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
)
61 uint64_t block_type
= *(uint64_t *)vbuf
;
63 if (block_type
== ZBT_LEAF
|| block_type
== BSWAP_64(ZBT_LEAF
))
64 zap_leaf_byteswap(vbuf
, size
);
66 /* it's a ptrtbl block */
67 byteswap_uint64_array(vbuf
, size
);
72 fzap_upgrade(zap_t
*zap
, dmu_tx_t
*tx
, zap_flags_t flags
)
74 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
75 zap
->zap_ismicro
= FALSE
;
77 zap
->zap_dbu
.dbu_evict_func_sync
= zap_evict_sync
;
78 zap
->zap_dbu
.dbu_evict_func_async
= NULL
;
80 mutex_init(&zap
->zap_f
.zap_num_entries_mtx
, 0, MUTEX_DEFAULT
, 0);
81 zap
->zap_f
.zap_block_shift
= highbit64(zap
->zap_dbuf
->db_size
) - 1;
83 zap_phys_t
*zp
= zap_f_phys(zap
);
85 * explicitly zero it since it might be coming from an
86 * initialized microzap
88 bzero(zap
->zap_dbuf
->db_data
, zap
->zap_dbuf
->db_size
);
89 zp
->zap_block_type
= ZBT_HEADER
;
90 zp
->zap_magic
= ZAP_MAGIC
;
92 zp
->zap_ptrtbl
.zt_shift
= ZAP_EMBEDDED_PTRTBL_SHIFT(zap
);
94 zp
->zap_freeblk
= 2; /* block 1 will be the first leaf */
95 zp
->zap_num_leafs
= 1;
96 zp
->zap_num_entries
= 0;
97 zp
->zap_salt
= zap
->zap_salt
;
98 zp
->zap_normflags
= zap
->zap_normflags
;
99 zp
->zap_flags
= flags
;
101 /* block 1 will be the first leaf */
102 for (int i
= 0; i
< (1<<zp
->zap_ptrtbl
.zt_shift
); i
++)
103 ZAP_EMBEDDED_PTRTBL_ENT(zap
, i
) = 1;
106 * set up block 1 - the first leaf
109 VERIFY0(dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
110 1<<FZAP_BLOCK_SHIFT(zap
), FTAG
, &db
, DMU_READ_NO_PREFETCH
));
111 dmu_buf_will_dirty(db
, tx
);
113 zap_leaf_t
*l
= kmem_zalloc(sizeof (zap_leaf_t
), KM_SLEEP
);
116 zap_leaf_init(l
, zp
->zap_normflags
!= 0);
118 kmem_free(l
, sizeof (zap_leaf_t
));
119 dmu_buf_rele(db
, FTAG
);
123 zap_tryupgradedir(zap_t
*zap
, dmu_tx_t
*tx
)
125 if (RW_WRITE_HELD(&zap
->zap_rwlock
))
127 if (rw_tryupgrade(&zap
->zap_rwlock
)) {
128 dmu_buf_will_dirty(zap
->zap_dbuf
, tx
);
135 * Generic routines for dealing with the pointer & cookie tables.
139 zap_table_grow(zap_t
*zap
, zap_table_phys_t
*tbl
,
140 void (*transfer_func
)(const uint64_t *src
, uint64_t *dst
, int n
),
144 int bs
= FZAP_BLOCK_SHIFT(zap
);
145 int hepb
= 1<<(bs
-4);
146 /* hepb = half the number of entries in a block */
148 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
149 ASSERT(tbl
->zt_blk
!= 0);
150 ASSERT(tbl
->zt_numblks
> 0);
152 if (tbl
->zt_nextblk
!= 0) {
153 newblk
= tbl
->zt_nextblk
;
155 newblk
= zap_allocate_blocks(zap
, tbl
->zt_numblks
* 2);
156 tbl
->zt_nextblk
= newblk
;
157 ASSERT0(tbl
->zt_blks_copied
);
158 dmu_prefetch(zap
->zap_objset
, zap
->zap_object
, 0,
159 tbl
->zt_blk
<< bs
, tbl
->zt_numblks
<< bs
,
160 ZIO_PRIORITY_SYNC_READ
);
164 * Copy the ptrtbl from the old to new location.
167 uint64_t b
= tbl
->zt_blks_copied
;
169 int err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
170 (tbl
->zt_blk
+ b
) << bs
, FTAG
, &db_old
, DMU_READ_NO_PREFETCH
);
174 /* first half of entries in old[b] go to new[2*b+0] */
176 VERIFY0(dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
177 (newblk
+ 2*b
+0) << bs
, FTAG
, &db_new
, DMU_READ_NO_PREFETCH
));
178 dmu_buf_will_dirty(db_new
, tx
);
179 transfer_func(db_old
->db_data
, db_new
->db_data
, hepb
);
180 dmu_buf_rele(db_new
, FTAG
);
182 /* second half of entries in old[b] go to new[2*b+1] */
183 VERIFY0(dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
184 (newblk
+ 2*b
+1) << bs
, FTAG
, &db_new
, DMU_READ_NO_PREFETCH
));
185 dmu_buf_will_dirty(db_new
, tx
);
186 transfer_func((uint64_t *)db_old
->db_data
+ hepb
,
187 db_new
->db_data
, hepb
);
188 dmu_buf_rele(db_new
, FTAG
);
190 dmu_buf_rele(db_old
, FTAG
);
192 tbl
->zt_blks_copied
++;
194 dprintf("copied block %llu of %llu\n",
195 tbl
->zt_blks_copied
, tbl
->zt_numblks
);
197 if (tbl
->zt_blks_copied
== tbl
->zt_numblks
) {
198 (void) dmu_free_range(zap
->zap_objset
, zap
->zap_object
,
199 tbl
->zt_blk
<< bs
, tbl
->zt_numblks
<< bs
, tx
);
201 tbl
->zt_blk
= newblk
;
202 tbl
->zt_numblks
*= 2;
205 tbl
->zt_blks_copied
= 0;
207 dprintf("finished; numblocks now %llu (%uk entries)\n",
208 tbl
->zt_numblks
, 1<<(tbl
->zt_shift
-10));
215 zap_table_store(zap_t
*zap
, zap_table_phys_t
*tbl
, uint64_t idx
, uint64_t val
,
218 int bs
= FZAP_BLOCK_SHIFT(zap
);
220 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
221 ASSERT(tbl
->zt_blk
!= 0);
223 dprintf("storing %llx at index %llx\n", val
, idx
);
225 uint64_t blk
= idx
>> (bs
-3);
226 uint64_t off
= idx
& ((1<<(bs
-3))-1);
229 int err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
230 (tbl
->zt_blk
+ blk
) << bs
, FTAG
, &db
, DMU_READ_NO_PREFETCH
);
233 dmu_buf_will_dirty(db
, tx
);
235 if (tbl
->zt_nextblk
!= 0) {
236 uint64_t idx2
= idx
* 2;
237 uint64_t blk2
= idx2
>> (bs
-3);
238 uint64_t off2
= idx2
& ((1<<(bs
-3))-1);
241 err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
242 (tbl
->zt_nextblk
+ blk2
) << bs
, FTAG
, &db2
,
243 DMU_READ_NO_PREFETCH
);
245 dmu_buf_rele(db
, FTAG
);
248 dmu_buf_will_dirty(db2
, tx
);
249 ((uint64_t *)db2
->db_data
)[off2
] = val
;
250 ((uint64_t *)db2
->db_data
)[off2
+1] = val
;
251 dmu_buf_rele(db2
, FTAG
);
254 ((uint64_t *)db
->db_data
)[off
] = val
;
255 dmu_buf_rele(db
, FTAG
);
261 zap_table_load(zap_t
*zap
, zap_table_phys_t
*tbl
, uint64_t idx
, uint64_t *valp
)
263 int bs
= FZAP_BLOCK_SHIFT(zap
);
265 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
267 uint64_t blk
= idx
>> (bs
-3);
268 uint64_t off
= idx
& ((1<<(bs
-3))-1);
271 * Note: this is equivalent to dmu_buf_hold(), but we use
272 * _dnode_enter / _by_dnode because it's faster because we don't
273 * have to hold the dnode.
275 dnode_t
*dn
= dmu_buf_dnode_enter(zap
->zap_dbuf
);
277 int err
= dmu_buf_hold_by_dnode(dn
,
278 (tbl
->zt_blk
+ blk
) << bs
, FTAG
, &db
, DMU_READ_NO_PREFETCH
);
279 dmu_buf_dnode_exit(zap
->zap_dbuf
);
282 *valp
= ((uint64_t *)db
->db_data
)[off
];
283 dmu_buf_rele(db
, FTAG
);
285 if (tbl
->zt_nextblk
!= 0) {
287 * read the nextblk for the sake of i/o error checking,
288 * so that zap_table_load() will catch errors for
291 blk
= (idx
*2) >> (bs
-3);
293 dn
= dmu_buf_dnode_enter(zap
->zap_dbuf
);
294 err
= dmu_buf_hold_by_dnode(dn
,
295 (tbl
->zt_nextblk
+ blk
) << bs
, FTAG
, &db
,
296 DMU_READ_NO_PREFETCH
);
297 dmu_buf_dnode_exit(zap
->zap_dbuf
);
299 dmu_buf_rele(db
, FTAG
);
305 * Routines for growing the ptrtbl.
309 zap_ptrtbl_transfer(const uint64_t *src
, uint64_t *dst
, int n
)
311 for (int i
= 0; i
< n
; i
++) {
312 uint64_t lb
= src
[i
];
319 zap_grow_ptrtbl(zap_t
*zap
, dmu_tx_t
*tx
)
322 * The pointer table should never use more hash bits than we
323 * have (otherwise we'd be using useless zero bits to index it).
324 * If we are within 2 bits of running out, stop growing, since
325 * this is already an aberrant condition.
327 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
>= zap_hashbits(zap
) - 2)
328 return (SET_ERROR(ENOSPC
));
330 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
== 0) {
332 * We are outgrowing the "embedded" ptrtbl (the one
333 * stored in the header block). Give it its own entire
334 * block, which will double the size of the ptrtbl.
336 ASSERT3U(zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
, ==,
337 ZAP_EMBEDDED_PTRTBL_SHIFT(zap
));
338 ASSERT0(zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
);
340 uint64_t newblk
= zap_allocate_blocks(zap
, 1);
342 int err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
343 newblk
<< FZAP_BLOCK_SHIFT(zap
), FTAG
, &db_new
,
344 DMU_READ_NO_PREFETCH
);
347 dmu_buf_will_dirty(db_new
, tx
);
348 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap
, 0),
349 db_new
->db_data
, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap
));
350 dmu_buf_rele(db_new
, FTAG
);
352 zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
= newblk
;
353 zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
= 1;
354 zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
++;
356 ASSERT3U(1ULL << zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
, ==,
357 zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
<<
358 (FZAP_BLOCK_SHIFT(zap
)-3));
362 return (zap_table_grow(zap
, &zap_f_phys(zap
)->zap_ptrtbl
,
363 zap_ptrtbl_transfer
, tx
));
368 zap_increment_num_entries(zap_t
*zap
, int delta
, dmu_tx_t
*tx
)
370 dmu_buf_will_dirty(zap
->zap_dbuf
, tx
);
371 mutex_enter(&zap
->zap_f
.zap_num_entries_mtx
);
372 ASSERT(delta
> 0 || zap_f_phys(zap
)->zap_num_entries
>= -delta
);
373 zap_f_phys(zap
)->zap_num_entries
+= delta
;
374 mutex_exit(&zap
->zap_f
.zap_num_entries_mtx
);
378 zap_allocate_blocks(zap_t
*zap
, int nblocks
)
380 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
381 uint64_t newblk
= zap_f_phys(zap
)->zap_freeblk
;
382 zap_f_phys(zap
)->zap_freeblk
+= nblocks
;
387 zap_leaf_evict_sync(void *dbu
)
391 rw_destroy(&l
->l_rwlock
);
392 kmem_free(l
, sizeof (zap_leaf_t
));
396 zap_create_leaf(zap_t
*zap
, dmu_tx_t
*tx
)
398 zap_leaf_t
*l
= kmem_zalloc(sizeof (zap_leaf_t
), KM_SLEEP
);
400 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
402 rw_init(&l
->l_rwlock
, NULL
, RW_NOLOCKDEP
, NULL
);
403 rw_enter(&l
->l_rwlock
, RW_WRITER
);
404 l
->l_blkid
= zap_allocate_blocks(zap
, 1);
407 VERIFY0(dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
408 l
->l_blkid
<< FZAP_BLOCK_SHIFT(zap
), NULL
, &l
->l_dbuf
,
409 DMU_READ_NO_PREFETCH
));
410 dmu_buf_init_user(&l
->l_dbu
, zap_leaf_evict_sync
, NULL
, &l
->l_dbuf
);
411 VERIFY3P(NULL
, ==, dmu_buf_set_user(l
->l_dbuf
, &l
->l_dbu
));
412 dmu_buf_will_dirty(l
->l_dbuf
, tx
);
414 zap_leaf_init(l
, zap
->zap_normflags
!= 0);
416 zap_f_phys(zap
)->zap_num_leafs
++;
422 fzap_count(zap_t
*zap
, uint64_t *count
)
424 ASSERT(!zap
->zap_ismicro
);
425 mutex_enter(&zap
->zap_f
.zap_num_entries_mtx
); /* unnecessary */
426 *count
= zap_f_phys(zap
)->zap_num_entries
;
427 mutex_exit(&zap
->zap_f
.zap_num_entries_mtx
);
432 * Routines for obtaining zap_leaf_t's
436 zap_put_leaf(zap_leaf_t
*l
)
438 rw_exit(&l
->l_rwlock
);
439 dmu_buf_rele(l
->l_dbuf
, NULL
);
443 zap_open_leaf(uint64_t blkid
, dmu_buf_t
*db
)
447 zap_leaf_t
*l
= kmem_zalloc(sizeof (zap_leaf_t
), KM_SLEEP
);
448 rw_init(&l
->l_rwlock
, NULL
, RW_DEFAULT
, NULL
);
449 rw_enter(&l
->l_rwlock
, RW_WRITER
);
451 l
->l_bs
= highbit64(db
->db_size
) - 1;
454 dmu_buf_init_user(&l
->l_dbu
, zap_leaf_evict_sync
, NULL
, &l
->l_dbuf
);
455 zap_leaf_t
*winner
= dmu_buf_set_user(db
, &l
->l_dbu
);
457 rw_exit(&l
->l_rwlock
);
458 if (winner
!= NULL
) {
459 /* someone else set it first */
460 zap_leaf_evict_sync(&l
->l_dbu
);
465 * lhr_pad was previously used for the next leaf in the leaf
466 * chain. There should be no chained leafs (as we have removed
469 ASSERT0(zap_leaf_phys(l
)->l_hdr
.lh_pad1
);
472 * There should be more hash entries than there can be
473 * chunks to put in the hash table
475 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l
), >, ZAP_LEAF_NUMCHUNKS(l
) / 3);
477 /* The chunks should begin at the end of the hash table */
478 ASSERT3P(&ZAP_LEAF_CHUNK(l
, 0), ==, (zap_leaf_chunk_t
*)
479 &zap_leaf_phys(l
)->l_hash
[ZAP_LEAF_HASH_NUMENTRIES(l
)]);
481 /* The chunks should end at the end of the block */
482 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l
, ZAP_LEAF_NUMCHUNKS(l
)) -
483 (uintptr_t)zap_leaf_phys(l
), ==, l
->l_dbuf
->db_size
);
489 zap_get_leaf_byblk(zap_t
*zap
, uint64_t blkid
, dmu_tx_t
*tx
, krw_t lt
,
494 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
497 * If system crashed just after dmu_free_long_range in zfs_rmnode, we
498 * would be left with an empty xattr dir in delete queue. blkid=0
499 * would be passed in when doing zfs_purgedir. If that's the case we
500 * should just return immediately. The underlying objects should
501 * already be freed, so this should be perfectly fine.
504 return (SET_ERROR(ENOENT
));
506 int bs
= FZAP_BLOCK_SHIFT(zap
);
507 dnode_t
*dn
= dmu_buf_dnode_enter(zap
->zap_dbuf
);
508 int err
= dmu_buf_hold_by_dnode(dn
,
509 blkid
<< bs
, NULL
, &db
, DMU_READ_NO_PREFETCH
);
510 dmu_buf_dnode_exit(zap
->zap_dbuf
);
514 ASSERT3U(db
->db_object
, ==, zap
->zap_object
);
515 ASSERT3U(db
->db_offset
, ==, blkid
<< bs
);
516 ASSERT3U(db
->db_size
, ==, 1 << bs
);
519 zap_leaf_t
*l
= dmu_buf_get_user(db
);
522 l
= zap_open_leaf(blkid
, db
);
524 rw_enter(&l
->l_rwlock
, lt
);
526 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
527 * causing ASSERT below to fail.
530 dmu_buf_will_dirty(db
, tx
);
531 ASSERT3U(l
->l_blkid
, ==, blkid
);
532 ASSERT3P(l
->l_dbuf
, ==, db
);
533 ASSERT3U(zap_leaf_phys(l
)->l_hdr
.lh_block_type
, ==, ZBT_LEAF
);
534 ASSERT3U(zap_leaf_phys(l
)->l_hdr
.lh_magic
, ==, ZAP_LEAF_MAGIC
);
541 zap_idx_to_blk(zap_t
*zap
, uint64_t idx
, uint64_t *valp
)
543 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
545 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
== 0) {
547 (1ULL << zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
));
548 *valp
= ZAP_EMBEDDED_PTRTBL_ENT(zap
, idx
);
551 return (zap_table_load(zap
, &zap_f_phys(zap
)->zap_ptrtbl
,
557 zap_set_idx_to_blk(zap_t
*zap
, uint64_t idx
, uint64_t blk
, dmu_tx_t
*tx
)
560 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
562 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
== 0) {
563 ZAP_EMBEDDED_PTRTBL_ENT(zap
, idx
) = blk
;
566 return (zap_table_store(zap
, &zap_f_phys(zap
)->zap_ptrtbl
,
572 zap_deref_leaf(zap_t
*zap
, uint64_t h
, dmu_tx_t
*tx
, krw_t lt
, zap_leaf_t
**lp
)
576 ASSERT(zap
->zap_dbuf
== NULL
||
577 zap_f_phys(zap
) == zap
->zap_dbuf
->db_data
);
579 /* Reality check for corrupt zap objects (leaf or header). */
580 if ((zap_f_phys(zap
)->zap_block_type
!= ZBT_LEAF
&&
581 zap_f_phys(zap
)->zap_block_type
!= ZBT_HEADER
) ||
582 zap_f_phys(zap
)->zap_magic
!= ZAP_MAGIC
) {
583 return (SET_ERROR(EIO
));
586 uint64_t idx
= ZAP_HASH_IDX(h
, zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
587 int err
= zap_idx_to_blk(zap
, idx
, &blk
);
590 err
= zap_get_leaf_byblk(zap
, blk
, tx
, lt
, lp
);
593 ZAP_HASH_IDX(h
, zap_leaf_phys(*lp
)->l_hdr
.lh_prefix_len
) ==
594 zap_leaf_phys(*lp
)->l_hdr
.lh_prefix
);
599 zap_expand_leaf(zap_name_t
*zn
, zap_leaf_t
*l
,
600 void *tag
, dmu_tx_t
*tx
, zap_leaf_t
**lp
)
602 zap_t
*zap
= zn
->zn_zap
;
603 uint64_t hash
= zn
->zn_hash
;
605 int old_prefix_len
= zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
;
607 ASSERT3U(old_prefix_len
, <=, zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
608 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
610 ASSERT3U(ZAP_HASH_IDX(hash
, old_prefix_len
), ==,
611 zap_leaf_phys(l
)->l_hdr
.lh_prefix
);
613 if (zap_tryupgradedir(zap
, tx
) == 0 ||
614 old_prefix_len
== zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
) {
615 /* We failed to upgrade, or need to grow the pointer table */
616 objset_t
*os
= zap
->zap_objset
;
617 uint64_t object
= zap
->zap_object
;
620 zap_unlockdir(zap
, tag
);
621 err
= zap_lockdir(os
, object
, tx
, RW_WRITER
,
622 FALSE
, FALSE
, tag
, &zn
->zn_zap
);
626 ASSERT(!zap
->zap_ismicro
);
628 while (old_prefix_len
==
629 zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
) {
630 err
= zap_grow_ptrtbl(zap
, tx
);
635 err
= zap_deref_leaf(zap
, hash
, tx
, RW_WRITER
, &l
);
639 if (zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
!= old_prefix_len
) {
640 /* it split while our locks were down */
645 ASSERT(RW_WRITE_HELD(&zap
->zap_rwlock
));
646 ASSERT3U(old_prefix_len
, <, zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
647 ASSERT3U(ZAP_HASH_IDX(hash
, old_prefix_len
), ==,
648 zap_leaf_phys(l
)->l_hdr
.lh_prefix
);
650 int prefix_diff
= zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
-
651 (old_prefix_len
+ 1);
653 (ZAP_HASH_IDX(hash
, old_prefix_len
+ 1) | 1) << prefix_diff
;
655 /* check for i/o errors before doing zap_leaf_split */
656 for (int i
= 0; i
< (1ULL << prefix_diff
); i
++) {
658 err
= zap_idx_to_blk(zap
, sibling
+ i
, &blk
);
661 ASSERT3U(blk
, ==, l
->l_blkid
);
664 zap_leaf_t
*nl
= zap_create_leaf(zap
, tx
);
665 zap_leaf_split(l
, nl
, zap
->zap_normflags
!= 0);
667 /* set sibling pointers */
668 for (int i
= 0; i
< (1ULL << prefix_diff
); i
++) {
669 err
= zap_set_idx_to_blk(zap
, sibling
+ i
, nl
->l_blkid
, tx
);
670 ASSERT0(err
); /* we checked for i/o errors above */
673 ASSERT3U(zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
, >, 0);
675 if (hash
& (1ULL << (64 - zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
))) {
676 /* we want the sibling */
688 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t
*zn
, zap_leaf_t
*l
,
689 void *tag
, dmu_tx_t
*tx
)
691 zap_t
*zap
= zn
->zn_zap
;
692 int shift
= zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
;
693 int leaffull
= (zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
== shift
&&
694 zap_leaf_phys(l
)->l_hdr
.lh_nfree
< ZAP_LEAF_LOW_WATER
);
698 if (leaffull
|| zap_f_phys(zap
)->zap_ptrtbl
.zt_nextblk
) {
700 * We are in the middle of growing the pointer table, or
701 * this leaf will soon make us grow it.
703 if (zap_tryupgradedir(zap
, tx
) == 0) {
704 objset_t
*os
= zap
->zap_objset
;
705 uint64_t zapobj
= zap
->zap_object
;
707 zap_unlockdir(zap
, tag
);
708 int err
= zap_lockdir(os
, zapobj
, tx
,
709 RW_WRITER
, FALSE
, FALSE
, tag
, &zn
->zn_zap
);
715 /* could have finished growing while our locks were down */
716 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
== shift
)
717 (void) zap_grow_ptrtbl(zap
, tx
);
722 fzap_checkname(zap_name_t
*zn
)
724 if (zn
->zn_key_orig_numints
* zn
->zn_key_intlen
> ZAP_MAXNAMELEN
)
725 return (SET_ERROR(ENAMETOOLONG
));
730 fzap_checksize(uint64_t integer_size
, uint64_t num_integers
)
732 /* Only integer sizes supported by C */
733 switch (integer_size
) {
740 return (SET_ERROR(EINVAL
));
743 if (integer_size
* num_integers
> ZAP_MAXVALUELEN
)
744 return (SET_ERROR(E2BIG
));
750 fzap_check(zap_name_t
*zn
, uint64_t integer_size
, uint64_t num_integers
)
752 int err
= fzap_checkname(zn
);
755 return (fzap_checksize(integer_size
, num_integers
));
759 * Routines for manipulating attributes.
762 fzap_lookup(zap_name_t
*zn
,
763 uint64_t integer_size
, uint64_t num_integers
, void *buf
,
764 char *realname
, int rn_len
, boolean_t
*ncp
)
767 zap_entry_handle_t zeh
;
769 int err
= fzap_checkname(zn
);
773 err
= zap_deref_leaf(zn
->zn_zap
, zn
->zn_hash
, NULL
, RW_READER
, &l
);
776 err
= zap_leaf_lookup(l
, zn
, &zeh
);
778 if ((err
= fzap_checksize(integer_size
, num_integers
)) != 0) {
783 err
= zap_entry_read(&zeh
, integer_size
, num_integers
, buf
);
784 (void) zap_entry_read_name(zn
->zn_zap
, &zeh
, rn_len
, realname
);
786 *ncp
= zap_entry_normalization_conflict(&zeh
,
787 zn
, NULL
, zn
->zn_zap
);
796 fzap_add_cd(zap_name_t
*zn
,
797 uint64_t integer_size
, uint64_t num_integers
,
798 const void *val
, uint32_t cd
, void *tag
, dmu_tx_t
*tx
)
802 zap_entry_handle_t zeh
;
803 zap_t
*zap
= zn
->zn_zap
;
805 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
806 ASSERT(!zap
->zap_ismicro
);
807 ASSERT(fzap_check(zn
, integer_size
, num_integers
) == 0);
809 err
= zap_deref_leaf(zap
, zn
->zn_hash
, tx
, RW_WRITER
, &l
);
813 err
= zap_leaf_lookup(l
, zn
, &zeh
);
815 err
= SET_ERROR(EEXIST
);
821 err
= zap_entry_create(l
, zn
, cd
,
822 integer_size
, num_integers
, val
, &zeh
);
825 zap_increment_num_entries(zap
, 1, tx
);
826 } else if (err
== EAGAIN
) {
827 err
= zap_expand_leaf(zn
, l
, tag
, tx
, &l
);
828 zap
= zn
->zn_zap
; /* zap_expand_leaf() may change zap */
831 } else if (err
== ENOSPC
) {
833 * If we failed to expand the leaf, then bailout
834 * as there is no point trying
835 * zap_put_leaf_maybe_grow_ptrtbl().
843 zap_put_leaf_maybe_grow_ptrtbl(zn
, l
, tag
, tx
);
848 fzap_add(zap_name_t
*zn
,
849 uint64_t integer_size
, uint64_t num_integers
,
850 const void *val
, void *tag
, dmu_tx_t
*tx
)
852 int err
= fzap_check(zn
, integer_size
, num_integers
);
856 return (fzap_add_cd(zn
, integer_size
, num_integers
,
857 val
, ZAP_NEED_CD
, tag
, tx
));
861 fzap_update(zap_name_t
*zn
,
862 int integer_size
, uint64_t num_integers
, const void *val
,
863 void *tag
, dmu_tx_t
*tx
)
868 zap_entry_handle_t zeh
;
869 zap_t
*zap
= zn
->zn_zap
;
871 ASSERT(RW_LOCK_HELD(&zap
->zap_rwlock
));
872 err
= fzap_check(zn
, integer_size
, num_integers
);
876 err
= zap_deref_leaf(zap
, zn
->zn_hash
, tx
, RW_WRITER
, &l
);
880 err
= zap_leaf_lookup(l
, zn
, &zeh
);
881 create
= (err
== ENOENT
);
882 ASSERT(err
== 0 || err
== ENOENT
);
885 err
= zap_entry_create(l
, zn
, ZAP_NEED_CD
,
886 integer_size
, num_integers
, val
, &zeh
);
888 zap_increment_num_entries(zap
, 1, tx
);
890 err
= zap_entry_update(&zeh
, integer_size
, num_integers
, val
);
894 err
= zap_expand_leaf(zn
, l
, tag
, tx
, &l
);
895 zap
= zn
->zn_zap
; /* zap_expand_leaf() may change zap */
901 zap_put_leaf_maybe_grow_ptrtbl(zn
, l
, tag
, tx
);
906 fzap_length(zap_name_t
*zn
,
907 uint64_t *integer_size
, uint64_t *num_integers
)
911 zap_entry_handle_t zeh
;
913 err
= zap_deref_leaf(zn
->zn_zap
, zn
->zn_hash
, NULL
, RW_READER
, &l
);
916 err
= zap_leaf_lookup(l
, zn
, &zeh
);
920 if (integer_size
!= 0)
921 *integer_size
= zeh
.zeh_integer_size
;
922 if (num_integers
!= 0)
923 *num_integers
= zeh
.zeh_num_integers
;
930 fzap_remove(zap_name_t
*zn
, dmu_tx_t
*tx
)
934 zap_entry_handle_t zeh
;
936 err
= zap_deref_leaf(zn
->zn_zap
, zn
->zn_hash
, tx
, RW_WRITER
, &l
);
939 err
= zap_leaf_lookup(l
, zn
, &zeh
);
941 zap_entry_remove(&zeh
);
942 zap_increment_num_entries(zn
->zn_zap
, -1, tx
);
949 fzap_prefetch(zap_name_t
*zn
)
952 zap_t
*zap
= zn
->zn_zap
;
954 uint64_t idx
= ZAP_HASH_IDX(zn
->zn_hash
,
955 zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
);
956 if (zap_idx_to_blk(zap
, idx
, &blk
) != 0)
958 int bs
= FZAP_BLOCK_SHIFT(zap
);
959 dmu_prefetch(zap
->zap_objset
, zap
->zap_object
, 0, blk
<< bs
, 1 << bs
,
960 ZIO_PRIORITY_SYNC_READ
);
964 * Helper functions for consumers.
968 zap_create_link(objset_t
*os
, dmu_object_type_t ot
, uint64_t parent_obj
,
969 const char *name
, dmu_tx_t
*tx
)
971 return (zap_create_link_dnsize(os
, ot
, parent_obj
, name
, 0, tx
));
975 zap_create_link_dnsize(objset_t
*os
, dmu_object_type_t ot
, uint64_t parent_obj
,
976 const char *name
, int dnodesize
, dmu_tx_t
*tx
)
980 new_obj
= zap_create_dnsize(os
, ot
, DMU_OT_NONE
, 0, dnodesize
, tx
);
981 VERIFY(new_obj
!= 0);
982 VERIFY0(zap_add(os
, parent_obj
, name
, sizeof (uint64_t), 1, &new_obj
,
989 zap_value_search(objset_t
*os
, uint64_t zapobj
, uint64_t value
, uint64_t mask
,
998 zap_attribute_t
*za
= kmem_alloc(sizeof (*za
), KM_SLEEP
);
999 for (zap_cursor_init(&zc
, os
, zapobj
);
1000 (err
= zap_cursor_retrieve(&zc
, za
)) == 0;
1001 zap_cursor_advance(&zc
)) {
1002 if ((za
->za_first_integer
& mask
) == (value
& mask
)) {
1003 (void) strcpy(name
, za
->za_name
);
1007 zap_cursor_fini(&zc
);
1008 kmem_free(za
, sizeof (*za
));
1013 zap_join(objset_t
*os
, uint64_t fromobj
, uint64_t intoobj
, dmu_tx_t
*tx
)
1018 zap_attribute_t
*za
= kmem_alloc(sizeof (*za
), KM_SLEEP
);
1019 for (zap_cursor_init(&zc
, os
, fromobj
);
1020 zap_cursor_retrieve(&zc
, za
) == 0;
1021 (void) zap_cursor_advance(&zc
)) {
1022 if (za
->za_integer_length
!= 8 || za
->za_num_integers
!= 1) {
1023 err
= SET_ERROR(EINVAL
);
1026 err
= zap_add(os
, intoobj
, za
->za_name
,
1027 8, 1, &za
->za_first_integer
, tx
);
1031 zap_cursor_fini(&zc
);
1032 kmem_free(za
, sizeof (*za
));
1037 zap_join_key(objset_t
*os
, uint64_t fromobj
, uint64_t intoobj
,
1038 uint64_t value
, dmu_tx_t
*tx
)
1043 zap_attribute_t
*za
= kmem_alloc(sizeof (*za
), KM_SLEEP
);
1044 for (zap_cursor_init(&zc
, os
, fromobj
);
1045 zap_cursor_retrieve(&zc
, za
) == 0;
1046 (void) zap_cursor_advance(&zc
)) {
1047 if (za
->za_integer_length
!= 8 || za
->za_num_integers
!= 1) {
1048 err
= SET_ERROR(EINVAL
);
1051 err
= zap_add(os
, intoobj
, za
->za_name
,
1056 zap_cursor_fini(&zc
);
1057 kmem_free(za
, sizeof (*za
));
1062 zap_join_increment(objset_t
*os
, uint64_t fromobj
, uint64_t intoobj
,
1068 zap_attribute_t
*za
= kmem_alloc(sizeof (*za
), KM_SLEEP
);
1069 for (zap_cursor_init(&zc
, os
, fromobj
);
1070 zap_cursor_retrieve(&zc
, za
) == 0;
1071 (void) zap_cursor_advance(&zc
)) {
1074 if (za
->za_integer_length
!= 8 || za
->za_num_integers
!= 1) {
1075 err
= SET_ERROR(EINVAL
);
1079 err
= zap_lookup(os
, intoobj
, za
->za_name
, 8, 1, &delta
);
1080 if (err
!= 0 && err
!= ENOENT
)
1082 delta
+= za
->za_first_integer
;
1083 err
= zap_update(os
, intoobj
, za
->za_name
, 8, 1, &delta
, tx
);
1087 zap_cursor_fini(&zc
);
1088 kmem_free(za
, sizeof (*za
));
1093 zap_add_int(objset_t
*os
, uint64_t obj
, uint64_t value
, dmu_tx_t
*tx
)
1097 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)value
);
1098 return (zap_add(os
, obj
, name
, 8, 1, &value
, tx
));
1102 zap_remove_int(objset_t
*os
, uint64_t obj
, uint64_t value
, dmu_tx_t
*tx
)
1106 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)value
);
1107 return (zap_remove(os
, obj
, name
, tx
));
1111 zap_lookup_int(objset_t
*os
, uint64_t obj
, uint64_t value
)
1115 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)value
);
1116 return (zap_lookup(os
, obj
, name
, 8, 1, &value
));
1120 zap_add_int_key(objset_t
*os
, uint64_t obj
,
1121 uint64_t key
, uint64_t value
, dmu_tx_t
*tx
)
1125 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1126 return (zap_add(os
, obj
, name
, 8, 1, &value
, tx
));
1130 zap_update_int_key(objset_t
*os
, uint64_t obj
,
1131 uint64_t key
, uint64_t value
, dmu_tx_t
*tx
)
1135 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1136 return (zap_update(os
, obj
, name
, 8, 1, &value
, tx
));
1140 zap_lookup_int_key(objset_t
*os
, uint64_t obj
, uint64_t key
, uint64_t *valuep
)
1144 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1145 return (zap_lookup(os
, obj
, name
, 8, 1, valuep
));
1149 zap_increment(objset_t
*os
, uint64_t obj
, const char *name
, int64_t delta
,
1157 int err
= zap_lookup(os
, obj
, name
, 8, 1, &value
);
1158 if (err
!= 0 && err
!= ENOENT
)
1162 err
= zap_remove(os
, obj
, name
, tx
);
1164 err
= zap_update(os
, obj
, name
, 8, 1, &value
, tx
);
1169 zap_increment_int(objset_t
*os
, uint64_t obj
, uint64_t key
, int64_t delta
,
1174 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)key
);
1175 return (zap_increment(os
, obj
, name
, delta
, tx
));
1179 * Routines for iterating over the attributes.
1183 fzap_cursor_retrieve(zap_t
*zap
, zap_cursor_t
*zc
, zap_attribute_t
*za
)
1186 zap_entry_handle_t zeh
;
1189 /* retrieve the next entry at or after zc_hash/zc_cd */
1190 /* if no entry, return ENOENT */
1193 (ZAP_HASH_IDX(zc
->zc_hash
,
1194 zap_leaf_phys(zc
->zc_leaf
)->l_hdr
.lh_prefix_len
) !=
1195 zap_leaf_phys(zc
->zc_leaf
)->l_hdr
.lh_prefix
)) {
1196 rw_enter(&zc
->zc_leaf
->l_rwlock
, RW_READER
);
1197 zap_put_leaf(zc
->zc_leaf
);
1202 if (zc
->zc_leaf
== NULL
) {
1203 err
= zap_deref_leaf(zap
, zc
->zc_hash
, NULL
, RW_READER
,
1208 rw_enter(&zc
->zc_leaf
->l_rwlock
, RW_READER
);
1212 err
= zap_leaf_lookup_closest(l
, zc
->zc_hash
, zc
->zc_cd
, &zeh
);
1214 if (err
== ENOENT
) {
1215 if (zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
== 0) {
1216 zc
->zc_hash
= -1ULL;
1219 uint64_t nocare
= (1ULL <<
1220 (64 - zap_leaf_phys(l
)->l_hdr
.lh_prefix_len
)) - 1;
1222 zc
->zc_hash
= (zc
->zc_hash
& ~nocare
) + nocare
+ 1;
1225 if (zc
->zc_hash
== 0) {
1226 zc
->zc_hash
= -1ULL;
1228 zap_put_leaf(zc
->zc_leaf
);
1236 zc
->zc_hash
= zeh
.zeh_hash
;
1237 zc
->zc_cd
= zeh
.zeh_cd
;
1238 za
->za_integer_length
= zeh
.zeh_integer_size
;
1239 za
->za_num_integers
= zeh
.zeh_num_integers
;
1240 if (zeh
.zeh_num_integers
== 0) {
1241 za
->za_first_integer
= 0;
1243 err
= zap_entry_read(&zeh
, 8, 1, &za
->za_first_integer
);
1244 ASSERT(err
== 0 || err
== EOVERFLOW
);
1246 err
= zap_entry_read_name(zap
, &zeh
,
1247 sizeof (za
->za_name
), za
->za_name
);
1250 za
->za_normalization_conflict
=
1251 zap_entry_normalization_conflict(&zeh
,
1252 NULL
, za
->za_name
, zap
);
1254 rw_exit(&zc
->zc_leaf
->l_rwlock
);
1259 zap_stats_ptrtbl(zap_t
*zap
, uint64_t *tbl
, int len
, zap_stats_t
*zs
)
1261 uint64_t lastblk
= 0;
1264 * NB: if a leaf has more pointers than an entire ptrtbl block
1265 * can hold, then it'll be accounted for more than once, since
1266 * we won't have lastblk.
1268 for (int i
= 0; i
< len
; i
++) {
1271 if (tbl
[i
] == lastblk
)
1275 int err
= zap_get_leaf_byblk(zap
, tbl
[i
], NULL
, RW_READER
, &l
);
1277 zap_leaf_stats(zap
, l
, zs
);
1284 fzap_get_stats(zap_t
*zap
, zap_stats_t
*zs
)
1286 int bs
= FZAP_BLOCK_SHIFT(zap
);
1287 zs
->zs_blocksize
= 1ULL << bs
;
1290 * Set zap_phys_t fields
1292 zs
->zs_num_leafs
= zap_f_phys(zap
)->zap_num_leafs
;
1293 zs
->zs_num_entries
= zap_f_phys(zap
)->zap_num_entries
;
1294 zs
->zs_num_blocks
= zap_f_phys(zap
)->zap_freeblk
;
1295 zs
->zs_block_type
= zap_f_phys(zap
)->zap_block_type
;
1296 zs
->zs_magic
= zap_f_phys(zap
)->zap_magic
;
1297 zs
->zs_salt
= zap_f_phys(zap
)->zap_salt
;
1300 * Set zap_ptrtbl fields
1302 zs
->zs_ptrtbl_len
= 1ULL << zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
;
1303 zs
->zs_ptrtbl_nextblk
= zap_f_phys(zap
)->zap_ptrtbl
.zt_nextblk
;
1304 zs
->zs_ptrtbl_blks_copied
=
1305 zap_f_phys(zap
)->zap_ptrtbl
.zt_blks_copied
;
1306 zs
->zs_ptrtbl_zt_blk
= zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
;
1307 zs
->zs_ptrtbl_zt_numblks
= zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
;
1308 zs
->zs_ptrtbl_zt_shift
= zap_f_phys(zap
)->zap_ptrtbl
.zt_shift
;
1310 if (zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
== 0) {
1311 /* the ptrtbl is entirely in the header block. */
1312 zap_stats_ptrtbl(zap
, &ZAP_EMBEDDED_PTRTBL_ENT(zap
, 0),
1313 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap
), zs
);
1315 dmu_prefetch(zap
->zap_objset
, zap
->zap_object
, 0,
1316 zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
<< bs
,
1317 zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
<< bs
,
1318 ZIO_PRIORITY_SYNC_READ
);
1320 for (int b
= 0; b
< zap_f_phys(zap
)->zap_ptrtbl
.zt_numblks
;
1325 err
= dmu_buf_hold(zap
->zap_objset
, zap
->zap_object
,
1326 (zap_f_phys(zap
)->zap_ptrtbl
.zt_blk
+ b
) << bs
,
1327 FTAG
, &db
, DMU_READ_NO_PREFETCH
);
1329 zap_stats_ptrtbl(zap
, db
->db_data
,
1331 dmu_buf_rele(db
, FTAG
);