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) 2013 by Delphix. All rights reserved.
26 /* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
32 #include <sys/dnode.h>
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_pool.h>
39 #include <sys/dsl_synctask.h>
40 #include <sys/dsl_prop.h>
41 #include <sys/dmu_zfetch.h>
42 #include <sys/zfs_ioctl.h>
44 #include <sys/zio_checksum.h>
45 #include <sys/zio_compress.h>
48 #include <sys/zfs_znode.h>
52 * Enable/disable nopwrite feature.
54 int zfs_nopwrite_enabled = 1;
55 SYSCTL_DECL(_vfs_zfs);
56 TUNABLE_INT("vfs.zfs.nopwrite_enabled", &zfs_nopwrite_enabled);
57 SYSCTL_INT(_vfs_zfs, OID_AUTO, nopwrite_enabled, CTLFLAG_RDTUN,
58 &zfs_nopwrite_enabled, 0, "Enable nopwrite feature");
60 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
61 { DMU_BSWAP_UINT8, TRUE, "unallocated" },
62 { DMU_BSWAP_ZAP, TRUE, "object directory" },
63 { DMU_BSWAP_UINT64, TRUE, "object array" },
64 { DMU_BSWAP_UINT8, TRUE, "packed nvlist" },
65 { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" },
66 { DMU_BSWAP_UINT64, TRUE, "bpobj" },
67 { DMU_BSWAP_UINT64, TRUE, "bpobj header" },
68 { DMU_BSWAP_UINT64, TRUE, "SPA space map header" },
69 { DMU_BSWAP_UINT64, TRUE, "SPA space map" },
70 { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" },
71 { DMU_BSWAP_DNODE, TRUE, "DMU dnode" },
72 { DMU_BSWAP_OBJSET, TRUE, "DMU objset" },
73 { DMU_BSWAP_UINT64, TRUE, "DSL directory" },
74 { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"},
75 { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" },
76 { DMU_BSWAP_ZAP, TRUE, "DSL props" },
77 { DMU_BSWAP_UINT64, TRUE, "DSL dataset" },
78 { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" },
79 { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" },
80 { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" },
81 { DMU_BSWAP_ZAP, TRUE, "ZFS directory" },
82 { DMU_BSWAP_ZAP, TRUE, "ZFS master node" },
83 { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" },
84 { DMU_BSWAP_UINT8, FALSE, "zvol object" },
85 { DMU_BSWAP_ZAP, TRUE, "zvol prop" },
86 { DMU_BSWAP_UINT8, FALSE, "other uint8[]" },
87 { DMU_BSWAP_UINT64, FALSE, "other uint64[]" },
88 { DMU_BSWAP_ZAP, TRUE, "other ZAP" },
89 { DMU_BSWAP_ZAP, TRUE, "persistent error log" },
90 { DMU_BSWAP_UINT8, TRUE, "SPA history" },
91 { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" },
92 { DMU_BSWAP_ZAP, TRUE, "Pool properties" },
93 { DMU_BSWAP_ZAP, TRUE, "DSL permissions" },
94 { DMU_BSWAP_ACL, TRUE, "ZFS ACL" },
95 { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" },
96 { DMU_BSWAP_UINT8, TRUE, "FUID table" },
97 { DMU_BSWAP_UINT64, TRUE, "FUID table size" },
98 { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"},
99 { DMU_BSWAP_ZAP, TRUE, "scan work queue" },
100 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" },
101 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" },
102 { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"},
103 { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" },
104 { DMU_BSWAP_ZAP, TRUE, "DDT statistics" },
105 { DMU_BSWAP_UINT8, TRUE, "System attributes" },
106 { DMU_BSWAP_ZAP, TRUE, "SA master node" },
107 { DMU_BSWAP_ZAP, TRUE, "SA attr registration" },
108 { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" },
109 { DMU_BSWAP_ZAP, TRUE, "scan translations" },
110 { DMU_BSWAP_UINT8, FALSE, "deduplicated block" },
111 { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" },
112 { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" },
113 { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" },
114 { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" }
117 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
118 { byteswap_uint8_array, "uint8" },
119 { byteswap_uint16_array, "uint16" },
120 { byteswap_uint32_array, "uint32" },
121 { byteswap_uint64_array, "uint64" },
122 { zap_byteswap, "zap" },
123 { dnode_buf_byteswap, "dnode" },
124 { dmu_objset_byteswap, "objset" },
125 { zfs_znode_byteswap, "znode" },
126 { zfs_oldacl_byteswap, "oldacl" },
127 { zfs_acl_byteswap, "acl" }
131 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
132 void *tag, dmu_buf_t **dbp, int flags)
138 int db_flags = DB_RF_CANFAIL;
140 if (flags & DMU_READ_NO_PREFETCH)
141 db_flags |= DB_RF_NOPREFETCH;
143 err = dnode_hold(os, object, FTAG, &dn);
146 blkid = dbuf_whichblock(dn, offset);
147 rw_enter(&dn->dn_struct_rwlock, RW_READER);
148 db = dbuf_hold(dn, blkid, tag);
149 rw_exit(&dn->dn_struct_rwlock);
151 err = SET_ERROR(EIO);
153 err = dbuf_read(db, NULL, db_flags);
160 dnode_rele(dn, FTAG);
161 *dbp = &db->db; /* NULL db plus first field offset is NULL */
168 return (DN_MAX_BONUSLEN);
172 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
174 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
181 if (dn->dn_bonus != db) {
182 error = SET_ERROR(EINVAL);
183 } else if (newsize < 0 || newsize > db_fake->db_size) {
184 error = SET_ERROR(EINVAL);
186 dnode_setbonuslen(dn, newsize, tx);
195 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
197 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
204 if (!DMU_OT_IS_VALID(type)) {
205 error = SET_ERROR(EINVAL);
206 } else if (dn->dn_bonus != db) {
207 error = SET_ERROR(EINVAL);
209 dnode_setbonus_type(dn, type, tx);
218 dmu_get_bonustype(dmu_buf_t *db_fake)
220 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
222 dmu_object_type_t type;
226 type = dn->dn_bonustype;
233 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
238 error = dnode_hold(os, object, FTAG, &dn);
239 dbuf_rm_spill(dn, tx);
240 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
241 dnode_rm_spill(dn, tx);
242 rw_exit(&dn->dn_struct_rwlock);
243 dnode_rele(dn, FTAG);
248 * returns ENOENT, EIO, or 0.
251 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
257 error = dnode_hold(os, object, FTAG, &dn);
261 rw_enter(&dn->dn_struct_rwlock, RW_READER);
262 if (dn->dn_bonus == NULL) {
263 rw_exit(&dn->dn_struct_rwlock);
264 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
265 if (dn->dn_bonus == NULL)
266 dbuf_create_bonus(dn);
270 /* as long as the bonus buf is held, the dnode will be held */
271 if (refcount_add(&db->db_holds, tag) == 1) {
272 VERIFY(dnode_add_ref(dn, db));
273 (void) atomic_inc_32_nv(&dn->dn_dbufs_count);
277 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
278 * hold and incrementing the dbuf count to ensure that dnode_move() sees
279 * a dnode hold for every dbuf.
281 rw_exit(&dn->dn_struct_rwlock);
283 dnode_rele(dn, FTAG);
285 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
292 * returns ENOENT, EIO, or 0.
294 * This interface will allocate a blank spill dbuf when a spill blk
295 * doesn't already exist on the dnode.
297 * if you only want to find an already existing spill db, then
298 * dmu_spill_hold_existing() should be used.
301 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
303 dmu_buf_impl_t *db = NULL;
306 if ((flags & DB_RF_HAVESTRUCT) == 0)
307 rw_enter(&dn->dn_struct_rwlock, RW_READER);
309 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
311 if ((flags & DB_RF_HAVESTRUCT) == 0)
312 rw_exit(&dn->dn_struct_rwlock);
315 err = dbuf_read(db, NULL, flags);
324 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
326 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
333 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
334 err = SET_ERROR(EINVAL);
336 rw_enter(&dn->dn_struct_rwlock, RW_READER);
338 if (!dn->dn_have_spill) {
339 err = SET_ERROR(ENOENT);
341 err = dmu_spill_hold_by_dnode(dn,
342 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
345 rw_exit(&dn->dn_struct_rwlock);
353 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
355 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
361 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
368 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
369 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
370 * and can induce severe lock contention when writing to several files
371 * whose dnodes are in the same block.
374 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
375 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
377 dsl_pool_t *dp = NULL;
379 uint64_t blkid, nblks, i;
385 ASSERT(length <= DMU_MAX_ACCESS);
387 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
388 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
389 dbuf_flags |= DB_RF_NOPREFETCH;
391 rw_enter(&dn->dn_struct_rwlock, RW_READER);
392 if (dn->dn_datablkshift) {
393 int blkshift = dn->dn_datablkshift;
394 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
395 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
397 if (offset + length > dn->dn_datablksz) {
398 zfs_panic_recover("zfs: accessing past end of object "
399 "%llx/%llx (size=%u access=%llu+%llu)",
400 (longlong_t)dn->dn_objset->
401 os_dsl_dataset->ds_object,
402 (longlong_t)dn->dn_object, dn->dn_datablksz,
403 (longlong_t)offset, (longlong_t)length);
404 rw_exit(&dn->dn_struct_rwlock);
405 return (SET_ERROR(EIO));
409 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
411 if (dn->dn_objset->os_dsl_dataset)
412 dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
414 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
415 blkid = dbuf_whichblock(dn, offset);
416 for (i = 0; i < nblks; i++) {
417 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
419 rw_exit(&dn->dn_struct_rwlock);
420 dmu_buf_rele_array(dbp, nblks, tag);
422 return (SET_ERROR(EIO));
424 /* initiate async i/o */
426 (void) dbuf_read(db, zio, dbuf_flags);
429 curthread->td_ru.ru_oublock++;
433 rw_exit(&dn->dn_struct_rwlock);
435 /* wait for async i/o */
437 /* track read overhead when we are in sync context */
438 if (dp && dsl_pool_sync_context(dp))
439 dp->dp_read_overhead += gethrtime() - start;
441 dmu_buf_rele_array(dbp, nblks, tag);
445 /* wait for other io to complete */
447 for (i = 0; i < nblks; i++) {
448 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
449 mutex_enter(&db->db_mtx);
450 while (db->db_state == DB_READ ||
451 db->db_state == DB_FILL)
452 cv_wait(&db->db_changed, &db->db_mtx);
453 if (db->db_state == DB_UNCACHED)
454 err = SET_ERROR(EIO);
455 mutex_exit(&db->db_mtx);
457 dmu_buf_rele_array(dbp, nblks, tag);
469 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
470 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
475 err = dnode_hold(os, object, FTAG, &dn);
479 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
480 numbufsp, dbpp, DMU_READ_PREFETCH);
482 dnode_rele(dn, FTAG);
488 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
489 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
491 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
497 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
498 numbufsp, dbpp, DMU_READ_PREFETCH);
505 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
508 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
513 for (i = 0; i < numbufs; i++) {
515 dbuf_rele(dbp[i], tag);
518 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
522 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
528 if (zfs_prefetch_disable)
531 if (len == 0) { /* they're interested in the bonus buffer */
532 dn = DMU_META_DNODE(os);
534 if (object == 0 || object >= DN_MAX_OBJECT)
537 rw_enter(&dn->dn_struct_rwlock, RW_READER);
538 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
539 dbuf_prefetch(dn, blkid);
540 rw_exit(&dn->dn_struct_rwlock);
545 * XXX - Note, if the dnode for the requested object is not
546 * already cached, we will do a *synchronous* read in the
547 * dnode_hold() call. The same is true for any indirects.
549 err = dnode_hold(os, object, FTAG, &dn);
553 rw_enter(&dn->dn_struct_rwlock, RW_READER);
554 if (dn->dn_datablkshift) {
555 int blkshift = dn->dn_datablkshift;
556 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
557 P2ALIGN(offset, 1<<blkshift)) >> blkshift;
559 nblks = (offset < dn->dn_datablksz);
563 blkid = dbuf_whichblock(dn, offset);
564 for (i = 0; i < nblks; i++)
565 dbuf_prefetch(dn, blkid+i);
568 rw_exit(&dn->dn_struct_rwlock);
570 dnode_rele(dn, FTAG);
574 * Get the next "chunk" of file data to free. We traverse the file from
575 * the end so that the file gets shorter over time (if we crashes in the
576 * middle, this will leave us in a better state). We find allocated file
577 * data by simply searching the allocated level 1 indirects.
580 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
582 uint64_t len = *start - limit;
584 uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
586 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
588 ASSERT(limit <= *start);
590 if (len <= iblkrange * maxblks) {
594 ASSERT(ISP2(iblkrange));
596 while (*start > limit && blkcnt < maxblks) {
599 /* find next allocated L1 indirect */
600 err = dnode_next_offset(dn,
601 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
603 /* if there are no more, then we are done */
612 /* reset offset to end of "next" block back */
613 *start = P2ALIGN(*start, iblkrange);
623 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
624 uint64_t length, boolean_t free_dnode)
627 uint64_t object_size, start, end, len;
628 boolean_t trunc = (length == DMU_OBJECT_END);
631 align = 1 << dn->dn_datablkshift;
633 object_size = align == 1 ? dn->dn_datablksz :
634 (dn->dn_maxblkid + 1) << dn->dn_datablkshift;
636 end = offset + length;
637 if (trunc || end > object_size)
641 length = end - offset;
645 /* assert(offset <= start) */
646 err = get_next_chunk(dn, &start, offset);
649 len = trunc ? DMU_OBJECT_END : end - start;
651 tx = dmu_tx_create(os);
652 dmu_tx_hold_free(tx, dn->dn_object, start, len);
653 err = dmu_tx_assign(tx, TXG_WAIT);
659 dnode_free_range(dn, start, trunc ? -1 : len, tx);
661 if (start == 0 && free_dnode) {
666 length -= end - start;
675 dmu_free_long_range(objset_t *os, uint64_t object,
676 uint64_t offset, uint64_t length)
681 err = dnode_hold(os, object, FTAG, &dn);
684 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
685 dnode_rele(dn, FTAG);
690 dmu_free_object(objset_t *os, uint64_t object)
696 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
700 if (dn->dn_nlevels == 1) {
701 tx = dmu_tx_create(os);
702 dmu_tx_hold_bonus(tx, object);
703 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END);
704 err = dmu_tx_assign(tx, TXG_WAIT);
706 dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
713 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
715 dnode_rele(dn, FTAG);
720 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
721 uint64_t size, dmu_tx_t *tx)
724 int err = dnode_hold(os, object, FTAG, &dn);
727 ASSERT(offset < UINT64_MAX);
728 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
729 dnode_free_range(dn, offset, size, tx);
730 dnode_rele(dn, FTAG);
735 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
736 void *buf, uint32_t flags)
742 err = dnode_hold(os, object, FTAG, &dn);
747 * Deal with odd block sizes, where there can't be data past the first
748 * block. If we ever do the tail block optimization, we will need to
749 * handle that here as well.
751 if (dn->dn_maxblkid == 0) {
752 int newsz = offset > dn->dn_datablksz ? 0 :
753 MIN(size, dn->dn_datablksz - offset);
754 bzero((char *)buf + newsz, size - newsz);
759 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
763 * NB: we could do this block-at-a-time, but it's nice
764 * to be reading in parallel.
766 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
767 TRUE, FTAG, &numbufs, &dbp, flags);
771 for (i = 0; i < numbufs; i++) {
774 dmu_buf_t *db = dbp[i];
778 bufoff = offset - db->db_offset;
779 tocpy = (int)MIN(db->db_size - bufoff, size);
781 bcopy((char *)db->db_data + bufoff, buf, tocpy);
785 buf = (char *)buf + tocpy;
787 dmu_buf_rele_array(dbp, numbufs, FTAG);
789 dnode_rele(dn, FTAG);
794 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
795 const void *buf, dmu_tx_t *tx)
803 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
804 FALSE, FTAG, &numbufs, &dbp));
806 for (i = 0; i < numbufs; i++) {
809 dmu_buf_t *db = dbp[i];
813 bufoff = offset - db->db_offset;
814 tocpy = (int)MIN(db->db_size - bufoff, size);
816 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
818 if (tocpy == db->db_size)
819 dmu_buf_will_fill(db, tx);
821 dmu_buf_will_dirty(db, tx);
823 bcopy(buf, (char *)db->db_data + bufoff, tocpy);
825 if (tocpy == db->db_size)
826 dmu_buf_fill_done(db, tx);
830 buf = (char *)buf + tocpy;
832 dmu_buf_rele_array(dbp, numbufs, FTAG);
836 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
845 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
846 FALSE, FTAG, &numbufs, &dbp));
848 for (i = 0; i < numbufs; i++) {
849 dmu_buf_t *db = dbp[i];
851 dmu_buf_will_not_fill(db, tx);
853 dmu_buf_rele_array(dbp, numbufs, FTAG);
857 * DMU support for xuio
859 kstat_t *xuio_ksp = NULL;
862 dmu_xuio_init(xuio_t *xuio, int nblk)
865 uio_t *uio = &xuio->xu_uio;
867 uio->uio_iovcnt = nblk;
868 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
870 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
872 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
873 priv->iovp = uio->uio_iov;
874 XUIO_XUZC_PRIV(xuio) = priv;
876 if (XUIO_XUZC_RW(xuio) == UIO_READ)
877 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
879 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
885 dmu_xuio_fini(xuio_t *xuio)
887 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
888 int nblk = priv->cnt;
890 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
891 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
892 kmem_free(priv, sizeof (dmu_xuio_t));
894 if (XUIO_XUZC_RW(xuio) == UIO_READ)
895 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
897 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
901 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
902 * and increase priv->next by 1.
905 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
908 uio_t *uio = &xuio->xu_uio;
909 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
910 int i = priv->next++;
912 ASSERT(i < priv->cnt);
913 ASSERT(off + n <= arc_buf_size(abuf));
914 iov = uio->uio_iov + i;
915 iov->iov_base = (char *)abuf->b_data + off;
917 priv->bufs[i] = abuf;
922 dmu_xuio_cnt(xuio_t *xuio)
924 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
929 dmu_xuio_arcbuf(xuio_t *xuio, int i)
931 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
933 ASSERT(i < priv->cnt);
934 return (priv->bufs[i]);
938 dmu_xuio_clear(xuio_t *xuio, int i)
940 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
942 ASSERT(i < priv->cnt);
943 priv->bufs[i] = NULL;
949 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
950 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
952 if (xuio_ksp != NULL) {
953 xuio_ksp->ks_data = &xuio_stats;
954 kstat_install(xuio_ksp);
961 if (xuio_ksp != NULL) {
962 kstat_delete(xuio_ksp);
968 xuio_stat_wbuf_copied()
970 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
974 xuio_stat_wbuf_nocopy()
976 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
981 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
988 * NB: we could do this block-at-a-time, but it's nice
989 * to be reading in parallel.
991 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
997 if (uio->uio_extflg == UIO_XUIO)
998 xuio = (xuio_t *)uio;
1001 for (i = 0; i < numbufs; i++) {
1004 dmu_buf_t *db = dbp[i];
1008 bufoff = uio->uio_loffset - db->db_offset;
1009 tocpy = (int)MIN(db->db_size - bufoff, size);
1012 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1013 arc_buf_t *dbuf_abuf = dbi->db_buf;
1014 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1015 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1017 uio->uio_resid -= tocpy;
1018 uio->uio_loffset += tocpy;
1021 if (abuf == dbuf_abuf)
1022 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1024 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1026 err = uiomove((char *)db->db_data + bufoff, tocpy,
1034 dmu_buf_rele_array(dbp, numbufs, FTAG);
1040 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1047 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1048 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1052 for (i = 0; i < numbufs; i++) {
1055 dmu_buf_t *db = dbp[i];
1059 bufoff = uio->uio_loffset - db->db_offset;
1060 tocpy = (int)MIN(db->db_size - bufoff, size);
1062 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1064 if (tocpy == db->db_size)
1065 dmu_buf_will_fill(db, tx);
1067 dmu_buf_will_dirty(db, tx);
1070 * XXX uiomove could block forever (eg. nfs-backed
1071 * pages). There needs to be a uiolockdown() function
1072 * to lock the pages in memory, so that uiomove won't
1075 err = uiomove((char *)db->db_data + bufoff, tocpy,
1078 if (tocpy == db->db_size)
1079 dmu_buf_fill_done(db, tx);
1087 dmu_buf_rele_array(dbp, numbufs, FTAG);
1092 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1095 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1104 err = dmu_write_uio_dnode(dn, uio, size, tx);
1111 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1120 err = dnode_hold(os, object, FTAG, &dn);
1124 err = dmu_write_uio_dnode(dn, uio, size, tx);
1126 dnode_rele(dn, FTAG);
1133 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1134 page_t *pp, dmu_tx_t *tx)
1143 err = dmu_buf_hold_array(os, object, offset, size,
1144 FALSE, FTAG, &numbufs, &dbp);
1148 for (i = 0; i < numbufs; i++) {
1149 int tocpy, copied, thiscpy;
1151 dmu_buf_t *db = dbp[i];
1155 ASSERT3U(db->db_size, >=, PAGESIZE);
1157 bufoff = offset - db->db_offset;
1158 tocpy = (int)MIN(db->db_size - bufoff, size);
1160 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1162 if (tocpy == db->db_size)
1163 dmu_buf_will_fill(db, tx);
1165 dmu_buf_will_dirty(db, tx);
1167 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1168 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
1169 thiscpy = MIN(PAGESIZE, tocpy - copied);
1170 va = zfs_map_page(pp, S_READ);
1171 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1172 zfs_unmap_page(pp, va);
1177 if (tocpy == db->db_size)
1178 dmu_buf_fill_done(db, tx);
1183 dmu_buf_rele_array(dbp, numbufs, FTAG);
1190 * Allocate a loaned anonymous arc buffer.
1193 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1195 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1198 DB_GET_SPA(&spa, db);
1199 return (arc_loan_buf(spa, size));
1203 * Free a loaned arc buffer.
1206 dmu_return_arcbuf(arc_buf_t *buf)
1208 arc_return_buf(buf, FTAG);
1209 VERIFY(arc_buf_remove_ref(buf, FTAG));
1213 * When possible directly assign passed loaned arc buffer to a dbuf.
1214 * If this is not possible copy the contents of passed arc buf via
1218 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1221 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1224 uint32_t blksz = (uint32_t)arc_buf_size(buf);
1227 DB_DNODE_ENTER(dbuf);
1228 dn = DB_DNODE(dbuf);
1229 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1230 blkid = dbuf_whichblock(dn, offset);
1231 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1232 rw_exit(&dn->dn_struct_rwlock);
1233 DB_DNODE_EXIT(dbuf);
1235 if (offset == db->db.db_offset && blksz == db->db.db_size) {
1236 dbuf_assign_arcbuf(db, buf, tx);
1237 dbuf_rele(db, FTAG);
1242 DB_DNODE_ENTER(dbuf);
1243 dn = DB_DNODE(dbuf);
1245 object = dn->dn_object;
1246 DB_DNODE_EXIT(dbuf);
1248 dbuf_rele(db, FTAG);
1249 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1250 dmu_return_arcbuf(buf);
1251 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1256 dbuf_dirty_record_t *dsa_dr;
1257 dmu_sync_cb_t *dsa_done;
1264 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1266 dmu_sync_arg_t *dsa = varg;
1267 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1268 blkptr_t *bp = zio->io_bp;
1270 if (zio->io_error == 0) {
1271 if (BP_IS_HOLE(bp)) {
1273 * A block of zeros may compress to a hole, but the
1274 * block size still needs to be known for replay.
1276 BP_SET_LSIZE(bp, db->db_size);
1278 ASSERT(BP_GET_LEVEL(bp) == 0);
1285 dmu_sync_late_arrival_ready(zio_t *zio)
1287 dmu_sync_ready(zio, NULL, zio->io_private);
1292 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1294 dmu_sync_arg_t *dsa = varg;
1295 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1296 dmu_buf_impl_t *db = dr->dr_dbuf;
1298 mutex_enter(&db->db_mtx);
1299 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1300 if (zio->io_error == 0) {
1301 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1302 if (dr->dt.dl.dr_nopwrite) {
1303 blkptr_t *bp = zio->io_bp;
1304 blkptr_t *bp_orig = &zio->io_bp_orig;
1305 uint8_t chksum = BP_GET_CHECKSUM(bp_orig);
1307 ASSERT(BP_EQUAL(bp, bp_orig));
1308 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1309 ASSERT(zio_checksum_table[chksum].ci_dedup);
1311 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1312 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1313 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1314 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
1315 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1317 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1319 cv_broadcast(&db->db_changed);
1320 mutex_exit(&db->db_mtx);
1322 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1324 kmem_free(dsa, sizeof (*dsa));
1328 dmu_sync_late_arrival_done(zio_t *zio)
1330 blkptr_t *bp = zio->io_bp;
1331 dmu_sync_arg_t *dsa = zio->io_private;
1332 blkptr_t *bp_orig = &zio->io_bp_orig;
1334 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1336 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1337 * then there is nothing to do here. Otherwise, free the
1338 * newly allocated block in this txg.
1340 if (zio->io_flags & ZIO_FLAG_NOPWRITE) {
1341 ASSERT(BP_EQUAL(bp, bp_orig));
1343 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1344 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1345 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1346 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1350 dmu_tx_commit(dsa->dsa_tx);
1352 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1354 kmem_free(dsa, sizeof (*dsa));
1358 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1359 zio_prop_t *zp, zbookmark_t *zb)
1361 dmu_sync_arg_t *dsa;
1364 tx = dmu_tx_create(os);
1365 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1366 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1368 /* Make zl_get_data do txg_waited_synced() */
1369 return (SET_ERROR(EIO));
1372 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1374 dsa->dsa_done = done;
1378 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1379 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
1380 dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa,
1381 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
1387 * Intent log support: sync the block associated with db to disk.
1388 * N.B. and XXX: the caller is responsible for making sure that the
1389 * data isn't changing while dmu_sync() is writing it.
1393 * EEXIST: this txg has already been synced, so there's nothing to do.
1394 * The caller should not log the write.
1396 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1397 * The caller should not log the write.
1399 * EALREADY: this block is already in the process of being synced.
1400 * The caller should track its progress (somehow).
1402 * EIO: could not do the I/O.
1403 * The caller should do a txg_wait_synced().
1405 * 0: the I/O has been initiated.
1406 * The caller should log this blkptr in the done callback.
1407 * It is possible that the I/O will fail, in which case
1408 * the error will be reported to the done callback and
1409 * propagated to pio from zio_done().
1412 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1414 blkptr_t *bp = zgd->zgd_bp;
1415 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1416 objset_t *os = db->db_objset;
1417 dsl_dataset_t *ds = os->os_dsl_dataset;
1418 dbuf_dirty_record_t *dr;
1419 dmu_sync_arg_t *dsa;
1424 ASSERT(pio != NULL);
1427 SET_BOOKMARK(&zb, ds->ds_object,
1428 db->db.db_object, db->db_level, db->db_blkid);
1432 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1436 * If we're frozen (running ziltest), we always need to generate a bp.
1438 if (txg > spa_freeze_txg(os->os_spa))
1439 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1442 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1443 * and us. If we determine that this txg is not yet syncing,
1444 * but it begins to sync a moment later, that's OK because the
1445 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1447 mutex_enter(&db->db_mtx);
1449 if (txg <= spa_last_synced_txg(os->os_spa)) {
1451 * This txg has already synced. There's nothing to do.
1453 mutex_exit(&db->db_mtx);
1454 return (SET_ERROR(EEXIST));
1457 if (txg <= spa_syncing_txg(os->os_spa)) {
1459 * This txg is currently syncing, so we can't mess with
1460 * the dirty record anymore; just write a new log block.
1462 mutex_exit(&db->db_mtx);
1463 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1466 dr = db->db_last_dirty;
1467 while (dr && dr->dr_txg != txg)
1472 * There's no dr for this dbuf, so it must have been freed.
1473 * There's no need to log writes to freed blocks, so we're done.
1475 mutex_exit(&db->db_mtx);
1476 return (SET_ERROR(ENOENT));
1479 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
1482 * Assume the on-disk data is X, the current syncing data is Y,
1483 * and the current in-memory data is Z (currently in dmu_sync).
1484 * X and Z are identical but Y is has been modified. Normally,
1485 * when X and Z are the same we will perform a nopwrite but if Y
1486 * is different we must disable nopwrite since the resulting write
1487 * of Y to disk can free the block containing X. If we allowed a
1488 * nopwrite to occur the block pointing to Z would reference a freed
1489 * block. Since this is a rare case we simplify this by disabling
1490 * nopwrite if the current dmu_sync-ing dbuf has been modified in
1491 * a previous transaction.
1494 zp.zp_nopwrite = B_FALSE;
1496 ASSERT(dr->dr_txg == txg);
1497 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1498 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1500 * We have already issued a sync write for this buffer,
1501 * or this buffer has already been synced. It could not
1502 * have been dirtied since, or we would have cleared the state.
1504 mutex_exit(&db->db_mtx);
1505 return (SET_ERROR(EALREADY));
1508 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1509 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1510 mutex_exit(&db->db_mtx);
1512 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1514 dsa->dsa_done = done;
1518 zio_nowait(arc_write(pio, os->os_spa, txg,
1519 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
1520 DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, dmu_sync_done,
1521 dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
1527 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1533 err = dnode_hold(os, object, FTAG, &dn);
1536 err = dnode_set_blksz(dn, size, ibs, tx);
1537 dnode_rele(dn, FTAG);
1542 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1547 /* XXX assumes dnode_hold will not get an i/o error */
1548 (void) dnode_hold(os, object, FTAG, &dn);
1549 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
1550 dn->dn_checksum = checksum;
1551 dnode_setdirty(dn, tx);
1552 dnode_rele(dn, FTAG);
1556 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1561 /* XXX assumes dnode_hold will not get an i/o error */
1562 (void) dnode_hold(os, object, FTAG, &dn);
1563 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
1564 dn->dn_compress = compress;
1565 dnode_setdirty(dn, tx);
1566 dnode_rele(dn, FTAG);
1569 int zfs_mdcomp_disable = 0;
1570 TUNABLE_INT("vfs.zfs.mdcomp_disable", &zfs_mdcomp_disable);
1571 SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RW,
1572 &zfs_mdcomp_disable, 0, "Disable metadata compression");
1575 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1577 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1578 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
1580 enum zio_checksum checksum = os->os_checksum;
1581 enum zio_compress compress = os->os_compress;
1582 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1583 boolean_t dedup = B_FALSE;
1584 boolean_t nopwrite = B_FALSE;
1585 boolean_t dedup_verify = os->os_dedup_verify;
1586 int copies = os->os_copies;
1589 * We maintain different write policies for each of the following
1592 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
1593 * 3. all other level 0 blocks
1597 * XXX -- we should design a compression algorithm
1598 * that specializes in arrays of bps.
1600 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
1604 * Metadata always gets checksummed. If the data
1605 * checksum is multi-bit correctable, and it's not a
1606 * ZBT-style checksum, then it's suitable for metadata
1607 * as well. Otherwise, the metadata checksum defaults
1610 if (zio_checksum_table[checksum].ci_correctable < 1 ||
1611 zio_checksum_table[checksum].ci_eck)
1612 checksum = ZIO_CHECKSUM_FLETCHER_4;
1613 } else if (wp & WP_NOFILL) {
1617 * If we're writing preallocated blocks, we aren't actually
1618 * writing them so don't set any policy properties. These
1619 * blocks are currently only used by an external subsystem
1620 * outside of zfs (i.e. dump) and not written by the zio
1623 compress = ZIO_COMPRESS_OFF;
1624 checksum = ZIO_CHECKSUM_OFF;
1626 compress = zio_compress_select(dn->dn_compress, compress);
1628 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
1629 zio_checksum_select(dn->dn_checksum, checksum) :
1633 * Determine dedup setting. If we are in dmu_sync(),
1634 * we won't actually dedup now because that's all
1635 * done in syncing context; but we do want to use the
1636 * dedup checkum. If the checksum is not strong
1637 * enough to ensure unique signatures, force
1640 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
1641 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
1642 if (!zio_checksum_table[checksum].ci_dedup)
1643 dedup_verify = B_TRUE;
1647 * Enable nopwrite if we have a cryptographically secure
1648 * checksum that has no known collisions (i.e. SHA-256)
1649 * and compression is enabled. We don't enable nopwrite if
1650 * dedup is enabled as the two features are mutually exclusive.
1652 nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup &&
1653 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
1656 zp->zp_checksum = checksum;
1657 zp->zp_compress = compress;
1658 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
1659 zp->zp_level = level;
1660 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa));
1661 zp->zp_dedup = dedup;
1662 zp->zp_dedup_verify = dedup && dedup_verify;
1663 zp->zp_nopwrite = nopwrite;
1667 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1672 err = dnode_hold(os, object, FTAG, &dn);
1676 * Sync any current changes before
1677 * we go trundling through the block pointers.
1679 for (i = 0; i < TXG_SIZE; i++) {
1680 if (list_link_active(&dn->dn_dirty_link[i]))
1683 if (i != TXG_SIZE) {
1684 dnode_rele(dn, FTAG);
1685 txg_wait_synced(dmu_objset_pool(os), 0);
1686 err = dnode_hold(os, object, FTAG, &dn);
1691 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1692 dnode_rele(dn, FTAG);
1698 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1702 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1703 mutex_enter(&dn->dn_mtx);
1707 doi->doi_data_block_size = dn->dn_datablksz;
1708 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1709 1ULL << dn->dn_indblkshift : 0;
1710 doi->doi_type = dn->dn_type;
1711 doi->doi_bonus_type = dn->dn_bonustype;
1712 doi->doi_bonus_size = dn->dn_bonuslen;
1713 doi->doi_indirection = dn->dn_nlevels;
1714 doi->doi_checksum = dn->dn_checksum;
1715 doi->doi_compress = dn->dn_compress;
1716 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1717 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
1718 doi->doi_fill_count = 0;
1719 for (int i = 0; i < dnp->dn_nblkptr; i++)
1720 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
1722 mutex_exit(&dn->dn_mtx);
1723 rw_exit(&dn->dn_struct_rwlock);
1727 * Get information on a DMU object.
1728 * If doi is NULL, just indicates whether the object exists.
1731 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1734 int err = dnode_hold(os, object, FTAG, &dn);
1740 dmu_object_info_from_dnode(dn, doi);
1742 dnode_rele(dn, FTAG);
1747 * As above, but faster; can be used when you have a held dbuf in hand.
1750 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
1752 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1755 dmu_object_info_from_dnode(DB_DNODE(db), doi);
1760 * Faster still when you only care about the size.
1761 * This is specifically optimized for zfs_getattr().
1764 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
1765 u_longlong_t *nblk512)
1767 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
1773 *blksize = dn->dn_datablksz;
1774 /* add 1 for dnode space */
1775 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1776 SPA_MINBLOCKSHIFT) + 1;
1781 byteswap_uint64_array(void *vbuf, size_t size)
1783 uint64_t *buf = vbuf;
1784 size_t count = size >> 3;
1787 ASSERT((size & 7) == 0);
1789 for (i = 0; i < count; i++)
1790 buf[i] = BSWAP_64(buf[i]);
1794 byteswap_uint32_array(void *vbuf, size_t size)
1796 uint32_t *buf = vbuf;
1797 size_t count = size >> 2;
1800 ASSERT((size & 3) == 0);
1802 for (i = 0; i < count; i++)
1803 buf[i] = BSWAP_32(buf[i]);
1807 byteswap_uint16_array(void *vbuf, size_t size)
1809 uint16_t *buf = vbuf;
1810 size_t count = size >> 1;
1813 ASSERT((size & 1) == 0);
1815 for (i = 0; i < count; i++)
1816 buf[i] = BSWAP_16(buf[i]);
1821 byteswap_uint8_array(void *vbuf, size_t size)
1842 arc_fini(); /* arc depends on l2arc, so arc must go first */