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) 2011, 2020 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
28 * Copyright (c) 2019 Datto Inc.
29 * Copyright (c) 2019, Klara Inc.
30 * Copyright (c) 2019, Allan Jude
34 #include <sys/dmu_impl.h>
35 #include <sys/dmu_tx.h>
37 #include <sys/dnode.h>
38 #include <sys/zfs_context.h>
39 #include <sys/dmu_objset.h>
40 #include <sys/dmu_traverse.h>
41 #include <sys/dsl_dataset.h>
42 #include <sys/dsl_dir.h>
43 #include <sys/dsl_pool.h>
44 #include <sys/dsl_synctask.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dmu_zfetch.h>
47 #include <sys/zfs_ioctl.h>
49 #include <sys/zio_checksum.h>
50 #include <sys/zio_compress.h>
52 #include <sys/zfeature.h>
54 #include <sys/trace_zfs.h>
55 #include <sys/zfs_rlock.h>
57 #include <sys/vmsystm.h>
58 #include <sys/zfs_znode.h>
62 * Enable/disable nopwrite feature.
64 int zfs_nopwrite_enabled = 1;
67 * Tunable to control percentage of dirtied L1 blocks from frees allowed into
68 * one TXG. After this threshold is crossed, additional dirty blocks from frees
69 * will wait until the next TXG.
70 * A value of zero will disable this throttle.
72 unsigned long zfs_per_txg_dirty_frees_percent = 5;
75 * Enable/disable forcing txg sync when dirty in dmu_offset_next.
77 int zfs_dmu_offset_next_sync = 0;
80 * Limit the amount we can prefetch with one call to this amount. This
81 * helps to limit the amount of memory that can be used by prefetching.
82 * Larger objects should be prefetched a bit at a time.
84 int dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE;
86 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
87 {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "unallocated" },
88 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "object directory" },
89 {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "object array" },
90 {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "packed nvlist" },
91 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "packed nvlist size" },
92 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj" },
93 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj header" },
94 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map header" },
95 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA space map" },
96 {DMU_BSWAP_UINT64, TRUE, FALSE, TRUE, "ZIL intent log" },
97 {DMU_BSWAP_DNODE, TRUE, FALSE, TRUE, "DMU dnode" },
98 {DMU_BSWAP_OBJSET, TRUE, TRUE, FALSE, "DMU objset" },
99 {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL directory" },
100 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL directory child map"},
101 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset snap map" },
102 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL props" },
103 {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL dataset" },
104 {DMU_BSWAP_ZNODE, TRUE, FALSE, FALSE, "ZFS znode" },
105 {DMU_BSWAP_OLDACL, TRUE, FALSE, TRUE, "ZFS V0 ACL" },
106 {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "ZFS plain file" },
107 {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS directory" },
108 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "ZFS master node" },
109 {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS delete queue" },
110 {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "zvol object" },
111 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "zvol prop" },
112 {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "other uint8[]" },
113 {DMU_BSWAP_UINT64, FALSE, FALSE, TRUE, "other uint64[]" },
114 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "other ZAP" },
115 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "persistent error log" },
116 {DMU_BSWAP_UINT8, TRUE, FALSE, FALSE, "SPA history" },
117 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "SPA history offsets" },
118 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "Pool properties" },
119 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL permissions" },
120 {DMU_BSWAP_ACL, TRUE, FALSE, TRUE, "ZFS ACL" },
121 {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "ZFS SYSACL" },
122 {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "FUID table" },
123 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "FUID table size" },
124 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dataset next clones"},
125 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan work queue" },
126 {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project used" },
127 {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "ZFS user/group/project quota"},
128 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "snapshot refcount tags"},
129 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT ZAP algorithm" },
130 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "DDT statistics" },
131 {DMU_BSWAP_UINT8, TRUE, FALSE, TRUE, "System attributes" },
132 {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA master node" },
133 {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr registration" },
134 {DMU_BSWAP_ZAP, TRUE, FALSE, TRUE, "SA attr layouts" },
135 {DMU_BSWAP_ZAP, TRUE, FALSE, FALSE, "scan translations" },
136 {DMU_BSWAP_UINT8, FALSE, FALSE, TRUE, "deduplicated block" },
137 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL deadlist map" },
138 {DMU_BSWAP_UINT64, TRUE, TRUE, FALSE, "DSL deadlist map hdr" },
139 {DMU_BSWAP_ZAP, TRUE, TRUE, FALSE, "DSL dir clones" },
140 {DMU_BSWAP_UINT64, TRUE, FALSE, FALSE, "bpobj subobj" }
143 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
144 { byteswap_uint8_array, "uint8" },
145 { byteswap_uint16_array, "uint16" },
146 { byteswap_uint32_array, "uint32" },
147 { byteswap_uint64_array, "uint64" },
148 { zap_byteswap, "zap" },
149 { dnode_buf_byteswap, "dnode" },
150 { dmu_objset_byteswap, "objset" },
151 { zfs_znode_byteswap, "znode" },
152 { zfs_oldacl_byteswap, "oldacl" },
153 { zfs_acl_byteswap, "acl" }
157 dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset,
158 void *tag, dmu_buf_t **dbp)
163 rw_enter(&dn->dn_struct_rwlock, RW_READER);
164 blkid = dbuf_whichblock(dn, 0, offset);
165 db = dbuf_hold(dn, blkid, tag);
166 rw_exit(&dn->dn_struct_rwlock);
170 return (SET_ERROR(EIO));
177 dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
178 void *tag, dmu_buf_t **dbp)
185 err = dnode_hold(os, object, FTAG, &dn);
188 rw_enter(&dn->dn_struct_rwlock, RW_READER);
189 blkid = dbuf_whichblock(dn, 0, offset);
190 db = dbuf_hold(dn, blkid, tag);
191 rw_exit(&dn->dn_struct_rwlock);
192 dnode_rele(dn, FTAG);
196 return (SET_ERROR(EIO));
204 dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
205 void *tag, dmu_buf_t **dbp, int flags)
208 int db_flags = DB_RF_CANFAIL;
210 if (flags & DMU_READ_NO_PREFETCH)
211 db_flags |= DB_RF_NOPREFETCH;
212 if (flags & DMU_READ_NO_DECRYPT)
213 db_flags |= DB_RF_NO_DECRYPT;
215 err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp);
217 dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
218 err = dbuf_read(db, NULL, db_flags);
229 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
230 void *tag, dmu_buf_t **dbp, int flags)
233 int db_flags = DB_RF_CANFAIL;
235 if (flags & DMU_READ_NO_PREFETCH)
236 db_flags |= DB_RF_NOPREFETCH;
237 if (flags & DMU_READ_NO_DECRYPT)
238 db_flags |= DB_RF_NO_DECRYPT;
240 err = dmu_buf_hold_noread(os, object, offset, tag, dbp);
242 dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
243 err = dbuf_read(db, NULL, db_flags);
256 return (DN_OLD_MAX_BONUSLEN);
260 dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx)
262 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
269 if (dn->dn_bonus != db) {
270 error = SET_ERROR(EINVAL);
271 } else if (newsize < 0 || newsize > db_fake->db_size) {
272 error = SET_ERROR(EINVAL);
274 dnode_setbonuslen(dn, newsize, tx);
283 dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx)
285 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
292 if (!DMU_OT_IS_VALID(type)) {
293 error = SET_ERROR(EINVAL);
294 } else if (dn->dn_bonus != db) {
295 error = SET_ERROR(EINVAL);
297 dnode_setbonus_type(dn, type, tx);
306 dmu_get_bonustype(dmu_buf_t *db_fake)
308 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
310 dmu_object_type_t type;
314 type = dn->dn_bonustype;
321 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx)
326 error = dnode_hold(os, object, FTAG, &dn);
327 dbuf_rm_spill(dn, tx);
328 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
329 dnode_rm_spill(dn, tx);
330 rw_exit(&dn->dn_struct_rwlock);
331 dnode_rele(dn, FTAG);
336 * Lookup and hold the bonus buffer for the provided dnode. If the dnode
337 * has not yet been allocated a new bonus dbuf a will be allocated.
338 * Returns ENOENT, EIO, or 0.
340 int dmu_bonus_hold_by_dnode(dnode_t *dn, void *tag, dmu_buf_t **dbp,
345 uint32_t db_flags = DB_RF_MUST_SUCCEED;
347 if (flags & DMU_READ_NO_PREFETCH)
348 db_flags |= DB_RF_NOPREFETCH;
349 if (flags & DMU_READ_NO_DECRYPT)
350 db_flags |= DB_RF_NO_DECRYPT;
352 rw_enter(&dn->dn_struct_rwlock, RW_READER);
353 if (dn->dn_bonus == NULL) {
354 rw_exit(&dn->dn_struct_rwlock);
355 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
356 if (dn->dn_bonus == NULL)
357 dbuf_create_bonus(dn);
361 /* as long as the bonus buf is held, the dnode will be held */
362 if (zfs_refcount_add(&db->db_holds, tag) == 1) {
363 VERIFY(dnode_add_ref(dn, db));
364 atomic_inc_32(&dn->dn_dbufs_count);
368 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
369 * hold and incrementing the dbuf count to ensure that dnode_move() sees
370 * a dnode hold for every dbuf.
372 rw_exit(&dn->dn_struct_rwlock);
374 error = dbuf_read(db, NULL, db_flags);
376 dnode_evict_bonus(dn);
387 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
392 error = dnode_hold(os, object, FTAG, &dn);
396 error = dmu_bonus_hold_by_dnode(dn, tag, dbp, DMU_READ_NO_PREFETCH);
397 dnode_rele(dn, FTAG);
403 * returns ENOENT, EIO, or 0.
405 * This interface will allocate a blank spill dbuf when a spill blk
406 * doesn't already exist on the dnode.
408 * if you only want to find an already existing spill db, then
409 * dmu_spill_hold_existing() should be used.
412 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
414 dmu_buf_impl_t *db = NULL;
417 if ((flags & DB_RF_HAVESTRUCT) == 0)
418 rw_enter(&dn->dn_struct_rwlock, RW_READER);
420 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
422 if ((flags & DB_RF_HAVESTRUCT) == 0)
423 rw_exit(&dn->dn_struct_rwlock);
427 return (SET_ERROR(EIO));
429 err = dbuf_read(db, NULL, flags);
440 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
442 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
449 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
450 err = SET_ERROR(EINVAL);
452 rw_enter(&dn->dn_struct_rwlock, RW_READER);
454 if (!dn->dn_have_spill) {
455 err = SET_ERROR(ENOENT);
457 err = dmu_spill_hold_by_dnode(dn,
458 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
461 rw_exit(&dn->dn_struct_rwlock);
469 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, uint32_t flags, void *tag,
472 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
475 uint32_t db_flags = DB_RF_CANFAIL;
477 if (flags & DMU_READ_NO_DECRYPT)
478 db_flags |= DB_RF_NO_DECRYPT;
482 err = dmu_spill_hold_by_dnode(dn, db_flags, tag, dbp);
489 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
490 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
491 * and can induce severe lock contention when writing to several files
492 * whose dnodes are in the same block.
495 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
496 boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
499 uint64_t blkid, nblks, i;
504 ASSERT(length <= DMU_MAX_ACCESS);
507 * Note: We directly notify the prefetch code of this read, so that
508 * we can tell it about the multi-block read. dbuf_read() only knows
509 * about the one block it is accessing.
511 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT |
514 rw_enter(&dn->dn_struct_rwlock, RW_READER);
515 if (dn->dn_datablkshift) {
516 int blkshift = dn->dn_datablkshift;
517 nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) -
518 P2ALIGN(offset, 1ULL << blkshift)) >> blkshift;
520 if (offset + length > dn->dn_datablksz) {
521 zfs_panic_recover("zfs: accessing past end of object "
522 "%llx/%llx (size=%u access=%llu+%llu)",
523 (longlong_t)dn->dn_objset->
524 os_dsl_dataset->ds_object,
525 (longlong_t)dn->dn_object, dn->dn_datablksz,
526 (longlong_t)offset, (longlong_t)length);
527 rw_exit(&dn->dn_struct_rwlock);
528 return (SET_ERROR(EIO));
532 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
534 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
535 blkid = dbuf_whichblock(dn, 0, offset);
536 for (i = 0; i < nblks; i++) {
537 dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag);
539 rw_exit(&dn->dn_struct_rwlock);
540 dmu_buf_rele_array(dbp, nblks, tag);
542 return (SET_ERROR(EIO));
545 /* initiate async i/o */
547 (void) dbuf_read(db, zio, dbuf_flags);
551 if ((flags & DMU_READ_NO_PREFETCH) == 0 &&
552 DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) {
553 dmu_zfetch(&dn->dn_zfetch, blkid, nblks,
554 read && DNODE_IS_CACHEABLE(dn), B_TRUE);
556 rw_exit(&dn->dn_struct_rwlock);
558 /* wait for async i/o */
561 dmu_buf_rele_array(dbp, nblks, tag);
565 /* wait for other io to complete */
567 for (i = 0; i < nblks; i++) {
568 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
569 mutex_enter(&db->db_mtx);
570 while (db->db_state == DB_READ ||
571 db->db_state == DB_FILL)
572 cv_wait(&db->db_changed, &db->db_mtx);
573 if (db->db_state == DB_UNCACHED)
574 err = SET_ERROR(EIO);
575 mutex_exit(&db->db_mtx);
577 dmu_buf_rele_array(dbp, nblks, tag);
589 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
590 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
595 err = dnode_hold(os, object, FTAG, &dn);
599 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
600 numbufsp, dbpp, DMU_READ_PREFETCH);
602 dnode_rele(dn, FTAG);
608 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
609 uint64_t length, boolean_t read, void *tag, int *numbufsp,
612 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
618 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
619 numbufsp, dbpp, DMU_READ_PREFETCH);
626 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
629 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
634 for (i = 0; i < numbufs; i++) {
636 dbuf_rele(dbp[i], tag);
639 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
643 * Issue prefetch i/os for the given blocks. If level is greater than 0, the
644 * indirect blocks prefetched will be those that point to the blocks containing
645 * the data starting at offset, and continuing to offset + len.
647 * Note that if the indirect blocks above the blocks being prefetched are not
648 * in cache, they will be asynchronously read in.
651 dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
652 uint64_t len, zio_priority_t pri)
658 if (len == 0) { /* they're interested in the bonus buffer */
659 dn = DMU_META_DNODE(os);
661 if (object == 0 || object >= DN_MAX_OBJECT)
664 rw_enter(&dn->dn_struct_rwlock, RW_READER);
665 blkid = dbuf_whichblock(dn, level,
666 object * sizeof (dnode_phys_t));
667 dbuf_prefetch(dn, level, blkid, pri, 0);
668 rw_exit(&dn->dn_struct_rwlock);
673 * See comment before the definition of dmu_prefetch_max.
675 len = MIN(len, dmu_prefetch_max);
678 * XXX - Note, if the dnode for the requested object is not
679 * already cached, we will do a *synchronous* read in the
680 * dnode_hold() call. The same is true for any indirects.
682 err = dnode_hold(os, object, FTAG, &dn);
687 * offset + len - 1 is the last byte we want to prefetch for, and offset
688 * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the
689 * last block we want to prefetch, and dbuf_whichblock(dn, level,
690 * offset) is the first. Then the number we need to prefetch is the
693 rw_enter(&dn->dn_struct_rwlock, RW_READER);
694 if (level > 0 || dn->dn_datablkshift != 0) {
695 nblks = dbuf_whichblock(dn, level, offset + len - 1) -
696 dbuf_whichblock(dn, level, offset) + 1;
698 nblks = (offset < dn->dn_datablksz);
702 blkid = dbuf_whichblock(dn, level, offset);
703 for (int i = 0; i < nblks; i++)
704 dbuf_prefetch(dn, level, blkid + i, pri, 0);
706 rw_exit(&dn->dn_struct_rwlock);
708 dnode_rele(dn, FTAG);
712 * Get the next "chunk" of file data to free. We traverse the file from
713 * the end so that the file gets shorter over time (if we crashes in the
714 * middle, this will leave us in a better state). We find allocated file
715 * data by simply searching the allocated level 1 indirects.
717 * On input, *start should be the first offset that does not need to be
718 * freed (e.g. "offset + length"). On return, *start will be the first
719 * offset that should be freed and l1blks is set to the number of level 1
720 * indirect blocks found within the chunk.
723 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks)
726 uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
727 /* bytes of data covered by a level-1 indirect block */
728 uint64_t iblkrange = (uint64_t)dn->dn_datablksz *
729 EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
731 ASSERT3U(minimum, <=, *start);
734 * Check if we can free the entire range assuming that all of the
735 * L1 blocks in this range have data. If we can, we use this
736 * worst case value as an estimate so we can avoid having to look
737 * at the object's actual data.
739 uint64_t total_l1blks =
740 (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) /
742 if (total_l1blks <= maxblks) {
743 *l1blks = total_l1blks;
747 ASSERT(ISP2(iblkrange));
749 for (blks = 0; *start > minimum && blks < maxblks; blks++) {
753 * dnode_next_offset(BACKWARDS) will find an allocated L1
754 * indirect block at or before the input offset. We must
755 * decrement *start so that it is at the end of the region
760 err = dnode_next_offset(dn,
761 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
763 /* if there are no indirect blocks before start, we are done */
767 } else if (err != 0) {
772 /* set start to the beginning of this L1 indirect */
773 *start = P2ALIGN(*start, iblkrange);
775 if (*start < minimum)
783 * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set,
784 * otherwise return false.
785 * Used below in dmu_free_long_range_impl() to enable abort when unmounting
789 dmu_objset_zfs_unmounting(objset_t *os)
792 if (dmu_objset_type(os) == DMU_OST_ZFS)
793 return (zfs_get_vfs_flag_unmounted(os));
799 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
802 uint64_t object_size;
804 uint64_t dirty_frees_threshold;
805 dsl_pool_t *dp = dmu_objset_pool(os);
808 return (SET_ERROR(EINVAL));
810 object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
811 if (offset >= object_size)
814 if (zfs_per_txg_dirty_frees_percent <= 100)
815 dirty_frees_threshold =
816 zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100;
818 dirty_frees_threshold = zfs_dirty_data_max / 20;
820 if (length == DMU_OBJECT_END || offset + length > object_size)
821 length = object_size - offset;
823 while (length != 0) {
824 uint64_t chunk_end, chunk_begin, chunk_len;
828 if (dmu_objset_zfs_unmounting(dn->dn_objset))
829 return (SET_ERROR(EINTR));
831 chunk_end = chunk_begin = offset + length;
833 /* move chunk_begin backwards to the beginning of this chunk */
834 err = get_next_chunk(dn, &chunk_begin, offset, &l1blks);
837 ASSERT3U(chunk_begin, >=, offset);
838 ASSERT3U(chunk_begin, <=, chunk_end);
840 chunk_len = chunk_end - chunk_begin;
842 tx = dmu_tx_create(os);
843 dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len);
846 * Mark this transaction as typically resulting in a net
847 * reduction in space used.
849 dmu_tx_mark_netfree(tx);
850 err = dmu_tx_assign(tx, TXG_WAIT);
856 uint64_t txg = dmu_tx_get_txg(tx);
858 mutex_enter(&dp->dp_lock);
859 uint64_t long_free_dirty =
860 dp->dp_long_free_dirty_pertxg[txg & TXG_MASK];
861 mutex_exit(&dp->dp_lock);
864 * To avoid filling up a TXG with just frees, wait for
865 * the next TXG to open before freeing more chunks if
866 * we have reached the threshold of frees.
868 if (dirty_frees_threshold != 0 &&
869 long_free_dirty >= dirty_frees_threshold) {
870 DMU_TX_STAT_BUMP(dmu_tx_dirty_frees_delay);
872 txg_wait_open(dp, 0, B_TRUE);
877 * In order to prevent unnecessary write throttling, for each
878 * TXG, we track the cumulative size of L1 blocks being dirtied
879 * in dnode_free_range() below. We compare this number to a
880 * tunable threshold, past which we prevent new L1 dirty freeing
881 * blocks from being added into the open TXG. See
882 * dmu_free_long_range_impl() for details. The threshold
883 * prevents write throttle activation due to dirty freeing L1
884 * blocks taking up a large percentage of zfs_dirty_data_max.
886 mutex_enter(&dp->dp_lock);
887 dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] +=
888 l1blks << dn->dn_indblkshift;
889 mutex_exit(&dp->dp_lock);
890 DTRACE_PROBE3(free__long__range,
891 uint64_t, long_free_dirty, uint64_t, chunk_len,
893 dnode_free_range(dn, chunk_begin, chunk_len, tx);
903 dmu_free_long_range(objset_t *os, uint64_t object,
904 uint64_t offset, uint64_t length)
909 err = dnode_hold(os, object, FTAG, &dn);
912 err = dmu_free_long_range_impl(os, dn, offset, length);
915 * It is important to zero out the maxblkid when freeing the entire
916 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
917 * will take the fast path, and (b) dnode_reallocate() can verify
918 * that the entire file has been freed.
920 if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
923 dnode_rele(dn, FTAG);
928 dmu_free_long_object(objset_t *os, uint64_t object)
933 err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
937 tx = dmu_tx_create(os);
938 dmu_tx_hold_bonus(tx, object);
939 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
940 dmu_tx_mark_netfree(tx);
941 err = dmu_tx_assign(tx, TXG_WAIT);
944 err = dmu_object_free(os, object, tx);
955 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
956 uint64_t size, dmu_tx_t *tx)
959 int err = dnode_hold(os, object, FTAG, &dn);
962 ASSERT(offset < UINT64_MAX);
963 ASSERT(size == DMU_OBJECT_END || size <= UINT64_MAX - offset);
964 dnode_free_range(dn, offset, size, tx);
965 dnode_rele(dn, FTAG);
970 dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size,
971 void *buf, uint32_t flags)
974 int numbufs, err = 0;
977 * Deal with odd block sizes, where there can't be data past the first
978 * block. If we ever do the tail block optimization, we will need to
979 * handle that here as well.
981 if (dn->dn_maxblkid == 0) {
982 uint64_t newsz = offset > dn->dn_datablksz ? 0 :
983 MIN(size, dn->dn_datablksz - offset);
984 bzero((char *)buf + newsz, size - newsz);
989 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
993 * NB: we could do this block-at-a-time, but it's nice
994 * to be reading in parallel.
996 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
997 TRUE, FTAG, &numbufs, &dbp, flags);
1001 for (i = 0; i < numbufs; i++) {
1004 dmu_buf_t *db = dbp[i];
1008 bufoff = offset - db->db_offset;
1009 tocpy = MIN(db->db_size - bufoff, size);
1011 (void) memcpy(buf, (char *)db->db_data + bufoff, tocpy);
1015 buf = (char *)buf + tocpy;
1017 dmu_buf_rele_array(dbp, numbufs, FTAG);
1023 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1024 void *buf, uint32_t flags)
1029 err = dnode_hold(os, object, FTAG, &dn);
1033 err = dmu_read_impl(dn, offset, size, buf, flags);
1034 dnode_rele(dn, FTAG);
1039 dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
1042 return (dmu_read_impl(dn, offset, size, buf, flags));
1046 dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size,
1047 const void *buf, dmu_tx_t *tx)
1051 for (i = 0; i < numbufs; i++) {
1054 dmu_buf_t *db = dbp[i];
1058 bufoff = offset - db->db_offset;
1059 tocpy = MIN(db->db_size - bufoff, size);
1061 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1063 if (tocpy == db->db_size)
1064 dmu_buf_will_fill(db, tx);
1066 dmu_buf_will_dirty(db, tx);
1068 (void) memcpy((char *)db->db_data + bufoff, buf, tocpy);
1070 if (tocpy == db->db_size)
1071 dmu_buf_fill_done(db, tx);
1075 buf = (char *)buf + tocpy;
1080 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1081 const void *buf, dmu_tx_t *tx)
1089 VERIFY0(dmu_buf_hold_array(os, object, offset, size,
1090 FALSE, FTAG, &numbufs, &dbp));
1091 dmu_write_impl(dbp, numbufs, offset, size, buf, tx);
1092 dmu_buf_rele_array(dbp, numbufs, FTAG);
1096 * Note: Lustre is an external consumer of this interface.
1099 dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
1100 const void *buf, dmu_tx_t *tx)
1108 VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size,
1109 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH));
1110 dmu_write_impl(dbp, numbufs, offset, size, buf, tx);
1111 dmu_buf_rele_array(dbp, numbufs, FTAG);
1115 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1124 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
1125 FALSE, FTAG, &numbufs, &dbp));
1127 for (i = 0; i < numbufs; i++) {
1128 dmu_buf_t *db = dbp[i];
1130 dmu_buf_will_not_fill(db, tx);
1132 dmu_buf_rele_array(dbp, numbufs, FTAG);
1136 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
1137 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
1138 int compressed_size, int byteorder, dmu_tx_t *tx)
1142 ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
1143 ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
1144 VERIFY0(dmu_buf_hold_noread(os, object, offset,
1147 dmu_buf_write_embedded(db,
1148 data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
1149 uncompressed_size, compressed_size, byteorder, tx);
1151 dmu_buf_rele(db, FTAG);
1155 dmu_redact(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1161 VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG,
1163 for (i = 0; i < numbufs; i++)
1164 dmu_buf_redact(dbp[i], tx);
1165 dmu_buf_rele_array(dbp, numbufs, FTAG);
1169 * DMU support for xuio
1171 kstat_t *xuio_ksp = NULL;
1173 typedef struct xuio_stats {
1174 /* loaned yet not returned arc_buf */
1175 kstat_named_t xuiostat_onloan_rbuf;
1176 kstat_named_t xuiostat_onloan_wbuf;
1177 /* whether a copy is made when loaning out a read buffer */
1178 kstat_named_t xuiostat_rbuf_copied;
1179 kstat_named_t xuiostat_rbuf_nocopy;
1180 /* whether a copy is made when assigning a write buffer */
1181 kstat_named_t xuiostat_wbuf_copied;
1182 kstat_named_t xuiostat_wbuf_nocopy;
1185 static xuio_stats_t xuio_stats = {
1186 { "onloan_read_buf", KSTAT_DATA_UINT64 },
1187 { "onloan_write_buf", KSTAT_DATA_UINT64 },
1188 { "read_buf_copied", KSTAT_DATA_UINT64 },
1189 { "read_buf_nocopy", KSTAT_DATA_UINT64 },
1190 { "write_buf_copied", KSTAT_DATA_UINT64 },
1191 { "write_buf_nocopy", KSTAT_DATA_UINT64 }
1194 #define XUIOSTAT_INCR(stat, val) \
1195 atomic_add_64(&xuio_stats.stat.value.ui64, (val))
1196 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
1198 #ifdef HAVE_UIO_ZEROCOPY
1200 dmu_xuio_init(xuio_t *xuio, int nblk)
1203 uio_t *uio = &xuio->xu_uio;
1205 uio->uio_iovcnt = nblk;
1206 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
1208 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
1210 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
1211 priv->iovp = (iovec_t *)uio->uio_iov;
1212 XUIO_XUZC_PRIV(xuio) = priv;
1214 if (XUIO_XUZC_RW(xuio) == UIO_READ)
1215 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
1217 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
1223 dmu_xuio_fini(xuio_t *xuio)
1225 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1226 int nblk = priv->cnt;
1228 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
1229 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
1230 kmem_free(priv, sizeof (dmu_xuio_t));
1232 if (XUIO_XUZC_RW(xuio) == UIO_READ)
1233 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
1235 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
1239 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
1240 * and increase priv->next by 1.
1243 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
1246 uio_t *uio = &xuio->xu_uio;
1247 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1248 int i = priv->next++;
1250 ASSERT(i < priv->cnt);
1251 ASSERT(off + n <= arc_buf_lsize(abuf));
1252 iov = (iovec_t *)uio->uio_iov + i;
1253 iov->iov_base = (char *)abuf->b_data + off;
1255 priv->bufs[i] = abuf;
1260 dmu_xuio_cnt(xuio_t *xuio)
1262 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1267 dmu_xuio_arcbuf(xuio_t *xuio, int i)
1269 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1271 ASSERT(i < priv->cnt);
1272 return (priv->bufs[i]);
1276 dmu_xuio_clear(xuio_t *xuio, int i)
1278 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1280 ASSERT(i < priv->cnt);
1281 priv->bufs[i] = NULL;
1283 #endif /* HAVE_UIO_ZEROCOPY */
1286 xuio_stat_init(void)
1288 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
1289 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
1290 KSTAT_FLAG_VIRTUAL);
1291 if (xuio_ksp != NULL) {
1292 xuio_ksp->ks_data = &xuio_stats;
1293 kstat_install(xuio_ksp);
1298 xuio_stat_fini(void)
1300 if (xuio_ksp != NULL) {
1301 kstat_delete(xuio_ksp);
1307 xuio_stat_wbuf_copied(void)
1309 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1313 xuio_stat_wbuf_nocopy(void)
1315 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1320 dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
1323 int numbufs, i, err;
1324 #ifdef HAVE_UIO_ZEROCOPY
1325 xuio_t *xuio = NULL;
1329 * NB: we could do this block-at-a-time, but it's nice
1330 * to be reading in parallel.
1332 err = dmu_buf_hold_array_by_dnode(dn, uio_offset(uio), size,
1333 TRUE, FTAG, &numbufs, &dbp, 0);
1337 for (i = 0; i < numbufs; i++) {
1340 dmu_buf_t *db = dbp[i];
1344 bufoff = uio_offset(uio) - db->db_offset;
1345 tocpy = MIN(db->db_size - bufoff, size);
1347 #ifdef HAVE_UIO_ZEROCOPY
1349 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1350 arc_buf_t *dbuf_abuf = dbi->db_buf;
1351 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1352 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1354 uio_advance(uio, tocpy);
1356 if (abuf == dbuf_abuf)
1357 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1359 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1363 err = vn_io_fault_uiomove((char *)db->db_data + bufoff,
1366 err = uiomove((char *)db->db_data + bufoff, tocpy,
1374 dmu_buf_rele_array(dbp, numbufs, FTAG);
1380 * Read 'size' bytes into the uio buffer.
1381 * From object zdb->db_object.
1382 * Starting at offset uio->uio_loffset.
1384 * If the caller already has a dbuf in the target object
1385 * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1386 * because we don't have to find the dnode_t for the object.
1389 dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size)
1391 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1400 err = dmu_read_uio_dnode(dn, uio, size);
1407 * Read 'size' bytes into the uio buffer.
1408 * From the specified object
1409 * Starting at offset uio->uio_loffset.
1412 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1420 err = dnode_hold(os, object, FTAG, &dn);
1424 err = dmu_read_uio_dnode(dn, uio, size);
1426 dnode_rele(dn, FTAG);
1432 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1439 err = dmu_buf_hold_array_by_dnode(dn, uio_offset(uio), size,
1440 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1444 for (i = 0; i < numbufs; i++) {
1447 dmu_buf_t *db = dbp[i];
1451 bufoff = uio_offset(uio) - db->db_offset;
1452 tocpy = MIN(db->db_size - bufoff, size);
1454 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1456 if (tocpy == db->db_size)
1457 dmu_buf_will_fill(db, tx);
1459 dmu_buf_will_dirty(db, tx);
1462 * XXX uiomove could block forever (eg.nfs-backed
1463 * pages). There needs to be a uiolockdown() function
1464 * to lock the pages in memory, so that uiomove won't
1468 err = vn_io_fault_uiomove((char *)db->db_data + bufoff,
1471 err = uiomove((char *)db->db_data + bufoff, tocpy,
1474 if (tocpy == db->db_size)
1475 dmu_buf_fill_done(db, tx);
1483 dmu_buf_rele_array(dbp, numbufs, FTAG);
1488 * Write 'size' bytes from the uio buffer.
1489 * To object zdb->db_object.
1490 * Starting at offset uio->uio_loffset.
1492 * If the caller already has a dbuf in the target object
1493 * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1494 * because we don't have to find the dnode_t for the object.
1497 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1500 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1509 err = dmu_write_uio_dnode(dn, uio, size, tx);
1516 * Write 'size' bytes from the uio buffer.
1517 * To the specified object.
1518 * Starting at offset uio->uio_loffset.
1521 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1530 err = dnode_hold(os, object, FTAG, &dn);
1534 err = dmu_write_uio_dnode(dn, uio, size, tx);
1536 dnode_rele(dn, FTAG);
1540 #endif /* _KERNEL */
1543 * Allocate a loaned anonymous arc buffer.
1546 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1548 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1550 return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size));
1554 * Free a loaned arc buffer.
1557 dmu_return_arcbuf(arc_buf_t *buf)
1559 arc_return_buf(buf, FTAG);
1560 arc_buf_destroy(buf, FTAG);
1564 * When possible directly assign passed loaned arc buffer to a dbuf.
1565 * If this is not possible copy the contents of passed arc buf via
1569 dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf,
1573 objset_t *os = dn->dn_objset;
1574 uint64_t object = dn->dn_object;
1575 uint32_t blksz = (uint32_t)arc_buf_lsize(buf);
1578 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1579 blkid = dbuf_whichblock(dn, 0, offset);
1580 db = dbuf_hold(dn, blkid, FTAG);
1582 return (SET_ERROR(EIO));
1583 rw_exit(&dn->dn_struct_rwlock);
1586 * We can only assign if the offset is aligned, the arc buf is the
1587 * same size as the dbuf, and the dbuf is not metadata.
1589 if (offset == db->db.db_offset && blksz == db->db.db_size) {
1590 dbuf_assign_arcbuf(db, buf, tx);
1591 dbuf_rele(db, FTAG);
1593 /* compressed bufs must always be assignable to their dbuf */
1594 ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF);
1595 ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED));
1597 dbuf_rele(db, FTAG);
1598 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1599 dmu_return_arcbuf(buf);
1600 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1607 dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1611 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1613 DB_DNODE_ENTER(dbuf);
1614 err = dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf), offset, buf, tx);
1615 DB_DNODE_EXIT(dbuf);
1621 dbuf_dirty_record_t *dsa_dr;
1622 dmu_sync_cb_t *dsa_done;
1629 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1631 dmu_sync_arg_t *dsa = varg;
1632 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1633 blkptr_t *bp = zio->io_bp;
1635 if (zio->io_error == 0) {
1636 if (BP_IS_HOLE(bp)) {
1638 * A block of zeros may compress to a hole, but the
1639 * block size still needs to be known for replay.
1641 BP_SET_LSIZE(bp, db->db_size);
1642 } else if (!BP_IS_EMBEDDED(bp)) {
1643 ASSERT(BP_GET_LEVEL(bp) == 0);
1650 dmu_sync_late_arrival_ready(zio_t *zio)
1652 dmu_sync_ready(zio, NULL, zio->io_private);
1657 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1659 dmu_sync_arg_t *dsa = varg;
1660 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1661 dmu_buf_impl_t *db = dr->dr_dbuf;
1662 zgd_t *zgd = dsa->dsa_zgd;
1665 * Record the vdev(s) backing this blkptr so they can be flushed after
1666 * the writes for the lwb have completed.
1668 if (zio->io_error == 0) {
1669 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1672 mutex_enter(&db->db_mtx);
1673 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1674 if (zio->io_error == 0) {
1675 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
1676 if (dr->dt.dl.dr_nopwrite) {
1677 blkptr_t *bp = zio->io_bp;
1678 blkptr_t *bp_orig = &zio->io_bp_orig;
1679 uint8_t chksum = BP_GET_CHECKSUM(bp_orig);
1681 ASSERT(BP_EQUAL(bp, bp_orig));
1682 VERIFY(BP_EQUAL(bp, db->db_blkptr));
1683 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
1684 VERIFY(zio_checksum_table[chksum].ci_flags &
1685 ZCHECKSUM_FLAG_NOPWRITE);
1687 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1688 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1689 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1692 * Old style holes are filled with all zeros, whereas
1693 * new-style holes maintain their lsize, type, level,
1694 * and birth time (see zio_write_compress). While we
1695 * need to reset the BP_SET_LSIZE() call that happened
1696 * in dmu_sync_ready for old style holes, we do *not*
1697 * want to wipe out the information contained in new
1698 * style holes. Thus, only zero out the block pointer if
1699 * it's an old style hole.
1701 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) &&
1702 dr->dt.dl.dr_overridden_by.blk_birth == 0)
1703 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1705 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1707 cv_broadcast(&db->db_changed);
1708 mutex_exit(&db->db_mtx);
1710 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1712 kmem_free(dsa, sizeof (*dsa));
1716 dmu_sync_late_arrival_done(zio_t *zio)
1718 blkptr_t *bp = zio->io_bp;
1719 dmu_sync_arg_t *dsa = zio->io_private;
1720 zgd_t *zgd = dsa->dsa_zgd;
1722 if (zio->io_error == 0) {
1724 * Record the vdev(s) backing this blkptr so they can be
1725 * flushed after the writes for the lwb have completed.
1727 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1729 if (!BP_IS_HOLE(bp)) {
1730 blkptr_t *bp_orig __maybe_unused = &zio->io_bp_orig;
1731 ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE));
1732 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
1733 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1734 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1735 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1739 dmu_tx_commit(dsa->dsa_tx);
1741 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1743 abd_put(zio->io_abd);
1744 kmem_free(dsa, sizeof (*dsa));
1748 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1749 zio_prop_t *zp, zbookmark_phys_t *zb)
1751 dmu_sync_arg_t *dsa;
1754 tx = dmu_tx_create(os);
1755 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1756 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1758 /* Make zl_get_data do txg_waited_synced() */
1759 return (SET_ERROR(EIO));
1763 * In order to prevent the zgd's lwb from being free'd prior to
1764 * dmu_sync_late_arrival_done() being called, we have to ensure
1765 * the lwb's "max txg" takes this tx's txg into account.
1767 zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx));
1769 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1771 dsa->dsa_done = done;
1776 * Since we are currently syncing this txg, it's nontrivial to
1777 * determine what BP to nopwrite against, so we disable nopwrite.
1779 * When syncing, the db_blkptr is initially the BP of the previous
1780 * txg. We can not nopwrite against it because it will be changed
1781 * (this is similar to the non-late-arrival case where the dbuf is
1782 * dirty in a future txg).
1784 * Then dbuf_write_ready() sets bp_blkptr to the location we will write.
1785 * We can not nopwrite against it because although the BP will not
1786 * (typically) be changed, the data has not yet been persisted to this
1789 * Finally, when dbuf_write_done() is called, it is theoretically
1790 * possible to always nopwrite, because the data that was written in
1791 * this txg is the same data that we are trying to write. However we
1792 * would need to check that this dbuf is not dirty in any future
1793 * txg's (as we do in the normal dmu_sync() path). For simplicity, we
1794 * don't nopwrite in this case.
1796 zp->zp_nopwrite = B_FALSE;
1798 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1799 abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size),
1800 zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp,
1801 dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done,
1802 dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
1808 * Intent log support: sync the block associated with db to disk.
1809 * N.B. and XXX: the caller is responsible for making sure that the
1810 * data isn't changing while dmu_sync() is writing it.
1814 * EEXIST: this txg has already been synced, so there's nothing to do.
1815 * The caller should not log the write.
1817 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1818 * The caller should not log the write.
1820 * EALREADY: this block is already in the process of being synced.
1821 * The caller should track its progress (somehow).
1823 * EIO: could not do the I/O.
1824 * The caller should do a txg_wait_synced().
1826 * 0: the I/O has been initiated.
1827 * The caller should log this blkptr in the done callback.
1828 * It is possible that the I/O will fail, in which case
1829 * the error will be reported to the done callback and
1830 * propagated to pio from zio_done().
1833 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1835 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1836 objset_t *os = db->db_objset;
1837 dsl_dataset_t *ds = os->os_dsl_dataset;
1838 dbuf_dirty_record_t *dr, *dr_next;
1839 dmu_sync_arg_t *dsa;
1840 zbookmark_phys_t zb;
1844 ASSERT(pio != NULL);
1847 SET_BOOKMARK(&zb, ds->ds_object,
1848 db->db.db_object, db->db_level, db->db_blkid);
1852 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
1856 * If we're frozen (running ziltest), we always need to generate a bp.
1858 if (txg > spa_freeze_txg(os->os_spa))
1859 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1862 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1863 * and us. If we determine that this txg is not yet syncing,
1864 * but it begins to sync a moment later, that's OK because the
1865 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1867 mutex_enter(&db->db_mtx);
1869 if (txg <= spa_last_synced_txg(os->os_spa)) {
1871 * This txg has already synced. There's nothing to do.
1873 mutex_exit(&db->db_mtx);
1874 return (SET_ERROR(EEXIST));
1877 if (txg <= spa_syncing_txg(os->os_spa)) {
1879 * This txg is currently syncing, so we can't mess with
1880 * the dirty record anymore; just write a new log block.
1882 mutex_exit(&db->db_mtx);
1883 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1886 dr = dbuf_find_dirty_eq(db, txg);
1890 * There's no dr for this dbuf, so it must have been freed.
1891 * There's no need to log writes to freed blocks, so we're done.
1893 mutex_exit(&db->db_mtx);
1894 return (SET_ERROR(ENOENT));
1897 dr_next = list_next(&db->db_dirty_records, dr);
1898 ASSERT(dr_next == NULL || dr_next->dr_txg < txg);
1900 if (db->db_blkptr != NULL) {
1902 * We need to fill in zgd_bp with the current blkptr so that
1903 * the nopwrite code can check if we're writing the same
1904 * data that's already on disk. We can only nopwrite if we
1905 * are sure that after making the copy, db_blkptr will not
1906 * change until our i/o completes. We ensure this by
1907 * holding the db_mtx, and only allowing nopwrite if the
1908 * block is not already dirty (see below). This is verified
1909 * by dmu_sync_done(), which VERIFYs that the db_blkptr has
1912 *zgd->zgd_bp = *db->db_blkptr;
1916 * Assume the on-disk data is X, the current syncing data (in
1917 * txg - 1) is Y, and the current in-memory data is Z (currently
1920 * We usually want to perform a nopwrite if X and Z are the
1921 * same. However, if Y is different (i.e. the BP is going to
1922 * change before this write takes effect), then a nopwrite will
1923 * be incorrect - we would override with X, which could have
1924 * been freed when Y was written.
1926 * (Note that this is not a concern when we are nop-writing from
1927 * syncing context, because X and Y must be identical, because
1928 * all previous txgs have been synced.)
1930 * Therefore, we disable nopwrite if the current BP could change
1931 * before this TXG. There are two ways it could change: by
1932 * being dirty (dr_next is non-NULL), or by being freed
1933 * (dnode_block_freed()). This behavior is verified by
1934 * zio_done(), which VERIFYs that the override BP is identical
1935 * to the on-disk BP.
1939 if (dr_next != NULL || dnode_block_freed(dn, db->db_blkid))
1940 zp.zp_nopwrite = B_FALSE;
1943 ASSERT(dr->dr_txg == txg);
1944 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1945 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1947 * We have already issued a sync write for this buffer,
1948 * or this buffer has already been synced. It could not
1949 * have been dirtied since, or we would have cleared the state.
1951 mutex_exit(&db->db_mtx);
1952 return (SET_ERROR(EALREADY));
1955 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1956 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1957 mutex_exit(&db->db_mtx);
1959 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1961 dsa->dsa_done = done;
1965 zio_nowait(arc_write(pio, os->os_spa, txg,
1966 zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
1967 &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa,
1968 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
1974 dmu_object_set_nlevels(objset_t *os, uint64_t object, int nlevels, dmu_tx_t *tx)
1979 err = dnode_hold(os, object, FTAG, &dn);
1982 err = dnode_set_nlevels(dn, nlevels, tx);
1983 dnode_rele(dn, FTAG);
1988 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1994 err = dnode_hold(os, object, FTAG, &dn);
1997 err = dnode_set_blksz(dn, size, ibs, tx);
1998 dnode_rele(dn, FTAG);
2003 dmu_object_set_maxblkid(objset_t *os, uint64_t object, uint64_t maxblkid,
2009 err = dnode_hold(os, object, FTAG, &dn);
2012 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
2013 dnode_new_blkid(dn, maxblkid, tx, B_FALSE, B_TRUE);
2014 rw_exit(&dn->dn_struct_rwlock);
2015 dnode_rele(dn, FTAG);
2020 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
2026 * Send streams include each object's checksum function. This
2027 * check ensures that the receiving system can understand the
2028 * checksum function transmitted.
2030 ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
2032 VERIFY0(dnode_hold(os, object, FTAG, &dn));
2033 ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
2034 dn->dn_checksum = checksum;
2035 dnode_setdirty(dn, tx);
2036 dnode_rele(dn, FTAG);
2040 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
2046 * Send streams include each object's compression function. This
2047 * check ensures that the receiving system can understand the
2048 * compression function transmitted.
2050 ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
2052 VERIFY0(dnode_hold(os, object, FTAG, &dn));
2053 dn->dn_compress = compress;
2054 dnode_setdirty(dn, tx);
2055 dnode_rele(dn, FTAG);
2059 * When the "redundant_metadata" property is set to "most", only indirect
2060 * blocks of this level and higher will have an additional ditto block.
2062 int zfs_redundant_metadata_most_ditto_level = 2;
2065 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
2067 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
2068 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
2070 enum zio_checksum checksum = os->os_checksum;
2071 enum zio_compress compress = os->os_compress;
2072 uint8_t complevel = os->os_complevel;
2073 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
2074 boolean_t dedup = B_FALSE;
2075 boolean_t nopwrite = B_FALSE;
2076 boolean_t dedup_verify = os->os_dedup_verify;
2077 boolean_t encrypt = B_FALSE;
2078 int copies = os->os_copies;
2081 * We maintain different write policies for each of the following
2084 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
2085 * 3. all other level 0 blocks
2089 * XXX -- we should design a compression algorithm
2090 * that specializes in arrays of bps.
2092 compress = zio_compress_select(os->os_spa,
2093 ZIO_COMPRESS_ON, ZIO_COMPRESS_ON);
2096 * Metadata always gets checksummed. If the data
2097 * checksum is multi-bit correctable, and it's not a
2098 * ZBT-style checksum, then it's suitable for metadata
2099 * as well. Otherwise, the metadata checksum defaults
2102 if (!(zio_checksum_table[checksum].ci_flags &
2103 ZCHECKSUM_FLAG_METADATA) ||
2104 (zio_checksum_table[checksum].ci_flags &
2105 ZCHECKSUM_FLAG_EMBEDDED))
2106 checksum = ZIO_CHECKSUM_FLETCHER_4;
2108 if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
2109 (os->os_redundant_metadata ==
2110 ZFS_REDUNDANT_METADATA_MOST &&
2111 (level >= zfs_redundant_metadata_most_ditto_level ||
2112 DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))))
2114 } else if (wp & WP_NOFILL) {
2118 * If we're writing preallocated blocks, we aren't actually
2119 * writing them so don't set any policy properties. These
2120 * blocks are currently only used by an external subsystem
2121 * outside of zfs (i.e. dump) and not written by the zio
2124 compress = ZIO_COMPRESS_OFF;
2125 checksum = ZIO_CHECKSUM_OFF;
2127 compress = zio_compress_select(os->os_spa, dn->dn_compress,
2129 complevel = zio_complevel_select(os->os_spa, compress,
2130 complevel, complevel);
2132 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
2133 zio_checksum_select(dn->dn_checksum, checksum) :
2137 * Determine dedup setting. If we are in dmu_sync(),
2138 * we won't actually dedup now because that's all
2139 * done in syncing context; but we do want to use the
2140 * dedup checksum. If the checksum is not strong
2141 * enough to ensure unique signatures, force
2144 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
2145 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
2146 if (!(zio_checksum_table[checksum].ci_flags &
2147 ZCHECKSUM_FLAG_DEDUP))
2148 dedup_verify = B_TRUE;
2152 * Enable nopwrite if we have secure enough checksum
2153 * algorithm (see comment in zio_nop_write) and
2154 * compression is enabled. We don't enable nopwrite if
2155 * dedup is enabled as the two features are mutually
2158 nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags &
2159 ZCHECKSUM_FLAG_NOPWRITE) &&
2160 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
2164 * All objects in an encrypted objset are protected from modification
2165 * via a MAC. Encrypted objects store their IV and salt in the last DVA
2166 * in the bp, so we cannot use all copies. Encrypted objects are also
2167 * not subject to nopwrite since writing the same data will still
2168 * result in a new ciphertext. Only encrypted blocks can be dedup'd
2169 * to avoid ambiguity in the dedup code since the DDT does not store
2172 if (os->os_encrypted && (wp & WP_NOFILL) == 0) {
2175 if (DMU_OT_IS_ENCRYPTED(type)) {
2176 copies = MIN(copies, SPA_DVAS_PER_BP - 1);
2183 (type == DMU_OT_DNODE || type == DMU_OT_OBJSET)) {
2184 compress = ZIO_COMPRESS_EMPTY;
2188 zp->zp_compress = compress;
2189 zp->zp_complevel = complevel;
2190 zp->zp_checksum = checksum;
2191 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
2192 zp->zp_level = level;
2193 zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
2194 zp->zp_dedup = dedup;
2195 zp->zp_dedup_verify = dedup && dedup_verify;
2196 zp->zp_nopwrite = nopwrite;
2197 zp->zp_encrypt = encrypt;
2198 zp->zp_byteorder = ZFS_HOST_BYTEORDER;
2199 bzero(zp->zp_salt, ZIO_DATA_SALT_LEN);
2200 bzero(zp->zp_iv, ZIO_DATA_IV_LEN);
2201 bzero(zp->zp_mac, ZIO_DATA_MAC_LEN);
2202 zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ?
2203 os->os_zpl_special_smallblock : 0;
2205 ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT);
2209 * This function is only called from zfs_holey_common() for zpl_llseek()
2210 * in order to determine the location of holes. In order to accurately
2211 * report holes all dirty data must be synced to disk. This causes extremely
2212 * poor performance when seeking for holes in a dirty file. As a compromise,
2213 * only provide hole data when the dnode is clean. When a dnode is dirty
2214 * report the dnode as having no holes which is always a safe thing to do.
2217 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
2221 boolean_t clean = B_TRUE;
2223 err = dnode_hold(os, object, FTAG, &dn);
2228 * Check if dnode is dirty
2230 for (i = 0; i < TXG_SIZE; i++) {
2231 if (multilist_link_active(&dn->dn_dirty_link[i])) {
2238 * If compatibility option is on, sync any current changes before
2239 * we go trundling through the block pointers.
2241 if (!clean && zfs_dmu_offset_next_sync) {
2243 dnode_rele(dn, FTAG);
2244 txg_wait_synced(dmu_objset_pool(os), 0);
2245 err = dnode_hold(os, object, FTAG, &dn);
2251 err = dnode_next_offset(dn,
2252 (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
2254 err = SET_ERROR(EBUSY);
2256 dnode_rele(dn, FTAG);
2262 __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
2264 dnode_phys_t *dnp = dn->dn_phys;
2266 doi->doi_data_block_size = dn->dn_datablksz;
2267 doi->doi_metadata_block_size = dn->dn_indblkshift ?
2268 1ULL << dn->dn_indblkshift : 0;
2269 doi->doi_type = dn->dn_type;
2270 doi->doi_bonus_type = dn->dn_bonustype;
2271 doi->doi_bonus_size = dn->dn_bonuslen;
2272 doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT;
2273 doi->doi_indirection = dn->dn_nlevels;
2274 doi->doi_checksum = dn->dn_checksum;
2275 doi->doi_compress = dn->dn_compress;
2276 doi->doi_nblkptr = dn->dn_nblkptr;
2277 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
2278 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
2279 doi->doi_fill_count = 0;
2280 for (int i = 0; i < dnp->dn_nblkptr; i++)
2281 doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
2285 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
2287 rw_enter(&dn->dn_struct_rwlock, RW_READER);
2288 mutex_enter(&dn->dn_mtx);
2290 __dmu_object_info_from_dnode(dn, doi);
2292 mutex_exit(&dn->dn_mtx);
2293 rw_exit(&dn->dn_struct_rwlock);
2297 * Get information on a DMU object.
2298 * If doi is NULL, just indicates whether the object exists.
2301 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
2304 int err = dnode_hold(os, object, FTAG, &dn);
2310 dmu_object_info_from_dnode(dn, doi);
2312 dnode_rele(dn, FTAG);
2317 * As above, but faster; can be used when you have a held dbuf in hand.
2320 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
2322 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2325 dmu_object_info_from_dnode(DB_DNODE(db), doi);
2330 * Faster still when you only care about the size.
2333 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
2334 u_longlong_t *nblk512)
2336 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2342 *blksize = dn->dn_datablksz;
2343 /* add in number of slots used for the dnode itself */
2344 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
2345 SPA_MINBLOCKSHIFT) + dn->dn_num_slots;
2350 dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize)
2352 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2357 *dnsize = dn->dn_num_slots << DNODE_SHIFT;
2362 byteswap_uint64_array(void *vbuf, size_t size)
2364 uint64_t *buf = vbuf;
2365 size_t count = size >> 3;
2368 ASSERT((size & 7) == 0);
2370 for (i = 0; i < count; i++)
2371 buf[i] = BSWAP_64(buf[i]);
2375 byteswap_uint32_array(void *vbuf, size_t size)
2377 uint32_t *buf = vbuf;
2378 size_t count = size >> 2;
2381 ASSERT((size & 3) == 0);
2383 for (i = 0; i < count; i++)
2384 buf[i] = BSWAP_32(buf[i]);
2388 byteswap_uint16_array(void *vbuf, size_t size)
2390 uint16_t *buf = vbuf;
2391 size_t count = size >> 1;
2394 ASSERT((size & 1) == 0);
2396 for (i = 0; i < count; i++)
2397 buf[i] = BSWAP_16(buf[i]);
2402 byteswap_uint8_array(void *vbuf, size_t size)
2425 arc_fini(); /* arc depends on l2arc, so arc must go first */
2438 EXPORT_SYMBOL(dmu_bonus_hold);
2439 EXPORT_SYMBOL(dmu_bonus_hold_by_dnode);
2440 EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus);
2441 EXPORT_SYMBOL(dmu_buf_rele_array);
2442 EXPORT_SYMBOL(dmu_prefetch);
2443 EXPORT_SYMBOL(dmu_free_range);
2444 EXPORT_SYMBOL(dmu_free_long_range);
2445 EXPORT_SYMBOL(dmu_free_long_object);
2446 EXPORT_SYMBOL(dmu_read);
2447 EXPORT_SYMBOL(dmu_read_by_dnode);
2448 EXPORT_SYMBOL(dmu_write);
2449 EXPORT_SYMBOL(dmu_write_by_dnode);
2450 EXPORT_SYMBOL(dmu_prealloc);
2451 EXPORT_SYMBOL(dmu_object_info);
2452 EXPORT_SYMBOL(dmu_object_info_from_dnode);
2453 EXPORT_SYMBOL(dmu_object_info_from_db);
2454 EXPORT_SYMBOL(dmu_object_size_from_db);
2455 EXPORT_SYMBOL(dmu_object_dnsize_from_db);
2456 EXPORT_SYMBOL(dmu_object_set_nlevels);
2457 EXPORT_SYMBOL(dmu_object_set_blocksize);
2458 EXPORT_SYMBOL(dmu_object_set_maxblkid);
2459 EXPORT_SYMBOL(dmu_object_set_checksum);
2460 EXPORT_SYMBOL(dmu_object_set_compress);
2461 EXPORT_SYMBOL(dmu_offset_next);
2462 EXPORT_SYMBOL(dmu_write_policy);
2463 EXPORT_SYMBOL(dmu_sync);
2464 EXPORT_SYMBOL(dmu_request_arcbuf);
2465 EXPORT_SYMBOL(dmu_return_arcbuf);
2466 EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode);
2467 EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf);
2468 EXPORT_SYMBOL(dmu_buf_hold);
2469 EXPORT_SYMBOL(dmu_ot);
2472 ZFS_MODULE_PARAM(zfs, zfs_, nopwrite_enabled, INT, ZMOD_RW,
2473 "Enable NOP writes");
2475 ZFS_MODULE_PARAM(zfs, zfs_, per_txg_dirty_frees_percent, ULONG, ZMOD_RW,
2476 "Percentage of dirtied blocks from frees in one TXG");
2478 ZFS_MODULE_PARAM(zfs, zfs_, dmu_offset_next_sync, INT, ZMOD_RW,
2479 "Enable forcing txg sync to find holes");
2481 ZFS_MODULE_PARAM(zfs, , dmu_prefetch_max, INT, ZMOD_RW,
2482 "Limit one prefetch call to this size");