;
2070 if (zil_replaying(zilog, tx))
2073 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2074 lr = (lr_truncate_t *)&itx->itx_lr;
2075 lr->lr_foid = ZVOL_OBJ;
2076 lr->lr_offset = off;
2077 lr->lr_length = len;
2079 itx->itx_sync = (sync || zv->zv_sync_cnt != 0);
2080 zil_itx_assign(zilog, itx, tx);
2085 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2086 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2090 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2093 struct dk_callback *dkc;
2097 mutex_enter(&zfsdev_state_lock);
2099 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2102 mutex_exit(&zfsdev_state_lock);
2103 return (SET_ERROR(ENXIO));
2105 ASSERT(zv->zv_total_opens > 0);
2111 struct dk_cinfo dki;
2113 bzero(&dki, sizeof (dki));
2114 (void) strcpy(dki.dki_cname, "zvol");
2115 (void) strcpy(dki.dki_dname, "zvol");
2116 dki.dki_ctype = DKC_UNKNOWN;
2117 dki.dki_unit = getminor(dev);
2118 dki.dki_maxtransfer =
2119 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2120 mutex_exit(&zfsdev_state_lock);
2121 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2122 error = SET_ERROR(EFAULT);
2126 case DKIOCGMEDIAINFO:
2128 struct dk_minfo dkm;
2130 bzero(&dkm, sizeof (dkm));
2131 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2132 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2133 dkm.dki_media_type = DK_UNKNOWN;
2134 mutex_exit(&zfsdev_state_lock);
2135 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2136 error = SET_ERROR(EFAULT);
2140 case DKIOCGMEDIAINFOEXT:
2142 struct dk_minfo_ext dkmext;
2144 bzero(&dkmext, sizeof (dkmext));
2145 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2146 dkmext.dki_pbsize = zv->zv_volblocksize;
2147 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2148 dkmext.dki_media_type = DK_UNKNOWN;
2149 mutex_exit(&zfsdev_state_lock);
2150 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2151 error = SET_ERROR(EFAULT);
2157 uint64_t vs = zv->zv_volsize;
2158 uint8_t bs = zv->zv_min_bs;
2160 mutex_exit(&zfsdev_state_lock);
2161 error = zvol_getefi((void *)arg, flag, vs, bs);
2165 case DKIOCFLUSHWRITECACHE:
2166 dkc = (struct dk_callback *)arg;
2167 mutex_exit(&zfsdev_state_lock);
2168 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2169 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2170 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2177 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2178 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2180 error = SET_ERROR(EFAULT);
2186 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2188 error = SET_ERROR(EFAULT);
2192 zv->zv_flags |= ZVOL_WCE;
2193 mutex_exit(&zfsdev_state_lock);
2195 zv->zv_flags &= ~ZVOL_WCE;
2196 mutex_exit(&zfsdev_state_lock);
2197 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2205 * commands using these (like prtvtoc) expect ENOTSUP
2206 * since we're emulating an EFI label
2208 error = SET_ERROR(ENOTSUP);
2212 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2214 error = zvol_dumpify(zv);
2215 zfs_range_unlock(rl);
2219 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2221 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2223 error = zvol_dump_fini(zv);
2224 zfs_range_unlock(rl);
2232 if (!zvol_unmap_enabled)
2235 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2236 error = SET_ERROR(EFAULT);
2241 * Apply Postel's Law to length-checking. If they overshoot,
2242 * just blank out until the end, if there's a need to blank
2245 if (df.df_start >= zv->zv_volsize)
2246 break; /* No need to do anything... */
2248 mutex_exit(&zfsdev_state_lock);
2250 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2252 tx = dmu_tx_create(zv->zv_objset);
2253 dmu_tx_mark_netfree(tx);
2254 error = dmu_tx_assign(tx, TXG_WAIT);
2258 zvol_log_truncate(zv, tx, df.df_start,
2259 df.df_length, B_TRUE);
2261 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2262 df.df_start, df.df_length);
2265 zfs_range_unlock(rl);
2269 * If the write-cache is disabled or 'sync' property
2270 * is set to 'always' then treat this as a synchronous
2271 * operation (i.e. commit to zil).
2273 if (!(zv->zv_flags & ZVOL_WCE) ||
2274 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2275 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2278 * If the caller really wants synchronous writes, and
2279 * can't wait for them, don't return until the write
2282 if (df.df_flags & DF_WAIT_SYNC) {
2284 dmu_objset_pool(zv->zv_objset), 0);
2291 error = SET_ERROR(ENOTTY);
2295 mutex_exit(&zfsdev_state_lock);
2298 #endif /* illumos */
2303 return (zvol_minors != 0);
2309 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2312 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2314 ZFS_LOG(1, "ZVOL Initialized.");
2322 mutex_destroy(&zfsdev_state_lock);
2324 ddi_soft_state_fini(&zfsdev_state);
2325 ZFS_LOG(1, "ZVOL Deinitialized.");
2331 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2333 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2335 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2342 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2344 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2346 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2350 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2354 objset_t *os = zv->zv_objset;
2355 spa_t *spa = dmu_objset_spa(os);
2356 vdev_t *vd = spa->spa_root_vdev;
2357 nvlist_t *nv = NULL;
2358 uint64_t version = spa_version(spa);
2359 uint64_t checksum, compress, refresrv, vbs, dedup;
2361 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2362 ASSERT(vd->vdev_ops == &vdev_root_ops);
2364 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2368 /* wait for dmu_free_long_range to actually free the blocks */
2369 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2372 * If the pool on which the dump device is being initialized has more
2373 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2374 * enabled. If so, bump that feature's counter to indicate that the
2375 * feature is active. We also check the vdev type to handle the
2377 * # zpool create test raidz disk1 disk2 disk3
2378 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2379 * the raidz vdev itself has 3 children.
2381 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2382 if (!spa_feature_is_enabled(spa,
2383 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2384 return (SET_ERROR(ENOTSUP));
2385 (void) dsl_sync_task(spa_name(spa),
2386 zfs_mvdev_dump_feature_check,
2387 zfs_mvdev_dump_activate_feature_sync, NULL,
2388 2, ZFS_SPACE_CHECK_RESERVED);
2392 error = dsl_prop_get_integer(zv->zv_name,
2393 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2395 error = dsl_prop_get_integer(zv->zv_name,
2396 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2400 error = dsl_prop_get_integer(zv->zv_name,
2401 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2405 error = dsl_prop_get_integer(zv->zv_name,
2406 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2409 if (version >= SPA_VERSION_DEDUP && error == 0) {
2410 error = dsl_prop_get_integer(zv->zv_name,
2411 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2417 tx = dmu_tx_create(os);
2418 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2419 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2420 error = dmu_tx_assign(tx, TXG_WAIT);
2427 * If we are resizing the dump device then we only need to
2428 * update the refreservation to match the newly updated
2429 * zvolsize. Otherwise, we save off the original state of the
2430 * zvol so that we can restore them if the zvol is ever undumpified.
2433 error = zap_update(os, ZVOL_ZAP_OBJ,
2434 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2435 &zv->zv_volsize, tx);
2437 error = zap_update(os, ZVOL_ZAP_OBJ,
2438 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2441 error = zap_update(os, ZVOL_ZAP_OBJ,
2442 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2446 error = zap_update(os, ZVOL_ZAP_OBJ,
2447 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2451 error = zap_update(os, ZVOL_ZAP_OBJ,
2452 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2456 error = dmu_object_set_blocksize(
2457 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2459 if (version >= SPA_VERSION_DEDUP && error == 0) {
2460 error = zap_update(os, ZVOL_ZAP_OBJ,
2461 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2465 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2470 * We only need update the zvol's property if we are initializing
2471 * the dump area for the first time.
2473 if (error == 0 && !resize) {
2475 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2476 * function. Otherwise, use the old default -- OFF.
2478 checksum = spa_feature_is_active(spa,
2479 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2482 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2483 VERIFY(nvlist_add_uint64(nv,
2484 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2485 VERIFY(nvlist_add_uint64(nv,
2486 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2487 ZIO_COMPRESS_OFF) == 0);
2488 VERIFY(nvlist_add_uint64(nv,
2489 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2491 if (version >= SPA_VERSION_DEDUP) {
2492 VERIFY(nvlist_add_uint64(nv,
2493 zfs_prop_to_name(ZFS_PROP_DEDUP),
2494 ZIO_CHECKSUM_OFF) == 0);
2497 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2502 /* Allocate the space for the dump */
2504 error = zvol_prealloc(zv);
2509 zvol_dumpify(zvol_state_t *zv)
2512 uint64_t dumpsize = 0;
2514 objset_t *os = zv->zv_objset;
2516 if (zv->zv_flags & ZVOL_RDONLY)
2517 return (SET_ERROR(EROFS));
2519 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2520 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2521 boolean_t resize = (dumpsize > 0);
2523 if ((error = zvol_dump_init(zv, resize)) != 0) {
2524 (void) zvol_dump_fini(zv);
2530 * Build up our lba mapping.
2532 error = zvol_get_lbas(zv);
2534 (void) zvol_dump_fini(zv);
2538 tx = dmu_tx_create(os);
2539 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2540 error = dmu_tx_assign(tx, TXG_WAIT);
2543 (void) zvol_dump_fini(zv);
2547 zv->zv_flags |= ZVOL_DUMPIFIED;
2548 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2549 &zv->zv_volsize, tx);
2553 (void) zvol_dump_fini(zv);
2557 txg_wait_synced(dmu_objset_pool(os), 0);
2562 zvol_dump_fini(zvol_state_t *zv)
2565 objset_t *os = zv->zv_objset;
2568 uint64_t checksum, compress, refresrv, vbs, dedup;
2569 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2572 * Attempt to restore the zvol back to its pre-dumpified state.
2573 * This is a best-effort attempt as it's possible that not all
2574 * of these properties were initialized during the dumpify process
2575 * (i.e. error during zvol_dump_init).
2578 tx = dmu_tx_create(os);
2579 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2580 error = dmu_tx_assign(tx, TXG_WAIT);
2585 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2588 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2589 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2590 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2591 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2592 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2593 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2594 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2595 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2597 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2598 (void) nvlist_add_uint64(nv,
2599 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2600 (void) nvlist_add_uint64(nv,
2601 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2602 (void) nvlist_add_uint64(nv,
2603 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2604 if (version >= SPA_VERSION_DEDUP &&
2605 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2606 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2607 (void) nvlist_add_uint64(nv,
2608 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2610 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2614 zvol_free_extents(zv);
2615 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2616 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2617 /* wait for dmu_free_long_range to actually free the blocks */
2618 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2619 tx = dmu_tx_create(os);
2620 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2621 error = dmu_tx_assign(tx, TXG_WAIT);
2626 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2627 zv->zv_volblocksize = vbs;
2632 #else /* !illumos */
2635 zvol_geom_run(zvol_state_t *zv)
2637 struct g_provider *pp;
2639 pp = zv->zv_provider;
2640 g_error_provider(pp, 0);
2642 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2643 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2647 zvol_geom_destroy(zvol_state_t *zv)
2649 struct g_provider *pp;
2651 g_topology_assert();
2653 mtx_lock(&zv->zv_queue_mtx);
2655 wakeup_one(&zv->zv_queue);
2656 while (zv->zv_state != 2)
2657 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2658 mtx_destroy(&zv->zv_queue_mtx);
2660 pp = zv->zv_provider;
2661 zv->zv_provider = NULL;
2663 g_wither_geom(pp->geom, ENXIO);
2667 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2669 int count, error, flags;
2671 g_topology_assert();
2674 * To make it easier we expect either open or close, but not both
2677 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2678 (acr <= 0 && acw <= 0 && ace <= 0),
2679 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2680 pp->name, acr, acw, ace));
2682 if (pp->private == NULL) {
2683 if (acr <= 0 && acw <= 0 && ace <= 0)
2689 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2690 * because GEOM already handles that and handles it a bit differently.
2691 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2692 * only one exclusive consumer, no matter if it is reader or writer.
2693 * I like better the way GEOM works so I'll leave it for GEOM to
2694 * decide what to do.
2697 count = acr + acw + ace;
2702 if (acr != 0 || ace != 0)
2707 g_topology_unlock();
2709 error = zvol_open(pp, flags, count);
2711 error = zvol_close(pp, flags, -count);
2717 zvol_geom_start(struct bio *bp)
2722 zv = bp->bio_to->private;
2724 switch (bp->bio_cmd) {
2726 if (!THREAD_CAN_SLEEP())
2728 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2729 g_io_deliver(bp, 0);
2734 if (!THREAD_CAN_SLEEP())
2739 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2740 uint64_t refd, avail, usedobjs, availobjs, val;
2742 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2744 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2745 dmu_objset_space(zv->zv_objset, &refd, &avail,
2746 &usedobjs, &availobjs);
2747 if (g_handleattr_off_t(bp, "blocksavail",
2750 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2751 dmu_objset_space(zv->zv_objset, &refd, &avail,
2752 &usedobjs, &availobjs);
2753 if (g_handleattr_off_t(bp, "blocksused",
2756 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2757 avail = metaslab_class_get_space(spa_normal_class(spa));
2758 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2759 if (g_handleattr_off_t(bp, "poolblocksavail",
2762 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2763 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2764 if (g_handleattr_off_t(bp, "poolblocksused",
2771 g_io_deliver(bp, EOPNOTSUPP);
2777 mtx_lock(&zv->zv_queue_mtx);
2778 first = (bioq_first(&zv->zv_queue) == NULL);
2779 bioq_insert_tail(&zv->zv_queue, bp);
2780 mtx_unlock(&zv->zv_queue_mtx);
2782 wakeup_one(&zv->zv_queue);
2786 zvol_geom_worker(void *arg)
2791 thread_lock(curthread);
2792 sched_prio(curthread, PRIBIO);
2793 thread_unlock(curthread);
2797 mtx_lock(&zv->zv_queue_mtx);
2798 bp = bioq_takefirst(&zv->zv_queue);
2800 if (zv->zv_state == 1) {
2802 wakeup(&zv->zv_state);
2803 mtx_unlock(&zv->zv_queue_mtx);
2806 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2810 mtx_unlock(&zv->zv_queue_mtx);
2811 switch (bp->bio_cmd) {
2813 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2814 g_io_deliver(bp, 0);
2822 g_io_deliver(bp, EOPNOTSUPP);
2828 extern boolean_t dataset_name_hidden(const char *name);
2831 zvol_create_snapshots(objset_t *os, const char *name)
2833 uint64_t cookie, obj;
2838 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2841 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2846 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2847 if (len >= MAXPATHLEN) {
2848 dmu_objset_rele(os, FTAG);
2849 error = ENAMETOOLONG;
2853 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2854 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2855 sname + len, &obj, &cookie, NULL);
2856 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2858 if (error == ENOENT)
2863 error = zvol_create_minor(sname);
2864 if (error != 0 && error != EEXIST) {
2865 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2871 kmem_free(sname, MAXPATHLEN);
2876 zvol_create_minors(const char *name)
2883 if (dataset_name_hidden(name))
2886 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2887 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2891 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2892 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2893 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2894 error = zvol_create_minor(name);
2895 if (error == 0 || error == EEXIST) {
2896 error = zvol_create_snapshots(os, name);
2898 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2901 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2902 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2905 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2906 dmu_objset_rele(os, FTAG);
2910 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2911 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2912 dmu_objset_rele(os, FTAG);
2913 kmem_free(osname, MAXPATHLEN);
2916 p = osname + strlen(osname);
2917 len = MAXPATHLEN - (p - osname);
2920 /* Prefetch the datasets. */
2922 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2923 if (!dataset_name_hidden(osname))
2924 (void) dmu_objset_prefetch(osname, NULL);
2929 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2931 dmu_objset_rele(os, FTAG);
2932 (void)zvol_create_minors(osname);
2933 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2934 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2940 dmu_objset_rele(os, FTAG);
2941 kmem_free(osname, MAXPATHLEN);
2946 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2949 struct g_provider *pp;
2952 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2954 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2956 pp = zv->zv_provider;
2961 zv->zv_provider = NULL;
2962 g_wither_provider(pp, ENXIO);
2964 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2965 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2966 pp->sectorsize = DEV_BSIZE;
2967 pp->mediasize = zv->zv_volsize;
2969 zv->zv_provider = pp;
2970 g_error_provider(pp, 0);
2971 g_topology_unlock();
2972 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2973 struct make_dev_args args;
2975 if ((dev = zv->zv_dev) != NULL) {
2978 if (zv->zv_total_opens > 0) {
2979 zv->zv_flags &= ~ZVOL_EXCL;
2980 zv->zv_total_opens = 0;
2981 zvol_last_close(zv);
2985 make_dev_args_init(&args);
2986 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
2987 args.mda_devsw = &zvol_cdevsw;
2989 args.mda_uid = UID_ROOT;
2990 args.mda_gid = GID_OPERATOR;
2991 args.mda_mode = 0640;
2992 args.mda_si_drv2 = zv;
2993 if (make_dev_s(&args, &zv->zv_dev,
2994 "%s/%s", ZVOL_DRIVER, newname) == 0)
2995 zv->zv_dev->si_iosize_max = MAXPHYS;
2997 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3001 zvol_rename_minors(const char *oldname, const char *newname)
3003 char name[MAXPATHLEN];
3004 struct g_provider *pp;
3006 size_t oldnamelen, newnamelen;
3009 boolean_t locked = B_FALSE;
3011 oldnamelen = strlen(oldname);
3012 newnamelen = strlen(newname);
3015 /* See comment in zvol_open(). */
3016 if (!MUTEX_HELD(&zfsdev_state_lock)) {
3017 mutex_enter(&zfsdev_state_lock);
3021 LIST_FOREACH(zv, &all_zvols, zv_links) {
3022 if (strcmp(zv->zv_name, oldname) == 0) {
3023 zvol_rename_minor(zv, newname);
3024 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3025 (zv->zv_name[oldnamelen] == '/' ||
3026 zv->zv_name[oldnamelen] == '@')) {
3027 snprintf(name, sizeof(name), "%s%c%s", newname,
3028 zv->zv_name[oldnamelen],
3029 zv->zv_name + oldnamelen + 1);
3030 zvol_rename_minor(zv, name);
3035 mutex_exit(&zfsdev_state_lock);
3040 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3042 zvol_state_t *zv = dev->si_drv2;
3045 mutex_enter(&zfsdev_state_lock);
3046 if (zv->zv_total_opens == 0)
3047 err = zvol_first_open(zv);
3049 mutex_exit(&zfsdev_state_lock);
3052 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3053 err = SET_ERROR(EROFS);
3056 if (zv->zv_flags & ZVOL_EXCL) {
3057 err = SET_ERROR(EBUSY);
3061 if (flags & FEXCL) {
3062 if (zv->zv_total_opens != 0) {
3063 err = SET_ERROR(EBUSY);
3066 zv->zv_flags |= ZVOL_EXCL;
3070 zv->zv_total_opens++;
3071 if (flags & (FSYNC | FDSYNC)) {
3073 if (zv->zv_sync_cnt == 1)
3074 zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
3076 mutex_exit(&zfsdev_state_lock);
3079 if (zv->zv_total_opens == 0)
3080 zvol_last_close(zv);
3081 mutex_exit(&zfsdev_state_lock);
3086 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3088 zvol_state_t *zv = dev->si_drv2;
3090 mutex_enter(&zfsdev_state_lock);
3091 if (zv->zv_flags & ZVOL_EXCL) {
3092 ASSERT(zv->zv_total_opens == 1);
3093 zv->zv_flags &= ~ZVOL_EXCL;
3097 * If the open count is zero, this is a spurious close.
3098 * That indicates a bug in the kernel / DDI framework.
3100 ASSERT(zv->zv_total_opens != 0);
3103 * You may get multiple opens, but only one close.
3105 zv->zv_total_opens--;
3106 if (flags & (FSYNC | FDSYNC))
3109 if (zv->zv_total_opens == 0)
3110 zvol_last_close(zv);
3112 mutex_exit(&zfsdev_state_lock);
3117 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3121 off_t offset, length;
3128 KASSERT(zv->zv_total_opens > 0,
3129 ("Device with zero access count in zvol_d_ioctl"));
3131 i = IOCPARM_LEN(cmd);
3133 case DIOCGSECTORSIZE:
3134 *(u_int *)data = DEV_BSIZE;
3136 case DIOCGMEDIASIZE:
3137 *(off_t *)data = zv->zv_volsize;
3140 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3143 if (!zvol_unmap_enabled)
3146 offset = ((off_t *)data)[0];
3147 length = ((off_t *)data)[1];
3148 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3149 offset < 0 || offset >= zv->zv_volsize ||
3151 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3157 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3158 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3159 error = dmu_tx_assign(tx, TXG_WAIT);
3164 sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
3165 zvol_log_truncate(zv, tx, offset, length, sync);
3167 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3170 zfs_range_unlock(rl);
3172 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3174 case DIOCGSTRIPESIZE:
3175 *(off_t *)data = zv->zv_volblocksize;
3177 case DIOCGSTRIPEOFFSET:
3181 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3182 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3183 uint64_t refd, avail, usedobjs, availobjs;
3185 if (strcmp(arg->name, "GEOM::candelete") == 0)
3187 else if (strcmp(arg->name, "blocksavail") == 0) {
3188 dmu_objset_space(zv->zv_objset, &refd, &avail,
3189 &usedobjs, &availobjs);
3190 arg->value.off = avail / DEV_BSIZE;
3191 } else if (strcmp(arg->name, "blocksused") == 0) {
3192 dmu_objset_space(zv->zv_objset, &refd, &avail,
3193 &usedobjs, &availobjs);
3194 arg->value.off = refd / DEV_BSIZE;
3195 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3196 avail = metaslab_class_get_space(spa_normal_class(spa));
3197 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3198 arg->value.off = avail / DEV_BSIZE;
3199 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3200 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3201 arg->value.off = refd / DEV_BSIZE;
3208 off_t *off = (off_t *)data;
3212 hole = (cmd == FIOSEEKHOLE);
3214 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3224 #endif /* illumos */