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, 2016 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) 2014, 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>
52 #include <sys/vmsystm.h>
53 #include <sys/zfs_znode.h>
57 * Enable/disable nopwrite feature.
59 int zfs_nopwrite_enabled
= 1;
61 const dmu_object_type_info_t dmu_ot
[DMU_OT_NUMTYPES
] = {
62 { DMU_BSWAP_UINT8
, TRUE
, "unallocated" },
63 { DMU_BSWAP_ZAP
, TRUE
, "object directory" },
64 { DMU_BSWAP_UINT64
, TRUE
, "object array" },
65 { DMU_BSWAP_UINT8
, TRUE
, "packed nvlist" },
66 { DMU_BSWAP_UINT64
, TRUE
, "packed nvlist size" },
67 { DMU_BSWAP_UINT64
, TRUE
, "bpobj" },
68 { DMU_BSWAP_UINT64
, TRUE
, "bpobj header" },
69 { DMU_BSWAP_UINT64
, TRUE
, "SPA space map header" },
70 { DMU_BSWAP_UINT64
, TRUE
, "SPA space map" },
71 { DMU_BSWAP_UINT64
, TRUE
, "ZIL intent log" },
72 { DMU_BSWAP_DNODE
, TRUE
, "DMU dnode" },
73 { DMU_BSWAP_OBJSET
, TRUE
, "DMU objset" },
74 { DMU_BSWAP_UINT64
, TRUE
, "DSL directory" },
75 { DMU_BSWAP_ZAP
, TRUE
, "DSL directory child map"},
76 { DMU_BSWAP_ZAP
, TRUE
, "DSL dataset snap map" },
77 { DMU_BSWAP_ZAP
, TRUE
, "DSL props" },
78 { DMU_BSWAP_UINT64
, TRUE
, "DSL dataset" },
79 { DMU_BSWAP_ZNODE
, TRUE
, "ZFS znode" },
80 { DMU_BSWAP_OLDACL
, TRUE
, "ZFS V0 ACL" },
81 { DMU_BSWAP_UINT8
, FALSE
, "ZFS plain file" },
82 { DMU_BSWAP_ZAP
, TRUE
, "ZFS directory" },
83 { DMU_BSWAP_ZAP
, TRUE
, "ZFS master node" },
84 { DMU_BSWAP_ZAP
, TRUE
, "ZFS delete queue" },
85 { DMU_BSWAP_UINT8
, FALSE
, "zvol object" },
86 { DMU_BSWAP_ZAP
, TRUE
, "zvol prop" },
87 { DMU_BSWAP_UINT8
, FALSE
, "other uint8[]" },
88 { DMU_BSWAP_UINT64
, FALSE
, "other uint64[]" },
89 { DMU_BSWAP_ZAP
, TRUE
, "other ZAP" },
90 { DMU_BSWAP_ZAP
, TRUE
, "persistent error log" },
91 { DMU_BSWAP_UINT8
, TRUE
, "SPA history" },
92 { DMU_BSWAP_UINT64
, TRUE
, "SPA history offsets" },
93 { DMU_BSWAP_ZAP
, TRUE
, "Pool properties" },
94 { DMU_BSWAP_ZAP
, TRUE
, "DSL permissions" },
95 { DMU_BSWAP_ACL
, TRUE
, "ZFS ACL" },
96 { DMU_BSWAP_UINT8
, TRUE
, "ZFS SYSACL" },
97 { DMU_BSWAP_UINT8
, TRUE
, "FUID table" },
98 { DMU_BSWAP_UINT64
, TRUE
, "FUID table size" },
99 { DMU_BSWAP_ZAP
, TRUE
, "DSL dataset next clones"},
100 { DMU_BSWAP_ZAP
, TRUE
, "scan work queue" },
101 { DMU_BSWAP_ZAP
, TRUE
, "ZFS user/group used" },
102 { DMU_BSWAP_ZAP
, TRUE
, "ZFS user/group quota" },
103 { DMU_BSWAP_ZAP
, TRUE
, "snapshot refcount tags"},
104 { DMU_BSWAP_ZAP
, TRUE
, "DDT ZAP algorithm" },
105 { DMU_BSWAP_ZAP
, TRUE
, "DDT statistics" },
106 { DMU_BSWAP_UINT8
, TRUE
, "System attributes" },
107 { DMU_BSWAP_ZAP
, TRUE
, "SA master node" },
108 { DMU_BSWAP_ZAP
, TRUE
, "SA attr registration" },
109 { DMU_BSWAP_ZAP
, TRUE
, "SA attr layouts" },
110 { DMU_BSWAP_ZAP
, TRUE
, "scan translations" },
111 { DMU_BSWAP_UINT8
, FALSE
, "deduplicated block" },
112 { DMU_BSWAP_ZAP
, TRUE
, "DSL deadlist map" },
113 { DMU_BSWAP_UINT64
, TRUE
, "DSL deadlist map hdr" },
114 { DMU_BSWAP_ZAP
, TRUE
, "DSL dir clones" },
115 { DMU_BSWAP_UINT64
, TRUE
, "bpobj subobj" }
118 const dmu_object_byteswap_info_t dmu_ot_byteswap
[DMU_BSWAP_NUMFUNCS
] = {
119 { byteswap_uint8_array
, "uint8" },
120 { byteswap_uint16_array
, "uint16" },
121 { byteswap_uint32_array
, "uint32" },
122 { byteswap_uint64_array
, "uint64" },
123 { zap_byteswap
, "zap" },
124 { dnode_buf_byteswap
, "dnode" },
125 { dmu_objset_byteswap
, "objset" },
126 { zfs_znode_byteswap
, "znode" },
127 { zfs_oldacl_byteswap
, "oldacl" },
128 { zfs_acl_byteswap
, "acl" }
132 dmu_buf_hold_noread_by_dnode(dnode_t
*dn
, uint64_t offset
,
133 void *tag
, dmu_buf_t
**dbp
)
138 blkid
= dbuf_whichblock(dn
, 0, offset
);
139 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
140 db
= dbuf_hold(dn
, blkid
, tag
);
141 rw_exit(&dn
->dn_struct_rwlock
);
145 return (SET_ERROR(EIO
));
152 dmu_buf_hold_noread(objset_t
*os
, uint64_t object
, uint64_t offset
,
153 void *tag
, dmu_buf_t
**dbp
)
160 err
= dnode_hold(os
, object
, FTAG
, &dn
);
163 blkid
= dbuf_whichblock(dn
, 0, offset
);
164 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
165 db
= dbuf_hold(dn
, blkid
, tag
);
166 rw_exit(&dn
->dn_struct_rwlock
);
167 dnode_rele(dn
, FTAG
);
171 return (SET_ERROR(EIO
));
179 dmu_buf_hold_by_dnode(dnode_t
*dn
, uint64_t offset
,
180 void *tag
, dmu_buf_t
**dbp
, int flags
)
183 int db_flags
= DB_RF_CANFAIL
;
185 if (flags
& DMU_READ_NO_PREFETCH
)
186 db_flags
|= DB_RF_NOPREFETCH
;
188 err
= dmu_buf_hold_noread_by_dnode(dn
, offset
, tag
, dbp
);
190 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)(*dbp
);
191 err
= dbuf_read(db
, NULL
, db_flags
);
202 dmu_buf_hold(objset_t
*os
, uint64_t object
, uint64_t offset
,
203 void *tag
, dmu_buf_t
**dbp
, int flags
)
206 int db_flags
= DB_RF_CANFAIL
;
208 if (flags
& DMU_READ_NO_PREFETCH
)
209 db_flags
|= DB_RF_NOPREFETCH
;
211 err
= dmu_buf_hold_noread(os
, object
, offset
, tag
, dbp
);
213 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)(*dbp
);
214 err
= dbuf_read(db
, NULL
, db_flags
);
227 return (DN_OLD_MAX_BONUSLEN
);
231 dmu_set_bonus(dmu_buf_t
*db_fake
, int newsize
, dmu_tx_t
*tx
)
233 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
240 if (dn
->dn_bonus
!= db
) {
241 error
= SET_ERROR(EINVAL
);
242 } else if (newsize
< 0 || newsize
> db_fake
->db_size
) {
243 error
= SET_ERROR(EINVAL
);
245 dnode_setbonuslen(dn
, newsize
, tx
);
254 dmu_set_bonustype(dmu_buf_t
*db_fake
, dmu_object_type_t type
, dmu_tx_t
*tx
)
256 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
263 if (!DMU_OT_IS_VALID(type
)) {
264 error
= SET_ERROR(EINVAL
);
265 } else if (dn
->dn_bonus
!= db
) {
266 error
= SET_ERROR(EINVAL
);
268 dnode_setbonus_type(dn
, type
, tx
);
277 dmu_get_bonustype(dmu_buf_t
*db_fake
)
279 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
281 dmu_object_type_t type
;
285 type
= dn
->dn_bonustype
;
292 dmu_rm_spill(objset_t
*os
, uint64_t object
, dmu_tx_t
*tx
)
297 error
= dnode_hold(os
, object
, FTAG
, &dn
);
298 dbuf_rm_spill(dn
, tx
);
299 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
300 dnode_rm_spill(dn
, tx
);
301 rw_exit(&dn
->dn_struct_rwlock
);
302 dnode_rele(dn
, FTAG
);
307 * returns ENOENT, EIO, or 0.
310 dmu_bonus_hold(objset_t
*os
, uint64_t object
, void *tag
, dmu_buf_t
**dbp
)
316 error
= dnode_hold(os
, object
, FTAG
, &dn
);
320 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
321 if (dn
->dn_bonus
== NULL
) {
322 rw_exit(&dn
->dn_struct_rwlock
);
323 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
324 if (dn
->dn_bonus
== NULL
)
325 dbuf_create_bonus(dn
);
329 /* as long as the bonus buf is held, the dnode will be held */
330 if (refcount_add(&db
->db_holds
, tag
) == 1) {
331 VERIFY(dnode_add_ref(dn
, db
));
332 atomic_inc_32(&dn
->dn_dbufs_count
);
336 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's
337 * hold and incrementing the dbuf count to ensure that dnode_move() sees
338 * a dnode hold for every dbuf.
340 rw_exit(&dn
->dn_struct_rwlock
);
342 dnode_rele(dn
, FTAG
);
344 VERIFY(0 == dbuf_read(db
, NULL
, DB_RF_MUST_SUCCEED
| DB_RF_NOPREFETCH
));
351 * returns ENOENT, EIO, or 0.
353 * This interface will allocate a blank spill dbuf when a spill blk
354 * doesn't already exist on the dnode.
356 * if you only want to find an already existing spill db, then
357 * dmu_spill_hold_existing() should be used.
360 dmu_spill_hold_by_dnode(dnode_t
*dn
, uint32_t flags
, void *tag
, dmu_buf_t
**dbp
)
362 dmu_buf_impl_t
*db
= NULL
;
365 if ((flags
& DB_RF_HAVESTRUCT
) == 0)
366 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
368 db
= dbuf_hold(dn
, DMU_SPILL_BLKID
, tag
);
370 if ((flags
& DB_RF_HAVESTRUCT
) == 0)
371 rw_exit(&dn
->dn_struct_rwlock
);
375 return (SET_ERROR(EIO
));
377 err
= dbuf_read(db
, NULL
, flags
);
388 dmu_spill_hold_existing(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
)
390 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)bonus
;
397 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_SA
) {
398 err
= SET_ERROR(EINVAL
);
400 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
402 if (!dn
->dn_have_spill
) {
403 err
= SET_ERROR(ENOENT
);
405 err
= dmu_spill_hold_by_dnode(dn
,
406 DB_RF_HAVESTRUCT
| DB_RF_CANFAIL
, tag
, dbp
);
409 rw_exit(&dn
->dn_struct_rwlock
);
417 dmu_spill_hold_by_bonus(dmu_buf_t
*bonus
, void *tag
, dmu_buf_t
**dbp
)
419 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)bonus
;
425 err
= dmu_spill_hold_by_dnode(dn
, DB_RF_CANFAIL
, tag
, dbp
);
432 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
433 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
434 * and can induce severe lock contention when writing to several files
435 * whose dnodes are in the same block.
438 dmu_buf_hold_array_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t length
,
439 boolean_t read
, void *tag
, int *numbufsp
, dmu_buf_t
***dbpp
, uint32_t flags
)
442 uint64_t blkid
, nblks
, i
;
447 ASSERT(length
<= DMU_MAX_ACCESS
);
450 * Note: We directly notify the prefetch code of this read, so that
451 * we can tell it about the multi-block read. dbuf_read() only knows
452 * about the one block it is accessing.
454 dbuf_flags
= DB_RF_CANFAIL
| DB_RF_NEVERWAIT
| DB_RF_HAVESTRUCT
|
457 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
458 if (dn
->dn_datablkshift
) {
459 int blkshift
= dn
->dn_datablkshift
;
460 nblks
= (P2ROUNDUP(offset
+ length
, 1ULL << blkshift
) -
461 P2ALIGN(offset
, 1ULL << blkshift
)) >> blkshift
;
463 if (offset
+ length
> dn
->dn_datablksz
) {
464 zfs_panic_recover("zfs: accessing past end of object "
465 "%llx/%llx (size=%u access=%llu+%llu)",
466 (longlong_t
)dn
->dn_objset
->
467 os_dsl_dataset
->ds_object
,
468 (longlong_t
)dn
->dn_object
, dn
->dn_datablksz
,
469 (longlong_t
)offset
, (longlong_t
)length
);
470 rw_exit(&dn
->dn_struct_rwlock
);
471 return (SET_ERROR(EIO
));
475 dbp
= kmem_zalloc(sizeof (dmu_buf_t
*) * nblks
, KM_SLEEP
);
477 zio
= zio_root(dn
->dn_objset
->os_spa
, NULL
, NULL
, ZIO_FLAG_CANFAIL
);
478 blkid
= dbuf_whichblock(dn
, 0, offset
);
479 for (i
= 0; i
< nblks
; i
++) {
480 dmu_buf_impl_t
*db
= dbuf_hold(dn
, blkid
+ i
, tag
);
482 rw_exit(&dn
->dn_struct_rwlock
);
483 dmu_buf_rele_array(dbp
, nblks
, tag
);
485 return (SET_ERROR(EIO
));
488 /* initiate async i/o */
490 (void) dbuf_read(db
, zio
, dbuf_flags
);
494 if ((flags
& DMU_READ_NO_PREFETCH
) == 0 &&
495 DNODE_META_IS_CACHEABLE(dn
) && length
<= zfetch_array_rd_sz
) {
496 dmu_zfetch(&dn
->dn_zfetch
, blkid
, nblks
,
497 read
&& DNODE_IS_CACHEABLE(dn
));
499 rw_exit(&dn
->dn_struct_rwlock
);
501 /* wait for async i/o */
504 dmu_buf_rele_array(dbp
, nblks
, tag
);
508 /* wait for other io to complete */
510 for (i
= 0; i
< nblks
; i
++) {
511 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)dbp
[i
];
512 mutex_enter(&db
->db_mtx
);
513 while (db
->db_state
== DB_READ
||
514 db
->db_state
== DB_FILL
)
515 cv_wait(&db
->db_changed
, &db
->db_mtx
);
516 if (db
->db_state
== DB_UNCACHED
)
517 err
= SET_ERROR(EIO
);
518 mutex_exit(&db
->db_mtx
);
520 dmu_buf_rele_array(dbp
, nblks
, tag
);
532 dmu_buf_hold_array(objset_t
*os
, uint64_t object
, uint64_t offset
,
533 uint64_t length
, int read
, void *tag
, int *numbufsp
, dmu_buf_t
***dbpp
)
538 err
= dnode_hold(os
, object
, FTAG
, &dn
);
542 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, length
, read
, tag
,
543 numbufsp
, dbpp
, DMU_READ_PREFETCH
);
545 dnode_rele(dn
, FTAG
);
551 dmu_buf_hold_array_by_bonus(dmu_buf_t
*db_fake
, uint64_t offset
,
552 uint64_t length
, boolean_t read
, void *tag
, int *numbufsp
,
555 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
561 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, length
, read
, tag
,
562 numbufsp
, dbpp
, DMU_READ_PREFETCH
);
569 dmu_buf_rele_array(dmu_buf_t
**dbp_fake
, int numbufs
, void *tag
)
572 dmu_buf_impl_t
**dbp
= (dmu_buf_impl_t
**)dbp_fake
;
577 for (i
= 0; i
< numbufs
; i
++) {
579 dbuf_rele(dbp
[i
], tag
);
582 kmem_free(dbp
, sizeof (dmu_buf_t
*) * numbufs
);
586 * Issue prefetch i/os for the given blocks. If level is greater than 0, the
587 * indirect blocks prefeteched will be those that point to the blocks containing
588 * the data starting at offset, and continuing to offset + len.
590 * Note that if the indirect blocks above the blocks being prefetched are not in
591 * cache, they will be asychronously read in.
594 dmu_prefetch(objset_t
*os
, uint64_t object
, int64_t level
, uint64_t offset
,
595 uint64_t len
, zio_priority_t pri
)
601 if (len
== 0) { /* they're interested in the bonus buffer */
602 dn
= DMU_META_DNODE(os
);
604 if (object
== 0 || object
>= DN_MAX_OBJECT
)
607 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
608 blkid
= dbuf_whichblock(dn
, level
,
609 object
* sizeof (dnode_phys_t
));
610 dbuf_prefetch(dn
, level
, blkid
, pri
, 0);
611 rw_exit(&dn
->dn_struct_rwlock
);
616 * XXX - Note, if the dnode for the requested object is not
617 * already cached, we will do a *synchronous* read in the
618 * dnode_hold() call. The same is true for any indirects.
620 err
= dnode_hold(os
, object
, FTAG
, &dn
);
624 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
626 * offset + len - 1 is the last byte we want to prefetch for, and offset
627 * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the
628 * last block we want to prefetch, and dbuf_whichblock(dn, level,
629 * offset) is the first. Then the number we need to prefetch is the
632 if (level
> 0 || dn
->dn_datablkshift
!= 0) {
633 nblks
= dbuf_whichblock(dn
, level
, offset
+ len
- 1) -
634 dbuf_whichblock(dn
, level
, offset
) + 1;
636 nblks
= (offset
< dn
->dn_datablksz
);
642 blkid
= dbuf_whichblock(dn
, level
, offset
);
643 for (i
= 0; i
< nblks
; i
++)
644 dbuf_prefetch(dn
, level
, blkid
+ i
, pri
, 0);
647 rw_exit(&dn
->dn_struct_rwlock
);
649 dnode_rele(dn
, FTAG
);
653 * Get the next "chunk" of file data to free. We traverse the file from
654 * the end so that the file gets shorter over time (if we crashes in the
655 * middle, this will leave us in a better state). We find allocated file
656 * data by simply searching the allocated level 1 indirects.
658 * On input, *start should be the first offset that does not need to be
659 * freed (e.g. "offset + length"). On return, *start will be the first
660 * offset that should be freed.
663 get_next_chunk(dnode_t
*dn
, uint64_t *start
, uint64_t minimum
)
665 uint64_t maxblks
= DMU_MAX_ACCESS
>> (dn
->dn_indblkshift
+ 1);
666 /* bytes of data covered by a level-1 indirect block */
668 dn
->dn_datablksz
* EPB(dn
->dn_indblkshift
, SPA_BLKPTRSHIFT
);
671 ASSERT3U(minimum
, <=, *start
);
673 if (*start
- minimum
<= iblkrange
* maxblks
) {
677 ASSERT(ISP2(iblkrange
));
679 for (blks
= 0; *start
> minimum
&& blks
< maxblks
; blks
++) {
683 * dnode_next_offset(BACKWARDS) will find an allocated L1
684 * indirect block at or before the input offset. We must
685 * decrement *start so that it is at the end of the region
689 err
= dnode_next_offset(dn
,
690 DNODE_FIND_BACKWARDS
, start
, 2, 1, 0);
692 /* if there are no indirect blocks before start, we are done */
696 } else if (err
!= 0) {
700 /* set start to the beginning of this L1 indirect */
701 *start
= P2ALIGN(*start
, iblkrange
);
703 if (*start
< minimum
)
709 dmu_free_long_range_impl(objset_t
*os
, dnode_t
*dn
, uint64_t offset
,
712 uint64_t object_size
;
716 return (SET_ERROR(EINVAL
));
718 object_size
= (dn
->dn_maxblkid
+ 1) * dn
->dn_datablksz
;
719 if (offset
>= object_size
)
722 if (length
== DMU_OBJECT_END
|| offset
+ length
> object_size
)
723 length
= object_size
- offset
;
725 while (length
!= 0) {
726 uint64_t chunk_end
, chunk_begin
;
729 chunk_end
= chunk_begin
= offset
+ length
;
731 /* move chunk_begin backwards to the beginning of this chunk */
732 err
= get_next_chunk(dn
, &chunk_begin
, offset
);
735 ASSERT3U(chunk_begin
, >=, offset
);
736 ASSERT3U(chunk_begin
, <=, chunk_end
);
738 tx
= dmu_tx_create(os
);
739 dmu_tx_hold_free(tx
, dn
->dn_object
,
740 chunk_begin
, chunk_end
- chunk_begin
);
743 * Mark this transaction as typically resulting in a net
744 * reduction in space used.
746 dmu_tx_mark_netfree(tx
);
747 err
= dmu_tx_assign(tx
, TXG_WAIT
);
752 dnode_free_range(dn
, chunk_begin
, chunk_end
- chunk_begin
, tx
);
755 length
-= chunk_end
- chunk_begin
;
761 dmu_free_long_range(objset_t
*os
, uint64_t object
,
762 uint64_t offset
, uint64_t length
)
767 err
= dnode_hold(os
, object
, FTAG
, &dn
);
770 err
= dmu_free_long_range_impl(os
, dn
, offset
, length
);
773 * It is important to zero out the maxblkid when freeing the entire
774 * file, so that (a) subsequent calls to dmu_free_long_range_impl()
775 * will take the fast path, and (b) dnode_reallocate() can verify
776 * that the entire file has been freed.
778 if (err
== 0 && offset
== 0 && length
== DMU_OBJECT_END
)
781 dnode_rele(dn
, FTAG
);
786 dmu_free_long_object(objset_t
*os
, uint64_t object
)
791 err
= dmu_free_long_range(os
, object
, 0, DMU_OBJECT_END
);
795 tx
= dmu_tx_create(os
);
796 dmu_tx_hold_bonus(tx
, object
);
797 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
798 dmu_tx_mark_netfree(tx
);
799 err
= dmu_tx_assign(tx
, TXG_WAIT
);
801 err
= dmu_object_free(os
, object
, tx
);
811 dmu_free_range(objset_t
*os
, uint64_t object
, uint64_t offset
,
812 uint64_t size
, dmu_tx_t
*tx
)
815 int err
= dnode_hold(os
, object
, FTAG
, &dn
);
818 ASSERT(offset
< UINT64_MAX
);
819 ASSERT(size
== -1ULL || size
<= UINT64_MAX
- offset
);
820 dnode_free_range(dn
, offset
, size
, tx
);
821 dnode_rele(dn
, FTAG
);
826 dmu_read_impl(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
827 void *buf
, uint32_t flags
)
830 int numbufs
, err
= 0;
833 * Deal with odd block sizes, where there can't be data past the first
834 * block. If we ever do the tail block optimization, we will need to
835 * handle that here as well.
837 if (dn
->dn_maxblkid
== 0) {
838 uint64_t newsz
= offset
> dn
->dn_datablksz
? 0 :
839 MIN(size
, dn
->dn_datablksz
- offset
);
840 bzero((char *)buf
+ newsz
, size
- newsz
);
845 uint64_t mylen
= MIN(size
, DMU_MAX_ACCESS
/ 2);
849 * NB: we could do this block-at-a-time, but it's nice
850 * to be reading in parallel.
852 err
= dmu_buf_hold_array_by_dnode(dn
, offset
, mylen
,
853 TRUE
, FTAG
, &numbufs
, &dbp
, flags
);
857 for (i
= 0; i
< numbufs
; i
++) {
860 dmu_buf_t
*db
= dbp
[i
];
864 bufoff
= offset
- db
->db_offset
;
865 tocpy
= MIN(db
->db_size
- bufoff
, size
);
867 (void) memcpy(buf
, (char *)db
->db_data
+ bufoff
, tocpy
);
871 buf
= (char *)buf
+ tocpy
;
873 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
879 dmu_read(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
880 void *buf
, uint32_t flags
)
885 err
= dnode_hold(os
, object
, FTAG
, &dn
);
889 err
= dmu_read_impl(dn
, offset
, size
, buf
, flags
);
890 dnode_rele(dn
, FTAG
);
895 dmu_read_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
, void *buf
,
898 return (dmu_read_impl(dn
, offset
, size
, buf
, flags
));
902 dmu_write_impl(dmu_buf_t
**dbp
, int numbufs
, uint64_t offset
, uint64_t size
,
903 const void *buf
, dmu_tx_t
*tx
)
907 for (i
= 0; i
< numbufs
; i
++) {
910 dmu_buf_t
*db
= dbp
[i
];
914 bufoff
= offset
- db
->db_offset
;
915 tocpy
= MIN(db
->db_size
- bufoff
, size
);
917 ASSERT(i
== 0 || i
== numbufs
-1 || tocpy
== db
->db_size
);
919 if (tocpy
== db
->db_size
)
920 dmu_buf_will_fill(db
, tx
);
922 dmu_buf_will_dirty(db
, tx
);
924 (void) memcpy((char *)db
->db_data
+ bufoff
, buf
, tocpy
);
926 if (tocpy
== db
->db_size
)
927 dmu_buf_fill_done(db
, tx
);
931 buf
= (char *)buf
+ tocpy
;
936 dmu_write(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
937 const void *buf
, dmu_tx_t
*tx
)
945 VERIFY0(dmu_buf_hold_array(os
, object
, offset
, size
,
946 FALSE
, FTAG
, &numbufs
, &dbp
));
947 dmu_write_impl(dbp
, numbufs
, offset
, size
, buf
, tx
);
948 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
952 dmu_write_by_dnode(dnode_t
*dn
, uint64_t offset
, uint64_t size
,
953 const void *buf
, dmu_tx_t
*tx
)
961 VERIFY0(dmu_buf_hold_array_by_dnode(dn
, offset
, size
,
962 FALSE
, FTAG
, &numbufs
, &dbp
, DMU_READ_PREFETCH
));
963 dmu_write_impl(dbp
, numbufs
, offset
, size
, buf
, tx
);
964 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
968 dmu_prealloc(objset_t
*os
, uint64_t object
, uint64_t offset
, uint64_t size
,
977 VERIFY(0 == dmu_buf_hold_array(os
, object
, offset
, size
,
978 FALSE
, FTAG
, &numbufs
, &dbp
));
980 for (i
= 0; i
< numbufs
; i
++) {
981 dmu_buf_t
*db
= dbp
[i
];
983 dmu_buf_will_not_fill(db
, tx
);
985 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
989 dmu_write_embedded(objset_t
*os
, uint64_t object
, uint64_t offset
,
990 void *data
, uint8_t etype
, uint8_t comp
, int uncompressed_size
,
991 int compressed_size
, int byteorder
, dmu_tx_t
*tx
)
995 ASSERT3U(etype
, <, NUM_BP_EMBEDDED_TYPES
);
996 ASSERT3U(comp
, <, ZIO_COMPRESS_FUNCTIONS
);
997 VERIFY0(dmu_buf_hold_noread(os
, object
, offset
,
1000 dmu_buf_write_embedded(db
,
1001 data
, (bp_embedded_type_t
)etype
, (enum zio_compress
)comp
,
1002 uncompressed_size
, compressed_size
, byteorder
, tx
);
1004 dmu_buf_rele(db
, FTAG
);
1008 * DMU support for xuio
1010 kstat_t
*xuio_ksp
= NULL
;
1012 typedef struct xuio_stats
{
1013 /* loaned yet not returned arc_buf */
1014 kstat_named_t xuiostat_onloan_rbuf
;
1015 kstat_named_t xuiostat_onloan_wbuf
;
1016 /* whether a copy is made when loaning out a read buffer */
1017 kstat_named_t xuiostat_rbuf_copied
;
1018 kstat_named_t xuiostat_rbuf_nocopy
;
1019 /* whether a copy is made when assigning a write buffer */
1020 kstat_named_t xuiostat_wbuf_copied
;
1021 kstat_named_t xuiostat_wbuf_nocopy
;
1024 static xuio_stats_t xuio_stats
= {
1025 { "onloan_read_buf", KSTAT_DATA_UINT64
},
1026 { "onloan_write_buf", KSTAT_DATA_UINT64
},
1027 { "read_buf_copied", KSTAT_DATA_UINT64
},
1028 { "read_buf_nocopy", KSTAT_DATA_UINT64
},
1029 { "write_buf_copied", KSTAT_DATA_UINT64
},
1030 { "write_buf_nocopy", KSTAT_DATA_UINT64
}
1033 #define XUIOSTAT_INCR(stat, val) \
1034 atomic_add_64(&xuio_stats.stat.value.ui64, (val))
1035 #define XUIOSTAT_BUMP(stat) XUIOSTAT_INCR(stat, 1)
1037 #ifdef HAVE_UIO_ZEROCOPY
1039 dmu_xuio_init(xuio_t
*xuio
, int nblk
)
1042 uio_t
*uio
= &xuio
->xu_uio
;
1044 uio
->uio_iovcnt
= nblk
;
1045 uio
->uio_iov
= kmem_zalloc(nblk
* sizeof (iovec_t
), KM_SLEEP
);
1047 priv
= kmem_zalloc(sizeof (dmu_xuio_t
), KM_SLEEP
);
1049 priv
->bufs
= kmem_zalloc(nblk
* sizeof (arc_buf_t
*), KM_SLEEP
);
1050 priv
->iovp
= (iovec_t
*)uio
->uio_iov
;
1051 XUIO_XUZC_PRIV(xuio
) = priv
;
1053 if (XUIO_XUZC_RW(xuio
) == UIO_READ
)
1054 XUIOSTAT_INCR(xuiostat_onloan_rbuf
, nblk
);
1056 XUIOSTAT_INCR(xuiostat_onloan_wbuf
, nblk
);
1062 dmu_xuio_fini(xuio_t
*xuio
)
1064 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1065 int nblk
= priv
->cnt
;
1067 kmem_free(priv
->iovp
, nblk
* sizeof (iovec_t
));
1068 kmem_free(priv
->bufs
, nblk
* sizeof (arc_buf_t
*));
1069 kmem_free(priv
, sizeof (dmu_xuio_t
));
1071 if (XUIO_XUZC_RW(xuio
) == UIO_READ
)
1072 XUIOSTAT_INCR(xuiostat_onloan_rbuf
, -nblk
);
1074 XUIOSTAT_INCR(xuiostat_onloan_wbuf
, -nblk
);
1078 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
1079 * and increase priv->next by 1.
1082 dmu_xuio_add(xuio_t
*xuio
, arc_buf_t
*abuf
, offset_t off
, size_t n
)
1085 uio_t
*uio
= &xuio
->xu_uio
;
1086 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1087 int i
= priv
->next
++;
1089 ASSERT(i
< priv
->cnt
);
1090 ASSERT(off
+ n
<= arc_buf_lsize(abuf
));
1091 iov
= (iovec_t
*)uio
->uio_iov
+ i
;
1092 iov
->iov_base
= (char *)abuf
->b_data
+ off
;
1094 priv
->bufs
[i
] = abuf
;
1099 dmu_xuio_cnt(xuio_t
*xuio
)
1101 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1106 dmu_xuio_arcbuf(xuio_t
*xuio
, int i
)
1108 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1110 ASSERT(i
< priv
->cnt
);
1111 return (priv
->bufs
[i
]);
1115 dmu_xuio_clear(xuio_t
*xuio
, int i
)
1117 dmu_xuio_t
*priv
= XUIO_XUZC_PRIV(xuio
);
1119 ASSERT(i
< priv
->cnt
);
1120 priv
->bufs
[i
] = NULL
;
1122 #endif /* HAVE_UIO_ZEROCOPY */
1125 xuio_stat_init(void)
1127 xuio_ksp
= kstat_create("zfs", 0, "xuio_stats", "misc",
1128 KSTAT_TYPE_NAMED
, sizeof (xuio_stats
) / sizeof (kstat_named_t
),
1129 KSTAT_FLAG_VIRTUAL
);
1130 if (xuio_ksp
!= NULL
) {
1131 xuio_ksp
->ks_data
= &xuio_stats
;
1132 kstat_install(xuio_ksp
);
1137 xuio_stat_fini(void)
1139 if (xuio_ksp
!= NULL
) {
1140 kstat_delete(xuio_ksp
);
1146 xuio_stat_wbuf_copied(void)
1148 XUIOSTAT_BUMP(xuiostat_wbuf_copied
);
1152 xuio_stat_wbuf_nocopy(void)
1154 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy
);
1159 dmu_read_uio_dnode(dnode_t
*dn
, uio_t
*uio
, uint64_t size
)
1162 int numbufs
, i
, err
;
1163 #ifdef HAVE_UIO_ZEROCOPY
1164 xuio_t
*xuio
= NULL
;
1168 * NB: we could do this block-at-a-time, but it's nice
1169 * to be reading in parallel.
1171 err
= dmu_buf_hold_array_by_dnode(dn
, uio
->uio_loffset
, size
,
1172 TRUE
, FTAG
, &numbufs
, &dbp
, 0);
1176 for (i
= 0; i
< numbufs
; i
++) {
1179 dmu_buf_t
*db
= dbp
[i
];
1183 bufoff
= uio
->uio_loffset
- db
->db_offset
;
1184 tocpy
= MIN(db
->db_size
- bufoff
, size
);
1186 #ifdef HAVE_UIO_ZEROCOPY
1188 dmu_buf_impl_t
*dbi
= (dmu_buf_impl_t
*)db
;
1189 arc_buf_t
*dbuf_abuf
= dbi
->db_buf
;
1190 arc_buf_t
*abuf
= dbuf_loan_arcbuf(dbi
);
1191 err
= dmu_xuio_add(xuio
, abuf
, bufoff
, tocpy
);
1193 uio
->uio_resid
-= tocpy
;
1194 uio
->uio_loffset
+= tocpy
;
1197 if (abuf
== dbuf_abuf
)
1198 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy
);
1200 XUIOSTAT_BUMP(xuiostat_rbuf_copied
);
1203 err
= uiomove((char *)db
->db_data
+ bufoff
, tocpy
,
1210 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1216 * Read 'size' bytes into the uio buffer.
1217 * From object zdb->db_object.
1218 * Starting at offset uio->uio_loffset.
1220 * If the caller already has a dbuf in the target object
1221 * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(),
1222 * because we don't have to find the dnode_t for the object.
1225 dmu_read_uio_dbuf(dmu_buf_t
*zdb
, uio_t
*uio
, uint64_t size
)
1227 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)zdb
;
1236 err
= dmu_read_uio_dnode(dn
, uio
, size
);
1243 * Read 'size' bytes into the uio buffer.
1244 * From the specified object
1245 * Starting at offset uio->uio_loffset.
1248 dmu_read_uio(objset_t
*os
, uint64_t object
, uio_t
*uio
, uint64_t size
)
1256 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1260 err
= dmu_read_uio_dnode(dn
, uio
, size
);
1262 dnode_rele(dn
, FTAG
);
1268 dmu_write_uio_dnode(dnode_t
*dn
, uio_t
*uio
, uint64_t size
, dmu_tx_t
*tx
)
1275 err
= dmu_buf_hold_array_by_dnode(dn
, uio
->uio_loffset
, size
,
1276 FALSE
, FTAG
, &numbufs
, &dbp
, DMU_READ_PREFETCH
);
1280 for (i
= 0; i
< numbufs
; i
++) {
1283 dmu_buf_t
*db
= dbp
[i
];
1287 bufoff
= uio
->uio_loffset
- db
->db_offset
;
1288 tocpy
= MIN(db
->db_size
- bufoff
, size
);
1290 ASSERT(i
== 0 || i
== numbufs
-1 || tocpy
== db
->db_size
);
1292 if (tocpy
== db
->db_size
)
1293 dmu_buf_will_fill(db
, tx
);
1295 dmu_buf_will_dirty(db
, tx
);
1298 * XXX uiomove could block forever (eg.nfs-backed
1299 * pages). There needs to be a uiolockdown() function
1300 * to lock the pages in memory, so that uiomove won't
1303 err
= uiomove((char *)db
->db_data
+ bufoff
, tocpy
,
1306 if (tocpy
== db
->db_size
)
1307 dmu_buf_fill_done(db
, tx
);
1315 dmu_buf_rele_array(dbp
, numbufs
, FTAG
);
1320 * Write 'size' bytes from the uio buffer.
1321 * To object zdb->db_object.
1322 * Starting at offset uio->uio_loffset.
1324 * If the caller already has a dbuf in the target object
1325 * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(),
1326 * because we don't have to find the dnode_t for the object.
1329 dmu_write_uio_dbuf(dmu_buf_t
*zdb
, uio_t
*uio
, uint64_t size
,
1332 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)zdb
;
1341 err
= dmu_write_uio_dnode(dn
, uio
, size
, tx
);
1348 * Write 'size' bytes from the uio buffer.
1349 * To the specified object.
1350 * Starting at offset uio->uio_loffset.
1353 dmu_write_uio(objset_t
*os
, uint64_t object
, uio_t
*uio
, uint64_t size
,
1362 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1366 err
= dmu_write_uio_dnode(dn
, uio
, size
, tx
);
1368 dnode_rele(dn
, FTAG
);
1372 #endif /* _KERNEL */
1375 * Allocate a loaned anonymous arc buffer.
1378 dmu_request_arcbuf(dmu_buf_t
*handle
, int size
)
1380 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)handle
;
1382 return (arc_loan_buf(db
->db_objset
->os_spa
, B_FALSE
, size
));
1386 * Free a loaned arc buffer.
1389 dmu_return_arcbuf(arc_buf_t
*buf
)
1391 arc_return_buf(buf
, FTAG
);
1392 arc_buf_destroy(buf
, FTAG
);
1396 * When possible directly assign passed loaned arc buffer to a dbuf.
1397 * If this is not possible copy the contents of passed arc buf via
1401 dmu_assign_arcbuf(dmu_buf_t
*handle
, uint64_t offset
, arc_buf_t
*buf
,
1404 dmu_buf_impl_t
*dbuf
= (dmu_buf_impl_t
*)handle
;
1407 uint32_t blksz
= (uint32_t)arc_buf_lsize(buf
);
1410 DB_DNODE_ENTER(dbuf
);
1411 dn
= DB_DNODE(dbuf
);
1412 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1413 blkid
= dbuf_whichblock(dn
, 0, offset
);
1414 VERIFY((db
= dbuf_hold(dn
, blkid
, FTAG
)) != NULL
);
1415 rw_exit(&dn
->dn_struct_rwlock
);
1416 DB_DNODE_EXIT(dbuf
);
1419 * We can only assign if the offset is aligned, the arc buf is the
1420 * same size as the dbuf, and the dbuf is not metadata.
1422 if (offset
== db
->db
.db_offset
&& blksz
== db
->db
.db_size
) {
1423 dbuf_assign_arcbuf(db
, buf
, tx
);
1424 dbuf_rele(db
, FTAG
);
1429 /* compressed bufs must always be assignable to their dbuf */
1430 ASSERT3U(arc_get_compression(buf
), ==, ZIO_COMPRESS_OFF
);
1431 ASSERT(!(buf
->b_flags
& ARC_BUF_FLAG_COMPRESSED
));
1433 DB_DNODE_ENTER(dbuf
);
1434 dn
= DB_DNODE(dbuf
);
1436 object
= dn
->dn_object
;
1437 DB_DNODE_EXIT(dbuf
);
1439 dbuf_rele(db
, FTAG
);
1440 dmu_write(os
, object
, offset
, blksz
, buf
->b_data
, tx
);
1441 dmu_return_arcbuf(buf
);
1442 XUIOSTAT_BUMP(xuiostat_wbuf_copied
);
1447 dbuf_dirty_record_t
*dsa_dr
;
1448 dmu_sync_cb_t
*dsa_done
;
1455 dmu_sync_ready(zio_t
*zio
, arc_buf_t
*buf
, void *varg
)
1457 dmu_sync_arg_t
*dsa
= varg
;
1458 dmu_buf_t
*db
= dsa
->dsa_zgd
->zgd_db
;
1459 blkptr_t
*bp
= zio
->io_bp
;
1461 if (zio
->io_error
== 0) {
1462 if (BP_IS_HOLE(bp
)) {
1464 * A block of zeros may compress to a hole, but the
1465 * block size still needs to be known for replay.
1467 BP_SET_LSIZE(bp
, db
->db_size
);
1468 } else if (!BP_IS_EMBEDDED(bp
)) {
1469 ASSERT(BP_GET_LEVEL(bp
) == 0);
1476 dmu_sync_late_arrival_ready(zio_t
*zio
)
1478 dmu_sync_ready(zio
, NULL
, zio
->io_private
);
1483 dmu_sync_done(zio_t
*zio
, arc_buf_t
*buf
, void *varg
)
1485 dmu_sync_arg_t
*dsa
= varg
;
1486 dbuf_dirty_record_t
*dr
= dsa
->dsa_dr
;
1487 dmu_buf_impl_t
*db
= dr
->dr_dbuf
;
1489 mutex_enter(&db
->db_mtx
);
1490 ASSERT(dr
->dt
.dl
.dr_override_state
== DR_IN_DMU_SYNC
);
1491 if (zio
->io_error
== 0) {
1492 dr
->dt
.dl
.dr_nopwrite
= !!(zio
->io_flags
& ZIO_FLAG_NOPWRITE
);
1493 if (dr
->dt
.dl
.dr_nopwrite
) {
1494 ASSERTV(blkptr_t
*bp
= zio
->io_bp
);
1495 ASSERTV(blkptr_t
*bp_orig
= &zio
->io_bp_orig
);
1496 ASSERTV(uint8_t chksum
= BP_GET_CHECKSUM(bp_orig
));
1498 ASSERT(BP_EQUAL(bp
, bp_orig
));
1499 ASSERT(zio
->io_prop
.zp_compress
!= ZIO_COMPRESS_OFF
);
1500 ASSERT(zio_checksum_table
[chksum
].ci_flags
&
1501 ZCHECKSUM_FLAG_NOPWRITE
);
1503 dr
->dt
.dl
.dr_overridden_by
= *zio
->io_bp
;
1504 dr
->dt
.dl
.dr_override_state
= DR_OVERRIDDEN
;
1505 dr
->dt
.dl
.dr_copies
= zio
->io_prop
.zp_copies
;
1508 * Old style holes are filled with all zeros, whereas
1509 * new-style holes maintain their lsize, type, level,
1510 * and birth time (see zio_write_compress). While we
1511 * need to reset the BP_SET_LSIZE() call that happened
1512 * in dmu_sync_ready for old style holes, we do *not*
1513 * want to wipe out the information contained in new
1514 * style holes. Thus, only zero out the block pointer if
1515 * it's an old style hole.
1517 if (BP_IS_HOLE(&dr
->dt
.dl
.dr_overridden_by
) &&
1518 dr
->dt
.dl
.dr_overridden_by
.blk_birth
== 0)
1519 BP_ZERO(&dr
->dt
.dl
.dr_overridden_by
);
1521 dr
->dt
.dl
.dr_override_state
= DR_NOT_OVERRIDDEN
;
1523 cv_broadcast(&db
->db_changed
);
1524 mutex_exit(&db
->db_mtx
);
1526 dsa
->dsa_done(dsa
->dsa_zgd
, zio
->io_error
);
1528 kmem_free(dsa
, sizeof (*dsa
));
1532 dmu_sync_late_arrival_done(zio_t
*zio
)
1534 blkptr_t
*bp
= zio
->io_bp
;
1535 dmu_sync_arg_t
*dsa
= zio
->io_private
;
1536 ASSERTV(blkptr_t
*bp_orig
= &zio
->io_bp_orig
);
1538 if (zio
->io_error
== 0 && !BP_IS_HOLE(bp
)) {
1540 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE)
1541 * then there is nothing to do here. Otherwise, free the
1542 * newly allocated block in this txg.
1544 if (zio
->io_flags
& ZIO_FLAG_NOPWRITE
) {
1545 ASSERT(BP_EQUAL(bp
, bp_orig
));
1547 ASSERT(BP_IS_HOLE(bp_orig
) || !BP_EQUAL(bp
, bp_orig
));
1548 ASSERT(zio
->io_bp
->blk_birth
== zio
->io_txg
);
1549 ASSERT(zio
->io_txg
> spa_syncing_txg(zio
->io_spa
));
1550 zio_free(zio
->io_spa
, zio
->io_txg
, zio
->io_bp
);
1554 dmu_tx_commit(dsa
->dsa_tx
);
1556 dsa
->dsa_done(dsa
->dsa_zgd
, zio
->io_error
);
1558 abd_put(zio
->io_abd
);
1559 kmem_free(dsa
, sizeof (*dsa
));
1563 dmu_sync_late_arrival(zio_t
*pio
, objset_t
*os
, dmu_sync_cb_t
*done
, zgd_t
*zgd
,
1564 zio_prop_t
*zp
, zbookmark_phys_t
*zb
)
1566 dmu_sync_arg_t
*dsa
;
1569 tx
= dmu_tx_create(os
);
1570 dmu_tx_hold_space(tx
, zgd
->zgd_db
->db_size
);
1571 if (dmu_tx_assign(tx
, TXG_WAIT
) != 0) {
1573 /* Make zl_get_data do txg_waited_synced() */
1574 return (SET_ERROR(EIO
));
1577 dsa
= kmem_alloc(sizeof (dmu_sync_arg_t
), KM_SLEEP
);
1579 dsa
->dsa_done
= done
;
1583 zio_nowait(zio_write(pio
, os
->os_spa
, dmu_tx_get_txg(tx
), zgd
->zgd_bp
,
1584 abd_get_from_buf(zgd
->zgd_db
->db_data
, zgd
->zgd_db
->db_size
),
1585 zgd
->zgd_db
->db_size
, zgd
->zgd_db
->db_size
, zp
,
1586 dmu_sync_late_arrival_ready
, NULL
, NULL
, dmu_sync_late_arrival_done
,
1587 dsa
, ZIO_PRIORITY_SYNC_WRITE
, ZIO_FLAG_CANFAIL
, zb
));
1593 * Intent log support: sync the block associated with db to disk.
1594 * N.B. and XXX: the caller is responsible for making sure that the
1595 * data isn't changing while dmu_sync() is writing it.
1599 * EEXIST: this txg has already been synced, so there's nothing to do.
1600 * The caller should not log the write.
1602 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1603 * The caller should not log the write.
1605 * EALREADY: this block is already in the process of being synced.
1606 * The caller should track its progress (somehow).
1608 * EIO: could not do the I/O.
1609 * The caller should do a txg_wait_synced().
1611 * 0: the I/O has been initiated.
1612 * The caller should log this blkptr in the done callback.
1613 * It is possible that the I/O will fail, in which case
1614 * the error will be reported to the done callback and
1615 * propagated to pio from zio_done().
1618 dmu_sync(zio_t
*pio
, uint64_t txg
, dmu_sync_cb_t
*done
, zgd_t
*zgd
)
1620 blkptr_t
*bp
= zgd
->zgd_bp
;
1621 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)zgd
->zgd_db
;
1622 objset_t
*os
= db
->db_objset
;
1623 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1624 dbuf_dirty_record_t
*dr
;
1625 dmu_sync_arg_t
*dsa
;
1626 zbookmark_phys_t zb
;
1630 ASSERT(pio
!= NULL
);
1633 SET_BOOKMARK(&zb
, ds
->ds_object
,
1634 db
->db
.db_object
, db
->db_level
, db
->db_blkid
);
1638 dmu_write_policy(os
, dn
, db
->db_level
, WP_DMU_SYNC
,
1639 ZIO_COMPRESS_INHERIT
, &zp
);
1643 * If we're frozen (running ziltest), we always need to generate a bp.
1645 if (txg
> spa_freeze_txg(os
->os_spa
))
1646 return (dmu_sync_late_arrival(pio
, os
, done
, zgd
, &zp
, &zb
));
1649 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1650 * and us. If we determine that this txg is not yet syncing,
1651 * but it begins to sync a moment later, that's OK because the
1652 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1654 mutex_enter(&db
->db_mtx
);
1656 if (txg
<= spa_last_synced_txg(os
->os_spa
)) {
1658 * This txg has already synced. There's nothing to do.
1660 mutex_exit(&db
->db_mtx
);
1661 return (SET_ERROR(EEXIST
));
1664 if (txg
<= spa_syncing_txg(os
->os_spa
)) {
1666 * This txg is currently syncing, so we can't mess with
1667 * the dirty record anymore; just write a new log block.
1669 mutex_exit(&db
->db_mtx
);
1670 return (dmu_sync_late_arrival(pio
, os
, done
, zgd
, &zp
, &zb
));
1673 dr
= db
->db_last_dirty
;
1674 while (dr
&& dr
->dr_txg
!= txg
)
1679 * There's no dr for this dbuf, so it must have been freed.
1680 * There's no need to log writes to freed blocks, so we're done.
1682 mutex_exit(&db
->db_mtx
);
1683 return (SET_ERROR(ENOENT
));
1686 ASSERT(dr
->dr_next
== NULL
|| dr
->dr_next
->dr_txg
< txg
);
1689 * Assume the on-disk data is X, the current syncing data (in
1690 * txg - 1) is Y, and the current in-memory data is Z (currently
1693 * We usually want to perform a nopwrite if X and Z are the
1694 * same. However, if Y is different (i.e. the BP is going to
1695 * change before this write takes effect), then a nopwrite will
1696 * be incorrect - we would override with X, which could have
1697 * been freed when Y was written.
1699 * (Note that this is not a concern when we are nop-writing from
1700 * syncing context, because X and Y must be identical, because
1701 * all previous txgs have been synced.)
1703 * Therefore, we disable nopwrite if the current BP could change
1704 * before this TXG. There are two ways it could change: by
1705 * being dirty (dr_next is non-NULL), or by being freed
1706 * (dnode_block_freed()). This behavior is verified by
1707 * zio_done(), which VERIFYs that the override BP is identical
1708 * to the on-disk BP.
1712 if (dr
->dr_next
!= NULL
|| dnode_block_freed(dn
, db
->db_blkid
))
1713 zp
.zp_nopwrite
= B_FALSE
;
1716 ASSERT(dr
->dr_txg
== txg
);
1717 if (dr
->dt
.dl
.dr_override_state
== DR_IN_DMU_SYNC
||
1718 dr
->dt
.dl
.dr_override_state
== DR_OVERRIDDEN
) {
1720 * We have already issued a sync write for this buffer,
1721 * or this buffer has already been synced. It could not
1722 * have been dirtied since, or we would have cleared the state.
1724 mutex_exit(&db
->db_mtx
);
1725 return (SET_ERROR(EALREADY
));
1728 ASSERT(dr
->dt
.dl
.dr_override_state
== DR_NOT_OVERRIDDEN
);
1729 dr
->dt
.dl
.dr_override_state
= DR_IN_DMU_SYNC
;
1730 mutex_exit(&db
->db_mtx
);
1732 dsa
= kmem_alloc(sizeof (dmu_sync_arg_t
), KM_SLEEP
);
1734 dsa
->dsa_done
= done
;
1738 zio_nowait(arc_write(pio
, os
->os_spa
, txg
,
1739 bp
, dr
->dt
.dl
.dr_data
, DBUF_IS_L2CACHEABLE(db
),
1740 &zp
, dmu_sync_ready
, NULL
, NULL
, dmu_sync_done
, dsa
,
1741 ZIO_PRIORITY_SYNC_WRITE
, ZIO_FLAG_CANFAIL
, &zb
));
1747 dmu_object_set_blocksize(objset_t
*os
, uint64_t object
, uint64_t size
, int ibs
,
1753 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1756 err
= dnode_set_blksz(dn
, size
, ibs
, tx
);
1757 dnode_rele(dn
, FTAG
);
1762 dmu_object_set_checksum(objset_t
*os
, uint64_t object
, uint8_t checksum
,
1768 * Send streams include each object's checksum function. This
1769 * check ensures that the receiving system can understand the
1770 * checksum function transmitted.
1772 ASSERT3U(checksum
, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
1774 VERIFY0(dnode_hold(os
, object
, FTAG
, &dn
));
1775 ASSERT3U(checksum
, <, ZIO_CHECKSUM_FUNCTIONS
);
1776 dn
->dn_checksum
= checksum
;
1777 dnode_setdirty(dn
, tx
);
1778 dnode_rele(dn
, FTAG
);
1782 dmu_object_set_compress(objset_t
*os
, uint64_t object
, uint8_t compress
,
1788 * Send streams include each object's compression function. This
1789 * check ensures that the receiving system can understand the
1790 * compression function transmitted.
1792 ASSERT3U(compress
, <, ZIO_COMPRESS_LEGACY_FUNCTIONS
);
1794 VERIFY0(dnode_hold(os
, object
, FTAG
, &dn
));
1795 dn
->dn_compress
= compress
;
1796 dnode_setdirty(dn
, tx
);
1797 dnode_rele(dn
, FTAG
);
1800 int zfs_mdcomp_disable
= 0;
1803 * When the "redundant_metadata" property is set to "most", only indirect
1804 * blocks of this level and higher will have an additional ditto block.
1806 int zfs_redundant_metadata_most_ditto_level
= 2;
1809 dmu_write_policy(objset_t
*os
, dnode_t
*dn
, int level
, int wp
,
1810 enum zio_compress override_compress
, zio_prop_t
*zp
)
1812 dmu_object_type_t type
= dn
? dn
->dn_type
: DMU_OT_OBJSET
;
1813 boolean_t ismd
= (level
> 0 || DMU_OT_IS_METADATA(type
) ||
1815 enum zio_checksum checksum
= os
->os_checksum
;
1816 enum zio_compress compress
= os
->os_compress
;
1817 enum zio_checksum dedup_checksum
= os
->os_dedup_checksum
;
1818 boolean_t dedup
= B_FALSE
;
1819 boolean_t nopwrite
= B_FALSE
;
1820 boolean_t dedup_verify
= os
->os_dedup_verify
;
1821 int copies
= os
->os_copies
;
1824 * We maintain different write policies for each of the following
1827 * 2. preallocated blocks (i.e. level-0 blocks of a dump device)
1828 * 3. all other level 0 blocks
1831 if (zfs_mdcomp_disable
) {
1832 compress
= ZIO_COMPRESS_EMPTY
;
1835 * XXX -- we should design a compression algorithm
1836 * that specializes in arrays of bps.
1838 compress
= zio_compress_select(os
->os_spa
,
1839 ZIO_COMPRESS_ON
, ZIO_COMPRESS_ON
);
1843 * Metadata always gets checksummed. If the data
1844 * checksum is multi-bit correctable, and it's not a
1845 * ZBT-style checksum, then it's suitable for metadata
1846 * as well. Otherwise, the metadata checksum defaults
1849 if (!(zio_checksum_table
[checksum
].ci_flags
&
1850 ZCHECKSUM_FLAG_METADATA
) ||
1851 (zio_checksum_table
[checksum
].ci_flags
&
1852 ZCHECKSUM_FLAG_EMBEDDED
))
1853 checksum
= ZIO_CHECKSUM_FLETCHER_4
;
1855 if (os
->os_redundant_metadata
== ZFS_REDUNDANT_METADATA_ALL
||
1856 (os
->os_redundant_metadata
==
1857 ZFS_REDUNDANT_METADATA_MOST
&&
1858 (level
>= zfs_redundant_metadata_most_ditto_level
||
1859 DMU_OT_IS_METADATA(type
) || (wp
& WP_SPILL
))))
1861 } else if (wp
& WP_NOFILL
) {
1865 * If we're writing preallocated blocks, we aren't actually
1866 * writing them so don't set any policy properties. These
1867 * blocks are currently only used by an external subsystem
1868 * outside of zfs (i.e. dump) and not written by the zio
1871 compress
= ZIO_COMPRESS_OFF
;
1872 checksum
= ZIO_CHECKSUM_OFF
;
1874 compress
= zio_compress_select(os
->os_spa
, dn
->dn_compress
,
1877 checksum
= (dedup_checksum
== ZIO_CHECKSUM_OFF
) ?
1878 zio_checksum_select(dn
->dn_checksum
, checksum
) :
1882 * Determine dedup setting. If we are in dmu_sync(),
1883 * we won't actually dedup now because that's all
1884 * done in syncing context; but we do want to use the
1885 * dedup checkum. If the checksum is not strong
1886 * enough to ensure unique signatures, force
1889 if (dedup_checksum
!= ZIO_CHECKSUM_OFF
) {
1890 dedup
= (wp
& WP_DMU_SYNC
) ? B_FALSE
: B_TRUE
;
1891 if (!(zio_checksum_table
[checksum
].ci_flags
&
1892 ZCHECKSUM_FLAG_DEDUP
))
1893 dedup_verify
= B_TRUE
;
1897 * Enable nopwrite if we have secure enough checksum
1898 * algorithm (see comment in zio_nop_write) and
1899 * compression is enabled. We don't enable nopwrite if
1900 * dedup is enabled as the two features are mutually
1903 nopwrite
= (!dedup
&& (zio_checksum_table
[checksum
].ci_flags
&
1904 ZCHECKSUM_FLAG_NOPWRITE
) &&
1905 compress
!= ZIO_COMPRESS_OFF
&& zfs_nopwrite_enabled
);
1908 zp
->zp_checksum
= checksum
;
1911 * If we're writing a pre-compressed buffer, the compression type we use
1912 * must match the data. If it hasn't been compressed yet, then we should
1913 * use the value dictated by the policies above.
1915 zp
->zp_compress
= override_compress
!= ZIO_COMPRESS_INHERIT
1916 ? override_compress
: compress
;
1917 ASSERT3U(zp
->zp_compress
, !=, ZIO_COMPRESS_INHERIT
);
1919 zp
->zp_type
= (wp
& WP_SPILL
) ? dn
->dn_bonustype
: type
;
1920 zp
->zp_level
= level
;
1921 zp
->zp_copies
= MIN(copies
, spa_max_replication(os
->os_spa
));
1922 zp
->zp_dedup
= dedup
;
1923 zp
->zp_dedup_verify
= dedup
&& dedup_verify
;
1924 zp
->zp_nopwrite
= nopwrite
;
1928 dmu_offset_next(objset_t
*os
, uint64_t object
, boolean_t hole
, uint64_t *off
)
1933 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1937 * Sync any current changes before
1938 * we go trundling through the block pointers.
1940 for (i
= 0; i
< TXG_SIZE
; i
++) {
1941 if (list_link_active(&dn
->dn_dirty_link
[i
]))
1944 if (i
!= TXG_SIZE
) {
1945 dnode_rele(dn
, FTAG
);
1946 txg_wait_synced(dmu_objset_pool(os
), 0);
1947 err
= dnode_hold(os
, object
, FTAG
, &dn
);
1952 err
= dnode_next_offset(dn
, (hole
? DNODE_FIND_HOLE
: 0), off
, 1, 1, 0);
1953 dnode_rele(dn
, FTAG
);
1959 __dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
)
1961 dnode_phys_t
*dnp
= dn
->dn_phys
;
1964 doi
->doi_data_block_size
= dn
->dn_datablksz
;
1965 doi
->doi_metadata_block_size
= dn
->dn_indblkshift
?
1966 1ULL << dn
->dn_indblkshift
: 0;
1967 doi
->doi_type
= dn
->dn_type
;
1968 doi
->doi_bonus_type
= dn
->dn_bonustype
;
1969 doi
->doi_bonus_size
= dn
->dn_bonuslen
;
1970 doi
->doi_dnodesize
= dn
->dn_num_slots
<< DNODE_SHIFT
;
1971 doi
->doi_indirection
= dn
->dn_nlevels
;
1972 doi
->doi_checksum
= dn
->dn_checksum
;
1973 doi
->doi_compress
= dn
->dn_compress
;
1974 doi
->doi_nblkptr
= dn
->dn_nblkptr
;
1975 doi
->doi_physical_blocks_512
= (DN_USED_BYTES(dnp
) + 256) >> 9;
1976 doi
->doi_max_offset
= (dn
->dn_maxblkid
+ 1) * dn
->dn_datablksz
;
1977 doi
->doi_fill_count
= 0;
1978 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1979 doi
->doi_fill_count
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1983 dmu_object_info_from_dnode(dnode_t
*dn
, dmu_object_info_t
*doi
)
1985 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1986 mutex_enter(&dn
->dn_mtx
);
1988 __dmu_object_info_from_dnode(dn
, doi
);
1990 mutex_exit(&dn
->dn_mtx
);
1991 rw_exit(&dn
->dn_struct_rwlock
);
1995 * Get information on a DMU object.
1996 * If doi is NULL, just indicates whether the object exists.
1999 dmu_object_info(objset_t
*os
, uint64_t object
, dmu_object_info_t
*doi
)
2002 int err
= dnode_hold(os
, object
, FTAG
, &dn
);
2008 dmu_object_info_from_dnode(dn
, doi
);
2010 dnode_rele(dn
, FTAG
);
2015 * As above, but faster; can be used when you have a held dbuf in hand.
2018 dmu_object_info_from_db(dmu_buf_t
*db_fake
, dmu_object_info_t
*doi
)
2020 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2023 dmu_object_info_from_dnode(DB_DNODE(db
), doi
);
2028 * Faster still when you only care about the size.
2029 * This is specifically optimized for zfs_getattr().
2032 dmu_object_size_from_db(dmu_buf_t
*db_fake
, uint32_t *blksize
,
2033 u_longlong_t
*nblk512
)
2035 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2041 *blksize
= dn
->dn_datablksz
;
2042 /* add in number of slots used for the dnode itself */
2043 *nblk512
= ((DN_USED_BYTES(dn
->dn_phys
) + SPA_MINBLOCKSIZE
/2) >>
2044 SPA_MINBLOCKSHIFT
) + dn
->dn_num_slots
;
2049 dmu_object_dnsize_from_db(dmu_buf_t
*db_fake
, int *dnsize
)
2051 dmu_buf_impl_t
*db
= (dmu_buf_impl_t
*)db_fake
;
2056 *dnsize
= dn
->dn_num_slots
<< DNODE_SHIFT
;
2061 byteswap_uint64_array(void *vbuf
, size_t size
)
2063 uint64_t *buf
= vbuf
;
2064 size_t count
= size
>> 3;
2067 ASSERT((size
& 7) == 0);
2069 for (i
= 0; i
< count
; i
++)
2070 buf
[i
] = BSWAP_64(buf
[i
]);
2074 byteswap_uint32_array(void *vbuf
, size_t size
)
2076 uint32_t *buf
= vbuf
;
2077 size_t count
= size
>> 2;
2080 ASSERT((size
& 3) == 0);
2082 for (i
= 0; i
< count
; i
++)
2083 buf
[i
] = BSWAP_32(buf
[i
]);
2087 byteswap_uint16_array(void *vbuf
, size_t size
)
2089 uint16_t *buf
= vbuf
;
2090 size_t count
= size
>> 1;
2093 ASSERT((size
& 1) == 0);
2095 for (i
= 0; i
< count
; i
++)
2096 buf
[i
] = BSWAP_16(buf
[i
]);
2101 byteswap_uint8_array(void *vbuf
, size_t size
)
2124 arc_fini(); /* arc depends on l2arc, so arc must go first */
2137 #if defined(_KERNEL) && defined(HAVE_SPL)
2138 EXPORT_SYMBOL(dmu_bonus_hold
);
2139 EXPORT_SYMBOL(dmu_buf_hold_array_by_bonus
);
2140 EXPORT_SYMBOL(dmu_buf_rele_array
);
2141 EXPORT_SYMBOL(dmu_prefetch
);
2142 EXPORT_SYMBOL(dmu_free_range
);
2143 EXPORT_SYMBOL(dmu_free_long_range
);
2144 EXPORT_SYMBOL(dmu_free_long_object
);
2145 EXPORT_SYMBOL(dmu_read
);
2146 EXPORT_SYMBOL(dmu_read_by_dnode
);
2147 EXPORT_SYMBOL(dmu_write
);
2148 EXPORT_SYMBOL(dmu_write_by_dnode
);
2149 EXPORT_SYMBOL(dmu_prealloc
);
2150 EXPORT_SYMBOL(dmu_object_info
);
2151 EXPORT_SYMBOL(dmu_object_info_from_dnode
);
2152 EXPORT_SYMBOL(dmu_object_info_from_db
);
2153 EXPORT_SYMBOL(dmu_object_size_from_db
);
2154 EXPORT_SYMBOL(dmu_object_dnsize_from_db
);
2155 EXPORT_SYMBOL(dmu_object_set_blocksize
);
2156 EXPORT_SYMBOL(dmu_object_set_checksum
);
2157 EXPORT_SYMBOL(dmu_object_set_compress
);
2158 EXPORT_SYMBOL(dmu_write_policy
);
2159 EXPORT_SYMBOL(dmu_sync
);
2160 EXPORT_SYMBOL(dmu_request_arcbuf
);
2161 EXPORT_SYMBOL(dmu_return_arcbuf
);
2162 EXPORT_SYMBOL(dmu_assign_arcbuf
);
2163 EXPORT_SYMBOL(dmu_buf_hold
);
2164 EXPORT_SYMBOL(dmu_ot
);
2166 module_param(zfs_mdcomp_disable
, int, 0644);
2167 MODULE_PARM_DESC(zfs_mdcomp_disable
, "Disable meta data compression");
2169 module_param(zfs_nopwrite_enabled
, int, 0644);
2170 MODULE_PARM_DESC(zfs_nopwrite_enabled
, "Enable NOP writes");