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, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2019 Datto Inc.
26 /* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */
27 /* Copyright (c) 2013, Joyent, Inc. All rights reserved. */
28 /* Copyright 2016 Nexenta Systems, Inc. All rights reserved. */
31 #include <sys/dmu_impl.h>
32 #include <sys/dmu_tx.h>
34 #include <sys/dnode.h>
35 #include <sys/zfs_context.h>
36 #include <sys/dmu_objset.h>
37 #include <sys/dmu_traverse.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/dsl_prop.h>
43 #include <sys/dmu_zfetch.h>
44 #include <sys/zfs_ioctl.h>
46 #include <sys/zio_checksum.h>
47 #include <sys/zio_compress.h>
49 #include <sys/zfeature.h>
52 #include <sys/racct.h>
54 #include <sys/zfs_znode.h>
58 * Enable/disable nopwrite feature.
60 int zfs_nopwrite_enabled = 1;
61 SYSCTL_DECL(_vfs_zfs);
62 SYSCTL_INT(_vfs_zfs, OID_AUTO, nopwrite_enabled, CTLFLAG_RDTUN,
63 &zfs_nopwrite_enabled, 0, "Enable nopwrite feature");
66 * Tunable to control percentage of dirtied L1 blocks from frees allowed into
67 * one TXG. After this threshold is crossed, additional dirty blocks from frees
68 * will wait until the next TXG.
69 * A value of zero will disable this throttle.
71 uint32_t zfs_per_txg_dirty_frees_percent = 5;
72 SYSCTL_INT(_vfs_zfs, OID_AUTO, per_txg_dirty_frees_percent, CTLFLAG_RWTUN,
73 &zfs_per_txg_dirty_frees_percent, 0,
74 "Percentage of dirtied indirect blocks from frees allowed in one txg");
77 * This can be used for testing, to ensure that certain actions happen
78 * while in the middle of a remap (which might otherwise complete too
81 int zfs_object_remap_one_indirect_delay_ticks = 0;
84 * Limit the amount we can prefetch with one call to this amount. This
85 * helps to limit the amount of memory that can be used by prefetching.
86 * Larger objects should be prefetched a bit at a time.
88 uint64_t dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE;
90 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
91 { DMU_BSWAP_UINT8, TRUE, FALSE, "unallocated" },
92 { DMU_BSWAP_ZAP, TRUE, TRUE, "object directory" },
93 { DMU_BSWAP_UINT64, TRUE, TRUE, "object array" },
94 { DMU_BSWAP_UINT8, TRUE, FALSE, "packed nvlist" },
95 { DMU_BSWAP_UINT64, TRUE, FALSE, "packed nvlist size" },
96 { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj" },
97 { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj header" },
98 { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA space map header" },
99 { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA space map" },
100 { DMU_BSWAP_UINT64, TRUE, FALSE, "ZIL intent log" },
101 { DMU_BSWAP_DNODE, TRUE, FALSE, "DMU dnode" },
102 { DMU_BSWAP_OBJSET, TRUE, TRUE, "DMU objset" },
103 { DMU_BSWAP_UINT64, TRUE, TRUE, "DSL directory" },
104 { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL directory child map" },
105 { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL dataset snap map" },
106 { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL props" },
107 { DMU_BSWAP_UINT64, TRUE, TRUE, "DSL dataset" },
108 { DMU_BSWAP_ZNODE, TRUE, FALSE, "ZFS znode" },
109 { DMU_BSWAP_OLDACL, TRUE, FALSE, "ZFS V0 ACL" },
110 { DMU_BSWAP_UINT8, FALSE, FALSE, "ZFS plain file" },
111 { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS directory" },
112 { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS master node" },
113 { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS delete queue" },
114 { DMU_BSWAP_UINT8, FALSE, FALSE, "zvol object" },
115 { DMU_BSWAP_ZAP, TRUE, FALSE, "zvol prop" },
116 { DMU_BSWAP_UINT8, FALSE, FALSE, "other uint8[]" },
117 { DMU_BSWAP_UINT64, FALSE, FALSE, "other uint64[]" },
118 { DMU_BSWAP_ZAP, TRUE, FALSE, "other ZAP" },
119 { DMU_BSWAP_ZAP, TRUE, FALSE, "persistent error log" },
120 { DMU_BSWAP_UINT8, TRUE, FALSE, "SPA history" },
121 { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA history offsets" },
122 { DMU_BSWAP_ZAP, TRUE, TRUE, "Pool properties" },
123 { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL permissions" },
124 { DMU_BSWAP_ACL, TRUE, FALSE, "ZFS ACL" },
125 { DMU_BSWAP_UINT8, TRUE, FALSE, "ZFS SYSACL" },
126 { DMU_BSWAP_UINT8, TRUE, FALSE, "FUID table" },
127 { DMU_BSWAP_UINT64, TRUE, FALSE, "FUID table size" },
128 { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL dataset next clones" },
129 { DMU_BSWAP_ZAP, TRUE, FALSE, "scan work queue" },
130 { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS user/group used" },
131 { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS user/group quota" },
132 { DMU_BSWAP_ZAP, TRUE, TRUE, "snapshot refcount tags" },
133 { DMU_BSWAP_ZAP, TRUE, FALSE, "DDT ZAP algorithm" },
134 { DMU_BSWAP_ZAP, TRUE, FALSE, "DDT statistics" },
135 { DMU_BSWAP_UINT8, TRUE, FALSE, "System attributes" },
136 { DMU_BSWAP_ZAP, TRUE, FALSE, "SA master node" },
137 { DMU_BSWAP_ZAP, TRUE, FALSE, "SA attr registration" },
138 { DMU_BSWAP_ZAP, TRUE, FALSE, "SA attr layouts" },
139 { DMU_BSWAP_ZAP, TRUE, FALSE, "scan translations" },
140 { DMU_BSWAP_UINT8, FALSE, FALSE, "deduplicated block" },
141 { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL deadlist map" },
142 { DMU_BSWAP_UINT64, TRUE, TRUE, "DSL deadlist map hdr" },
143 { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL dir clones" },
144 { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj subobj" }
147 const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
148 { byteswap_uint8_array, "uint8" },
149 { byteswap_uint16_array, "uint16" },
150 { byteswap_uint32_array, "uint32" },
151 { byteswap_uint64_array, "uint64" },
152 { zap_byteswap, "zap" },
153 { dnode_buf_byteswap, "dnode" },
154 { dmu_objset_byteswap, "objset" },
155 { zfs_znode_byteswap, "znode" },
156 { zfs_oldacl_byteswap, "oldacl" },
157 { zfs_acl_byteswap, "acl" }
161 dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset,
162 void *tag, dmu_buf_t **dbp)
167 blkid = dbuf_whichblock(dn, 0, offset);
168 rw_enter(&dn->dn_struct_rwlock, RW_READER);
169 db = dbuf_hold(dn, blkid, tag);
170 rw_exit(&dn->dn_struct_rwlock);
174 return (SET_ERROR(EIO));
181 dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
182 void *tag, dmu_buf_t **dbp)
189 err = dnode_hold(os, object, FTAG, &dn);
192 blkid = dbuf_whichblock(dn, 0, offset);
193 rw_enter(&dn->dn_struct_rwlock, RW_READER);
194 db = dbuf_hold(dn, blkid, tag);
195 rw_exit(&dn->dn_struct_rwlock);
196 dnode_rele(dn, FTAG);
200 return (SET_ERROR(EIO));
208 dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
209 void *tag, dmu_buf_t **dbp, int flags)
212 int db_flags = DB_RF_CANFAIL;
214 if (flags & DMU_READ_NO_PREFETCH)
215 db_flags |= DB_RF_NOPREFETCH;
217 err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp);
219 dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
220 err = dbuf_read(db, NULL, db_flags);
231 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
232 void *tag, dmu_buf_t **dbp, int flags)
235 int db_flags = DB_RF_CANFAIL;
237 if (flags & DMU_READ_NO_PREFETCH)
238 db_flags |= DB_RF_NOPREFETCH;
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 * returns ENOENT, EIO, or 0.
339 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
345 error = dnode_hold(os, object, FTAG, &dn);
349 rw_enter(&dn->dn_struct_rwlock, RW_READER);
350 if (dn->dn_bonus == NULL) {
351 rw_exit(&dn->dn_struct_rwlock);
352 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
353 if (dn->dn_bonus == NULL)
354 dbuf_create_bonus(dn);
358 /* as long as the bonus buf is held, the dnode will be held */
359 if (zfs_refcount_add(&db->db_holds, tag) == 1) {
360 VERIFY(dnode_add_ref(dn, db));
361 atomic_inc_32(&dn->dn_dbufs_count);
365 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
366 * hold and incrementing the dbuf count to ensure that dnode_move() sees
367 * a dnode hold for every dbuf.
369 rw_exit(&dn->dn_struct_rwlock);
371 dnode_rele(dn, FTAG);
373 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH));
380 * returns ENOENT, EIO, or 0.
382 * This interface will allocate a blank spill dbuf when a spill blk
383 * doesn't already exist on the dnode.
385 * if you only want to find an already existing spill db, then
386 * dmu_spill_hold_existing() should be used.
389 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp)
391 dmu_buf_impl_t *db = NULL;
394 if ((flags & DB_RF_HAVESTRUCT) == 0)
395 rw_enter(&dn->dn_struct_rwlock, RW_READER);
397 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag);
399 if ((flags & DB_RF_HAVESTRUCT) == 0)
400 rw_exit(&dn->dn_struct_rwlock);
403 err = dbuf_read(db, NULL, flags);
412 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
414 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
421 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) {
422 err = SET_ERROR(EINVAL);
424 rw_enter(&dn->dn_struct_rwlock, RW_READER);
426 if (!dn->dn_have_spill) {
427 err = SET_ERROR(ENOENT);
429 err = dmu_spill_hold_by_dnode(dn,
430 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp);
433 rw_exit(&dn->dn_struct_rwlock);
441 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp)
443 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus;
449 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp);
456 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
457 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
458 * and can induce severe lock contention when writing to several files
459 * whose dnodes are in the same block.
462 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
463 boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
466 uint64_t blkid, nblks, i;
471 ASSERT(length <= DMU_MAX_ACCESS);
474 * Note: We directly notify the prefetch code of this read, so that
475 * we can tell it about the multi-block read. dbuf_read() only knows
476 * about the one block it is accessing.
478 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT |
481 rw_enter(&dn->dn_struct_rwlock, RW_READER);
482 if (dn->dn_datablkshift) {
483 int blkshift = dn->dn_datablkshift;
484 nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) -
485 P2ALIGN(offset, 1ULL << blkshift)) >> blkshift;
487 if (offset + length > dn->dn_datablksz) {
488 zfs_panic_recover("zfs: accessing past end of object "
489 "%llx/%llx (size=%u access=%llu+%llu)",
490 (longlong_t)dn->dn_objset->
491 os_dsl_dataset->ds_object,
492 (longlong_t)dn->dn_object, dn->dn_datablksz,
493 (longlong_t)offset, (longlong_t)length);
494 rw_exit(&dn->dn_struct_rwlock);
495 return (SET_ERROR(EIO));
499 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
501 #if defined(_KERNEL) && defined(RACCT)
502 if (racct_enable && !read) {
504 racct_add_force(curproc, RACCT_WRITEBPS, length);
505 racct_add_force(curproc, RACCT_WRITEIOPS, nblks);
506 PROC_UNLOCK(curproc);
510 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
511 blkid = dbuf_whichblock(dn, 0, offset);
512 for (i = 0; i < nblks; i++) {
513 dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag);
515 rw_exit(&dn->dn_struct_rwlock);
516 dmu_buf_rele_array(dbp, nblks, tag);
518 return (SET_ERROR(EIO));
521 /* initiate async i/o */
523 (void) dbuf_read(db, zio, dbuf_flags);
526 curthread->td_ru.ru_oublock++;
531 if ((flags & DMU_READ_NO_PREFETCH) == 0 &&
532 DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) {
533 dmu_zfetch(&dn->dn_zfetch, blkid, nblks,
534 read && DNODE_IS_CACHEABLE(dn));
536 rw_exit(&dn->dn_struct_rwlock);
538 /* wait for async i/o */
541 dmu_buf_rele_array(dbp, nblks, tag);
545 /* wait for other io to complete */
547 for (i = 0; i < nblks; i++) {
548 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
549 mutex_enter(&db->db_mtx);
550 while (db->db_state == DB_READ ||
551 db->db_state == DB_FILL)
552 cv_wait(&db->db_changed, &db->db_mtx);
553 if (db->db_state == DB_UNCACHED)
554 err = SET_ERROR(EIO);
555 mutex_exit(&db->db_mtx);
557 dmu_buf_rele_array(dbp, nblks, tag);
569 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
570 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
575 err = dnode_hold(os, object, FTAG, &dn);
579 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
580 numbufsp, dbpp, DMU_READ_PREFETCH);
582 dnode_rele(dn, FTAG);
588 dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset,
589 uint64_t length, boolean_t read, void *tag, int *numbufsp,
592 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
598 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
599 numbufsp, dbpp, DMU_READ_PREFETCH);
606 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
609 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
614 for (i = 0; i < numbufs; i++) {
616 dbuf_rele(dbp[i], tag);
619 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
623 * Issue prefetch i/os for the given blocks. If level is greater than 0, the
624 * indirect blocks prefeteched will be those that point to the blocks containing
625 * the data starting at offset, and continuing to offset + len.
627 * Note that if the indirect blocks above the blocks being prefetched are not in
628 * cache, they will be asychronously read in.
631 dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset,
632 uint64_t len, zio_priority_t pri)
638 if (len == 0) { /* they're interested in the bonus buffer */
639 dn = DMU_META_DNODE(os);
641 if (object == 0 || object >= DN_MAX_OBJECT)
644 rw_enter(&dn->dn_struct_rwlock, RW_READER);
645 blkid = dbuf_whichblock(dn, level,
646 object * sizeof (dnode_phys_t));
647 dbuf_prefetch(dn, level, blkid, pri, 0);
648 rw_exit(&dn->dn_struct_rwlock);
653 * See comment before the definition of dmu_prefetch_max.
655 len = MIN(len, dmu_prefetch_max);
658 * XXX - Note, if the dnode for the requested object is not
659 * already cached, we will do a *synchronous* read in the
660 * dnode_hold() call. The same is true for any indirects.
662 err = dnode_hold(os, object, FTAG, &dn);
666 rw_enter(&dn->dn_struct_rwlock, RW_READER);
668 * offset + len - 1 is the last byte we want to prefetch for, and offset
669 * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the
670 * last block we want to prefetch, and dbuf_whichblock(dn, level,
671 * offset) is the first. Then the number we need to prefetch is the
674 if (level > 0 || dn->dn_datablkshift != 0) {
675 nblks = dbuf_whichblock(dn, level, offset + len - 1) -
676 dbuf_whichblock(dn, level, offset) + 1;
678 nblks = (offset < dn->dn_datablksz);
682 blkid = dbuf_whichblock(dn, level, offset);
683 for (int i = 0; i < nblks; i++)
684 dbuf_prefetch(dn, level, blkid + i, pri, 0);
687 rw_exit(&dn->dn_struct_rwlock);
689 dnode_rele(dn, FTAG);
693 * Get the next "chunk" of file data to free. We traverse the file from
694 * the end so that the file gets shorter over time (if we crashes in the
695 * middle, this will leave us in a better state). We find allocated file
696 * data by simply searching the allocated level 1 indirects.
698 * On input, *start should be the first offset that does not need to be
699 * freed (e.g. "offset + length"). On return, *start will be the first
700 * offset that should be freed and l1blks is set to the number of level 1
701 * indirect blocks found within the chunk.
704 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks)
707 uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1);
708 /* bytes of data covered by a level-1 indirect block */
710 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
712 ASSERT3U(minimum, <=, *start);
715 * Check if we can free the entire range assuming that all of the
716 * L1 blocks in this range have data. If we can, we use this
717 * worst case value as an estimate so we can avoid having to look
718 * at the object's actual data.
720 uint64_t total_l1blks =
721 (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) /
723 if (total_l1blks <= maxblks) {
724 *l1blks = total_l1blks;
728 ASSERT(ISP2(iblkrange));
730 for (blks = 0; *start > minimum && blks < maxblks; blks++) {
734 * dnode_next_offset(BACKWARDS) will find an allocated L1
735 * indirect block at or before the input offset. We must
736 * decrement *start so that it is at the end of the region
741 err = dnode_next_offset(dn,
742 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
744 /* if there are no indirect blocks before start, we are done */
748 } else if (err != 0) {
753 /* set start to the beginning of this L1 indirect */
754 *start = P2ALIGN(*start, iblkrange);
756 if (*start < minimum)
764 * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set,
765 * otherwise return false.
766 * Used below in dmu_free_long_range_impl() to enable abort when unmounting
770 dmu_objset_zfs_unmounting(objset_t *os)
773 if (dmu_objset_type(os) == DMU_OST_ZFS)
774 return (zfs_get_vfs_flag_unmounted(os));
780 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
783 uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
785 uint64_t dirty_frees_threshold;
786 dsl_pool_t *dp = dmu_objset_pool(os);
788 if (offset >= object_size)
791 if (zfs_per_txg_dirty_frees_percent <= 100)
792 dirty_frees_threshold =
793 zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100;
795 dirty_frees_threshold = zfs_dirty_data_max / 20;
797 if (length == DMU_OBJECT_END || offset + length > object_size)
798 length = object_size - offset;
800 while (length != 0) {
801 uint64_t chunk_end, chunk_begin, chunk_len;
805 if (dmu_objset_zfs_unmounting(dn->dn_objset))
806 return (SET_ERROR(EINTR));
808 chunk_end = chunk_begin = offset + length;
810 /* move chunk_begin backwards to the beginning of this chunk */
811 err = get_next_chunk(dn, &chunk_begin, offset, &l1blks);
814 ASSERT3U(chunk_begin, >=, offset);
815 ASSERT3U(chunk_begin, <=, chunk_end);
817 chunk_len = chunk_end - chunk_begin;
819 tx = dmu_tx_create(os);
820 dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len);
823 * Mark this transaction as typically resulting in a net
824 * reduction in space used.
826 dmu_tx_mark_netfree(tx);
827 err = dmu_tx_assign(tx, TXG_WAIT);
833 uint64_t txg = dmu_tx_get_txg(tx);
835 mutex_enter(&dp->dp_lock);
836 uint64_t long_free_dirty =
837 dp->dp_long_free_dirty_pertxg[txg & TXG_MASK];
838 mutex_exit(&dp->dp_lock);
841 * To avoid filling up a TXG with just frees, wait for
842 * the next TXG to open before freeing more chunks if
843 * we have reached the threshold of frees.
845 if (dirty_frees_threshold != 0 &&
846 long_free_dirty >= dirty_frees_threshold) {
848 txg_wait_open(dp, 0);
853 * In order to prevent unnecessary write throttling, for each
854 * TXG, we track the cumulative size of L1 blocks being dirtied
855 * in dnode_free_range() below. We compare this number to a
856 * tunable threshold, past which we prevent new L1 dirty freeing
857 * blocks from being added into the open TXG. See
858 * dmu_free_long_range_impl() for details. The threshold
859 * prevents write throttle activation due to dirty freeing L1
860 * blocks taking up a large percentage of zfs_dirty_data_max.
862 mutex_enter(&dp->dp_lock);
863 dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] +=
864 l1blks << dn->dn_indblkshift;
865 mutex_exit(&dp->dp_lock);
866 DTRACE_PROBE3(free__long__range,
867 uint64_t, long_free_dirty, uint64_t, chunk_len,
869 dnode_free_range(dn, chunk_begin, chunk_len, tx);
878 dmu_free_long_range(objset_t *os, uint64_t object,
879 uint64_t offset, uint64_t length)
884 err = dnode_hold(os, object, FTAG, &dn);
887 err = dmu_free_long_range_impl(os, dn, offset, length);
890 * It is important to zero out the maxblkid when freeing the entire
891 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
892 * will take the fast path, and (b) dnode_reallocate() can verify
893 * that the entire file has been freed.
895 if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
898 dnode_rele(dn, FTAG);
903 dmu_free_long_object(objset_t *os, uint64_t object)
908 err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END);
912 tx = dmu_tx_create(os);
913 dmu_tx_hold_bonus(tx, object);
914 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
915 dmu_tx_mark_netfree(tx);
916 err = dmu_tx_assign(tx, TXG_WAIT);
918 err = dmu_object_free(os, object, tx);
928 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
929 uint64_t size, dmu_tx_t *tx)
932 int err = dnode_hold(os, object, FTAG, &dn);
935 ASSERT(offset < UINT64_MAX);
936 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
937 dnode_free_range(dn, offset, size, tx);
938 dnode_rele(dn, FTAG);
943 dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size,
944 void *buf, uint32_t flags)
947 int numbufs, err = 0;
950 * Deal with odd block sizes, where there can't be data past the first
951 * block. If we ever do the tail block optimization, we will need to
952 * handle that here as well.
954 if (dn->dn_maxblkid == 0) {
955 int newsz = offset > dn->dn_datablksz ? 0 :
956 MIN(size, dn->dn_datablksz - offset);
957 bzero((char *)buf + newsz, size - newsz);
962 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
966 * NB: we could do this block-at-a-time, but it's nice
967 * to be reading in parallel.
969 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
970 TRUE, FTAG, &numbufs, &dbp, flags);
974 for (i = 0; i < numbufs; i++) {
977 dmu_buf_t *db = dbp[i];
981 bufoff = offset - db->db_offset;
982 tocpy = (int)MIN(db->db_size - bufoff, size);
984 bcopy((char *)db->db_data + bufoff, buf, tocpy);
988 buf = (char *)buf + tocpy;
990 dmu_buf_rele_array(dbp, numbufs, FTAG);
996 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
997 void *buf, uint32_t flags)
1002 err = dnode_hold(os, object, FTAG, &dn);
1006 err = dmu_read_impl(dn, offset, size, buf, flags);
1007 dnode_rele(dn, FTAG);
1012 dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf,
1015 return (dmu_read_impl(dn, offset, size, buf, flags));
1019 dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size,
1020 const void *buf, dmu_tx_t *tx)
1024 for (i = 0; i < numbufs; i++) {
1027 dmu_buf_t *db = dbp[i];
1031 bufoff = offset - db->db_offset;
1032 tocpy = (int)MIN(db->db_size - bufoff, size);
1034 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1036 if (tocpy == db->db_size)
1037 dmu_buf_will_fill(db, tx);
1039 dmu_buf_will_dirty(db, tx);
1041 bcopy(buf, (char *)db->db_data + bufoff, tocpy);
1043 if (tocpy == db->db_size)
1044 dmu_buf_fill_done(db, tx);
1048 buf = (char *)buf + tocpy;
1053 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1054 const void *buf, dmu_tx_t *tx)
1062 VERIFY0(dmu_buf_hold_array(os, object, offset, size,
1063 FALSE, FTAG, &numbufs, &dbp));
1064 dmu_write_impl(dbp, numbufs, offset, size, buf, tx);
1065 dmu_buf_rele_array(dbp, numbufs, FTAG);
1069 dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size,
1070 const void *buf, dmu_tx_t *tx)
1078 VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size,
1079 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH));
1080 dmu_write_impl(dbp, numbufs, offset, size, buf, tx);
1081 dmu_buf_rele_array(dbp, numbufs, FTAG);
1085 dmu_object_remap_one_indirect(objset_t *os, dnode_t *dn,
1086 uint64_t last_removal_txg, uint64_t offset)
1088 uint64_t l1blkid = dbuf_whichblock(dn, 1, offset);
1091 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1092 dmu_buf_impl_t *dbuf = dbuf_hold_level(dn, 1, l1blkid, FTAG);
1093 ASSERT3P(dbuf, !=, NULL);
1096 * If the block hasn't been written yet, this default will ensure
1097 * we don't try to remap it.
1099 uint64_t birth = UINT64_MAX;
1100 ASSERT3U(last_removal_txg, !=, UINT64_MAX);
1101 if (dbuf->db_blkptr != NULL)
1102 birth = dbuf->db_blkptr->blk_birth;
1103 rw_exit(&dn->dn_struct_rwlock);
1106 * If this L1 was already written after the last removal, then we've
1107 * already tried to remap it.
1109 if (birth <= last_removal_txg &&
1110 dbuf_read(dbuf, NULL, DB_RF_MUST_SUCCEED) == 0 &&
1111 dbuf_can_remap(dbuf)) {
1112 dmu_tx_t *tx = dmu_tx_create(os);
1113 dmu_tx_hold_remap_l1indirect(tx, dn->dn_object);
1114 err = dmu_tx_assign(tx, TXG_WAIT);
1116 (void) dbuf_dirty(dbuf, tx);
1123 dbuf_rele(dbuf, FTAG);
1125 delay(zfs_object_remap_one_indirect_delay_ticks);
1131 * Remap all blockpointers in the object, if possible, so that they reference
1132 * only concrete vdevs.
1134 * To do this, iterate over the L0 blockpointers and remap any that reference
1135 * an indirect vdev. Note that we only examine L0 blockpointers; since we
1136 * cannot guarantee that we can remap all blockpointer anyways (due to split
1137 * blocks), we do not want to make the code unnecessarily complicated to
1138 * catch the unlikely case that there is an L1 block on an indirect vdev that
1139 * contains no indirect blockpointers.
1142 dmu_object_remap_indirects(objset_t *os, uint64_t object,
1143 uint64_t last_removal_txg)
1145 uint64_t offset, l1span;
1149 err = dnode_hold(os, object, FTAG, &dn);
1154 if (dn->dn_nlevels <= 1) {
1155 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1156 err = SET_ERROR(EINTR);
1160 * If the dnode has no indirect blocks, we cannot dirty them.
1161 * We still want to remap the blkptr(s) in the dnode if
1162 * appropriate, so mark it as dirty.
1164 if (err == 0 && dnode_needs_remap(dn)) {
1165 dmu_tx_t *tx = dmu_tx_create(os);
1166 dmu_tx_hold_bonus(tx, dn->dn_object);
1167 if ((err = dmu_tx_assign(tx, TXG_WAIT)) == 0) {
1168 dnode_setdirty(dn, tx);
1175 dnode_rele(dn, FTAG);
1180 l1span = 1ULL << (dn->dn_indblkshift - SPA_BLKPTRSHIFT +
1181 dn->dn_datablkshift);
1183 * Find the next L1 indirect that is not a hole.
1185 while (dnode_next_offset(dn, 0, &offset, 2, 1, 0) == 0) {
1186 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1187 err = SET_ERROR(EINTR);
1190 if ((err = dmu_object_remap_one_indirect(os, dn,
1191 last_removal_txg, offset)) != 0) {
1197 dnode_rele(dn, FTAG);
1202 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1211 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
1212 FALSE, FTAG, &numbufs, &dbp));
1214 for (i = 0; i < numbufs; i++) {
1215 dmu_buf_t *db = dbp[i];
1217 dmu_buf_will_not_fill(db, tx);
1219 dmu_buf_rele_array(dbp, numbufs, FTAG);
1223 dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
1224 void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
1225 int compressed_size, int byteorder, dmu_tx_t *tx)
1229 ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
1230 ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
1231 VERIFY0(dmu_buf_hold_noread(os, object, offset,
1234 dmu_buf_write_embedded(db,
1235 data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
1236 uncompressed_size, compressed_size, byteorder, tx);
1238 dmu_buf_rele(db, FTAG);
1242 * DMU support for xuio
1244 kstat_t *xuio_ksp = NULL;
1247 dmu_xuio_init(xuio_t *xuio, int nblk)
1250 uio_t *uio = &xuio->xu_uio;
1252 uio->uio_iovcnt = nblk;
1253 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
1255 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
1257 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
1258 priv->iovp = uio->uio_iov;
1259 XUIO_XUZC_PRIV(xuio) = priv;
1261 if (XUIO_XUZC_RW(xuio) == UIO_READ)
1262 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
1264 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
1270 dmu_xuio_fini(xuio_t *xuio)
1272 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1273 int nblk = priv->cnt;
1275 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
1276 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
1277 kmem_free(priv, sizeof (dmu_xuio_t));
1279 if (XUIO_XUZC_RW(xuio) == UIO_READ)
1280 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
1282 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
1286 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
1287 * and increase priv->next by 1.
1290 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
1293 uio_t *uio = &xuio->xu_uio;
1294 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1295 int i = priv->next++;
1297 ASSERT(i < priv->cnt);
1298 ASSERT(off + n <= arc_buf_lsize(abuf));
1299 iov = uio->uio_iov + i;
1300 iov->iov_base = (char *)abuf->b_data + off;
1302 priv->bufs[i] = abuf;
1307 dmu_xuio_cnt(xuio_t *xuio)
1309 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1314 dmu_xuio_arcbuf(xuio_t *xuio, int i)
1316 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1318 ASSERT(i < priv->cnt);
1319 return (priv->bufs[i]);
1323 dmu_xuio_clear(xuio_t *xuio, int i)
1325 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
1327 ASSERT(i < priv->cnt);
1328 priv->bufs[i] = NULL;
1332 xuio_stat_init(void)
1334 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
1335 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
1336 KSTAT_FLAG_VIRTUAL);
1337 if (xuio_ksp != NULL) {
1338 xuio_ksp->ks_data = &xuio_stats;
1339 kstat_install(xuio_ksp);
1344 xuio_stat_fini(void)
1346 if (xuio_ksp != NULL) {
1347 kstat_delete(xuio_ksp);
1353 xuio_stat_wbuf_copied(void)
1355 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1359 xuio_stat_wbuf_nocopy(void)
1361 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
1366 dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size)
1369 int numbufs, i, err;
1370 xuio_t *xuio = NULL;
1373 * NB: we could do this block-at-a-time, but it's nice
1374 * to be reading in parallel.
1376 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1377 TRUE, FTAG, &numbufs, &dbp, 0);
1382 if (uio->uio_extflg == UIO_XUIO)
1383 xuio = (xuio_t *)uio;
1386 for (i = 0; i < numbufs; i++) {
1389 dmu_buf_t *db = dbp[i];
1393 bufoff = uio->uio_loffset - db->db_offset;
1394 tocpy = (int)MIN(db->db_size - bufoff, size);
1397 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
1398 arc_buf_t *dbuf_abuf = dbi->db_buf;
1399 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
1400 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
1402 uio->uio_resid -= tocpy;
1403 uio->uio_loffset += tocpy;
1406 if (abuf == dbuf_abuf)
1407 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
1409 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
1412 err = uiomove((char *)db->db_data + bufoff, tocpy,
1415 err = vn_io_fault_uiomove((char *)db->db_data + bufoff,
1424 dmu_buf_rele_array(dbp, numbufs, FTAG);
1430 * Read 'size' bytes into the uio buffer.
1431 * From object zdb->db_object.
1432 * Starting at offset uio->uio_loffset.
1434 * If the caller already has a dbuf in the target object
1435 * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1436 * because we don't have to find the dnode_t for the object.
1439 dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size)
1441 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1450 err = dmu_read_uio_dnode(dn, uio, size);
1457 * Read 'size' bytes into the uio buffer.
1458 * From the specified object
1459 * Starting at offset uio->uio_loffset.
1462 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
1470 err = dnode_hold(os, object, FTAG, &dn);
1474 err = dmu_read_uio_dnode(dn, uio, size);
1476 dnode_rele(dn, FTAG);
1482 dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx)
1489 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size,
1490 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH);
1494 for (i = 0; i < numbufs; i++) {
1497 dmu_buf_t *db = dbp[i];
1501 bufoff = uio->uio_loffset - db->db_offset;
1502 tocpy = (int)MIN(db->db_size - bufoff, size);
1504 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1506 if (tocpy == db->db_size)
1507 dmu_buf_will_fill(db, tx);
1509 dmu_buf_will_dirty(db, tx);
1513 * XXX uiomove could block forever (eg. nfs-backed
1514 * pages). There needs to be a uiolockdown() function
1515 * to lock the pages in memory, so that uiomove won't
1518 err = uiomove((char *)db->db_data + bufoff, tocpy,
1521 err = vn_io_fault_uiomove((char *)db->db_data + bufoff, tocpy,
1525 if (tocpy == db->db_size)
1526 dmu_buf_fill_done(db, tx);
1534 dmu_buf_rele_array(dbp, numbufs, FTAG);
1539 * Write 'size' bytes from the uio buffer.
1540 * To object zdb->db_object.
1541 * Starting at offset uio->uio_loffset.
1543 * If the caller already has a dbuf in the target object
1544 * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1545 * because we don't have to find the dnode_t for the object.
1548 dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size,
1551 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb;
1560 err = dmu_write_uio_dnode(dn, uio, size, tx);
1567 * Write 'size' bytes from the uio buffer.
1568 * To the specified object.
1569 * Starting at offset uio->uio_loffset.
1572 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
1581 err = dnode_hold(os, object, FTAG, &dn);
1585 err = dmu_write_uio_dnode(dn, uio, size, tx);
1587 dnode_rele(dn, FTAG);
1594 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1595 page_t *pp, dmu_tx_t *tx)
1604 err = dmu_buf_hold_array(os, object, offset, size,
1605 FALSE, FTAG, &numbufs, &dbp);
1609 for (i = 0; i < numbufs; i++) {
1610 int tocpy, copied, thiscpy;
1612 dmu_buf_t *db = dbp[i];
1616 ASSERT3U(db->db_size, >=, PAGESIZE);
1618 bufoff = offset - db->db_offset;
1619 tocpy = (int)MIN(db->db_size - bufoff, size);
1621 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1623 if (tocpy == db->db_size)
1624 dmu_buf_will_fill(db, tx);
1626 dmu_buf_will_dirty(db, tx);
1628 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1629 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
1630 thiscpy = MIN(PAGESIZE, tocpy - copied);
1631 va = zfs_map_page(pp, S_READ);
1632 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1633 zfs_unmap_page(pp, va);
1638 if (tocpy == db->db_size)
1639 dmu_buf_fill_done(db, tx);
1644 dmu_buf_rele_array(dbp, numbufs, FTAG);
1648 #else /* !illumos */
1651 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
1652 vm_page_t *ma, dmu_tx_t *tx)
1662 err = dmu_buf_hold_array(os, object, offset, size,
1663 FALSE, FTAG, &numbufs, &dbp);
1667 for (i = 0; i < numbufs; i++) {
1668 int tocpy, copied, thiscpy;
1670 dmu_buf_t *db = dbp[i];
1674 ASSERT3U(db->db_size, >=, PAGESIZE);
1676 bufoff = offset - db->db_offset;
1677 tocpy = (int)MIN(db->db_size - bufoff, size);
1679 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
1681 if (tocpy == db->db_size)
1682 dmu_buf_will_fill(db, tx);
1684 dmu_buf_will_dirty(db, tx);
1686 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
1687 ASSERT3U(ptoa((*ma)->pindex), ==, db->db_offset + bufoff);
1688 thiscpy = MIN(PAGESIZE, tocpy - copied);
1689 va = zfs_map_page(*ma, &sf);
1690 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
1696 if (tocpy == db->db_size)
1697 dmu_buf_fill_done(db, tx);
1702 dmu_buf_rele_array(dbp, numbufs, FTAG);
1707 dmu_read_pages(objset_t *os, uint64_t object, vm_page_t *ma, int count,
1708 int *rbehind, int *rahead, int last_size)
1717 int bufoff, pgoff, tocpy;
1721 ASSERT3U(ma[0]->pindex + count - 1, ==, ma[count - 1]->pindex);
1722 ASSERT(last_size <= PAGE_SIZE);
1724 err = dmu_buf_hold_array(os, object, IDX_TO_OFF(ma[0]->pindex),
1725 IDX_TO_OFF(count - 1) + last_size, TRUE, FTAG, &numbufs, &dbp);
1730 IMPLY(last_size < PAGE_SIZE, *rahead == 0);
1731 if (dbp[0]->db_offset != 0 || numbufs > 1) {
1732 for (i = 0; i < numbufs; i++) {
1733 ASSERT(ISP2(dbp[i]->db_size));
1734 ASSERT((dbp[i]->db_offset % dbp[i]->db_size) == 0);
1735 ASSERT3U(dbp[i]->db_size, ==, dbp[0]->db_size);
1740 vmobj = ma[0]->object;
1741 zfs_vmobject_wlock(vmobj);
1744 for (i = 0; i < *rbehind; i++) {
1745 m = vm_page_grab(vmobj, ma[0]->pindex - 1 - i,
1746 VM_ALLOC_NORMAL | VM_ALLOC_NOWAIT | VM_ALLOC_NOBUSY);
1749 if (m->valid != 0) {
1750 ASSERT3U(m->valid, ==, VM_PAGE_BITS_ALL);
1753 ASSERT(m->dirty == 0);
1754 ASSERT(!pmap_page_is_mapped(m));
1756 ASSERT(db->db_size > PAGE_SIZE);
1757 bufoff = IDX_TO_OFF(m->pindex) % db->db_size;
1758 va = zfs_map_page(m, &sf);
1759 bcopy((char *)db->db_data + bufoff, va, PAGESIZE);
1761 m->valid = VM_PAGE_BITS_ALL;
1763 if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1764 vm_page_activate(m);
1766 vm_page_deactivate(m);
1771 bufoff = IDX_TO_OFF(ma[0]->pindex) % db->db_size;
1773 for (mi = 0, di = 0; mi < count && di < numbufs; ) {
1776 if (m != bogus_page) {
1777 vm_page_assert_xbusied(m);
1778 ASSERT(m->valid == 0);
1779 ASSERT(m->dirty == 0);
1780 ASSERT(!pmap_page_is_mapped(m));
1781 va = zfs_map_page(m, &sf);
1787 if (m != bogus_page) {
1788 ASSERT3U(IDX_TO_OFF(m->pindex) + pgoff, ==,
1789 db->db_offset + bufoff);
1793 * We do not need to clamp the copy size by the file
1794 * size as the last block is zero-filled beyond the
1795 * end of file anyway.
1797 tocpy = MIN(db->db_size - bufoff, PAGESIZE - pgoff);
1798 if (m != bogus_page)
1799 bcopy((char *)db->db_data + bufoff, va + pgoff, tocpy);
1802 ASSERT(pgoff <= PAGESIZE);
1803 if (pgoff == PAGESIZE) {
1804 if (m != bogus_page) {
1806 m->valid = VM_PAGE_BITS_ALL;
1814 ASSERT(bufoff <= db->db_size);
1815 if (bufoff == db->db_size) {
1816 ASSERT(di < numbufs);
1824 * Three possibilities:
1825 * - last requested page ends at a buffer boundary and , thus,
1826 * all pages and buffers have been iterated;
1827 * - all requested pages are filled, but the last buffer
1828 * has not been exhausted;
1829 * the read-ahead is possible only in this case;
1830 * - all buffers have been read, but the last page has not been
1832 * this is only possible if the file has only a single buffer
1833 * with a size that is not a multiple of the page size.
1836 ASSERT(di >= numbufs - 1);
1837 IMPLY(*rahead != 0, di == numbufs - 1);
1838 IMPLY(*rahead != 0, bufoff != 0);
1841 if (di == numbufs) {
1842 ASSERT(mi >= count - 1);
1843 ASSERT(*rahead == 0);
1844 IMPLY(pgoff == 0, mi == count);
1846 ASSERT(mi == count - 1);
1847 ASSERT((dbp[0]->db_size & PAGE_MASK) != 0);
1852 ASSERT(m != bogus_page);
1853 bzero(va + pgoff, PAGESIZE - pgoff);
1855 m->valid = VM_PAGE_BITS_ALL;
1858 for (i = 0; i < *rahead; i++) {
1859 m = vm_page_grab(vmobj, ma[count - 1]->pindex + 1 + i,
1860 VM_ALLOC_NORMAL | VM_ALLOC_NOWAIT | VM_ALLOC_NOBUSY);
1863 if (m->valid != 0) {
1864 ASSERT3U(m->valid, ==, VM_PAGE_BITS_ALL);
1867 ASSERT(m->dirty == 0);
1868 ASSERT(!pmap_page_is_mapped(m));
1870 ASSERT(db->db_size > PAGE_SIZE);
1871 bufoff = IDX_TO_OFF(m->pindex) % db->db_size;
1872 tocpy = MIN(db->db_size - bufoff, PAGESIZE);
1873 va = zfs_map_page(m, &sf);
1874 bcopy((char *)db->db_data + bufoff, va, tocpy);
1875 if (tocpy < PAGESIZE) {
1876 ASSERT(i == *rahead - 1);
1877 ASSERT((db->db_size & PAGE_MASK) != 0);
1878 bzero(va + tocpy, PAGESIZE - tocpy);
1881 m->valid = VM_PAGE_BITS_ALL;
1883 if ((m->busy_lock & VPB_BIT_WAITERS) != 0)
1884 vm_page_activate(m);
1886 vm_page_deactivate(m);
1890 zfs_vmobject_wunlock(vmobj);
1892 dmu_buf_rele_array(dbp, numbufs, FTAG);
1895 #endif /* illumos */
1896 #endif /* _KERNEL */
1899 * Allocate a loaned anonymous arc buffer.
1902 dmu_request_arcbuf(dmu_buf_t *handle, int size)
1904 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle;
1906 return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size));
1910 * Free a loaned arc buffer.
1913 dmu_return_arcbuf(arc_buf_t *buf)
1915 arc_return_buf(buf, FTAG);
1916 arc_buf_destroy(buf, FTAG);
1920 * When possible directly assign passed loaned arc buffer to a dbuf.
1921 * If this is not possible copy the contents of passed arc buf via
1925 dmu_assign_arcbuf_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf,
1929 uint32_t blksz = (uint32_t)arc_buf_lsize(buf);
1932 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1933 blkid = dbuf_whichblock(dn, 0, offset);
1934 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1935 rw_exit(&dn->dn_struct_rwlock);
1938 * We can only assign if the offset is aligned, the arc buf is the
1939 * same size as the dbuf, and the dbuf is not metadata.
1941 if (offset == db->db.db_offset && blksz == db->db.db_size) {
1943 curthread->td_ru.ru_oublock++;
1947 racct_add_force(curproc, RACCT_WRITEBPS, blksz);
1948 racct_add_force(curproc, RACCT_WRITEIOPS, 1);
1949 PROC_UNLOCK(curproc);
1952 #endif /* _KERNEL */
1953 dbuf_assign_arcbuf(db, buf, tx);
1954 dbuf_rele(db, FTAG);
1959 /* compressed bufs must always be assignable to their dbuf */
1960 ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF);
1961 ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED));
1964 object = dn->dn_object;
1966 dbuf_rele(db, FTAG);
1967 dmu_write(os, object, offset, blksz, buf->b_data, tx);
1968 dmu_return_arcbuf(buf);
1969 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1974 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
1977 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle;
1979 DB_DNODE_ENTER(dbuf);
1980 dmu_assign_arcbuf_dnode(DB_DNODE(dbuf), offset, buf, tx);
1981 DB_DNODE_EXIT(dbuf);
1985 dbuf_dirty_record_t *dsa_dr;
1986 dmu_sync_cb_t *dsa_done;
1993 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1995 dmu_sync_arg_t *dsa = varg;
1996 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1997 blkptr_t *bp = zio->io_bp;
1999 if (zio->io_error == 0) {
2000 if (BP_IS_HOLE(bp)) {
2002 * A block of zeros may compress to a hole, but the
2003 * block size still needs to be known for replay.
2005 BP_SET_LSIZE(bp, db->db_size);
2006 } else if (!BP_IS_EMBEDDED(bp)) {
2007 ASSERT(BP_GET_LEVEL(bp) == 0);
2014 dmu_sync_late_arrival_ready(zio_t *zio)
2016 dmu_sync_ready(zio, NULL, zio->io_private);
2021 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
2023 dmu_sync_arg_t *dsa = varg;
2024 dbuf_dirty_record_t *dr = dsa->dsa_dr;
2025 dmu_buf_impl_t *db = dr->dr_dbuf;
2026 zgd_t *zgd = dsa->dsa_zgd;
2029 * Record the vdev(s) backing this blkptr so they can be flushed after
2030 * the writes for the lwb have completed.
2032 if (zio->io_error == 0) {
2033 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
2036 mutex_enter(&db->db_mtx);
2037 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
2038 if (zio->io_error == 0) {
2039 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE);
2040 if (dr->dt.dl.dr_nopwrite) {
2041 blkptr_t *bp = zio->io_bp;
2042 blkptr_t *bp_orig = &zio->io_bp_orig;
2043 uint8_t chksum = BP_GET_CHECKSUM(bp_orig);
2045 ASSERT(BP_EQUAL(bp, bp_orig));
2046 VERIFY(BP_EQUAL(bp, db->db_blkptr));
2047 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF);
2048 ASSERT(zio_checksum_table[chksum].ci_flags &
2049 ZCHECKSUM_FLAG_NOPWRITE);
2051 dr->dt.dl.dr_overridden_by = *zio->io_bp;
2052 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
2053 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
2056 * Old style holes are filled with all zeros, whereas
2057 * new-style holes maintain their lsize, type, level,
2058 * and birth time (see zio_write_compress). While we
2059 * need to reset the BP_SET_LSIZE() call that happened
2060 * in dmu_sync_ready for old style holes, we do *not*
2061 * want to wipe out the information contained in new
2062 * style holes. Thus, only zero out the block pointer if
2063 * it's an old style hole.
2065 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) &&
2066 dr->dt.dl.dr_overridden_by.blk_birth == 0)
2067 BP_ZERO(&dr->dt.dl.dr_overridden_by);
2069 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
2071 cv_broadcast(&db->db_changed);
2072 mutex_exit(&db->db_mtx);
2074 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
2076 kmem_free(dsa, sizeof (*dsa));
2080 dmu_sync_late_arrival_done(zio_t *zio)
2082 blkptr_t *bp = zio->io_bp;
2083 dmu_sync_arg_t *dsa = zio->io_private;
2084 blkptr_t *bp_orig = &zio->io_bp_orig;
2085 zgd_t *zgd = dsa->dsa_zgd;
2087 if (zio->io_error == 0) {
2089 * Record the vdev(s) backing this blkptr so they can be
2090 * flushed after the writes for the lwb have completed.
2092 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
2094 if (!BP_IS_HOLE(bp)) {
2095 ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE));
2096 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig));
2097 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
2098 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
2099 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
2103 dmu_tx_commit(dsa->dsa_tx);
2105 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
2107 abd_put(zio->io_abd);
2108 kmem_free(dsa, sizeof (*dsa));
2112 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
2113 zio_prop_t *zp, zbookmark_phys_t *zb)
2115 dmu_sync_arg_t *dsa;
2118 tx = dmu_tx_create(os);
2119 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
2120 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
2122 /* Make zl_get_data do txg_waited_synced() */
2123 return (SET_ERROR(EIO));
2127 * In order to prevent the zgd's lwb from being free'd prior to
2128 * dmu_sync_late_arrival_done() being called, we have to ensure
2129 * the lwb's "max txg" takes this tx's txg into account.
2131 zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx));
2133 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
2135 dsa->dsa_done = done;
2140 * Since we are currently syncing this txg, it's nontrivial to
2141 * determine what BP to nopwrite against, so we disable nopwrite.
2143 * When syncing, the db_blkptr is initially the BP of the previous
2144 * txg. We can not nopwrite against it because it will be changed
2145 * (this is similar to the non-late-arrival case where the dbuf is
2146 * dirty in a future txg).
2148 * Then dbuf_write_ready() sets bp_blkptr to the location we will write.
2149 * We can not nopwrite against it because although the BP will not
2150 * (typically) be changed, the data has not yet been persisted to this
2153 * Finally, when dbuf_write_done() is called, it is theoretically
2154 * possible to always nopwrite, because the data that was written in
2155 * this txg is the same data that we are trying to write. However we
2156 * would need to check that this dbuf is not dirty in any future
2157 * txg's (as we do in the normal dmu_sync() path). For simplicity, we
2158 * don't nopwrite in this case.
2160 zp->zp_nopwrite = B_FALSE;
2162 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
2163 abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size),
2164 zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp,
2165 dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done,
2166 dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
2172 * Intent log support: sync the block associated with db to disk.
2173 * N.B. and XXX: the caller is responsible for making sure that the
2174 * data isn't changing while dmu_sync() is writing it.
2178 * EEXIST: this txg has already been synced, so there's nothing to do.
2179 * The caller should not log the write.
2181 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
2182 * The caller should not log the write.
2184 * EALREADY: this block is already in the process of being synced.
2185 * The caller should track its progress (somehow).
2187 * EIO: could not do the I/O.
2188 * The caller should do a txg_wait_synced().
2190 * 0: the I/O has been initiated.
2191 * The caller should log this blkptr in the done callback.
2192 * It is possible that the I/O will fail, in which case
2193 * the error will be reported to the done callback and
2194 * propagated to pio from zio_done().
2197 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
2199 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
2200 objset_t *os = db->db_objset;
2201 dsl_dataset_t *ds = os->os_dsl_dataset;
2202 dbuf_dirty_record_t *dr;
2203 dmu_sync_arg_t *dsa;
2204 zbookmark_phys_t zb;
2208 ASSERT(pio != NULL);
2211 SET_BOOKMARK(&zb, ds->ds_object,
2212 db->db.db_object, db->db_level, db->db_blkid);
2216 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp);
2220 * If we're frozen (running ziltest), we always need to generate a bp.
2222 if (txg > spa_freeze_txg(os->os_spa))
2223 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
2226 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
2227 * and us. If we determine that this txg is not yet syncing,
2228 * but it begins to sync a moment later, that's OK because the
2229 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
2231 mutex_enter(&db->db_mtx);
2233 if (txg <= spa_last_synced_txg(os->os_spa)) {
2235 * This txg has already synced. There's nothing to do.
2237 mutex_exit(&db->db_mtx);
2238 return (SET_ERROR(EEXIST));
2241 if (txg <= spa_syncing_txg(os->os_spa)) {
2243 * This txg is currently syncing, so we can't mess with
2244 * the dirty record anymore; just write a new log block.
2246 mutex_exit(&db->db_mtx);
2247 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
2250 dr = db->db_last_dirty;
2251 while (dr && dr->dr_txg != txg)
2256 * There's no dr for this dbuf, so it must have been freed.
2257 * There's no need to log writes to freed blocks, so we're done.
2259 mutex_exit(&db->db_mtx);
2260 return (SET_ERROR(ENOENT));
2263 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg);
2265 if (db->db_blkptr != NULL) {
2267 * We need to fill in zgd_bp with the current blkptr so that
2268 * the nopwrite code can check if we're writing the same
2269 * data that's already on disk. We can only nopwrite if we
2270 * are sure that after making the copy, db_blkptr will not
2271 * change until our i/o completes. We ensure this by
2272 * holding the db_mtx, and only allowing nopwrite if the
2273 * block is not already dirty (see below). This is verified
2274 * by dmu_sync_done(), which VERIFYs that the db_blkptr has
2277 *zgd->zgd_bp = *db->db_blkptr;
2281 * Assume the on-disk data is X, the current syncing data (in
2282 * txg - 1) is Y, and the current in-memory data is Z (currently
2285 * We usually want to perform a nopwrite if X and Z are the
2286 * same. However, if Y is different (i.e. the BP is going to
2287 * change before this write takes effect), then a nopwrite will
2288 * be incorrect - we would override with X, which could have
2289 * been freed when Y was written.
2291 * (Note that this is not a concern when we are nop-writing from
2292 * syncing context, because X and Y must be identical, because
2293 * all previous txgs have been synced.)
2295 * Therefore, we disable nopwrite if the current BP could change
2296 * before this TXG. There are two ways it could change: by
2297 * being dirty (dr_next is non-NULL), or by being freed
2298 * (dnode_block_freed()). This behavior is verified by
2299 * zio_done(), which VERIFYs that the override BP is identical
2300 * to the on-disk BP.
2304 if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid))
2305 zp.zp_nopwrite = B_FALSE;
2308 ASSERT(dr->dr_txg == txg);
2309 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
2310 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
2312 * We have already issued a sync write for this buffer,
2313 * or this buffer has already been synced. It could not
2314 * have been dirtied since, or we would have cleared the state.
2316 mutex_exit(&db->db_mtx);
2317 return (SET_ERROR(EALREADY));
2320 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
2321 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
2322 mutex_exit(&db->db_mtx);
2324 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
2326 dsa->dsa_done = done;
2330 zio_nowait(arc_write(pio, os->os_spa, txg,
2331 zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db),
2332 &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa,
2333 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
2339 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
2345 err = dnode_hold(os, object, FTAG, &dn);
2348 err = dnode_set_blksz(dn, size, ibs, tx);
2349 dnode_rele(dn, FTAG);
2354 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
2360 * Send streams include each object's checksum function. This
2361 * check ensures that the receiving system can understand the
2362 * checksum function transmitted.
2364 ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
2366 VERIFY0(dnode_hold(os, object, FTAG, &dn));
2367 ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
2368 dn->dn_checksum = checksum;
2369 dnode_setdirty(dn, tx);
2370 dnode_rele(dn, FTAG);
2374 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
2380 * Send streams include each object's compression function. This
2381 * check ensures that the receiving system can understand the
2382 * compression function transmitted.
2384 ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
2386 VERIFY0(dnode_hold(os, object, FTAG, &dn));
2387 dn->dn_compress = compress;
2388 dnode_setdirty(dn, tx);
2389 dnode_rele(dn, FTAG);
2392 int zfs_mdcomp_disable = 0;
2393 SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RWTUN,
2394 &zfs_mdcomp_disable, 0, "Disable metadata compression");
2397 * When the "redundant_metadata" property is set to "most", only indirect
2398 * blocks of this level and higher will have an additional ditto block.
2400 int zfs_redundant_metadata_most_ditto_level = 2;
2403 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
2405 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
2406 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) ||
2408 enum zio_checksum checksum = os->os_checksum;
2409 enum zio_compress compress = os->os_compress;
2410 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
2411 boolean_t dedup = B_FALSE;
2412 boolean_t nopwrite = B_FALSE;
2413 boolean_t dedup_verify = os->os_dedup_verify;
2414 int copies = os->os_copies;
2417 * We maintain different write policies for each of the following
2420 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
2421 * 3. all other level 0 blocks
2424 if (zfs_mdcomp_disable) {
2425 compress = ZIO_COMPRESS_EMPTY;
2428 * XXX -- we should design a compression algorithm
2429 * that specializes in arrays of bps.
2431 compress = zio_compress_select(os->os_spa,
2432 ZIO_COMPRESS_ON, ZIO_COMPRESS_ON);
2436 * Metadata always gets checksummed. If the data
2437 * checksum is multi-bit correctable, and it's not a
2438 * ZBT-style checksum, then it's suitable for metadata
2439 * as well. Otherwise, the metadata checksum defaults
2442 if (!(zio_checksum_table[checksum].ci_flags &
2443 ZCHECKSUM_FLAG_METADATA) ||
2444 (zio_checksum_table[checksum].ci_flags &
2445 ZCHECKSUM_FLAG_EMBEDDED))
2446 checksum = ZIO_CHECKSUM_FLETCHER_4;
2448 if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
2449 (os->os_redundant_metadata ==
2450 ZFS_REDUNDANT_METADATA_MOST &&
2451 (level >= zfs_redundant_metadata_most_ditto_level ||
2452 DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))))
2454 } else if (wp & WP_NOFILL) {
2458 * If we're writing preallocated blocks, we aren't actually
2459 * writing them so don't set any policy properties. These
2460 * blocks are currently only used by an external subsystem
2461 * outside of zfs (i.e. dump) and not written by the zio
2464 compress = ZIO_COMPRESS_OFF;
2465 checksum = ZIO_CHECKSUM_NOPARITY;
2467 compress = zio_compress_select(os->os_spa, dn->dn_compress,
2470 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ?
2471 zio_checksum_select(dn->dn_checksum, checksum) :
2475 * Determine dedup setting. If we are in dmu_sync(),
2476 * we won't actually dedup now because that's all
2477 * done in syncing context; but we do want to use the
2478 * dedup checkum. If the checksum is not strong
2479 * enough to ensure unique signatures, force
2482 if (dedup_checksum != ZIO_CHECKSUM_OFF) {
2483 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE;
2484 if (!(zio_checksum_table[checksum].ci_flags &
2485 ZCHECKSUM_FLAG_DEDUP))
2486 dedup_verify = B_TRUE;
2490 * Enable nopwrite if we have secure enough checksum
2491 * algorithm (see comment in zio_nop_write) and
2492 * compression is enabled. We don't enable nopwrite if
2493 * dedup is enabled as the two features are mutually
2496 nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags &
2497 ZCHECKSUM_FLAG_NOPWRITE) &&
2498 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
2501 zp->zp_checksum = checksum;
2502 zp->zp_compress = compress;
2503 ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT);
2505 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
2506 zp->zp_level = level;
2507 zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
2508 zp->zp_dedup = dedup;
2509 zp->zp_dedup_verify = dedup && dedup_verify;
2510 zp->zp_nopwrite = nopwrite;
2511 zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ?
2512 os->os_zpl_special_smallblock : 0;
2516 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
2522 * Sync any current changes before
2523 * we go trundling through the block pointers.
2525 err = dmu_object_wait_synced(os, object);
2530 err = dnode_hold(os, object, FTAG, &dn);
2535 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
2536 dnode_rele(dn, FTAG);
2542 * Given the ZFS object, if it contains any dirty nodes
2543 * this function flushes all dirty blocks to disk. This
2544 * ensures the DMU object info is updated. A more efficient
2545 * future version might just find the TXG with the maximum
2546 * ID and wait for that to be synced.
2549 dmu_object_wait_synced(objset_t *os, uint64_t object)
2554 error = dnode_hold(os, object, FTAG, &dn);
2559 for (i = 0; i < TXG_SIZE; i++) {
2560 if (list_link_active(&dn->dn_dirty_link[i])) {
2564 dnode_rele(dn, FTAG);
2565 if (i != TXG_SIZE) {
2566 txg_wait_synced(dmu_objset_pool(os), 0);
2573 __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
2575 dnode_phys_t *dnp = dn->dn_phys;
2577 doi->doi_data_block_size = dn->dn_datablksz;
2578 doi->doi_metadata_block_size = dn->dn_indblkshift ?
2579 1ULL << dn->dn_indblkshift : 0;
2580 doi->doi_type = dn->dn_type;
2581 doi->doi_bonus_type = dn->dn_bonustype;
2582 doi->doi_bonus_size = dn->dn_bonuslen;
2583 doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT;
2584 doi->doi_indirection = dn->dn_nlevels;
2585 doi->doi_checksum = dn->dn_checksum;
2586 doi->doi_compress = dn->dn_compress;
2587 doi->doi_nblkptr = dn->dn_nblkptr;
2588 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
2589 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
2590 doi->doi_fill_count = 0;
2591 for (int i = 0; i < dnp->dn_nblkptr; i++)
2592 doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
2596 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
2598 rw_enter(&dn->dn_struct_rwlock, RW_READER);
2599 mutex_enter(&dn->dn_mtx);
2601 __dmu_object_info_from_dnode(dn, doi);
2603 mutex_exit(&dn->dn_mtx);
2604 rw_exit(&dn->dn_struct_rwlock);
2608 * Get information on a DMU object.
2609 * If doi is NULL, just indicates whether the object exists.
2612 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
2615 int err = dnode_hold(os, object, FTAG, &dn);
2621 dmu_object_info_from_dnode(dn, doi);
2623 dnode_rele(dn, FTAG);
2628 * As above, but faster; can be used when you have a held dbuf in hand.
2631 dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi)
2633 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2636 dmu_object_info_from_dnode(DB_DNODE(db), doi);
2641 * Faster still when you only care about the size.
2642 * This is specifically optimized for zfs_getattr().
2645 dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
2646 u_longlong_t *nblk512)
2648 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2654 *blksize = dn->dn_datablksz;
2655 /* add in number of slots used for the dnode itself */
2656 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
2657 SPA_MINBLOCKSHIFT) + dn->dn_num_slots;
2662 dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize)
2664 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
2669 *dnsize = dn->dn_num_slots << DNODE_SHIFT;
2674 byteswap_uint64_array(void *vbuf, size_t size)
2676 uint64_t *buf = vbuf;
2677 size_t count = size >> 3;
2680 ASSERT((size & 7) == 0);
2682 for (i = 0; i < count; i++)
2683 buf[i] = BSWAP_64(buf[i]);
2687 byteswap_uint32_array(void *vbuf, size_t size)
2689 uint32_t *buf = vbuf;
2690 size_t count = size >> 2;
2693 ASSERT((size & 3) == 0);
2695 for (i = 0; i < count; i++)
2696 buf[i] = BSWAP_32(buf[i]);
2700 byteswap_uint16_array(void *vbuf, size_t size)
2702 uint16_t *buf = vbuf;
2703 size_t count = size >> 1;
2706 ASSERT((size & 1) == 0);
2708 for (i = 0; i < count; i++)
2709 buf[i] = BSWAP_16(buf[i]);
2714 byteswap_uint8_array(void *vbuf, size_t size)
2728 zio_compress_init();
2737 arc_fini(); /* arc depends on l2arc, so arc must go first */
2740 zio_compress_fini();