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.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
27 #include <sys/dmu_impl.h>
28 #include <sys/dmu_tx.h>
30 #include <sys/dnode.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dmu_zfetch.h>
40 #include <sys/zfs_ioctl.h>
42 #include <sys/zio_checksum.h>
45 #include <sys/zfs_znode.h>
48 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
49 { DMU_BSWAP_UINT8, TRUE, "unallocated" },
50 { DMU_BSWAP_ZAP, TRUE, "object directory" },
51 { DMU_BSWAP_UINT64, TRUE, "object array" },
52 { DMU_BSWAP_UINT8, TRUE, "packed nvlist" },
53 { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" },
54 { DMU_BSWAP_UINT64, TRUE, "bpobj" },
55 { DMU_BSWAP_UINT64, TRUE, "bpobj header" },
56 { DMU_BSWAP_UINT64, TRUE, "SPA space map header" },
57 { DMU_BSWAP_UINT64, TRUE, "SPA space map" },
58 { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" },
59 { DMU_BSWAP_DNODE, TRUE, "DMU dnode" },
60 { DMU_BSWAP_OBJSET, TRUE, "DMU objset" },
61 { DMU_BSWAP_UINT64, TRUE, "DSL directory" },
62 { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"},
63 { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" },
64 { DMU_BSWAP_ZAP, TRUE, "DSL props" },
65 { DMU_BSWAP_UINT64, TRUE, "DSL dataset" },
66 { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" },
67 { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" },
68 { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" },
69 { DMU_BSWAP_ZAP, TRUE, "ZFS directory" },
70 { DMU_BSWAP_ZAP, TRUE, "ZFS master node" },
71 { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" },
72 { DMU_BSWAP_UINT8, FALSE, "zvol object" },
73 { DMU_BSWAP_ZAP, TRUE, "zvol prop" },
74 { DMU_BSWAP_UINT8, FALSE, "other uint8[]" },
75 { DMU_BSWAP_UINT64, FALSE, "other uint64[]" },
76 { DMU_BSWAP_ZAP, TRUE, "other ZAP" },
77 { DMU_BSWAP_ZAP, TRUE, "persistent error log" },
78 { DMU_BSWAP_UINT8, TRUE, "SPA history" },
79 { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" },
80 { DMU_BSWAP_ZAP, TRUE, "Pool properties" },
81 { DMU_BSWAP_ZAP, TRUE, "DSL permissions" },
82 { DMU_BSWAP_ACL, TRUE, "ZFS ACL" },
83 { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" },
84 { DMU_BSWAP_UINT8, TRUE, "FUID table" },
85 { DMU_BSWAP_UINT64, TRUE, "FUID table size" },
86 { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"},
87 { DMU_BSWAP_ZAP, TRUE, "scan work queue" },
88 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" },
89 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" },
90 { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"},
91 { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" },
92 { DMU_BSWAP_ZAP, TRUE, "DDT statistics" },
93 { DMU_BSWAP_UINT8, TRUE, "System attributes" },
94 { DMU_BSWAP_ZAP, TRUE, "SA master node" },
95 { DMU_BSWAP_ZAP, TRUE, "SA attr registration" },
96 { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" },
97 { DMU_BSWAP_ZAP, TRUE, "scan translations" },
98 { DMU_BSWAP_UINT8, FALSE, "deduplicated block" },
99 { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" },
100 { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" },
101 { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" },
102 { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" }
105 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
106 { byteswap_uint8_array, "uint8" },
107 { byteswap_uint16_array, "uint16" },
108 { byteswap_uint32_array, "uint32" },
109 { byteswap_uint64_array, "uint64" },
110 { zap_byteswap, "zap" },
111 { dnode_buf_byteswap, "dnode" },
112 { dmu_objset_byteswap, "objset" },
113 { zfs_znode_byteswap, "znode" },
114 { zfs_oldacl_byteswap, "oldacl" },
115 { zfs_acl_byteswap, "acl" }
119 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
120 void *tag, dmu_buf_t **dbp, int flags)
126 int db_flags = DB_RF_CANFAIL;
128 if (flags & DMU_READ_NO_PREFETCH)
129 db_flags |= DB_RF_NOPREFETCH;
131 err = dnode_hold(os, object, FTAG, &dn);
134 blkid = dbuf_whichblock(dn, offset);
135 rw_enter(&dn->dn_struct_rwlock, RW_READER);
136 db = dbuf_hold(dn, blkid, tag);
137 rw_exit(&dn->dn_struct_rwlock);
141 err = dbuf_read(db, NULL, db_flags);
148 dnode_rele(dn, FTAG);
149 *dbp = &db->db; /* NULL db plus first field offset is NULL */
156 return (DN_MAX_BONUSLEN);
160 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
162 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
169 if (dn->dn_bonus != db) {
171 } else if (newsize < 0 || newsize > db_fake->db_size) {
174 dnode_setbonuslen(dn, newsize, tx);
183 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
185 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
192 if (!DMU_OT_IS_VALID(type)) {
194 } else if (dn->dn_bonus != db) {
197 dnode_setbonus_type(dn, type, tx);
206 dmu_get_bonustype(dmu_buf_t *db_fake)
208 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
210 dmu_object_type_t type;
214 type = dn->dn_bonustype;
221 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
226 error = dnode_hold(os, object, FTAG, &dn);
227 dbuf_rm_spill(dn, tx);
228 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
229 dnode_rm_spill(dn, tx);
230 rw_exit(&dn->dn_struct_rwlock);
231 dnode_rele(dn, FTAG);
236 * returns ENOENT, EIO, or 0.
239 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
245 error = dnode_hold(os, object, FTAG, &dn);
249 rw_enter(&dn->dn_struct_rwlock, RW_READER);
250 if (dn->dn_bonus == NULL) {
251 rw_exit(&dn->dn_struct_rwlock);
252 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
253 if (dn->dn_bonus == NULL)
254 dbuf_create_bonus(dn);
258 /* as long as the bonus buf is held, the dnode will be held */
259 if (refcount_add(&db->db_holds, tag) == 1) {
260 VERIFY(dnode_add_ref(dn, db));
261 (void) atomic_inc_32_nv(&dn->dn_dbufs_count);
265 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
266 * hold and incrementing the dbuf count to ensure that dnode_move() sees
267 * a dnode hold for every dbuf.
269 rw_exit(&dn->dn_struct_rwlock);
271 dnode_rele(dn, FTAG);
273 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
280 * returns ENOENT, EIO, or 0.
282 * This interface will allocate a blank spill dbuf when a spill blk
283 * doesn't already exist on the dnode.
285 * if you only want to find an already existing spill db, then
286 * dmu_spill_hold_existing() should be used.
289 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
291 dmu_buf_impl_t *db = NULL;
294 if ((flags & DB_RF_HAVESTRUCT) == 0)
295 rw_enter(&dn->dn_struct_rwlock, RW_READER);
297 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
299 if ((flags & DB_RF_HAVESTRUCT) == 0)
300 rw_exit(&dn->dn_struct_rwlock);
303 err = dbuf_read(db, NULL, flags);
312 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
314 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
321 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
324 rw_enter(&dn->dn_struct_rwlock, RW_READER);
326 if (!dn->dn_have_spill) {
329 err = dmu_spill_hold_by_dnode(dn,
330 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
333 rw_exit(&dn->dn_struct_rwlock);
341 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
343 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
349 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
356 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
357 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
358 * and can induce severe lock contention when writing to several files
359 * whose dnodes are in the same block.
362 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
363 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
365 dsl_pool_t *dp = NULL;
367 uint64_t blkid, nblks, i;
373 ASSERT(length <= DMU_MAX_ACCESS);
375 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
376 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
377 dbuf_flags |= DB_RF_NOPREFETCH;
379 rw_enter(&dn->dn_struct_rwlock, RW_READER);
380 if (dn->dn_datablkshift) {
381 int blkshift = dn->dn_datablkshift;
382 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
383 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
385 if (offset + length > dn->dn_datablksz) {
386 zfs_panic_recover("zfs: accessing past end of object "
387 "%llx/%llx (size=%u access=%llu+%llu)",
388 (longlong_t)dn->dn_objset->
389 os_dsl_dataset->ds_object,
390 (longlong_t)dn->dn_object, dn->dn_datablksz,
391 (longlong_t)offset, (longlong_t)length);
392 rw_exit(&dn->dn_struct_rwlock);
397 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
399 if (dn->dn_objset->os_dsl_dataset)
400 dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
401 if (dp && dsl_pool_sync_context(dp))
403 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
404 blkid = dbuf_whichblock(dn, offset);
405 for (i = 0; i < nblks; i++) {
406 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
408 rw_exit(&dn->dn_struct_rwlock);
409 dmu_buf_rele_array(dbp, nblks, tag);
413 /* initiate async i/o */
415 (void) dbuf_read(db, zio, dbuf_flags);
418 curthread->td_ru.ru_oublock++;
422 rw_exit(&dn->dn_struct_rwlock);
424 /* wait for async i/o */
426 /* track read overhead when we are in sync context */
427 if (dp && dsl_pool_sync_context(dp))
428 dp->dp_read_overhead += gethrtime() - start;
430 dmu_buf_rele_array(dbp, nblks, tag);
434 /* wait for other io to complete */
436 for (i = 0; i < nblks; i++) {
437 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
438 mutex_enter(&db->db_mtx);
439 while (db->db_state == DB_READ ||
440 db->db_state == DB_FILL)
441 cv_wait(&db->db_changed, &db->db_mtx);
442 if (db->db_state == DB_UNCACHED)
444 mutex_exit(&db->db_mtx);
446 dmu_buf_rele_array(dbp, nblks, tag);
458 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
459 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
464 err = dnode_hold(os, object, FTAG, &dn);
468 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
469 numbufsp, dbpp, DMU_READ_PREFETCH);
471 dnode_rele(dn, FTAG);
477 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
478 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
480 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
486 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
487 numbufsp, dbpp, DMU_READ_PREFETCH);
494 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
497 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
502 for (i = 0; i < numbufs; i++) {
504 dbuf_rele(dbp[i], tag);
507 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
511 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
517 if (zfs_prefetch_disable)
520 if (len == 0) { /* they're interested in the bonus buffer */
521 dn = DMU_META_DNODE(os);
523 if (object == 0 || object >= DN_MAX_OBJECT)
526 rw_enter(&dn->dn_struct_rwlock, RW_READER);
527 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
528 dbuf_prefetch(dn, blkid);
529 rw_exit(&dn->dn_struct_rwlock);
534 * XXX - Note, if the dnode for the requested object is not
535 * already cached, we will do a *synchronous* read in the
536 * dnode_hold() call. The same is true for any indirects.
538 err = dnode_hold(os, object, FTAG, &dn);
542 rw_enter(&dn->dn_struct_rwlock, RW_READER);
543 if (dn->dn_datablkshift) {
544 int blkshift = dn->dn_datablkshift;
545 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
546 P2ALIGN(offset, 1<<blkshift)) >> blkshift;
548 nblks = (offset < dn->dn_datablksz);
552 blkid = dbuf_whichblock(dn, offset);
553 for (i = 0; i < nblks; i++)
554 dbuf_prefetch(dn, blkid+i);
557 rw_exit(&dn->dn_struct_rwlock);
559 dnode_rele(dn, FTAG);
563 * Get the next "chunk" of file data to free. We traverse the file from
564 * the end so that the file gets shorter over time (if we crashes in the
565 * middle, this will leave us in a better state). We find allocated file
566 * data by simply searching the allocated level 1 indirects.
569 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
571 uint64_t len = *start - limit;
573 uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
575 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
577 ASSERT(limit <= *start);
579 if (len <= iblkrange * maxblks) {
583 ASSERT(ISP2(iblkrange));
585 while (*start > limit && blkcnt < maxblks) {
588 /* find next allocated L1 indirect */
589 err = dnode_next_offset(dn,
590 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
592 /* if there are no more, then we are done */
601 /* reset offset to end of "next" block back */
602 *start = P2ALIGN(*start, iblkrange);
612 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
613 uint64_t length, boolean_t free_dnode)
616 uint64_t object_size, start, end, len;
617 boolean_t trunc = (length == DMU_OBJECT_END);
620 align = 1 << dn->dn_datablkshift;
622 object_size = align == 1 ? dn->dn_datablksz :
623 (dn->dn_maxblkid + 1) << dn->dn_datablkshift;
625 end = offset + length;
626 if (trunc || end > object_size)
630 length = end - offset;
634 /* assert(offset <= start) */
635 err = get_next_chunk(dn, &start, offset);
638 len = trunc ? DMU_OBJECT_END : end - start;
640 tx = dmu_tx_create(os);
641 dmu_tx_hold_free(tx, dn->dn_object, start, len);
642 err = dmu_tx_assign(tx, TXG_WAIT);
648 dnode_free_range(dn, start, trunc ? -1 : len, tx);
650 if (start == 0 && free_dnode) {
655 length -= end - start;
664 dmu_free_long_range(objset_t *os, uint64_t object,
665 uint64_t offset, uint64_t length)
670 err = dnode_hold(os, object, FTAG, &dn);
673 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
674 dnode_rele(dn, FTAG);
679 dmu_free_object(objset_t *os, uint64_t object)
685 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
689 if (dn->dn_nlevels == 1) {
690 tx = dmu_tx_create(os);
691 dmu_tx_hold_bonus(tx, object);
692 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END);
693 err = dmu_tx_assign(tx, TXG_WAIT);
695 dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
702 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
704 dnode_rele(dn, FTAG);
709 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
710 uint64_t size, dmu_tx_t *tx)
713 int err = dnode_hold(os, object, FTAG, &dn);
716 ASSERT(offset < UINT64_MAX);
717 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
718 dnode_free_range(dn, offset, size, tx);
719 dnode_rele(dn, FTAG);
724 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
725 void *buf, uint32_t flags)
731 err = dnode_hold(os, object, FTAG, &dn);
736 * Deal with odd block sizes, where there can't be data past the first
737 * block. If we ever do the tail block optimization, we will need to
738 * handle that here as well.
740 if (dn->dn_maxblkid == 0) {
741 int newsz = offset > dn->dn_datablksz ? 0 :
742 MIN(size, dn->dn_datablksz - offset);
743 bzero((char *)buf + newsz, size - newsz);
748 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
752 * NB: we could do this block-at-a-time, but it's nice
753 * to be reading in parallel.
755 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
756 TRUE, FTAG, &numbufs, &dbp, flags);
760 for (i = 0; i < numbufs; i++) {
763 dmu_buf_t *db = dbp[i];
767 bufoff = offset - db->db_offset;
768 tocpy = (int)MIN(db->db_size - bufoff, size);
770 bcopy((char *)db->db_data + bufoff, buf, tocpy);
774 buf = (char *)buf + tocpy;
776 dmu_buf_rele_array(dbp, numbufs, FTAG);
778 dnode_rele(dn, FTAG);
783 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
784 const void *buf, dmu_tx_t *tx)
792 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
793 FALSE, FTAG, &numbufs, &dbp));
795 for (i = 0; i < numbufs; i++) {
798 dmu_buf_t *db = dbp[i];
802 bufoff = offset - db->db_offset;
803 tocpy = (int)MIN(db->db_size - bufoff, size);
805 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
807 if (tocpy == db->db_size)
808 dmu_buf_will_fill(db, tx);
810 dmu_buf_will_dirty(db, tx);
812 bcopy(buf, (char *)db->db_data + bufoff, tocpy);
814 if (tocpy == db->db_size)
815 dmu_buf_fill_done(db, tx);
819 buf = (char *)buf + tocpy;
821 dmu_buf_rele_array(dbp, numbufs, FTAG);
825 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
834 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
835 FALSE, FTAG, &numbufs, &dbp));
837 for (i = 0; i < numbufs; i++) {
838 dmu_buf_t *db = dbp[i];
840 dmu_buf_will_not_fill(db, tx);
842 dmu_buf_rele_array(dbp, numbufs, FTAG);
846 * DMU support for xuio
848 kstat_t *xuio_ksp = NULL;
851 dmu_xuio_init(xuio_t *xuio, int nblk)
854 uio_t *uio = &xuio->xu_uio;
856 uio->uio_iovcnt = nblk;
857 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
859 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
861 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
862 priv->iovp = uio->uio_iov;
863 XUIO_XUZC_PRIV(xuio) = priv;
865 if (XUIO_XUZC_RW(xuio) == UIO_READ)
866 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
868 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
874 dmu_xuio_fini(xuio_t *xuio)
876 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
877 int nblk = priv->cnt;
879 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
880 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
881 kmem_free(priv, sizeof (dmu_xuio_t));
883 if (XUIO_XUZC_RW(xuio) == UIO_READ)
884 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
886 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
890 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
891 * and increase priv->next by 1.
894 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
897 uio_t *uio = &xuio->xu_uio;
898 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
899 int i = priv->next++;
901 ASSERT(i < priv->cnt);
902 ASSERT(off + n <= arc_buf_size(abuf));
903 iov = uio->uio_iov + i;
904 iov->iov_base = (char *)abuf->b_data + off;
906 priv->bufs[i] = abuf;
911 dmu_xuio_cnt(xuio_t *xuio)
913 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
918 dmu_xuio_arcbuf(xuio_t *xuio, int i)
920 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
922 ASSERT(i < priv->cnt);
923 return (priv->bufs[i]);
927 dmu_xuio_clear(xuio_t *xuio, int i)
929 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
931 ASSERT(i < priv->cnt);
932 priv->bufs[i] = NULL;
938 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
939 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
941 if (xuio_ksp != NULL) {
942 xuio_ksp->ks_data = &xuio_stats;
943 kstat_install(xuio_ksp);
950 if (xuio_ksp != NULL) {
951 kstat_delete(xuio_ksp);
957 xuio_stat_wbuf_copied()
959 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
963 xuio_stat_wbuf_nocopy()
965 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
970 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
977 * NB: we could do this block-at-a-time, but it's nice
978 * to be reading in parallel.
980 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
986 if (uio->uio_extflg == UIO_XUIO)
987 xuio = (xuio_t *)uio;
990 for (i = 0; i < numbufs; i++) {
993 dmu_buf_t *db = dbp[i];
997 bufoff = uio->uio_loffset - db->db_offset;
998 tocpy = (int)MIN(db->db_size - bufoff, size);
1001 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1002 arc_buf_t *dbuf_abuf = dbi->db_buf;
1003 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1004 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1006 uio->uio_resid -= tocpy;
1007 uio->uio_loffset += tocpy;
1010 if (abuf == dbuf_abuf)
1011 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1013 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1015 err = uiomove((char *)db->db_data + bufoff, tocpy,
1023 dmu_buf_rele_array(dbp, numbufs, FTAG);
1029 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1036 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1037 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1041 for (i = 0; i < numbufs; i++) {
1044 dmu_buf_t *db = dbp[i];
1048 bufoff = uio->uio_loffset - db->db_offset;
1049 tocpy = (int)MIN(db->db_size - bufoff, size);
1051 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1053 if (tocpy == db->db_size)
1054 dmu_buf_will_fill(db, tx);
1056 dmu_buf_will_dirty(db, tx);
1059 * XXX uiomove could block forever (eg. nfs-backed
1060 * pages). There needs to be a uiolockdown() function
1061 * to lock the pages in memory, so that uiomove won't
1064 err = uiomove((char *)db->db_data + bufoff, tocpy,
1067 if (tocpy == db->db_size)
1068 dmu_buf_fill_done(db, tx);
1076 dmu_buf_rele_array(dbp, numbufs, FTAG);
1081 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1084 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1093 err = dmu_write_uio_dnode(dn, uio, size, tx);
1100 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1109 err = dnode_hold(os, object, FTAG, &dn);
1113 err = dmu_write_uio_dnode(dn, uio, size, tx);
1115 dnode_rele(dn, FTAG);
1122 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1123 page_t *pp, dmu_tx_t *tx)
1132 err = dmu_buf_hold_array(os, object, offset, size,
1133 FALSE, FTAG, &numbufs, &dbp);
1137 for (i = 0; i < numbufs; i++) {
1138 int tocpy, copied, thiscpy;
1140 dmu_buf_t *db = dbp[i];
1144 ASSERT3U(db->db_size, >=, PAGESIZE);
1146 bufoff = offset - db->db_offset;
1147 tocpy = (int)MIN(db->db_size - bufoff, size);
1149 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1151 if (tocpy == db->db_size)
1152 dmu_buf_will_fill(db, tx);
1154 dmu_buf_will_dirty(db, tx);
1156 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1157 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
1158 thiscpy = MIN(PAGESIZE, tocpy - copied);
1159 va = zfs_map_page(pp, S_READ);
1160 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1161 zfs_unmap_page(pp, va);
1166 if (tocpy == db->db_size)
1167 dmu_buf_fill_done(db, tx);
1172 dmu_buf_rele_array(dbp, numbufs, FTAG);
1179 * Allocate a loaned anonymous arc buffer.
1182 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1184 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1187 DB_GET_SPA(&spa, db);
1188 return (arc_loan_buf(spa, size));
1192 * Free a loaned arc buffer.
1195 dmu_return_arcbuf(arc_buf_t *buf)
1197 arc_return_buf(buf, FTAG);
1198 VERIFY(arc_buf_remove_ref(buf, FTAG) == 1);
1202 * When possible directly assign passed loaned arc buffer to a dbuf.
1203 * If this is not possible copy the contents of passed arc buf via
1207 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1210 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1213 uint32_t blksz = (uint32_t)arc_buf_size(buf);
1216 DB_DNODE_ENTER(dbuf);
1217 dn = DB_DNODE(dbuf);
1218 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1219 blkid = dbuf_whichblock(dn, offset);
1220 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1221 rw_exit(&dn->dn_struct_rwlock);
1222 DB_DNODE_EXIT(dbuf);
1224 if (offset == db->db.db_offset && blksz == db->db.db_size) {
1225 dbuf_assign_arcbuf(db, buf, tx);
1226 dbuf_rele(db, FTAG);
1231 DB_DNODE_ENTER(dbuf);
1232 dn = DB_DNODE(dbuf);
1234 object = dn->dn_object;
1235 DB_DNODE_EXIT(dbuf);
1237 dbuf_rele(db, FTAG);
1238 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1239 dmu_return_arcbuf(buf);
1240 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1245 dbuf_dirty_record_t *dsa_dr;
1246 dmu_sync_cb_t *dsa_done;
1253 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1255 dmu_sync_arg_t *dsa = varg;
1256 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1257 blkptr_t *bp = zio->io_bp;
1259 if (zio->io_error == 0) {
1260 if (BP_IS_HOLE(bp)) {
1262 * A block of zeros may compress to a hole, but the
1263 * block size still needs to be known for replay.
1265 BP_SET_LSIZE(bp, db->db_size);
1267 ASSERT(BP_GET_LEVEL(bp) == 0);
1274 dmu_sync_late_arrival_ready(zio_t *zio)
1276 dmu_sync_ready(zio, NULL, zio->io_private);
1281 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1283 dmu_sync_arg_t *dsa = varg;
1284 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1285 dmu_buf_impl_t *db = dr->dr_dbuf;
1287 mutex_enter(&db->db_mtx);
1288 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1289 if (zio->io_error == 0) {
1290 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1291 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1292 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1293 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
1294 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1296 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1298 cv_broadcast(&db->db_changed);
1299 mutex_exit(&db->db_mtx);
1301 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1303 kmem_free(dsa, sizeof (*dsa));
1307 dmu_sync_late_arrival_done(zio_t *zio)
1309 blkptr_t *bp = zio->io_bp;
1310 dmu_sync_arg_t *dsa = zio->io_private;
1312 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1313 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1314 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1315 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1318 dmu_tx_commit(dsa->dsa_tx);
1320 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1322 kmem_free(dsa, sizeof (*dsa));
1326 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1327 zio_prop_t *zp, zbookmark_t *zb)
1329 dmu_sync_arg_t *dsa;
1332 tx = dmu_tx_create(os);
1333 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1334 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1336 return (EIO); /* Make zl_get_data do txg_waited_synced() */
1339 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1341 dsa->dsa_done = done;
1345 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1346 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
1347 dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa,
1348 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
1354 * Intent log support: sync the block associated with db to disk.
1355 * N.B. and XXX: the caller is responsible for making sure that the
1356 * data isn't changing while dmu_sync() is writing it.
1360 * EEXIST: this txg has already been synced, so there's nothing to to.
1361 * The caller should not log the write.
1363 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1364 * The caller should not log the write.
1366 * EALREADY: this block is already in the process of being synced.
1367 * The caller should track its progress (somehow).
1369 * EIO: could not do the I/O.
1370 * The caller should do a txg_wait_synced().
1372 * 0: the I/O has been initiated.
1373 * The caller should log this blkptr in the done callback.
1374 * It is possible that the I/O will fail, in which case
1375 * the error will be reported to the done callback and
1376 * propagated to pio from zio_done().
1379 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1381 blkptr_t *bp = zgd->zgd_bp;
1382 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1383 objset_t *os = db->db_objset;
1384 dsl_dataset_t *ds = os->os_dsl_dataset;
1385 dbuf_dirty_record_t *dr;
1386 dmu_sync_arg_t *dsa;
1391 ASSERT(pio != NULL);
1392 ASSERT(BP_IS_HOLE(bp));
1395 SET_BOOKMARK(&zb, ds->ds_object,
1396 db->db.db_object, db->db_level, db->db_blkid);
1400 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1404 * If we're frozen (running ziltest), we always need to generate a bp.
1406 if (txg > spa_freeze_txg(os->os_spa))
1407 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1410 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1411 * and us. If we determine that this txg is not yet syncing,
1412 * but it begins to sync a moment later, that's OK because the
1413 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1415 mutex_enter(&db->db_mtx);
1417 if (txg <= spa_last_synced_txg(os->os_spa)) {
1419 * This txg has already synced. There's nothing to do.
1421 mutex_exit(&db->db_mtx);
1425 if (txg <= spa_syncing_txg(os->os_spa)) {
1427 * This txg is currently syncing, so we can't mess with
1428 * the dirty record anymore; just write a new log block.
1430 mutex_exit(&db->db_mtx);
1431 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1434 dr = db->db_last_dirty;
1435 while (dr && dr->dr_txg != txg)
1440 * There's no dr for this dbuf, so it must have been freed.
1441 * There's no need to log writes to freed blocks, so we're done.
1443 mutex_exit(&db->db_mtx);
1447 ASSERT(dr->dr_txg == txg);
1448 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1449 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1451 * We have already issued a sync write for this buffer,
1452 * or this buffer has already been synced. It could not
1453 * have been dirtied since, or we would have cleared the state.
1455 mutex_exit(&db->db_mtx);
1459 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1460 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1461 mutex_exit(&db->db_mtx);
1463 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1465 dsa->dsa_done = done;
1469 zio_nowait(arc_write(pio, os->os_spa, txg,
1470 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp,
1471 dmu_sync_ready, dmu_sync_done, dsa,
1472 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
1478 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1484 err = dnode_hold(os, object, FTAG, &dn);
1487 err = dnode_set_blksz(dn, size, ibs, tx);
1488 dnode_rele(dn, FTAG);
1493 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1498 /* XXX assumes dnode_hold will not get an i/o error */
1499 (void) dnode_hold(os, object, FTAG, &dn);
1500 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
1501 dn->dn_checksum = checksum;
1502 dnode_setdirty(dn, tx);
1503 dnode_rele(dn, FTAG);
1507 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1512 /* XXX assumes dnode_hold will not get an i/o error */
1513 (void) dnode_hold(os, object, FTAG, &dn);
1514 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
1515 dn->dn_compress = compress;
1516 dnode_setdirty(dn, tx);
1517 dnode_rele(dn, FTAG);
1520 int zfs_mdcomp_disable = 0;
1521 TUNABLE_INT("vfs.zfs.mdcomp_disable", &zfs_mdcomp_disable);
1522 SYSCTL_DECL(_vfs_zfs);
1523 SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RW,
1524 &zfs_mdcomp_disable, 0, "Disable metadata compression");
1527 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1529 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1530 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1532 enum zio_checksum checksum = os->os_checksum;
1533 enum zio_compress compress = os->os_compress;
1534 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1536 boolean_t dedup_verify = os->os_dedup_verify;
1537 int copies = os->os_copies;
1540 * Determine checksum setting.
1544 * Metadata always gets checksummed. If the data
1545 * checksum is multi-bit correctable, and it's not a
1546 * ZBT-style checksum, then it's suitable for metadata
1547 * as well. Otherwise, the metadata checksum defaults
1550 if (zio_checksum_table[checksum].ci_correctable < 1 ||
1551 zio_checksum_table[checksum].ci_eck)
1552 checksum = ZIO_CHECKSUM_FLETCHER_4;
1554 checksum = zio_checksum_select(dn->dn_checksum, checksum);
1558 * Determine compression setting.
1562 * XXX -- we should design a compression algorithm
1563 * that specializes in arrays of bps.
1565 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
1568 compress = zio_compress_select(dn->dn_compress, compress);
1572 * Determine dedup setting. If we are in dmu_sync(), we won't
1573 * actually dedup now because that's all done in syncing context;
1574 * but we do want to use the dedup checkum. If the checksum is not
1575 * strong enough to ensure unique signatures, force dedup_verify.
1577 dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF);
1579 checksum = dedup_checksum;
1580 if (!zio_checksum_table[checksum].ci_dedup)
1584 if (wp & WP_DMU_SYNC)
1587 if (wp & WP_NOFILL) {
1588 ASSERT(!ismd && level == 0);
1589 checksum = ZIO_CHECKSUM_OFF;
1590 compress = ZIO_COMPRESS_OFF;
1594 zp->zp_checksum = checksum;
1595 zp->zp_compress = compress;
1596 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
1597 zp->zp_level = level;
1598 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa));
1599 zp->zp_dedup = dedup;
1600 zp->zp_dedup_verify = dedup && dedup_verify;
1604 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1609 err = dnode_hold(os, object, FTAG, &dn);
1613 * Sync any current changes before
1614 * we go trundling through the block pointers.
1616 for (i = 0; i < TXG_SIZE; i++) {
1617 if (list_link_active(&dn->dn_dirty_link[i]))
1620 if (i != TXG_SIZE) {
1621 dnode_rele(dn, FTAG);
1622 txg_wait_synced(dmu_objset_pool(os), 0);
1623 err = dnode_hold(os, object, FTAG, &dn);
1628 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1629 dnode_rele(dn, FTAG);
1635 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1639 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1640 mutex_enter(&dn->dn_mtx);
1644 doi->doi_data_block_size = dn->dn_datablksz;
1645 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1646 1ULL << dn->dn_indblkshift : 0;
1647 doi->doi_type = dn->dn_type;
1648 doi->doi_bonus_type = dn->dn_bonustype;
1649 doi->doi_bonus_size = dn->dn_bonuslen;
1650 doi->doi_indirection = dn->dn_nlevels;
1651 doi->doi_checksum = dn->dn_checksum;
1652 doi->doi_compress = dn->dn_compress;
1653 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1654 doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz;
1655 doi->doi_fill_count = 0;
1656 for (int i = 0; i < dnp->dn_nblkptr; i++)
1657 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
1659 mutex_exit(&dn->dn_mtx);
1660 rw_exit(&dn->dn_struct_rwlock);
1664 * Get information on a DMU object.
1665 * If doi is NULL, just indicates whether the object exists.
1668 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1671 int err = dnode_hold(os, object, FTAG, &dn);
1677 dmu_object_info_from_dnode(dn, doi);
1679 dnode_rele(dn, FTAG);
1684 * As above, but faster; can be used when you have a held dbuf in hand.
1687 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1689 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1692 dmu_object_info_from_dnode(DB_DNODE(db), doi);
1697 * Faster still when you only care about the size.
1698 * This is specifically optimized for zfs_getattr().
1701 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1702 u_longlong_t *nblk512)
1704 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1710 *blksize = dn->dn_datablksz;
1711 /* add 1 for dnode space */
1712 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1713 SPA_MINBLOCKSHIFT) + 1;
1718 byteswap_uint64_array(void *vbuf, size_t size)
1720 uint64_t *buf = vbuf;
1721 size_t count = size >> 3;
1724 ASSERT((size & 7) == 0);
1726 for (i = 0; i < count; i++)
1727 buf[i] = BSWAP_64(buf[i]);
1731 byteswap_uint32_array(void *vbuf, size_t size)
1733 uint32_t *buf = vbuf;
1734 size_t count = size >> 2;
1737 ASSERT((size & 3) == 0);
1739 for (i = 0; i < count; i++)
1740 buf[i] = BSWAP_32(buf[i]);
1744 byteswap_uint16_array(void *vbuf, size_t size)
1746 uint16_t *buf = vbuf;
1747 size_t count = size >> 1;
1750 ASSERT((size & 1) == 0);
1752 for (i = 0; i < count; i++)
1753 buf[i] = BSWAP_16(buf[i]);
1758 byteswap_uint8_array(void *vbuf, size_t size)