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.
28 /* Portions Copyright 2010 Robert Milkowski */
29 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
32 * ZFS volume emulation driver.
34 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
35 * Volumes are accessed through the symbolic links named:
37 * /dev/zvol/dsk/<pool_name>/<dataset_name>
38 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
40 * These links are created by the /dev filesystem (sdev_zvolops.c).
41 * Volumes are persistent through reboot. No user command needs to be
42 * run before opening and using a device.
45 * On FreeBSD ZVOLs are simply GEOM providers like any other storage device
49 #include <sys/types.h>
50 #include <sys/param.h>
51 #include <sys/kernel.h>
52 #include <sys/errno.h>
58 #include <sys/cmn_err.h>
63 #include <sys/dmu_traverse.h>
64 #include <sys/dnode.h>
65 #include <sys/dsl_dataset.h>
66 #include <sys/dsl_prop.h>
68 #include <sys/byteorder.h>
69 #include <sys/sunddi.h>
70 #include <sys/dirent.h>
71 #include <sys/policy.h>
72 #include <sys/fs/zfs.h>
73 #include <sys/zfs_ioctl.h>
75 #include <sys/refcount.h>
76 #include <sys/zfs_znode.h>
77 #include <sys/zfs_rlock.h>
78 #include <sys/vdev_impl.h>
80 #include <sys/zil_impl.h>
81 #include <geom/geom.h>
83 #include "zfs_namecheck.h"
85 struct g_class zfs_zvol_class = {
90 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
93 static char *zvol_tag = "zvol_tag";
95 #define ZVOL_DUMPSIZE "dumpsize"
98 * The spa_namespace_lock protects the zfsdev_state structure from being
99 * modified while it's being used, e.g. an open that comes in before a
100 * create finishes. It also protects temporary opens of the dataset so that,
101 * e.g., an open doesn't get a spurious EBUSY.
103 static uint32_t zvol_minors;
105 typedef struct zvol_extent {
107 dva_t ze_dva; /* dva associated with this extent */
108 uint64_t ze_nblks; /* number of blocks in extent */
112 * The in-core state of each volume.
114 typedef struct zvol_state {
115 char zv_name[MAXPATHLEN]; /* pool/dd name */
116 uint64_t zv_volsize; /* amount of space we advertise */
117 uint64_t zv_volblocksize; /* volume block size */
118 struct g_provider *zv_provider; /* GEOM provider */
119 uint8_t zv_min_bs; /* minimum addressable block shift */
120 uint8_t zv_flags; /* readonly, dumpified, etc. */
121 objset_t *zv_objset; /* objset handle */
122 uint32_t zv_total_opens; /* total open count */
123 zilog_t *zv_zilog; /* ZIL handle */
124 list_t zv_extents; /* List of extents for dump */
125 znode_t zv_znode; /* for range locking */
126 dmu_buf_t *zv_dbuf; /* bonus handle */
128 struct bio_queue_head zv_queue;
129 struct mtx zv_queue_mtx; /* zv_queue mutex */
133 * zvol specific flags
135 #define ZVOL_RDONLY 0x1
136 #define ZVOL_DUMPIFIED 0x2
137 #define ZVOL_EXCL 0x4
141 * zvol maximum transfer in one DMU tx.
143 int zvol_maxphys = DMU_MAX_ACCESS/2;
145 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
146 nvlist_t *, nvlist_t **);
147 static int zvol_remove_zv(zvol_state_t *);
148 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
149 static int zvol_dumpify(zvol_state_t *zv);
150 static int zvol_dump_fini(zvol_state_t *zv);
151 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
153 static zvol_state_t *zvol_geom_create(const char *name);
154 static void zvol_geom_run(zvol_state_t *zv);
155 static void zvol_geom_destroy(zvol_state_t *zv);
156 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
157 static void zvol_geom_start(struct bio *bp);
158 static void zvol_geom_worker(void *arg);
161 zvol_size_changed(zvol_state_t *zv)
164 dev_t dev = makedevice(maj, min);
166 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
167 "Size", volsize) == DDI_SUCCESS);
168 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
169 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
171 /* Notify specfs to invalidate the cached size */
172 spec_size_invalidate(dev, VBLK);
173 spec_size_invalidate(dev, VCHR);
175 struct g_provider *pp;
177 pp = zv->zv_provider;
180 if (zv->zv_volsize == pp->mediasize)
183 * Changing provider size is not really supported by GEOM, but it
184 * should be safe when provider is closed.
186 if (zv->zv_total_opens > 0)
188 pp->mediasize = zv->zv_volsize;
193 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
198 if (volsize % blocksize != 0)
202 if (volsize - 1 > SPEC_MAXOFFSET_T)
209 zvol_check_volblocksize(uint64_t volblocksize)
211 if (volblocksize < SPA_MINBLOCKSIZE ||
212 volblocksize > SPA_MAXBLOCKSIZE ||
220 zvol_get_stats(objset_t *os, nvlist_t *nv)
223 dmu_object_info_t doi;
226 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
230 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
232 error = dmu_object_info(os, ZVOL_OBJ, &doi);
235 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
236 doi.doi_data_block_size);
242 static zvol_state_t *
243 zvol_minor_lookup(const char *name)
245 struct g_provider *pp;
247 zvol_state_t *zv = NULL;
249 ASSERT(MUTEX_HELD(&spa_namespace_lock));
252 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) {
253 pp = LIST_FIRST(&gp->provider);
259 if (strcmp(zv->zv_name, name) == 0)
264 return (gp != NULL ? zv : NULL);
267 /* extent mapping arg */
275 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
276 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
278 struct maparg *ma = arg;
280 int bs = ma->ma_zv->zv_volblocksize;
282 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
285 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
288 /* Abort immediately if we have encountered gang blocks */
293 * See if the block is at the end of the previous extent.
295 ze = list_tail(&ma->ma_zv->zv_extents);
297 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
298 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
299 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
304 dprintf_bp(bp, "%s", "next blkptr:");
306 /* start a new extent */
307 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
308 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
310 list_insert_tail(&ma->ma_zv->zv_extents, ze);
315 zvol_free_extents(zvol_state_t *zv)
319 while (ze = list_head(&zv->zv_extents)) {
320 list_remove(&zv->zv_extents, ze);
321 kmem_free(ze, sizeof (zvol_extent_t));
326 zvol_get_lbas(zvol_state_t *zv)
328 objset_t *os = zv->zv_objset;
334 zvol_free_extents(zv);
336 /* commit any in-flight changes before traversing the dataset */
337 txg_wait_synced(dmu_objset_pool(os), 0);
338 err = traverse_dataset(dmu_objset_ds(os), 0,
339 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
340 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
341 zvol_free_extents(zv);
342 return (err ? err : EIO);
350 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
352 zfs_creat_t *zct = arg;
353 nvlist_t *nvprops = zct->zct_props;
355 uint64_t volblocksize, volsize;
357 VERIFY(nvlist_lookup_uint64(nvprops,
358 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
359 if (nvlist_lookup_uint64(nvprops,
360 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
361 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
364 * These properties must be removed from the list so the generic
365 * property setting step won't apply to them.
367 VERIFY(nvlist_remove_all(nvprops,
368 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
369 (void) nvlist_remove_all(nvprops,
370 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
372 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
376 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
380 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
385 * Replay a TX_WRITE ZIL transaction that didn't get committed
386 * after a system failure
389 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
391 objset_t *os = zv->zv_objset;
392 char *data = (char *)(lr + 1); /* data follows lr_write_t */
393 uint64_t offset, length;
398 byteswap_uint64_array(lr, sizeof (*lr));
400 offset = lr->lr_offset;
401 length = lr->lr_length;
403 /* If it's a dmu_sync() block, write the whole block */
404 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
405 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
406 if (length < blocksize) {
407 offset -= offset % blocksize;
412 tx = dmu_tx_create(os);
413 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
414 error = dmu_tx_assign(tx, TXG_WAIT);
418 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
427 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
433 * Callback vectors for replaying records.
434 * Only TX_WRITE is needed for zvol.
436 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
437 zvol_replay_err, /* 0 no such transaction type */
438 zvol_replay_err, /* TX_CREATE */
439 zvol_replay_err, /* TX_MKDIR */
440 zvol_replay_err, /* TX_MKXATTR */
441 zvol_replay_err, /* TX_SYMLINK */
442 zvol_replay_err, /* TX_REMOVE */
443 zvol_replay_err, /* TX_RMDIR */
444 zvol_replay_err, /* TX_LINK */
445 zvol_replay_err, /* TX_RENAME */
446 zvol_replay_write, /* TX_WRITE */
447 zvol_replay_err, /* TX_TRUNCATE */
448 zvol_replay_err, /* TX_SETATTR */
449 zvol_replay_err, /* TX_ACL */
450 zvol_replay_err, /* TX_CREATE_ACL */
451 zvol_replay_err, /* TX_CREATE_ATTR */
452 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
453 zvol_replay_err, /* TX_MKDIR_ACL */
454 zvol_replay_err, /* TX_MKDIR_ATTR */
455 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
456 zvol_replay_err, /* TX_WRITE2 */
461 zvol_name2minor(const char *name, minor_t *minor)
465 mutex_enter(&spa_namespace_lock);
466 zv = zvol_minor_lookup(name);
468 *minor = zv->zv_minor;
469 mutex_exit(&spa_namespace_lock);
470 return (zv ? 0 : -1);
475 * Create a minor node (plus a whole lot more) for the specified volume.
478 zvol_create_minor(const char *name)
480 zfs_soft_state_t *zs;
483 dmu_object_info_t doi;
486 ZFS_LOG(1, "Creating ZVOL %s...", name);
488 mutex_enter(&spa_namespace_lock);
490 if (zvol_minor_lookup(name) != NULL) {
491 mutex_exit(&spa_namespace_lock);
495 /* lie and say we're read-only */
496 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
499 mutex_exit(&spa_namespace_lock);
504 if ((minor = zfsdev_minor_alloc()) == 0) {
505 dmu_objset_disown(os, FTAG);
506 mutex_exit(&spa_namespace_lock);
510 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
511 dmu_objset_disown(os, FTAG);
512 mutex_exit(&spa_namespace_lock);
515 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
518 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
520 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
521 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
522 ddi_soft_state_free(zfsdev_state, minor);
523 dmu_objset_disown(os, FTAG);
524 mutex_exit(&spa_namespace_lock);
528 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
530 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
531 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
532 ddi_remove_minor_node(zfs_dip, chrbuf);
533 ddi_soft_state_free(zfsdev_state, minor);
534 dmu_objset_disown(os, FTAG);
535 mutex_exit(&spa_namespace_lock);
539 zs = ddi_get_soft_state(zfsdev_state, minor);
540 zs->zss_type = ZSST_ZVOL;
541 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
546 zv = zvol_geom_create(name);
549 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
550 zv->zv_min_bs = DEV_BSHIFT;
552 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
553 zv->zv_flags |= ZVOL_RDONLY;
554 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
555 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
556 sizeof (rl_t), offsetof(rl_t, r_node));
557 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
558 offsetof(zvol_extent_t, ze_node));
559 /* get and cache the blocksize */
560 error = dmu_object_info(os, ZVOL_OBJ, &doi);
562 zv->zv_volblocksize = doi.doi_data_block_size;
564 if (spa_writeable(dmu_objset_spa(os))) {
565 if (zil_replay_disable)
566 zil_destroy(dmu_objset_zil(os), B_FALSE);
568 zil_replay(os, zv, zvol_replay_vector);
570 dmu_objset_disown(os, FTAG);
571 zv->zv_objset = NULL;
575 mutex_exit(&spa_namespace_lock);
582 ZFS_LOG(1, "ZVOL %s created.", name);
588 * Remove minor node for the specified volume.
591 zvol_remove_zv(zvol_state_t *zv)
594 minor_t minor = zv->zv_minor;
597 ASSERT(MUTEX_HELD(&spa_namespace_lock));
598 if (zv->zv_total_opens != 0)
601 ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name);
604 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
605 ddi_remove_minor_node(zfs_dip, nmbuf);
608 avl_destroy(&zv->zv_znode.z_range_avl);
609 mutex_destroy(&zv->zv_znode.z_range_lock);
611 zvol_geom_destroy(zv);
618 zvol_remove_minor(const char *name)
623 mutex_enter(&spa_namespace_lock);
624 if ((zv = zvol_minor_lookup(name)) == NULL) {
625 mutex_exit(&spa_namespace_lock);
629 rc = zvol_remove_zv(zv);
631 mutex_exit(&spa_namespace_lock);
636 zvol_first_open(zvol_state_t *zv)
643 /* lie and say we're read-only */
644 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
649 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
652 dmu_objset_disown(os, zvol_tag);
656 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
658 dmu_objset_disown(os, zvol_tag);
661 zv->zv_volsize = volsize;
662 zv->zv_zilog = zil_open(os, zvol_get_data);
663 zvol_size_changed(zv);
665 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
667 if (readonly || dmu_objset_is_snapshot(os) ||
668 !spa_writeable(dmu_objset_spa(os)))
669 zv->zv_flags |= ZVOL_RDONLY;
671 zv->zv_flags &= ~ZVOL_RDONLY;
676 zvol_last_close(zvol_state_t *zv)
678 zil_close(zv->zv_zilog);
680 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
682 dmu_objset_disown(zv->zv_objset, zvol_tag);
683 zv->zv_objset = NULL;
688 zvol_prealloc(zvol_state_t *zv)
690 objset_t *os = zv->zv_objset;
692 uint64_t refd, avail, usedobjs, availobjs;
693 uint64_t resid = zv->zv_volsize;
696 /* Check the space usage before attempting to allocate the space */
697 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
698 if (avail < zv->zv_volsize)
701 /* Free old extents if they exist */
702 zvol_free_extents(zv);
706 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
708 tx = dmu_tx_create(os);
709 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
710 error = dmu_tx_assign(tx, TXG_WAIT);
713 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
716 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
721 txg_wait_synced(dmu_objset_pool(os), 0);
728 zvol_update_volsize(objset_t *os, uint64_t volsize)
733 ASSERT(MUTEX_HELD(&spa_namespace_lock));
735 tx = dmu_tx_create(os);
736 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
737 error = dmu_tx_assign(tx, TXG_WAIT);
743 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
748 error = dmu_free_long_range(os,
749 ZVOL_OBJ, volsize, DMU_OBJECT_END);
754 zvol_remove_minors(const char *name)
756 struct g_geom *gp, *gptmp;
757 struct g_provider *pp;
761 namelen = strlen(name);
764 mutex_enter(&spa_namespace_lock);
767 LIST_FOREACH_SAFE(gp, &zfs_zvol_class.geom, geom, gptmp) {
768 pp = LIST_FIRST(&gp->provider);
774 if (strcmp(zv->zv_name, name) == 0 ||
775 (strncmp(zv->zv_name, name, namelen) == 0 &&
776 zv->zv_name[namelen] == '/')) {
777 (void) zvol_remove_zv(zv);
782 mutex_exit(&spa_namespace_lock);
787 zvol_set_volsize(const char *name, major_t maj, uint64_t volsize)
789 zvol_state_t *zv = NULL;
792 dmu_object_info_t doi;
793 uint64_t old_volsize = 0ULL;
796 mutex_enter(&spa_namespace_lock);
797 zv = zvol_minor_lookup(name);
798 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
799 mutex_exit(&spa_namespace_lock);
803 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
804 (error = zvol_check_volsize(volsize,
805 doi.doi_data_block_size)) != 0)
808 VERIFY(dsl_prop_get_integer(name, "readonly", &readonly,
815 error = zvol_update_volsize(os, volsize);
817 * Reinitialize the dump area to the new size. If we
818 * failed to resize the dump area then restore it back to
821 if (zv && error == 0) {
823 if (zv->zv_flags & ZVOL_DUMPIFIED) {
824 old_volsize = zv->zv_volsize;
825 zv->zv_volsize = volsize;
826 if ((error = zvol_dumpify(zv)) != 0 ||
827 (error = dumpvp_resize()) != 0) {
828 (void) zvol_update_volsize(os, old_volsize);
829 zv->zv_volsize = old_volsize;
830 error = zvol_dumpify(zv);
833 #endif /* ZVOL_DUMP */
835 zv->zv_volsize = volsize;
836 zvol_size_changed(zv);
842 * Generate a LUN expansion event.
844 if (zv && error == 0) {
847 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
849 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
852 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
853 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
855 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
856 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
859 kmem_free(physpath, MAXPATHLEN);
864 dmu_objset_rele(os, FTAG);
866 mutex_exit(&spa_namespace_lock);
873 zvol_open(struct g_provider *pp, int flag, int count)
878 mutex_enter(&spa_namespace_lock);
882 mutex_exit(&spa_namespace_lock);
886 if (zv->zv_total_opens == 0)
887 err = zvol_first_open(zv);
889 mutex_exit(&spa_namespace_lock);
892 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
896 if (zv->zv_flags & ZVOL_EXCL) {
902 if (zv->zv_total_opens != 0) {
906 zv->zv_flags |= ZVOL_EXCL;
910 zv->zv_total_opens += count;
911 mutex_exit(&spa_namespace_lock);
915 if (zv->zv_total_opens == 0)
917 mutex_exit(&spa_namespace_lock);
923 zvol_close(struct g_provider *pp, int flag, int count)
928 mutex_enter(&spa_namespace_lock);
932 mutex_exit(&spa_namespace_lock);
936 if (zv->zv_flags & ZVOL_EXCL) {
937 ASSERT(zv->zv_total_opens == 1);
938 zv->zv_flags &= ~ZVOL_EXCL;
942 * If the open count is zero, this is a spurious close.
943 * That indicates a bug in the kernel / DDI framework.
945 ASSERT(zv->zv_total_opens != 0);
948 * You may get multiple opens, but only one close.
950 zv->zv_total_opens -= count;
952 if (zv->zv_total_opens == 0)
955 mutex_exit(&spa_namespace_lock);
960 zvol_get_done(zgd_t *zgd, int error)
963 dmu_buf_rele(zgd->zgd_db, zgd);
965 zfs_range_unlock(zgd->zgd_rl);
967 if (error == 0 && zgd->zgd_bp)
968 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
970 kmem_free(zgd, sizeof (zgd_t));
974 * Get data to generate a TX_WRITE intent log record.
977 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
979 zvol_state_t *zv = arg;
980 objset_t *os = zv->zv_objset;
981 uint64_t object = ZVOL_OBJ;
982 uint64_t offset = lr->lr_offset;
983 uint64_t size = lr->lr_length; /* length of user data */
984 blkptr_t *bp = &lr->lr_blkptr;
992 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
993 zgd->zgd_zilog = zv->zv_zilog;
994 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
997 * Write records come in two flavors: immediate and indirect.
998 * For small writes it's cheaper to store the data with the
999 * log record (immediate); for large writes it's cheaper to
1000 * sync the data and get a pointer to it (indirect) so that
1001 * we don't have to write the data twice.
1003 if (buf != NULL) { /* immediate write */
1004 error = dmu_read(os, object, offset, size, buf,
1005 DMU_READ_NO_PREFETCH);
1007 size = zv->zv_volblocksize;
1008 offset = P2ALIGN(offset, size);
1009 error = dmu_buf_hold(os, object, offset, zgd, &db,
1010 DMU_READ_NO_PREFETCH);
1015 ASSERT(db->db_offset == offset);
1016 ASSERT(db->db_size == size);
1018 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1019 zvol_get_done, zgd);
1026 zvol_get_done(zgd, error);
1032 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1034 * We store data in the log buffers if it's small enough.
1035 * Otherwise we will later flush the data out via dmu_sync().
1037 ssize_t zvol_immediate_write_sz = 32768;
1040 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1043 uint32_t blocksize = zv->zv_volblocksize;
1044 zilog_t *zilog = zv->zv_zilog;
1046 ssize_t immediate_write_sz;
1048 if (zil_replaying(zilog, tx))
1051 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1052 ? 0 : zvol_immediate_write_sz;
1054 slogging = spa_has_slogs(zilog->zl_spa) &&
1055 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1061 itx_wr_state_t write_state;
1064 * Unlike zfs_log_write() we can be called with
1065 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1067 if (blocksize > immediate_write_sz && !slogging &&
1068 resid >= blocksize && off % blocksize == 0) {
1069 write_state = WR_INDIRECT; /* uses dmu_sync */
1072 write_state = WR_COPIED;
1073 len = MIN(ZIL_MAX_LOG_DATA, resid);
1075 write_state = WR_NEED_COPY;
1076 len = MIN(ZIL_MAX_LOG_DATA, resid);
1079 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1080 (write_state == WR_COPIED ? len : 0));
1081 lr = (lr_write_t *)&itx->itx_lr;
1082 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1083 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1084 zil_itx_destroy(itx);
1085 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1086 lr = (lr_write_t *)&itx->itx_lr;
1087 write_state = WR_NEED_COPY;
1090 itx->itx_wr_state = write_state;
1091 if (write_state == WR_NEED_COPY)
1092 itx->itx_sod += len;
1093 lr->lr_foid = ZVOL_OBJ;
1094 lr->lr_offset = off;
1095 lr->lr_length = len;
1097 BP_ZERO(&lr->lr_blkptr);
1099 itx->itx_private = zv;
1100 itx->itx_sync = sync;
1102 zil_itx_assign(zilog, itx, tx);
1111 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t size,
1112 boolean_t doread, boolean_t isdump)
1118 for (c = 0; c < vd->vdev_children; c++) {
1119 ASSERT(vd->vdev_ops == &vdev_mirror_ops ||
1120 vd->vdev_ops == &vdev_replacing_ops ||
1121 vd->vdev_ops == &vdev_spare_ops);
1122 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1123 addr, offset, size, doread, isdump);
1126 } else if (doread) {
1131 if (!vd->vdev_ops->vdev_op_leaf)
1132 return (numerrors < vd->vdev_children ? 0 : EIO);
1134 if (doread && !vdev_readable(vd))
1136 else if (!doread && !vdev_writeable(vd))
1140 ASSERT3P(dvd, !=, NULL);
1141 offset += VDEV_LABEL_START_SIZE;
1143 if (ddi_in_panic() || isdump) {
1147 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1150 return (vdev_disk_physio(dvd->vd_lh, addr, size, offset,
1151 doread ? B_READ : B_WRITE));
1156 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1157 boolean_t doread, boolean_t isdump)
1162 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1164 /* Must be sector aligned, and not stradle a block boundary. */
1165 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1166 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1169 ASSERT(size <= zv->zv_volblocksize);
1171 /* Locate the extent this belongs to */
1172 ze = list_head(&zv->zv_extents);
1173 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1174 offset -= ze->ze_nblks * zv->zv_volblocksize;
1175 ze = list_next(&zv->zv_extents, ze);
1178 if (!ddi_in_panic())
1179 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1181 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1182 offset += DVA_GET_OFFSET(&ze->ze_dva);
1183 error = zvol_dumpio_vdev(vd, addr, offset, size, doread, isdump);
1185 if (!ddi_in_panic())
1186 spa_config_exit(spa, SCL_STATE, FTAG);
1193 zvol_strategy(struct bio *bp)
1195 zvol_state_t *zv = bp->bio_to->private;
1196 uint64_t off, volsize;
1202 boolean_t doread = (bp->bio_cmd == BIO_READ);
1206 g_io_deliver(bp, ENXIO);
1210 if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
1211 g_io_deliver(bp, EROFS);
1215 off = bp->bio_offset;
1216 volsize = zv->zv_volsize;
1221 addr = bp->bio_data;
1222 resid = bp->bio_length;
1224 if (resid > 0 && (off < 0 || off >= volsize)) {
1225 g_io_deliver(bp, EIO);
1229 sync = !doread && zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
1232 * There must be no buffer changes when doing a dmu_sync() because
1233 * we can't change the data whilst calculating the checksum.
1235 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1236 doread ? RL_READER : RL_WRITER);
1238 while (resid != 0 && off < volsize) {
1239 size_t size = MIN(resid, zvol_maxphys);
1241 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1244 dmu_tx_t *tx = dmu_tx_create(os);
1245 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1246 error = dmu_tx_assign(tx, TXG_WAIT);
1250 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1251 zvol_log_write(zv, tx, off, size, sync);
1256 /* convert checksum errors into IO errors */
1257 if (error == ECKSUM)
1265 zfs_range_unlock(rl);
1267 bp->bio_completed = bp->bio_length - resid;
1268 if (bp->bio_completed < bp->bio_length)
1269 bp->bio_error = (off > volsize ? EINVAL : error);
1272 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1273 g_io_deliver(bp, 0);
1280 * Set the buffer count to the zvol maximum transfer.
1281 * Using our own routine instead of the default minphys()
1282 * means that for larger writes we write bigger buffers on X86
1283 * (128K instead of 56K) and flush the disk write cache less often
1284 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1285 * 56K on X86 and 128K on sparc).
1288 zvol_minphys(struct buf *bp)
1290 if (bp->b_bcount > zvol_maxphys)
1291 bp->b_bcount = zvol_maxphys;
1295 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1297 minor_t minor = getminor(dev);
1304 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1308 boff = ldbtob(blkno);
1309 resid = ldbtob(nblocks);
1311 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1314 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1315 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1328 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1330 minor_t minor = getminor(dev);
1336 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1340 volsize = zv->zv_volsize;
1341 if (uio->uio_resid > 0 &&
1342 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1345 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1346 error = physio(zvol_strategy, NULL, dev, B_READ,
1351 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1353 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1354 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1356 /* don't read past the end */
1357 if (bytes > volsize - uio->uio_loffset)
1358 bytes = volsize - uio->uio_loffset;
1360 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1362 /* convert checksum errors into IO errors */
1363 if (error == ECKSUM)
1368 zfs_range_unlock(rl);
1374 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1376 minor_t minor = getminor(dev);
1383 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1387 volsize = zv->zv_volsize;
1388 if (uio->uio_resid > 0 &&
1389 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1392 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1393 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1398 sync = !(zv->zv_flags & ZVOL_WCE) ||
1399 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1401 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1403 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1404 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1405 uint64_t off = uio->uio_loffset;
1406 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1408 if (bytes > volsize - off) /* don't write past the end */
1409 bytes = volsize - off;
1411 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1412 error = dmu_tx_assign(tx, TXG_WAIT);
1417 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1419 zvol_log_write(zv, tx, off, bytes, sync);
1425 zfs_range_unlock(rl);
1427 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1432 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1434 struct uuid uuid = EFI_RESERVED;
1435 efi_gpe_t gpe = { 0 };
1441 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1443 ptr = (char *)(uintptr_t)efi.dki_data_64;
1444 length = efi.dki_length;
1446 * Some clients may attempt to request a PMBR for the
1447 * zvol. Currently this interface will return EINVAL to
1448 * such requests. These requests could be supported by
1449 * adding a check for lba == 0 and consing up an appropriate
1452 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1455 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1456 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1457 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1459 if (efi.dki_lba == 1) {
1460 efi_gpt_t gpt = { 0 };
1462 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1463 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1464 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1465 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1466 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1467 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1468 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1469 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1470 gpt.efi_gpt_SizeOfPartitionEntry =
1471 LE_32(sizeof (efi_gpe_t));
1472 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1473 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1474 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1475 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1476 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1479 ptr += sizeof (gpt);
1480 length -= sizeof (gpt);
1482 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1489 * BEGIN entry points to allow external callers access to the volume.
1492 * Return the volume parameters needed for access from an external caller.
1493 * These values are invariant as long as the volume is held open.
1496 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1497 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1498 void **rl_hdl, void **bonus_hdl)
1502 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1505 if (zv->zv_flags & ZVOL_DUMPIFIED)
1508 ASSERT(blksize && max_xfer_len && minor_hdl &&
1509 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1511 *blksize = zv->zv_volblocksize;
1512 *max_xfer_len = (uint64_t)zvol_maxphys;
1514 *objset_hdl = zv->zv_objset;
1515 *zil_hdl = zv->zv_zilog;
1516 *rl_hdl = &zv->zv_znode;
1517 *bonus_hdl = zv->zv_dbuf;
1522 * Return the current volume size to an external caller.
1523 * The size can change while the volume is open.
1526 zvol_get_volume_size(void *minor_hdl)
1528 zvol_state_t *zv = minor_hdl;
1530 return (zv->zv_volsize);
1534 * Return the current WCE setting to an external caller.
1535 * The WCE setting can change while the volume is open.
1538 zvol_get_volume_wce(void *minor_hdl)
1540 zvol_state_t *zv = minor_hdl;
1542 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1546 * Entry point for external callers to zvol_log_write
1549 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1552 zvol_state_t *zv = minor_hdl;
1554 zvol_log_write(zv, tx, off, resid, sync);
1557 * END entry points to allow external callers access to the volume.
1561 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
1565 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1568 struct dk_cinfo dki;
1569 struct dk_minfo dkm;
1570 struct dk_callback *dkc;
1574 mutex_enter(&spa_namespace_lock);
1576 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1579 mutex_exit(&spa_namespace_lock);
1582 ASSERT(zv->zv_total_opens > 0);
1587 bzero(&dki, sizeof (dki));
1588 (void) strcpy(dki.dki_cname, "zvol");
1589 (void) strcpy(dki.dki_dname, "zvol");
1590 dki.dki_ctype = DKC_UNKNOWN;
1591 dki.dki_unit = getminor(dev);
1592 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1593 mutex_exit(&spa_namespace_lock);
1594 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1598 case DKIOCGMEDIAINFO:
1599 bzero(&dkm, sizeof (dkm));
1600 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1601 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1602 dkm.dki_media_type = DK_UNKNOWN;
1603 mutex_exit(&spa_namespace_lock);
1604 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1610 uint64_t vs = zv->zv_volsize;
1611 uint8_t bs = zv->zv_min_bs;
1613 mutex_exit(&spa_namespace_lock);
1614 error = zvol_getefi((void *)arg, flag, vs, bs);
1618 case DKIOCFLUSHWRITECACHE:
1619 dkc = (struct dk_callback *)arg;
1620 mutex_exit(&spa_namespace_lock);
1621 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1622 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1623 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1630 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1631 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1639 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1645 zv->zv_flags |= ZVOL_WCE;
1646 mutex_exit(&spa_namespace_lock);
1648 zv->zv_flags &= ~ZVOL_WCE;
1649 mutex_exit(&spa_namespace_lock);
1650 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1658 * commands using these (like prtvtoc) expect ENOTSUP
1659 * since we're emulating an EFI label
1665 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1667 error = zvol_dumpify(zv);
1668 zfs_range_unlock(rl);
1672 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1674 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1676 error = zvol_dump_fini(zv);
1677 zfs_range_unlock(rl);
1685 mutex_exit(&spa_namespace_lock);
1693 return (zvol_minors != 0);
1699 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1701 ZFS_LOG(1, "ZVOL Initialized.");
1707 ddi_soft_state_fini(&zfsdev_state);
1708 ZFS_LOG(1, "ZVOL Deinitialized.");
1713 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1717 objset_t *os = zv->zv_objset;
1718 nvlist_t *nv = NULL;
1719 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1721 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1722 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1724 /* wait for dmu_free_long_range to actually free the blocks */
1725 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1727 tx = dmu_tx_create(os);
1728 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1729 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1730 error = dmu_tx_assign(tx, TXG_WAIT);
1737 * If we are resizing the dump device then we only need to
1738 * update the refreservation to match the newly updated
1739 * zvolsize. Otherwise, we save off the original state of the
1740 * zvol so that we can restore them if the zvol is ever undumpified.
1743 error = zap_update(os, ZVOL_ZAP_OBJ,
1744 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1745 &zv->zv_volsize, tx);
1747 uint64_t checksum, compress, refresrv, vbs, dedup;
1749 error = dsl_prop_get_integer(zv->zv_name,
1750 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1751 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1752 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1753 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1754 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1755 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1756 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1757 if (version >= SPA_VERSION_DEDUP) {
1758 error = error ? error :
1759 dsl_prop_get_integer(zv->zv_name,
1760 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1763 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1764 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1766 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1767 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1768 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1769 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1771 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1772 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1774 error = error ? error : dmu_object_set_blocksize(
1775 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1776 if (version >= SPA_VERSION_DEDUP) {
1777 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1778 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1782 zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1787 * We only need update the zvol's property if we are initializing
1788 * the dump area for the first time.
1791 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1792 VERIFY(nvlist_add_uint64(nv,
1793 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1794 VERIFY(nvlist_add_uint64(nv,
1795 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1796 ZIO_COMPRESS_OFF) == 0);
1797 VERIFY(nvlist_add_uint64(nv,
1798 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1799 ZIO_CHECKSUM_OFF) == 0);
1800 if (version >= SPA_VERSION_DEDUP) {
1801 VERIFY(nvlist_add_uint64(nv,
1802 zfs_prop_to_name(ZFS_PROP_DEDUP),
1803 ZIO_CHECKSUM_OFF) == 0);
1806 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1814 /* Allocate the space for the dump */
1815 error = zvol_prealloc(zv);
1820 zvol_dumpify(zvol_state_t *zv)
1823 uint64_t dumpsize = 0;
1825 objset_t *os = zv->zv_objset;
1827 if (zv->zv_flags & ZVOL_RDONLY)
1830 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
1831 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
1832 boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE;
1834 if ((error = zvol_dump_init(zv, resize)) != 0) {
1835 (void) zvol_dump_fini(zv);
1841 * Build up our lba mapping.
1843 error = zvol_get_lbas(zv);
1845 (void) zvol_dump_fini(zv);
1849 tx = dmu_tx_create(os);
1850 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1851 error = dmu_tx_assign(tx, TXG_WAIT);
1854 (void) zvol_dump_fini(zv);
1858 zv->zv_flags |= ZVOL_DUMPIFIED;
1859 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
1860 &zv->zv_volsize, tx);
1864 (void) zvol_dump_fini(zv);
1868 txg_wait_synced(dmu_objset_pool(os), 0);
1873 zvol_dump_fini(zvol_state_t *zv)
1876 objset_t *os = zv->zv_objset;
1879 uint64_t checksum, compress, refresrv, vbs, dedup;
1880 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
1883 * Attempt to restore the zvol back to its pre-dumpified state.
1884 * This is a best-effort attempt as it's possible that not all
1885 * of these properties were initialized during the dumpify process
1886 * (i.e. error during zvol_dump_init).
1889 tx = dmu_tx_create(os);
1890 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1891 error = dmu_tx_assign(tx, TXG_WAIT);
1896 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
1899 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1900 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
1901 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1902 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
1903 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1904 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
1905 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1906 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
1908 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1909 (void) nvlist_add_uint64(nv,
1910 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
1911 (void) nvlist_add_uint64(nv,
1912 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
1913 (void) nvlist_add_uint64(nv,
1914 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
1915 if (version >= SPA_VERSION_DEDUP &&
1916 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1917 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
1918 (void) nvlist_add_uint64(nv,
1919 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
1921 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
1925 zvol_free_extents(zv);
1926 zv->zv_flags &= ~ZVOL_DUMPIFIED;
1927 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
1928 /* wait for dmu_free_long_range to actually free the blocks */
1929 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1930 tx = dmu_tx_create(os);
1931 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1932 error = dmu_tx_assign(tx, TXG_WAIT);
1937 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
1938 zv->zv_volblocksize = vbs;
1945 static zvol_state_t *
1946 zvol_geom_create(const char *name)
1948 struct g_provider *pp;
1952 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
1953 gp->start = zvol_geom_start;
1954 gp->access = zvol_geom_access;
1955 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
1956 pp->sectorsize = DEV_BSIZE;
1958 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
1959 zv->zv_provider = pp;
1961 bioq_init(&zv->zv_queue);
1962 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
1970 zvol_geom_run(zvol_state_t *zv)
1972 struct g_provider *pp;
1974 pp = zv->zv_provider;
1975 g_error_provider(pp, 0);
1977 kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0,
1978 "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER));
1982 zvol_geom_destroy(zvol_state_t *zv)
1984 struct g_provider *pp;
1986 g_topology_assert();
1988 mtx_lock(&zv->zv_queue_mtx);
1990 wakeup_one(&zv->zv_queue);
1991 while (zv->zv_state != 2)
1992 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
1993 mtx_destroy(&zv->zv_queue_mtx);
1995 pp = zv->zv_provider;
1996 zv->zv_provider = NULL;
1998 g_wither_geom(pp->geom, ENXIO);
2000 kmem_free(zv, sizeof(*zv));
2004 zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace)
2006 int count, error, flags;
2008 g_topology_assert();
2011 * To make it easier we expect either open or close, but not both
2014 KASSERT((acr >= 0 && acw >= 0 && ace >= 0) ||
2015 (acr <= 0 && acw <= 0 && ace <= 0),
2016 ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).",
2017 pp->name, acr, acw, ace));
2019 if (pp->private == NULL) {
2020 if (acr <= 0 && acw <= 0 && ace <= 0)
2026 * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0,
2027 * because GEOM already handles that and handles it a bit differently.
2028 * GEOM allows for multiple read/exclusive consumers and ZFS allows
2029 * only one exclusive consumer, no matter if it is reader or writer.
2030 * I like better the way GEOM works so I'll leave it for GEOM to
2031 * decide what to do.
2034 count = acr + acw + ace;
2039 if (acr != 0 || ace != 0)
2044 g_topology_unlock();
2046 error = zvol_open(pp, flags, count);
2048 error = zvol_close(pp, flags, -count);
2054 zvol_geom_start(struct bio *bp)
2059 switch (bp->bio_cmd) {
2063 zv = bp->bio_to->private;
2065 mtx_lock(&zv->zv_queue_mtx);
2066 first = (bioq_first(&zv->zv_queue) == NULL);
2067 bioq_insert_tail(&zv->zv_queue, bp);
2068 mtx_unlock(&zv->zv_queue_mtx);
2070 wakeup_one(&zv->zv_queue);
2075 g_io_deliver(bp, EOPNOTSUPP);
2081 zvol_geom_worker(void *arg)
2086 thread_lock(curthread);
2087 sched_prio(curthread, PRIBIO);
2088 thread_unlock(curthread);
2092 mtx_lock(&zv->zv_queue_mtx);
2093 bp = bioq_takefirst(&zv->zv_queue);
2095 if (zv->zv_state == 1) {
2097 wakeup(&zv->zv_state);
2098 mtx_unlock(&zv->zv_queue_mtx);
2101 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
2105 mtx_unlock(&zv->zv_queue_mtx);
2106 switch (bp->bio_cmd) {
2108 zil_commit(zv->zv_zilog, ZVOL_OBJ);
2109 g_io_deliver(bp, 0);
2119 extern boolean_t dataset_name_hidden(const char *name);
2122 zvol_create_snapshots(objset_t *os, const char *name)
2124 uint64_t cookie, obj;
2129 sname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2131 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL,
2135 len = snprintf(sname, MAXPATHLEN, "%s@", name);
2136 if (len >= MAXPATHLEN) {
2137 dmu_objset_rele(os, FTAG);
2138 error = ENAMETOOLONG;
2142 error = dmu_snapshot_list_next(os, MAXPATHLEN - len,
2143 sname + len, &obj, &cookie, NULL);
2145 if (error == ENOENT)
2150 if ((error = zvol_create_minor(sname)) != 0) {
2151 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2157 kmem_free(sname, MAXPATHLEN);
2162 zvol_create_minors(const char *name)
2169 if (dataset_name_hidden(name))
2172 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2173 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2177 if (dmu_objset_type(os) == DMU_OST_ZVOL) {
2178 if ((error = zvol_create_minor(name)) == 0)
2179 error = zvol_create_snapshots(os, name);
2181 printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n",
2184 dmu_objset_rele(os, FTAG);
2187 if (dmu_objset_type(os) != DMU_OST_ZFS) {
2188 dmu_objset_rele(os, FTAG);
2192 osname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2193 if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) {
2194 dmu_objset_rele(os, FTAG);
2195 kmem_free(osname, MAXPATHLEN);
2198 p = osname + strlen(osname);
2199 len = MAXPATHLEN - (p - osname);
2201 /* Prefetch the datasets. */
2203 while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) {
2204 if (!dataset_name_hidden(osname))
2205 (void) dmu_objset_prefetch(osname, NULL);
2209 while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL,
2211 dmu_objset_rele(os, FTAG);
2212 (void)zvol_create_minors(osname);
2213 if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) {
2214 printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n",
2220 dmu_objset_rele(os, FTAG);
2221 kmem_free(osname, MAXPATHLEN);
2226 zvol_rename_minor(struct g_geom *gp, const char *newname)
2228 struct g_provider *pp;
2231 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2232 g_topology_assert();
2234 pp = LIST_FIRST(&gp->provider);
2239 zv->zv_provider = NULL;
2240 g_wither_provider(pp, ENXIO);
2242 pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
2243 pp->sectorsize = DEV_BSIZE;
2244 pp->mediasize = zv->zv_volsize;
2246 zv->zv_provider = pp;
2247 strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
2248 g_error_provider(pp, 0);
2252 zvol_rename_minors(const char *oldname, const char *newname)
2254 char name[MAXPATHLEN];
2255 struct g_provider *pp;
2257 size_t oldnamelen, newnamelen;
2261 oldnamelen = strlen(oldname);
2262 newnamelen = strlen(newname);
2265 mutex_enter(&spa_namespace_lock);
2268 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) {
2269 pp = LIST_FIRST(&gp->provider);
2275 if (strcmp(zv->zv_name, oldname) == 0) {
2276 zvol_rename_minor(gp, newname);
2277 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
2278 (zv->zv_name[oldnamelen] == '/' ||
2279 zv->zv_name[oldnamelen] == '@')) {
2280 snprintf(name, sizeof(name), "%s%c%s", newname,
2281 zv->zv_name[oldnamelen],
2282 zv->zv_name + oldnamelen + 1);
2283 zvol_rename_minor(gp, name);
2287 g_topology_unlock();
2288 mutex_exit(&spa_namespace_lock);