4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2006-2010 Pawel Jakub Dawidek <pjd@FreeBSD.org>
25 * All rights reserved.
27 * Portions Copyright 2010 Robert Milkowski
29 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
30 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
31 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32 * Copyright (c) 2014 Integros [integros.com]
35 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
38 * ZFS volume emulation driver.
40 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
41 * Volumes are accessed through the symbolic links named:
43 * /dev/zvol/dsk/<pool_name>/<dataset_name>
44 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
46 * These links are created by the /dev filesystem (sdev_zvolops.c).
47 * Volumes are persistent through reboot. No user command needs to be
48 * run before opening and using a device.
51 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
55 #include <sys/types.h>
56 #include <sys/param.h>
57 #include <sys/kernel.h>
58 #include <sys/errno.h>
64 #include <sys/cmn_err.h>
68 #include <sys/spa_impl.h>
71 #include <sys/dmu_traverse.h>
72 #include <sys/dnode.h>
73 #include <sys/dsl_dataset.h>
74 #include <sys/dsl_prop.h>
76 #include <sys/byteorder.h>
77 #include <sys/sunddi.h>
78 #include <sys/dirent.h>
79 #include <sys/policy.h>
80 #include <sys/queue.h>
81 #include <sys/fs/zfs.h>
82 #include <sys/zfs_ioctl.h>
84 #include <sys/refcount.h>
85 #include <sys/zfs_znode.h>
86 #include <sys/zfs_rlock.h>
87 #include <sys/vdev_impl.h>
88 #include <sys/vdev_raidz.h>
90 #include <sys/zil_impl.h>
92 #include <sys/dmu_tx.h>
93 #include <sys/zfeature.h>
94 #include <sys/zio_checksum.h>
95 #include <sys/filio.h>
97 #include <geom/geom.h>
99 #include "zfs_namecheck.h"
102 struct g_class zfs_zvol_class = {
104 .version = G_VERSION,
107 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
111 static char *zvol_tag = "zvol_tag";
113 #define ZVOL_DUMPSIZE "dumpsize"
116 * This lock protects the zfsdev_state structure from being modified
117 * while it's being used, e.g. an open that comes in before a create
118 * finishes. It also protects temporary opens of the dataset so that,
119 * e.g., an open doesn't get a spurious EBUSY.
122 kmutex_t zfsdev_state_lock;
125 * In FreeBSD we've replaced the upstream zfsdev_state_lock with the
126 * spa_namespace_lock in the ZVOL code.
128 #define zfsdev_state_lock spa_namespace_lock
130 static uint32_t zvol_minors;
133 SYSCTL_DECL(_vfs_zfs);
134 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
135 static int volmode = ZFS_VOLMODE_GEOM;
136 TUNABLE_INT("vfs.zfs.vol.mode", &volmode);
137 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
138 "Expose as GEOM providers (1), device files (2) or neither");
141 typedef struct zvol_extent {
143 dva_t ze_dva; /* dva associated with this extent */
144 uint64_t ze_nblks; /* number of blocks in extent */
148 * The in-core state of each volume.
150 typedef struct zvol_state {
152 LIST_ENTRY(zvol_state) zv_links;
154 char zv_name[MAXPATHLEN]; /* pool/dd name */
155 uint64_t zv_volsize; /* amount of space we advertise */
156 uint64_t zv_volblocksize; /* volume block size */
158 minor_t zv_minor; /* minor number */
160 struct cdev *zv_dev; /* non-GEOM device */
161 struct g_provider *zv_provider; /* GEOM provider */
163 uint8_t zv_min_bs; /* minimum addressable block shift */
164 uint8_t zv_flags; /* readonly, dumpified, etc. */
165 objset_t *zv_objset; /* objset handle */
167 uint32_t zv_open_count[OTYPCNT]; /* open counts */
169 uint32_t zv_total_opens; /* total open count */
170 zilog_t *zv_zilog; /* ZIL handle */
171 list_t zv_extents; /* List of extents for dump */
172 znode_t zv_znode; /* for range locking */
173 dmu_buf_t *zv_dbuf; /* bonus handle */
176 int zv_volmode; /* Provide GEOM or cdev */
177 struct bio_queue_head zv_queue;
178 struct mtx zv_queue_mtx; /* zv_queue mutex */
183 static LIST_HEAD(, zvol_state) all_zvols;
186 * zvol specific flags
188 #define ZVOL_RDONLY 0x1
189 #define ZVOL_DUMPIFIED 0x2
190 #define ZVOL_EXCL 0x4
194 * zvol maximum transfer in one DMU tx.
196 int zvol_maxphys = DMU_MAX_ACCESS/2;
199 * Toggle unmap functionality.
201 boolean_t zvol_unmap_enabled = B_TRUE;
203 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
204 &zvol_unmap_enabled, 0,
205 "Enable UNMAP functionality");
207 static d_open_t zvol_d_open;
208 static d_close_t zvol_d_close;
209 static d_read_t zvol_read;
210 static d_write_t zvol_write;
211 static d_ioctl_t zvol_d_ioctl;
212 static d_strategy_t zvol_strategy;
214 static struct cdevsw zvol_cdevsw = {
215 .d_version = D_VERSION,
216 .d_open = zvol_d_open,
217 .d_close = zvol_d_close,
219 .d_write = zvol_write,
220 .d_ioctl = zvol_d_ioctl,
221 .d_strategy = zvol_strategy,
223 .d_flags = D_DISK | D_TRACKCLOSE,
226 static void zvol_geom_run(zvol_state_t *zv);
227 static void zvol_geom_destroy(zvol_state_t *zv);
228 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
229 static void zvol_geom_start(struct bio *bp);
230 static void zvol_geom_worker(void *arg);
231 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
232 uint64_t len, boolean_t sync);
233 #endif /* !illumos */
235 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
236 nvlist_t *, nvlist_t *);
237 static int zvol_remove_zv(zvol_state_t *);
238 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
239 static int zvol_dumpify(zvol_state_t *zv);
240 static int zvol_dump_fini(zvol_state_t *zv);
241 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
244 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
247 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
249 zv->zv_volsize = volsize;
250 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
251 "Size", volsize) == DDI_SUCCESS);
252 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
253 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
255 /* Notify specfs to invalidate the cached size */
256 spec_size_invalidate(dev, VBLK);
257 spec_size_invalidate(dev, VCHR);
259 zv->zv_volsize = volsize;
260 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
261 struct g_provider *pp;
263 pp = zv->zv_provider;
267 g_resize_provider(pp, zv->zv_volsize);
274 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
277 return (SET_ERROR(EINVAL));
279 if (volsize % blocksize != 0)
280 return (SET_ERROR(EINVAL));
283 if (volsize - 1 > SPEC_MAXOFFSET_T)
284 return (SET_ERROR(EOVERFLOW));
290 zvol_check_volblocksize(uint64_t volblocksize)
292 if (volblocksize < SPA_MINBLOCKSIZE ||
293 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
295 return (SET_ERROR(EDOM));
301 zvol_get_stats(objset_t *os, nvlist_t *nv)
304 dmu_object_info_t doi;
307 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
311 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
313 error = dmu_object_info(os, ZVOL_OBJ, &doi);
316 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
317 doi.doi_data_block_size);
323 static zvol_state_t *
324 zvol_minor_lookup(const char *name)
331 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
334 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
335 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
339 LIST_FOREACH(zv, &all_zvols, zv_links) {
341 if (strcmp(zv->zv_name, name) == 0)
348 /* extent mapping arg */
356 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
357 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
359 struct maparg *ma = arg;
361 int bs = ma->ma_zv->zv_volblocksize;
363 if (bp == NULL || BP_IS_HOLE(bp) ||
364 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
367 VERIFY(!BP_IS_EMBEDDED(bp));
369 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
372 /* Abort immediately if we have encountered gang blocks */
374 return (SET_ERROR(EFRAGS));
377 * See if the block is at the end of the previous extent.
379 ze = list_tail(&ma->ma_zv->zv_extents);
381 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
382 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
383 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
388 dprintf_bp(bp, "%s", "next blkptr:");
390 /* start a new extent */
391 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
392 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
394 list_insert_tail(&ma->ma_zv->zv_extents, ze);
399 zvol_free_extents(zvol_state_t *zv)
403 while (ze = list_head(&zv->zv_extents)) {
404 list_remove(&zv->zv_extents, ze);
405 kmem_free(ze, sizeof (zvol_extent_t));
410 zvol_get_lbas(zvol_state_t *zv)
412 objset_t *os = zv->zv_objset;
418 zvol_free_extents(zv);
420 /* commit any in-flight changes before traversing the dataset */
421 txg_wait_synced(dmu_objset_pool(os), 0);
422 err = traverse_dataset(dmu_objset_ds(os), 0,
423 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
424 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
425 zvol_free_extents(zv);
426 return (err ? err : EIO);
434 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
436 zfs_creat_t *zct = arg;
437 nvlist_t *nvprops = zct->zct_props;
439 uint64_t volblocksize, volsize;
441 VERIFY(nvlist_lookup_uint64(nvprops,
442 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
443 if (nvlist_lookup_uint64(nvprops,
444 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
445 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
448 * These properties must be removed from the list so the generic
449 * property setting step won't apply to them.
451 VERIFY(nvlist_remove_all(nvprops,
452 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
453 (void) nvlist_remove_all(nvprops,
454 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
456 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
460 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
464 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
469 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
470 * implement DKIOCFREE/free-long-range.
473 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
475 uint64_t offset, length;
478 byteswap_uint64_array(lr, sizeof (*lr));
480 offset = lr->lr_offset;
481 length = lr->lr_length;
483 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
487 * Replay a TX_WRITE ZIL transaction that didn't get committed
488 * after a system failure
491 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
493 objset_t *os = zv->zv_objset;
494 char *data = (char *)(lr + 1); /* data follows lr_write_t */
495 uint64_t offset, length;
500 byteswap_uint64_array(lr, sizeof (*lr));
502 offset = lr->lr_offset;
503 length = lr->lr_length;
505 /* If it's a dmu_sync() block, write the whole block */
506 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
507 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
508 if (length < blocksize) {
509 offset -= offset % blocksize;
514 tx = dmu_tx_create(os);
515 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
516 error = dmu_tx_assign(tx, TXG_WAIT);
520 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
529 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
531 return (SET_ERROR(ENOTSUP));
535 * Callback vectors for replaying records.
536 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
538 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
539 zvol_replay_err, /* 0 no such transaction type */
540 zvol_replay_err, /* TX_CREATE */
541 zvol_replay_err, /* TX_MKDIR */
542 zvol_replay_err, /* TX_MKXATTR */
543 zvol_replay_err, /* TX_SYMLINK */
544 zvol_replay_err, /* TX_REMOVE */
545 zvol_replay_err, /* TX_RMDIR */
546 zvol_replay_err, /* TX_LINK */
547 zvol_replay_err, /* TX_RENAME */
548 zvol_replay_write, /* TX_WRITE */
549 zvol_replay_truncate, /* TX_TRUNCATE */
550 zvol_replay_err, /* TX_SETATTR */
551 zvol_replay_err, /* TX_ACL */
552 zvol_replay_err, /* TX_CREATE_ACL */
553 zvol_replay_err, /* TX_CREATE_ATTR */
554 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
555 zvol_replay_err, /* TX_MKDIR_ACL */
556 zvol_replay_err, /* TX_MKDIR_ATTR */
557 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
558 zvol_replay_err, /* TX_WRITE2 */
563 zvol_name2minor(const char *name, minor_t *minor)
567 mutex_enter(&zfsdev_state_lock);
568 zv = zvol_minor_lookup(name);
570 *minor = zv->zv_minor;
571 mutex_exit(&zfsdev_state_lock);
572 return (zv ? 0 : -1);
577 * Create a minor node (plus a whole lot more) for the specified volume.
580 zvol_create_minor(const char *name)
582 zfs_soft_state_t *zs;
585 dmu_object_info_t doi;
588 char chrbuf[30], blkbuf[30];
590 struct g_provider *pp;
592 uint64_t volsize, mode;
597 ZFS_LOG(1, "Creating ZVOL %s...", name);
600 mutex_enter(&zfsdev_state_lock);
602 if (zvol_minor_lookup(name) != NULL) {
603 mutex_exit(&zfsdev_state_lock);
604 return (SET_ERROR(EEXIST));
607 /* lie and say we're read-only */
608 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
611 mutex_exit(&zfsdev_state_lock);
616 if ((minor = zfsdev_minor_alloc()) == 0) {
617 dmu_objset_disown(os, FTAG);
618 mutex_exit(&zfsdev_state_lock);
619 return (SET_ERROR(ENXIO));
622 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
623 dmu_objset_disown(os, FTAG);
624 mutex_exit(&zfsdev_state_lock);
625 return (SET_ERROR(EAGAIN));
627 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
630 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
632 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
633 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
634 ddi_soft_state_free(zfsdev_state, minor);
635 dmu_objset_disown(os, FTAG);
636 mutex_exit(&zfsdev_state_lock);
637 return (SET_ERROR(EAGAIN));
640 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
642 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
643 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
644 ddi_remove_minor_node(zfs_dip, chrbuf);
645 ddi_soft_state_free(zfsdev_state, minor);
646 dmu_objset_disown(os, FTAG);
647 mutex_exit(&zfsdev_state_lock);
648 return (SET_ERROR(EAGAIN));
651 zs = ddi_get_soft_state(zfsdev_state, minor);
652 zs->zss_type = ZSST_ZVOL;
653 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
656 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
658 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
660 kmem_free(zv, sizeof(*zv));
661 dmu_objset_disown(os, zvol_tag);
662 mutex_exit(&zfsdev_state_lock);
665 error = dsl_prop_get_integer(name,
666 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
667 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
671 zv->zv_volsize = volsize;
672 zv->zv_volmode = mode;
673 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
675 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
676 gp->start = zvol_geom_start;
677 gp->access = zvol_geom_access;
678 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
679 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
680 pp->sectorsize = DEV_BSIZE;
681 pp->mediasize = zv->zv_volsize;
684 zv->zv_provider = pp;
685 bioq_init(&zv->zv_queue);
686 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
687 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
688 struct make_dev_args args;
690 make_dev_args_init(&args);
691 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
692 args.mda_devsw = &zvol_cdevsw;
694 args.mda_uid = UID_ROOT;
695 args.mda_gid = GID_OPERATOR;
696 args.mda_mode = 0640;
697 args.mda_si_drv2 = zv;
698 error = make_dev_s(&args, &zv->zv_dev,
699 "%s/%s", ZVOL_DRIVER, name);
701 kmem_free(zv, sizeof(*zv));
702 dmu_objset_disown(os, FTAG);
703 mutex_exit(&zfsdev_state_lock);
706 zv->zv_dev->si_iosize_max = MAXPHYS;
708 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
711 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
712 zv->zv_min_bs = DEV_BSHIFT;
714 zv->zv_minor = minor;
717 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
718 zv->zv_flags |= ZVOL_RDONLY;
719 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
720 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
721 sizeof (rl_t), offsetof(rl_t, r_node));
722 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
723 offsetof(zvol_extent_t, ze_node));
724 /* get and cache the blocksize */
725 error = dmu_object_info(os, ZVOL_OBJ, &doi);
727 zv->zv_volblocksize = doi.doi_data_block_size;
729 if (spa_writeable(dmu_objset_spa(os))) {
730 if (zil_replay_disable)
731 zil_destroy(dmu_objset_zil(os), B_FALSE);
733 zil_replay(os, zv, zvol_replay_vector);
735 dmu_objset_disown(os, FTAG);
736 zv->zv_objset = NULL;
740 mutex_exit(&zfsdev_state_lock);
742 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
748 ZFS_LOG(1, "ZVOL %s created.", name);
755 * Remove minor node for the specified volume.
758 zvol_remove_zv(zvol_state_t *zv)
762 minor_t minor = zv->zv_minor;
765 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
766 if (zv->zv_total_opens != 0)
767 return (SET_ERROR(EBUSY));
770 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
771 ddi_remove_minor_node(zfs_dip, nmbuf);
773 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
774 ddi_remove_minor_node(zfs_dip, nmbuf);
776 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
778 LIST_REMOVE(zv, zv_links);
779 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
781 zvol_geom_destroy(zv);
783 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV)
784 destroy_dev(zv->zv_dev);
787 avl_destroy(&zv->zv_znode.z_range_avl);
788 mutex_destroy(&zv->zv_znode.z_range_lock);
790 kmem_free(zv, sizeof (zvol_state_t));
792 ddi_soft_state_free(zfsdev_state, minor);
799 zvol_remove_minor(const char *name)
804 mutex_enter(&zfsdev_state_lock);
805 if ((zv = zvol_minor_lookup(name)) == NULL) {
806 mutex_exit(&zfsdev_state_lock);
807 return (SET_ERROR(ENXIO));
809 rc = zvol_remove_zv(zv);
810 mutex_exit(&zfsdev_state_lock);
815 zvol_first_open(zvol_state_t *zv)
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 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
838 dmu_objset_disown(os, zvol_tag);
842 zvol_size_changed(zv, volsize);
843 zv->zv_zilog = zil_open(os, zvol_get_data);
845 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
847 if (readonly || dmu_objset_is_snapshot(os) ||
848 !spa_writeable(dmu_objset_spa(os)))
849 zv->zv_flags |= ZVOL_RDONLY;
851 zv->zv_flags &= ~ZVOL_RDONLY;
856 zvol_last_close(zvol_state_t *zv)
858 zil_close(zv->zv_zilog);
861 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
867 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
868 !(zv->zv_flags & ZVOL_RDONLY))
869 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
870 dmu_objset_evict_dbufs(zv->zv_objset);
872 dmu_objset_disown(zv->zv_objset, zvol_tag);
873 zv->zv_objset = NULL;
878 zvol_prealloc(zvol_state_t *zv)
880 objset_t *os = zv->zv_objset;
882 uint64_t refd, avail, usedobjs, availobjs;
883 uint64_t resid = zv->zv_volsize;
886 /* Check the space usage before attempting to allocate the space */
887 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
888 if (avail < zv->zv_volsize)
889 return (SET_ERROR(ENOSPC));
891 /* Free old extents if they exist */
892 zvol_free_extents(zv);
896 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
898 tx = dmu_tx_create(os);
899 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
900 error = dmu_tx_assign(tx, TXG_WAIT);
903 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
906 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
911 txg_wait_synced(dmu_objset_pool(os), 0);
918 zvol_update_volsize(objset_t *os, uint64_t volsize)
923 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
925 tx = dmu_tx_create(os);
926 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
927 dmu_tx_mark_netfree(tx);
928 error = dmu_tx_assign(tx, TXG_WAIT);
934 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
939 error = dmu_free_long_range(os,
940 ZVOL_OBJ, volsize, DMU_OBJECT_END);
945 zvol_remove_minors(const char *name)
952 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
953 (void) strncpy(namebuf, name, strlen(name));
954 (void) strcat(namebuf, "/");
955 mutex_enter(&zfsdev_state_lock);
956 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
958 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
961 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
962 (void) zvol_remove_zv(zv);
964 kmem_free(namebuf, strlen(name) + 2);
966 mutex_exit(&zfsdev_state_lock);
968 zvol_state_t *zv, *tzv;
971 namelen = strlen(name);
974 mutex_enter(&zfsdev_state_lock);
976 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
977 if (strcmp(zv->zv_name, name) == 0 ||
978 (strncmp(zv->zv_name, name, namelen) == 0 &&
979 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
980 zv->zv_name[namelen] == '@'))) {
981 (void) zvol_remove_zv(zv);
985 mutex_exit(&zfsdev_state_lock);
991 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
993 uint64_t old_volsize = 0ULL;
996 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
999 * Reinitialize the dump area to the new size. If we
1000 * failed to resize the dump area then restore it back to
1001 * its original size. We must set the new volsize prior
1002 * to calling dumpvp_resize() to ensure that the devices'
1003 * size(9P) is not visible by the dump subsystem.
1005 old_volsize = zv->zv_volsize;
1006 zvol_size_changed(zv, volsize);
1009 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1010 if ((error = zvol_dumpify(zv)) != 0 ||
1011 (error = dumpvp_resize()) != 0) {
1014 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1015 zvol_size_changed(zv, old_volsize);
1016 dumpify_error = zvol_dumpify(zv);
1017 error = dumpify_error ? dumpify_error : error;
1020 #endif /* ZVOL_DUMP */
1024 * Generate a LUN expansion event.
1029 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1031 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1034 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1035 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1037 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1038 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1041 kmem_free(physpath, MAXPATHLEN);
1043 #endif /* illumos */
1048 zvol_set_volsize(const char *name, uint64_t volsize)
1050 zvol_state_t *zv = NULL;
1053 dmu_object_info_t doi;
1055 boolean_t owned = B_FALSE;
1057 error = dsl_prop_get_integer(name,
1058 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1062 return (SET_ERROR(EROFS));
1064 mutex_enter(&zfsdev_state_lock);
1065 zv = zvol_minor_lookup(name);
1067 if (zv == NULL || zv->zv_objset == NULL) {
1068 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1070 mutex_exit(&zfsdev_state_lock);
1080 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1081 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1084 error = zvol_update_volsize(os, volsize);
1086 if (error == 0 && zv != NULL)
1087 error = zvol_update_live_volsize(zv, volsize);
1090 dmu_objset_disown(os, FTAG);
1092 zv->zv_objset = NULL;
1094 mutex_exit(&zfsdev_state_lock);
1101 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1104 zvol_open(struct g_provider *pp, int flag, int count)
1111 mutex_enter(&zfsdev_state_lock);
1113 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1115 mutex_exit(&zfsdev_state_lock);
1116 return (SET_ERROR(ENXIO));
1119 if (zv->zv_total_opens == 0)
1120 err = zvol_first_open(zv);
1122 mutex_exit(&zfsdev_state_lock);
1125 #else /* !illumos */
1126 boolean_t locked = B_FALSE;
1129 * Protect against recursively entering spa_namespace_lock
1130 * when spa_open() is used for a pool on a (local) ZVOL(s).
1131 * This is needed since we replaced upstream zfsdev_state_lock
1132 * with spa_namespace_lock in the ZVOL code.
1133 * We are using the same trick as spa_open().
1134 * Note that calls in zvol_first_open which need to resolve
1135 * pool name to a spa object will enter spa_open()
1136 * recursively, but that function already has all the
1137 * necessary protection.
1139 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1140 mutex_enter(&zfsdev_state_lock);
1147 mutex_exit(&zfsdev_state_lock);
1148 return (SET_ERROR(ENXIO));
1151 if (zv->zv_total_opens == 0) {
1152 err = zvol_first_open(zv);
1155 mutex_exit(&zfsdev_state_lock);
1158 pp->mediasize = zv->zv_volsize;
1159 pp->stripeoffset = 0;
1160 pp->stripesize = zv->zv_volblocksize;
1162 #endif /* illumos */
1163 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1164 err = SET_ERROR(EROFS);
1167 if (zv->zv_flags & ZVOL_EXCL) {
1168 err = SET_ERROR(EBUSY);
1173 if (zv->zv_total_opens != 0) {
1174 err = SET_ERROR(EBUSY);
1177 zv->zv_flags |= ZVOL_EXCL;
1182 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1183 zv->zv_open_count[otyp]++;
1184 zv->zv_total_opens++;
1186 mutex_exit(&zfsdev_state_lock);
1188 zv->zv_total_opens += count;
1190 mutex_exit(&zfsdev_state_lock);
1195 if (zv->zv_total_opens == 0)
1196 zvol_last_close(zv);
1198 mutex_exit(&zfsdev_state_lock);
1201 mutex_exit(&zfsdev_state_lock);
1209 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1211 minor_t minor = getminor(dev);
1215 mutex_enter(&zfsdev_state_lock);
1217 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1219 mutex_exit(&zfsdev_state_lock);
1220 #else /* !illumos */
1222 zvol_close(struct g_provider *pp, int flag, int count)
1226 boolean_t locked = B_FALSE;
1228 /* See comment in zvol_open(). */
1229 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1230 mutex_enter(&zfsdev_state_lock);
1237 mutex_exit(&zfsdev_state_lock);
1238 #endif /* illumos */
1239 return (SET_ERROR(ENXIO));
1242 if (zv->zv_flags & ZVOL_EXCL) {
1243 ASSERT(zv->zv_total_opens == 1);
1244 zv->zv_flags &= ~ZVOL_EXCL;
1248 * If the open count is zero, this is a spurious close.
1249 * That indicates a bug in the kernel / DDI framework.
1252 ASSERT(zv->zv_open_count[otyp] != 0);
1254 ASSERT(zv->zv_total_opens != 0);
1257 * You may get multiple opens, but only one close.
1260 zv->zv_open_count[otyp]--;
1261 zv->zv_total_opens--;
1263 zv->zv_total_opens -= count;
1266 if (zv->zv_total_opens == 0)
1267 zvol_last_close(zv);
1270 mutex_exit(&zfsdev_state_lock);
1273 mutex_exit(&zfsdev_state_lock);
1279 zvol_get_done(zgd_t *zgd, int error)
1282 dmu_buf_rele(zgd->zgd_db, zgd);
1284 zfs_range_unlock(zgd->zgd_rl);
1286 if (error == 0 && zgd->zgd_bp)
1287 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1289 kmem_free(zgd, sizeof (zgd_t));
1293 * Get data to generate a TX_WRITE intent log record.
1296 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1298 zvol_state_t *zv = arg;
1299 objset_t *os = zv->zv_objset;
1300 uint64_t object = ZVOL_OBJ;
1301 uint64_t offset = lr->lr_offset;
1302 uint64_t size = lr->lr_length; /* length of user data */
1303 blkptr_t *bp = &lr->lr_blkptr;
1308 ASSERT(zio != NULL);
1311 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1312 zgd->zgd_zilog = zv->zv_zilog;
1313 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1316 * Write records come in two flavors: immediate and indirect.
1317 * For small writes it's cheaper to store the data with the
1318 * log record (immediate); for large writes it's cheaper to
1319 * sync the data and get a pointer to it (indirect) so that
1320 * we don't have to write the data twice.
1322 if (buf != NULL) { /* immediate write */
1323 error = dmu_read(os, object, offset, size, buf,
1324 DMU_READ_NO_PREFETCH);
1326 size = zv->zv_volblocksize;
1327 offset = P2ALIGN(offset, size);
1328 error = dmu_buf_hold(os, object, offset, zgd, &db,
1329 DMU_READ_NO_PREFETCH);
1331 blkptr_t *obp = dmu_buf_get_blkptr(db);
1333 ASSERT(BP_IS_HOLE(bp));
1340 ASSERT(db->db_offset == offset);
1341 ASSERT(db->db_size == size);
1343 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1344 zvol_get_done, zgd);
1351 zvol_get_done(zgd, error);
1357 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1359 * We store data in the log buffers if it's small enough.
1360 * Otherwise we will later flush the data out via dmu_sync().
1362 ssize_t zvol_immediate_write_sz = 32768;
1365 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1368 uint32_t blocksize = zv->zv_volblocksize;
1369 zilog_t *zilog = zv->zv_zilog;
1371 ssize_t immediate_write_sz;
1373 if (zil_replaying(zilog, tx))
1376 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1377 ? 0 : zvol_immediate_write_sz;
1379 slogging = spa_has_slogs(zilog->zl_spa) &&
1380 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1386 itx_wr_state_t write_state;
1389 * Unlike zfs_log_write() we can be called with
1390 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1392 if (blocksize > immediate_write_sz && !slogging &&
1393 resid >= blocksize && off % blocksize == 0) {
1394 write_state = WR_INDIRECT; /* uses dmu_sync */
1397 write_state = WR_COPIED;
1398 len = MIN(ZIL_MAX_LOG_DATA, resid);
1400 write_state = WR_NEED_COPY;
1401 len = MIN(ZIL_MAX_LOG_DATA, resid);
1404 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1405 (write_state == WR_COPIED ? len : 0));
1406 lr = (lr_write_t *)&itx->itx_lr;
1407 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1408 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1409 zil_itx_destroy(itx);
1410 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1411 lr = (lr_write_t *)&itx->itx_lr;
1412 write_state = WR_NEED_COPY;
1415 itx->itx_wr_state = write_state;
1416 if (write_state == WR_NEED_COPY)
1417 itx->itx_sod += len;
1418 lr->lr_foid = ZVOL_OBJ;
1419 lr->lr_offset = off;
1420 lr->lr_length = len;
1422 BP_ZERO(&lr->lr_blkptr);
1424 itx->itx_private = zv;
1425 itx->itx_sync = sync;
1427 zil_itx_assign(zilog, itx, tx);
1436 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1437 uint64_t size, boolean_t doread, boolean_t isdump)
1443 if (vd->vdev_ops == &vdev_mirror_ops ||
1444 vd->vdev_ops == &vdev_replacing_ops ||
1445 vd->vdev_ops == &vdev_spare_ops) {
1446 for (c = 0; c < vd->vdev_children; c++) {
1447 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1448 addr, offset, origoffset, size, doread, isdump);
1451 } else if (doread) {
1457 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1458 return (numerrors < vd->vdev_children ? 0 : EIO);
1460 if (doread && !vdev_readable(vd))
1461 return (SET_ERROR(EIO));
1462 else if (!doread && !vdev_writeable(vd))
1463 return (SET_ERROR(EIO));
1465 if (vd->vdev_ops == &vdev_raidz_ops) {
1466 return (vdev_raidz_physio(vd,
1467 addr, size, offset, origoffset, doread, isdump));
1470 offset += VDEV_LABEL_START_SIZE;
1472 if (ddi_in_panic() || isdump) {
1475 return (SET_ERROR(EIO));
1477 ASSERT3P(dvd, !=, NULL);
1478 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1482 ASSERT3P(dvd, !=, NULL);
1483 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1484 offset, doread ? B_READ : B_WRITE));
1489 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1490 boolean_t doread, boolean_t isdump)
1495 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1497 /* Must be sector aligned, and not stradle a block boundary. */
1498 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1499 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1500 return (SET_ERROR(EINVAL));
1502 ASSERT(size <= zv->zv_volblocksize);
1504 /* Locate the extent this belongs to */
1505 ze = list_head(&zv->zv_extents);
1506 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1507 offset -= ze->ze_nblks * zv->zv_volblocksize;
1508 ze = list_next(&zv->zv_extents, ze);
1512 return (SET_ERROR(EINVAL));
1514 if (!ddi_in_panic())
1515 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1517 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1518 offset += DVA_GET_OFFSET(&ze->ze_dva);
1519 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1520 size, doread, isdump);
1522 if (!ddi_in_panic())
1523 spa_config_exit(spa, SCL_STATE, FTAG);
1529 zvol_strategy(buf_t *bp)
1531 zfs_soft_state_t *zs = NULL;
1532 #else /* !illumos */
1534 zvol_strategy(struct bio *bp)
1536 #endif /* illumos */
1538 uint64_t off, volsize;
1545 boolean_t doread = bp->b_flags & B_READ;
1547 boolean_t doread = 0;
1549 boolean_t is_dumpified;
1553 if (getminor(bp->b_edev) == 0) {
1554 error = SET_ERROR(EINVAL);
1556 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1558 error = SET_ERROR(ENXIO);
1559 else if (zs->zss_type != ZSST_ZVOL)
1560 error = SET_ERROR(EINVAL);
1564 bioerror(bp, error);
1571 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1572 bioerror(bp, EROFS);
1577 off = ldbtob(bp->b_blkno);
1578 #else /* !illumos */
1580 zv = bp->bio_to->private;
1582 zv = bp->bio_dev->si_drv2;
1585 error = SET_ERROR(ENXIO);
1589 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1590 error = SET_ERROR(EROFS);
1594 switch (bp->bio_cmd) {
1607 off = bp->bio_offset;
1608 #endif /* illumos */
1609 volsize = zv->zv_volsize;
1616 addr = bp->b_un.b_addr;
1617 resid = bp->b_bcount;
1619 if (resid > 0 && (off < 0 || off >= volsize)) {
1625 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1626 sync = ((!(bp->b_flags & B_ASYNC) &&
1627 !(zv->zv_flags & ZVOL_WCE)) ||
1628 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1629 !doread && !is_dumpified;
1630 #else /* !illumos */
1631 addr = bp->bio_data;
1632 resid = bp->bio_length;
1634 if (resid > 0 && (off < 0 || off >= volsize)) {
1635 error = SET_ERROR(EIO);
1639 is_dumpified = B_FALSE;
1640 sync = !doread && !is_dumpified &&
1641 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1642 #endif /* illumos */
1645 * There must be no buffer changes when doing a dmu_sync() because
1646 * we can't change the data whilst calculating the checksum.
1648 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1649 doread ? RL_READER : RL_WRITER);
1652 if (bp->bio_cmd == BIO_DELETE) {
1653 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1654 error = dmu_tx_assign(tx, TXG_WAIT);
1658 zvol_log_truncate(zv, tx, off, resid, B_TRUE);
1660 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1667 while (resid != 0 && off < volsize) {
1668 size_t size = MIN(resid, zvol_maxphys);
1671 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1672 error = zvol_dumpio(zv, addr, off, size,
1674 } else if (doread) {
1678 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1681 dmu_tx_t *tx = dmu_tx_create(os);
1682 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1683 error = dmu_tx_assign(tx, TXG_WAIT);
1687 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1688 zvol_log_write(zv, tx, off, size, sync);
1693 /* convert checksum errors into IO errors */
1694 if (error == ECKSUM)
1695 error = SET_ERROR(EIO);
1705 zfs_range_unlock(rl);
1708 if ((bp->b_resid = resid) == bp->b_bcount)
1709 bioerror(bp, off > volsize ? EINVAL : error);
1712 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1716 #else /* !illumos */
1717 bp->bio_completed = bp->bio_length - resid;
1718 if (bp->bio_completed < bp->bio_length && off > volsize)
1723 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1727 g_io_deliver(bp, error);
1729 biofinish(bp, NULL, error);
1730 #endif /* illumos */
1735 * Set the buffer count to the zvol maximum transfer.
1736 * Using our own routine instead of the default minphys()
1737 * means that for larger writes we write bigger buffers on X86
1738 * (128K instead of 56K) and flush the disk write cache less often
1739 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1740 * 56K on X86 and 128K on sparc).
1743 zvol_minphys(struct buf *bp)
1745 if (bp->b_bcount > zvol_maxphys)
1746 bp->b_bcount = zvol_maxphys;
1750 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1752 minor_t minor = getminor(dev);
1759 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1761 return (SET_ERROR(ENXIO));
1763 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1764 return (SET_ERROR(EINVAL));
1766 boff = ldbtob(blkno);
1767 resid = ldbtob(nblocks);
1769 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1772 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1773 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1786 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1788 minor_t minor = getminor(dev);
1789 #else /* !illumos */
1791 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1793 #endif /* illumos */
1800 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1802 return (SET_ERROR(ENXIO));
1807 volsize = zv->zv_volsize;
1808 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1809 if (uio->uio_resid > 0 &&
1810 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1811 return (SET_ERROR(EIO));
1814 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1815 error = physio(zvol_strategy, NULL, dev, B_READ,
1821 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1823 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1824 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1826 /* don't read past the end */
1827 if (bytes > volsize - uio->uio_loffset)
1828 bytes = volsize - uio->uio_loffset;
1830 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
1832 /* convert checksum errors into IO errors */
1833 if (error == ECKSUM)
1834 error = SET_ERROR(EIO);
1838 zfs_range_unlock(rl);
1845 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1847 minor_t minor = getminor(dev);
1848 #else /* !illumos */
1850 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1852 #endif /* illumos */
1860 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1862 return (SET_ERROR(ENXIO));
1867 volsize = zv->zv_volsize;
1868 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1869 if (uio->uio_resid > 0 &&
1870 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1871 return (SET_ERROR(EIO));
1874 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1875 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1880 sync = !(zv->zv_flags & ZVOL_WCE) ||
1882 sync = (ioflag & IO_SYNC) ||
1884 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1886 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1888 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1889 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1890 uint64_t off = uio->uio_loffset;
1891 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1893 if (bytes > volsize - off) /* don't write past the end */
1894 bytes = volsize - off;
1896 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1897 error = dmu_tx_assign(tx, TXG_WAIT);
1902 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1904 zvol_log_write(zv, tx, off, bytes, sync);
1910 zfs_range_unlock(rl);
1912 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1918 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1920 struct uuid uuid = EFI_RESERVED;
1921 efi_gpe_t gpe = { 0 };
1927 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1928 return (SET_ERROR(EFAULT));
1929 ptr = (char *)(uintptr_t)efi.dki_data_64;
1930 length = efi.dki_length;
1932 * Some clients may attempt to request a PMBR for the
1933 * zvol. Currently this interface will return EINVAL to
1934 * such requests. These requests could be supported by
1935 * adding a check for lba == 0 and consing up an appropriate
1938 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1939 return (SET_ERROR(EINVAL));
1941 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1942 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1943 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1945 if (efi.dki_lba == 1) {
1946 efi_gpt_t gpt = { 0 };
1948 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1949 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1950 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1951 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1952 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1953 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1954 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1955 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1956 gpt.efi_gpt_SizeOfPartitionEntry =
1957 LE_32(sizeof (efi_gpe_t));
1958 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1959 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1960 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1961 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1962 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1964 return (SET_ERROR(EFAULT));
1965 ptr += sizeof (gpt);
1966 length -= sizeof (gpt);
1968 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1970 return (SET_ERROR(EFAULT));
1975 * BEGIN entry points to allow external callers access to the volume.
1978 * Return the volume parameters needed for access from an external caller.
1979 * These values are invariant as long as the volume is held open.
1982 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1983 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1984 void **rl_hdl, void **bonus_hdl)
1988 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1990 return (SET_ERROR(ENXIO));
1991 if (zv->zv_flags & ZVOL_DUMPIFIED)
1992 return (SET_ERROR(ENXIO));
1994 ASSERT(blksize && max_xfer_len && minor_hdl &&
1995 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1997 *blksize = zv->zv_volblocksize;
1998 *max_xfer_len = (uint64_t)zvol_maxphys;
2000 *objset_hdl = zv->zv_objset;
2001 *zil_hdl = zv->zv_zilog;
2002 *rl_hdl = &zv->zv_znode;
2003 *bonus_hdl = zv->zv_dbuf;
2008 * Return the current volume size to an external caller.
2009 * The size can change while the volume is open.
2012 zvol_get_volume_size(void *minor_hdl)
2014 zvol_state_t *zv = minor_hdl;
2016 return (zv->zv_volsize);
2020 * Return the current WCE setting to an external caller.
2021 * The WCE setting can change while the volume is open.
2024 zvol_get_volume_wce(void *minor_hdl)
2026 zvol_state_t *zv = minor_hdl;
2028 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2032 * Entry point for external callers to zvol_log_write
2035 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2038 zvol_state_t *zv = minor_hdl;
2040 zvol_log_write(zv, tx, off, resid, sync);
2043 * END entry points to allow external callers access to the volume.
2045 #endif /* illumos */
2048 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2051 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2056 zilog_t *zilog = zv->zv_zilog;
2058 if (zil_replaying(zilog, tx))
2061 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2062 lr = (lr_truncate_t *)&itx->itx_lr;
2063 lr->lr_foid = ZVOL_OBJ;
2064 lr->lr_offset = off;
2065 lr->lr_length = len;
2067 itx->itx_sync = sync;
2068 zil_itx_assign(zilog, itx, tx);
2073 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2074 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2078 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2081 struct dk_callback *dkc;
2085 mutex_enter(&zfsdev_state_lock);
2087 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2090 mutex_exit(&zfsdev_state_lock);
2091 return (SET_ERROR(ENXIO));
2093 ASSERT(zv->zv_total_opens > 0);
2099 struct dk_cinfo dki;
2101 bzero(&dki, sizeof (dki));
2102 (void) strcpy(dki.dki_cname, "zvol");
2103 (void) strcpy(dki.dki_dname, "zvol");
2104 dki.dki_ctype = DKC_UNKNOWN;
2105 dki.dki_unit = getminor(dev);
2106 dki.dki_maxtransfer =
2107 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2108 mutex_exit(&zfsdev_state_lock);
2109 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2110 error = SET_ERROR(EFAULT);
2114 case DKIOCGMEDIAINFO:
2116 struct dk_minfo dkm;
2118 bzero(&dkm, sizeof (dkm));
2119 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2120 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2121 dkm.dki_media_type = DK_UNKNOWN;
2122 mutex_exit(&zfsdev_state_lock);
2123 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2124 error = SET_ERROR(EFAULT);
2128 case DKIOCGMEDIAINFOEXT:
2130 struct dk_minfo_ext dkmext;
2132 bzero(&dkmext, sizeof (dkmext));
2133 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2134 dkmext.dki_pbsize = zv->zv_volblocksize;
2135 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2136 dkmext.dki_media_type = DK_UNKNOWN;
2137 mutex_exit(&zfsdev_state_lock);
2138 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2139 error = SET_ERROR(EFAULT);
2145 uint64_t vs = zv->zv_volsize;
2146 uint8_t bs = zv->zv_min_bs;
2148 mutex_exit(&zfsdev_state_lock);
2149 error = zvol_getefi((void *)arg, flag, vs, bs);
2153 case DKIOCFLUSHWRITECACHE:
2154 dkc = (struct dk_callback *)arg;
2155 mutex_exit(&zfsdev_state_lock);
2156 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2157 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2158 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2165 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2166 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2168 error = SET_ERROR(EFAULT);
2174 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2176 error = SET_ERROR(EFAULT);
2180 zv->zv_flags |= ZVOL_WCE;
2181 mutex_exit(&zfsdev_state_lock);
2183 zv->zv_flags &= ~ZVOL_WCE;
2184 mutex_exit(&zfsdev_state_lock);
2185 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2193 * commands using these (like prtvtoc) expect ENOTSUP
2194 * since we're emulating an EFI label
2196 error = SET_ERROR(ENOTSUP);
2200 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2202 error = zvol_dumpify(zv);
2203 zfs_range_unlock(rl);
2207 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2209 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2211 error = zvol_dump_fini(zv);
2212 zfs_range_unlock(rl);
2220 if (!zvol_unmap_enabled)
2223 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2224 error = SET_ERROR(EFAULT);
2229 * Apply Postel's Law to length-checking. If they overshoot,
2230 * just blank out until the end, if there's a need to blank
2233 if (df.df_start >= zv->zv_volsize)
2234 break; /* No need to do anything... */
2236 mutex_exit(&zfsdev_state_lock);
2238 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2240 tx = dmu_tx_create(zv->zv_objset);
2241 dmu_tx_mark_netfree(tx);
2242 error = dmu_tx_assign(tx, TXG_WAIT);
2246 zvol_log_truncate(zv, tx, df.df_start,
2247 df.df_length, B_TRUE);
2249 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2250 df.df_start, df.df_length);
2253 zfs_range_unlock(rl);
2257 * If the write-cache is disabled or 'sync' property
2258 * is set to 'always' then treat this as a synchronous
2259 * operation (i.e. commit to zil).
2261 if (!(zv->zv_flags & ZVOL_WCE) ||
2262 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2263 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2266 * If the caller really wants synchronous writes, and
2267 * can't wait for them, don't return until the write
2270 if (df.df_flags & DF_WAIT_SYNC) {
2272 dmu_objset_pool(zv->zv_objset), 0);
2279 error = SET_ERROR(ENOTTY);
2283 mutex_exit(&zfsdev_state_lock);
2286 #endif /* illumos */
2291 return (zvol_minors != 0);
2297 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2300 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2302 ZFS_LOG(1, "ZVOL Initialized.");
2310 mutex_destroy(&zfsdev_state_lock);
2312 ddi_soft_state_fini(&zfsdev_state);
2313 ZFS_LOG(1, "ZVOL Deinitialized.");
2319 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2321 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2323 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2330 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2332 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2334 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2338 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2342 objset_t *os = zv->zv_objset;
2343 spa_t *spa = dmu_objset_spa(os);
2344 vdev_t *vd = spa->spa_root_vdev;
2345 nvlist_t *nv = NULL;
2346 uint64_t version = spa_version(spa);
2347 uint64_t checksum, compress, refresrv, vbs, dedup;
2349 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2350 ASSERT(vd->vdev_ops == &vdev_root_ops);
2352 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2356 /* wait for dmu_free_long_range to actually free the blocks */
2357 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2360 * If the pool on which the dump device is being initialized has more
2361 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2362 * enabled. If so, bump that feature's counter to indicate that the
2363 * feature is active. We also check the vdev type to handle the
2365 * # zpool create test raidz disk1 disk2 disk3
2366 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2367 * the raidz vdev itself has 3 children.
2369 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2370 if (!spa_feature_is_enabled(spa,
2371 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2372 return (SET_ERROR(ENOTSUP));
2373 (void) dsl_sync_task(spa_name(spa),
2374 zfs_mvdev_dump_feature_check,
2375 zfs_mvdev_dump_activate_feature_sync, NULL,
2376 2, ZFS_SPACE_CHECK_RESERVED);
2380 error = dsl_prop_get_integer(zv->zv_name,
2381 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2383 error = dsl_prop_get_integer(zv->zv_name,
2384 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2388 error = dsl_prop_get_integer(zv->zv_name,
2389 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2393 error = dsl_prop_get_integer(zv->zv_name,
2394 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2397 if (version >= SPA_VERSION_DEDUP && error == 0) {
2398 error = dsl_prop_get_integer(zv->zv_name,
2399 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2405 tx = dmu_tx_create(os);
2406 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2407 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2408 error = dmu_tx_assign(tx, TXG_WAIT);
2415 * If we are resizing the dump device then we only need to
2416 * update the refreservation to match the newly updated
2417 * zvolsize. Otherwise, we save off the original state of the
2418 * zvol so that we can restore them if the zvol is ever undumpified.
2421 error = zap_update(os, ZVOL_ZAP_OBJ,
2422 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2423 &zv->zv_volsize, tx);
2425 error = zap_update(os, ZVOL_ZAP_OBJ,
2426 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2429 error = zap_update(os, ZVOL_ZAP_OBJ,
2430 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2434 error = zap_update(os, ZVOL_ZAP_OBJ,
2435 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2439 error = zap_update(os, ZVOL_ZAP_OBJ,
2440 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2444 error = dmu_object_set_blocksize(
2445 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2447 if (version >= SPA_VERSION_DEDUP && error == 0) {
2448 error = zap_update(os, ZVOL_ZAP_OBJ,
2449 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2453 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2458 * We only need update the zvol's property if we are initializing
2459 * the dump area for the first time.
2461 if (error == 0 && !resize) {
2463 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2464 * function. Otherwise, use the old default -- OFF.
2466 checksum = spa_feature_is_active(spa,
2467 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2470 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2471 VERIFY(nvlist_add_uint64(nv,
2472 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2473 VERIFY(nvlist_add_uint64(nv,
2474 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2475 ZIO_COMPRESS_OFF) == 0);
2476 VERIFY(nvlist_add_uint64(nv,
2477 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2479 if (version >= SPA_VERSION_DEDUP) {
2480 VERIFY(nvlist_add_uint64(nv,
2481 zfs_prop_to_name(ZFS_PROP_DEDUP),
2482 ZIO_CHECKSUM_OFF) == 0);
2485 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2490 /* Allocate the space for the dump */
2492 error = zvol_prealloc(zv);
2497 zvol_dumpify(zvol_state_t *zv)
2500 uint64_t dumpsize = 0;
2502 objset_t *os = zv->zv_objset;
2504 if (zv->zv_flags & ZVOL_RDONLY)
2505 return (SET_ERROR(EROFS));
2507 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2508 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2509 boolean_t resize = (dumpsize > 0);
2511 if ((error = zvol_dump_init(zv, resize)) != 0) {
2512 (void) zvol_dump_fini(zv);
2518 * Build up our lba mapping.
2520 error = zvol_get_lbas(zv);
2522 (void) zvol_dump_fini(zv);
2526 tx = dmu_tx_create(os);
2527 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2528 error = dmu_tx_assign(tx, TXG_WAIT);
2531 (void) zvol_dump_fini(zv);
2535 zv->zv_flags |= ZVOL_DUMPIFIED;
2536 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2537 &zv->zv_volsize, tx);
2541 (void) zvol_dump_fini(zv);
2545 txg_wait_synced(dmu_objset_pool(os), 0);
2550 zvol_dump_fini(zvol_state_t *zv)
2553 objset_t *os = zv->zv_objset;
2556 uint64_t checksum, compress, refresrv, vbs, dedup;
2557 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2560 * Attempt to restore the zvol back to its pre-dumpified state.
2561 * This is a best-effort attempt as it's possible that not all
2562 * of these properties were initialized during the dumpify process
2563 * (i.e. error during zvol_dump_init).
2566 tx = dmu_tx_create(os);
2567 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2568 error = dmu_tx_assign(tx, TXG_WAIT);
2573 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2576 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2577 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2578 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2579 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2580 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2581 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2582 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2583 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2585 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2586 (void) nvlist_add_uint64(nv,
2587 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2588 (void) nvlist_add_uint64(nv,
2589 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2590 (void) nvlist_add_uint64(nv,
2591 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2592 if (version >= SPA_VERSION_DEDUP &&
2593 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2594 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2595 (void) nvlist_add_uint64(nv,
2596 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2598 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2602 zvol_free_extents(zv);
2603 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2604 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2605 /* wait for dmu_free_long_range to actually free the blocks */
2606 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2607 tx = dmu_tx_create(os);
2608 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2609 error = dmu_tx_assign(tx, TXG_WAIT);
2614 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2615 zv->zv_volblocksize = vbs;
2620 #else /* !illumos */
2623 zvol_geom_run(zvol_state_t *zv)
2625 struct g_provider *pp;
2627 pp = zv->zv_provider;
2628 g_error_provider(pp, 0);
2630 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2631 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2635 zvol_geom_destroy(zvol_state_t *zv)
2637 struct g_provider *pp;
2639 g_topology_assert();
2641 mtx_lock(&zv->zv_queue_mtx);
2643 wakeup_one(&zv->zv_queue);
2644 while (zv->zv_state != 2)
2645 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2646 mtx_destroy(&zv->zv_queue_mtx);
2648 pp = zv->zv_provider;
2649 zv->zv_provider = NULL;
2651 g_wither_geom(pp->geom, ENXIO);
2655 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2657 int count, error, flags;
2659 g_topology_assert();
2662 * To make it easier we expect either open or close, but not both
2665 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2666 (acr <= 0 && acw <= 0 && ace <= 0),
2667 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2668 pp->name, acr, acw, ace));
2670 if (pp->private == NULL) {
2671 if (acr <= 0 && acw <= 0 && ace <= 0)
2677 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2678 * because GEOM already handles that and handles it a bit differently.
2679 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2680 * only one exclusive consumer, no matter if it is reader or writer.
2681 * I like better the way GEOM works so I'll leave it for GEOM to
2682 * decide what to do.
2685 count = acr + acw + ace;
2690 if (acr != 0 || ace != 0)
2695 g_topology_unlock();
2697 error = zvol_open(pp, flags, count);
2699 error = zvol_close(pp, flags, -count);
2705 zvol_geom_start(struct bio *bp)
2710 zv = bp->bio_to->private;
2712 switch (bp->bio_cmd) {
2714 if (!THREAD_CAN_SLEEP())
2716 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2717 g_io_deliver(bp, 0);
2722 if (!THREAD_CAN_SLEEP())
2727 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2728 uint64_t refd, avail, usedobjs, availobjs, val;
2730 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2732 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2733 dmu_objset_space(zv->zv_objset, &refd, &avail,
2734 &usedobjs, &availobjs);
2735 if (g_handleattr_off_t(bp, "blocksavail",
2738 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2739 dmu_objset_space(zv->zv_objset, &refd, &avail,
2740 &usedobjs, &availobjs);
2741 if (g_handleattr_off_t(bp, "blocksused",
2744 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2745 avail = metaslab_class_get_space(spa_normal_class(spa));
2746 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2747 if (g_handleattr_off_t(bp, "poolblocksavail",
2750 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2751 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2752 if (g_handleattr_off_t(bp, "poolblocksused",
2759 g_io_deliver(bp, EOPNOTSUPP);
2765 mtx_lock(&zv->zv_queue_mtx);
2766 first = (bioq_first(&zv->zv_queue) == NULL);
2767 bioq_insert_tail(&zv->zv_queue, bp);
2768 mtx_unlock(&zv->zv_queue_mtx);
2770 wakeup_one(&zv->zv_queue);
2774 zvol_geom_worker(void *arg)
2779 thread_lock(curthread);
2780 sched_prio(curthread, PRIBIO);
2781 thread_unlock(curthread);
2785 mtx_lock(&zv->zv_queue_mtx);
2786 bp = bioq_takefirst(&zv->zv_queue);
2788 if (zv->zv_state == 1) {
2790 wakeup(&zv->zv_state);
2791 mtx_unlock(&zv->zv_queue_mtx);
2794 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2798 mtx_unlock(&zv->zv_queue_mtx);
2799 switch (bp->bio_cmd) {
2801 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2802 g_io_deliver(bp, 0);
2810 g_io_deliver(bp, EOPNOTSUPP);
2816 extern boolean_t dataset_name_hidden(const char *name);
2819 zvol_create_snapshots(objset_t *os, const char *name)
2821 uint64_t cookie, obj;
2826 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2829 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2834 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2835 if (len >= MAXPATHLEN) {
2836 dmu_objset_rele(os, FTAG);
2837 error = ENAMETOOLONG;
2841 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2842 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2843 sname + len, &obj, &cookie, NULL);
2844 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2846 if (error == ENOENT)
2851 error = zvol_create_minor(sname);
2852 if (error != 0 && error != EEXIST) {
2853 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2859 kmem_free(sname, MAXPATHLEN);
2864 zvol_create_minors(const char *name)
2871 if (dataset_name_hidden(name))
2874 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2875 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2879 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2880 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2881 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2882 error = zvol_create_minor(name);
2883 if (error == 0 || error == EEXIST) {
2884 error = zvol_create_snapshots(os, name);
2886 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2889 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2890 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2893 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2894 dmu_objset_rele(os, FTAG);
2898 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2899 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2900 dmu_objset_rele(os, FTAG);
2901 kmem_free(osname, MAXPATHLEN);
2904 p = osname + strlen(osname);
2905 len = MAXPATHLEN - (p - osname);
2908 /* Prefetch the datasets. */
2910 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2911 if (!dataset_name_hidden(osname))
2912 (void) dmu_objset_prefetch(osname, NULL);
2917 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2919 dmu_objset_rele(os, FTAG);
2920 (void)zvol_create_minors(osname);
2921 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2922 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2928 dmu_objset_rele(os, FTAG);
2929 kmem_free(osname, MAXPATHLEN);
2934 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2937 struct g_provider *pp;
2940 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2942 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2944 pp = zv->zv_provider;
2949 zv->zv_provider = NULL;
2950 g_wither_provider(pp, ENXIO);
2952 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2953 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2954 pp->sectorsize = DEV_BSIZE;
2955 pp->mediasize = zv->zv_volsize;
2957 zv->zv_provider = pp;
2958 g_error_provider(pp, 0);
2959 g_topology_unlock();
2960 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2961 struct make_dev_args args;
2964 ASSERT(dev != NULL);
2967 if (zv->zv_total_opens > 0) {
2968 zv->zv_flags &= ~ZVOL_EXCL;
2969 zv->zv_total_opens = 0;
2970 zvol_last_close(zv);
2973 make_dev_args_init(&args);
2974 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
2975 args.mda_devsw = &zvol_cdevsw;
2977 args.mda_uid = UID_ROOT;
2978 args.mda_gid = GID_OPERATOR;
2979 args.mda_mode = 0640;
2980 args.mda_si_drv2 = zv;
2981 if (make_dev_s(&args, &zv->zv_dev,
2982 "%s/%s", ZVOL_DRIVER, newname) == 0)
2983 zv->zv_dev->si_iosize_max = MAXPHYS;
2985 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
2989 zvol_rename_minors(const char *oldname, const char *newname)
2991 char name[MAXPATHLEN];
2992 struct g_provider *pp;
2994 size_t oldnamelen, newnamelen;
2997 boolean_t locked = B_FALSE;
2999 oldnamelen = strlen(oldname);
3000 newnamelen = strlen(newname);
3003 /* See comment in zvol_open(). */
3004 if (!MUTEX_HELD(&zfsdev_state_lock)) {
3005 mutex_enter(&zfsdev_state_lock);
3009 LIST_FOREACH(zv, &all_zvols, zv_links) {
3010 if (strcmp(zv->zv_name, oldname) == 0) {
3011 zvol_rename_minor(zv, newname);
3012 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3013 (zv->zv_name[oldnamelen] == '/' ||
3014 zv->zv_name[oldnamelen] == '@')) {
3015 snprintf(name, sizeof(name), "%s%c%s", newname,
3016 zv->zv_name[oldnamelen],
3017 zv->zv_name + oldnamelen + 1);
3018 zvol_rename_minor(zv, name);
3023 mutex_exit(&zfsdev_state_lock);
3028 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3030 zvol_state_t *zv = dev->si_drv2;
3033 mutex_enter(&zfsdev_state_lock);
3034 if (zv->zv_total_opens == 0)
3035 err = zvol_first_open(zv);
3037 mutex_exit(&zfsdev_state_lock);
3040 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3041 err = SET_ERROR(EROFS);
3044 if (zv->zv_flags & ZVOL_EXCL) {
3045 err = SET_ERROR(EBUSY);
3049 if (flags & FEXCL) {
3050 if (zv->zv_total_opens != 0) {
3051 err = SET_ERROR(EBUSY);
3054 zv->zv_flags |= ZVOL_EXCL;
3058 zv->zv_total_opens++;
3059 mutex_exit(&zfsdev_state_lock);
3062 if (zv->zv_total_opens == 0)
3063 zvol_last_close(zv);
3064 mutex_exit(&zfsdev_state_lock);
3069 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3071 zvol_state_t *zv = dev->si_drv2;
3073 mutex_enter(&zfsdev_state_lock);
3074 if (zv->zv_flags & ZVOL_EXCL) {
3075 ASSERT(zv->zv_total_opens == 1);
3076 zv->zv_flags &= ~ZVOL_EXCL;
3080 * If the open count is zero, this is a spurious close.
3081 * That indicates a bug in the kernel / DDI framework.
3083 ASSERT(zv->zv_total_opens != 0);
3086 * You may get multiple opens, but only one close.
3088 zv->zv_total_opens--;
3090 if (zv->zv_total_opens == 0)
3091 zvol_last_close(zv);
3093 mutex_exit(&zfsdev_state_lock);
3098 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3102 off_t offset, length, chunk;
3109 KASSERT(zv->zv_total_opens > 0,
3110 ("Device with zero access count in zvol_d_ioctl"));
3112 i = IOCPARM_LEN(cmd);
3114 case DIOCGSECTORSIZE:
3115 *(u_int *)data = DEV_BSIZE;
3117 case DIOCGMEDIASIZE:
3118 *(off_t *)data = zv->zv_volsize;
3121 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3124 if (!zvol_unmap_enabled)
3127 offset = ((off_t *)data)[0];
3128 length = ((off_t *)data)[1];
3129 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3130 offset < 0 || offset >= zv->zv_volsize ||
3132 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3138 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3139 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3140 error = dmu_tx_assign(tx, TXG_WAIT);
3144 zvol_log_truncate(zv, tx, offset, length, B_TRUE);
3146 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3149 zfs_range_unlock(rl);
3150 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
3151 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3153 case DIOCGSTRIPESIZE:
3154 *(off_t *)data = zv->zv_volblocksize;
3156 case DIOCGSTRIPEOFFSET:
3160 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3161 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3162 uint64_t refd, avail, usedobjs, availobjs;
3164 if (strcmp(arg->name, "GEOM::candelete") == 0)
3166 else if (strcmp(arg->name, "blocksavail") == 0) {
3167 dmu_objset_space(zv->zv_objset, &refd, &avail,
3168 &usedobjs, &availobjs);
3169 arg->value.off = avail / DEV_BSIZE;
3170 } else if (strcmp(arg->name, "blocksused") == 0) {
3171 dmu_objset_space(zv->zv_objset, &refd, &avail,
3172 &usedobjs, &availobjs);
3173 arg->value.off = refd / DEV_BSIZE;
3174 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3175 avail = metaslab_class_get_space(spa_normal_class(spa));
3176 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3177 arg->value.off = avail / DEV_BSIZE;
3178 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3179 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3180 arg->value.off = refd / DEV_BSIZE;
3187 off_t *off = (off_t *)data;
3191 hole = (cmd == FIOSEEKHOLE);
3193 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3203 #endif /* illumos */