4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 #include <sys/zfs_context.h>
27 #include <sys/dnode.h>
29 #include <sys/dmu_impl.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/dmu_objset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_dataset.h>
36 #include <sys/dmu_zfetch.h>
38 static int free_range_compar(const void *node1
, const void *node2
);
40 static kmem_cache_t
*dnode_cache
;
42 static dnode_phys_t dnode_phys_zero
;
44 int zfs_default_bs
= SPA_MINBLOCKSHIFT
;
45 int zfs_default_ibs
= DN_MAX_INDBLKSHIFT
;
49 dnode_cons(void *arg
, void *unused
, int kmflag
)
53 bzero(dn
, sizeof (dnode_t
));
55 rw_init(&dn
->dn_struct_rwlock
, NULL
, RW_DEFAULT
, NULL
);
56 mutex_init(&dn
->dn_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
57 mutex_init(&dn
->dn_dbufs_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
58 cv_init(&dn
->dn_notxholds
, NULL
, CV_DEFAULT
, NULL
);
60 refcount_create(&dn
->dn_holds
);
61 refcount_create(&dn
->dn_tx_holds
);
63 for (i
= 0; i
< TXG_SIZE
; i
++) {
64 avl_create(&dn
->dn_ranges
[i
], free_range_compar
,
65 sizeof (free_range_t
),
66 offsetof(struct free_range
, fr_node
));
67 list_create(&dn
->dn_dirty_records
[i
],
68 sizeof (dbuf_dirty_record_t
),
69 offsetof(dbuf_dirty_record_t
, dr_dirty_node
));
72 list_create(&dn
->dn_dbufs
, sizeof (dmu_buf_impl_t
),
73 offsetof(dmu_buf_impl_t
, db_link
));
80 dnode_dest(void *arg
, void *unused
)
85 rw_destroy(&dn
->dn_struct_rwlock
);
86 mutex_destroy(&dn
->dn_mtx
);
87 mutex_destroy(&dn
->dn_dbufs_mtx
);
88 cv_destroy(&dn
->dn_notxholds
);
89 refcount_destroy(&dn
->dn_holds
);
90 refcount_destroy(&dn
->dn_tx_holds
);
92 for (i
= 0; i
< TXG_SIZE
; i
++) {
93 avl_destroy(&dn
->dn_ranges
[i
]);
94 list_destroy(&dn
->dn_dirty_records
[i
]);
97 list_destroy(&dn
->dn_dbufs
);
103 dnode_cache
= kmem_cache_create("dnode_t",
105 0, dnode_cons
, dnode_dest
, NULL
, NULL
, NULL
, 0);
111 kmem_cache_destroy(dnode_cache
);
117 dnode_verify(dnode_t
*dn
)
119 int drop_struct_lock
= FALSE
;
122 ASSERT(dn
->dn_objset
);
124 ASSERT(dn
->dn_phys
->dn_type
< DMU_OT_NUMTYPES
);
126 if (!(zfs_flags
& ZFS_DEBUG_DNODE_VERIFY
))
129 if (!RW_WRITE_HELD(&dn
->dn_struct_rwlock
)) {
130 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
131 drop_struct_lock
= TRUE
;
133 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
|| dn
->dn_allocated_txg
!= 0) {
135 ASSERT3U(dn
->dn_indblkshift
, >=, 0);
136 ASSERT3U(dn
->dn_indblkshift
, <=, SPA_MAXBLOCKSHIFT
);
137 if (dn
->dn_datablkshift
) {
138 ASSERT3U(dn
->dn_datablkshift
, >=, SPA_MINBLOCKSHIFT
);
139 ASSERT3U(dn
->dn_datablkshift
, <=, SPA_MAXBLOCKSHIFT
);
140 ASSERT3U(1<<dn
->dn_datablkshift
, ==, dn
->dn_datablksz
);
142 ASSERT3U(dn
->dn_nlevels
, <=, 30);
143 ASSERT3U(dn
->dn_type
, <=, DMU_OT_NUMTYPES
);
144 ASSERT3U(dn
->dn_nblkptr
, >=, 1);
145 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
146 ASSERT3U(dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
147 ASSERT3U(dn
->dn_datablksz
, ==,
148 dn
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
149 ASSERT3U(ISP2(dn
->dn_datablksz
), ==, dn
->dn_datablkshift
!= 0);
150 ASSERT3U((dn
->dn_nblkptr
- 1) * sizeof (blkptr_t
) +
151 dn
->dn_bonuslen
, <=, DN_MAX_BONUSLEN
);
152 for (i
= 0; i
< TXG_SIZE
; i
++) {
153 ASSERT3U(dn
->dn_next_nlevels
[i
], <=, dn
->dn_nlevels
);
156 if (dn
->dn_phys
->dn_type
!= DMU_OT_NONE
)
157 ASSERT3U(dn
->dn_phys
->dn_nlevels
, <=, dn
->dn_nlevels
);
158 ASSERT(DMU_OBJECT_IS_SPECIAL(dn
->dn_object
) || dn
->dn_dbuf
!= NULL
);
159 if (dn
->dn_dbuf
!= NULL
) {
160 ASSERT3P(dn
->dn_phys
, ==,
161 (dnode_phys_t
*)dn
->dn_dbuf
->db
.db_data
+
162 (dn
->dn_object
% (dn
->dn_dbuf
->db
.db_size
>> DNODE_SHIFT
)));
164 if (drop_struct_lock
)
165 rw_exit(&dn
->dn_struct_rwlock
);
170 dnode_byteswap(dnode_phys_t
*dnp
)
172 uint64_t *buf64
= (void*)&dnp
->dn_blkptr
;
175 if (dnp
->dn_type
== DMU_OT_NONE
) {
176 bzero(dnp
, sizeof (dnode_phys_t
));
180 dnp
->dn_datablkszsec
= BSWAP_16(dnp
->dn_datablkszsec
);
181 dnp
->dn_bonuslen
= BSWAP_16(dnp
->dn_bonuslen
);
182 dnp
->dn_maxblkid
= BSWAP_64(dnp
->dn_maxblkid
);
183 dnp
->dn_used
= BSWAP_64(dnp
->dn_used
);
186 * dn_nblkptr is only one byte, so it's OK to read it in either
187 * byte order. We can't read dn_bouslen.
189 ASSERT(dnp
->dn_indblkshift
<= SPA_MAXBLOCKSHIFT
);
190 ASSERT(dnp
->dn_nblkptr
<= DN_MAX_NBLKPTR
);
191 for (i
= 0; i
< dnp
->dn_nblkptr
* sizeof (blkptr_t
)/8; i
++)
192 buf64
[i
] = BSWAP_64(buf64
[i
]);
195 * OK to check dn_bonuslen for zero, because it won't matter if
196 * we have the wrong byte order. This is necessary because the
197 * dnode dnode is smaller than a regular dnode.
199 if (dnp
->dn_bonuslen
!= 0) {
201 * Note that the bonus length calculated here may be
202 * longer than the actual bonus buffer. This is because
203 * we always put the bonus buffer after the last block
204 * pointer (instead of packing it against the end of the
207 int off
= (dnp
->dn_nblkptr
-1) * sizeof (blkptr_t
);
208 size_t len
= DN_MAX_BONUSLEN
- off
;
209 ASSERT3U(dnp
->dn_bonustype
, <, DMU_OT_NUMTYPES
);
210 dmu_ot
[dnp
->dn_bonustype
].ot_byteswap(dnp
->dn_bonus
+ off
, len
);
213 /* Swap SPILL block if we have one */
214 if (dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
)
215 byteswap_uint64_array(&dnp
->dn_spill
, sizeof (blkptr_t
));
220 dnode_buf_byteswap(void *vbuf
, size_t size
)
222 dnode_phys_t
*buf
= vbuf
;
225 ASSERT3U(sizeof (dnode_phys_t
), ==, (1<<DNODE_SHIFT
));
226 ASSERT((size
& (sizeof (dnode_phys_t
)-1)) == 0);
228 size
>>= DNODE_SHIFT
;
229 for (i
= 0; i
< size
; i
++) {
236 free_range_compar(const void *node1
, const void *node2
)
238 const free_range_t
*rp1
= node1
;
239 const free_range_t
*rp2
= node2
;
241 if (rp1
->fr_blkid
< rp2
->fr_blkid
)
243 else if (rp1
->fr_blkid
> rp2
->fr_blkid
)
249 dnode_setbonuslen(dnode_t
*dn
, int newsize
, dmu_tx_t
*tx
)
251 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
253 dnode_setdirty(dn
, tx
);
254 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
255 ASSERT3U(newsize
, <=, DN_MAX_BONUSLEN
-
256 (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
));
257 dn
->dn_bonuslen
= newsize
;
259 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = DN_ZERO_BONUSLEN
;
261 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
262 rw_exit(&dn
->dn_struct_rwlock
);
266 dnode_setbonus_type(dnode_t
*dn
, dmu_object_type_t newtype
, dmu_tx_t
*tx
)
268 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
269 dnode_setdirty(dn
, tx
);
270 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
271 dn
->dn_bonustype
= newtype
;
272 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
273 rw_exit(&dn
->dn_struct_rwlock
);
277 dnode_rm_spill(dnode_t
*dn
, dmu_tx_t
*tx
)
279 ASSERT3U(refcount_count(&dn
->dn_holds
), >=, 1);
280 ASSERT(RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
281 dnode_setdirty(dn
, tx
);
282 dn
->dn_rm_spillblk
[tx
->tx_txg
&TXG_MASK
] = DN_KILL_SPILLBLK
;
283 dn
->dn_have_spill
= B_FALSE
;
287 dnode_setdblksz(dnode_t
*dn
, int size
)
289 ASSERT3U(P2PHASE(size
, SPA_MINBLOCKSIZE
), ==, 0);
290 ASSERT3U(size
, <=, SPA_MAXBLOCKSIZE
);
291 ASSERT3U(size
, >=, SPA_MINBLOCKSIZE
);
292 ASSERT3U(size
>> SPA_MINBLOCKSHIFT
, <,
293 1<<(sizeof (dn
->dn_phys
->dn_datablkszsec
) * 8));
294 dn
->dn_datablksz
= size
;
295 dn
->dn_datablkszsec
= size
>> SPA_MINBLOCKSHIFT
;
296 dn
->dn_datablkshift
= ISP2(size
) ? highbit(size
- 1) : 0;
300 dnode_create(objset_t
*os
, dnode_phys_t
*dnp
, dmu_buf_impl_t
*db
,
303 dnode_t
*dn
= kmem_cache_alloc(dnode_cache
, KM_SLEEP
);
304 (void) dnode_cons(dn
, NULL
, 0); /* XXX */
307 dn
->dn_object
= object
;
311 if (dnp
->dn_datablkszsec
)
312 dnode_setdblksz(dn
, dnp
->dn_datablkszsec
<< SPA_MINBLOCKSHIFT
);
313 dn
->dn_indblkshift
= dnp
->dn_indblkshift
;
314 dn
->dn_nlevels
= dnp
->dn_nlevels
;
315 dn
->dn_type
= dnp
->dn_type
;
316 dn
->dn_nblkptr
= dnp
->dn_nblkptr
;
317 dn
->dn_checksum
= dnp
->dn_checksum
;
318 dn
->dn_compress
= dnp
->dn_compress
;
319 dn
->dn_bonustype
= dnp
->dn_bonustype
;
320 dn
->dn_bonuslen
= dnp
->dn_bonuslen
;
321 dn
->dn_maxblkid
= dnp
->dn_maxblkid
;
322 dn
->dn_have_spill
= ((dnp
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) != 0);
325 dmu_zfetch_init(&dn
->dn_zfetch
, dn
);
327 ASSERT(dn
->dn_phys
->dn_type
< DMU_OT_NUMTYPES
);
328 mutex_enter(&os
->os_lock
);
329 list_insert_head(&os
->os_dnodes
, dn
);
330 mutex_exit(&os
->os_lock
);
332 arc_space_consume(sizeof (dnode_t
), ARC_SPACE_OTHER
);
337 dnode_destroy(dnode_t
*dn
)
339 objset_t
*os
= dn
->dn_objset
;
344 for (i
= 0; i
< TXG_SIZE
; i
++) {
345 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
346 ASSERT(NULL
== list_head(&dn
->dn_dirty_records
[i
]));
347 ASSERT(0 == avl_numnodes(&dn
->dn_ranges
[i
]));
349 ASSERT(NULL
== list_head(&dn
->dn_dbufs
));
351 ASSERT((dn
->dn_id_flags
& DN_ID_NEW_EXIST
) == 0);
353 mutex_enter(&os
->os_lock
);
354 list_remove(&os
->os_dnodes
, dn
);
355 mutex_exit(&os
->os_lock
);
357 if (dn
->dn_dirtyctx_firstset
) {
358 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
359 dn
->dn_dirtyctx_firstset
= NULL
;
361 dmu_zfetch_rele(&dn
->dn_zfetch
);
363 mutex_enter(&dn
->dn_bonus
->db_mtx
);
364 dbuf_evict(dn
->dn_bonus
);
367 kmem_cache_free(dnode_cache
, dn
);
368 arc_space_return(sizeof (dnode_t
), ARC_SPACE_OTHER
);
372 dnode_allocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
, int ibs
,
373 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
378 blocksize
= 1 << zfs_default_bs
;
379 else if (blocksize
> SPA_MAXBLOCKSIZE
)
380 blocksize
= SPA_MAXBLOCKSIZE
;
382 blocksize
= P2ROUNDUP(blocksize
, SPA_MINBLOCKSIZE
);
385 ibs
= zfs_default_ibs
;
387 ibs
= MIN(MAX(ibs
, DN_MIN_INDBLKSHIFT
), DN_MAX_INDBLKSHIFT
);
389 dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn
->dn_objset
,
390 dn
->dn_object
, tx
->tx_txg
, blocksize
, ibs
);
392 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
393 ASSERT(bcmp(dn
->dn_phys
, &dnode_phys_zero
, sizeof (dnode_phys_t
)) == 0);
394 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
);
395 ASSERT(ot
!= DMU_OT_NONE
);
396 ASSERT3U(ot
, <, DMU_OT_NUMTYPES
);
397 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
398 (bonustype
== DMU_OT_SA
&& bonuslen
== 0) ||
399 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0));
400 ASSERT3U(bonustype
, <, DMU_OT_NUMTYPES
);
401 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
402 ASSERT(dn
->dn_type
== DMU_OT_NONE
);
403 ASSERT3U(dn
->dn_maxblkid
, ==, 0);
404 ASSERT3U(dn
->dn_allocated_txg
, ==, 0);
405 ASSERT3U(dn
->dn_assigned_txg
, ==, 0);
406 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
407 ASSERT3U(refcount_count(&dn
->dn_holds
), <=, 1);
408 ASSERT3P(list_head(&dn
->dn_dbufs
), ==, NULL
);
410 for (i
= 0; i
< TXG_SIZE
; i
++) {
411 ASSERT3U(dn
->dn_next_nlevels
[i
], ==, 0);
412 ASSERT3U(dn
->dn_next_indblkshift
[i
], ==, 0);
413 ASSERT3U(dn
->dn_next_bonuslen
[i
], ==, 0);
414 ASSERT3U(dn
->dn_next_bonustype
[i
], ==, 0);
415 ASSERT3U(dn
->dn_rm_spillblk
[i
], ==, 0);
416 ASSERT3U(dn
->dn_next_blksz
[i
], ==, 0);
417 ASSERT(!list_link_active(&dn
->dn_dirty_link
[i
]));
418 ASSERT3P(list_head(&dn
->dn_dirty_records
[i
]), ==, NULL
);
419 ASSERT3U(avl_numnodes(&dn
->dn_ranges
[i
]), ==, 0);
423 dnode_setdblksz(dn
, blocksize
);
424 dn
->dn_indblkshift
= ibs
;
426 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
430 ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
431 dn
->dn_bonustype
= bonustype
;
432 dn
->dn_bonuslen
= bonuslen
;
433 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
434 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
438 if (dn
->dn_dirtyctx_firstset
) {
439 kmem_free(dn
->dn_dirtyctx_firstset
, 1);
440 dn
->dn_dirtyctx_firstset
= NULL
;
443 dn
->dn_allocated_txg
= tx
->tx_txg
;
446 dnode_setdirty(dn
, tx
);
447 dn
->dn_next_indblkshift
[tx
->tx_txg
& TXG_MASK
] = ibs
;
448 dn
->dn_next_bonuslen
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonuslen
;
449 dn
->dn_next_bonustype
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_bonustype
;
450 dn
->dn_next_blksz
[tx
->tx_txg
& TXG_MASK
] = dn
->dn_datablksz
;
454 dnode_reallocate(dnode_t
*dn
, dmu_object_type_t ot
, int blocksize
,
455 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
459 ASSERT3U(blocksize
, >=, SPA_MINBLOCKSIZE
);
460 ASSERT3U(blocksize
, <=, SPA_MAXBLOCKSIZE
);
461 ASSERT3U(blocksize
% SPA_MINBLOCKSIZE
, ==, 0);
462 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
|| dmu_tx_private_ok(tx
));
463 ASSERT(tx
->tx_txg
!= 0);
464 ASSERT((bonustype
== DMU_OT_NONE
&& bonuslen
== 0) ||
465 (bonustype
!= DMU_OT_NONE
&& bonuslen
!= 0) ||
466 (bonustype
== DMU_OT_SA
&& bonuslen
== 0));
467 ASSERT3U(bonustype
, <, DMU_OT_NUMTYPES
);
468 ASSERT3U(bonuslen
, <=, DN_MAX_BONUSLEN
);
470 /* clean up any unreferenced dbufs */
471 dnode_evict_dbufs(dn
);
475 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
476 dnode_setdirty(dn
, tx
);
477 if (dn
->dn_datablksz
!= blocksize
) {
478 /* change blocksize */
479 ASSERT(dn
->dn_maxblkid
== 0 &&
480 (BP_IS_HOLE(&dn
->dn_phys
->dn_blkptr
[0]) ||
481 dnode_block_freed(dn
, 0)));
482 dnode_setdblksz(dn
, blocksize
);
483 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = blocksize
;
485 if (dn
->dn_bonuslen
!= bonuslen
)
486 dn
->dn_next_bonuslen
[tx
->tx_txg
&TXG_MASK
] = bonuslen
;
488 if (bonustype
== DMU_OT_SA
) /* Maximize bonus space for SA */
491 nblkptr
= 1 + ((DN_MAX_BONUSLEN
- bonuslen
) >> SPA_BLKPTRSHIFT
);
492 if (dn
->dn_bonustype
!= bonustype
)
493 dn
->dn_next_bonustype
[tx
->tx_txg
&TXG_MASK
] = bonustype
;
494 if (dn
->dn_nblkptr
!= nblkptr
)
495 dn
->dn_next_nblkptr
[tx
->tx_txg
&TXG_MASK
] = nblkptr
;
496 if (dn
->dn_phys
->dn_flags
& DNODE_FLAG_SPILL_BLKPTR
) {
497 dbuf_rm_spill(dn
, tx
);
498 dnode_rm_spill(dn
, tx
);
500 rw_exit(&dn
->dn_struct_rwlock
);
505 /* change bonus size and type */
506 mutex_enter(&dn
->dn_mtx
);
507 dn
->dn_bonustype
= bonustype
;
508 dn
->dn_bonuslen
= bonuslen
;
509 dn
->dn_nblkptr
= nblkptr
;
510 dn
->dn_checksum
= ZIO_CHECKSUM_INHERIT
;
511 dn
->dn_compress
= ZIO_COMPRESS_INHERIT
;
512 ASSERT3U(dn
->dn_nblkptr
, <=, DN_MAX_NBLKPTR
);
514 /* fix up the bonus db_size */
516 dn
->dn_bonus
->db
.db_size
=
517 DN_MAX_BONUSLEN
- (dn
->dn_nblkptr
-1) * sizeof (blkptr_t
);
518 ASSERT(dn
->dn_bonuslen
<= dn
->dn_bonus
->db
.db_size
);
521 dn
->dn_allocated_txg
= tx
->tx_txg
;
522 mutex_exit(&dn
->dn_mtx
);
526 dnode_special_close(dnode_t
*dn
)
529 * Wait for final references to the dnode to clear. This can
530 * only happen if the arc is asyncronously evicting state that
531 * has a hold on this dnode while we are trying to evict this
534 while (refcount_count(&dn
->dn_holds
) > 0)
540 dnode_special_open(objset_t
*os
, dnode_phys_t
*dnp
, uint64_t object
)
542 dnode_t
*dn
= dnode_create(os
, dnp
, NULL
, object
);
548 dnode_buf_pageout(dmu_buf_t
*db
, void *arg
)
550 dnode_t
**children_dnodes
= arg
;
552 int epb
= db
->db_size
>> DNODE_SHIFT
;
554 for (i
= 0; i
< epb
; i
++) {
555 dnode_t
*dn
= children_dnodes
[i
];
562 * If there are holds on this dnode, then there should
563 * be holds on the dnode's containing dbuf as well; thus
564 * it wouldn't be eligable for eviction and this function
565 * would not have been called.
567 ASSERT(refcount_is_zero(&dn
->dn_holds
));
568 ASSERT(list_head(&dn
->dn_dbufs
) == NULL
);
569 ASSERT(refcount_is_zero(&dn
->dn_tx_holds
));
571 for (n
= 0; n
< TXG_SIZE
; n
++)
572 ASSERT(!list_link_active(&dn
->dn_dirty_link
[n
]));
574 children_dnodes
[i
] = NULL
;
577 kmem_free(children_dnodes
, epb
* sizeof (dnode_t
*));
582 * EINVAL - invalid object number.
584 * succeeds even for free dnodes.
587 dnode_hold_impl(objset_t
*os
, uint64_t object
, int flag
,
588 void *tag
, dnode_t
**dnp
)
591 int drop_struct_lock
= FALSE
;
596 dnode_t
**children_dnodes
;
599 * If you are holding the spa config lock as writer, you shouldn't
600 * be asking the DMU to do *anything* unless it's the root pool
601 * which may require us to read from the root filesystem while
602 * holding some (not all) of the locks as writer.
604 ASSERT(spa_config_held(os
->os_spa
, SCL_ALL
, RW_WRITER
) == 0 ||
605 (spa_is_root(os
->os_spa
) &&
606 spa_config_held(os
->os_spa
, SCL_STATE
, RW_WRITER
) &&
607 !spa_config_held(os
->os_spa
, SCL_ZIO
, RW_WRITER
)));
609 if (object
== DMU_USERUSED_OBJECT
|| object
== DMU_GROUPUSED_OBJECT
) {
610 dn
= (object
== DMU_USERUSED_OBJECT
) ?
611 os
->os_userused_dnode
: os
->os_groupused_dnode
;
615 if ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
)
617 if ((flag
& DNODE_MUST_BE_FREE
) && type
!= DMU_OT_NONE
)
620 (void) refcount_add(&dn
->dn_holds
, tag
);
625 if (object
== 0 || object
>= DN_MAX_OBJECT
)
628 mdn
= os
->os_meta_dnode
;
632 if (!RW_WRITE_HELD(&mdn
->dn_struct_rwlock
)) {
633 rw_enter(&mdn
->dn_struct_rwlock
, RW_READER
);
634 drop_struct_lock
= TRUE
;
637 blk
= dbuf_whichblock(mdn
, object
* sizeof (dnode_phys_t
));
639 db
= dbuf_hold(mdn
, blk
, FTAG
);
640 if (drop_struct_lock
)
641 rw_exit(&mdn
->dn_struct_rwlock
);
644 err
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
);
650 ASSERT3U(db
->db
.db_size
, >=, 1<<DNODE_SHIFT
);
651 epb
= db
->db
.db_size
>> DNODE_SHIFT
;
653 idx
= object
& (epb
-1);
655 children_dnodes
= dmu_buf_get_user(&db
->db
);
656 if (children_dnodes
== NULL
) {
658 children_dnodes
= kmem_zalloc(epb
* sizeof (dnode_t
*),
660 if (winner
= dmu_buf_set_user(&db
->db
, children_dnodes
, NULL
,
661 dnode_buf_pageout
)) {
662 kmem_free(children_dnodes
, epb
* sizeof (dnode_t
*));
663 children_dnodes
= winner
;
667 if ((dn
= children_dnodes
[idx
]) == NULL
) {
668 dnode_phys_t
*dnp
= (dnode_phys_t
*)db
->db
.db_data
+idx
;
671 dn
= dnode_create(os
, dnp
, db
, object
);
672 winner
= atomic_cas_ptr(&children_dnodes
[idx
], NULL
, dn
);
673 if (winner
!= NULL
) {
679 mutex_enter(&dn
->dn_mtx
);
681 if (dn
->dn_free_txg
||
682 ((flag
& DNODE_MUST_BE_ALLOCATED
) && type
== DMU_OT_NONE
) ||
683 ((flag
& DNODE_MUST_BE_FREE
) &&
684 (type
!= DMU_OT_NONE
|| !refcount_is_zero(&dn
->dn_holds
)))) {
685 mutex_exit(&dn
->dn_mtx
);
687 return (type
== DMU_OT_NONE
? ENOENT
: EEXIST
);
689 mutex_exit(&dn
->dn_mtx
);
691 if (refcount_add(&dn
->dn_holds
, tag
) == 1)
692 dbuf_add_ref(db
, dn
);
695 ASSERT3P(dn
->dn_dbuf
, ==, db
);
696 ASSERT3U(dn
->dn_object
, ==, object
);
704 * Return held dnode if the object is allocated, NULL if not.
707 dnode_hold(objset_t
*os
, uint64_t object
, void *tag
, dnode_t
**dnp
)
709 return (dnode_hold_impl(os
, object
, DNODE_MUST_BE_ALLOCATED
, tag
, dnp
));
713 * Can only add a reference if there is already at least one
714 * reference on the dnode. Returns FALSE if unable to add a
718 dnode_add_ref(dnode_t
*dn
, void *tag
)
720 mutex_enter(&dn
->dn_mtx
);
721 if (refcount_is_zero(&dn
->dn_holds
)) {
722 mutex_exit(&dn
->dn_mtx
);
725 VERIFY(1 < refcount_add(&dn
->dn_holds
, tag
));
726 mutex_exit(&dn
->dn_mtx
);
731 dnode_rele(dnode_t
*dn
, void *tag
)
735 mutex_enter(&dn
->dn_mtx
);
736 refs
= refcount_remove(&dn
->dn_holds
, tag
);
737 mutex_exit(&dn
->dn_mtx
);
738 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
739 if (refs
== 0 && dn
->dn_dbuf
)
740 dbuf_rele(dn
->dn_dbuf
, dn
);
744 dnode_setdirty(dnode_t
*dn
, dmu_tx_t
*tx
)
746 objset_t
*os
= dn
->dn_objset
;
747 uint64_t txg
= tx
->tx_txg
;
749 if (DMU_OBJECT_IS_SPECIAL(dn
->dn_object
)) {
750 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
757 mutex_enter(&dn
->dn_mtx
);
758 ASSERT(dn
->dn_phys
->dn_type
|| dn
->dn_allocated_txg
);
759 /* ASSERT(dn->dn_free_txg == 0 || dn->dn_free_txg >= txg); */
760 mutex_exit(&dn
->dn_mtx
);
764 * Determine old uid/gid when necessary
766 dmu_objset_userquota_get_ids(dn
, B_TRUE
, tx
);
768 mutex_enter(&os
->os_lock
);
771 * If we are already marked dirty, we're done.
773 if (list_link_active(&dn
->dn_dirty_link
[txg
& TXG_MASK
])) {
774 mutex_exit(&os
->os_lock
);
778 ASSERT(!refcount_is_zero(&dn
->dn_holds
) || list_head(&dn
->dn_dbufs
));
779 ASSERT(dn
->dn_datablksz
!= 0);
780 ASSERT3U(dn
->dn_next_bonuslen
[txg
&TXG_MASK
], ==, 0);
781 ASSERT3U(dn
->dn_next_blksz
[txg
&TXG_MASK
], ==, 0);
782 ASSERT3U(dn
->dn_next_bonustype
[txg
&TXG_MASK
], ==, 0);
784 dprintf_ds(os
->os_dsl_dataset
, "obj=%llu txg=%llu\n",
787 if (dn
->dn_free_txg
> 0 && dn
->dn_free_txg
<= txg
) {
788 list_insert_tail(&os
->os_free_dnodes
[txg
&TXG_MASK
], dn
);
790 list_insert_tail(&os
->os_dirty_dnodes
[txg
&TXG_MASK
], dn
);
793 mutex_exit(&os
->os_lock
);
796 * The dnode maintains a hold on its containing dbuf as
797 * long as there are holds on it. Each instantiated child
798 * dbuf maintaines a hold on the dnode. When the last child
799 * drops its hold, the dnode will drop its hold on the
800 * containing dbuf. We add a "dirty hold" here so that the
801 * dnode will hang around after we finish processing its
804 VERIFY(dnode_add_ref(dn
, (void *)(uintptr_t)tx
->tx_txg
));
806 (void) dbuf_dirty(dn
->dn_dbuf
, tx
);
808 dsl_dataset_dirty(os
->os_dsl_dataset
, tx
);
812 dnode_free(dnode_t
*dn
, dmu_tx_t
*tx
)
814 int txgoff
= tx
->tx_txg
& TXG_MASK
;
816 dprintf("dn=%p txg=%llu\n", dn
, tx
->tx_txg
);
818 /* we should be the only holder... hopefully */
819 /* ASSERT3U(refcount_count(&dn->dn_holds), ==, 1); */
821 mutex_enter(&dn
->dn_mtx
);
822 if (dn
->dn_type
== DMU_OT_NONE
|| dn
->dn_free_txg
) {
823 mutex_exit(&dn
->dn_mtx
);
826 dn
->dn_free_txg
= tx
->tx_txg
;
827 mutex_exit(&dn
->dn_mtx
);
830 * If the dnode is already dirty, it needs to be moved from
831 * the dirty list to the free list.
833 mutex_enter(&dn
->dn_objset
->os_lock
);
834 if (list_link_active(&dn
->dn_dirty_link
[txgoff
])) {
835 list_remove(&dn
->dn_objset
->os_dirty_dnodes
[txgoff
], dn
);
836 list_insert_tail(&dn
->dn_objset
->os_free_dnodes
[txgoff
], dn
);
837 mutex_exit(&dn
->dn_objset
->os_lock
);
839 mutex_exit(&dn
->dn_objset
->os_lock
);
840 dnode_setdirty(dn
, tx
);
845 * Try to change the block size for the indicated dnode. This can only
846 * succeed if there are no blocks allocated or dirty beyond first block
849 dnode_set_blksz(dnode_t
*dn
, uint64_t size
, int ibs
, dmu_tx_t
*tx
)
851 dmu_buf_impl_t
*db
, *db_next
;
855 size
= SPA_MINBLOCKSIZE
;
856 if (size
> SPA_MAXBLOCKSIZE
)
857 size
= SPA_MAXBLOCKSIZE
;
859 size
= P2ROUNDUP(size
, SPA_MINBLOCKSIZE
);
861 if (ibs
== dn
->dn_indblkshift
)
864 if (size
>> SPA_MINBLOCKSHIFT
== dn
->dn_datablkszsec
&& ibs
== 0)
867 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
869 /* Check for any allocated blocks beyond the first */
870 if (dn
->dn_phys
->dn_maxblkid
!= 0)
873 mutex_enter(&dn
->dn_dbufs_mtx
);
874 for (db
= list_head(&dn
->dn_dbufs
); db
; db
= db_next
) {
875 db_next
= list_next(&dn
->dn_dbufs
, db
);
877 if (db
->db_blkid
!= 0 && db
->db_blkid
!= DMU_BONUS_BLKID
&&
878 db
->db_blkid
!= DMU_SPILL_BLKID
) {
879 mutex_exit(&dn
->dn_dbufs_mtx
);
883 mutex_exit(&dn
->dn_dbufs_mtx
);
885 if (ibs
&& dn
->dn_nlevels
!= 1)
888 /* resize the old block */
889 err
= dbuf_hold_impl(dn
, 0, 0, TRUE
, FTAG
, &db
);
891 dbuf_new_size(db
, size
, tx
);
892 else if (err
!= ENOENT
)
895 dnode_setdblksz(dn
, size
);
896 dnode_setdirty(dn
, tx
);
897 dn
->dn_next_blksz
[tx
->tx_txg
&TXG_MASK
] = size
;
899 dn
->dn_indblkshift
= ibs
;
900 dn
->dn_next_indblkshift
[tx
->tx_txg
&TXG_MASK
] = ibs
;
902 /* rele after we have fixed the blocksize in the dnode */
906 rw_exit(&dn
->dn_struct_rwlock
);
910 rw_exit(&dn
->dn_struct_rwlock
);
914 /* read-holding callers must not rely on the lock being continuously held */
916 dnode_new_blkid(dnode_t
*dn
, uint64_t blkid
, dmu_tx_t
*tx
, boolean_t have_read
)
918 uint64_t txgoff
= tx
->tx_txg
& TXG_MASK
;
919 int epbs
, new_nlevels
;
922 ASSERT(blkid
!= DMU_BONUS_BLKID
);
925 RW_READ_HELD(&dn
->dn_struct_rwlock
) :
926 RW_WRITE_HELD(&dn
->dn_struct_rwlock
));
929 * if we have a read-lock, check to see if we need to do any work
930 * before upgrading to a write-lock.
933 if (blkid
<= dn
->dn_maxblkid
)
936 if (!rw_tryupgrade(&dn
->dn_struct_rwlock
)) {
937 rw_exit(&dn
->dn_struct_rwlock
);
938 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
942 if (blkid
<= dn
->dn_maxblkid
)
945 dn
->dn_maxblkid
= blkid
;
948 * Compute the number of levels necessary to support the new maxblkid.
951 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
952 for (sz
= dn
->dn_nblkptr
;
953 sz
<= blkid
&& sz
>= dn
->dn_nblkptr
; sz
<<= epbs
)
956 if (new_nlevels
> dn
->dn_nlevels
) {
957 int old_nlevels
= dn
->dn_nlevels
;
960 dbuf_dirty_record_t
*new, *dr
, *dr_next
;
962 dn
->dn_nlevels
= new_nlevels
;
964 ASSERT3U(new_nlevels
, >, dn
->dn_next_nlevels
[txgoff
]);
965 dn
->dn_next_nlevels
[txgoff
] = new_nlevels
;
967 /* dirty the left indirects */
968 db
= dbuf_hold_level(dn
, old_nlevels
, 0, FTAG
);
970 new = dbuf_dirty(db
, tx
);
973 /* transfer the dirty records to the new indirect */
974 mutex_enter(&dn
->dn_mtx
);
975 mutex_enter(&new->dt
.di
.dr_mtx
);
976 list
= &dn
->dn_dirty_records
[txgoff
];
977 for (dr
= list_head(list
); dr
; dr
= dr_next
) {
978 dr_next
= list_next(&dn
->dn_dirty_records
[txgoff
], dr
);
979 if (dr
->dr_dbuf
->db_level
!= new_nlevels
-1 &&
980 dr
->dr_dbuf
->db_blkid
!= DMU_BONUS_BLKID
&&
981 dr
->dr_dbuf
->db_blkid
!= DMU_SPILL_BLKID
) {
982 ASSERT(dr
->dr_dbuf
->db_level
== old_nlevels
-1);
983 list_remove(&dn
->dn_dirty_records
[txgoff
], dr
);
984 list_insert_tail(&new->dt
.di
.dr_children
, dr
);
988 mutex_exit(&new->dt
.di
.dr_mtx
);
989 mutex_exit(&dn
->dn_mtx
);
994 rw_downgrade(&dn
->dn_struct_rwlock
);
998 dnode_clear_range(dnode_t
*dn
, uint64_t blkid
, uint64_t nblks
, dmu_tx_t
*tx
)
1000 avl_tree_t
*tree
= &dn
->dn_ranges
[tx
->tx_txg
&TXG_MASK
];
1003 free_range_t rp_tofind
;
1004 uint64_t endblk
= blkid
+ nblks
;
1006 ASSERT(MUTEX_HELD(&dn
->dn_mtx
));
1007 ASSERT(nblks
<= UINT64_MAX
- blkid
); /* no overflow */
1009 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1010 blkid
, nblks
, tx
->tx_txg
);
1011 rp_tofind
.fr_blkid
= blkid
;
1012 rp
= avl_find(tree
, &rp_tofind
, &where
);
1014 rp
= avl_nearest(tree
, where
, AVL_BEFORE
);
1016 rp
= avl_nearest(tree
, where
, AVL_AFTER
);
1018 while (rp
&& (rp
->fr_blkid
<= blkid
+ nblks
)) {
1019 uint64_t fr_endblk
= rp
->fr_blkid
+ rp
->fr_nblks
;
1020 free_range_t
*nrp
= AVL_NEXT(tree
, rp
);
1022 if (blkid
<= rp
->fr_blkid
&& endblk
>= fr_endblk
) {
1023 /* clear this entire range */
1024 avl_remove(tree
, rp
);
1025 kmem_free(rp
, sizeof (free_range_t
));
1026 } else if (blkid
<= rp
->fr_blkid
&&
1027 endblk
> rp
->fr_blkid
&& endblk
< fr_endblk
) {
1028 /* clear the beginning of this range */
1029 rp
->fr_blkid
= endblk
;
1030 rp
->fr_nblks
= fr_endblk
- endblk
;
1031 } else if (blkid
> rp
->fr_blkid
&& blkid
< fr_endblk
&&
1032 endblk
>= fr_endblk
) {
1033 /* clear the end of this range */
1034 rp
->fr_nblks
= blkid
- rp
->fr_blkid
;
1035 } else if (blkid
> rp
->fr_blkid
&& endblk
< fr_endblk
) {
1036 /* clear a chunk out of this range */
1037 free_range_t
*new_rp
=
1038 kmem_alloc(sizeof (free_range_t
), KM_SLEEP
);
1040 new_rp
->fr_blkid
= endblk
;
1041 new_rp
->fr_nblks
= fr_endblk
- endblk
;
1042 avl_insert_here(tree
, new_rp
, rp
, AVL_AFTER
);
1043 rp
->fr_nblks
= blkid
- rp
->fr_blkid
;
1045 /* there may be no overlap */
1051 dnode_free_range(dnode_t
*dn
, uint64_t off
, uint64_t len
, dmu_tx_t
*tx
)
1054 uint64_t blkoff
, blkid
, nblks
;
1055 int blksz
, blkshift
, head
, tail
;
1059 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1060 blksz
= dn
->dn_datablksz
;
1061 blkshift
= dn
->dn_datablkshift
;
1062 epbs
= dn
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1065 len
= UINT64_MAX
- off
;
1070 * First, block align the region to free:
1073 head
= P2NPHASE(off
, blksz
);
1074 blkoff
= P2PHASE(off
, blksz
);
1075 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1078 ASSERT(dn
->dn_maxblkid
== 0);
1079 if (off
== 0 && len
>= blksz
) {
1080 /* Freeing the whole block; fast-track this request */
1084 } else if (off
>= blksz
) {
1085 /* Freeing past end-of-data */
1088 /* Freeing part of the block. */
1090 ASSERT3U(head
, >, 0);
1094 /* zero out any partial block data at the start of the range */
1096 ASSERT3U(blkoff
+ head
, ==, blksz
);
1099 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
), TRUE
,
1103 /* don't dirty if it isn't on disk and isn't dirty */
1104 if (db
->db_last_dirty
||
1105 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1106 rw_exit(&dn
->dn_struct_rwlock
);
1107 dbuf_will_dirty(db
, tx
);
1108 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1109 data
= db
->db
.db_data
;
1110 bzero(data
+ blkoff
, head
);
1112 dbuf_rele(db
, FTAG
);
1118 /* If the range was less than one block, we're done */
1122 /* If the remaining range is past end of file, we're done */
1123 if ((off
>> blkshift
) > dn
->dn_maxblkid
)
1126 ASSERT(ISP2(blksz
));
1130 tail
= P2PHASE(len
, blksz
);
1132 ASSERT3U(P2PHASE(off
, blksz
), ==, 0);
1133 /* zero out any partial block data at the end of the range */
1137 if (dbuf_hold_impl(dn
, 0, dbuf_whichblock(dn
, off
+len
),
1138 TRUE
, FTAG
, &db
) == 0) {
1139 /* don't dirty if not on disk and not dirty */
1140 if (db
->db_last_dirty
||
1141 (db
->db_blkptr
&& !BP_IS_HOLE(db
->db_blkptr
))) {
1142 rw_exit(&dn
->dn_struct_rwlock
);
1143 dbuf_will_dirty(db
, tx
);
1144 rw_enter(&dn
->dn_struct_rwlock
, RW_WRITER
);
1145 bzero(db
->db
.db_data
, tail
);
1147 dbuf_rele(db
, FTAG
);
1152 /* If the range did not include a full block, we are done */
1156 ASSERT(IS_P2ALIGNED(off
, blksz
));
1157 ASSERT(trunc
|| IS_P2ALIGNED(len
, blksz
));
1158 blkid
= off
>> blkshift
;
1159 nblks
= len
>> blkshift
;
1164 * Read in and mark all the level-1 indirects dirty,
1165 * so that they will stay in memory until syncing phase.
1166 * Always dirty the first and last indirect to make sure
1167 * we dirty all the partial indirects.
1169 if (dn
->dn_nlevels
> 1) {
1170 uint64_t i
, first
, last
;
1171 int shift
= epbs
+ dn
->dn_datablkshift
;
1173 first
= blkid
>> epbs
;
1174 if (db
= dbuf_hold_level(dn
, 1, first
, FTAG
)) {
1175 dbuf_will_dirty(db
, tx
);
1176 dbuf_rele(db
, FTAG
);
1179 last
= dn
->dn_maxblkid
>> epbs
;
1181 last
= (blkid
+ nblks
- 1) >> epbs
;
1182 if (last
> first
&& (db
= dbuf_hold_level(dn
, 1, last
, FTAG
))) {
1183 dbuf_will_dirty(db
, tx
);
1184 dbuf_rele(db
, FTAG
);
1186 for (i
= first
+ 1; i
< last
; i
++) {
1187 uint64_t ibyte
= i
<< shift
;
1190 err
= dnode_next_offset(dn
,
1191 DNODE_FIND_HAVELOCK
, &ibyte
, 1, 1, 0);
1193 if (err
== ESRCH
|| i
>= last
)
1196 db
= dbuf_hold_level(dn
, 1, i
, FTAG
);
1198 dbuf_will_dirty(db
, tx
);
1199 dbuf_rele(db
, FTAG
);
1205 * Add this range to the dnode range list.
1206 * We will finish up this free operation in the syncing phase.
1208 mutex_enter(&dn
->dn_mtx
);
1209 dnode_clear_range(dn
, blkid
, nblks
, tx
);
1211 free_range_t
*rp
, *found
;
1213 avl_tree_t
*tree
= &dn
->dn_ranges
[tx
->tx_txg
&TXG_MASK
];
1215 /* Add new range to dn_ranges */
1216 rp
= kmem_alloc(sizeof (free_range_t
), KM_SLEEP
);
1217 rp
->fr_blkid
= blkid
;
1218 rp
->fr_nblks
= nblks
;
1219 found
= avl_find(tree
, rp
, &where
);
1220 ASSERT(found
== NULL
);
1221 avl_insert(tree
, rp
, where
);
1222 dprintf_dnode(dn
, "blkid=%llu nblks=%llu txg=%llu\n",
1223 blkid
, nblks
, tx
->tx_txg
);
1225 mutex_exit(&dn
->dn_mtx
);
1227 dbuf_free_range(dn
, blkid
, blkid
+ nblks
- 1, tx
);
1228 dnode_setdirty(dn
, tx
);
1230 if (trunc
&& dn
->dn_maxblkid
>= (off
>> blkshift
))
1231 dn
->dn_maxblkid
= (off
>> blkshift
? (off
>> blkshift
) - 1 : 0);
1233 rw_exit(&dn
->dn_struct_rwlock
);
1237 dnode_spill_freed(dnode_t
*dn
)
1241 mutex_enter(&dn
->dn_mtx
);
1242 for (i
= 0; i
< TXG_SIZE
; i
++) {
1243 if (dn
->dn_rm_spillblk
[i
] == DN_KILL_SPILLBLK
)
1246 mutex_exit(&dn
->dn_mtx
);
1247 return (i
< TXG_SIZE
);
1250 /* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1252 dnode_block_freed(dnode_t
*dn
, uint64_t blkid
)
1254 free_range_t range_tofind
;
1255 void *dp
= spa_get_dsl(dn
->dn_objset
->os_spa
);
1258 if (blkid
== DMU_BONUS_BLKID
)
1262 * If we're in the process of opening the pool, dp will not be
1263 * set yet, but there shouldn't be anything dirty.
1268 if (dn
->dn_free_txg
)
1271 if (blkid
== DMU_SPILL_BLKID
)
1272 return (dnode_spill_freed(dn
));
1274 range_tofind
.fr_blkid
= blkid
;
1275 mutex_enter(&dn
->dn_mtx
);
1276 for (i
= 0; i
< TXG_SIZE
; i
++) {
1277 free_range_t
*range_found
;
1280 range_found
= avl_find(&dn
->dn_ranges
[i
], &range_tofind
, &idx
);
1282 ASSERT(range_found
->fr_nblks
> 0);
1285 range_found
= avl_nearest(&dn
->dn_ranges
[i
], idx
, AVL_BEFORE
);
1287 range_found
->fr_blkid
+ range_found
->fr_nblks
> blkid
)
1290 mutex_exit(&dn
->dn_mtx
);
1291 return (i
< TXG_SIZE
);
1294 /* call from syncing context when we actually write/free space for this dnode */
1296 dnode_diduse_space(dnode_t
*dn
, int64_t delta
)
1299 dprintf_dnode(dn
, "dn=%p dnp=%p used=%llu delta=%lld\n",
1301 (u_longlong_t
)dn
->dn_phys
->dn_used
,
1304 mutex_enter(&dn
->dn_mtx
);
1305 space
= DN_USED_BYTES(dn
->dn_phys
);
1307 ASSERT3U(space
+ delta
, >=, space
); /* no overflow */
1309 ASSERT3U(space
, >=, -delta
); /* no underflow */
1312 if (spa_version(dn
->dn_objset
->os_spa
) < SPA_VERSION_DNODE_BYTES
) {
1313 ASSERT((dn
->dn_phys
->dn_flags
& DNODE_FLAG_USED_BYTES
) == 0);
1314 ASSERT3U(P2PHASE(space
, 1<<DEV_BSHIFT
), ==, 0);
1315 dn
->dn_phys
->dn_used
= space
>> DEV_BSHIFT
;
1317 dn
->dn_phys
->dn_used
= space
;
1318 dn
->dn_phys
->dn_flags
|= DNODE_FLAG_USED_BYTES
;
1320 mutex_exit(&dn
->dn_mtx
);
1324 * Call when we think we're going to write/free space in open context.
1325 * Be conservative (ie. OK to write less than this or free more than
1326 * this, but don't write more or free less).
1329 dnode_willuse_space(dnode_t
*dn
, int64_t space
, dmu_tx_t
*tx
)
1331 objset_t
*os
= dn
->dn_objset
;
1332 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1335 space
= spa_get_asize(os
->os_spa
, space
);
1338 dsl_dir_willuse_space(ds
->ds_dir
, space
, tx
);
1340 dmu_tx_willuse_space(tx
, space
);
1344 * This function scans a block at the indicated "level" looking for
1345 * a hole or data (depending on 'flags'). If level > 0, then we are
1346 * scanning an indirect block looking at its pointers. If level == 0,
1347 * then we are looking at a block of dnodes. If we don't find what we
1348 * are looking for in the block, we return ESRCH. Otherwise, return
1349 * with *offset pointing to the beginning (if searching forwards) or
1350 * end (if searching backwards) of the range covered by the block
1351 * pointer we matched on (or dnode).
1353 * The basic search algorithm used below by dnode_next_offset() is to
1354 * use this function to search up the block tree (widen the search) until
1355 * we find something (i.e., we don't return ESRCH) and then search back
1356 * down the tree (narrow the search) until we reach our original search
1360 dnode_next_offset_level(dnode_t
*dn
, int flags
, uint64_t *offset
,
1361 int lvl
, uint64_t blkfill
, uint64_t txg
)
1363 dmu_buf_impl_t
*db
= NULL
;
1365 uint64_t epbs
= dn
->dn_phys
->dn_indblkshift
- SPA_BLKPTRSHIFT
;
1366 uint64_t epb
= 1ULL << epbs
;
1367 uint64_t minfill
, maxfill
;
1369 int i
, inc
, error
, span
;
1371 dprintf("probing object %llu offset %llx level %d of %u\n",
1372 dn
->dn_object
, *offset
, lvl
, dn
->dn_phys
->dn_nlevels
);
1374 hole
= ((flags
& DNODE_FIND_HOLE
) != 0);
1375 inc
= (flags
& DNODE_FIND_BACKWARDS
) ? -1 : 1;
1376 ASSERT(txg
== 0 || !hole
);
1378 if (lvl
== dn
->dn_phys
->dn_nlevels
) {
1380 epb
= dn
->dn_phys
->dn_nblkptr
;
1381 data
= dn
->dn_phys
->dn_blkptr
;
1383 uint64_t blkid
= dbuf_whichblock(dn
, *offset
) >> (epbs
* lvl
);
1384 error
= dbuf_hold_impl(dn
, lvl
, blkid
, TRUE
, FTAG
, &db
);
1386 if (error
!= ENOENT
)
1391 * This can only happen when we are searching up
1392 * the block tree for data. We don't really need to
1393 * adjust the offset, as we will just end up looking
1394 * at the pointer to this block in its parent, and its
1395 * going to be unallocated, so we will skip over it.
1399 error
= dbuf_read(db
, NULL
, DB_RF_CANFAIL
| DB_RF_HAVESTRUCT
);
1401 dbuf_rele(db
, FTAG
);
1404 data
= db
->db
.db_data
;
1408 (db
->db_blkptr
== NULL
|| db
->db_blkptr
->blk_birth
<= txg
)) {
1410 * This can only happen when we are searching up the tree
1411 * and these conditions mean that we need to keep climbing.
1414 } else if (lvl
== 0) {
1415 dnode_phys_t
*dnp
= data
;
1417 ASSERT(dn
->dn_type
== DMU_OT_DNODE
);
1419 for (i
= (*offset
>> span
) & (blkfill
- 1);
1420 i
>= 0 && i
< blkfill
; i
+= inc
) {
1421 if ((dnp
[i
].dn_type
== DMU_OT_NONE
) == hole
)
1423 *offset
+= (1ULL << span
) * inc
;
1425 if (i
< 0 || i
== blkfill
)
1428 blkptr_t
*bp
= data
;
1429 uint64_t start
= *offset
;
1430 span
= (lvl
- 1) * epbs
+ dn
->dn_datablkshift
;
1432 maxfill
= blkfill
<< ((lvl
- 1) * epbs
);
1439 *offset
= *offset
>> span
;
1440 for (i
= BF64_GET(*offset
, 0, epbs
);
1441 i
>= 0 && i
< epb
; i
+= inc
) {
1442 if (bp
[i
].blk_fill
>= minfill
&&
1443 bp
[i
].blk_fill
<= maxfill
&&
1444 (hole
|| bp
[i
].blk_birth
> txg
))
1446 if (inc
> 0 || *offset
> 0)
1449 *offset
= *offset
<< span
;
1451 /* traversing backwards; position offset at the end */
1452 ASSERT3U(*offset
, <=, start
);
1453 *offset
= MIN(*offset
+ (1ULL << span
) - 1, start
);
1454 } else if (*offset
< start
) {
1457 if (i
< 0 || i
>= epb
)
1462 dbuf_rele(db
, FTAG
);
1468 * Find the next hole, data, or sparse region at or after *offset.
1469 * The value 'blkfill' tells us how many items we expect to find
1470 * in an L0 data block; this value is 1 for normal objects,
1471 * DNODES_PER_BLOCK for the meta dnode, and some fraction of
1472 * DNODES_PER_BLOCK when searching for sparse regions thereof.
1476 * dnode_next_offset(dn, flags, offset, 1, 1, 0);
1477 * Finds the next/previous hole/data in a file.
1478 * Used in dmu_offset_next().
1480 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg);
1481 * Finds the next free/allocated dnode an objset's meta-dnode.
1482 * Only finds objects that have new contents since txg (ie.
1483 * bonus buffer changes and content removal are ignored).
1484 * Used in dmu_object_next().
1486 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0);
1487 * Finds the next L2 meta-dnode bp that's at most 1/4 full.
1488 * Used in dmu_object_alloc().
1491 dnode_next_offset(dnode_t
*dn
, int flags
, uint64_t *offset
,
1492 int minlvl
, uint64_t blkfill
, uint64_t txg
)
1494 uint64_t initial_offset
= *offset
;
1498 if (!(flags
& DNODE_FIND_HAVELOCK
))
1499 rw_enter(&dn
->dn_struct_rwlock
, RW_READER
);
1501 if (dn
->dn_phys
->dn_nlevels
== 0) {
1506 if (dn
->dn_datablkshift
== 0) {
1507 if (*offset
< dn
->dn_datablksz
) {
1508 if (flags
& DNODE_FIND_HOLE
)
1509 *offset
= dn
->dn_datablksz
;
1516 maxlvl
= dn
->dn_phys
->dn_nlevels
;
1518 for (lvl
= minlvl
; lvl
<= maxlvl
; lvl
++) {
1519 error
= dnode_next_offset_level(dn
,
1520 flags
, offset
, lvl
, blkfill
, txg
);
1525 while (error
== 0 && --lvl
>= minlvl
) {
1526 error
= dnode_next_offset_level(dn
,
1527 flags
, offset
, lvl
, blkfill
, txg
);
1530 if (error
== 0 && (flags
& DNODE_FIND_BACKWARDS
?
1531 initial_offset
< *offset
: initial_offset
> *offset
))
1534 if (!(flags
& DNODE_FIND_HAVELOCK
))
1535 rw_exit(&dn
->dn_struct_rwlock
);