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) 2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
23 * All rights reserved.
26 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
27 * Use is subject to license terms.
31 * ZFS volume emulation driver.
33 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
34 * Volumes are accessed through the symbolic links named:
36 * /dev/zvol/dsk/<pool_name>/<dataset_name>
37 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
39 * These links are created by the ZFS-specific devfsadm link generator.
40 * Volumes are persistent through reboot. No user command needs to be
41 * run before opening and using a device.
44 #include <sys/types.h>
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/errno.h>
53 #include <sys/cmn_err.h>
58 #include <sys/dmu_traverse.h>
59 #include <sys/dnode.h>
60 #include <sys/dsl_dataset.h>
61 #include <sys/dsl_prop.h>
63 #include <sys/byteorder.h>
64 #include <sys/sunddi.h>
65 #include <sys/dirent.h>
66 #include <sys/policy.h>
67 #include <sys/fs/zfs.h>
68 #include <sys/zfs_ioctl.h>
70 #include <sys/refcount.h>
71 #include <sys/zfs_znode.h>
72 #include <sys/zfs_rlock.h>
73 #include <sys/vdev_impl.h>
75 #include <sys/zil_impl.h>
76 #include <geom/geom.h>
78 #include "zfs_namecheck.h"
80 #define ZVOL_DUMPSIZE "dumpsize"
82 struct g_class zfs_zvol_class = {
87 DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol);
90 * This lock protects the zvol_state structure from being modified
91 * while it's being used, e.g. an open that comes in before a create
92 * finishes. It also protects temporary opens of the dataset so that,
93 * e.g., an open doesn't get a spurious EBUSY.
95 static kmutex_t zvol_state_lock;
96 static uint32_t zvol_minors;
98 typedef struct zvol_extent {
100 dva_t ze_dva; /* dva associated with this extent */
101 uint64_t ze_nblks; /* number of blocks in extent */
105 * The in-core state of each volume.
107 typedef struct zvol_state {
108 char zv_name[MAXPATHLEN]; /* pool/dd name */
109 uint64_t zv_volsize; /* amount of space we advertise */
110 uint64_t zv_volblocksize; /* volume block size */
111 struct g_provider *zv_provider; /* GEOM provider */
112 uint8_t zv_min_bs; /* minimum addressable block shift */
113 uint8_t zv_flags; /* readonly; dumpified */
114 objset_t *zv_objset; /* objset handle */
115 uint32_t zv_mode; /* DS_MODE_* flags at open time */
116 uint32_t zv_total_opens; /* total open count */
117 zilog_t *zv_zilog; /* ZIL handle */
118 list_t zv_extents; /* List of extents for dump */
119 znode_t zv_znode; /* for range locking */
121 struct bio_queue_head zv_queue;
122 struct mtx zv_queue_mtx; /* zv_queue mutex */
126 * zvol specific flags
128 #define ZVOL_RDONLY 0x1
129 #define ZVOL_DUMPIFIED 0x2
130 #define ZVOL_EXCL 0x4
133 * zvol maximum transfer in one DMU tx.
135 int zvol_maxphys = DMU_MAX_ACCESS/2;
137 extern int zfs_set_prop_nvlist(const char *, nvlist_t *);
138 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
139 static int zvol_dumpify(zvol_state_t *zv);
140 static int zvol_dump_fini(zvol_state_t *zv);
141 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
144 zvol_size_changed(zvol_state_t *zv, major_t maj)
146 struct g_provider *pp;
150 pp = zv->zv_provider;
153 if (zv->zv_volsize == pp->mediasize)
156 * Changing provider size is not really supported by GEOM, but it
157 * should be safe when provider is closed.
159 if (zv->zv_total_opens > 0)
161 pp->mediasize = zv->zv_volsize;
165 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
170 if (volsize % blocksize != 0)
174 if (volsize - 1 > SPEC_MAXOFFSET_T)
181 zvol_check_volblocksize(uint64_t volblocksize)
183 if (volblocksize < SPA_MINBLOCKSIZE ||
184 volblocksize > SPA_MAXBLOCKSIZE ||
192 zvol_readonly_changed_cb(void *arg, uint64_t newval)
194 zvol_state_t *zv = arg;
197 zv->zv_flags |= ZVOL_RDONLY;
199 zv->zv_flags &= ~ZVOL_RDONLY;
203 zvol_get_stats(objset_t *os, nvlist_t *nv)
206 dmu_object_info_t doi;
210 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
214 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
216 error = dmu_object_info(os, ZVOL_OBJ, &doi);
219 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
220 doi.doi_data_block_size);
226 static zvol_state_t *
227 zvol_minor_lookup(const char *name)
229 struct g_provider *pp;
233 ASSERT(MUTEX_HELD(&zvol_state_lock));
235 LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) {
236 LIST_FOREACH(pp, &gp->provider, provider) {
237 if (strcmp(pp->name + sizeof(ZVOL_DEV_DIR), name) == 0)
238 return (pp->private);
246 zvol_access(struct g_provider *pp, int acr, int acw, int ace)
251 mutex_enter(&zvol_state_lock);
255 if (acr <= 0 && acw <= 0 && ace <= 0)
257 mutex_exit(&zvol_state_lock);
261 ASSERT(zv->zv_objset != NULL);
264 ((zv->zv_flags & ZVOL_RDONLY) ||
265 (zv->zv_mode & DS_MODE_READONLY))) {
266 mutex_exit(&zvol_state_lock);
270 zv->zv_total_opens += acr + acw + ace;
271 zvol_size_changed(zv, 0);
273 mutex_exit(&zvol_state_lock);
279 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
281 * We store data in the log buffers if it's small enough.
282 * Otherwise we will later flush the data out via dmu_sync().
284 ssize_t zvol_immediate_write_sz = 32768;
287 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t len)
289 uint32_t blocksize = zv->zv_volblocksize;
290 zilog_t *zilog = zv->zv_zilog;
293 if (zilog->zl_replay) {
294 dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
295 zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] =
296 zilog->zl_replaying_seq;
301 ssize_t nbytes = MIN(len, blocksize - P2PHASE(off, blocksize));
302 itx_t *itx = zil_itx_create(TX_WRITE, sizeof (*lr));
305 len > zvol_immediate_write_sz ? WR_INDIRECT : WR_NEED_COPY;
306 itx->itx_private = zv;
307 lr = (lr_write_t *)&itx->itx_lr;
308 lr->lr_foid = ZVOL_OBJ;
310 lr->lr_length = nbytes;
311 lr->lr_blkoff = off - P2ALIGN_TYPED(off, blocksize, uint64_t);
312 BP_ZERO(&lr->lr_blkptr);
314 (void) zil_itx_assign(zilog, itx, tx);
321 zvol_start(struct bio *bp)
325 switch (bp->bio_cmd) {
329 zv = bp->bio_to->private;
331 mtx_lock(&zv->zv_queue_mtx);
332 bioq_insert_tail(&zv->zv_queue, bp);
333 wakeup_one(&zv->zv_queue);
334 mtx_unlock(&zv->zv_queue_mtx);
337 if (g_handleattr_int(bp, "ZFS::iszvol", 1))
342 g_io_deliver(bp, EOPNOTSUPP);
348 zvol_serve_one(zvol_state_t *zv, struct bio *bp)
350 uint64_t off, volsize;
356 boolean_t doread = (bp->bio_cmd == BIO_READ);
358 off = bp->bio_offset;
359 volsize = zv->zv_volsize;
365 resid = bp->bio_length;
370 * There must be no buffer changes when doing a dmu_sync() because
371 * we can't change the data whilst calculating the checksum.
372 * A better approach than a per zvol rwlock would be to lock ranges.
374 rl = zfs_range_lock(&zv->zv_znode, off, resid,
375 doread ? RL_READER : RL_WRITER);
377 while (resid != 0 && off < volsize) {
378 size_t size = MIN(resid, zvol_maxphys); /* zvol_maxphys per tx */
380 if (size > volsize - off) /* don't write past the end */
381 size = volsize - off;
384 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
387 dmu_tx_t *tx = dmu_tx_create(os);
388 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
389 error = dmu_tx_assign(tx, TXG_WAIT);
393 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
394 zvol_log_write(zv, tx, off, size);
399 /* convert checksum errors into IO errors */
408 zfs_range_unlock(rl);
410 bp->bio_completed = bp->bio_length - resid;
411 if (bp->bio_completed < bp->bio_length)
412 bp->bio_error = (off > volsize ? EINVAL : error);
416 zvol_worker(void *arg)
421 thread_lock(curthread);
422 sched_prio(curthread, PRIBIO);
423 thread_unlock(curthread);
427 mtx_lock(&zv->zv_queue_mtx);
428 bp = bioq_takefirst(&zv->zv_queue);
430 if (zv->zv_state == 1) {
432 wakeup(&zv->zv_state);
433 mtx_unlock(&zv->zv_queue_mtx);
436 msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP,
440 mtx_unlock(&zv->zv_queue_mtx);
441 switch (bp->bio_cmd) {
446 zvol_serve_one(zv, bp);
450 if (bp->bio_cmd == BIO_FLUSH && !zil_disable)
451 zil_commit(zv->zv_zilog, UINT64_MAX, ZVOL_OBJ);
453 g_io_deliver(bp, bp->bio_error);
457 /* extent mapping arg */
465 zvol_map_block(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
466 const dnode_phys_t *dnp, void *arg)
468 struct maparg *ma = arg;
470 int bs = ma->ma_zv->zv_volblocksize;
472 if (bp == NULL || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
475 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
478 /* Abort immediately if we have encountered gang blocks */
483 * See if the block is at the end of the previous extent.
485 ze = list_tail(&ma->ma_zv->zv_extents);
487 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
488 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
489 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
494 dprintf_bp(bp, "%s", "next blkptr:");
496 /* start a new extent */
497 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
498 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
500 list_insert_tail(&ma->ma_zv->zv_extents, ze);
505 zvol_free_extents(zvol_state_t *zv)
509 while (ze = list_head(&zv->zv_extents)) {
510 list_remove(&zv->zv_extents, ze);
511 kmem_free(ze, sizeof (zvol_extent_t));
516 zvol_get_lbas(zvol_state_t *zv)
523 zvol_free_extents(zv);
525 err = traverse_dataset(dmu_objset_ds(zv->zv_objset), 0,
526 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
527 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
528 zvol_free_extents(zv);
529 return (err ? err : EIO);
537 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
539 zfs_creat_t *zct = arg;
540 nvlist_t *nvprops = zct->zct_props;
542 uint64_t volblocksize, volsize;
544 VERIFY(nvlist_lookup_uint64(nvprops,
545 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
546 if (nvlist_lookup_uint64(nvprops,
547 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
548 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
551 * These properties must be removed from the list so the generic
552 * property setting step won't apply to them.
554 VERIFY(nvlist_remove_all(nvprops,
555 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
556 (void) nvlist_remove_all(nvprops,
557 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
559 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
563 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
567 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
572 * Replay a TX_WRITE ZIL transaction that didn't get committed
573 * after a system failure
576 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
578 objset_t *os = zv->zv_objset;
579 char *data = (char *)(lr + 1); /* data follows lr_write_t */
580 uint64_t off = lr->lr_offset;
581 uint64_t len = lr->lr_length;
586 byteswap_uint64_array(lr, sizeof (*lr));
588 /* If it's a dmu_sync() block get the data and write the whole block */
589 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t))
590 zil_get_replay_data(dmu_objset_zil(os), lr);
592 tx = dmu_tx_create(os);
593 dmu_tx_hold_write(tx, ZVOL_OBJ, off, len);
594 error = dmu_tx_assign(tx, TXG_WAIT);
598 dmu_write(os, ZVOL_OBJ, off, len, data, tx);
607 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
613 * Callback vectors for replaying records.
614 * Only TX_WRITE is needed for zvol.
616 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
617 zvol_replay_err, /* 0 no such transaction type */
618 zvol_replay_err, /* TX_CREATE */
619 zvol_replay_err, /* TX_MKDIR */
620 zvol_replay_err, /* TX_MKXATTR */
621 zvol_replay_err, /* TX_SYMLINK */
622 zvol_replay_err, /* TX_REMOVE */
623 zvol_replay_err, /* TX_RMDIR */
624 zvol_replay_err, /* TX_LINK */
625 zvol_replay_err, /* TX_RENAME */
626 zvol_replay_write, /* TX_WRITE */
627 zvol_replay_err, /* TX_TRUNCATE */
628 zvol_replay_err, /* TX_SETATTR */
629 zvol_replay_err, /* TX_ACL */
630 zvol_replay_err, /* TX_CREATE_ACL */
631 zvol_replay_err, /* TX_CREATE_ATTR */
632 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
633 zvol_replay_err, /* TX_MKDIR_ACL */
634 zvol_replay_err, /* TX_MKDIR_ATTR */
635 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
636 zvol_replay_err, /* TX_WRITE2 */
640 * Create a minor node (plus a whole lot more) for the specified volume.
643 zvol_create_minor(const char *name, major_t maj)
645 struct g_provider *pp;
649 dmu_object_info_t doi;
651 int ds_mode = DS_MODE_OWNER;
656 mutex_enter(&zvol_state_lock);
658 if ((zv = zvol_minor_lookup(name)) != NULL) {
663 if (strchr(name, '@') != 0)
664 ds_mode |= DS_MODE_READONLY;
666 error = dmu_objset_open(name, DMU_OST_ZVOL, ds_mode, &os);
670 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
672 dmu_objset_close(os);
676 gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
677 gp->start = zvol_start;
678 gp->access = zvol_access;
679 pp = g_new_providerf(gp, "%s/%s", ZVOL_DEV_DIR, name);
680 pp->mediasize = volsize;
681 pp->sectorsize = DEV_BSIZE;
683 zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
684 (void) strcpy(zv->zv_name, name);
685 zv->zv_min_bs = DEV_BSHIFT;
686 zv->zv_provider = pp;
687 zv->zv_volsize = pp->mediasize;
689 zv->zv_mode = ds_mode;
690 zv->zv_zilog = zil_open(os, zvol_get_data);
691 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
692 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
693 sizeof (rl_t), offsetof(rl_t, r_node));
694 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
695 offsetof(zvol_extent_t, ze_node));
696 /* get and cache the blocksize */
697 error = dmu_object_info(os, ZVOL_OBJ, &doi);
699 zv->zv_volblocksize = doi.doi_data_block_size;
701 zil_replay(os, zv, zvol_replay_vector);
703 /* XXX this should handle the possible i/o error */
704 VERIFY(dsl_prop_register(dmu_objset_ds(zv->zv_objset),
705 "readonly", zvol_readonly_changed_cb, zv) == 0);
708 g_error_provider(pp, 0);
710 bioq_init(&zv->zv_queue);
711 mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
713 kproc_kthread_add(zvol_worker, zv, &zfsproc, NULL, 0, 0, "zfskern",
714 "zvol %s", pp->name + strlen(ZVOL_DEV_DIR) + 1);
718 mutex_exit(&zvol_state_lock);
726 * Remove minor node for the specified volume.
729 zvol_remove_minor(const char *name)
731 struct g_provider *pp;
737 mutex_enter(&zvol_state_lock);
739 if ((zv = zvol_minor_lookup(name)) == NULL) {
744 if (zv->zv_total_opens != 0) {
749 VERIFY(dsl_prop_unregister(dmu_objset_ds(zv->zv_objset),
750 "readonly", zvol_readonly_changed_cb, zv) == 0);
752 mtx_lock(&zv->zv_queue_mtx);
754 wakeup_one(&zv->zv_queue);
755 while (zv->zv_state != 2)
756 msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0);
757 mtx_unlock(&zv->zv_queue_mtx);
758 mtx_destroy(&zv->zv_queue_mtx);
760 pp = zv->zv_provider;
762 g_wither_geom(pp->geom, ENXIO);
764 zil_close(zv->zv_zilog);
766 dmu_objset_close(zv->zv_objset);
767 zv->zv_objset = NULL;
768 avl_destroy(&zv->zv_znode.z_range_avl);
769 mutex_destroy(&zv->zv_znode.z_range_lock);
771 kmem_free(zv, sizeof(*zv));
775 mutex_exit(&zvol_state_lock);
783 zvol_prealloc(zvol_state_t *zv)
785 objset_t *os = zv->zv_objset;
788 uint64_t refd, avail, usedobjs, availobjs;
789 uint64_t resid = zv->zv_volsize;
792 /* Check the space usage before attempting to allocate the space */
793 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
794 if (avail < zv->zv_volsize)
797 /* Free old extents if they exist */
798 zvol_free_extents(zv);
800 /* allocate the blocks by writing each one */
801 data = kmem_zalloc(SPA_MAXBLOCKSIZE, KM_SLEEP);
805 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
807 tx = dmu_tx_create(os);
808 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
809 error = dmu_tx_assign(tx, TXG_WAIT);
812 kmem_free(data, SPA_MAXBLOCKSIZE);
813 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
816 dmu_write(os, ZVOL_OBJ, off, bytes, data, tx);
821 kmem_free(data, SPA_MAXBLOCKSIZE);
822 txg_wait_synced(dmu_objset_pool(os), 0);
828 zvol_update_volsize(zvol_state_t *zv, major_t maj, uint64_t volsize)
833 ASSERT(MUTEX_HELD(&zvol_state_lock));
835 tx = dmu_tx_create(zv->zv_objset);
836 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
837 error = dmu_tx_assign(tx, TXG_WAIT);
843 error = zap_update(zv->zv_objset, ZVOL_ZAP_OBJ, "size", 8, 1,
848 error = dmu_free_long_range(zv->zv_objset,
849 ZVOL_OBJ, volsize, DMU_OBJECT_END);
852 * If we are using a faked-up state (zv_provider == NULL) then don't
853 * try to update the in-core zvol state.
855 if (error == 0 && zv->zv_provider) {
856 zv->zv_volsize = volsize;
857 zvol_size_changed(zv, maj);
863 zvol_set_volsize(const char *name, major_t maj, uint64_t volsize)
867 dmu_object_info_t doi;
868 uint64_t old_volsize = 0ULL;
869 zvol_state_t state = { 0 };
873 mutex_enter(&zvol_state_lock);
875 if ((zv = zvol_minor_lookup(name)) == NULL) {
877 * If we are doing a "zfs clone -o volsize=", then the
878 * minor node won't exist yet.
880 error = dmu_objset_open(name, DMU_OST_ZVOL, DS_MODE_OWNER,
886 old_volsize = zv->zv_volsize;
888 if ((error = dmu_object_info(zv->zv_objset, ZVOL_OBJ, &doi)) != 0 ||
889 (error = zvol_check_volsize(volsize,
890 doi.doi_data_block_size)) != 0)
893 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY)) {
898 error = zvol_update_volsize(zv, maj, volsize);
902 * Reinitialize the dump area to the new size. If we
903 * failed to resize the dump area then restore the it back to
904 * it's original size.
906 if (error == 0 && zv->zv_flags & ZVOL_DUMPIFIED) {
907 if ((error = zvol_dumpify(zv)) != 0 ||
908 (error = dumpvp_resize()) != 0) {
909 (void) zvol_update_volsize(zv, maj, old_volsize);
910 error = zvol_dumpify(zv);
917 dmu_objset_close(state.zv_objset);
919 mutex_exit(&zvol_state_lock);
927 zvol_set_volblocksize(const char *name, uint64_t volblocksize)
935 mutex_enter(&zvol_state_lock);
937 if ((zv = zvol_minor_lookup(name)) == NULL) {
941 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY)) {
946 tx = dmu_tx_create(zv->zv_objset);
947 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
948 error = dmu_tx_assign(tx, TXG_WAIT);
952 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
953 volblocksize, 0, tx);
954 if (error == ENOTSUP)
958 zv->zv_volblocksize = volblocksize;
961 mutex_exit(&zvol_state_lock);
969 zvol_get_done(dmu_buf_t *db, void *vzgd)
971 zgd_t *zgd = (zgd_t *)vzgd;
972 rl_t *rl = zgd->zgd_rl;
974 dmu_buf_rele(db, vzgd);
975 zfs_range_unlock(rl);
976 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
977 kmem_free(zgd, sizeof (zgd_t));
981 * Get data to generate a TX_WRITE intent log record.
984 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
986 zvol_state_t *zv = arg;
987 objset_t *os = zv->zv_objset;
991 uint64_t boff; /* block starting offset */
992 int dlen = lr->lr_length; /* length of user data */
999 * Write records come in two flavors: immediate and indirect.
1000 * For small writes it's cheaper to store the data with the
1001 * log record (immediate); for large writes it's cheaper to
1002 * sync the data and get a pointer to it (indirect) so that
1003 * we don't have to write the data twice.
1005 if (buf != NULL) /* immediate write */
1006 return (dmu_read(os, ZVOL_OBJ, lr->lr_offset, dlen, buf,
1007 DMU_READ_NO_PREFETCH));
1009 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
1010 zgd->zgd_zilog = zv->zv_zilog;
1011 zgd->zgd_bp = &lr->lr_blkptr;
1014 * Lock the range of the block to ensure that when the data is
1015 * written out and its checksum is being calculated that no other
1016 * thread can change the block.
1018 boff = P2ALIGN_TYPED(lr->lr_offset, zv->zv_volblocksize, uint64_t);
1019 rl = zfs_range_lock(&zv->zv_znode, boff, zv->zv_volblocksize,
1023 VERIFY(0 == dmu_buf_hold(os, ZVOL_OBJ, lr->lr_offset, zgd, &db));
1025 error = dmu_sync(zio, db, &lr->lr_blkptr,
1026 lr->lr_common.lrc_txg, zvol_get_done, zgd);
1029 * dmu_sync() can compress a block of zeros to a null blkptr
1030 * but the block size still needs to be passed through to
1033 BP_SET_LSIZE(&lr->lr_blkptr, db->db_size);
1034 zil_add_block(zv->zv_zilog, &lr->lr_blkptr);
1038 * If we get EINPROGRESS, then we need to wait for a
1039 * write IO initiated by dmu_sync() to complete before
1040 * we can release this dbuf. We will finish everything
1041 * up in the zvol_get_done() callback.
1043 if (error == EINPROGRESS)
1045 dmu_buf_rele(db, zgd);
1046 zfs_range_unlock(rl);
1047 kmem_free(zgd, sizeof (zgd_t));
1054 return (zvol_minors != 0);
1060 mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
1061 ZFS_LOG(1, "ZVOL Initialized.");
1067 mutex_destroy(&zvol_state_lock);
1068 ZFS_LOG(1, "ZVOL Deinitialized.");
1072 zvol_is_swap(zvol_state_t *zv)
1075 boolean_t ret = B_FALSE;
1081 devpathlen = strlen(ZVOL_FULL_DEV_DIR) + strlen(zv->zv_name) + 1;
1082 devpath = kmem_alloc(devpathlen, KM_SLEEP);
1083 (void) sprintf(devpath, "%s%s", ZVOL_FULL_DEV_DIR, zv->zv_name);
1084 error = lookupname(devpath, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
1085 kmem_free(devpath, devpathlen);
1087 ret = !error && IS_SWAPVP(common_specvp(vp));
1097 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1101 objset_t *os = zv->zv_objset;
1102 nvlist_t *nv = NULL;
1104 ASSERT(MUTEX_HELD(&zvol_state_lock));
1106 tx = dmu_tx_create(os);
1107 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1108 error = dmu_tx_assign(tx, TXG_WAIT);
1115 * If we are resizing the dump device then we only need to
1116 * update the refreservation to match the newly updated
1117 * zvolsize. Otherwise, we save off the original state of the
1118 * zvol so that we can restore them if the zvol is ever undumpified.
1121 error = zap_update(os, ZVOL_ZAP_OBJ,
1122 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1123 &zv->zv_volsize, tx);
1125 uint64_t checksum, compress, refresrv, vbs;
1127 error = dsl_prop_get_integer(zv->zv_name,
1128 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1129 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1130 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1131 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1132 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1133 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1134 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1136 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1137 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1139 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1140 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1141 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1142 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1144 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1145 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1150 /* Truncate the file */
1152 error = dmu_free_long_range(zv->zv_objset,
1153 ZVOL_OBJ, 0, DMU_OBJECT_END);
1159 * We only need update the zvol's property if we are initializing
1160 * the dump area for the first time.
1163 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1164 VERIFY(nvlist_add_uint64(nv,
1165 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1166 VERIFY(nvlist_add_uint64(nv,
1167 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1168 ZIO_COMPRESS_OFF) == 0);
1169 VERIFY(nvlist_add_uint64(nv,
1170 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1171 ZIO_CHECKSUM_OFF) == 0);
1172 VERIFY(nvlist_add_uint64(nv,
1173 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE),
1174 SPA_MAXBLOCKSIZE) == 0);
1176 error = zfs_set_prop_nvlist(zv->zv_name, nv);
1183 /* Allocate the space for the dump */
1184 error = zvol_prealloc(zv);
1189 zvol_dumpify(zvol_state_t *zv)
1192 uint64_t dumpsize = 0;
1194 objset_t *os = zv->zv_objset;
1196 if (zv->zv_flags & ZVOL_RDONLY || (zv->zv_mode & DS_MODE_READONLY))
1200 * We do not support swap devices acting as dump devices.
1202 if (zvol_is_swap(zv))
1205 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
1206 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
1207 boolean_t resize = (dumpsize > 0) ? B_TRUE : B_FALSE;
1209 if ((error = zvol_dump_init(zv, resize)) != 0) {
1210 (void) zvol_dump_fini(zv);
1216 * Build up our lba mapping.
1218 error = zvol_get_lbas(zv);
1220 (void) zvol_dump_fini(zv);
1224 tx = dmu_tx_create(os);
1225 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1226 error = dmu_tx_assign(tx, TXG_WAIT);
1229 (void) zvol_dump_fini(zv);
1233 zv->zv_flags |= ZVOL_DUMPIFIED;
1234 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
1235 &zv->zv_volsize, tx);
1239 (void) zvol_dump_fini(zv);
1243 txg_wait_synced(dmu_objset_pool(os), 0);
1248 zvol_dump_fini(zvol_state_t *zv)
1251 objset_t *os = zv->zv_objset;
1254 uint64_t checksum, compress, refresrv, vbs;
1257 * Attempt to restore the zvol back to its pre-dumpified state.
1258 * This is a best-effort attempt as it's possible that not all
1259 * of these properties were initialized during the dumpify process
1260 * (i.e. error during zvol_dump_init).
1263 tx = dmu_tx_create(os);
1264 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1265 error = dmu_tx_assign(tx, TXG_WAIT);
1270 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
1273 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1274 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
1275 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1276 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
1277 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1278 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
1279 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
1280 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
1282 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1283 (void) nvlist_add_uint64(nv,
1284 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
1285 (void) nvlist_add_uint64(nv,
1286 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
1287 (void) nvlist_add_uint64(nv,
1288 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
1289 (void) nvlist_add_uint64(nv,
1290 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), vbs);
1291 (void) zfs_set_prop_nvlist(zv->zv_name, nv);
1294 zvol_free_extents(zv);
1295 zv->zv_flags &= ~ZVOL_DUMPIFIED;
1296 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);