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.
25 /* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */
26 /* Copyright (c) 2013, Joyent, Inc. All rights reserved. */
27 /* Copyright 2016 Nexenta Systems, Inc. All rights reserved. */
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
33 #include <sys/dnode.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_pool.h>
40 #include <sys/dsl_synctask.h>
41 #include <sys/dsl_prop.h>
42 #include <sys/dmu_zfetch.h>
43 #include <sys/zfs_ioctl.h>
45 #include <sys/zio_checksum.h>
46 #include <sys/zio_compress.h>
48 #include <sys/zfeature.h>
51 #include <sys/vmsystm.h>
52 #include <sys/zfs_znode.h>
56 * Enable/disable nopwrite feature.
58 int zfs_nopwrite_enabled
= 1;
61 * Tunable to control percentage of dirtied blocks from frees in one TXG.
62 * After this threshold is crossed, additional dirty blocks from frees
63 * wait until the next TXG.
64 * A value of zero will disable this throttle.
66 uint32_t zfs_per_txg_dirty_frees_percent
= 30;
69 * This can be used for testing, to ensure that certain actions happen
70 * while in the middle of a remap (which might otherwise complete too
73 int zfs_object_remap_one_indirect_delay_ticks
= 0;
75 const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
] = {
76 { DMU_BSWAP_UINT8
, TRUE
, "unallocated" },
77 { DMU_BSWAP_ZAP
, TRUE
, "object directory" },
78 { DMU_BSWAP_UINT64
, TRUE
, "object array" },
79 { DMU_BSWAP_UINT8
, TRUE
, "packed nvlist" },
80 { DMU_BSWAP_UINT64
, TRUE
, "packed nvlist size" },
81 { DMU_BSWAP_UINT64
, TRUE
, "bpobj" },
82 { DMU_BSWAP_UINT64
, TRUE
, "bpobj header" },
83 { DMU_BSWAP_UINT64
, TRUE
, "SPA space map header" },
84 { DMU_BSWAP_UINT64
, TRUE
, "SPA space map" },
85 { DMU_BSWAP_UINT64
, TRUE
, "ZIL intent log" },
86 { DMU_BSWAP_DNODE
, TRUE
, "DMU dnode" },
87 { DMU_BSWAP_OBJSET
, TRUE
, "DMU objset" },
88 { DMU_BSWAP_UINT64
, TRUE
, "DSL directory" },
89 { DMU_BSWAP_ZAP
, TRUE
, "DSL directory child map"},
90 { DMU_BSWAP_ZAP
, TRUE
, "DSL dataset snap map" },
91 { DMU_BSWAP_ZAP
, TRUE
, "DSL props" },
92 { DMU_BSWAP_UINT64
, TRUE
, "DSL dataset" },
93 { DMU_BSWAP_ZNODE
, TRUE
, "ZFS znode" },
94 { DMU_BSWAP_OLDACL
, TRUE
, "ZFS V0 ACL" },
95 { DMU_BSWAP_UINT8
, FALSE
, "ZFS plain file" },
96 { DMU_BSWAP_ZAP
, TRUE
, "ZFS directory" },
97 { DMU_BSWAP_ZAP
, TRUE
, "ZFS master node" },
98 { DMU_BSWAP_ZAP
, TRUE
, "ZFS delete queue" },
99 { DMU_BSWAP_UINT8
, FALSE
, "zvol object" },
100 { DMU_BSWAP_ZAP
, TRUE
, "zvol prop" },
101 { DMU_BSWAP_UINT8
, FALSE
, "other uint8[]" },
102 { DMU_BSWAP_UINT64
, FALSE
, "other uint64[]" },
103 { DMU_BSWAP_ZAP
, TRUE
, "other ZAP" },
104 { DMU_BSWAP_ZAP
, TRUE
, "persistent error log" },
105 { DMU_BSWAP_UINT8
, TRUE
, "SPA history" },
106 { DMU_BSWAP_UINT64
, TRUE
, "SPA history offsets" },
107 { DMU_BSWAP_ZAP
, TRUE
, "Pool properties" },
108 { DMU_BSWAP_ZAP
, TRUE
, "DSL permissions" },
109 { DMU_BSWAP_ACL
, TRUE
, "ZFS ACL" },
110 { DMU_BSWAP_UINT8
, TRUE
, "ZFS SYSACL" },
111 { DMU_BSWAP_UINT8
, TRUE
, "FUID table" },
112 { DMU_BSWAP_UINT64
, TRUE
, "FUID table size" },
113 { DMU_BSWAP_ZAP
, TRUE
, "DSL dataset next clones"},
114 { DMU_BSWAP_ZAP
, TRUE
, "scan work queue" },
115 { DMU_BSWAP_ZAP
, TRUE
, "ZFS user/group used" },
116 { DMU_BSWAP_ZAP
, TRUE
, "ZFS user/group quota" },
117 { DMU_BSWAP_ZAP
, TRUE
, "snapshot refcount tags"},
118 { DMU_BSWAP_ZAP
, TRUE
, "DDT ZAP algorithm" },
119 { DMU_BSWAP_ZAP
, TRUE
, "DDT statistics" },
120 { DMU_BSWAP_UINT8
, TRUE
, "System attributes" },
121 { DMU_BSWAP_ZAP
, TRUE
, "SA master node" },
122 { DMU_BSWAP_ZAP
, TRUE
, "SA attr registration" },
123 { DMU_BSWAP_ZAP
, TRUE
, "SA attr layouts" },
124 { DMU_BSWAP_ZAP
, TRUE
, "scan translations" },
125 { DMU_BSWAP_UINT8
, FALSE
, "deduplicated block" },
126 { DMU_BSWAP_ZAP
, TRUE
, "DSL deadlist map" },
127 { DMU_BSWAP_UINT64
, TRUE
, "DSL deadlist map hdr" },
128 { DMU_BSWAP_ZAP
, TRUE
, "DSL dir clones" },
129 { DMU_BSWAP_UINT64
, TRUE
, "bpobj subobj" }
132 const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
] = {
133 { byteswap_uint8_array
, "uint8" },
134 { byteswap_uint16_array
, "uint16" },
135 { byteswap_uint32_array
, "uint32" },
136 { byteswap_uint64_array
, "uint64" },
137 { zap_byteswap
, "zap" },
138 { dnode_buf_byteswap
, "dnode" },
139 { dmu_objset_byteswap
, "objset" },
140 { zfs_znode_byteswap
, "znode" },
141 { zfs_oldacl_byteswap
, "oldacl" },
142 { zfs_acl_byteswap
, "acl" }
146 dmu_buf_hold_noread_by_dnode(dnode_t
*dn
, uint64_t offset
,
147 void *tag
, dmu_buf_t
**dbp
)
152 blkid
= dbuf_whichblock(dn
, 0, offset
);
153 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
154 db
= dbuf_hold(dn
, blkid
, tag
);
155 rw_exit(&dn
->dn_struct_rwlock
);
159 return (SET_ERROR(EIO
));
166 dmu_buf_hold_noread(objset_t
*os
, uint64_t object
, uint64_t offset
,
167 void *tag
, dmu_buf_t
**dbp
)
174 err
= dnode_hold(os
, object
, FTAG
, &dn
);
177 blkid
= dbuf_whichblock(dn
, 0, offset
);
178 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
179 db
= dbuf_hold(dn
, blkid
, tag
);
180 rw_exit(&dn
->dn_struct_rwlock
);
181 dnode_rele(dn
, FTAG
);
185 return (SET_ERROR(EIO
));
193 dmu_buf_hold_by_dnode(dnode_t
*dn
, uint64_t offset
,
194 void *tag
, dmu_buf_t
**dbp
, int flags
)
197 int db_flags
= DB_RF_CANFAIL
;
199 if (flags
& DMU_READ_NO_PREFETCH
)
200 db_flags
|= DB_RF_NOPREFETCH
;
202 err
= dmu_buf_hold_noread_by_dnode(dn
, offset
, tag
, dbp
);
204 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)(*dbp
);
205 err
= dbuf_read(db
, NULL
, db_flags
);
216 dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
217 void *tag
, dmu_buf_t
**dbp
, int flags
)
220 int db_flags
= DB_RF_CANFAIL
;
222 if (flags
& DMU_READ_NO_PREFETCH
)
223 db_flags
|= DB_RF_NOPREFETCH
;
225 err
= dmu_buf_hold_noread(os
, object
, offset
, tag
, dbp
);
227 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)(*dbp
);
228 err
= dbuf_read(db
, NULL
, db_flags
);
241 return (DN_MAX_BONUSLEN
);
245 dmu_set_bonus(dmu_buf_t
*db_fake
, int newsize
, dmu_tx_t
*tx
)
247 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
254 if (dn
->dn_bonus
!= db
) {
255 error
= SET_ERROR(EINVAL
);
256 } else if (newsize
< 0 || newsize
> db_fake
->db_size
) {
257 error
= SET_ERROR(EINVAL
);
259 dnode_setbonuslen(dn
, newsize
, tx
);
268 dmu_set_bonustype(dmu_buf_t
*db_fake
, dmu_object_type_t type
, dmu_tx_t
*tx
)
270 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
277 if (!DMU_OT_IS_VALID(type
)) {
278 error
= SET_ERROR(EINVAL
);
279 } else if (dn
->dn_bonus
!= db
) {
280 error
= SET_ERROR(EINVAL
);
282 dnode_setbonus_type(dn
, type
, tx
);
291 dmu_get_bonustype(dmu_buf_t
*db_fake
)
293 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
295 dmu_object_type_t type
;
299 type
= dn
->dn_bonustype
;
306 dmu_rm_spill(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
)
311 error
= dnode_hold(os
, object
, FTAG
, &dn
);
312 dbuf_rm_spill(dn
, tx
);
313 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
314 dnode_rm_spill(dn
, tx
);
315 rw_exit(&dn
->dn_struct_rwlock
);
316 dnode_rele(dn
, FTAG
);
321 * returns ENOENT, EIO, or 0.
324 dmu_bonus_hold(objset_t
*os
, uint64_t object
, void *tag
, dmu_buf_t
**dbp
)
330 error
= dnode_hold(os
, object
, FTAG
, &dn
);
334 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
335 if (dn
->dn_bonus
== NULL
) {
336 rw_exit(&dn
->dn_struct_rwlock
);
337 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
338 if (dn
->dn_bonus
== NULL
)
339 dbuf_create_bonus(dn
);
343 /* as long as the bonus buf is held, the dnode will be held */
344 if (refcount_add(&db
->db_holds
, tag
) == 1) {
345 VERIFY(dnode_add_ref(dn
, db
));
346 atomic_inc_32(&dn
->dn_dbufs_count
);
350 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
351 * hold and incrementing the dbuf count to ensure that dnode_move() sees
352 * a dnode hold for every dbuf.
354 rw_exit(&dn
->dn_struct_rwlock
);
356 dnode_rele(dn
, FTAG
);
358 VERIFY(0 == dbuf_read(db
, NULL
, DB_RF_MUST_SUCCEED
| DB_RF_NOPREFETCH
));
365 * returns ENOENT, EIO, or 0.
367 * This interface will allocate a blank spill dbuf when a spill blk
368 * doesn't already exist on the dnode.
370 * if you only want to find an already existing spill db, then
371 * dmu_spill_hold_existing() should be used.
374 dmu_spill_hold_by_dnode(dnode_t
*dn
, uint32_t flags
, void *tag
, dmu_buf_t
**dbp
)
376 dmu_buf_impl_t
*db
= NULL
;
379 if ((flags
& DB_RF_HAVESTRUCT
) == 0)
380 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
382 db
= dbuf_hold(dn
, DMU_SPILL_BLKID
, tag
);
384 if ((flags
& DB_RF_HAVESTRUCT
) == 0)
385 rw_exit(&dn
->dn_struct_rwlock
);
388 err
= dbuf_read(db
, NULL
, flags
);
397 dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
)
399 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)bonus
;
406 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_SA
) {
407 err
= SET_ERROR(EINVAL
);
409 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
411 if (!dn
->dn_have_spill
) {
412 err
= SET_ERROR(ENOENT
);
414 err
= dmu_spill_hold_by_dnode(dn
,
415 DB_RF_HAVESTRUCT
| DB_RF_CANFAIL
, tag
, dbp
);
418 rw_exit(&dn
->dn_struct_rwlock
);
426 dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
)
428 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)bonus
;
434 err
= dmu_spill_hold_by_dnode(dn
, DB_RF_CANFAIL
, tag
, dbp
);
441 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
442 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
443 * and can induce severe lock contention when writing to several files
444 * whose dnodes are in the same block.
447 dmu_buf_hold_array_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t length
,
448 boolean_t read
, void *tag
, int *numbufsp
, dmu_buf_t
***dbpp
, uint32_t flags
)
451 uint64_t blkid
, nblks
, i
;
456 ASSERT(length
<= DMU_MAX_ACCESS
);
459 * Note: We directly notify the prefetch code of this read, so that
460 * we can tell it about the multi-block read. dbuf_read() only knows
461 * about the one block it is accessing.
463 dbuf_flags
= DB_RF_CANFAIL
| DB_RF_NEVERWAIT
| DB_RF_HAVESTRUCT
|
466 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
467 if (dn
->dn_datablkshift
) {
468 int blkshift
= dn
->dn_datablkshift
;
469 nblks
= (P2ROUNDUP(offset
+ length
, 1ULL << blkshift
) -
470 P2ALIGN(offset
, 1ULL << blkshift
)) >> blkshift
;
472 if (offset
+ length
> dn
->dn_datablksz
) {
473 zfs_panic_recover("zfs: accessing past end of object "
474 "%llx/%llx (size=%u access=%llu+%llu)",
475 (longlong_t
)dn
->dn_objset
->
476 os_dsl_dataset
->ds_object
,
477 (longlong_t
)dn
->dn_object
, dn
->dn_datablksz
,
478 (longlong_t
)offset
, (longlong_t
)length
);
479 rw_exit(&dn
->dn_struct_rwlock
);
480 return (SET_ERROR(EIO
));
484 dbp
= kmem_zalloc(sizeof (dmu_buf_t
*) * nblks
, KM_SLEEP
);
486 zio
= zio_root(dn
->dn_objset
->os_spa
, NULL
, NULL
, ZIO_FLAG_CANFAIL
);
487 blkid
= dbuf_whichblock(dn
, 0, offset
);
488 for (i
= 0; i
< nblks
; i
++) {
489 dmu_buf_impl_t
*db
= dbuf_hold(dn
, blkid
+ i
, tag
);
491 rw_exit(&dn
->dn_struct_rwlock
);
492 dmu_buf_rele_array(dbp
, nblks
, tag
);
494 return (SET_ERROR(EIO
));
497 /* initiate async i/o */
499 (void) dbuf_read(db
, zio
, dbuf_flags
);
503 if ((flags
& DMU_READ_NO_PREFETCH
) == 0 &&
504 DNODE_META_IS_CACHEABLE(dn
) && length
<= zfetch_array_rd_sz
) {
505 dmu_zfetch(&dn
->dn_zfetch
, blkid
, nblks
,
506 read
&& DNODE_IS_CACHEABLE(dn
));
508 rw_exit(&dn
->dn_struct_rwlock
);
510 /* wait for async i/o */
513 dmu_buf_rele_array(dbp
, nblks
, tag
);
517 /* wait for other io to complete */
519 for (i
= 0; i
< nblks
; i
++) {
520 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)dbp
[i
];
521 mutex_enter(&db
->db_mtx
);
522 while (db
->db_state
== DB_READ
||
523 db
->db_state
== DB_FILL
)
524 cv_wait(&db
->db_changed
, &db
->db_mtx
);
525 if (db
->db_state
== DB_UNCACHED
)
526 err
= SET_ERROR(EIO
);
527 mutex_exit(&db
->db_mtx
);
529 dmu_buf_rele_array(dbp
, nblks
, tag
);
541 dmu_buf_hold_array(objset_t
*os
, uint64_t object
, uint64_t offset
,
542 uint64_t length
, int read
, void *tag
, int *numbufsp
, dmu_buf_t
***dbpp
)
547 err
= dnode_hold(os
, object
, FTAG
, &dn
);
551 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, length
, read
, tag
,
552 numbufsp
, dbpp
, DMU_READ_PREFETCH
);
554 dnode_rele(dn
, FTAG
);
560 dmu_buf_hold_array_by_bonus(dmu_buf_t
*db_fake
, uint64_t offset
,
561 uint64_t length
, boolean_t read
, void *tag
, int *numbufsp
,
564 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
570 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, length
, read
, tag
,
571 numbufsp
, dbpp
, DMU_READ_PREFETCH
);
578 dmu_buf_rele_array(dmu_buf_t
**dbp_fake
, int numbufs
, void *tag
)
581 dmu_buf_impl_t
**dbp
= (dmu_buf_impl_t
**)dbp_fake
;
586 for (i
= 0; i
< numbufs
; i
++) {
588 dbuf_rele(dbp
[i
], tag
);
591 kmem_free(dbp
, sizeof (dmu_buf_t
*) * numbufs
);
595 * Issue prefetch i/os for the given blocks. If level is greater than 0, the
596 * indirect blocks prefeteched will be those that point to the blocks containing
597 * the data starting at offset, and continuing to offset + len.
599 * Note that if the indirect blocks above the blocks being prefetched are not in
600 * cache, they will be asychronously read in.
603 dmu_prefetch(objset_t
*os
, uint64_t object
, int64_t level
, uint64_t offset
,
604 uint64_t len
, zio_priority_t pri
)
610 if (len
== 0) { /* they're interested in the bonus buffer */
611 dn
= DMU_META_DNODE(os
);
613 if (object
== 0 || object
>= DN_MAX_OBJECT
)
616 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
617 blkid
= dbuf_whichblock(dn
, level
,
618 object
* sizeof (dnode_phys_t
));
619 dbuf_prefetch(dn
, level
, blkid
, pri
, 0);
620 rw_exit(&dn
->dn_struct_rwlock
);
625 * XXX - Note, if the dnode for the requested object is not
626 * already cached, we will do a *synchronous* read in the
627 * dnode_hold() call. The same is true for any indirects.
629 err
= dnode_hold(os
, object
, FTAG
, &dn
);
633 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
635 * offset + len - 1 is the last byte we want to prefetch for, and offset
636 * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the
637 * last block we want to prefetch, and dbuf_whichblock(dn, level,
638 * offset) is the first. Then the number we need to prefetch is the
641 if (level
> 0 || dn
->dn_datablkshift
!= 0) {
642 nblks
= dbuf_whichblock(dn
, level
, offset
+ len
- 1) -
643 dbuf_whichblock(dn
, level
, offset
) + 1;
645 nblks
= (offset
< dn
->dn_datablksz
);
649 blkid
= dbuf_whichblock(dn
, level
, offset
);
650 for (int i
= 0; i
< nblks
; i
++)
651 dbuf_prefetch(dn
, level
, blkid
+ i
, pri
, 0);
654 rw_exit(&dn
->dn_struct_rwlock
);
656 dnode_rele(dn
, FTAG
);
660 * Get the next "chunk" of file data to free. We traverse the file from
661 * the end so that the file gets shorter over time (if we crashes in the
662 * middle, this will leave us in a better state). We find allocated file
663 * data by simply searching the allocated level 1 indirects.
665 * On input, *start should be the first offset that does not need to be
666 * freed (e.g. "offset + length"). On return, *start will be the first
667 * offset that should be freed.
670 get_next_chunk(dnode_t
*dn
, uint64_t *start
, uint64_t minimum
)
672 uint64_t maxblks
= DMU_MAX_ACCESS
>> (dn
->dn_indblkshift
+ 1);
673 /* bytes of data covered by a level-1 indirect block */
675 dn
->dn_datablksz
* EPB(dn
->dn_indblkshift
, SPA_BLKPTRSHIFT
);
677 ASSERT3U(minimum
, <=, *start
);
679 if (*start
- minimum
<= iblkrange
* maxblks
) {
683 ASSERT(ISP2(iblkrange
));
685 for (uint64_t blks
= 0; *start
> minimum
&& blks
< maxblks
; blks
++) {
689 * dnode_next_offset(BACKWARDS) will find an allocated L1
690 * indirect block at or before the input offset. We must
691 * decrement *start so that it is at the end of the region
695 err
= dnode_next_offset(dn
,
696 DNODE_FIND_BACKWARDS
, start
, 2, 1, 0);
698 /* if there are no indirect blocks before start, we are done */
702 } else if (err
!= 0) {
706 /* set start to the beginning of this L1 indirect */
707 *start
= P2ALIGN(*start
, iblkrange
);
709 if (*start
< minimum
)
715 * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set,
716 * otherwise return false.
717 * Used below in dmu_free_long_range_impl() to enable abort when unmounting
721 dmu_objset_zfs_unmounting(objset_t
*os
)
724 if (dmu_objset_type(os
) == DMU_OST_ZFS
)
725 return (zfs_get_vfs_flag_unmounted(os
));
731 dmu_free_long_range_impl(objset_t
*os
, dnode_t
*dn
, uint64_t offset
,
734 uint64_t object_size
= (dn
->dn_maxblkid
+ 1) * dn
->dn_datablksz
;
736 uint64_t dirty_frees_threshold
;
737 dsl_pool_t
*dp
= dmu_objset_pool(os
);
739 if (offset
>= object_size
)
742 if (zfs_per_txg_dirty_frees_percent
<= 100)
743 dirty_frees_threshold
=
744 zfs_per_txg_dirty_frees_percent
* zfs_dirty_data_max
/ 100;
746 dirty_frees_threshold
= zfs_dirty_data_max
/ 4;
748 if (length
== DMU_OBJECT_END
|| offset
+ length
> object_size
)
749 length
= object_size
- offset
;
751 while (length
!= 0) {
752 uint64_t chunk_end
, chunk_begin
, chunk_len
;
753 uint64_t long_free_dirty_all_txgs
= 0;
756 if (dmu_objset_zfs_unmounting(dn
->dn_objset
))
757 return (SET_ERROR(EINTR
));
759 chunk_end
= chunk_begin
= offset
+ length
;
761 /* move chunk_begin backwards to the beginning of this chunk */
762 err
= get_next_chunk(dn
, &chunk_begin
, offset
);
765 ASSERT3U(chunk_begin
, >=, offset
);
766 ASSERT3U(chunk_begin
, <=, chunk_end
);
768 chunk_len
= chunk_end
- chunk_begin
;
770 mutex_enter(&dp
->dp_lock
);
771 for (int t
= 0; t
< TXG_SIZE
; t
++) {
772 long_free_dirty_all_txgs
+=
773 dp
->dp_long_free_dirty_pertxg
[t
];
775 mutex_exit(&dp
->dp_lock
);
778 * To avoid filling up a TXG with just frees wait for
779 * the next TXG to open before freeing more chunks if
780 * we have reached the threshold of frees
782 if (dirty_frees_threshold
!= 0 &&
783 long_free_dirty_all_txgs
>= dirty_frees_threshold
) {
784 txg_wait_open(dp
, 0);
788 tx
= dmu_tx_create(os
);
789 dmu_tx_hold_free(tx
, dn
->dn_object
, chunk_begin
, chunk_len
);
792 * Mark this transaction as typically resulting in a net
793 * reduction in space used.
795 dmu_tx_mark_netfree(tx
);
796 err
= dmu_tx_assign(tx
, TXG_WAIT
);
802 mutex_enter(&dp
->dp_lock
);
803 dp
->dp_long_free_dirty_pertxg
[dmu_tx_get_txg(tx
) & TXG_MASK
] +=
805 mutex_exit(&dp
->dp_lock
);
806 DTRACE_PROBE3(free__long__range
,
807 uint64_t, long_free_dirty_all_txgs
, uint64_t, chunk_len
,
808 uint64_t, dmu_tx_get_txg(tx
));
809 dnode_free_range(dn
, chunk_begin
, chunk_len
, tx
);
818 dmu_free_long_range(objset_t
*os
, uint64_t object
,
819 uint64_t offset
, uint64_t length
)
824 err
= dnode_hold(os
, object
, FTAG
, &dn
);
827 err
= dmu_free_long_range_impl(os
, dn
, offset
, length
);
830 * It is important to zero out the maxblkid when freeing the entire
831 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
832 * will take the fast path, and (b) dnode_reallocate() can verify
833 * that the entire file has been freed.
835 if (err
== 0 && offset
== 0 && length
== DMU_OBJECT_END
)
838 dnode_rele(dn
, FTAG
);
843 dmu_free_long_object(objset_t
*os
, uint64_t object
)
848 err
= dmu_free_long_range(os
, object
, 0, DMU_OBJECT_END
);
852 tx
= dmu_tx_create(os
);
853 dmu_tx_hold_bonus(tx
, object
);
854 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
855 dmu_tx_mark_netfree(tx
);
856 err
= dmu_tx_assign(tx
, TXG_WAIT
);
858 err
= dmu_object_free(os
, object
, tx
);
868 dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
869 uint64_t size
, dmu_tx_t
*tx
)
872 int err
= dnode_hold(os
, object
, FTAG
, &dn
);
875 ASSERT(offset
< UINT64_MAX
);
876 ASSERT(size
== -1ULL || size
<= UINT64_MAX
- offset
);
877 dnode_free_range(dn
, offset
, size
, tx
);
878 dnode_rele(dn
, FTAG
);
883 dmu_read_impl(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
884 void *buf
, uint32_t flags
)
887 int numbufs
, err
= 0;
890 * Deal with odd block sizes, where there can't be data past the first
891 * block. If we ever do the tail block optimization, we will need to
892 * handle that here as well.
894 if (dn
->dn_maxblkid
== 0) {
895 int newsz
= offset
> dn
->dn_datablksz
? 0 :
896 MIN(size
, dn
->dn_datablksz
- offset
);
897 bzero((char *)buf
+ newsz
, size
- newsz
);
902 uint64_t mylen
= MIN(size
, DMU_MAX_ACCESS
/ 2);
906 * NB: we could do this block-at-a-time, but it's nice
907 * to be reading in parallel.
909 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, mylen
,
910 TRUE
, FTAG
, &numbufs
, &dbp
, flags
);
914 for (i
= 0; i
< numbufs
; i
++) {
917 dmu_buf_t
*db
= dbp
[i
];
921 bufoff
= offset
- db
->db_offset
;
922 tocpy
= (int)MIN(db
->db_size
- bufoff
, size
);
924 bcopy((char *)db
->db_data
+ bufoff
, buf
, tocpy
);
928 buf
= (char *)buf
+ tocpy
;
930 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
936 dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
937 void *buf
, uint32_t flags
)
942 err
= dnode_hold(os
, object
, FTAG
, &dn
);
946 err
= dmu_read_impl(dn
, offset
, size
, buf
, flags
);
947 dnode_rele(dn
, FTAG
);
952 dmu_read_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
, void *buf
,
955 return (dmu_read_impl(dn
, offset
, size
, buf
, flags
));
959 dmu_write_impl(dmu_buf_t
**dbp
, int numbufs
, uint64_t offset
, uint64_t size
,
960 const void *buf
, dmu_tx_t
*tx
)
964 for (i
= 0; i
< numbufs
; i
++) {
967 dmu_buf_t
*db
= dbp
[i
];
971 bufoff
= offset
- db
->db_offset
;
972 tocpy
= (int)MIN(db
->db_size
- bufoff
, size
);
974 ASSERT(i
== 0 || i
== numbufs
-1 || tocpy
== db
->db_size
);
976 if (tocpy
== db
->db_size
)
977 dmu_buf_will_fill(db
, tx
);
979 dmu_buf_will_dirty(db
, tx
);
981 bcopy(buf
, (char *)db
->db_data
+ bufoff
, tocpy
);
983 if (tocpy
== db
->db_size
)
984 dmu_buf_fill_done(db
, tx
);
988 buf
= (char *)buf
+ tocpy
;
993 dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
994 const void *buf
, dmu_tx_t
*tx
)
1002 VERIFY0(dmu_buf_hold_array(os
, object
, offset
, size
,
1003 FALSE
, FTAG
, &numbufs
, &dbp
));
1004 dmu_write_impl(dbp
, numbufs
, offset
, size
, buf
, tx
);
1005 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1009 dmu_write_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
1010 const void *buf
, dmu_tx_t
*tx
)
1018 VERIFY0(dmu_buf_hold_array_by_dnode(dn
, offset
, size
,
1019 FALSE
, FTAG
, &numbufs
, &dbp
, DMU_READ_PREFETCH
));
1020 dmu_write_impl(dbp
, numbufs
, offset
, size
, buf
, tx
);
1021 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1025 dmu_object_remap_one_indirect(objset_t
*os
, dnode_t
*dn
,
1026 uint64_t last_removal_txg
, uint64_t offset
)
1028 uint64_t l1blkid
= dbuf_whichblock(dn
, 1, offset
);
1031 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1032 dmu_buf_impl_t
*dbuf
= dbuf_hold_level(dn
, 1, l1blkid
, FTAG
);
1033 ASSERT3P(dbuf
, !=, NULL
);
1036 * If the block hasn't been written yet, this default will ensure
1037 * we don't try to remap it.
1039 uint64_t birth
= UINT64_MAX
;
1040 ASSERT3U(last_removal_txg
, !=, UINT64_MAX
);
1041 if (dbuf
->db_blkptr
!= NULL
)
1042 birth
= dbuf
->db_blkptr
->blk_birth
;
1043 rw_exit(&dn
->dn_struct_rwlock
);
1046 * If this L1 was already written after the last removal, then we've
1047 * already tried to remap it.
1049 if (birth
<= last_removal_txg
&&
1050 dbuf_read(dbuf
, NULL
, DB_RF_MUST_SUCCEED
) == 0 &&
1051 dbuf_can_remap(dbuf
)) {
1052 dmu_tx_t
*tx
= dmu_tx_create(os
);
1053 dmu_tx_hold_remap_l1indirect(tx
, dn
->dn_object
);
1054 err
= dmu_tx_assign(tx
, TXG_WAIT
);
1056 (void) dbuf_dirty(dbuf
, tx
);
1063 dbuf_rele(dbuf
, FTAG
);
1065 delay(zfs_object_remap_one_indirect_delay_ticks
);
1071 * Remap all blockpointers in the object, if possible, so that they reference
1072 * only concrete vdevs.
1074 * To do this, iterate over the L0 blockpointers and remap any that reference
1075 * an indirect vdev. Note that we only examine L0 blockpointers; since we
1076 * cannot guarantee that we can remap all blockpointer anyways (due to split
1077 * blocks), we do not want to make the code unnecessarily complicated to
1078 * catch the unlikely case that there is an L1 block on an indirect vdev that
1079 * contains no indirect blockpointers.
1082 dmu_object_remap_indirects(objset_t
*os
, uint64_t object
,
1083 uint64_t last_removal_txg
)
1085 uint64_t offset
, l1span
;
1089 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1094 if (dn
->dn_nlevels
<= 1) {
1095 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
1096 err
= SET_ERROR(EINTR
);
1100 * If the dnode has no indirect blocks, we cannot dirty them.
1101 * We still want to remap the blkptr(s) in the dnode if
1102 * appropriate, so mark it as dirty.
1104 if (err
== 0 && dnode_needs_remap(dn
)) {
1105 dmu_tx_t
*tx
= dmu_tx_create(os
);
1106 dmu_tx_hold_bonus(tx
, dn
->dn_object
);
1107 if ((err
= dmu_tx_assign(tx
, TXG_WAIT
)) == 0) {
1108 dnode_setdirty(dn
, tx
);
1115 dnode_rele(dn
, FTAG
);
1120 l1span
= 1ULL << (dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
+
1121 dn
->dn_datablkshift
);
1123 * Find the next L1 indirect that is not a hole.
1125 while (dnode_next_offset(dn
, 0, &offset
, 2, 1, 0) == 0) {
1126 if (issig(JUSTLOOKING
) && issig(FORREAL
)) {
1127 err
= SET_ERROR(EINTR
);
1130 if ((err
= dmu_object_remap_one_indirect(os
, dn
,
1131 last_removal_txg
, offset
)) != 0) {
1137 dnode_rele(dn
, FTAG
);
1142 dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
1151 VERIFY(0 == dmu_buf_hold_array(os
, object
, offset
, size
,
1152 FALSE
, FTAG
, &numbufs
, &dbp
));
1154 for (i
= 0; i
< numbufs
; i
++) {
1155 dmu_buf_t
*db
= dbp
[i
];
1157 dmu_buf_will_not_fill(db
, tx
);
1159 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1163 dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
1164 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
1165 int compressed_size
, int byteorder
, dmu_tx_t
*tx
)
1169 ASSERT3U(etype
, <, NUM_BP_EMBEDDED_TYPES
);
1170 ASSERT3U(comp
, <, ZIO_COMPRESS_FUNCTIONS
);
1171 VERIFY0(dmu_buf_hold_noread(os
, object
, offset
,
1174 dmu_buf_write_embedded(db
,
1175 data
, (bp_embedded_type_t
)etype
, (enum zio_compress
)comp
,
1176 uncompressed_size
, compressed_size
, byteorder
, tx
);
1178 dmu_buf_rele(db
, FTAG
);
1182 * DMU support for xuio
1184 kstat_t
*xuio_ksp
= NULL
;
1187 dmu_xuio_init(xuio_t
*xuio
, int nblk
)
1190 uio_t
*uio
= &xuio
->xu_uio
;
1192 uio
->uio_iovcnt
= nblk
;
1193 uio
->uio_iov
= kmem_zalloc(nblk
* sizeof (iovec_t
), KM_SLEEP
);
1195 priv
= kmem_zalloc(sizeof (dmu_xuio_t
), KM_SLEEP
);
1197 priv
->bufs
= kmem_zalloc(nblk
* sizeof (arc_buf_t
*), KM_SLEEP
);
1198 priv
->iovp
= uio
->uio_iov
;
1199 XUIO_XUZC_PRIV(xuio
) = priv
;
1201 if (XUIO_XUZC_RW(xuio
) == UIO_READ
)
1202 XUIOSTAT_INCR(xuiostat_onloan_rbuf
, nblk
);
1204 XUIOSTAT_INCR(xuiostat_onloan_wbuf
, nblk
);
1210 dmu_xuio_fini(xuio_t
*xuio
)
1212 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1213 int nblk
= priv
->cnt
;
1215 kmem_free(priv
->iovp
, nblk
* sizeof (iovec_t
));
1216 kmem_free(priv
->bufs
, nblk
* sizeof (arc_buf_t
*));
1217 kmem_free(priv
, sizeof (dmu_xuio_t
));
1219 if (XUIO_XUZC_RW(xuio
) == UIO_READ
)
1220 XUIOSTAT_INCR(xuiostat_onloan_rbuf
, -nblk
);
1222 XUIOSTAT_INCR(xuiostat_onloan_wbuf
, -nblk
);
1226 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
1227 * and increase priv->next by 1.
1230 dmu_xuio_add(xuio_t
*xuio
, arc_buf_t
*abuf
, offset_t off
, size_t n
)
1233 uio_t
*uio
= &xuio
->xu_uio
;
1234 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1235 int i
= priv
->next
++;
1237 ASSERT(i
< priv
->cnt
);
1238 ASSERT(off
+ n
<= arc_buf_lsize(abuf
));
1239 iov
= uio
->uio_iov
+ i
;
1240 iov
->iov_base
= (char *)abuf
->b_data
+ off
;
1242 priv
->bufs
[i
] = abuf
;
1247 dmu_xuio_cnt(xuio_t
*xuio
)
1249 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1254 dmu_xuio_arcbuf(xuio_t
*xuio
, int i
)
1256 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1258 ASSERT(i
< priv
->cnt
);
1259 return (priv
->bufs
[i
]);
1263 dmu_xuio_clear(xuio_t
*xuio
, int i
)
1265 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1267 ASSERT(i
< priv
->cnt
);
1268 priv
->bufs
[i
] = NULL
;
1272 xuio_stat_init(void)
1274 xuio_ksp
= kstat_create("zfs", 0, "xuio_stats", "misc",
1275 KSTAT_TYPE_NAMED
, sizeof (xuio_stats
) / sizeof (kstat_named_t
),
1276 KSTAT_FLAG_VIRTUAL
);
1277 if (xuio_ksp
!= NULL
) {
1278 xuio_ksp
->ks_data
= &xuio_stats
;
1279 kstat_install(xuio_ksp
);
1284 xuio_stat_fini(void)
1286 if (xuio_ksp
!= NULL
) {
1287 kstat_delete(xuio_ksp
);
1293 xuio_stat_wbuf_copied(void)
1295 XUIOSTAT_BUMP(xuiostat_wbuf_copied
);
1299 xuio_stat_wbuf_nocopy(void)
1301 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy
);
1306 dmu_read_uio_dnode(dnode_t
*dn
, uio_t
*uio
, uint64_t size
)
1309 int numbufs
, i
, err
;
1310 xuio_t
*xuio
= NULL
;
1313 * NB: we could do this block-at-a-time, but it's nice
1314 * to be reading in parallel.
1316 err
= dmu_buf_hold_array_by_dnode(dn
, uio
->uio_loffset
, size
,
1317 TRUE
, FTAG
, &numbufs
, &dbp
, 0);
1321 if (uio
->uio_extflg
== UIO_XUIO
)
1322 xuio
= (xuio_t
*)uio
;
1324 for (i
= 0; i
< numbufs
; i
++) {
1327 dmu_buf_t
*db
= dbp
[i
];
1331 bufoff
= uio
->uio_loffset
- db
->db_offset
;
1332 tocpy
= (int)MIN(db
->db_size
- bufoff
, size
);
1335 dmu_buf_impl_t
*dbi
= (dmu_buf_impl_t
*)db
;
1336 arc_buf_t
*dbuf_abuf
= dbi
->db_buf
;
1337 arc_buf_t
*abuf
= dbuf_loan_arcbuf(dbi
);
1338 err
= dmu_xuio_add(xuio
, abuf
, bufoff
, tocpy
);
1340 uio
->uio_resid
-= tocpy
;
1341 uio
->uio_loffset
+= tocpy
;
1344 if (abuf
== dbuf_abuf
)
1345 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy
);
1347 XUIOSTAT_BUMP(xuiostat_rbuf_copied
);
1349 err
= uiomove((char *)db
->db_data
+ bufoff
, tocpy
,
1357 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1363 * Read 'size' bytes into the uio buffer.
1364 * From object zdb->db_object.
1365 * Starting at offset uio->uio_loffset.
1367 * If the caller already has a dbuf in the target object
1368 * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1369 * because we don't have to find the dnode_t for the object.
1372 dmu_read_uio_dbuf(dmu_buf_t
*zdb
, uio_t
*uio
, uint64_t size
)
1374 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)zdb
;
1383 err
= dmu_read_uio_dnode(dn
, uio
, size
);
1390 * Read 'size' bytes into the uio buffer.
1391 * From the specified object
1392 * Starting at offset uio->uio_loffset.
1395 dmu_read_uio(objset_t
*os
, uint64_t object
, uio_t
*uio
, uint64_t size
)
1403 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1407 err
= dmu_read_uio_dnode(dn
, uio
, size
);
1409 dnode_rele(dn
, FTAG
);
1415 dmu_write_uio_dnode(dnode_t
*dn
, uio_t
*uio
, uint64_t size
, dmu_tx_t
*tx
)
1422 err
= dmu_buf_hold_array_by_dnode(dn
, uio
->uio_loffset
, size
,
1423 FALSE
, FTAG
, &numbufs
, &dbp
, DMU_READ_PREFETCH
);
1427 for (i
= 0; i
< numbufs
; i
++) {
1430 dmu_buf_t
*db
= dbp
[i
];
1434 bufoff
= uio
->uio_loffset
- db
->db_offset
;
1435 tocpy
= (int)MIN(db
->db_size
- bufoff
, size
);
1437 ASSERT(i
== 0 || i
== numbufs
-1 || tocpy
== db
->db_size
);
1439 if (tocpy
== db
->db_size
)
1440 dmu_buf_will_fill(db
, tx
);
1442 dmu_buf_will_dirty(db
, tx
);
1445 * XXX uiomove could block forever (eg. nfs-backed
1446 * pages). There needs to be a uiolockdown() function
1447 * to lock the pages in memory, so that uiomove won't
1450 err
= uiomove((char *)db
->db_data
+ bufoff
, tocpy
,
1453 if (tocpy
== db
->db_size
)
1454 dmu_buf_fill_done(db
, tx
);
1462 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1467 * Write 'size' bytes from the uio buffer.
1468 * To object zdb->db_object.
1469 * Starting at offset uio->uio_loffset.
1471 * If the caller already has a dbuf in the target object
1472 * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1473 * because we don't have to find the dnode_t for the object.
1476 dmu_write_uio_dbuf(dmu_buf_t
*zdb
, uio_t
*uio
, uint64_t size
,
1479 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)zdb
;
1488 err
= dmu_write_uio_dnode(dn
, uio
, size
, tx
);
1495 * Write 'size' bytes from the uio buffer.
1496 * To the specified object.
1497 * Starting at offset uio->uio_loffset.
1500 dmu_write_uio(objset_t
*os
, uint64_t object
, uio_t
*uio
, uint64_t size
,
1509 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1513 err
= dmu_write_uio_dnode(dn
, uio
, size
, tx
);
1515 dnode_rele(dn
, FTAG
);
1521 dmu_write_pages(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
1522 page_t
*pp
, dmu_tx_t
*tx
)
1531 err
= dmu_buf_hold_array(os
, object
, offset
, size
,
1532 FALSE
, FTAG
, &numbufs
, &dbp
);
1536 for (i
= 0; i
< numbufs
; i
++) {
1537 int tocpy
, copied
, thiscpy
;
1539 dmu_buf_t
*db
= dbp
[i
];
1543 ASSERT3U(db
->db_size
, >=, PAGESIZE
);
1545 bufoff
= offset
- db
->db_offset
;
1546 tocpy
= (int)MIN(db
->db_size
- bufoff
, size
);
1548 ASSERT(i
== 0 || i
== numbufs
-1 || tocpy
== db
->db_size
);
1550 if (tocpy
== db
->db_size
)
1551 dmu_buf_will_fill(db
, tx
);
1553 dmu_buf_will_dirty(db
, tx
);
1555 for (copied
= 0; copied
< tocpy
; copied
+= PAGESIZE
) {
1556 ASSERT3U(pp
->p_offset
, ==, db
->db_offset
+ bufoff
);
1557 thiscpy
= MIN(PAGESIZE
, tocpy
- copied
);
1558 va
= zfs_map_page(pp
, S_READ
);
1559 bcopy(va
, (char *)db
->db_data
+ bufoff
, thiscpy
);
1560 zfs_unmap_page(pp
, va
);
1565 if (tocpy
== db
->db_size
)
1566 dmu_buf_fill_done(db
, tx
);
1571 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1577 * Allocate a loaned anonymous arc buffer.
1580 dmu_request_arcbuf(dmu_buf_t
*handle
, int size
)
1582 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)handle
;
1584 return (arc_loan_buf(db
->db_objset
->os_spa
, B_FALSE
, size
));
1588 * Free a loaned arc buffer.
1591 dmu_return_arcbuf(arc_buf_t
*buf
)
1593 arc_return_buf(buf
, FTAG
);
1594 arc_buf_destroy(buf
, FTAG
);
1598 * When possible directly assign passed loaned arc buffer to a dbuf.
1599 * If this is not possible copy the contents of passed arc buf via
1603 dmu_assign_arcbuf(dmu_buf_t
*handle
, uint64_t offset
, arc_buf_t
*buf
,
1606 dmu_buf_impl_t
*dbuf
= (dmu_buf_impl_t
*)handle
;
1609 uint32_t blksz
= (uint32_t)arc_buf_lsize(buf
);
1612 DB_DNODE_ENTER(dbuf
);
1613 dn
= DB_DNODE(dbuf
);
1614 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1615 blkid
= dbuf_whichblock(dn
, 0, offset
);
1616 VERIFY((db
= dbuf_hold(dn
, blkid
, FTAG
)) != NULL
);
1617 rw_exit(&dn
->dn_struct_rwlock
);
1618 DB_DNODE_EXIT(dbuf
);
1621 * We can only assign if the offset is aligned, the arc buf is the
1622 * same size as the dbuf, and the dbuf is not metadata.
1624 if (offset
== db
->db
.db_offset
&& blksz
== db
->db
.db_size
) {
1625 dbuf_assign_arcbuf(db
, buf
, tx
);
1626 dbuf_rele(db
, FTAG
);
1631 /* compressed bufs must always be assignable to their dbuf */
1632 ASSERT3U(arc_get_compression(buf
), ==, ZIO_COMPRESS_OFF
);
1633 ASSERT(!(buf
->b_flags
& ARC_BUF_FLAG_COMPRESSED
));
1635 DB_DNODE_ENTER(dbuf
);
1636 dn
= DB_DNODE(dbuf
);
1638 object
= dn
->dn_object
;
1639 DB_DNODE_EXIT(dbuf
);
1641 dbuf_rele(db
, FTAG
);
1642 dmu_write(os
, object
, offset
, blksz
, buf
->b_data
, tx
);
1643 dmu_return_arcbuf(buf
);
1644 XUIOSTAT_BUMP(xuiostat_wbuf_copied
);
1649 dbuf_dirty_record_t
*dsa_dr
;
1650 dmu_sync_cb_t
*dsa_done
;
1657 dmu_sync_ready(zio_t
*zio
, arc_buf_t
*buf
, void *varg
)
1659 dmu_sync_arg_t
*dsa
= varg
;
1660 dmu_buf_t
*db
= dsa
->dsa_zgd
->zgd_db
;
1661 blkptr_t
*bp
= zio
->io_bp
;
1663 if (zio
->io_error
== 0) {
1664 if (BP_IS_HOLE(bp
)) {
1666 * A block of zeros may compress to a hole, but the
1667 * block size still needs to be known for replay.
1669 BP_SET_LSIZE(bp
, db
->db_size
);
1670 } else if (!BP_IS_EMBEDDED(bp
)) {
1671 ASSERT(BP_GET_LEVEL(bp
) == 0);
1678 dmu_sync_late_arrival_ready(zio_t
*zio
)
1680 dmu_sync_ready(zio
, NULL
, zio
->io_private
);
1685 dmu_sync_done(zio_t
*zio
, arc_buf_t
*buf
, void *varg
)
1687 dmu_sync_arg_t
*dsa
= varg
;
1688 dbuf_dirty_record_t
*dr
= dsa
->dsa_dr
;
1689 dmu_buf_impl_t
*db
= dr
->dr_dbuf
;
1691 mutex_enter(&db
->db_mtx
);
1692 ASSERT(dr
->dt
.dl
.dr_override_state
== DR_IN_DMU_SYNC
);
1693 if (zio
->io_error
== 0) {
1694 dr
->dt
.dl
.dr_nopwrite
= !!(zio
->io_flags
& ZIO_FLAG_NOPWRITE
);
1695 if (dr
->dt
.dl
.dr_nopwrite
) {
1696 blkptr_t
*bp
= zio
->io_bp
;
1697 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1698 uint8_t chksum
= BP_GET_CHECKSUM(bp_orig
);
1700 ASSERT(BP_EQUAL(bp
, bp_orig
));
1701 VERIFY(BP_EQUAL(bp
, db
->db_blkptr
));
1702 ASSERT(zio
->io_prop
.zp_compress
!= ZIO_COMPRESS_OFF
);
1703 ASSERT(zio_checksum_table
[chksum
].ci_flags
&
1704 ZCHECKSUM_FLAG_NOPWRITE
);
1706 dr
->dt
.dl
.dr_overridden_by
= *zio
->io_bp
;
1707 dr
->dt
.dl
.dr_override_state
= DR_OVERRIDDEN
;
1708 dr
->dt
.dl
.dr_copies
= zio
->io_prop
.zp_copies
;
1711 * Old style holes are filled with all zeros, whereas
1712 * new-style holes maintain their lsize, type, level,
1713 * and birth time (see zio_write_compress). While we
1714 * need to reset the BP_SET_LSIZE() call that happened
1715 * in dmu_sync_ready for old style holes, we do *not*
1716 * want to wipe out the information contained in new
1717 * style holes. Thus, only zero out the block pointer if
1718 * it's an old style hole.
1720 if (BP_IS_HOLE(&dr
->dt
.dl
.dr_overridden_by
) &&
1721 dr
->dt
.dl
.dr_overridden_by
.blk_birth
== 0)
1722 BP_ZERO(&dr
->dt
.dl
.dr_overridden_by
);
1724 dr
->dt
.dl
.dr_override_state
= DR_NOT_OVERRIDDEN
;
1726 cv_broadcast(&db
->db_changed
);
1727 mutex_exit(&db
->db_mtx
);
1729 dsa
->dsa_done(dsa
->dsa_zgd
, zio
->io_error
);
1731 kmem_free(dsa
, sizeof (*dsa
));
1735 dmu_sync_late_arrival_done(zio_t
*zio
)
1737 blkptr_t
*bp
= zio
->io_bp
;
1738 dmu_sync_arg_t
*dsa
= zio
->io_private
;
1739 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1741 if (zio
->io_error
== 0 && !BP_IS_HOLE(bp
)) {
1742 ASSERT(!(zio
->io_flags
& ZIO_FLAG_NOPWRITE
));
1743 ASSERT(BP_IS_HOLE(bp_orig
) || !BP_EQUAL(bp
, bp_orig
));
1744 ASSERT(zio
->io_bp
->blk_birth
== zio
->io_txg
);
1745 ASSERT(zio
->io_txg
> spa_syncing_txg(zio
->io_spa
));
1746 zio_free(zio
->io_spa
, zio
->io_txg
, zio
->io_bp
);
1749 dmu_tx_commit(dsa
->dsa_tx
);
1751 dsa
->dsa_done(dsa
->dsa_zgd
, zio
->io_error
);
1753 abd_put(zio
->io_abd
);
1754 kmem_free(dsa
, sizeof (*dsa
));
1758 dmu_sync_late_arrival(zio_t
*pio
, objset_t
*os
, dmu_sync_cb_t
*done
, zgd_t
*zgd
,
1759 zio_prop_t
*zp
, zbookmark_phys_t
*zb
)
1761 dmu_sync_arg_t
*dsa
;
1764 tx
= dmu_tx_create(os
);
1765 dmu_tx_hold_space(tx
, zgd
->zgd_db
->db_size
);
1766 if (dmu_tx_assign(tx
, TXG_WAIT
) != 0) {
1768 /* Make zl_get_data do txg_waited_synced() */
1769 return (SET_ERROR(EIO
));
1773 * In order to prevent the zgd's lwb from being free'd prior to
1774 * dmu_sync_late_arrival_done() being called, we have to ensure
1775 * the lwb's "max txg" takes this tx's txg into account.
1777 zil_lwb_add_txg(zgd
->zgd_lwb
, dmu_tx_get_txg(tx
));
1779 dsa
= kmem_alloc(sizeof (dmu_sync_arg_t
), KM_SLEEP
);
1781 dsa
->dsa_done
= done
;
1786 * Since we are currently syncing this txg, it's nontrivial to
1787 * determine what BP to nopwrite against, so we disable nopwrite.
1789 * When syncing, the db_blkptr is initially the BP of the previous
1790 * txg. We can not nopwrite against it because it will be changed
1791 * (this is similar to the non-late-arrival case where the dbuf is
1792 * dirty in a future txg).
1794 * Then dbuf_write_ready() sets bp_blkptr to the location we will write.
1795 * We can not nopwrite against it because although the BP will not
1796 * (typically) be changed, the data has not yet been persisted to this
1799 * Finally, when dbuf_write_done() is called, it is theoretically
1800 * possible to always nopwrite, because the data that was written in
1801 * this txg is the same data that we are trying to write. However we
1802 * would need to check that this dbuf is not dirty in any future
1803 * txg's (as we do in the normal dmu_sync() path). For simplicity, we
1804 * don't nopwrite in this case.
1806 zp
->zp_nopwrite
= B_FALSE
;
1808 zio_nowait(zio_write(pio
, os
->os_spa
, dmu_tx_get_txg(tx
), zgd
->zgd_bp
,
1809 abd_get_from_buf(zgd
->zgd_db
->db_data
, zgd
->zgd_db
->db_size
),
1810 zgd
->zgd_db
->db_size
, zgd
->zgd_db
->db_size
, zp
,
1811 dmu_sync_late_arrival_ready
, NULL
, NULL
, dmu_sync_late_arrival_done
,
1812 dsa
, ZIO_PRIORITY_SYNC_WRITE
, ZIO_FLAG_CANFAIL
, zb
));
1818 * Intent log support: sync the block associated with db to disk.
1819 * N.B. and XXX: the caller is responsible for making sure that the
1820 * data isn't changing while dmu_sync() is writing it.
1824 * EEXIST: this txg has already been synced, so there's nothing to do.
1825 * The caller should not log the write.
1827 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1828 * The caller should not log the write.
1830 * EALREADY: this block is already in the process of being synced.
1831 * The caller should track its progress (somehow).
1833 * EIO: could not do the I/O.
1834 * The caller should do a txg_wait_synced().
1836 * 0: the I/O has been initiated.
1837 * The caller should log this blkptr in the done callback.
1838 * It is possible that the I/O will fail, in which case
1839 * the error will be reported to the done callback and
1840 * propagated to pio from zio_done().
1843 dmu_sync(zio_t
*pio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
)
1845 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)zgd
->zgd_db
;
1846 objset_t
*os
= db
->db_objset
;
1847 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1848 dbuf_dirty_record_t
*dr
;
1849 dmu_sync_arg_t
*dsa
;
1850 zbookmark_phys_t zb
;
1854 ASSERT(pio
!= NULL
);
1857 SET_BOOKMARK(&zb
, ds
->ds_object
,
1858 db
->db
.db_object
, db
->db_level
, db
->db_blkid
);
1862 dmu_write_policy(os
, dn
, db
->db_level
, WP_DMU_SYNC
, &zp
);
1866 * If we're frozen (running ziltest), we always need to generate a bp.
1868 if (txg
> spa_freeze_txg(os
->os_spa
))
1869 return (dmu_sync_late_arrival(pio
, os
, done
, zgd
, &zp
, &zb
));
1872 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1873 * and us. If we determine that this txg is not yet syncing,
1874 * but it begins to sync a moment later, that's OK because the
1875 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1877 mutex_enter(&db
->db_mtx
);
1879 if (txg
<= spa_last_synced_txg(os
->os_spa
)) {
1881 * This txg has already synced. There's nothing to do.
1883 mutex_exit(&db
->db_mtx
);
1884 return (SET_ERROR(EEXIST
));
1887 if (txg
<= spa_syncing_txg(os
->os_spa
)) {
1889 * This txg is currently syncing, so we can't mess with
1890 * the dirty record anymore; just write a new log block.
1892 mutex_exit(&db
->db_mtx
);
1893 return (dmu_sync_late_arrival(pio
, os
, done
, zgd
, &zp
, &zb
));
1896 dr
= db
->db_last_dirty
;
1897 while (dr
&& dr
->dr_txg
!= txg
)
1902 * There's no dr for this dbuf, so it must have been freed.
1903 * There's no need to log writes to freed blocks, so we're done.
1905 mutex_exit(&db
->db_mtx
);
1906 return (SET_ERROR(ENOENT
));
1909 ASSERT(dr
->dr_next
== NULL
|| dr
->dr_next
->dr_txg
< txg
);
1911 if (db
->db_blkptr
!= NULL
) {
1913 * We need to fill in zgd_bp with the current blkptr so that
1914 * the nopwrite code can check if we're writing the same
1915 * data that's already on disk. We can only nopwrite if we
1916 * are sure that after making the copy, db_blkptr will not
1917 * change until our i/o completes. We ensure this by
1918 * holding the db_mtx, and only allowing nopwrite if the
1919 * block is not already dirty (see below). This is verified
1920 * by dmu_sync_done(), which VERIFYs that the db_blkptr has
1923 *zgd
->zgd_bp
= *db
->db_blkptr
;
1927 * Assume the on-disk data is X, the current syncing data (in
1928 * txg - 1) is Y, and the current in-memory data is Z (currently
1931 * We usually want to perform a nopwrite if X and Z are the
1932 * same. However, if Y is different (i.e. the BP is going to
1933 * change before this write takes effect), then a nopwrite will
1934 * be incorrect - we would override with X, which could have
1935 * been freed when Y was written.
1937 * (Note that this is not a concern when we are nop-writing from
1938 * syncing context, because X and Y must be identical, because
1939 * all previous txgs have been synced.)
1941 * Therefore, we disable nopwrite if the current BP could change
1942 * before this TXG. There are two ways it could change: by
1943 * being dirty (dr_next is non-NULL), or by being freed
1944 * (dnode_block_freed()). This behavior is verified by
1945 * zio_done(), which VERIFYs that the override BP is identical
1946 * to the on-disk BP.
1950 if (dr
->dr_next
!= NULL
|| dnode_block_freed(dn
, db
->db_blkid
))
1951 zp
.zp_nopwrite
= B_FALSE
;
1954 ASSERT(dr
->dr_txg
== txg
);
1955 if (dr
->dt
.dl
.dr_override_state
== DR_IN_DMU_SYNC
||
1956 dr
->dt
.dl
.dr_override_state
== DR_OVERRIDDEN
) {
1958 * We have already issued a sync write for this buffer,
1959 * or this buffer has already been synced. It could not
1960 * have been dirtied since, or we would have cleared the state.
1962 mutex_exit(&db
->db_mtx
);
1963 return (SET_ERROR(EALREADY
));
1966 ASSERT(dr
->dt
.dl
.dr_override_state
== DR_NOT_OVERRIDDEN
);
1967 dr
->dt
.dl
.dr_override_state
= DR_IN_DMU_SYNC
;
1968 mutex_exit(&db
->db_mtx
);
1970 dsa
= kmem_alloc(sizeof (dmu_sync_arg_t
), KM_SLEEP
);
1972 dsa
->dsa_done
= done
;
1976 zio_nowait(arc_write(pio
, os
->os_spa
, txg
,
1977 zgd
->zgd_bp
, dr
->dt
.dl
.dr_data
, DBUF_IS_L2CACHEABLE(db
),
1978 &zp
, dmu_sync_ready
, NULL
, NULL
, dmu_sync_done
, dsa
,
1979 ZIO_PRIORITY_SYNC_WRITE
, ZIO_FLAG_CANFAIL
, &zb
));
1985 dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
, int ibs
,
1991 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1994 err
= dnode_set_blksz(dn
, size
, ibs
, tx
);
1995 dnode_rele(dn
, FTAG
);
2000 dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
2006 * Send streams include each object's checksum function. This
2007 * check ensures that the receiving system can understand the
2008 * checksum function transmitted.
2010 ASSERT3U(checksum
, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
2012 VERIFY0(dnode_hold(os
, object
, FTAG
, &dn
));
2013 ASSERT3U(checksum
, <, ZIO_CHECKSUM_FUNCTIONS
);
2014 dn
->dn_checksum
= checksum
;
2015 dnode_setdirty(dn
, tx
);
2016 dnode_rele(dn
, FTAG
);
2020 dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
2026 * Send streams include each object's compression function. This
2027 * check ensures that the receiving system can understand the
2028 * compression function transmitted.
2030 ASSERT3U(compress
, <, ZIO_COMPRESS_LEGACY_FUNCTIONS
);
2032 VERIFY0(dnode_hold(os
, object
, FTAG
, &dn
));
2033 dn
->dn_compress
= compress
;
2034 dnode_setdirty(dn
, tx
);
2035 dnode_rele(dn
, FTAG
);
2038 int zfs_mdcomp_disable
= 0;
2041 * When the "redundant_metadata" property is set to "most", only indirect
2042 * blocks of this level and higher will have an additional ditto block.
2044 int zfs_redundant_metadata_most_ditto_level
= 2;
2047 dmu_write_policy(objset_t
*os
, dnode_t
*dn
, int level
, int wp
, zio_prop_t
*zp
)
2049 dmu_object_type_t type
= dn
? dn
->dn_type
: DMU_OT_OBJSET
;
2050 boolean_t ismd
= (level
> 0 || DMU_OT_IS_METADATA(type
) ||
2052 enum zio_checksum checksum
= os
->os_checksum
;
2053 enum zio_compress compress
= os
->os_compress
;
2054 enum zio_checksum dedup_checksum
= os
->os_dedup_checksum
;
2055 boolean_t dedup
= B_FALSE
;
2056 boolean_t nopwrite
= B_FALSE
;
2057 boolean_t dedup_verify
= os
->os_dedup_verify
;
2058 int copies
= os
->os_copies
;
2061 * We maintain different write policies for each of the following
2064 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
2065 * 3. all other level 0 blocks
2068 if (zfs_mdcomp_disable
) {
2069 compress
= ZIO_COMPRESS_EMPTY
;
2072 * XXX -- we should design a compression algorithm
2073 * that specializes in arrays of bps.
2075 compress
= zio_compress_select(os
->os_spa
,
2076 ZIO_COMPRESS_ON
, ZIO_COMPRESS_ON
);
2080 * Metadata always gets checksummed. If the data
2081 * checksum is multi-bit correctable, and it's not a
2082 * ZBT-style checksum, then it's suitable for metadata
2083 * as well. Otherwise, the metadata checksum defaults
2086 if (!(zio_checksum_table
[checksum
].ci_flags
&
2087 ZCHECKSUM_FLAG_METADATA
) ||
2088 (zio_checksum_table
[checksum
].ci_flags
&
2089 ZCHECKSUM_FLAG_EMBEDDED
))
2090 checksum
= ZIO_CHECKSUM_FLETCHER_4
;
2092 if (os
->os_redundant_metadata
== ZFS_REDUNDANT_METADATA_ALL
||
2093 (os
->os_redundant_metadata
==
2094 ZFS_REDUNDANT_METADATA_MOST
&&
2095 (level
>= zfs_redundant_metadata_most_ditto_level
||
2096 DMU_OT_IS_METADATA(type
) || (wp
& WP_SPILL
))))
2098 } else if (wp
& WP_NOFILL
) {
2102 * If we're writing preallocated blocks, we aren't actually
2103 * writing them so don't set any policy properties. These
2104 * blocks are currently only used by an external subsystem
2105 * outside of zfs (i.e. dump) and not written by the zio
2108 compress
= ZIO_COMPRESS_OFF
;
2109 checksum
= ZIO_CHECKSUM_NOPARITY
;
2111 compress
= zio_compress_select(os
->os_spa
, dn
->dn_compress
,
2114 checksum
= (dedup_checksum
== ZIO_CHECKSUM_OFF
) ?
2115 zio_checksum_select(dn
->dn_checksum
, checksum
) :
2119 * Determine dedup setting. If we are in dmu_sync(),
2120 * we won't actually dedup now because that's all
2121 * done in syncing context; but we do want to use the
2122 * dedup checkum. If the checksum is not strong
2123 * enough to ensure unique signatures, force
2126 if (dedup_checksum
!= ZIO_CHECKSUM_OFF
) {
2127 dedup
= (wp
& WP_DMU_SYNC
) ? B_FALSE
: B_TRUE
;
2128 if (!(zio_checksum_table
[checksum
].ci_flags
&
2129 ZCHECKSUM_FLAG_DEDUP
))
2130 dedup_verify
= B_TRUE
;
2134 * Enable nopwrite if we have secure enough checksum
2135 * algorithm (see comment in zio_nop_write) and
2136 * compression is enabled. We don't enable nopwrite if
2137 * dedup is enabled as the two features are mutually
2140 nopwrite
= (!dedup
&& (zio_checksum_table
[checksum
].ci_flags
&
2141 ZCHECKSUM_FLAG_NOPWRITE
) &&
2142 compress
!= ZIO_COMPRESS_OFF
&& zfs_nopwrite_enabled
);
2145 zp
->zp_checksum
= checksum
;
2146 zp
->zp_compress
= compress
;
2147 ASSERT3U(zp
->zp_compress
, !=, ZIO_COMPRESS_INHERIT
);
2149 zp
->zp_type
= (wp
& WP_SPILL
) ? dn
->dn_bonustype
: type
;
2150 zp
->zp_level
= level
;
2151 zp
->zp_copies
= MIN(copies
, spa_max_replication(os
->os_spa
));
2152 zp
->zp_dedup
= dedup
;
2153 zp
->zp_dedup_verify
= dedup
&& dedup_verify
;
2154 zp
->zp_nopwrite
= nopwrite
;
2158 dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
, uint64_t *off
)
2164 * Sync any current changes before
2165 * we go trundling through the block pointers.
2167 err
= dmu_object_wait_synced(os
, object
);
2172 err
= dnode_hold(os
, object
, FTAG
, &dn
);
2177 err
= dnode_next_offset(dn
, (hole
? DNODE_FIND_HOLE
: 0), off
, 1, 1, 0);
2178 dnode_rele(dn
, FTAG
);
2184 * Given the ZFS object, if it contains any dirty nodes
2185 * this function flushes all dirty blocks to disk. This
2186 * ensures the DMU object info is updated. A more efficient
2187 * future version might just find the TXG with the maximum
2188 * ID and wait for that to be synced.
2191 dmu_object_wait_synced(objset_t
*os
, uint64_t object
)
2196 error
= dnode_hold(os
, object
, FTAG
, &dn
);
2201 for (i
= 0; i
< TXG_SIZE
; i
++) {
2202 if (list_link_active(&dn
->dn_dirty_link
[i
])) {
2206 dnode_rele(dn
, FTAG
);
2207 if (i
!= TXG_SIZE
) {
2208 txg_wait_synced(dmu_objset_pool(os
), 0);
2215 dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
)
2219 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
2220 mutex_enter(&dn
->dn_mtx
);
2224 doi
->doi_data_block_size
= dn
->dn_datablksz
;
2225 doi
->doi_metadata_block_size
= dn
->dn_indblkshift
?
2226 1ULL << dn
->dn_indblkshift
: 0;
2227 doi
->doi_type
= dn
->dn_type
;
2228 doi
->doi_bonus_type
= dn
->dn_bonustype
;
2229 doi
->doi_bonus_size
= dn
->dn_bonuslen
;
2230 doi
->doi_indirection
= dn
->dn_nlevels
;
2231 doi
->doi_checksum
= dn
->dn_checksum
;
2232 doi
->doi_compress
= dn
->dn_compress
;
2233 doi
->doi_nblkptr
= dn
->dn_nblkptr
;
2234 doi
->doi_physical_blocks_512
= (DN_USED_BYTES(dnp
) + 256) >> 9;
2235 doi
->doi_max_offset
= (dn
->dn_maxblkid
+ 1) * dn
->dn_datablksz
;
2236 doi
->doi_fill_count
= 0;
2237 for (int i
= 0; i
< dnp
->dn_nblkptr
; i
++)
2238 doi
->doi_fill_count
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
2240 mutex_exit(&dn
->dn_mtx
);
2241 rw_exit(&dn
->dn_struct_rwlock
);
2245 * Get information on a DMU object.
2246 * If doi is NULL, just indicates whether the object exists.
2249 dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
)
2252 int err
= dnode_hold(os
, object
, FTAG
, &dn
);
2258 dmu_object_info_from_dnode(dn
, doi
);
2260 dnode_rele(dn
, FTAG
);
2265 * As above, but faster; can be used when you have a held dbuf in hand.
2268 dmu_object_info_from_db(dmu_buf_t
*db_fake
, dmu_object_info_t
*doi
)
2270 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2273 dmu_object_info_from_dnode(DB_DNODE(db
), doi
);
2278 * Faster still when you only care about the size.
2279 * This is specifically optimized for zfs_getattr().
2282 dmu_object_size_from_db(dmu_buf_t
*db_fake
, uint32_t *blksize
,
2283 u_longlong_t
*nblk512
)
2285 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2291 *blksize
= dn
->dn_datablksz
;
2292 /* add 1 for dnode space */
2293 *nblk512
= ((DN_USED_BYTES(dn
->dn_phys
) + SPA_MINBLOCKSIZE
/2) >>
2294 SPA_MINBLOCKSHIFT
) + 1;
2299 byteswap_uint64_array(void *vbuf
, size_t size
)
2301 uint64_t *buf
= vbuf
;
2302 size_t count
= size
>> 3;
2305 ASSERT((size
& 7) == 0);
2307 for (i
= 0; i
< count
; i
++)
2308 buf
[i
] = BSWAP_64(buf
[i
]);
2312 byteswap_uint32_array(void *vbuf
, size_t size
)
2314 uint32_t *buf
= vbuf
;
2315 size_t count
= size
>> 2;
2318 ASSERT((size
& 3) == 0);
2320 for (i
= 0; i
< count
; i
++)
2321 buf
[i
] = BSWAP_32(buf
[i
]);
2325 byteswap_uint16_array(void *vbuf
, size_t size
)
2327 uint16_t *buf
= vbuf
;
2328 size_t count
= size
>> 1;
2331 ASSERT((size
& 1) == 0);
2333 for (i
= 0; i
< count
; i
++)
2334 buf
[i
] = BSWAP_16(buf
[i
]);
2339 byteswap_uint8_array(void *vbuf
, size_t size
)
2361 arc_fini(); /* arc depends on l2arc, so arc must go first */