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];
591 struct g_provider *pp;
593 uint64_t volsize, mode;
598 ZFS_LOG(1, "Creating ZVOL %s...", name);
601 mutex_enter(&zfsdev_state_lock);
603 if (zvol_minor_lookup(name) != NULL) {
604 mutex_exit(&zfsdev_state_lock);
605 return (SET_ERROR(EEXIST));
608 /* lie and say we're read-only */
609 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
612 mutex_exit(&zfsdev_state_lock);
617 if ((minor = zfsdev_minor_alloc()) == 0) {
618 dmu_objset_disown(os, FTAG);
619 mutex_exit(&zfsdev_state_lock);
620 return (SET_ERROR(ENXIO));
623 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
624 dmu_objset_disown(os, FTAG);
625 mutex_exit(&zfsdev_state_lock);
626 return (SET_ERROR(EAGAIN));
628 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
631 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
633 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
634 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
635 ddi_soft_state_free(zfsdev_state, minor);
636 dmu_objset_disown(os, FTAG);
637 mutex_exit(&zfsdev_state_lock);
638 return (SET_ERROR(EAGAIN));
641 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
643 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
644 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
645 ddi_remove_minor_node(zfs_dip, chrbuf);
646 ddi_soft_state_free(zfsdev_state, minor);
647 dmu_objset_disown(os, FTAG);
648 mutex_exit(&zfsdev_state_lock);
649 return (SET_ERROR(EAGAIN));
652 zs = ddi_get_soft_state(zfsdev_state, minor);
653 zs->zss_type = ZSST_ZVOL;
654 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
657 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
659 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
661 kmem_free(zv, sizeof(*zv));
662 dmu_objset_disown(os, zvol_tag);
663 mutex_exit(&zfsdev_state_lock);
666 error = dsl_prop_get_integer(name,
667 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
668 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
672 zv->zv_volsize = volsize;
673 zv->zv_volmode = mode;
674 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
676 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
677 gp->start = zvol_geom_start;
678 gp->access = zvol_geom_access;
679 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
680 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
681 pp->sectorsize = DEV_BSIZE;
682 pp->mediasize = zv->zv_volsize;
685 zv->zv_provider = pp;
686 bioq_init(&zv->zv_queue);
687 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
688 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
689 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
690 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
691 0640, "%s/%s", ZVOL_DRIVER, name) != 0) {
692 kmem_free(zv, sizeof(*zv));
693 dmu_objset_disown(os, FTAG);
694 mutex_exit(&zfsdev_state_lock);
695 return (SET_ERROR(ENXIO));
698 dev->si_iosize_max = MAXPHYS;
701 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
704 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
705 zv->zv_min_bs = DEV_BSHIFT;
707 zv->zv_minor = minor;
710 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
711 zv->zv_flags |= ZVOL_RDONLY;
712 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
713 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
714 sizeof (rl_t), offsetof(rl_t, r_node));
715 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
716 offsetof(zvol_extent_t, ze_node));
717 /* get and cache the blocksize */
718 error = dmu_object_info(os, ZVOL_OBJ, &doi);
720 zv->zv_volblocksize = doi.doi_data_block_size;
722 if (spa_writeable(dmu_objset_spa(os))) {
723 if (zil_replay_disable)
724 zil_destroy(dmu_objset_zil(os), B_FALSE);
726 zil_replay(os, zv, zvol_replay_vector);
728 dmu_objset_disown(os, FTAG);
729 zv->zv_objset = NULL;
733 mutex_exit(&zfsdev_state_lock);
735 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
741 ZFS_LOG(1, "ZVOL %s created.", name);
748 * Remove minor node for the specified volume.
751 zvol_remove_zv(zvol_state_t *zv)
755 minor_t minor = zv->zv_minor;
758 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
759 if (zv->zv_total_opens != 0)
760 return (SET_ERROR(EBUSY));
763 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
764 ddi_remove_minor_node(zfs_dip, nmbuf);
766 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
767 ddi_remove_minor_node(zfs_dip, nmbuf);
769 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
771 LIST_REMOVE(zv, zv_links);
772 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
774 zvol_geom_destroy(zv);
776 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV)
777 destroy_dev(zv->zv_dev);
780 avl_destroy(&zv->zv_znode.z_range_avl);
781 mutex_destroy(&zv->zv_znode.z_range_lock);
783 kmem_free(zv, sizeof (zvol_state_t));
785 ddi_soft_state_free(zfsdev_state, minor);
792 zvol_remove_minor(const char *name)
797 mutex_enter(&zfsdev_state_lock);
798 if ((zv = zvol_minor_lookup(name)) == NULL) {
799 mutex_exit(&zfsdev_state_lock);
800 return (SET_ERROR(ENXIO));
802 rc = zvol_remove_zv(zv);
803 mutex_exit(&zfsdev_state_lock);
808 zvol_first_open(zvol_state_t *zv)
815 /* lie and say we're read-only */
816 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
822 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
825 dmu_objset_disown(os, zvol_tag);
829 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
831 dmu_objset_disown(os, zvol_tag);
835 zvol_size_changed(zv, volsize);
836 zv->zv_zilog = zil_open(os, zvol_get_data);
838 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
840 if (readonly || dmu_objset_is_snapshot(os) ||
841 !spa_writeable(dmu_objset_spa(os)))
842 zv->zv_flags |= ZVOL_RDONLY;
844 zv->zv_flags &= ~ZVOL_RDONLY;
849 zvol_last_close(zvol_state_t *zv)
851 zil_close(zv->zv_zilog);
854 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
860 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
861 !(zv->zv_flags & ZVOL_RDONLY))
862 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
863 dmu_objset_evict_dbufs(zv->zv_objset);
865 dmu_objset_disown(zv->zv_objset, zvol_tag);
866 zv->zv_objset = NULL;
871 zvol_prealloc(zvol_state_t *zv)
873 objset_t *os = zv->zv_objset;
875 uint64_t refd, avail, usedobjs, availobjs;
876 uint64_t resid = zv->zv_volsize;
879 /* Check the space usage before attempting to allocate the space */
880 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
881 if (avail < zv->zv_volsize)
882 return (SET_ERROR(ENOSPC));
884 /* Free old extents if they exist */
885 zvol_free_extents(zv);
889 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
891 tx = dmu_tx_create(os);
892 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
893 error = dmu_tx_assign(tx, TXG_WAIT);
896 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
899 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
904 txg_wait_synced(dmu_objset_pool(os), 0);
911 zvol_update_volsize(objset_t *os, uint64_t volsize)
916 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
918 tx = dmu_tx_create(os);
919 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
920 dmu_tx_mark_netfree(tx);
921 error = dmu_tx_assign(tx, TXG_WAIT);
927 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
932 error = dmu_free_long_range(os,
933 ZVOL_OBJ, volsize, DMU_OBJECT_END);
938 zvol_remove_minors(const char *name)
945 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
946 (void) strncpy(namebuf, name, strlen(name));
947 (void) strcat(namebuf, "/");
948 mutex_enter(&zfsdev_state_lock);
949 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
951 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
954 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
955 (void) zvol_remove_zv(zv);
957 kmem_free(namebuf, strlen(name) + 2);
959 mutex_exit(&zfsdev_state_lock);
961 zvol_state_t *zv, *tzv;
964 namelen = strlen(name);
967 mutex_enter(&zfsdev_state_lock);
969 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
970 if (strcmp(zv->zv_name, name) == 0 ||
971 (strncmp(zv->zv_name, name, namelen) == 0 &&
972 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
973 zv->zv_name[namelen] == '@'))) {
974 (void) zvol_remove_zv(zv);
978 mutex_exit(&zfsdev_state_lock);
984 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
986 uint64_t old_volsize = 0ULL;
989 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
992 * Reinitialize the dump area to the new size. If we
993 * failed to resize the dump area then restore it back to
994 * its original size. We must set the new volsize prior
995 * to calling dumpvp_resize() to ensure that the devices'
996 * size(9P) is not visible by the dump subsystem.
998 old_volsize = zv->zv_volsize;
999 zvol_size_changed(zv, volsize);
1002 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1003 if ((error = zvol_dumpify(zv)) != 0 ||
1004 (error = dumpvp_resize()) != 0) {
1007 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1008 zvol_size_changed(zv, old_volsize);
1009 dumpify_error = zvol_dumpify(zv);
1010 error = dumpify_error ? dumpify_error : error;
1013 #endif /* ZVOL_DUMP */
1017 * Generate a LUN expansion event.
1022 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1024 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1027 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1028 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1030 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1031 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1034 kmem_free(physpath, MAXPATHLEN);
1036 #endif /* illumos */
1041 zvol_set_volsize(const char *name, uint64_t volsize)
1043 zvol_state_t *zv = NULL;
1046 dmu_object_info_t doi;
1048 boolean_t owned = B_FALSE;
1050 error = dsl_prop_get_integer(name,
1051 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1055 return (SET_ERROR(EROFS));
1057 mutex_enter(&zfsdev_state_lock);
1058 zv = zvol_minor_lookup(name);
1060 if (zv == NULL || zv->zv_objset == NULL) {
1061 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1063 mutex_exit(&zfsdev_state_lock);
1073 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1074 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1077 error = zvol_update_volsize(os, volsize);
1079 if (error == 0 && zv != NULL)
1080 error = zvol_update_live_volsize(zv, volsize);
1083 dmu_objset_disown(os, FTAG);
1085 zv->zv_objset = NULL;
1087 mutex_exit(&zfsdev_state_lock);
1094 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1097 zvol_open(struct g_provider *pp, int flag, int count)
1104 mutex_enter(&zfsdev_state_lock);
1106 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1108 mutex_exit(&zfsdev_state_lock);
1109 return (SET_ERROR(ENXIO));
1112 if (zv->zv_total_opens == 0)
1113 err = zvol_first_open(zv);
1115 mutex_exit(&zfsdev_state_lock);
1118 #else /* !illumos */
1119 boolean_t locked = B_FALSE;
1122 * Protect against recursively entering spa_namespace_lock
1123 * when spa_open() is used for a pool on a (local) ZVOL(s).
1124 * This is needed since we replaced upstream zfsdev_state_lock
1125 * with spa_namespace_lock in the ZVOL code.
1126 * We are using the same trick as spa_open().
1127 * Note that calls in zvol_first_open which need to resolve
1128 * pool name to a spa object will enter spa_open()
1129 * recursively, but that function already has all the
1130 * necessary protection.
1132 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1133 mutex_enter(&zfsdev_state_lock);
1140 mutex_exit(&zfsdev_state_lock);
1141 return (SET_ERROR(ENXIO));
1144 if (zv->zv_total_opens == 0) {
1145 err = zvol_first_open(zv);
1148 mutex_exit(&zfsdev_state_lock);
1151 pp->mediasize = zv->zv_volsize;
1152 pp->stripeoffset = 0;
1153 pp->stripesize = zv->zv_volblocksize;
1155 #endif /* illumos */
1156 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1157 err = SET_ERROR(EROFS);
1160 if (zv->zv_flags & ZVOL_EXCL) {
1161 err = SET_ERROR(EBUSY);
1166 if (zv->zv_total_opens != 0) {
1167 err = SET_ERROR(EBUSY);
1170 zv->zv_flags |= ZVOL_EXCL;
1175 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1176 zv->zv_open_count[otyp]++;
1177 zv->zv_total_opens++;
1179 mutex_exit(&zfsdev_state_lock);
1181 zv->zv_total_opens += count;
1183 mutex_exit(&zfsdev_state_lock);
1188 if (zv->zv_total_opens == 0)
1189 zvol_last_close(zv);
1191 mutex_exit(&zfsdev_state_lock);
1194 mutex_exit(&zfsdev_state_lock);
1202 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1204 minor_t minor = getminor(dev);
1208 mutex_enter(&zfsdev_state_lock);
1210 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1212 mutex_exit(&zfsdev_state_lock);
1213 #else /* !illumos */
1215 zvol_close(struct g_provider *pp, int flag, int count)
1219 boolean_t locked = B_FALSE;
1221 /* See comment in zvol_open(). */
1222 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1223 mutex_enter(&zfsdev_state_lock);
1230 mutex_exit(&zfsdev_state_lock);
1231 #endif /* illumos */
1232 return (SET_ERROR(ENXIO));
1235 if (zv->zv_flags & ZVOL_EXCL) {
1236 ASSERT(zv->zv_total_opens == 1);
1237 zv->zv_flags &= ~ZVOL_EXCL;
1241 * If the open count is zero, this is a spurious close.
1242 * That indicates a bug in the kernel / DDI framework.
1245 ASSERT(zv->zv_open_count[otyp] != 0);
1247 ASSERT(zv->zv_total_opens != 0);
1250 * You may get multiple opens, but only one close.
1253 zv->zv_open_count[otyp]--;
1254 zv->zv_total_opens--;
1256 zv->zv_total_opens -= count;
1259 if (zv->zv_total_opens == 0)
1260 zvol_last_close(zv);
1263 mutex_exit(&zfsdev_state_lock);
1266 mutex_exit(&zfsdev_state_lock);
1272 zvol_get_done(zgd_t *zgd, int error)
1275 dmu_buf_rele(zgd->zgd_db, zgd);
1277 zfs_range_unlock(zgd->zgd_rl);
1279 if (error == 0 && zgd->zgd_bp)
1280 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1282 kmem_free(zgd, sizeof (zgd_t));
1286 * Get data to generate a TX_WRITE intent log record.
1289 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1291 zvol_state_t *zv = arg;
1292 objset_t *os = zv->zv_objset;
1293 uint64_t object = ZVOL_OBJ;
1294 uint64_t offset = lr->lr_offset;
1295 uint64_t size = lr->lr_length; /* length of user data */
1296 blkptr_t *bp = &lr->lr_blkptr;
1301 ASSERT(zio != NULL);
1304 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1305 zgd->zgd_zilog = zv->zv_zilog;
1306 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1309 * Write records come in two flavors: immediate and indirect.
1310 * For small writes it's cheaper to store the data with the
1311 * log record (immediate); for large writes it's cheaper to
1312 * sync the data and get a pointer to it (indirect) so that
1313 * we don't have to write the data twice.
1315 if (buf != NULL) { /* immediate write */
1316 error = dmu_read(os, object, offset, size, buf,
1317 DMU_READ_NO_PREFETCH);
1319 size = zv->zv_volblocksize;
1320 offset = P2ALIGN(offset, size);
1321 error = dmu_buf_hold(os, object, offset, zgd, &db,
1322 DMU_READ_NO_PREFETCH);
1324 blkptr_t *obp = dmu_buf_get_blkptr(db);
1326 ASSERT(BP_IS_HOLE(bp));
1333 ASSERT(db->db_offset == offset);
1334 ASSERT(db->db_size == size);
1336 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1337 zvol_get_done, zgd);
1344 zvol_get_done(zgd, error);
1350 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1352 * We store data in the log buffers if it's small enough.
1353 * Otherwise we will later flush the data out via dmu_sync().
1355 ssize_t zvol_immediate_write_sz = 32768;
1358 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1361 uint32_t blocksize = zv->zv_volblocksize;
1362 zilog_t *zilog = zv->zv_zilog;
1364 ssize_t immediate_write_sz;
1366 if (zil_replaying(zilog, tx))
1369 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1370 ? 0 : zvol_immediate_write_sz;
1372 slogging = spa_has_slogs(zilog->zl_spa) &&
1373 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1379 itx_wr_state_t write_state;
1382 * Unlike zfs_log_write() we can be called with
1383 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1385 if (blocksize > immediate_write_sz && !slogging &&
1386 resid >= blocksize && off % blocksize == 0) {
1387 write_state = WR_INDIRECT; /* uses dmu_sync */
1390 write_state = WR_COPIED;
1391 len = MIN(ZIL_MAX_LOG_DATA, resid);
1393 write_state = WR_NEED_COPY;
1394 len = MIN(ZIL_MAX_LOG_DATA, resid);
1397 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1398 (write_state == WR_COPIED ? len : 0));
1399 lr = (lr_write_t *)&itx->itx_lr;
1400 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1401 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1402 zil_itx_destroy(itx);
1403 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1404 lr = (lr_write_t *)&itx->itx_lr;
1405 write_state = WR_NEED_COPY;
1408 itx->itx_wr_state = write_state;
1409 if (write_state == WR_NEED_COPY)
1410 itx->itx_sod += len;
1411 lr->lr_foid = ZVOL_OBJ;
1412 lr->lr_offset = off;
1413 lr->lr_length = len;
1415 BP_ZERO(&lr->lr_blkptr);
1417 itx->itx_private = zv;
1418 itx->itx_sync = sync;
1420 zil_itx_assign(zilog, itx, tx);
1429 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1430 uint64_t size, boolean_t doread, boolean_t isdump)
1436 if (vd->vdev_ops == &vdev_mirror_ops ||
1437 vd->vdev_ops == &vdev_replacing_ops ||
1438 vd->vdev_ops == &vdev_spare_ops) {
1439 for (c = 0; c < vd->vdev_children; c++) {
1440 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1441 addr, offset, origoffset, size, doread, isdump);
1444 } else if (doread) {
1450 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1451 return (numerrors < vd->vdev_children ? 0 : EIO);
1453 if (doread && !vdev_readable(vd))
1454 return (SET_ERROR(EIO));
1455 else if (!doread && !vdev_writeable(vd))
1456 return (SET_ERROR(EIO));
1458 if (vd->vdev_ops == &vdev_raidz_ops) {
1459 return (vdev_raidz_physio(vd,
1460 addr, size, offset, origoffset, doread, isdump));
1463 offset += VDEV_LABEL_START_SIZE;
1465 if (ddi_in_panic() || isdump) {
1468 return (SET_ERROR(EIO));
1470 ASSERT3P(dvd, !=, NULL);
1471 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1475 ASSERT3P(dvd, !=, NULL);
1476 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1477 offset, doread ? B_READ : B_WRITE));
1482 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1483 boolean_t doread, boolean_t isdump)
1488 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1490 /* Must be sector aligned, and not stradle a block boundary. */
1491 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1492 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1493 return (SET_ERROR(EINVAL));
1495 ASSERT(size <= zv->zv_volblocksize);
1497 /* Locate the extent this belongs to */
1498 ze = list_head(&zv->zv_extents);
1499 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1500 offset -= ze->ze_nblks * zv->zv_volblocksize;
1501 ze = list_next(&zv->zv_extents, ze);
1505 return (SET_ERROR(EINVAL));
1507 if (!ddi_in_panic())
1508 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1510 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1511 offset += DVA_GET_OFFSET(&ze->ze_dva);
1512 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1513 size, doread, isdump);
1515 if (!ddi_in_panic())
1516 spa_config_exit(spa, SCL_STATE, FTAG);
1522 zvol_strategy(buf_t *bp)
1524 zfs_soft_state_t *zs = NULL;
1525 #else /* !illumos */
1527 zvol_strategy(struct bio *bp)
1529 #endif /* illumos */
1531 uint64_t off, volsize;
1538 boolean_t doread = bp->b_flags & B_READ;
1540 boolean_t doread = 0;
1542 boolean_t is_dumpified;
1546 if (getminor(bp->b_edev) == 0) {
1547 error = SET_ERROR(EINVAL);
1549 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1551 error = SET_ERROR(ENXIO);
1552 else if (zs->zss_type != ZSST_ZVOL)
1553 error = SET_ERROR(EINVAL);
1557 bioerror(bp, error);
1564 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1565 bioerror(bp, EROFS);
1570 off = ldbtob(bp->b_blkno);
1571 #else /* !illumos */
1573 zv = bp->bio_to->private;
1575 zv = bp->bio_dev->si_drv2;
1578 error = SET_ERROR(ENXIO);
1582 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1583 error = SET_ERROR(EROFS);
1587 switch (bp->bio_cmd) {
1600 off = bp->bio_offset;
1601 #endif /* illumos */
1602 volsize = zv->zv_volsize;
1609 addr = bp->b_un.b_addr;
1610 resid = bp->b_bcount;
1612 if (resid > 0 && (off < 0 || off >= volsize)) {
1618 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1619 sync = ((!(bp->b_flags & B_ASYNC) &&
1620 !(zv->zv_flags & ZVOL_WCE)) ||
1621 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1622 !doread && !is_dumpified;
1623 #else /* !illumos */
1624 addr = bp->bio_data;
1625 resid = bp->bio_length;
1627 if (resid > 0 && (off < 0 || off >= volsize)) {
1628 error = SET_ERROR(EIO);
1632 is_dumpified = B_FALSE;
1633 sync = !doread && !is_dumpified &&
1634 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1635 #endif /* illumos */
1638 * There must be no buffer changes when doing a dmu_sync() because
1639 * we can't change the data whilst calculating the checksum.
1641 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1642 doread ? RL_READER : RL_WRITER);
1645 if (bp->bio_cmd == BIO_DELETE) {
1646 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1647 error = dmu_tx_assign(tx, TXG_WAIT);
1651 zvol_log_truncate(zv, tx, off, resid, B_TRUE);
1653 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1660 while (resid != 0 && off < volsize) {
1661 size_t size = MIN(resid, zvol_maxphys);
1664 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1665 error = zvol_dumpio(zv, addr, off, size,
1667 } else if (doread) {
1671 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1674 dmu_tx_t *tx = dmu_tx_create(os);
1675 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1676 error = dmu_tx_assign(tx, TXG_WAIT);
1680 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1681 zvol_log_write(zv, tx, off, size, sync);
1686 /* convert checksum errors into IO errors */
1687 if (error == ECKSUM)
1688 error = SET_ERROR(EIO);
1698 zfs_range_unlock(rl);
1701 if ((bp->b_resid = resid) == bp->b_bcount)
1702 bioerror(bp, off > volsize ? EINVAL : error);
1705 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1709 #else /* !illumos */
1710 bp->bio_completed = bp->bio_length - resid;
1711 if (bp->bio_completed < bp->bio_length && off > volsize)
1716 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1720 g_io_deliver(bp, error);
1722 biofinish(bp, NULL, error);
1723 #endif /* illumos */
1728 * Set the buffer count to the zvol maximum transfer.
1729 * Using our own routine instead of the default minphys()
1730 * means that for larger writes we write bigger buffers on X86
1731 * (128K instead of 56K) and flush the disk write cache less often
1732 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1733 * 56K on X86 and 128K on sparc).
1736 zvol_minphys(struct buf *bp)
1738 if (bp->b_bcount > zvol_maxphys)
1739 bp->b_bcount = zvol_maxphys;
1743 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1745 minor_t minor = getminor(dev);
1752 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1754 return (SET_ERROR(ENXIO));
1756 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1757 return (SET_ERROR(EINVAL));
1759 boff = ldbtob(blkno);
1760 resid = ldbtob(nblocks);
1762 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1765 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1766 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1779 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1781 minor_t minor = getminor(dev);
1782 #else /* !illumos */
1784 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1786 #endif /* illumos */
1793 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1795 return (SET_ERROR(ENXIO));
1800 volsize = zv->zv_volsize;
1801 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1802 if (uio->uio_resid > 0 &&
1803 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1804 return (SET_ERROR(EIO));
1807 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1808 error = physio(zvol_strategy, NULL, dev, B_READ,
1814 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1816 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1817 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1819 /* don't read past the end */
1820 if (bytes > volsize - uio->uio_loffset)
1821 bytes = volsize - uio->uio_loffset;
1823 error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
1825 /* convert checksum errors into IO errors */
1826 if (error == ECKSUM)
1827 error = SET_ERROR(EIO);
1831 zfs_range_unlock(rl);
1838 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1840 minor_t minor = getminor(dev);
1841 #else /* !illumos */
1843 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1845 #endif /* illumos */
1853 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1855 return (SET_ERROR(ENXIO));
1860 volsize = zv->zv_volsize;
1861 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1862 if (uio->uio_resid > 0 &&
1863 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1864 return (SET_ERROR(EIO));
1867 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1868 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1873 sync = !(zv->zv_flags & ZVOL_WCE) ||
1875 sync = (ioflag & IO_SYNC) ||
1877 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1879 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1881 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1882 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1883 uint64_t off = uio->uio_loffset;
1884 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1886 if (bytes > volsize - off) /* don't write past the end */
1887 bytes = volsize - off;
1889 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1890 error = dmu_tx_assign(tx, TXG_WAIT);
1895 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1897 zvol_log_write(zv, tx, off, bytes, sync);
1903 zfs_range_unlock(rl);
1905 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1911 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1913 struct uuid uuid = EFI_RESERVED;
1914 efi_gpe_t gpe = { 0 };
1920 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1921 return (SET_ERROR(EFAULT));
1922 ptr = (char *)(uintptr_t)efi.dki_data_64;
1923 length = efi.dki_length;
1925 * Some clients may attempt to request a PMBR for the
1926 * zvol. Currently this interface will return EINVAL to
1927 * such requests. These requests could be supported by
1928 * adding a check for lba == 0 and consing up an appropriate
1931 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1932 return (SET_ERROR(EINVAL));
1934 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1935 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1936 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1938 if (efi.dki_lba == 1) {
1939 efi_gpt_t gpt = { 0 };
1941 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1942 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1943 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1944 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1945 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1946 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1947 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1948 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1949 gpt.efi_gpt_SizeOfPartitionEntry =
1950 LE_32(sizeof (efi_gpe_t));
1951 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1952 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1953 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1954 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1955 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1957 return (SET_ERROR(EFAULT));
1958 ptr += sizeof (gpt);
1959 length -= sizeof (gpt);
1961 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1963 return (SET_ERROR(EFAULT));
1968 * BEGIN entry points to allow external callers access to the volume.
1971 * Return the volume parameters needed for access from an external caller.
1972 * These values are invariant as long as the volume is held open.
1975 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1976 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1977 void **rl_hdl, void **bonus_hdl)
1981 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1983 return (SET_ERROR(ENXIO));
1984 if (zv->zv_flags & ZVOL_DUMPIFIED)
1985 return (SET_ERROR(ENXIO));
1987 ASSERT(blksize && max_xfer_len && minor_hdl &&
1988 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1990 *blksize = zv->zv_volblocksize;
1991 *max_xfer_len = (uint64_t)zvol_maxphys;
1993 *objset_hdl = zv->zv_objset;
1994 *zil_hdl = zv->zv_zilog;
1995 *rl_hdl = &zv->zv_znode;
1996 *bonus_hdl = zv->zv_dbuf;
2001 * Return the current volume size to an external caller.
2002 * The size can change while the volume is open.
2005 zvol_get_volume_size(void *minor_hdl)
2007 zvol_state_t *zv = minor_hdl;
2009 return (zv->zv_volsize);
2013 * Return the current WCE setting to an external caller.
2014 * The WCE setting can change while the volume is open.
2017 zvol_get_volume_wce(void *minor_hdl)
2019 zvol_state_t *zv = minor_hdl;
2021 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2025 * Entry point for external callers to zvol_log_write
2028 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2031 zvol_state_t *zv = minor_hdl;
2033 zvol_log_write(zv, tx, off, resid, sync);
2036 * END entry points to allow external callers access to the volume.
2038 #endif /* illumos */
2041 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2044 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2049 zilog_t *zilog = zv->zv_zilog;
2051 if (zil_replaying(zilog, tx))
2054 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2055 lr = (lr_truncate_t *)&itx->itx_lr;
2056 lr->lr_foid = ZVOL_OBJ;
2057 lr->lr_offset = off;
2058 lr->lr_length = len;
2060 itx->itx_sync = sync;
2061 zil_itx_assign(zilog, itx, tx);
2066 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2067 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2071 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2074 struct dk_callback *dkc;
2078 mutex_enter(&zfsdev_state_lock);
2080 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2083 mutex_exit(&zfsdev_state_lock);
2084 return (SET_ERROR(ENXIO));
2086 ASSERT(zv->zv_total_opens > 0);
2092 struct dk_cinfo dki;
2094 bzero(&dki, sizeof (dki));
2095 (void) strcpy(dki.dki_cname, "zvol");
2096 (void) strcpy(dki.dki_dname, "zvol");
2097 dki.dki_ctype = DKC_UNKNOWN;
2098 dki.dki_unit = getminor(dev);
2099 dki.dki_maxtransfer =
2100 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2101 mutex_exit(&zfsdev_state_lock);
2102 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2103 error = SET_ERROR(EFAULT);
2107 case DKIOCGMEDIAINFO:
2109 struct dk_minfo dkm;
2111 bzero(&dkm, sizeof (dkm));
2112 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2113 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2114 dkm.dki_media_type = DK_UNKNOWN;
2115 mutex_exit(&zfsdev_state_lock);
2116 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2117 error = SET_ERROR(EFAULT);
2121 case DKIOCGMEDIAINFOEXT:
2123 struct dk_minfo_ext dkmext;
2125 bzero(&dkmext, sizeof (dkmext));
2126 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2127 dkmext.dki_pbsize = zv->zv_volblocksize;
2128 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2129 dkmext.dki_media_type = DK_UNKNOWN;
2130 mutex_exit(&zfsdev_state_lock);
2131 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2132 error = SET_ERROR(EFAULT);
2138 uint64_t vs = zv->zv_volsize;
2139 uint8_t bs = zv->zv_min_bs;
2141 mutex_exit(&zfsdev_state_lock);
2142 error = zvol_getefi((void *)arg, flag, vs, bs);
2146 case DKIOCFLUSHWRITECACHE:
2147 dkc = (struct dk_callback *)arg;
2148 mutex_exit(&zfsdev_state_lock);
2149 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2150 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2151 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2158 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2159 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2161 error = SET_ERROR(EFAULT);
2167 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2169 error = SET_ERROR(EFAULT);
2173 zv->zv_flags |= ZVOL_WCE;
2174 mutex_exit(&zfsdev_state_lock);
2176 zv->zv_flags &= ~ZVOL_WCE;
2177 mutex_exit(&zfsdev_state_lock);
2178 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2186 * commands using these (like prtvtoc) expect ENOTSUP
2187 * since we're emulating an EFI label
2189 error = SET_ERROR(ENOTSUP);
2193 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2195 error = zvol_dumpify(zv);
2196 zfs_range_unlock(rl);
2200 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2202 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
2204 error = zvol_dump_fini(zv);
2205 zfs_range_unlock(rl);
2213 if (!zvol_unmap_enabled)
2216 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
2217 error = SET_ERROR(EFAULT);
2222 * Apply Postel's Law to length-checking. If they overshoot,
2223 * just blank out until the end, if there's a need to blank
2226 if (df.df_start >= zv->zv_volsize)
2227 break; /* No need to do anything... */
2229 mutex_exit(&zfsdev_state_lock);
2231 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
2233 tx = dmu_tx_create(zv->zv_objset);
2234 dmu_tx_mark_netfree(tx);
2235 error = dmu_tx_assign(tx, TXG_WAIT);
2239 zvol_log_truncate(zv, tx, df.df_start,
2240 df.df_length, B_TRUE);
2242 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
2243 df.df_start, df.df_length);
2246 zfs_range_unlock(rl);
2250 * If the write-cache is disabled or 'sync' property
2251 * is set to 'always' then treat this as a synchronous
2252 * operation (i.e. commit to zil).
2254 if (!(zv->zv_flags & ZVOL_WCE) ||
2255 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
2256 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2259 * If the caller really wants synchronous writes, and
2260 * can't wait for them, don't return until the write
2263 if (df.df_flags & DF_WAIT_SYNC) {
2265 dmu_objset_pool(zv->zv_objset), 0);
2272 error = SET_ERROR(ENOTTY);
2276 mutex_exit(&zfsdev_state_lock);
2279 #endif /* illumos */
2284 return (zvol_minors != 0);
2290 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2293 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2295 ZFS_LOG(1, "ZVOL Initialized.");
2303 mutex_destroy(&zfsdev_state_lock);
2305 ddi_soft_state_fini(&zfsdev_state);
2306 ZFS_LOG(1, "ZVOL Deinitialized.");
2312 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2314 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2316 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2323 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2325 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2327 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2331 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2335 objset_t *os = zv->zv_objset;
2336 spa_t *spa = dmu_objset_spa(os);
2337 vdev_t *vd = spa->spa_root_vdev;
2338 nvlist_t *nv = NULL;
2339 uint64_t version = spa_version(spa);
2340 uint64_t checksum, compress, refresrv, vbs, dedup;
2342 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2343 ASSERT(vd->vdev_ops == &vdev_root_ops);
2345 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2349 /* wait for dmu_free_long_range to actually free the blocks */
2350 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2353 * If the pool on which the dump device is being initialized has more
2354 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2355 * enabled. If so, bump that feature's counter to indicate that the
2356 * feature is active. We also check the vdev type to handle the
2358 * # zpool create test raidz disk1 disk2 disk3
2359 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2360 * the raidz vdev itself has 3 children.
2362 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2363 if (!spa_feature_is_enabled(spa,
2364 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2365 return (SET_ERROR(ENOTSUP));
2366 (void) dsl_sync_task(spa_name(spa),
2367 zfs_mvdev_dump_feature_check,
2368 zfs_mvdev_dump_activate_feature_sync, NULL,
2369 2, ZFS_SPACE_CHECK_RESERVED);
2373 error = dsl_prop_get_integer(zv->zv_name,
2374 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2376 error = dsl_prop_get_integer(zv->zv_name,
2377 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2381 error = dsl_prop_get_integer(zv->zv_name,
2382 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2386 error = dsl_prop_get_integer(zv->zv_name,
2387 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2390 if (version >= SPA_VERSION_DEDUP && error == 0) {
2391 error = dsl_prop_get_integer(zv->zv_name,
2392 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2398 tx = dmu_tx_create(os);
2399 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2400 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2401 error = dmu_tx_assign(tx, TXG_WAIT);
2408 * If we are resizing the dump device then we only need to
2409 * update the refreservation to match the newly updated
2410 * zvolsize. Otherwise, we save off the original state of the
2411 * zvol so that we can restore them if the zvol is ever undumpified.
2414 error = zap_update(os, ZVOL_ZAP_OBJ,
2415 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2416 &zv->zv_volsize, tx);
2418 error = zap_update(os, ZVOL_ZAP_OBJ,
2419 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2422 error = zap_update(os, ZVOL_ZAP_OBJ,
2423 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2427 error = zap_update(os, ZVOL_ZAP_OBJ,
2428 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2432 error = zap_update(os, ZVOL_ZAP_OBJ,
2433 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2437 error = dmu_object_set_blocksize(
2438 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2440 if (version >= SPA_VERSION_DEDUP && error == 0) {
2441 error = zap_update(os, ZVOL_ZAP_OBJ,
2442 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2446 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2451 * We only need update the zvol's property if we are initializing
2452 * the dump area for the first time.
2454 if (error == 0 && !resize) {
2456 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2457 * function. Otherwise, use the old default -- OFF.
2459 checksum = spa_feature_is_active(spa,
2460 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2463 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2464 VERIFY(nvlist_add_uint64(nv,
2465 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2466 VERIFY(nvlist_add_uint64(nv,
2467 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2468 ZIO_COMPRESS_OFF) == 0);
2469 VERIFY(nvlist_add_uint64(nv,
2470 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2472 if (version >= SPA_VERSION_DEDUP) {
2473 VERIFY(nvlist_add_uint64(nv,
2474 zfs_prop_to_name(ZFS_PROP_DEDUP),
2475 ZIO_CHECKSUM_OFF) == 0);
2478 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2483 /* Allocate the space for the dump */
2485 error = zvol_prealloc(zv);
2490 zvol_dumpify(zvol_state_t *zv)
2493 uint64_t dumpsize = 0;
2495 objset_t *os = zv->zv_objset;
2497 if (zv->zv_flags & ZVOL_RDONLY)
2498 return (SET_ERROR(EROFS));
2500 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2501 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2502 boolean_t resize = (dumpsize > 0);
2504 if ((error = zvol_dump_init(zv, resize)) != 0) {
2505 (void) zvol_dump_fini(zv);
2511 * Build up our lba mapping.
2513 error = zvol_get_lbas(zv);
2515 (void) zvol_dump_fini(zv);
2519 tx = dmu_tx_create(os);
2520 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2521 error = dmu_tx_assign(tx, TXG_WAIT);
2524 (void) zvol_dump_fini(zv);
2528 zv->zv_flags |= ZVOL_DUMPIFIED;
2529 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2530 &zv->zv_volsize, tx);
2534 (void) zvol_dump_fini(zv);
2538 txg_wait_synced(dmu_objset_pool(os), 0);
2543 zvol_dump_fini(zvol_state_t *zv)
2546 objset_t *os = zv->zv_objset;
2549 uint64_t checksum, compress, refresrv, vbs, dedup;
2550 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2553 * Attempt to restore the zvol back to its pre-dumpified state.
2554 * This is a best-effort attempt as it's possible that not all
2555 * of these properties were initialized during the dumpify process
2556 * (i.e. error during zvol_dump_init).
2559 tx = dmu_tx_create(os);
2560 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2561 error = dmu_tx_assign(tx, TXG_WAIT);
2566 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2569 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2570 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2571 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2572 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2573 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2574 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2575 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2576 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2578 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2579 (void) nvlist_add_uint64(nv,
2580 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2581 (void) nvlist_add_uint64(nv,
2582 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2583 (void) nvlist_add_uint64(nv,
2584 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2585 if (version >= SPA_VERSION_DEDUP &&
2586 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2587 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2588 (void) nvlist_add_uint64(nv,
2589 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2591 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2595 zvol_free_extents(zv);
2596 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2597 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2598 /* wait for dmu_free_long_range to actually free the blocks */
2599 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2600 tx = dmu_tx_create(os);
2601 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2602 error = dmu_tx_assign(tx, TXG_WAIT);
2607 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2608 zv->zv_volblocksize = vbs;
2613 #else /* !illumos */
2616 zvol_geom_run(zvol_state_t *zv)
2618 struct g_provider *pp;
2620 pp = zv->zv_provider;
2621 g_error_provider(pp, 0);
2623 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
2624 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2628 zvol_geom_destroy(zvol_state_t *zv)
2630 struct g_provider *pp;
2632 g_topology_assert();
2634 mtx_lock(&zv->zv_queue_mtx);
2636 wakeup_one(&zv->zv_queue);
2637 while (zv->zv_state != 2)
2638 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2639 mtx_destroy(&zv->zv_queue_mtx);
2641 pp = zv->zv_provider;
2642 zv->zv_provider = NULL;
2644 g_wither_geom(pp->geom, ENXIO);
2648 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2650 int count, error, flags;
2652 g_topology_assert();
2655 * To make it easier we expect either open or close, but not both
2658 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2659 (acr <= 0 && acw <= 0 && ace <= 0),
2660 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2661 pp->name, acr, acw, ace));
2663 if (pp->private == NULL) {
2664 if (acr <= 0 && acw <= 0 && ace <= 0)
2670 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2671 * because GEOM already handles that and handles it a bit differently.
2672 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2673 * only one exclusive consumer, no matter if it is reader or writer.
2674 * I like better the way GEOM works so I'll leave it for GEOM to
2675 * decide what to do.
2678 count = acr + acw + ace;
2683 if (acr != 0 || ace != 0)
2688 g_topology_unlock();
2690 error = zvol_open(pp, flags, count);
2692 error = zvol_close(pp, flags, -count);
2698 zvol_geom_start(struct bio *bp)
2703 zv = bp->bio_to->private;
2705 switch (bp->bio_cmd) {
2707 if (!THREAD_CAN_SLEEP())
2709 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2710 g_io_deliver(bp, 0);
2715 if (!THREAD_CAN_SLEEP())
2720 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2721 uint64_t refd, avail, usedobjs, availobjs, val;
2723 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2725 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2726 dmu_objset_space(zv->zv_objset, &refd, &avail,
2727 &usedobjs, &availobjs);
2728 if (g_handleattr_off_t(bp, "blocksavail",
2731 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2732 dmu_objset_space(zv->zv_objset, &refd, &avail,
2733 &usedobjs, &availobjs);
2734 if (g_handleattr_off_t(bp, "blocksused",
2737 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2738 avail = metaslab_class_get_space(spa_normal_class(spa));
2739 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2740 if (g_handleattr_off_t(bp, "poolblocksavail",
2743 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2744 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2745 if (g_handleattr_off_t(bp, "poolblocksused",
2752 g_io_deliver(bp, EOPNOTSUPP);
2758 mtx_lock(&zv->zv_queue_mtx);
2759 first = (bioq_first(&zv->zv_queue) == NULL);
2760 bioq_insert_tail(&zv->zv_queue, bp);
2761 mtx_unlock(&zv->zv_queue_mtx);
2763 wakeup_one(&zv->zv_queue);
2767 zvol_geom_worker(void *arg)
2772 thread_lock(curthread);
2773 sched_prio(curthread, PRIBIO);
2774 thread_unlock(curthread);
2778 mtx_lock(&zv->zv_queue_mtx);
2779 bp = bioq_takefirst(&zv->zv_queue);
2781 if (zv->zv_state == 1) {
2783 wakeup(&zv->zv_state);
2784 mtx_unlock(&zv->zv_queue_mtx);
2787 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2791 mtx_unlock(&zv->zv_queue_mtx);
2792 switch (bp->bio_cmd) {
2794 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2795 g_io_deliver(bp, 0);
2803 g_io_deliver(bp, EOPNOTSUPP);
2809 extern boolean_t dataset_name_hidden(const char *name);
2812 zvol_create_snapshots(objset_t *os, const char *name)
2814 uint64_t cookie, obj;
2819 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2822 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2827 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2828 if (len >= MAXPATHLEN) {
2829 dmu_objset_rele(os, FTAG);
2830 error = ENAMETOOLONG;
2834 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2835 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2836 sname + len, &obj, &cookie, NULL);
2837 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2839 if (error == ENOENT)
2844 error = zvol_create_minor(sname);
2845 if (error != 0 && error != EEXIST) {
2846 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2852 kmem_free(sname, MAXPATHLEN);
2857 zvol_create_minors(const char *name)
2864 if (dataset_name_hidden(name))
2867 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2868 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2872 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2873 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2874 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2875 error = zvol_create_minor(name);
2876 if (error == 0 || error == EEXIST) {
2877 error = zvol_create_snapshots(os, name);
2879 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2882 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2883 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2886 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2887 dmu_objset_rele(os, FTAG);
2891 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2892 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2893 dmu_objset_rele(os, FTAG);
2894 kmem_free(osname, MAXPATHLEN);
2897 p = osname + strlen(osname);
2898 len = MAXPATHLEN - (p - osname);
2901 /* Prefetch the datasets. */
2903 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2904 if (!dataset_name_hidden(osname))
2905 (void) dmu_objset_prefetch(osname, NULL);
2910 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2912 dmu_objset_rele(os, FTAG);
2913 (void)zvol_create_minors(osname);
2914 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2915 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2921 dmu_objset_rele(os, FTAG);
2922 kmem_free(osname, MAXPATHLEN);
2927 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2930 struct g_provider *pp;
2933 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2935 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2937 pp = zv->zv_provider;
2942 zv->zv_provider = NULL;
2943 g_wither_provider(pp, ENXIO);
2945 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2946 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
2947 pp->sectorsize = DEV_BSIZE;
2948 pp->mediasize = zv->zv_volsize;
2950 zv->zv_provider = pp;
2951 g_error_provider(pp, 0);
2952 g_topology_unlock();
2953 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
2955 ASSERT(dev != NULL);
2959 if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
2960 &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
2961 0640, "%s/%s", ZVOL_DRIVER, newname) == 0) {
2963 dev->si_iosize_max = MAXPHYS;
2967 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
2971 zvol_rename_minors(const char *oldname, const char *newname)
2973 char name[MAXPATHLEN];
2974 struct g_provider *pp;
2976 size_t oldnamelen, newnamelen;
2979 boolean_t locked = B_FALSE;
2981 oldnamelen = strlen(oldname);
2982 newnamelen = strlen(newname);
2985 /* See comment in zvol_open(). */
2986 if (!MUTEX_HELD(&zfsdev_state_lock)) {
2987 mutex_enter(&zfsdev_state_lock);
2991 LIST_FOREACH(zv, &all_zvols, zv_links) {
2992 if (strcmp(zv->zv_name, oldname) == 0) {
2993 zvol_rename_minor(zv, newname);
2994 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
2995 (zv->zv_name[oldnamelen] == '/' ||
2996 zv->zv_name[oldnamelen] == '@')) {
2997 snprintf(name, sizeof(name), "%s%c%s", newname,
2998 zv->zv_name[oldnamelen],
2999 zv->zv_name + oldnamelen + 1);
3000 zvol_rename_minor(zv, name);
3005 mutex_exit(&zfsdev_state_lock);
3010 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3015 mutex_enter(&zfsdev_state_lock);
3018 mutex_exit(&zfsdev_state_lock);
3019 return(ENXIO); /* zvol_create_minor() not done yet */
3022 if (zv->zv_total_opens == 0)
3023 err = zvol_first_open(zv);
3025 mutex_exit(&zfsdev_state_lock);
3028 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3029 err = SET_ERROR(EROFS);
3032 if (zv->zv_flags & ZVOL_EXCL) {
3033 err = SET_ERROR(EBUSY);
3037 if (flags & FEXCL) {
3038 if (zv->zv_total_opens != 0) {
3039 err = SET_ERROR(EBUSY);
3042 zv->zv_flags |= ZVOL_EXCL;
3046 zv->zv_total_opens++;
3047 mutex_exit(&zfsdev_state_lock);
3050 if (zv->zv_total_opens == 0)
3051 zvol_last_close(zv);
3052 mutex_exit(&zfsdev_state_lock);
3057 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3062 mutex_enter(&zfsdev_state_lock);
3065 mutex_exit(&zfsdev_state_lock);
3069 if (zv->zv_flags & ZVOL_EXCL) {
3070 ASSERT(zv->zv_total_opens == 1);
3071 zv->zv_flags &= ~ZVOL_EXCL;
3075 * If the open count is zero, this is a spurious close.
3076 * That indicates a bug in the kernel / DDI framework.
3078 ASSERT(zv->zv_total_opens != 0);
3081 * You may get multiple opens, but only one close.
3083 zv->zv_total_opens--;
3085 if (zv->zv_total_opens == 0)
3086 zvol_last_close(zv);
3088 mutex_exit(&zfsdev_state_lock);
3093 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3097 off_t offset, length, chunk;
3104 KASSERT(zv->zv_total_opens > 0,
3105 ("Device with zero access count in zvol_d_ioctl"));
3107 i = IOCPARM_LEN(cmd);
3109 case DIOCGSECTORSIZE:
3110 *(u_int *)data = DEV_BSIZE;
3112 case DIOCGMEDIASIZE:
3113 *(off_t *)data = zv->zv_volsize;
3116 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3119 if (!zvol_unmap_enabled)
3122 offset = ((off_t *)data)[0];
3123 length = ((off_t *)data)[1];
3124 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3125 offset < 0 || offset >= zv->zv_volsize ||
3127 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3133 rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
3134 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3135 error = dmu_tx_assign(tx, TXG_WAIT);
3139 zvol_log_truncate(zv, tx, offset, length, B_TRUE);
3141 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3144 zfs_range_unlock(rl);
3145 if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
3146 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3148 case DIOCGSTRIPESIZE:
3149 *(off_t *)data = zv->zv_volblocksize;
3151 case DIOCGSTRIPEOFFSET:
3155 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3156 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3157 uint64_t refd, avail, usedobjs, availobjs;
3159 if (strcmp(arg->name, "GEOM::candelete") == 0)
3161 else if (strcmp(arg->name, "blocksavail") == 0) {
3162 dmu_objset_space(zv->zv_objset, &refd, &avail,
3163 &usedobjs, &availobjs);
3164 arg->value.off = avail / DEV_BSIZE;
3165 } else if (strcmp(arg->name, "blocksused") == 0) {
3166 dmu_objset_space(zv->zv_objset, &refd, &avail,
3167 &usedobjs, &availobjs);
3168 arg->value.off = refd / DEV_BSIZE;
3169 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3170 avail = metaslab_class_get_space(spa_normal_class(spa));
3171 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3172 arg->value.off = avail / DEV_BSIZE;
3173 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3174 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3175 arg->value.off = refd / DEV_BSIZE;
3182 off_t *off = (off_t *)data;
3186 hole = (cmd == FIOSEEKHOLE);
3188 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3198 #endif /* illumos */