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) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2016, Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
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>
51 #include <sys/trace_dmu.h>
52 #include <sys/zfs_rlock.h>
54 #include <sys/vmsystm.h>
55 #include <sys/zfs_znode.h>
59 * Enable/disable nopwrite feature.
61 int zfs_nopwrite_enabled
= 1;
64 * Tunable to control percentage of dirtied blocks from frees in one TXG.
65 * After this threshold is crossed, additional dirty blocks from frees
66 * wait until the next TXG.
67 * A value of zero will disable this throttle.
69 unsigned long zfs_per_txg_dirty_frees_percent
= 30;
72 * Enable/disable forcing txg sync when dirty in dmu_offset_next.
74 int zfs_dmu_offset_next_sync
= 0;
76 const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
] = {
77 { DMU_BSWAP_UINT8
, TRUE
, FALSE
, "unallocated" },
78 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "object directory" },
79 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "object array" },
80 { DMU_BSWAP_UINT8
, TRUE
, FALSE
, "packed nvlist" },
81 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "packed nvlist size" },
82 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "bpobj" },
83 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "bpobj header" },
84 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "SPA space map header" },
85 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "SPA space map" },
86 { DMU_BSWAP_UINT64
, TRUE
, TRUE
, "ZIL intent log" },
87 { DMU_BSWAP_DNODE
, TRUE
, TRUE
, "DMU dnode" },
88 { DMU_BSWAP_OBJSET
, TRUE
, FALSE
, "DMU objset" },
89 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "DSL directory" },
90 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DSL directory child map"},
91 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DSL dataset snap map" },
92 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DSL props" },
93 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "DSL dataset" },
94 { DMU_BSWAP_ZNODE
, TRUE
, FALSE
, "ZFS znode" },
95 { DMU_BSWAP_OLDACL
, TRUE
, TRUE
, "ZFS V0 ACL" },
96 { DMU_BSWAP_UINT8
, FALSE
, TRUE
, "ZFS plain file" },
97 { DMU_BSWAP_ZAP
, TRUE
, TRUE
, "ZFS directory" },
98 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "ZFS master node" },
99 { DMU_BSWAP_ZAP
, TRUE
, TRUE
, "ZFS delete queue" },
100 { DMU_BSWAP_UINT8
, FALSE
, TRUE
, "zvol object" },
101 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "zvol prop" },
102 { DMU_BSWAP_UINT8
, FALSE
, TRUE
, "other uint8[]" },
103 { DMU_BSWAP_UINT64
, FALSE
, TRUE
, "other uint64[]" },
104 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "other ZAP" },
105 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "persistent error log" },
106 { DMU_BSWAP_UINT8
, TRUE
, FALSE
, "SPA history" },
107 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "SPA history offsets" },
108 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "Pool properties" },
109 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DSL permissions" },
110 { DMU_BSWAP_ACL
, TRUE
, TRUE
, "ZFS ACL" },
111 { DMU_BSWAP_UINT8
, TRUE
, TRUE
, "ZFS SYSACL" },
112 { DMU_BSWAP_UINT8
, TRUE
, TRUE
, "FUID table" },
113 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "FUID table size" },
114 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DSL dataset next clones"},
115 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "scan work queue" },
116 { DMU_BSWAP_ZAP
, TRUE
, TRUE
, "ZFS user/group/project used" },
117 { DMU_BSWAP_ZAP
, TRUE
, TRUE
, "ZFS user/group/project quota"},
118 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "snapshot refcount tags"},
119 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DDT ZAP algorithm" },
120 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DDT statistics" },
121 { DMU_BSWAP_UINT8
, TRUE
, TRUE
, "System attributes" },
122 { DMU_BSWAP_ZAP
, TRUE
, TRUE
, "SA master node" },
123 { DMU_BSWAP_ZAP
, TRUE
, TRUE
, "SA attr registration" },
124 { DMU_BSWAP_ZAP
, TRUE
, TRUE
, "SA attr layouts" },
125 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "scan translations" },
126 { DMU_BSWAP_UINT8
, FALSE
, TRUE
, "deduplicated block" },
127 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DSL deadlist map" },
128 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "DSL deadlist map hdr" },
129 { DMU_BSWAP_ZAP
, TRUE
, FALSE
, "DSL dir clones" },
130 { DMU_BSWAP_UINT64
, TRUE
, FALSE
, "bpobj subobj" }
133 const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
] = {
134 { byteswap_uint8_array
, "uint8" },
135 { byteswap_uint16_array
, "uint16" },
136 { byteswap_uint32_array
, "uint32" },
137 { byteswap_uint64_array
, "uint64" },
138 { zap_byteswap
, "zap" },
139 { dnode_buf_byteswap
, "dnode" },
140 { dmu_objset_byteswap
, "objset" },
141 { zfs_znode_byteswap
, "znode" },
142 { zfs_oldacl_byteswap
, "oldacl" },
143 { zfs_acl_byteswap
, "acl" }
147 dmu_buf_hold_noread_by_dnode(dnode_t
*dn
, uint64_t offset
,
148 void *tag
, dmu_buf_t
**dbp
)
153 blkid
= dbuf_whichblock(dn
, 0, offset
);
154 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
155 db
= dbuf_hold(dn
, blkid
, tag
);
156 rw_exit(&dn
->dn_struct_rwlock
);
160 return (SET_ERROR(EIO
));
167 dmu_buf_hold_noread(objset_t
*os
, uint64_t object
, uint64_t offset
,
168 void *tag
, dmu_buf_t
**dbp
)
175 err
= dnode_hold(os
, object
, FTAG
, &dn
);
178 blkid
= dbuf_whichblock(dn
, 0, offset
);
179 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
180 db
= dbuf_hold(dn
, blkid
, tag
);
181 rw_exit(&dn
->dn_struct_rwlock
);
182 dnode_rele(dn
, FTAG
);
186 return (SET_ERROR(EIO
));
194 dmu_buf_hold_by_dnode(dnode_t
*dn
, uint64_t offset
,
195 void *tag
, dmu_buf_t
**dbp
, int flags
)
198 int db_flags
= DB_RF_CANFAIL
;
200 if (flags
& DMU_READ_NO_PREFETCH
)
201 db_flags
|= DB_RF_NOPREFETCH
;
202 if (flags
& DMU_READ_NO_DECRYPT
)
203 db_flags
|= DB_RF_NO_DECRYPT
;
205 err
= dmu_buf_hold_noread_by_dnode(dn
, offset
, tag
, dbp
);
207 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)(*dbp
);
208 err
= dbuf_read(db
, NULL
, db_flags
);
219 dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
220 void *tag
, dmu_buf_t
**dbp
, int flags
)
223 int db_flags
= DB_RF_CANFAIL
;
225 if (flags
& DMU_READ_NO_PREFETCH
)
226 db_flags
|= DB_RF_NOPREFETCH
;
227 if (flags
& DMU_READ_NO_DECRYPT
)
228 db_flags
|= DB_RF_NO_DECRYPT
;
230 err
= dmu_buf_hold_noread(os
, object
, offset
, tag
, dbp
);
232 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)(*dbp
);
233 err
= dbuf_read(db
, NULL
, db_flags
);
246 return (DN_OLD_MAX_BONUSLEN
);
250 dmu_set_bonus(dmu_buf_t
*db_fake
, int newsize
, dmu_tx_t
*tx
)
252 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
259 if (dn
->dn_bonus
!= db
) {
260 error
= SET_ERROR(EINVAL
);
261 } else if (newsize
< 0 || newsize
> db_fake
->db_size
) {
262 error
= SET_ERROR(EINVAL
);
264 dnode_setbonuslen(dn
, newsize
, tx
);
273 dmu_set_bonustype(dmu_buf_t
*db_fake
, dmu_object_type_t type
, dmu_tx_t
*tx
)
275 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
282 if (!DMU_OT_IS_VALID(type
)) {
283 error
= SET_ERROR(EINVAL
);
284 } else if (dn
->dn_bonus
!= db
) {
285 error
= SET_ERROR(EINVAL
);
287 dnode_setbonus_type(dn
, type
, tx
);
296 dmu_get_bonustype(dmu_buf_t
*db_fake
)
298 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
300 dmu_object_type_t type
;
304 type
= dn
->dn_bonustype
;
311 dmu_rm_spill(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
)
316 error
= dnode_hold(os
, object
, FTAG
, &dn
);
317 dbuf_rm_spill(dn
, tx
);
318 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
319 dnode_rm_spill(dn
, tx
);
320 rw_exit(&dn
->dn_struct_rwlock
);
321 dnode_rele(dn
, FTAG
);
326 * returns ENOENT, EIO, or 0.
329 dmu_bonus_hold_impl(objset_t
*os
, uint64_t object
, void *tag
, uint32_t flags
,
335 uint32_t db_flags
= DB_RF_MUST_SUCCEED
;
337 if (flags
& DMU_READ_NO_PREFETCH
)
338 db_flags
|= DB_RF_NOPREFETCH
;
339 if (flags
& DMU_READ_NO_DECRYPT
)
340 db_flags
|= DB_RF_NO_DECRYPT
;
342 error
= dnode_hold(os
, object
, FTAG
, &dn
);
346 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
347 if (dn
->dn_bonus
== NULL
) {
348 rw_exit(&dn
->dn_struct_rwlock
);
349 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
350 if (dn
->dn_bonus
== NULL
)
351 dbuf_create_bonus(dn
);
355 /* as long as the bonus buf is held, the dnode will be held */
356 if (refcount_add(&db
->db_holds
, tag
) == 1) {
357 VERIFY(dnode_add_ref(dn
, db
));
358 atomic_inc_32(&dn
->dn_dbufs_count
);
362 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
363 * hold and incrementing the dbuf count to ensure that dnode_move() sees
364 * a dnode hold for every dbuf.
366 rw_exit(&dn
->dn_struct_rwlock
);
368 dnode_rele(dn
, FTAG
);
370 error
= dbuf_read(db
, NULL
, db_flags
);
372 dnode_evict_bonus(dn
);
383 dmu_bonus_hold(objset_t
*os
, uint64_t obj
, void *tag
, dmu_buf_t
**dbp
)
385 return (dmu_bonus_hold_impl(os
, obj
, tag
, DMU_READ_NO_PREFETCH
, dbp
));
389 * returns ENOENT, EIO, or 0.
391 * This interface will allocate a blank spill dbuf when a spill blk
392 * doesn't already exist on the dnode.
394 * if you only want to find an already existing spill db, then
395 * dmu_spill_hold_existing() should be used.
398 dmu_spill_hold_by_dnode(dnode_t
*dn
, uint32_t flags
, void *tag
, dmu_buf_t
**dbp
)
400 dmu_buf_impl_t
*db
= NULL
;
403 if ((flags
& DB_RF_HAVESTRUCT
) == 0)
404 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
406 db
= dbuf_hold(dn
, DMU_SPILL_BLKID
, tag
);
408 if ((flags
& DB_RF_HAVESTRUCT
) == 0)
409 rw_exit(&dn
->dn_struct_rwlock
);
413 return (SET_ERROR(EIO
));
415 err
= dbuf_read(db
, NULL
, flags
);
426 dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
)
428 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)bonus
;
435 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_SA
) {
436 err
= SET_ERROR(EINVAL
);
438 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
440 if (!dn
->dn_have_spill
) {
441 err
= SET_ERROR(ENOENT
);
443 err
= dmu_spill_hold_by_dnode(dn
,
444 DB_RF_HAVESTRUCT
| DB_RF_CANFAIL
, tag
, dbp
);
447 rw_exit(&dn
->dn_struct_rwlock
);
455 dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
)
457 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)bonus
;
463 err
= dmu_spill_hold_by_dnode(dn
, DB_RF_CANFAIL
, tag
, dbp
);
470 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
471 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
472 * and can induce severe lock contention when writing to several files
473 * whose dnodes are in the same block.
476 dmu_buf_hold_array_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t length
,
477 boolean_t read
, void *tag
, int *numbufsp
, dmu_buf_t
***dbpp
, uint32_t flags
)
480 uint64_t blkid
, nblks
, i
;
485 ASSERT(length
<= DMU_MAX_ACCESS
);
488 * Note: We directly notify the prefetch code of this read, so that
489 * we can tell it about the multi-block read. dbuf_read() only knows
490 * about the one block it is accessing.
492 dbuf_flags
= DB_RF_CANFAIL
| DB_RF_NEVERWAIT
| DB_RF_HAVESTRUCT
|
495 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
496 if (dn
->dn_datablkshift
) {
497 int blkshift
= dn
->dn_datablkshift
;
498 nblks
= (P2ROUNDUP(offset
+ length
, 1ULL << blkshift
) -
499 P2ALIGN(offset
, 1ULL << blkshift
)) >> blkshift
;
501 if (offset
+ length
> dn
->dn_datablksz
) {
502 zfs_panic_recover("zfs: accessing past end of object "
503 "%llx/%llx (size=%u access=%llu+%llu)",
504 (longlong_t
)dn
->dn_objset
->
505 os_dsl_dataset
->ds_object
,
506 (longlong_t
)dn
->dn_object
, dn
->dn_datablksz
,
507 (longlong_t
)offset
, (longlong_t
)length
);
508 rw_exit(&dn
->dn_struct_rwlock
);
509 return (SET_ERROR(EIO
));
513 dbp
= kmem_zalloc(sizeof (dmu_buf_t
*) * nblks
, KM_SLEEP
);
515 zio
= zio_root(dn
->dn_objset
->os_spa
, NULL
, NULL
, ZIO_FLAG_CANFAIL
);
516 blkid
= dbuf_whichblock(dn
, 0, offset
);
517 for (i
= 0; i
< nblks
; i
++) {
518 dmu_buf_impl_t
*db
= dbuf_hold(dn
, blkid
+ i
, tag
);
520 rw_exit(&dn
->dn_struct_rwlock
);
521 dmu_buf_rele_array(dbp
, nblks
, tag
);
523 return (SET_ERROR(EIO
));
526 /* initiate async i/o */
528 (void) dbuf_read(db
, zio
, dbuf_flags
);
532 if ((flags
& DMU_READ_NO_PREFETCH
) == 0 &&
533 DNODE_META_IS_CACHEABLE(dn
) && length
<= zfetch_array_rd_sz
) {
534 dmu_zfetch(&dn
->dn_zfetch
, blkid
, nblks
,
535 read
&& DNODE_IS_CACHEABLE(dn
));
537 rw_exit(&dn
->dn_struct_rwlock
);
539 /* wait for async i/o */
542 dmu_buf_rele_array(dbp
, nblks
, tag
);
546 /* wait for other io to complete */
548 for (i
= 0; i
< nblks
; i
++) {
549 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)dbp
[i
];
550 mutex_enter(&db
->db_mtx
);
551 while (db
->db_state
== DB_READ
||
552 db
->db_state
== DB_FILL
)
553 cv_wait(&db
->db_changed
, &db
->db_mtx
);
554 if (db
->db_state
== DB_UNCACHED
)
555 err
= SET_ERROR(EIO
);
556 mutex_exit(&db
->db_mtx
);
558 dmu_buf_rele_array(dbp
, nblks
, tag
);
570 dmu_buf_hold_array(objset_t
*os
, uint64_t object
, uint64_t offset
,
571 uint64_t length
, int read
, void *tag
, int *numbufsp
, dmu_buf_t
***dbpp
)
576 err
= dnode_hold(os
, object
, FTAG
, &dn
);
580 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, length
, read
, tag
,
581 numbufsp
, dbpp
, DMU_READ_PREFETCH
);
583 dnode_rele(dn
, FTAG
);
589 dmu_buf_hold_array_by_bonus(dmu_buf_t
*db_fake
, uint64_t offset
,
590 uint64_t length
, boolean_t read
, void *tag
, int *numbufsp
,
593 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
599 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, length
, read
, tag
,
600 numbufsp
, dbpp
, DMU_READ_PREFETCH
);
607 dmu_buf_rele_array(dmu_buf_t
**dbp_fake
, int numbufs
, void *tag
)
610 dmu_buf_impl_t
**dbp
= (dmu_buf_impl_t
**)dbp_fake
;
615 for (i
= 0; i
< numbufs
; i
++) {
617 dbuf_rele(dbp
[i
], tag
);
620 kmem_free(dbp
, sizeof (dmu_buf_t
*) * numbufs
);
624 * Issue prefetch i/os for the given blocks. If level is greater than 0, the
625 * indirect blocks prefeteched will be those that point to the blocks containing
626 * the data starting at offset, and continuing to offset + len.
628 * Note that if the indirect blocks above the blocks being prefetched are not
629 * in cache, they will be asychronously read in.
632 dmu_prefetch(objset_t
*os
, uint64_t object
, int64_t level
, uint64_t offset
,
633 uint64_t len
, zio_priority_t pri
)
639 if (len
== 0) { /* they're interested in the bonus buffer */
640 dn
= DMU_META_DNODE(os
);
642 if (object
== 0 || object
>= DN_MAX_OBJECT
)
645 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
646 blkid
= dbuf_whichblock(dn
, level
,
647 object
* sizeof (dnode_phys_t
));
648 dbuf_prefetch(dn
, level
, blkid
, pri
, 0);
649 rw_exit(&dn
->dn_struct_rwlock
);
654 * XXX - Note, if the dnode for the requested object is not
655 * already cached, we will do a *synchronous* read in the
656 * dnode_hold() call. The same is true for any indirects.
658 err
= dnode_hold(os
, object
, FTAG
, &dn
);
662 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
664 * offset + len - 1 is the last byte we want to prefetch for, and offset
665 * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the
666 * last block we want to prefetch, and dbuf_whichblock(dn, level,
667 * offset) is the first. Then the number we need to prefetch is the
670 if (level
> 0 || dn
->dn_datablkshift
!= 0) {
671 nblks
= dbuf_whichblock(dn
, level
, offset
+ len
- 1) -
672 dbuf_whichblock(dn
, level
, offset
) + 1;
674 nblks
= (offset
< dn
->dn_datablksz
);
678 blkid
= dbuf_whichblock(dn
, level
, offset
);
679 for (int i
= 0; i
< nblks
; i
++)
680 dbuf_prefetch(dn
, level
, blkid
+ i
, pri
, 0);
683 rw_exit(&dn
->dn_struct_rwlock
);
685 dnode_rele(dn
, FTAG
);
689 * Get the next "chunk" of file data to free. We traverse the file from
690 * the end so that the file gets shorter over time (if we crashes in the
691 * middle, this will leave us in a better state). We find allocated file
692 * data by simply searching the allocated level 1 indirects.
694 * On input, *start should be the first offset that does not need to be
695 * freed (e.g. "offset + length"). On return, *start will be the first
696 * offset that should be freed.
699 get_next_chunk(dnode_t
*dn
, uint64_t *start
, uint64_t minimum
)
701 uint64_t maxblks
= DMU_MAX_ACCESS
>> (dn
->dn_indblkshift
+ 1);
702 /* bytes of data covered by a level-1 indirect block */
704 dn
->dn_datablksz
* EPB(dn
->dn_indblkshift
, SPA_BLKPTRSHIFT
);
706 ASSERT3U(minimum
, <=, *start
);
708 if (*start
- minimum
<= iblkrange
* maxblks
) {
712 ASSERT(ISP2(iblkrange
));
714 for (uint64_t blks
= 0; *start
> minimum
&& blks
< maxblks
; blks
++) {
718 * dnode_next_offset(BACKWARDS) will find an allocated L1
719 * indirect block at or before the input offset. We must
720 * decrement *start so that it is at the end of the region
724 err
= dnode_next_offset(dn
,
725 DNODE_FIND_BACKWARDS
, start
, 2, 1, 0);
727 /* if there are no indirect blocks before start, we are done */
731 } else if (err
!= 0) {
735 /* set start to the beginning of this L1 indirect */
736 *start
= P2ALIGN(*start
, iblkrange
);
738 if (*start
< minimum
)
744 * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set,
745 * otherwise return false.
746 * Used below in dmu_free_long_range_impl() to enable abort when unmounting
750 dmu_objset_zfs_unmounting(objset_t
*os
)
753 if (dmu_objset_type(os
) == DMU_OST_ZFS
)
754 return (zfs_get_vfs_flag_unmounted(os
));
760 dmu_free_long_range_impl(objset_t
*os
, dnode_t
*dn
, uint64_t offset
,
761 uint64_t length
, boolean_t raw
)
763 uint64_t object_size
;
765 uint64_t dirty_frees_threshold
;
766 dsl_pool_t
*dp
= dmu_objset_pool(os
);
769 return (SET_ERROR(EINVAL
));
771 object_size
= (dn
->dn_maxblkid
+ 1) * dn
->dn_datablksz
;
772 if (offset
>= object_size
)
775 if (zfs_per_txg_dirty_frees_percent
<= 100)
776 dirty_frees_threshold
=
777 zfs_per_txg_dirty_frees_percent
* zfs_dirty_data_max
/ 100;
779 dirty_frees_threshold
= zfs_dirty_data_max
/ 4;
781 if (length
== DMU_OBJECT_END
|| offset
+ length
> object_size
)
782 length
= object_size
- offset
;
784 while (length
!= 0) {
785 uint64_t chunk_end
, chunk_begin
, chunk_len
;
786 uint64_t long_free_dirty_all_txgs
= 0;
789 if (dmu_objset_zfs_unmounting(dn
->dn_objset
))
790 return (SET_ERROR(EINTR
));
792 chunk_end
= chunk_begin
= offset
+ length
;
794 /* move chunk_begin backwards to the beginning of this chunk */
795 err
= get_next_chunk(dn
, &chunk_begin
, offset
);
798 ASSERT3U(chunk_begin
, >=, offset
);
799 ASSERT3U(chunk_begin
, <=, chunk_end
);
801 chunk_len
= chunk_end
- chunk_begin
;
803 mutex_enter(&dp
->dp_lock
);
804 for (int t
= 0; t
< TXG_SIZE
; t
++) {
805 long_free_dirty_all_txgs
+=
806 dp
->dp_long_free_dirty_pertxg
[t
];
808 mutex_exit(&dp
->dp_lock
);
811 * To avoid filling up a TXG with just frees wait for
812 * the next TXG to open before freeing more chunks if
813 * we have reached the threshold of frees
815 if (dirty_frees_threshold
!= 0 &&
816 long_free_dirty_all_txgs
>= dirty_frees_threshold
) {
817 txg_wait_open(dp
, 0);
821 tx
= dmu_tx_create(os
);
822 dmu_tx_hold_free(tx
, dn
->dn_object
, chunk_begin
, chunk_len
);
825 * Mark this transaction as typically resulting in a net
826 * reduction in space used.
828 dmu_tx_mark_netfree(tx
);
829 err
= dmu_tx_assign(tx
, TXG_WAIT
);
835 mutex_enter(&dp
->dp_lock
);
836 dp
->dp_long_free_dirty_pertxg
[dmu_tx_get_txg(tx
) & TXG_MASK
] +=
838 mutex_exit(&dp
->dp_lock
);
839 DTRACE_PROBE3(free__long__range
,
840 uint64_t, long_free_dirty_all_txgs
, uint64_t, chunk_len
,
841 uint64_t, dmu_tx_get_txg(tx
));
842 dnode_free_range(dn
, chunk_begin
, chunk_len
, tx
);
844 /* if this is a raw free, mark the dirty record as such */
846 dbuf_dirty_record_t
*dr
= dn
->dn_dbuf
->db_last_dirty
;
848 while (dr
!= NULL
&& dr
->dr_txg
> tx
->tx_txg
)
850 if (dr
!= NULL
&& dr
->dr_txg
== tx
->tx_txg
)
851 dr
->dt
.dl
.dr_raw
= B_TRUE
;
862 dmu_free_long_range(objset_t
*os
, uint64_t object
,
863 uint64_t offset
, uint64_t length
)
868 err
= dnode_hold(os
, object
, FTAG
, &dn
);
871 err
= dmu_free_long_range_impl(os
, dn
, offset
, length
, B_FALSE
);
874 * It is important to zero out the maxblkid when freeing the entire
875 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
876 * will take the fast path, and (b) dnode_reallocate() can verify
877 * that the entire file has been freed.
879 if (err
== 0 && offset
== 0 && length
== DMU_OBJECT_END
)
882 dnode_rele(dn
, FTAG
);
887 * This function is equivalent to dmu_free_long_range(), but also
888 * marks the new dirty record as a raw write.
891 dmu_free_long_range_raw(objset_t
*os
, uint64_t object
,
892 uint64_t offset
, uint64_t length
)
897 err
= dnode_hold(os
, object
, FTAG
, &dn
);
900 err
= dmu_free_long_range_impl(os
, dn
, offset
, length
, B_TRUE
);
903 * It is important to zero out the maxblkid when freeing the entire
904 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
905 * will take the fast path, and (b) dnode_reallocate() can verify
906 * that the entire file has been freed.
908 if (err
== 0 && offset
== 0 && length
== DMU_OBJECT_END
)
911 dnode_rele(dn
, FTAG
);
916 dmu_free_long_object_impl(objset_t
*os
, uint64_t object
, boolean_t raw
)
921 err
= dmu_free_long_range(os
, object
, 0, DMU_OBJECT_END
);
925 tx
= dmu_tx_create(os
);
926 dmu_tx_hold_bonus(tx
, object
);
927 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
928 dmu_tx_mark_netfree(tx
);
929 err
= dmu_tx_assign(tx
, TXG_WAIT
);
932 err
= dmu_object_dirty_raw(os
, object
, tx
);
934 err
= dmu_object_free(os
, object
, tx
);
945 dmu_free_long_object(objset_t
*os
, uint64_t object
)
947 return (dmu_free_long_object_impl(os
, object
, B_FALSE
));
951 dmu_free_long_object_raw(objset_t
*os
, uint64_t object
)
953 return (dmu_free_long_object_impl(os
, object
, B_TRUE
));
958 dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
959 uint64_t size
, dmu_tx_t
*tx
)
962 int err
= dnode_hold(os
, object
, FTAG
, &dn
);
965 ASSERT(offset
< UINT64_MAX
);
966 ASSERT(size
== DMU_OBJECT_END
|| size
<= UINT64_MAX
- offset
);
967 dnode_free_range(dn
, offset
, size
, tx
);
968 dnode_rele(dn
, FTAG
);
973 dmu_read_impl(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
974 void *buf
, uint32_t flags
)
977 int numbufs
, err
= 0;
980 * Deal with odd block sizes, where there can't be data past the first
981 * block. If we ever do the tail block optimization, we will need to
982 * handle that here as well.
984 if (dn
->dn_maxblkid
== 0) {
985 uint64_t newsz
= offset
> dn
->dn_datablksz
? 0 :
986 MIN(size
, dn
->dn_datablksz
- offset
);
987 bzero((char *)buf
+ newsz
, size
- newsz
);
992 uint64_t mylen
= MIN(size
, DMU_MAX_ACCESS
/ 2);
996 * NB: we could do this block-at-a-time, but it's nice
997 * to be reading in parallel.
999 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, mylen
,
1000 TRUE
, FTAG
, &numbufs
, &dbp
, flags
);
1004 for (i
= 0; i
< numbufs
; i
++) {
1007 dmu_buf_t
*db
= dbp
[i
];
1011 bufoff
= offset
- db
->db_offset
;
1012 tocpy
= MIN(db
->db_size
- bufoff
, size
);
1014 (void) memcpy(buf
, (char *)db
->db_data
+ bufoff
, tocpy
);
1018 buf
= (char *)buf
+ tocpy
;
1020 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1026 dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
1027 void *buf
, uint32_t flags
)
1032 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1036 err
= dmu_read_impl(dn
, offset
, size
, buf
, flags
);
1037 dnode_rele(dn
, FTAG
);
1042 dmu_read_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
, void *buf
,
1045 return (dmu_read_impl(dn
, offset
, size
, buf
, flags
));
1049 dmu_write_impl(dmu_buf_t
**dbp
, int numbufs
, uint64_t offset
, uint64_t size
,
1050 const void *buf
, dmu_tx_t
*tx
)
1054 for (i
= 0; i
< numbufs
; i
++) {
1057 dmu_buf_t
*db
= dbp
[i
];
1061 bufoff
= offset
- db
->db_offset
;
1062 tocpy
= MIN(db
->db_size
- bufoff
, size
);
1064 ASSERT(i
== 0 || i
== numbufs
-1 || tocpy
== db
->db_size
);
1066 if (tocpy
== db
->db_size
)
1067 dmu_buf_will_fill(db
, tx
);
1069 dmu_buf_will_dirty(db
, tx
);
1071 (void) memcpy((char *)db
->db_data
+ bufoff
, buf
, tocpy
);
1073 if (tocpy
== db
->db_size
)
1074 dmu_buf_fill_done(db
, tx
);
1078 buf
= (char *)buf
+ tocpy
;
1083 dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
1084 const void *buf
, dmu_tx_t
*tx
)
1092 VERIFY0(dmu_buf_hold_array(os
, object
, offset
, size
,
1093 FALSE
, FTAG
, &numbufs
, &dbp
));
1094 dmu_write_impl(dbp
, numbufs
, offset
, size
, buf
, tx
);
1095 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1099 dmu_write_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
1100 const void *buf
, dmu_tx_t
*tx
)
1108 VERIFY0(dmu_buf_hold_array_by_dnode(dn
, offset
, size
,
1109 FALSE
, FTAG
, &numbufs
, &dbp
, DMU_READ_PREFETCH
));
1110 dmu_write_impl(dbp
, numbufs
, offset
, size
, buf
, tx
);
1111 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1115 dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
1124 VERIFY(0 == dmu_buf_hold_array(os
, object
, offset
, size
,
1125 FALSE
, FTAG
, &numbufs
, &dbp
));
1127 for (i
= 0; i
< numbufs
; i
++) {
1128 dmu_buf_t
*db
= dbp
[i
];
1130 dmu_buf_will_not_fill(db
, tx
);
1132 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1136 dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
1137 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
1138 int compressed_size
, int byteorder
, dmu_tx_t
*tx
)
1142 ASSERT3U(etype
, <, NUM_BP_EMBEDDED_TYPES
);
1143 ASSERT3U(comp
, <, ZIO_COMPRESS_FUNCTIONS
);
1144 VERIFY0(dmu_buf_hold_noread(os
, object
, offset
,
1147 dmu_buf_write_embedded(db
,
1148 data
, (bp_embedded_type_t
)etype
, (enum zio_compress
)comp
,
1149 uncompressed_size
, compressed_size
, byteorder
, tx
);
1151 dmu_buf_rele(db
, FTAG
);
1155 * DMU support for xuio
1157 kstat_t
*xuio_ksp
= NULL
;
1159 typedef struct xuio_stats
{
1160 /* loaned yet not returned arc_buf */
1161 kstat_named_t xuiostat_onloan_rbuf
;
1162 kstat_named_t xuiostat_onloan_wbuf
;
1163 /* whether a copy is made when loaning out a read buffer */
1164 kstat_named_t xuiostat_rbuf_copied
;
1165 kstat_named_t xuiostat_rbuf_nocopy
;
1166 /* whether a copy is made when assigning a write buffer */
1167 kstat_named_t xuiostat_wbuf_copied
;
1168 kstat_named_t xuiostat_wbuf_nocopy
;
1171 static xuio_stats_t xuio_stats
= {
1172 { "onloan_read_buf", KSTAT_DATA_UINT64
},
1173 { "onloan_write_buf", KSTAT_DATA_UINT64
},
1174 { "read_buf_copied", KSTAT_DATA_UINT64
},
1175 { "read_buf_nocopy", KSTAT_DATA_UINT64
},
1176 { "write_buf_copied", KSTAT_DATA_UINT64
},
1177 { "write_buf_nocopy", KSTAT_DATA_UINT64
}
1180 #define XUIOSTAT_INCR(stat, val) \
1181 atomic_add_64(&xuio_stats.stat.value.ui64, (val))
1182 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
1184 #ifdef HAVE_UIO_ZEROCOPY
1186 dmu_xuio_init(xuio_t
*xuio
, int nblk
)
1189 uio_t
*uio
= &xuio
->xu_uio
;
1191 uio
->uio_iovcnt
= nblk
;
1192 uio
->uio_iov
= kmem_zalloc(nblk
* sizeof (iovec_t
), KM_SLEEP
);
1194 priv
= kmem_zalloc(sizeof (dmu_xuio_t
), KM_SLEEP
);
1196 priv
->bufs
= kmem_zalloc(nblk
* sizeof (arc_buf_t
*), KM_SLEEP
);
1197 priv
->iovp
= (iovec_t
*)uio
->uio_iov
;
1198 XUIO_XUZC_PRIV(xuio
) = priv
;
1200 if (XUIO_XUZC_RW(xuio
) == UIO_READ
)
1201 XUIOSTAT_INCR(xuiostat_onloan_rbuf
, nblk
);
1203 XUIOSTAT_INCR(xuiostat_onloan_wbuf
, nblk
);
1209 dmu_xuio_fini(xuio_t
*xuio
)
1211 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1212 int nblk
= priv
->cnt
;
1214 kmem_free(priv
->iovp
, nblk
* sizeof (iovec_t
));
1215 kmem_free(priv
->bufs
, nblk
* sizeof (arc_buf_t
*));
1216 kmem_free(priv
, sizeof (dmu_xuio_t
));
1218 if (XUIO_XUZC_RW(xuio
) == UIO_READ
)
1219 XUIOSTAT_INCR(xuiostat_onloan_rbuf
, -nblk
);
1221 XUIOSTAT_INCR(xuiostat_onloan_wbuf
, -nblk
);
1225 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
1226 * and increase priv->next by 1.
1229 dmu_xuio_add(xuio_t
*xuio
, arc_buf_t
*abuf
, offset_t off
, size_t n
)
1232 uio_t
*uio
= &xuio
->xu_uio
;
1233 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1234 int i
= priv
->next
++;
1236 ASSERT(i
< priv
->cnt
);
1237 ASSERT(off
+ n
<= arc_buf_lsize(abuf
));
1238 iov
= (iovec_t
*)uio
->uio_iov
+ i
;
1239 iov
->iov_base
= (char *)abuf
->b_data
+ off
;
1241 priv
->bufs
[i
] = abuf
;
1246 dmu_xuio_cnt(xuio_t
*xuio
)
1248 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1253 dmu_xuio_arcbuf(xuio_t
*xuio
, int i
)
1255 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1257 ASSERT(i
< priv
->cnt
);
1258 return (priv
->bufs
[i
]);
1262 dmu_xuio_clear(xuio_t
*xuio
, int i
)
1264 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1266 ASSERT(i
< priv
->cnt
);
1267 priv
->bufs
[i
] = NULL
;
1269 #endif /* HAVE_UIO_ZEROCOPY */
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 #ifdef HAVE_UIO_ZEROCOPY
1311 xuio_t
*xuio
= NULL
;
1315 * NB: we could do this block-at-a-time, but it's nice
1316 * to be reading in parallel.
1318 err
= dmu_buf_hold_array_by_dnode(dn
, uio
->uio_loffset
, size
,
1319 TRUE
, FTAG
, &numbufs
, &dbp
, 0);
1323 for (i
= 0; i
< numbufs
; i
++) {
1326 dmu_buf_t
*db
= dbp
[i
];
1330 bufoff
= uio
->uio_loffset
- db
->db_offset
;
1331 tocpy
= MIN(db
->db_size
- bufoff
, size
);
1333 #ifdef HAVE_UIO_ZEROCOPY
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
);
1350 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
= 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
);
1519 #endif /* _KERNEL */
1522 * Allocate a loaned anonymous arc buffer.
1525 dmu_request_arcbuf(dmu_buf_t
*handle
, int size
)
1527 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)handle
;
1529 return (arc_loan_buf(db
->db_objset
->os_spa
, B_FALSE
, size
));
1533 * Free a loaned arc buffer.
1536 dmu_return_arcbuf(arc_buf_t
*buf
)
1538 arc_return_buf(buf
, FTAG
);
1539 arc_buf_destroy(buf
, FTAG
);
1543 dmu_convert_to_raw(dmu_buf_t
*handle
, boolean_t byteorder
, const uint8_t *salt
,
1544 const uint8_t *iv
, const uint8_t *mac
, dmu_tx_t
*tx
)
1546 dmu_object_type_t type
;
1547 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)handle
;
1548 uint64_t dsobj
= dmu_objset_id(db
->db_objset
);
1550 ASSERT3P(db
->db_buf
, !=, NULL
);
1551 ASSERT3U(dsobj
, !=, 0);
1553 dmu_buf_will_change_crypt_params(handle
, tx
);
1556 type
= DB_DNODE(db
)->dn_type
;
1560 * This technically violates the assumption the dmu code makes
1561 * that dnode blocks are only released in syncing context.
1563 (void) arc_release(db
->db_buf
, db
);
1564 arc_convert_to_raw(db
->db_buf
, dsobj
, byteorder
, type
, salt
, iv
, mac
);
1568 dmu_copy_from_buf(objset_t
*os
, uint64_t object
, uint64_t offset
,
1569 dmu_buf_t
*handle
, dmu_tx_t
*tx
)
1571 dmu_buf_t
*dst_handle
;
1572 dmu_buf_impl_t
*dstdb
;
1573 dmu_buf_impl_t
*srcdb
= (dmu_buf_impl_t
*)handle
;
1576 boolean_t byteorder
;
1577 uint8_t salt
[ZIO_DATA_SALT_LEN
];
1578 uint8_t iv
[ZIO_DATA_IV_LEN
];
1579 uint8_t mac
[ZIO_DATA_MAC_LEN
];
1581 ASSERT3P(srcdb
->db_buf
, !=, NULL
);
1583 /* hold the db that we want to write to */
1584 VERIFY0(dmu_buf_hold(os
, object
, offset
, FTAG
, &dst_handle
,
1585 DMU_READ_NO_DECRYPT
));
1586 dstdb
= (dmu_buf_impl_t
*)dst_handle
;
1587 datalen
= arc_buf_size(srcdb
->db_buf
);
1589 /* allocated an arc buffer that matches the type of srcdb->db_buf */
1590 if (arc_is_encrypted(srcdb
->db_buf
)) {
1591 arc_get_raw_params(srcdb
->db_buf
, &byteorder
, salt
, iv
, mac
);
1592 abuf
= arc_loan_raw_buf(os
->os_spa
, dmu_objset_id(os
),
1593 byteorder
, salt
, iv
, mac
, DB_DNODE(dstdb
)->dn_type
,
1594 datalen
, arc_buf_lsize(srcdb
->db_buf
),
1595 arc_get_compression(srcdb
->db_buf
));
1597 /* we won't get a compressed db back from dmu_buf_hold() */
1598 ASSERT3U(arc_get_compression(srcdb
->db_buf
),
1599 ==, ZIO_COMPRESS_OFF
);
1600 abuf
= arc_loan_buf(os
->os_spa
,
1601 DMU_OT_IS_METADATA(DB_DNODE(dstdb
)->dn_type
), datalen
);
1604 ASSERT3U(datalen
, ==, arc_buf_size(abuf
));
1606 /* copy the data to the new buffer and assign it to the dstdb */
1607 bcopy(srcdb
->db_buf
->b_data
, abuf
->b_data
, datalen
);
1608 dbuf_assign_arcbuf(dstdb
, abuf
, tx
);
1609 dmu_buf_rele(dst_handle
, FTAG
);
1613 * When possible directly assign passed loaned arc buffer to a dbuf.
1614 * If this is not possible copy the contents of passed arc buf via
1618 dmu_assign_arcbuf_by_dnode(dnode_t
*dn
, uint64_t offset
, arc_buf_t
*buf
,
1622 objset_t
*os
= dn
->dn_objset
;
1623 uint64_t object
= dn
->dn_object
;
1624 uint32_t blksz
= (uint32_t)arc_buf_lsize(buf
);
1627 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1628 blkid
= dbuf_whichblock(dn
, 0, offset
);
1629 VERIFY((db
= dbuf_hold(dn
, blkid
, FTAG
)) != NULL
);
1630 rw_exit(&dn
->dn_struct_rwlock
);
1633 * We can only assign if the offset is aligned, the arc buf is the
1634 * same size as the dbuf, and the dbuf is not metadata.
1636 if (offset
== db
->db
.db_offset
&& blksz
== db
->db
.db_size
) {
1637 dbuf_assign_arcbuf(db
, buf
, tx
);
1638 dbuf_rele(db
, FTAG
);
1640 /* compressed bufs must always be assignable to their dbuf */
1641 ASSERT3U(arc_get_compression(buf
), ==, ZIO_COMPRESS_OFF
);
1642 ASSERT(!(buf
->b_flags
& ARC_BUF_FLAG_COMPRESSED
));
1644 dbuf_rele(db
, FTAG
);
1645 dmu_write(os
, object
, offset
, blksz
, buf
->b_data
, tx
);
1646 dmu_return_arcbuf(buf
);
1647 XUIOSTAT_BUMP(xuiostat_wbuf_copied
);
1652 dmu_assign_arcbuf_by_dbuf(dmu_buf_t
*handle
, uint64_t offset
, arc_buf_t
*buf
,
1655 dmu_buf_impl_t
*dbuf
= (dmu_buf_impl_t
*)handle
;
1657 DB_DNODE_ENTER(dbuf
);
1658 dmu_assign_arcbuf_by_dnode(DB_DNODE(dbuf
), offset
, buf
, tx
);
1659 DB_DNODE_EXIT(dbuf
);
1663 dbuf_dirty_record_t
*dsa_dr
;
1664 dmu_sync_cb_t
*dsa_done
;
1671 dmu_sync_ready(zio_t
*zio
, arc_buf_t
*buf
, void *varg
)
1673 dmu_sync_arg_t
*dsa
= varg
;
1674 dmu_buf_t
*db
= dsa
->dsa_zgd
->zgd_db
;
1675 blkptr_t
*bp
= zio
->io_bp
;
1677 if (zio
->io_error
== 0) {
1678 if (BP_IS_HOLE(bp
)) {
1680 * A block of zeros may compress to a hole, but the
1681 * block size still needs to be known for replay.
1683 BP_SET_LSIZE(bp
, db
->db_size
);
1684 } else if (!BP_IS_EMBEDDED(bp
)) {
1685 ASSERT(BP_GET_LEVEL(bp
) == 0);
1692 dmu_sync_late_arrival_ready(zio_t
*zio
)
1694 dmu_sync_ready(zio
, NULL
, zio
->io_private
);
1699 dmu_sync_done(zio_t
*zio
, arc_buf_t
*buf
, void *varg
)
1701 dmu_sync_arg_t
*dsa
= varg
;
1702 dbuf_dirty_record_t
*dr
= dsa
->dsa_dr
;
1703 dmu_buf_impl_t
*db
= dr
->dr_dbuf
;
1705 mutex_enter(&db
->db_mtx
);
1706 ASSERT(dr
->dt
.dl
.dr_override_state
== DR_IN_DMU_SYNC
);
1707 if (zio
->io_error
== 0) {
1708 dr
->dt
.dl
.dr_nopwrite
= !!(zio
->io_flags
& ZIO_FLAG_NOPWRITE
);
1709 if (dr
->dt
.dl
.dr_nopwrite
) {
1710 blkptr_t
*bp
= zio
->io_bp
;
1711 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1712 uint8_t chksum
= BP_GET_CHECKSUM(bp_orig
);
1714 ASSERT(BP_EQUAL(bp
, bp_orig
));
1715 VERIFY(BP_EQUAL(bp
, db
->db_blkptr
));
1716 ASSERT(zio
->io_prop
.zp_compress
!= ZIO_COMPRESS_OFF
);
1717 VERIFY(zio_checksum_table
[chksum
].ci_flags
&
1718 ZCHECKSUM_FLAG_NOPWRITE
);
1720 dr
->dt
.dl
.dr_overridden_by
= *zio
->io_bp
;
1721 dr
->dt
.dl
.dr_override_state
= DR_OVERRIDDEN
;
1722 dr
->dt
.dl
.dr_copies
= zio
->io_prop
.zp_copies
;
1725 * Old style holes are filled with all zeros, whereas
1726 * new-style holes maintain their lsize, type, level,
1727 * and birth time (see zio_write_compress). While we
1728 * need to reset the BP_SET_LSIZE() call that happened
1729 * in dmu_sync_ready for old style holes, we do *not*
1730 * want to wipe out the information contained in new
1731 * style holes. Thus, only zero out the block pointer if
1732 * it's an old style hole.
1734 if (BP_IS_HOLE(&dr
->dt
.dl
.dr_overridden_by
) &&
1735 dr
->dt
.dl
.dr_overridden_by
.blk_birth
== 0)
1736 BP_ZERO(&dr
->dt
.dl
.dr_overridden_by
);
1738 dr
->dt
.dl
.dr_override_state
= DR_NOT_OVERRIDDEN
;
1740 cv_broadcast(&db
->db_changed
);
1741 mutex_exit(&db
->db_mtx
);
1743 dsa
->dsa_done(dsa
->dsa_zgd
, zio
->io_error
);
1745 kmem_free(dsa
, sizeof (*dsa
));
1749 dmu_sync_late_arrival_done(zio_t
*zio
)
1751 blkptr_t
*bp
= zio
->io_bp
;
1752 dmu_sync_arg_t
*dsa
= zio
->io_private
;
1753 ASSERTV(blkptr_t
*bp_orig
= &zio
->io_bp_orig
);
1755 if (zio
->io_error
== 0 && !BP_IS_HOLE(bp
)) {
1756 ASSERT(!(zio
->io_flags
& ZIO_FLAG_NOPWRITE
));
1757 ASSERT(BP_IS_HOLE(bp_orig
) || !BP_EQUAL(bp
, bp_orig
));
1758 ASSERT(zio
->io_bp
->blk_birth
== zio
->io_txg
);
1759 ASSERT(zio
->io_txg
> spa_syncing_txg(zio
->io_spa
));
1760 zio_free(zio
->io_spa
, zio
->io_txg
, zio
->io_bp
);
1763 dmu_tx_commit(dsa
->dsa_tx
);
1765 dsa
->dsa_done(dsa
->dsa_zgd
, zio
->io_error
);
1767 abd_put(zio
->io_abd
);
1768 kmem_free(dsa
, sizeof (*dsa
));
1772 dmu_sync_late_arrival(zio_t
*pio
, objset_t
*os
, dmu_sync_cb_t
*done
, zgd_t
*zgd
,
1773 zio_prop_t
*zp
, zbookmark_phys_t
*zb
)
1775 dmu_sync_arg_t
*dsa
;
1778 tx
= dmu_tx_create(os
);
1779 dmu_tx_hold_space(tx
, zgd
->zgd_db
->db_size
);
1780 if (dmu_tx_assign(tx
, TXG_WAIT
) != 0) {
1782 /* Make zl_get_data do txg_waited_synced() */
1783 return (SET_ERROR(EIO
));
1787 * In order to prevent the zgd's lwb from being free'd prior to
1788 * dmu_sync_late_arrival_done() being called, we have to ensure
1789 * the lwb's "max txg" takes this tx's txg into account.
1791 zil_lwb_add_txg(zgd
->zgd_lwb
, dmu_tx_get_txg(tx
));
1793 dsa
= kmem_alloc(sizeof (dmu_sync_arg_t
), KM_SLEEP
);
1795 dsa
->dsa_done
= done
;
1800 * Since we are currently syncing this txg, it's nontrivial to
1801 * determine what BP to nopwrite against, so we disable nopwrite.
1803 * When syncing, the db_blkptr is initially the BP of the previous
1804 * txg. We can not nopwrite against it because it will be changed
1805 * (this is similar to the non-late-arrival case where the dbuf is
1806 * dirty in a future txg).
1808 * Then dbuf_write_ready() sets bp_blkptr to the location we will write.
1809 * We can not nopwrite against it because although the BP will not
1810 * (typically) be changed, the data has not yet been persisted to this
1813 * Finally, when dbuf_write_done() is called, it is theoretically
1814 * possible to always nopwrite, because the data that was written in
1815 * this txg is the same data that we are trying to write. However we
1816 * would need to check that this dbuf is not dirty in any future
1817 * txg's (as we do in the normal dmu_sync() path). For simplicity, we
1818 * don't nopwrite in this case.
1820 zp
->zp_nopwrite
= B_FALSE
;
1822 zio_nowait(zio_write(pio
, os
->os_spa
, dmu_tx_get_txg(tx
), zgd
->zgd_bp
,
1823 abd_get_from_buf(zgd
->zgd_db
->db_data
, zgd
->zgd_db
->db_size
),
1824 zgd
->zgd_db
->db_size
, zgd
->zgd_db
->db_size
, zp
,
1825 dmu_sync_late_arrival_ready
, NULL
, NULL
, dmu_sync_late_arrival_done
,
1826 dsa
, ZIO_PRIORITY_SYNC_WRITE
, ZIO_FLAG_CANFAIL
, zb
));
1832 * Intent log support: sync the block associated with db to disk.
1833 * N.B. and XXX: the caller is responsible for making sure that the
1834 * data isn't changing while dmu_sync() is writing it.
1838 * EEXIST: this txg has already been synced, so there's nothing to do.
1839 * The caller should not log the write.
1841 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1842 * The caller should not log the write.
1844 * EALREADY: this block is already in the process of being synced.
1845 * The caller should track its progress (somehow).
1847 * EIO: could not do the I/O.
1848 * The caller should do a txg_wait_synced().
1850 * 0: the I/O has been initiated.
1851 * The caller should log this blkptr in the done callback.
1852 * It is possible that the I/O will fail, in which case
1853 * the error will be reported to the done callback and
1854 * propagated to pio from zio_done().
1857 dmu_sync(zio_t
*pio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
)
1859 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)zgd
->zgd_db
;
1860 objset_t
*os
= db
->db_objset
;
1861 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1862 dbuf_dirty_record_t
*dr
;
1863 dmu_sync_arg_t
*dsa
;
1864 zbookmark_phys_t zb
;
1868 ASSERT(pio
!= NULL
);
1871 /* dbuf is within the locked range */
1872 ASSERT3U(db
->db
.db_offset
, >=, zgd
->zgd_rl
->r_off
);
1873 ASSERT3U(db
->db
.db_offset
+ db
->db
.db_size
, <=,
1874 zgd
->zgd_rl
->r_off
+ zgd
->zgd_rl
->r_len
);
1876 SET_BOOKMARK(&zb
, ds
->ds_object
,
1877 db
->db
.db_object
, db
->db_level
, db
->db_blkid
);
1881 dmu_write_policy(os
, dn
, db
->db_level
, WP_DMU_SYNC
, &zp
);
1885 * If we're frozen (running ziltest), we always need to generate a bp.
1887 if (txg
> spa_freeze_txg(os
->os_spa
))
1888 return (dmu_sync_late_arrival(pio
, os
, done
, zgd
, &zp
, &zb
));
1891 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1892 * and us. If we determine that this txg is not yet syncing,
1893 * but it begins to sync a moment later, that's OK because the
1894 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1896 mutex_enter(&db
->db_mtx
);
1898 if (txg
<= spa_last_synced_txg(os
->os_spa
)) {
1900 * This txg has already synced. There's nothing to do.
1902 mutex_exit(&db
->db_mtx
);
1903 return (SET_ERROR(EEXIST
));
1906 if (txg
<= spa_syncing_txg(os
->os_spa
)) {
1908 * This txg is currently syncing, so we can't mess with
1909 * the dirty record anymore; just write a new log block.
1911 mutex_exit(&db
->db_mtx
);
1912 return (dmu_sync_late_arrival(pio
, os
, done
, zgd
, &zp
, &zb
));
1915 dr
= db
->db_last_dirty
;
1916 while (dr
&& dr
->dr_txg
!= txg
)
1921 * There's no dr for this dbuf, so it must have been freed.
1922 * There's no need to log writes to freed blocks, so we're done.
1924 mutex_exit(&db
->db_mtx
);
1925 return (SET_ERROR(ENOENT
));
1928 ASSERT(dr
->dr_next
== NULL
|| dr
->dr_next
->dr_txg
< txg
);
1930 if (db
->db_blkptr
!= NULL
) {
1932 * We need to fill in zgd_bp with the current blkptr so that
1933 * the nopwrite code can check if we're writing the same
1934 * data that's already on disk. We can only nopwrite if we
1935 * are sure that after making the copy, db_blkptr will not
1936 * change until our i/o completes. We ensure this by
1937 * holding the db_mtx, and only allowing nopwrite if the
1938 * block is not already dirty (see below). This is verified
1939 * by dmu_sync_done(), which VERIFYs that the db_blkptr has
1942 *zgd
->zgd_bp
= *db
->db_blkptr
;
1946 * Assume the on-disk data is X, the current syncing data (in
1947 * txg - 1) is Y, and the current in-memory data is Z (currently
1950 * We usually want to perform a nopwrite if X and Z are the
1951 * same. However, if Y is different (i.e. the BP is going to
1952 * change before this write takes effect), then a nopwrite will
1953 * be incorrect - we would override with X, which could have
1954 * been freed when Y was written.
1956 * (Note that this is not a concern when we are nop-writing from
1957 * syncing context, because X and Y must be identical, because
1958 * all previous txgs have been synced.)
1960 * Therefore, we disable nopwrite if the current BP could change
1961 * before this TXG. There are two ways it could change: by
1962 * being dirty (dr_next is non-NULL), or by being freed
1963 * (dnode_block_freed()). This behavior is verified by
1964 * zio_done(), which VERIFYs that the override BP is identical
1965 * to the on-disk BP.
1969 if (dr
->dr_next
!= NULL
|| dnode_block_freed(dn
, db
->db_blkid
))
1970 zp
.zp_nopwrite
= B_FALSE
;
1973 ASSERT(dr
->dr_txg
== txg
);
1974 if (dr
->dt
.dl
.dr_override_state
== DR_IN_DMU_SYNC
||
1975 dr
->dt
.dl
.dr_override_state
== DR_OVERRIDDEN
) {
1977 * We have already issued a sync write for this buffer,
1978 * or this buffer has already been synced. It could not
1979 * have been dirtied since, or we would have cleared the state.
1981 mutex_exit(&db
->db_mtx
);
1982 return (SET_ERROR(EALREADY
));
1985 ASSERT(dr
->dt
.dl
.dr_override_state
== DR_NOT_OVERRIDDEN
);
1986 dr
->dt
.dl
.dr_override_state
= DR_IN_DMU_SYNC
;
1987 mutex_exit(&db
->db_mtx
);
1989 dsa
= kmem_alloc(sizeof (dmu_sync_arg_t
), KM_SLEEP
);
1991 dsa
->dsa_done
= done
;
1995 zio_nowait(arc_write(pio
, os
->os_spa
, txg
,
1996 zgd
->zgd_bp
, dr
->dt
.dl
.dr_data
, DBUF_IS_L2CACHEABLE(db
),
1997 &zp
, dmu_sync_ready
, NULL
, NULL
, dmu_sync_done
, dsa
,
1998 ZIO_PRIORITY_SYNC_WRITE
, ZIO_FLAG_CANFAIL
, &zb
));
2004 dmu_object_set_nlevels(objset_t
*os
, uint64_t object
, int nlevels
, dmu_tx_t
*tx
)
2009 err
= dnode_hold(os
, object
, FTAG
, &dn
);
2012 err
= dnode_set_nlevels(dn
, nlevels
, tx
);
2013 dnode_rele(dn
, FTAG
);
2018 dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
, int ibs
,
2024 err
= dnode_hold(os
, object
, FTAG
, &dn
);
2027 err
= dnode_set_blksz(dn
, size
, ibs
, tx
);
2028 dnode_rele(dn
, FTAG
);
2033 dmu_object_set_maxblkid(objset_t
*os
, uint64_t object
, uint64_t maxblkid
,
2039 err
= dnode_hold(os
, object
, FTAG
, &dn
);
2042 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
2043 dnode_new_blkid(dn
, maxblkid
, tx
, B_FALSE
);
2044 rw_exit(&dn
->dn_struct_rwlock
);
2045 dnode_rele(dn
, FTAG
);
2050 dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
2056 * Send streams include each object's checksum function. This
2057 * check ensures that the receiving system can understand the
2058 * checksum function transmitted.
2060 ASSERT3U(checksum
, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
2062 VERIFY0(dnode_hold(os
, object
, FTAG
, &dn
));
2063 ASSERT3U(checksum
, <, ZIO_CHECKSUM_FUNCTIONS
);
2064 dn
->dn_checksum
= checksum
;
2065 dnode_setdirty(dn
, tx
);
2066 dnode_rele(dn
, FTAG
);
2070 dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
2076 * Send streams include each object's compression function. This
2077 * check ensures that the receiving system can understand the
2078 * compression function transmitted.
2080 ASSERT3U(compress
, <, ZIO_COMPRESS_LEGACY_FUNCTIONS
);
2082 VERIFY0(dnode_hold(os
, object
, FTAG
, &dn
));
2083 dn
->dn_compress
= compress
;
2084 dnode_setdirty(dn
, tx
);
2085 dnode_rele(dn
, FTAG
);
2089 * Dirty an object and set the dirty record's raw flag. This is used
2090 * when writing raw data to an object that will not effect the
2091 * encryption parameters, specifically during raw receives.
2094 dmu_object_dirty_raw(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
)
2099 err
= dnode_hold(os
, object
, FTAG
, &dn
);
2102 dmu_buf_will_change_crypt_params((dmu_buf_t
*)dn
->dn_dbuf
, tx
);
2103 dnode_rele(dn
, FTAG
);
2107 int zfs_mdcomp_disable
= 0;
2110 * When the "redundant_metadata" property is set to "most", only indirect
2111 * blocks of this level and higher will have an additional ditto block.
2113 int zfs_redundant_metadata_most_ditto_level
= 2;
2116 dmu_write_policy(objset_t
*os
, dnode_t
*dn
, int level
, int wp
, zio_prop_t
*zp
)
2118 dmu_object_type_t type
= dn
? dn
->dn_type
: DMU_OT_OBJSET
;
2119 boolean_t ismd
= (level
> 0 || DMU_OT_IS_METADATA(type
) ||
2121 enum zio_checksum checksum
= os
->os_checksum
;
2122 enum zio_compress compress
= os
->os_compress
;
2123 enum zio_checksum dedup_checksum
= os
->os_dedup_checksum
;
2124 boolean_t dedup
= B_FALSE
;
2125 boolean_t nopwrite
= B_FALSE
;
2126 boolean_t dedup_verify
= os
->os_dedup_verify
;
2127 boolean_t encrypt
= B_FALSE
;
2128 int copies
= os
->os_copies
;
2131 * We maintain different write policies for each of the following
2134 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
2135 * 3. all other level 0 blocks
2138 if (zfs_mdcomp_disable
) {
2139 compress
= ZIO_COMPRESS_EMPTY
;
2142 * XXX -- we should design a compression algorithm
2143 * that specializes in arrays of bps.
2145 compress
= zio_compress_select(os
->os_spa
,
2146 ZIO_COMPRESS_ON
, ZIO_COMPRESS_ON
);
2150 * Metadata always gets checksummed. If the data
2151 * checksum is multi-bit correctable, and it's not a
2152 * ZBT-style checksum, then it's suitable for metadata
2153 * as well. Otherwise, the metadata checksum defaults
2156 if (!(zio_checksum_table
[checksum
].ci_flags
&
2157 ZCHECKSUM_FLAG_METADATA
) ||
2158 (zio_checksum_table
[checksum
].ci_flags
&
2159 ZCHECKSUM_FLAG_EMBEDDED
))
2160 checksum
= ZIO_CHECKSUM_FLETCHER_4
;
2162 if (os
->os_redundant_metadata
== ZFS_REDUNDANT_METADATA_ALL
||
2163 (os
->os_redundant_metadata
==
2164 ZFS_REDUNDANT_METADATA_MOST
&&
2165 (level
>= zfs_redundant_metadata_most_ditto_level
||
2166 DMU_OT_IS_METADATA(type
) || (wp
& WP_SPILL
))))
2168 } else if (wp
& WP_NOFILL
) {
2172 * If we're writing preallocated blocks, we aren't actually
2173 * writing them so don't set any policy properties. These
2174 * blocks are currently only used by an external subsystem
2175 * outside of zfs (i.e. dump) and not written by the zio
2178 compress
= ZIO_COMPRESS_OFF
;
2179 checksum
= ZIO_CHECKSUM_OFF
;
2181 compress
= zio_compress_select(os
->os_spa
, dn
->dn_compress
,
2184 checksum
= (dedup_checksum
== ZIO_CHECKSUM_OFF
) ?
2185 zio_checksum_select(dn
->dn_checksum
, checksum
) :
2189 * Determine dedup setting. If we are in dmu_sync(),
2190 * we won't actually dedup now because that's all
2191 * done in syncing context; but we do want to use the
2192 * dedup checkum. If the checksum is not strong
2193 * enough to ensure unique signatures, force
2196 if (dedup_checksum
!= ZIO_CHECKSUM_OFF
) {
2197 dedup
= (wp
& WP_DMU_SYNC
) ? B_FALSE
: B_TRUE
;
2198 if (!(zio_checksum_table
[checksum
].ci_flags
&
2199 ZCHECKSUM_FLAG_DEDUP
))
2200 dedup_verify
= B_TRUE
;
2204 * Enable nopwrite if we have secure enough checksum
2205 * algorithm (see comment in zio_nop_write) and
2206 * compression is enabled. We don't enable nopwrite if
2207 * dedup is enabled as the two features are mutually
2210 nopwrite
= (!dedup
&& (zio_checksum_table
[checksum
].ci_flags
&
2211 ZCHECKSUM_FLAG_NOPWRITE
) &&
2212 compress
!= ZIO_COMPRESS_OFF
&& zfs_nopwrite_enabled
);
2216 * All objects in an encrypted objset are protected from modification
2217 * via a MAC. Encrypted objects store their IV and salt in the last DVA
2218 * in the bp, so we cannot use all copies. Encrypted objects are also
2219 * not subject to nopwrite since writing the same data will still
2220 * result in a new ciphertext. Only encrypted blocks can be dedup'd
2221 * to avoid ambiguity in the dedup code since the DDT does not store
2224 if (os
->os_encrypted
&& (wp
& WP_NOFILL
) == 0) {
2227 if (DMU_OT_IS_ENCRYPTED(type
)) {
2228 copies
= MIN(copies
, SPA_DVAS_PER_BP
- 1);
2235 (type
== DMU_OT_DNODE
|| type
== DMU_OT_OBJSET
)) {
2236 compress
= ZIO_COMPRESS_EMPTY
;
2240 zp
->zp_compress
= compress
;
2241 zp
->zp_checksum
= checksum
;
2242 zp
->zp_type
= (wp
& WP_SPILL
) ? dn
->dn_bonustype
: type
;
2243 zp
->zp_level
= level
;
2244 zp
->zp_copies
= MIN(copies
, spa_max_replication(os
->os_spa
));
2245 zp
->zp_dedup
= dedup
;
2246 zp
->zp_dedup_verify
= dedup
&& dedup_verify
;
2247 zp
->zp_nopwrite
= nopwrite
;
2248 zp
->zp_encrypt
= encrypt
;
2249 zp
->zp_byteorder
= ZFS_HOST_BYTEORDER
;
2250 bzero(zp
->zp_salt
, ZIO_DATA_SALT_LEN
);
2251 bzero(zp
->zp_iv
, ZIO_DATA_IV_LEN
);
2252 bzero(zp
->zp_mac
, ZIO_DATA_MAC_LEN
);
2254 ASSERT3U(zp
->zp_compress
, !=, ZIO_COMPRESS_INHERIT
);
2258 * This function is only called from zfs_holey_common() for zpl_llseek()
2259 * in order to determine the location of holes. In order to accurately
2260 * report holes all dirty data must be synced to disk. This causes extremely
2261 * poor performance when seeking for holes in a dirty file. As a compromise,
2262 * only provide hole data when the dnode is clean. When a dnode is dirty
2263 * report the dnode as having no holes which is always a safe thing to do.
2266 dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
, uint64_t *off
)
2270 boolean_t clean
= B_TRUE
;
2272 err
= dnode_hold(os
, object
, FTAG
, &dn
);
2277 * Check if dnode is dirty
2279 for (i
= 0; i
< TXG_SIZE
; i
++) {
2280 if (list_link_active(&dn
->dn_dirty_link
[i
])) {
2287 * If compatibility option is on, sync any current changes before
2288 * we go trundling through the block pointers.
2290 if (!clean
&& zfs_dmu_offset_next_sync
) {
2292 dnode_rele(dn
, FTAG
);
2293 txg_wait_synced(dmu_objset_pool(os
), 0);
2294 err
= dnode_hold(os
, object
, FTAG
, &dn
);
2300 err
= dnode_next_offset(dn
,
2301 (hole
? DNODE_FIND_HOLE
: 0), off
, 1, 1, 0);
2303 err
= SET_ERROR(EBUSY
);
2305 dnode_rele(dn
, FTAG
);
2311 __dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
)
2313 dnode_phys_t
*dnp
= dn
->dn_phys
;
2315 doi
->doi_data_block_size
= dn
->dn_datablksz
;
2316 doi
->doi_metadata_block_size
= dn
->dn_indblkshift
?
2317 1ULL << dn
->dn_indblkshift
: 0;
2318 doi
->doi_type
= dn
->dn_type
;
2319 doi
->doi_bonus_type
= dn
->dn_bonustype
;
2320 doi
->doi_bonus_size
= dn
->dn_bonuslen
;
2321 doi
->doi_dnodesize
= dn
->dn_num_slots
<< DNODE_SHIFT
;
2322 doi
->doi_indirection
= dn
->dn_nlevels
;
2323 doi
->doi_checksum
= dn
->dn_checksum
;
2324 doi
->doi_compress
= dn
->dn_compress
;
2325 doi
->doi_nblkptr
= dn
->dn_nblkptr
;
2326 doi
->doi_physical_blocks_512
= (DN_USED_BYTES(dnp
) + 256) >> 9;
2327 doi
->doi_max_offset
= (dn
->dn_maxblkid
+ 1) * dn
->dn_datablksz
;
2328 doi
->doi_fill_count
= 0;
2329 for (int i
= 0; i
< dnp
->dn_nblkptr
; i
++)
2330 doi
->doi_fill_count
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
2334 dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
)
2336 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
2337 mutex_enter(&dn
->dn_mtx
);
2339 __dmu_object_info_from_dnode(dn
, doi
);
2341 mutex_exit(&dn
->dn_mtx
);
2342 rw_exit(&dn
->dn_struct_rwlock
);
2346 * Get information on a DMU object.
2347 * If doi is NULL, just indicates whether the object exists.
2350 dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
)
2353 int err
= dnode_hold(os
, object
, FTAG
, &dn
);
2359 dmu_object_info_from_dnode(dn
, doi
);
2361 dnode_rele(dn
, FTAG
);
2366 * As above, but faster; can be used when you have a held dbuf in hand.
2369 dmu_object_info_from_db(dmu_buf_t
*db_fake
, dmu_object_info_t
*doi
)
2371 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2374 dmu_object_info_from_dnode(DB_DNODE(db
), doi
);
2379 * Faster still when you only care about the size.
2380 * This is specifically optimized for zfs_getattr().
2383 dmu_object_size_from_db(dmu_buf_t
*db_fake
, uint32_t *blksize
,
2384 u_longlong_t
*nblk512
)
2386 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2392 *blksize
= dn
->dn_datablksz
;
2393 /* add in number of slots used for the dnode itself */
2394 *nblk512
= ((DN_USED_BYTES(dn
->dn_phys
) + SPA_MINBLOCKSIZE
/2) >>
2395 SPA_MINBLOCKSHIFT
) + dn
->dn_num_slots
;
2400 dmu_object_dnsize_from_db(dmu_buf_t
*db_fake
, int *dnsize
)
2402 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2407 *dnsize
= dn
->dn_num_slots
<< DNODE_SHIFT
;
2412 byteswap_uint64_array(void *vbuf
, size_t size
)
2414 uint64_t *buf
= vbuf
;
2415 size_t count
= size
>> 3;
2418 ASSERT((size
& 7) == 0);
2420 for (i
= 0; i
< count
; i
++)
2421 buf
[i
] = BSWAP_64(buf
[i
]);
2425 byteswap_uint32_array(void *vbuf
, size_t size
)
2427 uint32_t *buf
= vbuf
;
2428 size_t count
= size
>> 2;
2431 ASSERT((size
& 3) == 0);
2433 for (i
= 0; i
< count
; i
++)
2434 buf
[i
] = BSWAP_32(buf
[i
]);
2438 byteswap_uint16_array(void *vbuf
, size_t size
)
2440 uint16_t *buf
= vbuf
;
2441 size_t count
= size
>> 1;
2444 ASSERT((size
& 1) == 0);
2446 for (i
= 0; i
< count
; i
++)
2447 buf
[i
] = BSWAP_16(buf
[i
]);
2452 byteswap_uint8_array(void *vbuf
, size_t size
)
2475 arc_fini(); /* arc depends on l2arc, so arc must go first */
2488 #if defined(_KERNEL) && defined(HAVE_SPL)
2489 EXPORT_SYMBOL(dmu_bonus_hold
);
2490 EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus
);
2491 EXPORT_SYMBOL(dmu_buf_rele_array
);
2492 EXPORT_SYMBOL(dmu_prefetch
);
2493 EXPORT_SYMBOL(dmu_free_range
);
2494 EXPORT_SYMBOL(dmu_free_long_range
);
2495 EXPORT_SYMBOL(dmu_free_long_range_raw
);
2496 EXPORT_SYMBOL(dmu_free_long_object
);
2497 EXPORT_SYMBOL(dmu_free_long_object_raw
);
2498 EXPORT_SYMBOL(dmu_read
);
2499 EXPORT_SYMBOL(dmu_read_by_dnode
);
2500 EXPORT_SYMBOL(dmu_write
);
2501 EXPORT_SYMBOL(dmu_write_by_dnode
);
2502 EXPORT_SYMBOL(dmu_prealloc
);
2503 EXPORT_SYMBOL(dmu_object_info
);
2504 EXPORT_SYMBOL(dmu_object_info_from_dnode
);
2505 EXPORT_SYMBOL(dmu_object_info_from_db
);
2506 EXPORT_SYMBOL(dmu_object_size_from_db
);
2507 EXPORT_SYMBOL(dmu_object_dnsize_from_db
);
2508 EXPORT_SYMBOL(dmu_object_set_nlevels
);
2509 EXPORT_SYMBOL(dmu_object_set_blocksize
);
2510 EXPORT_SYMBOL(dmu_object_set_maxblkid
);
2511 EXPORT_SYMBOL(dmu_object_set_checksum
);
2512 EXPORT_SYMBOL(dmu_object_set_compress
);
2513 EXPORT_SYMBOL(dmu_write_policy
);
2514 EXPORT_SYMBOL(dmu_sync
);
2515 EXPORT_SYMBOL(dmu_request_arcbuf
);
2516 EXPORT_SYMBOL(dmu_return_arcbuf
);
2517 EXPORT_SYMBOL(dmu_assign_arcbuf_by_dnode
);
2518 EXPORT_SYMBOL(dmu_assign_arcbuf_by_dbuf
);
2519 EXPORT_SYMBOL(dmu_buf_hold
);
2520 EXPORT_SYMBOL(dmu_ot
);
2523 module_param(zfs_mdcomp_disable
, int, 0644);
2524 MODULE_PARM_DESC(zfs_mdcomp_disable
, "Disable meta data compression");
2526 module_param(zfs_nopwrite_enabled
, int, 0644);
2527 MODULE_PARM_DESC(zfs_nopwrite_enabled
, "Enable NOP writes");
2529 module_param(zfs_per_txg_dirty_frees_percent
, ulong
, 0644);
2530 MODULE_PARM_DESC(zfs_per_txg_dirty_frees_percent
,
2531 "percentage of dirtied blocks from frees in one TXG");
2533 module_param(zfs_dmu_offset_next_sync
, int, 0644);
2534 MODULE_PARM_DESC(zfs_dmu_offset_next_sync
,
2535 "Enable forcing txg sync to find holes");