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.
28 #include <sys/zfs_context.h>
30 #include <sys/refcount.h>
31 #include <sys/zap_impl.h>
32 #include <sys/zap_leaf.h>
37 #include <sys/sunddi.h>
40 static int mzap_upgrade(zap_t
**zapp
, dmu_tx_t
*tx
, zap_flags_t flags
);
43 zap_getflags(zap_t
*zap
)
47 return (zap
->zap_u
.zap_fat
.zap_phys
->zap_flags
);
51 zap_hashbits(zap_t
*zap
)
53 if (zap_getflags(zap
) & ZAP_FLAG_HASH64
)
62 if (zap_getflags(zap
) & ZAP_FLAG_HASH64
)
69 zap_hash(zap_name_t
*zn
)
71 zap_t
*zap
= zn
->zn_zap
;
74 if (zap_getflags(zap
) & ZAP_FLAG_PRE_HASHED_KEY
) {
75 ASSERT(zap_getflags(zap
) & ZAP_FLAG_UINT64_KEY
);
76 h
= *(uint64_t *)zn
->zn_key_orig
;
80 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
82 if (zap_getflags(zap
) & ZAP_FLAG_UINT64_KEY
) {
84 const uint64_t *wp
= zn
->zn_key_norm
;
86 ASSERT(zn
->zn_key_intlen
== 8);
87 for (i
= 0; i
< zn
->zn_key_norm_numints
; wp
++, i
++) {
91 for (j
= 0; j
< zn
->zn_key_intlen
; j
++) {
93 zfs_crc64_table
[(h
^ word
) & 0xFF];
99 const uint8_t *cp
= zn
->zn_key_norm
;
102 * We previously stored the terminating null on
103 * disk, but didn't hash it, so we need to
104 * continue to not hash it. (The
105 * zn_key_*_numints includes the terminating
106 * null for non-binary keys.)
108 len
= zn
->zn_key_norm_numints
- 1;
110 ASSERT(zn
->zn_key_intlen
== 1);
111 for (i
= 0; i
< len
; cp
++, i
++) {
113 zfs_crc64_table
[(h
^ *cp
) & 0xFF];
118 * Don't use all 64 bits, since we need some in the cookie for
119 * the collision differentiator. We MUST use the high bits,
120 * since those are the ones that we first pay attention to when
121 * chosing the bucket.
123 h
&= ~((1ULL << (64 - zap_hashbits(zap
))) - 1);
129 zap_normalize(zap_t
*zap
, const char *name
, char *namenorm
)
131 size_t inlen
, outlen
;
134 ASSERT(!(zap_getflags(zap
) & ZAP_FLAG_UINT64_KEY
));
136 inlen
= strlen(name
) + 1;
137 outlen
= ZAP_MAXNAMELEN
;
140 (void) u8_textprep_str((char *)name
, &inlen
, namenorm
, &outlen
,
141 zap
->zap_normflags
| U8_TEXTPREP_IGNORE_NULL
|
142 U8_TEXTPREP_IGNORE_INVALID
, U8_UNICODE_LATEST
, &err
);
148 zap_match(zap_name_t
*zn
, const char *matchname
)
150 ASSERT(!(zap_getflags(zn
->zn_zap
) & ZAP_FLAG_UINT64_KEY
));
152 if (zn
->zn_matchtype
== MT_FIRST
) {
153 char norm
[ZAP_MAXNAMELEN
];
155 if (zap_normalize(zn
->zn_zap
, matchname
, norm
) != 0)
158 return (strcmp(zn
->zn_key_norm
, norm
) == 0);
160 /* MT_BEST or MT_EXACT */
161 return (strcmp(zn
->zn_key_orig
, matchname
) == 0);
166 zap_name_free(zap_name_t
*zn
)
168 kmem_free(zn
, sizeof (zap_name_t
));
172 zap_name_alloc(zap_t
*zap
, const char *key
, matchtype_t mt
)
174 zap_name_t
*zn
= kmem_alloc(sizeof (zap_name_t
), KM_SLEEP
);
177 zn
->zn_key_intlen
= sizeof (*key
);
178 zn
->zn_key_orig
= key
;
179 zn
->zn_key_orig_numints
= strlen(zn
->zn_key_orig
) + 1;
180 zn
->zn_matchtype
= mt
;
181 if (zap
->zap_normflags
) {
182 if (zap_normalize(zap
, key
, zn
->zn_normbuf
) != 0) {
186 zn
->zn_key_norm
= zn
->zn_normbuf
;
187 zn
->zn_key_norm_numints
= strlen(zn
->zn_key_norm
) + 1;
189 if (mt
!= MT_EXACT
) {
193 zn
->zn_key_norm
= zn
->zn_key_orig
;
194 zn
->zn_key_norm_numints
= zn
->zn_key_orig_numints
;
197 zn
->zn_hash
= zap_hash(zn
);
202 zap_name_alloc_uint64(zap_t
*zap
, const uint64_t *key
, int numints
)
204 zap_name_t
*zn
= kmem_alloc(sizeof (zap_name_t
), KM_SLEEP
);
206 ASSERT(zap
->zap_normflags
== 0);
208 zn
->zn_key_intlen
= sizeof (*key
);
209 zn
->zn_key_orig
= zn
->zn_key_norm
= key
;
210 zn
->zn_key_orig_numints
= zn
->zn_key_norm_numints
= numints
;
211 zn
->zn_matchtype
= MT_EXACT
;
213 zn
->zn_hash
= zap_hash(zn
);
218 mzap_byteswap(mzap_phys_t
*buf
, size_t size
)
221 buf
->mz_block_type
= BSWAP_64(buf
->mz_block_type
);
222 buf
->mz_salt
= BSWAP_64(buf
->mz_salt
);
223 buf
->mz_normflags
= BSWAP_64(buf
->mz_normflags
);
224 max
= (size
/ MZAP_ENT_LEN
) - 1;
225 for (i
= 0; i
< max
; i
++) {
226 buf
->mz_chunk
[i
].mze_value
=
227 BSWAP_64(buf
->mz_chunk
[i
].mze_value
);
228 buf
->mz_chunk
[i
].mze_cd
=
229 BSWAP_32(buf
->mz_chunk
[i
].mze_cd
);
234 zap_byteswap(void *buf
, size_t size
)
238 block_type
= *(uint64_t *)buf
;
240 if (block_type
== ZBT_MICRO
|| block_type
== BSWAP_64(ZBT_MICRO
)) {
241 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
242 mzap_byteswap(buf
, size
);
244 fzap_byteswap(buf
, size
);
249 mze_compare(const void *arg1
, const void *arg2
)
251 const mzap_ent_t
*mze1
= arg1
;
252 const mzap_ent_t
*mze2
= arg2
;
254 if (mze1
->mze_hash
> mze2
->mze_hash
)
256 if (mze1
->mze_hash
< mze2
->mze_hash
)
258 if (mze1
->mze_cd
> mze2
->mze_cd
)
260 if (mze1
->mze_cd
< mze2
->mze_cd
)
266 mze_insert(zap_t
*zap
, int chunkid
, uint64_t hash
)
270 ASSERT(zap
->zap_ismicro
);
271 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
273 mze
= kmem_alloc(sizeof (mzap_ent_t
), KM_SLEEP
);
274 mze
->mze_chunkid
= chunkid
;
275 mze
->mze_hash
= hash
;
276 mze
->mze_cd
= MZE_PHYS(zap
, mze
)->mze_cd
;
277 ASSERT(MZE_PHYS(zap
, mze
)->mze_name
[0] != 0);
278 avl_add(&zap
->zap_m
.zap_avl
, mze
);
282 mze_find(zap_name_t
*zn
)
284 mzap_ent_t mze_tofind
;
287 avl_tree_t
*avl
= &zn
->zn_zap
->zap_m
.zap_avl
;
289 ASSERT(zn
->zn_zap
->zap_ismicro
);
290 ASSERT(RW_LOCK_HELD(&zn
->zn_zap
->zap_rwlock
));
292 mze_tofind
.mze_hash
= zn
->zn_hash
;
293 mze_tofind
.mze_cd
= 0;
296 mze
= avl_find(avl
, &mze_tofind
, &idx
);
298 mze
= avl_nearest(avl
, idx
, AVL_AFTER
);
299 for (; mze
&& mze
->mze_hash
== zn
->zn_hash
; mze
= AVL_NEXT(avl
, mze
)) {
300 ASSERT3U(mze
->mze_cd
, ==, MZE_PHYS(zn
->zn_zap
, mze
)->mze_cd
);
301 if (zap_match(zn
, MZE_PHYS(zn
->zn_zap
, mze
)->mze_name
))
304 if (zn
->zn_matchtype
== MT_BEST
) {
305 zn
->zn_matchtype
= MT_FIRST
;
312 mze_find_unused_cd(zap_t
*zap
, uint64_t hash
)
314 mzap_ent_t mze_tofind
;
317 avl_tree_t
*avl
= &zap
->zap_m
.zap_avl
;
320 ASSERT(zap
->zap_ismicro
);
321 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
323 mze_tofind
.mze_hash
= hash
;
324 mze_tofind
.mze_cd
= 0;
327 for (mze
= avl_find(avl
, &mze_tofind
, &idx
);
328 mze
&& mze
->mze_hash
== hash
; mze
= AVL_NEXT(avl
, mze
)) {
329 if (mze
->mze_cd
!= cd
)
338 mze_remove(zap_t
*zap
, mzap_ent_t
*mze
)
340 ASSERT(zap
->zap_ismicro
);
341 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
343 avl_remove(&zap
->zap_m
.zap_avl
, mze
);
344 kmem_free(mze
, sizeof (mzap_ent_t
));
348 mze_destroy(zap_t
*zap
)
351 void *avlcookie
= NULL
;
353 while ((mze
= avl_destroy_nodes(&zap
->zap_m
.zap_avl
, &avlcookie
)))
354 kmem_free(mze
, sizeof (mzap_ent_t
));
355 avl_destroy(&zap
->zap_m
.zap_avl
);
359 mzap_open(objset_t
*os
, uint64_t obj
, dmu_buf_t
*db
)
365 ASSERT3U(MZAP_ENT_LEN
, ==, sizeof (mzap_ent_phys_t
));
367 zap
= kmem_zalloc(sizeof (zap_t
), KM_SLEEP
);
368 rw_init(&zap
->zap_rwlock
, NULL
, RW_DEFAULT
, NULL
);
369 rw_enter(&zap
->zap_rwlock
, RW_WRITER
);
370 zap
->zap_objset
= os
;
371 zap
->zap_object
= obj
;
374 if (*(uint64_t *)db
->db_data
!= ZBT_MICRO
) {
375 mutex_init(&zap
->zap_f
.zap_num_entries_mtx
, 0, 0, 0);
376 zap
->zap_f
.zap_block_shift
= highbit(db
->db_size
) - 1;
378 zap
->zap_ismicro
= TRUE
;
382 * Make sure that zap_ismicro is set before we let others see
383 * it, because zap_lockdir() checks zap_ismicro without the lock
386 winner
= dmu_buf_set_user(db
, zap
, &zap
->zap_m
.zap_phys
, zap_evict
);
388 if (winner
!= NULL
) {
389 rw_exit(&zap
->zap_rwlock
);
390 rw_destroy(&zap
->zap_rwlock
);
391 if (!zap
->zap_ismicro
)
392 mutex_destroy(&zap
->zap_f
.zap_num_entries_mtx
);
393 kmem_free(zap
, sizeof (zap_t
));
397 if (zap
->zap_ismicro
) {
398 zap
->zap_salt
= zap
->zap_m
.zap_phys
->mz_salt
;
399 zap
->zap_normflags
= zap
->zap_m
.zap_phys
->mz_normflags
;
400 zap
->zap_m
.zap_num_chunks
= db
->db_size
/ MZAP_ENT_LEN
- 1;
401 avl_create(&zap
->zap_m
.zap_avl
, mze_compare
,
402 sizeof (mzap_ent_t
), offsetof(mzap_ent_t
, mze_node
));
404 for (i
= 0; i
< zap
->zap_m
.zap_num_chunks
; i
++) {
405 mzap_ent_phys_t
*mze
=
406 &zap
->zap_m
.zap_phys
->mz_chunk
[i
];
407 if (mze
->mze_name
[0]) {
410 zap
->zap_m
.zap_num_entries
++;
411 zn
= zap_name_alloc(zap
, mze
->mze_name
,
413 mze_insert(zap
, i
, zn
->zn_hash
);
418 zap
->zap_salt
= zap
->zap_f
.zap_phys
->zap_salt
;
419 zap
->zap_normflags
= zap
->zap_f
.zap_phys
->zap_normflags
;
421 ASSERT3U(sizeof (struct zap_leaf_header
), ==,
422 2*ZAP_LEAF_CHUNKSIZE
);
425 * The embedded pointer table should not overlap the
428 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap
, 0), >,
429 &zap
->zap_f
.zap_phys
->zap_salt
);
432 * The embedded pointer table should end at the end of
435 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap
,
436 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap
)) -
437 (uintptr_t)zap
->zap_f
.zap_phys
, ==,
438 zap
->zap_dbuf
->db_size
);
440 rw_exit(&zap
->zap_rwlock
);
445 zap_lockdir(objset_t
*os
, uint64_t obj
, dmu_tx_t
*tx
,
446 krw_t lti
, boolean_t fatreader
, boolean_t adding
, zap_t
**zapp
)
455 err
= dmu_buf_hold(os
, obj
, 0, NULL
, &db
, DMU_READ_NO_PREFETCH
);
461 dmu_object_info_t doi
;
462 dmu_object_info_from_db(db
, &doi
);
463 ASSERT(dmu_ot
[doi
.doi_type
].ot_byteswap
== zap_byteswap
);
467 zap
= dmu_buf_get_user(db
);
469 zap
= mzap_open(os
, obj
, db
);
472 * We're checking zap_ismicro without the lock held, in order to
473 * tell what type of lock we want. Once we have some sort of
474 * lock, see if it really is the right type. In practice this
475 * can only be different if it was upgraded from micro to fat,
476 * and micro wanted WRITER but fat only needs READER.
478 lt
= (!zap
->zap_ismicro
&& fatreader
) ? RW_READER
: lti
;
479 rw_enter(&zap
->zap_rwlock
, lt
);
480 if (lt
!= ((!zap
->zap_ismicro
&& fatreader
) ? RW_READER
: lti
)) {
481 /* it was upgraded, now we only need reader */
482 ASSERT(lt
== RW_WRITER
);
484 (!zap
->zap_ismicro
&& fatreader
) ? RW_READER
: lti
);
485 rw_downgrade(&zap
->zap_rwlock
);
489 zap
->zap_objset
= os
;
492 dmu_buf_will_dirty(db
, tx
);
494 ASSERT3P(zap
->zap_dbuf
, ==, db
);
496 ASSERT(!zap
->zap_ismicro
||
497 zap
->zap_m
.zap_num_entries
<= zap
->zap_m
.zap_num_chunks
);
498 if (zap
->zap_ismicro
&& tx
&& adding
&&
499 zap
->zap_m
.zap_num_entries
== zap
->zap_m
.zap_num_chunks
) {
500 uint64_t newsz
= db
->db_size
+ SPA_MINBLOCKSIZE
;
501 if (newsz
> MZAP_MAX_BLKSZ
) {
502 dprintf("upgrading obj %llu: num_entries=%u\n",
503 obj
, zap
->zap_m
.zap_num_entries
);
505 return (mzap_upgrade(zapp
, tx
, 0));
507 err
= dmu_object_set_blocksize(os
, obj
, newsz
, 0, tx
);
508 ASSERT3U(err
, ==, 0);
509 zap
->zap_m
.zap_num_chunks
=
510 db
->db_size
/ MZAP_ENT_LEN
- 1;
518 zap_unlockdir(zap_t
*zap
)
520 rw_exit(&zap
->zap_rwlock
);
521 dmu_buf_rele(zap
->zap_dbuf
, NULL
);
525 mzap_upgrade(zap_t
**zapp
, dmu_tx_t
*tx
, zap_flags_t flags
)
532 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
534 sz
= zap
->zap_dbuf
->db_size
;
535 mzp
= kmem_alloc(sz
, KM_SLEEP
);
536 bcopy(zap
->zap_dbuf
->db_data
, mzp
, sz
);
537 nchunks
= zap
->zap_m
.zap_num_chunks
;
540 err
= dmu_object_set_blocksize(zap
->zap_objset
, zap
->zap_object
,
541 1ULL << fzap_default_block_shift
, 0, tx
);
548 dprintf("upgrading obj=%llu with %u chunks\n",
549 zap
->zap_object
, nchunks
);
550 /* XXX destroy the avl later, so we can use the stored hash value */
553 fzap_upgrade(zap
, tx
, flags
);
555 for (i
= 0; i
< nchunks
; i
++) {
556 mzap_ent_phys_t
*mze
= &mzp
->mz_chunk
[i
];
558 if (mze
->mze_name
[0] == 0)
560 dprintf("adding %s=%llu\n",
561 mze
->mze_name
, mze
->mze_value
);
562 zn
= zap_name_alloc(zap
, mze
->mze_name
, MT_EXACT
);
563 err
= fzap_add_cd(zn
, 8, 1, &mze
->mze_value
, mze
->mze_cd
, tx
);
564 zap
= zn
->zn_zap
; /* fzap_add_cd() may change zap */
575 mzap_create_impl(objset_t
*os
, uint64_t obj
, int normflags
, zap_flags_t flags
,
581 VERIFY(0 == dmu_buf_hold(os
, obj
, 0, FTAG
, &db
, DMU_READ_NO_PREFETCH
));
585 dmu_object_info_t doi
;
586 dmu_object_info_from_db(db
, &doi
);
587 ASSERT(dmu_ot
[doi
.doi_type
].ot_byteswap
== zap_byteswap
);
591 dmu_buf_will_dirty(db
, tx
);
593 zp
->mz_block_type
= ZBT_MICRO
;
594 zp
->mz_salt
= ((uintptr_t)db
^ (uintptr_t)tx
^ (obj
<< 1)) | 1ULL;
595 zp
->mz_normflags
= normflags
;
596 dmu_buf_rele(db
, FTAG
);
600 /* Only fat zap supports flags; upgrade immediately. */
601 VERIFY(0 == zap_lockdir(os
, obj
, tx
, RW_WRITER
,
602 B_FALSE
, B_FALSE
, &zap
));
603 VERIFY3U(0, ==, mzap_upgrade(&zap
, tx
, flags
));
609 zap_create_claim(objset_t
*os
, uint64_t obj
, dmu_object_type_t ot
,
610 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
612 return (zap_create_claim_norm(os
, obj
,
613 0, ot
, bonustype
, bonuslen
, tx
));
617 zap_create_claim_norm(objset_t
*os
, uint64_t obj
, int normflags
,
618 dmu_object_type_t ot
,
619 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
623 err
= dmu_object_claim(os
, obj
, ot
, 0, bonustype
, bonuslen
, tx
);
626 mzap_create_impl(os
, obj
, normflags
, 0, tx
);
631 zap_create(objset_t
*os
, dmu_object_type_t ot
,
632 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
634 return (zap_create_norm(os
, 0, ot
, bonustype
, bonuslen
, tx
));
638 zap_create_norm(objset_t
*os
, int normflags
, dmu_object_type_t ot
,
639 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
641 uint64_t obj
= dmu_object_alloc(os
, ot
, 0, bonustype
, bonuslen
, tx
);
643 mzap_create_impl(os
, obj
, normflags
, 0, tx
);
648 zap_create_flags(objset_t
*os
, int normflags
, zap_flags_t flags
,
649 dmu_object_type_t ot
, int leaf_blockshift
, int indirect_blockshift
,
650 dmu_object_type_t bonustype
, int bonuslen
, dmu_tx_t
*tx
)
652 uint64_t obj
= dmu_object_alloc(os
, ot
, 0, bonustype
, bonuslen
, tx
);
654 ASSERT(leaf_blockshift
>= SPA_MINBLOCKSHIFT
&&
655 leaf_blockshift
<= SPA_MAXBLOCKSHIFT
&&
656 indirect_blockshift
>= SPA_MINBLOCKSHIFT
&&
657 indirect_blockshift
<= SPA_MAXBLOCKSHIFT
);
659 VERIFY(dmu_object_set_blocksize(os
, obj
,
660 1ULL << leaf_blockshift
, indirect_blockshift
, tx
) == 0);
662 mzap_create_impl(os
, obj
, normflags
, flags
, tx
);
667 zap_destroy(objset_t
*os
, uint64_t zapobj
, dmu_tx_t
*tx
)
670 * dmu_object_free will free the object number and free the
671 * data. Freeing the data will cause our pageout function to be
672 * called, which will destroy our data (zap_leaf_t's and zap_t).
675 return (dmu_object_free(os
, zapobj
, tx
));
680 zap_evict(dmu_buf_t
*db
, void *vzap
)
684 rw_destroy(&zap
->zap_rwlock
);
686 if (zap
->zap_ismicro
)
689 mutex_destroy(&zap
->zap_f
.zap_num_entries_mtx
);
691 kmem_free(zap
, sizeof (zap_t
));
695 zap_count(objset_t
*os
, uint64_t zapobj
, uint64_t *count
)
700 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
703 if (!zap
->zap_ismicro
) {
704 err
= fzap_count(zap
, count
);
706 *count
= zap
->zap_m
.zap_num_entries
;
713 * zn may be NULL; if not specified, it will be computed if needed.
714 * See also the comment above zap_entry_normalization_conflict().
717 mzap_normalization_conflict(zap_t
*zap
, zap_name_t
*zn
, mzap_ent_t
*mze
)
720 int direction
= AVL_BEFORE
;
721 boolean_t allocdzn
= B_FALSE
;
723 if (zap
->zap_normflags
== 0)
727 for (other
= avl_walk(&zap
->zap_m
.zap_avl
, mze
, direction
);
728 other
&& other
->mze_hash
== mze
->mze_hash
;
729 other
= avl_walk(&zap
->zap_m
.zap_avl
, other
, direction
)) {
732 zn
= zap_name_alloc(zap
, MZE_PHYS(zap
, mze
)->mze_name
,
736 if (zap_match(zn
, MZE_PHYS(zap
, other
)->mze_name
)) {
743 if (direction
== AVL_BEFORE
) {
744 direction
= AVL_AFTER
;
754 * Routines for manipulating attributes.
758 zap_lookup(objset_t
*os
, uint64_t zapobj
, const char *name
,
759 uint64_t integer_size
, uint64_t num_integers
, void *buf
)
761 return (zap_lookup_norm(os
, zapobj
, name
, integer_size
,
762 num_integers
, buf
, MT_EXACT
, NULL
, 0, NULL
));
766 zap_lookup_norm(objset_t
*os
, uint64_t zapobj
, const char *name
,
767 uint64_t integer_size
, uint64_t num_integers
, void *buf
,
768 matchtype_t mt
, char *realname
, int rn_len
,
776 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
779 zn
= zap_name_alloc(zap
, name
, mt
);
785 if (!zap
->zap_ismicro
) {
786 err
= fzap_lookup(zn
, integer_size
, num_integers
, buf
,
787 realname
, rn_len
, ncp
);
793 if (num_integers
< 1) {
795 } else if (integer_size
!= 8) {
799 MZE_PHYS(zap
, mze
)->mze_value
;
800 (void) strlcpy(realname
,
801 MZE_PHYS(zap
, mze
)->mze_name
, rn_len
);
803 *ncp
= mzap_normalization_conflict(zap
,
815 zap_prefetch_uint64(objset_t
*os
, uint64_t zapobj
, const uint64_t *key
,
822 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
825 zn
= zap_name_alloc_uint64(zap
, key
, key_numints
);
838 zap_lookup_uint64(objset_t
*os
, uint64_t zapobj
, const uint64_t *key
,
839 int key_numints
, uint64_t integer_size
, uint64_t num_integers
, void *buf
)
845 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
848 zn
= zap_name_alloc_uint64(zap
, key
, key_numints
);
854 err
= fzap_lookup(zn
, integer_size
, num_integers
, buf
,
862 zap_contains(objset_t
*os
, uint64_t zapobj
, const char *name
)
864 int err
= (zap_lookup_norm(os
, zapobj
, name
, 0,
865 0, NULL
, MT_EXACT
, NULL
, 0, NULL
));
866 if (err
== EOVERFLOW
|| err
== EINVAL
)
867 err
= 0; /* found, but skipped reading the value */
872 zap_length(objset_t
*os
, uint64_t zapobj
, const char *name
,
873 uint64_t *integer_size
, uint64_t *num_integers
)
880 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
883 zn
= zap_name_alloc(zap
, name
, MT_EXACT
);
888 if (!zap
->zap_ismicro
) {
889 err
= fzap_length(zn
, integer_size
, num_integers
);
907 zap_length_uint64(objset_t
*os
, uint64_t zapobj
, const uint64_t *key
,
908 int key_numints
, uint64_t *integer_size
, uint64_t *num_integers
)
914 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
917 zn
= zap_name_alloc_uint64(zap
, key
, key_numints
);
922 err
= fzap_length(zn
, integer_size
, num_integers
);
929 mzap_addent(zap_name_t
*zn
, uint64_t value
)
932 zap_t
*zap
= zn
->zn_zap
;
933 int start
= zap
->zap_m
.zap_alloc_next
;
936 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
939 for (i
= 0; i
< zap
->zap_m
.zap_num_chunks
; i
++) {
940 ASSERTV(mzap_ent_phys_t
*mze
=&zap
->zap_m
.zap_phys
->mz_chunk
[i
]);
941 ASSERT(strcmp(zn
->zn_key_orig
, mze
->mze_name
) != 0);
945 cd
= mze_find_unused_cd(zap
, zn
->zn_hash
);
946 /* given the limited size of the microzap, this can't happen */
947 ASSERT(cd
< zap_maxcd(zap
));
950 for (i
= start
; i
< zap
->zap_m
.zap_num_chunks
; i
++) {
951 mzap_ent_phys_t
*mze
= &zap
->zap_m
.zap_phys
->mz_chunk
[i
];
952 if (mze
->mze_name
[0] == 0) {
953 mze
->mze_value
= value
;
955 (void) strcpy(mze
->mze_name
, zn
->zn_key_orig
);
956 zap
->zap_m
.zap_num_entries
++;
957 zap
->zap_m
.zap_alloc_next
= i
+1;
958 if (zap
->zap_m
.zap_alloc_next
==
959 zap
->zap_m
.zap_num_chunks
)
960 zap
->zap_m
.zap_alloc_next
= 0;
961 mze_insert(zap
, i
, zn
->zn_hash
);
969 ASSERT(!"out of entries!");
973 zap_add(objset_t
*os
, uint64_t zapobj
, const char *key
,
974 int integer_size
, uint64_t num_integers
,
975 const void *val
, dmu_tx_t
*tx
)
980 const uint64_t *intval
= val
;
983 err
= zap_lockdir(os
, zapobj
, tx
, RW_WRITER
, TRUE
, TRUE
, &zap
);
986 zn
= zap_name_alloc(zap
, key
, MT_EXACT
);
991 if (!zap
->zap_ismicro
) {
992 err
= fzap_add(zn
, integer_size
, num_integers
, val
, tx
);
993 zap
= zn
->zn_zap
; /* fzap_add() may change zap */
994 } else if (integer_size
!= 8 || num_integers
!= 1 ||
995 strlen(key
) >= MZAP_NAME_LEN
) {
996 err
= mzap_upgrade(&zn
->zn_zap
, tx
, 0);
998 err
= fzap_add(zn
, integer_size
, num_integers
, val
, tx
);
999 zap
= zn
->zn_zap
; /* fzap_add() may change zap */
1005 mzap_addent(zn
, *intval
);
1008 ASSERT(zap
== zn
->zn_zap
);
1010 if (zap
!= NULL
) /* may be NULL if fzap_add() failed */
1016 zap_add_uint64(objset_t
*os
, uint64_t zapobj
, const uint64_t *key
,
1017 int key_numints
, int integer_size
, uint64_t num_integers
,
1018 const void *val
, dmu_tx_t
*tx
)
1024 err
= zap_lockdir(os
, zapobj
, tx
, RW_WRITER
, TRUE
, TRUE
, &zap
);
1027 zn
= zap_name_alloc_uint64(zap
, key
, key_numints
);
1032 err
= fzap_add(zn
, integer_size
, num_integers
, val
, tx
);
1033 zap
= zn
->zn_zap
; /* fzap_add() may change zap */
1035 if (zap
!= NULL
) /* may be NULL if fzap_add() failed */
1041 zap_update(objset_t
*os
, uint64_t zapobj
, const char *name
,
1042 int integer_size
, uint64_t num_integers
, const void *val
, dmu_tx_t
*tx
)
1046 const uint64_t *intval
= val
;
1054 * If there is an old value, it shouldn't change across the
1055 * lockdir (eg, due to bprewrite's xlation).
1057 if (integer_size
== 8 && num_integers
== 1)
1058 (void) zap_lookup(os
, zapobj
, name
, 8, 1, &oldval
);
1061 err
= zap_lockdir(os
, zapobj
, tx
, RW_WRITER
, TRUE
, TRUE
, &zap
);
1064 zn
= zap_name_alloc(zap
, name
, MT_EXACT
);
1069 if (!zap
->zap_ismicro
) {
1070 err
= fzap_update(zn
, integer_size
, num_integers
, val
, tx
);
1071 zap
= zn
->zn_zap
; /* fzap_update() may change zap */
1072 } else if (integer_size
!= 8 || num_integers
!= 1 ||
1073 strlen(name
) >= MZAP_NAME_LEN
) {
1074 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1075 zapobj
, integer_size
, num_integers
, name
);
1076 err
= mzap_upgrade(&zn
->zn_zap
, tx
, 0);
1078 err
= fzap_update(zn
, integer_size
, num_integers
,
1080 zap
= zn
->zn_zap
; /* fzap_update() may change zap */
1084 ASSERT3U(MZE_PHYS(zap
, mze
)->mze_value
, ==, oldval
);
1085 MZE_PHYS(zap
, mze
)->mze_value
= *intval
;
1087 mzap_addent(zn
, *intval
);
1090 ASSERT(zap
== zn
->zn_zap
);
1092 if (zap
!= NULL
) /* may be NULL if fzap_upgrade() failed */
1098 zap_update_uint64(objset_t
*os
, uint64_t zapobj
, const uint64_t *key
,
1100 int integer_size
, uint64_t num_integers
, const void *val
, dmu_tx_t
*tx
)
1106 err
= zap_lockdir(os
, zapobj
, tx
, RW_WRITER
, TRUE
, TRUE
, &zap
);
1109 zn
= zap_name_alloc_uint64(zap
, key
, key_numints
);
1114 err
= fzap_update(zn
, integer_size
, num_integers
, val
, tx
);
1115 zap
= zn
->zn_zap
; /* fzap_update() may change zap */
1117 if (zap
!= NULL
) /* may be NULL if fzap_upgrade() failed */
1123 zap_remove(objset_t
*os
, uint64_t zapobj
, const char *name
, dmu_tx_t
*tx
)
1125 return (zap_remove_norm(os
, zapobj
, name
, MT_EXACT
, tx
));
1129 zap_remove_norm(objset_t
*os
, uint64_t zapobj
, const char *name
,
1130 matchtype_t mt
, dmu_tx_t
*tx
)
1137 err
= zap_lockdir(os
, zapobj
, tx
, RW_WRITER
, TRUE
, FALSE
, &zap
);
1140 zn
= zap_name_alloc(zap
, name
, mt
);
1145 if (!zap
->zap_ismicro
) {
1146 err
= fzap_remove(zn
, tx
);
1152 zap
->zap_m
.zap_num_entries
--;
1153 bzero(&zap
->zap_m
.zap_phys
->mz_chunk
[mze
->mze_chunkid
],
1154 sizeof (mzap_ent_phys_t
));
1155 mze_remove(zap
, mze
);
1164 zap_remove_uint64(objset_t
*os
, uint64_t zapobj
, const uint64_t *key
,
1165 int key_numints
, dmu_tx_t
*tx
)
1171 err
= zap_lockdir(os
, zapobj
, tx
, RW_WRITER
, TRUE
, FALSE
, &zap
);
1174 zn
= zap_name_alloc_uint64(zap
, key
, key_numints
);
1179 err
= fzap_remove(zn
, tx
);
1186 * Routines for iterating over the attributes.
1190 zap_cursor_init_serialized(zap_cursor_t
*zc
, objset_t
*os
, uint64_t zapobj
,
1191 uint64_t serialized
)
1196 zc
->zc_zapobj
= zapobj
;
1197 zc
->zc_serialized
= serialized
;
1203 zap_cursor_init(zap_cursor_t
*zc
, objset_t
*os
, uint64_t zapobj
)
1205 zap_cursor_init_serialized(zc
, os
, zapobj
, 0);
1209 zap_cursor_fini(zap_cursor_t
*zc
)
1212 rw_enter(&zc
->zc_zap
->zap_rwlock
, RW_READER
);
1213 zap_unlockdir(zc
->zc_zap
);
1217 rw_enter(&zc
->zc_leaf
->l_rwlock
, RW_READER
);
1218 zap_put_leaf(zc
->zc_leaf
);
1221 zc
->zc_objset
= NULL
;
1225 zap_cursor_serialize(zap_cursor_t
*zc
)
1227 if (zc
->zc_hash
== -1ULL)
1229 if (zc
->zc_zap
== NULL
)
1230 return (zc
->zc_serialized
);
1231 ASSERT((zc
->zc_hash
& zap_maxcd(zc
->zc_zap
)) == 0);
1232 ASSERT(zc
->zc_cd
< zap_maxcd(zc
->zc_zap
));
1235 * We want to keep the high 32 bits of the cursor zero if we can, so
1236 * that 32-bit programs can access this. So usually use a small
1237 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1240 * [ collision differentiator | zap_hashbits()-bit hash value ]
1242 return ((zc
->zc_hash
>> (64 - zap_hashbits(zc
->zc_zap
))) |
1243 ((uint64_t)zc
->zc_cd
<< zap_hashbits(zc
->zc_zap
)));
1247 zap_cursor_retrieve(zap_cursor_t
*zc
, zap_attribute_t
*za
)
1251 mzap_ent_t mze_tofind
;
1254 if (zc
->zc_hash
== -1ULL)
1257 if (zc
->zc_zap
== NULL
) {
1259 err
= zap_lockdir(zc
->zc_objset
, zc
->zc_zapobj
, NULL
,
1260 RW_READER
, TRUE
, FALSE
, &zc
->zc_zap
);
1265 * To support zap_cursor_init_serialized, advance, retrieve,
1266 * we must add to the existing zc_cd, which may already
1267 * be 1 due to the zap_cursor_advance.
1269 ASSERT(zc
->zc_hash
== 0);
1270 hb
= zap_hashbits(zc
->zc_zap
);
1271 zc
->zc_hash
= zc
->zc_serialized
<< (64 - hb
);
1272 zc
->zc_cd
+= zc
->zc_serialized
>> hb
;
1273 if (zc
->zc_cd
>= zap_maxcd(zc
->zc_zap
)) /* corrupt serialized */
1276 rw_enter(&zc
->zc_zap
->zap_rwlock
, RW_READER
);
1278 if (!zc
->zc_zap
->zap_ismicro
) {
1279 err
= fzap_cursor_retrieve(zc
->zc_zap
, zc
, za
);
1283 mze_tofind
.mze_hash
= zc
->zc_hash
;
1284 mze_tofind
.mze_cd
= zc
->zc_cd
;
1286 mze
= avl_find(&zc
->zc_zap
->zap_m
.zap_avl
, &mze_tofind
, &idx
);
1288 mze
= avl_nearest(&zc
->zc_zap
->zap_m
.zap_avl
,
1292 mzap_ent_phys_t
*mzep
= MZE_PHYS(zc
->zc_zap
, mze
);
1293 ASSERT3U(mze
->mze_cd
, ==, mzep
->mze_cd
);
1294 za
->za_normalization_conflict
=
1295 mzap_normalization_conflict(zc
->zc_zap
, NULL
, mze
);
1296 za
->za_integer_length
= 8;
1297 za
->za_num_integers
= 1;
1298 za
->za_first_integer
= mzep
->mze_value
;
1299 (void) strcpy(za
->za_name
, mzep
->mze_name
);
1300 zc
->zc_hash
= mze
->mze_hash
;
1301 zc
->zc_cd
= mze
->mze_cd
;
1304 zc
->zc_hash
= -1ULL;
1307 rw_exit(&zc
->zc_zap
->zap_rwlock
);
1312 zap_cursor_advance(zap_cursor_t
*zc
)
1314 if (zc
->zc_hash
== -1ULL)
1320 zap_cursor_move_to_key(zap_cursor_t
*zc
, const char *name
, matchtype_t mt
)
1326 if (zc
->zc_zap
== NULL
) {
1327 err
= zap_lockdir(zc
->zc_objset
, zc
->zc_zapobj
, NULL
,
1328 RW_READER
, TRUE
, FALSE
, &zc
->zc_zap
);
1332 rw_enter(&zc
->zc_zap
->zap_rwlock
, RW_READER
);
1335 zn
= zap_name_alloc(zc
->zc_zap
, name
, mt
);
1337 rw_exit(&zc
->zc_zap
->zap_rwlock
);
1341 if (!zc
->zc_zap
->zap_ismicro
) {
1342 err
= fzap_cursor_move_to_key(zc
, zn
);
1349 zc
->zc_hash
= mze
->mze_hash
;
1350 zc
->zc_cd
= mze
->mze_cd
;
1355 rw_exit(&zc
->zc_zap
->zap_rwlock
);
1360 zap_get_stats(objset_t
*os
, uint64_t zapobj
, zap_stats_t
*zs
)
1365 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
1369 bzero(zs
, sizeof (zap_stats_t
));
1371 if (zap
->zap_ismicro
) {
1372 zs
->zs_blocksize
= zap
->zap_dbuf
->db_size
;
1373 zs
->zs_num_entries
= zap
->zap_m
.zap_num_entries
;
1374 zs
->zs_num_blocks
= 1;
1376 fzap_get_stats(zap
, zs
);
1383 zap_count_write(objset_t
*os
, uint64_t zapobj
, const char *name
, int add
,
1384 uint64_t *towrite
, uint64_t *tooverwrite
)
1391 * Since, we don't have a name, we cannot figure out which blocks will
1392 * be affected in this operation. So, account for the worst case :
1393 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1394 * - 4 new blocks written if adding:
1395 * - 2 blocks for possibly split leaves,
1396 * - 2 grown ptrtbl blocks
1398 * This also accomodates the case where an add operation to a fairly
1399 * large microzap results in a promotion to fatzap.
1402 *towrite
+= (3 + (add
? 4 : 0)) * SPA_MAXBLOCKSIZE
;
1407 * We lock the zap with adding == FALSE. Because, if we pass
1408 * the actual value of add, it could trigger a mzap_upgrade().
1409 * At present we are just evaluating the possibility of this operation
1410 * and hence we donot want to trigger an upgrade.
1412 err
= zap_lockdir(os
, zapobj
, NULL
, RW_READER
, TRUE
, FALSE
, &zap
);
1416 if (!zap
->zap_ismicro
) {
1417 zap_name_t
*zn
= zap_name_alloc(zap
, name
, MT_EXACT
);
1419 err
= fzap_count_write(zn
, add
, towrite
,
1424 * We treat this case as similar to (name == NULL)
1426 *towrite
+= (3 + (add
? 4 : 0)) * SPA_MAXBLOCKSIZE
;
1430 * We are here if (name != NULL) and this is a micro-zap.
1431 * We account for the header block depending on whether it
1434 * Incase of an add-operation it is hard to find out
1435 * if this add will promote this microzap to fatzap.
1436 * Hence, we consider the worst case and account for the
1437 * blocks assuming this microzap would be promoted to a
1440 * 1 block overwritten : header block
1441 * 4 new blocks written : 2 new split leaf, 2 grown
1444 if (dmu_buf_freeable(zap
->zap_dbuf
))
1445 *tooverwrite
+= SPA_MAXBLOCKSIZE
;
1447 *towrite
+= SPA_MAXBLOCKSIZE
;
1450 *towrite
+= 4 * SPA_MAXBLOCKSIZE
;