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.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 #include <sys/zfs_context.h>
29 #include <sys/dnode.h>
31 #include <sys/dmu_impl.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dsl_dataset.h>
38 #include <sys/dmu_zfetch.h>
39 #include <sys/range_tree.h>
40 #include <sys/trace_dnode.h>
42 static kmem_cache_t
*dnode_cache
;
44 * Define DNODE_STATS to turn on statistic gathering. By default, it is only
45 * turned on when DEBUG is also defined.
52 #define DNODE_STAT_ADD(stat) ((stat)++)
54 #define DNODE_STAT_ADD(stat) /* nothing */
55 #endif /* DNODE_STATS */
57 ASSERTV(static dnode_phys_t dnode_phys_zero
);
59 int zfs_default_bs
= SPA_MINBLOCKSHIFT
;
60 int zfs_default_ibs
= DN_MAX_INDBLKSHIFT
;
63 static kmem_cbrc_t
dnode_move(void *, void *, size_t, void *);
67 dbuf_compare(const void *x1
, const void *x2
)
69 const dmu_buf_impl_t
*d1
= x1
;
70 const dmu_buf_impl_t
*d2
= x2
;
72 if (d1
->db_level
< d2
->db_level
) {
75 if (d1
->db_level
> d2
->db_level
) {
79 if (d1
->db_blkid
< d2
->db_blkid
) {
82 if (d1
->db_blkid
> d2
->db_blkid
) {
86 if (d1
->db_state
< d2
->db_state
) {
89 if (d1
->db_state
> d2
->db_state
) {
93 ASSERT3S(d1
->db_state
, !=, DB_SEARCH
);
94 ASSERT3S(d2
->db_state
, !=, DB_SEARCH
);
96 if ((uintptr_t)d1
< (uintptr_t)d2
) {
99 if ((uintptr_t)d1
> (uintptr_t)d2
) {
107 dnode_cons(void *arg
, void *unused
, int kmflag
)
112 rw_init(&dn
->dn_struct_rwlock
, NULL
, RW_DEFAULT
, NULL
);
113 mutex_init(&dn
->dn_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
114 mutex_init(&dn
->dn_dbufs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
115 cv_init(&dn
->dn_notxholds
, NULL
, CV_DEFAULT
, NULL
);
118 * Every dbuf has a reference, and dropping a tracked reference is
119 * O(number of references), so don't track dn_holds.
121 refcount_create_untracked(&dn
->dn_holds
);
122 refcount_create(&dn
->dn_tx_holds
);
123 list_link_init(&dn
->dn_link
);
125 bzero(&dn
->dn_next_nblkptr
[0], sizeof (dn
->dn_next_nblkptr
));
126 bzero(&dn
->dn_next_nlevels
[0], sizeof (dn
->dn_next_nlevels
));
127 bzero(&dn
->dn_next_indblkshift
[0], sizeof (dn
->dn_next_indblkshift
));
128 bzero(&dn
->dn_next_bonustype
[0], sizeof (dn
->dn_next_bonustype
));
129 bzero(&dn
->dn_rm_spillblk
[0], sizeof (dn
->dn_rm_spillblk
));
130 bzero(&dn
->dn_next_bonuslen
[0], sizeof (dn
->dn_next_bonuslen
));
131 bzero(&dn
->dn_next_blksz
[0], sizeof (dn
->dn_next_blksz
));
133 for (i
= 0; i
< TXG_SIZE
; i
++) {
134 list_link_init(&dn
->dn_dirty_link
[i
]);
135 dn
->dn_free_ranges
[i
] = NULL
;
136 list_create(&dn
->dn_dirty_records
[i
],
137 sizeof (dbuf_dirty_record_t
),
138 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
141 dn
->dn_allocated_txg
= 0;
143 dn
->dn_assigned_txg
= 0;
145 dn
->dn_dirtyctx_firstset
= NULL
;
147 dn
->dn_have_spill
= B_FALSE
;
157 dn
->dn_dbufs_count
= 0;
158 dn
->dn_unlisted_l0_blkid
= 0;
159 avl_create(&dn
->dn_dbufs
, dbuf_compare
, sizeof (dmu_buf_impl_t
),
160 offsetof(dmu_buf_impl_t
, db_link
));
168 dnode_dest(void *arg
, void *unused
)
173 rw_destroy(&dn
->dn_struct_rwlock
);
174 mutex_destroy(&dn
->dn_mtx
);
175 mutex_destroy(&dn
->dn_dbufs_mtx
);
176 cv_destroy(&dn
->dn_notxholds
);
177 refcount_destroy(&dn
->dn_holds
);
178 refcount_destroy(&dn
->dn_tx_holds
);
179 ASSERT(!list_link_active(&dn
->dn_link
));
181 for (i
= 0; i
< TXG_SIZE
; i
++) {
182 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
183 ASSERT3P(dn
->dn_free_ranges
[i
], ==, NULL
);
184 list_destroy(&dn
->dn_dirty_records
[i
]);
185 ASSERT0(dn
->dn_next_nblkptr
[i
]);
186 ASSERT0(dn
->dn_next_nlevels
[i
]);
187 ASSERT0(dn
->dn_next_indblkshift
[i
]);
188 ASSERT0(dn
->dn_next_bonustype
[i
]);
189 ASSERT0(dn
->dn_rm_spillblk
[i
]);
190 ASSERT0(dn
->dn_next_bonuslen
[i
]);
191 ASSERT0(dn
->dn_next_blksz
[i
]);
194 ASSERT0(dn
->dn_allocated_txg
);
195 ASSERT0(dn
->dn_free_txg
);
196 ASSERT0(dn
->dn_assigned_txg
);
197 ASSERT0(dn
->dn_dirtyctx
);
198 ASSERT3P(dn
->dn_dirtyctx_firstset
, ==, NULL
);
199 ASSERT3P(dn
->dn_bonus
, ==, NULL
);
200 ASSERT(!dn
->dn_have_spill
);
201 ASSERT3P(dn
->dn_zio
, ==, NULL
);
202 ASSERT0(dn
->dn_oldused
);
203 ASSERT0(dn
->dn_oldflags
);
204 ASSERT0(dn
->dn_olduid
);
205 ASSERT0(dn
->dn_oldgid
);
206 ASSERT0(dn
->dn_newuid
);
207 ASSERT0(dn
->dn_newgid
);
208 ASSERT0(dn
->dn_id_flags
);
210 ASSERT0(dn
->dn_dbufs_count
);
211 ASSERT0(dn
->dn_unlisted_l0_blkid
);
212 avl_destroy(&dn
->dn_dbufs
);
218 ASSERT(dnode_cache
== NULL
);
219 dnode_cache
= kmem_cache_create("dnode_t", sizeof (dnode_t
),
220 0, dnode_cons
, dnode_dest
, NULL
, NULL
, NULL
, 0);
221 kmem_cache_set_move(dnode_cache
, dnode_move
);
227 kmem_cache_destroy(dnode_cache
);
234 dnode_verify(dnode_t
*dn
)
236 int drop_struct_lock
= FALSE
;
239 ASSERT(dn
->dn_objset
);
240 ASSERT(dn
->dn_handle
->dnh_dnode
== dn
);
242 ASSERT(DMU_OT_IS_VALID(dn
->dn_phys
->dn_type
));
244 if (!(zfs_flags
& ZFS_DEBUG_DNODE_VERIFY
))
247 if (!RW_WRITE_HELD(&dn
->dn_struct_rwlock
)) {
248 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
249 drop_struct_lock
= TRUE
;
251 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
|| dn
->dn_allocated_txg
!= 0) {
253 ASSERT3U(dn
->dn_indblkshift
, <=, SPA_MAXBLOCKSHIFT
);
254 if (dn
->dn_datablkshift
) {
255 ASSERT3U(dn
->dn_datablkshift
, >=, SPA_MINBLOCKSHIFT
);
256 ASSERT3U(dn
->dn_datablkshift
, <=, SPA_MAXBLOCKSHIFT
);
257 ASSERT3U(1<<dn
->dn_datablkshift
, ==, dn
->dn_datablksz
);
259 ASSERT3U(dn
->dn_nlevels
, <=, 30);
260 ASSERT(DMU_OT_IS_VALID(dn
->dn_type
));
261 ASSERT3U(dn
->dn_nblkptr
, >=, 1);
262 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
263 ASSERT3U(dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
264 ASSERT3U(dn
->dn_datablksz
, ==,
265 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
266 ASSERT3U(ISP2(dn
->dn_datablksz
), ==, dn
->dn_datablkshift
!= 0);
267 ASSERT3U((dn
->dn_nblkptr
- 1) * sizeof (blkptr_t
) +
268 dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
269 for (i
= 0; i
< TXG_SIZE
; i
++) {
270 ASSERT3U(dn
->dn_next_nlevels
[i
], <=, dn
->dn_nlevels
);
273 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
)
274 ASSERT3U(dn
->dn_phys
->dn_nlevels
, <=, dn
->dn_nlevels
);
275 ASSERT(DMU_OBJECT_IS_SPECIAL(dn
->dn_object
) || dn
->dn_dbuf
!= NULL
);
276 if (dn
->dn_dbuf
!= NULL
) {
277 ASSERT3P(dn
->dn_phys
, ==,
278 (dnode_phys_t
*)dn
->dn_dbuf
->db
.db_data
+
279 (dn
->dn_object
% (dn
->dn_dbuf
->db
.db_size
>> DNODE_SHIFT
)));
281 if (drop_struct_lock
)
282 rw_exit(&dn
->dn_struct_rwlock
);
287 dnode_byteswap(dnode_phys_t
*dnp
)
289 uint64_t *buf64
= (void*)&dnp
->dn_blkptr
;
292 if (dnp
->dn_type
== DMU_OT_NONE
) {
293 bzero(dnp
, sizeof (dnode_phys_t
));
297 dnp
->dn_datablkszsec
= BSWAP_16(dnp
->dn_datablkszsec
);
298 dnp
->dn_bonuslen
= BSWAP_16(dnp
->dn_bonuslen
);
299 dnp
->dn_maxblkid
= BSWAP_64(dnp
->dn_maxblkid
);
300 dnp
->dn_used
= BSWAP_64(dnp
->dn_used
);
303 * dn_nblkptr is only one byte, so it's OK to read it in either
304 * byte order. We can't read dn_bouslen.
306 ASSERT(dnp
->dn_indblkshift
<= SPA_MAXBLOCKSHIFT
);
307 ASSERT(dnp
->dn_nblkptr
<= DN_MAX_NBLKPTR
);
308 for (i
= 0; i
< dnp
->dn_nblkptr
* sizeof (blkptr_t
)/8; i
++)
309 buf64
[i
] = BSWAP_64(buf64
[i
]);
312 * OK to check dn_bonuslen for zero, because it won't matter if
313 * we have the wrong byte order. This is necessary because the
314 * dnode dnode is smaller than a regular dnode.
316 if (dnp
->dn_bonuslen
!= 0) {
318 * Note that the bonus length calculated here may be
319 * longer than the actual bonus buffer. This is because
320 * we always put the bonus buffer after the last block
321 * pointer (instead of packing it against the end of the
324 int off
= (dnp
->dn_nblkptr
-1) * sizeof (blkptr_t
);
325 size_t len
= DN_MAX_BONUSLEN
- off
;
326 dmu_object_byteswap_t byteswap
;
327 ASSERT(DMU_OT_IS_VALID(dnp
->dn_bonustype
));
328 byteswap
= DMU_OT_BYTESWAP(dnp
->dn_bonustype
);
329 dmu_ot_byteswap
[byteswap
].ob_func(dnp
->dn_bonus
+ off
, len
);
332 /* Swap SPILL block if we have one */
333 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)
334 byteswap_uint64_array(&dnp
->dn_spill
, sizeof (blkptr_t
));
339 dnode_buf_byteswap(void *vbuf
, size_t size
)
341 dnode_phys_t
*buf
= vbuf
;
344 ASSERT3U(sizeof (dnode_phys_t
), ==, (1<<DNODE_SHIFT
));
345 ASSERT((size
& (sizeof (dnode_phys_t
)-1)) == 0);
347 size
>>= DNODE_SHIFT
;
348 for (i
= 0; i
< size
; i
++) {
355 dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
)
357 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
359 dnode_setdirty(dn
, tx
);
360 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
361 ASSERT3U(newsize
, <=, DN_MAX_BONUSLEN
-
362 (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
));
363 dn
->dn_bonuslen
= newsize
;
365 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = DN_ZERO_BONUSLEN
;
367 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
368 rw_exit(&dn
->dn_struct_rwlock
);
372 dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t newtype
, dmu_tx_t
*tx
)
374 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
375 dnode_setdirty(dn
, tx
);
376 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
377 dn
->dn_bonustype
= newtype
;
378 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
379 rw_exit(&dn
->dn_struct_rwlock
);
383 dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
)
385 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
386 ASSERT(RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
387 dnode_setdirty(dn
, tx
);
388 dn
->dn_rm_spillblk
[tx
->tx_txg
&TXG_MASK
] = DN_KILL_SPILLBLK
;
389 dn
->dn_have_spill
= B_FALSE
;
393 dnode_setdblksz(dnode_t
*dn
, int size
)
395 ASSERT0(P2PHASE(size
, SPA_MINBLOCKSIZE
));
396 ASSERT3U(size
, <=, SPA_MAXBLOCKSIZE
);
397 ASSERT3U(size
, >=, SPA_MINBLOCKSIZE
);
398 ASSERT3U(size
>> SPA_MINBLOCKSHIFT
, <,
399 1<<(sizeof (dn
->dn_phys
->dn_datablkszsec
) * 8));
400 dn
->dn_datablksz
= size
;
401 dn
->dn_datablkszsec
= size
>> SPA_MINBLOCKSHIFT
;
402 dn
->dn_datablkshift
= ISP2(size
) ? highbit64(size
- 1) : 0;
406 dnode_create(objset_t
*os
, dnode_phys_t
*dnp
, dmu_buf_impl_t
*db
,
407 uint64_t object
, dnode_handle_t
*dnh
)
411 dn
= kmem_cache_alloc(dnode_cache
, KM_SLEEP
);
412 ASSERT(!POINTER_IS_VALID(dn
->dn_objset
));
416 * Defer setting dn_objset until the dnode is ready to be a candidate
417 * for the dnode_move() callback.
419 dn
->dn_object
= object
;
424 if (dnp
->dn_datablkszsec
) {
425 dnode_setdblksz(dn
, dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
427 dn
->dn_datablksz
= 0;
428 dn
->dn_datablkszsec
= 0;
429 dn
->dn_datablkshift
= 0;
431 dn
->dn_indblkshift
= dnp
->dn_indblkshift
;
432 dn
->dn_nlevels
= dnp
->dn_nlevels
;
433 dn
->dn_type
= dnp
->dn_type
;
434 dn
->dn_nblkptr
= dnp
->dn_nblkptr
;
435 dn
->dn_checksum
= dnp
->dn_checksum
;
436 dn
->dn_compress
= dnp
->dn_compress
;
437 dn
->dn_bonustype
= dnp
->dn_bonustype
;
438 dn
->dn_bonuslen
= dnp
->dn_bonuslen
;
439 dn
->dn_maxblkid
= dnp
->dn_maxblkid
;
440 dn
->dn_have_spill
= ((dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) != 0);
443 dmu_zfetch_init(&dn
->dn_zfetch
, dn
);
445 ASSERT(DMU_OT_IS_VALID(dn
->dn_phys
->dn_type
));
447 mutex_enter(&os
->os_lock
);
448 if (dnh
->dnh_dnode
!= NULL
) {
449 /* Lost the allocation race. */
450 mutex_exit(&os
->os_lock
);
451 kmem_cache_free(dnode_cache
, dn
);
452 return (dnh
->dnh_dnode
);
456 * Exclude special dnodes from os_dnodes so an empty os_dnodes
457 * signifies that the special dnodes have no references from
458 * their children (the entries in os_dnodes). This allows
459 * dnode_destroy() to easily determine if the last child has
460 * been removed and then complete eviction of the objset.
462 if (!DMU_OBJECT_IS_SPECIAL(object
))
463 list_insert_head(&os
->os_dnodes
, dn
);
467 * Everything else must be valid before assigning dn_objset
468 * makes the dnode eligible for dnode_move().
473 mutex_exit(&os
->os_lock
);
475 arc_space_consume(sizeof (dnode_t
), ARC_SPACE_OTHER
);
480 * Caller must be holding the dnode handle, which is released upon return.
483 dnode_destroy(dnode_t
*dn
)
485 objset_t
*os
= dn
->dn_objset
;
486 boolean_t complete_os_eviction
= B_FALSE
;
488 ASSERT((dn
->dn_id_flags
& DN_ID_NEW_EXIST
) == 0);
490 mutex_enter(&os
->os_lock
);
491 POINTER_INVALIDATE(&dn
->dn_objset
);
492 if (!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
493 list_remove(&os
->os_dnodes
, dn
);
494 complete_os_eviction
=
495 list_is_empty(&os
->os_dnodes
) &&
496 list_link_active(&os
->os_evicting_node
);
498 mutex_exit(&os
->os_lock
);
500 /* the dnode can no longer move, so we can release the handle */
501 zrl_remove(&dn
->dn_handle
->dnh_zrlock
);
503 dn
->dn_allocated_txg
= 0;
505 dn
->dn_assigned_txg
= 0;
508 if (dn
->dn_dirtyctx_firstset
!= NULL
) {
509 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
510 dn
->dn_dirtyctx_firstset
= NULL
;
512 if (dn
->dn_bonus
!= NULL
) {
513 mutex_enter(&dn
->dn_bonus
->db_mtx
);
514 dbuf_evict(dn
->dn_bonus
);
519 dn
->dn_have_spill
= B_FALSE
;
527 dn
->dn_unlisted_l0_blkid
= 0;
529 dmu_zfetch_rele(&dn
->dn_zfetch
);
530 kmem_cache_free(dnode_cache
, dn
);
531 arc_space_return(sizeof (dnode_t
), ARC_SPACE_OTHER
);
533 if (complete_os_eviction
)
534 dmu_objset_evict_done(os
);
538 dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
539 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
544 blocksize
= 1 << zfs_default_bs
;
545 else if (blocksize
> SPA_MAXBLOCKSIZE
)
546 blocksize
= SPA_MAXBLOCKSIZE
;
548 blocksize
= P2ROUNDUP(blocksize
, SPA_MINBLOCKSIZE
);
551 ibs
= zfs_default_ibs
;
553 ibs
= MIN(MAX(ibs
, DN_MIN_INDBLKSHIFT
), DN_MAX_INDBLKSHIFT
);
555 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn
->dn_objset
,
556 dn
->dn_object
, tx
->tx_txg
, blocksize
, ibs
);
558 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
559 ASSERT(bcmp(dn
->dn_phys
, &dnode_phys_zero
, sizeof (dnode_phys_t
)) == 0);
560 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
);
561 ASSERT(ot
!= DMU_OT_NONE
);
562 ASSERT(DMU_OT_IS_VALID(ot
));
563 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
564 (bonustype
== DMU_OT_SA
&& bonuslen
== 0) ||
565 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0));
566 ASSERT(DMU_OT_IS_VALID(bonustype
));
567 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
568 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
569 ASSERT0(dn
->dn_maxblkid
);
570 ASSERT0(dn
->dn_allocated_txg
);
571 ASSERT0(dn
->dn_assigned_txg
);
572 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
573 ASSERT3U(refcount_count(&dn
->dn_holds
), <=, 1);
574 ASSERT(avl_is_empty(&dn
->dn_dbufs
));
576 for (i
= 0; i
< TXG_SIZE
; i
++) {
577 ASSERT0(dn
->dn_next_nblkptr
[i
]);
578 ASSERT0(dn
->dn_next_nlevels
[i
]);
579 ASSERT0(dn
->dn_next_indblkshift
[i
]);
580 ASSERT0(dn
->dn_next_bonuslen
[i
]);
581 ASSERT0(dn
->dn_next_bonustype
[i
]);
582 ASSERT0(dn
->dn_rm_spillblk
[i
]);
583 ASSERT0(dn
->dn_next_blksz
[i
]);
584 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
585 ASSERT3P(list_head(&dn
->dn_dirty_records
[i
]), ==, NULL
);
586 ASSERT3P(dn
->dn_free_ranges
[i
], ==, NULL
);
590 dnode_setdblksz(dn
, blocksize
);
591 dn
->dn_indblkshift
= ibs
;
593 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
597 ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
598 dn
->dn_bonustype
= bonustype
;
599 dn
->dn_bonuslen
= bonuslen
;
600 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
601 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
605 if (dn
->dn_dirtyctx_firstset
) {
606 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
607 dn
->dn_dirtyctx_firstset
= NULL
;
610 dn
->dn_allocated_txg
= tx
->tx_txg
;
613 dnode_setdirty(dn
, tx
);
614 dn
->dn_next_indblkshift
[tx
->tx_txg
& TXG_MASK
] = ibs
;
615 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
616 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
617 dn
->dn_next_blksz
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_datablksz
;
621 dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
622 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
626 ASSERT3U(blocksize
, >=, SPA_MINBLOCKSIZE
);
627 ASSERT3U(blocksize
, <=, SPA_MAXBLOCKSIZE
);
628 ASSERT0(blocksize
% SPA_MINBLOCKSIZE
);
629 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
|| dmu_tx_private_ok(tx
));
630 ASSERT(tx
->tx_txg
!= 0);
631 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
632 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0) ||
633 (bonustype
== DMU_OT_SA
&& bonuslen
== 0));
634 ASSERT(DMU_OT_IS_VALID(bonustype
));
635 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
637 /* clean up any unreferenced dbufs */
638 dnode_evict_dbufs(dn
);
642 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
643 dnode_setdirty(dn
, tx
);
644 if (dn
->dn_datablksz
!= blocksize
) {
645 /* change blocksize */
646 ASSERT(dn
->dn_maxblkid
== 0 &&
647 (BP_IS_HOLE(&dn
->dn_phys
->dn_blkptr
[0]) ||
648 dnode_block_freed(dn
, 0)));
649 dnode_setdblksz(dn
, blocksize
);
650 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = blocksize
;
652 if (dn
->dn_bonuslen
!= bonuslen
)
653 dn
->dn_next_bonuslen
[tx
->tx_txg
&TXG_MASK
] = bonuslen
;
655 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
658 nblkptr
= 1 + ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
659 if (dn
->dn_bonustype
!= bonustype
)
660 dn
->dn_next_bonustype
[tx
->tx_txg
&TXG_MASK
] = bonustype
;
661 if (dn
->dn_nblkptr
!= nblkptr
)
662 dn
->dn_next_nblkptr
[tx
->tx_txg
&TXG_MASK
] = nblkptr
;
663 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
664 dbuf_rm_spill(dn
, tx
);
665 dnode_rm_spill(dn
, tx
);
667 rw_exit(&dn
->dn_struct_rwlock
);
672 /* change bonus size and type */
673 mutex_enter(&dn
->dn_mtx
);
674 dn
->dn_bonustype
= bonustype
;
675 dn
->dn_bonuslen
= bonuslen
;
676 dn
->dn_nblkptr
= nblkptr
;
677 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
678 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
679 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
681 /* fix up the bonus db_size */
683 dn
->dn_bonus
->db
.db_size
=
684 DN_MAX_BONUSLEN
- (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
);
685 ASSERT(dn
->dn_bonuslen
<= dn
->dn_bonus
->db
.db_size
);
688 dn
->dn_allocated_txg
= tx
->tx_txg
;
689 mutex_exit(&dn
->dn_mtx
);
695 uint64_t dms_dnode_invalid
;
696 uint64_t dms_dnode_recheck1
;
697 uint64_t dms_dnode_recheck2
;
698 uint64_t dms_dnode_special
;
699 uint64_t dms_dnode_handle
;
700 uint64_t dms_dnode_rwlock
;
701 uint64_t dms_dnode_active
;
703 #endif /* DNODE_STATS */
706 dnode_move_impl(dnode_t
*odn
, dnode_t
*ndn
)
710 ASSERT(!RW_LOCK_HELD(&odn
->dn_struct_rwlock
));
711 ASSERT(MUTEX_NOT_HELD(&odn
->dn_mtx
));
712 ASSERT(MUTEX_NOT_HELD(&odn
->dn_dbufs_mtx
));
713 ASSERT(!RW_LOCK_HELD(&odn
->dn_zfetch
.zf_rwlock
));
716 ndn
->dn_objset
= odn
->dn_objset
;
717 ndn
->dn_object
= odn
->dn_object
;
718 ndn
->dn_dbuf
= odn
->dn_dbuf
;
719 ndn
->dn_handle
= odn
->dn_handle
;
720 ndn
->dn_phys
= odn
->dn_phys
;
721 ndn
->dn_type
= odn
->dn_type
;
722 ndn
->dn_bonuslen
= odn
->dn_bonuslen
;
723 ndn
->dn_bonustype
= odn
->dn_bonustype
;
724 ndn
->dn_nblkptr
= odn
->dn_nblkptr
;
725 ndn
->dn_checksum
= odn
->dn_checksum
;
726 ndn
->dn_compress
= odn
->dn_compress
;
727 ndn
->dn_nlevels
= odn
->dn_nlevels
;
728 ndn
->dn_indblkshift
= odn
->dn_indblkshift
;
729 ndn
->dn_datablkshift
= odn
->dn_datablkshift
;
730 ndn
->dn_datablkszsec
= odn
->dn_datablkszsec
;
731 ndn
->dn_datablksz
= odn
->dn_datablksz
;
732 ndn
->dn_maxblkid
= odn
->dn_maxblkid
;
733 bcopy(&odn
->dn_next_nblkptr
[0], &ndn
->dn_next_nblkptr
[0],
734 sizeof (odn
->dn_next_nblkptr
));
735 bcopy(&odn
->dn_next_nlevels
[0], &ndn
->dn_next_nlevels
[0],
736 sizeof (odn
->dn_next_nlevels
));
737 bcopy(&odn
->dn_next_indblkshift
[0], &ndn
->dn_next_indblkshift
[0],
738 sizeof (odn
->dn_next_indblkshift
));
739 bcopy(&odn
->dn_next_bonustype
[0], &ndn
->dn_next_bonustype
[0],
740 sizeof (odn
->dn_next_bonustype
));
741 bcopy(&odn
->dn_rm_spillblk
[0], &ndn
->dn_rm_spillblk
[0],
742 sizeof (odn
->dn_rm_spillblk
));
743 bcopy(&odn
->dn_next_bonuslen
[0], &ndn
->dn_next_bonuslen
[0],
744 sizeof (odn
->dn_next_bonuslen
));
745 bcopy(&odn
->dn_next_blksz
[0], &ndn
->dn_next_blksz
[0],
746 sizeof (odn
->dn_next_blksz
));
747 for (i
= 0; i
< TXG_SIZE
; i
++) {
748 list_move_tail(&ndn
->dn_dirty_records
[i
],
749 &odn
->dn_dirty_records
[i
]);
751 bcopy(&odn
->dn_free_ranges
[0], &ndn
->dn_free_ranges
[0],
752 sizeof (odn
->dn_free_ranges
));
753 ndn
->dn_allocated_txg
= odn
->dn_allocated_txg
;
754 ndn
->dn_free_txg
= odn
->dn_free_txg
;
755 ndn
->dn_assigned_txg
= odn
->dn_assigned_txg
;
756 ndn
->dn_dirtyctx
= odn
->dn_dirtyctx
;
757 ndn
->dn_dirtyctx_firstset
= odn
->dn_dirtyctx_firstset
;
758 ASSERT(refcount_count(&odn
->dn_tx_holds
) == 0);
759 refcount_transfer(&ndn
->dn_holds
, &odn
->dn_holds
);
760 ASSERT(avl_is_empty(&ndn
->dn_dbufs
));
761 avl_swap(&ndn
->dn_dbufs
, &odn
->dn_dbufs
);
762 ndn
->dn_dbufs_count
= odn
->dn_dbufs_count
;
763 ndn
->dn_unlisted_l0_blkid
= odn
->dn_unlisted_l0_blkid
;
764 ndn
->dn_bonus
= odn
->dn_bonus
;
765 ndn
->dn_have_spill
= odn
->dn_have_spill
;
766 ndn
->dn_zio
= odn
->dn_zio
;
767 ndn
->dn_oldused
= odn
->dn_oldused
;
768 ndn
->dn_oldflags
= odn
->dn_oldflags
;
769 ndn
->dn_olduid
= odn
->dn_olduid
;
770 ndn
->dn_oldgid
= odn
->dn_oldgid
;
771 ndn
->dn_newuid
= odn
->dn_newuid
;
772 ndn
->dn_newgid
= odn
->dn_newgid
;
773 ndn
->dn_id_flags
= odn
->dn_id_flags
;
774 dmu_zfetch_init(&ndn
->dn_zfetch
, NULL
);
775 list_move_tail(&ndn
->dn_zfetch
.zf_stream
, &odn
->dn_zfetch
.zf_stream
);
776 ndn
->dn_zfetch
.zf_dnode
= odn
->dn_zfetch
.zf_dnode
;
777 ndn
->dn_zfetch
.zf_stream_cnt
= odn
->dn_zfetch
.zf_stream_cnt
;
778 ndn
->dn_zfetch
.zf_alloc_fail
= odn
->dn_zfetch
.zf_alloc_fail
;
781 * Update back pointers. Updating the handle fixes the back pointer of
782 * every descendant dbuf as well as the bonus dbuf.
784 ASSERT(ndn
->dn_handle
->dnh_dnode
== odn
);
785 ndn
->dn_handle
->dnh_dnode
= ndn
;
786 if (ndn
->dn_zfetch
.zf_dnode
== odn
) {
787 ndn
->dn_zfetch
.zf_dnode
= ndn
;
791 * Invalidate the original dnode by clearing all of its back pointers.
794 odn
->dn_handle
= NULL
;
795 avl_create(&odn
->dn_dbufs
, dbuf_compare
, sizeof (dmu_buf_impl_t
),
796 offsetof(dmu_buf_impl_t
, db_link
));
797 odn
->dn_dbufs_count
= 0;
798 odn
->dn_unlisted_l0_blkid
= 0;
799 odn
->dn_bonus
= NULL
;
800 odn
->dn_zfetch
.zf_dnode
= NULL
;
803 * Set the low bit of the objset pointer to ensure that dnode_move()
804 * recognizes the dnode as invalid in any subsequent callback.
806 POINTER_INVALIDATE(&odn
->dn_objset
);
809 * Satisfy the destructor.
811 for (i
= 0; i
< TXG_SIZE
; i
++) {
812 list_create(&odn
->dn_dirty_records
[i
],
813 sizeof (dbuf_dirty_record_t
),
814 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
815 odn
->dn_free_ranges
[i
] = NULL
;
816 odn
->dn_next_nlevels
[i
] = 0;
817 odn
->dn_next_indblkshift
[i
] = 0;
818 odn
->dn_next_bonustype
[i
] = 0;
819 odn
->dn_rm_spillblk
[i
] = 0;
820 odn
->dn_next_bonuslen
[i
] = 0;
821 odn
->dn_next_blksz
[i
] = 0;
823 odn
->dn_allocated_txg
= 0;
824 odn
->dn_free_txg
= 0;
825 odn
->dn_assigned_txg
= 0;
826 odn
->dn_dirtyctx
= 0;
827 odn
->dn_dirtyctx_firstset
= NULL
;
828 odn
->dn_have_spill
= B_FALSE
;
831 odn
->dn_oldflags
= 0;
836 odn
->dn_id_flags
= 0;
842 odn
->dn_moved
= (uint8_t)-1;
847 dnode_move(void *buf
, void *newbuf
, size_t size
, void *arg
)
849 dnode_t
*odn
= buf
, *ndn
= newbuf
;
855 * The dnode is on the objset's list of known dnodes if the objset
856 * pointer is valid. We set the low bit of the objset pointer when
857 * freeing the dnode to invalidate it, and the memory patterns written
858 * by kmem (baddcafe and deadbeef) set at least one of the two low bits.
859 * A newly created dnode sets the objset pointer last of all to indicate
860 * that the dnode is known and in a valid state to be moved by this
864 if (!POINTER_IS_VALID(os
)) {
865 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_invalid
);
866 return (KMEM_CBRC_DONT_KNOW
);
870 * Ensure that the objset does not go away during the move.
872 rw_enter(&os_lock
, RW_WRITER
);
873 if (os
!= odn
->dn_objset
) {
875 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck1
);
876 return (KMEM_CBRC_DONT_KNOW
);
880 * If the dnode is still valid, then so is the objset. We know that no
881 * valid objset can be freed while we hold os_lock, so we can safely
882 * ensure that the objset remains in use.
884 mutex_enter(&os
->os_lock
);
887 * Recheck the objset pointer in case the dnode was removed just before
888 * acquiring the lock.
890 if (os
!= odn
->dn_objset
) {
891 mutex_exit(&os
->os_lock
);
893 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck2
);
894 return (KMEM_CBRC_DONT_KNOW
);
898 * At this point we know that as long as we hold os->os_lock, the dnode
899 * cannot be freed and fields within the dnode can be safely accessed.
900 * The objset listing this dnode cannot go away as long as this dnode is
904 if (DMU_OBJECT_IS_SPECIAL(odn
->dn_object
)) {
905 mutex_exit(&os
->os_lock
);
906 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_special
);
907 return (KMEM_CBRC_NO
);
909 ASSERT(odn
->dn_dbuf
!= NULL
); /* only "special" dnodes have no parent */
912 * Lock the dnode handle to prevent the dnode from obtaining any new
913 * holds. This also prevents the descendant dbufs and the bonus dbuf
914 * from accessing the dnode, so that we can discount their holds. The
915 * handle is safe to access because we know that while the dnode cannot
916 * go away, neither can its handle. Once we hold dnh_zrlock, we can
917 * safely move any dnode referenced only by dbufs.
919 if (!zrl_tryenter(&odn
->dn_handle
->dnh_zrlock
)) {
920 mutex_exit(&os
->os_lock
);
921 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_handle
);
922 return (KMEM_CBRC_LATER
);
926 * Ensure a consistent view of the dnode's holds and the dnode's dbufs.
927 * We need to guarantee that there is a hold for every dbuf in order to
928 * determine whether the dnode is actively referenced. Falsely matching
929 * a dbuf to an active hold would lead to an unsafe move. It's possible
930 * that a thread already having an active dnode hold is about to add a
931 * dbuf, and we can't compare hold and dbuf counts while the add is in
934 if (!rw_tryenter(&odn
->dn_struct_rwlock
, RW_WRITER
)) {
935 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
936 mutex_exit(&os
->os_lock
);
937 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_rwlock
);
938 return (KMEM_CBRC_LATER
);
942 * A dbuf may be removed (evicted) without an active dnode hold. In that
943 * case, the dbuf count is decremented under the handle lock before the
944 * dbuf's hold is released. This order ensures that if we count the hold
945 * after the dbuf is removed but before its hold is released, we will
946 * treat the unmatched hold as active and exit safely. If we count the
947 * hold before the dbuf is removed, the hold is discounted, and the
948 * removal is blocked until the move completes.
950 refcount
= refcount_count(&odn
->dn_holds
);
951 ASSERT(refcount
>= 0);
952 dbufs
= odn
->dn_dbufs_count
;
954 /* We can't have more dbufs than dnode holds. */
955 ASSERT3U(dbufs
, <=, refcount
);
956 DTRACE_PROBE3(dnode__move
, dnode_t
*, odn
, int64_t, refcount
,
959 if (refcount
> dbufs
) {
960 rw_exit(&odn
->dn_struct_rwlock
);
961 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
962 mutex_exit(&os
->os_lock
);
963 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_active
);
964 return (KMEM_CBRC_LATER
);
967 rw_exit(&odn
->dn_struct_rwlock
);
970 * At this point we know that anyone with a hold on the dnode is not
971 * actively referencing it. The dnode is known and in a valid state to
972 * move. We're holding the locks needed to execute the critical section.
974 dnode_move_impl(odn
, ndn
);
976 list_link_replace(&odn
->dn_link
, &ndn
->dn_link
);
977 /* If the dnode was safe to move, the refcount cannot have changed. */
978 ASSERT(refcount
== refcount_count(&ndn
->dn_holds
));
979 ASSERT(dbufs
== ndn
->dn_dbufs_count
);
980 zrl_exit(&ndn
->dn_handle
->dnh_zrlock
); /* handle has moved */
981 mutex_exit(&os
->os_lock
);
983 return (KMEM_CBRC_YES
);
988 dnode_special_close(dnode_handle_t
*dnh
)
990 dnode_t
*dn
= dnh
->dnh_dnode
;
993 * Wait for final references to the dnode to clear. This can
994 * only happen if the arc is asyncronously evicting state that
995 * has a hold on this dnode while we are trying to evict this
998 while (refcount_count(&dn
->dn_holds
) > 0)
1000 ASSERT(dn
->dn_dbuf
== NULL
||
1001 dmu_buf_get_user(&dn
->dn_dbuf
->db
) == NULL
);
1002 zrl_add(&dnh
->dnh_zrlock
);
1003 dnode_destroy(dn
); /* implicit zrl_remove() */
1004 zrl_destroy(&dnh
->dnh_zrlock
);
1005 dnh
->dnh_dnode
= NULL
;
1009 dnode_special_open(objset_t
*os
, dnode_phys_t
*dnp
, uint64_t object
,
1010 dnode_handle_t
*dnh
)
1014 dn
= dnode_create(os
, dnp
, NULL
, object
, dnh
);
1015 zrl_init(&dnh
->dnh_zrlock
);
1020 dnode_buf_pageout(void *dbu
)
1022 dnode_children_t
*children_dnodes
= dbu
;
1025 for (i
= 0; i
< children_dnodes
->dnc_count
; i
++) {
1026 dnode_handle_t
*dnh
= &children_dnodes
->dnc_children
[i
];
1030 * The dnode handle lock guards against the dnode moving to
1031 * another valid address, so there is no need here to guard
1032 * against changes to or from NULL.
1034 if (dnh
->dnh_dnode
== NULL
) {
1035 zrl_destroy(&dnh
->dnh_zrlock
);
1039 zrl_add(&dnh
->dnh_zrlock
);
1040 dn
= dnh
->dnh_dnode
;
1042 * If there are holds on this dnode, then there should
1043 * be holds on the dnode's containing dbuf as well; thus
1044 * it wouldn't be eligible for eviction and this function
1045 * would not have been called.
1047 ASSERT(refcount_is_zero(&dn
->dn_holds
));
1048 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
1050 dnode_destroy(dn
); /* implicit zrl_remove() */
1051 zrl_destroy(&dnh
->dnh_zrlock
);
1052 dnh
->dnh_dnode
= NULL
;
1054 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
1055 children_dnodes
->dnc_count
* sizeof (dnode_handle_t
));
1060 * EINVAL - invalid object number.
1062 * succeeds even for free dnodes.
1065 dnode_hold_impl(objset_t
*os
, uint64_t object
, int flag
,
1066 void *tag
, dnode_t
**dnp
)
1069 int drop_struct_lock
= FALSE
;
1074 dnode_children_t
*children_dnodes
;
1075 dnode_handle_t
*dnh
;
1078 * If you are holding the spa config lock as writer, you shouldn't
1079 * be asking the DMU to do *anything* unless it's the root pool
1080 * which may require us to read from the root filesystem while
1081 * holding some (not all) of the locks as writer.
1083 ASSERT(spa_config_held(os
->os_spa
, SCL_ALL
, RW_WRITER
) == 0 ||
1084 (spa_is_root(os
->os_spa
) &&
1085 spa_config_held(os
->os_spa
, SCL_STATE
, RW_WRITER
)));
1087 if (object
== DMU_USERUSED_OBJECT
|| object
== DMU_GROUPUSED_OBJECT
) {
1088 dn
= (object
== DMU_USERUSED_OBJECT
) ?
1089 DMU_USERUSED_DNODE(os
) : DMU_GROUPUSED_DNODE(os
);
1091 return (SET_ERROR(ENOENT
));
1093 if ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
)
1094 return (SET_ERROR(ENOENT
));
1095 if ((flag
& DNODE_MUST_BE_FREE
) && type
!= DMU_OT_NONE
)
1096 return (SET_ERROR(EEXIST
));
1098 (void) refcount_add(&dn
->dn_holds
, tag
);
1103 if (object
== 0 || object
>= DN_MAX_OBJECT
)
1104 return (SET_ERROR(EINVAL
));
1106 mdn
= DMU_META_DNODE(os
);
1107 ASSERT(mdn
->dn_object
== DMU_META_DNODE_OBJECT
);
1111 if (!RW_WRITE_HELD(&mdn
->dn_struct_rwlock
)) {
1112 rw_enter(&mdn
->dn_struct_rwlock
, RW_READER
);
1113 drop_struct_lock
= TRUE
;
1116 blk
= dbuf_whichblock(mdn
, object
* sizeof (dnode_phys_t
));
1118 db
= dbuf_hold(mdn
, blk
, FTAG
);
1119 if (drop_struct_lock
)
1120 rw_exit(&mdn
->dn_struct_rwlock
);
1122 return (SET_ERROR(EIO
));
1123 err
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
);
1125 dbuf_rele(db
, FTAG
);
1129 ASSERT3U(db
->db
.db_size
, >=, 1<<DNODE_SHIFT
);
1130 epb
= db
->db
.db_size
>> DNODE_SHIFT
;
1132 idx
= object
& (epb
-1);
1134 ASSERT(DB_DNODE(db
)->dn_type
== DMU_OT_DNODE
);
1135 children_dnodes
= dmu_buf_get_user(&db
->db
);
1136 if (children_dnodes
== NULL
) {
1138 dnode_children_t
*winner
;
1139 children_dnodes
= kmem_zalloc(sizeof (dnode_children_t
) +
1140 epb
* sizeof (dnode_handle_t
), KM_SLEEP
);
1141 children_dnodes
->dnc_count
= epb
;
1142 dnh
= &children_dnodes
->dnc_children
[0];
1143 for (i
= 0; i
< epb
; i
++) {
1144 zrl_init(&dnh
[i
].dnh_zrlock
);
1146 dmu_buf_init_user(&children_dnodes
->dnc_dbu
,
1147 dnode_buf_pageout
, NULL
);
1148 winner
= dmu_buf_set_user(&db
->db
, &children_dnodes
->dnc_dbu
);
1149 if (winner
!= NULL
) {
1151 for (i
= 0; i
< epb
; i
++) {
1152 zrl_destroy(&dnh
[i
].dnh_zrlock
);
1155 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
1156 epb
* sizeof (dnode_handle_t
));
1157 children_dnodes
= winner
;
1160 ASSERT(children_dnodes
->dnc_count
== epb
);
1162 dnh
= &children_dnodes
->dnc_children
[idx
];
1163 zrl_add(&dnh
->dnh_zrlock
);
1164 dn
= dnh
->dnh_dnode
;
1166 dnode_phys_t
*phys
= (dnode_phys_t
*)db
->db
.db_data
+idx
;
1168 dn
= dnode_create(os
, phys
, db
, object
, dnh
);
1171 mutex_enter(&dn
->dn_mtx
);
1173 if (dn
->dn_free_txg
||
1174 ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
) ||
1175 ((flag
& DNODE_MUST_BE_FREE
) &&
1176 (type
!= DMU_OT_NONE
|| !refcount_is_zero(&dn
->dn_holds
)))) {
1177 mutex_exit(&dn
->dn_mtx
);
1178 zrl_remove(&dnh
->dnh_zrlock
);
1179 dbuf_rele(db
, FTAG
);
1180 return (type
== DMU_OT_NONE
? ENOENT
: EEXIST
);
1182 if (refcount_add(&dn
->dn_holds
, tag
) == 1)
1183 dbuf_add_ref(db
, dnh
);
1184 mutex_exit(&dn
->dn_mtx
);
1186 /* Now we can rely on the hold to prevent the dnode from moving. */
1187 zrl_remove(&dnh
->dnh_zrlock
);
1190 ASSERT3P(dn
->dn_dbuf
, ==, db
);
1191 ASSERT3U(dn
->dn_object
, ==, object
);
1192 dbuf_rele(db
, FTAG
);
1199 * Return held dnode if the object is allocated, NULL if not.
1202 dnode_hold(objset_t
*os
, uint64_t object
, void *tag
, dnode_t
**dnp
)
1204 return (dnode_hold_impl(os
, object
, DNODE_MUST_BE_ALLOCATED
, tag
, dnp
));
1208 * Can only add a reference if there is already at least one
1209 * reference on the dnode. Returns FALSE if unable to add a
1213 dnode_add_ref(dnode_t
*dn
, void *tag
)
1215 mutex_enter(&dn
->dn_mtx
);
1216 if (refcount_is_zero(&dn
->dn_holds
)) {
1217 mutex_exit(&dn
->dn_mtx
);
1220 VERIFY(1 < refcount_add(&dn
->dn_holds
, tag
));
1221 mutex_exit(&dn
->dn_mtx
);
1226 dnode_rele(dnode_t
*dn
, void *tag
)
1228 mutex_enter(&dn
->dn_mtx
);
1229 dnode_rele_and_unlock(dn
, tag
);
1233 dnode_rele_and_unlock(dnode_t
*dn
, void *tag
)
1236 /* Get while the hold prevents the dnode from moving. */
1237 dmu_buf_impl_t
*db
= dn
->dn_dbuf
;
1238 dnode_handle_t
*dnh
= dn
->dn_handle
;
1240 refs
= refcount_remove(&dn
->dn_holds
, tag
);
1241 mutex_exit(&dn
->dn_mtx
);
1244 * It's unsafe to release the last hold on a dnode by dnode_rele() or
1245 * indirectly by dbuf_rele() while relying on the dnode handle to
1246 * prevent the dnode from moving, since releasing the last hold could
1247 * result in the dnode's parent dbuf evicting its dnode handles. For
1248 * that reason anyone calling dnode_rele() or dbuf_rele() without some
1249 * other direct or indirect hold on the dnode must first drop the dnode
1252 ASSERT(refs
> 0 || dnh
->dnh_zrlock
.zr_owner
!= curthread
);
1254 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
1255 if (refs
== 0 && db
!= NULL
) {
1257 * Another thread could add a hold to the dnode handle in
1258 * dnode_hold_impl() while holding the parent dbuf. Since the
1259 * hold on the parent dbuf prevents the handle from being
1260 * destroyed, the hold on the handle is OK. We can't yet assert
1261 * that the handle has zero references, but that will be
1262 * asserted anyway when the handle gets destroyed.
1269 dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
)
1271 objset_t
*os
= dn
->dn_objset
;
1272 uint64_t txg
= tx
->tx_txg
;
1274 if (DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
1275 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1282 mutex_enter(&dn
->dn_mtx
);
1283 ASSERT(dn
->dn_phys
->dn_type
|| dn
->dn_allocated_txg
);
1284 ASSERT(dn
->dn_free_txg
== 0 || dn
->dn_free_txg
>= txg
);
1285 mutex_exit(&dn
->dn_mtx
);
1289 * Determine old uid/gid when necessary
1291 dmu_objset_userquota_get_ids(dn
, B_TRUE
, tx
);
1293 mutex_enter(&os
->os_lock
);
1296 * If we are already marked dirty, we're done.
1298 if (list_link_active(&dn
->dn_dirty_link
[txg
& TXG_MASK
])) {
1299 mutex_exit(&os
->os_lock
);
1303 ASSERT(!refcount_is_zero(&dn
->dn_holds
) ||
1304 !avl_is_empty(&dn
->dn_dbufs
));
1305 ASSERT(dn
->dn_datablksz
!= 0);
1306 ASSERT0(dn
->dn_next_bonuslen
[txg
&TXG_MASK
]);
1307 ASSERT0(dn
->dn_next_blksz
[txg
&TXG_MASK
]);
1308 ASSERT0(dn
->dn_next_bonustype
[txg
&TXG_MASK
]);
1310 dprintf_ds(os
->os_dsl_dataset
, "obj=%llu txg=%llu\n",
1311 dn
->dn_object
, txg
);
1313 if (dn
->dn_free_txg
> 0 && dn
->dn_free_txg
<= txg
) {
1314 list_insert_tail(&os
->os_free_dnodes
[txg
&TXG_MASK
], dn
);
1316 list_insert_tail(&os
->os_dirty_dnodes
[txg
&TXG_MASK
], dn
);
1319 mutex_exit(&os
->os_lock
);
1322 * The dnode maintains a hold on its containing dbuf as
1323 * long as there are holds on it. Each instantiated child
1324 * dbuf maintains a hold on the dnode. When the last child
1325 * drops its hold, the dnode will drop its hold on the
1326 * containing dbuf. We add a "dirty hold" here so that the
1327 * dnode will hang around after we finish processing its
1330 VERIFY(dnode_add_ref(dn
, (void *)(uintptr_t)tx
->tx_txg
));
1332 (void) dbuf_dirty(dn
->dn_dbuf
, tx
);
1334 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1338 dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
)
1340 int txgoff
= tx
->tx_txg
& TXG_MASK
;
1342 dprintf("dn=%p txg=%llu\n", dn
, tx
->tx_txg
);
1344 /* we should be the only holder... hopefully */
1345 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */
1347 mutex_enter(&dn
->dn_mtx
);
1348 if (dn
->dn_type
== DMU_OT_NONE
|| dn
->dn_free_txg
) {
1349 mutex_exit(&dn
->dn_mtx
);
1352 dn
->dn_free_txg
= tx
->tx_txg
;
1353 mutex_exit(&dn
->dn_mtx
);
1356 * If the dnode is already dirty, it needs to be moved from
1357 * the dirty list to the free list.
1359 mutex_enter(&dn
->dn_objset
->os_lock
);
1360 if (list_link_active(&dn
->dn_dirty_link
[txgoff
])) {
1361 list_remove(&dn
->dn_objset
->os_dirty_dnodes
[txgoff
], dn
);
1362 list_insert_tail(&dn
->dn_objset
->os_free_dnodes
[txgoff
], dn
);
1363 mutex_exit(&dn
->dn_objset
->os_lock
);
1365 mutex_exit(&dn
->dn_objset
->os_lock
);
1366 dnode_setdirty(dn
, tx
);
1371 * Try to change the block size for the indicated dnode. This can only
1372 * succeed if there are no blocks allocated or dirty beyond first block
1375 dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
)
1381 size
= SPA_MINBLOCKSIZE
;
1382 if (size
> SPA_MAXBLOCKSIZE
)
1383 size
= SPA_MAXBLOCKSIZE
;
1385 size
= P2ROUNDUP(size
, SPA_MINBLOCKSIZE
);
1387 if (ibs
== dn
->dn_indblkshift
)
1390 if (size
>> SPA_MINBLOCKSHIFT
== dn
->dn_datablkszsec
&& ibs
== 0)
1393 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1395 /* Check for any allocated blocks beyond the first */
1396 if (dn
->dn_maxblkid
!= 0)
1399 mutex_enter(&dn
->dn_dbufs_mtx
);
1400 for (db
= avl_first(&dn
->dn_dbufs
); db
!= NULL
;
1401 db
= AVL_NEXT(&dn
->dn_dbufs
, db
)) {
1402 if (db
->db_blkid
!= 0 && db
->db_blkid
!= DMU_BONUS_BLKID
&&
1403 db
->db_blkid
!= DMU_SPILL_BLKID
) {
1404 mutex_exit(&dn
->dn_dbufs_mtx
);
1408 mutex_exit(&dn
->dn_dbufs_mtx
);
1410 if (ibs
&& dn
->dn_nlevels
!= 1)
1413 /* resize the old block */
1414 err
= dbuf_hold_impl(dn
, 0, 0, TRUE
, FTAG
, &db
);
1416 dbuf_new_size(db
, size
, tx
);
1417 else if (err
!= ENOENT
)
1420 dnode_setdblksz(dn
, size
);
1421 dnode_setdirty(dn
, tx
);
1422 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = size
;
1424 dn
->dn_indblkshift
= ibs
;
1425 dn
->dn_next_indblkshift
[tx
->tx_txg
&TXG_MASK
] = ibs
;
1427 /* rele after we have fixed the blocksize in the dnode */
1429 dbuf_rele(db
, FTAG
);
1431 rw_exit(&dn
->dn_struct_rwlock
);
1435 rw_exit(&dn
->dn_struct_rwlock
);
1436 return (SET_ERROR(ENOTSUP
));
1439 /* read-holding callers must not rely on the lock being continuously held */
1441 dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t have_read
)
1443 uint64_t txgoff
= tx
->tx_txg
& TXG_MASK
;
1444 int epbs
, new_nlevels
;
1447 ASSERT(blkid
!= DMU_BONUS_BLKID
);
1450 RW_READ_HELD(&dn
->dn_struct_rwlock
) :
1451 RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
1454 * if we have a read-lock, check to see if we need to do any work
1455 * before upgrading to a write-lock.
1458 if (blkid
<= dn
->dn_maxblkid
)
1461 if (!rw_tryupgrade(&dn
->dn_struct_rwlock
)) {
1462 rw_exit(&dn
->dn_struct_rwlock
);
1463 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1467 if (blkid
<= dn
->dn_maxblkid
)
1470 dn
->dn_maxblkid
= blkid
;
1473 * Compute the number of levels necessary to support the new maxblkid.
1476 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1477 for (sz
= dn
->dn_nblkptr
;
1478 sz
<= blkid
&& sz
>= dn
->dn_nblkptr
; sz
<<= epbs
)
1481 if (new_nlevels
> dn
->dn_nlevels
) {
1482 int old_nlevels
= dn
->dn_nlevels
;
1485 dbuf_dirty_record_t
*new, *dr
, *dr_next
;
1487 dn
->dn_nlevels
= new_nlevels
;
1489 ASSERT3U(new_nlevels
, >, dn
->dn_next_nlevels
[txgoff
]);
1490 dn
->dn_next_nlevels
[txgoff
] = new_nlevels
;
1492 /* dirty the left indirects */
1493 db
= dbuf_hold_level(dn
, old_nlevels
, 0, FTAG
);
1495 new = dbuf_dirty(db
, tx
);
1496 dbuf_rele(db
, FTAG
);
1498 /* transfer the dirty records to the new indirect */
1499 mutex_enter(&dn
->dn_mtx
);
1500 mutex_enter(&new->dt
.di
.dr_mtx
);
1501 list
= &dn
->dn_dirty_records
[txgoff
];
1502 for (dr
= list_head(list
); dr
; dr
= dr_next
) {
1503 dr_next
= list_next(&dn
->dn_dirty_records
[txgoff
], dr
);
1504 if (dr
->dr_dbuf
->db_level
!= new_nlevels
-1 &&
1505 dr
->dr_dbuf
->db_blkid
!= DMU_BONUS_BLKID
&&
1506 dr
->dr_dbuf
->db_blkid
!= DMU_SPILL_BLKID
) {
1507 ASSERT(dr
->dr_dbuf
->db_level
== old_nlevels
-1);
1508 list_remove(&dn
->dn_dirty_records
[txgoff
], dr
);
1509 list_insert_tail(&new->dt
.di
.dr_children
, dr
);
1510 dr
->dr_parent
= new;
1513 mutex_exit(&new->dt
.di
.dr_mtx
);
1514 mutex_exit(&dn
->dn_mtx
);
1519 rw_downgrade(&dn
->dn_struct_rwlock
);
1523 dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
)
1526 uint64_t blkoff
, blkid
, nblks
;
1527 int blksz
, blkshift
, head
, tail
;
1531 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1532 blksz
= dn
->dn_datablksz
;
1533 blkshift
= dn
->dn_datablkshift
;
1534 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1536 if (len
== DMU_OBJECT_END
) {
1537 len
= UINT64_MAX
- off
;
1542 * First, block align the region to free:
1545 head
= P2NPHASE(off
, blksz
);
1546 blkoff
= P2PHASE(off
, blksz
);
1547 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1550 ASSERT(dn
->dn_maxblkid
== 0);
1551 if (off
== 0 && len
>= blksz
) {
1553 * Freeing the whole block; fast-track this request.
1554 * Note that we won't dirty any indirect blocks,
1555 * which is fine because we will be freeing the entire
1556 * file and thus all indirect blocks will be freed
1557 * by free_children().
1562 } else if (off
>= blksz
) {
1563 /* Freeing past end-of-data */
1566 /* Freeing part of the block. */
1568 ASSERT3U(head
, >, 0);
1572 /* zero out any partial block data at the start of the range */
1574 ASSERT3U(blkoff
+ head
, ==, blksz
);
1577 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
), TRUE
,
1581 /* don't dirty if it isn't on disk and isn't dirty */
1582 if (db
->db_last_dirty
||
1583 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1584 rw_exit(&dn
->dn_struct_rwlock
);
1585 dmu_buf_will_dirty(&db
->db
, tx
);
1586 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1587 data
= db
->db
.db_data
;
1588 bzero(data
+ blkoff
, head
);
1590 dbuf_rele(db
, FTAG
);
1596 /* If the range was less than one block, we're done */
1600 /* If the remaining range is past end of file, we're done */
1601 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1604 ASSERT(ISP2(blksz
));
1608 tail
= P2PHASE(len
, blksz
);
1610 ASSERT0(P2PHASE(off
, blksz
));
1611 /* zero out any partial block data at the end of the range */
1615 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
+len
),
1616 TRUE
, FTAG
, &db
) == 0) {
1617 /* don't dirty if not on disk and not dirty */
1618 if (db
->db_last_dirty
||
1619 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1620 rw_exit(&dn
->dn_struct_rwlock
);
1621 dmu_buf_will_dirty(&db
->db
, tx
);
1622 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1623 bzero(db
->db
.db_data
, tail
);
1625 dbuf_rele(db
, FTAG
);
1630 /* If the range did not include a full block, we are done */
1634 ASSERT(IS_P2ALIGNED(off
, blksz
));
1635 ASSERT(trunc
|| IS_P2ALIGNED(len
, blksz
));
1636 blkid
= off
>> blkshift
;
1637 nblks
= len
>> blkshift
;
1642 * Dirty the first and last indirect blocks, as they (and/or their
1643 * parents) will need to be written out if they were only
1644 * partially freed. Interior indirect blocks will be themselves freed,
1645 * by free_children(), so they need not be dirtied. Note that these
1646 * interior blocks have already been prefetched by dmu_tx_hold_free().
1648 if (dn
->dn_nlevels
> 1) {
1649 uint64_t first
, last
;
1651 first
= blkid
>> epbs
;
1652 if ((db
= dbuf_hold_level(dn
, 1, first
, FTAG
))) {
1653 dmu_buf_will_dirty(&db
->db
, tx
);
1654 dbuf_rele(db
, FTAG
);
1657 last
= dn
->dn_maxblkid
>> epbs
;
1659 last
= (blkid
+ nblks
- 1) >> epbs
;
1660 if (last
> first
&& (db
= dbuf_hold_level(dn
, 1, last
, FTAG
))) {
1661 dmu_buf_will_dirty(&db
->db
, tx
);
1662 dbuf_rele(db
, FTAG
);
1668 * Add this range to the dnode range list.
1669 * We will finish up this free operation in the syncing phase.
1671 mutex_enter(&dn
->dn_mtx
);
1673 int txgoff
= tx
->tx_txg
& TXG_MASK
;
1674 if (dn
->dn_free_ranges
[txgoff
] == NULL
) {
1675 dn
->dn_free_ranges
[txgoff
] =
1676 range_tree_create(NULL
, NULL
, &dn
->dn_mtx
);
1678 range_tree_clear(dn
->dn_free_ranges
[txgoff
], blkid
, nblks
);
1679 range_tree_add(dn
->dn_free_ranges
[txgoff
], blkid
, nblks
);
1681 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1682 blkid
, nblks
, tx
->tx_txg
);
1683 mutex_exit(&dn
->dn_mtx
);
1685 dbuf_free_range(dn
, blkid
, blkid
+ nblks
- 1, tx
);
1686 dnode_setdirty(dn
, tx
);
1689 rw_exit(&dn
->dn_struct_rwlock
);
1693 dnode_spill_freed(dnode_t
*dn
)
1697 mutex_enter(&dn
->dn_mtx
);
1698 for (i
= 0; i
< TXG_SIZE
; i
++) {
1699 if (dn
->dn_rm_spillblk
[i
] == DN_KILL_SPILLBLK
)
1702 mutex_exit(&dn
->dn_mtx
);
1703 return (i
< TXG_SIZE
);
1706 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1708 dnode_block_freed(dnode_t
*dn
, uint64_t blkid
)
1710 void *dp
= spa_get_dsl(dn
->dn_objset
->os_spa
);
1713 if (blkid
== DMU_BONUS_BLKID
)
1717 * If we're in the process of opening the pool, dp will not be
1718 * set yet, but there shouldn't be anything dirty.
1723 if (dn
->dn_free_txg
)
1726 if (blkid
== DMU_SPILL_BLKID
)
1727 return (dnode_spill_freed(dn
));
1729 mutex_enter(&dn
->dn_mtx
);
1730 for (i
= 0; i
< TXG_SIZE
; i
++) {
1731 if (dn
->dn_free_ranges
[i
] != NULL
&&
1732 range_tree_contains(dn
->dn_free_ranges
[i
], blkid
, 1))
1735 mutex_exit(&dn
->dn_mtx
);
1736 return (i
< TXG_SIZE
);
1739 /* call from syncing context when we actually write/free space for this dnode */
1741 dnode_diduse_space(dnode_t
*dn
, int64_t delta
)
1744 dprintf_dnode(dn
, "dn=%p dnp=%p used=%llu delta=%lld\n",
1746 (u_longlong_t
)dn
->dn_phys
->dn_used
,
1749 mutex_enter(&dn
->dn_mtx
);
1750 space
= DN_USED_BYTES(dn
->dn_phys
);
1752 ASSERT3U(space
+ delta
, >=, space
); /* no overflow */
1754 ASSERT3U(space
, >=, -delta
); /* no underflow */
1757 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_DNODE_BYTES
) {
1758 ASSERT((dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) == 0);
1759 ASSERT0(P2PHASE(space
, 1<<DEV_BSHIFT
));
1760 dn
->dn_phys
->dn_used
= space
>> DEV_BSHIFT
;
1762 dn
->dn_phys
->dn_used
= space
;
1763 dn
->dn_phys
->dn_flags
|= DNODE_FLAG_USED_BYTES
;
1765 mutex_exit(&dn
->dn_mtx
);
1769 * Call when we think we're going to write/free space in open context to track
1770 * the amount of memory in use by the currently open txg.
1773 dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
)
1775 objset_t
*os
= dn
->dn_objset
;
1776 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1777 int64_t aspace
= spa_get_asize(os
->os_spa
, space
);
1780 dsl_dir_willuse_space(ds
->ds_dir
, aspace
, tx
);
1781 dsl_pool_dirty_space(dmu_tx_pool(tx
), space
, tx
);
1784 dmu_tx_willuse_space(tx
, aspace
);
1788 * Scans a block at the indicated "level" looking for a hole or data,
1789 * depending on 'flags'.
1791 * If level > 0, then we are scanning an indirect block looking at its
1792 * pointers. If level == 0, then we are looking at a block of dnodes.
1794 * If we don't find what we are looking for in the block, we return ESRCH.
1795 * Otherwise, return with *offset pointing to the beginning (if searching
1796 * forwards) or end (if searching backwards) of the range covered by the
1797 * block pointer we matched on (or dnode).
1799 * The basic search algorithm used below by dnode_next_offset() is to
1800 * use this function to search up the block tree (widen the search) until
1801 * we find something (i.e., we don't return ESRCH) and then search back
1802 * down the tree (narrow the search) until we reach our original search
1806 dnode_next_offset_level(dnode_t
*dn
, int flags
, uint64_t *offset
,
1807 int lvl
, uint64_t blkfill
, uint64_t txg
)
1809 dmu_buf_impl_t
*db
= NULL
;
1811 uint64_t epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1812 uint64_t epb
= 1ULL << epbs
;
1813 uint64_t minfill
, maxfill
;
1815 int i
, inc
, error
, span
;
1817 dprintf("probing object %llu offset %llx level %d of %u\n",
1818 dn
->dn_object
, *offset
, lvl
, dn
->dn_phys
->dn_nlevels
);
1820 hole
= ((flags
& DNODE_FIND_HOLE
) != 0);
1821 inc
= (flags
& DNODE_FIND_BACKWARDS
) ? -1 : 1;
1822 ASSERT(txg
== 0 || !hole
);
1824 if (lvl
== dn
->dn_phys
->dn_nlevels
) {
1826 epb
= dn
->dn_phys
->dn_nblkptr
;
1827 data
= dn
->dn_phys
->dn_blkptr
;
1829 uint64_t blkid
= dbuf_whichblock(dn
, *offset
) >> (epbs
* lvl
);
1830 error
= dbuf_hold_impl(dn
, lvl
, blkid
, TRUE
, FTAG
, &db
);
1832 if (error
!= ENOENT
)
1837 * This can only happen when we are searching up
1838 * the block tree for data. We don't really need to
1839 * adjust the offset, as we will just end up looking
1840 * at the pointer to this block in its parent, and its
1841 * going to be unallocated, so we will skip over it.
1843 return (SET_ERROR(ESRCH
));
1845 error
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
| DB_RF_HAVESTRUCT
);
1847 dbuf_rele(db
, FTAG
);
1850 data
= db
->db
.db_data
;
1854 if (db
!= NULL
&& txg
!= 0 && (db
->db_blkptr
== NULL
||
1855 db
->db_blkptr
->blk_birth
<= txg
||
1856 BP_IS_HOLE(db
->db_blkptr
))) {
1858 * This can only happen when we are searching up the tree
1859 * and these conditions mean that we need to keep climbing.
1861 error
= SET_ERROR(ESRCH
);
1862 } else if (lvl
== 0) {
1863 dnode_phys_t
*dnp
= data
;
1865 ASSERT(dn
->dn_type
== DMU_OT_DNODE
);
1867 for (i
= (*offset
>> span
) & (blkfill
- 1);
1868 i
>= 0 && i
< blkfill
; i
+= inc
) {
1869 if ((dnp
[i
].dn_type
== DMU_OT_NONE
) == hole
)
1871 *offset
+= (1ULL << span
) * inc
;
1873 if (i
< 0 || i
== blkfill
)
1874 error
= SET_ERROR(ESRCH
);
1876 blkptr_t
*bp
= data
;
1877 uint64_t start
= *offset
;
1878 span
= (lvl
- 1) * epbs
+ dn
->dn_datablkshift
;
1880 maxfill
= blkfill
<< ((lvl
- 1) * epbs
);
1887 *offset
= *offset
>> span
;
1888 for (i
= BF64_GET(*offset
, 0, epbs
);
1889 i
>= 0 && i
< epb
; i
+= inc
) {
1890 if (BP_GET_FILL(&bp
[i
]) >= minfill
&&
1891 BP_GET_FILL(&bp
[i
]) <= maxfill
&&
1892 (hole
|| bp
[i
].blk_birth
> txg
))
1894 if (inc
> 0 || *offset
> 0)
1897 *offset
= *offset
<< span
;
1899 /* traversing backwards; position offset at the end */
1900 ASSERT3U(*offset
, <=, start
);
1901 *offset
= MIN(*offset
+ (1ULL << span
) - 1, start
);
1902 } else if (*offset
< start
) {
1905 if (i
< 0 || i
>= epb
)
1906 error
= SET_ERROR(ESRCH
);
1910 dbuf_rele(db
, FTAG
);
1916 * Find the next hole, data, or sparse region at or after *offset.
1917 * The value 'blkfill' tells us how many items we expect to find
1918 * in an L0 data block; this value is 1 for normal objects,
1919 * DNODES_PER_BLOCK for the meta dnode, and some fraction of
1920 * DNODES_PER_BLOCK when searching for sparse regions thereof.
1924 * dnode_next_offset(dn, flags, offset, 1, 1, 0);
1925 * Finds the next/previous hole/data in a file.
1926 * Used in dmu_offset_next().
1928 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg);
1929 * Finds the next free/allocated dnode an objset's meta-dnode.
1930 * Only finds objects that have new contents since txg (ie.
1931 * bonus buffer changes and content removal are ignored).
1932 * Used in dmu_object_next().
1934 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0);
1935 * Finds the next L2 meta-dnode bp that's at most 1/4 full.
1936 * Used in dmu_object_alloc().
1939 dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *offset
,
1940 int minlvl
, uint64_t blkfill
, uint64_t txg
)
1942 uint64_t initial_offset
= *offset
;
1946 if (!(flags
& DNODE_FIND_HAVELOCK
))
1947 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1949 if (dn
->dn_phys
->dn_nlevels
== 0) {
1950 error
= SET_ERROR(ESRCH
);
1954 if (dn
->dn_datablkshift
== 0) {
1955 if (*offset
< dn
->dn_datablksz
) {
1956 if (flags
& DNODE_FIND_HOLE
)
1957 *offset
= dn
->dn_datablksz
;
1959 error
= SET_ERROR(ESRCH
);
1964 maxlvl
= dn
->dn_phys
->dn_nlevels
;
1966 for (lvl
= minlvl
; lvl
<= maxlvl
; lvl
++) {
1967 error
= dnode_next_offset_level(dn
,
1968 flags
, offset
, lvl
, blkfill
, txg
);
1973 while (error
== 0 && --lvl
>= minlvl
) {
1974 error
= dnode_next_offset_level(dn
,
1975 flags
, offset
, lvl
, blkfill
, txg
);
1979 * There's always a "virtual hole" at the end of the object, even
1980 * if all BP's which physically exist are non-holes.
1982 if ((flags
& DNODE_FIND_HOLE
) && error
== ESRCH
&& txg
== 0 &&
1983 minlvl
== 1 && blkfill
== 1 && !(flags
& DNODE_FIND_BACKWARDS
)) {
1987 if (error
== 0 && (flags
& DNODE_FIND_BACKWARDS
?
1988 initial_offset
< *offset
: initial_offset
> *offset
))
1989 error
= SET_ERROR(ESRCH
);
1991 if (!(flags
& DNODE_FIND_HAVELOCK
))
1992 rw_exit(&dn
->dn_struct_rwlock
);