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 2017 Nexenta Systems, Inc. All rights reserved.
30 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
31 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
32 * Copyright (c) 2014 Integros [integros.com]
33 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
36 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
39 * ZFS volume emulation driver.
41 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
42 * Volumes are accessed through the symbolic links named:
44 * /dev/zvol/dsk/<pool_name>/<dataset_name>
45 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
47 * These links are created by the /dev filesystem (sdev_zvolops.c).
48 * Volumes are persistent through reboot. No user command needs to be
49 * run before opening and using a device.
52 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
56 #include <sys/types.h>
57 #include <sys/param.h>
58 #include <sys/kernel.h>
59 #include <sys/errno.h>
65 #include <sys/cmn_err.h>
69 #include <sys/spa_impl.h>
72 #include <sys/dmu_traverse.h>
73 #include <sys/dnode.h>
74 #include <sys/dsl_dataset.h>
75 #include <sys/dsl_prop.h>
77 #include <sys/byteorder.h>
78 #include <sys/sunddi.h>
79 #include <sys/dirent.h>
80 #include <sys/policy.h>
81 #include <sys/queue.h>
82 #include <sys/fs/zfs.h>
83 #include <sys/zfs_ioctl.h>
85 #include <sys/refcount.h>
86 #include <sys/zfs_znode.h>
87 #include <sys/zfs_rlock.h>
88 #include <sys/vdev_impl.h>
89 #include <sys/vdev_raidz.h>
91 #include <sys/zil_impl.h>
93 #include <sys/dmu_tx.h>
94 #include <sys/zfeature.h>
95 #include <sys/zio_checksum.h>
96 #include <sys/zil_impl.h>
97 #include <sys/filio.h>
98 #include <sys/zfs_rlock.h>
100 #include <geom/geom.h>
102 #include "zfs_namecheck.h"
105 struct g_class zfs_zvol_class = {
107 .version = G_VERSION,
110 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
114 static char *zvol_tag = "zvol_tag";
116 #define ZVOL_DUMPSIZE "dumpsize"
119 * This lock protects the zfsdev_state structure from being modified
120 * while it's being used, e.g. an open that comes in before a create
121 * finishes. It also protects temporary opens of the dataset so that,
122 * e.g., an open doesn't get a spurious EBUSY.
125 kmutex_t zfsdev_state_lock;
128 * In FreeBSD we've replaced the upstream zfsdev_state_lock with the
129 * spa_namespace_lock in the ZVOL code.
131 #define zfsdev_state_lock spa_namespace_lock
133 static uint32_t zvol_minors;
136 SYSCTL_DECL(_vfs_zfs);
137 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
139 static int volmode = ZFS_VOLMODE_GEOM;
140 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
141 "Expose as GEOM providers (1), device files (2) or neither");
142 static boolean_t zpool_on_zvol = B_FALSE;
143 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, recursive, CTLFLAG_RWTUN, &zpool_on_zvol, 0,
144 "Allow zpools to use zvols as vdevs (DANGEROUS)");
147 typedef struct zvol_extent {
149 dva_t ze_dva; /* dva associated with this extent */
150 uint64_t ze_nblks; /* number of blocks in extent */
154 * The in-core state of each volume.
156 typedef struct zvol_state {
158 LIST_ENTRY(zvol_state) zv_links;
160 char zv_name[MAXPATHLEN]; /* pool/dd name */
161 uint64_t zv_volsize; /* amount of space we advertise */
162 uint64_t zv_volblocksize; /* volume block size */
164 minor_t zv_minor; /* minor number */
166 struct cdev *zv_dev; /* non-GEOM device */
167 struct g_provider *zv_provider; /* GEOM provider */
169 uint8_t zv_min_bs; /* minimum addressable block shift */
170 uint8_t zv_flags; /* readonly, dumpified, etc. */
171 objset_t *zv_objset; /* objset handle */
173 uint32_t zv_open_count[OTYPCNT]; /* open counts */
175 uint32_t zv_total_opens; /* total open count */
176 uint32_t zv_sync_cnt; /* synchronous open count */
177 zilog_t *zv_zilog; /* ZIL handle */
178 list_t zv_extents; /* List of extents for dump */
179 rangelock_t zv_rangelock;
180 dnode_t *zv_dn; /* dnode hold */
183 int zv_volmode; /* Provide GEOM or cdev */
184 struct bio_queue_head zv_queue;
185 struct mtx zv_queue_mtx; /* zv_queue mutex */
190 ZVOL_ASYNC_CREATE_MINORS,
191 ZVOL_ASYNC_REMOVE_MINORS,
192 ZVOL_ASYNC_RENAME_MINORS,
198 char pool[ZFS_MAX_DATASET_NAME_LEN];
199 char name1[ZFS_MAX_DATASET_NAME_LEN];
200 char name2[ZFS_MAX_DATASET_NAME_LEN];
204 static LIST_HEAD(, zvol_state) all_zvols;
207 * zvol specific flags
209 #define ZVOL_RDONLY 0x1
210 #define ZVOL_DUMPIFIED 0x2
211 #define ZVOL_EXCL 0x4
215 * zvol maximum transfer in one DMU tx.
217 int zvol_maxphys = DMU_MAX_ACCESS/2;
220 * Toggle unmap functionality.
222 boolean_t zvol_unmap_enabled = B_TRUE;
225 * If true, unmaps requested as synchronous are executed synchronously,
226 * otherwise all unmaps are asynchronous.
228 boolean_t zvol_unmap_sync_enabled = B_FALSE;
231 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN,
232 &zvol_unmap_enabled, 0,
233 "Enable UNMAP functionality");
235 SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_sync_enabled, CTLFLAG_RWTUN,
236 &zvol_unmap_sync_enabled, 0,
237 "UNMAPs requested as sync are executed synchronously");
239 static d_open_t zvol_d_open;
240 static d_close_t zvol_d_close;
241 static d_read_t zvol_read;
242 static d_write_t zvol_write;
243 static d_ioctl_t zvol_d_ioctl;
244 static d_strategy_t zvol_strategy;
246 static struct cdevsw zvol_cdevsw = {
247 .d_version = D_VERSION,
248 .d_open = zvol_d_open,
249 .d_close = zvol_d_close,
251 .d_write = zvol_write,
252 .d_ioctl = zvol_d_ioctl,
253 .d_strategy = zvol_strategy,
255 .d_flags = D_DISK | D_TRACKCLOSE,
258 static void zvol_geom_run(zvol_state_t *zv);
259 static void zvol_geom_destroy(zvol_state_t *zv);
260 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
261 static void zvol_geom_start(struct bio *bp);
262 static void zvol_geom_worker(void *arg);
263 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
264 uint64_t len, boolean_t sync);
265 #endif /* !illumos */
267 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
268 nvlist_t *, nvlist_t *);
269 static int zvol_remove_zv(zvol_state_t *);
270 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
271 struct lwb *lwb, zio_t *zio);
272 static int zvol_dumpify(zvol_state_t *zv);
273 static int zvol_dump_fini(zvol_state_t *zv);
274 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
277 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
280 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
282 zv->zv_volsize = volsize;
283 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
284 "Size", volsize) == DDI_SUCCESS);
285 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
286 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
288 /* Notify specfs to invalidate the cached size */
289 spec_size_invalidate(dev, VBLK);
290 spec_size_invalidate(dev, VCHR);
292 zv->zv_volsize = volsize;
293 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
294 struct g_provider *pp;
296 pp = zv->zv_provider;
302 * Do not invoke resize event when initial size was zero.
303 * ZVOL initializes the size on first open, this is not
306 if (pp->mediasize == 0)
307 pp->mediasize = zv->zv_volsize;
309 g_resize_provider(pp, zv->zv_volsize);
316 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
319 return (SET_ERROR(EINVAL));
321 if (volsize % blocksize != 0)
322 return (SET_ERROR(EINVAL));
325 if (volsize - 1 > SPEC_MAXOFFSET_T)
326 return (SET_ERROR(EOVERFLOW));
332 zvol_check_volblocksize(uint64_t volblocksize)
334 if (volblocksize < SPA_MINBLOCKSIZE ||
335 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
337 return (SET_ERROR(EDOM));
343 zvol_get_stats(objset_t *os, nvlist_t *nv)
346 dmu_object_info_t doi;
349 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
353 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
355 error = dmu_object_info(os, ZVOL_OBJ, &doi);
358 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
359 doi.doi_data_block_size);
365 static zvol_state_t *
366 zvol_minor_lookup(const char *name)
373 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
376 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
377 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
381 LIST_FOREACH(zv, &all_zvols, zv_links) {
383 if (strcmp(zv->zv_name, name) == 0)
390 /* extent mapping arg */
398 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
399 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
401 struct maparg *ma = arg;
403 int bs = ma->ma_zv->zv_volblocksize;
405 if (bp == NULL || BP_IS_HOLE(bp) ||
406 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
409 VERIFY(!BP_IS_EMBEDDED(bp));
411 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
414 /* Abort immediately if we have encountered gang blocks */
416 return (SET_ERROR(EFRAGS));
419 * See if the block is at the end of the previous extent.
421 ze = list_tail(&ma->ma_zv->zv_extents);
423 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
424 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
425 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
430 dprintf_bp(bp, "%s", "next blkptr:");
432 /* start a new extent */
433 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
434 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
436 list_insert_tail(&ma->ma_zv->zv_extents, ze);
441 zvol_free_extents(zvol_state_t *zv)
445 while (ze = list_head(&zv->zv_extents)) {
446 list_remove(&zv->zv_extents, ze);
447 kmem_free(ze, sizeof (zvol_extent_t));
452 zvol_get_lbas(zvol_state_t *zv)
454 objset_t *os = zv->zv_objset;
460 zvol_free_extents(zv);
462 /* commit any in-flight changes before traversing the dataset */
463 txg_wait_synced(dmu_objset_pool(os), 0);
464 err = traverse_dataset(dmu_objset_ds(os), 0,
465 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
466 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
467 zvol_free_extents(zv);
468 return (err ? err : EIO);
476 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
478 zfs_creat_t *zct = arg;
479 nvlist_t *nvprops = zct->zct_props;
481 uint64_t volblocksize, volsize;
483 VERIFY(nvlist_lookup_uint64(nvprops,
484 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
485 if (nvlist_lookup_uint64(nvprops,
486 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
487 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
490 * These properties must be removed from the list so the generic
491 * property setting step won't apply to them.
493 VERIFY(nvlist_remove_all(nvprops,
494 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
495 (void) nvlist_remove_all(nvprops,
496 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
498 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
502 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
506 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
511 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
512 * implement DKIOCFREE/free-long-range.
515 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
517 zvol_state_t *zv = arg1;
518 lr_truncate_t *lr = arg2;
519 uint64_t offset, length;
522 byteswap_uint64_array(lr, sizeof (*lr));
524 offset = lr->lr_offset;
525 length = lr->lr_length;
527 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
531 * Replay a TX_WRITE ZIL transaction that didn't get committed
532 * after a system failure
535 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
537 zvol_state_t *zv = arg1;
538 lr_write_t *lr = arg2;
539 objset_t *os = zv->zv_objset;
540 char *data = (char *)(lr + 1); /* data follows lr_write_t */
541 uint64_t offset, length;
546 byteswap_uint64_array(lr, sizeof (*lr));
548 offset = lr->lr_offset;
549 length = lr->lr_length;
551 /* If it's a dmu_sync() block, write the whole block */
552 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
553 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
554 if (length < blocksize) {
555 offset -= offset % blocksize;
560 tx = dmu_tx_create(os);
561 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
562 error = dmu_tx_assign(tx, TXG_WAIT);
566 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
575 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
577 return (SET_ERROR(ENOTSUP));
581 * Callback vectors for replaying records.
582 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
584 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
585 zvol_replay_err, /* 0 no such transaction type */
586 zvol_replay_err, /* TX_CREATE */
587 zvol_replay_err, /* TX_MKDIR */
588 zvol_replay_err, /* TX_MKXATTR */
589 zvol_replay_err, /* TX_SYMLINK */
590 zvol_replay_err, /* TX_REMOVE */
591 zvol_replay_err, /* TX_RMDIR */
592 zvol_replay_err, /* TX_LINK */
593 zvol_replay_err, /* TX_RENAME */
594 zvol_replay_write, /* TX_WRITE */
595 zvol_replay_truncate, /* TX_TRUNCATE */
596 zvol_replay_err, /* TX_SETATTR */
597 zvol_replay_err, /* TX_ACL */
598 zvol_replay_err, /* TX_CREATE_ACL */
599 zvol_replay_err, /* TX_CREATE_ATTR */
600 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
601 zvol_replay_err, /* TX_MKDIR_ACL */
602 zvol_replay_err, /* TX_MKDIR_ATTR */
603 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
604 zvol_replay_err, /* TX_WRITE2 */
609 zvol_name2minor(const char *name, minor_t *minor)
613 mutex_enter(&zfsdev_state_lock);
614 zv = zvol_minor_lookup(name);
616 *minor = zv->zv_minor;
617 mutex_exit(&zfsdev_state_lock);
618 return (zv ? 0 : -1);
623 * Create a minor node (plus a whole lot more) for the specified volume.
626 zvol_create_minor(const char *name)
628 zfs_soft_state_t *zs;
632 dmu_object_info_t doi;
634 char chrbuf[30], blkbuf[30];
636 struct g_provider *pp;
643 ZFS_LOG(1, "Creating ZVOL %s...", name);
646 mutex_enter(&zfsdev_state_lock);
648 if (zvol_minor_lookup(name) != NULL) {
649 mutex_exit(&zfsdev_state_lock);
650 return (SET_ERROR(EEXIST));
653 /* lie and say we're read-only */
654 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
657 mutex_exit(&zfsdev_state_lock);
662 if ((minor = zfsdev_minor_alloc()) == 0) {
663 dmu_objset_disown(os, FTAG);
664 mutex_exit(&zfsdev_state_lock);
665 return (SET_ERROR(ENXIO));
668 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
669 dmu_objset_disown(os, FTAG);
670 mutex_exit(&zfsdev_state_lock);
671 return (SET_ERROR(EAGAIN));
673 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
676 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
678 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
679 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
680 ddi_soft_state_free(zfsdev_state, minor);
681 dmu_objset_disown(os, FTAG);
682 mutex_exit(&zfsdev_state_lock);
683 return (SET_ERROR(EAGAIN));
686 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
688 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
689 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
690 ddi_remove_minor_node(zfs_dip, chrbuf);
691 ddi_soft_state_free(zfsdev_state, minor);
692 dmu_objset_disown(os, FTAG);
693 mutex_exit(&zfsdev_state_lock);
694 return (SET_ERROR(EAGAIN));
697 zs = ddi_get_soft_state(zfsdev_state, minor);
698 zs->zss_type = ZSST_ZVOL;
699 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
702 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
704 error = dsl_prop_get_integer(name,
705 zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
706 if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
709 zv->zv_volmode = mode;
710 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
712 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
713 gp->start = zvol_geom_start;
714 gp->access = zvol_geom_access;
715 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
716 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
717 pp->sectorsize = DEV_BSIZE;
721 zv->zv_provider = pp;
722 bioq_init(&zv->zv_queue);
723 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
724 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
725 struct make_dev_args args;
727 make_dev_args_init(&args);
728 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
729 args.mda_devsw = &zvol_cdevsw;
731 args.mda_uid = UID_ROOT;
732 args.mda_gid = GID_OPERATOR;
733 args.mda_mode = 0640;
734 args.mda_si_drv2 = zv;
735 error = make_dev_s(&args, &zv->zv_dev,
736 "%s/%s", ZVOL_DRIVER, name);
738 kmem_free(zv, sizeof(*zv));
739 dmu_objset_disown(os, FTAG);
740 mutex_exit(&zfsdev_state_lock);
743 zv->zv_dev->si_iosize_max = MAXPHYS;
745 LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
748 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
749 zv->zv_min_bs = DEV_BSHIFT;
751 zv->zv_minor = minor;
754 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
755 zv->zv_flags |= ZVOL_RDONLY;
756 rangelock_init(&zv->zv_rangelock, NULL, NULL);
757 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
758 offsetof(zvol_extent_t, ze_node));
760 /* get and cache the blocksize */
761 error = dmu_object_info(os, ZVOL_OBJ, &doi);
763 zv->zv_volblocksize = doi.doi_data_block_size;
766 if (spa_writeable(dmu_objset_spa(os))) {
767 if (zil_replay_disable)
768 zil_destroy(dmu_objset_zil(os), B_FALSE);
770 zil_replay(os, zv, zvol_replay_vector);
772 dmu_objset_disown(os, FTAG);
773 zv->zv_objset = NULL;
777 mutex_exit(&zfsdev_state_lock);
779 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
784 ZFS_LOG(1, "ZVOL %s created.", name);
791 * Remove minor node for the specified volume.
794 zvol_remove_zv(zvol_state_t *zv)
798 minor_t minor = zv->zv_minor;
801 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
802 if (zv->zv_total_opens != 0)
803 return (SET_ERROR(EBUSY));
806 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
807 ddi_remove_minor_node(zfs_dip, nmbuf);
809 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
810 ddi_remove_minor_node(zfs_dip, nmbuf);
812 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
814 LIST_REMOVE(zv, zv_links);
815 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
817 zvol_geom_destroy(zv);
819 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
820 if (zv->zv_dev != NULL)
821 destroy_dev(zv->zv_dev);
825 rangelock_fini(&zv->zv_rangelock);
827 kmem_free(zv, sizeof (zvol_state_t));
829 ddi_soft_state_free(zfsdev_state, minor);
836 zvol_first_open(zvol_state_t *zv)
838 dmu_object_info_t doi;
844 /* lie and say we're read-only */
845 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
851 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
854 dmu_objset_disown(os, zvol_tag);
858 /* get and cache the blocksize */
859 error = dmu_object_info(os, ZVOL_OBJ, &doi);
862 dmu_objset_disown(os, zvol_tag);
865 zv->zv_volblocksize = doi.doi_data_block_size;
867 error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
869 dmu_objset_disown(os, zvol_tag);
873 zvol_size_changed(zv, volsize);
874 zv->zv_zilog = zil_open(os, zvol_get_data);
876 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
878 if (readonly || dmu_objset_is_snapshot(os) ||
879 !spa_writeable(dmu_objset_spa(os)))
880 zv->zv_flags |= ZVOL_RDONLY;
882 zv->zv_flags &= ~ZVOL_RDONLY;
887 zvol_last_close(zvol_state_t *zv)
889 zil_close(zv->zv_zilog);
892 dnode_rele(zv->zv_dn, zvol_tag);
898 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
899 !(zv->zv_flags & ZVOL_RDONLY))
900 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
901 dmu_objset_evict_dbufs(zv->zv_objset);
903 dmu_objset_disown(zv->zv_objset, zvol_tag);
904 zv->zv_objset = NULL;
909 zvol_prealloc(zvol_state_t *zv)
911 objset_t *os = zv->zv_objset;
913 uint64_t refd, avail, usedobjs, availobjs;
914 uint64_t resid = zv->zv_volsize;
917 /* Check the space usage before attempting to allocate the space */
918 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
919 if (avail < zv->zv_volsize)
920 return (SET_ERROR(ENOSPC));
922 /* Free old extents if they exist */
923 zvol_free_extents(zv);
927 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
929 tx = dmu_tx_create(os);
930 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
931 error = dmu_tx_assign(tx, TXG_WAIT);
934 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
937 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
942 txg_wait_synced(dmu_objset_pool(os), 0);
949 zvol_update_volsize(objset_t *os, uint64_t volsize)
954 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
956 tx = dmu_tx_create(os);
957 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
958 dmu_tx_mark_netfree(tx);
959 error = dmu_tx_assign(tx, TXG_WAIT);
965 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
970 error = dmu_free_long_range(os,
971 ZVOL_OBJ, volsize, DMU_OBJECT_END);
976 zvol_remove_minors_impl(const char *name)
983 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
984 (void) strncpy(namebuf, name, strlen(name));
985 (void) strcat(namebuf, "/");
986 mutex_enter(&zfsdev_state_lock);
987 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
989 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
992 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
993 (void) zvol_remove_zv(zv);
995 kmem_free(namebuf, strlen(name) + 2);
997 mutex_exit(&zfsdev_state_lock);
999 zvol_state_t *zv, *tzv;
1002 namelen = strlen(name);
1004 mutex_enter(&zfsdev_state_lock);
1006 LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
1007 if (strcmp(zv->zv_name, name) == 0 ||
1008 (strncmp(zv->zv_name, name, namelen) == 0 &&
1009 strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' ||
1010 zv->zv_name[namelen] == '@'))) {
1011 (void) zvol_remove_zv(zv);
1015 mutex_exit(&zfsdev_state_lock);
1016 #endif /* illumos */
1020 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
1022 uint64_t old_volsize = 0ULL;
1025 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1028 * Reinitialize the dump area to the new size. If we
1029 * failed to resize the dump area then restore it back to
1030 * its original size. We must set the new volsize prior
1031 * to calling dumpvp_resize() to ensure that the devices'
1032 * size(9P) is not visible by the dump subsystem.
1034 old_volsize = zv->zv_volsize;
1035 zvol_size_changed(zv, volsize);
1038 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1039 if ((error = zvol_dumpify(zv)) != 0 ||
1040 (error = dumpvp_resize()) != 0) {
1043 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
1044 zvol_size_changed(zv, old_volsize);
1045 dumpify_error = zvol_dumpify(zv);
1046 error = dumpify_error ? dumpify_error : error;
1049 #endif /* ZVOL_DUMP */
1053 * Generate a LUN expansion event.
1058 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1060 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
1063 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1064 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
1066 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
1067 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
1070 kmem_free(physpath, MAXPATHLEN);
1072 #endif /* illumos */
1077 zvol_set_volsize(const char *name, uint64_t volsize)
1079 zvol_state_t *zv = NULL;
1082 dmu_object_info_t doi;
1084 boolean_t owned = B_FALSE;
1086 error = dsl_prop_get_integer(name,
1087 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
1091 return (SET_ERROR(EROFS));
1093 mutex_enter(&zfsdev_state_lock);
1094 zv = zvol_minor_lookup(name);
1096 if (zv == NULL || zv->zv_objset == NULL) {
1097 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
1099 mutex_exit(&zfsdev_state_lock);
1109 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
1110 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
1113 error = zvol_update_volsize(os, volsize);
1115 if (error == 0 && zv != NULL)
1116 error = zvol_update_live_volsize(zv, volsize);
1119 dmu_objset_disown(os, FTAG);
1121 zv->zv_objset = NULL;
1123 mutex_exit(&zfsdev_state_lock);
1130 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
1133 zvol_open(struct g_provider *pp, int flag, int count)
1140 mutex_enter(&zfsdev_state_lock);
1142 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
1144 mutex_exit(&zfsdev_state_lock);
1145 return (SET_ERROR(ENXIO));
1148 if (zv->zv_total_opens == 0)
1149 err = zvol_first_open(zv);
1151 mutex_exit(&zfsdev_state_lock);
1154 #else /* !illumos */
1155 boolean_t locked = B_FALSE;
1157 if (!zpool_on_zvol && tsd_get(zfs_geom_probe_vdev_key) != NULL) {
1159 * if zfs_geom_probe_vdev_key is set, that means that zfs is
1160 * attempting to probe geom providers while looking for a
1161 * replacement for a missing VDEV. In this case, the
1162 * spa_namespace_lock will not be held, but it is still illegal
1163 * to use a zvol as a vdev. Deadlocks can result if another
1164 * thread has spa_namespace_lock
1166 return (EOPNOTSUPP);
1169 * Protect against recursively entering spa_namespace_lock
1170 * when spa_open() is used for a pool on a (local) ZVOL(s).
1171 * This is needed since we replaced upstream zfsdev_state_lock
1172 * with spa_namespace_lock in the ZVOL code.
1173 * We are using the same trick as spa_open().
1174 * Note that calls in zvol_first_open which need to resolve
1175 * pool name to a spa object will enter spa_open()
1176 * recursively, but that function already has all the
1177 * necessary protection.
1179 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1180 mutex_enter(&zfsdev_state_lock);
1187 mutex_exit(&zfsdev_state_lock);
1188 return (SET_ERROR(ENXIO));
1191 if (zv->zv_total_opens == 0) {
1192 err = zvol_first_open(zv);
1195 mutex_exit(&zfsdev_state_lock);
1198 pp->mediasize = zv->zv_volsize;
1199 pp->stripeoffset = 0;
1200 pp->stripesize = zv->zv_volblocksize;
1202 #endif /* illumos */
1203 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
1204 err = SET_ERROR(EROFS);
1207 if (zv->zv_flags & ZVOL_EXCL) {
1208 err = SET_ERROR(EBUSY);
1213 if (zv->zv_total_opens != 0) {
1214 err = SET_ERROR(EBUSY);
1217 zv->zv_flags |= ZVOL_EXCL;
1222 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
1223 zv->zv_open_count[otyp]++;
1224 zv->zv_total_opens++;
1226 mutex_exit(&zfsdev_state_lock);
1228 zv->zv_total_opens += count;
1230 mutex_exit(&zfsdev_state_lock);
1235 if (zv->zv_total_opens == 0)
1236 zvol_last_close(zv);
1238 mutex_exit(&zfsdev_state_lock);
1241 mutex_exit(&zfsdev_state_lock);
1249 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
1251 minor_t minor = getminor(dev);
1255 mutex_enter(&zfsdev_state_lock);
1257 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1259 mutex_exit(&zfsdev_state_lock);
1260 #else /* !illumos */
1262 zvol_close(struct g_provider *pp, int flag, int count)
1266 boolean_t locked = B_FALSE;
1268 /* See comment in zvol_open(). */
1269 if (!MUTEX_HELD(&zfsdev_state_lock)) {
1270 mutex_enter(&zfsdev_state_lock);
1277 mutex_exit(&zfsdev_state_lock);
1278 #endif /* illumos */
1279 return (SET_ERROR(ENXIO));
1282 if (zv->zv_flags & ZVOL_EXCL) {
1283 ASSERT(zv->zv_total_opens == 1);
1284 zv->zv_flags &= ~ZVOL_EXCL;
1288 * If the open count is zero, this is a spurious close.
1289 * That indicates a bug in the kernel / DDI framework.
1292 ASSERT(zv->zv_open_count[otyp] != 0);
1294 ASSERT(zv->zv_total_opens != 0);
1297 * You may get multiple opens, but only one close.
1300 zv->zv_open_count[otyp]--;
1301 zv->zv_total_opens--;
1303 zv->zv_total_opens -= count;
1306 if (zv->zv_total_opens == 0)
1307 zvol_last_close(zv);
1310 mutex_exit(&zfsdev_state_lock);
1313 mutex_exit(&zfsdev_state_lock);
1320 zvol_get_done(zgd_t *zgd, int error)
1323 dmu_buf_rele(zgd->zgd_db, zgd);
1325 rangelock_exit(zgd->zgd_lr);
1327 kmem_free(zgd, sizeof (zgd_t));
1331 * Get data to generate a TX_WRITE intent log record.
1334 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1336 zvol_state_t *zv = arg;
1337 uint64_t offset = lr->lr_offset;
1338 uint64_t size = lr->lr_length; /* length of user data */
1343 ASSERT3P(lwb, !=, NULL);
1344 ASSERT3P(zio, !=, NULL);
1345 ASSERT3U(size, !=, 0);
1347 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1351 * Write records come in two flavors: immediate and indirect.
1352 * For small writes it's cheaper to store the data with the
1353 * log record (immediate); for large writes it's cheaper to
1354 * sync the data and get a pointer to it (indirect) so that
1355 * we don't have to write the data twice.
1357 if (buf != NULL) { /* immediate write */
1358 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1360 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1361 DMU_READ_NO_PREFETCH);
1362 } else { /* indirect write */
1364 * Have to lock the whole block to ensure when it's written out
1365 * and its checksum is being calculated that no one can change
1366 * the data. Contrarily to zfs_get_data we need not re-check
1367 * blocksize after we get the lock because it cannot be changed.
1369 size = zv->zv_volblocksize;
1370 offset = P2ALIGN(offset, size);
1371 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1373 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1374 DMU_READ_NO_PREFETCH);
1376 blkptr_t *bp = &lr->lr_blkptr;
1381 ASSERT(db->db_offset == offset);
1382 ASSERT(db->db_size == size);
1384 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1385 zvol_get_done, zgd);
1392 zvol_get_done(zgd, error);
1398 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1400 * We store data in the log buffers if it's small enough.
1401 * Otherwise we will later flush the data out via dmu_sync().
1403 ssize_t zvol_immediate_write_sz = 32768;
1405 SYSCTL_LONG(_vfs_zfs_vol, OID_AUTO, immediate_write_sz, CTLFLAG_RWTUN,
1406 &zvol_immediate_write_sz, 0, "Minimal size for indirect log write");
1410 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1413 uint32_t blocksize = zv->zv_volblocksize;
1414 zilog_t *zilog = zv->zv_zilog;
1415 itx_wr_state_t write_state;
1417 if (zil_replaying(zilog, tx))
1420 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1421 write_state = WR_INDIRECT;
1422 else if (!spa_has_slogs(zilog->zl_spa) &&
1423 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1424 write_state = WR_INDIRECT;
1426 write_state = WR_COPIED;
1428 write_state = WR_NEED_COPY;
1433 itx_wr_state_t wr_state = write_state;
1434 ssize_t len = resid;
1436 if (wr_state == WR_COPIED && resid > zil_max_copied_data(zilog))
1437 wr_state = WR_NEED_COPY;
1438 else if (wr_state == WR_INDIRECT)
1439 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1441 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1442 (wr_state == WR_COPIED ? len : 0));
1443 lr = (lr_write_t *)&itx->itx_lr;
1444 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1445 off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1446 zil_itx_destroy(itx);
1447 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1448 lr = (lr_write_t *)&itx->itx_lr;
1449 wr_state = WR_NEED_COPY;
1452 itx->itx_wr_state = wr_state;
1453 lr->lr_foid = ZVOL_OBJ;
1454 lr->lr_offset = off;
1455 lr->lr_length = len;
1457 BP_ZERO(&lr->lr_blkptr);
1459 itx->itx_private = zv;
1461 if (!sync && (zv->zv_sync_cnt == 0))
1462 itx->itx_sync = B_FALSE;
1464 zil_itx_assign(zilog, itx, tx);
1473 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1474 uint64_t size, boolean_t doread, boolean_t isdump)
1480 if (vd->vdev_ops == &vdev_mirror_ops ||
1481 vd->vdev_ops == &vdev_replacing_ops ||
1482 vd->vdev_ops == &vdev_spare_ops) {
1483 for (c = 0; c < vd->vdev_children; c++) {
1484 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1485 addr, offset, origoffset, size, doread, isdump);
1488 } else if (doread) {
1494 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1495 return (numerrors < vd->vdev_children ? 0 : EIO);
1497 if (doread && !vdev_readable(vd))
1498 return (SET_ERROR(EIO));
1499 else if (!doread && !vdev_writeable(vd))
1500 return (SET_ERROR(EIO));
1502 if (vd->vdev_ops == &vdev_raidz_ops) {
1503 return (vdev_raidz_physio(vd,
1504 addr, size, offset, origoffset, doread, isdump));
1507 offset += VDEV_LABEL_START_SIZE;
1509 if (ddi_in_panic() || isdump) {
1512 return (SET_ERROR(EIO));
1514 ASSERT3P(dvd, !=, NULL);
1515 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1519 ASSERT3P(dvd, !=, NULL);
1520 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1521 offset, doread ? B_READ : B_WRITE));
1526 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1527 boolean_t doread, boolean_t isdump)
1532 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1534 /* Must be sector aligned, and not stradle a block boundary. */
1535 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1536 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1537 return (SET_ERROR(EINVAL));
1539 ASSERT(size <= zv->zv_volblocksize);
1541 /* Locate the extent this belongs to */
1542 ze = list_head(&zv->zv_extents);
1543 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1544 offset -= ze->ze_nblks * zv->zv_volblocksize;
1545 ze = list_next(&zv->zv_extents, ze);
1549 return (SET_ERROR(EINVAL));
1551 if (!ddi_in_panic())
1552 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1554 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1555 offset += DVA_GET_OFFSET(&ze->ze_dva);
1556 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1557 size, doread, isdump);
1559 if (!ddi_in_panic())
1560 spa_config_exit(spa, SCL_STATE, FTAG);
1566 zvol_strategy(buf_t *bp)
1568 zfs_soft_state_t *zs = NULL;
1569 #else /* !illumos */
1571 zvol_strategy(struct bio *bp)
1573 #endif /* illumos */
1575 uint64_t off, volsize;
1581 boolean_t doread = bp->b_flags & B_READ;
1583 boolean_t doread = 0;
1585 boolean_t is_dumpified;
1589 if (getminor(bp->b_edev) == 0) {
1590 error = SET_ERROR(EINVAL);
1592 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1594 error = SET_ERROR(ENXIO);
1595 else if (zs->zss_type != ZSST_ZVOL)
1596 error = SET_ERROR(EINVAL);
1600 bioerror(bp, error);
1607 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1608 bioerror(bp, EROFS);
1613 off = ldbtob(bp->b_blkno);
1614 #else /* !illumos */
1616 zv = bp->bio_to->private;
1618 zv = bp->bio_dev->si_drv2;
1621 error = SET_ERROR(ENXIO);
1625 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1626 error = SET_ERROR(EROFS);
1630 switch (bp->bio_cmd) {
1643 off = bp->bio_offset;
1644 #endif /* illumos */
1645 volsize = zv->zv_volsize;
1652 addr = bp->b_un.b_addr;
1653 resid = bp->b_bcount;
1655 if (resid > 0 && (off < 0 || off >= volsize)) {
1661 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1662 sync = ((!(bp->b_flags & B_ASYNC) &&
1663 !(zv->zv_flags & ZVOL_WCE)) ||
1664 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1665 !doread && !is_dumpified;
1666 #else /* !illumos */
1667 addr = bp->bio_data;
1668 resid = bp->bio_length;
1670 if (resid > 0 && (off < 0 || off >= volsize)) {
1671 error = SET_ERROR(EIO);
1675 is_dumpified = B_FALSE;
1676 sync = !doread && !is_dumpified &&
1677 zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1678 #endif /* illumos */
1681 * There must be no buffer changes when doing a dmu_sync() because
1682 * we can't change the data whilst calculating the checksum.
1684 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock, off, resid,
1685 doread ? RL_READER : RL_WRITER);
1688 if (bp->bio_cmd == BIO_DELETE) {
1689 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1690 error = dmu_tx_assign(tx, TXG_WAIT);
1694 zvol_log_truncate(zv, tx, off, resid, sync);
1696 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1703 while (resid != 0 && off < volsize) {
1704 size_t size = MIN(resid, zvol_maxphys);
1707 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1708 error = zvol_dumpio(zv, addr, off, size,
1710 } else if (doread) {
1714 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1717 dmu_tx_t *tx = dmu_tx_create(os);
1718 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1719 error = dmu_tx_assign(tx, TXG_WAIT);
1723 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1724 zvol_log_write(zv, tx, off, size, sync);
1729 /* convert checksum errors into IO errors */
1730 if (error == ECKSUM)
1731 error = SET_ERROR(EIO);
1744 if ((bp->b_resid = resid) == bp->b_bcount)
1745 bioerror(bp, off > volsize ? EINVAL : error);
1748 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1752 #else /* !illumos */
1753 bp->bio_completed = bp->bio_length - resid;
1754 if (bp->bio_completed < bp->bio_length && off > volsize)
1759 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1763 g_io_deliver(bp, error);
1765 biofinish(bp, NULL, error);
1766 #endif /* illumos */
1771 * Set the buffer count to the zvol maximum transfer.
1772 * Using our own routine instead of the default minphys()
1773 * means that for larger writes we write bigger buffers on X86
1774 * (128K instead of 56K) and flush the disk write cache less often
1775 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1776 * 56K on X86 and 128K on sparc).
1779 zvol_minphys(struct buf *bp)
1781 if (bp->b_bcount > zvol_maxphys)
1782 bp->b_bcount = zvol_maxphys;
1786 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1788 minor_t minor = getminor(dev);
1795 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1797 return (SET_ERROR(ENXIO));
1799 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1800 return (SET_ERROR(EINVAL));
1802 boff = ldbtob(blkno);
1803 resid = ldbtob(nblocks);
1805 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1808 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1809 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1822 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1824 minor_t minor = getminor(dev);
1825 #else /* !illumos */
1827 zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
1829 #endif /* illumos */
1835 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1837 return (SET_ERROR(ENXIO));
1842 volsize = zv->zv_volsize;
1843 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1844 if (uio->uio_resid > 0 &&
1845 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1846 return (SET_ERROR(EIO));
1849 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1850 error = physio(zvol_strategy, NULL, dev, B_READ,
1856 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1857 uio->uio_loffset, uio->uio_resid, RL_READER);
1858 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1859 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1861 /* don't read past the end */
1862 if (bytes > volsize - uio->uio_loffset)
1863 bytes = volsize - uio->uio_loffset;
1865 error = dmu_read_uio_dnode(zv->zv_dn, uio, bytes);
1867 /* convert checksum errors into IO errors */
1868 if (error == ECKSUM)
1869 error = SET_ERROR(EIO);
1881 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1883 minor_t minor = getminor(dev);
1884 #else /* !illumos */
1886 zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
1888 #endif /* illumos */
1895 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1897 return (SET_ERROR(ENXIO));
1902 volsize = zv->zv_volsize;
1903 /* uio_loffset == volsize isn't an error as its required for EOF processing. */
1904 if (uio->uio_resid > 0 &&
1905 (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
1906 return (SET_ERROR(EIO));
1909 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1910 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1915 sync = !(zv->zv_flags & ZVOL_WCE) ||
1917 sync = (ioflag & IO_SYNC) ||
1919 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1921 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1922 uio->uio_loffset, uio->uio_resid, RL_WRITER);
1923 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1924 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1925 uint64_t off = uio->uio_loffset;
1926 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1928 if (bytes > volsize - off) /* don't write past the end */
1929 bytes = volsize - off;
1931 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1932 error = dmu_tx_assign(tx, TXG_WAIT);
1937 error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1939 zvol_log_write(zv, tx, off, bytes, sync);
1948 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1954 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1956 struct uuid uuid = EFI_RESERVED;
1957 efi_gpe_t gpe = { 0 };
1963 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1964 return (SET_ERROR(EFAULT));
1965 ptr = (char *)(uintptr_t)efi.dki_data_64;
1966 length = efi.dki_length;
1968 * Some clients may attempt to request a PMBR for the
1969 * zvol. Currently this interface will return EINVAL to
1970 * such requests. These requests could be supported by
1971 * adding a check for lba == 0 and consing up an appropriate
1974 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1975 return (SET_ERROR(EINVAL));
1977 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1978 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1979 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1981 if (efi.dki_lba == 1) {
1982 efi_gpt_t gpt = { 0 };
1984 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1985 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1986 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1987 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1988 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1989 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1990 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1991 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1992 gpt.efi_gpt_SizeOfPartitionEntry =
1993 LE_32(sizeof (efi_gpe_t));
1994 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1995 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1996 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1997 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1998 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
2000 return (SET_ERROR(EFAULT));
2001 ptr += sizeof (gpt);
2002 length -= sizeof (gpt);
2004 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
2006 return (SET_ERROR(EFAULT));
2011 * BEGIN entry points to allow external callers access to the volume.
2014 * Return the volume parameters needed for access from an external caller.
2015 * These values are invariant as long as the volume is held open.
2018 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
2019 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
2020 void **rl_hdl, void **dnode_hdl)
2024 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
2026 return (SET_ERROR(ENXIO));
2027 if (zv->zv_flags & ZVOL_DUMPIFIED)
2028 return (SET_ERROR(ENXIO));
2030 ASSERT(blksize && max_xfer_len && minor_hdl &&
2031 objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
2033 *blksize = zv->zv_volblocksize;
2034 *max_xfer_len = (uint64_t)zvol_maxphys;
2036 *objset_hdl = zv->zv_objset;
2037 *zil_hdl = zv->zv_zilog;
2038 *rl_hdl = &zv->zv_rangelock;
2039 *dnode_hdl = zv->zv_dn;
2044 * Return the current volume size to an external caller.
2045 * The size can change while the volume is open.
2048 zvol_get_volume_size(void *minor_hdl)
2050 zvol_state_t *zv = minor_hdl;
2052 return (zv->zv_volsize);
2056 * Return the current WCE setting to an external caller.
2057 * The WCE setting can change while the volume is open.
2060 zvol_get_volume_wce(void *minor_hdl)
2062 zvol_state_t *zv = minor_hdl;
2064 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
2068 * Entry point for external callers to zvol_log_write
2071 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
2074 zvol_state_t *zv = minor_hdl;
2076 zvol_log_write(zv, tx, off, resid, sync);
2079 * END entry points to allow external callers access to the volume.
2081 #endif /* illumos */
2084 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
2087 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
2092 zilog_t *zilog = zv->zv_zilog;
2094 if (zil_replaying(zilog, tx))
2097 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2098 lr = (lr_truncate_t *)&itx->itx_lr;
2099 lr->lr_foid = ZVOL_OBJ;
2100 lr->lr_offset = off;
2101 lr->lr_length = len;
2103 itx->itx_sync = (sync || zv->zv_sync_cnt != 0);
2104 zil_itx_assign(zilog, itx, tx);
2109 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
2110 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
2114 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
2117 struct dk_callback *dkc;
2121 mutex_enter(&zfsdev_state_lock);
2123 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
2126 mutex_exit(&zfsdev_state_lock);
2127 return (SET_ERROR(ENXIO));
2129 ASSERT(zv->zv_total_opens > 0);
2135 struct dk_cinfo dki;
2137 bzero(&dki, sizeof (dki));
2138 (void) strcpy(dki.dki_cname, "zvol");
2139 (void) strcpy(dki.dki_dname, "zvol");
2140 dki.dki_ctype = DKC_UNKNOWN;
2141 dki.dki_unit = getminor(dev);
2142 dki.dki_maxtransfer =
2143 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
2144 mutex_exit(&zfsdev_state_lock);
2145 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
2146 error = SET_ERROR(EFAULT);
2150 case DKIOCGMEDIAINFO:
2152 struct dk_minfo dkm;
2154 bzero(&dkm, sizeof (dkm));
2155 dkm.dki_lbsize = 1U << zv->zv_min_bs;
2156 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2157 dkm.dki_media_type = DK_UNKNOWN;
2158 mutex_exit(&zfsdev_state_lock);
2159 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
2160 error = SET_ERROR(EFAULT);
2164 case DKIOCGMEDIAINFOEXT:
2166 struct dk_minfo_ext dkmext;
2168 bzero(&dkmext, sizeof (dkmext));
2169 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
2170 dkmext.dki_pbsize = zv->zv_volblocksize;
2171 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
2172 dkmext.dki_media_type = DK_UNKNOWN;
2173 mutex_exit(&zfsdev_state_lock);
2174 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
2175 error = SET_ERROR(EFAULT);
2181 uint64_t vs = zv->zv_volsize;
2182 uint8_t bs = zv->zv_min_bs;
2184 mutex_exit(&zfsdev_state_lock);
2185 error = zvol_getefi((void *)arg, flag, vs, bs);
2189 case DKIOCFLUSHWRITECACHE:
2190 dkc = (struct dk_callback *)arg;
2191 mutex_exit(&zfsdev_state_lock);
2192 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2193 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
2194 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
2201 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
2202 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
2204 error = SET_ERROR(EFAULT);
2210 if (ddi_copyin((void *)arg, &wce, sizeof (int),
2212 error = SET_ERROR(EFAULT);
2216 zv->zv_flags |= ZVOL_WCE;
2217 mutex_exit(&zfsdev_state_lock);
2219 zv->zv_flags &= ~ZVOL_WCE;
2220 mutex_exit(&zfsdev_state_lock);
2221 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2229 * commands using these (like prtvtoc) expect ENOTSUP
2230 * since we're emulating an EFI label
2232 error = SET_ERROR(ENOTSUP);
2236 lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
2238 error = zvol_dumpify(zv);
2243 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
2245 lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
2247 error = zvol_dump_fini(zv);
2253 dkioc_free_list_t *dfl;
2256 if (!zvol_unmap_enabled)
2259 if (!(flag & FKIOCTL)) {
2260 error = dfl_copyin((void *)arg, &dfl, flag, KM_SLEEP);
2264 dfl = (dkioc_free_list_t *)arg;
2265 ASSERT3U(dfl->dfl_num_exts, <=, DFL_COPYIN_MAX_EXTS);
2266 if (dfl->dfl_num_exts > DFL_COPYIN_MAX_EXTS) {
2267 error = SET_ERROR(EINVAL);
2272 mutex_exit(&zfsdev_state_lock);
2274 for (int i = 0; i < dfl->dfl_num_exts; i++) {
2275 uint64_t start = dfl->dfl_exts[i].dfle_start,
2276 length = dfl->dfl_exts[i].dfle_length,
2277 end = start + length;
2280 * Apply Postel's Law to length-checking. If they
2281 * overshoot, just blank out until the end, if there's
2282 * a need to blank out anything.
2284 if (start >= zv->zv_volsize)
2285 continue; /* No need to do anything... */
2286 if (end > zv->zv_volsize) {
2287 end = DMU_OBJECT_END;
2288 length = end - start;
2291 lr = rangelock_enter(&zv->zv_rangelock, start, length,
2293 tx = dmu_tx_create(zv->zv_objset);
2294 error = dmu_tx_assign(tx, TXG_WAIT);
2298 zvol_log_truncate(zv, tx, start, length,
2301 error = dmu_free_long_range(zv->zv_objset,
2302 ZVOL_OBJ, start, length);
2312 * If the write-cache is disabled, 'sync' property
2313 * is set to 'always', or if the caller is asking for
2314 * a synchronous free, commit this operation to the zil.
2315 * This will sync any previous uncommitted writes to the
2317 * Can be overridden by the zvol_unmap_sync_enabled tunable.
2319 if ((error == 0) && zvol_unmap_sync_enabled &&
2320 (!(zv->zv_flags & ZVOL_WCE) ||
2321 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
2322 (dfl->dfl_flags & DF_WAIT_SYNC))) {
2323 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2326 if (!(flag & FKIOCTL))
2333 error = SET_ERROR(ENOTTY);
2337 mutex_exit(&zfsdev_state_lock);
2340 #endif /* illumos */
2345 return (zvol_minors != 0);
2351 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
2354 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
2356 ZFS_LOG(1, "ZVOL Initialized.");
2364 mutex_destroy(&zfsdev_state_lock);
2366 ddi_soft_state_fini(&zfsdev_state);
2367 ZFS_LOG(1, "ZVOL Deinitialized.");
2373 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
2375 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2377 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2384 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
2386 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
2388 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
2392 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
2396 objset_t *os = zv->zv_objset;
2397 spa_t *spa = dmu_objset_spa(os);
2398 vdev_t *vd = spa->spa_root_vdev;
2399 nvlist_t *nv = NULL;
2400 uint64_t version = spa_version(spa);
2401 uint64_t checksum, compress, refresrv, vbs, dedup;
2403 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2404 ASSERT(vd->vdev_ops == &vdev_root_ops);
2406 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
2410 /* wait for dmu_free_long_range to actually free the blocks */
2411 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2414 * If the pool on which the dump device is being initialized has more
2415 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
2416 * enabled. If so, bump that feature's counter to indicate that the
2417 * feature is active. We also check the vdev type to handle the
2419 * # zpool create test raidz disk1 disk2 disk3
2420 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
2421 * the raidz vdev itself has 3 children.
2423 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
2424 if (!spa_feature_is_enabled(spa,
2425 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
2426 return (SET_ERROR(ENOTSUP));
2427 (void) dsl_sync_task(spa_name(spa),
2428 zfs_mvdev_dump_feature_check,
2429 zfs_mvdev_dump_activate_feature_sync, NULL,
2430 2, ZFS_SPACE_CHECK_RESERVED);
2434 error = dsl_prop_get_integer(zv->zv_name,
2435 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
2437 error = dsl_prop_get_integer(zv->zv_name,
2438 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
2442 error = dsl_prop_get_integer(zv->zv_name,
2443 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2447 error = dsl_prop_get_integer(zv->zv_name,
2448 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
2451 if (version >= SPA_VERSION_DEDUP && error == 0) {
2452 error = dsl_prop_get_integer(zv->zv_name,
2453 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
2459 tx = dmu_tx_create(os);
2460 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2461 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2462 error = dmu_tx_assign(tx, TXG_WAIT);
2469 * If we are resizing the dump device then we only need to
2470 * update the refreservation to match the newly updated
2471 * zvolsize. Otherwise, we save off the original state of the
2472 * zvol so that we can restore them if the zvol is ever undumpified.
2475 error = zap_update(os, ZVOL_ZAP_OBJ,
2476 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2477 &zv->zv_volsize, tx);
2479 error = zap_update(os, ZVOL_ZAP_OBJ,
2480 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2483 error = zap_update(os, ZVOL_ZAP_OBJ,
2484 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2488 error = zap_update(os, ZVOL_ZAP_OBJ,
2489 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2493 error = zap_update(os, ZVOL_ZAP_OBJ,
2494 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2498 error = dmu_object_set_blocksize(
2499 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2501 if (version >= SPA_VERSION_DEDUP && error == 0) {
2502 error = zap_update(os, ZVOL_ZAP_OBJ,
2503 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2507 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2512 * We only need update the zvol's property if we are initializing
2513 * the dump area for the first time.
2515 if (error == 0 && !resize) {
2517 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2518 * function. Otherwise, use the old default -- OFF.
2520 checksum = spa_feature_is_active(spa,
2521 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2524 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2525 VERIFY(nvlist_add_uint64(nv,
2526 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2527 VERIFY(nvlist_add_uint64(nv,
2528 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2529 ZIO_COMPRESS_OFF) == 0);
2530 VERIFY(nvlist_add_uint64(nv,
2531 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2533 if (version >= SPA_VERSION_DEDUP) {
2534 VERIFY(nvlist_add_uint64(nv,
2535 zfs_prop_to_name(ZFS_PROP_DEDUP),
2536 ZIO_CHECKSUM_OFF) == 0);
2539 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2544 /* Allocate the space for the dump */
2546 error = zvol_prealloc(zv);
2551 zvol_dumpify(zvol_state_t *zv)
2554 uint64_t dumpsize = 0;
2556 objset_t *os = zv->zv_objset;
2558 if (zv->zv_flags & ZVOL_RDONLY)
2559 return (SET_ERROR(EROFS));
2561 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2562 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2563 boolean_t resize = (dumpsize > 0);
2565 if ((error = zvol_dump_init(zv, resize)) != 0) {
2566 (void) zvol_dump_fini(zv);
2572 * Build up our lba mapping.
2574 error = zvol_get_lbas(zv);
2576 (void) zvol_dump_fini(zv);
2580 tx = dmu_tx_create(os);
2581 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2582 error = dmu_tx_assign(tx, TXG_WAIT);
2585 (void) zvol_dump_fini(zv);
2589 zv->zv_flags |= ZVOL_DUMPIFIED;
2590 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2591 &zv->zv_volsize, tx);
2595 (void) zvol_dump_fini(zv);
2599 txg_wait_synced(dmu_objset_pool(os), 0);
2604 zvol_dump_fini(zvol_state_t *zv)
2607 objset_t *os = zv->zv_objset;
2610 uint64_t checksum, compress, refresrv, vbs, dedup;
2611 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2614 * Attempt to restore the zvol back to its pre-dumpified state.
2615 * This is a best-effort attempt as it's possible that not all
2616 * of these properties were initialized during the dumpify process
2617 * (i.e. error during zvol_dump_init).
2620 tx = dmu_tx_create(os);
2621 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2622 error = dmu_tx_assign(tx, TXG_WAIT);
2627 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2630 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2631 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2632 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2633 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2634 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2635 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2636 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2637 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2639 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2640 (void) nvlist_add_uint64(nv,
2641 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2642 (void) nvlist_add_uint64(nv,
2643 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2644 (void) nvlist_add_uint64(nv,
2645 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2646 if (version >= SPA_VERSION_DEDUP &&
2647 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2648 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2649 (void) nvlist_add_uint64(nv,
2650 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2652 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2656 zvol_free_extents(zv);
2657 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2658 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2659 /* wait for dmu_free_long_range to actually free the blocks */
2660 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2661 tx = dmu_tx_create(os);
2662 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2663 error = dmu_tx_assign(tx, TXG_WAIT);
2668 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2669 zv->zv_volblocksize = vbs;
2674 #else /* !illumos */
2677 zvol_geom_run(zvol_state_t *zv)
2679 struct g_provider *pp;
2681 pp = zv->zv_provider;
2682 g_error_provider(pp, 0);
2684 kproc_kthread_add(zvol_geom_worker, zv, &system_proc, NULL, 0, 0,
2685 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
2689 zvol_geom_destroy(zvol_state_t *zv)
2691 struct g_provider *pp;
2693 g_topology_assert();
2695 mtx_lock(&zv->zv_queue_mtx);
2697 wakeup_one(&zv->zv_queue);
2698 while (zv->zv_state != 2)
2699 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
2700 mtx_destroy(&zv->zv_queue_mtx);
2702 pp = zv->zv_provider;
2703 zv->zv_provider = NULL;
2705 g_wither_geom(pp->geom, ENXIO);
2709 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2711 int count, error, flags;
2713 g_topology_assert();
2716 * To make it easier we expect either open or close, but not both
2719 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2720 (acr <= 0 && acw <= 0 && ace <= 0),
2721 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2722 pp->name, acr, acw, ace));
2724 if (pp->private == NULL) {
2725 if (acr <= 0 && acw <= 0 && ace <= 0)
2731 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2732 * because GEOM already handles that and handles it a bit differently.
2733 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2734 * only one exclusive consumer, no matter if it is reader or writer.
2735 * I like better the way GEOM works so I'll leave it for GEOM to
2736 * decide what to do.
2739 count = acr + acw + ace;
2744 if (acr != 0 || ace != 0)
2749 g_topology_unlock();
2751 error = zvol_open(pp, flags, count);
2753 error = zvol_close(pp, flags, -count);
2759 zvol_geom_start(struct bio *bp)
2764 zv = bp->bio_to->private;
2766 switch (bp->bio_cmd) {
2768 if (!THREAD_CAN_SLEEP())
2770 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2771 g_io_deliver(bp, 0);
2776 if (!THREAD_CAN_SLEEP())
2781 spa_t *spa = dmu_objset_spa(zv->zv_objset);
2782 uint64_t refd, avail, usedobjs, availobjs, val;
2784 if (g_handleattr_int(bp, "GEOM::candelete", 1))
2786 if (strcmp(bp->bio_attribute, "blocksavail") == 0) {
2787 dmu_objset_space(zv->zv_objset, &refd, &avail,
2788 &usedobjs, &availobjs);
2789 if (g_handleattr_off_t(bp, "blocksavail",
2792 } else if (strcmp(bp->bio_attribute, "blocksused") == 0) {
2793 dmu_objset_space(zv->zv_objset, &refd, &avail,
2794 &usedobjs, &availobjs);
2795 if (g_handleattr_off_t(bp, "blocksused",
2798 } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) {
2799 avail = metaslab_class_get_space(spa_normal_class(spa));
2800 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
2801 if (g_handleattr_off_t(bp, "poolblocksavail",
2804 } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) {
2805 refd = metaslab_class_get_alloc(spa_normal_class(spa));
2806 if (g_handleattr_off_t(bp, "poolblocksused",
2813 g_io_deliver(bp, EOPNOTSUPP);
2819 mtx_lock(&zv->zv_queue_mtx);
2820 first = (bioq_first(&zv->zv_queue) == NULL);
2821 bioq_insert_tail(&zv->zv_queue, bp);
2822 mtx_unlock(&zv->zv_queue_mtx);
2824 wakeup_one(&zv->zv_queue);
2828 zvol_geom_worker(void *arg)
2833 thread_lock(curthread);
2834 sched_prio(curthread, PRIBIO);
2835 thread_unlock(curthread);
2839 mtx_lock(&zv->zv_queue_mtx);
2840 bp = bioq_takefirst(&zv->zv_queue);
2842 if (zv->zv_state == 1) {
2844 wakeup(&zv->zv_state);
2845 mtx_unlock(&zv->zv_queue_mtx);
2848 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2852 mtx_unlock(&zv->zv_queue_mtx);
2853 switch (bp->bio_cmd) {
2855 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2856 g_io_deliver(bp, 0);
2864 g_io_deliver(bp, EOPNOTSUPP);
2870 extern boolean_t dataset_name_hidden(const char *name);
2873 zvol_create_snapshots(objset_t *os, const char *name)
2875 uint64_t cookie, obj;
2880 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2883 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2888 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2889 if (len >= MAXPATHLEN) {
2890 dmu_objset_rele(os, FTAG);
2891 error = ENAMETOOLONG;
2895 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2896 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2897 sname + len, &obj, &cookie, NULL);
2898 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2900 if (error == ENOENT)
2905 error = zvol_create_minor(sname);
2906 if (error != 0 && error != EEXIST) {
2907 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2913 kmem_free(sname, MAXPATHLEN);
2918 zvol_create_minors_impl(const char *name)
2925 if (dataset_name_hidden(name))
2928 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2929 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2933 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2934 dsl_dataset_long_hold(os->os_dsl_dataset, FTAG);
2935 dsl_pool_rele(dmu_objset_pool(os), FTAG);
2936 error = zvol_create_minor(name);
2937 if (error == 0 || error == EEXIST) {
2938 error = zvol_create_snapshots(os, name);
2940 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2943 dsl_dataset_long_rele(os->os_dsl_dataset, FTAG);
2944 dsl_dataset_rele(os->os_dsl_dataset, FTAG);
2947 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2948 dmu_objset_rele(os, FTAG);
2952 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2953 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2954 dmu_objset_rele(os, FTAG);
2955 kmem_free(osname, MAXPATHLEN);
2958 p = osname + strlen(osname);
2959 len = MAXPATHLEN - (p - osname);
2962 /* Prefetch the datasets. */
2964 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2965 if (!dataset_name_hidden(osname))
2966 (void) dmu_objset_prefetch(osname, NULL);
2971 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2973 dmu_objset_rele(os, FTAG);
2974 (void)zvol_create_minors_impl(osname);
2975 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2976 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2982 dmu_objset_rele(os, FTAG);
2983 kmem_free(osname, MAXPATHLEN);
2988 zvol_rename_minor(zvol_state_t *zv, const char *newname)
2991 struct g_provider *pp;
2994 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
2996 if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
2998 pp = zv->zv_provider;
3003 zv->zv_provider = NULL;
3004 g_wither_provider(pp, ENXIO);
3006 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
3007 pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
3008 pp->sectorsize = DEV_BSIZE;
3009 pp->mediasize = zv->zv_volsize;
3011 zv->zv_provider = pp;
3012 g_error_provider(pp, 0);
3013 g_topology_unlock();
3014 } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
3015 struct make_dev_args args;
3017 if ((dev = zv->zv_dev) != NULL) {
3020 if (zv->zv_total_opens > 0) {
3021 zv->zv_flags &= ~ZVOL_EXCL;
3022 zv->zv_total_opens = 0;
3023 zvol_last_close(zv);
3027 make_dev_args_init(&args);
3028 args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
3029 args.mda_devsw = &zvol_cdevsw;
3031 args.mda_uid = UID_ROOT;
3032 args.mda_gid = GID_OPERATOR;
3033 args.mda_mode = 0640;
3034 args.mda_si_drv2 = zv;
3035 if (make_dev_s(&args, &zv->zv_dev,
3036 "%s/%s", ZVOL_DRIVER, newname) == 0)
3037 zv->zv_dev->si_iosize_max = MAXPHYS;
3039 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
3043 zvol_rename_minors_impl(const char *oldname, const char *newname)
3045 char name[MAXPATHLEN];
3046 struct g_provider *pp;
3048 size_t oldnamelen, newnamelen;
3051 boolean_t locked = B_FALSE;
3053 oldnamelen = strlen(oldname);
3054 newnamelen = strlen(newname);
3056 /* See comment in zvol_open(). */
3057 if (!MUTEX_HELD(&zfsdev_state_lock)) {
3058 mutex_enter(&zfsdev_state_lock);
3062 LIST_FOREACH(zv, &all_zvols, zv_links) {
3063 if (strcmp(zv->zv_name, oldname) == 0) {
3064 zvol_rename_minor(zv, newname);
3065 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
3066 (zv->zv_name[oldnamelen] == '/' ||
3067 zv->zv_name[oldnamelen] == '@')) {
3068 snprintf(name, sizeof(name), "%s%c%s", newname,
3069 zv->zv_name[oldnamelen],
3070 zv->zv_name + oldnamelen + 1);
3071 zvol_rename_minor(zv, name);
3076 mutex_exit(&zfsdev_state_lock);
3079 static zvol_task_t *
3080 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2)
3085 task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP);
3087 delim = strchr(name1, '/');
3088 strlcpy(task->pool, name1, delim ? (delim - name1 + 1) : MAXNAMELEN);
3090 strlcpy(task->name1, name1, MAXNAMELEN);
3092 strlcpy(task->name2, name2, MAXNAMELEN);
3098 zvol_task_free(zvol_task_t *task)
3100 kmem_free(task, sizeof (zvol_task_t));
3104 * The worker thread function performed asynchronously.
3107 zvol_task_cb(void *param)
3109 zvol_task_t *task = (zvol_task_t *)param;
3112 case ZVOL_ASYNC_CREATE_MINORS:
3113 (void) zvol_create_minors_impl(task->name1);
3115 case ZVOL_ASYNC_REMOVE_MINORS:
3116 zvol_remove_minors_impl(task->name1);
3118 case ZVOL_ASYNC_RENAME_MINORS:
3119 zvol_rename_minors_impl(task->name1, task->name2);
3126 zvol_task_free(task);
3130 zvol_minors_helper(spa_t *spa, zvol_async_op_t op, const char *name1,
3135 if (dataset_name_hidden(name1))
3137 if (name2 != NULL && dataset_name_hidden(name2))
3139 task = zvol_task_alloc(op, name1, name2);
3140 (void)taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
3144 zvol_create_minors(spa_t *spa, const char *name)
3146 zvol_minors_helper(spa, ZVOL_ASYNC_CREATE_MINORS, name, NULL);
3150 zvol_remove_minors(spa_t *spa, const char *name)
3152 zvol_minors_helper(spa, ZVOL_ASYNC_REMOVE_MINORS, name, NULL);
3156 zvol_rename_minors(spa_t *spa, const char *oldname, const char *newname)
3158 zvol_minors_helper(spa, ZVOL_ASYNC_RENAME_MINORS, oldname, newname);
3162 zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
3164 zvol_state_t *zv = dev->si_drv2;
3167 mutex_enter(&zfsdev_state_lock);
3168 if (zv->zv_total_opens == 0)
3169 err = zvol_first_open(zv);
3171 mutex_exit(&zfsdev_state_lock);
3174 if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
3175 err = SET_ERROR(EROFS);
3178 if (zv->zv_flags & ZVOL_EXCL) {
3179 err = SET_ERROR(EBUSY);
3183 if (flags & FEXCL) {
3184 if (zv->zv_total_opens != 0) {
3185 err = SET_ERROR(EBUSY);
3188 zv->zv_flags |= ZVOL_EXCL;
3192 zv->zv_total_opens++;
3193 if (flags & (FSYNC | FDSYNC)) {
3195 if (zv->zv_sync_cnt == 1)
3196 zil_async_to_sync(zv->zv_zilog, ZVOL_OBJ);
3198 mutex_exit(&zfsdev_state_lock);
3201 if (zv->zv_total_opens == 0)
3202 zvol_last_close(zv);
3203 mutex_exit(&zfsdev_state_lock);
3208 zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
3210 zvol_state_t *zv = dev->si_drv2;
3212 mutex_enter(&zfsdev_state_lock);
3213 if (zv->zv_flags & ZVOL_EXCL) {
3214 ASSERT(zv->zv_total_opens == 1);
3215 zv->zv_flags &= ~ZVOL_EXCL;
3219 * If the open count is zero, this is a spurious close.
3220 * That indicates a bug in the kernel / DDI framework.
3222 ASSERT(zv->zv_total_opens != 0);
3225 * You may get multiple opens, but only one close.
3227 zv->zv_total_opens--;
3228 if (flags & (FSYNC | FDSYNC))
3231 if (zv->zv_total_opens == 0)
3232 zvol_last_close(zv);
3234 mutex_exit(&zfsdev_state_lock);
3239 zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
3243 off_t offset, length;
3250 KASSERT(zv->zv_total_opens > 0,
3251 ("Device with zero access count in zvol_d_ioctl"));
3253 i = IOCPARM_LEN(cmd);
3255 case DIOCGSECTORSIZE:
3256 *(u_int *)data = DEV_BSIZE;
3258 case DIOCGMEDIASIZE:
3259 *(off_t *)data = zv->zv_volsize;
3262 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3265 if (!zvol_unmap_enabled)
3268 offset = ((off_t *)data)[0];
3269 length = ((off_t *)data)[1];
3270 if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
3271 offset < 0 || offset >= zv->zv_volsize ||
3273 printf("%s: offset=%jd length=%jd\n", __func__, offset,
3279 lr = rangelock_enter(&zv->zv_rangelock, offset, length,
3281 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
3282 error = dmu_tx_assign(tx, TXG_WAIT);
3287 sync = (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
3288 zvol_log_truncate(zv, tx, offset, length, sync);
3290 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
3295 zil_commit(zv->zv_zilog, ZVOL_OBJ);
3297 case DIOCGSTRIPESIZE:
3298 *(off_t *)data = zv->zv_volblocksize;
3300 case DIOCGSTRIPEOFFSET:
3304 spa_t *spa = dmu_objset_spa(zv->zv_objset);
3305 struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
3306 uint64_t refd, avail, usedobjs, availobjs;
3308 if (strcmp(arg->name, "GEOM::candelete") == 0)
3310 else if (strcmp(arg->name, "blocksavail") == 0) {
3311 dmu_objset_space(zv->zv_objset, &refd, &avail,
3312 &usedobjs, &availobjs);
3313 arg->value.off = avail / DEV_BSIZE;
3314 } else if (strcmp(arg->name, "blocksused") == 0) {
3315 dmu_objset_space(zv->zv_objset, &refd, &avail,
3316 &usedobjs, &availobjs);
3317 arg->value.off = refd / DEV_BSIZE;
3318 } else if (strcmp(arg->name, "poolblocksavail") == 0) {
3319 avail = metaslab_class_get_space(spa_normal_class(spa));
3320 avail -= metaslab_class_get_alloc(spa_normal_class(spa));
3321 arg->value.off = avail / DEV_BSIZE;
3322 } else if (strcmp(arg->name, "poolblocksused") == 0) {
3323 refd = metaslab_class_get_alloc(spa_normal_class(spa));
3324 arg->value.off = refd / DEV_BSIZE;
3331 off_t *off = (off_t *)data;
3335 hole = (cmd == FIOSEEKHOLE);
3337 error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff);
3347 #endif /* illumos */