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) 2012, 2014 by Delphix. All rights reserved.
26 #include <sys/zfs_context.h>
28 #include <sys/dnode.h>
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_dataset.h>
37 #include <sys/dmu_zfetch.h>
38 #include <sys/range_tree.h>
40 static kmem_cache_t
*dnode_cache
;
42 * Define DNODE_STATS to turn on statistic gathering. By default, it is only
43 * turned on when DEBUG is also defined.
50 #define DNODE_STAT_ADD(stat) ((stat)++)
52 #define DNODE_STAT_ADD(stat) /* nothing */
53 #endif /* DNODE_STATS */
55 ASSERTV(static dnode_phys_t dnode_phys_zero
);
57 int zfs_default_bs
= SPA_MINBLOCKSHIFT
;
58 int zfs_default_ibs
= DN_MAX_INDBLKSHIFT
;
61 static kmem_cbrc_t
dnode_move(void *, void *, size_t, void *);
66 dnode_cons(void *arg
, void *unused
, int kmflag
)
71 rw_init(&dn
->dn_struct_rwlock
, NULL
, RW_DEFAULT
, NULL
);
72 mutex_init(&dn
->dn_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
73 mutex_init(&dn
->dn_dbufs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
74 cv_init(&dn
->dn_notxholds
, NULL
, CV_DEFAULT
, NULL
);
77 * Every dbuf has a reference, and dropping a tracked reference is
78 * O(number of references), so don't track dn_holds.
80 refcount_create_untracked(&dn
->dn_holds
);
81 refcount_create(&dn
->dn_tx_holds
);
82 list_link_init(&dn
->dn_link
);
84 bzero(&dn
->dn_next_nblkptr
[0], sizeof (dn
->dn_next_nblkptr
));
85 bzero(&dn
->dn_next_nlevels
[0], sizeof (dn
->dn_next_nlevels
));
86 bzero(&dn
->dn_next_indblkshift
[0], sizeof (dn
->dn_next_indblkshift
));
87 bzero(&dn
->dn_next_bonustype
[0], sizeof (dn
->dn_next_bonustype
));
88 bzero(&dn
->dn_rm_spillblk
[0], sizeof (dn
->dn_rm_spillblk
));
89 bzero(&dn
->dn_next_bonuslen
[0], sizeof (dn
->dn_next_bonuslen
));
90 bzero(&dn
->dn_next_blksz
[0], sizeof (dn
->dn_next_blksz
));
92 for (i
= 0; i
< TXG_SIZE
; i
++) {
93 list_link_init(&dn
->dn_dirty_link
[i
]);
94 dn
->dn_free_ranges
[i
] = NULL
;
95 list_create(&dn
->dn_dirty_records
[i
],
96 sizeof (dbuf_dirty_record_t
),
97 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
100 dn
->dn_allocated_txg
= 0;
102 dn
->dn_assigned_txg
= 0;
104 dn
->dn_dirtyctx_firstset
= NULL
;
106 dn
->dn_have_spill
= B_FALSE
;
116 dn
->dn_dbufs_count
= 0;
117 dn
->dn_unlisted_l0_blkid
= 0;
118 list_create(&dn
->dn_dbufs
, sizeof (dmu_buf_impl_t
),
119 offsetof(dmu_buf_impl_t
, db_link
));
127 dnode_dest(void *arg
, void *unused
)
132 rw_destroy(&dn
->dn_struct_rwlock
);
133 mutex_destroy(&dn
->dn_mtx
);
134 mutex_destroy(&dn
->dn_dbufs_mtx
);
135 cv_destroy(&dn
->dn_notxholds
);
136 refcount_destroy(&dn
->dn_holds
);
137 refcount_destroy(&dn
->dn_tx_holds
);
138 ASSERT(!list_link_active(&dn
->dn_link
));
140 for (i
= 0; i
< TXG_SIZE
; i
++) {
141 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
142 ASSERT3P(dn
->dn_free_ranges
[i
], ==, NULL
);
143 list_destroy(&dn
->dn_dirty_records
[i
]);
144 ASSERT0(dn
->dn_next_nblkptr
[i
]);
145 ASSERT0(dn
->dn_next_nlevels
[i
]);
146 ASSERT0(dn
->dn_next_indblkshift
[i
]);
147 ASSERT0(dn
->dn_next_bonustype
[i
]);
148 ASSERT0(dn
->dn_rm_spillblk
[i
]);
149 ASSERT0(dn
->dn_next_bonuslen
[i
]);
150 ASSERT0(dn
->dn_next_blksz
[i
]);
153 ASSERT0(dn
->dn_allocated_txg
);
154 ASSERT0(dn
->dn_free_txg
);
155 ASSERT0(dn
->dn_assigned_txg
);
156 ASSERT0(dn
->dn_dirtyctx
);
157 ASSERT3P(dn
->dn_dirtyctx_firstset
, ==, NULL
);
158 ASSERT3P(dn
->dn_bonus
, ==, NULL
);
159 ASSERT(!dn
->dn_have_spill
);
160 ASSERT3P(dn
->dn_zio
, ==, NULL
);
161 ASSERT0(dn
->dn_oldused
);
162 ASSERT0(dn
->dn_oldflags
);
163 ASSERT0(dn
->dn_olduid
);
164 ASSERT0(dn
->dn_oldgid
);
165 ASSERT0(dn
->dn_newuid
);
166 ASSERT0(dn
->dn_newgid
);
167 ASSERT0(dn
->dn_id_flags
);
169 ASSERT0(dn
->dn_dbufs_count
);
170 ASSERT0(dn
->dn_unlisted_l0_blkid
);
171 list_destroy(&dn
->dn_dbufs
);
177 ASSERT(dnode_cache
== NULL
);
178 dnode_cache
= kmem_cache_create("dnode_t", sizeof (dnode_t
),
179 0, dnode_cons
, dnode_dest
, NULL
, NULL
, NULL
, 0);
180 kmem_cache_set_move(dnode_cache
, dnode_move
);
186 kmem_cache_destroy(dnode_cache
);
193 dnode_verify(dnode_t
*dn
)
195 int drop_struct_lock
= FALSE
;
198 ASSERT(dn
->dn_objset
);
199 ASSERT(dn
->dn_handle
->dnh_dnode
== dn
);
201 ASSERT(DMU_OT_IS_VALID(dn
->dn_phys
->dn_type
));
203 if (!(zfs_flags
& ZFS_DEBUG_DNODE_VERIFY
))
206 if (!RW_WRITE_HELD(&dn
->dn_struct_rwlock
)) {
207 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
208 drop_struct_lock
= TRUE
;
210 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
|| dn
->dn_allocated_txg
!= 0) {
212 ASSERT3U(dn
->dn_indblkshift
, <=, SPA_MAXBLOCKSHIFT
);
213 if (dn
->dn_datablkshift
) {
214 ASSERT3U(dn
->dn_datablkshift
, >=, SPA_MINBLOCKSHIFT
);
215 ASSERT3U(dn
->dn_datablkshift
, <=, SPA_MAXBLOCKSHIFT
);
216 ASSERT3U(1<<dn
->dn_datablkshift
, ==, dn
->dn_datablksz
);
218 ASSERT3U(dn
->dn_nlevels
, <=, 30);
219 ASSERT(DMU_OT_IS_VALID(dn
->dn_type
));
220 ASSERT3U(dn
->dn_nblkptr
, >=, 1);
221 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
222 ASSERT3U(dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
223 ASSERT3U(dn
->dn_datablksz
, ==,
224 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
225 ASSERT3U(ISP2(dn
->dn_datablksz
), ==, dn
->dn_datablkshift
!= 0);
226 ASSERT3U((dn
->dn_nblkptr
- 1) * sizeof (blkptr_t
) +
227 dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
228 for (i
= 0; i
< TXG_SIZE
; i
++) {
229 ASSERT3U(dn
->dn_next_nlevels
[i
], <=, dn
->dn_nlevels
);
232 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
)
233 ASSERT3U(dn
->dn_phys
->dn_nlevels
, <=, dn
->dn_nlevels
);
234 ASSERT(DMU_OBJECT_IS_SPECIAL(dn
->dn_object
) || dn
->dn_dbuf
!= NULL
);
235 if (dn
->dn_dbuf
!= NULL
) {
236 ASSERT3P(dn
->dn_phys
, ==,
237 (dnode_phys_t
*)dn
->dn_dbuf
->db
.db_data
+
238 (dn
->dn_object
% (dn
->dn_dbuf
->db
.db_size
>> DNODE_SHIFT
)));
240 if (drop_struct_lock
)
241 rw_exit(&dn
->dn_struct_rwlock
);
246 dnode_byteswap(dnode_phys_t
*dnp
)
248 uint64_t *buf64
= (void*)&dnp
->dn_blkptr
;
251 if (dnp
->dn_type
== DMU_OT_NONE
) {
252 bzero(dnp
, sizeof (dnode_phys_t
));
256 dnp
->dn_datablkszsec
= BSWAP_16(dnp
->dn_datablkszsec
);
257 dnp
->dn_bonuslen
= BSWAP_16(dnp
->dn_bonuslen
);
258 dnp
->dn_maxblkid
= BSWAP_64(dnp
->dn_maxblkid
);
259 dnp
->dn_used
= BSWAP_64(dnp
->dn_used
);
262 * dn_nblkptr is only one byte, so it's OK to read it in either
263 * byte order. We can't read dn_bouslen.
265 ASSERT(dnp
->dn_indblkshift
<= SPA_MAXBLOCKSHIFT
);
266 ASSERT(dnp
->dn_nblkptr
<= DN_MAX_NBLKPTR
);
267 for (i
= 0; i
< dnp
->dn_nblkptr
* sizeof (blkptr_t
)/8; i
++)
268 buf64
[i
] = BSWAP_64(buf64
[i
]);
271 * OK to check dn_bonuslen for zero, because it won't matter if
272 * we have the wrong byte order. This is necessary because the
273 * dnode dnode is smaller than a regular dnode.
275 if (dnp
->dn_bonuslen
!= 0) {
277 * Note that the bonus length calculated here may be
278 * longer than the actual bonus buffer. This is because
279 * we always put the bonus buffer after the last block
280 * pointer (instead of packing it against the end of the
283 int off
= (dnp
->dn_nblkptr
-1) * sizeof (blkptr_t
);
284 size_t len
= DN_MAX_BONUSLEN
- off
;
285 dmu_object_byteswap_t byteswap
;
286 ASSERT(DMU_OT_IS_VALID(dnp
->dn_bonustype
));
287 byteswap
= DMU_OT_BYTESWAP(dnp
->dn_bonustype
);
288 dmu_ot_byteswap
[byteswap
].ob_func(dnp
->dn_bonus
+ off
, len
);
291 /* Swap SPILL block if we have one */
292 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)
293 byteswap_uint64_array(&dnp
->dn_spill
, sizeof (blkptr_t
));
298 dnode_buf_byteswap(void *vbuf
, size_t size
)
300 dnode_phys_t
*buf
= vbuf
;
303 ASSERT3U(sizeof (dnode_phys_t
), ==, (1<<DNODE_SHIFT
));
304 ASSERT((size
& (sizeof (dnode_phys_t
)-1)) == 0);
306 size
>>= DNODE_SHIFT
;
307 for (i
= 0; i
< size
; i
++) {
314 dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
)
316 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
318 dnode_setdirty(dn
, tx
);
319 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
320 ASSERT3U(newsize
, <=, DN_MAX_BONUSLEN
-
321 (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
));
322 dn
->dn_bonuslen
= newsize
;
324 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = DN_ZERO_BONUSLEN
;
326 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
327 rw_exit(&dn
->dn_struct_rwlock
);
331 dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t newtype
, dmu_tx_t
*tx
)
333 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
334 dnode_setdirty(dn
, tx
);
335 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
336 dn
->dn_bonustype
= newtype
;
337 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
338 rw_exit(&dn
->dn_struct_rwlock
);
342 dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
)
344 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
345 ASSERT(RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
346 dnode_setdirty(dn
, tx
);
347 dn
->dn_rm_spillblk
[tx
->tx_txg
&TXG_MASK
] = DN_KILL_SPILLBLK
;
348 dn
->dn_have_spill
= B_FALSE
;
352 dnode_setdblksz(dnode_t
*dn
, int size
)
354 ASSERT0(P2PHASE(size
, SPA_MINBLOCKSIZE
));
355 ASSERT3U(size
, <=, SPA_MAXBLOCKSIZE
);
356 ASSERT3U(size
, >=, SPA_MINBLOCKSIZE
);
357 ASSERT3U(size
>> SPA_MINBLOCKSHIFT
, <,
358 1<<(sizeof (dn
->dn_phys
->dn_datablkszsec
) * 8));
359 dn
->dn_datablksz
= size
;
360 dn
->dn_datablkszsec
= size
>> SPA_MINBLOCKSHIFT
;
361 dn
->dn_datablkshift
= ISP2(size
) ? highbit64(size
- 1) : 0;
365 dnode_create(objset_t
*os
, dnode_phys_t
*dnp
, dmu_buf_impl_t
*db
,
366 uint64_t object
, dnode_handle_t
*dnh
)
368 dnode_t
*dn
= kmem_cache_alloc(dnode_cache
, KM_PUSHPAGE
);
370 ASSERT(!POINTER_IS_VALID(dn
->dn_objset
));
374 * Defer setting dn_objset until the dnode is ready to be a candidate
375 * for the dnode_move() callback.
377 dn
->dn_object
= object
;
382 if (dnp
->dn_datablkszsec
) {
383 dnode_setdblksz(dn
, dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
385 dn
->dn_datablksz
= 0;
386 dn
->dn_datablkszsec
= 0;
387 dn
->dn_datablkshift
= 0;
389 dn
->dn_indblkshift
= dnp
->dn_indblkshift
;
390 dn
->dn_nlevels
= dnp
->dn_nlevels
;
391 dn
->dn_type
= dnp
->dn_type
;
392 dn
->dn_nblkptr
= dnp
->dn_nblkptr
;
393 dn
->dn_checksum
= dnp
->dn_checksum
;
394 dn
->dn_compress
= dnp
->dn_compress
;
395 dn
->dn_bonustype
= dnp
->dn_bonustype
;
396 dn
->dn_bonuslen
= dnp
->dn_bonuslen
;
397 dn
->dn_maxblkid
= dnp
->dn_maxblkid
;
398 dn
->dn_have_spill
= ((dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) != 0);
401 dmu_zfetch_init(&dn
->dn_zfetch
, dn
);
403 ASSERT(DMU_OT_IS_VALID(dn
->dn_phys
->dn_type
));
405 mutex_enter(&os
->os_lock
);
406 list_insert_head(&os
->os_dnodes
, dn
);
409 * Everything else must be valid before assigning dn_objset makes the
410 * dnode eligible for dnode_move().
413 mutex_exit(&os
->os_lock
);
415 arc_space_consume(sizeof (dnode_t
), ARC_SPACE_OTHER
);
420 * Caller must be holding the dnode handle, which is released upon return.
423 dnode_destroy(dnode_t
*dn
)
425 objset_t
*os
= dn
->dn_objset
;
427 ASSERT((dn
->dn_id_flags
& DN_ID_NEW_EXIST
) == 0);
429 mutex_enter(&os
->os_lock
);
430 POINTER_INVALIDATE(&dn
->dn_objset
);
431 list_remove(&os
->os_dnodes
, dn
);
432 mutex_exit(&os
->os_lock
);
434 /* the dnode can no longer move, so we can release the handle */
435 zrl_remove(&dn
->dn_handle
->dnh_zrlock
);
437 dn
->dn_allocated_txg
= 0;
439 dn
->dn_assigned_txg
= 0;
442 if (dn
->dn_dirtyctx_firstset
!= NULL
) {
443 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
444 dn
->dn_dirtyctx_firstset
= NULL
;
446 if (dn
->dn_bonus
!= NULL
) {
447 mutex_enter(&dn
->dn_bonus
->db_mtx
);
448 dbuf_evict(dn
->dn_bonus
);
453 dn
->dn_have_spill
= B_FALSE
;
461 dn
->dn_unlisted_l0_blkid
= 0;
463 dmu_zfetch_rele(&dn
->dn_zfetch
);
464 kmem_cache_free(dnode_cache
, dn
);
465 arc_space_return(sizeof (dnode_t
), ARC_SPACE_OTHER
);
469 dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
470 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
475 blocksize
= 1 << zfs_default_bs
;
476 else if (blocksize
> SPA_MAXBLOCKSIZE
)
477 blocksize
= SPA_MAXBLOCKSIZE
;
479 blocksize
= P2ROUNDUP(blocksize
, SPA_MINBLOCKSIZE
);
482 ibs
= zfs_default_ibs
;
484 ibs
= MIN(MAX(ibs
, DN_MIN_INDBLKSHIFT
), DN_MAX_INDBLKSHIFT
);
486 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn
->dn_objset
,
487 dn
->dn_object
, tx
->tx_txg
, blocksize
, ibs
);
489 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
490 ASSERT(bcmp(dn
->dn_phys
, &dnode_phys_zero
, sizeof (dnode_phys_t
)) == 0);
491 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
);
492 ASSERT(ot
!= DMU_OT_NONE
);
493 ASSERT(DMU_OT_IS_VALID(ot
));
494 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
495 (bonustype
== DMU_OT_SA
&& bonuslen
== 0) ||
496 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0));
497 ASSERT(DMU_OT_IS_VALID(bonustype
));
498 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
499 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
500 ASSERT0(dn
->dn_maxblkid
);
501 ASSERT0(dn
->dn_allocated_txg
);
502 ASSERT0(dn
->dn_assigned_txg
);
503 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
504 ASSERT3U(refcount_count(&dn
->dn_holds
), <=, 1);
505 ASSERT3P(list_head(&dn
->dn_dbufs
), ==, NULL
);
507 for (i
= 0; i
< TXG_SIZE
; i
++) {
508 ASSERT0(dn
->dn_next_nblkptr
[i
]);
509 ASSERT0(dn
->dn_next_nlevels
[i
]);
510 ASSERT0(dn
->dn_next_indblkshift
[i
]);
511 ASSERT0(dn
->dn_next_bonuslen
[i
]);
512 ASSERT0(dn
->dn_next_bonustype
[i
]);
513 ASSERT0(dn
->dn_rm_spillblk
[i
]);
514 ASSERT0(dn
->dn_next_blksz
[i
]);
515 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
516 ASSERT3P(list_head(&dn
->dn_dirty_records
[i
]), ==, NULL
);
517 ASSERT3P(dn
->dn_free_ranges
[i
], ==, NULL
);
521 dnode_setdblksz(dn
, blocksize
);
522 dn
->dn_indblkshift
= ibs
;
524 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
528 ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
529 dn
->dn_bonustype
= bonustype
;
530 dn
->dn_bonuslen
= bonuslen
;
531 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
532 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
536 if (dn
->dn_dirtyctx_firstset
) {
537 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
538 dn
->dn_dirtyctx_firstset
= NULL
;
541 dn
->dn_allocated_txg
= tx
->tx_txg
;
544 dnode_setdirty(dn
, tx
);
545 dn
->dn_next_indblkshift
[tx
->tx_txg
& TXG_MASK
] = ibs
;
546 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
547 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
548 dn
->dn_next_blksz
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_datablksz
;
552 dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
553 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
557 ASSERT3U(blocksize
, >=, SPA_MINBLOCKSIZE
);
558 ASSERT3U(blocksize
, <=, SPA_MAXBLOCKSIZE
);
559 ASSERT0(blocksize
% SPA_MINBLOCKSIZE
);
560 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
|| dmu_tx_private_ok(tx
));
561 ASSERT(tx
->tx_txg
!= 0);
562 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
563 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0) ||
564 (bonustype
== DMU_OT_SA
&& bonuslen
== 0));
565 ASSERT(DMU_OT_IS_VALID(bonustype
));
566 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
568 /* clean up any unreferenced dbufs */
569 dnode_evict_dbufs(dn
);
573 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
574 dnode_setdirty(dn
, tx
);
575 if (dn
->dn_datablksz
!= blocksize
) {
576 /* change blocksize */
577 ASSERT(dn
->dn_maxblkid
== 0 &&
578 (BP_IS_HOLE(&dn
->dn_phys
->dn_blkptr
[0]) ||
579 dnode_block_freed(dn
, 0)));
580 dnode_setdblksz(dn
, blocksize
);
581 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = blocksize
;
583 if (dn
->dn_bonuslen
!= bonuslen
)
584 dn
->dn_next_bonuslen
[tx
->tx_txg
&TXG_MASK
] = bonuslen
;
586 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
589 nblkptr
= 1 + ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
590 if (dn
->dn_bonustype
!= bonustype
)
591 dn
->dn_next_bonustype
[tx
->tx_txg
&TXG_MASK
] = bonustype
;
592 if (dn
->dn_nblkptr
!= nblkptr
)
593 dn
->dn_next_nblkptr
[tx
->tx_txg
&TXG_MASK
] = nblkptr
;
594 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
595 dbuf_rm_spill(dn
, tx
);
596 dnode_rm_spill(dn
, tx
);
598 rw_exit(&dn
->dn_struct_rwlock
);
603 /* change bonus size and type */
604 mutex_enter(&dn
->dn_mtx
);
605 dn
->dn_bonustype
= bonustype
;
606 dn
->dn_bonuslen
= bonuslen
;
607 dn
->dn_nblkptr
= nblkptr
;
608 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
609 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
610 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
612 /* fix up the bonus db_size */
614 dn
->dn_bonus
->db
.db_size
=
615 DN_MAX_BONUSLEN
- (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
);
616 ASSERT(dn
->dn_bonuslen
<= dn
->dn_bonus
->db
.db_size
);
619 dn
->dn_allocated_txg
= tx
->tx_txg
;
620 mutex_exit(&dn
->dn_mtx
);
626 uint64_t dms_dnode_invalid
;
627 uint64_t dms_dnode_recheck1
;
628 uint64_t dms_dnode_recheck2
;
629 uint64_t dms_dnode_special
;
630 uint64_t dms_dnode_handle
;
631 uint64_t dms_dnode_rwlock
;
632 uint64_t dms_dnode_active
;
634 #endif /* DNODE_STATS */
637 dnode_move_impl(dnode_t
*odn
, dnode_t
*ndn
)
641 ASSERT(!RW_LOCK_HELD(&odn
->dn_struct_rwlock
));
642 ASSERT(MUTEX_NOT_HELD(&odn
->dn_mtx
));
643 ASSERT(MUTEX_NOT_HELD(&odn
->dn_dbufs_mtx
));
644 ASSERT(!RW_LOCK_HELD(&odn
->dn_zfetch
.zf_rwlock
));
647 ndn
->dn_objset
= odn
->dn_objset
;
648 ndn
->dn_object
= odn
->dn_object
;
649 ndn
->dn_dbuf
= odn
->dn_dbuf
;
650 ndn
->dn_handle
= odn
->dn_handle
;
651 ndn
->dn_phys
= odn
->dn_phys
;
652 ndn
->dn_type
= odn
->dn_type
;
653 ndn
->dn_bonuslen
= odn
->dn_bonuslen
;
654 ndn
->dn_bonustype
= odn
->dn_bonustype
;
655 ndn
->dn_nblkptr
= odn
->dn_nblkptr
;
656 ndn
->dn_checksum
= odn
->dn_checksum
;
657 ndn
->dn_compress
= odn
->dn_compress
;
658 ndn
->dn_nlevels
= odn
->dn_nlevels
;
659 ndn
->dn_indblkshift
= odn
->dn_indblkshift
;
660 ndn
->dn_datablkshift
= odn
->dn_datablkshift
;
661 ndn
->dn_datablkszsec
= odn
->dn_datablkszsec
;
662 ndn
->dn_datablksz
= odn
->dn_datablksz
;
663 ndn
->dn_maxblkid
= odn
->dn_maxblkid
;
664 bcopy(&odn
->dn_next_nblkptr
[0], &ndn
->dn_next_nblkptr
[0],
665 sizeof (odn
->dn_next_nblkptr
));
666 bcopy(&odn
->dn_next_nlevels
[0], &ndn
->dn_next_nlevels
[0],
667 sizeof (odn
->dn_next_nlevels
));
668 bcopy(&odn
->dn_next_indblkshift
[0], &ndn
->dn_next_indblkshift
[0],
669 sizeof (odn
->dn_next_indblkshift
));
670 bcopy(&odn
->dn_next_bonustype
[0], &ndn
->dn_next_bonustype
[0],
671 sizeof (odn
->dn_next_bonustype
));
672 bcopy(&odn
->dn_rm_spillblk
[0], &ndn
->dn_rm_spillblk
[0],
673 sizeof (odn
->dn_rm_spillblk
));
674 bcopy(&odn
->dn_next_bonuslen
[0], &ndn
->dn_next_bonuslen
[0],
675 sizeof (odn
->dn_next_bonuslen
));
676 bcopy(&odn
->dn_next_blksz
[0], &ndn
->dn_next_blksz
[0],
677 sizeof (odn
->dn_next_blksz
));
678 for (i
= 0; i
< TXG_SIZE
; i
++) {
679 list_move_tail(&ndn
->dn_dirty_records
[i
],
680 &odn
->dn_dirty_records
[i
]);
682 bcopy(&odn
->dn_free_ranges
[0], &ndn
->dn_free_ranges
[0],
683 sizeof (odn
->dn_free_ranges
));
684 ndn
->dn_allocated_txg
= odn
->dn_allocated_txg
;
685 ndn
->dn_free_txg
= odn
->dn_free_txg
;
686 ndn
->dn_assigned_txg
= odn
->dn_assigned_txg
;
687 ndn
->dn_dirtyctx
= odn
->dn_dirtyctx
;
688 ndn
->dn_dirtyctx_firstset
= odn
->dn_dirtyctx_firstset
;
689 ASSERT(refcount_count(&odn
->dn_tx_holds
) == 0);
690 refcount_transfer(&ndn
->dn_holds
, &odn
->dn_holds
);
691 ASSERT(list_is_empty(&ndn
->dn_dbufs
));
692 list_move_tail(&ndn
->dn_dbufs
, &odn
->dn_dbufs
);
693 ndn
->dn_dbufs_count
= odn
->dn_dbufs_count
;
694 ndn
->dn_unlisted_l0_blkid
= odn
->dn_unlisted_l0_blkid
;
695 ndn
->dn_bonus
= odn
->dn_bonus
;
696 ndn
->dn_have_spill
= odn
->dn_have_spill
;
697 ndn
->dn_zio
= odn
->dn_zio
;
698 ndn
->dn_oldused
= odn
->dn_oldused
;
699 ndn
->dn_oldflags
= odn
->dn_oldflags
;
700 ndn
->dn_olduid
= odn
->dn_olduid
;
701 ndn
->dn_oldgid
= odn
->dn_oldgid
;
702 ndn
->dn_newuid
= odn
->dn_newuid
;
703 ndn
->dn_newgid
= odn
->dn_newgid
;
704 ndn
->dn_id_flags
= odn
->dn_id_flags
;
705 dmu_zfetch_init(&ndn
->dn_zfetch
, NULL
);
706 list_move_tail(&ndn
->dn_zfetch
.zf_stream
, &odn
->dn_zfetch
.zf_stream
);
707 ndn
->dn_zfetch
.zf_dnode
= odn
->dn_zfetch
.zf_dnode
;
708 ndn
->dn_zfetch
.zf_stream_cnt
= odn
->dn_zfetch
.zf_stream_cnt
;
709 ndn
->dn_zfetch
.zf_alloc_fail
= odn
->dn_zfetch
.zf_alloc_fail
;
712 * Update back pointers. Updating the handle fixes the back pointer of
713 * every descendant dbuf as well as the bonus dbuf.
715 ASSERT(ndn
->dn_handle
->dnh_dnode
== odn
);
716 ndn
->dn_handle
->dnh_dnode
= ndn
;
717 if (ndn
->dn_zfetch
.zf_dnode
== odn
) {
718 ndn
->dn_zfetch
.zf_dnode
= ndn
;
722 * Invalidate the original dnode by clearing all of its back pointers.
725 odn
->dn_handle
= NULL
;
726 list_create(&odn
->dn_dbufs
, sizeof (dmu_buf_impl_t
),
727 offsetof(dmu_buf_impl_t
, db_link
));
728 odn
->dn_dbufs_count
= 0;
729 odn
->dn_unlisted_l0_blkid
= 0;
730 odn
->dn_bonus
= NULL
;
731 odn
->dn_zfetch
.zf_dnode
= NULL
;
734 * Set the low bit of the objset pointer to ensure that dnode_move()
735 * recognizes the dnode as invalid in any subsequent callback.
737 POINTER_INVALIDATE(&odn
->dn_objset
);
740 * Satisfy the destructor.
742 for (i
= 0; i
< TXG_SIZE
; i
++) {
743 list_create(&odn
->dn_dirty_records
[i
],
744 sizeof (dbuf_dirty_record_t
),
745 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
746 odn
->dn_free_ranges
[i
] = NULL
;
747 odn
->dn_next_nlevels
[i
] = 0;
748 odn
->dn_next_indblkshift
[i
] = 0;
749 odn
->dn_next_bonustype
[i
] = 0;
750 odn
->dn_rm_spillblk
[i
] = 0;
751 odn
->dn_next_bonuslen
[i
] = 0;
752 odn
->dn_next_blksz
[i
] = 0;
754 odn
->dn_allocated_txg
= 0;
755 odn
->dn_free_txg
= 0;
756 odn
->dn_assigned_txg
= 0;
757 odn
->dn_dirtyctx
= 0;
758 odn
->dn_dirtyctx_firstset
= NULL
;
759 odn
->dn_have_spill
= B_FALSE
;
762 odn
->dn_oldflags
= 0;
767 odn
->dn_id_flags
= 0;
773 odn
->dn_moved
= (uint8_t)-1;
778 dnode_move(void *buf
, void *newbuf
, size_t size
, void *arg
)
780 dnode_t
*odn
= buf
, *ndn
= newbuf
;
786 * The dnode is on the objset's list of known dnodes if the objset
787 * pointer is valid. We set the low bit of the objset pointer when
788 * freeing the dnode to invalidate it, and the memory patterns written
789 * by kmem (baddcafe and deadbeef) set at least one of the two low bits.
790 * A newly created dnode sets the objset pointer last of all to indicate
791 * that the dnode is known and in a valid state to be moved by this
795 if (!POINTER_IS_VALID(os
)) {
796 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_invalid
);
797 return (KMEM_CBRC_DONT_KNOW
);
801 * Ensure that the objset does not go away during the move.
803 rw_enter(&os_lock
, RW_WRITER
);
804 if (os
!= odn
->dn_objset
) {
806 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck1
);
807 return (KMEM_CBRC_DONT_KNOW
);
811 * If the dnode is still valid, then so is the objset. We know that no
812 * valid objset can be freed while we hold os_lock, so we can safely
813 * ensure that the objset remains in use.
815 mutex_enter(&os
->os_lock
);
818 * Recheck the objset pointer in case the dnode was removed just before
819 * acquiring the lock.
821 if (os
!= odn
->dn_objset
) {
822 mutex_exit(&os
->os_lock
);
824 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck2
);
825 return (KMEM_CBRC_DONT_KNOW
);
829 * At this point we know that as long as we hold os->os_lock, the dnode
830 * cannot be freed and fields within the dnode can be safely accessed.
831 * The objset listing this dnode cannot go away as long as this dnode is
835 if (DMU_OBJECT_IS_SPECIAL(odn
->dn_object
)) {
836 mutex_exit(&os
->os_lock
);
837 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_special
);
838 return (KMEM_CBRC_NO
);
840 ASSERT(odn
->dn_dbuf
!= NULL
); /* only "special" dnodes have no parent */
843 * Lock the dnode handle to prevent the dnode from obtaining any new
844 * holds. This also prevents the descendant dbufs and the bonus dbuf
845 * from accessing the dnode, so that we can discount their holds. The
846 * handle is safe to access because we know that while the dnode cannot
847 * go away, neither can its handle. Once we hold dnh_zrlock, we can
848 * safely move any dnode referenced only by dbufs.
850 if (!zrl_tryenter(&odn
->dn_handle
->dnh_zrlock
)) {
851 mutex_exit(&os
->os_lock
);
852 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_handle
);
853 return (KMEM_CBRC_LATER
);
857 * Ensure a consistent view of the dnode's holds and the dnode's dbufs.
858 * We need to guarantee that there is a hold for every dbuf in order to
859 * determine whether the dnode is actively referenced. Falsely matching
860 * a dbuf to an active hold would lead to an unsafe move. It's possible
861 * that a thread already having an active dnode hold is about to add a
862 * dbuf, and we can't compare hold and dbuf counts while the add is in
865 if (!rw_tryenter(&odn
->dn_struct_rwlock
, RW_WRITER
)) {
866 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
867 mutex_exit(&os
->os_lock
);
868 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_rwlock
);
869 return (KMEM_CBRC_LATER
);
873 * A dbuf may be removed (evicted) without an active dnode hold. In that
874 * case, the dbuf count is decremented under the handle lock before the
875 * dbuf's hold is released. This order ensures that if we count the hold
876 * after the dbuf is removed but before its hold is released, we will
877 * treat the unmatched hold as active and exit safely. If we count the
878 * hold before the dbuf is removed, the hold is discounted, and the
879 * removal is blocked until the move completes.
881 refcount
= refcount_count(&odn
->dn_holds
);
882 ASSERT(refcount
>= 0);
883 dbufs
= odn
->dn_dbufs_count
;
885 /* We can't have more dbufs than dnode holds. */
886 ASSERT3U(dbufs
, <=, refcount
);
887 DTRACE_PROBE3(dnode__move
, dnode_t
*, odn
, int64_t, refcount
,
890 if (refcount
> dbufs
) {
891 rw_exit(&odn
->dn_struct_rwlock
);
892 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
893 mutex_exit(&os
->os_lock
);
894 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_active
);
895 return (KMEM_CBRC_LATER
);
898 rw_exit(&odn
->dn_struct_rwlock
);
901 * At this point we know that anyone with a hold on the dnode is not
902 * actively referencing it. The dnode is known and in a valid state to
903 * move. We're holding the locks needed to execute the critical section.
905 dnode_move_impl(odn
, ndn
);
907 list_link_replace(&odn
->dn_link
, &ndn
->dn_link
);
908 /* If the dnode was safe to move, the refcount cannot have changed. */
909 ASSERT(refcount
== refcount_count(&ndn
->dn_holds
));
910 ASSERT(dbufs
== ndn
->dn_dbufs_count
);
911 zrl_exit(&ndn
->dn_handle
->dnh_zrlock
); /* handle has moved */
912 mutex_exit(&os
->os_lock
);
914 return (KMEM_CBRC_YES
);
919 dnode_special_close(dnode_handle_t
*dnh
)
921 dnode_t
*dn
= dnh
->dnh_dnode
;
924 * Wait for final references to the dnode to clear. This can
925 * only happen if the arc is asyncronously evicting state that
926 * has a hold on this dnode while we are trying to evict this
929 while (refcount_count(&dn
->dn_holds
) > 0)
931 zrl_add(&dnh
->dnh_zrlock
);
932 dnode_destroy(dn
); /* implicit zrl_remove() */
933 zrl_destroy(&dnh
->dnh_zrlock
);
934 dnh
->dnh_dnode
= NULL
;
938 dnode_special_open(objset_t
*os
, dnode_phys_t
*dnp
, uint64_t object
,
941 dnode_t
*dn
= dnode_create(os
, dnp
, NULL
, object
, dnh
);
943 zrl_init(&dnh
->dnh_zrlock
);
949 dnode_buf_pageout(dmu_buf_t
*db
, void *arg
)
951 dnode_children_t
*children_dnodes
= arg
;
953 int epb
= db
->db_size
>> DNODE_SHIFT
;
955 ASSERT(epb
== children_dnodes
->dnc_count
);
957 for (i
= 0; i
< epb
; i
++) {
958 dnode_handle_t
*dnh
= &children_dnodes
->dnc_children
[i
];
962 * The dnode handle lock guards against the dnode moving to
963 * another valid address, so there is no need here to guard
964 * against changes to or from NULL.
966 if (dnh
->dnh_dnode
== NULL
) {
967 zrl_destroy(&dnh
->dnh_zrlock
);
971 zrl_add(&dnh
->dnh_zrlock
);
974 * If there are holds on this dnode, then there should
975 * be holds on the dnode's containing dbuf as well; thus
976 * it wouldn't be eligible for eviction and this function
977 * would not have been called.
979 ASSERT(refcount_is_zero(&dn
->dn_holds
));
980 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
982 dnode_destroy(dn
); /* implicit zrl_remove() */
983 zrl_destroy(&dnh
->dnh_zrlock
);
984 dnh
->dnh_dnode
= NULL
;
986 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
987 (epb
- 1) * sizeof (dnode_handle_t
));
992 * EINVAL - invalid object number.
994 * succeeds even for free dnodes.
997 dnode_hold_impl(objset_t
*os
, uint64_t object
, int flag
,
998 void *tag
, dnode_t
**dnp
)
1001 int drop_struct_lock
= FALSE
;
1006 dnode_children_t
*children_dnodes
;
1007 dnode_handle_t
*dnh
;
1010 * If you are holding the spa config lock as writer, you shouldn't
1011 * be asking the DMU to do *anything* unless it's the root pool
1012 * which may require us to read from the root filesystem while
1013 * holding some (not all) of the locks as writer.
1015 ASSERT(spa_config_held(os
->os_spa
, SCL_ALL
, RW_WRITER
) == 0 ||
1016 (spa_is_root(os
->os_spa
) &&
1017 spa_config_held(os
->os_spa
, SCL_STATE
, RW_WRITER
)));
1019 if (object
== DMU_USERUSED_OBJECT
|| object
== DMU_GROUPUSED_OBJECT
) {
1020 dn
= (object
== DMU_USERUSED_OBJECT
) ?
1021 DMU_USERUSED_DNODE(os
) : DMU_GROUPUSED_DNODE(os
);
1023 return (SET_ERROR(ENOENT
));
1025 if ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
)
1026 return (SET_ERROR(ENOENT
));
1027 if ((flag
& DNODE_MUST_BE_FREE
) && type
!= DMU_OT_NONE
)
1028 return (SET_ERROR(EEXIST
));
1030 (void) refcount_add(&dn
->dn_holds
, tag
);
1035 if (object
== 0 || object
>= DN_MAX_OBJECT
)
1036 return (SET_ERROR(EINVAL
));
1038 mdn
= DMU_META_DNODE(os
);
1039 ASSERT(mdn
->dn_object
== DMU_META_DNODE_OBJECT
);
1043 if (!RW_WRITE_HELD(&mdn
->dn_struct_rwlock
)) {
1044 rw_enter(&mdn
->dn_struct_rwlock
, RW_READER
);
1045 drop_struct_lock
= TRUE
;
1048 blk
= dbuf_whichblock(mdn
, object
* sizeof (dnode_phys_t
));
1050 db
= dbuf_hold(mdn
, blk
, FTAG
);
1051 if (drop_struct_lock
)
1052 rw_exit(&mdn
->dn_struct_rwlock
);
1054 return (SET_ERROR(EIO
));
1055 err
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
);
1057 dbuf_rele(db
, FTAG
);
1061 ASSERT3U(db
->db
.db_size
, >=, 1<<DNODE_SHIFT
);
1062 epb
= db
->db
.db_size
>> DNODE_SHIFT
;
1064 idx
= object
& (epb
-1);
1066 ASSERT(DB_DNODE(db
)->dn_type
== DMU_OT_DNODE
);
1067 children_dnodes
= dmu_buf_get_user(&db
->db
);
1068 if (children_dnodes
== NULL
) {
1070 dnode_children_t
*winner
;
1071 children_dnodes
= kmem_alloc(sizeof (dnode_children_t
) +
1072 (epb
- 1) * sizeof (dnode_handle_t
),
1073 KM_PUSHPAGE
| KM_NODEBUG
);
1074 children_dnodes
->dnc_count
= epb
;
1075 dnh
= &children_dnodes
->dnc_children
[0];
1076 for (i
= 0; i
< epb
; i
++) {
1077 zrl_init(&dnh
[i
].dnh_zrlock
);
1078 dnh
[i
].dnh_dnode
= NULL
;
1080 if ((winner
= dmu_buf_set_user(&db
->db
, children_dnodes
, NULL
,
1081 dnode_buf_pageout
))) {
1082 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
1083 (epb
- 1) * sizeof (dnode_handle_t
));
1084 children_dnodes
= winner
;
1087 ASSERT(children_dnodes
->dnc_count
== epb
);
1089 dnh
= &children_dnodes
->dnc_children
[idx
];
1090 zrl_add(&dnh
->dnh_zrlock
);
1091 if ((dn
= dnh
->dnh_dnode
) == NULL
) {
1092 dnode_phys_t
*phys
= (dnode_phys_t
*)db
->db
.db_data
+idx
;
1095 dn
= dnode_create(os
, phys
, db
, object
, dnh
);
1096 winner
= atomic_cas_ptr(&dnh
->dnh_dnode
, NULL
, dn
);
1097 if (winner
!= NULL
) {
1098 zrl_add(&dnh
->dnh_zrlock
);
1099 dnode_destroy(dn
); /* implicit zrl_remove() */
1104 mutex_enter(&dn
->dn_mtx
);
1106 if (dn
->dn_free_txg
||
1107 ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
) ||
1108 ((flag
& DNODE_MUST_BE_FREE
) &&
1109 (type
!= DMU_OT_NONE
|| !refcount_is_zero(&dn
->dn_holds
)))) {
1110 mutex_exit(&dn
->dn_mtx
);
1111 zrl_remove(&dnh
->dnh_zrlock
);
1112 dbuf_rele(db
, FTAG
);
1113 return (type
== DMU_OT_NONE
? ENOENT
: EEXIST
);
1115 mutex_exit(&dn
->dn_mtx
);
1117 if (refcount_add(&dn
->dn_holds
, tag
) == 1)
1118 dbuf_add_ref(db
, dnh
);
1119 /* Now we can rely on the hold to prevent the dnode from moving. */
1120 zrl_remove(&dnh
->dnh_zrlock
);
1123 ASSERT3P(dn
->dn_dbuf
, ==, db
);
1124 ASSERT3U(dn
->dn_object
, ==, object
);
1125 dbuf_rele(db
, FTAG
);
1132 * Return held dnode if the object is allocated, NULL if not.
1135 dnode_hold(objset_t
*os
, uint64_t object
, void *tag
, dnode_t
**dnp
)
1137 return (dnode_hold_impl(os
, object
, DNODE_MUST_BE_ALLOCATED
, tag
, dnp
));
1141 * Can only add a reference if there is already at least one
1142 * reference on the dnode. Returns FALSE if unable to add a
1146 dnode_add_ref(dnode_t
*dn
, void *tag
)
1148 mutex_enter(&dn
->dn_mtx
);
1149 if (refcount_is_zero(&dn
->dn_holds
)) {
1150 mutex_exit(&dn
->dn_mtx
);
1153 VERIFY(1 < refcount_add(&dn
->dn_holds
, tag
));
1154 mutex_exit(&dn
->dn_mtx
);
1159 dnode_rele(dnode_t
*dn
, void *tag
)
1162 /* Get while the hold prevents the dnode from moving. */
1163 dmu_buf_impl_t
*db
= dn
->dn_dbuf
;
1164 dnode_handle_t
*dnh
= dn
->dn_handle
;
1166 mutex_enter(&dn
->dn_mtx
);
1167 refs
= refcount_remove(&dn
->dn_holds
, tag
);
1168 mutex_exit(&dn
->dn_mtx
);
1171 * It's unsafe to release the last hold on a dnode by dnode_rele() or
1172 * indirectly by dbuf_rele() while relying on the dnode handle to
1173 * prevent the dnode from moving, since releasing the last hold could
1174 * result in the dnode's parent dbuf evicting its dnode handles. For
1175 * that reason anyone calling dnode_rele() or dbuf_rele() without some
1176 * other direct or indirect hold on the dnode must first drop the dnode
1179 ASSERT(refs
> 0 || dnh
->dnh_zrlock
.zr_owner
!= curthread
);
1181 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
1182 if (refs
== 0 && db
!= NULL
) {
1184 * Another thread could add a hold to the dnode handle in
1185 * dnode_hold_impl() while holding the parent dbuf. Since the
1186 * hold on the parent dbuf prevents the handle from being
1187 * destroyed, the hold on the handle is OK. We can't yet assert
1188 * that the handle has zero references, but that will be
1189 * asserted anyway when the handle gets destroyed.
1196 dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
)
1198 objset_t
*os
= dn
->dn_objset
;
1199 uint64_t txg
= tx
->tx_txg
;
1201 if (DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
1202 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1209 mutex_enter(&dn
->dn_mtx
);
1210 ASSERT(dn
->dn_phys
->dn_type
|| dn
->dn_allocated_txg
);
1211 ASSERT(dn
->dn_free_txg
== 0 || dn
->dn_free_txg
>= txg
);
1212 mutex_exit(&dn
->dn_mtx
);
1216 * Determine old uid/gid when necessary
1218 dmu_objset_userquota_get_ids(dn
, B_TRUE
, tx
);
1220 mutex_enter(&os
->os_lock
);
1223 * If we are already marked dirty, we're done.
1225 if (list_link_active(&dn
->dn_dirty_link
[txg
& TXG_MASK
])) {
1226 mutex_exit(&os
->os_lock
);
1230 ASSERT(!refcount_is_zero(&dn
->dn_holds
) || list_head(&dn
->dn_dbufs
));
1231 ASSERT(dn
->dn_datablksz
!= 0);
1232 ASSERT0(dn
->dn_next_bonuslen
[txg
&TXG_MASK
]);
1233 ASSERT0(dn
->dn_next_blksz
[txg
&TXG_MASK
]);
1234 ASSERT0(dn
->dn_next_bonustype
[txg
&TXG_MASK
]);
1236 dprintf_ds(os
->os_dsl_dataset
, "obj=%llu txg=%llu\n",
1237 dn
->dn_object
, txg
);
1239 if (dn
->dn_free_txg
> 0 && dn
->dn_free_txg
<= txg
) {
1240 list_insert_tail(&os
->os_free_dnodes
[txg
&TXG_MASK
], dn
);
1242 list_insert_tail(&os
->os_dirty_dnodes
[txg
&TXG_MASK
], dn
);
1245 mutex_exit(&os
->os_lock
);
1248 * The dnode maintains a hold on its containing dbuf as
1249 * long as there are holds on it. Each instantiated child
1250 * dbuf maintains a hold on the dnode. When the last child
1251 * drops its hold, the dnode will drop its hold on the
1252 * containing dbuf. We add a "dirty hold" here so that the
1253 * dnode will hang around after we finish processing its
1256 VERIFY(dnode_add_ref(dn
, (void *)(uintptr_t)tx
->tx_txg
));
1258 (void) dbuf_dirty(dn
->dn_dbuf
, tx
);
1260 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1264 dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
)
1266 int txgoff
= tx
->tx_txg
& TXG_MASK
;
1268 dprintf("dn=%p txg=%llu\n", dn
, tx
->tx_txg
);
1270 /* we should be the only holder... hopefully */
1271 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */
1273 mutex_enter(&dn
->dn_mtx
);
1274 if (dn
->dn_type
== DMU_OT_NONE
|| dn
->dn_free_txg
) {
1275 mutex_exit(&dn
->dn_mtx
);
1278 dn
->dn_free_txg
= tx
->tx_txg
;
1279 mutex_exit(&dn
->dn_mtx
);
1282 * If the dnode is already dirty, it needs to be moved from
1283 * the dirty list to the free list.
1285 mutex_enter(&dn
->dn_objset
->os_lock
);
1286 if (list_link_active(&dn
->dn_dirty_link
[txgoff
])) {
1287 list_remove(&dn
->dn_objset
->os_dirty_dnodes
[txgoff
], dn
);
1288 list_insert_tail(&dn
->dn_objset
->os_free_dnodes
[txgoff
], dn
);
1289 mutex_exit(&dn
->dn_objset
->os_lock
);
1291 mutex_exit(&dn
->dn_objset
->os_lock
);
1292 dnode_setdirty(dn
, tx
);
1297 * Try to change the block size for the indicated dnode. This can only
1298 * succeed if there are no blocks allocated or dirty beyond first block
1301 dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
)
1303 dmu_buf_impl_t
*db
, *db_next
;
1307 size
= SPA_MINBLOCKSIZE
;
1308 if (size
> SPA_MAXBLOCKSIZE
)
1309 size
= SPA_MAXBLOCKSIZE
;
1311 size
= P2ROUNDUP(size
, SPA_MINBLOCKSIZE
);
1313 if (ibs
== dn
->dn_indblkshift
)
1316 if (size
>> SPA_MINBLOCKSHIFT
== dn
->dn_datablkszsec
&& ibs
== 0)
1319 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1321 /* Check for any allocated blocks beyond the first */
1322 if (dn
->dn_maxblkid
!= 0)
1325 mutex_enter(&dn
->dn_dbufs_mtx
);
1326 for (db
= list_head(&dn
->dn_dbufs
); db
; db
= db_next
) {
1327 db_next
= list_next(&dn
->dn_dbufs
, db
);
1329 if (db
->db_blkid
!= 0 && db
->db_blkid
!= DMU_BONUS_BLKID
&&
1330 db
->db_blkid
!= DMU_SPILL_BLKID
) {
1331 mutex_exit(&dn
->dn_dbufs_mtx
);
1335 mutex_exit(&dn
->dn_dbufs_mtx
);
1337 if (ibs
&& dn
->dn_nlevels
!= 1)
1340 /* resize the old block */
1341 err
= dbuf_hold_impl(dn
, 0, 0, TRUE
, FTAG
, &db
);
1343 dbuf_new_size(db
, size
, tx
);
1344 else if (err
!= ENOENT
)
1347 dnode_setdblksz(dn
, size
);
1348 dnode_setdirty(dn
, tx
);
1349 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = size
;
1351 dn
->dn_indblkshift
= ibs
;
1352 dn
->dn_next_indblkshift
[tx
->tx_txg
&TXG_MASK
] = ibs
;
1354 /* rele after we have fixed the blocksize in the dnode */
1356 dbuf_rele(db
, FTAG
);
1358 rw_exit(&dn
->dn_struct_rwlock
);
1362 rw_exit(&dn
->dn_struct_rwlock
);
1363 return (SET_ERROR(ENOTSUP
));
1366 /* read-holding callers must not rely on the lock being continuously held */
1368 dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t have_read
)
1370 uint64_t txgoff
= tx
->tx_txg
& TXG_MASK
;
1371 int epbs
, new_nlevels
;
1374 ASSERT(blkid
!= DMU_BONUS_BLKID
);
1377 RW_READ_HELD(&dn
->dn_struct_rwlock
) :
1378 RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
1381 * if we have a read-lock, check to see if we need to do any work
1382 * before upgrading to a write-lock.
1385 if (blkid
<= dn
->dn_maxblkid
)
1388 if (!rw_tryupgrade(&dn
->dn_struct_rwlock
)) {
1389 rw_exit(&dn
->dn_struct_rwlock
);
1390 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1394 if (blkid
<= dn
->dn_maxblkid
)
1397 dn
->dn_maxblkid
= blkid
;
1400 * Compute the number of levels necessary to support the new maxblkid.
1403 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1404 for (sz
= dn
->dn_nblkptr
;
1405 sz
<= blkid
&& sz
>= dn
->dn_nblkptr
; sz
<<= epbs
)
1408 if (new_nlevels
> dn
->dn_nlevels
) {
1409 int old_nlevels
= dn
->dn_nlevels
;
1412 dbuf_dirty_record_t
*new, *dr
, *dr_next
;
1414 dn
->dn_nlevels
= new_nlevels
;
1416 ASSERT3U(new_nlevels
, >, dn
->dn_next_nlevels
[txgoff
]);
1417 dn
->dn_next_nlevels
[txgoff
] = new_nlevels
;
1419 /* dirty the left indirects */
1420 db
= dbuf_hold_level(dn
, old_nlevels
, 0, FTAG
);
1422 new = dbuf_dirty(db
, tx
);
1423 dbuf_rele(db
, FTAG
);
1425 /* transfer the dirty records to the new indirect */
1426 mutex_enter(&dn
->dn_mtx
);
1427 mutex_enter(&new->dt
.di
.dr_mtx
);
1428 list
= &dn
->dn_dirty_records
[txgoff
];
1429 for (dr
= list_head(list
); dr
; dr
= dr_next
) {
1430 dr_next
= list_next(&dn
->dn_dirty_records
[txgoff
], dr
);
1431 if (dr
->dr_dbuf
->db_level
!= new_nlevels
-1 &&
1432 dr
->dr_dbuf
->db_blkid
!= DMU_BONUS_BLKID
&&
1433 dr
->dr_dbuf
->db_blkid
!= DMU_SPILL_BLKID
) {
1434 ASSERT(dr
->dr_dbuf
->db_level
== old_nlevels
-1);
1435 list_remove(&dn
->dn_dirty_records
[txgoff
], dr
);
1436 list_insert_tail(&new->dt
.di
.dr_children
, dr
);
1437 dr
->dr_parent
= new;
1440 mutex_exit(&new->dt
.di
.dr_mtx
);
1441 mutex_exit(&dn
->dn_mtx
);
1446 rw_downgrade(&dn
->dn_struct_rwlock
);
1450 dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
)
1453 uint64_t blkoff
, blkid
, nblks
;
1454 int blksz
, blkshift
, head
, tail
;
1458 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1459 blksz
= dn
->dn_datablksz
;
1460 blkshift
= dn
->dn_datablkshift
;
1461 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1463 if (len
== DMU_OBJECT_END
) {
1464 len
= UINT64_MAX
- off
;
1469 * First, block align the region to free:
1472 head
= P2NPHASE(off
, blksz
);
1473 blkoff
= P2PHASE(off
, blksz
);
1474 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1477 ASSERT(dn
->dn_maxblkid
== 0);
1478 if (off
== 0 && len
>= blksz
) {
1480 * Freeing the whole block; fast-track this request.
1481 * Note that we won't dirty any indirect blocks,
1482 * which is fine because we will be freeing the entire
1483 * file and thus all indirect blocks will be freed
1484 * by free_children().
1489 } else if (off
>= blksz
) {
1490 /* Freeing past end-of-data */
1493 /* Freeing part of the block. */
1495 ASSERT3U(head
, >, 0);
1499 /* zero out any partial block data at the start of the range */
1501 ASSERT3U(blkoff
+ head
, ==, blksz
);
1504 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
), TRUE
,
1508 /* don't dirty if it isn't on disk and isn't dirty */
1509 if (db
->db_last_dirty
||
1510 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1511 rw_exit(&dn
->dn_struct_rwlock
);
1512 dmu_buf_will_dirty(&db
->db
, tx
);
1513 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1514 data
= db
->db
.db_data
;
1515 bzero(data
+ blkoff
, head
);
1517 dbuf_rele(db
, FTAG
);
1523 /* If the range was less than one block, we're done */
1527 /* If the remaining range is past end of file, we're done */
1528 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1531 ASSERT(ISP2(blksz
));
1535 tail
= P2PHASE(len
, blksz
);
1537 ASSERT0(P2PHASE(off
, blksz
));
1538 /* zero out any partial block data at the end of the range */
1542 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
+len
),
1543 TRUE
, FTAG
, &db
) == 0) {
1544 /* don't dirty if not on disk and not dirty */
1545 if (db
->db_last_dirty
||
1546 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1547 rw_exit(&dn
->dn_struct_rwlock
);
1548 dmu_buf_will_dirty(&db
->db
, tx
);
1549 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1550 bzero(db
->db
.db_data
, tail
);
1552 dbuf_rele(db
, FTAG
);
1557 /* If the range did not include a full block, we are done */
1561 ASSERT(IS_P2ALIGNED(off
, blksz
));
1562 ASSERT(trunc
|| IS_P2ALIGNED(len
, blksz
));
1563 blkid
= off
>> blkshift
;
1564 nblks
= len
>> blkshift
;
1569 * Dirty the first and last indirect blocks, as they (and/or their
1570 * parents) will need to be written out if they were only
1571 * partially freed. Interior indirect blocks will be themselves freed,
1572 * by free_children(), so they need not be dirtied. Note that these
1573 * interior blocks have already been prefetched by dmu_tx_hold_free().
1575 if (dn
->dn_nlevels
> 1) {
1576 uint64_t first
, last
;
1578 first
= blkid
>> epbs
;
1579 if ((db
= dbuf_hold_level(dn
, 1, first
, FTAG
))) {
1580 dmu_buf_will_dirty(&db
->db
, tx
);
1581 dbuf_rele(db
, FTAG
);
1584 last
= dn
->dn_maxblkid
>> epbs
;
1586 last
= (blkid
+ nblks
- 1) >> epbs
;
1587 if (last
> first
&& (db
= dbuf_hold_level(dn
, 1, last
, FTAG
))) {
1588 dmu_buf_will_dirty(&db
->db
, tx
);
1589 dbuf_rele(db
, FTAG
);
1595 * Add this range to the dnode range list.
1596 * We will finish up this free operation in the syncing phase.
1598 mutex_enter(&dn
->dn_mtx
);
1600 int txgoff
= tx
->tx_txg
& TXG_MASK
;
1601 if (dn
->dn_free_ranges
[txgoff
] == NULL
) {
1602 dn
->dn_free_ranges
[txgoff
] =
1603 range_tree_create(NULL
, NULL
, &dn
->dn_mtx
);
1605 range_tree_clear(dn
->dn_free_ranges
[txgoff
], blkid
, nblks
);
1606 range_tree_add(dn
->dn_free_ranges
[txgoff
], blkid
, nblks
);
1608 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1609 blkid
, nblks
, tx
->tx_txg
);
1610 mutex_exit(&dn
->dn_mtx
);
1612 dbuf_free_range(dn
, blkid
, blkid
+ nblks
- 1, tx
);
1613 dnode_setdirty(dn
, tx
);
1616 rw_exit(&dn
->dn_struct_rwlock
);
1620 dnode_spill_freed(dnode_t
*dn
)
1624 mutex_enter(&dn
->dn_mtx
);
1625 for (i
= 0; i
< TXG_SIZE
; i
++) {
1626 if (dn
->dn_rm_spillblk
[i
] == DN_KILL_SPILLBLK
)
1629 mutex_exit(&dn
->dn_mtx
);
1630 return (i
< TXG_SIZE
);
1633 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1635 dnode_block_freed(dnode_t
*dn
, uint64_t blkid
)
1637 void *dp
= spa_get_dsl(dn
->dn_objset
->os_spa
);
1640 if (blkid
== DMU_BONUS_BLKID
)
1644 * If we're in the process of opening the pool, dp will not be
1645 * set yet, but there shouldn't be anything dirty.
1650 if (dn
->dn_free_txg
)
1653 if (blkid
== DMU_SPILL_BLKID
)
1654 return (dnode_spill_freed(dn
));
1656 mutex_enter(&dn
->dn_mtx
);
1657 for (i
= 0; i
< TXG_SIZE
; i
++) {
1658 if (dn
->dn_free_ranges
[i
] != NULL
&&
1659 range_tree_contains(dn
->dn_free_ranges
[i
], blkid
, 1))
1662 mutex_exit(&dn
->dn_mtx
);
1663 return (i
< TXG_SIZE
);
1666 /* call from syncing context when we actually write/free space for this dnode */
1668 dnode_diduse_space(dnode_t
*dn
, int64_t delta
)
1671 dprintf_dnode(dn
, "dn=%p dnp=%p used=%llu delta=%lld\n",
1673 (u_longlong_t
)dn
->dn_phys
->dn_used
,
1676 mutex_enter(&dn
->dn_mtx
);
1677 space
= DN_USED_BYTES(dn
->dn_phys
);
1679 ASSERT3U(space
+ delta
, >=, space
); /* no overflow */
1681 ASSERT3U(space
, >=, -delta
); /* no underflow */
1684 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_DNODE_BYTES
) {
1685 ASSERT((dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) == 0);
1686 ASSERT0(P2PHASE(space
, 1<<DEV_BSHIFT
));
1687 dn
->dn_phys
->dn_used
= space
>> DEV_BSHIFT
;
1689 dn
->dn_phys
->dn_used
= space
;
1690 dn
->dn_phys
->dn_flags
|= DNODE_FLAG_USED_BYTES
;
1692 mutex_exit(&dn
->dn_mtx
);
1696 * Call when we think we're going to write/free space in open context to track
1697 * the amount of memory in use by the currently open txg.
1700 dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
)
1702 objset_t
*os
= dn
->dn_objset
;
1703 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1704 int64_t aspace
= spa_get_asize(os
->os_spa
, space
);
1707 dsl_dir_willuse_space(ds
->ds_dir
, aspace
, tx
);
1708 dsl_pool_dirty_space(dmu_tx_pool(tx
), space
, tx
);
1711 dmu_tx_willuse_space(tx
, aspace
);
1715 * Scans a block at the indicated "level" looking for a hole or data,
1716 * depending on 'flags'.
1718 * If level > 0, then we are scanning an indirect block looking at its
1719 * pointers. If level == 0, then we are looking at a block of dnodes.
1721 * If we don't find what we are looking for in the block, we return ESRCH.
1722 * Otherwise, return with *offset pointing to the beginning (if searching
1723 * forwards) or end (if searching backwards) of the range covered by the
1724 * block pointer we matched on (or dnode).
1726 * The basic search algorithm used below by dnode_next_offset() is to
1727 * use this function to search up the block tree (widen the search) until
1728 * we find something (i.e., we don't return ESRCH) and then search back
1729 * down the tree (narrow the search) until we reach our original search
1733 dnode_next_offset_level(dnode_t
*dn
, int flags
, uint64_t *offset
,
1734 int lvl
, uint64_t blkfill
, uint64_t txg
)
1736 dmu_buf_impl_t
*db
= NULL
;
1738 uint64_t epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1739 uint64_t epb
= 1ULL << epbs
;
1740 uint64_t minfill
, maxfill
;
1742 int i
, inc
, error
, span
;
1744 dprintf("probing object %llu offset %llx level %d of %u\n",
1745 dn
->dn_object
, *offset
, lvl
, dn
->dn_phys
->dn_nlevels
);
1747 hole
= ((flags
& DNODE_FIND_HOLE
) != 0);
1748 inc
= (flags
& DNODE_FIND_BACKWARDS
) ? -1 : 1;
1749 ASSERT(txg
== 0 || !hole
);
1751 if (lvl
== dn
->dn_phys
->dn_nlevels
) {
1753 epb
= dn
->dn_phys
->dn_nblkptr
;
1754 data
= dn
->dn_phys
->dn_blkptr
;
1756 uint64_t blkid
= dbuf_whichblock(dn
, *offset
) >> (epbs
* lvl
);
1757 error
= dbuf_hold_impl(dn
, lvl
, blkid
, TRUE
, FTAG
, &db
);
1759 if (error
!= ENOENT
)
1764 * This can only happen when we are searching up
1765 * the block tree for data. We don't really need to
1766 * adjust the offset, as we will just end up looking
1767 * at the pointer to this block in its parent, and its
1768 * going to be unallocated, so we will skip over it.
1770 return (SET_ERROR(ESRCH
));
1772 error
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
| DB_RF_HAVESTRUCT
);
1774 dbuf_rele(db
, FTAG
);
1777 data
= db
->db
.db_data
;
1781 if (db
!= NULL
&& txg
!= 0 && (db
->db_blkptr
== NULL
||
1782 db
->db_blkptr
->blk_birth
<= txg
||
1783 BP_IS_HOLE(db
->db_blkptr
))) {
1785 * This can only happen when we are searching up the tree
1786 * and these conditions mean that we need to keep climbing.
1788 error
= SET_ERROR(ESRCH
);
1789 } else if (lvl
== 0) {
1790 dnode_phys_t
*dnp
= data
;
1792 ASSERT(dn
->dn_type
== DMU_OT_DNODE
);
1794 for (i
= (*offset
>> span
) & (blkfill
- 1);
1795 i
>= 0 && i
< blkfill
; i
+= inc
) {
1796 if ((dnp
[i
].dn_type
== DMU_OT_NONE
) == hole
)
1798 *offset
+= (1ULL << span
) * inc
;
1800 if (i
< 0 || i
== blkfill
)
1801 error
= SET_ERROR(ESRCH
);
1803 blkptr_t
*bp
= data
;
1804 uint64_t start
= *offset
;
1805 span
= (lvl
- 1) * epbs
+ dn
->dn_datablkshift
;
1807 maxfill
= blkfill
<< ((lvl
- 1) * epbs
);
1814 *offset
= *offset
>> span
;
1815 for (i
= BF64_GET(*offset
, 0, epbs
);
1816 i
>= 0 && i
< epb
; i
+= inc
) {
1817 if (BP_GET_FILL(&bp
[i
]) >= minfill
&&
1818 BP_GET_FILL(&bp
[i
]) <= maxfill
&&
1819 (hole
|| bp
[i
].blk_birth
> txg
))
1821 if (inc
> 0 || *offset
> 0)
1824 *offset
= *offset
<< span
;
1826 /* traversing backwards; position offset at the end */
1827 ASSERT3U(*offset
, <=, start
);
1828 *offset
= MIN(*offset
+ (1ULL << span
) - 1, start
);
1829 } else if (*offset
< start
) {
1832 if (i
< 0 || i
>= epb
)
1833 error
= SET_ERROR(ESRCH
);
1837 dbuf_rele(db
, FTAG
);
1843 * Find the next hole, data, or sparse region at or after *offset.
1844 * The value 'blkfill' tells us how many items we expect to find
1845 * in an L0 data block; this value is 1 for normal objects,
1846 * DNODES_PER_BLOCK for the meta dnode, and some fraction of
1847 * DNODES_PER_BLOCK when searching for sparse regions thereof.
1851 * dnode_next_offset(dn, flags, offset, 1, 1, 0);
1852 * Finds the next/previous hole/data in a file.
1853 * Used in dmu_offset_next().
1855 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg);
1856 * Finds the next free/allocated dnode an objset's meta-dnode.
1857 * Only finds objects that have new contents since txg (ie.
1858 * bonus buffer changes and content removal are ignored).
1859 * Used in dmu_object_next().
1861 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0);
1862 * Finds the next L2 meta-dnode bp that's at most 1/4 full.
1863 * Used in dmu_object_alloc().
1866 dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *offset
,
1867 int minlvl
, uint64_t blkfill
, uint64_t txg
)
1869 uint64_t initial_offset
= *offset
;
1873 if (!(flags
& DNODE_FIND_HAVELOCK
))
1874 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1876 if (dn
->dn_phys
->dn_nlevels
== 0) {
1877 error
= SET_ERROR(ESRCH
);
1881 if (dn
->dn_datablkshift
== 0) {
1882 if (*offset
< dn
->dn_datablksz
) {
1883 if (flags
& DNODE_FIND_HOLE
)
1884 *offset
= dn
->dn_datablksz
;
1886 error
= SET_ERROR(ESRCH
);
1891 maxlvl
= dn
->dn_phys
->dn_nlevels
;
1893 for (lvl
= minlvl
; lvl
<= maxlvl
; lvl
++) {
1894 error
= dnode_next_offset_level(dn
,
1895 flags
, offset
, lvl
, blkfill
, txg
);
1900 while (error
== 0 && --lvl
>= minlvl
) {
1901 error
= dnode_next_offset_level(dn
,
1902 flags
, offset
, lvl
, blkfill
, txg
);
1905 if (error
== 0 && (flags
& DNODE_FIND_BACKWARDS
?
1906 initial_offset
< *offset
: initial_offset
> *offset
))
1907 error
= SET_ERROR(ESRCH
);
1909 if (!(flags
& DNODE_FIND_HAVELOCK
))
1910 rw_exit(&dn
->dn_struct_rwlock
);