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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #include <sys/zfs_context.h>
28 #include <sys/dnode.h>
30 #include <sys/dmu_impl.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_dataset.h>
37 #include <sys/dmu_zfetch.h>
39 static int free_range_compar(const void *node1
, const void *node2
);
41 static kmem_cache_t
*dnode_cache
;
43 static dnode_phys_t dnode_phys_zero
;
45 int zfs_default_bs
= SPA_MINBLOCKSHIFT
;
46 int zfs_default_ibs
= DN_MAX_INDBLKSHIFT
;
50 dnode_cons(void *arg
, void *unused
, int kmflag
)
54 bzero(dn
, sizeof (dnode_t
));
56 rw_init(&dn
->dn_struct_rwlock
, NULL
, RW_DEFAULT
, NULL
);
57 mutex_init(&dn
->dn_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
58 mutex_init(&dn
->dn_dbufs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
59 cv_init(&dn
->dn_notxholds
, NULL
, CV_DEFAULT
, NULL
);
61 refcount_create(&dn
->dn_holds
);
62 refcount_create(&dn
->dn_tx_holds
);
64 for (i
= 0; i
< TXG_SIZE
; i
++) {
65 avl_create(&dn
->dn_ranges
[i
], free_range_compar
,
66 sizeof (free_range_t
),
67 offsetof(struct free_range
, fr_node
));
68 list_create(&dn
->dn_dirty_records
[i
],
69 sizeof (dbuf_dirty_record_t
),
70 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
73 list_create(&dn
->dn_dbufs
, sizeof (dmu_buf_impl_t
),
74 offsetof(dmu_buf_impl_t
, db_link
));
81 dnode_dest(void *arg
, void *unused
)
86 rw_destroy(&dn
->dn_struct_rwlock
);
87 mutex_destroy(&dn
->dn_mtx
);
88 mutex_destroy(&dn
->dn_dbufs_mtx
);
89 cv_destroy(&dn
->dn_notxholds
);
90 refcount_destroy(&dn
->dn_holds
);
91 refcount_destroy(&dn
->dn_tx_holds
);
93 for (i
= 0; i
< TXG_SIZE
; i
++) {
94 avl_destroy(&dn
->dn_ranges
[i
]);
95 list_destroy(&dn
->dn_dirty_records
[i
]);
98 list_destroy(&dn
->dn_dbufs
);
104 dnode_cache
= kmem_cache_create("dnode_t",
106 0, dnode_cons
, dnode_dest
, NULL
, NULL
, NULL
, 0);
112 kmem_cache_destroy(dnode_cache
);
118 dnode_verify(dnode_t
*dn
)
120 int drop_struct_lock
= FALSE
;
123 ASSERT(dn
->dn_objset
);
125 ASSERT(dn
->dn_phys
->dn_type
< DMU_OT_NUMTYPES
);
127 if (!(zfs_flags
& ZFS_DEBUG_DNODE_VERIFY
))
130 if (!RW_WRITE_HELD(&dn
->dn_struct_rwlock
)) {
131 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
132 drop_struct_lock
= TRUE
;
134 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
|| dn
->dn_allocated_txg
!= 0) {
136 ASSERT3U(dn
->dn_indblkshift
, >=, 0);
137 ASSERT3U(dn
->dn_indblkshift
, <=, SPA_MAXBLOCKSHIFT
);
138 if (dn
->dn_datablkshift
) {
139 ASSERT3U(dn
->dn_datablkshift
, >=, SPA_MINBLOCKSHIFT
);
140 ASSERT3U(dn
->dn_datablkshift
, <=, SPA_MAXBLOCKSHIFT
);
141 ASSERT3U(1<<dn
->dn_datablkshift
, ==, dn
->dn_datablksz
);
143 ASSERT3U(dn
->dn_nlevels
, <=, 30);
144 ASSERT3U(dn
->dn_type
, <=, DMU_OT_NUMTYPES
);
145 ASSERT3U(dn
->dn_nblkptr
, >=, 1);
146 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
147 ASSERT3U(dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
148 ASSERT3U(dn
->dn_datablksz
, ==,
149 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
150 ASSERT3U(ISP2(dn
->dn_datablksz
), ==, dn
->dn_datablkshift
!= 0);
151 ASSERT3U((dn
->dn_nblkptr
- 1) * sizeof (blkptr_t
) +
152 dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
153 for (i
= 0; i
< TXG_SIZE
; i
++) {
154 ASSERT3U(dn
->dn_next_nlevels
[i
], <=, dn
->dn_nlevels
);
157 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
)
158 ASSERT3U(dn
->dn_phys
->dn_nlevels
, <=, dn
->dn_nlevels
);
159 ASSERT(DMU_OBJECT_IS_SPECIAL(dn
->dn_object
) || dn
->dn_dbuf
!= NULL
);
160 if (dn
->dn_dbuf
!= NULL
) {
161 ASSERT3P(dn
->dn_phys
, ==,
162 (dnode_phys_t
*)dn
->dn_dbuf
->db
.db_data
+
163 (dn
->dn_object
% (dn
->dn_dbuf
->db
.db_size
>> DNODE_SHIFT
)));
165 if (drop_struct_lock
)
166 rw_exit(&dn
->dn_struct_rwlock
);
171 dnode_byteswap(dnode_phys_t
*dnp
)
173 uint64_t *buf64
= (void*)&dnp
->dn_blkptr
;
176 if (dnp
->dn_type
== DMU_OT_NONE
) {
177 bzero(dnp
, sizeof (dnode_phys_t
));
181 dnp
->dn_datablkszsec
= BSWAP_16(dnp
->dn_datablkszsec
);
182 dnp
->dn_bonuslen
= BSWAP_16(dnp
->dn_bonuslen
);
183 dnp
->dn_maxblkid
= BSWAP_64(dnp
->dn_maxblkid
);
184 dnp
->dn_used
= BSWAP_64(dnp
->dn_used
);
187 * dn_nblkptr is only one byte, so it's OK to read it in either
188 * byte order. We can't read dn_bouslen.
190 ASSERT(dnp
->dn_indblkshift
<= SPA_MAXBLOCKSHIFT
);
191 ASSERT(dnp
->dn_nblkptr
<= DN_MAX_NBLKPTR
);
192 for (i
= 0; i
< dnp
->dn_nblkptr
* sizeof (blkptr_t
)/8; i
++)
193 buf64
[i
] = BSWAP_64(buf64
[i
]);
196 * OK to check dn_bonuslen for zero, because it won't matter if
197 * we have the wrong byte order. This is necessary because the
198 * dnode dnode is smaller than a regular dnode.
200 if (dnp
->dn_bonuslen
!= 0) {
202 * Note that the bonus length calculated here may be
203 * longer than the actual bonus buffer. This is because
204 * we always put the bonus buffer after the last block
205 * pointer (instead of packing it against the end of the
208 int off
= (dnp
->dn_nblkptr
-1) * sizeof (blkptr_t
);
209 size_t len
= DN_MAX_BONUSLEN
- off
;
210 ASSERT3U(dnp
->dn_bonustype
, <, DMU_OT_NUMTYPES
);
211 dmu_ot
[dnp
->dn_bonustype
].ot_byteswap(dnp
->dn_bonus
+ off
, len
);
216 dnode_buf_byteswap(void *vbuf
, size_t size
)
218 dnode_phys_t
*buf
= vbuf
;
221 ASSERT3U(sizeof (dnode_phys_t
), ==, (1<<DNODE_SHIFT
));
222 ASSERT((size
& (sizeof (dnode_phys_t
)-1)) == 0);
224 size
>>= DNODE_SHIFT
;
225 for (i
= 0; i
< size
; i
++) {
232 free_range_compar(const void *node1
, const void *node2
)
234 const free_range_t
*rp1
= node1
;
235 const free_range_t
*rp2
= node2
;
237 if (rp1
->fr_blkid
< rp2
->fr_blkid
)
239 else if (rp1
->fr_blkid
> rp2
->fr_blkid
)
245 dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
)
247 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
249 dnode_setdirty(dn
, tx
);
250 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
251 ASSERT3U(newsize
, <=, DN_MAX_BONUSLEN
-
252 (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
));
253 dn
->dn_bonuslen
= newsize
;
255 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = DN_ZERO_BONUSLEN
;
257 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
258 rw_exit(&dn
->dn_struct_rwlock
);
262 dnode_setdblksz(dnode_t
*dn
, int size
)
264 ASSERT3U(P2PHASE(size
, SPA_MINBLOCKSIZE
), ==, 0);
265 ASSERT3U(size
, <=, SPA_MAXBLOCKSIZE
);
266 ASSERT3U(size
, >=, SPA_MINBLOCKSIZE
);
267 ASSERT3U(size
>> SPA_MINBLOCKSHIFT
, <,
268 1<<(sizeof (dn
->dn_phys
->dn_datablkszsec
) * 8));
269 dn
->dn_datablksz
= size
;
270 dn
->dn_datablkszsec
= size
>> SPA_MINBLOCKSHIFT
;
271 dn
->dn_datablkshift
= ISP2(size
) ? highbit(size
- 1) : 0;
275 dnode_create(objset_impl_t
*os
, dnode_phys_t
*dnp
, dmu_buf_impl_t
*db
,
278 dnode_t
*dn
= kmem_cache_alloc(dnode_cache
, KM_SLEEP
);
279 (void) dnode_cons(dn
, NULL
, 0); /* XXX */
282 dn
->dn_object
= object
;
286 if (dnp
->dn_datablkszsec
)
287 dnode_setdblksz(dn
, dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
288 dn
->dn_indblkshift
= dnp
->dn_indblkshift
;
289 dn
->dn_nlevels
= dnp
->dn_nlevels
;
290 dn
->dn_type
= dnp
->dn_type
;
291 dn
->dn_nblkptr
= dnp
->dn_nblkptr
;
292 dn
->dn_checksum
= dnp
->dn_checksum
;
293 dn
->dn_compress
= dnp
->dn_compress
;
294 dn
->dn_bonustype
= dnp
->dn_bonustype
;
295 dn
->dn_bonuslen
= dnp
->dn_bonuslen
;
296 dn
->dn_maxblkid
= dnp
->dn_maxblkid
;
298 dmu_zfetch_init(&dn
->dn_zfetch
, dn
);
300 ASSERT(dn
->dn_phys
->dn_type
< DMU_OT_NUMTYPES
);
301 mutex_enter(&os
->os_lock
);
302 list_insert_head(&os
->os_dnodes
, dn
);
303 mutex_exit(&os
->os_lock
);
305 arc_space_consume(sizeof (dnode_t
), ARC_SPACE_OTHER
);
310 dnode_destroy(dnode_t
*dn
)
312 objset_impl_t
*os
= dn
->dn_objset
;
317 for (i
= 0; i
< TXG_SIZE
; i
++) {
318 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
319 ASSERT(NULL
== list_head(&dn
->dn_dirty_records
[i
]));
320 ASSERT(0 == avl_numnodes(&dn
->dn_ranges
[i
]));
322 ASSERT(NULL
== list_head(&dn
->dn_dbufs
));
324 ASSERT(dn
->dn_oldphys
== NULL
);
326 mutex_enter(&os
->os_lock
);
327 list_remove(&os
->os_dnodes
, dn
);
328 mutex_exit(&os
->os_lock
);
330 if (dn
->dn_dirtyctx_firstset
) {
331 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
332 dn
->dn_dirtyctx_firstset
= NULL
;
334 dmu_zfetch_rele(&dn
->dn_zfetch
);
336 mutex_enter(&dn
->dn_bonus
->db_mtx
);
337 dbuf_evict(dn
->dn_bonus
);
340 kmem_cache_free(dnode_cache
, dn
);
341 arc_space_return(sizeof (dnode_t
), ARC_SPACE_OTHER
);
345 dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
346 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
351 blocksize
= 1 << zfs_default_bs
;
352 else if (blocksize
> SPA_MAXBLOCKSIZE
)
353 blocksize
= SPA_MAXBLOCKSIZE
;
355 blocksize
= P2ROUNDUP(blocksize
, SPA_MINBLOCKSIZE
);
358 ibs
= zfs_default_ibs
;
360 ibs
= MIN(MAX(ibs
, DN_MIN_INDBLKSHIFT
), DN_MAX_INDBLKSHIFT
);
362 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn
->dn_objset
,
363 dn
->dn_object
, tx
->tx_txg
, blocksize
, ibs
);
365 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
366 ASSERT(bcmp(dn
->dn_phys
, &dnode_phys_zero
, sizeof (dnode_phys_t
)) == 0);
367 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
);
368 ASSERT(ot
!= DMU_OT_NONE
);
369 ASSERT3U(ot
, <, DMU_OT_NUMTYPES
);
370 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
371 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0));
372 ASSERT3U(bonustype
, <, DMU_OT_NUMTYPES
);
373 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
374 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
375 ASSERT3U(dn
->dn_maxblkid
, ==, 0);
376 ASSERT3U(dn
->dn_allocated_txg
, ==, 0);
377 ASSERT3U(dn
->dn_assigned_txg
, ==, 0);
378 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
379 ASSERT3U(refcount_count(&dn
->dn_holds
), <=, 1);
380 ASSERT3P(list_head(&dn
->dn_dbufs
), ==, NULL
);
382 for (i
= 0; i
< TXG_SIZE
; i
++) {
383 ASSERT3U(dn
->dn_next_nlevels
[i
], ==, 0);
384 ASSERT3U(dn
->dn_next_indblkshift
[i
], ==, 0);
385 ASSERT3U(dn
->dn_next_bonuslen
[i
], ==, 0);
386 ASSERT3U(dn
->dn_next_blksz
[i
], ==, 0);
387 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
388 ASSERT3P(list_head(&dn
->dn_dirty_records
[i
]), ==, NULL
);
389 ASSERT3U(avl_numnodes(&dn
->dn_ranges
[i
]), ==, 0);
393 dnode_setdblksz(dn
, blocksize
);
394 dn
->dn_indblkshift
= ibs
;
396 dn
->dn_nblkptr
= 1 + ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
397 dn
->dn_bonustype
= bonustype
;
398 dn
->dn_bonuslen
= bonuslen
;
399 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
400 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
404 if (dn
->dn_dirtyctx_firstset
) {
405 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
406 dn
->dn_dirtyctx_firstset
= NULL
;
409 dn
->dn_allocated_txg
= tx
->tx_txg
;
411 dnode_setdirty(dn
, tx
);
412 dn
->dn_next_indblkshift
[tx
->tx_txg
& TXG_MASK
] = ibs
;
413 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
414 dn
->dn_next_blksz
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_datablksz
;
418 dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
419 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
423 ASSERT3U(blocksize
, >=, SPA_MINBLOCKSIZE
);
424 ASSERT3U(blocksize
, <=, SPA_MAXBLOCKSIZE
);
425 ASSERT3U(blocksize
% SPA_MINBLOCKSIZE
, ==, 0);
426 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
|| dmu_tx_private_ok(tx
));
427 ASSERT(tx
->tx_txg
!= 0);
428 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
429 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0));
430 ASSERT3U(bonustype
, <, DMU_OT_NUMTYPES
);
431 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
433 /* clean up any unreferenced dbufs */
434 dnode_evict_dbufs(dn
);
436 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
437 dnode_setdirty(dn
, tx
);
438 if (dn
->dn_datablksz
!= blocksize
) {
439 /* change blocksize */
440 ASSERT(dn
->dn_maxblkid
== 0 &&
441 (BP_IS_HOLE(&dn
->dn_phys
->dn_blkptr
[0]) ||
442 dnode_block_freed(dn
, 0)));
443 dnode_setdblksz(dn
, blocksize
);
444 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = blocksize
;
446 if (dn
->dn_bonuslen
!= bonuslen
)
447 dn
->dn_next_bonuslen
[tx
->tx_txg
&TXG_MASK
] = bonuslen
;
448 nblkptr
= 1 + ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
449 if (dn
->dn_nblkptr
!= nblkptr
)
450 dn
->dn_next_nblkptr
[tx
->tx_txg
&TXG_MASK
] = nblkptr
;
451 rw_exit(&dn
->dn_struct_rwlock
);
456 /* change bonus size and type */
457 mutex_enter(&dn
->dn_mtx
);
458 dn
->dn_bonustype
= bonustype
;
459 dn
->dn_bonuslen
= bonuslen
;
460 dn
->dn_nblkptr
= nblkptr
;
461 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
462 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
463 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
465 /* fix up the bonus db_size */
467 dn
->dn_bonus
->db
.db_size
=
468 DN_MAX_BONUSLEN
- (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
);
469 ASSERT(dn
->dn_bonuslen
<= dn
->dn_bonus
->db
.db_size
);
472 dn
->dn_allocated_txg
= tx
->tx_txg
;
473 mutex_exit(&dn
->dn_mtx
);
477 dnode_special_close(dnode_t
*dn
)
480 * Wait for final references to the dnode to clear. This can
481 * only happen if the arc is asyncronously evicting state that
482 * has a hold on this dnode while we are trying to evict this
485 while (refcount_count(&dn
->dn_holds
) > 0)
491 dnode_special_open(objset_impl_t
*os
, dnode_phys_t
*dnp
, uint64_t object
)
493 dnode_t
*dn
= dnode_create(os
, dnp
, NULL
, object
);
499 dnode_buf_pageout(dmu_buf_t
*db
, void *arg
)
501 dnode_t
**children_dnodes
= arg
;
503 int epb
= db
->db_size
>> DNODE_SHIFT
;
505 for (i
= 0; i
< epb
; i
++) {
506 dnode_t
*dn
= children_dnodes
[i
];
513 * If there are holds on this dnode, then there should
514 * be holds on the dnode's containing dbuf as well; thus
515 * it wouldn't be eligable for eviction and this function
516 * would not have been called.
518 ASSERT(refcount_is_zero(&dn
->dn_holds
));
519 ASSERT(list_head(&dn
->dn_dbufs
) == NULL
);
520 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
522 for (n
= 0; n
< TXG_SIZE
; n
++)
523 ASSERT(!list_link_active(&dn
->dn_dirty_link
[n
]));
525 children_dnodes
[i
] = NULL
;
528 kmem_free(children_dnodes
, epb
* sizeof (dnode_t
*));
533 * EINVAL - invalid object number.
535 * succeeds even for free dnodes.
538 dnode_hold_impl(objset_impl_t
*os
, uint64_t object
, int flag
,
539 void *tag
, dnode_t
**dnp
)
542 int drop_struct_lock
= FALSE
;
547 dnode_t
**children_dnodes
;
550 * If you are holding the spa config lock as writer, you shouldn't
551 * be asking the DMU to do *anything*.
553 ASSERT(spa_config_held(os
->os_spa
, SCL_ALL
, RW_WRITER
) == 0);
555 if (object
== DMU_USERUSED_OBJECT
|| object
== DMU_GROUPUSED_OBJECT
) {
556 dn
= (object
== DMU_USERUSED_OBJECT
) ?
557 os
->os_userused_dnode
: os
->os_groupused_dnode
;
561 if ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
)
563 if ((flag
& DNODE_MUST_BE_FREE
) && type
!= DMU_OT_NONE
)
566 (void) refcount_add(&dn
->dn_holds
, tag
);
571 if (object
== 0 || object
>= DN_MAX_OBJECT
)
574 mdn
= os
->os_meta_dnode
;
578 if (!RW_WRITE_HELD(&mdn
->dn_struct_rwlock
)) {
579 rw_enter(&mdn
->dn_struct_rwlock
, RW_READER
);
580 drop_struct_lock
= TRUE
;
583 blk
= dbuf_whichblock(mdn
, object
* sizeof (dnode_phys_t
));
585 db
= dbuf_hold(mdn
, blk
, FTAG
);
586 if (drop_struct_lock
)
587 rw_exit(&mdn
->dn_struct_rwlock
);
590 err
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
);
596 ASSERT3U(db
->db
.db_size
, >=, 1<<DNODE_SHIFT
);
597 epb
= db
->db
.db_size
>> DNODE_SHIFT
;
599 idx
= object
& (epb
-1);
601 children_dnodes
= dmu_buf_get_user(&db
->db
);
602 if (children_dnodes
== NULL
) {
604 children_dnodes
= kmem_zalloc(epb
* sizeof (dnode_t
*),
606 if (winner
= dmu_buf_set_user(&db
->db
, children_dnodes
, NULL
,
607 dnode_buf_pageout
)) {
608 kmem_free(children_dnodes
, epb
* sizeof (dnode_t
*));
609 children_dnodes
= winner
;
613 if ((dn
= children_dnodes
[idx
]) == NULL
) {
614 dnode_phys_t
*dnp
= (dnode_phys_t
*)db
->db
.db_data
+idx
;
617 dn
= dnode_create(os
, dnp
, db
, object
);
618 winner
= atomic_cas_ptr(&children_dnodes
[idx
], NULL
, dn
);
619 if (winner
!= NULL
) {
625 mutex_enter(&dn
->dn_mtx
);
627 if (dn
->dn_free_txg
||
628 ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
) ||
629 ((flag
& DNODE_MUST_BE_FREE
) &&
630 (type
!= DMU_OT_NONE
|| dn
->dn_oldphys
))) {
631 mutex_exit(&dn
->dn_mtx
);
633 return (type
== DMU_OT_NONE
? ENOENT
: EEXIST
);
635 mutex_exit(&dn
->dn_mtx
);
637 if (refcount_add(&dn
->dn_holds
, tag
) == 1)
638 dbuf_add_ref(db
, dn
);
641 ASSERT3P(dn
->dn_dbuf
, ==, db
);
642 ASSERT3U(dn
->dn_object
, ==, object
);
650 * Return held dnode if the object is allocated, NULL if not.
653 dnode_hold(objset_impl_t
*os
, uint64_t object
, void *tag
, dnode_t
**dnp
)
655 return (dnode_hold_impl(os
, object
, DNODE_MUST_BE_ALLOCATED
, tag
, dnp
));
659 * Can only add a reference if there is already at least one
660 * reference on the dnode. Returns FALSE if unable to add a
664 dnode_add_ref(dnode_t
*dn
, void *tag
)
666 mutex_enter(&dn
->dn_mtx
);
667 if (refcount_is_zero(&dn
->dn_holds
)) {
668 mutex_exit(&dn
->dn_mtx
);
671 VERIFY(1 < refcount_add(&dn
->dn_holds
, tag
));
672 mutex_exit(&dn
->dn_mtx
);
677 dnode_rele(dnode_t
*dn
, void *tag
)
681 mutex_enter(&dn
->dn_mtx
);
682 refs
= refcount_remove(&dn
->dn_holds
, tag
);
683 mutex_exit(&dn
->dn_mtx
);
684 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
685 if (refs
== 0 && dn
->dn_dbuf
)
686 dbuf_rele(dn
->dn_dbuf
, dn
);
690 dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
)
692 objset_impl_t
*os
= dn
->dn_objset
;
693 uint64_t txg
= tx
->tx_txg
;
695 if (DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
696 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
703 mutex_enter(&dn
->dn_mtx
);
704 ASSERT(dn
->dn_phys
->dn_type
|| dn
->dn_allocated_txg
);
705 /* ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); */
706 mutex_exit(&dn
->dn_mtx
);
709 mutex_enter(&os
->os_lock
);
712 * If we are already marked dirty, we're done.
714 if (list_link_active(&dn
->dn_dirty_link
[txg
& TXG_MASK
])) {
715 mutex_exit(&os
->os_lock
);
719 ASSERT(!refcount_is_zero(&dn
->dn_holds
) || list_head(&dn
->dn_dbufs
));
720 ASSERT(dn
->dn_datablksz
!= 0);
721 ASSERT3U(dn
->dn_next_bonuslen
[txg
&TXG_MASK
], ==, 0);
722 ASSERT3U(dn
->dn_next_blksz
[txg
&TXG_MASK
], ==, 0);
724 dprintf_ds(os
->os_dsl_dataset
, "obj=%llu txg=%llu\n",
727 if (dn
->dn_free_txg
> 0 && dn
->dn_free_txg
<= txg
) {
728 list_insert_tail(&os
->os_free_dnodes
[txg
&TXG_MASK
], dn
);
730 list_insert_tail(&os
->os_dirty_dnodes
[txg
&TXG_MASK
], dn
);
733 mutex_exit(&os
->os_lock
);
736 * The dnode maintains a hold on its containing dbuf as
737 * long as there are holds on it. Each instantiated child
738 * dbuf maintaines a hold on the dnode. When the last child
739 * drops its hold, the dnode will drop its hold on the
740 * containing dbuf. We add a "dirty hold" here so that the
741 * dnode will hang around after we finish processing its
744 VERIFY(dnode_add_ref(dn
, (void *)(uintptr_t)tx
->tx_txg
));
746 (void) dbuf_dirty(dn
->dn_dbuf
, tx
);
748 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
752 dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
)
754 int txgoff
= tx
->tx_txg
& TXG_MASK
;
756 dprintf("dn=%p txg=%llu\n", dn
, tx
->tx_txg
);
758 /* we should be the only holder... hopefully */
759 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */
761 mutex_enter(&dn
->dn_mtx
);
762 if (dn
->dn_type
== DMU_OT_NONE
|| dn
->dn_free_txg
) {
763 mutex_exit(&dn
->dn_mtx
);
766 dn
->dn_free_txg
= tx
->tx_txg
;
767 mutex_exit(&dn
->dn_mtx
);
770 * If the dnode is already dirty, it needs to be moved from
771 * the dirty list to the free list.
773 mutex_enter(&dn
->dn_objset
->os_lock
);
774 if (list_link_active(&dn
->dn_dirty_link
[txgoff
])) {
775 list_remove(&dn
->dn_objset
->os_dirty_dnodes
[txgoff
], dn
);
776 list_insert_tail(&dn
->dn_objset
->os_free_dnodes
[txgoff
], dn
);
777 mutex_exit(&dn
->dn_objset
->os_lock
);
779 mutex_exit(&dn
->dn_objset
->os_lock
);
780 dnode_setdirty(dn
, tx
);
785 * Try to change the block size for the indicated dnode. This can only
786 * succeed if there are no blocks allocated or dirty beyond first block
789 dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
)
791 dmu_buf_impl_t
*db
, *db_next
;
795 size
= SPA_MINBLOCKSIZE
;
796 if (size
> SPA_MAXBLOCKSIZE
)
797 size
= SPA_MAXBLOCKSIZE
;
799 size
= P2ROUNDUP(size
, SPA_MINBLOCKSIZE
);
801 if (ibs
== dn
->dn_indblkshift
)
804 if (size
>> SPA_MINBLOCKSHIFT
== dn
->dn_datablkszsec
&& ibs
== 0)
807 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
809 /* Check for any allocated blocks beyond the first */
810 if (dn
->dn_phys
->dn_maxblkid
!= 0)
813 mutex_enter(&dn
->dn_dbufs_mtx
);
814 for (db
= list_head(&dn
->dn_dbufs
); db
; db
= db_next
) {
815 db_next
= list_next(&dn
->dn_dbufs
, db
);
817 if (db
->db_blkid
!= 0 && db
->db_blkid
!= DB_BONUS_BLKID
) {
818 mutex_exit(&dn
->dn_dbufs_mtx
);
822 mutex_exit(&dn
->dn_dbufs_mtx
);
824 if (ibs
&& dn
->dn_nlevels
!= 1)
827 /* resize the old block */
828 err
= dbuf_hold_impl(dn
, 0, 0, TRUE
, FTAG
, &db
);
830 dbuf_new_size(db
, size
, tx
);
831 else if (err
!= ENOENT
)
834 dnode_setdblksz(dn
, size
);
835 dnode_setdirty(dn
, tx
);
836 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = size
;
838 dn
->dn_indblkshift
= ibs
;
839 dn
->dn_next_indblkshift
[tx
->tx_txg
&TXG_MASK
] = ibs
;
841 /* rele after we have fixed the blocksize in the dnode */
845 rw_exit(&dn
->dn_struct_rwlock
);
849 rw_exit(&dn
->dn_struct_rwlock
);
853 /* read-holding callers must not rely on the lock being continuously held */
855 dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t have_read
)
857 uint64_t txgoff
= tx
->tx_txg
& TXG_MASK
;
858 int epbs
, new_nlevels
;
861 ASSERT(blkid
!= DB_BONUS_BLKID
);
864 RW_READ_HELD(&dn
->dn_struct_rwlock
) :
865 RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
868 * if we have a read-lock, check to see if we need to do any work
869 * before upgrading to a write-lock.
872 if (blkid
<= dn
->dn_maxblkid
)
875 if (!rw_tryupgrade(&dn
->dn_struct_rwlock
)) {
876 rw_exit(&dn
->dn_struct_rwlock
);
877 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
881 if (blkid
<= dn
->dn_maxblkid
)
884 dn
->dn_maxblkid
= blkid
;
887 * Compute the number of levels necessary to support the new maxblkid.
890 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
891 for (sz
= dn
->dn_nblkptr
;
892 sz
<= blkid
&& sz
>= dn
->dn_nblkptr
; sz
<<= epbs
)
895 if (new_nlevels
> dn
->dn_nlevels
) {
896 int old_nlevels
= dn
->dn_nlevels
;
899 dbuf_dirty_record_t
*new, *dr
, *dr_next
;
901 dn
->dn_nlevels
= new_nlevels
;
903 ASSERT3U(new_nlevels
, >, dn
->dn_next_nlevels
[txgoff
]);
904 dn
->dn_next_nlevels
[txgoff
] = new_nlevels
;
906 /* dirty the left indirects */
907 db
= dbuf_hold_level(dn
, old_nlevels
, 0, FTAG
);
908 new = dbuf_dirty(db
, tx
);
911 /* transfer the dirty records to the new indirect */
912 mutex_enter(&dn
->dn_mtx
);
913 mutex_enter(&new->dt
.di
.dr_mtx
);
914 list
= &dn
->dn_dirty_records
[txgoff
];
915 for (dr
= list_head(list
); dr
; dr
= dr_next
) {
916 dr_next
= list_next(&dn
->dn_dirty_records
[txgoff
], dr
);
917 if (dr
->dr_dbuf
->db_level
!= new_nlevels
-1 &&
918 dr
->dr_dbuf
->db_blkid
!= DB_BONUS_BLKID
) {
919 ASSERT(dr
->dr_dbuf
->db_level
== old_nlevels
-1);
920 list_remove(&dn
->dn_dirty_records
[txgoff
], dr
);
921 list_insert_tail(&new->dt
.di
.dr_children
, dr
);
925 mutex_exit(&new->dt
.di
.dr_mtx
);
926 mutex_exit(&dn
->dn_mtx
);
931 rw_downgrade(&dn
->dn_struct_rwlock
);
935 dnode_clear_range(dnode_t
*dn
, uint64_t blkid
, uint64_t nblks
, dmu_tx_t
*tx
)
937 avl_tree_t
*tree
= &dn
->dn_ranges
[tx
->tx_txg
&TXG_MASK
];
940 free_range_t rp_tofind
;
941 uint64_t endblk
= blkid
+ nblks
;
943 ASSERT(MUTEX_HELD(&dn
->dn_mtx
));
944 ASSERT(nblks
<= UINT64_MAX
- blkid
); /* no overflow */
946 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
947 blkid
, nblks
, tx
->tx_txg
);
948 rp_tofind
.fr_blkid
= blkid
;
949 rp
= avl_find(tree
, &rp_tofind
, &where
);
951 rp
= avl_nearest(tree
, where
, AVL_BEFORE
);
953 rp
= avl_nearest(tree
, where
, AVL_AFTER
);
955 while (rp
&& (rp
->fr_blkid
<= blkid
+ nblks
)) {
956 uint64_t fr_endblk
= rp
->fr_blkid
+ rp
->fr_nblks
;
957 free_range_t
*nrp
= AVL_NEXT(tree
, rp
);
959 if (blkid
<= rp
->fr_blkid
&& endblk
>= fr_endblk
) {
960 /* clear this entire range */
961 avl_remove(tree
, rp
);
962 kmem_free(rp
, sizeof (free_range_t
));
963 } else if (blkid
<= rp
->fr_blkid
&&
964 endblk
> rp
->fr_blkid
&& endblk
< fr_endblk
) {
965 /* clear the beginning of this range */
966 rp
->fr_blkid
= endblk
;
967 rp
->fr_nblks
= fr_endblk
- endblk
;
968 } else if (blkid
> rp
->fr_blkid
&& blkid
< fr_endblk
&&
969 endblk
>= fr_endblk
) {
970 /* clear the end of this range */
971 rp
->fr_nblks
= blkid
- rp
->fr_blkid
;
972 } else if (blkid
> rp
->fr_blkid
&& endblk
< fr_endblk
) {
973 /* clear a chunk out of this range */
974 free_range_t
*new_rp
=
975 kmem_alloc(sizeof (free_range_t
), KM_SLEEP
);
977 new_rp
->fr_blkid
= endblk
;
978 new_rp
->fr_nblks
= fr_endblk
- endblk
;
979 avl_insert_here(tree
, new_rp
, rp
, AVL_AFTER
);
980 rp
->fr_nblks
= blkid
- rp
->fr_blkid
;
982 /* there may be no overlap */
988 dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
)
991 uint64_t blkoff
, blkid
, nblks
;
992 int blksz
, blkshift
, head
, tail
;
996 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
997 blksz
= dn
->dn_datablksz
;
998 blkshift
= dn
->dn_datablkshift
;
999 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1002 len
= UINT64_MAX
- off
;
1007 * First, block align the region to free:
1010 head
= P2NPHASE(off
, blksz
);
1011 blkoff
= P2PHASE(off
, blksz
);
1012 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1015 ASSERT(dn
->dn_maxblkid
== 0);
1016 if (off
== 0 && len
>= blksz
) {
1017 /* Freeing the whole block; fast-track this request */
1021 } else if (off
>= blksz
) {
1022 /* Freeing past end-of-data */
1025 /* Freeing part of the block. */
1027 ASSERT3U(head
, >, 0);
1031 /* zero out any partial block data at the start of the range */
1033 ASSERT3U(blkoff
+ head
, ==, blksz
);
1036 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
), TRUE
,
1040 /* don't dirty if it isn't on disk and isn't dirty */
1041 if (db
->db_last_dirty
||
1042 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1043 rw_exit(&dn
->dn_struct_rwlock
);
1044 dbuf_will_dirty(db
, tx
);
1045 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1046 data
= db
->db
.db_data
;
1047 bzero(data
+ blkoff
, head
);
1049 dbuf_rele(db
, FTAG
);
1055 /* If the range was less than one block, we're done */
1059 /* If the remaining range is past end of file, we're done */
1060 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1063 ASSERT(ISP2(blksz
));
1067 tail
= P2PHASE(len
, blksz
);
1069 ASSERT3U(P2PHASE(off
, blksz
), ==, 0);
1070 /* zero out any partial block data at the end of the range */
1074 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
+len
),
1075 TRUE
, FTAG
, &db
) == 0) {
1076 /* don't dirty if not on disk and not dirty */
1077 if (db
->db_last_dirty
||
1078 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1079 rw_exit(&dn
->dn_struct_rwlock
);
1080 dbuf_will_dirty(db
, tx
);
1081 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1082 bzero(db
->db
.db_data
, tail
);
1084 dbuf_rele(db
, FTAG
);
1089 /* If the range did not include a full block, we are done */
1093 ASSERT(IS_P2ALIGNED(off
, blksz
));
1094 ASSERT(trunc
|| IS_P2ALIGNED(len
, blksz
));
1095 blkid
= off
>> blkshift
;
1096 nblks
= len
>> blkshift
;
1101 * Read in and mark all the level-1 indirects dirty,
1102 * so that they will stay in memory until syncing phase.
1103 * Always dirty the first and last indirect to make sure
1104 * we dirty all the partial indirects.
1106 if (dn
->dn_nlevels
> 1) {
1107 uint64_t i
, first
, last
;
1108 int shift
= epbs
+ dn
->dn_datablkshift
;
1110 first
= blkid
>> epbs
;
1111 if (db
= dbuf_hold_level(dn
, 1, first
, FTAG
)) {
1112 dbuf_will_dirty(db
, tx
);
1113 dbuf_rele(db
, FTAG
);
1116 last
= dn
->dn_maxblkid
>> epbs
;
1118 last
= (blkid
+ nblks
- 1) >> epbs
;
1119 if (last
> first
&& (db
= dbuf_hold_level(dn
, 1, last
, FTAG
))) {
1120 dbuf_will_dirty(db
, tx
);
1121 dbuf_rele(db
, FTAG
);
1123 for (i
= first
+ 1; i
< last
; i
++) {
1124 uint64_t ibyte
= i
<< shift
;
1127 err
= dnode_next_offset(dn
,
1128 DNODE_FIND_HAVELOCK
, &ibyte
, 1, 1, 0);
1130 if (err
== ESRCH
|| i
>= last
)
1133 db
= dbuf_hold_level(dn
, 1, i
, FTAG
);
1135 dbuf_will_dirty(db
, tx
);
1136 dbuf_rele(db
, FTAG
);
1142 * Add this range to the dnode range list.
1143 * We will finish up this free operation in the syncing phase.
1145 mutex_enter(&dn
->dn_mtx
);
1146 dnode_clear_range(dn
, blkid
, nblks
, tx
);
1148 free_range_t
*rp
, *found
;
1150 avl_tree_t
*tree
= &dn
->dn_ranges
[tx
->tx_txg
&TXG_MASK
];
1152 /* Add new range to dn_ranges */
1153 rp
= kmem_alloc(sizeof (free_range_t
), KM_SLEEP
);
1154 rp
->fr_blkid
= blkid
;
1155 rp
->fr_nblks
= nblks
;
1156 found
= avl_find(tree
, rp
, &where
);
1157 ASSERT(found
== NULL
);
1158 avl_insert(tree
, rp
, where
);
1159 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1160 blkid
, nblks
, tx
->tx_txg
);
1162 mutex_exit(&dn
->dn_mtx
);
1164 dbuf_free_range(dn
, blkid
, blkid
+ nblks
- 1, tx
);
1165 dnode_setdirty(dn
, tx
);
1167 if (trunc
&& dn
->dn_maxblkid
>= (off
>> blkshift
))
1168 dn
->dn_maxblkid
= (off
>> blkshift
? (off
>> blkshift
) - 1 : 0);
1170 rw_exit(&dn
->dn_struct_rwlock
);
1173 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1175 dnode_block_freed(dnode_t
*dn
, uint64_t blkid
)
1177 free_range_t range_tofind
;
1178 void *dp
= spa_get_dsl(dn
->dn_objset
->os_spa
);
1181 if (blkid
== DB_BONUS_BLKID
)
1185 * If we're in the process of opening the pool, dp will not be
1186 * set yet, but there shouldn't be anything dirty.
1191 if (dn
->dn_free_txg
)
1194 range_tofind
.fr_blkid
= blkid
;
1195 mutex_enter(&dn
->dn_mtx
);
1196 for (i
= 0; i
< TXG_SIZE
; i
++) {
1197 free_range_t
*range_found
;
1200 range_found
= avl_find(&dn
->dn_ranges
[i
], &range_tofind
, &idx
);
1202 ASSERT(range_found
->fr_nblks
> 0);
1205 range_found
= avl_nearest(&dn
->dn_ranges
[i
], idx
, AVL_BEFORE
);
1207 range_found
->fr_blkid
+ range_found
->fr_nblks
> blkid
)
1210 mutex_exit(&dn
->dn_mtx
);
1211 return (i
< TXG_SIZE
);
1214 /* call from syncing context when we actually write/free space for this dnode */
1216 dnode_diduse_space(dnode_t
*dn
, int64_t delta
)
1219 dprintf_dnode(dn
, "dn=%p dnp=%p used=%llu delta=%lld\n",
1221 (u_longlong_t
)dn
->dn_phys
->dn_used
,
1224 mutex_enter(&dn
->dn_mtx
);
1225 space
= DN_USED_BYTES(dn
->dn_phys
);
1227 ASSERT3U(space
+ delta
, >=, space
); /* no overflow */
1229 ASSERT3U(space
, >=, -delta
); /* no underflow */
1232 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_DNODE_BYTES
) {
1233 ASSERT((dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) == 0);
1234 ASSERT3U(P2PHASE(space
, 1<<DEV_BSHIFT
), ==, 0);
1235 dn
->dn_phys
->dn_used
= space
>> DEV_BSHIFT
;
1237 dn
->dn_phys
->dn_used
= space
;
1238 dn
->dn_phys
->dn_flags
|= DNODE_FLAG_USED_BYTES
;
1240 mutex_exit(&dn
->dn_mtx
);
1244 * Call when we think we're going to write/free space in open context.
1245 * Be conservative (ie. OK to write less than this or free more than
1246 * this, but don't write more or free less).
1249 dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
)
1251 objset_impl_t
*os
= dn
->dn_objset
;
1252 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1255 space
= spa_get_asize(os
->os_spa
, space
);
1258 dsl_dir_willuse_space(ds
->ds_dir
, space
, tx
);
1260 dmu_tx_willuse_space(tx
, space
);
1264 * This function scans a block at the indicated "level" looking for
1265 * a hole or data (depending on 'flags'). If level > 0, then we are
1266 * scanning an indirect block looking at its pointers. If level == 0,
1267 * then we are looking at a block of dnodes. If we don't find what we
1268 * are looking for in the block, we return ESRCH. Otherwise, return
1269 * with *offset pointing to the beginning (if searching forwards) or
1270 * end (if searching backwards) of the range covered by the block
1271 * pointer we matched on (or dnode).
1273 * The basic search algorithm used below by dnode_next_offset() is to
1274 * use this function to search up the block tree (widen the search) until
1275 * we find something (i.e., we don't return ESRCH) and then search back
1276 * down the tree (narrow the search) until we reach our original search
1280 dnode_next_offset_level(dnode_t
*dn
, int flags
, uint64_t *offset
,
1281 int lvl
, uint64_t blkfill
, uint64_t txg
)
1283 dmu_buf_impl_t
*db
= NULL
;
1285 uint64_t epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1286 uint64_t epb
= 1ULL << epbs
;
1287 uint64_t minfill
, maxfill
;
1289 int i
, inc
, error
, span
;
1291 dprintf("probing object %llu offset %llx level %d of %u\n",
1292 dn
->dn_object
, *offset
, lvl
, dn
->dn_phys
->dn_nlevels
);
1294 hole
= ((flags
& DNODE_FIND_HOLE
) != 0);
1295 inc
= (flags
& DNODE_FIND_BACKWARDS
) ? -1 : 1;
1296 ASSERT(txg
== 0 || !hole
);
1298 if (lvl
== dn
->dn_phys
->dn_nlevels
) {
1300 epb
= dn
->dn_phys
->dn_nblkptr
;
1301 data
= dn
->dn_phys
->dn_blkptr
;
1303 uint64_t blkid
= dbuf_whichblock(dn
, *offset
) >> (epbs
* lvl
);
1304 error
= dbuf_hold_impl(dn
, lvl
, blkid
, TRUE
, FTAG
, &db
);
1306 if (error
!= ENOENT
)
1311 * This can only happen when we are searching up
1312 * the block tree for data. We don't really need to
1313 * adjust the offset, as we will just end up looking
1314 * at the pointer to this block in its parent, and its
1315 * going to be unallocated, so we will skip over it.
1319 error
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
| DB_RF_HAVESTRUCT
);
1321 dbuf_rele(db
, FTAG
);
1324 data
= db
->db
.db_data
;
1328 (db
->db_blkptr
== NULL
|| db
->db_blkptr
->blk_birth
<= txg
)) {
1330 * This can only happen when we are searching up the tree
1331 * and these conditions mean that we need to keep climbing.
1334 } else if (lvl
== 0) {
1335 dnode_phys_t
*dnp
= data
;
1337 ASSERT(dn
->dn_type
== DMU_OT_DNODE
);
1339 for (i
= (*offset
>> span
) & (blkfill
- 1);
1340 i
>= 0 && i
< blkfill
; i
+= inc
) {
1341 if ((dnp
[i
].dn_type
== DMU_OT_NONE
) == hole
)
1343 *offset
+= (1ULL << span
) * inc
;
1345 if (i
< 0 || i
== blkfill
)
1348 blkptr_t
*bp
= data
;
1349 uint64_t start
= *offset
;
1350 span
= (lvl
- 1) * epbs
+ dn
->dn_datablkshift
;
1352 maxfill
= blkfill
<< ((lvl
- 1) * epbs
);
1359 *offset
= *offset
>> span
;
1360 for (i
= BF64_GET(*offset
, 0, epbs
);
1361 i
>= 0 && i
< epb
; i
+= inc
) {
1362 if (bp
[i
].blk_fill
>= minfill
&&
1363 bp
[i
].blk_fill
<= maxfill
&&
1364 (hole
|| bp
[i
].blk_birth
> txg
))
1366 if (inc
> 0 || *offset
> 0)
1369 *offset
= *offset
<< span
;
1371 /* traversing backwards; position offset at the end */
1372 ASSERT3U(*offset
, <=, start
);
1373 *offset
= MIN(*offset
+ (1ULL << span
) - 1, start
);
1374 } else if (*offset
< start
) {
1377 if (i
< 0 || i
>= epb
)
1382 dbuf_rele(db
, FTAG
);
1388 * Find the next hole, data, or sparse region at or after *offset.
1389 * The value 'blkfill' tells us how many items we expect to find
1390 * in an L0 data block; this value is 1 for normal objects,
1391 * DNODES_PER_BLOCK for the meta dnode, and some fraction of
1392 * DNODES_PER_BLOCK when searching for sparse regions thereof.
1396 * dnode_next_offset(dn, flags, offset, 1, 1, 0);
1397 * Finds the next/previous hole/data in a file.
1398 * Used in dmu_offset_next().
1400 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg);
1401 * Finds the next free/allocated dnode an objset's meta-dnode.
1402 * Only finds objects that have new contents since txg (ie.
1403 * bonus buffer changes and content removal are ignored).
1404 * Used in dmu_object_next().
1406 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0);
1407 * Finds the next L2 meta-dnode bp that's at most 1/4 full.
1408 * Used in dmu_object_alloc().
1411 dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *offset
,
1412 int minlvl
, uint64_t blkfill
, uint64_t txg
)
1414 uint64_t initial_offset
= *offset
;
1418 if (!(flags
& DNODE_FIND_HAVELOCK
))
1419 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1421 if (dn
->dn_phys
->dn_nlevels
== 0) {
1426 if (dn
->dn_datablkshift
== 0) {
1427 if (*offset
< dn
->dn_datablksz
) {
1428 if (flags
& DNODE_FIND_HOLE
)
1429 *offset
= dn
->dn_datablksz
;
1436 maxlvl
= dn
->dn_phys
->dn_nlevels
;
1438 for (lvl
= minlvl
; lvl
<= maxlvl
; lvl
++) {
1439 error
= dnode_next_offset_level(dn
,
1440 flags
, offset
, lvl
, blkfill
, txg
);
1445 while (error
== 0 && --lvl
>= minlvl
) {
1446 error
= dnode_next_offset_level(dn
,
1447 flags
, offset
, lvl
, blkfill
, txg
);
1450 if (error
== 0 && (flags
& DNODE_FIND_BACKWARDS
?
1451 initial_offset
< *offset
: initial_offset
> *offset
))
1454 if (!(flags
& DNODE_FIND_HAVELOCK
))
1455 rw_exit(&dn
->dn_struct_rwlock
);