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.
25 #include <sys/zfs_context.h>
27 #include <sys/dnode.h>
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dmu_objset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_dataset.h>
36 #include <sys/dmu_zfetch.h>
38 static int free_range_compar(const void *node1
, const void *node2
);
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 static dnode_phys_t dnode_phys_zero
;
57 int zfs_default_bs
= SPA_MINBLOCKSHIFT
;
58 int zfs_default_ibs
= DN_MAX_INDBLKSHIFT
;
60 static kmem_cbrc_t
dnode_move(void *, void *, size_t, void *);
64 dnode_cons(void *arg
, void *unused
, int kmflag
)
69 rw_init(&dn
->dn_struct_rwlock
, NULL
, RW_DEFAULT
, NULL
);
70 mutex_init(&dn
->dn_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
71 mutex_init(&dn
->dn_dbufs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
72 cv_init(&dn
->dn_notxholds
, NULL
, CV_DEFAULT
, NULL
);
74 refcount_create(&dn
->dn_holds
);
75 refcount_create(&dn
->dn_tx_holds
);
76 list_link_init(&dn
->dn_link
);
78 bzero(&dn
->dn_next_nblkptr
[0], sizeof (dn
->dn_next_nblkptr
));
79 bzero(&dn
->dn_next_nlevels
[0], sizeof (dn
->dn_next_nlevels
));
80 bzero(&dn
->dn_next_indblkshift
[0], sizeof (dn
->dn_next_indblkshift
));
81 bzero(&dn
->dn_next_bonustype
[0], sizeof (dn
->dn_next_bonustype
));
82 bzero(&dn
->dn_rm_spillblk
[0], sizeof (dn
->dn_rm_spillblk
));
83 bzero(&dn
->dn_next_bonuslen
[0], sizeof (dn
->dn_next_bonuslen
));
84 bzero(&dn
->dn_next_blksz
[0], sizeof (dn
->dn_next_blksz
));
86 for (i
= 0; i
< TXG_SIZE
; i
++) {
87 list_link_init(&dn
->dn_dirty_link
[i
]);
88 avl_create(&dn
->dn_ranges
[i
], free_range_compar
,
89 sizeof (free_range_t
),
90 offsetof(struct free_range
, fr_node
));
91 list_create(&dn
->dn_dirty_records
[i
],
92 sizeof (dbuf_dirty_record_t
),
93 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
96 dn
->dn_allocated_txg
= 0;
98 dn
->dn_assigned_txg
= 0;
100 dn
->dn_dirtyctx_firstset
= NULL
;
102 dn
->dn_have_spill
= B_FALSE
;
112 dn
->dn_dbufs_count
= 0;
113 list_create(&dn
->dn_dbufs
, sizeof (dmu_buf_impl_t
),
114 offsetof(dmu_buf_impl_t
, db_link
));
122 dnode_dest(void *arg
, void *unused
)
127 rw_destroy(&dn
->dn_struct_rwlock
);
128 mutex_destroy(&dn
->dn_mtx
);
129 mutex_destroy(&dn
->dn_dbufs_mtx
);
130 cv_destroy(&dn
->dn_notxholds
);
131 refcount_destroy(&dn
->dn_holds
);
132 refcount_destroy(&dn
->dn_tx_holds
);
133 ASSERT(!list_link_active(&dn
->dn_link
));
135 for (i
= 0; i
< TXG_SIZE
; i
++) {
136 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
137 avl_destroy(&dn
->dn_ranges
[i
]);
138 list_destroy(&dn
->dn_dirty_records
[i
]);
139 ASSERT3U(dn
->dn_next_nblkptr
[i
], ==, 0);
140 ASSERT3U(dn
->dn_next_nlevels
[i
], ==, 0);
141 ASSERT3U(dn
->dn_next_indblkshift
[i
], ==, 0);
142 ASSERT3U(dn
->dn_next_bonustype
[i
], ==, 0);
143 ASSERT3U(dn
->dn_rm_spillblk
[i
], ==, 0);
144 ASSERT3U(dn
->dn_next_bonuslen
[i
], ==, 0);
145 ASSERT3U(dn
->dn_next_blksz
[i
], ==, 0);
148 ASSERT3U(dn
->dn_allocated_txg
, ==, 0);
149 ASSERT3U(dn
->dn_free_txg
, ==, 0);
150 ASSERT3U(dn
->dn_assigned_txg
, ==, 0);
151 ASSERT3U(dn
->dn_dirtyctx
, ==, 0);
152 ASSERT3P(dn
->dn_dirtyctx_firstset
, ==, NULL
);
153 ASSERT3P(dn
->dn_bonus
, ==, NULL
);
154 ASSERT(!dn
->dn_have_spill
);
155 ASSERT3P(dn
->dn_zio
, ==, NULL
);
156 ASSERT3U(dn
->dn_oldused
, ==, 0);
157 ASSERT3U(dn
->dn_oldflags
, ==, 0);
158 ASSERT3U(dn
->dn_olduid
, ==, 0);
159 ASSERT3U(dn
->dn_oldgid
, ==, 0);
160 ASSERT3U(dn
->dn_newuid
, ==, 0);
161 ASSERT3U(dn
->dn_newgid
, ==, 0);
162 ASSERT3U(dn
->dn_id_flags
, ==, 0);
164 ASSERT3U(dn
->dn_dbufs_count
, ==, 0);
165 list_destroy(&dn
->dn_dbufs
);
171 ASSERT(dnode_cache
== NULL
);
172 dnode_cache
= kmem_cache_create("dnode_t",
174 0, dnode_cons
, dnode_dest
, NULL
, NULL
, NULL
, 0);
175 kmem_cache_set_move(dnode_cache
, dnode_move
);
181 kmem_cache_destroy(dnode_cache
);
188 dnode_verify(dnode_t
*dn
)
190 int drop_struct_lock
= FALSE
;
193 ASSERT(dn
->dn_objset
);
194 ASSERT(dn
->dn_handle
->dnh_dnode
== dn
);
196 ASSERT(dn
->dn_phys
->dn_type
< DMU_OT_NUMTYPES
);
198 if (!(zfs_flags
& ZFS_DEBUG_DNODE_VERIFY
))
201 if (!RW_WRITE_HELD(&dn
->dn_struct_rwlock
)) {
202 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
203 drop_struct_lock
= TRUE
;
205 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
|| dn
->dn_allocated_txg
!= 0) {
207 ASSERT3U(dn
->dn_indblkshift
, <=, SPA_MAXBLOCKSHIFT
);
208 if (dn
->dn_datablkshift
) {
209 ASSERT3U(dn
->dn_datablkshift
, >=, SPA_MINBLOCKSHIFT
);
210 ASSERT3U(dn
->dn_datablkshift
, <=, SPA_MAXBLOCKSHIFT
);
211 ASSERT3U(1<<dn
->dn_datablkshift
, ==, dn
->dn_datablksz
);
213 ASSERT3U(dn
->dn_nlevels
, <=, 30);
214 ASSERT3U(dn
->dn_type
, <=, DMU_OT_NUMTYPES
);
215 ASSERT3U(dn
->dn_nblkptr
, >=, 1);
216 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
217 ASSERT3U(dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
218 ASSERT3U(dn
->dn_datablksz
, ==,
219 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
220 ASSERT3U(ISP2(dn
->dn_datablksz
), ==, dn
->dn_datablkshift
!= 0);
221 ASSERT3U((dn
->dn_nblkptr
- 1) * sizeof (blkptr_t
) +
222 dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
223 for (i
= 0; i
< TXG_SIZE
; i
++) {
224 ASSERT3U(dn
->dn_next_nlevels
[i
], <=, dn
->dn_nlevels
);
227 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
)
228 ASSERT3U(dn
->dn_phys
->dn_nlevels
, <=, dn
->dn_nlevels
);
229 ASSERT(DMU_OBJECT_IS_SPECIAL(dn
->dn_object
) || dn
->dn_dbuf
!= NULL
);
230 if (dn
->dn_dbuf
!= NULL
) {
231 ASSERT3P(dn
->dn_phys
, ==,
232 (dnode_phys_t
*)dn
->dn_dbuf
->db
.db_data
+
233 (dn
->dn_object
% (dn
->dn_dbuf
->db
.db_size
>> DNODE_SHIFT
)));
235 if (drop_struct_lock
)
236 rw_exit(&dn
->dn_struct_rwlock
);
241 dnode_byteswap(dnode_phys_t
*dnp
)
243 uint64_t *buf64
= (void*)&dnp
->dn_blkptr
;
246 if (dnp
->dn_type
== DMU_OT_NONE
) {
247 bzero(dnp
, sizeof (dnode_phys_t
));
251 dnp
->dn_datablkszsec
= BSWAP_16(dnp
->dn_datablkszsec
);
252 dnp
->dn_bonuslen
= BSWAP_16(dnp
->dn_bonuslen
);
253 dnp
->dn_maxblkid
= BSWAP_64(dnp
->dn_maxblkid
);
254 dnp
->dn_used
= BSWAP_64(dnp
->dn_used
);
257 * dn_nblkptr is only one byte, so it's OK to read it in either
258 * byte order. We can't read dn_bouslen.
260 ASSERT(dnp
->dn_indblkshift
<= SPA_MAXBLOCKSHIFT
);
261 ASSERT(dnp
->dn_nblkptr
<= DN_MAX_NBLKPTR
);
262 for (i
= 0; i
< dnp
->dn_nblkptr
* sizeof (blkptr_t
)/8; i
++)
263 buf64
[i
] = BSWAP_64(buf64
[i
]);
266 * OK to check dn_bonuslen for zero, because it won't matter if
267 * we have the wrong byte order. This is necessary because the
268 * dnode dnode is smaller than a regular dnode.
270 if (dnp
->dn_bonuslen
!= 0) {
272 * Note that the bonus length calculated here may be
273 * longer than the actual bonus buffer. This is because
274 * we always put the bonus buffer after the last block
275 * pointer (instead of packing it against the end of the
278 int off
= (dnp
->dn_nblkptr
-1) * sizeof (blkptr_t
);
279 size_t len
= DN_MAX_BONUSLEN
- off
;
280 ASSERT3U(dnp
->dn_bonustype
, <, DMU_OT_NUMTYPES
);
281 dmu_ot
[dnp
->dn_bonustype
].ot_byteswap(dnp
->dn_bonus
+ off
, len
);
284 /* Swap SPILL block if we have one */
285 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)
286 byteswap_uint64_array(&dnp
->dn_spill
, sizeof (blkptr_t
));
291 dnode_buf_byteswap(void *vbuf
, size_t size
)
293 dnode_phys_t
*buf
= vbuf
;
296 ASSERT3U(sizeof (dnode_phys_t
), ==, (1<<DNODE_SHIFT
));
297 ASSERT((size
& (sizeof (dnode_phys_t
)-1)) == 0);
299 size
>>= DNODE_SHIFT
;
300 for (i
= 0; i
< size
; i
++) {
307 free_range_compar(const void *node1
, const void *node2
)
309 const free_range_t
*rp1
= node1
;
310 const free_range_t
*rp2
= node2
;
312 if (rp1
->fr_blkid
< rp2
->fr_blkid
)
314 else if (rp1
->fr_blkid
> rp2
->fr_blkid
)
320 dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
)
322 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
324 dnode_setdirty(dn
, tx
);
325 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
326 ASSERT3U(newsize
, <=, DN_MAX_BONUSLEN
-
327 (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
));
328 dn
->dn_bonuslen
= newsize
;
330 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = DN_ZERO_BONUSLEN
;
332 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
333 rw_exit(&dn
->dn_struct_rwlock
);
337 dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t newtype
, dmu_tx_t
*tx
)
339 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
340 dnode_setdirty(dn
, tx
);
341 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
342 dn
->dn_bonustype
= newtype
;
343 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
344 rw_exit(&dn
->dn_struct_rwlock
);
348 dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
)
350 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
351 ASSERT(RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
352 dnode_setdirty(dn
, tx
);
353 dn
->dn_rm_spillblk
[tx
->tx_txg
&TXG_MASK
] = DN_KILL_SPILLBLK
;
354 dn
->dn_have_spill
= B_FALSE
;
358 dnode_setdblksz(dnode_t
*dn
, int size
)
360 ASSERT3U(P2PHASE(size
, SPA_MINBLOCKSIZE
), ==, 0);
361 ASSERT3U(size
, <=, SPA_MAXBLOCKSIZE
);
362 ASSERT3U(size
, >=, SPA_MINBLOCKSIZE
);
363 ASSERT3U(size
>> SPA_MINBLOCKSHIFT
, <,
364 1<<(sizeof (dn
->dn_phys
->dn_datablkszsec
) * 8));
365 dn
->dn_datablksz
= size
;
366 dn
->dn_datablkszsec
= size
>> SPA_MINBLOCKSHIFT
;
367 dn
->dn_datablkshift
= ISP2(size
) ? highbit(size
- 1) : 0;
371 dnode_create(objset_t
*os
, dnode_phys_t
*dnp
, dmu_buf_impl_t
*db
,
372 uint64_t object
, dnode_handle_t
*dnh
)
374 dnode_t
*dn
= kmem_cache_alloc(dnode_cache
, KM_SLEEP
);
376 ASSERT(!POINTER_IS_VALID(dn
->dn_objset
));
380 * Defer setting dn_objset until the dnode is ready to be a candidate
381 * for the dnode_move() callback.
383 dn
->dn_object
= object
;
388 if (dnp
->dn_datablkszsec
) {
389 dnode_setdblksz(dn
, dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
391 dn
->dn_datablksz
= 0;
392 dn
->dn_datablkszsec
= 0;
393 dn
->dn_datablkshift
= 0;
395 dn
->dn_indblkshift
= dnp
->dn_indblkshift
;
396 dn
->dn_nlevels
= dnp
->dn_nlevels
;
397 dn
->dn_type
= dnp
->dn_type
;
398 dn
->dn_nblkptr
= dnp
->dn_nblkptr
;
399 dn
->dn_checksum
= dnp
->dn_checksum
;
400 dn
->dn_compress
= dnp
->dn_compress
;
401 dn
->dn_bonustype
= dnp
->dn_bonustype
;
402 dn
->dn_bonuslen
= dnp
->dn_bonuslen
;
403 dn
->dn_maxblkid
= dnp
->dn_maxblkid
;
404 dn
->dn_have_spill
= ((dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) != 0);
407 dmu_zfetch_init(&dn
->dn_zfetch
, dn
);
409 ASSERT(dn
->dn_phys
->dn_type
< DMU_OT_NUMTYPES
);
411 mutex_enter(&os
->os_lock
);
412 list_insert_head(&os
->os_dnodes
, dn
);
415 * Everything else must be valid before assigning dn_objset makes the
416 * dnode eligible for dnode_move().
419 mutex_exit(&os
->os_lock
);
421 arc_space_consume(sizeof (dnode_t
), ARC_SPACE_OTHER
);
426 * Caller must be holding the dnode handle, which is released upon return.
429 dnode_destroy(dnode_t
*dn
)
431 objset_t
*os
= dn
->dn_objset
;
433 ASSERT((dn
->dn_id_flags
& DN_ID_NEW_EXIST
) == 0);
435 mutex_enter(&os
->os_lock
);
436 POINTER_INVALIDATE(&dn
->dn_objset
);
437 list_remove(&os
->os_dnodes
, dn
);
438 mutex_exit(&os
->os_lock
);
440 /* the dnode can no longer move, so we can release the handle */
441 zrl_remove(&dn
->dn_handle
->dnh_zrlock
);
443 dn
->dn_allocated_txg
= 0;
445 dn
->dn_assigned_txg
= 0;
448 if (dn
->dn_dirtyctx_firstset
!= NULL
) {
449 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
450 dn
->dn_dirtyctx_firstset
= NULL
;
452 if (dn
->dn_bonus
!= NULL
) {
453 mutex_enter(&dn
->dn_bonus
->db_mtx
);
454 dbuf_evict(dn
->dn_bonus
);
459 dn
->dn_have_spill
= B_FALSE
;
468 dmu_zfetch_rele(&dn
->dn_zfetch
);
469 kmem_cache_free(dnode_cache
, dn
);
470 arc_space_return(sizeof (dnode_t
), ARC_SPACE_OTHER
);
474 dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
475 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
480 blocksize
= 1 << zfs_default_bs
;
481 else if (blocksize
> SPA_MAXBLOCKSIZE
)
482 blocksize
= SPA_MAXBLOCKSIZE
;
484 blocksize
= P2ROUNDUP(blocksize
, SPA_MINBLOCKSIZE
);
487 ibs
= zfs_default_ibs
;
489 ibs
= MIN(MAX(ibs
, DN_MIN_INDBLKSHIFT
), DN_MAX_INDBLKSHIFT
);
491 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn
->dn_objset
,
492 dn
->dn_object
, tx
->tx_txg
, blocksize
, ibs
);
494 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
495 ASSERT(bcmp(dn
->dn_phys
, &dnode_phys_zero
, sizeof (dnode_phys_t
)) == 0);
496 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
);
497 ASSERT(ot
!= DMU_OT_NONE
);
498 ASSERT3U(ot
, <, DMU_OT_NUMTYPES
);
499 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
500 (bonustype
== DMU_OT_SA
&& bonuslen
== 0) ||
501 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0));
502 ASSERT3U(bonustype
, <, DMU_OT_NUMTYPES
);
503 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
504 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
505 ASSERT3U(dn
->dn_maxblkid
, ==, 0);
506 ASSERT3U(dn
->dn_allocated_txg
, ==, 0);
507 ASSERT3U(dn
->dn_assigned_txg
, ==, 0);
508 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
509 ASSERT3U(refcount_count(&dn
->dn_holds
), <=, 1);
510 ASSERT3P(list_head(&dn
->dn_dbufs
), ==, NULL
);
512 for (i
= 0; i
< TXG_SIZE
; i
++) {
513 ASSERT3U(dn
->dn_next_nblkptr
[i
], ==, 0);
514 ASSERT3U(dn
->dn_next_nlevels
[i
], ==, 0);
515 ASSERT3U(dn
->dn_next_indblkshift
[i
], ==, 0);
516 ASSERT3U(dn
->dn_next_bonuslen
[i
], ==, 0);
517 ASSERT3U(dn
->dn_next_bonustype
[i
], ==, 0);
518 ASSERT3U(dn
->dn_rm_spillblk
[i
], ==, 0);
519 ASSERT3U(dn
->dn_next_blksz
[i
], ==, 0);
520 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
521 ASSERT3P(list_head(&dn
->dn_dirty_records
[i
]), ==, NULL
);
522 ASSERT3U(avl_numnodes(&dn
->dn_ranges
[i
]), ==, 0);
526 dnode_setdblksz(dn
, blocksize
);
527 dn
->dn_indblkshift
= ibs
;
529 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
533 ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
534 dn
->dn_bonustype
= bonustype
;
535 dn
->dn_bonuslen
= bonuslen
;
536 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
537 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
541 if (dn
->dn_dirtyctx_firstset
) {
542 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
543 dn
->dn_dirtyctx_firstset
= NULL
;
546 dn
->dn_allocated_txg
= tx
->tx_txg
;
549 dnode_setdirty(dn
, tx
);
550 dn
->dn_next_indblkshift
[tx
->tx_txg
& TXG_MASK
] = ibs
;
551 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
552 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
553 dn
->dn_next_blksz
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_datablksz
;
557 dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
558 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
562 ASSERT3U(blocksize
, >=, SPA_MINBLOCKSIZE
);
563 ASSERT3U(blocksize
, <=, SPA_MAXBLOCKSIZE
);
564 ASSERT3U(blocksize
% SPA_MINBLOCKSIZE
, ==, 0);
565 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
|| dmu_tx_private_ok(tx
));
566 ASSERT(tx
->tx_txg
!= 0);
567 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
568 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0) ||
569 (bonustype
== DMU_OT_SA
&& bonuslen
== 0));
570 ASSERT3U(bonustype
, <, DMU_OT_NUMTYPES
);
571 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
573 /* clean up any unreferenced dbufs */
574 dnode_evict_dbufs(dn
);
578 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
579 dnode_setdirty(dn
, tx
);
580 if (dn
->dn_datablksz
!= blocksize
) {
581 /* change blocksize */
582 ASSERT(dn
->dn_maxblkid
== 0 &&
583 (BP_IS_HOLE(&dn
->dn_phys
->dn_blkptr
[0]) ||
584 dnode_block_freed(dn
, 0)));
585 dnode_setdblksz(dn
, blocksize
);
586 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = blocksize
;
588 if (dn
->dn_bonuslen
!= bonuslen
)
589 dn
->dn_next_bonuslen
[tx
->tx_txg
&TXG_MASK
] = bonuslen
;
591 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
594 nblkptr
= 1 + ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
595 if (dn
->dn_bonustype
!= bonustype
)
596 dn
->dn_next_bonustype
[tx
->tx_txg
&TXG_MASK
] = bonustype
;
597 if (dn
->dn_nblkptr
!= nblkptr
)
598 dn
->dn_next_nblkptr
[tx
->tx_txg
&TXG_MASK
] = nblkptr
;
599 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
600 dbuf_rm_spill(dn
, tx
);
601 dnode_rm_spill(dn
, tx
);
603 rw_exit(&dn
->dn_struct_rwlock
);
608 /* change bonus size and type */
609 mutex_enter(&dn
->dn_mtx
);
610 dn
->dn_bonustype
= bonustype
;
611 dn
->dn_bonuslen
= bonuslen
;
612 dn
->dn_nblkptr
= nblkptr
;
613 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
614 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
615 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
617 /* fix up the bonus db_size */
619 dn
->dn_bonus
->db
.db_size
=
620 DN_MAX_BONUSLEN
- (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
);
621 ASSERT(dn
->dn_bonuslen
<= dn
->dn_bonus
->db
.db_size
);
624 dn
->dn_allocated_txg
= tx
->tx_txg
;
625 mutex_exit(&dn
->dn_mtx
);
630 uint64_t dms_dnode_invalid
;
631 uint64_t dms_dnode_recheck1
;
632 uint64_t dms_dnode_recheck2
;
633 uint64_t dms_dnode_special
;
634 uint64_t dms_dnode_handle
;
635 uint64_t dms_dnode_rwlock
;
636 uint64_t dms_dnode_active
;
638 #endif /* DNODE_STATS */
641 dnode_move_impl(dnode_t
*odn
, dnode_t
*ndn
)
645 ASSERT(!RW_LOCK_HELD(&odn
->dn_struct_rwlock
));
646 ASSERT(MUTEX_NOT_HELD(&odn
->dn_mtx
));
647 ASSERT(MUTEX_NOT_HELD(&odn
->dn_dbufs_mtx
));
648 ASSERT(!RW_LOCK_HELD(&odn
->dn_zfetch
.zf_rwlock
));
651 ndn
->dn_objset
= odn
->dn_objset
;
652 ndn
->dn_object
= odn
->dn_object
;
653 ndn
->dn_dbuf
= odn
->dn_dbuf
;
654 ndn
->dn_handle
= odn
->dn_handle
;
655 ndn
->dn_phys
= odn
->dn_phys
;
656 ndn
->dn_type
= odn
->dn_type
;
657 ndn
->dn_bonuslen
= odn
->dn_bonuslen
;
658 ndn
->dn_bonustype
= odn
->dn_bonustype
;
659 ndn
->dn_nblkptr
= odn
->dn_nblkptr
;
660 ndn
->dn_checksum
= odn
->dn_checksum
;
661 ndn
->dn_compress
= odn
->dn_compress
;
662 ndn
->dn_nlevels
= odn
->dn_nlevels
;
663 ndn
->dn_indblkshift
= odn
->dn_indblkshift
;
664 ndn
->dn_datablkshift
= odn
->dn_datablkshift
;
665 ndn
->dn_datablkszsec
= odn
->dn_datablkszsec
;
666 ndn
->dn_datablksz
= odn
->dn_datablksz
;
667 ndn
->dn_maxblkid
= odn
->dn_maxblkid
;
668 bcopy(&odn
->dn_next_nblkptr
[0], &ndn
->dn_next_nblkptr
[0],
669 sizeof (odn
->dn_next_nblkptr
));
670 bcopy(&odn
->dn_next_nlevels
[0], &ndn
->dn_next_nlevels
[0],
671 sizeof (odn
->dn_next_nlevels
));
672 bcopy(&odn
->dn_next_indblkshift
[0], &ndn
->dn_next_indblkshift
[0],
673 sizeof (odn
->dn_next_indblkshift
));
674 bcopy(&odn
->dn_next_bonustype
[0], &ndn
->dn_next_bonustype
[0],
675 sizeof (odn
->dn_next_bonustype
));
676 bcopy(&odn
->dn_rm_spillblk
[0], &ndn
->dn_rm_spillblk
[0],
677 sizeof (odn
->dn_rm_spillblk
));
678 bcopy(&odn
->dn_next_bonuslen
[0], &ndn
->dn_next_bonuslen
[0],
679 sizeof (odn
->dn_next_bonuslen
));
680 bcopy(&odn
->dn_next_blksz
[0], &ndn
->dn_next_blksz
[0],
681 sizeof (odn
->dn_next_blksz
));
682 for (i
= 0; i
< TXG_SIZE
; i
++) {
683 list_move_tail(&ndn
->dn_dirty_records
[i
],
684 &odn
->dn_dirty_records
[i
]);
686 bcopy(&odn
->dn_ranges
[0], &ndn
->dn_ranges
[0], sizeof (odn
->dn_ranges
));
687 ndn
->dn_allocated_txg
= odn
->dn_allocated_txg
;
688 ndn
->dn_free_txg
= odn
->dn_free_txg
;
689 ndn
->dn_assigned_txg
= odn
->dn_assigned_txg
;
690 ndn
->dn_dirtyctx
= odn
->dn_dirtyctx
;
691 ndn
->dn_dirtyctx_firstset
= odn
->dn_dirtyctx_firstset
;
692 ASSERT(refcount_count(&odn
->dn_tx_holds
) == 0);
693 refcount_transfer(&ndn
->dn_holds
, &odn
->dn_holds
);
694 ASSERT(list_is_empty(&ndn
->dn_dbufs
));
695 list_move_tail(&ndn
->dn_dbufs
, &odn
->dn_dbufs
);
696 ndn
->dn_dbufs_count
= odn
->dn_dbufs_count
;
697 ndn
->dn_bonus
= odn
->dn_bonus
;
698 ndn
->dn_have_spill
= odn
->dn_have_spill
;
699 ndn
->dn_zio
= odn
->dn_zio
;
700 ndn
->dn_oldused
= odn
->dn_oldused
;
701 ndn
->dn_oldflags
= odn
->dn_oldflags
;
702 ndn
->dn_olduid
= odn
->dn_olduid
;
703 ndn
->dn_oldgid
= odn
->dn_oldgid
;
704 ndn
->dn_newuid
= odn
->dn_newuid
;
705 ndn
->dn_newgid
= odn
->dn_newgid
;
706 ndn
->dn_id_flags
= odn
->dn_id_flags
;
707 dmu_zfetch_init(&ndn
->dn_zfetch
, NULL
);
708 list_move_tail(&ndn
->dn_zfetch
.zf_stream
, &odn
->dn_zfetch
.zf_stream
);
709 ndn
->dn_zfetch
.zf_dnode
= odn
->dn_zfetch
.zf_dnode
;
710 ndn
->dn_zfetch
.zf_stream_cnt
= odn
->dn_zfetch
.zf_stream_cnt
;
711 ndn
->dn_zfetch
.zf_alloc_fail
= odn
->dn_zfetch
.zf_alloc_fail
;
714 * Update back pointers. Updating the handle fixes the back pointer of
715 * every descendant dbuf as well as the bonus dbuf.
717 ASSERT(ndn
->dn_handle
->dnh_dnode
== odn
);
718 ndn
->dn_handle
->dnh_dnode
= ndn
;
719 if (ndn
->dn_zfetch
.zf_dnode
== odn
) {
720 ndn
->dn_zfetch
.zf_dnode
= ndn
;
724 * Invalidate the original dnode by clearing all of its back pointers.
727 odn
->dn_handle
= NULL
;
728 list_create(&odn
->dn_dbufs
, sizeof (dmu_buf_impl_t
),
729 offsetof(dmu_buf_impl_t
, db_link
));
730 odn
->dn_dbufs_count
= 0;
731 odn
->dn_bonus
= NULL
;
732 odn
->dn_zfetch
.zf_dnode
= NULL
;
735 * Set the low bit of the objset pointer to ensure that dnode_move()
736 * recognizes the dnode as invalid in any subsequent callback.
738 POINTER_INVALIDATE(&odn
->dn_objset
);
741 * Satisfy the destructor.
743 for (i
= 0; i
< TXG_SIZE
; i
++) {
744 list_create(&odn
->dn_dirty_records
[i
],
745 sizeof (dbuf_dirty_record_t
),
746 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
747 odn
->dn_ranges
[i
].avl_root
= NULL
;
748 odn
->dn_ranges
[i
].avl_numnodes
= 0;
749 odn
->dn_next_nlevels
[i
] = 0;
750 odn
->dn_next_indblkshift
[i
] = 0;
751 odn
->dn_next_bonustype
[i
] = 0;
752 odn
->dn_rm_spillblk
[i
] = 0;
753 odn
->dn_next_bonuslen
[i
] = 0;
754 odn
->dn_next_blksz
[i
] = 0;
756 odn
->dn_allocated_txg
= 0;
757 odn
->dn_free_txg
= 0;
758 odn
->dn_assigned_txg
= 0;
759 odn
->dn_dirtyctx
= 0;
760 odn
->dn_dirtyctx_firstset
= NULL
;
761 odn
->dn_have_spill
= B_FALSE
;
764 odn
->dn_oldflags
= 0;
769 odn
->dn_id_flags
= 0;
775 odn
->dn_moved
= (uint8_t)-1;
781 dnode_move(void *buf
, void *newbuf
, size_t size
, void *arg
)
783 dnode_t
*odn
= buf
, *ndn
= newbuf
;
789 * The dnode is on the objset's list of known dnodes if the objset
790 * pointer is valid. We set the low bit of the objset pointer when
791 * freeing the dnode to invalidate it, and the memory patterns written
792 * by kmem (baddcafe and deadbeef) set at least one of the two low bits.
793 * A newly created dnode sets the objset pointer last of all to indicate
794 * that the dnode is known and in a valid state to be moved by this
798 if (!POINTER_IS_VALID(os
)) {
799 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_invalid
);
800 return (KMEM_CBRC_DONT_KNOW
);
804 * Ensure that the objset does not go away during the move.
806 rw_enter(&os_lock
, RW_WRITER
);
807 if (os
!= odn
->dn_objset
) {
809 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck1
);
810 return (KMEM_CBRC_DONT_KNOW
);
814 * If the dnode is still valid, then so is the objset. We know that no
815 * valid objset can be freed while we hold os_lock, so we can safely
816 * ensure that the objset remains in use.
818 mutex_enter(&os
->os_lock
);
821 * Recheck the objset pointer in case the dnode was removed just before
822 * acquiring the lock.
824 if (os
!= odn
->dn_objset
) {
825 mutex_exit(&os
->os_lock
);
827 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck2
);
828 return (KMEM_CBRC_DONT_KNOW
);
832 * At this point we know that as long as we hold os->os_lock, the dnode
833 * cannot be freed and fields within the dnode can be safely accessed.
834 * The objset listing this dnode cannot go away as long as this dnode is
838 if (DMU_OBJECT_IS_SPECIAL(odn
->dn_object
)) {
839 mutex_exit(&os
->os_lock
);
840 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_special
);
841 return (KMEM_CBRC_NO
);
843 ASSERT(odn
->dn_dbuf
!= NULL
); /* only "special" dnodes have no parent */
846 * Lock the dnode handle to prevent the dnode from obtaining any new
847 * holds. This also prevents the descendant dbufs and the bonus dbuf
848 * from accessing the dnode, so that we can discount their holds. The
849 * handle is safe to access because we know that while the dnode cannot
850 * go away, neither can its handle. Once we hold dnh_zrlock, we can
851 * safely move any dnode referenced only by dbufs.
853 if (!zrl_tryenter(&odn
->dn_handle
->dnh_zrlock
)) {
854 mutex_exit(&os
->os_lock
);
855 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_handle
);
856 return (KMEM_CBRC_LATER
);
860 * Ensure a consistent view of the dnode's holds and the dnode's dbufs.
861 * We need to guarantee that there is a hold for every dbuf in order to
862 * determine whether the dnode is actively referenced. Falsely matching
863 * a dbuf to an active hold would lead to an unsafe move. It's possible
864 * that a thread already having an active dnode hold is about to add a
865 * dbuf, and we can't compare hold and dbuf counts while the add is in
868 if (!rw_tryenter(&odn
->dn_struct_rwlock
, RW_WRITER
)) {
869 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
870 mutex_exit(&os
->os_lock
);
871 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_rwlock
);
872 return (KMEM_CBRC_LATER
);
876 * A dbuf may be removed (evicted) without an active dnode hold. In that
877 * case, the dbuf count is decremented under the handle lock before the
878 * dbuf's hold is released. This order ensures that if we count the hold
879 * after the dbuf is removed but before its hold is released, we will
880 * treat the unmatched hold as active and exit safely. If we count the
881 * hold before the dbuf is removed, the hold is discounted, and the
882 * removal is blocked until the move completes.
884 refcount
= refcount_count(&odn
->dn_holds
);
885 ASSERT(refcount
>= 0);
886 dbufs
= odn
->dn_dbufs_count
;
888 /* We can't have more dbufs than dnode holds. */
889 ASSERT3U(dbufs
, <=, refcount
);
890 DTRACE_PROBE3(dnode__move
, dnode_t
*, odn
, int64_t, refcount
,
893 if (refcount
> dbufs
) {
894 rw_exit(&odn
->dn_struct_rwlock
);
895 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
896 mutex_exit(&os
->os_lock
);
897 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_active
);
898 return (KMEM_CBRC_LATER
);
901 rw_exit(&odn
->dn_struct_rwlock
);
904 * At this point we know that anyone with a hold on the dnode is not
905 * actively referencing it. The dnode is known and in a valid state to
906 * move. We're holding the locks needed to execute the critical section.
908 dnode_move_impl(odn
, ndn
);
910 list_link_replace(&odn
->dn_link
, &ndn
->dn_link
);
911 /* If the dnode was safe to move, the refcount cannot have changed. */
912 ASSERT(refcount
== refcount_count(&ndn
->dn_holds
));
913 ASSERT(dbufs
== ndn
->dn_dbufs_count
);
914 zrl_exit(&ndn
->dn_handle
->dnh_zrlock
); /* handle has moved */
915 mutex_exit(&os
->os_lock
);
917 return (KMEM_CBRC_YES
);
922 dnode_special_close(dnode_handle_t
*dnh
)
924 dnode_t
*dn
= dnh
->dnh_dnode
;
927 * Wait for final references to the dnode to clear. This can
928 * only happen if the arc is asyncronously evicting state that
929 * has a hold on this dnode while we are trying to evict this
932 while (refcount_count(&dn
->dn_holds
) > 0)
934 zrl_add(&dnh
->dnh_zrlock
);
935 dnode_destroy(dn
); /* implicit zrl_remove() */
936 zrl_destroy(&dnh
->dnh_zrlock
);
937 dnh
->dnh_dnode
= NULL
;
941 dnode_special_open(objset_t
*os
, dnode_phys_t
*dnp
, uint64_t object
,
944 dnode_t
*dn
= dnode_create(os
, dnp
, NULL
, object
, dnh
);
946 zrl_init(&dnh
->dnh_zrlock
);
952 dnode_buf_pageout(dmu_buf_t
*db
, void *arg
)
954 dnode_children_t
*children_dnodes
= arg
;
956 int epb
= db
->db_size
>> DNODE_SHIFT
;
958 ASSERT(epb
== children_dnodes
->dnc_count
);
960 for (i
= 0; i
< epb
; i
++) {
961 dnode_handle_t
*dnh
= &children_dnodes
->dnc_children
[i
];
965 * The dnode handle lock guards against the dnode moving to
966 * another valid address, so there is no need here to guard
967 * against changes to or from NULL.
969 if (dnh
->dnh_dnode
== NULL
) {
970 zrl_destroy(&dnh
->dnh_zrlock
);
974 zrl_add(&dnh
->dnh_zrlock
);
977 * If there are holds on this dnode, then there should
978 * be holds on the dnode's containing dbuf as well; thus
979 * it wouldn't be eligible for eviction and this function
980 * would not have been called.
982 ASSERT(refcount_is_zero(&dn
->dn_holds
));
983 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
985 dnode_destroy(dn
); /* implicit zrl_remove() */
986 zrl_destroy(&dnh
->dnh_zrlock
);
987 dnh
->dnh_dnode
= NULL
;
989 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
990 (epb
- 1) * sizeof (dnode_handle_t
));
995 * EINVAL - invalid object number.
997 * succeeds even for free dnodes.
1000 dnode_hold_impl(objset_t
*os
, uint64_t object
, int flag
,
1001 void *tag
, dnode_t
**dnp
)
1004 int drop_struct_lock
= FALSE
;
1009 dnode_children_t
*children_dnodes
;
1010 dnode_handle_t
*dnh
;
1013 * If you are holding the spa config lock as writer, you shouldn't
1014 * be asking the DMU to do *anything* unless it's the root pool
1015 * which may require us to read from the root filesystem while
1016 * holding some (not all) of the locks as writer.
1018 ASSERT(spa_config_held(os
->os_spa
, SCL_ALL
, RW_WRITER
) == 0 ||
1019 (spa_is_root(os
->os_spa
) &&
1020 spa_config_held(os
->os_spa
, SCL_STATE
, RW_WRITER
)));
1022 if (object
== DMU_USERUSED_OBJECT
|| object
== DMU_GROUPUSED_OBJECT
) {
1023 dn
= (object
== DMU_USERUSED_OBJECT
) ?
1024 DMU_USERUSED_DNODE(os
) : DMU_GROUPUSED_DNODE(os
);
1028 if ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
)
1030 if ((flag
& DNODE_MUST_BE_FREE
) && type
!= DMU_OT_NONE
)
1033 (void) refcount_add(&dn
->dn_holds
, tag
);
1038 if (object
== 0 || object
>= DN_MAX_OBJECT
)
1041 mdn
= DMU_META_DNODE(os
);
1042 ASSERT(mdn
->dn_object
== DMU_META_DNODE_OBJECT
);
1046 if (!RW_WRITE_HELD(&mdn
->dn_struct_rwlock
)) {
1047 rw_enter(&mdn
->dn_struct_rwlock
, RW_READER
);
1048 drop_struct_lock
= TRUE
;
1051 blk
= dbuf_whichblock(mdn
, object
* sizeof (dnode_phys_t
));
1053 db
= dbuf_hold(mdn
, blk
, FTAG
);
1054 if (drop_struct_lock
)
1055 rw_exit(&mdn
->dn_struct_rwlock
);
1058 err
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
);
1060 dbuf_rele(db
, FTAG
);
1064 ASSERT3U(db
->db
.db_size
, >=, 1<<DNODE_SHIFT
);
1065 epb
= db
->db
.db_size
>> DNODE_SHIFT
;
1067 idx
= object
& (epb
-1);
1069 ASSERT(DB_DNODE(db
)->dn_type
== DMU_OT_DNODE
);
1070 children_dnodes
= dmu_buf_get_user(&db
->db
);
1071 if (children_dnodes
== NULL
) {
1073 dnode_children_t
*winner
;
1074 children_dnodes
= kmem_alloc(sizeof (dnode_children_t
) +
1075 (epb
- 1) * sizeof (dnode_handle_t
), KM_SLEEP
);
1076 children_dnodes
->dnc_count
= epb
;
1077 dnh
= &children_dnodes
->dnc_children
[0];
1078 for (i
= 0; i
< epb
; i
++) {
1079 zrl_init(&dnh
[i
].dnh_zrlock
);
1080 dnh
[i
].dnh_dnode
= NULL
;
1082 if ((winner
= dmu_buf_set_user(&db
->db
, children_dnodes
, NULL
,
1083 dnode_buf_pageout
))) {
1084 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
1085 (epb
- 1) * sizeof (dnode_handle_t
));
1086 children_dnodes
= winner
;
1089 ASSERT(children_dnodes
->dnc_count
== epb
);
1091 dnh
= &children_dnodes
->dnc_children
[idx
];
1092 zrl_add(&dnh
->dnh_zrlock
);
1093 if ((dn
= dnh
->dnh_dnode
) == NULL
) {
1094 dnode_phys_t
*phys
= (dnode_phys_t
*)db
->db
.db_data
+idx
;
1097 dn
= dnode_create(os
, phys
, db
, object
, dnh
);
1098 winner
= atomic_cas_ptr(&dnh
->dnh_dnode
, NULL
, dn
);
1099 if (winner
!= NULL
) {
1100 zrl_add(&dnh
->dnh_zrlock
);
1101 dnode_destroy(dn
); /* implicit zrl_remove() */
1106 mutex_enter(&dn
->dn_mtx
);
1108 if (dn
->dn_free_txg
||
1109 ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
) ||
1110 ((flag
& DNODE_MUST_BE_FREE
) &&
1111 (type
!= DMU_OT_NONE
|| !refcount_is_zero(&dn
->dn_holds
)))) {
1112 mutex_exit(&dn
->dn_mtx
);
1113 zrl_remove(&dnh
->dnh_zrlock
);
1114 dbuf_rele(db
, FTAG
);
1115 return (type
== DMU_OT_NONE
? ENOENT
: EEXIST
);
1117 mutex_exit(&dn
->dn_mtx
);
1119 if (refcount_add(&dn
->dn_holds
, tag
) == 1)
1120 dbuf_add_ref(db
, dnh
);
1121 /* Now we can rely on the hold to prevent the dnode from moving. */
1122 zrl_remove(&dnh
->dnh_zrlock
);
1125 ASSERT3P(dn
->dn_dbuf
, ==, db
);
1126 ASSERT3U(dn
->dn_object
, ==, object
);
1127 dbuf_rele(db
, FTAG
);
1134 * Return held dnode if the object is allocated, NULL if not.
1137 dnode_hold(objset_t
*os
, uint64_t object
, void *tag
, dnode_t
**dnp
)
1139 return (dnode_hold_impl(os
, object
, DNODE_MUST_BE_ALLOCATED
, tag
, dnp
));
1143 * Can only add a reference if there is already at least one
1144 * reference on the dnode. Returns FALSE if unable to add a
1148 dnode_add_ref(dnode_t
*dn
, void *tag
)
1150 mutex_enter(&dn
->dn_mtx
);
1151 if (refcount_is_zero(&dn
->dn_holds
)) {
1152 mutex_exit(&dn
->dn_mtx
);
1155 VERIFY(1 < refcount_add(&dn
->dn_holds
, tag
));
1156 mutex_exit(&dn
->dn_mtx
);
1161 dnode_rele(dnode_t
*dn
, void *tag
)
1164 /* Get while the hold prevents the dnode from moving. */
1165 dmu_buf_impl_t
*db
= dn
->dn_dbuf
;
1166 dnode_handle_t
*dnh
= dn
->dn_handle
;
1168 mutex_enter(&dn
->dn_mtx
);
1169 refs
= refcount_remove(&dn
->dn_holds
, tag
);
1170 mutex_exit(&dn
->dn_mtx
);
1173 * It's unsafe to release the last hold on a dnode by dnode_rele() or
1174 * indirectly by dbuf_rele() while relying on the dnode handle to
1175 * prevent the dnode from moving, since releasing the last hold could
1176 * result in the dnode's parent dbuf evicting its dnode handles. For
1177 * that reason anyone calling dnode_rele() or dbuf_rele() without some
1178 * other direct or indirect hold on the dnode must first drop the dnode
1181 ASSERT(refs
> 0 || dnh
->dnh_zrlock
.zr_owner
!= curthread
);
1183 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
1184 if (refs
== 0 && db
!= NULL
) {
1186 * Another thread could add a hold to the dnode handle in
1187 * dnode_hold_impl() while holding the parent dbuf. Since the
1188 * hold on the parent dbuf prevents the handle from being
1189 * destroyed, the hold on the handle is OK. We can't yet assert
1190 * that the handle has zero references, but that will be
1191 * asserted anyway when the handle gets destroyed.
1198 dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
)
1200 objset_t
*os
= dn
->dn_objset
;
1201 uint64_t txg
= tx
->tx_txg
;
1203 if (DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
1204 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1211 mutex_enter(&dn
->dn_mtx
);
1212 ASSERT(dn
->dn_phys
->dn_type
|| dn
->dn_allocated_txg
);
1213 ASSERT(dn
->dn_free_txg
== 0 || dn
->dn_free_txg
>= txg
);
1214 mutex_exit(&dn
->dn_mtx
);
1218 * Determine old uid/gid when necessary
1220 dmu_objset_userquota_get_ids(dn
, B_TRUE
, tx
);
1222 mutex_enter(&os
->os_lock
);
1225 * If we are already marked dirty, we're done.
1227 if (list_link_active(&dn
->dn_dirty_link
[txg
& TXG_MASK
])) {
1228 mutex_exit(&os
->os_lock
);
1232 ASSERT(!refcount_is_zero(&dn
->dn_holds
) || list_head(&dn
->dn_dbufs
));
1233 ASSERT(dn
->dn_datablksz
!= 0);
1234 ASSERT3U(dn
->dn_next_bonuslen
[txg
&TXG_MASK
], ==, 0);
1235 ASSERT3U(dn
->dn_next_blksz
[txg
&TXG_MASK
], ==, 0);
1236 ASSERT3U(dn
->dn_next_bonustype
[txg
&TXG_MASK
], ==, 0);
1238 dprintf_ds(os
->os_dsl_dataset
, "obj=%llu txg=%llu\n",
1239 dn
->dn_object
, txg
);
1241 if (dn
->dn_free_txg
> 0 && dn
->dn_free_txg
<= txg
) {
1242 list_insert_tail(&os
->os_free_dnodes
[txg
&TXG_MASK
], dn
);
1244 list_insert_tail(&os
->os_dirty_dnodes
[txg
&TXG_MASK
], dn
);
1247 mutex_exit(&os
->os_lock
);
1250 * The dnode maintains a hold on its containing dbuf as
1251 * long as there are holds on it. Each instantiated child
1252 * dbuf maintains a hold on the dnode. When the last child
1253 * drops its hold, the dnode will drop its hold on the
1254 * containing dbuf. We add a "dirty hold" here so that the
1255 * dnode will hang around after we finish processing its
1258 VERIFY(dnode_add_ref(dn
, (void *)(uintptr_t)tx
->tx_txg
));
1260 (void) dbuf_dirty(dn
->dn_dbuf
, tx
);
1262 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1266 dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
)
1268 int txgoff
= tx
->tx_txg
& TXG_MASK
;
1270 dprintf("dn=%p txg=%llu\n", dn
, tx
->tx_txg
);
1272 /* we should be the only holder... hopefully */
1273 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */
1275 mutex_enter(&dn
->dn_mtx
);
1276 if (dn
->dn_type
== DMU_OT_NONE
|| dn
->dn_free_txg
) {
1277 mutex_exit(&dn
->dn_mtx
);
1280 dn
->dn_free_txg
= tx
->tx_txg
;
1281 mutex_exit(&dn
->dn_mtx
);
1284 * If the dnode is already dirty, it needs to be moved from
1285 * the dirty list to the free list.
1287 mutex_enter(&dn
->dn_objset
->os_lock
);
1288 if (list_link_active(&dn
->dn_dirty_link
[txgoff
])) {
1289 list_remove(&dn
->dn_objset
->os_dirty_dnodes
[txgoff
], dn
);
1290 list_insert_tail(&dn
->dn_objset
->os_free_dnodes
[txgoff
], dn
);
1291 mutex_exit(&dn
->dn_objset
->os_lock
);
1293 mutex_exit(&dn
->dn_objset
->os_lock
);
1294 dnode_setdirty(dn
, tx
);
1299 * Try to change the block size for the indicated dnode. This can only
1300 * succeed if there are no blocks allocated or dirty beyond first block
1303 dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
)
1305 dmu_buf_impl_t
*db
, *db_next
;
1309 size
= SPA_MINBLOCKSIZE
;
1310 if (size
> SPA_MAXBLOCKSIZE
)
1311 size
= SPA_MAXBLOCKSIZE
;
1313 size
= P2ROUNDUP(size
, SPA_MINBLOCKSIZE
);
1315 if (ibs
== dn
->dn_indblkshift
)
1318 if (size
>> SPA_MINBLOCKSHIFT
== dn
->dn_datablkszsec
&& ibs
== 0)
1321 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1323 /* Check for any allocated blocks beyond the first */
1324 if (dn
->dn_phys
->dn_maxblkid
!= 0)
1327 mutex_enter(&dn
->dn_dbufs_mtx
);
1328 for (db
= list_head(&dn
->dn_dbufs
); db
; db
= db_next
) {
1329 db_next
= list_next(&dn
->dn_dbufs
, db
);
1331 if (db
->db_blkid
!= 0 && db
->db_blkid
!= DMU_BONUS_BLKID
&&
1332 db
->db_blkid
!= DMU_SPILL_BLKID
) {
1333 mutex_exit(&dn
->dn_dbufs_mtx
);
1337 mutex_exit(&dn
->dn_dbufs_mtx
);
1339 if (ibs
&& dn
->dn_nlevels
!= 1)
1342 /* resize the old block */
1343 err
= dbuf_hold_impl(dn
, 0, 0, TRUE
, FTAG
, &db
);
1345 dbuf_new_size(db
, size
, tx
);
1346 else if (err
!= ENOENT
)
1349 dnode_setdblksz(dn
, size
);
1350 dnode_setdirty(dn
, tx
);
1351 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = size
;
1353 dn
->dn_indblkshift
= ibs
;
1354 dn
->dn_next_indblkshift
[tx
->tx_txg
&TXG_MASK
] = ibs
;
1356 /* rele after we have fixed the blocksize in the dnode */
1358 dbuf_rele(db
, FTAG
);
1360 rw_exit(&dn
->dn_struct_rwlock
);
1364 rw_exit(&dn
->dn_struct_rwlock
);
1368 /* read-holding callers must not rely on the lock being continuously held */
1370 dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t have_read
)
1372 uint64_t txgoff
= tx
->tx_txg
& TXG_MASK
;
1373 int epbs
, new_nlevels
;
1376 ASSERT(blkid
!= DMU_BONUS_BLKID
);
1379 RW_READ_HELD(&dn
->dn_struct_rwlock
) :
1380 RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
1383 * if we have a read-lock, check to see if we need to do any work
1384 * before upgrading to a write-lock.
1387 if (blkid
<= dn
->dn_maxblkid
)
1390 if (!rw_tryupgrade(&dn
->dn_struct_rwlock
)) {
1391 rw_exit(&dn
->dn_struct_rwlock
);
1392 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1396 if (blkid
<= dn
->dn_maxblkid
)
1399 dn
->dn_maxblkid
= blkid
;
1402 * Compute the number of levels necessary to support the new maxblkid.
1405 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1406 for (sz
= dn
->dn_nblkptr
;
1407 sz
<= blkid
&& sz
>= dn
->dn_nblkptr
; sz
<<= epbs
)
1410 if (new_nlevels
> dn
->dn_nlevels
) {
1411 int old_nlevels
= dn
->dn_nlevels
;
1414 dbuf_dirty_record_t
*new, *dr
, *dr_next
;
1416 dn
->dn_nlevels
= new_nlevels
;
1418 ASSERT3U(new_nlevels
, >, dn
->dn_next_nlevels
[txgoff
]);
1419 dn
->dn_next_nlevels
[txgoff
] = new_nlevels
;
1421 /* dirty the left indirects */
1422 db
= dbuf_hold_level(dn
, old_nlevels
, 0, FTAG
);
1424 new = dbuf_dirty(db
, tx
);
1425 dbuf_rele(db
, FTAG
);
1427 /* transfer the dirty records to the new indirect */
1428 mutex_enter(&dn
->dn_mtx
);
1429 mutex_enter(&new->dt
.di
.dr_mtx
);
1430 list
= &dn
->dn_dirty_records
[txgoff
];
1431 for (dr
= list_head(list
); dr
; dr
= dr_next
) {
1432 dr_next
= list_next(&dn
->dn_dirty_records
[txgoff
], dr
);
1433 if (dr
->dr_dbuf
->db_level
!= new_nlevels
-1 &&
1434 dr
->dr_dbuf
->db_blkid
!= DMU_BONUS_BLKID
&&
1435 dr
->dr_dbuf
->db_blkid
!= DMU_SPILL_BLKID
) {
1436 ASSERT(dr
->dr_dbuf
->db_level
== old_nlevels
-1);
1437 list_remove(&dn
->dn_dirty_records
[txgoff
], dr
);
1438 list_insert_tail(&new->dt
.di
.dr_children
, dr
);
1439 dr
->dr_parent
= new;
1442 mutex_exit(&new->dt
.di
.dr_mtx
);
1443 mutex_exit(&dn
->dn_mtx
);
1448 rw_downgrade(&dn
->dn_struct_rwlock
);
1452 dnode_clear_range(dnode_t
*dn
, uint64_t blkid
, uint64_t nblks
, dmu_tx_t
*tx
)
1454 avl_tree_t
*tree
= &dn
->dn_ranges
[tx
->tx_txg
&TXG_MASK
];
1457 free_range_t rp_tofind
;
1458 uint64_t endblk
= blkid
+ nblks
;
1460 ASSERT(MUTEX_HELD(&dn
->dn_mtx
));
1461 ASSERT(nblks
<= UINT64_MAX
- blkid
); /* no overflow */
1463 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1464 blkid
, nblks
, tx
->tx_txg
);
1465 rp_tofind
.fr_blkid
= blkid
;
1466 rp
= avl_find(tree
, &rp_tofind
, &where
);
1468 rp
= avl_nearest(tree
, where
, AVL_BEFORE
);
1470 rp
= avl_nearest(tree
, where
, AVL_AFTER
);
1472 while (rp
&& (rp
->fr_blkid
<= blkid
+ nblks
)) {
1473 uint64_t fr_endblk
= rp
->fr_blkid
+ rp
->fr_nblks
;
1474 free_range_t
*nrp
= AVL_NEXT(tree
, rp
);
1476 if (blkid
<= rp
->fr_blkid
&& endblk
>= fr_endblk
) {
1477 /* clear this entire range */
1478 avl_remove(tree
, rp
);
1479 kmem_free(rp
, sizeof (free_range_t
));
1480 } else if (blkid
<= rp
->fr_blkid
&&
1481 endblk
> rp
->fr_blkid
&& endblk
< fr_endblk
) {
1482 /* clear the beginning of this range */
1483 rp
->fr_blkid
= endblk
;
1484 rp
->fr_nblks
= fr_endblk
- endblk
;
1485 } else if (blkid
> rp
->fr_blkid
&& blkid
< fr_endblk
&&
1486 endblk
>= fr_endblk
) {
1487 /* clear the end of this range */
1488 rp
->fr_nblks
= blkid
- rp
->fr_blkid
;
1489 } else if (blkid
> rp
->fr_blkid
&& endblk
< fr_endblk
) {
1490 /* clear a chunk out of this range */
1491 free_range_t
*new_rp
=
1492 kmem_alloc(sizeof (free_range_t
), KM_SLEEP
);
1494 new_rp
->fr_blkid
= endblk
;
1495 new_rp
->fr_nblks
= fr_endblk
- endblk
;
1496 avl_insert_here(tree
, new_rp
, rp
, AVL_AFTER
);
1497 rp
->fr_nblks
= blkid
- rp
->fr_blkid
;
1499 /* there may be no overlap */
1505 dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
)
1508 uint64_t blkoff
, blkid
, nblks
;
1509 int blksz
, blkshift
, head
, tail
;
1513 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1514 blksz
= dn
->dn_datablksz
;
1515 blkshift
= dn
->dn_datablkshift
;
1516 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1519 len
= UINT64_MAX
- off
;
1524 * First, block align the region to free:
1527 head
= P2NPHASE(off
, blksz
);
1528 blkoff
= P2PHASE(off
, blksz
);
1529 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1532 ASSERT(dn
->dn_maxblkid
== 0);
1533 if (off
== 0 && len
>= blksz
) {
1534 /* Freeing the whole block; fast-track this request */
1538 } else if (off
>= blksz
) {
1539 /* Freeing past end-of-data */
1542 /* Freeing part of the block. */
1544 ASSERT3U(head
, >, 0);
1548 /* zero out any partial block data at the start of the range */
1550 ASSERT3U(blkoff
+ head
, ==, blksz
);
1553 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
), TRUE
,
1557 /* don't dirty if it isn't on disk and isn't dirty */
1558 if (db
->db_last_dirty
||
1559 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1560 rw_exit(&dn
->dn_struct_rwlock
);
1561 dbuf_will_dirty(db
, tx
);
1562 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1563 data
= db
->db
.db_data
;
1564 bzero(data
+ blkoff
, head
);
1566 dbuf_rele(db
, FTAG
);
1572 /* If the range was less than one block, we're done */
1576 /* If the remaining range is past end of file, we're done */
1577 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1580 ASSERT(ISP2(blksz
));
1584 tail
= P2PHASE(len
, blksz
);
1586 ASSERT3U(P2PHASE(off
, blksz
), ==, 0);
1587 /* zero out any partial block data at the end of the range */
1591 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
+len
),
1592 TRUE
, FTAG
, &db
) == 0) {
1593 /* don't dirty if not on disk and not dirty */
1594 if (db
->db_last_dirty
||
1595 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1596 rw_exit(&dn
->dn_struct_rwlock
);
1597 dbuf_will_dirty(db
, tx
);
1598 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1599 bzero(db
->db
.db_data
, tail
);
1601 dbuf_rele(db
, FTAG
);
1606 /* If the range did not include a full block, we are done */
1610 ASSERT(IS_P2ALIGNED(off
, blksz
));
1611 ASSERT(trunc
|| IS_P2ALIGNED(len
, blksz
));
1612 blkid
= off
>> blkshift
;
1613 nblks
= len
>> blkshift
;
1618 * Read in and mark all the level-1 indirects dirty,
1619 * so that they will stay in memory until syncing phase.
1620 * Always dirty the first and last indirect to make sure
1621 * we dirty all the partial indirects.
1623 if (dn
->dn_nlevels
> 1) {
1624 uint64_t i
, first
, last
;
1625 int shift
= epbs
+ dn
->dn_datablkshift
;
1627 first
= blkid
>> epbs
;
1628 if ((db
= dbuf_hold_level(dn
, 1, first
, FTAG
))) {
1629 dbuf_will_dirty(db
, tx
);
1630 dbuf_rele(db
, FTAG
);
1633 last
= dn
->dn_maxblkid
>> epbs
;
1635 last
= (blkid
+ nblks
- 1) >> epbs
;
1636 if (last
> first
&& (db
= dbuf_hold_level(dn
, 1, last
, FTAG
))) {
1637 dbuf_will_dirty(db
, tx
);
1638 dbuf_rele(db
, FTAG
);
1640 for (i
= first
+ 1; i
< last
; i
++) {
1641 uint64_t ibyte
= i
<< shift
;
1644 err
= dnode_next_offset(dn
,
1645 DNODE_FIND_HAVELOCK
, &ibyte
, 1, 1, 0);
1647 if (err
== ESRCH
|| i
>= last
)
1650 db
= dbuf_hold_level(dn
, 1, i
, FTAG
);
1652 dbuf_will_dirty(db
, tx
);
1653 dbuf_rele(db
, FTAG
);
1659 * Add this range to the dnode range list.
1660 * We will finish up this free operation in the syncing phase.
1662 mutex_enter(&dn
->dn_mtx
);
1663 dnode_clear_range(dn
, blkid
, nblks
, tx
);
1665 free_range_t
*rp
, *found
;
1667 avl_tree_t
*tree
= &dn
->dn_ranges
[tx
->tx_txg
&TXG_MASK
];
1669 /* Add new range to dn_ranges */
1670 rp
= kmem_alloc(sizeof (free_range_t
), KM_SLEEP
);
1671 rp
->fr_blkid
= blkid
;
1672 rp
->fr_nblks
= nblks
;
1673 found
= avl_find(tree
, rp
, &where
);
1674 ASSERT(found
== NULL
);
1675 avl_insert(tree
, rp
, where
);
1676 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1677 blkid
, nblks
, tx
->tx_txg
);
1679 mutex_exit(&dn
->dn_mtx
);
1681 dbuf_free_range(dn
, blkid
, blkid
+ nblks
- 1, tx
);
1682 dnode_setdirty(dn
, tx
);
1684 if (trunc
&& dn
->dn_maxblkid
>= (off
>> blkshift
))
1685 dn
->dn_maxblkid
= (off
>> blkshift
? (off
>> blkshift
) - 1 : 0);
1687 rw_exit(&dn
->dn_struct_rwlock
);
1691 dnode_spill_freed(dnode_t
*dn
)
1695 mutex_enter(&dn
->dn_mtx
);
1696 for (i
= 0; i
< TXG_SIZE
; i
++) {
1697 if (dn
->dn_rm_spillblk
[i
] == DN_KILL_SPILLBLK
)
1700 mutex_exit(&dn
->dn_mtx
);
1701 return (i
< TXG_SIZE
);
1704 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1706 dnode_block_freed(dnode_t
*dn
, uint64_t blkid
)
1708 free_range_t range_tofind
;
1709 void *dp
= spa_get_dsl(dn
->dn_objset
->os_spa
);
1712 if (blkid
== DMU_BONUS_BLKID
)
1716 * If we're in the process of opening the pool, dp will not be
1717 * set yet, but there shouldn't be anything dirty.
1722 if (dn
->dn_free_txg
)
1725 if (blkid
== DMU_SPILL_BLKID
)
1726 return (dnode_spill_freed(dn
));
1728 range_tofind
.fr_blkid
= blkid
;
1729 mutex_enter(&dn
->dn_mtx
);
1730 for (i
= 0; i
< TXG_SIZE
; i
++) {
1731 free_range_t
*range_found
;
1734 range_found
= avl_find(&dn
->dn_ranges
[i
], &range_tofind
, &idx
);
1736 ASSERT(range_found
->fr_nblks
> 0);
1739 range_found
= avl_nearest(&dn
->dn_ranges
[i
], idx
, AVL_BEFORE
);
1741 range_found
->fr_blkid
+ range_found
->fr_nblks
> blkid
)
1744 mutex_exit(&dn
->dn_mtx
);
1745 return (i
< TXG_SIZE
);
1748 /* call from syncing context when we actually write/free space for this dnode */
1750 dnode_diduse_space(dnode_t
*dn
, int64_t delta
)
1753 dprintf_dnode(dn
, "dn=%p dnp=%p used=%llu delta=%lld\n",
1755 (u_longlong_t
)dn
->dn_phys
->dn_used
,
1758 mutex_enter(&dn
->dn_mtx
);
1759 space
= DN_USED_BYTES(dn
->dn_phys
);
1761 ASSERT3U(space
+ delta
, >=, space
); /* no overflow */
1763 ASSERT3U(space
, >=, -delta
); /* no underflow */
1766 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_DNODE_BYTES
) {
1767 ASSERT((dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) == 0);
1768 ASSERT3U(P2PHASE(space
, 1<<DEV_BSHIFT
), ==, 0);
1769 dn
->dn_phys
->dn_used
= space
>> DEV_BSHIFT
;
1771 dn
->dn_phys
->dn_used
= space
;
1772 dn
->dn_phys
->dn_flags
|= DNODE_FLAG_USED_BYTES
;
1774 mutex_exit(&dn
->dn_mtx
);
1778 * Call when we think we're going to write/free space in open context.
1779 * Be conservative (ie. OK to write less than this or free more than
1780 * this, but don't write more or free less).
1783 dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
)
1785 objset_t
*os
= dn
->dn_objset
;
1786 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1789 space
= spa_get_asize(os
->os_spa
, space
);
1792 dsl_dir_willuse_space(ds
->ds_dir
, space
, tx
);
1794 dmu_tx_willuse_space(tx
, space
);
1798 * This function scans a block at the indicated "level" looking for
1799 * a hole or data (depending on 'flags'). If level > 0, then we are
1800 * scanning an indirect block looking at its pointers. If level == 0,
1801 * then we are looking at a block of dnodes. If we don't find what we
1802 * are looking for in the block, we return ESRCH. Otherwise, return
1803 * with *offset pointing to the beginning (if searching forwards) or
1804 * end (if searching backwards) of the range covered by the block
1805 * pointer we matched on (or dnode).
1807 * The basic search algorithm used below by dnode_next_offset() is to
1808 * use this function to search up the block tree (widen the search) until
1809 * we find something (i.e., we don't return ESRCH) and then search back
1810 * down the tree (narrow the search) until we reach our original search
1814 dnode_next_offset_level(dnode_t
*dn
, int flags
, uint64_t *offset
,
1815 int lvl
, uint64_t blkfill
, uint64_t txg
)
1817 dmu_buf_impl_t
*db
= NULL
;
1819 uint64_t epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1820 uint64_t epb
= 1ULL << epbs
;
1821 uint64_t minfill
, maxfill
;
1823 int i
, inc
, error
, span
;
1825 dprintf("probing object %llu offset %llx level %d of %u\n",
1826 dn
->dn_object
, *offset
, lvl
, dn
->dn_phys
->dn_nlevels
);
1828 hole
= ((flags
& DNODE_FIND_HOLE
) != 0);
1829 inc
= (flags
& DNODE_FIND_BACKWARDS
) ? -1 : 1;
1830 ASSERT(txg
== 0 || !hole
);
1832 if (lvl
== dn
->dn_phys
->dn_nlevels
) {
1834 epb
= dn
->dn_phys
->dn_nblkptr
;
1835 data
= dn
->dn_phys
->dn_blkptr
;
1837 uint64_t blkid
= dbuf_whichblock(dn
, *offset
) >> (epbs
* lvl
);
1838 error
= dbuf_hold_impl(dn
, lvl
, blkid
, TRUE
, FTAG
, &db
);
1840 if (error
!= ENOENT
)
1845 * This can only happen when we are searching up
1846 * the block tree for data. We don't really need to
1847 * adjust the offset, as we will just end up looking
1848 * at the pointer to this block in its parent, and its
1849 * going to be unallocated, so we will skip over it.
1853 error
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
| DB_RF_HAVESTRUCT
);
1855 dbuf_rele(db
, FTAG
);
1858 data
= db
->db
.db_data
;
1862 (db
->db_blkptr
== NULL
|| db
->db_blkptr
->blk_birth
<= txg
)) {
1864 * This can only happen when we are searching up the tree
1865 * and these conditions mean that we need to keep climbing.
1868 } else if (lvl
== 0) {
1869 dnode_phys_t
*dnp
= data
;
1871 ASSERT(dn
->dn_type
== DMU_OT_DNODE
);
1873 for (i
= (*offset
>> span
) & (blkfill
- 1);
1874 i
>= 0 && i
< blkfill
; i
+= inc
) {
1875 if ((dnp
[i
].dn_type
== DMU_OT_NONE
) == hole
)
1877 *offset
+= (1ULL << span
) * inc
;
1879 if (i
< 0 || i
== blkfill
)
1882 blkptr_t
*bp
= data
;
1883 uint64_t start
= *offset
;
1884 span
= (lvl
- 1) * epbs
+ dn
->dn_datablkshift
;
1886 maxfill
= blkfill
<< ((lvl
- 1) * epbs
);
1893 *offset
= *offset
>> span
;
1894 for (i
= BF64_GET(*offset
, 0, epbs
);
1895 i
>= 0 && i
< epb
; i
+= inc
) {
1896 if (bp
[i
].blk_fill
>= minfill
&&
1897 bp
[i
].blk_fill
<= maxfill
&&
1898 (hole
|| bp
[i
].blk_birth
> txg
))
1900 if (inc
> 0 || *offset
> 0)
1903 *offset
= *offset
<< span
;
1905 /* traversing backwards; position offset at the end */
1906 ASSERT3U(*offset
, <=, start
);
1907 *offset
= MIN(*offset
+ (1ULL << span
) - 1, start
);
1908 } else if (*offset
< start
) {
1911 if (i
< 0 || i
>= epb
)
1916 dbuf_rele(db
, FTAG
);
1922 * Find the next hole, data, or sparse region at or after *offset.
1923 * The value 'blkfill' tells us how many items we expect to find
1924 * in an L0 data block; this value is 1 for normal objects,
1925 * DNODES_PER_BLOCK for the meta dnode, and some fraction of
1926 * DNODES_PER_BLOCK when searching for sparse regions thereof.
1930 * dnode_next_offset(dn, flags, offset, 1, 1, 0);
1931 * Finds the next/previous hole/data in a file.
1932 * Used in dmu_offset_next().
1934 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg);
1935 * Finds the next free/allocated dnode an objset's meta-dnode.
1936 * Only finds objects that have new contents since txg (ie.
1937 * bonus buffer changes and content removal are ignored).
1938 * Used in dmu_object_next().
1940 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0);
1941 * Finds the next L2 meta-dnode bp that's at most 1/4 full.
1942 * Used in dmu_object_alloc().
1945 dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *offset
,
1946 int minlvl
, uint64_t blkfill
, uint64_t txg
)
1948 uint64_t initial_offset
= *offset
;
1952 if (!(flags
& DNODE_FIND_HAVELOCK
))
1953 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1955 if (dn
->dn_phys
->dn_nlevels
== 0) {
1960 if (dn
->dn_datablkshift
== 0) {
1961 if (*offset
< dn
->dn_datablksz
) {
1962 if (flags
& DNODE_FIND_HOLE
)
1963 *offset
= dn
->dn_datablksz
;
1970 maxlvl
= dn
->dn_phys
->dn_nlevels
;
1972 for (lvl
= minlvl
; lvl
<= maxlvl
; lvl
++) {
1973 error
= dnode_next_offset_level(dn
,
1974 flags
, offset
, lvl
, blkfill
, txg
);
1979 while (error
== 0 && --lvl
>= minlvl
) {
1980 error
= dnode_next_offset_level(dn
,
1981 flags
, offset
, lvl
, blkfill
, txg
);
1984 if (error
== 0 && (flags
& DNODE_FIND_BACKWARDS
?
1985 initial_offset
< *offset
: initial_offset
> *offset
))
1988 if (!(flags
& DNODE_FIND_HAVELOCK
))
1989 rw_exit(&dn
->dn_struct_rwlock
);