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, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2017 RackTop Systems.
29 #include <sys/zfs_context.h>
31 #include <sys/dnode.h>
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/dsl_dataset.h>
40 #include <sys/dmu_zfetch.h>
41 #include <sys/range_tree.h>
43 static kmem_cache_t
*dnode_cache
;
45 * Define DNODE_STATS to turn on statistic gathering. By default, it is only
46 * turned on when DEBUG is also defined.
53 #define DNODE_STAT_ADD(stat) ((stat)++)
55 #define DNODE_STAT_ADD(stat) /* nothing */
56 #endif /* DNODE_STATS */
58 static dnode_phys_t dnode_phys_zero
;
60 int zfs_default_bs
= SPA_MINBLOCKSHIFT
;
61 int zfs_default_ibs
= DN_MAX_INDBLKSHIFT
;
64 static kmem_cbrc_t
dnode_move(void *, void *, size_t, void *);
68 dbuf_compare(const void *x1
, const void *x2
)
70 const dmu_buf_impl_t
*d1
= x1
;
71 const dmu_buf_impl_t
*d2
= x2
;
73 if (d1
->db_level
< d2
->db_level
) {
76 if (d1
->db_level
> d2
->db_level
) {
80 if (d1
->db_blkid
< d2
->db_blkid
) {
83 if (d1
->db_blkid
> d2
->db_blkid
) {
87 if (d1
->db_state
== DB_SEARCH
) {
88 ASSERT3S(d2
->db_state
, !=, DB_SEARCH
);
90 } else if (d2
->db_state
== DB_SEARCH
) {
91 ASSERT3S(d1
->db_state
, !=, DB_SEARCH
);
95 if ((uintptr_t)d1
< (uintptr_t)d2
) {
98 if ((uintptr_t)d1
> (uintptr_t)d2
) {
106 dnode_cons(void *arg
, void *unused
, int kmflag
)
111 rw_init(&dn
->dn_struct_rwlock
, NULL
, RW_DEFAULT
, NULL
);
112 mutex_init(&dn
->dn_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
113 mutex_init(&dn
->dn_dbufs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
114 cv_init(&dn
->dn_notxholds
, NULL
, CV_DEFAULT
, NULL
);
117 * Every dbuf has a reference, and dropping a tracked reference is
118 * O(number of references), so don't track dn_holds.
120 refcount_create_untracked(&dn
->dn_holds
);
121 refcount_create(&dn
->dn_tx_holds
);
122 list_link_init(&dn
->dn_link
);
124 bzero(&dn
->dn_next_nblkptr
[0], sizeof (dn
->dn_next_nblkptr
));
125 bzero(&dn
->dn_next_nlevels
[0], sizeof (dn
->dn_next_nlevels
));
126 bzero(&dn
->dn_next_indblkshift
[0], sizeof (dn
->dn_next_indblkshift
));
127 bzero(&dn
->dn_next_bonustype
[0], sizeof (dn
->dn_next_bonustype
));
128 bzero(&dn
->dn_rm_spillblk
[0], sizeof (dn
->dn_rm_spillblk
));
129 bzero(&dn
->dn_next_bonuslen
[0], sizeof (dn
->dn_next_bonuslen
));
130 bzero(&dn
->dn_next_blksz
[0], sizeof (dn
->dn_next_blksz
));
132 for (i
= 0; i
< TXG_SIZE
; i
++) {
133 list_link_init(&dn
->dn_dirty_link
[i
]);
134 dn
->dn_free_ranges
[i
] = NULL
;
135 list_create(&dn
->dn_dirty_records
[i
],
136 sizeof (dbuf_dirty_record_t
),
137 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
140 dn
->dn_allocated_txg
= 0;
142 dn
->dn_assigned_txg
= 0;
144 dn
->dn_dirtyctx_firstset
= NULL
;
146 dn
->dn_have_spill
= B_FALSE
;
156 dn
->dn_dbufs_count
= 0;
157 avl_create(&dn
->dn_dbufs
, dbuf_compare
, sizeof (dmu_buf_impl_t
),
158 offsetof(dmu_buf_impl_t
, db_link
));
166 dnode_dest(void *arg
, void *unused
)
171 rw_destroy(&dn
->dn_struct_rwlock
);
172 mutex_destroy(&dn
->dn_mtx
);
173 mutex_destroy(&dn
->dn_dbufs_mtx
);
174 cv_destroy(&dn
->dn_notxholds
);
175 refcount_destroy(&dn
->dn_holds
);
176 refcount_destroy(&dn
->dn_tx_holds
);
177 ASSERT(!list_link_active(&dn
->dn_link
));
179 for (i
= 0; i
< TXG_SIZE
; i
++) {
180 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
181 ASSERT3P(dn
->dn_free_ranges
[i
], ==, NULL
);
182 list_destroy(&dn
->dn_dirty_records
[i
]);
183 ASSERT0(dn
->dn_next_nblkptr
[i
]);
184 ASSERT0(dn
->dn_next_nlevels
[i
]);
185 ASSERT0(dn
->dn_next_indblkshift
[i
]);
186 ASSERT0(dn
->dn_next_bonustype
[i
]);
187 ASSERT0(dn
->dn_rm_spillblk
[i
]);
188 ASSERT0(dn
->dn_next_bonuslen
[i
]);
189 ASSERT0(dn
->dn_next_blksz
[i
]);
192 ASSERT0(dn
->dn_allocated_txg
);
193 ASSERT0(dn
->dn_free_txg
);
194 ASSERT0(dn
->dn_assigned_txg
);
195 ASSERT0(dn
->dn_dirtyctx
);
196 ASSERT3P(dn
->dn_dirtyctx_firstset
, ==, NULL
);
197 ASSERT3P(dn
->dn_bonus
, ==, NULL
);
198 ASSERT(!dn
->dn_have_spill
);
199 ASSERT3P(dn
->dn_zio
, ==, NULL
);
200 ASSERT0(dn
->dn_oldused
);
201 ASSERT0(dn
->dn_oldflags
);
202 ASSERT0(dn
->dn_olduid
);
203 ASSERT0(dn
->dn_oldgid
);
204 ASSERT0(dn
->dn_newuid
);
205 ASSERT0(dn
->dn_newgid
);
206 ASSERT0(dn
->dn_id_flags
);
208 ASSERT0(dn
->dn_dbufs_count
);
209 avl_destroy(&dn
->dn_dbufs
);
215 ASSERT(dnode_cache
== NULL
);
216 dnode_cache
= kmem_cache_create("dnode_t",
218 0, dnode_cons
, dnode_dest
, NULL
, NULL
, NULL
, 0);
220 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
, >=, 0);
254 ASSERT3U(dn
->dn_indblkshift
, <=, SPA_MAXBLOCKSHIFT
);
255 if (dn
->dn_datablkshift
) {
256 ASSERT3U(dn
->dn_datablkshift
, >=, SPA_MINBLOCKSHIFT
);
257 ASSERT3U(dn
->dn_datablkshift
, <=, SPA_MAXBLOCKSHIFT
);
258 ASSERT3U(1<<dn
->dn_datablkshift
, ==, dn
->dn_datablksz
);
260 ASSERT3U(dn
->dn_nlevels
, <=, 30);
261 ASSERT(DMU_OT_IS_VALID(dn
->dn_type
));
262 ASSERT3U(dn
->dn_nblkptr
, >=, 1);
263 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
264 ASSERT3U(dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
265 ASSERT3U(dn
->dn_datablksz
, ==,
266 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
267 ASSERT3U(ISP2(dn
->dn_datablksz
), ==, dn
->dn_datablkshift
!= 0);
268 ASSERT3U((dn
->dn_nblkptr
- 1) * sizeof (blkptr_t
) +
269 dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
270 for (i
= 0; i
< TXG_SIZE
; i
++) {
271 ASSERT3U(dn
->dn_next_nlevels
[i
], <=, dn
->dn_nlevels
);
274 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
)
275 ASSERT3U(dn
->dn_phys
->dn_nlevels
, <=, dn
->dn_nlevels
);
276 ASSERT(DMU_OBJECT_IS_SPECIAL(dn
->dn_object
) || dn
->dn_dbuf
!= NULL
);
277 if (dn
->dn_dbuf
!= NULL
) {
278 ASSERT3P(dn
->dn_phys
, ==,
279 (dnode_phys_t
*)dn
->dn_dbuf
->db
.db_data
+
280 (dn
->dn_object
% (dn
->dn_dbuf
->db
.db_size
>> DNODE_SHIFT
)));
282 if (drop_struct_lock
)
283 rw_exit(&dn
->dn_struct_rwlock
);
288 dnode_byteswap(dnode_phys_t
*dnp
)
290 uint64_t *buf64
= (void*)&dnp
->dn_blkptr
;
293 if (dnp
->dn_type
== DMU_OT_NONE
) {
294 bzero(dnp
, sizeof (dnode_phys_t
));
298 dnp
->dn_datablkszsec
= BSWAP_16(dnp
->dn_datablkszsec
);
299 dnp
->dn_bonuslen
= BSWAP_16(dnp
->dn_bonuslen
);
300 dnp
->dn_maxblkid
= BSWAP_64(dnp
->dn_maxblkid
);
301 dnp
->dn_used
= BSWAP_64(dnp
->dn_used
);
304 * dn_nblkptr is only one byte, so it's OK to read it in either
305 * byte order. We can't read dn_bouslen.
307 ASSERT(dnp
->dn_indblkshift
<= SPA_MAXBLOCKSHIFT
);
308 ASSERT(dnp
->dn_nblkptr
<= DN_MAX_NBLKPTR
);
309 for (i
= 0; i
< dnp
->dn_nblkptr
* sizeof (blkptr_t
)/8; i
++)
310 buf64
[i
] = BSWAP_64(buf64
[i
]);
313 * OK to check dn_bonuslen for zero, because it won't matter if
314 * we have the wrong byte order. This is necessary because the
315 * dnode dnode is smaller than a regular dnode.
317 if (dnp
->dn_bonuslen
!= 0) {
319 * Note that the bonus length calculated here may be
320 * longer than the actual bonus buffer. This is because
321 * we always put the bonus buffer after the last block
322 * pointer (instead of packing it against the end of the
325 int off
= (dnp
->dn_nblkptr
-1) * sizeof (blkptr_t
);
326 size_t len
= DN_MAX_BONUSLEN
- off
;
327 ASSERT(DMU_OT_IS_VALID(dnp
->dn_bonustype
));
328 dmu_object_byteswap_t byteswap
=
329 DMU_OT_BYTESWAP(dnp
->dn_bonustype
);
330 dmu_ot_byteswap
[byteswap
].ob_func(dnp
->dn_bonus
+ off
, len
);
333 /* Swap SPILL block if we have one */
334 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)
335 byteswap_uint64_array(&dnp
->dn_spill
, sizeof (blkptr_t
));
340 dnode_buf_byteswap(void *vbuf
, size_t size
)
342 dnode_phys_t
*buf
= vbuf
;
345 ASSERT3U(sizeof (dnode_phys_t
), ==, (1<<DNODE_SHIFT
));
346 ASSERT((size
& (sizeof (dnode_phys_t
)-1)) == 0);
348 size
>>= DNODE_SHIFT
;
349 for (i
= 0; i
< size
; i
++) {
356 dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
)
358 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
360 dnode_setdirty(dn
, tx
);
361 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
362 ASSERT3U(newsize
, <=, DN_MAX_BONUSLEN
-
363 (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
));
364 dn
->dn_bonuslen
= newsize
;
366 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = DN_ZERO_BONUSLEN
;
368 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
369 rw_exit(&dn
->dn_struct_rwlock
);
373 dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t newtype
, dmu_tx_t
*tx
)
375 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
376 dnode_setdirty(dn
, tx
);
377 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
378 dn
->dn_bonustype
= newtype
;
379 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
380 rw_exit(&dn
->dn_struct_rwlock
);
384 dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
)
386 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
387 ASSERT(RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
388 dnode_setdirty(dn
, tx
);
389 dn
->dn_rm_spillblk
[tx
->tx_txg
&TXG_MASK
] = DN_KILL_SPILLBLK
;
390 dn
->dn_have_spill
= B_FALSE
;
394 dnode_setdblksz(dnode_t
*dn
, int size
)
396 ASSERT0(P2PHASE(size
, SPA_MINBLOCKSIZE
));
397 ASSERT3U(size
, <=, SPA_MAXBLOCKSIZE
);
398 ASSERT3U(size
, >=, SPA_MINBLOCKSIZE
);
399 ASSERT3U(size
>> SPA_MINBLOCKSHIFT
, <,
400 1<<(sizeof (dn
->dn_phys
->dn_datablkszsec
) * 8));
401 dn
->dn_datablksz
= size
;
402 dn
->dn_datablkszsec
= size
>> SPA_MINBLOCKSHIFT
;
403 dn
->dn_datablkshift
= ISP2(size
) ? highbit64(size
- 1) : 0;
407 dnode_create(objset_t
*os
, dnode_phys_t
*dnp
, dmu_buf_impl_t
*db
,
408 uint64_t object
, dnode_handle_t
*dnh
)
412 dn
= kmem_cache_alloc(dnode_cache
, KM_SLEEP
);
414 ASSERT(!POINTER_IS_VALID(dn
->dn_objset
));
419 * Defer setting dn_objset until the dnode is ready to be a candidate
420 * for the dnode_move() callback.
422 dn
->dn_object
= object
;
427 if (dnp
->dn_datablkszsec
) {
428 dnode_setdblksz(dn
, dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
430 dn
->dn_datablksz
= 0;
431 dn
->dn_datablkszsec
= 0;
432 dn
->dn_datablkshift
= 0;
434 dn
->dn_indblkshift
= dnp
->dn_indblkshift
;
435 dn
->dn_nlevels
= dnp
->dn_nlevels
;
436 dn
->dn_type
= dnp
->dn_type
;
437 dn
->dn_nblkptr
= dnp
->dn_nblkptr
;
438 dn
->dn_checksum
= dnp
->dn_checksum
;
439 dn
->dn_compress
= dnp
->dn_compress
;
440 dn
->dn_bonustype
= dnp
->dn_bonustype
;
441 dn
->dn_bonuslen
= dnp
->dn_bonuslen
;
442 dn
->dn_maxblkid
= dnp
->dn_maxblkid
;
443 dn
->dn_have_spill
= ((dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) != 0);
446 dmu_zfetch_init(&dn
->dn_zfetch
, dn
);
448 ASSERT(DMU_OT_IS_VALID(dn
->dn_phys
->dn_type
));
450 mutex_enter(&os
->os_lock
);
451 if (dnh
->dnh_dnode
!= NULL
) {
452 /* Lost the allocation race. */
453 mutex_exit(&os
->os_lock
);
454 kmem_cache_free(dnode_cache
, dn
);
455 return (dnh
->dnh_dnode
);
459 * Exclude special dnodes from os_dnodes so an empty os_dnodes
460 * signifies that the special dnodes have no references from
461 * their children (the entries in os_dnodes). This allows
462 * dnode_destroy() to easily determine if the last child has
463 * been removed and then complete eviction of the objset.
465 if (!DMU_OBJECT_IS_SPECIAL(object
))
466 list_insert_head(&os
->os_dnodes
, dn
);
470 * Everything else must be valid before assigning dn_objset
471 * makes the dnode eligible for dnode_move().
476 mutex_exit(&os
->os_lock
);
478 arc_space_consume(sizeof (dnode_t
), ARC_SPACE_OTHER
);
483 * Caller must be holding the dnode handle, which is released upon return.
486 dnode_destroy(dnode_t
*dn
)
488 objset_t
*os
= dn
->dn_objset
;
489 boolean_t complete_os_eviction
= B_FALSE
;
491 ASSERT((dn
->dn_id_flags
& DN_ID_NEW_EXIST
) == 0);
493 mutex_enter(&os
->os_lock
);
494 POINTER_INVALIDATE(&dn
->dn_objset
);
495 if (!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
496 list_remove(&os
->os_dnodes
, dn
);
497 complete_os_eviction
=
498 list_is_empty(&os
->os_dnodes
) &&
499 list_link_active(&os
->os_evicting_node
);
501 mutex_exit(&os
->os_lock
);
503 /* the dnode can no longer move, so we can release the handle */
504 zrl_remove(&dn
->dn_handle
->dnh_zrlock
);
506 dn
->dn_allocated_txg
= 0;
508 dn
->dn_assigned_txg
= 0;
511 if (dn
->dn_dirtyctx_firstset
!= NULL
) {
512 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
513 dn
->dn_dirtyctx_firstset
= NULL
;
515 if (dn
->dn_bonus
!= NULL
) {
516 mutex_enter(&dn
->dn_bonus
->db_mtx
);
517 dbuf_destroy(dn
->dn_bonus
);
522 dn
->dn_have_spill
= B_FALSE
;
531 dmu_zfetch_fini(&dn
->dn_zfetch
);
532 kmem_cache_free(dnode_cache
, dn
);
533 arc_space_return(sizeof (dnode_t
), ARC_SPACE_OTHER
);
535 if (complete_os_eviction
)
536 dmu_objset_evict_done(os
);
540 dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
541 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
545 ASSERT3U(blocksize
, <=,
546 spa_maxblocksize(dmu_objset_spa(dn
->dn_objset
)));
548 blocksize
= 1 << zfs_default_bs
;
550 blocksize
= P2ROUNDUP(blocksize
, SPA_MINBLOCKSIZE
);
553 ibs
= zfs_default_ibs
;
555 ibs
= MIN(MAX(ibs
, DN_MIN_INDBLKSHIFT
), DN_MAX_INDBLKSHIFT
);
557 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn
->dn_objset
,
558 dn
->dn_object
, tx
->tx_txg
, blocksize
, ibs
);
560 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
561 ASSERT(bcmp(dn
->dn_phys
, &dnode_phys_zero
, sizeof (dnode_phys_t
)) == 0);
562 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
);
563 ASSERT(ot
!= DMU_OT_NONE
);
564 ASSERT(DMU_OT_IS_VALID(ot
));
565 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
566 (bonustype
== DMU_OT_SA
&& bonuslen
== 0) ||
567 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0));
568 ASSERT(DMU_OT_IS_VALID(bonustype
));
569 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
570 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
571 ASSERT0(dn
->dn_maxblkid
);
572 ASSERT0(dn
->dn_allocated_txg
);
573 ASSERT0(dn
->dn_assigned_txg
);
574 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
575 ASSERT3U(refcount_count(&dn
->dn_holds
), <=, 1);
576 ASSERT(avl_is_empty(&dn
->dn_dbufs
));
578 for (i
= 0; i
< TXG_SIZE
; i
++) {
579 ASSERT0(dn
->dn_next_nblkptr
[i
]);
580 ASSERT0(dn
->dn_next_nlevels
[i
]);
581 ASSERT0(dn
->dn_next_indblkshift
[i
]);
582 ASSERT0(dn
->dn_next_bonuslen
[i
]);
583 ASSERT0(dn
->dn_next_bonustype
[i
]);
584 ASSERT0(dn
->dn_rm_spillblk
[i
]);
585 ASSERT0(dn
->dn_next_blksz
[i
]);
586 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
587 ASSERT3P(list_head(&dn
->dn_dirty_records
[i
]), ==, NULL
);
588 ASSERT3P(dn
->dn_free_ranges
[i
], ==, NULL
);
592 dnode_setdblksz(dn
, blocksize
);
593 dn
->dn_indblkshift
= ibs
;
595 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
599 ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
600 dn
->dn_bonustype
= bonustype
;
601 dn
->dn_bonuslen
= bonuslen
;
602 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
603 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
607 if (dn
->dn_dirtyctx_firstset
) {
608 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
609 dn
->dn_dirtyctx_firstset
= NULL
;
612 dn
->dn_allocated_txg
= tx
->tx_txg
;
615 dnode_setdirty(dn
, tx
);
616 dn
->dn_next_indblkshift
[tx
->tx_txg
& TXG_MASK
] = ibs
;
617 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
618 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
619 dn
->dn_next_blksz
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_datablksz
;
623 dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
624 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
628 ASSERT3U(blocksize
, >=, SPA_MINBLOCKSIZE
);
629 ASSERT3U(blocksize
, <=,
630 spa_maxblocksize(dmu_objset_spa(dn
->dn_objset
)));
631 ASSERT0(blocksize
% SPA_MINBLOCKSIZE
);
632 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
|| dmu_tx_private_ok(tx
));
633 ASSERT(tx
->tx_txg
!= 0);
634 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
635 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0) ||
636 (bonustype
== DMU_OT_SA
&& bonuslen
== 0));
637 ASSERT(DMU_OT_IS_VALID(bonustype
));
638 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
640 /* clean up any unreferenced dbufs */
641 dnode_evict_dbufs(dn
);
645 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
646 dnode_setdirty(dn
, tx
);
647 if (dn
->dn_datablksz
!= blocksize
) {
648 /* change blocksize */
649 ASSERT(dn
->dn_maxblkid
== 0 &&
650 (BP_IS_HOLE(&dn
->dn_phys
->dn_blkptr
[0]) ||
651 dnode_block_freed(dn
, 0)));
652 dnode_setdblksz(dn
, blocksize
);
653 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = blocksize
;
655 if (dn
->dn_bonuslen
!= bonuslen
)
656 dn
->dn_next_bonuslen
[tx
->tx_txg
&TXG_MASK
] = bonuslen
;
658 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
661 nblkptr
= 1 + ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
662 if (dn
->dn_bonustype
!= bonustype
)
663 dn
->dn_next_bonustype
[tx
->tx_txg
&TXG_MASK
] = bonustype
;
664 if (dn
->dn_nblkptr
!= nblkptr
)
665 dn
->dn_next_nblkptr
[tx
->tx_txg
&TXG_MASK
] = nblkptr
;
666 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
667 dbuf_rm_spill(dn
, tx
);
668 dnode_rm_spill(dn
, tx
);
670 rw_exit(&dn
->dn_struct_rwlock
);
675 /* change bonus size and type */
676 mutex_enter(&dn
->dn_mtx
);
677 dn
->dn_bonustype
= bonustype
;
678 dn
->dn_bonuslen
= bonuslen
;
679 dn
->dn_nblkptr
= nblkptr
;
680 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
681 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
682 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
684 /* fix up the bonus db_size */
686 dn
->dn_bonus
->db
.db_size
=
687 DN_MAX_BONUSLEN
- (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
);
688 ASSERT(dn
->dn_bonuslen
<= dn
->dn_bonus
->db
.db_size
);
691 dn
->dn_allocated_txg
= tx
->tx_txg
;
692 mutex_exit(&dn
->dn_mtx
);
697 uint64_t dms_dnode_invalid
;
698 uint64_t dms_dnode_recheck1
;
699 uint64_t dms_dnode_recheck2
;
700 uint64_t dms_dnode_special
;
701 uint64_t dms_dnode_handle
;
702 uint64_t dms_dnode_rwlock
;
703 uint64_t dms_dnode_active
;
705 #endif /* DNODE_STATS */
709 dnode_move_impl(dnode_t
*odn
, dnode_t
*ndn
)
713 ASSERT(!RW_LOCK_HELD(&odn
->dn_struct_rwlock
));
714 ASSERT(MUTEX_NOT_HELD(&odn
->dn_mtx
));
715 ASSERT(MUTEX_NOT_HELD(&odn
->dn_dbufs_mtx
));
716 ASSERT(!RW_LOCK_HELD(&odn
->dn_zfetch
.zf_rwlock
));
719 ndn
->dn_objset
= odn
->dn_objset
;
720 ndn
->dn_object
= odn
->dn_object
;
721 ndn
->dn_dbuf
= odn
->dn_dbuf
;
722 ndn
->dn_handle
= odn
->dn_handle
;
723 ndn
->dn_phys
= odn
->dn_phys
;
724 ndn
->dn_type
= odn
->dn_type
;
725 ndn
->dn_bonuslen
= odn
->dn_bonuslen
;
726 ndn
->dn_bonustype
= odn
->dn_bonustype
;
727 ndn
->dn_nblkptr
= odn
->dn_nblkptr
;
728 ndn
->dn_checksum
= odn
->dn_checksum
;
729 ndn
->dn_compress
= odn
->dn_compress
;
730 ndn
->dn_nlevels
= odn
->dn_nlevels
;
731 ndn
->dn_indblkshift
= odn
->dn_indblkshift
;
732 ndn
->dn_datablkshift
= odn
->dn_datablkshift
;
733 ndn
->dn_datablkszsec
= odn
->dn_datablkszsec
;
734 ndn
->dn_datablksz
= odn
->dn_datablksz
;
735 ndn
->dn_maxblkid
= odn
->dn_maxblkid
;
736 bcopy(&odn
->dn_next_nblkptr
[0], &ndn
->dn_next_nblkptr
[0],
737 sizeof (odn
->dn_next_nblkptr
));
738 bcopy(&odn
->dn_next_nlevels
[0], &ndn
->dn_next_nlevels
[0],
739 sizeof (odn
->dn_next_nlevels
));
740 bcopy(&odn
->dn_next_indblkshift
[0], &ndn
->dn_next_indblkshift
[0],
741 sizeof (odn
->dn_next_indblkshift
));
742 bcopy(&odn
->dn_next_bonustype
[0], &ndn
->dn_next_bonustype
[0],
743 sizeof (odn
->dn_next_bonustype
));
744 bcopy(&odn
->dn_rm_spillblk
[0], &ndn
->dn_rm_spillblk
[0],
745 sizeof (odn
->dn_rm_spillblk
));
746 bcopy(&odn
->dn_next_bonuslen
[0], &ndn
->dn_next_bonuslen
[0],
747 sizeof (odn
->dn_next_bonuslen
));
748 bcopy(&odn
->dn_next_blksz
[0], &ndn
->dn_next_blksz
[0],
749 sizeof (odn
->dn_next_blksz
));
750 for (i
= 0; i
< TXG_SIZE
; i
++) {
751 list_move_tail(&ndn
->dn_dirty_records
[i
],
752 &odn
->dn_dirty_records
[i
]);
754 bcopy(&odn
->dn_free_ranges
[0], &ndn
->dn_free_ranges
[0],
755 sizeof (odn
->dn_free_ranges
));
756 ndn
->dn_allocated_txg
= odn
->dn_allocated_txg
;
757 ndn
->dn_free_txg
= odn
->dn_free_txg
;
758 ndn
->dn_assigned_txg
= odn
->dn_assigned_txg
;
759 ndn
->dn_dirtyctx
= odn
->dn_dirtyctx
;
760 ndn
->dn_dirtyctx_firstset
= odn
->dn_dirtyctx_firstset
;
761 ASSERT(refcount_count(&odn
->dn_tx_holds
) == 0);
762 refcount_transfer(&ndn
->dn_holds
, &odn
->dn_holds
);
763 ASSERT(avl_is_empty(&ndn
->dn_dbufs
));
764 avl_swap(&ndn
->dn_dbufs
, &odn
->dn_dbufs
);
765 ndn
->dn_dbufs_count
= odn
->dn_dbufs_count
;
766 ndn
->dn_bonus
= odn
->dn_bonus
;
767 ndn
->dn_have_spill
= odn
->dn_have_spill
;
768 ndn
->dn_zio
= odn
->dn_zio
;
769 ndn
->dn_oldused
= odn
->dn_oldused
;
770 ndn
->dn_oldflags
= odn
->dn_oldflags
;
771 ndn
->dn_olduid
= odn
->dn_olduid
;
772 ndn
->dn_oldgid
= odn
->dn_oldgid
;
773 ndn
->dn_newuid
= odn
->dn_newuid
;
774 ndn
->dn_newgid
= odn
->dn_newgid
;
775 ndn
->dn_id_flags
= odn
->dn_id_flags
;
776 dmu_zfetch_init(&ndn
->dn_zfetch
, NULL
);
777 list_move_tail(&ndn
->dn_zfetch
.zf_stream
, &odn
->dn_zfetch
.zf_stream
);
778 ndn
->dn_zfetch
.zf_dnode
= odn
->dn_zfetch
.zf_dnode
;
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_bonus
= NULL
;
799 odn
->dn_zfetch
.zf_dnode
= NULL
;
802 * Set the low bit of the objset pointer to ensure that dnode_move()
803 * recognizes the dnode as invalid in any subsequent callback.
805 POINTER_INVALIDATE(&odn
->dn_objset
);
808 * Satisfy the destructor.
810 for (i
= 0; i
< TXG_SIZE
; i
++) {
811 list_create(&odn
->dn_dirty_records
[i
],
812 sizeof (dbuf_dirty_record_t
),
813 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
814 odn
->dn_free_ranges
[i
] = NULL
;
815 odn
->dn_next_nlevels
[i
] = 0;
816 odn
->dn_next_indblkshift
[i
] = 0;
817 odn
->dn_next_bonustype
[i
] = 0;
818 odn
->dn_rm_spillblk
[i
] = 0;
819 odn
->dn_next_bonuslen
[i
] = 0;
820 odn
->dn_next_blksz
[i
] = 0;
822 odn
->dn_allocated_txg
= 0;
823 odn
->dn_free_txg
= 0;
824 odn
->dn_assigned_txg
= 0;
825 odn
->dn_dirtyctx
= 0;
826 odn
->dn_dirtyctx_firstset
= NULL
;
827 odn
->dn_have_spill
= B_FALSE
;
830 odn
->dn_oldflags
= 0;
835 odn
->dn_id_flags
= 0;
841 odn
->dn_moved
= (uint8_t)-1;
846 dnode_move(void *buf
, void *newbuf
, size_t size
, void *arg
)
848 dnode_t
*odn
= buf
, *ndn
= newbuf
;
854 * The dnode is on the objset's list of known dnodes if the objset
855 * pointer is valid. We set the low bit of the objset pointer when
856 * freeing the dnode to invalidate it, and the memory patterns written
857 * by kmem (baddcafe and deadbeef) set at least one of the two low bits.
858 * A newly created dnode sets the objset pointer last of all to indicate
859 * that the dnode is known and in a valid state to be moved by this
863 if (!POINTER_IS_VALID(os
)) {
864 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_invalid
);
865 return (KMEM_CBRC_DONT_KNOW
);
869 * Ensure that the objset does not go away during the move.
871 rw_enter(&os_lock
, RW_WRITER
);
872 if (os
!= odn
->dn_objset
) {
874 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck1
);
875 return (KMEM_CBRC_DONT_KNOW
);
879 * If the dnode is still valid, then so is the objset. We know that no
880 * valid objset can be freed while we hold os_lock, so we can safely
881 * ensure that the objset remains in use.
883 mutex_enter(&os
->os_lock
);
886 * Recheck the objset pointer in case the dnode was removed just before
887 * acquiring the lock.
889 if (os
!= odn
->dn_objset
) {
890 mutex_exit(&os
->os_lock
);
892 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_recheck2
);
893 return (KMEM_CBRC_DONT_KNOW
);
897 * At this point we know that as long as we hold os->os_lock, the dnode
898 * cannot be freed and fields within the dnode can be safely accessed.
899 * The objset listing this dnode cannot go away as long as this dnode is
903 if (DMU_OBJECT_IS_SPECIAL(odn
->dn_object
)) {
904 mutex_exit(&os
->os_lock
);
905 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_special
);
906 return (KMEM_CBRC_NO
);
908 ASSERT(odn
->dn_dbuf
!= NULL
); /* only "special" dnodes have no parent */
911 * Lock the dnode handle to prevent the dnode from obtaining any new
912 * holds. This also prevents the descendant dbufs and the bonus dbuf
913 * from accessing the dnode, so that we can discount their holds. The
914 * handle is safe to access because we know that while the dnode cannot
915 * go away, neither can its handle. Once we hold dnh_zrlock, we can
916 * safely move any dnode referenced only by dbufs.
918 if (!zrl_tryenter(&odn
->dn_handle
->dnh_zrlock
)) {
919 mutex_exit(&os
->os_lock
);
920 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_handle
);
921 return (KMEM_CBRC_LATER
);
925 * Ensure a consistent view of the dnode's holds and the dnode's dbufs.
926 * We need to guarantee that there is a hold for every dbuf in order to
927 * determine whether the dnode is actively referenced. Falsely matching
928 * a dbuf to an active hold would lead to an unsafe move. It's possible
929 * that a thread already having an active dnode hold is about to add a
930 * dbuf, and we can't compare hold and dbuf counts while the add is in
933 if (!rw_tryenter(&odn
->dn_struct_rwlock
, RW_WRITER
)) {
934 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
935 mutex_exit(&os
->os_lock
);
936 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_rwlock
);
937 return (KMEM_CBRC_LATER
);
941 * A dbuf may be removed (evicted) without an active dnode hold. In that
942 * case, the dbuf count is decremented under the handle lock before the
943 * dbuf's hold is released. This order ensures that if we count the hold
944 * after the dbuf is removed but before its hold is released, we will
945 * treat the unmatched hold as active and exit safely. If we count the
946 * hold before the dbuf is removed, the hold is discounted, and the
947 * removal is blocked until the move completes.
949 refcount
= refcount_count(&odn
->dn_holds
);
950 ASSERT(refcount
>= 0);
951 dbufs
= odn
->dn_dbufs_count
;
953 /* We can't have more dbufs than dnode holds. */
954 ASSERT3U(dbufs
, <=, refcount
);
955 DTRACE_PROBE3(dnode__move
, dnode_t
*, odn
, int64_t, refcount
,
958 if (refcount
> dbufs
) {
959 rw_exit(&odn
->dn_struct_rwlock
);
960 zrl_exit(&odn
->dn_handle
->dnh_zrlock
);
961 mutex_exit(&os
->os_lock
);
962 DNODE_STAT_ADD(dnode_move_stats
.dms_dnode_active
);
963 return (KMEM_CBRC_LATER
);
966 rw_exit(&odn
->dn_struct_rwlock
);
969 * At this point we know that anyone with a hold on the dnode is not
970 * actively referencing it. The dnode is known and in a valid state to
971 * move. We're holding the locks needed to execute the critical section.
973 dnode_move_impl(odn
, ndn
);
975 list_link_replace(&odn
->dn_link
, &ndn
->dn_link
);
976 /* If the dnode was safe to move, the refcount cannot have changed. */
977 ASSERT(refcount
== refcount_count(&ndn
->dn_holds
));
978 ASSERT(dbufs
== ndn
->dn_dbufs_count
);
979 zrl_exit(&ndn
->dn_handle
->dnh_zrlock
); /* handle has moved */
980 mutex_exit(&os
->os_lock
);
982 return (KMEM_CBRC_YES
);
987 dnode_special_close(dnode_handle_t
*dnh
)
989 dnode_t
*dn
= dnh
->dnh_dnode
;
992 * Wait for final references to the dnode to clear. This can
993 * only happen if the arc is asyncronously evicting state that
994 * has a hold on this dnode while we are trying to evict this
997 while (refcount_count(&dn
->dn_holds
) > 0)
999 ASSERT(dn
->dn_dbuf
== NULL
||
1000 dmu_buf_get_user(&dn
->dn_dbuf
->db
) == NULL
);
1001 zrl_add(&dnh
->dnh_zrlock
);
1002 dnode_destroy(dn
); /* implicit zrl_remove() */
1003 zrl_destroy(&dnh
->dnh_zrlock
);
1004 dnh
->dnh_dnode
= NULL
;
1008 dnode_special_open(objset_t
*os
, dnode_phys_t
*dnp
, uint64_t object
,
1009 dnode_handle_t
*dnh
)
1013 dn
= dnode_create(os
, dnp
, NULL
, object
, dnh
);
1014 zrl_init(&dnh
->dnh_zrlock
);
1019 dnode_buf_evict_async(void *dbu
)
1021 dnode_children_t
*children_dnodes
= dbu
;
1024 for (i
= 0; i
< children_dnodes
->dnc_count
; i
++) {
1025 dnode_handle_t
*dnh
= &children_dnodes
->dnc_children
[i
];
1029 * The dnode handle lock guards against the dnode moving to
1030 * another valid address, so there is no need here to guard
1031 * against changes to or from NULL.
1033 if (dnh
->dnh_dnode
== NULL
) {
1034 zrl_destroy(&dnh
->dnh_zrlock
);
1038 zrl_add(&dnh
->dnh_zrlock
);
1039 dn
= dnh
->dnh_dnode
;
1041 * If there are holds on this dnode, then there should
1042 * be holds on the dnode's containing dbuf as well; thus
1043 * it wouldn't be eligible for eviction and this function
1044 * would not have been called.
1046 ASSERT(refcount_is_zero(&dn
->dn_holds
));
1047 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
1049 dnode_destroy(dn
); /* implicit zrl_remove() */
1050 zrl_destroy(&dnh
->dnh_zrlock
);
1051 dnh
->dnh_dnode
= NULL
;
1053 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
1054 children_dnodes
->dnc_count
* sizeof (dnode_handle_t
));
1059 * EINVAL - invalid object number.
1061 * succeeds even for free dnodes.
1064 dnode_hold_impl(objset_t
*os
, uint64_t object
, int flag
,
1065 void *tag
, dnode_t
**dnp
)
1068 int drop_struct_lock
= FALSE
;
1073 dnode_children_t
*children_dnodes
;
1074 dnode_handle_t
*dnh
;
1077 * If you are holding the spa config lock as writer, you shouldn't
1078 * be asking the DMU to do *anything* unless it's the root pool
1079 * which may require us to read from the root filesystem while
1080 * holding some (not all) of the locks as writer.
1082 ASSERT(spa_config_held(os
->os_spa
, SCL_ALL
, RW_WRITER
) == 0 ||
1083 (spa_is_root(os
->os_spa
) &&
1084 spa_config_held(os
->os_spa
, SCL_STATE
, RW_WRITER
)));
1086 ASSERT((flag
& DNODE_MUST_BE_ALLOCATED
) || (flag
& DNODE_MUST_BE_FREE
));
1088 if (object
== DMU_USERUSED_OBJECT
|| object
== DMU_GROUPUSED_OBJECT
) {
1089 dn
= (object
== DMU_USERUSED_OBJECT
) ?
1090 DMU_USERUSED_DNODE(os
) : DMU_GROUPUSED_DNODE(os
);
1092 return (SET_ERROR(ENOENT
));
1094 if ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
)
1095 return (SET_ERROR(ENOENT
));
1096 if ((flag
& DNODE_MUST_BE_FREE
) && type
!= DMU_OT_NONE
)
1097 return (SET_ERROR(EEXIST
));
1099 (void) refcount_add(&dn
->dn_holds
, tag
);
1104 if (object
== 0 || object
>= DN_MAX_OBJECT
)
1105 return (SET_ERROR(EINVAL
));
1107 mdn
= DMU_META_DNODE(os
);
1108 ASSERT(mdn
->dn_object
== DMU_META_DNODE_OBJECT
);
1112 if (!RW_WRITE_HELD(&mdn
->dn_struct_rwlock
)) {
1113 rw_enter(&mdn
->dn_struct_rwlock
, RW_READER
);
1114 drop_struct_lock
= TRUE
;
1117 blk
= dbuf_whichblock(mdn
, 0, object
* sizeof (dnode_phys_t
));
1119 db
= dbuf_hold(mdn
, blk
, FTAG
);
1120 if (drop_struct_lock
)
1121 rw_exit(&mdn
->dn_struct_rwlock
);
1123 return (SET_ERROR(EIO
));
1124 err
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
);
1126 dbuf_rele(db
, FTAG
);
1130 ASSERT3U(db
->db
.db_size
, >=, 1<<DNODE_SHIFT
);
1131 epb
= db
->db
.db_size
>> DNODE_SHIFT
;
1133 idx
= object
& (epb
-1);
1135 ASSERT(DB_DNODE(db
)->dn_type
== DMU_OT_DNODE
);
1136 children_dnodes
= dmu_buf_get_user(&db
->db
);
1137 if (children_dnodes
== NULL
) {
1139 dnode_children_t
*winner
;
1140 children_dnodes
= kmem_zalloc(sizeof (dnode_children_t
) +
1141 epb
* sizeof (dnode_handle_t
), KM_SLEEP
);
1142 children_dnodes
->dnc_count
= epb
;
1143 dnh
= &children_dnodes
->dnc_children
[0];
1144 for (i
= 0; i
< epb
; i
++) {
1145 zrl_init(&dnh
[i
].dnh_zrlock
);
1147 dmu_buf_init_user(&children_dnodes
->dnc_dbu
, NULL
,
1148 dnode_buf_evict_async
, NULL
);
1149 winner
= dmu_buf_set_user(&db
->db
, &children_dnodes
->dnc_dbu
);
1150 if (winner
!= NULL
) {
1152 for (i
= 0; i
< epb
; i
++) {
1153 zrl_destroy(&dnh
[i
].dnh_zrlock
);
1156 kmem_free(children_dnodes
, sizeof (dnode_children_t
) +
1157 epb
* sizeof (dnode_handle_t
));
1158 children_dnodes
= winner
;
1161 ASSERT(children_dnodes
->dnc_count
== epb
);
1163 dnh
= &children_dnodes
->dnc_children
[idx
];
1164 zrl_add(&dnh
->dnh_zrlock
);
1165 dn
= dnh
->dnh_dnode
;
1167 dnode_phys_t
*phys
= (dnode_phys_t
*)db
->db
.db_data
+idx
;
1169 dn
= dnode_create(os
, phys
, db
, object
, dnh
);
1172 mutex_enter(&dn
->dn_mtx
);
1174 if (dn
->dn_free_txg
||
1175 ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
) ||
1176 ((flag
& DNODE_MUST_BE_FREE
) &&
1177 (type
!= DMU_OT_NONE
|| !refcount_is_zero(&dn
->dn_holds
)))) {
1178 mutex_exit(&dn
->dn_mtx
);
1179 zrl_remove(&dnh
->dnh_zrlock
);
1180 dbuf_rele(db
, FTAG
);
1181 return ((flag
& DNODE_MUST_BE_ALLOCATED
) ? ENOENT
: EEXIST
);
1183 if (refcount_add(&dn
->dn_holds
, tag
) == 1)
1184 dbuf_add_ref(db
, dnh
);
1185 mutex_exit(&dn
->dn_mtx
);
1187 /* Now we can rely on the hold to prevent the dnode from moving. */
1188 zrl_remove(&dnh
->dnh_zrlock
);
1191 ASSERT3P(dn
->dn_dbuf
, ==, db
);
1192 ASSERT3U(dn
->dn_object
, ==, object
);
1193 dbuf_rele(db
, FTAG
);
1200 * Return held dnode if the object is allocated, NULL if not.
1203 dnode_hold(objset_t
*os
, uint64_t object
, void *tag
, dnode_t
**dnp
)
1205 return (dnode_hold_impl(os
, object
, DNODE_MUST_BE_ALLOCATED
, tag
, dnp
));
1209 * Can only add a reference if there is already at least one
1210 * reference on the dnode. Returns FALSE if unable to add a
1214 dnode_add_ref(dnode_t
*dn
, void *tag
)
1216 mutex_enter(&dn
->dn_mtx
);
1217 if (refcount_is_zero(&dn
->dn_holds
)) {
1218 mutex_exit(&dn
->dn_mtx
);
1221 VERIFY(1 < refcount_add(&dn
->dn_holds
, tag
));
1222 mutex_exit(&dn
->dn_mtx
);
1227 dnode_rele(dnode_t
*dn
, void *tag
)
1229 mutex_enter(&dn
->dn_mtx
);
1230 dnode_rele_and_unlock(dn
, tag
);
1234 dnode_rele_and_unlock(dnode_t
*dn
, void *tag
)
1237 /* Get while the hold prevents the dnode from moving. */
1238 dmu_buf_impl_t
*db
= dn
->dn_dbuf
;
1239 dnode_handle_t
*dnh
= dn
->dn_handle
;
1241 refs
= refcount_remove(&dn
->dn_holds
, tag
);
1242 mutex_exit(&dn
->dn_mtx
);
1245 * It's unsafe to release the last hold on a dnode by dnode_rele() or
1246 * indirectly by dbuf_rele() while relying on the dnode handle to
1247 * prevent the dnode from moving, since releasing the last hold could
1248 * result in the dnode's parent dbuf evicting its dnode handles. For
1249 * that reason anyone calling dnode_rele() or dbuf_rele() without some
1250 * other direct or indirect hold on the dnode must first drop the dnode
1253 ASSERT(refs
> 0 || dnh
->dnh_zrlock
.zr_owner
!= curthread
);
1255 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
1256 if (refs
== 0 && db
!= NULL
) {
1258 * Another thread could add a hold to the dnode handle in
1259 * dnode_hold_impl() while holding the parent dbuf. Since the
1260 * hold on the parent dbuf prevents the handle from being
1261 * destroyed, the hold on the handle is OK. We can't yet assert
1262 * that the handle has zero references, but that will be
1263 * asserted anyway when the handle gets destroyed.
1270 dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
)
1272 objset_t
*os
= dn
->dn_objset
;
1273 uint64_t txg
= tx
->tx_txg
;
1275 if (DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
1276 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1283 mutex_enter(&dn
->dn_mtx
);
1284 ASSERT(dn
->dn_phys
->dn_type
|| dn
->dn_allocated_txg
);
1285 ASSERT(dn
->dn_free_txg
== 0 || dn
->dn_free_txg
>= txg
);
1286 mutex_exit(&dn
->dn_mtx
);
1290 * Determine old uid/gid when necessary
1292 dmu_objset_userquota_get_ids(dn
, B_TRUE
, tx
);
1294 multilist_t
*dirtylist
= os
->os_dirty_dnodes
[txg
& TXG_MASK
];
1295 multilist_sublist_t
*mls
= multilist_sublist_lock_obj(dirtylist
, dn
);
1298 * If we are already marked dirty, we're done.
1300 if (list_link_active(&dn
->dn_dirty_link
[txg
& TXG_MASK
])) {
1301 multilist_sublist_unlock(mls
);
1305 ASSERT(!refcount_is_zero(&dn
->dn_holds
) ||
1306 !avl_is_empty(&dn
->dn_dbufs
));
1307 ASSERT(dn
->dn_datablksz
!= 0);
1308 ASSERT0(dn
->dn_next_bonuslen
[txg
&TXG_MASK
]);
1309 ASSERT0(dn
->dn_next_blksz
[txg
&TXG_MASK
]);
1310 ASSERT0(dn
->dn_next_bonustype
[txg
&TXG_MASK
]);
1312 dprintf_ds(os
->os_dsl_dataset
, "obj=%llu txg=%llu\n",
1313 dn
->dn_object
, txg
);
1315 multilist_sublist_insert_head(mls
, dn
);
1317 multilist_sublist_unlock(mls
);
1320 * The dnode maintains a hold on its containing dbuf as
1321 * long as there are holds on it. Each instantiated child
1322 * dbuf maintains a hold on the dnode. When the last child
1323 * drops its hold, the dnode will drop its hold on the
1324 * containing dbuf. We add a "dirty hold" here so that the
1325 * dnode will hang around after we finish processing its
1328 VERIFY(dnode_add_ref(dn
, (void *)(uintptr_t)tx
->tx_txg
));
1330 (void) dbuf_dirty(dn
->dn_dbuf
, tx
);
1332 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
1336 dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
)
1338 mutex_enter(&dn
->dn_mtx
);
1339 if (dn
->dn_type
== DMU_OT_NONE
|| dn
->dn_free_txg
) {
1340 mutex_exit(&dn
->dn_mtx
);
1343 dn
->dn_free_txg
= tx
->tx_txg
;
1344 mutex_exit(&dn
->dn_mtx
);
1346 dnode_setdirty(dn
, tx
);
1350 * Try to change the block size for the indicated dnode. This can only
1351 * succeed if there are no blocks allocated or dirty beyond first block
1354 dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
)
1359 ASSERT3U(size
, <=, spa_maxblocksize(dmu_objset_spa(dn
->dn_objset
)));
1361 size
= SPA_MINBLOCKSIZE
;
1363 size
= P2ROUNDUP(size
, SPA_MINBLOCKSIZE
);
1365 if (ibs
== dn
->dn_indblkshift
)
1368 if (size
>> SPA_MINBLOCKSHIFT
== dn
->dn_datablkszsec
&& ibs
== 0)
1371 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1373 /* Check for any allocated blocks beyond the first */
1374 if (dn
->dn_maxblkid
!= 0)
1377 mutex_enter(&dn
->dn_dbufs_mtx
);
1378 for (db
= avl_first(&dn
->dn_dbufs
); db
!= NULL
;
1379 db
= AVL_NEXT(&dn
->dn_dbufs
, db
)) {
1380 if (db
->db_blkid
!= 0 && db
->db_blkid
!= DMU_BONUS_BLKID
&&
1381 db
->db_blkid
!= DMU_SPILL_BLKID
) {
1382 mutex_exit(&dn
->dn_dbufs_mtx
);
1386 mutex_exit(&dn
->dn_dbufs_mtx
);
1388 if (ibs
&& dn
->dn_nlevels
!= 1)
1391 /* resize the old block */
1392 err
= dbuf_hold_impl(dn
, 0, 0, TRUE
, FALSE
, FTAG
, &db
);
1394 dbuf_new_size(db
, size
, tx
);
1395 else if (err
!= ENOENT
)
1398 dnode_setdblksz(dn
, size
);
1399 dnode_setdirty(dn
, tx
);
1400 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = size
;
1402 dn
->dn_indblkshift
= ibs
;
1403 dn
->dn_next_indblkshift
[tx
->tx_txg
&TXG_MASK
] = ibs
;
1405 /* rele after we have fixed the blocksize in the dnode */
1407 dbuf_rele(db
, FTAG
);
1409 rw_exit(&dn
->dn_struct_rwlock
);
1413 rw_exit(&dn
->dn_struct_rwlock
);
1414 return (SET_ERROR(ENOTSUP
));
1417 /* read-holding callers must not rely on the lock being continuously held */
1419 dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t have_read
)
1421 uint64_t txgoff
= tx
->tx_txg
& TXG_MASK
;
1422 int epbs
, new_nlevels
;
1425 ASSERT(blkid
!= DMU_BONUS_BLKID
);
1428 RW_READ_HELD(&dn
->dn_struct_rwlock
) :
1429 RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
1432 * if we have a read-lock, check to see if we need to do any work
1433 * before upgrading to a write-lock.
1436 if (blkid
<= dn
->dn_maxblkid
)
1439 if (!rw_tryupgrade(&dn
->dn_struct_rwlock
)) {
1440 rw_exit(&dn
->dn_struct_rwlock
);
1441 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1445 if (blkid
<= dn
->dn_maxblkid
)
1448 dn
->dn_maxblkid
= blkid
;
1451 * Compute the number of levels necessary to support the new maxblkid.
1454 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1455 for (sz
= dn
->dn_nblkptr
;
1456 sz
<= blkid
&& sz
>= dn
->dn_nblkptr
; sz
<<= epbs
)
1459 if (new_nlevels
> dn
->dn_nlevels
) {
1460 int old_nlevels
= dn
->dn_nlevels
;
1463 dbuf_dirty_record_t
*new, *dr
, *dr_next
;
1465 dn
->dn_nlevels
= new_nlevels
;
1467 ASSERT3U(new_nlevels
, >, dn
->dn_next_nlevels
[txgoff
]);
1468 dn
->dn_next_nlevels
[txgoff
] = new_nlevels
;
1470 /* dirty the left indirects */
1471 db
= dbuf_hold_level(dn
, old_nlevels
, 0, FTAG
);
1473 new = dbuf_dirty(db
, tx
);
1474 dbuf_rele(db
, FTAG
);
1476 /* transfer the dirty records to the new indirect */
1477 mutex_enter(&dn
->dn_mtx
);
1478 mutex_enter(&new->dt
.di
.dr_mtx
);
1479 list
= &dn
->dn_dirty_records
[txgoff
];
1480 for (dr
= list_head(list
); dr
; dr
= dr_next
) {
1481 dr_next
= list_next(&dn
->dn_dirty_records
[txgoff
], dr
);
1482 if (dr
->dr_dbuf
->db_level
!= new_nlevels
-1 &&
1483 dr
->dr_dbuf
->db_blkid
!= DMU_BONUS_BLKID
&&
1484 dr
->dr_dbuf
->db_blkid
!= DMU_SPILL_BLKID
) {
1485 ASSERT(dr
->dr_dbuf
->db_level
== old_nlevels
-1);
1486 list_remove(&dn
->dn_dirty_records
[txgoff
], dr
);
1487 list_insert_tail(&new->dt
.di
.dr_children
, dr
);
1488 dr
->dr_parent
= new;
1491 mutex_exit(&new->dt
.di
.dr_mtx
);
1492 mutex_exit(&dn
->dn_mtx
);
1497 rw_downgrade(&dn
->dn_struct_rwlock
);
1501 dnode_dirty_l1(dnode_t
*dn
, uint64_t l1blkid
, dmu_tx_t
*tx
)
1503 dmu_buf_impl_t
*db
= dbuf_hold_level(dn
, 1, l1blkid
, FTAG
);
1505 dmu_buf_will_dirty(&db
->db
, tx
);
1506 dbuf_rele(db
, FTAG
);
1511 dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
)
1514 uint64_t blkoff
, blkid
, nblks
;
1515 int blksz
, blkshift
, head
, tail
;
1519 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1520 blksz
= dn
->dn_datablksz
;
1521 blkshift
= dn
->dn_datablkshift
;
1522 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1524 if (len
== DMU_OBJECT_END
) {
1525 len
= UINT64_MAX
- off
;
1530 * First, block align the region to free:
1533 head
= P2NPHASE(off
, blksz
);
1534 blkoff
= P2PHASE(off
, blksz
);
1535 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1538 ASSERT(dn
->dn_maxblkid
== 0);
1539 if (off
== 0 && len
>= blksz
) {
1541 * Freeing the whole block; fast-track this request.
1542 * Note that we won't dirty any indirect blocks,
1543 * which is fine because we will be freeing the entire
1544 * file and thus all indirect blocks will be freed
1545 * by free_children().
1550 } else if (off
>= blksz
) {
1551 /* Freeing past end-of-data */
1554 /* Freeing part of the block. */
1556 ASSERT3U(head
, >, 0);
1560 /* zero out any partial block data at the start of the range */
1562 ASSERT3U(blkoff
+ head
, ==, blksz
);
1565 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, 0, off
),
1566 TRUE
, FALSE
, FTAG
, &db
) == 0) {
1569 /* don't dirty if it isn't on disk and isn't dirty */
1570 if (db
->db_last_dirty
||
1571 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1572 rw_exit(&dn
->dn_struct_rwlock
);
1573 dmu_buf_will_dirty(&db
->db
, tx
);
1574 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1575 data
= db
->db
.db_data
;
1576 bzero(data
+ blkoff
, head
);
1578 dbuf_rele(db
, FTAG
);
1584 /* If the range was less than one block, we're done */
1588 /* If the remaining range is past end of file, we're done */
1589 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1592 ASSERT(ISP2(blksz
));
1596 tail
= P2PHASE(len
, blksz
);
1598 ASSERT0(P2PHASE(off
, blksz
));
1599 /* zero out any partial block data at the end of the range */
1603 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, 0, off
+len
),
1604 TRUE
, FALSE
, FTAG
, &db
) == 0) {
1605 /* don't dirty if not on disk and not dirty */
1606 if (db
->db_last_dirty
||
1607 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1608 rw_exit(&dn
->dn_struct_rwlock
);
1609 dmu_buf_will_dirty(&db
->db
, tx
);
1610 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1611 bzero(db
->db
.db_data
, tail
);
1613 dbuf_rele(db
, FTAG
);
1618 /* If the range did not include a full block, we are done */
1622 ASSERT(IS_P2ALIGNED(off
, blksz
));
1623 ASSERT(trunc
|| IS_P2ALIGNED(len
, blksz
));
1624 blkid
= off
>> blkshift
;
1625 nblks
= len
>> blkshift
;
1630 * Dirty all the indirect blocks in this range. Note that only
1631 * the first and last indirect blocks can actually be written
1632 * (if they were partially freed) -- they must be dirtied, even if
1633 * they do not exist on disk yet. The interior blocks will
1634 * be freed by free_children(), so they will not actually be written.
1635 * Even though these interior blocks will not be written, we
1636 * dirty them for two reasons:
1638 * - It ensures that the indirect blocks remain in memory until
1639 * syncing context. (They have already been prefetched by
1640 * dmu_tx_hold_free(), so we don't have to worry about reading
1641 * them serially here.)
1643 * - The dirty space accounting will put pressure on the txg sync
1644 * mechanism to begin syncing, and to delay transactions if there
1645 * is a large amount of freeing. Even though these indirect
1646 * blocks will not be written, we could need to write the same
1647 * amount of space if we copy the freed BPs into deadlists.
1649 if (dn
->dn_nlevels
> 1) {
1650 uint64_t first
, last
;
1652 first
= blkid
>> epbs
;
1653 dnode_dirty_l1(dn
, first
, tx
);
1655 last
= dn
->dn_maxblkid
>> epbs
;
1657 last
= (blkid
+ nblks
- 1) >> epbs
;
1659 dnode_dirty_l1(dn
, last
, tx
);
1661 int shift
= dn
->dn_datablkshift
+ dn
->dn_indblkshift
-
1663 for (uint64_t i
= first
+ 1; i
< last
; i
++) {
1665 * Set i to the blockid of the next non-hole
1666 * level-1 indirect block at or after i. Note
1667 * that dnode_next_offset() operates in terms of
1668 * level-0-equivalent bytes.
1670 uint64_t ibyte
= i
<< shift
;
1671 int err
= dnode_next_offset(dn
, DNODE_FIND_HAVELOCK
,
1678 * Normally we should not see an error, either
1679 * from dnode_next_offset() or dbuf_hold_level()
1680 * (except for ESRCH from dnode_next_offset).
1681 * If there is an i/o error, then when we read
1682 * this block in syncing context, it will use
1683 * ZIO_FLAG_MUSTSUCCEED, and thus hang/panic according
1684 * to the "failmode" property. dnode_next_offset()
1685 * doesn't have a flag to indicate MUSTSUCCEED.
1690 dnode_dirty_l1(dn
, i
, tx
);
1696 * Add this range to the dnode range list.
1697 * We will finish up this free operation in the syncing phase.
1699 mutex_enter(&dn
->dn_mtx
);
1700 int txgoff
= tx
->tx_txg
& TXG_MASK
;
1701 if (dn
->dn_free_ranges
[txgoff
] == NULL
) {
1702 dn
->dn_free_ranges
[txgoff
] = range_tree_create(NULL
, NULL
);
1704 range_tree_clear(dn
->dn_free_ranges
[txgoff
], blkid
, nblks
);
1705 range_tree_add(dn
->dn_free_ranges
[txgoff
], blkid
, nblks
);
1706 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1707 blkid
, nblks
, tx
->tx_txg
);
1708 mutex_exit(&dn
->dn_mtx
);
1710 dbuf_free_range(dn
, blkid
, blkid
+ nblks
- 1, tx
);
1711 dnode_setdirty(dn
, tx
);
1714 rw_exit(&dn
->dn_struct_rwlock
);
1718 dnode_spill_freed(dnode_t
*dn
)
1722 mutex_enter(&dn
->dn_mtx
);
1723 for (i
= 0; i
< TXG_SIZE
; i
++) {
1724 if (dn
->dn_rm_spillblk
[i
] == DN_KILL_SPILLBLK
)
1727 mutex_exit(&dn
->dn_mtx
);
1728 return (i
< TXG_SIZE
);
1731 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1733 dnode_block_freed(dnode_t
*dn
, uint64_t blkid
)
1735 void *dp
= spa_get_dsl(dn
->dn_objset
->os_spa
);
1738 if (blkid
== DMU_BONUS_BLKID
)
1742 * If we're in the process of opening the pool, dp will not be
1743 * set yet, but there shouldn't be anything dirty.
1748 if (dn
->dn_free_txg
)
1751 if (blkid
== DMU_SPILL_BLKID
)
1752 return (dnode_spill_freed(dn
));
1754 mutex_enter(&dn
->dn_mtx
);
1755 for (i
= 0; i
< TXG_SIZE
; i
++) {
1756 if (dn
->dn_free_ranges
[i
] != NULL
&&
1757 range_tree_contains(dn
->dn_free_ranges
[i
], blkid
, 1))
1760 mutex_exit(&dn
->dn_mtx
);
1761 return (i
< TXG_SIZE
);
1764 /* call from syncing context when we actually write/free space for this dnode */
1766 dnode_diduse_space(dnode_t
*dn
, int64_t delta
)
1769 dprintf_dnode(dn
, "dn=%p dnp=%p used=%llu delta=%lld\n",
1771 (u_longlong_t
)dn
->dn_phys
->dn_used
,
1774 mutex_enter(&dn
->dn_mtx
);
1775 space
= DN_USED_BYTES(dn
->dn_phys
);
1777 ASSERT3U(space
+ delta
, >=, space
); /* no overflow */
1779 ASSERT3U(space
, >=, -delta
); /* no underflow */
1782 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_DNODE_BYTES
) {
1783 ASSERT((dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) == 0);
1784 ASSERT0(P2PHASE(space
, 1<<DEV_BSHIFT
));
1785 dn
->dn_phys
->dn_used
= space
>> DEV_BSHIFT
;
1787 dn
->dn_phys
->dn_used
= space
;
1788 dn
->dn_phys
->dn_flags
|= DNODE_FLAG_USED_BYTES
;
1790 mutex_exit(&dn
->dn_mtx
);
1794 * Scans a block at the indicated "level" looking for a hole or data,
1795 * depending on 'flags'.
1797 * If level > 0, then we are scanning an indirect block looking at its
1798 * pointers. If level == 0, then we are looking at a block of dnodes.
1800 * If we don't find what we are looking for in the block, we return ESRCH.
1801 * Otherwise, return with *offset pointing to the beginning (if searching
1802 * forwards) or end (if searching backwards) of the range covered by the
1803 * block pointer we matched on (or dnode).
1805 * The basic search algorithm used below by dnode_next_offset() is to
1806 * use this function to search up the block tree (widen the search) until
1807 * we find something (i.e., we don't return ESRCH) and then search back
1808 * down the tree (narrow the search) until we reach our original search
1812 dnode_next_offset_level(dnode_t
*dn
, int flags
, uint64_t *offset
,
1813 int lvl
, uint64_t blkfill
, uint64_t txg
)
1815 dmu_buf_impl_t
*db
= NULL
;
1817 uint64_t epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1818 uint64_t epb
= 1ULL << epbs
;
1819 uint64_t minfill
, maxfill
;
1821 int i
, inc
, error
, span
;
1823 dprintf("probing object %llu offset %llx level %d of %u\n",
1824 dn
->dn_object
, *offset
, lvl
, dn
->dn_phys
->dn_nlevels
);
1826 hole
= ((flags
& DNODE_FIND_HOLE
) != 0);
1827 inc
= (flags
& DNODE_FIND_BACKWARDS
) ? -1 : 1;
1828 ASSERT(txg
== 0 || !hole
);
1830 if (lvl
== dn
->dn_phys
->dn_nlevels
) {
1832 epb
= dn
->dn_phys
->dn_nblkptr
;
1833 data
= dn
->dn_phys
->dn_blkptr
;
1835 uint64_t blkid
= dbuf_whichblock(dn
, lvl
, *offset
);
1836 error
= dbuf_hold_impl(dn
, lvl
, blkid
, TRUE
, FALSE
, FTAG
, &db
);
1838 if (error
!= ENOENT
)
1843 * This can only happen when we are searching up
1844 * the block tree for data. We don't really need to
1845 * adjust the offset, as we will just end up looking
1846 * at the pointer to this block in its parent, and its
1847 * going to be unallocated, so we will skip over it.
1849 return (SET_ERROR(ESRCH
));
1851 error
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
| DB_RF_HAVESTRUCT
);
1853 dbuf_rele(db
, FTAG
);
1856 data
= db
->db
.db_data
;
1860 if (db
!= NULL
&& txg
!= 0 && (db
->db_blkptr
== NULL
||
1861 db
->db_blkptr
->blk_birth
<= txg
||
1862 BP_IS_HOLE(db
->db_blkptr
))) {
1864 * This can only happen when we are searching up the tree
1865 * and these conditions mean that we need to keep climbing.
1867 error
= SET_ERROR(ESRCH
);
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
)
1880 error
= SET_ERROR(ESRCH
);
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_GET_FILL(&bp
[i
]) >= minfill
&&
1897 BP_GET_FILL(&bp
[i
]) <= 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
)
1912 error
= SET_ERROR(ESRCH
);
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) {
1956 error
= SET_ERROR(ESRCH
);
1960 if (dn
->dn_datablkshift
== 0) {
1961 if (*offset
< dn
->dn_datablksz
) {
1962 if (flags
& DNODE_FIND_HOLE
)
1963 *offset
= dn
->dn_datablksz
;
1965 error
= SET_ERROR(ESRCH
);
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
);
1985 * There's always a "virtual hole" at the end of the object, even
1986 * if all BP's which physically exist are non-holes.
1988 if ((flags
& DNODE_FIND_HOLE
) && error
== ESRCH
&& txg
== 0 &&
1989 minlvl
== 1 && blkfill
== 1 && !(flags
& DNODE_FIND_BACKWARDS
)) {
1993 if (error
== 0 && (flags
& DNODE_FIND_BACKWARDS
?
1994 initial_offset
< *offset
: initial_offset
> *offset
))
1995 error
= SET_ERROR(ESRCH
);
1997 if (!(flags
& DNODE_FIND_HAVELOCK
))
1998 rw_exit(&dn
->dn_struct_rwlock
);