4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * This file contains the top half of the zfs directory structure
29 * implementation. The bottom half is in zap_leaf.c.
31 * The zdir is an extendable hash data structure. There is a table of
32 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are
33 * each a constant size and hold a variable number of directory entries.
34 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c.
36 * The pointer table holds a power of 2 number of pointers.
37 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to
38 * by the pointer at index i in the table holds entries whose hash value
39 * has a zd_prefix_len - bit prefix
44 #include <sys/zfs_context.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/fs/zfs.h>
48 #include <sys/refcount.h>
49 #include <sys/zap_impl.h>
50 #include <sys/zap_leaf.h>
52 int fzap_default_block_shift
= 14; /* 16k blocksize */
54 extern inline zap_phys_t
*zap_f_phys(zap_t
*zap
);
56 static uint64_t zap_allocate_blocks(zap_t
*zap
, int nblocks
);
59 fzap_byteswap(void *vbuf
, size_t size
)
63 block_type
= *(uint64_t *)vbuf
;
65 if (block_type
== ZBT_LEAF
|| block_type
== BSWAP_64(ZBT_LEAF
))
66 zap_leaf_byteswap(vbuf
, size
);
68 /* it's a ptrtbl block */
69 byteswap_uint64_array(vbuf
, size
);
74 fzap_upgrade(zap_t
*zap
, dmu_tx_t
*tx
, zap_flags_t flags
)
81 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
82 zap
->zap_ismicro
= FALSE
;
84 zap
->zap_dbu
.dbu_evict_func_sync
= zap_evict_sync
;
85 zap
->zap_dbu
.dbu_evict_func_async
= NULL
;
87 mutex_init(&zap
->zap_f
.zap_num_entries_mtx
, 0, MUTEX_DEFAULT
, 0);
88 zap
->zap_f
.zap_block_shift
= highbit64(zap
->zap_dbuf
->db_size
) - 1;
92 * explicitly zero it since it might be coming from an
93 * initialized microzap
95 bzero(zap
->zap_dbuf
->db_data
, zap
->zap_dbuf
->db_size
);
96 zp
->zap_block_type
= ZBT_HEADER
;
97 zp
->zap_magic
= ZAP_MAGIC
;
99 zp
->zap_ptrtbl
.zt_shift
= ZAP_EMBEDDED_PTRTBL_SHIFT(zap
);
101 zp
->zap_freeblk
= 2; /* block 1 will be the first leaf */
102 zp
->zap_num_leafs
= 1;
103 zp
->zap_num_entries
= 0;
104 zp
->zap_salt
= zap
->zap_salt
;
105 zp
->zap_normflags
= zap
->zap_normflags
;
106 zp
->zap_flags
= flags
;
108 /* block 1 will be the first leaf */
109 for (i
= 0; i
< (1<<zp
->zap_ptrtbl
.zt_shift
); i
++)
110 ZAP_EMBEDDED_PTRTBL_ENT(zap
, i
) = 1;
113 * set up block 1 - the first leaf
115 VERIFY(0 == dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
116 1<<FZAP_BLOCK_SHIFT(zap
), FTAG
, &db
, DMU_READ_NO_PREFETCH
));
117 dmu_buf_will_dirty(db
, tx
);
119 l
= kmem_zalloc(sizeof (zap_leaf_t
), KM_SLEEP
);
122 zap_leaf_init(l
, zp
->zap_normflags
!= 0);
124 kmem_free(l
, sizeof (zap_leaf_t
));
125 dmu_buf_rele(db
, FTAG
);
129 zap_tryupgradedir(zap_t
*zap
, dmu_tx_t
*tx
)
131 if (RW_WRITE_HELD(&zap
->zap_rwlock
))
133 if (rw_tryupgrade(&zap
->zap_rwlock
)) {
134 dmu_buf_will_dirty(zap
->zap_dbuf
, tx
);
141 * Generic routines for dealing with the pointer & cookie tables.
145 zap_table_grow(zap_t
*zap
, zap_table_phys_t
*tbl
,
146 void (*transfer_func
)(const uint64_t *src
, uint64_t *dst
, int n
),
150 dmu_buf_t
*db_old
, *db_new
;
152 int bs
= FZAP_BLOCK_SHIFT(zap
);
153 int hepb
= 1<<(bs
-4);
154 /* hepb = half the number of entries in a block */
156 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
157 ASSERT(tbl
->zt_blk
!= 0);
158 ASSERT(tbl
->zt_numblks
> 0);
160 if (tbl
->zt_nextblk
!= 0) {
161 newblk
= tbl
->zt_nextblk
;
163 newblk
= zap_allocate_blocks(zap
, tbl
->zt_numblks
* 2);
164 tbl
->zt_nextblk
= newblk
;
165 ASSERT0(tbl
->zt_blks_copied
);
166 dmu_prefetch(zap
->zap_objset
, zap
->zap_object
, 0,
167 tbl
->zt_blk
<< bs
, tbl
->zt_numblks
<< bs
,
168 ZIO_PRIORITY_SYNC_READ
);
172 * Copy the ptrtbl from the old to new location.
175 b
= tbl
->zt_blks_copied
;
176 err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
177 (tbl
->zt_blk
+ b
) << bs
, FTAG
, &db_old
, DMU_READ_NO_PREFETCH
);
181 /* first half of entries in old[b] go to new[2*b+0] */
182 VERIFY(0 == dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
183 (newblk
+ 2*b
+0) << bs
, FTAG
, &db_new
, DMU_READ_NO_PREFETCH
));
184 dmu_buf_will_dirty(db_new
, tx
);
185 transfer_func(db_old
->db_data
, db_new
->db_data
, hepb
);
186 dmu_buf_rele(db_new
, FTAG
);
188 /* second half of entries in old[b] go to new[2*b+1] */
189 VERIFY(0 == dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
190 (newblk
+ 2*b
+1) << bs
, FTAG
, &db_new
, DMU_READ_NO_PREFETCH
));
191 dmu_buf_will_dirty(db_new
, tx
);
192 transfer_func((uint64_t *)db_old
->db_data
+ hepb
,
193 db_new
->db_data
, hepb
);
194 dmu_buf_rele(db_new
, FTAG
);
196 dmu_buf_rele(db_old
, FTAG
);
198 tbl
->zt_blks_copied
++;
200 dprintf("copied block %llu of %llu\n",
201 tbl
->zt_blks_copied
, tbl
->zt_numblks
);
203 if (tbl
->zt_blks_copied
== tbl
->zt_numblks
) {
204 (void) dmu_free_range(zap
->zap_objset
, zap
->zap_object
,
205 tbl
->zt_blk
<< bs
, tbl
->zt_numblks
<< bs
, tx
);
207 tbl
->zt_blk
= newblk
;
208 tbl
->zt_numblks
*= 2;
211 tbl
->zt_blks_copied
= 0;
213 dprintf("finished; numblocks now %llu (%uk entries)\n",
214 tbl
->zt_numblks
, 1<<(tbl
->zt_shift
-10));
221 zap_table_store(zap_t
*zap
, zap_table_phys_t
*tbl
, uint64_t idx
, uint64_t val
,
226 int bs
= FZAP_BLOCK_SHIFT(zap
);
229 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
230 ASSERT(tbl
->zt_blk
!= 0);
232 dprintf("storing %llx at index %llx\n", val
, idx
);
235 off
= idx
& ((1<<(bs
-3))-1);
237 err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
238 (tbl
->zt_blk
+ blk
) << bs
, FTAG
, &db
, DMU_READ_NO_PREFETCH
);
241 dmu_buf_will_dirty(db
, tx
);
243 if (tbl
->zt_nextblk
!= 0) {
244 uint64_t idx2
= idx
* 2;
245 uint64_t blk2
= idx2
>> (bs
-3);
246 uint64_t off2
= idx2
& ((1<<(bs
-3))-1);
249 err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
250 (tbl
->zt_nextblk
+ blk2
) << bs
, FTAG
, &db2
,
251 DMU_READ_NO_PREFETCH
);
253 dmu_buf_rele(db
, FTAG
);
256 dmu_buf_will_dirty(db2
, tx
);
257 ((uint64_t *)db2
->db_data
)[off2
] = val
;
258 ((uint64_t *)db2
->db_data
)[off2
+1] = val
;
259 dmu_buf_rele(db2
, FTAG
);
262 ((uint64_t *)db
->db_data
)[off
] = val
;
263 dmu_buf_rele(db
, FTAG
);
269 zap_table_load(zap_t
*zap
, zap_table_phys_t
*tbl
, uint64_t idx
, uint64_t *valp
)
275 int bs
= FZAP_BLOCK_SHIFT(zap
);
277 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
280 off
= idx
& ((1<<(bs
-3))-1);
283 * Note: this is equivalent to dmu_buf_hold(), but we use
284 * _dnode_enter / _by_dnode because it's faster because we don't
285 * have to hold the dnode.
287 dn
= dmu_buf_dnode_enter(zap
->zap_dbuf
);
288 err
= dmu_buf_hold_by_dnode(dn
,
289 (tbl
->zt_blk
+ blk
) << bs
, FTAG
, &db
, DMU_READ_NO_PREFETCH
);
290 dmu_buf_dnode_exit(zap
->zap_dbuf
);
293 *valp
= ((uint64_t *)db
->db_data
)[off
];
294 dmu_buf_rele(db
, FTAG
);
296 if (tbl
->zt_nextblk
!= 0) {
298 * read the nextblk for the sake of i/o error checking,
299 * so that zap_table_load() will catch errors for
302 blk
= (idx
*2) >> (bs
-3);
304 dn
= dmu_buf_dnode_enter(zap
->zap_dbuf
);
305 err
= dmu_buf_hold_by_dnode(dn
,
306 (tbl
->zt_nextblk
+ blk
) << bs
, FTAG
, &db
,
307 DMU_READ_NO_PREFETCH
);
308 dmu_buf_dnode_exit(zap
->zap_dbuf
);
310 dmu_buf_rele(db
, FTAG
);
316 * Routines for growing the ptrtbl.
320 zap_ptrtbl_transfer(const uint64_t *src
, uint64_t *dst
, int n
)
323 for (i
= 0; i
< n
; i
++) {
324 uint64_t lb
= src
[i
];
331 zap_grow_ptrtbl(zap_t
*zap
, dmu_tx_t
*tx
)
334 * The pointer table should never use more hash bits than we
335 * have (otherwise we'd be using useless zero bits to index it).
336 * If we are within 2 bits of running out, stop growing, since
337 * this is already an aberrant condition.
339 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
>= zap_hashbits(zap
) - 2)
340 return (SET_ERROR(ENOSPC
));
342 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
== 0) {
344 * We are outgrowing the "embedded" ptrtbl (the one
345 * stored in the header block). Give it its own entire
346 * block, which will double the size of the ptrtbl.
352 ASSERT3U(zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
, ==,
353 ZAP_EMBEDDED_PTRTBL_SHIFT(zap
));
354 ASSERT0(zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
);
356 newblk
= zap_allocate_blocks(zap
, 1);
357 err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
358 newblk
<< FZAP_BLOCK_SHIFT(zap
), FTAG
, &db_new
,
359 DMU_READ_NO_PREFETCH
);
362 dmu_buf_will_dirty(db_new
, tx
);
363 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap
, 0),
364 db_new
->db_data
, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap
));
365 dmu_buf_rele(db_new
, FTAG
);
367 zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
= newblk
;
368 zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
= 1;
369 zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
++;
371 ASSERT3U(1ULL << zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
, ==,
372 zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
<<
373 (FZAP_BLOCK_SHIFT(zap
)-3));
377 return (zap_table_grow(zap
, &zap_f_phys(zap
)->zap_ptrtbl
,
378 zap_ptrtbl_transfer
, tx
));
383 zap_increment_num_entries(zap_t
*zap
, int delta
, dmu_tx_t
*tx
)
385 dmu_buf_will_dirty(zap
->zap_dbuf
, tx
);
386 mutex_enter(&zap
->zap_f
.zap_num_entries_mtx
);
387 ASSERT(delta
> 0 || zap_f_phys(zap
)->zap_num_entries
>= -delta
);
388 zap_f_phys(zap
)->zap_num_entries
+= delta
;
389 mutex_exit(&zap
->zap_f
.zap_num_entries_mtx
);
393 zap_allocate_blocks(zap_t
*zap
, int nblocks
)
396 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
397 newblk
= zap_f_phys(zap
)->zap_freeblk
;
398 zap_f_phys(zap
)->zap_freeblk
+= nblocks
;
403 zap_leaf_evict_sync(void *dbu
)
407 rw_destroy(&l
->l_rwlock
);
408 kmem_free(l
, sizeof (zap_leaf_t
));
412 zap_create_leaf(zap_t
*zap
, dmu_tx_t
*tx
)
415 zap_leaf_t
*l
= kmem_zalloc(sizeof (zap_leaf_t
), KM_SLEEP
);
417 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
419 rw_init(&l
->l_rwlock
, NULL
, RW_NOLOCKDEP
, NULL
);
420 rw_enter(&l
->l_rwlock
, RW_WRITER
);
421 l
->l_blkid
= zap_allocate_blocks(zap
, 1);
424 VERIFY(0 == dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
425 l
->l_blkid
<< FZAP_BLOCK_SHIFT(zap
), NULL
, &l
->l_dbuf
,
426 DMU_READ_NO_PREFETCH
));
427 dmu_buf_init_user(&l
->l_dbu
, zap_leaf_evict_sync
, NULL
, &l
->l_dbuf
);
428 winner
= dmu_buf_set_user(l
->l_dbuf
, &l
->l_dbu
);
429 ASSERT(winner
== NULL
);
430 dmu_buf_will_dirty(l
->l_dbuf
, tx
);
432 zap_leaf_init(l
, zap
->zap_normflags
!= 0);
434 zap_f_phys(zap
)->zap_num_leafs
++;
440 fzap_count(zap_t
*zap
, uint64_t *count
)
442 ASSERT(!zap
->zap_ismicro
);
443 mutex_enter(&zap
->zap_f
.zap_num_entries_mtx
); /* unnecessary */
444 *count
= zap_f_phys(zap
)->zap_num_entries
;
445 mutex_exit(&zap
->zap_f
.zap_num_entries_mtx
);
450 * Routines for obtaining zap_leaf_t's
454 zap_put_leaf(zap_leaf_t
*l
)
456 rw_exit(&l
->l_rwlock
);
457 dmu_buf_rele(l
->l_dbuf
, NULL
);
461 zap_open_leaf(uint64_t blkid
, dmu_buf_t
*db
)
463 zap_leaf_t
*l
, *winner
;
467 l
= kmem_zalloc(sizeof (zap_leaf_t
), KM_SLEEP
);
468 rw_init(&l
->l_rwlock
, NULL
, RW_DEFAULT
, NULL
);
469 rw_enter(&l
->l_rwlock
, RW_WRITER
);
471 l
->l_bs
= highbit64(db
->db_size
) - 1;
474 dmu_buf_init_user(&l
->l_dbu
, zap_leaf_evict_sync
, NULL
, &l
->l_dbuf
);
475 winner
= dmu_buf_set_user(db
, &l
->l_dbu
);
477 rw_exit(&l
->l_rwlock
);
478 if (winner
!= NULL
) {
479 /* someone else set it first */
480 zap_leaf_evict_sync(&l
->l_dbu
);
485 * lhr_pad was previously used for the next leaf in the leaf
486 * chain. There should be no chained leafs (as we have removed
489 ASSERT0(zap_leaf_phys(l
)->l_hdr
.lh_pad1
);
492 * There should be more hash entries than there can be
493 * chunks to put in the hash table
495 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l
), >, ZAP_LEAF_NUMCHUNKS(l
) / 3);
497 /* The chunks should begin at the end of the hash table */
498 ASSERT3P(&ZAP_LEAF_CHUNK(l
, 0), ==, (zap_leaf_chunk_t
*)
499 &zap_leaf_phys(l
)->l_hash
[ZAP_LEAF_HASH_NUMENTRIES(l
)]);
501 /* The chunks should end at the end of the block */
502 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l
, ZAP_LEAF_NUMCHUNKS(l
)) -
503 (uintptr_t)zap_leaf_phys(l
), ==, l
->l_dbuf
->db_size
);
509 zap_get_leaf_byblk(zap_t
*zap
, uint64_t blkid
, dmu_tx_t
*tx
, krw_t lt
,
514 int bs
= FZAP_BLOCK_SHIFT(zap
);
518 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
521 * If system crashed just after dmu_free_long_range in zfs_rmnode, we
522 * would be left with an empty xattr dir in delete queue. blkid=0
523 * would be passed in when doing zfs_purgedir. If that's the case we
524 * should just return immediately. The underlying objects should
525 * already be freed, so this should be perfectly fine.
528 return (SET_ERROR(ENOENT
));
530 dn
= dmu_buf_dnode_enter(zap
->zap_dbuf
);
531 err
= dmu_buf_hold_by_dnode(dn
,
532 blkid
<< bs
, NULL
, &db
, DMU_READ_NO_PREFETCH
);
533 dmu_buf_dnode_exit(zap
->zap_dbuf
);
537 ASSERT3U(db
->db_object
, ==, zap
->zap_object
);
538 ASSERT3U(db
->db_offset
, ==, blkid
<< bs
);
539 ASSERT3U(db
->db_size
, ==, 1 << bs
);
542 l
= dmu_buf_get_user(db
);
545 l
= zap_open_leaf(blkid
, db
);
547 rw_enter(&l
->l_rwlock
, lt
);
549 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
550 * causing ASSERT below to fail.
553 dmu_buf_will_dirty(db
, tx
);
554 ASSERT3U(l
->l_blkid
, ==, blkid
);
555 ASSERT3P(l
->l_dbuf
, ==, db
);
556 ASSERT3U(zap_leaf_phys(l
)->l_hdr
.lh_block_type
, ==, ZBT_LEAF
);
557 ASSERT3U(zap_leaf_phys(l
)->l_hdr
.lh_magic
, ==, ZAP_LEAF_MAGIC
);
564 zap_idx_to_blk(zap_t
*zap
, uint64_t idx
, uint64_t *valp
)
566 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
568 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
== 0) {
570 (1ULL << zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
));
571 *valp
= ZAP_EMBEDDED_PTRTBL_ENT(zap
, idx
);
574 return (zap_table_load(zap
, &zap_f_phys(zap
)->zap_ptrtbl
,
580 zap_set_idx_to_blk(zap_t
*zap
, uint64_t idx
, uint64_t blk
, dmu_tx_t
*tx
)
583 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
585 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
== 0) {
586 ZAP_EMBEDDED_PTRTBL_ENT(zap
, idx
) = blk
;
589 return (zap_table_store(zap
, &zap_f_phys(zap
)->zap_ptrtbl
,
595 zap_deref_leaf(zap_t
*zap
, uint64_t h
, dmu_tx_t
*tx
, krw_t lt
, zap_leaf_t
**lp
)
600 ASSERT(zap
->zap_dbuf
== NULL
||
601 zap_f_phys(zap
) == zap
->zap_dbuf
->db_data
);
603 /* Reality check for corrupt zap objects (leaf or header). */
604 if ((zap_f_phys(zap
)->zap_block_type
!= ZBT_LEAF
&&
605 zap_f_phys(zap
)->zap_block_type
!= ZBT_HEADER
) ||
606 zap_f_phys(zap
)->zap_magic
!= ZAP_MAGIC
) {
607 return (SET_ERROR(EIO
));
609 idx
= ZAP_HASH_IDX(h
, zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
610 err
= zap_idx_to_blk(zap
, idx
, &blk
);
613 err
= zap_get_leaf_byblk(zap
, blk
, tx
, lt
, lp
);
616 ZAP_HASH_IDX(h
, zap_leaf_phys(*lp
)->l_hdr
.lh_prefix_len
) ==
617 zap_leaf_phys(*lp
)->l_hdr
.lh_prefix
);
622 zap_expand_leaf(zap_name_t
*zn
, zap_leaf_t
*l
,
623 void *tag
, dmu_tx_t
*tx
, zap_leaf_t
**lp
)
625 zap_t
*zap
= zn
->zn_zap
;
626 uint64_t hash
= zn
->zn_hash
;
628 int prefix_diff
, i
, err
;
630 int old_prefix_len
= zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
;
632 ASSERT3U(old_prefix_len
, <=, zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
633 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
635 ASSERT3U(ZAP_HASH_IDX(hash
, old_prefix_len
), ==,
636 zap_leaf_phys(l
)->l_hdr
.lh_prefix
);
638 if (zap_tryupgradedir(zap
, tx
) == 0 ||
639 old_prefix_len
== zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
) {
640 /* We failed to upgrade, or need to grow the pointer table */
641 objset_t
*os
= zap
->zap_objset
;
642 uint64_t object
= zap
->zap_object
;
645 zap_unlockdir(zap
, tag
);
646 err
= zap_lockdir(os
, object
, tx
, RW_WRITER
,
647 FALSE
, FALSE
, tag
, &zn
->zn_zap
);
651 ASSERT(!zap
->zap_ismicro
);
653 while (old_prefix_len
==
654 zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
) {
655 err
= zap_grow_ptrtbl(zap
, tx
);
660 err
= zap_deref_leaf(zap
, hash
, tx
, RW_WRITER
, &l
);
664 if (zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
!= old_prefix_len
) {
665 /* it split while our locks were down */
670 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
671 ASSERT3U(old_prefix_len
, <, zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
672 ASSERT3U(ZAP_HASH_IDX(hash
, old_prefix_len
), ==,
673 zap_leaf_phys(l
)->l_hdr
.lh_prefix
);
675 prefix_diff
= zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
-
676 (old_prefix_len
+ 1);
677 sibling
= (ZAP_HASH_IDX(hash
, old_prefix_len
+ 1) | 1) << prefix_diff
;
679 /* check for i/o errors before doing zap_leaf_split */
680 for (i
= 0; i
< (1ULL<<prefix_diff
); i
++) {
682 err
= zap_idx_to_blk(zap
, sibling
+i
, &blk
);
685 ASSERT3U(blk
, ==, l
->l_blkid
);
688 nl
= zap_create_leaf(zap
, tx
);
689 zap_leaf_split(l
, nl
, zap
->zap_normflags
!= 0);
691 /* set sibling pointers */
692 for (i
= 0; i
< (1ULL << prefix_diff
); i
++) {
693 err
= zap_set_idx_to_blk(zap
, sibling
+i
, nl
->l_blkid
, tx
);
694 ASSERT0(err
); /* we checked for i/o errors above */
697 ASSERT3U(zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
, >, 0);
699 if (hash
& (1ULL << (64 - zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
))) {
700 /* we want the sibling */
712 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t
*zn
, zap_leaf_t
*l
,
713 void *tag
, dmu_tx_t
*tx
)
715 zap_t
*zap
= zn
->zn_zap
;
716 int shift
= zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
;
717 int leaffull
= (zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
== shift
&&
718 zap_leaf_phys(l
)->l_hdr
.lh_nfree
< ZAP_LEAF_LOW_WATER
);
722 if (leaffull
|| zap_f_phys(zap
)->zap_ptrtbl
.zt_nextblk
) {
726 * We are in the middle of growing the pointer table, or
727 * this leaf will soon make us grow it.
729 if (zap_tryupgradedir(zap
, tx
) == 0) {
730 objset_t
*os
= zap
->zap_objset
;
731 uint64_t zapobj
= zap
->zap_object
;
733 zap_unlockdir(zap
, tag
);
734 err
= zap_lockdir(os
, zapobj
, tx
,
735 RW_WRITER
, FALSE
, FALSE
, tag
, &zn
->zn_zap
);
741 /* could have finished growing while our locks were down */
742 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
== shift
)
743 (void) zap_grow_ptrtbl(zap
, tx
);
748 fzap_checkname(zap_name_t
*zn
)
750 if (zn
->zn_key_orig_numints
* zn
->zn_key_intlen
> ZAP_MAXNAMELEN
)
751 return (SET_ERROR(ENAMETOOLONG
));
756 fzap_checksize(uint64_t integer_size
, uint64_t num_integers
)
758 /* Only integer sizes supported by C */
759 switch (integer_size
) {
766 return (SET_ERROR(EINVAL
));
769 if (integer_size
* num_integers
> ZAP_MAXVALUELEN
)
770 return (SET_ERROR(E2BIG
));
776 fzap_check(zap_name_t
*zn
, uint64_t integer_size
, uint64_t num_integers
)
780 if ((err
= fzap_checkname(zn
)) != 0)
782 return (fzap_checksize(integer_size
, num_integers
));
786 * Routines for manipulating attributes.
789 fzap_lookup(zap_name_t
*zn
,
790 uint64_t integer_size
, uint64_t num_integers
, void *buf
,
791 char *realname
, int rn_len
, boolean_t
*ncp
)
795 zap_entry_handle_t zeh
;
797 if ((err
= fzap_checkname(zn
)) != 0)
800 err
= zap_deref_leaf(zn
->zn_zap
, zn
->zn_hash
, NULL
, RW_READER
, &l
);
803 err
= zap_leaf_lookup(l
, zn
, &zeh
);
805 if ((err
= fzap_checksize(integer_size
, num_integers
)) != 0) {
810 err
= zap_entry_read(&zeh
, integer_size
, num_integers
, buf
);
811 (void) zap_entry_read_name(zn
->zn_zap
, &zeh
, rn_len
, realname
);
813 *ncp
= zap_entry_normalization_conflict(&zeh
,
814 zn
, NULL
, zn
->zn_zap
);
823 fzap_add_cd(zap_name_t
*zn
,
824 uint64_t integer_size
, uint64_t num_integers
,
825 const void *val
, uint32_t cd
, void *tag
, dmu_tx_t
*tx
)
829 zap_entry_handle_t zeh
;
830 zap_t
*zap
= zn
->zn_zap
;
832 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
833 ASSERT(!zap
->zap_ismicro
);
834 ASSERT(fzap_check(zn
, integer_size
, num_integers
) == 0);
836 err
= zap_deref_leaf(zap
, zn
->zn_hash
, tx
, RW_WRITER
, &l
);
840 err
= zap_leaf_lookup(l
, zn
, &zeh
);
842 err
= SET_ERROR(EEXIST
);
848 err
= zap_entry_create(l
, zn
, cd
,
849 integer_size
, num_integers
, val
, &zeh
);
852 zap_increment_num_entries(zap
, 1, tx
);
853 } else if (err
== EAGAIN
) {
854 err
= zap_expand_leaf(zn
, l
, tag
, tx
, &l
);
855 zap
= zn
->zn_zap
; /* zap_expand_leaf() may change zap */
862 zap_put_leaf_maybe_grow_ptrtbl(zn
, l
, tag
, tx
);
867 fzap_add(zap_name_t
*zn
,
868 uint64_t integer_size
, uint64_t num_integers
,
869 const void *val
, void *tag
, dmu_tx_t
*tx
)
871 int err
= fzap_check(zn
, integer_size
, num_integers
);
875 return (fzap_add_cd(zn
, integer_size
, num_integers
,
876 val
, ZAP_NEED_CD
, tag
, tx
));
880 fzap_update(zap_name_t
*zn
,
881 int integer_size
, uint64_t num_integers
, const void *val
,
882 void *tag
, dmu_tx_t
*tx
)
886 zap_entry_handle_t zeh
;
887 zap_t
*zap
= zn
->zn_zap
;
889 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
890 err
= fzap_check(zn
, integer_size
, num_integers
);
894 err
= zap_deref_leaf(zap
, zn
->zn_hash
, tx
, RW_WRITER
, &l
);
898 err
= zap_leaf_lookup(l
, zn
, &zeh
);
899 create
= (err
== ENOENT
);
900 ASSERT(err
== 0 || err
== ENOENT
);
903 err
= zap_entry_create(l
, zn
, ZAP_NEED_CD
,
904 integer_size
, num_integers
, val
, &zeh
);
906 zap_increment_num_entries(zap
, 1, tx
);
908 err
= zap_entry_update(&zeh
, integer_size
, num_integers
, val
);
912 err
= zap_expand_leaf(zn
, l
, tag
, tx
, &l
);
913 zap
= zn
->zn_zap
; /* zap_expand_leaf() may change zap */
919 zap_put_leaf_maybe_grow_ptrtbl(zn
, l
, tag
, tx
);
924 fzap_length(zap_name_t
*zn
,
925 uint64_t *integer_size
, uint64_t *num_integers
)
929 zap_entry_handle_t zeh
;
931 err
= zap_deref_leaf(zn
->zn_zap
, zn
->zn_hash
, NULL
, RW_READER
, &l
);
934 err
= zap_leaf_lookup(l
, zn
, &zeh
);
939 *integer_size
= zeh
.zeh_integer_size
;
941 *num_integers
= zeh
.zeh_num_integers
;
948 fzap_remove(zap_name_t
*zn
, dmu_tx_t
*tx
)
952 zap_entry_handle_t zeh
;
954 err
= zap_deref_leaf(zn
->zn_zap
, zn
->zn_hash
, tx
, RW_WRITER
, &l
);
957 err
= zap_leaf_lookup(l
, zn
, &zeh
);
959 zap_entry_remove(&zeh
);
960 zap_increment_num_entries(zn
->zn_zap
, -1, tx
);
967 fzap_prefetch(zap_name_t
*zn
)
970 zap_t
*zap
= zn
->zn_zap
;
973 idx
= ZAP_HASH_IDX(zn
->zn_hash
,
974 zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
975 if (zap_idx_to_blk(zap
, idx
, &blk
) != 0)
977 bs
= FZAP_BLOCK_SHIFT(zap
);
978 dmu_prefetch(zap
->zap_objset
, zap
->zap_object
, 0, blk
<< bs
, 1 << bs
,
979 ZIO_PRIORITY_SYNC_READ
);
983 * Helper functions for consumers.
987 zap_create_link(objset_t
*os
, dmu_object_type_t ot
, uint64_t parent_obj
,
988 const char *name
, dmu_tx_t
*tx
)
990 return (zap_create_link_dnsize(os
, ot
, parent_obj
, name
, 0, tx
));
994 zap_create_link_dnsize(objset_t
*os
, dmu_object_type_t ot
, uint64_t parent_obj
,
995 const char *name
, int dnodesize
, dmu_tx_t
*tx
)
999 VERIFY((new_obj
= zap_create_dnsize(os
, ot
, DMU_OT_NONE
, 0,
1000 dnodesize
, tx
)) > 0);
1001 VERIFY0(zap_add(os
, parent_obj
, name
, sizeof (uint64_t), 1, &new_obj
,
1008 zap_value_search(objset_t
*os
, uint64_t zapobj
, uint64_t value
, uint64_t mask
,
1012 zap_attribute_t
*za
;
1018 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1019 for (zap_cursor_init(&zc
, os
, zapobj
);
1020 (err
= zap_cursor_retrieve(&zc
, za
)) == 0;
1021 zap_cursor_advance(&zc
)) {
1022 if ((za
->za_first_integer
& mask
) == (value
& mask
)) {
1023 (void) strcpy(name
, za
->za_name
);
1027 zap_cursor_fini(&zc
);
1028 kmem_free(za
, sizeof (zap_attribute_t
));
1033 zap_join(objset_t
*os
, uint64_t fromobj
, uint64_t intoobj
, dmu_tx_t
*tx
)
1040 for (zap_cursor_init(&zc
, os
, fromobj
);
1041 zap_cursor_retrieve(&zc
, &za
) == 0;
1042 (void) zap_cursor_advance(&zc
)) {
1043 if (za
.za_integer_length
!= 8 || za
.za_num_integers
!= 1) {
1044 err
= SET_ERROR(EINVAL
);
1047 err
= zap_add(os
, intoobj
, za
.za_name
,
1048 8, 1, &za
.za_first_integer
, tx
);
1052 zap_cursor_fini(&zc
);
1057 zap_join_key(objset_t
*os
, uint64_t fromobj
, uint64_t intoobj
,
1058 uint64_t value
, dmu_tx_t
*tx
)
1065 for (zap_cursor_init(&zc
, os
, fromobj
);
1066 zap_cursor_retrieve(&zc
, &za
) == 0;
1067 (void) zap_cursor_advance(&zc
)) {
1068 if (za
.za_integer_length
!= 8 || za
.za_num_integers
!= 1) {
1069 err
= SET_ERROR(EINVAL
);
1072 err
= zap_add(os
, intoobj
, za
.za_name
,
1077 zap_cursor_fini(&zc
);
1082 zap_join_increment(objset_t
*os
, uint64_t fromobj
, uint64_t intoobj
,
1090 for (zap_cursor_init(&zc
, os
, fromobj
);
1091 zap_cursor_retrieve(&zc
, &za
) == 0;
1092 (void) zap_cursor_advance(&zc
)) {
1095 if (za
.za_integer_length
!= 8 || za
.za_num_integers
!= 1) {
1096 err
= SET_ERROR(EINVAL
);
1100 err
= zap_lookup(os
, intoobj
, za
.za_name
, 8, 1, &delta
);
1101 if (err
!= 0 && err
!= ENOENT
)
1103 delta
+= za
.za_first_integer
;
1104 err
= zap_update(os
, intoobj
, za
.za_name
, 8, 1, &delta
, tx
);
1108 zap_cursor_fini(&zc
);
1113 zap_add_int(objset_t
*os
, uint64_t obj
, uint64_t value
, dmu_tx_t
*tx
)
1117 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)value
);
1118 return (zap_add(os
, obj
, name
, 8, 1, &value
, tx
));
1122 zap_remove_int(objset_t
*os
, uint64_t obj
, uint64_t value
, dmu_tx_t
*tx
)
1126 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)value
);
1127 return (zap_remove(os
, obj
, name
, tx
));
1131 zap_lookup_int(objset_t
*os
, uint64_t obj
, uint64_t value
)
1135 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)value
);
1136 return (zap_lookup(os
, obj
, name
, 8, 1, &value
));
1140 zap_add_int_key(objset_t
*os
, uint64_t obj
,
1141 uint64_t key
, uint64_t value
, dmu_tx_t
*tx
)
1145 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1146 return (zap_add(os
, obj
, name
, 8, 1, &value
, tx
));
1150 zap_update_int_key(objset_t
*os
, uint64_t obj
,
1151 uint64_t key
, uint64_t value
, dmu_tx_t
*tx
)
1155 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1156 return (zap_update(os
, obj
, name
, 8, 1, &value
, tx
));
1160 zap_lookup_int_key(objset_t
*os
, uint64_t obj
, uint64_t key
, uint64_t *valuep
)
1164 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1165 return (zap_lookup(os
, obj
, name
, 8, 1, valuep
));
1169 zap_increment(objset_t
*os
, uint64_t obj
, const char *name
, int64_t delta
,
1178 err
= zap_lookup(os
, obj
, name
, 8, 1, &value
);
1179 if (err
!= 0 && err
!= ENOENT
)
1183 err
= zap_remove(os
, obj
, name
, tx
);
1185 err
= zap_update(os
, obj
, name
, 8, 1, &value
, tx
);
1190 zap_increment_int(objset_t
*os
, uint64_t obj
, uint64_t key
, int64_t delta
,
1195 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1196 return (zap_increment(os
, obj
, name
, delta
, tx
));
1200 * Routines for iterating over the attributes.
1204 fzap_cursor_retrieve(zap_t
*zap
, zap_cursor_t
*zc
, zap_attribute_t
*za
)
1207 zap_entry_handle_t zeh
;
1210 /* retrieve the next entry at or after zc_hash/zc_cd */
1211 /* if no entry, return ENOENT */
1214 (ZAP_HASH_IDX(zc
->zc_hash
,
1215 zap_leaf_phys(zc
->zc_leaf
)->l_hdr
.lh_prefix_len
) !=
1216 zap_leaf_phys(zc
->zc_leaf
)->l_hdr
.lh_prefix
)) {
1217 rw_enter(&zc
->zc_leaf
->l_rwlock
, RW_READER
);
1218 zap_put_leaf(zc
->zc_leaf
);
1223 if (zc
->zc_leaf
== NULL
) {
1224 err
= zap_deref_leaf(zap
, zc
->zc_hash
, NULL
, RW_READER
,
1229 rw_enter(&zc
->zc_leaf
->l_rwlock
, RW_READER
);
1233 err
= zap_leaf_lookup_closest(l
, zc
->zc_hash
, zc
->zc_cd
, &zeh
);
1235 if (err
== ENOENT
) {
1236 if (zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
== 0) {
1237 zc
->zc_hash
= -1ULL;
1240 uint64_t nocare
= (1ULL <<
1241 (64 - zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
)) - 1;
1243 zc
->zc_hash
= (zc
->zc_hash
& ~nocare
) + nocare
+ 1;
1246 if (zc
->zc_hash
== 0) {
1247 zc
->zc_hash
= -1ULL;
1249 zap_put_leaf(zc
->zc_leaf
);
1257 zc
->zc_hash
= zeh
.zeh_hash
;
1258 zc
->zc_cd
= zeh
.zeh_cd
;
1259 za
->za_integer_length
= zeh
.zeh_integer_size
;
1260 za
->za_num_integers
= zeh
.zeh_num_integers
;
1261 if (zeh
.zeh_num_integers
== 0) {
1262 za
->za_first_integer
= 0;
1264 err
= zap_entry_read(&zeh
, 8, 1, &za
->za_first_integer
);
1265 ASSERT(err
== 0 || err
== EOVERFLOW
);
1267 err
= zap_entry_read_name(zap
, &zeh
,
1268 sizeof (za
->za_name
), za
->za_name
);
1271 za
->za_normalization_conflict
=
1272 zap_entry_normalization_conflict(&zeh
,
1273 NULL
, za
->za_name
, zap
);
1275 rw_exit(&zc
->zc_leaf
->l_rwlock
);
1280 zap_stats_ptrtbl(zap_t
*zap
, uint64_t *tbl
, int len
, zap_stats_t
*zs
)
1283 uint64_t lastblk
= 0;
1286 * NB: if a leaf has more pointers than an entire ptrtbl block
1287 * can hold, then it'll be accounted for more than once, since
1288 * we won't have lastblk.
1290 for (i
= 0; i
< len
; i
++) {
1293 if (tbl
[i
] == lastblk
)
1297 err
= zap_get_leaf_byblk(zap
, tbl
[i
], NULL
, RW_READER
, &l
);
1299 zap_leaf_stats(zap
, l
, zs
);
1306 fzap_get_stats(zap_t
*zap
, zap_stats_t
*zs
)
1308 int bs
= FZAP_BLOCK_SHIFT(zap
);
1309 zs
->zs_blocksize
= 1ULL << bs
;
1312 * Set zap_phys_t fields
1314 zs
->zs_num_leafs
= zap_f_phys(zap
)->zap_num_leafs
;
1315 zs
->zs_num_entries
= zap_f_phys(zap
)->zap_num_entries
;
1316 zs
->zs_num_blocks
= zap_f_phys(zap
)->zap_freeblk
;
1317 zs
->zs_block_type
= zap_f_phys(zap
)->zap_block_type
;
1318 zs
->zs_magic
= zap_f_phys(zap
)->zap_magic
;
1319 zs
->zs_salt
= zap_f_phys(zap
)->zap_salt
;
1322 * Set zap_ptrtbl fields
1324 zs
->zs_ptrtbl_len
= 1ULL << zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
;
1325 zs
->zs_ptrtbl_nextblk
= zap_f_phys(zap
)->zap_ptrtbl
.zt_nextblk
;
1326 zs
->zs_ptrtbl_blks_copied
=
1327 zap_f_phys(zap
)->zap_ptrtbl
.zt_blks_copied
;
1328 zs
->zs_ptrtbl_zt_blk
= zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
;
1329 zs
->zs_ptrtbl_zt_numblks
= zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
;
1330 zs
->zs_ptrtbl_zt_shift
= zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
;
1332 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
== 0) {
1333 /* the ptrtbl is entirely in the header block. */
1334 zap_stats_ptrtbl(zap
, &ZAP_EMBEDDED_PTRTBL_ENT(zap
, 0),
1335 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap
), zs
);
1339 dmu_prefetch(zap
->zap_objset
, zap
->zap_object
, 0,
1340 zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
<< bs
,
1341 zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
<< bs
,
1342 ZIO_PRIORITY_SYNC_READ
);
1344 for (b
= 0; b
< zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
;
1349 err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
1350 (zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
+ b
) << bs
,
1351 FTAG
, &db
, DMU_READ_NO_PREFETCH
);
1353 zap_stats_ptrtbl(zap
, db
->db_data
,
1355 dmu_buf_rele(db
, FTAG
);