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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 #include <sys/zfs_context.h>
30 #include <sys/dnode.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dataset.h>
36 #include <sys/range_tree.h>
37 #include <sys/zfeature.h>
40 dnode_increase_indirection(dnode_t
*dn
, dmu_tx_t
*tx
)
43 int txgoff
= tx
->tx_txg
& TXG_MASK
;
44 int nblkptr
= dn
->dn_phys
->dn_nblkptr
;
45 int old_toplvl
= dn
->dn_phys
->dn_nlevels
- 1;
46 int new_level
= dn
->dn_next_nlevels
[txgoff
];
49 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
51 /* this dnode can't be paged out because it's dirty */
52 ASSERT(dn
->dn_phys
->dn_type
!= DMU_OT_NONE
);
53 ASSERT(RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
54 ASSERT(new_level
> 1 && dn
->dn_phys
->dn_nlevels
> 0);
56 db
= dbuf_hold_level(dn
, dn
->dn_phys
->dn_nlevels
, 0, FTAG
);
59 dn
->dn_phys
->dn_nlevels
= new_level
;
60 dprintf("os=%p obj=%llu, increase to %d\n", dn
->dn_objset
,
61 dn
->dn_object
, dn
->dn_phys
->dn_nlevels
);
63 /* check for existing blkptrs in the dnode */
64 for (i
= 0; i
< nblkptr
; i
++)
65 if (!BP_IS_HOLE(&dn
->dn_phys
->dn_blkptr
[i
]))
68 /* transfer dnode's block pointers to new indirect block */
69 (void) dbuf_read(db
, NULL
, DB_RF_MUST_SUCCEED
|DB_RF_HAVESTRUCT
);
70 ASSERT(db
->db
.db_data
);
71 ASSERT(arc_released(db
->db_buf
));
72 ASSERT3U(sizeof (blkptr_t
) * nblkptr
, <=, db
->db
.db_size
);
73 bcopy(dn
->dn_phys
->dn_blkptr
, db
->db
.db_data
,
74 sizeof (blkptr_t
) * nblkptr
);
75 arc_buf_freeze(db
->db_buf
);
78 /* set dbuf's parent pointers to new indirect buf */
79 for (i
= 0; i
< nblkptr
; i
++) {
80 dmu_buf_impl_t
*child
= dbuf_find(dn
, old_toplvl
, i
);
85 DB_DNODE_ENTER(child
);
86 ASSERT3P(DB_DNODE(child
), ==, dn
);
89 if (child
->db_parent
&& child
->db_parent
!= dn
->dn_dbuf
) {
90 ASSERT(child
->db_parent
->db_level
== db
->db_level
);
91 ASSERT(child
->db_blkptr
!=
92 &dn
->dn_phys
->dn_blkptr
[child
->db_blkid
]);
93 mutex_exit(&child
->db_mtx
);
96 ASSERT(child
->db_parent
== NULL
||
97 child
->db_parent
== dn
->dn_dbuf
);
99 child
->db_parent
= db
;
100 dbuf_add_ref(db
, child
);
102 child
->db_blkptr
= (blkptr_t
*)db
->db
.db_data
+ i
;
104 child
->db_blkptr
= NULL
;
105 dprintf_dbuf_bp(child
, child
->db_blkptr
,
106 "changed db_blkptr to new indirect %s", "");
108 mutex_exit(&child
->db_mtx
);
111 bzero(dn
->dn_phys
->dn_blkptr
, sizeof (blkptr_t
) * nblkptr
);
115 rw_exit(&dn
->dn_struct_rwlock
);
119 free_blocks(dnode_t
*dn
, blkptr_t
*bp
, int num
, dmu_tx_t
*tx
)
121 dsl_dataset_t
*ds
= dn
->dn_objset
->os_dsl_dataset
;
122 uint64_t bytesfreed
= 0;
124 dprintf("ds=%p obj=%llx num=%d\n", ds
, dn
->dn_object
, num
);
126 for (int i
= 0; i
< num
; i
++, bp
++) {
130 bytesfreed
+= dsl_dataset_block_kill(ds
, bp
, tx
, B_FALSE
);
131 ASSERT3U(bytesfreed
, <=, DN_USED_BYTES(dn
->dn_phys
));
134 * Save some useful information on the holes being
135 * punched, including logical size, type, and indirection
136 * level. Retaining birth time enables detection of when
137 * holes are punched for reducing the number of free
138 * records transmitted during a zfs send.
141 uint64_t lsize
= BP_GET_LSIZE(bp
);
142 dmu_object_type_t type
= BP_GET_TYPE(bp
);
143 uint64_t lvl
= BP_GET_LEVEL(bp
);
145 bzero(bp
, sizeof (blkptr_t
));
147 if (spa_feature_is_active(dn
->dn_objset
->os_spa
,
148 SPA_FEATURE_HOLE_BIRTH
)) {
149 BP_SET_LSIZE(bp
, lsize
);
150 BP_SET_TYPE(bp
, type
);
151 BP_SET_LEVEL(bp
, lvl
);
152 BP_SET_BIRTH(bp
, dmu_tx_get_txg(tx
), 0);
155 dnode_diduse_space(dn
, -bytesfreed
);
160 free_verify(dmu_buf_impl_t
*db
, uint64_t start
, uint64_t end
, dmu_tx_t
*tx
)
164 uint64_t txg
= tx
->tx_txg
;
169 epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
170 off
= start
- (db
->db_blkid
* 1<<epbs
);
171 num
= end
- start
+ 1;
173 ASSERT3U(off
, >=, 0);
174 ASSERT3U(num
, >=, 0);
175 ASSERT3U(db
->db_level
, >, 0);
176 ASSERT3U(db
->db
.db_size
, ==, 1 << dn
->dn_phys
->dn_indblkshift
);
177 ASSERT3U(off
+num
, <=, db
->db
.db_size
>> SPA_BLKPTRSHIFT
);
178 ASSERT(db
->db_blkptr
!= NULL
);
180 for (i
= off
; i
< off
+num
; i
++) {
182 dmu_buf_impl_t
*child
;
183 dbuf_dirty_record_t
*dr
;
186 ASSERT(db
->db_level
== 1);
188 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
189 err
= dbuf_hold_impl(dn
, db
->db_level
-1,
190 (db
->db_blkid
<< epbs
) + i
, TRUE
, FTAG
, &child
);
191 rw_exit(&dn
->dn_struct_rwlock
);
195 ASSERT(child
->db_level
== 0);
196 dr
= child
->db_last_dirty
;
197 while (dr
&& dr
->dr_txg
> txg
)
199 ASSERT(dr
== NULL
|| dr
->dr_txg
== txg
);
201 /* data_old better be zeroed */
203 buf
= dr
->dt
.dl
.dr_data
->b_data
;
204 for (j
= 0; j
< child
->db
.db_size
>> 3; j
++) {
206 panic("freed data not zero: "
207 "child=%p i=%d off=%d num=%d\n",
208 (void *)child
, i
, off
, num
);
214 * db_data better be zeroed unless it's dirty in a
217 mutex_enter(&child
->db_mtx
);
218 buf
= child
->db
.db_data
;
219 if (buf
!= NULL
&& child
->db_state
!= DB_FILL
&&
220 child
->db_last_dirty
== NULL
) {
221 for (j
= 0; j
< child
->db
.db_size
>> 3; j
++) {
223 panic("freed data not zero: "
224 "child=%p i=%d off=%d num=%d\n",
225 (void *)child
, i
, off
, num
);
229 mutex_exit(&child
->db_mtx
);
231 dbuf_rele(child
, FTAG
);
238 free_children(dmu_buf_impl_t
*db
, uint64_t blkid
, uint64_t nblks
,
243 dmu_buf_impl_t
*subdb
;
244 uint64_t start
, end
, dbstart
, dbend
, i
;
248 * There is a small possibility that this block will not be cached:
249 * 1 - if level > 1 and there are no children with level <= 1
250 * 2 - if this block was evicted since we read it from
251 * dmu_tx_hold_free().
253 if (db
->db_state
!= DB_CACHED
)
254 (void) dbuf_read(db
, NULL
, DB_RF_MUST_SUCCEED
);
261 epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
262 shift
= (db
->db_level
- 1) * epbs
;
263 dbstart
= db
->db_blkid
<< epbs
;
264 start
= blkid
>> shift
;
265 if (dbstart
< start
) {
266 bp
+= start
- dbstart
;
270 dbend
= ((db
->db_blkid
+ 1) << epbs
) - 1;
271 end
= (blkid
+ nblks
- 1) >> shift
;
275 ASSERT3U(start
, <=, end
);
277 if (db
->db_level
== 1) {
278 FREE_VERIFY(db
, start
, end
, tx
);
279 free_blocks(dn
, bp
, end
-start
+1, tx
);
281 for (i
= start
; i
<= end
; i
++, bp
++) {
284 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
285 VERIFY0(dbuf_hold_impl(dn
, db
->db_level
- 1,
286 i
, B_TRUE
, FTAG
, &subdb
));
287 rw_exit(&dn
->dn_struct_rwlock
);
288 ASSERT3P(bp
, ==, subdb
->db_blkptr
);
290 free_children(subdb
, blkid
, nblks
, tx
);
291 dbuf_rele(subdb
, FTAG
);
295 /* If this whole block is free, free ourself too. */
296 for (i
= 0, bp
= db
->db
.db_data
; i
< 1 << epbs
; i
++, bp
++) {
300 if (i
== 1 << epbs
) {
301 /* didn't find any non-holes */
302 bzero(db
->db
.db_data
, db
->db
.db_size
);
303 free_blocks(dn
, db
->db_blkptr
, 1, tx
);
306 * Partial block free; must be marked dirty so that it
307 * will be written out.
309 ASSERT(db
->db_dirtycnt
> 0);
313 arc_buf_freeze(db
->db_buf
);
317 * Traverse the indicated range of the provided file
318 * and "free" all the blocks contained there.
321 dnode_sync_free_range_impl(dnode_t
*dn
, uint64_t blkid
, uint64_t nblks
,
324 blkptr_t
*bp
= dn
->dn_phys
->dn_blkptr
;
325 int dnlevel
= dn
->dn_phys
->dn_nlevels
;
326 boolean_t trunc
= B_FALSE
;
328 if (blkid
> dn
->dn_phys
->dn_maxblkid
)
331 ASSERT(dn
->dn_phys
->dn_maxblkid
< UINT64_MAX
);
332 if (blkid
+ nblks
> dn
->dn_phys
->dn_maxblkid
) {
333 nblks
= dn
->dn_phys
->dn_maxblkid
- blkid
+ 1;
337 /* There are no indirect blocks in the object */
339 if (blkid
>= dn
->dn_phys
->dn_nblkptr
) {
340 /* this range was never made persistent */
343 ASSERT3U(blkid
+ nblks
, <=, dn
->dn_phys
->dn_nblkptr
);
344 free_blocks(dn
, bp
+ blkid
, nblks
, tx
);
346 int shift
= (dnlevel
- 1) *
347 (dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
);
348 int start
= blkid
>> shift
;
349 int end
= (blkid
+ nblks
- 1) >> shift
;
352 ASSERT(start
< dn
->dn_phys
->dn_nblkptr
);
354 for (int i
= start
; i
<= end
; i
++, bp
++) {
357 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
358 VERIFY0(dbuf_hold_impl(dn
, dnlevel
- 1, i
,
360 rw_exit(&dn
->dn_struct_rwlock
);
362 free_children(db
, blkid
, nblks
, tx
);
368 dn
->dn_phys
->dn_maxblkid
= blkid
== 0 ? 0 : blkid
- 1;
370 uint64_t off
= (dn
->dn_phys
->dn_maxblkid
+ 1) *
371 (dn
->dn_phys
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
372 ASSERT(off
< dn
->dn_phys
->dn_maxblkid
||
373 dn
->dn_phys
->dn_maxblkid
== 0 ||
374 dnode_next_offset(dn
, 0, &off
, 1, 1, 0) != 0);
378 typedef struct dnode_sync_free_range_arg
{
379 dnode_t
*dsfra_dnode
;
381 } dnode_sync_free_range_arg_t
;
384 dnode_sync_free_range(void *arg
, uint64_t blkid
, uint64_t nblks
)
386 dnode_sync_free_range_arg_t
*dsfra
= arg
;
387 dnode_t
*dn
= dsfra
->dsfra_dnode
;
389 mutex_exit(&dn
->dn_mtx
);
390 dnode_sync_free_range_impl(dn
, blkid
, nblks
, dsfra
->dsfra_tx
);
391 mutex_enter(&dn
->dn_mtx
);
395 * Try to kick all the dnode's dbufs out of the cache...
398 dnode_evict_dbufs(dnode_t
*dn
)
400 dmu_buf_impl_t db_marker
;
401 dmu_buf_impl_t
*db
, *db_next
;
403 mutex_enter(&dn
->dn_dbufs_mtx
);
404 for (db
= avl_first(&dn
->dn_dbufs
); db
!= NULL
; db
= db_next
) {
408 ASSERT3P(DB_DNODE(db
), ==, dn
);
412 mutex_enter(&db
->db_mtx
);
413 if (db
->db_state
!= DB_EVICTING
&&
414 refcount_is_zero(&db
->db_holds
)) {
415 db_marker
.db_level
= db
->db_level
;
416 db_marker
.db_blkid
= db
->db_blkid
;
417 db_marker
.db_state
= DB_SEARCH
;
418 avl_insert_here(&dn
->dn_dbufs
, &db_marker
, db
,
423 db_next
= AVL_NEXT(&dn
->dn_dbufs
, &db_marker
);
424 avl_remove(&dn
->dn_dbufs
, &db_marker
);
426 mutex_exit(&db
->db_mtx
);
427 db_next
= AVL_NEXT(&dn
->dn_dbufs
, db
);
430 mutex_exit(&dn
->dn_dbufs_mtx
);
432 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
433 if (dn
->dn_bonus
&& refcount_is_zero(&dn
->dn_bonus
->db_holds
)) {
434 mutex_enter(&dn
->dn_bonus
->db_mtx
);
435 dbuf_evict(dn
->dn_bonus
);
438 rw_exit(&dn
->dn_struct_rwlock
);
442 dnode_undirty_dbufs(list_t
*list
)
444 dbuf_dirty_record_t
*dr
;
446 while (dr
= list_head(list
)) {
447 dmu_buf_impl_t
*db
= dr
->dr_dbuf
;
448 uint64_t txg
= dr
->dr_txg
;
450 if (db
->db_level
!= 0)
451 dnode_undirty_dbufs(&dr
->dt
.di
.dr_children
);
453 mutex_enter(&db
->db_mtx
);
454 /* XXX - use dbuf_undirty()? */
455 list_remove(list
, dr
);
456 ASSERT(db
->db_last_dirty
== dr
);
457 db
->db_last_dirty
= NULL
;
458 db
->db_dirtycnt
-= 1;
459 if (db
->db_level
== 0) {
460 ASSERT(db
->db_blkid
== DMU_BONUS_BLKID
||
461 dr
->dt
.dl
.dr_data
== db
->db_buf
);
464 mutex_destroy(&dr
->dt
.di
.dr_mtx
);
465 list_destroy(&dr
->dt
.di
.dr_children
);
467 kmem_free(dr
, sizeof (dbuf_dirty_record_t
));
468 dbuf_rele_and_unlock(db
, (void *)(uintptr_t)txg
);
473 dnode_sync_free(dnode_t
*dn
, dmu_tx_t
*tx
)
475 int txgoff
= tx
->tx_txg
& TXG_MASK
;
477 ASSERT(dmu_tx_is_syncing(tx
));
480 * Our contents should have been freed in dnode_sync() by the
481 * free range record inserted by the caller of dnode_free().
483 ASSERT0(DN_USED_BYTES(dn
->dn_phys
));
484 ASSERT(BP_IS_HOLE(dn
->dn_phys
->dn_blkptr
));
486 dnode_undirty_dbufs(&dn
->dn_dirty_records
[txgoff
]);
487 dnode_evict_dbufs(dn
);
488 ASSERT(avl_is_empty(&dn
->dn_dbufs
));
491 * XXX - It would be nice to assert this, but we may still
492 * have residual holds from async evictions from the arc...
494 * zfs_obj_to_path() also depends on this being
497 * ASSERT3U(refcount_count(&dn->dn_holds), ==, 1);
500 /* Undirty next bits */
501 dn
->dn_next_nlevels
[txgoff
] = 0;
502 dn
->dn_next_indblkshift
[txgoff
] = 0;
503 dn
->dn_next_blksz
[txgoff
] = 0;
505 /* ASSERT(blkptrs are zero); */
506 ASSERT(dn
->dn_phys
->dn_type
!= DMU_OT_NONE
);
507 ASSERT(dn
->dn_type
!= DMU_OT_NONE
);
509 ASSERT(dn
->dn_free_txg
> 0);
510 if (dn
->dn_allocated_txg
!= dn
->dn_free_txg
)
511 dmu_buf_will_dirty(&dn
->dn_dbuf
->db
, tx
);
512 bzero(dn
->dn_phys
, sizeof (dnode_phys_t
));
514 mutex_enter(&dn
->dn_mtx
);
515 dn
->dn_type
= DMU_OT_NONE
;
517 dn
->dn_allocated_txg
= 0;
519 dn
->dn_have_spill
= B_FALSE
;
520 mutex_exit(&dn
->dn_mtx
);
522 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
524 dnode_rele(dn
, (void *)(uintptr_t)tx
->tx_txg
);
526 * Now that we've released our hold, the dnode may
527 * be evicted, so we musn't access it.
532 * Write out the dnode's dirty buffers.
535 dnode_sync(dnode_t
*dn
, dmu_tx_t
*tx
)
537 dnode_phys_t
*dnp
= dn
->dn_phys
;
538 int txgoff
= tx
->tx_txg
& TXG_MASK
;
539 list_t
*list
= &dn
->dn_dirty_records
[txgoff
];
540 static const dnode_phys_t zerodn
= { 0 };
541 boolean_t kill_spill
= B_FALSE
;
543 ASSERT(dmu_tx_is_syncing(tx
));
544 ASSERT(dnp
->dn_type
!= DMU_OT_NONE
|| dn
->dn_allocated_txg
);
545 ASSERT(dnp
->dn_type
!= DMU_OT_NONE
||
546 bcmp(dnp
, &zerodn
, DNODE_SIZE
) == 0);
549 ASSERT(dn
->dn_dbuf
== NULL
|| arc_released(dn
->dn_dbuf
->db_buf
));
551 if (dmu_objset_userused_enabled(dn
->dn_objset
) &&
552 !DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
553 mutex_enter(&dn
->dn_mtx
);
554 dn
->dn_oldused
= DN_USED_BYTES(dn
->dn_phys
);
555 dn
->dn_oldflags
= dn
->dn_phys
->dn_flags
;
556 dn
->dn_phys
->dn_flags
|= DNODE_FLAG_USERUSED_ACCOUNTED
;
557 mutex_exit(&dn
->dn_mtx
);
558 dmu_objset_userquota_get_ids(dn
, B_FALSE
, tx
);
560 /* Once we account for it, we should always account for it. */
561 ASSERT(!(dn
->dn_phys
->dn_flags
&
562 DNODE_FLAG_USERUSED_ACCOUNTED
));
565 mutex_enter(&dn
->dn_mtx
);
566 if (dn
->dn_allocated_txg
== tx
->tx_txg
) {
567 /* The dnode is newly allocated or reallocated */
568 if (dnp
->dn_type
== DMU_OT_NONE
) {
569 /* this is a first alloc, not a realloc */
571 dnp
->dn_nblkptr
= dn
->dn_nblkptr
;
574 dnp
->dn_type
= dn
->dn_type
;
575 dnp
->dn_bonustype
= dn
->dn_bonustype
;
576 dnp
->dn_bonuslen
= dn
->dn_bonuslen
;
578 ASSERT(dnp
->dn_nlevels
> 1 ||
579 BP_IS_HOLE(&dnp
->dn_blkptr
[0]) ||
580 BP_IS_EMBEDDED(&dnp
->dn_blkptr
[0]) ||
581 BP_GET_LSIZE(&dnp
->dn_blkptr
[0]) ==
582 dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
583 ASSERT(dnp
->dn_nlevels
< 2 ||
584 BP_IS_HOLE(&dnp
->dn_blkptr
[0]) ||
585 BP_GET_LSIZE(&dnp
->dn_blkptr
[0]) == 1 << dnp
->dn_indblkshift
);
587 if (dn
->dn_next_type
[txgoff
] != 0) {
588 dnp
->dn_type
= dn
->dn_type
;
589 dn
->dn_next_type
[txgoff
] = 0;
592 if (dn
->dn_next_blksz
[txgoff
] != 0) {
593 ASSERT(P2PHASE(dn
->dn_next_blksz
[txgoff
],
594 SPA_MINBLOCKSIZE
) == 0);
595 ASSERT(BP_IS_HOLE(&dnp
->dn_blkptr
[0]) ||
596 dn
->dn_maxblkid
== 0 || list_head(list
) != NULL
||
597 dn
->dn_next_blksz
[txgoff
] >> SPA_MINBLOCKSHIFT
==
598 dnp
->dn_datablkszsec
||
599 range_tree_space(dn
->dn_free_ranges
[txgoff
]) != 0);
600 dnp
->dn_datablkszsec
=
601 dn
->dn_next_blksz
[txgoff
] >> SPA_MINBLOCKSHIFT
;
602 dn
->dn_next_blksz
[txgoff
] = 0;
605 if (dn
->dn_next_bonuslen
[txgoff
] != 0) {
606 if (dn
->dn_next_bonuslen
[txgoff
] == DN_ZERO_BONUSLEN
)
607 dnp
->dn_bonuslen
= 0;
609 dnp
->dn_bonuslen
= dn
->dn_next_bonuslen
[txgoff
];
610 ASSERT(dnp
->dn_bonuslen
<= DN_MAX_BONUSLEN
);
611 dn
->dn_next_bonuslen
[txgoff
] = 0;
614 if (dn
->dn_next_bonustype
[txgoff
] != 0) {
615 ASSERT(DMU_OT_IS_VALID(dn
->dn_next_bonustype
[txgoff
]));
616 dnp
->dn_bonustype
= dn
->dn_next_bonustype
[txgoff
];
617 dn
->dn_next_bonustype
[txgoff
] = 0;
620 boolean_t freeing_dnode
= dn
->dn_free_txg
> 0 &&
621 dn
->dn_free_txg
<= tx
->tx_txg
;
624 * Remove the spill block if we have been explicitly asked to
625 * remove it, or if the object is being removed.
627 if (dn
->dn_rm_spillblk
[txgoff
] || freeing_dnode
) {
628 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)
630 dn
->dn_rm_spillblk
[txgoff
] = 0;
633 if (dn
->dn_next_indblkshift
[txgoff
] != 0) {
634 ASSERT(dnp
->dn_nlevels
== 1);
635 dnp
->dn_indblkshift
= dn
->dn_next_indblkshift
[txgoff
];
636 dn
->dn_next_indblkshift
[txgoff
] = 0;
640 * Just take the live (open-context) values for checksum and compress.
641 * Strictly speaking it's a future leak, but nothing bad happens if we
642 * start using the new checksum or compress algorithm a little early.
644 dnp
->dn_checksum
= dn
->dn_checksum
;
645 dnp
->dn_compress
= dn
->dn_compress
;
647 mutex_exit(&dn
->dn_mtx
);
650 free_blocks(dn
, &dn
->dn_phys
->dn_spill
, 1, tx
);
651 mutex_enter(&dn
->dn_mtx
);
652 dnp
->dn_flags
&= ~DNODE_FLAG_SPILL_BLKPTR
;
653 mutex_exit(&dn
->dn_mtx
);
656 /* process all the "freed" ranges in the file */
657 if (dn
->dn_free_ranges
[txgoff
] != NULL
) {
658 dnode_sync_free_range_arg_t dsfra
;
659 dsfra
.dsfra_dnode
= dn
;
661 mutex_enter(&dn
->dn_mtx
);
662 range_tree_vacate(dn
->dn_free_ranges
[txgoff
],
663 dnode_sync_free_range
, &dsfra
);
664 range_tree_destroy(dn
->dn_free_ranges
[txgoff
]);
665 dn
->dn_free_ranges
[txgoff
] = NULL
;
666 mutex_exit(&dn
->dn_mtx
);
670 dnode_sync_free(dn
, tx
);
674 if (dn
->dn_next_nlevels
[txgoff
]) {
675 dnode_increase_indirection(dn
, tx
);
676 dn
->dn_next_nlevels
[txgoff
] = 0;
679 if (dn
->dn_next_nblkptr
[txgoff
]) {
680 /* this should only happen on a realloc */
681 ASSERT(dn
->dn_allocated_txg
== tx
->tx_txg
);
682 if (dn
->dn_next_nblkptr
[txgoff
] > dnp
->dn_nblkptr
) {
683 /* zero the new blkptrs we are gaining */
684 bzero(dnp
->dn_blkptr
+ dnp
->dn_nblkptr
,
686 (dn
->dn_next_nblkptr
[txgoff
] - dnp
->dn_nblkptr
));
690 ASSERT(dn
->dn_next_nblkptr
[txgoff
] < dnp
->dn_nblkptr
);
691 /* the blkptrs we are losing better be unallocated */
692 for (i
= dn
->dn_next_nblkptr
[txgoff
];
693 i
< dnp
->dn_nblkptr
; i
++)
694 ASSERT(BP_IS_HOLE(&dnp
->dn_blkptr
[i
]));
697 mutex_enter(&dn
->dn_mtx
);
698 dnp
->dn_nblkptr
= dn
->dn_next_nblkptr
[txgoff
];
699 dn
->dn_next_nblkptr
[txgoff
] = 0;
700 mutex_exit(&dn
->dn_mtx
);
703 dbuf_sync_list(list
, tx
);
705 if (!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
706 ASSERT3P(list_head(list
), ==, NULL
);
707 dnode_rele(dn
, (void *)(uintptr_t)tx
->tx_txg
);
711 * Although we have dropped our reference to the dnode, it
712 * can't be evicted until its written, and we haven't yet
713 * initiated the IO for the dnode's dbuf.